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);
538 * Search and remove all stale devices (which are not mounted).
539 * When both inputs are NULL, it will search and release all stale devices.
541 * @devt: Optional. When provided will it release all unmounted devices
542 * matching this devt only.
543 * @skip_device: Optional. Will skip this device when searching for the stale
546 * Return: 0 for success or if @devt is 0.
547 * -EBUSY if @devt is a mounted device.
548 * -ENOENT if @devt does not match any device in the list.
550 static int btrfs_free_stale_devices(dev_t devt, struct btrfs_device *skip_device)
552 struct btrfs_fs_devices *fs_devices, *tmp_fs_devices;
553 struct btrfs_device *device, *tmp_device;
556 lockdep_assert_held(&uuid_mutex);
561 list_for_each_entry_safe(fs_devices, tmp_fs_devices, &fs_uuids, fs_list) {
563 mutex_lock(&fs_devices->device_list_mutex);
564 list_for_each_entry_safe(device, tmp_device,
565 &fs_devices->devices, dev_list) {
566 if (skip_device && skip_device == device)
568 if (devt && devt != device->devt)
570 if (fs_devices->opened) {
571 /* for an already deleted device return 0 */
572 if (devt && ret != 0)
577 /* delete the stale device */
578 fs_devices->num_devices--;
579 list_del(&device->dev_list);
580 btrfs_free_device(device);
584 mutex_unlock(&fs_devices->device_list_mutex);
586 if (fs_devices->num_devices == 0) {
587 btrfs_sysfs_remove_fsid(fs_devices);
588 list_del(&fs_devices->fs_list);
589 free_fs_devices(fs_devices);
597 * This is only used on mount, and we are protected from competing things
598 * messing with our fs_devices by the uuid_mutex, thus we do not need the
599 * fs_devices->device_list_mutex here.
601 static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices,
602 struct btrfs_device *device, fmode_t flags,
605 struct block_device *bdev;
606 struct btrfs_super_block *disk_super;
615 ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
620 devid = btrfs_stack_device_id(&disk_super->dev_item);
621 if (devid != device->devid)
622 goto error_free_page;
624 if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE))
625 goto error_free_page;
627 device->generation = btrfs_super_generation(disk_super);
629 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
630 if (btrfs_super_incompat_flags(disk_super) &
631 BTRFS_FEATURE_INCOMPAT_METADATA_UUID) {
633 "BTRFS: Invalid seeding and uuid-changed device detected\n");
634 goto error_free_page;
637 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
638 fs_devices->seeding = true;
640 if (bdev_read_only(bdev))
641 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
643 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
646 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
647 fs_devices->rotating = true;
650 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
651 device->mode = flags;
653 fs_devices->open_devices++;
654 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
655 device->devid != BTRFS_DEV_REPLACE_DEVID) {
656 fs_devices->rw_devices++;
657 list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list);
659 btrfs_release_disk_super(disk_super);
664 btrfs_release_disk_super(disk_super);
665 blkdev_put(bdev, flags);
671 * Handle scanned device having its CHANGING_FSID_V2 flag set and the fs_devices
672 * being created with a disk that has already completed its fsid change. Such
673 * disk can belong to an fs which has its FSID changed or to one which doesn't.
674 * Handle both cases here.
676 static struct btrfs_fs_devices *find_fsid_inprogress(
677 struct btrfs_super_block *disk_super)
679 struct btrfs_fs_devices *fs_devices;
681 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
682 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid,
683 BTRFS_FSID_SIZE) != 0 &&
684 memcmp(fs_devices->metadata_uuid, disk_super->fsid,
685 BTRFS_FSID_SIZE) == 0 && !fs_devices->fsid_change) {
690 return find_fsid(disk_super->fsid, NULL);
694 static struct btrfs_fs_devices *find_fsid_changed(
695 struct btrfs_super_block *disk_super)
697 struct btrfs_fs_devices *fs_devices;
700 * Handles the case where scanned device is part of an fs that had
701 * multiple successful changes of FSID but currently device didn't
702 * observe it. Meaning our fsid will be different than theirs. We need
703 * to handle two subcases :
704 * 1 - The fs still continues to have different METADATA/FSID uuids.
705 * 2 - The fs is switched back to its original FSID (METADATA/FSID
708 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
710 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid,
711 BTRFS_FSID_SIZE) != 0 &&
712 memcmp(fs_devices->metadata_uuid, disk_super->metadata_uuid,
713 BTRFS_FSID_SIZE) == 0 &&
714 memcmp(fs_devices->fsid, disk_super->fsid,
715 BTRFS_FSID_SIZE) != 0)
718 /* Unchanged UUIDs */
719 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid,
720 BTRFS_FSID_SIZE) == 0 &&
721 memcmp(fs_devices->fsid, disk_super->metadata_uuid,
722 BTRFS_FSID_SIZE) == 0)
729 static struct btrfs_fs_devices *find_fsid_reverted_metadata(
730 struct btrfs_super_block *disk_super)
732 struct btrfs_fs_devices *fs_devices;
735 * Handle the case where the scanned device is part of an fs whose last
736 * metadata UUID change reverted it to the original FSID. At the same
737 * time * fs_devices was first created by another constitutent device
738 * which didn't fully observe the operation. This results in an
739 * btrfs_fs_devices created with metadata/fsid different AND
740 * btrfs_fs_devices::fsid_change set AND the metadata_uuid of the
741 * fs_devices equal to the FSID of the disk.
743 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
744 if (memcmp(fs_devices->fsid, fs_devices->metadata_uuid,
745 BTRFS_FSID_SIZE) != 0 &&
746 memcmp(fs_devices->metadata_uuid, disk_super->fsid,
747 BTRFS_FSID_SIZE) == 0 &&
748 fs_devices->fsid_change)
755 * Add new device to list of registered devices
758 * device pointer which was just added or updated when successful
759 * error pointer when failed
761 static noinline struct btrfs_device *device_list_add(const char *path,
762 struct btrfs_super_block *disk_super,
763 bool *new_device_added)
765 struct btrfs_device *device;
766 struct btrfs_fs_devices *fs_devices = NULL;
767 struct rcu_string *name;
768 u64 found_transid = btrfs_super_generation(disk_super);
769 u64 devid = btrfs_stack_device_id(&disk_super->dev_item);
772 bool has_metadata_uuid = (btrfs_super_incompat_flags(disk_super) &
773 BTRFS_FEATURE_INCOMPAT_METADATA_UUID);
774 bool fsid_change_in_progress = (btrfs_super_flags(disk_super) &
775 BTRFS_SUPER_FLAG_CHANGING_FSID_V2);
777 error = lookup_bdev(path, &path_devt);
779 return ERR_PTR(error);
781 if (fsid_change_in_progress) {
782 if (!has_metadata_uuid)
783 fs_devices = find_fsid_inprogress(disk_super);
785 fs_devices = find_fsid_changed(disk_super);
786 } else if (has_metadata_uuid) {
787 fs_devices = find_fsid_with_metadata_uuid(disk_super);
789 fs_devices = find_fsid_reverted_metadata(disk_super);
791 fs_devices = find_fsid(disk_super->fsid, NULL);
796 if (has_metadata_uuid)
797 fs_devices = alloc_fs_devices(disk_super->fsid,
798 disk_super->metadata_uuid);
800 fs_devices = alloc_fs_devices(disk_super->fsid, NULL);
802 if (IS_ERR(fs_devices))
803 return ERR_CAST(fs_devices);
805 fs_devices->fsid_change = fsid_change_in_progress;
807 mutex_lock(&fs_devices->device_list_mutex);
808 list_add(&fs_devices->fs_list, &fs_uuids);
812 struct btrfs_dev_lookup_args args = {
814 .uuid = disk_super->dev_item.uuid,
817 mutex_lock(&fs_devices->device_list_mutex);
818 device = btrfs_find_device(fs_devices, &args);
821 * If this disk has been pulled into an fs devices created by
822 * a device which had the CHANGING_FSID_V2 flag then replace the
823 * metadata_uuid/fsid values of the fs_devices.
825 if (fs_devices->fsid_change &&
826 found_transid > fs_devices->latest_generation) {
827 memcpy(fs_devices->fsid, disk_super->fsid,
830 if (has_metadata_uuid)
831 memcpy(fs_devices->metadata_uuid,
832 disk_super->metadata_uuid,
835 memcpy(fs_devices->metadata_uuid,
836 disk_super->fsid, BTRFS_FSID_SIZE);
838 fs_devices->fsid_change = false;
843 if (fs_devices->opened) {
844 mutex_unlock(&fs_devices->device_list_mutex);
845 return ERR_PTR(-EBUSY);
848 device = btrfs_alloc_device(NULL, &devid,
849 disk_super->dev_item.uuid);
850 if (IS_ERR(device)) {
851 mutex_unlock(&fs_devices->device_list_mutex);
852 /* we can safely leave the fs_devices entry around */
856 name = rcu_string_strdup(path, GFP_NOFS);
858 btrfs_free_device(device);
859 mutex_unlock(&fs_devices->device_list_mutex);
860 return ERR_PTR(-ENOMEM);
862 rcu_assign_pointer(device->name, name);
863 device->devt = path_devt;
865 list_add_rcu(&device->dev_list, &fs_devices->devices);
866 fs_devices->num_devices++;
868 device->fs_devices = fs_devices;
869 *new_device_added = true;
871 if (disk_super->label[0])
873 "BTRFS: device label %s devid %llu transid %llu %s scanned by %s (%d)\n",
874 disk_super->label, devid, found_transid, path,
875 current->comm, task_pid_nr(current));
878 "BTRFS: device fsid %pU devid %llu transid %llu %s scanned by %s (%d)\n",
879 disk_super->fsid, devid, found_transid, path,
880 current->comm, task_pid_nr(current));
882 } else if (!device->name || strcmp(device->name->str, path)) {
884 * When FS is already mounted.
885 * 1. If you are here and if the device->name is NULL that
886 * means this device was missing at time of FS mount.
887 * 2. If you are here and if the device->name is different
888 * from 'path' that means either
889 * a. The same device disappeared and reappeared with
891 * b. The missing-disk-which-was-replaced, has
894 * We must allow 1 and 2a above. But 2b would be a spurious
897 * Further in case of 1 and 2a above, the disk at 'path'
898 * would have missed some transaction when it was away and
899 * in case of 2a the stale bdev has to be updated as well.
900 * 2b must not be allowed at all time.
904 * For now, we do allow update to btrfs_fs_device through the
905 * btrfs dev scan cli after FS has been mounted. We're still
906 * tracking a problem where systems fail mount by subvolume id
907 * when we reject replacement on a mounted FS.
909 if (!fs_devices->opened && found_transid < device->generation) {
911 * That is if the FS is _not_ mounted and if you
912 * are here, that means there is more than one
913 * disk with same uuid and devid.We keep the one
914 * with larger generation number or the last-in if
915 * generation are equal.
917 mutex_unlock(&fs_devices->device_list_mutex);
918 return ERR_PTR(-EEXIST);
922 * We are going to replace the device path for a given devid,
923 * make sure it's the same device if the device is mounted
925 * NOTE: the device->fs_info may not be reliable here so pass
926 * in a NULL to message helpers instead. This avoids a possible
927 * use-after-free when the fs_info and fs_info->sb are already
931 if (device->devt != path_devt) {
932 mutex_unlock(&fs_devices->device_list_mutex);
933 btrfs_warn_in_rcu(NULL,
934 "duplicate device %s devid %llu generation %llu scanned by %s (%d)",
935 path, devid, found_transid,
937 task_pid_nr(current));
938 return ERR_PTR(-EEXIST);
940 btrfs_info_in_rcu(NULL,
941 "devid %llu device path %s changed to %s scanned by %s (%d)",
942 devid, rcu_str_deref(device->name),
944 task_pid_nr(current));
947 name = rcu_string_strdup(path, GFP_NOFS);
949 mutex_unlock(&fs_devices->device_list_mutex);
950 return ERR_PTR(-ENOMEM);
952 rcu_string_free(device->name);
953 rcu_assign_pointer(device->name, name);
954 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
955 fs_devices->missing_devices--;
956 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
958 device->devt = path_devt;
962 * Unmount does not free the btrfs_device struct but would zero
963 * generation along with most of the other members. So just update
964 * it back. We need it to pick the disk with largest generation
967 if (!fs_devices->opened) {
968 device->generation = found_transid;
969 fs_devices->latest_generation = max_t(u64, found_transid,
970 fs_devices->latest_generation);
973 fs_devices->total_devices = btrfs_super_num_devices(disk_super);
975 mutex_unlock(&fs_devices->device_list_mutex);
979 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
981 struct btrfs_fs_devices *fs_devices;
982 struct btrfs_device *device;
983 struct btrfs_device *orig_dev;
986 lockdep_assert_held(&uuid_mutex);
988 fs_devices = alloc_fs_devices(orig->fsid, NULL);
989 if (IS_ERR(fs_devices))
992 fs_devices->total_devices = orig->total_devices;
994 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
995 struct rcu_string *name;
997 device = btrfs_alloc_device(NULL, &orig_dev->devid,
999 if (IS_ERR(device)) {
1000 ret = PTR_ERR(device);
1005 * This is ok to do without rcu read locked because we hold the
1006 * uuid mutex so nothing we touch in here is going to disappear.
1008 if (orig_dev->name) {
1009 name = rcu_string_strdup(orig_dev->name->str,
1012 btrfs_free_device(device);
1016 rcu_assign_pointer(device->name, name);
1019 list_add(&device->dev_list, &fs_devices->devices);
1020 device->fs_devices = fs_devices;
1021 fs_devices->num_devices++;
1025 free_fs_devices(fs_devices);
1026 return ERR_PTR(ret);
1029 static void __btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices,
1030 struct btrfs_device **latest_dev)
1032 struct btrfs_device *device, *next;
1034 /* This is the initialized path, it is safe to release the devices. */
1035 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
1036 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)) {
1037 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
1038 &device->dev_state) &&
1039 !test_bit(BTRFS_DEV_STATE_MISSING,
1040 &device->dev_state) &&
1042 device->generation > (*latest_dev)->generation)) {
1043 *latest_dev = device;
1049 * We have already validated the presence of BTRFS_DEV_REPLACE_DEVID,
1050 * in btrfs_init_dev_replace() so just continue.
1052 if (device->devid == BTRFS_DEV_REPLACE_DEVID)
1056 blkdev_put(device->bdev, device->mode);
1057 device->bdev = NULL;
1058 fs_devices->open_devices--;
1060 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1061 list_del_init(&device->dev_alloc_list);
1062 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
1063 fs_devices->rw_devices--;
1065 list_del_init(&device->dev_list);
1066 fs_devices->num_devices--;
1067 btrfs_free_device(device);
1073 * After we have read the system tree and know devids belonging to this
1074 * filesystem, remove the device which does not belong there.
1076 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices)
1078 struct btrfs_device *latest_dev = NULL;
1079 struct btrfs_fs_devices *seed_dev;
1081 mutex_lock(&uuid_mutex);
1082 __btrfs_free_extra_devids(fs_devices, &latest_dev);
1084 list_for_each_entry(seed_dev, &fs_devices->seed_list, seed_list)
1085 __btrfs_free_extra_devids(seed_dev, &latest_dev);
1087 fs_devices->latest_dev = latest_dev;
1089 mutex_unlock(&uuid_mutex);
1092 static void btrfs_close_bdev(struct btrfs_device *device)
1097 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1098 sync_blockdev(device->bdev);
1099 invalidate_bdev(device->bdev);
1102 blkdev_put(device->bdev, device->mode);
1105 static void btrfs_close_one_device(struct btrfs_device *device)
1107 struct btrfs_fs_devices *fs_devices = device->fs_devices;
1109 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
1110 device->devid != BTRFS_DEV_REPLACE_DEVID) {
1111 list_del_init(&device->dev_alloc_list);
1112 fs_devices->rw_devices--;
1115 if (device->devid == BTRFS_DEV_REPLACE_DEVID)
1116 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
1118 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
1119 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
1120 fs_devices->missing_devices--;
1123 btrfs_close_bdev(device);
1125 fs_devices->open_devices--;
1126 device->bdev = NULL;
1128 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
1129 btrfs_destroy_dev_zone_info(device);
1131 device->fs_info = NULL;
1132 atomic_set(&device->dev_stats_ccnt, 0);
1133 extent_io_tree_release(&device->alloc_state);
1136 * Reset the flush error record. We might have a transient flush error
1137 * in this mount, and if so we aborted the current transaction and set
1138 * the fs to an error state, guaranteeing no super blocks can be further
1139 * committed. However that error might be transient and if we unmount the
1140 * filesystem and mount it again, we should allow the mount to succeed
1141 * (btrfs_check_rw_degradable() should not fail) - if after mounting the
1142 * filesystem again we still get flush errors, then we will again abort
1143 * any transaction and set the error state, guaranteeing no commits of
1144 * unsafe super blocks.
1146 device->last_flush_error = 0;
1148 /* Verify the device is back in a pristine state */
1149 ASSERT(!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state));
1150 ASSERT(!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state));
1151 ASSERT(list_empty(&device->dev_alloc_list));
1152 ASSERT(list_empty(&device->post_commit_list));
1155 static void close_fs_devices(struct btrfs_fs_devices *fs_devices)
1157 struct btrfs_device *device, *tmp;
1159 lockdep_assert_held(&uuid_mutex);
1161 if (--fs_devices->opened > 0)
1164 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list)
1165 btrfs_close_one_device(device);
1167 WARN_ON(fs_devices->open_devices);
1168 WARN_ON(fs_devices->rw_devices);
1169 fs_devices->opened = 0;
1170 fs_devices->seeding = false;
1171 fs_devices->fs_info = NULL;
1174 void btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
1177 struct btrfs_fs_devices *tmp;
1179 mutex_lock(&uuid_mutex);
1180 close_fs_devices(fs_devices);
1181 if (!fs_devices->opened)
1182 list_splice_init(&fs_devices->seed_list, &list);
1184 list_for_each_entry_safe(fs_devices, tmp, &list, seed_list) {
1185 close_fs_devices(fs_devices);
1186 list_del(&fs_devices->seed_list);
1187 free_fs_devices(fs_devices);
1189 mutex_unlock(&uuid_mutex);
1192 static int open_fs_devices(struct btrfs_fs_devices *fs_devices,
1193 fmode_t flags, void *holder)
1195 struct btrfs_device *device;
1196 struct btrfs_device *latest_dev = NULL;
1197 struct btrfs_device *tmp_device;
1199 flags |= FMODE_EXCL;
1201 list_for_each_entry_safe(device, tmp_device, &fs_devices->devices,
1205 ret = btrfs_open_one_device(fs_devices, device, flags, holder);
1207 (!latest_dev || device->generation > latest_dev->generation)) {
1208 latest_dev = device;
1209 } else if (ret == -ENODATA) {
1210 fs_devices->num_devices--;
1211 list_del(&device->dev_list);
1212 btrfs_free_device(device);
1215 if (fs_devices->open_devices == 0)
1218 fs_devices->opened = 1;
1219 fs_devices->latest_dev = latest_dev;
1220 fs_devices->total_rw_bytes = 0;
1221 fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_REGULAR;
1222 fs_devices->read_policy = BTRFS_READ_POLICY_PID;
1227 static int devid_cmp(void *priv, const struct list_head *a,
1228 const struct list_head *b)
1230 const struct btrfs_device *dev1, *dev2;
1232 dev1 = list_entry(a, struct btrfs_device, dev_list);
1233 dev2 = list_entry(b, struct btrfs_device, dev_list);
1235 if (dev1->devid < dev2->devid)
1237 else if (dev1->devid > dev2->devid)
1242 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1243 fmode_t flags, void *holder)
1247 lockdep_assert_held(&uuid_mutex);
1249 * The device_list_mutex cannot be taken here in case opening the
1250 * underlying device takes further locks like open_mutex.
1252 * We also don't need the lock here as this is called during mount and
1253 * exclusion is provided by uuid_mutex
1256 if (fs_devices->opened) {
1257 fs_devices->opened++;
1260 list_sort(NULL, &fs_devices->devices, devid_cmp);
1261 ret = open_fs_devices(fs_devices, flags, holder);
1267 void btrfs_release_disk_super(struct btrfs_super_block *super)
1269 struct page *page = virt_to_page(super);
1274 static struct btrfs_super_block *btrfs_read_disk_super(struct block_device *bdev,
1275 u64 bytenr, u64 bytenr_orig)
1277 struct btrfs_super_block *disk_super;
1282 /* make sure our super fits in the device */
1283 if (bytenr + PAGE_SIZE >= bdev_nr_bytes(bdev))
1284 return ERR_PTR(-EINVAL);
1286 /* make sure our super fits in the page */
1287 if (sizeof(*disk_super) > PAGE_SIZE)
1288 return ERR_PTR(-EINVAL);
1290 /* make sure our super doesn't straddle pages on disk */
1291 index = bytenr >> PAGE_SHIFT;
1292 if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_SHIFT != index)
1293 return ERR_PTR(-EINVAL);
1295 /* pull in the page with our super */
1296 page = read_cache_page_gfp(bdev->bd_inode->i_mapping, index, GFP_KERNEL);
1299 return ERR_CAST(page);
1301 p = page_address(page);
1303 /* align our pointer to the offset of the super block */
1304 disk_super = p + offset_in_page(bytenr);
1306 if (btrfs_super_bytenr(disk_super) != bytenr_orig ||
1307 btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
1308 btrfs_release_disk_super(p);
1309 return ERR_PTR(-EINVAL);
1312 if (disk_super->label[0] && disk_super->label[BTRFS_LABEL_SIZE - 1])
1313 disk_super->label[BTRFS_LABEL_SIZE - 1] = 0;
1318 int btrfs_forget_devices(dev_t devt)
1322 mutex_lock(&uuid_mutex);
1323 ret = btrfs_free_stale_devices(devt, NULL);
1324 mutex_unlock(&uuid_mutex);
1330 * Look for a btrfs signature on a device. This may be called out of the mount path
1331 * and we are not allowed to call set_blocksize during the scan. The superblock
1332 * is read via pagecache
1334 struct btrfs_device *btrfs_scan_one_device(const char *path, fmode_t flags,
1337 struct btrfs_super_block *disk_super;
1338 bool new_device_added = false;
1339 struct btrfs_device *device = NULL;
1340 struct block_device *bdev;
1341 u64 bytenr, bytenr_orig;
1344 lockdep_assert_held(&uuid_mutex);
1347 * we would like to check all the supers, but that would make
1348 * a btrfs mount succeed after a mkfs from a different FS.
1349 * So, we need to add a special mount option to scan for
1350 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1352 flags |= FMODE_EXCL;
1354 bdev = blkdev_get_by_path(path, flags, holder);
1356 return ERR_CAST(bdev);
1358 bytenr_orig = btrfs_sb_offset(0);
1359 ret = btrfs_sb_log_location_bdev(bdev, 0, READ, &bytenr);
1361 device = ERR_PTR(ret);
1362 goto error_bdev_put;
1365 disk_super = btrfs_read_disk_super(bdev, bytenr, bytenr_orig);
1366 if (IS_ERR(disk_super)) {
1367 device = ERR_CAST(disk_super);
1368 goto error_bdev_put;
1371 device = device_list_add(path, disk_super, &new_device_added);
1372 if (!IS_ERR(device) && new_device_added)
1373 btrfs_free_stale_devices(device->devt, device);
1375 btrfs_release_disk_super(disk_super);
1378 blkdev_put(bdev, flags);
1384 * Try to find a chunk that intersects [start, start + len] range and when one
1385 * such is found, record the end of it in *start
1387 static bool contains_pending_extent(struct btrfs_device *device, u64 *start,
1390 u64 physical_start, physical_end;
1392 lockdep_assert_held(&device->fs_info->chunk_mutex);
1394 if (!find_first_extent_bit(&device->alloc_state, *start,
1395 &physical_start, &physical_end,
1396 CHUNK_ALLOCATED, NULL)) {
1398 if (in_range(physical_start, *start, len) ||
1399 in_range(*start, physical_start,
1400 physical_end - physical_start)) {
1401 *start = physical_end + 1;
1408 static u64 dev_extent_search_start(struct btrfs_device *device, u64 start)
1410 switch (device->fs_devices->chunk_alloc_policy) {
1411 case BTRFS_CHUNK_ALLOC_REGULAR:
1413 * We don't want to overwrite the superblock on the drive nor
1414 * any area used by the boot loader (grub for example), so we
1415 * make sure to start at an offset of at least 1MB.
1417 return max_t(u64, start, SZ_1M);
1418 case BTRFS_CHUNK_ALLOC_ZONED:
1420 * We don't care about the starting region like regular
1421 * allocator, because we anyway use/reserve the first two zones
1422 * for superblock logging.
1424 return ALIGN(start, device->zone_info->zone_size);
1430 static bool dev_extent_hole_check_zoned(struct btrfs_device *device,
1431 u64 *hole_start, u64 *hole_size,
1434 u64 zone_size = device->zone_info->zone_size;
1437 bool changed = false;
1439 ASSERT(IS_ALIGNED(*hole_start, zone_size));
1441 while (*hole_size > 0) {
1442 pos = btrfs_find_allocatable_zones(device, *hole_start,
1443 *hole_start + *hole_size,
1445 if (pos != *hole_start) {
1446 *hole_size = *hole_start + *hole_size - pos;
1449 if (*hole_size < num_bytes)
1453 ret = btrfs_ensure_empty_zones(device, pos, num_bytes);
1455 /* Range is ensured to be empty */
1459 /* Given hole range was invalid (outside of device) */
1460 if (ret == -ERANGE) {
1461 *hole_start += *hole_size;
1466 *hole_start += zone_size;
1467 *hole_size -= zone_size;
1475 * dev_extent_hole_check - check if specified hole is suitable for allocation
1476 * @device: the device which we have the hole
1477 * @hole_start: starting position of the hole
1478 * @hole_size: the size of the hole
1479 * @num_bytes: the size of the free space that we need
1481 * This function may modify @hole_start and @hole_size to reflect the suitable
1482 * position for allocation. Returns 1 if hole position is updated, 0 otherwise.
1484 static bool dev_extent_hole_check(struct btrfs_device *device, u64 *hole_start,
1485 u64 *hole_size, u64 num_bytes)
1487 bool changed = false;
1488 u64 hole_end = *hole_start + *hole_size;
1492 * Check before we set max_hole_start, otherwise we could end up
1493 * sending back this offset anyway.
1495 if (contains_pending_extent(device, hole_start, *hole_size)) {
1496 if (hole_end >= *hole_start)
1497 *hole_size = hole_end - *hole_start;
1503 switch (device->fs_devices->chunk_alloc_policy) {
1504 case BTRFS_CHUNK_ALLOC_REGULAR:
1505 /* No extra check */
1507 case BTRFS_CHUNK_ALLOC_ZONED:
1508 if (dev_extent_hole_check_zoned(device, hole_start,
1509 hole_size, num_bytes)) {
1512 * The changed hole can contain pending extent.
1513 * Loop again to check that.
1529 * find_free_dev_extent_start - find free space in the specified device
1530 * @device: the device which we search the free space in
1531 * @num_bytes: the size of the free space that we need
1532 * @search_start: the position from which to begin the search
1533 * @start: store the start of the free space.
1534 * @len: the size of the free space. that we find, or the size
1535 * of the max free space if we don't find suitable free space
1537 * this uses a pretty simple search, the expectation is that it is
1538 * called very infrequently and that a given device has a small number
1541 * @start is used to store the start of the free space if we find. But if we
1542 * don't find suitable free space, it will be used to store the start position
1543 * of the max free space.
1545 * @len is used to store the size of the free space that we find.
1546 * But if we don't find suitable free space, it is used to store the size of
1547 * the max free space.
1549 * NOTE: This function will search *commit* root of device tree, and does extra
1550 * check to ensure dev extents are not double allocated.
1551 * This makes the function safe to allocate dev extents but may not report
1552 * correct usable device space, as device extent freed in current transaction
1553 * is not reported as available.
1555 static int find_free_dev_extent_start(struct btrfs_device *device,
1556 u64 num_bytes, u64 search_start, u64 *start,
1559 struct btrfs_fs_info *fs_info = device->fs_info;
1560 struct btrfs_root *root = fs_info->dev_root;
1561 struct btrfs_key key;
1562 struct btrfs_dev_extent *dev_extent;
1563 struct btrfs_path *path;
1568 u64 search_end = device->total_bytes;
1571 struct extent_buffer *l;
1573 search_start = dev_extent_search_start(device, search_start);
1575 WARN_ON(device->zone_info &&
1576 !IS_ALIGNED(num_bytes, device->zone_info->zone_size));
1578 path = btrfs_alloc_path();
1582 max_hole_start = search_start;
1586 if (search_start >= search_end ||
1587 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1592 path->reada = READA_FORWARD;
1593 path->search_commit_root = 1;
1594 path->skip_locking = 1;
1596 key.objectid = device->devid;
1597 key.offset = search_start;
1598 key.type = BTRFS_DEV_EXTENT_KEY;
1600 ret = btrfs_search_backwards(root, &key, path);
1606 slot = path->slots[0];
1607 if (slot >= btrfs_header_nritems(l)) {
1608 ret = btrfs_next_leaf(root, path);
1616 btrfs_item_key_to_cpu(l, &key, slot);
1618 if (key.objectid < device->devid)
1621 if (key.objectid > device->devid)
1624 if (key.type != BTRFS_DEV_EXTENT_KEY)
1627 if (key.offset > search_start) {
1628 hole_size = key.offset - search_start;
1629 dev_extent_hole_check(device, &search_start, &hole_size,
1632 if (hole_size > max_hole_size) {
1633 max_hole_start = search_start;
1634 max_hole_size = hole_size;
1638 * If this free space is greater than which we need,
1639 * it must be the max free space that we have found
1640 * until now, so max_hole_start must point to the start
1641 * of this free space and the length of this free space
1642 * is stored in max_hole_size. Thus, we return
1643 * max_hole_start and max_hole_size and go back to the
1646 if (hole_size >= num_bytes) {
1652 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1653 extent_end = key.offset + btrfs_dev_extent_length(l,
1655 if (extent_end > search_start)
1656 search_start = extent_end;
1663 * At this point, search_start should be the end of
1664 * allocated dev extents, and when shrinking the device,
1665 * search_end may be smaller than search_start.
1667 if (search_end > search_start) {
1668 hole_size = search_end - search_start;
1669 if (dev_extent_hole_check(device, &search_start, &hole_size,
1671 btrfs_release_path(path);
1675 if (hole_size > max_hole_size) {
1676 max_hole_start = search_start;
1677 max_hole_size = hole_size;
1682 if (max_hole_size < num_bytes)
1688 btrfs_free_path(path);
1689 *start = max_hole_start;
1691 *len = max_hole_size;
1695 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
1696 u64 *start, u64 *len)
1698 /* FIXME use last free of some kind */
1699 return find_free_dev_extent_start(device, num_bytes, 0, start, len);
1702 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1703 struct btrfs_device *device,
1704 u64 start, u64 *dev_extent_len)
1706 struct btrfs_fs_info *fs_info = device->fs_info;
1707 struct btrfs_root *root = fs_info->dev_root;
1709 struct btrfs_path *path;
1710 struct btrfs_key key;
1711 struct btrfs_key found_key;
1712 struct extent_buffer *leaf = NULL;
1713 struct btrfs_dev_extent *extent = NULL;
1715 path = btrfs_alloc_path();
1719 key.objectid = device->devid;
1721 key.type = BTRFS_DEV_EXTENT_KEY;
1723 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1725 ret = btrfs_previous_item(root, path, key.objectid,
1726 BTRFS_DEV_EXTENT_KEY);
1729 leaf = path->nodes[0];
1730 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1731 extent = btrfs_item_ptr(leaf, path->slots[0],
1732 struct btrfs_dev_extent);
1733 BUG_ON(found_key.offset > start || found_key.offset +
1734 btrfs_dev_extent_length(leaf, extent) < start);
1736 btrfs_release_path(path);
1738 } else if (ret == 0) {
1739 leaf = path->nodes[0];
1740 extent = btrfs_item_ptr(leaf, path->slots[0],
1741 struct btrfs_dev_extent);
1746 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1748 ret = btrfs_del_item(trans, root, path);
1750 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1752 btrfs_free_path(path);
1756 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1758 struct extent_map_tree *em_tree;
1759 struct extent_map *em;
1763 em_tree = &fs_info->mapping_tree;
1764 read_lock(&em_tree->lock);
1765 n = rb_last(&em_tree->map.rb_root);
1767 em = rb_entry(n, struct extent_map, rb_node);
1768 ret = em->start + em->len;
1770 read_unlock(&em_tree->lock);
1775 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1779 struct btrfs_key key;
1780 struct btrfs_key found_key;
1781 struct btrfs_path *path;
1783 path = btrfs_alloc_path();
1787 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1788 key.type = BTRFS_DEV_ITEM_KEY;
1789 key.offset = (u64)-1;
1791 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1797 btrfs_err(fs_info, "corrupted chunk tree devid -1 matched");
1802 ret = btrfs_previous_item(fs_info->chunk_root, path,
1803 BTRFS_DEV_ITEMS_OBJECTID,
1804 BTRFS_DEV_ITEM_KEY);
1808 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1810 *devid_ret = found_key.offset + 1;
1814 btrfs_free_path(path);
1819 * the device information is stored in the chunk root
1820 * the btrfs_device struct should be fully filled in
1822 static int btrfs_add_dev_item(struct btrfs_trans_handle *trans,
1823 struct btrfs_device *device)
1826 struct btrfs_path *path;
1827 struct btrfs_dev_item *dev_item;
1828 struct extent_buffer *leaf;
1829 struct btrfs_key key;
1832 path = btrfs_alloc_path();
1836 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1837 key.type = BTRFS_DEV_ITEM_KEY;
1838 key.offset = device->devid;
1840 btrfs_reserve_chunk_metadata(trans, true);
1841 ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path,
1842 &key, sizeof(*dev_item));
1843 btrfs_trans_release_chunk_metadata(trans);
1847 leaf = path->nodes[0];
1848 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1850 btrfs_set_device_id(leaf, dev_item, device->devid);
1851 btrfs_set_device_generation(leaf, dev_item, 0);
1852 btrfs_set_device_type(leaf, dev_item, device->type);
1853 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1854 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1855 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1856 btrfs_set_device_total_bytes(leaf, dev_item,
1857 btrfs_device_get_disk_total_bytes(device));
1858 btrfs_set_device_bytes_used(leaf, dev_item,
1859 btrfs_device_get_bytes_used(device));
1860 btrfs_set_device_group(leaf, dev_item, 0);
1861 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1862 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1863 btrfs_set_device_start_offset(leaf, dev_item, 0);
1865 ptr = btrfs_device_uuid(dev_item);
1866 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1867 ptr = btrfs_device_fsid(dev_item);
1868 write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid,
1869 ptr, BTRFS_FSID_SIZE);
1870 btrfs_mark_buffer_dirty(leaf);
1874 btrfs_free_path(path);
1879 * Function to update ctime/mtime for a given device path.
1880 * Mainly used for ctime/mtime based probe like libblkid.
1882 * We don't care about errors here, this is just to be kind to userspace.
1884 static void update_dev_time(const char *device_path)
1887 struct timespec64 now;
1890 ret = kern_path(device_path, LOOKUP_FOLLOW, &path);
1894 now = current_time(d_inode(path.dentry));
1895 inode_update_time(d_inode(path.dentry), &now, S_MTIME | S_CTIME);
1899 static int btrfs_rm_dev_item(struct btrfs_trans_handle *trans,
1900 struct btrfs_device *device)
1902 struct btrfs_root *root = device->fs_info->chunk_root;
1904 struct btrfs_path *path;
1905 struct btrfs_key key;
1907 path = btrfs_alloc_path();
1911 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1912 key.type = BTRFS_DEV_ITEM_KEY;
1913 key.offset = device->devid;
1915 btrfs_reserve_chunk_metadata(trans, false);
1916 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1917 btrfs_trans_release_chunk_metadata(trans);
1924 ret = btrfs_del_item(trans, root, path);
1926 btrfs_free_path(path);
1931 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1932 * filesystem. It's up to the caller to adjust that number regarding eg. device
1935 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1943 seq = read_seqbegin(&fs_info->profiles_lock);
1945 all_avail = fs_info->avail_data_alloc_bits |
1946 fs_info->avail_system_alloc_bits |
1947 fs_info->avail_metadata_alloc_bits;
1948 } while (read_seqretry(&fs_info->profiles_lock, seq));
1950 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1951 if (!(all_avail & btrfs_raid_array[i].bg_flag))
1954 if (num_devices < btrfs_raid_array[i].devs_min)
1955 return btrfs_raid_array[i].mindev_error;
1961 static struct btrfs_device * btrfs_find_next_active_device(
1962 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device)
1964 struct btrfs_device *next_device;
1966 list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1967 if (next_device != device &&
1968 !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state)
1969 && next_device->bdev)
1977 * Helper function to check if the given device is part of s_bdev / latest_dev
1978 * and replace it with the provided or the next active device, in the context
1979 * where this function called, there should be always be another device (or
1980 * this_dev) which is active.
1982 void __cold btrfs_assign_next_active_device(struct btrfs_device *device,
1983 struct btrfs_device *next_device)
1985 struct btrfs_fs_info *fs_info = device->fs_info;
1988 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
1990 ASSERT(next_device);
1992 if (fs_info->sb->s_bdev &&
1993 (fs_info->sb->s_bdev == device->bdev))
1994 fs_info->sb->s_bdev = next_device->bdev;
1996 if (fs_info->fs_devices->latest_dev->bdev == device->bdev)
1997 fs_info->fs_devices->latest_dev = next_device;
2001 * Return btrfs_fs_devices::num_devices excluding the device that's being
2002 * currently replaced.
2004 static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info)
2006 u64 num_devices = fs_info->fs_devices->num_devices;
2008 down_read(&fs_info->dev_replace.rwsem);
2009 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
2010 ASSERT(num_devices > 1);
2013 up_read(&fs_info->dev_replace.rwsem);
2018 void btrfs_scratch_superblocks(struct btrfs_fs_info *fs_info,
2019 struct block_device *bdev,
2020 const char *device_path)
2022 struct btrfs_super_block *disk_super;
2028 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; copy_num++) {
2032 disk_super = btrfs_read_dev_one_super(bdev, copy_num);
2033 if (IS_ERR(disk_super))
2036 if (bdev_is_zoned(bdev)) {
2037 btrfs_reset_sb_log_zones(bdev, copy_num);
2041 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
2043 page = virt_to_page(disk_super);
2044 set_page_dirty(page);
2046 /* write_on_page() unlocks the page */
2047 ret = write_one_page(page);
2050 "error clearing superblock number %d (%d)",
2052 btrfs_release_disk_super(disk_super);
2056 /* Notify udev that device has changed */
2057 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
2059 /* Update ctime/mtime for device path for libblkid */
2060 update_dev_time(device_path);
2063 int btrfs_rm_device(struct btrfs_fs_info *fs_info,
2064 struct btrfs_dev_lookup_args *args,
2065 struct block_device **bdev, fmode_t *mode)
2067 struct btrfs_trans_handle *trans;
2068 struct btrfs_device *device;
2069 struct btrfs_fs_devices *cur_devices;
2070 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2074 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
2075 btrfs_err(fs_info, "device remove not supported on extent tree v2 yet");
2080 * The device list in fs_devices is accessed without locks (neither
2081 * uuid_mutex nor device_list_mutex) as it won't change on a mounted
2082 * filesystem and another device rm cannot run.
2084 num_devices = btrfs_num_devices(fs_info);
2086 ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
2090 device = btrfs_find_device(fs_info->fs_devices, args);
2093 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2099 if (btrfs_pinned_by_swapfile(fs_info, device)) {
2100 btrfs_warn_in_rcu(fs_info,
2101 "cannot remove device %s (devid %llu) due to active swapfile",
2102 rcu_str_deref(device->name), device->devid);
2106 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
2107 return BTRFS_ERROR_DEV_TGT_REPLACE;
2109 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
2110 fs_info->fs_devices->rw_devices == 1)
2111 return BTRFS_ERROR_DEV_ONLY_WRITABLE;
2113 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2114 mutex_lock(&fs_info->chunk_mutex);
2115 list_del_init(&device->dev_alloc_list);
2116 device->fs_devices->rw_devices--;
2117 mutex_unlock(&fs_info->chunk_mutex);
2120 ret = btrfs_shrink_device(device, 0);
2124 trans = btrfs_start_transaction(fs_info->chunk_root, 0);
2125 if (IS_ERR(trans)) {
2126 ret = PTR_ERR(trans);
2130 ret = btrfs_rm_dev_item(trans, device);
2132 /* Any error in dev item removal is critical */
2134 "failed to remove device item for devid %llu: %d",
2135 device->devid, ret);
2136 btrfs_abort_transaction(trans, ret);
2137 btrfs_end_transaction(trans);
2141 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2142 btrfs_scrub_cancel_dev(device);
2145 * the device list mutex makes sure that we don't change
2146 * the device list while someone else is writing out all
2147 * the device supers. Whoever is writing all supers, should
2148 * lock the device list mutex before getting the number of
2149 * devices in the super block (super_copy). Conversely,
2150 * whoever updates the number of devices in the super block
2151 * (super_copy) should hold the device list mutex.
2155 * In normal cases the cur_devices == fs_devices. But in case
2156 * of deleting a seed device, the cur_devices should point to
2157 * its own fs_devices listed under the fs_devices->seed_list.
2159 cur_devices = device->fs_devices;
2160 mutex_lock(&fs_devices->device_list_mutex);
2161 list_del_rcu(&device->dev_list);
2163 cur_devices->num_devices--;
2164 cur_devices->total_devices--;
2165 /* Update total_devices of the parent fs_devices if it's seed */
2166 if (cur_devices != fs_devices)
2167 fs_devices->total_devices--;
2169 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
2170 cur_devices->missing_devices--;
2172 btrfs_assign_next_active_device(device, NULL);
2175 cur_devices->open_devices--;
2176 /* remove sysfs entry */
2177 btrfs_sysfs_remove_device(device);
2180 num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
2181 btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
2182 mutex_unlock(&fs_devices->device_list_mutex);
2185 * At this point, the device is zero sized and detached from the
2186 * devices list. All that's left is to zero out the old supers and
2189 * We cannot call btrfs_close_bdev() here because we're holding the sb
2190 * write lock, and blkdev_put() will pull in the ->open_mutex on the
2191 * block device and it's dependencies. Instead just flush the device
2192 * and let the caller do the final blkdev_put.
2194 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2195 btrfs_scratch_superblocks(fs_info, device->bdev,
2198 sync_blockdev(device->bdev);
2199 invalidate_bdev(device->bdev);
2203 *bdev = device->bdev;
2204 *mode = device->mode;
2206 btrfs_free_device(device);
2209 * This can happen if cur_devices is the private seed devices list. We
2210 * cannot call close_fs_devices() here because it expects the uuid_mutex
2211 * to be held, but in fact we don't need that for the private
2212 * seed_devices, we can simply decrement cur_devices->opened and then
2213 * remove it from our list and free the fs_devices.
2215 if (cur_devices->num_devices == 0) {
2216 list_del_init(&cur_devices->seed_list);
2217 ASSERT(cur_devices->opened == 1);
2218 cur_devices->opened--;
2219 free_fs_devices(cur_devices);
2222 ret = btrfs_commit_transaction(trans);
2227 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2228 mutex_lock(&fs_info->chunk_mutex);
2229 list_add(&device->dev_alloc_list,
2230 &fs_devices->alloc_list);
2231 device->fs_devices->rw_devices++;
2232 mutex_unlock(&fs_info->chunk_mutex);
2237 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev)
2239 struct btrfs_fs_devices *fs_devices;
2241 lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex);
2244 * in case of fs with no seed, srcdev->fs_devices will point
2245 * to fs_devices of fs_info. However when the dev being replaced is
2246 * a seed dev it will point to the seed's local fs_devices. In short
2247 * srcdev will have its correct fs_devices in both the cases.
2249 fs_devices = srcdev->fs_devices;
2251 list_del_rcu(&srcdev->dev_list);
2252 list_del(&srcdev->dev_alloc_list);
2253 fs_devices->num_devices--;
2254 if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state))
2255 fs_devices->missing_devices--;
2257 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state))
2258 fs_devices->rw_devices--;
2261 fs_devices->open_devices--;
2264 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev)
2266 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2268 mutex_lock(&uuid_mutex);
2270 btrfs_close_bdev(srcdev);
2272 btrfs_free_device(srcdev);
2274 /* if this is no devs we rather delete the fs_devices */
2275 if (!fs_devices->num_devices) {
2277 * On a mounted FS, num_devices can't be zero unless it's a
2278 * seed. In case of a seed device being replaced, the replace
2279 * target added to the sprout FS, so there will be no more
2280 * device left under the seed FS.
2282 ASSERT(fs_devices->seeding);
2284 list_del_init(&fs_devices->seed_list);
2285 close_fs_devices(fs_devices);
2286 free_fs_devices(fs_devices);
2288 mutex_unlock(&uuid_mutex);
2291 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev)
2293 struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices;
2295 mutex_lock(&fs_devices->device_list_mutex);
2297 btrfs_sysfs_remove_device(tgtdev);
2300 fs_devices->open_devices--;
2302 fs_devices->num_devices--;
2304 btrfs_assign_next_active_device(tgtdev, NULL);
2306 list_del_rcu(&tgtdev->dev_list);
2308 mutex_unlock(&fs_devices->device_list_mutex);
2310 btrfs_scratch_superblocks(tgtdev->fs_info, tgtdev->bdev,
2313 btrfs_close_bdev(tgtdev);
2315 btrfs_free_device(tgtdev);
2319 * Populate args from device at path
2321 * @fs_info: the filesystem
2322 * @args: the args to populate
2323 * @path: the path to the device
2325 * This will read the super block of the device at @path and populate @args with
2326 * the devid, fsid, and uuid. This is meant to be used for ioctls that need to
2327 * lookup a device to operate on, but need to do it before we take any locks.
2328 * This properly handles the special case of "missing" that a user may pass in,
2329 * and does some basic sanity checks. The caller must make sure that @path is
2330 * properly NUL terminated before calling in, and must call
2331 * btrfs_put_dev_args_from_path() in order to free up the temporary fsid and
2334 * Return: 0 for success, -errno for failure
2336 int btrfs_get_dev_args_from_path(struct btrfs_fs_info *fs_info,
2337 struct btrfs_dev_lookup_args *args,
2340 struct btrfs_super_block *disk_super;
2341 struct block_device *bdev;
2344 if (!path || !path[0])
2346 if (!strcmp(path, "missing")) {
2347 args->missing = true;
2351 args->uuid = kzalloc(BTRFS_UUID_SIZE, GFP_KERNEL);
2352 args->fsid = kzalloc(BTRFS_FSID_SIZE, GFP_KERNEL);
2353 if (!args->uuid || !args->fsid) {
2354 btrfs_put_dev_args_from_path(args);
2358 ret = btrfs_get_bdev_and_sb(path, FMODE_READ, fs_info->bdev_holder, 0,
2359 &bdev, &disk_super);
2362 args->devid = btrfs_stack_device_id(&disk_super->dev_item);
2363 memcpy(args->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE);
2364 if (btrfs_fs_incompat(fs_info, METADATA_UUID))
2365 memcpy(args->fsid, disk_super->metadata_uuid, BTRFS_FSID_SIZE);
2367 memcpy(args->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
2368 btrfs_release_disk_super(disk_super);
2369 blkdev_put(bdev, FMODE_READ);
2374 * Only use this jointly with btrfs_get_dev_args_from_path() because we will
2375 * allocate our ->uuid and ->fsid pointers, everybody else uses local variables
2376 * that don't need to be freed.
2378 void btrfs_put_dev_args_from_path(struct btrfs_dev_lookup_args *args)
2386 struct btrfs_device *btrfs_find_device_by_devspec(
2387 struct btrfs_fs_info *fs_info, u64 devid,
2388 const char *device_path)
2390 BTRFS_DEV_LOOKUP_ARGS(args);
2391 struct btrfs_device *device;
2396 device = btrfs_find_device(fs_info->fs_devices, &args);
2398 return ERR_PTR(-ENOENT);
2402 ret = btrfs_get_dev_args_from_path(fs_info, &args, device_path);
2404 return ERR_PTR(ret);
2405 device = btrfs_find_device(fs_info->fs_devices, &args);
2406 btrfs_put_dev_args_from_path(&args);
2408 return ERR_PTR(-ENOENT);
2412 static struct btrfs_fs_devices *btrfs_init_sprout(struct btrfs_fs_info *fs_info)
2414 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2415 struct btrfs_fs_devices *old_devices;
2416 struct btrfs_fs_devices *seed_devices;
2418 lockdep_assert_held(&uuid_mutex);
2419 if (!fs_devices->seeding)
2420 return ERR_PTR(-EINVAL);
2423 * Private copy of the seed devices, anchored at
2424 * fs_info->fs_devices->seed_list
2426 seed_devices = alloc_fs_devices(NULL, NULL);
2427 if (IS_ERR(seed_devices))
2428 return seed_devices;
2431 * It's necessary to retain a copy of the original seed fs_devices in
2432 * fs_uuids so that filesystems which have been seeded can successfully
2433 * reference the seed device from open_seed_devices. This also supports
2436 old_devices = clone_fs_devices(fs_devices);
2437 if (IS_ERR(old_devices)) {
2438 kfree(seed_devices);
2442 list_add(&old_devices->fs_list, &fs_uuids);
2444 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2445 seed_devices->opened = 1;
2446 INIT_LIST_HEAD(&seed_devices->devices);
2447 INIT_LIST_HEAD(&seed_devices->alloc_list);
2448 mutex_init(&seed_devices->device_list_mutex);
2450 return seed_devices;
2454 * Splice seed devices into the sprout fs_devices.
2455 * Generate a new fsid for the sprouted read-write filesystem.
2457 static void btrfs_setup_sprout(struct btrfs_fs_info *fs_info,
2458 struct btrfs_fs_devices *seed_devices)
2460 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2461 struct btrfs_super_block *disk_super = fs_info->super_copy;
2462 struct btrfs_device *device;
2466 * We are updating the fsid, the thread leading to device_list_add()
2467 * could race, so uuid_mutex is needed.
2469 lockdep_assert_held(&uuid_mutex);
2472 * The threads listed below may traverse dev_list but can do that without
2473 * device_list_mutex:
2474 * - All device ops and balance - as we are in btrfs_exclop_start.
2475 * - Various dev_list readers - are using RCU.
2476 * - btrfs_ioctl_fitrim() - is using RCU.
2478 * For-read threads as below are using device_list_mutex:
2479 * - Readonly scrub btrfs_scrub_dev()
2480 * - Readonly scrub btrfs_scrub_progress()
2481 * - btrfs_get_dev_stats()
2483 lockdep_assert_held(&fs_devices->device_list_mutex);
2485 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2487 list_for_each_entry(device, &seed_devices->devices, dev_list)
2488 device->fs_devices = seed_devices;
2490 fs_devices->seeding = false;
2491 fs_devices->num_devices = 0;
2492 fs_devices->open_devices = 0;
2493 fs_devices->missing_devices = 0;
2494 fs_devices->rotating = false;
2495 list_add(&seed_devices->seed_list, &fs_devices->seed_list);
2497 generate_random_uuid(fs_devices->fsid);
2498 memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE);
2499 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2501 super_flags = btrfs_super_flags(disk_super) &
2502 ~BTRFS_SUPER_FLAG_SEEDING;
2503 btrfs_set_super_flags(disk_super, super_flags);
2507 * Store the expected generation for seed devices in device items.
2509 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans)
2511 BTRFS_DEV_LOOKUP_ARGS(args);
2512 struct btrfs_fs_info *fs_info = trans->fs_info;
2513 struct btrfs_root *root = fs_info->chunk_root;
2514 struct btrfs_path *path;
2515 struct extent_buffer *leaf;
2516 struct btrfs_dev_item *dev_item;
2517 struct btrfs_device *device;
2518 struct btrfs_key key;
2519 u8 fs_uuid[BTRFS_FSID_SIZE];
2520 u8 dev_uuid[BTRFS_UUID_SIZE];
2523 path = btrfs_alloc_path();
2527 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2529 key.type = BTRFS_DEV_ITEM_KEY;
2532 btrfs_reserve_chunk_metadata(trans, false);
2533 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2534 btrfs_trans_release_chunk_metadata(trans);
2538 leaf = path->nodes[0];
2540 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2541 ret = btrfs_next_leaf(root, path);
2546 leaf = path->nodes[0];
2547 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2548 btrfs_release_path(path);
2552 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2553 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2554 key.type != BTRFS_DEV_ITEM_KEY)
2557 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2558 struct btrfs_dev_item);
2559 args.devid = btrfs_device_id(leaf, dev_item);
2560 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2562 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2564 args.uuid = dev_uuid;
2565 args.fsid = fs_uuid;
2566 device = btrfs_find_device(fs_info->fs_devices, &args);
2567 BUG_ON(!device); /* Logic error */
2569 if (device->fs_devices->seeding) {
2570 btrfs_set_device_generation(leaf, dev_item,
2571 device->generation);
2572 btrfs_mark_buffer_dirty(leaf);
2580 btrfs_free_path(path);
2584 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2586 struct btrfs_root *root = fs_info->dev_root;
2587 struct btrfs_trans_handle *trans;
2588 struct btrfs_device *device;
2589 struct block_device *bdev;
2590 struct super_block *sb = fs_info->sb;
2591 struct rcu_string *name;
2592 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2593 struct btrfs_fs_devices *seed_devices;
2594 u64 orig_super_total_bytes;
2595 u64 orig_super_num_devices;
2597 bool seeding_dev = false;
2598 bool locked = false;
2600 if (sb_rdonly(sb) && !fs_devices->seeding)
2603 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2604 fs_info->bdev_holder);
2606 return PTR_ERR(bdev);
2608 if (!btrfs_check_device_zone_type(fs_info, bdev)) {
2613 if (fs_devices->seeding) {
2615 down_write(&sb->s_umount);
2616 mutex_lock(&uuid_mutex);
2620 sync_blockdev(bdev);
2623 list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
2624 if (device->bdev == bdev) {
2632 device = btrfs_alloc_device(fs_info, NULL, NULL);
2633 if (IS_ERR(device)) {
2634 /* we can safely leave the fs_devices entry around */
2635 ret = PTR_ERR(device);
2639 name = rcu_string_strdup(device_path, GFP_KERNEL);
2642 goto error_free_device;
2644 rcu_assign_pointer(device->name, name);
2646 device->fs_info = fs_info;
2647 device->bdev = bdev;
2648 ret = lookup_bdev(device_path, &device->devt);
2650 goto error_free_device;
2652 ret = btrfs_get_dev_zone_info(device, false);
2654 goto error_free_device;
2656 trans = btrfs_start_transaction(root, 0);
2657 if (IS_ERR(trans)) {
2658 ret = PTR_ERR(trans);
2659 goto error_free_zone;
2662 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
2663 device->generation = trans->transid;
2664 device->io_width = fs_info->sectorsize;
2665 device->io_align = fs_info->sectorsize;
2666 device->sector_size = fs_info->sectorsize;
2667 device->total_bytes =
2668 round_down(bdev_nr_bytes(bdev), fs_info->sectorsize);
2669 device->disk_total_bytes = device->total_bytes;
2670 device->commit_total_bytes = device->total_bytes;
2671 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2672 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
2673 device->mode = FMODE_EXCL;
2674 device->dev_stats_valid = 1;
2675 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2678 btrfs_clear_sb_rdonly(sb);
2680 /* GFP_KERNEL allocation must not be under device_list_mutex */
2681 seed_devices = btrfs_init_sprout(fs_info);
2682 if (IS_ERR(seed_devices)) {
2683 ret = PTR_ERR(seed_devices);
2684 btrfs_abort_transaction(trans, ret);
2689 mutex_lock(&fs_devices->device_list_mutex);
2691 btrfs_setup_sprout(fs_info, seed_devices);
2692 btrfs_assign_next_active_device(fs_info->fs_devices->latest_dev,
2696 device->fs_devices = fs_devices;
2698 mutex_lock(&fs_info->chunk_mutex);
2699 list_add_rcu(&device->dev_list, &fs_devices->devices);
2700 list_add(&device->dev_alloc_list, &fs_devices->alloc_list);
2701 fs_devices->num_devices++;
2702 fs_devices->open_devices++;
2703 fs_devices->rw_devices++;
2704 fs_devices->total_devices++;
2705 fs_devices->total_rw_bytes += device->total_bytes;
2707 atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2709 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2710 fs_devices->rotating = true;
2712 orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
2713 btrfs_set_super_total_bytes(fs_info->super_copy,
2714 round_down(orig_super_total_bytes + device->total_bytes,
2715 fs_info->sectorsize));
2717 orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy);
2718 btrfs_set_super_num_devices(fs_info->super_copy,
2719 orig_super_num_devices + 1);
2722 * we've got more storage, clear any full flags on the space
2725 btrfs_clear_space_info_full(fs_info);
2727 mutex_unlock(&fs_info->chunk_mutex);
2729 /* Add sysfs device entry */
2730 btrfs_sysfs_add_device(device);
2732 mutex_unlock(&fs_devices->device_list_mutex);
2735 mutex_lock(&fs_info->chunk_mutex);
2736 ret = init_first_rw_device(trans);
2737 mutex_unlock(&fs_info->chunk_mutex);
2739 btrfs_abort_transaction(trans, ret);
2744 ret = btrfs_add_dev_item(trans, device);
2746 btrfs_abort_transaction(trans, ret);
2751 ret = btrfs_finish_sprout(trans);
2753 btrfs_abort_transaction(trans, ret);
2758 * fs_devices now represents the newly sprouted filesystem and
2759 * its fsid has been changed by btrfs_sprout_splice().
2761 btrfs_sysfs_update_sprout_fsid(fs_devices);
2764 ret = btrfs_commit_transaction(trans);
2767 mutex_unlock(&uuid_mutex);
2768 up_write(&sb->s_umount);
2771 if (ret) /* transaction commit */
2774 ret = btrfs_relocate_sys_chunks(fs_info);
2776 btrfs_handle_fs_error(fs_info, ret,
2777 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2778 trans = btrfs_attach_transaction(root);
2779 if (IS_ERR(trans)) {
2780 if (PTR_ERR(trans) == -ENOENT)
2782 ret = PTR_ERR(trans);
2786 ret = btrfs_commit_transaction(trans);
2790 * Now that we have written a new super block to this device, check all
2791 * other fs_devices list if device_path alienates any other scanned
2793 * We can ignore the return value as it typically returns -EINVAL and
2794 * only succeeds if the device was an alien.
2796 btrfs_forget_devices(device->devt);
2798 /* Update ctime/mtime for blkid or udev */
2799 update_dev_time(device_path);
2804 btrfs_sysfs_remove_device(device);
2805 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2806 mutex_lock(&fs_info->chunk_mutex);
2807 list_del_rcu(&device->dev_list);
2808 list_del(&device->dev_alloc_list);
2809 fs_info->fs_devices->num_devices--;
2810 fs_info->fs_devices->open_devices--;
2811 fs_info->fs_devices->rw_devices--;
2812 fs_info->fs_devices->total_devices--;
2813 fs_info->fs_devices->total_rw_bytes -= device->total_bytes;
2814 atomic64_sub(device->total_bytes, &fs_info->free_chunk_space);
2815 btrfs_set_super_total_bytes(fs_info->super_copy,
2816 orig_super_total_bytes);
2817 btrfs_set_super_num_devices(fs_info->super_copy,
2818 orig_super_num_devices);
2819 mutex_unlock(&fs_info->chunk_mutex);
2820 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2823 btrfs_set_sb_rdonly(sb);
2825 btrfs_end_transaction(trans);
2827 btrfs_destroy_dev_zone_info(device);
2829 btrfs_free_device(device);
2831 blkdev_put(bdev, FMODE_EXCL);
2833 mutex_unlock(&uuid_mutex);
2834 up_write(&sb->s_umount);
2839 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2840 struct btrfs_device *device)
2843 struct btrfs_path *path;
2844 struct btrfs_root *root = device->fs_info->chunk_root;
2845 struct btrfs_dev_item *dev_item;
2846 struct extent_buffer *leaf;
2847 struct btrfs_key key;
2849 path = btrfs_alloc_path();
2853 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2854 key.type = BTRFS_DEV_ITEM_KEY;
2855 key.offset = device->devid;
2857 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2866 leaf = path->nodes[0];
2867 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2869 btrfs_set_device_id(leaf, dev_item, device->devid);
2870 btrfs_set_device_type(leaf, dev_item, device->type);
2871 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2872 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2873 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2874 btrfs_set_device_total_bytes(leaf, dev_item,
2875 btrfs_device_get_disk_total_bytes(device));
2876 btrfs_set_device_bytes_used(leaf, dev_item,
2877 btrfs_device_get_bytes_used(device));
2878 btrfs_mark_buffer_dirty(leaf);
2881 btrfs_free_path(path);
2885 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2886 struct btrfs_device *device, u64 new_size)
2888 struct btrfs_fs_info *fs_info = device->fs_info;
2889 struct btrfs_super_block *super_copy = fs_info->super_copy;
2894 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2897 new_size = round_down(new_size, fs_info->sectorsize);
2899 mutex_lock(&fs_info->chunk_mutex);
2900 old_total = btrfs_super_total_bytes(super_copy);
2901 diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2903 if (new_size <= device->total_bytes ||
2904 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2905 mutex_unlock(&fs_info->chunk_mutex);
2909 btrfs_set_super_total_bytes(super_copy,
2910 round_down(old_total + diff, fs_info->sectorsize));
2911 device->fs_devices->total_rw_bytes += diff;
2913 btrfs_device_set_total_bytes(device, new_size);
2914 btrfs_device_set_disk_total_bytes(device, new_size);
2915 btrfs_clear_space_info_full(device->fs_info);
2916 if (list_empty(&device->post_commit_list))
2917 list_add_tail(&device->post_commit_list,
2918 &trans->transaction->dev_update_list);
2919 mutex_unlock(&fs_info->chunk_mutex);
2921 btrfs_reserve_chunk_metadata(trans, false);
2922 ret = btrfs_update_device(trans, device);
2923 btrfs_trans_release_chunk_metadata(trans);
2928 static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
2930 struct btrfs_fs_info *fs_info = trans->fs_info;
2931 struct btrfs_root *root = fs_info->chunk_root;
2933 struct btrfs_path *path;
2934 struct btrfs_key key;
2936 path = btrfs_alloc_path();
2940 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2941 key.offset = chunk_offset;
2942 key.type = BTRFS_CHUNK_ITEM_KEY;
2944 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2947 else if (ret > 0) { /* Logic error or corruption */
2948 btrfs_handle_fs_error(fs_info, -ENOENT,
2949 "Failed lookup while freeing chunk.");
2954 ret = btrfs_del_item(trans, root, path);
2956 btrfs_handle_fs_error(fs_info, ret,
2957 "Failed to delete chunk item.");
2959 btrfs_free_path(path);
2963 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2965 struct btrfs_super_block *super_copy = fs_info->super_copy;
2966 struct btrfs_disk_key *disk_key;
2967 struct btrfs_chunk *chunk;
2974 struct btrfs_key key;
2976 lockdep_assert_held(&fs_info->chunk_mutex);
2977 array_size = btrfs_super_sys_array_size(super_copy);
2979 ptr = super_copy->sys_chunk_array;
2982 while (cur < array_size) {
2983 disk_key = (struct btrfs_disk_key *)ptr;
2984 btrfs_disk_key_to_cpu(&key, disk_key);
2986 len = sizeof(*disk_key);
2988 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2989 chunk = (struct btrfs_chunk *)(ptr + len);
2990 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2991 len += btrfs_chunk_item_size(num_stripes);
2996 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID &&
2997 key.offset == chunk_offset) {
2998 memmove(ptr, ptr + len, array_size - (cur + len));
3000 btrfs_set_super_sys_array_size(super_copy, array_size);
3010 * btrfs_get_chunk_map() - Find the mapping containing the given logical extent.
3011 * @logical: Logical block offset in bytes.
3012 * @length: Length of extent in bytes.
3014 * Return: Chunk mapping or ERR_PTR.
3016 struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info,
3017 u64 logical, u64 length)
3019 struct extent_map_tree *em_tree;
3020 struct extent_map *em;
3022 em_tree = &fs_info->mapping_tree;
3023 read_lock(&em_tree->lock);
3024 em = lookup_extent_mapping(em_tree, logical, length);
3025 read_unlock(&em_tree->lock);
3028 btrfs_crit(fs_info, "unable to find logical %llu length %llu",
3030 return ERR_PTR(-EINVAL);
3033 if (em->start > logical || em->start + em->len < logical) {
3035 "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
3036 logical, length, em->start, em->start + em->len);
3037 free_extent_map(em);
3038 return ERR_PTR(-EINVAL);
3041 /* callers are responsible for dropping em's ref. */
3045 static int remove_chunk_item(struct btrfs_trans_handle *trans,
3046 struct map_lookup *map, u64 chunk_offset)
3051 * Removing chunk items and updating the device items in the chunks btree
3052 * requires holding the chunk_mutex.
3053 * See the comment at btrfs_chunk_alloc() for the details.
3055 lockdep_assert_held(&trans->fs_info->chunk_mutex);
3057 for (i = 0; i < map->num_stripes; i++) {
3060 ret = btrfs_update_device(trans, map->stripes[i].dev);
3065 return btrfs_free_chunk(trans, chunk_offset);
3068 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
3070 struct btrfs_fs_info *fs_info = trans->fs_info;
3071 struct extent_map *em;
3072 struct map_lookup *map;
3073 u64 dev_extent_len = 0;
3075 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
3077 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
3080 * This is a logic error, but we don't want to just rely on the
3081 * user having built with ASSERT enabled, so if ASSERT doesn't
3082 * do anything we still error out.
3087 map = em->map_lookup;
3090 * First delete the device extent items from the devices btree.
3091 * We take the device_list_mutex to avoid racing with the finishing phase
3092 * of a device replace operation. See the comment below before acquiring
3093 * fs_info->chunk_mutex. Note that here we do not acquire the chunk_mutex
3094 * because that can result in a deadlock when deleting the device extent
3095 * items from the devices btree - COWing an extent buffer from the btree
3096 * may result in allocating a new metadata chunk, which would attempt to
3097 * lock again fs_info->chunk_mutex.
3099 mutex_lock(&fs_devices->device_list_mutex);
3100 for (i = 0; i < map->num_stripes; i++) {
3101 struct btrfs_device *device = map->stripes[i].dev;
3102 ret = btrfs_free_dev_extent(trans, device,
3103 map->stripes[i].physical,
3106 mutex_unlock(&fs_devices->device_list_mutex);
3107 btrfs_abort_transaction(trans, ret);
3111 if (device->bytes_used > 0) {
3112 mutex_lock(&fs_info->chunk_mutex);
3113 btrfs_device_set_bytes_used(device,
3114 device->bytes_used - dev_extent_len);
3115 atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
3116 btrfs_clear_space_info_full(fs_info);
3117 mutex_unlock(&fs_info->chunk_mutex);
3120 mutex_unlock(&fs_devices->device_list_mutex);
3123 * We acquire fs_info->chunk_mutex for 2 reasons:
3125 * 1) Just like with the first phase of the chunk allocation, we must
3126 * reserve system space, do all chunk btree updates and deletions, and
3127 * update the system chunk array in the superblock while holding this
3128 * mutex. This is for similar reasons as explained on the comment at
3129 * the top of btrfs_chunk_alloc();
3131 * 2) Prevent races with the final phase of a device replace operation
3132 * that replaces the device object associated with the map's stripes,
3133 * because the device object's id can change at any time during that
3134 * final phase of the device replace operation
3135 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the
3136 * replaced device and then see it with an ID of
3137 * BTRFS_DEV_REPLACE_DEVID, which would cause a failure when updating
3138 * the device item, which does not exists on the chunk btree.
3139 * The finishing phase of device replace acquires both the
3140 * device_list_mutex and the chunk_mutex, in that order, so we are
3141 * safe by just acquiring the chunk_mutex.
3143 trans->removing_chunk = true;
3144 mutex_lock(&fs_info->chunk_mutex);
3146 check_system_chunk(trans, map->type);
3148 ret = remove_chunk_item(trans, map, chunk_offset);
3150 * Normally we should not get -ENOSPC since we reserved space before
3151 * through the call to check_system_chunk().
3153 * Despite our system space_info having enough free space, we may not
3154 * be able to allocate extents from its block groups, because all have
3155 * an incompatible profile, which will force us to allocate a new system
3156 * block group with the right profile, or right after we called
3157 * check_system_space() above, a scrub turned the only system block group
3158 * with enough free space into RO mode.
3159 * This is explained with more detail at do_chunk_alloc().
3161 * So if we get -ENOSPC, allocate a new system chunk and retry once.
3163 if (ret == -ENOSPC) {
3164 const u64 sys_flags = btrfs_system_alloc_profile(fs_info);
3165 struct btrfs_block_group *sys_bg;
3167 sys_bg = btrfs_create_chunk(trans, sys_flags);
3168 if (IS_ERR(sys_bg)) {
3169 ret = PTR_ERR(sys_bg);
3170 btrfs_abort_transaction(trans, ret);
3174 ret = btrfs_chunk_alloc_add_chunk_item(trans, sys_bg);
3176 btrfs_abort_transaction(trans, ret);
3180 ret = remove_chunk_item(trans, map, chunk_offset);
3182 btrfs_abort_transaction(trans, ret);
3186 btrfs_abort_transaction(trans, ret);
3190 trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
3192 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
3193 ret = btrfs_del_sys_chunk(fs_info, chunk_offset);
3195 btrfs_abort_transaction(trans, ret);
3200 mutex_unlock(&fs_info->chunk_mutex);
3201 trans->removing_chunk = false;
3204 * We are done with chunk btree updates and deletions, so release the
3205 * system space we previously reserved (with check_system_chunk()).
3207 btrfs_trans_release_chunk_metadata(trans);
3209 ret = btrfs_remove_block_group(trans, chunk_offset, em);
3211 btrfs_abort_transaction(trans, ret);
3216 if (trans->removing_chunk) {
3217 mutex_unlock(&fs_info->chunk_mutex);
3218 trans->removing_chunk = false;
3221 free_extent_map(em);
3225 int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
3227 struct btrfs_root *root = fs_info->chunk_root;
3228 struct btrfs_trans_handle *trans;
3229 struct btrfs_block_group *block_group;
3233 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
3235 "relocate: not supported on extent tree v2 yet");
3240 * Prevent races with automatic removal of unused block groups.
3241 * After we relocate and before we remove the chunk with offset
3242 * chunk_offset, automatic removal of the block group can kick in,
3243 * resulting in a failure when calling btrfs_remove_chunk() below.
3245 * Make sure to acquire this mutex before doing a tree search (dev
3246 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
3247 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
3248 * we release the path used to search the chunk/dev tree and before
3249 * the current task acquires this mutex and calls us.
3251 lockdep_assert_held(&fs_info->reclaim_bgs_lock);
3253 /* step one, relocate all the extents inside this chunk */
3254 btrfs_scrub_pause(fs_info);
3255 ret = btrfs_relocate_block_group(fs_info, chunk_offset);
3256 btrfs_scrub_continue(fs_info);
3260 block_group = btrfs_lookup_block_group(fs_info, chunk_offset);
3263 btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group);
3264 length = block_group->length;
3265 btrfs_put_block_group(block_group);
3268 * On a zoned file system, discard the whole block group, this will
3269 * trigger a REQ_OP_ZONE_RESET operation on the device zone. If
3270 * resetting the zone fails, don't treat it as a fatal problem from the
3271 * filesystem's point of view.
3273 if (btrfs_is_zoned(fs_info)) {
3274 ret = btrfs_discard_extent(fs_info, chunk_offset, length, NULL);
3277 "failed to reset zone %llu after relocation",
3281 trans = btrfs_start_trans_remove_block_group(root->fs_info,
3283 if (IS_ERR(trans)) {
3284 ret = PTR_ERR(trans);
3285 btrfs_handle_fs_error(root->fs_info, ret, NULL);
3290 * step two, delete the device extents and the
3291 * chunk tree entries
3293 ret = btrfs_remove_chunk(trans, chunk_offset);
3294 btrfs_end_transaction(trans);
3298 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
3300 struct btrfs_root *chunk_root = fs_info->chunk_root;
3301 struct btrfs_path *path;
3302 struct extent_buffer *leaf;
3303 struct btrfs_chunk *chunk;
3304 struct btrfs_key key;
3305 struct btrfs_key found_key;
3307 bool retried = false;
3311 path = btrfs_alloc_path();
3316 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3317 key.offset = (u64)-1;
3318 key.type = BTRFS_CHUNK_ITEM_KEY;
3321 mutex_lock(&fs_info->reclaim_bgs_lock);
3322 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3324 mutex_unlock(&fs_info->reclaim_bgs_lock);
3327 BUG_ON(ret == 0); /* Corruption */
3329 ret = btrfs_previous_item(chunk_root, path, key.objectid,
3332 mutex_unlock(&fs_info->reclaim_bgs_lock);
3338 leaf = path->nodes[0];
3339 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3341 chunk = btrfs_item_ptr(leaf, path->slots[0],
3342 struct btrfs_chunk);
3343 chunk_type = btrfs_chunk_type(leaf, chunk);
3344 btrfs_release_path(path);
3346 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3347 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3353 mutex_unlock(&fs_info->reclaim_bgs_lock);
3355 if (found_key.offset == 0)
3357 key.offset = found_key.offset - 1;
3360 if (failed && !retried) {
3364 } else if (WARN_ON(failed && retried)) {
3368 btrfs_free_path(path);
3373 * return 1 : allocate a data chunk successfully,
3374 * return <0: errors during allocating a data chunk,
3375 * return 0 : no need to allocate a data chunk.
3377 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info,
3380 struct btrfs_block_group *cache;
3384 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3386 chunk_type = cache->flags;
3387 btrfs_put_block_group(cache);
3389 if (!(chunk_type & BTRFS_BLOCK_GROUP_DATA))
3392 spin_lock(&fs_info->data_sinfo->lock);
3393 bytes_used = fs_info->data_sinfo->bytes_used;
3394 spin_unlock(&fs_info->data_sinfo->lock);
3397 struct btrfs_trans_handle *trans;
3400 trans = btrfs_join_transaction(fs_info->tree_root);
3402 return PTR_ERR(trans);
3404 ret = btrfs_force_chunk_alloc(trans, BTRFS_BLOCK_GROUP_DATA);
3405 btrfs_end_transaction(trans);
3414 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3415 struct btrfs_balance_control *bctl)
3417 struct btrfs_root *root = fs_info->tree_root;
3418 struct btrfs_trans_handle *trans;
3419 struct btrfs_balance_item *item;
3420 struct btrfs_disk_balance_args disk_bargs;
3421 struct btrfs_path *path;
3422 struct extent_buffer *leaf;
3423 struct btrfs_key key;
3426 path = btrfs_alloc_path();
3430 trans = btrfs_start_transaction(root, 0);
3431 if (IS_ERR(trans)) {
3432 btrfs_free_path(path);
3433 return PTR_ERR(trans);
3436 key.objectid = BTRFS_BALANCE_OBJECTID;
3437 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3440 ret = btrfs_insert_empty_item(trans, root, path, &key,
3445 leaf = path->nodes[0];
3446 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3448 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3450 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3451 btrfs_set_balance_data(leaf, item, &disk_bargs);
3452 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3453 btrfs_set_balance_meta(leaf, item, &disk_bargs);
3454 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3455 btrfs_set_balance_sys(leaf, item, &disk_bargs);
3457 btrfs_set_balance_flags(leaf, item, bctl->flags);
3459 btrfs_mark_buffer_dirty(leaf);
3461 btrfs_free_path(path);
3462 err = btrfs_commit_transaction(trans);
3468 static int del_balance_item(struct btrfs_fs_info *fs_info)
3470 struct btrfs_root *root = fs_info->tree_root;
3471 struct btrfs_trans_handle *trans;
3472 struct btrfs_path *path;
3473 struct btrfs_key key;
3476 path = btrfs_alloc_path();
3480 trans = btrfs_start_transaction_fallback_global_rsv(root, 0);
3481 if (IS_ERR(trans)) {
3482 btrfs_free_path(path);
3483 return PTR_ERR(trans);
3486 key.objectid = BTRFS_BALANCE_OBJECTID;
3487 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3490 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3498 ret = btrfs_del_item(trans, root, path);
3500 btrfs_free_path(path);
3501 err = btrfs_commit_transaction(trans);
3508 * This is a heuristic used to reduce the number of chunks balanced on
3509 * resume after balance was interrupted.
3511 static void update_balance_args(struct btrfs_balance_control *bctl)
3514 * Turn on soft mode for chunk types that were being converted.
3516 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3517 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3518 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3519 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3520 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3521 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3524 * Turn on usage filter if is not already used. The idea is
3525 * that chunks that we have already balanced should be
3526 * reasonably full. Don't do it for chunks that are being
3527 * converted - that will keep us from relocating unconverted
3528 * (albeit full) chunks.
3530 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3531 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3532 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3533 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3534 bctl->data.usage = 90;
3536 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3537 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3538 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3539 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3540 bctl->sys.usage = 90;
3542 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3543 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3544 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3545 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3546 bctl->meta.usage = 90;
3551 * Clear the balance status in fs_info and delete the balance item from disk.
3553 static void reset_balance_state(struct btrfs_fs_info *fs_info)
3555 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3558 BUG_ON(!fs_info->balance_ctl);
3560 spin_lock(&fs_info->balance_lock);
3561 fs_info->balance_ctl = NULL;
3562 spin_unlock(&fs_info->balance_lock);
3565 ret = del_balance_item(fs_info);
3567 btrfs_handle_fs_error(fs_info, ret, NULL);
3571 * Balance filters. Return 1 if chunk should be filtered out
3572 * (should not be balanced).
3574 static int chunk_profiles_filter(u64 chunk_type,
3575 struct btrfs_balance_args *bargs)
3577 chunk_type = chunk_to_extended(chunk_type) &
3578 BTRFS_EXTENDED_PROFILE_MASK;
3580 if (bargs->profiles & chunk_type)
3586 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3587 struct btrfs_balance_args *bargs)
3589 struct btrfs_block_group *cache;
3591 u64 user_thresh_min;
3592 u64 user_thresh_max;
3595 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3596 chunk_used = cache->used;
3598 if (bargs->usage_min == 0)
3599 user_thresh_min = 0;
3601 user_thresh_min = div_factor_fine(cache->length,
3604 if (bargs->usage_max == 0)
3605 user_thresh_max = 1;
3606 else if (bargs->usage_max > 100)
3607 user_thresh_max = cache->length;
3609 user_thresh_max = div_factor_fine(cache->length,
3612 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3615 btrfs_put_block_group(cache);
3619 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3620 u64 chunk_offset, struct btrfs_balance_args *bargs)
3622 struct btrfs_block_group *cache;
3623 u64 chunk_used, user_thresh;
3626 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3627 chunk_used = cache->used;
3629 if (bargs->usage_min == 0)
3631 else if (bargs->usage > 100)
3632 user_thresh = cache->length;
3634 user_thresh = div_factor_fine(cache->length, bargs->usage);
3636 if (chunk_used < user_thresh)
3639 btrfs_put_block_group(cache);
3643 static int chunk_devid_filter(struct extent_buffer *leaf,
3644 struct btrfs_chunk *chunk,
3645 struct btrfs_balance_args *bargs)
3647 struct btrfs_stripe *stripe;
3648 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3651 for (i = 0; i < num_stripes; i++) {
3652 stripe = btrfs_stripe_nr(chunk, i);
3653 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3660 static u64 calc_data_stripes(u64 type, int num_stripes)
3662 const int index = btrfs_bg_flags_to_raid_index(type);
3663 const int ncopies = btrfs_raid_array[index].ncopies;
3664 const int nparity = btrfs_raid_array[index].nparity;
3666 return (num_stripes - nparity) / ncopies;
3669 /* [pstart, pend) */
3670 static int chunk_drange_filter(struct extent_buffer *leaf,
3671 struct btrfs_chunk *chunk,
3672 struct btrfs_balance_args *bargs)
3674 struct btrfs_stripe *stripe;
3675 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3682 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3685 type = btrfs_chunk_type(leaf, chunk);
3686 factor = calc_data_stripes(type, num_stripes);
3688 for (i = 0; i < num_stripes; i++) {
3689 stripe = btrfs_stripe_nr(chunk, i);
3690 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3693 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3694 stripe_length = btrfs_chunk_length(leaf, chunk);
3695 stripe_length = div_u64(stripe_length, factor);
3697 if (stripe_offset < bargs->pend &&
3698 stripe_offset + stripe_length > bargs->pstart)
3705 /* [vstart, vend) */
3706 static int chunk_vrange_filter(struct extent_buffer *leaf,
3707 struct btrfs_chunk *chunk,
3709 struct btrfs_balance_args *bargs)
3711 if (chunk_offset < bargs->vend &&
3712 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3713 /* at least part of the chunk is inside this vrange */
3719 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3720 struct btrfs_chunk *chunk,
3721 struct btrfs_balance_args *bargs)
3723 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3725 if (bargs->stripes_min <= num_stripes
3726 && num_stripes <= bargs->stripes_max)
3732 static int chunk_soft_convert_filter(u64 chunk_type,
3733 struct btrfs_balance_args *bargs)
3735 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3738 chunk_type = chunk_to_extended(chunk_type) &
3739 BTRFS_EXTENDED_PROFILE_MASK;
3741 if (bargs->target == chunk_type)
3747 static int should_balance_chunk(struct extent_buffer *leaf,
3748 struct btrfs_chunk *chunk, u64 chunk_offset)
3750 struct btrfs_fs_info *fs_info = leaf->fs_info;
3751 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3752 struct btrfs_balance_args *bargs = NULL;
3753 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3756 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3757 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3761 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3762 bargs = &bctl->data;
3763 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3765 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3766 bargs = &bctl->meta;
3768 /* profiles filter */
3769 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3770 chunk_profiles_filter(chunk_type, bargs)) {
3775 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3776 chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3778 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3779 chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3784 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3785 chunk_devid_filter(leaf, chunk, bargs)) {
3789 /* drange filter, makes sense only with devid filter */
3790 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3791 chunk_drange_filter(leaf, chunk, bargs)) {
3796 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3797 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3801 /* stripes filter */
3802 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3803 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3807 /* soft profile changing mode */
3808 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3809 chunk_soft_convert_filter(chunk_type, bargs)) {
3814 * limited by count, must be the last filter
3816 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3817 if (bargs->limit == 0)
3821 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3823 * Same logic as the 'limit' filter; the minimum cannot be
3824 * determined here because we do not have the global information
3825 * about the count of all chunks that satisfy the filters.
3827 if (bargs->limit_max == 0)
3836 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3838 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3839 struct btrfs_root *chunk_root = fs_info->chunk_root;
3841 struct btrfs_chunk *chunk;
3842 struct btrfs_path *path = NULL;
3843 struct btrfs_key key;
3844 struct btrfs_key found_key;
3845 struct extent_buffer *leaf;
3848 int enospc_errors = 0;
3849 bool counting = true;
3850 /* The single value limit and min/max limits use the same bytes in the */
3851 u64 limit_data = bctl->data.limit;
3852 u64 limit_meta = bctl->meta.limit;
3853 u64 limit_sys = bctl->sys.limit;
3857 int chunk_reserved = 0;
3859 path = btrfs_alloc_path();
3865 /* zero out stat counters */
3866 spin_lock(&fs_info->balance_lock);
3867 memset(&bctl->stat, 0, sizeof(bctl->stat));
3868 spin_unlock(&fs_info->balance_lock);
3872 * The single value limit and min/max limits use the same bytes
3875 bctl->data.limit = limit_data;
3876 bctl->meta.limit = limit_meta;
3877 bctl->sys.limit = limit_sys;
3879 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3880 key.offset = (u64)-1;
3881 key.type = BTRFS_CHUNK_ITEM_KEY;
3884 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3885 atomic_read(&fs_info->balance_cancel_req)) {
3890 mutex_lock(&fs_info->reclaim_bgs_lock);
3891 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3893 mutex_unlock(&fs_info->reclaim_bgs_lock);
3898 * this shouldn't happen, it means the last relocate
3902 BUG(); /* FIXME break ? */
3904 ret = btrfs_previous_item(chunk_root, path, 0,
3905 BTRFS_CHUNK_ITEM_KEY);
3907 mutex_unlock(&fs_info->reclaim_bgs_lock);
3912 leaf = path->nodes[0];
3913 slot = path->slots[0];
3914 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3916 if (found_key.objectid != key.objectid) {
3917 mutex_unlock(&fs_info->reclaim_bgs_lock);
3921 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3922 chunk_type = btrfs_chunk_type(leaf, chunk);
3925 spin_lock(&fs_info->balance_lock);
3926 bctl->stat.considered++;
3927 spin_unlock(&fs_info->balance_lock);
3930 ret = should_balance_chunk(leaf, chunk, found_key.offset);
3932 btrfs_release_path(path);
3934 mutex_unlock(&fs_info->reclaim_bgs_lock);
3939 mutex_unlock(&fs_info->reclaim_bgs_lock);
3940 spin_lock(&fs_info->balance_lock);
3941 bctl->stat.expected++;
3942 spin_unlock(&fs_info->balance_lock);
3944 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3946 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3948 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3955 * Apply limit_min filter, no need to check if the LIMITS
3956 * filter is used, limit_min is 0 by default
3958 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3959 count_data < bctl->data.limit_min)
3960 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3961 count_meta < bctl->meta.limit_min)
3962 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3963 count_sys < bctl->sys.limit_min)) {
3964 mutex_unlock(&fs_info->reclaim_bgs_lock);
3968 if (!chunk_reserved) {
3970 * We may be relocating the only data chunk we have,
3971 * which could potentially end up with losing data's
3972 * raid profile, so lets allocate an empty one in
3975 ret = btrfs_may_alloc_data_chunk(fs_info,
3978 mutex_unlock(&fs_info->reclaim_bgs_lock);
3980 } else if (ret == 1) {
3985 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3986 mutex_unlock(&fs_info->reclaim_bgs_lock);
3987 if (ret == -ENOSPC) {
3989 } else if (ret == -ETXTBSY) {
3991 "skipping relocation of block group %llu due to active swapfile",
3997 spin_lock(&fs_info->balance_lock);
3998 bctl->stat.completed++;
3999 spin_unlock(&fs_info->balance_lock);
4002 if (found_key.offset == 0)
4004 key.offset = found_key.offset - 1;
4008 btrfs_release_path(path);
4013 btrfs_free_path(path);
4014 if (enospc_errors) {
4015 btrfs_info(fs_info, "%d enospc errors during balance",
4025 * alloc_profile_is_valid - see if a given profile is valid and reduced
4026 * @flags: profile to validate
4027 * @extended: if true @flags is treated as an extended profile
4029 static int alloc_profile_is_valid(u64 flags, int extended)
4031 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
4032 BTRFS_BLOCK_GROUP_PROFILE_MASK);
4034 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
4036 /* 1) check that all other bits are zeroed */
4040 /* 2) see if profile is reduced */
4042 return !extended; /* "0" is valid for usual profiles */
4044 return has_single_bit_set(flags);
4047 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
4049 /* cancel requested || normal exit path */
4050 return atomic_read(&fs_info->balance_cancel_req) ||
4051 (atomic_read(&fs_info->balance_pause_req) == 0 &&
4052 atomic_read(&fs_info->balance_cancel_req) == 0);
4056 * Validate target profile against allowed profiles and return true if it's OK.
4057 * Otherwise print the error message and return false.
4059 static inline int validate_convert_profile(struct btrfs_fs_info *fs_info,
4060 const struct btrfs_balance_args *bargs,
4061 u64 allowed, const char *type)
4063 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
4066 if (fs_info->sectorsize < PAGE_SIZE &&
4067 bargs->target & BTRFS_BLOCK_GROUP_RAID56_MASK) {
4069 "RAID56 is not yet supported for sectorsize %u with page size %lu",
4070 fs_info->sectorsize, PAGE_SIZE);
4073 /* Profile is valid and does not have bits outside of the allowed set */
4074 if (alloc_profile_is_valid(bargs->target, 1) &&
4075 (bargs->target & ~allowed) == 0)
4078 btrfs_err(fs_info, "balance: invalid convert %s profile %s",
4079 type, btrfs_bg_type_to_raid_name(bargs->target));
4084 * Fill @buf with textual description of balance filter flags @bargs, up to
4085 * @size_buf including the terminating null. The output may be trimmed if it
4086 * does not fit into the provided buffer.
4088 static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf,
4092 u32 size_bp = size_buf;
4094 u64 flags = bargs->flags;
4095 char tmp_buf[128] = {'\0'};
4100 #define CHECK_APPEND_NOARG(a) \
4102 ret = snprintf(bp, size_bp, (a)); \
4103 if (ret < 0 || ret >= size_bp) \
4104 goto out_overflow; \
4109 #define CHECK_APPEND_1ARG(a, v1) \
4111 ret = snprintf(bp, size_bp, (a), (v1)); \
4112 if (ret < 0 || ret >= size_bp) \
4113 goto out_overflow; \
4118 #define CHECK_APPEND_2ARG(a, v1, v2) \
4120 ret = snprintf(bp, size_bp, (a), (v1), (v2)); \
4121 if (ret < 0 || ret >= size_bp) \
4122 goto out_overflow; \
4127 if (flags & BTRFS_BALANCE_ARGS_CONVERT)
4128 CHECK_APPEND_1ARG("convert=%s,",
4129 btrfs_bg_type_to_raid_name(bargs->target));
4131 if (flags & BTRFS_BALANCE_ARGS_SOFT)
4132 CHECK_APPEND_NOARG("soft,");
4134 if (flags & BTRFS_BALANCE_ARGS_PROFILES) {
4135 btrfs_describe_block_groups(bargs->profiles, tmp_buf,
4137 CHECK_APPEND_1ARG("profiles=%s,", tmp_buf);
4140 if (flags & BTRFS_BALANCE_ARGS_USAGE)
4141 CHECK_APPEND_1ARG("usage=%llu,", bargs->usage);
4143 if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE)
4144 CHECK_APPEND_2ARG("usage=%u..%u,",
4145 bargs->usage_min, bargs->usage_max);
4147 if (flags & BTRFS_BALANCE_ARGS_DEVID)
4148 CHECK_APPEND_1ARG("devid=%llu,", bargs->devid);
4150 if (flags & BTRFS_BALANCE_ARGS_DRANGE)
4151 CHECK_APPEND_2ARG("drange=%llu..%llu,",
4152 bargs->pstart, bargs->pend);
4154 if (flags & BTRFS_BALANCE_ARGS_VRANGE)
4155 CHECK_APPEND_2ARG("vrange=%llu..%llu,",
4156 bargs->vstart, bargs->vend);
4158 if (flags & BTRFS_BALANCE_ARGS_LIMIT)
4159 CHECK_APPEND_1ARG("limit=%llu,", bargs->limit);
4161 if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)
4162 CHECK_APPEND_2ARG("limit=%u..%u,",
4163 bargs->limit_min, bargs->limit_max);
4165 if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE)
4166 CHECK_APPEND_2ARG("stripes=%u..%u,",
4167 bargs->stripes_min, bargs->stripes_max);
4169 #undef CHECK_APPEND_2ARG
4170 #undef CHECK_APPEND_1ARG
4171 #undef CHECK_APPEND_NOARG
4175 if (size_bp < size_buf)
4176 buf[size_buf - size_bp - 1] = '\0'; /* remove last , */
4181 static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info)
4183 u32 size_buf = 1024;
4184 char tmp_buf[192] = {'\0'};
4187 u32 size_bp = size_buf;
4189 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4191 buf = kzalloc(size_buf, GFP_KERNEL);
4197 #define CHECK_APPEND_1ARG(a, v1) \
4199 ret = snprintf(bp, size_bp, (a), (v1)); \
4200 if (ret < 0 || ret >= size_bp) \
4201 goto out_overflow; \
4206 if (bctl->flags & BTRFS_BALANCE_FORCE)
4207 CHECK_APPEND_1ARG("%s", "-f ");
4209 if (bctl->flags & BTRFS_BALANCE_DATA) {
4210 describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf));
4211 CHECK_APPEND_1ARG("-d%s ", tmp_buf);
4214 if (bctl->flags & BTRFS_BALANCE_METADATA) {
4215 describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf));
4216 CHECK_APPEND_1ARG("-m%s ", tmp_buf);
4219 if (bctl->flags & BTRFS_BALANCE_SYSTEM) {
4220 describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf));
4221 CHECK_APPEND_1ARG("-s%s ", tmp_buf);
4224 #undef CHECK_APPEND_1ARG
4228 if (size_bp < size_buf)
4229 buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */
4230 btrfs_info(fs_info, "balance: %s %s",
4231 (bctl->flags & BTRFS_BALANCE_RESUME) ?
4232 "resume" : "start", buf);
4238 * Should be called with balance mutexe held
4240 int btrfs_balance(struct btrfs_fs_info *fs_info,
4241 struct btrfs_balance_control *bctl,
4242 struct btrfs_ioctl_balance_args *bargs)
4244 u64 meta_target, data_target;
4250 bool reducing_redundancy;
4253 if (btrfs_fs_closing(fs_info) ||
4254 atomic_read(&fs_info->balance_pause_req) ||
4255 btrfs_should_cancel_balance(fs_info)) {
4260 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
4261 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
4265 * In case of mixed groups both data and meta should be picked,
4266 * and identical options should be given for both of them.
4268 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
4269 if (mixed && (bctl->flags & allowed)) {
4270 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
4271 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
4272 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
4274 "balance: mixed groups data and metadata options must be the same");
4281 * rw_devices will not change at the moment, device add/delete/replace
4284 num_devices = fs_info->fs_devices->rw_devices;
4287 * SINGLE profile on-disk has no profile bit, but in-memory we have a
4288 * special bit for it, to make it easier to distinguish. Thus we need
4289 * to set it manually, or balance would refuse the profile.
4291 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
4292 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++)
4293 if (num_devices >= btrfs_raid_array[i].devs_min)
4294 allowed |= btrfs_raid_array[i].bg_flag;
4296 if (!validate_convert_profile(fs_info, &bctl->data, allowed, "data") ||
4297 !validate_convert_profile(fs_info, &bctl->meta, allowed, "metadata") ||
4298 !validate_convert_profile(fs_info, &bctl->sys, allowed, "system")) {
4304 * Allow to reduce metadata or system integrity only if force set for
4305 * profiles with redundancy (copies, parity)
4308 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) {
4309 if (btrfs_raid_array[i].ncopies >= 2 ||
4310 btrfs_raid_array[i].tolerated_failures >= 1)
4311 allowed |= btrfs_raid_array[i].bg_flag;
4314 seq = read_seqbegin(&fs_info->profiles_lock);
4316 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4317 (fs_info->avail_system_alloc_bits & allowed) &&
4318 !(bctl->sys.target & allowed)) ||
4319 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4320 (fs_info->avail_metadata_alloc_bits & allowed) &&
4321 !(bctl->meta.target & allowed)))
4322 reducing_redundancy = true;
4324 reducing_redundancy = false;
4326 /* if we're not converting, the target field is uninitialized */
4327 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4328 bctl->meta.target : fs_info->avail_metadata_alloc_bits;
4329 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4330 bctl->data.target : fs_info->avail_data_alloc_bits;
4331 } while (read_seqretry(&fs_info->profiles_lock, seq));
4333 if (reducing_redundancy) {
4334 if (bctl->flags & BTRFS_BALANCE_FORCE) {
4336 "balance: force reducing metadata redundancy");
4339 "balance: reduces metadata redundancy, use --force if you want this");
4345 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <
4346 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {
4348 "balance: metadata profile %s has lower redundancy than data profile %s",
4349 btrfs_bg_type_to_raid_name(meta_target),
4350 btrfs_bg_type_to_raid_name(data_target));
4353 ret = insert_balance_item(fs_info, bctl);
4354 if (ret && ret != -EEXIST)
4357 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
4358 BUG_ON(ret == -EEXIST);
4359 BUG_ON(fs_info->balance_ctl);
4360 spin_lock(&fs_info->balance_lock);
4361 fs_info->balance_ctl = bctl;
4362 spin_unlock(&fs_info->balance_lock);
4364 BUG_ON(ret != -EEXIST);
4365 spin_lock(&fs_info->balance_lock);
4366 update_balance_args(bctl);
4367 spin_unlock(&fs_info->balance_lock);
4370 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4371 set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4372 describe_balance_start_or_resume(fs_info);
4373 mutex_unlock(&fs_info->balance_mutex);
4375 ret = __btrfs_balance(fs_info);
4377 mutex_lock(&fs_info->balance_mutex);
4378 if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req)) {
4379 btrfs_info(fs_info, "balance: paused");
4380 btrfs_exclop_balance(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED);
4383 * Balance can be canceled by:
4385 * - Regular cancel request
4386 * Then ret == -ECANCELED and balance_cancel_req > 0
4388 * - Fatal signal to "btrfs" process
4389 * Either the signal caught by wait_reserve_ticket() and callers
4390 * got -EINTR, or caught by btrfs_should_cancel_balance() and
4392 * Either way, in this case balance_cancel_req = 0, and
4393 * ret == -EINTR or ret == -ECANCELED.
4395 * So here we only check the return value to catch canceled balance.
4397 else if (ret == -ECANCELED || ret == -EINTR)
4398 btrfs_info(fs_info, "balance: canceled");
4400 btrfs_info(fs_info, "balance: ended with status: %d", ret);
4402 clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4405 memset(bargs, 0, sizeof(*bargs));
4406 btrfs_update_ioctl_balance_args(fs_info, bargs);
4409 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
4410 balance_need_close(fs_info)) {
4411 reset_balance_state(fs_info);
4412 btrfs_exclop_finish(fs_info);
4415 wake_up(&fs_info->balance_wait_q);
4419 if (bctl->flags & BTRFS_BALANCE_RESUME)
4420 reset_balance_state(fs_info);
4423 btrfs_exclop_finish(fs_info);
4428 static int balance_kthread(void *data)
4430 struct btrfs_fs_info *fs_info = data;
4433 sb_start_write(fs_info->sb);
4434 mutex_lock(&fs_info->balance_mutex);
4435 if (fs_info->balance_ctl)
4436 ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL);
4437 mutex_unlock(&fs_info->balance_mutex);
4438 sb_end_write(fs_info->sb);
4443 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
4445 struct task_struct *tsk;
4447 mutex_lock(&fs_info->balance_mutex);
4448 if (!fs_info->balance_ctl) {
4449 mutex_unlock(&fs_info->balance_mutex);
4452 mutex_unlock(&fs_info->balance_mutex);
4454 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
4455 btrfs_info(fs_info, "balance: resume skipped");
4459 spin_lock(&fs_info->super_lock);
4460 ASSERT(fs_info->exclusive_operation == BTRFS_EXCLOP_BALANCE_PAUSED);
4461 fs_info->exclusive_operation = BTRFS_EXCLOP_BALANCE;
4462 spin_unlock(&fs_info->super_lock);
4464 * A ro->rw remount sequence should continue with the paused balance
4465 * regardless of who pauses it, system or the user as of now, so set
4468 spin_lock(&fs_info->balance_lock);
4469 fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME;
4470 spin_unlock(&fs_info->balance_lock);
4472 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
4473 return PTR_ERR_OR_ZERO(tsk);
4476 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
4478 struct btrfs_balance_control *bctl;
4479 struct btrfs_balance_item *item;
4480 struct btrfs_disk_balance_args disk_bargs;
4481 struct btrfs_path *path;
4482 struct extent_buffer *leaf;
4483 struct btrfs_key key;
4486 path = btrfs_alloc_path();
4490 key.objectid = BTRFS_BALANCE_OBJECTID;
4491 key.type = BTRFS_TEMPORARY_ITEM_KEY;
4494 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4497 if (ret > 0) { /* ret = -ENOENT; */
4502 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4508 leaf = path->nodes[0];
4509 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
4511 bctl->flags = btrfs_balance_flags(leaf, item);
4512 bctl->flags |= BTRFS_BALANCE_RESUME;
4514 btrfs_balance_data(leaf, item, &disk_bargs);
4515 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4516 btrfs_balance_meta(leaf, item, &disk_bargs);
4517 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4518 btrfs_balance_sys(leaf, item, &disk_bargs);
4519 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4522 * This should never happen, as the paused balance state is recovered
4523 * during mount without any chance of other exclusive ops to collide.
4525 * This gives the exclusive op status to balance and keeps in paused
4526 * state until user intervention (cancel or umount). If the ownership
4527 * cannot be assigned, show a message but do not fail. The balance
4528 * is in a paused state and must have fs_info::balance_ctl properly
4531 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED))
4533 "balance: cannot set exclusive op status, resume manually");
4535 btrfs_release_path(path);
4537 mutex_lock(&fs_info->balance_mutex);
4538 BUG_ON(fs_info->balance_ctl);
4539 spin_lock(&fs_info->balance_lock);
4540 fs_info->balance_ctl = bctl;
4541 spin_unlock(&fs_info->balance_lock);
4542 mutex_unlock(&fs_info->balance_mutex);
4544 btrfs_free_path(path);
4548 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4552 mutex_lock(&fs_info->balance_mutex);
4553 if (!fs_info->balance_ctl) {
4554 mutex_unlock(&fs_info->balance_mutex);
4558 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4559 atomic_inc(&fs_info->balance_pause_req);
4560 mutex_unlock(&fs_info->balance_mutex);
4562 wait_event(fs_info->balance_wait_q,
4563 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4565 mutex_lock(&fs_info->balance_mutex);
4566 /* we are good with balance_ctl ripped off from under us */
4567 BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4568 atomic_dec(&fs_info->balance_pause_req);
4573 mutex_unlock(&fs_info->balance_mutex);
4577 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4579 mutex_lock(&fs_info->balance_mutex);
4580 if (!fs_info->balance_ctl) {
4581 mutex_unlock(&fs_info->balance_mutex);
4586 * A paused balance with the item stored on disk can be resumed at
4587 * mount time if the mount is read-write. Otherwise it's still paused
4588 * and we must not allow cancelling as it deletes the item.
4590 if (sb_rdonly(fs_info->sb)) {
4591 mutex_unlock(&fs_info->balance_mutex);
4595 atomic_inc(&fs_info->balance_cancel_req);
4597 * if we are running just wait and return, balance item is
4598 * deleted in btrfs_balance in this case
4600 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4601 mutex_unlock(&fs_info->balance_mutex);
4602 wait_event(fs_info->balance_wait_q,
4603 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4604 mutex_lock(&fs_info->balance_mutex);
4606 mutex_unlock(&fs_info->balance_mutex);
4608 * Lock released to allow other waiters to continue, we'll
4609 * reexamine the status again.
4611 mutex_lock(&fs_info->balance_mutex);
4613 if (fs_info->balance_ctl) {
4614 reset_balance_state(fs_info);
4615 btrfs_exclop_finish(fs_info);
4616 btrfs_info(fs_info, "balance: canceled");
4620 BUG_ON(fs_info->balance_ctl ||
4621 test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4622 atomic_dec(&fs_info->balance_cancel_req);
4623 mutex_unlock(&fs_info->balance_mutex);
4627 int btrfs_uuid_scan_kthread(void *data)
4629 struct btrfs_fs_info *fs_info = data;
4630 struct btrfs_root *root = fs_info->tree_root;
4631 struct btrfs_key key;
4632 struct btrfs_path *path = NULL;
4634 struct extent_buffer *eb;
4636 struct btrfs_root_item root_item;
4638 struct btrfs_trans_handle *trans = NULL;
4639 bool closing = false;
4641 path = btrfs_alloc_path();
4648 key.type = BTRFS_ROOT_ITEM_KEY;
4652 if (btrfs_fs_closing(fs_info)) {
4656 ret = btrfs_search_forward(root, &key, path,
4657 BTRFS_OLDEST_GENERATION);
4664 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4665 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4666 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4667 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4670 eb = path->nodes[0];
4671 slot = path->slots[0];
4672 item_size = btrfs_item_size(eb, slot);
4673 if (item_size < sizeof(root_item))
4676 read_extent_buffer(eb, &root_item,
4677 btrfs_item_ptr_offset(eb, slot),
4678 (int)sizeof(root_item));
4679 if (btrfs_root_refs(&root_item) == 0)
4682 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4683 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4687 btrfs_release_path(path);
4689 * 1 - subvol uuid item
4690 * 1 - received_subvol uuid item
4692 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4693 if (IS_ERR(trans)) {
4694 ret = PTR_ERR(trans);
4702 btrfs_release_path(path);
4703 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4704 ret = btrfs_uuid_tree_add(trans, root_item.uuid,
4705 BTRFS_UUID_KEY_SUBVOL,
4708 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4714 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4715 ret = btrfs_uuid_tree_add(trans,
4716 root_item.received_uuid,
4717 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4720 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4727 btrfs_release_path(path);
4729 ret = btrfs_end_transaction(trans);
4735 if (key.offset < (u64)-1) {
4737 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4739 key.type = BTRFS_ROOT_ITEM_KEY;
4740 } else if (key.objectid < (u64)-1) {
4742 key.type = BTRFS_ROOT_ITEM_KEY;
4751 btrfs_free_path(path);
4752 if (trans && !IS_ERR(trans))
4753 btrfs_end_transaction(trans);
4755 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4757 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4758 up(&fs_info->uuid_tree_rescan_sem);
4762 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4764 struct btrfs_trans_handle *trans;
4765 struct btrfs_root *tree_root = fs_info->tree_root;
4766 struct btrfs_root *uuid_root;
4767 struct task_struct *task;
4774 trans = btrfs_start_transaction(tree_root, 2);
4776 return PTR_ERR(trans);
4778 uuid_root = btrfs_create_tree(trans, BTRFS_UUID_TREE_OBJECTID);
4779 if (IS_ERR(uuid_root)) {
4780 ret = PTR_ERR(uuid_root);
4781 btrfs_abort_transaction(trans, ret);
4782 btrfs_end_transaction(trans);
4786 fs_info->uuid_root = uuid_root;
4788 ret = btrfs_commit_transaction(trans);
4792 down(&fs_info->uuid_tree_rescan_sem);
4793 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4795 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4796 btrfs_warn(fs_info, "failed to start uuid_scan task");
4797 up(&fs_info->uuid_tree_rescan_sem);
4798 return PTR_ERR(task);
4805 * shrinking a device means finding all of the device extents past
4806 * the new size, and then following the back refs to the chunks.
4807 * The chunk relocation code actually frees the device extent
4809 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4811 struct btrfs_fs_info *fs_info = device->fs_info;
4812 struct btrfs_root *root = fs_info->dev_root;
4813 struct btrfs_trans_handle *trans;
4814 struct btrfs_dev_extent *dev_extent = NULL;
4815 struct btrfs_path *path;
4821 bool retried = false;
4822 struct extent_buffer *l;
4823 struct btrfs_key key;
4824 struct btrfs_super_block *super_copy = fs_info->super_copy;
4825 u64 old_total = btrfs_super_total_bytes(super_copy);
4826 u64 old_size = btrfs_device_get_total_bytes(device);
4830 new_size = round_down(new_size, fs_info->sectorsize);
4832 diff = round_down(old_size - new_size, fs_info->sectorsize);
4834 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4837 path = btrfs_alloc_path();
4841 path->reada = READA_BACK;
4843 trans = btrfs_start_transaction(root, 0);
4844 if (IS_ERR(trans)) {
4845 btrfs_free_path(path);
4846 return PTR_ERR(trans);
4849 mutex_lock(&fs_info->chunk_mutex);
4851 btrfs_device_set_total_bytes(device, new_size);
4852 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4853 device->fs_devices->total_rw_bytes -= diff;
4854 atomic64_sub(diff, &fs_info->free_chunk_space);
4858 * Once the device's size has been set to the new size, ensure all
4859 * in-memory chunks are synced to disk so that the loop below sees them
4860 * and relocates them accordingly.
4862 if (contains_pending_extent(device, &start, diff)) {
4863 mutex_unlock(&fs_info->chunk_mutex);
4864 ret = btrfs_commit_transaction(trans);
4868 mutex_unlock(&fs_info->chunk_mutex);
4869 btrfs_end_transaction(trans);
4873 key.objectid = device->devid;
4874 key.offset = (u64)-1;
4875 key.type = BTRFS_DEV_EXTENT_KEY;
4878 mutex_lock(&fs_info->reclaim_bgs_lock);
4879 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4881 mutex_unlock(&fs_info->reclaim_bgs_lock);
4885 ret = btrfs_previous_item(root, path, 0, key.type);
4887 mutex_unlock(&fs_info->reclaim_bgs_lock);
4891 btrfs_release_path(path);
4896 slot = path->slots[0];
4897 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4899 if (key.objectid != device->devid) {
4900 mutex_unlock(&fs_info->reclaim_bgs_lock);
4901 btrfs_release_path(path);
4905 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4906 length = btrfs_dev_extent_length(l, dev_extent);
4908 if (key.offset + length <= new_size) {
4909 mutex_unlock(&fs_info->reclaim_bgs_lock);
4910 btrfs_release_path(path);
4914 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4915 btrfs_release_path(path);
4918 * We may be relocating the only data chunk we have,
4919 * which could potentially end up with losing data's
4920 * raid profile, so lets allocate an empty one in
4923 ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset);
4925 mutex_unlock(&fs_info->reclaim_bgs_lock);
4929 ret = btrfs_relocate_chunk(fs_info, chunk_offset);
4930 mutex_unlock(&fs_info->reclaim_bgs_lock);
4931 if (ret == -ENOSPC) {
4934 if (ret == -ETXTBSY) {
4936 "could not shrink block group %llu due to active swapfile",
4941 } while (key.offset-- > 0);
4943 if (failed && !retried) {
4947 } else if (failed && retried) {
4952 /* Shrinking succeeded, else we would be at "done". */
4953 trans = btrfs_start_transaction(root, 0);
4954 if (IS_ERR(trans)) {
4955 ret = PTR_ERR(trans);
4959 mutex_lock(&fs_info->chunk_mutex);
4960 /* Clear all state bits beyond the shrunk device size */
4961 clear_extent_bits(&device->alloc_state, new_size, (u64)-1,
4964 btrfs_device_set_disk_total_bytes(device, new_size);
4965 if (list_empty(&device->post_commit_list))
4966 list_add_tail(&device->post_commit_list,
4967 &trans->transaction->dev_update_list);
4969 WARN_ON(diff > old_total);
4970 btrfs_set_super_total_bytes(super_copy,
4971 round_down(old_total - diff, fs_info->sectorsize));
4972 mutex_unlock(&fs_info->chunk_mutex);
4974 btrfs_reserve_chunk_metadata(trans, false);
4975 /* Now btrfs_update_device() will change the on-disk size. */
4976 ret = btrfs_update_device(trans, device);
4977 btrfs_trans_release_chunk_metadata(trans);
4979 btrfs_abort_transaction(trans, ret);
4980 btrfs_end_transaction(trans);
4982 ret = btrfs_commit_transaction(trans);
4985 btrfs_free_path(path);
4987 mutex_lock(&fs_info->chunk_mutex);
4988 btrfs_device_set_total_bytes(device, old_size);
4989 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
4990 device->fs_devices->total_rw_bytes += diff;
4991 atomic64_add(diff, &fs_info->free_chunk_space);
4992 mutex_unlock(&fs_info->chunk_mutex);
4997 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
4998 struct btrfs_key *key,
4999 struct btrfs_chunk *chunk, int item_size)
5001 struct btrfs_super_block *super_copy = fs_info->super_copy;
5002 struct btrfs_disk_key disk_key;
5006 lockdep_assert_held(&fs_info->chunk_mutex);
5008 array_size = btrfs_super_sys_array_size(super_copy);
5009 if (array_size + item_size + sizeof(disk_key)
5010 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
5013 ptr = super_copy->sys_chunk_array + array_size;
5014 btrfs_cpu_key_to_disk(&disk_key, key);
5015 memcpy(ptr, &disk_key, sizeof(disk_key));
5016 ptr += sizeof(disk_key);
5017 memcpy(ptr, chunk, item_size);
5018 item_size += sizeof(disk_key);
5019 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
5025 * sort the devices in descending order by max_avail, total_avail
5027 static int btrfs_cmp_device_info(const void *a, const void *b)
5029 const struct btrfs_device_info *di_a = a;
5030 const struct btrfs_device_info *di_b = b;
5032 if (di_a->max_avail > di_b->max_avail)
5034 if (di_a->max_avail < di_b->max_avail)
5036 if (di_a->total_avail > di_b->total_avail)
5038 if (di_a->total_avail < di_b->total_avail)
5043 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
5045 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
5048 btrfs_set_fs_incompat(info, RAID56);
5051 static void check_raid1c34_incompat_flag(struct btrfs_fs_info *info, u64 type)
5053 if (!(type & (BTRFS_BLOCK_GROUP_RAID1C3 | BTRFS_BLOCK_GROUP_RAID1C4)))
5056 btrfs_set_fs_incompat(info, RAID1C34);
5060 * Structure used internally for btrfs_create_chunk() function.
5061 * Wraps needed parameters.
5063 struct alloc_chunk_ctl {
5066 /* Total number of stripes to allocate */
5068 /* sub_stripes info for map */
5070 /* Stripes per device */
5072 /* Maximum number of devices to use */
5074 /* Minimum number of devices to use */
5076 /* ndevs has to be a multiple of this */
5078 /* Number of copies */
5080 /* Number of stripes worth of bytes to store parity information */
5082 u64 max_stripe_size;
5090 static void init_alloc_chunk_ctl_policy_regular(
5091 struct btrfs_fs_devices *fs_devices,
5092 struct alloc_chunk_ctl *ctl)
5094 u64 type = ctl->type;
5096 if (type & BTRFS_BLOCK_GROUP_DATA) {
5097 ctl->max_stripe_size = SZ_1G;
5098 ctl->max_chunk_size = BTRFS_MAX_DATA_CHUNK_SIZE;
5099 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
5100 /* For larger filesystems, use larger metadata chunks */
5101 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
5102 ctl->max_stripe_size = SZ_1G;
5104 ctl->max_stripe_size = SZ_256M;
5105 ctl->max_chunk_size = ctl->max_stripe_size;
5106 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
5107 ctl->max_stripe_size = SZ_32M;
5108 ctl->max_chunk_size = 2 * ctl->max_stripe_size;
5109 ctl->devs_max = min_t(int, ctl->devs_max,
5110 BTRFS_MAX_DEVS_SYS_CHUNK);
5115 /* We don't want a chunk larger than 10% of writable space */
5116 ctl->max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
5117 ctl->max_chunk_size);
5118 ctl->dev_extent_min = BTRFS_STRIPE_LEN * ctl->dev_stripes;
5121 static void init_alloc_chunk_ctl_policy_zoned(
5122 struct btrfs_fs_devices *fs_devices,
5123 struct alloc_chunk_ctl *ctl)
5125 u64 zone_size = fs_devices->fs_info->zone_size;
5127 int min_num_stripes = ctl->devs_min * ctl->dev_stripes;
5128 int min_data_stripes = (min_num_stripes - ctl->nparity) / ctl->ncopies;
5129 u64 min_chunk_size = min_data_stripes * zone_size;
5130 u64 type = ctl->type;
5132 ctl->max_stripe_size = zone_size;
5133 if (type & BTRFS_BLOCK_GROUP_DATA) {
5134 ctl->max_chunk_size = round_down(BTRFS_MAX_DATA_CHUNK_SIZE,
5136 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
5137 ctl->max_chunk_size = ctl->max_stripe_size;
5138 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
5139 ctl->max_chunk_size = 2 * ctl->max_stripe_size;
5140 ctl->devs_max = min_t(int, ctl->devs_max,
5141 BTRFS_MAX_DEVS_SYS_CHUNK);
5146 /* We don't want a chunk larger than 10% of writable space */
5147 limit = max(round_down(div_factor(fs_devices->total_rw_bytes, 1),
5150 ctl->max_chunk_size = min(limit, ctl->max_chunk_size);
5151 ctl->dev_extent_min = zone_size * ctl->dev_stripes;
5154 static void init_alloc_chunk_ctl(struct btrfs_fs_devices *fs_devices,
5155 struct alloc_chunk_ctl *ctl)
5157 int index = btrfs_bg_flags_to_raid_index(ctl->type);
5159 ctl->sub_stripes = btrfs_raid_array[index].sub_stripes;
5160 ctl->dev_stripes = btrfs_raid_array[index].dev_stripes;
5161 ctl->devs_max = btrfs_raid_array[index].devs_max;
5163 ctl->devs_max = BTRFS_MAX_DEVS(fs_devices->fs_info);
5164 ctl->devs_min = btrfs_raid_array[index].devs_min;
5165 ctl->devs_increment = btrfs_raid_array[index].devs_increment;
5166 ctl->ncopies = btrfs_raid_array[index].ncopies;
5167 ctl->nparity = btrfs_raid_array[index].nparity;
5170 switch (fs_devices->chunk_alloc_policy) {
5171 case BTRFS_CHUNK_ALLOC_REGULAR:
5172 init_alloc_chunk_ctl_policy_regular(fs_devices, ctl);
5174 case BTRFS_CHUNK_ALLOC_ZONED:
5175 init_alloc_chunk_ctl_policy_zoned(fs_devices, ctl);
5182 static int gather_device_info(struct btrfs_fs_devices *fs_devices,
5183 struct alloc_chunk_ctl *ctl,
5184 struct btrfs_device_info *devices_info)
5186 struct btrfs_fs_info *info = fs_devices->fs_info;
5187 struct btrfs_device *device;
5189 u64 dev_extent_want = ctl->max_stripe_size * ctl->dev_stripes;
5196 * in the first pass through the devices list, we gather information
5197 * about the available holes on each device.
5199 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
5200 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
5202 "BTRFS: read-only device in alloc_list\n");
5206 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
5207 &device->dev_state) ||
5208 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
5211 if (device->total_bytes > device->bytes_used)
5212 total_avail = device->total_bytes - device->bytes_used;
5216 /* If there is no space on this device, skip it. */
5217 if (total_avail < ctl->dev_extent_min)
5220 ret = find_free_dev_extent(device, dev_extent_want, &dev_offset,
5222 if (ret && ret != -ENOSPC)
5226 max_avail = dev_extent_want;
5228 if (max_avail < ctl->dev_extent_min) {
5229 if (btrfs_test_opt(info, ENOSPC_DEBUG))
5231 "%s: devid %llu has no free space, have=%llu want=%llu",
5232 __func__, device->devid, max_avail,
5233 ctl->dev_extent_min);
5237 if (ndevs == fs_devices->rw_devices) {
5238 WARN(1, "%s: found more than %llu devices\n",
5239 __func__, fs_devices->rw_devices);
5242 devices_info[ndevs].dev_offset = dev_offset;
5243 devices_info[ndevs].max_avail = max_avail;
5244 devices_info[ndevs].total_avail = total_avail;
5245 devices_info[ndevs].dev = device;
5251 * now sort the devices by hole size / available space
5253 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
5254 btrfs_cmp_device_info, NULL);
5259 static int decide_stripe_size_regular(struct alloc_chunk_ctl *ctl,
5260 struct btrfs_device_info *devices_info)
5262 /* Number of stripes that count for block group size */
5266 * The primary goal is to maximize the number of stripes, so use as
5267 * many devices as possible, even if the stripes are not maximum sized.
5269 * The DUP profile stores more than one stripe per device, the
5270 * max_avail is the total size so we have to adjust.
5272 ctl->stripe_size = div_u64(devices_info[ctl->ndevs - 1].max_avail,
5274 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5276 /* This will have to be fixed for RAID1 and RAID10 over more drives */
5277 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5280 * Use the number of data stripes to figure out how big this chunk is
5281 * really going to be in terms of logical address space, and compare
5282 * that answer with the max chunk size. If it's higher, we try to
5283 * reduce stripe_size.
5285 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) {
5287 * Reduce stripe_size, round it up to a 16MB boundary again and
5288 * then use it, unless it ends up being even bigger than the
5289 * previous value we had already.
5291 ctl->stripe_size = min(round_up(div_u64(ctl->max_chunk_size,
5292 data_stripes), SZ_16M),
5296 /* Align to BTRFS_STRIPE_LEN */
5297 ctl->stripe_size = round_down(ctl->stripe_size, BTRFS_STRIPE_LEN);
5298 ctl->chunk_size = ctl->stripe_size * data_stripes;
5303 static int decide_stripe_size_zoned(struct alloc_chunk_ctl *ctl,
5304 struct btrfs_device_info *devices_info)
5306 u64 zone_size = devices_info[0].dev->zone_info->zone_size;
5307 /* Number of stripes that count for block group size */
5311 * It should hold because:
5312 * dev_extent_min == dev_extent_want == zone_size * dev_stripes
5314 ASSERT(devices_info[ctl->ndevs - 1].max_avail == ctl->dev_extent_min);
5316 ctl->stripe_size = zone_size;
5317 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5318 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5320 /* stripe_size is fixed in zoned filesysmte. Reduce ndevs instead. */
5321 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) {
5322 ctl->ndevs = div_u64(div_u64(ctl->max_chunk_size * ctl->ncopies,
5323 ctl->stripe_size) + ctl->nparity,
5325 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5326 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5327 ASSERT(ctl->stripe_size * data_stripes <= ctl->max_chunk_size);
5330 ctl->chunk_size = ctl->stripe_size * data_stripes;
5335 static int decide_stripe_size(struct btrfs_fs_devices *fs_devices,
5336 struct alloc_chunk_ctl *ctl,
5337 struct btrfs_device_info *devices_info)
5339 struct btrfs_fs_info *info = fs_devices->fs_info;
5342 * Round down to number of usable stripes, devs_increment can be any
5343 * number so we can't use round_down() that requires power of 2, while
5344 * rounddown is safe.
5346 ctl->ndevs = rounddown(ctl->ndevs, ctl->devs_increment);
5348 if (ctl->ndevs < ctl->devs_min) {
5349 if (btrfs_test_opt(info, ENOSPC_DEBUG)) {
5351 "%s: not enough devices with free space: have=%d minimum required=%d",
5352 __func__, ctl->ndevs, ctl->devs_min);
5357 ctl->ndevs = min(ctl->ndevs, ctl->devs_max);
5359 switch (fs_devices->chunk_alloc_policy) {
5360 case BTRFS_CHUNK_ALLOC_REGULAR:
5361 return decide_stripe_size_regular(ctl, devices_info);
5362 case BTRFS_CHUNK_ALLOC_ZONED:
5363 return decide_stripe_size_zoned(ctl, devices_info);
5369 static struct btrfs_block_group *create_chunk(struct btrfs_trans_handle *trans,
5370 struct alloc_chunk_ctl *ctl,
5371 struct btrfs_device_info *devices_info)
5373 struct btrfs_fs_info *info = trans->fs_info;
5374 struct map_lookup *map = NULL;
5375 struct extent_map_tree *em_tree;
5376 struct btrfs_block_group *block_group;
5377 struct extent_map *em;
5378 u64 start = ctl->start;
5379 u64 type = ctl->type;
5384 map = kmalloc(map_lookup_size(ctl->num_stripes), GFP_NOFS);
5386 return ERR_PTR(-ENOMEM);
5387 map->num_stripes = ctl->num_stripes;
5389 for (i = 0; i < ctl->ndevs; ++i) {
5390 for (j = 0; j < ctl->dev_stripes; ++j) {
5391 int s = i * ctl->dev_stripes + j;
5392 map->stripes[s].dev = devices_info[i].dev;
5393 map->stripes[s].physical = devices_info[i].dev_offset +
5394 j * ctl->stripe_size;
5397 map->stripe_len = BTRFS_STRIPE_LEN;
5398 map->io_align = BTRFS_STRIPE_LEN;
5399 map->io_width = BTRFS_STRIPE_LEN;
5401 map->sub_stripes = ctl->sub_stripes;
5403 trace_btrfs_chunk_alloc(info, map, start, ctl->chunk_size);
5405 em = alloc_extent_map();
5408 return ERR_PTR(-ENOMEM);
5410 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
5411 em->map_lookup = map;
5413 em->len = ctl->chunk_size;
5414 em->block_start = 0;
5415 em->block_len = em->len;
5416 em->orig_block_len = ctl->stripe_size;
5418 em_tree = &info->mapping_tree;
5419 write_lock(&em_tree->lock);
5420 ret = add_extent_mapping(em_tree, em, 0);
5422 write_unlock(&em_tree->lock);
5423 free_extent_map(em);
5424 return ERR_PTR(ret);
5426 write_unlock(&em_tree->lock);
5428 block_group = btrfs_make_block_group(trans, 0, type, start, ctl->chunk_size);
5429 if (IS_ERR(block_group))
5430 goto error_del_extent;
5432 for (i = 0; i < map->num_stripes; i++) {
5433 struct btrfs_device *dev = map->stripes[i].dev;
5435 btrfs_device_set_bytes_used(dev,
5436 dev->bytes_used + ctl->stripe_size);
5437 if (list_empty(&dev->post_commit_list))
5438 list_add_tail(&dev->post_commit_list,
5439 &trans->transaction->dev_update_list);
5442 atomic64_sub(ctl->stripe_size * map->num_stripes,
5443 &info->free_chunk_space);
5445 free_extent_map(em);
5446 check_raid56_incompat_flag(info, type);
5447 check_raid1c34_incompat_flag(info, type);
5452 write_lock(&em_tree->lock);
5453 remove_extent_mapping(em_tree, em);
5454 write_unlock(&em_tree->lock);
5456 /* One for our allocation */
5457 free_extent_map(em);
5458 /* One for the tree reference */
5459 free_extent_map(em);
5464 struct btrfs_block_group *btrfs_create_chunk(struct btrfs_trans_handle *trans,
5467 struct btrfs_fs_info *info = trans->fs_info;
5468 struct btrfs_fs_devices *fs_devices = info->fs_devices;
5469 struct btrfs_device_info *devices_info = NULL;
5470 struct alloc_chunk_ctl ctl;
5471 struct btrfs_block_group *block_group;
5474 lockdep_assert_held(&info->chunk_mutex);
5476 if (!alloc_profile_is_valid(type, 0)) {
5478 return ERR_PTR(-EINVAL);
5481 if (list_empty(&fs_devices->alloc_list)) {
5482 if (btrfs_test_opt(info, ENOSPC_DEBUG))
5483 btrfs_debug(info, "%s: no writable device", __func__);
5484 return ERR_PTR(-ENOSPC);
5487 if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
5488 btrfs_err(info, "invalid chunk type 0x%llx requested", type);
5490 return ERR_PTR(-EINVAL);
5493 ctl.start = find_next_chunk(info);
5495 init_alloc_chunk_ctl(fs_devices, &ctl);
5497 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
5500 return ERR_PTR(-ENOMEM);
5502 ret = gather_device_info(fs_devices, &ctl, devices_info);
5504 block_group = ERR_PTR(ret);
5508 ret = decide_stripe_size(fs_devices, &ctl, devices_info);
5510 block_group = ERR_PTR(ret);
5514 block_group = create_chunk(trans, &ctl, devices_info);
5517 kfree(devices_info);
5522 * This function, btrfs_chunk_alloc_add_chunk_item(), typically belongs to the
5523 * phase 1 of chunk allocation. It belongs to phase 2 only when allocating system
5526 * See the comment at btrfs_chunk_alloc() for details about the chunk allocation
5529 int btrfs_chunk_alloc_add_chunk_item(struct btrfs_trans_handle *trans,
5530 struct btrfs_block_group *bg)
5532 struct btrfs_fs_info *fs_info = trans->fs_info;
5533 struct btrfs_root *chunk_root = fs_info->chunk_root;
5534 struct btrfs_key key;
5535 struct btrfs_chunk *chunk;
5536 struct btrfs_stripe *stripe;
5537 struct extent_map *em;
5538 struct map_lookup *map;
5544 * We take the chunk_mutex for 2 reasons:
5546 * 1) Updates and insertions in the chunk btree must be done while holding
5547 * the chunk_mutex, as well as updating the system chunk array in the
5548 * superblock. See the comment on top of btrfs_chunk_alloc() for the
5551 * 2) To prevent races with the final phase of a device replace operation
5552 * that replaces the device object associated with the map's stripes,
5553 * because the device object's id can change at any time during that
5554 * final phase of the device replace operation
5555 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the
5556 * replaced device and then see it with an ID of BTRFS_DEV_REPLACE_DEVID,
5557 * which would cause a failure when updating the device item, which does
5558 * not exists, or persisting a stripe of the chunk item with such ID.
5559 * Here we can't use the device_list_mutex because our caller already
5560 * has locked the chunk_mutex, and the final phase of device replace
5561 * acquires both mutexes - first the device_list_mutex and then the
5562 * chunk_mutex. Using any of those two mutexes protects us from a
5563 * concurrent device replace.
5565 lockdep_assert_held(&fs_info->chunk_mutex);
5567 em = btrfs_get_chunk_map(fs_info, bg->start, bg->length);
5570 btrfs_abort_transaction(trans, ret);
5574 map = em->map_lookup;
5575 item_size = btrfs_chunk_item_size(map->num_stripes);
5577 chunk = kzalloc(item_size, GFP_NOFS);
5580 btrfs_abort_transaction(trans, ret);
5584 for (i = 0; i < map->num_stripes; i++) {
5585 struct btrfs_device *device = map->stripes[i].dev;
5587 ret = btrfs_update_device(trans, device);
5592 stripe = &chunk->stripe;
5593 for (i = 0; i < map->num_stripes; i++) {
5594 struct btrfs_device *device = map->stripes[i].dev;
5595 const u64 dev_offset = map->stripes[i].physical;
5597 btrfs_set_stack_stripe_devid(stripe, device->devid);
5598 btrfs_set_stack_stripe_offset(stripe, dev_offset);
5599 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
5603 btrfs_set_stack_chunk_length(chunk, bg->length);
5604 btrfs_set_stack_chunk_owner(chunk, BTRFS_EXTENT_TREE_OBJECTID);
5605 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
5606 btrfs_set_stack_chunk_type(chunk, map->type);
5607 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
5608 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
5609 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
5610 btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
5611 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
5613 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
5614 key.type = BTRFS_CHUNK_ITEM_KEY;
5615 key.offset = bg->start;
5617 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
5621 bg->chunk_item_inserted = 1;
5623 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
5624 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
5631 free_extent_map(em);
5635 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans)
5637 struct btrfs_fs_info *fs_info = trans->fs_info;
5639 struct btrfs_block_group *meta_bg;
5640 struct btrfs_block_group *sys_bg;
5643 * When adding a new device for sprouting, the seed device is read-only
5644 * so we must first allocate a metadata and a system chunk. But before
5645 * adding the block group items to the extent, device and chunk btrees,
5648 * 1) Create both chunks without doing any changes to the btrees, as
5649 * otherwise we would get -ENOSPC since the block groups from the
5650 * seed device are read-only;
5652 * 2) Add the device item for the new sprout device - finishing the setup
5653 * of a new block group requires updating the device item in the chunk
5654 * btree, so it must exist when we attempt to do it. The previous step
5655 * ensures this does not fail with -ENOSPC.
5657 * After that we can add the block group items to their btrees:
5658 * update existing device item in the chunk btree, add a new block group
5659 * item to the extent btree, add a new chunk item to the chunk btree and
5660 * finally add the new device extent items to the devices btree.
5663 alloc_profile = btrfs_metadata_alloc_profile(fs_info);
5664 meta_bg = btrfs_create_chunk(trans, alloc_profile);
5665 if (IS_ERR(meta_bg))
5666 return PTR_ERR(meta_bg);
5668 alloc_profile = btrfs_system_alloc_profile(fs_info);
5669 sys_bg = btrfs_create_chunk(trans, alloc_profile);
5671 return PTR_ERR(sys_bg);
5676 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
5678 const int index = btrfs_bg_flags_to_raid_index(map->type);
5680 return btrfs_raid_array[index].tolerated_failures;
5683 bool btrfs_chunk_writeable(struct btrfs_fs_info *fs_info, u64 chunk_offset)
5685 struct extent_map *em;
5686 struct map_lookup *map;
5691 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
5695 map = em->map_lookup;
5696 for (i = 0; i < map->num_stripes; i++) {
5697 if (test_bit(BTRFS_DEV_STATE_MISSING,
5698 &map->stripes[i].dev->dev_state)) {
5702 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE,
5703 &map->stripes[i].dev->dev_state)) {
5710 * If the number of missing devices is larger than max errors, we can
5711 * not write the data into that chunk successfully.
5713 if (miss_ndevs > btrfs_chunk_max_errors(map))
5716 free_extent_map(em);
5720 void btrfs_mapping_tree_free(struct extent_map_tree *tree)
5722 struct extent_map *em;
5725 write_lock(&tree->lock);
5726 em = lookup_extent_mapping(tree, 0, (u64)-1);
5728 remove_extent_mapping(tree, em);
5729 write_unlock(&tree->lock);
5733 free_extent_map(em);
5734 /* once for the tree */
5735 free_extent_map(em);
5739 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5741 struct extent_map *em;
5742 struct map_lookup *map;
5745 em = btrfs_get_chunk_map(fs_info, logical, len);
5748 * We could return errors for these cases, but that could get
5749 * ugly and we'd probably do the same thing which is just not do
5750 * anything else and exit, so return 1 so the callers don't try
5751 * to use other copies.
5755 map = em->map_lookup;
5756 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1_MASK))
5757 ret = map->num_stripes;
5758 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5759 ret = map->sub_stripes;
5760 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5762 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5764 * There could be two corrupted data stripes, we need
5765 * to loop retry in order to rebuild the correct data.
5767 * Fail a stripe at a time on every retry except the
5768 * stripe under reconstruction.
5770 ret = map->num_stripes;
5773 free_extent_map(em);
5775 down_read(&fs_info->dev_replace.rwsem);
5776 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) &&
5777 fs_info->dev_replace.tgtdev)
5779 up_read(&fs_info->dev_replace.rwsem);
5784 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
5787 struct extent_map *em;
5788 struct map_lookup *map;
5789 unsigned long len = fs_info->sectorsize;
5791 em = btrfs_get_chunk_map(fs_info, logical, len);
5793 if (!WARN_ON(IS_ERR(em))) {
5794 map = em->map_lookup;
5795 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5796 len = map->stripe_len * nr_data_stripes(map);
5797 free_extent_map(em);
5802 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5804 struct extent_map *em;
5805 struct map_lookup *map;
5808 em = btrfs_get_chunk_map(fs_info, logical, len);
5810 if(!WARN_ON(IS_ERR(em))) {
5811 map = em->map_lookup;
5812 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5814 free_extent_map(em);
5819 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5820 struct map_lookup *map, int first,
5821 int dev_replace_is_ongoing)
5825 int preferred_mirror;
5827 struct btrfs_device *srcdev;
5830 (BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10)));
5832 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5833 num_stripes = map->sub_stripes;
5835 num_stripes = map->num_stripes;
5837 switch (fs_info->fs_devices->read_policy) {
5839 /* Shouldn't happen, just warn and use pid instead of failing */
5840 btrfs_warn_rl(fs_info,
5841 "unknown read_policy type %u, reset to pid",
5842 fs_info->fs_devices->read_policy);
5843 fs_info->fs_devices->read_policy = BTRFS_READ_POLICY_PID;
5845 case BTRFS_READ_POLICY_PID:
5846 preferred_mirror = first + (current->pid % num_stripes);
5850 if (dev_replace_is_ongoing &&
5851 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5852 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5853 srcdev = fs_info->dev_replace.srcdev;
5858 * try to avoid the drive that is the source drive for a
5859 * dev-replace procedure, only choose it if no other non-missing
5860 * mirror is available
5862 for (tolerance = 0; tolerance < 2; tolerance++) {
5863 if (map->stripes[preferred_mirror].dev->bdev &&
5864 (tolerance || map->stripes[preferred_mirror].dev != srcdev))
5865 return preferred_mirror;
5866 for (i = first; i < first + num_stripes; i++) {
5867 if (map->stripes[i].dev->bdev &&
5868 (tolerance || map->stripes[i].dev != srcdev))
5873 /* we couldn't find one that doesn't fail. Just return something
5874 * and the io error handling code will clean up eventually
5876 return preferred_mirror;
5879 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5880 static void sort_parity_stripes(struct btrfs_io_context *bioc, int num_stripes)
5887 for (i = 0; i < num_stripes - 1; i++) {
5888 /* Swap if parity is on a smaller index */
5889 if (bioc->raid_map[i] > bioc->raid_map[i + 1]) {
5890 swap(bioc->stripes[i], bioc->stripes[i + 1]);
5891 swap(bioc->raid_map[i], bioc->raid_map[i + 1]);
5898 static struct btrfs_io_context *alloc_btrfs_io_context(struct btrfs_fs_info *fs_info,
5902 struct btrfs_io_context *bioc = kzalloc(
5903 /* The size of btrfs_io_context */
5904 sizeof(struct btrfs_io_context) +
5905 /* Plus the variable array for the stripes */
5906 sizeof(struct btrfs_io_stripe) * (total_stripes) +
5907 /* Plus the variable array for the tgt dev */
5908 sizeof(int) * (real_stripes) +
5910 * Plus the raid_map, which includes both the tgt dev
5913 sizeof(u64) * (total_stripes),
5914 GFP_NOFS|__GFP_NOFAIL);
5916 atomic_set(&bioc->error, 0);
5917 refcount_set(&bioc->refs, 1);
5919 bioc->fs_info = fs_info;
5920 bioc->tgtdev_map = (int *)(bioc->stripes + total_stripes);
5921 bioc->raid_map = (u64 *)(bioc->tgtdev_map + real_stripes);
5926 void btrfs_get_bioc(struct btrfs_io_context *bioc)
5928 WARN_ON(!refcount_read(&bioc->refs));
5929 refcount_inc(&bioc->refs);
5932 void btrfs_put_bioc(struct btrfs_io_context *bioc)
5936 if (refcount_dec_and_test(&bioc->refs))
5940 /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */
5942 * Please note that, discard won't be sent to target device of device
5945 static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info,
5946 u64 logical, u64 *length_ret,
5947 struct btrfs_io_context **bioc_ret)
5949 struct extent_map *em;
5950 struct map_lookup *map;
5951 struct btrfs_io_context *bioc;
5952 u64 length = *length_ret;
5956 u64 stripe_end_offset;
5963 u32 sub_stripes = 0;
5964 u64 stripes_per_dev = 0;
5965 u32 remaining_stripes = 0;
5966 u32 last_stripe = 0;
5970 /* Discard always returns a bioc. */
5973 em = btrfs_get_chunk_map(fs_info, logical, length);
5977 map = em->map_lookup;
5978 /* we don't discard raid56 yet */
5979 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5984 offset = logical - em->start;
5985 length = min_t(u64, em->start + em->len - logical, length);
5986 *length_ret = length;
5988 stripe_len = map->stripe_len;
5990 * stripe_nr counts the total number of stripes we have to stride
5991 * to get to this block
5993 stripe_nr = div64_u64(offset, stripe_len);
5995 /* stripe_offset is the offset of this block in its stripe */
5996 stripe_offset = offset - stripe_nr * stripe_len;
5998 stripe_nr_end = round_up(offset + length, map->stripe_len);
5999 stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len);
6000 stripe_cnt = stripe_nr_end - stripe_nr;
6001 stripe_end_offset = stripe_nr_end * map->stripe_len -
6004 * after this, stripe_nr is the number of stripes on this
6005 * device we have to walk to find the data, and stripe_index is
6006 * the number of our device in the stripe array
6010 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
6011 BTRFS_BLOCK_GROUP_RAID10)) {
6012 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
6015 sub_stripes = map->sub_stripes;
6017 factor = map->num_stripes / sub_stripes;
6018 num_stripes = min_t(u64, map->num_stripes,
6019 sub_stripes * stripe_cnt);
6020 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
6021 stripe_index *= sub_stripes;
6022 stripes_per_dev = div_u64_rem(stripe_cnt, factor,
6023 &remaining_stripes);
6024 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
6025 last_stripe *= sub_stripes;
6026 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1_MASK |
6027 BTRFS_BLOCK_GROUP_DUP)) {
6028 num_stripes = map->num_stripes;
6030 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
6034 bioc = alloc_btrfs_io_context(fs_info, num_stripes, 0);
6040 for (i = 0; i < num_stripes; i++) {
6041 bioc->stripes[i].physical =
6042 map->stripes[stripe_index].physical +
6043 stripe_offset + stripe_nr * map->stripe_len;
6044 bioc->stripes[i].dev = map->stripes[stripe_index].dev;
6046 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
6047 BTRFS_BLOCK_GROUP_RAID10)) {
6048 bioc->stripes[i].length = stripes_per_dev *
6051 if (i / sub_stripes < remaining_stripes)
6052 bioc->stripes[i].length += map->stripe_len;
6055 * Special for the first stripe and
6058 * |-------|...|-------|
6062 if (i < sub_stripes)
6063 bioc->stripes[i].length -= stripe_offset;
6065 if (stripe_index >= last_stripe &&
6066 stripe_index <= (last_stripe +
6068 bioc->stripes[i].length -= stripe_end_offset;
6070 if (i == sub_stripes - 1)
6073 bioc->stripes[i].length = length;
6077 if (stripe_index == map->num_stripes) {
6084 bioc->map_type = map->type;
6085 bioc->num_stripes = num_stripes;
6087 free_extent_map(em);
6092 * In dev-replace case, for repair case (that's the only case where the mirror
6093 * is selected explicitly when calling btrfs_map_block), blocks left of the
6094 * left cursor can also be read from the target drive.
6096 * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the
6098 * For READ, it also needs to be supported using the same mirror number.
6100 * If the requested block is not left of the left cursor, EIO is returned. This
6101 * can happen because btrfs_num_copies() returns one more in the dev-replace
6104 static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info,
6105 u64 logical, u64 length,
6106 u64 srcdev_devid, int *mirror_num,
6109 struct btrfs_io_context *bioc = NULL;
6111 int index_srcdev = 0;
6113 u64 physical_of_found = 0;
6117 ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS,
6118 logical, &length, &bioc, 0, 0);
6120 ASSERT(bioc == NULL);
6124 num_stripes = bioc->num_stripes;
6125 if (*mirror_num > num_stripes) {
6127 * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror,
6128 * that means that the requested area is not left of the left
6131 btrfs_put_bioc(bioc);
6136 * process the rest of the function using the mirror_num of the source
6137 * drive. Therefore look it up first. At the end, patch the device
6138 * pointer to the one of the target drive.
6140 for (i = 0; i < num_stripes; i++) {
6141 if (bioc->stripes[i].dev->devid != srcdev_devid)
6145 * In case of DUP, in order to keep it simple, only add the
6146 * mirror with the lowest physical address
6149 physical_of_found <= bioc->stripes[i].physical)
6154 physical_of_found = bioc->stripes[i].physical;
6157 btrfs_put_bioc(bioc);
6163 *mirror_num = index_srcdev + 1;
6164 *physical = physical_of_found;
6168 static bool is_block_group_to_copy(struct btrfs_fs_info *fs_info, u64 logical)
6170 struct btrfs_block_group *cache;
6173 /* Non zoned filesystem does not use "to_copy" flag */
6174 if (!btrfs_is_zoned(fs_info))
6177 cache = btrfs_lookup_block_group(fs_info, logical);
6179 spin_lock(&cache->lock);
6180 ret = cache->to_copy;
6181 spin_unlock(&cache->lock);
6183 btrfs_put_block_group(cache);
6187 static void handle_ops_on_dev_replace(enum btrfs_map_op op,
6188 struct btrfs_io_context **bioc_ret,
6189 struct btrfs_dev_replace *dev_replace,
6191 int *num_stripes_ret, int *max_errors_ret)
6193 struct btrfs_io_context *bioc = *bioc_ret;
6194 u64 srcdev_devid = dev_replace->srcdev->devid;
6195 int tgtdev_indexes = 0;
6196 int num_stripes = *num_stripes_ret;
6197 int max_errors = *max_errors_ret;
6200 if (op == BTRFS_MAP_WRITE) {
6201 int index_where_to_add;
6204 * A block group which have "to_copy" set will eventually
6205 * copied by dev-replace process. We can avoid cloning IO here.
6207 if (is_block_group_to_copy(dev_replace->srcdev->fs_info, logical))
6211 * duplicate the write operations while the dev replace
6212 * procedure is running. Since the copying of the old disk to
6213 * the new disk takes place at run time while the filesystem is
6214 * mounted writable, the regular write operations to the old
6215 * disk have to be duplicated to go to the new disk as well.
6217 * Note that device->missing is handled by the caller, and that
6218 * the write to the old disk is already set up in the stripes
6221 index_where_to_add = num_stripes;
6222 for (i = 0; i < num_stripes; i++) {
6223 if (bioc->stripes[i].dev->devid == srcdev_devid) {
6224 /* write to new disk, too */
6225 struct btrfs_io_stripe *new =
6226 bioc->stripes + index_where_to_add;
6227 struct btrfs_io_stripe *old =
6230 new->physical = old->physical;
6231 new->length = old->length;
6232 new->dev = dev_replace->tgtdev;
6233 bioc->tgtdev_map[i] = index_where_to_add;
6234 index_where_to_add++;
6239 num_stripes = index_where_to_add;
6240 } else if (op == BTRFS_MAP_GET_READ_MIRRORS) {
6241 int index_srcdev = 0;
6243 u64 physical_of_found = 0;
6246 * During the dev-replace procedure, the target drive can also
6247 * be used to read data in case it is needed to repair a corrupt
6248 * block elsewhere. This is possible if the requested area is
6249 * left of the left cursor. In this area, the target drive is a
6250 * full copy of the source drive.
6252 for (i = 0; i < num_stripes; i++) {
6253 if (bioc->stripes[i].dev->devid == srcdev_devid) {
6255 * In case of DUP, in order to keep it simple,
6256 * only add the mirror with the lowest physical
6260 physical_of_found <= bioc->stripes[i].physical)
6264 physical_of_found = bioc->stripes[i].physical;
6268 struct btrfs_io_stripe *tgtdev_stripe =
6269 bioc->stripes + num_stripes;
6271 tgtdev_stripe->physical = physical_of_found;
6272 tgtdev_stripe->length =
6273 bioc->stripes[index_srcdev].length;
6274 tgtdev_stripe->dev = dev_replace->tgtdev;
6275 bioc->tgtdev_map[index_srcdev] = num_stripes;
6282 *num_stripes_ret = num_stripes;
6283 *max_errors_ret = max_errors;
6284 bioc->num_tgtdevs = tgtdev_indexes;
6288 static bool need_full_stripe(enum btrfs_map_op op)
6290 return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS);
6294 * Calculate the geometry of a particular (address, len) tuple. This
6295 * information is used to calculate how big a particular bio can get before it
6296 * straddles a stripe.
6298 * @fs_info: the filesystem
6299 * @em: mapping containing the logical extent
6300 * @op: type of operation - write or read
6301 * @logical: address that we want to figure out the geometry of
6302 * @io_geom: pointer used to return values
6304 * Returns < 0 in case a chunk for the given logical address cannot be found,
6305 * usually shouldn't happen unless @logical is corrupted, 0 otherwise.
6307 int btrfs_get_io_geometry(struct btrfs_fs_info *fs_info, struct extent_map *em,
6308 enum btrfs_map_op op, u64 logical,
6309 struct btrfs_io_geometry *io_geom)
6311 struct map_lookup *map;
6317 u64 raid56_full_stripe_start = (u64)-1;
6320 ASSERT(op != BTRFS_MAP_DISCARD);
6322 map = em->map_lookup;
6323 /* Offset of this logical address in the chunk */
6324 offset = logical - em->start;
6325 /* Len of a stripe in a chunk */
6326 stripe_len = map->stripe_len;
6327 /* Stripe where this block falls in */
6328 stripe_nr = div64_u64(offset, stripe_len);
6329 /* Offset of stripe in the chunk */
6330 stripe_offset = stripe_nr * stripe_len;
6331 if (offset < stripe_offset) {
6333 "stripe math has gone wrong, stripe_offset=%llu offset=%llu start=%llu logical=%llu stripe_len=%llu",
6334 stripe_offset, offset, em->start, logical, stripe_len);
6338 /* stripe_offset is the offset of this block in its stripe */
6339 stripe_offset = offset - stripe_offset;
6340 data_stripes = nr_data_stripes(map);
6342 /* Only stripe based profiles needs to check against stripe length. */
6343 if (map->type & BTRFS_BLOCK_GROUP_STRIPE_MASK) {
6344 u64 max_len = stripe_len - stripe_offset;
6347 * In case of raid56, we need to know the stripe aligned start
6349 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6350 unsigned long full_stripe_len = stripe_len * data_stripes;
6351 raid56_full_stripe_start = offset;
6354 * Allow a write of a full stripe, but make sure we
6355 * don't allow straddling of stripes
6357 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
6359 raid56_full_stripe_start *= full_stripe_len;
6362 * For writes to RAID[56], allow a full stripeset across
6363 * all disks. For other RAID types and for RAID[56]
6364 * reads, just allow a single stripe (on a single disk).
6366 if (op == BTRFS_MAP_WRITE) {
6367 max_len = stripe_len * data_stripes -
6368 (offset - raid56_full_stripe_start);
6371 len = min_t(u64, em->len - offset, max_len);
6373 len = em->len - offset;
6377 io_geom->offset = offset;
6378 io_geom->stripe_len = stripe_len;
6379 io_geom->stripe_nr = stripe_nr;
6380 io_geom->stripe_offset = stripe_offset;
6381 io_geom->raid56_stripe_offset = raid56_full_stripe_start;
6386 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
6387 enum btrfs_map_op op,
6388 u64 logical, u64 *length,
6389 struct btrfs_io_context **bioc_ret,
6390 int mirror_num, int need_raid_map)
6392 struct extent_map *em;
6393 struct map_lookup *map;
6403 int tgtdev_indexes = 0;
6404 struct btrfs_io_context *bioc = NULL;
6405 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
6406 int dev_replace_is_ongoing = 0;
6407 int num_alloc_stripes;
6408 int patch_the_first_stripe_for_dev_replace = 0;
6409 u64 physical_to_patch_in_first_stripe = 0;
6410 u64 raid56_full_stripe_start = (u64)-1;
6411 struct btrfs_io_geometry geom;
6414 ASSERT(op != BTRFS_MAP_DISCARD);
6416 em = btrfs_get_chunk_map(fs_info, logical, *length);
6417 ASSERT(!IS_ERR(em));
6419 ret = btrfs_get_io_geometry(fs_info, em, op, logical, &geom);
6423 map = em->map_lookup;
6426 stripe_len = geom.stripe_len;
6427 stripe_nr = geom.stripe_nr;
6428 stripe_offset = geom.stripe_offset;
6429 raid56_full_stripe_start = geom.raid56_stripe_offset;
6430 data_stripes = nr_data_stripes(map);
6432 down_read(&dev_replace->rwsem);
6433 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
6435 * Hold the semaphore for read during the whole operation, write is
6436 * requested at commit time but must wait.
6438 if (!dev_replace_is_ongoing)
6439 up_read(&dev_replace->rwsem);
6441 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
6442 !need_full_stripe(op) && dev_replace->tgtdev != NULL) {
6443 ret = get_extra_mirror_from_replace(fs_info, logical, *length,
6444 dev_replace->srcdev->devid,
6446 &physical_to_patch_in_first_stripe);
6450 patch_the_first_stripe_for_dev_replace = 1;
6451 } else if (mirror_num > map->num_stripes) {
6457 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
6458 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
6460 if (!need_full_stripe(op))
6462 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1_MASK) {
6463 if (need_full_stripe(op))
6464 num_stripes = map->num_stripes;
6465 else if (mirror_num)
6466 stripe_index = mirror_num - 1;
6468 stripe_index = find_live_mirror(fs_info, map, 0,
6469 dev_replace_is_ongoing);
6470 mirror_num = stripe_index + 1;
6473 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
6474 if (need_full_stripe(op)) {
6475 num_stripes = map->num_stripes;
6476 } else if (mirror_num) {
6477 stripe_index = mirror_num - 1;
6482 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
6483 u32 factor = map->num_stripes / map->sub_stripes;
6485 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
6486 stripe_index *= map->sub_stripes;
6488 if (need_full_stripe(op))
6489 num_stripes = map->sub_stripes;
6490 else if (mirror_num)
6491 stripe_index += mirror_num - 1;
6493 int old_stripe_index = stripe_index;
6494 stripe_index = find_live_mirror(fs_info, map,
6496 dev_replace_is_ongoing);
6497 mirror_num = stripe_index - old_stripe_index + 1;
6500 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6501 if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) {
6502 /* push stripe_nr back to the start of the full stripe */
6503 stripe_nr = div64_u64(raid56_full_stripe_start,
6504 stripe_len * data_stripes);
6506 /* RAID[56] write or recovery. Return all stripes */
6507 num_stripes = map->num_stripes;
6508 max_errors = nr_parity_stripes(map);
6510 *length = map->stripe_len;
6515 * Mirror #0 or #1 means the original data block.
6516 * Mirror #2 is RAID5 parity block.
6517 * Mirror #3 is RAID6 Q block.
6519 stripe_nr = div_u64_rem(stripe_nr,
6520 data_stripes, &stripe_index);
6522 stripe_index = data_stripes + mirror_num - 2;
6524 /* We distribute the parity blocks across stripes */
6525 div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
6527 if (!need_full_stripe(op) && mirror_num <= 1)
6532 * after this, stripe_nr is the number of stripes on this
6533 * device we have to walk to find the data, and stripe_index is
6534 * the number of our device in the stripe array
6536 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
6538 mirror_num = stripe_index + 1;
6540 if (stripe_index >= map->num_stripes) {
6542 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
6543 stripe_index, map->num_stripes);
6548 num_alloc_stripes = num_stripes;
6549 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) {
6550 if (op == BTRFS_MAP_WRITE)
6551 num_alloc_stripes <<= 1;
6552 if (op == BTRFS_MAP_GET_READ_MIRRORS)
6553 num_alloc_stripes++;
6554 tgtdev_indexes = num_stripes;
6557 bioc = alloc_btrfs_io_context(fs_info, num_alloc_stripes, tgtdev_indexes);
6563 for (i = 0; i < num_stripes; i++) {
6564 bioc->stripes[i].physical = map->stripes[stripe_index].physical +
6565 stripe_offset + stripe_nr * map->stripe_len;
6566 bioc->stripes[i].dev = map->stripes[stripe_index].dev;
6570 /* Build raid_map */
6571 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map &&
6572 (need_full_stripe(op) || mirror_num > 1)) {
6576 /* Work out the disk rotation on this stripe-set */
6577 div_u64_rem(stripe_nr, num_stripes, &rot);
6579 /* Fill in the logical address of each stripe */
6580 tmp = stripe_nr * data_stripes;
6581 for (i = 0; i < data_stripes; i++)
6582 bioc->raid_map[(i + rot) % num_stripes] =
6583 em->start + (tmp + i) * map->stripe_len;
6585 bioc->raid_map[(i + rot) % map->num_stripes] = RAID5_P_STRIPE;
6586 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
6587 bioc->raid_map[(i + rot + 1) % num_stripes] =
6590 sort_parity_stripes(bioc, num_stripes);
6593 if (need_full_stripe(op))
6594 max_errors = btrfs_chunk_max_errors(map);
6596 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
6597 need_full_stripe(op)) {
6598 handle_ops_on_dev_replace(op, &bioc, dev_replace, logical,
6599 &num_stripes, &max_errors);
6603 bioc->map_type = map->type;
6604 bioc->num_stripes = num_stripes;
6605 bioc->max_errors = max_errors;
6606 bioc->mirror_num = mirror_num;
6609 * this is the case that REQ_READ && dev_replace_is_ongoing &&
6610 * mirror_num == num_stripes + 1 && dev_replace target drive is
6611 * available as a mirror
6613 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
6614 WARN_ON(num_stripes > 1);
6615 bioc->stripes[0].dev = dev_replace->tgtdev;
6616 bioc->stripes[0].physical = physical_to_patch_in_first_stripe;
6617 bioc->mirror_num = map->num_stripes + 1;
6620 if (dev_replace_is_ongoing) {
6621 lockdep_assert_held(&dev_replace->rwsem);
6622 /* Unlock and let waiting writers proceed */
6623 up_read(&dev_replace->rwsem);
6625 free_extent_map(em);
6629 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6630 u64 logical, u64 *length,
6631 struct btrfs_io_context **bioc_ret, int mirror_num)
6633 if (op == BTRFS_MAP_DISCARD)
6634 return __btrfs_map_block_for_discard(fs_info, logical,
6637 return __btrfs_map_block(fs_info, op, logical, length, bioc_ret,
6641 /* For Scrub/replace */
6642 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6643 u64 logical, u64 *length,
6644 struct btrfs_io_context **bioc_ret)
6646 return __btrfs_map_block(fs_info, op, logical, length, bioc_ret, 0, 1);
6649 static inline void btrfs_end_bioc(struct btrfs_io_context *bioc, struct bio *bio)
6651 bio->bi_private = bioc->private;
6652 bio->bi_end_io = bioc->end_io;
6655 btrfs_put_bioc(bioc);
6658 static void btrfs_end_bio(struct bio *bio)
6660 struct btrfs_io_context *bioc = bio->bi_private;
6661 int is_orig_bio = 0;
6663 if (bio->bi_status) {
6664 atomic_inc(&bioc->error);
6665 if (bio->bi_status == BLK_STS_IOERR ||
6666 bio->bi_status == BLK_STS_TARGET) {
6667 struct btrfs_device *dev = btrfs_bio(bio)->device;
6670 if (btrfs_op(bio) == BTRFS_MAP_WRITE)
6671 btrfs_dev_stat_inc_and_print(dev,
6672 BTRFS_DEV_STAT_WRITE_ERRS);
6673 else if (!(bio->bi_opf & REQ_RAHEAD))
6674 btrfs_dev_stat_inc_and_print(dev,
6675 BTRFS_DEV_STAT_READ_ERRS);
6676 if (bio->bi_opf & REQ_PREFLUSH)
6677 btrfs_dev_stat_inc_and_print(dev,
6678 BTRFS_DEV_STAT_FLUSH_ERRS);
6682 if (bio == bioc->orig_bio)
6685 btrfs_bio_counter_dec(bioc->fs_info);
6687 if (atomic_dec_and_test(&bioc->stripes_pending)) {
6690 bio = bioc->orig_bio;
6693 btrfs_bio(bio)->mirror_num = bioc->mirror_num;
6694 /* only send an error to the higher layers if it is
6695 * beyond the tolerance of the btrfs bio
6697 if (atomic_read(&bioc->error) > bioc->max_errors) {
6698 bio->bi_status = BLK_STS_IOERR;
6701 * this bio is actually up to date, we didn't
6702 * go over the max number of errors
6704 bio->bi_status = BLK_STS_OK;
6707 btrfs_end_bioc(bioc, bio);
6708 } else if (!is_orig_bio) {
6713 static void submit_stripe_bio(struct btrfs_io_context *bioc, struct bio *bio,
6714 u64 physical, struct btrfs_device *dev)
6716 struct btrfs_fs_info *fs_info = bioc->fs_info;
6718 bio->bi_private = bioc;
6719 btrfs_bio(bio)->device = dev;
6720 bio->bi_end_io = btrfs_end_bio;
6721 bio->bi_iter.bi_sector = physical >> 9;
6723 * For zone append writing, bi_sector must point the beginning of the
6726 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
6727 if (btrfs_dev_is_sequential(dev, physical)) {
6728 u64 zone_start = round_down(physical, fs_info->zone_size);
6730 bio->bi_iter.bi_sector = zone_start >> SECTOR_SHIFT;
6732 bio->bi_opf &= ~REQ_OP_ZONE_APPEND;
6733 bio->bi_opf |= REQ_OP_WRITE;
6736 btrfs_debug_in_rcu(fs_info,
6737 "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
6738 bio_op(bio), bio->bi_opf, bio->bi_iter.bi_sector,
6739 (unsigned long)dev->bdev->bd_dev, rcu_str_deref(dev->name),
6740 dev->devid, bio->bi_iter.bi_size);
6741 bio_set_dev(bio, dev->bdev);
6743 btrfs_bio_counter_inc_noblocked(fs_info);
6745 btrfsic_submit_bio(bio);
6748 static void bioc_error(struct btrfs_io_context *bioc, struct bio *bio, u64 logical)
6750 atomic_inc(&bioc->error);
6751 if (atomic_dec_and_test(&bioc->stripes_pending)) {
6752 /* Should be the original bio. */
6753 WARN_ON(bio != bioc->orig_bio);
6755 btrfs_bio(bio)->mirror_num = bioc->mirror_num;
6756 bio->bi_iter.bi_sector = logical >> 9;
6757 if (atomic_read(&bioc->error) > bioc->max_errors)
6758 bio->bi_status = BLK_STS_IOERR;
6760 bio->bi_status = BLK_STS_OK;
6761 btrfs_end_bioc(bioc, bio);
6765 blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
6768 struct btrfs_device *dev;
6769 struct bio *first_bio = bio;
6770 u64 logical = bio->bi_iter.bi_sector << 9;
6776 struct btrfs_io_context *bioc = NULL;
6778 length = bio->bi_iter.bi_size;
6779 map_length = length;
6781 btrfs_bio_counter_inc_blocked(fs_info);
6782 ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical,
6783 &map_length, &bioc, mirror_num, 1);
6785 btrfs_bio_counter_dec(fs_info);
6786 return errno_to_blk_status(ret);
6789 total_devs = bioc->num_stripes;
6790 bioc->orig_bio = first_bio;
6791 bioc->private = first_bio->bi_private;
6792 bioc->end_io = first_bio->bi_end_io;
6793 atomic_set(&bioc->stripes_pending, bioc->num_stripes);
6795 if ((bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6796 ((btrfs_op(bio) == BTRFS_MAP_WRITE) || (mirror_num > 1))) {
6797 /* In this case, map_length has been set to the length of
6798 a single stripe; not the whole write */
6799 if (btrfs_op(bio) == BTRFS_MAP_WRITE) {
6800 ret = raid56_parity_write(bio, bioc, map_length);
6802 ret = raid56_parity_recover(bio, bioc, map_length,
6806 btrfs_bio_counter_dec(fs_info);
6807 return errno_to_blk_status(ret);
6810 if (map_length < length) {
6812 "mapping failed logical %llu bio len %llu len %llu",
6813 logical, length, map_length);
6817 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6818 dev = bioc->stripes[dev_nr].dev;
6819 if (!dev || !dev->bdev || test_bit(BTRFS_DEV_STATE_MISSING,
6821 (btrfs_op(first_bio) == BTRFS_MAP_WRITE &&
6822 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) {
6823 bioc_error(bioc, first_bio, logical);
6827 if (dev_nr < total_devs - 1)
6828 bio = btrfs_bio_clone(first_bio);
6832 submit_stripe_bio(bioc, bio, bioc->stripes[dev_nr].physical, dev);
6834 btrfs_bio_counter_dec(fs_info);
6838 static bool dev_args_match_fs_devices(const struct btrfs_dev_lookup_args *args,
6839 const struct btrfs_fs_devices *fs_devices)
6841 if (args->fsid == NULL)
6843 if (memcmp(fs_devices->metadata_uuid, args->fsid, BTRFS_FSID_SIZE) == 0)
6848 static bool dev_args_match_device(const struct btrfs_dev_lookup_args *args,
6849 const struct btrfs_device *device)
6851 ASSERT((args->devid != (u64)-1) || args->missing);
6853 if ((args->devid != (u64)-1) && device->devid != args->devid)
6855 if (args->uuid && memcmp(device->uuid, args->uuid, BTRFS_UUID_SIZE) != 0)
6859 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state) &&
6866 * Find a device specified by @devid or @uuid in the list of @fs_devices, or
6869 * If devid and uuid are both specified, the match must be exact, otherwise
6870 * only devid is used.
6872 struct btrfs_device *btrfs_find_device(const struct btrfs_fs_devices *fs_devices,
6873 const struct btrfs_dev_lookup_args *args)
6875 struct btrfs_device *device;
6876 struct btrfs_fs_devices *seed_devs;
6878 if (dev_args_match_fs_devices(args, fs_devices)) {
6879 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6880 if (dev_args_match_device(args, device))
6885 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
6886 if (!dev_args_match_fs_devices(args, seed_devs))
6888 list_for_each_entry(device, &seed_devs->devices, dev_list) {
6889 if (dev_args_match_device(args, device))
6897 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices,
6898 u64 devid, u8 *dev_uuid)
6900 struct btrfs_device *device;
6901 unsigned int nofs_flag;
6904 * We call this under the chunk_mutex, so we want to use NOFS for this
6905 * allocation, however we don't want to change btrfs_alloc_device() to
6906 * always do NOFS because we use it in a lot of other GFP_KERNEL safe
6909 nofs_flag = memalloc_nofs_save();
6910 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6911 memalloc_nofs_restore(nofs_flag);
6915 list_add(&device->dev_list, &fs_devices->devices);
6916 device->fs_devices = fs_devices;
6917 fs_devices->num_devices++;
6919 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6920 fs_devices->missing_devices++;
6926 * btrfs_alloc_device - allocate struct btrfs_device
6927 * @fs_info: used only for generating a new devid, can be NULL if
6928 * devid is provided (i.e. @devid != NULL).
6929 * @devid: a pointer to devid for this device. If NULL a new devid
6931 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6934 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6935 * on error. Returned struct is not linked onto any lists and must be
6936 * destroyed with btrfs_free_device.
6938 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6942 struct btrfs_device *dev;
6945 if (WARN_ON(!devid && !fs_info))
6946 return ERR_PTR(-EINVAL);
6948 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
6950 return ERR_PTR(-ENOMEM);
6953 * Preallocate a bio that's always going to be used for flushing device
6954 * barriers and matches the device lifespan
6956 dev->flush_bio = bio_kmalloc(GFP_KERNEL, 0);
6957 if (!dev->flush_bio) {
6959 return ERR_PTR(-ENOMEM);
6962 INIT_LIST_HEAD(&dev->dev_list);
6963 INIT_LIST_HEAD(&dev->dev_alloc_list);
6964 INIT_LIST_HEAD(&dev->post_commit_list);
6966 atomic_set(&dev->dev_stats_ccnt, 0);
6967 btrfs_device_data_ordered_init(dev);
6968 extent_io_tree_init(fs_info, &dev->alloc_state,
6969 IO_TREE_DEVICE_ALLOC_STATE, NULL);
6976 ret = find_next_devid(fs_info, &tmp);
6978 btrfs_free_device(dev);
6979 return ERR_PTR(ret);
6985 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6987 generate_random_uuid(dev->uuid);
6992 static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info,
6993 u64 devid, u8 *uuid, bool error)
6996 btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing",
6999 btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing",
7003 static u64 calc_stripe_length(u64 type, u64 chunk_len, int num_stripes)
7005 const int data_stripes = calc_data_stripes(type, num_stripes);
7007 return div_u64(chunk_len, data_stripes);
7010 #if BITS_PER_LONG == 32
7012 * Due to page cache limit, metadata beyond BTRFS_32BIT_MAX_FILE_SIZE
7013 * can't be accessed on 32bit systems.
7015 * This function do mount time check to reject the fs if it already has
7016 * metadata chunk beyond that limit.
7018 static int check_32bit_meta_chunk(struct btrfs_fs_info *fs_info,
7019 u64 logical, u64 length, u64 type)
7021 if (!(type & BTRFS_BLOCK_GROUP_METADATA))
7024 if (logical + length < MAX_LFS_FILESIZE)
7027 btrfs_err_32bit_limit(fs_info);
7032 * This is to give early warning for any metadata chunk reaching
7033 * BTRFS_32BIT_EARLY_WARN_THRESHOLD.
7034 * Although we can still access the metadata, it's not going to be possible
7035 * once the limit is reached.
7037 static void warn_32bit_meta_chunk(struct btrfs_fs_info *fs_info,
7038 u64 logical, u64 length, u64 type)
7040 if (!(type & BTRFS_BLOCK_GROUP_METADATA))
7043 if (logical + length < BTRFS_32BIT_EARLY_WARN_THRESHOLD)
7046 btrfs_warn_32bit_limit(fs_info);
7050 static struct btrfs_device *handle_missing_device(struct btrfs_fs_info *fs_info,
7051 u64 devid, u8 *uuid)
7053 struct btrfs_device *dev;
7055 if (!btrfs_test_opt(fs_info, DEGRADED)) {
7056 btrfs_report_missing_device(fs_info, devid, uuid, true);
7057 return ERR_PTR(-ENOENT);
7060 dev = add_missing_dev(fs_info->fs_devices, devid, uuid);
7062 btrfs_err(fs_info, "failed to init missing device %llu: %ld",
7063 devid, PTR_ERR(dev));
7066 btrfs_report_missing_device(fs_info, devid, uuid, false);
7071 static int read_one_chunk(struct btrfs_key *key, struct extent_buffer *leaf,
7072 struct btrfs_chunk *chunk)
7074 BTRFS_DEV_LOOKUP_ARGS(args);
7075 struct btrfs_fs_info *fs_info = leaf->fs_info;
7076 struct extent_map_tree *map_tree = &fs_info->mapping_tree;
7077 struct map_lookup *map;
7078 struct extent_map *em;
7083 u8 uuid[BTRFS_UUID_SIZE];
7088 logical = key->offset;
7089 length = btrfs_chunk_length(leaf, chunk);
7090 type = btrfs_chunk_type(leaf, chunk);
7091 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
7093 #if BITS_PER_LONG == 32
7094 ret = check_32bit_meta_chunk(fs_info, logical, length, type);
7097 warn_32bit_meta_chunk(fs_info, logical, length, type);
7101 * Only need to verify chunk item if we're reading from sys chunk array,
7102 * as chunk item in tree block is already verified by tree-checker.
7104 if (leaf->start == BTRFS_SUPER_INFO_OFFSET) {
7105 ret = btrfs_check_chunk_valid(leaf, chunk, logical);
7110 read_lock(&map_tree->lock);
7111 em = lookup_extent_mapping(map_tree, logical, 1);
7112 read_unlock(&map_tree->lock);
7114 /* already mapped? */
7115 if (em && em->start <= logical && em->start + em->len > logical) {
7116 free_extent_map(em);
7119 free_extent_map(em);
7122 em = alloc_extent_map();
7125 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
7127 free_extent_map(em);
7131 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
7132 em->map_lookup = map;
7133 em->start = logical;
7136 em->block_start = 0;
7137 em->block_len = em->len;
7139 map->num_stripes = num_stripes;
7140 map->io_width = btrfs_chunk_io_width(leaf, chunk);
7141 map->io_align = btrfs_chunk_io_align(leaf, chunk);
7142 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
7144 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
7145 map->verified_stripes = 0;
7146 em->orig_block_len = calc_stripe_length(type, em->len,
7148 for (i = 0; i < num_stripes; i++) {
7149 map->stripes[i].physical =
7150 btrfs_stripe_offset_nr(leaf, chunk, i);
7151 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
7153 read_extent_buffer(leaf, uuid, (unsigned long)
7154 btrfs_stripe_dev_uuid_nr(chunk, i),
7157 map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices, &args);
7158 if (!map->stripes[i].dev) {
7159 map->stripes[i].dev = handle_missing_device(fs_info,
7161 if (IS_ERR(map->stripes[i].dev)) {
7162 free_extent_map(em);
7163 return PTR_ERR(map->stripes[i].dev);
7167 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
7168 &(map->stripes[i].dev->dev_state));
7171 write_lock(&map_tree->lock);
7172 ret = add_extent_mapping(map_tree, em, 0);
7173 write_unlock(&map_tree->lock);
7176 "failed to add chunk map, start=%llu len=%llu: %d",
7177 em->start, em->len, ret);
7179 free_extent_map(em);
7184 static void fill_device_from_item(struct extent_buffer *leaf,
7185 struct btrfs_dev_item *dev_item,
7186 struct btrfs_device *device)
7190 device->devid = btrfs_device_id(leaf, dev_item);
7191 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
7192 device->total_bytes = device->disk_total_bytes;
7193 device->commit_total_bytes = device->disk_total_bytes;
7194 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
7195 device->commit_bytes_used = device->bytes_used;
7196 device->type = btrfs_device_type(leaf, dev_item);
7197 device->io_align = btrfs_device_io_align(leaf, dev_item);
7198 device->io_width = btrfs_device_io_width(leaf, dev_item);
7199 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
7200 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
7201 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
7203 ptr = btrfs_device_uuid(dev_item);
7204 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
7207 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info,
7210 struct btrfs_fs_devices *fs_devices;
7213 lockdep_assert_held(&uuid_mutex);
7216 /* This will match only for multi-device seed fs */
7217 list_for_each_entry(fs_devices, &fs_info->fs_devices->seed_list, seed_list)
7218 if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE))
7222 fs_devices = find_fsid(fsid, NULL);
7224 if (!btrfs_test_opt(fs_info, DEGRADED))
7225 return ERR_PTR(-ENOENT);
7227 fs_devices = alloc_fs_devices(fsid, NULL);
7228 if (IS_ERR(fs_devices))
7231 fs_devices->seeding = true;
7232 fs_devices->opened = 1;
7237 * Upon first call for a seed fs fsid, just create a private copy of the
7238 * respective fs_devices and anchor it at fs_info->fs_devices->seed_list
7240 fs_devices = clone_fs_devices(fs_devices);
7241 if (IS_ERR(fs_devices))
7244 ret = open_fs_devices(fs_devices, FMODE_READ, fs_info->bdev_holder);
7246 free_fs_devices(fs_devices);
7247 return ERR_PTR(ret);
7250 if (!fs_devices->seeding) {
7251 close_fs_devices(fs_devices);
7252 free_fs_devices(fs_devices);
7253 return ERR_PTR(-EINVAL);
7256 list_add(&fs_devices->seed_list, &fs_info->fs_devices->seed_list);
7261 static int read_one_dev(struct extent_buffer *leaf,
7262 struct btrfs_dev_item *dev_item)
7264 BTRFS_DEV_LOOKUP_ARGS(args);
7265 struct btrfs_fs_info *fs_info = leaf->fs_info;
7266 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7267 struct btrfs_device *device;
7270 u8 fs_uuid[BTRFS_FSID_SIZE];
7271 u8 dev_uuid[BTRFS_UUID_SIZE];
7273 devid = args.devid = btrfs_device_id(leaf, dev_item);
7274 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
7276 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
7278 args.uuid = dev_uuid;
7279 args.fsid = fs_uuid;
7281 if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) {
7282 fs_devices = open_seed_devices(fs_info, fs_uuid);
7283 if (IS_ERR(fs_devices))
7284 return PTR_ERR(fs_devices);
7287 device = btrfs_find_device(fs_info->fs_devices, &args);
7289 if (!btrfs_test_opt(fs_info, DEGRADED)) {
7290 btrfs_report_missing_device(fs_info, devid,
7295 device = add_missing_dev(fs_devices, devid, dev_uuid);
7296 if (IS_ERR(device)) {
7298 "failed to add missing dev %llu: %ld",
7299 devid, PTR_ERR(device));
7300 return PTR_ERR(device);
7302 btrfs_report_missing_device(fs_info, devid, dev_uuid, false);
7304 if (!device->bdev) {
7305 if (!btrfs_test_opt(fs_info, DEGRADED)) {
7306 btrfs_report_missing_device(fs_info,
7307 devid, dev_uuid, true);
7310 btrfs_report_missing_device(fs_info, devid,
7314 if (!device->bdev &&
7315 !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
7317 * this happens when a device that was properly setup
7318 * in the device info lists suddenly goes bad.
7319 * device->bdev is NULL, and so we have to set
7320 * device->missing to one here
7322 device->fs_devices->missing_devices++;
7323 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
7326 /* Move the device to its own fs_devices */
7327 if (device->fs_devices != fs_devices) {
7328 ASSERT(test_bit(BTRFS_DEV_STATE_MISSING,
7329 &device->dev_state));
7331 list_move(&device->dev_list, &fs_devices->devices);
7332 device->fs_devices->num_devices--;
7333 fs_devices->num_devices++;
7335 device->fs_devices->missing_devices--;
7336 fs_devices->missing_devices++;
7338 device->fs_devices = fs_devices;
7342 if (device->fs_devices != fs_info->fs_devices) {
7343 BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state));
7344 if (device->generation !=
7345 btrfs_device_generation(leaf, dev_item))
7349 fill_device_from_item(leaf, dev_item, device);
7351 u64 max_total_bytes = bdev_nr_bytes(device->bdev);
7353 if (device->total_bytes > max_total_bytes) {
7355 "device total_bytes should be at most %llu but found %llu",
7356 max_total_bytes, device->total_bytes);
7360 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
7361 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
7362 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
7363 device->fs_devices->total_rw_bytes += device->total_bytes;
7364 atomic64_add(device->total_bytes - device->bytes_used,
7365 &fs_info->free_chunk_space);
7371 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
7373 struct btrfs_root *root = fs_info->tree_root;
7374 struct btrfs_super_block *super_copy = fs_info->super_copy;
7375 struct extent_buffer *sb;
7376 struct btrfs_disk_key *disk_key;
7377 struct btrfs_chunk *chunk;
7379 unsigned long sb_array_offset;
7386 struct btrfs_key key;
7388 ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize);
7390 * This will create extent buffer of nodesize, superblock size is
7391 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
7392 * overallocate but we can keep it as-is, only the first page is used.
7394 sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET,
7395 root->root_key.objectid, 0);
7398 set_extent_buffer_uptodate(sb);
7400 * The sb extent buffer is artificial and just used to read the system array.
7401 * set_extent_buffer_uptodate() call does not properly mark all it's
7402 * pages up-to-date when the page is larger: extent does not cover the
7403 * whole page and consequently check_page_uptodate does not find all
7404 * the page's extents up-to-date (the hole beyond sb),
7405 * write_extent_buffer then triggers a WARN_ON.
7407 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
7408 * but sb spans only this function. Add an explicit SetPageUptodate call
7409 * to silence the warning eg. on PowerPC 64.
7411 if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
7412 SetPageUptodate(sb->pages[0]);
7414 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
7415 array_size = btrfs_super_sys_array_size(super_copy);
7417 array_ptr = super_copy->sys_chunk_array;
7418 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
7421 while (cur_offset < array_size) {
7422 disk_key = (struct btrfs_disk_key *)array_ptr;
7423 len = sizeof(*disk_key);
7424 if (cur_offset + len > array_size)
7425 goto out_short_read;
7427 btrfs_disk_key_to_cpu(&key, disk_key);
7430 sb_array_offset += len;
7433 if (key.type != BTRFS_CHUNK_ITEM_KEY) {
7435 "unexpected item type %u in sys_array at offset %u",
7436 (u32)key.type, cur_offset);
7441 chunk = (struct btrfs_chunk *)sb_array_offset;
7443 * At least one btrfs_chunk with one stripe must be present,
7444 * exact stripe count check comes afterwards
7446 len = btrfs_chunk_item_size(1);
7447 if (cur_offset + len > array_size)
7448 goto out_short_read;
7450 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
7453 "invalid number of stripes %u in sys_array at offset %u",
7454 num_stripes, cur_offset);
7459 type = btrfs_chunk_type(sb, chunk);
7460 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
7462 "invalid chunk type %llu in sys_array at offset %u",
7468 len = btrfs_chunk_item_size(num_stripes);
7469 if (cur_offset + len > array_size)
7470 goto out_short_read;
7472 ret = read_one_chunk(&key, sb, chunk);
7477 sb_array_offset += len;
7480 clear_extent_buffer_uptodate(sb);
7481 free_extent_buffer_stale(sb);
7485 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
7487 clear_extent_buffer_uptodate(sb);
7488 free_extent_buffer_stale(sb);
7493 * Check if all chunks in the fs are OK for read-write degraded mount
7495 * If the @failing_dev is specified, it's accounted as missing.
7497 * Return true if all chunks meet the minimal RW mount requirements.
7498 * Return false if any chunk doesn't meet the minimal RW mount requirements.
7500 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
7501 struct btrfs_device *failing_dev)
7503 struct extent_map_tree *map_tree = &fs_info->mapping_tree;
7504 struct extent_map *em;
7508 read_lock(&map_tree->lock);
7509 em = lookup_extent_mapping(map_tree, 0, (u64)-1);
7510 read_unlock(&map_tree->lock);
7511 /* No chunk at all? Return false anyway */
7517 struct map_lookup *map;
7522 map = em->map_lookup;
7524 btrfs_get_num_tolerated_disk_barrier_failures(
7526 for (i = 0; i < map->num_stripes; i++) {
7527 struct btrfs_device *dev = map->stripes[i].dev;
7529 if (!dev || !dev->bdev ||
7530 test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) ||
7531 dev->last_flush_error)
7533 else if (failing_dev && failing_dev == dev)
7536 if (missing > max_tolerated) {
7539 "chunk %llu missing %d devices, max tolerance is %d for writable mount",
7540 em->start, missing, max_tolerated);
7541 free_extent_map(em);
7545 next_start = extent_map_end(em);
7546 free_extent_map(em);
7548 read_lock(&map_tree->lock);
7549 em = lookup_extent_mapping(map_tree, next_start,
7550 (u64)(-1) - next_start);
7551 read_unlock(&map_tree->lock);
7557 static void readahead_tree_node_children(struct extent_buffer *node)
7560 const int nr_items = btrfs_header_nritems(node);
7562 for (i = 0; i < nr_items; i++)
7563 btrfs_readahead_node_child(node, i);
7566 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
7568 struct btrfs_root *root = fs_info->chunk_root;
7569 struct btrfs_path *path;
7570 struct extent_buffer *leaf;
7571 struct btrfs_key key;
7572 struct btrfs_key found_key;
7576 u64 last_ra_node = 0;
7578 path = btrfs_alloc_path();
7583 * uuid_mutex is needed only if we are mounting a sprout FS
7584 * otherwise we don't need it.
7586 mutex_lock(&uuid_mutex);
7589 * It is possible for mount and umount to race in such a way that
7590 * we execute this code path, but open_fs_devices failed to clear
7591 * total_rw_bytes. We certainly want it cleared before reading the
7592 * device items, so clear it here.
7594 fs_info->fs_devices->total_rw_bytes = 0;
7597 * Lockdep complains about possible circular locking dependency between
7598 * a disk's open_mutex (struct gendisk.open_mutex), the rw semaphores
7599 * used for freeze procection of a fs (struct super_block.s_writers),
7600 * which we take when starting a transaction, and extent buffers of the
7601 * chunk tree if we call read_one_dev() while holding a lock on an
7602 * extent buffer of the chunk tree. Since we are mounting the filesystem
7603 * and at this point there can't be any concurrent task modifying the
7604 * chunk tree, to keep it simple, just skip locking on the chunk tree.
7606 ASSERT(!test_bit(BTRFS_FS_OPEN, &fs_info->flags));
7607 path->skip_locking = 1;
7610 * Read all device items, and then all the chunk items. All
7611 * device items are found before any chunk item (their object id
7612 * is smaller than the lowest possible object id for a chunk
7613 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
7615 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
7618 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7622 struct extent_buffer *node;
7624 leaf = path->nodes[0];
7625 slot = path->slots[0];
7626 if (slot >= btrfs_header_nritems(leaf)) {
7627 ret = btrfs_next_leaf(root, path);
7634 node = path->nodes[1];
7636 if (last_ra_node != node->start) {
7637 readahead_tree_node_children(node);
7638 last_ra_node = node->start;
7641 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7642 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
7643 struct btrfs_dev_item *dev_item;
7644 dev_item = btrfs_item_ptr(leaf, slot,
7645 struct btrfs_dev_item);
7646 ret = read_one_dev(leaf, dev_item);
7650 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
7651 struct btrfs_chunk *chunk;
7654 * We are only called at mount time, so no need to take
7655 * fs_info->chunk_mutex. Plus, to avoid lockdep warnings,
7656 * we always lock first fs_info->chunk_mutex before
7657 * acquiring any locks on the chunk tree. This is a
7658 * requirement for chunk allocation, see the comment on
7659 * top of btrfs_chunk_alloc() for details.
7661 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
7662 ret = read_one_chunk(&found_key, leaf, chunk);
7670 * After loading chunk tree, we've got all device information,
7671 * do another round of validation checks.
7673 if (total_dev != fs_info->fs_devices->total_devices) {
7675 "super_num_devices %llu mismatch with num_devices %llu found here",
7676 btrfs_super_num_devices(fs_info->super_copy),
7681 if (btrfs_super_total_bytes(fs_info->super_copy) <
7682 fs_info->fs_devices->total_rw_bytes) {
7684 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
7685 btrfs_super_total_bytes(fs_info->super_copy),
7686 fs_info->fs_devices->total_rw_bytes);
7692 mutex_unlock(&uuid_mutex);
7694 btrfs_free_path(path);
7698 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
7700 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
7701 struct btrfs_device *device;
7703 fs_devices->fs_info = fs_info;
7705 mutex_lock(&fs_devices->device_list_mutex);
7706 list_for_each_entry(device, &fs_devices->devices, dev_list)
7707 device->fs_info = fs_info;
7709 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
7710 list_for_each_entry(device, &seed_devs->devices, dev_list)
7711 device->fs_info = fs_info;
7713 seed_devs->fs_info = fs_info;
7715 mutex_unlock(&fs_devices->device_list_mutex);
7718 static u64 btrfs_dev_stats_value(const struct extent_buffer *eb,
7719 const struct btrfs_dev_stats_item *ptr,
7724 read_extent_buffer(eb, &val,
7725 offsetof(struct btrfs_dev_stats_item, values) +
7726 ((unsigned long)ptr) + (index * sizeof(u64)),
7731 static void btrfs_set_dev_stats_value(struct extent_buffer *eb,
7732 struct btrfs_dev_stats_item *ptr,
7735 write_extent_buffer(eb, &val,
7736 offsetof(struct btrfs_dev_stats_item, values) +
7737 ((unsigned long)ptr) + (index * sizeof(u64)),
7741 static int btrfs_device_init_dev_stats(struct btrfs_device *device,
7742 struct btrfs_path *path)
7744 struct btrfs_dev_stats_item *ptr;
7745 struct extent_buffer *eb;
7746 struct btrfs_key key;
7750 if (!device->fs_info->dev_root)
7753 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7754 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7755 key.offset = device->devid;
7756 ret = btrfs_search_slot(NULL, device->fs_info->dev_root, &key, path, 0, 0);
7758 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7759 btrfs_dev_stat_set(device, i, 0);
7760 device->dev_stats_valid = 1;
7761 btrfs_release_path(path);
7762 return ret < 0 ? ret : 0;
7764 slot = path->slots[0];
7765 eb = path->nodes[0];
7766 item_size = btrfs_item_size(eb, slot);
7768 ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_stats_item);
7770 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7771 if (item_size >= (1 + i) * sizeof(__le64))
7772 btrfs_dev_stat_set(device, i,
7773 btrfs_dev_stats_value(eb, ptr, i));
7775 btrfs_dev_stat_set(device, i, 0);
7778 device->dev_stats_valid = 1;
7779 btrfs_dev_stat_print_on_load(device);
7780 btrfs_release_path(path);
7785 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
7787 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
7788 struct btrfs_device *device;
7789 struct btrfs_path *path = NULL;
7792 path = btrfs_alloc_path();
7796 mutex_lock(&fs_devices->device_list_mutex);
7797 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7798 ret = btrfs_device_init_dev_stats(device, path);
7802 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
7803 list_for_each_entry(device, &seed_devs->devices, dev_list) {
7804 ret = btrfs_device_init_dev_stats(device, path);
7810 mutex_unlock(&fs_devices->device_list_mutex);
7812 btrfs_free_path(path);
7816 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
7817 struct btrfs_device *device)
7819 struct btrfs_fs_info *fs_info = trans->fs_info;
7820 struct btrfs_root *dev_root = fs_info->dev_root;
7821 struct btrfs_path *path;
7822 struct btrfs_key key;
7823 struct extent_buffer *eb;
7824 struct btrfs_dev_stats_item *ptr;
7828 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7829 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7830 key.offset = device->devid;
7832 path = btrfs_alloc_path();
7835 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
7837 btrfs_warn_in_rcu(fs_info,
7838 "error %d while searching for dev_stats item for device %s",
7839 ret, rcu_str_deref(device->name));
7844 btrfs_item_size(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
7845 /* need to delete old one and insert a new one */
7846 ret = btrfs_del_item(trans, dev_root, path);
7848 btrfs_warn_in_rcu(fs_info,
7849 "delete too small dev_stats item for device %s failed %d",
7850 rcu_str_deref(device->name), ret);
7857 /* need to insert a new item */
7858 btrfs_release_path(path);
7859 ret = btrfs_insert_empty_item(trans, dev_root, path,
7860 &key, sizeof(*ptr));
7862 btrfs_warn_in_rcu(fs_info,
7863 "insert dev_stats item for device %s failed %d",
7864 rcu_str_deref(device->name), ret);
7869 eb = path->nodes[0];
7870 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7871 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7872 btrfs_set_dev_stats_value(eb, ptr, i,
7873 btrfs_dev_stat_read(device, i));
7874 btrfs_mark_buffer_dirty(eb);
7877 btrfs_free_path(path);
7882 * called from commit_transaction. Writes all changed device stats to disk.
7884 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans)
7886 struct btrfs_fs_info *fs_info = trans->fs_info;
7887 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7888 struct btrfs_device *device;
7892 mutex_lock(&fs_devices->device_list_mutex);
7893 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7894 stats_cnt = atomic_read(&device->dev_stats_ccnt);
7895 if (!device->dev_stats_valid || stats_cnt == 0)
7900 * There is a LOAD-LOAD control dependency between the value of
7901 * dev_stats_ccnt and updating the on-disk values which requires
7902 * reading the in-memory counters. Such control dependencies
7903 * require explicit read memory barriers.
7905 * This memory barriers pairs with smp_mb__before_atomic in
7906 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full
7907 * barrier implied by atomic_xchg in
7908 * btrfs_dev_stats_read_and_reset
7912 ret = update_dev_stat_item(trans, device);
7914 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7916 mutex_unlock(&fs_devices->device_list_mutex);
7921 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7923 btrfs_dev_stat_inc(dev, index);
7924 btrfs_dev_stat_print_on_error(dev);
7927 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
7929 if (!dev->dev_stats_valid)
7931 btrfs_err_rl_in_rcu(dev->fs_info,
7932 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7933 rcu_str_deref(dev->name),
7934 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7935 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7936 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7937 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7938 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7941 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7945 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7946 if (btrfs_dev_stat_read(dev, i) != 0)
7948 if (i == BTRFS_DEV_STAT_VALUES_MAX)
7949 return; /* all values == 0, suppress message */
7951 btrfs_info_in_rcu(dev->fs_info,
7952 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7953 rcu_str_deref(dev->name),
7954 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7955 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7956 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7957 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7958 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7961 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
7962 struct btrfs_ioctl_get_dev_stats *stats)
7964 BTRFS_DEV_LOOKUP_ARGS(args);
7965 struct btrfs_device *dev;
7966 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7969 mutex_lock(&fs_devices->device_list_mutex);
7970 args.devid = stats->devid;
7971 dev = btrfs_find_device(fs_info->fs_devices, &args);
7972 mutex_unlock(&fs_devices->device_list_mutex);
7975 btrfs_warn(fs_info, "get dev_stats failed, device not found");
7977 } else if (!dev->dev_stats_valid) {
7978 btrfs_warn(fs_info, "get dev_stats failed, not yet valid");
7980 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7981 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7982 if (stats->nr_items > i)
7984 btrfs_dev_stat_read_and_reset(dev, i);
7986 btrfs_dev_stat_set(dev, i, 0);
7988 btrfs_info(fs_info, "device stats zeroed by %s (%d)",
7989 current->comm, task_pid_nr(current));
7991 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7992 if (stats->nr_items > i)
7993 stats->values[i] = btrfs_dev_stat_read(dev, i);
7995 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7996 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
8001 * Update the size and bytes used for each device where it changed. This is
8002 * delayed since we would otherwise get errors while writing out the
8005 * Must be invoked during transaction commit.
8007 void btrfs_commit_device_sizes(struct btrfs_transaction *trans)
8009 struct btrfs_device *curr, *next;
8011 ASSERT(trans->state == TRANS_STATE_COMMIT_DOING);
8013 if (list_empty(&trans->dev_update_list))
8017 * We don't need the device_list_mutex here. This list is owned by the
8018 * transaction and the transaction must complete before the device is
8021 mutex_lock(&trans->fs_info->chunk_mutex);
8022 list_for_each_entry_safe(curr, next, &trans->dev_update_list,
8024 list_del_init(&curr->post_commit_list);
8025 curr->commit_total_bytes = curr->disk_total_bytes;
8026 curr->commit_bytes_used = curr->bytes_used;
8028 mutex_unlock(&trans->fs_info->chunk_mutex);
8032 * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10.
8034 int btrfs_bg_type_to_factor(u64 flags)
8036 const int index = btrfs_bg_flags_to_raid_index(flags);
8038 return btrfs_raid_array[index].ncopies;
8043 static int verify_one_dev_extent(struct btrfs_fs_info *fs_info,
8044 u64 chunk_offset, u64 devid,
8045 u64 physical_offset, u64 physical_len)
8047 struct btrfs_dev_lookup_args args = { .devid = devid };
8048 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
8049 struct extent_map *em;
8050 struct map_lookup *map;
8051 struct btrfs_device *dev;
8057 read_lock(&em_tree->lock);
8058 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
8059 read_unlock(&em_tree->lock);
8063 "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk",
8064 physical_offset, devid);
8069 map = em->map_lookup;
8070 stripe_len = calc_stripe_length(map->type, em->len, map->num_stripes);
8071 if (physical_len != stripe_len) {
8073 "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu",
8074 physical_offset, devid, em->start, physical_len,
8080 for (i = 0; i < map->num_stripes; i++) {
8081 if (map->stripes[i].dev->devid == devid &&
8082 map->stripes[i].physical == physical_offset) {
8084 if (map->verified_stripes >= map->num_stripes) {
8086 "too many dev extents for chunk %llu found",
8091 map->verified_stripes++;
8097 "dev extent physical offset %llu devid %llu has no corresponding chunk",
8098 physical_offset, devid);
8102 /* Make sure no dev extent is beyond device boundary */
8103 dev = btrfs_find_device(fs_info->fs_devices, &args);
8105 btrfs_err(fs_info, "failed to find devid %llu", devid);
8110 if (physical_offset + physical_len > dev->disk_total_bytes) {
8112 "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu",
8113 devid, physical_offset, physical_len,
8114 dev->disk_total_bytes);
8119 if (dev->zone_info) {
8120 u64 zone_size = dev->zone_info->zone_size;
8122 if (!IS_ALIGNED(physical_offset, zone_size) ||
8123 !IS_ALIGNED(physical_len, zone_size)) {
8125 "zoned: dev extent devid %llu physical offset %llu len %llu is not aligned to device zone",
8126 devid, physical_offset, physical_len);
8133 free_extent_map(em);
8137 static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info)
8139 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
8140 struct extent_map *em;
8141 struct rb_node *node;
8144 read_lock(&em_tree->lock);
8145 for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) {
8146 em = rb_entry(node, struct extent_map, rb_node);
8147 if (em->map_lookup->num_stripes !=
8148 em->map_lookup->verified_stripes) {
8150 "chunk %llu has missing dev extent, have %d expect %d",
8151 em->start, em->map_lookup->verified_stripes,
8152 em->map_lookup->num_stripes);
8158 read_unlock(&em_tree->lock);
8163 * Ensure that all dev extents are mapped to correct chunk, otherwise
8164 * later chunk allocation/free would cause unexpected behavior.
8166 * NOTE: This will iterate through the whole device tree, which should be of
8167 * the same size level as the chunk tree. This slightly increases mount time.
8169 int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info)
8171 struct btrfs_path *path;
8172 struct btrfs_root *root = fs_info->dev_root;
8173 struct btrfs_key key;
8175 u64 prev_dev_ext_end = 0;
8179 * We don't have a dev_root because we mounted with ignorebadroots and
8180 * failed to load the root, so we want to skip the verification in this
8183 * However if the dev root is fine, but the tree itself is corrupted
8184 * we'd still fail to mount. This verification is only to make sure
8185 * writes can happen safely, so instead just bypass this check
8186 * completely in the case of IGNOREBADROOTS.
8188 if (btrfs_test_opt(fs_info, IGNOREBADROOTS))
8192 key.type = BTRFS_DEV_EXTENT_KEY;
8195 path = btrfs_alloc_path();
8199 path->reada = READA_FORWARD;
8200 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
8204 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
8205 ret = btrfs_next_leaf(root, path);
8208 /* No dev extents at all? Not good */
8215 struct extent_buffer *leaf = path->nodes[0];
8216 struct btrfs_dev_extent *dext;
8217 int slot = path->slots[0];
8219 u64 physical_offset;
8223 btrfs_item_key_to_cpu(leaf, &key, slot);
8224 if (key.type != BTRFS_DEV_EXTENT_KEY)
8226 devid = key.objectid;
8227 physical_offset = key.offset;
8229 dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent);
8230 chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext);
8231 physical_len = btrfs_dev_extent_length(leaf, dext);
8233 /* Check if this dev extent overlaps with the previous one */
8234 if (devid == prev_devid && physical_offset < prev_dev_ext_end) {
8236 "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu",
8237 devid, physical_offset, prev_dev_ext_end);
8242 ret = verify_one_dev_extent(fs_info, chunk_offset, devid,
8243 physical_offset, physical_len);
8247 prev_dev_ext_end = physical_offset + physical_len;
8249 ret = btrfs_next_item(root, path);
8258 /* Ensure all chunks have corresponding dev extents */
8259 ret = verify_chunk_dev_extent_mapping(fs_info);
8261 btrfs_free_path(path);
8266 * Check whether the given block group or device is pinned by any inode being
8267 * used as a swapfile.
8269 bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr)
8271 struct btrfs_swapfile_pin *sp;
8272 struct rb_node *node;
8274 spin_lock(&fs_info->swapfile_pins_lock);
8275 node = fs_info->swapfile_pins.rb_node;
8277 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
8279 node = node->rb_left;
8280 else if (ptr > sp->ptr)
8281 node = node->rb_right;
8285 spin_unlock(&fs_info->swapfile_pins_lock);
8286 return node != NULL;
8289 static int relocating_repair_kthread(void *data)
8291 struct btrfs_block_group *cache = (struct btrfs_block_group *)data;
8292 struct btrfs_fs_info *fs_info = cache->fs_info;
8296 target = cache->start;
8297 btrfs_put_block_group(cache);
8299 sb_start_write(fs_info->sb);
8300 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) {
8302 "zoned: skip relocating block group %llu to repair: EBUSY",
8304 sb_end_write(fs_info->sb);
8308 mutex_lock(&fs_info->reclaim_bgs_lock);
8310 /* Ensure block group still exists */
8311 cache = btrfs_lookup_block_group(fs_info, target);
8315 if (!cache->relocating_repair)
8318 ret = btrfs_may_alloc_data_chunk(fs_info, target);
8323 "zoned: relocating block group %llu to repair IO failure",
8325 ret = btrfs_relocate_chunk(fs_info, target);
8329 btrfs_put_block_group(cache);
8330 mutex_unlock(&fs_info->reclaim_bgs_lock);
8331 btrfs_exclop_finish(fs_info);
8332 sb_end_write(fs_info->sb);
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");