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>
17 #include <linux/namei.h>
20 #include "extent_map.h"
22 #include "transaction.h"
23 #include "print-tree.h"
26 #include "async-thread.h"
27 #include "check-integrity.h"
28 #include "rcu-string.h"
29 #include "dev-replace.h"
31 #include "tree-checker.h"
32 #include "space-info.h"
33 #include "block-group.h"
37 #define BTRFS_BLOCK_GROUP_STRIPE_MASK (BTRFS_BLOCK_GROUP_RAID0 | \
38 BTRFS_BLOCK_GROUP_RAID10 | \
39 BTRFS_BLOCK_GROUP_RAID56_MASK)
41 const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
42 [BTRFS_RAID_RAID10] = {
45 .devs_max = 0, /* 0 == as many as possible */
47 .tolerated_failures = 1,
51 .raid_name = "raid10",
52 .bg_flag = BTRFS_BLOCK_GROUP_RAID10,
53 .mindev_error = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET,
55 [BTRFS_RAID_RAID1] = {
60 .tolerated_failures = 1,
65 .bg_flag = BTRFS_BLOCK_GROUP_RAID1,
66 .mindev_error = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET,
68 [BTRFS_RAID_RAID1C3] = {
73 .tolerated_failures = 2,
77 .raid_name = "raid1c3",
78 .bg_flag = BTRFS_BLOCK_GROUP_RAID1C3,
79 .mindev_error = BTRFS_ERROR_DEV_RAID1C3_MIN_NOT_MET,
81 [BTRFS_RAID_RAID1C4] = {
86 .tolerated_failures = 3,
90 .raid_name = "raid1c4",
91 .bg_flag = BTRFS_BLOCK_GROUP_RAID1C4,
92 .mindev_error = BTRFS_ERROR_DEV_RAID1C4_MIN_NOT_MET,
99 .tolerated_failures = 0,
104 .bg_flag = BTRFS_BLOCK_GROUP_DUP,
107 [BTRFS_RAID_RAID0] = {
112 .tolerated_failures = 0,
116 .raid_name = "raid0",
117 .bg_flag = BTRFS_BLOCK_GROUP_RAID0,
120 [BTRFS_RAID_SINGLE] = {
125 .tolerated_failures = 0,
129 .raid_name = "single",
133 [BTRFS_RAID_RAID5] = {
138 .tolerated_failures = 1,
142 .raid_name = "raid5",
143 .bg_flag = BTRFS_BLOCK_GROUP_RAID5,
144 .mindev_error = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET,
146 [BTRFS_RAID_RAID6] = {
151 .tolerated_failures = 2,
155 .raid_name = "raid6",
156 .bg_flag = BTRFS_BLOCK_GROUP_RAID6,
157 .mindev_error = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET,
162 * Convert block group flags (BTRFS_BLOCK_GROUP_*) to btrfs_raid_types, which
163 * can be used as index to access btrfs_raid_array[].
165 enum btrfs_raid_types __attribute_const__ btrfs_bg_flags_to_raid_index(u64 flags)
167 if (flags & BTRFS_BLOCK_GROUP_RAID10)
168 return BTRFS_RAID_RAID10;
169 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
170 return BTRFS_RAID_RAID1;
171 else if (flags & BTRFS_BLOCK_GROUP_RAID1C3)
172 return BTRFS_RAID_RAID1C3;
173 else if (flags & BTRFS_BLOCK_GROUP_RAID1C4)
174 return BTRFS_RAID_RAID1C4;
175 else if (flags & BTRFS_BLOCK_GROUP_DUP)
176 return BTRFS_RAID_DUP;
177 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
178 return BTRFS_RAID_RAID0;
179 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
180 return BTRFS_RAID_RAID5;
181 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
182 return BTRFS_RAID_RAID6;
184 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
187 const char *btrfs_bg_type_to_raid_name(u64 flags)
189 const int index = btrfs_bg_flags_to_raid_index(flags);
191 if (index >= BTRFS_NR_RAID_TYPES)
194 return btrfs_raid_array[index].raid_name;
198 * Fill @buf with textual description of @bg_flags, no more than @size_buf
199 * bytes including terminating null byte.
201 void btrfs_describe_block_groups(u64 bg_flags, char *buf, u32 size_buf)
206 u64 flags = bg_flags;
207 u32 size_bp = size_buf;
214 #define DESCRIBE_FLAG(flag, desc) \
216 if (flags & (flag)) { \
217 ret = snprintf(bp, size_bp, "%s|", (desc)); \
218 if (ret < 0 || ret >= size_bp) \
226 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_DATA, "data");
227 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_SYSTEM, "system");
228 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_METADATA, "metadata");
230 DESCRIBE_FLAG(BTRFS_AVAIL_ALLOC_BIT_SINGLE, "single");
231 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
232 DESCRIBE_FLAG(btrfs_raid_array[i].bg_flag,
233 btrfs_raid_array[i].raid_name);
237 ret = snprintf(bp, size_bp, "0x%llx|", flags);
241 if (size_bp < size_buf)
242 buf[size_buf - size_bp - 1] = '\0'; /* remove last | */
245 * The text is trimmed, it's up to the caller to provide sufficiently
251 static int init_first_rw_device(struct btrfs_trans_handle *trans);
252 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info);
253 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
254 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
255 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
256 enum btrfs_map_op op,
257 u64 logical, u64 *length,
258 struct btrfs_io_context **bioc_ret,
259 int mirror_num, int need_raid_map);
265 * There are several mutexes that protect manipulation of devices and low-level
266 * structures like chunks but not block groups, extents or files
268 * uuid_mutex (global lock)
269 * ------------------------
270 * protects the fs_uuids list that tracks all per-fs fs_devices, resulting from
271 * the SCAN_DEV ioctl registration or from mount either implicitly (the first
272 * device) or requested by the device= mount option
274 * the mutex can be very coarse and can cover long-running operations
276 * protects: updates to fs_devices counters like missing devices, rw devices,
277 * seeding, structure cloning, opening/closing devices at mount/umount time
279 * global::fs_devs - add, remove, updates to the global list
281 * does not protect: manipulation of the fs_devices::devices list in general
282 * but in mount context it could be used to exclude list modifications by eg.
285 * btrfs_device::name - renames (write side), read is RCU
287 * fs_devices::device_list_mutex (per-fs, with RCU)
288 * ------------------------------------------------
289 * protects updates to fs_devices::devices, ie. adding and deleting
291 * simple list traversal with read-only actions can be done with RCU protection
293 * may be used to exclude some operations from running concurrently without any
294 * modifications to the list (see write_all_supers)
296 * Is not required at mount and close times, because our device list is
297 * protected by the uuid_mutex at that point.
301 * protects balance structures (status, state) and context accessed from
302 * several places (internally, ioctl)
306 * protects chunks, adding or removing during allocation, trim or when a new
307 * device is added/removed. Additionally it also protects post_commit_list of
308 * individual devices, since they can be added to the transaction's
309 * post_commit_list only with chunk_mutex held.
313 * a big lock that is held by the cleaner thread and prevents running subvolume
314 * cleaning together with relocation or delayed iputs
326 * Exclusive operations
327 * ====================
329 * Maintains the exclusivity of the following operations that apply to the
330 * whole filesystem and cannot run in parallel.
335 * - Device replace (*)
338 * The device operations (as above) can be in one of the following states:
344 * Only device operations marked with (*) can go into the Paused state for the
347 * - ioctl (only Balance can be Paused through ioctl)
348 * - filesystem remounted as read-only
349 * - filesystem unmounted and mounted as read-only
350 * - system power-cycle and filesystem mounted as read-only
351 * - filesystem or device errors leading to forced read-only
353 * The status of exclusive operation is set and cleared atomically.
354 * During the course of Paused state, fs_info::exclusive_operation remains set.
355 * A device operation in Paused or Running state can be canceled or resumed
356 * either by ioctl (Balance only) or when remounted as read-write.
357 * The exclusive status is cleared when the device operation is canceled or
361 DEFINE_MUTEX(uuid_mutex);
362 static LIST_HEAD(fs_uuids);
363 struct list_head * __attribute_const__ btrfs_get_fs_uuids(void)
369 * alloc_fs_devices - allocate struct btrfs_fs_devices
370 * @fsid: if not NULL, copy the UUID to fs_devices::fsid
371 * @metadata_fsid: if not NULL, copy the UUID to fs_devices::metadata_fsid
373 * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR().
374 * The returned struct is not linked onto any lists and can be destroyed with
375 * kfree() right away.
377 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid,
378 const u8 *metadata_fsid)
380 struct btrfs_fs_devices *fs_devs;
382 fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL);
384 return ERR_PTR(-ENOMEM);
386 mutex_init(&fs_devs->device_list_mutex);
388 INIT_LIST_HEAD(&fs_devs->devices);
389 INIT_LIST_HEAD(&fs_devs->alloc_list);
390 INIT_LIST_HEAD(&fs_devs->fs_list);
391 INIT_LIST_HEAD(&fs_devs->seed_list);
393 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
396 memcpy(fs_devs->metadata_uuid, metadata_fsid, BTRFS_FSID_SIZE);
398 memcpy(fs_devs->metadata_uuid, fsid, BTRFS_FSID_SIZE);
403 void btrfs_free_device(struct btrfs_device *device)
405 WARN_ON(!list_empty(&device->post_commit_list));
406 rcu_string_free(device->name);
407 extent_io_tree_release(&device->alloc_state);
408 bio_put(device->flush_bio);
409 btrfs_destroy_dev_zone_info(device);
413 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
415 struct btrfs_device *device;
416 WARN_ON(fs_devices->opened);
417 while (!list_empty(&fs_devices->devices)) {
418 device = list_entry(fs_devices->devices.next,
419 struct btrfs_device, dev_list);
420 list_del(&device->dev_list);
421 btrfs_free_device(device);
426 void __exit btrfs_cleanup_fs_uuids(void)
428 struct btrfs_fs_devices *fs_devices;
430 while (!list_empty(&fs_uuids)) {
431 fs_devices = list_entry(fs_uuids.next,
432 struct btrfs_fs_devices, fs_list);
433 list_del(&fs_devices->fs_list);
434 free_fs_devices(fs_devices);
438 static noinline struct btrfs_fs_devices *find_fsid(
439 const u8 *fsid, const u8 *metadata_fsid)
441 struct btrfs_fs_devices *fs_devices;
445 /* Handle non-split brain cases */
446 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
448 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0
449 && memcmp(metadata_fsid, fs_devices->metadata_uuid,
450 BTRFS_FSID_SIZE) == 0)
453 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
460 static struct btrfs_fs_devices *find_fsid_with_metadata_uuid(
461 struct btrfs_super_block *disk_super)
464 struct btrfs_fs_devices *fs_devices;
467 * Handle scanned device having completed its fsid change but
468 * belonging to a fs_devices that was created by first scanning
469 * a device which didn't have its fsid/metadata_uuid changed
470 * at all and the CHANGING_FSID_V2 flag set.
472 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
473 if (fs_devices->fsid_change &&
474 memcmp(disk_super->metadata_uuid, fs_devices->fsid,
475 BTRFS_FSID_SIZE) == 0 &&
476 memcmp(fs_devices->fsid, fs_devices->metadata_uuid,
477 BTRFS_FSID_SIZE) == 0) {
482 * Handle scanned device having completed its fsid change but
483 * belonging to a fs_devices that was created by a device that
484 * has an outdated pair of fsid/metadata_uuid and
485 * CHANGING_FSID_V2 flag set.
487 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
488 if (fs_devices->fsid_change &&
489 memcmp(fs_devices->metadata_uuid,
490 fs_devices->fsid, BTRFS_FSID_SIZE) != 0 &&
491 memcmp(disk_super->metadata_uuid, fs_devices->metadata_uuid,
492 BTRFS_FSID_SIZE) == 0) {
497 return find_fsid(disk_super->fsid, disk_super->metadata_uuid);
502 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
503 int flush, struct block_device **bdev,
504 struct btrfs_super_block **disk_super)
508 *bdev = blkdev_get_by_path(device_path, flags, holder);
511 ret = PTR_ERR(*bdev);
516 sync_blockdev(*bdev);
517 ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE);
519 blkdev_put(*bdev, flags);
522 invalidate_bdev(*bdev);
523 *disk_super = btrfs_read_dev_super(*bdev);
524 if (IS_ERR(*disk_super)) {
525 ret = PTR_ERR(*disk_super);
526 blkdev_put(*bdev, flags);
537 static bool device_path_matched(const char *path, struct btrfs_device *device)
542 found = strcmp(rcu_str_deref(device->name), path);
549 * Search and remove all stale (devices which are not mounted) devices.
550 * When both inputs are NULL, it will search and release all stale devices.
551 * path: Optional. When provided will it release all unmounted devices
552 * matching this path only.
553 * skip_dev: Optional. Will skip this device when searching for the stale
555 * Return: 0 for success or if @path is NULL.
556 * -EBUSY if @path is a mounted device.
557 * -ENOENT if @path does not match any device in the list.
559 static int btrfs_free_stale_devices(const char *path,
560 struct btrfs_device *skip_device)
562 struct btrfs_fs_devices *fs_devices, *tmp_fs_devices;
563 struct btrfs_device *device, *tmp_device;
566 lockdep_assert_held(&uuid_mutex);
571 list_for_each_entry_safe(fs_devices, tmp_fs_devices, &fs_uuids, fs_list) {
573 mutex_lock(&fs_devices->device_list_mutex);
574 list_for_each_entry_safe(device, tmp_device,
575 &fs_devices->devices, dev_list) {
576 if (skip_device && skip_device == device)
578 if (path && !device->name)
580 if (path && !device_path_matched(path, device))
582 if (fs_devices->opened) {
583 /* for an already deleted device return 0 */
584 if (path && ret != 0)
589 /* delete the stale device */
590 fs_devices->num_devices--;
591 list_del(&device->dev_list);
592 btrfs_free_device(device);
596 mutex_unlock(&fs_devices->device_list_mutex);
598 if (fs_devices->num_devices == 0) {
599 btrfs_sysfs_remove_fsid(fs_devices);
600 list_del(&fs_devices->fs_list);
601 free_fs_devices(fs_devices);
609 * This is only used on mount, and we are protected from competing things
610 * messing with our fs_devices by the uuid_mutex, thus we do not need the
611 * fs_devices->device_list_mutex here.
613 static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices,
614 struct btrfs_device *device, fmode_t flags,
617 struct request_queue *q;
618 struct block_device *bdev;
619 struct btrfs_super_block *disk_super;
628 ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
633 devid = btrfs_stack_device_id(&disk_super->dev_item);
634 if (devid != device->devid)
635 goto error_free_page;
637 if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE))
638 goto error_free_page;
640 device->generation = btrfs_super_generation(disk_super);
642 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
643 if (btrfs_super_incompat_flags(disk_super) &
644 BTRFS_FEATURE_INCOMPAT_METADATA_UUID) {
646 "BTRFS: Invalid seeding and uuid-changed device detected\n");
647 goto error_free_page;
650 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
651 fs_devices->seeding = true;
653 if (bdev_read_only(bdev))
654 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
656 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
659 q = bdev_get_queue(bdev);
660 if (!blk_queue_nonrot(q))
661 fs_devices->rotating = true;
664 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
665 device->mode = flags;
667 fs_devices->open_devices++;
668 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
669 device->devid != BTRFS_DEV_REPLACE_DEVID) {
670 fs_devices->rw_devices++;
671 list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list);
673 btrfs_release_disk_super(disk_super);
678 btrfs_release_disk_super(disk_super);
679 blkdev_put(bdev, flags);
685 * Handle scanned device having its CHANGING_FSID_V2 flag set and the fs_devices
686 * being created with a disk that has already completed its fsid change. Such
687 * disk can belong to an fs which has its FSID changed or to one which doesn't.
688 * Handle both cases here.
690 static struct btrfs_fs_devices *find_fsid_inprogress(
691 struct btrfs_super_block *disk_super)
693 struct btrfs_fs_devices *fs_devices;
695 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
696 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid,
697 BTRFS_FSID_SIZE) != 0 &&
698 memcmp(fs_devices->metadata_uuid, disk_super->fsid,
699 BTRFS_FSID_SIZE) == 0 && !fs_devices->fsid_change) {
704 return find_fsid(disk_super->fsid, NULL);
708 static struct btrfs_fs_devices *find_fsid_changed(
709 struct btrfs_super_block *disk_super)
711 struct btrfs_fs_devices *fs_devices;
714 * Handles the case where scanned device is part of an fs that had
715 * multiple successful changes of FSID but currently device didn't
716 * observe it. Meaning our fsid will be different than theirs. We need
717 * to handle two subcases :
718 * 1 - The fs still continues to have different METADATA/FSID uuids.
719 * 2 - The fs is switched back to its original FSID (METADATA/FSID
722 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
724 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid,
725 BTRFS_FSID_SIZE) != 0 &&
726 memcmp(fs_devices->metadata_uuid, disk_super->metadata_uuid,
727 BTRFS_FSID_SIZE) == 0 &&
728 memcmp(fs_devices->fsid, disk_super->fsid,
729 BTRFS_FSID_SIZE) != 0)
732 /* Unchanged UUIDs */
733 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid,
734 BTRFS_FSID_SIZE) == 0 &&
735 memcmp(fs_devices->fsid, disk_super->metadata_uuid,
736 BTRFS_FSID_SIZE) == 0)
743 static struct btrfs_fs_devices *find_fsid_reverted_metadata(
744 struct btrfs_super_block *disk_super)
746 struct btrfs_fs_devices *fs_devices;
749 * Handle the case where the scanned device is part of an fs whose last
750 * metadata UUID change reverted it to the original FSID. At the same
751 * time * fs_devices was first created by another constitutent device
752 * which didn't fully observe the operation. This results in an
753 * btrfs_fs_devices created with metadata/fsid different AND
754 * btrfs_fs_devices::fsid_change set AND the metadata_uuid of the
755 * fs_devices equal to the FSID of the disk.
757 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
758 if (memcmp(fs_devices->fsid, fs_devices->metadata_uuid,
759 BTRFS_FSID_SIZE) != 0 &&
760 memcmp(fs_devices->metadata_uuid, disk_super->fsid,
761 BTRFS_FSID_SIZE) == 0 &&
762 fs_devices->fsid_change)
769 * Add new device to list of registered devices
772 * device pointer which was just added or updated when successful
773 * error pointer when failed
775 static noinline struct btrfs_device *device_list_add(const char *path,
776 struct btrfs_super_block *disk_super,
777 bool *new_device_added)
779 struct btrfs_device *device;
780 struct btrfs_fs_devices *fs_devices = NULL;
781 struct rcu_string *name;
782 u64 found_transid = btrfs_super_generation(disk_super);
783 u64 devid = btrfs_stack_device_id(&disk_super->dev_item);
784 bool has_metadata_uuid = (btrfs_super_incompat_flags(disk_super) &
785 BTRFS_FEATURE_INCOMPAT_METADATA_UUID);
786 bool fsid_change_in_progress = (btrfs_super_flags(disk_super) &
787 BTRFS_SUPER_FLAG_CHANGING_FSID_V2);
789 if (fsid_change_in_progress) {
790 if (!has_metadata_uuid)
791 fs_devices = find_fsid_inprogress(disk_super);
793 fs_devices = find_fsid_changed(disk_super);
794 } else if (has_metadata_uuid) {
795 fs_devices = find_fsid_with_metadata_uuid(disk_super);
797 fs_devices = find_fsid_reverted_metadata(disk_super);
799 fs_devices = find_fsid(disk_super->fsid, NULL);
804 if (has_metadata_uuid)
805 fs_devices = alloc_fs_devices(disk_super->fsid,
806 disk_super->metadata_uuid);
808 fs_devices = alloc_fs_devices(disk_super->fsid, NULL);
810 if (IS_ERR(fs_devices))
811 return ERR_CAST(fs_devices);
813 fs_devices->fsid_change = fsid_change_in_progress;
815 mutex_lock(&fs_devices->device_list_mutex);
816 list_add(&fs_devices->fs_list, &fs_uuids);
820 struct btrfs_dev_lookup_args args = {
822 .uuid = disk_super->dev_item.uuid,
825 mutex_lock(&fs_devices->device_list_mutex);
826 device = btrfs_find_device(fs_devices, &args);
829 * If this disk has been pulled into an fs devices created by
830 * a device which had the CHANGING_FSID_V2 flag then replace the
831 * metadata_uuid/fsid values of the fs_devices.
833 if (fs_devices->fsid_change &&
834 found_transid > fs_devices->latest_generation) {
835 memcpy(fs_devices->fsid, disk_super->fsid,
838 if (has_metadata_uuid)
839 memcpy(fs_devices->metadata_uuid,
840 disk_super->metadata_uuid,
843 memcpy(fs_devices->metadata_uuid,
844 disk_super->fsid, BTRFS_FSID_SIZE);
846 fs_devices->fsid_change = false;
851 if (fs_devices->opened) {
852 mutex_unlock(&fs_devices->device_list_mutex);
853 return ERR_PTR(-EBUSY);
856 device = btrfs_alloc_device(NULL, &devid,
857 disk_super->dev_item.uuid);
858 if (IS_ERR(device)) {
859 mutex_unlock(&fs_devices->device_list_mutex);
860 /* we can safely leave the fs_devices entry around */
864 name = rcu_string_strdup(path, GFP_NOFS);
866 btrfs_free_device(device);
867 mutex_unlock(&fs_devices->device_list_mutex);
868 return ERR_PTR(-ENOMEM);
870 rcu_assign_pointer(device->name, name);
872 list_add_rcu(&device->dev_list, &fs_devices->devices);
873 fs_devices->num_devices++;
875 device->fs_devices = fs_devices;
876 *new_device_added = true;
878 if (disk_super->label[0])
880 "BTRFS: device label %s devid %llu transid %llu %s scanned by %s (%d)\n",
881 disk_super->label, devid, found_transid, path,
882 current->comm, task_pid_nr(current));
885 "BTRFS: device fsid %pU devid %llu transid %llu %s scanned by %s (%d)\n",
886 disk_super->fsid, devid, found_transid, path,
887 current->comm, task_pid_nr(current));
889 } else if (!device->name || strcmp(device->name->str, path)) {
891 * When FS is already mounted.
892 * 1. If you are here and if the device->name is NULL that
893 * means this device was missing at time of FS mount.
894 * 2. If you are here and if the device->name is different
895 * from 'path' that means either
896 * a. The same device disappeared and reappeared with
898 * b. The missing-disk-which-was-replaced, has
901 * We must allow 1 and 2a above. But 2b would be a spurious
904 * Further in case of 1 and 2a above, the disk at 'path'
905 * would have missed some transaction when it was away and
906 * in case of 2a the stale bdev has to be updated as well.
907 * 2b must not be allowed at all time.
911 * For now, we do allow update to btrfs_fs_device through the
912 * btrfs dev scan cli after FS has been mounted. We're still
913 * tracking a problem where systems fail mount by subvolume id
914 * when we reject replacement on a mounted FS.
916 if (!fs_devices->opened && found_transid < device->generation) {
918 * That is if the FS is _not_ mounted and if you
919 * are here, that means there is more than one
920 * disk with same uuid and devid.We keep the one
921 * with larger generation number or the last-in if
922 * generation are equal.
924 mutex_unlock(&fs_devices->device_list_mutex);
925 return ERR_PTR(-EEXIST);
929 * We are going to replace the device path for a given devid,
930 * make sure it's the same device if the device is mounted
936 error = lookup_bdev(path, &path_dev);
938 mutex_unlock(&fs_devices->device_list_mutex);
939 return ERR_PTR(error);
942 if (device->bdev->bd_dev != path_dev) {
943 mutex_unlock(&fs_devices->device_list_mutex);
945 * device->fs_info may not be reliable here, so
946 * pass in a NULL instead. This avoids a
947 * possible use-after-free when the fs_info and
948 * fs_info->sb are already torn down.
950 btrfs_warn_in_rcu(NULL,
951 "duplicate device %s devid %llu generation %llu scanned by %s (%d)",
952 path, devid, found_transid,
954 task_pid_nr(current));
955 return ERR_PTR(-EEXIST);
957 btrfs_info_in_rcu(device->fs_info,
958 "devid %llu device path %s changed to %s scanned by %s (%d)",
959 devid, rcu_str_deref(device->name),
961 task_pid_nr(current));
964 name = rcu_string_strdup(path, GFP_NOFS);
966 mutex_unlock(&fs_devices->device_list_mutex);
967 return ERR_PTR(-ENOMEM);
969 rcu_string_free(device->name);
970 rcu_assign_pointer(device->name, name);
971 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
972 fs_devices->missing_devices--;
973 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
978 * Unmount does not free the btrfs_device struct but would zero
979 * generation along with most of the other members. So just update
980 * it back. We need it to pick the disk with largest generation
983 if (!fs_devices->opened) {
984 device->generation = found_transid;
985 fs_devices->latest_generation = max_t(u64, found_transid,
986 fs_devices->latest_generation);
989 fs_devices->total_devices = btrfs_super_num_devices(disk_super);
991 mutex_unlock(&fs_devices->device_list_mutex);
995 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
997 struct btrfs_fs_devices *fs_devices;
998 struct btrfs_device *device;
999 struct btrfs_device *orig_dev;
1002 lockdep_assert_held(&uuid_mutex);
1004 fs_devices = alloc_fs_devices(orig->fsid, NULL);
1005 if (IS_ERR(fs_devices))
1008 fs_devices->total_devices = orig->total_devices;
1010 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
1011 struct rcu_string *name;
1013 device = btrfs_alloc_device(NULL, &orig_dev->devid,
1015 if (IS_ERR(device)) {
1016 ret = PTR_ERR(device);
1021 * This is ok to do without rcu read locked because we hold the
1022 * uuid mutex so nothing we touch in here is going to disappear.
1024 if (orig_dev->name) {
1025 name = rcu_string_strdup(orig_dev->name->str,
1028 btrfs_free_device(device);
1032 rcu_assign_pointer(device->name, name);
1035 list_add(&device->dev_list, &fs_devices->devices);
1036 device->fs_devices = fs_devices;
1037 fs_devices->num_devices++;
1041 free_fs_devices(fs_devices);
1042 return ERR_PTR(ret);
1045 static void __btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices,
1046 struct btrfs_device **latest_dev)
1048 struct btrfs_device *device, *next;
1050 /* This is the initialized path, it is safe to release the devices. */
1051 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
1052 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)) {
1053 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
1054 &device->dev_state) &&
1055 !test_bit(BTRFS_DEV_STATE_MISSING,
1056 &device->dev_state) &&
1058 device->generation > (*latest_dev)->generation)) {
1059 *latest_dev = device;
1065 * We have already validated the presence of BTRFS_DEV_REPLACE_DEVID,
1066 * in btrfs_init_dev_replace() so just continue.
1068 if (device->devid == BTRFS_DEV_REPLACE_DEVID)
1072 blkdev_put(device->bdev, device->mode);
1073 device->bdev = NULL;
1074 fs_devices->open_devices--;
1076 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1077 list_del_init(&device->dev_alloc_list);
1078 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
1079 fs_devices->rw_devices--;
1081 list_del_init(&device->dev_list);
1082 fs_devices->num_devices--;
1083 btrfs_free_device(device);
1089 * After we have read the system tree and know devids belonging to this
1090 * filesystem, remove the device which does not belong there.
1092 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices)
1094 struct btrfs_device *latest_dev = NULL;
1095 struct btrfs_fs_devices *seed_dev;
1097 mutex_lock(&uuid_mutex);
1098 __btrfs_free_extra_devids(fs_devices, &latest_dev);
1100 list_for_each_entry(seed_dev, &fs_devices->seed_list, seed_list)
1101 __btrfs_free_extra_devids(seed_dev, &latest_dev);
1103 fs_devices->latest_dev = latest_dev;
1105 mutex_unlock(&uuid_mutex);
1108 static void btrfs_close_bdev(struct btrfs_device *device)
1113 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1114 sync_blockdev(device->bdev);
1115 invalidate_bdev(device->bdev);
1118 blkdev_put(device->bdev, device->mode);
1121 static void btrfs_close_one_device(struct btrfs_device *device)
1123 struct btrfs_fs_devices *fs_devices = device->fs_devices;
1125 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
1126 device->devid != BTRFS_DEV_REPLACE_DEVID) {
1127 list_del_init(&device->dev_alloc_list);
1128 fs_devices->rw_devices--;
1131 if (device->devid == BTRFS_DEV_REPLACE_DEVID)
1132 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
1134 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
1135 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
1136 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);
1152 * Reset the flush error record. We might have a transient flush error
1153 * in this mount, and if so we aborted the current transaction and set
1154 * the fs to an error state, guaranteeing no super blocks can be further
1155 * committed. However that error might be transient and if we unmount the
1156 * filesystem and mount it again, we should allow the mount to succeed
1157 * (btrfs_check_rw_degradable() should not fail) - if after mounting the
1158 * filesystem again we still get flush errors, then we will again abort
1159 * any transaction and set the error state, guaranteeing no commits of
1160 * unsafe super blocks.
1162 device->last_flush_error = 0;
1164 /* Verify the device is back in a pristine state */
1165 ASSERT(!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state));
1166 ASSERT(!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state));
1167 ASSERT(list_empty(&device->dev_alloc_list));
1168 ASSERT(list_empty(&device->post_commit_list));
1171 static void close_fs_devices(struct btrfs_fs_devices *fs_devices)
1173 struct btrfs_device *device, *tmp;
1175 lockdep_assert_held(&uuid_mutex);
1177 if (--fs_devices->opened > 0)
1180 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list)
1181 btrfs_close_one_device(device);
1183 WARN_ON(fs_devices->open_devices);
1184 WARN_ON(fs_devices->rw_devices);
1185 fs_devices->opened = 0;
1186 fs_devices->seeding = false;
1187 fs_devices->fs_info = NULL;
1190 void btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
1193 struct btrfs_fs_devices *tmp;
1195 mutex_lock(&uuid_mutex);
1196 close_fs_devices(fs_devices);
1197 if (!fs_devices->opened)
1198 list_splice_init(&fs_devices->seed_list, &list);
1200 list_for_each_entry_safe(fs_devices, tmp, &list, seed_list) {
1201 close_fs_devices(fs_devices);
1202 list_del(&fs_devices->seed_list);
1203 free_fs_devices(fs_devices);
1205 mutex_unlock(&uuid_mutex);
1208 static int open_fs_devices(struct btrfs_fs_devices *fs_devices,
1209 fmode_t flags, void *holder)
1211 struct btrfs_device *device;
1212 struct btrfs_device *latest_dev = NULL;
1213 struct btrfs_device *tmp_device;
1215 flags |= FMODE_EXCL;
1217 list_for_each_entry_safe(device, tmp_device, &fs_devices->devices,
1221 ret = btrfs_open_one_device(fs_devices, device, flags, holder);
1223 (!latest_dev || device->generation > latest_dev->generation)) {
1224 latest_dev = device;
1225 } else if (ret == -ENODATA) {
1226 fs_devices->num_devices--;
1227 list_del(&device->dev_list);
1228 btrfs_free_device(device);
1231 if (fs_devices->open_devices == 0)
1234 fs_devices->opened = 1;
1235 fs_devices->latest_dev = latest_dev;
1236 fs_devices->total_rw_bytes = 0;
1237 fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_REGULAR;
1238 fs_devices->read_policy = BTRFS_READ_POLICY_PID;
1243 static int devid_cmp(void *priv, const struct list_head *a,
1244 const struct list_head *b)
1246 const struct btrfs_device *dev1, *dev2;
1248 dev1 = list_entry(a, struct btrfs_device, dev_list);
1249 dev2 = list_entry(b, struct btrfs_device, dev_list);
1251 if (dev1->devid < dev2->devid)
1253 else if (dev1->devid > dev2->devid)
1258 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1259 fmode_t flags, void *holder)
1263 lockdep_assert_held(&uuid_mutex);
1265 * The device_list_mutex cannot be taken here in case opening the
1266 * underlying device takes further locks like open_mutex.
1268 * We also don't need the lock here as this is called during mount and
1269 * exclusion is provided by uuid_mutex
1272 if (fs_devices->opened) {
1273 fs_devices->opened++;
1276 list_sort(NULL, &fs_devices->devices, devid_cmp);
1277 ret = open_fs_devices(fs_devices, flags, holder);
1283 void btrfs_release_disk_super(struct btrfs_super_block *super)
1285 struct page *page = virt_to_page(super);
1290 static struct btrfs_super_block *btrfs_read_disk_super(struct block_device *bdev,
1291 u64 bytenr, u64 bytenr_orig)
1293 struct btrfs_super_block *disk_super;
1298 /* make sure our super fits in the device */
1299 if (bytenr + PAGE_SIZE >= bdev_nr_bytes(bdev))
1300 return ERR_PTR(-EINVAL);
1302 /* make sure our super fits in the page */
1303 if (sizeof(*disk_super) > PAGE_SIZE)
1304 return ERR_PTR(-EINVAL);
1306 /* make sure our super doesn't straddle pages on disk */
1307 index = bytenr >> PAGE_SHIFT;
1308 if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_SHIFT != index)
1309 return ERR_PTR(-EINVAL);
1311 /* pull in the page with our super */
1312 page = read_cache_page_gfp(bdev->bd_inode->i_mapping, index, GFP_KERNEL);
1315 return ERR_CAST(page);
1317 p = page_address(page);
1319 /* align our pointer to the offset of the super block */
1320 disk_super = p + offset_in_page(bytenr);
1322 if (btrfs_super_bytenr(disk_super) != bytenr_orig ||
1323 btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
1324 btrfs_release_disk_super(p);
1325 return ERR_PTR(-EINVAL);
1328 if (disk_super->label[0] && disk_super->label[BTRFS_LABEL_SIZE - 1])
1329 disk_super->label[BTRFS_LABEL_SIZE - 1] = 0;
1334 int btrfs_forget_devices(const char *path)
1338 mutex_lock(&uuid_mutex);
1339 ret = btrfs_free_stale_devices(strlen(path) ? path : NULL, NULL);
1340 mutex_unlock(&uuid_mutex);
1346 * Look for a btrfs signature on a device. This may be called out of the mount path
1347 * and we are not allowed to call set_blocksize during the scan. The superblock
1348 * is read via pagecache
1350 struct btrfs_device *btrfs_scan_one_device(const char *path, fmode_t flags,
1353 struct btrfs_super_block *disk_super;
1354 bool new_device_added = false;
1355 struct btrfs_device *device = NULL;
1356 struct block_device *bdev;
1357 u64 bytenr, bytenr_orig;
1360 lockdep_assert_held(&uuid_mutex);
1363 * we would like to check all the supers, but that would make
1364 * a btrfs mount succeed after a mkfs from a different FS.
1365 * So, we need to add a special mount option to scan for
1366 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1368 flags |= FMODE_EXCL;
1370 bdev = blkdev_get_by_path(path, flags, holder);
1372 return ERR_CAST(bdev);
1374 bytenr_orig = btrfs_sb_offset(0);
1375 ret = btrfs_sb_log_location_bdev(bdev, 0, READ, &bytenr);
1377 device = ERR_PTR(ret);
1378 goto error_bdev_put;
1381 disk_super = btrfs_read_disk_super(bdev, bytenr, bytenr_orig);
1382 if (IS_ERR(disk_super)) {
1383 device = ERR_CAST(disk_super);
1384 goto error_bdev_put;
1387 device = device_list_add(path, disk_super, &new_device_added);
1388 if (!IS_ERR(device)) {
1389 if (new_device_added)
1390 btrfs_free_stale_devices(path, device);
1393 btrfs_release_disk_super(disk_super);
1396 blkdev_put(bdev, flags);
1402 * Try to find a chunk that intersects [start, start + len] range and when one
1403 * such is found, record the end of it in *start
1405 static bool contains_pending_extent(struct btrfs_device *device, u64 *start,
1408 u64 physical_start, physical_end;
1410 lockdep_assert_held(&device->fs_info->chunk_mutex);
1412 if (!find_first_extent_bit(&device->alloc_state, *start,
1413 &physical_start, &physical_end,
1414 CHUNK_ALLOCATED, NULL)) {
1416 if (in_range(physical_start, *start, len) ||
1417 in_range(*start, physical_start,
1418 physical_end - physical_start)) {
1419 *start = physical_end + 1;
1426 static u64 dev_extent_search_start(struct btrfs_device *device, u64 start)
1428 switch (device->fs_devices->chunk_alloc_policy) {
1429 case BTRFS_CHUNK_ALLOC_REGULAR:
1431 * We don't want to overwrite the superblock on the drive nor
1432 * any area used by the boot loader (grub for example), so we
1433 * make sure to start at an offset of at least 1MB.
1435 return max_t(u64, start, SZ_1M);
1436 case BTRFS_CHUNK_ALLOC_ZONED:
1438 * We don't care about the starting region like regular
1439 * allocator, because we anyway use/reserve the first two zones
1440 * for superblock logging.
1442 return ALIGN(start, device->zone_info->zone_size);
1448 static bool dev_extent_hole_check_zoned(struct btrfs_device *device,
1449 u64 *hole_start, u64 *hole_size,
1452 u64 zone_size = device->zone_info->zone_size;
1455 bool changed = false;
1457 ASSERT(IS_ALIGNED(*hole_start, zone_size));
1459 while (*hole_size > 0) {
1460 pos = btrfs_find_allocatable_zones(device, *hole_start,
1461 *hole_start + *hole_size,
1463 if (pos != *hole_start) {
1464 *hole_size = *hole_start + *hole_size - pos;
1467 if (*hole_size < num_bytes)
1471 ret = btrfs_ensure_empty_zones(device, pos, num_bytes);
1473 /* Range is ensured to be empty */
1477 /* Given hole range was invalid (outside of device) */
1478 if (ret == -ERANGE) {
1479 *hole_start += *hole_size;
1484 *hole_start += zone_size;
1485 *hole_size -= zone_size;
1493 * dev_extent_hole_check - check if specified hole is suitable for allocation
1494 * @device: the device which we have the hole
1495 * @hole_start: starting position of the hole
1496 * @hole_size: the size of the hole
1497 * @num_bytes: the size of the free space that we need
1499 * This function may modify @hole_start and @hole_size to reflect the suitable
1500 * position for allocation. Returns 1 if hole position is updated, 0 otherwise.
1502 static bool dev_extent_hole_check(struct btrfs_device *device, u64 *hole_start,
1503 u64 *hole_size, u64 num_bytes)
1505 bool changed = false;
1506 u64 hole_end = *hole_start + *hole_size;
1510 * Check before we set max_hole_start, otherwise we could end up
1511 * sending back this offset anyway.
1513 if (contains_pending_extent(device, hole_start, *hole_size)) {
1514 if (hole_end >= *hole_start)
1515 *hole_size = hole_end - *hole_start;
1521 switch (device->fs_devices->chunk_alloc_policy) {
1522 case BTRFS_CHUNK_ALLOC_REGULAR:
1523 /* No extra check */
1525 case BTRFS_CHUNK_ALLOC_ZONED:
1526 if (dev_extent_hole_check_zoned(device, hole_start,
1527 hole_size, num_bytes)) {
1530 * The changed hole can contain pending extent.
1531 * Loop again to check that.
1547 * find_free_dev_extent_start - find free space in the specified device
1548 * @device: the device which we search the free space in
1549 * @num_bytes: the size of the free space that we need
1550 * @search_start: the position from which to begin the search
1551 * @start: store the start of the free space.
1552 * @len: the size of the free space. that we find, or the size
1553 * of the max free space if we don't find suitable free space
1555 * this uses a pretty simple search, the expectation is that it is
1556 * called very infrequently and that a given device has a small number
1559 * @start is used to store the start of the free space if we find. But if we
1560 * don't find suitable free space, it will be used to store the start position
1561 * of the max free space.
1563 * @len is used to store the size of the free space that we find.
1564 * But if we don't find suitable free space, it is used to store the size of
1565 * the max free space.
1567 * NOTE: This function will search *commit* root of device tree, and does extra
1568 * check to ensure dev extents are not double allocated.
1569 * This makes the function safe to allocate dev extents but may not report
1570 * correct usable device space, as device extent freed in current transaction
1571 * is not reported as available.
1573 static int find_free_dev_extent_start(struct btrfs_device *device,
1574 u64 num_bytes, u64 search_start, u64 *start,
1577 struct btrfs_fs_info *fs_info = device->fs_info;
1578 struct btrfs_root *root = fs_info->dev_root;
1579 struct btrfs_key key;
1580 struct btrfs_dev_extent *dev_extent;
1581 struct btrfs_path *path;
1586 u64 search_end = device->total_bytes;
1589 struct extent_buffer *l;
1591 search_start = dev_extent_search_start(device, search_start);
1593 WARN_ON(device->zone_info &&
1594 !IS_ALIGNED(num_bytes, device->zone_info->zone_size));
1596 path = btrfs_alloc_path();
1600 max_hole_start = search_start;
1604 if (search_start >= search_end ||
1605 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1610 path->reada = READA_FORWARD;
1611 path->search_commit_root = 1;
1612 path->skip_locking = 1;
1614 key.objectid = device->devid;
1615 key.offset = search_start;
1616 key.type = BTRFS_DEV_EXTENT_KEY;
1618 ret = btrfs_search_backwards(root, &key, path);
1624 slot = path->slots[0];
1625 if (slot >= btrfs_header_nritems(l)) {
1626 ret = btrfs_next_leaf(root, path);
1634 btrfs_item_key_to_cpu(l, &key, slot);
1636 if (key.objectid < device->devid)
1639 if (key.objectid > device->devid)
1642 if (key.type != BTRFS_DEV_EXTENT_KEY)
1645 if (key.offset > search_start) {
1646 hole_size = key.offset - search_start;
1647 dev_extent_hole_check(device, &search_start, &hole_size,
1650 if (hole_size > max_hole_size) {
1651 max_hole_start = search_start;
1652 max_hole_size = hole_size;
1656 * If this free space is greater than which we need,
1657 * it must be the max free space that we have found
1658 * until now, so max_hole_start must point to the start
1659 * of this free space and the length of this free space
1660 * is stored in max_hole_size. Thus, we return
1661 * max_hole_start and max_hole_size and go back to the
1664 if (hole_size >= num_bytes) {
1670 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1671 extent_end = key.offset + btrfs_dev_extent_length(l,
1673 if (extent_end > search_start)
1674 search_start = extent_end;
1681 * At this point, search_start should be the end of
1682 * allocated dev extents, and when shrinking the device,
1683 * search_end may be smaller than search_start.
1685 if (search_end > search_start) {
1686 hole_size = search_end - search_start;
1687 if (dev_extent_hole_check(device, &search_start, &hole_size,
1689 btrfs_release_path(path);
1693 if (hole_size > max_hole_size) {
1694 max_hole_start = search_start;
1695 max_hole_size = hole_size;
1700 if (max_hole_size < num_bytes)
1706 btrfs_free_path(path);
1707 *start = max_hole_start;
1709 *len = max_hole_size;
1713 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
1714 u64 *start, u64 *len)
1716 /* FIXME use last free of some kind */
1717 return find_free_dev_extent_start(device, num_bytes, 0, start, len);
1720 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1721 struct btrfs_device *device,
1722 u64 start, u64 *dev_extent_len)
1724 struct btrfs_fs_info *fs_info = device->fs_info;
1725 struct btrfs_root *root = fs_info->dev_root;
1727 struct btrfs_path *path;
1728 struct btrfs_key key;
1729 struct btrfs_key found_key;
1730 struct extent_buffer *leaf = NULL;
1731 struct btrfs_dev_extent *extent = NULL;
1733 path = btrfs_alloc_path();
1737 key.objectid = device->devid;
1739 key.type = BTRFS_DEV_EXTENT_KEY;
1741 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1743 ret = btrfs_previous_item(root, path, key.objectid,
1744 BTRFS_DEV_EXTENT_KEY);
1747 leaf = path->nodes[0];
1748 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1749 extent = btrfs_item_ptr(leaf, path->slots[0],
1750 struct btrfs_dev_extent);
1751 BUG_ON(found_key.offset > start || found_key.offset +
1752 btrfs_dev_extent_length(leaf, extent) < start);
1754 btrfs_release_path(path);
1756 } else if (ret == 0) {
1757 leaf = path->nodes[0];
1758 extent = btrfs_item_ptr(leaf, path->slots[0],
1759 struct btrfs_dev_extent);
1764 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1766 ret = btrfs_del_item(trans, root, path);
1768 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1770 btrfs_free_path(path);
1774 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1776 struct extent_map_tree *em_tree;
1777 struct extent_map *em;
1781 em_tree = &fs_info->mapping_tree;
1782 read_lock(&em_tree->lock);
1783 n = rb_last(&em_tree->map.rb_root);
1785 em = rb_entry(n, struct extent_map, rb_node);
1786 ret = em->start + em->len;
1788 read_unlock(&em_tree->lock);
1793 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1797 struct btrfs_key key;
1798 struct btrfs_key found_key;
1799 struct btrfs_path *path;
1801 path = btrfs_alloc_path();
1805 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1806 key.type = BTRFS_DEV_ITEM_KEY;
1807 key.offset = (u64)-1;
1809 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1815 btrfs_err(fs_info, "corrupted chunk tree devid -1 matched");
1820 ret = btrfs_previous_item(fs_info->chunk_root, path,
1821 BTRFS_DEV_ITEMS_OBJECTID,
1822 BTRFS_DEV_ITEM_KEY);
1826 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1828 *devid_ret = found_key.offset + 1;
1832 btrfs_free_path(path);
1837 * the device information is stored in the chunk root
1838 * the btrfs_device struct should be fully filled in
1840 static int btrfs_add_dev_item(struct btrfs_trans_handle *trans,
1841 struct btrfs_device *device)
1844 struct btrfs_path *path;
1845 struct btrfs_dev_item *dev_item;
1846 struct extent_buffer *leaf;
1847 struct btrfs_key key;
1850 path = btrfs_alloc_path();
1854 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1855 key.type = BTRFS_DEV_ITEM_KEY;
1856 key.offset = device->devid;
1858 btrfs_reserve_chunk_metadata(trans, true);
1859 ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path,
1860 &key, sizeof(*dev_item));
1861 btrfs_trans_release_chunk_metadata(trans);
1865 leaf = path->nodes[0];
1866 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1868 btrfs_set_device_id(leaf, dev_item, device->devid);
1869 btrfs_set_device_generation(leaf, dev_item, 0);
1870 btrfs_set_device_type(leaf, dev_item, device->type);
1871 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1872 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1873 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1874 btrfs_set_device_total_bytes(leaf, dev_item,
1875 btrfs_device_get_disk_total_bytes(device));
1876 btrfs_set_device_bytes_used(leaf, dev_item,
1877 btrfs_device_get_bytes_used(device));
1878 btrfs_set_device_group(leaf, dev_item, 0);
1879 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1880 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1881 btrfs_set_device_start_offset(leaf, dev_item, 0);
1883 ptr = btrfs_device_uuid(dev_item);
1884 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1885 ptr = btrfs_device_fsid(dev_item);
1886 write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid,
1887 ptr, BTRFS_FSID_SIZE);
1888 btrfs_mark_buffer_dirty(leaf);
1892 btrfs_free_path(path);
1897 * Function to update ctime/mtime for a given device path.
1898 * Mainly used for ctime/mtime based probe like libblkid.
1900 * We don't care about errors here, this is just to be kind to userspace.
1902 static void update_dev_time(const char *device_path)
1905 struct timespec64 now;
1908 ret = kern_path(device_path, LOOKUP_FOLLOW, &path);
1912 now = current_time(d_inode(path.dentry));
1913 inode_update_time(d_inode(path.dentry), &now, S_MTIME | S_CTIME);
1917 static int btrfs_rm_dev_item(struct btrfs_device *device)
1919 struct btrfs_root *root = device->fs_info->chunk_root;
1921 struct btrfs_path *path;
1922 struct btrfs_key key;
1923 struct btrfs_trans_handle *trans;
1925 path = btrfs_alloc_path();
1929 trans = btrfs_start_transaction(root, 0);
1930 if (IS_ERR(trans)) {
1931 btrfs_free_path(path);
1932 return PTR_ERR(trans);
1934 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1935 key.type = BTRFS_DEV_ITEM_KEY;
1936 key.offset = device->devid;
1938 btrfs_reserve_chunk_metadata(trans, false);
1939 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1940 btrfs_trans_release_chunk_metadata(trans);
1944 btrfs_abort_transaction(trans, ret);
1945 btrfs_end_transaction(trans);
1949 ret = btrfs_del_item(trans, root, path);
1951 btrfs_abort_transaction(trans, ret);
1952 btrfs_end_transaction(trans);
1956 btrfs_free_path(path);
1958 ret = btrfs_commit_transaction(trans);
1963 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1964 * filesystem. It's up to the caller to adjust that number regarding eg. device
1967 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1975 seq = read_seqbegin(&fs_info->profiles_lock);
1977 all_avail = fs_info->avail_data_alloc_bits |
1978 fs_info->avail_system_alloc_bits |
1979 fs_info->avail_metadata_alloc_bits;
1980 } while (read_seqretry(&fs_info->profiles_lock, seq));
1982 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1983 if (!(all_avail & btrfs_raid_array[i].bg_flag))
1986 if (num_devices < btrfs_raid_array[i].devs_min)
1987 return btrfs_raid_array[i].mindev_error;
1993 static struct btrfs_device * btrfs_find_next_active_device(
1994 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device)
1996 struct btrfs_device *next_device;
1998 list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1999 if (next_device != device &&
2000 !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state)
2001 && next_device->bdev)
2009 * Helper function to check if the given device is part of s_bdev / latest_dev
2010 * and replace it with the provided or the next active device, in the context
2011 * where this function called, there should be always be another device (or
2012 * this_dev) which is active.
2014 void __cold btrfs_assign_next_active_device(struct btrfs_device *device,
2015 struct btrfs_device *next_device)
2017 struct btrfs_fs_info *fs_info = device->fs_info;
2020 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
2022 ASSERT(next_device);
2024 if (fs_info->sb->s_bdev &&
2025 (fs_info->sb->s_bdev == device->bdev))
2026 fs_info->sb->s_bdev = next_device->bdev;
2028 if (fs_info->fs_devices->latest_dev->bdev == device->bdev)
2029 fs_info->fs_devices->latest_dev = next_device;
2033 * Return btrfs_fs_devices::num_devices excluding the device that's being
2034 * currently replaced.
2036 static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info)
2038 u64 num_devices = fs_info->fs_devices->num_devices;
2040 down_read(&fs_info->dev_replace.rwsem);
2041 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
2042 ASSERT(num_devices > 1);
2045 up_read(&fs_info->dev_replace.rwsem);
2050 void btrfs_scratch_superblocks(struct btrfs_fs_info *fs_info,
2051 struct block_device *bdev,
2052 const char *device_path)
2054 struct btrfs_super_block *disk_super;
2060 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; copy_num++) {
2064 disk_super = btrfs_read_dev_one_super(bdev, copy_num);
2065 if (IS_ERR(disk_super))
2068 if (bdev_is_zoned(bdev)) {
2069 btrfs_reset_sb_log_zones(bdev, copy_num);
2073 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
2075 page = virt_to_page(disk_super);
2076 set_page_dirty(page);
2078 /* write_on_page() unlocks the page */
2079 ret = write_one_page(page);
2082 "error clearing superblock number %d (%d)",
2084 btrfs_release_disk_super(disk_super);
2088 /* Notify udev that device has changed */
2089 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
2091 /* Update ctime/mtime for device path for libblkid */
2092 update_dev_time(device_path);
2095 int btrfs_rm_device(struct btrfs_fs_info *fs_info,
2096 struct btrfs_dev_lookup_args *args,
2097 struct block_device **bdev, fmode_t *mode)
2099 struct btrfs_device *device;
2100 struct btrfs_fs_devices *cur_devices;
2101 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2106 * The device list in fs_devices is accessed without locks (neither
2107 * uuid_mutex nor device_list_mutex) as it won't change on a mounted
2108 * filesystem and another device rm cannot run.
2110 num_devices = btrfs_num_devices(fs_info);
2112 ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
2116 device = btrfs_find_device(fs_info->fs_devices, args);
2119 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2125 if (btrfs_pinned_by_swapfile(fs_info, device)) {
2126 btrfs_warn_in_rcu(fs_info,
2127 "cannot remove device %s (devid %llu) due to active swapfile",
2128 rcu_str_deref(device->name), device->devid);
2133 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2134 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
2138 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
2139 fs_info->fs_devices->rw_devices == 1) {
2140 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
2144 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2145 mutex_lock(&fs_info->chunk_mutex);
2146 list_del_init(&device->dev_alloc_list);
2147 device->fs_devices->rw_devices--;
2148 mutex_unlock(&fs_info->chunk_mutex);
2151 ret = btrfs_shrink_device(device, 0);
2156 * TODO: the superblock still includes this device in its num_devices
2157 * counter although write_all_supers() is not locked out. This
2158 * could give a filesystem state which requires a degraded mount.
2160 ret = btrfs_rm_dev_item(device);
2164 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2165 btrfs_scrub_cancel_dev(device);
2168 * the device list mutex makes sure that we don't change
2169 * the device list while someone else is writing out all
2170 * the device supers. Whoever is writing all supers, should
2171 * lock the device list mutex before getting the number of
2172 * devices in the super block (super_copy). Conversely,
2173 * whoever updates the number of devices in the super block
2174 * (super_copy) should hold the device list mutex.
2178 * In normal cases the cur_devices == fs_devices. But in case
2179 * of deleting a seed device, the cur_devices should point to
2180 * its own fs_devices listed under the fs_devices->seed_list.
2182 cur_devices = device->fs_devices;
2183 mutex_lock(&fs_devices->device_list_mutex);
2184 list_del_rcu(&device->dev_list);
2186 cur_devices->num_devices--;
2187 cur_devices->total_devices--;
2188 /* Update total_devices of the parent fs_devices if it's seed */
2189 if (cur_devices != fs_devices)
2190 fs_devices->total_devices--;
2192 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
2193 cur_devices->missing_devices--;
2195 btrfs_assign_next_active_device(device, NULL);
2198 cur_devices->open_devices--;
2199 /* remove sysfs entry */
2200 btrfs_sysfs_remove_device(device);
2203 num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
2204 btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
2205 mutex_unlock(&fs_devices->device_list_mutex);
2208 * At this point, the device is zero sized and detached from the
2209 * devices list. All that's left is to zero out the old supers and
2212 * We cannot call btrfs_close_bdev() here because we're holding the sb
2213 * write lock, and blkdev_put() will pull in the ->open_mutex on the
2214 * block device and it's dependencies. Instead just flush the device
2215 * and let the caller do the final blkdev_put.
2217 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2218 btrfs_scratch_superblocks(fs_info, device->bdev,
2221 sync_blockdev(device->bdev);
2222 invalidate_bdev(device->bdev);
2226 *bdev = device->bdev;
2227 *mode = device->mode;
2229 btrfs_free_device(device);
2232 * This can happen if cur_devices is the private seed devices list. We
2233 * cannot call close_fs_devices() here because it expects the uuid_mutex
2234 * to be held, but in fact we don't need that for the private
2235 * seed_devices, we can simply decrement cur_devices->opened and then
2236 * remove it from our list and free the fs_devices.
2238 if (cur_devices->num_devices == 0) {
2239 list_del_init(&cur_devices->seed_list);
2240 ASSERT(cur_devices->opened == 1);
2241 cur_devices->opened--;
2242 free_fs_devices(cur_devices);
2249 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2250 mutex_lock(&fs_info->chunk_mutex);
2251 list_add(&device->dev_alloc_list,
2252 &fs_devices->alloc_list);
2253 device->fs_devices->rw_devices++;
2254 mutex_unlock(&fs_info->chunk_mutex);
2259 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev)
2261 struct btrfs_fs_devices *fs_devices;
2263 lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex);
2266 * in case of fs with no seed, srcdev->fs_devices will point
2267 * to fs_devices of fs_info. However when the dev being replaced is
2268 * a seed dev it will point to the seed's local fs_devices. In short
2269 * srcdev will have its correct fs_devices in both the cases.
2271 fs_devices = srcdev->fs_devices;
2273 list_del_rcu(&srcdev->dev_list);
2274 list_del(&srcdev->dev_alloc_list);
2275 fs_devices->num_devices--;
2276 if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state))
2277 fs_devices->missing_devices--;
2279 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state))
2280 fs_devices->rw_devices--;
2283 fs_devices->open_devices--;
2286 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev)
2288 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2290 mutex_lock(&uuid_mutex);
2292 btrfs_close_bdev(srcdev);
2294 btrfs_free_device(srcdev);
2296 /* if this is no devs we rather delete the fs_devices */
2297 if (!fs_devices->num_devices) {
2299 * On a mounted FS, num_devices can't be zero unless it's a
2300 * seed. In case of a seed device being replaced, the replace
2301 * target added to the sprout FS, so there will be no more
2302 * device left under the seed FS.
2304 ASSERT(fs_devices->seeding);
2306 list_del_init(&fs_devices->seed_list);
2307 close_fs_devices(fs_devices);
2308 free_fs_devices(fs_devices);
2310 mutex_unlock(&uuid_mutex);
2313 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev)
2315 struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices;
2317 mutex_lock(&fs_devices->device_list_mutex);
2319 btrfs_sysfs_remove_device(tgtdev);
2322 fs_devices->open_devices--;
2324 fs_devices->num_devices--;
2326 btrfs_assign_next_active_device(tgtdev, NULL);
2328 list_del_rcu(&tgtdev->dev_list);
2330 mutex_unlock(&fs_devices->device_list_mutex);
2332 btrfs_scratch_superblocks(tgtdev->fs_info, tgtdev->bdev,
2335 btrfs_close_bdev(tgtdev);
2337 btrfs_free_device(tgtdev);
2341 * Populate args from device at path
2343 * @fs_info: the filesystem
2344 * @args: the args to populate
2345 * @path: the path to the device
2347 * This will read the super block of the device at @path and populate @args with
2348 * the devid, fsid, and uuid. This is meant to be used for ioctls that need to
2349 * lookup a device to operate on, but need to do it before we take any locks.
2350 * This properly handles the special case of "missing" that a user may pass in,
2351 * and does some basic sanity checks. The caller must make sure that @path is
2352 * properly NUL terminated before calling in, and must call
2353 * btrfs_put_dev_args_from_path() in order to free up the temporary fsid and
2356 * Return: 0 for success, -errno for failure
2358 int btrfs_get_dev_args_from_path(struct btrfs_fs_info *fs_info,
2359 struct btrfs_dev_lookup_args *args,
2362 struct btrfs_super_block *disk_super;
2363 struct block_device *bdev;
2366 if (!path || !path[0])
2368 if (!strcmp(path, "missing")) {
2369 args->missing = true;
2373 args->uuid = kzalloc(BTRFS_UUID_SIZE, GFP_KERNEL);
2374 args->fsid = kzalloc(BTRFS_FSID_SIZE, GFP_KERNEL);
2375 if (!args->uuid || !args->fsid) {
2376 btrfs_put_dev_args_from_path(args);
2380 ret = btrfs_get_bdev_and_sb(path, FMODE_READ, fs_info->bdev_holder, 0,
2381 &bdev, &disk_super);
2384 args->devid = btrfs_stack_device_id(&disk_super->dev_item);
2385 memcpy(args->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE);
2386 if (btrfs_fs_incompat(fs_info, METADATA_UUID))
2387 memcpy(args->fsid, disk_super->metadata_uuid, BTRFS_FSID_SIZE);
2389 memcpy(args->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
2390 btrfs_release_disk_super(disk_super);
2391 blkdev_put(bdev, FMODE_READ);
2396 * Only use this jointly with btrfs_get_dev_args_from_path() because we will
2397 * allocate our ->uuid and ->fsid pointers, everybody else uses local variables
2398 * that don't need to be freed.
2400 void btrfs_put_dev_args_from_path(struct btrfs_dev_lookup_args *args)
2408 struct btrfs_device *btrfs_find_device_by_devspec(
2409 struct btrfs_fs_info *fs_info, u64 devid,
2410 const char *device_path)
2412 BTRFS_DEV_LOOKUP_ARGS(args);
2413 struct btrfs_device *device;
2418 device = btrfs_find_device(fs_info->fs_devices, &args);
2420 return ERR_PTR(-ENOENT);
2424 ret = btrfs_get_dev_args_from_path(fs_info, &args, device_path);
2426 return ERR_PTR(ret);
2427 device = btrfs_find_device(fs_info->fs_devices, &args);
2428 btrfs_put_dev_args_from_path(&args);
2430 return ERR_PTR(-ENOENT);
2434 static struct btrfs_fs_devices *btrfs_init_sprout(struct btrfs_fs_info *fs_info)
2436 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2437 struct btrfs_fs_devices *old_devices;
2438 struct btrfs_fs_devices *seed_devices;
2440 lockdep_assert_held(&uuid_mutex);
2441 if (!fs_devices->seeding)
2442 return ERR_PTR(-EINVAL);
2445 * Private copy of the seed devices, anchored at
2446 * fs_info->fs_devices->seed_list
2448 seed_devices = alloc_fs_devices(NULL, NULL);
2449 if (IS_ERR(seed_devices))
2450 return seed_devices;
2453 * It's necessary to retain a copy of the original seed fs_devices in
2454 * fs_uuids so that filesystems which have been seeded can successfully
2455 * reference the seed device from open_seed_devices. This also supports
2458 old_devices = clone_fs_devices(fs_devices);
2459 if (IS_ERR(old_devices)) {
2460 kfree(seed_devices);
2464 list_add(&old_devices->fs_list, &fs_uuids);
2466 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2467 seed_devices->opened = 1;
2468 INIT_LIST_HEAD(&seed_devices->devices);
2469 INIT_LIST_HEAD(&seed_devices->alloc_list);
2470 mutex_init(&seed_devices->device_list_mutex);
2472 return seed_devices;
2476 * Splice seed devices into the sprout fs_devices.
2477 * Generate a new fsid for the sprouted read-write filesystem.
2479 static void btrfs_setup_sprout(struct btrfs_fs_info *fs_info,
2480 struct btrfs_fs_devices *seed_devices)
2482 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2483 struct btrfs_super_block *disk_super = fs_info->super_copy;
2484 struct btrfs_device *device;
2488 * We are updating the fsid, the thread leading to device_list_add()
2489 * could race, so uuid_mutex is needed.
2491 lockdep_assert_held(&uuid_mutex);
2494 * The threads listed below may traverse dev_list but can do that without
2495 * device_list_mutex:
2496 * - All device ops and balance - as we are in btrfs_exclop_start.
2497 * - Various dev_list readers - are using RCU.
2498 * - btrfs_ioctl_fitrim() - is using RCU.
2500 * For-read threads as below are using device_list_mutex:
2501 * - Readonly scrub btrfs_scrub_dev()
2502 * - Readonly scrub btrfs_scrub_progress()
2503 * - btrfs_get_dev_stats()
2505 lockdep_assert_held(&fs_devices->device_list_mutex);
2507 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2509 list_for_each_entry(device, &seed_devices->devices, dev_list)
2510 device->fs_devices = seed_devices;
2512 fs_devices->seeding = false;
2513 fs_devices->num_devices = 0;
2514 fs_devices->open_devices = 0;
2515 fs_devices->missing_devices = 0;
2516 fs_devices->rotating = false;
2517 list_add(&seed_devices->seed_list, &fs_devices->seed_list);
2519 generate_random_uuid(fs_devices->fsid);
2520 memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE);
2521 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2523 super_flags = btrfs_super_flags(disk_super) &
2524 ~BTRFS_SUPER_FLAG_SEEDING;
2525 btrfs_set_super_flags(disk_super, super_flags);
2529 * Store the expected generation for seed devices in device items.
2531 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans)
2533 BTRFS_DEV_LOOKUP_ARGS(args);
2534 struct btrfs_fs_info *fs_info = trans->fs_info;
2535 struct btrfs_root *root = fs_info->chunk_root;
2536 struct btrfs_path *path;
2537 struct extent_buffer *leaf;
2538 struct btrfs_dev_item *dev_item;
2539 struct btrfs_device *device;
2540 struct btrfs_key key;
2541 u8 fs_uuid[BTRFS_FSID_SIZE];
2542 u8 dev_uuid[BTRFS_UUID_SIZE];
2545 path = btrfs_alloc_path();
2549 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2551 key.type = BTRFS_DEV_ITEM_KEY;
2554 btrfs_reserve_chunk_metadata(trans, false);
2555 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2556 btrfs_trans_release_chunk_metadata(trans);
2560 leaf = path->nodes[0];
2562 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2563 ret = btrfs_next_leaf(root, path);
2568 leaf = path->nodes[0];
2569 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2570 btrfs_release_path(path);
2574 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2575 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2576 key.type != BTRFS_DEV_ITEM_KEY)
2579 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2580 struct btrfs_dev_item);
2581 args.devid = btrfs_device_id(leaf, dev_item);
2582 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2584 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2586 args.uuid = dev_uuid;
2587 args.fsid = fs_uuid;
2588 device = btrfs_find_device(fs_info->fs_devices, &args);
2589 BUG_ON(!device); /* Logic error */
2591 if (device->fs_devices->seeding) {
2592 btrfs_set_device_generation(leaf, dev_item,
2593 device->generation);
2594 btrfs_mark_buffer_dirty(leaf);
2602 btrfs_free_path(path);
2606 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2608 struct btrfs_root *root = fs_info->dev_root;
2609 struct request_queue *q;
2610 struct btrfs_trans_handle *trans;
2611 struct btrfs_device *device;
2612 struct block_device *bdev;
2613 struct super_block *sb = fs_info->sb;
2614 struct rcu_string *name;
2615 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2616 struct btrfs_fs_devices *seed_devices;
2617 u64 orig_super_total_bytes;
2618 u64 orig_super_num_devices;
2620 bool seeding_dev = false;
2621 bool locked = false;
2623 if (sb_rdonly(sb) && !fs_devices->seeding)
2626 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2627 fs_info->bdev_holder);
2629 return PTR_ERR(bdev);
2631 if (!btrfs_check_device_zone_type(fs_info, bdev)) {
2636 if (fs_devices->seeding) {
2638 down_write(&sb->s_umount);
2639 mutex_lock(&uuid_mutex);
2643 sync_blockdev(bdev);
2646 list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
2647 if (device->bdev == bdev) {
2655 device = btrfs_alloc_device(fs_info, NULL, NULL);
2656 if (IS_ERR(device)) {
2657 /* we can safely leave the fs_devices entry around */
2658 ret = PTR_ERR(device);
2662 name = rcu_string_strdup(device_path, GFP_KERNEL);
2665 goto error_free_device;
2667 rcu_assign_pointer(device->name, name);
2669 device->fs_info = fs_info;
2670 device->bdev = bdev;
2672 ret = btrfs_get_dev_zone_info(device, false);
2674 goto error_free_device;
2676 trans = btrfs_start_transaction(root, 0);
2677 if (IS_ERR(trans)) {
2678 ret = PTR_ERR(trans);
2679 goto error_free_zone;
2682 q = bdev_get_queue(bdev);
2683 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
2684 device->generation = trans->transid;
2685 device->io_width = fs_info->sectorsize;
2686 device->io_align = fs_info->sectorsize;
2687 device->sector_size = fs_info->sectorsize;
2688 device->total_bytes =
2689 round_down(bdev_nr_bytes(bdev), fs_info->sectorsize);
2690 device->disk_total_bytes = device->total_bytes;
2691 device->commit_total_bytes = device->total_bytes;
2692 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2693 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
2694 device->mode = FMODE_EXCL;
2695 device->dev_stats_valid = 1;
2696 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2699 btrfs_clear_sb_rdonly(sb);
2701 /* GFP_KERNEL allocation must not be under device_list_mutex */
2702 seed_devices = btrfs_init_sprout(fs_info);
2703 if (IS_ERR(seed_devices)) {
2704 ret = PTR_ERR(seed_devices);
2705 btrfs_abort_transaction(trans, ret);
2710 mutex_lock(&fs_devices->device_list_mutex);
2712 btrfs_setup_sprout(fs_info, seed_devices);
2713 btrfs_assign_next_active_device(fs_info->fs_devices->latest_dev,
2717 device->fs_devices = fs_devices;
2719 mutex_lock(&fs_info->chunk_mutex);
2720 list_add_rcu(&device->dev_list, &fs_devices->devices);
2721 list_add(&device->dev_alloc_list, &fs_devices->alloc_list);
2722 fs_devices->num_devices++;
2723 fs_devices->open_devices++;
2724 fs_devices->rw_devices++;
2725 fs_devices->total_devices++;
2726 fs_devices->total_rw_bytes += device->total_bytes;
2728 atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2730 if (!blk_queue_nonrot(q))
2731 fs_devices->rotating = true;
2733 orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
2734 btrfs_set_super_total_bytes(fs_info->super_copy,
2735 round_down(orig_super_total_bytes + device->total_bytes,
2736 fs_info->sectorsize));
2738 orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy);
2739 btrfs_set_super_num_devices(fs_info->super_copy,
2740 orig_super_num_devices + 1);
2743 * we've got more storage, clear any full flags on the space
2746 btrfs_clear_space_info_full(fs_info);
2748 mutex_unlock(&fs_info->chunk_mutex);
2750 /* Add sysfs device entry */
2751 btrfs_sysfs_add_device(device);
2753 mutex_unlock(&fs_devices->device_list_mutex);
2756 mutex_lock(&fs_info->chunk_mutex);
2757 ret = init_first_rw_device(trans);
2758 mutex_unlock(&fs_info->chunk_mutex);
2760 btrfs_abort_transaction(trans, ret);
2765 ret = btrfs_add_dev_item(trans, device);
2767 btrfs_abort_transaction(trans, ret);
2772 ret = btrfs_finish_sprout(trans);
2774 btrfs_abort_transaction(trans, ret);
2779 * fs_devices now represents the newly sprouted filesystem and
2780 * its fsid has been changed by btrfs_sprout_splice().
2782 btrfs_sysfs_update_sprout_fsid(fs_devices);
2785 ret = btrfs_commit_transaction(trans);
2788 mutex_unlock(&uuid_mutex);
2789 up_write(&sb->s_umount);
2792 if (ret) /* transaction commit */
2795 ret = btrfs_relocate_sys_chunks(fs_info);
2797 btrfs_handle_fs_error(fs_info, ret,
2798 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2799 trans = btrfs_attach_transaction(root);
2800 if (IS_ERR(trans)) {
2801 if (PTR_ERR(trans) == -ENOENT)
2803 ret = PTR_ERR(trans);
2807 ret = btrfs_commit_transaction(trans);
2811 * Now that we have written a new super block to this device, check all
2812 * other fs_devices list if device_path alienates any other scanned
2814 * We can ignore the return value as it typically returns -EINVAL and
2815 * only succeeds if the device was an alien.
2817 btrfs_forget_devices(device_path);
2819 /* Update ctime/mtime for blkid or udev */
2820 update_dev_time(device_path);
2825 btrfs_sysfs_remove_device(device);
2826 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2827 mutex_lock(&fs_info->chunk_mutex);
2828 list_del_rcu(&device->dev_list);
2829 list_del(&device->dev_alloc_list);
2830 fs_info->fs_devices->num_devices--;
2831 fs_info->fs_devices->open_devices--;
2832 fs_info->fs_devices->rw_devices--;
2833 fs_info->fs_devices->total_devices--;
2834 fs_info->fs_devices->total_rw_bytes -= device->total_bytes;
2835 atomic64_sub(device->total_bytes, &fs_info->free_chunk_space);
2836 btrfs_set_super_total_bytes(fs_info->super_copy,
2837 orig_super_total_bytes);
2838 btrfs_set_super_num_devices(fs_info->super_copy,
2839 orig_super_num_devices);
2840 mutex_unlock(&fs_info->chunk_mutex);
2841 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2844 btrfs_set_sb_rdonly(sb);
2846 btrfs_end_transaction(trans);
2848 btrfs_destroy_dev_zone_info(device);
2850 btrfs_free_device(device);
2852 blkdev_put(bdev, FMODE_EXCL);
2854 mutex_unlock(&uuid_mutex);
2855 up_write(&sb->s_umount);
2860 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2861 struct btrfs_device *device)
2864 struct btrfs_path *path;
2865 struct btrfs_root *root = device->fs_info->chunk_root;
2866 struct btrfs_dev_item *dev_item;
2867 struct extent_buffer *leaf;
2868 struct btrfs_key key;
2870 path = btrfs_alloc_path();
2874 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2875 key.type = BTRFS_DEV_ITEM_KEY;
2876 key.offset = device->devid;
2878 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2887 leaf = path->nodes[0];
2888 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2890 btrfs_set_device_id(leaf, dev_item, device->devid);
2891 btrfs_set_device_type(leaf, dev_item, device->type);
2892 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2893 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2894 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2895 btrfs_set_device_total_bytes(leaf, dev_item,
2896 btrfs_device_get_disk_total_bytes(device));
2897 btrfs_set_device_bytes_used(leaf, dev_item,
2898 btrfs_device_get_bytes_used(device));
2899 btrfs_mark_buffer_dirty(leaf);
2902 btrfs_free_path(path);
2906 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2907 struct btrfs_device *device, u64 new_size)
2909 struct btrfs_fs_info *fs_info = device->fs_info;
2910 struct btrfs_super_block *super_copy = fs_info->super_copy;
2915 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2918 new_size = round_down(new_size, fs_info->sectorsize);
2920 mutex_lock(&fs_info->chunk_mutex);
2921 old_total = btrfs_super_total_bytes(super_copy);
2922 diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2924 if (new_size <= device->total_bytes ||
2925 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2926 mutex_unlock(&fs_info->chunk_mutex);
2930 btrfs_set_super_total_bytes(super_copy,
2931 round_down(old_total + diff, fs_info->sectorsize));
2932 device->fs_devices->total_rw_bytes += diff;
2934 btrfs_device_set_total_bytes(device, new_size);
2935 btrfs_device_set_disk_total_bytes(device, new_size);
2936 btrfs_clear_space_info_full(device->fs_info);
2937 if (list_empty(&device->post_commit_list))
2938 list_add_tail(&device->post_commit_list,
2939 &trans->transaction->dev_update_list);
2940 mutex_unlock(&fs_info->chunk_mutex);
2942 btrfs_reserve_chunk_metadata(trans, false);
2943 ret = btrfs_update_device(trans, device);
2944 btrfs_trans_release_chunk_metadata(trans);
2949 static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
2951 struct btrfs_fs_info *fs_info = trans->fs_info;
2952 struct btrfs_root *root = fs_info->chunk_root;
2954 struct btrfs_path *path;
2955 struct btrfs_key key;
2957 path = btrfs_alloc_path();
2961 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2962 key.offset = chunk_offset;
2963 key.type = BTRFS_CHUNK_ITEM_KEY;
2965 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2968 else if (ret > 0) { /* Logic error or corruption */
2969 btrfs_handle_fs_error(fs_info, -ENOENT,
2970 "Failed lookup while freeing chunk.");
2975 ret = btrfs_del_item(trans, root, path);
2977 btrfs_handle_fs_error(fs_info, ret,
2978 "Failed to delete chunk item.");
2980 btrfs_free_path(path);
2984 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2986 struct btrfs_super_block *super_copy = fs_info->super_copy;
2987 struct btrfs_disk_key *disk_key;
2988 struct btrfs_chunk *chunk;
2995 struct btrfs_key key;
2997 lockdep_assert_held(&fs_info->chunk_mutex);
2998 array_size = btrfs_super_sys_array_size(super_copy);
3000 ptr = super_copy->sys_chunk_array;
3003 while (cur < array_size) {
3004 disk_key = (struct btrfs_disk_key *)ptr;
3005 btrfs_disk_key_to_cpu(&key, disk_key);
3007 len = sizeof(*disk_key);
3009 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
3010 chunk = (struct btrfs_chunk *)(ptr + len);
3011 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
3012 len += btrfs_chunk_item_size(num_stripes);
3017 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID &&
3018 key.offset == chunk_offset) {
3019 memmove(ptr, ptr + len, array_size - (cur + len));
3021 btrfs_set_super_sys_array_size(super_copy, array_size);
3031 * btrfs_get_chunk_map() - Find the mapping containing the given logical extent.
3032 * @logical: Logical block offset in bytes.
3033 * @length: Length of extent in bytes.
3035 * Return: Chunk mapping or ERR_PTR.
3037 struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info,
3038 u64 logical, u64 length)
3040 struct extent_map_tree *em_tree;
3041 struct extent_map *em;
3043 em_tree = &fs_info->mapping_tree;
3044 read_lock(&em_tree->lock);
3045 em = lookup_extent_mapping(em_tree, logical, length);
3046 read_unlock(&em_tree->lock);
3049 btrfs_crit(fs_info, "unable to find logical %llu length %llu",
3051 return ERR_PTR(-EINVAL);
3054 if (em->start > logical || em->start + em->len < logical) {
3056 "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
3057 logical, length, em->start, em->start + em->len);
3058 free_extent_map(em);
3059 return ERR_PTR(-EINVAL);
3062 /* callers are responsible for dropping em's ref. */
3066 static int remove_chunk_item(struct btrfs_trans_handle *trans,
3067 struct map_lookup *map, u64 chunk_offset)
3072 * Removing chunk items and updating the device items in the chunks btree
3073 * requires holding the chunk_mutex.
3074 * See the comment at btrfs_chunk_alloc() for the details.
3076 lockdep_assert_held(&trans->fs_info->chunk_mutex);
3078 for (i = 0; i < map->num_stripes; i++) {
3081 ret = btrfs_update_device(trans, map->stripes[i].dev);
3086 return btrfs_free_chunk(trans, chunk_offset);
3089 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
3091 struct btrfs_fs_info *fs_info = trans->fs_info;
3092 struct extent_map *em;
3093 struct map_lookup *map;
3094 u64 dev_extent_len = 0;
3096 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
3098 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
3101 * This is a logic error, but we don't want to just rely on the
3102 * user having built with ASSERT enabled, so if ASSERT doesn't
3103 * do anything we still error out.
3108 map = em->map_lookup;
3111 * First delete the device extent items from the devices btree.
3112 * We take the device_list_mutex to avoid racing with the finishing phase
3113 * of a device replace operation. See the comment below before acquiring
3114 * fs_info->chunk_mutex. Note that here we do not acquire the chunk_mutex
3115 * because that can result in a deadlock when deleting the device extent
3116 * items from the devices btree - COWing an extent buffer from the btree
3117 * may result in allocating a new metadata chunk, which would attempt to
3118 * lock again fs_info->chunk_mutex.
3120 mutex_lock(&fs_devices->device_list_mutex);
3121 for (i = 0; i < map->num_stripes; i++) {
3122 struct btrfs_device *device = map->stripes[i].dev;
3123 ret = btrfs_free_dev_extent(trans, device,
3124 map->stripes[i].physical,
3127 mutex_unlock(&fs_devices->device_list_mutex);
3128 btrfs_abort_transaction(trans, ret);
3132 if (device->bytes_used > 0) {
3133 mutex_lock(&fs_info->chunk_mutex);
3134 btrfs_device_set_bytes_used(device,
3135 device->bytes_used - dev_extent_len);
3136 atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
3137 btrfs_clear_space_info_full(fs_info);
3138 mutex_unlock(&fs_info->chunk_mutex);
3141 mutex_unlock(&fs_devices->device_list_mutex);
3144 * We acquire fs_info->chunk_mutex for 2 reasons:
3146 * 1) Just like with the first phase of the chunk allocation, we must
3147 * reserve system space, do all chunk btree updates and deletions, and
3148 * update the system chunk array in the superblock while holding this
3149 * mutex. This is for similar reasons as explained on the comment at
3150 * the top of btrfs_chunk_alloc();
3152 * 2) Prevent races with the final phase of a device replace operation
3153 * that replaces the device object associated with the map's stripes,
3154 * because the device object's id can change at any time during that
3155 * final phase of the device replace operation
3156 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the
3157 * replaced device and then see it with an ID of
3158 * BTRFS_DEV_REPLACE_DEVID, which would cause a failure when updating
3159 * the device item, which does not exists on the chunk btree.
3160 * The finishing phase of device replace acquires both the
3161 * device_list_mutex and the chunk_mutex, in that order, so we are
3162 * safe by just acquiring the chunk_mutex.
3164 trans->removing_chunk = true;
3165 mutex_lock(&fs_info->chunk_mutex);
3167 check_system_chunk(trans, map->type);
3169 ret = remove_chunk_item(trans, map, chunk_offset);
3171 * Normally we should not get -ENOSPC since we reserved space before
3172 * through the call to check_system_chunk().
3174 * Despite our system space_info having enough free space, we may not
3175 * be able to allocate extents from its block groups, because all have
3176 * an incompatible profile, which will force us to allocate a new system
3177 * block group with the right profile, or right after we called
3178 * check_system_space() above, a scrub turned the only system block group
3179 * with enough free space into RO mode.
3180 * This is explained with more detail at do_chunk_alloc().
3182 * So if we get -ENOSPC, allocate a new system chunk and retry once.
3184 if (ret == -ENOSPC) {
3185 const u64 sys_flags = btrfs_system_alloc_profile(fs_info);
3186 struct btrfs_block_group *sys_bg;
3188 sys_bg = btrfs_create_chunk(trans, sys_flags);
3189 if (IS_ERR(sys_bg)) {
3190 ret = PTR_ERR(sys_bg);
3191 btrfs_abort_transaction(trans, ret);
3195 ret = btrfs_chunk_alloc_add_chunk_item(trans, sys_bg);
3197 btrfs_abort_transaction(trans, ret);
3201 ret = remove_chunk_item(trans, map, chunk_offset);
3203 btrfs_abort_transaction(trans, ret);
3207 btrfs_abort_transaction(trans, ret);
3211 trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
3213 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
3214 ret = btrfs_del_sys_chunk(fs_info, chunk_offset);
3216 btrfs_abort_transaction(trans, ret);
3221 mutex_unlock(&fs_info->chunk_mutex);
3222 trans->removing_chunk = false;
3225 * We are done with chunk btree updates and deletions, so release the
3226 * system space we previously reserved (with check_system_chunk()).
3228 btrfs_trans_release_chunk_metadata(trans);
3230 ret = btrfs_remove_block_group(trans, chunk_offset, em);
3232 btrfs_abort_transaction(trans, ret);
3237 if (trans->removing_chunk) {
3238 mutex_unlock(&fs_info->chunk_mutex);
3239 trans->removing_chunk = false;
3242 free_extent_map(em);
3246 int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
3248 struct btrfs_root *root = fs_info->chunk_root;
3249 struct btrfs_trans_handle *trans;
3250 struct btrfs_block_group *block_group;
3255 * Prevent races with automatic removal of unused block groups.
3256 * After we relocate and before we remove the chunk with offset
3257 * chunk_offset, automatic removal of the block group can kick in,
3258 * resulting in a failure when calling btrfs_remove_chunk() below.
3260 * Make sure to acquire this mutex before doing a tree search (dev
3261 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
3262 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
3263 * we release the path used to search the chunk/dev tree and before
3264 * the current task acquires this mutex and calls us.
3266 lockdep_assert_held(&fs_info->reclaim_bgs_lock);
3268 /* step one, relocate all the extents inside this chunk */
3269 btrfs_scrub_pause(fs_info);
3270 ret = btrfs_relocate_block_group(fs_info, chunk_offset);
3271 btrfs_scrub_continue(fs_info);
3275 block_group = btrfs_lookup_block_group(fs_info, chunk_offset);
3278 btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group);
3279 length = block_group->length;
3280 btrfs_put_block_group(block_group);
3283 * On a zoned file system, discard the whole block group, this will
3284 * trigger a REQ_OP_ZONE_RESET operation on the device zone. If
3285 * resetting the zone fails, don't treat it as a fatal problem from the
3286 * filesystem's point of view.
3288 if (btrfs_is_zoned(fs_info)) {
3289 ret = btrfs_discard_extent(fs_info, chunk_offset, length, NULL);
3292 "failed to reset zone %llu after relocation",
3296 trans = btrfs_start_trans_remove_block_group(root->fs_info,
3298 if (IS_ERR(trans)) {
3299 ret = PTR_ERR(trans);
3300 btrfs_handle_fs_error(root->fs_info, ret, NULL);
3305 * step two, delete the device extents and the
3306 * chunk tree entries
3308 ret = btrfs_remove_chunk(trans, chunk_offset);
3309 btrfs_end_transaction(trans);
3313 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
3315 struct btrfs_root *chunk_root = fs_info->chunk_root;
3316 struct btrfs_path *path;
3317 struct extent_buffer *leaf;
3318 struct btrfs_chunk *chunk;
3319 struct btrfs_key key;
3320 struct btrfs_key found_key;
3322 bool retried = false;
3326 path = btrfs_alloc_path();
3331 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3332 key.offset = (u64)-1;
3333 key.type = BTRFS_CHUNK_ITEM_KEY;
3336 mutex_lock(&fs_info->reclaim_bgs_lock);
3337 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3339 mutex_unlock(&fs_info->reclaim_bgs_lock);
3342 BUG_ON(ret == 0); /* Corruption */
3344 ret = btrfs_previous_item(chunk_root, path, key.objectid,
3347 mutex_unlock(&fs_info->reclaim_bgs_lock);
3353 leaf = path->nodes[0];
3354 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3356 chunk = btrfs_item_ptr(leaf, path->slots[0],
3357 struct btrfs_chunk);
3358 chunk_type = btrfs_chunk_type(leaf, chunk);
3359 btrfs_release_path(path);
3361 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3362 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3368 mutex_unlock(&fs_info->reclaim_bgs_lock);
3370 if (found_key.offset == 0)
3372 key.offset = found_key.offset - 1;
3375 if (failed && !retried) {
3379 } else if (WARN_ON(failed && retried)) {
3383 btrfs_free_path(path);
3388 * return 1 : allocate a data chunk successfully,
3389 * return <0: errors during allocating a data chunk,
3390 * return 0 : no need to allocate a data chunk.
3392 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info,
3395 struct btrfs_block_group *cache;
3399 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3401 chunk_type = cache->flags;
3402 btrfs_put_block_group(cache);
3404 if (!(chunk_type & BTRFS_BLOCK_GROUP_DATA))
3407 spin_lock(&fs_info->data_sinfo->lock);
3408 bytes_used = fs_info->data_sinfo->bytes_used;
3409 spin_unlock(&fs_info->data_sinfo->lock);
3412 struct btrfs_trans_handle *trans;
3415 trans = btrfs_join_transaction(fs_info->tree_root);
3417 return PTR_ERR(trans);
3419 ret = btrfs_force_chunk_alloc(trans, BTRFS_BLOCK_GROUP_DATA);
3420 btrfs_end_transaction(trans);
3429 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3430 struct btrfs_balance_control *bctl)
3432 struct btrfs_root *root = fs_info->tree_root;
3433 struct btrfs_trans_handle *trans;
3434 struct btrfs_balance_item *item;
3435 struct btrfs_disk_balance_args disk_bargs;
3436 struct btrfs_path *path;
3437 struct extent_buffer *leaf;
3438 struct btrfs_key key;
3441 path = btrfs_alloc_path();
3445 trans = btrfs_start_transaction(root, 0);
3446 if (IS_ERR(trans)) {
3447 btrfs_free_path(path);
3448 return PTR_ERR(trans);
3451 key.objectid = BTRFS_BALANCE_OBJECTID;
3452 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3455 ret = btrfs_insert_empty_item(trans, root, path, &key,
3460 leaf = path->nodes[0];
3461 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3463 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3465 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3466 btrfs_set_balance_data(leaf, item, &disk_bargs);
3467 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3468 btrfs_set_balance_meta(leaf, item, &disk_bargs);
3469 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3470 btrfs_set_balance_sys(leaf, item, &disk_bargs);
3472 btrfs_set_balance_flags(leaf, item, bctl->flags);
3474 btrfs_mark_buffer_dirty(leaf);
3476 btrfs_free_path(path);
3477 err = btrfs_commit_transaction(trans);
3483 static int del_balance_item(struct btrfs_fs_info *fs_info)
3485 struct btrfs_root *root = fs_info->tree_root;
3486 struct btrfs_trans_handle *trans;
3487 struct btrfs_path *path;
3488 struct btrfs_key key;
3491 path = btrfs_alloc_path();
3495 trans = btrfs_start_transaction_fallback_global_rsv(root, 0);
3496 if (IS_ERR(trans)) {
3497 btrfs_free_path(path);
3498 return PTR_ERR(trans);
3501 key.objectid = BTRFS_BALANCE_OBJECTID;
3502 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3505 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3513 ret = btrfs_del_item(trans, root, path);
3515 btrfs_free_path(path);
3516 err = btrfs_commit_transaction(trans);
3523 * This is a heuristic used to reduce the number of chunks balanced on
3524 * resume after balance was interrupted.
3526 static void update_balance_args(struct btrfs_balance_control *bctl)
3529 * Turn on soft mode for chunk types that were being converted.
3531 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3532 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3533 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3534 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3535 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3536 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3539 * Turn on usage filter if is not already used. The idea is
3540 * that chunks that we have already balanced should be
3541 * reasonably full. Don't do it for chunks that are being
3542 * converted - that will keep us from relocating unconverted
3543 * (albeit full) chunks.
3545 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3546 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3547 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3548 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3549 bctl->data.usage = 90;
3551 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3552 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3553 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3554 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3555 bctl->sys.usage = 90;
3557 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3558 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3559 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3560 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3561 bctl->meta.usage = 90;
3566 * Clear the balance status in fs_info and delete the balance item from disk.
3568 static void reset_balance_state(struct btrfs_fs_info *fs_info)
3570 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3573 BUG_ON(!fs_info->balance_ctl);
3575 spin_lock(&fs_info->balance_lock);
3576 fs_info->balance_ctl = NULL;
3577 spin_unlock(&fs_info->balance_lock);
3580 ret = del_balance_item(fs_info);
3582 btrfs_handle_fs_error(fs_info, ret, NULL);
3586 * Balance filters. Return 1 if chunk should be filtered out
3587 * (should not be balanced).
3589 static int chunk_profiles_filter(u64 chunk_type,
3590 struct btrfs_balance_args *bargs)
3592 chunk_type = chunk_to_extended(chunk_type) &
3593 BTRFS_EXTENDED_PROFILE_MASK;
3595 if (bargs->profiles & chunk_type)
3601 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3602 struct btrfs_balance_args *bargs)
3604 struct btrfs_block_group *cache;
3606 u64 user_thresh_min;
3607 u64 user_thresh_max;
3610 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3611 chunk_used = cache->used;
3613 if (bargs->usage_min == 0)
3614 user_thresh_min = 0;
3616 user_thresh_min = div_factor_fine(cache->length,
3619 if (bargs->usage_max == 0)
3620 user_thresh_max = 1;
3621 else if (bargs->usage_max > 100)
3622 user_thresh_max = cache->length;
3624 user_thresh_max = div_factor_fine(cache->length,
3627 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3630 btrfs_put_block_group(cache);
3634 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3635 u64 chunk_offset, struct btrfs_balance_args *bargs)
3637 struct btrfs_block_group *cache;
3638 u64 chunk_used, user_thresh;
3641 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3642 chunk_used = cache->used;
3644 if (bargs->usage_min == 0)
3646 else if (bargs->usage > 100)
3647 user_thresh = cache->length;
3649 user_thresh = div_factor_fine(cache->length, bargs->usage);
3651 if (chunk_used < user_thresh)
3654 btrfs_put_block_group(cache);
3658 static int chunk_devid_filter(struct extent_buffer *leaf,
3659 struct btrfs_chunk *chunk,
3660 struct btrfs_balance_args *bargs)
3662 struct btrfs_stripe *stripe;
3663 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3666 for (i = 0; i < num_stripes; i++) {
3667 stripe = btrfs_stripe_nr(chunk, i);
3668 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3675 static u64 calc_data_stripes(u64 type, int num_stripes)
3677 const int index = btrfs_bg_flags_to_raid_index(type);
3678 const int ncopies = btrfs_raid_array[index].ncopies;
3679 const int nparity = btrfs_raid_array[index].nparity;
3681 return (num_stripes - nparity) / ncopies;
3684 /* [pstart, pend) */
3685 static int chunk_drange_filter(struct extent_buffer *leaf,
3686 struct btrfs_chunk *chunk,
3687 struct btrfs_balance_args *bargs)
3689 struct btrfs_stripe *stripe;
3690 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3697 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3700 type = btrfs_chunk_type(leaf, chunk);
3701 factor = calc_data_stripes(type, num_stripes);
3703 for (i = 0; i < num_stripes; i++) {
3704 stripe = btrfs_stripe_nr(chunk, i);
3705 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3708 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3709 stripe_length = btrfs_chunk_length(leaf, chunk);
3710 stripe_length = div_u64(stripe_length, factor);
3712 if (stripe_offset < bargs->pend &&
3713 stripe_offset + stripe_length > bargs->pstart)
3720 /* [vstart, vend) */
3721 static int chunk_vrange_filter(struct extent_buffer *leaf,
3722 struct btrfs_chunk *chunk,
3724 struct btrfs_balance_args *bargs)
3726 if (chunk_offset < bargs->vend &&
3727 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3728 /* at least part of the chunk is inside this vrange */
3734 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3735 struct btrfs_chunk *chunk,
3736 struct btrfs_balance_args *bargs)
3738 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3740 if (bargs->stripes_min <= num_stripes
3741 && num_stripes <= bargs->stripes_max)
3747 static int chunk_soft_convert_filter(u64 chunk_type,
3748 struct btrfs_balance_args *bargs)
3750 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3753 chunk_type = chunk_to_extended(chunk_type) &
3754 BTRFS_EXTENDED_PROFILE_MASK;
3756 if (bargs->target == chunk_type)
3762 static int should_balance_chunk(struct extent_buffer *leaf,
3763 struct btrfs_chunk *chunk, u64 chunk_offset)
3765 struct btrfs_fs_info *fs_info = leaf->fs_info;
3766 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3767 struct btrfs_balance_args *bargs = NULL;
3768 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3771 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3772 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3776 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3777 bargs = &bctl->data;
3778 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3780 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3781 bargs = &bctl->meta;
3783 /* profiles filter */
3784 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3785 chunk_profiles_filter(chunk_type, bargs)) {
3790 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3791 chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3793 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3794 chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3799 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3800 chunk_devid_filter(leaf, chunk, bargs)) {
3804 /* drange filter, makes sense only with devid filter */
3805 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3806 chunk_drange_filter(leaf, chunk, bargs)) {
3811 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3812 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3816 /* stripes filter */
3817 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3818 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3822 /* soft profile changing mode */
3823 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3824 chunk_soft_convert_filter(chunk_type, bargs)) {
3829 * limited by count, must be the last filter
3831 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3832 if (bargs->limit == 0)
3836 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3838 * Same logic as the 'limit' filter; the minimum cannot be
3839 * determined here because we do not have the global information
3840 * about the count of all chunks that satisfy the filters.
3842 if (bargs->limit_max == 0)
3851 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3853 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3854 struct btrfs_root *chunk_root = fs_info->chunk_root;
3856 struct btrfs_chunk *chunk;
3857 struct btrfs_path *path = NULL;
3858 struct btrfs_key key;
3859 struct btrfs_key found_key;
3860 struct extent_buffer *leaf;
3863 int enospc_errors = 0;
3864 bool counting = true;
3865 /* The single value limit and min/max limits use the same bytes in the */
3866 u64 limit_data = bctl->data.limit;
3867 u64 limit_meta = bctl->meta.limit;
3868 u64 limit_sys = bctl->sys.limit;
3872 int chunk_reserved = 0;
3874 path = btrfs_alloc_path();
3880 /* zero out stat counters */
3881 spin_lock(&fs_info->balance_lock);
3882 memset(&bctl->stat, 0, sizeof(bctl->stat));
3883 spin_unlock(&fs_info->balance_lock);
3887 * The single value limit and min/max limits use the same bytes
3890 bctl->data.limit = limit_data;
3891 bctl->meta.limit = limit_meta;
3892 bctl->sys.limit = limit_sys;
3894 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3895 key.offset = (u64)-1;
3896 key.type = BTRFS_CHUNK_ITEM_KEY;
3899 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3900 atomic_read(&fs_info->balance_cancel_req)) {
3905 mutex_lock(&fs_info->reclaim_bgs_lock);
3906 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3908 mutex_unlock(&fs_info->reclaim_bgs_lock);
3913 * this shouldn't happen, it means the last relocate
3917 BUG(); /* FIXME break ? */
3919 ret = btrfs_previous_item(chunk_root, path, 0,
3920 BTRFS_CHUNK_ITEM_KEY);
3922 mutex_unlock(&fs_info->reclaim_bgs_lock);
3927 leaf = path->nodes[0];
3928 slot = path->slots[0];
3929 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3931 if (found_key.objectid != key.objectid) {
3932 mutex_unlock(&fs_info->reclaim_bgs_lock);
3936 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3937 chunk_type = btrfs_chunk_type(leaf, chunk);
3940 spin_lock(&fs_info->balance_lock);
3941 bctl->stat.considered++;
3942 spin_unlock(&fs_info->balance_lock);
3945 ret = should_balance_chunk(leaf, chunk, found_key.offset);
3947 btrfs_release_path(path);
3949 mutex_unlock(&fs_info->reclaim_bgs_lock);
3954 mutex_unlock(&fs_info->reclaim_bgs_lock);
3955 spin_lock(&fs_info->balance_lock);
3956 bctl->stat.expected++;
3957 spin_unlock(&fs_info->balance_lock);
3959 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3961 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3963 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3970 * Apply limit_min filter, no need to check if the LIMITS
3971 * filter is used, limit_min is 0 by default
3973 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3974 count_data < bctl->data.limit_min)
3975 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3976 count_meta < bctl->meta.limit_min)
3977 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3978 count_sys < bctl->sys.limit_min)) {
3979 mutex_unlock(&fs_info->reclaim_bgs_lock);
3983 if (!chunk_reserved) {
3985 * We may be relocating the only data chunk we have,
3986 * which could potentially end up with losing data's
3987 * raid profile, so lets allocate an empty one in
3990 ret = btrfs_may_alloc_data_chunk(fs_info,
3993 mutex_unlock(&fs_info->reclaim_bgs_lock);
3995 } else if (ret == 1) {
4000 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
4001 mutex_unlock(&fs_info->reclaim_bgs_lock);
4002 if (ret == -ENOSPC) {
4004 } else if (ret == -ETXTBSY) {
4006 "skipping relocation of block group %llu due to active swapfile",
4012 spin_lock(&fs_info->balance_lock);
4013 bctl->stat.completed++;
4014 spin_unlock(&fs_info->balance_lock);
4017 if (found_key.offset == 0)
4019 key.offset = found_key.offset - 1;
4023 btrfs_release_path(path);
4028 btrfs_free_path(path);
4029 if (enospc_errors) {
4030 btrfs_info(fs_info, "%d enospc errors during balance",
4040 * alloc_profile_is_valid - see if a given profile is valid and reduced
4041 * @flags: profile to validate
4042 * @extended: if true @flags is treated as an extended profile
4044 static int alloc_profile_is_valid(u64 flags, int extended)
4046 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
4047 BTRFS_BLOCK_GROUP_PROFILE_MASK);
4049 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
4051 /* 1) check that all other bits are zeroed */
4055 /* 2) see if profile is reduced */
4057 return !extended; /* "0" is valid for usual profiles */
4059 return has_single_bit_set(flags);
4062 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
4064 /* cancel requested || normal exit path */
4065 return atomic_read(&fs_info->balance_cancel_req) ||
4066 (atomic_read(&fs_info->balance_pause_req) == 0 &&
4067 atomic_read(&fs_info->balance_cancel_req) == 0);
4071 * Validate target profile against allowed profiles and return true if it's OK.
4072 * Otherwise print the error message and return false.
4074 static inline int validate_convert_profile(struct btrfs_fs_info *fs_info,
4075 const struct btrfs_balance_args *bargs,
4076 u64 allowed, const char *type)
4078 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
4081 if (fs_info->sectorsize < PAGE_SIZE &&
4082 bargs->target & BTRFS_BLOCK_GROUP_RAID56_MASK) {
4084 "RAID56 is not yet supported for sectorsize %u with page size %lu",
4085 fs_info->sectorsize, PAGE_SIZE);
4088 /* Profile is valid and does not have bits outside of the allowed set */
4089 if (alloc_profile_is_valid(bargs->target, 1) &&
4090 (bargs->target & ~allowed) == 0)
4093 btrfs_err(fs_info, "balance: invalid convert %s profile %s",
4094 type, btrfs_bg_type_to_raid_name(bargs->target));
4099 * Fill @buf with textual description of balance filter flags @bargs, up to
4100 * @size_buf including the terminating null. The output may be trimmed if it
4101 * does not fit into the provided buffer.
4103 static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf,
4107 u32 size_bp = size_buf;
4109 u64 flags = bargs->flags;
4110 char tmp_buf[128] = {'\0'};
4115 #define CHECK_APPEND_NOARG(a) \
4117 ret = snprintf(bp, size_bp, (a)); \
4118 if (ret < 0 || ret >= size_bp) \
4119 goto out_overflow; \
4124 #define CHECK_APPEND_1ARG(a, v1) \
4126 ret = snprintf(bp, size_bp, (a), (v1)); \
4127 if (ret < 0 || ret >= size_bp) \
4128 goto out_overflow; \
4133 #define CHECK_APPEND_2ARG(a, v1, v2) \
4135 ret = snprintf(bp, size_bp, (a), (v1), (v2)); \
4136 if (ret < 0 || ret >= size_bp) \
4137 goto out_overflow; \
4142 if (flags & BTRFS_BALANCE_ARGS_CONVERT)
4143 CHECK_APPEND_1ARG("convert=%s,",
4144 btrfs_bg_type_to_raid_name(bargs->target));
4146 if (flags & BTRFS_BALANCE_ARGS_SOFT)
4147 CHECK_APPEND_NOARG("soft,");
4149 if (flags & BTRFS_BALANCE_ARGS_PROFILES) {
4150 btrfs_describe_block_groups(bargs->profiles, tmp_buf,
4152 CHECK_APPEND_1ARG("profiles=%s,", tmp_buf);
4155 if (flags & BTRFS_BALANCE_ARGS_USAGE)
4156 CHECK_APPEND_1ARG("usage=%llu,", bargs->usage);
4158 if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE)
4159 CHECK_APPEND_2ARG("usage=%u..%u,",
4160 bargs->usage_min, bargs->usage_max);
4162 if (flags & BTRFS_BALANCE_ARGS_DEVID)
4163 CHECK_APPEND_1ARG("devid=%llu,", bargs->devid);
4165 if (flags & BTRFS_BALANCE_ARGS_DRANGE)
4166 CHECK_APPEND_2ARG("drange=%llu..%llu,",
4167 bargs->pstart, bargs->pend);
4169 if (flags & BTRFS_BALANCE_ARGS_VRANGE)
4170 CHECK_APPEND_2ARG("vrange=%llu..%llu,",
4171 bargs->vstart, bargs->vend);
4173 if (flags & BTRFS_BALANCE_ARGS_LIMIT)
4174 CHECK_APPEND_1ARG("limit=%llu,", bargs->limit);
4176 if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)
4177 CHECK_APPEND_2ARG("limit=%u..%u,",
4178 bargs->limit_min, bargs->limit_max);
4180 if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE)
4181 CHECK_APPEND_2ARG("stripes=%u..%u,",
4182 bargs->stripes_min, bargs->stripes_max);
4184 #undef CHECK_APPEND_2ARG
4185 #undef CHECK_APPEND_1ARG
4186 #undef CHECK_APPEND_NOARG
4190 if (size_bp < size_buf)
4191 buf[size_buf - size_bp - 1] = '\0'; /* remove last , */
4196 static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info)
4198 u32 size_buf = 1024;
4199 char tmp_buf[192] = {'\0'};
4202 u32 size_bp = size_buf;
4204 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4206 buf = kzalloc(size_buf, GFP_KERNEL);
4212 #define CHECK_APPEND_1ARG(a, v1) \
4214 ret = snprintf(bp, size_bp, (a), (v1)); \
4215 if (ret < 0 || ret >= size_bp) \
4216 goto out_overflow; \
4221 if (bctl->flags & BTRFS_BALANCE_FORCE)
4222 CHECK_APPEND_1ARG("%s", "-f ");
4224 if (bctl->flags & BTRFS_BALANCE_DATA) {
4225 describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf));
4226 CHECK_APPEND_1ARG("-d%s ", tmp_buf);
4229 if (bctl->flags & BTRFS_BALANCE_METADATA) {
4230 describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf));
4231 CHECK_APPEND_1ARG("-m%s ", tmp_buf);
4234 if (bctl->flags & BTRFS_BALANCE_SYSTEM) {
4235 describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf));
4236 CHECK_APPEND_1ARG("-s%s ", tmp_buf);
4239 #undef CHECK_APPEND_1ARG
4243 if (size_bp < size_buf)
4244 buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */
4245 btrfs_info(fs_info, "balance: %s %s",
4246 (bctl->flags & BTRFS_BALANCE_RESUME) ?
4247 "resume" : "start", buf);
4253 * Should be called with balance mutexe held
4255 int btrfs_balance(struct btrfs_fs_info *fs_info,
4256 struct btrfs_balance_control *bctl,
4257 struct btrfs_ioctl_balance_args *bargs)
4259 u64 meta_target, data_target;
4265 bool reducing_redundancy;
4268 if (btrfs_fs_closing(fs_info) ||
4269 atomic_read(&fs_info->balance_pause_req) ||
4270 btrfs_should_cancel_balance(fs_info)) {
4275 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
4276 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
4280 * In case of mixed groups both data and meta should be picked,
4281 * and identical options should be given for both of them.
4283 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
4284 if (mixed && (bctl->flags & allowed)) {
4285 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
4286 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
4287 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
4289 "balance: mixed groups data and metadata options must be the same");
4296 * rw_devices will not change at the moment, device add/delete/replace
4299 num_devices = fs_info->fs_devices->rw_devices;
4302 * SINGLE profile on-disk has no profile bit, but in-memory we have a
4303 * special bit for it, to make it easier to distinguish. Thus we need
4304 * to set it manually, or balance would refuse the profile.
4306 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
4307 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++)
4308 if (num_devices >= btrfs_raid_array[i].devs_min)
4309 allowed |= btrfs_raid_array[i].bg_flag;
4311 if (!validate_convert_profile(fs_info, &bctl->data, allowed, "data") ||
4312 !validate_convert_profile(fs_info, &bctl->meta, allowed, "metadata") ||
4313 !validate_convert_profile(fs_info, &bctl->sys, allowed, "system")) {
4319 * Allow to reduce metadata or system integrity only if force set for
4320 * profiles with redundancy (copies, parity)
4323 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) {
4324 if (btrfs_raid_array[i].ncopies >= 2 ||
4325 btrfs_raid_array[i].tolerated_failures >= 1)
4326 allowed |= btrfs_raid_array[i].bg_flag;
4329 seq = read_seqbegin(&fs_info->profiles_lock);
4331 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4332 (fs_info->avail_system_alloc_bits & allowed) &&
4333 !(bctl->sys.target & allowed)) ||
4334 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4335 (fs_info->avail_metadata_alloc_bits & allowed) &&
4336 !(bctl->meta.target & allowed)))
4337 reducing_redundancy = true;
4339 reducing_redundancy = false;
4341 /* if we're not converting, the target field is uninitialized */
4342 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4343 bctl->meta.target : fs_info->avail_metadata_alloc_bits;
4344 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4345 bctl->data.target : fs_info->avail_data_alloc_bits;
4346 } while (read_seqretry(&fs_info->profiles_lock, seq));
4348 if (reducing_redundancy) {
4349 if (bctl->flags & BTRFS_BALANCE_FORCE) {
4351 "balance: force reducing metadata redundancy");
4354 "balance: reduces metadata redundancy, use --force if you want this");
4360 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <
4361 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {
4363 "balance: metadata profile %s has lower redundancy than data profile %s",
4364 btrfs_bg_type_to_raid_name(meta_target),
4365 btrfs_bg_type_to_raid_name(data_target));
4368 ret = insert_balance_item(fs_info, bctl);
4369 if (ret && ret != -EEXIST)
4372 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
4373 BUG_ON(ret == -EEXIST);
4374 BUG_ON(fs_info->balance_ctl);
4375 spin_lock(&fs_info->balance_lock);
4376 fs_info->balance_ctl = bctl;
4377 spin_unlock(&fs_info->balance_lock);
4379 BUG_ON(ret != -EEXIST);
4380 spin_lock(&fs_info->balance_lock);
4381 update_balance_args(bctl);
4382 spin_unlock(&fs_info->balance_lock);
4385 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4386 set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4387 describe_balance_start_or_resume(fs_info);
4388 mutex_unlock(&fs_info->balance_mutex);
4390 ret = __btrfs_balance(fs_info);
4392 mutex_lock(&fs_info->balance_mutex);
4393 if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req)) {
4394 btrfs_info(fs_info, "balance: paused");
4395 btrfs_exclop_balance(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED);
4398 * Balance can be canceled by:
4400 * - Regular cancel request
4401 * Then ret == -ECANCELED and balance_cancel_req > 0
4403 * - Fatal signal to "btrfs" process
4404 * Either the signal caught by wait_reserve_ticket() and callers
4405 * got -EINTR, or caught by btrfs_should_cancel_balance() and
4407 * Either way, in this case balance_cancel_req = 0, and
4408 * ret == -EINTR or ret == -ECANCELED.
4410 * So here we only check the return value to catch canceled balance.
4412 else if (ret == -ECANCELED || ret == -EINTR)
4413 btrfs_info(fs_info, "balance: canceled");
4415 btrfs_info(fs_info, "balance: ended with status: %d", ret);
4417 clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4420 memset(bargs, 0, sizeof(*bargs));
4421 btrfs_update_ioctl_balance_args(fs_info, bargs);
4424 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
4425 balance_need_close(fs_info)) {
4426 reset_balance_state(fs_info);
4427 btrfs_exclop_finish(fs_info);
4430 wake_up(&fs_info->balance_wait_q);
4434 if (bctl->flags & BTRFS_BALANCE_RESUME)
4435 reset_balance_state(fs_info);
4438 btrfs_exclop_finish(fs_info);
4443 static int balance_kthread(void *data)
4445 struct btrfs_fs_info *fs_info = data;
4448 mutex_lock(&fs_info->balance_mutex);
4449 if (fs_info->balance_ctl)
4450 ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL);
4451 mutex_unlock(&fs_info->balance_mutex);
4456 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
4458 struct task_struct *tsk;
4460 mutex_lock(&fs_info->balance_mutex);
4461 if (!fs_info->balance_ctl) {
4462 mutex_unlock(&fs_info->balance_mutex);
4465 mutex_unlock(&fs_info->balance_mutex);
4467 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
4468 btrfs_info(fs_info, "balance: resume skipped");
4472 spin_lock(&fs_info->super_lock);
4473 ASSERT(fs_info->exclusive_operation == BTRFS_EXCLOP_BALANCE_PAUSED);
4474 fs_info->exclusive_operation = BTRFS_EXCLOP_BALANCE;
4475 spin_unlock(&fs_info->super_lock);
4477 * A ro->rw remount sequence should continue with the paused balance
4478 * regardless of who pauses it, system or the user as of now, so set
4481 spin_lock(&fs_info->balance_lock);
4482 fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME;
4483 spin_unlock(&fs_info->balance_lock);
4485 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
4486 return PTR_ERR_OR_ZERO(tsk);
4489 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
4491 struct btrfs_balance_control *bctl;
4492 struct btrfs_balance_item *item;
4493 struct btrfs_disk_balance_args disk_bargs;
4494 struct btrfs_path *path;
4495 struct extent_buffer *leaf;
4496 struct btrfs_key key;
4499 path = btrfs_alloc_path();
4503 key.objectid = BTRFS_BALANCE_OBJECTID;
4504 key.type = BTRFS_TEMPORARY_ITEM_KEY;
4507 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4510 if (ret > 0) { /* ret = -ENOENT; */
4515 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4521 leaf = path->nodes[0];
4522 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
4524 bctl->flags = btrfs_balance_flags(leaf, item);
4525 bctl->flags |= BTRFS_BALANCE_RESUME;
4527 btrfs_balance_data(leaf, item, &disk_bargs);
4528 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4529 btrfs_balance_meta(leaf, item, &disk_bargs);
4530 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4531 btrfs_balance_sys(leaf, item, &disk_bargs);
4532 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4535 * This should never happen, as the paused balance state is recovered
4536 * during mount without any chance of other exclusive ops to collide.
4538 * This gives the exclusive op status to balance and keeps in paused
4539 * state until user intervention (cancel or umount). If the ownership
4540 * cannot be assigned, show a message but do not fail. The balance
4541 * is in a paused state and must have fs_info::balance_ctl properly
4544 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED))
4546 "balance: cannot set exclusive op status, resume manually");
4548 btrfs_release_path(path);
4550 mutex_lock(&fs_info->balance_mutex);
4551 BUG_ON(fs_info->balance_ctl);
4552 spin_lock(&fs_info->balance_lock);
4553 fs_info->balance_ctl = bctl;
4554 spin_unlock(&fs_info->balance_lock);
4555 mutex_unlock(&fs_info->balance_mutex);
4557 btrfs_free_path(path);
4561 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4565 mutex_lock(&fs_info->balance_mutex);
4566 if (!fs_info->balance_ctl) {
4567 mutex_unlock(&fs_info->balance_mutex);
4571 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4572 atomic_inc(&fs_info->balance_pause_req);
4573 mutex_unlock(&fs_info->balance_mutex);
4575 wait_event(fs_info->balance_wait_q,
4576 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4578 mutex_lock(&fs_info->balance_mutex);
4579 /* we are good with balance_ctl ripped off from under us */
4580 BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4581 atomic_dec(&fs_info->balance_pause_req);
4586 mutex_unlock(&fs_info->balance_mutex);
4590 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4592 mutex_lock(&fs_info->balance_mutex);
4593 if (!fs_info->balance_ctl) {
4594 mutex_unlock(&fs_info->balance_mutex);
4599 * A paused balance with the item stored on disk can be resumed at
4600 * mount time if the mount is read-write. Otherwise it's still paused
4601 * and we must not allow cancelling as it deletes the item.
4603 if (sb_rdonly(fs_info->sb)) {
4604 mutex_unlock(&fs_info->balance_mutex);
4608 atomic_inc(&fs_info->balance_cancel_req);
4610 * if we are running just wait and return, balance item is
4611 * deleted in btrfs_balance in this case
4613 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4614 mutex_unlock(&fs_info->balance_mutex);
4615 wait_event(fs_info->balance_wait_q,
4616 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4617 mutex_lock(&fs_info->balance_mutex);
4619 mutex_unlock(&fs_info->balance_mutex);
4621 * Lock released to allow other waiters to continue, we'll
4622 * reexamine the status again.
4624 mutex_lock(&fs_info->balance_mutex);
4626 if (fs_info->balance_ctl) {
4627 reset_balance_state(fs_info);
4628 btrfs_exclop_finish(fs_info);
4629 btrfs_info(fs_info, "balance: canceled");
4633 BUG_ON(fs_info->balance_ctl ||
4634 test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4635 atomic_dec(&fs_info->balance_cancel_req);
4636 mutex_unlock(&fs_info->balance_mutex);
4640 int btrfs_uuid_scan_kthread(void *data)
4642 struct btrfs_fs_info *fs_info = data;
4643 struct btrfs_root *root = fs_info->tree_root;
4644 struct btrfs_key key;
4645 struct btrfs_path *path = NULL;
4647 struct extent_buffer *eb;
4649 struct btrfs_root_item root_item;
4651 struct btrfs_trans_handle *trans = NULL;
4652 bool closing = false;
4654 path = btrfs_alloc_path();
4661 key.type = BTRFS_ROOT_ITEM_KEY;
4665 if (btrfs_fs_closing(fs_info)) {
4669 ret = btrfs_search_forward(root, &key, path,
4670 BTRFS_OLDEST_GENERATION);
4677 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4678 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4679 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4680 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4683 eb = path->nodes[0];
4684 slot = path->slots[0];
4685 item_size = btrfs_item_size(eb, slot);
4686 if (item_size < sizeof(root_item))
4689 read_extent_buffer(eb, &root_item,
4690 btrfs_item_ptr_offset(eb, slot),
4691 (int)sizeof(root_item));
4692 if (btrfs_root_refs(&root_item) == 0)
4695 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4696 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4700 btrfs_release_path(path);
4702 * 1 - subvol uuid item
4703 * 1 - received_subvol uuid item
4705 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4706 if (IS_ERR(trans)) {
4707 ret = PTR_ERR(trans);
4715 btrfs_release_path(path);
4716 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4717 ret = btrfs_uuid_tree_add(trans, root_item.uuid,
4718 BTRFS_UUID_KEY_SUBVOL,
4721 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4727 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4728 ret = btrfs_uuid_tree_add(trans,
4729 root_item.received_uuid,
4730 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4733 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4740 btrfs_release_path(path);
4742 ret = btrfs_end_transaction(trans);
4748 if (key.offset < (u64)-1) {
4750 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4752 key.type = BTRFS_ROOT_ITEM_KEY;
4753 } else if (key.objectid < (u64)-1) {
4755 key.type = BTRFS_ROOT_ITEM_KEY;
4764 btrfs_free_path(path);
4765 if (trans && !IS_ERR(trans))
4766 btrfs_end_transaction(trans);
4768 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4770 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4771 up(&fs_info->uuid_tree_rescan_sem);
4775 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4777 struct btrfs_trans_handle *trans;
4778 struct btrfs_root *tree_root = fs_info->tree_root;
4779 struct btrfs_root *uuid_root;
4780 struct task_struct *task;
4787 trans = btrfs_start_transaction(tree_root, 2);
4789 return PTR_ERR(trans);
4791 uuid_root = btrfs_create_tree(trans, BTRFS_UUID_TREE_OBJECTID);
4792 if (IS_ERR(uuid_root)) {
4793 ret = PTR_ERR(uuid_root);
4794 btrfs_abort_transaction(trans, ret);
4795 btrfs_end_transaction(trans);
4799 fs_info->uuid_root = uuid_root;
4801 ret = btrfs_commit_transaction(trans);
4805 down(&fs_info->uuid_tree_rescan_sem);
4806 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4808 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4809 btrfs_warn(fs_info, "failed to start uuid_scan task");
4810 up(&fs_info->uuid_tree_rescan_sem);
4811 return PTR_ERR(task);
4818 * shrinking a device means finding all of the device extents past
4819 * the new size, and then following the back refs to the chunks.
4820 * The chunk relocation code actually frees the device extent
4822 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4824 struct btrfs_fs_info *fs_info = device->fs_info;
4825 struct btrfs_root *root = fs_info->dev_root;
4826 struct btrfs_trans_handle *trans;
4827 struct btrfs_dev_extent *dev_extent = NULL;
4828 struct btrfs_path *path;
4834 bool retried = false;
4835 struct extent_buffer *l;
4836 struct btrfs_key key;
4837 struct btrfs_super_block *super_copy = fs_info->super_copy;
4838 u64 old_total = btrfs_super_total_bytes(super_copy);
4839 u64 old_size = btrfs_device_get_total_bytes(device);
4843 new_size = round_down(new_size, fs_info->sectorsize);
4845 diff = round_down(old_size - new_size, fs_info->sectorsize);
4847 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4850 path = btrfs_alloc_path();
4854 path->reada = READA_BACK;
4856 trans = btrfs_start_transaction(root, 0);
4857 if (IS_ERR(trans)) {
4858 btrfs_free_path(path);
4859 return PTR_ERR(trans);
4862 mutex_lock(&fs_info->chunk_mutex);
4864 btrfs_device_set_total_bytes(device, new_size);
4865 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4866 device->fs_devices->total_rw_bytes -= diff;
4867 atomic64_sub(diff, &fs_info->free_chunk_space);
4871 * Once the device's size has been set to the new size, ensure all
4872 * in-memory chunks are synced to disk so that the loop below sees them
4873 * and relocates them accordingly.
4875 if (contains_pending_extent(device, &start, diff)) {
4876 mutex_unlock(&fs_info->chunk_mutex);
4877 ret = btrfs_commit_transaction(trans);
4881 mutex_unlock(&fs_info->chunk_mutex);
4882 btrfs_end_transaction(trans);
4886 key.objectid = device->devid;
4887 key.offset = (u64)-1;
4888 key.type = BTRFS_DEV_EXTENT_KEY;
4891 mutex_lock(&fs_info->reclaim_bgs_lock);
4892 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4894 mutex_unlock(&fs_info->reclaim_bgs_lock);
4898 ret = btrfs_previous_item(root, path, 0, key.type);
4900 mutex_unlock(&fs_info->reclaim_bgs_lock);
4904 btrfs_release_path(path);
4909 slot = path->slots[0];
4910 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4912 if (key.objectid != device->devid) {
4913 mutex_unlock(&fs_info->reclaim_bgs_lock);
4914 btrfs_release_path(path);
4918 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4919 length = btrfs_dev_extent_length(l, dev_extent);
4921 if (key.offset + length <= new_size) {
4922 mutex_unlock(&fs_info->reclaim_bgs_lock);
4923 btrfs_release_path(path);
4927 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4928 btrfs_release_path(path);
4931 * We may be relocating the only data chunk we have,
4932 * which could potentially end up with losing data's
4933 * raid profile, so lets allocate an empty one in
4936 ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset);
4938 mutex_unlock(&fs_info->reclaim_bgs_lock);
4942 ret = btrfs_relocate_chunk(fs_info, chunk_offset);
4943 mutex_unlock(&fs_info->reclaim_bgs_lock);
4944 if (ret == -ENOSPC) {
4947 if (ret == -ETXTBSY) {
4949 "could not shrink block group %llu due to active swapfile",
4954 } while (key.offset-- > 0);
4956 if (failed && !retried) {
4960 } else if (failed && retried) {
4965 /* Shrinking succeeded, else we would be at "done". */
4966 trans = btrfs_start_transaction(root, 0);
4967 if (IS_ERR(trans)) {
4968 ret = PTR_ERR(trans);
4972 mutex_lock(&fs_info->chunk_mutex);
4973 /* Clear all state bits beyond the shrunk device size */
4974 clear_extent_bits(&device->alloc_state, new_size, (u64)-1,
4977 btrfs_device_set_disk_total_bytes(device, new_size);
4978 if (list_empty(&device->post_commit_list))
4979 list_add_tail(&device->post_commit_list,
4980 &trans->transaction->dev_update_list);
4982 WARN_ON(diff > old_total);
4983 btrfs_set_super_total_bytes(super_copy,
4984 round_down(old_total - diff, fs_info->sectorsize));
4985 mutex_unlock(&fs_info->chunk_mutex);
4987 btrfs_reserve_chunk_metadata(trans, false);
4988 /* Now btrfs_update_device() will change the on-disk size. */
4989 ret = btrfs_update_device(trans, device);
4990 btrfs_trans_release_chunk_metadata(trans);
4992 btrfs_abort_transaction(trans, ret);
4993 btrfs_end_transaction(trans);
4995 ret = btrfs_commit_transaction(trans);
4998 btrfs_free_path(path);
5000 mutex_lock(&fs_info->chunk_mutex);
5001 btrfs_device_set_total_bytes(device, old_size);
5002 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
5003 device->fs_devices->total_rw_bytes += diff;
5004 atomic64_add(diff, &fs_info->free_chunk_space);
5005 mutex_unlock(&fs_info->chunk_mutex);
5010 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
5011 struct btrfs_key *key,
5012 struct btrfs_chunk *chunk, int item_size)
5014 struct btrfs_super_block *super_copy = fs_info->super_copy;
5015 struct btrfs_disk_key disk_key;
5019 lockdep_assert_held(&fs_info->chunk_mutex);
5021 array_size = btrfs_super_sys_array_size(super_copy);
5022 if (array_size + item_size + sizeof(disk_key)
5023 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
5026 ptr = super_copy->sys_chunk_array + array_size;
5027 btrfs_cpu_key_to_disk(&disk_key, key);
5028 memcpy(ptr, &disk_key, sizeof(disk_key));
5029 ptr += sizeof(disk_key);
5030 memcpy(ptr, chunk, item_size);
5031 item_size += sizeof(disk_key);
5032 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
5038 * sort the devices in descending order by max_avail, total_avail
5040 static int btrfs_cmp_device_info(const void *a, const void *b)
5042 const struct btrfs_device_info *di_a = a;
5043 const struct btrfs_device_info *di_b = b;
5045 if (di_a->max_avail > di_b->max_avail)
5047 if (di_a->max_avail < di_b->max_avail)
5049 if (di_a->total_avail > di_b->total_avail)
5051 if (di_a->total_avail < di_b->total_avail)
5056 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
5058 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
5061 btrfs_set_fs_incompat(info, RAID56);
5064 static void check_raid1c34_incompat_flag(struct btrfs_fs_info *info, u64 type)
5066 if (!(type & (BTRFS_BLOCK_GROUP_RAID1C3 | BTRFS_BLOCK_GROUP_RAID1C4)))
5069 btrfs_set_fs_incompat(info, RAID1C34);
5073 * Structure used internally for btrfs_create_chunk() function.
5074 * Wraps needed parameters.
5076 struct alloc_chunk_ctl {
5079 /* Total number of stripes to allocate */
5081 /* sub_stripes info for map */
5083 /* Stripes per device */
5085 /* Maximum number of devices to use */
5087 /* Minimum number of devices to use */
5089 /* ndevs has to be a multiple of this */
5091 /* Number of copies */
5093 /* Number of stripes worth of bytes to store parity information */
5095 u64 max_stripe_size;
5103 static void init_alloc_chunk_ctl_policy_regular(
5104 struct btrfs_fs_devices *fs_devices,
5105 struct alloc_chunk_ctl *ctl)
5107 u64 type = ctl->type;
5109 if (type & BTRFS_BLOCK_GROUP_DATA) {
5110 ctl->max_stripe_size = SZ_1G;
5111 ctl->max_chunk_size = BTRFS_MAX_DATA_CHUNK_SIZE;
5112 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
5113 /* For larger filesystems, use larger metadata chunks */
5114 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
5115 ctl->max_stripe_size = SZ_1G;
5117 ctl->max_stripe_size = SZ_256M;
5118 ctl->max_chunk_size = ctl->max_stripe_size;
5119 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
5120 ctl->max_stripe_size = SZ_32M;
5121 ctl->max_chunk_size = 2 * ctl->max_stripe_size;
5122 ctl->devs_max = min_t(int, ctl->devs_max,
5123 BTRFS_MAX_DEVS_SYS_CHUNK);
5128 /* We don't want a chunk larger than 10% of writable space */
5129 ctl->max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
5130 ctl->max_chunk_size);
5131 ctl->dev_extent_min = BTRFS_STRIPE_LEN * ctl->dev_stripes;
5134 static void init_alloc_chunk_ctl_policy_zoned(
5135 struct btrfs_fs_devices *fs_devices,
5136 struct alloc_chunk_ctl *ctl)
5138 u64 zone_size = fs_devices->fs_info->zone_size;
5140 int min_num_stripes = ctl->devs_min * ctl->dev_stripes;
5141 int min_data_stripes = (min_num_stripes - ctl->nparity) / ctl->ncopies;
5142 u64 min_chunk_size = min_data_stripes * zone_size;
5143 u64 type = ctl->type;
5145 ctl->max_stripe_size = zone_size;
5146 if (type & BTRFS_BLOCK_GROUP_DATA) {
5147 ctl->max_chunk_size = round_down(BTRFS_MAX_DATA_CHUNK_SIZE,
5149 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
5150 ctl->max_chunk_size = ctl->max_stripe_size;
5151 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
5152 ctl->max_chunk_size = 2 * ctl->max_stripe_size;
5153 ctl->devs_max = min_t(int, ctl->devs_max,
5154 BTRFS_MAX_DEVS_SYS_CHUNK);
5159 /* We don't want a chunk larger than 10% of writable space */
5160 limit = max(round_down(div_factor(fs_devices->total_rw_bytes, 1),
5163 ctl->max_chunk_size = min(limit, ctl->max_chunk_size);
5164 ctl->dev_extent_min = zone_size * ctl->dev_stripes;
5167 static void init_alloc_chunk_ctl(struct btrfs_fs_devices *fs_devices,
5168 struct alloc_chunk_ctl *ctl)
5170 int index = btrfs_bg_flags_to_raid_index(ctl->type);
5172 ctl->sub_stripes = btrfs_raid_array[index].sub_stripes;
5173 ctl->dev_stripes = btrfs_raid_array[index].dev_stripes;
5174 ctl->devs_max = btrfs_raid_array[index].devs_max;
5176 ctl->devs_max = BTRFS_MAX_DEVS(fs_devices->fs_info);
5177 ctl->devs_min = btrfs_raid_array[index].devs_min;
5178 ctl->devs_increment = btrfs_raid_array[index].devs_increment;
5179 ctl->ncopies = btrfs_raid_array[index].ncopies;
5180 ctl->nparity = btrfs_raid_array[index].nparity;
5183 switch (fs_devices->chunk_alloc_policy) {
5184 case BTRFS_CHUNK_ALLOC_REGULAR:
5185 init_alloc_chunk_ctl_policy_regular(fs_devices, ctl);
5187 case BTRFS_CHUNK_ALLOC_ZONED:
5188 init_alloc_chunk_ctl_policy_zoned(fs_devices, ctl);
5195 static int gather_device_info(struct btrfs_fs_devices *fs_devices,
5196 struct alloc_chunk_ctl *ctl,
5197 struct btrfs_device_info *devices_info)
5199 struct btrfs_fs_info *info = fs_devices->fs_info;
5200 struct btrfs_device *device;
5202 u64 dev_extent_want = ctl->max_stripe_size * ctl->dev_stripes;
5209 * in the first pass through the devices list, we gather information
5210 * about the available holes on each device.
5212 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
5213 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
5215 "BTRFS: read-only device in alloc_list\n");
5219 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
5220 &device->dev_state) ||
5221 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
5224 if (device->total_bytes > device->bytes_used)
5225 total_avail = device->total_bytes - device->bytes_used;
5229 /* If there is no space on this device, skip it. */
5230 if (total_avail < ctl->dev_extent_min)
5233 ret = find_free_dev_extent(device, dev_extent_want, &dev_offset,
5235 if (ret && ret != -ENOSPC)
5239 max_avail = dev_extent_want;
5241 if (max_avail < ctl->dev_extent_min) {
5242 if (btrfs_test_opt(info, ENOSPC_DEBUG))
5244 "%s: devid %llu has no free space, have=%llu want=%llu",
5245 __func__, device->devid, max_avail,
5246 ctl->dev_extent_min);
5250 if (ndevs == fs_devices->rw_devices) {
5251 WARN(1, "%s: found more than %llu devices\n",
5252 __func__, fs_devices->rw_devices);
5255 devices_info[ndevs].dev_offset = dev_offset;
5256 devices_info[ndevs].max_avail = max_avail;
5257 devices_info[ndevs].total_avail = total_avail;
5258 devices_info[ndevs].dev = device;
5264 * now sort the devices by hole size / available space
5266 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
5267 btrfs_cmp_device_info, NULL);
5272 static int decide_stripe_size_regular(struct alloc_chunk_ctl *ctl,
5273 struct btrfs_device_info *devices_info)
5275 /* Number of stripes that count for block group size */
5279 * The primary goal is to maximize the number of stripes, so use as
5280 * many devices as possible, even if the stripes are not maximum sized.
5282 * The DUP profile stores more than one stripe per device, the
5283 * max_avail is the total size so we have to adjust.
5285 ctl->stripe_size = div_u64(devices_info[ctl->ndevs - 1].max_avail,
5287 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5289 /* This will have to be fixed for RAID1 and RAID10 over more drives */
5290 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5293 * Use the number of data stripes to figure out how big this chunk is
5294 * really going to be in terms of logical address space, and compare
5295 * that answer with the max chunk size. If it's higher, we try to
5296 * reduce stripe_size.
5298 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) {
5300 * Reduce stripe_size, round it up to a 16MB boundary again and
5301 * then use it, unless it ends up being even bigger than the
5302 * previous value we had already.
5304 ctl->stripe_size = min(round_up(div_u64(ctl->max_chunk_size,
5305 data_stripes), SZ_16M),
5309 /* Align to BTRFS_STRIPE_LEN */
5310 ctl->stripe_size = round_down(ctl->stripe_size, BTRFS_STRIPE_LEN);
5311 ctl->chunk_size = ctl->stripe_size * data_stripes;
5316 static int decide_stripe_size_zoned(struct alloc_chunk_ctl *ctl,
5317 struct btrfs_device_info *devices_info)
5319 u64 zone_size = devices_info[0].dev->zone_info->zone_size;
5320 /* Number of stripes that count for block group size */
5324 * It should hold because:
5325 * dev_extent_min == dev_extent_want == zone_size * dev_stripes
5327 ASSERT(devices_info[ctl->ndevs - 1].max_avail == ctl->dev_extent_min);
5329 ctl->stripe_size = zone_size;
5330 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5331 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5333 /* stripe_size is fixed in zoned filesysmte. Reduce ndevs instead. */
5334 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) {
5335 ctl->ndevs = div_u64(div_u64(ctl->max_chunk_size * ctl->ncopies,
5336 ctl->stripe_size) + ctl->nparity,
5338 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5339 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5340 ASSERT(ctl->stripe_size * data_stripes <= ctl->max_chunk_size);
5343 ctl->chunk_size = ctl->stripe_size * data_stripes;
5348 static int decide_stripe_size(struct btrfs_fs_devices *fs_devices,
5349 struct alloc_chunk_ctl *ctl,
5350 struct btrfs_device_info *devices_info)
5352 struct btrfs_fs_info *info = fs_devices->fs_info;
5355 * Round down to number of usable stripes, devs_increment can be any
5356 * number so we can't use round_down() that requires power of 2, while
5357 * rounddown is safe.
5359 ctl->ndevs = rounddown(ctl->ndevs, ctl->devs_increment);
5361 if (ctl->ndevs < ctl->devs_min) {
5362 if (btrfs_test_opt(info, ENOSPC_DEBUG)) {
5364 "%s: not enough devices with free space: have=%d minimum required=%d",
5365 __func__, ctl->ndevs, ctl->devs_min);
5370 ctl->ndevs = min(ctl->ndevs, ctl->devs_max);
5372 switch (fs_devices->chunk_alloc_policy) {
5373 case BTRFS_CHUNK_ALLOC_REGULAR:
5374 return decide_stripe_size_regular(ctl, devices_info);
5375 case BTRFS_CHUNK_ALLOC_ZONED:
5376 return decide_stripe_size_zoned(ctl, devices_info);
5382 static struct btrfs_block_group *create_chunk(struct btrfs_trans_handle *trans,
5383 struct alloc_chunk_ctl *ctl,
5384 struct btrfs_device_info *devices_info)
5386 struct btrfs_fs_info *info = trans->fs_info;
5387 struct map_lookup *map = NULL;
5388 struct extent_map_tree *em_tree;
5389 struct btrfs_block_group *block_group;
5390 struct extent_map *em;
5391 u64 start = ctl->start;
5392 u64 type = ctl->type;
5397 map = kmalloc(map_lookup_size(ctl->num_stripes), GFP_NOFS);
5399 return ERR_PTR(-ENOMEM);
5400 map->num_stripes = ctl->num_stripes;
5402 for (i = 0; i < ctl->ndevs; ++i) {
5403 for (j = 0; j < ctl->dev_stripes; ++j) {
5404 int s = i * ctl->dev_stripes + j;
5405 map->stripes[s].dev = devices_info[i].dev;
5406 map->stripes[s].physical = devices_info[i].dev_offset +
5407 j * ctl->stripe_size;
5410 map->stripe_len = BTRFS_STRIPE_LEN;
5411 map->io_align = BTRFS_STRIPE_LEN;
5412 map->io_width = BTRFS_STRIPE_LEN;
5414 map->sub_stripes = ctl->sub_stripes;
5416 trace_btrfs_chunk_alloc(info, map, start, ctl->chunk_size);
5418 em = alloc_extent_map();
5421 return ERR_PTR(-ENOMEM);
5423 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
5424 em->map_lookup = map;
5426 em->len = ctl->chunk_size;
5427 em->block_start = 0;
5428 em->block_len = em->len;
5429 em->orig_block_len = ctl->stripe_size;
5431 em_tree = &info->mapping_tree;
5432 write_lock(&em_tree->lock);
5433 ret = add_extent_mapping(em_tree, em, 0);
5435 write_unlock(&em_tree->lock);
5436 free_extent_map(em);
5437 return ERR_PTR(ret);
5439 write_unlock(&em_tree->lock);
5441 block_group = btrfs_make_block_group(trans, 0, type, start, ctl->chunk_size);
5442 if (IS_ERR(block_group))
5443 goto error_del_extent;
5445 for (i = 0; i < map->num_stripes; i++) {
5446 struct btrfs_device *dev = map->stripes[i].dev;
5448 btrfs_device_set_bytes_used(dev,
5449 dev->bytes_used + ctl->stripe_size);
5450 if (list_empty(&dev->post_commit_list))
5451 list_add_tail(&dev->post_commit_list,
5452 &trans->transaction->dev_update_list);
5455 atomic64_sub(ctl->stripe_size * map->num_stripes,
5456 &info->free_chunk_space);
5458 free_extent_map(em);
5459 check_raid56_incompat_flag(info, type);
5460 check_raid1c34_incompat_flag(info, type);
5465 write_lock(&em_tree->lock);
5466 remove_extent_mapping(em_tree, em);
5467 write_unlock(&em_tree->lock);
5469 /* One for our allocation */
5470 free_extent_map(em);
5471 /* One for the tree reference */
5472 free_extent_map(em);
5477 struct btrfs_block_group *btrfs_create_chunk(struct btrfs_trans_handle *trans,
5480 struct btrfs_fs_info *info = trans->fs_info;
5481 struct btrfs_fs_devices *fs_devices = info->fs_devices;
5482 struct btrfs_device_info *devices_info = NULL;
5483 struct alloc_chunk_ctl ctl;
5484 struct btrfs_block_group *block_group;
5487 lockdep_assert_held(&info->chunk_mutex);
5489 if (!alloc_profile_is_valid(type, 0)) {
5491 return ERR_PTR(-EINVAL);
5494 if (list_empty(&fs_devices->alloc_list)) {
5495 if (btrfs_test_opt(info, ENOSPC_DEBUG))
5496 btrfs_debug(info, "%s: no writable device", __func__);
5497 return ERR_PTR(-ENOSPC);
5500 if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
5501 btrfs_err(info, "invalid chunk type 0x%llx requested", type);
5503 return ERR_PTR(-EINVAL);
5506 ctl.start = find_next_chunk(info);
5508 init_alloc_chunk_ctl(fs_devices, &ctl);
5510 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
5513 return ERR_PTR(-ENOMEM);
5515 ret = gather_device_info(fs_devices, &ctl, devices_info);
5517 block_group = ERR_PTR(ret);
5521 ret = decide_stripe_size(fs_devices, &ctl, devices_info);
5523 block_group = ERR_PTR(ret);
5527 block_group = create_chunk(trans, &ctl, devices_info);
5530 kfree(devices_info);
5535 * This function, btrfs_chunk_alloc_add_chunk_item(), typically belongs to the
5536 * phase 1 of chunk allocation. It belongs to phase 2 only when allocating system
5539 * See the comment at btrfs_chunk_alloc() for details about the chunk allocation
5542 int btrfs_chunk_alloc_add_chunk_item(struct btrfs_trans_handle *trans,
5543 struct btrfs_block_group *bg)
5545 struct btrfs_fs_info *fs_info = trans->fs_info;
5546 struct btrfs_root *chunk_root = fs_info->chunk_root;
5547 struct btrfs_key key;
5548 struct btrfs_chunk *chunk;
5549 struct btrfs_stripe *stripe;
5550 struct extent_map *em;
5551 struct map_lookup *map;
5557 * We take the chunk_mutex for 2 reasons:
5559 * 1) Updates and insertions in the chunk btree must be done while holding
5560 * the chunk_mutex, as well as updating the system chunk array in the
5561 * superblock. See the comment on top of btrfs_chunk_alloc() for the
5564 * 2) To prevent races with the final phase of a device replace operation
5565 * that replaces the device object associated with the map's stripes,
5566 * because the device object's id can change at any time during that
5567 * final phase of the device replace operation
5568 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the
5569 * replaced device and then see it with an ID of BTRFS_DEV_REPLACE_DEVID,
5570 * which would cause a failure when updating the device item, which does
5571 * not exists, or persisting a stripe of the chunk item with such ID.
5572 * Here we can't use the device_list_mutex because our caller already
5573 * has locked the chunk_mutex, and the final phase of device replace
5574 * acquires both mutexes - first the device_list_mutex and then the
5575 * chunk_mutex. Using any of those two mutexes protects us from a
5576 * concurrent device replace.
5578 lockdep_assert_held(&fs_info->chunk_mutex);
5580 em = btrfs_get_chunk_map(fs_info, bg->start, bg->length);
5583 btrfs_abort_transaction(trans, ret);
5587 map = em->map_lookup;
5588 item_size = btrfs_chunk_item_size(map->num_stripes);
5590 chunk = kzalloc(item_size, GFP_NOFS);
5593 btrfs_abort_transaction(trans, ret);
5597 for (i = 0; i < map->num_stripes; i++) {
5598 struct btrfs_device *device = map->stripes[i].dev;
5600 ret = btrfs_update_device(trans, device);
5605 stripe = &chunk->stripe;
5606 for (i = 0; i < map->num_stripes; i++) {
5607 struct btrfs_device *device = map->stripes[i].dev;
5608 const u64 dev_offset = map->stripes[i].physical;
5610 btrfs_set_stack_stripe_devid(stripe, device->devid);
5611 btrfs_set_stack_stripe_offset(stripe, dev_offset);
5612 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
5616 btrfs_set_stack_chunk_length(chunk, bg->length);
5617 btrfs_set_stack_chunk_owner(chunk, BTRFS_EXTENT_TREE_OBJECTID);
5618 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
5619 btrfs_set_stack_chunk_type(chunk, map->type);
5620 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
5621 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
5622 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
5623 btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
5624 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
5626 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
5627 key.type = BTRFS_CHUNK_ITEM_KEY;
5628 key.offset = bg->start;
5630 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
5634 bg->chunk_item_inserted = 1;
5636 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
5637 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
5644 free_extent_map(em);
5648 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans)
5650 struct btrfs_fs_info *fs_info = trans->fs_info;
5652 struct btrfs_block_group *meta_bg;
5653 struct btrfs_block_group *sys_bg;
5656 * When adding a new device for sprouting, the seed device is read-only
5657 * so we must first allocate a metadata and a system chunk. But before
5658 * adding the block group items to the extent, device and chunk btrees,
5661 * 1) Create both chunks without doing any changes to the btrees, as
5662 * otherwise we would get -ENOSPC since the block groups from the
5663 * seed device are read-only;
5665 * 2) Add the device item for the new sprout device - finishing the setup
5666 * of a new block group requires updating the device item in the chunk
5667 * btree, so it must exist when we attempt to do it. The previous step
5668 * ensures this does not fail with -ENOSPC.
5670 * After that we can add the block group items to their btrees:
5671 * update existing device item in the chunk btree, add a new block group
5672 * item to the extent btree, add a new chunk item to the chunk btree and
5673 * finally add the new device extent items to the devices btree.
5676 alloc_profile = btrfs_metadata_alloc_profile(fs_info);
5677 meta_bg = btrfs_create_chunk(trans, alloc_profile);
5678 if (IS_ERR(meta_bg))
5679 return PTR_ERR(meta_bg);
5681 alloc_profile = btrfs_system_alloc_profile(fs_info);
5682 sys_bg = btrfs_create_chunk(trans, alloc_profile);
5684 return PTR_ERR(sys_bg);
5689 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
5691 const int index = btrfs_bg_flags_to_raid_index(map->type);
5693 return btrfs_raid_array[index].tolerated_failures;
5696 bool btrfs_chunk_writeable(struct btrfs_fs_info *fs_info, u64 chunk_offset)
5698 struct extent_map *em;
5699 struct map_lookup *map;
5704 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
5708 map = em->map_lookup;
5709 for (i = 0; i < map->num_stripes; i++) {
5710 if (test_bit(BTRFS_DEV_STATE_MISSING,
5711 &map->stripes[i].dev->dev_state)) {
5715 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE,
5716 &map->stripes[i].dev->dev_state)) {
5723 * If the number of missing devices is larger than max errors, we can
5724 * not write the data into that chunk successfully.
5726 if (miss_ndevs > btrfs_chunk_max_errors(map))
5729 free_extent_map(em);
5733 void btrfs_mapping_tree_free(struct extent_map_tree *tree)
5735 struct extent_map *em;
5738 write_lock(&tree->lock);
5739 em = lookup_extent_mapping(tree, 0, (u64)-1);
5741 remove_extent_mapping(tree, em);
5742 write_unlock(&tree->lock);
5746 free_extent_map(em);
5747 /* once for the tree */
5748 free_extent_map(em);
5752 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5754 struct extent_map *em;
5755 struct map_lookup *map;
5758 em = btrfs_get_chunk_map(fs_info, logical, len);
5761 * We could return errors for these cases, but that could get
5762 * ugly and we'd probably do the same thing which is just not do
5763 * anything else and exit, so return 1 so the callers don't try
5764 * to use other copies.
5768 map = em->map_lookup;
5769 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1_MASK))
5770 ret = map->num_stripes;
5771 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5772 ret = map->sub_stripes;
5773 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5775 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5777 * There could be two corrupted data stripes, we need
5778 * to loop retry in order to rebuild the correct data.
5780 * Fail a stripe at a time on every retry except the
5781 * stripe under reconstruction.
5783 ret = map->num_stripes;
5786 free_extent_map(em);
5788 down_read(&fs_info->dev_replace.rwsem);
5789 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) &&
5790 fs_info->dev_replace.tgtdev)
5792 up_read(&fs_info->dev_replace.rwsem);
5797 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
5800 struct extent_map *em;
5801 struct map_lookup *map;
5802 unsigned long len = fs_info->sectorsize;
5804 em = btrfs_get_chunk_map(fs_info, logical, len);
5806 if (!WARN_ON(IS_ERR(em))) {
5807 map = em->map_lookup;
5808 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5809 len = map->stripe_len * nr_data_stripes(map);
5810 free_extent_map(em);
5815 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5817 struct extent_map *em;
5818 struct map_lookup *map;
5821 em = btrfs_get_chunk_map(fs_info, logical, len);
5823 if(!WARN_ON(IS_ERR(em))) {
5824 map = em->map_lookup;
5825 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5827 free_extent_map(em);
5832 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5833 struct map_lookup *map, int first,
5834 int dev_replace_is_ongoing)
5838 int preferred_mirror;
5840 struct btrfs_device *srcdev;
5843 (BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10)));
5845 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5846 num_stripes = map->sub_stripes;
5848 num_stripes = map->num_stripes;
5850 switch (fs_info->fs_devices->read_policy) {
5852 /* Shouldn't happen, just warn and use pid instead of failing */
5853 btrfs_warn_rl(fs_info,
5854 "unknown read_policy type %u, reset to pid",
5855 fs_info->fs_devices->read_policy);
5856 fs_info->fs_devices->read_policy = BTRFS_READ_POLICY_PID;
5858 case BTRFS_READ_POLICY_PID:
5859 preferred_mirror = first + (current->pid % num_stripes);
5863 if (dev_replace_is_ongoing &&
5864 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5865 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5866 srcdev = fs_info->dev_replace.srcdev;
5871 * try to avoid the drive that is the source drive for a
5872 * dev-replace procedure, only choose it if no other non-missing
5873 * mirror is available
5875 for (tolerance = 0; tolerance < 2; tolerance++) {
5876 if (map->stripes[preferred_mirror].dev->bdev &&
5877 (tolerance || map->stripes[preferred_mirror].dev != srcdev))
5878 return preferred_mirror;
5879 for (i = first; i < first + num_stripes; i++) {
5880 if (map->stripes[i].dev->bdev &&
5881 (tolerance || map->stripes[i].dev != srcdev))
5886 /* we couldn't find one that doesn't fail. Just return something
5887 * and the io error handling code will clean up eventually
5889 return preferred_mirror;
5892 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5893 static void sort_parity_stripes(struct btrfs_io_context *bioc, int num_stripes)
5900 for (i = 0; i < num_stripes - 1; i++) {
5901 /* Swap if parity is on a smaller index */
5902 if (bioc->raid_map[i] > bioc->raid_map[i + 1]) {
5903 swap(bioc->stripes[i], bioc->stripes[i + 1]);
5904 swap(bioc->raid_map[i], bioc->raid_map[i + 1]);
5911 static struct btrfs_io_context *alloc_btrfs_io_context(struct btrfs_fs_info *fs_info,
5915 struct btrfs_io_context *bioc = kzalloc(
5916 /* The size of btrfs_io_context */
5917 sizeof(struct btrfs_io_context) +
5918 /* Plus the variable array for the stripes */
5919 sizeof(struct btrfs_io_stripe) * (total_stripes) +
5920 /* Plus the variable array for the tgt dev */
5921 sizeof(int) * (real_stripes) +
5923 * Plus the raid_map, which includes both the tgt dev
5926 sizeof(u64) * (total_stripes),
5927 GFP_NOFS|__GFP_NOFAIL);
5929 atomic_set(&bioc->error, 0);
5930 refcount_set(&bioc->refs, 1);
5932 bioc->fs_info = fs_info;
5933 bioc->tgtdev_map = (int *)(bioc->stripes + total_stripes);
5934 bioc->raid_map = (u64 *)(bioc->tgtdev_map + real_stripes);
5939 void btrfs_get_bioc(struct btrfs_io_context *bioc)
5941 WARN_ON(!refcount_read(&bioc->refs));
5942 refcount_inc(&bioc->refs);
5945 void btrfs_put_bioc(struct btrfs_io_context *bioc)
5949 if (refcount_dec_and_test(&bioc->refs))
5953 /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */
5955 * Please note that, discard won't be sent to target device of device
5958 static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info,
5959 u64 logical, u64 *length_ret,
5960 struct btrfs_io_context **bioc_ret)
5962 struct extent_map *em;
5963 struct map_lookup *map;
5964 struct btrfs_io_context *bioc;
5965 u64 length = *length_ret;
5969 u64 stripe_end_offset;
5976 u32 sub_stripes = 0;
5977 u64 stripes_per_dev = 0;
5978 u32 remaining_stripes = 0;
5979 u32 last_stripe = 0;
5983 /* Discard always returns a bioc. */
5986 em = btrfs_get_chunk_map(fs_info, logical, length);
5990 map = em->map_lookup;
5991 /* we don't discard raid56 yet */
5992 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5997 offset = logical - em->start;
5998 length = min_t(u64, em->start + em->len - logical, length);
5999 *length_ret = length;
6001 stripe_len = map->stripe_len;
6003 * stripe_nr counts the total number of stripes we have to stride
6004 * to get to this block
6006 stripe_nr = div64_u64(offset, stripe_len);
6008 /* stripe_offset is the offset of this block in its stripe */
6009 stripe_offset = offset - stripe_nr * stripe_len;
6011 stripe_nr_end = round_up(offset + length, map->stripe_len);
6012 stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len);
6013 stripe_cnt = stripe_nr_end - stripe_nr;
6014 stripe_end_offset = stripe_nr_end * map->stripe_len -
6017 * after this, stripe_nr is the number of stripes on this
6018 * device we have to walk to find the data, and stripe_index is
6019 * the number of our device in the stripe array
6023 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
6024 BTRFS_BLOCK_GROUP_RAID10)) {
6025 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
6028 sub_stripes = map->sub_stripes;
6030 factor = map->num_stripes / sub_stripes;
6031 num_stripes = min_t(u64, map->num_stripes,
6032 sub_stripes * stripe_cnt);
6033 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
6034 stripe_index *= sub_stripes;
6035 stripes_per_dev = div_u64_rem(stripe_cnt, factor,
6036 &remaining_stripes);
6037 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
6038 last_stripe *= sub_stripes;
6039 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1_MASK |
6040 BTRFS_BLOCK_GROUP_DUP)) {
6041 num_stripes = map->num_stripes;
6043 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
6047 bioc = alloc_btrfs_io_context(fs_info, num_stripes, 0);
6053 for (i = 0; i < num_stripes; i++) {
6054 bioc->stripes[i].physical =
6055 map->stripes[stripe_index].physical +
6056 stripe_offset + stripe_nr * map->stripe_len;
6057 bioc->stripes[i].dev = map->stripes[stripe_index].dev;
6059 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
6060 BTRFS_BLOCK_GROUP_RAID10)) {
6061 bioc->stripes[i].length = stripes_per_dev *
6064 if (i / sub_stripes < remaining_stripes)
6065 bioc->stripes[i].length += map->stripe_len;
6068 * Special for the first stripe and
6071 * |-------|...|-------|
6075 if (i < sub_stripes)
6076 bioc->stripes[i].length -= stripe_offset;
6078 if (stripe_index >= last_stripe &&
6079 stripe_index <= (last_stripe +
6081 bioc->stripes[i].length -= stripe_end_offset;
6083 if (i == sub_stripes - 1)
6086 bioc->stripes[i].length = length;
6090 if (stripe_index == map->num_stripes) {
6097 bioc->map_type = map->type;
6098 bioc->num_stripes = num_stripes;
6100 free_extent_map(em);
6105 * In dev-replace case, for repair case (that's the only case where the mirror
6106 * is selected explicitly when calling btrfs_map_block), blocks left of the
6107 * left cursor can also be read from the target drive.
6109 * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the
6111 * For READ, it also needs to be supported using the same mirror number.
6113 * If the requested block is not left of the left cursor, EIO is returned. This
6114 * can happen because btrfs_num_copies() returns one more in the dev-replace
6117 static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info,
6118 u64 logical, u64 length,
6119 u64 srcdev_devid, int *mirror_num,
6122 struct btrfs_io_context *bioc = NULL;
6124 int index_srcdev = 0;
6126 u64 physical_of_found = 0;
6130 ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS,
6131 logical, &length, &bioc, 0, 0);
6133 ASSERT(bioc == NULL);
6137 num_stripes = bioc->num_stripes;
6138 if (*mirror_num > num_stripes) {
6140 * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror,
6141 * that means that the requested area is not left of the left
6144 btrfs_put_bioc(bioc);
6149 * process the rest of the function using the mirror_num of the source
6150 * drive. Therefore look it up first. At the end, patch the device
6151 * pointer to the one of the target drive.
6153 for (i = 0; i < num_stripes; i++) {
6154 if (bioc->stripes[i].dev->devid != srcdev_devid)
6158 * In case of DUP, in order to keep it simple, only add the
6159 * mirror with the lowest physical address
6162 physical_of_found <= bioc->stripes[i].physical)
6167 physical_of_found = bioc->stripes[i].physical;
6170 btrfs_put_bioc(bioc);
6176 *mirror_num = index_srcdev + 1;
6177 *physical = physical_of_found;
6181 static bool is_block_group_to_copy(struct btrfs_fs_info *fs_info, u64 logical)
6183 struct btrfs_block_group *cache;
6186 /* Non zoned filesystem does not use "to_copy" flag */
6187 if (!btrfs_is_zoned(fs_info))
6190 cache = btrfs_lookup_block_group(fs_info, logical);
6192 spin_lock(&cache->lock);
6193 ret = cache->to_copy;
6194 spin_unlock(&cache->lock);
6196 btrfs_put_block_group(cache);
6200 static void handle_ops_on_dev_replace(enum btrfs_map_op op,
6201 struct btrfs_io_context **bioc_ret,
6202 struct btrfs_dev_replace *dev_replace,
6204 int *num_stripes_ret, int *max_errors_ret)
6206 struct btrfs_io_context *bioc = *bioc_ret;
6207 u64 srcdev_devid = dev_replace->srcdev->devid;
6208 int tgtdev_indexes = 0;
6209 int num_stripes = *num_stripes_ret;
6210 int max_errors = *max_errors_ret;
6213 if (op == BTRFS_MAP_WRITE) {
6214 int index_where_to_add;
6217 * A block group which have "to_copy" set will eventually
6218 * copied by dev-replace process. We can avoid cloning IO here.
6220 if (is_block_group_to_copy(dev_replace->srcdev->fs_info, logical))
6224 * duplicate the write operations while the dev replace
6225 * procedure is running. Since the copying of the old disk to
6226 * the new disk takes place at run time while the filesystem is
6227 * mounted writable, the regular write operations to the old
6228 * disk have to be duplicated to go to the new disk as well.
6230 * Note that device->missing is handled by the caller, and that
6231 * the write to the old disk is already set up in the stripes
6234 index_where_to_add = num_stripes;
6235 for (i = 0; i < num_stripes; i++) {
6236 if (bioc->stripes[i].dev->devid == srcdev_devid) {
6237 /* write to new disk, too */
6238 struct btrfs_io_stripe *new =
6239 bioc->stripes + index_where_to_add;
6240 struct btrfs_io_stripe *old =
6243 new->physical = old->physical;
6244 new->length = old->length;
6245 new->dev = dev_replace->tgtdev;
6246 bioc->tgtdev_map[i] = index_where_to_add;
6247 index_where_to_add++;
6252 num_stripes = index_where_to_add;
6253 } else if (op == BTRFS_MAP_GET_READ_MIRRORS) {
6254 int index_srcdev = 0;
6256 u64 physical_of_found = 0;
6259 * During the dev-replace procedure, the target drive can also
6260 * be used to read data in case it is needed to repair a corrupt
6261 * block elsewhere. This is possible if the requested area is
6262 * left of the left cursor. In this area, the target drive is a
6263 * full copy of the source drive.
6265 for (i = 0; i < num_stripes; i++) {
6266 if (bioc->stripes[i].dev->devid == srcdev_devid) {
6268 * In case of DUP, in order to keep it simple,
6269 * only add the mirror with the lowest physical
6273 physical_of_found <= bioc->stripes[i].physical)
6277 physical_of_found = bioc->stripes[i].physical;
6281 struct btrfs_io_stripe *tgtdev_stripe =
6282 bioc->stripes + num_stripes;
6284 tgtdev_stripe->physical = physical_of_found;
6285 tgtdev_stripe->length =
6286 bioc->stripes[index_srcdev].length;
6287 tgtdev_stripe->dev = dev_replace->tgtdev;
6288 bioc->tgtdev_map[index_srcdev] = num_stripes;
6295 *num_stripes_ret = num_stripes;
6296 *max_errors_ret = max_errors;
6297 bioc->num_tgtdevs = tgtdev_indexes;
6301 static bool need_full_stripe(enum btrfs_map_op op)
6303 return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS);
6307 * Calculate the geometry of a particular (address, len) tuple. This
6308 * information is used to calculate how big a particular bio can get before it
6309 * straddles a stripe.
6311 * @fs_info: the filesystem
6312 * @em: mapping containing the logical extent
6313 * @op: type of operation - write or read
6314 * @logical: address that we want to figure out the geometry of
6315 * @io_geom: pointer used to return values
6317 * Returns < 0 in case a chunk for the given logical address cannot be found,
6318 * usually shouldn't happen unless @logical is corrupted, 0 otherwise.
6320 int btrfs_get_io_geometry(struct btrfs_fs_info *fs_info, struct extent_map *em,
6321 enum btrfs_map_op op, u64 logical,
6322 struct btrfs_io_geometry *io_geom)
6324 struct map_lookup *map;
6330 u64 raid56_full_stripe_start = (u64)-1;
6333 ASSERT(op != BTRFS_MAP_DISCARD);
6335 map = em->map_lookup;
6336 /* Offset of this logical address in the chunk */
6337 offset = logical - em->start;
6338 /* Len of a stripe in a chunk */
6339 stripe_len = map->stripe_len;
6340 /* Stripe where this block falls in */
6341 stripe_nr = div64_u64(offset, stripe_len);
6342 /* Offset of stripe in the chunk */
6343 stripe_offset = stripe_nr * stripe_len;
6344 if (offset < stripe_offset) {
6346 "stripe math has gone wrong, stripe_offset=%llu offset=%llu start=%llu logical=%llu stripe_len=%llu",
6347 stripe_offset, offset, em->start, logical, stripe_len);
6351 /* stripe_offset is the offset of this block in its stripe */
6352 stripe_offset = offset - stripe_offset;
6353 data_stripes = nr_data_stripes(map);
6355 /* Only stripe based profiles needs to check against stripe length. */
6356 if (map->type & BTRFS_BLOCK_GROUP_STRIPE_MASK) {
6357 u64 max_len = stripe_len - stripe_offset;
6360 * In case of raid56, we need to know the stripe aligned start
6362 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6363 unsigned long full_stripe_len = stripe_len * data_stripes;
6364 raid56_full_stripe_start = offset;
6367 * Allow a write of a full stripe, but make sure we
6368 * don't allow straddling of stripes
6370 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
6372 raid56_full_stripe_start *= full_stripe_len;
6375 * For writes to RAID[56], allow a full stripeset across
6376 * all disks. For other RAID types and for RAID[56]
6377 * reads, just allow a single stripe (on a single disk).
6379 if (op == BTRFS_MAP_WRITE) {
6380 max_len = stripe_len * data_stripes -
6381 (offset - raid56_full_stripe_start);
6384 len = min_t(u64, em->len - offset, max_len);
6386 len = em->len - offset;
6390 io_geom->offset = offset;
6391 io_geom->stripe_len = stripe_len;
6392 io_geom->stripe_nr = stripe_nr;
6393 io_geom->stripe_offset = stripe_offset;
6394 io_geom->raid56_stripe_offset = raid56_full_stripe_start;
6399 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
6400 enum btrfs_map_op op,
6401 u64 logical, u64 *length,
6402 struct btrfs_io_context **bioc_ret,
6403 int mirror_num, int need_raid_map)
6405 struct extent_map *em;
6406 struct map_lookup *map;
6416 int tgtdev_indexes = 0;
6417 struct btrfs_io_context *bioc = NULL;
6418 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
6419 int dev_replace_is_ongoing = 0;
6420 int num_alloc_stripes;
6421 int patch_the_first_stripe_for_dev_replace = 0;
6422 u64 physical_to_patch_in_first_stripe = 0;
6423 u64 raid56_full_stripe_start = (u64)-1;
6424 struct btrfs_io_geometry geom;
6427 ASSERT(op != BTRFS_MAP_DISCARD);
6429 em = btrfs_get_chunk_map(fs_info, logical, *length);
6430 ASSERT(!IS_ERR(em));
6432 ret = btrfs_get_io_geometry(fs_info, em, op, logical, &geom);
6436 map = em->map_lookup;
6439 stripe_len = geom.stripe_len;
6440 stripe_nr = geom.stripe_nr;
6441 stripe_offset = geom.stripe_offset;
6442 raid56_full_stripe_start = geom.raid56_stripe_offset;
6443 data_stripes = nr_data_stripes(map);
6445 down_read(&dev_replace->rwsem);
6446 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
6448 * Hold the semaphore for read during the whole operation, write is
6449 * requested at commit time but must wait.
6451 if (!dev_replace_is_ongoing)
6452 up_read(&dev_replace->rwsem);
6454 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
6455 !need_full_stripe(op) && dev_replace->tgtdev != NULL) {
6456 ret = get_extra_mirror_from_replace(fs_info, logical, *length,
6457 dev_replace->srcdev->devid,
6459 &physical_to_patch_in_first_stripe);
6463 patch_the_first_stripe_for_dev_replace = 1;
6464 } else if (mirror_num > map->num_stripes) {
6470 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
6471 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
6473 if (!need_full_stripe(op))
6475 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1_MASK) {
6476 if (need_full_stripe(op))
6477 num_stripes = map->num_stripes;
6478 else if (mirror_num)
6479 stripe_index = mirror_num - 1;
6481 stripe_index = find_live_mirror(fs_info, map, 0,
6482 dev_replace_is_ongoing);
6483 mirror_num = stripe_index + 1;
6486 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
6487 if (need_full_stripe(op)) {
6488 num_stripes = map->num_stripes;
6489 } else if (mirror_num) {
6490 stripe_index = mirror_num - 1;
6495 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
6496 u32 factor = map->num_stripes / map->sub_stripes;
6498 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
6499 stripe_index *= map->sub_stripes;
6501 if (need_full_stripe(op))
6502 num_stripes = map->sub_stripes;
6503 else if (mirror_num)
6504 stripe_index += mirror_num - 1;
6506 int old_stripe_index = stripe_index;
6507 stripe_index = find_live_mirror(fs_info, map,
6509 dev_replace_is_ongoing);
6510 mirror_num = stripe_index - old_stripe_index + 1;
6513 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6514 if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) {
6515 /* push stripe_nr back to the start of the full stripe */
6516 stripe_nr = div64_u64(raid56_full_stripe_start,
6517 stripe_len * data_stripes);
6519 /* RAID[56] write or recovery. Return all stripes */
6520 num_stripes = map->num_stripes;
6521 max_errors = nr_parity_stripes(map);
6523 *length = map->stripe_len;
6528 * Mirror #0 or #1 means the original data block.
6529 * Mirror #2 is RAID5 parity block.
6530 * Mirror #3 is RAID6 Q block.
6532 stripe_nr = div_u64_rem(stripe_nr,
6533 data_stripes, &stripe_index);
6535 stripe_index = data_stripes + mirror_num - 2;
6537 /* We distribute the parity blocks across stripes */
6538 div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
6540 if (!need_full_stripe(op) && mirror_num <= 1)
6545 * after this, stripe_nr is the number of stripes on this
6546 * device we have to walk to find the data, and stripe_index is
6547 * the number of our device in the stripe array
6549 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
6551 mirror_num = stripe_index + 1;
6553 if (stripe_index >= map->num_stripes) {
6555 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
6556 stripe_index, map->num_stripes);
6561 num_alloc_stripes = num_stripes;
6562 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) {
6563 if (op == BTRFS_MAP_WRITE)
6564 num_alloc_stripes <<= 1;
6565 if (op == BTRFS_MAP_GET_READ_MIRRORS)
6566 num_alloc_stripes++;
6567 tgtdev_indexes = num_stripes;
6570 bioc = alloc_btrfs_io_context(fs_info, num_alloc_stripes, tgtdev_indexes);
6576 for (i = 0; i < num_stripes; i++) {
6577 bioc->stripes[i].physical = map->stripes[stripe_index].physical +
6578 stripe_offset + stripe_nr * map->stripe_len;
6579 bioc->stripes[i].dev = map->stripes[stripe_index].dev;
6583 /* Build raid_map */
6584 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map &&
6585 (need_full_stripe(op) || mirror_num > 1)) {
6589 /* Work out the disk rotation on this stripe-set */
6590 div_u64_rem(stripe_nr, num_stripes, &rot);
6592 /* Fill in the logical address of each stripe */
6593 tmp = stripe_nr * data_stripes;
6594 for (i = 0; i < data_stripes; i++)
6595 bioc->raid_map[(i + rot) % num_stripes] =
6596 em->start + (tmp + i) * map->stripe_len;
6598 bioc->raid_map[(i + rot) % map->num_stripes] = RAID5_P_STRIPE;
6599 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
6600 bioc->raid_map[(i + rot + 1) % num_stripes] =
6603 sort_parity_stripes(bioc, num_stripes);
6606 if (need_full_stripe(op))
6607 max_errors = btrfs_chunk_max_errors(map);
6609 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
6610 need_full_stripe(op)) {
6611 handle_ops_on_dev_replace(op, &bioc, dev_replace, logical,
6612 &num_stripes, &max_errors);
6616 bioc->map_type = map->type;
6617 bioc->num_stripes = num_stripes;
6618 bioc->max_errors = max_errors;
6619 bioc->mirror_num = mirror_num;
6622 * this is the case that REQ_READ && dev_replace_is_ongoing &&
6623 * mirror_num == num_stripes + 1 && dev_replace target drive is
6624 * available as a mirror
6626 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
6627 WARN_ON(num_stripes > 1);
6628 bioc->stripes[0].dev = dev_replace->tgtdev;
6629 bioc->stripes[0].physical = physical_to_patch_in_first_stripe;
6630 bioc->mirror_num = map->num_stripes + 1;
6633 if (dev_replace_is_ongoing) {
6634 lockdep_assert_held(&dev_replace->rwsem);
6635 /* Unlock and let waiting writers proceed */
6636 up_read(&dev_replace->rwsem);
6638 free_extent_map(em);
6642 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6643 u64 logical, u64 *length,
6644 struct btrfs_io_context **bioc_ret, int mirror_num)
6646 if (op == BTRFS_MAP_DISCARD)
6647 return __btrfs_map_block_for_discard(fs_info, logical,
6650 return __btrfs_map_block(fs_info, op, logical, length, bioc_ret,
6654 /* For Scrub/replace */
6655 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6656 u64 logical, u64 *length,
6657 struct btrfs_io_context **bioc_ret)
6659 return __btrfs_map_block(fs_info, op, logical, length, bioc_ret, 0, 1);
6662 static inline void btrfs_end_bioc(struct btrfs_io_context *bioc, struct bio *bio)
6664 bio->bi_private = bioc->private;
6665 bio->bi_end_io = bioc->end_io;
6668 btrfs_put_bioc(bioc);
6671 static void btrfs_end_bio(struct bio *bio)
6673 struct btrfs_io_context *bioc = bio->bi_private;
6674 int is_orig_bio = 0;
6676 if (bio->bi_status) {
6677 atomic_inc(&bioc->error);
6678 if (bio->bi_status == BLK_STS_IOERR ||
6679 bio->bi_status == BLK_STS_TARGET) {
6680 struct btrfs_device *dev = btrfs_bio(bio)->device;
6683 if (btrfs_op(bio) == BTRFS_MAP_WRITE)
6684 btrfs_dev_stat_inc_and_print(dev,
6685 BTRFS_DEV_STAT_WRITE_ERRS);
6686 else if (!(bio->bi_opf & REQ_RAHEAD))
6687 btrfs_dev_stat_inc_and_print(dev,
6688 BTRFS_DEV_STAT_READ_ERRS);
6689 if (bio->bi_opf & REQ_PREFLUSH)
6690 btrfs_dev_stat_inc_and_print(dev,
6691 BTRFS_DEV_STAT_FLUSH_ERRS);
6695 if (bio == bioc->orig_bio)
6698 btrfs_bio_counter_dec(bioc->fs_info);
6700 if (atomic_dec_and_test(&bioc->stripes_pending)) {
6703 bio = bioc->orig_bio;
6706 btrfs_bio(bio)->mirror_num = bioc->mirror_num;
6707 /* only send an error to the higher layers if it is
6708 * beyond the tolerance of the btrfs bio
6710 if (atomic_read(&bioc->error) > bioc->max_errors) {
6711 bio->bi_status = BLK_STS_IOERR;
6714 * this bio is actually up to date, we didn't
6715 * go over the max number of errors
6717 bio->bi_status = BLK_STS_OK;
6720 btrfs_end_bioc(bioc, bio);
6721 } else if (!is_orig_bio) {
6726 static void submit_stripe_bio(struct btrfs_io_context *bioc, struct bio *bio,
6727 u64 physical, struct btrfs_device *dev)
6729 struct btrfs_fs_info *fs_info = bioc->fs_info;
6731 bio->bi_private = bioc;
6732 btrfs_bio(bio)->device = dev;
6733 bio->bi_end_io = btrfs_end_bio;
6734 bio->bi_iter.bi_sector = physical >> 9;
6736 * For zone append writing, bi_sector must point the beginning of the
6739 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
6740 if (btrfs_dev_is_sequential(dev, physical)) {
6741 u64 zone_start = round_down(physical, fs_info->zone_size);
6743 bio->bi_iter.bi_sector = zone_start >> SECTOR_SHIFT;
6745 bio->bi_opf &= ~REQ_OP_ZONE_APPEND;
6746 bio->bi_opf |= REQ_OP_WRITE;
6749 btrfs_debug_in_rcu(fs_info,
6750 "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
6751 bio_op(bio), bio->bi_opf, bio->bi_iter.bi_sector,
6752 (unsigned long)dev->bdev->bd_dev, rcu_str_deref(dev->name),
6753 dev->devid, bio->bi_iter.bi_size);
6754 bio_set_dev(bio, dev->bdev);
6756 btrfs_bio_counter_inc_noblocked(fs_info);
6758 btrfsic_submit_bio(bio);
6761 static void bioc_error(struct btrfs_io_context *bioc, struct bio *bio, u64 logical)
6763 atomic_inc(&bioc->error);
6764 if (atomic_dec_and_test(&bioc->stripes_pending)) {
6765 /* Should be the original bio. */
6766 WARN_ON(bio != bioc->orig_bio);
6768 btrfs_bio(bio)->mirror_num = bioc->mirror_num;
6769 bio->bi_iter.bi_sector = logical >> 9;
6770 if (atomic_read(&bioc->error) > bioc->max_errors)
6771 bio->bi_status = BLK_STS_IOERR;
6773 bio->bi_status = BLK_STS_OK;
6774 btrfs_end_bioc(bioc, bio);
6778 blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
6781 struct btrfs_device *dev;
6782 struct bio *first_bio = bio;
6783 u64 logical = bio->bi_iter.bi_sector << 9;
6789 struct btrfs_io_context *bioc = NULL;
6791 length = bio->bi_iter.bi_size;
6792 map_length = length;
6794 btrfs_bio_counter_inc_blocked(fs_info);
6795 ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical,
6796 &map_length, &bioc, mirror_num, 1);
6798 btrfs_bio_counter_dec(fs_info);
6799 return errno_to_blk_status(ret);
6802 total_devs = bioc->num_stripes;
6803 bioc->orig_bio = first_bio;
6804 bioc->private = first_bio->bi_private;
6805 bioc->end_io = first_bio->bi_end_io;
6806 atomic_set(&bioc->stripes_pending, bioc->num_stripes);
6808 if ((bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6809 ((btrfs_op(bio) == BTRFS_MAP_WRITE) || (mirror_num > 1))) {
6810 /* In this case, map_length has been set to the length of
6811 a single stripe; not the whole write */
6812 if (btrfs_op(bio) == BTRFS_MAP_WRITE) {
6813 ret = raid56_parity_write(bio, bioc, map_length);
6815 ret = raid56_parity_recover(bio, bioc, map_length,
6819 btrfs_bio_counter_dec(fs_info);
6820 return errno_to_blk_status(ret);
6823 if (map_length < length) {
6825 "mapping failed logical %llu bio len %llu len %llu",
6826 logical, length, map_length);
6830 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6831 dev = bioc->stripes[dev_nr].dev;
6832 if (!dev || !dev->bdev || test_bit(BTRFS_DEV_STATE_MISSING,
6834 (btrfs_op(first_bio) == BTRFS_MAP_WRITE &&
6835 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) {
6836 bioc_error(bioc, first_bio, logical);
6840 if (dev_nr < total_devs - 1)
6841 bio = btrfs_bio_clone(first_bio);
6845 submit_stripe_bio(bioc, bio, bioc->stripes[dev_nr].physical, dev);
6847 btrfs_bio_counter_dec(fs_info);
6851 static bool dev_args_match_fs_devices(const struct btrfs_dev_lookup_args *args,
6852 const struct btrfs_fs_devices *fs_devices)
6854 if (args->fsid == NULL)
6856 if (memcmp(fs_devices->metadata_uuid, args->fsid, BTRFS_FSID_SIZE) == 0)
6861 static bool dev_args_match_device(const struct btrfs_dev_lookup_args *args,
6862 const struct btrfs_device *device)
6864 ASSERT((args->devid != (u64)-1) || args->missing);
6866 if ((args->devid != (u64)-1) && device->devid != args->devid)
6868 if (args->uuid && memcmp(device->uuid, args->uuid, BTRFS_UUID_SIZE) != 0)
6872 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state) &&
6879 * Find a device specified by @devid or @uuid in the list of @fs_devices, or
6882 * If devid and uuid are both specified, the match must be exact, otherwise
6883 * only devid is used.
6885 struct btrfs_device *btrfs_find_device(const struct btrfs_fs_devices *fs_devices,
6886 const struct btrfs_dev_lookup_args *args)
6888 struct btrfs_device *device;
6889 struct btrfs_fs_devices *seed_devs;
6891 if (dev_args_match_fs_devices(args, fs_devices)) {
6892 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6893 if (dev_args_match_device(args, device))
6898 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
6899 if (!dev_args_match_fs_devices(args, seed_devs))
6901 list_for_each_entry(device, &seed_devs->devices, dev_list) {
6902 if (dev_args_match_device(args, device))
6910 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices,
6911 u64 devid, u8 *dev_uuid)
6913 struct btrfs_device *device;
6914 unsigned int nofs_flag;
6917 * We call this under the chunk_mutex, so we want to use NOFS for this
6918 * allocation, however we don't want to change btrfs_alloc_device() to
6919 * always do NOFS because we use it in a lot of other GFP_KERNEL safe
6922 nofs_flag = memalloc_nofs_save();
6923 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6924 memalloc_nofs_restore(nofs_flag);
6928 list_add(&device->dev_list, &fs_devices->devices);
6929 device->fs_devices = fs_devices;
6930 fs_devices->num_devices++;
6932 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6933 fs_devices->missing_devices++;
6939 * btrfs_alloc_device - allocate struct btrfs_device
6940 * @fs_info: used only for generating a new devid, can be NULL if
6941 * devid is provided (i.e. @devid != NULL).
6942 * @devid: a pointer to devid for this device. If NULL a new devid
6944 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6947 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6948 * on error. Returned struct is not linked onto any lists and must be
6949 * destroyed with btrfs_free_device.
6951 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6955 struct btrfs_device *dev;
6958 if (WARN_ON(!devid && !fs_info))
6959 return ERR_PTR(-EINVAL);
6961 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
6963 return ERR_PTR(-ENOMEM);
6966 * Preallocate a bio that's always going to be used for flushing device
6967 * barriers and matches the device lifespan
6969 dev->flush_bio = bio_kmalloc(GFP_KERNEL, 0);
6970 if (!dev->flush_bio) {
6972 return ERR_PTR(-ENOMEM);
6975 INIT_LIST_HEAD(&dev->dev_list);
6976 INIT_LIST_HEAD(&dev->dev_alloc_list);
6977 INIT_LIST_HEAD(&dev->post_commit_list);
6979 atomic_set(&dev->dev_stats_ccnt, 0);
6980 btrfs_device_data_ordered_init(dev);
6981 extent_io_tree_init(fs_info, &dev->alloc_state,
6982 IO_TREE_DEVICE_ALLOC_STATE, NULL);
6989 ret = find_next_devid(fs_info, &tmp);
6991 btrfs_free_device(dev);
6992 return ERR_PTR(ret);
6998 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
7000 generate_random_uuid(dev->uuid);
7005 static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info,
7006 u64 devid, u8 *uuid, bool error)
7009 btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing",
7012 btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing",
7016 static u64 calc_stripe_length(u64 type, u64 chunk_len, int num_stripes)
7018 const int data_stripes = calc_data_stripes(type, num_stripes);
7020 return div_u64(chunk_len, data_stripes);
7023 #if BITS_PER_LONG == 32
7025 * Due to page cache limit, metadata beyond BTRFS_32BIT_MAX_FILE_SIZE
7026 * can't be accessed on 32bit systems.
7028 * This function do mount time check to reject the fs if it already has
7029 * metadata chunk beyond that limit.
7031 static int check_32bit_meta_chunk(struct btrfs_fs_info *fs_info,
7032 u64 logical, u64 length, u64 type)
7034 if (!(type & BTRFS_BLOCK_GROUP_METADATA))
7037 if (logical + length < MAX_LFS_FILESIZE)
7040 btrfs_err_32bit_limit(fs_info);
7045 * This is to give early warning for any metadata chunk reaching
7046 * BTRFS_32BIT_EARLY_WARN_THRESHOLD.
7047 * Although we can still access the metadata, it's not going to be possible
7048 * once the limit is reached.
7050 static void warn_32bit_meta_chunk(struct btrfs_fs_info *fs_info,
7051 u64 logical, u64 length, u64 type)
7053 if (!(type & BTRFS_BLOCK_GROUP_METADATA))
7056 if (logical + length < BTRFS_32BIT_EARLY_WARN_THRESHOLD)
7059 btrfs_warn_32bit_limit(fs_info);
7063 static int read_one_chunk(struct btrfs_key *key, struct extent_buffer *leaf,
7064 struct btrfs_chunk *chunk)
7066 BTRFS_DEV_LOOKUP_ARGS(args);
7067 struct btrfs_fs_info *fs_info = leaf->fs_info;
7068 struct extent_map_tree *map_tree = &fs_info->mapping_tree;
7069 struct map_lookup *map;
7070 struct extent_map *em;
7075 u8 uuid[BTRFS_UUID_SIZE];
7080 logical = key->offset;
7081 length = btrfs_chunk_length(leaf, chunk);
7082 type = btrfs_chunk_type(leaf, chunk);
7083 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
7085 #if BITS_PER_LONG == 32
7086 ret = check_32bit_meta_chunk(fs_info, logical, length, type);
7089 warn_32bit_meta_chunk(fs_info, logical, length, type);
7093 * Only need to verify chunk item if we're reading from sys chunk array,
7094 * as chunk item in tree block is already verified by tree-checker.
7096 if (leaf->start == BTRFS_SUPER_INFO_OFFSET) {
7097 ret = btrfs_check_chunk_valid(leaf, chunk, logical);
7102 read_lock(&map_tree->lock);
7103 em = lookup_extent_mapping(map_tree, logical, 1);
7104 read_unlock(&map_tree->lock);
7106 /* already mapped? */
7107 if (em && em->start <= logical && em->start + em->len > logical) {
7108 free_extent_map(em);
7111 free_extent_map(em);
7114 em = alloc_extent_map();
7117 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
7119 free_extent_map(em);
7123 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
7124 em->map_lookup = map;
7125 em->start = logical;
7128 em->block_start = 0;
7129 em->block_len = em->len;
7131 map->num_stripes = num_stripes;
7132 map->io_width = btrfs_chunk_io_width(leaf, chunk);
7133 map->io_align = btrfs_chunk_io_align(leaf, chunk);
7134 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
7136 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
7137 map->verified_stripes = 0;
7138 em->orig_block_len = calc_stripe_length(type, em->len,
7140 for (i = 0; i < num_stripes; i++) {
7141 map->stripes[i].physical =
7142 btrfs_stripe_offset_nr(leaf, chunk, i);
7143 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
7145 read_extent_buffer(leaf, uuid, (unsigned long)
7146 btrfs_stripe_dev_uuid_nr(chunk, i),
7149 map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices, &args);
7150 if (!map->stripes[i].dev &&
7151 !btrfs_test_opt(fs_info, DEGRADED)) {
7152 free_extent_map(em);
7153 btrfs_report_missing_device(fs_info, devid, uuid, true);
7156 if (!map->stripes[i].dev) {
7157 map->stripes[i].dev =
7158 add_missing_dev(fs_info->fs_devices, devid,
7160 if (IS_ERR(map->stripes[i].dev)) {
7161 free_extent_map(em);
7163 "failed to init missing dev %llu: %ld",
7164 devid, PTR_ERR(map->stripes[i].dev));
7165 return PTR_ERR(map->stripes[i].dev);
7167 btrfs_report_missing_device(fs_info, devid, uuid, false);
7169 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
7170 &(map->stripes[i].dev->dev_state));
7174 write_lock(&map_tree->lock);
7175 ret = add_extent_mapping(map_tree, em, 0);
7176 write_unlock(&map_tree->lock);
7179 "failed to add chunk map, start=%llu len=%llu: %d",
7180 em->start, em->len, ret);
7182 free_extent_map(em);
7187 static void fill_device_from_item(struct extent_buffer *leaf,
7188 struct btrfs_dev_item *dev_item,
7189 struct btrfs_device *device)
7193 device->devid = btrfs_device_id(leaf, dev_item);
7194 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
7195 device->total_bytes = device->disk_total_bytes;
7196 device->commit_total_bytes = device->disk_total_bytes;
7197 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
7198 device->commit_bytes_used = device->bytes_used;
7199 device->type = btrfs_device_type(leaf, dev_item);
7200 device->io_align = btrfs_device_io_align(leaf, dev_item);
7201 device->io_width = btrfs_device_io_width(leaf, dev_item);
7202 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
7203 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
7204 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
7206 ptr = btrfs_device_uuid(dev_item);
7207 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
7210 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info,
7213 struct btrfs_fs_devices *fs_devices;
7216 lockdep_assert_held(&uuid_mutex);
7219 /* This will match only for multi-device seed fs */
7220 list_for_each_entry(fs_devices, &fs_info->fs_devices->seed_list, seed_list)
7221 if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE))
7225 fs_devices = find_fsid(fsid, NULL);
7227 if (!btrfs_test_opt(fs_info, DEGRADED))
7228 return ERR_PTR(-ENOENT);
7230 fs_devices = alloc_fs_devices(fsid, NULL);
7231 if (IS_ERR(fs_devices))
7234 fs_devices->seeding = true;
7235 fs_devices->opened = 1;
7240 * Upon first call for a seed fs fsid, just create a private copy of the
7241 * respective fs_devices and anchor it at fs_info->fs_devices->seed_list
7243 fs_devices = clone_fs_devices(fs_devices);
7244 if (IS_ERR(fs_devices))
7247 ret = open_fs_devices(fs_devices, FMODE_READ, fs_info->bdev_holder);
7249 free_fs_devices(fs_devices);
7250 return ERR_PTR(ret);
7253 if (!fs_devices->seeding) {
7254 close_fs_devices(fs_devices);
7255 free_fs_devices(fs_devices);
7256 return ERR_PTR(-EINVAL);
7259 list_add(&fs_devices->seed_list, &fs_info->fs_devices->seed_list);
7264 static int read_one_dev(struct extent_buffer *leaf,
7265 struct btrfs_dev_item *dev_item)
7267 BTRFS_DEV_LOOKUP_ARGS(args);
7268 struct btrfs_fs_info *fs_info = leaf->fs_info;
7269 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7270 struct btrfs_device *device;
7273 u8 fs_uuid[BTRFS_FSID_SIZE];
7274 u8 dev_uuid[BTRFS_UUID_SIZE];
7276 devid = args.devid = btrfs_device_id(leaf, dev_item);
7277 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
7279 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
7281 args.uuid = dev_uuid;
7282 args.fsid = fs_uuid;
7284 if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) {
7285 fs_devices = open_seed_devices(fs_info, fs_uuid);
7286 if (IS_ERR(fs_devices))
7287 return PTR_ERR(fs_devices);
7290 device = btrfs_find_device(fs_info->fs_devices, &args);
7292 if (!btrfs_test_opt(fs_info, DEGRADED)) {
7293 btrfs_report_missing_device(fs_info, devid,
7298 device = add_missing_dev(fs_devices, devid, dev_uuid);
7299 if (IS_ERR(device)) {
7301 "failed to add missing dev %llu: %ld",
7302 devid, PTR_ERR(device));
7303 return PTR_ERR(device);
7305 btrfs_report_missing_device(fs_info, devid, dev_uuid, false);
7307 if (!device->bdev) {
7308 if (!btrfs_test_opt(fs_info, DEGRADED)) {
7309 btrfs_report_missing_device(fs_info,
7310 devid, dev_uuid, true);
7313 btrfs_report_missing_device(fs_info, devid,
7317 if (!device->bdev &&
7318 !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
7320 * this happens when a device that was properly setup
7321 * in the device info lists suddenly goes bad.
7322 * device->bdev is NULL, and so we have to set
7323 * device->missing to one here
7325 device->fs_devices->missing_devices++;
7326 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
7329 /* Move the device to its own fs_devices */
7330 if (device->fs_devices != fs_devices) {
7331 ASSERT(test_bit(BTRFS_DEV_STATE_MISSING,
7332 &device->dev_state));
7334 list_move(&device->dev_list, &fs_devices->devices);
7335 device->fs_devices->num_devices--;
7336 fs_devices->num_devices++;
7338 device->fs_devices->missing_devices--;
7339 fs_devices->missing_devices++;
7341 device->fs_devices = fs_devices;
7345 if (device->fs_devices != fs_info->fs_devices) {
7346 BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state));
7347 if (device->generation !=
7348 btrfs_device_generation(leaf, dev_item))
7352 fill_device_from_item(leaf, dev_item, device);
7354 u64 max_total_bytes = bdev_nr_bytes(device->bdev);
7356 if (device->total_bytes > max_total_bytes) {
7358 "device total_bytes should be at most %llu but found %llu",
7359 max_total_bytes, device->total_bytes);
7363 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
7364 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
7365 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
7366 device->fs_devices->total_rw_bytes += device->total_bytes;
7367 atomic64_add(device->total_bytes - device->bytes_used,
7368 &fs_info->free_chunk_space);
7374 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
7376 struct btrfs_root *root = fs_info->tree_root;
7377 struct btrfs_super_block *super_copy = fs_info->super_copy;
7378 struct extent_buffer *sb;
7379 struct btrfs_disk_key *disk_key;
7380 struct btrfs_chunk *chunk;
7382 unsigned long sb_array_offset;
7389 struct btrfs_key key;
7391 ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize);
7393 * This will create extent buffer of nodesize, superblock size is
7394 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
7395 * overallocate but we can keep it as-is, only the first page is used.
7397 sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET,
7398 root->root_key.objectid, 0);
7401 set_extent_buffer_uptodate(sb);
7403 * The sb extent buffer is artificial and just used to read the system array.
7404 * set_extent_buffer_uptodate() call does not properly mark all it's
7405 * pages up-to-date when the page is larger: extent does not cover the
7406 * whole page and consequently check_page_uptodate does not find all
7407 * the page's extents up-to-date (the hole beyond sb),
7408 * write_extent_buffer then triggers a WARN_ON.
7410 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
7411 * but sb spans only this function. Add an explicit SetPageUptodate call
7412 * to silence the warning eg. on PowerPC 64.
7414 if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
7415 SetPageUptodate(sb->pages[0]);
7417 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
7418 array_size = btrfs_super_sys_array_size(super_copy);
7420 array_ptr = super_copy->sys_chunk_array;
7421 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
7424 while (cur_offset < array_size) {
7425 disk_key = (struct btrfs_disk_key *)array_ptr;
7426 len = sizeof(*disk_key);
7427 if (cur_offset + len > array_size)
7428 goto out_short_read;
7430 btrfs_disk_key_to_cpu(&key, disk_key);
7433 sb_array_offset += len;
7436 if (key.type != BTRFS_CHUNK_ITEM_KEY) {
7438 "unexpected item type %u in sys_array at offset %u",
7439 (u32)key.type, cur_offset);
7444 chunk = (struct btrfs_chunk *)sb_array_offset;
7446 * At least one btrfs_chunk with one stripe must be present,
7447 * exact stripe count check comes afterwards
7449 len = btrfs_chunk_item_size(1);
7450 if (cur_offset + len > array_size)
7451 goto out_short_read;
7453 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
7456 "invalid number of stripes %u in sys_array at offset %u",
7457 num_stripes, cur_offset);
7462 type = btrfs_chunk_type(sb, chunk);
7463 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
7465 "invalid chunk type %llu in sys_array at offset %u",
7471 len = btrfs_chunk_item_size(num_stripes);
7472 if (cur_offset + len > array_size)
7473 goto out_short_read;
7475 ret = read_one_chunk(&key, sb, chunk);
7480 sb_array_offset += len;
7483 clear_extent_buffer_uptodate(sb);
7484 free_extent_buffer_stale(sb);
7488 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
7490 clear_extent_buffer_uptodate(sb);
7491 free_extent_buffer_stale(sb);
7496 * Check if all chunks in the fs are OK for read-write degraded mount
7498 * If the @failing_dev is specified, it's accounted as missing.
7500 * Return true if all chunks meet the minimal RW mount requirements.
7501 * Return false if any chunk doesn't meet the minimal RW mount requirements.
7503 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
7504 struct btrfs_device *failing_dev)
7506 struct extent_map_tree *map_tree = &fs_info->mapping_tree;
7507 struct extent_map *em;
7511 read_lock(&map_tree->lock);
7512 em = lookup_extent_mapping(map_tree, 0, (u64)-1);
7513 read_unlock(&map_tree->lock);
7514 /* No chunk at all? Return false anyway */
7520 struct map_lookup *map;
7525 map = em->map_lookup;
7527 btrfs_get_num_tolerated_disk_barrier_failures(
7529 for (i = 0; i < map->num_stripes; i++) {
7530 struct btrfs_device *dev = map->stripes[i].dev;
7532 if (!dev || !dev->bdev ||
7533 test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) ||
7534 dev->last_flush_error)
7536 else if (failing_dev && failing_dev == dev)
7539 if (missing > max_tolerated) {
7542 "chunk %llu missing %d devices, max tolerance is %d for writable mount",
7543 em->start, missing, max_tolerated);
7544 free_extent_map(em);
7548 next_start = extent_map_end(em);
7549 free_extent_map(em);
7551 read_lock(&map_tree->lock);
7552 em = lookup_extent_mapping(map_tree, next_start,
7553 (u64)(-1) - next_start);
7554 read_unlock(&map_tree->lock);
7560 static void readahead_tree_node_children(struct extent_buffer *node)
7563 const int nr_items = btrfs_header_nritems(node);
7565 for (i = 0; i < nr_items; i++)
7566 btrfs_readahead_node_child(node, i);
7569 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
7571 struct btrfs_root *root = fs_info->chunk_root;
7572 struct btrfs_path *path;
7573 struct extent_buffer *leaf;
7574 struct btrfs_key key;
7575 struct btrfs_key found_key;
7579 u64 last_ra_node = 0;
7581 path = btrfs_alloc_path();
7586 * uuid_mutex is needed only if we are mounting a sprout FS
7587 * otherwise we don't need it.
7589 mutex_lock(&uuid_mutex);
7592 * It is possible for mount and umount to race in such a way that
7593 * we execute this code path, but open_fs_devices failed to clear
7594 * total_rw_bytes. We certainly want it cleared before reading the
7595 * device items, so clear it here.
7597 fs_info->fs_devices->total_rw_bytes = 0;
7600 * Lockdep complains about possible circular locking dependency between
7601 * a disk's open_mutex (struct gendisk.open_mutex), the rw semaphores
7602 * used for freeze procection of a fs (struct super_block.s_writers),
7603 * which we take when starting a transaction, and extent buffers of the
7604 * chunk tree if we call read_one_dev() while holding a lock on an
7605 * extent buffer of the chunk tree. Since we are mounting the filesystem
7606 * and at this point there can't be any concurrent task modifying the
7607 * chunk tree, to keep it simple, just skip locking on the chunk tree.
7609 ASSERT(!test_bit(BTRFS_FS_OPEN, &fs_info->flags));
7610 path->skip_locking = 1;
7613 * Read all device items, and then all the chunk items. All
7614 * device items are found before any chunk item (their object id
7615 * is smaller than the lowest possible object id for a chunk
7616 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
7618 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
7621 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7625 struct extent_buffer *node;
7627 leaf = path->nodes[0];
7628 slot = path->slots[0];
7629 if (slot >= btrfs_header_nritems(leaf)) {
7630 ret = btrfs_next_leaf(root, path);
7637 node = path->nodes[1];
7639 if (last_ra_node != node->start) {
7640 readahead_tree_node_children(node);
7641 last_ra_node = node->start;
7644 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7645 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
7646 struct btrfs_dev_item *dev_item;
7647 dev_item = btrfs_item_ptr(leaf, slot,
7648 struct btrfs_dev_item);
7649 ret = read_one_dev(leaf, dev_item);
7653 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
7654 struct btrfs_chunk *chunk;
7657 * We are only called at mount time, so no need to take
7658 * fs_info->chunk_mutex. Plus, to avoid lockdep warnings,
7659 * we always lock first fs_info->chunk_mutex before
7660 * acquiring any locks on the chunk tree. This is a
7661 * requirement for chunk allocation, see the comment on
7662 * top of btrfs_chunk_alloc() for details.
7664 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
7665 ret = read_one_chunk(&found_key, leaf, chunk);
7673 * After loading chunk tree, we've got all device information,
7674 * do another round of validation checks.
7676 if (total_dev != fs_info->fs_devices->total_devices) {
7678 "super_num_devices %llu mismatch with num_devices %llu found here",
7679 btrfs_super_num_devices(fs_info->super_copy),
7684 if (btrfs_super_total_bytes(fs_info->super_copy) <
7685 fs_info->fs_devices->total_rw_bytes) {
7687 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
7688 btrfs_super_total_bytes(fs_info->super_copy),
7689 fs_info->fs_devices->total_rw_bytes);
7695 mutex_unlock(&uuid_mutex);
7697 btrfs_free_path(path);
7701 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
7703 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
7704 struct btrfs_device *device;
7706 fs_devices->fs_info = fs_info;
7708 mutex_lock(&fs_devices->device_list_mutex);
7709 list_for_each_entry(device, &fs_devices->devices, dev_list)
7710 device->fs_info = fs_info;
7712 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
7713 list_for_each_entry(device, &seed_devs->devices, dev_list)
7714 device->fs_info = fs_info;
7716 seed_devs->fs_info = fs_info;
7718 mutex_unlock(&fs_devices->device_list_mutex);
7721 static u64 btrfs_dev_stats_value(const struct extent_buffer *eb,
7722 const struct btrfs_dev_stats_item *ptr,
7727 read_extent_buffer(eb, &val,
7728 offsetof(struct btrfs_dev_stats_item, values) +
7729 ((unsigned long)ptr) + (index * sizeof(u64)),
7734 static void btrfs_set_dev_stats_value(struct extent_buffer *eb,
7735 struct btrfs_dev_stats_item *ptr,
7738 write_extent_buffer(eb, &val,
7739 offsetof(struct btrfs_dev_stats_item, values) +
7740 ((unsigned long)ptr) + (index * sizeof(u64)),
7744 static int btrfs_device_init_dev_stats(struct btrfs_device *device,
7745 struct btrfs_path *path)
7747 struct btrfs_dev_stats_item *ptr;
7748 struct extent_buffer *eb;
7749 struct btrfs_key key;
7753 if (!device->fs_info->dev_root)
7756 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7757 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7758 key.offset = device->devid;
7759 ret = btrfs_search_slot(NULL, device->fs_info->dev_root, &key, path, 0, 0);
7761 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7762 btrfs_dev_stat_set(device, i, 0);
7763 device->dev_stats_valid = 1;
7764 btrfs_release_path(path);
7765 return ret < 0 ? ret : 0;
7767 slot = path->slots[0];
7768 eb = path->nodes[0];
7769 item_size = btrfs_item_size(eb, slot);
7771 ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_stats_item);
7773 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7774 if (item_size >= (1 + i) * sizeof(__le64))
7775 btrfs_dev_stat_set(device, i,
7776 btrfs_dev_stats_value(eb, ptr, i));
7778 btrfs_dev_stat_set(device, i, 0);
7781 device->dev_stats_valid = 1;
7782 btrfs_dev_stat_print_on_load(device);
7783 btrfs_release_path(path);
7788 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
7790 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
7791 struct btrfs_device *device;
7792 struct btrfs_path *path = NULL;
7795 path = btrfs_alloc_path();
7799 mutex_lock(&fs_devices->device_list_mutex);
7800 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7801 ret = btrfs_device_init_dev_stats(device, path);
7805 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
7806 list_for_each_entry(device, &seed_devs->devices, dev_list) {
7807 ret = btrfs_device_init_dev_stats(device, path);
7813 mutex_unlock(&fs_devices->device_list_mutex);
7815 btrfs_free_path(path);
7819 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
7820 struct btrfs_device *device)
7822 struct btrfs_fs_info *fs_info = trans->fs_info;
7823 struct btrfs_root *dev_root = fs_info->dev_root;
7824 struct btrfs_path *path;
7825 struct btrfs_key key;
7826 struct extent_buffer *eb;
7827 struct btrfs_dev_stats_item *ptr;
7831 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7832 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7833 key.offset = device->devid;
7835 path = btrfs_alloc_path();
7838 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
7840 btrfs_warn_in_rcu(fs_info,
7841 "error %d while searching for dev_stats item for device %s",
7842 ret, rcu_str_deref(device->name));
7847 btrfs_item_size(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
7848 /* need to delete old one and insert a new one */
7849 ret = btrfs_del_item(trans, dev_root, path);
7851 btrfs_warn_in_rcu(fs_info,
7852 "delete too small dev_stats item for device %s failed %d",
7853 rcu_str_deref(device->name), ret);
7860 /* need to insert a new item */
7861 btrfs_release_path(path);
7862 ret = btrfs_insert_empty_item(trans, dev_root, path,
7863 &key, sizeof(*ptr));
7865 btrfs_warn_in_rcu(fs_info,
7866 "insert dev_stats item for device %s failed %d",
7867 rcu_str_deref(device->name), ret);
7872 eb = path->nodes[0];
7873 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7874 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7875 btrfs_set_dev_stats_value(eb, ptr, i,
7876 btrfs_dev_stat_read(device, i));
7877 btrfs_mark_buffer_dirty(eb);
7880 btrfs_free_path(path);
7885 * called from commit_transaction. Writes all changed device stats to disk.
7887 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans)
7889 struct btrfs_fs_info *fs_info = trans->fs_info;
7890 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7891 struct btrfs_device *device;
7895 mutex_lock(&fs_devices->device_list_mutex);
7896 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7897 stats_cnt = atomic_read(&device->dev_stats_ccnt);
7898 if (!device->dev_stats_valid || stats_cnt == 0)
7903 * There is a LOAD-LOAD control dependency between the value of
7904 * dev_stats_ccnt and updating the on-disk values which requires
7905 * reading the in-memory counters. Such control dependencies
7906 * require explicit read memory barriers.
7908 * This memory barriers pairs with smp_mb__before_atomic in
7909 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full
7910 * barrier implied by atomic_xchg in
7911 * btrfs_dev_stats_read_and_reset
7915 ret = update_dev_stat_item(trans, device);
7917 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7919 mutex_unlock(&fs_devices->device_list_mutex);
7924 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7926 btrfs_dev_stat_inc(dev, index);
7927 btrfs_dev_stat_print_on_error(dev);
7930 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
7932 if (!dev->dev_stats_valid)
7934 btrfs_err_rl_in_rcu(dev->fs_info,
7935 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7936 rcu_str_deref(dev->name),
7937 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7938 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7939 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7940 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7941 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7944 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7948 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7949 if (btrfs_dev_stat_read(dev, i) != 0)
7951 if (i == BTRFS_DEV_STAT_VALUES_MAX)
7952 return; /* all values == 0, suppress message */
7954 btrfs_info_in_rcu(dev->fs_info,
7955 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7956 rcu_str_deref(dev->name),
7957 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7958 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7959 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7960 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7961 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7964 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
7965 struct btrfs_ioctl_get_dev_stats *stats)
7967 BTRFS_DEV_LOOKUP_ARGS(args);
7968 struct btrfs_device *dev;
7969 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7972 mutex_lock(&fs_devices->device_list_mutex);
7973 args.devid = stats->devid;
7974 dev = btrfs_find_device(fs_info->fs_devices, &args);
7975 mutex_unlock(&fs_devices->device_list_mutex);
7978 btrfs_warn(fs_info, "get dev_stats failed, device not found");
7980 } else if (!dev->dev_stats_valid) {
7981 btrfs_warn(fs_info, "get dev_stats failed, not yet valid");
7983 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7984 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7985 if (stats->nr_items > i)
7987 btrfs_dev_stat_read_and_reset(dev, i);
7989 btrfs_dev_stat_set(dev, i, 0);
7991 btrfs_info(fs_info, "device stats zeroed by %s (%d)",
7992 current->comm, task_pid_nr(current));
7994 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7995 if (stats->nr_items > i)
7996 stats->values[i] = btrfs_dev_stat_read(dev, i);
7998 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7999 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
8004 * Update the size and bytes used for each device where it changed. This is
8005 * delayed since we would otherwise get errors while writing out the
8008 * Must be invoked during transaction commit.
8010 void btrfs_commit_device_sizes(struct btrfs_transaction *trans)
8012 struct btrfs_device *curr, *next;
8014 ASSERT(trans->state == TRANS_STATE_COMMIT_DOING);
8016 if (list_empty(&trans->dev_update_list))
8020 * We don't need the device_list_mutex here. This list is owned by the
8021 * transaction and the transaction must complete before the device is
8024 mutex_lock(&trans->fs_info->chunk_mutex);
8025 list_for_each_entry_safe(curr, next, &trans->dev_update_list,
8027 list_del_init(&curr->post_commit_list);
8028 curr->commit_total_bytes = curr->disk_total_bytes;
8029 curr->commit_bytes_used = curr->bytes_used;
8031 mutex_unlock(&trans->fs_info->chunk_mutex);
8035 * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10.
8037 int btrfs_bg_type_to_factor(u64 flags)
8039 const int index = btrfs_bg_flags_to_raid_index(flags);
8041 return btrfs_raid_array[index].ncopies;
8046 static int verify_one_dev_extent(struct btrfs_fs_info *fs_info,
8047 u64 chunk_offset, u64 devid,
8048 u64 physical_offset, u64 physical_len)
8050 struct btrfs_dev_lookup_args args = { .devid = devid };
8051 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
8052 struct extent_map *em;
8053 struct map_lookup *map;
8054 struct btrfs_device *dev;
8060 read_lock(&em_tree->lock);
8061 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
8062 read_unlock(&em_tree->lock);
8066 "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk",
8067 physical_offset, devid);
8072 map = em->map_lookup;
8073 stripe_len = calc_stripe_length(map->type, em->len, map->num_stripes);
8074 if (physical_len != stripe_len) {
8076 "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu",
8077 physical_offset, devid, em->start, physical_len,
8083 for (i = 0; i < map->num_stripes; i++) {
8084 if (map->stripes[i].dev->devid == devid &&
8085 map->stripes[i].physical == physical_offset) {
8087 if (map->verified_stripes >= map->num_stripes) {
8089 "too many dev extents for chunk %llu found",
8094 map->verified_stripes++;
8100 "dev extent physical offset %llu devid %llu has no corresponding chunk",
8101 physical_offset, devid);
8105 /* Make sure no dev extent is beyond device boundary */
8106 dev = btrfs_find_device(fs_info->fs_devices, &args);
8108 btrfs_err(fs_info, "failed to find devid %llu", devid);
8113 if (physical_offset + physical_len > dev->disk_total_bytes) {
8115 "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu",
8116 devid, physical_offset, physical_len,
8117 dev->disk_total_bytes);
8122 if (dev->zone_info) {
8123 u64 zone_size = dev->zone_info->zone_size;
8125 if (!IS_ALIGNED(physical_offset, zone_size) ||
8126 !IS_ALIGNED(physical_len, zone_size)) {
8128 "zoned: dev extent devid %llu physical offset %llu len %llu is not aligned to device zone",
8129 devid, physical_offset, physical_len);
8136 free_extent_map(em);
8140 static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info)
8142 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
8143 struct extent_map *em;
8144 struct rb_node *node;
8147 read_lock(&em_tree->lock);
8148 for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) {
8149 em = rb_entry(node, struct extent_map, rb_node);
8150 if (em->map_lookup->num_stripes !=
8151 em->map_lookup->verified_stripes) {
8153 "chunk %llu has missing dev extent, have %d expect %d",
8154 em->start, em->map_lookup->verified_stripes,
8155 em->map_lookup->num_stripes);
8161 read_unlock(&em_tree->lock);
8166 * Ensure that all dev extents are mapped to correct chunk, otherwise
8167 * later chunk allocation/free would cause unexpected behavior.
8169 * NOTE: This will iterate through the whole device tree, which should be of
8170 * the same size level as the chunk tree. This slightly increases mount time.
8172 int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info)
8174 struct btrfs_path *path;
8175 struct btrfs_root *root = fs_info->dev_root;
8176 struct btrfs_key key;
8178 u64 prev_dev_ext_end = 0;
8182 * We don't have a dev_root because we mounted with ignorebadroots and
8183 * failed to load the root, so we want to skip the verification in this
8186 * However if the dev root is fine, but the tree itself is corrupted
8187 * we'd still fail to mount. This verification is only to make sure
8188 * writes can happen safely, so instead just bypass this check
8189 * completely in the case of IGNOREBADROOTS.
8191 if (btrfs_test_opt(fs_info, IGNOREBADROOTS))
8195 key.type = BTRFS_DEV_EXTENT_KEY;
8198 path = btrfs_alloc_path();
8202 path->reada = READA_FORWARD;
8203 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
8207 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
8208 ret = btrfs_next_leaf(root, path);
8211 /* No dev extents at all? Not good */
8218 struct extent_buffer *leaf = path->nodes[0];
8219 struct btrfs_dev_extent *dext;
8220 int slot = path->slots[0];
8222 u64 physical_offset;
8226 btrfs_item_key_to_cpu(leaf, &key, slot);
8227 if (key.type != BTRFS_DEV_EXTENT_KEY)
8229 devid = key.objectid;
8230 physical_offset = key.offset;
8232 dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent);
8233 chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext);
8234 physical_len = btrfs_dev_extent_length(leaf, dext);
8236 /* Check if this dev extent overlaps with the previous one */
8237 if (devid == prev_devid && physical_offset < prev_dev_ext_end) {
8239 "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu",
8240 devid, physical_offset, prev_dev_ext_end);
8245 ret = verify_one_dev_extent(fs_info, chunk_offset, devid,
8246 physical_offset, physical_len);
8250 prev_dev_ext_end = physical_offset + physical_len;
8252 ret = btrfs_next_item(root, path);
8261 /* Ensure all chunks have corresponding dev extents */
8262 ret = verify_chunk_dev_extent_mapping(fs_info);
8264 btrfs_free_path(path);
8269 * Check whether the given block group or device is pinned by any inode being
8270 * used as a swapfile.
8272 bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr)
8274 struct btrfs_swapfile_pin *sp;
8275 struct rb_node *node;
8277 spin_lock(&fs_info->swapfile_pins_lock);
8278 node = fs_info->swapfile_pins.rb_node;
8280 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
8282 node = node->rb_left;
8283 else if (ptr > sp->ptr)
8284 node = node->rb_right;
8288 spin_unlock(&fs_info->swapfile_pins_lock);
8289 return node != NULL;
8292 static int relocating_repair_kthread(void *data)
8294 struct btrfs_block_group *cache = (struct btrfs_block_group *)data;
8295 struct btrfs_fs_info *fs_info = cache->fs_info;
8299 target = cache->start;
8300 btrfs_put_block_group(cache);
8302 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) {
8304 "zoned: skip relocating block group %llu to repair: EBUSY",
8309 mutex_lock(&fs_info->reclaim_bgs_lock);
8311 /* Ensure block group still exists */
8312 cache = btrfs_lookup_block_group(fs_info, target);
8316 if (!cache->relocating_repair)
8319 ret = btrfs_may_alloc_data_chunk(fs_info, target);
8324 "zoned: relocating block group %llu to repair IO failure",
8326 ret = btrfs_relocate_chunk(fs_info, target);
8330 btrfs_put_block_group(cache);
8331 mutex_unlock(&fs_info->reclaim_bgs_lock);
8332 btrfs_exclop_finish(fs_info);
8337 bool btrfs_repair_one_zone(struct btrfs_fs_info *fs_info, u64 logical)
8339 struct btrfs_block_group *cache;
8341 if (!btrfs_is_zoned(fs_info))
8344 /* Do not attempt to repair in degraded state */
8345 if (btrfs_test_opt(fs_info, DEGRADED))
8348 cache = btrfs_lookup_block_group(fs_info, logical);
8352 spin_lock(&cache->lock);
8353 if (cache->relocating_repair) {
8354 spin_unlock(&cache->lock);
8355 btrfs_put_block_group(cache);
8358 cache->relocating_repair = 1;
8359 spin_unlock(&cache->lock);
8361 kthread_run(relocating_repair_kthread, cache,
8362 "btrfs-relocating-repair");