1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
6 #include <linux/sched.h>
7 #include <linux/sched/mm.h>
9 #include <linux/slab.h>
10 #include <linux/blkdev.h>
11 #include <linux/ratelimit.h>
12 #include <linux/kthread.h>
13 #include <linux/raid/pq.h>
14 #include <linux/semaphore.h>
15 #include <linux/uuid.h>
16 #include <linux/list_sort.h>
19 #include "extent_map.h"
21 #include "transaction.h"
22 #include "print-tree.h"
25 #include "async-thread.h"
26 #include "check-integrity.h"
27 #include "rcu-string.h"
28 #include "dev-replace.h"
30 #include "tree-checker.h"
31 #include "space-info.h"
32 #include "block-group.h"
36 const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
37 [BTRFS_RAID_RAID10] = {
40 .devs_max = 0, /* 0 == as many as possible */
42 .tolerated_failures = 1,
46 .raid_name = "raid10",
47 .bg_flag = BTRFS_BLOCK_GROUP_RAID10,
48 .mindev_error = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET,
50 [BTRFS_RAID_RAID1] = {
55 .tolerated_failures = 1,
60 .bg_flag = BTRFS_BLOCK_GROUP_RAID1,
61 .mindev_error = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET,
63 [BTRFS_RAID_RAID1C3] = {
68 .tolerated_failures = 2,
72 .raid_name = "raid1c3",
73 .bg_flag = BTRFS_BLOCK_GROUP_RAID1C3,
74 .mindev_error = BTRFS_ERROR_DEV_RAID1C3_MIN_NOT_MET,
76 [BTRFS_RAID_RAID1C4] = {
81 .tolerated_failures = 3,
85 .raid_name = "raid1c4",
86 .bg_flag = BTRFS_BLOCK_GROUP_RAID1C4,
87 .mindev_error = BTRFS_ERROR_DEV_RAID1C4_MIN_NOT_MET,
94 .tolerated_failures = 0,
99 .bg_flag = BTRFS_BLOCK_GROUP_DUP,
102 [BTRFS_RAID_RAID0] = {
107 .tolerated_failures = 0,
111 .raid_name = "raid0",
112 .bg_flag = BTRFS_BLOCK_GROUP_RAID0,
115 [BTRFS_RAID_SINGLE] = {
120 .tolerated_failures = 0,
124 .raid_name = "single",
128 [BTRFS_RAID_RAID5] = {
133 .tolerated_failures = 1,
137 .raid_name = "raid5",
138 .bg_flag = BTRFS_BLOCK_GROUP_RAID5,
139 .mindev_error = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET,
141 [BTRFS_RAID_RAID6] = {
146 .tolerated_failures = 2,
150 .raid_name = "raid6",
151 .bg_flag = BTRFS_BLOCK_GROUP_RAID6,
152 .mindev_error = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET,
156 const char *btrfs_bg_type_to_raid_name(u64 flags)
158 const int index = btrfs_bg_flags_to_raid_index(flags);
160 if (index >= BTRFS_NR_RAID_TYPES)
163 return btrfs_raid_array[index].raid_name;
167 * Fill @buf with textual description of @bg_flags, no more than @size_buf
168 * bytes including terminating null byte.
170 void btrfs_describe_block_groups(u64 bg_flags, char *buf, u32 size_buf)
175 u64 flags = bg_flags;
176 u32 size_bp = size_buf;
183 #define DESCRIBE_FLAG(flag, desc) \
185 if (flags & (flag)) { \
186 ret = snprintf(bp, size_bp, "%s|", (desc)); \
187 if (ret < 0 || ret >= size_bp) \
195 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_DATA, "data");
196 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_SYSTEM, "system");
197 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_METADATA, "metadata");
199 DESCRIBE_FLAG(BTRFS_AVAIL_ALLOC_BIT_SINGLE, "single");
200 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
201 DESCRIBE_FLAG(btrfs_raid_array[i].bg_flag,
202 btrfs_raid_array[i].raid_name);
206 ret = snprintf(bp, size_bp, "0x%llx|", flags);
210 if (size_bp < size_buf)
211 buf[size_buf - size_bp - 1] = '\0'; /* remove last | */
214 * The text is trimmed, it's up to the caller to provide sufficiently
220 static int init_first_rw_device(struct btrfs_trans_handle *trans);
221 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info);
222 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
223 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
224 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
225 enum btrfs_map_op op,
226 u64 logical, u64 *length,
227 struct btrfs_bio **bbio_ret,
228 int mirror_num, int need_raid_map);
234 * There are several mutexes that protect manipulation of devices and low-level
235 * structures like chunks but not block groups, extents or files
237 * uuid_mutex (global lock)
238 * ------------------------
239 * protects the fs_uuids list that tracks all per-fs fs_devices, resulting from
240 * the SCAN_DEV ioctl registration or from mount either implicitly (the first
241 * device) or requested by the device= mount option
243 * the mutex can be very coarse and can cover long-running operations
245 * protects: updates to fs_devices counters like missing devices, rw devices,
246 * seeding, structure cloning, opening/closing devices at mount/umount time
248 * global::fs_devs - add, remove, updates to the global list
250 * does not protect: manipulation of the fs_devices::devices list in general
251 * but in mount context it could be used to exclude list modifications by eg.
254 * btrfs_device::name - renames (write side), read is RCU
256 * fs_devices::device_list_mutex (per-fs, with RCU)
257 * ------------------------------------------------
258 * protects updates to fs_devices::devices, ie. adding and deleting
260 * simple list traversal with read-only actions can be done with RCU protection
262 * may be used to exclude some operations from running concurrently without any
263 * modifications to the list (see write_all_supers)
265 * Is not required at mount and close times, because our device list is
266 * protected by the uuid_mutex at that point.
270 * protects balance structures (status, state) and context accessed from
271 * several places (internally, ioctl)
275 * protects chunks, adding or removing during allocation, trim or when a new
276 * device is added/removed. Additionally it also protects post_commit_list of
277 * individual devices, since they can be added to the transaction's
278 * post_commit_list only with chunk_mutex held.
282 * a big lock that is held by the cleaner thread and prevents running subvolume
283 * cleaning together with relocation or delayed iputs
295 * Exclusive operations
296 * ====================
298 * Maintains the exclusivity of the following operations that apply to the
299 * whole filesystem and cannot run in parallel.
304 * - Device replace (*)
307 * The device operations (as above) can be in one of the following states:
313 * Only device operations marked with (*) can go into the Paused state for the
316 * - ioctl (only Balance can be Paused through ioctl)
317 * - filesystem remounted as read-only
318 * - filesystem unmounted and mounted as read-only
319 * - system power-cycle and filesystem mounted as read-only
320 * - filesystem or device errors leading to forced read-only
322 * The status of exclusive operation is set and cleared atomically.
323 * During the course of Paused state, fs_info::exclusive_operation remains set.
324 * A device operation in Paused or Running state can be canceled or resumed
325 * either by ioctl (Balance only) or when remounted as read-write.
326 * The exclusive status is cleared when the device operation is canceled or
330 DEFINE_MUTEX(uuid_mutex);
331 static LIST_HEAD(fs_uuids);
332 struct list_head * __attribute_const__ btrfs_get_fs_uuids(void)
338 * alloc_fs_devices - allocate struct btrfs_fs_devices
339 * @fsid: if not NULL, copy the UUID to fs_devices::fsid
340 * @metadata_fsid: if not NULL, copy the UUID to fs_devices::metadata_fsid
342 * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR().
343 * The returned struct is not linked onto any lists and can be destroyed with
344 * kfree() right away.
346 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid,
347 const u8 *metadata_fsid)
349 struct btrfs_fs_devices *fs_devs;
351 fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL);
353 return ERR_PTR(-ENOMEM);
355 mutex_init(&fs_devs->device_list_mutex);
357 INIT_LIST_HEAD(&fs_devs->devices);
358 INIT_LIST_HEAD(&fs_devs->alloc_list);
359 INIT_LIST_HEAD(&fs_devs->fs_list);
360 INIT_LIST_HEAD(&fs_devs->seed_list);
362 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
365 memcpy(fs_devs->metadata_uuid, metadata_fsid, BTRFS_FSID_SIZE);
367 memcpy(fs_devs->metadata_uuid, fsid, BTRFS_FSID_SIZE);
372 void btrfs_free_device(struct btrfs_device *device)
374 WARN_ON(!list_empty(&device->post_commit_list));
375 rcu_string_free(device->name);
376 extent_io_tree_release(&device->alloc_state);
377 bio_put(device->flush_bio);
378 btrfs_destroy_dev_zone_info(device);
382 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
384 struct btrfs_device *device;
385 WARN_ON(fs_devices->opened);
386 while (!list_empty(&fs_devices->devices)) {
387 device = list_entry(fs_devices->devices.next,
388 struct btrfs_device, dev_list);
389 list_del(&device->dev_list);
390 btrfs_free_device(device);
395 void __exit btrfs_cleanup_fs_uuids(void)
397 struct btrfs_fs_devices *fs_devices;
399 while (!list_empty(&fs_uuids)) {
400 fs_devices = list_entry(fs_uuids.next,
401 struct btrfs_fs_devices, fs_list);
402 list_del(&fs_devices->fs_list);
403 free_fs_devices(fs_devices);
408 * Returns a pointer to a new btrfs_device on success; ERR_PTR() on error.
409 * Returned struct is not linked onto any lists and must be destroyed using
412 static struct btrfs_device *__alloc_device(struct btrfs_fs_info *fs_info)
414 struct btrfs_device *dev;
416 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
418 return ERR_PTR(-ENOMEM);
421 * Preallocate a bio that's always going to be used for flushing device
422 * barriers and matches the device lifespan
424 dev->flush_bio = bio_alloc_bioset(GFP_KERNEL, 0, NULL);
425 if (!dev->flush_bio) {
427 return ERR_PTR(-ENOMEM);
430 INIT_LIST_HEAD(&dev->dev_list);
431 INIT_LIST_HEAD(&dev->dev_alloc_list);
432 INIT_LIST_HEAD(&dev->post_commit_list);
434 atomic_set(&dev->reada_in_flight, 0);
435 atomic_set(&dev->dev_stats_ccnt, 0);
436 btrfs_device_data_ordered_init(dev, fs_info);
437 INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
438 INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
439 extent_io_tree_init(fs_info, &dev->alloc_state,
440 IO_TREE_DEVICE_ALLOC_STATE, NULL);
445 static noinline struct btrfs_fs_devices *find_fsid(
446 const u8 *fsid, const u8 *metadata_fsid)
448 struct btrfs_fs_devices *fs_devices;
452 /* Handle non-split brain cases */
453 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
455 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0
456 && memcmp(metadata_fsid, fs_devices->metadata_uuid,
457 BTRFS_FSID_SIZE) == 0)
460 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
467 static struct btrfs_fs_devices *find_fsid_with_metadata_uuid(
468 struct btrfs_super_block *disk_super)
471 struct btrfs_fs_devices *fs_devices;
474 * Handle scanned device having completed its fsid change but
475 * belonging to a fs_devices that was created by first scanning
476 * a device which didn't have its fsid/metadata_uuid changed
477 * at all and the CHANGING_FSID_V2 flag set.
479 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
480 if (fs_devices->fsid_change &&
481 memcmp(disk_super->metadata_uuid, fs_devices->fsid,
482 BTRFS_FSID_SIZE) == 0 &&
483 memcmp(fs_devices->fsid, fs_devices->metadata_uuid,
484 BTRFS_FSID_SIZE) == 0) {
489 * Handle scanned device having completed its fsid change but
490 * belonging to a fs_devices that was created by a device that
491 * has an outdated pair of fsid/metadata_uuid and
492 * CHANGING_FSID_V2 flag set.
494 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
495 if (fs_devices->fsid_change &&
496 memcmp(fs_devices->metadata_uuid,
497 fs_devices->fsid, BTRFS_FSID_SIZE) != 0 &&
498 memcmp(disk_super->metadata_uuid, fs_devices->metadata_uuid,
499 BTRFS_FSID_SIZE) == 0) {
504 return find_fsid(disk_super->fsid, disk_super->metadata_uuid);
509 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
510 int flush, struct block_device **bdev,
511 struct btrfs_super_block **disk_super)
515 *bdev = blkdev_get_by_path(device_path, flags, holder);
518 ret = PTR_ERR(*bdev);
523 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
524 ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE);
526 blkdev_put(*bdev, flags);
529 invalidate_bdev(*bdev);
530 *disk_super = btrfs_read_dev_super(*bdev);
531 if (IS_ERR(*disk_super)) {
532 ret = PTR_ERR(*disk_super);
533 blkdev_put(*bdev, flags);
544 static bool device_path_matched(const char *path, struct btrfs_device *device)
549 found = strcmp(rcu_str_deref(device->name), path);
556 * Search and remove all stale (devices which are not mounted) devices.
557 * When both inputs are NULL, it will search and release all stale devices.
558 * path: Optional. When provided will it release all unmounted devices
559 * matching this path only.
560 * skip_dev: Optional. Will skip this device when searching for the stale
562 * Return: 0 for success or if @path is NULL.
563 * -EBUSY if @path is a mounted device.
564 * -ENOENT if @path does not match any device in the list.
566 static int btrfs_free_stale_devices(const char *path,
567 struct btrfs_device *skip_device)
569 struct btrfs_fs_devices *fs_devices, *tmp_fs_devices;
570 struct btrfs_device *device, *tmp_device;
576 list_for_each_entry_safe(fs_devices, tmp_fs_devices, &fs_uuids, fs_list) {
578 mutex_lock(&fs_devices->device_list_mutex);
579 list_for_each_entry_safe(device, tmp_device,
580 &fs_devices->devices, dev_list) {
581 if (skip_device && skip_device == device)
583 if (path && !device->name)
585 if (path && !device_path_matched(path, device))
587 if (fs_devices->opened) {
588 /* for an already deleted device return 0 */
589 if (path && ret != 0)
594 /* delete the stale device */
595 fs_devices->num_devices--;
596 list_del(&device->dev_list);
597 btrfs_free_device(device);
601 mutex_unlock(&fs_devices->device_list_mutex);
603 if (fs_devices->num_devices == 0) {
604 btrfs_sysfs_remove_fsid(fs_devices);
605 list_del(&fs_devices->fs_list);
606 free_fs_devices(fs_devices);
614 * This is only used on mount, and we are protected from competing things
615 * messing with our fs_devices by the uuid_mutex, thus we do not need the
616 * fs_devices->device_list_mutex here.
618 static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices,
619 struct btrfs_device *device, fmode_t flags,
622 struct request_queue *q;
623 struct block_device *bdev;
624 struct btrfs_super_block *disk_super;
633 ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
638 devid = btrfs_stack_device_id(&disk_super->dev_item);
639 if (devid != device->devid)
640 goto error_free_page;
642 if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE))
643 goto error_free_page;
645 device->generation = btrfs_super_generation(disk_super);
647 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
648 if (btrfs_super_incompat_flags(disk_super) &
649 BTRFS_FEATURE_INCOMPAT_METADATA_UUID) {
651 "BTRFS: Invalid seeding and uuid-changed device detected\n");
652 goto error_free_page;
655 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
656 fs_devices->seeding = true;
658 if (bdev_read_only(bdev))
659 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
661 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
664 q = bdev_get_queue(bdev);
665 if (!blk_queue_nonrot(q))
666 fs_devices->rotating = true;
669 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
670 device->mode = flags;
672 ret = btrfs_get_dev_zone_info(device);
674 goto error_free_page;
676 fs_devices->open_devices++;
677 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
678 device->devid != BTRFS_DEV_REPLACE_DEVID) {
679 fs_devices->rw_devices++;
680 list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list);
682 btrfs_release_disk_super(disk_super);
687 btrfs_release_disk_super(disk_super);
688 blkdev_put(bdev, flags);
694 * Handle scanned device having its CHANGING_FSID_V2 flag set and the fs_devices
695 * being created with a disk that has already completed its fsid change. Such
696 * disk can belong to an fs which has its FSID changed or to one which doesn't.
697 * Handle both cases here.
699 static struct btrfs_fs_devices *find_fsid_inprogress(
700 struct btrfs_super_block *disk_super)
702 struct btrfs_fs_devices *fs_devices;
704 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
705 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid,
706 BTRFS_FSID_SIZE) != 0 &&
707 memcmp(fs_devices->metadata_uuid, disk_super->fsid,
708 BTRFS_FSID_SIZE) == 0 && !fs_devices->fsid_change) {
713 return find_fsid(disk_super->fsid, NULL);
717 static struct btrfs_fs_devices *find_fsid_changed(
718 struct btrfs_super_block *disk_super)
720 struct btrfs_fs_devices *fs_devices;
723 * Handles the case where scanned device is part of an fs that had
724 * multiple successful changes of FSID but curently device didn't
725 * observe it. Meaning our fsid will be different than theirs. We need
726 * to handle two subcases :
727 * 1 - The fs still continues to have different METADATA/FSID uuids.
728 * 2 - The fs is switched back to its original FSID (METADATA/FSID
731 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
733 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid,
734 BTRFS_FSID_SIZE) != 0 &&
735 memcmp(fs_devices->metadata_uuid, disk_super->metadata_uuid,
736 BTRFS_FSID_SIZE) == 0 &&
737 memcmp(fs_devices->fsid, disk_super->fsid,
738 BTRFS_FSID_SIZE) != 0)
741 /* Unchanged UUIDs */
742 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid,
743 BTRFS_FSID_SIZE) == 0 &&
744 memcmp(fs_devices->fsid, disk_super->metadata_uuid,
745 BTRFS_FSID_SIZE) == 0)
752 static struct btrfs_fs_devices *find_fsid_reverted_metadata(
753 struct btrfs_super_block *disk_super)
755 struct btrfs_fs_devices *fs_devices;
758 * Handle the case where the scanned device is part of an fs whose last
759 * metadata UUID change reverted it to the original FSID. At the same
760 * time * fs_devices was first created by another constitutent device
761 * which didn't fully observe the operation. This results in an
762 * btrfs_fs_devices created with metadata/fsid different AND
763 * btrfs_fs_devices::fsid_change set AND the metadata_uuid of the
764 * fs_devices equal to the FSID of the disk.
766 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
767 if (memcmp(fs_devices->fsid, fs_devices->metadata_uuid,
768 BTRFS_FSID_SIZE) != 0 &&
769 memcmp(fs_devices->metadata_uuid, disk_super->fsid,
770 BTRFS_FSID_SIZE) == 0 &&
771 fs_devices->fsid_change)
778 * Add new device to list of registered devices
781 * device pointer which was just added or updated when successful
782 * error pointer when failed
784 static noinline struct btrfs_device *device_list_add(const char *path,
785 struct btrfs_super_block *disk_super,
786 bool *new_device_added)
788 struct btrfs_device *device;
789 struct btrfs_fs_devices *fs_devices = NULL;
790 struct rcu_string *name;
791 u64 found_transid = btrfs_super_generation(disk_super);
792 u64 devid = btrfs_stack_device_id(&disk_super->dev_item);
793 bool has_metadata_uuid = (btrfs_super_incompat_flags(disk_super) &
794 BTRFS_FEATURE_INCOMPAT_METADATA_UUID);
795 bool fsid_change_in_progress = (btrfs_super_flags(disk_super) &
796 BTRFS_SUPER_FLAG_CHANGING_FSID_V2);
798 if (fsid_change_in_progress) {
799 if (!has_metadata_uuid)
800 fs_devices = find_fsid_inprogress(disk_super);
802 fs_devices = find_fsid_changed(disk_super);
803 } else if (has_metadata_uuid) {
804 fs_devices = find_fsid_with_metadata_uuid(disk_super);
806 fs_devices = find_fsid_reverted_metadata(disk_super);
808 fs_devices = find_fsid(disk_super->fsid, NULL);
813 if (has_metadata_uuid)
814 fs_devices = alloc_fs_devices(disk_super->fsid,
815 disk_super->metadata_uuid);
817 fs_devices = alloc_fs_devices(disk_super->fsid, NULL);
819 if (IS_ERR(fs_devices))
820 return ERR_CAST(fs_devices);
822 fs_devices->fsid_change = fsid_change_in_progress;
824 mutex_lock(&fs_devices->device_list_mutex);
825 list_add(&fs_devices->fs_list, &fs_uuids);
829 mutex_lock(&fs_devices->device_list_mutex);
830 device = btrfs_find_device(fs_devices, devid,
831 disk_super->dev_item.uuid, NULL);
834 * If this disk has been pulled into an fs devices created by
835 * a device which had the CHANGING_FSID_V2 flag then replace the
836 * metadata_uuid/fsid values of the fs_devices.
838 if (fs_devices->fsid_change &&
839 found_transid > fs_devices->latest_generation) {
840 memcpy(fs_devices->fsid, disk_super->fsid,
843 if (has_metadata_uuid)
844 memcpy(fs_devices->metadata_uuid,
845 disk_super->metadata_uuid,
848 memcpy(fs_devices->metadata_uuid,
849 disk_super->fsid, BTRFS_FSID_SIZE);
851 fs_devices->fsid_change = false;
856 if (fs_devices->opened) {
857 mutex_unlock(&fs_devices->device_list_mutex);
858 return ERR_PTR(-EBUSY);
861 device = btrfs_alloc_device(NULL, &devid,
862 disk_super->dev_item.uuid);
863 if (IS_ERR(device)) {
864 mutex_unlock(&fs_devices->device_list_mutex);
865 /* we can safely leave the fs_devices entry around */
869 name = rcu_string_strdup(path, GFP_NOFS);
871 btrfs_free_device(device);
872 mutex_unlock(&fs_devices->device_list_mutex);
873 return ERR_PTR(-ENOMEM);
875 rcu_assign_pointer(device->name, name);
877 list_add_rcu(&device->dev_list, &fs_devices->devices);
878 fs_devices->num_devices++;
880 device->fs_devices = fs_devices;
881 *new_device_added = true;
883 if (disk_super->label[0])
885 "BTRFS: device label %s devid %llu transid %llu %s scanned by %s (%d)\n",
886 disk_super->label, devid, found_transid, path,
887 current->comm, task_pid_nr(current));
890 "BTRFS: device fsid %pU devid %llu transid %llu %s scanned by %s (%d)\n",
891 disk_super->fsid, devid, found_transid, path,
892 current->comm, task_pid_nr(current));
894 } else if (!device->name || strcmp(device->name->str, path)) {
896 * When FS is already mounted.
897 * 1. If you are here and if the device->name is NULL that
898 * means this device was missing at time of FS mount.
899 * 2. If you are here and if the device->name is different
900 * from 'path' that means either
901 * a. The same device disappeared and reappeared with
903 * b. The missing-disk-which-was-replaced, has
906 * We must allow 1 and 2a above. But 2b would be a spurious
909 * Further in case of 1 and 2a above, the disk at 'path'
910 * would have missed some transaction when it was away and
911 * in case of 2a the stale bdev has to be updated as well.
912 * 2b must not be allowed at all time.
916 * For now, we do allow update to btrfs_fs_device through the
917 * btrfs dev scan cli after FS has been mounted. We're still
918 * tracking a problem where systems fail mount by subvolume id
919 * when we reject replacement on a mounted FS.
921 if (!fs_devices->opened && found_transid < device->generation) {
923 * That is if the FS is _not_ mounted and if you
924 * are here, that means there is more than one
925 * disk with same uuid and devid.We keep the one
926 * with larger generation number or the last-in if
927 * generation are equal.
929 mutex_unlock(&fs_devices->device_list_mutex);
930 return ERR_PTR(-EEXIST);
934 * We are going to replace the device path for a given devid,
935 * make sure it's the same device if the device is mounted
938 struct block_device *path_bdev;
940 path_bdev = lookup_bdev(path);
941 if (IS_ERR(path_bdev)) {
942 mutex_unlock(&fs_devices->device_list_mutex);
943 return ERR_CAST(path_bdev);
946 if (device->bdev != path_bdev) {
948 mutex_unlock(&fs_devices->device_list_mutex);
950 * device->fs_info may not be reliable here, so
951 * pass in a NULL instead. This avoids a
952 * possible use-after-free when the fs_info and
953 * fs_info->sb are already torn down.
955 btrfs_warn_in_rcu(NULL,
956 "duplicate device %s devid %llu generation %llu scanned by %s (%d)",
957 path, devid, found_transid,
959 task_pid_nr(current));
960 return ERR_PTR(-EEXIST);
963 btrfs_info_in_rcu(device->fs_info,
964 "devid %llu device path %s changed to %s scanned by %s (%d)",
965 devid, rcu_str_deref(device->name),
967 task_pid_nr(current));
970 name = rcu_string_strdup(path, GFP_NOFS);
972 mutex_unlock(&fs_devices->device_list_mutex);
973 return ERR_PTR(-ENOMEM);
975 rcu_string_free(device->name);
976 rcu_assign_pointer(device->name, name);
977 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
978 fs_devices->missing_devices--;
979 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
984 * Unmount does not free the btrfs_device struct but would zero
985 * generation along with most of the other members. So just update
986 * it back. We need it to pick the disk with largest generation
989 if (!fs_devices->opened) {
990 device->generation = found_transid;
991 fs_devices->latest_generation = max_t(u64, found_transid,
992 fs_devices->latest_generation);
995 fs_devices->total_devices = btrfs_super_num_devices(disk_super);
997 mutex_unlock(&fs_devices->device_list_mutex);
1001 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
1003 struct btrfs_fs_devices *fs_devices;
1004 struct btrfs_device *device;
1005 struct btrfs_device *orig_dev;
1008 fs_devices = alloc_fs_devices(orig->fsid, NULL);
1009 if (IS_ERR(fs_devices))
1012 mutex_lock(&orig->device_list_mutex);
1013 fs_devices->total_devices = orig->total_devices;
1015 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
1016 struct rcu_string *name;
1018 device = btrfs_alloc_device(NULL, &orig_dev->devid,
1020 if (IS_ERR(device)) {
1021 ret = PTR_ERR(device);
1026 * This is ok to do without rcu read locked because we hold the
1027 * uuid mutex so nothing we touch in here is going to disappear.
1029 if (orig_dev->name) {
1030 name = rcu_string_strdup(orig_dev->name->str,
1033 btrfs_free_device(device);
1037 rcu_assign_pointer(device->name, name);
1040 list_add(&device->dev_list, &fs_devices->devices);
1041 device->fs_devices = fs_devices;
1042 fs_devices->num_devices++;
1044 mutex_unlock(&orig->device_list_mutex);
1047 mutex_unlock(&orig->device_list_mutex);
1048 free_fs_devices(fs_devices);
1049 return ERR_PTR(ret);
1052 static void __btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices,
1053 struct btrfs_device **latest_dev)
1055 struct btrfs_device *device, *next;
1057 /* This is the initialized path, it is safe to release the devices. */
1058 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
1059 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)) {
1060 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
1061 &device->dev_state) &&
1062 !test_bit(BTRFS_DEV_STATE_MISSING,
1063 &device->dev_state) &&
1065 device->generation > (*latest_dev)->generation)) {
1066 *latest_dev = device;
1072 * We have already validated the presence of BTRFS_DEV_REPLACE_DEVID,
1073 * in btrfs_init_dev_replace() so just continue.
1075 if (device->devid == BTRFS_DEV_REPLACE_DEVID)
1079 blkdev_put(device->bdev, device->mode);
1080 device->bdev = NULL;
1081 fs_devices->open_devices--;
1083 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1084 list_del_init(&device->dev_alloc_list);
1085 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
1087 list_del_init(&device->dev_list);
1088 fs_devices->num_devices--;
1089 btrfs_free_device(device);
1095 * After we have read the system tree and know devids belonging to this
1096 * filesystem, remove the device which does not belong there.
1098 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices)
1100 struct btrfs_device *latest_dev = NULL;
1101 struct btrfs_fs_devices *seed_dev;
1103 mutex_lock(&uuid_mutex);
1104 __btrfs_free_extra_devids(fs_devices, &latest_dev);
1106 list_for_each_entry(seed_dev, &fs_devices->seed_list, seed_list)
1107 __btrfs_free_extra_devids(seed_dev, &latest_dev);
1109 fs_devices->latest_bdev = latest_dev->bdev;
1111 mutex_unlock(&uuid_mutex);
1114 static void btrfs_close_bdev(struct btrfs_device *device)
1119 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1120 sync_blockdev(device->bdev);
1121 invalidate_bdev(device->bdev);
1124 blkdev_put(device->bdev, device->mode);
1127 static void btrfs_close_one_device(struct btrfs_device *device)
1129 struct btrfs_fs_devices *fs_devices = device->fs_devices;
1131 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
1132 device->devid != BTRFS_DEV_REPLACE_DEVID) {
1133 list_del_init(&device->dev_alloc_list);
1134 fs_devices->rw_devices--;
1137 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
1138 fs_devices->missing_devices--;
1140 btrfs_close_bdev(device);
1142 fs_devices->open_devices--;
1143 device->bdev = NULL;
1145 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
1146 btrfs_destroy_dev_zone_info(device);
1148 device->fs_info = NULL;
1149 atomic_set(&device->dev_stats_ccnt, 0);
1150 extent_io_tree_release(&device->alloc_state);
1152 /* Verify the device is back in a pristine state */
1153 ASSERT(!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state));
1154 ASSERT(!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state));
1155 ASSERT(list_empty(&device->dev_alloc_list));
1156 ASSERT(list_empty(&device->post_commit_list));
1157 ASSERT(atomic_read(&device->reada_in_flight) == 0);
1160 static void close_fs_devices(struct btrfs_fs_devices *fs_devices)
1162 struct btrfs_device *device, *tmp;
1164 lockdep_assert_held(&uuid_mutex);
1166 if (--fs_devices->opened > 0)
1169 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list)
1170 btrfs_close_one_device(device);
1172 WARN_ON(fs_devices->open_devices);
1173 WARN_ON(fs_devices->rw_devices);
1174 fs_devices->opened = 0;
1175 fs_devices->seeding = false;
1176 fs_devices->fs_info = NULL;
1179 void btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
1182 struct btrfs_fs_devices *tmp;
1184 mutex_lock(&uuid_mutex);
1185 close_fs_devices(fs_devices);
1186 if (!fs_devices->opened)
1187 list_splice_init(&fs_devices->seed_list, &list);
1189 list_for_each_entry_safe(fs_devices, tmp, &list, seed_list) {
1190 close_fs_devices(fs_devices);
1191 list_del(&fs_devices->seed_list);
1192 free_fs_devices(fs_devices);
1194 mutex_unlock(&uuid_mutex);
1197 static int open_fs_devices(struct btrfs_fs_devices *fs_devices,
1198 fmode_t flags, void *holder)
1200 struct btrfs_device *device;
1201 struct btrfs_device *latest_dev = NULL;
1202 struct btrfs_device *tmp_device;
1204 flags |= FMODE_EXCL;
1206 list_for_each_entry_safe(device, tmp_device, &fs_devices->devices,
1210 ret = btrfs_open_one_device(fs_devices, device, flags, holder);
1212 (!latest_dev || device->generation > latest_dev->generation)) {
1213 latest_dev = device;
1214 } else if (ret == -ENODATA) {
1215 fs_devices->num_devices--;
1216 list_del(&device->dev_list);
1217 btrfs_free_device(device);
1220 if (fs_devices->open_devices == 0)
1223 fs_devices->opened = 1;
1224 fs_devices->latest_bdev = latest_dev->bdev;
1225 fs_devices->total_rw_bytes = 0;
1226 fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_REGULAR;
1227 fs_devices->read_policy = BTRFS_READ_POLICY_PID;
1232 static int devid_cmp(void *priv, struct list_head *a, struct list_head *b)
1234 struct btrfs_device *dev1, *dev2;
1236 dev1 = list_entry(a, struct btrfs_device, dev_list);
1237 dev2 = list_entry(b, struct btrfs_device, dev_list);
1239 if (dev1->devid < dev2->devid)
1241 else if (dev1->devid > dev2->devid)
1246 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1247 fmode_t flags, void *holder)
1251 lockdep_assert_held(&uuid_mutex);
1253 * The device_list_mutex cannot be taken here in case opening the
1254 * underlying device takes further locks like bd_mutex.
1256 * We also don't need the lock here as this is called during mount and
1257 * exclusion is provided by uuid_mutex
1260 if (fs_devices->opened) {
1261 fs_devices->opened++;
1264 list_sort(NULL, &fs_devices->devices, devid_cmp);
1265 ret = open_fs_devices(fs_devices, flags, holder);
1271 void btrfs_release_disk_super(struct btrfs_super_block *super)
1273 struct page *page = virt_to_page(super);
1278 static struct btrfs_super_block *btrfs_read_disk_super(struct block_device *bdev,
1279 u64 bytenr, u64 bytenr_orig)
1281 struct btrfs_super_block *disk_super;
1286 /* make sure our super fits in the device */
1287 if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1288 return ERR_PTR(-EINVAL);
1290 /* make sure our super fits in the page */
1291 if (sizeof(*disk_super) > PAGE_SIZE)
1292 return ERR_PTR(-EINVAL);
1294 /* make sure our super doesn't straddle pages on disk */
1295 index = bytenr >> PAGE_SHIFT;
1296 if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_SHIFT != index)
1297 return ERR_PTR(-EINVAL);
1299 /* pull in the page with our super */
1300 page = read_cache_page_gfp(bdev->bd_inode->i_mapping, index, GFP_KERNEL);
1303 return ERR_CAST(page);
1305 p = page_address(page);
1307 /* align our pointer to the offset of the super block */
1308 disk_super = p + offset_in_page(bytenr);
1310 if (btrfs_super_bytenr(disk_super) != bytenr_orig ||
1311 btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
1312 btrfs_release_disk_super(p);
1313 return ERR_PTR(-EINVAL);
1316 if (disk_super->label[0] && disk_super->label[BTRFS_LABEL_SIZE - 1])
1317 disk_super->label[BTRFS_LABEL_SIZE - 1] = 0;
1322 int btrfs_forget_devices(const char *path)
1326 mutex_lock(&uuid_mutex);
1327 ret = btrfs_free_stale_devices(strlen(path) ? path : NULL, NULL);
1328 mutex_unlock(&uuid_mutex);
1334 * Look for a btrfs signature on a device. This may be called out of the mount path
1335 * and we are not allowed to call set_blocksize during the scan. The superblock
1336 * is read via pagecache
1338 struct btrfs_device *btrfs_scan_one_device(const char *path, fmode_t flags,
1341 struct btrfs_super_block *disk_super;
1342 bool new_device_added = false;
1343 struct btrfs_device *device = NULL;
1344 struct block_device *bdev;
1345 u64 bytenr, bytenr_orig;
1348 lockdep_assert_held(&uuid_mutex);
1351 * we would like to check all the supers, but that would make
1352 * a btrfs mount succeed after a mkfs from a different FS.
1353 * So, we need to add a special mount option to scan for
1354 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1356 flags |= FMODE_EXCL;
1358 bdev = blkdev_get_by_path(path, flags, holder);
1360 return ERR_CAST(bdev);
1362 bytenr_orig = btrfs_sb_offset(0);
1363 ret = btrfs_sb_log_location_bdev(bdev, 0, READ, &bytenr);
1365 return ERR_PTR(ret);
1367 disk_super = btrfs_read_disk_super(bdev, bytenr, bytenr_orig);
1368 if (IS_ERR(disk_super)) {
1369 device = ERR_CAST(disk_super);
1370 goto error_bdev_put;
1373 device = device_list_add(path, disk_super, &new_device_added);
1374 if (!IS_ERR(device)) {
1375 if (new_device_added)
1376 btrfs_free_stale_devices(path, device);
1379 btrfs_release_disk_super(disk_super);
1382 blkdev_put(bdev, flags);
1388 * Try to find a chunk that intersects [start, start + len] range and when one
1389 * such is found, record the end of it in *start
1391 static bool contains_pending_extent(struct btrfs_device *device, u64 *start,
1394 u64 physical_start, physical_end;
1396 lockdep_assert_held(&device->fs_info->chunk_mutex);
1398 if (!find_first_extent_bit(&device->alloc_state, *start,
1399 &physical_start, &physical_end,
1400 CHUNK_ALLOCATED, NULL)) {
1402 if (in_range(physical_start, *start, len) ||
1403 in_range(*start, physical_start,
1404 physical_end - physical_start)) {
1405 *start = physical_end + 1;
1412 static u64 dev_extent_search_start(struct btrfs_device *device, u64 start)
1414 switch (device->fs_devices->chunk_alloc_policy) {
1415 case BTRFS_CHUNK_ALLOC_REGULAR:
1417 * We don't want to overwrite the superblock on the drive nor
1418 * any area used by the boot loader (grub for example), so we
1419 * make sure to start at an offset of at least 1MB.
1421 return max_t(u64, start, SZ_1M);
1428 * dev_extent_hole_check - check if specified hole is suitable for allocation
1429 * @device: the device which we have the hole
1430 * @hole_start: starting position of the hole
1431 * @hole_size: the size of the hole
1432 * @num_bytes: the size of the free space that we need
1434 * This function may modify @hole_start and @hole_end to reflect the suitable
1435 * position for allocation. Returns 1 if hole position is updated, 0 otherwise.
1437 static bool dev_extent_hole_check(struct btrfs_device *device, u64 *hole_start,
1438 u64 *hole_size, u64 num_bytes)
1440 bool changed = false;
1441 u64 hole_end = *hole_start + *hole_size;
1444 * Check before we set max_hole_start, otherwise we could end up
1445 * sending back this offset anyway.
1447 if (contains_pending_extent(device, hole_start, *hole_size)) {
1448 if (hole_end >= *hole_start)
1449 *hole_size = hole_end - *hole_start;
1455 switch (device->fs_devices->chunk_alloc_policy) {
1456 case BTRFS_CHUNK_ALLOC_REGULAR:
1457 /* No extra check */
1467 * find_free_dev_extent_start - find free space in the specified device
1468 * @device: the device which we search the free space in
1469 * @num_bytes: the size of the free space that we need
1470 * @search_start: the position from which to begin the search
1471 * @start: store the start of the free space.
1472 * @len: the size of the free space. that we find, or the size
1473 * of the max free space if we don't find suitable free space
1475 * this uses a pretty simple search, the expectation is that it is
1476 * called very infrequently and that a given device has a small number
1479 * @start is used to store the start of the free space if we find. But if we
1480 * don't find suitable free space, it will be used to store the start position
1481 * of the max free space.
1483 * @len is used to store the size of the free space that we find.
1484 * But if we don't find suitable free space, it is used to store the size of
1485 * the max free space.
1487 * NOTE: This function will search *commit* root of device tree, and does extra
1488 * check to ensure dev extents are not double allocated.
1489 * This makes the function safe to allocate dev extents but may not report
1490 * correct usable device space, as device extent freed in current transaction
1491 * is not reported as avaiable.
1493 static int find_free_dev_extent_start(struct btrfs_device *device,
1494 u64 num_bytes, u64 search_start, u64 *start,
1497 struct btrfs_fs_info *fs_info = device->fs_info;
1498 struct btrfs_root *root = fs_info->dev_root;
1499 struct btrfs_key key;
1500 struct btrfs_dev_extent *dev_extent;
1501 struct btrfs_path *path;
1506 u64 search_end = device->total_bytes;
1509 struct extent_buffer *l;
1511 search_start = dev_extent_search_start(device, search_start);
1513 path = btrfs_alloc_path();
1517 max_hole_start = search_start;
1521 if (search_start >= search_end ||
1522 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1527 path->reada = READA_FORWARD;
1528 path->search_commit_root = 1;
1529 path->skip_locking = 1;
1531 key.objectid = device->devid;
1532 key.offset = search_start;
1533 key.type = BTRFS_DEV_EXTENT_KEY;
1535 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1539 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1546 slot = path->slots[0];
1547 if (slot >= btrfs_header_nritems(l)) {
1548 ret = btrfs_next_leaf(root, path);
1556 btrfs_item_key_to_cpu(l, &key, slot);
1558 if (key.objectid < device->devid)
1561 if (key.objectid > device->devid)
1564 if (key.type != BTRFS_DEV_EXTENT_KEY)
1567 if (key.offset > search_start) {
1568 hole_size = key.offset - search_start;
1569 dev_extent_hole_check(device, &search_start, &hole_size,
1572 if (hole_size > max_hole_size) {
1573 max_hole_start = search_start;
1574 max_hole_size = hole_size;
1578 * If this free space is greater than which we need,
1579 * it must be the max free space that we have found
1580 * until now, so max_hole_start must point to the start
1581 * of this free space and the length of this free space
1582 * is stored in max_hole_size. Thus, we return
1583 * max_hole_start and max_hole_size and go back to the
1586 if (hole_size >= num_bytes) {
1592 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1593 extent_end = key.offset + btrfs_dev_extent_length(l,
1595 if (extent_end > search_start)
1596 search_start = extent_end;
1603 * At this point, search_start should be the end of
1604 * allocated dev extents, and when shrinking the device,
1605 * search_end may be smaller than search_start.
1607 if (search_end > search_start) {
1608 hole_size = search_end - search_start;
1609 if (dev_extent_hole_check(device, &search_start, &hole_size,
1611 btrfs_release_path(path);
1615 if (hole_size > max_hole_size) {
1616 max_hole_start = search_start;
1617 max_hole_size = hole_size;
1622 if (max_hole_size < num_bytes)
1628 btrfs_free_path(path);
1629 *start = max_hole_start;
1631 *len = max_hole_size;
1635 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
1636 u64 *start, u64 *len)
1638 /* FIXME use last free of some kind */
1639 return find_free_dev_extent_start(device, num_bytes, 0, start, len);
1642 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1643 struct btrfs_device *device,
1644 u64 start, u64 *dev_extent_len)
1646 struct btrfs_fs_info *fs_info = device->fs_info;
1647 struct btrfs_root *root = fs_info->dev_root;
1649 struct btrfs_path *path;
1650 struct btrfs_key key;
1651 struct btrfs_key found_key;
1652 struct extent_buffer *leaf = NULL;
1653 struct btrfs_dev_extent *extent = NULL;
1655 path = btrfs_alloc_path();
1659 key.objectid = device->devid;
1661 key.type = BTRFS_DEV_EXTENT_KEY;
1663 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1665 ret = btrfs_previous_item(root, path, key.objectid,
1666 BTRFS_DEV_EXTENT_KEY);
1669 leaf = path->nodes[0];
1670 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1671 extent = btrfs_item_ptr(leaf, path->slots[0],
1672 struct btrfs_dev_extent);
1673 BUG_ON(found_key.offset > start || found_key.offset +
1674 btrfs_dev_extent_length(leaf, extent) < start);
1676 btrfs_release_path(path);
1678 } else if (ret == 0) {
1679 leaf = path->nodes[0];
1680 extent = btrfs_item_ptr(leaf, path->slots[0],
1681 struct btrfs_dev_extent);
1683 btrfs_handle_fs_error(fs_info, ret, "Slot search failed");
1687 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1689 ret = btrfs_del_item(trans, root, path);
1691 btrfs_handle_fs_error(fs_info, ret,
1692 "Failed to remove dev extent item");
1694 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1697 btrfs_free_path(path);
1701 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1702 struct btrfs_device *device,
1703 u64 chunk_offset, u64 start, u64 num_bytes)
1706 struct btrfs_path *path;
1707 struct btrfs_fs_info *fs_info = device->fs_info;
1708 struct btrfs_root *root = fs_info->dev_root;
1709 struct btrfs_dev_extent *extent;
1710 struct extent_buffer *leaf;
1711 struct btrfs_key key;
1713 WARN_ON(!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state));
1714 WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state));
1715 path = btrfs_alloc_path();
1719 key.objectid = device->devid;
1721 key.type = BTRFS_DEV_EXTENT_KEY;
1722 ret = btrfs_insert_empty_item(trans, root, path, &key,
1727 leaf = path->nodes[0];
1728 extent = btrfs_item_ptr(leaf, path->slots[0],
1729 struct btrfs_dev_extent);
1730 btrfs_set_dev_extent_chunk_tree(leaf, extent,
1731 BTRFS_CHUNK_TREE_OBJECTID);
1732 btrfs_set_dev_extent_chunk_objectid(leaf, extent,
1733 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
1734 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1736 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1737 btrfs_mark_buffer_dirty(leaf);
1739 btrfs_free_path(path);
1743 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1745 struct extent_map_tree *em_tree;
1746 struct extent_map *em;
1750 em_tree = &fs_info->mapping_tree;
1751 read_lock(&em_tree->lock);
1752 n = rb_last(&em_tree->map.rb_root);
1754 em = rb_entry(n, struct extent_map, rb_node);
1755 ret = em->start + em->len;
1757 read_unlock(&em_tree->lock);
1762 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1766 struct btrfs_key key;
1767 struct btrfs_key found_key;
1768 struct btrfs_path *path;
1770 path = btrfs_alloc_path();
1774 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1775 key.type = BTRFS_DEV_ITEM_KEY;
1776 key.offset = (u64)-1;
1778 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1784 btrfs_err(fs_info, "corrupted chunk tree devid -1 matched");
1789 ret = btrfs_previous_item(fs_info->chunk_root, path,
1790 BTRFS_DEV_ITEMS_OBJECTID,
1791 BTRFS_DEV_ITEM_KEY);
1795 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1797 *devid_ret = found_key.offset + 1;
1801 btrfs_free_path(path);
1806 * the device information is stored in the chunk root
1807 * the btrfs_device struct should be fully filled in
1809 static int btrfs_add_dev_item(struct btrfs_trans_handle *trans,
1810 struct btrfs_device *device)
1813 struct btrfs_path *path;
1814 struct btrfs_dev_item *dev_item;
1815 struct extent_buffer *leaf;
1816 struct btrfs_key key;
1819 path = btrfs_alloc_path();
1823 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1824 key.type = BTRFS_DEV_ITEM_KEY;
1825 key.offset = device->devid;
1827 ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path,
1828 &key, sizeof(*dev_item));
1832 leaf = path->nodes[0];
1833 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1835 btrfs_set_device_id(leaf, dev_item, device->devid);
1836 btrfs_set_device_generation(leaf, dev_item, 0);
1837 btrfs_set_device_type(leaf, dev_item, device->type);
1838 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1839 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1840 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1841 btrfs_set_device_total_bytes(leaf, dev_item,
1842 btrfs_device_get_disk_total_bytes(device));
1843 btrfs_set_device_bytes_used(leaf, dev_item,
1844 btrfs_device_get_bytes_used(device));
1845 btrfs_set_device_group(leaf, dev_item, 0);
1846 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1847 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1848 btrfs_set_device_start_offset(leaf, dev_item, 0);
1850 ptr = btrfs_device_uuid(dev_item);
1851 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1852 ptr = btrfs_device_fsid(dev_item);
1853 write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid,
1854 ptr, BTRFS_FSID_SIZE);
1855 btrfs_mark_buffer_dirty(leaf);
1859 btrfs_free_path(path);
1864 * Function to update ctime/mtime for a given device path.
1865 * Mainly used for ctime/mtime based probe like libblkid.
1867 static void update_dev_time(const char *path_name)
1871 filp = filp_open(path_name, O_RDWR, 0);
1874 file_update_time(filp);
1875 filp_close(filp, NULL);
1878 static int btrfs_rm_dev_item(struct btrfs_device *device)
1880 struct btrfs_root *root = device->fs_info->chunk_root;
1882 struct btrfs_path *path;
1883 struct btrfs_key key;
1884 struct btrfs_trans_handle *trans;
1886 path = btrfs_alloc_path();
1890 trans = btrfs_start_transaction(root, 0);
1891 if (IS_ERR(trans)) {
1892 btrfs_free_path(path);
1893 return PTR_ERR(trans);
1895 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1896 key.type = BTRFS_DEV_ITEM_KEY;
1897 key.offset = device->devid;
1899 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1903 btrfs_abort_transaction(trans, ret);
1904 btrfs_end_transaction(trans);
1908 ret = btrfs_del_item(trans, root, path);
1910 btrfs_abort_transaction(trans, ret);
1911 btrfs_end_transaction(trans);
1915 btrfs_free_path(path);
1917 ret = btrfs_commit_transaction(trans);
1922 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1923 * filesystem. It's up to the caller to adjust that number regarding eg. device
1926 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1934 seq = read_seqbegin(&fs_info->profiles_lock);
1936 all_avail = fs_info->avail_data_alloc_bits |
1937 fs_info->avail_system_alloc_bits |
1938 fs_info->avail_metadata_alloc_bits;
1939 } while (read_seqretry(&fs_info->profiles_lock, seq));
1941 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1942 if (!(all_avail & btrfs_raid_array[i].bg_flag))
1945 if (num_devices < btrfs_raid_array[i].devs_min) {
1946 int ret = btrfs_raid_array[i].mindev_error;
1956 static struct btrfs_device * btrfs_find_next_active_device(
1957 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device)
1959 struct btrfs_device *next_device;
1961 list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1962 if (next_device != device &&
1963 !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state)
1964 && next_device->bdev)
1972 * Helper function to check if the given device is part of s_bdev / latest_bdev
1973 * and replace it with the provided or the next active device, in the context
1974 * where this function called, there should be always be another device (or
1975 * this_dev) which is active.
1977 void __cold btrfs_assign_next_active_device(struct btrfs_device *device,
1978 struct btrfs_device *next_device)
1980 struct btrfs_fs_info *fs_info = device->fs_info;
1983 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
1985 ASSERT(next_device);
1987 if (fs_info->sb->s_bdev &&
1988 (fs_info->sb->s_bdev == device->bdev))
1989 fs_info->sb->s_bdev = next_device->bdev;
1991 if (fs_info->fs_devices->latest_bdev == device->bdev)
1992 fs_info->fs_devices->latest_bdev = next_device->bdev;
1996 * Return btrfs_fs_devices::num_devices excluding the device that's being
1997 * currently replaced.
1999 static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info)
2001 u64 num_devices = fs_info->fs_devices->num_devices;
2003 down_read(&fs_info->dev_replace.rwsem);
2004 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
2005 ASSERT(num_devices > 1);
2008 up_read(&fs_info->dev_replace.rwsem);
2013 void btrfs_scratch_superblocks(struct btrfs_fs_info *fs_info,
2014 struct block_device *bdev,
2015 const char *device_path)
2017 struct btrfs_super_block *disk_super;
2023 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; copy_num++) {
2027 disk_super = btrfs_read_dev_one_super(bdev, copy_num);
2028 if (IS_ERR(disk_super))
2031 if (bdev_is_zoned(bdev)) {
2032 btrfs_reset_sb_log_zones(bdev, copy_num);
2036 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
2038 page = virt_to_page(disk_super);
2039 set_page_dirty(page);
2041 /* write_on_page() unlocks the page */
2042 ret = write_one_page(page);
2045 "error clearing superblock number %d (%d)",
2047 btrfs_release_disk_super(disk_super);
2051 /* Notify udev that device has changed */
2052 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
2054 /* Update ctime/mtime for device path for libblkid */
2055 update_dev_time(device_path);
2058 int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path,
2061 struct btrfs_device *device;
2062 struct btrfs_fs_devices *cur_devices;
2063 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2067 mutex_lock(&uuid_mutex);
2069 num_devices = btrfs_num_devices(fs_info);
2071 ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
2075 device = btrfs_find_device_by_devspec(fs_info, devid, device_path);
2077 if (IS_ERR(device)) {
2078 if (PTR_ERR(device) == -ENOENT &&
2079 strcmp(device_path, "missing") == 0)
2080 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2082 ret = PTR_ERR(device);
2086 if (btrfs_pinned_by_swapfile(fs_info, device)) {
2087 btrfs_warn_in_rcu(fs_info,
2088 "cannot remove device %s (devid %llu) due to active swapfile",
2089 rcu_str_deref(device->name), device->devid);
2094 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2095 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
2099 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
2100 fs_info->fs_devices->rw_devices == 1) {
2101 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
2105 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2106 mutex_lock(&fs_info->chunk_mutex);
2107 list_del_init(&device->dev_alloc_list);
2108 device->fs_devices->rw_devices--;
2109 mutex_unlock(&fs_info->chunk_mutex);
2112 mutex_unlock(&uuid_mutex);
2113 ret = btrfs_shrink_device(device, 0);
2115 btrfs_reada_remove_dev(device);
2116 mutex_lock(&uuid_mutex);
2121 * TODO: the superblock still includes this device in its num_devices
2122 * counter although write_all_supers() is not locked out. This
2123 * could give a filesystem state which requires a degraded mount.
2125 ret = btrfs_rm_dev_item(device);
2129 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2130 btrfs_scrub_cancel_dev(device);
2133 * the device list mutex makes sure that we don't change
2134 * the device list while someone else is writing out all
2135 * the device supers. Whoever is writing all supers, should
2136 * lock the device list mutex before getting the number of
2137 * devices in the super block (super_copy). Conversely,
2138 * whoever updates the number of devices in the super block
2139 * (super_copy) should hold the device list mutex.
2143 * In normal cases the cur_devices == fs_devices. But in case
2144 * of deleting a seed device, the cur_devices should point to
2145 * its own fs_devices listed under the fs_devices->seed.
2147 cur_devices = device->fs_devices;
2148 mutex_lock(&fs_devices->device_list_mutex);
2149 list_del_rcu(&device->dev_list);
2151 cur_devices->num_devices--;
2152 cur_devices->total_devices--;
2153 /* Update total_devices of the parent fs_devices if it's seed */
2154 if (cur_devices != fs_devices)
2155 fs_devices->total_devices--;
2157 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
2158 cur_devices->missing_devices--;
2160 btrfs_assign_next_active_device(device, NULL);
2163 cur_devices->open_devices--;
2164 /* remove sysfs entry */
2165 btrfs_sysfs_remove_device(device);
2168 num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
2169 btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
2170 mutex_unlock(&fs_devices->device_list_mutex);
2173 * at this point, the device is zero sized and detached from
2174 * the devices list. All that's left is to zero out the old
2175 * supers and free the device.
2177 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2178 btrfs_scratch_superblocks(fs_info, device->bdev,
2181 btrfs_close_bdev(device);
2183 btrfs_free_device(device);
2185 if (cur_devices->open_devices == 0) {
2186 list_del_init(&cur_devices->seed_list);
2187 close_fs_devices(cur_devices);
2188 free_fs_devices(cur_devices);
2192 mutex_unlock(&uuid_mutex);
2196 btrfs_reada_undo_remove_dev(device);
2197 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2198 mutex_lock(&fs_info->chunk_mutex);
2199 list_add(&device->dev_alloc_list,
2200 &fs_devices->alloc_list);
2201 device->fs_devices->rw_devices++;
2202 mutex_unlock(&fs_info->chunk_mutex);
2207 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev)
2209 struct btrfs_fs_devices *fs_devices;
2211 lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex);
2214 * in case of fs with no seed, srcdev->fs_devices will point
2215 * to fs_devices of fs_info. However when the dev being replaced is
2216 * a seed dev it will point to the seed's local fs_devices. In short
2217 * srcdev will have its correct fs_devices in both the cases.
2219 fs_devices = srcdev->fs_devices;
2221 list_del_rcu(&srcdev->dev_list);
2222 list_del(&srcdev->dev_alloc_list);
2223 fs_devices->num_devices--;
2224 if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state))
2225 fs_devices->missing_devices--;
2227 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state))
2228 fs_devices->rw_devices--;
2231 fs_devices->open_devices--;
2234 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev)
2236 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2238 mutex_lock(&uuid_mutex);
2240 btrfs_close_bdev(srcdev);
2242 btrfs_free_device(srcdev);
2244 /* if this is no devs we rather delete the fs_devices */
2245 if (!fs_devices->num_devices) {
2247 * On a mounted FS, num_devices can't be zero unless it's a
2248 * seed. In case of a seed device being replaced, the replace
2249 * target added to the sprout FS, so there will be no more
2250 * device left under the seed FS.
2252 ASSERT(fs_devices->seeding);
2254 list_del_init(&fs_devices->seed_list);
2255 close_fs_devices(fs_devices);
2256 free_fs_devices(fs_devices);
2258 mutex_unlock(&uuid_mutex);
2261 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev)
2263 struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices;
2265 mutex_lock(&fs_devices->device_list_mutex);
2267 btrfs_sysfs_remove_device(tgtdev);
2270 fs_devices->open_devices--;
2272 fs_devices->num_devices--;
2274 btrfs_assign_next_active_device(tgtdev, NULL);
2276 list_del_rcu(&tgtdev->dev_list);
2278 mutex_unlock(&fs_devices->device_list_mutex);
2281 * The update_dev_time() with in btrfs_scratch_superblocks()
2282 * may lead to a call to btrfs_show_devname() which will try
2283 * to hold device_list_mutex. And here this device
2284 * is already out of device list, so we don't have to hold
2285 * the device_list_mutex lock.
2287 btrfs_scratch_superblocks(tgtdev->fs_info, tgtdev->bdev,
2290 btrfs_close_bdev(tgtdev);
2292 btrfs_free_device(tgtdev);
2295 static struct btrfs_device *btrfs_find_device_by_path(
2296 struct btrfs_fs_info *fs_info, const char *device_path)
2299 struct btrfs_super_block *disk_super;
2302 struct block_device *bdev;
2303 struct btrfs_device *device;
2305 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2306 fs_info->bdev_holder, 0, &bdev, &disk_super);
2308 return ERR_PTR(ret);
2310 devid = btrfs_stack_device_id(&disk_super->dev_item);
2311 dev_uuid = disk_super->dev_item.uuid;
2312 if (btrfs_fs_incompat(fs_info, METADATA_UUID))
2313 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
2314 disk_super->metadata_uuid);
2316 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
2319 btrfs_release_disk_super(disk_super);
2321 device = ERR_PTR(-ENOENT);
2322 blkdev_put(bdev, FMODE_READ);
2327 * Lookup a device given by device id, or the path if the id is 0.
2329 struct btrfs_device *btrfs_find_device_by_devspec(
2330 struct btrfs_fs_info *fs_info, u64 devid,
2331 const char *device_path)
2333 struct btrfs_device *device;
2336 device = btrfs_find_device(fs_info->fs_devices, devid, NULL,
2339 return ERR_PTR(-ENOENT);
2343 if (!device_path || !device_path[0])
2344 return ERR_PTR(-EINVAL);
2346 if (strcmp(device_path, "missing") == 0) {
2347 /* Find first missing device */
2348 list_for_each_entry(device, &fs_info->fs_devices->devices,
2350 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
2351 &device->dev_state) && !device->bdev)
2354 return ERR_PTR(-ENOENT);
2357 return btrfs_find_device_by_path(fs_info, device_path);
2361 * does all the dirty work required for changing file system's UUID.
2363 static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info)
2365 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2366 struct btrfs_fs_devices *old_devices;
2367 struct btrfs_fs_devices *seed_devices;
2368 struct btrfs_super_block *disk_super = fs_info->super_copy;
2369 struct btrfs_device *device;
2372 lockdep_assert_held(&uuid_mutex);
2373 if (!fs_devices->seeding)
2377 * Private copy of the seed devices, anchored at
2378 * fs_info->fs_devices->seed_list
2380 seed_devices = alloc_fs_devices(NULL, NULL);
2381 if (IS_ERR(seed_devices))
2382 return PTR_ERR(seed_devices);
2385 * It's necessary to retain a copy of the original seed fs_devices in
2386 * fs_uuids so that filesystems which have been seeded can successfully
2387 * reference the seed device from open_seed_devices. This also supports
2390 old_devices = clone_fs_devices(fs_devices);
2391 if (IS_ERR(old_devices)) {
2392 kfree(seed_devices);
2393 return PTR_ERR(old_devices);
2396 list_add(&old_devices->fs_list, &fs_uuids);
2398 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2399 seed_devices->opened = 1;
2400 INIT_LIST_HEAD(&seed_devices->devices);
2401 INIT_LIST_HEAD(&seed_devices->alloc_list);
2402 mutex_init(&seed_devices->device_list_mutex);
2404 mutex_lock(&fs_devices->device_list_mutex);
2405 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2407 list_for_each_entry(device, &seed_devices->devices, dev_list)
2408 device->fs_devices = seed_devices;
2410 fs_devices->seeding = false;
2411 fs_devices->num_devices = 0;
2412 fs_devices->open_devices = 0;
2413 fs_devices->missing_devices = 0;
2414 fs_devices->rotating = false;
2415 list_add(&seed_devices->seed_list, &fs_devices->seed_list);
2417 generate_random_uuid(fs_devices->fsid);
2418 memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE);
2419 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2420 mutex_unlock(&fs_devices->device_list_mutex);
2422 super_flags = btrfs_super_flags(disk_super) &
2423 ~BTRFS_SUPER_FLAG_SEEDING;
2424 btrfs_set_super_flags(disk_super, super_flags);
2430 * Store the expected generation for seed devices in device items.
2432 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans)
2434 struct btrfs_fs_info *fs_info = trans->fs_info;
2435 struct btrfs_root *root = fs_info->chunk_root;
2436 struct btrfs_path *path;
2437 struct extent_buffer *leaf;
2438 struct btrfs_dev_item *dev_item;
2439 struct btrfs_device *device;
2440 struct btrfs_key key;
2441 u8 fs_uuid[BTRFS_FSID_SIZE];
2442 u8 dev_uuid[BTRFS_UUID_SIZE];
2446 path = btrfs_alloc_path();
2450 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2452 key.type = BTRFS_DEV_ITEM_KEY;
2455 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2459 leaf = path->nodes[0];
2461 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2462 ret = btrfs_next_leaf(root, path);
2467 leaf = path->nodes[0];
2468 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2469 btrfs_release_path(path);
2473 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2474 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2475 key.type != BTRFS_DEV_ITEM_KEY)
2478 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2479 struct btrfs_dev_item);
2480 devid = btrfs_device_id(leaf, dev_item);
2481 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2483 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2485 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
2487 BUG_ON(!device); /* Logic error */
2489 if (device->fs_devices->seeding) {
2490 btrfs_set_device_generation(leaf, dev_item,
2491 device->generation);
2492 btrfs_mark_buffer_dirty(leaf);
2500 btrfs_free_path(path);
2504 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2506 struct btrfs_root *root = fs_info->dev_root;
2507 struct request_queue *q;
2508 struct btrfs_trans_handle *trans;
2509 struct btrfs_device *device;
2510 struct block_device *bdev;
2511 struct super_block *sb = fs_info->sb;
2512 struct rcu_string *name;
2513 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2514 u64 orig_super_total_bytes;
2515 u64 orig_super_num_devices;
2516 int seeding_dev = 0;
2518 bool locked = false;
2520 if (sb_rdonly(sb) && !fs_devices->seeding)
2523 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2524 fs_info->bdev_holder);
2526 return PTR_ERR(bdev);
2528 if (!btrfs_check_device_zone_type(fs_info, bdev)) {
2533 if (fs_devices->seeding) {
2535 down_write(&sb->s_umount);
2536 mutex_lock(&uuid_mutex);
2540 sync_blockdev(bdev);
2543 list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
2544 if (device->bdev == bdev) {
2552 device = btrfs_alloc_device(fs_info, NULL, NULL);
2553 if (IS_ERR(device)) {
2554 /* we can safely leave the fs_devices entry around */
2555 ret = PTR_ERR(device);
2559 name = rcu_string_strdup(device_path, GFP_KERNEL);
2562 goto error_free_device;
2564 rcu_assign_pointer(device->name, name);
2566 device->fs_info = fs_info;
2567 device->bdev = bdev;
2569 ret = btrfs_get_dev_zone_info(device);
2571 goto error_free_device;
2573 trans = btrfs_start_transaction(root, 0);
2574 if (IS_ERR(trans)) {
2575 ret = PTR_ERR(trans);
2576 goto error_free_zone;
2579 q = bdev_get_queue(bdev);
2580 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
2581 device->generation = trans->transid;
2582 device->io_width = fs_info->sectorsize;
2583 device->io_align = fs_info->sectorsize;
2584 device->sector_size = fs_info->sectorsize;
2585 device->total_bytes = round_down(i_size_read(bdev->bd_inode),
2586 fs_info->sectorsize);
2587 device->disk_total_bytes = device->total_bytes;
2588 device->commit_total_bytes = device->total_bytes;
2589 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2590 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
2591 device->mode = FMODE_EXCL;
2592 device->dev_stats_valid = 1;
2593 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2596 sb->s_flags &= ~SB_RDONLY;
2597 ret = btrfs_prepare_sprout(fs_info);
2599 btrfs_abort_transaction(trans, ret);
2604 device->fs_devices = fs_devices;
2606 mutex_lock(&fs_devices->device_list_mutex);
2607 mutex_lock(&fs_info->chunk_mutex);
2608 list_add_rcu(&device->dev_list, &fs_devices->devices);
2609 list_add(&device->dev_alloc_list, &fs_devices->alloc_list);
2610 fs_devices->num_devices++;
2611 fs_devices->open_devices++;
2612 fs_devices->rw_devices++;
2613 fs_devices->total_devices++;
2614 fs_devices->total_rw_bytes += device->total_bytes;
2616 atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2618 if (!blk_queue_nonrot(q))
2619 fs_devices->rotating = true;
2621 orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
2622 btrfs_set_super_total_bytes(fs_info->super_copy,
2623 round_down(orig_super_total_bytes + device->total_bytes,
2624 fs_info->sectorsize));
2626 orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy);
2627 btrfs_set_super_num_devices(fs_info->super_copy,
2628 orig_super_num_devices + 1);
2631 * we've got more storage, clear any full flags on the space
2634 btrfs_clear_space_info_full(fs_info);
2636 mutex_unlock(&fs_info->chunk_mutex);
2638 /* Add sysfs device entry */
2639 btrfs_sysfs_add_device(device);
2641 mutex_unlock(&fs_devices->device_list_mutex);
2644 mutex_lock(&fs_info->chunk_mutex);
2645 ret = init_first_rw_device(trans);
2646 mutex_unlock(&fs_info->chunk_mutex);
2648 btrfs_abort_transaction(trans, ret);
2653 ret = btrfs_add_dev_item(trans, device);
2655 btrfs_abort_transaction(trans, ret);
2660 ret = btrfs_finish_sprout(trans);
2662 btrfs_abort_transaction(trans, ret);
2667 * fs_devices now represents the newly sprouted filesystem and
2668 * its fsid has been changed by btrfs_prepare_sprout
2670 btrfs_sysfs_update_sprout_fsid(fs_devices);
2673 ret = btrfs_commit_transaction(trans);
2676 mutex_unlock(&uuid_mutex);
2677 up_write(&sb->s_umount);
2680 if (ret) /* transaction commit */
2683 ret = btrfs_relocate_sys_chunks(fs_info);
2685 btrfs_handle_fs_error(fs_info, ret,
2686 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2687 trans = btrfs_attach_transaction(root);
2688 if (IS_ERR(trans)) {
2689 if (PTR_ERR(trans) == -ENOENT)
2691 ret = PTR_ERR(trans);
2695 ret = btrfs_commit_transaction(trans);
2699 * Now that we have written a new super block to this device, check all
2700 * other fs_devices list if device_path alienates any other scanned
2702 * We can ignore the return value as it typically returns -EINVAL and
2703 * only succeeds if the device was an alien.
2705 btrfs_forget_devices(device_path);
2707 /* Update ctime/mtime for blkid or udev */
2708 update_dev_time(device_path);
2713 btrfs_sysfs_remove_device(device);
2714 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2715 mutex_lock(&fs_info->chunk_mutex);
2716 list_del_rcu(&device->dev_list);
2717 list_del(&device->dev_alloc_list);
2718 fs_info->fs_devices->num_devices--;
2719 fs_info->fs_devices->open_devices--;
2720 fs_info->fs_devices->rw_devices--;
2721 fs_info->fs_devices->total_devices--;
2722 fs_info->fs_devices->total_rw_bytes -= device->total_bytes;
2723 atomic64_sub(device->total_bytes, &fs_info->free_chunk_space);
2724 btrfs_set_super_total_bytes(fs_info->super_copy,
2725 orig_super_total_bytes);
2726 btrfs_set_super_num_devices(fs_info->super_copy,
2727 orig_super_num_devices);
2728 mutex_unlock(&fs_info->chunk_mutex);
2729 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2732 sb->s_flags |= SB_RDONLY;
2734 btrfs_end_transaction(trans);
2736 btrfs_destroy_dev_zone_info(device);
2738 btrfs_free_device(device);
2740 blkdev_put(bdev, FMODE_EXCL);
2742 mutex_unlock(&uuid_mutex);
2743 up_write(&sb->s_umount);
2748 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2749 struct btrfs_device *device)
2752 struct btrfs_path *path;
2753 struct btrfs_root *root = device->fs_info->chunk_root;
2754 struct btrfs_dev_item *dev_item;
2755 struct extent_buffer *leaf;
2756 struct btrfs_key key;
2758 path = btrfs_alloc_path();
2762 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2763 key.type = BTRFS_DEV_ITEM_KEY;
2764 key.offset = device->devid;
2766 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2775 leaf = path->nodes[0];
2776 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2778 btrfs_set_device_id(leaf, dev_item, device->devid);
2779 btrfs_set_device_type(leaf, dev_item, device->type);
2780 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2781 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2782 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2783 btrfs_set_device_total_bytes(leaf, dev_item,
2784 btrfs_device_get_disk_total_bytes(device));
2785 btrfs_set_device_bytes_used(leaf, dev_item,
2786 btrfs_device_get_bytes_used(device));
2787 btrfs_mark_buffer_dirty(leaf);
2790 btrfs_free_path(path);
2794 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2795 struct btrfs_device *device, u64 new_size)
2797 struct btrfs_fs_info *fs_info = device->fs_info;
2798 struct btrfs_super_block *super_copy = fs_info->super_copy;
2802 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2805 new_size = round_down(new_size, fs_info->sectorsize);
2807 mutex_lock(&fs_info->chunk_mutex);
2808 old_total = btrfs_super_total_bytes(super_copy);
2809 diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2811 if (new_size <= device->total_bytes ||
2812 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2813 mutex_unlock(&fs_info->chunk_mutex);
2817 btrfs_set_super_total_bytes(super_copy,
2818 round_down(old_total + diff, fs_info->sectorsize));
2819 device->fs_devices->total_rw_bytes += diff;
2821 btrfs_device_set_total_bytes(device, new_size);
2822 btrfs_device_set_disk_total_bytes(device, new_size);
2823 btrfs_clear_space_info_full(device->fs_info);
2824 if (list_empty(&device->post_commit_list))
2825 list_add_tail(&device->post_commit_list,
2826 &trans->transaction->dev_update_list);
2827 mutex_unlock(&fs_info->chunk_mutex);
2829 return btrfs_update_device(trans, device);
2832 static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
2834 struct btrfs_fs_info *fs_info = trans->fs_info;
2835 struct btrfs_root *root = fs_info->chunk_root;
2837 struct btrfs_path *path;
2838 struct btrfs_key key;
2840 path = btrfs_alloc_path();
2844 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2845 key.offset = chunk_offset;
2846 key.type = BTRFS_CHUNK_ITEM_KEY;
2848 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2851 else if (ret > 0) { /* Logic error or corruption */
2852 btrfs_handle_fs_error(fs_info, -ENOENT,
2853 "Failed lookup while freeing chunk.");
2858 ret = btrfs_del_item(trans, root, path);
2860 btrfs_handle_fs_error(fs_info, ret,
2861 "Failed to delete chunk item.");
2863 btrfs_free_path(path);
2867 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2869 struct btrfs_super_block *super_copy = fs_info->super_copy;
2870 struct btrfs_disk_key *disk_key;
2871 struct btrfs_chunk *chunk;
2878 struct btrfs_key key;
2880 mutex_lock(&fs_info->chunk_mutex);
2881 array_size = btrfs_super_sys_array_size(super_copy);
2883 ptr = super_copy->sys_chunk_array;
2886 while (cur < array_size) {
2887 disk_key = (struct btrfs_disk_key *)ptr;
2888 btrfs_disk_key_to_cpu(&key, disk_key);
2890 len = sizeof(*disk_key);
2892 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2893 chunk = (struct btrfs_chunk *)(ptr + len);
2894 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2895 len += btrfs_chunk_item_size(num_stripes);
2900 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID &&
2901 key.offset == chunk_offset) {
2902 memmove(ptr, ptr + len, array_size - (cur + len));
2904 btrfs_set_super_sys_array_size(super_copy, array_size);
2910 mutex_unlock(&fs_info->chunk_mutex);
2915 * btrfs_get_chunk_map() - Find the mapping containing the given logical extent.
2916 * @logical: Logical block offset in bytes.
2917 * @length: Length of extent in bytes.
2919 * Return: Chunk mapping or ERR_PTR.
2921 struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info,
2922 u64 logical, u64 length)
2924 struct extent_map_tree *em_tree;
2925 struct extent_map *em;
2927 em_tree = &fs_info->mapping_tree;
2928 read_lock(&em_tree->lock);
2929 em = lookup_extent_mapping(em_tree, logical, length);
2930 read_unlock(&em_tree->lock);
2933 btrfs_crit(fs_info, "unable to find logical %llu length %llu",
2935 return ERR_PTR(-EINVAL);
2938 if (em->start > logical || em->start + em->len < logical) {
2940 "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
2941 logical, length, em->start, em->start + em->len);
2942 free_extent_map(em);
2943 return ERR_PTR(-EINVAL);
2946 /* callers are responsible for dropping em's ref. */
2950 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
2952 struct btrfs_fs_info *fs_info = trans->fs_info;
2953 struct extent_map *em;
2954 struct map_lookup *map;
2955 u64 dev_extent_len = 0;
2957 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2959 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
2962 * This is a logic error, but we don't want to just rely on the
2963 * user having built with ASSERT enabled, so if ASSERT doesn't
2964 * do anything we still error out.
2969 map = em->map_lookup;
2970 mutex_lock(&fs_info->chunk_mutex);
2971 check_system_chunk(trans, map->type);
2972 mutex_unlock(&fs_info->chunk_mutex);
2975 * Take the device list mutex to prevent races with the final phase of
2976 * a device replace operation that replaces the device object associated
2977 * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2979 mutex_lock(&fs_devices->device_list_mutex);
2980 for (i = 0; i < map->num_stripes; i++) {
2981 struct btrfs_device *device = map->stripes[i].dev;
2982 ret = btrfs_free_dev_extent(trans, device,
2983 map->stripes[i].physical,
2986 mutex_unlock(&fs_devices->device_list_mutex);
2987 btrfs_abort_transaction(trans, ret);
2991 if (device->bytes_used > 0) {
2992 mutex_lock(&fs_info->chunk_mutex);
2993 btrfs_device_set_bytes_used(device,
2994 device->bytes_used - dev_extent_len);
2995 atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
2996 btrfs_clear_space_info_full(fs_info);
2997 mutex_unlock(&fs_info->chunk_mutex);
3000 ret = btrfs_update_device(trans, device);
3002 mutex_unlock(&fs_devices->device_list_mutex);
3003 btrfs_abort_transaction(trans, ret);
3007 mutex_unlock(&fs_devices->device_list_mutex);
3009 ret = btrfs_free_chunk(trans, chunk_offset);
3011 btrfs_abort_transaction(trans, ret);
3015 trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
3017 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
3018 ret = btrfs_del_sys_chunk(fs_info, chunk_offset);
3020 btrfs_abort_transaction(trans, ret);
3025 ret = btrfs_remove_block_group(trans, chunk_offset, em);
3027 btrfs_abort_transaction(trans, ret);
3033 free_extent_map(em);
3037 static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
3039 struct btrfs_root *root = fs_info->chunk_root;
3040 struct btrfs_trans_handle *trans;
3041 struct btrfs_block_group *block_group;
3045 * Prevent races with automatic removal of unused block groups.
3046 * After we relocate and before we remove the chunk with offset
3047 * chunk_offset, automatic removal of the block group can kick in,
3048 * resulting in a failure when calling btrfs_remove_chunk() below.
3050 * Make sure to acquire this mutex before doing a tree search (dev
3051 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
3052 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
3053 * we release the path used to search the chunk/dev tree and before
3054 * the current task acquires this mutex and calls us.
3056 lockdep_assert_held(&fs_info->delete_unused_bgs_mutex);
3058 /* step one, relocate all the extents inside this chunk */
3059 btrfs_scrub_pause(fs_info);
3060 ret = btrfs_relocate_block_group(fs_info, chunk_offset);
3061 btrfs_scrub_continue(fs_info);
3065 block_group = btrfs_lookup_block_group(fs_info, chunk_offset);
3068 btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group);
3069 btrfs_put_block_group(block_group);
3071 trans = btrfs_start_trans_remove_block_group(root->fs_info,
3073 if (IS_ERR(trans)) {
3074 ret = PTR_ERR(trans);
3075 btrfs_handle_fs_error(root->fs_info, ret, NULL);
3080 * step two, delete the device extents and the
3081 * chunk tree entries
3083 ret = btrfs_remove_chunk(trans, chunk_offset);
3084 btrfs_end_transaction(trans);
3088 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
3090 struct btrfs_root *chunk_root = fs_info->chunk_root;
3091 struct btrfs_path *path;
3092 struct extent_buffer *leaf;
3093 struct btrfs_chunk *chunk;
3094 struct btrfs_key key;
3095 struct btrfs_key found_key;
3097 bool retried = false;
3101 path = btrfs_alloc_path();
3106 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3107 key.offset = (u64)-1;
3108 key.type = BTRFS_CHUNK_ITEM_KEY;
3111 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3112 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3114 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3117 BUG_ON(ret == 0); /* Corruption */
3119 ret = btrfs_previous_item(chunk_root, path, key.objectid,
3122 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3128 leaf = path->nodes[0];
3129 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3131 chunk = btrfs_item_ptr(leaf, path->slots[0],
3132 struct btrfs_chunk);
3133 chunk_type = btrfs_chunk_type(leaf, chunk);
3134 btrfs_release_path(path);
3136 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3137 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3143 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3145 if (found_key.offset == 0)
3147 key.offset = found_key.offset - 1;
3150 if (failed && !retried) {
3154 } else if (WARN_ON(failed && retried)) {
3158 btrfs_free_path(path);
3163 * return 1 : allocate a data chunk successfully,
3164 * return <0: errors during allocating a data chunk,
3165 * return 0 : no need to allocate a data chunk.
3167 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info,
3170 struct btrfs_block_group *cache;
3174 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3176 chunk_type = cache->flags;
3177 btrfs_put_block_group(cache);
3179 if (!(chunk_type & BTRFS_BLOCK_GROUP_DATA))
3182 spin_lock(&fs_info->data_sinfo->lock);
3183 bytes_used = fs_info->data_sinfo->bytes_used;
3184 spin_unlock(&fs_info->data_sinfo->lock);
3187 struct btrfs_trans_handle *trans;
3190 trans = btrfs_join_transaction(fs_info->tree_root);
3192 return PTR_ERR(trans);
3194 ret = btrfs_force_chunk_alloc(trans, BTRFS_BLOCK_GROUP_DATA);
3195 btrfs_end_transaction(trans);
3204 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3205 struct btrfs_balance_control *bctl)
3207 struct btrfs_root *root = fs_info->tree_root;
3208 struct btrfs_trans_handle *trans;
3209 struct btrfs_balance_item *item;
3210 struct btrfs_disk_balance_args disk_bargs;
3211 struct btrfs_path *path;
3212 struct extent_buffer *leaf;
3213 struct btrfs_key key;
3216 path = btrfs_alloc_path();
3220 trans = btrfs_start_transaction(root, 0);
3221 if (IS_ERR(trans)) {
3222 btrfs_free_path(path);
3223 return PTR_ERR(trans);
3226 key.objectid = BTRFS_BALANCE_OBJECTID;
3227 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3230 ret = btrfs_insert_empty_item(trans, root, path, &key,
3235 leaf = path->nodes[0];
3236 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3238 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3240 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3241 btrfs_set_balance_data(leaf, item, &disk_bargs);
3242 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3243 btrfs_set_balance_meta(leaf, item, &disk_bargs);
3244 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3245 btrfs_set_balance_sys(leaf, item, &disk_bargs);
3247 btrfs_set_balance_flags(leaf, item, bctl->flags);
3249 btrfs_mark_buffer_dirty(leaf);
3251 btrfs_free_path(path);
3252 err = btrfs_commit_transaction(trans);
3258 static int del_balance_item(struct btrfs_fs_info *fs_info)
3260 struct btrfs_root *root = fs_info->tree_root;
3261 struct btrfs_trans_handle *trans;
3262 struct btrfs_path *path;
3263 struct btrfs_key key;
3266 path = btrfs_alloc_path();
3270 trans = btrfs_start_transaction_fallback_global_rsv(root, 0);
3271 if (IS_ERR(trans)) {
3272 btrfs_free_path(path);
3273 return PTR_ERR(trans);
3276 key.objectid = BTRFS_BALANCE_OBJECTID;
3277 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3280 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3288 ret = btrfs_del_item(trans, root, path);
3290 btrfs_free_path(path);
3291 err = btrfs_commit_transaction(trans);
3298 * This is a heuristic used to reduce the number of chunks balanced on
3299 * resume after balance was interrupted.
3301 static void update_balance_args(struct btrfs_balance_control *bctl)
3304 * Turn on soft mode for chunk types that were being converted.
3306 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3307 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3308 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3309 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3310 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3311 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3314 * Turn on usage filter if is not already used. The idea is
3315 * that chunks that we have already balanced should be
3316 * reasonably full. Don't do it for chunks that are being
3317 * converted - that will keep us from relocating unconverted
3318 * (albeit full) chunks.
3320 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3321 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3322 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3323 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3324 bctl->data.usage = 90;
3326 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3327 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3328 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3329 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3330 bctl->sys.usage = 90;
3332 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3333 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3334 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3335 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3336 bctl->meta.usage = 90;
3341 * Clear the balance status in fs_info and delete the balance item from disk.
3343 static void reset_balance_state(struct btrfs_fs_info *fs_info)
3345 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3348 BUG_ON(!fs_info->balance_ctl);
3350 spin_lock(&fs_info->balance_lock);
3351 fs_info->balance_ctl = NULL;
3352 spin_unlock(&fs_info->balance_lock);
3355 ret = del_balance_item(fs_info);
3357 btrfs_handle_fs_error(fs_info, ret, NULL);
3361 * Balance filters. Return 1 if chunk should be filtered out
3362 * (should not be balanced).
3364 static int chunk_profiles_filter(u64 chunk_type,
3365 struct btrfs_balance_args *bargs)
3367 chunk_type = chunk_to_extended(chunk_type) &
3368 BTRFS_EXTENDED_PROFILE_MASK;
3370 if (bargs->profiles & chunk_type)
3376 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3377 struct btrfs_balance_args *bargs)
3379 struct btrfs_block_group *cache;
3381 u64 user_thresh_min;
3382 u64 user_thresh_max;
3385 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3386 chunk_used = cache->used;
3388 if (bargs->usage_min == 0)
3389 user_thresh_min = 0;
3391 user_thresh_min = div_factor_fine(cache->length,
3394 if (bargs->usage_max == 0)
3395 user_thresh_max = 1;
3396 else if (bargs->usage_max > 100)
3397 user_thresh_max = cache->length;
3399 user_thresh_max = div_factor_fine(cache->length,
3402 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3405 btrfs_put_block_group(cache);
3409 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3410 u64 chunk_offset, struct btrfs_balance_args *bargs)
3412 struct btrfs_block_group *cache;
3413 u64 chunk_used, user_thresh;
3416 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3417 chunk_used = cache->used;
3419 if (bargs->usage_min == 0)
3421 else if (bargs->usage > 100)
3422 user_thresh = cache->length;
3424 user_thresh = div_factor_fine(cache->length, bargs->usage);
3426 if (chunk_used < user_thresh)
3429 btrfs_put_block_group(cache);
3433 static int chunk_devid_filter(struct extent_buffer *leaf,
3434 struct btrfs_chunk *chunk,
3435 struct btrfs_balance_args *bargs)
3437 struct btrfs_stripe *stripe;
3438 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3441 for (i = 0; i < num_stripes; i++) {
3442 stripe = btrfs_stripe_nr(chunk, i);
3443 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3450 static u64 calc_data_stripes(u64 type, int num_stripes)
3452 const int index = btrfs_bg_flags_to_raid_index(type);
3453 const int ncopies = btrfs_raid_array[index].ncopies;
3454 const int nparity = btrfs_raid_array[index].nparity;
3457 return num_stripes - nparity;
3459 return num_stripes / ncopies;
3462 /* [pstart, pend) */
3463 static int chunk_drange_filter(struct extent_buffer *leaf,
3464 struct btrfs_chunk *chunk,
3465 struct btrfs_balance_args *bargs)
3467 struct btrfs_stripe *stripe;
3468 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3475 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3478 type = btrfs_chunk_type(leaf, chunk);
3479 factor = calc_data_stripes(type, num_stripes);
3481 for (i = 0; i < num_stripes; i++) {
3482 stripe = btrfs_stripe_nr(chunk, i);
3483 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3486 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3487 stripe_length = btrfs_chunk_length(leaf, chunk);
3488 stripe_length = div_u64(stripe_length, factor);
3490 if (stripe_offset < bargs->pend &&
3491 stripe_offset + stripe_length > bargs->pstart)
3498 /* [vstart, vend) */
3499 static int chunk_vrange_filter(struct extent_buffer *leaf,
3500 struct btrfs_chunk *chunk,
3502 struct btrfs_balance_args *bargs)
3504 if (chunk_offset < bargs->vend &&
3505 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3506 /* at least part of the chunk is inside this vrange */
3512 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3513 struct btrfs_chunk *chunk,
3514 struct btrfs_balance_args *bargs)
3516 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3518 if (bargs->stripes_min <= num_stripes
3519 && num_stripes <= bargs->stripes_max)
3525 static int chunk_soft_convert_filter(u64 chunk_type,
3526 struct btrfs_balance_args *bargs)
3528 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3531 chunk_type = chunk_to_extended(chunk_type) &
3532 BTRFS_EXTENDED_PROFILE_MASK;
3534 if (bargs->target == chunk_type)
3540 static int should_balance_chunk(struct extent_buffer *leaf,
3541 struct btrfs_chunk *chunk, u64 chunk_offset)
3543 struct btrfs_fs_info *fs_info = leaf->fs_info;
3544 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3545 struct btrfs_balance_args *bargs = NULL;
3546 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3549 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3550 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3554 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3555 bargs = &bctl->data;
3556 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3558 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3559 bargs = &bctl->meta;
3561 /* profiles filter */
3562 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3563 chunk_profiles_filter(chunk_type, bargs)) {
3568 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3569 chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3571 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3572 chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3577 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3578 chunk_devid_filter(leaf, chunk, bargs)) {
3582 /* drange filter, makes sense only with devid filter */
3583 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3584 chunk_drange_filter(leaf, chunk, bargs)) {
3589 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3590 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3594 /* stripes filter */
3595 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3596 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3600 /* soft profile changing mode */
3601 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3602 chunk_soft_convert_filter(chunk_type, bargs)) {
3607 * limited by count, must be the last filter
3609 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3610 if (bargs->limit == 0)
3614 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3616 * Same logic as the 'limit' filter; the minimum cannot be
3617 * determined here because we do not have the global information
3618 * about the count of all chunks that satisfy the filters.
3620 if (bargs->limit_max == 0)
3629 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3631 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3632 struct btrfs_root *chunk_root = fs_info->chunk_root;
3634 struct btrfs_chunk *chunk;
3635 struct btrfs_path *path = NULL;
3636 struct btrfs_key key;
3637 struct btrfs_key found_key;
3638 struct extent_buffer *leaf;
3641 int enospc_errors = 0;
3642 bool counting = true;
3643 /* The single value limit and min/max limits use the same bytes in the */
3644 u64 limit_data = bctl->data.limit;
3645 u64 limit_meta = bctl->meta.limit;
3646 u64 limit_sys = bctl->sys.limit;
3650 int chunk_reserved = 0;
3652 path = btrfs_alloc_path();
3658 /* zero out stat counters */
3659 spin_lock(&fs_info->balance_lock);
3660 memset(&bctl->stat, 0, sizeof(bctl->stat));
3661 spin_unlock(&fs_info->balance_lock);
3665 * The single value limit and min/max limits use the same bytes
3668 bctl->data.limit = limit_data;
3669 bctl->meta.limit = limit_meta;
3670 bctl->sys.limit = limit_sys;
3672 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3673 key.offset = (u64)-1;
3674 key.type = BTRFS_CHUNK_ITEM_KEY;
3677 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3678 atomic_read(&fs_info->balance_cancel_req)) {
3683 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3684 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3686 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3691 * this shouldn't happen, it means the last relocate
3695 BUG(); /* FIXME break ? */
3697 ret = btrfs_previous_item(chunk_root, path, 0,
3698 BTRFS_CHUNK_ITEM_KEY);
3700 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3705 leaf = path->nodes[0];
3706 slot = path->slots[0];
3707 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3709 if (found_key.objectid != key.objectid) {
3710 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3714 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3715 chunk_type = btrfs_chunk_type(leaf, chunk);
3718 spin_lock(&fs_info->balance_lock);
3719 bctl->stat.considered++;
3720 spin_unlock(&fs_info->balance_lock);
3723 ret = should_balance_chunk(leaf, chunk, found_key.offset);
3725 btrfs_release_path(path);
3727 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3732 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3733 spin_lock(&fs_info->balance_lock);
3734 bctl->stat.expected++;
3735 spin_unlock(&fs_info->balance_lock);
3737 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3739 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3741 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3748 * Apply limit_min filter, no need to check if the LIMITS
3749 * filter is used, limit_min is 0 by default
3751 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3752 count_data < bctl->data.limit_min)
3753 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3754 count_meta < bctl->meta.limit_min)
3755 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3756 count_sys < bctl->sys.limit_min)) {
3757 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3761 if (!chunk_reserved) {
3763 * We may be relocating the only data chunk we have,
3764 * which could potentially end up with losing data's
3765 * raid profile, so lets allocate an empty one in
3768 ret = btrfs_may_alloc_data_chunk(fs_info,
3771 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3773 } else if (ret == 1) {
3778 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3779 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3780 if (ret == -ENOSPC) {
3782 } else if (ret == -ETXTBSY) {
3784 "skipping relocation of block group %llu due to active swapfile",
3790 spin_lock(&fs_info->balance_lock);
3791 bctl->stat.completed++;
3792 spin_unlock(&fs_info->balance_lock);
3795 if (found_key.offset == 0)
3797 key.offset = found_key.offset - 1;
3801 btrfs_release_path(path);
3806 btrfs_free_path(path);
3807 if (enospc_errors) {
3808 btrfs_info(fs_info, "%d enospc errors during balance",
3818 * alloc_profile_is_valid - see if a given profile is valid and reduced
3819 * @flags: profile to validate
3820 * @extended: if true @flags is treated as an extended profile
3822 static int alloc_profile_is_valid(u64 flags, int extended)
3824 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3825 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3827 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3829 /* 1) check that all other bits are zeroed */
3833 /* 2) see if profile is reduced */
3835 return !extended; /* "0" is valid for usual profiles */
3837 return has_single_bit_set(flags);
3840 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3842 /* cancel requested || normal exit path */
3843 return atomic_read(&fs_info->balance_cancel_req) ||
3844 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3845 atomic_read(&fs_info->balance_cancel_req) == 0);
3849 * Validate target profile against allowed profiles and return true if it's OK.
3850 * Otherwise print the error message and return false.
3852 static inline int validate_convert_profile(struct btrfs_fs_info *fs_info,
3853 const struct btrfs_balance_args *bargs,
3854 u64 allowed, const char *type)
3856 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3859 /* Profile is valid and does not have bits outside of the allowed set */
3860 if (alloc_profile_is_valid(bargs->target, 1) &&
3861 (bargs->target & ~allowed) == 0)
3864 btrfs_err(fs_info, "balance: invalid convert %s profile %s",
3865 type, btrfs_bg_type_to_raid_name(bargs->target));
3870 * Fill @buf with textual description of balance filter flags @bargs, up to
3871 * @size_buf including the terminating null. The output may be trimmed if it
3872 * does not fit into the provided buffer.
3874 static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf,
3878 u32 size_bp = size_buf;
3880 u64 flags = bargs->flags;
3881 char tmp_buf[128] = {'\0'};
3886 #define CHECK_APPEND_NOARG(a) \
3888 ret = snprintf(bp, size_bp, (a)); \
3889 if (ret < 0 || ret >= size_bp) \
3890 goto out_overflow; \
3895 #define CHECK_APPEND_1ARG(a, v1) \
3897 ret = snprintf(bp, size_bp, (a), (v1)); \
3898 if (ret < 0 || ret >= size_bp) \
3899 goto out_overflow; \
3904 #define CHECK_APPEND_2ARG(a, v1, v2) \
3906 ret = snprintf(bp, size_bp, (a), (v1), (v2)); \
3907 if (ret < 0 || ret >= size_bp) \
3908 goto out_overflow; \
3913 if (flags & BTRFS_BALANCE_ARGS_CONVERT)
3914 CHECK_APPEND_1ARG("convert=%s,",
3915 btrfs_bg_type_to_raid_name(bargs->target));
3917 if (flags & BTRFS_BALANCE_ARGS_SOFT)
3918 CHECK_APPEND_NOARG("soft,");
3920 if (flags & BTRFS_BALANCE_ARGS_PROFILES) {
3921 btrfs_describe_block_groups(bargs->profiles, tmp_buf,
3923 CHECK_APPEND_1ARG("profiles=%s,", tmp_buf);
3926 if (flags & BTRFS_BALANCE_ARGS_USAGE)
3927 CHECK_APPEND_1ARG("usage=%llu,", bargs->usage);
3929 if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE)
3930 CHECK_APPEND_2ARG("usage=%u..%u,",
3931 bargs->usage_min, bargs->usage_max);
3933 if (flags & BTRFS_BALANCE_ARGS_DEVID)
3934 CHECK_APPEND_1ARG("devid=%llu,", bargs->devid);
3936 if (flags & BTRFS_BALANCE_ARGS_DRANGE)
3937 CHECK_APPEND_2ARG("drange=%llu..%llu,",
3938 bargs->pstart, bargs->pend);
3940 if (flags & BTRFS_BALANCE_ARGS_VRANGE)
3941 CHECK_APPEND_2ARG("vrange=%llu..%llu,",
3942 bargs->vstart, bargs->vend);
3944 if (flags & BTRFS_BALANCE_ARGS_LIMIT)
3945 CHECK_APPEND_1ARG("limit=%llu,", bargs->limit);
3947 if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)
3948 CHECK_APPEND_2ARG("limit=%u..%u,",
3949 bargs->limit_min, bargs->limit_max);
3951 if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE)
3952 CHECK_APPEND_2ARG("stripes=%u..%u,",
3953 bargs->stripes_min, bargs->stripes_max);
3955 #undef CHECK_APPEND_2ARG
3956 #undef CHECK_APPEND_1ARG
3957 #undef CHECK_APPEND_NOARG
3961 if (size_bp < size_buf)
3962 buf[size_buf - size_bp - 1] = '\0'; /* remove last , */
3967 static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info)
3969 u32 size_buf = 1024;
3970 char tmp_buf[192] = {'\0'};
3973 u32 size_bp = size_buf;
3975 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3977 buf = kzalloc(size_buf, GFP_KERNEL);
3983 #define CHECK_APPEND_1ARG(a, v1) \
3985 ret = snprintf(bp, size_bp, (a), (v1)); \
3986 if (ret < 0 || ret >= size_bp) \
3987 goto out_overflow; \
3992 if (bctl->flags & BTRFS_BALANCE_FORCE)
3993 CHECK_APPEND_1ARG("%s", "-f ");
3995 if (bctl->flags & BTRFS_BALANCE_DATA) {
3996 describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf));
3997 CHECK_APPEND_1ARG("-d%s ", tmp_buf);
4000 if (bctl->flags & BTRFS_BALANCE_METADATA) {
4001 describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf));
4002 CHECK_APPEND_1ARG("-m%s ", tmp_buf);
4005 if (bctl->flags & BTRFS_BALANCE_SYSTEM) {
4006 describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf));
4007 CHECK_APPEND_1ARG("-s%s ", tmp_buf);
4010 #undef CHECK_APPEND_1ARG
4014 if (size_bp < size_buf)
4015 buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */
4016 btrfs_info(fs_info, "balance: %s %s",
4017 (bctl->flags & BTRFS_BALANCE_RESUME) ?
4018 "resume" : "start", buf);
4024 * Should be called with balance mutexe held
4026 int btrfs_balance(struct btrfs_fs_info *fs_info,
4027 struct btrfs_balance_control *bctl,
4028 struct btrfs_ioctl_balance_args *bargs)
4030 u64 meta_target, data_target;
4036 bool reducing_redundancy;
4039 if (btrfs_fs_closing(fs_info) ||
4040 atomic_read(&fs_info->balance_pause_req) ||
4041 btrfs_should_cancel_balance(fs_info)) {
4046 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
4047 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
4051 * In case of mixed groups both data and meta should be picked,
4052 * and identical options should be given for both of them.
4054 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
4055 if (mixed && (bctl->flags & allowed)) {
4056 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
4057 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
4058 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
4060 "balance: mixed groups data and metadata options must be the same");
4067 * rw_devices will not change at the moment, device add/delete/replace
4070 num_devices = fs_info->fs_devices->rw_devices;
4073 * SINGLE profile on-disk has no profile bit, but in-memory we have a
4074 * special bit for it, to make it easier to distinguish. Thus we need
4075 * to set it manually, or balance would refuse the profile.
4077 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
4078 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++)
4079 if (num_devices >= btrfs_raid_array[i].devs_min)
4080 allowed |= btrfs_raid_array[i].bg_flag;
4082 if (!validate_convert_profile(fs_info, &bctl->data, allowed, "data") ||
4083 !validate_convert_profile(fs_info, &bctl->meta, allowed, "metadata") ||
4084 !validate_convert_profile(fs_info, &bctl->sys, allowed, "system")) {
4090 * Allow to reduce metadata or system integrity only if force set for
4091 * profiles with redundancy (copies, parity)
4094 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) {
4095 if (btrfs_raid_array[i].ncopies >= 2 ||
4096 btrfs_raid_array[i].tolerated_failures >= 1)
4097 allowed |= btrfs_raid_array[i].bg_flag;
4100 seq = read_seqbegin(&fs_info->profiles_lock);
4102 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4103 (fs_info->avail_system_alloc_bits & allowed) &&
4104 !(bctl->sys.target & allowed)) ||
4105 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4106 (fs_info->avail_metadata_alloc_bits & allowed) &&
4107 !(bctl->meta.target & allowed)))
4108 reducing_redundancy = true;
4110 reducing_redundancy = false;
4112 /* if we're not converting, the target field is uninitialized */
4113 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4114 bctl->meta.target : fs_info->avail_metadata_alloc_bits;
4115 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4116 bctl->data.target : fs_info->avail_data_alloc_bits;
4117 } while (read_seqretry(&fs_info->profiles_lock, seq));
4119 if (reducing_redundancy) {
4120 if (bctl->flags & BTRFS_BALANCE_FORCE) {
4122 "balance: force reducing metadata redundancy");
4125 "balance: reduces metadata redundancy, use --force if you want this");
4131 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <
4132 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {
4134 "balance: metadata profile %s has lower redundancy than data profile %s",
4135 btrfs_bg_type_to_raid_name(meta_target),
4136 btrfs_bg_type_to_raid_name(data_target));
4139 if (fs_info->send_in_progress) {
4140 btrfs_warn_rl(fs_info,
4141 "cannot run balance while send operations are in progress (%d in progress)",
4142 fs_info->send_in_progress);
4147 ret = insert_balance_item(fs_info, bctl);
4148 if (ret && ret != -EEXIST)
4151 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
4152 BUG_ON(ret == -EEXIST);
4153 BUG_ON(fs_info->balance_ctl);
4154 spin_lock(&fs_info->balance_lock);
4155 fs_info->balance_ctl = bctl;
4156 spin_unlock(&fs_info->balance_lock);
4158 BUG_ON(ret != -EEXIST);
4159 spin_lock(&fs_info->balance_lock);
4160 update_balance_args(bctl);
4161 spin_unlock(&fs_info->balance_lock);
4164 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4165 set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4166 describe_balance_start_or_resume(fs_info);
4167 mutex_unlock(&fs_info->balance_mutex);
4169 ret = __btrfs_balance(fs_info);
4171 mutex_lock(&fs_info->balance_mutex);
4172 if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req))
4173 btrfs_info(fs_info, "balance: paused");
4175 * Balance can be canceled by:
4177 * - Regular cancel request
4178 * Then ret == -ECANCELED and balance_cancel_req > 0
4180 * - Fatal signal to "btrfs" process
4181 * Either the signal caught by wait_reserve_ticket() and callers
4182 * got -EINTR, or caught by btrfs_should_cancel_balance() and
4184 * Either way, in this case balance_cancel_req = 0, and
4185 * ret == -EINTR or ret == -ECANCELED.
4187 * So here we only check the return value to catch canceled balance.
4189 else if (ret == -ECANCELED || ret == -EINTR)
4190 btrfs_info(fs_info, "balance: canceled");
4192 btrfs_info(fs_info, "balance: ended with status: %d", ret);
4194 clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4197 memset(bargs, 0, sizeof(*bargs));
4198 btrfs_update_ioctl_balance_args(fs_info, bargs);
4201 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
4202 balance_need_close(fs_info)) {
4203 reset_balance_state(fs_info);
4204 btrfs_exclop_finish(fs_info);
4207 wake_up(&fs_info->balance_wait_q);
4211 if (bctl->flags & BTRFS_BALANCE_RESUME)
4212 reset_balance_state(fs_info);
4215 btrfs_exclop_finish(fs_info);
4220 static int balance_kthread(void *data)
4222 struct btrfs_fs_info *fs_info = data;
4225 mutex_lock(&fs_info->balance_mutex);
4226 if (fs_info->balance_ctl)
4227 ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL);
4228 mutex_unlock(&fs_info->balance_mutex);
4233 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
4235 struct task_struct *tsk;
4237 mutex_lock(&fs_info->balance_mutex);
4238 if (!fs_info->balance_ctl) {
4239 mutex_unlock(&fs_info->balance_mutex);
4242 mutex_unlock(&fs_info->balance_mutex);
4244 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
4245 btrfs_info(fs_info, "balance: resume skipped");
4250 * A ro->rw remount sequence should continue with the paused balance
4251 * regardless of who pauses it, system or the user as of now, so set
4254 spin_lock(&fs_info->balance_lock);
4255 fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME;
4256 spin_unlock(&fs_info->balance_lock);
4258 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
4259 return PTR_ERR_OR_ZERO(tsk);
4262 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
4264 struct btrfs_balance_control *bctl;
4265 struct btrfs_balance_item *item;
4266 struct btrfs_disk_balance_args disk_bargs;
4267 struct btrfs_path *path;
4268 struct extent_buffer *leaf;
4269 struct btrfs_key key;
4272 path = btrfs_alloc_path();
4276 key.objectid = BTRFS_BALANCE_OBJECTID;
4277 key.type = BTRFS_TEMPORARY_ITEM_KEY;
4280 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4283 if (ret > 0) { /* ret = -ENOENT; */
4288 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4294 leaf = path->nodes[0];
4295 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
4297 bctl->flags = btrfs_balance_flags(leaf, item);
4298 bctl->flags |= BTRFS_BALANCE_RESUME;
4300 btrfs_balance_data(leaf, item, &disk_bargs);
4301 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4302 btrfs_balance_meta(leaf, item, &disk_bargs);
4303 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4304 btrfs_balance_sys(leaf, item, &disk_bargs);
4305 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4308 * This should never happen, as the paused balance state is recovered
4309 * during mount without any chance of other exclusive ops to collide.
4311 * This gives the exclusive op status to balance and keeps in paused
4312 * state until user intervention (cancel or umount). If the ownership
4313 * cannot be assigned, show a message but do not fail. The balance
4314 * is in a paused state and must have fs_info::balance_ctl properly
4317 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE))
4319 "balance: cannot set exclusive op status, resume manually");
4321 mutex_lock(&fs_info->balance_mutex);
4322 BUG_ON(fs_info->balance_ctl);
4323 spin_lock(&fs_info->balance_lock);
4324 fs_info->balance_ctl = bctl;
4325 spin_unlock(&fs_info->balance_lock);
4326 mutex_unlock(&fs_info->balance_mutex);
4328 btrfs_free_path(path);
4332 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4336 mutex_lock(&fs_info->balance_mutex);
4337 if (!fs_info->balance_ctl) {
4338 mutex_unlock(&fs_info->balance_mutex);
4342 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4343 atomic_inc(&fs_info->balance_pause_req);
4344 mutex_unlock(&fs_info->balance_mutex);
4346 wait_event(fs_info->balance_wait_q,
4347 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4349 mutex_lock(&fs_info->balance_mutex);
4350 /* we are good with balance_ctl ripped off from under us */
4351 BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4352 atomic_dec(&fs_info->balance_pause_req);
4357 mutex_unlock(&fs_info->balance_mutex);
4361 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4363 mutex_lock(&fs_info->balance_mutex);
4364 if (!fs_info->balance_ctl) {
4365 mutex_unlock(&fs_info->balance_mutex);
4370 * A paused balance with the item stored on disk can be resumed at
4371 * mount time if the mount is read-write. Otherwise it's still paused
4372 * and we must not allow cancelling as it deletes the item.
4374 if (sb_rdonly(fs_info->sb)) {
4375 mutex_unlock(&fs_info->balance_mutex);
4379 atomic_inc(&fs_info->balance_cancel_req);
4381 * if we are running just wait and return, balance item is
4382 * deleted in btrfs_balance in this case
4384 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4385 mutex_unlock(&fs_info->balance_mutex);
4386 wait_event(fs_info->balance_wait_q,
4387 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4388 mutex_lock(&fs_info->balance_mutex);
4390 mutex_unlock(&fs_info->balance_mutex);
4392 * Lock released to allow other waiters to continue, we'll
4393 * reexamine the status again.
4395 mutex_lock(&fs_info->balance_mutex);
4397 if (fs_info->balance_ctl) {
4398 reset_balance_state(fs_info);
4399 btrfs_exclop_finish(fs_info);
4400 btrfs_info(fs_info, "balance: canceled");
4404 BUG_ON(fs_info->balance_ctl ||
4405 test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4406 atomic_dec(&fs_info->balance_cancel_req);
4407 mutex_unlock(&fs_info->balance_mutex);
4411 int btrfs_uuid_scan_kthread(void *data)
4413 struct btrfs_fs_info *fs_info = data;
4414 struct btrfs_root *root = fs_info->tree_root;
4415 struct btrfs_key key;
4416 struct btrfs_path *path = NULL;
4418 struct extent_buffer *eb;
4420 struct btrfs_root_item root_item;
4422 struct btrfs_trans_handle *trans = NULL;
4423 bool closing = false;
4425 path = btrfs_alloc_path();
4432 key.type = BTRFS_ROOT_ITEM_KEY;
4436 if (btrfs_fs_closing(fs_info)) {
4440 ret = btrfs_search_forward(root, &key, path,
4441 BTRFS_OLDEST_GENERATION);
4448 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4449 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4450 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4451 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4454 eb = path->nodes[0];
4455 slot = path->slots[0];
4456 item_size = btrfs_item_size_nr(eb, slot);
4457 if (item_size < sizeof(root_item))
4460 read_extent_buffer(eb, &root_item,
4461 btrfs_item_ptr_offset(eb, slot),
4462 (int)sizeof(root_item));
4463 if (btrfs_root_refs(&root_item) == 0)
4466 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4467 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4471 btrfs_release_path(path);
4473 * 1 - subvol uuid item
4474 * 1 - received_subvol uuid item
4476 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4477 if (IS_ERR(trans)) {
4478 ret = PTR_ERR(trans);
4486 btrfs_release_path(path);
4487 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4488 ret = btrfs_uuid_tree_add(trans, root_item.uuid,
4489 BTRFS_UUID_KEY_SUBVOL,
4492 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4498 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4499 ret = btrfs_uuid_tree_add(trans,
4500 root_item.received_uuid,
4501 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4504 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4511 btrfs_release_path(path);
4513 ret = btrfs_end_transaction(trans);
4519 if (key.offset < (u64)-1) {
4521 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4523 key.type = BTRFS_ROOT_ITEM_KEY;
4524 } else if (key.objectid < (u64)-1) {
4526 key.type = BTRFS_ROOT_ITEM_KEY;
4535 btrfs_free_path(path);
4536 if (trans && !IS_ERR(trans))
4537 btrfs_end_transaction(trans);
4539 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4541 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4542 up(&fs_info->uuid_tree_rescan_sem);
4546 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4548 struct btrfs_trans_handle *trans;
4549 struct btrfs_root *tree_root = fs_info->tree_root;
4550 struct btrfs_root *uuid_root;
4551 struct task_struct *task;
4558 trans = btrfs_start_transaction(tree_root, 2);
4560 return PTR_ERR(trans);
4562 uuid_root = btrfs_create_tree(trans, BTRFS_UUID_TREE_OBJECTID);
4563 if (IS_ERR(uuid_root)) {
4564 ret = PTR_ERR(uuid_root);
4565 btrfs_abort_transaction(trans, ret);
4566 btrfs_end_transaction(trans);
4570 fs_info->uuid_root = uuid_root;
4572 ret = btrfs_commit_transaction(trans);
4576 down(&fs_info->uuid_tree_rescan_sem);
4577 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4579 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4580 btrfs_warn(fs_info, "failed to start uuid_scan task");
4581 up(&fs_info->uuid_tree_rescan_sem);
4582 return PTR_ERR(task);
4589 * shrinking a device means finding all of the device extents past
4590 * the new size, and then following the back refs to the chunks.
4591 * The chunk relocation code actually frees the device extent
4593 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4595 struct btrfs_fs_info *fs_info = device->fs_info;
4596 struct btrfs_root *root = fs_info->dev_root;
4597 struct btrfs_trans_handle *trans;
4598 struct btrfs_dev_extent *dev_extent = NULL;
4599 struct btrfs_path *path;
4605 bool retried = false;
4606 struct extent_buffer *l;
4607 struct btrfs_key key;
4608 struct btrfs_super_block *super_copy = fs_info->super_copy;
4609 u64 old_total = btrfs_super_total_bytes(super_copy);
4610 u64 old_size = btrfs_device_get_total_bytes(device);
4614 new_size = round_down(new_size, fs_info->sectorsize);
4616 diff = round_down(old_size - new_size, fs_info->sectorsize);
4618 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4621 path = btrfs_alloc_path();
4625 path->reada = READA_BACK;
4627 trans = btrfs_start_transaction(root, 0);
4628 if (IS_ERR(trans)) {
4629 btrfs_free_path(path);
4630 return PTR_ERR(trans);
4633 mutex_lock(&fs_info->chunk_mutex);
4635 btrfs_device_set_total_bytes(device, new_size);
4636 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4637 device->fs_devices->total_rw_bytes -= diff;
4638 atomic64_sub(diff, &fs_info->free_chunk_space);
4642 * Once the device's size has been set to the new size, ensure all
4643 * in-memory chunks are synced to disk so that the loop below sees them
4644 * and relocates them accordingly.
4646 if (contains_pending_extent(device, &start, diff)) {
4647 mutex_unlock(&fs_info->chunk_mutex);
4648 ret = btrfs_commit_transaction(trans);
4652 mutex_unlock(&fs_info->chunk_mutex);
4653 btrfs_end_transaction(trans);
4657 key.objectid = device->devid;
4658 key.offset = (u64)-1;
4659 key.type = BTRFS_DEV_EXTENT_KEY;
4662 mutex_lock(&fs_info->delete_unused_bgs_mutex);
4663 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4665 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4669 ret = btrfs_previous_item(root, path, 0, key.type);
4671 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4676 btrfs_release_path(path);
4681 slot = path->slots[0];
4682 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4684 if (key.objectid != device->devid) {
4685 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4686 btrfs_release_path(path);
4690 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4691 length = btrfs_dev_extent_length(l, dev_extent);
4693 if (key.offset + length <= new_size) {
4694 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4695 btrfs_release_path(path);
4699 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4700 btrfs_release_path(path);
4703 * We may be relocating the only data chunk we have,
4704 * which could potentially end up with losing data's
4705 * raid profile, so lets allocate an empty one in
4708 ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset);
4710 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4714 ret = btrfs_relocate_chunk(fs_info, chunk_offset);
4715 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4716 if (ret == -ENOSPC) {
4719 if (ret == -ETXTBSY) {
4721 "could not shrink block group %llu due to active swapfile",
4726 } while (key.offset-- > 0);
4728 if (failed && !retried) {
4732 } else if (failed && retried) {
4737 /* Shrinking succeeded, else we would be at "done". */
4738 trans = btrfs_start_transaction(root, 0);
4739 if (IS_ERR(trans)) {
4740 ret = PTR_ERR(trans);
4744 mutex_lock(&fs_info->chunk_mutex);
4745 /* Clear all state bits beyond the shrunk device size */
4746 clear_extent_bits(&device->alloc_state, new_size, (u64)-1,
4749 btrfs_device_set_disk_total_bytes(device, new_size);
4750 if (list_empty(&device->post_commit_list))
4751 list_add_tail(&device->post_commit_list,
4752 &trans->transaction->dev_update_list);
4754 WARN_ON(diff > old_total);
4755 btrfs_set_super_total_bytes(super_copy,
4756 round_down(old_total - diff, fs_info->sectorsize));
4757 mutex_unlock(&fs_info->chunk_mutex);
4759 /* Now btrfs_update_device() will change the on-disk size. */
4760 ret = btrfs_update_device(trans, device);
4762 btrfs_abort_transaction(trans, ret);
4763 btrfs_end_transaction(trans);
4765 ret = btrfs_commit_transaction(trans);
4768 btrfs_free_path(path);
4770 mutex_lock(&fs_info->chunk_mutex);
4771 btrfs_device_set_total_bytes(device, old_size);
4772 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
4773 device->fs_devices->total_rw_bytes += diff;
4774 atomic64_add(diff, &fs_info->free_chunk_space);
4775 mutex_unlock(&fs_info->chunk_mutex);
4780 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
4781 struct btrfs_key *key,
4782 struct btrfs_chunk *chunk, int item_size)
4784 struct btrfs_super_block *super_copy = fs_info->super_copy;
4785 struct btrfs_disk_key disk_key;
4789 mutex_lock(&fs_info->chunk_mutex);
4790 array_size = btrfs_super_sys_array_size(super_copy);
4791 if (array_size + item_size + sizeof(disk_key)
4792 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4793 mutex_unlock(&fs_info->chunk_mutex);
4797 ptr = super_copy->sys_chunk_array + array_size;
4798 btrfs_cpu_key_to_disk(&disk_key, key);
4799 memcpy(ptr, &disk_key, sizeof(disk_key));
4800 ptr += sizeof(disk_key);
4801 memcpy(ptr, chunk, item_size);
4802 item_size += sizeof(disk_key);
4803 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4804 mutex_unlock(&fs_info->chunk_mutex);
4810 * sort the devices in descending order by max_avail, total_avail
4812 static int btrfs_cmp_device_info(const void *a, const void *b)
4814 const struct btrfs_device_info *di_a = a;
4815 const struct btrfs_device_info *di_b = b;
4817 if (di_a->max_avail > di_b->max_avail)
4819 if (di_a->max_avail < di_b->max_avail)
4821 if (di_a->total_avail > di_b->total_avail)
4823 if (di_a->total_avail < di_b->total_avail)
4828 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4830 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4833 btrfs_set_fs_incompat(info, RAID56);
4836 static void check_raid1c34_incompat_flag(struct btrfs_fs_info *info, u64 type)
4838 if (!(type & (BTRFS_BLOCK_GROUP_RAID1C3 | BTRFS_BLOCK_GROUP_RAID1C4)))
4841 btrfs_set_fs_incompat(info, RAID1C34);
4845 * Structure used internally for __btrfs_alloc_chunk() function.
4846 * Wraps needed parameters.
4848 struct alloc_chunk_ctl {
4851 /* Total number of stripes to allocate */
4853 /* sub_stripes info for map */
4855 /* Stripes per device */
4857 /* Maximum number of devices to use */
4859 /* Minimum number of devices to use */
4861 /* ndevs has to be a multiple of this */
4863 /* Number of copies */
4865 /* Number of stripes worth of bytes to store parity information */
4867 u64 max_stripe_size;
4875 static void init_alloc_chunk_ctl_policy_regular(
4876 struct btrfs_fs_devices *fs_devices,
4877 struct alloc_chunk_ctl *ctl)
4879 u64 type = ctl->type;
4881 if (type & BTRFS_BLOCK_GROUP_DATA) {
4882 ctl->max_stripe_size = SZ_1G;
4883 ctl->max_chunk_size = BTRFS_MAX_DATA_CHUNK_SIZE;
4884 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4885 /* For larger filesystems, use larger metadata chunks */
4886 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
4887 ctl->max_stripe_size = SZ_1G;
4889 ctl->max_stripe_size = SZ_256M;
4890 ctl->max_chunk_size = ctl->max_stripe_size;
4891 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4892 ctl->max_stripe_size = SZ_32M;
4893 ctl->max_chunk_size = 2 * ctl->max_stripe_size;
4894 ctl->devs_max = min_t(int, ctl->devs_max,
4895 BTRFS_MAX_DEVS_SYS_CHUNK);
4900 /* We don't want a chunk larger than 10% of writable space */
4901 ctl->max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4902 ctl->max_chunk_size);
4903 ctl->dev_extent_min = BTRFS_STRIPE_LEN * ctl->dev_stripes;
4906 static void init_alloc_chunk_ctl(struct btrfs_fs_devices *fs_devices,
4907 struct alloc_chunk_ctl *ctl)
4909 int index = btrfs_bg_flags_to_raid_index(ctl->type);
4911 ctl->sub_stripes = btrfs_raid_array[index].sub_stripes;
4912 ctl->dev_stripes = btrfs_raid_array[index].dev_stripes;
4913 ctl->devs_max = btrfs_raid_array[index].devs_max;
4915 ctl->devs_max = BTRFS_MAX_DEVS(fs_devices->fs_info);
4916 ctl->devs_min = btrfs_raid_array[index].devs_min;
4917 ctl->devs_increment = btrfs_raid_array[index].devs_increment;
4918 ctl->ncopies = btrfs_raid_array[index].ncopies;
4919 ctl->nparity = btrfs_raid_array[index].nparity;
4922 switch (fs_devices->chunk_alloc_policy) {
4923 case BTRFS_CHUNK_ALLOC_REGULAR:
4924 init_alloc_chunk_ctl_policy_regular(fs_devices, ctl);
4931 static int gather_device_info(struct btrfs_fs_devices *fs_devices,
4932 struct alloc_chunk_ctl *ctl,
4933 struct btrfs_device_info *devices_info)
4935 struct btrfs_fs_info *info = fs_devices->fs_info;
4936 struct btrfs_device *device;
4938 u64 dev_extent_want = ctl->max_stripe_size * ctl->dev_stripes;
4945 * in the first pass through the devices list, we gather information
4946 * about the available holes on each device.
4948 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
4949 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4951 "BTRFS: read-only device in alloc_list\n");
4955 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
4956 &device->dev_state) ||
4957 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4960 if (device->total_bytes > device->bytes_used)
4961 total_avail = device->total_bytes - device->bytes_used;
4965 /* If there is no space on this device, skip it. */
4966 if (total_avail < ctl->dev_extent_min)
4969 ret = find_free_dev_extent(device, dev_extent_want, &dev_offset,
4971 if (ret && ret != -ENOSPC)
4975 max_avail = dev_extent_want;
4977 if (max_avail < ctl->dev_extent_min) {
4978 if (btrfs_test_opt(info, ENOSPC_DEBUG))
4980 "%s: devid %llu has no free space, have=%llu want=%llu",
4981 __func__, device->devid, max_avail,
4982 ctl->dev_extent_min);
4986 if (ndevs == fs_devices->rw_devices) {
4987 WARN(1, "%s: found more than %llu devices\n",
4988 __func__, fs_devices->rw_devices);
4991 devices_info[ndevs].dev_offset = dev_offset;
4992 devices_info[ndevs].max_avail = max_avail;
4993 devices_info[ndevs].total_avail = total_avail;
4994 devices_info[ndevs].dev = device;
5000 * now sort the devices by hole size / available space
5002 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
5003 btrfs_cmp_device_info, NULL);
5008 static int decide_stripe_size_regular(struct alloc_chunk_ctl *ctl,
5009 struct btrfs_device_info *devices_info)
5011 /* Number of stripes that count for block group size */
5015 * The primary goal is to maximize the number of stripes, so use as
5016 * many devices as possible, even if the stripes are not maximum sized.
5018 * The DUP profile stores more than one stripe per device, the
5019 * max_avail is the total size so we have to adjust.
5021 ctl->stripe_size = div_u64(devices_info[ctl->ndevs - 1].max_avail,
5023 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5025 /* This will have to be fixed for RAID1 and RAID10 over more drives */
5026 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5029 * Use the number of data stripes to figure out how big this chunk is
5030 * really going to be in terms of logical address space, and compare
5031 * that answer with the max chunk size. If it's higher, we try to
5032 * reduce stripe_size.
5034 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) {
5036 * Reduce stripe_size, round it up to a 16MB boundary again and
5037 * then use it, unless it ends up being even bigger than the
5038 * previous value we had already.
5040 ctl->stripe_size = min(round_up(div_u64(ctl->max_chunk_size,
5041 data_stripes), SZ_16M),
5045 /* Align to BTRFS_STRIPE_LEN */
5046 ctl->stripe_size = round_down(ctl->stripe_size, BTRFS_STRIPE_LEN);
5047 ctl->chunk_size = ctl->stripe_size * data_stripes;
5052 static int decide_stripe_size(struct btrfs_fs_devices *fs_devices,
5053 struct alloc_chunk_ctl *ctl,
5054 struct btrfs_device_info *devices_info)
5056 struct btrfs_fs_info *info = fs_devices->fs_info;
5059 * Round down to number of usable stripes, devs_increment can be any
5060 * number so we can't use round_down() that requires power of 2, while
5061 * rounddown is safe.
5063 ctl->ndevs = rounddown(ctl->ndevs, ctl->devs_increment);
5065 if (ctl->ndevs < ctl->devs_min) {
5066 if (btrfs_test_opt(info, ENOSPC_DEBUG)) {
5068 "%s: not enough devices with free space: have=%d minimum required=%d",
5069 __func__, ctl->ndevs, ctl->devs_min);
5074 ctl->ndevs = min(ctl->ndevs, ctl->devs_max);
5076 switch (fs_devices->chunk_alloc_policy) {
5077 case BTRFS_CHUNK_ALLOC_REGULAR:
5078 return decide_stripe_size_regular(ctl, devices_info);
5084 static int create_chunk(struct btrfs_trans_handle *trans,
5085 struct alloc_chunk_ctl *ctl,
5086 struct btrfs_device_info *devices_info)
5088 struct btrfs_fs_info *info = trans->fs_info;
5089 struct map_lookup *map = NULL;
5090 struct extent_map_tree *em_tree;
5091 struct extent_map *em;
5092 u64 start = ctl->start;
5093 u64 type = ctl->type;
5098 map = kmalloc(map_lookup_size(ctl->num_stripes), GFP_NOFS);
5101 map->num_stripes = ctl->num_stripes;
5103 for (i = 0; i < ctl->ndevs; ++i) {
5104 for (j = 0; j < ctl->dev_stripes; ++j) {
5105 int s = i * ctl->dev_stripes + j;
5106 map->stripes[s].dev = devices_info[i].dev;
5107 map->stripes[s].physical = devices_info[i].dev_offset +
5108 j * ctl->stripe_size;
5111 map->stripe_len = BTRFS_STRIPE_LEN;
5112 map->io_align = BTRFS_STRIPE_LEN;
5113 map->io_width = BTRFS_STRIPE_LEN;
5115 map->sub_stripes = ctl->sub_stripes;
5117 trace_btrfs_chunk_alloc(info, map, start, ctl->chunk_size);
5119 em = alloc_extent_map();
5124 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
5125 em->map_lookup = map;
5127 em->len = ctl->chunk_size;
5128 em->block_start = 0;
5129 em->block_len = em->len;
5130 em->orig_block_len = ctl->stripe_size;
5132 em_tree = &info->mapping_tree;
5133 write_lock(&em_tree->lock);
5134 ret = add_extent_mapping(em_tree, em, 0);
5136 write_unlock(&em_tree->lock);
5137 free_extent_map(em);
5140 write_unlock(&em_tree->lock);
5142 ret = btrfs_make_block_group(trans, 0, type, start, ctl->chunk_size);
5144 goto error_del_extent;
5146 for (i = 0; i < map->num_stripes; i++) {
5147 struct btrfs_device *dev = map->stripes[i].dev;
5149 btrfs_device_set_bytes_used(dev,
5150 dev->bytes_used + ctl->stripe_size);
5151 if (list_empty(&dev->post_commit_list))
5152 list_add_tail(&dev->post_commit_list,
5153 &trans->transaction->dev_update_list);
5156 atomic64_sub(ctl->stripe_size * map->num_stripes,
5157 &info->free_chunk_space);
5159 free_extent_map(em);
5160 check_raid56_incompat_flag(info, type);
5161 check_raid1c34_incompat_flag(info, type);
5166 write_lock(&em_tree->lock);
5167 remove_extent_mapping(em_tree, em);
5168 write_unlock(&em_tree->lock);
5170 /* One for our allocation */
5171 free_extent_map(em);
5172 /* One for the tree reference */
5173 free_extent_map(em);
5178 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, u64 type)
5180 struct btrfs_fs_info *info = trans->fs_info;
5181 struct btrfs_fs_devices *fs_devices = info->fs_devices;
5182 struct btrfs_device_info *devices_info = NULL;
5183 struct alloc_chunk_ctl ctl;
5186 lockdep_assert_held(&info->chunk_mutex);
5188 if (!alloc_profile_is_valid(type, 0)) {
5193 if (list_empty(&fs_devices->alloc_list)) {
5194 if (btrfs_test_opt(info, ENOSPC_DEBUG))
5195 btrfs_debug(info, "%s: no writable device", __func__);
5199 if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
5200 btrfs_err(info, "invalid chunk type 0x%llx requested", type);
5205 ctl.start = find_next_chunk(info);
5207 init_alloc_chunk_ctl(fs_devices, &ctl);
5209 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
5214 ret = gather_device_info(fs_devices, &ctl, devices_info);
5218 ret = decide_stripe_size(fs_devices, &ctl, devices_info);
5222 ret = create_chunk(trans, &ctl, devices_info);
5225 kfree(devices_info);
5230 * Chunk allocation falls into two parts. The first part does work
5231 * that makes the new allocated chunk usable, but does not do any operation
5232 * that modifies the chunk tree. The second part does the work that
5233 * requires modifying the chunk tree. This division is important for the
5234 * bootstrap process of adding storage to a seed btrfs.
5236 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
5237 u64 chunk_offset, u64 chunk_size)
5239 struct btrfs_fs_info *fs_info = trans->fs_info;
5240 struct btrfs_root *extent_root = fs_info->extent_root;
5241 struct btrfs_root *chunk_root = fs_info->chunk_root;
5242 struct btrfs_key key;
5243 struct btrfs_device *device;
5244 struct btrfs_chunk *chunk;
5245 struct btrfs_stripe *stripe;
5246 struct extent_map *em;
5247 struct map_lookup *map;
5254 em = btrfs_get_chunk_map(fs_info, chunk_offset, chunk_size);
5258 map = em->map_lookup;
5259 item_size = btrfs_chunk_item_size(map->num_stripes);
5260 stripe_size = em->orig_block_len;
5262 chunk = kzalloc(item_size, GFP_NOFS);
5269 * Take the device list mutex to prevent races with the final phase of
5270 * a device replace operation that replaces the device object associated
5271 * with the map's stripes, because the device object's id can change
5272 * at any time during that final phase of the device replace operation
5273 * (dev-replace.c:btrfs_dev_replace_finishing()).
5275 mutex_lock(&fs_info->fs_devices->device_list_mutex);
5276 for (i = 0; i < map->num_stripes; i++) {
5277 device = map->stripes[i].dev;
5278 dev_offset = map->stripes[i].physical;
5280 ret = btrfs_update_device(trans, device);
5283 ret = btrfs_alloc_dev_extent(trans, device, chunk_offset,
5284 dev_offset, stripe_size);
5289 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
5293 stripe = &chunk->stripe;
5294 for (i = 0; i < map->num_stripes; i++) {
5295 device = map->stripes[i].dev;
5296 dev_offset = map->stripes[i].physical;
5298 btrfs_set_stack_stripe_devid(stripe, device->devid);
5299 btrfs_set_stack_stripe_offset(stripe, dev_offset);
5300 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
5303 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
5305 btrfs_set_stack_chunk_length(chunk, chunk_size);
5306 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
5307 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
5308 btrfs_set_stack_chunk_type(chunk, map->type);
5309 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
5310 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
5311 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
5312 btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
5313 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
5315 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
5316 key.type = BTRFS_CHUNK_ITEM_KEY;
5317 key.offset = chunk_offset;
5319 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
5320 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
5322 * TODO: Cleanup of inserted chunk root in case of
5325 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
5330 free_extent_map(em);
5334 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans)
5336 struct btrfs_fs_info *fs_info = trans->fs_info;
5340 alloc_profile = btrfs_metadata_alloc_profile(fs_info);
5341 ret = btrfs_alloc_chunk(trans, alloc_profile);
5345 alloc_profile = btrfs_system_alloc_profile(fs_info);
5346 ret = btrfs_alloc_chunk(trans, alloc_profile);
5350 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
5352 const int index = btrfs_bg_flags_to_raid_index(map->type);
5354 return btrfs_raid_array[index].tolerated_failures;
5357 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset)
5359 struct extent_map *em;
5360 struct map_lookup *map;
5365 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
5369 map = em->map_lookup;
5370 for (i = 0; i < map->num_stripes; i++) {
5371 if (test_bit(BTRFS_DEV_STATE_MISSING,
5372 &map->stripes[i].dev->dev_state)) {
5376 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE,
5377 &map->stripes[i].dev->dev_state)) {
5384 * If the number of missing devices is larger than max errors,
5385 * we can not write the data into that chunk successfully, so
5388 if (miss_ndevs > btrfs_chunk_max_errors(map))
5391 free_extent_map(em);
5395 void btrfs_mapping_tree_free(struct extent_map_tree *tree)
5397 struct extent_map *em;
5400 write_lock(&tree->lock);
5401 em = lookup_extent_mapping(tree, 0, (u64)-1);
5403 remove_extent_mapping(tree, em);
5404 write_unlock(&tree->lock);
5408 free_extent_map(em);
5409 /* once for the tree */
5410 free_extent_map(em);
5414 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5416 struct extent_map *em;
5417 struct map_lookup *map;
5420 em = btrfs_get_chunk_map(fs_info, logical, len);
5423 * We could return errors for these cases, but that could get
5424 * ugly and we'd probably do the same thing which is just not do
5425 * anything else and exit, so return 1 so the callers don't try
5426 * to use other copies.
5430 map = em->map_lookup;
5431 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1_MASK))
5432 ret = map->num_stripes;
5433 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5434 ret = map->sub_stripes;
5435 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5437 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5439 * There could be two corrupted data stripes, we need
5440 * to loop retry in order to rebuild the correct data.
5442 * Fail a stripe at a time on every retry except the
5443 * stripe under reconstruction.
5445 ret = map->num_stripes;
5448 free_extent_map(em);
5450 down_read(&fs_info->dev_replace.rwsem);
5451 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) &&
5452 fs_info->dev_replace.tgtdev)
5454 up_read(&fs_info->dev_replace.rwsem);
5459 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
5462 struct extent_map *em;
5463 struct map_lookup *map;
5464 unsigned long len = fs_info->sectorsize;
5466 em = btrfs_get_chunk_map(fs_info, logical, len);
5468 if (!WARN_ON(IS_ERR(em))) {
5469 map = em->map_lookup;
5470 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5471 len = map->stripe_len * nr_data_stripes(map);
5472 free_extent_map(em);
5477 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5479 struct extent_map *em;
5480 struct map_lookup *map;
5483 em = btrfs_get_chunk_map(fs_info, logical, len);
5485 if(!WARN_ON(IS_ERR(em))) {
5486 map = em->map_lookup;
5487 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5489 free_extent_map(em);
5494 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5495 struct map_lookup *map, int first,
5496 int dev_replace_is_ongoing)
5500 int preferred_mirror;
5502 struct btrfs_device *srcdev;
5505 (BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10)));
5507 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5508 num_stripes = map->sub_stripes;
5510 num_stripes = map->num_stripes;
5512 switch (fs_info->fs_devices->read_policy) {
5514 /* Shouldn't happen, just warn and use pid instead of failing */
5515 btrfs_warn_rl(fs_info,
5516 "unknown read_policy type %u, reset to pid",
5517 fs_info->fs_devices->read_policy);
5518 fs_info->fs_devices->read_policy = BTRFS_READ_POLICY_PID;
5520 case BTRFS_READ_POLICY_PID:
5521 preferred_mirror = first + (current->pid % num_stripes);
5525 if (dev_replace_is_ongoing &&
5526 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5527 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5528 srcdev = fs_info->dev_replace.srcdev;
5533 * try to avoid the drive that is the source drive for a
5534 * dev-replace procedure, only choose it if no other non-missing
5535 * mirror is available
5537 for (tolerance = 0; tolerance < 2; tolerance++) {
5538 if (map->stripes[preferred_mirror].dev->bdev &&
5539 (tolerance || map->stripes[preferred_mirror].dev != srcdev))
5540 return preferred_mirror;
5541 for (i = first; i < first + num_stripes; i++) {
5542 if (map->stripes[i].dev->bdev &&
5543 (tolerance || map->stripes[i].dev != srcdev))
5548 /* we couldn't find one that doesn't fail. Just return something
5549 * and the io error handling code will clean up eventually
5551 return preferred_mirror;
5554 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5555 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5562 for (i = 0; i < num_stripes - 1; i++) {
5563 /* Swap if parity is on a smaller index */
5564 if (bbio->raid_map[i] > bbio->raid_map[i + 1]) {
5565 swap(bbio->stripes[i], bbio->stripes[i + 1]);
5566 swap(bbio->raid_map[i], bbio->raid_map[i + 1]);
5573 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5575 struct btrfs_bio *bbio = kzalloc(
5576 /* the size of the btrfs_bio */
5577 sizeof(struct btrfs_bio) +
5578 /* plus the variable array for the stripes */
5579 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5580 /* plus the variable array for the tgt dev */
5581 sizeof(int) * (real_stripes) +
5583 * plus the raid_map, which includes both the tgt dev
5586 sizeof(u64) * (total_stripes),
5587 GFP_NOFS|__GFP_NOFAIL);
5589 atomic_set(&bbio->error, 0);
5590 refcount_set(&bbio->refs, 1);
5592 bbio->tgtdev_map = (int *)(bbio->stripes + total_stripes);
5593 bbio->raid_map = (u64 *)(bbio->tgtdev_map + real_stripes);
5598 void btrfs_get_bbio(struct btrfs_bio *bbio)
5600 WARN_ON(!refcount_read(&bbio->refs));
5601 refcount_inc(&bbio->refs);
5604 void btrfs_put_bbio(struct btrfs_bio *bbio)
5608 if (refcount_dec_and_test(&bbio->refs))
5612 /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */
5614 * Please note that, discard won't be sent to target device of device
5617 static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info,
5618 u64 logical, u64 *length_ret,
5619 struct btrfs_bio **bbio_ret)
5621 struct extent_map *em;
5622 struct map_lookup *map;
5623 struct btrfs_bio *bbio;
5624 u64 length = *length_ret;
5628 u64 stripe_end_offset;
5635 u32 sub_stripes = 0;
5636 u64 stripes_per_dev = 0;
5637 u32 remaining_stripes = 0;
5638 u32 last_stripe = 0;
5642 /* discard always return a bbio */
5645 em = btrfs_get_chunk_map(fs_info, logical, length);
5649 map = em->map_lookup;
5650 /* we don't discard raid56 yet */
5651 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5656 offset = logical - em->start;
5657 length = min_t(u64, em->start + em->len - logical, length);
5658 *length_ret = length;
5660 stripe_len = map->stripe_len;
5662 * stripe_nr counts the total number of stripes we have to stride
5663 * to get to this block
5665 stripe_nr = div64_u64(offset, stripe_len);
5667 /* stripe_offset is the offset of this block in its stripe */
5668 stripe_offset = offset - stripe_nr * stripe_len;
5670 stripe_nr_end = round_up(offset + length, map->stripe_len);
5671 stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len);
5672 stripe_cnt = stripe_nr_end - stripe_nr;
5673 stripe_end_offset = stripe_nr_end * map->stripe_len -
5676 * after this, stripe_nr is the number of stripes on this
5677 * device we have to walk to find the data, and stripe_index is
5678 * the number of our device in the stripe array
5682 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5683 BTRFS_BLOCK_GROUP_RAID10)) {
5684 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5687 sub_stripes = map->sub_stripes;
5689 factor = map->num_stripes / sub_stripes;
5690 num_stripes = min_t(u64, map->num_stripes,
5691 sub_stripes * stripe_cnt);
5692 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5693 stripe_index *= sub_stripes;
5694 stripes_per_dev = div_u64_rem(stripe_cnt, factor,
5695 &remaining_stripes);
5696 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5697 last_stripe *= sub_stripes;
5698 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1_MASK |
5699 BTRFS_BLOCK_GROUP_DUP)) {
5700 num_stripes = map->num_stripes;
5702 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5706 bbio = alloc_btrfs_bio(num_stripes, 0);
5712 for (i = 0; i < num_stripes; i++) {
5713 bbio->stripes[i].physical =
5714 map->stripes[stripe_index].physical +
5715 stripe_offset + stripe_nr * map->stripe_len;
5716 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5718 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5719 BTRFS_BLOCK_GROUP_RAID10)) {
5720 bbio->stripes[i].length = stripes_per_dev *
5723 if (i / sub_stripes < remaining_stripes)
5724 bbio->stripes[i].length +=
5728 * Special for the first stripe and
5731 * |-------|...|-------|
5735 if (i < sub_stripes)
5736 bbio->stripes[i].length -=
5739 if (stripe_index >= last_stripe &&
5740 stripe_index <= (last_stripe +
5742 bbio->stripes[i].length -=
5745 if (i == sub_stripes - 1)
5748 bbio->stripes[i].length = length;
5752 if (stripe_index == map->num_stripes) {
5759 bbio->map_type = map->type;
5760 bbio->num_stripes = num_stripes;
5762 free_extent_map(em);
5767 * In dev-replace case, for repair case (that's the only case where the mirror
5768 * is selected explicitly when calling btrfs_map_block), blocks left of the
5769 * left cursor can also be read from the target drive.
5771 * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the
5773 * For READ, it also needs to be supported using the same mirror number.
5775 * If the requested block is not left of the left cursor, EIO is returned. This
5776 * can happen because btrfs_num_copies() returns one more in the dev-replace
5779 static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info,
5780 u64 logical, u64 length,
5781 u64 srcdev_devid, int *mirror_num,
5784 struct btrfs_bio *bbio = NULL;
5786 int index_srcdev = 0;
5788 u64 physical_of_found = 0;
5792 ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS,
5793 logical, &length, &bbio, 0, 0);
5795 ASSERT(bbio == NULL);
5799 num_stripes = bbio->num_stripes;
5800 if (*mirror_num > num_stripes) {
5802 * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror,
5803 * that means that the requested area is not left of the left
5806 btrfs_put_bbio(bbio);
5811 * process the rest of the function using the mirror_num of the source
5812 * drive. Therefore look it up first. At the end, patch the device
5813 * pointer to the one of the target drive.
5815 for (i = 0; i < num_stripes; i++) {
5816 if (bbio->stripes[i].dev->devid != srcdev_devid)
5820 * In case of DUP, in order to keep it simple, only add the
5821 * mirror with the lowest physical address
5824 physical_of_found <= bbio->stripes[i].physical)
5829 physical_of_found = bbio->stripes[i].physical;
5832 btrfs_put_bbio(bbio);
5838 *mirror_num = index_srcdev + 1;
5839 *physical = physical_of_found;
5843 static void handle_ops_on_dev_replace(enum btrfs_map_op op,
5844 struct btrfs_bio **bbio_ret,
5845 struct btrfs_dev_replace *dev_replace,
5846 int *num_stripes_ret, int *max_errors_ret)
5848 struct btrfs_bio *bbio = *bbio_ret;
5849 u64 srcdev_devid = dev_replace->srcdev->devid;
5850 int tgtdev_indexes = 0;
5851 int num_stripes = *num_stripes_ret;
5852 int max_errors = *max_errors_ret;
5855 if (op == BTRFS_MAP_WRITE) {
5856 int index_where_to_add;
5859 * duplicate the write operations while the dev replace
5860 * procedure is running. Since the copying of the old disk to
5861 * the new disk takes place at run time while the filesystem is
5862 * mounted writable, the regular write operations to the old
5863 * disk have to be duplicated to go to the new disk as well.
5865 * Note that device->missing is handled by the caller, and that
5866 * the write to the old disk is already set up in the stripes
5869 index_where_to_add = num_stripes;
5870 for (i = 0; i < num_stripes; i++) {
5871 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5872 /* write to new disk, too */
5873 struct btrfs_bio_stripe *new =
5874 bbio->stripes + index_where_to_add;
5875 struct btrfs_bio_stripe *old =
5878 new->physical = old->physical;
5879 new->length = old->length;
5880 new->dev = dev_replace->tgtdev;
5881 bbio->tgtdev_map[i] = index_where_to_add;
5882 index_where_to_add++;
5887 num_stripes = index_where_to_add;
5888 } else if (op == BTRFS_MAP_GET_READ_MIRRORS) {
5889 int index_srcdev = 0;
5891 u64 physical_of_found = 0;
5894 * During the dev-replace procedure, the target drive can also
5895 * be used to read data in case it is needed to repair a corrupt
5896 * block elsewhere. This is possible if the requested area is
5897 * left of the left cursor. In this area, the target drive is a
5898 * full copy of the source drive.
5900 for (i = 0; i < num_stripes; i++) {
5901 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5903 * In case of DUP, in order to keep it simple,
5904 * only add the mirror with the lowest physical
5908 physical_of_found <=
5909 bbio->stripes[i].physical)
5913 physical_of_found = bbio->stripes[i].physical;
5917 struct btrfs_bio_stripe *tgtdev_stripe =
5918 bbio->stripes + num_stripes;
5920 tgtdev_stripe->physical = physical_of_found;
5921 tgtdev_stripe->length =
5922 bbio->stripes[index_srcdev].length;
5923 tgtdev_stripe->dev = dev_replace->tgtdev;
5924 bbio->tgtdev_map[index_srcdev] = num_stripes;
5931 *num_stripes_ret = num_stripes;
5932 *max_errors_ret = max_errors;
5933 bbio->num_tgtdevs = tgtdev_indexes;
5937 static bool need_full_stripe(enum btrfs_map_op op)
5939 return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS);
5943 * btrfs_get_io_geometry - calculates the geomery of a particular (address, len)
5944 * tuple. This information is used to calculate how big a
5945 * particular bio can get before it straddles a stripe.
5947 * @fs_info - the filesystem
5948 * @logical - address that we want to figure out the geometry of
5949 * @len - the length of IO we are going to perform, starting at @logical
5950 * @op - type of operation - write or read
5951 * @io_geom - pointer used to return values
5953 * Returns < 0 in case a chunk for the given logical address cannot be found,
5954 * usually shouldn't happen unless @logical is corrupted, 0 otherwise.
5956 int btrfs_get_io_geometry(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
5957 u64 logical, u64 len, struct btrfs_io_geometry *io_geom)
5959 struct extent_map *em;
5960 struct map_lookup *map;
5965 u64 raid56_full_stripe_start = (u64)-1;
5969 ASSERT(op != BTRFS_MAP_DISCARD);
5971 em = btrfs_get_chunk_map(fs_info, logical, len);
5975 map = em->map_lookup;
5976 /* Offset of this logical address in the chunk */
5977 offset = logical - em->start;
5978 /* Len of a stripe in a chunk */
5979 stripe_len = map->stripe_len;
5980 /* Stripe wher this block falls in */
5981 stripe_nr = div64_u64(offset, stripe_len);
5982 /* Offset of stripe in the chunk */
5983 stripe_offset = stripe_nr * stripe_len;
5984 if (offset < stripe_offset) {
5986 "stripe math has gone wrong, stripe_offset=%llu offset=%llu start=%llu logical=%llu stripe_len=%llu",
5987 stripe_offset, offset, em->start, logical, stripe_len);
5992 /* stripe_offset is the offset of this block in its stripe */
5993 stripe_offset = offset - stripe_offset;
5994 data_stripes = nr_data_stripes(map);
5996 if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5997 u64 max_len = stripe_len - stripe_offset;
6000 * In case of raid56, we need to know the stripe aligned start
6002 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6003 unsigned long full_stripe_len = stripe_len * data_stripes;
6004 raid56_full_stripe_start = offset;
6007 * Allow a write of a full stripe, but make sure we
6008 * don't allow straddling of stripes
6010 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
6012 raid56_full_stripe_start *= full_stripe_len;
6015 * For writes to RAID[56], allow a full stripeset across
6016 * all disks. For other RAID types and for RAID[56]
6017 * reads, just allow a single stripe (on a single disk).
6019 if (op == BTRFS_MAP_WRITE) {
6020 max_len = stripe_len * data_stripes -
6021 (offset - raid56_full_stripe_start);
6024 len = min_t(u64, em->len - offset, max_len);
6026 len = em->len - offset;
6030 io_geom->offset = offset;
6031 io_geom->stripe_len = stripe_len;
6032 io_geom->stripe_nr = stripe_nr;
6033 io_geom->stripe_offset = stripe_offset;
6034 io_geom->raid56_stripe_offset = raid56_full_stripe_start;
6038 free_extent_map(em);
6042 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
6043 enum btrfs_map_op op,
6044 u64 logical, u64 *length,
6045 struct btrfs_bio **bbio_ret,
6046 int mirror_num, int need_raid_map)
6048 struct extent_map *em;
6049 struct map_lookup *map;
6059 int tgtdev_indexes = 0;
6060 struct btrfs_bio *bbio = NULL;
6061 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
6062 int dev_replace_is_ongoing = 0;
6063 int num_alloc_stripes;
6064 int patch_the_first_stripe_for_dev_replace = 0;
6065 u64 physical_to_patch_in_first_stripe = 0;
6066 u64 raid56_full_stripe_start = (u64)-1;
6067 struct btrfs_io_geometry geom;
6070 ASSERT(op != BTRFS_MAP_DISCARD);
6072 ret = btrfs_get_io_geometry(fs_info, op, logical, *length, &geom);
6076 em = btrfs_get_chunk_map(fs_info, logical, *length);
6077 ASSERT(!IS_ERR(em));
6078 map = em->map_lookup;
6081 stripe_len = geom.stripe_len;
6082 stripe_nr = geom.stripe_nr;
6083 stripe_offset = geom.stripe_offset;
6084 raid56_full_stripe_start = geom.raid56_stripe_offset;
6085 data_stripes = nr_data_stripes(map);
6087 down_read(&dev_replace->rwsem);
6088 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
6090 * Hold the semaphore for read during the whole operation, write is
6091 * requested at commit time but must wait.
6093 if (!dev_replace_is_ongoing)
6094 up_read(&dev_replace->rwsem);
6096 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
6097 !need_full_stripe(op) && dev_replace->tgtdev != NULL) {
6098 ret = get_extra_mirror_from_replace(fs_info, logical, *length,
6099 dev_replace->srcdev->devid,
6101 &physical_to_patch_in_first_stripe);
6105 patch_the_first_stripe_for_dev_replace = 1;
6106 } else if (mirror_num > map->num_stripes) {
6112 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
6113 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
6115 if (!need_full_stripe(op))
6117 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1_MASK) {
6118 if (need_full_stripe(op))
6119 num_stripes = map->num_stripes;
6120 else if (mirror_num)
6121 stripe_index = mirror_num - 1;
6123 stripe_index = find_live_mirror(fs_info, map, 0,
6124 dev_replace_is_ongoing);
6125 mirror_num = stripe_index + 1;
6128 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
6129 if (need_full_stripe(op)) {
6130 num_stripes = map->num_stripes;
6131 } else if (mirror_num) {
6132 stripe_index = mirror_num - 1;
6137 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
6138 u32 factor = map->num_stripes / map->sub_stripes;
6140 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
6141 stripe_index *= map->sub_stripes;
6143 if (need_full_stripe(op))
6144 num_stripes = map->sub_stripes;
6145 else if (mirror_num)
6146 stripe_index += mirror_num - 1;
6148 int old_stripe_index = stripe_index;
6149 stripe_index = find_live_mirror(fs_info, map,
6151 dev_replace_is_ongoing);
6152 mirror_num = stripe_index - old_stripe_index + 1;
6155 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6156 if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) {
6157 /* push stripe_nr back to the start of the full stripe */
6158 stripe_nr = div64_u64(raid56_full_stripe_start,
6159 stripe_len * data_stripes);
6161 /* RAID[56] write or recovery. Return all stripes */
6162 num_stripes = map->num_stripes;
6163 max_errors = nr_parity_stripes(map);
6165 *length = map->stripe_len;
6170 * Mirror #0 or #1 means the original data block.
6171 * Mirror #2 is RAID5 parity block.
6172 * Mirror #3 is RAID6 Q block.
6174 stripe_nr = div_u64_rem(stripe_nr,
6175 data_stripes, &stripe_index);
6177 stripe_index = data_stripes + mirror_num - 2;
6179 /* We distribute the parity blocks across stripes */
6180 div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
6182 if (!need_full_stripe(op) && mirror_num <= 1)
6187 * after this, stripe_nr is the number of stripes on this
6188 * device we have to walk to find the data, and stripe_index is
6189 * the number of our device in the stripe array
6191 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
6193 mirror_num = stripe_index + 1;
6195 if (stripe_index >= map->num_stripes) {
6197 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
6198 stripe_index, map->num_stripes);
6203 num_alloc_stripes = num_stripes;
6204 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) {
6205 if (op == BTRFS_MAP_WRITE)
6206 num_alloc_stripes <<= 1;
6207 if (op == BTRFS_MAP_GET_READ_MIRRORS)
6208 num_alloc_stripes++;
6209 tgtdev_indexes = num_stripes;
6212 bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
6218 for (i = 0; i < num_stripes; i++) {
6219 bbio->stripes[i].physical = map->stripes[stripe_index].physical +
6220 stripe_offset + stripe_nr * map->stripe_len;
6221 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
6225 /* build raid_map */
6226 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map &&
6227 (need_full_stripe(op) || mirror_num > 1)) {
6231 /* Work out the disk rotation on this stripe-set */
6232 div_u64_rem(stripe_nr, num_stripes, &rot);
6234 /* Fill in the logical address of each stripe */
6235 tmp = stripe_nr * data_stripes;
6236 for (i = 0; i < data_stripes; i++)
6237 bbio->raid_map[(i+rot) % num_stripes] =
6238 em->start + (tmp + i) * map->stripe_len;
6240 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
6241 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
6242 bbio->raid_map[(i+rot+1) % num_stripes] =
6245 sort_parity_stripes(bbio, num_stripes);
6248 if (need_full_stripe(op))
6249 max_errors = btrfs_chunk_max_errors(map);
6251 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
6252 need_full_stripe(op)) {
6253 handle_ops_on_dev_replace(op, &bbio, dev_replace, &num_stripes,
6258 bbio->map_type = map->type;
6259 bbio->num_stripes = num_stripes;
6260 bbio->max_errors = max_errors;
6261 bbio->mirror_num = mirror_num;
6264 * this is the case that REQ_READ && dev_replace_is_ongoing &&
6265 * mirror_num == num_stripes + 1 && dev_replace target drive is
6266 * available as a mirror
6268 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
6269 WARN_ON(num_stripes > 1);
6270 bbio->stripes[0].dev = dev_replace->tgtdev;
6271 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
6272 bbio->mirror_num = map->num_stripes + 1;
6275 if (dev_replace_is_ongoing) {
6276 lockdep_assert_held(&dev_replace->rwsem);
6277 /* Unlock and let waiting writers proceed */
6278 up_read(&dev_replace->rwsem);
6280 free_extent_map(em);
6284 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6285 u64 logical, u64 *length,
6286 struct btrfs_bio **bbio_ret, int mirror_num)
6288 if (op == BTRFS_MAP_DISCARD)
6289 return __btrfs_map_block_for_discard(fs_info, logical,
6292 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret,
6296 /* For Scrub/replace */
6297 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6298 u64 logical, u64 *length,
6299 struct btrfs_bio **bbio_ret)
6301 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1);
6304 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
6306 bio->bi_private = bbio->private;
6307 bio->bi_end_io = bbio->end_io;
6310 btrfs_put_bbio(bbio);
6313 static void btrfs_end_bio(struct bio *bio)
6315 struct btrfs_bio *bbio = bio->bi_private;
6316 int is_orig_bio = 0;
6318 if (bio->bi_status) {
6319 atomic_inc(&bbio->error);
6320 if (bio->bi_status == BLK_STS_IOERR ||
6321 bio->bi_status == BLK_STS_TARGET) {
6322 struct btrfs_device *dev = btrfs_io_bio(bio)->device;
6325 if (bio_op(bio) == REQ_OP_WRITE)
6326 btrfs_dev_stat_inc_and_print(dev,
6327 BTRFS_DEV_STAT_WRITE_ERRS);
6328 else if (!(bio->bi_opf & REQ_RAHEAD))
6329 btrfs_dev_stat_inc_and_print(dev,
6330 BTRFS_DEV_STAT_READ_ERRS);
6331 if (bio->bi_opf & REQ_PREFLUSH)
6332 btrfs_dev_stat_inc_and_print(dev,
6333 BTRFS_DEV_STAT_FLUSH_ERRS);
6337 if (bio == bbio->orig_bio)
6340 btrfs_bio_counter_dec(bbio->fs_info);
6342 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6345 bio = bbio->orig_bio;
6348 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6349 /* only send an error to the higher layers if it is
6350 * beyond the tolerance of the btrfs bio
6352 if (atomic_read(&bbio->error) > bbio->max_errors) {
6353 bio->bi_status = BLK_STS_IOERR;
6356 * this bio is actually up to date, we didn't
6357 * go over the max number of errors
6359 bio->bi_status = BLK_STS_OK;
6362 btrfs_end_bbio(bbio, bio);
6363 } else if (!is_orig_bio) {
6368 static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio,
6369 u64 physical, struct btrfs_device *dev)
6371 struct btrfs_fs_info *fs_info = bbio->fs_info;
6373 bio->bi_private = bbio;
6374 btrfs_io_bio(bio)->device = dev;
6375 bio->bi_end_io = btrfs_end_bio;
6376 bio->bi_iter.bi_sector = physical >> 9;
6377 btrfs_debug_in_rcu(fs_info,
6378 "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
6379 bio_op(bio), bio->bi_opf, bio->bi_iter.bi_sector,
6380 (unsigned long)dev->bdev->bd_dev, rcu_str_deref(dev->name),
6381 dev->devid, bio->bi_iter.bi_size);
6382 bio_set_dev(bio, dev->bdev);
6384 btrfs_bio_counter_inc_noblocked(fs_info);
6386 btrfsic_submit_bio(bio);
6389 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6391 atomic_inc(&bbio->error);
6392 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6393 /* Should be the original bio. */
6394 WARN_ON(bio != bbio->orig_bio);
6396 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6397 bio->bi_iter.bi_sector = logical >> 9;
6398 if (atomic_read(&bbio->error) > bbio->max_errors)
6399 bio->bi_status = BLK_STS_IOERR;
6401 bio->bi_status = BLK_STS_OK;
6402 btrfs_end_bbio(bbio, bio);
6406 blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
6409 struct btrfs_device *dev;
6410 struct bio *first_bio = bio;
6411 u64 logical = bio->bi_iter.bi_sector << 9;
6417 struct btrfs_bio *bbio = NULL;
6419 length = bio->bi_iter.bi_size;
6420 map_length = length;
6422 btrfs_bio_counter_inc_blocked(fs_info);
6423 ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical,
6424 &map_length, &bbio, mirror_num, 1);
6426 btrfs_bio_counter_dec(fs_info);
6427 return errno_to_blk_status(ret);
6430 total_devs = bbio->num_stripes;
6431 bbio->orig_bio = first_bio;
6432 bbio->private = first_bio->bi_private;
6433 bbio->end_io = first_bio->bi_end_io;
6434 bbio->fs_info = fs_info;
6435 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6437 if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6438 ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) {
6439 /* In this case, map_length has been set to the length of
6440 a single stripe; not the whole write */
6441 if (bio_op(bio) == REQ_OP_WRITE) {
6442 ret = raid56_parity_write(fs_info, bio, bbio,
6445 ret = raid56_parity_recover(fs_info, bio, bbio,
6446 map_length, mirror_num, 1);
6449 btrfs_bio_counter_dec(fs_info);
6450 return errno_to_blk_status(ret);
6453 if (map_length < length) {
6455 "mapping failed logical %llu bio len %llu len %llu",
6456 logical, length, map_length);
6460 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6461 dev = bbio->stripes[dev_nr].dev;
6462 if (!dev || !dev->bdev || test_bit(BTRFS_DEV_STATE_MISSING,
6464 (bio_op(first_bio) == REQ_OP_WRITE &&
6465 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) {
6466 bbio_error(bbio, first_bio, logical);
6470 if (dev_nr < total_devs - 1)
6471 bio = btrfs_bio_clone(first_bio);
6475 submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical, dev);
6477 btrfs_bio_counter_dec(fs_info);
6482 * Find a device specified by @devid or @uuid in the list of @fs_devices, or
6485 * If devid and uuid are both specified, the match must be exact, otherwise
6486 * only devid is used.
6488 * If @seed is true, traverse through the seed devices.
6490 struct btrfs_device *btrfs_find_device(struct btrfs_fs_devices *fs_devices,
6491 u64 devid, u8 *uuid, u8 *fsid)
6493 struct btrfs_device *device;
6494 struct btrfs_fs_devices *seed_devs;
6496 if (!fsid || !memcmp(fs_devices->metadata_uuid, fsid, BTRFS_FSID_SIZE)) {
6497 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6498 if (device->devid == devid &&
6499 (!uuid || memcmp(device->uuid, uuid,
6500 BTRFS_UUID_SIZE) == 0))
6505 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
6507 !memcmp(seed_devs->metadata_uuid, fsid, BTRFS_FSID_SIZE)) {
6508 list_for_each_entry(device, &seed_devs->devices,
6510 if (device->devid == devid &&
6511 (!uuid || memcmp(device->uuid, uuid,
6512 BTRFS_UUID_SIZE) == 0))
6521 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices,
6522 u64 devid, u8 *dev_uuid)
6524 struct btrfs_device *device;
6525 unsigned int nofs_flag;
6528 * We call this under the chunk_mutex, so we want to use NOFS for this
6529 * allocation, however we don't want to change btrfs_alloc_device() to
6530 * always do NOFS because we use it in a lot of other GFP_KERNEL safe
6533 nofs_flag = memalloc_nofs_save();
6534 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6535 memalloc_nofs_restore(nofs_flag);
6539 list_add(&device->dev_list, &fs_devices->devices);
6540 device->fs_devices = fs_devices;
6541 fs_devices->num_devices++;
6543 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6544 fs_devices->missing_devices++;
6550 * btrfs_alloc_device - allocate struct btrfs_device
6551 * @fs_info: used only for generating a new devid, can be NULL if
6552 * devid is provided (i.e. @devid != NULL).
6553 * @devid: a pointer to devid for this device. If NULL a new devid
6555 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6558 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6559 * on error. Returned struct is not linked onto any lists and must be
6560 * destroyed with btrfs_free_device.
6562 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6566 struct btrfs_device *dev;
6569 if (WARN_ON(!devid && !fs_info))
6570 return ERR_PTR(-EINVAL);
6572 dev = __alloc_device(fs_info);
6581 ret = find_next_devid(fs_info, &tmp);
6583 btrfs_free_device(dev);
6584 return ERR_PTR(ret);
6590 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6592 generate_random_uuid(dev->uuid);
6597 static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info,
6598 u64 devid, u8 *uuid, bool error)
6601 btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing",
6604 btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing",
6608 static u64 calc_stripe_length(u64 type, u64 chunk_len, int num_stripes)
6610 int index = btrfs_bg_flags_to_raid_index(type);
6611 int ncopies = btrfs_raid_array[index].ncopies;
6612 const int nparity = btrfs_raid_array[index].nparity;
6616 data_stripes = num_stripes - nparity;
6618 data_stripes = num_stripes / ncopies;
6620 return div_u64(chunk_len, data_stripes);
6623 static int read_one_chunk(struct btrfs_key *key, struct extent_buffer *leaf,
6624 struct btrfs_chunk *chunk)
6626 struct btrfs_fs_info *fs_info = leaf->fs_info;
6627 struct extent_map_tree *map_tree = &fs_info->mapping_tree;
6628 struct map_lookup *map;
6629 struct extent_map *em;
6633 u8 uuid[BTRFS_UUID_SIZE];
6638 logical = key->offset;
6639 length = btrfs_chunk_length(leaf, chunk);
6640 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6643 * Only need to verify chunk item if we're reading from sys chunk array,
6644 * as chunk item in tree block is already verified by tree-checker.
6646 if (leaf->start == BTRFS_SUPER_INFO_OFFSET) {
6647 ret = btrfs_check_chunk_valid(leaf, chunk, logical);
6652 read_lock(&map_tree->lock);
6653 em = lookup_extent_mapping(map_tree, logical, 1);
6654 read_unlock(&map_tree->lock);
6656 /* already mapped? */
6657 if (em && em->start <= logical && em->start + em->len > logical) {
6658 free_extent_map(em);
6661 free_extent_map(em);
6664 em = alloc_extent_map();
6667 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6669 free_extent_map(em);
6673 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6674 em->map_lookup = map;
6675 em->start = logical;
6678 em->block_start = 0;
6679 em->block_len = em->len;
6681 map->num_stripes = num_stripes;
6682 map->io_width = btrfs_chunk_io_width(leaf, chunk);
6683 map->io_align = btrfs_chunk_io_align(leaf, chunk);
6684 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6685 map->type = btrfs_chunk_type(leaf, chunk);
6686 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6687 map->verified_stripes = 0;
6688 em->orig_block_len = calc_stripe_length(map->type, em->len,
6690 for (i = 0; i < num_stripes; i++) {
6691 map->stripes[i].physical =
6692 btrfs_stripe_offset_nr(leaf, chunk, i);
6693 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6694 read_extent_buffer(leaf, uuid, (unsigned long)
6695 btrfs_stripe_dev_uuid_nr(chunk, i),
6697 map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices,
6699 if (!map->stripes[i].dev &&
6700 !btrfs_test_opt(fs_info, DEGRADED)) {
6701 free_extent_map(em);
6702 btrfs_report_missing_device(fs_info, devid, uuid, true);
6705 if (!map->stripes[i].dev) {
6706 map->stripes[i].dev =
6707 add_missing_dev(fs_info->fs_devices, devid,
6709 if (IS_ERR(map->stripes[i].dev)) {
6710 free_extent_map(em);
6712 "failed to init missing dev %llu: %ld",
6713 devid, PTR_ERR(map->stripes[i].dev));
6714 return PTR_ERR(map->stripes[i].dev);
6716 btrfs_report_missing_device(fs_info, devid, uuid, false);
6718 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
6719 &(map->stripes[i].dev->dev_state));
6723 write_lock(&map_tree->lock);
6724 ret = add_extent_mapping(map_tree, em, 0);
6725 write_unlock(&map_tree->lock);
6728 "failed to add chunk map, start=%llu len=%llu: %d",
6729 em->start, em->len, ret);
6731 free_extent_map(em);
6736 static void fill_device_from_item(struct extent_buffer *leaf,
6737 struct btrfs_dev_item *dev_item,
6738 struct btrfs_device *device)
6742 device->devid = btrfs_device_id(leaf, dev_item);
6743 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6744 device->total_bytes = device->disk_total_bytes;
6745 device->commit_total_bytes = device->disk_total_bytes;
6746 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6747 device->commit_bytes_used = device->bytes_used;
6748 device->type = btrfs_device_type(leaf, dev_item);
6749 device->io_align = btrfs_device_io_align(leaf, dev_item);
6750 device->io_width = btrfs_device_io_width(leaf, dev_item);
6751 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6752 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6753 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
6755 ptr = btrfs_device_uuid(dev_item);
6756 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6759 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info,
6762 struct btrfs_fs_devices *fs_devices;
6765 lockdep_assert_held(&uuid_mutex);
6768 /* This will match only for multi-device seed fs */
6769 list_for_each_entry(fs_devices, &fs_info->fs_devices->seed_list, seed_list)
6770 if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE))
6774 fs_devices = find_fsid(fsid, NULL);
6776 if (!btrfs_test_opt(fs_info, DEGRADED))
6777 return ERR_PTR(-ENOENT);
6779 fs_devices = alloc_fs_devices(fsid, NULL);
6780 if (IS_ERR(fs_devices))
6783 fs_devices->seeding = true;
6784 fs_devices->opened = 1;
6789 * Upon first call for a seed fs fsid, just create a private copy of the
6790 * respective fs_devices and anchor it at fs_info->fs_devices->seed_list
6792 fs_devices = clone_fs_devices(fs_devices);
6793 if (IS_ERR(fs_devices))
6796 ret = open_fs_devices(fs_devices, FMODE_READ, fs_info->bdev_holder);
6798 free_fs_devices(fs_devices);
6799 return ERR_PTR(ret);
6802 if (!fs_devices->seeding) {
6803 close_fs_devices(fs_devices);
6804 free_fs_devices(fs_devices);
6805 return ERR_PTR(-EINVAL);
6808 list_add(&fs_devices->seed_list, &fs_info->fs_devices->seed_list);
6813 static int read_one_dev(struct extent_buffer *leaf,
6814 struct btrfs_dev_item *dev_item)
6816 struct btrfs_fs_info *fs_info = leaf->fs_info;
6817 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6818 struct btrfs_device *device;
6821 u8 fs_uuid[BTRFS_FSID_SIZE];
6822 u8 dev_uuid[BTRFS_UUID_SIZE];
6824 devid = btrfs_device_id(leaf, dev_item);
6825 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6827 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6830 if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) {
6831 fs_devices = open_seed_devices(fs_info, fs_uuid);
6832 if (IS_ERR(fs_devices))
6833 return PTR_ERR(fs_devices);
6836 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
6839 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6840 btrfs_report_missing_device(fs_info, devid,
6845 device = add_missing_dev(fs_devices, devid, dev_uuid);
6846 if (IS_ERR(device)) {
6848 "failed to add missing dev %llu: %ld",
6849 devid, PTR_ERR(device));
6850 return PTR_ERR(device);
6852 btrfs_report_missing_device(fs_info, devid, dev_uuid, false);
6854 if (!device->bdev) {
6855 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6856 btrfs_report_missing_device(fs_info,
6857 devid, dev_uuid, true);
6860 btrfs_report_missing_device(fs_info, devid,
6864 if (!device->bdev &&
6865 !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
6867 * this happens when a device that was properly setup
6868 * in the device info lists suddenly goes bad.
6869 * device->bdev is NULL, and so we have to set
6870 * device->missing to one here
6872 device->fs_devices->missing_devices++;
6873 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6876 /* Move the device to its own fs_devices */
6877 if (device->fs_devices != fs_devices) {
6878 ASSERT(test_bit(BTRFS_DEV_STATE_MISSING,
6879 &device->dev_state));
6881 list_move(&device->dev_list, &fs_devices->devices);
6882 device->fs_devices->num_devices--;
6883 fs_devices->num_devices++;
6885 device->fs_devices->missing_devices--;
6886 fs_devices->missing_devices++;
6888 device->fs_devices = fs_devices;
6892 if (device->fs_devices != fs_info->fs_devices) {
6893 BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state));
6894 if (device->generation !=
6895 btrfs_device_generation(leaf, dev_item))
6899 fill_device_from_item(leaf, dev_item, device);
6901 u64 max_total_bytes = i_size_read(device->bdev->bd_inode);
6903 if (device->total_bytes > max_total_bytes) {
6905 "device total_bytes should be at most %llu but found %llu",
6906 max_total_bytes, device->total_bytes);
6910 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
6911 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
6912 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
6913 device->fs_devices->total_rw_bytes += device->total_bytes;
6914 atomic64_add(device->total_bytes - device->bytes_used,
6915 &fs_info->free_chunk_space);
6921 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
6923 struct btrfs_root *root = fs_info->tree_root;
6924 struct btrfs_super_block *super_copy = fs_info->super_copy;
6925 struct extent_buffer *sb;
6926 struct btrfs_disk_key *disk_key;
6927 struct btrfs_chunk *chunk;
6929 unsigned long sb_array_offset;
6936 struct btrfs_key key;
6938 ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize);
6940 * This will create extent buffer of nodesize, superblock size is
6941 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6942 * overallocate but we can keep it as-is, only the first page is used.
6944 sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET,
6945 root->root_key.objectid, 0);
6948 set_extent_buffer_uptodate(sb);
6950 * The sb extent buffer is artificial and just used to read the system array.
6951 * set_extent_buffer_uptodate() call does not properly mark all it's
6952 * pages up-to-date when the page is larger: extent does not cover the
6953 * whole page and consequently check_page_uptodate does not find all
6954 * the page's extents up-to-date (the hole beyond sb),
6955 * write_extent_buffer then triggers a WARN_ON.
6957 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6958 * but sb spans only this function. Add an explicit SetPageUptodate call
6959 * to silence the warning eg. on PowerPC 64.
6961 if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
6962 SetPageUptodate(sb->pages[0]);
6964 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6965 array_size = btrfs_super_sys_array_size(super_copy);
6967 array_ptr = super_copy->sys_chunk_array;
6968 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6971 while (cur_offset < array_size) {
6972 disk_key = (struct btrfs_disk_key *)array_ptr;
6973 len = sizeof(*disk_key);
6974 if (cur_offset + len > array_size)
6975 goto out_short_read;
6977 btrfs_disk_key_to_cpu(&key, disk_key);
6980 sb_array_offset += len;
6983 if (key.type != BTRFS_CHUNK_ITEM_KEY) {
6985 "unexpected item type %u in sys_array at offset %u",
6986 (u32)key.type, cur_offset);
6991 chunk = (struct btrfs_chunk *)sb_array_offset;
6993 * At least one btrfs_chunk with one stripe must be present,
6994 * exact stripe count check comes afterwards
6996 len = btrfs_chunk_item_size(1);
6997 if (cur_offset + len > array_size)
6998 goto out_short_read;
7000 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
7003 "invalid number of stripes %u in sys_array at offset %u",
7004 num_stripes, cur_offset);
7009 type = btrfs_chunk_type(sb, chunk);
7010 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
7012 "invalid chunk type %llu in sys_array at offset %u",
7018 len = btrfs_chunk_item_size(num_stripes);
7019 if (cur_offset + len > array_size)
7020 goto out_short_read;
7022 ret = read_one_chunk(&key, sb, chunk);
7027 sb_array_offset += len;
7030 clear_extent_buffer_uptodate(sb);
7031 free_extent_buffer_stale(sb);
7035 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
7037 clear_extent_buffer_uptodate(sb);
7038 free_extent_buffer_stale(sb);
7043 * Check if all chunks in the fs are OK for read-write degraded mount
7045 * If the @failing_dev is specified, it's accounted as missing.
7047 * Return true if all chunks meet the minimal RW mount requirements.
7048 * Return false if any chunk doesn't meet the minimal RW mount requirements.
7050 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
7051 struct btrfs_device *failing_dev)
7053 struct extent_map_tree *map_tree = &fs_info->mapping_tree;
7054 struct extent_map *em;
7058 read_lock(&map_tree->lock);
7059 em = lookup_extent_mapping(map_tree, 0, (u64)-1);
7060 read_unlock(&map_tree->lock);
7061 /* No chunk at all? Return false anyway */
7067 struct map_lookup *map;
7072 map = em->map_lookup;
7074 btrfs_get_num_tolerated_disk_barrier_failures(
7076 for (i = 0; i < map->num_stripes; i++) {
7077 struct btrfs_device *dev = map->stripes[i].dev;
7079 if (!dev || !dev->bdev ||
7080 test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) ||
7081 dev->last_flush_error)
7083 else if (failing_dev && failing_dev == dev)
7086 if (missing > max_tolerated) {
7089 "chunk %llu missing %d devices, max tolerance is %d for writable mount",
7090 em->start, missing, max_tolerated);
7091 free_extent_map(em);
7095 next_start = extent_map_end(em);
7096 free_extent_map(em);
7098 read_lock(&map_tree->lock);
7099 em = lookup_extent_mapping(map_tree, next_start,
7100 (u64)(-1) - next_start);
7101 read_unlock(&map_tree->lock);
7107 static void readahead_tree_node_children(struct extent_buffer *node)
7110 const int nr_items = btrfs_header_nritems(node);
7112 for (i = 0; i < nr_items; i++)
7113 btrfs_readahead_node_child(node, i);
7116 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
7118 struct btrfs_root *root = fs_info->chunk_root;
7119 struct btrfs_path *path;
7120 struct extent_buffer *leaf;
7121 struct btrfs_key key;
7122 struct btrfs_key found_key;
7126 u64 last_ra_node = 0;
7128 path = btrfs_alloc_path();
7133 * uuid_mutex is needed only if we are mounting a sprout FS
7134 * otherwise we don't need it.
7136 mutex_lock(&uuid_mutex);
7139 * It is possible for mount and umount to race in such a way that
7140 * we execute this code path, but open_fs_devices failed to clear
7141 * total_rw_bytes. We certainly want it cleared before reading the
7142 * device items, so clear it here.
7144 fs_info->fs_devices->total_rw_bytes = 0;
7147 * Read all device items, and then all the chunk items. All
7148 * device items are found before any chunk item (their object id
7149 * is smaller than the lowest possible object id for a chunk
7150 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
7152 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
7155 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7159 struct extent_buffer *node;
7161 leaf = path->nodes[0];
7162 slot = path->slots[0];
7163 if (slot >= btrfs_header_nritems(leaf)) {
7164 ret = btrfs_next_leaf(root, path);
7172 * The nodes on level 1 are not locked but we don't need to do
7173 * that during mount time as nothing else can access the tree
7175 node = path->nodes[1];
7177 if (last_ra_node != node->start) {
7178 readahead_tree_node_children(node);
7179 last_ra_node = node->start;
7182 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7183 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
7184 struct btrfs_dev_item *dev_item;
7185 dev_item = btrfs_item_ptr(leaf, slot,
7186 struct btrfs_dev_item);
7187 ret = read_one_dev(leaf, dev_item);
7191 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
7192 struct btrfs_chunk *chunk;
7193 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
7194 mutex_lock(&fs_info->chunk_mutex);
7195 ret = read_one_chunk(&found_key, leaf, chunk);
7196 mutex_unlock(&fs_info->chunk_mutex);
7204 * After loading chunk tree, we've got all device information,
7205 * do another round of validation checks.
7207 if (total_dev != fs_info->fs_devices->total_devices) {
7209 "super_num_devices %llu mismatch with num_devices %llu found here",
7210 btrfs_super_num_devices(fs_info->super_copy),
7215 if (btrfs_super_total_bytes(fs_info->super_copy) <
7216 fs_info->fs_devices->total_rw_bytes) {
7218 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
7219 btrfs_super_total_bytes(fs_info->super_copy),
7220 fs_info->fs_devices->total_rw_bytes);
7226 mutex_unlock(&uuid_mutex);
7228 btrfs_free_path(path);
7232 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
7234 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
7235 struct btrfs_device *device;
7237 fs_devices->fs_info = fs_info;
7239 mutex_lock(&fs_devices->device_list_mutex);
7240 list_for_each_entry(device, &fs_devices->devices, dev_list)
7241 device->fs_info = fs_info;
7243 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
7244 list_for_each_entry(device, &seed_devs->devices, dev_list)
7245 device->fs_info = fs_info;
7247 seed_devs->fs_info = fs_info;
7249 mutex_unlock(&fs_devices->device_list_mutex);
7252 static u64 btrfs_dev_stats_value(const struct extent_buffer *eb,
7253 const struct btrfs_dev_stats_item *ptr,
7258 read_extent_buffer(eb, &val,
7259 offsetof(struct btrfs_dev_stats_item, values) +
7260 ((unsigned long)ptr) + (index * sizeof(u64)),
7265 static void btrfs_set_dev_stats_value(struct extent_buffer *eb,
7266 struct btrfs_dev_stats_item *ptr,
7269 write_extent_buffer(eb, &val,
7270 offsetof(struct btrfs_dev_stats_item, values) +
7271 ((unsigned long)ptr) + (index * sizeof(u64)),
7275 static int btrfs_device_init_dev_stats(struct btrfs_device *device,
7276 struct btrfs_path *path)
7278 struct btrfs_dev_stats_item *ptr;
7279 struct extent_buffer *eb;
7280 struct btrfs_key key;
7284 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7285 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7286 key.offset = device->devid;
7287 ret = btrfs_search_slot(NULL, device->fs_info->dev_root, &key, path, 0, 0);
7289 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7290 btrfs_dev_stat_set(device, i, 0);
7291 device->dev_stats_valid = 1;
7292 btrfs_release_path(path);
7293 return ret < 0 ? ret : 0;
7295 slot = path->slots[0];
7296 eb = path->nodes[0];
7297 item_size = btrfs_item_size_nr(eb, slot);
7299 ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_stats_item);
7301 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7302 if (item_size >= (1 + i) * sizeof(__le64))
7303 btrfs_dev_stat_set(device, i,
7304 btrfs_dev_stats_value(eb, ptr, i));
7306 btrfs_dev_stat_set(device, i, 0);
7309 device->dev_stats_valid = 1;
7310 btrfs_dev_stat_print_on_load(device);
7311 btrfs_release_path(path);
7316 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
7318 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
7319 struct btrfs_device *device;
7320 struct btrfs_path *path = NULL;
7323 path = btrfs_alloc_path();
7327 mutex_lock(&fs_devices->device_list_mutex);
7328 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7329 ret = btrfs_device_init_dev_stats(device, path);
7333 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
7334 list_for_each_entry(device, &seed_devs->devices, dev_list) {
7335 ret = btrfs_device_init_dev_stats(device, path);
7341 mutex_unlock(&fs_devices->device_list_mutex);
7343 btrfs_free_path(path);
7347 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
7348 struct btrfs_device *device)
7350 struct btrfs_fs_info *fs_info = trans->fs_info;
7351 struct btrfs_root *dev_root = fs_info->dev_root;
7352 struct btrfs_path *path;
7353 struct btrfs_key key;
7354 struct extent_buffer *eb;
7355 struct btrfs_dev_stats_item *ptr;
7359 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7360 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7361 key.offset = device->devid;
7363 path = btrfs_alloc_path();
7366 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
7368 btrfs_warn_in_rcu(fs_info,
7369 "error %d while searching for dev_stats item for device %s",
7370 ret, rcu_str_deref(device->name));
7375 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
7376 /* need to delete old one and insert a new one */
7377 ret = btrfs_del_item(trans, dev_root, path);
7379 btrfs_warn_in_rcu(fs_info,
7380 "delete too small dev_stats item for device %s failed %d",
7381 rcu_str_deref(device->name), ret);
7388 /* need to insert a new item */
7389 btrfs_release_path(path);
7390 ret = btrfs_insert_empty_item(trans, dev_root, path,
7391 &key, sizeof(*ptr));
7393 btrfs_warn_in_rcu(fs_info,
7394 "insert dev_stats item for device %s failed %d",
7395 rcu_str_deref(device->name), ret);
7400 eb = path->nodes[0];
7401 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7402 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7403 btrfs_set_dev_stats_value(eb, ptr, i,
7404 btrfs_dev_stat_read(device, i));
7405 btrfs_mark_buffer_dirty(eb);
7408 btrfs_free_path(path);
7413 * called from commit_transaction. Writes all changed device stats to disk.
7415 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans)
7417 struct btrfs_fs_info *fs_info = trans->fs_info;
7418 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7419 struct btrfs_device *device;
7423 mutex_lock(&fs_devices->device_list_mutex);
7424 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7425 stats_cnt = atomic_read(&device->dev_stats_ccnt);
7426 if (!device->dev_stats_valid || stats_cnt == 0)
7431 * There is a LOAD-LOAD control dependency between the value of
7432 * dev_stats_ccnt and updating the on-disk values which requires
7433 * reading the in-memory counters. Such control dependencies
7434 * require explicit read memory barriers.
7436 * This memory barriers pairs with smp_mb__before_atomic in
7437 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full
7438 * barrier implied by atomic_xchg in
7439 * btrfs_dev_stats_read_and_reset
7443 ret = update_dev_stat_item(trans, device);
7445 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7447 mutex_unlock(&fs_devices->device_list_mutex);
7452 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7454 btrfs_dev_stat_inc(dev, index);
7455 btrfs_dev_stat_print_on_error(dev);
7458 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
7460 if (!dev->dev_stats_valid)
7462 btrfs_err_rl_in_rcu(dev->fs_info,
7463 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7464 rcu_str_deref(dev->name),
7465 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7466 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7467 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7468 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7469 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7472 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7476 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7477 if (btrfs_dev_stat_read(dev, i) != 0)
7479 if (i == BTRFS_DEV_STAT_VALUES_MAX)
7480 return; /* all values == 0, suppress message */
7482 btrfs_info_in_rcu(dev->fs_info,
7483 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7484 rcu_str_deref(dev->name),
7485 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7486 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7487 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7488 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7489 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7492 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
7493 struct btrfs_ioctl_get_dev_stats *stats)
7495 struct btrfs_device *dev;
7496 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7499 mutex_lock(&fs_devices->device_list_mutex);
7500 dev = btrfs_find_device(fs_info->fs_devices, stats->devid, NULL, NULL);
7501 mutex_unlock(&fs_devices->device_list_mutex);
7504 btrfs_warn(fs_info, "get dev_stats failed, device not found");
7506 } else if (!dev->dev_stats_valid) {
7507 btrfs_warn(fs_info, "get dev_stats failed, not yet valid");
7509 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7510 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7511 if (stats->nr_items > i)
7513 btrfs_dev_stat_read_and_reset(dev, i);
7515 btrfs_dev_stat_set(dev, i, 0);
7517 btrfs_info(fs_info, "device stats zeroed by %s (%d)",
7518 current->comm, task_pid_nr(current));
7520 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7521 if (stats->nr_items > i)
7522 stats->values[i] = btrfs_dev_stat_read(dev, i);
7524 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7525 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7530 * Update the size and bytes used for each device where it changed. This is
7531 * delayed since we would otherwise get errors while writing out the
7534 * Must be invoked during transaction commit.
7536 void btrfs_commit_device_sizes(struct btrfs_transaction *trans)
7538 struct btrfs_device *curr, *next;
7540 ASSERT(trans->state == TRANS_STATE_COMMIT_DOING);
7542 if (list_empty(&trans->dev_update_list))
7546 * We don't need the device_list_mutex here. This list is owned by the
7547 * transaction and the transaction must complete before the device is
7550 mutex_lock(&trans->fs_info->chunk_mutex);
7551 list_for_each_entry_safe(curr, next, &trans->dev_update_list,
7553 list_del_init(&curr->post_commit_list);
7554 curr->commit_total_bytes = curr->disk_total_bytes;
7555 curr->commit_bytes_used = curr->bytes_used;
7557 mutex_unlock(&trans->fs_info->chunk_mutex);
7561 * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10.
7563 int btrfs_bg_type_to_factor(u64 flags)
7565 const int index = btrfs_bg_flags_to_raid_index(flags);
7567 return btrfs_raid_array[index].ncopies;
7572 static int verify_one_dev_extent(struct btrfs_fs_info *fs_info,
7573 u64 chunk_offset, u64 devid,
7574 u64 physical_offset, u64 physical_len)
7576 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
7577 struct extent_map *em;
7578 struct map_lookup *map;
7579 struct btrfs_device *dev;
7585 read_lock(&em_tree->lock);
7586 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
7587 read_unlock(&em_tree->lock);
7591 "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk",
7592 physical_offset, devid);
7597 map = em->map_lookup;
7598 stripe_len = calc_stripe_length(map->type, em->len, map->num_stripes);
7599 if (physical_len != stripe_len) {
7601 "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu",
7602 physical_offset, devid, em->start, physical_len,
7608 for (i = 0; i < map->num_stripes; i++) {
7609 if (map->stripes[i].dev->devid == devid &&
7610 map->stripes[i].physical == physical_offset) {
7612 if (map->verified_stripes >= map->num_stripes) {
7614 "too many dev extents for chunk %llu found",
7619 map->verified_stripes++;
7625 "dev extent physical offset %llu devid %llu has no corresponding chunk",
7626 physical_offset, devid);
7630 /* Make sure no dev extent is beyond device bondary */
7631 dev = btrfs_find_device(fs_info->fs_devices, devid, NULL, NULL);
7633 btrfs_err(fs_info, "failed to find devid %llu", devid);
7638 if (physical_offset + physical_len > dev->disk_total_bytes) {
7640 "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu",
7641 devid, physical_offset, physical_len,
7642 dev->disk_total_bytes);
7647 free_extent_map(em);
7651 static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info)
7653 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
7654 struct extent_map *em;
7655 struct rb_node *node;
7658 read_lock(&em_tree->lock);
7659 for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) {
7660 em = rb_entry(node, struct extent_map, rb_node);
7661 if (em->map_lookup->num_stripes !=
7662 em->map_lookup->verified_stripes) {
7664 "chunk %llu has missing dev extent, have %d expect %d",
7665 em->start, em->map_lookup->verified_stripes,
7666 em->map_lookup->num_stripes);
7672 read_unlock(&em_tree->lock);
7677 * Ensure that all dev extents are mapped to correct chunk, otherwise
7678 * later chunk allocation/free would cause unexpected behavior.
7680 * NOTE: This will iterate through the whole device tree, which should be of
7681 * the same size level as the chunk tree. This slightly increases mount time.
7683 int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info)
7685 struct btrfs_path *path;
7686 struct btrfs_root *root = fs_info->dev_root;
7687 struct btrfs_key key;
7689 u64 prev_dev_ext_end = 0;
7693 * We don't have a dev_root because we mounted with ignorebadroots and
7694 * failed to load the root, so we want to skip the verification in this
7697 * However if the dev root is fine, but the tree itself is corrupted
7698 * we'd still fail to mount. This verification is only to make sure
7699 * writes can happen safely, so instead just bypass this check
7700 * completely in the case of IGNOREBADROOTS.
7702 if (btrfs_test_opt(fs_info, IGNOREBADROOTS))
7706 key.type = BTRFS_DEV_EXTENT_KEY;
7709 path = btrfs_alloc_path();
7713 path->reada = READA_FORWARD;
7714 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7718 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
7719 ret = btrfs_next_item(root, path);
7722 /* No dev extents at all? Not good */
7729 struct extent_buffer *leaf = path->nodes[0];
7730 struct btrfs_dev_extent *dext;
7731 int slot = path->slots[0];
7733 u64 physical_offset;
7737 btrfs_item_key_to_cpu(leaf, &key, slot);
7738 if (key.type != BTRFS_DEV_EXTENT_KEY)
7740 devid = key.objectid;
7741 physical_offset = key.offset;
7743 dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent);
7744 chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext);
7745 physical_len = btrfs_dev_extent_length(leaf, dext);
7747 /* Check if this dev extent overlaps with the previous one */
7748 if (devid == prev_devid && physical_offset < prev_dev_ext_end) {
7750 "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu",
7751 devid, physical_offset, prev_dev_ext_end);
7756 ret = verify_one_dev_extent(fs_info, chunk_offset, devid,
7757 physical_offset, physical_len);
7761 prev_dev_ext_end = physical_offset + physical_len;
7763 ret = btrfs_next_item(root, path);
7772 /* Ensure all chunks have corresponding dev extents */
7773 ret = verify_chunk_dev_extent_mapping(fs_info);
7775 btrfs_free_path(path);
7780 * Check whether the given block group or device is pinned by any inode being
7781 * used as a swapfile.
7783 bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr)
7785 struct btrfs_swapfile_pin *sp;
7786 struct rb_node *node;
7788 spin_lock(&fs_info->swapfile_pins_lock);
7789 node = fs_info->swapfile_pins.rb_node;
7791 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
7793 node = node->rb_left;
7794 else if (ptr > sp->ptr)
7795 node = node->rb_right;
7799 spin_unlock(&fs_info->swapfile_pins_lock);
7800 return node != NULL;