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);
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
941 error = lookup_bdev(path, &path_dev);
943 mutex_unlock(&fs_devices->device_list_mutex);
944 return ERR_PTR(error);
947 if (device->bdev->bd_dev != path_dev) {
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);
962 btrfs_info_in_rcu(device->fs_info,
963 "devid %llu device path %s changed to %s scanned by %s (%d)",
964 devid, rcu_str_deref(device->name),
966 task_pid_nr(current));
969 name = rcu_string_strdup(path, GFP_NOFS);
971 mutex_unlock(&fs_devices->device_list_mutex);
972 return ERR_PTR(-ENOMEM);
974 rcu_string_free(device->name);
975 rcu_assign_pointer(device->name, name);
976 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
977 fs_devices->missing_devices--;
978 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
983 * Unmount does not free the btrfs_device struct but would zero
984 * generation along with most of the other members. So just update
985 * it back. We need it to pick the disk with largest generation
988 if (!fs_devices->opened) {
989 device->generation = found_transid;
990 fs_devices->latest_generation = max_t(u64, found_transid,
991 fs_devices->latest_generation);
994 fs_devices->total_devices = btrfs_super_num_devices(disk_super);
996 mutex_unlock(&fs_devices->device_list_mutex);
1000 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
1002 struct btrfs_fs_devices *fs_devices;
1003 struct btrfs_device *device;
1004 struct btrfs_device *orig_dev;
1007 fs_devices = alloc_fs_devices(orig->fsid, NULL);
1008 if (IS_ERR(fs_devices))
1011 mutex_lock(&orig->device_list_mutex);
1012 fs_devices->total_devices = orig->total_devices;
1014 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
1015 struct rcu_string *name;
1017 device = btrfs_alloc_device(NULL, &orig_dev->devid,
1019 if (IS_ERR(device)) {
1020 ret = PTR_ERR(device);
1025 * This is ok to do without rcu read locked because we hold the
1026 * uuid mutex so nothing we touch in here is going to disappear.
1028 if (orig_dev->name) {
1029 name = rcu_string_strdup(orig_dev->name->str,
1032 btrfs_free_device(device);
1036 rcu_assign_pointer(device->name, name);
1039 list_add(&device->dev_list, &fs_devices->devices);
1040 device->fs_devices = fs_devices;
1041 fs_devices->num_devices++;
1043 mutex_unlock(&orig->device_list_mutex);
1046 mutex_unlock(&orig->device_list_mutex);
1047 free_fs_devices(fs_devices);
1048 return ERR_PTR(ret);
1051 static void __btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices,
1052 struct btrfs_device **latest_dev)
1054 struct btrfs_device *device, *next;
1056 /* This is the initialized path, it is safe to release the devices. */
1057 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
1058 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)) {
1059 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
1060 &device->dev_state) &&
1061 !test_bit(BTRFS_DEV_STATE_MISSING,
1062 &device->dev_state) &&
1064 device->generation > (*latest_dev)->generation)) {
1065 *latest_dev = device;
1071 * We have already validated the presence of BTRFS_DEV_REPLACE_DEVID,
1072 * in btrfs_init_dev_replace() so just continue.
1074 if (device->devid == BTRFS_DEV_REPLACE_DEVID)
1078 blkdev_put(device->bdev, device->mode);
1079 device->bdev = NULL;
1080 fs_devices->open_devices--;
1082 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1083 list_del_init(&device->dev_alloc_list);
1084 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
1086 list_del_init(&device->dev_list);
1087 fs_devices->num_devices--;
1088 btrfs_free_device(device);
1094 * After we have read the system tree and know devids belonging to this
1095 * filesystem, remove the device which does not belong there.
1097 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices)
1099 struct btrfs_device *latest_dev = NULL;
1100 struct btrfs_fs_devices *seed_dev;
1102 mutex_lock(&uuid_mutex);
1103 __btrfs_free_extra_devids(fs_devices, &latest_dev);
1105 list_for_each_entry(seed_dev, &fs_devices->seed_list, seed_list)
1106 __btrfs_free_extra_devids(seed_dev, &latest_dev);
1108 fs_devices->latest_bdev = latest_dev->bdev;
1110 mutex_unlock(&uuid_mutex);
1113 static void btrfs_close_bdev(struct btrfs_device *device)
1118 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1119 sync_blockdev(device->bdev);
1120 invalidate_bdev(device->bdev);
1123 blkdev_put(device->bdev, device->mode);
1126 static void btrfs_close_one_device(struct btrfs_device *device)
1128 struct btrfs_fs_devices *fs_devices = device->fs_devices;
1130 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
1131 device->devid != BTRFS_DEV_REPLACE_DEVID) {
1132 list_del_init(&device->dev_alloc_list);
1133 fs_devices->rw_devices--;
1136 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
1137 fs_devices->missing_devices--;
1139 btrfs_close_bdev(device);
1141 fs_devices->open_devices--;
1142 device->bdev = NULL;
1144 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
1145 btrfs_destroy_dev_zone_info(device);
1147 device->fs_info = NULL;
1148 atomic_set(&device->dev_stats_ccnt, 0);
1149 extent_io_tree_release(&device->alloc_state);
1151 /* Verify the device is back in a pristine state */
1152 ASSERT(!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state));
1153 ASSERT(!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state));
1154 ASSERT(list_empty(&device->dev_alloc_list));
1155 ASSERT(list_empty(&device->post_commit_list));
1156 ASSERT(atomic_read(&device->reada_in_flight) == 0);
1159 static void close_fs_devices(struct btrfs_fs_devices *fs_devices)
1161 struct btrfs_device *device, *tmp;
1163 lockdep_assert_held(&uuid_mutex);
1165 if (--fs_devices->opened > 0)
1168 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list)
1169 btrfs_close_one_device(device);
1171 WARN_ON(fs_devices->open_devices);
1172 WARN_ON(fs_devices->rw_devices);
1173 fs_devices->opened = 0;
1174 fs_devices->seeding = false;
1175 fs_devices->fs_info = NULL;
1178 void btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
1181 struct btrfs_fs_devices *tmp;
1183 mutex_lock(&uuid_mutex);
1184 close_fs_devices(fs_devices);
1185 if (!fs_devices->opened)
1186 list_splice_init(&fs_devices->seed_list, &list);
1188 list_for_each_entry_safe(fs_devices, tmp, &list, seed_list) {
1189 close_fs_devices(fs_devices);
1190 list_del(&fs_devices->seed_list);
1191 free_fs_devices(fs_devices);
1193 mutex_unlock(&uuid_mutex);
1196 static int open_fs_devices(struct btrfs_fs_devices *fs_devices,
1197 fmode_t flags, void *holder)
1199 struct btrfs_device *device;
1200 struct btrfs_device *latest_dev = NULL;
1201 struct btrfs_device *tmp_device;
1203 flags |= FMODE_EXCL;
1205 list_for_each_entry_safe(device, tmp_device, &fs_devices->devices,
1209 ret = btrfs_open_one_device(fs_devices, device, flags, holder);
1211 (!latest_dev || device->generation > latest_dev->generation)) {
1212 latest_dev = device;
1213 } else if (ret == -ENODATA) {
1214 fs_devices->num_devices--;
1215 list_del(&device->dev_list);
1216 btrfs_free_device(device);
1219 if (fs_devices->open_devices == 0)
1222 fs_devices->opened = 1;
1223 fs_devices->latest_bdev = latest_dev->bdev;
1224 fs_devices->total_rw_bytes = 0;
1225 fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_REGULAR;
1226 fs_devices->read_policy = BTRFS_READ_POLICY_PID;
1231 static int devid_cmp(void *priv, struct list_head *a, struct list_head *b)
1233 struct btrfs_device *dev1, *dev2;
1235 dev1 = list_entry(a, struct btrfs_device, dev_list);
1236 dev2 = list_entry(b, struct btrfs_device, dev_list);
1238 if (dev1->devid < dev2->devid)
1240 else if (dev1->devid > dev2->devid)
1245 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1246 fmode_t flags, void *holder)
1250 lockdep_assert_held(&uuid_mutex);
1252 * The device_list_mutex cannot be taken here in case opening the
1253 * underlying device takes further locks like bd_mutex.
1255 * We also don't need the lock here as this is called during mount and
1256 * exclusion is provided by uuid_mutex
1259 if (fs_devices->opened) {
1260 fs_devices->opened++;
1263 list_sort(NULL, &fs_devices->devices, devid_cmp);
1264 ret = open_fs_devices(fs_devices, flags, holder);
1270 void btrfs_release_disk_super(struct btrfs_super_block *super)
1272 struct page *page = virt_to_page(super);
1277 static struct btrfs_super_block *btrfs_read_disk_super(struct block_device *bdev,
1278 u64 bytenr, u64 bytenr_orig)
1280 struct btrfs_super_block *disk_super;
1285 /* make sure our super fits in the device */
1286 if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1287 return ERR_PTR(-EINVAL);
1289 /* make sure our super fits in the page */
1290 if (sizeof(*disk_super) > PAGE_SIZE)
1291 return ERR_PTR(-EINVAL);
1293 /* make sure our super doesn't straddle pages on disk */
1294 index = bytenr >> PAGE_SHIFT;
1295 if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_SHIFT != index)
1296 return ERR_PTR(-EINVAL);
1298 /* pull in the page with our super */
1299 page = read_cache_page_gfp(bdev->bd_inode->i_mapping, index, GFP_KERNEL);
1302 return ERR_CAST(page);
1304 p = page_address(page);
1306 /* align our pointer to the offset of the super block */
1307 disk_super = p + offset_in_page(bytenr);
1309 if (btrfs_super_bytenr(disk_super) != bytenr_orig ||
1310 btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
1311 btrfs_release_disk_super(p);
1312 return ERR_PTR(-EINVAL);
1315 if (disk_super->label[0] && disk_super->label[BTRFS_LABEL_SIZE - 1])
1316 disk_super->label[BTRFS_LABEL_SIZE - 1] = 0;
1321 int btrfs_forget_devices(const char *path)
1325 mutex_lock(&uuid_mutex);
1326 ret = btrfs_free_stale_devices(strlen(path) ? path : NULL, NULL);
1327 mutex_unlock(&uuid_mutex);
1333 * Look for a btrfs signature on a device. This may be called out of the mount path
1334 * and we are not allowed to call set_blocksize during the scan. The superblock
1335 * is read via pagecache
1337 struct btrfs_device *btrfs_scan_one_device(const char *path, fmode_t flags,
1340 struct btrfs_super_block *disk_super;
1341 bool new_device_added = false;
1342 struct btrfs_device *device = NULL;
1343 struct block_device *bdev;
1344 u64 bytenr, bytenr_orig;
1347 lockdep_assert_held(&uuid_mutex);
1350 * we would like to check all the supers, but that would make
1351 * a btrfs mount succeed after a mkfs from a different FS.
1352 * So, we need to add a special mount option to scan for
1353 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1355 flags |= FMODE_EXCL;
1357 bdev = blkdev_get_by_path(path, flags, holder);
1359 return ERR_CAST(bdev);
1361 bytenr_orig = btrfs_sb_offset(0);
1362 ret = btrfs_sb_log_location_bdev(bdev, 0, READ, &bytenr);
1364 return ERR_PTR(ret);
1366 disk_super = btrfs_read_disk_super(bdev, bytenr, bytenr_orig);
1367 if (IS_ERR(disk_super)) {
1368 device = ERR_CAST(disk_super);
1369 goto error_bdev_put;
1372 device = device_list_add(path, disk_super, &new_device_added);
1373 if (!IS_ERR(device)) {
1374 if (new_device_added)
1375 btrfs_free_stale_devices(path, device);
1378 btrfs_release_disk_super(disk_super);
1381 blkdev_put(bdev, flags);
1387 * Try to find a chunk that intersects [start, start + len] range and when one
1388 * such is found, record the end of it in *start
1390 static bool contains_pending_extent(struct btrfs_device *device, u64 *start,
1393 u64 physical_start, physical_end;
1395 lockdep_assert_held(&device->fs_info->chunk_mutex);
1397 if (!find_first_extent_bit(&device->alloc_state, *start,
1398 &physical_start, &physical_end,
1399 CHUNK_ALLOCATED, NULL)) {
1401 if (in_range(physical_start, *start, len) ||
1402 in_range(*start, physical_start,
1403 physical_end - physical_start)) {
1404 *start = physical_end + 1;
1411 static u64 dev_extent_search_start(struct btrfs_device *device, u64 start)
1413 switch (device->fs_devices->chunk_alloc_policy) {
1414 case BTRFS_CHUNK_ALLOC_REGULAR:
1416 * We don't want to overwrite the superblock on the drive nor
1417 * any area used by the boot loader (grub for example), so we
1418 * make sure to start at an offset of at least 1MB.
1420 return max_t(u64, start, SZ_1M);
1427 * dev_extent_hole_check - check if specified hole is suitable for allocation
1428 * @device: the device which we have the hole
1429 * @hole_start: starting position of the hole
1430 * @hole_size: the size of the hole
1431 * @num_bytes: the size of the free space that we need
1433 * This function may modify @hole_start and @hole_end to reflect the suitable
1434 * position for allocation. Returns 1 if hole position is updated, 0 otherwise.
1436 static bool dev_extent_hole_check(struct btrfs_device *device, u64 *hole_start,
1437 u64 *hole_size, u64 num_bytes)
1439 bool changed = false;
1440 u64 hole_end = *hole_start + *hole_size;
1443 * Check before we set max_hole_start, otherwise we could end up
1444 * sending back this offset anyway.
1446 if (contains_pending_extent(device, hole_start, *hole_size)) {
1447 if (hole_end >= *hole_start)
1448 *hole_size = hole_end - *hole_start;
1454 switch (device->fs_devices->chunk_alloc_policy) {
1455 case BTRFS_CHUNK_ALLOC_REGULAR:
1456 /* No extra check */
1466 * find_free_dev_extent_start - find free space in the specified device
1467 * @device: the device which we search the free space in
1468 * @num_bytes: the size of the free space that we need
1469 * @search_start: the position from which to begin the search
1470 * @start: store the start of the free space.
1471 * @len: the size of the free space. that we find, or the size
1472 * of the max free space if we don't find suitable free space
1474 * this uses a pretty simple search, the expectation is that it is
1475 * called very infrequently and that a given device has a small number
1478 * @start is used to store the start of the free space if we find. But if we
1479 * don't find suitable free space, it will be used to store the start position
1480 * of the max free space.
1482 * @len is used to store the size of the free space that we find.
1483 * But if we don't find suitable free space, it is used to store the size of
1484 * the max free space.
1486 * NOTE: This function will search *commit* root of device tree, and does extra
1487 * check to ensure dev extents are not double allocated.
1488 * This makes the function safe to allocate dev extents but may not report
1489 * correct usable device space, as device extent freed in current transaction
1490 * is not reported as avaiable.
1492 static int find_free_dev_extent_start(struct btrfs_device *device,
1493 u64 num_bytes, u64 search_start, u64 *start,
1496 struct btrfs_fs_info *fs_info = device->fs_info;
1497 struct btrfs_root *root = fs_info->dev_root;
1498 struct btrfs_key key;
1499 struct btrfs_dev_extent *dev_extent;
1500 struct btrfs_path *path;
1505 u64 search_end = device->total_bytes;
1508 struct extent_buffer *l;
1510 search_start = dev_extent_search_start(device, search_start);
1512 path = btrfs_alloc_path();
1516 max_hole_start = search_start;
1520 if (search_start >= search_end ||
1521 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1526 path->reada = READA_FORWARD;
1527 path->search_commit_root = 1;
1528 path->skip_locking = 1;
1530 key.objectid = device->devid;
1531 key.offset = search_start;
1532 key.type = BTRFS_DEV_EXTENT_KEY;
1534 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1538 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1545 slot = path->slots[0];
1546 if (slot >= btrfs_header_nritems(l)) {
1547 ret = btrfs_next_leaf(root, path);
1555 btrfs_item_key_to_cpu(l, &key, slot);
1557 if (key.objectid < device->devid)
1560 if (key.objectid > device->devid)
1563 if (key.type != BTRFS_DEV_EXTENT_KEY)
1566 if (key.offset > search_start) {
1567 hole_size = key.offset - search_start;
1568 dev_extent_hole_check(device, &search_start, &hole_size,
1571 if (hole_size > max_hole_size) {
1572 max_hole_start = search_start;
1573 max_hole_size = hole_size;
1577 * If this free space is greater than which we need,
1578 * it must be the max free space that we have found
1579 * until now, so max_hole_start must point to the start
1580 * of this free space and the length of this free space
1581 * is stored in max_hole_size. Thus, we return
1582 * max_hole_start and max_hole_size and go back to the
1585 if (hole_size >= num_bytes) {
1591 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1592 extent_end = key.offset + btrfs_dev_extent_length(l,
1594 if (extent_end > search_start)
1595 search_start = extent_end;
1602 * At this point, search_start should be the end of
1603 * allocated dev extents, and when shrinking the device,
1604 * search_end may be smaller than search_start.
1606 if (search_end > search_start) {
1607 hole_size = search_end - search_start;
1608 if (dev_extent_hole_check(device, &search_start, &hole_size,
1610 btrfs_release_path(path);
1614 if (hole_size > max_hole_size) {
1615 max_hole_start = search_start;
1616 max_hole_size = hole_size;
1621 if (max_hole_size < num_bytes)
1627 btrfs_free_path(path);
1628 *start = max_hole_start;
1630 *len = max_hole_size;
1634 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
1635 u64 *start, u64 *len)
1637 /* FIXME use last free of some kind */
1638 return find_free_dev_extent_start(device, num_bytes, 0, start, len);
1641 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1642 struct btrfs_device *device,
1643 u64 start, u64 *dev_extent_len)
1645 struct btrfs_fs_info *fs_info = device->fs_info;
1646 struct btrfs_root *root = fs_info->dev_root;
1648 struct btrfs_path *path;
1649 struct btrfs_key key;
1650 struct btrfs_key found_key;
1651 struct extent_buffer *leaf = NULL;
1652 struct btrfs_dev_extent *extent = NULL;
1654 path = btrfs_alloc_path();
1658 key.objectid = device->devid;
1660 key.type = BTRFS_DEV_EXTENT_KEY;
1662 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1664 ret = btrfs_previous_item(root, path, key.objectid,
1665 BTRFS_DEV_EXTENT_KEY);
1668 leaf = path->nodes[0];
1669 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1670 extent = btrfs_item_ptr(leaf, path->slots[0],
1671 struct btrfs_dev_extent);
1672 BUG_ON(found_key.offset > start || found_key.offset +
1673 btrfs_dev_extent_length(leaf, extent) < start);
1675 btrfs_release_path(path);
1677 } else if (ret == 0) {
1678 leaf = path->nodes[0];
1679 extent = btrfs_item_ptr(leaf, path->slots[0],
1680 struct btrfs_dev_extent);
1682 btrfs_handle_fs_error(fs_info, ret, "Slot search failed");
1686 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1688 ret = btrfs_del_item(trans, root, path);
1690 btrfs_handle_fs_error(fs_info, ret,
1691 "Failed to remove dev extent item");
1693 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1696 btrfs_free_path(path);
1700 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1701 struct btrfs_device *device,
1702 u64 chunk_offset, u64 start, u64 num_bytes)
1705 struct btrfs_path *path;
1706 struct btrfs_fs_info *fs_info = device->fs_info;
1707 struct btrfs_root *root = fs_info->dev_root;
1708 struct btrfs_dev_extent *extent;
1709 struct extent_buffer *leaf;
1710 struct btrfs_key key;
1712 WARN_ON(!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state));
1713 WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state));
1714 path = btrfs_alloc_path();
1718 key.objectid = device->devid;
1720 key.type = BTRFS_DEV_EXTENT_KEY;
1721 ret = btrfs_insert_empty_item(trans, root, path, &key,
1726 leaf = path->nodes[0];
1727 extent = btrfs_item_ptr(leaf, path->slots[0],
1728 struct btrfs_dev_extent);
1729 btrfs_set_dev_extent_chunk_tree(leaf, extent,
1730 BTRFS_CHUNK_TREE_OBJECTID);
1731 btrfs_set_dev_extent_chunk_objectid(leaf, extent,
1732 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
1733 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1735 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1736 btrfs_mark_buffer_dirty(leaf);
1738 btrfs_free_path(path);
1742 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1744 struct extent_map_tree *em_tree;
1745 struct extent_map *em;
1749 em_tree = &fs_info->mapping_tree;
1750 read_lock(&em_tree->lock);
1751 n = rb_last(&em_tree->map.rb_root);
1753 em = rb_entry(n, struct extent_map, rb_node);
1754 ret = em->start + em->len;
1756 read_unlock(&em_tree->lock);
1761 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1765 struct btrfs_key key;
1766 struct btrfs_key found_key;
1767 struct btrfs_path *path;
1769 path = btrfs_alloc_path();
1773 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1774 key.type = BTRFS_DEV_ITEM_KEY;
1775 key.offset = (u64)-1;
1777 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1783 btrfs_err(fs_info, "corrupted chunk tree devid -1 matched");
1788 ret = btrfs_previous_item(fs_info->chunk_root, path,
1789 BTRFS_DEV_ITEMS_OBJECTID,
1790 BTRFS_DEV_ITEM_KEY);
1794 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1796 *devid_ret = found_key.offset + 1;
1800 btrfs_free_path(path);
1805 * the device information is stored in the chunk root
1806 * the btrfs_device struct should be fully filled in
1808 static int btrfs_add_dev_item(struct btrfs_trans_handle *trans,
1809 struct btrfs_device *device)
1812 struct btrfs_path *path;
1813 struct btrfs_dev_item *dev_item;
1814 struct extent_buffer *leaf;
1815 struct btrfs_key key;
1818 path = btrfs_alloc_path();
1822 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1823 key.type = BTRFS_DEV_ITEM_KEY;
1824 key.offset = device->devid;
1826 ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path,
1827 &key, sizeof(*dev_item));
1831 leaf = path->nodes[0];
1832 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1834 btrfs_set_device_id(leaf, dev_item, device->devid);
1835 btrfs_set_device_generation(leaf, dev_item, 0);
1836 btrfs_set_device_type(leaf, dev_item, device->type);
1837 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1838 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1839 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1840 btrfs_set_device_total_bytes(leaf, dev_item,
1841 btrfs_device_get_disk_total_bytes(device));
1842 btrfs_set_device_bytes_used(leaf, dev_item,
1843 btrfs_device_get_bytes_used(device));
1844 btrfs_set_device_group(leaf, dev_item, 0);
1845 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1846 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1847 btrfs_set_device_start_offset(leaf, dev_item, 0);
1849 ptr = btrfs_device_uuid(dev_item);
1850 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1851 ptr = btrfs_device_fsid(dev_item);
1852 write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid,
1853 ptr, BTRFS_FSID_SIZE);
1854 btrfs_mark_buffer_dirty(leaf);
1858 btrfs_free_path(path);
1863 * Function to update ctime/mtime for a given device path.
1864 * Mainly used for ctime/mtime based probe like libblkid.
1866 static void update_dev_time(const char *path_name)
1870 filp = filp_open(path_name, O_RDWR, 0);
1873 file_update_time(filp);
1874 filp_close(filp, NULL);
1877 static int btrfs_rm_dev_item(struct btrfs_device *device)
1879 struct btrfs_root *root = device->fs_info->chunk_root;
1881 struct btrfs_path *path;
1882 struct btrfs_key key;
1883 struct btrfs_trans_handle *trans;
1885 path = btrfs_alloc_path();
1889 trans = btrfs_start_transaction(root, 0);
1890 if (IS_ERR(trans)) {
1891 btrfs_free_path(path);
1892 return PTR_ERR(trans);
1894 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1895 key.type = BTRFS_DEV_ITEM_KEY;
1896 key.offset = device->devid;
1898 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1902 btrfs_abort_transaction(trans, ret);
1903 btrfs_end_transaction(trans);
1907 ret = btrfs_del_item(trans, root, path);
1909 btrfs_abort_transaction(trans, ret);
1910 btrfs_end_transaction(trans);
1914 btrfs_free_path(path);
1916 ret = btrfs_commit_transaction(trans);
1921 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1922 * filesystem. It's up to the caller to adjust that number regarding eg. device
1925 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1933 seq = read_seqbegin(&fs_info->profiles_lock);
1935 all_avail = fs_info->avail_data_alloc_bits |
1936 fs_info->avail_system_alloc_bits |
1937 fs_info->avail_metadata_alloc_bits;
1938 } while (read_seqretry(&fs_info->profiles_lock, seq));
1940 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1941 if (!(all_avail & btrfs_raid_array[i].bg_flag))
1944 if (num_devices < btrfs_raid_array[i].devs_min) {
1945 int ret = btrfs_raid_array[i].mindev_error;
1955 static struct btrfs_device * btrfs_find_next_active_device(
1956 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device)
1958 struct btrfs_device *next_device;
1960 list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1961 if (next_device != device &&
1962 !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state)
1963 && next_device->bdev)
1971 * Helper function to check if the given device is part of s_bdev / latest_bdev
1972 * and replace it with the provided or the next active device, in the context
1973 * where this function called, there should be always be another device (or
1974 * this_dev) which is active.
1976 void __cold btrfs_assign_next_active_device(struct btrfs_device *device,
1977 struct btrfs_device *next_device)
1979 struct btrfs_fs_info *fs_info = device->fs_info;
1982 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
1984 ASSERT(next_device);
1986 if (fs_info->sb->s_bdev &&
1987 (fs_info->sb->s_bdev == device->bdev))
1988 fs_info->sb->s_bdev = next_device->bdev;
1990 if (fs_info->fs_devices->latest_bdev == device->bdev)
1991 fs_info->fs_devices->latest_bdev = next_device->bdev;
1995 * Return btrfs_fs_devices::num_devices excluding the device that's being
1996 * currently replaced.
1998 static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info)
2000 u64 num_devices = fs_info->fs_devices->num_devices;
2002 down_read(&fs_info->dev_replace.rwsem);
2003 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
2004 ASSERT(num_devices > 1);
2007 up_read(&fs_info->dev_replace.rwsem);
2012 void btrfs_scratch_superblocks(struct btrfs_fs_info *fs_info,
2013 struct block_device *bdev,
2014 const char *device_path)
2016 struct btrfs_super_block *disk_super;
2022 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; copy_num++) {
2026 disk_super = btrfs_read_dev_one_super(bdev, copy_num);
2027 if (IS_ERR(disk_super))
2030 if (bdev_is_zoned(bdev)) {
2031 btrfs_reset_sb_log_zones(bdev, copy_num);
2035 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
2037 page = virt_to_page(disk_super);
2038 set_page_dirty(page);
2040 /* write_on_page() unlocks the page */
2041 ret = write_one_page(page);
2044 "error clearing superblock number %d (%d)",
2046 btrfs_release_disk_super(disk_super);
2050 /* Notify udev that device has changed */
2051 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
2053 /* Update ctime/mtime for device path for libblkid */
2054 update_dev_time(device_path);
2057 int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path,
2060 struct btrfs_device *device;
2061 struct btrfs_fs_devices *cur_devices;
2062 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2066 mutex_lock(&uuid_mutex);
2068 num_devices = btrfs_num_devices(fs_info);
2070 ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
2074 device = btrfs_find_device_by_devspec(fs_info, devid, device_path);
2076 if (IS_ERR(device)) {
2077 if (PTR_ERR(device) == -ENOENT &&
2078 strcmp(device_path, "missing") == 0)
2079 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2081 ret = PTR_ERR(device);
2085 if (btrfs_pinned_by_swapfile(fs_info, device)) {
2086 btrfs_warn_in_rcu(fs_info,
2087 "cannot remove device %s (devid %llu) due to active swapfile",
2088 rcu_str_deref(device->name), device->devid);
2093 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2094 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
2098 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
2099 fs_info->fs_devices->rw_devices == 1) {
2100 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
2104 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2105 mutex_lock(&fs_info->chunk_mutex);
2106 list_del_init(&device->dev_alloc_list);
2107 device->fs_devices->rw_devices--;
2108 mutex_unlock(&fs_info->chunk_mutex);
2111 mutex_unlock(&uuid_mutex);
2112 ret = btrfs_shrink_device(device, 0);
2114 btrfs_reada_remove_dev(device);
2115 mutex_lock(&uuid_mutex);
2120 * TODO: the superblock still includes this device in its num_devices
2121 * counter although write_all_supers() is not locked out. This
2122 * could give a filesystem state which requires a degraded mount.
2124 ret = btrfs_rm_dev_item(device);
2128 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2129 btrfs_scrub_cancel_dev(device);
2132 * the device list mutex makes sure that we don't change
2133 * the device list while someone else is writing out all
2134 * the device supers. Whoever is writing all supers, should
2135 * lock the device list mutex before getting the number of
2136 * devices in the super block (super_copy). Conversely,
2137 * whoever updates the number of devices in the super block
2138 * (super_copy) should hold the device list mutex.
2142 * In normal cases the cur_devices == fs_devices. But in case
2143 * of deleting a seed device, the cur_devices should point to
2144 * its own fs_devices listed under the fs_devices->seed.
2146 cur_devices = device->fs_devices;
2147 mutex_lock(&fs_devices->device_list_mutex);
2148 list_del_rcu(&device->dev_list);
2150 cur_devices->num_devices--;
2151 cur_devices->total_devices--;
2152 /* Update total_devices of the parent fs_devices if it's seed */
2153 if (cur_devices != fs_devices)
2154 fs_devices->total_devices--;
2156 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
2157 cur_devices->missing_devices--;
2159 btrfs_assign_next_active_device(device, NULL);
2162 cur_devices->open_devices--;
2163 /* remove sysfs entry */
2164 btrfs_sysfs_remove_device(device);
2167 num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
2168 btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
2169 mutex_unlock(&fs_devices->device_list_mutex);
2172 * at this point, the device is zero sized and detached from
2173 * the devices list. All that's left is to zero out the old
2174 * supers and free the device.
2176 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2177 btrfs_scratch_superblocks(fs_info, device->bdev,
2180 btrfs_close_bdev(device);
2182 btrfs_free_device(device);
2184 if (cur_devices->open_devices == 0) {
2185 list_del_init(&cur_devices->seed_list);
2186 close_fs_devices(cur_devices);
2187 free_fs_devices(cur_devices);
2191 mutex_unlock(&uuid_mutex);
2195 btrfs_reada_undo_remove_dev(device);
2196 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2197 mutex_lock(&fs_info->chunk_mutex);
2198 list_add(&device->dev_alloc_list,
2199 &fs_devices->alloc_list);
2200 device->fs_devices->rw_devices++;
2201 mutex_unlock(&fs_info->chunk_mutex);
2206 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev)
2208 struct btrfs_fs_devices *fs_devices;
2210 lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex);
2213 * in case of fs with no seed, srcdev->fs_devices will point
2214 * to fs_devices of fs_info. However when the dev being replaced is
2215 * a seed dev it will point to the seed's local fs_devices. In short
2216 * srcdev will have its correct fs_devices in both the cases.
2218 fs_devices = srcdev->fs_devices;
2220 list_del_rcu(&srcdev->dev_list);
2221 list_del(&srcdev->dev_alloc_list);
2222 fs_devices->num_devices--;
2223 if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state))
2224 fs_devices->missing_devices--;
2226 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state))
2227 fs_devices->rw_devices--;
2230 fs_devices->open_devices--;
2233 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev)
2235 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2237 mutex_lock(&uuid_mutex);
2239 btrfs_close_bdev(srcdev);
2241 btrfs_free_device(srcdev);
2243 /* if this is no devs we rather delete the fs_devices */
2244 if (!fs_devices->num_devices) {
2246 * On a mounted FS, num_devices can't be zero unless it's a
2247 * seed. In case of a seed device being replaced, the replace
2248 * target added to the sprout FS, so there will be no more
2249 * device left under the seed FS.
2251 ASSERT(fs_devices->seeding);
2253 list_del_init(&fs_devices->seed_list);
2254 close_fs_devices(fs_devices);
2255 free_fs_devices(fs_devices);
2257 mutex_unlock(&uuid_mutex);
2260 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev)
2262 struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices;
2264 mutex_lock(&fs_devices->device_list_mutex);
2266 btrfs_sysfs_remove_device(tgtdev);
2269 fs_devices->open_devices--;
2271 fs_devices->num_devices--;
2273 btrfs_assign_next_active_device(tgtdev, NULL);
2275 list_del_rcu(&tgtdev->dev_list);
2277 mutex_unlock(&fs_devices->device_list_mutex);
2280 * The update_dev_time() with in btrfs_scratch_superblocks()
2281 * may lead to a call to btrfs_show_devname() which will try
2282 * to hold device_list_mutex. And here this device
2283 * is already out of device list, so we don't have to hold
2284 * the device_list_mutex lock.
2286 btrfs_scratch_superblocks(tgtdev->fs_info, tgtdev->bdev,
2289 btrfs_close_bdev(tgtdev);
2291 btrfs_free_device(tgtdev);
2294 static struct btrfs_device *btrfs_find_device_by_path(
2295 struct btrfs_fs_info *fs_info, const char *device_path)
2298 struct btrfs_super_block *disk_super;
2301 struct block_device *bdev;
2302 struct btrfs_device *device;
2304 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2305 fs_info->bdev_holder, 0, &bdev, &disk_super);
2307 return ERR_PTR(ret);
2309 devid = btrfs_stack_device_id(&disk_super->dev_item);
2310 dev_uuid = disk_super->dev_item.uuid;
2311 if (btrfs_fs_incompat(fs_info, METADATA_UUID))
2312 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
2313 disk_super->metadata_uuid);
2315 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
2318 btrfs_release_disk_super(disk_super);
2320 device = ERR_PTR(-ENOENT);
2321 blkdev_put(bdev, FMODE_READ);
2326 * Lookup a device given by device id, or the path if the id is 0.
2328 struct btrfs_device *btrfs_find_device_by_devspec(
2329 struct btrfs_fs_info *fs_info, u64 devid,
2330 const char *device_path)
2332 struct btrfs_device *device;
2335 device = btrfs_find_device(fs_info->fs_devices, devid, NULL,
2338 return ERR_PTR(-ENOENT);
2342 if (!device_path || !device_path[0])
2343 return ERR_PTR(-EINVAL);
2345 if (strcmp(device_path, "missing") == 0) {
2346 /* Find first missing device */
2347 list_for_each_entry(device, &fs_info->fs_devices->devices,
2349 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
2350 &device->dev_state) && !device->bdev)
2353 return ERR_PTR(-ENOENT);
2356 return btrfs_find_device_by_path(fs_info, device_path);
2360 * does all the dirty work required for changing file system's UUID.
2362 static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info)
2364 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2365 struct btrfs_fs_devices *old_devices;
2366 struct btrfs_fs_devices *seed_devices;
2367 struct btrfs_super_block *disk_super = fs_info->super_copy;
2368 struct btrfs_device *device;
2371 lockdep_assert_held(&uuid_mutex);
2372 if (!fs_devices->seeding)
2376 * Private copy of the seed devices, anchored at
2377 * fs_info->fs_devices->seed_list
2379 seed_devices = alloc_fs_devices(NULL, NULL);
2380 if (IS_ERR(seed_devices))
2381 return PTR_ERR(seed_devices);
2384 * It's necessary to retain a copy of the original seed fs_devices in
2385 * fs_uuids so that filesystems which have been seeded can successfully
2386 * reference the seed device from open_seed_devices. This also supports
2389 old_devices = clone_fs_devices(fs_devices);
2390 if (IS_ERR(old_devices)) {
2391 kfree(seed_devices);
2392 return PTR_ERR(old_devices);
2395 list_add(&old_devices->fs_list, &fs_uuids);
2397 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2398 seed_devices->opened = 1;
2399 INIT_LIST_HEAD(&seed_devices->devices);
2400 INIT_LIST_HEAD(&seed_devices->alloc_list);
2401 mutex_init(&seed_devices->device_list_mutex);
2403 mutex_lock(&fs_devices->device_list_mutex);
2404 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2406 list_for_each_entry(device, &seed_devices->devices, dev_list)
2407 device->fs_devices = seed_devices;
2409 fs_devices->seeding = false;
2410 fs_devices->num_devices = 0;
2411 fs_devices->open_devices = 0;
2412 fs_devices->missing_devices = 0;
2413 fs_devices->rotating = false;
2414 list_add(&seed_devices->seed_list, &fs_devices->seed_list);
2416 generate_random_uuid(fs_devices->fsid);
2417 memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE);
2418 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2419 mutex_unlock(&fs_devices->device_list_mutex);
2421 super_flags = btrfs_super_flags(disk_super) &
2422 ~BTRFS_SUPER_FLAG_SEEDING;
2423 btrfs_set_super_flags(disk_super, super_flags);
2429 * Store the expected generation for seed devices in device items.
2431 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans)
2433 struct btrfs_fs_info *fs_info = trans->fs_info;
2434 struct btrfs_root *root = fs_info->chunk_root;
2435 struct btrfs_path *path;
2436 struct extent_buffer *leaf;
2437 struct btrfs_dev_item *dev_item;
2438 struct btrfs_device *device;
2439 struct btrfs_key key;
2440 u8 fs_uuid[BTRFS_FSID_SIZE];
2441 u8 dev_uuid[BTRFS_UUID_SIZE];
2445 path = btrfs_alloc_path();
2449 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2451 key.type = BTRFS_DEV_ITEM_KEY;
2454 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2458 leaf = path->nodes[0];
2460 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2461 ret = btrfs_next_leaf(root, path);
2466 leaf = path->nodes[0];
2467 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2468 btrfs_release_path(path);
2472 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2473 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2474 key.type != BTRFS_DEV_ITEM_KEY)
2477 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2478 struct btrfs_dev_item);
2479 devid = btrfs_device_id(leaf, dev_item);
2480 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2482 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2484 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
2486 BUG_ON(!device); /* Logic error */
2488 if (device->fs_devices->seeding) {
2489 btrfs_set_device_generation(leaf, dev_item,
2490 device->generation);
2491 btrfs_mark_buffer_dirty(leaf);
2499 btrfs_free_path(path);
2503 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2505 struct btrfs_root *root = fs_info->dev_root;
2506 struct request_queue *q;
2507 struct btrfs_trans_handle *trans;
2508 struct btrfs_device *device;
2509 struct block_device *bdev;
2510 struct super_block *sb = fs_info->sb;
2511 struct rcu_string *name;
2512 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2513 u64 orig_super_total_bytes;
2514 u64 orig_super_num_devices;
2515 int seeding_dev = 0;
2517 bool locked = false;
2519 if (sb_rdonly(sb) && !fs_devices->seeding)
2522 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2523 fs_info->bdev_holder);
2525 return PTR_ERR(bdev);
2527 if (!btrfs_check_device_zone_type(fs_info, bdev)) {
2532 if (fs_devices->seeding) {
2534 down_write(&sb->s_umount);
2535 mutex_lock(&uuid_mutex);
2539 sync_blockdev(bdev);
2542 list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
2543 if (device->bdev == bdev) {
2551 device = btrfs_alloc_device(fs_info, NULL, NULL);
2552 if (IS_ERR(device)) {
2553 /* we can safely leave the fs_devices entry around */
2554 ret = PTR_ERR(device);
2558 name = rcu_string_strdup(device_path, GFP_KERNEL);
2561 goto error_free_device;
2563 rcu_assign_pointer(device->name, name);
2565 device->fs_info = fs_info;
2566 device->bdev = bdev;
2568 ret = btrfs_get_dev_zone_info(device);
2570 goto error_free_device;
2572 trans = btrfs_start_transaction(root, 0);
2573 if (IS_ERR(trans)) {
2574 ret = PTR_ERR(trans);
2575 goto error_free_zone;
2578 q = bdev_get_queue(bdev);
2579 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
2580 device->generation = trans->transid;
2581 device->io_width = fs_info->sectorsize;
2582 device->io_align = fs_info->sectorsize;
2583 device->sector_size = fs_info->sectorsize;
2584 device->total_bytes = round_down(i_size_read(bdev->bd_inode),
2585 fs_info->sectorsize);
2586 device->disk_total_bytes = device->total_bytes;
2587 device->commit_total_bytes = device->total_bytes;
2588 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2589 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
2590 device->mode = FMODE_EXCL;
2591 device->dev_stats_valid = 1;
2592 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2595 btrfs_clear_sb_rdonly(sb);
2596 ret = btrfs_prepare_sprout(fs_info);
2598 btrfs_abort_transaction(trans, ret);
2603 device->fs_devices = fs_devices;
2605 mutex_lock(&fs_devices->device_list_mutex);
2606 mutex_lock(&fs_info->chunk_mutex);
2607 list_add_rcu(&device->dev_list, &fs_devices->devices);
2608 list_add(&device->dev_alloc_list, &fs_devices->alloc_list);
2609 fs_devices->num_devices++;
2610 fs_devices->open_devices++;
2611 fs_devices->rw_devices++;
2612 fs_devices->total_devices++;
2613 fs_devices->total_rw_bytes += device->total_bytes;
2615 atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2617 if (!blk_queue_nonrot(q))
2618 fs_devices->rotating = true;
2620 orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
2621 btrfs_set_super_total_bytes(fs_info->super_copy,
2622 round_down(orig_super_total_bytes + device->total_bytes,
2623 fs_info->sectorsize));
2625 orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy);
2626 btrfs_set_super_num_devices(fs_info->super_copy,
2627 orig_super_num_devices + 1);
2630 * we've got more storage, clear any full flags on the space
2633 btrfs_clear_space_info_full(fs_info);
2635 mutex_unlock(&fs_info->chunk_mutex);
2637 /* Add sysfs device entry */
2638 btrfs_sysfs_add_device(device);
2640 mutex_unlock(&fs_devices->device_list_mutex);
2643 mutex_lock(&fs_info->chunk_mutex);
2644 ret = init_first_rw_device(trans);
2645 mutex_unlock(&fs_info->chunk_mutex);
2647 btrfs_abort_transaction(trans, ret);
2652 ret = btrfs_add_dev_item(trans, device);
2654 btrfs_abort_transaction(trans, ret);
2659 ret = btrfs_finish_sprout(trans);
2661 btrfs_abort_transaction(trans, ret);
2666 * fs_devices now represents the newly sprouted filesystem and
2667 * its fsid has been changed by btrfs_prepare_sprout
2669 btrfs_sysfs_update_sprout_fsid(fs_devices);
2672 ret = btrfs_commit_transaction(trans);
2675 mutex_unlock(&uuid_mutex);
2676 up_write(&sb->s_umount);
2679 if (ret) /* transaction commit */
2682 ret = btrfs_relocate_sys_chunks(fs_info);
2684 btrfs_handle_fs_error(fs_info, ret,
2685 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2686 trans = btrfs_attach_transaction(root);
2687 if (IS_ERR(trans)) {
2688 if (PTR_ERR(trans) == -ENOENT)
2690 ret = PTR_ERR(trans);
2694 ret = btrfs_commit_transaction(trans);
2698 * Now that we have written a new super block to this device, check all
2699 * other fs_devices list if device_path alienates any other scanned
2701 * We can ignore the return value as it typically returns -EINVAL and
2702 * only succeeds if the device was an alien.
2704 btrfs_forget_devices(device_path);
2706 /* Update ctime/mtime for blkid or udev */
2707 update_dev_time(device_path);
2712 btrfs_sysfs_remove_device(device);
2713 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2714 mutex_lock(&fs_info->chunk_mutex);
2715 list_del_rcu(&device->dev_list);
2716 list_del(&device->dev_alloc_list);
2717 fs_info->fs_devices->num_devices--;
2718 fs_info->fs_devices->open_devices--;
2719 fs_info->fs_devices->rw_devices--;
2720 fs_info->fs_devices->total_devices--;
2721 fs_info->fs_devices->total_rw_bytes -= device->total_bytes;
2722 atomic64_sub(device->total_bytes, &fs_info->free_chunk_space);
2723 btrfs_set_super_total_bytes(fs_info->super_copy,
2724 orig_super_total_bytes);
2725 btrfs_set_super_num_devices(fs_info->super_copy,
2726 orig_super_num_devices);
2727 mutex_unlock(&fs_info->chunk_mutex);
2728 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2731 btrfs_set_sb_rdonly(sb);
2733 btrfs_end_transaction(trans);
2735 btrfs_destroy_dev_zone_info(device);
2737 btrfs_free_device(device);
2739 blkdev_put(bdev, FMODE_EXCL);
2741 mutex_unlock(&uuid_mutex);
2742 up_write(&sb->s_umount);
2747 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2748 struct btrfs_device *device)
2751 struct btrfs_path *path;
2752 struct btrfs_root *root = device->fs_info->chunk_root;
2753 struct btrfs_dev_item *dev_item;
2754 struct extent_buffer *leaf;
2755 struct btrfs_key key;
2757 path = btrfs_alloc_path();
2761 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2762 key.type = BTRFS_DEV_ITEM_KEY;
2763 key.offset = device->devid;
2765 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2774 leaf = path->nodes[0];
2775 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2777 btrfs_set_device_id(leaf, dev_item, device->devid);
2778 btrfs_set_device_type(leaf, dev_item, device->type);
2779 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2780 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2781 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2782 btrfs_set_device_total_bytes(leaf, dev_item,
2783 btrfs_device_get_disk_total_bytes(device));
2784 btrfs_set_device_bytes_used(leaf, dev_item,
2785 btrfs_device_get_bytes_used(device));
2786 btrfs_mark_buffer_dirty(leaf);
2789 btrfs_free_path(path);
2793 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2794 struct btrfs_device *device, u64 new_size)
2796 struct btrfs_fs_info *fs_info = device->fs_info;
2797 struct btrfs_super_block *super_copy = fs_info->super_copy;
2801 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2804 new_size = round_down(new_size, fs_info->sectorsize);
2806 mutex_lock(&fs_info->chunk_mutex);
2807 old_total = btrfs_super_total_bytes(super_copy);
2808 diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2810 if (new_size <= device->total_bytes ||
2811 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2812 mutex_unlock(&fs_info->chunk_mutex);
2816 btrfs_set_super_total_bytes(super_copy,
2817 round_down(old_total + diff, fs_info->sectorsize));
2818 device->fs_devices->total_rw_bytes += diff;
2820 btrfs_device_set_total_bytes(device, new_size);
2821 btrfs_device_set_disk_total_bytes(device, new_size);
2822 btrfs_clear_space_info_full(device->fs_info);
2823 if (list_empty(&device->post_commit_list))
2824 list_add_tail(&device->post_commit_list,
2825 &trans->transaction->dev_update_list);
2826 mutex_unlock(&fs_info->chunk_mutex);
2828 return btrfs_update_device(trans, device);
2831 static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
2833 struct btrfs_fs_info *fs_info = trans->fs_info;
2834 struct btrfs_root *root = fs_info->chunk_root;
2836 struct btrfs_path *path;
2837 struct btrfs_key key;
2839 path = btrfs_alloc_path();
2843 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2844 key.offset = chunk_offset;
2845 key.type = BTRFS_CHUNK_ITEM_KEY;
2847 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2850 else if (ret > 0) { /* Logic error or corruption */
2851 btrfs_handle_fs_error(fs_info, -ENOENT,
2852 "Failed lookup while freeing chunk.");
2857 ret = btrfs_del_item(trans, root, path);
2859 btrfs_handle_fs_error(fs_info, ret,
2860 "Failed to delete chunk item.");
2862 btrfs_free_path(path);
2866 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2868 struct btrfs_super_block *super_copy = fs_info->super_copy;
2869 struct btrfs_disk_key *disk_key;
2870 struct btrfs_chunk *chunk;
2877 struct btrfs_key key;
2879 mutex_lock(&fs_info->chunk_mutex);
2880 array_size = btrfs_super_sys_array_size(super_copy);
2882 ptr = super_copy->sys_chunk_array;
2885 while (cur < array_size) {
2886 disk_key = (struct btrfs_disk_key *)ptr;
2887 btrfs_disk_key_to_cpu(&key, disk_key);
2889 len = sizeof(*disk_key);
2891 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2892 chunk = (struct btrfs_chunk *)(ptr + len);
2893 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2894 len += btrfs_chunk_item_size(num_stripes);
2899 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID &&
2900 key.offset == chunk_offset) {
2901 memmove(ptr, ptr + len, array_size - (cur + len));
2903 btrfs_set_super_sys_array_size(super_copy, array_size);
2909 mutex_unlock(&fs_info->chunk_mutex);
2914 * btrfs_get_chunk_map() - Find the mapping containing the given logical extent.
2915 * @logical: Logical block offset in bytes.
2916 * @length: Length of extent in bytes.
2918 * Return: Chunk mapping or ERR_PTR.
2920 struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info,
2921 u64 logical, u64 length)
2923 struct extent_map_tree *em_tree;
2924 struct extent_map *em;
2926 em_tree = &fs_info->mapping_tree;
2927 read_lock(&em_tree->lock);
2928 em = lookup_extent_mapping(em_tree, logical, length);
2929 read_unlock(&em_tree->lock);
2932 btrfs_crit(fs_info, "unable to find logical %llu length %llu",
2934 return ERR_PTR(-EINVAL);
2937 if (em->start > logical || em->start + em->len < logical) {
2939 "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
2940 logical, length, em->start, em->start + em->len);
2941 free_extent_map(em);
2942 return ERR_PTR(-EINVAL);
2945 /* callers are responsible for dropping em's ref. */
2949 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
2951 struct btrfs_fs_info *fs_info = trans->fs_info;
2952 struct extent_map *em;
2953 struct map_lookup *map;
2954 u64 dev_extent_len = 0;
2956 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2958 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
2961 * This is a logic error, but we don't want to just rely on the
2962 * user having built with ASSERT enabled, so if ASSERT doesn't
2963 * do anything we still error out.
2968 map = em->map_lookup;
2969 mutex_lock(&fs_info->chunk_mutex);
2970 check_system_chunk(trans, map->type);
2971 mutex_unlock(&fs_info->chunk_mutex);
2974 * Take the device list mutex to prevent races with the final phase of
2975 * a device replace operation that replaces the device object associated
2976 * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2978 mutex_lock(&fs_devices->device_list_mutex);
2979 for (i = 0; i < map->num_stripes; i++) {
2980 struct btrfs_device *device = map->stripes[i].dev;
2981 ret = btrfs_free_dev_extent(trans, device,
2982 map->stripes[i].physical,
2985 mutex_unlock(&fs_devices->device_list_mutex);
2986 btrfs_abort_transaction(trans, ret);
2990 if (device->bytes_used > 0) {
2991 mutex_lock(&fs_info->chunk_mutex);
2992 btrfs_device_set_bytes_used(device,
2993 device->bytes_used - dev_extent_len);
2994 atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
2995 btrfs_clear_space_info_full(fs_info);
2996 mutex_unlock(&fs_info->chunk_mutex);
2999 ret = btrfs_update_device(trans, device);
3001 mutex_unlock(&fs_devices->device_list_mutex);
3002 btrfs_abort_transaction(trans, ret);
3006 mutex_unlock(&fs_devices->device_list_mutex);
3008 ret = btrfs_free_chunk(trans, chunk_offset);
3010 btrfs_abort_transaction(trans, ret);
3014 trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
3016 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
3017 ret = btrfs_del_sys_chunk(fs_info, chunk_offset);
3019 btrfs_abort_transaction(trans, ret);
3024 ret = btrfs_remove_block_group(trans, chunk_offset, em);
3026 btrfs_abort_transaction(trans, ret);
3032 free_extent_map(em);
3036 static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
3038 struct btrfs_root *root = fs_info->chunk_root;
3039 struct btrfs_trans_handle *trans;
3040 struct btrfs_block_group *block_group;
3044 * Prevent races with automatic removal of unused block groups.
3045 * After we relocate and before we remove the chunk with offset
3046 * chunk_offset, automatic removal of the block group can kick in,
3047 * resulting in a failure when calling btrfs_remove_chunk() below.
3049 * Make sure to acquire this mutex before doing a tree search (dev
3050 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
3051 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
3052 * we release the path used to search the chunk/dev tree and before
3053 * the current task acquires this mutex and calls us.
3055 lockdep_assert_held(&fs_info->delete_unused_bgs_mutex);
3057 /* step one, relocate all the extents inside this chunk */
3058 btrfs_scrub_pause(fs_info);
3059 ret = btrfs_relocate_block_group(fs_info, chunk_offset);
3060 btrfs_scrub_continue(fs_info);
3064 block_group = btrfs_lookup_block_group(fs_info, chunk_offset);
3067 btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group);
3068 btrfs_put_block_group(block_group);
3070 trans = btrfs_start_trans_remove_block_group(root->fs_info,
3072 if (IS_ERR(trans)) {
3073 ret = PTR_ERR(trans);
3074 btrfs_handle_fs_error(root->fs_info, ret, NULL);
3079 * step two, delete the device extents and the
3080 * chunk tree entries
3082 ret = btrfs_remove_chunk(trans, chunk_offset);
3083 btrfs_end_transaction(trans);
3087 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
3089 struct btrfs_root *chunk_root = fs_info->chunk_root;
3090 struct btrfs_path *path;
3091 struct extent_buffer *leaf;
3092 struct btrfs_chunk *chunk;
3093 struct btrfs_key key;
3094 struct btrfs_key found_key;
3096 bool retried = false;
3100 path = btrfs_alloc_path();
3105 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3106 key.offset = (u64)-1;
3107 key.type = BTRFS_CHUNK_ITEM_KEY;
3110 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3111 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3113 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3116 BUG_ON(ret == 0); /* Corruption */
3118 ret = btrfs_previous_item(chunk_root, path, key.objectid,
3121 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3127 leaf = path->nodes[0];
3128 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3130 chunk = btrfs_item_ptr(leaf, path->slots[0],
3131 struct btrfs_chunk);
3132 chunk_type = btrfs_chunk_type(leaf, chunk);
3133 btrfs_release_path(path);
3135 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3136 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3142 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3144 if (found_key.offset == 0)
3146 key.offset = found_key.offset - 1;
3149 if (failed && !retried) {
3153 } else if (WARN_ON(failed && retried)) {
3157 btrfs_free_path(path);
3162 * return 1 : allocate a data chunk successfully,
3163 * return <0: errors during allocating a data chunk,
3164 * return 0 : no need to allocate a data chunk.
3166 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info,
3169 struct btrfs_block_group *cache;
3173 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3175 chunk_type = cache->flags;
3176 btrfs_put_block_group(cache);
3178 if (!(chunk_type & BTRFS_BLOCK_GROUP_DATA))
3181 spin_lock(&fs_info->data_sinfo->lock);
3182 bytes_used = fs_info->data_sinfo->bytes_used;
3183 spin_unlock(&fs_info->data_sinfo->lock);
3186 struct btrfs_trans_handle *trans;
3189 trans = btrfs_join_transaction(fs_info->tree_root);
3191 return PTR_ERR(trans);
3193 ret = btrfs_force_chunk_alloc(trans, BTRFS_BLOCK_GROUP_DATA);
3194 btrfs_end_transaction(trans);
3203 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3204 struct btrfs_balance_control *bctl)
3206 struct btrfs_root *root = fs_info->tree_root;
3207 struct btrfs_trans_handle *trans;
3208 struct btrfs_balance_item *item;
3209 struct btrfs_disk_balance_args disk_bargs;
3210 struct btrfs_path *path;
3211 struct extent_buffer *leaf;
3212 struct btrfs_key key;
3215 path = btrfs_alloc_path();
3219 trans = btrfs_start_transaction(root, 0);
3220 if (IS_ERR(trans)) {
3221 btrfs_free_path(path);
3222 return PTR_ERR(trans);
3225 key.objectid = BTRFS_BALANCE_OBJECTID;
3226 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3229 ret = btrfs_insert_empty_item(trans, root, path, &key,
3234 leaf = path->nodes[0];
3235 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3237 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3239 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3240 btrfs_set_balance_data(leaf, item, &disk_bargs);
3241 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3242 btrfs_set_balance_meta(leaf, item, &disk_bargs);
3243 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3244 btrfs_set_balance_sys(leaf, item, &disk_bargs);
3246 btrfs_set_balance_flags(leaf, item, bctl->flags);
3248 btrfs_mark_buffer_dirty(leaf);
3250 btrfs_free_path(path);
3251 err = btrfs_commit_transaction(trans);
3257 static int del_balance_item(struct btrfs_fs_info *fs_info)
3259 struct btrfs_root *root = fs_info->tree_root;
3260 struct btrfs_trans_handle *trans;
3261 struct btrfs_path *path;
3262 struct btrfs_key key;
3265 path = btrfs_alloc_path();
3269 trans = btrfs_start_transaction_fallback_global_rsv(root, 0);
3270 if (IS_ERR(trans)) {
3271 btrfs_free_path(path);
3272 return PTR_ERR(trans);
3275 key.objectid = BTRFS_BALANCE_OBJECTID;
3276 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3279 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3287 ret = btrfs_del_item(trans, root, path);
3289 btrfs_free_path(path);
3290 err = btrfs_commit_transaction(trans);
3297 * This is a heuristic used to reduce the number of chunks balanced on
3298 * resume after balance was interrupted.
3300 static void update_balance_args(struct btrfs_balance_control *bctl)
3303 * Turn on soft mode for chunk types that were being converted.
3305 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3306 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3307 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3308 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3309 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3310 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3313 * Turn on usage filter if is not already used. The idea is
3314 * that chunks that we have already balanced should be
3315 * reasonably full. Don't do it for chunks that are being
3316 * converted - that will keep us from relocating unconverted
3317 * (albeit full) chunks.
3319 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3320 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3321 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3322 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3323 bctl->data.usage = 90;
3325 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3326 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3327 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3328 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3329 bctl->sys.usage = 90;
3331 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3332 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3333 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3334 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3335 bctl->meta.usage = 90;
3340 * Clear the balance status in fs_info and delete the balance item from disk.
3342 static void reset_balance_state(struct btrfs_fs_info *fs_info)
3344 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3347 BUG_ON(!fs_info->balance_ctl);
3349 spin_lock(&fs_info->balance_lock);
3350 fs_info->balance_ctl = NULL;
3351 spin_unlock(&fs_info->balance_lock);
3354 ret = del_balance_item(fs_info);
3356 btrfs_handle_fs_error(fs_info, ret, NULL);
3360 * Balance filters. Return 1 if chunk should be filtered out
3361 * (should not be balanced).
3363 static int chunk_profiles_filter(u64 chunk_type,
3364 struct btrfs_balance_args *bargs)
3366 chunk_type = chunk_to_extended(chunk_type) &
3367 BTRFS_EXTENDED_PROFILE_MASK;
3369 if (bargs->profiles & chunk_type)
3375 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3376 struct btrfs_balance_args *bargs)
3378 struct btrfs_block_group *cache;
3380 u64 user_thresh_min;
3381 u64 user_thresh_max;
3384 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3385 chunk_used = cache->used;
3387 if (bargs->usage_min == 0)
3388 user_thresh_min = 0;
3390 user_thresh_min = div_factor_fine(cache->length,
3393 if (bargs->usage_max == 0)
3394 user_thresh_max = 1;
3395 else if (bargs->usage_max > 100)
3396 user_thresh_max = cache->length;
3398 user_thresh_max = div_factor_fine(cache->length,
3401 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3404 btrfs_put_block_group(cache);
3408 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3409 u64 chunk_offset, struct btrfs_balance_args *bargs)
3411 struct btrfs_block_group *cache;
3412 u64 chunk_used, user_thresh;
3415 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3416 chunk_used = cache->used;
3418 if (bargs->usage_min == 0)
3420 else if (bargs->usage > 100)
3421 user_thresh = cache->length;
3423 user_thresh = div_factor_fine(cache->length, bargs->usage);
3425 if (chunk_used < user_thresh)
3428 btrfs_put_block_group(cache);
3432 static int chunk_devid_filter(struct extent_buffer *leaf,
3433 struct btrfs_chunk *chunk,
3434 struct btrfs_balance_args *bargs)
3436 struct btrfs_stripe *stripe;
3437 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3440 for (i = 0; i < num_stripes; i++) {
3441 stripe = btrfs_stripe_nr(chunk, i);
3442 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3449 static u64 calc_data_stripes(u64 type, int num_stripes)
3451 const int index = btrfs_bg_flags_to_raid_index(type);
3452 const int ncopies = btrfs_raid_array[index].ncopies;
3453 const int nparity = btrfs_raid_array[index].nparity;
3456 return num_stripes - nparity;
3458 return num_stripes / ncopies;
3461 /* [pstart, pend) */
3462 static int chunk_drange_filter(struct extent_buffer *leaf,
3463 struct btrfs_chunk *chunk,
3464 struct btrfs_balance_args *bargs)
3466 struct btrfs_stripe *stripe;
3467 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3474 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3477 type = btrfs_chunk_type(leaf, chunk);
3478 factor = calc_data_stripes(type, num_stripes);
3480 for (i = 0; i < num_stripes; i++) {
3481 stripe = btrfs_stripe_nr(chunk, i);
3482 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3485 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3486 stripe_length = btrfs_chunk_length(leaf, chunk);
3487 stripe_length = div_u64(stripe_length, factor);
3489 if (stripe_offset < bargs->pend &&
3490 stripe_offset + stripe_length > bargs->pstart)
3497 /* [vstart, vend) */
3498 static int chunk_vrange_filter(struct extent_buffer *leaf,
3499 struct btrfs_chunk *chunk,
3501 struct btrfs_balance_args *bargs)
3503 if (chunk_offset < bargs->vend &&
3504 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3505 /* at least part of the chunk is inside this vrange */
3511 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3512 struct btrfs_chunk *chunk,
3513 struct btrfs_balance_args *bargs)
3515 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3517 if (bargs->stripes_min <= num_stripes
3518 && num_stripes <= bargs->stripes_max)
3524 static int chunk_soft_convert_filter(u64 chunk_type,
3525 struct btrfs_balance_args *bargs)
3527 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3530 chunk_type = chunk_to_extended(chunk_type) &
3531 BTRFS_EXTENDED_PROFILE_MASK;
3533 if (bargs->target == chunk_type)
3539 static int should_balance_chunk(struct extent_buffer *leaf,
3540 struct btrfs_chunk *chunk, u64 chunk_offset)
3542 struct btrfs_fs_info *fs_info = leaf->fs_info;
3543 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3544 struct btrfs_balance_args *bargs = NULL;
3545 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3548 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3549 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3553 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3554 bargs = &bctl->data;
3555 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3557 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3558 bargs = &bctl->meta;
3560 /* profiles filter */
3561 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3562 chunk_profiles_filter(chunk_type, bargs)) {
3567 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3568 chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3570 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3571 chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3576 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3577 chunk_devid_filter(leaf, chunk, bargs)) {
3581 /* drange filter, makes sense only with devid filter */
3582 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3583 chunk_drange_filter(leaf, chunk, bargs)) {
3588 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3589 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3593 /* stripes filter */
3594 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3595 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3599 /* soft profile changing mode */
3600 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3601 chunk_soft_convert_filter(chunk_type, bargs)) {
3606 * limited by count, must be the last filter
3608 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3609 if (bargs->limit == 0)
3613 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3615 * Same logic as the 'limit' filter; the minimum cannot be
3616 * determined here because we do not have the global information
3617 * about the count of all chunks that satisfy the filters.
3619 if (bargs->limit_max == 0)
3628 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3630 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3631 struct btrfs_root *chunk_root = fs_info->chunk_root;
3633 struct btrfs_chunk *chunk;
3634 struct btrfs_path *path = NULL;
3635 struct btrfs_key key;
3636 struct btrfs_key found_key;
3637 struct extent_buffer *leaf;
3640 int enospc_errors = 0;
3641 bool counting = true;
3642 /* The single value limit and min/max limits use the same bytes in the */
3643 u64 limit_data = bctl->data.limit;
3644 u64 limit_meta = bctl->meta.limit;
3645 u64 limit_sys = bctl->sys.limit;
3649 int chunk_reserved = 0;
3651 path = btrfs_alloc_path();
3657 /* zero out stat counters */
3658 spin_lock(&fs_info->balance_lock);
3659 memset(&bctl->stat, 0, sizeof(bctl->stat));
3660 spin_unlock(&fs_info->balance_lock);
3664 * The single value limit and min/max limits use the same bytes
3667 bctl->data.limit = limit_data;
3668 bctl->meta.limit = limit_meta;
3669 bctl->sys.limit = limit_sys;
3671 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3672 key.offset = (u64)-1;
3673 key.type = BTRFS_CHUNK_ITEM_KEY;
3676 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3677 atomic_read(&fs_info->balance_cancel_req)) {
3682 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3683 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3685 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3690 * this shouldn't happen, it means the last relocate
3694 BUG(); /* FIXME break ? */
3696 ret = btrfs_previous_item(chunk_root, path, 0,
3697 BTRFS_CHUNK_ITEM_KEY);
3699 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3704 leaf = path->nodes[0];
3705 slot = path->slots[0];
3706 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3708 if (found_key.objectid != key.objectid) {
3709 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3713 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3714 chunk_type = btrfs_chunk_type(leaf, chunk);
3717 spin_lock(&fs_info->balance_lock);
3718 bctl->stat.considered++;
3719 spin_unlock(&fs_info->balance_lock);
3722 ret = should_balance_chunk(leaf, chunk, found_key.offset);
3724 btrfs_release_path(path);
3726 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3731 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3732 spin_lock(&fs_info->balance_lock);
3733 bctl->stat.expected++;
3734 spin_unlock(&fs_info->balance_lock);
3736 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3738 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3740 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3747 * Apply limit_min filter, no need to check if the LIMITS
3748 * filter is used, limit_min is 0 by default
3750 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3751 count_data < bctl->data.limit_min)
3752 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3753 count_meta < bctl->meta.limit_min)
3754 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3755 count_sys < bctl->sys.limit_min)) {
3756 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3760 if (!chunk_reserved) {
3762 * We may be relocating the only data chunk we have,
3763 * which could potentially end up with losing data's
3764 * raid profile, so lets allocate an empty one in
3767 ret = btrfs_may_alloc_data_chunk(fs_info,
3770 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3772 } else if (ret == 1) {
3777 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3778 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3779 if (ret == -ENOSPC) {
3781 } else if (ret == -ETXTBSY) {
3783 "skipping relocation of block group %llu due to active swapfile",
3789 spin_lock(&fs_info->balance_lock);
3790 bctl->stat.completed++;
3791 spin_unlock(&fs_info->balance_lock);
3794 if (found_key.offset == 0)
3796 key.offset = found_key.offset - 1;
3800 btrfs_release_path(path);
3805 btrfs_free_path(path);
3806 if (enospc_errors) {
3807 btrfs_info(fs_info, "%d enospc errors during balance",
3817 * alloc_profile_is_valid - see if a given profile is valid and reduced
3818 * @flags: profile to validate
3819 * @extended: if true @flags is treated as an extended profile
3821 static int alloc_profile_is_valid(u64 flags, int extended)
3823 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3824 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3826 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3828 /* 1) check that all other bits are zeroed */
3832 /* 2) see if profile is reduced */
3834 return !extended; /* "0" is valid for usual profiles */
3836 return has_single_bit_set(flags);
3839 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3841 /* cancel requested || normal exit path */
3842 return atomic_read(&fs_info->balance_cancel_req) ||
3843 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3844 atomic_read(&fs_info->balance_cancel_req) == 0);
3848 * Validate target profile against allowed profiles and return true if it's OK.
3849 * Otherwise print the error message and return false.
3851 static inline int validate_convert_profile(struct btrfs_fs_info *fs_info,
3852 const struct btrfs_balance_args *bargs,
3853 u64 allowed, const char *type)
3855 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3858 /* Profile is valid and does not have bits outside of the allowed set */
3859 if (alloc_profile_is_valid(bargs->target, 1) &&
3860 (bargs->target & ~allowed) == 0)
3863 btrfs_err(fs_info, "balance: invalid convert %s profile %s",
3864 type, btrfs_bg_type_to_raid_name(bargs->target));
3869 * Fill @buf with textual description of balance filter flags @bargs, up to
3870 * @size_buf including the terminating null. The output may be trimmed if it
3871 * does not fit into the provided buffer.
3873 static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf,
3877 u32 size_bp = size_buf;
3879 u64 flags = bargs->flags;
3880 char tmp_buf[128] = {'\0'};
3885 #define CHECK_APPEND_NOARG(a) \
3887 ret = snprintf(bp, size_bp, (a)); \
3888 if (ret < 0 || ret >= size_bp) \
3889 goto out_overflow; \
3894 #define CHECK_APPEND_1ARG(a, v1) \
3896 ret = snprintf(bp, size_bp, (a), (v1)); \
3897 if (ret < 0 || ret >= size_bp) \
3898 goto out_overflow; \
3903 #define CHECK_APPEND_2ARG(a, v1, v2) \
3905 ret = snprintf(bp, size_bp, (a), (v1), (v2)); \
3906 if (ret < 0 || ret >= size_bp) \
3907 goto out_overflow; \
3912 if (flags & BTRFS_BALANCE_ARGS_CONVERT)
3913 CHECK_APPEND_1ARG("convert=%s,",
3914 btrfs_bg_type_to_raid_name(bargs->target));
3916 if (flags & BTRFS_BALANCE_ARGS_SOFT)
3917 CHECK_APPEND_NOARG("soft,");
3919 if (flags & BTRFS_BALANCE_ARGS_PROFILES) {
3920 btrfs_describe_block_groups(bargs->profiles, tmp_buf,
3922 CHECK_APPEND_1ARG("profiles=%s,", tmp_buf);
3925 if (flags & BTRFS_BALANCE_ARGS_USAGE)
3926 CHECK_APPEND_1ARG("usage=%llu,", bargs->usage);
3928 if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE)
3929 CHECK_APPEND_2ARG("usage=%u..%u,",
3930 bargs->usage_min, bargs->usage_max);
3932 if (flags & BTRFS_BALANCE_ARGS_DEVID)
3933 CHECK_APPEND_1ARG("devid=%llu,", bargs->devid);
3935 if (flags & BTRFS_BALANCE_ARGS_DRANGE)
3936 CHECK_APPEND_2ARG("drange=%llu..%llu,",
3937 bargs->pstart, bargs->pend);
3939 if (flags & BTRFS_BALANCE_ARGS_VRANGE)
3940 CHECK_APPEND_2ARG("vrange=%llu..%llu,",
3941 bargs->vstart, bargs->vend);
3943 if (flags & BTRFS_BALANCE_ARGS_LIMIT)
3944 CHECK_APPEND_1ARG("limit=%llu,", bargs->limit);
3946 if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)
3947 CHECK_APPEND_2ARG("limit=%u..%u,",
3948 bargs->limit_min, bargs->limit_max);
3950 if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE)
3951 CHECK_APPEND_2ARG("stripes=%u..%u,",
3952 bargs->stripes_min, bargs->stripes_max);
3954 #undef CHECK_APPEND_2ARG
3955 #undef CHECK_APPEND_1ARG
3956 #undef CHECK_APPEND_NOARG
3960 if (size_bp < size_buf)
3961 buf[size_buf - size_bp - 1] = '\0'; /* remove last , */
3966 static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info)
3968 u32 size_buf = 1024;
3969 char tmp_buf[192] = {'\0'};
3972 u32 size_bp = size_buf;
3974 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3976 buf = kzalloc(size_buf, GFP_KERNEL);
3982 #define CHECK_APPEND_1ARG(a, v1) \
3984 ret = snprintf(bp, size_bp, (a), (v1)); \
3985 if (ret < 0 || ret >= size_bp) \
3986 goto out_overflow; \
3991 if (bctl->flags & BTRFS_BALANCE_FORCE)
3992 CHECK_APPEND_1ARG("%s", "-f ");
3994 if (bctl->flags & BTRFS_BALANCE_DATA) {
3995 describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf));
3996 CHECK_APPEND_1ARG("-d%s ", tmp_buf);
3999 if (bctl->flags & BTRFS_BALANCE_METADATA) {
4000 describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf));
4001 CHECK_APPEND_1ARG("-m%s ", tmp_buf);
4004 if (bctl->flags & BTRFS_BALANCE_SYSTEM) {
4005 describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf));
4006 CHECK_APPEND_1ARG("-s%s ", tmp_buf);
4009 #undef CHECK_APPEND_1ARG
4013 if (size_bp < size_buf)
4014 buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */
4015 btrfs_info(fs_info, "balance: %s %s",
4016 (bctl->flags & BTRFS_BALANCE_RESUME) ?
4017 "resume" : "start", buf);
4023 * Should be called with balance mutexe held
4025 int btrfs_balance(struct btrfs_fs_info *fs_info,
4026 struct btrfs_balance_control *bctl,
4027 struct btrfs_ioctl_balance_args *bargs)
4029 u64 meta_target, data_target;
4035 bool reducing_redundancy;
4038 if (btrfs_fs_closing(fs_info) ||
4039 atomic_read(&fs_info->balance_pause_req) ||
4040 btrfs_should_cancel_balance(fs_info)) {
4045 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
4046 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
4050 * In case of mixed groups both data and meta should be picked,
4051 * and identical options should be given for both of them.
4053 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
4054 if (mixed && (bctl->flags & allowed)) {
4055 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
4056 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
4057 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
4059 "balance: mixed groups data and metadata options must be the same");
4066 * rw_devices will not change at the moment, device add/delete/replace
4069 num_devices = fs_info->fs_devices->rw_devices;
4072 * SINGLE profile on-disk has no profile bit, but in-memory we have a
4073 * special bit for it, to make it easier to distinguish. Thus we need
4074 * to set it manually, or balance would refuse the profile.
4076 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
4077 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++)
4078 if (num_devices >= btrfs_raid_array[i].devs_min)
4079 allowed |= btrfs_raid_array[i].bg_flag;
4081 if (!validate_convert_profile(fs_info, &bctl->data, allowed, "data") ||
4082 !validate_convert_profile(fs_info, &bctl->meta, allowed, "metadata") ||
4083 !validate_convert_profile(fs_info, &bctl->sys, allowed, "system")) {
4089 * Allow to reduce metadata or system integrity only if force set for
4090 * profiles with redundancy (copies, parity)
4093 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) {
4094 if (btrfs_raid_array[i].ncopies >= 2 ||
4095 btrfs_raid_array[i].tolerated_failures >= 1)
4096 allowed |= btrfs_raid_array[i].bg_flag;
4099 seq = read_seqbegin(&fs_info->profiles_lock);
4101 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4102 (fs_info->avail_system_alloc_bits & allowed) &&
4103 !(bctl->sys.target & allowed)) ||
4104 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4105 (fs_info->avail_metadata_alloc_bits & allowed) &&
4106 !(bctl->meta.target & allowed)))
4107 reducing_redundancy = true;
4109 reducing_redundancy = false;
4111 /* if we're not converting, the target field is uninitialized */
4112 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4113 bctl->meta.target : fs_info->avail_metadata_alloc_bits;
4114 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4115 bctl->data.target : fs_info->avail_data_alloc_bits;
4116 } while (read_seqretry(&fs_info->profiles_lock, seq));
4118 if (reducing_redundancy) {
4119 if (bctl->flags & BTRFS_BALANCE_FORCE) {
4121 "balance: force reducing metadata redundancy");
4124 "balance: reduces metadata redundancy, use --force if you want this");
4130 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <
4131 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {
4133 "balance: metadata profile %s has lower redundancy than data profile %s",
4134 btrfs_bg_type_to_raid_name(meta_target),
4135 btrfs_bg_type_to_raid_name(data_target));
4138 if (fs_info->send_in_progress) {
4139 btrfs_warn_rl(fs_info,
4140 "cannot run balance while send operations are in progress (%d in progress)",
4141 fs_info->send_in_progress);
4146 ret = insert_balance_item(fs_info, bctl);
4147 if (ret && ret != -EEXIST)
4150 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
4151 BUG_ON(ret == -EEXIST);
4152 BUG_ON(fs_info->balance_ctl);
4153 spin_lock(&fs_info->balance_lock);
4154 fs_info->balance_ctl = bctl;
4155 spin_unlock(&fs_info->balance_lock);
4157 BUG_ON(ret != -EEXIST);
4158 spin_lock(&fs_info->balance_lock);
4159 update_balance_args(bctl);
4160 spin_unlock(&fs_info->balance_lock);
4163 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4164 set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4165 describe_balance_start_or_resume(fs_info);
4166 mutex_unlock(&fs_info->balance_mutex);
4168 ret = __btrfs_balance(fs_info);
4170 mutex_lock(&fs_info->balance_mutex);
4171 if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req))
4172 btrfs_info(fs_info, "balance: paused");
4174 * Balance can be canceled by:
4176 * - Regular cancel request
4177 * Then ret == -ECANCELED and balance_cancel_req > 0
4179 * - Fatal signal to "btrfs" process
4180 * Either the signal caught by wait_reserve_ticket() and callers
4181 * got -EINTR, or caught by btrfs_should_cancel_balance() and
4183 * Either way, in this case balance_cancel_req = 0, and
4184 * ret == -EINTR or ret == -ECANCELED.
4186 * So here we only check the return value to catch canceled balance.
4188 else if (ret == -ECANCELED || ret == -EINTR)
4189 btrfs_info(fs_info, "balance: canceled");
4191 btrfs_info(fs_info, "balance: ended with status: %d", ret);
4193 clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4196 memset(bargs, 0, sizeof(*bargs));
4197 btrfs_update_ioctl_balance_args(fs_info, bargs);
4200 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
4201 balance_need_close(fs_info)) {
4202 reset_balance_state(fs_info);
4203 btrfs_exclop_finish(fs_info);
4206 wake_up(&fs_info->balance_wait_q);
4210 if (bctl->flags & BTRFS_BALANCE_RESUME)
4211 reset_balance_state(fs_info);
4214 btrfs_exclop_finish(fs_info);
4219 static int balance_kthread(void *data)
4221 struct btrfs_fs_info *fs_info = data;
4224 mutex_lock(&fs_info->balance_mutex);
4225 if (fs_info->balance_ctl)
4226 ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL);
4227 mutex_unlock(&fs_info->balance_mutex);
4232 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
4234 struct task_struct *tsk;
4236 mutex_lock(&fs_info->balance_mutex);
4237 if (!fs_info->balance_ctl) {
4238 mutex_unlock(&fs_info->balance_mutex);
4241 mutex_unlock(&fs_info->balance_mutex);
4243 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
4244 btrfs_info(fs_info, "balance: resume skipped");
4249 * A ro->rw remount sequence should continue with the paused balance
4250 * regardless of who pauses it, system or the user as of now, so set
4253 spin_lock(&fs_info->balance_lock);
4254 fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME;
4255 spin_unlock(&fs_info->balance_lock);
4257 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
4258 return PTR_ERR_OR_ZERO(tsk);
4261 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
4263 struct btrfs_balance_control *bctl;
4264 struct btrfs_balance_item *item;
4265 struct btrfs_disk_balance_args disk_bargs;
4266 struct btrfs_path *path;
4267 struct extent_buffer *leaf;
4268 struct btrfs_key key;
4271 path = btrfs_alloc_path();
4275 key.objectid = BTRFS_BALANCE_OBJECTID;
4276 key.type = BTRFS_TEMPORARY_ITEM_KEY;
4279 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4282 if (ret > 0) { /* ret = -ENOENT; */
4287 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4293 leaf = path->nodes[0];
4294 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
4296 bctl->flags = btrfs_balance_flags(leaf, item);
4297 bctl->flags |= BTRFS_BALANCE_RESUME;
4299 btrfs_balance_data(leaf, item, &disk_bargs);
4300 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4301 btrfs_balance_meta(leaf, item, &disk_bargs);
4302 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4303 btrfs_balance_sys(leaf, item, &disk_bargs);
4304 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4307 * This should never happen, as the paused balance state is recovered
4308 * during mount without any chance of other exclusive ops to collide.
4310 * This gives the exclusive op status to balance and keeps in paused
4311 * state until user intervention (cancel or umount). If the ownership
4312 * cannot be assigned, show a message but do not fail. The balance
4313 * is in a paused state and must have fs_info::balance_ctl properly
4316 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE))
4318 "balance: cannot set exclusive op status, resume manually");
4320 btrfs_release_path(path);
4322 mutex_lock(&fs_info->balance_mutex);
4323 BUG_ON(fs_info->balance_ctl);
4324 spin_lock(&fs_info->balance_lock);
4325 fs_info->balance_ctl = bctl;
4326 spin_unlock(&fs_info->balance_lock);
4327 mutex_unlock(&fs_info->balance_mutex);
4329 btrfs_free_path(path);
4333 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4337 mutex_lock(&fs_info->balance_mutex);
4338 if (!fs_info->balance_ctl) {
4339 mutex_unlock(&fs_info->balance_mutex);
4343 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4344 atomic_inc(&fs_info->balance_pause_req);
4345 mutex_unlock(&fs_info->balance_mutex);
4347 wait_event(fs_info->balance_wait_q,
4348 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4350 mutex_lock(&fs_info->balance_mutex);
4351 /* we are good with balance_ctl ripped off from under us */
4352 BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4353 atomic_dec(&fs_info->balance_pause_req);
4358 mutex_unlock(&fs_info->balance_mutex);
4362 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4364 mutex_lock(&fs_info->balance_mutex);
4365 if (!fs_info->balance_ctl) {
4366 mutex_unlock(&fs_info->balance_mutex);
4371 * A paused balance with the item stored on disk can be resumed at
4372 * mount time if the mount is read-write. Otherwise it's still paused
4373 * and we must not allow cancelling as it deletes the item.
4375 if (sb_rdonly(fs_info->sb)) {
4376 mutex_unlock(&fs_info->balance_mutex);
4380 atomic_inc(&fs_info->balance_cancel_req);
4382 * if we are running just wait and return, balance item is
4383 * deleted in btrfs_balance in this case
4385 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4386 mutex_unlock(&fs_info->balance_mutex);
4387 wait_event(fs_info->balance_wait_q,
4388 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4389 mutex_lock(&fs_info->balance_mutex);
4391 mutex_unlock(&fs_info->balance_mutex);
4393 * Lock released to allow other waiters to continue, we'll
4394 * reexamine the status again.
4396 mutex_lock(&fs_info->balance_mutex);
4398 if (fs_info->balance_ctl) {
4399 reset_balance_state(fs_info);
4400 btrfs_exclop_finish(fs_info);
4401 btrfs_info(fs_info, "balance: canceled");
4405 BUG_ON(fs_info->balance_ctl ||
4406 test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4407 atomic_dec(&fs_info->balance_cancel_req);
4408 mutex_unlock(&fs_info->balance_mutex);
4412 int btrfs_uuid_scan_kthread(void *data)
4414 struct btrfs_fs_info *fs_info = data;
4415 struct btrfs_root *root = fs_info->tree_root;
4416 struct btrfs_key key;
4417 struct btrfs_path *path = NULL;
4419 struct extent_buffer *eb;
4421 struct btrfs_root_item root_item;
4423 struct btrfs_trans_handle *trans = NULL;
4424 bool closing = false;
4426 path = btrfs_alloc_path();
4433 key.type = BTRFS_ROOT_ITEM_KEY;
4437 if (btrfs_fs_closing(fs_info)) {
4441 ret = btrfs_search_forward(root, &key, path,
4442 BTRFS_OLDEST_GENERATION);
4449 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4450 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4451 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4452 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4455 eb = path->nodes[0];
4456 slot = path->slots[0];
4457 item_size = btrfs_item_size_nr(eb, slot);
4458 if (item_size < sizeof(root_item))
4461 read_extent_buffer(eb, &root_item,
4462 btrfs_item_ptr_offset(eb, slot),
4463 (int)sizeof(root_item));
4464 if (btrfs_root_refs(&root_item) == 0)
4467 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4468 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4472 btrfs_release_path(path);
4474 * 1 - subvol uuid item
4475 * 1 - received_subvol uuid item
4477 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4478 if (IS_ERR(trans)) {
4479 ret = PTR_ERR(trans);
4487 btrfs_release_path(path);
4488 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4489 ret = btrfs_uuid_tree_add(trans, root_item.uuid,
4490 BTRFS_UUID_KEY_SUBVOL,
4493 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4499 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4500 ret = btrfs_uuid_tree_add(trans,
4501 root_item.received_uuid,
4502 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4505 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4512 btrfs_release_path(path);
4514 ret = btrfs_end_transaction(trans);
4520 if (key.offset < (u64)-1) {
4522 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4524 key.type = BTRFS_ROOT_ITEM_KEY;
4525 } else if (key.objectid < (u64)-1) {
4527 key.type = BTRFS_ROOT_ITEM_KEY;
4536 btrfs_free_path(path);
4537 if (trans && !IS_ERR(trans))
4538 btrfs_end_transaction(trans);
4540 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4542 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4543 up(&fs_info->uuid_tree_rescan_sem);
4547 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4549 struct btrfs_trans_handle *trans;
4550 struct btrfs_root *tree_root = fs_info->tree_root;
4551 struct btrfs_root *uuid_root;
4552 struct task_struct *task;
4559 trans = btrfs_start_transaction(tree_root, 2);
4561 return PTR_ERR(trans);
4563 uuid_root = btrfs_create_tree(trans, BTRFS_UUID_TREE_OBJECTID);
4564 if (IS_ERR(uuid_root)) {
4565 ret = PTR_ERR(uuid_root);
4566 btrfs_abort_transaction(trans, ret);
4567 btrfs_end_transaction(trans);
4571 fs_info->uuid_root = uuid_root;
4573 ret = btrfs_commit_transaction(trans);
4577 down(&fs_info->uuid_tree_rescan_sem);
4578 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4580 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4581 btrfs_warn(fs_info, "failed to start uuid_scan task");
4582 up(&fs_info->uuid_tree_rescan_sem);
4583 return PTR_ERR(task);
4590 * shrinking a device means finding all of the device extents past
4591 * the new size, and then following the back refs to the chunks.
4592 * The chunk relocation code actually frees the device extent
4594 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4596 struct btrfs_fs_info *fs_info = device->fs_info;
4597 struct btrfs_root *root = fs_info->dev_root;
4598 struct btrfs_trans_handle *trans;
4599 struct btrfs_dev_extent *dev_extent = NULL;
4600 struct btrfs_path *path;
4606 bool retried = false;
4607 struct extent_buffer *l;
4608 struct btrfs_key key;
4609 struct btrfs_super_block *super_copy = fs_info->super_copy;
4610 u64 old_total = btrfs_super_total_bytes(super_copy);
4611 u64 old_size = btrfs_device_get_total_bytes(device);
4615 new_size = round_down(new_size, fs_info->sectorsize);
4617 diff = round_down(old_size - new_size, fs_info->sectorsize);
4619 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4622 path = btrfs_alloc_path();
4626 path->reada = READA_BACK;
4628 trans = btrfs_start_transaction(root, 0);
4629 if (IS_ERR(trans)) {
4630 btrfs_free_path(path);
4631 return PTR_ERR(trans);
4634 mutex_lock(&fs_info->chunk_mutex);
4636 btrfs_device_set_total_bytes(device, new_size);
4637 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4638 device->fs_devices->total_rw_bytes -= diff;
4639 atomic64_sub(diff, &fs_info->free_chunk_space);
4643 * Once the device's size has been set to the new size, ensure all
4644 * in-memory chunks are synced to disk so that the loop below sees them
4645 * and relocates them accordingly.
4647 if (contains_pending_extent(device, &start, diff)) {
4648 mutex_unlock(&fs_info->chunk_mutex);
4649 ret = btrfs_commit_transaction(trans);
4653 mutex_unlock(&fs_info->chunk_mutex);
4654 btrfs_end_transaction(trans);
4658 key.objectid = device->devid;
4659 key.offset = (u64)-1;
4660 key.type = BTRFS_DEV_EXTENT_KEY;
4663 mutex_lock(&fs_info->delete_unused_bgs_mutex);
4664 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4666 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4670 ret = btrfs_previous_item(root, path, 0, key.type);
4672 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;