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 const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
38 [BTRFS_RAID_RAID10] = {
41 .devs_max = 0, /* 0 == as many as possible */
43 .tolerated_failures = 1,
47 .raid_name = "raid10",
48 .bg_flag = BTRFS_BLOCK_GROUP_RAID10,
49 .mindev_error = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET,
51 [BTRFS_RAID_RAID1] = {
56 .tolerated_failures = 1,
61 .bg_flag = BTRFS_BLOCK_GROUP_RAID1,
62 .mindev_error = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET,
64 [BTRFS_RAID_RAID1C3] = {
69 .tolerated_failures = 2,
73 .raid_name = "raid1c3",
74 .bg_flag = BTRFS_BLOCK_GROUP_RAID1C3,
75 .mindev_error = BTRFS_ERROR_DEV_RAID1C3_MIN_NOT_MET,
77 [BTRFS_RAID_RAID1C4] = {
82 .tolerated_failures = 3,
86 .raid_name = "raid1c4",
87 .bg_flag = BTRFS_BLOCK_GROUP_RAID1C4,
88 .mindev_error = BTRFS_ERROR_DEV_RAID1C4_MIN_NOT_MET,
95 .tolerated_failures = 0,
100 .bg_flag = BTRFS_BLOCK_GROUP_DUP,
103 [BTRFS_RAID_RAID0] = {
108 .tolerated_failures = 0,
112 .raid_name = "raid0",
113 .bg_flag = BTRFS_BLOCK_GROUP_RAID0,
116 [BTRFS_RAID_SINGLE] = {
121 .tolerated_failures = 0,
125 .raid_name = "single",
129 [BTRFS_RAID_RAID5] = {
134 .tolerated_failures = 1,
138 .raid_name = "raid5",
139 .bg_flag = BTRFS_BLOCK_GROUP_RAID5,
140 .mindev_error = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET,
142 [BTRFS_RAID_RAID6] = {
147 .tolerated_failures = 2,
151 .raid_name = "raid6",
152 .bg_flag = BTRFS_BLOCK_GROUP_RAID6,
153 .mindev_error = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET,
158 * Convert block group flags (BTRFS_BLOCK_GROUP_*) to btrfs_raid_types, which
159 * can be used as index to access btrfs_raid_array[].
161 enum btrfs_raid_types __attribute_const__ btrfs_bg_flags_to_raid_index(u64 flags)
163 if (flags & BTRFS_BLOCK_GROUP_RAID10)
164 return BTRFS_RAID_RAID10;
165 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
166 return BTRFS_RAID_RAID1;
167 else if (flags & BTRFS_BLOCK_GROUP_RAID1C3)
168 return BTRFS_RAID_RAID1C3;
169 else if (flags & BTRFS_BLOCK_GROUP_RAID1C4)
170 return BTRFS_RAID_RAID1C4;
171 else if (flags & BTRFS_BLOCK_GROUP_DUP)
172 return BTRFS_RAID_DUP;
173 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
174 return BTRFS_RAID_RAID0;
175 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
176 return BTRFS_RAID_RAID5;
177 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
178 return BTRFS_RAID_RAID6;
180 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
183 const char *btrfs_bg_type_to_raid_name(u64 flags)
185 const int index = btrfs_bg_flags_to_raid_index(flags);
187 if (index >= BTRFS_NR_RAID_TYPES)
190 return btrfs_raid_array[index].raid_name;
194 * Fill @buf with textual description of @bg_flags, no more than @size_buf
195 * bytes including terminating null byte.
197 void btrfs_describe_block_groups(u64 bg_flags, char *buf, u32 size_buf)
202 u64 flags = bg_flags;
203 u32 size_bp = size_buf;
210 #define DESCRIBE_FLAG(flag, desc) \
212 if (flags & (flag)) { \
213 ret = snprintf(bp, size_bp, "%s|", (desc)); \
214 if (ret < 0 || ret >= size_bp) \
222 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_DATA, "data");
223 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_SYSTEM, "system");
224 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_METADATA, "metadata");
226 DESCRIBE_FLAG(BTRFS_AVAIL_ALLOC_BIT_SINGLE, "single");
227 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
228 DESCRIBE_FLAG(btrfs_raid_array[i].bg_flag,
229 btrfs_raid_array[i].raid_name);
233 ret = snprintf(bp, size_bp, "0x%llx|", flags);
237 if (size_bp < size_buf)
238 buf[size_buf - size_bp - 1] = '\0'; /* remove last | */
241 * The text is trimmed, it's up to the caller to provide sufficiently
247 static int init_first_rw_device(struct btrfs_trans_handle *trans);
248 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info);
249 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
250 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
251 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
252 enum btrfs_map_op op,
253 u64 logical, u64 *length,
254 struct btrfs_io_context **bioc_ret,
255 int mirror_num, int need_raid_map);
261 * There are several mutexes that protect manipulation of devices and low-level
262 * structures like chunks but not block groups, extents or files
264 * uuid_mutex (global lock)
265 * ------------------------
266 * protects the fs_uuids list that tracks all per-fs fs_devices, resulting from
267 * the SCAN_DEV ioctl registration or from mount either implicitly (the first
268 * device) or requested by the device= mount option
270 * the mutex can be very coarse and can cover long-running operations
272 * protects: updates to fs_devices counters like missing devices, rw devices,
273 * seeding, structure cloning, opening/closing devices at mount/umount time
275 * global::fs_devs - add, remove, updates to the global list
277 * does not protect: manipulation of the fs_devices::devices list in general
278 * but in mount context it could be used to exclude list modifications by eg.
281 * btrfs_device::name - renames (write side), read is RCU
283 * fs_devices::device_list_mutex (per-fs, with RCU)
284 * ------------------------------------------------
285 * protects updates to fs_devices::devices, ie. adding and deleting
287 * simple list traversal with read-only actions can be done with RCU protection
289 * may be used to exclude some operations from running concurrently without any
290 * modifications to the list (see write_all_supers)
292 * Is not required at mount and close times, because our device list is
293 * protected by the uuid_mutex at that point.
297 * protects balance structures (status, state) and context accessed from
298 * several places (internally, ioctl)
302 * protects chunks, adding or removing during allocation, trim or when a new
303 * device is added/removed. Additionally it also protects post_commit_list of
304 * individual devices, since they can be added to the transaction's
305 * post_commit_list only with chunk_mutex held.
309 * a big lock that is held by the cleaner thread and prevents running subvolume
310 * cleaning together with relocation or delayed iputs
322 * Exclusive operations
323 * ====================
325 * Maintains the exclusivity of the following operations that apply to the
326 * whole filesystem and cannot run in parallel.
331 * - Device replace (*)
334 * The device operations (as above) can be in one of the following states:
340 * Only device operations marked with (*) can go into the Paused state for the
343 * - ioctl (only Balance can be Paused through ioctl)
344 * - filesystem remounted as read-only
345 * - filesystem unmounted and mounted as read-only
346 * - system power-cycle and filesystem mounted as read-only
347 * - filesystem or device errors leading to forced read-only
349 * The status of exclusive operation is set and cleared atomically.
350 * During the course of Paused state, fs_info::exclusive_operation remains set.
351 * A device operation in Paused or Running state can be canceled or resumed
352 * either by ioctl (Balance only) or when remounted as read-write.
353 * The exclusive status is cleared when the device operation is canceled or
357 DEFINE_MUTEX(uuid_mutex);
358 static LIST_HEAD(fs_uuids);
359 struct list_head * __attribute_const__ btrfs_get_fs_uuids(void)
365 * alloc_fs_devices - allocate struct btrfs_fs_devices
366 * @fsid: if not NULL, copy the UUID to fs_devices::fsid
367 * @metadata_fsid: if not NULL, copy the UUID to fs_devices::metadata_fsid
369 * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR().
370 * The returned struct is not linked onto any lists and can be destroyed with
371 * kfree() right away.
373 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid,
374 const u8 *metadata_fsid)
376 struct btrfs_fs_devices *fs_devs;
378 fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL);
380 return ERR_PTR(-ENOMEM);
382 mutex_init(&fs_devs->device_list_mutex);
384 INIT_LIST_HEAD(&fs_devs->devices);
385 INIT_LIST_HEAD(&fs_devs->alloc_list);
386 INIT_LIST_HEAD(&fs_devs->fs_list);
387 INIT_LIST_HEAD(&fs_devs->seed_list);
389 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
392 memcpy(fs_devs->metadata_uuid, metadata_fsid, BTRFS_FSID_SIZE);
394 memcpy(fs_devs->metadata_uuid, fsid, BTRFS_FSID_SIZE);
399 void btrfs_free_device(struct btrfs_device *device)
401 WARN_ON(!list_empty(&device->post_commit_list));
402 rcu_string_free(device->name);
403 extent_io_tree_release(&device->alloc_state);
404 bio_put(device->flush_bio);
405 btrfs_destroy_dev_zone_info(device);
409 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
411 struct btrfs_device *device;
412 WARN_ON(fs_devices->opened);
413 while (!list_empty(&fs_devices->devices)) {
414 device = list_entry(fs_devices->devices.next,
415 struct btrfs_device, dev_list);
416 list_del(&device->dev_list);
417 btrfs_free_device(device);
422 void __exit btrfs_cleanup_fs_uuids(void)
424 struct btrfs_fs_devices *fs_devices;
426 while (!list_empty(&fs_uuids)) {
427 fs_devices = list_entry(fs_uuids.next,
428 struct btrfs_fs_devices, fs_list);
429 list_del(&fs_devices->fs_list);
430 free_fs_devices(fs_devices);
434 static noinline struct btrfs_fs_devices *find_fsid(
435 const u8 *fsid, const u8 *metadata_fsid)
437 struct btrfs_fs_devices *fs_devices;
441 /* Handle non-split brain cases */
442 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
444 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0
445 && memcmp(metadata_fsid, fs_devices->metadata_uuid,
446 BTRFS_FSID_SIZE) == 0)
449 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
456 static struct btrfs_fs_devices *find_fsid_with_metadata_uuid(
457 struct btrfs_super_block *disk_super)
460 struct btrfs_fs_devices *fs_devices;
463 * Handle scanned device having completed its fsid change but
464 * belonging to a fs_devices that was created by first scanning
465 * a device which didn't have its fsid/metadata_uuid changed
466 * at all and the CHANGING_FSID_V2 flag set.
468 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
469 if (fs_devices->fsid_change &&
470 memcmp(disk_super->metadata_uuid, fs_devices->fsid,
471 BTRFS_FSID_SIZE) == 0 &&
472 memcmp(fs_devices->fsid, fs_devices->metadata_uuid,
473 BTRFS_FSID_SIZE) == 0) {
478 * Handle scanned device having completed its fsid change but
479 * belonging to a fs_devices that was created by a device that
480 * has an outdated pair of fsid/metadata_uuid and
481 * CHANGING_FSID_V2 flag set.
483 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
484 if (fs_devices->fsid_change &&
485 memcmp(fs_devices->metadata_uuid,
486 fs_devices->fsid, BTRFS_FSID_SIZE) != 0 &&
487 memcmp(disk_super->metadata_uuid, fs_devices->metadata_uuid,
488 BTRFS_FSID_SIZE) == 0) {
493 return find_fsid(disk_super->fsid, disk_super->metadata_uuid);
498 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
499 int flush, struct block_device **bdev,
500 struct btrfs_super_block **disk_super)
504 *bdev = blkdev_get_by_path(device_path, flags, holder);
507 ret = PTR_ERR(*bdev);
512 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
513 ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE);
515 blkdev_put(*bdev, flags);
518 invalidate_bdev(*bdev);
519 *disk_super = btrfs_read_dev_super(*bdev);
520 if (IS_ERR(*disk_super)) {
521 ret = PTR_ERR(*disk_super);
522 blkdev_put(*bdev, flags);
534 * Check if the device in the path matches the device in the given struct device.
537 * true If it is the same device.
538 * false If it is not the same device or on error.
540 static bool device_matched(const struct btrfs_device *device, const char *path)
548 * If we are looking for a device with the matching dev_t, then skip
549 * device without a name (a missing device).
554 device_name = kzalloc(BTRFS_PATH_NAME_MAX, GFP_KERNEL);
559 scnprintf(device_name, BTRFS_PATH_NAME_MAX, "%s", rcu_str_deref(device->name));
562 ret = lookup_bdev(device_name, &dev_old);
567 ret = lookup_bdev(path, &dev_new);
571 if (dev_old == dev_new)
578 * Search and remove all stale (devices which are not mounted) devices.
579 * When both inputs are NULL, it will search and release all stale devices.
580 * path: Optional. When provided will it release all unmounted devices
581 * matching this path only.
582 * skip_dev: Optional. Will skip this device when searching for the stale
584 * Return: 0 for success or if @path is NULL.
585 * -EBUSY if @path is a mounted device.
586 * -ENOENT if @path does not match any device in the list.
588 static int btrfs_free_stale_devices(const char *path,
589 struct btrfs_device *skip_device)
591 struct btrfs_fs_devices *fs_devices, *tmp_fs_devices;
592 struct btrfs_device *device, *tmp_device;
595 lockdep_assert_held(&uuid_mutex);
600 list_for_each_entry_safe(fs_devices, tmp_fs_devices, &fs_uuids, fs_list) {
602 mutex_lock(&fs_devices->device_list_mutex);
603 list_for_each_entry_safe(device, tmp_device,
604 &fs_devices->devices, dev_list) {
605 if (skip_device && skip_device == device)
607 if (path && !device_matched(device, path))
609 if (fs_devices->opened) {
610 /* for an already deleted device return 0 */
611 if (path && ret != 0)
616 /* delete the stale device */
617 fs_devices->num_devices--;
618 list_del(&device->dev_list);
619 btrfs_free_device(device);
623 mutex_unlock(&fs_devices->device_list_mutex);
625 if (fs_devices->num_devices == 0) {
626 btrfs_sysfs_remove_fsid(fs_devices);
627 list_del(&fs_devices->fs_list);
628 free_fs_devices(fs_devices);
636 * This is only used on mount, and we are protected from competing things
637 * messing with our fs_devices by the uuid_mutex, thus we do not need the
638 * fs_devices->device_list_mutex here.
640 static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices,
641 struct btrfs_device *device, fmode_t flags,
644 struct request_queue *q;
645 struct block_device *bdev;
646 struct btrfs_super_block *disk_super;
655 ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
660 devid = btrfs_stack_device_id(&disk_super->dev_item);
661 if (devid != device->devid)
662 goto error_free_page;
664 if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE))
665 goto error_free_page;
667 device->generation = btrfs_super_generation(disk_super);
669 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
670 if (btrfs_super_incompat_flags(disk_super) &
671 BTRFS_FEATURE_INCOMPAT_METADATA_UUID) {
673 "BTRFS: Invalid seeding and uuid-changed device detected\n");
674 goto error_free_page;
677 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
678 fs_devices->seeding = true;
680 if (bdev_read_only(bdev))
681 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
683 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
686 q = bdev_get_queue(bdev);
687 if (!blk_queue_nonrot(q))
688 fs_devices->rotating = true;
691 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
692 device->mode = flags;
694 fs_devices->open_devices++;
695 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
696 device->devid != BTRFS_DEV_REPLACE_DEVID) {
697 fs_devices->rw_devices++;
698 list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list);
700 btrfs_release_disk_super(disk_super);
705 btrfs_release_disk_super(disk_super);
706 blkdev_put(bdev, flags);
712 * Handle scanned device having its CHANGING_FSID_V2 flag set and the fs_devices
713 * being created with a disk that has already completed its fsid change. Such
714 * disk can belong to an fs which has its FSID changed or to one which doesn't.
715 * Handle both cases here.
717 static struct btrfs_fs_devices *find_fsid_inprogress(
718 struct btrfs_super_block *disk_super)
720 struct btrfs_fs_devices *fs_devices;
722 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
723 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid,
724 BTRFS_FSID_SIZE) != 0 &&
725 memcmp(fs_devices->metadata_uuid, disk_super->fsid,
726 BTRFS_FSID_SIZE) == 0 && !fs_devices->fsid_change) {
731 return find_fsid(disk_super->fsid, NULL);
735 static struct btrfs_fs_devices *find_fsid_changed(
736 struct btrfs_super_block *disk_super)
738 struct btrfs_fs_devices *fs_devices;
741 * Handles the case where scanned device is part of an fs that had
742 * multiple successful changes of FSID but currently device didn't
743 * observe it. Meaning our fsid will be different than theirs. We need
744 * to handle two subcases :
745 * 1 - The fs still continues to have different METADATA/FSID uuids.
746 * 2 - The fs is switched back to its original FSID (METADATA/FSID
749 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
751 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid,
752 BTRFS_FSID_SIZE) != 0 &&
753 memcmp(fs_devices->metadata_uuid, disk_super->metadata_uuid,
754 BTRFS_FSID_SIZE) == 0 &&
755 memcmp(fs_devices->fsid, disk_super->fsid,
756 BTRFS_FSID_SIZE) != 0)
759 /* Unchanged UUIDs */
760 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid,
761 BTRFS_FSID_SIZE) == 0 &&
762 memcmp(fs_devices->fsid, disk_super->metadata_uuid,
763 BTRFS_FSID_SIZE) == 0)
770 static struct btrfs_fs_devices *find_fsid_reverted_metadata(
771 struct btrfs_super_block *disk_super)
773 struct btrfs_fs_devices *fs_devices;
776 * Handle the case where the scanned device is part of an fs whose last
777 * metadata UUID change reverted it to the original FSID. At the same
778 * time * fs_devices was first created by another constitutent device
779 * which didn't fully observe the operation. This results in an
780 * btrfs_fs_devices created with metadata/fsid different AND
781 * btrfs_fs_devices::fsid_change set AND the metadata_uuid of the
782 * fs_devices equal to the FSID of the disk.
784 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
785 if (memcmp(fs_devices->fsid, fs_devices->metadata_uuid,
786 BTRFS_FSID_SIZE) != 0 &&
787 memcmp(fs_devices->metadata_uuid, disk_super->fsid,
788 BTRFS_FSID_SIZE) == 0 &&
789 fs_devices->fsid_change)
796 * Add new device to list of registered devices
799 * device pointer which was just added or updated when successful
800 * error pointer when failed
802 static noinline struct btrfs_device *device_list_add(const char *path,
803 struct btrfs_super_block *disk_super,
804 bool *new_device_added)
806 struct btrfs_device *device;
807 struct btrfs_fs_devices *fs_devices = NULL;
808 struct rcu_string *name;
809 u64 found_transid = btrfs_super_generation(disk_super);
810 u64 devid = btrfs_stack_device_id(&disk_super->dev_item);
811 bool has_metadata_uuid = (btrfs_super_incompat_flags(disk_super) &
812 BTRFS_FEATURE_INCOMPAT_METADATA_UUID);
813 bool fsid_change_in_progress = (btrfs_super_flags(disk_super) &
814 BTRFS_SUPER_FLAG_CHANGING_FSID_V2);
816 if (fsid_change_in_progress) {
817 if (!has_metadata_uuid)
818 fs_devices = find_fsid_inprogress(disk_super);
820 fs_devices = find_fsid_changed(disk_super);
821 } else if (has_metadata_uuid) {
822 fs_devices = find_fsid_with_metadata_uuid(disk_super);
824 fs_devices = find_fsid_reverted_metadata(disk_super);
826 fs_devices = find_fsid(disk_super->fsid, NULL);
831 if (has_metadata_uuid)
832 fs_devices = alloc_fs_devices(disk_super->fsid,
833 disk_super->metadata_uuid);
835 fs_devices = alloc_fs_devices(disk_super->fsid, NULL);
837 if (IS_ERR(fs_devices))
838 return ERR_CAST(fs_devices);
840 fs_devices->fsid_change = fsid_change_in_progress;
842 mutex_lock(&fs_devices->device_list_mutex);
843 list_add(&fs_devices->fs_list, &fs_uuids);
847 struct btrfs_dev_lookup_args args = {
849 .uuid = disk_super->dev_item.uuid,
852 mutex_lock(&fs_devices->device_list_mutex);
853 device = btrfs_find_device(fs_devices, &args);
856 * If this disk has been pulled into an fs devices created by
857 * a device which had the CHANGING_FSID_V2 flag then replace the
858 * metadata_uuid/fsid values of the fs_devices.
860 if (fs_devices->fsid_change &&
861 found_transid > fs_devices->latest_generation) {
862 memcpy(fs_devices->fsid, disk_super->fsid,
865 if (has_metadata_uuid)
866 memcpy(fs_devices->metadata_uuid,
867 disk_super->metadata_uuid,
870 memcpy(fs_devices->metadata_uuid,
871 disk_super->fsid, BTRFS_FSID_SIZE);
873 fs_devices->fsid_change = false;
878 if (fs_devices->opened) {
879 mutex_unlock(&fs_devices->device_list_mutex);
880 return ERR_PTR(-EBUSY);
883 device = btrfs_alloc_device(NULL, &devid,
884 disk_super->dev_item.uuid);
885 if (IS_ERR(device)) {
886 mutex_unlock(&fs_devices->device_list_mutex);
887 /* we can safely leave the fs_devices entry around */
891 name = rcu_string_strdup(path, GFP_NOFS);
893 btrfs_free_device(device);
894 mutex_unlock(&fs_devices->device_list_mutex);
895 return ERR_PTR(-ENOMEM);
897 rcu_assign_pointer(device->name, name);
899 list_add_rcu(&device->dev_list, &fs_devices->devices);
900 fs_devices->num_devices++;
902 device->fs_devices = fs_devices;
903 *new_device_added = true;
905 if (disk_super->label[0])
907 "BTRFS: device label %s devid %llu transid %llu %s scanned by %s (%d)\n",
908 disk_super->label, devid, found_transid, path,
909 current->comm, task_pid_nr(current));
912 "BTRFS: device fsid %pU devid %llu transid %llu %s scanned by %s (%d)\n",
913 disk_super->fsid, devid, found_transid, path,
914 current->comm, task_pid_nr(current));
916 } else if (!device->name || strcmp(device->name->str, path)) {
918 * When FS is already mounted.
919 * 1. If you are here and if the device->name is NULL that
920 * means this device was missing at time of FS mount.
921 * 2. If you are here and if the device->name is different
922 * from 'path' that means either
923 * a. The same device disappeared and reappeared with
925 * b. The missing-disk-which-was-replaced, has
928 * We must allow 1 and 2a above. But 2b would be a spurious
931 * Further in case of 1 and 2a above, the disk at 'path'
932 * would have missed some transaction when it was away and
933 * in case of 2a the stale bdev has to be updated as well.
934 * 2b must not be allowed at all time.
938 * For now, we do allow update to btrfs_fs_device through the
939 * btrfs dev scan cli after FS has been mounted. We're still
940 * tracking a problem where systems fail mount by subvolume id
941 * when we reject replacement on a mounted FS.
943 if (!fs_devices->opened && found_transid < device->generation) {
945 * That is if the FS is _not_ mounted and if you
946 * are here, that means there is more than one
947 * disk with same uuid and devid.We keep the one
948 * with larger generation number or the last-in if
949 * generation are equal.
951 mutex_unlock(&fs_devices->device_list_mutex);
952 return ERR_PTR(-EEXIST);
956 * We are going to replace the device path for a given devid,
957 * make sure it's the same device if the device is mounted
959 * NOTE: the device->fs_info may not be reliable here so pass
960 * in a NULL to message helpers instead. This avoids a possible
961 * use-after-free when the fs_info and fs_info->sb are already
968 error = lookup_bdev(path, &path_dev);
970 mutex_unlock(&fs_devices->device_list_mutex);
971 return ERR_PTR(error);
974 if (device->bdev->bd_dev != path_dev) {
975 mutex_unlock(&fs_devices->device_list_mutex);
976 btrfs_warn_in_rcu(NULL,
977 "duplicate device %s devid %llu generation %llu scanned by %s (%d)",
978 path, devid, found_transid,
980 task_pid_nr(current));
981 return ERR_PTR(-EEXIST);
983 btrfs_info_in_rcu(NULL,
984 "devid %llu device path %s changed to %s scanned by %s (%d)",
985 devid, rcu_str_deref(device->name),
987 task_pid_nr(current));
990 name = rcu_string_strdup(path, GFP_NOFS);
992 mutex_unlock(&fs_devices->device_list_mutex);
993 return ERR_PTR(-ENOMEM);
995 rcu_string_free(device->name);
996 rcu_assign_pointer(device->name, name);
997 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
998 fs_devices->missing_devices--;
999 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
1004 * Unmount does not free the btrfs_device struct but would zero
1005 * generation along with most of the other members. So just update
1006 * it back. We need it to pick the disk with largest generation
1009 if (!fs_devices->opened) {
1010 device->generation = found_transid;
1011 fs_devices->latest_generation = max_t(u64, found_transid,
1012 fs_devices->latest_generation);
1015 fs_devices->total_devices = btrfs_super_num_devices(disk_super);
1017 mutex_unlock(&fs_devices->device_list_mutex);
1021 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
1023 struct btrfs_fs_devices *fs_devices;
1024 struct btrfs_device *device;
1025 struct btrfs_device *orig_dev;
1028 lockdep_assert_held(&uuid_mutex);
1030 fs_devices = alloc_fs_devices(orig->fsid, NULL);
1031 if (IS_ERR(fs_devices))
1034 fs_devices->total_devices = orig->total_devices;
1036 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
1037 struct rcu_string *name;
1039 device = btrfs_alloc_device(NULL, &orig_dev->devid,
1041 if (IS_ERR(device)) {
1042 ret = PTR_ERR(device);
1047 * This is ok to do without rcu read locked because we hold the
1048 * uuid mutex so nothing we touch in here is going to disappear.
1050 if (orig_dev->name) {
1051 name = rcu_string_strdup(orig_dev->name->str,
1054 btrfs_free_device(device);
1058 rcu_assign_pointer(device->name, name);
1061 list_add(&device->dev_list, &fs_devices->devices);
1062 device->fs_devices = fs_devices;
1063 fs_devices->num_devices++;
1067 free_fs_devices(fs_devices);
1068 return ERR_PTR(ret);
1071 static void __btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices,
1072 struct btrfs_device **latest_dev)
1074 struct btrfs_device *device, *next;
1076 /* This is the initialized path, it is safe to release the devices. */
1077 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
1078 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)) {
1079 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
1080 &device->dev_state) &&
1081 !test_bit(BTRFS_DEV_STATE_MISSING,
1082 &device->dev_state) &&
1084 device->generation > (*latest_dev)->generation)) {
1085 *latest_dev = device;
1091 * We have already validated the presence of BTRFS_DEV_REPLACE_DEVID,
1092 * in btrfs_init_dev_replace() so just continue.
1094 if (device->devid == BTRFS_DEV_REPLACE_DEVID)
1098 blkdev_put(device->bdev, device->mode);
1099 device->bdev = NULL;
1100 fs_devices->open_devices--;
1102 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1103 list_del_init(&device->dev_alloc_list);
1104 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
1105 fs_devices->rw_devices--;
1107 list_del_init(&device->dev_list);
1108 fs_devices->num_devices--;
1109 btrfs_free_device(device);
1115 * After we have read the system tree and know devids belonging to this
1116 * filesystem, remove the device which does not belong there.
1118 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices)
1120 struct btrfs_device *latest_dev = NULL;
1121 struct btrfs_fs_devices *seed_dev;
1123 mutex_lock(&uuid_mutex);
1124 __btrfs_free_extra_devids(fs_devices, &latest_dev);
1126 list_for_each_entry(seed_dev, &fs_devices->seed_list, seed_list)
1127 __btrfs_free_extra_devids(seed_dev, &latest_dev);
1129 fs_devices->latest_dev = latest_dev;
1131 mutex_unlock(&uuid_mutex);
1134 static void btrfs_close_bdev(struct btrfs_device *device)
1139 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1140 sync_blockdev(device->bdev);
1141 invalidate_bdev(device->bdev);
1144 blkdev_put(device->bdev, device->mode);
1147 static void btrfs_close_one_device(struct btrfs_device *device)
1149 struct btrfs_fs_devices *fs_devices = device->fs_devices;
1151 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
1152 device->devid != BTRFS_DEV_REPLACE_DEVID) {
1153 list_del_init(&device->dev_alloc_list);
1154 fs_devices->rw_devices--;
1157 if (device->devid == BTRFS_DEV_REPLACE_DEVID)
1158 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
1160 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
1161 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
1162 fs_devices->missing_devices--;
1165 btrfs_close_bdev(device);
1167 fs_devices->open_devices--;
1168 device->bdev = NULL;
1170 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
1171 btrfs_destroy_dev_zone_info(device);
1173 device->fs_info = NULL;
1174 atomic_set(&device->dev_stats_ccnt, 0);
1175 extent_io_tree_release(&device->alloc_state);
1178 * Reset the flush error record. We might have a transient flush error
1179 * in this mount, and if so we aborted the current transaction and set
1180 * the fs to an error state, guaranteeing no super blocks can be further
1181 * committed. However that error might be transient and if we unmount the
1182 * filesystem and mount it again, we should allow the mount to succeed
1183 * (btrfs_check_rw_degradable() should not fail) - if after mounting the
1184 * filesystem again we still get flush errors, then we will again abort
1185 * any transaction and set the error state, guaranteeing no commits of
1186 * unsafe super blocks.
1188 device->last_flush_error = 0;
1190 /* Verify the device is back in a pristine state */
1191 ASSERT(!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state));
1192 ASSERT(!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state));
1193 ASSERT(list_empty(&device->dev_alloc_list));
1194 ASSERT(list_empty(&device->post_commit_list));
1195 ASSERT(atomic_read(&device->reada_in_flight) == 0);
1198 static void close_fs_devices(struct btrfs_fs_devices *fs_devices)
1200 struct btrfs_device *device, *tmp;
1202 lockdep_assert_held(&uuid_mutex);
1204 if (--fs_devices->opened > 0)
1207 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list)
1208 btrfs_close_one_device(device);
1210 WARN_ON(fs_devices->open_devices);
1211 WARN_ON(fs_devices->rw_devices);
1212 fs_devices->opened = 0;
1213 fs_devices->seeding = false;
1214 fs_devices->fs_info = NULL;
1217 void btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
1220 struct btrfs_fs_devices *tmp;
1222 mutex_lock(&uuid_mutex);
1223 close_fs_devices(fs_devices);
1224 if (!fs_devices->opened)
1225 list_splice_init(&fs_devices->seed_list, &list);
1227 list_for_each_entry_safe(fs_devices, tmp, &list, seed_list) {
1228 close_fs_devices(fs_devices);
1229 list_del(&fs_devices->seed_list);
1230 free_fs_devices(fs_devices);
1232 mutex_unlock(&uuid_mutex);
1235 static int open_fs_devices(struct btrfs_fs_devices *fs_devices,
1236 fmode_t flags, void *holder)
1238 struct btrfs_device *device;
1239 struct btrfs_device *latest_dev = NULL;
1240 struct btrfs_device *tmp_device;
1242 flags |= FMODE_EXCL;
1244 list_for_each_entry_safe(device, tmp_device, &fs_devices->devices,
1248 ret = btrfs_open_one_device(fs_devices, device, flags, holder);
1250 (!latest_dev || device->generation > latest_dev->generation)) {
1251 latest_dev = device;
1252 } else if (ret == -ENODATA) {
1253 fs_devices->num_devices--;
1254 list_del(&device->dev_list);
1255 btrfs_free_device(device);
1258 if (fs_devices->open_devices == 0)
1261 fs_devices->opened = 1;
1262 fs_devices->latest_dev = latest_dev;
1263 fs_devices->total_rw_bytes = 0;
1264 fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_REGULAR;
1265 fs_devices->read_policy = BTRFS_READ_POLICY_PID;
1270 static int devid_cmp(void *priv, const struct list_head *a,
1271 const struct list_head *b)
1273 const struct btrfs_device *dev1, *dev2;
1275 dev1 = list_entry(a, struct btrfs_device, dev_list);
1276 dev2 = list_entry(b, struct btrfs_device, dev_list);
1278 if (dev1->devid < dev2->devid)
1280 else if (dev1->devid > dev2->devid)
1285 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1286 fmode_t flags, void *holder)
1290 lockdep_assert_held(&uuid_mutex);
1292 * The device_list_mutex cannot be taken here in case opening the
1293 * underlying device takes further locks like open_mutex.
1295 * We also don't need the lock here as this is called during mount and
1296 * exclusion is provided by uuid_mutex
1299 if (fs_devices->opened) {
1300 fs_devices->opened++;
1303 list_sort(NULL, &fs_devices->devices, devid_cmp);
1304 ret = open_fs_devices(fs_devices, flags, holder);
1310 void btrfs_release_disk_super(struct btrfs_super_block *super)
1312 struct page *page = virt_to_page(super);
1317 static struct btrfs_super_block *btrfs_read_disk_super(struct block_device *bdev,
1318 u64 bytenr, u64 bytenr_orig)
1320 struct btrfs_super_block *disk_super;
1325 /* make sure our super fits in the device */
1326 if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1327 return ERR_PTR(-EINVAL);
1329 /* make sure our super fits in the page */
1330 if (sizeof(*disk_super) > PAGE_SIZE)
1331 return ERR_PTR(-EINVAL);
1333 /* make sure our super doesn't straddle pages on disk */
1334 index = bytenr >> PAGE_SHIFT;
1335 if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_SHIFT != index)
1336 return ERR_PTR(-EINVAL);
1338 /* pull in the page with our super */
1339 page = read_cache_page_gfp(bdev->bd_inode->i_mapping, index, GFP_KERNEL);
1342 return ERR_CAST(page);
1344 p = page_address(page);
1346 /* align our pointer to the offset of the super block */
1347 disk_super = p + offset_in_page(bytenr);
1349 if (btrfs_super_bytenr(disk_super) != bytenr_orig ||
1350 btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
1351 btrfs_release_disk_super(p);
1352 return ERR_PTR(-EINVAL);
1355 if (disk_super->label[0] && disk_super->label[BTRFS_LABEL_SIZE - 1])
1356 disk_super->label[BTRFS_LABEL_SIZE - 1] = 0;
1361 int btrfs_forget_devices(const char *path)
1365 mutex_lock(&uuid_mutex);
1366 ret = btrfs_free_stale_devices(strlen(path) ? path : NULL, NULL);
1367 mutex_unlock(&uuid_mutex);
1373 * Look for a btrfs signature on a device. This may be called out of the mount path
1374 * and we are not allowed to call set_blocksize during the scan. The superblock
1375 * is read via pagecache
1377 struct btrfs_device *btrfs_scan_one_device(const char *path, fmode_t flags,
1380 struct btrfs_super_block *disk_super;
1381 bool new_device_added = false;
1382 struct btrfs_device *device = NULL;
1383 struct block_device *bdev;
1384 u64 bytenr, bytenr_orig;
1387 lockdep_assert_held(&uuid_mutex);
1390 * we would like to check all the supers, but that would make
1391 * a btrfs mount succeed after a mkfs from a different FS.
1392 * So, we need to add a special mount option to scan for
1393 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1395 flags |= FMODE_EXCL;
1397 bdev = blkdev_get_by_path(path, flags, holder);
1399 return ERR_CAST(bdev);
1401 bytenr_orig = btrfs_sb_offset(0);
1402 ret = btrfs_sb_log_location_bdev(bdev, 0, READ, &bytenr);
1404 device = ERR_PTR(ret);
1405 goto error_bdev_put;
1408 disk_super = btrfs_read_disk_super(bdev, bytenr, bytenr_orig);
1409 if (IS_ERR(disk_super)) {
1410 device = ERR_CAST(disk_super);
1411 goto error_bdev_put;
1414 device = device_list_add(path, disk_super, &new_device_added);
1415 if (!IS_ERR(device)) {
1416 if (new_device_added)
1417 btrfs_free_stale_devices(path, device);
1420 btrfs_release_disk_super(disk_super);
1423 blkdev_put(bdev, flags);
1429 * Try to find a chunk that intersects [start, start + len] range and when one
1430 * such is found, record the end of it in *start
1432 static bool contains_pending_extent(struct btrfs_device *device, u64 *start,
1435 u64 physical_start, physical_end;
1437 lockdep_assert_held(&device->fs_info->chunk_mutex);
1439 if (!find_first_extent_bit(&device->alloc_state, *start,
1440 &physical_start, &physical_end,
1441 CHUNK_ALLOCATED, NULL)) {
1443 if (in_range(physical_start, *start, len) ||
1444 in_range(*start, physical_start,
1445 physical_end - physical_start)) {
1446 *start = physical_end + 1;
1453 static u64 dev_extent_search_start(struct btrfs_device *device, u64 start)
1455 switch (device->fs_devices->chunk_alloc_policy) {
1456 case BTRFS_CHUNK_ALLOC_REGULAR:
1458 * We don't want to overwrite the superblock on the drive nor
1459 * any area used by the boot loader (grub for example), so we
1460 * make sure to start at an offset of at least 1MB.
1462 return max_t(u64, start, SZ_1M);
1463 case BTRFS_CHUNK_ALLOC_ZONED:
1465 * We don't care about the starting region like regular
1466 * allocator, because we anyway use/reserve the first two zones
1467 * for superblock logging.
1469 return ALIGN(start, device->zone_info->zone_size);
1475 static bool dev_extent_hole_check_zoned(struct btrfs_device *device,
1476 u64 *hole_start, u64 *hole_size,
1479 u64 zone_size = device->zone_info->zone_size;
1482 bool changed = false;
1484 ASSERT(IS_ALIGNED(*hole_start, zone_size));
1486 while (*hole_size > 0) {
1487 pos = btrfs_find_allocatable_zones(device, *hole_start,
1488 *hole_start + *hole_size,
1490 if (pos != *hole_start) {
1491 *hole_size = *hole_start + *hole_size - pos;
1494 if (*hole_size < num_bytes)
1498 ret = btrfs_ensure_empty_zones(device, pos, num_bytes);
1500 /* Range is ensured to be empty */
1504 /* Given hole range was invalid (outside of device) */
1505 if (ret == -ERANGE) {
1506 *hole_start += *hole_size;
1511 *hole_start += zone_size;
1512 *hole_size -= zone_size;
1520 * dev_extent_hole_check - check if specified hole is suitable for allocation
1521 * @device: the device which we have the hole
1522 * @hole_start: starting position of the hole
1523 * @hole_size: the size of the hole
1524 * @num_bytes: the size of the free space that we need
1526 * This function may modify @hole_start and @hole_size to reflect the suitable
1527 * position for allocation. Returns 1 if hole position is updated, 0 otherwise.
1529 static bool dev_extent_hole_check(struct btrfs_device *device, u64 *hole_start,
1530 u64 *hole_size, u64 num_bytes)
1532 bool changed = false;
1533 u64 hole_end = *hole_start + *hole_size;
1537 * Check before we set max_hole_start, otherwise we could end up
1538 * sending back this offset anyway.
1540 if (contains_pending_extent(device, hole_start, *hole_size)) {
1541 if (hole_end >= *hole_start)
1542 *hole_size = hole_end - *hole_start;
1548 switch (device->fs_devices->chunk_alloc_policy) {
1549 case BTRFS_CHUNK_ALLOC_REGULAR:
1550 /* No extra check */
1552 case BTRFS_CHUNK_ALLOC_ZONED:
1553 if (dev_extent_hole_check_zoned(device, hole_start,
1554 hole_size, num_bytes)) {
1557 * The changed hole can contain pending extent.
1558 * Loop again to check that.
1574 * find_free_dev_extent_start - find free space in the specified device
1575 * @device: the device which we search the free space in
1576 * @num_bytes: the size of the free space that we need
1577 * @search_start: the position from which to begin the search
1578 * @start: store the start of the free space.
1579 * @len: the size of the free space. that we find, or the size
1580 * of the max free space if we don't find suitable free space
1582 * this uses a pretty simple search, the expectation is that it is
1583 * called very infrequently and that a given device has a small number
1586 * @start is used to store the start of the free space if we find. But if we
1587 * don't find suitable free space, it will be used to store the start position
1588 * of the max free space.
1590 * @len is used to store the size of the free space that we find.
1591 * But if we don't find suitable free space, it is used to store the size of
1592 * the max free space.
1594 * NOTE: This function will search *commit* root of device tree, and does extra
1595 * check to ensure dev extents are not double allocated.
1596 * This makes the function safe to allocate dev extents but may not report
1597 * correct usable device space, as device extent freed in current transaction
1598 * is not reported as available.
1600 static int find_free_dev_extent_start(struct btrfs_device *device,
1601 u64 num_bytes, u64 search_start, u64 *start,
1604 struct btrfs_fs_info *fs_info = device->fs_info;
1605 struct btrfs_root *root = fs_info->dev_root;
1606 struct btrfs_key key;
1607 struct btrfs_dev_extent *dev_extent;
1608 struct btrfs_path *path;
1613 u64 search_end = device->total_bytes;
1616 struct extent_buffer *l;
1618 search_start = dev_extent_search_start(device, search_start);
1620 WARN_ON(device->zone_info &&
1621 !IS_ALIGNED(num_bytes, device->zone_info->zone_size));
1623 path = btrfs_alloc_path();
1627 max_hole_start = search_start;
1631 if (search_start >= search_end ||
1632 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1637 path->reada = READA_FORWARD;
1638 path->search_commit_root = 1;
1639 path->skip_locking = 1;
1641 key.objectid = device->devid;
1642 key.offset = search_start;
1643 key.type = BTRFS_DEV_EXTENT_KEY;
1645 ret = btrfs_search_backwards(root, &key, path);
1651 slot = path->slots[0];
1652 if (slot >= btrfs_header_nritems(l)) {
1653 ret = btrfs_next_leaf(root, path);
1661 btrfs_item_key_to_cpu(l, &key, slot);
1663 if (key.objectid < device->devid)
1666 if (key.objectid > device->devid)
1669 if (key.type != BTRFS_DEV_EXTENT_KEY)
1672 if (key.offset > search_start) {
1673 hole_size = key.offset - search_start;
1674 dev_extent_hole_check(device, &search_start, &hole_size,
1677 if (hole_size > max_hole_size) {
1678 max_hole_start = search_start;
1679 max_hole_size = hole_size;
1683 * If this free space is greater than which we need,
1684 * it must be the max free space that we have found
1685 * until now, so max_hole_start must point to the start
1686 * of this free space and the length of this free space
1687 * is stored in max_hole_size. Thus, we return
1688 * max_hole_start and max_hole_size and go back to the
1691 if (hole_size >= num_bytes) {
1697 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1698 extent_end = key.offset + btrfs_dev_extent_length(l,
1700 if (extent_end > search_start)
1701 search_start = extent_end;
1708 * At this point, search_start should be the end of
1709 * allocated dev extents, and when shrinking the device,
1710 * search_end may be smaller than search_start.
1712 if (search_end > search_start) {
1713 hole_size = search_end - search_start;
1714 if (dev_extent_hole_check(device, &search_start, &hole_size,
1716 btrfs_release_path(path);
1720 if (hole_size > max_hole_size) {
1721 max_hole_start = search_start;
1722 max_hole_size = hole_size;
1727 if (max_hole_size < num_bytes)
1733 btrfs_free_path(path);
1734 *start = max_hole_start;
1736 *len = max_hole_size;
1740 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
1741 u64 *start, u64 *len)
1743 /* FIXME use last free of some kind */
1744 return find_free_dev_extent_start(device, num_bytes, 0, start, len);
1747 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1748 struct btrfs_device *device,
1749 u64 start, u64 *dev_extent_len)
1751 struct btrfs_fs_info *fs_info = device->fs_info;
1752 struct btrfs_root *root = fs_info->dev_root;
1754 struct btrfs_path *path;
1755 struct btrfs_key key;
1756 struct btrfs_key found_key;
1757 struct extent_buffer *leaf = NULL;
1758 struct btrfs_dev_extent *extent = NULL;
1760 path = btrfs_alloc_path();
1764 key.objectid = device->devid;
1766 key.type = BTRFS_DEV_EXTENT_KEY;
1768 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1770 ret = btrfs_previous_item(root, path, key.objectid,
1771 BTRFS_DEV_EXTENT_KEY);
1774 leaf = path->nodes[0];
1775 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1776 extent = btrfs_item_ptr(leaf, path->slots[0],
1777 struct btrfs_dev_extent);
1778 BUG_ON(found_key.offset > start || found_key.offset +
1779 btrfs_dev_extent_length(leaf, extent) < start);
1781 btrfs_release_path(path);
1783 } else if (ret == 0) {
1784 leaf = path->nodes[0];
1785 extent = btrfs_item_ptr(leaf, path->slots[0],
1786 struct btrfs_dev_extent);
1791 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1793 ret = btrfs_del_item(trans, root, path);
1795 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1797 btrfs_free_path(path);
1801 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1803 struct extent_map_tree *em_tree;
1804 struct extent_map *em;
1808 em_tree = &fs_info->mapping_tree;
1809 read_lock(&em_tree->lock);
1810 n = rb_last(&em_tree->map.rb_root);
1812 em = rb_entry(n, struct extent_map, rb_node);
1813 ret = em->start + em->len;
1815 read_unlock(&em_tree->lock);
1820 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1824 struct btrfs_key key;
1825 struct btrfs_key found_key;
1826 struct btrfs_path *path;
1828 path = btrfs_alloc_path();
1832 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1833 key.type = BTRFS_DEV_ITEM_KEY;
1834 key.offset = (u64)-1;
1836 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1842 btrfs_err(fs_info, "corrupted chunk tree devid -1 matched");
1847 ret = btrfs_previous_item(fs_info->chunk_root, path,
1848 BTRFS_DEV_ITEMS_OBJECTID,
1849 BTRFS_DEV_ITEM_KEY);
1853 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1855 *devid_ret = found_key.offset + 1;
1859 btrfs_free_path(path);
1864 * the device information is stored in the chunk root
1865 * the btrfs_device struct should be fully filled in
1867 static int btrfs_add_dev_item(struct btrfs_trans_handle *trans,
1868 struct btrfs_device *device)
1871 struct btrfs_path *path;
1872 struct btrfs_dev_item *dev_item;
1873 struct extent_buffer *leaf;
1874 struct btrfs_key key;
1877 path = btrfs_alloc_path();
1881 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1882 key.type = BTRFS_DEV_ITEM_KEY;
1883 key.offset = device->devid;
1885 btrfs_reserve_chunk_metadata(trans, true);
1886 ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path,
1887 &key, sizeof(*dev_item));
1888 btrfs_trans_release_chunk_metadata(trans);
1892 leaf = path->nodes[0];
1893 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1895 btrfs_set_device_id(leaf, dev_item, device->devid);
1896 btrfs_set_device_generation(leaf, dev_item, 0);
1897 btrfs_set_device_type(leaf, dev_item, device->type);
1898 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1899 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1900 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1901 btrfs_set_device_total_bytes(leaf, dev_item,
1902 btrfs_device_get_disk_total_bytes(device));
1903 btrfs_set_device_bytes_used(leaf, dev_item,
1904 btrfs_device_get_bytes_used(device));
1905 btrfs_set_device_group(leaf, dev_item, 0);
1906 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1907 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1908 btrfs_set_device_start_offset(leaf, dev_item, 0);
1910 ptr = btrfs_device_uuid(dev_item);
1911 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1912 ptr = btrfs_device_fsid(dev_item);
1913 write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid,
1914 ptr, BTRFS_FSID_SIZE);
1915 btrfs_mark_buffer_dirty(leaf);
1919 btrfs_free_path(path);
1924 * Function to update ctime/mtime for a given device path.
1925 * Mainly used for ctime/mtime based probe like libblkid.
1927 * We don't care about errors here, this is just to be kind to userspace.
1929 static void update_dev_time(const char *device_path)
1932 struct timespec64 now;
1935 ret = kern_path(device_path, LOOKUP_FOLLOW, &path);
1939 now = current_time(d_inode(path.dentry));
1940 inode_update_time(d_inode(path.dentry), &now, S_MTIME | S_CTIME);
1944 static int btrfs_rm_dev_item(struct btrfs_trans_handle *trans,
1945 struct btrfs_device *device)
1947 struct btrfs_root *root = device->fs_info->chunk_root;
1949 struct btrfs_path *path;
1950 struct btrfs_key key;
1952 path = btrfs_alloc_path();
1956 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1957 key.type = BTRFS_DEV_ITEM_KEY;
1958 key.offset = device->devid;
1960 btrfs_reserve_chunk_metadata(trans, false);
1961 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1962 btrfs_trans_release_chunk_metadata(trans);
1969 ret = btrfs_del_item(trans, root, path);
1971 btrfs_free_path(path);
1976 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1977 * filesystem. It's up to the caller to adjust that number regarding eg. device
1980 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1988 seq = read_seqbegin(&fs_info->profiles_lock);
1990 all_avail = fs_info->avail_data_alloc_bits |
1991 fs_info->avail_system_alloc_bits |
1992 fs_info->avail_metadata_alloc_bits;
1993 } while (read_seqretry(&fs_info->profiles_lock, seq));
1995 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1996 if (!(all_avail & btrfs_raid_array[i].bg_flag))
1999 if (num_devices < btrfs_raid_array[i].devs_min)
2000 return btrfs_raid_array[i].mindev_error;
2006 static struct btrfs_device * btrfs_find_next_active_device(
2007 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device)
2009 struct btrfs_device *next_device;
2011 list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
2012 if (next_device != device &&
2013 !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state)
2014 && next_device->bdev)
2022 * Helper function to check if the given device is part of s_bdev / latest_dev
2023 * and replace it with the provided or the next active device, in the context
2024 * where this function called, there should be always be another device (or
2025 * this_dev) which is active.
2027 void __cold btrfs_assign_next_active_device(struct btrfs_device *device,
2028 struct btrfs_device *next_device)
2030 struct btrfs_fs_info *fs_info = device->fs_info;
2033 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
2035 ASSERT(next_device);
2037 if (fs_info->sb->s_bdev &&
2038 (fs_info->sb->s_bdev == device->bdev))
2039 fs_info->sb->s_bdev = next_device->bdev;
2041 if (fs_info->fs_devices->latest_dev->bdev == device->bdev)
2042 fs_info->fs_devices->latest_dev = next_device;
2046 * Return btrfs_fs_devices::num_devices excluding the device that's being
2047 * currently replaced.
2049 static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info)
2051 u64 num_devices = fs_info->fs_devices->num_devices;
2053 down_read(&fs_info->dev_replace.rwsem);
2054 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
2055 ASSERT(num_devices > 1);
2058 up_read(&fs_info->dev_replace.rwsem);
2063 void btrfs_scratch_superblocks(struct btrfs_fs_info *fs_info,
2064 struct block_device *bdev,
2065 const char *device_path)
2067 struct btrfs_super_block *disk_super;
2073 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; copy_num++) {
2077 disk_super = btrfs_read_dev_one_super(bdev, copy_num, false);
2078 if (IS_ERR(disk_super))
2081 if (bdev_is_zoned(bdev)) {
2082 btrfs_reset_sb_log_zones(bdev, copy_num);
2086 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
2088 page = virt_to_page(disk_super);
2089 set_page_dirty(page);
2091 /* write_on_page() unlocks the page */
2092 ret = write_one_page(page);
2095 "error clearing superblock number %d (%d)",
2097 btrfs_release_disk_super(disk_super);
2101 /* Notify udev that device has changed */
2102 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
2104 /* Update ctime/mtime for device path for libblkid */
2105 update_dev_time(device_path);
2108 int btrfs_rm_device(struct btrfs_fs_info *fs_info,
2109 struct btrfs_dev_lookup_args *args,
2110 struct block_device **bdev, fmode_t *mode)
2112 struct btrfs_trans_handle *trans;
2113 struct btrfs_device *device;
2114 struct btrfs_fs_devices *cur_devices;
2115 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2120 * The device list in fs_devices is accessed without locks (neither
2121 * uuid_mutex nor device_list_mutex) as it won't change on a mounted
2122 * filesystem and another device rm cannot run.
2124 num_devices = btrfs_num_devices(fs_info);
2126 ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
2130 device = btrfs_find_device(fs_info->fs_devices, args);
2133 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2139 if (btrfs_pinned_by_swapfile(fs_info, device)) {
2140 btrfs_warn_in_rcu(fs_info,
2141 "cannot remove device %s (devid %llu) due to active swapfile",
2142 rcu_str_deref(device->name), device->devid);
2146 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
2147 return BTRFS_ERROR_DEV_TGT_REPLACE;
2149 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
2150 fs_info->fs_devices->rw_devices == 1)
2151 return BTRFS_ERROR_DEV_ONLY_WRITABLE;
2153 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2154 mutex_lock(&fs_info->chunk_mutex);
2155 list_del_init(&device->dev_alloc_list);
2156 device->fs_devices->rw_devices--;
2157 mutex_unlock(&fs_info->chunk_mutex);
2160 ret = btrfs_shrink_device(device, 0);
2162 btrfs_reada_remove_dev(device);
2166 trans = btrfs_start_transaction(fs_info->chunk_root, 0);
2167 if (IS_ERR(trans)) {
2168 ret = PTR_ERR(trans);
2172 ret = btrfs_rm_dev_item(trans, device);
2174 /* Any error in dev item removal is critical */
2176 "failed to remove device item for devid %llu: %d",
2177 device->devid, ret);
2178 btrfs_abort_transaction(trans, ret);
2179 btrfs_end_transaction(trans);
2183 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2184 btrfs_scrub_cancel_dev(device);
2187 * the device list mutex makes sure that we don't change
2188 * the device list while someone else is writing out all
2189 * the device supers. Whoever is writing all supers, should
2190 * lock the device list mutex before getting the number of
2191 * devices in the super block (super_copy). Conversely,
2192 * whoever updates the number of devices in the super block
2193 * (super_copy) should hold the device list mutex.
2197 * In normal cases the cur_devices == fs_devices. But in case
2198 * of deleting a seed device, the cur_devices should point to
2199 * its own fs_devices listed under the fs_devices->seed.
2201 cur_devices = device->fs_devices;
2202 mutex_lock(&fs_devices->device_list_mutex);
2203 list_del_rcu(&device->dev_list);
2205 cur_devices->num_devices--;
2206 cur_devices->total_devices--;
2207 /* Update total_devices of the parent fs_devices if it's seed */
2208 if (cur_devices != fs_devices)
2209 fs_devices->total_devices--;
2211 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
2212 cur_devices->missing_devices--;
2214 btrfs_assign_next_active_device(device, NULL);
2217 cur_devices->open_devices--;
2218 /* remove sysfs entry */
2219 btrfs_sysfs_remove_device(device);
2222 num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
2223 btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
2224 mutex_unlock(&fs_devices->device_list_mutex);
2227 * At this point, the device is zero sized and detached from the
2228 * devices list. All that's left is to zero out the old supers and
2231 * We cannot call btrfs_close_bdev() here because we're holding the sb
2232 * write lock, and blkdev_put() will pull in the ->open_mutex on the
2233 * block device and it's dependencies. Instead just flush the device
2234 * and let the caller do the final blkdev_put.
2236 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2237 btrfs_scratch_superblocks(fs_info, device->bdev,
2240 sync_blockdev(device->bdev);
2241 invalidate_bdev(device->bdev);
2245 *bdev = device->bdev;
2246 *mode = device->mode;
2248 btrfs_free_device(device);
2250 if (cur_devices->open_devices == 0) {
2251 list_del_init(&cur_devices->seed_list);
2252 close_fs_devices(cur_devices);
2253 free_fs_devices(cur_devices);
2256 ret = btrfs_commit_transaction(trans);
2261 btrfs_reada_undo_remove_dev(device);
2262 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2263 mutex_lock(&fs_info->chunk_mutex);
2264 list_add(&device->dev_alloc_list,
2265 &fs_devices->alloc_list);
2266 device->fs_devices->rw_devices++;
2267 mutex_unlock(&fs_info->chunk_mutex);
2272 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev)
2274 struct btrfs_fs_devices *fs_devices;
2276 lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex);
2279 * in case of fs with no seed, srcdev->fs_devices will point
2280 * to fs_devices of fs_info. However when the dev being replaced is
2281 * a seed dev it will point to the seed's local fs_devices. In short
2282 * srcdev will have its correct fs_devices in both the cases.
2284 fs_devices = srcdev->fs_devices;
2286 list_del_rcu(&srcdev->dev_list);
2287 list_del(&srcdev->dev_alloc_list);
2288 fs_devices->num_devices--;
2289 if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state))
2290 fs_devices->missing_devices--;
2292 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state))
2293 fs_devices->rw_devices--;
2296 fs_devices->open_devices--;
2299 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev)
2301 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2303 mutex_lock(&uuid_mutex);
2305 btrfs_close_bdev(srcdev);
2307 btrfs_free_device(srcdev);
2309 /* if this is no devs we rather delete the fs_devices */
2310 if (!fs_devices->num_devices) {
2312 * On a mounted FS, num_devices can't be zero unless it's a
2313 * seed. In case of a seed device being replaced, the replace
2314 * target added to the sprout FS, so there will be no more
2315 * device left under the seed FS.
2317 ASSERT(fs_devices->seeding);
2319 list_del_init(&fs_devices->seed_list);
2320 close_fs_devices(fs_devices);
2321 free_fs_devices(fs_devices);
2323 mutex_unlock(&uuid_mutex);
2326 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev)
2328 struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices;
2330 mutex_lock(&fs_devices->device_list_mutex);
2332 btrfs_sysfs_remove_device(tgtdev);
2335 fs_devices->open_devices--;
2337 fs_devices->num_devices--;
2339 btrfs_assign_next_active_device(tgtdev, NULL);
2341 list_del_rcu(&tgtdev->dev_list);
2343 mutex_unlock(&fs_devices->device_list_mutex);
2345 btrfs_scratch_superblocks(tgtdev->fs_info, tgtdev->bdev,
2348 btrfs_close_bdev(tgtdev);
2350 btrfs_free_device(tgtdev);
2354 * Populate args from device at path
2356 * @fs_info: the filesystem
2357 * @args: the args to populate
2358 * @path: the path to the device
2360 * This will read the super block of the device at @path and populate @args with
2361 * the devid, fsid, and uuid. This is meant to be used for ioctls that need to
2362 * lookup a device to operate on, but need to do it before we take any locks.
2363 * This properly handles the special case of "missing" that a user may pass in,
2364 * and does some basic sanity checks. The caller must make sure that @path is
2365 * properly NUL terminated before calling in, and must call
2366 * btrfs_put_dev_args_from_path() in order to free up the temporary fsid and
2369 * Return: 0 for success, -errno for failure
2371 int btrfs_get_dev_args_from_path(struct btrfs_fs_info *fs_info,
2372 struct btrfs_dev_lookup_args *args,
2375 struct btrfs_super_block *disk_super;
2376 struct block_device *bdev;
2379 if (!path || !path[0])
2381 if (!strcmp(path, "missing")) {
2382 args->missing = true;
2386 args->uuid = kzalloc(BTRFS_UUID_SIZE, GFP_KERNEL);
2387 args->fsid = kzalloc(BTRFS_FSID_SIZE, GFP_KERNEL);
2388 if (!args->uuid || !args->fsid) {
2389 btrfs_put_dev_args_from_path(args);
2393 ret = btrfs_get_bdev_and_sb(path, FMODE_READ, fs_info->bdev_holder, 0,
2394 &bdev, &disk_super);
2396 btrfs_put_dev_args_from_path(args);
2400 args->devid = btrfs_stack_device_id(&disk_super->dev_item);
2401 memcpy(args->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE);
2402 if (btrfs_fs_incompat(fs_info, METADATA_UUID))
2403 memcpy(args->fsid, disk_super->metadata_uuid, BTRFS_FSID_SIZE);
2405 memcpy(args->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
2406 btrfs_release_disk_super(disk_super);
2407 blkdev_put(bdev, FMODE_READ);
2412 * Only use this jointly with btrfs_get_dev_args_from_path() because we will
2413 * allocate our ->uuid and ->fsid pointers, everybody else uses local variables
2414 * that don't need to be freed.
2416 void btrfs_put_dev_args_from_path(struct btrfs_dev_lookup_args *args)
2424 struct btrfs_device *btrfs_find_device_by_devspec(
2425 struct btrfs_fs_info *fs_info, u64 devid,
2426 const char *device_path)
2428 BTRFS_DEV_LOOKUP_ARGS(args);
2429 struct btrfs_device *device;
2434 device = btrfs_find_device(fs_info->fs_devices, &args);
2436 return ERR_PTR(-ENOENT);
2440 ret = btrfs_get_dev_args_from_path(fs_info, &args, device_path);
2442 return ERR_PTR(ret);
2443 device = btrfs_find_device(fs_info->fs_devices, &args);
2444 btrfs_put_dev_args_from_path(&args);
2446 return ERR_PTR(-ENOENT);
2451 * does all the dirty work required for changing file system's UUID.
2453 static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info)
2455 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2456 struct btrfs_fs_devices *old_devices;
2457 struct btrfs_fs_devices *seed_devices;
2458 struct btrfs_super_block *disk_super = fs_info->super_copy;
2459 struct btrfs_device *device;
2462 lockdep_assert_held(&uuid_mutex);
2463 if (!fs_devices->seeding)
2467 * Private copy of the seed devices, anchored at
2468 * fs_info->fs_devices->seed_list
2470 seed_devices = alloc_fs_devices(NULL, NULL);
2471 if (IS_ERR(seed_devices))
2472 return PTR_ERR(seed_devices);
2475 * It's necessary to retain a copy of the original seed fs_devices in
2476 * fs_uuids so that filesystems which have been seeded can successfully
2477 * reference the seed device from open_seed_devices. This also supports
2480 old_devices = clone_fs_devices(fs_devices);
2481 if (IS_ERR(old_devices)) {
2482 kfree(seed_devices);
2483 return PTR_ERR(old_devices);
2486 list_add(&old_devices->fs_list, &fs_uuids);
2488 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2489 seed_devices->opened = 1;
2490 INIT_LIST_HEAD(&seed_devices->devices);
2491 INIT_LIST_HEAD(&seed_devices->alloc_list);
2492 mutex_init(&seed_devices->device_list_mutex);
2494 mutex_lock(&fs_devices->device_list_mutex);
2495 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2497 list_for_each_entry(device, &seed_devices->devices, dev_list)
2498 device->fs_devices = seed_devices;
2500 fs_devices->seeding = false;
2501 fs_devices->num_devices = 0;
2502 fs_devices->open_devices = 0;
2503 fs_devices->missing_devices = 0;
2504 fs_devices->rotating = false;
2505 list_add(&seed_devices->seed_list, &fs_devices->seed_list);
2507 generate_random_uuid(fs_devices->fsid);
2508 memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE);
2509 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2510 mutex_unlock(&fs_devices->device_list_mutex);
2512 super_flags = btrfs_super_flags(disk_super) &
2513 ~BTRFS_SUPER_FLAG_SEEDING;
2514 btrfs_set_super_flags(disk_super, super_flags);
2520 * Store the expected generation for seed devices in device items.
2522 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans)
2524 BTRFS_DEV_LOOKUP_ARGS(args);
2525 struct btrfs_fs_info *fs_info = trans->fs_info;
2526 struct btrfs_root *root = fs_info->chunk_root;
2527 struct btrfs_path *path;
2528 struct extent_buffer *leaf;
2529 struct btrfs_dev_item *dev_item;
2530 struct btrfs_device *device;
2531 struct btrfs_key key;
2532 u8 fs_uuid[BTRFS_FSID_SIZE];
2533 u8 dev_uuid[BTRFS_UUID_SIZE];
2536 path = btrfs_alloc_path();
2540 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2542 key.type = BTRFS_DEV_ITEM_KEY;
2545 btrfs_reserve_chunk_metadata(trans, false);
2546 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2547 btrfs_trans_release_chunk_metadata(trans);
2551 leaf = path->nodes[0];
2553 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2554 ret = btrfs_next_leaf(root, path);
2559 leaf = path->nodes[0];
2560 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2561 btrfs_release_path(path);
2565 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2566 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2567 key.type != BTRFS_DEV_ITEM_KEY)
2570 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2571 struct btrfs_dev_item);
2572 args.devid = btrfs_device_id(leaf, dev_item);
2573 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2575 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2577 args.uuid = dev_uuid;
2578 args.fsid = fs_uuid;
2579 device = btrfs_find_device(fs_info->fs_devices, &args);
2580 BUG_ON(!device); /* Logic error */
2582 if (device->fs_devices->seeding) {
2583 btrfs_set_device_generation(leaf, dev_item,
2584 device->generation);
2585 btrfs_mark_buffer_dirty(leaf);
2593 btrfs_free_path(path);
2597 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2599 struct btrfs_root *root = fs_info->dev_root;
2600 struct request_queue *q;
2601 struct btrfs_trans_handle *trans;
2602 struct btrfs_device *device;
2603 struct block_device *bdev;
2604 struct super_block *sb = fs_info->sb;
2605 struct rcu_string *name;
2606 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2607 u64 orig_super_total_bytes;
2608 u64 orig_super_num_devices;
2609 int seeding_dev = 0;
2611 bool locked = false;
2613 if (sb_rdonly(sb) && !fs_devices->seeding)
2616 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2617 fs_info->bdev_holder);
2619 return PTR_ERR(bdev);
2621 if (!btrfs_check_device_zone_type(fs_info, bdev)) {
2626 if (fs_devices->seeding) {
2628 down_write(&sb->s_umount);
2629 mutex_lock(&uuid_mutex);
2633 sync_blockdev(bdev);
2636 list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
2637 if (device->bdev == bdev) {
2645 device = btrfs_alloc_device(fs_info, NULL, NULL);
2646 if (IS_ERR(device)) {
2647 /* we can safely leave the fs_devices entry around */
2648 ret = PTR_ERR(device);
2652 name = rcu_string_strdup(device_path, GFP_KERNEL);
2655 goto error_free_device;
2657 rcu_assign_pointer(device->name, name);
2659 device->fs_info = fs_info;
2660 device->bdev = bdev;
2662 ret = btrfs_get_dev_zone_info(device, false);
2664 goto error_free_device;
2666 trans = btrfs_start_transaction(root, 0);
2667 if (IS_ERR(trans)) {
2668 ret = PTR_ERR(trans);
2669 goto error_free_zone;
2672 q = bdev_get_queue(bdev);
2673 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
2674 device->generation = trans->transid;
2675 device->io_width = fs_info->sectorsize;
2676 device->io_align = fs_info->sectorsize;
2677 device->sector_size = fs_info->sectorsize;
2678 device->total_bytes = round_down(i_size_read(bdev->bd_inode),
2679 fs_info->sectorsize);
2680 device->disk_total_bytes = device->total_bytes;
2681 device->commit_total_bytes = device->total_bytes;
2682 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2683 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
2684 device->mode = FMODE_EXCL;
2685 device->dev_stats_valid = 1;
2686 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2689 btrfs_clear_sb_rdonly(sb);
2690 ret = btrfs_prepare_sprout(fs_info);
2692 btrfs_abort_transaction(trans, ret);
2695 btrfs_assign_next_active_device(fs_info->fs_devices->latest_dev,
2699 device->fs_devices = fs_devices;
2701 mutex_lock(&fs_devices->device_list_mutex);
2702 mutex_lock(&fs_info->chunk_mutex);
2703 list_add_rcu(&device->dev_list, &fs_devices->devices);
2704 list_add(&device->dev_alloc_list, &fs_devices->alloc_list);
2705 fs_devices->num_devices++;
2706 fs_devices->open_devices++;
2707 fs_devices->rw_devices++;
2708 fs_devices->total_devices++;
2709 fs_devices->total_rw_bytes += device->total_bytes;
2711 atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2713 if (!blk_queue_nonrot(q))
2714 fs_devices->rotating = true;
2716 orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
2717 btrfs_set_super_total_bytes(fs_info->super_copy,
2718 round_down(orig_super_total_bytes + device->total_bytes,
2719 fs_info->sectorsize));
2721 orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy);
2722 btrfs_set_super_num_devices(fs_info->super_copy,
2723 orig_super_num_devices + 1);
2726 * we've got more storage, clear any full flags on the space
2729 btrfs_clear_space_info_full(fs_info);
2731 mutex_unlock(&fs_info->chunk_mutex);
2733 /* Add sysfs device entry */
2734 btrfs_sysfs_add_device(device);
2736 mutex_unlock(&fs_devices->device_list_mutex);
2739 mutex_lock(&fs_info->chunk_mutex);
2740 ret = init_first_rw_device(trans);
2741 mutex_unlock(&fs_info->chunk_mutex);
2743 btrfs_abort_transaction(trans, ret);
2748 ret = btrfs_add_dev_item(trans, device);
2750 btrfs_abort_transaction(trans, ret);
2755 ret = btrfs_finish_sprout(trans);
2757 btrfs_abort_transaction(trans, ret);
2762 * fs_devices now represents the newly sprouted filesystem and
2763 * its fsid has been changed by btrfs_prepare_sprout
2765 btrfs_sysfs_update_sprout_fsid(fs_devices);
2768 ret = btrfs_commit_transaction(trans);
2771 mutex_unlock(&uuid_mutex);
2772 up_write(&sb->s_umount);
2775 if (ret) /* transaction commit */
2778 ret = btrfs_relocate_sys_chunks(fs_info);
2780 btrfs_handle_fs_error(fs_info, ret,
2781 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2782 trans = btrfs_attach_transaction(root);
2783 if (IS_ERR(trans)) {
2784 if (PTR_ERR(trans) == -ENOENT)
2786 ret = PTR_ERR(trans);
2790 ret = btrfs_commit_transaction(trans);
2794 * Now that we have written a new super block to this device, check all
2795 * other fs_devices list if device_path alienates any other scanned
2797 * We can ignore the return value as it typically returns -EINVAL and
2798 * only succeeds if the device was an alien.
2800 btrfs_forget_devices(device_path);
2802 /* Update ctime/mtime for blkid or udev */
2803 update_dev_time(device_path);
2808 btrfs_sysfs_remove_device(device);
2809 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2810 mutex_lock(&fs_info->chunk_mutex);
2811 list_del_rcu(&device->dev_list);
2812 list_del(&device->dev_alloc_list);
2813 fs_info->fs_devices->num_devices--;
2814 fs_info->fs_devices->open_devices--;
2815 fs_info->fs_devices->rw_devices--;
2816 fs_info->fs_devices->total_devices--;
2817 fs_info->fs_devices->total_rw_bytes -= device->total_bytes;
2818 atomic64_sub(device->total_bytes, &fs_info->free_chunk_space);
2819 btrfs_set_super_total_bytes(fs_info->super_copy,
2820 orig_super_total_bytes);
2821 btrfs_set_super_num_devices(fs_info->super_copy,
2822 orig_super_num_devices);
2823 mutex_unlock(&fs_info->chunk_mutex);
2824 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2827 btrfs_set_sb_rdonly(sb);
2829 btrfs_end_transaction(trans);
2831 btrfs_destroy_dev_zone_info(device);
2833 btrfs_free_device(device);
2835 blkdev_put(bdev, FMODE_EXCL);
2837 mutex_unlock(&uuid_mutex);
2838 up_write(&sb->s_umount);
2843 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2844 struct btrfs_device *device)
2847 struct btrfs_path *path;
2848 struct btrfs_root *root = device->fs_info->chunk_root;
2849 struct btrfs_dev_item *dev_item;
2850 struct extent_buffer *leaf;
2851 struct btrfs_key key;
2853 path = btrfs_alloc_path();
2857 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2858 key.type = BTRFS_DEV_ITEM_KEY;
2859 key.offset = device->devid;
2861 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2870 leaf = path->nodes[0];
2871 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2873 btrfs_set_device_id(leaf, dev_item, device->devid);
2874 btrfs_set_device_type(leaf, dev_item, device->type);
2875 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2876 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2877 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2878 btrfs_set_device_total_bytes(leaf, dev_item,
2879 btrfs_device_get_disk_total_bytes(device));
2880 btrfs_set_device_bytes_used(leaf, dev_item,
2881 btrfs_device_get_bytes_used(device));
2882 btrfs_mark_buffer_dirty(leaf);
2885 btrfs_free_path(path);
2889 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2890 struct btrfs_device *device, u64 new_size)
2892 struct btrfs_fs_info *fs_info = device->fs_info;
2893 struct btrfs_super_block *super_copy = fs_info->super_copy;
2898 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2901 new_size = round_down(new_size, fs_info->sectorsize);
2903 mutex_lock(&fs_info->chunk_mutex);
2904 old_total = btrfs_super_total_bytes(super_copy);
2905 diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2907 if (new_size <= device->total_bytes ||
2908 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2909 mutex_unlock(&fs_info->chunk_mutex);
2913 btrfs_set_super_total_bytes(super_copy,
2914 round_down(old_total + diff, fs_info->sectorsize));
2915 device->fs_devices->total_rw_bytes += diff;
2917 btrfs_device_set_total_bytes(device, new_size);
2918 btrfs_device_set_disk_total_bytes(device, new_size);
2919 btrfs_clear_space_info_full(device->fs_info);
2920 if (list_empty(&device->post_commit_list))
2921 list_add_tail(&device->post_commit_list,
2922 &trans->transaction->dev_update_list);
2923 mutex_unlock(&fs_info->chunk_mutex);
2925 btrfs_reserve_chunk_metadata(trans, false);
2926 ret = btrfs_update_device(trans, device);
2927 btrfs_trans_release_chunk_metadata(trans);
2932 static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
2934 struct btrfs_fs_info *fs_info = trans->fs_info;
2935 struct btrfs_root *root = fs_info->chunk_root;
2937 struct btrfs_path *path;
2938 struct btrfs_key key;
2940 path = btrfs_alloc_path();
2944 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2945 key.offset = chunk_offset;
2946 key.type = BTRFS_CHUNK_ITEM_KEY;
2948 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2951 else if (ret > 0) { /* Logic error or corruption */
2952 btrfs_handle_fs_error(fs_info, -ENOENT,
2953 "Failed lookup while freeing chunk.");
2958 ret = btrfs_del_item(trans, root, path);
2960 btrfs_handle_fs_error(fs_info, ret,
2961 "Failed to delete chunk item.");
2963 btrfs_free_path(path);
2967 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2969 struct btrfs_super_block *super_copy = fs_info->super_copy;
2970 struct btrfs_disk_key *disk_key;
2971 struct btrfs_chunk *chunk;
2978 struct btrfs_key key;
2980 lockdep_assert_held(&fs_info->chunk_mutex);
2981 array_size = btrfs_super_sys_array_size(super_copy);
2983 ptr = super_copy->sys_chunk_array;
2986 while (cur < array_size) {
2987 disk_key = (struct btrfs_disk_key *)ptr;
2988 btrfs_disk_key_to_cpu(&key, disk_key);
2990 len = sizeof(*disk_key);
2992 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2993 chunk = (struct btrfs_chunk *)(ptr + len);
2994 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2995 len += btrfs_chunk_item_size(num_stripes);
3000 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID &&
3001 key.offset == chunk_offset) {
3002 memmove(ptr, ptr + len, array_size - (cur + len));
3004 btrfs_set_super_sys_array_size(super_copy, array_size);
3014 * btrfs_get_chunk_map() - Find the mapping containing the given logical extent.
3015 * @logical: Logical block offset in bytes.
3016 * @length: Length of extent in bytes.
3018 * Return: Chunk mapping or ERR_PTR.
3020 struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info,
3021 u64 logical, u64 length)
3023 struct extent_map_tree *em_tree;
3024 struct extent_map *em;
3026 em_tree = &fs_info->mapping_tree;
3027 read_lock(&em_tree->lock);
3028 em = lookup_extent_mapping(em_tree, logical, length);
3029 read_unlock(&em_tree->lock);
3032 btrfs_crit(fs_info, "unable to find logical %llu length %llu",
3034 return ERR_PTR(-EINVAL);
3037 if (em->start > logical || em->start + em->len < logical) {
3039 "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
3040 logical, length, em->start, em->start + em->len);
3041 free_extent_map(em);
3042 return ERR_PTR(-EINVAL);
3045 /* callers are responsible for dropping em's ref. */
3049 static int remove_chunk_item(struct btrfs_trans_handle *trans,
3050 struct map_lookup *map, u64 chunk_offset)
3055 * Removing chunk items and updating the device items in the chunks btree
3056 * requires holding the chunk_mutex.
3057 * See the comment at btrfs_chunk_alloc() for the details.
3059 lockdep_assert_held(&trans->fs_info->chunk_mutex);
3061 for (i = 0; i < map->num_stripes; i++) {
3064 ret = btrfs_update_device(trans, map->stripes[i].dev);
3069 return btrfs_free_chunk(trans, chunk_offset);
3072 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
3074 struct btrfs_fs_info *fs_info = trans->fs_info;
3075 struct extent_map *em;
3076 struct map_lookup *map;
3077 u64 dev_extent_len = 0;
3079 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
3081 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
3084 * This is a logic error, but we don't want to just rely on the
3085 * user having built with ASSERT enabled, so if ASSERT doesn't
3086 * do anything we still error out.
3091 map = em->map_lookup;
3094 * First delete the device extent items from the devices btree.
3095 * We take the device_list_mutex to avoid racing with the finishing phase
3096 * of a device replace operation. See the comment below before acquiring
3097 * fs_info->chunk_mutex. Note that here we do not acquire the chunk_mutex
3098 * because that can result in a deadlock when deleting the device extent
3099 * items from the devices btree - COWing an extent buffer from the btree
3100 * may result in allocating a new metadata chunk, which would attempt to
3101 * lock again fs_info->chunk_mutex.
3103 mutex_lock(&fs_devices->device_list_mutex);
3104 for (i = 0; i < map->num_stripes; i++) {
3105 struct btrfs_device *device = map->stripes[i].dev;
3106 ret = btrfs_free_dev_extent(trans, device,
3107 map->stripes[i].physical,
3110 mutex_unlock(&fs_devices->device_list_mutex);
3111 btrfs_abort_transaction(trans, ret);
3115 if (device->bytes_used > 0) {
3116 mutex_lock(&fs_info->chunk_mutex);
3117 btrfs_device_set_bytes_used(device,
3118 device->bytes_used - dev_extent_len);
3119 atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
3120 btrfs_clear_space_info_full(fs_info);
3121 mutex_unlock(&fs_info->chunk_mutex);
3124 mutex_unlock(&fs_devices->device_list_mutex);
3127 * We acquire fs_info->chunk_mutex for 2 reasons:
3129 * 1) Just like with the first phase of the chunk allocation, we must
3130 * reserve system space, do all chunk btree updates and deletions, and
3131 * update the system chunk array in the superblock while holding this
3132 * mutex. This is for similar reasons as explained on the comment at
3133 * the top of btrfs_chunk_alloc();
3135 * 2) Prevent races with the final phase of a device replace operation
3136 * that replaces the device object associated with the map's stripes,
3137 * because the device object's id can change at any time during that
3138 * final phase of the device replace operation
3139 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the
3140 * replaced device and then see it with an ID of
3141 * BTRFS_DEV_REPLACE_DEVID, which would cause a failure when updating
3142 * the device item, which does not exists on the chunk btree.
3143 * The finishing phase of device replace acquires both the
3144 * device_list_mutex and the chunk_mutex, in that order, so we are
3145 * safe by just acquiring the chunk_mutex.
3147 trans->removing_chunk = true;
3148 mutex_lock(&fs_info->chunk_mutex);
3150 check_system_chunk(trans, map->type);
3152 ret = remove_chunk_item(trans, map, chunk_offset);
3154 * Normally we should not get -ENOSPC since we reserved space before
3155 * through the call to check_system_chunk().
3157 * Despite our system space_info having enough free space, we may not
3158 * be able to allocate extents from its block groups, because all have
3159 * an incompatible profile, which will force us to allocate a new system
3160 * block group with the right profile, or right after we called
3161 * check_system_space() above, a scrub turned the only system block group
3162 * with enough free space into RO mode.
3163 * This is explained with more detail at do_chunk_alloc().
3165 * So if we get -ENOSPC, allocate a new system chunk and retry once.
3167 if (ret == -ENOSPC) {
3168 const u64 sys_flags = btrfs_system_alloc_profile(fs_info);
3169 struct btrfs_block_group *sys_bg;
3171 sys_bg = btrfs_create_chunk(trans, sys_flags);
3172 if (IS_ERR(sys_bg)) {
3173 ret = PTR_ERR(sys_bg);
3174 btrfs_abort_transaction(trans, ret);
3178 ret = btrfs_chunk_alloc_add_chunk_item(trans, sys_bg);
3180 btrfs_abort_transaction(trans, ret);
3184 ret = remove_chunk_item(trans, map, chunk_offset);
3186 btrfs_abort_transaction(trans, ret);
3190 btrfs_abort_transaction(trans, ret);
3194 trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
3196 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
3197 ret = btrfs_del_sys_chunk(fs_info, chunk_offset);
3199 btrfs_abort_transaction(trans, ret);
3204 mutex_unlock(&fs_info->chunk_mutex);
3205 trans->removing_chunk = false;
3208 * We are done with chunk btree updates and deletions, so release the
3209 * system space we previously reserved (with check_system_chunk()).
3211 btrfs_trans_release_chunk_metadata(trans);
3213 ret = btrfs_remove_block_group(trans, chunk_offset, em);
3215 btrfs_abort_transaction(trans, ret);
3220 if (trans->removing_chunk) {
3221 mutex_unlock(&fs_info->chunk_mutex);
3222 trans->removing_chunk = false;
3225 free_extent_map(em);
3229 int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
3231 struct btrfs_root *root = fs_info->chunk_root;
3232 struct btrfs_trans_handle *trans;
3233 struct btrfs_block_group *block_group;
3238 * Prevent races with automatic removal of unused block groups.
3239 * After we relocate and before we remove the chunk with offset
3240 * chunk_offset, automatic removal of the block group can kick in,
3241 * resulting in a failure when calling btrfs_remove_chunk() below.
3243 * Make sure to acquire this mutex before doing a tree search (dev
3244 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
3245 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
3246 * we release the path used to search the chunk/dev tree and before
3247 * the current task acquires this mutex and calls us.
3249 lockdep_assert_held(&fs_info->reclaim_bgs_lock);
3251 /* step one, relocate all the extents inside this chunk */
3252 btrfs_scrub_pause(fs_info);
3253 ret = btrfs_relocate_block_group(fs_info, chunk_offset);
3254 btrfs_scrub_continue(fs_info);
3258 block_group = btrfs_lookup_block_group(fs_info, chunk_offset);
3261 btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group);
3262 length = block_group->length;
3263 btrfs_put_block_group(block_group);
3266 * On a zoned file system, discard the whole block group, this will
3267 * trigger a REQ_OP_ZONE_RESET operation on the device zone. If
3268 * resetting the zone fails, don't treat it as a fatal problem from the
3269 * filesystem's point of view.
3271 if (btrfs_is_zoned(fs_info)) {
3272 ret = btrfs_discard_extent(fs_info, chunk_offset, length, NULL);
3275 "failed to reset zone %llu after relocation",
3279 trans = btrfs_start_trans_remove_block_group(root->fs_info,
3281 if (IS_ERR(trans)) {
3282 ret = PTR_ERR(trans);
3283 btrfs_handle_fs_error(root->fs_info, ret, NULL);
3288 * step two, delete the device extents and the
3289 * chunk tree entries
3291 ret = btrfs_remove_chunk(trans, chunk_offset);
3292 btrfs_end_transaction(trans);
3296 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
3298 struct btrfs_root *chunk_root = fs_info->chunk_root;
3299 struct btrfs_path *path;
3300 struct extent_buffer *leaf;
3301 struct btrfs_chunk *chunk;
3302 struct btrfs_key key;
3303 struct btrfs_key found_key;
3305 bool retried = false;
3309 path = btrfs_alloc_path();
3314 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3315 key.offset = (u64)-1;
3316 key.type = BTRFS_CHUNK_ITEM_KEY;
3319 mutex_lock(&fs_info->reclaim_bgs_lock);
3320 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3322 mutex_unlock(&fs_info->reclaim_bgs_lock);
3325 BUG_ON(ret == 0); /* Corruption */
3327 ret = btrfs_previous_item(chunk_root, path, key.objectid,
3330 mutex_unlock(&fs_info->reclaim_bgs_lock);
3336 leaf = path->nodes[0];
3337 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3339 chunk = btrfs_item_ptr(leaf, path->slots[0],
3340 struct btrfs_chunk);
3341 chunk_type = btrfs_chunk_type(leaf, chunk);
3342 btrfs_release_path(path);
3344 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3345 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3351 mutex_unlock(&fs_info->reclaim_bgs_lock);
3353 if (found_key.offset == 0)
3355 key.offset = found_key.offset - 1;
3358 if (failed && !retried) {
3362 } else if (WARN_ON(failed && retried)) {
3366 btrfs_free_path(path);
3371 * return 1 : allocate a data chunk successfully,
3372 * return <0: errors during allocating a data chunk,
3373 * return 0 : no need to allocate a data chunk.
3375 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info,
3378 struct btrfs_block_group *cache;
3382 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3384 chunk_type = cache->flags;
3385 btrfs_put_block_group(cache);
3387 if (!(chunk_type & BTRFS_BLOCK_GROUP_DATA))
3390 spin_lock(&fs_info->data_sinfo->lock);
3391 bytes_used = fs_info->data_sinfo->bytes_used;
3392 spin_unlock(&fs_info->data_sinfo->lock);
3395 struct btrfs_trans_handle *trans;
3398 trans = btrfs_join_transaction(fs_info->tree_root);
3400 return PTR_ERR(trans);
3402 ret = btrfs_force_chunk_alloc(trans, BTRFS_BLOCK_GROUP_DATA);
3403 btrfs_end_transaction(trans);
3412 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3413 struct btrfs_balance_control *bctl)
3415 struct btrfs_root *root = fs_info->tree_root;
3416 struct btrfs_trans_handle *trans;
3417 struct btrfs_balance_item *item;
3418 struct btrfs_disk_balance_args disk_bargs;
3419 struct btrfs_path *path;
3420 struct extent_buffer *leaf;
3421 struct btrfs_key key;
3424 path = btrfs_alloc_path();
3428 trans = btrfs_start_transaction(root, 0);
3429 if (IS_ERR(trans)) {
3430 btrfs_free_path(path);
3431 return PTR_ERR(trans);
3434 key.objectid = BTRFS_BALANCE_OBJECTID;
3435 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3438 ret = btrfs_insert_empty_item(trans, root, path, &key,
3443 leaf = path->nodes[0];
3444 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3446 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3448 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3449 btrfs_set_balance_data(leaf, item, &disk_bargs);
3450 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3451 btrfs_set_balance_meta(leaf, item, &disk_bargs);
3452 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3453 btrfs_set_balance_sys(leaf, item, &disk_bargs);
3455 btrfs_set_balance_flags(leaf, item, bctl->flags);
3457 btrfs_mark_buffer_dirty(leaf);
3459 btrfs_free_path(path);
3460 err = btrfs_commit_transaction(trans);
3466 static int del_balance_item(struct btrfs_fs_info *fs_info)
3468 struct btrfs_root *root = fs_info->tree_root;
3469 struct btrfs_trans_handle *trans;
3470 struct btrfs_path *path;
3471 struct btrfs_key key;
3474 path = btrfs_alloc_path();
3478 trans = btrfs_start_transaction_fallback_global_rsv(root, 0);
3479 if (IS_ERR(trans)) {
3480 btrfs_free_path(path);
3481 return PTR_ERR(trans);
3484 key.objectid = BTRFS_BALANCE_OBJECTID;
3485 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3488 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3496 ret = btrfs_del_item(trans, root, path);
3498 btrfs_free_path(path);
3499 err = btrfs_commit_transaction(trans);
3506 * This is a heuristic used to reduce the number of chunks balanced on
3507 * resume after balance was interrupted.
3509 static void update_balance_args(struct btrfs_balance_control *bctl)
3512 * Turn on soft mode for chunk types that were being converted.
3514 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3515 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3516 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3517 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3518 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3519 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3522 * Turn on usage filter if is not already used. The idea is
3523 * that chunks that we have already balanced should be
3524 * reasonably full. Don't do it for chunks that are being
3525 * converted - that will keep us from relocating unconverted
3526 * (albeit full) chunks.
3528 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3529 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3530 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3531 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3532 bctl->data.usage = 90;
3534 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3535 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3536 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3537 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3538 bctl->sys.usage = 90;
3540 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3541 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3542 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3543 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3544 bctl->meta.usage = 90;
3549 * Clear the balance status in fs_info and delete the balance item from disk.
3551 static void reset_balance_state(struct btrfs_fs_info *fs_info)
3553 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3556 BUG_ON(!fs_info->balance_ctl);
3558 spin_lock(&fs_info->balance_lock);
3559 fs_info->balance_ctl = NULL;
3560 spin_unlock(&fs_info->balance_lock);
3563 ret = del_balance_item(fs_info);
3565 btrfs_handle_fs_error(fs_info, ret, NULL);
3569 * Balance filters. Return 1 if chunk should be filtered out
3570 * (should not be balanced).
3572 static int chunk_profiles_filter(u64 chunk_type,
3573 struct btrfs_balance_args *bargs)
3575 chunk_type = chunk_to_extended(chunk_type) &
3576 BTRFS_EXTENDED_PROFILE_MASK;
3578 if (bargs->profiles & chunk_type)
3584 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3585 struct btrfs_balance_args *bargs)
3587 struct btrfs_block_group *cache;
3589 u64 user_thresh_min;
3590 u64 user_thresh_max;
3593 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3594 chunk_used = cache->used;
3596 if (bargs->usage_min == 0)
3597 user_thresh_min = 0;
3599 user_thresh_min = div_factor_fine(cache->length,
3602 if (bargs->usage_max == 0)
3603 user_thresh_max = 1;
3604 else if (bargs->usage_max > 100)
3605 user_thresh_max = cache->length;
3607 user_thresh_max = div_factor_fine(cache->length,
3610 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3613 btrfs_put_block_group(cache);
3617 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3618 u64 chunk_offset, struct btrfs_balance_args *bargs)
3620 struct btrfs_block_group *cache;
3621 u64 chunk_used, user_thresh;
3624 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3625 chunk_used = cache->used;
3627 if (bargs->usage_min == 0)
3629 else if (bargs->usage > 100)
3630 user_thresh = cache->length;
3632 user_thresh = div_factor_fine(cache->length, bargs->usage);
3634 if (chunk_used < user_thresh)
3637 btrfs_put_block_group(cache);
3641 static int chunk_devid_filter(struct extent_buffer *leaf,
3642 struct btrfs_chunk *chunk,
3643 struct btrfs_balance_args *bargs)
3645 struct btrfs_stripe *stripe;
3646 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3649 for (i = 0; i < num_stripes; i++) {
3650 stripe = btrfs_stripe_nr(chunk, i);
3651 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3658 static u64 calc_data_stripes(u64 type, int num_stripes)
3660 const int index = btrfs_bg_flags_to_raid_index(type);
3661 const int ncopies = btrfs_raid_array[index].ncopies;
3662 const int nparity = btrfs_raid_array[index].nparity;
3664 return (num_stripes - nparity) / ncopies;
3667 /* [pstart, pend) */
3668 static int chunk_drange_filter(struct extent_buffer *leaf,
3669 struct btrfs_chunk *chunk,
3670 struct btrfs_balance_args *bargs)
3672 struct btrfs_stripe *stripe;
3673 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3680 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3683 type = btrfs_chunk_type(leaf, chunk);
3684 factor = calc_data_stripes(type, num_stripes);
3686 for (i = 0; i < num_stripes; i++) {
3687 stripe = btrfs_stripe_nr(chunk, i);
3688 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3691 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3692 stripe_length = btrfs_chunk_length(leaf, chunk);
3693 stripe_length = div_u64(stripe_length, factor);
3695 if (stripe_offset < bargs->pend &&
3696 stripe_offset + stripe_length > bargs->pstart)
3703 /* [vstart, vend) */
3704 static int chunk_vrange_filter(struct extent_buffer *leaf,
3705 struct btrfs_chunk *chunk,
3707 struct btrfs_balance_args *bargs)
3709 if (chunk_offset < bargs->vend &&
3710 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3711 /* at least part of the chunk is inside this vrange */
3717 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3718 struct btrfs_chunk *chunk,
3719 struct btrfs_balance_args *bargs)
3721 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3723 if (bargs->stripes_min <= num_stripes
3724 && num_stripes <= bargs->stripes_max)
3730 static int chunk_soft_convert_filter(u64 chunk_type,
3731 struct btrfs_balance_args *bargs)
3733 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3736 chunk_type = chunk_to_extended(chunk_type) &
3737 BTRFS_EXTENDED_PROFILE_MASK;
3739 if (bargs->target == chunk_type)
3745 static int should_balance_chunk(struct extent_buffer *leaf,
3746 struct btrfs_chunk *chunk, u64 chunk_offset)
3748 struct btrfs_fs_info *fs_info = leaf->fs_info;
3749 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3750 struct btrfs_balance_args *bargs = NULL;
3751 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3754 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3755 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3759 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3760 bargs = &bctl->data;
3761 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3763 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3764 bargs = &bctl->meta;
3766 /* profiles filter */
3767 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3768 chunk_profiles_filter(chunk_type, bargs)) {
3773 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3774 chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3776 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3777 chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3782 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3783 chunk_devid_filter(leaf, chunk, bargs)) {
3787 /* drange filter, makes sense only with devid filter */
3788 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3789 chunk_drange_filter(leaf, chunk, bargs)) {
3794 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3795 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3799 /* stripes filter */
3800 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3801 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3805 /* soft profile changing mode */
3806 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3807 chunk_soft_convert_filter(chunk_type, bargs)) {
3812 * limited by count, must be the last filter
3814 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3815 if (bargs->limit == 0)
3819 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3821 * Same logic as the 'limit' filter; the minimum cannot be
3822 * determined here because we do not have the global information
3823 * about the count of all chunks that satisfy the filters.
3825 if (bargs->limit_max == 0)
3834 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3836 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3837 struct btrfs_root *chunk_root = fs_info->chunk_root;
3839 struct btrfs_chunk *chunk;
3840 struct btrfs_path *path = NULL;
3841 struct btrfs_key key;
3842 struct btrfs_key found_key;
3843 struct extent_buffer *leaf;
3846 int enospc_errors = 0;
3847 bool counting = true;
3848 /* The single value limit and min/max limits use the same bytes in the */
3849 u64 limit_data = bctl->data.limit;
3850 u64 limit_meta = bctl->meta.limit;
3851 u64 limit_sys = bctl->sys.limit;
3855 int chunk_reserved = 0;
3857 path = btrfs_alloc_path();
3863 /* zero out stat counters */
3864 spin_lock(&fs_info->balance_lock);
3865 memset(&bctl->stat, 0, sizeof(bctl->stat));
3866 spin_unlock(&fs_info->balance_lock);
3870 * The single value limit and min/max limits use the same bytes
3873 bctl->data.limit = limit_data;
3874 bctl->meta.limit = limit_meta;
3875 bctl->sys.limit = limit_sys;
3877 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3878 key.offset = (u64)-1;
3879 key.type = BTRFS_CHUNK_ITEM_KEY;
3882 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3883 atomic_read(&fs_info->balance_cancel_req)) {
3888 mutex_lock(&fs_info->reclaim_bgs_lock);
3889 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3891 mutex_unlock(&fs_info->reclaim_bgs_lock);
3896 * this shouldn't happen, it means the last relocate
3900 BUG(); /* FIXME break ? */
3902 ret = btrfs_previous_item(chunk_root, path, 0,
3903 BTRFS_CHUNK_ITEM_KEY);
3905 mutex_unlock(&fs_info->reclaim_bgs_lock);
3910 leaf = path->nodes[0];
3911 slot = path->slots[0];
3912 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3914 if (found_key.objectid != key.objectid) {
3915 mutex_unlock(&fs_info->reclaim_bgs_lock);
3919 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3920 chunk_type = btrfs_chunk_type(leaf, chunk);
3923 spin_lock(&fs_info->balance_lock);
3924 bctl->stat.considered++;
3925 spin_unlock(&fs_info->balance_lock);
3928 ret = should_balance_chunk(leaf, chunk, found_key.offset);
3930 btrfs_release_path(path);
3932 mutex_unlock(&fs_info->reclaim_bgs_lock);
3937 mutex_unlock(&fs_info->reclaim_bgs_lock);
3938 spin_lock(&fs_info->balance_lock);
3939 bctl->stat.expected++;
3940 spin_unlock(&fs_info->balance_lock);
3942 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3944 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3946 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3953 * Apply limit_min filter, no need to check if the LIMITS
3954 * filter is used, limit_min is 0 by default
3956 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3957 count_data < bctl->data.limit_min)
3958 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3959 count_meta < bctl->meta.limit_min)
3960 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3961 count_sys < bctl->sys.limit_min)) {
3962 mutex_unlock(&fs_info->reclaim_bgs_lock);
3966 if (!chunk_reserved) {
3968 * We may be relocating the only data chunk we have,
3969 * which could potentially end up with losing data's
3970 * raid profile, so lets allocate an empty one in
3973 ret = btrfs_may_alloc_data_chunk(fs_info,
3976 mutex_unlock(&fs_info->reclaim_bgs_lock);
3978 } else if (ret == 1) {
3983 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3984 mutex_unlock(&fs_info->reclaim_bgs_lock);
3985 if (ret == -ENOSPC) {
3987 } else if (ret == -ETXTBSY) {
3989 "skipping relocation of block group %llu due to active swapfile",
3995 spin_lock(&fs_info->balance_lock);
3996 bctl->stat.completed++;
3997 spin_unlock(&fs_info->balance_lock);
4000 if (found_key.offset == 0)
4002 key.offset = found_key.offset - 1;
4006 btrfs_release_path(path);
4011 btrfs_free_path(path);
4012 if (enospc_errors) {
4013 btrfs_info(fs_info, "%d enospc errors during balance",
4023 * alloc_profile_is_valid - see if a given profile is valid and reduced
4024 * @flags: profile to validate
4025 * @extended: if true @flags is treated as an extended profile
4027 static int alloc_profile_is_valid(u64 flags, int extended)
4029 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
4030 BTRFS_BLOCK_GROUP_PROFILE_MASK);
4032 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
4034 /* 1) check that all other bits are zeroed */
4038 /* 2) see if profile is reduced */
4040 return !extended; /* "0" is valid for usual profiles */
4042 return has_single_bit_set(flags);
4045 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
4047 /* cancel requested || normal exit path */
4048 return atomic_read(&fs_info->balance_cancel_req) ||
4049 (atomic_read(&fs_info->balance_pause_req) == 0 &&
4050 atomic_read(&fs_info->balance_cancel_req) == 0);
4054 * Validate target profile against allowed profiles and return true if it's OK.
4055 * Otherwise print the error message and return false.
4057 static inline int validate_convert_profile(struct btrfs_fs_info *fs_info,
4058 const struct btrfs_balance_args *bargs,
4059 u64 allowed, const char *type)
4061 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
4064 if (fs_info->sectorsize < PAGE_SIZE &&
4065 bargs->target & BTRFS_BLOCK_GROUP_RAID56_MASK) {
4067 "RAID56 is not yet supported for sectorsize %u with page size %lu",
4068 fs_info->sectorsize, PAGE_SIZE);
4071 /* Profile is valid and does not have bits outside of the allowed set */
4072 if (alloc_profile_is_valid(bargs->target, 1) &&
4073 (bargs->target & ~allowed) == 0)
4076 btrfs_err(fs_info, "balance: invalid convert %s profile %s",
4077 type, btrfs_bg_type_to_raid_name(bargs->target));
4082 * Fill @buf with textual description of balance filter flags @bargs, up to
4083 * @size_buf including the terminating null. The output may be trimmed if it
4084 * does not fit into the provided buffer.
4086 static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf,
4090 u32 size_bp = size_buf;
4092 u64 flags = bargs->flags;
4093 char tmp_buf[128] = {'\0'};
4098 #define CHECK_APPEND_NOARG(a) \
4100 ret = snprintf(bp, size_bp, (a)); \
4101 if (ret < 0 || ret >= size_bp) \
4102 goto out_overflow; \
4107 #define CHECK_APPEND_1ARG(a, v1) \
4109 ret = snprintf(bp, size_bp, (a), (v1)); \
4110 if (ret < 0 || ret >= size_bp) \
4111 goto out_overflow; \
4116 #define CHECK_APPEND_2ARG(a, v1, v2) \
4118 ret = snprintf(bp, size_bp, (a), (v1), (v2)); \
4119 if (ret < 0 || ret >= size_bp) \
4120 goto out_overflow; \
4125 if (flags & BTRFS_BALANCE_ARGS_CONVERT)
4126 CHECK_APPEND_1ARG("convert=%s,",
4127 btrfs_bg_type_to_raid_name(bargs->target));
4129 if (flags & BTRFS_BALANCE_ARGS_SOFT)
4130 CHECK_APPEND_NOARG("soft,");
4132 if (flags & BTRFS_BALANCE_ARGS_PROFILES) {
4133 btrfs_describe_block_groups(bargs->profiles, tmp_buf,
4135 CHECK_APPEND_1ARG("profiles=%s,", tmp_buf);
4138 if (flags & BTRFS_BALANCE_ARGS_USAGE)
4139 CHECK_APPEND_1ARG("usage=%llu,", bargs->usage);
4141 if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE)
4142 CHECK_APPEND_2ARG("usage=%u..%u,",
4143 bargs->usage_min, bargs->usage_max);
4145 if (flags & BTRFS_BALANCE_ARGS_DEVID)
4146 CHECK_APPEND_1ARG("devid=%llu,", bargs->devid);
4148 if (flags & BTRFS_BALANCE_ARGS_DRANGE)
4149 CHECK_APPEND_2ARG("drange=%llu..%llu,",
4150 bargs->pstart, bargs->pend);
4152 if (flags & BTRFS_BALANCE_ARGS_VRANGE)
4153 CHECK_APPEND_2ARG("vrange=%llu..%llu,",
4154 bargs->vstart, bargs->vend);
4156 if (flags & BTRFS_BALANCE_ARGS_LIMIT)
4157 CHECK_APPEND_1ARG("limit=%llu,", bargs->limit);
4159 if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)
4160 CHECK_APPEND_2ARG("limit=%u..%u,",
4161 bargs->limit_min, bargs->limit_max);
4163 if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE)
4164 CHECK_APPEND_2ARG("stripes=%u..%u,",
4165 bargs->stripes_min, bargs->stripes_max);
4167 #undef CHECK_APPEND_2ARG
4168 #undef CHECK_APPEND_1ARG
4169 #undef CHECK_APPEND_NOARG
4173 if (size_bp < size_buf)
4174 buf[size_buf - size_bp - 1] = '\0'; /* remove last , */
4179 static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info)
4181 u32 size_buf = 1024;
4182 char tmp_buf[192] = {'\0'};
4185 u32 size_bp = size_buf;
4187 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4189 buf = kzalloc(size_buf, GFP_KERNEL);
4195 #define CHECK_APPEND_1ARG(a, v1) \
4197 ret = snprintf(bp, size_bp, (a), (v1)); \
4198 if (ret < 0 || ret >= size_bp) \
4199 goto out_overflow; \
4204 if (bctl->flags & BTRFS_BALANCE_FORCE)
4205 CHECK_APPEND_1ARG("%s", "-f ");
4207 if (bctl->flags & BTRFS_BALANCE_DATA) {
4208 describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf));
4209 CHECK_APPEND_1ARG("-d%s ", tmp_buf);
4212 if (bctl->flags & BTRFS_BALANCE_METADATA) {
4213 describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf));
4214 CHECK_APPEND_1ARG("-m%s ", tmp_buf);
4217 if (bctl->flags & BTRFS_BALANCE_SYSTEM) {
4218 describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf));
4219 CHECK_APPEND_1ARG("-s%s ", tmp_buf);
4222 #undef CHECK_APPEND_1ARG
4226 if (size_bp < size_buf)
4227 buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */
4228 btrfs_info(fs_info, "balance: %s %s",
4229 (bctl->flags & BTRFS_BALANCE_RESUME) ?
4230 "resume" : "start", buf);
4236 * Should be called with balance mutexe held
4238 int btrfs_balance(struct btrfs_fs_info *fs_info,
4239 struct btrfs_balance_control *bctl,
4240 struct btrfs_ioctl_balance_args *bargs)
4242 u64 meta_target, data_target;
4248 bool reducing_redundancy;
4251 if (btrfs_fs_closing(fs_info) ||
4252 atomic_read(&fs_info->balance_pause_req) ||
4253 btrfs_should_cancel_balance(fs_info)) {
4258 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
4259 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
4263 * In case of mixed groups both data and meta should be picked,
4264 * and identical options should be given for both of them.
4266 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
4267 if (mixed && (bctl->flags & allowed)) {
4268 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
4269 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
4270 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
4272 "balance: mixed groups data and metadata options must be the same");
4279 * rw_devices will not change at the moment, device add/delete/replace
4282 num_devices = fs_info->fs_devices->rw_devices;
4285 * SINGLE profile on-disk has no profile bit, but in-memory we have a
4286 * special bit for it, to make it easier to distinguish. Thus we need
4287 * to set it manually, or balance would refuse the profile.
4289 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
4290 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++)
4291 if (num_devices >= btrfs_raid_array[i].devs_min)
4292 allowed |= btrfs_raid_array[i].bg_flag;
4294 if (!validate_convert_profile(fs_info, &bctl->data, allowed, "data") ||
4295 !validate_convert_profile(fs_info, &bctl->meta, allowed, "metadata") ||
4296 !validate_convert_profile(fs_info, &bctl->sys, allowed, "system")) {
4302 * Allow to reduce metadata or system integrity only if force set for
4303 * profiles with redundancy (copies, parity)
4306 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) {
4307 if (btrfs_raid_array[i].ncopies >= 2 ||
4308 btrfs_raid_array[i].tolerated_failures >= 1)
4309 allowed |= btrfs_raid_array[i].bg_flag;
4312 seq = read_seqbegin(&fs_info->profiles_lock);
4314 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4315 (fs_info->avail_system_alloc_bits & allowed) &&
4316 !(bctl->sys.target & allowed)) ||
4317 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4318 (fs_info->avail_metadata_alloc_bits & allowed) &&
4319 !(bctl->meta.target & allowed)))
4320 reducing_redundancy = true;
4322 reducing_redundancy = false;
4324 /* if we're not converting, the target field is uninitialized */
4325 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4326 bctl->meta.target : fs_info->avail_metadata_alloc_bits;
4327 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4328 bctl->data.target : fs_info->avail_data_alloc_bits;
4329 } while (read_seqretry(&fs_info->profiles_lock, seq));
4331 if (reducing_redundancy) {
4332 if (bctl->flags & BTRFS_BALANCE_FORCE) {
4334 "balance: force reducing metadata redundancy");
4337 "balance: reduces metadata redundancy, use --force if you want this");
4343 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <
4344 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {
4346 "balance: metadata profile %s has lower redundancy than data profile %s",
4347 btrfs_bg_type_to_raid_name(meta_target),
4348 btrfs_bg_type_to_raid_name(data_target));
4351 ret = insert_balance_item(fs_info, bctl);
4352 if (ret && ret != -EEXIST)
4355 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
4356 BUG_ON(ret == -EEXIST);
4357 BUG_ON(fs_info->balance_ctl);
4358 spin_lock(&fs_info->balance_lock);
4359 fs_info->balance_ctl = bctl;
4360 spin_unlock(&fs_info->balance_lock);
4362 BUG_ON(ret != -EEXIST);
4363 spin_lock(&fs_info->balance_lock);
4364 update_balance_args(bctl);
4365 spin_unlock(&fs_info->balance_lock);
4368 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4369 set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4370 describe_balance_start_or_resume(fs_info);
4371 mutex_unlock(&fs_info->balance_mutex);
4373 ret = __btrfs_balance(fs_info);
4375 mutex_lock(&fs_info->balance_mutex);
4376 if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req))
4377 btrfs_info(fs_info, "balance: paused");
4379 * Balance can be canceled by:
4381 * - Regular cancel request
4382 * Then ret == -ECANCELED and balance_cancel_req > 0
4384 * - Fatal signal to "btrfs" process
4385 * Either the signal caught by wait_reserve_ticket() and callers
4386 * got -EINTR, or caught by btrfs_should_cancel_balance() and
4388 * Either way, in this case balance_cancel_req = 0, and
4389 * ret == -EINTR or ret == -ECANCELED.
4391 * So here we only check the return value to catch canceled balance.
4393 else if (ret == -ECANCELED || ret == -EINTR)
4394 btrfs_info(fs_info, "balance: canceled");
4396 btrfs_info(fs_info, "balance: ended with status: %d", ret);
4398 clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4401 memset(bargs, 0, sizeof(*bargs));
4402 btrfs_update_ioctl_balance_args(fs_info, bargs);
4405 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
4406 balance_need_close(fs_info)) {
4407 reset_balance_state(fs_info);
4408 btrfs_exclop_finish(fs_info);
4411 wake_up(&fs_info->balance_wait_q);
4415 if (bctl->flags & BTRFS_BALANCE_RESUME)
4416 reset_balance_state(fs_info);
4419 btrfs_exclop_finish(fs_info);
4424 static int balance_kthread(void *data)
4426 struct btrfs_fs_info *fs_info = data;
4429 sb_start_write(fs_info->sb);
4430 mutex_lock(&fs_info->balance_mutex);
4431 if (fs_info->balance_ctl)
4432 ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL);
4433 mutex_unlock(&fs_info->balance_mutex);
4434 sb_end_write(fs_info->sb);
4439 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
4441 struct task_struct *tsk;
4443 mutex_lock(&fs_info->balance_mutex);
4444 if (!fs_info->balance_ctl) {
4445 mutex_unlock(&fs_info->balance_mutex);
4448 mutex_unlock(&fs_info->balance_mutex);
4450 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
4451 btrfs_info(fs_info, "balance: resume skipped");
4456 * A ro->rw remount sequence should continue with the paused balance
4457 * regardless of who pauses it, system or the user as of now, so set
4460 spin_lock(&fs_info->balance_lock);
4461 fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME;
4462 spin_unlock(&fs_info->balance_lock);
4464 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
4465 return PTR_ERR_OR_ZERO(tsk);
4468 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
4470 struct btrfs_balance_control *bctl;
4471 struct btrfs_balance_item *item;
4472 struct btrfs_disk_balance_args disk_bargs;
4473 struct btrfs_path *path;
4474 struct extent_buffer *leaf;
4475 struct btrfs_key key;
4478 path = btrfs_alloc_path();
4482 key.objectid = BTRFS_BALANCE_OBJECTID;
4483 key.type = BTRFS_TEMPORARY_ITEM_KEY;
4486 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4489 if (ret > 0) { /* ret = -ENOENT; */
4494 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4500 leaf = path->nodes[0];
4501 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
4503 bctl->flags = btrfs_balance_flags(leaf, item);
4504 bctl->flags |= BTRFS_BALANCE_RESUME;
4506 btrfs_balance_data(leaf, item, &disk_bargs);
4507 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4508 btrfs_balance_meta(leaf, item, &disk_bargs);
4509 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4510 btrfs_balance_sys(leaf, item, &disk_bargs);
4511 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4514 * This should never happen, as the paused balance state is recovered
4515 * during mount without any chance of other exclusive ops to collide.
4517 * This gives the exclusive op status to balance and keeps in paused
4518 * state until user intervention (cancel or umount). If the ownership
4519 * cannot be assigned, show a message but do not fail. The balance
4520 * is in a paused state and must have fs_info::balance_ctl properly
4523 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE))
4525 "balance: cannot set exclusive op status, resume manually");
4527 btrfs_release_path(path);
4529 mutex_lock(&fs_info->balance_mutex);
4530 BUG_ON(fs_info->balance_ctl);
4531 spin_lock(&fs_info->balance_lock);
4532 fs_info->balance_ctl = bctl;
4533 spin_unlock(&fs_info->balance_lock);
4534 mutex_unlock(&fs_info->balance_mutex);
4536 btrfs_free_path(path);
4540 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4544 mutex_lock(&fs_info->balance_mutex);
4545 if (!fs_info->balance_ctl) {
4546 mutex_unlock(&fs_info->balance_mutex);
4550 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4551 atomic_inc(&fs_info->balance_pause_req);
4552 mutex_unlock(&fs_info->balance_mutex);
4554 wait_event(fs_info->balance_wait_q,
4555 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4557 mutex_lock(&fs_info->balance_mutex);
4558 /* we are good with balance_ctl ripped off from under us */
4559 BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4560 atomic_dec(&fs_info->balance_pause_req);
4565 mutex_unlock(&fs_info->balance_mutex);
4569 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4571 mutex_lock(&fs_info->balance_mutex);
4572 if (!fs_info->balance_ctl) {
4573 mutex_unlock(&fs_info->balance_mutex);
4578 * A paused balance with the item stored on disk can be resumed at
4579 * mount time if the mount is read-write. Otherwise it's still paused
4580 * and we must not allow cancelling as it deletes the item.
4582 if (sb_rdonly(fs_info->sb)) {
4583 mutex_unlock(&fs_info->balance_mutex);
4587 atomic_inc(&fs_info->balance_cancel_req);
4589 * if we are running just wait and return, balance item is
4590 * deleted in btrfs_balance in this case
4592 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4593 mutex_unlock(&fs_info->balance_mutex);
4594 wait_event(fs_info->balance_wait_q,
4595 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4596 mutex_lock(&fs_info->balance_mutex);
4598 mutex_unlock(&fs_info->balance_mutex);
4600 * Lock released to allow other waiters to continue, we'll
4601 * reexamine the status again.
4603 mutex_lock(&fs_info->balance_mutex);
4605 if (fs_info->balance_ctl) {
4606 reset_balance_state(fs_info);
4607 btrfs_exclop_finish(fs_info);
4608 btrfs_info(fs_info, "balance: canceled");
4612 BUG_ON(fs_info->balance_ctl ||
4613 test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4614 atomic_dec(&fs_info->balance_cancel_req);
4615 mutex_unlock(&fs_info->balance_mutex);
4619 int btrfs_uuid_scan_kthread(void *data)
4621 struct btrfs_fs_info *fs_info = data;
4622 struct btrfs_root *root = fs_info->tree_root;
4623 struct btrfs_key key;
4624 struct btrfs_path *path = NULL;
4626 struct extent_buffer *eb;
4628 struct btrfs_root_item root_item;
4630 struct btrfs_trans_handle *trans = NULL;
4631 bool closing = false;
4633 path = btrfs_alloc_path();
4640 key.type = BTRFS_ROOT_ITEM_KEY;
4644 if (btrfs_fs_closing(fs_info)) {
4648 ret = btrfs_search_forward(root, &key, path,
4649 BTRFS_OLDEST_GENERATION);
4656 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4657 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4658 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4659 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4662 eb = path->nodes[0];
4663 slot = path->slots[0];
4664 item_size = btrfs_item_size_nr(eb, slot);
4665 if (item_size < sizeof(root_item))
4668 read_extent_buffer(eb, &root_item,
4669 btrfs_item_ptr_offset(eb, slot),
4670 (int)sizeof(root_item));
4671 if (btrfs_root_refs(&root_item) == 0)
4674 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4675 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4679 btrfs_release_path(path);
4681 * 1 - subvol uuid item
4682 * 1 - received_subvol uuid item
4684 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4685 if (IS_ERR(trans)) {
4686 ret = PTR_ERR(trans);
4694 btrfs_release_path(path);
4695 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4696 ret = btrfs_uuid_tree_add(trans, root_item.uuid,
4697 BTRFS_UUID_KEY_SUBVOL,
4700 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4706 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4707 ret = btrfs_uuid_tree_add(trans,
4708 root_item.received_uuid,
4709 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4712 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4719 btrfs_release_path(path);
4721 ret = btrfs_end_transaction(trans);
4727 if (key.offset < (u64)-1) {
4729 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4731 key.type = BTRFS_ROOT_ITEM_KEY;
4732 } else if (key.objectid < (u64)-1) {
4734 key.type = BTRFS_ROOT_ITEM_KEY;
4743 btrfs_free_path(path);
4744 if (trans && !IS_ERR(trans))
4745 btrfs_end_transaction(trans);
4747 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4749 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4750 up(&fs_info->uuid_tree_rescan_sem);
4754 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4756 struct btrfs_trans_handle *trans;
4757 struct btrfs_root *tree_root = fs_info->tree_root;
4758 struct btrfs_root *uuid_root;
4759 struct task_struct *task;
4766 trans = btrfs_start_transaction(tree_root, 2);
4768 return PTR_ERR(trans);
4770 uuid_root = btrfs_create_tree(trans, BTRFS_UUID_TREE_OBJECTID);
4771 if (IS_ERR(uuid_root)) {
4772 ret = PTR_ERR(uuid_root);
4773 btrfs_abort_transaction(trans, ret);
4774 btrfs_end_transaction(trans);
4778 fs_info->uuid_root = uuid_root;
4780 ret = btrfs_commit_transaction(trans);
4784 down(&fs_info->uuid_tree_rescan_sem);
4785 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4787 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4788 btrfs_warn(fs_info, "failed to start uuid_scan task");
4789 up(&fs_info->uuid_tree_rescan_sem);
4790 return PTR_ERR(task);
4797 * shrinking a device means finding all of the device extents past
4798 * the new size, and then following the back refs to the chunks.
4799 * The chunk relocation code actually frees the device extent
4801 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4803 struct btrfs_fs_info *fs_info = device->fs_info;
4804 struct btrfs_root *root = fs_info->dev_root;
4805 struct btrfs_trans_handle *trans;
4806 struct btrfs_dev_extent *dev_extent = NULL;
4807 struct btrfs_path *path;
4813 bool retried = false;
4814 struct extent_buffer *l;
4815 struct btrfs_key key;
4816 struct btrfs_super_block *super_copy = fs_info->super_copy;
4817 u64 old_total = btrfs_super_total_bytes(super_copy);
4818 u64 old_size = btrfs_device_get_total_bytes(device);
4822 new_size = round_down(new_size, fs_info->sectorsize);
4824 diff = round_down(old_size - new_size, fs_info->sectorsize);
4826 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4829 path = btrfs_alloc_path();
4833 path->reada = READA_BACK;
4835 trans = btrfs_start_transaction(root, 0);
4836 if (IS_ERR(trans)) {
4837 btrfs_free_path(path);
4838 return PTR_ERR(trans);
4841 mutex_lock(&fs_info->chunk_mutex);
4843 btrfs_device_set_total_bytes(device, new_size);
4844 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4845 device->fs_devices->total_rw_bytes -= diff;
4846 atomic64_sub(diff, &fs_info->free_chunk_space);
4850 * Once the device's size has been set to the new size, ensure all
4851 * in-memory chunks are synced to disk so that the loop below sees them
4852 * and relocates them accordingly.
4854 if (contains_pending_extent(device, &start, diff)) {
4855 mutex_unlock(&fs_info->chunk_mutex);
4856 ret = btrfs_commit_transaction(trans);
4860 mutex_unlock(&fs_info->chunk_mutex);
4861 btrfs_end_transaction(trans);
4865 key.objectid = device->devid;
4866 key.offset = (u64)-1;
4867 key.type = BTRFS_DEV_EXTENT_KEY;
4870 mutex_lock(&fs_info->reclaim_bgs_lock);
4871 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4873 mutex_unlock(&fs_info->reclaim_bgs_lock);
4877 ret = btrfs_previous_item(root, path, 0, key.type);
4879 mutex_unlock(&fs_info->reclaim_bgs_lock);
4883 btrfs_release_path(path);
4888 slot = path->slots[0];
4889 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4891 if (key.objectid != device->devid) {
4892 mutex_unlock(&fs_info->reclaim_bgs_lock);
4893 btrfs_release_path(path);
4897 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4898 length = btrfs_dev_extent_length(l, dev_extent);
4900 if (key.offset + length <= new_size) {
4901 mutex_unlock(&fs_info->reclaim_bgs_lock);
4902 btrfs_release_path(path);
4906 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4907 btrfs_release_path(path);
4910 * We may be relocating the only data chunk we have,
4911 * which could potentially end up with losing data's
4912 * raid profile, so lets allocate an empty one in
4915 ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset);
4917 mutex_unlock(&fs_info->reclaim_bgs_lock);
4921 ret = btrfs_relocate_chunk(fs_info, chunk_offset);
4922 mutex_unlock(&fs_info->reclaim_bgs_lock);
4923 if (ret == -ENOSPC) {
4926 if (ret == -ETXTBSY) {
4928 "could not shrink block group %llu due to active swapfile",
4933 } while (key.offset-- > 0);
4935 if (failed && !retried) {
4939 } else if (failed && retried) {
4944 /* Shrinking succeeded, else we would be at "done". */
4945 trans = btrfs_start_transaction(root, 0);
4946 if (IS_ERR(trans)) {
4947 ret = PTR_ERR(trans);
4951 mutex_lock(&fs_info->chunk_mutex);
4952 /* Clear all state bits beyond the shrunk device size */
4953 clear_extent_bits(&device->alloc_state, new_size, (u64)-1,
4956 btrfs_device_set_disk_total_bytes(device, new_size);
4957 if (list_empty(&device->post_commit_list))
4958 list_add_tail(&device->post_commit_list,
4959 &trans->transaction->dev_update_list);
4961 WARN_ON(diff > old_total);
4962 btrfs_set_super_total_bytes(super_copy,
4963 round_down(old_total - diff, fs_info->sectorsize));
4964 mutex_unlock(&fs_info->chunk_mutex);
4966 btrfs_reserve_chunk_metadata(trans, false);
4967 /* Now btrfs_update_device() will change the on-disk size. */
4968 ret = btrfs_update_device(trans, device);
4969 btrfs_trans_release_chunk_metadata(trans);
4971 btrfs_abort_transaction(trans, ret);
4972 btrfs_end_transaction(trans);
4974 ret = btrfs_commit_transaction(trans);
4977 btrfs_free_path(path);
4979 mutex_lock(&fs_info->chunk_mutex);
4980 btrfs_device_set_total_bytes(device, old_size);
4981 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
4982 device->fs_devices->total_rw_bytes += diff;
4983 atomic64_add(diff, &fs_info->free_chunk_space);
4984 mutex_unlock(&fs_info->chunk_mutex);
4989 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
4990 struct btrfs_key *key,
4991 struct btrfs_chunk *chunk, int item_size)
4993 struct btrfs_super_block *super_copy = fs_info->super_copy;
4994 struct btrfs_disk_key disk_key;
4998 lockdep_assert_held(&fs_info->chunk_mutex);
5000 array_size = btrfs_super_sys_array_size(super_copy);
5001 if (array_size + item_size + sizeof(disk_key)
5002 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
5005 ptr = super_copy->sys_chunk_array + array_size;
5006 btrfs_cpu_key_to_disk(&disk_key, key);
5007 memcpy(ptr, &disk_key, sizeof(disk_key));
5008 ptr += sizeof(disk_key);
5009 memcpy(ptr, chunk, item_size);
5010 item_size += sizeof(disk_key);
5011 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
5017 * sort the devices in descending order by max_avail, total_avail
5019 static int btrfs_cmp_device_info(const void *a, const void *b)
5021 const struct btrfs_device_info *di_a = a;
5022 const struct btrfs_device_info *di_b = b;
5024 if (di_a->max_avail > di_b->max_avail)
5026 if (di_a->max_avail < di_b->max_avail)
5028 if (di_a->total_avail > di_b->total_avail)
5030 if (di_a->total_avail < di_b->total_avail)
5035 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
5037 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
5040 btrfs_set_fs_incompat(info, RAID56);
5043 static void check_raid1c34_incompat_flag(struct btrfs_fs_info *info, u64 type)
5045 if (!(type & (BTRFS_BLOCK_GROUP_RAID1C3 | BTRFS_BLOCK_GROUP_RAID1C4)))
5048 btrfs_set_fs_incompat(info, RAID1C34);
5052 * Structure used internally for btrfs_create_chunk() function.
5053 * Wraps needed parameters.
5055 struct alloc_chunk_ctl {
5058 /* Total number of stripes to allocate */
5060 /* sub_stripes info for map */
5062 /* Stripes per device */
5064 /* Maximum number of devices to use */
5066 /* Minimum number of devices to use */
5068 /* ndevs has to be a multiple of this */
5070 /* Number of copies */
5072 /* Number of stripes worth of bytes to store parity information */
5074 u64 max_stripe_size;
5082 static void init_alloc_chunk_ctl_policy_regular(
5083 struct btrfs_fs_devices *fs_devices,
5084 struct alloc_chunk_ctl *ctl)
5086 u64 type = ctl->type;
5088 if (type & BTRFS_BLOCK_GROUP_DATA) {
5089 ctl->max_stripe_size = SZ_1G;
5090 ctl->max_chunk_size = BTRFS_MAX_DATA_CHUNK_SIZE;
5091 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
5092 /* For larger filesystems, use larger metadata chunks */
5093 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
5094 ctl->max_stripe_size = SZ_1G;
5096 ctl->max_stripe_size = SZ_256M;
5097 ctl->max_chunk_size = ctl->max_stripe_size;
5098 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
5099 ctl->max_stripe_size = SZ_32M;
5100 ctl->max_chunk_size = 2 * ctl->max_stripe_size;
5101 ctl->devs_max = min_t(int, ctl->devs_max,
5102 BTRFS_MAX_DEVS_SYS_CHUNK);
5107 /* We don't want a chunk larger than 10% of writable space */
5108 ctl->max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
5109 ctl->max_chunk_size);
5110 ctl->dev_extent_min = BTRFS_STRIPE_LEN * ctl->dev_stripes;
5113 static void init_alloc_chunk_ctl_policy_zoned(
5114 struct btrfs_fs_devices *fs_devices,
5115 struct alloc_chunk_ctl *ctl)
5117 u64 zone_size = fs_devices->fs_info->zone_size;
5119 int min_num_stripes = ctl->devs_min * ctl->dev_stripes;
5120 int min_data_stripes = (min_num_stripes - ctl->nparity) / ctl->ncopies;
5121 u64 min_chunk_size = min_data_stripes * zone_size;
5122 u64 type = ctl->type;
5124 ctl->max_stripe_size = zone_size;
5125 if (type & BTRFS_BLOCK_GROUP_DATA) {
5126 ctl->max_chunk_size = round_down(BTRFS_MAX_DATA_CHUNK_SIZE,
5128 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
5129 ctl->max_chunk_size = ctl->max_stripe_size;
5130 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
5131 ctl->max_chunk_size = 2 * ctl->max_stripe_size;
5132 ctl->devs_max = min_t(int, ctl->devs_max,
5133 BTRFS_MAX_DEVS_SYS_CHUNK);
5138 /* We don't want a chunk larger than 10% of writable space */
5139 limit = max(round_down(div_factor(fs_devices->total_rw_bytes, 1),
5142 ctl->max_chunk_size = min(limit, ctl->max_chunk_size);
5143 ctl->dev_extent_min = zone_size * ctl->dev_stripes;
5146 static void init_alloc_chunk_ctl(struct btrfs_fs_devices *fs_devices,
5147 struct alloc_chunk_ctl *ctl)
5149 int index = btrfs_bg_flags_to_raid_index(ctl->type);
5151 ctl->sub_stripes = btrfs_raid_array[index].sub_stripes;
5152 ctl->dev_stripes = btrfs_raid_array[index].dev_stripes;
5153 ctl->devs_max = btrfs_raid_array[index].devs_max;
5155 ctl->devs_max = BTRFS_MAX_DEVS(fs_devices->fs_info);
5156 ctl->devs_min = btrfs_raid_array[index].devs_min;
5157 ctl->devs_increment = btrfs_raid_array[index].devs_increment;
5158 ctl->ncopies = btrfs_raid_array[index].ncopies;
5159 ctl->nparity = btrfs_raid_array[index].nparity;
5162 switch (fs_devices->chunk_alloc_policy) {
5163 case BTRFS_CHUNK_ALLOC_REGULAR:
5164 init_alloc_chunk_ctl_policy_regular(fs_devices, ctl);
5166 case BTRFS_CHUNK_ALLOC_ZONED:
5167 init_alloc_chunk_ctl_policy_zoned(fs_devices, ctl);
5174 static int gather_device_info(struct btrfs_fs_devices *fs_devices,
5175 struct alloc_chunk_ctl *ctl,
5176 struct btrfs_device_info *devices_info)
5178 struct btrfs_fs_info *info = fs_devices->fs_info;
5179 struct btrfs_device *device;
5181 u64 dev_extent_want = ctl->max_stripe_size * ctl->dev_stripes;
5188 * in the first pass through the devices list, we gather information
5189 * about the available holes on each device.
5191 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
5192 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
5194 "BTRFS: read-only device in alloc_list\n");
5198 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
5199 &device->dev_state) ||
5200 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
5203 if (device->total_bytes > device->bytes_used)
5204 total_avail = device->total_bytes - device->bytes_used;
5208 /* If there is no space on this device, skip it. */
5209 if (total_avail < ctl->dev_extent_min)
5212 ret = find_free_dev_extent(device, dev_extent_want, &dev_offset,
5214 if (ret && ret != -ENOSPC)
5218 max_avail = dev_extent_want;
5220 if (max_avail < ctl->dev_extent_min) {
5221 if (btrfs_test_opt(info, ENOSPC_DEBUG))
5223 "%s: devid %llu has no free space, have=%llu want=%llu",
5224 __func__, device->devid, max_avail,
5225 ctl->dev_extent_min);
5229 if (ndevs == fs_devices->rw_devices) {
5230 WARN(1, "%s: found more than %llu devices\n",
5231 __func__, fs_devices->rw_devices);
5234 devices_info[ndevs].dev_offset = dev_offset;
5235 devices_info[ndevs].max_avail = max_avail;
5236 devices_info[ndevs].total_avail = total_avail;
5237 devices_info[ndevs].dev = device;
5243 * now sort the devices by hole size / available space
5245 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
5246 btrfs_cmp_device_info, NULL);
5251 static int decide_stripe_size_regular(struct alloc_chunk_ctl *ctl,
5252 struct btrfs_device_info *devices_info)
5254 /* Number of stripes that count for block group size */
5258 * The primary goal is to maximize the number of stripes, so use as
5259 * many devices as possible, even if the stripes are not maximum sized.
5261 * The DUP profile stores more than one stripe per device, the
5262 * max_avail is the total size so we have to adjust.
5264 ctl->stripe_size = div_u64(devices_info[ctl->ndevs - 1].max_avail,
5266 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5268 /* This will have to be fixed for RAID1 and RAID10 over more drives */
5269 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5272 * Use the number of data stripes to figure out how big this chunk is
5273 * really going to be in terms of logical address space, and compare
5274 * that answer with the max chunk size. If it's higher, we try to
5275 * reduce stripe_size.
5277 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) {
5279 * Reduce stripe_size, round it up to a 16MB boundary again and
5280 * then use it, unless it ends up being even bigger than the
5281 * previous value we had already.
5283 ctl->stripe_size = min(round_up(div_u64(ctl->max_chunk_size,
5284 data_stripes), SZ_16M),
5288 /* Align to BTRFS_STRIPE_LEN */
5289 ctl->stripe_size = round_down(ctl->stripe_size, BTRFS_STRIPE_LEN);
5290 ctl->chunk_size = ctl->stripe_size * data_stripes;
5295 static int decide_stripe_size_zoned(struct alloc_chunk_ctl *ctl,
5296 struct btrfs_device_info *devices_info)
5298 u64 zone_size = devices_info[0].dev->zone_info->zone_size;
5299 /* Number of stripes that count for block group size */
5303 * It should hold because:
5304 * dev_extent_min == dev_extent_want == zone_size * dev_stripes
5306 ASSERT(devices_info[ctl->ndevs - 1].max_avail == ctl->dev_extent_min);
5308 ctl->stripe_size = zone_size;
5309 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5310 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5312 /* stripe_size is fixed in zoned filesysmte. Reduce ndevs instead. */
5313 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) {
5314 ctl->ndevs = div_u64(div_u64(ctl->max_chunk_size * ctl->ncopies,
5315 ctl->stripe_size) + ctl->nparity,
5317 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5318 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5319 ASSERT(ctl->stripe_size * data_stripes <= ctl->max_chunk_size);
5322 ctl->chunk_size = ctl->stripe_size * data_stripes;
5327 static int decide_stripe_size(struct btrfs_fs_devices *fs_devices,
5328 struct alloc_chunk_ctl *ctl,
5329 struct btrfs_device_info *devices_info)
5331 struct btrfs_fs_info *info = fs_devices->fs_info;
5334 * Round down to number of usable stripes, devs_increment can be any
5335 * number so we can't use round_down() that requires power of 2, while
5336 * rounddown is safe.
5338 ctl->ndevs = rounddown(ctl->ndevs, ctl->devs_increment);
5340 if (ctl->ndevs < ctl->devs_min) {
5341 if (btrfs_test_opt(info, ENOSPC_DEBUG)) {
5343 "%s: not enough devices with free space: have=%d minimum required=%d",
5344 __func__, ctl->ndevs, ctl->devs_min);
5349 ctl->ndevs = min(ctl->ndevs, ctl->devs_max);
5351 switch (fs_devices->chunk_alloc_policy) {
5352 case BTRFS_CHUNK_ALLOC_REGULAR:
5353 return decide_stripe_size_regular(ctl, devices_info);
5354 case BTRFS_CHUNK_ALLOC_ZONED:
5355 return decide_stripe_size_zoned(ctl, devices_info);
5361 static struct btrfs_block_group *create_chunk(struct btrfs_trans_handle *trans,
5362 struct alloc_chunk_ctl *ctl,
5363 struct btrfs_device_info *devices_info)
5365 struct btrfs_fs_info *info = trans->fs_info;
5366 struct map_lookup *map = NULL;
5367 struct extent_map_tree *em_tree;
5368 struct btrfs_block_group *block_group;
5369 struct extent_map *em;
5370 u64 start = ctl->start;
5371 u64 type = ctl->type;
5376 map = kmalloc(map_lookup_size(ctl->num_stripes), GFP_NOFS);
5378 return ERR_PTR(-ENOMEM);
5379 map->num_stripes = ctl->num_stripes;
5381 for (i = 0; i < ctl->ndevs; ++i) {
5382 for (j = 0; j < ctl->dev_stripes; ++j) {
5383 int s = i * ctl->dev_stripes + j;
5384 map->stripes[s].dev = devices_info[i].dev;
5385 map->stripes[s].physical = devices_info[i].dev_offset +
5386 j * ctl->stripe_size;
5389 map->stripe_len = BTRFS_STRIPE_LEN;
5390 map->io_align = BTRFS_STRIPE_LEN;
5391 map->io_width = BTRFS_STRIPE_LEN;
5393 map->sub_stripes = ctl->sub_stripes;
5395 trace_btrfs_chunk_alloc(info, map, start, ctl->chunk_size);
5397 em = alloc_extent_map();
5400 return ERR_PTR(-ENOMEM);
5402 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
5403 em->map_lookup = map;
5405 em->len = ctl->chunk_size;
5406 em->block_start = 0;
5407 em->block_len = em->len;
5408 em->orig_block_len = ctl->stripe_size;
5410 em_tree = &info->mapping_tree;
5411 write_lock(&em_tree->lock);
5412 ret = add_extent_mapping(em_tree, em, 0);
5414 write_unlock(&em_tree->lock);
5415 free_extent_map(em);
5416 return ERR_PTR(ret);
5418 write_unlock(&em_tree->lock);
5420 block_group = btrfs_make_block_group(trans, 0, type, start, ctl->chunk_size);
5421 if (IS_ERR(block_group))
5422 goto error_del_extent;
5424 for (i = 0; i < map->num_stripes; i++) {
5425 struct btrfs_device *dev = map->stripes[i].dev;
5427 btrfs_device_set_bytes_used(dev,
5428 dev->bytes_used + ctl->stripe_size);
5429 if (list_empty(&dev->post_commit_list))
5430 list_add_tail(&dev->post_commit_list,
5431 &trans->transaction->dev_update_list);
5434 atomic64_sub(ctl->stripe_size * map->num_stripes,
5435 &info->free_chunk_space);
5437 free_extent_map(em);
5438 check_raid56_incompat_flag(info, type);
5439 check_raid1c34_incompat_flag(info, type);
5444 write_lock(&em_tree->lock);
5445 remove_extent_mapping(em_tree, em);
5446 write_unlock(&em_tree->lock);
5448 /* One for our allocation */
5449 free_extent_map(em);
5450 /* One for the tree reference */
5451 free_extent_map(em);
5456 struct btrfs_block_group *btrfs_create_chunk(struct btrfs_trans_handle *trans,
5459 struct btrfs_fs_info *info = trans->fs_info;
5460 struct btrfs_fs_devices *fs_devices = info->fs_devices;
5461 struct btrfs_device_info *devices_info = NULL;
5462 struct alloc_chunk_ctl ctl;
5463 struct btrfs_block_group *block_group;
5466 lockdep_assert_held(&info->chunk_mutex);
5468 if (!alloc_profile_is_valid(type, 0)) {
5470 return ERR_PTR(-EINVAL);
5473 if (list_empty(&fs_devices->alloc_list)) {
5474 if (btrfs_test_opt(info, ENOSPC_DEBUG))
5475 btrfs_debug(info, "%s: no writable device", __func__);
5476 return ERR_PTR(-ENOSPC);
5479 if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
5480 btrfs_err(info, "invalid chunk type 0x%llx requested", type);
5482 return ERR_PTR(-EINVAL);
5485 ctl.start = find_next_chunk(info);
5487 init_alloc_chunk_ctl(fs_devices, &ctl);
5489 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
5492 return ERR_PTR(-ENOMEM);
5494 ret = gather_device_info(fs_devices, &ctl, devices_info);
5496 block_group = ERR_PTR(ret);
5500 ret = decide_stripe_size(fs_devices, &ctl, devices_info);
5502 block_group = ERR_PTR(ret);
5506 block_group = create_chunk(trans, &ctl, devices_info);
5509 kfree(devices_info);
5514 * This function, btrfs_chunk_alloc_add_chunk_item(), typically belongs to the
5515 * phase 1 of chunk allocation. It belongs to phase 2 only when allocating system
5518 * See the comment at btrfs_chunk_alloc() for details about the chunk allocation
5521 int btrfs_chunk_alloc_add_chunk_item(struct btrfs_trans_handle *trans,
5522 struct btrfs_block_group *bg)
5524 struct btrfs_fs_info *fs_info = trans->fs_info;
5525 struct btrfs_root *extent_root = fs_info->extent_root;
5526 struct btrfs_root *chunk_root = fs_info->chunk_root;
5527 struct btrfs_key key;
5528 struct btrfs_chunk *chunk;
5529 struct btrfs_stripe *stripe;
5530 struct extent_map *em;
5531 struct map_lookup *map;
5537 * We take the chunk_mutex for 2 reasons:
5539 * 1) Updates and insertions in the chunk btree must be done while holding
5540 * the chunk_mutex, as well as updating the system chunk array in the
5541 * superblock. See the comment on top of btrfs_chunk_alloc() for the
5544 * 2) To prevent races with the final phase of a device replace operation
5545 * that replaces the device object associated with the map's stripes,
5546 * because the device object's id can change at any time during that
5547 * final phase of the device replace operation
5548 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the
5549 * replaced device and then see it with an ID of BTRFS_DEV_REPLACE_DEVID,
5550 * which would cause a failure when updating the device item, which does
5551 * not exists, or persisting a stripe of the chunk item with such ID.
5552 * Here we can't use the device_list_mutex because our caller already
5553 * has locked the chunk_mutex, and the final phase of device replace
5554 * acquires both mutexes - first the device_list_mutex and then the
5555 * chunk_mutex. Using any of those two mutexes protects us from a
5556 * concurrent device replace.
5558 lockdep_assert_held(&fs_info->chunk_mutex);
5560 em = btrfs_get_chunk_map(fs_info, bg->start, bg->length);
5563 btrfs_abort_transaction(trans, ret);
5567 map = em->map_lookup;
5568 item_size = btrfs_chunk_item_size(map->num_stripes);
5570 chunk = kzalloc(item_size, GFP_NOFS);
5573 btrfs_abort_transaction(trans, ret);
5577 for (i = 0; i < map->num_stripes; i++) {
5578 struct btrfs_device *device = map->stripes[i].dev;
5580 ret = btrfs_update_device(trans, device);
5585 stripe = &chunk->stripe;
5586 for (i = 0; i < map->num_stripes; i++) {
5587 struct btrfs_device *device = map->stripes[i].dev;
5588 const u64 dev_offset = map->stripes[i].physical;
5590 btrfs_set_stack_stripe_devid(stripe, device->devid);
5591 btrfs_set_stack_stripe_offset(stripe, dev_offset);
5592 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
5596 btrfs_set_stack_chunk_length(chunk, bg->length);
5597 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
5598 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
5599 btrfs_set_stack_chunk_type(chunk, map->type);
5600 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
5601 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
5602 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
5603 btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
5604 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
5606 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
5607 key.type = BTRFS_CHUNK_ITEM_KEY;
5608 key.offset = bg->start;
5610 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
5614 bg->chunk_item_inserted = 1;
5616 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
5617 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
5624 free_extent_map(em);
5628 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans)
5630 struct btrfs_fs_info *fs_info = trans->fs_info;
5632 struct btrfs_block_group *meta_bg;
5633 struct btrfs_block_group *sys_bg;
5636 * When adding a new device for sprouting, the seed device is read-only
5637 * so we must first allocate a metadata and a system chunk. But before
5638 * adding the block group items to the extent, device and chunk btrees,
5641 * 1) Create both chunks without doing any changes to the btrees, as
5642 * otherwise we would get -ENOSPC since the block groups from the
5643 * seed device are read-only;
5645 * 2) Add the device item for the new sprout device - finishing the setup
5646 * of a new block group requires updating the device item in the chunk
5647 * btree, so it must exist when we attempt to do it. The previous step
5648 * ensures this does not fail with -ENOSPC.
5650 * After that we can add the block group items to their btrees:
5651 * update existing device item in the chunk btree, add a new block group
5652 * item to the extent btree, add a new chunk item to the chunk btree and
5653 * finally add the new device extent items to the devices btree.
5656 alloc_profile = btrfs_metadata_alloc_profile(fs_info);
5657 meta_bg = btrfs_create_chunk(trans, alloc_profile);
5658 if (IS_ERR(meta_bg))
5659 return PTR_ERR(meta_bg);
5661 alloc_profile = btrfs_system_alloc_profile(fs_info);
5662 sys_bg = btrfs_create_chunk(trans, alloc_profile);
5664 return PTR_ERR(sys_bg);
5669 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
5671 const int index = btrfs_bg_flags_to_raid_index(map->type);
5673 return btrfs_raid_array[index].tolerated_failures;
5676 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset)
5678 struct extent_map *em;
5679 struct map_lookup *map;
5684 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
5688 map = em->map_lookup;
5689 for (i = 0; i < map->num_stripes; i++) {
5690 if (test_bit(BTRFS_DEV_STATE_MISSING,
5691 &map->stripes[i].dev->dev_state)) {
5695 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE,
5696 &map->stripes[i].dev->dev_state)) {
5703 * If the number of missing devices is larger than max errors,
5704 * we can not write the data into that chunk successfully, so
5707 if (miss_ndevs > btrfs_chunk_max_errors(map))
5710 free_extent_map(em);
5714 void btrfs_mapping_tree_free(struct extent_map_tree *tree)
5716 struct extent_map *em;
5719 write_lock(&tree->lock);
5720 em = lookup_extent_mapping(tree, 0, (u64)-1);
5722 remove_extent_mapping(tree, em);
5723 write_unlock(&tree->lock);
5727 free_extent_map(em);
5728 /* once for the tree */
5729 free_extent_map(em);
5733 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5735 struct extent_map *em;
5736 struct map_lookup *map;
5739 em = btrfs_get_chunk_map(fs_info, logical, len);
5742 * We could return errors for these cases, but that could get
5743 * ugly and we'd probably do the same thing which is just not do
5744 * anything else and exit, so return 1 so the callers don't try
5745 * to use other copies.
5749 map = em->map_lookup;
5750 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1_MASK))
5751 ret = map->num_stripes;
5752 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5753 ret = map->sub_stripes;
5754 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5756 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5758 * There could be two corrupted data stripes, we need
5759 * to loop retry in order to rebuild the correct data.
5761 * Fail a stripe at a time on every retry except the
5762 * stripe under reconstruction.
5764 ret = map->num_stripes;
5767 free_extent_map(em);
5769 down_read(&fs_info->dev_replace.rwsem);
5770 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) &&
5771 fs_info->dev_replace.tgtdev)
5773 up_read(&fs_info->dev_replace.rwsem);
5778 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
5781 struct extent_map *em;
5782 struct map_lookup *map;
5783 unsigned long len = fs_info->sectorsize;
5785 em = btrfs_get_chunk_map(fs_info, logical, len);
5787 if (!WARN_ON(IS_ERR(em))) {
5788 map = em->map_lookup;
5789 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5790 len = map->stripe_len * nr_data_stripes(map);
5791 free_extent_map(em);
5796 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5798 struct extent_map *em;
5799 struct map_lookup *map;
5802 em = btrfs_get_chunk_map(fs_info, logical, len);
5804 if(!WARN_ON(IS_ERR(em))) {
5805 map = em->map_lookup;
5806 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5808 free_extent_map(em);
5813 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5814 struct map_lookup *map, int first,
5815 int dev_replace_is_ongoing)
5819 int preferred_mirror;
5821 struct btrfs_device *srcdev;
5824 (BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10)));
5826 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5827 num_stripes = map->sub_stripes;
5829 num_stripes = map->num_stripes;
5831 switch (fs_info->fs_devices->read_policy) {
5833 /* Shouldn't happen, just warn and use pid instead of failing */
5834 btrfs_warn_rl(fs_info,
5835 "unknown read_policy type %u, reset to pid",
5836 fs_info->fs_devices->read_policy);
5837 fs_info->fs_devices->read_policy = BTRFS_READ_POLICY_PID;
5839 case BTRFS_READ_POLICY_PID:
5840 preferred_mirror = first + (current->pid % num_stripes);
5844 if (dev_replace_is_ongoing &&
5845 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5846 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5847 srcdev = fs_info->dev_replace.srcdev;
5852 * try to avoid the drive that is the source drive for a
5853 * dev-replace procedure, only choose it if no other non-missing
5854 * mirror is available
5856 for (tolerance = 0; tolerance < 2; tolerance++) {
5857 if (map->stripes[preferred_mirror].dev->bdev &&
5858 (tolerance || map->stripes[preferred_mirror].dev != srcdev))
5859 return preferred_mirror;
5860 for (i = first; i < first + num_stripes; i++) {
5861 if (map->stripes[i].dev->bdev &&
5862 (tolerance || map->stripes[i].dev != srcdev))
5867 /* we couldn't find one that doesn't fail. Just return something
5868 * and the io error handling code will clean up eventually
5870 return preferred_mirror;
5873 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5874 static void sort_parity_stripes(struct btrfs_io_context *bioc, int num_stripes)
5881 for (i = 0; i < num_stripes - 1; i++) {
5882 /* Swap if parity is on a smaller index */
5883 if (bioc->raid_map[i] > bioc->raid_map[i + 1]) {
5884 swap(bioc->stripes[i], bioc->stripes[i + 1]);
5885 swap(bioc->raid_map[i], bioc->raid_map[i + 1]);
5892 static struct btrfs_io_context *alloc_btrfs_io_context(int total_stripes,
5895 struct btrfs_io_context *bioc = kzalloc(
5896 /* The size of btrfs_io_context */
5897 sizeof(struct btrfs_io_context) +
5898 /* Plus the variable array for the stripes */
5899 sizeof(struct btrfs_io_stripe) * (total_stripes) +
5900 /* Plus the variable array for the tgt dev */
5901 sizeof(int) * (real_stripes) +
5903 * Plus the raid_map, which includes both the tgt dev
5906 sizeof(u64) * (total_stripes),
5907 GFP_NOFS|__GFP_NOFAIL);
5909 atomic_set(&bioc->error, 0);
5910 refcount_set(&bioc->refs, 1);
5912 bioc->tgtdev_map = (int *)(bioc->stripes + total_stripes);
5913 bioc->raid_map = (u64 *)(bioc->tgtdev_map + real_stripes);
5918 void btrfs_get_bioc(struct btrfs_io_context *bioc)
5920 WARN_ON(!refcount_read(&bioc->refs));
5921 refcount_inc(&bioc->refs);
5924 void btrfs_put_bioc(struct btrfs_io_context *bioc)
5928 if (refcount_dec_and_test(&bioc->refs))
5932 /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */
5934 * Please note that, discard won't be sent to target device of device
5937 static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info,
5938 u64 logical, u64 *length_ret,
5939 struct btrfs_io_context **bioc_ret)
5941 struct extent_map *em;
5942 struct map_lookup *map;
5943 struct btrfs_io_context *bioc;
5944 u64 length = *length_ret;
5948 u64 stripe_end_offset;
5955 u32 sub_stripes = 0;
5956 u64 stripes_per_dev = 0;
5957 u32 remaining_stripes = 0;
5958 u32 last_stripe = 0;
5962 /* Discard always returns a bioc. */
5965 em = btrfs_get_chunk_map(fs_info, logical, length);
5969 map = em->map_lookup;
5970 /* we don't discard raid56 yet */
5971 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5976 offset = logical - em->start;
5977 length = min_t(u64, em->start + em->len - logical, length);
5978 *length_ret = length;
5980 stripe_len = map->stripe_len;
5982 * stripe_nr counts the total number of stripes we have to stride
5983 * to get to this block
5985 stripe_nr = div64_u64(offset, stripe_len);
5987 /* stripe_offset is the offset of this block in its stripe */
5988 stripe_offset = offset - stripe_nr * stripe_len;
5990 stripe_nr_end = round_up(offset + length, map->stripe_len);
5991 stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len);
5992 stripe_cnt = stripe_nr_end - stripe_nr;
5993 stripe_end_offset = stripe_nr_end * map->stripe_len -
5996 * after this, stripe_nr is the number of stripes on this
5997 * device we have to walk to find the data, and stripe_index is
5998 * the number of our device in the stripe array
6002 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
6003 BTRFS_BLOCK_GROUP_RAID10)) {
6004 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
6007 sub_stripes = map->sub_stripes;
6009 factor = map->num_stripes / sub_stripes;
6010 num_stripes = min_t(u64, map->num_stripes,
6011 sub_stripes * stripe_cnt);
6012 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
6013 stripe_index *= sub_stripes;
6014 stripes_per_dev = div_u64_rem(stripe_cnt, factor,
6015 &remaining_stripes);
6016 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
6017 last_stripe *= sub_stripes;
6018 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1_MASK |
6019 BTRFS_BLOCK_GROUP_DUP)) {
6020 num_stripes = map->num_stripes;
6022 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
6026 bioc = alloc_btrfs_io_context(num_stripes, 0);
6032 for (i = 0; i < num_stripes; i++) {
6033 bioc->stripes[i].physical =
6034 map->stripes[stripe_index].physical +
6035 stripe_offset + stripe_nr * map->stripe_len;
6036 bioc->stripes[i].dev = map->stripes[stripe_index].dev;
6038 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
6039 BTRFS_BLOCK_GROUP_RAID10)) {
6040 bioc->stripes[i].length = stripes_per_dev *
6043 if (i / sub_stripes < remaining_stripes)
6044 bioc->stripes[i].length += map->stripe_len;
6047 * Special for the first stripe and
6050 * |-------|...|-------|
6054 if (i < sub_stripes)
6055 bioc->stripes[i].length -= stripe_offset;
6057 if (stripe_index >= last_stripe &&
6058 stripe_index <= (last_stripe +
6060 bioc->stripes[i].length -= stripe_end_offset;
6062 if (i == sub_stripes - 1)
6065 bioc->stripes[i].length = length;
6069 if (stripe_index == map->num_stripes) {
6076 bioc->map_type = map->type;
6077 bioc->num_stripes = num_stripes;
6079 free_extent_map(em);
6084 * In dev-replace case, for repair case (that's the only case where the mirror
6085 * is selected explicitly when calling btrfs_map_block), blocks left of the
6086 * left cursor can also be read from the target drive.
6088 * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the
6090 * For READ, it also needs to be supported using the same mirror number.
6092 * If the requested block is not left of the left cursor, EIO is returned. This
6093 * can happen because btrfs_num_copies() returns one more in the dev-replace
6096 static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info,
6097 u64 logical, u64 length,
6098 u64 srcdev_devid, int *mirror_num,
6101 struct btrfs_io_context *bioc = NULL;
6103 int index_srcdev = 0;
6105 u64 physical_of_found = 0;
6109 ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS,
6110 logical, &length, &bioc, 0, 0);
6112 ASSERT(bioc == NULL);
6116 num_stripes = bioc->num_stripes;
6117 if (*mirror_num > num_stripes) {
6119 * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror,
6120 * that means that the requested area is not left of the left
6123 btrfs_put_bioc(bioc);
6128 * process the rest of the function using the mirror_num of the source
6129 * drive. Therefore look it up first. At the end, patch the device
6130 * pointer to the one of the target drive.
6132 for (i = 0; i < num_stripes; i++) {
6133 if (bioc->stripes[i].dev->devid != srcdev_devid)
6137 * In case of DUP, in order to keep it simple, only add the
6138 * mirror with the lowest physical address
6141 physical_of_found <= bioc->stripes[i].physical)
6146 physical_of_found = bioc->stripes[i].physical;
6149 btrfs_put_bioc(bioc);
6155 *mirror_num = index_srcdev + 1;
6156 *physical = physical_of_found;
6160 static bool is_block_group_to_copy(struct btrfs_fs_info *fs_info, u64 logical)
6162 struct btrfs_block_group *cache;
6165 /* Non zoned filesystem does not use "to_copy" flag */
6166 if (!btrfs_is_zoned(fs_info))
6169 cache = btrfs_lookup_block_group(fs_info, logical);
6171 spin_lock(&cache->lock);
6172 ret = cache->to_copy;
6173 spin_unlock(&cache->lock);
6175 btrfs_put_block_group(cache);
6179 static void handle_ops_on_dev_replace(enum btrfs_map_op op,
6180 struct btrfs_io_context **bioc_ret,
6181 struct btrfs_dev_replace *dev_replace,
6183 int *num_stripes_ret, int *max_errors_ret)
6185 struct btrfs_io_context *bioc = *bioc_ret;
6186 u64 srcdev_devid = dev_replace->srcdev->devid;
6187 int tgtdev_indexes = 0;
6188 int num_stripes = *num_stripes_ret;
6189 int max_errors = *max_errors_ret;
6192 if (op == BTRFS_MAP_WRITE) {
6193 int index_where_to_add;
6196 * A block group which have "to_copy" set will eventually
6197 * copied by dev-replace process. We can avoid cloning IO here.
6199 if (is_block_group_to_copy(dev_replace->srcdev->fs_info, logical))
6203 * duplicate the write operations while the dev replace
6204 * procedure is running. Since the copying of the old disk to
6205 * the new disk takes place at run time while the filesystem is
6206 * mounted writable, the regular write operations to the old
6207 * disk have to be duplicated to go to the new disk as well.
6209 * Note that device->missing is handled by the caller, and that
6210 * the write to the old disk is already set up in the stripes
6213 index_where_to_add = num_stripes;
6214 for (i = 0; i < num_stripes; i++) {
6215 if (bioc->stripes[i].dev->devid == srcdev_devid) {
6216 /* write to new disk, too */
6217 struct btrfs_io_stripe *new =
6218 bioc->stripes + index_where_to_add;
6219 struct btrfs_io_stripe *old =
6222 new->physical = old->physical;
6223 new->length = old->length;
6224 new->dev = dev_replace->tgtdev;
6225 bioc->tgtdev_map[i] = index_where_to_add;
6226 index_where_to_add++;
6231 num_stripes = index_where_to_add;
6232 } else if (op == BTRFS_MAP_GET_READ_MIRRORS) {
6233 int index_srcdev = 0;
6235 u64 physical_of_found = 0;
6238 * During the dev-replace procedure, the target drive can also
6239 * be used to read data in case it is needed to repair a corrupt
6240 * block elsewhere. This is possible if the requested area is
6241 * left of the left cursor. In this area, the target drive is a
6242 * full copy of the source drive.
6244 for (i = 0; i < num_stripes; i++) {
6245 if (bioc->stripes[i].dev->devid == srcdev_devid) {
6247 * In case of DUP, in order to keep it simple,
6248 * only add the mirror with the lowest physical
6252 physical_of_found <= bioc->stripes[i].physical)
6256 physical_of_found = bioc->stripes[i].physical;
6260 struct btrfs_io_stripe *tgtdev_stripe =
6261 bioc->stripes + num_stripes;
6263 tgtdev_stripe->physical = physical_of_found;
6264 tgtdev_stripe->length =
6265 bioc->stripes[index_srcdev].length;
6266 tgtdev_stripe->dev = dev_replace->tgtdev;
6267 bioc->tgtdev_map[index_srcdev] = num_stripes;
6274 *num_stripes_ret = num_stripes;
6275 *max_errors_ret = max_errors;
6276 bioc->num_tgtdevs = tgtdev_indexes;
6280 static bool need_full_stripe(enum btrfs_map_op op)
6282 return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS);
6286 * Calculate the geometry of a particular (address, len) tuple. This
6287 * information is used to calculate how big a particular bio can get before it
6288 * straddles a stripe.
6290 * @fs_info: the filesystem
6291 * @em: mapping containing the logical extent
6292 * @op: type of operation - write or read
6293 * @logical: address that we want to figure out the geometry of
6294 * @io_geom: pointer used to return values
6296 * Returns < 0 in case a chunk for the given logical address cannot be found,
6297 * usually shouldn't happen unless @logical is corrupted, 0 otherwise.
6299 int btrfs_get_io_geometry(struct btrfs_fs_info *fs_info, struct extent_map *em,
6300 enum btrfs_map_op op, u64 logical,
6301 struct btrfs_io_geometry *io_geom)
6303 struct map_lookup *map;
6309 u64 raid56_full_stripe_start = (u64)-1;
6312 ASSERT(op != BTRFS_MAP_DISCARD);
6314 map = em->map_lookup;
6315 /* Offset of this logical address in the chunk */
6316 offset = logical - em->start;
6317 /* Len of a stripe in a chunk */
6318 stripe_len = map->stripe_len;
6319 /* Stripe where this block falls in */
6320 stripe_nr = div64_u64(offset, stripe_len);
6321 /* Offset of stripe in the chunk */
6322 stripe_offset = stripe_nr * stripe_len;
6323 if (offset < stripe_offset) {
6325 "stripe math has gone wrong, stripe_offset=%llu offset=%llu start=%llu logical=%llu stripe_len=%llu",
6326 stripe_offset, offset, em->start, logical, stripe_len);
6330 /* stripe_offset is the offset of this block in its stripe */
6331 stripe_offset = offset - stripe_offset;
6332 data_stripes = nr_data_stripes(map);
6334 if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
6335 u64 max_len = stripe_len - stripe_offset;
6338 * In case of raid56, we need to know the stripe aligned start
6340 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6341 unsigned long full_stripe_len = stripe_len * data_stripes;
6342 raid56_full_stripe_start = offset;
6345 * Allow a write of a full stripe, but make sure we
6346 * don't allow straddling of stripes
6348 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
6350 raid56_full_stripe_start *= full_stripe_len;
6353 * For writes to RAID[56], allow a full stripeset across
6354 * all disks. For other RAID types and for RAID[56]
6355 * reads, just allow a single stripe (on a single disk).
6357 if (op == BTRFS_MAP_WRITE) {
6358 max_len = stripe_len * data_stripes -
6359 (offset - raid56_full_stripe_start);
6362 len = min_t(u64, em->len - offset, max_len);
6364 len = em->len - offset;
6368 io_geom->offset = offset;
6369 io_geom->stripe_len = stripe_len;
6370 io_geom->stripe_nr = stripe_nr;
6371 io_geom->stripe_offset = stripe_offset;
6372 io_geom->raid56_stripe_offset = raid56_full_stripe_start;
6377 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
6378 enum btrfs_map_op op,
6379 u64 logical, u64 *length,
6380 struct btrfs_io_context **bioc_ret,
6381 int mirror_num, int need_raid_map)
6383 struct extent_map *em;
6384 struct map_lookup *map;
6394 int tgtdev_indexes = 0;
6395 struct btrfs_io_context *bioc = NULL;
6396 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
6397 int dev_replace_is_ongoing = 0;
6398 int num_alloc_stripes;
6399 int patch_the_first_stripe_for_dev_replace = 0;
6400 u64 physical_to_patch_in_first_stripe = 0;
6401 u64 raid56_full_stripe_start = (u64)-1;
6402 struct btrfs_io_geometry geom;
6405 ASSERT(op != BTRFS_MAP_DISCARD);
6407 em = btrfs_get_chunk_map(fs_info, logical, *length);
6408 ASSERT(!IS_ERR(em));
6410 ret = btrfs_get_io_geometry(fs_info, em, op, logical, &geom);
6414 map = em->map_lookup;
6417 stripe_len = geom.stripe_len;
6418 stripe_nr = geom.stripe_nr;
6419 stripe_offset = geom.stripe_offset;
6420 raid56_full_stripe_start = geom.raid56_stripe_offset;
6421 data_stripes = nr_data_stripes(map);
6423 down_read(&dev_replace->rwsem);
6424 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
6426 * Hold the semaphore for read during the whole operation, write is
6427 * requested at commit time but must wait.
6429 if (!dev_replace_is_ongoing)
6430 up_read(&dev_replace->rwsem);
6432 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
6433 !need_full_stripe(op) && dev_replace->tgtdev != NULL) {
6434 ret = get_extra_mirror_from_replace(fs_info, logical, *length,
6435 dev_replace->srcdev->devid,
6437 &physical_to_patch_in_first_stripe);
6441 patch_the_first_stripe_for_dev_replace = 1;
6442 } else if (mirror_num > map->num_stripes) {
6448 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
6449 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
6451 if (!need_full_stripe(op))
6453 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1_MASK) {
6454 if (need_full_stripe(op))
6455 num_stripes = map->num_stripes;
6456 else if (mirror_num)
6457 stripe_index = mirror_num - 1;
6459 stripe_index = find_live_mirror(fs_info, map, 0,
6460 dev_replace_is_ongoing);
6461 mirror_num = stripe_index + 1;
6464 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
6465 if (need_full_stripe(op)) {
6466 num_stripes = map->num_stripes;
6467 } else if (mirror_num) {
6468 stripe_index = mirror_num - 1;
6473 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
6474 u32 factor = map->num_stripes / map->sub_stripes;
6476 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
6477 stripe_index *= map->sub_stripes;
6479 if (need_full_stripe(op))
6480 num_stripes = map->sub_stripes;
6481 else if (mirror_num)
6482 stripe_index += mirror_num - 1;
6484 int old_stripe_index = stripe_index;
6485 stripe_index = find_live_mirror(fs_info, map,
6487 dev_replace_is_ongoing);
6488 mirror_num = stripe_index - old_stripe_index + 1;
6491 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6492 if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) {
6493 /* push stripe_nr back to the start of the full stripe */
6494 stripe_nr = div64_u64(raid56_full_stripe_start,
6495 stripe_len * data_stripes);
6497 /* RAID[56] write or recovery. Return all stripes */
6498 num_stripes = map->num_stripes;
6499 max_errors = nr_parity_stripes(map);
6501 *length = map->stripe_len;
6506 * Mirror #0 or #1 means the original data block.
6507 * Mirror #2 is RAID5 parity block.
6508 * Mirror #3 is RAID6 Q block.
6510 stripe_nr = div_u64_rem(stripe_nr,
6511 data_stripes, &stripe_index);
6513 stripe_index = data_stripes + mirror_num - 2;
6515 /* We distribute the parity blocks across stripes */
6516 div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
6518 if (!need_full_stripe(op) && mirror_num <= 1)
6523 * after this, stripe_nr is the number of stripes on this
6524 * device we have to walk to find the data, and stripe_index is
6525 * the number of our device in the stripe array
6527 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
6529 mirror_num = stripe_index + 1;
6531 if (stripe_index >= map->num_stripes) {
6533 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
6534 stripe_index, map->num_stripes);
6539 num_alloc_stripes = num_stripes;
6540 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) {
6541 if (op == BTRFS_MAP_WRITE)
6542 num_alloc_stripes <<= 1;
6543 if (op == BTRFS_MAP_GET_READ_MIRRORS)
6544 num_alloc_stripes++;
6545 tgtdev_indexes = num_stripes;
6548 bioc = alloc_btrfs_io_context(num_alloc_stripes, tgtdev_indexes);
6554 for (i = 0; i < num_stripes; i++) {
6555 bioc->stripes[i].physical = map->stripes[stripe_index].physical +
6556 stripe_offset + stripe_nr * map->stripe_len;
6557 bioc->stripes[i].dev = map->stripes[stripe_index].dev;
6561 /* Build raid_map */
6562 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map &&
6563 (need_full_stripe(op) || mirror_num > 1)) {
6567 /* Work out the disk rotation on this stripe-set */
6568 div_u64_rem(stripe_nr, num_stripes, &rot);
6570 /* Fill in the logical address of each stripe */
6571 tmp = stripe_nr * data_stripes;
6572 for (i = 0; i < data_stripes; i++)
6573 bioc->raid_map[(i + rot) % num_stripes] =
6574 em->start + (tmp + i) * map->stripe_len;
6576 bioc->raid_map[(i + rot) % map->num_stripes] = RAID5_P_STRIPE;
6577 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
6578 bioc->raid_map[(i + rot + 1) % num_stripes] =
6581 sort_parity_stripes(bioc, num_stripes);
6584 if (need_full_stripe(op))
6585 max_errors = btrfs_chunk_max_errors(map);
6587 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
6588 need_full_stripe(op)) {
6589 handle_ops_on_dev_replace(op, &bioc, dev_replace, logical,
6590 &num_stripes, &max_errors);
6594 bioc->map_type = map->type;
6595 bioc->num_stripes = num_stripes;
6596 bioc->max_errors = max_errors;
6597 bioc->mirror_num = mirror_num;
6600 * this is the case that REQ_READ && dev_replace_is_ongoing &&
6601 * mirror_num == num_stripes + 1 && dev_replace target drive is
6602 * available as a mirror
6604 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
6605 WARN_ON(num_stripes > 1);
6606 bioc->stripes[0].dev = dev_replace->tgtdev;
6607 bioc->stripes[0].physical = physical_to_patch_in_first_stripe;
6608 bioc->mirror_num = map->num_stripes + 1;
6611 if (dev_replace_is_ongoing) {
6612 lockdep_assert_held(&dev_replace->rwsem);
6613 /* Unlock and let waiting writers proceed */
6614 up_read(&dev_replace->rwsem);
6616 free_extent_map(em);
6620 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6621 u64 logical, u64 *length,
6622 struct btrfs_io_context **bioc_ret, int mirror_num)
6624 if (op == BTRFS_MAP_DISCARD)
6625 return __btrfs_map_block_for_discard(fs_info, logical,
6628 return __btrfs_map_block(fs_info, op, logical, length, bioc_ret,
6632 /* For Scrub/replace */
6633 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6634 u64 logical, u64 *length,
6635 struct btrfs_io_context **bioc_ret)
6637 return __btrfs_map_block(fs_info, op, logical, length, bioc_ret, 0, 1);
6640 static inline void btrfs_end_bioc(struct btrfs_io_context *bioc, struct bio *bio)
6642 bio->bi_private = bioc->private;
6643 bio->bi_end_io = bioc->end_io;
6646 btrfs_put_bioc(bioc);
6649 static void btrfs_end_bio(struct bio *bio)
6651 struct btrfs_io_context *bioc = bio->bi_private;
6652 int is_orig_bio = 0;
6654 if (bio->bi_status) {
6655 atomic_inc(&bioc->error);
6656 if (bio->bi_status == BLK_STS_IOERR ||
6657 bio->bi_status == BLK_STS_TARGET) {
6658 struct btrfs_device *dev = btrfs_io_bio(bio)->device;
6661 if (btrfs_op(bio) == BTRFS_MAP_WRITE)
6662 btrfs_dev_stat_inc_and_print(dev,
6663 BTRFS_DEV_STAT_WRITE_ERRS);
6664 else if (!(bio->bi_opf & REQ_RAHEAD))
6665 btrfs_dev_stat_inc_and_print(dev,
6666 BTRFS_DEV_STAT_READ_ERRS);
6667 if (bio->bi_opf & REQ_PREFLUSH)
6668 btrfs_dev_stat_inc_and_print(dev,
6669 BTRFS_DEV_STAT_FLUSH_ERRS);
6673 if (bio == bioc->orig_bio)
6676 btrfs_bio_counter_dec(bioc->fs_info);
6678 if (atomic_dec_and_test(&bioc->stripes_pending)) {
6681 bio = bioc->orig_bio;
6684 btrfs_io_bio(bio)->mirror_num = bioc->mirror_num;
6685 /* only send an error to the higher layers if it is
6686 * beyond the tolerance of the btrfs bio
6688 if (atomic_read(&bioc->error) > bioc->max_errors) {
6689 bio->bi_status = BLK_STS_IOERR;
6692 * this bio is actually up to date, we didn't
6693 * go over the max number of errors
6695 bio->bi_status = BLK_STS_OK;
6698 btrfs_end_bioc(bioc, bio);
6699 } else if (!is_orig_bio) {
6704 static void submit_stripe_bio(struct btrfs_io_context *bioc, struct bio *bio,
6705 u64 physical, struct btrfs_device *dev)
6707 struct btrfs_fs_info *fs_info = bioc->fs_info;
6709 bio->bi_private = bioc;
6710 btrfs_io_bio(bio)->device = dev;
6711 bio->bi_end_io = btrfs_end_bio;
6712 bio->bi_iter.bi_sector = physical >> 9;
6714 * For zone append writing, bi_sector must point the beginning of the
6717 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
6718 if (btrfs_dev_is_sequential(dev, physical)) {
6719 u64 zone_start = round_down(physical, fs_info->zone_size);
6721 bio->bi_iter.bi_sector = zone_start >> SECTOR_SHIFT;
6723 bio->bi_opf &= ~REQ_OP_ZONE_APPEND;
6724 bio->bi_opf |= REQ_OP_WRITE;
6727 btrfs_debug_in_rcu(fs_info,
6728 "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
6729 bio_op(bio), bio->bi_opf, bio->bi_iter.bi_sector,
6730 (unsigned long)dev->bdev->bd_dev, rcu_str_deref(dev->name),
6731 dev->devid, bio->bi_iter.bi_size);
6732 bio_set_dev(bio, dev->bdev);
6734 btrfs_bio_counter_inc_noblocked(fs_info);
6736 btrfsic_submit_bio(bio);
6739 static void bioc_error(struct btrfs_io_context *bioc, struct bio *bio, u64 logical)
6741 atomic_inc(&bioc->error);
6742 if (atomic_dec_and_test(&bioc->stripes_pending)) {
6743 /* Should be the original bio. */
6744 WARN_ON(bio != bioc->orig_bio);
6746 btrfs_io_bio(bio)->mirror_num = bioc->mirror_num;
6747 bio->bi_iter.bi_sector = logical >> 9;
6748 if (atomic_read(&bioc->error) > bioc->max_errors)
6749 bio->bi_status = BLK_STS_IOERR;
6751 bio->bi_status = BLK_STS_OK;
6752 btrfs_end_bioc(bioc, bio);
6756 blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
6759 struct btrfs_device *dev;
6760 struct bio *first_bio = bio;
6761 u64 logical = bio->bi_iter.bi_sector << 9;
6767 struct btrfs_io_context *bioc = NULL;
6769 length = bio->bi_iter.bi_size;
6770 map_length = length;
6772 btrfs_bio_counter_inc_blocked(fs_info);
6773 ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical,
6774 &map_length, &bioc, mirror_num, 1);
6776 btrfs_bio_counter_dec(fs_info);
6777 return errno_to_blk_status(ret);
6780 total_devs = bioc->num_stripes;
6781 bioc->orig_bio = first_bio;
6782 bioc->private = first_bio->bi_private;
6783 bioc->end_io = first_bio->bi_end_io;
6784 bioc->fs_info = fs_info;
6785 atomic_set(&bioc->stripes_pending, bioc->num_stripes);
6787 if ((bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6788 ((btrfs_op(bio) == BTRFS_MAP_WRITE) || (mirror_num > 1))) {
6789 /* In this case, map_length has been set to the length of
6790 a single stripe; not the whole write */
6791 if (btrfs_op(bio) == BTRFS_MAP_WRITE) {
6792 ret = raid56_parity_write(fs_info, bio, bioc,
6795 ret = raid56_parity_recover(fs_info, bio, bioc,
6796 map_length, mirror_num, 1);
6799 btrfs_bio_counter_dec(fs_info);
6800 return errno_to_blk_status(ret);
6803 if (map_length < length) {
6805 "mapping failed logical %llu bio len %llu len %llu",
6806 logical, length, map_length);
6810 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6811 dev = bioc->stripes[dev_nr].dev;
6812 if (!dev || !dev->bdev || test_bit(BTRFS_DEV_STATE_MISSING,
6814 (btrfs_op(first_bio) == BTRFS_MAP_WRITE &&
6815 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) {
6816 bioc_error(bioc, first_bio, logical);
6820 if (dev_nr < total_devs - 1)
6821 bio = btrfs_bio_clone(first_bio);
6825 submit_stripe_bio(bioc, bio, bioc->stripes[dev_nr].physical, dev);
6827 btrfs_bio_counter_dec(fs_info);
6831 static bool dev_args_match_fs_devices(const struct btrfs_dev_lookup_args *args,
6832 const struct btrfs_fs_devices *fs_devices)
6834 if (args->fsid == NULL)
6836 if (memcmp(fs_devices->metadata_uuid, args->fsid, BTRFS_FSID_SIZE) == 0)
6841 static bool dev_args_match_device(const struct btrfs_dev_lookup_args *args,
6842 const struct btrfs_device *device)
6844 if (args->missing) {
6845 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state) &&
6851 if (device->devid != args->devid)
6853 if (args->uuid && memcmp(device->uuid, args->uuid, BTRFS_UUID_SIZE) != 0)
6859 * Find a device specified by @devid or @uuid in the list of @fs_devices, or
6862 * If devid and uuid are both specified, the match must be exact, otherwise
6863 * only devid is used.
6865 struct btrfs_device *btrfs_find_device(const struct btrfs_fs_devices *fs_devices,
6866 const struct btrfs_dev_lookup_args *args)
6868 struct btrfs_device *device;
6869 struct btrfs_fs_devices *seed_devs;
6871 if (dev_args_match_fs_devices(args, fs_devices)) {
6872 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6873 if (dev_args_match_device(args, device))
6878 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
6879 if (!dev_args_match_fs_devices(args, seed_devs))
6881 list_for_each_entry(device, &seed_devs->devices, dev_list) {
6882 if (dev_args_match_device(args, device))
6890 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices,
6891 u64 devid, u8 *dev_uuid)
6893 struct btrfs_device *device;
6894 unsigned int nofs_flag;
6897 * We call this under the chunk_mutex, so we want to use NOFS for this
6898 * allocation, however we don't want to change btrfs_alloc_device() to
6899 * always do NOFS because we use it in a lot of other GFP_KERNEL safe
6902 nofs_flag = memalloc_nofs_save();
6903 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6904 memalloc_nofs_restore(nofs_flag);
6908 list_add(&device->dev_list, &fs_devices->devices);
6909 device->fs_devices = fs_devices;
6910 fs_devices->num_devices++;
6912 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6913 fs_devices->missing_devices++;
6919 * btrfs_alloc_device - allocate struct btrfs_device
6920 * @fs_info: used only for generating a new devid, can be NULL if
6921 * devid is provided (i.e. @devid != NULL).
6922 * @devid: a pointer to devid for this device. If NULL a new devid
6924 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6927 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6928 * on error. Returned struct is not linked onto any lists and must be
6929 * destroyed with btrfs_free_device.
6931 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6935 struct btrfs_device *dev;
6938 if (WARN_ON(!devid && !fs_info))
6939 return ERR_PTR(-EINVAL);
6941 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
6943 return ERR_PTR(-ENOMEM);
6946 * Preallocate a bio that's always going to be used for flushing device
6947 * barriers and matches the device lifespan
6949 dev->flush_bio = bio_kmalloc(GFP_KERNEL, 0);
6950 if (!dev->flush_bio) {
6952 return ERR_PTR(-ENOMEM);
6955 INIT_LIST_HEAD(&dev->dev_list);
6956 INIT_LIST_HEAD(&dev->dev_alloc_list);
6957 INIT_LIST_HEAD(&dev->post_commit_list);
6959 atomic_set(&dev->reada_in_flight, 0);
6960 atomic_set(&dev->dev_stats_ccnt, 0);
6961 btrfs_device_data_ordered_init(dev);
6962 INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
6963 INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
6964 extent_io_tree_init(fs_info, &dev->alloc_state,
6965 IO_TREE_DEVICE_ALLOC_STATE, NULL);
6972 ret = find_next_devid(fs_info, &tmp);
6974 btrfs_free_device(dev);
6975 return ERR_PTR(ret);
6981 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6983 generate_random_uuid(dev->uuid);
6988 static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info,
6989 u64 devid, u8 *uuid, bool error)
6992 btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing",
6995 btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing",
6999 static u64 calc_stripe_length(u64 type, u64 chunk_len, int num_stripes)
7001 const int data_stripes = calc_data_stripes(type, num_stripes);
7003 return div_u64(chunk_len, data_stripes);
7006 #if BITS_PER_LONG == 32
7008 * Due to page cache limit, metadata beyond BTRFS_32BIT_MAX_FILE_SIZE
7009 * can't be accessed on 32bit systems.
7011 * This function do mount time check to reject the fs if it already has
7012 * metadata chunk beyond that limit.
7014 static int check_32bit_meta_chunk(struct btrfs_fs_info *fs_info,
7015 u64 logical, u64 length, u64 type)
7017 if (!(type & BTRFS_BLOCK_GROUP_METADATA))
7020 if (logical + length < MAX_LFS_FILESIZE)
7023 btrfs_err_32bit_limit(fs_info);
7028 * This is to give early warning for any metadata chunk reaching
7029 * BTRFS_32BIT_EARLY_WARN_THRESHOLD.
7030 * Although we can still access the metadata, it's not going to be possible
7031 * once the limit is reached.
7033 static void warn_32bit_meta_chunk(struct btrfs_fs_info *fs_info,
7034 u64 logical, u64 length, u64 type)
7036 if (!(type & BTRFS_BLOCK_GROUP_METADATA))
7039 if (logical + length < BTRFS_32BIT_EARLY_WARN_THRESHOLD)
7042 btrfs_warn_32bit_limit(fs_info);
7046 static struct btrfs_device *handle_missing_device(struct btrfs_fs_info *fs_info,
7047 u64 devid, u8 *uuid)
7049 struct btrfs_device *dev;
7051 if (!btrfs_test_opt(fs_info, DEGRADED)) {
7052 btrfs_report_missing_device(fs_info, devid, uuid, true);
7053 return ERR_PTR(-ENOENT);
7056 dev = add_missing_dev(fs_info->fs_devices, devid, uuid);
7058 btrfs_err(fs_info, "failed to init missing device %llu: %ld",
7059 devid, PTR_ERR(dev));
7062 btrfs_report_missing_device(fs_info, devid, uuid, false);
7067 static int read_one_chunk(struct btrfs_key *key, struct extent_buffer *leaf,
7068 struct btrfs_chunk *chunk)
7070 BTRFS_DEV_LOOKUP_ARGS(args);
7071 struct btrfs_fs_info *fs_info = leaf->fs_info;
7072 struct extent_map_tree *map_tree = &fs_info->mapping_tree;
7073 struct map_lookup *map;
7074 struct extent_map *em;
7079 u8 uuid[BTRFS_UUID_SIZE];
7084 logical = key->offset;
7085 length = btrfs_chunk_length(leaf, chunk);
7086 type = btrfs_chunk_type(leaf, chunk);
7087 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
7089 #if BITS_PER_LONG == 32
7090 ret = check_32bit_meta_chunk(fs_info, logical, length, type);
7093 warn_32bit_meta_chunk(fs_info, logical, length, type);
7097 * Only need to verify chunk item if we're reading from sys chunk array,
7098 * as chunk item in tree block is already verified by tree-checker.
7100 if (leaf->start == BTRFS_SUPER_INFO_OFFSET) {
7101 ret = btrfs_check_chunk_valid(leaf, chunk, logical);
7106 read_lock(&map_tree->lock);
7107 em = lookup_extent_mapping(map_tree, logical, 1);
7108 read_unlock(&map_tree->lock);
7110 /* already mapped? */
7111 if (em && em->start <= logical && em->start + em->len > logical) {
7112 free_extent_map(em);
7115 free_extent_map(em);
7118 em = alloc_extent_map();
7121 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
7123 free_extent_map(em);
7127 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
7128 em->map_lookup = map;
7129 em->start = logical;
7132 em->block_start = 0;
7133 em->block_len = em->len;
7135 map->num_stripes = num_stripes;
7136 map->io_width = btrfs_chunk_io_width(leaf, chunk);
7137 map->io_align = btrfs_chunk_io_align(leaf, chunk);
7138 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
7140 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
7141 map->verified_stripes = 0;
7142 em->orig_block_len = calc_stripe_length(type, em->len,
7144 for (i = 0; i < num_stripes; i++) {
7145 map->stripes[i].physical =
7146 btrfs_stripe_offset_nr(leaf, chunk, i);
7147 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
7149 read_extent_buffer(leaf, uuid, (unsigned long)
7150 btrfs_stripe_dev_uuid_nr(chunk, i),
7153 map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices, &args);
7154 if (!map->stripes[i].dev) {
7155 map->stripes[i].dev = handle_missing_device(fs_info,
7157 if (IS_ERR(map->stripes[i].dev)) {
7158 ret = PTR_ERR(map->stripes[i].dev);
7159 free_extent_map(em);
7164 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
7165 &(map->stripes[i].dev->dev_state));
7168 write_lock(&map_tree->lock);
7169 ret = add_extent_mapping(map_tree, em, 0);
7170 write_unlock(&map_tree->lock);
7173 "failed to add chunk map, start=%llu len=%llu: %d",
7174 em->start, em->len, ret);
7176 free_extent_map(em);
7181 static void fill_device_from_item(struct extent_buffer *leaf,
7182 struct btrfs_dev_item *dev_item,
7183 struct btrfs_device *device)
7187 device->devid = btrfs_device_id(leaf, dev_item);
7188 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
7189 device->total_bytes = device->disk_total_bytes;
7190 device->commit_total_bytes = device->disk_total_bytes;
7191 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
7192 device->commit_bytes_used = device->bytes_used;
7193 device->type = btrfs_device_type(leaf, dev_item);
7194 device->io_align = btrfs_device_io_align(leaf, dev_item);
7195 device->io_width = btrfs_device_io_width(leaf, dev_item);
7196 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
7197 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
7198 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
7200 ptr = btrfs_device_uuid(dev_item);
7201 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
7204 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info,
7207 struct btrfs_fs_devices *fs_devices;
7210 lockdep_assert_held(&uuid_mutex);
7213 /* This will match only for multi-device seed fs */
7214 list_for_each_entry(fs_devices, &fs_info->fs_devices->seed_list, seed_list)
7215 if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE))
7219 fs_devices = find_fsid(fsid, NULL);
7221 if (!btrfs_test_opt(fs_info, DEGRADED))
7222 return ERR_PTR(-ENOENT);
7224 fs_devices = alloc_fs_devices(fsid, NULL);
7225 if (IS_ERR(fs_devices))
7228 fs_devices->seeding = true;
7229 fs_devices->opened = 1;
7234 * Upon first call for a seed fs fsid, just create a private copy of the
7235 * respective fs_devices and anchor it at fs_info->fs_devices->seed_list
7237 fs_devices = clone_fs_devices(fs_devices);
7238 if (IS_ERR(fs_devices))
7241 ret = open_fs_devices(fs_devices, FMODE_READ, fs_info->bdev_holder);
7243 free_fs_devices(fs_devices);
7244 return ERR_PTR(ret);
7247 if (!fs_devices->seeding) {
7248 close_fs_devices(fs_devices);
7249 free_fs_devices(fs_devices);
7250 return ERR_PTR(-EINVAL);
7253 list_add(&fs_devices->seed_list, &fs_info->fs_devices->seed_list);
7258 static int read_one_dev(struct extent_buffer *leaf,
7259 struct btrfs_dev_item *dev_item)
7261 BTRFS_DEV_LOOKUP_ARGS(args);
7262 struct btrfs_fs_info *fs_info = leaf->fs_info;
7263 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7264 struct btrfs_device *device;
7267 u8 fs_uuid[BTRFS_FSID_SIZE];
7268 u8 dev_uuid[BTRFS_UUID_SIZE];
7270 devid = args.devid = btrfs_device_id(leaf, dev_item);
7271 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
7273 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
7275 args.uuid = dev_uuid;
7276 args.fsid = fs_uuid;
7278 if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) {
7279 fs_devices = open_seed_devices(fs_info, fs_uuid);
7280 if (IS_ERR(fs_devices))
7281 return PTR_ERR(fs_devices);
7284 device = btrfs_find_device(fs_info->fs_devices, &args);
7286 if (!btrfs_test_opt(fs_info, DEGRADED)) {
7287 btrfs_report_missing_device(fs_info, devid,
7292 device = add_missing_dev(fs_devices, devid, dev_uuid);
7293 if (IS_ERR(device)) {
7295 "failed to add missing dev %llu: %ld",
7296 devid, PTR_ERR(device));
7297 return PTR_ERR(device);
7299 btrfs_report_missing_device(fs_info, devid, dev_uuid, false);
7301 if (!device->bdev) {
7302 if (!btrfs_test_opt(fs_info, DEGRADED)) {
7303 btrfs_report_missing_device(fs_info,
7304 devid, dev_uuid, true);
7307 btrfs_report_missing_device(fs_info, devid,
7311 if (!device->bdev &&
7312 !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
7314 * this happens when a device that was properly setup
7315 * in the device info lists suddenly goes bad.
7316 * device->bdev is NULL, and so we have to set
7317 * device->missing to one here
7319 device->fs_devices->missing_devices++;
7320 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
7323 /* Move the device to its own fs_devices */
7324 if (device->fs_devices != fs_devices) {
7325 ASSERT(test_bit(BTRFS_DEV_STATE_MISSING,
7326 &device->dev_state));
7328 list_move(&device->dev_list, &fs_devices->devices);
7329 device->fs_devices->num_devices--;
7330 fs_devices->num_devices++;
7332 device->fs_devices->missing_devices--;
7333 fs_devices->missing_devices++;
7335 device->fs_devices = fs_devices;
7339 if (device->fs_devices != fs_info->fs_devices) {
7340 BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state));
7341 if (device->generation !=
7342 btrfs_device_generation(leaf, dev_item))
7346 fill_device_from_item(leaf, dev_item, device);
7348 u64 max_total_bytes = i_size_read(device->bdev->bd_inode);
7350 if (device->total_bytes > max_total_bytes) {
7352 "device total_bytes should be at most %llu but found %llu",
7353 max_total_bytes, device->total_bytes);
7357 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
7358 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
7359 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
7360 device->fs_devices->total_rw_bytes += device->total_bytes;
7361 atomic64_add(device->total_bytes - device->bytes_used,
7362 &fs_info->free_chunk_space);
7368 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
7370 struct btrfs_root *root = fs_info->tree_root;
7371 struct btrfs_super_block *super_copy = fs_info->super_copy;
7372 struct extent_buffer *sb;
7373 struct btrfs_disk_key *disk_key;
7374 struct btrfs_chunk *chunk;
7376 unsigned long sb_array_offset;
7383 struct btrfs_key key;
7385 ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize);
7387 * This will create extent buffer of nodesize, superblock size is
7388 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
7389 * overallocate but we can keep it as-is, only the first page is used.
7391 sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET,
7392 root->root_key.objectid, 0);
7395 set_extent_buffer_uptodate(sb);
7397 * The sb extent buffer is artificial and just used to read the system array.
7398 * set_extent_buffer_uptodate() call does not properly mark all it's
7399 * pages up-to-date when the page is larger: extent does not cover the
7400 * whole page and consequently check_page_uptodate does not find all
7401 * the page's extents up-to-date (the hole beyond sb),
7402 * write_extent_buffer then triggers a WARN_ON.
7404 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
7405 * but sb spans only this function. Add an explicit SetPageUptodate call
7406 * to silence the warning eg. on PowerPC 64.
7408 if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
7409 SetPageUptodate(sb->pages[0]);
7411 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
7412 array_size = btrfs_super_sys_array_size(super_copy);
7414 array_ptr = super_copy->sys_chunk_array;
7415 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
7418 while (cur_offset < array_size) {
7419 disk_key = (struct btrfs_disk_key *)array_ptr;
7420 len = sizeof(*disk_key);
7421 if (cur_offset + len > array_size)
7422 goto out_short_read;
7424 btrfs_disk_key_to_cpu(&key, disk_key);
7427 sb_array_offset += len;
7430 if (key.type != BTRFS_CHUNK_ITEM_KEY) {
7432 "unexpected item type %u in sys_array at offset %u",
7433 (u32)key.type, cur_offset);
7438 chunk = (struct btrfs_chunk *)sb_array_offset;
7440 * At least one btrfs_chunk with one stripe must be present,
7441 * exact stripe count check comes afterwards
7443 len = btrfs_chunk_item_size(1);
7444 if (cur_offset + len > array_size)
7445 goto out_short_read;
7447 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
7450 "invalid number of stripes %u in sys_array at offset %u",
7451 num_stripes, cur_offset);
7456 type = btrfs_chunk_type(sb, chunk);
7457 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
7459 "invalid chunk type %llu in sys_array at offset %u",
7465 len = btrfs_chunk_item_size(num_stripes);
7466 if (cur_offset + len > array_size)
7467 goto out_short_read;
7469 ret = read_one_chunk(&key, sb, chunk);
7474 sb_array_offset += len;
7477 clear_extent_buffer_uptodate(sb);
7478 free_extent_buffer_stale(sb);
7482 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
7484 clear_extent_buffer_uptodate(sb);
7485 free_extent_buffer_stale(sb);
7490 * Check if all chunks in the fs are OK for read-write degraded mount
7492 * If the @failing_dev is specified, it's accounted as missing.
7494 * Return true if all chunks meet the minimal RW mount requirements.
7495 * Return false if any chunk doesn't meet the minimal RW mount requirements.
7497 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
7498 struct btrfs_device *failing_dev)
7500 struct extent_map_tree *map_tree = &fs_info->mapping_tree;
7501 struct extent_map *em;
7505 read_lock(&map_tree->lock);
7506 em = lookup_extent_mapping(map_tree, 0, (u64)-1);
7507 read_unlock(&map_tree->lock);
7508 /* No chunk at all? Return false anyway */
7514 struct map_lookup *map;
7519 map = em->map_lookup;
7521 btrfs_get_num_tolerated_disk_barrier_failures(
7523 for (i = 0; i < map->num_stripes; i++) {
7524 struct btrfs_device *dev = map->stripes[i].dev;
7526 if (!dev || !dev->bdev ||
7527 test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) ||
7528 dev->last_flush_error)
7530 else if (failing_dev && failing_dev == dev)
7533 if (missing > max_tolerated) {
7536 "chunk %llu missing %d devices, max tolerance is %d for writable mount",
7537 em->start, missing, max_tolerated);
7538 free_extent_map(em);
7542 next_start = extent_map_end(em);
7543 free_extent_map(em);
7545 read_lock(&map_tree->lock);
7546 em = lookup_extent_mapping(map_tree, next_start,
7547 (u64)(-1) - next_start);
7548 read_unlock(&map_tree->lock);
7554 static void readahead_tree_node_children(struct extent_buffer *node)
7557 const int nr_items = btrfs_header_nritems(node);
7559 for (i = 0; i < nr_items; i++)
7560 btrfs_readahead_node_child(node, i);
7563 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
7565 struct btrfs_root *root = fs_info->chunk_root;
7566 struct btrfs_path *path;
7567 struct extent_buffer *leaf;
7568 struct btrfs_key key;
7569 struct btrfs_key found_key;
7573 u64 last_ra_node = 0;
7575 path = btrfs_alloc_path();
7580 * uuid_mutex is needed only if we are mounting a sprout FS
7581 * otherwise we don't need it.
7583 mutex_lock(&uuid_mutex);
7586 * It is possible for mount and umount to race in such a way that
7587 * we execute this code path, but open_fs_devices failed to clear
7588 * total_rw_bytes. We certainly want it cleared before reading the
7589 * device items, so clear it here.
7591 fs_info->fs_devices->total_rw_bytes = 0;
7594 * Lockdep complains about possible circular locking dependency between
7595 * a disk's open_mutex (struct gendisk.open_mutex), the rw semaphores
7596 * used for freeze procection of a fs (struct super_block.s_writers),
7597 * which we take when starting a transaction, and extent buffers of the
7598 * chunk tree if we call read_one_dev() while holding a lock on an
7599 * extent buffer of the chunk tree. Since we are mounting the filesystem
7600 * and at this point there can't be any concurrent task modifying the
7601 * chunk tree, to keep it simple, just skip locking on the chunk tree.
7603 ASSERT(!test_bit(BTRFS_FS_OPEN, &fs_info->flags));
7604 path->skip_locking = 1;
7607 * Read all device items, and then all the chunk items. All
7608 * device items are found before any chunk item (their object id
7609 * is smaller than the lowest possible object id for a chunk
7610 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
7612 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
7615 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7619 struct extent_buffer *node;
7621 leaf = path->nodes[0];
7622 slot = path->slots[0];
7623 if (slot >= btrfs_header_nritems(leaf)) {
7624 ret = btrfs_next_leaf(root, path);
7631 node = path->nodes[1];
7633 if (last_ra_node != node->start) {
7634 readahead_tree_node_children(node);
7635 last_ra_node = node->start;
7638 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7639 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
7640 struct btrfs_dev_item *dev_item;
7641 dev_item = btrfs_item_ptr(leaf, slot,
7642 struct btrfs_dev_item);
7643 ret = read_one_dev(leaf, dev_item);
7647 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
7648 struct btrfs_chunk *chunk;
7651 * We are only called at mount time, so no need to take
7652 * fs_info->chunk_mutex. Plus, to avoid lockdep warnings,
7653 * we always lock first fs_info->chunk_mutex before
7654 * acquiring any locks on the chunk tree. This is a
7655 * requirement for chunk allocation, see the comment on
7656 * top of btrfs_chunk_alloc() for details.
7658 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
7659 ret = read_one_chunk(&found_key, leaf, chunk);
7667 * After loading chunk tree, we've got all device information,
7668 * do another round of validation checks.
7670 if (total_dev != fs_info->fs_devices->total_devices) {
7672 "super block num_devices %llu mismatch with DEV_ITEM count %llu, will be repaired on next transaction commit",
7673 btrfs_super_num_devices(fs_info->super_copy),
7675 fs_info->fs_devices->total_devices = total_dev;
7676 btrfs_set_super_num_devices(fs_info->super_copy, total_dev);
7678 if (btrfs_super_total_bytes(fs_info->super_copy) <
7679 fs_info->fs_devices->total_rw_bytes) {
7681 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
7682 btrfs_super_total_bytes(fs_info->super_copy),
7683 fs_info->fs_devices->total_rw_bytes);
7689 mutex_unlock(&uuid_mutex);
7691 btrfs_free_path(path);
7695 int btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
7697 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
7698 struct btrfs_device *device;
7701 fs_devices->fs_info = fs_info;
7703 mutex_lock(&fs_devices->device_list_mutex);
7704 list_for_each_entry(device, &fs_devices->devices, dev_list)
7705 device->fs_info = fs_info;
7707 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
7708 list_for_each_entry(device, &seed_devs->devices, dev_list) {
7709 device->fs_info = fs_info;
7710 ret = btrfs_get_dev_zone_info(device, false);
7715 seed_devs->fs_info = fs_info;
7717 mutex_unlock(&fs_devices->device_list_mutex);
7722 static u64 btrfs_dev_stats_value(const struct extent_buffer *eb,
7723 const struct btrfs_dev_stats_item *ptr,
7728 read_extent_buffer(eb, &val,
7729 offsetof(struct btrfs_dev_stats_item, values) +
7730 ((unsigned long)ptr) + (index * sizeof(u64)),
7735 static void btrfs_set_dev_stats_value(struct extent_buffer *eb,
7736 struct btrfs_dev_stats_item *ptr,
7739 write_extent_buffer(eb, &val,
7740 offsetof(struct btrfs_dev_stats_item, values) +
7741 ((unsigned long)ptr) + (index * sizeof(u64)),
7745 static int btrfs_device_init_dev_stats(struct btrfs_device *device,
7746 struct btrfs_path *path)
7748 struct btrfs_dev_stats_item *ptr;
7749 struct extent_buffer *eb;
7750 struct btrfs_key key;
7754 if (!device->fs_info->dev_root)
7757 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7758 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7759 key.offset = device->devid;
7760 ret = btrfs_search_slot(NULL, device->fs_info->dev_root, &key, path, 0, 0);
7762 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7763 btrfs_dev_stat_set(device, i, 0);
7764 device->dev_stats_valid = 1;
7765 btrfs_release_path(path);
7766 return ret < 0 ? ret : 0;
7768 slot = path->slots[0];
7769 eb = path->nodes[0];
7770 item_size = btrfs_item_size_nr(eb, slot);
7772 ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_stats_item);
7774 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7775 if (item_size >= (1 + i) * sizeof(__le64))
7776 btrfs_dev_stat_set(device, i,
7777 btrfs_dev_stats_value(eb, ptr, i));
7779 btrfs_dev_stat_set(device, i, 0);
7782 device->dev_stats_valid = 1;
7783 btrfs_dev_stat_print_on_load(device);
7784 btrfs_release_path(path);
7789 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
7791 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
7792 struct btrfs_device *device;
7793 struct btrfs_path *path = NULL;
7796 path = btrfs_alloc_path();
7800 mutex_lock(&fs_devices->device_list_mutex);
7801 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7802 ret = btrfs_device_init_dev_stats(device, path);
7806 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
7807 list_for_each_entry(device, &seed_devs->devices, dev_list) {
7808 ret = btrfs_device_init_dev_stats(device, path);
7814 mutex_unlock(&fs_devices->device_list_mutex);
7816 btrfs_free_path(path);
7820 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
7821 struct btrfs_device *device)
7823 struct btrfs_fs_info *fs_info = trans->fs_info;
7824 struct btrfs_root *dev_root = fs_info->dev_root;
7825 struct btrfs_path *path;
7826 struct btrfs_key key;
7827 struct extent_buffer *eb;
7828 struct btrfs_dev_stats_item *ptr;
7832 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7833 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7834 key.offset = device->devid;
7836 path = btrfs_alloc_path();
7839 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
7841 btrfs_warn_in_rcu(fs_info,
7842 "error %d while searching for dev_stats item for device %s",
7843 ret, rcu_str_deref(device->name));
7848 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
7849 /* need to delete old one and insert a new one */
7850 ret = btrfs_del_item(trans, dev_root, path);
7852 btrfs_warn_in_rcu(fs_info,
7853 "delete too small dev_stats item for device %s failed %d",
7854 rcu_str_deref(device->name), ret);
7861 /* need to insert a new item */
7862 btrfs_release_path(path);
7863 ret = btrfs_insert_empty_item(trans, dev_root, path,
7864 &key, sizeof(*ptr));
7866 btrfs_warn_in_rcu(fs_info,
7867 "insert dev_stats item for device %s failed %d",
7868 rcu_str_deref(device->name), ret);
7873 eb = path->nodes[0];
7874 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7875 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7876 btrfs_set_dev_stats_value(eb, ptr, i,
7877 btrfs_dev_stat_read(device, i));
7878 btrfs_mark_buffer_dirty(eb);
7881 btrfs_free_path(path);
7886 * called from commit_transaction. Writes all changed device stats to disk.
7888 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans)
7890 struct btrfs_fs_info *fs_info = trans->fs_info;
7891 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7892 struct btrfs_device *device;
7896 mutex_lock(&fs_devices->device_list_mutex);
7897 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7898 stats_cnt = atomic_read(&device->dev_stats_ccnt);
7899 if (!device->dev_stats_valid || stats_cnt == 0)
7904 * There is a LOAD-LOAD control dependency between the value of
7905 * dev_stats_ccnt and updating the on-disk values which requires
7906 * reading the in-memory counters. Such control dependencies
7907 * require explicit read memory barriers.
7909 * This memory barriers pairs with smp_mb__before_atomic in
7910 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full
7911 * barrier implied by atomic_xchg in
7912 * btrfs_dev_stats_read_and_reset
7916 ret = update_dev_stat_item(trans, device);
7918 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7920 mutex_unlock(&fs_devices->device_list_mutex);
7925 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7927 btrfs_dev_stat_inc(dev, index);
7928 btrfs_dev_stat_print_on_error(dev);
7931 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
7933 if (!dev->dev_stats_valid)
7935 btrfs_err_rl_in_rcu(dev->fs_info,
7936 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7937 rcu_str_deref(dev->name),
7938 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7939 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7940 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7941 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7942 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7945 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7949 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7950 if (btrfs_dev_stat_read(dev, i) != 0)
7952 if (i == BTRFS_DEV_STAT_VALUES_MAX)
7953 return; /* all values == 0, suppress message */
7955 btrfs_info_in_rcu(dev->fs_info,
7956 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7957 rcu_str_deref(dev->name),
7958 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7959 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7960 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7961 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7962 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7965 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
7966 struct btrfs_ioctl_get_dev_stats *stats)
7968 BTRFS_DEV_LOOKUP_ARGS(args);
7969 struct btrfs_device *dev;
7970 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7973 mutex_lock(&fs_devices->device_list_mutex);
7974 args.devid = stats->devid;
7975 dev = btrfs_find_device(fs_info->fs_devices, &args);
7976 mutex_unlock(&fs_devices->device_list_mutex);
7979 btrfs_warn(fs_info, "get dev_stats failed, device not found");
7981 } else if (!dev->dev_stats_valid) {
7982 btrfs_warn(fs_info, "get dev_stats failed, not yet valid");
7984 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7985 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7986 if (stats->nr_items > i)
7988 btrfs_dev_stat_read_and_reset(dev, i);
7990 btrfs_dev_stat_set(dev, i, 0);
7992 btrfs_info(fs_info, "device stats zeroed by %s (%d)",
7993 current->comm, task_pid_nr(current));
7995 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7996 if (stats->nr_items > i)
7997 stats->values[i] = btrfs_dev_stat_read(dev, i);
7999 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
8000 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
8005 * Update the size and bytes used for each device where it changed. This is
8006 * delayed since we would otherwise get errors while writing out the
8009 * Must be invoked during transaction commit.
8011 void btrfs_commit_device_sizes(struct btrfs_transaction *trans)
8013 struct btrfs_device *curr, *next;
8015 ASSERT(trans->state == TRANS_STATE_COMMIT_DOING);
8017 if (list_empty(&trans->dev_update_list))
8021 * We don't need the device_list_mutex here. This list is owned by the
8022 * transaction and the transaction must complete before the device is
8025 mutex_lock(&trans->fs_info->chunk_mutex);
8026 list_for_each_entry_safe(curr, next, &trans->dev_update_list,
8028 list_del_init(&curr->post_commit_list);
8029 curr->commit_total_bytes = curr->disk_total_bytes;
8030 curr->commit_bytes_used = curr->bytes_used;
8032 mutex_unlock(&trans->fs_info->chunk_mutex);
8036 * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10.
8038 int btrfs_bg_type_to_factor(u64 flags)
8040 const int index = btrfs_bg_flags_to_raid_index(flags);
8042 return btrfs_raid_array[index].ncopies;
8047 static int verify_one_dev_extent(struct btrfs_fs_info *fs_info,
8048 u64 chunk_offset, u64 devid,
8049 u64 physical_offset, u64 physical_len)
8051 struct btrfs_dev_lookup_args args = { .devid = devid };
8052 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
8053 struct extent_map *em;
8054 struct map_lookup *map;
8055 struct btrfs_device *dev;
8061 read_lock(&em_tree->lock);
8062 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
8063 read_unlock(&em_tree->lock);
8067 "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk",
8068 physical_offset, devid);
8073 map = em->map_lookup;
8074 stripe_len = calc_stripe_length(map->type, em->len, map->num_stripes);
8075 if (physical_len != stripe_len) {
8077 "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu",
8078 physical_offset, devid, em->start, physical_len,
8084 for (i = 0; i < map->num_stripes; i++) {
8085 if (map->stripes[i].dev->devid == devid &&
8086 map->stripes[i].physical == physical_offset) {
8088 if (map->verified_stripes >= map->num_stripes) {
8090 "too many dev extents for chunk %llu found",
8095 map->verified_stripes++;
8101 "dev extent physical offset %llu devid %llu has no corresponding chunk",
8102 physical_offset, devid);
8106 /* Make sure no dev extent is beyond device boundary */
8107 dev = btrfs_find_device(fs_info->fs_devices, &args);
8109 btrfs_err(fs_info, "failed to find devid %llu", devid);
8114 if (physical_offset + physical_len > dev->disk_total_bytes) {
8116 "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu",
8117 devid, physical_offset, physical_len,
8118 dev->disk_total_bytes);
8123 if (dev->zone_info) {
8124 u64 zone_size = dev->zone_info->zone_size;
8126 if (!IS_ALIGNED(physical_offset, zone_size) ||
8127 !IS_ALIGNED(physical_len, zone_size)) {
8129 "zoned: dev extent devid %llu physical offset %llu len %llu is not aligned to device zone",
8130 devid, physical_offset, physical_len);
8137 free_extent_map(em);
8141 static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info)
8143 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
8144 struct extent_map *em;
8145 struct rb_node *node;
8148 read_lock(&em_tree->lock);
8149 for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) {
8150 em = rb_entry(node, struct extent_map, rb_node);
8151 if (em->map_lookup->num_stripes !=
8152 em->map_lookup->verified_stripes) {
8154 "chunk %llu has missing dev extent, have %d expect %d",
8155 em->start, em->map_lookup->verified_stripes,
8156 em->map_lookup->num_stripes);
8162 read_unlock(&em_tree->lock);
8167 * Ensure that all dev extents are mapped to correct chunk, otherwise
8168 * later chunk allocation/free would cause unexpected behavior.
8170 * NOTE: This will iterate through the whole device tree, which should be of
8171 * the same size level as the chunk tree. This slightly increases mount time.
8173 int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info)
8175 struct btrfs_path *path;
8176 struct btrfs_root *root = fs_info->dev_root;
8177 struct btrfs_key key;
8179 u64 prev_dev_ext_end = 0;
8183 * We don't have a dev_root because we mounted with ignorebadroots and
8184 * failed to load the root, so we want to skip the verification in this
8187 * However if the dev root is fine, but the tree itself is corrupted
8188 * we'd still fail to mount. This verification is only to make sure
8189 * writes can happen safely, so instead just bypass this check
8190 * completely in the case of IGNOREBADROOTS.
8192 if (btrfs_test_opt(fs_info, IGNOREBADROOTS))
8196 key.type = BTRFS_DEV_EXTENT_KEY;
8199 path = btrfs_alloc_path();
8203 path->reada = READA_FORWARD;
8204 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
8208 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
8209 ret = btrfs_next_leaf(root, path);
8212 /* No dev extents at all? Not good */
8219 struct extent_buffer *leaf = path->nodes[0];
8220 struct btrfs_dev_extent *dext;
8221 int slot = path->slots[0];
8223 u64 physical_offset;
8227 btrfs_item_key_to_cpu(leaf, &key, slot);
8228 if (key.type != BTRFS_DEV_EXTENT_KEY)
8230 devid = key.objectid;
8231 physical_offset = key.offset;
8233 dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent);
8234 chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext);
8235 physical_len = btrfs_dev_extent_length(leaf, dext);
8237 /* Check if this dev extent overlaps with the previous one */
8238 if (devid == prev_devid && physical_offset < prev_dev_ext_end) {
8240 "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu",
8241 devid, physical_offset, prev_dev_ext_end);
8246 ret = verify_one_dev_extent(fs_info, chunk_offset, devid,
8247 physical_offset, physical_len);
8251 prev_dev_ext_end = physical_offset + physical_len;
8253 ret = btrfs_next_item(root, path);
8262 /* Ensure all chunks have corresponding dev extents */
8263 ret = verify_chunk_dev_extent_mapping(fs_info);
8265 btrfs_free_path(path);
8270 * Check whether the given block group or device is pinned by any inode being
8271 * used as a swapfile.
8273 bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr)
8275 struct btrfs_swapfile_pin *sp;
8276 struct rb_node *node;
8278 spin_lock(&fs_info->swapfile_pins_lock);
8279 node = fs_info->swapfile_pins.rb_node;
8281 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
8283 node = node->rb_left;
8284 else if (ptr > sp->ptr)
8285 node = node->rb_right;
8289 spin_unlock(&fs_info->swapfile_pins_lock);
8290 return node != NULL;
8293 static int relocating_repair_kthread(void *data)
8295 struct btrfs_block_group *cache = (struct btrfs_block_group *)data;
8296 struct btrfs_fs_info *fs_info = cache->fs_info;
8300 target = cache->start;
8301 btrfs_put_block_group(cache);
8303 sb_start_write(fs_info->sb);
8304 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) {
8306 "zoned: skip relocating block group %llu to repair: EBUSY",
8308 sb_end_write(fs_info->sb);
8312 mutex_lock(&fs_info->reclaim_bgs_lock);
8314 /* Ensure block group still exists */
8315 cache = btrfs_lookup_block_group(fs_info, target);
8319 if (!cache->relocating_repair)
8322 ret = btrfs_may_alloc_data_chunk(fs_info, target);
8327 "zoned: relocating block group %llu to repair IO failure",
8329 ret = btrfs_relocate_chunk(fs_info, target);
8333 btrfs_put_block_group(cache);
8334 mutex_unlock(&fs_info->reclaim_bgs_lock);
8335 btrfs_exclop_finish(fs_info);
8336 sb_end_write(fs_info->sb);
8341 int btrfs_repair_one_zone(struct btrfs_fs_info *fs_info, u64 logical)
8343 struct btrfs_block_group *cache;
8345 /* Do not attempt to repair in degraded state */
8346 if (btrfs_test_opt(fs_info, DEGRADED))
8349 cache = btrfs_lookup_block_group(fs_info, logical);
8353 spin_lock(&cache->lock);
8354 if (cache->relocating_repair) {
8355 spin_unlock(&cache->lock);
8356 btrfs_put_block_group(cache);
8359 cache->relocating_repair = 1;
8360 spin_unlock(&cache->lock);
8362 kthread_run(relocating_repair_kthread, cache,
8363 "btrfs-relocating-repair");