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 device = ERR_PTR(ret);
1371 goto error_bdev_put;
1374 disk_super = btrfs_read_disk_super(bdev, bytenr, bytenr_orig);
1375 if (IS_ERR(disk_super)) {
1376 device = ERR_CAST(disk_super);
1377 goto error_bdev_put;
1380 device = device_list_add(path, disk_super, &new_device_added);
1381 if (!IS_ERR(device)) {
1382 if (new_device_added)
1383 btrfs_free_stale_devices(path, device);
1386 btrfs_release_disk_super(disk_super);
1389 blkdev_put(bdev, flags);
1395 * Try to find a chunk that intersects [start, start + len] range and when one
1396 * such is found, record the end of it in *start
1398 static bool contains_pending_extent(struct btrfs_device *device, u64 *start,
1401 u64 physical_start, physical_end;
1403 lockdep_assert_held(&device->fs_info->chunk_mutex);
1405 if (!find_first_extent_bit(&device->alloc_state, *start,
1406 &physical_start, &physical_end,
1407 CHUNK_ALLOCATED, NULL)) {
1409 if (in_range(physical_start, *start, len) ||
1410 in_range(*start, physical_start,
1411 physical_end - physical_start)) {
1412 *start = physical_end + 1;
1419 static u64 dev_extent_search_start(struct btrfs_device *device, u64 start)
1421 switch (device->fs_devices->chunk_alloc_policy) {
1422 case BTRFS_CHUNK_ALLOC_REGULAR:
1424 * We don't want to overwrite the superblock on the drive nor
1425 * any area used by the boot loader (grub for example), so we
1426 * make sure to start at an offset of at least 1MB.
1428 return max_t(u64, start, SZ_1M);
1429 case BTRFS_CHUNK_ALLOC_ZONED:
1431 * We don't care about the starting region like regular
1432 * allocator, because we anyway use/reserve the first two zones
1433 * for superblock logging.
1435 return ALIGN(start, device->zone_info->zone_size);
1441 static bool dev_extent_hole_check_zoned(struct btrfs_device *device,
1442 u64 *hole_start, u64 *hole_size,
1445 u64 zone_size = device->zone_info->zone_size;
1448 bool changed = false;
1450 ASSERT(IS_ALIGNED(*hole_start, zone_size));
1452 while (*hole_size > 0) {
1453 pos = btrfs_find_allocatable_zones(device, *hole_start,
1454 *hole_start + *hole_size,
1456 if (pos != *hole_start) {
1457 *hole_size = *hole_start + *hole_size - pos;
1460 if (*hole_size < num_bytes)
1464 ret = btrfs_ensure_empty_zones(device, pos, num_bytes);
1466 /* Range is ensured to be empty */
1470 /* Given hole range was invalid (outside of device) */
1471 if (ret == -ERANGE) {
1472 *hole_start += *hole_size;
1477 *hole_start += zone_size;
1478 *hole_size -= zone_size;
1486 * dev_extent_hole_check - check if specified hole is suitable for allocation
1487 * @device: the device which we have the hole
1488 * @hole_start: starting position of the hole
1489 * @hole_size: the size of the hole
1490 * @num_bytes: the size of the free space that we need
1492 * This function may modify @hole_start and @hole_size to reflect the suitable
1493 * position for allocation. Returns 1 if hole position is updated, 0 otherwise.
1495 static bool dev_extent_hole_check(struct btrfs_device *device, u64 *hole_start,
1496 u64 *hole_size, u64 num_bytes)
1498 bool changed = false;
1499 u64 hole_end = *hole_start + *hole_size;
1503 * Check before we set max_hole_start, otherwise we could end up
1504 * sending back this offset anyway.
1506 if (contains_pending_extent(device, hole_start, *hole_size)) {
1507 if (hole_end >= *hole_start)
1508 *hole_size = hole_end - *hole_start;
1514 switch (device->fs_devices->chunk_alloc_policy) {
1515 case BTRFS_CHUNK_ALLOC_REGULAR:
1516 /* No extra check */
1518 case BTRFS_CHUNK_ALLOC_ZONED:
1519 if (dev_extent_hole_check_zoned(device, hole_start,
1520 hole_size, num_bytes)) {
1523 * The changed hole can contain pending extent.
1524 * Loop again to check that.
1540 * find_free_dev_extent_start - find free space in the specified device
1541 * @device: the device which we search the free space in
1542 * @num_bytes: the size of the free space that we need
1543 * @search_start: the position from which to begin the search
1544 * @start: store the start of the free space.
1545 * @len: the size of the free space. that we find, or the size
1546 * of the max free space if we don't find suitable free space
1548 * this uses a pretty simple search, the expectation is that it is
1549 * called very infrequently and that a given device has a small number
1552 * @start is used to store the start of the free space if we find. But if we
1553 * don't find suitable free space, it will be used to store the start position
1554 * of the max free space.
1556 * @len is used to store the size of the free space that we find.
1557 * But if we don't find suitable free space, it is used to store the size of
1558 * the max free space.
1560 * NOTE: This function will search *commit* root of device tree, and does extra
1561 * check to ensure dev extents are not double allocated.
1562 * This makes the function safe to allocate dev extents but may not report
1563 * correct usable device space, as device extent freed in current transaction
1564 * is not reported as available.
1566 static int find_free_dev_extent_start(struct btrfs_device *device,
1567 u64 num_bytes, u64 search_start, u64 *start,
1570 struct btrfs_fs_info *fs_info = device->fs_info;
1571 struct btrfs_root *root = fs_info->dev_root;
1572 struct btrfs_key key;
1573 struct btrfs_dev_extent *dev_extent;
1574 struct btrfs_path *path;
1579 u64 search_end = device->total_bytes;
1582 struct extent_buffer *l;
1584 search_start = dev_extent_search_start(device, search_start);
1586 WARN_ON(device->zone_info &&
1587 !IS_ALIGNED(num_bytes, device->zone_info->zone_size));
1589 path = btrfs_alloc_path();
1593 max_hole_start = search_start;
1597 if (search_start >= search_end ||
1598 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1603 path->reada = READA_FORWARD;
1604 path->search_commit_root = 1;
1605 path->skip_locking = 1;
1607 key.objectid = device->devid;
1608 key.offset = search_start;
1609 key.type = BTRFS_DEV_EXTENT_KEY;
1611 ret = btrfs_search_backwards(root, &key, path);
1617 slot = path->slots[0];
1618 if (slot >= btrfs_header_nritems(l)) {
1619 ret = btrfs_next_leaf(root, path);
1627 btrfs_item_key_to_cpu(l, &key, slot);
1629 if (key.objectid < device->devid)
1632 if (key.objectid > device->devid)
1635 if (key.type != BTRFS_DEV_EXTENT_KEY)
1638 if (key.offset > search_start) {
1639 hole_size = key.offset - search_start;
1640 dev_extent_hole_check(device, &search_start, &hole_size,
1643 if (hole_size > max_hole_size) {
1644 max_hole_start = search_start;
1645 max_hole_size = hole_size;
1649 * If this free space is greater than which we need,
1650 * it must be the max free space that we have found
1651 * until now, so max_hole_start must point to the start
1652 * of this free space and the length of this free space
1653 * is stored in max_hole_size. Thus, we return
1654 * max_hole_start and max_hole_size and go back to the
1657 if (hole_size >= num_bytes) {
1663 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1664 extent_end = key.offset + btrfs_dev_extent_length(l,
1666 if (extent_end > search_start)
1667 search_start = extent_end;
1674 * At this point, search_start should be the end of
1675 * allocated dev extents, and when shrinking the device,
1676 * search_end may be smaller than search_start.
1678 if (search_end > search_start) {
1679 hole_size = search_end - search_start;
1680 if (dev_extent_hole_check(device, &search_start, &hole_size,
1682 btrfs_release_path(path);
1686 if (hole_size > max_hole_size) {
1687 max_hole_start = search_start;
1688 max_hole_size = hole_size;
1693 if (max_hole_size < num_bytes)
1699 btrfs_free_path(path);
1700 *start = max_hole_start;
1702 *len = max_hole_size;
1706 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
1707 u64 *start, u64 *len)
1709 /* FIXME use last free of some kind */
1710 return find_free_dev_extent_start(device, num_bytes, 0, start, len);
1713 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1714 struct btrfs_device *device,
1715 u64 start, u64 *dev_extent_len)
1717 struct btrfs_fs_info *fs_info = device->fs_info;
1718 struct btrfs_root *root = fs_info->dev_root;
1720 struct btrfs_path *path;
1721 struct btrfs_key key;
1722 struct btrfs_key found_key;
1723 struct extent_buffer *leaf = NULL;
1724 struct btrfs_dev_extent *extent = NULL;
1726 path = btrfs_alloc_path();
1730 key.objectid = device->devid;
1732 key.type = BTRFS_DEV_EXTENT_KEY;
1734 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1736 ret = btrfs_previous_item(root, path, key.objectid,
1737 BTRFS_DEV_EXTENT_KEY);
1740 leaf = path->nodes[0];
1741 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1742 extent = btrfs_item_ptr(leaf, path->slots[0],
1743 struct btrfs_dev_extent);
1744 BUG_ON(found_key.offset > start || found_key.offset +
1745 btrfs_dev_extent_length(leaf, extent) < start);
1747 btrfs_release_path(path);
1749 } else if (ret == 0) {
1750 leaf = path->nodes[0];
1751 extent = btrfs_item_ptr(leaf, path->slots[0],
1752 struct btrfs_dev_extent);
1757 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1759 ret = btrfs_del_item(trans, root, path);
1761 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1763 btrfs_free_path(path);
1767 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1769 struct extent_map_tree *em_tree;
1770 struct extent_map *em;
1774 em_tree = &fs_info->mapping_tree;
1775 read_lock(&em_tree->lock);
1776 n = rb_last(&em_tree->map.rb_root);
1778 em = rb_entry(n, struct extent_map, rb_node);
1779 ret = em->start + em->len;
1781 read_unlock(&em_tree->lock);
1786 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1790 struct btrfs_key key;
1791 struct btrfs_key found_key;
1792 struct btrfs_path *path;
1794 path = btrfs_alloc_path();
1798 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1799 key.type = BTRFS_DEV_ITEM_KEY;
1800 key.offset = (u64)-1;
1802 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1808 btrfs_err(fs_info, "corrupted chunk tree devid -1 matched");
1813 ret = btrfs_previous_item(fs_info->chunk_root, path,
1814 BTRFS_DEV_ITEMS_OBJECTID,
1815 BTRFS_DEV_ITEM_KEY);
1819 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1821 *devid_ret = found_key.offset + 1;
1825 btrfs_free_path(path);
1830 * the device information is stored in the chunk root
1831 * the btrfs_device struct should be fully filled in
1833 static int btrfs_add_dev_item(struct btrfs_trans_handle *trans,
1834 struct btrfs_device *device)
1837 struct btrfs_path *path;
1838 struct btrfs_dev_item *dev_item;
1839 struct extent_buffer *leaf;
1840 struct btrfs_key key;
1843 path = btrfs_alloc_path();
1847 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1848 key.type = BTRFS_DEV_ITEM_KEY;
1849 key.offset = device->devid;
1851 ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path,
1852 &key, sizeof(*dev_item));
1856 leaf = path->nodes[0];
1857 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1859 btrfs_set_device_id(leaf, dev_item, device->devid);
1860 btrfs_set_device_generation(leaf, dev_item, 0);
1861 btrfs_set_device_type(leaf, dev_item, device->type);
1862 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1863 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1864 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1865 btrfs_set_device_total_bytes(leaf, dev_item,
1866 btrfs_device_get_disk_total_bytes(device));
1867 btrfs_set_device_bytes_used(leaf, dev_item,
1868 btrfs_device_get_bytes_used(device));
1869 btrfs_set_device_group(leaf, dev_item, 0);
1870 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1871 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1872 btrfs_set_device_start_offset(leaf, dev_item, 0);
1874 ptr = btrfs_device_uuid(dev_item);
1875 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1876 ptr = btrfs_device_fsid(dev_item);
1877 write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid,
1878 ptr, BTRFS_FSID_SIZE);
1879 btrfs_mark_buffer_dirty(leaf);
1883 btrfs_free_path(path);
1888 * Function to update ctime/mtime for a given device path.
1889 * Mainly used for ctime/mtime based probe like libblkid.
1891 * We don't care about errors here, this is just to be kind to userspace.
1893 static void update_dev_time(const char *device_path)
1896 struct timespec64 now;
1899 ret = kern_path(device_path, LOOKUP_FOLLOW, &path);
1903 now = current_time(d_inode(path.dentry));
1904 inode_update_time(d_inode(path.dentry), &now, S_MTIME | S_CTIME);
1908 static int btrfs_rm_dev_item(struct btrfs_device *device)
1910 struct btrfs_root *root = device->fs_info->chunk_root;
1912 struct btrfs_path *path;
1913 struct btrfs_key key;
1914 struct btrfs_trans_handle *trans;
1916 path = btrfs_alloc_path();
1920 trans = btrfs_start_transaction(root, 0);
1921 if (IS_ERR(trans)) {
1922 btrfs_free_path(path);
1923 return PTR_ERR(trans);
1925 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1926 key.type = BTRFS_DEV_ITEM_KEY;
1927 key.offset = device->devid;
1929 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1933 btrfs_abort_transaction(trans, ret);
1934 btrfs_end_transaction(trans);
1938 ret = btrfs_del_item(trans, root, path);
1940 btrfs_abort_transaction(trans, ret);
1941 btrfs_end_transaction(trans);
1945 btrfs_free_path(path);
1947 ret = btrfs_commit_transaction(trans);
1952 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1953 * filesystem. It's up to the caller to adjust that number regarding eg. device
1956 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1964 seq = read_seqbegin(&fs_info->profiles_lock);
1966 all_avail = fs_info->avail_data_alloc_bits |
1967 fs_info->avail_system_alloc_bits |
1968 fs_info->avail_metadata_alloc_bits;
1969 } while (read_seqretry(&fs_info->profiles_lock, seq));
1971 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1972 if (!(all_avail & btrfs_raid_array[i].bg_flag))
1975 if (num_devices < btrfs_raid_array[i].devs_min)
1976 return btrfs_raid_array[i].mindev_error;
1982 static struct btrfs_device * btrfs_find_next_active_device(
1983 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device)
1985 struct btrfs_device *next_device;
1987 list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1988 if (next_device != device &&
1989 !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state)
1990 && next_device->bdev)
1998 * Helper function to check if the given device is part of s_bdev / latest_dev
1999 * and replace it with the provided or the next active device, in the context
2000 * where this function called, there should be always be another device (or
2001 * this_dev) which is active.
2003 void __cold btrfs_assign_next_active_device(struct btrfs_device *device,
2004 struct btrfs_device *next_device)
2006 struct btrfs_fs_info *fs_info = device->fs_info;
2009 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
2011 ASSERT(next_device);
2013 if (fs_info->sb->s_bdev &&
2014 (fs_info->sb->s_bdev == device->bdev))
2015 fs_info->sb->s_bdev = next_device->bdev;
2017 if (fs_info->fs_devices->latest_dev->bdev == device->bdev)
2018 fs_info->fs_devices->latest_dev = next_device;
2022 * Return btrfs_fs_devices::num_devices excluding the device that's being
2023 * currently replaced.
2025 static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info)
2027 u64 num_devices = fs_info->fs_devices->num_devices;
2029 down_read(&fs_info->dev_replace.rwsem);
2030 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
2031 ASSERT(num_devices > 1);
2034 up_read(&fs_info->dev_replace.rwsem);
2039 void btrfs_scratch_superblocks(struct btrfs_fs_info *fs_info,
2040 struct block_device *bdev,
2041 const char *device_path)
2043 struct btrfs_super_block *disk_super;
2049 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; copy_num++) {
2053 disk_super = btrfs_read_dev_one_super(bdev, copy_num);
2054 if (IS_ERR(disk_super))
2057 if (bdev_is_zoned(bdev)) {
2058 btrfs_reset_sb_log_zones(bdev, copy_num);
2062 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
2064 page = virt_to_page(disk_super);
2065 set_page_dirty(page);
2067 /* write_on_page() unlocks the page */
2068 ret = write_one_page(page);
2071 "error clearing superblock number %d (%d)",
2073 btrfs_release_disk_super(disk_super);
2077 /* Notify udev that device has changed */
2078 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
2080 /* Update ctime/mtime for device path for libblkid */
2081 update_dev_time(device_path);
2084 int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path,
2085 u64 devid, struct block_device **bdev, fmode_t *mode)
2087 struct btrfs_device *device;
2088 struct btrfs_fs_devices *cur_devices;
2089 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2094 * The device list in fs_devices is accessed without locks (neither
2095 * uuid_mutex nor device_list_mutex) as it won't change on a mounted
2096 * filesystem and another device rm cannot run.
2098 num_devices = btrfs_num_devices(fs_info);
2100 ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
2104 device = btrfs_find_device_by_devspec(fs_info, devid, device_path);
2106 if (IS_ERR(device)) {
2107 if (PTR_ERR(device) == -ENOENT &&
2108 device_path && strcmp(device_path, "missing") == 0)
2109 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2111 ret = PTR_ERR(device);
2115 if (btrfs_pinned_by_swapfile(fs_info, device)) {
2116 btrfs_warn_in_rcu(fs_info,
2117 "cannot remove device %s (devid %llu) due to active swapfile",
2118 rcu_str_deref(device->name), device->devid);
2123 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2124 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
2128 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
2129 fs_info->fs_devices->rw_devices == 1) {
2130 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
2134 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2135 mutex_lock(&fs_info->chunk_mutex);
2136 list_del_init(&device->dev_alloc_list);
2137 device->fs_devices->rw_devices--;
2138 mutex_unlock(&fs_info->chunk_mutex);
2141 ret = btrfs_shrink_device(device, 0);
2143 btrfs_reada_remove_dev(device);
2148 * TODO: the superblock still includes this device in its num_devices
2149 * counter although write_all_supers() is not locked out. This
2150 * could give a filesystem state which requires a degraded mount.
2152 ret = btrfs_rm_dev_item(device);
2156 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2157 btrfs_scrub_cancel_dev(device);
2160 * the device list mutex makes sure that we don't change
2161 * the device list while someone else is writing out all
2162 * the device supers. Whoever is writing all supers, should
2163 * lock the device list mutex before getting the number of
2164 * devices in the super block (super_copy). Conversely,
2165 * whoever updates the number of devices in the super block
2166 * (super_copy) should hold the device list mutex.
2170 * In normal cases the cur_devices == fs_devices. But in case
2171 * of deleting a seed device, the cur_devices should point to
2172 * its own fs_devices listed under the fs_devices->seed.
2174 cur_devices = device->fs_devices;
2175 mutex_lock(&fs_devices->device_list_mutex);
2176 list_del_rcu(&device->dev_list);
2178 cur_devices->num_devices--;
2179 cur_devices->total_devices--;
2180 /* Update total_devices of the parent fs_devices if it's seed */
2181 if (cur_devices != fs_devices)
2182 fs_devices->total_devices--;
2184 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
2185 cur_devices->missing_devices--;
2187 btrfs_assign_next_active_device(device, NULL);
2190 cur_devices->open_devices--;
2191 /* remove sysfs entry */
2192 btrfs_sysfs_remove_device(device);
2195 num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
2196 btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
2197 mutex_unlock(&fs_devices->device_list_mutex);
2200 * At this point, the device is zero sized and detached from the
2201 * devices list. All that's left is to zero out the old supers and
2204 * We cannot call btrfs_close_bdev() here because we're holding the sb
2205 * write lock, and blkdev_put() will pull in the ->open_mutex on the
2206 * block device and it's dependencies. Instead just flush the device
2207 * and let the caller do the final blkdev_put.
2209 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2210 btrfs_scratch_superblocks(fs_info, device->bdev,
2213 sync_blockdev(device->bdev);
2214 invalidate_bdev(device->bdev);
2218 *bdev = device->bdev;
2219 *mode = device->mode;
2221 btrfs_free_device(device);
2223 if (cur_devices->open_devices == 0) {
2224 list_del_init(&cur_devices->seed_list);
2225 close_fs_devices(cur_devices);
2226 free_fs_devices(cur_devices);
2233 btrfs_reada_undo_remove_dev(device);
2234 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2235 mutex_lock(&fs_info->chunk_mutex);
2236 list_add(&device->dev_alloc_list,
2237 &fs_devices->alloc_list);
2238 device->fs_devices->rw_devices++;
2239 mutex_unlock(&fs_info->chunk_mutex);
2244 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev)
2246 struct btrfs_fs_devices *fs_devices;
2248 lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex);
2251 * in case of fs with no seed, srcdev->fs_devices will point
2252 * to fs_devices of fs_info. However when the dev being replaced is
2253 * a seed dev it will point to the seed's local fs_devices. In short
2254 * srcdev will have its correct fs_devices in both the cases.
2256 fs_devices = srcdev->fs_devices;
2258 list_del_rcu(&srcdev->dev_list);
2259 list_del(&srcdev->dev_alloc_list);
2260 fs_devices->num_devices--;
2261 if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state))
2262 fs_devices->missing_devices--;
2264 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state))
2265 fs_devices->rw_devices--;
2268 fs_devices->open_devices--;
2271 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev)
2273 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2275 mutex_lock(&uuid_mutex);
2277 btrfs_close_bdev(srcdev);
2279 btrfs_free_device(srcdev);
2281 /* if this is no devs we rather delete the fs_devices */
2282 if (!fs_devices->num_devices) {
2284 * On a mounted FS, num_devices can't be zero unless it's a
2285 * seed. In case of a seed device being replaced, the replace
2286 * target added to the sprout FS, so there will be no more
2287 * device left under the seed FS.
2289 ASSERT(fs_devices->seeding);
2291 list_del_init(&fs_devices->seed_list);
2292 close_fs_devices(fs_devices);
2293 free_fs_devices(fs_devices);
2295 mutex_unlock(&uuid_mutex);
2298 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev)
2300 struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices;
2302 mutex_lock(&fs_devices->device_list_mutex);
2304 btrfs_sysfs_remove_device(tgtdev);
2307 fs_devices->open_devices--;
2309 fs_devices->num_devices--;
2311 btrfs_assign_next_active_device(tgtdev, NULL);
2313 list_del_rcu(&tgtdev->dev_list);
2315 mutex_unlock(&fs_devices->device_list_mutex);
2317 btrfs_scratch_superblocks(tgtdev->fs_info, tgtdev->bdev,
2320 btrfs_close_bdev(tgtdev);
2322 btrfs_free_device(tgtdev);
2325 static struct btrfs_device *btrfs_find_device_by_path(
2326 struct btrfs_fs_info *fs_info, const char *device_path)
2329 struct btrfs_super_block *disk_super;
2332 struct block_device *bdev;
2333 struct btrfs_device *device;
2335 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2336 fs_info->bdev_holder, 0, &bdev, &disk_super);
2338 return ERR_PTR(ret);
2340 devid = btrfs_stack_device_id(&disk_super->dev_item);
2341 dev_uuid = disk_super->dev_item.uuid;
2342 if (btrfs_fs_incompat(fs_info, METADATA_UUID))
2343 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
2344 disk_super->metadata_uuid);
2346 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
2349 btrfs_release_disk_super(disk_super);
2351 device = ERR_PTR(-ENOENT);
2352 blkdev_put(bdev, FMODE_READ);
2357 * Lookup a device given by device id, or the path if the id is 0.
2359 struct btrfs_device *btrfs_find_device_by_devspec(
2360 struct btrfs_fs_info *fs_info, u64 devid,
2361 const char *device_path)
2363 struct btrfs_device *device;
2366 device = btrfs_find_device(fs_info->fs_devices, devid, NULL,
2369 return ERR_PTR(-ENOENT);
2373 if (!device_path || !device_path[0])
2374 return ERR_PTR(-EINVAL);
2376 if (strcmp(device_path, "missing") == 0) {
2377 /* Find first missing device */
2378 list_for_each_entry(device, &fs_info->fs_devices->devices,
2380 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
2381 &device->dev_state) && !device->bdev)
2384 return ERR_PTR(-ENOENT);
2387 return btrfs_find_device_by_path(fs_info, device_path);
2391 * does all the dirty work required for changing file system's UUID.
2393 static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info)
2395 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2396 struct btrfs_fs_devices *old_devices;
2397 struct btrfs_fs_devices *seed_devices;
2398 struct btrfs_super_block *disk_super = fs_info->super_copy;
2399 struct btrfs_device *device;
2402 lockdep_assert_held(&uuid_mutex);
2403 if (!fs_devices->seeding)
2407 * Private copy of the seed devices, anchored at
2408 * fs_info->fs_devices->seed_list
2410 seed_devices = alloc_fs_devices(NULL, NULL);
2411 if (IS_ERR(seed_devices))
2412 return PTR_ERR(seed_devices);
2415 * It's necessary to retain a copy of the original seed fs_devices in
2416 * fs_uuids so that filesystems which have been seeded can successfully
2417 * reference the seed device from open_seed_devices. This also supports
2420 old_devices = clone_fs_devices(fs_devices);
2421 if (IS_ERR(old_devices)) {
2422 kfree(seed_devices);
2423 return PTR_ERR(old_devices);
2426 list_add(&old_devices->fs_list, &fs_uuids);
2428 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2429 seed_devices->opened = 1;
2430 INIT_LIST_HEAD(&seed_devices->devices);
2431 INIT_LIST_HEAD(&seed_devices->alloc_list);
2432 mutex_init(&seed_devices->device_list_mutex);
2434 mutex_lock(&fs_devices->device_list_mutex);
2435 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2437 list_for_each_entry(device, &seed_devices->devices, dev_list)
2438 device->fs_devices = seed_devices;
2440 fs_devices->seeding = false;
2441 fs_devices->num_devices = 0;
2442 fs_devices->open_devices = 0;
2443 fs_devices->missing_devices = 0;
2444 fs_devices->rotating = false;
2445 list_add(&seed_devices->seed_list, &fs_devices->seed_list);
2447 generate_random_uuid(fs_devices->fsid);
2448 memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE);
2449 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2450 mutex_unlock(&fs_devices->device_list_mutex);
2452 super_flags = btrfs_super_flags(disk_super) &
2453 ~BTRFS_SUPER_FLAG_SEEDING;
2454 btrfs_set_super_flags(disk_super, super_flags);
2460 * Store the expected generation for seed devices in device items.
2462 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans)
2464 struct btrfs_fs_info *fs_info = trans->fs_info;
2465 struct btrfs_root *root = fs_info->chunk_root;
2466 struct btrfs_path *path;
2467 struct extent_buffer *leaf;
2468 struct btrfs_dev_item *dev_item;
2469 struct btrfs_device *device;
2470 struct btrfs_key key;
2471 u8 fs_uuid[BTRFS_FSID_SIZE];
2472 u8 dev_uuid[BTRFS_UUID_SIZE];
2476 path = btrfs_alloc_path();
2480 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2482 key.type = BTRFS_DEV_ITEM_KEY;
2485 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2489 leaf = path->nodes[0];
2491 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2492 ret = btrfs_next_leaf(root, path);
2497 leaf = path->nodes[0];
2498 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2499 btrfs_release_path(path);
2503 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2504 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2505 key.type != BTRFS_DEV_ITEM_KEY)
2508 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2509 struct btrfs_dev_item);
2510 devid = btrfs_device_id(leaf, dev_item);
2511 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2513 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2515 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
2517 BUG_ON(!device); /* Logic error */
2519 if (device->fs_devices->seeding) {
2520 btrfs_set_device_generation(leaf, dev_item,
2521 device->generation);
2522 btrfs_mark_buffer_dirty(leaf);
2530 btrfs_free_path(path);
2534 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2536 struct btrfs_root *root = fs_info->dev_root;
2537 struct request_queue *q;
2538 struct btrfs_trans_handle *trans;
2539 struct btrfs_device *device;
2540 struct block_device *bdev;
2541 struct super_block *sb = fs_info->sb;
2542 struct rcu_string *name;
2543 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2544 u64 orig_super_total_bytes;
2545 u64 orig_super_num_devices;
2546 int seeding_dev = 0;
2548 bool locked = false;
2550 if (sb_rdonly(sb) && !fs_devices->seeding)
2553 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2554 fs_info->bdev_holder);
2556 return PTR_ERR(bdev);
2558 if (!btrfs_check_device_zone_type(fs_info, bdev)) {
2563 if (fs_devices->seeding) {
2565 down_write(&sb->s_umount);
2566 mutex_lock(&uuid_mutex);
2570 sync_blockdev(bdev);
2573 list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
2574 if (device->bdev == bdev) {
2582 device = btrfs_alloc_device(fs_info, NULL, NULL);
2583 if (IS_ERR(device)) {
2584 /* we can safely leave the fs_devices entry around */
2585 ret = PTR_ERR(device);
2589 name = rcu_string_strdup(device_path, GFP_KERNEL);
2592 goto error_free_device;
2594 rcu_assign_pointer(device->name, name);
2596 device->fs_info = fs_info;
2597 device->bdev = bdev;
2599 ret = btrfs_get_dev_zone_info(device, false);
2601 goto error_free_device;
2603 trans = btrfs_start_transaction(root, 0);
2604 if (IS_ERR(trans)) {
2605 ret = PTR_ERR(trans);
2606 goto error_free_zone;
2609 q = bdev_get_queue(bdev);
2610 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
2611 device->generation = trans->transid;
2612 device->io_width = fs_info->sectorsize;
2613 device->io_align = fs_info->sectorsize;
2614 device->sector_size = fs_info->sectorsize;
2615 device->total_bytes = round_down(i_size_read(bdev->bd_inode),
2616 fs_info->sectorsize);
2617 device->disk_total_bytes = device->total_bytes;
2618 device->commit_total_bytes = device->total_bytes;
2619 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2620 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
2621 device->mode = FMODE_EXCL;
2622 device->dev_stats_valid = 1;
2623 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2626 btrfs_clear_sb_rdonly(sb);
2627 ret = btrfs_prepare_sprout(fs_info);
2629 btrfs_abort_transaction(trans, ret);
2632 btrfs_assign_next_active_device(fs_info->fs_devices->latest_dev,
2636 device->fs_devices = fs_devices;
2638 mutex_lock(&fs_devices->device_list_mutex);
2639 mutex_lock(&fs_info->chunk_mutex);
2640 list_add_rcu(&device->dev_list, &fs_devices->devices);
2641 list_add(&device->dev_alloc_list, &fs_devices->alloc_list);
2642 fs_devices->num_devices++;
2643 fs_devices->open_devices++;
2644 fs_devices->rw_devices++;
2645 fs_devices->total_devices++;
2646 fs_devices->total_rw_bytes += device->total_bytes;
2648 atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2650 if (!blk_queue_nonrot(q))
2651 fs_devices->rotating = true;
2653 orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
2654 btrfs_set_super_total_bytes(fs_info->super_copy,
2655 round_down(orig_super_total_bytes + device->total_bytes,
2656 fs_info->sectorsize));
2658 orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy);
2659 btrfs_set_super_num_devices(fs_info->super_copy,
2660 orig_super_num_devices + 1);
2663 * we've got more storage, clear any full flags on the space
2666 btrfs_clear_space_info_full(fs_info);
2668 mutex_unlock(&fs_info->chunk_mutex);
2670 /* Add sysfs device entry */
2671 btrfs_sysfs_add_device(device);
2673 mutex_unlock(&fs_devices->device_list_mutex);
2676 mutex_lock(&fs_info->chunk_mutex);
2677 ret = init_first_rw_device(trans);
2678 mutex_unlock(&fs_info->chunk_mutex);
2680 btrfs_abort_transaction(trans, ret);
2685 ret = btrfs_add_dev_item(trans, device);
2687 btrfs_abort_transaction(trans, ret);
2692 ret = btrfs_finish_sprout(trans);
2694 btrfs_abort_transaction(trans, ret);
2699 * fs_devices now represents the newly sprouted filesystem and
2700 * its fsid has been changed by btrfs_prepare_sprout
2702 btrfs_sysfs_update_sprout_fsid(fs_devices);
2705 ret = btrfs_commit_transaction(trans);
2708 mutex_unlock(&uuid_mutex);
2709 up_write(&sb->s_umount);
2712 if (ret) /* transaction commit */
2715 ret = btrfs_relocate_sys_chunks(fs_info);
2717 btrfs_handle_fs_error(fs_info, ret,
2718 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2719 trans = btrfs_attach_transaction(root);
2720 if (IS_ERR(trans)) {
2721 if (PTR_ERR(trans) == -ENOENT)
2723 ret = PTR_ERR(trans);
2727 ret = btrfs_commit_transaction(trans);
2731 * Now that we have written a new super block to this device, check all
2732 * other fs_devices list if device_path alienates any other scanned
2734 * We can ignore the return value as it typically returns -EINVAL and
2735 * only succeeds if the device was an alien.
2737 btrfs_forget_devices(device_path);
2739 /* Update ctime/mtime for blkid or udev */
2740 update_dev_time(device_path);
2745 btrfs_sysfs_remove_device(device);
2746 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2747 mutex_lock(&fs_info->chunk_mutex);
2748 list_del_rcu(&device->dev_list);
2749 list_del(&device->dev_alloc_list);
2750 fs_info->fs_devices->num_devices--;
2751 fs_info->fs_devices->open_devices--;
2752 fs_info->fs_devices->rw_devices--;
2753 fs_info->fs_devices->total_devices--;
2754 fs_info->fs_devices->total_rw_bytes -= device->total_bytes;
2755 atomic64_sub(device->total_bytes, &fs_info->free_chunk_space);
2756 btrfs_set_super_total_bytes(fs_info->super_copy,
2757 orig_super_total_bytes);
2758 btrfs_set_super_num_devices(fs_info->super_copy,
2759 orig_super_num_devices);
2760 mutex_unlock(&fs_info->chunk_mutex);
2761 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2764 btrfs_set_sb_rdonly(sb);
2766 btrfs_end_transaction(trans);
2768 btrfs_destroy_dev_zone_info(device);
2770 btrfs_free_device(device);
2772 blkdev_put(bdev, FMODE_EXCL);
2774 mutex_unlock(&uuid_mutex);
2775 up_write(&sb->s_umount);
2780 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2781 struct btrfs_device *device)
2784 struct btrfs_path *path;
2785 struct btrfs_root *root = device->fs_info->chunk_root;
2786 struct btrfs_dev_item *dev_item;
2787 struct extent_buffer *leaf;
2788 struct btrfs_key key;
2790 path = btrfs_alloc_path();
2794 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2795 key.type = BTRFS_DEV_ITEM_KEY;
2796 key.offset = device->devid;
2798 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2807 leaf = path->nodes[0];
2808 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2810 btrfs_set_device_id(leaf, dev_item, device->devid);
2811 btrfs_set_device_type(leaf, dev_item, device->type);
2812 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2813 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2814 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2815 btrfs_set_device_total_bytes(leaf, dev_item,
2816 btrfs_device_get_disk_total_bytes(device));
2817 btrfs_set_device_bytes_used(leaf, dev_item,
2818 btrfs_device_get_bytes_used(device));
2819 btrfs_mark_buffer_dirty(leaf);
2822 btrfs_free_path(path);
2826 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2827 struct btrfs_device *device, u64 new_size)
2829 struct btrfs_fs_info *fs_info = device->fs_info;
2830 struct btrfs_super_block *super_copy = fs_info->super_copy;
2834 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2837 new_size = round_down(new_size, fs_info->sectorsize);
2839 mutex_lock(&fs_info->chunk_mutex);
2840 old_total = btrfs_super_total_bytes(super_copy);
2841 diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2843 if (new_size <= device->total_bytes ||
2844 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2845 mutex_unlock(&fs_info->chunk_mutex);
2849 btrfs_set_super_total_bytes(super_copy,
2850 round_down(old_total + diff, fs_info->sectorsize));
2851 device->fs_devices->total_rw_bytes += diff;
2853 btrfs_device_set_total_bytes(device, new_size);
2854 btrfs_device_set_disk_total_bytes(device, new_size);
2855 btrfs_clear_space_info_full(device->fs_info);
2856 if (list_empty(&device->post_commit_list))
2857 list_add_tail(&device->post_commit_list,
2858 &trans->transaction->dev_update_list);
2859 mutex_unlock(&fs_info->chunk_mutex);
2861 return btrfs_update_device(trans, device);
2864 static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
2866 struct btrfs_fs_info *fs_info = trans->fs_info;
2867 struct btrfs_root *root = fs_info->chunk_root;
2869 struct btrfs_path *path;
2870 struct btrfs_key key;
2872 path = btrfs_alloc_path();
2876 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2877 key.offset = chunk_offset;
2878 key.type = BTRFS_CHUNK_ITEM_KEY;
2880 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2883 else if (ret > 0) { /* Logic error or corruption */
2884 btrfs_handle_fs_error(fs_info, -ENOENT,
2885 "Failed lookup while freeing chunk.");
2890 ret = btrfs_del_item(trans, root, path);
2892 btrfs_handle_fs_error(fs_info, ret,
2893 "Failed to delete chunk item.");
2895 btrfs_free_path(path);
2899 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2901 struct btrfs_super_block *super_copy = fs_info->super_copy;
2902 struct btrfs_disk_key *disk_key;
2903 struct btrfs_chunk *chunk;
2910 struct btrfs_key key;
2912 lockdep_assert_held(&fs_info->chunk_mutex);
2913 array_size = btrfs_super_sys_array_size(super_copy);
2915 ptr = super_copy->sys_chunk_array;
2918 while (cur < array_size) {
2919 disk_key = (struct btrfs_disk_key *)ptr;
2920 btrfs_disk_key_to_cpu(&key, disk_key);
2922 len = sizeof(*disk_key);
2924 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2925 chunk = (struct btrfs_chunk *)(ptr + len);
2926 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2927 len += btrfs_chunk_item_size(num_stripes);
2932 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID &&
2933 key.offset == chunk_offset) {
2934 memmove(ptr, ptr + len, array_size - (cur + len));
2936 btrfs_set_super_sys_array_size(super_copy, array_size);
2946 * btrfs_get_chunk_map() - Find the mapping containing the given logical extent.
2947 * @logical: Logical block offset in bytes.
2948 * @length: Length of extent in bytes.
2950 * Return: Chunk mapping or ERR_PTR.
2952 struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info,
2953 u64 logical, u64 length)
2955 struct extent_map_tree *em_tree;
2956 struct extent_map *em;
2958 em_tree = &fs_info->mapping_tree;
2959 read_lock(&em_tree->lock);
2960 em = lookup_extent_mapping(em_tree, logical, length);
2961 read_unlock(&em_tree->lock);
2964 btrfs_crit(fs_info, "unable to find logical %llu length %llu",
2966 return ERR_PTR(-EINVAL);
2969 if (em->start > logical || em->start + em->len < logical) {
2971 "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
2972 logical, length, em->start, em->start + em->len);
2973 free_extent_map(em);
2974 return ERR_PTR(-EINVAL);
2977 /* callers are responsible for dropping em's ref. */
2981 static int remove_chunk_item(struct btrfs_trans_handle *trans,
2982 struct map_lookup *map, u64 chunk_offset)
2987 * Removing chunk items and updating the device items in the chunks btree
2988 * requires holding the chunk_mutex.
2989 * See the comment at btrfs_chunk_alloc() for the details.
2991 lockdep_assert_held(&trans->fs_info->chunk_mutex);
2993 for (i = 0; i < map->num_stripes; i++) {
2996 ret = btrfs_update_device(trans, map->stripes[i].dev);
3001 return btrfs_free_chunk(trans, chunk_offset);
3004 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
3006 struct btrfs_fs_info *fs_info = trans->fs_info;
3007 struct extent_map *em;
3008 struct map_lookup *map;
3009 u64 dev_extent_len = 0;
3011 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
3013 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
3016 * This is a logic error, but we don't want to just rely on the
3017 * user having built with ASSERT enabled, so if ASSERT doesn't
3018 * do anything we still error out.
3023 map = em->map_lookup;
3026 * First delete the device extent items from the devices btree.
3027 * We take the device_list_mutex to avoid racing with the finishing phase
3028 * of a device replace operation. See the comment below before acquiring
3029 * fs_info->chunk_mutex. Note that here we do not acquire the chunk_mutex
3030 * because that can result in a deadlock when deleting the device extent
3031 * items from the devices btree - COWing an extent buffer from the btree
3032 * may result in allocating a new metadata chunk, which would attempt to
3033 * lock again fs_info->chunk_mutex.
3035 mutex_lock(&fs_devices->device_list_mutex);
3036 for (i = 0; i < map->num_stripes; i++) {
3037 struct btrfs_device *device = map->stripes[i].dev;
3038 ret = btrfs_free_dev_extent(trans, device,
3039 map->stripes[i].physical,
3042 mutex_unlock(&fs_devices->device_list_mutex);
3043 btrfs_abort_transaction(trans, ret);
3047 if (device->bytes_used > 0) {
3048 mutex_lock(&fs_info->chunk_mutex);
3049 btrfs_device_set_bytes_used(device,
3050 device->bytes_used - dev_extent_len);
3051 atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
3052 btrfs_clear_space_info_full(fs_info);
3053 mutex_unlock(&fs_info->chunk_mutex);
3056 mutex_unlock(&fs_devices->device_list_mutex);
3059 * We acquire fs_info->chunk_mutex for 2 reasons:
3061 * 1) Just like with the first phase of the chunk allocation, we must
3062 * reserve system space, do all chunk btree updates and deletions, and
3063 * update the system chunk array in the superblock while holding this
3064 * mutex. This is for similar reasons as explained on the comment at
3065 * the top of btrfs_chunk_alloc();
3067 * 2) Prevent races with the final phase of a device replace operation
3068 * that replaces the device object associated with the map's stripes,
3069 * because the device object's id can change at any time during that
3070 * final phase of the device replace operation
3071 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the
3072 * replaced device and then see it with an ID of
3073 * BTRFS_DEV_REPLACE_DEVID, which would cause a failure when updating
3074 * the device item, which does not exists on the chunk btree.
3075 * The finishing phase of device replace acquires both the
3076 * device_list_mutex and the chunk_mutex, in that order, so we are
3077 * safe by just acquiring the chunk_mutex.
3079 trans->removing_chunk = true;
3080 mutex_lock(&fs_info->chunk_mutex);
3082 check_system_chunk(trans, map->type);
3084 ret = remove_chunk_item(trans, map, chunk_offset);
3086 * Normally we should not get -ENOSPC since we reserved space before
3087 * through the call to check_system_chunk().
3089 * Despite our system space_info having enough free space, we may not
3090 * be able to allocate extents from its block groups, because all have
3091 * an incompatible profile, which will force us to allocate a new system
3092 * block group with the right profile, or right after we called
3093 * check_system_space() above, a scrub turned the only system block group
3094 * with enough free space into RO mode.
3095 * This is explained with more detail at do_chunk_alloc().
3097 * So if we get -ENOSPC, allocate a new system chunk and retry once.
3099 if (ret == -ENOSPC) {
3100 const u64 sys_flags = btrfs_system_alloc_profile(fs_info);
3101 struct btrfs_block_group *sys_bg;
3103 sys_bg = btrfs_alloc_chunk(trans, sys_flags);
3104 if (IS_ERR(sys_bg)) {
3105 ret = PTR_ERR(sys_bg);
3106 btrfs_abort_transaction(trans, ret);
3110 ret = btrfs_chunk_alloc_add_chunk_item(trans, sys_bg);
3112 btrfs_abort_transaction(trans, ret);
3116 ret = remove_chunk_item(trans, map, chunk_offset);
3118 btrfs_abort_transaction(trans, ret);
3122 btrfs_abort_transaction(trans, ret);
3126 trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
3128 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
3129 ret = btrfs_del_sys_chunk(fs_info, chunk_offset);
3131 btrfs_abort_transaction(trans, ret);
3136 mutex_unlock(&fs_info->chunk_mutex);
3137 trans->removing_chunk = false;
3140 * We are done with chunk btree updates and deletions, so release the
3141 * system space we previously reserved (with check_system_chunk()).
3143 btrfs_trans_release_chunk_metadata(trans);
3145 ret = btrfs_remove_block_group(trans, chunk_offset, em);
3147 btrfs_abort_transaction(trans, ret);
3152 if (trans->removing_chunk) {
3153 mutex_unlock(&fs_info->chunk_mutex);
3154 trans->removing_chunk = false;
3157 free_extent_map(em);
3161 int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
3163 struct btrfs_root *root = fs_info->chunk_root;
3164 struct btrfs_trans_handle *trans;
3165 struct btrfs_block_group *block_group;
3170 * Prevent races with automatic removal of unused block groups.
3171 * After we relocate and before we remove the chunk with offset
3172 * chunk_offset, automatic removal of the block group can kick in,
3173 * resulting in a failure when calling btrfs_remove_chunk() below.
3175 * Make sure to acquire this mutex before doing a tree search (dev
3176 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
3177 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
3178 * we release the path used to search the chunk/dev tree and before
3179 * the current task acquires this mutex and calls us.
3181 lockdep_assert_held(&fs_info->reclaim_bgs_lock);
3183 /* step one, relocate all the extents inside this chunk */
3184 btrfs_scrub_pause(fs_info);
3185 ret = btrfs_relocate_block_group(fs_info, chunk_offset);
3186 btrfs_scrub_continue(fs_info);
3190 block_group = btrfs_lookup_block_group(fs_info, chunk_offset);
3193 btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group);
3194 length = block_group->length;
3195 btrfs_put_block_group(block_group);
3198 * On a zoned file system, discard the whole block group, this will
3199 * trigger a REQ_OP_ZONE_RESET operation on the device zone. If
3200 * resetting the zone fails, don't treat it as a fatal problem from the
3201 * filesystem's point of view.
3203 if (btrfs_is_zoned(fs_info)) {
3204 ret = btrfs_discard_extent(fs_info, chunk_offset, length, NULL);
3207 "failed to reset zone %llu after relocation",
3211 trans = btrfs_start_trans_remove_block_group(root->fs_info,
3213 if (IS_ERR(trans)) {
3214 ret = PTR_ERR(trans);
3215 btrfs_handle_fs_error(root->fs_info, ret, NULL);
3220 * step two, delete the device extents and the
3221 * chunk tree entries
3223 ret = btrfs_remove_chunk(trans, chunk_offset);
3224 btrfs_end_transaction(trans);
3228 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
3230 struct btrfs_root *chunk_root = fs_info->chunk_root;
3231 struct btrfs_path *path;
3232 struct extent_buffer *leaf;
3233 struct btrfs_chunk *chunk;
3234 struct btrfs_key key;
3235 struct btrfs_key found_key;
3237 bool retried = false;
3241 path = btrfs_alloc_path();
3246 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3247 key.offset = (u64)-1;
3248 key.type = BTRFS_CHUNK_ITEM_KEY;
3251 mutex_lock(&fs_info->reclaim_bgs_lock);
3252 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3254 mutex_unlock(&fs_info->reclaim_bgs_lock);
3257 BUG_ON(ret == 0); /* Corruption */
3259 ret = btrfs_previous_item(chunk_root, path, key.objectid,
3262 mutex_unlock(&fs_info->reclaim_bgs_lock);
3268 leaf = path->nodes[0];
3269 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3271 chunk = btrfs_item_ptr(leaf, path->slots[0],
3272 struct btrfs_chunk);
3273 chunk_type = btrfs_chunk_type(leaf, chunk);
3274 btrfs_release_path(path);
3276 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3277 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3283 mutex_unlock(&fs_info->reclaim_bgs_lock);
3285 if (found_key.offset == 0)
3287 key.offset = found_key.offset - 1;
3290 if (failed && !retried) {
3294 } else if (WARN_ON(failed && retried)) {
3298 btrfs_free_path(path);
3303 * return 1 : allocate a data chunk successfully,
3304 * return <0: errors during allocating a data chunk,
3305 * return 0 : no need to allocate a data chunk.
3307 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info,
3310 struct btrfs_block_group *cache;
3314 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3316 chunk_type = cache->flags;
3317 btrfs_put_block_group(cache);
3319 if (!(chunk_type & BTRFS_BLOCK_GROUP_DATA))
3322 spin_lock(&fs_info->data_sinfo->lock);
3323 bytes_used = fs_info->data_sinfo->bytes_used;
3324 spin_unlock(&fs_info->data_sinfo->lock);
3327 struct btrfs_trans_handle *trans;
3330 trans = btrfs_join_transaction(fs_info->tree_root);
3332 return PTR_ERR(trans);
3334 ret = btrfs_force_chunk_alloc(trans, BTRFS_BLOCK_GROUP_DATA);
3335 btrfs_end_transaction(trans);
3344 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3345 struct btrfs_balance_control *bctl)
3347 struct btrfs_root *root = fs_info->tree_root;
3348 struct btrfs_trans_handle *trans;
3349 struct btrfs_balance_item *item;
3350 struct btrfs_disk_balance_args disk_bargs;
3351 struct btrfs_path *path;
3352 struct extent_buffer *leaf;
3353 struct btrfs_key key;
3356 path = btrfs_alloc_path();
3360 trans = btrfs_start_transaction(root, 0);
3361 if (IS_ERR(trans)) {
3362 btrfs_free_path(path);
3363 return PTR_ERR(trans);
3366 key.objectid = BTRFS_BALANCE_OBJECTID;
3367 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3370 ret = btrfs_insert_empty_item(trans, root, path, &key,
3375 leaf = path->nodes[0];
3376 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3378 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3380 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3381 btrfs_set_balance_data(leaf, item, &disk_bargs);
3382 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3383 btrfs_set_balance_meta(leaf, item, &disk_bargs);
3384 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3385 btrfs_set_balance_sys(leaf, item, &disk_bargs);
3387 btrfs_set_balance_flags(leaf, item, bctl->flags);
3389 btrfs_mark_buffer_dirty(leaf);
3391 btrfs_free_path(path);
3392 err = btrfs_commit_transaction(trans);
3398 static int del_balance_item(struct btrfs_fs_info *fs_info)
3400 struct btrfs_root *root = fs_info->tree_root;
3401 struct btrfs_trans_handle *trans;
3402 struct btrfs_path *path;
3403 struct btrfs_key key;
3406 path = btrfs_alloc_path();
3410 trans = btrfs_start_transaction_fallback_global_rsv(root, 0);
3411 if (IS_ERR(trans)) {
3412 btrfs_free_path(path);
3413 return PTR_ERR(trans);
3416 key.objectid = BTRFS_BALANCE_OBJECTID;
3417 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3420 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3428 ret = btrfs_del_item(trans, root, path);
3430 btrfs_free_path(path);
3431 err = btrfs_commit_transaction(trans);
3438 * This is a heuristic used to reduce the number of chunks balanced on
3439 * resume after balance was interrupted.
3441 static void update_balance_args(struct btrfs_balance_control *bctl)
3444 * Turn on soft mode for chunk types that were being converted.
3446 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3447 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3448 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3449 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3450 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3451 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3454 * Turn on usage filter if is not already used. The idea is
3455 * that chunks that we have already balanced should be
3456 * reasonably full. Don't do it for chunks that are being
3457 * converted - that will keep us from relocating unconverted
3458 * (albeit full) chunks.
3460 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3461 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3462 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3463 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3464 bctl->data.usage = 90;
3466 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3467 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3468 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3469 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3470 bctl->sys.usage = 90;
3472 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3473 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3474 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3475 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3476 bctl->meta.usage = 90;
3481 * Clear the balance status in fs_info and delete the balance item from disk.
3483 static void reset_balance_state(struct btrfs_fs_info *fs_info)
3485 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3488 BUG_ON(!fs_info->balance_ctl);
3490 spin_lock(&fs_info->balance_lock);
3491 fs_info->balance_ctl = NULL;
3492 spin_unlock(&fs_info->balance_lock);
3495 ret = del_balance_item(fs_info);
3497 btrfs_handle_fs_error(fs_info, ret, NULL);
3501 * Balance filters. Return 1 if chunk should be filtered out
3502 * (should not be balanced).
3504 static int chunk_profiles_filter(u64 chunk_type,
3505 struct btrfs_balance_args *bargs)
3507 chunk_type = chunk_to_extended(chunk_type) &
3508 BTRFS_EXTENDED_PROFILE_MASK;
3510 if (bargs->profiles & chunk_type)
3516 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3517 struct btrfs_balance_args *bargs)
3519 struct btrfs_block_group *cache;
3521 u64 user_thresh_min;
3522 u64 user_thresh_max;
3525 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3526 chunk_used = cache->used;
3528 if (bargs->usage_min == 0)
3529 user_thresh_min = 0;
3531 user_thresh_min = div_factor_fine(cache->length,
3534 if (bargs->usage_max == 0)
3535 user_thresh_max = 1;
3536 else if (bargs->usage_max > 100)
3537 user_thresh_max = cache->length;
3539 user_thresh_max = div_factor_fine(cache->length,
3542 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3545 btrfs_put_block_group(cache);
3549 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3550 u64 chunk_offset, struct btrfs_balance_args *bargs)
3552 struct btrfs_block_group *cache;
3553 u64 chunk_used, user_thresh;
3556 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3557 chunk_used = cache->used;
3559 if (bargs->usage_min == 0)
3561 else if (bargs->usage > 100)
3562 user_thresh = cache->length;
3564 user_thresh = div_factor_fine(cache->length, bargs->usage);
3566 if (chunk_used < user_thresh)
3569 btrfs_put_block_group(cache);
3573 static int chunk_devid_filter(struct extent_buffer *leaf,
3574 struct btrfs_chunk *chunk,
3575 struct btrfs_balance_args *bargs)
3577 struct btrfs_stripe *stripe;
3578 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3581 for (i = 0; i < num_stripes; i++) {
3582 stripe = btrfs_stripe_nr(chunk, i);
3583 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3590 static u64 calc_data_stripes(u64 type, int num_stripes)
3592 const int index = btrfs_bg_flags_to_raid_index(type);
3593 const int ncopies = btrfs_raid_array[index].ncopies;
3594 const int nparity = btrfs_raid_array[index].nparity;
3596 return (num_stripes - nparity) / ncopies;
3599 /* [pstart, pend) */
3600 static int chunk_drange_filter(struct extent_buffer *leaf,
3601 struct btrfs_chunk *chunk,
3602 struct btrfs_balance_args *bargs)
3604 struct btrfs_stripe *stripe;
3605 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3612 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3615 type = btrfs_chunk_type(leaf, chunk);
3616 factor = calc_data_stripes(type, num_stripes);
3618 for (i = 0; i < num_stripes; i++) {
3619 stripe = btrfs_stripe_nr(chunk, i);
3620 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3623 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3624 stripe_length = btrfs_chunk_length(leaf, chunk);
3625 stripe_length = div_u64(stripe_length, factor);
3627 if (stripe_offset < bargs->pend &&
3628 stripe_offset + stripe_length > bargs->pstart)
3635 /* [vstart, vend) */
3636 static int chunk_vrange_filter(struct extent_buffer *leaf,
3637 struct btrfs_chunk *chunk,
3639 struct btrfs_balance_args *bargs)
3641 if (chunk_offset < bargs->vend &&
3642 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3643 /* at least part of the chunk is inside this vrange */
3649 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3650 struct btrfs_chunk *chunk,
3651 struct btrfs_balance_args *bargs)
3653 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3655 if (bargs->stripes_min <= num_stripes
3656 && num_stripes <= bargs->stripes_max)
3662 static int chunk_soft_convert_filter(u64 chunk_type,
3663 struct btrfs_balance_args *bargs)
3665 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3668 chunk_type = chunk_to_extended(chunk_type) &
3669 BTRFS_EXTENDED_PROFILE_MASK;
3671 if (bargs->target == chunk_type)
3677 static int should_balance_chunk(struct extent_buffer *leaf,
3678 struct btrfs_chunk *chunk, u64 chunk_offset)
3680 struct btrfs_fs_info *fs_info = leaf->fs_info;
3681 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3682 struct btrfs_balance_args *bargs = NULL;
3683 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3686 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3687 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3691 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3692 bargs = &bctl->data;
3693 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3695 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3696 bargs = &bctl->meta;
3698 /* profiles filter */
3699 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3700 chunk_profiles_filter(chunk_type, bargs)) {
3705 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3706 chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3708 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3709 chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3714 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3715 chunk_devid_filter(leaf, chunk, bargs)) {
3719 /* drange filter, makes sense only with devid filter */
3720 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3721 chunk_drange_filter(leaf, chunk, bargs)) {
3726 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3727 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3731 /* stripes filter */
3732 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3733 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3737 /* soft profile changing mode */
3738 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3739 chunk_soft_convert_filter(chunk_type, bargs)) {
3744 * limited by count, must be the last filter
3746 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3747 if (bargs->limit == 0)
3751 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3753 * Same logic as the 'limit' filter; the minimum cannot be
3754 * determined here because we do not have the global information
3755 * about the count of all chunks that satisfy the filters.
3757 if (bargs->limit_max == 0)
3766 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3768 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3769 struct btrfs_root *chunk_root = fs_info->chunk_root;
3771 struct btrfs_chunk *chunk;
3772 struct btrfs_path *path = NULL;
3773 struct btrfs_key key;
3774 struct btrfs_key found_key;
3775 struct extent_buffer *leaf;
3778 int enospc_errors = 0;
3779 bool counting = true;
3780 /* The single value limit and min/max limits use the same bytes in the */
3781 u64 limit_data = bctl->data.limit;
3782 u64 limit_meta = bctl->meta.limit;
3783 u64 limit_sys = bctl->sys.limit;
3787 int chunk_reserved = 0;
3789 path = btrfs_alloc_path();
3795 /* zero out stat counters */
3796 spin_lock(&fs_info->balance_lock);
3797 memset(&bctl->stat, 0, sizeof(bctl->stat));
3798 spin_unlock(&fs_info->balance_lock);
3802 * The single value limit and min/max limits use the same bytes
3805 bctl->data.limit = limit_data;
3806 bctl->meta.limit = limit_meta;
3807 bctl->sys.limit = limit_sys;
3809 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3810 key.offset = (u64)-1;
3811 key.type = BTRFS_CHUNK_ITEM_KEY;
3814 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3815 atomic_read(&fs_info->balance_cancel_req)) {
3820 mutex_lock(&fs_info->reclaim_bgs_lock);
3821 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3823 mutex_unlock(&fs_info->reclaim_bgs_lock);
3828 * this shouldn't happen, it means the last relocate
3832 BUG(); /* FIXME break ? */
3834 ret = btrfs_previous_item(chunk_root, path, 0,
3835 BTRFS_CHUNK_ITEM_KEY);
3837 mutex_unlock(&fs_info->reclaim_bgs_lock);
3842 leaf = path->nodes[0];
3843 slot = path->slots[0];
3844 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3846 if (found_key.objectid != key.objectid) {
3847 mutex_unlock(&fs_info->reclaim_bgs_lock);
3851 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3852 chunk_type = btrfs_chunk_type(leaf, chunk);
3855 spin_lock(&fs_info->balance_lock);
3856 bctl->stat.considered++;
3857 spin_unlock(&fs_info->balance_lock);
3860 ret = should_balance_chunk(leaf, chunk, found_key.offset);
3862 btrfs_release_path(path);
3864 mutex_unlock(&fs_info->reclaim_bgs_lock);
3869 mutex_unlock(&fs_info->reclaim_bgs_lock);
3870 spin_lock(&fs_info->balance_lock);
3871 bctl->stat.expected++;
3872 spin_unlock(&fs_info->balance_lock);
3874 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3876 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3878 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3885 * Apply limit_min filter, no need to check if the LIMITS
3886 * filter is used, limit_min is 0 by default
3888 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3889 count_data < bctl->data.limit_min)
3890 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3891 count_meta < bctl->meta.limit_min)
3892 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3893 count_sys < bctl->sys.limit_min)) {
3894 mutex_unlock(&fs_info->reclaim_bgs_lock);
3898 if (!chunk_reserved) {
3900 * We may be relocating the only data chunk we have,
3901 * which could potentially end up with losing data's
3902 * raid profile, so lets allocate an empty one in
3905 ret = btrfs_may_alloc_data_chunk(fs_info,
3908 mutex_unlock(&fs_info->reclaim_bgs_lock);
3910 } else if (ret == 1) {
3915 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3916 mutex_unlock(&fs_info->reclaim_bgs_lock);
3917 if (ret == -ENOSPC) {
3919 } else if (ret == -ETXTBSY) {
3921 "skipping relocation of block group %llu due to active swapfile",
3927 spin_lock(&fs_info->balance_lock);
3928 bctl->stat.completed++;
3929 spin_unlock(&fs_info->balance_lock);
3932 if (found_key.offset == 0)
3934 key.offset = found_key.offset - 1;
3938 btrfs_release_path(path);
3943 btrfs_free_path(path);
3944 if (enospc_errors) {
3945 btrfs_info(fs_info, "%d enospc errors during balance",
3955 * alloc_profile_is_valid - see if a given profile is valid and reduced
3956 * @flags: profile to validate
3957 * @extended: if true @flags is treated as an extended profile
3959 static int alloc_profile_is_valid(u64 flags, int extended)
3961 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3962 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3964 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3966 /* 1) check that all other bits are zeroed */
3970 /* 2) see if profile is reduced */
3972 return !extended; /* "0" is valid for usual profiles */
3974 return has_single_bit_set(flags);
3977 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3979 /* cancel requested || normal exit path */
3980 return atomic_read(&fs_info->balance_cancel_req) ||
3981 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3982 atomic_read(&fs_info->balance_cancel_req) == 0);
3986 * Validate target profile against allowed profiles and return true if it's OK.
3987 * Otherwise print the error message and return false.
3989 static inline int validate_convert_profile(struct btrfs_fs_info *fs_info,
3990 const struct btrfs_balance_args *bargs,
3991 u64 allowed, const char *type)
3993 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3996 if (fs_info->sectorsize < PAGE_SIZE &&
3997 bargs->target & BTRFS_BLOCK_GROUP_RAID56_MASK) {
3999 "RAID56 is not yet supported for sectorsize %u with page size %lu",
4000 fs_info->sectorsize, PAGE_SIZE);
4003 /* Profile is valid and does not have bits outside of the allowed set */
4004 if (alloc_profile_is_valid(bargs->target, 1) &&
4005 (bargs->target & ~allowed) == 0)
4008 btrfs_err(fs_info, "balance: invalid convert %s profile %s",
4009 type, btrfs_bg_type_to_raid_name(bargs->target));
4014 * Fill @buf with textual description of balance filter flags @bargs, up to
4015 * @size_buf including the terminating null. The output may be trimmed if it
4016 * does not fit into the provided buffer.
4018 static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf,
4022 u32 size_bp = size_buf;
4024 u64 flags = bargs->flags;
4025 char tmp_buf[128] = {'\0'};
4030 #define CHECK_APPEND_NOARG(a) \
4032 ret = snprintf(bp, size_bp, (a)); \
4033 if (ret < 0 || ret >= size_bp) \
4034 goto out_overflow; \
4039 #define CHECK_APPEND_1ARG(a, v1) \
4041 ret = snprintf(bp, size_bp, (a), (v1)); \
4042 if (ret < 0 || ret >= size_bp) \
4043 goto out_overflow; \
4048 #define CHECK_APPEND_2ARG(a, v1, v2) \
4050 ret = snprintf(bp, size_bp, (a), (v1), (v2)); \
4051 if (ret < 0 || ret >= size_bp) \
4052 goto out_overflow; \
4057 if (flags & BTRFS_BALANCE_ARGS_CONVERT)
4058 CHECK_APPEND_1ARG("convert=%s,",
4059 btrfs_bg_type_to_raid_name(bargs->target));
4061 if (flags & BTRFS_BALANCE_ARGS_SOFT)
4062 CHECK_APPEND_NOARG("soft,");
4064 if (flags & BTRFS_BALANCE_ARGS_PROFILES) {
4065 btrfs_describe_block_groups(bargs->profiles, tmp_buf,
4067 CHECK_APPEND_1ARG("profiles=%s,", tmp_buf);
4070 if (flags & BTRFS_BALANCE_ARGS_USAGE)
4071 CHECK_APPEND_1ARG("usage=%llu,", bargs->usage);
4073 if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE)
4074 CHECK_APPEND_2ARG("usage=%u..%u,",
4075 bargs->usage_min, bargs->usage_max);
4077 if (flags & BTRFS_BALANCE_ARGS_DEVID)
4078 CHECK_APPEND_1ARG("devid=%llu,", bargs->devid);
4080 if (flags & BTRFS_BALANCE_ARGS_DRANGE)
4081 CHECK_APPEND_2ARG("drange=%llu..%llu,",
4082 bargs->pstart, bargs->pend);
4084 if (flags & BTRFS_BALANCE_ARGS_VRANGE)
4085 CHECK_APPEND_2ARG("vrange=%llu..%llu,",
4086 bargs->vstart, bargs->vend);
4088 if (flags & BTRFS_BALANCE_ARGS_LIMIT)
4089 CHECK_APPEND_1ARG("limit=%llu,", bargs->limit);
4091 if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)
4092 CHECK_APPEND_2ARG("limit=%u..%u,",
4093 bargs->limit_min, bargs->limit_max);
4095 if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE)
4096 CHECK_APPEND_2ARG("stripes=%u..%u,",
4097 bargs->stripes_min, bargs->stripes_max);
4099 #undef CHECK_APPEND_2ARG
4100 #undef CHECK_APPEND_1ARG
4101 #undef CHECK_APPEND_NOARG
4105 if (size_bp < size_buf)
4106 buf[size_buf - size_bp - 1] = '\0'; /* remove last , */
4111 static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info)
4113 u32 size_buf = 1024;
4114 char tmp_buf[192] = {'\0'};
4117 u32 size_bp = size_buf;
4119 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4121 buf = kzalloc(size_buf, GFP_KERNEL);
4127 #define CHECK_APPEND_1ARG(a, v1) \
4129 ret = snprintf(bp, size_bp, (a), (v1)); \
4130 if (ret < 0 || ret >= size_bp) \
4131 goto out_overflow; \
4136 if (bctl->flags & BTRFS_BALANCE_FORCE)
4137 CHECK_APPEND_1ARG("%s", "-f ");
4139 if (bctl->flags & BTRFS_BALANCE_DATA) {
4140 describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf));
4141 CHECK_APPEND_1ARG("-d%s ", tmp_buf);
4144 if (bctl->flags & BTRFS_BALANCE_METADATA) {
4145 describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf));
4146 CHECK_APPEND_1ARG("-m%s ", tmp_buf);
4149 if (bctl->flags & BTRFS_BALANCE_SYSTEM) {
4150 describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf));
4151 CHECK_APPEND_1ARG("-s%s ", tmp_buf);
4154 #undef CHECK_APPEND_1ARG
4158 if (size_bp < size_buf)
4159 buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */
4160 btrfs_info(fs_info, "balance: %s %s",
4161 (bctl->flags & BTRFS_BALANCE_RESUME) ?
4162 "resume" : "start", buf);
4168 * Should be called with balance mutexe held
4170 int btrfs_balance(struct btrfs_fs_info *fs_info,
4171 struct btrfs_balance_control *bctl,
4172 struct btrfs_ioctl_balance_args *bargs)
4174 u64 meta_target, data_target;
4180 bool reducing_redundancy;
4183 if (btrfs_fs_closing(fs_info) ||
4184 atomic_read(&fs_info->balance_pause_req) ||
4185 btrfs_should_cancel_balance(fs_info)) {
4190 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
4191 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
4195 * In case of mixed groups both data and meta should be picked,
4196 * and identical options should be given for both of them.
4198 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
4199 if (mixed && (bctl->flags & allowed)) {
4200 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
4201 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
4202 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
4204 "balance: mixed groups data and metadata options must be the same");
4211 * rw_devices will not change at the moment, device add/delete/replace
4214 num_devices = fs_info->fs_devices->rw_devices;
4217 * SINGLE profile on-disk has no profile bit, but in-memory we have a
4218 * special bit for it, to make it easier to distinguish. Thus we need
4219 * to set it manually, or balance would refuse the profile.
4221 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
4222 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++)
4223 if (num_devices >= btrfs_raid_array[i].devs_min)
4224 allowed |= btrfs_raid_array[i].bg_flag;
4226 if (!validate_convert_profile(fs_info, &bctl->data, allowed, "data") ||
4227 !validate_convert_profile(fs_info, &bctl->meta, allowed, "metadata") ||
4228 !validate_convert_profile(fs_info, &bctl->sys, allowed, "system")) {
4234 * Allow to reduce metadata or system integrity only if force set for
4235 * profiles with redundancy (copies, parity)
4238 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) {
4239 if (btrfs_raid_array[i].ncopies >= 2 ||
4240 btrfs_raid_array[i].tolerated_failures >= 1)
4241 allowed |= btrfs_raid_array[i].bg_flag;
4244 seq = read_seqbegin(&fs_info->profiles_lock);
4246 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4247 (fs_info->avail_system_alloc_bits & allowed) &&
4248 !(bctl->sys.target & allowed)) ||
4249 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4250 (fs_info->avail_metadata_alloc_bits & allowed) &&
4251 !(bctl->meta.target & allowed)))
4252 reducing_redundancy = true;
4254 reducing_redundancy = false;
4256 /* if we're not converting, the target field is uninitialized */
4257 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4258 bctl->meta.target : fs_info->avail_metadata_alloc_bits;
4259 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4260 bctl->data.target : fs_info->avail_data_alloc_bits;
4261 } while (read_seqretry(&fs_info->profiles_lock, seq));
4263 if (reducing_redundancy) {
4264 if (bctl->flags & BTRFS_BALANCE_FORCE) {
4266 "balance: force reducing metadata redundancy");
4269 "balance: reduces metadata redundancy, use --force if you want this");
4275 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <
4276 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {
4278 "balance: metadata profile %s has lower redundancy than data profile %s",
4279 btrfs_bg_type_to_raid_name(meta_target),
4280 btrfs_bg_type_to_raid_name(data_target));
4283 ret = insert_balance_item(fs_info, bctl);
4284 if (ret && ret != -EEXIST)
4287 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
4288 BUG_ON(ret == -EEXIST);
4289 BUG_ON(fs_info->balance_ctl);
4290 spin_lock(&fs_info->balance_lock);
4291 fs_info->balance_ctl = bctl;
4292 spin_unlock(&fs_info->balance_lock);
4294 BUG_ON(ret != -EEXIST);
4295 spin_lock(&fs_info->balance_lock);
4296 update_balance_args(bctl);
4297 spin_unlock(&fs_info->balance_lock);
4300 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4301 set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4302 describe_balance_start_or_resume(fs_info);
4303 mutex_unlock(&fs_info->balance_mutex);
4305 ret = __btrfs_balance(fs_info);
4307 mutex_lock(&fs_info->balance_mutex);
4308 if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req))
4309 btrfs_info(fs_info, "balance: paused");
4311 * Balance can be canceled by:
4313 * - Regular cancel request
4314 * Then ret == -ECANCELED and balance_cancel_req > 0
4316 * - Fatal signal to "btrfs" process
4317 * Either the signal caught by wait_reserve_ticket() and callers
4318 * got -EINTR, or caught by btrfs_should_cancel_balance() and
4320 * Either way, in this case balance_cancel_req = 0, and
4321 * ret == -EINTR or ret == -ECANCELED.
4323 * So here we only check the return value to catch canceled balance.
4325 else if (ret == -ECANCELED || ret == -EINTR)
4326 btrfs_info(fs_info, "balance: canceled");
4328 btrfs_info(fs_info, "balance: ended with status: %d", ret);
4330 clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4333 memset(bargs, 0, sizeof(*bargs));
4334 btrfs_update_ioctl_balance_args(fs_info, bargs);
4337 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
4338 balance_need_close(fs_info)) {
4339 reset_balance_state(fs_info);
4340 btrfs_exclop_finish(fs_info);
4343 wake_up(&fs_info->balance_wait_q);
4347 if (bctl->flags & BTRFS_BALANCE_RESUME)
4348 reset_balance_state(fs_info);
4351 btrfs_exclop_finish(fs_info);
4356 static int balance_kthread(void *data)
4358 struct btrfs_fs_info *fs_info = data;
4361 mutex_lock(&fs_info->balance_mutex);
4362 if (fs_info->balance_ctl)
4363 ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL);
4364 mutex_unlock(&fs_info->balance_mutex);
4369 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
4371 struct task_struct *tsk;
4373 mutex_lock(&fs_info->balance_mutex);
4374 if (!fs_info->balance_ctl) {
4375 mutex_unlock(&fs_info->balance_mutex);
4378 mutex_unlock(&fs_info->balance_mutex);
4380 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
4381 btrfs_info(fs_info, "balance: resume skipped");
4386 * A ro->rw remount sequence should continue with the paused balance
4387 * regardless of who pauses it, system or the user as of now, so set
4390 spin_lock(&fs_info->balance_lock);
4391 fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME;
4392 spin_unlock(&fs_info->balance_lock);
4394 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
4395 return PTR_ERR_OR_ZERO(tsk);
4398 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
4400 struct btrfs_balance_control *bctl;
4401 struct btrfs_balance_item *item;
4402 struct btrfs_disk_balance_args disk_bargs;
4403 struct btrfs_path *path;
4404 struct extent_buffer *leaf;
4405 struct btrfs_key key;
4408 path = btrfs_alloc_path();
4412 key.objectid = BTRFS_BALANCE_OBJECTID;
4413 key.type = BTRFS_TEMPORARY_ITEM_KEY;
4416 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4419 if (ret > 0) { /* ret = -ENOENT; */
4424 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4430 leaf = path->nodes[0];
4431 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
4433 bctl->flags = btrfs_balance_flags(leaf, item);
4434 bctl->flags |= BTRFS_BALANCE_RESUME;
4436 btrfs_balance_data(leaf, item, &disk_bargs);
4437 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4438 btrfs_balance_meta(leaf, item, &disk_bargs);
4439 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4440 btrfs_balance_sys(leaf, item, &disk_bargs);
4441 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4444 * This should never happen, as the paused balance state is recovered
4445 * during mount without any chance of other exclusive ops to collide.
4447 * This gives the exclusive op status to balance and keeps in paused
4448 * state until user intervention (cancel or umount). If the ownership
4449 * cannot be assigned, show a message but do not fail. The balance
4450 * is in a paused state and must have fs_info::balance_ctl properly
4453 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE))
4455 "balance: cannot set exclusive op status, resume manually");
4457 btrfs_release_path(path);
4459 mutex_lock(&fs_info->balance_mutex);
4460 BUG_ON(fs_info->balance_ctl);
4461 spin_lock(&fs_info->balance_lock);
4462 fs_info->balance_ctl = bctl;
4463 spin_unlock(&fs_info->balance_lock);
4464 mutex_unlock(&fs_info->balance_mutex);
4466 btrfs_free_path(path);
4470 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4474 mutex_lock(&fs_info->balance_mutex);
4475 if (!fs_info->balance_ctl) {
4476 mutex_unlock(&fs_info->balance_mutex);
4480 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4481 atomic_inc(&fs_info->balance_pause_req);
4482 mutex_unlock(&fs_info->balance_mutex);
4484 wait_event(fs_info->balance_wait_q,
4485 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4487 mutex_lock(&fs_info->balance_mutex);
4488 /* we are good with balance_ctl ripped off from under us */
4489 BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4490 atomic_dec(&fs_info->balance_pause_req);
4495 mutex_unlock(&fs_info->balance_mutex);
4499 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4501 mutex_lock(&fs_info->balance_mutex);
4502 if (!fs_info->balance_ctl) {
4503 mutex_unlock(&fs_info->balance_mutex);
4508 * A paused balance with the item stored on disk can be resumed at
4509 * mount time if the mount is read-write. Otherwise it's still paused
4510 * and we must not allow cancelling as it deletes the item.
4512 if (sb_rdonly(fs_info->sb)) {
4513 mutex_unlock(&fs_info->balance_mutex);
4517 atomic_inc(&fs_info->balance_cancel_req);
4519 * if we are running just wait and return, balance item is
4520 * deleted in btrfs_balance in this case
4522 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4523 mutex_unlock(&fs_info->balance_mutex);
4524 wait_event(fs_info->balance_wait_q,
4525 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4526 mutex_lock(&fs_info->balance_mutex);
4528 mutex_unlock(&fs_info->balance_mutex);
4530 * Lock released to allow other waiters to continue, we'll
4531 * reexamine the status again.
4533 mutex_lock(&fs_info->balance_mutex);
4535 if (fs_info->balance_ctl) {
4536 reset_balance_state(fs_info);
4537 btrfs_exclop_finish(fs_info);
4538 btrfs_info(fs_info, "balance: canceled");
4542 BUG_ON(fs_info->balance_ctl ||
4543 test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4544 atomic_dec(&fs_info->balance_cancel_req);
4545 mutex_unlock(&fs_info->balance_mutex);
4549 int btrfs_uuid_scan_kthread(void *data)
4551 struct btrfs_fs_info *fs_info = data;
4552 struct btrfs_root *root = fs_info->tree_root;
4553 struct btrfs_key key;
4554 struct btrfs_path *path = NULL;
4556 struct extent_buffer *eb;
4558 struct btrfs_root_item root_item;
4560 struct btrfs_trans_handle *trans = NULL;
4561 bool closing = false;
4563 path = btrfs_alloc_path();
4570 key.type = BTRFS_ROOT_ITEM_KEY;
4574 if (btrfs_fs_closing(fs_info)) {
4578 ret = btrfs_search_forward(root, &key, path,
4579 BTRFS_OLDEST_GENERATION);
4586 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4587 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4588 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4589 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4592 eb = path->nodes[0];
4593 slot = path->slots[0];
4594 item_size = btrfs_item_size_nr(eb, slot);
4595 if (item_size < sizeof(root_item))
4598 read_extent_buffer(eb, &root_item,
4599 btrfs_item_ptr_offset(eb, slot),
4600 (int)sizeof(root_item));
4601 if (btrfs_root_refs(&root_item) == 0)
4604 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4605 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4609 btrfs_release_path(path);
4611 * 1 - subvol uuid item
4612 * 1 - received_subvol uuid item
4614 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4615 if (IS_ERR(trans)) {
4616 ret = PTR_ERR(trans);
4624 btrfs_release_path(path);
4625 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4626 ret = btrfs_uuid_tree_add(trans, root_item.uuid,
4627 BTRFS_UUID_KEY_SUBVOL,
4630 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4636 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4637 ret = btrfs_uuid_tree_add(trans,
4638 root_item.received_uuid,
4639 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4642 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4649 btrfs_release_path(path);
4651 ret = btrfs_end_transaction(trans);
4657 if (key.offset < (u64)-1) {
4659 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4661 key.type = BTRFS_ROOT_ITEM_KEY;
4662 } else if (key.objectid < (u64)-1) {
4664 key.type = BTRFS_ROOT_ITEM_KEY;
4673 btrfs_free_path(path);
4674 if (trans && !IS_ERR(trans))
4675 btrfs_end_transaction(trans);
4677 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4679 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4680 up(&fs_info->uuid_tree_rescan_sem);
4684 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4686 struct btrfs_trans_handle *trans;
4687 struct btrfs_root *tree_root = fs_info->tree_root;
4688 struct btrfs_root *uuid_root;
4689 struct task_struct *task;
4696 trans = btrfs_start_transaction(tree_root, 2);
4698 return PTR_ERR(trans);
4700 uuid_root = btrfs_create_tree(trans, BTRFS_UUID_TREE_OBJECTID);
4701 if (IS_ERR(uuid_root)) {
4702 ret = PTR_ERR(uuid_root);
4703 btrfs_abort_transaction(trans, ret);
4704 btrfs_end_transaction(trans);
4708 fs_info->uuid_root = uuid_root;
4710 ret = btrfs_commit_transaction(trans);
4714 down(&fs_info->uuid_tree_rescan_sem);
4715 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4717 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4718 btrfs_warn(fs_info, "failed to start uuid_scan task");
4719 up(&fs_info->uuid_tree_rescan_sem);
4720 return PTR_ERR(task);
4727 * shrinking a device means finding all of the device extents past
4728 * the new size, and then following the back refs to the chunks.
4729 * The chunk relocation code actually frees the device extent
4731 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4733 struct btrfs_fs_info *fs_info = device->fs_info;
4734 struct btrfs_root *root = fs_info->dev_root;
4735 struct btrfs_trans_handle *trans;
4736 struct btrfs_dev_extent *dev_extent = NULL;
4737 struct btrfs_path *path;
4743 bool retried = false;
4744 struct extent_buffer *l;
4745 struct btrfs_key key;
4746 struct btrfs_super_block *super_copy = fs_info->super_copy;
4747 u64 old_total = btrfs_super_total_bytes(super_copy);
4748 u64 old_size = btrfs_device_get_total_bytes(device);
4752 new_size = round_down(new_size, fs_info->sectorsize);
4754 diff = round_down(old_size - new_size, fs_info->sectorsize);
4756 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4759 path = btrfs_alloc_path();
4763 path->reada = READA_BACK;
4765 trans = btrfs_start_transaction(root, 0);
4766 if (IS_ERR(trans)) {
4767 btrfs_free_path(path);
4768 return PTR_ERR(trans);
4771 mutex_lock(&fs_info->chunk_mutex);
4773 btrfs_device_set_total_bytes(device, new_size);
4774 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4775 device->fs_devices->total_rw_bytes -= diff;
4776 atomic64_sub(diff, &fs_info->free_chunk_space);
4780 * Once the device's size has been set to the new size, ensure all
4781 * in-memory chunks are synced to disk so that the loop below sees them
4782 * and relocates them accordingly.
4784 if (contains_pending_extent(device, &start, diff)) {
4785 mutex_unlock(&fs_info->chunk_mutex);
4786 ret = btrfs_commit_transaction(trans);
4790 mutex_unlock(&fs_info->chunk_mutex);
4791 btrfs_end_transaction(trans);
4795 key.objectid = device->devid;
4796 key.offset = (u64)-1;
4797 key.type = BTRFS_DEV_EXTENT_KEY;
4800 mutex_lock(&fs_info->reclaim_bgs_lock);
4801 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4803 mutex_unlock(&fs_info->reclaim_bgs_lock);
4807 ret = btrfs_previous_item(root, path, 0, key.type);
4809 mutex_unlock(&fs_info->reclaim_bgs_lock);
4813 btrfs_release_path(path);
4818 slot = path->slots[0];
4819 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4821 if (key.objectid != device->devid) {
4822 mutex_unlock(&fs_info->reclaim_bgs_lock);
4823 btrfs_release_path(path);
4827 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4828 length = btrfs_dev_extent_length(l, dev_extent);
4830 if (key.offset + length <= new_size) {
4831 mutex_unlock(&fs_info->reclaim_bgs_lock);
4832 btrfs_release_path(path);
4836 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4837 btrfs_release_path(path);
4840 * We may be relocating the only data chunk we have,
4841 * which could potentially end up with losing data's
4842 * raid profile, so lets allocate an empty one in
4845 ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset);
4847 mutex_unlock(&fs_info->reclaim_bgs_lock);
4851 ret = btrfs_relocate_chunk(fs_info, chunk_offset);
4852 mutex_unlock(&fs_info->reclaim_bgs_lock);
4853 if (ret == -ENOSPC) {
4856 if (ret == -ETXTBSY) {
4858 "could not shrink block group %llu due to active swapfile",
4863 } while (key.offset-- > 0);
4865 if (failed && !retried) {
4869 } else if (failed && retried) {
4874 /* Shrinking succeeded, else we would be at "done". */
4875 trans = btrfs_start_transaction(root, 0);
4876 if (IS_ERR(trans)) {
4877 ret = PTR_ERR(trans);
4881 mutex_lock(&fs_info->chunk_mutex);
4882 /* Clear all state bits beyond the shrunk device size */
4883 clear_extent_bits(&device->alloc_state, new_size, (u64)-1,
4886 btrfs_device_set_disk_total_bytes(device, new_size);
4887 if (list_empty(&device->post_commit_list))
4888 list_add_tail(&device->post_commit_list,
4889 &trans->transaction->dev_update_list);
4891 WARN_ON(diff > old_total);
4892 btrfs_set_super_total_bytes(super_copy,
4893 round_down(old_total - diff, fs_info->sectorsize));
4894 mutex_unlock(&fs_info->chunk_mutex);
4896 /* Now btrfs_update_device() will change the on-disk size. */
4897 ret = btrfs_update_device(trans, device);
4899 btrfs_abort_transaction(trans, ret);
4900 btrfs_end_transaction(trans);
4902 ret = btrfs_commit_transaction(trans);
4905 btrfs_free_path(path);
4907 mutex_lock(&fs_info->chunk_mutex);
4908 btrfs_device_set_total_bytes(device, old_size);
4909 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
4910 device->fs_devices->total_rw_bytes += diff;
4911 atomic64_add(diff, &fs_info->free_chunk_space);
4912 mutex_unlock(&fs_info->chunk_mutex);
4917 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
4918 struct btrfs_key *key,
4919 struct btrfs_chunk *chunk, int item_size)
4921 struct btrfs_super_block *super_copy = fs_info->super_copy;
4922 struct btrfs_disk_key disk_key;
4926 lockdep_assert_held(&fs_info->chunk_mutex);
4928 array_size = btrfs_super_sys_array_size(super_copy);
4929 if (array_size + item_size + sizeof(disk_key)
4930 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
4933 ptr = super_copy->sys_chunk_array + array_size;
4934 btrfs_cpu_key_to_disk(&disk_key, key);
4935 memcpy(ptr, &disk_key, sizeof(disk_key));
4936 ptr += sizeof(disk_key);
4937 memcpy(ptr, chunk, item_size);
4938 item_size += sizeof(disk_key);
4939 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4945 * sort the devices in descending order by max_avail, total_avail
4947 static int btrfs_cmp_device_info(const void *a, const void *b)
4949 const struct btrfs_device_info *di_a = a;
4950 const struct btrfs_device_info *di_b = b;
4952 if (di_a->max_avail > di_b->max_avail)
4954 if (di_a->max_avail < di_b->max_avail)
4956 if (di_a->total_avail > di_b->total_avail)
4958 if (di_a->total_avail < di_b->total_avail)
4963 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4965 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4968 btrfs_set_fs_incompat(info, RAID56);
4971 static void check_raid1c34_incompat_flag(struct btrfs_fs_info *info, u64 type)
4973 if (!(type & (BTRFS_BLOCK_GROUP_RAID1C3 | BTRFS_BLOCK_GROUP_RAID1C4)))
4976 btrfs_set_fs_incompat(info, RAID1C34);
4980 * Structure used internally for __btrfs_alloc_chunk() function.
4981 * Wraps needed parameters.
4983 struct alloc_chunk_ctl {
4986 /* Total number of stripes to allocate */
4988 /* sub_stripes info for map */
4990 /* Stripes per device */
4992 /* Maximum number of devices to use */
4994 /* Minimum number of devices to use */
4996 /* ndevs has to be a multiple of this */
4998 /* Number of copies */
5000 /* Number of stripes worth of bytes to store parity information */
5002 u64 max_stripe_size;
5010 static void init_alloc_chunk_ctl_policy_regular(
5011 struct btrfs_fs_devices *fs_devices,
5012 struct alloc_chunk_ctl *ctl)
5014 u64 type = ctl->type;
5016 if (type & BTRFS_BLOCK_GROUP_DATA) {
5017 ctl->max_stripe_size = SZ_1G;
5018 ctl->max_chunk_size = BTRFS_MAX_DATA_CHUNK_SIZE;
5019 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
5020 /* For larger filesystems, use larger metadata chunks */
5021 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
5022 ctl->max_stripe_size = SZ_1G;
5024 ctl->max_stripe_size = SZ_256M;
5025 ctl->max_chunk_size = ctl->max_stripe_size;
5026 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
5027 ctl->max_stripe_size = SZ_32M;
5028 ctl->max_chunk_size = 2 * ctl->max_stripe_size;
5029 ctl->devs_max = min_t(int, ctl->devs_max,
5030 BTRFS_MAX_DEVS_SYS_CHUNK);
5035 /* We don't want a chunk larger than 10% of writable space */
5036 ctl->max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
5037 ctl->max_chunk_size);
5038 ctl->dev_extent_min = BTRFS_STRIPE_LEN * ctl->dev_stripes;
5041 static void init_alloc_chunk_ctl_policy_zoned(
5042 struct btrfs_fs_devices *fs_devices,
5043 struct alloc_chunk_ctl *ctl)
5045 u64 zone_size = fs_devices->fs_info->zone_size;
5047 int min_num_stripes = ctl->devs_min * ctl->dev_stripes;
5048 int min_data_stripes = (min_num_stripes - ctl->nparity) / ctl->ncopies;
5049 u64 min_chunk_size = min_data_stripes * zone_size;
5050 u64 type = ctl->type;
5052 ctl->max_stripe_size = zone_size;
5053 if (type & BTRFS_BLOCK_GROUP_DATA) {
5054 ctl->max_chunk_size = round_down(BTRFS_MAX_DATA_CHUNK_SIZE,
5056 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
5057 ctl->max_chunk_size = ctl->max_stripe_size;
5058 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
5059 ctl->max_chunk_size = 2 * ctl->max_stripe_size;
5060 ctl->devs_max = min_t(int, ctl->devs_max,
5061 BTRFS_MAX_DEVS_SYS_CHUNK);
5066 /* We don't want a chunk larger than 10% of writable space */
5067 limit = max(round_down(div_factor(fs_devices->total_rw_bytes, 1),
5070 ctl->max_chunk_size = min(limit, ctl->max_chunk_size);
5071 ctl->dev_extent_min = zone_size * ctl->dev_stripes;
5074 static void init_alloc_chunk_ctl(struct btrfs_fs_devices *fs_devices,
5075 struct alloc_chunk_ctl *ctl)
5077 int index = btrfs_bg_flags_to_raid_index(ctl->type);
5079 ctl->sub_stripes = btrfs_raid_array[index].sub_stripes;
5080 ctl->dev_stripes = btrfs_raid_array[index].dev_stripes;
5081 ctl->devs_max = btrfs_raid_array[index].devs_max;
5083 ctl->devs_max = BTRFS_MAX_DEVS(fs_devices->fs_info);
5084 ctl->devs_min = btrfs_raid_array[index].devs_min;
5085 ctl->devs_increment = btrfs_raid_array[index].devs_increment;
5086 ctl->ncopies = btrfs_raid_array[index].ncopies;
5087 ctl->nparity = btrfs_raid_array[index].nparity;
5090 switch (fs_devices->chunk_alloc_policy) {
5091 case BTRFS_CHUNK_ALLOC_REGULAR:
5092 init_alloc_chunk_ctl_policy_regular(fs_devices, ctl);
5094 case BTRFS_CHUNK_ALLOC_ZONED:
5095 init_alloc_chunk_ctl_policy_zoned(fs_devices, ctl);
5102 static int gather_device_info(struct btrfs_fs_devices *fs_devices,
5103 struct alloc_chunk_ctl *ctl,
5104 struct btrfs_device_info *devices_info)
5106 struct btrfs_fs_info *info = fs_devices->fs_info;
5107 struct btrfs_device *device;
5109 u64 dev_extent_want = ctl->max_stripe_size * ctl->dev_stripes;
5116 * in the first pass through the devices list, we gather information
5117 * about the available holes on each device.
5119 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
5120 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
5122 "BTRFS: read-only device in alloc_list\n");
5126 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
5127 &device->dev_state) ||
5128 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
5131 if (device->total_bytes > device->bytes_used)
5132 total_avail = device->total_bytes - device->bytes_used;
5136 /* If there is no space on this device, skip it. */
5137 if (total_avail < ctl->dev_extent_min)
5140 ret = find_free_dev_extent(device, dev_extent_want, &dev_offset,
5142 if (ret && ret != -ENOSPC)
5146 max_avail = dev_extent_want;
5148 if (max_avail < ctl->dev_extent_min) {
5149 if (btrfs_test_opt(info, ENOSPC_DEBUG))
5151 "%s: devid %llu has no free space, have=%llu want=%llu",
5152 __func__, device->devid, max_avail,
5153 ctl->dev_extent_min);
5157 if (ndevs == fs_devices->rw_devices) {
5158 WARN(1, "%s: found more than %llu devices\n",
5159 __func__, fs_devices->rw_devices);
5162 devices_info[ndevs].dev_offset = dev_offset;
5163 devices_info[ndevs].max_avail = max_avail;
5164 devices_info[ndevs].total_avail = total_avail;
5165 devices_info[ndevs].dev = device;
5171 * now sort the devices by hole size / available space
5173 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
5174 btrfs_cmp_device_info, NULL);
5179 static int decide_stripe_size_regular(struct alloc_chunk_ctl *ctl,
5180 struct btrfs_device_info *devices_info)
5182 /* Number of stripes that count for block group size */
5186 * The primary goal is to maximize the number of stripes, so use as
5187 * many devices as possible, even if the stripes are not maximum sized.
5189 * The DUP profile stores more than one stripe per device, the
5190 * max_avail is the total size so we have to adjust.
5192 ctl->stripe_size = div_u64(devices_info[ctl->ndevs - 1].max_avail,
5194 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5196 /* This will have to be fixed for RAID1 and RAID10 over more drives */
5197 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5200 * Use the number of data stripes to figure out how big this chunk is
5201 * really going to be in terms of logical address space, and compare
5202 * that answer with the max chunk size. If it's higher, we try to
5203 * reduce stripe_size.
5205 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) {
5207 * Reduce stripe_size, round it up to a 16MB boundary again and
5208 * then use it, unless it ends up being even bigger than the
5209 * previous value we had already.
5211 ctl->stripe_size = min(round_up(div_u64(ctl->max_chunk_size,
5212 data_stripes), SZ_16M),
5216 /* Align to BTRFS_STRIPE_LEN */
5217 ctl->stripe_size = round_down(ctl->stripe_size, BTRFS_STRIPE_LEN);
5218 ctl->chunk_size = ctl->stripe_size * data_stripes;
5223 static int decide_stripe_size_zoned(struct alloc_chunk_ctl *ctl,
5224 struct btrfs_device_info *devices_info)
5226 u64 zone_size = devices_info[0].dev->zone_info->zone_size;
5227 /* Number of stripes that count for block group size */
5231 * It should hold because:
5232 * dev_extent_min == dev_extent_want == zone_size * dev_stripes
5234 ASSERT(devices_info[ctl->ndevs - 1].max_avail == ctl->dev_extent_min);
5236 ctl->stripe_size = zone_size;
5237 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5238 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5240 /* stripe_size is fixed in zoned filesysmte. Reduce ndevs instead. */
5241 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) {
5242 ctl->ndevs = div_u64(div_u64(ctl->max_chunk_size * ctl->ncopies,
5243 ctl->stripe_size) + ctl->nparity,
5245 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5246 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5247 ASSERT(ctl->stripe_size * data_stripes <= ctl->max_chunk_size);
5250 ctl->chunk_size = ctl->stripe_size * data_stripes;
5255 static int decide_stripe_size(struct btrfs_fs_devices *fs_devices,
5256 struct alloc_chunk_ctl *ctl,
5257 struct btrfs_device_info *devices_info)
5259 struct btrfs_fs_info *info = fs_devices->fs_info;
5262 * Round down to number of usable stripes, devs_increment can be any
5263 * number so we can't use round_down() that requires power of 2, while
5264 * rounddown is safe.
5266 ctl->ndevs = rounddown(ctl->ndevs, ctl->devs_increment);
5268 if (ctl->ndevs < ctl->devs_min) {
5269 if (btrfs_test_opt(info, ENOSPC_DEBUG)) {
5271 "%s: not enough devices with free space: have=%d minimum required=%d",
5272 __func__, ctl->ndevs, ctl->devs_min);
5277 ctl->ndevs = min(ctl->ndevs, ctl->devs_max);
5279 switch (fs_devices->chunk_alloc_policy) {
5280 case BTRFS_CHUNK_ALLOC_REGULAR:
5281 return decide_stripe_size_regular(ctl, devices_info);
5282 case BTRFS_CHUNK_ALLOC_ZONED:
5283 return decide_stripe_size_zoned(ctl, devices_info);
5289 static struct btrfs_block_group *create_chunk(struct btrfs_trans_handle *trans,
5290 struct alloc_chunk_ctl *ctl,
5291 struct btrfs_device_info *devices_info)
5293 struct btrfs_fs_info *info = trans->fs_info;
5294 struct map_lookup *map = NULL;
5295 struct extent_map_tree *em_tree;
5296 struct btrfs_block_group *block_group;
5297 struct extent_map *em;
5298 u64 start = ctl->start;
5299 u64 type = ctl->type;
5304 map = kmalloc(map_lookup_size(ctl->num_stripes), GFP_NOFS);
5306 return ERR_PTR(-ENOMEM);
5307 map->num_stripes = ctl->num_stripes;
5309 for (i = 0; i < ctl->ndevs; ++i) {
5310 for (j = 0; j < ctl->dev_stripes; ++j) {
5311 int s = i * ctl->dev_stripes + j;
5312 map->stripes[s].dev = devices_info[i].dev;
5313 map->stripes[s].physical = devices_info[i].dev_offset +
5314 j * ctl->stripe_size;
5317 map->stripe_len = BTRFS_STRIPE_LEN;
5318 map->io_align = BTRFS_STRIPE_LEN;
5319 map->io_width = BTRFS_STRIPE_LEN;
5321 map->sub_stripes = ctl->sub_stripes;
5323 trace_btrfs_chunk_alloc(info, map, start, ctl->chunk_size);
5325 em = alloc_extent_map();
5328 return ERR_PTR(-ENOMEM);
5330 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
5331 em->map_lookup = map;
5333 em->len = ctl->chunk_size;
5334 em->block_start = 0;
5335 em->block_len = em->len;
5336 em->orig_block_len = ctl->stripe_size;
5338 em_tree = &info->mapping_tree;
5339 write_lock(&em_tree->lock);
5340 ret = add_extent_mapping(em_tree, em, 0);
5342 write_unlock(&em_tree->lock);
5343 free_extent_map(em);
5344 return ERR_PTR(ret);
5346 write_unlock(&em_tree->lock);
5348 block_group = btrfs_make_block_group(trans, 0, type, start, ctl->chunk_size);
5349 if (IS_ERR(block_group))
5350 goto error_del_extent;
5352 for (i = 0; i < map->num_stripes; i++) {
5353 struct btrfs_device *dev = map->stripes[i].dev;
5355 btrfs_device_set_bytes_used(dev,
5356 dev->bytes_used + ctl->stripe_size);
5357 if (list_empty(&dev->post_commit_list))
5358 list_add_tail(&dev->post_commit_list,
5359 &trans->transaction->dev_update_list);
5362 atomic64_sub(ctl->stripe_size * map->num_stripes,
5363 &info->free_chunk_space);
5365 free_extent_map(em);
5366 check_raid56_incompat_flag(info, type);
5367 check_raid1c34_incompat_flag(info, type);
5372 write_lock(&em_tree->lock);
5373 remove_extent_mapping(em_tree, em);
5374 write_unlock(&em_tree->lock);
5376 /* One for our allocation */
5377 free_extent_map(em);
5378 /* One for the tree reference */
5379 free_extent_map(em);
5384 struct btrfs_block_group *btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
5387 struct btrfs_fs_info *info = trans->fs_info;
5388 struct btrfs_fs_devices *fs_devices = info->fs_devices;
5389 struct btrfs_device_info *devices_info = NULL;
5390 struct alloc_chunk_ctl ctl;
5391 struct btrfs_block_group *block_group;
5394 lockdep_assert_held(&info->chunk_mutex);
5396 if (!alloc_profile_is_valid(type, 0)) {
5398 return ERR_PTR(-EINVAL);
5401 if (list_empty(&fs_devices->alloc_list)) {
5402 if (btrfs_test_opt(info, ENOSPC_DEBUG))
5403 btrfs_debug(info, "%s: no writable device", __func__);
5404 return ERR_PTR(-ENOSPC);
5407 if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
5408 btrfs_err(info, "invalid chunk type 0x%llx requested", type);
5410 return ERR_PTR(-EINVAL);
5413 ctl.start = find_next_chunk(info);
5415 init_alloc_chunk_ctl(fs_devices, &ctl);
5417 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
5420 return ERR_PTR(-ENOMEM);
5422 ret = gather_device_info(fs_devices, &ctl, devices_info);
5424 block_group = ERR_PTR(ret);
5428 ret = decide_stripe_size(fs_devices, &ctl, devices_info);
5430 block_group = ERR_PTR(ret);
5434 block_group = create_chunk(trans, &ctl, devices_info);
5437 kfree(devices_info);
5442 * This function, btrfs_chunk_alloc_add_chunk_item(), typically belongs to the
5443 * phase 1 of chunk allocation. It belongs to phase 2 only when allocating system
5446 * See the comment at btrfs_chunk_alloc() for details about the chunk allocation
5449 int btrfs_chunk_alloc_add_chunk_item(struct btrfs_trans_handle *trans,
5450 struct btrfs_block_group *bg)
5452 struct btrfs_fs_info *fs_info = trans->fs_info;
5453 struct btrfs_root *extent_root = fs_info->extent_root;
5454 struct btrfs_root *chunk_root = fs_info->chunk_root;
5455 struct btrfs_key key;
5456 struct btrfs_chunk *chunk;
5457 struct btrfs_stripe *stripe;
5458 struct extent_map *em;
5459 struct map_lookup *map;
5465 * We take the chunk_mutex for 2 reasons:
5467 * 1) Updates and insertions in the chunk btree must be done while holding
5468 * the chunk_mutex, as well as updating the system chunk array in the
5469 * superblock. See the comment on top of btrfs_chunk_alloc() for the
5472 * 2) To prevent races with the final phase of a device replace operation
5473 * that replaces the device object associated with the map's stripes,
5474 * because the device object's id can change at any time during that
5475 * final phase of the device replace operation
5476 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the
5477 * replaced device and then see it with an ID of BTRFS_DEV_REPLACE_DEVID,
5478 * which would cause a failure when updating the device item, which does
5479 * not exists, or persisting a stripe of the chunk item with such ID.
5480 * Here we can't use the device_list_mutex because our caller already
5481 * has locked the chunk_mutex, and the final phase of device replace
5482 * acquires both mutexes - first the device_list_mutex and then the
5483 * chunk_mutex. Using any of those two mutexes protects us from a
5484 * concurrent device replace.
5486 lockdep_assert_held(&fs_info->chunk_mutex);
5488 em = btrfs_get_chunk_map(fs_info, bg->start, bg->length);
5491 btrfs_abort_transaction(trans, ret);
5495 map = em->map_lookup;
5496 item_size = btrfs_chunk_item_size(map->num_stripes);
5498 chunk = kzalloc(item_size, GFP_NOFS);
5501 btrfs_abort_transaction(trans, ret);
5505 for (i = 0; i < map->num_stripes; i++) {
5506 struct btrfs_device *device = map->stripes[i].dev;
5508 ret = btrfs_update_device(trans, device);
5513 stripe = &chunk->stripe;
5514 for (i = 0; i < map->num_stripes; i++) {
5515 struct btrfs_device *device = map->stripes[i].dev;
5516 const u64 dev_offset = map->stripes[i].physical;
5518 btrfs_set_stack_stripe_devid(stripe, device->devid);
5519 btrfs_set_stack_stripe_offset(stripe, dev_offset);
5520 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
5524 btrfs_set_stack_chunk_length(chunk, bg->length);
5525 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
5526 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
5527 btrfs_set_stack_chunk_type(chunk, map->type);
5528 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
5529 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
5530 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
5531 btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
5532 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
5534 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
5535 key.type = BTRFS_CHUNK_ITEM_KEY;
5536 key.offset = bg->start;
5538 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
5542 bg->chunk_item_inserted = 1;
5544 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
5545 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
5552 free_extent_map(em);
5556 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans)
5558 struct btrfs_fs_info *fs_info = trans->fs_info;
5560 struct btrfs_block_group *meta_bg;
5561 struct btrfs_block_group *sys_bg;
5564 * When adding a new device for sprouting, the seed device is read-only
5565 * so we must first allocate a metadata and a system chunk. But before
5566 * adding the block group items to the extent, device and chunk btrees,
5569 * 1) Create both chunks without doing any changes to the btrees, as
5570 * otherwise we would get -ENOSPC since the block groups from the
5571 * seed device are read-only;
5573 * 2) Add the device item for the new sprout device - finishing the setup
5574 * of a new block group requires updating the device item in the chunk
5575 * btree, so it must exist when we attempt to do it. The previous step
5576 * ensures this does not fail with -ENOSPC.
5578 * After that we can add the block group items to their btrees:
5579 * update existing device item in the chunk btree, add a new block group
5580 * item to the extent btree, add a new chunk item to the chunk btree and
5581 * finally add the new device extent items to the devices btree.
5584 alloc_profile = btrfs_metadata_alloc_profile(fs_info);
5585 meta_bg = btrfs_alloc_chunk(trans, alloc_profile);
5586 if (IS_ERR(meta_bg))
5587 return PTR_ERR(meta_bg);
5589 alloc_profile = btrfs_system_alloc_profile(fs_info);
5590 sys_bg = btrfs_alloc_chunk(trans, alloc_profile);
5592 return PTR_ERR(sys_bg);
5597 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
5599 const int index = btrfs_bg_flags_to_raid_index(map->type);
5601 return btrfs_raid_array[index].tolerated_failures;
5604 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset)
5606 struct extent_map *em;
5607 struct map_lookup *map;
5612 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
5616 map = em->map_lookup;
5617 for (i = 0; i < map->num_stripes; i++) {
5618 if (test_bit(BTRFS_DEV_STATE_MISSING,
5619 &map->stripes[i].dev->dev_state)) {
5623 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE,
5624 &map->stripes[i].dev->dev_state)) {
5631 * If the number of missing devices is larger than max errors,
5632 * we can not write the data into that chunk successfully, so
5635 if (miss_ndevs > btrfs_chunk_max_errors(map))
5638 free_extent_map(em);
5642 void btrfs_mapping_tree_free(struct extent_map_tree *tree)
5644 struct extent_map *em;
5647 write_lock(&tree->lock);
5648 em = lookup_extent_mapping(tree, 0, (u64)-1);
5650 remove_extent_mapping(tree, em);
5651 write_unlock(&tree->lock);
5655 free_extent_map(em);
5656 /* once for the tree */
5657 free_extent_map(em);
5661 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5663 struct extent_map *em;
5664 struct map_lookup *map;
5667 em = btrfs_get_chunk_map(fs_info, logical, len);
5670 * We could return errors for these cases, but that could get
5671 * ugly and we'd probably do the same thing which is just not do
5672 * anything else and exit, so return 1 so the callers don't try
5673 * to use other copies.
5677 map = em->map_lookup;
5678 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1_MASK))
5679 ret = map->num_stripes;
5680 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5681 ret = map->sub_stripes;
5682 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5684 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5686 * There could be two corrupted data stripes, we need
5687 * to loop retry in order to rebuild the correct data.
5689 * Fail a stripe at a time on every retry except the
5690 * stripe under reconstruction.
5692 ret = map->num_stripes;
5695 free_extent_map(em);
5697 down_read(&fs_info->dev_replace.rwsem);
5698 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) &&
5699 fs_info->dev_replace.tgtdev)
5701 up_read(&fs_info->dev_replace.rwsem);
5706 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
5709 struct extent_map *em;
5710 struct map_lookup *map;
5711 unsigned long len = fs_info->sectorsize;
5713 em = btrfs_get_chunk_map(fs_info, logical, len);
5715 if (!WARN_ON(IS_ERR(em))) {
5716 map = em->map_lookup;
5717 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5718 len = map->stripe_len * nr_data_stripes(map);
5719 free_extent_map(em);
5724 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5726 struct extent_map *em;
5727 struct map_lookup *map;
5730 em = btrfs_get_chunk_map(fs_info, logical, len);
5732 if(!WARN_ON(IS_ERR(em))) {
5733 map = em->map_lookup;
5734 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5736 free_extent_map(em);
5741 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5742 struct map_lookup *map, int first,
5743 int dev_replace_is_ongoing)
5747 int preferred_mirror;
5749 struct btrfs_device *srcdev;
5752 (BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10)));
5754 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5755 num_stripes = map->sub_stripes;
5757 num_stripes = map->num_stripes;
5759 switch (fs_info->fs_devices->read_policy) {
5761 /* Shouldn't happen, just warn and use pid instead of failing */
5762 btrfs_warn_rl(fs_info,
5763 "unknown read_policy type %u, reset to pid",
5764 fs_info->fs_devices->read_policy);
5765 fs_info->fs_devices->read_policy = BTRFS_READ_POLICY_PID;
5767 case BTRFS_READ_POLICY_PID:
5768 preferred_mirror = first + (current->pid % num_stripes);
5772 if (dev_replace_is_ongoing &&
5773 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5774 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5775 srcdev = fs_info->dev_replace.srcdev;
5780 * try to avoid the drive that is the source drive for a
5781 * dev-replace procedure, only choose it if no other non-missing
5782 * mirror is available
5784 for (tolerance = 0; tolerance < 2; tolerance++) {
5785 if (map->stripes[preferred_mirror].dev->bdev &&
5786 (tolerance || map->stripes[preferred_mirror].dev != srcdev))
5787 return preferred_mirror;
5788 for (i = first; i < first + num_stripes; i++) {
5789 if (map->stripes[i].dev->bdev &&
5790 (tolerance || map->stripes[i].dev != srcdev))
5795 /* we couldn't find one that doesn't fail. Just return something
5796 * and the io error handling code will clean up eventually
5798 return preferred_mirror;
5801 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5802 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5809 for (i = 0; i < num_stripes - 1; i++) {
5810 /* Swap if parity is on a smaller index */
5811 if (bbio->raid_map[i] > bbio->raid_map[i + 1]) {
5812 swap(bbio->stripes[i], bbio->stripes[i + 1]);
5813 swap(bbio->raid_map[i], bbio->raid_map[i + 1]);
5820 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5822 struct btrfs_bio *bbio = kzalloc(
5823 /* the size of the btrfs_bio */
5824 sizeof(struct btrfs_bio) +
5825 /* plus the variable array for the stripes */
5826 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5827 /* plus the variable array for the tgt dev */
5828 sizeof(int) * (real_stripes) +
5830 * plus the raid_map, which includes both the tgt dev
5833 sizeof(u64) * (total_stripes),
5834 GFP_NOFS|__GFP_NOFAIL);
5836 atomic_set(&bbio->error, 0);
5837 refcount_set(&bbio->refs, 1);
5839 bbio->tgtdev_map = (int *)(bbio->stripes + total_stripes);
5840 bbio->raid_map = (u64 *)(bbio->tgtdev_map + real_stripes);
5845 void btrfs_get_bbio(struct btrfs_bio *bbio)
5847 WARN_ON(!refcount_read(&bbio->refs));
5848 refcount_inc(&bbio->refs);
5851 void btrfs_put_bbio(struct btrfs_bio *bbio)
5855 if (refcount_dec_and_test(&bbio->refs))
5859 /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */
5861 * Please note that, discard won't be sent to target device of device
5864 static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info,
5865 u64 logical, u64 *length_ret,
5866 struct btrfs_bio **bbio_ret)
5868 struct extent_map *em;
5869 struct map_lookup *map;
5870 struct btrfs_bio *bbio;
5871 u64 length = *length_ret;
5875 u64 stripe_end_offset;
5882 u32 sub_stripes = 0;
5883 u64 stripes_per_dev = 0;
5884 u32 remaining_stripes = 0;
5885 u32 last_stripe = 0;
5889 /* discard always return a bbio */
5892 em = btrfs_get_chunk_map(fs_info, logical, length);
5896 map = em->map_lookup;
5897 /* we don't discard raid56 yet */
5898 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5903 offset = logical - em->start;
5904 length = min_t(u64, em->start + em->len - logical, length);
5905 *length_ret = length;
5907 stripe_len = map->stripe_len;
5909 * stripe_nr counts the total number of stripes we have to stride
5910 * to get to this block
5912 stripe_nr = div64_u64(offset, stripe_len);
5914 /* stripe_offset is the offset of this block in its stripe */
5915 stripe_offset = offset - stripe_nr * stripe_len;
5917 stripe_nr_end = round_up(offset + length, map->stripe_len);
5918 stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len);
5919 stripe_cnt = stripe_nr_end - stripe_nr;
5920 stripe_end_offset = stripe_nr_end * map->stripe_len -
5923 * after this, stripe_nr is the number of stripes on this
5924 * device we have to walk to find the data, and stripe_index is
5925 * the number of our device in the stripe array
5929 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5930 BTRFS_BLOCK_GROUP_RAID10)) {
5931 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5934 sub_stripes = map->sub_stripes;
5936 factor = map->num_stripes / sub_stripes;
5937 num_stripes = min_t(u64, map->num_stripes,
5938 sub_stripes * stripe_cnt);
5939 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5940 stripe_index *= sub_stripes;
5941 stripes_per_dev = div_u64_rem(stripe_cnt, factor,
5942 &remaining_stripes);
5943 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5944 last_stripe *= sub_stripes;
5945 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1_MASK |
5946 BTRFS_BLOCK_GROUP_DUP)) {
5947 num_stripes = map->num_stripes;
5949 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5953 bbio = alloc_btrfs_bio(num_stripes, 0);
5959 for (i = 0; i < num_stripes; i++) {
5960 bbio->stripes[i].physical =
5961 map->stripes[stripe_index].physical +
5962 stripe_offset + stripe_nr * map->stripe_len;
5963 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5965 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5966 BTRFS_BLOCK_GROUP_RAID10)) {
5967 bbio->stripes[i].length = stripes_per_dev *
5970 if (i / sub_stripes < remaining_stripes)
5971 bbio->stripes[i].length +=
5975 * Special for the first stripe and
5978 * |-------|...|-------|
5982 if (i < sub_stripes)
5983 bbio->stripes[i].length -=
5986 if (stripe_index >= last_stripe &&
5987 stripe_index <= (last_stripe +
5989 bbio->stripes[i].length -=
5992 if (i == sub_stripes - 1)
5995 bbio->stripes[i].length = length;
5999 if (stripe_index == map->num_stripes) {
6006 bbio->map_type = map->type;
6007 bbio->num_stripes = num_stripes;
6009 free_extent_map(em);
6014 * In dev-replace case, for repair case (that's the only case where the mirror
6015 * is selected explicitly when calling btrfs_map_block), blocks left of the
6016 * left cursor can also be read from the target drive.
6018 * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the
6020 * For READ, it also needs to be supported using the same mirror number.
6022 * If the requested block is not left of the left cursor, EIO is returned. This
6023 * can happen because btrfs_num_copies() returns one more in the dev-replace
6026 static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info,
6027 u64 logical, u64 length,
6028 u64 srcdev_devid, int *mirror_num,
6031 struct btrfs_bio *bbio = NULL;
6033 int index_srcdev = 0;
6035 u64 physical_of_found = 0;
6039 ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS,
6040 logical, &length, &bbio, 0, 0);
6042 ASSERT(bbio == NULL);
6046 num_stripes = bbio->num_stripes;
6047 if (*mirror_num > num_stripes) {
6049 * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror,
6050 * that means that the requested area is not left of the left
6053 btrfs_put_bbio(bbio);
6058 * process the rest of the function using the mirror_num of the source
6059 * drive. Therefore look it up first. At the end, patch the device
6060 * pointer to the one of the target drive.
6062 for (i = 0; i < num_stripes; i++) {
6063 if (bbio->stripes[i].dev->devid != srcdev_devid)
6067 * In case of DUP, in order to keep it simple, only add the
6068 * mirror with the lowest physical address
6071 physical_of_found <= bbio->stripes[i].physical)
6076 physical_of_found = bbio->stripes[i].physical;
6079 btrfs_put_bbio(bbio);
6085 *mirror_num = index_srcdev + 1;
6086 *physical = physical_of_found;
6090 static bool is_block_group_to_copy(struct btrfs_fs_info *fs_info, u64 logical)
6092 struct btrfs_block_group *cache;
6095 /* Non zoned filesystem does not use "to_copy" flag */
6096 if (!btrfs_is_zoned(fs_info))
6099 cache = btrfs_lookup_block_group(fs_info, logical);
6101 spin_lock(&cache->lock);
6102 ret = cache->to_copy;
6103 spin_unlock(&cache->lock);
6105 btrfs_put_block_group(cache);
6109 static void handle_ops_on_dev_replace(enum btrfs_map_op op,
6110 struct btrfs_bio **bbio_ret,
6111 struct btrfs_dev_replace *dev_replace,
6113 int *num_stripes_ret, int *max_errors_ret)
6115 struct btrfs_bio *bbio = *bbio_ret;
6116 u64 srcdev_devid = dev_replace->srcdev->devid;
6117 int tgtdev_indexes = 0;
6118 int num_stripes = *num_stripes_ret;
6119 int max_errors = *max_errors_ret;
6122 if (op == BTRFS_MAP_WRITE) {
6123 int index_where_to_add;
6126 * A block group which have "to_copy" set will eventually
6127 * copied by dev-replace process. We can avoid cloning IO here.
6129 if (is_block_group_to_copy(dev_replace->srcdev->fs_info, logical))
6133 * duplicate the write operations while the dev replace
6134 * procedure is running. Since the copying of the old disk to
6135 * the new disk takes place at run time while the filesystem is
6136 * mounted writable, the regular write operations to the old
6137 * disk have to be duplicated to go to the new disk as well.
6139 * Note that device->missing is handled by the caller, and that
6140 * the write to the old disk is already set up in the stripes
6143 index_where_to_add = num_stripes;
6144 for (i = 0; i < num_stripes; i++) {
6145 if (bbio->stripes[i].dev->devid == srcdev_devid) {
6146 /* write to new disk, too */
6147 struct btrfs_bio_stripe *new =
6148 bbio->stripes + index_where_to_add;
6149 struct btrfs_bio_stripe *old =
6152 new->physical = old->physical;
6153 new->length = old->length;
6154 new->dev = dev_replace->tgtdev;
6155 bbio->tgtdev_map[i] = index_where_to_add;
6156 index_where_to_add++;
6161 num_stripes = index_where_to_add;
6162 } else if (op == BTRFS_MAP_GET_READ_MIRRORS) {
6163 int index_srcdev = 0;
6165 u64 physical_of_found = 0;
6168 * During the dev-replace procedure, the target drive can also
6169 * be used to read data in case it is needed to repair a corrupt
6170 * block elsewhere. This is possible if the requested area is
6171 * left of the left cursor. In this area, the target drive is a
6172 * full copy of the source drive.
6174 for (i = 0; i < num_stripes; i++) {
6175 if (bbio->stripes[i].dev->devid == srcdev_devid) {
6177 * In case of DUP, in order to keep it simple,
6178 * only add the mirror with the lowest physical
6182 physical_of_found <=
6183 bbio->stripes[i].physical)
6187 physical_of_found = bbio->stripes[i].physical;
6191 struct btrfs_bio_stripe *tgtdev_stripe =
6192 bbio->stripes + num_stripes;
6194 tgtdev_stripe->physical = physical_of_found;
6195 tgtdev_stripe->length =
6196 bbio->stripes[index_srcdev].length;
6197 tgtdev_stripe->dev = dev_replace->tgtdev;
6198 bbio->tgtdev_map[index_srcdev] = num_stripes;
6205 *num_stripes_ret = num_stripes;
6206 *max_errors_ret = max_errors;
6207 bbio->num_tgtdevs = tgtdev_indexes;
6211 static bool need_full_stripe(enum btrfs_map_op op)
6213 return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS);
6217 * Calculate the geometry of a particular (address, len) tuple. This
6218 * information is used to calculate how big a particular bio can get before it
6219 * straddles a stripe.
6221 * @fs_info: the filesystem
6222 * @em: mapping containing the logical extent
6223 * @op: type of operation - write or read
6224 * @logical: address that we want to figure out the geometry of
6225 * @io_geom: pointer used to return values
6227 * Returns < 0 in case a chunk for the given logical address cannot be found,
6228 * usually shouldn't happen unless @logical is corrupted, 0 otherwise.
6230 int btrfs_get_io_geometry(struct btrfs_fs_info *fs_info, struct extent_map *em,
6231 enum btrfs_map_op op, u64 logical,
6232 struct btrfs_io_geometry *io_geom)
6234 struct map_lookup *map;
6240 u64 raid56_full_stripe_start = (u64)-1;
6243 ASSERT(op != BTRFS_MAP_DISCARD);
6245 map = em->map_lookup;
6246 /* Offset of this logical address in the chunk */
6247 offset = logical - em->start;
6248 /* Len of a stripe in a chunk */
6249 stripe_len = map->stripe_len;
6250 /* Stripe where this block falls in */
6251 stripe_nr = div64_u64(offset, stripe_len);
6252 /* Offset of stripe in the chunk */
6253 stripe_offset = stripe_nr * stripe_len;
6254 if (offset < stripe_offset) {
6256 "stripe math has gone wrong, stripe_offset=%llu offset=%llu start=%llu logical=%llu stripe_len=%llu",
6257 stripe_offset, offset, em->start, logical, stripe_len);
6261 /* stripe_offset is the offset of this block in its stripe */
6262 stripe_offset = offset - stripe_offset;
6263 data_stripes = nr_data_stripes(map);
6265 if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
6266 u64 max_len = stripe_len - stripe_offset;
6269 * In case of raid56, we need to know the stripe aligned start
6271 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6272 unsigned long full_stripe_len = stripe_len * data_stripes;
6273 raid56_full_stripe_start = offset;
6276 * Allow a write of a full stripe, but make sure we
6277 * don't allow straddling of stripes
6279 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
6281 raid56_full_stripe_start *= full_stripe_len;
6284 * For writes to RAID[56], allow a full stripeset across
6285 * all disks. For other RAID types and for RAID[56]
6286 * reads, just allow a single stripe (on a single disk).
6288 if (op == BTRFS_MAP_WRITE) {
6289 max_len = stripe_len * data_stripes -
6290 (offset - raid56_full_stripe_start);
6293 len = min_t(u64, em->len - offset, max_len);
6295 len = em->len - offset;
6299 io_geom->offset = offset;
6300 io_geom->stripe_len = stripe_len;
6301 io_geom->stripe_nr = stripe_nr;
6302 io_geom->stripe_offset = stripe_offset;
6303 io_geom->raid56_stripe_offset = raid56_full_stripe_start;
6308 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
6309 enum btrfs_map_op op,
6310 u64 logical, u64 *length,
6311 struct btrfs_bio **bbio_ret,
6312 int mirror_num, int need_raid_map)
6314 struct extent_map *em;
6315 struct map_lookup *map;
6325 int tgtdev_indexes = 0;
6326 struct btrfs_bio *bbio = NULL;
6327 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
6328 int dev_replace_is_ongoing = 0;
6329 int num_alloc_stripes;
6330 int patch_the_first_stripe_for_dev_replace = 0;
6331 u64 physical_to_patch_in_first_stripe = 0;
6332 u64 raid56_full_stripe_start = (u64)-1;
6333 struct btrfs_io_geometry geom;
6336 ASSERT(op != BTRFS_MAP_DISCARD);
6338 em = btrfs_get_chunk_map(fs_info, logical, *length);
6339 ASSERT(!IS_ERR(em));
6341 ret = btrfs_get_io_geometry(fs_info, em, op, logical, &geom);
6345 map = em->map_lookup;
6348 stripe_len = geom.stripe_len;
6349 stripe_nr = geom.stripe_nr;
6350 stripe_offset = geom.stripe_offset;
6351 raid56_full_stripe_start = geom.raid56_stripe_offset;
6352 data_stripes = nr_data_stripes(map);
6354 down_read(&dev_replace->rwsem);
6355 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
6357 * Hold the semaphore for read during the whole operation, write is
6358 * requested at commit time but must wait.
6360 if (!dev_replace_is_ongoing)
6361 up_read(&dev_replace->rwsem);
6363 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
6364 !need_full_stripe(op) && dev_replace->tgtdev != NULL) {
6365 ret = get_extra_mirror_from_replace(fs_info, logical, *length,
6366 dev_replace->srcdev->devid,
6368 &physical_to_patch_in_first_stripe);
6372 patch_the_first_stripe_for_dev_replace = 1;
6373 } else if (mirror_num > map->num_stripes) {
6379 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
6380 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
6382 if (!need_full_stripe(op))
6384 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1_MASK) {
6385 if (need_full_stripe(op))
6386 num_stripes = map->num_stripes;
6387 else if (mirror_num)
6388 stripe_index = mirror_num - 1;
6390 stripe_index = find_live_mirror(fs_info, map, 0,
6391 dev_replace_is_ongoing);
6392 mirror_num = stripe_index + 1;
6395 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
6396 if (need_full_stripe(op)) {
6397 num_stripes = map->num_stripes;
6398 } else if (mirror_num) {
6399 stripe_index = mirror_num - 1;
6404 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
6405 u32 factor = map->num_stripes / map->sub_stripes;
6407 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
6408 stripe_index *= map->sub_stripes;
6410 if (need_full_stripe(op))
6411 num_stripes = map->sub_stripes;
6412 else if (mirror_num)
6413 stripe_index += mirror_num - 1;
6415 int old_stripe_index = stripe_index;
6416 stripe_index = find_live_mirror(fs_info, map,
6418 dev_replace_is_ongoing);
6419 mirror_num = stripe_index - old_stripe_index + 1;
6422 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6423 if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) {
6424 /* push stripe_nr back to the start of the full stripe */
6425 stripe_nr = div64_u64(raid56_full_stripe_start,
6426 stripe_len * data_stripes);
6428 /* RAID[56] write or recovery. Return all stripes */
6429 num_stripes = map->num_stripes;
6430 max_errors = nr_parity_stripes(map);
6432 *length = map->stripe_len;
6437 * Mirror #0 or #1 means the original data block.
6438 * Mirror #2 is RAID5 parity block.
6439 * Mirror #3 is RAID6 Q block.
6441 stripe_nr = div_u64_rem(stripe_nr,
6442 data_stripes, &stripe_index);
6444 stripe_index = data_stripes + mirror_num - 2;
6446 /* We distribute the parity blocks across stripes */
6447 div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
6449 if (!need_full_stripe(op) && mirror_num <= 1)
6454 * after this, stripe_nr is the number of stripes on this
6455 * device we have to walk to find the data, and stripe_index is
6456 * the number of our device in the stripe array
6458 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
6460 mirror_num = stripe_index + 1;
6462 if (stripe_index >= map->num_stripes) {
6464 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
6465 stripe_index, map->num_stripes);
6470 num_alloc_stripes = num_stripes;
6471 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) {
6472 if (op == BTRFS_MAP_WRITE)
6473 num_alloc_stripes <<= 1;
6474 if (op == BTRFS_MAP_GET_READ_MIRRORS)
6475 num_alloc_stripes++;
6476 tgtdev_indexes = num_stripes;
6479 bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
6485 for (i = 0; i < num_stripes; i++) {
6486 bbio->stripes[i].physical = map->stripes[stripe_index].physical +
6487 stripe_offset + stripe_nr * map->stripe_len;
6488 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
6492 /* build raid_map */
6493 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map &&
6494 (need_full_stripe(op) || mirror_num > 1)) {
6498 /* Work out the disk rotation on this stripe-set */
6499 div_u64_rem(stripe_nr, num_stripes, &rot);
6501 /* Fill in the logical address of each stripe */
6502 tmp = stripe_nr * data_stripes;
6503 for (i = 0; i < data_stripes; i++)
6504 bbio->raid_map[(i+rot) % num_stripes] =
6505 em->start + (tmp + i) * map->stripe_len;
6507 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
6508 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
6509 bbio->raid_map[(i+rot+1) % num_stripes] =
6512 sort_parity_stripes(bbio, num_stripes);
6515 if (need_full_stripe(op))
6516 max_errors = btrfs_chunk_max_errors(map);
6518 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
6519 need_full_stripe(op)) {
6520 handle_ops_on_dev_replace(op, &bbio, dev_replace, logical,
6521 &num_stripes, &max_errors);
6525 bbio->map_type = map->type;
6526 bbio->num_stripes = num_stripes;
6527 bbio->max_errors = max_errors;
6528 bbio->mirror_num = mirror_num;
6531 * this is the case that REQ_READ && dev_replace_is_ongoing &&
6532 * mirror_num == num_stripes + 1 && dev_replace target drive is
6533 * available as a mirror
6535 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
6536 WARN_ON(num_stripes > 1);
6537 bbio->stripes[0].dev = dev_replace->tgtdev;
6538 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
6539 bbio->mirror_num = map->num_stripes + 1;
6542 if (dev_replace_is_ongoing) {
6543 lockdep_assert_held(&dev_replace->rwsem);
6544 /* Unlock and let waiting writers proceed */
6545 up_read(&dev_replace->rwsem);
6547 free_extent_map(em);
6551 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6552 u64 logical, u64 *length,
6553 struct btrfs_bio **bbio_ret, int mirror_num)
6555 if (op == BTRFS_MAP_DISCARD)
6556 return __btrfs_map_block_for_discard(fs_info, logical,
6559 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret,
6563 /* For Scrub/replace */
6564 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6565 u64 logical, u64 *length,
6566 struct btrfs_bio **bbio_ret)
6568 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1);
6571 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
6573 bio->bi_private = bbio->private;
6574 bio->bi_end_io = bbio->end_io;
6577 btrfs_put_bbio(bbio);
6580 static void btrfs_end_bio(struct bio *bio)
6582 struct btrfs_bio *bbio = bio->bi_private;
6583 int is_orig_bio = 0;
6585 if (bio->bi_status) {
6586 atomic_inc(&bbio->error);
6587 if (bio->bi_status == BLK_STS_IOERR ||
6588 bio->bi_status == BLK_STS_TARGET) {
6589 struct btrfs_device *dev = btrfs_io_bio(bio)->device;
6592 if (btrfs_op(bio) == BTRFS_MAP_WRITE)
6593 btrfs_dev_stat_inc_and_print(dev,
6594 BTRFS_DEV_STAT_WRITE_ERRS);
6595 else if (!(bio->bi_opf & REQ_RAHEAD))
6596 btrfs_dev_stat_inc_and_print(dev,
6597 BTRFS_DEV_STAT_READ_ERRS);
6598 if (bio->bi_opf & REQ_PREFLUSH)
6599 btrfs_dev_stat_inc_and_print(dev,
6600 BTRFS_DEV_STAT_FLUSH_ERRS);
6604 if (bio == bbio->orig_bio)
6607 btrfs_bio_counter_dec(bbio->fs_info);
6609 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6612 bio = bbio->orig_bio;
6615 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6616 /* only send an error to the higher layers if it is
6617 * beyond the tolerance of the btrfs bio
6619 if (atomic_read(&bbio->error) > bbio->max_errors) {
6620 bio->bi_status = BLK_STS_IOERR;
6623 * this bio is actually up to date, we didn't
6624 * go over the max number of errors
6626 bio->bi_status = BLK_STS_OK;
6629 btrfs_end_bbio(bbio, bio);
6630 } else if (!is_orig_bio) {
6635 static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio,
6636 u64 physical, struct btrfs_device *dev)
6638 struct btrfs_fs_info *fs_info = bbio->fs_info;
6640 bio->bi_private = bbio;
6641 btrfs_io_bio(bio)->device = dev;
6642 bio->bi_end_io = btrfs_end_bio;
6643 bio->bi_iter.bi_sector = physical >> 9;
6645 * For zone append writing, bi_sector must point the beginning of the
6648 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
6649 if (btrfs_dev_is_sequential(dev, physical)) {
6650 u64 zone_start = round_down(physical, fs_info->zone_size);
6652 bio->bi_iter.bi_sector = zone_start >> SECTOR_SHIFT;
6654 bio->bi_opf &= ~REQ_OP_ZONE_APPEND;
6655 bio->bi_opf |= REQ_OP_WRITE;
6658 btrfs_debug_in_rcu(fs_info,
6659 "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
6660 bio_op(bio), bio->bi_opf, bio->bi_iter.bi_sector,
6661 (unsigned long)dev->bdev->bd_dev, rcu_str_deref(dev->name),
6662 dev->devid, bio->bi_iter.bi_size);
6663 bio_set_dev(bio, dev->bdev);
6665 btrfs_bio_counter_inc_noblocked(fs_info);
6667 btrfsic_submit_bio(bio);
6670 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6672 atomic_inc(&bbio->error);
6673 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6674 /* Should be the original bio. */
6675 WARN_ON(bio != bbio->orig_bio);
6677 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6678 bio->bi_iter.bi_sector = logical >> 9;
6679 if (atomic_read(&bbio->error) > bbio->max_errors)
6680 bio->bi_status = BLK_STS_IOERR;
6682 bio->bi_status = BLK_STS_OK;
6683 btrfs_end_bbio(bbio, bio);
6687 blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
6690 struct btrfs_device *dev;
6691 struct bio *first_bio = bio;
6692 u64 logical = bio->bi_iter.bi_sector << 9;
6698 struct btrfs_bio *bbio = NULL;
6700 length = bio->bi_iter.bi_size;
6701 map_length = length;
6703 btrfs_bio_counter_inc_blocked(fs_info);
6704 ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical,
6705 &map_length, &bbio, mirror_num, 1);
6707 btrfs_bio_counter_dec(fs_info);
6708 return errno_to_blk_status(ret);
6711 total_devs = bbio->num_stripes;
6712 bbio->orig_bio = first_bio;
6713 bbio->private = first_bio->bi_private;
6714 bbio->end_io = first_bio->bi_end_io;
6715 bbio->fs_info = fs_info;
6716 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6718 if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6719 ((btrfs_op(bio) == BTRFS_MAP_WRITE) || (mirror_num > 1))) {
6720 /* In this case, map_length has been set to the length of
6721 a single stripe; not the whole write */
6722 if (btrfs_op(bio) == BTRFS_MAP_WRITE) {
6723 ret = raid56_parity_write(fs_info, bio, bbio,
6726 ret = raid56_parity_recover(fs_info, bio, bbio,
6727 map_length, mirror_num, 1);
6730 btrfs_bio_counter_dec(fs_info);
6731 return errno_to_blk_status(ret);
6734 if (map_length < length) {
6736 "mapping failed logical %llu bio len %llu len %llu",
6737 logical, length, map_length);
6741 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6742 dev = bbio->stripes[dev_nr].dev;
6743 if (!dev || !dev->bdev || test_bit(BTRFS_DEV_STATE_MISSING,
6745 (btrfs_op(first_bio) == BTRFS_MAP_WRITE &&
6746 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) {
6747 bbio_error(bbio, first_bio, logical);
6751 if (dev_nr < total_devs - 1)
6752 bio = btrfs_bio_clone(first_bio);
6756 submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical, dev);
6758 btrfs_bio_counter_dec(fs_info);
6763 * Find a device specified by @devid or @uuid in the list of @fs_devices, or
6766 * If devid and uuid are both specified, the match must be exact, otherwise
6767 * only devid is used.
6769 struct btrfs_device *btrfs_find_device(struct btrfs_fs_devices *fs_devices,
6770 u64 devid, u8 *uuid, u8 *fsid)
6772 struct btrfs_device *device;
6773 struct btrfs_fs_devices *seed_devs;
6775 if (!fsid || !memcmp(fs_devices->metadata_uuid, fsid, BTRFS_FSID_SIZE)) {
6776 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6777 if (device->devid == devid &&
6778 (!uuid || memcmp(device->uuid, uuid,
6779 BTRFS_UUID_SIZE) == 0))
6784 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
6786 !memcmp(seed_devs->metadata_uuid, fsid, BTRFS_FSID_SIZE)) {
6787 list_for_each_entry(device, &seed_devs->devices,
6789 if (device->devid == devid &&
6790 (!uuid || memcmp(device->uuid, uuid,
6791 BTRFS_UUID_SIZE) == 0))
6800 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices,
6801 u64 devid, u8 *dev_uuid)
6803 struct btrfs_device *device;
6804 unsigned int nofs_flag;
6807 * We call this under the chunk_mutex, so we want to use NOFS for this
6808 * allocation, however we don't want to change btrfs_alloc_device() to
6809 * always do NOFS because we use it in a lot of other GFP_KERNEL safe
6812 nofs_flag = memalloc_nofs_save();
6813 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6814 memalloc_nofs_restore(nofs_flag);
6818 list_add(&device->dev_list, &fs_devices->devices);
6819 device->fs_devices = fs_devices;
6820 fs_devices->num_devices++;
6822 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6823 fs_devices->missing_devices++;
6829 * btrfs_alloc_device - allocate struct btrfs_device
6830 * @fs_info: used only for generating a new devid, can be NULL if
6831 * devid is provided (i.e. @devid != NULL).
6832 * @devid: a pointer to devid for this device. If NULL a new devid
6834 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6837 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6838 * on error. Returned struct is not linked onto any lists and must be
6839 * destroyed with btrfs_free_device.
6841 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6845 struct btrfs_device *dev;
6848 if (WARN_ON(!devid && !fs_info))
6849 return ERR_PTR(-EINVAL);
6851 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
6853 return ERR_PTR(-ENOMEM);
6856 * Preallocate a bio that's always going to be used for flushing device
6857 * barriers and matches the device lifespan
6859 dev->flush_bio = bio_kmalloc(GFP_KERNEL, 0);
6860 if (!dev->flush_bio) {
6862 return ERR_PTR(-ENOMEM);
6865 INIT_LIST_HEAD(&dev->dev_list);
6866 INIT_LIST_HEAD(&dev->dev_alloc_list);
6867 INIT_LIST_HEAD(&dev->post_commit_list);
6869 atomic_set(&dev->reada_in_flight, 0);
6870 atomic_set(&dev->dev_stats_ccnt, 0);
6871 btrfs_device_data_ordered_init(dev);
6872 INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
6873 INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
6874 extent_io_tree_init(fs_info, &dev->alloc_state,
6875 IO_TREE_DEVICE_ALLOC_STATE, NULL);
6882 ret = find_next_devid(fs_info, &tmp);
6884 btrfs_free_device(dev);
6885 return ERR_PTR(ret);
6891 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6893 generate_random_uuid(dev->uuid);
6898 static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info,
6899 u64 devid, u8 *uuid, bool error)
6902 btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing",
6905 btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing",
6909 static u64 calc_stripe_length(u64 type, u64 chunk_len, int num_stripes)
6911 const int data_stripes = calc_data_stripes(type, num_stripes);
6913 return div_u64(chunk_len, data_stripes);
6916 #if BITS_PER_LONG == 32
6918 * Due to page cache limit, metadata beyond BTRFS_32BIT_MAX_FILE_SIZE
6919 * can't be accessed on 32bit systems.
6921 * This function do mount time check to reject the fs if it already has
6922 * metadata chunk beyond that limit.
6924 static int check_32bit_meta_chunk(struct btrfs_fs_info *fs_info,
6925 u64 logical, u64 length, u64 type)
6927 if (!(type & BTRFS_BLOCK_GROUP_METADATA))
6930 if (logical + length < MAX_LFS_FILESIZE)
6933 btrfs_err_32bit_limit(fs_info);
6938 * This is to give early warning for any metadata chunk reaching
6939 * BTRFS_32BIT_EARLY_WARN_THRESHOLD.
6940 * Although we can still access the metadata, it's not going to be possible
6941 * once the limit is reached.
6943 static void warn_32bit_meta_chunk(struct btrfs_fs_info *fs_info,
6944 u64 logical, u64 length, u64 type)
6946 if (!(type & BTRFS_BLOCK_GROUP_METADATA))
6949 if (logical + length < BTRFS_32BIT_EARLY_WARN_THRESHOLD)
6952 btrfs_warn_32bit_limit(fs_info);
6956 static int read_one_chunk(struct btrfs_key *key, struct extent_buffer *leaf,
6957 struct btrfs_chunk *chunk)
6959 struct btrfs_fs_info *fs_info = leaf->fs_info;
6960 struct extent_map_tree *map_tree = &fs_info->mapping_tree;
6961 struct map_lookup *map;
6962 struct extent_map *em;
6967 u8 uuid[BTRFS_UUID_SIZE];
6972 logical = key->offset;
6973 length = btrfs_chunk_length(leaf, chunk);
6974 type = btrfs_chunk_type(leaf, chunk);
6975 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6977 #if BITS_PER_LONG == 32
6978 ret = check_32bit_meta_chunk(fs_info, logical, length, type);
6981 warn_32bit_meta_chunk(fs_info, logical, length, type);
6985 * Only need to verify chunk item if we're reading from sys chunk array,
6986 * as chunk item in tree block is already verified by tree-checker.
6988 if (leaf->start == BTRFS_SUPER_INFO_OFFSET) {
6989 ret = btrfs_check_chunk_valid(leaf, chunk, logical);
6994 read_lock(&map_tree->lock);
6995 em = lookup_extent_mapping(map_tree, logical, 1);
6996 read_unlock(&map_tree->lock);
6998 /* already mapped? */
6999 if (em && em->start <= logical && em->start + em->len > logical) {
7000 free_extent_map(em);
7003 free_extent_map(em);
7006 em = alloc_extent_map();
7009 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
7011 free_extent_map(em);
7015 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
7016 em->map_lookup = map;
7017 em->start = logical;
7020 em->block_start = 0;
7021 em->block_len = em->len;
7023 map->num_stripes = num_stripes;
7024 map->io_width = btrfs_chunk_io_width(leaf, chunk);
7025 map->io_align = btrfs_chunk_io_align(leaf, chunk);
7026 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
7028 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
7029 map->verified_stripes = 0;
7030 em->orig_block_len = calc_stripe_length(type, em->len,
7032 for (i = 0; i < num_stripes; i++) {
7033 map->stripes[i].physical =
7034 btrfs_stripe_offset_nr(leaf, chunk, i);
7035 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
7036 read_extent_buffer(leaf, uuid, (unsigned long)
7037 btrfs_stripe_dev_uuid_nr(chunk, i),
7039 map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices,
7041 if (!map->stripes[i].dev &&
7042 !btrfs_test_opt(fs_info, DEGRADED)) {
7043 free_extent_map(em);
7044 btrfs_report_missing_device(fs_info, devid, uuid, true);
7047 if (!map->stripes[i].dev) {
7048 map->stripes[i].dev =
7049 add_missing_dev(fs_info->fs_devices, devid,
7051 if (IS_ERR(map->stripes[i].dev)) {
7052 free_extent_map(em);
7054 "failed to init missing dev %llu: %ld",
7055 devid, PTR_ERR(map->stripes[i].dev));
7056 return PTR_ERR(map->stripes[i].dev);
7058 btrfs_report_missing_device(fs_info, devid, uuid, false);
7060 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
7061 &(map->stripes[i].dev->dev_state));
7065 write_lock(&map_tree->lock);
7066 ret = add_extent_mapping(map_tree, em, 0);
7067 write_unlock(&map_tree->lock);
7070 "failed to add chunk map, start=%llu len=%llu: %d",
7071 em->start, em->len, ret);
7073 free_extent_map(em);
7078 static void fill_device_from_item(struct extent_buffer *leaf,
7079 struct btrfs_dev_item *dev_item,
7080 struct btrfs_device *device)
7084 device->devid = btrfs_device_id(leaf, dev_item);
7085 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
7086 device->total_bytes = device->disk_total_bytes;
7087 device->commit_total_bytes = device->disk_total_bytes;
7088 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
7089 device->commit_bytes_used = device->bytes_used;
7090 device->type = btrfs_device_type(leaf, dev_item);
7091 device->io_align = btrfs_device_io_align(leaf, dev_item);
7092 device->io_width = btrfs_device_io_width(leaf, dev_item);
7093 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
7094 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
7095 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
7097 ptr = btrfs_device_uuid(dev_item);
7098 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
7101 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info,
7104 struct btrfs_fs_devices *fs_devices;
7107 lockdep_assert_held(&uuid_mutex);
7110 /* This will match only for multi-device seed fs */
7111 list_for_each_entry(fs_devices, &fs_info->fs_devices->seed_list, seed_list)
7112 if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE))
7116 fs_devices = find_fsid(fsid, NULL);
7118 if (!btrfs_test_opt(fs_info, DEGRADED))
7119 return ERR_PTR(-ENOENT);
7121 fs_devices = alloc_fs_devices(fsid, NULL);
7122 if (IS_ERR(fs_devices))
7125 fs_devices->seeding = true;
7126 fs_devices->opened = 1;
7131 * Upon first call for a seed fs fsid, just create a private copy of the
7132 * respective fs_devices and anchor it at fs_info->fs_devices->seed_list
7134 fs_devices = clone_fs_devices(fs_devices);
7135 if (IS_ERR(fs_devices))
7138 ret = open_fs_devices(fs_devices, FMODE_READ, fs_info->bdev_holder);
7140 free_fs_devices(fs_devices);
7141 return ERR_PTR(ret);
7144 if (!fs_devices->seeding) {
7145 close_fs_devices(fs_devices);
7146 free_fs_devices(fs_devices);
7147 return ERR_PTR(-EINVAL);
7150 list_add(&fs_devices->seed_list, &fs_info->fs_devices->seed_list);
7155 static int read_one_dev(struct extent_buffer *leaf,
7156 struct btrfs_dev_item *dev_item)
7158 struct btrfs_fs_info *fs_info = leaf->fs_info;
7159 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7160 struct btrfs_device *device;
7163 u8 fs_uuid[BTRFS_FSID_SIZE];
7164 u8 dev_uuid[BTRFS_UUID_SIZE];
7166 devid = btrfs_device_id(leaf, dev_item);
7167 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
7169 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
7172 if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) {
7173 fs_devices = open_seed_devices(fs_info, fs_uuid);
7174 if (IS_ERR(fs_devices))
7175 return PTR_ERR(fs_devices);
7178 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
7181 if (!btrfs_test_opt(fs_info, DEGRADED)) {
7182 btrfs_report_missing_device(fs_info, devid,
7187 device = add_missing_dev(fs_devices, devid, dev_uuid);
7188 if (IS_ERR(device)) {
7190 "failed to add missing dev %llu: %ld",
7191 devid, PTR_ERR(device));
7192 return PTR_ERR(device);
7194 btrfs_report_missing_device(fs_info, devid, dev_uuid, false);
7196 if (!device->bdev) {
7197 if (!btrfs_test_opt(fs_info, DEGRADED)) {
7198 btrfs_report_missing_device(fs_info,
7199 devid, dev_uuid, true);
7202 btrfs_report_missing_device(fs_info, devid,
7206 if (!device->bdev &&
7207 !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
7209 * this happens when a device that was properly setup
7210 * in the device info lists suddenly goes bad.
7211 * device->bdev is NULL, and so we have to set
7212 * device->missing to one here
7214 device->fs_devices->missing_devices++;
7215 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
7218 /* Move the device to its own fs_devices */
7219 if (device->fs_devices != fs_devices) {
7220 ASSERT(test_bit(BTRFS_DEV_STATE_MISSING,
7221 &device->dev_state));
7223 list_move(&device->dev_list, &fs_devices->devices);
7224 device->fs_devices->num_devices--;
7225 fs_devices->num_devices++;
7227 device->fs_devices->missing_devices--;
7228 fs_devices->missing_devices++;
7230 device->fs_devices = fs_devices;
7234 if (device->fs_devices != fs_info->fs_devices) {
7235 BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state));
7236 if (device->generation !=
7237 btrfs_device_generation(leaf, dev_item))
7241 fill_device_from_item(leaf, dev_item, device);
7243 u64 max_total_bytes = i_size_read(device->bdev->bd_inode);
7245 if (device->total_bytes > max_total_bytes) {
7247 "device total_bytes should be at most %llu but found %llu",
7248 max_total_bytes, device->total_bytes);
7252 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
7253 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
7254 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
7255 device->fs_devices->total_rw_bytes += device->total_bytes;
7256 atomic64_add(device->total_bytes - device->bytes_used,
7257 &fs_info->free_chunk_space);
7263 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
7265 struct btrfs_root *root = fs_info->tree_root;
7266 struct btrfs_super_block *super_copy = fs_info->super_copy;
7267 struct extent_buffer *sb;
7268 struct btrfs_disk_key *disk_key;
7269 struct btrfs_chunk *chunk;
7271 unsigned long sb_array_offset;
7278 struct btrfs_key key;
7280 ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize);
7282 * This will create extent buffer of nodesize, superblock size is
7283 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
7284 * overallocate but we can keep it as-is, only the first page is used.
7286 sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET,
7287 root->root_key.objectid, 0);
7290 set_extent_buffer_uptodate(sb);
7292 * The sb extent buffer is artificial and just used to read the system array.
7293 * set_extent_buffer_uptodate() call does not properly mark all it's
7294 * pages up-to-date when the page is larger: extent does not cover the
7295 * whole page and consequently check_page_uptodate does not find all
7296 * the page's extents up-to-date (the hole beyond sb),
7297 * write_extent_buffer then triggers a WARN_ON.
7299 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
7300 * but sb spans only this function. Add an explicit SetPageUptodate call
7301 * to silence the warning eg. on PowerPC 64.
7303 if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
7304 SetPageUptodate(sb->pages[0]);
7306 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
7307 array_size = btrfs_super_sys_array_size(super_copy);
7309 array_ptr = super_copy->sys_chunk_array;
7310 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
7313 while (cur_offset < array_size) {
7314 disk_key = (struct btrfs_disk_key *)array_ptr;
7315 len = sizeof(*disk_key);
7316 if (cur_offset + len > array_size)
7317 goto out_short_read;
7319 btrfs_disk_key_to_cpu(&key, disk_key);
7322 sb_array_offset += len;
7325 if (key.type != BTRFS_CHUNK_ITEM_KEY) {
7327 "unexpected item type %u in sys_array at offset %u",
7328 (u32)key.type, cur_offset);
7333 chunk = (struct btrfs_chunk *)sb_array_offset;
7335 * At least one btrfs_chunk with one stripe must be present,
7336 * exact stripe count check comes afterwards
7338 len = btrfs_chunk_item_size(1);
7339 if (cur_offset + len > array_size)
7340 goto out_short_read;
7342 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
7345 "invalid number of stripes %u in sys_array at offset %u",
7346 num_stripes, cur_offset);
7351 type = btrfs_chunk_type(sb, chunk);
7352 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
7354 "invalid chunk type %llu in sys_array at offset %u",
7360 len = btrfs_chunk_item_size(num_stripes);
7361 if (cur_offset + len > array_size)
7362 goto out_short_read;
7364 ret = read_one_chunk(&key, sb, chunk);
7369 sb_array_offset += len;
7372 clear_extent_buffer_uptodate(sb);
7373 free_extent_buffer_stale(sb);
7377 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
7379 clear_extent_buffer_uptodate(sb);
7380 free_extent_buffer_stale(sb);
7385 * Check if all chunks in the fs are OK for read-write degraded mount
7387 * If the @failing_dev is specified, it's accounted as missing.
7389 * Return true if all chunks meet the minimal RW mount requirements.
7390 * Return false if any chunk doesn't meet the minimal RW mount requirements.
7392 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
7393 struct btrfs_device *failing_dev)
7395 struct extent_map_tree *map_tree = &fs_info->mapping_tree;
7396 struct extent_map *em;
7400 read_lock(&map_tree->lock);
7401 em = lookup_extent_mapping(map_tree, 0, (u64)-1);
7402 read_unlock(&map_tree->lock);
7403 /* No chunk at all? Return false anyway */
7409 struct map_lookup *map;
7414 map = em->map_lookup;
7416 btrfs_get_num_tolerated_disk_barrier_failures(
7418 for (i = 0; i < map->num_stripes; i++) {
7419 struct btrfs_device *dev = map->stripes[i].dev;
7421 if (!dev || !dev->bdev ||
7422 test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) ||
7423 dev->last_flush_error)
7425 else if (failing_dev && failing_dev == dev)
7428 if (missing > max_tolerated) {
7431 "chunk %llu missing %d devices, max tolerance is %d for writable mount",
7432 em->start, missing, max_tolerated);
7433 free_extent_map(em);
7437 next_start = extent_map_end(em);
7438 free_extent_map(em);
7440 read_lock(&map_tree->lock);
7441 em = lookup_extent_mapping(map_tree, next_start,
7442 (u64)(-1) - next_start);
7443 read_unlock(&map_tree->lock);
7449 static void readahead_tree_node_children(struct extent_buffer *node)
7452 const int nr_items = btrfs_header_nritems(node);
7454 for (i = 0; i < nr_items; i++)
7455 btrfs_readahead_node_child(node, i);
7458 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
7460 struct btrfs_root *root = fs_info->chunk_root;
7461 struct btrfs_path *path;
7462 struct extent_buffer *leaf;
7463 struct btrfs_key key;
7464 struct btrfs_key found_key;
7468 u64 last_ra_node = 0;
7470 path = btrfs_alloc_path();
7475 * uuid_mutex is needed only if we are mounting a sprout FS
7476 * otherwise we don't need it.
7478 mutex_lock(&uuid_mutex);
7481 * It is possible for mount and umount to race in such a way that
7482 * we execute this code path, but open_fs_devices failed to clear
7483 * total_rw_bytes. We certainly want it cleared before reading the
7484 * device items, so clear it here.
7486 fs_info->fs_devices->total_rw_bytes = 0;
7489 * Lockdep complains about possible circular locking dependency between
7490 * a disk's open_mutex (struct gendisk.open_mutex), the rw semaphores
7491 * used for freeze procection of a fs (struct super_block.s_writers),
7492 * which we take when starting a transaction, and extent buffers of the
7493 * chunk tree if we call read_one_dev() while holding a lock on an
7494 * extent buffer of the chunk tree. Since we are mounting the filesystem
7495 * and at this point there can't be any concurrent task modifying the
7496 * chunk tree, to keep it simple, just skip locking on the chunk tree.
7498 ASSERT(!test_bit(BTRFS_FS_OPEN, &fs_info->flags));
7499 path->skip_locking = 1;
7502 * Read all device items, and then all the chunk items. All
7503 * device items are found before any chunk item (their object id
7504 * is smaller than the lowest possible object id for a chunk
7505 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
7507 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
7510 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7514 struct extent_buffer *node;
7516 leaf = path->nodes[0];
7517 slot = path->slots[0];
7518 if (slot >= btrfs_header_nritems(leaf)) {
7519 ret = btrfs_next_leaf(root, path);
7526 node = path->nodes[1];
7528 if (last_ra_node != node->start) {
7529 readahead_tree_node_children(node);
7530 last_ra_node = node->start;
7533 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7534 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
7535 struct btrfs_dev_item *dev_item;
7536 dev_item = btrfs_item_ptr(leaf, slot,
7537 struct btrfs_dev_item);
7538 ret = read_one_dev(leaf, dev_item);
7542 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
7543 struct btrfs_chunk *chunk;
7546 * We are only called at mount time, so no need to take
7547 * fs_info->chunk_mutex. Plus, to avoid lockdep warnings,
7548 * we always lock first fs_info->chunk_mutex before
7549 * acquiring any locks on the chunk tree. This is a
7550 * requirement for chunk allocation, see the comment on
7551 * top of btrfs_chunk_alloc() for details.
7553 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
7554 ret = read_one_chunk(&found_key, leaf, chunk);
7562 * After loading chunk tree, we've got all device information,
7563 * do another round of validation checks.
7565 if (total_dev != fs_info->fs_devices->total_devices) {
7567 "super_num_devices %llu mismatch with num_devices %llu found here",
7568 btrfs_super_num_devices(fs_info->super_copy),
7573 if (btrfs_super_total_bytes(fs_info->super_copy) <
7574 fs_info->fs_devices->total_rw_bytes) {
7576 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
7577 btrfs_super_total_bytes(fs_info->super_copy),
7578 fs_info->fs_devices->total_rw_bytes);
7584 mutex_unlock(&uuid_mutex);
7586 btrfs_free_path(path);
7590 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
7592 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
7593 struct btrfs_device *device;
7595 fs_devices->fs_info = fs_info;
7597 mutex_lock(&fs_devices->device_list_mutex);
7598 list_for_each_entry(device, &fs_devices->devices, dev_list)
7599 device->fs_info = fs_info;
7601 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
7602 list_for_each_entry(device, &seed_devs->devices, dev_list)
7603 device->fs_info = fs_info;
7605 seed_devs->fs_info = fs_info;
7607 mutex_unlock(&fs_devices->device_list_mutex);
7610 static u64 btrfs_dev_stats_value(const struct extent_buffer *eb,
7611 const struct btrfs_dev_stats_item *ptr,
7616 read_extent_buffer(eb, &val,
7617 offsetof(struct btrfs_dev_stats_item, values) +
7618 ((unsigned long)ptr) + (index * sizeof(u64)),
7623 static void btrfs_set_dev_stats_value(struct extent_buffer *eb,
7624 struct btrfs_dev_stats_item *ptr,
7627 write_extent_buffer(eb, &val,
7628 offsetof(struct btrfs_dev_stats_item, values) +
7629 ((unsigned long)ptr) + (index * sizeof(u64)),
7633 static int btrfs_device_init_dev_stats(struct btrfs_device *device,
7634 struct btrfs_path *path)
7636 struct btrfs_dev_stats_item *ptr;
7637 struct extent_buffer *eb;
7638 struct btrfs_key key;
7642 if (!device->fs_info->dev_root)
7645 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7646 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7647 key.offset = device->devid;
7648 ret = btrfs_search_slot(NULL, device->fs_info->dev_root, &key, path, 0, 0);
7650 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7651 btrfs_dev_stat_set(device, i, 0);
7652 device->dev_stats_valid = 1;
7653 btrfs_release_path(path);
7654 return ret < 0 ? ret : 0;
7656 slot = path->slots[0];
7657 eb = path->nodes[0];
7658 item_size = btrfs_item_size_nr(eb, slot);
7660 ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_stats_item);
7662 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7663 if (item_size >= (1 + i) * sizeof(__le64))
7664 btrfs_dev_stat_set(device, i,
7665 btrfs_dev_stats_value(eb, ptr, i));
7667 btrfs_dev_stat_set(device, i, 0);
7670 device->dev_stats_valid = 1;
7671 btrfs_dev_stat_print_on_load(device);
7672 btrfs_release_path(path);
7677 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
7679 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
7680 struct btrfs_device *device;
7681 struct btrfs_path *path = NULL;
7684 path = btrfs_alloc_path();
7688 mutex_lock(&fs_devices->device_list_mutex);
7689 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7690 ret = btrfs_device_init_dev_stats(device, path);
7694 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
7695 list_for_each_entry(device, &seed_devs->devices, dev_list) {
7696 ret = btrfs_device_init_dev_stats(device, path);
7702 mutex_unlock(&fs_devices->device_list_mutex);
7704 btrfs_free_path(path);
7708 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
7709 struct btrfs_device *device)
7711 struct btrfs_fs_info *fs_info = trans->fs_info;
7712 struct btrfs_root *dev_root = fs_info->dev_root;
7713 struct btrfs_path *path;
7714 struct btrfs_key key;
7715 struct extent_buffer *eb;
7716 struct btrfs_dev_stats_item *ptr;
7720 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7721 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7722 key.offset = device->devid;
7724 path = btrfs_alloc_path();
7727 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
7729 btrfs_warn_in_rcu(fs_info,
7730 "error %d while searching for dev_stats item for device %s",
7731 ret, rcu_str_deref(device->name));
7736 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
7737 /* need to delete old one and insert a new one */
7738 ret = btrfs_del_item(trans, dev_root, path);
7740 btrfs_warn_in_rcu(fs_info,
7741 "delete too small dev_stats item for device %s failed %d",
7742 rcu_str_deref(device->name), ret);
7749 /* need to insert a new item */
7750 btrfs_release_path(path);
7751 ret = btrfs_insert_empty_item(trans, dev_root, path,
7752 &key, sizeof(*ptr));
7754 btrfs_warn_in_rcu(fs_info,
7755 "insert dev_stats item for device %s failed %d",
7756 rcu_str_deref(device->name), ret);
7761 eb = path->nodes[0];
7762 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7763 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7764 btrfs_set_dev_stats_value(eb, ptr, i,
7765 btrfs_dev_stat_read(device, i));
7766 btrfs_mark_buffer_dirty(eb);
7769 btrfs_free_path(path);
7774 * called from commit_transaction. Writes all changed device stats to disk.
7776 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans)
7778 struct btrfs_fs_info *fs_info = trans->fs_info;
7779 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7780 struct btrfs_device *device;
7784 mutex_lock(&fs_devices->device_list_mutex);
7785 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7786 stats_cnt = atomic_read(&device->dev_stats_ccnt);
7787 if (!device->dev_stats_valid || stats_cnt == 0)
7792 * There is a LOAD-LOAD control dependency between the value of
7793 * dev_stats_ccnt and updating the on-disk values which requires
7794 * reading the in-memory counters. Such control dependencies
7795 * require explicit read memory barriers.
7797 * This memory barriers pairs with smp_mb__before_atomic in
7798 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full
7799 * barrier implied by atomic_xchg in
7800 * btrfs_dev_stats_read_and_reset
7804 ret = update_dev_stat_item(trans, device);
7806 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7808 mutex_unlock(&fs_devices->device_list_mutex);
7813 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7815 btrfs_dev_stat_inc(dev, index);
7816 btrfs_dev_stat_print_on_error(dev);
7819 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
7821 if (!dev->dev_stats_valid)
7823 btrfs_err_rl_in_rcu(dev->fs_info,
7824 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7825 rcu_str_deref(dev->name),
7826 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7827 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7828 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7829 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7830 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7833 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7837 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7838 if (btrfs_dev_stat_read(dev, i) != 0)
7840 if (i == BTRFS_DEV_STAT_VALUES_MAX)
7841 return; /* all values == 0, suppress message */
7843 btrfs_info_in_rcu(dev->fs_info,
7844 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7845 rcu_str_deref(dev->name),
7846 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7847 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7848 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7849 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7850 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7853 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
7854 struct btrfs_ioctl_get_dev_stats *stats)
7856 struct btrfs_device *dev;
7857 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7860 mutex_lock(&fs_devices->device_list_mutex);
7861 dev = btrfs_find_device(fs_info->fs_devices, stats->devid, NULL, NULL);
7862 mutex_unlock(&fs_devices->device_list_mutex);
7865 btrfs_warn(fs_info, "get dev_stats failed, device not found");
7867 } else if (!dev->dev_stats_valid) {
7868 btrfs_warn(fs_info, "get dev_stats failed, not yet valid");
7870 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7871 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7872 if (stats->nr_items > i)
7874 btrfs_dev_stat_read_and_reset(dev, i);
7876 btrfs_dev_stat_set(dev, i, 0);
7878 btrfs_info(fs_info, "device stats zeroed by %s (%d)",
7879 current->comm, task_pid_nr(current));
7881 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7882 if (stats->nr_items > i)
7883 stats->values[i] = btrfs_dev_stat_read(dev, i);
7885 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7886 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7891 * Update the size and bytes used for each device where it changed. This is
7892 * delayed since we would otherwise get errors while writing out the
7895 * Must be invoked during transaction commit.
7897 void btrfs_commit_device_sizes(struct btrfs_transaction *trans)
7899 struct btrfs_device *curr, *next;
7901 ASSERT(trans->state == TRANS_STATE_COMMIT_DOING);
7903 if (list_empty(&trans->dev_update_list))
7907 * We don't need the device_list_mutex here. This list is owned by the
7908 * transaction and the transaction must complete before the device is
7911 mutex_lock(&trans->fs_info->chunk_mutex);
7912 list_for_each_entry_safe(curr, next, &trans->dev_update_list,
7914 list_del_init(&curr->post_commit_list);
7915 curr->commit_total_bytes = curr->disk_total_bytes;
7916 curr->commit_bytes_used = curr->bytes_used;
7918 mutex_unlock(&trans->fs_info->chunk_mutex);
7922 * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10.
7924 int btrfs_bg_type_to_factor(u64 flags)
7926 const int index = btrfs_bg_flags_to_raid_index(flags);
7928 return btrfs_raid_array[index].ncopies;
7933 static int verify_one_dev_extent(struct btrfs_fs_info *fs_info,
7934 u64 chunk_offset, u64 devid,
7935 u64 physical_offset, u64 physical_len)
7937 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
7938 struct extent_map *em;
7939 struct map_lookup *map;
7940 struct btrfs_device *dev;
7946 read_lock(&em_tree->lock);
7947 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
7948 read_unlock(&em_tree->lock);
7952 "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk",
7953 physical_offset, devid);
7958 map = em->map_lookup;
7959 stripe_len = calc_stripe_length(map->type, em->len, map->num_stripes);
7960 if (physical_len != stripe_len) {
7962 "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu",
7963 physical_offset, devid, em->start, physical_len,
7969 for (i = 0; i < map->num_stripes; i++) {
7970 if (map->stripes[i].dev->devid == devid &&
7971 map->stripes[i].physical == physical_offset) {
7973 if (map->verified_stripes >= map->num_stripes) {
7975 "too many dev extents for chunk %llu found",
7980 map->verified_stripes++;
7986 "dev extent physical offset %llu devid %llu has no corresponding chunk",
7987 physical_offset, devid);
7991 /* Make sure no dev extent is beyond device boundary */
7992 dev = btrfs_find_device(fs_info->fs_devices, devid, NULL, NULL);
7994 btrfs_err(fs_info, "failed to find devid %llu", devid);
7999 if (physical_offset + physical_len > dev->disk_total_bytes) {
8001 "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu",
8002 devid, physical_offset, physical_len,
8003 dev->disk_total_bytes);
8008 if (dev->zone_info) {
8009 u64 zone_size = dev->zone_info->zone_size;
8011 if (!IS_ALIGNED(physical_offset, zone_size) ||
8012 !IS_ALIGNED(physical_len, zone_size)) {
8014 "zoned: dev extent devid %llu physical offset %llu len %llu is not aligned to device zone",
8015 devid, physical_offset, physical_len);
8022 free_extent_map(em);
8026 static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info)
8028 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
8029 struct extent_map *em;
8030 struct rb_node *node;
8033 read_lock(&em_tree->lock);
8034 for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) {
8035 em = rb_entry(node, struct extent_map, rb_node);
8036 if (em->map_lookup->num_stripes !=
8037 em->map_lookup->verified_stripes) {
8039 "chunk %llu has missing dev extent, have %d expect %d",
8040 em->start, em->map_lookup->verified_stripes,
8041 em->map_lookup->num_stripes);
8047 read_unlock(&em_tree->lock);
8052 * Ensure that all dev extents are mapped to correct chunk, otherwise
8053 * later chunk allocation/free would cause unexpected behavior.
8055 * NOTE: This will iterate through the whole device tree, which should be of
8056 * the same size level as the chunk tree. This slightly increases mount time.
8058 int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info)
8060 struct btrfs_path *path;
8061 struct btrfs_root *root = fs_info->dev_root;
8062 struct btrfs_key key;
8064 u64 prev_dev_ext_end = 0;
8068 * We don't have a dev_root because we mounted with ignorebadroots and
8069 * failed to load the root, so we want to skip the verification in this
8072 * However if the dev root is fine, but the tree itself is corrupted
8073 * we'd still fail to mount. This verification is only to make sure
8074 * writes can happen safely, so instead just bypass this check
8075 * completely in the case of IGNOREBADROOTS.
8077 if (btrfs_test_opt(fs_info, IGNOREBADROOTS))
8081 key.type = BTRFS_DEV_EXTENT_KEY;
8084 path = btrfs_alloc_path();
8088 path->reada = READA_FORWARD;
8089 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
8093 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
8094 ret = btrfs_next_leaf(root, path);
8097 /* No dev extents at all? Not good */
8104 struct extent_buffer *leaf = path->nodes[0];
8105 struct btrfs_dev_extent *dext;
8106 int slot = path->slots[0];
8108 u64 physical_offset;
8112 btrfs_item_key_to_cpu(leaf, &key, slot);
8113 if (key.type != BTRFS_DEV_EXTENT_KEY)
8115 devid = key.objectid;
8116 physical_offset = key.offset;
8118 dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent);
8119 chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext);
8120 physical_len = btrfs_dev_extent_length(leaf, dext);
8122 /* Check if this dev extent overlaps with the previous one */
8123 if (devid == prev_devid && physical_offset < prev_dev_ext_end) {
8125 "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu",
8126 devid, physical_offset, prev_dev_ext_end);
8131 ret = verify_one_dev_extent(fs_info, chunk_offset, devid,
8132 physical_offset, physical_len);
8136 prev_dev_ext_end = physical_offset + physical_len;
8138 ret = btrfs_next_item(root, path);
8147 /* Ensure all chunks have corresponding dev extents */
8148 ret = verify_chunk_dev_extent_mapping(fs_info);
8150 btrfs_free_path(path);
8155 * Check whether the given block group or device is pinned by any inode being
8156 * used as a swapfile.
8158 bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr)
8160 struct btrfs_swapfile_pin *sp;
8161 struct rb_node *node;
8163 spin_lock(&fs_info->swapfile_pins_lock);
8164 node = fs_info->swapfile_pins.rb_node;
8166 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
8168 node = node->rb_left;
8169 else if (ptr > sp->ptr)
8170 node = node->rb_right;
8174 spin_unlock(&fs_info->swapfile_pins_lock);
8175 return node != NULL;
8178 static int relocating_repair_kthread(void *data)
8180 struct btrfs_block_group *cache = (struct btrfs_block_group *)data;
8181 struct btrfs_fs_info *fs_info = cache->fs_info;
8185 target = cache->start;
8186 btrfs_put_block_group(cache);
8188 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) {
8190 "zoned: skip relocating block group %llu to repair: EBUSY",
8195 mutex_lock(&fs_info->reclaim_bgs_lock);
8197 /* Ensure block group still exists */
8198 cache = btrfs_lookup_block_group(fs_info, target);
8202 if (!cache->relocating_repair)
8205 ret = btrfs_may_alloc_data_chunk(fs_info, target);
8210 "zoned: relocating block group %llu to repair IO failure",
8212 ret = btrfs_relocate_chunk(fs_info, target);
8216 btrfs_put_block_group(cache);
8217 mutex_unlock(&fs_info->reclaim_bgs_lock);
8218 btrfs_exclop_finish(fs_info);
8223 int btrfs_repair_one_zone(struct btrfs_fs_info *fs_info, u64 logical)
8225 struct btrfs_block_group *cache;
8227 /* Do not attempt to repair in degraded state */
8228 if (btrfs_test_opt(fs_info, DEGRADED))
8231 cache = btrfs_lookup_block_group(fs_info, logical);
8235 spin_lock(&cache->lock);
8236 if (cache->relocating_repair) {
8237 spin_unlock(&cache->lock);
8238 btrfs_put_block_group(cache);
8241 cache->relocating_repair = 1;
8242 spin_unlock(&cache->lock);
8244 kthread_run(relocating_repair_kthread, cache,
8245 "btrfs-relocating-repair");