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;
1060 * We have already validated the presence of BTRFS_DEV_REPLACE_DEVID,
1061 * in btrfs_init_dev_replace() so just continue.
1063 if (device->devid == BTRFS_DEV_REPLACE_DEVID)
1067 blkdev_put(device->bdev, device->mode);
1068 device->bdev = NULL;
1069 fs_devices->open_devices--;
1071 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1072 list_del_init(&device->dev_alloc_list);
1073 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
1075 list_del_init(&device->dev_list);
1076 fs_devices->num_devices--;
1077 btrfs_free_device(device);
1083 * After we have read the system tree and know devids belonging to this
1084 * filesystem, remove the device which does not belong there.
1086 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices, int step)
1088 struct btrfs_device *latest_dev = NULL;
1089 struct btrfs_fs_devices *seed_dev;
1091 mutex_lock(&uuid_mutex);
1092 __btrfs_free_extra_devids(fs_devices, step, &latest_dev);
1094 list_for_each_entry(seed_dev, &fs_devices->seed_list, seed_list)
1095 __btrfs_free_extra_devids(seed_dev, step, &latest_dev);
1097 fs_devices->latest_bdev = latest_dev->bdev;
1099 mutex_unlock(&uuid_mutex);
1102 static void btrfs_close_bdev(struct btrfs_device *device)
1107 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1108 sync_blockdev(device->bdev);
1109 invalidate_bdev(device->bdev);
1112 blkdev_put(device->bdev, device->mode);
1115 static void btrfs_close_one_device(struct btrfs_device *device)
1117 struct btrfs_fs_devices *fs_devices = device->fs_devices;
1119 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
1120 device->devid != BTRFS_DEV_REPLACE_DEVID) {
1121 list_del_init(&device->dev_alloc_list);
1122 fs_devices->rw_devices--;
1125 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
1126 fs_devices->missing_devices--;
1128 btrfs_close_bdev(device);
1130 fs_devices->open_devices--;
1131 device->bdev = NULL;
1133 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
1135 device->fs_info = NULL;
1136 atomic_set(&device->dev_stats_ccnt, 0);
1137 extent_io_tree_release(&device->alloc_state);
1139 /* Verify the device is back in a pristine state */
1140 ASSERT(!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state));
1141 ASSERT(!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state));
1142 ASSERT(list_empty(&device->dev_alloc_list));
1143 ASSERT(list_empty(&device->post_commit_list));
1144 ASSERT(atomic_read(&device->reada_in_flight) == 0);
1147 static void close_fs_devices(struct btrfs_fs_devices *fs_devices)
1149 struct btrfs_device *device, *tmp;
1151 lockdep_assert_held(&uuid_mutex);
1153 if (--fs_devices->opened > 0)
1156 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list)
1157 btrfs_close_one_device(device);
1159 WARN_ON(fs_devices->open_devices);
1160 WARN_ON(fs_devices->rw_devices);
1161 fs_devices->opened = 0;
1162 fs_devices->seeding = false;
1163 fs_devices->fs_info = NULL;
1166 void btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
1169 struct btrfs_fs_devices *tmp;
1171 mutex_lock(&uuid_mutex);
1172 close_fs_devices(fs_devices);
1173 if (!fs_devices->opened)
1174 list_splice_init(&fs_devices->seed_list, &list);
1176 list_for_each_entry_safe(fs_devices, tmp, &list, seed_list) {
1177 close_fs_devices(fs_devices);
1178 list_del(&fs_devices->seed_list);
1179 free_fs_devices(fs_devices);
1181 mutex_unlock(&uuid_mutex);
1184 static int open_fs_devices(struct btrfs_fs_devices *fs_devices,
1185 fmode_t flags, void *holder)
1187 struct btrfs_device *device;
1188 struct btrfs_device *latest_dev = NULL;
1189 struct btrfs_device *tmp_device;
1191 flags |= FMODE_EXCL;
1193 list_for_each_entry_safe(device, tmp_device, &fs_devices->devices,
1197 ret = btrfs_open_one_device(fs_devices, device, flags, holder);
1199 (!latest_dev || device->generation > latest_dev->generation)) {
1200 latest_dev = device;
1201 } else if (ret == -ENODATA) {
1202 fs_devices->num_devices--;
1203 list_del(&device->dev_list);
1204 btrfs_free_device(device);
1207 if (fs_devices->open_devices == 0)
1210 fs_devices->opened = 1;
1211 fs_devices->latest_bdev = latest_dev->bdev;
1212 fs_devices->total_rw_bytes = 0;
1213 fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_REGULAR;
1218 static int devid_cmp(void *priv, struct list_head *a, struct list_head *b)
1220 struct btrfs_device *dev1, *dev2;
1222 dev1 = list_entry(a, struct btrfs_device, dev_list);
1223 dev2 = list_entry(b, struct btrfs_device, dev_list);
1225 if (dev1->devid < dev2->devid)
1227 else if (dev1->devid > dev2->devid)
1232 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1233 fmode_t flags, void *holder)
1237 lockdep_assert_held(&uuid_mutex);
1239 * The device_list_mutex cannot be taken here in case opening the
1240 * underlying device takes further locks like bd_mutex.
1242 * We also don't need the lock here as this is called during mount and
1243 * exclusion is provided by uuid_mutex
1246 if (fs_devices->opened) {
1247 fs_devices->opened++;
1250 list_sort(NULL, &fs_devices->devices, devid_cmp);
1251 ret = open_fs_devices(fs_devices, flags, holder);
1257 void btrfs_release_disk_super(struct btrfs_super_block *super)
1259 struct page *page = virt_to_page(super);
1264 static struct btrfs_super_block *btrfs_read_disk_super(struct block_device *bdev,
1267 struct btrfs_super_block *disk_super;
1272 /* make sure our super fits in the device */
1273 if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1274 return ERR_PTR(-EINVAL);
1276 /* make sure our super fits in the page */
1277 if (sizeof(*disk_super) > PAGE_SIZE)
1278 return ERR_PTR(-EINVAL);
1280 /* make sure our super doesn't straddle pages on disk */
1281 index = bytenr >> PAGE_SHIFT;
1282 if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_SHIFT != index)
1283 return ERR_PTR(-EINVAL);
1285 /* pull in the page with our super */
1286 page = read_cache_page_gfp(bdev->bd_inode->i_mapping, index, GFP_KERNEL);
1289 return ERR_CAST(page);
1291 p = page_address(page);
1293 /* align our pointer to the offset of the super block */
1294 disk_super = p + offset_in_page(bytenr);
1296 if (btrfs_super_bytenr(disk_super) != bytenr ||
1297 btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
1298 btrfs_release_disk_super(p);
1299 return ERR_PTR(-EINVAL);
1302 if (disk_super->label[0] && disk_super->label[BTRFS_LABEL_SIZE - 1])
1303 disk_super->label[BTRFS_LABEL_SIZE - 1] = 0;
1308 int btrfs_forget_devices(const char *path)
1312 mutex_lock(&uuid_mutex);
1313 ret = btrfs_free_stale_devices(strlen(path) ? path : NULL, NULL);
1314 mutex_unlock(&uuid_mutex);
1320 * Look for a btrfs signature on a device. This may be called out of the mount path
1321 * and we are not allowed to call set_blocksize during the scan. The superblock
1322 * is read via pagecache
1324 struct btrfs_device *btrfs_scan_one_device(const char *path, fmode_t flags,
1327 struct btrfs_super_block *disk_super;
1328 bool new_device_added = false;
1329 struct btrfs_device *device = NULL;
1330 struct block_device *bdev;
1333 lockdep_assert_held(&uuid_mutex);
1336 * we would like to check all the supers, but that would make
1337 * a btrfs mount succeed after a mkfs from a different FS.
1338 * So, we need to add a special mount option to scan for
1339 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1341 bytenr = btrfs_sb_offset(0);
1342 flags |= FMODE_EXCL;
1344 bdev = blkdev_get_by_path(path, flags, holder);
1346 return ERR_CAST(bdev);
1348 disk_super = btrfs_read_disk_super(bdev, bytenr);
1349 if (IS_ERR(disk_super)) {
1350 device = ERR_CAST(disk_super);
1351 goto error_bdev_put;
1354 device = device_list_add(path, disk_super, &new_device_added);
1355 if (!IS_ERR(device)) {
1356 if (new_device_added)
1357 btrfs_free_stale_devices(path, device);
1360 btrfs_release_disk_super(disk_super);
1363 blkdev_put(bdev, flags);
1369 * Try to find a chunk that intersects [start, start + len] range and when one
1370 * such is found, record the end of it in *start
1372 static bool contains_pending_extent(struct btrfs_device *device, u64 *start,
1375 u64 physical_start, physical_end;
1377 lockdep_assert_held(&device->fs_info->chunk_mutex);
1379 if (!find_first_extent_bit(&device->alloc_state, *start,
1380 &physical_start, &physical_end,
1381 CHUNK_ALLOCATED, NULL)) {
1383 if (in_range(physical_start, *start, len) ||
1384 in_range(*start, physical_start,
1385 physical_end - physical_start)) {
1386 *start = physical_end + 1;
1393 static u64 dev_extent_search_start(struct btrfs_device *device, u64 start)
1395 switch (device->fs_devices->chunk_alloc_policy) {
1396 case BTRFS_CHUNK_ALLOC_REGULAR:
1398 * We don't want to overwrite the superblock on the drive nor
1399 * any area used by the boot loader (grub for example), so we
1400 * make sure to start at an offset of at least 1MB.
1402 return max_t(u64, start, SZ_1M);
1409 * dev_extent_hole_check - check if specified hole is suitable for allocation
1410 * @device: the device which we have the hole
1411 * @hole_start: starting position of the hole
1412 * @hole_size: the size of the hole
1413 * @num_bytes: the size of the free space that we need
1415 * This function may modify @hole_start and @hole_end to reflect the suitable
1416 * position for allocation. Returns 1 if hole position is updated, 0 otherwise.
1418 static bool dev_extent_hole_check(struct btrfs_device *device, u64 *hole_start,
1419 u64 *hole_size, u64 num_bytes)
1421 bool changed = false;
1422 u64 hole_end = *hole_start + *hole_size;
1425 * Check before we set max_hole_start, otherwise we could end up
1426 * sending back this offset anyway.
1428 if (contains_pending_extent(device, hole_start, *hole_size)) {
1429 if (hole_end >= *hole_start)
1430 *hole_size = hole_end - *hole_start;
1436 switch (device->fs_devices->chunk_alloc_policy) {
1437 case BTRFS_CHUNK_ALLOC_REGULAR:
1438 /* No extra check */
1448 * find_free_dev_extent_start - find free space in the specified device
1449 * @device: the device which we search the free space in
1450 * @num_bytes: the size of the free space that we need
1451 * @search_start: the position from which to begin the search
1452 * @start: store the start of the free space.
1453 * @len: the size of the free space. that we find, or the size
1454 * of the max free space if we don't find suitable free space
1456 * this uses a pretty simple search, the expectation is that it is
1457 * called very infrequently and that a given device has a small number
1460 * @start is used to store the start of the free space if we find. But if we
1461 * don't find suitable free space, it will be used to store the start position
1462 * of the max free space.
1464 * @len is used to store the size of the free space that we find.
1465 * But if we don't find suitable free space, it is used to store the size of
1466 * the max free space.
1468 * NOTE: This function will search *commit* root of device tree, and does extra
1469 * check to ensure dev extents are not double allocated.
1470 * This makes the function safe to allocate dev extents but may not report
1471 * correct usable device space, as device extent freed in current transaction
1472 * is not reported as avaiable.
1474 static int find_free_dev_extent_start(struct btrfs_device *device,
1475 u64 num_bytes, u64 search_start, u64 *start,
1478 struct btrfs_fs_info *fs_info = device->fs_info;
1479 struct btrfs_root *root = fs_info->dev_root;
1480 struct btrfs_key key;
1481 struct btrfs_dev_extent *dev_extent;
1482 struct btrfs_path *path;
1487 u64 search_end = device->total_bytes;
1490 struct extent_buffer *l;
1492 search_start = dev_extent_search_start(device, search_start);
1494 path = btrfs_alloc_path();
1498 max_hole_start = search_start;
1502 if (search_start >= search_end ||
1503 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1508 path->reada = READA_FORWARD;
1509 path->search_commit_root = 1;
1510 path->skip_locking = 1;
1512 key.objectid = device->devid;
1513 key.offset = search_start;
1514 key.type = BTRFS_DEV_EXTENT_KEY;
1516 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1520 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1527 slot = path->slots[0];
1528 if (slot >= btrfs_header_nritems(l)) {
1529 ret = btrfs_next_leaf(root, path);
1537 btrfs_item_key_to_cpu(l, &key, slot);
1539 if (key.objectid < device->devid)
1542 if (key.objectid > device->devid)
1545 if (key.type != BTRFS_DEV_EXTENT_KEY)
1548 if (key.offset > search_start) {
1549 hole_size = key.offset - search_start;
1550 dev_extent_hole_check(device, &search_start, &hole_size,
1553 if (hole_size > max_hole_size) {
1554 max_hole_start = search_start;
1555 max_hole_size = hole_size;
1559 * If this free space is greater than which we need,
1560 * it must be the max free space that we have found
1561 * until now, so max_hole_start must point to the start
1562 * of this free space and the length of this free space
1563 * is stored in max_hole_size. Thus, we return
1564 * max_hole_start and max_hole_size and go back to the
1567 if (hole_size >= num_bytes) {
1573 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1574 extent_end = key.offset + btrfs_dev_extent_length(l,
1576 if (extent_end > search_start)
1577 search_start = extent_end;
1584 * At this point, search_start should be the end of
1585 * allocated dev extents, and when shrinking the device,
1586 * search_end may be smaller than search_start.
1588 if (search_end > search_start) {
1589 hole_size = search_end - search_start;
1590 if (dev_extent_hole_check(device, &search_start, &hole_size,
1592 btrfs_release_path(path);
1596 if (hole_size > max_hole_size) {
1597 max_hole_start = search_start;
1598 max_hole_size = hole_size;
1603 if (max_hole_size < num_bytes)
1609 btrfs_free_path(path);
1610 *start = max_hole_start;
1612 *len = max_hole_size;
1616 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
1617 u64 *start, u64 *len)
1619 /* FIXME use last free of some kind */
1620 return find_free_dev_extent_start(device, num_bytes, 0, start, len);
1623 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1624 struct btrfs_device *device,
1625 u64 start, u64 *dev_extent_len)
1627 struct btrfs_fs_info *fs_info = device->fs_info;
1628 struct btrfs_root *root = fs_info->dev_root;
1630 struct btrfs_path *path;
1631 struct btrfs_key key;
1632 struct btrfs_key found_key;
1633 struct extent_buffer *leaf = NULL;
1634 struct btrfs_dev_extent *extent = NULL;
1636 path = btrfs_alloc_path();
1640 key.objectid = device->devid;
1642 key.type = BTRFS_DEV_EXTENT_KEY;
1644 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1646 ret = btrfs_previous_item(root, path, key.objectid,
1647 BTRFS_DEV_EXTENT_KEY);
1650 leaf = path->nodes[0];
1651 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1652 extent = btrfs_item_ptr(leaf, path->slots[0],
1653 struct btrfs_dev_extent);
1654 BUG_ON(found_key.offset > start || found_key.offset +
1655 btrfs_dev_extent_length(leaf, extent) < start);
1657 btrfs_release_path(path);
1659 } else if (ret == 0) {
1660 leaf = path->nodes[0];
1661 extent = btrfs_item_ptr(leaf, path->slots[0],
1662 struct btrfs_dev_extent);
1664 btrfs_handle_fs_error(fs_info, ret, "Slot search failed");
1668 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1670 ret = btrfs_del_item(trans, root, path);
1672 btrfs_handle_fs_error(fs_info, ret,
1673 "Failed to remove dev extent item");
1675 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1678 btrfs_free_path(path);
1682 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1683 struct btrfs_device *device,
1684 u64 chunk_offset, u64 start, u64 num_bytes)
1687 struct btrfs_path *path;
1688 struct btrfs_fs_info *fs_info = device->fs_info;
1689 struct btrfs_root *root = fs_info->dev_root;
1690 struct btrfs_dev_extent *extent;
1691 struct extent_buffer *leaf;
1692 struct btrfs_key key;
1694 WARN_ON(!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state));
1695 WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state));
1696 path = btrfs_alloc_path();
1700 key.objectid = device->devid;
1702 key.type = BTRFS_DEV_EXTENT_KEY;
1703 ret = btrfs_insert_empty_item(trans, root, path, &key,
1708 leaf = path->nodes[0];
1709 extent = btrfs_item_ptr(leaf, path->slots[0],
1710 struct btrfs_dev_extent);
1711 btrfs_set_dev_extent_chunk_tree(leaf, extent,
1712 BTRFS_CHUNK_TREE_OBJECTID);
1713 btrfs_set_dev_extent_chunk_objectid(leaf, extent,
1714 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
1715 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1717 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1718 btrfs_mark_buffer_dirty(leaf);
1720 btrfs_free_path(path);
1724 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1726 struct extent_map_tree *em_tree;
1727 struct extent_map *em;
1731 em_tree = &fs_info->mapping_tree;
1732 read_lock(&em_tree->lock);
1733 n = rb_last(&em_tree->map.rb_root);
1735 em = rb_entry(n, struct extent_map, rb_node);
1736 ret = em->start + em->len;
1738 read_unlock(&em_tree->lock);
1743 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1747 struct btrfs_key key;
1748 struct btrfs_key found_key;
1749 struct btrfs_path *path;
1751 path = btrfs_alloc_path();
1755 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1756 key.type = BTRFS_DEV_ITEM_KEY;
1757 key.offset = (u64)-1;
1759 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1765 btrfs_err(fs_info, "corrupted chunk tree devid -1 matched");
1770 ret = btrfs_previous_item(fs_info->chunk_root, path,
1771 BTRFS_DEV_ITEMS_OBJECTID,
1772 BTRFS_DEV_ITEM_KEY);
1776 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1778 *devid_ret = found_key.offset + 1;
1782 btrfs_free_path(path);
1787 * the device information is stored in the chunk root
1788 * the btrfs_device struct should be fully filled in
1790 static int btrfs_add_dev_item(struct btrfs_trans_handle *trans,
1791 struct btrfs_device *device)
1794 struct btrfs_path *path;
1795 struct btrfs_dev_item *dev_item;
1796 struct extent_buffer *leaf;
1797 struct btrfs_key key;
1800 path = btrfs_alloc_path();
1804 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1805 key.type = BTRFS_DEV_ITEM_KEY;
1806 key.offset = device->devid;
1808 ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path,
1809 &key, sizeof(*dev_item));
1813 leaf = path->nodes[0];
1814 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1816 btrfs_set_device_id(leaf, dev_item, device->devid);
1817 btrfs_set_device_generation(leaf, dev_item, 0);
1818 btrfs_set_device_type(leaf, dev_item, device->type);
1819 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1820 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1821 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1822 btrfs_set_device_total_bytes(leaf, dev_item,
1823 btrfs_device_get_disk_total_bytes(device));
1824 btrfs_set_device_bytes_used(leaf, dev_item,
1825 btrfs_device_get_bytes_used(device));
1826 btrfs_set_device_group(leaf, dev_item, 0);
1827 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1828 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1829 btrfs_set_device_start_offset(leaf, dev_item, 0);
1831 ptr = btrfs_device_uuid(dev_item);
1832 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1833 ptr = btrfs_device_fsid(dev_item);
1834 write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid,
1835 ptr, BTRFS_FSID_SIZE);
1836 btrfs_mark_buffer_dirty(leaf);
1840 btrfs_free_path(path);
1845 * Function to update ctime/mtime for a given device path.
1846 * Mainly used for ctime/mtime based probe like libblkid.
1848 static void update_dev_time(const char *path_name)
1852 filp = filp_open(path_name, O_RDWR, 0);
1855 file_update_time(filp);
1856 filp_close(filp, NULL);
1859 static int btrfs_rm_dev_item(struct btrfs_device *device)
1861 struct btrfs_root *root = device->fs_info->chunk_root;
1863 struct btrfs_path *path;
1864 struct btrfs_key key;
1865 struct btrfs_trans_handle *trans;
1867 path = btrfs_alloc_path();
1871 trans = btrfs_start_transaction(root, 0);
1872 if (IS_ERR(trans)) {
1873 btrfs_free_path(path);
1874 return PTR_ERR(trans);
1876 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1877 key.type = BTRFS_DEV_ITEM_KEY;
1878 key.offset = device->devid;
1880 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1884 btrfs_abort_transaction(trans, ret);
1885 btrfs_end_transaction(trans);
1889 ret = btrfs_del_item(trans, root, path);
1891 btrfs_abort_transaction(trans, ret);
1892 btrfs_end_transaction(trans);
1896 btrfs_free_path(path);
1898 ret = btrfs_commit_transaction(trans);
1903 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1904 * filesystem. It's up to the caller to adjust that number regarding eg. device
1907 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1915 seq = read_seqbegin(&fs_info->profiles_lock);
1917 all_avail = fs_info->avail_data_alloc_bits |
1918 fs_info->avail_system_alloc_bits |
1919 fs_info->avail_metadata_alloc_bits;
1920 } while (read_seqretry(&fs_info->profiles_lock, seq));
1922 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1923 if (!(all_avail & btrfs_raid_array[i].bg_flag))
1926 if (num_devices < btrfs_raid_array[i].devs_min) {
1927 int ret = btrfs_raid_array[i].mindev_error;
1937 static struct btrfs_device * btrfs_find_next_active_device(
1938 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device)
1940 struct btrfs_device *next_device;
1942 list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1943 if (next_device != device &&
1944 !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state)
1945 && next_device->bdev)
1953 * Helper function to check if the given device is part of s_bdev / latest_bdev
1954 * and replace it with the provided or the next active device, in the context
1955 * where this function called, there should be always be another device (or
1956 * this_dev) which is active.
1958 void __cold btrfs_assign_next_active_device(struct btrfs_device *device,
1959 struct btrfs_device *next_device)
1961 struct btrfs_fs_info *fs_info = device->fs_info;
1964 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
1966 ASSERT(next_device);
1968 if (fs_info->sb->s_bdev &&
1969 (fs_info->sb->s_bdev == device->bdev))
1970 fs_info->sb->s_bdev = next_device->bdev;
1972 if (fs_info->fs_devices->latest_bdev == device->bdev)
1973 fs_info->fs_devices->latest_bdev = next_device->bdev;
1977 * Return btrfs_fs_devices::num_devices excluding the device that's being
1978 * currently replaced.
1980 static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info)
1982 u64 num_devices = fs_info->fs_devices->num_devices;
1984 down_read(&fs_info->dev_replace.rwsem);
1985 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
1986 ASSERT(num_devices > 1);
1989 up_read(&fs_info->dev_replace.rwsem);
1994 void btrfs_scratch_superblocks(struct btrfs_fs_info *fs_info,
1995 struct block_device *bdev,
1996 const char *device_path)
1998 struct btrfs_super_block *disk_super;
2004 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; copy_num++) {
2008 disk_super = btrfs_read_dev_one_super(bdev, copy_num);
2009 if (IS_ERR(disk_super))
2012 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
2014 page = virt_to_page(disk_super);
2015 set_page_dirty(page);
2017 /* write_on_page() unlocks the page */
2018 ret = write_one_page(page);
2021 "error clearing superblock number %d (%d)",
2023 btrfs_release_disk_super(disk_super);
2027 /* Notify udev that device has changed */
2028 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
2030 /* Update ctime/mtime for device path for libblkid */
2031 update_dev_time(device_path);
2034 int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path,
2037 struct btrfs_device *device;
2038 struct btrfs_fs_devices *cur_devices;
2039 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2043 mutex_lock(&uuid_mutex);
2045 num_devices = btrfs_num_devices(fs_info);
2047 ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
2051 device = btrfs_find_device_by_devspec(fs_info, devid, device_path);
2053 if (IS_ERR(device)) {
2054 if (PTR_ERR(device) == -ENOENT &&
2055 strcmp(device_path, "missing") == 0)
2056 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2058 ret = PTR_ERR(device);
2062 if (btrfs_pinned_by_swapfile(fs_info, device)) {
2063 btrfs_warn_in_rcu(fs_info,
2064 "cannot remove device %s (devid %llu) due to active swapfile",
2065 rcu_str_deref(device->name), device->devid);
2070 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2071 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
2075 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
2076 fs_info->fs_devices->rw_devices == 1) {
2077 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
2081 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2082 mutex_lock(&fs_info->chunk_mutex);
2083 list_del_init(&device->dev_alloc_list);
2084 device->fs_devices->rw_devices--;
2085 mutex_unlock(&fs_info->chunk_mutex);
2088 mutex_unlock(&uuid_mutex);
2089 ret = btrfs_shrink_device(device, 0);
2091 btrfs_reada_remove_dev(device);
2092 mutex_lock(&uuid_mutex);
2097 * TODO: the superblock still includes this device in its num_devices
2098 * counter although write_all_supers() is not locked out. This
2099 * could give a filesystem state which requires a degraded mount.
2101 ret = btrfs_rm_dev_item(device);
2105 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2106 btrfs_scrub_cancel_dev(device);
2109 * the device list mutex makes sure that we don't change
2110 * the device list while someone else is writing out all
2111 * the device supers. Whoever is writing all supers, should
2112 * lock the device list mutex before getting the number of
2113 * devices in the super block (super_copy). Conversely,
2114 * whoever updates the number of devices in the super block
2115 * (super_copy) should hold the device list mutex.
2119 * In normal cases the cur_devices == fs_devices. But in case
2120 * of deleting a seed device, the cur_devices should point to
2121 * its own fs_devices listed under the fs_devices->seed.
2123 cur_devices = device->fs_devices;
2124 mutex_lock(&fs_devices->device_list_mutex);
2125 list_del_rcu(&device->dev_list);
2127 cur_devices->num_devices--;
2128 cur_devices->total_devices--;
2129 /* Update total_devices of the parent fs_devices if it's seed */
2130 if (cur_devices != fs_devices)
2131 fs_devices->total_devices--;
2133 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
2134 cur_devices->missing_devices--;
2136 btrfs_assign_next_active_device(device, NULL);
2139 cur_devices->open_devices--;
2140 /* remove sysfs entry */
2141 btrfs_sysfs_remove_device(device);
2144 num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
2145 btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
2146 mutex_unlock(&fs_devices->device_list_mutex);
2149 * at this point, the device is zero sized and detached from
2150 * the devices list. All that's left is to zero out the old
2151 * supers and free the device.
2153 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2154 btrfs_scratch_superblocks(fs_info, device->bdev,
2157 btrfs_close_bdev(device);
2159 btrfs_free_device(device);
2161 if (cur_devices->open_devices == 0) {
2162 list_del_init(&cur_devices->seed_list);
2163 close_fs_devices(cur_devices);
2164 free_fs_devices(cur_devices);
2168 mutex_unlock(&uuid_mutex);
2172 btrfs_reada_undo_remove_dev(device);
2173 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2174 mutex_lock(&fs_info->chunk_mutex);
2175 list_add(&device->dev_alloc_list,
2176 &fs_devices->alloc_list);
2177 device->fs_devices->rw_devices++;
2178 mutex_unlock(&fs_info->chunk_mutex);
2183 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev)
2185 struct btrfs_fs_devices *fs_devices;
2187 lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex);
2190 * in case of fs with no seed, srcdev->fs_devices will point
2191 * to fs_devices of fs_info. However when the dev being replaced is
2192 * a seed dev it will point to the seed's local fs_devices. In short
2193 * srcdev will have its correct fs_devices in both the cases.
2195 fs_devices = srcdev->fs_devices;
2197 list_del_rcu(&srcdev->dev_list);
2198 list_del(&srcdev->dev_alloc_list);
2199 fs_devices->num_devices--;
2200 if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state))
2201 fs_devices->missing_devices--;
2203 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state))
2204 fs_devices->rw_devices--;
2207 fs_devices->open_devices--;
2210 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev)
2212 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2214 mutex_lock(&uuid_mutex);
2216 btrfs_close_bdev(srcdev);
2218 btrfs_free_device(srcdev);
2220 /* if this is no devs we rather delete the fs_devices */
2221 if (!fs_devices->num_devices) {
2223 * On a mounted FS, num_devices can't be zero unless it's a
2224 * seed. In case of a seed device being replaced, the replace
2225 * target added to the sprout FS, so there will be no more
2226 * device left under the seed FS.
2228 ASSERT(fs_devices->seeding);
2230 list_del_init(&fs_devices->seed_list);
2231 close_fs_devices(fs_devices);
2232 free_fs_devices(fs_devices);
2234 mutex_unlock(&uuid_mutex);
2237 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev)
2239 struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices;
2241 mutex_lock(&fs_devices->device_list_mutex);
2243 btrfs_sysfs_remove_device(tgtdev);
2246 fs_devices->open_devices--;
2248 fs_devices->num_devices--;
2250 btrfs_assign_next_active_device(tgtdev, NULL);
2252 list_del_rcu(&tgtdev->dev_list);
2254 mutex_unlock(&fs_devices->device_list_mutex);
2257 * The update_dev_time() with in btrfs_scratch_superblocks()
2258 * may lead to a call to btrfs_show_devname() which will try
2259 * to hold device_list_mutex. And here this device
2260 * is already out of device list, so we don't have to hold
2261 * the device_list_mutex lock.
2263 btrfs_scratch_superblocks(tgtdev->fs_info, tgtdev->bdev,
2266 btrfs_close_bdev(tgtdev);
2268 btrfs_free_device(tgtdev);
2271 static struct btrfs_device *btrfs_find_device_by_path(
2272 struct btrfs_fs_info *fs_info, const char *device_path)
2275 struct btrfs_super_block *disk_super;
2278 struct block_device *bdev;
2279 struct btrfs_device *device;
2281 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2282 fs_info->bdev_holder, 0, &bdev, &disk_super);
2284 return ERR_PTR(ret);
2286 devid = btrfs_stack_device_id(&disk_super->dev_item);
2287 dev_uuid = disk_super->dev_item.uuid;
2288 if (btrfs_fs_incompat(fs_info, METADATA_UUID))
2289 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
2290 disk_super->metadata_uuid, true);
2292 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
2293 disk_super->fsid, true);
2295 btrfs_release_disk_super(disk_super);
2297 device = ERR_PTR(-ENOENT);
2298 blkdev_put(bdev, FMODE_READ);
2303 * Lookup a device given by device id, or the path if the id is 0.
2305 struct btrfs_device *btrfs_find_device_by_devspec(
2306 struct btrfs_fs_info *fs_info, u64 devid,
2307 const char *device_path)
2309 struct btrfs_device *device;
2312 device = btrfs_find_device(fs_info->fs_devices, devid, NULL,
2315 return ERR_PTR(-ENOENT);
2319 if (!device_path || !device_path[0])
2320 return ERR_PTR(-EINVAL);
2322 if (strcmp(device_path, "missing") == 0) {
2323 /* Find first missing device */
2324 list_for_each_entry(device, &fs_info->fs_devices->devices,
2326 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
2327 &device->dev_state) && !device->bdev)
2330 return ERR_PTR(-ENOENT);
2333 return btrfs_find_device_by_path(fs_info, device_path);
2337 * does all the dirty work required for changing file system's UUID.
2339 static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info)
2341 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2342 struct btrfs_fs_devices *old_devices;
2343 struct btrfs_fs_devices *seed_devices;
2344 struct btrfs_super_block *disk_super = fs_info->super_copy;
2345 struct btrfs_device *device;
2348 lockdep_assert_held(&uuid_mutex);
2349 if (!fs_devices->seeding)
2353 * Private copy of the seed devices, anchored at
2354 * fs_info->fs_devices->seed_list
2356 seed_devices = alloc_fs_devices(NULL, NULL);
2357 if (IS_ERR(seed_devices))
2358 return PTR_ERR(seed_devices);
2361 * It's necessary to retain a copy of the original seed fs_devices in
2362 * fs_uuids so that filesystems which have been seeded can successfully
2363 * reference the seed device from open_seed_devices. This also supports
2366 old_devices = clone_fs_devices(fs_devices);
2367 if (IS_ERR(old_devices)) {
2368 kfree(seed_devices);
2369 return PTR_ERR(old_devices);
2372 list_add(&old_devices->fs_list, &fs_uuids);
2374 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2375 seed_devices->opened = 1;
2376 INIT_LIST_HEAD(&seed_devices->devices);
2377 INIT_LIST_HEAD(&seed_devices->alloc_list);
2378 mutex_init(&seed_devices->device_list_mutex);
2380 mutex_lock(&fs_devices->device_list_mutex);
2381 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2383 list_for_each_entry(device, &seed_devices->devices, dev_list)
2384 device->fs_devices = seed_devices;
2386 fs_devices->seeding = false;
2387 fs_devices->num_devices = 0;
2388 fs_devices->open_devices = 0;
2389 fs_devices->missing_devices = 0;
2390 fs_devices->rotating = false;
2391 list_add(&seed_devices->seed_list, &fs_devices->seed_list);
2393 generate_random_uuid(fs_devices->fsid);
2394 memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE);
2395 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2396 mutex_unlock(&fs_devices->device_list_mutex);
2398 super_flags = btrfs_super_flags(disk_super) &
2399 ~BTRFS_SUPER_FLAG_SEEDING;
2400 btrfs_set_super_flags(disk_super, super_flags);
2406 * Store the expected generation for seed devices in device items.
2408 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans)
2410 struct btrfs_fs_info *fs_info = trans->fs_info;
2411 struct btrfs_root *root = fs_info->chunk_root;
2412 struct btrfs_path *path;
2413 struct extent_buffer *leaf;
2414 struct btrfs_dev_item *dev_item;
2415 struct btrfs_device *device;
2416 struct btrfs_key key;
2417 u8 fs_uuid[BTRFS_FSID_SIZE];
2418 u8 dev_uuid[BTRFS_UUID_SIZE];
2422 path = btrfs_alloc_path();
2426 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2428 key.type = BTRFS_DEV_ITEM_KEY;
2431 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2435 leaf = path->nodes[0];
2437 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2438 ret = btrfs_next_leaf(root, path);
2443 leaf = path->nodes[0];
2444 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2445 btrfs_release_path(path);
2449 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2450 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2451 key.type != BTRFS_DEV_ITEM_KEY)
2454 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2455 struct btrfs_dev_item);
2456 devid = btrfs_device_id(leaf, dev_item);
2457 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2459 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2461 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
2463 BUG_ON(!device); /* Logic error */
2465 if (device->fs_devices->seeding) {
2466 btrfs_set_device_generation(leaf, dev_item,
2467 device->generation);
2468 btrfs_mark_buffer_dirty(leaf);
2476 btrfs_free_path(path);
2480 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2482 struct btrfs_root *root = fs_info->dev_root;
2483 struct request_queue *q;
2484 struct btrfs_trans_handle *trans;
2485 struct btrfs_device *device;
2486 struct block_device *bdev;
2487 struct super_block *sb = fs_info->sb;
2488 struct rcu_string *name;
2489 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2490 u64 orig_super_total_bytes;
2491 u64 orig_super_num_devices;
2492 int seeding_dev = 0;
2494 bool locked = false;
2496 if (sb_rdonly(sb) && !fs_devices->seeding)
2499 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2500 fs_info->bdev_holder);
2502 return PTR_ERR(bdev);
2504 if (fs_devices->seeding) {
2506 down_write(&sb->s_umount);
2507 mutex_lock(&uuid_mutex);
2511 sync_blockdev(bdev);
2514 list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
2515 if (device->bdev == bdev) {
2523 device = btrfs_alloc_device(fs_info, NULL, NULL);
2524 if (IS_ERR(device)) {
2525 /* we can safely leave the fs_devices entry around */
2526 ret = PTR_ERR(device);
2530 name = rcu_string_strdup(device_path, GFP_KERNEL);
2533 goto error_free_device;
2535 rcu_assign_pointer(device->name, name);
2537 trans = btrfs_start_transaction(root, 0);
2538 if (IS_ERR(trans)) {
2539 ret = PTR_ERR(trans);
2540 goto error_free_device;
2543 q = bdev_get_queue(bdev);
2544 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
2545 device->generation = trans->transid;
2546 device->io_width = fs_info->sectorsize;
2547 device->io_align = fs_info->sectorsize;
2548 device->sector_size = fs_info->sectorsize;
2549 device->total_bytes = round_down(i_size_read(bdev->bd_inode),
2550 fs_info->sectorsize);
2551 device->disk_total_bytes = device->total_bytes;
2552 device->commit_total_bytes = device->total_bytes;
2553 device->fs_info = fs_info;
2554 device->bdev = bdev;
2555 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2556 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
2557 device->mode = FMODE_EXCL;
2558 device->dev_stats_valid = 1;
2559 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2562 sb->s_flags &= ~SB_RDONLY;
2563 ret = btrfs_prepare_sprout(fs_info);
2565 btrfs_abort_transaction(trans, ret);
2570 device->fs_devices = fs_devices;
2572 mutex_lock(&fs_devices->device_list_mutex);
2573 mutex_lock(&fs_info->chunk_mutex);
2574 list_add_rcu(&device->dev_list, &fs_devices->devices);
2575 list_add(&device->dev_alloc_list, &fs_devices->alloc_list);
2576 fs_devices->num_devices++;
2577 fs_devices->open_devices++;
2578 fs_devices->rw_devices++;
2579 fs_devices->total_devices++;
2580 fs_devices->total_rw_bytes += device->total_bytes;
2582 atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2584 if (!blk_queue_nonrot(q))
2585 fs_devices->rotating = true;
2587 orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
2588 btrfs_set_super_total_bytes(fs_info->super_copy,
2589 round_down(orig_super_total_bytes + device->total_bytes,
2590 fs_info->sectorsize));
2592 orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy);
2593 btrfs_set_super_num_devices(fs_info->super_copy,
2594 orig_super_num_devices + 1);
2597 * we've got more storage, clear any full flags on the space
2600 btrfs_clear_space_info_full(fs_info);
2602 mutex_unlock(&fs_info->chunk_mutex);
2604 /* Add sysfs device entry */
2605 btrfs_sysfs_add_device(device);
2607 mutex_unlock(&fs_devices->device_list_mutex);
2610 mutex_lock(&fs_info->chunk_mutex);
2611 ret = init_first_rw_device(trans);
2612 mutex_unlock(&fs_info->chunk_mutex);
2614 btrfs_abort_transaction(trans, ret);
2619 ret = btrfs_add_dev_item(trans, device);
2621 btrfs_abort_transaction(trans, ret);
2626 ret = btrfs_finish_sprout(trans);
2628 btrfs_abort_transaction(trans, ret);
2633 * fs_devices now represents the newly sprouted filesystem and
2634 * its fsid has been changed by btrfs_prepare_sprout
2636 btrfs_sysfs_update_sprout_fsid(fs_devices);
2639 ret = btrfs_commit_transaction(trans);
2642 mutex_unlock(&uuid_mutex);
2643 up_write(&sb->s_umount);
2646 if (ret) /* transaction commit */
2649 ret = btrfs_relocate_sys_chunks(fs_info);
2651 btrfs_handle_fs_error(fs_info, ret,
2652 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2653 trans = btrfs_attach_transaction(root);
2654 if (IS_ERR(trans)) {
2655 if (PTR_ERR(trans) == -ENOENT)
2657 ret = PTR_ERR(trans);
2661 ret = btrfs_commit_transaction(trans);
2665 * Now that we have written a new super block to this device, check all
2666 * other fs_devices list if device_path alienates any other scanned
2668 * We can ignore the return value as it typically returns -EINVAL and
2669 * only succeeds if the device was an alien.
2671 btrfs_forget_devices(device_path);
2673 /* Update ctime/mtime for blkid or udev */
2674 update_dev_time(device_path);
2679 btrfs_sysfs_remove_device(device);
2680 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2681 mutex_lock(&fs_info->chunk_mutex);
2682 list_del_rcu(&device->dev_list);
2683 list_del(&device->dev_alloc_list);
2684 fs_info->fs_devices->num_devices--;
2685 fs_info->fs_devices->open_devices--;
2686 fs_info->fs_devices->rw_devices--;
2687 fs_info->fs_devices->total_devices--;
2688 fs_info->fs_devices->total_rw_bytes -= device->total_bytes;
2689 atomic64_sub(device->total_bytes, &fs_info->free_chunk_space);
2690 btrfs_set_super_total_bytes(fs_info->super_copy,
2691 orig_super_total_bytes);
2692 btrfs_set_super_num_devices(fs_info->super_copy,
2693 orig_super_num_devices);
2694 mutex_unlock(&fs_info->chunk_mutex);
2695 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2698 sb->s_flags |= SB_RDONLY;
2700 btrfs_end_transaction(trans);
2702 btrfs_free_device(device);
2704 blkdev_put(bdev, FMODE_EXCL);
2706 mutex_unlock(&uuid_mutex);
2707 up_write(&sb->s_umount);
2712 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2713 struct btrfs_device *device)
2716 struct btrfs_path *path;
2717 struct btrfs_root *root = device->fs_info->chunk_root;
2718 struct btrfs_dev_item *dev_item;
2719 struct extent_buffer *leaf;
2720 struct btrfs_key key;
2722 path = btrfs_alloc_path();
2726 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2727 key.type = BTRFS_DEV_ITEM_KEY;
2728 key.offset = device->devid;
2730 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2739 leaf = path->nodes[0];
2740 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2742 btrfs_set_device_id(leaf, dev_item, device->devid);
2743 btrfs_set_device_type(leaf, dev_item, device->type);
2744 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2745 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2746 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2747 btrfs_set_device_total_bytes(leaf, dev_item,
2748 btrfs_device_get_disk_total_bytes(device));
2749 btrfs_set_device_bytes_used(leaf, dev_item,
2750 btrfs_device_get_bytes_used(device));
2751 btrfs_mark_buffer_dirty(leaf);
2754 btrfs_free_path(path);
2758 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2759 struct btrfs_device *device, u64 new_size)
2761 struct btrfs_fs_info *fs_info = device->fs_info;
2762 struct btrfs_super_block *super_copy = fs_info->super_copy;
2766 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2769 new_size = round_down(new_size, fs_info->sectorsize);
2771 mutex_lock(&fs_info->chunk_mutex);
2772 old_total = btrfs_super_total_bytes(super_copy);
2773 diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2775 if (new_size <= device->total_bytes ||
2776 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2777 mutex_unlock(&fs_info->chunk_mutex);
2781 btrfs_set_super_total_bytes(super_copy,
2782 round_down(old_total + diff, fs_info->sectorsize));
2783 device->fs_devices->total_rw_bytes += diff;
2785 btrfs_device_set_total_bytes(device, new_size);
2786 btrfs_device_set_disk_total_bytes(device, new_size);
2787 btrfs_clear_space_info_full(device->fs_info);
2788 if (list_empty(&device->post_commit_list))
2789 list_add_tail(&device->post_commit_list,
2790 &trans->transaction->dev_update_list);
2791 mutex_unlock(&fs_info->chunk_mutex);
2793 return btrfs_update_device(trans, device);
2796 static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
2798 struct btrfs_fs_info *fs_info = trans->fs_info;
2799 struct btrfs_root *root = fs_info->chunk_root;
2801 struct btrfs_path *path;
2802 struct btrfs_key key;
2804 path = btrfs_alloc_path();
2808 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2809 key.offset = chunk_offset;
2810 key.type = BTRFS_CHUNK_ITEM_KEY;
2812 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2815 else if (ret > 0) { /* Logic error or corruption */
2816 btrfs_handle_fs_error(fs_info, -ENOENT,
2817 "Failed lookup while freeing chunk.");
2822 ret = btrfs_del_item(trans, root, path);
2824 btrfs_handle_fs_error(fs_info, ret,
2825 "Failed to delete chunk item.");
2827 btrfs_free_path(path);
2831 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2833 struct btrfs_super_block *super_copy = fs_info->super_copy;
2834 struct btrfs_disk_key *disk_key;
2835 struct btrfs_chunk *chunk;
2842 struct btrfs_key key;
2844 mutex_lock(&fs_info->chunk_mutex);
2845 array_size = btrfs_super_sys_array_size(super_copy);
2847 ptr = super_copy->sys_chunk_array;
2850 while (cur < array_size) {
2851 disk_key = (struct btrfs_disk_key *)ptr;
2852 btrfs_disk_key_to_cpu(&key, disk_key);
2854 len = sizeof(*disk_key);
2856 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2857 chunk = (struct btrfs_chunk *)(ptr + len);
2858 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2859 len += btrfs_chunk_item_size(num_stripes);
2864 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID &&
2865 key.offset == chunk_offset) {
2866 memmove(ptr, ptr + len, array_size - (cur + len));
2868 btrfs_set_super_sys_array_size(super_copy, array_size);
2874 mutex_unlock(&fs_info->chunk_mutex);
2879 * btrfs_get_chunk_map() - Find the mapping containing the given logical extent.
2880 * @logical: Logical block offset in bytes.
2881 * @length: Length of extent in bytes.
2883 * Return: Chunk mapping or ERR_PTR.
2885 struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info,
2886 u64 logical, u64 length)
2888 struct extent_map_tree *em_tree;
2889 struct extent_map *em;
2891 em_tree = &fs_info->mapping_tree;
2892 read_lock(&em_tree->lock);
2893 em = lookup_extent_mapping(em_tree, logical, length);
2894 read_unlock(&em_tree->lock);
2897 btrfs_crit(fs_info, "unable to find logical %llu length %llu",
2899 return ERR_PTR(-EINVAL);
2902 if (em->start > logical || em->start + em->len < logical) {
2904 "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
2905 logical, length, em->start, em->start + em->len);
2906 free_extent_map(em);
2907 return ERR_PTR(-EINVAL);
2910 /* callers are responsible for dropping em's ref. */
2914 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
2916 struct btrfs_fs_info *fs_info = trans->fs_info;
2917 struct extent_map *em;
2918 struct map_lookup *map;
2919 u64 dev_extent_len = 0;
2921 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2923 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
2926 * This is a logic error, but we don't want to just rely on the
2927 * user having built with ASSERT enabled, so if ASSERT doesn't
2928 * do anything we still error out.
2933 map = em->map_lookup;
2934 mutex_lock(&fs_info->chunk_mutex);
2935 check_system_chunk(trans, map->type);
2936 mutex_unlock(&fs_info->chunk_mutex);
2939 * Take the device list mutex to prevent races with the final phase of
2940 * a device replace operation that replaces the device object associated
2941 * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2943 mutex_lock(&fs_devices->device_list_mutex);
2944 for (i = 0; i < map->num_stripes; i++) {
2945 struct btrfs_device *device = map->stripes[i].dev;
2946 ret = btrfs_free_dev_extent(trans, device,
2947 map->stripes[i].physical,
2950 mutex_unlock(&fs_devices->device_list_mutex);
2951 btrfs_abort_transaction(trans, ret);
2955 if (device->bytes_used > 0) {
2956 mutex_lock(&fs_info->chunk_mutex);
2957 btrfs_device_set_bytes_used(device,
2958 device->bytes_used - dev_extent_len);
2959 atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
2960 btrfs_clear_space_info_full(fs_info);
2961 mutex_unlock(&fs_info->chunk_mutex);
2964 ret = btrfs_update_device(trans, device);
2966 mutex_unlock(&fs_devices->device_list_mutex);
2967 btrfs_abort_transaction(trans, ret);
2971 mutex_unlock(&fs_devices->device_list_mutex);
2973 ret = btrfs_free_chunk(trans, chunk_offset);
2975 btrfs_abort_transaction(trans, ret);
2979 trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
2981 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2982 ret = btrfs_del_sys_chunk(fs_info, chunk_offset);
2984 btrfs_abort_transaction(trans, ret);
2989 ret = btrfs_remove_block_group(trans, chunk_offset, em);
2991 btrfs_abort_transaction(trans, ret);
2997 free_extent_map(em);
3001 static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
3003 struct btrfs_root *root = fs_info->chunk_root;
3004 struct btrfs_trans_handle *trans;
3005 struct btrfs_block_group *block_group;
3009 * Prevent races with automatic removal of unused block groups.
3010 * After we relocate and before we remove the chunk with offset
3011 * chunk_offset, automatic removal of the block group can kick in,
3012 * resulting in a failure when calling btrfs_remove_chunk() below.
3014 * Make sure to acquire this mutex before doing a tree search (dev
3015 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
3016 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
3017 * we release the path used to search the chunk/dev tree and before
3018 * the current task acquires this mutex and calls us.
3020 lockdep_assert_held(&fs_info->delete_unused_bgs_mutex);
3022 /* step one, relocate all the extents inside this chunk */
3023 btrfs_scrub_pause(fs_info);
3024 ret = btrfs_relocate_block_group(fs_info, chunk_offset);
3025 btrfs_scrub_continue(fs_info);
3029 block_group = btrfs_lookup_block_group(fs_info, chunk_offset);
3032 btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group);
3033 btrfs_put_block_group(block_group);
3035 trans = btrfs_start_trans_remove_block_group(root->fs_info,
3037 if (IS_ERR(trans)) {
3038 ret = PTR_ERR(trans);
3039 btrfs_handle_fs_error(root->fs_info, ret, NULL);
3044 * step two, delete the device extents and the
3045 * chunk tree entries
3047 ret = btrfs_remove_chunk(trans, chunk_offset);
3048 btrfs_end_transaction(trans);
3052 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
3054 struct btrfs_root *chunk_root = fs_info->chunk_root;
3055 struct btrfs_path *path;
3056 struct extent_buffer *leaf;
3057 struct btrfs_chunk *chunk;
3058 struct btrfs_key key;
3059 struct btrfs_key found_key;
3061 bool retried = false;
3065 path = btrfs_alloc_path();
3070 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3071 key.offset = (u64)-1;
3072 key.type = BTRFS_CHUNK_ITEM_KEY;
3075 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3076 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3078 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3081 BUG_ON(ret == 0); /* Corruption */
3083 ret = btrfs_previous_item(chunk_root, path, key.objectid,
3086 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3092 leaf = path->nodes[0];
3093 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3095 chunk = btrfs_item_ptr(leaf, path->slots[0],
3096 struct btrfs_chunk);
3097 chunk_type = btrfs_chunk_type(leaf, chunk);
3098 btrfs_release_path(path);
3100 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3101 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3107 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3109 if (found_key.offset == 0)
3111 key.offset = found_key.offset - 1;
3114 if (failed && !retried) {
3118 } else if (WARN_ON(failed && retried)) {
3122 btrfs_free_path(path);
3127 * return 1 : allocate a data chunk successfully,
3128 * return <0: errors during allocating a data chunk,
3129 * return 0 : no need to allocate a data chunk.
3131 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info,
3134 struct btrfs_block_group *cache;
3138 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3140 chunk_type = cache->flags;
3141 btrfs_put_block_group(cache);
3143 if (!(chunk_type & BTRFS_BLOCK_GROUP_DATA))
3146 spin_lock(&fs_info->data_sinfo->lock);
3147 bytes_used = fs_info->data_sinfo->bytes_used;
3148 spin_unlock(&fs_info->data_sinfo->lock);
3151 struct btrfs_trans_handle *trans;
3154 trans = btrfs_join_transaction(fs_info->tree_root);
3156 return PTR_ERR(trans);
3158 ret = btrfs_force_chunk_alloc(trans, BTRFS_BLOCK_GROUP_DATA);
3159 btrfs_end_transaction(trans);
3168 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3169 struct btrfs_balance_control *bctl)
3171 struct btrfs_root *root = fs_info->tree_root;
3172 struct btrfs_trans_handle *trans;
3173 struct btrfs_balance_item *item;
3174 struct btrfs_disk_balance_args disk_bargs;
3175 struct btrfs_path *path;
3176 struct extent_buffer *leaf;
3177 struct btrfs_key key;
3180 path = btrfs_alloc_path();
3184 trans = btrfs_start_transaction(root, 0);
3185 if (IS_ERR(trans)) {
3186 btrfs_free_path(path);
3187 return PTR_ERR(trans);
3190 key.objectid = BTRFS_BALANCE_OBJECTID;
3191 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3194 ret = btrfs_insert_empty_item(trans, root, path, &key,
3199 leaf = path->nodes[0];
3200 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3202 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3204 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3205 btrfs_set_balance_data(leaf, item, &disk_bargs);
3206 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3207 btrfs_set_balance_meta(leaf, item, &disk_bargs);
3208 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3209 btrfs_set_balance_sys(leaf, item, &disk_bargs);
3211 btrfs_set_balance_flags(leaf, item, bctl->flags);
3213 btrfs_mark_buffer_dirty(leaf);
3215 btrfs_free_path(path);
3216 err = btrfs_commit_transaction(trans);
3222 static int del_balance_item(struct btrfs_fs_info *fs_info)
3224 struct btrfs_root *root = fs_info->tree_root;
3225 struct btrfs_trans_handle *trans;
3226 struct btrfs_path *path;
3227 struct btrfs_key key;
3230 path = btrfs_alloc_path();
3234 trans = btrfs_start_transaction_fallback_global_rsv(root, 0);
3235 if (IS_ERR(trans)) {
3236 btrfs_free_path(path);
3237 return PTR_ERR(trans);
3240 key.objectid = BTRFS_BALANCE_OBJECTID;
3241 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3244 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3252 ret = btrfs_del_item(trans, root, path);
3254 btrfs_free_path(path);
3255 err = btrfs_commit_transaction(trans);
3262 * This is a heuristic used to reduce the number of chunks balanced on
3263 * resume after balance was interrupted.
3265 static void update_balance_args(struct btrfs_balance_control *bctl)
3268 * Turn on soft mode for chunk types that were being converted.
3270 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3271 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3272 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3273 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3274 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3275 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3278 * Turn on usage filter if is not already used. The idea is
3279 * that chunks that we have already balanced should be
3280 * reasonably full. Don't do it for chunks that are being
3281 * converted - that will keep us from relocating unconverted
3282 * (albeit full) chunks.
3284 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3285 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3286 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3287 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3288 bctl->data.usage = 90;
3290 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3291 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3292 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3293 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3294 bctl->sys.usage = 90;
3296 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3297 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3298 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3299 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3300 bctl->meta.usage = 90;
3305 * Clear the balance status in fs_info and delete the balance item from disk.
3307 static void reset_balance_state(struct btrfs_fs_info *fs_info)
3309 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3312 BUG_ON(!fs_info->balance_ctl);
3314 spin_lock(&fs_info->balance_lock);
3315 fs_info->balance_ctl = NULL;
3316 spin_unlock(&fs_info->balance_lock);
3319 ret = del_balance_item(fs_info);
3321 btrfs_handle_fs_error(fs_info, ret, NULL);
3325 * Balance filters. Return 1 if chunk should be filtered out
3326 * (should not be balanced).
3328 static int chunk_profiles_filter(u64 chunk_type,
3329 struct btrfs_balance_args *bargs)
3331 chunk_type = chunk_to_extended(chunk_type) &
3332 BTRFS_EXTENDED_PROFILE_MASK;
3334 if (bargs->profiles & chunk_type)
3340 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3341 struct btrfs_balance_args *bargs)
3343 struct btrfs_block_group *cache;
3345 u64 user_thresh_min;
3346 u64 user_thresh_max;
3349 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3350 chunk_used = cache->used;
3352 if (bargs->usage_min == 0)
3353 user_thresh_min = 0;
3355 user_thresh_min = div_factor_fine(cache->length,
3358 if (bargs->usage_max == 0)
3359 user_thresh_max = 1;
3360 else if (bargs->usage_max > 100)
3361 user_thresh_max = cache->length;
3363 user_thresh_max = div_factor_fine(cache->length,
3366 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3369 btrfs_put_block_group(cache);
3373 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3374 u64 chunk_offset, struct btrfs_balance_args *bargs)
3376 struct btrfs_block_group *cache;
3377 u64 chunk_used, user_thresh;
3380 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3381 chunk_used = cache->used;
3383 if (bargs->usage_min == 0)
3385 else if (bargs->usage > 100)
3386 user_thresh = cache->length;
3388 user_thresh = div_factor_fine(cache->length, bargs->usage);
3390 if (chunk_used < user_thresh)
3393 btrfs_put_block_group(cache);
3397 static int chunk_devid_filter(struct extent_buffer *leaf,
3398 struct btrfs_chunk *chunk,
3399 struct btrfs_balance_args *bargs)
3401 struct btrfs_stripe *stripe;
3402 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3405 for (i = 0; i < num_stripes; i++) {
3406 stripe = btrfs_stripe_nr(chunk, i);
3407 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3414 static u64 calc_data_stripes(u64 type, int num_stripes)
3416 const int index = btrfs_bg_flags_to_raid_index(type);
3417 const int ncopies = btrfs_raid_array[index].ncopies;
3418 const int nparity = btrfs_raid_array[index].nparity;
3421 return num_stripes - nparity;
3423 return num_stripes / ncopies;
3426 /* [pstart, pend) */
3427 static int chunk_drange_filter(struct extent_buffer *leaf,
3428 struct btrfs_chunk *chunk,
3429 struct btrfs_balance_args *bargs)
3431 struct btrfs_stripe *stripe;
3432 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3439 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3442 type = btrfs_chunk_type(leaf, chunk);
3443 factor = calc_data_stripes(type, num_stripes);
3445 for (i = 0; i < num_stripes; i++) {
3446 stripe = btrfs_stripe_nr(chunk, i);
3447 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3450 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3451 stripe_length = btrfs_chunk_length(leaf, chunk);
3452 stripe_length = div_u64(stripe_length, factor);
3454 if (stripe_offset < bargs->pend &&
3455 stripe_offset + stripe_length > bargs->pstart)
3462 /* [vstart, vend) */
3463 static int chunk_vrange_filter(struct extent_buffer *leaf,
3464 struct btrfs_chunk *chunk,
3466 struct btrfs_balance_args *bargs)
3468 if (chunk_offset < bargs->vend &&
3469 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3470 /* at least part of the chunk is inside this vrange */
3476 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3477 struct btrfs_chunk *chunk,
3478 struct btrfs_balance_args *bargs)
3480 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3482 if (bargs->stripes_min <= num_stripes
3483 && num_stripes <= bargs->stripes_max)
3489 static int chunk_soft_convert_filter(u64 chunk_type,
3490 struct btrfs_balance_args *bargs)
3492 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3495 chunk_type = chunk_to_extended(chunk_type) &
3496 BTRFS_EXTENDED_PROFILE_MASK;
3498 if (bargs->target == chunk_type)
3504 static int should_balance_chunk(struct extent_buffer *leaf,
3505 struct btrfs_chunk *chunk, u64 chunk_offset)
3507 struct btrfs_fs_info *fs_info = leaf->fs_info;
3508 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3509 struct btrfs_balance_args *bargs = NULL;
3510 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3513 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3514 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3518 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3519 bargs = &bctl->data;
3520 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3522 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3523 bargs = &bctl->meta;
3525 /* profiles filter */
3526 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3527 chunk_profiles_filter(chunk_type, bargs)) {
3532 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3533 chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3535 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3536 chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3541 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3542 chunk_devid_filter(leaf, chunk, bargs)) {
3546 /* drange filter, makes sense only with devid filter */
3547 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3548 chunk_drange_filter(leaf, chunk, bargs)) {
3553 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3554 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3558 /* stripes filter */
3559 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3560 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3564 /* soft profile changing mode */
3565 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3566 chunk_soft_convert_filter(chunk_type, bargs)) {
3571 * limited by count, must be the last filter
3573 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3574 if (bargs->limit == 0)
3578 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3580 * Same logic as the 'limit' filter; the minimum cannot be
3581 * determined here because we do not have the global information
3582 * about the count of all chunks that satisfy the filters.
3584 if (bargs->limit_max == 0)
3593 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3595 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3596 struct btrfs_root *chunk_root = fs_info->chunk_root;
3598 struct btrfs_chunk *chunk;
3599 struct btrfs_path *path = NULL;
3600 struct btrfs_key key;
3601 struct btrfs_key found_key;
3602 struct extent_buffer *leaf;
3605 int enospc_errors = 0;
3606 bool counting = true;
3607 /* The single value limit and min/max limits use the same bytes in the */
3608 u64 limit_data = bctl->data.limit;
3609 u64 limit_meta = bctl->meta.limit;
3610 u64 limit_sys = bctl->sys.limit;
3614 int chunk_reserved = 0;
3616 path = btrfs_alloc_path();
3622 /* zero out stat counters */
3623 spin_lock(&fs_info->balance_lock);
3624 memset(&bctl->stat, 0, sizeof(bctl->stat));
3625 spin_unlock(&fs_info->balance_lock);
3629 * The single value limit and min/max limits use the same bytes
3632 bctl->data.limit = limit_data;
3633 bctl->meta.limit = limit_meta;
3634 bctl->sys.limit = limit_sys;
3636 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3637 key.offset = (u64)-1;
3638 key.type = BTRFS_CHUNK_ITEM_KEY;
3641 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3642 atomic_read(&fs_info->balance_cancel_req)) {
3647 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3648 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3650 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3655 * this shouldn't happen, it means the last relocate
3659 BUG(); /* FIXME break ? */
3661 ret = btrfs_previous_item(chunk_root, path, 0,
3662 BTRFS_CHUNK_ITEM_KEY);
3664 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3669 leaf = path->nodes[0];
3670 slot = path->slots[0];
3671 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3673 if (found_key.objectid != key.objectid) {
3674 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3678 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3679 chunk_type = btrfs_chunk_type(leaf, chunk);
3682 spin_lock(&fs_info->balance_lock);
3683 bctl->stat.considered++;
3684 spin_unlock(&fs_info->balance_lock);
3687 ret = should_balance_chunk(leaf, chunk, found_key.offset);
3689 btrfs_release_path(path);
3691 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3696 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3697 spin_lock(&fs_info->balance_lock);
3698 bctl->stat.expected++;
3699 spin_unlock(&fs_info->balance_lock);
3701 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3703 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3705 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3712 * Apply limit_min filter, no need to check if the LIMITS
3713 * filter is used, limit_min is 0 by default
3715 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3716 count_data < bctl->data.limit_min)
3717 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3718 count_meta < bctl->meta.limit_min)
3719 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3720 count_sys < bctl->sys.limit_min)) {
3721 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3725 if (!chunk_reserved) {
3727 * We may be relocating the only data chunk we have,
3728 * which could potentially end up with losing data's
3729 * raid profile, so lets allocate an empty one in
3732 ret = btrfs_may_alloc_data_chunk(fs_info,
3735 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3737 } else if (ret == 1) {
3742 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3743 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3744 if (ret == -ENOSPC) {
3746 } else if (ret == -ETXTBSY) {
3748 "skipping relocation of block group %llu due to active swapfile",
3754 spin_lock(&fs_info->balance_lock);
3755 bctl->stat.completed++;
3756 spin_unlock(&fs_info->balance_lock);
3759 if (found_key.offset == 0)
3761 key.offset = found_key.offset - 1;
3765 btrfs_release_path(path);
3770 btrfs_free_path(path);
3771 if (enospc_errors) {
3772 btrfs_info(fs_info, "%d enospc errors during balance",
3782 * alloc_profile_is_valid - see if a given profile is valid and reduced
3783 * @flags: profile to validate
3784 * @extended: if true @flags is treated as an extended profile
3786 static int alloc_profile_is_valid(u64 flags, int extended)
3788 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3789 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3791 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3793 /* 1) check that all other bits are zeroed */
3797 /* 2) see if profile is reduced */
3799 return !extended; /* "0" is valid for usual profiles */
3801 return has_single_bit_set(flags);
3804 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3806 /* cancel requested || normal exit path */
3807 return atomic_read(&fs_info->balance_cancel_req) ||
3808 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3809 atomic_read(&fs_info->balance_cancel_req) == 0);
3813 * Validate target profile against allowed profiles and return true if it's OK.
3814 * Otherwise print the error message and return false.
3816 static inline int validate_convert_profile(struct btrfs_fs_info *fs_info,
3817 const struct btrfs_balance_args *bargs,
3818 u64 allowed, const char *type)
3820 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3823 /* Profile is valid and does not have bits outside of the allowed set */
3824 if (alloc_profile_is_valid(bargs->target, 1) &&
3825 (bargs->target & ~allowed) == 0)
3828 btrfs_err(fs_info, "balance: invalid convert %s profile %s",
3829 type, btrfs_bg_type_to_raid_name(bargs->target));
3834 * Fill @buf with textual description of balance filter flags @bargs, up to
3835 * @size_buf including the terminating null. The output may be trimmed if it
3836 * does not fit into the provided buffer.
3838 static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf,
3842 u32 size_bp = size_buf;
3844 u64 flags = bargs->flags;
3845 char tmp_buf[128] = {'\0'};
3850 #define CHECK_APPEND_NOARG(a) \
3852 ret = snprintf(bp, size_bp, (a)); \
3853 if (ret < 0 || ret >= size_bp) \
3854 goto out_overflow; \
3859 #define CHECK_APPEND_1ARG(a, v1) \
3861 ret = snprintf(bp, size_bp, (a), (v1)); \
3862 if (ret < 0 || ret >= size_bp) \
3863 goto out_overflow; \
3868 #define CHECK_APPEND_2ARG(a, v1, v2) \
3870 ret = snprintf(bp, size_bp, (a), (v1), (v2)); \
3871 if (ret < 0 || ret >= size_bp) \
3872 goto out_overflow; \
3877 if (flags & BTRFS_BALANCE_ARGS_CONVERT)
3878 CHECK_APPEND_1ARG("convert=%s,",
3879 btrfs_bg_type_to_raid_name(bargs->target));
3881 if (flags & BTRFS_BALANCE_ARGS_SOFT)
3882 CHECK_APPEND_NOARG("soft,");
3884 if (flags & BTRFS_BALANCE_ARGS_PROFILES) {
3885 btrfs_describe_block_groups(bargs->profiles, tmp_buf,
3887 CHECK_APPEND_1ARG("profiles=%s,", tmp_buf);
3890 if (flags & BTRFS_BALANCE_ARGS_USAGE)
3891 CHECK_APPEND_1ARG("usage=%llu,", bargs->usage);
3893 if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE)
3894 CHECK_APPEND_2ARG("usage=%u..%u,",
3895 bargs->usage_min, bargs->usage_max);
3897 if (flags & BTRFS_BALANCE_ARGS_DEVID)
3898 CHECK_APPEND_1ARG("devid=%llu,", bargs->devid);
3900 if (flags & BTRFS_BALANCE_ARGS_DRANGE)
3901 CHECK_APPEND_2ARG("drange=%llu..%llu,",
3902 bargs->pstart, bargs->pend);
3904 if (flags & BTRFS_BALANCE_ARGS_VRANGE)
3905 CHECK_APPEND_2ARG("vrange=%llu..%llu,",
3906 bargs->vstart, bargs->vend);
3908 if (flags & BTRFS_BALANCE_ARGS_LIMIT)
3909 CHECK_APPEND_1ARG("limit=%llu,", bargs->limit);
3911 if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)
3912 CHECK_APPEND_2ARG("limit=%u..%u,",
3913 bargs->limit_min, bargs->limit_max);
3915 if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE)
3916 CHECK_APPEND_2ARG("stripes=%u..%u,",
3917 bargs->stripes_min, bargs->stripes_max);
3919 #undef CHECK_APPEND_2ARG
3920 #undef CHECK_APPEND_1ARG
3921 #undef CHECK_APPEND_NOARG
3925 if (size_bp < size_buf)
3926 buf[size_buf - size_bp - 1] = '\0'; /* remove last , */
3931 static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info)
3933 u32 size_buf = 1024;
3934 char tmp_buf[192] = {'\0'};
3937 u32 size_bp = size_buf;
3939 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3941 buf = kzalloc(size_buf, GFP_KERNEL);
3947 #define CHECK_APPEND_1ARG(a, v1) \
3949 ret = snprintf(bp, size_bp, (a), (v1)); \
3950 if (ret < 0 || ret >= size_bp) \
3951 goto out_overflow; \
3956 if (bctl->flags & BTRFS_BALANCE_FORCE)
3957 CHECK_APPEND_1ARG("%s", "-f ");
3959 if (bctl->flags & BTRFS_BALANCE_DATA) {
3960 describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf));
3961 CHECK_APPEND_1ARG("-d%s ", tmp_buf);
3964 if (bctl->flags & BTRFS_BALANCE_METADATA) {
3965 describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf));
3966 CHECK_APPEND_1ARG("-m%s ", tmp_buf);
3969 if (bctl->flags & BTRFS_BALANCE_SYSTEM) {
3970 describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf));
3971 CHECK_APPEND_1ARG("-s%s ", tmp_buf);
3974 #undef CHECK_APPEND_1ARG
3978 if (size_bp < size_buf)
3979 buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */
3980 btrfs_info(fs_info, "balance: %s %s",
3981 (bctl->flags & BTRFS_BALANCE_RESUME) ?
3982 "resume" : "start", buf);
3988 * Should be called with balance mutexe held
3990 int btrfs_balance(struct btrfs_fs_info *fs_info,
3991 struct btrfs_balance_control *bctl,
3992 struct btrfs_ioctl_balance_args *bargs)
3994 u64 meta_target, data_target;
4000 bool reducing_redundancy;
4003 if (btrfs_fs_closing(fs_info) ||
4004 atomic_read(&fs_info->balance_pause_req) ||
4005 btrfs_should_cancel_balance(fs_info)) {
4010 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
4011 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
4015 * In case of mixed groups both data and meta should be picked,
4016 * and identical options should be given for both of them.
4018 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
4019 if (mixed && (bctl->flags & allowed)) {
4020 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
4021 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
4022 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
4024 "balance: mixed groups data and metadata options must be the same");
4031 * rw_devices will not change at the moment, device add/delete/replace
4034 num_devices = fs_info->fs_devices->rw_devices;
4037 * SINGLE profile on-disk has no profile bit, but in-memory we have a
4038 * special bit for it, to make it easier to distinguish. Thus we need
4039 * to set it manually, or balance would refuse the profile.
4041 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
4042 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++)
4043 if (num_devices >= btrfs_raid_array[i].devs_min)
4044 allowed |= btrfs_raid_array[i].bg_flag;
4046 if (!validate_convert_profile(fs_info, &bctl->data, allowed, "data") ||
4047 !validate_convert_profile(fs_info, &bctl->meta, allowed, "metadata") ||
4048 !validate_convert_profile(fs_info, &bctl->sys, allowed, "system")) {
4054 * Allow to reduce metadata or system integrity only if force set for
4055 * profiles with redundancy (copies, parity)
4058 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) {
4059 if (btrfs_raid_array[i].ncopies >= 2 ||
4060 btrfs_raid_array[i].tolerated_failures >= 1)
4061 allowed |= btrfs_raid_array[i].bg_flag;
4064 seq = read_seqbegin(&fs_info->profiles_lock);
4066 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4067 (fs_info->avail_system_alloc_bits & allowed) &&
4068 !(bctl->sys.target & allowed)) ||
4069 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4070 (fs_info->avail_metadata_alloc_bits & allowed) &&
4071 !(bctl->meta.target & allowed)))
4072 reducing_redundancy = true;
4074 reducing_redundancy = false;
4076 /* if we're not converting, the target field is uninitialized */
4077 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4078 bctl->meta.target : fs_info->avail_metadata_alloc_bits;
4079 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4080 bctl->data.target : fs_info->avail_data_alloc_bits;
4081 } while (read_seqretry(&fs_info->profiles_lock, seq));
4083 if (reducing_redundancy) {
4084 if (bctl->flags & BTRFS_BALANCE_FORCE) {
4086 "balance: force reducing metadata redundancy");
4089 "balance: reduces metadata redundancy, use --force if you want this");
4095 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <
4096 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {
4098 "balance: metadata profile %s has lower redundancy than data profile %s",
4099 btrfs_bg_type_to_raid_name(meta_target),
4100 btrfs_bg_type_to_raid_name(data_target));
4103 if (fs_info->send_in_progress) {
4104 btrfs_warn_rl(fs_info,
4105 "cannot run balance while send operations are in progress (%d in progress)",
4106 fs_info->send_in_progress);
4111 ret = insert_balance_item(fs_info, bctl);
4112 if (ret && ret != -EEXIST)
4115 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
4116 BUG_ON(ret == -EEXIST);
4117 BUG_ON(fs_info->balance_ctl);
4118 spin_lock(&fs_info->balance_lock);
4119 fs_info->balance_ctl = bctl;
4120 spin_unlock(&fs_info->balance_lock);
4122 BUG_ON(ret != -EEXIST);
4123 spin_lock(&fs_info->balance_lock);
4124 update_balance_args(bctl);
4125 spin_unlock(&fs_info->balance_lock);
4128 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4129 set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4130 describe_balance_start_or_resume(fs_info);
4131 mutex_unlock(&fs_info->balance_mutex);
4133 ret = __btrfs_balance(fs_info);
4135 mutex_lock(&fs_info->balance_mutex);
4136 if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req))
4137 btrfs_info(fs_info, "balance: paused");
4139 * Balance can be canceled by:
4141 * - Regular cancel request
4142 * Then ret == -ECANCELED and balance_cancel_req > 0
4144 * - Fatal signal to "btrfs" process
4145 * Either the signal caught by wait_reserve_ticket() and callers
4146 * got -EINTR, or caught by btrfs_should_cancel_balance() and
4148 * Either way, in this case balance_cancel_req = 0, and
4149 * ret == -EINTR or ret == -ECANCELED.
4151 * So here we only check the return value to catch canceled balance.
4153 else if (ret == -ECANCELED || ret == -EINTR)
4154 btrfs_info(fs_info, "balance: canceled");
4156 btrfs_info(fs_info, "balance: ended with status: %d", ret);
4158 clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4161 memset(bargs, 0, sizeof(*bargs));
4162 btrfs_update_ioctl_balance_args(fs_info, bargs);
4165 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
4166 balance_need_close(fs_info)) {
4167 reset_balance_state(fs_info);
4168 btrfs_exclop_finish(fs_info);
4171 wake_up(&fs_info->balance_wait_q);
4175 if (bctl->flags & BTRFS_BALANCE_RESUME)
4176 reset_balance_state(fs_info);
4179 btrfs_exclop_finish(fs_info);
4184 static int balance_kthread(void *data)
4186 struct btrfs_fs_info *fs_info = data;
4189 mutex_lock(&fs_info->balance_mutex);
4190 if (fs_info->balance_ctl)
4191 ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL);
4192 mutex_unlock(&fs_info->balance_mutex);
4197 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
4199 struct task_struct *tsk;
4201 mutex_lock(&fs_info->balance_mutex);
4202 if (!fs_info->balance_ctl) {
4203 mutex_unlock(&fs_info->balance_mutex);
4206 mutex_unlock(&fs_info->balance_mutex);
4208 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
4209 btrfs_info(fs_info, "balance: resume skipped");
4214 * A ro->rw remount sequence should continue with the paused balance
4215 * regardless of who pauses it, system or the user as of now, so set
4218 spin_lock(&fs_info->balance_lock);
4219 fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME;
4220 spin_unlock(&fs_info->balance_lock);
4222 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
4223 return PTR_ERR_OR_ZERO(tsk);
4226 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
4228 struct btrfs_balance_control *bctl;
4229 struct btrfs_balance_item *item;
4230 struct btrfs_disk_balance_args disk_bargs;
4231 struct btrfs_path *path;
4232 struct extent_buffer *leaf;
4233 struct btrfs_key key;
4236 path = btrfs_alloc_path();
4240 key.objectid = BTRFS_BALANCE_OBJECTID;
4241 key.type = BTRFS_TEMPORARY_ITEM_KEY;
4244 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4247 if (ret > 0) { /* ret = -ENOENT; */
4252 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4258 leaf = path->nodes[0];
4259 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
4261 bctl->flags = btrfs_balance_flags(leaf, item);
4262 bctl->flags |= BTRFS_BALANCE_RESUME;
4264 btrfs_balance_data(leaf, item, &disk_bargs);
4265 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4266 btrfs_balance_meta(leaf, item, &disk_bargs);
4267 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4268 btrfs_balance_sys(leaf, item, &disk_bargs);
4269 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4272 * This should never happen, as the paused balance state is recovered
4273 * during mount without any chance of other exclusive ops to collide.
4275 * This gives the exclusive op status to balance and keeps in paused
4276 * state until user intervention (cancel or umount). If the ownership
4277 * cannot be assigned, show a message but do not fail. The balance
4278 * is in a paused state and must have fs_info::balance_ctl properly
4281 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE))
4283 "balance: cannot set exclusive op status, resume manually");
4285 mutex_lock(&fs_info->balance_mutex);
4286 BUG_ON(fs_info->balance_ctl);
4287 spin_lock(&fs_info->balance_lock);
4288 fs_info->balance_ctl = bctl;
4289 spin_unlock(&fs_info->balance_lock);
4290 mutex_unlock(&fs_info->balance_mutex);
4292 btrfs_free_path(path);
4296 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4300 mutex_lock(&fs_info->balance_mutex);
4301 if (!fs_info->balance_ctl) {
4302 mutex_unlock(&fs_info->balance_mutex);
4306 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4307 atomic_inc(&fs_info->balance_pause_req);
4308 mutex_unlock(&fs_info->balance_mutex);
4310 wait_event(fs_info->balance_wait_q,
4311 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4313 mutex_lock(&fs_info->balance_mutex);
4314 /* we are good with balance_ctl ripped off from under us */
4315 BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4316 atomic_dec(&fs_info->balance_pause_req);
4321 mutex_unlock(&fs_info->balance_mutex);
4325 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4327 mutex_lock(&fs_info->balance_mutex);
4328 if (!fs_info->balance_ctl) {
4329 mutex_unlock(&fs_info->balance_mutex);
4334 * A paused balance with the item stored on disk can be resumed at
4335 * mount time if the mount is read-write. Otherwise it's still paused
4336 * and we must not allow cancelling as it deletes the item.
4338 if (sb_rdonly(fs_info->sb)) {
4339 mutex_unlock(&fs_info->balance_mutex);
4343 atomic_inc(&fs_info->balance_cancel_req);
4345 * if we are running just wait and return, balance item is
4346 * deleted in btrfs_balance in this case
4348 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4349 mutex_unlock(&fs_info->balance_mutex);
4350 wait_event(fs_info->balance_wait_q,
4351 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4352 mutex_lock(&fs_info->balance_mutex);
4354 mutex_unlock(&fs_info->balance_mutex);
4356 * Lock released to allow other waiters to continue, we'll
4357 * reexamine the status again.
4359 mutex_lock(&fs_info->balance_mutex);
4361 if (fs_info->balance_ctl) {
4362 reset_balance_state(fs_info);
4363 btrfs_exclop_finish(fs_info);
4364 btrfs_info(fs_info, "balance: canceled");
4368 BUG_ON(fs_info->balance_ctl ||
4369 test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4370 atomic_dec(&fs_info->balance_cancel_req);
4371 mutex_unlock(&fs_info->balance_mutex);
4375 int btrfs_uuid_scan_kthread(void *data)
4377 struct btrfs_fs_info *fs_info = data;
4378 struct btrfs_root *root = fs_info->tree_root;
4379 struct btrfs_key key;
4380 struct btrfs_path *path = NULL;
4382 struct extent_buffer *eb;
4384 struct btrfs_root_item root_item;
4386 struct btrfs_trans_handle *trans = NULL;
4387 bool closing = false;
4389 path = btrfs_alloc_path();
4396 key.type = BTRFS_ROOT_ITEM_KEY;
4400 if (btrfs_fs_closing(fs_info)) {
4404 ret = btrfs_search_forward(root, &key, path,
4405 BTRFS_OLDEST_GENERATION);
4412 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4413 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4414 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4415 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4418 eb = path->nodes[0];
4419 slot = path->slots[0];
4420 item_size = btrfs_item_size_nr(eb, slot);
4421 if (item_size < sizeof(root_item))
4424 read_extent_buffer(eb, &root_item,
4425 btrfs_item_ptr_offset(eb, slot),
4426 (int)sizeof(root_item));
4427 if (btrfs_root_refs(&root_item) == 0)
4430 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4431 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4435 btrfs_release_path(path);
4437 * 1 - subvol uuid item
4438 * 1 - received_subvol uuid item
4440 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4441 if (IS_ERR(trans)) {
4442 ret = PTR_ERR(trans);
4450 btrfs_release_path(path);
4451 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4452 ret = btrfs_uuid_tree_add(trans, root_item.uuid,
4453 BTRFS_UUID_KEY_SUBVOL,
4456 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4462 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4463 ret = btrfs_uuid_tree_add(trans,
4464 root_item.received_uuid,
4465 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4468 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4475 btrfs_release_path(path);
4477 ret = btrfs_end_transaction(trans);
4483 if (key.offset < (u64)-1) {
4485 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4487 key.type = BTRFS_ROOT_ITEM_KEY;
4488 } else if (key.objectid < (u64)-1) {
4490 key.type = BTRFS_ROOT_ITEM_KEY;
4499 btrfs_free_path(path);
4500 if (trans && !IS_ERR(trans))
4501 btrfs_end_transaction(trans);
4503 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4505 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4506 up(&fs_info->uuid_tree_rescan_sem);
4510 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4512 struct btrfs_trans_handle *trans;
4513 struct btrfs_root *tree_root = fs_info->tree_root;
4514 struct btrfs_root *uuid_root;
4515 struct task_struct *task;
4522 trans = btrfs_start_transaction(tree_root, 2);
4524 return PTR_ERR(trans);
4526 uuid_root = btrfs_create_tree(trans, BTRFS_UUID_TREE_OBJECTID);
4527 if (IS_ERR(uuid_root)) {
4528 ret = PTR_ERR(uuid_root);
4529 btrfs_abort_transaction(trans, ret);
4530 btrfs_end_transaction(trans);
4534 fs_info->uuid_root = uuid_root;
4536 ret = btrfs_commit_transaction(trans);
4540 down(&fs_info->uuid_tree_rescan_sem);
4541 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4543 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4544 btrfs_warn(fs_info, "failed to start uuid_scan task");
4545 up(&fs_info->uuid_tree_rescan_sem);
4546 return PTR_ERR(task);
4553 * shrinking a device means finding all of the device extents past
4554 * the new size, and then following the back refs to the chunks.
4555 * The chunk relocation code actually frees the device extent
4557 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4559 struct btrfs_fs_info *fs_info = device->fs_info;
4560 struct btrfs_root *root = fs_info->dev_root;
4561 struct btrfs_trans_handle *trans;
4562 struct btrfs_dev_extent *dev_extent = NULL;
4563 struct btrfs_path *path;
4569 bool retried = false;
4570 struct extent_buffer *l;
4571 struct btrfs_key key;
4572 struct btrfs_super_block *super_copy = fs_info->super_copy;
4573 u64 old_total = btrfs_super_total_bytes(super_copy);
4574 u64 old_size = btrfs_device_get_total_bytes(device);
4578 new_size = round_down(new_size, fs_info->sectorsize);
4580 diff = round_down(old_size - new_size, fs_info->sectorsize);
4582 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4585 path = btrfs_alloc_path();
4589 path->reada = READA_BACK;
4591 trans = btrfs_start_transaction(root, 0);
4592 if (IS_ERR(trans)) {
4593 btrfs_free_path(path);
4594 return PTR_ERR(trans);
4597 mutex_lock(&fs_info->chunk_mutex);
4599 btrfs_device_set_total_bytes(device, new_size);
4600 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4601 device->fs_devices->total_rw_bytes -= diff;
4602 atomic64_sub(diff, &fs_info->free_chunk_space);
4606 * Once the device's size has been set to the new size, ensure all
4607 * in-memory chunks are synced to disk so that the loop below sees them
4608 * and relocates them accordingly.
4610 if (contains_pending_extent(device, &start, diff)) {
4611 mutex_unlock(&fs_info->chunk_mutex);
4612 ret = btrfs_commit_transaction(trans);
4616 mutex_unlock(&fs_info->chunk_mutex);
4617 btrfs_end_transaction(trans);
4621 key.objectid = device->devid;
4622 key.offset = (u64)-1;
4623 key.type = BTRFS_DEV_EXTENT_KEY;
4626 mutex_lock(&fs_info->delete_unused_bgs_mutex);
4627 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4629 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4633 ret = btrfs_previous_item(root, path, 0, key.type);
4635 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4640 btrfs_release_path(path);
4645 slot = path->slots[0];
4646 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4648 if (key.objectid != device->devid) {
4649 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4650 btrfs_release_path(path);
4654 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4655 length = btrfs_dev_extent_length(l, dev_extent);
4657 if (key.offset + length <= new_size) {
4658 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4659 btrfs_release_path(path);
4663 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4664 btrfs_release_path(path);
4667 * We may be relocating the only data chunk we have,
4668 * which could potentially end up with losing data's
4669 * raid profile, so lets allocate an empty one in
4672 ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset);
4674 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4678 ret = btrfs_relocate_chunk(fs_info, chunk_offset);
4679 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4680 if (ret == -ENOSPC) {
4683 if (ret == -ETXTBSY) {
4685 "could not shrink block group %llu due to active swapfile",
4690 } while (key.offset-- > 0);
4692 if (failed && !retried) {
4696 } else if (failed && retried) {
4701 /* Shrinking succeeded, else we would be at "done". */
4702 trans = btrfs_start_transaction(root, 0);
4703 if (IS_ERR(trans)) {
4704 ret = PTR_ERR(trans);
4708 mutex_lock(&fs_info->chunk_mutex);
4709 /* Clear all state bits beyond the shrunk device size */
4710 clear_extent_bits(&device->alloc_state, new_size, (u64)-1,
4713 btrfs_device_set_disk_total_bytes(device, new_size);
4714 if (list_empty(&device->post_commit_list))
4715 list_add_tail(&device->post_commit_list,
4716 &trans->transaction->dev_update_list);
4718 WARN_ON(diff > old_total);
4719 btrfs_set_super_total_bytes(super_copy,
4720 round_down(old_total - diff, fs_info->sectorsize));
4721 mutex_unlock(&fs_info->chunk_mutex);
4723 /* Now btrfs_update_device() will change the on-disk size. */
4724 ret = btrfs_update_device(trans, device);
4726 btrfs_abort_transaction(trans, ret);
4727 btrfs_end_transaction(trans);
4729 ret = btrfs_commit_transaction(trans);
4732 btrfs_free_path(path);
4734 mutex_lock(&fs_info->chunk_mutex);
4735 btrfs_device_set_total_bytes(device, old_size);
4736 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
4737 device->fs_devices->total_rw_bytes += diff;
4738 atomic64_add(diff, &fs_info->free_chunk_space);
4739 mutex_unlock(&fs_info->chunk_mutex);
4744 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
4745 struct btrfs_key *key,
4746 struct btrfs_chunk *chunk, int item_size)
4748 struct btrfs_super_block *super_copy = fs_info->super_copy;
4749 struct btrfs_disk_key disk_key;
4753 mutex_lock(&fs_info->chunk_mutex);
4754 array_size = btrfs_super_sys_array_size(super_copy);
4755 if (array_size + item_size + sizeof(disk_key)
4756 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4757 mutex_unlock(&fs_info->chunk_mutex);
4761 ptr = super_copy->sys_chunk_array + array_size;
4762 btrfs_cpu_key_to_disk(&disk_key, key);
4763 memcpy(ptr, &disk_key, sizeof(disk_key));
4764 ptr += sizeof(disk_key);
4765 memcpy(ptr, chunk, item_size);
4766 item_size += sizeof(disk_key);
4767 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4768 mutex_unlock(&fs_info->chunk_mutex);
4774 * sort the devices in descending order by max_avail, total_avail
4776 static int btrfs_cmp_device_info(const void *a, const void *b)
4778 const struct btrfs_device_info *di_a = a;
4779 const struct btrfs_device_info *di_b = b;
4781 if (di_a->max_avail > di_b->max_avail)
4783 if (di_a->max_avail < di_b->max_avail)
4785 if (di_a->total_avail > di_b->total_avail)
4787 if (di_a->total_avail < di_b->total_avail)
4792 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4794 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4797 btrfs_set_fs_incompat(info, RAID56);
4800 static void check_raid1c34_incompat_flag(struct btrfs_fs_info *info, u64 type)
4802 if (!(type & (BTRFS_BLOCK_GROUP_RAID1C3 | BTRFS_BLOCK_GROUP_RAID1C4)))
4805 btrfs_set_fs_incompat(info, RAID1C34);
4809 * Structure used internally for __btrfs_alloc_chunk() function.
4810 * Wraps needed parameters.
4812 struct alloc_chunk_ctl {
4815 /* Total number of stripes to allocate */
4817 /* sub_stripes info for map */
4819 /* Stripes per device */
4821 /* Maximum number of devices to use */
4823 /* Minimum number of devices to use */
4825 /* ndevs has to be a multiple of this */
4827 /* Number of copies */
4829 /* Number of stripes worth of bytes to store parity information */
4831 u64 max_stripe_size;
4839 static void init_alloc_chunk_ctl_policy_regular(
4840 struct btrfs_fs_devices *fs_devices,
4841 struct alloc_chunk_ctl *ctl)
4843 u64 type = ctl->type;
4845 if (type & BTRFS_BLOCK_GROUP_DATA) {
4846 ctl->max_stripe_size = SZ_1G;
4847 ctl->max_chunk_size = BTRFS_MAX_DATA_CHUNK_SIZE;
4848 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4849 /* For larger filesystems, use larger metadata chunks */
4850 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
4851 ctl->max_stripe_size = SZ_1G;
4853 ctl->max_stripe_size = SZ_256M;
4854 ctl->max_chunk_size = ctl->max_stripe_size;
4855 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4856 ctl->max_stripe_size = SZ_32M;
4857 ctl->max_chunk_size = 2 * ctl->max_stripe_size;
4858 ctl->devs_max = min_t(int, ctl->devs_max,
4859 BTRFS_MAX_DEVS_SYS_CHUNK);
4864 /* We don't want a chunk larger than 10% of writable space */
4865 ctl->max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4866 ctl->max_chunk_size);
4867 ctl->dev_extent_min = BTRFS_STRIPE_LEN * ctl->dev_stripes;
4870 static void init_alloc_chunk_ctl(struct btrfs_fs_devices *fs_devices,
4871 struct alloc_chunk_ctl *ctl)
4873 int index = btrfs_bg_flags_to_raid_index(ctl->type);
4875 ctl->sub_stripes = btrfs_raid_array[index].sub_stripes;
4876 ctl->dev_stripes = btrfs_raid_array[index].dev_stripes;
4877 ctl->devs_max = btrfs_raid_array[index].devs_max;
4879 ctl->devs_max = BTRFS_MAX_DEVS(fs_devices->fs_info);
4880 ctl->devs_min = btrfs_raid_array[index].devs_min;
4881 ctl->devs_increment = btrfs_raid_array[index].devs_increment;
4882 ctl->ncopies = btrfs_raid_array[index].ncopies;
4883 ctl->nparity = btrfs_raid_array[index].nparity;
4886 switch (fs_devices->chunk_alloc_policy) {
4887 case BTRFS_CHUNK_ALLOC_REGULAR:
4888 init_alloc_chunk_ctl_policy_regular(fs_devices, ctl);
4895 static int gather_device_info(struct btrfs_fs_devices *fs_devices,
4896 struct alloc_chunk_ctl *ctl,
4897 struct btrfs_device_info *devices_info)
4899 struct btrfs_fs_info *info = fs_devices->fs_info;
4900 struct btrfs_device *device;
4902 u64 dev_extent_want = ctl->max_stripe_size * ctl->dev_stripes;
4909 * in the first pass through the devices list, we gather information
4910 * about the available holes on each device.
4912 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
4913 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4915 "BTRFS: read-only device in alloc_list\n");
4919 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
4920 &device->dev_state) ||
4921 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4924 if (device->total_bytes > device->bytes_used)
4925 total_avail = device->total_bytes - device->bytes_used;
4929 /* If there is no space on this device, skip it. */
4930 if (total_avail < ctl->dev_extent_min)
4933 ret = find_free_dev_extent(device, dev_extent_want, &dev_offset,
4935 if (ret && ret != -ENOSPC)
4939 max_avail = dev_extent_want;
4941 if (max_avail < ctl->dev_extent_min) {
4942 if (btrfs_test_opt(info, ENOSPC_DEBUG))
4944 "%s: devid %llu has no free space, have=%llu want=%llu",
4945 __func__, device->devid, max_avail,
4946 ctl->dev_extent_min);
4950 if (ndevs == fs_devices->rw_devices) {
4951 WARN(1, "%s: found more than %llu devices\n",
4952 __func__, fs_devices->rw_devices);
4955 devices_info[ndevs].dev_offset = dev_offset;
4956 devices_info[ndevs].max_avail = max_avail;
4957 devices_info[ndevs].total_avail = total_avail;
4958 devices_info[ndevs].dev = device;
4964 * now sort the devices by hole size / available space
4966 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4967 btrfs_cmp_device_info, NULL);
4972 static int decide_stripe_size_regular(struct alloc_chunk_ctl *ctl,
4973 struct btrfs_device_info *devices_info)
4975 /* Number of stripes that count for block group size */
4979 * The primary goal is to maximize the number of stripes, so use as
4980 * many devices as possible, even if the stripes are not maximum sized.
4982 * The DUP profile stores more than one stripe per device, the
4983 * max_avail is the total size so we have to adjust.
4985 ctl->stripe_size = div_u64(devices_info[ctl->ndevs - 1].max_avail,
4987 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
4989 /* This will have to be fixed for RAID1 and RAID10 over more drives */
4990 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
4993 * Use the number of data stripes to figure out how big this chunk is
4994 * really going to be in terms of logical address space, and compare
4995 * that answer with the max chunk size. If it's higher, we try to
4996 * reduce stripe_size.
4998 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) {
5000 * Reduce stripe_size, round it up to a 16MB boundary again and
5001 * then use it, unless it ends up being even bigger than the
5002 * previous value we had already.
5004 ctl->stripe_size = min(round_up(div_u64(ctl->max_chunk_size,
5005 data_stripes), SZ_16M),
5009 /* Align to BTRFS_STRIPE_LEN */
5010 ctl->stripe_size = round_down(ctl->stripe_size, BTRFS_STRIPE_LEN);
5011 ctl->chunk_size = ctl->stripe_size * data_stripes;
5016 static int decide_stripe_size(struct btrfs_fs_devices *fs_devices,
5017 struct alloc_chunk_ctl *ctl,
5018 struct btrfs_device_info *devices_info)
5020 struct btrfs_fs_info *info = fs_devices->fs_info;
5023 * Round down to number of usable stripes, devs_increment can be any
5024 * number so we can't use round_down() that requires power of 2, while
5025 * rounddown is safe.
5027 ctl->ndevs = rounddown(ctl->ndevs, ctl->devs_increment);
5029 if (ctl->ndevs < ctl->devs_min) {
5030 if (btrfs_test_opt(info, ENOSPC_DEBUG)) {
5032 "%s: not enough devices with free space: have=%d minimum required=%d",
5033 __func__, ctl->ndevs, ctl->devs_min);
5038 ctl->ndevs = min(ctl->ndevs, ctl->devs_max);
5040 switch (fs_devices->chunk_alloc_policy) {
5041 case BTRFS_CHUNK_ALLOC_REGULAR:
5042 return decide_stripe_size_regular(ctl, devices_info);
5048 static int create_chunk(struct btrfs_trans_handle *trans,
5049 struct alloc_chunk_ctl *ctl,
5050 struct btrfs_device_info *devices_info)
5052 struct btrfs_fs_info *info = trans->fs_info;
5053 struct map_lookup *map = NULL;
5054 struct extent_map_tree *em_tree;
5055 struct extent_map *em;
5056 u64 start = ctl->start;
5057 u64 type = ctl->type;
5062 map = kmalloc(map_lookup_size(ctl->num_stripes), GFP_NOFS);
5065 map->num_stripes = ctl->num_stripes;
5067 for (i = 0; i < ctl->ndevs; ++i) {
5068 for (j = 0; j < ctl->dev_stripes; ++j) {
5069 int s = i * ctl->dev_stripes + j;
5070 map->stripes[s].dev = devices_info[i].dev;
5071 map->stripes[s].physical = devices_info[i].dev_offset +
5072 j * ctl->stripe_size;
5075 map->stripe_len = BTRFS_STRIPE_LEN;
5076 map->io_align = BTRFS_STRIPE_LEN;
5077 map->io_width = BTRFS_STRIPE_LEN;
5079 map->sub_stripes = ctl->sub_stripes;
5081 trace_btrfs_chunk_alloc(info, map, start, ctl->chunk_size);
5083 em = alloc_extent_map();
5088 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
5089 em->map_lookup = map;
5091 em->len = ctl->chunk_size;
5092 em->block_start = 0;
5093 em->block_len = em->len;
5094 em->orig_block_len = ctl->stripe_size;
5096 em_tree = &info->mapping_tree;
5097 write_lock(&em_tree->lock);
5098 ret = add_extent_mapping(em_tree, em, 0);
5100 write_unlock(&em_tree->lock);
5101 free_extent_map(em);
5104 write_unlock(&em_tree->lock);
5106 ret = btrfs_make_block_group(trans, 0, type, start, ctl->chunk_size);
5108 goto error_del_extent;
5110 for (i = 0; i < map->num_stripes; i++) {
5111 struct btrfs_device *dev = map->stripes[i].dev;
5113 btrfs_device_set_bytes_used(dev,
5114 dev->bytes_used + ctl->stripe_size);
5115 if (list_empty(&dev->post_commit_list))
5116 list_add_tail(&dev->post_commit_list,
5117 &trans->transaction->dev_update_list);
5120 atomic64_sub(ctl->stripe_size * map->num_stripes,
5121 &info->free_chunk_space);
5123 free_extent_map(em);
5124 check_raid56_incompat_flag(info, type);
5125 check_raid1c34_incompat_flag(info, type);
5130 write_lock(&em_tree->lock);
5131 remove_extent_mapping(em_tree, em);
5132 write_unlock(&em_tree->lock);
5134 /* One for our allocation */
5135 free_extent_map(em);
5136 /* One for the tree reference */
5137 free_extent_map(em);
5142 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, u64 type)
5144 struct btrfs_fs_info *info = trans->fs_info;
5145 struct btrfs_fs_devices *fs_devices = info->fs_devices;
5146 struct btrfs_device_info *devices_info = NULL;
5147 struct alloc_chunk_ctl ctl;
5150 lockdep_assert_held(&info->chunk_mutex);
5152 if (!alloc_profile_is_valid(type, 0)) {
5157 if (list_empty(&fs_devices->alloc_list)) {
5158 if (btrfs_test_opt(info, ENOSPC_DEBUG))
5159 btrfs_debug(info, "%s: no writable device", __func__);
5163 if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
5164 btrfs_err(info, "invalid chunk type 0x%llx requested", type);
5169 ctl.start = find_next_chunk(info);
5171 init_alloc_chunk_ctl(fs_devices, &ctl);
5173 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
5178 ret = gather_device_info(fs_devices, &ctl, devices_info);
5182 ret = decide_stripe_size(fs_devices, &ctl, devices_info);
5186 ret = create_chunk(trans, &ctl, devices_info);
5189 kfree(devices_info);
5194 * Chunk allocation falls into two parts. The first part does work
5195 * that makes the new allocated chunk usable, but does not do any operation
5196 * that modifies the chunk tree. The second part does the work that
5197 * requires modifying the chunk tree. This division is important for the
5198 * bootstrap process of adding storage to a seed btrfs.
5200 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
5201 u64 chunk_offset, u64 chunk_size)
5203 struct btrfs_fs_info *fs_info = trans->fs_info;
5204 struct btrfs_root *extent_root = fs_info->extent_root;
5205 struct btrfs_root *chunk_root = fs_info->chunk_root;
5206 struct btrfs_key key;
5207 struct btrfs_device *device;
5208 struct btrfs_chunk *chunk;
5209 struct btrfs_stripe *stripe;
5210 struct extent_map *em;
5211 struct map_lookup *map;
5218 em = btrfs_get_chunk_map(fs_info, chunk_offset, chunk_size);
5222 map = em->map_lookup;
5223 item_size = btrfs_chunk_item_size(map->num_stripes);
5224 stripe_size = em->orig_block_len;
5226 chunk = kzalloc(item_size, GFP_NOFS);
5233 * Take the device list mutex to prevent races with the final phase of
5234 * a device replace operation that replaces the device object associated
5235 * with the map's stripes, because the device object's id can change
5236 * at any time during that final phase of the device replace operation
5237 * (dev-replace.c:btrfs_dev_replace_finishing()).
5239 mutex_lock(&fs_info->fs_devices->device_list_mutex);
5240 for (i = 0; i < map->num_stripes; i++) {
5241 device = map->stripes[i].dev;
5242 dev_offset = map->stripes[i].physical;
5244 ret = btrfs_update_device(trans, device);
5247 ret = btrfs_alloc_dev_extent(trans, device, chunk_offset,
5248 dev_offset, stripe_size);
5253 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
5257 stripe = &chunk->stripe;
5258 for (i = 0; i < map->num_stripes; i++) {
5259 device = map->stripes[i].dev;
5260 dev_offset = map->stripes[i].physical;
5262 btrfs_set_stack_stripe_devid(stripe, device->devid);
5263 btrfs_set_stack_stripe_offset(stripe, dev_offset);
5264 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
5267 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
5269 btrfs_set_stack_chunk_length(chunk, chunk_size);
5270 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
5271 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
5272 btrfs_set_stack_chunk_type(chunk, map->type);
5273 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
5274 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
5275 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
5276 btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
5277 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
5279 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
5280 key.type = BTRFS_CHUNK_ITEM_KEY;
5281 key.offset = chunk_offset;
5283 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
5284 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
5286 * TODO: Cleanup of inserted chunk root in case of
5289 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
5294 free_extent_map(em);
5298 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans)
5300 struct btrfs_fs_info *fs_info = trans->fs_info;
5304 alloc_profile = btrfs_metadata_alloc_profile(fs_info);
5305 ret = btrfs_alloc_chunk(trans, alloc_profile);
5309 alloc_profile = btrfs_system_alloc_profile(fs_info);
5310 ret = btrfs_alloc_chunk(trans, alloc_profile);
5314 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
5316 const int index = btrfs_bg_flags_to_raid_index(map->type);
5318 return btrfs_raid_array[index].tolerated_failures;
5321 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset)
5323 struct extent_map *em;
5324 struct map_lookup *map;
5329 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
5333 map = em->map_lookup;
5334 for (i = 0; i < map->num_stripes; i++) {
5335 if (test_bit(BTRFS_DEV_STATE_MISSING,
5336 &map->stripes[i].dev->dev_state)) {
5340 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE,
5341 &map->stripes[i].dev->dev_state)) {
5348 * If the number of missing devices is larger than max errors,
5349 * we can not write the data into that chunk successfully, so
5352 if (miss_ndevs > btrfs_chunk_max_errors(map))
5355 free_extent_map(em);
5359 void btrfs_mapping_tree_free(struct extent_map_tree *tree)
5361 struct extent_map *em;
5364 write_lock(&tree->lock);
5365 em = lookup_extent_mapping(tree, 0, (u64)-1);
5367 remove_extent_mapping(tree, em);
5368 write_unlock(&tree->lock);
5372 free_extent_map(em);
5373 /* once for the tree */
5374 free_extent_map(em);
5378 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5380 struct extent_map *em;
5381 struct map_lookup *map;
5384 em = btrfs_get_chunk_map(fs_info, logical, len);
5387 * We could return errors for these cases, but that could get
5388 * ugly and we'd probably do the same thing which is just not do
5389 * anything else and exit, so return 1 so the callers don't try
5390 * to use other copies.
5394 map = em->map_lookup;
5395 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1_MASK))
5396 ret = map->num_stripes;
5397 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5398 ret = map->sub_stripes;
5399 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5401 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5403 * There could be two corrupted data stripes, we need
5404 * to loop retry in order to rebuild the correct data.
5406 * Fail a stripe at a time on every retry except the
5407 * stripe under reconstruction.
5409 ret = map->num_stripes;
5412 free_extent_map(em);
5414 down_read(&fs_info->dev_replace.rwsem);
5415 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) &&
5416 fs_info->dev_replace.tgtdev)
5418 up_read(&fs_info->dev_replace.rwsem);
5423 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
5426 struct extent_map *em;
5427 struct map_lookup *map;
5428 unsigned long len = fs_info->sectorsize;
5430 em = btrfs_get_chunk_map(fs_info, logical, len);
5432 if (!WARN_ON(IS_ERR(em))) {
5433 map = em->map_lookup;
5434 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5435 len = map->stripe_len * nr_data_stripes(map);
5436 free_extent_map(em);
5441 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5443 struct extent_map *em;
5444 struct map_lookup *map;
5447 em = btrfs_get_chunk_map(fs_info, logical, len);
5449 if(!WARN_ON(IS_ERR(em))) {
5450 map = em->map_lookup;
5451 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5453 free_extent_map(em);
5458 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5459 struct map_lookup *map, int first,
5460 int dev_replace_is_ongoing)
5464 int preferred_mirror;
5466 struct btrfs_device *srcdev;
5469 (BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10)));
5471 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5472 num_stripes = map->sub_stripes;
5474 num_stripes = map->num_stripes;
5476 preferred_mirror = first + current->pid % num_stripes;
5478 if (dev_replace_is_ongoing &&
5479 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5480 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5481 srcdev = fs_info->dev_replace.srcdev;
5486 * try to avoid the drive that is the source drive for a
5487 * dev-replace procedure, only choose it if no other non-missing
5488 * mirror is available
5490 for (tolerance = 0; tolerance < 2; tolerance++) {
5491 if (map->stripes[preferred_mirror].dev->bdev &&
5492 (tolerance || map->stripes[preferred_mirror].dev != srcdev))
5493 return preferred_mirror;
5494 for (i = first; i < first + num_stripes; i++) {
5495 if (map->stripes[i].dev->bdev &&
5496 (tolerance || map->stripes[i].dev != srcdev))
5501 /* we couldn't find one that doesn't fail. Just return something
5502 * and the io error handling code will clean up eventually
5504 return preferred_mirror;
5507 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5508 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5515 for (i = 0; i < num_stripes - 1; i++) {
5516 /* Swap if parity is on a smaller index */
5517 if (bbio->raid_map[i] > bbio->raid_map[i + 1]) {
5518 swap(bbio->stripes[i], bbio->stripes[i + 1]);
5519 swap(bbio->raid_map[i], bbio->raid_map[i + 1]);
5526 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5528 struct btrfs_bio *bbio = kzalloc(
5529 /* the size of the btrfs_bio */
5530 sizeof(struct btrfs_bio) +
5531 /* plus the variable array for the stripes */
5532 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5533 /* plus the variable array for the tgt dev */
5534 sizeof(int) * (real_stripes) +
5536 * plus the raid_map, which includes both the tgt dev
5539 sizeof(u64) * (total_stripes),
5540 GFP_NOFS|__GFP_NOFAIL);
5542 atomic_set(&bbio->error, 0);
5543 refcount_set(&bbio->refs, 1);
5545 bbio->tgtdev_map = (int *)(bbio->stripes + total_stripes);
5546 bbio->raid_map = (u64 *)(bbio->tgtdev_map + real_stripes);
5551 void btrfs_get_bbio(struct btrfs_bio *bbio)
5553 WARN_ON(!refcount_read(&bbio->refs));
5554 refcount_inc(&bbio->refs);
5557 void btrfs_put_bbio(struct btrfs_bio *bbio)
5561 if (refcount_dec_and_test(&bbio->refs))
5565 /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */
5567 * Please note that, discard won't be sent to target device of device
5570 static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info,
5571 u64 logical, u64 *length_ret,
5572 struct btrfs_bio **bbio_ret)
5574 struct extent_map *em;
5575 struct map_lookup *map;
5576 struct btrfs_bio *bbio;
5577 u64 length = *length_ret;
5581 u64 stripe_end_offset;
5588 u32 sub_stripes = 0;
5589 u64 stripes_per_dev = 0;
5590 u32 remaining_stripes = 0;
5591 u32 last_stripe = 0;
5595 /* discard always return a bbio */
5598 em = btrfs_get_chunk_map(fs_info, logical, length);
5602 map = em->map_lookup;
5603 /* we don't discard raid56 yet */
5604 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5609 offset = logical - em->start;
5610 length = min_t(u64, em->start + em->len - logical, length);
5611 *length_ret = length;
5613 stripe_len = map->stripe_len;
5615 * stripe_nr counts the total number of stripes we have to stride
5616 * to get to this block
5618 stripe_nr = div64_u64(offset, stripe_len);
5620 /* stripe_offset is the offset of this block in its stripe */
5621 stripe_offset = offset - stripe_nr * stripe_len;
5623 stripe_nr_end = round_up(offset + length, map->stripe_len);
5624 stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len);
5625 stripe_cnt = stripe_nr_end - stripe_nr;
5626 stripe_end_offset = stripe_nr_end * map->stripe_len -
5629 * after this, stripe_nr is the number of stripes on this
5630 * device we have to walk to find the data, and stripe_index is
5631 * the number of our device in the stripe array
5635 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5636 BTRFS_BLOCK_GROUP_RAID10)) {
5637 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5640 sub_stripes = map->sub_stripes;
5642 factor = map->num_stripes / sub_stripes;
5643 num_stripes = min_t(u64, map->num_stripes,
5644 sub_stripes * stripe_cnt);
5645 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5646 stripe_index *= sub_stripes;
5647 stripes_per_dev = div_u64_rem(stripe_cnt, factor,
5648 &remaining_stripes);
5649 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5650 last_stripe *= sub_stripes;
5651 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1_MASK |
5652 BTRFS_BLOCK_GROUP_DUP)) {
5653 num_stripes = map->num_stripes;
5655 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5659 bbio = alloc_btrfs_bio(num_stripes, 0);
5665 for (i = 0; i < num_stripes; i++) {
5666 bbio->stripes[i].physical =
5667 map->stripes[stripe_index].physical +
5668 stripe_offset + stripe_nr * map->stripe_len;
5669 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5671 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5672 BTRFS_BLOCK_GROUP_RAID10)) {
5673 bbio->stripes[i].length = stripes_per_dev *
5676 if (i / sub_stripes < remaining_stripes)
5677 bbio->stripes[i].length +=
5681 * Special for the first stripe and
5684 * |-------|...|-------|
5688 if (i < sub_stripes)
5689 bbio->stripes[i].length -=
5692 if (stripe_index >= last_stripe &&
5693 stripe_index <= (last_stripe +
5695 bbio->stripes[i].length -=
5698 if (i == sub_stripes - 1)
5701 bbio->stripes[i].length = length;
5705 if (stripe_index == map->num_stripes) {
5712 bbio->map_type = map->type;
5713 bbio->num_stripes = num_stripes;
5715 free_extent_map(em);
5720 * In dev-replace case, for repair case (that's the only case where the mirror
5721 * is selected explicitly when calling btrfs_map_block), blocks left of the
5722 * left cursor can also be read from the target drive.
5724 * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the
5726 * For READ, it also needs to be supported using the same mirror number.
5728 * If the requested block is not left of the left cursor, EIO is returned. This
5729 * can happen because btrfs_num_copies() returns one more in the dev-replace
5732 static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info,
5733 u64 logical, u64 length,
5734 u64 srcdev_devid, int *mirror_num,
5737 struct btrfs_bio *bbio = NULL;
5739 int index_srcdev = 0;
5741 u64 physical_of_found = 0;
5745 ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS,
5746 logical, &length, &bbio, 0, 0);
5748 ASSERT(bbio == NULL);
5752 num_stripes = bbio->num_stripes;
5753 if (*mirror_num > num_stripes) {
5755 * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror,
5756 * that means that the requested area is not left of the left
5759 btrfs_put_bbio(bbio);
5764 * process the rest of the function using the mirror_num of the source
5765 * drive. Therefore look it up first. At the end, patch the device
5766 * pointer to the one of the target drive.
5768 for (i = 0; i < num_stripes; i++) {
5769 if (bbio->stripes[i].dev->devid != srcdev_devid)
5773 * In case of DUP, in order to keep it simple, only add the
5774 * mirror with the lowest physical address
5777 physical_of_found <= bbio->stripes[i].physical)
5782 physical_of_found = bbio->stripes[i].physical;
5785 btrfs_put_bbio(bbio);
5791 *mirror_num = index_srcdev + 1;
5792 *physical = physical_of_found;
5796 static void handle_ops_on_dev_replace(enum btrfs_map_op op,
5797 struct btrfs_bio **bbio_ret,
5798 struct btrfs_dev_replace *dev_replace,
5799 int *num_stripes_ret, int *max_errors_ret)
5801 struct btrfs_bio *bbio = *bbio_ret;
5802 u64 srcdev_devid = dev_replace->srcdev->devid;
5803 int tgtdev_indexes = 0;
5804 int num_stripes = *num_stripes_ret;
5805 int max_errors = *max_errors_ret;
5808 if (op == BTRFS_MAP_WRITE) {
5809 int index_where_to_add;
5812 * duplicate the write operations while the dev replace
5813 * procedure is running. Since the copying of the old disk to
5814 * the new disk takes place at run time while the filesystem is
5815 * mounted writable, the regular write operations to the old
5816 * disk have to be duplicated to go to the new disk as well.
5818 * Note that device->missing is handled by the caller, and that
5819 * the write to the old disk is already set up in the stripes
5822 index_where_to_add = num_stripes;
5823 for (i = 0; i < num_stripes; i++) {
5824 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5825 /* write to new disk, too */
5826 struct btrfs_bio_stripe *new =
5827 bbio->stripes + index_where_to_add;
5828 struct btrfs_bio_stripe *old =
5831 new->physical = old->physical;
5832 new->length = old->length;
5833 new->dev = dev_replace->tgtdev;
5834 bbio->tgtdev_map[i] = index_where_to_add;
5835 index_where_to_add++;
5840 num_stripes = index_where_to_add;
5841 } else if (op == BTRFS_MAP_GET_READ_MIRRORS) {
5842 int index_srcdev = 0;
5844 u64 physical_of_found = 0;
5847 * During the dev-replace procedure, the target drive can also
5848 * be used to read data in case it is needed to repair a corrupt
5849 * block elsewhere. This is possible if the requested area is
5850 * left of the left cursor. In this area, the target drive is a
5851 * full copy of the source drive.
5853 for (i = 0; i < num_stripes; i++) {
5854 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5856 * In case of DUP, in order to keep it simple,
5857 * only add the mirror with the lowest physical
5861 physical_of_found <=
5862 bbio->stripes[i].physical)
5866 physical_of_found = bbio->stripes[i].physical;
5870 struct btrfs_bio_stripe *tgtdev_stripe =
5871 bbio->stripes + num_stripes;
5873 tgtdev_stripe->physical = physical_of_found;
5874 tgtdev_stripe->length =
5875 bbio->stripes[index_srcdev].length;
5876 tgtdev_stripe->dev = dev_replace->tgtdev;
5877 bbio->tgtdev_map[index_srcdev] = num_stripes;
5884 *num_stripes_ret = num_stripes;
5885 *max_errors_ret = max_errors;
5886 bbio->num_tgtdevs = tgtdev_indexes;
5890 static bool need_full_stripe(enum btrfs_map_op op)
5892 return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS);
5896 * btrfs_get_io_geometry - calculates the geomery of a particular (address, len)
5897 * tuple. This information is used to calculate how big a
5898 * particular bio can get before it straddles a stripe.
5900 * @fs_info - the filesystem
5901 * @logical - address that we want to figure out the geometry of
5902 * @len - the length of IO we are going to perform, starting at @logical
5903 * @op - type of operation - write or read
5904 * @io_geom - pointer used to return values
5906 * Returns < 0 in case a chunk for the given logical address cannot be found,
5907 * usually shouldn't happen unless @logical is corrupted, 0 otherwise.
5909 int btrfs_get_io_geometry(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
5910 u64 logical, u64 len, struct btrfs_io_geometry *io_geom)
5912 struct extent_map *em;
5913 struct map_lookup *map;
5918 u64 raid56_full_stripe_start = (u64)-1;
5922 ASSERT(op != BTRFS_MAP_DISCARD);
5924 em = btrfs_get_chunk_map(fs_info, logical, len);
5928 map = em->map_lookup;
5929 /* Offset of this logical address in the chunk */
5930 offset = logical - em->start;
5931 /* Len of a stripe in a chunk */
5932 stripe_len = map->stripe_len;
5933 /* Stripe wher this block falls in */
5934 stripe_nr = div64_u64(offset, stripe_len);
5935 /* Offset of stripe in the chunk */
5936 stripe_offset = stripe_nr * stripe_len;
5937 if (offset < stripe_offset) {
5939 "stripe math has gone wrong, stripe_offset=%llu offset=%llu start=%llu logical=%llu stripe_len=%llu",
5940 stripe_offset, offset, em->start, logical, stripe_len);
5945 /* stripe_offset is the offset of this block in its stripe */
5946 stripe_offset = offset - stripe_offset;
5947 data_stripes = nr_data_stripes(map);
5949 if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5950 u64 max_len = stripe_len - stripe_offset;
5953 * In case of raid56, we need to know the stripe aligned start
5955 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5956 unsigned long full_stripe_len = stripe_len * data_stripes;
5957 raid56_full_stripe_start = offset;
5960 * Allow a write of a full stripe, but make sure we
5961 * don't allow straddling of stripes
5963 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5965 raid56_full_stripe_start *= full_stripe_len;
5968 * For writes to RAID[56], allow a full stripeset across
5969 * all disks. For other RAID types and for RAID[56]
5970 * reads, just allow a single stripe (on a single disk).
5972 if (op == BTRFS_MAP_WRITE) {
5973 max_len = stripe_len * data_stripes -
5974 (offset - raid56_full_stripe_start);
5977 len = min_t(u64, em->len - offset, max_len);
5979 len = em->len - offset;
5983 io_geom->offset = offset;
5984 io_geom->stripe_len = stripe_len;
5985 io_geom->stripe_nr = stripe_nr;
5986 io_geom->stripe_offset = stripe_offset;
5987 io_geom->raid56_stripe_offset = raid56_full_stripe_start;
5991 free_extent_map(em);
5995 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
5996 enum btrfs_map_op op,
5997 u64 logical, u64 *length,
5998 struct btrfs_bio **bbio_ret,
5999 int mirror_num, int need_raid_map)
6001 struct extent_map *em;
6002 struct map_lookup *map;
6012 int tgtdev_indexes = 0;
6013 struct btrfs_bio *bbio = NULL;
6014 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
6015 int dev_replace_is_ongoing = 0;
6016 int num_alloc_stripes;
6017 int patch_the_first_stripe_for_dev_replace = 0;
6018 u64 physical_to_patch_in_first_stripe = 0;
6019 u64 raid56_full_stripe_start = (u64)-1;
6020 struct btrfs_io_geometry geom;
6023 ASSERT(op != BTRFS_MAP_DISCARD);
6025 ret = btrfs_get_io_geometry(fs_info, op, logical, *length, &geom);
6029 em = btrfs_get_chunk_map(fs_info, logical, *length);
6030 ASSERT(!IS_ERR(em));
6031 map = em->map_lookup;
6034 stripe_len = geom.stripe_len;
6035 stripe_nr = geom.stripe_nr;
6036 stripe_offset = geom.stripe_offset;
6037 raid56_full_stripe_start = geom.raid56_stripe_offset;
6038 data_stripes = nr_data_stripes(map);
6040 down_read(&dev_replace->rwsem);
6041 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
6043 * Hold the semaphore for read during the whole operation, write is
6044 * requested at commit time but must wait.
6046 if (!dev_replace_is_ongoing)
6047 up_read(&dev_replace->rwsem);
6049 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
6050 !need_full_stripe(op) && dev_replace->tgtdev != NULL) {
6051 ret = get_extra_mirror_from_replace(fs_info, logical, *length,
6052 dev_replace->srcdev->devid,
6054 &physical_to_patch_in_first_stripe);
6058 patch_the_first_stripe_for_dev_replace = 1;
6059 } else if (mirror_num > map->num_stripes) {
6065 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
6066 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
6068 if (!need_full_stripe(op))
6070 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1_MASK) {
6071 if (need_full_stripe(op))
6072 num_stripes = map->num_stripes;
6073 else if (mirror_num)
6074 stripe_index = mirror_num - 1;
6076 stripe_index = find_live_mirror(fs_info, map, 0,
6077 dev_replace_is_ongoing);
6078 mirror_num = stripe_index + 1;
6081 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
6082 if (need_full_stripe(op)) {
6083 num_stripes = map->num_stripes;
6084 } else if (mirror_num) {
6085 stripe_index = mirror_num - 1;
6090 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
6091 u32 factor = map->num_stripes / map->sub_stripes;
6093 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
6094 stripe_index *= map->sub_stripes;
6096 if (need_full_stripe(op))
6097 num_stripes = map->sub_stripes;
6098 else if (mirror_num)
6099 stripe_index += mirror_num - 1;
6101 int old_stripe_index = stripe_index;
6102 stripe_index = find_live_mirror(fs_info, map,
6104 dev_replace_is_ongoing);
6105 mirror_num = stripe_index - old_stripe_index + 1;
6108 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6109 if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) {
6110 /* push stripe_nr back to the start of the full stripe */
6111 stripe_nr = div64_u64(raid56_full_stripe_start,
6112 stripe_len * data_stripes);
6114 /* RAID[56] write or recovery. Return all stripes */
6115 num_stripes = map->num_stripes;
6116 max_errors = nr_parity_stripes(map);
6118 *length = map->stripe_len;
6123 * Mirror #0 or #1 means the original data block.
6124 * Mirror #2 is RAID5 parity block.
6125 * Mirror #3 is RAID6 Q block.
6127 stripe_nr = div_u64_rem(stripe_nr,
6128 data_stripes, &stripe_index);
6130 stripe_index = data_stripes + mirror_num - 2;
6132 /* We distribute the parity blocks across stripes */
6133 div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
6135 if (!need_full_stripe(op) && mirror_num <= 1)
6140 * after this, stripe_nr is the number of stripes on this
6141 * device we have to walk to find the data, and stripe_index is
6142 * the number of our device in the stripe array
6144 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
6146 mirror_num = stripe_index + 1;
6148 if (stripe_index >= map->num_stripes) {
6150 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
6151 stripe_index, map->num_stripes);
6156 num_alloc_stripes = num_stripes;
6157 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) {
6158 if (op == BTRFS_MAP_WRITE)
6159 num_alloc_stripes <<= 1;
6160 if (op == BTRFS_MAP_GET_READ_MIRRORS)
6161 num_alloc_stripes++;
6162 tgtdev_indexes = num_stripes;
6165 bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
6171 for (i = 0; i < num_stripes; i++) {
6172 bbio->stripes[i].physical = map->stripes[stripe_index].physical +
6173 stripe_offset + stripe_nr * map->stripe_len;
6174 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
6178 /* build raid_map */
6179 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map &&
6180 (need_full_stripe(op) || mirror_num > 1)) {
6184 /* Work out the disk rotation on this stripe-set */
6185 div_u64_rem(stripe_nr, num_stripes, &rot);
6187 /* Fill in the logical address of each stripe */
6188 tmp = stripe_nr * data_stripes;
6189 for (i = 0; i < data_stripes; i++)
6190 bbio->raid_map[(i+rot) % num_stripes] =
6191 em->start + (tmp + i) * map->stripe_len;
6193 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
6194 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
6195 bbio->raid_map[(i+rot+1) % num_stripes] =
6198 sort_parity_stripes(bbio, num_stripes);
6201 if (need_full_stripe(op))
6202 max_errors = btrfs_chunk_max_errors(map);
6204 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
6205 need_full_stripe(op)) {
6206 handle_ops_on_dev_replace(op, &bbio, dev_replace, &num_stripes,
6211 bbio->map_type = map->type;
6212 bbio->num_stripes = num_stripes;
6213 bbio->max_errors = max_errors;
6214 bbio->mirror_num = mirror_num;
6217 * this is the case that REQ_READ && dev_replace_is_ongoing &&
6218 * mirror_num == num_stripes + 1 && dev_replace target drive is
6219 * available as a mirror
6221 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
6222 WARN_ON(num_stripes > 1);
6223 bbio->stripes[0].dev = dev_replace->tgtdev;
6224 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
6225 bbio->mirror_num = map->num_stripes + 1;
6228 if (dev_replace_is_ongoing) {
6229 lockdep_assert_held(&dev_replace->rwsem);
6230 /* Unlock and let waiting writers proceed */
6231 up_read(&dev_replace->rwsem);
6233 free_extent_map(em);
6237 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6238 u64 logical, u64 *length,
6239 struct btrfs_bio **bbio_ret, int mirror_num)
6241 if (op == BTRFS_MAP_DISCARD)
6242 return __btrfs_map_block_for_discard(fs_info, logical,
6245 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret,
6249 /* For Scrub/replace */
6250 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6251 u64 logical, u64 *length,
6252 struct btrfs_bio **bbio_ret)
6254 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1);
6257 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
6259 bio->bi_private = bbio->private;
6260 bio->bi_end_io = bbio->end_io;
6263 btrfs_put_bbio(bbio);
6266 static void btrfs_end_bio(struct bio *bio)
6268 struct btrfs_bio *bbio = bio->bi_private;
6269 int is_orig_bio = 0;
6271 if (bio->bi_status) {
6272 atomic_inc(&bbio->error);
6273 if (bio->bi_status == BLK_STS_IOERR ||
6274 bio->bi_status == BLK_STS_TARGET) {
6275 struct btrfs_device *dev = btrfs_io_bio(bio)->device;
6278 if (bio_op(bio) == REQ_OP_WRITE)
6279 btrfs_dev_stat_inc_and_print(dev,
6280 BTRFS_DEV_STAT_WRITE_ERRS);
6281 else if (!(bio->bi_opf & REQ_RAHEAD))
6282 btrfs_dev_stat_inc_and_print(dev,
6283 BTRFS_DEV_STAT_READ_ERRS);
6284 if (bio->bi_opf & REQ_PREFLUSH)
6285 btrfs_dev_stat_inc_and_print(dev,
6286 BTRFS_DEV_STAT_FLUSH_ERRS);
6290 if (bio == bbio->orig_bio)
6293 btrfs_bio_counter_dec(bbio->fs_info);
6295 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6298 bio = bbio->orig_bio;
6301 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6302 /* only send an error to the higher layers if it is
6303 * beyond the tolerance of the btrfs bio
6305 if (atomic_read(&bbio->error) > bbio->max_errors) {
6306 bio->bi_status = BLK_STS_IOERR;
6309 * this bio is actually up to date, we didn't
6310 * go over the max number of errors
6312 bio->bi_status = BLK_STS_OK;
6315 btrfs_end_bbio(bbio, bio);
6316 } else if (!is_orig_bio) {
6321 static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio,
6322 u64 physical, struct btrfs_device *dev)
6324 struct btrfs_fs_info *fs_info = bbio->fs_info;
6326 bio->bi_private = bbio;
6327 btrfs_io_bio(bio)->device = dev;
6328 bio->bi_end_io = btrfs_end_bio;
6329 bio->bi_iter.bi_sector = physical >> 9;
6330 btrfs_debug_in_rcu(fs_info,
6331 "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
6332 bio_op(bio), bio->bi_opf, (u64)bio->bi_iter.bi_sector,
6333 (unsigned long)dev->bdev->bd_dev, rcu_str_deref(dev->name),
6334 dev->devid, bio->bi_iter.bi_size);
6335 bio_set_dev(bio, dev->bdev);
6337 btrfs_bio_counter_inc_noblocked(fs_info);
6339 btrfsic_submit_bio(bio);
6342 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6344 atomic_inc(&bbio->error);
6345 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6346 /* Should be the original bio. */
6347 WARN_ON(bio != bbio->orig_bio);
6349 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6350 bio->bi_iter.bi_sector = logical >> 9;
6351 if (atomic_read(&bbio->error) > bbio->max_errors)
6352 bio->bi_status = BLK_STS_IOERR;
6354 bio->bi_status = BLK_STS_OK;
6355 btrfs_end_bbio(bbio, bio);
6359 blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
6362 struct btrfs_device *dev;
6363 struct bio *first_bio = bio;
6364 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
6370 struct btrfs_bio *bbio = NULL;
6372 length = bio->bi_iter.bi_size;
6373 map_length = length;
6375 btrfs_bio_counter_inc_blocked(fs_info);
6376 ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical,
6377 &map_length, &bbio, mirror_num, 1);
6379 btrfs_bio_counter_dec(fs_info);
6380 return errno_to_blk_status(ret);
6383 total_devs = bbio->num_stripes;
6384 bbio->orig_bio = first_bio;
6385 bbio->private = first_bio->bi_private;
6386 bbio->end_io = first_bio->bi_end_io;
6387 bbio->fs_info = fs_info;
6388 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6390 if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6391 ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) {
6392 /* In this case, map_length has been set to the length of
6393 a single stripe; not the whole write */
6394 if (bio_op(bio) == REQ_OP_WRITE) {
6395 ret = raid56_parity_write(fs_info, bio, bbio,
6398 ret = raid56_parity_recover(fs_info, bio, bbio,
6399 map_length, mirror_num, 1);
6402 btrfs_bio_counter_dec(fs_info);
6403 return errno_to_blk_status(ret);
6406 if (map_length < length) {
6408 "mapping failed logical %llu bio len %llu len %llu",
6409 logical, length, map_length);
6413 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6414 dev = bbio->stripes[dev_nr].dev;
6415 if (!dev || !dev->bdev || test_bit(BTRFS_DEV_STATE_MISSING,
6417 (bio_op(first_bio) == REQ_OP_WRITE &&
6418 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) {
6419 bbio_error(bbio, first_bio, logical);
6423 if (dev_nr < total_devs - 1)
6424 bio = btrfs_bio_clone(first_bio);
6428 submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical, dev);
6430 btrfs_bio_counter_dec(fs_info);
6435 * Find a device specified by @devid or @uuid in the list of @fs_devices, or
6438 * If devid and uuid are both specified, the match must be exact, otherwise
6439 * only devid is used.
6441 * If @seed is true, traverse through the seed devices.
6443 struct btrfs_device *btrfs_find_device(struct btrfs_fs_devices *fs_devices,
6444 u64 devid, u8 *uuid, u8 *fsid,
6447 struct btrfs_device *device;
6448 struct btrfs_fs_devices *seed_devs;
6450 if (!fsid || !memcmp(fs_devices->metadata_uuid, fsid, BTRFS_FSID_SIZE)) {
6451 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6452 if (device->devid == devid &&
6453 (!uuid || memcmp(device->uuid, uuid,
6454 BTRFS_UUID_SIZE) == 0))
6459 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
6461 !memcmp(seed_devs->metadata_uuid, fsid, BTRFS_FSID_SIZE)) {
6462 list_for_each_entry(device, &seed_devs->devices,
6464 if (device->devid == devid &&
6465 (!uuid || memcmp(device->uuid, uuid,
6466 BTRFS_UUID_SIZE) == 0))
6475 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices,
6476 u64 devid, u8 *dev_uuid)
6478 struct btrfs_device *device;
6479 unsigned int nofs_flag;
6482 * We call this under the chunk_mutex, so we want to use NOFS for this
6483 * allocation, however we don't want to change btrfs_alloc_device() to
6484 * always do NOFS because we use it in a lot of other GFP_KERNEL safe
6487 nofs_flag = memalloc_nofs_save();
6488 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6489 memalloc_nofs_restore(nofs_flag);
6493 list_add(&device->dev_list, &fs_devices->devices);
6494 device->fs_devices = fs_devices;
6495 fs_devices->num_devices++;
6497 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6498 fs_devices->missing_devices++;
6504 * btrfs_alloc_device - allocate struct btrfs_device
6505 * @fs_info: used only for generating a new devid, can be NULL if
6506 * devid is provided (i.e. @devid != NULL).
6507 * @devid: a pointer to devid for this device. If NULL a new devid
6509 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6512 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6513 * on error. Returned struct is not linked onto any lists and must be
6514 * destroyed with btrfs_free_device.
6516 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6520 struct btrfs_device *dev;
6523 if (WARN_ON(!devid && !fs_info))
6524 return ERR_PTR(-EINVAL);
6526 dev = __alloc_device(fs_info);
6535 ret = find_next_devid(fs_info, &tmp);
6537 btrfs_free_device(dev);
6538 return ERR_PTR(ret);
6544 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6546 generate_random_uuid(dev->uuid);
6551 static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info,
6552 u64 devid, u8 *uuid, bool error)
6555 btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing",
6558 btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing",
6562 static u64 calc_stripe_length(u64 type, u64 chunk_len, int num_stripes)
6564 int index = btrfs_bg_flags_to_raid_index(type);
6565 int ncopies = btrfs_raid_array[index].ncopies;
6566 const int nparity = btrfs_raid_array[index].nparity;
6570 data_stripes = num_stripes - nparity;
6572 data_stripes = num_stripes / ncopies;
6574 return div_u64(chunk_len, data_stripes);
6577 static int read_one_chunk(struct btrfs_key *key, struct extent_buffer *leaf,
6578 struct btrfs_chunk *chunk)
6580 struct btrfs_fs_info *fs_info = leaf->fs_info;
6581 struct extent_map_tree *map_tree = &fs_info->mapping_tree;
6582 struct map_lookup *map;
6583 struct extent_map *em;
6587 u8 uuid[BTRFS_UUID_SIZE];
6592 logical = key->offset;
6593 length = btrfs_chunk_length(leaf, chunk);
6594 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6597 * Only need to verify chunk item if we're reading from sys chunk array,
6598 * as chunk item in tree block is already verified by tree-checker.
6600 if (leaf->start == BTRFS_SUPER_INFO_OFFSET) {
6601 ret = btrfs_check_chunk_valid(leaf, chunk, logical);
6606 read_lock(&map_tree->lock);
6607 em = lookup_extent_mapping(map_tree, logical, 1);
6608 read_unlock(&map_tree->lock);
6610 /* already mapped? */
6611 if (em && em->start <= logical && em->start + em->len > logical) {
6612 free_extent_map(em);
6615 free_extent_map(em);
6618 em = alloc_extent_map();
6621 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6623 free_extent_map(em);
6627 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6628 em->map_lookup = map;
6629 em->start = logical;
6632 em->block_start = 0;
6633 em->block_len = em->len;
6635 map->num_stripes = num_stripes;
6636 map->io_width = btrfs_chunk_io_width(leaf, chunk);
6637 map->io_align = btrfs_chunk_io_align(leaf, chunk);
6638 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6639 map->type = btrfs_chunk_type(leaf, chunk);
6640 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6641 map->verified_stripes = 0;
6642 em->orig_block_len = calc_stripe_length(map->type, em->len,
6644 for (i = 0; i < num_stripes; i++) {
6645 map->stripes[i].physical =
6646 btrfs_stripe_offset_nr(leaf, chunk, i);
6647 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6648 read_extent_buffer(leaf, uuid, (unsigned long)
6649 btrfs_stripe_dev_uuid_nr(chunk, i),
6651 map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices,
6652 devid, uuid, NULL, true);
6653 if (!map->stripes[i].dev &&
6654 !btrfs_test_opt(fs_info, DEGRADED)) {
6655 free_extent_map(em);
6656 btrfs_report_missing_device(fs_info, devid, uuid, true);
6659 if (!map->stripes[i].dev) {
6660 map->stripes[i].dev =
6661 add_missing_dev(fs_info->fs_devices, devid,
6663 if (IS_ERR(map->stripes[i].dev)) {
6664 free_extent_map(em);
6666 "failed to init missing dev %llu: %ld",
6667 devid, PTR_ERR(map->stripes[i].dev));
6668 return PTR_ERR(map->stripes[i].dev);
6670 btrfs_report_missing_device(fs_info, devid, uuid, false);
6672 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
6673 &(map->stripes[i].dev->dev_state));
6677 write_lock(&map_tree->lock);
6678 ret = add_extent_mapping(map_tree, em, 0);
6679 write_unlock(&map_tree->lock);
6682 "failed to add chunk map, start=%llu len=%llu: %d",
6683 em->start, em->len, ret);
6685 free_extent_map(em);
6690 static void fill_device_from_item(struct extent_buffer *leaf,
6691 struct btrfs_dev_item *dev_item,
6692 struct btrfs_device *device)
6696 device->devid = btrfs_device_id(leaf, dev_item);
6697 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6698 device->total_bytes = device->disk_total_bytes;
6699 device->commit_total_bytes = device->disk_total_bytes;
6700 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6701 device->commit_bytes_used = device->bytes_used;
6702 device->type = btrfs_device_type(leaf, dev_item);
6703 device->io_align = btrfs_device_io_align(leaf, dev_item);
6704 device->io_width = btrfs_device_io_width(leaf, dev_item);
6705 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6706 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6707 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
6709 ptr = btrfs_device_uuid(dev_item);
6710 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6713 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info,
6716 struct btrfs_fs_devices *fs_devices;
6719 lockdep_assert_held(&uuid_mutex);
6722 /* This will match only for multi-device seed fs */
6723 list_for_each_entry(fs_devices, &fs_info->fs_devices->seed_list, seed_list)
6724 if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE))
6728 fs_devices = find_fsid(fsid, NULL);
6730 if (!btrfs_test_opt(fs_info, DEGRADED))
6731 return ERR_PTR(-ENOENT);
6733 fs_devices = alloc_fs_devices(fsid, NULL);
6734 if (IS_ERR(fs_devices))
6737 fs_devices->seeding = true;
6738 fs_devices->opened = 1;
6743 * Upon first call for a seed fs fsid, just create a private copy of the
6744 * respective fs_devices and anchor it at fs_info->fs_devices->seed_list
6746 fs_devices = clone_fs_devices(fs_devices);
6747 if (IS_ERR(fs_devices))
6750 ret = open_fs_devices(fs_devices, FMODE_READ, fs_info->bdev_holder);
6752 free_fs_devices(fs_devices);
6753 return ERR_PTR(ret);
6756 if (!fs_devices->seeding) {
6757 close_fs_devices(fs_devices);
6758 free_fs_devices(fs_devices);
6759 return ERR_PTR(-EINVAL);
6762 list_add(&fs_devices->seed_list, &fs_info->fs_devices->seed_list);
6767 static int read_one_dev(struct extent_buffer *leaf,
6768 struct btrfs_dev_item *dev_item)
6770 struct btrfs_fs_info *fs_info = leaf->fs_info;
6771 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6772 struct btrfs_device *device;
6775 u8 fs_uuid[BTRFS_FSID_SIZE];
6776 u8 dev_uuid[BTRFS_UUID_SIZE];
6778 devid = btrfs_device_id(leaf, dev_item);
6779 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6781 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6784 if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) {
6785 fs_devices = open_seed_devices(fs_info, fs_uuid);
6786 if (IS_ERR(fs_devices))
6787 return PTR_ERR(fs_devices);
6790 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
6793 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6794 btrfs_report_missing_device(fs_info, devid,
6799 device = add_missing_dev(fs_devices, devid, dev_uuid);
6800 if (IS_ERR(device)) {
6802 "failed to add missing dev %llu: %ld",
6803 devid, PTR_ERR(device));
6804 return PTR_ERR(device);
6806 btrfs_report_missing_device(fs_info, devid, dev_uuid, false);
6808 if (!device->bdev) {
6809 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6810 btrfs_report_missing_device(fs_info,
6811 devid, dev_uuid, true);
6814 btrfs_report_missing_device(fs_info, devid,
6818 if (!device->bdev &&
6819 !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
6821 * this happens when a device that was properly setup
6822 * in the device info lists suddenly goes bad.
6823 * device->bdev is NULL, and so we have to set
6824 * device->missing to one here
6826 device->fs_devices->missing_devices++;
6827 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6830 /* Move the device to its own fs_devices */
6831 if (device->fs_devices != fs_devices) {
6832 ASSERT(test_bit(BTRFS_DEV_STATE_MISSING,
6833 &device->dev_state));
6835 list_move(&device->dev_list, &fs_devices->devices);
6836 device->fs_devices->num_devices--;
6837 fs_devices->num_devices++;
6839 device->fs_devices->missing_devices--;
6840 fs_devices->missing_devices++;
6842 device->fs_devices = fs_devices;
6846 if (device->fs_devices != fs_info->fs_devices) {
6847 BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state));
6848 if (device->generation !=
6849 btrfs_device_generation(leaf, dev_item))
6853 fill_device_from_item(leaf, dev_item, device);
6854 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
6855 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
6856 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
6857 device->fs_devices->total_rw_bytes += device->total_bytes;
6858 atomic64_add(device->total_bytes - device->bytes_used,
6859 &fs_info->free_chunk_space);
6865 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
6867 struct btrfs_root *root = fs_info->tree_root;
6868 struct btrfs_super_block *super_copy = fs_info->super_copy;
6869 struct extent_buffer *sb;
6870 struct btrfs_disk_key *disk_key;
6871 struct btrfs_chunk *chunk;
6873 unsigned long sb_array_offset;
6880 struct btrfs_key key;
6882 ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize);
6884 * This will create extent buffer of nodesize, superblock size is
6885 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6886 * overallocate but we can keep it as-is, only the first page is used.
6888 sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET);
6891 set_extent_buffer_uptodate(sb);
6892 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6894 * The sb extent buffer is artificial and just used to read the system array.
6895 * set_extent_buffer_uptodate() call does not properly mark all it's
6896 * pages up-to-date when the page is larger: extent does not cover the
6897 * whole page and consequently check_page_uptodate does not find all
6898 * the page's extents up-to-date (the hole beyond sb),
6899 * write_extent_buffer then triggers a WARN_ON.
6901 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6902 * but sb spans only this function. Add an explicit SetPageUptodate call
6903 * to silence the warning eg. on PowerPC 64.
6905 if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
6906 SetPageUptodate(sb->pages[0]);
6908 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6909 array_size = btrfs_super_sys_array_size(super_copy);
6911 array_ptr = super_copy->sys_chunk_array;
6912 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6915 while (cur_offset < array_size) {
6916 disk_key = (struct btrfs_disk_key *)array_ptr;
6917 len = sizeof(*disk_key);
6918 if (cur_offset + len > array_size)
6919 goto out_short_read;
6921 btrfs_disk_key_to_cpu(&key, disk_key);
6924 sb_array_offset += len;
6927 if (key.type != BTRFS_CHUNK_ITEM_KEY) {
6929 "unexpected item type %u in sys_array at offset %u",
6930 (u32)key.type, cur_offset);
6935 chunk = (struct btrfs_chunk *)sb_array_offset;
6937 * At least one btrfs_chunk with one stripe must be present,
6938 * exact stripe count check comes afterwards
6940 len = btrfs_chunk_item_size(1);
6941 if (cur_offset + len > array_size)
6942 goto out_short_read;
6944 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6947 "invalid number of stripes %u in sys_array at offset %u",
6948 num_stripes, cur_offset);
6953 type = btrfs_chunk_type(sb, chunk);
6954 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
6956 "invalid chunk type %llu in sys_array at offset %u",
6962 len = btrfs_chunk_item_size(num_stripes);
6963 if (cur_offset + len > array_size)
6964 goto out_short_read;
6966 ret = read_one_chunk(&key, sb, chunk);
6971 sb_array_offset += len;
6974 clear_extent_buffer_uptodate(sb);
6975 free_extent_buffer_stale(sb);
6979 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
6981 clear_extent_buffer_uptodate(sb);
6982 free_extent_buffer_stale(sb);
6987 * Check if all chunks in the fs are OK for read-write degraded mount
6989 * If the @failing_dev is specified, it's accounted as missing.
6991 * Return true if all chunks meet the minimal RW mount requirements.
6992 * Return false if any chunk doesn't meet the minimal RW mount requirements.
6994 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
6995 struct btrfs_device *failing_dev)
6997 struct extent_map_tree *map_tree = &fs_info->mapping_tree;
6998 struct extent_map *em;
7002 read_lock(&map_tree->lock);
7003 em = lookup_extent_mapping(map_tree, 0, (u64)-1);
7004 read_unlock(&map_tree->lock);
7005 /* No chunk at all? Return false anyway */
7011 struct map_lookup *map;
7016 map = em->map_lookup;
7018 btrfs_get_num_tolerated_disk_barrier_failures(
7020 for (i = 0; i < map->num_stripes; i++) {
7021 struct btrfs_device *dev = map->stripes[i].dev;
7023 if (!dev || !dev->bdev ||
7024 test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) ||
7025 dev->last_flush_error)
7027 else if (failing_dev && failing_dev == dev)
7030 if (missing > max_tolerated) {
7033 "chunk %llu missing %d devices, max tolerance is %d for writable mount",
7034 em->start, missing, max_tolerated);
7035 free_extent_map(em);
7039 next_start = extent_map_end(em);
7040 free_extent_map(em);
7042 read_lock(&map_tree->lock);
7043 em = lookup_extent_mapping(map_tree, next_start,
7044 (u64)(-1) - next_start);
7045 read_unlock(&map_tree->lock);
7051 static void readahead_tree_node_children(struct extent_buffer *node)
7054 const int nr_items = btrfs_header_nritems(node);
7056 for (i = 0; i < nr_items; i++) {
7059 start = btrfs_node_blockptr(node, i);
7060 readahead_tree_block(node->fs_info, start);
7064 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
7066 struct btrfs_root *root = fs_info->chunk_root;
7067 struct btrfs_path *path;
7068 struct extent_buffer *leaf;
7069 struct btrfs_key key;
7070 struct btrfs_key found_key;
7074 u64 last_ra_node = 0;
7076 path = btrfs_alloc_path();
7081 * uuid_mutex is needed only if we are mounting a sprout FS
7082 * otherwise we don't need it.
7084 mutex_lock(&uuid_mutex);
7087 * It is possible for mount and umount to race in such a way that
7088 * we execute this code path, but open_fs_devices failed to clear
7089 * total_rw_bytes. We certainly want it cleared before reading the
7090 * device items, so clear it here.
7092 fs_info->fs_devices->total_rw_bytes = 0;
7095 * Read all device items, and then all the chunk items. All
7096 * device items are found before any chunk item (their object id
7097 * is smaller than the lowest possible object id for a chunk
7098 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
7100 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
7103 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7107 struct extent_buffer *node;
7109 leaf = path->nodes[0];
7110 slot = path->slots[0];
7111 if (slot >= btrfs_header_nritems(leaf)) {
7112 ret = btrfs_next_leaf(root, path);
7120 * The nodes on level 1 are not locked but we don't need to do
7121 * that during mount time as nothing else can access the tree
7123 node = path->nodes[1];
7125 if (last_ra_node != node->start) {
7126 readahead_tree_node_children(node);
7127 last_ra_node = node->start;
7130 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7131 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
7132 struct btrfs_dev_item *dev_item;
7133 dev_item = btrfs_item_ptr(leaf, slot,
7134 struct btrfs_dev_item);
7135 ret = read_one_dev(leaf, dev_item);
7139 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
7140 struct btrfs_chunk *chunk;
7141 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
7142 mutex_lock(&fs_info->chunk_mutex);
7143 ret = read_one_chunk(&found_key, leaf, chunk);
7144 mutex_unlock(&fs_info->chunk_mutex);
7152 * After loading chunk tree, we've got all device information,
7153 * do another round of validation checks.
7155 if (total_dev != fs_info->fs_devices->total_devices) {
7157 "super_num_devices %llu mismatch with num_devices %llu found here",
7158 btrfs_super_num_devices(fs_info->super_copy),
7163 if (btrfs_super_total_bytes(fs_info->super_copy) <
7164 fs_info->fs_devices->total_rw_bytes) {
7166 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
7167 btrfs_super_total_bytes(fs_info->super_copy),
7168 fs_info->fs_devices->total_rw_bytes);
7174 mutex_unlock(&uuid_mutex);
7176 btrfs_free_path(path);
7180 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
7182 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
7183 struct btrfs_device *device;
7185 fs_devices->fs_info = fs_info;
7187 mutex_lock(&fs_devices->device_list_mutex);
7188 list_for_each_entry(device, &fs_devices->devices, dev_list)
7189 device->fs_info = fs_info;
7191 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
7192 list_for_each_entry(device, &seed_devs->devices, dev_list)
7193 device->fs_info = fs_info;
7195 seed_devs->fs_info = fs_info;
7197 mutex_unlock(&fs_devices->device_list_mutex);
7200 static u64 btrfs_dev_stats_value(const struct extent_buffer *eb,
7201 const struct btrfs_dev_stats_item *ptr,
7206 read_extent_buffer(eb, &val,
7207 offsetof(struct btrfs_dev_stats_item, values) +
7208 ((unsigned long)ptr) + (index * sizeof(u64)),
7213 static void btrfs_set_dev_stats_value(struct extent_buffer *eb,
7214 struct btrfs_dev_stats_item *ptr,
7217 write_extent_buffer(eb, &val,
7218 offsetof(struct btrfs_dev_stats_item, values) +
7219 ((unsigned long)ptr) + (index * sizeof(u64)),
7223 static int btrfs_device_init_dev_stats(struct btrfs_device *device,
7224 struct btrfs_path *path)
7226 struct btrfs_dev_stats_item *ptr;
7227 struct extent_buffer *eb;
7228 struct btrfs_key key;
7232 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7233 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7234 key.offset = device->devid;
7235 ret = btrfs_search_slot(NULL, device->fs_info->dev_root, &key, path, 0, 0);
7237 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7238 btrfs_dev_stat_set(device, i, 0);
7239 device->dev_stats_valid = 1;
7240 btrfs_release_path(path);
7241 return ret < 0 ? ret : 0;
7243 slot = path->slots[0];
7244 eb = path->nodes[0];
7245 item_size = btrfs_item_size_nr(eb, slot);
7247 ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_stats_item);
7249 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7250 if (item_size >= (1 + i) * sizeof(__le64))
7251 btrfs_dev_stat_set(device, i,
7252 btrfs_dev_stats_value(eb, ptr, i));
7254 btrfs_dev_stat_set(device, i, 0);
7257 device->dev_stats_valid = 1;
7258 btrfs_dev_stat_print_on_load(device);
7259 btrfs_release_path(path);
7264 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
7266 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
7267 struct btrfs_device *device;
7268 struct btrfs_path *path = NULL;
7271 path = btrfs_alloc_path();
7275 mutex_lock(&fs_devices->device_list_mutex);
7276 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7277 ret = btrfs_device_init_dev_stats(device, path);
7281 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
7282 list_for_each_entry(device, &seed_devs->devices, dev_list) {
7283 ret = btrfs_device_init_dev_stats(device, path);
7289 mutex_unlock(&fs_devices->device_list_mutex);
7291 btrfs_free_path(path);
7295 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
7296 struct btrfs_device *device)
7298 struct btrfs_fs_info *fs_info = trans->fs_info;
7299 struct btrfs_root *dev_root = fs_info->dev_root;
7300 struct btrfs_path *path;
7301 struct btrfs_key key;
7302 struct extent_buffer *eb;
7303 struct btrfs_dev_stats_item *ptr;
7307 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7308 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7309 key.offset = device->devid;
7311 path = btrfs_alloc_path();
7314 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
7316 btrfs_warn_in_rcu(fs_info,
7317 "error %d while searching for dev_stats item for device %s",
7318 ret, rcu_str_deref(device->name));
7323 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
7324 /* need to delete old one and insert a new one */
7325 ret = btrfs_del_item(trans, dev_root, path);
7327 btrfs_warn_in_rcu(fs_info,
7328 "delete too small dev_stats item for device %s failed %d",
7329 rcu_str_deref(device->name), ret);
7336 /* need to insert a new item */
7337 btrfs_release_path(path);
7338 ret = btrfs_insert_empty_item(trans, dev_root, path,
7339 &key, sizeof(*ptr));
7341 btrfs_warn_in_rcu(fs_info,
7342 "insert dev_stats item for device %s failed %d",
7343 rcu_str_deref(device->name), ret);
7348 eb = path->nodes[0];
7349 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7350 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7351 btrfs_set_dev_stats_value(eb, ptr, i,
7352 btrfs_dev_stat_read(device, i));
7353 btrfs_mark_buffer_dirty(eb);
7356 btrfs_free_path(path);
7361 * called from commit_transaction. Writes all changed device stats to disk.
7363 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans)
7365 struct btrfs_fs_info *fs_info = trans->fs_info;
7366 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7367 struct btrfs_device *device;
7371 mutex_lock(&fs_devices->device_list_mutex);
7372 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7373 stats_cnt = atomic_read(&device->dev_stats_ccnt);
7374 if (!device->dev_stats_valid || stats_cnt == 0)
7379 * There is a LOAD-LOAD control dependency between the value of
7380 * dev_stats_ccnt and updating the on-disk values which requires
7381 * reading the in-memory counters. Such control dependencies
7382 * require explicit read memory barriers.
7384 * This memory barriers pairs with smp_mb__before_atomic in
7385 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full
7386 * barrier implied by atomic_xchg in
7387 * btrfs_dev_stats_read_and_reset
7391 ret = update_dev_stat_item(trans, device);
7393 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7395 mutex_unlock(&fs_devices->device_list_mutex);
7400 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7402 btrfs_dev_stat_inc(dev, index);
7403 btrfs_dev_stat_print_on_error(dev);
7406 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
7408 if (!dev->dev_stats_valid)
7410 btrfs_err_rl_in_rcu(dev->fs_info,
7411 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7412 rcu_str_deref(dev->name),
7413 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7414 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7415 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7416 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7417 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7420 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7424 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7425 if (btrfs_dev_stat_read(dev, i) != 0)
7427 if (i == BTRFS_DEV_STAT_VALUES_MAX)
7428 return; /* all values == 0, suppress message */
7430 btrfs_info_in_rcu(dev->fs_info,
7431 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7432 rcu_str_deref(dev->name),
7433 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7434 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7435 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7436 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7437 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7440 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
7441 struct btrfs_ioctl_get_dev_stats *stats)
7443 struct btrfs_device *dev;
7444 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7447 mutex_lock(&fs_devices->device_list_mutex);
7448 dev = btrfs_find_device(fs_info->fs_devices, stats->devid, NULL, NULL,
7450 mutex_unlock(&fs_devices->device_list_mutex);
7453 btrfs_warn(fs_info, "get dev_stats failed, device not found");
7455 } else if (!dev->dev_stats_valid) {
7456 btrfs_warn(fs_info, "get dev_stats failed, not yet valid");
7458 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7459 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7460 if (stats->nr_items > i)
7462 btrfs_dev_stat_read_and_reset(dev, i);
7464 btrfs_dev_stat_set(dev, i, 0);
7466 btrfs_info(fs_info, "device stats zeroed by %s (%d)",
7467 current->comm, task_pid_nr(current));
7469 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7470 if (stats->nr_items > i)
7471 stats->values[i] = btrfs_dev_stat_read(dev, i);
7473 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7474 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7479 * Update the size and bytes used for each device where it changed. This is
7480 * delayed since we would otherwise get errors while writing out the
7483 * Must be invoked during transaction commit.
7485 void btrfs_commit_device_sizes(struct btrfs_transaction *trans)
7487 struct btrfs_device *curr, *next;
7489 ASSERT(trans->state == TRANS_STATE_COMMIT_DOING);
7491 if (list_empty(&trans->dev_update_list))
7495 * We don't need the device_list_mutex here. This list is owned by the
7496 * transaction and the transaction must complete before the device is
7499 mutex_lock(&trans->fs_info->chunk_mutex);
7500 list_for_each_entry_safe(curr, next, &trans->dev_update_list,
7502 list_del_init(&curr->post_commit_list);
7503 curr->commit_total_bytes = curr->disk_total_bytes;
7504 curr->commit_bytes_used = curr->bytes_used;
7506 mutex_unlock(&trans->fs_info->chunk_mutex);
7510 * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10.
7512 int btrfs_bg_type_to_factor(u64 flags)
7514 const int index = btrfs_bg_flags_to_raid_index(flags);
7516 return btrfs_raid_array[index].ncopies;
7521 static int verify_one_dev_extent(struct btrfs_fs_info *fs_info,
7522 u64 chunk_offset, u64 devid,
7523 u64 physical_offset, u64 physical_len)
7525 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
7526 struct extent_map *em;
7527 struct map_lookup *map;
7528 struct btrfs_device *dev;
7534 read_lock(&em_tree->lock);
7535 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
7536 read_unlock(&em_tree->lock);
7540 "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk",
7541 physical_offset, devid);
7546 map = em->map_lookup;
7547 stripe_len = calc_stripe_length(map->type, em->len, map->num_stripes);
7548 if (physical_len != stripe_len) {
7550 "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu",
7551 physical_offset, devid, em->start, physical_len,
7557 for (i = 0; i < map->num_stripes; i++) {
7558 if (map->stripes[i].dev->devid == devid &&
7559 map->stripes[i].physical == physical_offset) {
7561 if (map->verified_stripes >= map->num_stripes) {
7563 "too many dev extents for chunk %llu found",
7568 map->verified_stripes++;
7574 "dev extent physical offset %llu devid %llu has no corresponding chunk",
7575 physical_offset, devid);
7579 /* Make sure no dev extent is beyond device bondary */
7580 dev = btrfs_find_device(fs_info->fs_devices, devid, NULL, NULL, true);
7582 btrfs_err(fs_info, "failed to find devid %llu", devid);
7587 /* It's possible this device is a dummy for seed device */
7588 if (dev->disk_total_bytes == 0) {
7589 struct btrfs_fs_devices *devs;
7591 devs = list_first_entry(&fs_info->fs_devices->seed_list,
7592 struct btrfs_fs_devices, seed_list);
7593 dev = btrfs_find_device(devs, devid, NULL, NULL, false);
7595 btrfs_err(fs_info, "failed to find seed devid %llu",
7602 if (physical_offset + physical_len > dev->disk_total_bytes) {
7604 "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu",
7605 devid, physical_offset, physical_len,
7606 dev->disk_total_bytes);
7611 free_extent_map(em);
7615 static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info)
7617 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
7618 struct extent_map *em;
7619 struct rb_node *node;
7622 read_lock(&em_tree->lock);
7623 for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) {
7624 em = rb_entry(node, struct extent_map, rb_node);
7625 if (em->map_lookup->num_stripes !=
7626 em->map_lookup->verified_stripes) {
7628 "chunk %llu has missing dev extent, have %d expect %d",
7629 em->start, em->map_lookup->verified_stripes,
7630 em->map_lookup->num_stripes);
7636 read_unlock(&em_tree->lock);
7641 * Ensure that all dev extents are mapped to correct chunk, otherwise
7642 * later chunk allocation/free would cause unexpected behavior.
7644 * NOTE: This will iterate through the whole device tree, which should be of
7645 * the same size level as the chunk tree. This slightly increases mount time.
7647 int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info)
7649 struct btrfs_path *path;
7650 struct btrfs_root *root = fs_info->dev_root;
7651 struct btrfs_key key;
7653 u64 prev_dev_ext_end = 0;
7657 key.type = BTRFS_DEV_EXTENT_KEY;
7660 path = btrfs_alloc_path();
7664 path->reada = READA_FORWARD;
7665 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7669 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
7670 ret = btrfs_next_item(root, path);
7673 /* No dev extents at all? Not good */
7680 struct extent_buffer *leaf = path->nodes[0];
7681 struct btrfs_dev_extent *dext;
7682 int slot = path->slots[0];
7684 u64 physical_offset;
7688 btrfs_item_key_to_cpu(leaf, &key, slot);
7689 if (key.type != BTRFS_DEV_EXTENT_KEY)
7691 devid = key.objectid;
7692 physical_offset = key.offset;
7694 dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent);
7695 chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext);
7696 physical_len = btrfs_dev_extent_length(leaf, dext);
7698 /* Check if this dev extent overlaps with the previous one */
7699 if (devid == prev_devid && physical_offset < prev_dev_ext_end) {
7701 "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu",
7702 devid, physical_offset, prev_dev_ext_end);
7707 ret = verify_one_dev_extent(fs_info, chunk_offset, devid,
7708 physical_offset, physical_len);
7712 prev_dev_ext_end = physical_offset + physical_len;
7714 ret = btrfs_next_item(root, path);
7723 /* Ensure all chunks have corresponding dev extents */
7724 ret = verify_chunk_dev_extent_mapping(fs_info);
7726 btrfs_free_path(path);
7731 * Check whether the given block group or device is pinned by any inode being
7732 * used as a swapfile.
7734 bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr)
7736 struct btrfs_swapfile_pin *sp;
7737 struct rb_node *node;
7739 spin_lock(&fs_info->swapfile_pins_lock);
7740 node = fs_info->swapfile_pins.rb_node;
7742 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
7744 node = node->rb_left;
7745 else if (ptr > sp->ptr)
7746 node = node->rb_right;
7750 spin_unlock(&fs_info->swapfile_pins_lock);
7751 return node != NULL;