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_bio **bbio_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);
533 static bool device_path_matched(const char *path, struct btrfs_device *device)
538 found = strcmp(rcu_str_deref(device->name), path);
545 * Search and remove all stale (devices which are not mounted) devices.
546 * When both inputs are NULL, it will search and release all stale devices.
547 * path: Optional. When provided will it release all unmounted devices
548 * matching this path only.
549 * skip_dev: Optional. Will skip this device when searching for the stale
551 * Return: 0 for success or if @path is NULL.
552 * -EBUSY if @path is a mounted device.
553 * -ENOENT if @path does not match any device in the list.
555 static int btrfs_free_stale_devices(const char *path,
556 struct btrfs_device *skip_device)
558 struct btrfs_fs_devices *fs_devices, *tmp_fs_devices;
559 struct btrfs_device *device, *tmp_device;
562 lockdep_assert_held(&uuid_mutex);
567 list_for_each_entry_safe(fs_devices, tmp_fs_devices, &fs_uuids, fs_list) {
569 mutex_lock(&fs_devices->device_list_mutex);
570 list_for_each_entry_safe(device, tmp_device,
571 &fs_devices->devices, dev_list) {
572 if (skip_device && skip_device == device)
574 if (path && !device->name)
576 if (path && !device_path_matched(path, device))
578 if (fs_devices->opened) {
579 /* for an already deleted device return 0 */
580 if (path && ret != 0)
585 /* delete the stale device */
586 fs_devices->num_devices--;
587 list_del(&device->dev_list);
588 btrfs_free_device(device);
592 mutex_unlock(&fs_devices->device_list_mutex);
594 if (fs_devices->num_devices == 0) {
595 btrfs_sysfs_remove_fsid(fs_devices);
596 list_del(&fs_devices->fs_list);
597 free_fs_devices(fs_devices);
605 * This is only used on mount, and we are protected from competing things
606 * messing with our fs_devices by the uuid_mutex, thus we do not need the
607 * fs_devices->device_list_mutex here.
609 static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices,
610 struct btrfs_device *device, fmode_t flags,
613 struct request_queue *q;
614 struct block_device *bdev;
615 struct btrfs_super_block *disk_super;
624 ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
629 devid = btrfs_stack_device_id(&disk_super->dev_item);
630 if (devid != device->devid)
631 goto error_free_page;
633 if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE))
634 goto error_free_page;
636 device->generation = btrfs_super_generation(disk_super);
638 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
639 if (btrfs_super_incompat_flags(disk_super) &
640 BTRFS_FEATURE_INCOMPAT_METADATA_UUID) {
642 "BTRFS: Invalid seeding and uuid-changed device detected\n");
643 goto error_free_page;
646 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
647 fs_devices->seeding = true;
649 if (bdev_read_only(bdev))
650 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
652 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
655 q = bdev_get_queue(bdev);
656 if (!blk_queue_nonrot(q))
657 fs_devices->rotating = true;
660 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
661 device->mode = flags;
663 fs_devices->open_devices++;
664 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
665 device->devid != BTRFS_DEV_REPLACE_DEVID) {
666 fs_devices->rw_devices++;
667 list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list);
669 btrfs_release_disk_super(disk_super);
674 btrfs_release_disk_super(disk_super);
675 blkdev_put(bdev, flags);
681 * Handle scanned device having its CHANGING_FSID_V2 flag set and the fs_devices
682 * being created with a disk that has already completed its fsid change. Such
683 * disk can belong to an fs which has its FSID changed or to one which doesn't.
684 * Handle both cases here.
686 static struct btrfs_fs_devices *find_fsid_inprogress(
687 struct btrfs_super_block *disk_super)
689 struct btrfs_fs_devices *fs_devices;
691 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
692 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid,
693 BTRFS_FSID_SIZE) != 0 &&
694 memcmp(fs_devices->metadata_uuid, disk_super->fsid,
695 BTRFS_FSID_SIZE) == 0 && !fs_devices->fsid_change) {
700 return find_fsid(disk_super->fsid, NULL);
704 static struct btrfs_fs_devices *find_fsid_changed(
705 struct btrfs_super_block *disk_super)
707 struct btrfs_fs_devices *fs_devices;
710 * Handles the case where scanned device is part of an fs that had
711 * multiple successful changes of FSID but currently device didn't
712 * observe it. Meaning our fsid will be different than theirs. We need
713 * to handle two subcases :
714 * 1 - The fs still continues to have different METADATA/FSID uuids.
715 * 2 - The fs is switched back to its original FSID (METADATA/FSID
718 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
720 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid,
721 BTRFS_FSID_SIZE) != 0 &&
722 memcmp(fs_devices->metadata_uuid, disk_super->metadata_uuid,
723 BTRFS_FSID_SIZE) == 0 &&
724 memcmp(fs_devices->fsid, disk_super->fsid,
725 BTRFS_FSID_SIZE) != 0)
728 /* Unchanged UUIDs */
729 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid,
730 BTRFS_FSID_SIZE) == 0 &&
731 memcmp(fs_devices->fsid, disk_super->metadata_uuid,
732 BTRFS_FSID_SIZE) == 0)
739 static struct btrfs_fs_devices *find_fsid_reverted_metadata(
740 struct btrfs_super_block *disk_super)
742 struct btrfs_fs_devices *fs_devices;
745 * Handle the case where the scanned device is part of an fs whose last
746 * metadata UUID change reverted it to the original FSID. At the same
747 * time * fs_devices was first created by another constitutent device
748 * which didn't fully observe the operation. This results in an
749 * btrfs_fs_devices created with metadata/fsid different AND
750 * btrfs_fs_devices::fsid_change set AND the metadata_uuid of the
751 * fs_devices equal to the FSID of the disk.
753 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
754 if (memcmp(fs_devices->fsid, fs_devices->metadata_uuid,
755 BTRFS_FSID_SIZE) != 0 &&
756 memcmp(fs_devices->metadata_uuid, disk_super->fsid,
757 BTRFS_FSID_SIZE) == 0 &&
758 fs_devices->fsid_change)
765 * Add new device to list of registered devices
768 * device pointer which was just added or updated when successful
769 * error pointer when failed
771 static noinline struct btrfs_device *device_list_add(const char *path,
772 struct btrfs_super_block *disk_super,
773 bool *new_device_added)
775 struct btrfs_device *device;
776 struct btrfs_fs_devices *fs_devices = NULL;
777 struct rcu_string *name;
778 u64 found_transid = btrfs_super_generation(disk_super);
779 u64 devid = btrfs_stack_device_id(&disk_super->dev_item);
780 bool has_metadata_uuid = (btrfs_super_incompat_flags(disk_super) &
781 BTRFS_FEATURE_INCOMPAT_METADATA_UUID);
782 bool fsid_change_in_progress = (btrfs_super_flags(disk_super) &
783 BTRFS_SUPER_FLAG_CHANGING_FSID_V2);
785 if (fsid_change_in_progress) {
786 if (!has_metadata_uuid)
787 fs_devices = find_fsid_inprogress(disk_super);
789 fs_devices = find_fsid_changed(disk_super);
790 } else if (has_metadata_uuid) {
791 fs_devices = find_fsid_with_metadata_uuid(disk_super);
793 fs_devices = find_fsid_reverted_metadata(disk_super);
795 fs_devices = find_fsid(disk_super->fsid, NULL);
800 if (has_metadata_uuid)
801 fs_devices = alloc_fs_devices(disk_super->fsid,
802 disk_super->metadata_uuid);
804 fs_devices = alloc_fs_devices(disk_super->fsid, NULL);
806 if (IS_ERR(fs_devices))
807 return ERR_CAST(fs_devices);
809 fs_devices->fsid_change = fsid_change_in_progress;
811 mutex_lock(&fs_devices->device_list_mutex);
812 list_add(&fs_devices->fs_list, &fs_uuids);
816 mutex_lock(&fs_devices->device_list_mutex);
817 device = btrfs_find_device(fs_devices, devid,
818 disk_super->dev_item.uuid, NULL);
821 * If this disk has been pulled into an fs devices created by
822 * a device which had the CHANGING_FSID_V2 flag then replace the
823 * metadata_uuid/fsid values of the fs_devices.
825 if (fs_devices->fsid_change &&
826 found_transid > fs_devices->latest_generation) {
827 memcpy(fs_devices->fsid, disk_super->fsid,
830 if (has_metadata_uuid)
831 memcpy(fs_devices->metadata_uuid,
832 disk_super->metadata_uuid,
835 memcpy(fs_devices->metadata_uuid,
836 disk_super->fsid, BTRFS_FSID_SIZE);
838 fs_devices->fsid_change = false;
843 if (fs_devices->opened) {
844 mutex_unlock(&fs_devices->device_list_mutex);
845 return ERR_PTR(-EBUSY);
848 device = btrfs_alloc_device(NULL, &devid,
849 disk_super->dev_item.uuid);
850 if (IS_ERR(device)) {
851 mutex_unlock(&fs_devices->device_list_mutex);
852 /* we can safely leave the fs_devices entry around */
856 name = rcu_string_strdup(path, GFP_NOFS);
858 btrfs_free_device(device);
859 mutex_unlock(&fs_devices->device_list_mutex);
860 return ERR_PTR(-ENOMEM);
862 rcu_assign_pointer(device->name, name);
864 list_add_rcu(&device->dev_list, &fs_devices->devices);
865 fs_devices->num_devices++;
867 device->fs_devices = fs_devices;
868 *new_device_added = true;
870 if (disk_super->label[0])
872 "BTRFS: device label %s devid %llu transid %llu %s scanned by %s (%d)\n",
873 disk_super->label, devid, found_transid, path,
874 current->comm, task_pid_nr(current));
877 "BTRFS: device fsid %pU devid %llu transid %llu %s scanned by %s (%d)\n",
878 disk_super->fsid, devid, found_transid, path,
879 current->comm, task_pid_nr(current));
881 } else if (!device->name || strcmp(device->name->str, path)) {
883 * When FS is already mounted.
884 * 1. If you are here and if the device->name is NULL that
885 * means this device was missing at time of FS mount.
886 * 2. If you are here and if the device->name is different
887 * from 'path' that means either
888 * a. The same device disappeared and reappeared with
890 * b. The missing-disk-which-was-replaced, has
893 * We must allow 1 and 2a above. But 2b would be a spurious
896 * Further in case of 1 and 2a above, the disk at 'path'
897 * would have missed some transaction when it was away and
898 * in case of 2a the stale bdev has to be updated as well.
899 * 2b must not be allowed at all time.
903 * For now, we do allow update to btrfs_fs_device through the
904 * btrfs dev scan cli after FS has been mounted. We're still
905 * tracking a problem where systems fail mount by subvolume id
906 * when we reject replacement on a mounted FS.
908 if (!fs_devices->opened && found_transid < device->generation) {
910 * That is if the FS is _not_ mounted and if you
911 * are here, that means there is more than one
912 * disk with same uuid and devid.We keep the one
913 * with larger generation number or the last-in if
914 * generation are equal.
916 mutex_unlock(&fs_devices->device_list_mutex);
917 return ERR_PTR(-EEXIST);
921 * We are going to replace the device path for a given devid,
922 * make sure it's the same device if the device is mounted
928 error = lookup_bdev(path, &path_dev);
930 mutex_unlock(&fs_devices->device_list_mutex);
931 return ERR_PTR(error);
934 if (device->bdev->bd_dev != path_dev) {
935 mutex_unlock(&fs_devices->device_list_mutex);
937 * device->fs_info may not be reliable here, so
938 * pass in a NULL instead. This avoids a
939 * possible use-after-free when the fs_info and
940 * fs_info->sb are already torn down.
942 btrfs_warn_in_rcu(NULL,
943 "duplicate device %s devid %llu generation %llu scanned by %s (%d)",
944 path, devid, found_transid,
946 task_pid_nr(current));
947 return ERR_PTR(-EEXIST);
949 btrfs_info_in_rcu(device->fs_info,
950 "devid %llu device path %s changed to %s scanned by %s (%d)",
951 devid, rcu_str_deref(device->name),
953 task_pid_nr(current));
956 name = rcu_string_strdup(path, GFP_NOFS);
958 mutex_unlock(&fs_devices->device_list_mutex);
959 return ERR_PTR(-ENOMEM);
961 rcu_string_free(device->name);
962 rcu_assign_pointer(device->name, name);
963 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
964 fs_devices->missing_devices--;
965 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
970 * Unmount does not free the btrfs_device struct but would zero
971 * generation along with most of the other members. So just update
972 * it back. We need it to pick the disk with largest generation
975 if (!fs_devices->opened) {
976 device->generation = found_transid;
977 fs_devices->latest_generation = max_t(u64, found_transid,
978 fs_devices->latest_generation);
981 fs_devices->total_devices = btrfs_super_num_devices(disk_super);
983 mutex_unlock(&fs_devices->device_list_mutex);
987 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
989 struct btrfs_fs_devices *fs_devices;
990 struct btrfs_device *device;
991 struct btrfs_device *orig_dev;
994 lockdep_assert_held(&uuid_mutex);
996 fs_devices = alloc_fs_devices(orig->fsid, NULL);
997 if (IS_ERR(fs_devices))
1000 fs_devices->total_devices = orig->total_devices;
1002 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
1003 struct rcu_string *name;
1005 device = btrfs_alloc_device(NULL, &orig_dev->devid,
1007 if (IS_ERR(device)) {
1008 ret = PTR_ERR(device);
1013 * This is ok to do without rcu read locked because we hold the
1014 * uuid mutex so nothing we touch in here is going to disappear.
1016 if (orig_dev->name) {
1017 name = rcu_string_strdup(orig_dev->name->str,
1020 btrfs_free_device(device);
1024 rcu_assign_pointer(device->name, name);
1027 list_add(&device->dev_list, &fs_devices->devices);
1028 device->fs_devices = fs_devices;
1029 fs_devices->num_devices++;
1033 free_fs_devices(fs_devices);
1034 return ERR_PTR(ret);
1037 static void __btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices,
1038 struct btrfs_device **latest_dev)
1040 struct btrfs_device *device, *next;
1042 /* This is the initialized path, it is safe to release the devices. */
1043 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
1044 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)) {
1045 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
1046 &device->dev_state) &&
1047 !test_bit(BTRFS_DEV_STATE_MISSING,
1048 &device->dev_state) &&
1050 device->generation > (*latest_dev)->generation)) {
1051 *latest_dev = device;
1057 * We have already validated the presence of BTRFS_DEV_REPLACE_DEVID,
1058 * in btrfs_init_dev_replace() so just continue.
1060 if (device->devid == BTRFS_DEV_REPLACE_DEVID)
1064 blkdev_put(device->bdev, device->mode);
1065 device->bdev = NULL;
1066 fs_devices->open_devices--;
1068 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1069 list_del_init(&device->dev_alloc_list);
1070 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
1071 fs_devices->rw_devices--;
1073 list_del_init(&device->dev_list);
1074 fs_devices->num_devices--;
1075 btrfs_free_device(device);
1081 * After we have read the system tree and know devids belonging to this
1082 * filesystem, remove the device which does not belong there.
1084 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices)
1086 struct btrfs_device *latest_dev = NULL;
1087 struct btrfs_fs_devices *seed_dev;
1089 mutex_lock(&uuid_mutex);
1090 __btrfs_free_extra_devids(fs_devices, &latest_dev);
1092 list_for_each_entry(seed_dev, &fs_devices->seed_list, seed_list)
1093 __btrfs_free_extra_devids(seed_dev, &latest_dev);
1095 fs_devices->latest_dev = latest_dev;
1097 mutex_unlock(&uuid_mutex);
1100 static void btrfs_close_bdev(struct btrfs_device *device)
1105 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1106 sync_blockdev(device->bdev);
1107 invalidate_bdev(device->bdev);
1110 blkdev_put(device->bdev, device->mode);
1113 static void btrfs_close_one_device(struct btrfs_device *device)
1115 struct btrfs_fs_devices *fs_devices = device->fs_devices;
1117 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
1118 device->devid != BTRFS_DEV_REPLACE_DEVID) {
1119 list_del_init(&device->dev_alloc_list);
1120 fs_devices->rw_devices--;
1123 if (device->devid == BTRFS_DEV_REPLACE_DEVID)
1124 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
1126 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
1127 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
1128 fs_devices->missing_devices--;
1131 btrfs_close_bdev(device);
1133 fs_devices->open_devices--;
1134 device->bdev = NULL;
1136 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
1137 btrfs_destroy_dev_zone_info(device);
1139 device->fs_info = NULL;
1140 atomic_set(&device->dev_stats_ccnt, 0);
1141 extent_io_tree_release(&device->alloc_state);
1144 * Reset the flush error record. We might have a transient flush error
1145 * in this mount, and if so we aborted the current transaction and set
1146 * the fs to an error state, guaranteeing no super blocks can be further
1147 * committed. However that error might be transient and if we unmount the
1148 * filesystem and mount it again, we should allow the mount to succeed
1149 * (btrfs_check_rw_degradable() should not fail) - if after mounting the
1150 * filesystem again we still get flush errors, then we will again abort
1151 * any transaction and set the error state, guaranteeing no commits of
1152 * unsafe super blocks.
1154 device->last_flush_error = 0;
1156 /* Verify the device is back in a pristine state */
1157 ASSERT(!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state));
1158 ASSERT(!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state));
1159 ASSERT(list_empty(&device->dev_alloc_list));
1160 ASSERT(list_empty(&device->post_commit_list));
1161 ASSERT(atomic_read(&device->reada_in_flight) == 0);
1164 static void close_fs_devices(struct btrfs_fs_devices *fs_devices)
1166 struct btrfs_device *device, *tmp;
1168 lockdep_assert_held(&uuid_mutex);
1170 if (--fs_devices->opened > 0)
1173 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list)
1174 btrfs_close_one_device(device);
1176 WARN_ON(fs_devices->open_devices);
1177 WARN_ON(fs_devices->rw_devices);
1178 fs_devices->opened = 0;
1179 fs_devices->seeding = false;
1180 fs_devices->fs_info = NULL;
1183 void btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
1186 struct btrfs_fs_devices *tmp;
1188 mutex_lock(&uuid_mutex);
1189 close_fs_devices(fs_devices);
1190 if (!fs_devices->opened)
1191 list_splice_init(&fs_devices->seed_list, &list);
1193 list_for_each_entry_safe(fs_devices, tmp, &list, seed_list) {
1194 close_fs_devices(fs_devices);
1195 list_del(&fs_devices->seed_list);
1196 free_fs_devices(fs_devices);
1198 mutex_unlock(&uuid_mutex);
1201 static int open_fs_devices(struct btrfs_fs_devices *fs_devices,
1202 fmode_t flags, void *holder)
1204 struct btrfs_device *device;
1205 struct btrfs_device *latest_dev = NULL;
1206 struct btrfs_device *tmp_device;
1208 flags |= FMODE_EXCL;
1210 list_for_each_entry_safe(device, tmp_device, &fs_devices->devices,
1214 ret = btrfs_open_one_device(fs_devices, device, flags, holder);
1216 (!latest_dev || device->generation > latest_dev->generation)) {
1217 latest_dev = device;
1218 } else if (ret == -ENODATA) {
1219 fs_devices->num_devices--;
1220 list_del(&device->dev_list);
1221 btrfs_free_device(device);
1224 if (fs_devices->open_devices == 0)
1227 fs_devices->opened = 1;
1228 fs_devices->latest_dev = latest_dev;
1229 fs_devices->total_rw_bytes = 0;
1230 fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_REGULAR;
1231 fs_devices->read_policy = BTRFS_READ_POLICY_PID;
1236 static int devid_cmp(void *priv, const struct list_head *a,
1237 const struct list_head *b)
1239 const struct btrfs_device *dev1, *dev2;
1241 dev1 = list_entry(a, struct btrfs_device, dev_list);
1242 dev2 = list_entry(b, struct btrfs_device, dev_list);
1244 if (dev1->devid < dev2->devid)
1246 else if (dev1->devid > dev2->devid)
1251 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1252 fmode_t flags, void *holder)
1256 lockdep_assert_held(&uuid_mutex);
1258 * The device_list_mutex cannot be taken here in case opening the
1259 * underlying device takes further locks like open_mutex.
1261 * We also don't need the lock here as this is called during mount and
1262 * exclusion is provided by uuid_mutex
1265 if (fs_devices->opened) {
1266 fs_devices->opened++;
1269 list_sort(NULL, &fs_devices->devices, devid_cmp);
1270 ret = open_fs_devices(fs_devices, flags, holder);
1276 void btrfs_release_disk_super(struct btrfs_super_block *super)
1278 struct page *page = virt_to_page(super);
1283 static struct btrfs_super_block *btrfs_read_disk_super(struct block_device *bdev,
1284 u64 bytenr, u64 bytenr_orig)
1286 struct btrfs_super_block *disk_super;
1291 /* make sure our super fits in the device */
1292 if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1293 return ERR_PTR(-EINVAL);
1295 /* make sure our super fits in the page */
1296 if (sizeof(*disk_super) > PAGE_SIZE)
1297 return ERR_PTR(-EINVAL);
1299 /* make sure our super doesn't straddle pages on disk */
1300 index = bytenr >> PAGE_SHIFT;
1301 if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_SHIFT != index)
1302 return ERR_PTR(-EINVAL);
1304 /* pull in the page with our super */
1305 page = read_cache_page_gfp(bdev->bd_inode->i_mapping, index, GFP_KERNEL);
1308 return ERR_CAST(page);
1310 p = page_address(page);
1312 /* align our pointer to the offset of the super block */
1313 disk_super = p + offset_in_page(bytenr);
1315 if (btrfs_super_bytenr(disk_super) != bytenr_orig ||
1316 btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
1317 btrfs_release_disk_super(p);
1318 return ERR_PTR(-EINVAL);
1321 if (disk_super->label[0] && disk_super->label[BTRFS_LABEL_SIZE - 1])
1322 disk_super->label[BTRFS_LABEL_SIZE - 1] = 0;
1327 int btrfs_forget_devices(const char *path)
1331 mutex_lock(&uuid_mutex);
1332 ret = btrfs_free_stale_devices(strlen(path) ? path : NULL, NULL);
1333 mutex_unlock(&uuid_mutex);
1339 * Look for a btrfs signature on a device. This may be called out of the mount path
1340 * and we are not allowed to call set_blocksize during the scan. The superblock
1341 * is read via pagecache
1343 struct btrfs_device *btrfs_scan_one_device(const char *path, fmode_t flags,
1346 struct btrfs_super_block *disk_super;
1347 bool new_device_added = false;
1348 struct btrfs_device *device = NULL;
1349 struct block_device *bdev;
1350 u64 bytenr, bytenr_orig;
1353 lockdep_assert_held(&uuid_mutex);
1356 * we would like to check all the supers, but that would make
1357 * a btrfs mount succeed after a mkfs from a different FS.
1358 * So, we need to add a special mount option to scan for
1359 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1361 flags |= FMODE_EXCL;
1363 bdev = blkdev_get_by_path(path, flags, holder);
1365 return ERR_CAST(bdev);
1367 bytenr_orig = btrfs_sb_offset(0);
1368 ret = btrfs_sb_log_location_bdev(bdev, 0, READ, &bytenr);
1370 return ERR_PTR(ret);
1372 disk_super = btrfs_read_disk_super(bdev, bytenr, bytenr_orig);
1373 if (IS_ERR(disk_super)) {
1374 device = ERR_CAST(disk_super);
1375 goto error_bdev_put;
1378 device = device_list_add(path, disk_super, &new_device_added);
1379 if (!IS_ERR(device)) {
1380 if (new_device_added)
1381 btrfs_free_stale_devices(path, device);
1384 btrfs_release_disk_super(disk_super);
1387 blkdev_put(bdev, flags);
1393 * Try to find a chunk that intersects [start, start + len] range and when one
1394 * such is found, record the end of it in *start
1396 static bool contains_pending_extent(struct btrfs_device *device, u64 *start,
1399 u64 physical_start, physical_end;
1401 lockdep_assert_held(&device->fs_info->chunk_mutex);
1403 if (!find_first_extent_bit(&device->alloc_state, *start,
1404 &physical_start, &physical_end,
1405 CHUNK_ALLOCATED, NULL)) {
1407 if (in_range(physical_start, *start, len) ||
1408 in_range(*start, physical_start,
1409 physical_end - physical_start)) {
1410 *start = physical_end + 1;
1417 static u64 dev_extent_search_start(struct btrfs_device *device, u64 start)
1419 switch (device->fs_devices->chunk_alloc_policy) {
1420 case BTRFS_CHUNK_ALLOC_REGULAR:
1422 * We don't want to overwrite the superblock on the drive nor
1423 * any area used by the boot loader (grub for example), so we
1424 * make sure to start at an offset of at least 1MB.
1426 return max_t(u64, start, SZ_1M);
1427 case BTRFS_CHUNK_ALLOC_ZONED:
1429 * We don't care about the starting region like regular
1430 * allocator, because we anyway use/reserve the first two zones
1431 * for superblock logging.
1433 return ALIGN(start, device->zone_info->zone_size);
1439 static bool dev_extent_hole_check_zoned(struct btrfs_device *device,
1440 u64 *hole_start, u64 *hole_size,
1443 u64 zone_size = device->zone_info->zone_size;
1446 bool changed = false;
1448 ASSERT(IS_ALIGNED(*hole_start, zone_size));
1450 while (*hole_size > 0) {
1451 pos = btrfs_find_allocatable_zones(device, *hole_start,
1452 *hole_start + *hole_size,
1454 if (pos != *hole_start) {
1455 *hole_size = *hole_start + *hole_size - pos;
1458 if (*hole_size < num_bytes)
1462 ret = btrfs_ensure_empty_zones(device, pos, num_bytes);
1464 /* Range is ensured to be empty */
1468 /* Given hole range was invalid (outside of device) */
1469 if (ret == -ERANGE) {
1470 *hole_start += *hole_size;
1475 *hole_start += zone_size;
1476 *hole_size -= zone_size;
1484 * dev_extent_hole_check - check if specified hole is suitable for allocation
1485 * @device: the device which we have the hole
1486 * @hole_start: starting position of the hole
1487 * @hole_size: the size of the hole
1488 * @num_bytes: the size of the free space that we need
1490 * This function may modify @hole_start and @hole_size to reflect the suitable
1491 * position for allocation. Returns 1 if hole position is updated, 0 otherwise.
1493 static bool dev_extent_hole_check(struct btrfs_device *device, u64 *hole_start,
1494 u64 *hole_size, u64 num_bytes)
1496 bool changed = false;
1497 u64 hole_end = *hole_start + *hole_size;
1501 * Check before we set max_hole_start, otherwise we could end up
1502 * sending back this offset anyway.
1504 if (contains_pending_extent(device, hole_start, *hole_size)) {
1505 if (hole_end >= *hole_start)
1506 *hole_size = hole_end - *hole_start;
1512 switch (device->fs_devices->chunk_alloc_policy) {
1513 case BTRFS_CHUNK_ALLOC_REGULAR:
1514 /* No extra check */
1516 case BTRFS_CHUNK_ALLOC_ZONED:
1517 if (dev_extent_hole_check_zoned(device, hole_start,
1518 hole_size, num_bytes)) {
1521 * The changed hole can contain pending extent.
1522 * Loop again to check that.
1538 * find_free_dev_extent_start - find free space in the specified device
1539 * @device: the device which we search the free space in
1540 * @num_bytes: the size of the free space that we need
1541 * @search_start: the position from which to begin the search
1542 * @start: store the start of the free space.
1543 * @len: the size of the free space. that we find, or the size
1544 * of the max free space if we don't find suitable free space
1546 * this uses a pretty simple search, the expectation is that it is
1547 * called very infrequently and that a given device has a small number
1550 * @start is used to store the start of the free space if we find. But if we
1551 * don't find suitable free space, it will be used to store the start position
1552 * of the max free space.
1554 * @len is used to store the size of the free space that we find.
1555 * But if we don't find suitable free space, it is used to store the size of
1556 * the max free space.
1558 * NOTE: This function will search *commit* root of device tree, and does extra
1559 * check to ensure dev extents are not double allocated.
1560 * This makes the function safe to allocate dev extents but may not report
1561 * correct usable device space, as device extent freed in current transaction
1562 * is not reported as available.
1564 static int find_free_dev_extent_start(struct btrfs_device *device,
1565 u64 num_bytes, u64 search_start, u64 *start,
1568 struct btrfs_fs_info *fs_info = device->fs_info;
1569 struct btrfs_root *root = fs_info->dev_root;
1570 struct btrfs_key key;
1571 struct btrfs_dev_extent *dev_extent;
1572 struct btrfs_path *path;
1577 u64 search_end = device->total_bytes;
1580 struct extent_buffer *l;
1582 search_start = dev_extent_search_start(device, search_start);
1584 WARN_ON(device->zone_info &&
1585 !IS_ALIGNED(num_bytes, device->zone_info->zone_size));
1587 path = btrfs_alloc_path();
1591 max_hole_start = search_start;
1595 if (search_start >= search_end ||
1596 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1601 path->reada = READA_FORWARD;
1602 path->search_commit_root = 1;
1603 path->skip_locking = 1;
1605 key.objectid = device->devid;
1606 key.offset = search_start;
1607 key.type = BTRFS_DEV_EXTENT_KEY;
1609 ret = btrfs_search_backwards(root, &key, path);
1615 slot = path->slots[0];
1616 if (slot >= btrfs_header_nritems(l)) {
1617 ret = btrfs_next_leaf(root, path);
1625 btrfs_item_key_to_cpu(l, &key, slot);
1627 if (key.objectid < device->devid)
1630 if (key.objectid > device->devid)
1633 if (key.type != BTRFS_DEV_EXTENT_KEY)
1636 if (key.offset > search_start) {
1637 hole_size = key.offset - search_start;
1638 dev_extent_hole_check(device, &search_start, &hole_size,
1641 if (hole_size > max_hole_size) {
1642 max_hole_start = search_start;
1643 max_hole_size = hole_size;
1647 * If this free space is greater than which we need,
1648 * it must be the max free space that we have found
1649 * until now, so max_hole_start must point to the start
1650 * of this free space and the length of this free space
1651 * is stored in max_hole_size. Thus, we return
1652 * max_hole_start and max_hole_size and go back to the
1655 if (hole_size >= num_bytes) {
1661 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1662 extent_end = key.offset + btrfs_dev_extent_length(l,
1664 if (extent_end > search_start)
1665 search_start = extent_end;
1672 * At this point, search_start should be the end of
1673 * allocated dev extents, and when shrinking the device,
1674 * search_end may be smaller than search_start.
1676 if (search_end > search_start) {
1677 hole_size = search_end - search_start;
1678 if (dev_extent_hole_check(device, &search_start, &hole_size,
1680 btrfs_release_path(path);
1684 if (hole_size > max_hole_size) {
1685 max_hole_start = search_start;
1686 max_hole_size = hole_size;
1691 if (max_hole_size < num_bytes)
1697 btrfs_free_path(path);
1698 *start = max_hole_start;
1700 *len = max_hole_size;
1704 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
1705 u64 *start, u64 *len)
1707 /* FIXME use last free of some kind */
1708 return find_free_dev_extent_start(device, num_bytes, 0, start, len);
1711 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1712 struct btrfs_device *device,
1713 u64 start, u64 *dev_extent_len)
1715 struct btrfs_fs_info *fs_info = device->fs_info;
1716 struct btrfs_root *root = fs_info->dev_root;
1718 struct btrfs_path *path;
1719 struct btrfs_key key;
1720 struct btrfs_key found_key;
1721 struct extent_buffer *leaf = NULL;
1722 struct btrfs_dev_extent *extent = NULL;
1724 path = btrfs_alloc_path();
1728 key.objectid = device->devid;
1730 key.type = BTRFS_DEV_EXTENT_KEY;
1732 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1734 ret = btrfs_previous_item(root, path, key.objectid,
1735 BTRFS_DEV_EXTENT_KEY);
1738 leaf = path->nodes[0];
1739 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1740 extent = btrfs_item_ptr(leaf, path->slots[0],
1741 struct btrfs_dev_extent);
1742 BUG_ON(found_key.offset > start || found_key.offset +
1743 btrfs_dev_extent_length(leaf, extent) < start);
1745 btrfs_release_path(path);
1747 } else if (ret == 0) {
1748 leaf = path->nodes[0];
1749 extent = btrfs_item_ptr(leaf, path->slots[0],
1750 struct btrfs_dev_extent);
1755 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1757 ret = btrfs_del_item(trans, root, path);
1759 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1761 btrfs_free_path(path);
1765 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1767 struct extent_map_tree *em_tree;
1768 struct extent_map *em;
1772 em_tree = &fs_info->mapping_tree;
1773 read_lock(&em_tree->lock);
1774 n = rb_last(&em_tree->map.rb_root);
1776 em = rb_entry(n, struct extent_map, rb_node);
1777 ret = em->start + em->len;
1779 read_unlock(&em_tree->lock);
1784 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1788 struct btrfs_key key;
1789 struct btrfs_key found_key;
1790 struct btrfs_path *path;
1792 path = btrfs_alloc_path();
1796 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1797 key.type = BTRFS_DEV_ITEM_KEY;
1798 key.offset = (u64)-1;
1800 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1806 btrfs_err(fs_info, "corrupted chunk tree devid -1 matched");
1811 ret = btrfs_previous_item(fs_info->chunk_root, path,
1812 BTRFS_DEV_ITEMS_OBJECTID,
1813 BTRFS_DEV_ITEM_KEY);
1817 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1819 *devid_ret = found_key.offset + 1;
1823 btrfs_free_path(path);
1828 * the device information is stored in the chunk root
1829 * the btrfs_device struct should be fully filled in
1831 static int btrfs_add_dev_item(struct btrfs_trans_handle *trans,
1832 struct btrfs_device *device)
1835 struct btrfs_path *path;
1836 struct btrfs_dev_item *dev_item;
1837 struct extent_buffer *leaf;
1838 struct btrfs_key key;
1841 path = btrfs_alloc_path();
1845 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1846 key.type = BTRFS_DEV_ITEM_KEY;
1847 key.offset = device->devid;
1849 ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path,
1850 &key, sizeof(*dev_item));
1854 leaf = path->nodes[0];
1855 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1857 btrfs_set_device_id(leaf, dev_item, device->devid);
1858 btrfs_set_device_generation(leaf, dev_item, 0);
1859 btrfs_set_device_type(leaf, dev_item, device->type);
1860 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1861 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1862 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1863 btrfs_set_device_total_bytes(leaf, dev_item,
1864 btrfs_device_get_disk_total_bytes(device));
1865 btrfs_set_device_bytes_used(leaf, dev_item,
1866 btrfs_device_get_bytes_used(device));
1867 btrfs_set_device_group(leaf, dev_item, 0);
1868 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1869 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1870 btrfs_set_device_start_offset(leaf, dev_item, 0);
1872 ptr = btrfs_device_uuid(dev_item);
1873 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1874 ptr = btrfs_device_fsid(dev_item);
1875 write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid,
1876 ptr, BTRFS_FSID_SIZE);
1877 btrfs_mark_buffer_dirty(leaf);
1881 btrfs_free_path(path);
1886 * Function to update ctime/mtime for a given device path.
1887 * Mainly used for ctime/mtime based probe like libblkid.
1889 * We don't care about errors here, this is just to be kind to userspace.
1891 static void update_dev_time(const char *device_path)
1894 struct timespec64 now;
1897 ret = kern_path(device_path, LOOKUP_FOLLOW, &path);
1901 now = current_time(d_inode(path.dentry));
1902 inode_update_time(d_inode(path.dentry), &now, S_MTIME | S_CTIME);
1906 static int btrfs_rm_dev_item(struct btrfs_device *device)
1908 struct btrfs_root *root = device->fs_info->chunk_root;
1910 struct btrfs_path *path;
1911 struct btrfs_key key;
1912 struct btrfs_trans_handle *trans;
1914 path = btrfs_alloc_path();
1918 trans = btrfs_start_transaction(root, 0);
1919 if (IS_ERR(trans)) {
1920 btrfs_free_path(path);
1921 return PTR_ERR(trans);
1923 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1924 key.type = BTRFS_DEV_ITEM_KEY;
1925 key.offset = device->devid;
1927 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1931 btrfs_abort_transaction(trans, ret);
1932 btrfs_end_transaction(trans);
1936 ret = btrfs_del_item(trans, root, path);
1938 btrfs_abort_transaction(trans, ret);
1939 btrfs_end_transaction(trans);
1943 btrfs_free_path(path);
1945 ret = btrfs_commit_transaction(trans);
1950 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1951 * filesystem. It's up to the caller to adjust that number regarding eg. device
1954 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1962 seq = read_seqbegin(&fs_info->profiles_lock);
1964 all_avail = fs_info->avail_data_alloc_bits |
1965 fs_info->avail_system_alloc_bits |
1966 fs_info->avail_metadata_alloc_bits;
1967 } while (read_seqretry(&fs_info->profiles_lock, seq));
1969 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1970 if (!(all_avail & btrfs_raid_array[i].bg_flag))
1973 if (num_devices < btrfs_raid_array[i].devs_min)
1974 return btrfs_raid_array[i].mindev_error;
1980 static struct btrfs_device * btrfs_find_next_active_device(
1981 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device)
1983 struct btrfs_device *next_device;
1985 list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1986 if (next_device != device &&
1987 !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state)
1988 && next_device->bdev)
1996 * Helper function to check if the given device is part of s_bdev / latest_dev
1997 * and replace it with the provided or the next active device, in the context
1998 * where this function called, there should be always be another device (or
1999 * this_dev) which is active.
2001 void __cold btrfs_assign_next_active_device(struct btrfs_device *device,
2002 struct btrfs_device *next_device)
2004 struct btrfs_fs_info *fs_info = device->fs_info;
2007 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
2009 ASSERT(next_device);
2011 if (fs_info->sb->s_bdev &&
2012 (fs_info->sb->s_bdev == device->bdev))
2013 fs_info->sb->s_bdev = next_device->bdev;
2015 if (fs_info->fs_devices->latest_dev->bdev == device->bdev)
2016 fs_info->fs_devices->latest_dev = next_device;
2020 * Return btrfs_fs_devices::num_devices excluding the device that's being
2021 * currently replaced.
2023 static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info)
2025 u64 num_devices = fs_info->fs_devices->num_devices;
2027 down_read(&fs_info->dev_replace.rwsem);
2028 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
2029 ASSERT(num_devices > 1);
2032 up_read(&fs_info->dev_replace.rwsem);
2037 void btrfs_scratch_superblocks(struct btrfs_fs_info *fs_info,
2038 struct block_device *bdev,
2039 const char *device_path)
2041 struct btrfs_super_block *disk_super;
2047 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; copy_num++) {
2051 disk_super = btrfs_read_dev_one_super(bdev, copy_num);
2052 if (IS_ERR(disk_super))
2055 if (bdev_is_zoned(bdev)) {
2056 btrfs_reset_sb_log_zones(bdev, copy_num);
2060 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
2062 page = virt_to_page(disk_super);
2063 set_page_dirty(page);
2065 /* write_on_page() unlocks the page */
2066 ret = write_one_page(page);
2069 "error clearing superblock number %d (%d)",
2071 btrfs_release_disk_super(disk_super);
2075 /* Notify udev that device has changed */
2076 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
2078 /* Update ctime/mtime for device path for libblkid */
2079 update_dev_time(device_path);
2082 int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path,
2083 u64 devid, struct block_device **bdev, fmode_t *mode)
2085 struct btrfs_device *device;
2086 struct btrfs_fs_devices *cur_devices;
2087 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2092 * The device list in fs_devices is accessed without locks (neither
2093 * uuid_mutex nor device_list_mutex) as it won't change on a mounted
2094 * filesystem and another device rm cannot run.
2096 num_devices = btrfs_num_devices(fs_info);
2098 ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
2102 device = btrfs_find_device_by_devspec(fs_info, devid, device_path);
2104 if (IS_ERR(device)) {
2105 if (PTR_ERR(device) == -ENOENT &&
2106 device_path && strcmp(device_path, "missing") == 0)
2107 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2109 ret = PTR_ERR(device);
2113 if (btrfs_pinned_by_swapfile(fs_info, device)) {
2114 btrfs_warn_in_rcu(fs_info,
2115 "cannot remove device %s (devid %llu) due to active swapfile",
2116 rcu_str_deref(device->name), device->devid);
2121 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2122 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
2126 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
2127 fs_info->fs_devices->rw_devices == 1) {
2128 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
2132 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2133 mutex_lock(&fs_info->chunk_mutex);
2134 list_del_init(&device->dev_alloc_list);
2135 device->fs_devices->rw_devices--;
2136 mutex_unlock(&fs_info->chunk_mutex);
2139 ret = btrfs_shrink_device(device, 0);
2141 btrfs_reada_remove_dev(device);
2146 * TODO: the superblock still includes this device in its num_devices
2147 * counter although write_all_supers() is not locked out. This
2148 * could give a filesystem state which requires a degraded mount.
2150 ret = btrfs_rm_dev_item(device);
2154 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2155 btrfs_scrub_cancel_dev(device);
2158 * the device list mutex makes sure that we don't change
2159 * the device list while someone else is writing out all
2160 * the device supers. Whoever is writing all supers, should
2161 * lock the device list mutex before getting the number of
2162 * devices in the super block (super_copy). Conversely,
2163 * whoever updates the number of devices in the super block
2164 * (super_copy) should hold the device list mutex.
2168 * In normal cases the cur_devices == fs_devices. But in case
2169 * of deleting a seed device, the cur_devices should point to
2170 * its own fs_devices listed under the fs_devices->seed.
2172 cur_devices = device->fs_devices;
2173 mutex_lock(&fs_devices->device_list_mutex);
2174 list_del_rcu(&device->dev_list);
2176 cur_devices->num_devices--;
2177 cur_devices->total_devices--;
2178 /* Update total_devices of the parent fs_devices if it's seed */
2179 if (cur_devices != fs_devices)
2180 fs_devices->total_devices--;
2182 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
2183 cur_devices->missing_devices--;
2185 btrfs_assign_next_active_device(device, NULL);
2188 cur_devices->open_devices--;
2189 /* remove sysfs entry */
2190 btrfs_sysfs_remove_device(device);
2193 num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
2194 btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
2195 mutex_unlock(&fs_devices->device_list_mutex);
2198 * At this point, the device is zero sized and detached from the
2199 * devices list. All that's left is to zero out the old supers and
2202 * We cannot call btrfs_close_bdev() here because we're holding the sb
2203 * write lock, and blkdev_put() will pull in the ->open_mutex on the
2204 * block device and it's dependencies. Instead just flush the device
2205 * and let the caller do the final blkdev_put.
2207 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2208 btrfs_scratch_superblocks(fs_info, device->bdev,
2211 sync_blockdev(device->bdev);
2212 invalidate_bdev(device->bdev);
2216 *bdev = device->bdev;
2217 *mode = device->mode;
2219 btrfs_free_device(device);
2221 if (cur_devices->open_devices == 0) {
2222 list_del_init(&cur_devices->seed_list);
2223 close_fs_devices(cur_devices);
2224 free_fs_devices(cur_devices);
2231 btrfs_reada_undo_remove_dev(device);
2232 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2233 mutex_lock(&fs_info->chunk_mutex);
2234 list_add(&device->dev_alloc_list,
2235 &fs_devices->alloc_list);
2236 device->fs_devices->rw_devices++;
2237 mutex_unlock(&fs_info->chunk_mutex);
2242 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev)
2244 struct btrfs_fs_devices *fs_devices;
2246 lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex);
2249 * in case of fs with no seed, srcdev->fs_devices will point
2250 * to fs_devices of fs_info. However when the dev being replaced is
2251 * a seed dev it will point to the seed's local fs_devices. In short
2252 * srcdev will have its correct fs_devices in both the cases.
2254 fs_devices = srcdev->fs_devices;
2256 list_del_rcu(&srcdev->dev_list);
2257 list_del(&srcdev->dev_alloc_list);
2258 fs_devices->num_devices--;
2259 if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state))
2260 fs_devices->missing_devices--;
2262 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state))
2263 fs_devices->rw_devices--;
2266 fs_devices->open_devices--;
2269 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev)
2271 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2273 mutex_lock(&uuid_mutex);
2275 btrfs_close_bdev(srcdev);
2277 btrfs_free_device(srcdev);
2279 /* if this is no devs we rather delete the fs_devices */
2280 if (!fs_devices->num_devices) {
2282 * On a mounted FS, num_devices can't be zero unless it's a
2283 * seed. In case of a seed device being replaced, the replace
2284 * target added to the sprout FS, so there will be no more
2285 * device left under the seed FS.
2287 ASSERT(fs_devices->seeding);
2289 list_del_init(&fs_devices->seed_list);
2290 close_fs_devices(fs_devices);
2291 free_fs_devices(fs_devices);
2293 mutex_unlock(&uuid_mutex);
2296 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev)
2298 struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices;
2300 mutex_lock(&fs_devices->device_list_mutex);
2302 btrfs_sysfs_remove_device(tgtdev);
2305 fs_devices->open_devices--;
2307 fs_devices->num_devices--;
2309 btrfs_assign_next_active_device(tgtdev, NULL);
2311 list_del_rcu(&tgtdev->dev_list);
2313 mutex_unlock(&fs_devices->device_list_mutex);
2315 btrfs_scratch_superblocks(tgtdev->fs_info, tgtdev->bdev,
2318 btrfs_close_bdev(tgtdev);
2320 btrfs_free_device(tgtdev);
2323 static struct btrfs_device *btrfs_find_device_by_path(
2324 struct btrfs_fs_info *fs_info, const char *device_path)
2327 struct btrfs_super_block *disk_super;
2330 struct block_device *bdev;
2331 struct btrfs_device *device;
2333 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2334 fs_info->bdev_holder, 0, &bdev, &disk_super);
2336 return ERR_PTR(ret);
2338 devid = btrfs_stack_device_id(&disk_super->dev_item);
2339 dev_uuid = disk_super->dev_item.uuid;
2340 if (btrfs_fs_incompat(fs_info, METADATA_UUID))
2341 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
2342 disk_super->metadata_uuid);
2344 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
2347 btrfs_release_disk_super(disk_super);
2349 device = ERR_PTR(-ENOENT);
2350 blkdev_put(bdev, FMODE_READ);
2355 * Lookup a device given by device id, or the path if the id is 0.
2357 struct btrfs_device *btrfs_find_device_by_devspec(
2358 struct btrfs_fs_info *fs_info, u64 devid,
2359 const char *device_path)
2361 struct btrfs_device *device;
2364 device = btrfs_find_device(fs_info->fs_devices, devid, NULL,
2367 return ERR_PTR(-ENOENT);
2371 if (!device_path || !device_path[0])
2372 return ERR_PTR(-EINVAL);
2374 if (strcmp(device_path, "missing") == 0) {
2375 /* Find first missing device */
2376 list_for_each_entry(device, &fs_info->fs_devices->devices,
2378 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
2379 &device->dev_state) && !device->bdev)
2382 return ERR_PTR(-ENOENT);
2385 return btrfs_find_device_by_path(fs_info, device_path);
2389 * does all the dirty work required for changing file system's UUID.
2391 static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info)
2393 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2394 struct btrfs_fs_devices *old_devices;
2395 struct btrfs_fs_devices *seed_devices;
2396 struct btrfs_super_block *disk_super = fs_info->super_copy;
2397 struct btrfs_device *device;
2400 lockdep_assert_held(&uuid_mutex);
2401 if (!fs_devices->seeding)
2405 * Private copy of the seed devices, anchored at
2406 * fs_info->fs_devices->seed_list
2408 seed_devices = alloc_fs_devices(NULL, NULL);
2409 if (IS_ERR(seed_devices))
2410 return PTR_ERR(seed_devices);
2413 * It's necessary to retain a copy of the original seed fs_devices in
2414 * fs_uuids so that filesystems which have been seeded can successfully
2415 * reference the seed device from open_seed_devices. This also supports
2418 old_devices = clone_fs_devices(fs_devices);
2419 if (IS_ERR(old_devices)) {
2420 kfree(seed_devices);
2421 return PTR_ERR(old_devices);
2424 list_add(&old_devices->fs_list, &fs_uuids);
2426 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2427 seed_devices->opened = 1;
2428 INIT_LIST_HEAD(&seed_devices->devices);
2429 INIT_LIST_HEAD(&seed_devices->alloc_list);
2430 mutex_init(&seed_devices->device_list_mutex);
2432 mutex_lock(&fs_devices->device_list_mutex);
2433 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2435 list_for_each_entry(device, &seed_devices->devices, dev_list)
2436 device->fs_devices = seed_devices;
2438 fs_devices->seeding = false;
2439 fs_devices->num_devices = 0;
2440 fs_devices->open_devices = 0;
2441 fs_devices->missing_devices = 0;
2442 fs_devices->rotating = false;
2443 list_add(&seed_devices->seed_list, &fs_devices->seed_list);
2445 generate_random_uuid(fs_devices->fsid);
2446 memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE);
2447 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2448 mutex_unlock(&fs_devices->device_list_mutex);
2450 super_flags = btrfs_super_flags(disk_super) &
2451 ~BTRFS_SUPER_FLAG_SEEDING;
2452 btrfs_set_super_flags(disk_super, super_flags);
2458 * Store the expected generation for seed devices in device items.
2460 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans)
2462 struct btrfs_fs_info *fs_info = trans->fs_info;
2463 struct btrfs_root *root = fs_info->chunk_root;
2464 struct btrfs_path *path;
2465 struct extent_buffer *leaf;
2466 struct btrfs_dev_item *dev_item;
2467 struct btrfs_device *device;
2468 struct btrfs_key key;
2469 u8 fs_uuid[BTRFS_FSID_SIZE];
2470 u8 dev_uuid[BTRFS_UUID_SIZE];
2474 path = btrfs_alloc_path();
2478 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2480 key.type = BTRFS_DEV_ITEM_KEY;
2483 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2487 leaf = path->nodes[0];
2489 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2490 ret = btrfs_next_leaf(root, path);
2495 leaf = path->nodes[0];
2496 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2497 btrfs_release_path(path);
2501 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2502 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2503 key.type != BTRFS_DEV_ITEM_KEY)
2506 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2507 struct btrfs_dev_item);
2508 devid = btrfs_device_id(leaf, dev_item);
2509 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2511 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2513 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
2515 BUG_ON(!device); /* Logic error */
2517 if (device->fs_devices->seeding) {
2518 btrfs_set_device_generation(leaf, dev_item,
2519 device->generation);
2520 btrfs_mark_buffer_dirty(leaf);
2528 btrfs_free_path(path);
2532 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2534 struct btrfs_root *root = fs_info->dev_root;
2535 struct request_queue *q;
2536 struct btrfs_trans_handle *trans;
2537 struct btrfs_device *device;
2538 struct block_device *bdev;
2539 struct super_block *sb = fs_info->sb;
2540 struct rcu_string *name;
2541 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2542 u64 orig_super_total_bytes;
2543 u64 orig_super_num_devices;
2544 int seeding_dev = 0;
2546 bool locked = false;
2548 if (sb_rdonly(sb) && !fs_devices->seeding)
2551 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2552 fs_info->bdev_holder);
2554 return PTR_ERR(bdev);
2556 if (!btrfs_check_device_zone_type(fs_info, bdev)) {
2561 if (fs_devices->seeding) {
2563 down_write(&sb->s_umount);
2564 mutex_lock(&uuid_mutex);
2568 sync_blockdev(bdev);
2571 list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
2572 if (device->bdev == bdev) {
2580 device = btrfs_alloc_device(fs_info, NULL, NULL);
2581 if (IS_ERR(device)) {
2582 /* we can safely leave the fs_devices entry around */
2583 ret = PTR_ERR(device);
2587 name = rcu_string_strdup(device_path, GFP_KERNEL);
2590 goto error_free_device;
2592 rcu_assign_pointer(device->name, name);
2594 device->fs_info = fs_info;
2595 device->bdev = bdev;
2597 ret = btrfs_get_dev_zone_info(device);
2599 goto error_free_device;
2601 trans = btrfs_start_transaction(root, 0);
2602 if (IS_ERR(trans)) {
2603 ret = PTR_ERR(trans);
2604 goto error_free_zone;
2607 q = bdev_get_queue(bdev);
2608 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
2609 device->generation = trans->transid;
2610 device->io_width = fs_info->sectorsize;
2611 device->io_align = fs_info->sectorsize;
2612 device->sector_size = fs_info->sectorsize;
2613 device->total_bytes = round_down(i_size_read(bdev->bd_inode),
2614 fs_info->sectorsize);
2615 device->disk_total_bytes = device->total_bytes;
2616 device->commit_total_bytes = device->total_bytes;
2617 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2618 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
2619 device->mode = FMODE_EXCL;
2620 device->dev_stats_valid = 1;
2621 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2624 btrfs_clear_sb_rdonly(sb);
2625 ret = btrfs_prepare_sprout(fs_info);
2627 btrfs_abort_transaction(trans, ret);
2630 btrfs_assign_next_active_device(fs_info->fs_devices->latest_dev,
2634 device->fs_devices = fs_devices;
2636 mutex_lock(&fs_devices->device_list_mutex);
2637 mutex_lock(&fs_info->chunk_mutex);
2638 list_add_rcu(&device->dev_list, &fs_devices->devices);
2639 list_add(&device->dev_alloc_list, &fs_devices->alloc_list);
2640 fs_devices->num_devices++;
2641 fs_devices->open_devices++;
2642 fs_devices->rw_devices++;
2643 fs_devices->total_devices++;
2644 fs_devices->total_rw_bytes += device->total_bytes;
2646 atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2648 if (!blk_queue_nonrot(q))
2649 fs_devices->rotating = true;
2651 orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
2652 btrfs_set_super_total_bytes(fs_info->super_copy,
2653 round_down(orig_super_total_bytes + device->total_bytes,
2654 fs_info->sectorsize));
2656 orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy);
2657 btrfs_set_super_num_devices(fs_info->super_copy,
2658 orig_super_num_devices + 1);
2661 * we've got more storage, clear any full flags on the space
2664 btrfs_clear_space_info_full(fs_info);
2666 mutex_unlock(&fs_info->chunk_mutex);
2668 /* Add sysfs device entry */
2669 btrfs_sysfs_add_device(device);
2671 mutex_unlock(&fs_devices->device_list_mutex);
2674 mutex_lock(&fs_info->chunk_mutex);
2675 ret = init_first_rw_device(trans);
2676 mutex_unlock(&fs_info->chunk_mutex);
2678 btrfs_abort_transaction(trans, ret);
2683 ret = btrfs_add_dev_item(trans, device);
2685 btrfs_abort_transaction(trans, ret);
2690 ret = btrfs_finish_sprout(trans);
2692 btrfs_abort_transaction(trans, ret);
2697 * fs_devices now represents the newly sprouted filesystem and
2698 * its fsid has been changed by btrfs_prepare_sprout
2700 btrfs_sysfs_update_sprout_fsid(fs_devices);
2703 ret = btrfs_commit_transaction(trans);
2706 mutex_unlock(&uuid_mutex);
2707 up_write(&sb->s_umount);
2710 if (ret) /* transaction commit */
2713 ret = btrfs_relocate_sys_chunks(fs_info);
2715 btrfs_handle_fs_error(fs_info, ret,
2716 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2717 trans = btrfs_attach_transaction(root);
2718 if (IS_ERR(trans)) {
2719 if (PTR_ERR(trans) == -ENOENT)
2721 ret = PTR_ERR(trans);
2725 ret = btrfs_commit_transaction(trans);
2729 * Now that we have written a new super block to this device, check all
2730 * other fs_devices list if device_path alienates any other scanned
2732 * We can ignore the return value as it typically returns -EINVAL and
2733 * only succeeds if the device was an alien.
2735 btrfs_forget_devices(device_path);
2737 /* Update ctime/mtime for blkid or udev */
2738 update_dev_time(device_path);
2743 btrfs_sysfs_remove_device(device);
2744 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2745 mutex_lock(&fs_info->chunk_mutex);
2746 list_del_rcu(&device->dev_list);
2747 list_del(&device->dev_alloc_list);
2748 fs_info->fs_devices->num_devices--;
2749 fs_info->fs_devices->open_devices--;
2750 fs_info->fs_devices->rw_devices--;
2751 fs_info->fs_devices->total_devices--;
2752 fs_info->fs_devices->total_rw_bytes -= device->total_bytes;
2753 atomic64_sub(device->total_bytes, &fs_info->free_chunk_space);
2754 btrfs_set_super_total_bytes(fs_info->super_copy,
2755 orig_super_total_bytes);
2756 btrfs_set_super_num_devices(fs_info->super_copy,
2757 orig_super_num_devices);
2758 mutex_unlock(&fs_info->chunk_mutex);
2759 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2762 btrfs_set_sb_rdonly(sb);
2764 btrfs_end_transaction(trans);
2766 btrfs_destroy_dev_zone_info(device);
2768 btrfs_free_device(device);
2770 blkdev_put(bdev, FMODE_EXCL);
2772 mutex_unlock(&uuid_mutex);
2773 up_write(&sb->s_umount);
2778 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2779 struct btrfs_device *device)
2782 struct btrfs_path *path;
2783 struct btrfs_root *root = device->fs_info->chunk_root;
2784 struct btrfs_dev_item *dev_item;
2785 struct extent_buffer *leaf;
2786 struct btrfs_key key;
2788 path = btrfs_alloc_path();
2792 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2793 key.type = BTRFS_DEV_ITEM_KEY;
2794 key.offset = device->devid;
2796 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2805 leaf = path->nodes[0];
2806 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2808 btrfs_set_device_id(leaf, dev_item, device->devid);
2809 btrfs_set_device_type(leaf, dev_item, device->type);
2810 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2811 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2812 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2813 btrfs_set_device_total_bytes(leaf, dev_item,
2814 btrfs_device_get_disk_total_bytes(device));
2815 btrfs_set_device_bytes_used(leaf, dev_item,
2816 btrfs_device_get_bytes_used(device));
2817 btrfs_mark_buffer_dirty(leaf);
2820 btrfs_free_path(path);
2824 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2825 struct btrfs_device *device, u64 new_size)
2827 struct btrfs_fs_info *fs_info = device->fs_info;
2828 struct btrfs_super_block *super_copy = fs_info->super_copy;
2832 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2835 new_size = round_down(new_size, fs_info->sectorsize);
2837 mutex_lock(&fs_info->chunk_mutex);
2838 old_total = btrfs_super_total_bytes(super_copy);
2839 diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2841 if (new_size <= device->total_bytes ||
2842 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2843 mutex_unlock(&fs_info->chunk_mutex);
2847 btrfs_set_super_total_bytes(super_copy,
2848 round_down(old_total + diff, fs_info->sectorsize));
2849 device->fs_devices->total_rw_bytes += diff;
2851 btrfs_device_set_total_bytes(device, new_size);
2852 btrfs_device_set_disk_total_bytes(device, new_size);
2853 btrfs_clear_space_info_full(device->fs_info);
2854 if (list_empty(&device->post_commit_list))
2855 list_add_tail(&device->post_commit_list,
2856 &trans->transaction->dev_update_list);
2857 mutex_unlock(&fs_info->chunk_mutex);
2859 return btrfs_update_device(trans, device);
2862 static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
2864 struct btrfs_fs_info *fs_info = trans->fs_info;
2865 struct btrfs_root *root = fs_info->chunk_root;
2867 struct btrfs_path *path;
2868 struct btrfs_key key;
2870 path = btrfs_alloc_path();
2874 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2875 key.offset = chunk_offset;
2876 key.type = BTRFS_CHUNK_ITEM_KEY;
2878 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2881 else if (ret > 0) { /* Logic error or corruption */
2882 btrfs_handle_fs_error(fs_info, -ENOENT,
2883 "Failed lookup while freeing chunk.");
2888 ret = btrfs_del_item(trans, root, path);
2890 btrfs_handle_fs_error(fs_info, ret,
2891 "Failed to delete chunk item.");
2893 btrfs_free_path(path);
2897 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2899 struct btrfs_super_block *super_copy = fs_info->super_copy;
2900 struct btrfs_disk_key *disk_key;
2901 struct btrfs_chunk *chunk;
2908 struct btrfs_key key;
2910 lockdep_assert_held(&fs_info->chunk_mutex);
2911 array_size = btrfs_super_sys_array_size(super_copy);
2913 ptr = super_copy->sys_chunk_array;
2916 while (cur < array_size) {
2917 disk_key = (struct btrfs_disk_key *)ptr;
2918 btrfs_disk_key_to_cpu(&key, disk_key);
2920 len = sizeof(*disk_key);
2922 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2923 chunk = (struct btrfs_chunk *)(ptr + len);
2924 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2925 len += btrfs_chunk_item_size(num_stripes);
2930 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID &&
2931 key.offset == chunk_offset) {
2932 memmove(ptr, ptr + len, array_size - (cur + len));
2934 btrfs_set_super_sys_array_size(super_copy, array_size);
2944 * btrfs_get_chunk_map() - Find the mapping containing the given logical extent.
2945 * @logical: Logical block offset in bytes.
2946 * @length: Length of extent in bytes.
2948 * Return: Chunk mapping or ERR_PTR.
2950 struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info,
2951 u64 logical, u64 length)
2953 struct extent_map_tree *em_tree;
2954 struct extent_map *em;
2956 em_tree = &fs_info->mapping_tree;
2957 read_lock(&em_tree->lock);
2958 em = lookup_extent_mapping(em_tree, logical, length);
2959 read_unlock(&em_tree->lock);
2962 btrfs_crit(fs_info, "unable to find logical %llu length %llu",
2964 return ERR_PTR(-EINVAL);
2967 if (em->start > logical || em->start + em->len < logical) {
2969 "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
2970 logical, length, em->start, em->start + em->len);
2971 free_extent_map(em);
2972 return ERR_PTR(-EINVAL);
2975 /* callers are responsible for dropping em's ref. */
2979 static int remove_chunk_item(struct btrfs_trans_handle *trans,
2980 struct map_lookup *map, u64 chunk_offset)
2985 * Removing chunk items and updating the device items in the chunks btree
2986 * requires holding the chunk_mutex.
2987 * See the comment at btrfs_chunk_alloc() for the details.
2989 lockdep_assert_held(&trans->fs_info->chunk_mutex);
2991 for (i = 0; i < map->num_stripes; i++) {
2994 ret = btrfs_update_device(trans, map->stripes[i].dev);
2999 return btrfs_free_chunk(trans, chunk_offset);
3002 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
3004 struct btrfs_fs_info *fs_info = trans->fs_info;
3005 struct extent_map *em;
3006 struct map_lookup *map;
3007 u64 dev_extent_len = 0;
3009 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
3011 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
3014 * This is a logic error, but we don't want to just rely on the
3015 * user having built with ASSERT enabled, so if ASSERT doesn't
3016 * do anything we still error out.
3021 map = em->map_lookup;
3024 * First delete the device extent items from the devices btree.
3025 * We take the device_list_mutex to avoid racing with the finishing phase
3026 * of a device replace operation. See the comment below before acquiring
3027 * fs_info->chunk_mutex. Note that here we do not acquire the chunk_mutex
3028 * because that can result in a deadlock when deleting the device extent
3029 * items from the devices btree - COWing an extent buffer from the btree
3030 * may result in allocating a new metadata chunk, which would attempt to
3031 * lock again fs_info->chunk_mutex.
3033 mutex_lock(&fs_devices->device_list_mutex);
3034 for (i = 0; i < map->num_stripes; i++) {
3035 struct btrfs_device *device = map->stripes[i].dev;
3036 ret = btrfs_free_dev_extent(trans, device,
3037 map->stripes[i].physical,
3040 mutex_unlock(&fs_devices->device_list_mutex);
3041 btrfs_abort_transaction(trans, ret);
3045 if (device->bytes_used > 0) {
3046 mutex_lock(&fs_info->chunk_mutex);
3047 btrfs_device_set_bytes_used(device,
3048 device->bytes_used - dev_extent_len);
3049 atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
3050 btrfs_clear_space_info_full(fs_info);
3051 mutex_unlock(&fs_info->chunk_mutex);
3054 mutex_unlock(&fs_devices->device_list_mutex);
3057 * We acquire fs_info->chunk_mutex for 2 reasons:
3059 * 1) Just like with the first phase of the chunk allocation, we must
3060 * reserve system space, do all chunk btree updates and deletions, and
3061 * update the system chunk array in the superblock while holding this
3062 * mutex. This is for similar reasons as explained on the comment at
3063 * the top of btrfs_chunk_alloc();
3065 * 2) Prevent races with the final phase of a device replace operation
3066 * that replaces the device object associated with the map's stripes,
3067 * because the device object's id can change at any time during that
3068 * final phase of the device replace operation
3069 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the
3070 * replaced device and then see it with an ID of
3071 * BTRFS_DEV_REPLACE_DEVID, which would cause a failure when updating
3072 * the device item, which does not exists on the chunk btree.
3073 * The finishing phase of device replace acquires both the
3074 * device_list_mutex and the chunk_mutex, in that order, so we are
3075 * safe by just acquiring the chunk_mutex.
3077 trans->removing_chunk = true;
3078 mutex_lock(&fs_info->chunk_mutex);
3080 check_system_chunk(trans, map->type);
3082 ret = remove_chunk_item(trans, map, chunk_offset);
3084 * Normally we should not get -ENOSPC since we reserved space before
3085 * through the call to check_system_chunk().
3087 * Despite our system space_info having enough free space, we may not
3088 * be able to allocate extents from its block groups, because all have
3089 * an incompatible profile, which will force us to allocate a new system
3090 * block group with the right profile, or right after we called
3091 * check_system_space() above, a scrub turned the only system block group
3092 * with enough free space into RO mode.
3093 * This is explained with more detail at do_chunk_alloc().
3095 * So if we get -ENOSPC, allocate a new system chunk and retry once.
3097 if (ret == -ENOSPC) {
3098 const u64 sys_flags = btrfs_system_alloc_profile(fs_info);
3099 struct btrfs_block_group *sys_bg;
3101 sys_bg = btrfs_alloc_chunk(trans, sys_flags);
3102 if (IS_ERR(sys_bg)) {
3103 ret = PTR_ERR(sys_bg);
3104 btrfs_abort_transaction(trans, ret);
3108 ret = btrfs_chunk_alloc_add_chunk_item(trans, sys_bg);
3110 btrfs_abort_transaction(trans, ret);
3114 ret = remove_chunk_item(trans, map, chunk_offset);
3116 btrfs_abort_transaction(trans, ret);
3120 btrfs_abort_transaction(trans, ret);
3124 trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
3126 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
3127 ret = btrfs_del_sys_chunk(fs_info, chunk_offset);
3129 btrfs_abort_transaction(trans, ret);
3134 mutex_unlock(&fs_info->chunk_mutex);
3135 trans->removing_chunk = false;
3138 * We are done with chunk btree updates and deletions, so release the
3139 * system space we previously reserved (with check_system_chunk()).
3141 btrfs_trans_release_chunk_metadata(trans);
3143 ret = btrfs_remove_block_group(trans, chunk_offset, em);
3145 btrfs_abort_transaction(trans, ret);
3150 if (trans->removing_chunk) {
3151 mutex_unlock(&fs_info->chunk_mutex);
3152 trans->removing_chunk = false;
3155 free_extent_map(em);
3159 int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
3161 struct btrfs_root *root = fs_info->chunk_root;
3162 struct btrfs_trans_handle *trans;
3163 struct btrfs_block_group *block_group;
3168 * Prevent races with automatic removal of unused block groups.
3169 * After we relocate and before we remove the chunk with offset
3170 * chunk_offset, automatic removal of the block group can kick in,
3171 * resulting in a failure when calling btrfs_remove_chunk() below.
3173 * Make sure to acquire this mutex before doing a tree search (dev
3174 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
3175 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
3176 * we release the path used to search the chunk/dev tree and before
3177 * the current task acquires this mutex and calls us.
3179 lockdep_assert_held(&fs_info->reclaim_bgs_lock);
3181 /* step one, relocate all the extents inside this chunk */
3182 btrfs_scrub_pause(fs_info);
3183 ret = btrfs_relocate_block_group(fs_info, chunk_offset);
3184 btrfs_scrub_continue(fs_info);
3188 block_group = btrfs_lookup_block_group(fs_info, chunk_offset);
3191 btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group);
3192 length = block_group->length;
3193 btrfs_put_block_group(block_group);
3196 * On a zoned file system, discard the whole block group, this will
3197 * trigger a REQ_OP_ZONE_RESET operation on the device zone. If
3198 * resetting the zone fails, don't treat it as a fatal problem from the
3199 * filesystem's point of view.
3201 if (btrfs_is_zoned(fs_info)) {
3202 ret = btrfs_discard_extent(fs_info, chunk_offset, length, NULL);
3205 "failed to reset zone %llu after relocation",
3209 trans = btrfs_start_trans_remove_block_group(root->fs_info,
3211 if (IS_ERR(trans)) {
3212 ret = PTR_ERR(trans);
3213 btrfs_handle_fs_error(root->fs_info, ret, NULL);
3218 * step two, delete the device extents and the
3219 * chunk tree entries
3221 ret = btrfs_remove_chunk(trans, chunk_offset);
3222 btrfs_end_transaction(trans);
3226 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
3228 struct btrfs_root *chunk_root = fs_info->chunk_root;
3229 struct btrfs_path *path;
3230 struct extent_buffer *leaf;
3231 struct btrfs_chunk *chunk;
3232 struct btrfs_key key;
3233 struct btrfs_key found_key;
3235 bool retried = false;
3239 path = btrfs_alloc_path();
3244 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3245 key.offset = (u64)-1;
3246 key.type = BTRFS_CHUNK_ITEM_KEY;
3249 mutex_lock(&fs_info->reclaim_bgs_lock);
3250 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3252 mutex_unlock(&fs_info->reclaim_bgs_lock);
3255 BUG_ON(ret == 0); /* Corruption */
3257 ret = btrfs_previous_item(chunk_root, path, key.objectid,
3260 mutex_unlock(&fs_info->reclaim_bgs_lock);
3266 leaf = path->nodes[0];
3267 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3269 chunk = btrfs_item_ptr(leaf, path->slots[0],
3270 struct btrfs_chunk);
3271 chunk_type = btrfs_chunk_type(leaf, chunk);
3272 btrfs_release_path(path);
3274 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3275 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3281 mutex_unlock(&fs_info->reclaim_bgs_lock);
3283 if (found_key.offset == 0)
3285 key.offset = found_key.offset - 1;
3288 if (failed && !retried) {
3292 } else if (WARN_ON(failed && retried)) {
3296 btrfs_free_path(path);
3301 * return 1 : allocate a data chunk successfully,
3302 * return <0: errors during allocating a data chunk,
3303 * return 0 : no need to allocate a data chunk.
3305 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info,
3308 struct btrfs_block_group *cache;
3312 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3314 chunk_type = cache->flags;
3315 btrfs_put_block_group(cache);
3317 if (!(chunk_type & BTRFS_BLOCK_GROUP_DATA))
3320 spin_lock(&fs_info->data_sinfo->lock);
3321 bytes_used = fs_info->data_sinfo->bytes_used;
3322 spin_unlock(&fs_info->data_sinfo->lock);
3325 struct btrfs_trans_handle *trans;
3328 trans = btrfs_join_transaction(fs_info->tree_root);
3330 return PTR_ERR(trans);
3332 ret = btrfs_force_chunk_alloc(trans, BTRFS_BLOCK_GROUP_DATA);
3333 btrfs_end_transaction(trans);
3342 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3343 struct btrfs_balance_control *bctl)
3345 struct btrfs_root *root = fs_info->tree_root;
3346 struct btrfs_trans_handle *trans;
3347 struct btrfs_balance_item *item;
3348 struct btrfs_disk_balance_args disk_bargs;
3349 struct btrfs_path *path;
3350 struct extent_buffer *leaf;
3351 struct btrfs_key key;
3354 path = btrfs_alloc_path();
3358 trans = btrfs_start_transaction(root, 0);
3359 if (IS_ERR(trans)) {
3360 btrfs_free_path(path);
3361 return PTR_ERR(trans);
3364 key.objectid = BTRFS_BALANCE_OBJECTID;
3365 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3368 ret = btrfs_insert_empty_item(trans, root, path, &key,
3373 leaf = path->nodes[0];
3374 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3376 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3378 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3379 btrfs_set_balance_data(leaf, item, &disk_bargs);
3380 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3381 btrfs_set_balance_meta(leaf, item, &disk_bargs);
3382 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3383 btrfs_set_balance_sys(leaf, item, &disk_bargs);
3385 btrfs_set_balance_flags(leaf, item, bctl->flags);
3387 btrfs_mark_buffer_dirty(leaf);
3389 btrfs_free_path(path);
3390 err = btrfs_commit_transaction(trans);
3396 static int del_balance_item(struct btrfs_fs_info *fs_info)
3398 struct btrfs_root *root = fs_info->tree_root;
3399 struct btrfs_trans_handle *trans;
3400 struct btrfs_path *path;
3401 struct btrfs_key key;
3404 path = btrfs_alloc_path();
3408 trans = btrfs_start_transaction_fallback_global_rsv(root, 0);
3409 if (IS_ERR(trans)) {
3410 btrfs_free_path(path);
3411 return PTR_ERR(trans);
3414 key.objectid = BTRFS_BALANCE_OBJECTID;
3415 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3418 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3426 ret = btrfs_del_item(trans, root, path);
3428 btrfs_free_path(path);
3429 err = btrfs_commit_transaction(trans);
3436 * This is a heuristic used to reduce the number of chunks balanced on
3437 * resume after balance was interrupted.
3439 static void update_balance_args(struct btrfs_balance_control *bctl)
3442 * Turn on soft mode for chunk types that were being converted.
3444 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3445 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3446 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3447 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3448 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3449 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3452 * Turn on usage filter if is not already used. The idea is
3453 * that chunks that we have already balanced should be
3454 * reasonably full. Don't do it for chunks that are being
3455 * converted - that will keep us from relocating unconverted
3456 * (albeit full) chunks.
3458 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3459 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3460 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3461 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3462 bctl->data.usage = 90;
3464 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3465 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3466 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3467 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3468 bctl->sys.usage = 90;
3470 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3471 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3472 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3473 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3474 bctl->meta.usage = 90;
3479 * Clear the balance status in fs_info and delete the balance item from disk.
3481 static void reset_balance_state(struct btrfs_fs_info *fs_info)
3483 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3486 BUG_ON(!fs_info->balance_ctl);
3488 spin_lock(&fs_info->balance_lock);
3489 fs_info->balance_ctl = NULL;
3490 spin_unlock(&fs_info->balance_lock);
3493 ret = del_balance_item(fs_info);
3495 btrfs_handle_fs_error(fs_info, ret, NULL);
3499 * Balance filters. Return 1 if chunk should be filtered out
3500 * (should not be balanced).
3502 static int chunk_profiles_filter(u64 chunk_type,
3503 struct btrfs_balance_args *bargs)
3505 chunk_type = chunk_to_extended(chunk_type) &
3506 BTRFS_EXTENDED_PROFILE_MASK;
3508 if (bargs->profiles & chunk_type)
3514 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3515 struct btrfs_balance_args *bargs)
3517 struct btrfs_block_group *cache;
3519 u64 user_thresh_min;
3520 u64 user_thresh_max;
3523 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3524 chunk_used = cache->used;
3526 if (bargs->usage_min == 0)
3527 user_thresh_min = 0;
3529 user_thresh_min = div_factor_fine(cache->length,
3532 if (bargs->usage_max == 0)
3533 user_thresh_max = 1;
3534 else if (bargs->usage_max > 100)
3535 user_thresh_max = cache->length;
3537 user_thresh_max = div_factor_fine(cache->length,
3540 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3543 btrfs_put_block_group(cache);
3547 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3548 u64 chunk_offset, struct btrfs_balance_args *bargs)
3550 struct btrfs_block_group *cache;
3551 u64 chunk_used, user_thresh;
3554 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3555 chunk_used = cache->used;
3557 if (bargs->usage_min == 0)
3559 else if (bargs->usage > 100)
3560 user_thresh = cache->length;
3562 user_thresh = div_factor_fine(cache->length, bargs->usage);
3564 if (chunk_used < user_thresh)
3567 btrfs_put_block_group(cache);
3571 static int chunk_devid_filter(struct extent_buffer *leaf,
3572 struct btrfs_chunk *chunk,
3573 struct btrfs_balance_args *bargs)
3575 struct btrfs_stripe *stripe;
3576 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3579 for (i = 0; i < num_stripes; i++) {
3580 stripe = btrfs_stripe_nr(chunk, i);
3581 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3588 static u64 calc_data_stripes(u64 type, int num_stripes)
3590 const int index = btrfs_bg_flags_to_raid_index(type);
3591 const int ncopies = btrfs_raid_array[index].ncopies;
3592 const int nparity = btrfs_raid_array[index].nparity;
3594 return (num_stripes - nparity) / ncopies;
3597 /* [pstart, pend) */
3598 static int chunk_drange_filter(struct extent_buffer *leaf,
3599 struct btrfs_chunk *chunk,
3600 struct btrfs_balance_args *bargs)
3602 struct btrfs_stripe *stripe;
3603 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3610 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3613 type = btrfs_chunk_type(leaf, chunk);
3614 factor = calc_data_stripes(type, num_stripes);
3616 for (i = 0; i < num_stripes; i++) {
3617 stripe = btrfs_stripe_nr(chunk, i);
3618 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3621 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3622 stripe_length = btrfs_chunk_length(leaf, chunk);
3623 stripe_length = div_u64(stripe_length, factor);
3625 if (stripe_offset < bargs->pend &&
3626 stripe_offset + stripe_length > bargs->pstart)
3633 /* [vstart, vend) */
3634 static int chunk_vrange_filter(struct extent_buffer *leaf,
3635 struct btrfs_chunk *chunk,
3637 struct btrfs_balance_args *bargs)
3639 if (chunk_offset < bargs->vend &&
3640 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3641 /* at least part of the chunk is inside this vrange */
3647 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3648 struct btrfs_chunk *chunk,
3649 struct btrfs_balance_args *bargs)
3651 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3653 if (bargs->stripes_min <= num_stripes
3654 && num_stripes <= bargs->stripes_max)
3660 static int chunk_soft_convert_filter(u64 chunk_type,
3661 struct btrfs_balance_args *bargs)
3663 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3666 chunk_type = chunk_to_extended(chunk_type) &
3667 BTRFS_EXTENDED_PROFILE_MASK;
3669 if (bargs->target == chunk_type)
3675 static int should_balance_chunk(struct extent_buffer *leaf,
3676 struct btrfs_chunk *chunk, u64 chunk_offset)
3678 struct btrfs_fs_info *fs_info = leaf->fs_info;
3679 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3680 struct btrfs_balance_args *bargs = NULL;
3681 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3684 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3685 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3689 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3690 bargs = &bctl->data;
3691 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3693 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3694 bargs = &bctl->meta;
3696 /* profiles filter */
3697 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3698 chunk_profiles_filter(chunk_type, bargs)) {
3703 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3704 chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3706 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3707 chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3712 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3713 chunk_devid_filter(leaf, chunk, bargs)) {
3717 /* drange filter, makes sense only with devid filter */
3718 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3719 chunk_drange_filter(leaf, chunk, bargs)) {
3724 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3725 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3729 /* stripes filter */
3730 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3731 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3735 /* soft profile changing mode */
3736 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3737 chunk_soft_convert_filter(chunk_type, bargs)) {
3742 * limited by count, must be the last filter
3744 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3745 if (bargs->limit == 0)
3749 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3751 * Same logic as the 'limit' filter; the minimum cannot be
3752 * determined here because we do not have the global information
3753 * about the count of all chunks that satisfy the filters.
3755 if (bargs->limit_max == 0)
3764 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3766 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3767 struct btrfs_root *chunk_root = fs_info->chunk_root;
3769 struct btrfs_chunk *chunk;
3770 struct btrfs_path *path = NULL;
3771 struct btrfs_key key;
3772 struct btrfs_key found_key;
3773 struct extent_buffer *leaf;
3776 int enospc_errors = 0;
3777 bool counting = true;
3778 /* The single value limit and min/max limits use the same bytes in the */
3779 u64 limit_data = bctl->data.limit;
3780 u64 limit_meta = bctl->meta.limit;
3781 u64 limit_sys = bctl->sys.limit;
3785 int chunk_reserved = 0;
3787 path = btrfs_alloc_path();
3793 /* zero out stat counters */
3794 spin_lock(&fs_info->balance_lock);
3795 memset(&bctl->stat, 0, sizeof(bctl->stat));
3796 spin_unlock(&fs_info->balance_lock);
3800 * The single value limit and min/max limits use the same bytes
3803 bctl->data.limit = limit_data;
3804 bctl->meta.limit = limit_meta;
3805 bctl->sys.limit = limit_sys;
3807 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3808 key.offset = (u64)-1;
3809 key.type = BTRFS_CHUNK_ITEM_KEY;
3812 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3813 atomic_read(&fs_info->balance_cancel_req)) {
3818 mutex_lock(&fs_info->reclaim_bgs_lock);
3819 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3821 mutex_unlock(&fs_info->reclaim_bgs_lock);
3826 * this shouldn't happen, it means the last relocate
3830 BUG(); /* FIXME break ? */
3832 ret = btrfs_previous_item(chunk_root, path, 0,
3833 BTRFS_CHUNK_ITEM_KEY);
3835 mutex_unlock(&fs_info->reclaim_bgs_lock);
3840 leaf = path->nodes[0];
3841 slot = path->slots[0];
3842 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3844 if (found_key.objectid != key.objectid) {
3845 mutex_unlock(&fs_info->reclaim_bgs_lock);
3849 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3850 chunk_type = btrfs_chunk_type(leaf, chunk);
3853 spin_lock(&fs_info->balance_lock);
3854 bctl->stat.considered++;
3855 spin_unlock(&fs_info->balance_lock);
3858 ret = should_balance_chunk(leaf, chunk, found_key.offset);
3860 btrfs_release_path(path);
3862 mutex_unlock(&fs_info->reclaim_bgs_lock);
3867 mutex_unlock(&fs_info->reclaim_bgs_lock);
3868 spin_lock(&fs_info->balance_lock);
3869 bctl->stat.expected++;
3870 spin_unlock(&fs_info->balance_lock);
3872 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3874 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3876 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3883 * Apply limit_min filter, no need to check if the LIMITS
3884 * filter is used, limit_min is 0 by default
3886 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3887 count_data < bctl->data.limit_min)
3888 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3889 count_meta < bctl->meta.limit_min)
3890 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3891 count_sys < bctl->sys.limit_min)) {
3892 mutex_unlock(&fs_info->reclaim_bgs_lock);
3896 if (!chunk_reserved) {
3898 * We may be relocating the only data chunk we have,
3899 * which could potentially end up with losing data's
3900 * raid profile, so lets allocate an empty one in
3903 ret = btrfs_may_alloc_data_chunk(fs_info,
3906 mutex_unlock(&fs_info->reclaim_bgs_lock);
3908 } else if (ret == 1) {
3913 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3914 mutex_unlock(&fs_info->reclaim_bgs_lock);
3915 if (ret == -ENOSPC) {
3917 } else if (ret == -ETXTBSY) {
3919 "skipping relocation of block group %llu due to active swapfile",
3925 spin_lock(&fs_info->balance_lock);
3926 bctl->stat.completed++;
3927 spin_unlock(&fs_info->balance_lock);
3930 if (found_key.offset == 0)
3932 key.offset = found_key.offset - 1;
3936 btrfs_release_path(path);
3941 btrfs_free_path(path);
3942 if (enospc_errors) {
3943 btrfs_info(fs_info, "%d enospc errors during balance",
3953 * alloc_profile_is_valid - see if a given profile is valid and reduced
3954 * @flags: profile to validate
3955 * @extended: if true @flags is treated as an extended profile
3957 static int alloc_profile_is_valid(u64 flags, int extended)
3959 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3960 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3962 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3964 /* 1) check that all other bits are zeroed */
3968 /* 2) see if profile is reduced */
3970 return !extended; /* "0" is valid for usual profiles */
3972 return has_single_bit_set(flags);
3975 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3977 /* cancel requested || normal exit path */
3978 return atomic_read(&fs_info->balance_cancel_req) ||
3979 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3980 atomic_read(&fs_info->balance_cancel_req) == 0);
3984 * Validate target profile against allowed profiles and return true if it's OK.
3985 * Otherwise print the error message and return false.
3987 static inline int validate_convert_profile(struct btrfs_fs_info *fs_info,
3988 const struct btrfs_balance_args *bargs,
3989 u64 allowed, const char *type)
3991 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3994 if (fs_info->sectorsize < PAGE_SIZE &&
3995 bargs->target & BTRFS_BLOCK_GROUP_RAID56_MASK) {
3997 "RAID56 is not yet supported for sectorsize %u with page size %lu",
3998 fs_info->sectorsize, PAGE_SIZE);
4001 /* Profile is valid and does not have bits outside of the allowed set */
4002 if (alloc_profile_is_valid(bargs->target, 1) &&
4003 (bargs->target & ~allowed) == 0)
4006 btrfs_err(fs_info, "balance: invalid convert %s profile %s",
4007 type, btrfs_bg_type_to_raid_name(bargs->target));
4012 * Fill @buf with textual description of balance filter flags @bargs, up to
4013 * @size_buf including the terminating null. The output may be trimmed if it
4014 * does not fit into the provided buffer.
4016 static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf,
4020 u32 size_bp = size_buf;
4022 u64 flags = bargs->flags;
4023 char tmp_buf[128] = {'\0'};
4028 #define CHECK_APPEND_NOARG(a) \
4030 ret = snprintf(bp, size_bp, (a)); \
4031 if (ret < 0 || ret >= size_bp) \
4032 goto out_overflow; \
4037 #define CHECK_APPEND_1ARG(a, v1) \
4039 ret = snprintf(bp, size_bp, (a), (v1)); \
4040 if (ret < 0 || ret >= size_bp) \
4041 goto out_overflow; \
4046 #define CHECK_APPEND_2ARG(a, v1, v2) \
4048 ret = snprintf(bp, size_bp, (a), (v1), (v2)); \
4049 if (ret < 0 || ret >= size_bp) \
4050 goto out_overflow; \
4055 if (flags & BTRFS_BALANCE_ARGS_CONVERT)
4056 CHECK_APPEND_1ARG("convert=%s,",
4057 btrfs_bg_type_to_raid_name(bargs->target));
4059 if (flags & BTRFS_BALANCE_ARGS_SOFT)
4060 CHECK_APPEND_NOARG("soft,");
4062 if (flags & BTRFS_BALANCE_ARGS_PROFILES) {
4063 btrfs_describe_block_groups(bargs->profiles, tmp_buf,
4065 CHECK_APPEND_1ARG("profiles=%s,", tmp_buf);
4068 if (flags & BTRFS_BALANCE_ARGS_USAGE)
4069 CHECK_APPEND_1ARG("usage=%llu,", bargs->usage);
4071 if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE)
4072 CHECK_APPEND_2ARG("usage=%u..%u,",
4073 bargs->usage_min, bargs->usage_max);
4075 if (flags & BTRFS_BALANCE_ARGS_DEVID)
4076 CHECK_APPEND_1ARG("devid=%llu,", bargs->devid);
4078 if (flags & BTRFS_BALANCE_ARGS_DRANGE)
4079 CHECK_APPEND_2ARG("drange=%llu..%llu,",
4080 bargs->pstart, bargs->pend);
4082 if (flags & BTRFS_BALANCE_ARGS_VRANGE)
4083 CHECK_APPEND_2ARG("vrange=%llu..%llu,",
4084 bargs->vstart, bargs->vend);
4086 if (flags & BTRFS_BALANCE_ARGS_LIMIT)
4087 CHECK_APPEND_1ARG("limit=%llu,", bargs->limit);
4089 if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)
4090 CHECK_APPEND_2ARG("limit=%u..%u,",
4091 bargs->limit_min, bargs->limit_max);
4093 if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE)
4094 CHECK_APPEND_2ARG("stripes=%u..%u,",
4095 bargs->stripes_min, bargs->stripes_max);
4097 #undef CHECK_APPEND_2ARG
4098 #undef CHECK_APPEND_1ARG
4099 #undef CHECK_APPEND_NOARG
4103 if (size_bp < size_buf)
4104 buf[size_buf - size_bp - 1] = '\0'; /* remove last , */
4109 static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info)
4111 u32 size_buf = 1024;
4112 char tmp_buf[192] = {'\0'};
4115 u32 size_bp = size_buf;
4117 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4119 buf = kzalloc(size_buf, GFP_KERNEL);
4125 #define CHECK_APPEND_1ARG(a, v1) \
4127 ret = snprintf(bp, size_bp, (a), (v1)); \
4128 if (ret < 0 || ret >= size_bp) \
4129 goto out_overflow; \
4134 if (bctl->flags & BTRFS_BALANCE_FORCE)
4135 CHECK_APPEND_1ARG("%s", "-f ");
4137 if (bctl->flags & BTRFS_BALANCE_DATA) {
4138 describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf));
4139 CHECK_APPEND_1ARG("-d%s ", tmp_buf);
4142 if (bctl->flags & BTRFS_BALANCE_METADATA) {
4143 describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf));
4144 CHECK_APPEND_1ARG("-m%s ", tmp_buf);
4147 if (bctl->flags & BTRFS_BALANCE_SYSTEM) {
4148 describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf));
4149 CHECK_APPEND_1ARG("-s%s ", tmp_buf);
4152 #undef CHECK_APPEND_1ARG
4156 if (size_bp < size_buf)
4157 buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */
4158 btrfs_info(fs_info, "balance: %s %s",
4159 (bctl->flags & BTRFS_BALANCE_RESUME) ?
4160 "resume" : "start", buf);
4166 * Should be called with balance mutexe held
4168 int btrfs_balance(struct btrfs_fs_info *fs_info,
4169 struct btrfs_balance_control *bctl,
4170 struct btrfs_ioctl_balance_args *bargs)
4172 u64 meta_target, data_target;
4178 bool reducing_redundancy;
4181 if (btrfs_fs_closing(fs_info) ||
4182 atomic_read(&fs_info->balance_pause_req) ||
4183 btrfs_should_cancel_balance(fs_info)) {
4188 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
4189 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
4193 * In case of mixed groups both data and meta should be picked,
4194 * and identical options should be given for both of them.
4196 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
4197 if (mixed && (bctl->flags & allowed)) {
4198 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
4199 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
4200 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
4202 "balance: mixed groups data and metadata options must be the same");
4209 * rw_devices will not change at the moment, device add/delete/replace
4212 num_devices = fs_info->fs_devices->rw_devices;
4215 * SINGLE profile on-disk has no profile bit, but in-memory we have a
4216 * special bit for it, to make it easier to distinguish. Thus we need
4217 * to set it manually, or balance would refuse the profile.
4219 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
4220 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++)
4221 if (num_devices >= btrfs_raid_array[i].devs_min)
4222 allowed |= btrfs_raid_array[i].bg_flag;
4224 if (!validate_convert_profile(fs_info, &bctl->data, allowed, "data") ||
4225 !validate_convert_profile(fs_info, &bctl->meta, allowed, "metadata") ||
4226 !validate_convert_profile(fs_info, &bctl->sys, allowed, "system")) {
4232 * Allow to reduce metadata or system integrity only if force set for
4233 * profiles with redundancy (copies, parity)
4236 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) {
4237 if (btrfs_raid_array[i].ncopies >= 2 ||
4238 btrfs_raid_array[i].tolerated_failures >= 1)
4239 allowed |= btrfs_raid_array[i].bg_flag;
4242 seq = read_seqbegin(&fs_info->profiles_lock);
4244 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4245 (fs_info->avail_system_alloc_bits & allowed) &&
4246 !(bctl->sys.target & allowed)) ||
4247 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4248 (fs_info->avail_metadata_alloc_bits & allowed) &&
4249 !(bctl->meta.target & allowed)))
4250 reducing_redundancy = true;
4252 reducing_redundancy = false;
4254 /* if we're not converting, the target field is uninitialized */
4255 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4256 bctl->meta.target : fs_info->avail_metadata_alloc_bits;
4257 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4258 bctl->data.target : fs_info->avail_data_alloc_bits;
4259 } while (read_seqretry(&fs_info->profiles_lock, seq));
4261 if (reducing_redundancy) {
4262 if (bctl->flags & BTRFS_BALANCE_FORCE) {
4264 "balance: force reducing metadata redundancy");
4267 "balance: reduces metadata redundancy, use --force if you want this");
4273 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <
4274 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {
4276 "balance: metadata profile %s has lower redundancy than data profile %s",
4277 btrfs_bg_type_to_raid_name(meta_target),
4278 btrfs_bg_type_to_raid_name(data_target));
4281 ret = insert_balance_item(fs_info, bctl);
4282 if (ret && ret != -EEXIST)
4285 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
4286 BUG_ON(ret == -EEXIST);
4287 BUG_ON(fs_info->balance_ctl);
4288 spin_lock(&fs_info->balance_lock);
4289 fs_info->balance_ctl = bctl;
4290 spin_unlock(&fs_info->balance_lock);
4292 BUG_ON(ret != -EEXIST);
4293 spin_lock(&fs_info->balance_lock);
4294 update_balance_args(bctl);
4295 spin_unlock(&fs_info->balance_lock);
4298 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4299 set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4300 describe_balance_start_or_resume(fs_info);
4301 mutex_unlock(&fs_info->balance_mutex);
4303 ret = __btrfs_balance(fs_info);
4305 mutex_lock(&fs_info->balance_mutex);
4306 if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req))
4307 btrfs_info(fs_info, "balance: paused");
4309 * Balance can be canceled by:
4311 * - Regular cancel request
4312 * Then ret == -ECANCELED and balance_cancel_req > 0
4314 * - Fatal signal to "btrfs" process
4315 * Either the signal caught by wait_reserve_ticket() and callers
4316 * got -EINTR, or caught by btrfs_should_cancel_balance() and
4318 * Either way, in this case balance_cancel_req = 0, and
4319 * ret == -EINTR or ret == -ECANCELED.
4321 * So here we only check the return value to catch canceled balance.
4323 else if (ret == -ECANCELED || ret == -EINTR)
4324 btrfs_info(fs_info, "balance: canceled");
4326 btrfs_info(fs_info, "balance: ended with status: %d", ret);
4328 clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4331 memset(bargs, 0, sizeof(*bargs));
4332 btrfs_update_ioctl_balance_args(fs_info, bargs);
4335 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
4336 balance_need_close(fs_info)) {
4337 reset_balance_state(fs_info);
4338 btrfs_exclop_finish(fs_info);
4341 wake_up(&fs_info->balance_wait_q);
4345 if (bctl->flags & BTRFS_BALANCE_RESUME)
4346 reset_balance_state(fs_info);
4349 btrfs_exclop_finish(fs_info);
4354 static int balance_kthread(void *data)
4356 struct btrfs_fs_info *fs_info = data;
4359 mutex_lock(&fs_info->balance_mutex);
4360 if (fs_info->balance_ctl)
4361 ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL);
4362 mutex_unlock(&fs_info->balance_mutex);
4367 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
4369 struct task_struct *tsk;
4371 mutex_lock(&fs_info->balance_mutex);
4372 if (!fs_info->balance_ctl) {
4373 mutex_unlock(&fs_info->balance_mutex);
4376 mutex_unlock(&fs_info->balance_mutex);
4378 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
4379 btrfs_info(fs_info, "balance: resume skipped");
4384 * A ro->rw remount sequence should continue with the paused balance
4385 * regardless of who pauses it, system or the user as of now, so set
4388 spin_lock(&fs_info->balance_lock);
4389 fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME;
4390 spin_unlock(&fs_info->balance_lock);
4392 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
4393 return PTR_ERR_OR_ZERO(tsk);
4396 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
4398 struct btrfs_balance_control *bctl;
4399 struct btrfs_balance_item *item;
4400 struct btrfs_disk_balance_args disk_bargs;
4401 struct btrfs_path *path;
4402 struct extent_buffer *leaf;
4403 struct btrfs_key key;
4406 path = btrfs_alloc_path();
4410 key.objectid = BTRFS_BALANCE_OBJECTID;
4411 key.type = BTRFS_TEMPORARY_ITEM_KEY;
4414 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4417 if (ret > 0) { /* ret = -ENOENT; */
4422 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4428 leaf = path->nodes[0];
4429 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
4431 bctl->flags = btrfs_balance_flags(leaf, item);
4432 bctl->flags |= BTRFS_BALANCE_RESUME;
4434 btrfs_balance_data(leaf, item, &disk_bargs);
4435 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4436 btrfs_balance_meta(leaf, item, &disk_bargs);
4437 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4438 btrfs_balance_sys(leaf, item, &disk_bargs);
4439 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4442 * This should never happen, as the paused balance state is recovered
4443 * during mount without any chance of other exclusive ops to collide.
4445 * This gives the exclusive op status to balance and keeps in paused
4446 * state until user intervention (cancel or umount). If the ownership
4447 * cannot be assigned, show a message but do not fail. The balance
4448 * is in a paused state and must have fs_info::balance_ctl properly
4451 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE))
4453 "balance: cannot set exclusive op status, resume manually");
4455 btrfs_release_path(path);
4457 mutex_lock(&fs_info->balance_mutex);
4458 BUG_ON(fs_info->balance_ctl);
4459 spin_lock(&fs_info->balance_lock);
4460 fs_info->balance_ctl = bctl;
4461 spin_unlock(&fs_info->balance_lock);
4462 mutex_unlock(&fs_info->balance_mutex);
4464 btrfs_free_path(path);
4468 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4472 mutex_lock(&fs_info->balance_mutex);
4473 if (!fs_info->balance_ctl) {
4474 mutex_unlock(&fs_info->balance_mutex);
4478 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4479 atomic_inc(&fs_info->balance_pause_req);
4480 mutex_unlock(&fs_info->balance_mutex);
4482 wait_event(fs_info->balance_wait_q,
4483 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4485 mutex_lock(&fs_info->balance_mutex);
4486 /* we are good with balance_ctl ripped off from under us */
4487 BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4488 atomic_dec(&fs_info->balance_pause_req);
4493 mutex_unlock(&fs_info->balance_mutex);
4497 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4499 mutex_lock(&fs_info->balance_mutex);
4500 if (!fs_info->balance_ctl) {
4501 mutex_unlock(&fs_info->balance_mutex);
4506 * A paused balance with the item stored on disk can be resumed at
4507 * mount time if the mount is read-write. Otherwise it's still paused
4508 * and we must not allow cancelling as it deletes the item.
4510 if (sb_rdonly(fs_info->sb)) {
4511 mutex_unlock(&fs_info->balance_mutex);
4515 atomic_inc(&fs_info->balance_cancel_req);
4517 * if we are running just wait and return, balance item is
4518 * deleted in btrfs_balance in this case
4520 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4521 mutex_unlock(&fs_info->balance_mutex);
4522 wait_event(fs_info->balance_wait_q,
4523 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4524 mutex_lock(&fs_info->balance_mutex);
4526 mutex_unlock(&fs_info->balance_mutex);
4528 * Lock released to allow other waiters to continue, we'll
4529 * reexamine the status again.
4531 mutex_lock(&fs_info->balance_mutex);
4533 if (fs_info->balance_ctl) {
4534 reset_balance_state(fs_info);
4535 btrfs_exclop_finish(fs_info);
4536 btrfs_info(fs_info, "balance: canceled");
4540 BUG_ON(fs_info->balance_ctl ||
4541 test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4542 atomic_dec(&fs_info->balance_cancel_req);
4543 mutex_unlock(&fs_info->balance_mutex);
4547 int btrfs_uuid_scan_kthread(void *data)
4549 struct btrfs_fs_info *fs_info = data;
4550 struct btrfs_root *root = fs_info->tree_root;
4551 struct btrfs_key key;
4552 struct btrfs_path *path = NULL;
4554 struct extent_buffer *eb;
4556 struct btrfs_root_item root_item;
4558 struct btrfs_trans_handle *trans = NULL;
4559 bool closing = false;
4561 path = btrfs_alloc_path();
4568 key.type = BTRFS_ROOT_ITEM_KEY;
4572 if (btrfs_fs_closing(fs_info)) {
4576 ret = btrfs_search_forward(root, &key, path,
4577 BTRFS_OLDEST_GENERATION);
4584 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4585 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4586 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4587 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4590 eb = path->nodes[0];
4591 slot = path->slots[0];
4592 item_size = btrfs_item_size_nr(eb, slot);
4593 if (item_size < sizeof(root_item))
4596 read_extent_buffer(eb, &root_item,
4597 btrfs_item_ptr_offset(eb, slot),
4598 (int)sizeof(root_item));
4599 if (btrfs_root_refs(&root_item) == 0)
4602 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4603 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4607 btrfs_release_path(path);
4609 * 1 - subvol uuid item
4610 * 1 - received_subvol uuid item
4612 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4613 if (IS_ERR(trans)) {
4614 ret = PTR_ERR(trans);
4622 btrfs_release_path(path);
4623 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4624 ret = btrfs_uuid_tree_add(trans, root_item.uuid,
4625 BTRFS_UUID_KEY_SUBVOL,
4628 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4634 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4635 ret = btrfs_uuid_tree_add(trans,
4636 root_item.received_uuid,
4637 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4640 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4647 btrfs_release_path(path);
4649 ret = btrfs_end_transaction(trans);
4655 if (key.offset < (u64)-1) {
4657 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4659 key.type = BTRFS_ROOT_ITEM_KEY;
4660 } else if (key.objectid < (u64)-1) {
4662 key.type = BTRFS_ROOT_ITEM_KEY;
4671 btrfs_free_path(path);
4672 if (trans && !IS_ERR(trans))
4673 btrfs_end_transaction(trans);
4675 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4677 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4678 up(&fs_info->uuid_tree_rescan_sem);
4682 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4684 struct btrfs_trans_handle *trans;
4685 struct btrfs_root *tree_root = fs_info->tree_root;
4686 struct btrfs_root *uuid_root;
4687 struct task_struct *task;
4694 trans = btrfs_start_transaction(tree_root, 2);
4696 return PTR_ERR(trans);
4698 uuid_root = btrfs_create_tree(trans, BTRFS_UUID_TREE_OBJECTID);
4699 if (IS_ERR(uuid_root)) {
4700 ret = PTR_ERR(uuid_root);
4701 btrfs_abort_transaction(trans, ret);
4702 btrfs_end_transaction(trans);
4706 fs_info->uuid_root = uuid_root;
4708 ret = btrfs_commit_transaction(trans);
4712 down(&fs_info->uuid_tree_rescan_sem);
4713 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4715 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4716 btrfs_warn(fs_info, "failed to start uuid_scan task");
4717 up(&fs_info->uuid_tree_rescan_sem);
4718 return PTR_ERR(task);
4725 * shrinking a device means finding all of the device extents past
4726 * the new size, and then following the back refs to the chunks.
4727 * The chunk relocation code actually frees the device extent
4729 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4731 struct btrfs_fs_info *fs_info = device->fs_info;
4732 struct btrfs_root *root = fs_info->dev_root;
4733 struct btrfs_trans_handle *trans;
4734 struct btrfs_dev_extent *dev_extent = NULL;
4735 struct btrfs_path *path;
4741 bool retried = false;
4742 struct extent_buffer *l;
4743 struct btrfs_key key;
4744 struct btrfs_super_block *super_copy = fs_info->super_copy;
4745 u64 old_total = btrfs_super_total_bytes(super_copy);
4746 u64 old_size = btrfs_device_get_total_bytes(device);
4750 new_size = round_down(new_size, fs_info->sectorsize);
4752 diff = round_down(old_size - new_size, fs_info->sectorsize);
4754 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4757 path = btrfs_alloc_path();
4761 path->reada = READA_BACK;
4763 trans = btrfs_start_transaction(root, 0);
4764 if (IS_ERR(trans)) {
4765 btrfs_free_path(path);
4766 return PTR_ERR(trans);
4769 mutex_lock(&fs_info->chunk_mutex);
4771 btrfs_device_set_total_bytes(device, new_size);
4772 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4773 device->fs_devices->total_rw_bytes -= diff;
4774 atomic64_sub(diff, &fs_info->free_chunk_space);
4778 * Once the device's size has been set to the new size, ensure all
4779 * in-memory chunks are synced to disk so that the loop below sees them
4780 * and relocates them accordingly.
4782 if (contains_pending_extent(device, &start, diff)) {
4783 mutex_unlock(&fs_info->chunk_mutex);
4784 ret = btrfs_commit_transaction(trans);
4788 mutex_unlock(&fs_info->chunk_mutex);
4789 btrfs_end_transaction(trans);
4793 key.objectid = device->devid;
4794 key.offset = (u64)-1;
4795 key.type = BTRFS_DEV_EXTENT_KEY;
4798 mutex_lock(&fs_info->reclaim_bgs_lock);
4799 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4801 mutex_unlock(&fs_info->reclaim_bgs_lock);
4805 ret = btrfs_previous_item(root, path, 0, key.type);
4807 mutex_unlock(&fs_info->reclaim_bgs_lock);
4811 btrfs_release_path(path);
4816 slot = path->slots[0];
4817 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4819 if (key.objectid != device->devid) {
4820 mutex_unlock(&fs_info->reclaim_bgs_lock);
4821 btrfs_release_path(path);
4825 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4826 length = btrfs_dev_extent_length(l, dev_extent);
4828 if (key.offset + length <= new_size) {
4829 mutex_unlock(&fs_info->reclaim_bgs_lock);
4830 btrfs_release_path(path);
4834 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4835 btrfs_release_path(path);
4838 * We may be relocating the only data chunk we have,
4839 * which could potentially end up with losing data's
4840 * raid profile, so lets allocate an empty one in
4843 ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset);
4845 mutex_unlock(&fs_info->reclaim_bgs_lock);
4849 ret = btrfs_relocate_chunk(fs_info, chunk_offset);
4850 mutex_unlock(&fs_info->reclaim_bgs_lock);
4851 if (ret == -ENOSPC) {
4854 if (ret == -ETXTBSY) {
4856 "could not shrink block group %llu due to active swapfile",
4861 } while (key.offset-- > 0);
4863 if (failed && !retried) {
4867 } else if (failed && retried) {
4872 /* Shrinking succeeded, else we would be at "done". */
4873 trans = btrfs_start_transaction(root, 0);
4874 if (IS_ERR(trans)) {
4875 ret = PTR_ERR(trans);
4879 mutex_lock(&fs_info->chunk_mutex);
4880 /* Clear all state bits beyond the shrunk device size */
4881 clear_extent_bits(&device->alloc_state, new_size, (u64)-1,
4884 btrfs_device_set_disk_total_bytes(device, new_size);
4885 if (list_empty(&device->post_commit_list))
4886 list_add_tail(&device->post_commit_list,
4887 &trans->transaction->dev_update_list);
4889 WARN_ON(diff > old_total);
4890 btrfs_set_super_total_bytes(super_copy,
4891 round_down(old_total - diff, fs_info->sectorsize));
4892 mutex_unlock(&fs_info->chunk_mutex);
4894 /* Now btrfs_update_device() will change the on-disk size. */
4895 ret = btrfs_update_device(trans, device);
4897 btrfs_abort_transaction(trans, ret);
4898 btrfs_end_transaction(trans);
4900 ret = btrfs_commit_transaction(trans);
4903 btrfs_free_path(path);
4905 mutex_lock(&fs_info->chunk_mutex);
4906 btrfs_device_set_total_bytes(device, old_size);
4907 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
4908 device->fs_devices->total_rw_bytes += diff;
4909 atomic64_add(diff, &fs_info->free_chunk_space);
4910 mutex_unlock(&fs_info->chunk_mutex);
4915 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
4916 struct btrfs_key *key,
4917 struct btrfs_chunk *chunk, int item_size)
4919 struct btrfs_super_block *super_copy = fs_info->super_copy;
4920 struct btrfs_disk_key disk_key;
4924 lockdep_assert_held(&fs_info->chunk_mutex);
4926 array_size = btrfs_super_sys_array_size(super_copy);
4927 if (array_size + item_size + sizeof(disk_key)
4928 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
4931 ptr = super_copy->sys_chunk_array + array_size;
4932 btrfs_cpu_key_to_disk(&disk_key, key);
4933 memcpy(ptr, &disk_key, sizeof(disk_key));
4934 ptr += sizeof(disk_key);
4935 memcpy(ptr, chunk, item_size);
4936 item_size += sizeof(disk_key);
4937 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4943 * sort the devices in descending order by max_avail, total_avail
4945 static int btrfs_cmp_device_info(const void *a, const void *b)
4947 const struct btrfs_device_info *di_a = a;
4948 const struct btrfs_device_info *di_b = b;
4950 if (di_a->max_avail > di_b->max_avail)
4952 if (di_a->max_avail < di_b->max_avail)
4954 if (di_a->total_avail > di_b->total_avail)
4956 if (di_a->total_avail < di_b->total_avail)
4961 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4963 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4966 btrfs_set_fs_incompat(info, RAID56);
4969 static void check_raid1c34_incompat_flag(struct btrfs_fs_info *info, u64 type)
4971 if (!(type & (BTRFS_BLOCK_GROUP_RAID1C3 | BTRFS_BLOCK_GROUP_RAID1C4)))
4974 btrfs_set_fs_incompat(info, RAID1C34);
4978 * Structure used internally for __btrfs_alloc_chunk() function.
4979 * Wraps needed parameters.
4981 struct alloc_chunk_ctl {
4984 /* Total number of stripes to allocate */
4986 /* sub_stripes info for map */
4988 /* Stripes per device */
4990 /* Maximum number of devices to use */
4992 /* Minimum number of devices to use */
4994 /* ndevs has to be a multiple of this */
4996 /* Number of copies */
4998 /* Number of stripes worth of bytes to store parity information */
5000 u64 max_stripe_size;
5008 static void init_alloc_chunk_ctl_policy_regular(
5009 struct btrfs_fs_devices *fs_devices,
5010 struct alloc_chunk_ctl *ctl)
5012 u64 type = ctl->type;
5014 if (type & BTRFS_BLOCK_GROUP_DATA) {
5015 ctl->max_stripe_size = SZ_1G;
5016 ctl->max_chunk_size = BTRFS_MAX_DATA_CHUNK_SIZE;
5017 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
5018 /* For larger filesystems, use larger metadata chunks */
5019 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
5020 ctl->max_stripe_size = SZ_1G;
5022 ctl->max_stripe_size = SZ_256M;
5023 ctl->max_chunk_size = ctl->max_stripe_size;
5024 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
5025 ctl->max_stripe_size = SZ_32M;
5026 ctl->max_chunk_size = 2 * ctl->max_stripe_size;
5027 ctl->devs_max = min_t(int, ctl->devs_max,
5028 BTRFS_MAX_DEVS_SYS_CHUNK);
5033 /* We don't want a chunk larger than 10% of writable space */
5034 ctl->max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
5035 ctl->max_chunk_size);
5036 ctl->dev_extent_min = BTRFS_STRIPE_LEN * ctl->dev_stripes;
5039 static void init_alloc_chunk_ctl_policy_zoned(
5040 struct btrfs_fs_devices *fs_devices,
5041 struct alloc_chunk_ctl *ctl)
5043 u64 zone_size = fs_devices->fs_info->zone_size;
5045 int min_num_stripes = ctl->devs_min * ctl->dev_stripes;
5046 int min_data_stripes = (min_num_stripes - ctl->nparity) / ctl->ncopies;
5047 u64 min_chunk_size = min_data_stripes * zone_size;
5048 u64 type = ctl->type;
5050 ctl->max_stripe_size = zone_size;
5051 if (type & BTRFS_BLOCK_GROUP_DATA) {
5052 ctl->max_chunk_size = round_down(BTRFS_MAX_DATA_CHUNK_SIZE,
5054 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
5055 ctl->max_chunk_size = ctl->max_stripe_size;
5056 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
5057 ctl->max_chunk_size = 2 * ctl->max_stripe_size;
5058 ctl->devs_max = min_t(int, ctl->devs_max,
5059 BTRFS_MAX_DEVS_SYS_CHUNK);
5064 /* We don't want a chunk larger than 10% of writable space */
5065 limit = max(round_down(div_factor(fs_devices->total_rw_bytes, 1),
5068 ctl->max_chunk_size = min(limit, ctl->max_chunk_size);
5069 ctl->dev_extent_min = zone_size * ctl->dev_stripes;
5072 static void init_alloc_chunk_ctl(struct btrfs_fs_devices *fs_devices,
5073 struct alloc_chunk_ctl *ctl)
5075 int index = btrfs_bg_flags_to_raid_index(ctl->type);
5077 ctl->sub_stripes = btrfs_raid_array[index].sub_stripes;
5078 ctl->dev_stripes = btrfs_raid_array[index].dev_stripes;
5079 ctl->devs_max = btrfs_raid_array[index].devs_max;
5081 ctl->devs_max = BTRFS_MAX_DEVS(fs_devices->fs_info);
5082 ctl->devs_min = btrfs_raid_array[index].devs_min;
5083 ctl->devs_increment = btrfs_raid_array[index].devs_increment;
5084 ctl->ncopies = btrfs_raid_array[index].ncopies;
5085 ctl->nparity = btrfs_raid_array[index].nparity;
5088 switch (fs_devices->chunk_alloc_policy) {
5089 case BTRFS_CHUNK_ALLOC_REGULAR:
5090 init_alloc_chunk_ctl_policy_regular(fs_devices, ctl);
5092 case BTRFS_CHUNK_ALLOC_ZONED:
5093 init_alloc_chunk_ctl_policy_zoned(fs_devices, ctl);
5100 static int gather_device_info(struct btrfs_fs_devices *fs_devices,
5101 struct alloc_chunk_ctl *ctl,
5102 struct btrfs_device_info *devices_info)
5104 struct btrfs_fs_info *info = fs_devices->fs_info;
5105 struct btrfs_device *device;
5107 u64 dev_extent_want = ctl->max_stripe_size * ctl->dev_stripes;
5114 * in the first pass through the devices list, we gather information
5115 * about the available holes on each device.
5117 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
5118 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
5120 "BTRFS: read-only device in alloc_list\n");
5124 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
5125 &device->dev_state) ||
5126 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
5129 if (device->total_bytes > device->bytes_used)
5130 total_avail = device->total_bytes - device->bytes_used;
5134 /* If there is no space on this device, skip it. */
5135 if (total_avail < ctl->dev_extent_min)
5138 ret = find_free_dev_extent(device, dev_extent_want, &dev_offset,
5140 if (ret && ret != -ENOSPC)
5144 max_avail = dev_extent_want;
5146 if (max_avail < ctl->dev_extent_min) {
5147 if (btrfs_test_opt(info, ENOSPC_DEBUG))
5149 "%s: devid %llu has no free space, have=%llu want=%llu",
5150 __func__, device->devid, max_avail,
5151 ctl->dev_extent_min);
5155 if (ndevs == fs_devices->rw_devices) {
5156 WARN(1, "%s: found more than %llu devices\n",
5157 __func__, fs_devices->rw_devices);
5160 devices_info[ndevs].dev_offset = dev_offset;
5161 devices_info[ndevs].max_avail = max_avail;
5162 devices_info[ndevs].total_avail = total_avail;
5163 devices_info[ndevs].dev = device;
5169 * now sort the devices by hole size / available space
5171 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
5172 btrfs_cmp_device_info, NULL);
5177 static int decide_stripe_size_regular(struct alloc_chunk_ctl *ctl,
5178 struct btrfs_device_info *devices_info)
5180 /* Number of stripes that count for block group size */
5184 * The primary goal is to maximize the number of stripes, so use as
5185 * many devices as possible, even if the stripes are not maximum sized.
5187 * The DUP profile stores more than one stripe per device, the
5188 * max_avail is the total size so we have to adjust.
5190 ctl->stripe_size = div_u64(devices_info[ctl->ndevs - 1].max_avail,
5192 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5194 /* This will have to be fixed for RAID1 and RAID10 over more drives */
5195 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5198 * Use the number of data stripes to figure out how big this chunk is
5199 * really going to be in terms of logical address space, and compare
5200 * that answer with the max chunk size. If it's higher, we try to
5201 * reduce stripe_size.
5203 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) {
5205 * Reduce stripe_size, round it up to a 16MB boundary again and
5206 * then use it, unless it ends up being even bigger than the
5207 * previous value we had already.
5209 ctl->stripe_size = min(round_up(div_u64(ctl->max_chunk_size,
5210 data_stripes), SZ_16M),
5214 /* Align to BTRFS_STRIPE_LEN */
5215 ctl->stripe_size = round_down(ctl->stripe_size, BTRFS_STRIPE_LEN);
5216 ctl->chunk_size = ctl->stripe_size * data_stripes;
5221 static int decide_stripe_size_zoned(struct alloc_chunk_ctl *ctl,
5222 struct btrfs_device_info *devices_info)
5224 u64 zone_size = devices_info[0].dev->zone_info->zone_size;
5225 /* Number of stripes that count for block group size */
5229 * It should hold because:
5230 * dev_extent_min == dev_extent_want == zone_size * dev_stripes
5232 ASSERT(devices_info[ctl->ndevs - 1].max_avail == ctl->dev_extent_min);
5234 ctl->stripe_size = zone_size;
5235 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5236 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5238 /* stripe_size is fixed in zoned filesysmte. Reduce ndevs instead. */
5239 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) {
5240 ctl->ndevs = div_u64(div_u64(ctl->max_chunk_size * ctl->ncopies,
5241 ctl->stripe_size) + ctl->nparity,
5243 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5244 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5245 ASSERT(ctl->stripe_size * data_stripes <= ctl->max_chunk_size);
5248 ctl->chunk_size = ctl->stripe_size * data_stripes;
5253 static int decide_stripe_size(struct btrfs_fs_devices *fs_devices,
5254 struct alloc_chunk_ctl *ctl,
5255 struct btrfs_device_info *devices_info)
5257 struct btrfs_fs_info *info = fs_devices->fs_info;
5260 * Round down to number of usable stripes, devs_increment can be any
5261 * number so we can't use round_down() that requires power of 2, while
5262 * rounddown is safe.
5264 ctl->ndevs = rounddown(ctl->ndevs, ctl->devs_increment);
5266 if (ctl->ndevs < ctl->devs_min) {
5267 if (btrfs_test_opt(info, ENOSPC_DEBUG)) {
5269 "%s: not enough devices with free space: have=%d minimum required=%d",
5270 __func__, ctl->ndevs, ctl->devs_min);
5275 ctl->ndevs = min(ctl->ndevs, ctl->devs_max);
5277 switch (fs_devices->chunk_alloc_policy) {
5278 case BTRFS_CHUNK_ALLOC_REGULAR:
5279 return decide_stripe_size_regular(ctl, devices_info);
5280 case BTRFS_CHUNK_ALLOC_ZONED:
5281 return decide_stripe_size_zoned(ctl, devices_info);
5287 static struct btrfs_block_group *create_chunk(struct btrfs_trans_handle *trans,
5288 struct alloc_chunk_ctl *ctl,
5289 struct btrfs_device_info *devices_info)
5291 struct btrfs_fs_info *info = trans->fs_info;
5292 struct map_lookup *map = NULL;
5293 struct extent_map_tree *em_tree;
5294 struct btrfs_block_group *block_group;
5295 struct extent_map *em;
5296 u64 start = ctl->start;
5297 u64 type = ctl->type;
5302 map = kmalloc(map_lookup_size(ctl->num_stripes), GFP_NOFS);
5304 return ERR_PTR(-ENOMEM);
5305 map->num_stripes = ctl->num_stripes;
5307 for (i = 0; i < ctl->ndevs; ++i) {
5308 for (j = 0; j < ctl->dev_stripes; ++j) {
5309 int s = i * ctl->dev_stripes + j;
5310 map->stripes[s].dev = devices_info[i].dev;
5311 map->stripes[s].physical = devices_info[i].dev_offset +
5312 j * ctl->stripe_size;
5315 map->stripe_len = BTRFS_STRIPE_LEN;
5316 map->io_align = BTRFS_STRIPE_LEN;
5317 map->io_width = BTRFS_STRIPE_LEN;
5319 map->sub_stripes = ctl->sub_stripes;
5321 trace_btrfs_chunk_alloc(info, map, start, ctl->chunk_size);
5323 em = alloc_extent_map();
5326 return ERR_PTR(-ENOMEM);
5328 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
5329 em->map_lookup = map;
5331 em->len = ctl->chunk_size;
5332 em->block_start = 0;
5333 em->block_len = em->len;
5334 em->orig_block_len = ctl->stripe_size;
5336 em_tree = &info->mapping_tree;
5337 write_lock(&em_tree->lock);
5338 ret = add_extent_mapping(em_tree, em, 0);
5340 write_unlock(&em_tree->lock);
5341 free_extent_map(em);
5342 return ERR_PTR(ret);
5344 write_unlock(&em_tree->lock);
5346 block_group = btrfs_make_block_group(trans, 0, type, start, ctl->chunk_size);
5347 if (IS_ERR(block_group))
5348 goto error_del_extent;
5350 for (i = 0; i < map->num_stripes; i++) {
5351 struct btrfs_device *dev = map->stripes[i].dev;
5353 btrfs_device_set_bytes_used(dev,
5354 dev->bytes_used + ctl->stripe_size);
5355 if (list_empty(&dev->post_commit_list))
5356 list_add_tail(&dev->post_commit_list,
5357 &trans->transaction->dev_update_list);
5360 atomic64_sub(ctl->stripe_size * map->num_stripes,
5361 &info->free_chunk_space);
5363 free_extent_map(em);
5364 check_raid56_incompat_flag(info, type);
5365 check_raid1c34_incompat_flag(info, type);
5370 write_lock(&em_tree->lock);
5371 remove_extent_mapping(em_tree, em);
5372 write_unlock(&em_tree->lock);
5374 /* One for our allocation */
5375 free_extent_map(em);
5376 /* One for the tree reference */
5377 free_extent_map(em);
5382 struct btrfs_block_group *btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
5385 struct btrfs_fs_info *info = trans->fs_info;
5386 struct btrfs_fs_devices *fs_devices = info->fs_devices;
5387 struct btrfs_device_info *devices_info = NULL;
5388 struct alloc_chunk_ctl ctl;
5389 struct btrfs_block_group *block_group;
5392 lockdep_assert_held(&info->chunk_mutex);
5394 if (!alloc_profile_is_valid(type, 0)) {
5396 return ERR_PTR(-EINVAL);
5399 if (list_empty(&fs_devices->alloc_list)) {
5400 if (btrfs_test_opt(info, ENOSPC_DEBUG))
5401 btrfs_debug(info, "%s: no writable device", __func__);
5402 return ERR_PTR(-ENOSPC);
5405 if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
5406 btrfs_err(info, "invalid chunk type 0x%llx requested", type);
5408 return ERR_PTR(-EINVAL);
5411 ctl.start = find_next_chunk(info);
5413 init_alloc_chunk_ctl(fs_devices, &ctl);
5415 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
5418 return ERR_PTR(-ENOMEM);
5420 ret = gather_device_info(fs_devices, &ctl, devices_info);
5422 block_group = ERR_PTR(ret);
5426 ret = decide_stripe_size(fs_devices, &ctl, devices_info);
5428 block_group = ERR_PTR(ret);
5432 block_group = create_chunk(trans, &ctl, devices_info);
5435 kfree(devices_info);
5440 * This function, btrfs_chunk_alloc_add_chunk_item(), typically belongs to the
5441 * phase 1 of chunk allocation. It belongs to phase 2 only when allocating system
5444 * See the comment at btrfs_chunk_alloc() for details about the chunk allocation
5447 int btrfs_chunk_alloc_add_chunk_item(struct btrfs_trans_handle *trans,
5448 struct btrfs_block_group *bg)
5450 struct btrfs_fs_info *fs_info = trans->fs_info;
5451 struct btrfs_root *extent_root = fs_info->extent_root;
5452 struct btrfs_root *chunk_root = fs_info->chunk_root;
5453 struct btrfs_key key;
5454 struct btrfs_chunk *chunk;
5455 struct btrfs_stripe *stripe;
5456 struct extent_map *em;
5457 struct map_lookup *map;
5463 * We take the chunk_mutex for 2 reasons:
5465 * 1) Updates and insertions in the chunk btree must be done while holding
5466 * the chunk_mutex, as well as updating the system chunk array in the
5467 * superblock. See the comment on top of btrfs_chunk_alloc() for the
5470 * 2) To prevent races with the final phase of a device replace operation
5471 * that replaces the device object associated with the map's stripes,
5472 * because the device object's id can change at any time during that
5473 * final phase of the device replace operation
5474 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the
5475 * replaced device and then see it with an ID of BTRFS_DEV_REPLACE_DEVID,
5476 * which would cause a failure when updating the device item, which does
5477 * not exists, or persisting a stripe of the chunk item with such ID.
5478 * Here we can't use the device_list_mutex because our caller already
5479 * has locked the chunk_mutex, and the final phase of device replace
5480 * acquires both mutexes - first the device_list_mutex and then the
5481 * chunk_mutex. Using any of those two mutexes protects us from a
5482 * concurrent device replace.
5484 lockdep_assert_held(&fs_info->chunk_mutex);
5486 em = btrfs_get_chunk_map(fs_info, bg->start, bg->length);
5489 btrfs_abort_transaction(trans, ret);
5493 map = em->map_lookup;
5494 item_size = btrfs_chunk_item_size(map->num_stripes);
5496 chunk = kzalloc(item_size, GFP_NOFS);
5499 btrfs_abort_transaction(trans, ret);
5503 for (i = 0; i < map->num_stripes; i++) {
5504 struct btrfs_device *device = map->stripes[i].dev;
5506 ret = btrfs_update_device(trans, device);
5511 stripe = &chunk->stripe;
5512 for (i = 0; i < map->num_stripes; i++) {
5513 struct btrfs_device *device = map->stripes[i].dev;
5514 const u64 dev_offset = map->stripes[i].physical;
5516 btrfs_set_stack_stripe_devid(stripe, device->devid);
5517 btrfs_set_stack_stripe_offset(stripe, dev_offset);
5518 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
5522 btrfs_set_stack_chunk_length(chunk, bg->length);
5523 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
5524 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
5525 btrfs_set_stack_chunk_type(chunk, map->type);
5526 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
5527 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
5528 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
5529 btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
5530 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
5532 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
5533 key.type = BTRFS_CHUNK_ITEM_KEY;
5534 key.offset = bg->start;
5536 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
5540 bg->chunk_item_inserted = 1;
5542 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
5543 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
5550 free_extent_map(em);
5554 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans)
5556 struct btrfs_fs_info *fs_info = trans->fs_info;
5558 struct btrfs_block_group *meta_bg;
5559 struct btrfs_block_group *sys_bg;
5562 * When adding a new device for sprouting, the seed device is read-only
5563 * so we must first allocate a metadata and a system chunk. But before
5564 * adding the block group items to the extent, device and chunk btrees,
5567 * 1) Create both chunks without doing any changes to the btrees, as
5568 * otherwise we would get -ENOSPC since the block groups from the
5569 * seed device are read-only;
5571 * 2) Add the device item for the new sprout device - finishing the setup
5572 * of a new block group requires updating the device item in the chunk
5573 * btree, so it must exist when we attempt to do it. The previous step
5574 * ensures this does not fail with -ENOSPC.
5576 * After that we can add the block group items to their btrees:
5577 * update existing device item in the chunk btree, add a new block group
5578 * item to the extent btree, add a new chunk item to the chunk btree and
5579 * finally add the new device extent items to the devices btree.
5582 alloc_profile = btrfs_metadata_alloc_profile(fs_info);
5583 meta_bg = btrfs_alloc_chunk(trans, alloc_profile);
5584 if (IS_ERR(meta_bg))
5585 return PTR_ERR(meta_bg);
5587 alloc_profile = btrfs_system_alloc_profile(fs_info);
5588 sys_bg = btrfs_alloc_chunk(trans, alloc_profile);
5590 return PTR_ERR(sys_bg);
5595 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
5597 const int index = btrfs_bg_flags_to_raid_index(map->type);
5599 return btrfs_raid_array[index].tolerated_failures;
5602 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset)
5604 struct extent_map *em;
5605 struct map_lookup *map;
5610 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
5614 map = em->map_lookup;
5615 for (i = 0; i < map->num_stripes; i++) {
5616 if (test_bit(BTRFS_DEV_STATE_MISSING,
5617 &map->stripes[i].dev->dev_state)) {
5621 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE,
5622 &map->stripes[i].dev->dev_state)) {
5629 * If the number of missing devices is larger than max errors,
5630 * we can not write the data into that chunk successfully, so
5633 if (miss_ndevs > btrfs_chunk_max_errors(map))
5636 free_extent_map(em);
5640 void btrfs_mapping_tree_free(struct extent_map_tree *tree)
5642 struct extent_map *em;
5645 write_lock(&tree->lock);
5646 em = lookup_extent_mapping(tree, 0, (u64)-1);
5648 remove_extent_mapping(tree, em);
5649 write_unlock(&tree->lock);
5653 free_extent_map(em);
5654 /* once for the tree */
5655 free_extent_map(em);
5659 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5661 struct extent_map *em;
5662 struct map_lookup *map;
5665 em = btrfs_get_chunk_map(fs_info, logical, len);
5668 * We could return errors for these cases, but that could get
5669 * ugly and we'd probably do the same thing which is just not do
5670 * anything else and exit, so return 1 so the callers don't try
5671 * to use other copies.
5675 map = em->map_lookup;
5676 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1_MASK))
5677 ret = map->num_stripes;
5678 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5679 ret = map->sub_stripes;
5680 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5682 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5684 * There could be two corrupted data stripes, we need
5685 * to loop retry in order to rebuild the correct data.
5687 * Fail a stripe at a time on every retry except the
5688 * stripe under reconstruction.
5690 ret = map->num_stripes;
5693 free_extent_map(em);
5695 down_read(&fs_info->dev_replace.rwsem);
5696 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) &&
5697 fs_info->dev_replace.tgtdev)
5699 up_read(&fs_info->dev_replace.rwsem);
5704 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
5707 struct extent_map *em;
5708 struct map_lookup *map;
5709 unsigned long len = fs_info->sectorsize;
5711 em = btrfs_get_chunk_map(fs_info, logical, len);
5713 if (!WARN_ON(IS_ERR(em))) {
5714 map = em->map_lookup;
5715 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5716 len = map->stripe_len * nr_data_stripes(map);
5717 free_extent_map(em);
5722 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5724 struct extent_map *em;
5725 struct map_lookup *map;
5728 em = btrfs_get_chunk_map(fs_info, logical, len);
5730 if(!WARN_ON(IS_ERR(em))) {
5731 map = em->map_lookup;
5732 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5734 free_extent_map(em);
5739 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5740 struct map_lookup *map, int first,
5741 int dev_replace_is_ongoing)
5745 int preferred_mirror;
5747 struct btrfs_device *srcdev;
5750 (BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10)));
5752 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5753 num_stripes = map->sub_stripes;
5755 num_stripes = map->num_stripes;
5757 switch (fs_info->fs_devices->read_policy) {
5759 /* Shouldn't happen, just warn and use pid instead of failing */
5760 btrfs_warn_rl(fs_info,
5761 "unknown read_policy type %u, reset to pid",
5762 fs_info->fs_devices->read_policy);
5763 fs_info->fs_devices->read_policy = BTRFS_READ_POLICY_PID;
5765 case BTRFS_READ_POLICY_PID:
5766 preferred_mirror = first + (current->pid % num_stripes);
5770 if (dev_replace_is_ongoing &&
5771 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5772 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5773 srcdev = fs_info->dev_replace.srcdev;
5778 * try to avoid the drive that is the source drive for a
5779 * dev-replace procedure, only choose it if no other non-missing
5780 * mirror is available
5782 for (tolerance = 0; tolerance < 2; tolerance++) {
5783 if (map->stripes[preferred_mirror].dev->bdev &&
5784 (tolerance || map->stripes[preferred_mirror].dev != srcdev))
5785 return preferred_mirror;
5786 for (i = first; i < first + num_stripes; i++) {
5787 if (map->stripes[i].dev->bdev &&
5788 (tolerance || map->stripes[i].dev != srcdev))
5793 /* we couldn't find one that doesn't fail. Just return something
5794 * and the io error handling code will clean up eventually
5796 return preferred_mirror;
5799 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5800 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5807 for (i = 0; i < num_stripes - 1; i++) {
5808 /* Swap if parity is on a smaller index */
5809 if (bbio->raid_map[i] > bbio->raid_map[i + 1]) {
5810 swap(bbio->stripes[i], bbio->stripes[i + 1]);
5811 swap(bbio->raid_map[i], bbio->raid_map[i + 1]);
5818 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5820 struct btrfs_bio *bbio = kzalloc(
5821 /* the size of the btrfs_bio */
5822 sizeof(struct btrfs_bio) +
5823 /* plus the variable array for the stripes */
5824 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5825 /* plus the variable array for the tgt dev */
5826 sizeof(int) * (real_stripes) +
5828 * plus the raid_map, which includes both the tgt dev
5831 sizeof(u64) * (total_stripes),
5832 GFP_NOFS|__GFP_NOFAIL);
5834 atomic_set(&bbio->error, 0);
5835 refcount_set(&bbio->refs, 1);
5837 bbio->tgtdev_map = (int *)(bbio->stripes + total_stripes);
5838 bbio->raid_map = (u64 *)(bbio->tgtdev_map + real_stripes);
5843 void btrfs_get_bbio(struct btrfs_bio *bbio)
5845 WARN_ON(!refcount_read(&bbio->refs));
5846 refcount_inc(&bbio->refs);
5849 void btrfs_put_bbio(struct btrfs_bio *bbio)
5853 if (refcount_dec_and_test(&bbio->refs))
5857 /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */
5859 * Please note that, discard won't be sent to target device of device
5862 static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info,
5863 u64 logical, u64 *length_ret,
5864 struct btrfs_bio **bbio_ret)
5866 struct extent_map *em;
5867 struct map_lookup *map;
5868 struct btrfs_bio *bbio;
5869 u64 length = *length_ret;
5873 u64 stripe_end_offset;
5880 u32 sub_stripes = 0;
5881 u64 stripes_per_dev = 0;
5882 u32 remaining_stripes = 0;
5883 u32 last_stripe = 0;
5887 /* discard always return a bbio */
5890 em = btrfs_get_chunk_map(fs_info, logical, length);
5894 map = em->map_lookup;
5895 /* we don't discard raid56 yet */
5896 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5901 offset = logical - em->start;
5902 length = min_t(u64, em->start + em->len - logical, length);
5903 *length_ret = length;
5905 stripe_len = map->stripe_len;
5907 * stripe_nr counts the total number of stripes we have to stride
5908 * to get to this block
5910 stripe_nr = div64_u64(offset, stripe_len);
5912 /* stripe_offset is the offset of this block in its stripe */
5913 stripe_offset = offset - stripe_nr * stripe_len;
5915 stripe_nr_end = round_up(offset + length, map->stripe_len);
5916 stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len);
5917 stripe_cnt = stripe_nr_end - stripe_nr;
5918 stripe_end_offset = stripe_nr_end * map->stripe_len -
5921 * after this, stripe_nr is the number of stripes on this
5922 * device we have to walk to find the data, and stripe_index is
5923 * the number of our device in the stripe array
5927 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5928 BTRFS_BLOCK_GROUP_RAID10)) {
5929 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5932 sub_stripes = map->sub_stripes;
5934 factor = map->num_stripes / sub_stripes;
5935 num_stripes = min_t(u64, map->num_stripes,
5936 sub_stripes * stripe_cnt);
5937 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5938 stripe_index *= sub_stripes;
5939 stripes_per_dev = div_u64_rem(stripe_cnt, factor,
5940 &remaining_stripes);
5941 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5942 last_stripe *= sub_stripes;
5943 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1_MASK |
5944 BTRFS_BLOCK_GROUP_DUP)) {
5945 num_stripes = map->num_stripes;
5947 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5951 bbio = alloc_btrfs_bio(num_stripes, 0);
5957 for (i = 0; i < num_stripes; i++) {
5958 bbio->stripes[i].physical =
5959 map->stripes[stripe_index].physical +
5960 stripe_offset + stripe_nr * map->stripe_len;
5961 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5963 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5964 BTRFS_BLOCK_GROUP_RAID10)) {
5965 bbio->stripes[i].length = stripes_per_dev *
5968 if (i / sub_stripes < remaining_stripes)
5969 bbio->stripes[i].length +=
5973 * Special for the first stripe and
5976 * |-------|...|-------|
5980 if (i < sub_stripes)
5981 bbio->stripes[i].length -=
5984 if (stripe_index >= last_stripe &&
5985 stripe_index <= (last_stripe +
5987 bbio->stripes[i].length -=
5990 if (i == sub_stripes - 1)
5993 bbio->stripes[i].length = length;
5997 if (stripe_index == map->num_stripes) {
6004 bbio->map_type = map->type;
6005 bbio->num_stripes = num_stripes;
6007 free_extent_map(em);
6012 * In dev-replace case, for repair case (that's the only case where the mirror
6013 * is selected explicitly when calling btrfs_map_block), blocks left of the
6014 * left cursor can also be read from the target drive.
6016 * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the
6018 * For READ, it also needs to be supported using the same mirror number.
6020 * If the requested block is not left of the left cursor, EIO is returned. This
6021 * can happen because btrfs_num_copies() returns one more in the dev-replace
6024 static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info,
6025 u64 logical, u64 length,
6026 u64 srcdev_devid, int *mirror_num,
6029 struct btrfs_bio *bbio = NULL;
6031 int index_srcdev = 0;
6033 u64 physical_of_found = 0;
6037 ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS,
6038 logical, &length, &bbio, 0, 0);
6040 ASSERT(bbio == NULL);
6044 num_stripes = bbio->num_stripes;
6045 if (*mirror_num > num_stripes) {
6047 * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror,
6048 * that means that the requested area is not left of the left
6051 btrfs_put_bbio(bbio);
6056 * process the rest of the function using the mirror_num of the source
6057 * drive. Therefore look it up first. At the end, patch the device
6058 * pointer to the one of the target drive.
6060 for (i = 0; i < num_stripes; i++) {
6061 if (bbio->stripes[i].dev->devid != srcdev_devid)
6065 * In case of DUP, in order to keep it simple, only add the
6066 * mirror with the lowest physical address
6069 physical_of_found <= bbio->stripes[i].physical)
6074 physical_of_found = bbio->stripes[i].physical;
6077 btrfs_put_bbio(bbio);
6083 *mirror_num = index_srcdev + 1;
6084 *physical = physical_of_found;
6088 static bool is_block_group_to_copy(struct btrfs_fs_info *fs_info, u64 logical)
6090 struct btrfs_block_group *cache;
6093 /* Non zoned filesystem does not use "to_copy" flag */
6094 if (!btrfs_is_zoned(fs_info))
6097 cache = btrfs_lookup_block_group(fs_info, logical);
6099 spin_lock(&cache->lock);
6100 ret = cache->to_copy;
6101 spin_unlock(&cache->lock);
6103 btrfs_put_block_group(cache);
6107 static void handle_ops_on_dev_replace(enum btrfs_map_op op,
6108 struct btrfs_bio **bbio_ret,
6109 struct btrfs_dev_replace *dev_replace,
6111 int *num_stripes_ret, int *max_errors_ret)
6113 struct btrfs_bio *bbio = *bbio_ret;
6114 u64 srcdev_devid = dev_replace->srcdev->devid;
6115 int tgtdev_indexes = 0;
6116 int num_stripes = *num_stripes_ret;
6117 int max_errors = *max_errors_ret;
6120 if (op == BTRFS_MAP_WRITE) {
6121 int index_where_to_add;
6124 * A block group which have "to_copy" set will eventually
6125 * copied by dev-replace process. We can avoid cloning IO here.
6127 if (is_block_group_to_copy(dev_replace->srcdev->fs_info, logical))
6131 * duplicate the write operations while the dev replace
6132 * procedure is running. Since the copying of the old disk to
6133 * the new disk takes place at run time while the filesystem is
6134 * mounted writable, the regular write operations to the old
6135 * disk have to be duplicated to go to the new disk as well.
6137 * Note that device->missing is handled by the caller, and that
6138 * the write to the old disk is already set up in the stripes
6141 index_where_to_add = num_stripes;
6142 for (i = 0; i < num_stripes; i++) {
6143 if (bbio->stripes[i].dev->devid == srcdev_devid) {
6144 /* write to new disk, too */
6145 struct btrfs_bio_stripe *new =
6146 bbio->stripes + index_where_to_add;
6147 struct btrfs_bio_stripe *old =
6150 new->physical = old->physical;
6151 new->length = old->length;
6152 new->dev = dev_replace->tgtdev;
6153 bbio->tgtdev_map[i] = index_where_to_add;
6154 index_where_to_add++;
6159 num_stripes = index_where_to_add;
6160 } else if (op == BTRFS_MAP_GET_READ_MIRRORS) {
6161 int index_srcdev = 0;
6163 u64 physical_of_found = 0;
6166 * During the dev-replace procedure, the target drive can also
6167 * be used to read data in case it is needed to repair a corrupt
6168 * block elsewhere. This is possible if the requested area is
6169 * left of the left cursor. In this area, the target drive is a
6170 * full copy of the source drive.
6172 for (i = 0; i < num_stripes; i++) {
6173 if (bbio->stripes[i].dev->devid == srcdev_devid) {
6175 * In case of DUP, in order to keep it simple,
6176 * only add the mirror with the lowest physical
6180 physical_of_found <=
6181 bbio->stripes[i].physical)
6185 physical_of_found = bbio->stripes[i].physical;
6189 struct btrfs_bio_stripe *tgtdev_stripe =
6190 bbio->stripes + num_stripes;
6192 tgtdev_stripe->physical = physical_of_found;
6193 tgtdev_stripe->length =
6194 bbio->stripes[index_srcdev].length;
6195 tgtdev_stripe->dev = dev_replace->tgtdev;
6196 bbio->tgtdev_map[index_srcdev] = num_stripes;
6203 *num_stripes_ret = num_stripes;
6204 *max_errors_ret = max_errors;
6205 bbio->num_tgtdevs = tgtdev_indexes;
6209 static bool need_full_stripe(enum btrfs_map_op op)
6211 return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS);
6215 * Calculate the geometry of a particular (address, len) tuple. This
6216 * information is used to calculate how big a particular bio can get before it
6217 * straddles a stripe.
6219 * @fs_info: the filesystem
6220 * @em: mapping containing the logical extent
6221 * @op: type of operation - write or read
6222 * @logical: address that we want to figure out the geometry of
6223 * @io_geom: pointer used to return values
6225 * Returns < 0 in case a chunk for the given logical address cannot be found,
6226 * usually shouldn't happen unless @logical is corrupted, 0 otherwise.
6228 int btrfs_get_io_geometry(struct btrfs_fs_info *fs_info, struct extent_map *em,
6229 enum btrfs_map_op op, u64 logical,
6230 struct btrfs_io_geometry *io_geom)
6232 struct map_lookup *map;
6238 u64 raid56_full_stripe_start = (u64)-1;
6241 ASSERT(op != BTRFS_MAP_DISCARD);
6243 map = em->map_lookup;
6244 /* Offset of this logical address in the chunk */
6245 offset = logical - em->start;
6246 /* Len of a stripe in a chunk */
6247 stripe_len = map->stripe_len;
6248 /* Stripe where this block falls in */
6249 stripe_nr = div64_u64(offset, stripe_len);
6250 /* Offset of stripe in the chunk */
6251 stripe_offset = stripe_nr * stripe_len;
6252 if (offset < stripe_offset) {
6254 "stripe math has gone wrong, stripe_offset=%llu offset=%llu start=%llu logical=%llu stripe_len=%llu",
6255 stripe_offset, offset, em->start, logical, stripe_len);
6259 /* stripe_offset is the offset of this block in its stripe */
6260 stripe_offset = offset - stripe_offset;
6261 data_stripes = nr_data_stripes(map);
6263 if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
6264 u64 max_len = stripe_len - stripe_offset;
6267 * In case of raid56, we need to know the stripe aligned start
6269 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6270 unsigned long full_stripe_len = stripe_len * data_stripes;
6271 raid56_full_stripe_start = offset;
6274 * Allow a write of a full stripe, but make sure we
6275 * don't allow straddling of stripes
6277 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
6279 raid56_full_stripe_start *= full_stripe_len;
6282 * For writes to RAID[56], allow a full stripeset across
6283 * all disks. For other RAID types and for RAID[56]
6284 * reads, just allow a single stripe (on a single disk).
6286 if (op == BTRFS_MAP_WRITE) {
6287 max_len = stripe_len * data_stripes -
6288 (offset - raid56_full_stripe_start);
6291 len = min_t(u64, em->len - offset, max_len);
6293 len = em->len - offset;
6297 io_geom->offset = offset;
6298 io_geom->stripe_len = stripe_len;
6299 io_geom->stripe_nr = stripe_nr;
6300 io_geom->stripe_offset = stripe_offset;
6301 io_geom->raid56_stripe_offset = raid56_full_stripe_start;
6306 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
6307 enum btrfs_map_op op,
6308 u64 logical, u64 *length,
6309 struct btrfs_bio **bbio_ret,
6310 int mirror_num, int need_raid_map)
6312 struct extent_map *em;
6313 struct map_lookup *map;
6323 int tgtdev_indexes = 0;
6324 struct btrfs_bio *bbio = NULL;
6325 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
6326 int dev_replace_is_ongoing = 0;
6327 int num_alloc_stripes;
6328 int patch_the_first_stripe_for_dev_replace = 0;
6329 u64 physical_to_patch_in_first_stripe = 0;
6330 u64 raid56_full_stripe_start = (u64)-1;
6331 struct btrfs_io_geometry geom;
6334 ASSERT(op != BTRFS_MAP_DISCARD);
6336 em = btrfs_get_chunk_map(fs_info, logical, *length);
6337 ASSERT(!IS_ERR(em));
6339 ret = btrfs_get_io_geometry(fs_info, em, op, logical, &geom);
6343 map = em->map_lookup;
6346 stripe_len = geom.stripe_len;
6347 stripe_nr = geom.stripe_nr;
6348 stripe_offset = geom.stripe_offset;
6349 raid56_full_stripe_start = geom.raid56_stripe_offset;
6350 data_stripes = nr_data_stripes(map);
6352 down_read(&dev_replace->rwsem);
6353 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
6355 * Hold the semaphore for read during the whole operation, write is
6356 * requested at commit time but must wait.
6358 if (!dev_replace_is_ongoing)
6359 up_read(&dev_replace->rwsem);
6361 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
6362 !need_full_stripe(op) && dev_replace->tgtdev != NULL) {
6363 ret = get_extra_mirror_from_replace(fs_info, logical, *length,
6364 dev_replace->srcdev->devid,
6366 &physical_to_patch_in_first_stripe);
6370 patch_the_first_stripe_for_dev_replace = 1;
6371 } else if (mirror_num > map->num_stripes) {
6377 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
6378 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
6380 if (!need_full_stripe(op))
6382 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1_MASK) {
6383 if (need_full_stripe(op))
6384 num_stripes = map->num_stripes;
6385 else if (mirror_num)
6386 stripe_index = mirror_num - 1;
6388 stripe_index = find_live_mirror(fs_info, map, 0,
6389 dev_replace_is_ongoing);
6390 mirror_num = stripe_index + 1;
6393 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
6394 if (need_full_stripe(op)) {
6395 num_stripes = map->num_stripes;
6396 } else if (mirror_num) {
6397 stripe_index = mirror_num - 1;
6402 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
6403 u32 factor = map->num_stripes / map->sub_stripes;
6405 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
6406 stripe_index *= map->sub_stripes;
6408 if (need_full_stripe(op))
6409 num_stripes = map->sub_stripes;
6410 else if (mirror_num)
6411 stripe_index += mirror_num - 1;
6413 int old_stripe_index = stripe_index;
6414 stripe_index = find_live_mirror(fs_info, map,
6416 dev_replace_is_ongoing);
6417 mirror_num = stripe_index - old_stripe_index + 1;
6420 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6421 if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) {
6422 /* push stripe_nr back to the start of the full stripe */
6423 stripe_nr = div64_u64(raid56_full_stripe_start,
6424 stripe_len * data_stripes);
6426 /* RAID[56] write or recovery. Return all stripes */
6427 num_stripes = map->num_stripes;
6428 max_errors = nr_parity_stripes(map);
6430 *length = map->stripe_len;
6435 * Mirror #0 or #1 means the original data block.
6436 * Mirror #2 is RAID5 parity block.
6437 * Mirror #3 is RAID6 Q block.
6439 stripe_nr = div_u64_rem(stripe_nr,
6440 data_stripes, &stripe_index);
6442 stripe_index = data_stripes + mirror_num - 2;
6444 /* We distribute the parity blocks across stripes */
6445 div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
6447 if (!need_full_stripe(op) && mirror_num <= 1)
6452 * after this, stripe_nr is the number of stripes on this
6453 * device we have to walk to find the data, and stripe_index is
6454 * the number of our device in the stripe array
6456 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
6458 mirror_num = stripe_index + 1;
6460 if (stripe_index >= map->num_stripes) {
6462 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
6463 stripe_index, map->num_stripes);
6468 num_alloc_stripes = num_stripes;
6469 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) {
6470 if (op == BTRFS_MAP_WRITE)
6471 num_alloc_stripes <<= 1;
6472 if (op == BTRFS_MAP_GET_READ_MIRRORS)
6473 num_alloc_stripes++;
6474 tgtdev_indexes = num_stripes;
6477 bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
6483 for (i = 0; i < num_stripes; i++) {
6484 bbio->stripes[i].physical = map->stripes[stripe_index].physical +
6485 stripe_offset + stripe_nr * map->stripe_len;
6486 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
6490 /* build raid_map */
6491 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map &&
6492 (need_full_stripe(op) || mirror_num > 1)) {
6496 /* Work out the disk rotation on this stripe-set */
6497 div_u64_rem(stripe_nr, num_stripes, &rot);
6499 /* Fill in the logical address of each stripe */
6500 tmp = stripe_nr * data_stripes;
6501 for (i = 0; i < data_stripes; i++)
6502 bbio->raid_map[(i+rot) % num_stripes] =
6503 em->start + (tmp + i) * map->stripe_len;
6505 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
6506 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
6507 bbio->raid_map[(i+rot+1) % num_stripes] =
6510 sort_parity_stripes(bbio, num_stripes);
6513 if (need_full_stripe(op))
6514 max_errors = btrfs_chunk_max_errors(map);
6516 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
6517 need_full_stripe(op)) {
6518 handle_ops_on_dev_replace(op, &bbio, dev_replace, logical,
6519 &num_stripes, &max_errors);
6523 bbio->map_type = map->type;
6524 bbio->num_stripes = num_stripes;
6525 bbio->max_errors = max_errors;
6526 bbio->mirror_num = mirror_num;
6529 * this is the case that REQ_READ && dev_replace_is_ongoing &&
6530 * mirror_num == num_stripes + 1 && dev_replace target drive is
6531 * available as a mirror
6533 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
6534 WARN_ON(num_stripes > 1);
6535 bbio->stripes[0].dev = dev_replace->tgtdev;
6536 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
6537 bbio->mirror_num = map->num_stripes + 1;
6540 if (dev_replace_is_ongoing) {
6541 lockdep_assert_held(&dev_replace->rwsem);
6542 /* Unlock and let waiting writers proceed */
6543 up_read(&dev_replace->rwsem);
6545 free_extent_map(em);
6549 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6550 u64 logical, u64 *length,
6551 struct btrfs_bio **bbio_ret, int mirror_num)
6553 if (op == BTRFS_MAP_DISCARD)
6554 return __btrfs_map_block_for_discard(fs_info, logical,
6557 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret,
6561 /* For Scrub/replace */
6562 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6563 u64 logical, u64 *length,
6564 struct btrfs_bio **bbio_ret)
6566 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1);
6569 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
6571 bio->bi_private = bbio->private;
6572 bio->bi_end_io = bbio->end_io;
6575 btrfs_put_bbio(bbio);
6578 static void btrfs_end_bio(struct bio *bio)
6580 struct btrfs_bio *bbio = bio->bi_private;
6581 int is_orig_bio = 0;
6583 if (bio->bi_status) {
6584 atomic_inc(&bbio->error);
6585 if (bio->bi_status == BLK_STS_IOERR ||
6586 bio->bi_status == BLK_STS_TARGET) {
6587 struct btrfs_device *dev = btrfs_io_bio(bio)->device;
6590 if (btrfs_op(bio) == BTRFS_MAP_WRITE)
6591 btrfs_dev_stat_inc_and_print(dev,
6592 BTRFS_DEV_STAT_WRITE_ERRS);
6593 else if (!(bio->bi_opf & REQ_RAHEAD))
6594 btrfs_dev_stat_inc_and_print(dev,
6595 BTRFS_DEV_STAT_READ_ERRS);
6596 if (bio->bi_opf & REQ_PREFLUSH)
6597 btrfs_dev_stat_inc_and_print(dev,
6598 BTRFS_DEV_STAT_FLUSH_ERRS);
6602 if (bio == bbio->orig_bio)
6605 btrfs_bio_counter_dec(bbio->fs_info);
6607 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6610 bio = bbio->orig_bio;
6613 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6614 /* only send an error to the higher layers if it is
6615 * beyond the tolerance of the btrfs bio
6617 if (atomic_read(&bbio->error) > bbio->max_errors) {
6618 bio->bi_status = BLK_STS_IOERR;
6621 * this bio is actually up to date, we didn't
6622 * go over the max number of errors
6624 bio->bi_status = BLK_STS_OK;
6627 btrfs_end_bbio(bbio, bio);
6628 } else if (!is_orig_bio) {
6633 static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio,
6634 u64 physical, struct btrfs_device *dev)
6636 struct btrfs_fs_info *fs_info = bbio->fs_info;
6638 bio->bi_private = bbio;
6639 btrfs_io_bio(bio)->device = dev;
6640 bio->bi_end_io = btrfs_end_bio;
6641 bio->bi_iter.bi_sector = physical >> 9;
6643 * For zone append writing, bi_sector must point the beginning of the
6646 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
6647 if (btrfs_dev_is_sequential(dev, physical)) {
6648 u64 zone_start = round_down(physical, fs_info->zone_size);
6650 bio->bi_iter.bi_sector = zone_start >> SECTOR_SHIFT;
6652 bio->bi_opf &= ~REQ_OP_ZONE_APPEND;
6653 bio->bi_opf |= REQ_OP_WRITE;
6656 btrfs_debug_in_rcu(fs_info,
6657 "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
6658 bio_op(bio), bio->bi_opf, bio->bi_iter.bi_sector,
6659 (unsigned long)dev->bdev->bd_dev, rcu_str_deref(dev->name),
6660 dev->devid, bio->bi_iter.bi_size);
6661 bio_set_dev(bio, dev->bdev);
6663 btrfs_bio_counter_inc_noblocked(fs_info);
6665 btrfsic_submit_bio(bio);
6668 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6670 atomic_inc(&bbio->error);
6671 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6672 /* Should be the original bio. */
6673 WARN_ON(bio != bbio->orig_bio);
6675 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6676 bio->bi_iter.bi_sector = logical >> 9;
6677 if (atomic_read(&bbio->error) > bbio->max_errors)
6678 bio->bi_status = BLK_STS_IOERR;
6680 bio->bi_status = BLK_STS_OK;
6681 btrfs_end_bbio(bbio, bio);
6685 blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
6688 struct btrfs_device *dev;
6689 struct bio *first_bio = bio;
6690 u64 logical = bio->bi_iter.bi_sector << 9;
6696 struct btrfs_bio *bbio = NULL;
6698 length = bio->bi_iter.bi_size;
6699 map_length = length;
6701 btrfs_bio_counter_inc_blocked(fs_info);
6702 ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical,
6703 &map_length, &bbio, mirror_num, 1);
6705 btrfs_bio_counter_dec(fs_info);
6706 return errno_to_blk_status(ret);
6709 total_devs = bbio->num_stripes;
6710 bbio->orig_bio = first_bio;
6711 bbio->private = first_bio->bi_private;
6712 bbio->end_io = first_bio->bi_end_io;
6713 bbio->fs_info = fs_info;
6714 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6716 if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6717 ((btrfs_op(bio) == BTRFS_MAP_WRITE) || (mirror_num > 1))) {
6718 /* In this case, map_length has been set to the length of
6719 a single stripe; not the whole write */
6720 if (btrfs_op(bio) == BTRFS_MAP_WRITE) {
6721 ret = raid56_parity_write(fs_info, bio, bbio,
6724 ret = raid56_parity_recover(fs_info, bio, bbio,
6725 map_length, mirror_num, 1);
6728 btrfs_bio_counter_dec(fs_info);
6729 return errno_to_blk_status(ret);
6732 if (map_length < length) {
6734 "mapping failed logical %llu bio len %llu len %llu",
6735 logical, length, map_length);
6739 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6740 dev = bbio->stripes[dev_nr].dev;
6741 if (!dev || !dev->bdev || test_bit(BTRFS_DEV_STATE_MISSING,
6743 (btrfs_op(first_bio) == BTRFS_MAP_WRITE &&
6744 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) {
6745 bbio_error(bbio, first_bio, logical);
6749 if (dev_nr < total_devs - 1)
6750 bio = btrfs_bio_clone(first_bio);
6754 submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical, dev);
6756 btrfs_bio_counter_dec(fs_info);
6761 * Find a device specified by @devid or @uuid in the list of @fs_devices, or
6764 * If devid and uuid are both specified, the match must be exact, otherwise
6765 * only devid is used.
6767 struct btrfs_device *btrfs_find_device(struct btrfs_fs_devices *fs_devices,
6768 u64 devid, u8 *uuid, u8 *fsid)
6770 struct btrfs_device *device;
6771 struct btrfs_fs_devices *seed_devs;
6773 if (!fsid || !memcmp(fs_devices->metadata_uuid, fsid, BTRFS_FSID_SIZE)) {
6774 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6775 if (device->devid == devid &&
6776 (!uuid || memcmp(device->uuid, uuid,
6777 BTRFS_UUID_SIZE) == 0))
6782 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
6784 !memcmp(seed_devs->metadata_uuid, fsid, BTRFS_FSID_SIZE)) {
6785 list_for_each_entry(device, &seed_devs->devices,
6787 if (device->devid == devid &&
6788 (!uuid || memcmp(device->uuid, uuid,
6789 BTRFS_UUID_SIZE) == 0))
6798 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices,
6799 u64 devid, u8 *dev_uuid)
6801 struct btrfs_device *device;
6802 unsigned int nofs_flag;
6805 * We call this under the chunk_mutex, so we want to use NOFS for this
6806 * allocation, however we don't want to change btrfs_alloc_device() to
6807 * always do NOFS because we use it in a lot of other GFP_KERNEL safe
6810 nofs_flag = memalloc_nofs_save();
6811 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6812 memalloc_nofs_restore(nofs_flag);
6816 list_add(&device->dev_list, &fs_devices->devices);
6817 device->fs_devices = fs_devices;
6818 fs_devices->num_devices++;
6820 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6821 fs_devices->missing_devices++;
6827 * btrfs_alloc_device - allocate struct btrfs_device
6828 * @fs_info: used only for generating a new devid, can be NULL if
6829 * devid is provided (i.e. @devid != NULL).
6830 * @devid: a pointer to devid for this device. If NULL a new devid
6832 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6835 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6836 * on error. Returned struct is not linked onto any lists and must be
6837 * destroyed with btrfs_free_device.
6839 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6843 struct btrfs_device *dev;
6846 if (WARN_ON(!devid && !fs_info))
6847 return ERR_PTR(-EINVAL);
6849 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
6851 return ERR_PTR(-ENOMEM);
6854 * Preallocate a bio that's always going to be used for flushing device
6855 * barriers and matches the device lifespan
6857 dev->flush_bio = bio_kmalloc(GFP_KERNEL, 0);
6858 if (!dev->flush_bio) {
6860 return ERR_PTR(-ENOMEM);
6863 INIT_LIST_HEAD(&dev->dev_list);
6864 INIT_LIST_HEAD(&dev->dev_alloc_list);
6865 INIT_LIST_HEAD(&dev->post_commit_list);
6867 atomic_set(&dev->reada_in_flight, 0);
6868 atomic_set(&dev->dev_stats_ccnt, 0);
6869 btrfs_device_data_ordered_init(dev);
6870 INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
6871 INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
6872 extent_io_tree_init(fs_info, &dev->alloc_state,
6873 IO_TREE_DEVICE_ALLOC_STATE, NULL);
6880 ret = find_next_devid(fs_info, &tmp);
6882 btrfs_free_device(dev);
6883 return ERR_PTR(ret);
6889 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6891 generate_random_uuid(dev->uuid);
6896 static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info,
6897 u64 devid, u8 *uuid, bool error)
6900 btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing",
6903 btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing",
6907 static u64 calc_stripe_length(u64 type, u64 chunk_len, int num_stripes)
6909 const int data_stripes = calc_data_stripes(type, num_stripes);
6911 return div_u64(chunk_len, data_stripes);
6914 #if BITS_PER_LONG == 32
6916 * Due to page cache limit, metadata beyond BTRFS_32BIT_MAX_FILE_SIZE
6917 * can't be accessed on 32bit systems.
6919 * This function do mount time check to reject the fs if it already has
6920 * metadata chunk beyond that limit.
6922 static int check_32bit_meta_chunk(struct btrfs_fs_info *fs_info,
6923 u64 logical, u64 length, u64 type)
6925 if (!(type & BTRFS_BLOCK_GROUP_METADATA))
6928 if (logical + length < MAX_LFS_FILESIZE)
6931 btrfs_err_32bit_limit(fs_info);
6936 * This is to give early warning for any metadata chunk reaching
6937 * BTRFS_32BIT_EARLY_WARN_THRESHOLD.
6938 * Although we can still access the metadata, it's not going to be possible
6939 * once the limit is reached.
6941 static void warn_32bit_meta_chunk(struct btrfs_fs_info *fs_info,
6942 u64 logical, u64 length, u64 type)
6944 if (!(type & BTRFS_BLOCK_GROUP_METADATA))
6947 if (logical + length < BTRFS_32BIT_EARLY_WARN_THRESHOLD)
6950 btrfs_warn_32bit_limit(fs_info);
6954 static int read_one_chunk(struct btrfs_key *key, struct extent_buffer *leaf,
6955 struct btrfs_chunk *chunk)
6957 struct btrfs_fs_info *fs_info = leaf->fs_info;
6958 struct extent_map_tree *map_tree = &fs_info->mapping_tree;
6959 struct map_lookup *map;
6960 struct extent_map *em;
6965 u8 uuid[BTRFS_UUID_SIZE];
6970 logical = key->offset;
6971 length = btrfs_chunk_length(leaf, chunk);
6972 type = btrfs_chunk_type(leaf, chunk);
6973 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6975 #if BITS_PER_LONG == 32
6976 ret = check_32bit_meta_chunk(fs_info, logical, length, type);
6979 warn_32bit_meta_chunk(fs_info, logical, length, type);
6983 * Only need to verify chunk item if we're reading from sys chunk array,
6984 * as chunk item in tree block is already verified by tree-checker.
6986 if (leaf->start == BTRFS_SUPER_INFO_OFFSET) {
6987 ret = btrfs_check_chunk_valid(leaf, chunk, logical);
6992 read_lock(&map_tree->lock);
6993 em = lookup_extent_mapping(map_tree, logical, 1);
6994 read_unlock(&map_tree->lock);
6996 /* already mapped? */
6997 if (em && em->start <= logical && em->start + em->len > logical) {
6998 free_extent_map(em);
7001 free_extent_map(em);
7004 em = alloc_extent_map();
7007 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
7009 free_extent_map(em);
7013 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
7014 em->map_lookup = map;
7015 em->start = logical;
7018 em->block_start = 0;
7019 em->block_len = em->len;
7021 map->num_stripes = num_stripes;
7022 map->io_width = btrfs_chunk_io_width(leaf, chunk);
7023 map->io_align = btrfs_chunk_io_align(leaf, chunk);
7024 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
7026 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
7027 map->verified_stripes = 0;
7028 em->orig_block_len = calc_stripe_length(type, em->len,
7030 for (i = 0; i < num_stripes; i++) {
7031 map->stripes[i].physical =
7032 btrfs_stripe_offset_nr(leaf, chunk, i);
7033 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
7034 read_extent_buffer(leaf, uuid, (unsigned long)
7035 btrfs_stripe_dev_uuid_nr(chunk, i),
7037 map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices,
7039 if (!map->stripes[i].dev &&
7040 !btrfs_test_opt(fs_info, DEGRADED)) {
7041 free_extent_map(em);
7042 btrfs_report_missing_device(fs_info, devid, uuid, true);
7045 if (!map->stripes[i].dev) {
7046 map->stripes[i].dev =
7047 add_missing_dev(fs_info->fs_devices, devid,
7049 if (IS_ERR(map->stripes[i].dev)) {
7050 free_extent_map(em);
7052 "failed to init missing dev %llu: %ld",
7053 devid, PTR_ERR(map->stripes[i].dev));
7054 return PTR_ERR(map->stripes[i].dev);
7056 btrfs_report_missing_device(fs_info, devid, uuid, false);
7058 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
7059 &(map->stripes[i].dev->dev_state));
7063 write_lock(&map_tree->lock);
7064 ret = add_extent_mapping(map_tree, em, 0);
7065 write_unlock(&map_tree->lock);
7068 "failed to add chunk map, start=%llu len=%llu: %d",
7069 em->start, em->len, ret);
7071 free_extent_map(em);
7076 static void fill_device_from_item(struct extent_buffer *leaf,
7077 struct btrfs_dev_item *dev_item,
7078 struct btrfs_device *device)
7082 device->devid = btrfs_device_id(leaf, dev_item);
7083 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
7084 device->total_bytes = device->disk_total_bytes;
7085 device->commit_total_bytes = device->disk_total_bytes;
7086 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
7087 device->commit_bytes_used = device->bytes_used;
7088 device->type = btrfs_device_type(leaf, dev_item);
7089 device->io_align = btrfs_device_io_align(leaf, dev_item);
7090 device->io_width = btrfs_device_io_width(leaf, dev_item);
7091 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
7092 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
7093 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
7095 ptr = btrfs_device_uuid(dev_item);
7096 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
7099 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info,
7102 struct btrfs_fs_devices *fs_devices;
7105 lockdep_assert_held(&uuid_mutex);
7108 /* This will match only for multi-device seed fs */
7109 list_for_each_entry(fs_devices, &fs_info->fs_devices->seed_list, seed_list)
7110 if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE))
7114 fs_devices = find_fsid(fsid, NULL);
7116 if (!btrfs_test_opt(fs_info, DEGRADED))
7117 return ERR_PTR(-ENOENT);
7119 fs_devices = alloc_fs_devices(fsid, NULL);
7120 if (IS_ERR(fs_devices))
7123 fs_devices->seeding = true;
7124 fs_devices->opened = 1;
7129 * Upon first call for a seed fs fsid, just create a private copy of the
7130 * respective fs_devices and anchor it at fs_info->fs_devices->seed_list
7132 fs_devices = clone_fs_devices(fs_devices);
7133 if (IS_ERR(fs_devices))
7136 ret = open_fs_devices(fs_devices, FMODE_READ, fs_info->bdev_holder);
7138 free_fs_devices(fs_devices);
7139 return ERR_PTR(ret);
7142 if (!fs_devices->seeding) {
7143 close_fs_devices(fs_devices);
7144 free_fs_devices(fs_devices);
7145 return ERR_PTR(-EINVAL);
7148 list_add(&fs_devices->seed_list, &fs_info->fs_devices->seed_list);
7153 static int read_one_dev(struct extent_buffer *leaf,
7154 struct btrfs_dev_item *dev_item)
7156 struct btrfs_fs_info *fs_info = leaf->fs_info;
7157 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7158 struct btrfs_device *device;
7161 u8 fs_uuid[BTRFS_FSID_SIZE];
7162 u8 dev_uuid[BTRFS_UUID_SIZE];
7164 devid = btrfs_device_id(leaf, dev_item);
7165 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
7167 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
7170 if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) {
7171 fs_devices = open_seed_devices(fs_info, fs_uuid);
7172 if (IS_ERR(fs_devices))
7173 return PTR_ERR(fs_devices);
7176 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
7179 if (!btrfs_test_opt(fs_info, DEGRADED)) {
7180 btrfs_report_missing_device(fs_info, devid,
7185 device = add_missing_dev(fs_devices, devid, dev_uuid);
7186 if (IS_ERR(device)) {
7188 "failed to add missing dev %llu: %ld",
7189 devid, PTR_ERR(device));
7190 return PTR_ERR(device);
7192 btrfs_report_missing_device(fs_info, devid, dev_uuid, false);
7194 if (!device->bdev) {
7195 if (!btrfs_test_opt(fs_info, DEGRADED)) {
7196 btrfs_report_missing_device(fs_info,
7197 devid, dev_uuid, true);
7200 btrfs_report_missing_device(fs_info, devid,
7204 if (!device->bdev &&
7205 !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
7207 * this happens when a device that was properly setup
7208 * in the device info lists suddenly goes bad.
7209 * device->bdev is NULL, and so we have to set
7210 * device->missing to one here
7212 device->fs_devices->missing_devices++;
7213 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
7216 /* Move the device to its own fs_devices */
7217 if (device->fs_devices != fs_devices) {
7218 ASSERT(test_bit(BTRFS_DEV_STATE_MISSING,
7219 &device->dev_state));
7221 list_move(&device->dev_list, &fs_devices->devices);
7222 device->fs_devices->num_devices--;
7223 fs_devices->num_devices++;
7225 device->fs_devices->missing_devices--;
7226 fs_devices->missing_devices++;
7228 device->fs_devices = fs_devices;
7232 if (device->fs_devices != fs_info->fs_devices) {
7233 BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state));
7234 if (device->generation !=
7235 btrfs_device_generation(leaf, dev_item))
7239 fill_device_from_item(leaf, dev_item, device);
7241 u64 max_total_bytes = i_size_read(device->bdev->bd_inode);
7243 if (device->total_bytes > max_total_bytes) {
7245 "device total_bytes should be at most %llu but found %llu",
7246 max_total_bytes, device->total_bytes);
7250 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
7251 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
7252 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
7253 device->fs_devices->total_rw_bytes += device->total_bytes;
7254 atomic64_add(device->total_bytes - device->bytes_used,
7255 &fs_info->free_chunk_space);
7261 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
7263 struct btrfs_root *root = fs_info->tree_root;
7264 struct btrfs_super_block *super_copy = fs_info->super_copy;
7265 struct extent_buffer *sb;
7266 struct btrfs_disk_key *disk_key;
7267 struct btrfs_chunk *chunk;
7269 unsigned long sb_array_offset;
7276 struct btrfs_key key;
7278 ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize);
7280 * This will create extent buffer of nodesize, superblock size is
7281 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
7282 * overallocate but we can keep it as-is, only the first page is used.
7284 sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET,
7285 root->root_key.objectid, 0);
7288 set_extent_buffer_uptodate(sb);
7290 * The sb extent buffer is artificial and just used to read the system array.
7291 * set_extent_buffer_uptodate() call does not properly mark all it's
7292 * pages up-to-date when the page is larger: extent does not cover the
7293 * whole page and consequently check_page_uptodate does not find all
7294 * the page's extents up-to-date (the hole beyond sb),
7295 * write_extent_buffer then triggers a WARN_ON.
7297 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
7298 * but sb spans only this function. Add an explicit SetPageUptodate call
7299 * to silence the warning eg. on PowerPC 64.
7301 if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
7302 SetPageUptodate(sb->pages[0]);
7304 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
7305 array_size = btrfs_super_sys_array_size(super_copy);
7307 array_ptr = super_copy->sys_chunk_array;
7308 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
7311 while (cur_offset < array_size) {
7312 disk_key = (struct btrfs_disk_key *)array_ptr;
7313 len = sizeof(*disk_key);
7314 if (cur_offset + len > array_size)
7315 goto out_short_read;
7317 btrfs_disk_key_to_cpu(&key, disk_key);
7320 sb_array_offset += len;
7323 if (key.type != BTRFS_CHUNK_ITEM_KEY) {
7325 "unexpected item type %u in sys_array at offset %u",
7326 (u32)key.type, cur_offset);
7331 chunk = (struct btrfs_chunk *)sb_array_offset;
7333 * At least one btrfs_chunk with one stripe must be present,
7334 * exact stripe count check comes afterwards
7336 len = btrfs_chunk_item_size(1);
7337 if (cur_offset + len > array_size)
7338 goto out_short_read;
7340 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
7343 "invalid number of stripes %u in sys_array at offset %u",
7344 num_stripes, cur_offset);
7349 type = btrfs_chunk_type(sb, chunk);
7350 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
7352 "invalid chunk type %llu in sys_array at offset %u",
7358 len = btrfs_chunk_item_size(num_stripes);
7359 if (cur_offset + len > array_size)
7360 goto out_short_read;
7362 ret = read_one_chunk(&key, sb, chunk);
7367 sb_array_offset += len;
7370 clear_extent_buffer_uptodate(sb);
7371 free_extent_buffer_stale(sb);
7375 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
7377 clear_extent_buffer_uptodate(sb);
7378 free_extent_buffer_stale(sb);
7383 * Check if all chunks in the fs are OK for read-write degraded mount
7385 * If the @failing_dev is specified, it's accounted as missing.
7387 * Return true if all chunks meet the minimal RW mount requirements.
7388 * Return false if any chunk doesn't meet the minimal RW mount requirements.
7390 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
7391 struct btrfs_device *failing_dev)
7393 struct extent_map_tree *map_tree = &fs_info->mapping_tree;
7394 struct extent_map *em;
7398 read_lock(&map_tree->lock);
7399 em = lookup_extent_mapping(map_tree, 0, (u64)-1);
7400 read_unlock(&map_tree->lock);
7401 /* No chunk at all? Return false anyway */
7407 struct map_lookup *map;
7412 map = em->map_lookup;
7414 btrfs_get_num_tolerated_disk_barrier_failures(
7416 for (i = 0; i < map->num_stripes; i++) {
7417 struct btrfs_device *dev = map->stripes[i].dev;
7419 if (!dev || !dev->bdev ||
7420 test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) ||
7421 dev->last_flush_error)
7423 else if (failing_dev && failing_dev == dev)
7426 if (missing > max_tolerated) {
7429 "chunk %llu missing %d devices, max tolerance is %d for writable mount",
7430 em->start, missing, max_tolerated);
7431 free_extent_map(em);
7435 next_start = extent_map_end(em);
7436 free_extent_map(em);
7438 read_lock(&map_tree->lock);
7439 em = lookup_extent_mapping(map_tree, next_start,
7440 (u64)(-1) - next_start);
7441 read_unlock(&map_tree->lock);
7447 static void readahead_tree_node_children(struct extent_buffer *node)
7450 const int nr_items = btrfs_header_nritems(node);
7452 for (i = 0; i < nr_items; i++)
7453 btrfs_readahead_node_child(node, i);
7456 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
7458 struct btrfs_root *root = fs_info->chunk_root;
7459 struct btrfs_path *path;
7460 struct extent_buffer *leaf;
7461 struct btrfs_key key;
7462 struct btrfs_key found_key;
7466 u64 last_ra_node = 0;
7468 path = btrfs_alloc_path();
7473 * uuid_mutex is needed only if we are mounting a sprout FS
7474 * otherwise we don't need it.
7476 mutex_lock(&uuid_mutex);
7479 * It is possible for mount and umount to race in such a way that
7480 * we execute this code path, but open_fs_devices failed to clear
7481 * total_rw_bytes. We certainly want it cleared before reading the
7482 * device items, so clear it here.
7484 fs_info->fs_devices->total_rw_bytes = 0;
7487 * Lockdep complains about possible circular locking dependency between
7488 * a disk's open_mutex (struct gendisk.open_mutex), the rw semaphores
7489 * used for freeze procection of a fs (struct super_block.s_writers),
7490 * which we take when starting a transaction, and extent buffers of the
7491 * chunk tree if we call read_one_dev() while holding a lock on an
7492 * extent buffer of the chunk tree. Since we are mounting the filesystem
7493 * and at this point there can't be any concurrent task modifying the
7494 * chunk tree, to keep it simple, just skip locking on the chunk tree.
7496 ASSERT(!test_bit(BTRFS_FS_OPEN, &fs_info->flags));
7497 path->skip_locking = 1;
7500 * Read all device items, and then all the chunk items. All
7501 * device items are found before any chunk item (their object id
7502 * is smaller than the lowest possible object id for a chunk
7503 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
7505 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
7508 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7512 struct extent_buffer *node;
7514 leaf = path->nodes[0];
7515 slot = path->slots[0];
7516 if (slot >= btrfs_header_nritems(leaf)) {
7517 ret = btrfs_next_leaf(root, path);
7524 node = path->nodes[1];
7526 if (last_ra_node != node->start) {
7527 readahead_tree_node_children(node);
7528 last_ra_node = node->start;
7531 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7532 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
7533 struct btrfs_dev_item *dev_item;
7534 dev_item = btrfs_item_ptr(leaf, slot,
7535 struct btrfs_dev_item);
7536 ret = read_one_dev(leaf, dev_item);
7540 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
7541 struct btrfs_chunk *chunk;
7544 * We are only called at mount time, so no need to take
7545 * fs_info->chunk_mutex. Plus, to avoid lockdep warnings,
7546 * we always lock first fs_info->chunk_mutex before
7547 * acquiring any locks on the chunk tree. This is a
7548 * requirement for chunk allocation, see the comment on
7549 * top of btrfs_chunk_alloc() for details.
7551 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
7552 ret = read_one_chunk(&found_key, leaf, chunk);
7560 * After loading chunk tree, we've got all device information,
7561 * do another round of validation checks.
7563 if (total_dev != fs_info->fs_devices->total_devices) {
7565 "super_num_devices %llu mismatch with num_devices %llu found here",
7566 btrfs_super_num_devices(fs_info->super_copy),
7571 if (btrfs_super_total_bytes(fs_info->super_copy) <
7572 fs_info->fs_devices->total_rw_bytes) {
7574 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
7575 btrfs_super_total_bytes(fs_info->super_copy),
7576 fs_info->fs_devices->total_rw_bytes);
7582 mutex_unlock(&uuid_mutex);
7584 btrfs_free_path(path);
7588 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
7590 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
7591 struct btrfs_device *device;
7593 fs_devices->fs_info = fs_info;
7595 mutex_lock(&fs_devices->device_list_mutex);
7596 list_for_each_entry(device, &fs_devices->devices, dev_list)
7597 device->fs_info = fs_info;
7599 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
7600 list_for_each_entry(device, &seed_devs->devices, dev_list)
7601 device->fs_info = fs_info;
7603 seed_devs->fs_info = fs_info;
7605 mutex_unlock(&fs_devices->device_list_mutex);
7608 static u64 btrfs_dev_stats_value(const struct extent_buffer *eb,
7609 const struct btrfs_dev_stats_item *ptr,
7614 read_extent_buffer(eb, &val,
7615 offsetof(struct btrfs_dev_stats_item, values) +
7616 ((unsigned long)ptr) + (index * sizeof(u64)),
7621 static void btrfs_set_dev_stats_value(struct extent_buffer *eb,
7622 struct btrfs_dev_stats_item *ptr,
7625 write_extent_buffer(eb, &val,
7626 offsetof(struct btrfs_dev_stats_item, values) +
7627 ((unsigned long)ptr) + (index * sizeof(u64)),
7631 static int btrfs_device_init_dev_stats(struct btrfs_device *device,
7632 struct btrfs_path *path)
7634 struct btrfs_dev_stats_item *ptr;
7635 struct extent_buffer *eb;
7636 struct btrfs_key key;
7640 if (!device->fs_info->dev_root)
7643 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7644 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7645 key.offset = device->devid;
7646 ret = btrfs_search_slot(NULL, device->fs_info->dev_root, &key, path, 0, 0);
7648 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7649 btrfs_dev_stat_set(device, i, 0);
7650 device->dev_stats_valid = 1;
7651 btrfs_release_path(path);
7652 return ret < 0 ? ret : 0;
7654 slot = path->slots[0];
7655 eb = path->nodes[0];
7656 item_size = btrfs_item_size_nr(eb, slot);
7658 ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_stats_item);
7660 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7661 if (item_size >= (1 + i) * sizeof(__le64))
7662 btrfs_dev_stat_set(device, i,
7663 btrfs_dev_stats_value(eb, ptr, i));
7665 btrfs_dev_stat_set(device, i, 0);
7668 device->dev_stats_valid = 1;
7669 btrfs_dev_stat_print_on_load(device);
7670 btrfs_release_path(path);
7675 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
7677 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
7678 struct btrfs_device *device;
7679 struct btrfs_path *path = NULL;
7682 path = btrfs_alloc_path();
7686 mutex_lock(&fs_devices->device_list_mutex);
7687 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7688 ret = btrfs_device_init_dev_stats(device, path);
7692 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
7693 list_for_each_entry(device, &seed_devs->devices, dev_list) {
7694 ret = btrfs_device_init_dev_stats(device, path);
7700 mutex_unlock(&fs_devices->device_list_mutex);
7702 btrfs_free_path(path);
7706 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
7707 struct btrfs_device *device)
7709 struct btrfs_fs_info *fs_info = trans->fs_info;
7710 struct btrfs_root *dev_root = fs_info->dev_root;
7711 struct btrfs_path *path;
7712 struct btrfs_key key;
7713 struct extent_buffer *eb;
7714 struct btrfs_dev_stats_item *ptr;
7718 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7719 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7720 key.offset = device->devid;
7722 path = btrfs_alloc_path();
7725 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
7727 btrfs_warn_in_rcu(fs_info,
7728 "error %d while searching for dev_stats item for device %s",
7729 ret, rcu_str_deref(device->name));
7734 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
7735 /* need to delete old one and insert a new one */
7736 ret = btrfs_del_item(trans, dev_root, path);
7738 btrfs_warn_in_rcu(fs_info,
7739 "delete too small dev_stats item for device %s failed %d",
7740 rcu_str_deref(device->name), ret);
7747 /* need to insert a new item */
7748 btrfs_release_path(path);
7749 ret = btrfs_insert_empty_item(trans, dev_root, path,
7750 &key, sizeof(*ptr));
7752 btrfs_warn_in_rcu(fs_info,
7753 "insert dev_stats item for device %s failed %d",
7754 rcu_str_deref(device->name), ret);
7759 eb = path->nodes[0];
7760 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7761 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7762 btrfs_set_dev_stats_value(eb, ptr, i,
7763 btrfs_dev_stat_read(device, i));
7764 btrfs_mark_buffer_dirty(eb);
7767 btrfs_free_path(path);
7772 * called from commit_transaction. Writes all changed device stats to disk.
7774 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans)
7776 struct btrfs_fs_info *fs_info = trans->fs_info;
7777 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7778 struct btrfs_device *device;
7782 mutex_lock(&fs_devices->device_list_mutex);
7783 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7784 stats_cnt = atomic_read(&device->dev_stats_ccnt);
7785 if (!device->dev_stats_valid || stats_cnt == 0)
7790 * There is a LOAD-LOAD control dependency between the value of
7791 * dev_stats_ccnt and updating the on-disk values which requires
7792 * reading the in-memory counters. Such control dependencies
7793 * require explicit read memory barriers.
7795 * This memory barriers pairs with smp_mb__before_atomic in
7796 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full
7797 * barrier implied by atomic_xchg in
7798 * btrfs_dev_stats_read_and_reset
7802 ret = update_dev_stat_item(trans, device);
7804 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7806 mutex_unlock(&fs_devices->device_list_mutex);
7811 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7813 btrfs_dev_stat_inc(dev, index);
7814 btrfs_dev_stat_print_on_error(dev);
7817 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
7819 if (!dev->dev_stats_valid)
7821 btrfs_err_rl_in_rcu(dev->fs_info,
7822 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7823 rcu_str_deref(dev->name),
7824 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7825 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7826 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7827 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7828 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7831 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7835 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7836 if (btrfs_dev_stat_read(dev, i) != 0)
7838 if (i == BTRFS_DEV_STAT_VALUES_MAX)
7839 return; /* all values == 0, suppress message */
7841 btrfs_info_in_rcu(dev->fs_info,
7842 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7843 rcu_str_deref(dev->name),
7844 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7845 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7846 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7847 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7848 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7851 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
7852 struct btrfs_ioctl_get_dev_stats *stats)
7854 struct btrfs_device *dev;
7855 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7858 mutex_lock(&fs_devices->device_list_mutex);
7859 dev = btrfs_find_device(fs_info->fs_devices, stats->devid, NULL, NULL);
7860 mutex_unlock(&fs_devices->device_list_mutex);
7863 btrfs_warn(fs_info, "get dev_stats failed, device not found");
7865 } else if (!dev->dev_stats_valid) {
7866 btrfs_warn(fs_info, "get dev_stats failed, not yet valid");
7868 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7869 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7870 if (stats->nr_items > i)
7872 btrfs_dev_stat_read_and_reset(dev, i);
7874 btrfs_dev_stat_set(dev, i, 0);
7876 btrfs_info(fs_info, "device stats zeroed by %s (%d)",
7877 current->comm, task_pid_nr(current));
7879 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7880 if (stats->nr_items > i)
7881 stats->values[i] = btrfs_dev_stat_read(dev, i);
7883 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7884 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7889 * Update the size and bytes used for each device where it changed. This is
7890 * delayed since we would otherwise get errors while writing out the
7893 * Must be invoked during transaction commit.
7895 void btrfs_commit_device_sizes(struct btrfs_transaction *trans)
7897 struct btrfs_device *curr, *next;
7899 ASSERT(trans->state == TRANS_STATE_COMMIT_DOING);
7901 if (list_empty(&trans->dev_update_list))
7905 * We don't need the device_list_mutex here. This list is owned by the
7906 * transaction and the transaction must complete before the device is
7909 mutex_lock(&trans->fs_info->chunk_mutex);
7910 list_for_each_entry_safe(curr, next, &trans->dev_update_list,
7912 list_del_init(&curr->post_commit_list);
7913 curr->commit_total_bytes = curr->disk_total_bytes;
7914 curr->commit_bytes_used = curr->bytes_used;
7916 mutex_unlock(&trans->fs_info->chunk_mutex);
7920 * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10.
7922 int btrfs_bg_type_to_factor(u64 flags)
7924 const int index = btrfs_bg_flags_to_raid_index(flags);
7926 return btrfs_raid_array[index].ncopies;
7931 static int verify_one_dev_extent(struct btrfs_fs_info *fs_info,
7932 u64 chunk_offset, u64 devid,
7933 u64 physical_offset, u64 physical_len)
7935 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
7936 struct extent_map *em;
7937 struct map_lookup *map;
7938 struct btrfs_device *dev;
7944 read_lock(&em_tree->lock);
7945 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
7946 read_unlock(&em_tree->lock);
7950 "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk",
7951 physical_offset, devid);
7956 map = em->map_lookup;
7957 stripe_len = calc_stripe_length(map->type, em->len, map->num_stripes);
7958 if (physical_len != stripe_len) {
7960 "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu",
7961 physical_offset, devid, em->start, physical_len,
7967 for (i = 0; i < map->num_stripes; i++) {
7968 if (map->stripes[i].dev->devid == devid &&
7969 map->stripes[i].physical == physical_offset) {
7971 if (map->verified_stripes >= map->num_stripes) {
7973 "too many dev extents for chunk %llu found",
7978 map->verified_stripes++;
7984 "dev extent physical offset %llu devid %llu has no corresponding chunk",
7985 physical_offset, devid);
7989 /* Make sure no dev extent is beyond device boundary */
7990 dev = btrfs_find_device(fs_info->fs_devices, devid, NULL, NULL);
7992 btrfs_err(fs_info, "failed to find devid %llu", devid);
7997 if (physical_offset + physical_len > dev->disk_total_bytes) {
7999 "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu",
8000 devid, physical_offset, physical_len,
8001 dev->disk_total_bytes);
8006 if (dev->zone_info) {
8007 u64 zone_size = dev->zone_info->zone_size;
8009 if (!IS_ALIGNED(physical_offset, zone_size) ||
8010 !IS_ALIGNED(physical_len, zone_size)) {
8012 "zoned: dev extent devid %llu physical offset %llu len %llu is not aligned to device zone",
8013 devid, physical_offset, physical_len);
8020 free_extent_map(em);
8024 static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info)
8026 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
8027 struct extent_map *em;
8028 struct rb_node *node;
8031 read_lock(&em_tree->lock);
8032 for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) {
8033 em = rb_entry(node, struct extent_map, rb_node);
8034 if (em->map_lookup->num_stripes !=
8035 em->map_lookup->verified_stripes) {
8037 "chunk %llu has missing dev extent, have %d expect %d",
8038 em->start, em->map_lookup->verified_stripes,
8039 em->map_lookup->num_stripes);
8045 read_unlock(&em_tree->lock);
8050 * Ensure that all dev extents are mapped to correct chunk, otherwise
8051 * later chunk allocation/free would cause unexpected behavior.
8053 * NOTE: This will iterate through the whole device tree, which should be of
8054 * the same size level as the chunk tree. This slightly increases mount time.
8056 int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info)
8058 struct btrfs_path *path;
8059 struct btrfs_root *root = fs_info->dev_root;
8060 struct btrfs_key key;
8062 u64 prev_dev_ext_end = 0;
8066 * We don't have a dev_root because we mounted with ignorebadroots and
8067 * failed to load the root, so we want to skip the verification in this
8070 * However if the dev root is fine, but the tree itself is corrupted
8071 * we'd still fail to mount. This verification is only to make sure
8072 * writes can happen safely, so instead just bypass this check
8073 * completely in the case of IGNOREBADROOTS.
8075 if (btrfs_test_opt(fs_info, IGNOREBADROOTS))
8079 key.type = BTRFS_DEV_EXTENT_KEY;
8082 path = btrfs_alloc_path();
8086 path->reada = READA_FORWARD;
8087 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
8091 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
8092 ret = btrfs_next_leaf(root, path);
8095 /* No dev extents at all? Not good */
8102 struct extent_buffer *leaf = path->nodes[0];
8103 struct btrfs_dev_extent *dext;
8104 int slot = path->slots[0];
8106 u64 physical_offset;
8110 btrfs_item_key_to_cpu(leaf, &key, slot);
8111 if (key.type != BTRFS_DEV_EXTENT_KEY)
8113 devid = key.objectid;
8114 physical_offset = key.offset;
8116 dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent);
8117 chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext);
8118 physical_len = btrfs_dev_extent_length(leaf, dext);
8120 /* Check if this dev extent overlaps with the previous one */
8121 if (devid == prev_devid && physical_offset < prev_dev_ext_end) {
8123 "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu",
8124 devid, physical_offset, prev_dev_ext_end);
8129 ret = verify_one_dev_extent(fs_info, chunk_offset, devid,
8130 physical_offset, physical_len);
8134 prev_dev_ext_end = physical_offset + physical_len;
8136 ret = btrfs_next_item(root, path);
8145 /* Ensure all chunks have corresponding dev extents */
8146 ret = verify_chunk_dev_extent_mapping(fs_info);
8148 btrfs_free_path(path);
8153 * Check whether the given block group or device is pinned by any inode being
8154 * used as a swapfile.
8156 bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr)
8158 struct btrfs_swapfile_pin *sp;
8159 struct rb_node *node;
8161 spin_lock(&fs_info->swapfile_pins_lock);
8162 node = fs_info->swapfile_pins.rb_node;
8164 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
8166 node = node->rb_left;
8167 else if (ptr > sp->ptr)
8168 node = node->rb_right;
8172 spin_unlock(&fs_info->swapfile_pins_lock);
8173 return node != NULL;
8176 static int relocating_repair_kthread(void *data)
8178 struct btrfs_block_group *cache = (struct btrfs_block_group *)data;
8179 struct btrfs_fs_info *fs_info = cache->fs_info;
8183 target = cache->start;
8184 btrfs_put_block_group(cache);
8186 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) {
8188 "zoned: skip relocating block group %llu to repair: EBUSY",
8193 mutex_lock(&fs_info->reclaim_bgs_lock);
8195 /* Ensure block group still exists */
8196 cache = btrfs_lookup_block_group(fs_info, target);
8200 if (!cache->relocating_repair)
8203 ret = btrfs_may_alloc_data_chunk(fs_info, target);
8208 "zoned: relocating block group %llu to repair IO failure",
8210 ret = btrfs_relocate_chunk(fs_info, target);
8214 btrfs_put_block_group(cache);
8215 mutex_unlock(&fs_info->reclaim_bgs_lock);
8216 btrfs_exclop_finish(fs_info);
8221 int btrfs_repair_one_zone(struct btrfs_fs_info *fs_info, u64 logical)
8223 struct btrfs_block_group *cache;
8225 /* Do not attempt to repair in degraded state */
8226 if (btrfs_test_opt(fs_info, DEGRADED))
8229 cache = btrfs_lookup_block_group(fs_info, logical);
8233 spin_lock(&cache->lock);
8234 if (cache->relocating_repair) {
8235 spin_unlock(&cache->lock);
8236 btrfs_put_block_group(cache);
8239 cache->relocating_repair = 1;
8240 spin_unlock(&cache->lock);
8242 kthread_run(relocating_repair_kthread, cache,
8243 "btrfs-relocating-repair");