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_device *device)
1901 struct btrfs_root *root = device->fs_info->chunk_root;
1903 struct btrfs_path *path;
1904 struct btrfs_key key;
1905 struct btrfs_trans_handle *trans;
1907 path = btrfs_alloc_path();
1911 trans = btrfs_start_transaction(root, 0);
1912 if (IS_ERR(trans)) {
1913 btrfs_free_path(path);
1914 return PTR_ERR(trans);
1916 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1917 key.type = BTRFS_DEV_ITEM_KEY;
1918 key.offset = device->devid;
1920 btrfs_reserve_chunk_metadata(trans, false);
1921 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1922 btrfs_trans_release_chunk_metadata(trans);
1926 btrfs_abort_transaction(trans, ret);
1927 btrfs_end_transaction(trans);
1931 ret = btrfs_del_item(trans, root, path);
1933 btrfs_abort_transaction(trans, ret);
1934 btrfs_end_transaction(trans);
1938 btrfs_free_path(path);
1940 ret = btrfs_commit_transaction(trans);
1945 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1946 * filesystem. It's up to the caller to adjust that number regarding eg. device
1949 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1957 seq = read_seqbegin(&fs_info->profiles_lock);
1959 all_avail = fs_info->avail_data_alloc_bits |
1960 fs_info->avail_system_alloc_bits |
1961 fs_info->avail_metadata_alloc_bits;
1962 } while (read_seqretry(&fs_info->profiles_lock, seq));
1964 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1965 if (!(all_avail & btrfs_raid_array[i].bg_flag))
1968 if (num_devices < btrfs_raid_array[i].devs_min)
1969 return btrfs_raid_array[i].mindev_error;
1975 static struct btrfs_device * btrfs_find_next_active_device(
1976 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device)
1978 struct btrfs_device *next_device;
1980 list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1981 if (next_device != device &&
1982 !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state)
1983 && next_device->bdev)
1991 * Helper function to check if the given device is part of s_bdev / latest_dev
1992 * and replace it with the provided or the next active device, in the context
1993 * where this function called, there should be always be another device (or
1994 * this_dev) which is active.
1996 void __cold btrfs_assign_next_active_device(struct btrfs_device *device,
1997 struct btrfs_device *next_device)
1999 struct btrfs_fs_info *fs_info = device->fs_info;
2002 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
2004 ASSERT(next_device);
2006 if (fs_info->sb->s_bdev &&
2007 (fs_info->sb->s_bdev == device->bdev))
2008 fs_info->sb->s_bdev = next_device->bdev;
2010 if (fs_info->fs_devices->latest_dev->bdev == device->bdev)
2011 fs_info->fs_devices->latest_dev = next_device;
2015 * Return btrfs_fs_devices::num_devices excluding the device that's being
2016 * currently replaced.
2018 static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info)
2020 u64 num_devices = fs_info->fs_devices->num_devices;
2022 down_read(&fs_info->dev_replace.rwsem);
2023 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
2024 ASSERT(num_devices > 1);
2027 up_read(&fs_info->dev_replace.rwsem);
2032 void btrfs_scratch_superblocks(struct btrfs_fs_info *fs_info,
2033 struct block_device *bdev,
2034 const char *device_path)
2036 struct btrfs_super_block *disk_super;
2042 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; copy_num++) {
2046 disk_super = btrfs_read_dev_one_super(bdev, copy_num);
2047 if (IS_ERR(disk_super))
2050 if (bdev_is_zoned(bdev)) {
2051 btrfs_reset_sb_log_zones(bdev, copy_num);
2055 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
2057 page = virt_to_page(disk_super);
2058 set_page_dirty(page);
2060 /* write_on_page() unlocks the page */
2061 ret = write_one_page(page);
2064 "error clearing superblock number %d (%d)",
2066 btrfs_release_disk_super(disk_super);
2070 /* Notify udev that device has changed */
2071 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
2073 /* Update ctime/mtime for device path for libblkid */
2074 update_dev_time(device_path);
2077 int btrfs_rm_device(struct btrfs_fs_info *fs_info,
2078 struct btrfs_dev_lookup_args *args,
2079 struct block_device **bdev, fmode_t *mode)
2081 struct btrfs_device *device;
2082 struct btrfs_fs_devices *cur_devices;
2083 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2087 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
2088 btrfs_err(fs_info, "device remove not supported on extent tree v2 yet");
2093 * The device list in fs_devices is accessed without locks (neither
2094 * uuid_mutex nor device_list_mutex) as it won't change on a mounted
2095 * filesystem and another device rm cannot run.
2097 num_devices = btrfs_num_devices(fs_info);
2099 ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
2103 device = btrfs_find_device(fs_info->fs_devices, args);
2106 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2112 if (btrfs_pinned_by_swapfile(fs_info, device)) {
2113 btrfs_warn_in_rcu(fs_info,
2114 "cannot remove device %s (devid %llu) due to active swapfile",
2115 rcu_str_deref(device->name), device->devid);
2120 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2121 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
2125 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
2126 fs_info->fs_devices->rw_devices == 1) {
2127 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
2131 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2132 mutex_lock(&fs_info->chunk_mutex);
2133 list_del_init(&device->dev_alloc_list);
2134 device->fs_devices->rw_devices--;
2135 mutex_unlock(&fs_info->chunk_mutex);
2138 ret = btrfs_shrink_device(device, 0);
2143 * TODO: the superblock still includes this device in its num_devices
2144 * counter although write_all_supers() is not locked out. This
2145 * could give a filesystem state which requires a degraded mount.
2147 ret = btrfs_rm_dev_item(device);
2151 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2152 btrfs_scrub_cancel_dev(device);
2155 * the device list mutex makes sure that we don't change
2156 * the device list while someone else is writing out all
2157 * the device supers. Whoever is writing all supers, should
2158 * lock the device list mutex before getting the number of
2159 * devices in the super block (super_copy). Conversely,
2160 * whoever updates the number of devices in the super block
2161 * (super_copy) should hold the device list mutex.
2165 * In normal cases the cur_devices == fs_devices. But in case
2166 * of deleting a seed device, the cur_devices should point to
2167 * its own fs_devices listed under the fs_devices->seed_list.
2169 cur_devices = device->fs_devices;
2170 mutex_lock(&fs_devices->device_list_mutex);
2171 list_del_rcu(&device->dev_list);
2173 cur_devices->num_devices--;
2174 cur_devices->total_devices--;
2175 /* Update total_devices of the parent fs_devices if it's seed */
2176 if (cur_devices != fs_devices)
2177 fs_devices->total_devices--;
2179 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
2180 cur_devices->missing_devices--;
2182 btrfs_assign_next_active_device(device, NULL);
2185 cur_devices->open_devices--;
2186 /* remove sysfs entry */
2187 btrfs_sysfs_remove_device(device);
2190 num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
2191 btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
2192 mutex_unlock(&fs_devices->device_list_mutex);
2195 * At this point, the device is zero sized and detached from the
2196 * devices list. All that's left is to zero out the old supers and
2199 * We cannot call btrfs_close_bdev() here because we're holding the sb
2200 * write lock, and blkdev_put() will pull in the ->open_mutex on the
2201 * block device and it's dependencies. Instead just flush the device
2202 * and let the caller do the final blkdev_put.
2204 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2205 btrfs_scratch_superblocks(fs_info, device->bdev,
2208 sync_blockdev(device->bdev);
2209 invalidate_bdev(device->bdev);
2213 *bdev = device->bdev;
2214 *mode = device->mode;
2216 btrfs_free_device(device);
2219 * This can happen if cur_devices is the private seed devices list. We
2220 * cannot call close_fs_devices() here because it expects the uuid_mutex
2221 * to be held, but in fact we don't need that for the private
2222 * seed_devices, we can simply decrement cur_devices->opened and then
2223 * remove it from our list and free the fs_devices.
2225 if (cur_devices->num_devices == 0) {
2226 list_del_init(&cur_devices->seed_list);
2227 ASSERT(cur_devices->opened == 1);
2228 cur_devices->opened--;
2229 free_fs_devices(cur_devices);
2236 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2237 mutex_lock(&fs_info->chunk_mutex);
2238 list_add(&device->dev_alloc_list,
2239 &fs_devices->alloc_list);
2240 device->fs_devices->rw_devices++;
2241 mutex_unlock(&fs_info->chunk_mutex);
2246 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev)
2248 struct btrfs_fs_devices *fs_devices;
2250 lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex);
2253 * in case of fs with no seed, srcdev->fs_devices will point
2254 * to fs_devices of fs_info. However when the dev being replaced is
2255 * a seed dev it will point to the seed's local fs_devices. In short
2256 * srcdev will have its correct fs_devices in both the cases.
2258 fs_devices = srcdev->fs_devices;
2260 list_del_rcu(&srcdev->dev_list);
2261 list_del(&srcdev->dev_alloc_list);
2262 fs_devices->num_devices--;
2263 if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state))
2264 fs_devices->missing_devices--;
2266 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state))
2267 fs_devices->rw_devices--;
2270 fs_devices->open_devices--;
2273 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev)
2275 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2277 mutex_lock(&uuid_mutex);
2279 btrfs_close_bdev(srcdev);
2281 btrfs_free_device(srcdev);
2283 /* if this is no devs we rather delete the fs_devices */
2284 if (!fs_devices->num_devices) {
2286 * On a mounted FS, num_devices can't be zero unless it's a
2287 * seed. In case of a seed device being replaced, the replace
2288 * target added to the sprout FS, so there will be no more
2289 * device left under the seed FS.
2291 ASSERT(fs_devices->seeding);
2293 list_del_init(&fs_devices->seed_list);
2294 close_fs_devices(fs_devices);
2295 free_fs_devices(fs_devices);
2297 mutex_unlock(&uuid_mutex);
2300 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev)
2302 struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices;
2304 mutex_lock(&fs_devices->device_list_mutex);
2306 btrfs_sysfs_remove_device(tgtdev);
2309 fs_devices->open_devices--;
2311 fs_devices->num_devices--;
2313 btrfs_assign_next_active_device(tgtdev, NULL);
2315 list_del_rcu(&tgtdev->dev_list);
2317 mutex_unlock(&fs_devices->device_list_mutex);
2319 btrfs_scratch_superblocks(tgtdev->fs_info, tgtdev->bdev,
2322 btrfs_close_bdev(tgtdev);
2324 btrfs_free_device(tgtdev);
2328 * Populate args from device at path
2330 * @fs_info: the filesystem
2331 * @args: the args to populate
2332 * @path: the path to the device
2334 * This will read the super block of the device at @path and populate @args with
2335 * the devid, fsid, and uuid. This is meant to be used for ioctls that need to
2336 * lookup a device to operate on, but need to do it before we take any locks.
2337 * This properly handles the special case of "missing" that a user may pass in,
2338 * and does some basic sanity checks. The caller must make sure that @path is
2339 * properly NUL terminated before calling in, and must call
2340 * btrfs_put_dev_args_from_path() in order to free up the temporary fsid and
2343 * Return: 0 for success, -errno for failure
2345 int btrfs_get_dev_args_from_path(struct btrfs_fs_info *fs_info,
2346 struct btrfs_dev_lookup_args *args,
2349 struct btrfs_super_block *disk_super;
2350 struct block_device *bdev;
2353 if (!path || !path[0])
2355 if (!strcmp(path, "missing")) {
2356 args->missing = true;
2360 args->uuid = kzalloc(BTRFS_UUID_SIZE, GFP_KERNEL);
2361 args->fsid = kzalloc(BTRFS_FSID_SIZE, GFP_KERNEL);
2362 if (!args->uuid || !args->fsid) {
2363 btrfs_put_dev_args_from_path(args);
2367 ret = btrfs_get_bdev_and_sb(path, FMODE_READ, fs_info->bdev_holder, 0,
2368 &bdev, &disk_super);
2371 args->devid = btrfs_stack_device_id(&disk_super->dev_item);
2372 memcpy(args->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE);
2373 if (btrfs_fs_incompat(fs_info, METADATA_UUID))
2374 memcpy(args->fsid, disk_super->metadata_uuid, BTRFS_FSID_SIZE);
2376 memcpy(args->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
2377 btrfs_release_disk_super(disk_super);
2378 blkdev_put(bdev, FMODE_READ);
2383 * Only use this jointly with btrfs_get_dev_args_from_path() because we will
2384 * allocate our ->uuid and ->fsid pointers, everybody else uses local variables
2385 * that don't need to be freed.
2387 void btrfs_put_dev_args_from_path(struct btrfs_dev_lookup_args *args)
2395 struct btrfs_device *btrfs_find_device_by_devspec(
2396 struct btrfs_fs_info *fs_info, u64 devid,
2397 const char *device_path)
2399 BTRFS_DEV_LOOKUP_ARGS(args);
2400 struct btrfs_device *device;
2405 device = btrfs_find_device(fs_info->fs_devices, &args);
2407 return ERR_PTR(-ENOENT);
2411 ret = btrfs_get_dev_args_from_path(fs_info, &args, device_path);
2413 return ERR_PTR(ret);
2414 device = btrfs_find_device(fs_info->fs_devices, &args);
2415 btrfs_put_dev_args_from_path(&args);
2417 return ERR_PTR(-ENOENT);
2421 static struct btrfs_fs_devices *btrfs_init_sprout(struct btrfs_fs_info *fs_info)
2423 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2424 struct btrfs_fs_devices *old_devices;
2425 struct btrfs_fs_devices *seed_devices;
2427 lockdep_assert_held(&uuid_mutex);
2428 if (!fs_devices->seeding)
2429 return ERR_PTR(-EINVAL);
2432 * Private copy of the seed devices, anchored at
2433 * fs_info->fs_devices->seed_list
2435 seed_devices = alloc_fs_devices(NULL, NULL);
2436 if (IS_ERR(seed_devices))
2437 return seed_devices;
2440 * It's necessary to retain a copy of the original seed fs_devices in
2441 * fs_uuids so that filesystems which have been seeded can successfully
2442 * reference the seed device from open_seed_devices. This also supports
2445 old_devices = clone_fs_devices(fs_devices);
2446 if (IS_ERR(old_devices)) {
2447 kfree(seed_devices);
2451 list_add(&old_devices->fs_list, &fs_uuids);
2453 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2454 seed_devices->opened = 1;
2455 INIT_LIST_HEAD(&seed_devices->devices);
2456 INIT_LIST_HEAD(&seed_devices->alloc_list);
2457 mutex_init(&seed_devices->device_list_mutex);
2459 return seed_devices;
2463 * Splice seed devices into the sprout fs_devices.
2464 * Generate a new fsid for the sprouted read-write filesystem.
2466 static void btrfs_setup_sprout(struct btrfs_fs_info *fs_info,
2467 struct btrfs_fs_devices *seed_devices)
2469 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2470 struct btrfs_super_block *disk_super = fs_info->super_copy;
2471 struct btrfs_device *device;
2475 * We are updating the fsid, the thread leading to device_list_add()
2476 * could race, so uuid_mutex is needed.
2478 lockdep_assert_held(&uuid_mutex);
2481 * The threads listed below may traverse dev_list but can do that without
2482 * device_list_mutex:
2483 * - All device ops and balance - as we are in btrfs_exclop_start.
2484 * - Various dev_list readers - are using RCU.
2485 * - btrfs_ioctl_fitrim() - is using RCU.
2487 * For-read threads as below are using device_list_mutex:
2488 * - Readonly scrub btrfs_scrub_dev()
2489 * - Readonly scrub btrfs_scrub_progress()
2490 * - btrfs_get_dev_stats()
2492 lockdep_assert_held(&fs_devices->device_list_mutex);
2494 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2496 list_for_each_entry(device, &seed_devices->devices, dev_list)
2497 device->fs_devices = seed_devices;
2499 fs_devices->seeding = false;
2500 fs_devices->num_devices = 0;
2501 fs_devices->open_devices = 0;
2502 fs_devices->missing_devices = 0;
2503 fs_devices->rotating = false;
2504 list_add(&seed_devices->seed_list, &fs_devices->seed_list);
2506 generate_random_uuid(fs_devices->fsid);
2507 memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE);
2508 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2510 super_flags = btrfs_super_flags(disk_super) &
2511 ~BTRFS_SUPER_FLAG_SEEDING;
2512 btrfs_set_super_flags(disk_super, super_flags);
2516 * Store the expected generation for seed devices in device items.
2518 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans)
2520 BTRFS_DEV_LOOKUP_ARGS(args);
2521 struct btrfs_fs_info *fs_info = trans->fs_info;
2522 struct btrfs_root *root = fs_info->chunk_root;
2523 struct btrfs_path *path;
2524 struct extent_buffer *leaf;
2525 struct btrfs_dev_item *dev_item;
2526 struct btrfs_device *device;
2527 struct btrfs_key key;
2528 u8 fs_uuid[BTRFS_FSID_SIZE];
2529 u8 dev_uuid[BTRFS_UUID_SIZE];
2532 path = btrfs_alloc_path();
2536 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2538 key.type = BTRFS_DEV_ITEM_KEY;
2541 btrfs_reserve_chunk_metadata(trans, false);
2542 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2543 btrfs_trans_release_chunk_metadata(trans);
2547 leaf = path->nodes[0];
2549 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2550 ret = btrfs_next_leaf(root, path);
2555 leaf = path->nodes[0];
2556 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2557 btrfs_release_path(path);
2561 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2562 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2563 key.type != BTRFS_DEV_ITEM_KEY)
2566 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2567 struct btrfs_dev_item);
2568 args.devid = btrfs_device_id(leaf, dev_item);
2569 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2571 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2573 args.uuid = dev_uuid;
2574 args.fsid = fs_uuid;
2575 device = btrfs_find_device(fs_info->fs_devices, &args);
2576 BUG_ON(!device); /* Logic error */
2578 if (device->fs_devices->seeding) {
2579 btrfs_set_device_generation(leaf, dev_item,
2580 device->generation);
2581 btrfs_mark_buffer_dirty(leaf);
2589 btrfs_free_path(path);
2593 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2595 struct btrfs_root *root = fs_info->dev_root;
2596 struct btrfs_trans_handle *trans;
2597 struct btrfs_device *device;
2598 struct block_device *bdev;
2599 struct super_block *sb = fs_info->sb;
2600 struct rcu_string *name;
2601 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2602 struct btrfs_fs_devices *seed_devices;
2603 u64 orig_super_total_bytes;
2604 u64 orig_super_num_devices;
2606 bool seeding_dev = false;
2607 bool locked = false;
2609 if (sb_rdonly(sb) && !fs_devices->seeding)
2612 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2613 fs_info->bdev_holder);
2615 return PTR_ERR(bdev);
2617 if (!btrfs_check_device_zone_type(fs_info, bdev)) {
2622 if (fs_devices->seeding) {
2624 down_write(&sb->s_umount);
2625 mutex_lock(&uuid_mutex);
2629 sync_blockdev(bdev);
2632 list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
2633 if (device->bdev == bdev) {
2641 device = btrfs_alloc_device(fs_info, NULL, NULL);
2642 if (IS_ERR(device)) {
2643 /* we can safely leave the fs_devices entry around */
2644 ret = PTR_ERR(device);
2648 name = rcu_string_strdup(device_path, GFP_KERNEL);
2651 goto error_free_device;
2653 rcu_assign_pointer(device->name, name);
2655 device->fs_info = fs_info;
2656 device->bdev = bdev;
2657 ret = lookup_bdev(device_path, &device->devt);
2659 goto error_free_device;
2661 ret = btrfs_get_dev_zone_info(device, false);
2663 goto error_free_device;
2665 trans = btrfs_start_transaction(root, 0);
2666 if (IS_ERR(trans)) {
2667 ret = PTR_ERR(trans);
2668 goto error_free_zone;
2671 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
2672 device->generation = trans->transid;
2673 device->io_width = fs_info->sectorsize;
2674 device->io_align = fs_info->sectorsize;
2675 device->sector_size = fs_info->sectorsize;
2676 device->total_bytes =
2677 round_down(bdev_nr_bytes(bdev), fs_info->sectorsize);
2678 device->disk_total_bytes = device->total_bytes;
2679 device->commit_total_bytes = device->total_bytes;
2680 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2681 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
2682 device->mode = FMODE_EXCL;
2683 device->dev_stats_valid = 1;
2684 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2687 btrfs_clear_sb_rdonly(sb);
2689 /* GFP_KERNEL allocation must not be under device_list_mutex */
2690 seed_devices = btrfs_init_sprout(fs_info);
2691 if (IS_ERR(seed_devices)) {
2692 ret = PTR_ERR(seed_devices);
2693 btrfs_abort_transaction(trans, ret);
2698 mutex_lock(&fs_devices->device_list_mutex);
2700 btrfs_setup_sprout(fs_info, seed_devices);
2701 btrfs_assign_next_active_device(fs_info->fs_devices->latest_dev,
2705 device->fs_devices = fs_devices;
2707 mutex_lock(&fs_info->chunk_mutex);
2708 list_add_rcu(&device->dev_list, &fs_devices->devices);
2709 list_add(&device->dev_alloc_list, &fs_devices->alloc_list);
2710 fs_devices->num_devices++;
2711 fs_devices->open_devices++;
2712 fs_devices->rw_devices++;
2713 fs_devices->total_devices++;
2714 fs_devices->total_rw_bytes += device->total_bytes;
2716 atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2718 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2719 fs_devices->rotating = true;
2721 orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
2722 btrfs_set_super_total_bytes(fs_info->super_copy,
2723 round_down(orig_super_total_bytes + device->total_bytes,
2724 fs_info->sectorsize));
2726 orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy);
2727 btrfs_set_super_num_devices(fs_info->super_copy,
2728 orig_super_num_devices + 1);
2731 * we've got more storage, clear any full flags on the space
2734 btrfs_clear_space_info_full(fs_info);
2736 mutex_unlock(&fs_info->chunk_mutex);
2738 /* Add sysfs device entry */
2739 btrfs_sysfs_add_device(device);
2741 mutex_unlock(&fs_devices->device_list_mutex);
2744 mutex_lock(&fs_info->chunk_mutex);
2745 ret = init_first_rw_device(trans);
2746 mutex_unlock(&fs_info->chunk_mutex);
2748 btrfs_abort_transaction(trans, ret);
2753 ret = btrfs_add_dev_item(trans, device);
2755 btrfs_abort_transaction(trans, ret);
2760 ret = btrfs_finish_sprout(trans);
2762 btrfs_abort_transaction(trans, ret);
2767 * fs_devices now represents the newly sprouted filesystem and
2768 * its fsid has been changed by btrfs_sprout_splice().
2770 btrfs_sysfs_update_sprout_fsid(fs_devices);
2773 ret = btrfs_commit_transaction(trans);
2776 mutex_unlock(&uuid_mutex);
2777 up_write(&sb->s_umount);
2780 if (ret) /* transaction commit */
2783 ret = btrfs_relocate_sys_chunks(fs_info);
2785 btrfs_handle_fs_error(fs_info, ret,
2786 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2787 trans = btrfs_attach_transaction(root);
2788 if (IS_ERR(trans)) {
2789 if (PTR_ERR(trans) == -ENOENT)
2791 ret = PTR_ERR(trans);
2795 ret = btrfs_commit_transaction(trans);
2799 * Now that we have written a new super block to this device, check all
2800 * other fs_devices list if device_path alienates any other scanned
2802 * We can ignore the return value as it typically returns -EINVAL and
2803 * only succeeds if the device was an alien.
2805 btrfs_forget_devices(device->devt);
2807 /* Update ctime/mtime for blkid or udev */
2808 update_dev_time(device_path);
2813 btrfs_sysfs_remove_device(device);
2814 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2815 mutex_lock(&fs_info->chunk_mutex);
2816 list_del_rcu(&device->dev_list);
2817 list_del(&device->dev_alloc_list);
2818 fs_info->fs_devices->num_devices--;
2819 fs_info->fs_devices->open_devices--;
2820 fs_info->fs_devices->rw_devices--;
2821 fs_info->fs_devices->total_devices--;
2822 fs_info->fs_devices->total_rw_bytes -= device->total_bytes;
2823 atomic64_sub(device->total_bytes, &fs_info->free_chunk_space);
2824 btrfs_set_super_total_bytes(fs_info->super_copy,
2825 orig_super_total_bytes);
2826 btrfs_set_super_num_devices(fs_info->super_copy,
2827 orig_super_num_devices);
2828 mutex_unlock(&fs_info->chunk_mutex);
2829 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2832 btrfs_set_sb_rdonly(sb);
2834 btrfs_end_transaction(trans);
2836 btrfs_destroy_dev_zone_info(device);
2838 btrfs_free_device(device);
2840 blkdev_put(bdev, FMODE_EXCL);
2842 mutex_unlock(&uuid_mutex);
2843 up_write(&sb->s_umount);
2848 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2849 struct btrfs_device *device)
2852 struct btrfs_path *path;
2853 struct btrfs_root *root = device->fs_info->chunk_root;
2854 struct btrfs_dev_item *dev_item;
2855 struct extent_buffer *leaf;
2856 struct btrfs_key key;
2858 path = btrfs_alloc_path();
2862 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2863 key.type = BTRFS_DEV_ITEM_KEY;
2864 key.offset = device->devid;
2866 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2875 leaf = path->nodes[0];
2876 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2878 btrfs_set_device_id(leaf, dev_item, device->devid);
2879 btrfs_set_device_type(leaf, dev_item, device->type);
2880 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2881 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2882 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2883 btrfs_set_device_total_bytes(leaf, dev_item,
2884 btrfs_device_get_disk_total_bytes(device));
2885 btrfs_set_device_bytes_used(leaf, dev_item,
2886 btrfs_device_get_bytes_used(device));
2887 btrfs_mark_buffer_dirty(leaf);
2890 btrfs_free_path(path);
2894 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2895 struct btrfs_device *device, u64 new_size)
2897 struct btrfs_fs_info *fs_info = device->fs_info;
2898 struct btrfs_super_block *super_copy = fs_info->super_copy;
2903 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2906 new_size = round_down(new_size, fs_info->sectorsize);
2908 mutex_lock(&fs_info->chunk_mutex);
2909 old_total = btrfs_super_total_bytes(super_copy);
2910 diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2912 if (new_size <= device->total_bytes ||
2913 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2914 mutex_unlock(&fs_info->chunk_mutex);
2918 btrfs_set_super_total_bytes(super_copy,
2919 round_down(old_total + diff, fs_info->sectorsize));
2920 device->fs_devices->total_rw_bytes += diff;
2922 btrfs_device_set_total_bytes(device, new_size);
2923 btrfs_device_set_disk_total_bytes(device, new_size);
2924 btrfs_clear_space_info_full(device->fs_info);
2925 if (list_empty(&device->post_commit_list))
2926 list_add_tail(&device->post_commit_list,
2927 &trans->transaction->dev_update_list);
2928 mutex_unlock(&fs_info->chunk_mutex);
2930 btrfs_reserve_chunk_metadata(trans, false);
2931 ret = btrfs_update_device(trans, device);
2932 btrfs_trans_release_chunk_metadata(trans);
2937 static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
2939 struct btrfs_fs_info *fs_info = trans->fs_info;
2940 struct btrfs_root *root = fs_info->chunk_root;
2942 struct btrfs_path *path;
2943 struct btrfs_key key;
2945 path = btrfs_alloc_path();
2949 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2950 key.offset = chunk_offset;
2951 key.type = BTRFS_CHUNK_ITEM_KEY;
2953 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2956 else if (ret > 0) { /* Logic error or corruption */
2957 btrfs_handle_fs_error(fs_info, -ENOENT,
2958 "Failed lookup while freeing chunk.");
2963 ret = btrfs_del_item(trans, root, path);
2965 btrfs_handle_fs_error(fs_info, ret,
2966 "Failed to delete chunk item.");
2968 btrfs_free_path(path);
2972 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2974 struct btrfs_super_block *super_copy = fs_info->super_copy;
2975 struct btrfs_disk_key *disk_key;
2976 struct btrfs_chunk *chunk;
2983 struct btrfs_key key;
2985 lockdep_assert_held(&fs_info->chunk_mutex);
2986 array_size = btrfs_super_sys_array_size(super_copy);
2988 ptr = super_copy->sys_chunk_array;
2991 while (cur < array_size) {
2992 disk_key = (struct btrfs_disk_key *)ptr;
2993 btrfs_disk_key_to_cpu(&key, disk_key);
2995 len = sizeof(*disk_key);
2997 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2998 chunk = (struct btrfs_chunk *)(ptr + len);
2999 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
3000 len += btrfs_chunk_item_size(num_stripes);
3005 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID &&
3006 key.offset == chunk_offset) {
3007 memmove(ptr, ptr + len, array_size - (cur + len));
3009 btrfs_set_super_sys_array_size(super_copy, array_size);
3019 * btrfs_get_chunk_map() - Find the mapping containing the given logical extent.
3020 * @logical: Logical block offset in bytes.
3021 * @length: Length of extent in bytes.
3023 * Return: Chunk mapping or ERR_PTR.
3025 struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info,
3026 u64 logical, u64 length)
3028 struct extent_map_tree *em_tree;
3029 struct extent_map *em;
3031 em_tree = &fs_info->mapping_tree;
3032 read_lock(&em_tree->lock);
3033 em = lookup_extent_mapping(em_tree, logical, length);
3034 read_unlock(&em_tree->lock);
3037 btrfs_crit(fs_info, "unable to find logical %llu length %llu",
3039 return ERR_PTR(-EINVAL);
3042 if (em->start > logical || em->start + em->len < logical) {
3044 "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
3045 logical, length, em->start, em->start + em->len);
3046 free_extent_map(em);
3047 return ERR_PTR(-EINVAL);
3050 /* callers are responsible for dropping em's ref. */
3054 static int remove_chunk_item(struct btrfs_trans_handle *trans,
3055 struct map_lookup *map, u64 chunk_offset)
3060 * Removing chunk items and updating the device items in the chunks btree
3061 * requires holding the chunk_mutex.
3062 * See the comment at btrfs_chunk_alloc() for the details.
3064 lockdep_assert_held(&trans->fs_info->chunk_mutex);
3066 for (i = 0; i < map->num_stripes; i++) {
3069 ret = btrfs_update_device(trans, map->stripes[i].dev);
3074 return btrfs_free_chunk(trans, chunk_offset);
3077 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
3079 struct btrfs_fs_info *fs_info = trans->fs_info;
3080 struct extent_map *em;
3081 struct map_lookup *map;
3082 u64 dev_extent_len = 0;
3084 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
3086 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
3089 * This is a logic error, but we don't want to just rely on the
3090 * user having built with ASSERT enabled, so if ASSERT doesn't
3091 * do anything we still error out.
3096 map = em->map_lookup;
3099 * First delete the device extent items from the devices btree.
3100 * We take the device_list_mutex to avoid racing with the finishing phase
3101 * of a device replace operation. See the comment below before acquiring
3102 * fs_info->chunk_mutex. Note that here we do not acquire the chunk_mutex
3103 * because that can result in a deadlock when deleting the device extent
3104 * items from the devices btree - COWing an extent buffer from the btree
3105 * may result in allocating a new metadata chunk, which would attempt to
3106 * lock again fs_info->chunk_mutex.
3108 mutex_lock(&fs_devices->device_list_mutex);
3109 for (i = 0; i < map->num_stripes; i++) {
3110 struct btrfs_device *device = map->stripes[i].dev;
3111 ret = btrfs_free_dev_extent(trans, device,
3112 map->stripes[i].physical,
3115 mutex_unlock(&fs_devices->device_list_mutex);
3116 btrfs_abort_transaction(trans, ret);
3120 if (device->bytes_used > 0) {
3121 mutex_lock(&fs_info->chunk_mutex);
3122 btrfs_device_set_bytes_used(device,
3123 device->bytes_used - dev_extent_len);
3124 atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
3125 btrfs_clear_space_info_full(fs_info);
3126 mutex_unlock(&fs_info->chunk_mutex);
3129 mutex_unlock(&fs_devices->device_list_mutex);
3132 * We acquire fs_info->chunk_mutex for 2 reasons:
3134 * 1) Just like with the first phase of the chunk allocation, we must
3135 * reserve system space, do all chunk btree updates and deletions, and
3136 * update the system chunk array in the superblock while holding this
3137 * mutex. This is for similar reasons as explained on the comment at
3138 * the top of btrfs_chunk_alloc();
3140 * 2) Prevent races with the final phase of a device replace operation
3141 * that replaces the device object associated with the map's stripes,
3142 * because the device object's id can change at any time during that
3143 * final phase of the device replace operation
3144 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the
3145 * replaced device and then see it with an ID of
3146 * BTRFS_DEV_REPLACE_DEVID, which would cause a failure when updating
3147 * the device item, which does not exists on the chunk btree.
3148 * The finishing phase of device replace acquires both the
3149 * device_list_mutex and the chunk_mutex, in that order, so we are
3150 * safe by just acquiring the chunk_mutex.
3152 trans->removing_chunk = true;
3153 mutex_lock(&fs_info->chunk_mutex);
3155 check_system_chunk(trans, map->type);
3157 ret = remove_chunk_item(trans, map, chunk_offset);
3159 * Normally we should not get -ENOSPC since we reserved space before
3160 * through the call to check_system_chunk().
3162 * Despite our system space_info having enough free space, we may not
3163 * be able to allocate extents from its block groups, because all have
3164 * an incompatible profile, which will force us to allocate a new system
3165 * block group with the right profile, or right after we called
3166 * check_system_space() above, a scrub turned the only system block group
3167 * with enough free space into RO mode.
3168 * This is explained with more detail at do_chunk_alloc().
3170 * So if we get -ENOSPC, allocate a new system chunk and retry once.
3172 if (ret == -ENOSPC) {
3173 const u64 sys_flags = btrfs_system_alloc_profile(fs_info);
3174 struct btrfs_block_group *sys_bg;
3176 sys_bg = btrfs_create_chunk(trans, sys_flags);
3177 if (IS_ERR(sys_bg)) {
3178 ret = PTR_ERR(sys_bg);
3179 btrfs_abort_transaction(trans, ret);
3183 ret = btrfs_chunk_alloc_add_chunk_item(trans, sys_bg);
3185 btrfs_abort_transaction(trans, ret);
3189 ret = remove_chunk_item(trans, map, chunk_offset);
3191 btrfs_abort_transaction(trans, ret);
3195 btrfs_abort_transaction(trans, ret);
3199 trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
3201 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
3202 ret = btrfs_del_sys_chunk(fs_info, chunk_offset);
3204 btrfs_abort_transaction(trans, ret);
3209 mutex_unlock(&fs_info->chunk_mutex);
3210 trans->removing_chunk = false;
3213 * We are done with chunk btree updates and deletions, so release the
3214 * system space we previously reserved (with check_system_chunk()).
3216 btrfs_trans_release_chunk_metadata(trans);
3218 ret = btrfs_remove_block_group(trans, chunk_offset, em);
3220 btrfs_abort_transaction(trans, ret);
3225 if (trans->removing_chunk) {
3226 mutex_unlock(&fs_info->chunk_mutex);
3227 trans->removing_chunk = false;
3230 free_extent_map(em);
3234 int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
3236 struct btrfs_root *root = fs_info->chunk_root;
3237 struct btrfs_trans_handle *trans;
3238 struct btrfs_block_group *block_group;
3242 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
3244 "relocate: not supported on extent tree v2 yet");
3249 * Prevent races with automatic removal of unused block groups.
3250 * After we relocate and before we remove the chunk with offset
3251 * chunk_offset, automatic removal of the block group can kick in,
3252 * resulting in a failure when calling btrfs_remove_chunk() below.
3254 * Make sure to acquire this mutex before doing a tree search (dev
3255 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
3256 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
3257 * we release the path used to search the chunk/dev tree and before
3258 * the current task acquires this mutex and calls us.
3260 lockdep_assert_held(&fs_info->reclaim_bgs_lock);
3262 /* step one, relocate all the extents inside this chunk */
3263 btrfs_scrub_pause(fs_info);
3264 ret = btrfs_relocate_block_group(fs_info, chunk_offset);
3265 btrfs_scrub_continue(fs_info);
3269 block_group = btrfs_lookup_block_group(fs_info, chunk_offset);
3272 btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group);
3273 length = block_group->length;
3274 btrfs_put_block_group(block_group);
3277 * On a zoned file system, discard the whole block group, this will
3278 * trigger a REQ_OP_ZONE_RESET operation on the device zone. If
3279 * resetting the zone fails, don't treat it as a fatal problem from the
3280 * filesystem's point of view.
3282 if (btrfs_is_zoned(fs_info)) {
3283 ret = btrfs_discard_extent(fs_info, chunk_offset, length, NULL);
3286 "failed to reset zone %llu after relocation",
3290 trans = btrfs_start_trans_remove_block_group(root->fs_info,
3292 if (IS_ERR(trans)) {
3293 ret = PTR_ERR(trans);
3294 btrfs_handle_fs_error(root->fs_info, ret, NULL);
3299 * step two, delete the device extents and the
3300 * chunk tree entries
3302 ret = btrfs_remove_chunk(trans, chunk_offset);
3303 btrfs_end_transaction(trans);
3307 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
3309 struct btrfs_root *chunk_root = fs_info->chunk_root;
3310 struct btrfs_path *path;
3311 struct extent_buffer *leaf;
3312 struct btrfs_chunk *chunk;
3313 struct btrfs_key key;
3314 struct btrfs_key found_key;
3316 bool retried = false;
3320 path = btrfs_alloc_path();
3325 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3326 key.offset = (u64)-1;
3327 key.type = BTRFS_CHUNK_ITEM_KEY;
3330 mutex_lock(&fs_info->reclaim_bgs_lock);
3331 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3333 mutex_unlock(&fs_info->reclaim_bgs_lock);
3336 BUG_ON(ret == 0); /* Corruption */
3338 ret = btrfs_previous_item(chunk_root, path, key.objectid,
3341 mutex_unlock(&fs_info->reclaim_bgs_lock);
3347 leaf = path->nodes[0];
3348 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3350 chunk = btrfs_item_ptr(leaf, path->slots[0],
3351 struct btrfs_chunk);
3352 chunk_type = btrfs_chunk_type(leaf, chunk);
3353 btrfs_release_path(path);
3355 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3356 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3362 mutex_unlock(&fs_info->reclaim_bgs_lock);
3364 if (found_key.offset == 0)
3366 key.offset = found_key.offset - 1;
3369 if (failed && !retried) {
3373 } else if (WARN_ON(failed && retried)) {
3377 btrfs_free_path(path);
3382 * return 1 : allocate a data chunk successfully,
3383 * return <0: errors during allocating a data chunk,
3384 * return 0 : no need to allocate a data chunk.
3386 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info,
3389 struct btrfs_block_group *cache;
3393 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3395 chunk_type = cache->flags;
3396 btrfs_put_block_group(cache);
3398 if (!(chunk_type & BTRFS_BLOCK_GROUP_DATA))
3401 spin_lock(&fs_info->data_sinfo->lock);
3402 bytes_used = fs_info->data_sinfo->bytes_used;
3403 spin_unlock(&fs_info->data_sinfo->lock);
3406 struct btrfs_trans_handle *trans;
3409 trans = btrfs_join_transaction(fs_info->tree_root);
3411 return PTR_ERR(trans);
3413 ret = btrfs_force_chunk_alloc(trans, BTRFS_BLOCK_GROUP_DATA);
3414 btrfs_end_transaction(trans);
3423 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3424 struct btrfs_balance_control *bctl)
3426 struct btrfs_root *root = fs_info->tree_root;
3427 struct btrfs_trans_handle *trans;
3428 struct btrfs_balance_item *item;
3429 struct btrfs_disk_balance_args disk_bargs;
3430 struct btrfs_path *path;
3431 struct extent_buffer *leaf;
3432 struct btrfs_key key;
3435 path = btrfs_alloc_path();
3439 trans = btrfs_start_transaction(root, 0);
3440 if (IS_ERR(trans)) {
3441 btrfs_free_path(path);
3442 return PTR_ERR(trans);
3445 key.objectid = BTRFS_BALANCE_OBJECTID;
3446 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3449 ret = btrfs_insert_empty_item(trans, root, path, &key,
3454 leaf = path->nodes[0];
3455 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3457 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3459 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3460 btrfs_set_balance_data(leaf, item, &disk_bargs);
3461 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3462 btrfs_set_balance_meta(leaf, item, &disk_bargs);
3463 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3464 btrfs_set_balance_sys(leaf, item, &disk_bargs);
3466 btrfs_set_balance_flags(leaf, item, bctl->flags);
3468 btrfs_mark_buffer_dirty(leaf);
3470 btrfs_free_path(path);
3471 err = btrfs_commit_transaction(trans);
3477 static int del_balance_item(struct btrfs_fs_info *fs_info)
3479 struct btrfs_root *root = fs_info->tree_root;
3480 struct btrfs_trans_handle *trans;
3481 struct btrfs_path *path;
3482 struct btrfs_key key;
3485 path = btrfs_alloc_path();
3489 trans = btrfs_start_transaction_fallback_global_rsv(root, 0);
3490 if (IS_ERR(trans)) {
3491 btrfs_free_path(path);
3492 return PTR_ERR(trans);
3495 key.objectid = BTRFS_BALANCE_OBJECTID;
3496 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3499 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3507 ret = btrfs_del_item(trans, root, path);
3509 btrfs_free_path(path);
3510 err = btrfs_commit_transaction(trans);
3517 * This is a heuristic used to reduce the number of chunks balanced on
3518 * resume after balance was interrupted.
3520 static void update_balance_args(struct btrfs_balance_control *bctl)
3523 * Turn on soft mode for chunk types that were being converted.
3525 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3526 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3527 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3528 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3529 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3530 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3533 * Turn on usage filter if is not already used. The idea is
3534 * that chunks that we have already balanced should be
3535 * reasonably full. Don't do it for chunks that are being
3536 * converted - that will keep us from relocating unconverted
3537 * (albeit full) chunks.
3539 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3540 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3541 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3542 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3543 bctl->data.usage = 90;
3545 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3546 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3547 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3548 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3549 bctl->sys.usage = 90;
3551 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3552 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3553 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3554 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3555 bctl->meta.usage = 90;
3560 * Clear the balance status in fs_info and delete the balance item from disk.
3562 static void reset_balance_state(struct btrfs_fs_info *fs_info)
3564 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3567 BUG_ON(!fs_info->balance_ctl);
3569 spin_lock(&fs_info->balance_lock);
3570 fs_info->balance_ctl = NULL;
3571 spin_unlock(&fs_info->balance_lock);
3574 ret = del_balance_item(fs_info);
3576 btrfs_handle_fs_error(fs_info, ret, NULL);
3580 * Balance filters. Return 1 if chunk should be filtered out
3581 * (should not be balanced).
3583 static int chunk_profiles_filter(u64 chunk_type,
3584 struct btrfs_balance_args *bargs)
3586 chunk_type = chunk_to_extended(chunk_type) &
3587 BTRFS_EXTENDED_PROFILE_MASK;
3589 if (bargs->profiles & chunk_type)
3595 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3596 struct btrfs_balance_args *bargs)
3598 struct btrfs_block_group *cache;
3600 u64 user_thresh_min;
3601 u64 user_thresh_max;
3604 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3605 chunk_used = cache->used;
3607 if (bargs->usage_min == 0)
3608 user_thresh_min = 0;
3610 user_thresh_min = div_factor_fine(cache->length,
3613 if (bargs->usage_max == 0)
3614 user_thresh_max = 1;
3615 else if (bargs->usage_max > 100)
3616 user_thresh_max = cache->length;
3618 user_thresh_max = div_factor_fine(cache->length,
3621 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3624 btrfs_put_block_group(cache);
3628 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3629 u64 chunk_offset, struct btrfs_balance_args *bargs)
3631 struct btrfs_block_group *cache;
3632 u64 chunk_used, user_thresh;
3635 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3636 chunk_used = cache->used;
3638 if (bargs->usage_min == 0)
3640 else if (bargs->usage > 100)
3641 user_thresh = cache->length;
3643 user_thresh = div_factor_fine(cache->length, bargs->usage);
3645 if (chunk_used < user_thresh)
3648 btrfs_put_block_group(cache);
3652 static int chunk_devid_filter(struct extent_buffer *leaf,
3653 struct btrfs_chunk *chunk,
3654 struct btrfs_balance_args *bargs)
3656 struct btrfs_stripe *stripe;
3657 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3660 for (i = 0; i < num_stripes; i++) {
3661 stripe = btrfs_stripe_nr(chunk, i);
3662 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3669 static u64 calc_data_stripes(u64 type, int num_stripes)
3671 const int index = btrfs_bg_flags_to_raid_index(type);
3672 const int ncopies = btrfs_raid_array[index].ncopies;
3673 const int nparity = btrfs_raid_array[index].nparity;
3675 return (num_stripes - nparity) / ncopies;
3678 /* [pstart, pend) */
3679 static int chunk_drange_filter(struct extent_buffer *leaf,
3680 struct btrfs_chunk *chunk,
3681 struct btrfs_balance_args *bargs)
3683 struct btrfs_stripe *stripe;
3684 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3691 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3694 type = btrfs_chunk_type(leaf, chunk);
3695 factor = calc_data_stripes(type, num_stripes);
3697 for (i = 0; i < num_stripes; i++) {
3698 stripe = btrfs_stripe_nr(chunk, i);
3699 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3702 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3703 stripe_length = btrfs_chunk_length(leaf, chunk);
3704 stripe_length = div_u64(stripe_length, factor);
3706 if (stripe_offset < bargs->pend &&
3707 stripe_offset + stripe_length > bargs->pstart)
3714 /* [vstart, vend) */
3715 static int chunk_vrange_filter(struct extent_buffer *leaf,
3716 struct btrfs_chunk *chunk,
3718 struct btrfs_balance_args *bargs)
3720 if (chunk_offset < bargs->vend &&
3721 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3722 /* at least part of the chunk is inside this vrange */
3728 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3729 struct btrfs_chunk *chunk,
3730 struct btrfs_balance_args *bargs)
3732 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3734 if (bargs->stripes_min <= num_stripes
3735 && num_stripes <= bargs->stripes_max)
3741 static int chunk_soft_convert_filter(u64 chunk_type,
3742 struct btrfs_balance_args *bargs)
3744 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3747 chunk_type = chunk_to_extended(chunk_type) &
3748 BTRFS_EXTENDED_PROFILE_MASK;
3750 if (bargs->target == chunk_type)
3756 static int should_balance_chunk(struct extent_buffer *leaf,
3757 struct btrfs_chunk *chunk, u64 chunk_offset)
3759 struct btrfs_fs_info *fs_info = leaf->fs_info;
3760 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3761 struct btrfs_balance_args *bargs = NULL;
3762 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3765 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3766 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3770 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3771 bargs = &bctl->data;
3772 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3774 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3775 bargs = &bctl->meta;
3777 /* profiles filter */
3778 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3779 chunk_profiles_filter(chunk_type, bargs)) {
3784 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3785 chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3787 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3788 chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3793 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3794 chunk_devid_filter(leaf, chunk, bargs)) {
3798 /* drange filter, makes sense only with devid filter */
3799 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3800 chunk_drange_filter(leaf, chunk, bargs)) {
3805 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3806 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3810 /* stripes filter */
3811 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3812 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3816 /* soft profile changing mode */
3817 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3818 chunk_soft_convert_filter(chunk_type, bargs)) {
3823 * limited by count, must be the last filter
3825 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3826 if (bargs->limit == 0)
3830 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3832 * Same logic as the 'limit' filter; the minimum cannot be
3833 * determined here because we do not have the global information
3834 * about the count of all chunks that satisfy the filters.
3836 if (bargs->limit_max == 0)
3845 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3847 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3848 struct btrfs_root *chunk_root = fs_info->chunk_root;
3850 struct btrfs_chunk *chunk;
3851 struct btrfs_path *path = NULL;
3852 struct btrfs_key key;
3853 struct btrfs_key found_key;
3854 struct extent_buffer *leaf;
3857 int enospc_errors = 0;
3858 bool counting = true;
3859 /* The single value limit and min/max limits use the same bytes in the */
3860 u64 limit_data = bctl->data.limit;
3861 u64 limit_meta = bctl->meta.limit;
3862 u64 limit_sys = bctl->sys.limit;
3866 int chunk_reserved = 0;
3868 path = btrfs_alloc_path();
3874 /* zero out stat counters */
3875 spin_lock(&fs_info->balance_lock);
3876 memset(&bctl->stat, 0, sizeof(bctl->stat));
3877 spin_unlock(&fs_info->balance_lock);
3881 * The single value limit and min/max limits use the same bytes
3884 bctl->data.limit = limit_data;
3885 bctl->meta.limit = limit_meta;
3886 bctl->sys.limit = limit_sys;
3888 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3889 key.offset = (u64)-1;
3890 key.type = BTRFS_CHUNK_ITEM_KEY;
3893 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3894 atomic_read(&fs_info->balance_cancel_req)) {
3899 mutex_lock(&fs_info->reclaim_bgs_lock);
3900 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3902 mutex_unlock(&fs_info->reclaim_bgs_lock);
3907 * this shouldn't happen, it means the last relocate
3911 BUG(); /* FIXME break ? */
3913 ret = btrfs_previous_item(chunk_root, path, 0,
3914 BTRFS_CHUNK_ITEM_KEY);
3916 mutex_unlock(&fs_info->reclaim_bgs_lock);
3921 leaf = path->nodes[0];
3922 slot = path->slots[0];
3923 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3925 if (found_key.objectid != key.objectid) {
3926 mutex_unlock(&fs_info->reclaim_bgs_lock);
3930 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3931 chunk_type = btrfs_chunk_type(leaf, chunk);
3934 spin_lock(&fs_info->balance_lock);
3935 bctl->stat.considered++;
3936 spin_unlock(&fs_info->balance_lock);
3939 ret = should_balance_chunk(leaf, chunk, found_key.offset);
3941 btrfs_release_path(path);
3943 mutex_unlock(&fs_info->reclaim_bgs_lock);
3948 mutex_unlock(&fs_info->reclaim_bgs_lock);
3949 spin_lock(&fs_info->balance_lock);
3950 bctl->stat.expected++;
3951 spin_unlock(&fs_info->balance_lock);
3953 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3955 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3957 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3964 * Apply limit_min filter, no need to check if the LIMITS
3965 * filter is used, limit_min is 0 by default
3967 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3968 count_data < bctl->data.limit_min)
3969 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3970 count_meta < bctl->meta.limit_min)
3971 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3972 count_sys < bctl->sys.limit_min)) {
3973 mutex_unlock(&fs_info->reclaim_bgs_lock);
3977 if (!chunk_reserved) {
3979 * We may be relocating the only data chunk we have,
3980 * which could potentially end up with losing data's
3981 * raid profile, so lets allocate an empty one in
3984 ret = btrfs_may_alloc_data_chunk(fs_info,
3987 mutex_unlock(&fs_info->reclaim_bgs_lock);
3989 } else if (ret == 1) {
3994 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3995 mutex_unlock(&fs_info->reclaim_bgs_lock);
3996 if (ret == -ENOSPC) {
3998 } else if (ret == -ETXTBSY) {
4000 "skipping relocation of block group %llu due to active swapfile",
4006 spin_lock(&fs_info->balance_lock);
4007 bctl->stat.completed++;
4008 spin_unlock(&fs_info->balance_lock);
4011 if (found_key.offset == 0)
4013 key.offset = found_key.offset - 1;
4017 btrfs_release_path(path);
4022 btrfs_free_path(path);
4023 if (enospc_errors) {
4024 btrfs_info(fs_info, "%d enospc errors during balance",
4034 * alloc_profile_is_valid - see if a given profile is valid and reduced
4035 * @flags: profile to validate
4036 * @extended: if true @flags is treated as an extended profile
4038 static int alloc_profile_is_valid(u64 flags, int extended)
4040 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
4041 BTRFS_BLOCK_GROUP_PROFILE_MASK);
4043 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
4045 /* 1) check that all other bits are zeroed */
4049 /* 2) see if profile is reduced */
4051 return !extended; /* "0" is valid for usual profiles */
4053 return has_single_bit_set(flags);
4056 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
4058 /* cancel requested || normal exit path */
4059 return atomic_read(&fs_info->balance_cancel_req) ||
4060 (atomic_read(&fs_info->balance_pause_req) == 0 &&
4061 atomic_read(&fs_info->balance_cancel_req) == 0);
4065 * Validate target profile against allowed profiles and return true if it's OK.
4066 * Otherwise print the error message and return false.
4068 static inline int validate_convert_profile(struct btrfs_fs_info *fs_info,
4069 const struct btrfs_balance_args *bargs,
4070 u64 allowed, const char *type)
4072 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
4075 if (fs_info->sectorsize < PAGE_SIZE &&
4076 bargs->target & BTRFS_BLOCK_GROUP_RAID56_MASK) {
4078 "RAID56 is not yet supported for sectorsize %u with page size %lu",
4079 fs_info->sectorsize, PAGE_SIZE);
4082 /* Profile is valid and does not have bits outside of the allowed set */
4083 if (alloc_profile_is_valid(bargs->target, 1) &&
4084 (bargs->target & ~allowed) == 0)
4087 btrfs_err(fs_info, "balance: invalid convert %s profile %s",
4088 type, btrfs_bg_type_to_raid_name(bargs->target));
4093 * Fill @buf with textual description of balance filter flags @bargs, up to
4094 * @size_buf including the terminating null. The output may be trimmed if it
4095 * does not fit into the provided buffer.
4097 static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf,
4101 u32 size_bp = size_buf;
4103 u64 flags = bargs->flags;
4104 char tmp_buf[128] = {'\0'};
4109 #define CHECK_APPEND_NOARG(a) \
4111 ret = snprintf(bp, size_bp, (a)); \
4112 if (ret < 0 || ret >= size_bp) \
4113 goto out_overflow; \
4118 #define CHECK_APPEND_1ARG(a, v1) \
4120 ret = snprintf(bp, size_bp, (a), (v1)); \
4121 if (ret < 0 || ret >= size_bp) \
4122 goto out_overflow; \
4127 #define CHECK_APPEND_2ARG(a, v1, v2) \
4129 ret = snprintf(bp, size_bp, (a), (v1), (v2)); \
4130 if (ret < 0 || ret >= size_bp) \
4131 goto out_overflow; \
4136 if (flags & BTRFS_BALANCE_ARGS_CONVERT)
4137 CHECK_APPEND_1ARG("convert=%s,",
4138 btrfs_bg_type_to_raid_name(bargs->target));
4140 if (flags & BTRFS_BALANCE_ARGS_SOFT)
4141 CHECK_APPEND_NOARG("soft,");
4143 if (flags & BTRFS_BALANCE_ARGS_PROFILES) {
4144 btrfs_describe_block_groups(bargs->profiles, tmp_buf,
4146 CHECK_APPEND_1ARG("profiles=%s,", tmp_buf);
4149 if (flags & BTRFS_BALANCE_ARGS_USAGE)
4150 CHECK_APPEND_1ARG("usage=%llu,", bargs->usage);
4152 if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE)
4153 CHECK_APPEND_2ARG("usage=%u..%u,",
4154 bargs->usage_min, bargs->usage_max);
4156 if (flags & BTRFS_BALANCE_ARGS_DEVID)
4157 CHECK_APPEND_1ARG("devid=%llu,", bargs->devid);
4159 if (flags & BTRFS_BALANCE_ARGS_DRANGE)
4160 CHECK_APPEND_2ARG("drange=%llu..%llu,",
4161 bargs->pstart, bargs->pend);
4163 if (flags & BTRFS_BALANCE_ARGS_VRANGE)
4164 CHECK_APPEND_2ARG("vrange=%llu..%llu,",
4165 bargs->vstart, bargs->vend);
4167 if (flags & BTRFS_BALANCE_ARGS_LIMIT)
4168 CHECK_APPEND_1ARG("limit=%llu,", bargs->limit);
4170 if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)
4171 CHECK_APPEND_2ARG("limit=%u..%u,",
4172 bargs->limit_min, bargs->limit_max);
4174 if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE)
4175 CHECK_APPEND_2ARG("stripes=%u..%u,",
4176 bargs->stripes_min, bargs->stripes_max);
4178 #undef CHECK_APPEND_2ARG
4179 #undef CHECK_APPEND_1ARG
4180 #undef CHECK_APPEND_NOARG
4184 if (size_bp < size_buf)
4185 buf[size_buf - size_bp - 1] = '\0'; /* remove last , */
4190 static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info)
4192 u32 size_buf = 1024;
4193 char tmp_buf[192] = {'\0'};
4196 u32 size_bp = size_buf;
4198 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4200 buf = kzalloc(size_buf, GFP_KERNEL);
4206 #define CHECK_APPEND_1ARG(a, v1) \
4208 ret = snprintf(bp, size_bp, (a), (v1)); \
4209 if (ret < 0 || ret >= size_bp) \
4210 goto out_overflow; \
4215 if (bctl->flags & BTRFS_BALANCE_FORCE)
4216 CHECK_APPEND_1ARG("%s", "-f ");
4218 if (bctl->flags & BTRFS_BALANCE_DATA) {
4219 describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf));
4220 CHECK_APPEND_1ARG("-d%s ", tmp_buf);
4223 if (bctl->flags & BTRFS_BALANCE_METADATA) {
4224 describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf));
4225 CHECK_APPEND_1ARG("-m%s ", tmp_buf);
4228 if (bctl->flags & BTRFS_BALANCE_SYSTEM) {
4229 describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf));
4230 CHECK_APPEND_1ARG("-s%s ", tmp_buf);
4233 #undef CHECK_APPEND_1ARG
4237 if (size_bp < size_buf)
4238 buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */
4239 btrfs_info(fs_info, "balance: %s %s",
4240 (bctl->flags & BTRFS_BALANCE_RESUME) ?
4241 "resume" : "start", buf);
4247 * Should be called with balance mutexe held
4249 int btrfs_balance(struct btrfs_fs_info *fs_info,
4250 struct btrfs_balance_control *bctl,
4251 struct btrfs_ioctl_balance_args *bargs)
4253 u64 meta_target, data_target;
4259 bool reducing_redundancy;
4262 if (btrfs_fs_closing(fs_info) ||
4263 atomic_read(&fs_info->balance_pause_req) ||
4264 btrfs_should_cancel_balance(fs_info)) {
4269 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
4270 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
4274 * In case of mixed groups both data and meta should be picked,
4275 * and identical options should be given for both of them.
4277 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
4278 if (mixed && (bctl->flags & allowed)) {
4279 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
4280 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
4281 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
4283 "balance: mixed groups data and metadata options must be the same");
4290 * rw_devices will not change at the moment, device add/delete/replace
4293 num_devices = fs_info->fs_devices->rw_devices;
4296 * SINGLE profile on-disk has no profile bit, but in-memory we have a
4297 * special bit for it, to make it easier to distinguish. Thus we need
4298 * to set it manually, or balance would refuse the profile.
4300 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
4301 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++)
4302 if (num_devices >= btrfs_raid_array[i].devs_min)
4303 allowed |= btrfs_raid_array[i].bg_flag;
4305 if (!validate_convert_profile(fs_info, &bctl->data, allowed, "data") ||
4306 !validate_convert_profile(fs_info, &bctl->meta, allowed, "metadata") ||
4307 !validate_convert_profile(fs_info, &bctl->sys, allowed, "system")) {
4313 * Allow to reduce metadata or system integrity only if force set for
4314 * profiles with redundancy (copies, parity)
4317 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) {
4318 if (btrfs_raid_array[i].ncopies >= 2 ||
4319 btrfs_raid_array[i].tolerated_failures >= 1)
4320 allowed |= btrfs_raid_array[i].bg_flag;
4323 seq = read_seqbegin(&fs_info->profiles_lock);
4325 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4326 (fs_info->avail_system_alloc_bits & allowed) &&
4327 !(bctl->sys.target & allowed)) ||
4328 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4329 (fs_info->avail_metadata_alloc_bits & allowed) &&
4330 !(bctl->meta.target & allowed)))
4331 reducing_redundancy = true;
4333 reducing_redundancy = false;
4335 /* if we're not converting, the target field is uninitialized */
4336 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4337 bctl->meta.target : fs_info->avail_metadata_alloc_bits;
4338 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4339 bctl->data.target : fs_info->avail_data_alloc_bits;
4340 } while (read_seqretry(&fs_info->profiles_lock, seq));
4342 if (reducing_redundancy) {
4343 if (bctl->flags & BTRFS_BALANCE_FORCE) {
4345 "balance: force reducing metadata redundancy");
4348 "balance: reduces metadata redundancy, use --force if you want this");
4354 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <
4355 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {
4357 "balance: metadata profile %s has lower redundancy than data profile %s",
4358 btrfs_bg_type_to_raid_name(meta_target),
4359 btrfs_bg_type_to_raid_name(data_target));
4362 ret = insert_balance_item(fs_info, bctl);
4363 if (ret && ret != -EEXIST)
4366 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
4367 BUG_ON(ret == -EEXIST);
4368 BUG_ON(fs_info->balance_ctl);
4369 spin_lock(&fs_info->balance_lock);
4370 fs_info->balance_ctl = bctl;
4371 spin_unlock(&fs_info->balance_lock);
4373 BUG_ON(ret != -EEXIST);
4374 spin_lock(&fs_info->balance_lock);
4375 update_balance_args(bctl);
4376 spin_unlock(&fs_info->balance_lock);
4379 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4380 set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4381 describe_balance_start_or_resume(fs_info);
4382 mutex_unlock(&fs_info->balance_mutex);
4384 ret = __btrfs_balance(fs_info);
4386 mutex_lock(&fs_info->balance_mutex);
4387 if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req)) {
4388 btrfs_info(fs_info, "balance: paused");
4389 btrfs_exclop_balance(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED);
4392 * Balance can be canceled by:
4394 * - Regular cancel request
4395 * Then ret == -ECANCELED and balance_cancel_req > 0
4397 * - Fatal signal to "btrfs" process
4398 * Either the signal caught by wait_reserve_ticket() and callers
4399 * got -EINTR, or caught by btrfs_should_cancel_balance() and
4401 * Either way, in this case balance_cancel_req = 0, and
4402 * ret == -EINTR or ret == -ECANCELED.
4404 * So here we only check the return value to catch canceled balance.
4406 else if (ret == -ECANCELED || ret == -EINTR)
4407 btrfs_info(fs_info, "balance: canceled");
4409 btrfs_info(fs_info, "balance: ended with status: %d", ret);
4411 clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4414 memset(bargs, 0, sizeof(*bargs));
4415 btrfs_update_ioctl_balance_args(fs_info, bargs);
4418 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
4419 balance_need_close(fs_info)) {
4420 reset_balance_state(fs_info);
4421 btrfs_exclop_finish(fs_info);
4424 wake_up(&fs_info->balance_wait_q);
4428 if (bctl->flags & BTRFS_BALANCE_RESUME)
4429 reset_balance_state(fs_info);
4432 btrfs_exclop_finish(fs_info);
4437 static int balance_kthread(void *data)
4439 struct btrfs_fs_info *fs_info = data;
4442 mutex_lock(&fs_info->balance_mutex);
4443 if (fs_info->balance_ctl)
4444 ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL);
4445 mutex_unlock(&fs_info->balance_mutex);
4450 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
4452 struct task_struct *tsk;
4454 mutex_lock(&fs_info->balance_mutex);
4455 if (!fs_info->balance_ctl) {
4456 mutex_unlock(&fs_info->balance_mutex);
4459 mutex_unlock(&fs_info->balance_mutex);
4461 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
4462 btrfs_info(fs_info, "balance: resume skipped");
4466 spin_lock(&fs_info->super_lock);
4467 ASSERT(fs_info->exclusive_operation == BTRFS_EXCLOP_BALANCE_PAUSED);
4468 fs_info->exclusive_operation = BTRFS_EXCLOP_BALANCE;
4469 spin_unlock(&fs_info->super_lock);
4471 * A ro->rw remount sequence should continue with the paused balance
4472 * regardless of who pauses it, system or the user as of now, so set
4475 spin_lock(&fs_info->balance_lock);
4476 fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME;
4477 spin_unlock(&fs_info->balance_lock);
4479 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
4480 return PTR_ERR_OR_ZERO(tsk);
4483 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
4485 struct btrfs_balance_control *bctl;
4486 struct btrfs_balance_item *item;
4487 struct btrfs_disk_balance_args disk_bargs;
4488 struct btrfs_path *path;
4489 struct extent_buffer *leaf;
4490 struct btrfs_key key;
4493 path = btrfs_alloc_path();
4497 key.objectid = BTRFS_BALANCE_OBJECTID;
4498 key.type = BTRFS_TEMPORARY_ITEM_KEY;
4501 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4504 if (ret > 0) { /* ret = -ENOENT; */
4509 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4515 leaf = path->nodes[0];
4516 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
4518 bctl->flags = btrfs_balance_flags(leaf, item);
4519 bctl->flags |= BTRFS_BALANCE_RESUME;
4521 btrfs_balance_data(leaf, item, &disk_bargs);
4522 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4523 btrfs_balance_meta(leaf, item, &disk_bargs);
4524 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4525 btrfs_balance_sys(leaf, item, &disk_bargs);
4526 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4529 * This should never happen, as the paused balance state is recovered
4530 * during mount without any chance of other exclusive ops to collide.
4532 * This gives the exclusive op status to balance and keeps in paused
4533 * state until user intervention (cancel or umount). If the ownership
4534 * cannot be assigned, show a message but do not fail. The balance
4535 * is in a paused state and must have fs_info::balance_ctl properly
4538 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED))
4540 "balance: cannot set exclusive op status, resume manually");
4542 btrfs_release_path(path);
4544 mutex_lock(&fs_info->balance_mutex);
4545 BUG_ON(fs_info->balance_ctl);
4546 spin_lock(&fs_info->balance_lock);
4547 fs_info->balance_ctl = bctl;
4548 spin_unlock(&fs_info->balance_lock);
4549 mutex_unlock(&fs_info->balance_mutex);
4551 btrfs_free_path(path);
4555 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4559 mutex_lock(&fs_info->balance_mutex);
4560 if (!fs_info->balance_ctl) {
4561 mutex_unlock(&fs_info->balance_mutex);
4565 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4566 atomic_inc(&fs_info->balance_pause_req);
4567 mutex_unlock(&fs_info->balance_mutex);
4569 wait_event(fs_info->balance_wait_q,
4570 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4572 mutex_lock(&fs_info->balance_mutex);
4573 /* we are good with balance_ctl ripped off from under us */
4574 BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4575 atomic_dec(&fs_info->balance_pause_req);
4580 mutex_unlock(&fs_info->balance_mutex);
4584 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4586 mutex_lock(&fs_info->balance_mutex);
4587 if (!fs_info->balance_ctl) {
4588 mutex_unlock(&fs_info->balance_mutex);
4593 * A paused balance with the item stored on disk can be resumed at
4594 * mount time if the mount is read-write. Otherwise it's still paused
4595 * and we must not allow cancelling as it deletes the item.
4597 if (sb_rdonly(fs_info->sb)) {
4598 mutex_unlock(&fs_info->balance_mutex);
4602 atomic_inc(&fs_info->balance_cancel_req);
4604 * if we are running just wait and return, balance item is
4605 * deleted in btrfs_balance in this case
4607 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4608 mutex_unlock(&fs_info->balance_mutex);
4609 wait_event(fs_info->balance_wait_q,
4610 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4611 mutex_lock(&fs_info->balance_mutex);
4613 mutex_unlock(&fs_info->balance_mutex);
4615 * Lock released to allow other waiters to continue, we'll
4616 * reexamine the status again.
4618 mutex_lock(&fs_info->balance_mutex);
4620 if (fs_info->balance_ctl) {
4621 reset_balance_state(fs_info);
4622 btrfs_exclop_finish(fs_info);
4623 btrfs_info(fs_info, "balance: canceled");
4627 BUG_ON(fs_info->balance_ctl ||
4628 test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4629 atomic_dec(&fs_info->balance_cancel_req);
4630 mutex_unlock(&fs_info->balance_mutex);
4634 int btrfs_uuid_scan_kthread(void *data)
4636 struct btrfs_fs_info *fs_info = data;
4637 struct btrfs_root *root = fs_info->tree_root;
4638 struct btrfs_key key;
4639 struct btrfs_path *path = NULL;
4641 struct extent_buffer *eb;
4643 struct btrfs_root_item root_item;
4645 struct btrfs_trans_handle *trans = NULL;
4646 bool closing = false;
4648 path = btrfs_alloc_path();
4655 key.type = BTRFS_ROOT_ITEM_KEY;
4659 if (btrfs_fs_closing(fs_info)) {
4663 ret = btrfs_search_forward(root, &key, path,
4664 BTRFS_OLDEST_GENERATION);
4671 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4672 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4673 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4674 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4677 eb = path->nodes[0];
4678 slot = path->slots[0];
4679 item_size = btrfs_item_size(eb, slot);
4680 if (item_size < sizeof(root_item))
4683 read_extent_buffer(eb, &root_item,
4684 btrfs_item_ptr_offset(eb, slot),
4685 (int)sizeof(root_item));
4686 if (btrfs_root_refs(&root_item) == 0)
4689 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4690 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4694 btrfs_release_path(path);
4696 * 1 - subvol uuid item
4697 * 1 - received_subvol uuid item
4699 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4700 if (IS_ERR(trans)) {
4701 ret = PTR_ERR(trans);
4709 btrfs_release_path(path);
4710 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4711 ret = btrfs_uuid_tree_add(trans, root_item.uuid,
4712 BTRFS_UUID_KEY_SUBVOL,
4715 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4721 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4722 ret = btrfs_uuid_tree_add(trans,
4723 root_item.received_uuid,
4724 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4727 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4734 btrfs_release_path(path);
4736 ret = btrfs_end_transaction(trans);
4742 if (key.offset < (u64)-1) {
4744 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4746 key.type = BTRFS_ROOT_ITEM_KEY;
4747 } else if (key.objectid < (u64)-1) {
4749 key.type = BTRFS_ROOT_ITEM_KEY;
4758 btrfs_free_path(path);
4759 if (trans && !IS_ERR(trans))
4760 btrfs_end_transaction(trans);
4762 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4764 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4765 up(&fs_info->uuid_tree_rescan_sem);
4769 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4771 struct btrfs_trans_handle *trans;
4772 struct btrfs_root *tree_root = fs_info->tree_root;
4773 struct btrfs_root *uuid_root;
4774 struct task_struct *task;
4781 trans = btrfs_start_transaction(tree_root, 2);
4783 return PTR_ERR(trans);
4785 uuid_root = btrfs_create_tree(trans, BTRFS_UUID_TREE_OBJECTID);
4786 if (IS_ERR(uuid_root)) {
4787 ret = PTR_ERR(uuid_root);
4788 btrfs_abort_transaction(trans, ret);
4789 btrfs_end_transaction(trans);
4793 fs_info->uuid_root = uuid_root;
4795 ret = btrfs_commit_transaction(trans);
4799 down(&fs_info->uuid_tree_rescan_sem);
4800 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4802 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4803 btrfs_warn(fs_info, "failed to start uuid_scan task");
4804 up(&fs_info->uuid_tree_rescan_sem);
4805 return PTR_ERR(task);
4812 * shrinking a device means finding all of the device extents past
4813 * the new size, and then following the back refs to the chunks.
4814 * The chunk relocation code actually frees the device extent
4816 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4818 struct btrfs_fs_info *fs_info = device->fs_info;
4819 struct btrfs_root *root = fs_info->dev_root;
4820 struct btrfs_trans_handle *trans;
4821 struct btrfs_dev_extent *dev_extent = NULL;
4822 struct btrfs_path *path;
4828 bool retried = false;
4829 struct extent_buffer *l;
4830 struct btrfs_key key;
4831 struct btrfs_super_block *super_copy = fs_info->super_copy;
4832 u64 old_total = btrfs_super_total_bytes(super_copy);
4833 u64 old_size = btrfs_device_get_total_bytes(device);
4837 new_size = round_down(new_size, fs_info->sectorsize);
4839 diff = round_down(old_size - new_size, fs_info->sectorsize);
4841 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4844 path = btrfs_alloc_path();
4848 path->reada = READA_BACK;
4850 trans = btrfs_start_transaction(root, 0);
4851 if (IS_ERR(trans)) {
4852 btrfs_free_path(path);
4853 return PTR_ERR(trans);
4856 mutex_lock(&fs_info->chunk_mutex);
4858 btrfs_device_set_total_bytes(device, new_size);
4859 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4860 device->fs_devices->total_rw_bytes -= diff;
4861 atomic64_sub(diff, &fs_info->free_chunk_space);
4865 * Once the device's size has been set to the new size, ensure all
4866 * in-memory chunks are synced to disk so that the loop below sees them
4867 * and relocates them accordingly.
4869 if (contains_pending_extent(device, &start, diff)) {
4870 mutex_unlock(&fs_info->chunk_mutex);
4871 ret = btrfs_commit_transaction(trans);
4875 mutex_unlock(&fs_info->chunk_mutex);
4876 btrfs_end_transaction(trans);
4880 key.objectid = device->devid;
4881 key.offset = (u64)-1;
4882 key.type = BTRFS_DEV_EXTENT_KEY;
4885 mutex_lock(&fs_info->reclaim_bgs_lock);
4886 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4888 mutex_unlock(&fs_info->reclaim_bgs_lock);
4892 ret = btrfs_previous_item(root, path, 0, key.type);
4894 mutex_unlock(&fs_info->reclaim_bgs_lock);
4898 btrfs_release_path(path);
4903 slot = path->slots[0];
4904 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4906 if (key.objectid != device->devid) {
4907 mutex_unlock(&fs_info->reclaim_bgs_lock);
4908 btrfs_release_path(path);
4912 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4913 length = btrfs_dev_extent_length(l, dev_extent);
4915 if (key.offset + length <= new_size) {
4916 mutex_unlock(&fs_info->reclaim_bgs_lock);
4917 btrfs_release_path(path);
4921 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4922 btrfs_release_path(path);
4925 * We may be relocating the only data chunk we have,
4926 * which could potentially end up with losing data's
4927 * raid profile, so lets allocate an empty one in
4930 ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset);
4932 mutex_unlock(&fs_info->reclaim_bgs_lock);
4936 ret = btrfs_relocate_chunk(fs_info, chunk_offset);
4937 mutex_unlock(&fs_info->reclaim_bgs_lock);
4938 if (ret == -ENOSPC) {
4941 if (ret == -ETXTBSY) {
4943 "could not shrink block group %llu due to active swapfile",
4948 } while (key.offset-- > 0);
4950 if (failed && !retried) {
4954 } else if (failed && retried) {
4959 /* Shrinking succeeded, else we would be at "done". */
4960 trans = btrfs_start_transaction(root, 0);
4961 if (IS_ERR(trans)) {
4962 ret = PTR_ERR(trans);
4966 mutex_lock(&fs_info->chunk_mutex);
4967 /* Clear all state bits beyond the shrunk device size */
4968 clear_extent_bits(&device->alloc_state, new_size, (u64)-1,
4971 btrfs_device_set_disk_total_bytes(device, new_size);
4972 if (list_empty(&device->post_commit_list))
4973 list_add_tail(&device->post_commit_list,
4974 &trans->transaction->dev_update_list);
4976 WARN_ON(diff > old_total);
4977 btrfs_set_super_total_bytes(super_copy,
4978 round_down(old_total - diff, fs_info->sectorsize));
4979 mutex_unlock(&fs_info->chunk_mutex);
4981 btrfs_reserve_chunk_metadata(trans, false);
4982 /* Now btrfs_update_device() will change the on-disk size. */
4983 ret = btrfs_update_device(trans, device);
4984 btrfs_trans_release_chunk_metadata(trans);
4986 btrfs_abort_transaction(trans, ret);
4987 btrfs_end_transaction(trans);
4989 ret = btrfs_commit_transaction(trans);
4992 btrfs_free_path(path);
4994 mutex_lock(&fs_info->chunk_mutex);
4995 btrfs_device_set_total_bytes(device, old_size);
4996 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
4997 device->fs_devices->total_rw_bytes += diff;
4998 atomic64_add(diff, &fs_info->free_chunk_space);
4999 mutex_unlock(&fs_info->chunk_mutex);
5004 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
5005 struct btrfs_key *key,
5006 struct btrfs_chunk *chunk, int item_size)
5008 struct btrfs_super_block *super_copy = fs_info->super_copy;
5009 struct btrfs_disk_key disk_key;
5013 lockdep_assert_held(&fs_info->chunk_mutex);
5015 array_size = btrfs_super_sys_array_size(super_copy);
5016 if (array_size + item_size + sizeof(disk_key)
5017 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
5020 ptr = super_copy->sys_chunk_array + array_size;
5021 btrfs_cpu_key_to_disk(&disk_key, key);
5022 memcpy(ptr, &disk_key, sizeof(disk_key));
5023 ptr += sizeof(disk_key);
5024 memcpy(ptr, chunk, item_size);
5025 item_size += sizeof(disk_key);
5026 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
5032 * sort the devices in descending order by max_avail, total_avail
5034 static int btrfs_cmp_device_info(const void *a, const void *b)
5036 const struct btrfs_device_info *di_a = a;
5037 const struct btrfs_device_info *di_b = b;
5039 if (di_a->max_avail > di_b->max_avail)
5041 if (di_a->max_avail < di_b->max_avail)
5043 if (di_a->total_avail > di_b->total_avail)
5045 if (di_a->total_avail < di_b->total_avail)
5050 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
5052 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
5055 btrfs_set_fs_incompat(info, RAID56);
5058 static void check_raid1c34_incompat_flag(struct btrfs_fs_info *info, u64 type)
5060 if (!(type & (BTRFS_BLOCK_GROUP_RAID1C3 | BTRFS_BLOCK_GROUP_RAID1C4)))
5063 btrfs_set_fs_incompat(info, RAID1C34);
5067 * Structure used internally for btrfs_create_chunk() function.
5068 * Wraps needed parameters.
5070 struct alloc_chunk_ctl {
5073 /* Total number of stripes to allocate */
5075 /* sub_stripes info for map */
5077 /* Stripes per device */
5079 /* Maximum number of devices to use */
5081 /* Minimum number of devices to use */
5083 /* ndevs has to be a multiple of this */
5085 /* Number of copies */
5087 /* Number of stripes worth of bytes to store parity information */
5089 u64 max_stripe_size;
5097 static void init_alloc_chunk_ctl_policy_regular(
5098 struct btrfs_fs_devices *fs_devices,
5099 struct alloc_chunk_ctl *ctl)
5101 u64 type = ctl->type;
5103 if (type & BTRFS_BLOCK_GROUP_DATA) {
5104 ctl->max_stripe_size = SZ_1G;
5105 ctl->max_chunk_size = BTRFS_MAX_DATA_CHUNK_SIZE;
5106 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
5107 /* For larger filesystems, use larger metadata chunks */
5108 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
5109 ctl->max_stripe_size = SZ_1G;
5111 ctl->max_stripe_size = SZ_256M;
5112 ctl->max_chunk_size = ctl->max_stripe_size;
5113 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
5114 ctl->max_stripe_size = SZ_32M;
5115 ctl->max_chunk_size = 2 * ctl->max_stripe_size;
5116 ctl->devs_max = min_t(int, ctl->devs_max,
5117 BTRFS_MAX_DEVS_SYS_CHUNK);
5122 /* We don't want a chunk larger than 10% of writable space */
5123 ctl->max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
5124 ctl->max_chunk_size);
5125 ctl->dev_extent_min = BTRFS_STRIPE_LEN * ctl->dev_stripes;
5128 static void init_alloc_chunk_ctl_policy_zoned(
5129 struct btrfs_fs_devices *fs_devices,
5130 struct alloc_chunk_ctl *ctl)
5132 u64 zone_size = fs_devices->fs_info->zone_size;
5134 int min_num_stripes = ctl->devs_min * ctl->dev_stripes;
5135 int min_data_stripes = (min_num_stripes - ctl->nparity) / ctl->ncopies;
5136 u64 min_chunk_size = min_data_stripes * zone_size;
5137 u64 type = ctl->type;
5139 ctl->max_stripe_size = zone_size;
5140 if (type & BTRFS_BLOCK_GROUP_DATA) {
5141 ctl->max_chunk_size = round_down(BTRFS_MAX_DATA_CHUNK_SIZE,
5143 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
5144 ctl->max_chunk_size = ctl->max_stripe_size;
5145 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
5146 ctl->max_chunk_size = 2 * ctl->max_stripe_size;
5147 ctl->devs_max = min_t(int, ctl->devs_max,
5148 BTRFS_MAX_DEVS_SYS_CHUNK);
5153 /* We don't want a chunk larger than 10% of writable space */
5154 limit = max(round_down(div_factor(fs_devices->total_rw_bytes, 1),
5157 ctl->max_chunk_size = min(limit, ctl->max_chunk_size);
5158 ctl->dev_extent_min = zone_size * ctl->dev_stripes;
5161 static void init_alloc_chunk_ctl(struct btrfs_fs_devices *fs_devices,
5162 struct alloc_chunk_ctl *ctl)
5164 int index = btrfs_bg_flags_to_raid_index(ctl->type);
5166 ctl->sub_stripes = btrfs_raid_array[index].sub_stripes;
5167 ctl->dev_stripes = btrfs_raid_array[index].dev_stripes;
5168 ctl->devs_max = btrfs_raid_array[index].devs_max;
5170 ctl->devs_max = BTRFS_MAX_DEVS(fs_devices->fs_info);
5171 ctl->devs_min = btrfs_raid_array[index].devs_min;
5172 ctl->devs_increment = btrfs_raid_array[index].devs_increment;
5173 ctl->ncopies = btrfs_raid_array[index].ncopies;
5174 ctl->nparity = btrfs_raid_array[index].nparity;
5177 switch (fs_devices->chunk_alloc_policy) {
5178 case BTRFS_CHUNK_ALLOC_REGULAR:
5179 init_alloc_chunk_ctl_policy_regular(fs_devices, ctl);
5181 case BTRFS_CHUNK_ALLOC_ZONED:
5182 init_alloc_chunk_ctl_policy_zoned(fs_devices, ctl);
5189 static int gather_device_info(struct btrfs_fs_devices *fs_devices,
5190 struct alloc_chunk_ctl *ctl,
5191 struct btrfs_device_info *devices_info)
5193 struct btrfs_fs_info *info = fs_devices->fs_info;
5194 struct btrfs_device *device;
5196 u64 dev_extent_want = ctl->max_stripe_size * ctl->dev_stripes;
5203 * in the first pass through the devices list, we gather information
5204 * about the available holes on each device.
5206 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
5207 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
5209 "BTRFS: read-only device in alloc_list\n");
5213 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
5214 &device->dev_state) ||
5215 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
5218 if (device->total_bytes > device->bytes_used)
5219 total_avail = device->total_bytes - device->bytes_used;
5223 /* If there is no space on this device, skip it. */
5224 if (total_avail < ctl->dev_extent_min)
5227 ret = find_free_dev_extent(device, dev_extent_want, &dev_offset,
5229 if (ret && ret != -ENOSPC)
5233 max_avail = dev_extent_want;
5235 if (max_avail < ctl->dev_extent_min) {
5236 if (btrfs_test_opt(info, ENOSPC_DEBUG))
5238 "%s: devid %llu has no free space, have=%llu want=%llu",
5239 __func__, device->devid, max_avail,
5240 ctl->dev_extent_min);
5244 if (ndevs == fs_devices->rw_devices) {
5245 WARN(1, "%s: found more than %llu devices\n",
5246 __func__, fs_devices->rw_devices);
5249 devices_info[ndevs].dev_offset = dev_offset;
5250 devices_info[ndevs].max_avail = max_avail;
5251 devices_info[ndevs].total_avail = total_avail;
5252 devices_info[ndevs].dev = device;
5258 * now sort the devices by hole size / available space
5260 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
5261 btrfs_cmp_device_info, NULL);
5266 static int decide_stripe_size_regular(struct alloc_chunk_ctl *ctl,
5267 struct btrfs_device_info *devices_info)
5269 /* Number of stripes that count for block group size */
5273 * The primary goal is to maximize the number of stripes, so use as
5274 * many devices as possible, even if the stripes are not maximum sized.
5276 * The DUP profile stores more than one stripe per device, the
5277 * max_avail is the total size so we have to adjust.
5279 ctl->stripe_size = div_u64(devices_info[ctl->ndevs - 1].max_avail,
5281 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5283 /* This will have to be fixed for RAID1 and RAID10 over more drives */
5284 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5287 * Use the number of data stripes to figure out how big this chunk is
5288 * really going to be in terms of logical address space, and compare
5289 * that answer with the max chunk size. If it's higher, we try to
5290 * reduce stripe_size.
5292 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) {
5294 * Reduce stripe_size, round it up to a 16MB boundary again and
5295 * then use it, unless it ends up being even bigger than the
5296 * previous value we had already.
5298 ctl->stripe_size = min(round_up(div_u64(ctl->max_chunk_size,
5299 data_stripes), SZ_16M),
5303 /* Align to BTRFS_STRIPE_LEN */
5304 ctl->stripe_size = round_down(ctl->stripe_size, BTRFS_STRIPE_LEN);
5305 ctl->chunk_size = ctl->stripe_size * data_stripes;
5310 static int decide_stripe_size_zoned(struct alloc_chunk_ctl *ctl,
5311 struct btrfs_device_info *devices_info)
5313 u64 zone_size = devices_info[0].dev->zone_info->zone_size;
5314 /* Number of stripes that count for block group size */
5318 * It should hold because:
5319 * dev_extent_min == dev_extent_want == zone_size * dev_stripes
5321 ASSERT(devices_info[ctl->ndevs - 1].max_avail == ctl->dev_extent_min);
5323 ctl->stripe_size = zone_size;
5324 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5325 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5327 /* stripe_size is fixed in zoned filesysmte. Reduce ndevs instead. */
5328 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) {
5329 ctl->ndevs = div_u64(div_u64(ctl->max_chunk_size * ctl->ncopies,
5330 ctl->stripe_size) + ctl->nparity,
5332 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5333 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5334 ASSERT(ctl->stripe_size * data_stripes <= ctl->max_chunk_size);
5337 ctl->chunk_size = ctl->stripe_size * data_stripes;
5342 static int decide_stripe_size(struct btrfs_fs_devices *fs_devices,
5343 struct alloc_chunk_ctl *ctl,
5344 struct btrfs_device_info *devices_info)
5346 struct btrfs_fs_info *info = fs_devices->fs_info;
5349 * Round down to number of usable stripes, devs_increment can be any
5350 * number so we can't use round_down() that requires power of 2, while
5351 * rounddown is safe.
5353 ctl->ndevs = rounddown(ctl->ndevs, ctl->devs_increment);
5355 if (ctl->ndevs < ctl->devs_min) {
5356 if (btrfs_test_opt(info, ENOSPC_DEBUG)) {
5358 "%s: not enough devices with free space: have=%d minimum required=%d",
5359 __func__, ctl->ndevs, ctl->devs_min);
5364 ctl->ndevs = min(ctl->ndevs, ctl->devs_max);
5366 switch (fs_devices->chunk_alloc_policy) {
5367 case BTRFS_CHUNK_ALLOC_REGULAR:
5368 return decide_stripe_size_regular(ctl, devices_info);
5369 case BTRFS_CHUNK_ALLOC_ZONED:
5370 return decide_stripe_size_zoned(ctl, devices_info);
5376 static struct btrfs_block_group *create_chunk(struct btrfs_trans_handle *trans,
5377 struct alloc_chunk_ctl *ctl,
5378 struct btrfs_device_info *devices_info)
5380 struct btrfs_fs_info *info = trans->fs_info;
5381 struct map_lookup *map = NULL;
5382 struct extent_map_tree *em_tree;
5383 struct btrfs_block_group *block_group;
5384 struct extent_map *em;
5385 u64 start = ctl->start;
5386 u64 type = ctl->type;
5391 map = kmalloc(map_lookup_size(ctl->num_stripes), GFP_NOFS);
5393 return ERR_PTR(-ENOMEM);
5394 map->num_stripes = ctl->num_stripes;
5396 for (i = 0; i < ctl->ndevs; ++i) {
5397 for (j = 0; j < ctl->dev_stripes; ++j) {
5398 int s = i * ctl->dev_stripes + j;
5399 map->stripes[s].dev = devices_info[i].dev;
5400 map->stripes[s].physical = devices_info[i].dev_offset +
5401 j * ctl->stripe_size;
5404 map->stripe_len = BTRFS_STRIPE_LEN;
5405 map->io_align = BTRFS_STRIPE_LEN;
5406 map->io_width = BTRFS_STRIPE_LEN;
5408 map->sub_stripes = ctl->sub_stripes;
5410 trace_btrfs_chunk_alloc(info, map, start, ctl->chunk_size);
5412 em = alloc_extent_map();
5415 return ERR_PTR(-ENOMEM);
5417 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
5418 em->map_lookup = map;
5420 em->len = ctl->chunk_size;
5421 em->block_start = 0;
5422 em->block_len = em->len;
5423 em->orig_block_len = ctl->stripe_size;
5425 em_tree = &info->mapping_tree;
5426 write_lock(&em_tree->lock);
5427 ret = add_extent_mapping(em_tree, em, 0);
5429 write_unlock(&em_tree->lock);
5430 free_extent_map(em);
5431 return ERR_PTR(ret);
5433 write_unlock(&em_tree->lock);
5435 block_group = btrfs_make_block_group(trans, 0, type, start, ctl->chunk_size);
5436 if (IS_ERR(block_group))
5437 goto error_del_extent;
5439 for (i = 0; i < map->num_stripes; i++) {
5440 struct btrfs_device *dev = map->stripes[i].dev;
5442 btrfs_device_set_bytes_used(dev,
5443 dev->bytes_used + ctl->stripe_size);
5444 if (list_empty(&dev->post_commit_list))
5445 list_add_tail(&dev->post_commit_list,
5446 &trans->transaction->dev_update_list);
5449 atomic64_sub(ctl->stripe_size * map->num_stripes,
5450 &info->free_chunk_space);
5452 free_extent_map(em);
5453 check_raid56_incompat_flag(info, type);
5454 check_raid1c34_incompat_flag(info, type);
5459 write_lock(&em_tree->lock);
5460 remove_extent_mapping(em_tree, em);
5461 write_unlock(&em_tree->lock);
5463 /* One for our allocation */
5464 free_extent_map(em);
5465 /* One for the tree reference */
5466 free_extent_map(em);
5471 struct btrfs_block_group *btrfs_create_chunk(struct btrfs_trans_handle *trans,
5474 struct btrfs_fs_info *info = trans->fs_info;
5475 struct btrfs_fs_devices *fs_devices = info->fs_devices;
5476 struct btrfs_device_info *devices_info = NULL;
5477 struct alloc_chunk_ctl ctl;
5478 struct btrfs_block_group *block_group;
5481 lockdep_assert_held(&info->chunk_mutex);
5483 if (!alloc_profile_is_valid(type, 0)) {
5485 return ERR_PTR(-EINVAL);
5488 if (list_empty(&fs_devices->alloc_list)) {
5489 if (btrfs_test_opt(info, ENOSPC_DEBUG))
5490 btrfs_debug(info, "%s: no writable device", __func__);
5491 return ERR_PTR(-ENOSPC);
5494 if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
5495 btrfs_err(info, "invalid chunk type 0x%llx requested", type);
5497 return ERR_PTR(-EINVAL);
5500 ctl.start = find_next_chunk(info);
5502 init_alloc_chunk_ctl(fs_devices, &ctl);
5504 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
5507 return ERR_PTR(-ENOMEM);
5509 ret = gather_device_info(fs_devices, &ctl, devices_info);
5511 block_group = ERR_PTR(ret);
5515 ret = decide_stripe_size(fs_devices, &ctl, devices_info);
5517 block_group = ERR_PTR(ret);
5521 block_group = create_chunk(trans, &ctl, devices_info);
5524 kfree(devices_info);
5529 * This function, btrfs_chunk_alloc_add_chunk_item(), typically belongs to the
5530 * phase 1 of chunk allocation. It belongs to phase 2 only when allocating system
5533 * See the comment at btrfs_chunk_alloc() for details about the chunk allocation
5536 int btrfs_chunk_alloc_add_chunk_item(struct btrfs_trans_handle *trans,
5537 struct btrfs_block_group *bg)
5539 struct btrfs_fs_info *fs_info = trans->fs_info;
5540 struct btrfs_root *chunk_root = fs_info->chunk_root;
5541 struct btrfs_key key;
5542 struct btrfs_chunk *chunk;
5543 struct btrfs_stripe *stripe;
5544 struct extent_map *em;
5545 struct map_lookup *map;
5551 * We take the chunk_mutex for 2 reasons:
5553 * 1) Updates and insertions in the chunk btree must be done while holding
5554 * the chunk_mutex, as well as updating the system chunk array in the
5555 * superblock. See the comment on top of btrfs_chunk_alloc() for the
5558 * 2) To prevent races with the final phase of a device replace operation
5559 * that replaces the device object associated with the map's stripes,
5560 * because the device object's id can change at any time during that
5561 * final phase of the device replace operation
5562 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the
5563 * replaced device and then see it with an ID of BTRFS_DEV_REPLACE_DEVID,
5564 * which would cause a failure when updating the device item, which does
5565 * not exists, or persisting a stripe of the chunk item with such ID.
5566 * Here we can't use the device_list_mutex because our caller already
5567 * has locked the chunk_mutex, and the final phase of device replace
5568 * acquires both mutexes - first the device_list_mutex and then the
5569 * chunk_mutex. Using any of those two mutexes protects us from a
5570 * concurrent device replace.
5572 lockdep_assert_held(&fs_info->chunk_mutex);
5574 em = btrfs_get_chunk_map(fs_info, bg->start, bg->length);
5577 btrfs_abort_transaction(trans, ret);
5581 map = em->map_lookup;
5582 item_size = btrfs_chunk_item_size(map->num_stripes);
5584 chunk = kzalloc(item_size, GFP_NOFS);
5587 btrfs_abort_transaction(trans, ret);
5591 for (i = 0; i < map->num_stripes; i++) {
5592 struct btrfs_device *device = map->stripes[i].dev;
5594 ret = btrfs_update_device(trans, device);
5599 stripe = &chunk->stripe;
5600 for (i = 0; i < map->num_stripes; i++) {
5601 struct btrfs_device *device = map->stripes[i].dev;
5602 const u64 dev_offset = map->stripes[i].physical;
5604 btrfs_set_stack_stripe_devid(stripe, device->devid);
5605 btrfs_set_stack_stripe_offset(stripe, dev_offset);
5606 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
5610 btrfs_set_stack_chunk_length(chunk, bg->length);
5611 btrfs_set_stack_chunk_owner(chunk, BTRFS_EXTENT_TREE_OBJECTID);
5612 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
5613 btrfs_set_stack_chunk_type(chunk, map->type);
5614 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
5615 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
5616 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
5617 btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
5618 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
5620 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
5621 key.type = BTRFS_CHUNK_ITEM_KEY;
5622 key.offset = bg->start;
5624 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
5628 bg->chunk_item_inserted = 1;
5630 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
5631 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
5638 free_extent_map(em);
5642 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans)
5644 struct btrfs_fs_info *fs_info = trans->fs_info;
5646 struct btrfs_block_group *meta_bg;
5647 struct btrfs_block_group *sys_bg;
5650 * When adding a new device for sprouting, the seed device is read-only
5651 * so we must first allocate a metadata and a system chunk. But before
5652 * adding the block group items to the extent, device and chunk btrees,
5655 * 1) Create both chunks without doing any changes to the btrees, as
5656 * otherwise we would get -ENOSPC since the block groups from the
5657 * seed device are read-only;
5659 * 2) Add the device item for the new sprout device - finishing the setup
5660 * of a new block group requires updating the device item in the chunk
5661 * btree, so it must exist when we attempt to do it. The previous step
5662 * ensures this does not fail with -ENOSPC.
5664 * After that we can add the block group items to their btrees:
5665 * update existing device item in the chunk btree, add a new block group
5666 * item to the extent btree, add a new chunk item to the chunk btree and
5667 * finally add the new device extent items to the devices btree.
5670 alloc_profile = btrfs_metadata_alloc_profile(fs_info);
5671 meta_bg = btrfs_create_chunk(trans, alloc_profile);
5672 if (IS_ERR(meta_bg))
5673 return PTR_ERR(meta_bg);
5675 alloc_profile = btrfs_system_alloc_profile(fs_info);
5676 sys_bg = btrfs_create_chunk(trans, alloc_profile);
5678 return PTR_ERR(sys_bg);
5683 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
5685 const int index = btrfs_bg_flags_to_raid_index(map->type);
5687 return btrfs_raid_array[index].tolerated_failures;
5690 bool btrfs_chunk_writeable(struct btrfs_fs_info *fs_info, u64 chunk_offset)
5692 struct extent_map *em;
5693 struct map_lookup *map;
5698 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
5702 map = em->map_lookup;
5703 for (i = 0; i < map->num_stripes; i++) {
5704 if (test_bit(BTRFS_DEV_STATE_MISSING,
5705 &map->stripes[i].dev->dev_state)) {
5709 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE,
5710 &map->stripes[i].dev->dev_state)) {
5717 * If the number of missing devices is larger than max errors, we can
5718 * not write the data into that chunk successfully.
5720 if (miss_ndevs > btrfs_chunk_max_errors(map))
5723 free_extent_map(em);
5727 void btrfs_mapping_tree_free(struct extent_map_tree *tree)
5729 struct extent_map *em;
5732 write_lock(&tree->lock);
5733 em = lookup_extent_mapping(tree, 0, (u64)-1);
5735 remove_extent_mapping(tree, em);
5736 write_unlock(&tree->lock);
5740 free_extent_map(em);
5741 /* once for the tree */
5742 free_extent_map(em);
5746 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5748 struct extent_map *em;
5749 struct map_lookup *map;
5752 em = btrfs_get_chunk_map(fs_info, logical, len);
5755 * We could return errors for these cases, but that could get
5756 * ugly and we'd probably do the same thing which is just not do
5757 * anything else and exit, so return 1 so the callers don't try
5758 * to use other copies.
5762 map = em->map_lookup;
5763 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1_MASK))
5764 ret = map->num_stripes;
5765 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5766 ret = map->sub_stripes;
5767 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5769 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5771 * There could be two corrupted data stripes, we need
5772 * to loop retry in order to rebuild the correct data.
5774 * Fail a stripe at a time on every retry except the
5775 * stripe under reconstruction.
5777 ret = map->num_stripes;
5780 free_extent_map(em);
5782 down_read(&fs_info->dev_replace.rwsem);
5783 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) &&
5784 fs_info->dev_replace.tgtdev)
5786 up_read(&fs_info->dev_replace.rwsem);
5791 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
5794 struct extent_map *em;
5795 struct map_lookup *map;
5796 unsigned long len = fs_info->sectorsize;
5798 em = btrfs_get_chunk_map(fs_info, logical, len);
5800 if (!WARN_ON(IS_ERR(em))) {
5801 map = em->map_lookup;
5802 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5803 len = map->stripe_len * nr_data_stripes(map);
5804 free_extent_map(em);
5809 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5811 struct extent_map *em;
5812 struct map_lookup *map;
5815 em = btrfs_get_chunk_map(fs_info, logical, len);
5817 if(!WARN_ON(IS_ERR(em))) {
5818 map = em->map_lookup;
5819 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5821 free_extent_map(em);
5826 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5827 struct map_lookup *map, int first,
5828 int dev_replace_is_ongoing)
5832 int preferred_mirror;
5834 struct btrfs_device *srcdev;
5837 (BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10)));
5839 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5840 num_stripes = map->sub_stripes;
5842 num_stripes = map->num_stripes;
5844 switch (fs_info->fs_devices->read_policy) {
5846 /* Shouldn't happen, just warn and use pid instead of failing */
5847 btrfs_warn_rl(fs_info,
5848 "unknown read_policy type %u, reset to pid",
5849 fs_info->fs_devices->read_policy);
5850 fs_info->fs_devices->read_policy = BTRFS_READ_POLICY_PID;
5852 case BTRFS_READ_POLICY_PID:
5853 preferred_mirror = first + (current->pid % num_stripes);
5857 if (dev_replace_is_ongoing &&
5858 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5859 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5860 srcdev = fs_info->dev_replace.srcdev;
5865 * try to avoid the drive that is the source drive for a
5866 * dev-replace procedure, only choose it if no other non-missing
5867 * mirror is available
5869 for (tolerance = 0; tolerance < 2; tolerance++) {
5870 if (map->stripes[preferred_mirror].dev->bdev &&
5871 (tolerance || map->stripes[preferred_mirror].dev != srcdev))
5872 return preferred_mirror;
5873 for (i = first; i < first + num_stripes; i++) {
5874 if (map->stripes[i].dev->bdev &&
5875 (tolerance || map->stripes[i].dev != srcdev))
5880 /* we couldn't find one that doesn't fail. Just return something
5881 * and the io error handling code will clean up eventually
5883 return preferred_mirror;
5886 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5887 static void sort_parity_stripes(struct btrfs_io_context *bioc, int num_stripes)
5894 for (i = 0; i < num_stripes - 1; i++) {
5895 /* Swap if parity is on a smaller index */
5896 if (bioc->raid_map[i] > bioc->raid_map[i + 1]) {
5897 swap(bioc->stripes[i], bioc->stripes[i + 1]);
5898 swap(bioc->raid_map[i], bioc->raid_map[i + 1]);
5905 static struct btrfs_io_context *alloc_btrfs_io_context(struct btrfs_fs_info *fs_info,
5909 struct btrfs_io_context *bioc = kzalloc(
5910 /* The size of btrfs_io_context */
5911 sizeof(struct btrfs_io_context) +
5912 /* Plus the variable array for the stripes */
5913 sizeof(struct btrfs_io_stripe) * (total_stripes) +
5914 /* Plus the variable array for the tgt dev */
5915 sizeof(int) * (real_stripes) +
5917 * Plus the raid_map, which includes both the tgt dev
5920 sizeof(u64) * (total_stripes),
5921 GFP_NOFS|__GFP_NOFAIL);
5923 atomic_set(&bioc->error, 0);
5924 refcount_set(&bioc->refs, 1);
5926 bioc->fs_info = fs_info;
5927 bioc->tgtdev_map = (int *)(bioc->stripes + total_stripes);
5928 bioc->raid_map = (u64 *)(bioc->tgtdev_map + real_stripes);
5933 void btrfs_get_bioc(struct btrfs_io_context *bioc)
5935 WARN_ON(!refcount_read(&bioc->refs));
5936 refcount_inc(&bioc->refs);
5939 void btrfs_put_bioc(struct btrfs_io_context *bioc)
5943 if (refcount_dec_and_test(&bioc->refs))
5947 /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */
5949 * Please note that, discard won't be sent to target device of device
5952 static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info,
5953 u64 logical, u64 *length_ret,
5954 struct btrfs_io_context **bioc_ret)
5956 struct extent_map *em;
5957 struct map_lookup *map;
5958 struct btrfs_io_context *bioc;
5959 u64 length = *length_ret;
5963 u64 stripe_end_offset;
5970 u32 sub_stripes = 0;
5971 u64 stripes_per_dev = 0;
5972 u32 remaining_stripes = 0;
5973 u32 last_stripe = 0;
5977 /* Discard always returns a bioc. */
5980 em = btrfs_get_chunk_map(fs_info, logical, length);
5984 map = em->map_lookup;
5985 /* we don't discard raid56 yet */
5986 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5991 offset = logical - em->start;
5992 length = min_t(u64, em->start + em->len - logical, length);
5993 *length_ret = length;
5995 stripe_len = map->stripe_len;
5997 * stripe_nr counts the total number of stripes we have to stride
5998 * to get to this block
6000 stripe_nr = div64_u64(offset, stripe_len);
6002 /* stripe_offset is the offset of this block in its stripe */
6003 stripe_offset = offset - stripe_nr * stripe_len;
6005 stripe_nr_end = round_up(offset + length, map->stripe_len);
6006 stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len);
6007 stripe_cnt = stripe_nr_end - stripe_nr;
6008 stripe_end_offset = stripe_nr_end * map->stripe_len -
6011 * after this, stripe_nr is the number of stripes on this
6012 * device we have to walk to find the data, and stripe_index is
6013 * the number of our device in the stripe array
6017 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
6018 BTRFS_BLOCK_GROUP_RAID10)) {
6019 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
6022 sub_stripes = map->sub_stripes;
6024 factor = map->num_stripes / sub_stripes;
6025 num_stripes = min_t(u64, map->num_stripes,
6026 sub_stripes * stripe_cnt);
6027 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
6028 stripe_index *= sub_stripes;
6029 stripes_per_dev = div_u64_rem(stripe_cnt, factor,
6030 &remaining_stripes);
6031 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
6032 last_stripe *= sub_stripes;
6033 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1_MASK |
6034 BTRFS_BLOCK_GROUP_DUP)) {
6035 num_stripes = map->num_stripes;
6037 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
6041 bioc = alloc_btrfs_io_context(fs_info, num_stripes, 0);
6047 for (i = 0; i < num_stripes; i++) {
6048 bioc->stripes[i].physical =
6049 map->stripes[stripe_index].physical +
6050 stripe_offset + stripe_nr * map->stripe_len;
6051 bioc->stripes[i].dev = map->stripes[stripe_index].dev;
6053 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
6054 BTRFS_BLOCK_GROUP_RAID10)) {
6055 bioc->stripes[i].length = stripes_per_dev *
6058 if (i / sub_stripes < remaining_stripes)
6059 bioc->stripes[i].length += map->stripe_len;
6062 * Special for the first stripe and
6065 * |-------|...|-------|
6069 if (i < sub_stripes)
6070 bioc->stripes[i].length -= stripe_offset;
6072 if (stripe_index >= last_stripe &&
6073 stripe_index <= (last_stripe +
6075 bioc->stripes[i].length -= stripe_end_offset;
6077 if (i == sub_stripes - 1)
6080 bioc->stripes[i].length = length;
6084 if (stripe_index == map->num_stripes) {
6091 bioc->map_type = map->type;
6092 bioc->num_stripes = num_stripes;
6094 free_extent_map(em);
6099 * In dev-replace case, for repair case (that's the only case where the mirror
6100 * is selected explicitly when calling btrfs_map_block), blocks left of the
6101 * left cursor can also be read from the target drive.
6103 * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the
6105 * For READ, it also needs to be supported using the same mirror number.
6107 * If the requested block is not left of the left cursor, EIO is returned. This
6108 * can happen because btrfs_num_copies() returns one more in the dev-replace
6111 static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info,
6112 u64 logical, u64 length,
6113 u64 srcdev_devid, int *mirror_num,
6116 struct btrfs_io_context *bioc = NULL;
6118 int index_srcdev = 0;
6120 u64 physical_of_found = 0;
6124 ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS,
6125 logical, &length, &bioc, 0, 0);
6127 ASSERT(bioc == NULL);
6131 num_stripes = bioc->num_stripes;
6132 if (*mirror_num > num_stripes) {
6134 * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror,
6135 * that means that the requested area is not left of the left
6138 btrfs_put_bioc(bioc);
6143 * process the rest of the function using the mirror_num of the source
6144 * drive. Therefore look it up first. At the end, patch the device
6145 * pointer to the one of the target drive.
6147 for (i = 0; i < num_stripes; i++) {
6148 if (bioc->stripes[i].dev->devid != srcdev_devid)
6152 * In case of DUP, in order to keep it simple, only add the
6153 * mirror with the lowest physical address
6156 physical_of_found <= bioc->stripes[i].physical)
6161 physical_of_found = bioc->stripes[i].physical;
6164 btrfs_put_bioc(bioc);
6170 *mirror_num = index_srcdev + 1;
6171 *physical = physical_of_found;
6175 static bool is_block_group_to_copy(struct btrfs_fs_info *fs_info, u64 logical)
6177 struct btrfs_block_group *cache;
6180 /* Non zoned filesystem does not use "to_copy" flag */
6181 if (!btrfs_is_zoned(fs_info))
6184 cache = btrfs_lookup_block_group(fs_info, logical);
6186 spin_lock(&cache->lock);
6187 ret = cache->to_copy;
6188 spin_unlock(&cache->lock);
6190 btrfs_put_block_group(cache);
6194 static void handle_ops_on_dev_replace(enum btrfs_map_op op,
6195 struct btrfs_io_context **bioc_ret,
6196 struct btrfs_dev_replace *dev_replace,
6198 int *num_stripes_ret, int *max_errors_ret)
6200 struct btrfs_io_context *bioc = *bioc_ret;
6201 u64 srcdev_devid = dev_replace->srcdev->devid;
6202 int tgtdev_indexes = 0;
6203 int num_stripes = *num_stripes_ret;
6204 int max_errors = *max_errors_ret;
6207 if (op == BTRFS_MAP_WRITE) {
6208 int index_where_to_add;
6211 * A block group which have "to_copy" set will eventually
6212 * copied by dev-replace process. We can avoid cloning IO here.
6214 if (is_block_group_to_copy(dev_replace->srcdev->fs_info, logical))
6218 * duplicate the write operations while the dev replace
6219 * procedure is running. Since the copying of the old disk to
6220 * the new disk takes place at run time while the filesystem is
6221 * mounted writable, the regular write operations to the old
6222 * disk have to be duplicated to go to the new disk as well.
6224 * Note that device->missing is handled by the caller, and that
6225 * the write to the old disk is already set up in the stripes
6228 index_where_to_add = num_stripes;
6229 for (i = 0; i < num_stripes; i++) {
6230 if (bioc->stripes[i].dev->devid == srcdev_devid) {
6231 /* write to new disk, too */
6232 struct btrfs_io_stripe *new =
6233 bioc->stripes + index_where_to_add;
6234 struct btrfs_io_stripe *old =
6237 new->physical = old->physical;
6238 new->length = old->length;
6239 new->dev = dev_replace->tgtdev;
6240 bioc->tgtdev_map[i] = index_where_to_add;
6241 index_where_to_add++;
6246 num_stripes = index_where_to_add;
6247 } else if (op == BTRFS_MAP_GET_READ_MIRRORS) {
6248 int index_srcdev = 0;
6250 u64 physical_of_found = 0;
6253 * During the dev-replace procedure, the target drive can also
6254 * be used to read data in case it is needed to repair a corrupt
6255 * block elsewhere. This is possible if the requested area is
6256 * left of the left cursor. In this area, the target drive is a
6257 * full copy of the source drive.
6259 for (i = 0; i < num_stripes; i++) {
6260 if (bioc->stripes[i].dev->devid == srcdev_devid) {
6262 * In case of DUP, in order to keep it simple,
6263 * only add the mirror with the lowest physical
6267 physical_of_found <= bioc->stripes[i].physical)
6271 physical_of_found = bioc->stripes[i].physical;
6275 struct btrfs_io_stripe *tgtdev_stripe =
6276 bioc->stripes + num_stripes;
6278 tgtdev_stripe->physical = physical_of_found;
6279 tgtdev_stripe->length =
6280 bioc->stripes[index_srcdev].length;
6281 tgtdev_stripe->dev = dev_replace->tgtdev;
6282 bioc->tgtdev_map[index_srcdev] = num_stripes;
6289 *num_stripes_ret = num_stripes;
6290 *max_errors_ret = max_errors;
6291 bioc->num_tgtdevs = tgtdev_indexes;
6295 static bool need_full_stripe(enum btrfs_map_op op)
6297 return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS);
6301 * Calculate the geometry of a particular (address, len) tuple. This
6302 * information is used to calculate how big a particular bio can get before it
6303 * straddles a stripe.
6305 * @fs_info: the filesystem
6306 * @em: mapping containing the logical extent
6307 * @op: type of operation - write or read
6308 * @logical: address that we want to figure out the geometry of
6309 * @io_geom: pointer used to return values
6311 * Returns < 0 in case a chunk for the given logical address cannot be found,
6312 * usually shouldn't happen unless @logical is corrupted, 0 otherwise.
6314 int btrfs_get_io_geometry(struct btrfs_fs_info *fs_info, struct extent_map *em,
6315 enum btrfs_map_op op, u64 logical,
6316 struct btrfs_io_geometry *io_geom)
6318 struct map_lookup *map;
6324 u64 raid56_full_stripe_start = (u64)-1;
6327 ASSERT(op != BTRFS_MAP_DISCARD);
6329 map = em->map_lookup;
6330 /* Offset of this logical address in the chunk */
6331 offset = logical - em->start;
6332 /* Len of a stripe in a chunk */
6333 stripe_len = map->stripe_len;
6334 /* Stripe where this block falls in */
6335 stripe_nr = div64_u64(offset, stripe_len);
6336 /* Offset of stripe in the chunk */
6337 stripe_offset = stripe_nr * stripe_len;
6338 if (offset < stripe_offset) {
6340 "stripe math has gone wrong, stripe_offset=%llu offset=%llu start=%llu logical=%llu stripe_len=%llu",
6341 stripe_offset, offset, em->start, logical, stripe_len);
6345 /* stripe_offset is the offset of this block in its stripe */
6346 stripe_offset = offset - stripe_offset;
6347 data_stripes = nr_data_stripes(map);
6349 /* Only stripe based profiles needs to check against stripe length. */
6350 if (map->type & BTRFS_BLOCK_GROUP_STRIPE_MASK) {
6351 u64 max_len = stripe_len - stripe_offset;
6354 * In case of raid56, we need to know the stripe aligned start
6356 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6357 unsigned long full_stripe_len = stripe_len * data_stripes;
6358 raid56_full_stripe_start = offset;
6361 * Allow a write of a full stripe, but make sure we
6362 * don't allow straddling of stripes
6364 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
6366 raid56_full_stripe_start *= full_stripe_len;
6369 * For writes to RAID[56], allow a full stripeset across
6370 * all disks. For other RAID types and for RAID[56]
6371 * reads, just allow a single stripe (on a single disk).
6373 if (op == BTRFS_MAP_WRITE) {
6374 max_len = stripe_len * data_stripes -
6375 (offset - raid56_full_stripe_start);
6378 len = min_t(u64, em->len - offset, max_len);
6380 len = em->len - offset;
6384 io_geom->offset = offset;
6385 io_geom->stripe_len = stripe_len;
6386 io_geom->stripe_nr = stripe_nr;
6387 io_geom->stripe_offset = stripe_offset;
6388 io_geom->raid56_stripe_offset = raid56_full_stripe_start;
6393 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
6394 enum btrfs_map_op op,
6395 u64 logical, u64 *length,
6396 struct btrfs_io_context **bioc_ret,
6397 int mirror_num, int need_raid_map)
6399 struct extent_map *em;
6400 struct map_lookup *map;
6410 int tgtdev_indexes = 0;
6411 struct btrfs_io_context *bioc = NULL;
6412 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
6413 int dev_replace_is_ongoing = 0;
6414 int num_alloc_stripes;
6415 int patch_the_first_stripe_for_dev_replace = 0;
6416 u64 physical_to_patch_in_first_stripe = 0;
6417 u64 raid56_full_stripe_start = (u64)-1;
6418 struct btrfs_io_geometry geom;
6421 ASSERT(op != BTRFS_MAP_DISCARD);
6423 em = btrfs_get_chunk_map(fs_info, logical, *length);
6424 ASSERT(!IS_ERR(em));
6426 ret = btrfs_get_io_geometry(fs_info, em, op, logical, &geom);
6430 map = em->map_lookup;
6433 stripe_len = geom.stripe_len;
6434 stripe_nr = geom.stripe_nr;
6435 stripe_offset = geom.stripe_offset;
6436 raid56_full_stripe_start = geom.raid56_stripe_offset;
6437 data_stripes = nr_data_stripes(map);
6439 down_read(&dev_replace->rwsem);
6440 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
6442 * Hold the semaphore for read during the whole operation, write is
6443 * requested at commit time but must wait.
6445 if (!dev_replace_is_ongoing)
6446 up_read(&dev_replace->rwsem);
6448 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
6449 !need_full_stripe(op) && dev_replace->tgtdev != NULL) {
6450 ret = get_extra_mirror_from_replace(fs_info, logical, *length,
6451 dev_replace->srcdev->devid,
6453 &physical_to_patch_in_first_stripe);
6457 patch_the_first_stripe_for_dev_replace = 1;
6458 } else if (mirror_num > map->num_stripes) {
6464 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
6465 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
6467 if (!need_full_stripe(op))
6469 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1_MASK) {
6470 if (need_full_stripe(op))
6471 num_stripes = map->num_stripes;
6472 else if (mirror_num)
6473 stripe_index = mirror_num - 1;
6475 stripe_index = find_live_mirror(fs_info, map, 0,
6476 dev_replace_is_ongoing);
6477 mirror_num = stripe_index + 1;
6480 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
6481 if (need_full_stripe(op)) {
6482 num_stripes = map->num_stripes;
6483 } else if (mirror_num) {
6484 stripe_index = mirror_num - 1;
6489 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
6490 u32 factor = map->num_stripes / map->sub_stripes;
6492 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
6493 stripe_index *= map->sub_stripes;
6495 if (need_full_stripe(op))
6496 num_stripes = map->sub_stripes;
6497 else if (mirror_num)
6498 stripe_index += mirror_num - 1;
6500 int old_stripe_index = stripe_index;
6501 stripe_index = find_live_mirror(fs_info, map,
6503 dev_replace_is_ongoing);
6504 mirror_num = stripe_index - old_stripe_index + 1;
6507 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6508 if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) {
6509 /* push stripe_nr back to the start of the full stripe */
6510 stripe_nr = div64_u64(raid56_full_stripe_start,
6511 stripe_len * data_stripes);
6513 /* RAID[56] write or recovery. Return all stripes */
6514 num_stripes = map->num_stripes;
6515 max_errors = nr_parity_stripes(map);
6517 *length = map->stripe_len;
6522 * Mirror #0 or #1 means the original data block.
6523 * Mirror #2 is RAID5 parity block.
6524 * Mirror #3 is RAID6 Q block.
6526 stripe_nr = div_u64_rem(stripe_nr,
6527 data_stripes, &stripe_index);
6529 stripe_index = data_stripes + mirror_num - 2;
6531 /* We distribute the parity blocks across stripes */
6532 div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
6534 if (!need_full_stripe(op) && mirror_num <= 1)
6539 * after this, stripe_nr is the number of stripes on this
6540 * device we have to walk to find the data, and stripe_index is
6541 * the number of our device in the stripe array
6543 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
6545 mirror_num = stripe_index + 1;
6547 if (stripe_index >= map->num_stripes) {
6549 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
6550 stripe_index, map->num_stripes);
6555 num_alloc_stripes = num_stripes;
6556 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) {
6557 if (op == BTRFS_MAP_WRITE)
6558 num_alloc_stripes <<= 1;
6559 if (op == BTRFS_MAP_GET_READ_MIRRORS)
6560 num_alloc_stripes++;
6561 tgtdev_indexes = num_stripes;
6564 bioc = alloc_btrfs_io_context(fs_info, num_alloc_stripes, tgtdev_indexes);
6570 for (i = 0; i < num_stripes; i++) {
6571 bioc->stripes[i].physical = map->stripes[stripe_index].physical +
6572 stripe_offset + stripe_nr * map->stripe_len;
6573 bioc->stripes[i].dev = map->stripes[stripe_index].dev;
6577 /* Build raid_map */
6578 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map &&
6579 (need_full_stripe(op) || mirror_num > 1)) {
6583 /* Work out the disk rotation on this stripe-set */
6584 div_u64_rem(stripe_nr, num_stripes, &rot);
6586 /* Fill in the logical address of each stripe */
6587 tmp = stripe_nr * data_stripes;
6588 for (i = 0; i < data_stripes; i++)
6589 bioc->raid_map[(i + rot) % num_stripes] =
6590 em->start + (tmp + i) * map->stripe_len;
6592 bioc->raid_map[(i + rot) % map->num_stripes] = RAID5_P_STRIPE;
6593 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
6594 bioc->raid_map[(i + rot + 1) % num_stripes] =
6597 sort_parity_stripes(bioc, num_stripes);
6600 if (need_full_stripe(op))
6601 max_errors = btrfs_chunk_max_errors(map);
6603 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
6604 need_full_stripe(op)) {
6605 handle_ops_on_dev_replace(op, &bioc, dev_replace, logical,
6606 &num_stripes, &max_errors);
6610 bioc->map_type = map->type;
6611 bioc->num_stripes = num_stripes;
6612 bioc->max_errors = max_errors;
6613 bioc->mirror_num = mirror_num;
6616 * this is the case that REQ_READ && dev_replace_is_ongoing &&
6617 * mirror_num == num_stripes + 1 && dev_replace target drive is
6618 * available as a mirror
6620 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
6621 WARN_ON(num_stripes > 1);
6622 bioc->stripes[0].dev = dev_replace->tgtdev;
6623 bioc->stripes[0].physical = physical_to_patch_in_first_stripe;
6624 bioc->mirror_num = map->num_stripes + 1;
6627 if (dev_replace_is_ongoing) {
6628 lockdep_assert_held(&dev_replace->rwsem);
6629 /* Unlock and let waiting writers proceed */
6630 up_read(&dev_replace->rwsem);
6632 free_extent_map(em);
6636 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6637 u64 logical, u64 *length,
6638 struct btrfs_io_context **bioc_ret, int mirror_num)
6640 if (op == BTRFS_MAP_DISCARD)
6641 return __btrfs_map_block_for_discard(fs_info, logical,
6644 return __btrfs_map_block(fs_info, op, logical, length, bioc_ret,
6648 /* For Scrub/replace */
6649 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6650 u64 logical, u64 *length,
6651 struct btrfs_io_context **bioc_ret)
6653 return __btrfs_map_block(fs_info, op, logical, length, bioc_ret, 0, 1);
6656 static inline void btrfs_end_bioc(struct btrfs_io_context *bioc, struct bio *bio)
6658 bio->bi_private = bioc->private;
6659 bio->bi_end_io = bioc->end_io;
6662 btrfs_put_bioc(bioc);
6665 static void btrfs_end_bio(struct bio *bio)
6667 struct btrfs_io_context *bioc = bio->bi_private;
6668 int is_orig_bio = 0;
6670 if (bio->bi_status) {
6671 atomic_inc(&bioc->error);
6672 if (bio->bi_status == BLK_STS_IOERR ||
6673 bio->bi_status == BLK_STS_TARGET) {
6674 struct btrfs_device *dev = btrfs_bio(bio)->device;
6677 if (btrfs_op(bio) == BTRFS_MAP_WRITE)
6678 btrfs_dev_stat_inc_and_print(dev,
6679 BTRFS_DEV_STAT_WRITE_ERRS);
6680 else if (!(bio->bi_opf & REQ_RAHEAD))
6681 btrfs_dev_stat_inc_and_print(dev,
6682 BTRFS_DEV_STAT_READ_ERRS);
6683 if (bio->bi_opf & REQ_PREFLUSH)
6684 btrfs_dev_stat_inc_and_print(dev,
6685 BTRFS_DEV_STAT_FLUSH_ERRS);
6689 if (bio == bioc->orig_bio)
6692 btrfs_bio_counter_dec(bioc->fs_info);
6694 if (atomic_dec_and_test(&bioc->stripes_pending)) {
6697 bio = bioc->orig_bio;
6700 btrfs_bio(bio)->mirror_num = bioc->mirror_num;
6701 /* only send an error to the higher layers if it is
6702 * beyond the tolerance of the btrfs bio
6704 if (atomic_read(&bioc->error) > bioc->max_errors) {
6705 bio->bi_status = BLK_STS_IOERR;
6708 * this bio is actually up to date, we didn't
6709 * go over the max number of errors
6711 bio->bi_status = BLK_STS_OK;
6714 btrfs_end_bioc(bioc, bio);
6715 } else if (!is_orig_bio) {
6720 static void submit_stripe_bio(struct btrfs_io_context *bioc, struct bio *bio,
6721 u64 physical, struct btrfs_device *dev)
6723 struct btrfs_fs_info *fs_info = bioc->fs_info;
6725 bio->bi_private = bioc;
6726 btrfs_bio(bio)->device = dev;
6727 bio->bi_end_io = btrfs_end_bio;
6728 bio->bi_iter.bi_sector = physical >> 9;
6730 * For zone append writing, bi_sector must point the beginning of the
6733 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
6734 if (btrfs_dev_is_sequential(dev, physical)) {
6735 u64 zone_start = round_down(physical, fs_info->zone_size);
6737 bio->bi_iter.bi_sector = zone_start >> SECTOR_SHIFT;
6739 bio->bi_opf &= ~REQ_OP_ZONE_APPEND;
6740 bio->bi_opf |= REQ_OP_WRITE;
6743 btrfs_debug_in_rcu(fs_info,
6744 "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
6745 bio_op(bio), bio->bi_opf, bio->bi_iter.bi_sector,
6746 (unsigned long)dev->bdev->bd_dev, rcu_str_deref(dev->name),
6747 dev->devid, bio->bi_iter.bi_size);
6748 bio_set_dev(bio, dev->bdev);
6750 btrfs_bio_counter_inc_noblocked(fs_info);
6752 btrfsic_submit_bio(bio);
6755 static void bioc_error(struct btrfs_io_context *bioc, struct bio *bio, u64 logical)
6757 atomic_inc(&bioc->error);
6758 if (atomic_dec_and_test(&bioc->stripes_pending)) {
6759 /* Should be the original bio. */
6760 WARN_ON(bio != bioc->orig_bio);
6762 btrfs_bio(bio)->mirror_num = bioc->mirror_num;
6763 bio->bi_iter.bi_sector = logical >> 9;
6764 if (atomic_read(&bioc->error) > bioc->max_errors)
6765 bio->bi_status = BLK_STS_IOERR;
6767 bio->bi_status = BLK_STS_OK;
6768 btrfs_end_bioc(bioc, bio);
6772 blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
6775 struct btrfs_device *dev;
6776 struct bio *first_bio = bio;
6777 u64 logical = bio->bi_iter.bi_sector << 9;
6783 struct btrfs_io_context *bioc = NULL;
6785 length = bio->bi_iter.bi_size;
6786 map_length = length;
6788 btrfs_bio_counter_inc_blocked(fs_info);
6789 ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical,
6790 &map_length, &bioc, mirror_num, 1);
6792 btrfs_bio_counter_dec(fs_info);
6793 return errno_to_blk_status(ret);
6796 total_devs = bioc->num_stripes;
6797 bioc->orig_bio = first_bio;
6798 bioc->private = first_bio->bi_private;
6799 bioc->end_io = first_bio->bi_end_io;
6800 atomic_set(&bioc->stripes_pending, bioc->num_stripes);
6802 if ((bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6803 ((btrfs_op(bio) == BTRFS_MAP_WRITE) || (mirror_num > 1))) {
6804 /* In this case, map_length has been set to the length of
6805 a single stripe; not the whole write */
6806 if (btrfs_op(bio) == BTRFS_MAP_WRITE) {
6807 ret = raid56_parity_write(bio, bioc, map_length);
6809 ret = raid56_parity_recover(bio, bioc, map_length,
6813 btrfs_bio_counter_dec(fs_info);
6814 return errno_to_blk_status(ret);
6817 if (map_length < length) {
6819 "mapping failed logical %llu bio len %llu len %llu",
6820 logical, length, map_length);
6824 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6825 dev = bioc->stripes[dev_nr].dev;
6826 if (!dev || !dev->bdev || test_bit(BTRFS_DEV_STATE_MISSING,
6828 (btrfs_op(first_bio) == BTRFS_MAP_WRITE &&
6829 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) {
6830 bioc_error(bioc, first_bio, logical);
6834 if (dev_nr < total_devs - 1)
6835 bio = btrfs_bio_clone(first_bio);
6839 submit_stripe_bio(bioc, bio, bioc->stripes[dev_nr].physical, dev);
6841 btrfs_bio_counter_dec(fs_info);
6845 static bool dev_args_match_fs_devices(const struct btrfs_dev_lookup_args *args,
6846 const struct btrfs_fs_devices *fs_devices)
6848 if (args->fsid == NULL)
6850 if (memcmp(fs_devices->metadata_uuid, args->fsid, BTRFS_FSID_SIZE) == 0)
6855 static bool dev_args_match_device(const struct btrfs_dev_lookup_args *args,
6856 const struct btrfs_device *device)
6858 ASSERT((args->devid != (u64)-1) || args->missing);
6860 if ((args->devid != (u64)-1) && device->devid != args->devid)
6862 if (args->uuid && memcmp(device->uuid, args->uuid, BTRFS_UUID_SIZE) != 0)
6866 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state) &&
6873 * Find a device specified by @devid or @uuid in the list of @fs_devices, or
6876 * If devid and uuid are both specified, the match must be exact, otherwise
6877 * only devid is used.
6879 struct btrfs_device *btrfs_find_device(const struct btrfs_fs_devices *fs_devices,
6880 const struct btrfs_dev_lookup_args *args)
6882 struct btrfs_device *device;
6883 struct btrfs_fs_devices *seed_devs;
6885 if (dev_args_match_fs_devices(args, fs_devices)) {
6886 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6887 if (dev_args_match_device(args, device))
6892 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
6893 if (!dev_args_match_fs_devices(args, seed_devs))
6895 list_for_each_entry(device, &seed_devs->devices, dev_list) {
6896 if (dev_args_match_device(args, device))
6904 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices,
6905 u64 devid, u8 *dev_uuid)
6907 struct btrfs_device *device;
6908 unsigned int nofs_flag;
6911 * We call this under the chunk_mutex, so we want to use NOFS for this
6912 * allocation, however we don't want to change btrfs_alloc_device() to
6913 * always do NOFS because we use it in a lot of other GFP_KERNEL safe
6916 nofs_flag = memalloc_nofs_save();
6917 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6918 memalloc_nofs_restore(nofs_flag);
6922 list_add(&device->dev_list, &fs_devices->devices);
6923 device->fs_devices = fs_devices;
6924 fs_devices->num_devices++;
6926 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6927 fs_devices->missing_devices++;
6933 * btrfs_alloc_device - allocate struct btrfs_device
6934 * @fs_info: used only for generating a new devid, can be NULL if
6935 * devid is provided (i.e. @devid != NULL).
6936 * @devid: a pointer to devid for this device. If NULL a new devid
6938 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6941 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6942 * on error. Returned struct is not linked onto any lists and must be
6943 * destroyed with btrfs_free_device.
6945 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6949 struct btrfs_device *dev;
6952 if (WARN_ON(!devid && !fs_info))
6953 return ERR_PTR(-EINVAL);
6955 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
6957 return ERR_PTR(-ENOMEM);
6960 * Preallocate a bio that's always going to be used for flushing device
6961 * barriers and matches the device lifespan
6963 dev->flush_bio = bio_kmalloc(GFP_KERNEL, 0);
6964 if (!dev->flush_bio) {
6966 return ERR_PTR(-ENOMEM);
6969 INIT_LIST_HEAD(&dev->dev_list);
6970 INIT_LIST_HEAD(&dev->dev_alloc_list);
6971 INIT_LIST_HEAD(&dev->post_commit_list);
6973 atomic_set(&dev->dev_stats_ccnt, 0);
6974 btrfs_device_data_ordered_init(dev);
6975 extent_io_tree_init(fs_info, &dev->alloc_state,
6976 IO_TREE_DEVICE_ALLOC_STATE, NULL);
6983 ret = find_next_devid(fs_info, &tmp);
6985 btrfs_free_device(dev);
6986 return ERR_PTR(ret);
6992 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6994 generate_random_uuid(dev->uuid);
6999 static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info,
7000 u64 devid, u8 *uuid, bool error)
7003 btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing",
7006 btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing",
7010 static u64 calc_stripe_length(u64 type, u64 chunk_len, int num_stripes)
7012 const int data_stripes = calc_data_stripes(type, num_stripes);
7014 return div_u64(chunk_len, data_stripes);
7017 #if BITS_PER_LONG == 32
7019 * Due to page cache limit, metadata beyond BTRFS_32BIT_MAX_FILE_SIZE
7020 * can't be accessed on 32bit systems.
7022 * This function do mount time check to reject the fs if it already has
7023 * metadata chunk beyond that limit.
7025 static int check_32bit_meta_chunk(struct btrfs_fs_info *fs_info,
7026 u64 logical, u64 length, u64 type)
7028 if (!(type & BTRFS_BLOCK_GROUP_METADATA))
7031 if (logical + length < MAX_LFS_FILESIZE)
7034 btrfs_err_32bit_limit(fs_info);
7039 * This is to give early warning for any metadata chunk reaching
7040 * BTRFS_32BIT_EARLY_WARN_THRESHOLD.
7041 * Although we can still access the metadata, it's not going to be possible
7042 * once the limit is reached.
7044 static void warn_32bit_meta_chunk(struct btrfs_fs_info *fs_info,
7045 u64 logical, u64 length, u64 type)
7047 if (!(type & BTRFS_BLOCK_GROUP_METADATA))
7050 if (logical + length < BTRFS_32BIT_EARLY_WARN_THRESHOLD)
7053 btrfs_warn_32bit_limit(fs_info);
7057 static struct btrfs_device *handle_missing_device(struct btrfs_fs_info *fs_info,
7058 u64 devid, u8 *uuid)
7060 struct btrfs_device *dev;
7062 if (!btrfs_test_opt(fs_info, DEGRADED)) {
7063 btrfs_report_missing_device(fs_info, devid, uuid, true);
7064 return ERR_PTR(-ENOENT);
7067 dev = add_missing_dev(fs_info->fs_devices, devid, uuid);
7069 btrfs_err(fs_info, "failed to init missing device %llu: %ld",
7070 devid, PTR_ERR(dev));
7073 btrfs_report_missing_device(fs_info, devid, uuid, false);
7078 static int read_one_chunk(struct btrfs_key *key, struct extent_buffer *leaf,
7079 struct btrfs_chunk *chunk)
7081 BTRFS_DEV_LOOKUP_ARGS(args);
7082 struct btrfs_fs_info *fs_info = leaf->fs_info;
7083 struct extent_map_tree *map_tree = &fs_info->mapping_tree;
7084 struct map_lookup *map;
7085 struct extent_map *em;
7090 u8 uuid[BTRFS_UUID_SIZE];
7095 logical = key->offset;
7096 length = btrfs_chunk_length(leaf, chunk);
7097 type = btrfs_chunk_type(leaf, chunk);
7098 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
7100 #if BITS_PER_LONG == 32
7101 ret = check_32bit_meta_chunk(fs_info, logical, length, type);
7104 warn_32bit_meta_chunk(fs_info, logical, length, type);
7108 * Only need to verify chunk item if we're reading from sys chunk array,
7109 * as chunk item in tree block is already verified by tree-checker.
7111 if (leaf->start == BTRFS_SUPER_INFO_OFFSET) {
7112 ret = btrfs_check_chunk_valid(leaf, chunk, logical);
7117 read_lock(&map_tree->lock);
7118 em = lookup_extent_mapping(map_tree, logical, 1);
7119 read_unlock(&map_tree->lock);
7121 /* already mapped? */
7122 if (em && em->start <= logical && em->start + em->len > logical) {
7123 free_extent_map(em);
7126 free_extent_map(em);
7129 em = alloc_extent_map();
7132 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
7134 free_extent_map(em);
7138 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
7139 em->map_lookup = map;
7140 em->start = logical;
7143 em->block_start = 0;
7144 em->block_len = em->len;
7146 map->num_stripes = num_stripes;
7147 map->io_width = btrfs_chunk_io_width(leaf, chunk);
7148 map->io_align = btrfs_chunk_io_align(leaf, chunk);
7149 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
7151 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
7152 map->verified_stripes = 0;
7153 em->orig_block_len = calc_stripe_length(type, em->len,
7155 for (i = 0; i < num_stripes; i++) {
7156 map->stripes[i].physical =
7157 btrfs_stripe_offset_nr(leaf, chunk, i);
7158 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
7160 read_extent_buffer(leaf, uuid, (unsigned long)
7161 btrfs_stripe_dev_uuid_nr(chunk, i),
7164 map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices, &args);
7165 if (!map->stripes[i].dev) {
7166 map->stripes[i].dev = handle_missing_device(fs_info,
7168 if (IS_ERR(map->stripes[i].dev)) {
7169 free_extent_map(em);
7170 return PTR_ERR(map->stripes[i].dev);
7174 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
7175 &(map->stripes[i].dev->dev_state));
7178 write_lock(&map_tree->lock);
7179 ret = add_extent_mapping(map_tree, em, 0);
7180 write_unlock(&map_tree->lock);
7183 "failed to add chunk map, start=%llu len=%llu: %d",
7184 em->start, em->len, ret);
7186 free_extent_map(em);
7191 static void fill_device_from_item(struct extent_buffer *leaf,
7192 struct btrfs_dev_item *dev_item,
7193 struct btrfs_device *device)
7197 device->devid = btrfs_device_id(leaf, dev_item);
7198 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
7199 device->total_bytes = device->disk_total_bytes;
7200 device->commit_total_bytes = device->disk_total_bytes;
7201 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
7202 device->commit_bytes_used = device->bytes_used;
7203 device->type = btrfs_device_type(leaf, dev_item);
7204 device->io_align = btrfs_device_io_align(leaf, dev_item);
7205 device->io_width = btrfs_device_io_width(leaf, dev_item);
7206 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
7207 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
7208 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
7210 ptr = btrfs_device_uuid(dev_item);
7211 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
7214 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info,
7217 struct btrfs_fs_devices *fs_devices;
7220 lockdep_assert_held(&uuid_mutex);
7223 /* This will match only for multi-device seed fs */
7224 list_for_each_entry(fs_devices, &fs_info->fs_devices->seed_list, seed_list)
7225 if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE))
7229 fs_devices = find_fsid(fsid, NULL);
7231 if (!btrfs_test_opt(fs_info, DEGRADED))
7232 return ERR_PTR(-ENOENT);
7234 fs_devices = alloc_fs_devices(fsid, NULL);
7235 if (IS_ERR(fs_devices))
7238 fs_devices->seeding = true;
7239 fs_devices->opened = 1;
7244 * Upon first call for a seed fs fsid, just create a private copy of the
7245 * respective fs_devices and anchor it at fs_info->fs_devices->seed_list
7247 fs_devices = clone_fs_devices(fs_devices);
7248 if (IS_ERR(fs_devices))
7251 ret = open_fs_devices(fs_devices, FMODE_READ, fs_info->bdev_holder);
7253 free_fs_devices(fs_devices);
7254 return ERR_PTR(ret);
7257 if (!fs_devices->seeding) {
7258 close_fs_devices(fs_devices);
7259 free_fs_devices(fs_devices);
7260 return ERR_PTR(-EINVAL);
7263 list_add(&fs_devices->seed_list, &fs_info->fs_devices->seed_list);
7268 static int read_one_dev(struct extent_buffer *leaf,
7269 struct btrfs_dev_item *dev_item)
7271 BTRFS_DEV_LOOKUP_ARGS(args);
7272 struct btrfs_fs_info *fs_info = leaf->fs_info;
7273 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7274 struct btrfs_device *device;
7277 u8 fs_uuid[BTRFS_FSID_SIZE];
7278 u8 dev_uuid[BTRFS_UUID_SIZE];
7280 devid = args.devid = btrfs_device_id(leaf, dev_item);
7281 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
7283 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
7285 args.uuid = dev_uuid;
7286 args.fsid = fs_uuid;
7288 if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) {
7289 fs_devices = open_seed_devices(fs_info, fs_uuid);
7290 if (IS_ERR(fs_devices))
7291 return PTR_ERR(fs_devices);
7294 device = btrfs_find_device(fs_info->fs_devices, &args);
7296 if (!btrfs_test_opt(fs_info, DEGRADED)) {
7297 btrfs_report_missing_device(fs_info, devid,
7302 device = add_missing_dev(fs_devices, devid, dev_uuid);
7303 if (IS_ERR(device)) {
7305 "failed to add missing dev %llu: %ld",
7306 devid, PTR_ERR(device));
7307 return PTR_ERR(device);
7309 btrfs_report_missing_device(fs_info, devid, dev_uuid, false);
7311 if (!device->bdev) {
7312 if (!btrfs_test_opt(fs_info, DEGRADED)) {
7313 btrfs_report_missing_device(fs_info,
7314 devid, dev_uuid, true);
7317 btrfs_report_missing_device(fs_info, devid,
7321 if (!device->bdev &&
7322 !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
7324 * this happens when a device that was properly setup
7325 * in the device info lists suddenly goes bad.
7326 * device->bdev is NULL, and so we have to set
7327 * device->missing to one here
7329 device->fs_devices->missing_devices++;
7330 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
7333 /* Move the device to its own fs_devices */
7334 if (device->fs_devices != fs_devices) {
7335 ASSERT(test_bit(BTRFS_DEV_STATE_MISSING,
7336 &device->dev_state));
7338 list_move(&device->dev_list, &fs_devices->devices);
7339 device->fs_devices->num_devices--;
7340 fs_devices->num_devices++;
7342 device->fs_devices->missing_devices--;
7343 fs_devices->missing_devices++;
7345 device->fs_devices = fs_devices;
7349 if (device->fs_devices != fs_info->fs_devices) {
7350 BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state));
7351 if (device->generation !=
7352 btrfs_device_generation(leaf, dev_item))
7356 fill_device_from_item(leaf, dev_item, device);
7358 u64 max_total_bytes = bdev_nr_bytes(device->bdev);
7360 if (device->total_bytes > max_total_bytes) {
7362 "device total_bytes should be at most %llu but found %llu",
7363 max_total_bytes, device->total_bytes);
7367 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
7368 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
7369 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
7370 device->fs_devices->total_rw_bytes += device->total_bytes;
7371 atomic64_add(device->total_bytes - device->bytes_used,
7372 &fs_info->free_chunk_space);
7378 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
7380 struct btrfs_root *root = fs_info->tree_root;
7381 struct btrfs_super_block *super_copy = fs_info->super_copy;
7382 struct extent_buffer *sb;
7383 struct btrfs_disk_key *disk_key;
7384 struct btrfs_chunk *chunk;
7386 unsigned long sb_array_offset;
7393 struct btrfs_key key;
7395 ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize);
7397 * This will create extent buffer of nodesize, superblock size is
7398 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
7399 * overallocate but we can keep it as-is, only the first page is used.
7401 sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET,
7402 root->root_key.objectid, 0);
7405 set_extent_buffer_uptodate(sb);
7407 * The sb extent buffer is artificial and just used to read the system array.
7408 * set_extent_buffer_uptodate() call does not properly mark all it's
7409 * pages up-to-date when the page is larger: extent does not cover the
7410 * whole page and consequently check_page_uptodate does not find all
7411 * the page's extents up-to-date (the hole beyond sb),
7412 * write_extent_buffer then triggers a WARN_ON.
7414 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
7415 * but sb spans only this function. Add an explicit SetPageUptodate call
7416 * to silence the warning eg. on PowerPC 64.
7418 if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
7419 SetPageUptodate(sb->pages[0]);
7421 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
7422 array_size = btrfs_super_sys_array_size(super_copy);
7424 array_ptr = super_copy->sys_chunk_array;
7425 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
7428 while (cur_offset < array_size) {
7429 disk_key = (struct btrfs_disk_key *)array_ptr;
7430 len = sizeof(*disk_key);
7431 if (cur_offset + len > array_size)
7432 goto out_short_read;
7434 btrfs_disk_key_to_cpu(&key, disk_key);
7437 sb_array_offset += len;
7440 if (key.type != BTRFS_CHUNK_ITEM_KEY) {
7442 "unexpected item type %u in sys_array at offset %u",
7443 (u32)key.type, cur_offset);
7448 chunk = (struct btrfs_chunk *)sb_array_offset;
7450 * At least one btrfs_chunk with one stripe must be present,
7451 * exact stripe count check comes afterwards
7453 len = btrfs_chunk_item_size(1);
7454 if (cur_offset + len > array_size)
7455 goto out_short_read;
7457 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
7460 "invalid number of stripes %u in sys_array at offset %u",
7461 num_stripes, cur_offset);
7466 type = btrfs_chunk_type(sb, chunk);
7467 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
7469 "invalid chunk type %llu in sys_array at offset %u",
7475 len = btrfs_chunk_item_size(num_stripes);
7476 if (cur_offset + len > array_size)
7477 goto out_short_read;
7479 ret = read_one_chunk(&key, sb, chunk);
7484 sb_array_offset += len;
7487 clear_extent_buffer_uptodate(sb);
7488 free_extent_buffer_stale(sb);
7492 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
7494 clear_extent_buffer_uptodate(sb);
7495 free_extent_buffer_stale(sb);
7500 * Check if all chunks in the fs are OK for read-write degraded mount
7502 * If the @failing_dev is specified, it's accounted as missing.
7504 * Return true if all chunks meet the minimal RW mount requirements.
7505 * Return false if any chunk doesn't meet the minimal RW mount requirements.
7507 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
7508 struct btrfs_device *failing_dev)
7510 struct extent_map_tree *map_tree = &fs_info->mapping_tree;
7511 struct extent_map *em;
7515 read_lock(&map_tree->lock);
7516 em = lookup_extent_mapping(map_tree, 0, (u64)-1);
7517 read_unlock(&map_tree->lock);
7518 /* No chunk at all? Return false anyway */
7524 struct map_lookup *map;
7529 map = em->map_lookup;
7531 btrfs_get_num_tolerated_disk_barrier_failures(
7533 for (i = 0; i < map->num_stripes; i++) {
7534 struct btrfs_device *dev = map->stripes[i].dev;
7536 if (!dev || !dev->bdev ||
7537 test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) ||
7538 dev->last_flush_error)
7540 else if (failing_dev && failing_dev == dev)
7543 if (missing > max_tolerated) {
7546 "chunk %llu missing %d devices, max tolerance is %d for writable mount",
7547 em->start, missing, max_tolerated);
7548 free_extent_map(em);
7552 next_start = extent_map_end(em);
7553 free_extent_map(em);
7555 read_lock(&map_tree->lock);
7556 em = lookup_extent_mapping(map_tree, next_start,
7557 (u64)(-1) - next_start);
7558 read_unlock(&map_tree->lock);
7564 static void readahead_tree_node_children(struct extent_buffer *node)
7567 const int nr_items = btrfs_header_nritems(node);
7569 for (i = 0; i < nr_items; i++)
7570 btrfs_readahead_node_child(node, i);
7573 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
7575 struct btrfs_root *root = fs_info->chunk_root;
7576 struct btrfs_path *path;
7577 struct extent_buffer *leaf;
7578 struct btrfs_key key;
7579 struct btrfs_key found_key;
7583 u64 last_ra_node = 0;
7585 path = btrfs_alloc_path();
7590 * uuid_mutex is needed only if we are mounting a sprout FS
7591 * otherwise we don't need it.
7593 mutex_lock(&uuid_mutex);
7596 * It is possible for mount and umount to race in such a way that
7597 * we execute this code path, but open_fs_devices failed to clear
7598 * total_rw_bytes. We certainly want it cleared before reading the
7599 * device items, so clear it here.
7601 fs_info->fs_devices->total_rw_bytes = 0;
7604 * Lockdep complains about possible circular locking dependency between
7605 * a disk's open_mutex (struct gendisk.open_mutex), the rw semaphores
7606 * used for freeze procection of a fs (struct super_block.s_writers),
7607 * which we take when starting a transaction, and extent buffers of the
7608 * chunk tree if we call read_one_dev() while holding a lock on an
7609 * extent buffer of the chunk tree. Since we are mounting the filesystem
7610 * and at this point there can't be any concurrent task modifying the
7611 * chunk tree, to keep it simple, just skip locking on the chunk tree.
7613 ASSERT(!test_bit(BTRFS_FS_OPEN, &fs_info->flags));
7614 path->skip_locking = 1;
7617 * Read all device items, and then all the chunk items. All
7618 * device items are found before any chunk item (their object id
7619 * is smaller than the lowest possible object id for a chunk
7620 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
7622 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
7625 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7629 struct extent_buffer *node;
7631 leaf = path->nodes[0];
7632 slot = path->slots[0];
7633 if (slot >= btrfs_header_nritems(leaf)) {
7634 ret = btrfs_next_leaf(root, path);
7641 node = path->nodes[1];
7643 if (last_ra_node != node->start) {
7644 readahead_tree_node_children(node);
7645 last_ra_node = node->start;
7648 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7649 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
7650 struct btrfs_dev_item *dev_item;
7651 dev_item = btrfs_item_ptr(leaf, slot,
7652 struct btrfs_dev_item);
7653 ret = read_one_dev(leaf, dev_item);
7657 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
7658 struct btrfs_chunk *chunk;
7661 * We are only called at mount time, so no need to take
7662 * fs_info->chunk_mutex. Plus, to avoid lockdep warnings,
7663 * we always lock first fs_info->chunk_mutex before
7664 * acquiring any locks on the chunk tree. This is a
7665 * requirement for chunk allocation, see the comment on
7666 * top of btrfs_chunk_alloc() for details.
7668 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
7669 ret = read_one_chunk(&found_key, leaf, chunk);
7677 * After loading chunk tree, we've got all device information,
7678 * do another round of validation checks.
7680 if (total_dev != fs_info->fs_devices->total_devices) {
7682 "super_num_devices %llu mismatch with num_devices %llu found here",
7683 btrfs_super_num_devices(fs_info->super_copy),
7688 if (btrfs_super_total_bytes(fs_info->super_copy) <
7689 fs_info->fs_devices->total_rw_bytes) {
7691 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
7692 btrfs_super_total_bytes(fs_info->super_copy),
7693 fs_info->fs_devices->total_rw_bytes);
7699 mutex_unlock(&uuid_mutex);
7701 btrfs_free_path(path);
7705 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
7707 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
7708 struct btrfs_device *device;
7710 fs_devices->fs_info = fs_info;
7712 mutex_lock(&fs_devices->device_list_mutex);
7713 list_for_each_entry(device, &fs_devices->devices, dev_list)
7714 device->fs_info = fs_info;
7716 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
7717 list_for_each_entry(device, &seed_devs->devices, dev_list)
7718 device->fs_info = fs_info;
7720 seed_devs->fs_info = fs_info;
7722 mutex_unlock(&fs_devices->device_list_mutex);
7725 static u64 btrfs_dev_stats_value(const struct extent_buffer *eb,
7726 const struct btrfs_dev_stats_item *ptr,
7731 read_extent_buffer(eb, &val,
7732 offsetof(struct btrfs_dev_stats_item, values) +
7733 ((unsigned long)ptr) + (index * sizeof(u64)),
7738 static void btrfs_set_dev_stats_value(struct extent_buffer *eb,
7739 struct btrfs_dev_stats_item *ptr,
7742 write_extent_buffer(eb, &val,
7743 offsetof(struct btrfs_dev_stats_item, values) +
7744 ((unsigned long)ptr) + (index * sizeof(u64)),
7748 static int btrfs_device_init_dev_stats(struct btrfs_device *device,
7749 struct btrfs_path *path)
7751 struct btrfs_dev_stats_item *ptr;
7752 struct extent_buffer *eb;
7753 struct btrfs_key key;
7757 if (!device->fs_info->dev_root)
7760 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7761 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7762 key.offset = device->devid;
7763 ret = btrfs_search_slot(NULL, device->fs_info->dev_root, &key, path, 0, 0);
7765 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7766 btrfs_dev_stat_set(device, i, 0);
7767 device->dev_stats_valid = 1;
7768 btrfs_release_path(path);
7769 return ret < 0 ? ret : 0;
7771 slot = path->slots[0];
7772 eb = path->nodes[0];
7773 item_size = btrfs_item_size(eb, slot);
7775 ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_stats_item);
7777 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7778 if (item_size >= (1 + i) * sizeof(__le64))
7779 btrfs_dev_stat_set(device, i,
7780 btrfs_dev_stats_value(eb, ptr, i));
7782 btrfs_dev_stat_set(device, i, 0);
7785 device->dev_stats_valid = 1;
7786 btrfs_dev_stat_print_on_load(device);
7787 btrfs_release_path(path);
7792 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
7794 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
7795 struct btrfs_device *device;
7796 struct btrfs_path *path = NULL;
7799 path = btrfs_alloc_path();
7803 mutex_lock(&fs_devices->device_list_mutex);
7804 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7805 ret = btrfs_device_init_dev_stats(device, path);
7809 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
7810 list_for_each_entry(device, &seed_devs->devices, dev_list) {
7811 ret = btrfs_device_init_dev_stats(device, path);
7817 mutex_unlock(&fs_devices->device_list_mutex);
7819 btrfs_free_path(path);
7823 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
7824 struct btrfs_device *device)
7826 struct btrfs_fs_info *fs_info = trans->fs_info;
7827 struct btrfs_root *dev_root = fs_info->dev_root;
7828 struct btrfs_path *path;
7829 struct btrfs_key key;
7830 struct extent_buffer *eb;
7831 struct btrfs_dev_stats_item *ptr;
7835 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7836 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7837 key.offset = device->devid;
7839 path = btrfs_alloc_path();
7842 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
7844 btrfs_warn_in_rcu(fs_info,
7845 "error %d while searching for dev_stats item for device %s",
7846 ret, rcu_str_deref(device->name));
7851 btrfs_item_size(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
7852 /* need to delete old one and insert a new one */
7853 ret = btrfs_del_item(trans, dev_root, path);
7855 btrfs_warn_in_rcu(fs_info,
7856 "delete too small dev_stats item for device %s failed %d",
7857 rcu_str_deref(device->name), ret);
7864 /* need to insert a new item */
7865 btrfs_release_path(path);
7866 ret = btrfs_insert_empty_item(trans, dev_root, path,
7867 &key, sizeof(*ptr));
7869 btrfs_warn_in_rcu(fs_info,
7870 "insert dev_stats item for device %s failed %d",
7871 rcu_str_deref(device->name), ret);
7876 eb = path->nodes[0];
7877 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7878 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7879 btrfs_set_dev_stats_value(eb, ptr, i,
7880 btrfs_dev_stat_read(device, i));
7881 btrfs_mark_buffer_dirty(eb);
7884 btrfs_free_path(path);
7889 * called from commit_transaction. Writes all changed device stats to disk.
7891 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans)
7893 struct btrfs_fs_info *fs_info = trans->fs_info;
7894 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7895 struct btrfs_device *device;
7899 mutex_lock(&fs_devices->device_list_mutex);
7900 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7901 stats_cnt = atomic_read(&device->dev_stats_ccnt);
7902 if (!device->dev_stats_valid || stats_cnt == 0)
7907 * There is a LOAD-LOAD control dependency between the value of
7908 * dev_stats_ccnt and updating the on-disk values which requires
7909 * reading the in-memory counters. Such control dependencies
7910 * require explicit read memory barriers.
7912 * This memory barriers pairs with smp_mb__before_atomic in
7913 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full
7914 * barrier implied by atomic_xchg in
7915 * btrfs_dev_stats_read_and_reset
7919 ret = update_dev_stat_item(trans, device);
7921 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7923 mutex_unlock(&fs_devices->device_list_mutex);
7928 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7930 btrfs_dev_stat_inc(dev, index);
7931 btrfs_dev_stat_print_on_error(dev);
7934 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
7936 if (!dev->dev_stats_valid)
7938 btrfs_err_rl_in_rcu(dev->fs_info,
7939 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7940 rcu_str_deref(dev->name),
7941 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7942 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7943 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7944 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7945 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7948 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7952 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7953 if (btrfs_dev_stat_read(dev, i) != 0)
7955 if (i == BTRFS_DEV_STAT_VALUES_MAX)
7956 return; /* all values == 0, suppress message */
7958 btrfs_info_in_rcu(dev->fs_info,
7959 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7960 rcu_str_deref(dev->name),
7961 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7962 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7963 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7964 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7965 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7968 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
7969 struct btrfs_ioctl_get_dev_stats *stats)
7971 BTRFS_DEV_LOOKUP_ARGS(args);
7972 struct btrfs_device *dev;
7973 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7976 mutex_lock(&fs_devices->device_list_mutex);
7977 args.devid = stats->devid;
7978 dev = btrfs_find_device(fs_info->fs_devices, &args);
7979 mutex_unlock(&fs_devices->device_list_mutex);
7982 btrfs_warn(fs_info, "get dev_stats failed, device not found");
7984 } else if (!dev->dev_stats_valid) {
7985 btrfs_warn(fs_info, "get dev_stats failed, not yet valid");
7987 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7988 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7989 if (stats->nr_items > i)
7991 btrfs_dev_stat_read_and_reset(dev, i);
7993 btrfs_dev_stat_set(dev, i, 0);
7995 btrfs_info(fs_info, "device stats zeroed by %s (%d)",
7996 current->comm, task_pid_nr(current));
7998 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7999 if (stats->nr_items > i)
8000 stats->values[i] = btrfs_dev_stat_read(dev, i);
8002 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
8003 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
8008 * Update the size and bytes used for each device where it changed. This is
8009 * delayed since we would otherwise get errors while writing out the
8012 * Must be invoked during transaction commit.
8014 void btrfs_commit_device_sizes(struct btrfs_transaction *trans)
8016 struct btrfs_device *curr, *next;
8018 ASSERT(trans->state == TRANS_STATE_COMMIT_DOING);
8020 if (list_empty(&trans->dev_update_list))
8024 * We don't need the device_list_mutex here. This list is owned by the
8025 * transaction and the transaction must complete before the device is
8028 mutex_lock(&trans->fs_info->chunk_mutex);
8029 list_for_each_entry_safe(curr, next, &trans->dev_update_list,
8031 list_del_init(&curr->post_commit_list);
8032 curr->commit_total_bytes = curr->disk_total_bytes;
8033 curr->commit_bytes_used = curr->bytes_used;
8035 mutex_unlock(&trans->fs_info->chunk_mutex);
8039 * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10.
8041 int btrfs_bg_type_to_factor(u64 flags)
8043 const int index = btrfs_bg_flags_to_raid_index(flags);
8045 return btrfs_raid_array[index].ncopies;
8050 static int verify_one_dev_extent(struct btrfs_fs_info *fs_info,
8051 u64 chunk_offset, u64 devid,
8052 u64 physical_offset, u64 physical_len)
8054 struct btrfs_dev_lookup_args args = { .devid = devid };
8055 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
8056 struct extent_map *em;
8057 struct map_lookup *map;
8058 struct btrfs_device *dev;
8064 read_lock(&em_tree->lock);
8065 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
8066 read_unlock(&em_tree->lock);
8070 "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk",
8071 physical_offset, devid);
8076 map = em->map_lookup;
8077 stripe_len = calc_stripe_length(map->type, em->len, map->num_stripes);
8078 if (physical_len != stripe_len) {
8080 "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu",
8081 physical_offset, devid, em->start, physical_len,
8087 for (i = 0; i < map->num_stripes; i++) {
8088 if (map->stripes[i].dev->devid == devid &&
8089 map->stripes[i].physical == physical_offset) {
8091 if (map->verified_stripes >= map->num_stripes) {
8093 "too many dev extents for chunk %llu found",
8098 map->verified_stripes++;
8104 "dev extent physical offset %llu devid %llu has no corresponding chunk",
8105 physical_offset, devid);
8109 /* Make sure no dev extent is beyond device boundary */
8110 dev = btrfs_find_device(fs_info->fs_devices, &args);
8112 btrfs_err(fs_info, "failed to find devid %llu", devid);
8117 if (physical_offset + physical_len > dev->disk_total_bytes) {
8119 "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu",
8120 devid, physical_offset, physical_len,
8121 dev->disk_total_bytes);
8126 if (dev->zone_info) {
8127 u64 zone_size = dev->zone_info->zone_size;
8129 if (!IS_ALIGNED(physical_offset, zone_size) ||
8130 !IS_ALIGNED(physical_len, zone_size)) {
8132 "zoned: dev extent devid %llu physical offset %llu len %llu is not aligned to device zone",
8133 devid, physical_offset, physical_len);
8140 free_extent_map(em);
8144 static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info)
8146 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
8147 struct extent_map *em;
8148 struct rb_node *node;
8151 read_lock(&em_tree->lock);
8152 for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) {
8153 em = rb_entry(node, struct extent_map, rb_node);
8154 if (em->map_lookup->num_stripes !=
8155 em->map_lookup->verified_stripes) {
8157 "chunk %llu has missing dev extent, have %d expect %d",
8158 em->start, em->map_lookup->verified_stripes,
8159 em->map_lookup->num_stripes);
8165 read_unlock(&em_tree->lock);
8170 * Ensure that all dev extents are mapped to correct chunk, otherwise
8171 * later chunk allocation/free would cause unexpected behavior.
8173 * NOTE: This will iterate through the whole device tree, which should be of
8174 * the same size level as the chunk tree. This slightly increases mount time.
8176 int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info)
8178 struct btrfs_path *path;
8179 struct btrfs_root *root = fs_info->dev_root;
8180 struct btrfs_key key;
8182 u64 prev_dev_ext_end = 0;
8186 * We don't have a dev_root because we mounted with ignorebadroots and
8187 * failed to load the root, so we want to skip the verification in this
8190 * However if the dev root is fine, but the tree itself is corrupted
8191 * we'd still fail to mount. This verification is only to make sure
8192 * writes can happen safely, so instead just bypass this check
8193 * completely in the case of IGNOREBADROOTS.
8195 if (btrfs_test_opt(fs_info, IGNOREBADROOTS))
8199 key.type = BTRFS_DEV_EXTENT_KEY;
8202 path = btrfs_alloc_path();
8206 path->reada = READA_FORWARD;
8207 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
8211 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
8212 ret = btrfs_next_leaf(root, path);
8215 /* No dev extents at all? Not good */
8222 struct extent_buffer *leaf = path->nodes[0];
8223 struct btrfs_dev_extent *dext;
8224 int slot = path->slots[0];
8226 u64 physical_offset;
8230 btrfs_item_key_to_cpu(leaf, &key, slot);
8231 if (key.type != BTRFS_DEV_EXTENT_KEY)
8233 devid = key.objectid;
8234 physical_offset = key.offset;
8236 dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent);
8237 chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext);
8238 physical_len = btrfs_dev_extent_length(leaf, dext);
8240 /* Check if this dev extent overlaps with the previous one */
8241 if (devid == prev_devid && physical_offset < prev_dev_ext_end) {
8243 "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu",
8244 devid, physical_offset, prev_dev_ext_end);
8249 ret = verify_one_dev_extent(fs_info, chunk_offset, devid,
8250 physical_offset, physical_len);
8254 prev_dev_ext_end = physical_offset + physical_len;
8256 ret = btrfs_next_item(root, path);
8265 /* Ensure all chunks have corresponding dev extents */
8266 ret = verify_chunk_dev_extent_mapping(fs_info);
8268 btrfs_free_path(path);
8273 * Check whether the given block group or device is pinned by any inode being
8274 * used as a swapfile.
8276 bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr)
8278 struct btrfs_swapfile_pin *sp;
8279 struct rb_node *node;
8281 spin_lock(&fs_info->swapfile_pins_lock);
8282 node = fs_info->swapfile_pins.rb_node;
8284 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
8286 node = node->rb_left;
8287 else if (ptr > sp->ptr)
8288 node = node->rb_right;
8292 spin_unlock(&fs_info->swapfile_pins_lock);
8293 return node != NULL;
8296 static int relocating_repair_kthread(void *data)
8298 struct btrfs_block_group *cache = (struct btrfs_block_group *)data;
8299 struct btrfs_fs_info *fs_info = cache->fs_info;
8303 target = cache->start;
8304 btrfs_put_block_group(cache);
8306 sb_start_write(fs_info->sb);
8307 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) {
8309 "zoned: skip relocating block group %llu to repair: EBUSY",
8311 sb_end_write(fs_info->sb);
8315 mutex_lock(&fs_info->reclaim_bgs_lock);
8317 /* Ensure block group still exists */
8318 cache = btrfs_lookup_block_group(fs_info, target);
8322 if (!cache->relocating_repair)
8325 ret = btrfs_may_alloc_data_chunk(fs_info, target);
8330 "zoned: relocating block group %llu to repair IO failure",
8332 ret = btrfs_relocate_chunk(fs_info, target);
8336 btrfs_put_block_group(cache);
8337 mutex_unlock(&fs_info->reclaim_bgs_lock);
8338 btrfs_exclop_finish(fs_info);
8339 sb_end_write(fs_info->sb);
8344 bool btrfs_repair_one_zone(struct btrfs_fs_info *fs_info, u64 logical)
8346 struct btrfs_block_group *cache;
8348 if (!btrfs_is_zoned(fs_info))
8351 /* Do not attempt to repair in degraded state */
8352 if (btrfs_test_opt(fs_info, DEGRADED))
8355 cache = btrfs_lookup_block_group(fs_info, logical);
8359 spin_lock(&cache->lock);
8360 if (cache->relocating_repair) {
8361 spin_unlock(&cache->lock);
8362 btrfs_put_block_group(cache);
8365 cache->relocating_repair = 1;
8366 spin_unlock(&cache->lock);
8368 kthread_run(relocating_repair_kthread, cache,
8369 "btrfs-relocating-repair");