1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
6 #include <linux/sched.h>
7 #include <linux/sched/mm.h>
9 #include <linux/slab.h>
10 #include <linux/blkdev.h>
11 #include <linux/ratelimit.h>
12 #include <linux/kthread.h>
13 #include <linux/raid/pq.h>
14 #include <linux/semaphore.h>
15 #include <linux/uuid.h>
16 #include <linux/list_sort.h>
17 #include <linux/namei.h>
20 #include "extent_map.h"
22 #include "transaction.h"
23 #include "print-tree.h"
26 #include "async-thread.h"
27 #include "check-integrity.h"
28 #include "rcu-string.h"
29 #include "dev-replace.h"
31 #include "tree-checker.h"
32 #include "space-info.h"
33 #include "block-group.h"
37 const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
38 [BTRFS_RAID_RAID10] = {
41 .devs_max = 0, /* 0 == as many as possible */
43 .tolerated_failures = 1,
47 .raid_name = "raid10",
48 .bg_flag = BTRFS_BLOCK_GROUP_RAID10,
49 .mindev_error = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET,
51 [BTRFS_RAID_RAID1] = {
56 .tolerated_failures = 1,
61 .bg_flag = BTRFS_BLOCK_GROUP_RAID1,
62 .mindev_error = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET,
64 [BTRFS_RAID_RAID1C3] = {
69 .tolerated_failures = 2,
73 .raid_name = "raid1c3",
74 .bg_flag = BTRFS_BLOCK_GROUP_RAID1C3,
75 .mindev_error = BTRFS_ERROR_DEV_RAID1C3_MIN_NOT_MET,
77 [BTRFS_RAID_RAID1C4] = {
82 .tolerated_failures = 3,
86 .raid_name = "raid1c4",
87 .bg_flag = BTRFS_BLOCK_GROUP_RAID1C4,
88 .mindev_error = BTRFS_ERROR_DEV_RAID1C4_MIN_NOT_MET,
95 .tolerated_failures = 0,
100 .bg_flag = BTRFS_BLOCK_GROUP_DUP,
103 [BTRFS_RAID_RAID0] = {
108 .tolerated_failures = 0,
112 .raid_name = "raid0",
113 .bg_flag = BTRFS_BLOCK_GROUP_RAID0,
116 [BTRFS_RAID_SINGLE] = {
121 .tolerated_failures = 0,
125 .raid_name = "single",
129 [BTRFS_RAID_RAID5] = {
134 .tolerated_failures = 1,
138 .raid_name = "raid5",
139 .bg_flag = BTRFS_BLOCK_GROUP_RAID5,
140 .mindev_error = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET,
142 [BTRFS_RAID_RAID6] = {
147 .tolerated_failures = 2,
151 .raid_name = "raid6",
152 .bg_flag = BTRFS_BLOCK_GROUP_RAID6,
153 .mindev_error = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET,
158 * Convert block group flags (BTRFS_BLOCK_GROUP_*) to btrfs_raid_types, which
159 * can be used as index to access btrfs_raid_array[].
161 enum btrfs_raid_types __attribute_const__ btrfs_bg_flags_to_raid_index(u64 flags)
163 if (flags & BTRFS_BLOCK_GROUP_RAID10)
164 return BTRFS_RAID_RAID10;
165 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
166 return BTRFS_RAID_RAID1;
167 else if (flags & BTRFS_BLOCK_GROUP_RAID1C3)
168 return BTRFS_RAID_RAID1C3;
169 else if (flags & BTRFS_BLOCK_GROUP_RAID1C4)
170 return BTRFS_RAID_RAID1C4;
171 else if (flags & BTRFS_BLOCK_GROUP_DUP)
172 return BTRFS_RAID_DUP;
173 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
174 return BTRFS_RAID_RAID0;
175 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
176 return BTRFS_RAID_RAID5;
177 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
178 return BTRFS_RAID_RAID6;
180 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
183 const char *btrfs_bg_type_to_raid_name(u64 flags)
185 const int index = btrfs_bg_flags_to_raid_index(flags);
187 if (index >= BTRFS_NR_RAID_TYPES)
190 return btrfs_raid_array[index].raid_name;
194 * Fill @buf with textual description of @bg_flags, no more than @size_buf
195 * bytes including terminating null byte.
197 void btrfs_describe_block_groups(u64 bg_flags, char *buf, u32 size_buf)
202 u64 flags = bg_flags;
203 u32 size_bp = size_buf;
210 #define DESCRIBE_FLAG(flag, desc) \
212 if (flags & (flag)) { \
213 ret = snprintf(bp, size_bp, "%s|", (desc)); \
214 if (ret < 0 || ret >= size_bp) \
222 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_DATA, "data");
223 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_SYSTEM, "system");
224 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_METADATA, "metadata");
226 DESCRIBE_FLAG(BTRFS_AVAIL_ALLOC_BIT_SINGLE, "single");
227 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
228 DESCRIBE_FLAG(btrfs_raid_array[i].bg_flag,
229 btrfs_raid_array[i].raid_name);
233 ret = snprintf(bp, size_bp, "0x%llx|", flags);
237 if (size_bp < size_buf)
238 buf[size_buf - size_bp - 1] = '\0'; /* remove last | */
241 * The text is trimmed, it's up to the caller to provide sufficiently
247 static int init_first_rw_device(struct btrfs_trans_handle *trans);
248 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info);
249 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
250 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
251 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
252 enum btrfs_map_op op,
253 u64 logical, u64 *length,
254 struct btrfs_bio **bbio_ret,
255 int mirror_num, int need_raid_map);
261 * There are several mutexes that protect manipulation of devices and low-level
262 * structures like chunks but not block groups, extents or files
264 * uuid_mutex (global lock)
265 * ------------------------
266 * protects the fs_uuids list that tracks all per-fs fs_devices, resulting from
267 * the SCAN_DEV ioctl registration or from mount either implicitly (the first
268 * device) or requested by the device= mount option
270 * the mutex can be very coarse and can cover long-running operations
272 * protects: updates to fs_devices counters like missing devices, rw devices,
273 * seeding, structure cloning, opening/closing devices at mount/umount time
275 * global::fs_devs - add, remove, updates to the global list
277 * does not protect: manipulation of the fs_devices::devices list in general
278 * but in mount context it could be used to exclude list modifications by eg.
281 * btrfs_device::name - renames (write side), read is RCU
283 * fs_devices::device_list_mutex (per-fs, with RCU)
284 * ------------------------------------------------
285 * protects updates to fs_devices::devices, ie. adding and deleting
287 * simple list traversal with read-only actions can be done with RCU protection
289 * may be used to exclude some operations from running concurrently without any
290 * modifications to the list (see write_all_supers)
292 * Is not required at mount and close times, because our device list is
293 * protected by the uuid_mutex at that point.
297 * protects balance structures (status, state) and context accessed from
298 * several places (internally, ioctl)
302 * protects chunks, adding or removing during allocation, trim or when a new
303 * device is added/removed. Additionally it also protects post_commit_list of
304 * individual devices, since they can be added to the transaction's
305 * post_commit_list only with chunk_mutex held.
309 * a big lock that is held by the cleaner thread and prevents running subvolume
310 * cleaning together with relocation or delayed iputs
322 * Exclusive operations
323 * ====================
325 * Maintains the exclusivity of the following operations that apply to the
326 * whole filesystem and cannot run in parallel.
331 * - Device replace (*)
334 * The device operations (as above) can be in one of the following states:
340 * Only device operations marked with (*) can go into the Paused state for the
343 * - ioctl (only Balance can be Paused through ioctl)
344 * - filesystem remounted as read-only
345 * - filesystem unmounted and mounted as read-only
346 * - system power-cycle and filesystem mounted as read-only
347 * - filesystem or device errors leading to forced read-only
349 * The status of exclusive operation is set and cleared atomically.
350 * During the course of Paused state, fs_info::exclusive_operation remains set.
351 * A device operation in Paused or Running state can be canceled or resumed
352 * either by ioctl (Balance only) or when remounted as read-write.
353 * The exclusive status is cleared when the device operation is canceled or
357 DEFINE_MUTEX(uuid_mutex);
358 static LIST_HEAD(fs_uuids);
359 struct list_head * __attribute_const__ btrfs_get_fs_uuids(void)
365 * alloc_fs_devices - allocate struct btrfs_fs_devices
366 * @fsid: if not NULL, copy the UUID to fs_devices::fsid
367 * @metadata_fsid: if not NULL, copy the UUID to fs_devices::metadata_fsid
369 * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR().
370 * The returned struct is not linked onto any lists and can be destroyed with
371 * kfree() right away.
373 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid,
374 const u8 *metadata_fsid)
376 struct btrfs_fs_devices *fs_devs;
378 fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL);
380 return ERR_PTR(-ENOMEM);
382 mutex_init(&fs_devs->device_list_mutex);
384 INIT_LIST_HEAD(&fs_devs->devices);
385 INIT_LIST_HEAD(&fs_devs->alloc_list);
386 INIT_LIST_HEAD(&fs_devs->fs_list);
387 INIT_LIST_HEAD(&fs_devs->seed_list);
389 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
392 memcpy(fs_devs->metadata_uuid, metadata_fsid, BTRFS_FSID_SIZE);
394 memcpy(fs_devs->metadata_uuid, fsid, BTRFS_FSID_SIZE);
399 void btrfs_free_device(struct btrfs_device *device)
401 WARN_ON(!list_empty(&device->post_commit_list));
402 rcu_string_free(device->name);
403 extent_io_tree_release(&device->alloc_state);
404 bio_put(device->flush_bio);
405 btrfs_destroy_dev_zone_info(device);
409 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
411 struct btrfs_device *device;
412 WARN_ON(fs_devices->opened);
413 while (!list_empty(&fs_devices->devices)) {
414 device = list_entry(fs_devices->devices.next,
415 struct btrfs_device, dev_list);
416 list_del(&device->dev_list);
417 btrfs_free_device(device);
422 void __exit btrfs_cleanup_fs_uuids(void)
424 struct btrfs_fs_devices *fs_devices;
426 while (!list_empty(&fs_uuids)) {
427 fs_devices = list_entry(fs_uuids.next,
428 struct btrfs_fs_devices, fs_list);
429 list_del(&fs_devices->fs_list);
430 free_fs_devices(fs_devices);
434 static noinline struct btrfs_fs_devices *find_fsid(
435 const u8 *fsid, const u8 *metadata_fsid)
437 struct btrfs_fs_devices *fs_devices;
441 /* Handle non-split brain cases */
442 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
444 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0
445 && memcmp(metadata_fsid, fs_devices->metadata_uuid,
446 BTRFS_FSID_SIZE) == 0)
449 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
456 static struct btrfs_fs_devices *find_fsid_with_metadata_uuid(
457 struct btrfs_super_block *disk_super)
460 struct btrfs_fs_devices *fs_devices;
463 * Handle scanned device having completed its fsid change but
464 * belonging to a fs_devices that was created by first scanning
465 * a device which didn't have its fsid/metadata_uuid changed
466 * at all and the CHANGING_FSID_V2 flag set.
468 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
469 if (fs_devices->fsid_change &&
470 memcmp(disk_super->metadata_uuid, fs_devices->fsid,
471 BTRFS_FSID_SIZE) == 0 &&
472 memcmp(fs_devices->fsid, fs_devices->metadata_uuid,
473 BTRFS_FSID_SIZE) == 0) {
478 * Handle scanned device having completed its fsid change but
479 * belonging to a fs_devices that was created by a device that
480 * has an outdated pair of fsid/metadata_uuid and
481 * CHANGING_FSID_V2 flag set.
483 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
484 if (fs_devices->fsid_change &&
485 memcmp(fs_devices->metadata_uuid,
486 fs_devices->fsid, BTRFS_FSID_SIZE) != 0 &&
487 memcmp(disk_super->metadata_uuid, fs_devices->metadata_uuid,
488 BTRFS_FSID_SIZE) == 0) {
493 return find_fsid(disk_super->fsid, disk_super->metadata_uuid);
498 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
499 int flush, struct block_device **bdev,
500 struct btrfs_super_block **disk_super)
504 *bdev = blkdev_get_by_path(device_path, flags, holder);
507 ret = PTR_ERR(*bdev);
512 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
513 ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE);
515 blkdev_put(*bdev, flags);
518 invalidate_bdev(*bdev);
519 *disk_super = btrfs_read_dev_super(*bdev);
520 if (IS_ERR(*disk_super)) {
521 ret = PTR_ERR(*disk_super);
522 blkdev_put(*bdev, flags);
533 static bool device_path_matched(const char *path, struct btrfs_device *device)
538 found = strcmp(rcu_str_deref(device->name), path);
545 * Search and remove all stale (devices which are not mounted) devices.
546 * When both inputs are NULL, it will search and release all stale devices.
547 * path: Optional. When provided will it release all unmounted devices
548 * matching this path only.
549 * skip_dev: Optional. Will skip this device when searching for the stale
551 * Return: 0 for success or if @path is NULL.
552 * -EBUSY if @path is a mounted device.
553 * -ENOENT if @path does not match any device in the list.
555 static int btrfs_free_stale_devices(const char *path,
556 struct btrfs_device *skip_device)
558 struct btrfs_fs_devices *fs_devices, *tmp_fs_devices;
559 struct btrfs_device *device, *tmp_device;
562 lockdep_assert_held(&uuid_mutex);
567 list_for_each_entry_safe(fs_devices, tmp_fs_devices, &fs_uuids, fs_list) {
569 mutex_lock(&fs_devices->device_list_mutex);
570 list_for_each_entry_safe(device, tmp_device,
571 &fs_devices->devices, dev_list) {
572 if (skip_device && skip_device == device)
574 if (path && !device->name)
576 if (path && !device_path_matched(path, device))
578 if (fs_devices->opened) {
579 /* for an already deleted device return 0 */
580 if (path && ret != 0)
585 /* delete the stale device */
586 fs_devices->num_devices--;
587 list_del(&device->dev_list);
588 btrfs_free_device(device);
592 mutex_unlock(&fs_devices->device_list_mutex);
594 if (fs_devices->num_devices == 0) {
595 btrfs_sysfs_remove_fsid(fs_devices);
596 list_del(&fs_devices->fs_list);
597 free_fs_devices(fs_devices);
605 * This is only used on mount, and we are protected from competing things
606 * messing with our fs_devices by the uuid_mutex, thus we do not need the
607 * fs_devices->device_list_mutex here.
609 static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices,
610 struct btrfs_device *device, fmode_t flags,
613 struct request_queue *q;
614 struct block_device *bdev;
615 struct btrfs_super_block *disk_super;
624 ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
629 devid = btrfs_stack_device_id(&disk_super->dev_item);
630 if (devid != device->devid)
631 goto error_free_page;
633 if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE))
634 goto error_free_page;
636 device->generation = btrfs_super_generation(disk_super);
638 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
639 if (btrfs_super_incompat_flags(disk_super) &
640 BTRFS_FEATURE_INCOMPAT_METADATA_UUID) {
642 "BTRFS: Invalid seeding and uuid-changed device detected\n");
643 goto error_free_page;
646 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
647 fs_devices->seeding = true;
649 if (bdev_read_only(bdev))
650 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
652 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
655 q = bdev_get_queue(bdev);
656 if (!blk_queue_nonrot(q))
657 fs_devices->rotating = true;
660 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
661 device->mode = flags;
663 fs_devices->open_devices++;
664 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
665 device->devid != BTRFS_DEV_REPLACE_DEVID) {
666 fs_devices->rw_devices++;
667 list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list);
669 btrfs_release_disk_super(disk_super);
674 btrfs_release_disk_super(disk_super);
675 blkdev_put(bdev, flags);
681 * Handle scanned device having its CHANGING_FSID_V2 flag set and the fs_devices
682 * being created with a disk that has already completed its fsid change. Such
683 * disk can belong to an fs which has its FSID changed or to one which doesn't.
684 * Handle both cases here.
686 static struct btrfs_fs_devices *find_fsid_inprogress(
687 struct btrfs_super_block *disk_super)
689 struct btrfs_fs_devices *fs_devices;
691 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
692 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid,
693 BTRFS_FSID_SIZE) != 0 &&
694 memcmp(fs_devices->metadata_uuid, disk_super->fsid,
695 BTRFS_FSID_SIZE) == 0 && !fs_devices->fsid_change) {
700 return find_fsid(disk_super->fsid, NULL);
704 static struct btrfs_fs_devices *find_fsid_changed(
705 struct btrfs_super_block *disk_super)
707 struct btrfs_fs_devices *fs_devices;
710 * Handles the case where scanned device is part of an fs that had
711 * multiple successful changes of FSID but currently device didn't
712 * observe it. Meaning our fsid will be different than theirs. We need
713 * to handle two subcases :
714 * 1 - The fs still continues to have different METADATA/FSID uuids.
715 * 2 - The fs is switched back to its original FSID (METADATA/FSID
718 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
720 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid,
721 BTRFS_FSID_SIZE) != 0 &&
722 memcmp(fs_devices->metadata_uuid, disk_super->metadata_uuid,
723 BTRFS_FSID_SIZE) == 0 &&
724 memcmp(fs_devices->fsid, disk_super->fsid,
725 BTRFS_FSID_SIZE) != 0)
728 /* Unchanged UUIDs */
729 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid,
730 BTRFS_FSID_SIZE) == 0 &&
731 memcmp(fs_devices->fsid, disk_super->metadata_uuid,
732 BTRFS_FSID_SIZE) == 0)
739 static struct btrfs_fs_devices *find_fsid_reverted_metadata(
740 struct btrfs_super_block *disk_super)
742 struct btrfs_fs_devices *fs_devices;
745 * Handle the case where the scanned device is part of an fs whose last
746 * metadata UUID change reverted it to the original FSID. At the same
747 * time * fs_devices was first created by another constitutent device
748 * which didn't fully observe the operation. This results in an
749 * btrfs_fs_devices created with metadata/fsid different AND
750 * btrfs_fs_devices::fsid_change set AND the metadata_uuid of the
751 * fs_devices equal to the FSID of the disk.
753 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
754 if (memcmp(fs_devices->fsid, fs_devices->metadata_uuid,
755 BTRFS_FSID_SIZE) != 0 &&
756 memcmp(fs_devices->metadata_uuid, disk_super->fsid,
757 BTRFS_FSID_SIZE) == 0 &&
758 fs_devices->fsid_change)
765 * Add new device to list of registered devices
768 * device pointer which was just added or updated when successful
769 * error pointer when failed
771 static noinline struct btrfs_device *device_list_add(const char *path,
772 struct btrfs_super_block *disk_super,
773 bool *new_device_added)
775 struct btrfs_device *device;
776 struct btrfs_fs_devices *fs_devices = NULL;
777 struct rcu_string *name;
778 u64 found_transid = btrfs_super_generation(disk_super);
779 u64 devid = btrfs_stack_device_id(&disk_super->dev_item);
780 bool has_metadata_uuid = (btrfs_super_incompat_flags(disk_super) &
781 BTRFS_FEATURE_INCOMPAT_METADATA_UUID);
782 bool fsid_change_in_progress = (btrfs_super_flags(disk_super) &
783 BTRFS_SUPER_FLAG_CHANGING_FSID_V2);
785 if (fsid_change_in_progress) {
786 if (!has_metadata_uuid)
787 fs_devices = find_fsid_inprogress(disk_super);
789 fs_devices = find_fsid_changed(disk_super);
790 } else if (has_metadata_uuid) {
791 fs_devices = find_fsid_with_metadata_uuid(disk_super);
793 fs_devices = find_fsid_reverted_metadata(disk_super);
795 fs_devices = find_fsid(disk_super->fsid, NULL);
800 if (has_metadata_uuid)
801 fs_devices = alloc_fs_devices(disk_super->fsid,
802 disk_super->metadata_uuid);
804 fs_devices = alloc_fs_devices(disk_super->fsid, NULL);
806 if (IS_ERR(fs_devices))
807 return ERR_CAST(fs_devices);
809 fs_devices->fsid_change = fsid_change_in_progress;
811 mutex_lock(&fs_devices->device_list_mutex);
812 list_add(&fs_devices->fs_list, &fs_uuids);
816 mutex_lock(&fs_devices->device_list_mutex);
817 device = btrfs_find_device(fs_devices, devid,
818 disk_super->dev_item.uuid, NULL);
821 * If this disk has been pulled into an fs devices created by
822 * a device which had the CHANGING_FSID_V2 flag then replace the
823 * metadata_uuid/fsid values of the fs_devices.
825 if (fs_devices->fsid_change &&
826 found_transid > fs_devices->latest_generation) {
827 memcpy(fs_devices->fsid, disk_super->fsid,
830 if (has_metadata_uuid)
831 memcpy(fs_devices->metadata_uuid,
832 disk_super->metadata_uuid,
835 memcpy(fs_devices->metadata_uuid,
836 disk_super->fsid, BTRFS_FSID_SIZE);
838 fs_devices->fsid_change = false;
843 if (fs_devices->opened) {
844 mutex_unlock(&fs_devices->device_list_mutex);
845 return ERR_PTR(-EBUSY);
848 device = btrfs_alloc_device(NULL, &devid,
849 disk_super->dev_item.uuid);
850 if (IS_ERR(device)) {
851 mutex_unlock(&fs_devices->device_list_mutex);
852 /* we can safely leave the fs_devices entry around */
856 name = rcu_string_strdup(path, GFP_NOFS);
858 btrfs_free_device(device);
859 mutex_unlock(&fs_devices->device_list_mutex);
860 return ERR_PTR(-ENOMEM);
862 rcu_assign_pointer(device->name, name);
864 list_add_rcu(&device->dev_list, &fs_devices->devices);
865 fs_devices->num_devices++;
867 device->fs_devices = fs_devices;
868 *new_device_added = true;
870 if (disk_super->label[0])
872 "BTRFS: device label %s devid %llu transid %llu %s scanned by %s (%d)\n",
873 disk_super->label, devid, found_transid, path,
874 current->comm, task_pid_nr(current));
877 "BTRFS: device fsid %pU devid %llu transid %llu %s scanned by %s (%d)\n",
878 disk_super->fsid, devid, found_transid, path,
879 current->comm, task_pid_nr(current));
881 } else if (!device->name || strcmp(device->name->str, path)) {
883 * When FS is already mounted.
884 * 1. If you are here and if the device->name is NULL that
885 * means this device was missing at time of FS mount.
886 * 2. If you are here and if the device->name is different
887 * from 'path' that means either
888 * a. The same device disappeared and reappeared with
890 * b. The missing-disk-which-was-replaced, has
893 * We must allow 1 and 2a above. But 2b would be a spurious
896 * Further in case of 1 and 2a above, the disk at 'path'
897 * would have missed some transaction when it was away and
898 * in case of 2a the stale bdev has to be updated as well.
899 * 2b must not be allowed at all time.
903 * For now, we do allow update to btrfs_fs_device through the
904 * btrfs dev scan cli after FS has been mounted. We're still
905 * tracking a problem where systems fail mount by subvolume id
906 * when we reject replacement on a mounted FS.
908 if (!fs_devices->opened && found_transid < device->generation) {
910 * That is if the FS is _not_ mounted and if you
911 * are here, that means there is more than one
912 * disk with same uuid and devid.We keep the one
913 * with larger generation number or the last-in if
914 * generation are equal.
916 mutex_unlock(&fs_devices->device_list_mutex);
917 return ERR_PTR(-EEXIST);
921 * We are going to replace the device path for a given devid,
922 * make sure it's the same device if the device is mounted
928 error = lookup_bdev(path, &path_dev);
930 mutex_unlock(&fs_devices->device_list_mutex);
931 return ERR_PTR(error);
934 if (device->bdev->bd_dev != path_dev) {
935 mutex_unlock(&fs_devices->device_list_mutex);
937 * device->fs_info may not be reliable here, so
938 * pass in a NULL instead. This avoids a
939 * possible use-after-free when the fs_info and
940 * fs_info->sb are already torn down.
942 btrfs_warn_in_rcu(NULL,
943 "duplicate device %s devid %llu generation %llu scanned by %s (%d)",
944 path, devid, found_transid,
946 task_pid_nr(current));
947 return ERR_PTR(-EEXIST);
949 btrfs_info_in_rcu(device->fs_info,
950 "devid %llu device path %s changed to %s scanned by %s (%d)",
951 devid, rcu_str_deref(device->name),
953 task_pid_nr(current));
956 name = rcu_string_strdup(path, GFP_NOFS);
958 mutex_unlock(&fs_devices->device_list_mutex);
959 return ERR_PTR(-ENOMEM);
961 rcu_string_free(device->name);
962 rcu_assign_pointer(device->name, name);
963 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
964 fs_devices->missing_devices--;
965 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
970 * Unmount does not free the btrfs_device struct but would zero
971 * generation along with most of the other members. So just update
972 * it back. We need it to pick the disk with largest generation
975 if (!fs_devices->opened) {
976 device->generation = found_transid;
977 fs_devices->latest_generation = max_t(u64, found_transid,
978 fs_devices->latest_generation);
981 fs_devices->total_devices = btrfs_super_num_devices(disk_super);
983 mutex_unlock(&fs_devices->device_list_mutex);
987 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
989 struct btrfs_fs_devices *fs_devices;
990 struct btrfs_device *device;
991 struct btrfs_device *orig_dev;
994 lockdep_assert_held(&uuid_mutex);
996 fs_devices = alloc_fs_devices(orig->fsid, NULL);
997 if (IS_ERR(fs_devices))
1000 fs_devices->total_devices = orig->total_devices;
1002 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
1003 struct rcu_string *name;
1005 device = btrfs_alloc_device(NULL, &orig_dev->devid,
1007 if (IS_ERR(device)) {
1008 ret = PTR_ERR(device);
1013 * This is ok to do without rcu read locked because we hold the
1014 * uuid mutex so nothing we touch in here is going to disappear.
1016 if (orig_dev->name) {
1017 name = rcu_string_strdup(orig_dev->name->str,
1020 btrfs_free_device(device);
1024 rcu_assign_pointer(device->name, name);
1027 list_add(&device->dev_list, &fs_devices->devices);
1028 device->fs_devices = fs_devices;
1029 fs_devices->num_devices++;
1033 free_fs_devices(fs_devices);
1034 return ERR_PTR(ret);
1037 static void __btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices,
1038 struct btrfs_device **latest_dev)
1040 struct btrfs_device *device, *next;
1042 /* This is the initialized path, it is safe to release the devices. */
1043 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
1044 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)) {
1045 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
1046 &device->dev_state) &&
1047 !test_bit(BTRFS_DEV_STATE_MISSING,
1048 &device->dev_state) &&
1050 device->generation > (*latest_dev)->generation)) {
1051 *latest_dev = device;
1057 * We have already validated the presence of BTRFS_DEV_REPLACE_DEVID,
1058 * in btrfs_init_dev_replace() so just continue.
1060 if (device->devid == BTRFS_DEV_REPLACE_DEVID)
1064 blkdev_put(device->bdev, device->mode);
1065 device->bdev = NULL;
1066 fs_devices->open_devices--;
1068 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1069 list_del_init(&device->dev_alloc_list);
1070 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
1071 fs_devices->rw_devices--;
1073 list_del_init(&device->dev_list);
1074 fs_devices->num_devices--;
1075 btrfs_free_device(device);
1081 * After we have read the system tree and know devids belonging to this
1082 * filesystem, remove the device which does not belong there.
1084 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices)
1086 struct btrfs_device *latest_dev = NULL;
1087 struct btrfs_fs_devices *seed_dev;
1089 mutex_lock(&uuid_mutex);
1090 __btrfs_free_extra_devids(fs_devices, &latest_dev);
1092 list_for_each_entry(seed_dev, &fs_devices->seed_list, seed_list)
1093 __btrfs_free_extra_devids(seed_dev, &latest_dev);
1095 fs_devices->latest_dev = latest_dev;
1097 mutex_unlock(&uuid_mutex);
1100 static void btrfs_close_bdev(struct btrfs_device *device)
1105 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1106 sync_blockdev(device->bdev);
1107 invalidate_bdev(device->bdev);
1110 blkdev_put(device->bdev, device->mode);
1113 static void btrfs_close_one_device(struct btrfs_device *device)
1115 struct btrfs_fs_devices *fs_devices = device->fs_devices;
1117 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
1118 device->devid != BTRFS_DEV_REPLACE_DEVID) {
1119 list_del_init(&device->dev_alloc_list);
1120 fs_devices->rw_devices--;
1123 if (device->devid == BTRFS_DEV_REPLACE_DEVID)
1124 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
1126 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
1127 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
1128 fs_devices->missing_devices--;
1131 btrfs_close_bdev(device);
1133 fs_devices->open_devices--;
1134 device->bdev = NULL;
1136 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
1137 btrfs_destroy_dev_zone_info(device);
1139 device->fs_info = NULL;
1140 atomic_set(&device->dev_stats_ccnt, 0);
1141 extent_io_tree_release(&device->alloc_state);
1144 * Reset the flush error record. We might have a transient flush error
1145 * in this mount, and if so we aborted the current transaction and set
1146 * the fs to an error state, guaranteeing no super blocks can be further
1147 * committed. However that error might be transient and if we unmount the
1148 * filesystem and mount it again, we should allow the mount to succeed
1149 * (btrfs_check_rw_degradable() should not fail) - if after mounting the
1150 * filesystem again we still get flush errors, then we will again abort
1151 * any transaction and set the error state, guaranteeing no commits of
1152 * unsafe super blocks.
1154 device->last_flush_error = 0;
1156 /* Verify the device is back in a pristine state */
1157 ASSERT(!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state));
1158 ASSERT(!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state));
1159 ASSERT(list_empty(&device->dev_alloc_list));
1160 ASSERT(list_empty(&device->post_commit_list));
1161 ASSERT(atomic_read(&device->reada_in_flight) == 0);
1164 static void close_fs_devices(struct btrfs_fs_devices *fs_devices)
1166 struct btrfs_device *device, *tmp;
1168 lockdep_assert_held(&uuid_mutex);
1170 if (--fs_devices->opened > 0)
1173 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list)
1174 btrfs_close_one_device(device);
1176 WARN_ON(fs_devices->open_devices);
1177 WARN_ON(fs_devices->rw_devices);
1178 fs_devices->opened = 0;
1179 fs_devices->seeding = false;
1180 fs_devices->fs_info = NULL;
1183 void btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
1186 struct btrfs_fs_devices *tmp;
1188 mutex_lock(&uuid_mutex);
1189 close_fs_devices(fs_devices);
1190 if (!fs_devices->opened)
1191 list_splice_init(&fs_devices->seed_list, &list);
1193 list_for_each_entry_safe(fs_devices, tmp, &list, seed_list) {
1194 close_fs_devices(fs_devices);
1195 list_del(&fs_devices->seed_list);
1196 free_fs_devices(fs_devices);
1198 mutex_unlock(&uuid_mutex);
1201 static int open_fs_devices(struct btrfs_fs_devices *fs_devices,
1202 fmode_t flags, void *holder)
1204 struct btrfs_device *device;
1205 struct btrfs_device *latest_dev = NULL;
1206 struct btrfs_device *tmp_device;
1208 flags |= FMODE_EXCL;
1210 list_for_each_entry_safe(device, tmp_device, &fs_devices->devices,
1214 ret = btrfs_open_one_device(fs_devices, device, flags, holder);
1216 (!latest_dev || device->generation > latest_dev->generation)) {
1217 latest_dev = device;
1218 } else if (ret == -ENODATA) {
1219 fs_devices->num_devices--;
1220 list_del(&device->dev_list);
1221 btrfs_free_device(device);
1224 if (fs_devices->open_devices == 0)
1227 fs_devices->opened = 1;
1228 fs_devices->latest_dev = latest_dev;
1229 fs_devices->total_rw_bytes = 0;
1230 fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_REGULAR;
1231 fs_devices->read_policy = BTRFS_READ_POLICY_PID;
1236 static int devid_cmp(void *priv, const struct list_head *a,
1237 const struct list_head *b)
1239 const struct btrfs_device *dev1, *dev2;
1241 dev1 = list_entry(a, struct btrfs_device, dev_list);
1242 dev2 = list_entry(b, struct btrfs_device, dev_list);
1244 if (dev1->devid < dev2->devid)
1246 else if (dev1->devid > dev2->devid)
1251 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1252 fmode_t flags, void *holder)
1256 lockdep_assert_held(&uuid_mutex);
1258 * The device_list_mutex cannot be taken here in case opening the
1259 * underlying device takes further locks like open_mutex.
1261 * We also don't need the lock here as this is called during mount and
1262 * exclusion is provided by uuid_mutex
1265 if (fs_devices->opened) {
1266 fs_devices->opened++;
1269 list_sort(NULL, &fs_devices->devices, devid_cmp);
1270 ret = open_fs_devices(fs_devices, flags, holder);
1276 void btrfs_release_disk_super(struct btrfs_super_block *super)
1278 struct page *page = virt_to_page(super);
1283 static struct btrfs_super_block *btrfs_read_disk_super(struct block_device *bdev,
1284 u64 bytenr, u64 bytenr_orig)
1286 struct btrfs_super_block *disk_super;
1291 /* make sure our super fits in the device */
1292 if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1293 return ERR_PTR(-EINVAL);
1295 /* make sure our super fits in the page */
1296 if (sizeof(*disk_super) > PAGE_SIZE)
1297 return ERR_PTR(-EINVAL);
1299 /* make sure our super doesn't straddle pages on disk */
1300 index = bytenr >> PAGE_SHIFT;
1301 if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_SHIFT != index)
1302 return ERR_PTR(-EINVAL);
1304 /* pull in the page with our super */
1305 page = read_cache_page_gfp(bdev->bd_inode->i_mapping, index, GFP_KERNEL);
1308 return ERR_CAST(page);
1310 p = page_address(page);
1312 /* align our pointer to the offset of the super block */
1313 disk_super = p + offset_in_page(bytenr);
1315 if (btrfs_super_bytenr(disk_super) != bytenr_orig ||
1316 btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
1317 btrfs_release_disk_super(p);
1318 return ERR_PTR(-EINVAL);
1321 if (disk_super->label[0] && disk_super->label[BTRFS_LABEL_SIZE - 1])
1322 disk_super->label[BTRFS_LABEL_SIZE - 1] = 0;
1327 int btrfs_forget_devices(const char *path)
1331 mutex_lock(&uuid_mutex);
1332 ret = btrfs_free_stale_devices(strlen(path) ? path : NULL, NULL);
1333 mutex_unlock(&uuid_mutex);
1339 * Look for a btrfs signature on a device. This may be called out of the mount path
1340 * and we are not allowed to call set_blocksize during the scan. The superblock
1341 * is read via pagecache
1343 struct btrfs_device *btrfs_scan_one_device(const char *path, fmode_t flags,
1346 struct btrfs_super_block *disk_super;
1347 bool new_device_added = false;
1348 struct btrfs_device *device = NULL;
1349 struct block_device *bdev;
1350 u64 bytenr, bytenr_orig;
1353 lockdep_assert_held(&uuid_mutex);
1356 * we would like to check all the supers, but that would make
1357 * a btrfs mount succeed after a mkfs from a different FS.
1358 * So, we need to add a special mount option to scan for
1359 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1361 flags |= FMODE_EXCL;
1363 bdev = blkdev_get_by_path(path, flags, holder);
1365 return ERR_CAST(bdev);
1367 bytenr_orig = btrfs_sb_offset(0);
1368 ret = btrfs_sb_log_location_bdev(bdev, 0, READ, &bytenr);
1370 return ERR_PTR(ret);
1372 disk_super = btrfs_read_disk_super(bdev, bytenr, bytenr_orig);
1373 if (IS_ERR(disk_super)) {
1374 device = ERR_CAST(disk_super);
1375 goto error_bdev_put;
1378 device = device_list_add(path, disk_super, &new_device_added);
1379 if (!IS_ERR(device)) {
1380 if (new_device_added)
1381 btrfs_free_stale_devices(path, device);
1384 btrfs_release_disk_super(disk_super);
1387 blkdev_put(bdev, flags);
1393 * Try to find a chunk that intersects [start, start + len] range and when one
1394 * such is found, record the end of it in *start
1396 static bool contains_pending_extent(struct btrfs_device *device, u64 *start,
1399 u64 physical_start, physical_end;
1401 lockdep_assert_held(&device->fs_info->chunk_mutex);
1403 if (!find_first_extent_bit(&device->alloc_state, *start,
1404 &physical_start, &physical_end,
1405 CHUNK_ALLOCATED, NULL)) {
1407 if (in_range(physical_start, *start, len) ||
1408 in_range(*start, physical_start,
1409 physical_end - physical_start)) {
1410 *start = physical_end + 1;
1417 static u64 dev_extent_search_start(struct btrfs_device *device, u64 start)
1419 switch (device->fs_devices->chunk_alloc_policy) {
1420 case BTRFS_CHUNK_ALLOC_REGULAR:
1422 * We don't want to overwrite the superblock on the drive nor
1423 * any area used by the boot loader (grub for example), so we
1424 * make sure to start at an offset of at least 1MB.
1426 return max_t(u64, start, SZ_1M);
1427 case BTRFS_CHUNK_ALLOC_ZONED:
1429 * We don't care about the starting region like regular
1430 * allocator, because we anyway use/reserve the first two zones
1431 * for superblock logging.
1433 return ALIGN(start, device->zone_info->zone_size);
1439 static bool dev_extent_hole_check_zoned(struct btrfs_device *device,
1440 u64 *hole_start, u64 *hole_size,
1443 u64 zone_size = device->zone_info->zone_size;
1446 bool changed = false;
1448 ASSERT(IS_ALIGNED(*hole_start, zone_size));
1450 while (*hole_size > 0) {
1451 pos = btrfs_find_allocatable_zones(device, *hole_start,
1452 *hole_start + *hole_size,
1454 if (pos != *hole_start) {
1455 *hole_size = *hole_start + *hole_size - pos;
1458 if (*hole_size < num_bytes)
1462 ret = btrfs_ensure_empty_zones(device, pos, num_bytes);
1464 /* Range is ensured to be empty */
1468 /* Given hole range was invalid (outside of device) */
1469 if (ret == -ERANGE) {
1470 *hole_start += *hole_size;
1475 *hole_start += zone_size;
1476 *hole_size -= zone_size;
1484 * dev_extent_hole_check - check if specified hole is suitable for allocation
1485 * @device: the device which we have the hole
1486 * @hole_start: starting position of the hole
1487 * @hole_size: the size of the hole
1488 * @num_bytes: the size of the free space that we need
1490 * This function may modify @hole_start and @hole_size to reflect the suitable
1491 * position for allocation. Returns 1 if hole position is updated, 0 otherwise.
1493 static bool dev_extent_hole_check(struct btrfs_device *device, u64 *hole_start,
1494 u64 *hole_size, u64 num_bytes)
1496 bool changed = false;
1497 u64 hole_end = *hole_start + *hole_size;
1501 * Check before we set max_hole_start, otherwise we could end up
1502 * sending back this offset anyway.
1504 if (contains_pending_extent(device, hole_start, *hole_size)) {
1505 if (hole_end >= *hole_start)
1506 *hole_size = hole_end - *hole_start;
1512 switch (device->fs_devices->chunk_alloc_policy) {
1513 case BTRFS_CHUNK_ALLOC_REGULAR:
1514 /* No extra check */
1516 case BTRFS_CHUNK_ALLOC_ZONED:
1517 if (dev_extent_hole_check_zoned(device, hole_start,
1518 hole_size, num_bytes)) {
1521 * The changed hole can contain pending extent.
1522 * Loop again to check that.
1538 * find_free_dev_extent_start - find free space in the specified device
1539 * @device: the device which we search the free space in
1540 * @num_bytes: the size of the free space that we need
1541 * @search_start: the position from which to begin the search
1542 * @start: store the start of the free space.
1543 * @len: the size of the free space. that we find, or the size
1544 * of the max free space if we don't find suitable free space
1546 * this uses a pretty simple search, the expectation is that it is
1547 * called very infrequently and that a given device has a small number
1550 * @start is used to store the start of the free space if we find. But if we
1551 * don't find suitable free space, it will be used to store the start position
1552 * of the max free space.
1554 * @len is used to store the size of the free space that we find.
1555 * But if we don't find suitable free space, it is used to store the size of
1556 * the max free space.
1558 * NOTE: This function will search *commit* root of device tree, and does extra
1559 * check to ensure dev extents are not double allocated.
1560 * This makes the function safe to allocate dev extents but may not report
1561 * correct usable device space, as device extent freed in current transaction
1562 * is not reported as available.
1564 static int find_free_dev_extent_start(struct btrfs_device *device,
1565 u64 num_bytes, u64 search_start, u64 *start,
1568 struct btrfs_fs_info *fs_info = device->fs_info;
1569 struct btrfs_root *root = fs_info->dev_root;
1570 struct btrfs_key key;
1571 struct btrfs_dev_extent *dev_extent;
1572 struct btrfs_path *path;
1577 u64 search_end = device->total_bytes;
1580 struct extent_buffer *l;
1582 search_start = dev_extent_search_start(device, search_start);
1584 WARN_ON(device->zone_info &&
1585 !IS_ALIGNED(num_bytes, device->zone_info->zone_size));
1587 path = btrfs_alloc_path();
1591 max_hole_start = search_start;
1595 if (search_start >= search_end ||
1596 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1601 path->reada = READA_FORWARD;
1602 path->search_commit_root = 1;
1603 path->skip_locking = 1;
1605 key.objectid = device->devid;
1606 key.offset = search_start;
1607 key.type = BTRFS_DEV_EXTENT_KEY;
1609 ret = btrfs_search_backwards(root, &key, path);
1615 slot = path->slots[0];
1616 if (slot >= btrfs_header_nritems(l)) {
1617 ret = btrfs_next_leaf(root, path);
1625 btrfs_item_key_to_cpu(l, &key, slot);
1627 if (key.objectid < device->devid)
1630 if (key.objectid > device->devid)
1633 if (key.type != BTRFS_DEV_EXTENT_KEY)
1636 if (key.offset > search_start) {
1637 hole_size = key.offset - search_start;
1638 dev_extent_hole_check(device, &search_start, &hole_size,
1641 if (hole_size > max_hole_size) {
1642 max_hole_start = search_start;
1643 max_hole_size = hole_size;
1647 * If this free space is greater than which we need,
1648 * it must be the max free space that we have found
1649 * until now, so max_hole_start must point to the start
1650 * of this free space and the length of this free space
1651 * is stored in max_hole_size. Thus, we return
1652 * max_hole_start and max_hole_size and go back to the
1655 if (hole_size >= num_bytes) {
1661 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1662 extent_end = key.offset + btrfs_dev_extent_length(l,
1664 if (extent_end > search_start)
1665 search_start = extent_end;
1672 * At this point, search_start should be the end of
1673 * allocated dev extents, and when shrinking the device,
1674 * search_end may be smaller than search_start.
1676 if (search_end > search_start) {
1677 hole_size = search_end - search_start;
1678 if (dev_extent_hole_check(device, &search_start, &hole_size,
1680 btrfs_release_path(path);
1684 if (hole_size > max_hole_size) {
1685 max_hole_start = search_start;
1686 max_hole_size = hole_size;
1691 if (max_hole_size < num_bytes)
1697 btrfs_free_path(path);
1698 *start = max_hole_start;
1700 *len = max_hole_size;
1704 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
1705 u64 *start, u64 *len)
1707 /* FIXME use last free of some kind */
1708 return find_free_dev_extent_start(device, num_bytes, 0, start, len);
1711 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1712 struct btrfs_device *device,
1713 u64 start, u64 *dev_extent_len)
1715 struct btrfs_fs_info *fs_info = device->fs_info;
1716 struct btrfs_root *root = fs_info->dev_root;
1718 struct btrfs_path *path;
1719 struct btrfs_key key;
1720 struct btrfs_key found_key;
1721 struct extent_buffer *leaf = NULL;
1722 struct btrfs_dev_extent *extent = NULL;
1724 path = btrfs_alloc_path();
1728 key.objectid = device->devid;
1730 key.type = BTRFS_DEV_EXTENT_KEY;
1732 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1734 ret = btrfs_previous_item(root, path, key.objectid,
1735 BTRFS_DEV_EXTENT_KEY);
1738 leaf = path->nodes[0];
1739 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1740 extent = btrfs_item_ptr(leaf, path->slots[0],
1741 struct btrfs_dev_extent);
1742 BUG_ON(found_key.offset > start || found_key.offset +
1743 btrfs_dev_extent_length(leaf, extent) < start);
1745 btrfs_release_path(path);
1747 } else if (ret == 0) {
1748 leaf = path->nodes[0];
1749 extent = btrfs_item_ptr(leaf, path->slots[0],
1750 struct btrfs_dev_extent);
1755 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1757 ret = btrfs_del_item(trans, root, path);
1759 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1761 btrfs_free_path(path);
1765 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1767 struct extent_map_tree *em_tree;
1768 struct extent_map *em;
1772 em_tree = &fs_info->mapping_tree;
1773 read_lock(&em_tree->lock);
1774 n = rb_last(&em_tree->map.rb_root);
1776 em = rb_entry(n, struct extent_map, rb_node);
1777 ret = em->start + em->len;
1779 read_unlock(&em_tree->lock);
1784 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1788 struct btrfs_key key;
1789 struct btrfs_key found_key;
1790 struct btrfs_path *path;
1792 path = btrfs_alloc_path();
1796 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1797 key.type = BTRFS_DEV_ITEM_KEY;
1798 key.offset = (u64)-1;
1800 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1806 btrfs_err(fs_info, "corrupted chunk tree devid -1 matched");
1811 ret = btrfs_previous_item(fs_info->chunk_root, path,
1812 BTRFS_DEV_ITEMS_OBJECTID,
1813 BTRFS_DEV_ITEM_KEY);
1817 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1819 *devid_ret = found_key.offset + 1;
1823 btrfs_free_path(path);
1828 * the device information is stored in the chunk root
1829 * the btrfs_device struct should be fully filled in
1831 static int btrfs_add_dev_item(struct btrfs_trans_handle *trans,
1832 struct btrfs_device *device)
1835 struct btrfs_path *path;
1836 struct btrfs_dev_item *dev_item;
1837 struct extent_buffer *leaf;
1838 struct btrfs_key key;
1841 path = btrfs_alloc_path();
1845 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1846 key.type = BTRFS_DEV_ITEM_KEY;
1847 key.offset = device->devid;
1849 ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path,
1850 &key, sizeof(*dev_item));
1854 leaf = path->nodes[0];
1855 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1857 btrfs_set_device_id(leaf, dev_item, device->devid);
1858 btrfs_set_device_generation(leaf, dev_item, 0);
1859 btrfs_set_device_type(leaf, dev_item, device->type);
1860 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1861 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1862 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1863 btrfs_set_device_total_bytes(leaf, dev_item,
1864 btrfs_device_get_disk_total_bytes(device));
1865 btrfs_set_device_bytes_used(leaf, dev_item,
1866 btrfs_device_get_bytes_used(device));
1867 btrfs_set_device_group(leaf, dev_item, 0);
1868 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1869 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1870 btrfs_set_device_start_offset(leaf, dev_item, 0);
1872 ptr = btrfs_device_uuid(dev_item);
1873 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1874 ptr = btrfs_device_fsid(dev_item);
1875 write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid,
1876 ptr, BTRFS_FSID_SIZE);
1877 btrfs_mark_buffer_dirty(leaf);
1881 btrfs_free_path(path);
1886 * Function to update ctime/mtime for a given device path.
1887 * Mainly used for ctime/mtime based probe like libblkid.
1889 * We don't care about errors here, this is just to be kind to userspace.
1891 static void update_dev_time(const char *device_path)
1894 struct timespec64 now;
1897 ret = kern_path(device_path, LOOKUP_FOLLOW, &path);
1901 now = current_time(d_inode(path.dentry));
1902 inode_update_time(d_inode(path.dentry), &now, S_MTIME | S_CTIME);
1906 static int btrfs_rm_dev_item(struct btrfs_device *device)
1908 struct btrfs_root *root = device->fs_info->chunk_root;
1910 struct btrfs_path *path;
1911 struct btrfs_key key;
1912 struct btrfs_trans_handle *trans;
1914 path = btrfs_alloc_path();
1918 trans = btrfs_start_transaction(root, 0);
1919 if (IS_ERR(trans)) {
1920 btrfs_free_path(path);
1921 return PTR_ERR(trans);
1923 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1924 key.type = BTRFS_DEV_ITEM_KEY;
1925 key.offset = device->devid;
1927 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1931 btrfs_abort_transaction(trans, ret);
1932 btrfs_end_transaction(trans);
1936 ret = btrfs_del_item(trans, root, path);
1938 btrfs_abort_transaction(trans, ret);
1939 btrfs_end_transaction(trans);
1943 btrfs_free_path(path);
1945 ret = btrfs_commit_transaction(trans);
1950 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1951 * filesystem. It's up to the caller to adjust that number regarding eg. device
1954 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1962 seq = read_seqbegin(&fs_info->profiles_lock);
1964 all_avail = fs_info->avail_data_alloc_bits |
1965 fs_info->avail_system_alloc_bits |
1966 fs_info->avail_metadata_alloc_bits;
1967 } while (read_seqretry(&fs_info->profiles_lock, seq));
1969 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1970 if (!(all_avail & btrfs_raid_array[i].bg_flag))
1973 if (num_devices < btrfs_raid_array[i].devs_min)
1974 return btrfs_raid_array[i].mindev_error;
1980 static struct btrfs_device * btrfs_find_next_active_device(
1981 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device)
1983 struct btrfs_device *next_device;
1985 list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1986 if (next_device != device &&
1987 !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state)
1988 && next_device->bdev)
1996 * Helper function to check if the given device is part of s_bdev / latest_dev
1997 * and replace it with the provided or the next active device, in the context
1998 * where this function called, there should be always be another device (or
1999 * this_dev) which is active.
2001 void __cold btrfs_assign_next_active_device(struct btrfs_device *device,
2002 struct btrfs_device *next_device)
2004 struct btrfs_fs_info *fs_info = device->fs_info;
2007 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
2009 ASSERT(next_device);
2011 if (fs_info->sb->s_bdev &&
2012 (fs_info->sb->s_bdev == device->bdev))
2013 fs_info->sb->s_bdev = next_device->bdev;
2015 if (fs_info->fs_devices->latest_dev->bdev == device->bdev)
2016 fs_info->fs_devices->latest_dev = next_device;
2020 * Return btrfs_fs_devices::num_devices excluding the device that's being
2021 * currently replaced.
2023 static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info)
2025 u64 num_devices = fs_info->fs_devices->num_devices;
2027 down_read(&fs_info->dev_replace.rwsem);
2028 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
2029 ASSERT(num_devices > 1);
2032 up_read(&fs_info->dev_replace.rwsem);
2037 void btrfs_scratch_superblocks(struct btrfs_fs_info *fs_info,
2038 struct block_device *bdev,
2039 const char *device_path)
2041 struct btrfs_super_block *disk_super;
2047 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; copy_num++) {
2051 disk_super = btrfs_read_dev_one_super(bdev, copy_num);
2052 if (IS_ERR(disk_super))
2055 if (bdev_is_zoned(bdev)) {
2056 btrfs_reset_sb_log_zones(bdev, copy_num);
2060 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
2062 page = virt_to_page(disk_super);
2063 set_page_dirty(page);
2065 /* write_on_page() unlocks the page */
2066 ret = write_one_page(page);
2069 "error clearing superblock number %d (%d)",
2071 btrfs_release_disk_super(disk_super);
2075 /* Notify udev that device has changed */
2076 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
2078 /* Update ctime/mtime for device path for libblkid */
2079 update_dev_time(device_path);
2082 int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path,
2083 u64 devid, struct block_device **bdev, fmode_t *mode)
2085 struct btrfs_device *device;
2086 struct btrfs_fs_devices *cur_devices;
2087 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2092 * The device list in fs_devices is accessed without locks (neither
2093 * uuid_mutex nor device_list_mutex) as it won't change on a mounted
2094 * filesystem and another device rm cannot run.
2096 num_devices = btrfs_num_devices(fs_info);
2098 ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
2102 device = btrfs_find_device_by_devspec(fs_info, devid, device_path);
2104 if (IS_ERR(device)) {
2105 if (PTR_ERR(device) == -ENOENT &&
2106 device_path && strcmp(device_path, "missing") == 0)
2107 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2109 ret = PTR_ERR(device);
2113 if (btrfs_pinned_by_swapfile(fs_info, device)) {
2114 btrfs_warn_in_rcu(fs_info,
2115 "cannot remove device %s (devid %llu) due to active swapfile",
2116 rcu_str_deref(device->name), device->devid);
2121 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2122 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
2126 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
2127 fs_info->fs_devices->rw_devices == 1) {
2128 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
2132 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2133 mutex_lock(&fs_info->chunk_mutex);
2134 list_del_init(&device->dev_alloc_list);
2135 device->fs_devices->rw_devices--;
2136 mutex_unlock(&fs_info->chunk_mutex);
2139 ret = btrfs_shrink_device(device, 0);
2141 btrfs_reada_remove_dev(device);
2146 * TODO: the superblock still includes this device in its num_devices
2147 * counter although write_all_supers() is not locked out. This
2148 * could give a filesystem state which requires a degraded mount.
2150 ret = btrfs_rm_dev_item(device);
2154 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2155 btrfs_scrub_cancel_dev(device);
2158 * the device list mutex makes sure that we don't change
2159 * the device list while someone else is writing out all
2160 * the device supers. Whoever is writing all supers, should
2161 * lock the device list mutex before getting the number of
2162 * devices in the super block (super_copy). Conversely,
2163 * whoever updates the number of devices in the super block
2164 * (super_copy) should hold the device list mutex.
2168 * In normal cases the cur_devices == fs_devices. But in case
2169 * of deleting a seed device, the cur_devices should point to
2170 * its own fs_devices listed under the fs_devices->seed.
2172 cur_devices = device->fs_devices;
2173 mutex_lock(&fs_devices->device_list_mutex);
2174 list_del_rcu(&device->dev_list);
2176 cur_devices->num_devices--;
2177 cur_devices->total_devices--;
2178 /* Update total_devices of the parent fs_devices if it's seed */
2179 if (cur_devices != fs_devices)
2180 fs_devices->total_devices--;
2182 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
2183 cur_devices->missing_devices--;
2185 btrfs_assign_next_active_device(device, NULL);
2188 cur_devices->open_devices--;
2189 /* remove sysfs entry */
2190 btrfs_sysfs_remove_device(device);
2193 num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
2194 btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
2195 mutex_unlock(&fs_devices->device_list_mutex);
2198 * At this point, the device is zero sized and detached from the
2199 * devices list. All that's left is to zero out the old supers and
2202 * We cannot call btrfs_close_bdev() here because we're holding the sb
2203 * write lock, and blkdev_put() will pull in the ->open_mutex on the
2204 * block device and it's dependencies. Instead just flush the device
2205 * and let the caller do the final blkdev_put.
2207 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2208 btrfs_scratch_superblocks(fs_info, device->bdev,
2211 sync_blockdev(device->bdev);
2212 invalidate_bdev(device->bdev);
2216 *bdev = device->bdev;
2217 *mode = device->mode;
2219 btrfs_free_device(device);
2221 if (cur_devices->open_devices == 0) {
2222 list_del_init(&cur_devices->seed_list);
2223 close_fs_devices(cur_devices);
2224 free_fs_devices(cur_devices);
2231 btrfs_reada_undo_remove_dev(device);
2232 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2233 mutex_lock(&fs_info->chunk_mutex);
2234 list_add(&device->dev_alloc_list,
2235 &fs_devices->alloc_list);
2236 device->fs_devices->rw_devices++;
2237 mutex_unlock(&fs_info->chunk_mutex);
2242 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev)
2244 struct btrfs_fs_devices *fs_devices;
2246 lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex);
2249 * in case of fs with no seed, srcdev->fs_devices will point
2250 * to fs_devices of fs_info. However when the dev being replaced is
2251 * a seed dev it will point to the seed's local fs_devices. In short
2252 * srcdev will have its correct fs_devices in both the cases.
2254 fs_devices = srcdev->fs_devices;
2256 list_del_rcu(&srcdev->dev_list);
2257 list_del(&srcdev->dev_alloc_list);
2258 fs_devices->num_devices--;
2259 if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state))
2260 fs_devices->missing_devices--;
2262 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state))
2263 fs_devices->rw_devices--;
2266 fs_devices->open_devices--;
2269 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev)
2271 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2273 mutex_lock(&uuid_mutex);
2275 btrfs_close_bdev(srcdev);
2277 btrfs_free_device(srcdev);
2279 /* if this is no devs we rather delete the fs_devices */
2280 if (!fs_devices->num_devices) {
2282 * On a mounted FS, num_devices can't be zero unless it's a
2283 * seed. In case of a seed device being replaced, the replace
2284 * target added to the sprout FS, so there will be no more
2285 * device left under the seed FS.
2287 ASSERT(fs_devices->seeding);
2289 list_del_init(&fs_devices->seed_list);
2290 close_fs_devices(fs_devices);
2291 free_fs_devices(fs_devices);
2293 mutex_unlock(&uuid_mutex);
2296 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev)
2298 struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices;
2300 mutex_lock(&fs_devices->device_list_mutex);
2302 btrfs_sysfs_remove_device(tgtdev);
2305 fs_devices->open_devices--;
2307 fs_devices->num_devices--;
2309 btrfs_assign_next_active_device(tgtdev, NULL);
2311 list_del_rcu(&tgtdev->dev_list);
2313 mutex_unlock(&fs_devices->device_list_mutex);
2316 * The update_dev_time() with in btrfs_scratch_superblocks()
2317 * may lead to a call to btrfs_show_devname() which will try
2318 * to hold device_list_mutex. And here this device
2319 * is already out of device list, so we don't have to hold
2320 * the device_list_mutex lock.
2322 btrfs_scratch_superblocks(tgtdev->fs_info, tgtdev->bdev,
2325 btrfs_close_bdev(tgtdev);
2327 btrfs_free_device(tgtdev);
2330 static struct btrfs_device *btrfs_find_device_by_path(
2331 struct btrfs_fs_info *fs_info, const char *device_path)
2334 struct btrfs_super_block *disk_super;
2337 struct block_device *bdev;
2338 struct btrfs_device *device;
2340 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2341 fs_info->bdev_holder, 0, &bdev, &disk_super);
2343 return ERR_PTR(ret);
2345 devid = btrfs_stack_device_id(&disk_super->dev_item);
2346 dev_uuid = disk_super->dev_item.uuid;
2347 if (btrfs_fs_incompat(fs_info, METADATA_UUID))
2348 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
2349 disk_super->metadata_uuid);
2351 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
2354 btrfs_release_disk_super(disk_super);
2356 device = ERR_PTR(-ENOENT);
2357 blkdev_put(bdev, FMODE_READ);
2362 * Lookup a device given by device id, or the path if the id is 0.
2364 struct btrfs_device *btrfs_find_device_by_devspec(
2365 struct btrfs_fs_info *fs_info, u64 devid,
2366 const char *device_path)
2368 struct btrfs_device *device;
2371 device = btrfs_find_device(fs_info->fs_devices, devid, NULL,
2374 return ERR_PTR(-ENOENT);
2378 if (!device_path || !device_path[0])
2379 return ERR_PTR(-EINVAL);
2381 if (strcmp(device_path, "missing") == 0) {
2382 /* Find first missing device */
2383 list_for_each_entry(device, &fs_info->fs_devices->devices,
2385 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
2386 &device->dev_state) && !device->bdev)
2389 return ERR_PTR(-ENOENT);
2392 return btrfs_find_device_by_path(fs_info, device_path);
2396 * does all the dirty work required for changing file system's UUID.
2398 static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info)
2400 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2401 struct btrfs_fs_devices *old_devices;
2402 struct btrfs_fs_devices *seed_devices;
2403 struct btrfs_super_block *disk_super = fs_info->super_copy;
2404 struct btrfs_device *device;
2407 lockdep_assert_held(&uuid_mutex);
2408 if (!fs_devices->seeding)
2412 * Private copy of the seed devices, anchored at
2413 * fs_info->fs_devices->seed_list
2415 seed_devices = alloc_fs_devices(NULL, NULL);
2416 if (IS_ERR(seed_devices))
2417 return PTR_ERR(seed_devices);
2420 * It's necessary to retain a copy of the original seed fs_devices in
2421 * fs_uuids so that filesystems which have been seeded can successfully
2422 * reference the seed device from open_seed_devices. This also supports
2425 old_devices = clone_fs_devices(fs_devices);
2426 if (IS_ERR(old_devices)) {
2427 kfree(seed_devices);
2428 return PTR_ERR(old_devices);
2431 list_add(&old_devices->fs_list, &fs_uuids);
2433 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2434 seed_devices->opened = 1;
2435 INIT_LIST_HEAD(&seed_devices->devices);
2436 INIT_LIST_HEAD(&seed_devices->alloc_list);
2437 mutex_init(&seed_devices->device_list_mutex);
2439 mutex_lock(&fs_devices->device_list_mutex);
2440 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2442 list_for_each_entry(device, &seed_devices->devices, dev_list)
2443 device->fs_devices = seed_devices;
2445 fs_devices->seeding = false;
2446 fs_devices->num_devices = 0;
2447 fs_devices->open_devices = 0;
2448 fs_devices->missing_devices = 0;
2449 fs_devices->rotating = false;
2450 list_add(&seed_devices->seed_list, &fs_devices->seed_list);
2452 generate_random_uuid(fs_devices->fsid);
2453 memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE);
2454 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2455 mutex_unlock(&fs_devices->device_list_mutex);
2457 super_flags = btrfs_super_flags(disk_super) &
2458 ~BTRFS_SUPER_FLAG_SEEDING;
2459 btrfs_set_super_flags(disk_super, super_flags);
2465 * Store the expected generation for seed devices in device items.
2467 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans)
2469 struct btrfs_fs_info *fs_info = trans->fs_info;
2470 struct btrfs_root *root = fs_info->chunk_root;
2471 struct btrfs_path *path;
2472 struct extent_buffer *leaf;
2473 struct btrfs_dev_item *dev_item;
2474 struct btrfs_device *device;
2475 struct btrfs_key key;
2476 u8 fs_uuid[BTRFS_FSID_SIZE];
2477 u8 dev_uuid[BTRFS_UUID_SIZE];
2481 path = btrfs_alloc_path();
2485 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2487 key.type = BTRFS_DEV_ITEM_KEY;
2490 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2494 leaf = path->nodes[0];
2496 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2497 ret = btrfs_next_leaf(root, path);
2502 leaf = path->nodes[0];
2503 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2504 btrfs_release_path(path);
2508 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2509 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2510 key.type != BTRFS_DEV_ITEM_KEY)
2513 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2514 struct btrfs_dev_item);
2515 devid = btrfs_device_id(leaf, dev_item);
2516 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2518 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2520 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
2522 BUG_ON(!device); /* Logic error */
2524 if (device->fs_devices->seeding) {
2525 btrfs_set_device_generation(leaf, dev_item,
2526 device->generation);
2527 btrfs_mark_buffer_dirty(leaf);
2535 btrfs_free_path(path);
2539 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2541 struct btrfs_root *root = fs_info->dev_root;
2542 struct request_queue *q;
2543 struct btrfs_trans_handle *trans;
2544 struct btrfs_device *device;
2545 struct block_device *bdev;
2546 struct super_block *sb = fs_info->sb;
2547 struct rcu_string *name;
2548 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2549 u64 orig_super_total_bytes;
2550 u64 orig_super_num_devices;
2551 int seeding_dev = 0;
2553 bool locked = false;
2555 if (sb_rdonly(sb) && !fs_devices->seeding)
2558 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2559 fs_info->bdev_holder);
2561 return PTR_ERR(bdev);
2563 if (!btrfs_check_device_zone_type(fs_info, bdev)) {
2568 if (fs_devices->seeding) {
2570 down_write(&sb->s_umount);
2571 mutex_lock(&uuid_mutex);
2575 sync_blockdev(bdev);
2578 list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
2579 if (device->bdev == bdev) {
2587 device = btrfs_alloc_device(fs_info, NULL, NULL);
2588 if (IS_ERR(device)) {
2589 /* we can safely leave the fs_devices entry around */
2590 ret = PTR_ERR(device);
2594 name = rcu_string_strdup(device_path, GFP_KERNEL);
2597 goto error_free_device;
2599 rcu_assign_pointer(device->name, name);
2601 device->fs_info = fs_info;
2602 device->bdev = bdev;
2604 ret = btrfs_get_dev_zone_info(device);
2606 goto error_free_device;
2608 trans = btrfs_start_transaction(root, 0);
2609 if (IS_ERR(trans)) {
2610 ret = PTR_ERR(trans);
2611 goto error_free_zone;
2614 q = bdev_get_queue(bdev);
2615 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
2616 device->generation = trans->transid;
2617 device->io_width = fs_info->sectorsize;
2618 device->io_align = fs_info->sectorsize;
2619 device->sector_size = fs_info->sectorsize;
2620 device->total_bytes = round_down(i_size_read(bdev->bd_inode),
2621 fs_info->sectorsize);
2622 device->disk_total_bytes = device->total_bytes;
2623 device->commit_total_bytes = device->total_bytes;
2624 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2625 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
2626 device->mode = FMODE_EXCL;
2627 device->dev_stats_valid = 1;
2628 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2631 btrfs_clear_sb_rdonly(sb);
2632 ret = btrfs_prepare_sprout(fs_info);
2634 btrfs_abort_transaction(trans, ret);
2637 btrfs_assign_next_active_device(fs_info->fs_devices->latest_dev,
2641 device->fs_devices = fs_devices;
2643 mutex_lock(&fs_devices->device_list_mutex);
2644 mutex_lock(&fs_info->chunk_mutex);
2645 list_add_rcu(&device->dev_list, &fs_devices->devices);
2646 list_add(&device->dev_alloc_list, &fs_devices->alloc_list);
2647 fs_devices->num_devices++;
2648 fs_devices->open_devices++;
2649 fs_devices->rw_devices++;
2650 fs_devices->total_devices++;
2651 fs_devices->total_rw_bytes += device->total_bytes;
2653 atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2655 if (!blk_queue_nonrot(q))
2656 fs_devices->rotating = true;
2658 orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
2659 btrfs_set_super_total_bytes(fs_info->super_copy,
2660 round_down(orig_super_total_bytes + device->total_bytes,
2661 fs_info->sectorsize));
2663 orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy);
2664 btrfs_set_super_num_devices(fs_info->super_copy,
2665 orig_super_num_devices + 1);
2668 * we've got more storage, clear any full flags on the space
2671 btrfs_clear_space_info_full(fs_info);
2673 mutex_unlock(&fs_info->chunk_mutex);
2675 /* Add sysfs device entry */
2676 btrfs_sysfs_add_device(device);
2678 mutex_unlock(&fs_devices->device_list_mutex);
2681 mutex_lock(&fs_info->chunk_mutex);
2682 ret = init_first_rw_device(trans);
2683 mutex_unlock(&fs_info->chunk_mutex);
2685 btrfs_abort_transaction(trans, ret);
2690 ret = btrfs_add_dev_item(trans, device);
2692 btrfs_abort_transaction(trans, ret);
2697 ret = btrfs_finish_sprout(trans);
2699 btrfs_abort_transaction(trans, ret);
2704 * fs_devices now represents the newly sprouted filesystem and
2705 * its fsid has been changed by btrfs_prepare_sprout
2707 btrfs_sysfs_update_sprout_fsid(fs_devices);
2710 ret = btrfs_commit_transaction(trans);
2713 mutex_unlock(&uuid_mutex);
2714 up_write(&sb->s_umount);
2717 if (ret) /* transaction commit */
2720 ret = btrfs_relocate_sys_chunks(fs_info);
2722 btrfs_handle_fs_error(fs_info, ret,
2723 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2724 trans = btrfs_attach_transaction(root);
2725 if (IS_ERR(trans)) {
2726 if (PTR_ERR(trans) == -ENOENT)
2728 ret = PTR_ERR(trans);
2732 ret = btrfs_commit_transaction(trans);
2736 * Now that we have written a new super block to this device, check all
2737 * other fs_devices list if device_path alienates any other scanned
2739 * We can ignore the return value as it typically returns -EINVAL and
2740 * only succeeds if the device was an alien.
2742 btrfs_forget_devices(device_path);
2744 /* Update ctime/mtime for blkid or udev */
2745 update_dev_time(device_path);
2750 btrfs_sysfs_remove_device(device);
2751 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2752 mutex_lock(&fs_info->chunk_mutex);
2753 list_del_rcu(&device->dev_list);
2754 list_del(&device->dev_alloc_list);
2755 fs_info->fs_devices->num_devices--;
2756 fs_info->fs_devices->open_devices--;
2757 fs_info->fs_devices->rw_devices--;
2758 fs_info->fs_devices->total_devices--;
2759 fs_info->fs_devices->total_rw_bytes -= device->total_bytes;
2760 atomic64_sub(device->total_bytes, &fs_info->free_chunk_space);
2761 btrfs_set_super_total_bytes(fs_info->super_copy,
2762 orig_super_total_bytes);
2763 btrfs_set_super_num_devices(fs_info->super_copy,
2764 orig_super_num_devices);
2765 mutex_unlock(&fs_info->chunk_mutex);
2766 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2769 btrfs_set_sb_rdonly(sb);
2771 btrfs_end_transaction(trans);
2773 btrfs_destroy_dev_zone_info(device);
2775 btrfs_free_device(device);
2777 blkdev_put(bdev, FMODE_EXCL);
2779 mutex_unlock(&uuid_mutex);
2780 up_write(&sb->s_umount);
2785 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2786 struct btrfs_device *device)
2789 struct btrfs_path *path;
2790 struct btrfs_root *root = device->fs_info->chunk_root;
2791 struct btrfs_dev_item *dev_item;
2792 struct extent_buffer *leaf;
2793 struct btrfs_key key;
2795 path = btrfs_alloc_path();
2799 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2800 key.type = BTRFS_DEV_ITEM_KEY;
2801 key.offset = device->devid;
2803 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2812 leaf = path->nodes[0];
2813 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2815 btrfs_set_device_id(leaf, dev_item, device->devid);
2816 btrfs_set_device_type(leaf, dev_item, device->type);
2817 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2818 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2819 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2820 btrfs_set_device_total_bytes(leaf, dev_item,
2821 btrfs_device_get_disk_total_bytes(device));
2822 btrfs_set_device_bytes_used(leaf, dev_item,
2823 btrfs_device_get_bytes_used(device));
2824 btrfs_mark_buffer_dirty(leaf);
2827 btrfs_free_path(path);
2831 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2832 struct btrfs_device *device, u64 new_size)
2834 struct btrfs_fs_info *fs_info = device->fs_info;
2835 struct btrfs_super_block *super_copy = fs_info->super_copy;
2839 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2842 new_size = round_down(new_size, fs_info->sectorsize);
2844 mutex_lock(&fs_info->chunk_mutex);
2845 old_total = btrfs_super_total_bytes(super_copy);
2846 diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2848 if (new_size <= device->total_bytes ||
2849 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2850 mutex_unlock(&fs_info->chunk_mutex);
2854 btrfs_set_super_total_bytes(super_copy,
2855 round_down(old_total + diff, fs_info->sectorsize));
2856 device->fs_devices->total_rw_bytes += diff;
2858 btrfs_device_set_total_bytes(device, new_size);
2859 btrfs_device_set_disk_total_bytes(device, new_size);
2860 btrfs_clear_space_info_full(device->fs_info);
2861 if (list_empty(&device->post_commit_list))
2862 list_add_tail(&device->post_commit_list,
2863 &trans->transaction->dev_update_list);
2864 mutex_unlock(&fs_info->chunk_mutex);
2866 return btrfs_update_device(trans, device);
2869 static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
2871 struct btrfs_fs_info *fs_info = trans->fs_info;
2872 struct btrfs_root *root = fs_info->chunk_root;
2874 struct btrfs_path *path;
2875 struct btrfs_key key;
2877 path = btrfs_alloc_path();
2881 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2882 key.offset = chunk_offset;
2883 key.type = BTRFS_CHUNK_ITEM_KEY;
2885 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2888 else if (ret > 0) { /* Logic error or corruption */
2889 btrfs_handle_fs_error(fs_info, -ENOENT,
2890 "Failed lookup while freeing chunk.");
2895 ret = btrfs_del_item(trans, root, path);
2897 btrfs_handle_fs_error(fs_info, ret,
2898 "Failed to delete chunk item.");
2900 btrfs_free_path(path);
2904 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2906 struct btrfs_super_block *super_copy = fs_info->super_copy;
2907 struct btrfs_disk_key *disk_key;
2908 struct btrfs_chunk *chunk;
2915 struct btrfs_key key;
2917 lockdep_assert_held(&fs_info->chunk_mutex);
2918 array_size = btrfs_super_sys_array_size(super_copy);
2920 ptr = super_copy->sys_chunk_array;
2923 while (cur < array_size) {
2924 disk_key = (struct btrfs_disk_key *)ptr;
2925 btrfs_disk_key_to_cpu(&key, disk_key);
2927 len = sizeof(*disk_key);
2929 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2930 chunk = (struct btrfs_chunk *)(ptr + len);
2931 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2932 len += btrfs_chunk_item_size(num_stripes);
2937 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID &&
2938 key.offset == chunk_offset) {
2939 memmove(ptr, ptr + len, array_size - (cur + len));
2941 btrfs_set_super_sys_array_size(super_copy, array_size);
2951 * btrfs_get_chunk_map() - Find the mapping containing the given logical extent.
2952 * @logical: Logical block offset in bytes.
2953 * @length: Length of extent in bytes.
2955 * Return: Chunk mapping or ERR_PTR.
2957 struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info,
2958 u64 logical, u64 length)
2960 struct extent_map_tree *em_tree;
2961 struct extent_map *em;
2963 em_tree = &fs_info->mapping_tree;
2964 read_lock(&em_tree->lock);
2965 em = lookup_extent_mapping(em_tree, logical, length);
2966 read_unlock(&em_tree->lock);
2969 btrfs_crit(fs_info, "unable to find logical %llu length %llu",
2971 return ERR_PTR(-EINVAL);
2974 if (em->start > logical || em->start + em->len < logical) {
2976 "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
2977 logical, length, em->start, em->start + em->len);
2978 free_extent_map(em);
2979 return ERR_PTR(-EINVAL);
2982 /* callers are responsible for dropping em's ref. */
2986 static int remove_chunk_item(struct btrfs_trans_handle *trans,
2987 struct map_lookup *map, u64 chunk_offset)
2992 * Removing chunk items and updating the device items in the chunks btree
2993 * requires holding the chunk_mutex.
2994 * See the comment at btrfs_chunk_alloc() for the details.
2996 lockdep_assert_held(&trans->fs_info->chunk_mutex);
2998 for (i = 0; i < map->num_stripes; i++) {
3001 ret = btrfs_update_device(trans, map->stripes[i].dev);
3006 return btrfs_free_chunk(trans, chunk_offset);
3009 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
3011 struct btrfs_fs_info *fs_info = trans->fs_info;
3012 struct extent_map *em;
3013 struct map_lookup *map;
3014 u64 dev_extent_len = 0;
3016 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
3018 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
3021 * This is a logic error, but we don't want to just rely on the
3022 * user having built with ASSERT enabled, so if ASSERT doesn't
3023 * do anything we still error out.
3028 map = em->map_lookup;
3031 * First delete the device extent items from the devices btree.
3032 * We take the device_list_mutex to avoid racing with the finishing phase
3033 * of a device replace operation. See the comment below before acquiring
3034 * fs_info->chunk_mutex. Note that here we do not acquire the chunk_mutex
3035 * because that can result in a deadlock when deleting the device extent
3036 * items from the devices btree - COWing an extent buffer from the btree
3037 * may result in allocating a new metadata chunk, which would attempt to
3038 * lock again fs_info->chunk_mutex.
3040 mutex_lock(&fs_devices->device_list_mutex);
3041 for (i = 0; i < map->num_stripes; i++) {
3042 struct btrfs_device *device = map->stripes[i].dev;
3043 ret = btrfs_free_dev_extent(trans, device,
3044 map->stripes[i].physical,
3047 mutex_unlock(&fs_devices->device_list_mutex);
3048 btrfs_abort_transaction(trans, ret);
3052 if (device->bytes_used > 0) {
3053 mutex_lock(&fs_info->chunk_mutex);
3054 btrfs_device_set_bytes_used(device,
3055 device->bytes_used - dev_extent_len);
3056 atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
3057 btrfs_clear_space_info_full(fs_info);
3058 mutex_unlock(&fs_info->chunk_mutex);
3061 mutex_unlock(&fs_devices->device_list_mutex);
3064 * We acquire fs_info->chunk_mutex for 2 reasons:
3066 * 1) Just like with the first phase of the chunk allocation, we must
3067 * reserve system space, do all chunk btree updates and deletions, and
3068 * update the system chunk array in the superblock while holding this
3069 * mutex. This is for similar reasons as explained on the comment at
3070 * the top of btrfs_chunk_alloc();
3072 * 2) Prevent races with the final phase of a device replace operation
3073 * that replaces the device object associated with the map's stripes,
3074 * because the device object's id can change at any time during that
3075 * final phase of the device replace operation
3076 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the
3077 * replaced device and then see it with an ID of
3078 * BTRFS_DEV_REPLACE_DEVID, which would cause a failure when updating
3079 * the device item, which does not exists on the chunk btree.
3080 * The finishing phase of device replace acquires both the
3081 * device_list_mutex and the chunk_mutex, in that order, so we are
3082 * safe by just acquiring the chunk_mutex.
3084 trans->removing_chunk = true;
3085 mutex_lock(&fs_info->chunk_mutex);
3087 check_system_chunk(trans, map->type);
3089 ret = remove_chunk_item(trans, map, chunk_offset);
3091 * Normally we should not get -ENOSPC since we reserved space before
3092 * through the call to check_system_chunk().
3094 * Despite our system space_info having enough free space, we may not
3095 * be able to allocate extents from its block groups, because all have
3096 * an incompatible profile, which will force us to allocate a new system
3097 * block group with the right profile, or right after we called
3098 * check_system_space() above, a scrub turned the only system block group
3099 * with enough free space into RO mode.
3100 * This is explained with more detail at do_chunk_alloc().
3102 * So if we get -ENOSPC, allocate a new system chunk and retry once.
3104 if (ret == -ENOSPC) {
3105 const u64 sys_flags = btrfs_system_alloc_profile(fs_info);
3106 struct btrfs_block_group *sys_bg;
3108 sys_bg = btrfs_alloc_chunk(trans, sys_flags);
3109 if (IS_ERR(sys_bg)) {
3110 ret = PTR_ERR(sys_bg);
3111 btrfs_abort_transaction(trans, ret);
3115 ret = btrfs_chunk_alloc_add_chunk_item(trans, sys_bg);
3117 btrfs_abort_transaction(trans, ret);
3121 ret = remove_chunk_item(trans, map, chunk_offset);
3123 btrfs_abort_transaction(trans, ret);
3127 btrfs_abort_transaction(trans, ret);
3131 trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
3133 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
3134 ret = btrfs_del_sys_chunk(fs_info, chunk_offset);
3136 btrfs_abort_transaction(trans, ret);
3141 mutex_unlock(&fs_info->chunk_mutex);
3142 trans->removing_chunk = false;
3145 * We are done with chunk btree updates and deletions, so release the
3146 * system space we previously reserved (with check_system_chunk()).
3148 btrfs_trans_release_chunk_metadata(trans);
3150 ret = btrfs_remove_block_group(trans, chunk_offset, em);
3152 btrfs_abort_transaction(trans, ret);
3157 if (trans->removing_chunk) {
3158 mutex_unlock(&fs_info->chunk_mutex);
3159 trans->removing_chunk = false;
3162 free_extent_map(em);
3166 int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
3168 struct btrfs_root *root = fs_info->chunk_root;
3169 struct btrfs_trans_handle *trans;
3170 struct btrfs_block_group *block_group;
3175 * Prevent races with automatic removal of unused block groups.
3176 * After we relocate and before we remove the chunk with offset
3177 * chunk_offset, automatic removal of the block group can kick in,
3178 * resulting in a failure when calling btrfs_remove_chunk() below.
3180 * Make sure to acquire this mutex before doing a tree search (dev
3181 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
3182 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
3183 * we release the path used to search the chunk/dev tree and before
3184 * the current task acquires this mutex and calls us.
3186 lockdep_assert_held(&fs_info->reclaim_bgs_lock);
3188 /* step one, relocate all the extents inside this chunk */
3189 btrfs_scrub_pause(fs_info);
3190 ret = btrfs_relocate_block_group(fs_info, chunk_offset);
3191 btrfs_scrub_continue(fs_info);
3195 block_group = btrfs_lookup_block_group(fs_info, chunk_offset);
3198 btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group);
3199 length = block_group->length;
3200 btrfs_put_block_group(block_group);
3203 * On a zoned file system, discard the whole block group, this will
3204 * trigger a REQ_OP_ZONE_RESET operation on the device zone. If
3205 * resetting the zone fails, don't treat it as a fatal problem from the
3206 * filesystem's point of view.
3208 if (btrfs_is_zoned(fs_info)) {
3209 ret = btrfs_discard_extent(fs_info, chunk_offset, length, NULL);
3212 "failed to reset zone %llu after relocation",
3216 trans = btrfs_start_trans_remove_block_group(root->fs_info,
3218 if (IS_ERR(trans)) {
3219 ret = PTR_ERR(trans);
3220 btrfs_handle_fs_error(root->fs_info, ret, NULL);
3225 * step two, delete the device extents and the
3226 * chunk tree entries
3228 ret = btrfs_remove_chunk(trans, chunk_offset);
3229 btrfs_end_transaction(trans);
3233 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
3235 struct btrfs_root *chunk_root = fs_info->chunk_root;
3236 struct btrfs_path *path;
3237 struct extent_buffer *leaf;
3238 struct btrfs_chunk *chunk;
3239 struct btrfs_key key;
3240 struct btrfs_key found_key;
3242 bool retried = false;
3246 path = btrfs_alloc_path();
3251 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3252 key.offset = (u64)-1;
3253 key.type = BTRFS_CHUNK_ITEM_KEY;
3256 mutex_lock(&fs_info->reclaim_bgs_lock);
3257 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3259 mutex_unlock(&fs_info->reclaim_bgs_lock);
3262 BUG_ON(ret == 0); /* Corruption */
3264 ret = btrfs_previous_item(chunk_root, path, key.objectid,
3267 mutex_unlock(&fs_info->reclaim_bgs_lock);
3273 leaf = path->nodes[0];
3274 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3276 chunk = btrfs_item_ptr(leaf, path->slots[0],
3277 struct btrfs_chunk);
3278 chunk_type = btrfs_chunk_type(leaf, chunk);
3279 btrfs_release_path(path);
3281 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3282 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3288 mutex_unlock(&fs_info->reclaim_bgs_lock);
3290 if (found_key.offset == 0)
3292 key.offset = found_key.offset - 1;
3295 if (failed && !retried) {
3299 } else if (WARN_ON(failed && retried)) {
3303 btrfs_free_path(path);
3308 * return 1 : allocate a data chunk successfully,
3309 * return <0: errors during allocating a data chunk,
3310 * return 0 : no need to allocate a data chunk.
3312 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info,
3315 struct btrfs_block_group *cache;
3319 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3321 chunk_type = cache->flags;
3322 btrfs_put_block_group(cache);
3324 if (!(chunk_type & BTRFS_BLOCK_GROUP_DATA))
3327 spin_lock(&fs_info->data_sinfo->lock);
3328 bytes_used = fs_info->data_sinfo->bytes_used;
3329 spin_unlock(&fs_info->data_sinfo->lock);
3332 struct btrfs_trans_handle *trans;
3335 trans = btrfs_join_transaction(fs_info->tree_root);
3337 return PTR_ERR(trans);
3339 ret = btrfs_force_chunk_alloc(trans, BTRFS_BLOCK_GROUP_DATA);
3340 btrfs_end_transaction(trans);
3349 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3350 struct btrfs_balance_control *bctl)
3352 struct btrfs_root *root = fs_info->tree_root;
3353 struct btrfs_trans_handle *trans;
3354 struct btrfs_balance_item *item;
3355 struct btrfs_disk_balance_args disk_bargs;
3356 struct btrfs_path *path;
3357 struct extent_buffer *leaf;
3358 struct btrfs_key key;
3361 path = btrfs_alloc_path();
3365 trans = btrfs_start_transaction(root, 0);
3366 if (IS_ERR(trans)) {
3367 btrfs_free_path(path);
3368 return PTR_ERR(trans);
3371 key.objectid = BTRFS_BALANCE_OBJECTID;
3372 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3375 ret = btrfs_insert_empty_item(trans, root, path, &key,
3380 leaf = path->nodes[0];
3381 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3383 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3385 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3386 btrfs_set_balance_data(leaf, item, &disk_bargs);
3387 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3388 btrfs_set_balance_meta(leaf, item, &disk_bargs);
3389 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3390 btrfs_set_balance_sys(leaf, item, &disk_bargs);
3392 btrfs_set_balance_flags(leaf, item, bctl->flags);
3394 btrfs_mark_buffer_dirty(leaf);
3396 btrfs_free_path(path);
3397 err = btrfs_commit_transaction(trans);
3403 static int del_balance_item(struct btrfs_fs_info *fs_info)
3405 struct btrfs_root *root = fs_info->tree_root;
3406 struct btrfs_trans_handle *trans;
3407 struct btrfs_path *path;
3408 struct btrfs_key key;
3411 path = btrfs_alloc_path();
3415 trans = btrfs_start_transaction_fallback_global_rsv(root, 0);
3416 if (IS_ERR(trans)) {
3417 btrfs_free_path(path);
3418 return PTR_ERR(trans);
3421 key.objectid = BTRFS_BALANCE_OBJECTID;
3422 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3425 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3433 ret = btrfs_del_item(trans, root, path);
3435 btrfs_free_path(path);
3436 err = btrfs_commit_transaction(trans);
3443 * This is a heuristic used to reduce the number of chunks balanced on
3444 * resume after balance was interrupted.
3446 static void update_balance_args(struct btrfs_balance_control *bctl)
3449 * Turn on soft mode for chunk types that were being converted.
3451 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3452 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3453 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3454 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3455 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3456 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3459 * Turn on usage filter if is not already used. The idea is
3460 * that chunks that we have already balanced should be
3461 * reasonably full. Don't do it for chunks that are being
3462 * converted - that will keep us from relocating unconverted
3463 * (albeit full) chunks.
3465 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3466 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3467 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3468 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3469 bctl->data.usage = 90;
3471 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3472 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3473 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3474 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3475 bctl->sys.usage = 90;
3477 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3478 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3479 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3480 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3481 bctl->meta.usage = 90;
3486 * Clear the balance status in fs_info and delete the balance item from disk.
3488 static void reset_balance_state(struct btrfs_fs_info *fs_info)
3490 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3493 BUG_ON(!fs_info->balance_ctl);
3495 spin_lock(&fs_info->balance_lock);
3496 fs_info->balance_ctl = NULL;
3497 spin_unlock(&fs_info->balance_lock);
3500 ret = del_balance_item(fs_info);
3502 btrfs_handle_fs_error(fs_info, ret, NULL);
3506 * Balance filters. Return 1 if chunk should be filtered out
3507 * (should not be balanced).
3509 static int chunk_profiles_filter(u64 chunk_type,
3510 struct btrfs_balance_args *bargs)
3512 chunk_type = chunk_to_extended(chunk_type) &
3513 BTRFS_EXTENDED_PROFILE_MASK;
3515 if (bargs->profiles & chunk_type)
3521 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3522 struct btrfs_balance_args *bargs)
3524 struct btrfs_block_group *cache;
3526 u64 user_thresh_min;
3527 u64 user_thresh_max;
3530 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3531 chunk_used = cache->used;
3533 if (bargs->usage_min == 0)
3534 user_thresh_min = 0;
3536 user_thresh_min = div_factor_fine(cache->length,
3539 if (bargs->usage_max == 0)
3540 user_thresh_max = 1;
3541 else if (bargs->usage_max > 100)
3542 user_thresh_max = cache->length;
3544 user_thresh_max = div_factor_fine(cache->length,
3547 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3550 btrfs_put_block_group(cache);
3554 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3555 u64 chunk_offset, struct btrfs_balance_args *bargs)
3557 struct btrfs_block_group *cache;
3558 u64 chunk_used, user_thresh;
3561 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3562 chunk_used = cache->used;
3564 if (bargs->usage_min == 0)
3566 else if (bargs->usage > 100)
3567 user_thresh = cache->length;
3569 user_thresh = div_factor_fine(cache->length, bargs->usage);
3571 if (chunk_used < user_thresh)
3574 btrfs_put_block_group(cache);
3578 static int chunk_devid_filter(struct extent_buffer *leaf,
3579 struct btrfs_chunk *chunk,
3580 struct btrfs_balance_args *bargs)
3582 struct btrfs_stripe *stripe;
3583 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3586 for (i = 0; i < num_stripes; i++) {
3587 stripe = btrfs_stripe_nr(chunk, i);
3588 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3595 static u64 calc_data_stripes(u64 type, int num_stripes)
3597 const int index = btrfs_bg_flags_to_raid_index(type);
3598 const int ncopies = btrfs_raid_array[index].ncopies;
3599 const int nparity = btrfs_raid_array[index].nparity;
3601 return (num_stripes - nparity) / ncopies;
3604 /* [pstart, pend) */
3605 static int chunk_drange_filter(struct extent_buffer *leaf,
3606 struct btrfs_chunk *chunk,
3607 struct btrfs_balance_args *bargs)
3609 struct btrfs_stripe *stripe;
3610 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3617 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3620 type = btrfs_chunk_type(leaf, chunk);
3621 factor = calc_data_stripes(type, num_stripes);
3623 for (i = 0; i < num_stripes; i++) {
3624 stripe = btrfs_stripe_nr(chunk, i);
3625 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3628 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3629 stripe_length = btrfs_chunk_length(leaf, chunk);
3630 stripe_length = div_u64(stripe_length, factor);
3632 if (stripe_offset < bargs->pend &&
3633 stripe_offset + stripe_length > bargs->pstart)
3640 /* [vstart, vend) */
3641 static int chunk_vrange_filter(struct extent_buffer *leaf,
3642 struct btrfs_chunk *chunk,
3644 struct btrfs_balance_args *bargs)
3646 if (chunk_offset < bargs->vend &&
3647 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3648 /* at least part of the chunk is inside this vrange */
3654 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3655 struct btrfs_chunk *chunk,
3656 struct btrfs_balance_args *bargs)
3658 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3660 if (bargs->stripes_min <= num_stripes
3661 && num_stripes <= bargs->stripes_max)
3667 static int chunk_soft_convert_filter(u64 chunk_type,
3668 struct btrfs_balance_args *bargs)
3670 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3673 chunk_type = chunk_to_extended(chunk_type) &
3674 BTRFS_EXTENDED_PROFILE_MASK;
3676 if (bargs->target == chunk_type)
3682 static int should_balance_chunk(struct extent_buffer *leaf,
3683 struct btrfs_chunk *chunk, u64 chunk_offset)
3685 struct btrfs_fs_info *fs_info = leaf->fs_info;
3686 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3687 struct btrfs_balance_args *bargs = NULL;
3688 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3691 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3692 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3696 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3697 bargs = &bctl->data;
3698 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3700 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3701 bargs = &bctl->meta;
3703 /* profiles filter */
3704 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3705 chunk_profiles_filter(chunk_type, bargs)) {
3710 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3711 chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3713 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3714 chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3719 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3720 chunk_devid_filter(leaf, chunk, bargs)) {
3724 /* drange filter, makes sense only with devid filter */
3725 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3726 chunk_drange_filter(leaf, chunk, bargs)) {
3731 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3732 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3736 /* stripes filter */
3737 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3738 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3742 /* soft profile changing mode */
3743 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3744 chunk_soft_convert_filter(chunk_type, bargs)) {
3749 * limited by count, must be the last filter
3751 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3752 if (bargs->limit == 0)
3756 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3758 * Same logic as the 'limit' filter; the minimum cannot be
3759 * determined here because we do not have the global information
3760 * about the count of all chunks that satisfy the filters.
3762 if (bargs->limit_max == 0)
3771 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3773 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3774 struct btrfs_root *chunk_root = fs_info->chunk_root;
3776 struct btrfs_chunk *chunk;
3777 struct btrfs_path *path = NULL;
3778 struct btrfs_key key;
3779 struct btrfs_key found_key;
3780 struct extent_buffer *leaf;
3783 int enospc_errors = 0;
3784 bool counting = true;
3785 /* The single value limit and min/max limits use the same bytes in the */
3786 u64 limit_data = bctl->data.limit;
3787 u64 limit_meta = bctl->meta.limit;
3788 u64 limit_sys = bctl->sys.limit;
3792 int chunk_reserved = 0;
3794 path = btrfs_alloc_path();
3800 /* zero out stat counters */
3801 spin_lock(&fs_info->balance_lock);
3802 memset(&bctl->stat, 0, sizeof(bctl->stat));
3803 spin_unlock(&fs_info->balance_lock);
3807 * The single value limit and min/max limits use the same bytes
3810 bctl->data.limit = limit_data;
3811 bctl->meta.limit = limit_meta;
3812 bctl->sys.limit = limit_sys;
3814 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3815 key.offset = (u64)-1;
3816 key.type = BTRFS_CHUNK_ITEM_KEY;
3819 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3820 atomic_read(&fs_info->balance_cancel_req)) {
3825 mutex_lock(&fs_info->reclaim_bgs_lock);
3826 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3828 mutex_unlock(&fs_info->reclaim_bgs_lock);
3833 * this shouldn't happen, it means the last relocate
3837 BUG(); /* FIXME break ? */
3839 ret = btrfs_previous_item(chunk_root, path, 0,
3840 BTRFS_CHUNK_ITEM_KEY);
3842 mutex_unlock(&fs_info->reclaim_bgs_lock);
3847 leaf = path->nodes[0];
3848 slot = path->slots[0];
3849 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3851 if (found_key.objectid != key.objectid) {
3852 mutex_unlock(&fs_info->reclaim_bgs_lock);
3856 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3857 chunk_type = btrfs_chunk_type(leaf, chunk);
3860 spin_lock(&fs_info->balance_lock);
3861 bctl->stat.considered++;
3862 spin_unlock(&fs_info->balance_lock);
3865 ret = should_balance_chunk(leaf, chunk, found_key.offset);
3867 btrfs_release_path(path);
3869 mutex_unlock(&fs_info->reclaim_bgs_lock);
3874 mutex_unlock(&fs_info->reclaim_bgs_lock);
3875 spin_lock(&fs_info->balance_lock);
3876 bctl->stat.expected++;
3877 spin_unlock(&fs_info->balance_lock);
3879 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3881 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3883 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3890 * Apply limit_min filter, no need to check if the LIMITS
3891 * filter is used, limit_min is 0 by default
3893 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3894 count_data < bctl->data.limit_min)
3895 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3896 count_meta < bctl->meta.limit_min)
3897 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3898 count_sys < bctl->sys.limit_min)) {
3899 mutex_unlock(&fs_info->reclaim_bgs_lock);
3903 if (!chunk_reserved) {
3905 * We may be relocating the only data chunk we have,
3906 * which could potentially end up with losing data's
3907 * raid profile, so lets allocate an empty one in
3910 ret = btrfs_may_alloc_data_chunk(fs_info,
3913 mutex_unlock(&fs_info->reclaim_bgs_lock);
3915 } else if (ret == 1) {
3920 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3921 mutex_unlock(&fs_info->reclaim_bgs_lock);
3922 if (ret == -ENOSPC) {
3924 } else if (ret == -ETXTBSY) {
3926 "skipping relocation of block group %llu due to active swapfile",
3932 spin_lock(&fs_info->balance_lock);
3933 bctl->stat.completed++;
3934 spin_unlock(&fs_info->balance_lock);
3937 if (found_key.offset == 0)
3939 key.offset = found_key.offset - 1;
3943 btrfs_release_path(path);
3948 btrfs_free_path(path);
3949 if (enospc_errors) {
3950 btrfs_info(fs_info, "%d enospc errors during balance",
3960 * alloc_profile_is_valid - see if a given profile is valid and reduced
3961 * @flags: profile to validate
3962 * @extended: if true @flags is treated as an extended profile
3964 static int alloc_profile_is_valid(u64 flags, int extended)
3966 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3967 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3969 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3971 /* 1) check that all other bits are zeroed */
3975 /* 2) see if profile is reduced */
3977 return !extended; /* "0" is valid for usual profiles */
3979 return has_single_bit_set(flags);
3982 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3984 /* cancel requested || normal exit path */
3985 return atomic_read(&fs_info->balance_cancel_req) ||
3986 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3987 atomic_read(&fs_info->balance_cancel_req) == 0);
3991 * Validate target profile against allowed profiles and return true if it's OK.
3992 * Otherwise print the error message and return false.
3994 static inline int validate_convert_profile(struct btrfs_fs_info *fs_info,
3995 const struct btrfs_balance_args *bargs,
3996 u64 allowed, const char *type)
3998 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
4001 if (fs_info->sectorsize < PAGE_SIZE &&
4002 bargs->target & BTRFS_BLOCK_GROUP_RAID56_MASK) {
4004 "RAID56 is not yet supported for sectorsize %u with page size %lu",
4005 fs_info->sectorsize, PAGE_SIZE);
4008 /* Profile is valid and does not have bits outside of the allowed set */
4009 if (alloc_profile_is_valid(bargs->target, 1) &&
4010 (bargs->target & ~allowed) == 0)
4013 btrfs_err(fs_info, "balance: invalid convert %s profile %s",
4014 type, btrfs_bg_type_to_raid_name(bargs->target));
4019 * Fill @buf with textual description of balance filter flags @bargs, up to
4020 * @size_buf including the terminating null. The output may be trimmed if it
4021 * does not fit into the provided buffer.
4023 static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf,
4027 u32 size_bp = size_buf;
4029 u64 flags = bargs->flags;
4030 char tmp_buf[128] = {'\0'};
4035 #define CHECK_APPEND_NOARG(a) \
4037 ret = snprintf(bp, size_bp, (a)); \
4038 if (ret < 0 || ret >= size_bp) \
4039 goto out_overflow; \
4044 #define CHECK_APPEND_1ARG(a, v1) \
4046 ret = snprintf(bp, size_bp, (a), (v1)); \
4047 if (ret < 0 || ret >= size_bp) \
4048 goto out_overflow; \
4053 #define CHECK_APPEND_2ARG(a, v1, v2) \
4055 ret = snprintf(bp, size_bp, (a), (v1), (v2)); \
4056 if (ret < 0 || ret >= size_bp) \
4057 goto out_overflow; \
4062 if (flags & BTRFS_BALANCE_ARGS_CONVERT)
4063 CHECK_APPEND_1ARG("convert=%s,",
4064 btrfs_bg_type_to_raid_name(bargs->target));
4066 if (flags & BTRFS_BALANCE_ARGS_SOFT)
4067 CHECK_APPEND_NOARG("soft,");
4069 if (flags & BTRFS_BALANCE_ARGS_PROFILES) {
4070 btrfs_describe_block_groups(bargs->profiles, tmp_buf,
4072 CHECK_APPEND_1ARG("profiles=%s,", tmp_buf);
4075 if (flags & BTRFS_BALANCE_ARGS_USAGE)
4076 CHECK_APPEND_1ARG("usage=%llu,", bargs->usage);
4078 if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE)
4079 CHECK_APPEND_2ARG("usage=%u..%u,",
4080 bargs->usage_min, bargs->usage_max);
4082 if (flags & BTRFS_BALANCE_ARGS_DEVID)
4083 CHECK_APPEND_1ARG("devid=%llu,", bargs->devid);
4085 if (flags & BTRFS_BALANCE_ARGS_DRANGE)
4086 CHECK_APPEND_2ARG("drange=%llu..%llu,",
4087 bargs->pstart, bargs->pend);
4089 if (flags & BTRFS_BALANCE_ARGS_VRANGE)
4090 CHECK_APPEND_2ARG("vrange=%llu..%llu,",
4091 bargs->vstart, bargs->vend);
4093 if (flags & BTRFS_BALANCE_ARGS_LIMIT)
4094 CHECK_APPEND_1ARG("limit=%llu,", bargs->limit);
4096 if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)
4097 CHECK_APPEND_2ARG("limit=%u..%u,",
4098 bargs->limit_min, bargs->limit_max);
4100 if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE)
4101 CHECK_APPEND_2ARG("stripes=%u..%u,",
4102 bargs->stripes_min, bargs->stripes_max);
4104 #undef CHECK_APPEND_2ARG
4105 #undef CHECK_APPEND_1ARG
4106 #undef CHECK_APPEND_NOARG
4110 if (size_bp < size_buf)
4111 buf[size_buf - size_bp - 1] = '\0'; /* remove last , */
4116 static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info)
4118 u32 size_buf = 1024;
4119 char tmp_buf[192] = {'\0'};
4122 u32 size_bp = size_buf;
4124 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4126 buf = kzalloc(size_buf, GFP_KERNEL);
4132 #define CHECK_APPEND_1ARG(a, v1) \
4134 ret = snprintf(bp, size_bp, (a), (v1)); \
4135 if (ret < 0 || ret >= size_bp) \
4136 goto out_overflow; \
4141 if (bctl->flags & BTRFS_BALANCE_FORCE)
4142 CHECK_APPEND_1ARG("%s", "-f ");
4144 if (bctl->flags & BTRFS_BALANCE_DATA) {
4145 describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf));
4146 CHECK_APPEND_1ARG("-d%s ", tmp_buf);
4149 if (bctl->flags & BTRFS_BALANCE_METADATA) {
4150 describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf));
4151 CHECK_APPEND_1ARG("-m%s ", tmp_buf);
4154 if (bctl->flags & BTRFS_BALANCE_SYSTEM) {
4155 describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf));
4156 CHECK_APPEND_1ARG("-s%s ", tmp_buf);
4159 #undef CHECK_APPEND_1ARG
4163 if (size_bp < size_buf)
4164 buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */
4165 btrfs_info(fs_info, "balance: %s %s",
4166 (bctl->flags & BTRFS_BALANCE_RESUME) ?
4167 "resume" : "start", buf);
4173 * Should be called with balance mutexe held
4175 int btrfs_balance(struct btrfs_fs_info *fs_info,
4176 struct btrfs_balance_control *bctl,
4177 struct btrfs_ioctl_balance_args *bargs)
4179 u64 meta_target, data_target;
4185 bool reducing_redundancy;
4188 if (btrfs_fs_closing(fs_info) ||
4189 atomic_read(&fs_info->balance_pause_req) ||
4190 btrfs_should_cancel_balance(fs_info)) {
4195 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
4196 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
4200 * In case of mixed groups both data and meta should be picked,
4201 * and identical options should be given for both of them.
4203 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
4204 if (mixed && (bctl->flags & allowed)) {
4205 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
4206 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
4207 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
4209 "balance: mixed groups data and metadata options must be the same");
4216 * rw_devices will not change at the moment, device add/delete/replace
4219 num_devices = fs_info->fs_devices->rw_devices;
4222 * SINGLE profile on-disk has no profile bit, but in-memory we have a
4223 * special bit for it, to make it easier to distinguish. Thus we need
4224 * to set it manually, or balance would refuse the profile.
4226 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
4227 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++)
4228 if (num_devices >= btrfs_raid_array[i].devs_min)
4229 allowed |= btrfs_raid_array[i].bg_flag;
4231 if (!validate_convert_profile(fs_info, &bctl->data, allowed, "data") ||
4232 !validate_convert_profile(fs_info, &bctl->meta, allowed, "metadata") ||
4233 !validate_convert_profile(fs_info, &bctl->sys, allowed, "system")) {
4239 * Allow to reduce metadata or system integrity only if force set for
4240 * profiles with redundancy (copies, parity)
4243 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) {
4244 if (btrfs_raid_array[i].ncopies >= 2 ||
4245 btrfs_raid_array[i].tolerated_failures >= 1)
4246 allowed |= btrfs_raid_array[i].bg_flag;
4249 seq = read_seqbegin(&fs_info->profiles_lock);
4251 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4252 (fs_info->avail_system_alloc_bits & allowed) &&
4253 !(bctl->sys.target & allowed)) ||
4254 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4255 (fs_info->avail_metadata_alloc_bits & allowed) &&
4256 !(bctl->meta.target & allowed)))
4257 reducing_redundancy = true;
4259 reducing_redundancy = false;
4261 /* if we're not converting, the target field is uninitialized */
4262 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4263 bctl->meta.target : fs_info->avail_metadata_alloc_bits;
4264 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4265 bctl->data.target : fs_info->avail_data_alloc_bits;
4266 } while (read_seqretry(&fs_info->profiles_lock, seq));
4268 if (reducing_redundancy) {
4269 if (bctl->flags & BTRFS_BALANCE_FORCE) {
4271 "balance: force reducing metadata redundancy");
4274 "balance: reduces metadata redundancy, use --force if you want this");
4280 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <
4281 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {
4283 "balance: metadata profile %s has lower redundancy than data profile %s",
4284 btrfs_bg_type_to_raid_name(meta_target),
4285 btrfs_bg_type_to_raid_name(data_target));
4288 ret = insert_balance_item(fs_info, bctl);
4289 if (ret && ret != -EEXIST)
4292 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
4293 BUG_ON(ret == -EEXIST);
4294 BUG_ON(fs_info->balance_ctl);
4295 spin_lock(&fs_info->balance_lock);
4296 fs_info->balance_ctl = bctl;
4297 spin_unlock(&fs_info->balance_lock);
4299 BUG_ON(ret != -EEXIST);
4300 spin_lock(&fs_info->balance_lock);
4301 update_balance_args(bctl);
4302 spin_unlock(&fs_info->balance_lock);
4305 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4306 set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4307 describe_balance_start_or_resume(fs_info);
4308 mutex_unlock(&fs_info->balance_mutex);
4310 ret = __btrfs_balance(fs_info);
4312 mutex_lock(&fs_info->balance_mutex);
4313 if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req))
4314 btrfs_info(fs_info, "balance: paused");
4316 * Balance can be canceled by:
4318 * - Regular cancel request
4319 * Then ret == -ECANCELED and balance_cancel_req > 0
4321 * - Fatal signal to "btrfs" process
4322 * Either the signal caught by wait_reserve_ticket() and callers
4323 * got -EINTR, or caught by btrfs_should_cancel_balance() and
4325 * Either way, in this case balance_cancel_req = 0, and
4326 * ret == -EINTR or ret == -ECANCELED.
4328 * So here we only check the return value to catch canceled balance.
4330 else if (ret == -ECANCELED || ret == -EINTR)
4331 btrfs_info(fs_info, "balance: canceled");
4333 btrfs_info(fs_info, "balance: ended with status: %d", ret);
4335 clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4338 memset(bargs, 0, sizeof(*bargs));
4339 btrfs_update_ioctl_balance_args(fs_info, bargs);
4342 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
4343 balance_need_close(fs_info)) {
4344 reset_balance_state(fs_info);
4345 btrfs_exclop_finish(fs_info);
4348 wake_up(&fs_info->balance_wait_q);
4352 if (bctl->flags & BTRFS_BALANCE_RESUME)
4353 reset_balance_state(fs_info);
4356 btrfs_exclop_finish(fs_info);
4361 static int balance_kthread(void *data)
4363 struct btrfs_fs_info *fs_info = data;
4366 mutex_lock(&fs_info->balance_mutex);
4367 if (fs_info->balance_ctl)
4368 ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL);
4369 mutex_unlock(&fs_info->balance_mutex);
4374 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
4376 struct task_struct *tsk;
4378 mutex_lock(&fs_info->balance_mutex);
4379 if (!fs_info->balance_ctl) {
4380 mutex_unlock(&fs_info->balance_mutex);
4383 mutex_unlock(&fs_info->balance_mutex);
4385 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
4386 btrfs_info(fs_info, "balance: resume skipped");
4391 * A ro->rw remount sequence should continue with the paused balance
4392 * regardless of who pauses it, system or the user as of now, so set
4395 spin_lock(&fs_info->balance_lock);
4396 fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME;
4397 spin_unlock(&fs_info->balance_lock);
4399 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
4400 return PTR_ERR_OR_ZERO(tsk);
4403 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
4405 struct btrfs_balance_control *bctl;
4406 struct btrfs_balance_item *item;
4407 struct btrfs_disk_balance_args disk_bargs;
4408 struct btrfs_path *path;
4409 struct extent_buffer *leaf;
4410 struct btrfs_key key;
4413 path = btrfs_alloc_path();
4417 key.objectid = BTRFS_BALANCE_OBJECTID;
4418 key.type = BTRFS_TEMPORARY_ITEM_KEY;
4421 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4424 if (ret > 0) { /* ret = -ENOENT; */
4429 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4435 leaf = path->nodes[0];
4436 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
4438 bctl->flags = btrfs_balance_flags(leaf, item);
4439 bctl->flags |= BTRFS_BALANCE_RESUME;
4441 btrfs_balance_data(leaf, item, &disk_bargs);
4442 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4443 btrfs_balance_meta(leaf, item, &disk_bargs);
4444 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4445 btrfs_balance_sys(leaf, item, &disk_bargs);
4446 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4449 * This should never happen, as the paused balance state is recovered
4450 * during mount without any chance of other exclusive ops to collide.
4452 * This gives the exclusive op status to balance and keeps in paused
4453 * state until user intervention (cancel or umount). If the ownership
4454 * cannot be assigned, show a message but do not fail. The balance
4455 * is in a paused state and must have fs_info::balance_ctl properly
4458 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE))
4460 "balance: cannot set exclusive op status, resume manually");
4462 btrfs_release_path(path);
4464 mutex_lock(&fs_info->balance_mutex);
4465 BUG_ON(fs_info->balance_ctl);
4466 spin_lock(&fs_info->balance_lock);
4467 fs_info->balance_ctl = bctl;
4468 spin_unlock(&fs_info->balance_lock);
4469 mutex_unlock(&fs_info->balance_mutex);
4471 btrfs_free_path(path);
4475 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4479 mutex_lock(&fs_info->balance_mutex);
4480 if (!fs_info->balance_ctl) {
4481 mutex_unlock(&fs_info->balance_mutex);
4485 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4486 atomic_inc(&fs_info->balance_pause_req);
4487 mutex_unlock(&fs_info->balance_mutex);
4489 wait_event(fs_info->balance_wait_q,
4490 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4492 mutex_lock(&fs_info->balance_mutex);
4493 /* we are good with balance_ctl ripped off from under us */
4494 BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4495 atomic_dec(&fs_info->balance_pause_req);
4500 mutex_unlock(&fs_info->balance_mutex);
4504 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4506 mutex_lock(&fs_info->balance_mutex);
4507 if (!fs_info->balance_ctl) {
4508 mutex_unlock(&fs_info->balance_mutex);
4513 * A paused balance with the item stored on disk can be resumed at
4514 * mount time if the mount is read-write. Otherwise it's still paused
4515 * and we must not allow cancelling as it deletes the item.
4517 if (sb_rdonly(fs_info->sb)) {
4518 mutex_unlock(&fs_info->balance_mutex);
4522 atomic_inc(&fs_info->balance_cancel_req);
4524 * if we are running just wait and return, balance item is
4525 * deleted in btrfs_balance in this case
4527 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4528 mutex_unlock(&fs_info->balance_mutex);
4529 wait_event(fs_info->balance_wait_q,
4530 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4531 mutex_lock(&fs_info->balance_mutex);
4533 mutex_unlock(&fs_info->balance_mutex);
4535 * Lock released to allow other waiters to continue, we'll
4536 * reexamine the status again.
4538 mutex_lock(&fs_info->balance_mutex);
4540 if (fs_info->balance_ctl) {
4541 reset_balance_state(fs_info);
4542 btrfs_exclop_finish(fs_info);
4543 btrfs_info(fs_info, "balance: canceled");
4547 BUG_ON(fs_info->balance_ctl ||
4548 test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4549 atomic_dec(&fs_info->balance_cancel_req);
4550 mutex_unlock(&fs_info->balance_mutex);
4554 int btrfs_uuid_scan_kthread(void *data)
4556 struct btrfs_fs_info *fs_info = data;
4557 struct btrfs_root *root = fs_info->tree_root;
4558 struct btrfs_key key;
4559 struct btrfs_path *path = NULL;
4561 struct extent_buffer *eb;
4563 struct btrfs_root_item root_item;
4565 struct btrfs_trans_handle *trans = NULL;
4566 bool closing = false;
4568 path = btrfs_alloc_path();
4575 key.type = BTRFS_ROOT_ITEM_KEY;
4579 if (btrfs_fs_closing(fs_info)) {
4583 ret = btrfs_search_forward(root, &key, path,
4584 BTRFS_OLDEST_GENERATION);
4591 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4592 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4593 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4594 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4597 eb = path->nodes[0];
4598 slot = path->slots[0];
4599 item_size = btrfs_item_size_nr(eb, slot);
4600 if (item_size < sizeof(root_item))
4603 read_extent_buffer(eb, &root_item,
4604 btrfs_item_ptr_offset(eb, slot),
4605 (int)sizeof(root_item));
4606 if (btrfs_root_refs(&root_item) == 0)
4609 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4610 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4614 btrfs_release_path(path);
4616 * 1 - subvol uuid item
4617 * 1 - received_subvol uuid item
4619 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4620 if (IS_ERR(trans)) {
4621 ret = PTR_ERR(trans);
4629 btrfs_release_path(path);
4630 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4631 ret = btrfs_uuid_tree_add(trans, root_item.uuid,
4632 BTRFS_UUID_KEY_SUBVOL,
4635 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4641 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4642 ret = btrfs_uuid_tree_add(trans,
4643 root_item.received_uuid,
4644 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4647 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4654 btrfs_release_path(path);
4656 ret = btrfs_end_transaction(trans);
4662 if (key.offset < (u64)-1) {
4664 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4666 key.type = BTRFS_ROOT_ITEM_KEY;
4667 } else if (key.objectid < (u64)-1) {
4669 key.type = BTRFS_ROOT_ITEM_KEY;
4678 btrfs_free_path(path);
4679 if (trans && !IS_ERR(trans))
4680 btrfs_end_transaction(trans);
4682 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4684 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4685 up(&fs_info->uuid_tree_rescan_sem);
4689 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4691 struct btrfs_trans_handle *trans;
4692 struct btrfs_root *tree_root = fs_info->tree_root;
4693 struct btrfs_root *uuid_root;
4694 struct task_struct *task;
4701 trans = btrfs_start_transaction(tree_root, 2);
4703 return PTR_ERR(trans);
4705 uuid_root = btrfs_create_tree(trans, BTRFS_UUID_TREE_OBJECTID);
4706 if (IS_ERR(uuid_root)) {
4707 ret = PTR_ERR(uuid_root);
4708 btrfs_abort_transaction(trans, ret);
4709 btrfs_end_transaction(trans);
4713 fs_info->uuid_root = uuid_root;
4715 ret = btrfs_commit_transaction(trans);
4719 down(&fs_info->uuid_tree_rescan_sem);
4720 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4722 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4723 btrfs_warn(fs_info, "failed to start uuid_scan task");
4724 up(&fs_info->uuid_tree_rescan_sem);
4725 return PTR_ERR(task);
4732 * shrinking a device means finding all of the device extents past
4733 * the new size, and then following the back refs to the chunks.
4734 * The chunk relocation code actually frees the device extent
4736 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4738 struct btrfs_fs_info *fs_info = device->fs_info;
4739 struct btrfs_root *root = fs_info->dev_root;
4740 struct btrfs_trans_handle *trans;
4741 struct btrfs_dev_extent *dev_extent = NULL;
4742 struct btrfs_path *path;
4748 bool retried = false;
4749 struct extent_buffer *l;
4750 struct btrfs_key key;
4751 struct btrfs_super_block *super_copy = fs_info->super_copy;
4752 u64 old_total = btrfs_super_total_bytes(super_copy);
4753 u64 old_size = btrfs_device_get_total_bytes(device);
4757 new_size = round_down(new_size, fs_info->sectorsize);
4759 diff = round_down(old_size - new_size, fs_info->sectorsize);
4761 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4764 path = btrfs_alloc_path();
4768 path->reada = READA_BACK;
4770 trans = btrfs_start_transaction(root, 0);
4771 if (IS_ERR(trans)) {
4772 btrfs_free_path(path);
4773 return PTR_ERR(trans);
4776 mutex_lock(&fs_info->chunk_mutex);
4778 btrfs_device_set_total_bytes(device, new_size);
4779 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4780 device->fs_devices->total_rw_bytes -= diff;
4781 atomic64_sub(diff, &fs_info->free_chunk_space);
4785 * Once the device's size has been set to the new size, ensure all
4786 * in-memory chunks are synced to disk so that the loop below sees them
4787 * and relocates them accordingly.
4789 if (contains_pending_extent(device, &start, diff)) {
4790 mutex_unlock(&fs_info->chunk_mutex);
4791 ret = btrfs_commit_transaction(trans);
4795 mutex_unlock(&fs_info->chunk_mutex);
4796 btrfs_end_transaction(trans);
4800 key.objectid = device->devid;
4801 key.offset = (u64)-1;
4802 key.type = BTRFS_DEV_EXTENT_KEY;
4805 mutex_lock(&fs_info->reclaim_bgs_lock);
4806 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4808 mutex_unlock(&fs_info->reclaim_bgs_lock);
4812 ret = btrfs_previous_item(root, path, 0, key.type);
4814 mutex_unlock(&fs_info->reclaim_bgs_lock);
4818 btrfs_release_path(path);
4823 slot = path->slots[0];
4824 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4826 if (key.objectid != device->devid) {
4827 mutex_unlock(&fs_info->reclaim_bgs_lock);
4828 btrfs_release_path(path);
4832 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4833 length = btrfs_dev_extent_length(l, dev_extent);
4835 if (key.offset + length <= new_size) {
4836 mutex_unlock(&fs_info->reclaim_bgs_lock);
4837 btrfs_release_path(path);
4841 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4842 btrfs_release_path(path);
4845 * We may be relocating the only data chunk we have,
4846 * which could potentially end up with losing data's
4847 * raid profile, so lets allocate an empty one in
4850 ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset);
4852 mutex_unlock(&fs_info->reclaim_bgs_lock);
4856 ret = btrfs_relocate_chunk(fs_info, chunk_offset);
4857 mutex_unlock(&fs_info->reclaim_bgs_lock);
4858 if (ret == -ENOSPC) {
4861 if (ret == -ETXTBSY) {
4863 "could not shrink block group %llu due to active swapfile",
4868 } while (key.offset-- > 0);
4870 if (failed && !retried) {
4874 } else if (failed && retried) {
4879 /* Shrinking succeeded, else we would be at "done". */
4880 trans = btrfs_start_transaction(root, 0);
4881 if (IS_ERR(trans)) {
4882 ret = PTR_ERR(trans);
4886 mutex_lock(&fs_info->chunk_mutex);
4887 /* Clear all state bits beyond the shrunk device size */
4888 clear_extent_bits(&device->alloc_state, new_size, (u64)-1,
4891 btrfs_device_set_disk_total_bytes(device, new_size);
4892 if (list_empty(&device->post_commit_list))
4893 list_add_tail(&device->post_commit_list,
4894 &trans->transaction->dev_update_list);
4896 WARN_ON(diff > old_total);
4897 btrfs_set_super_total_bytes(super_copy,
4898 round_down(old_total - diff, fs_info->sectorsize));
4899 mutex_unlock(&fs_info->chunk_mutex);
4901 /* Now btrfs_update_device() will change the on-disk size. */
4902 ret = btrfs_update_device(trans, device);
4904 btrfs_abort_transaction(trans, ret);
4905 btrfs_end_transaction(trans);
4907 ret = btrfs_commit_transaction(trans);
4910 btrfs_free_path(path);
4912 mutex_lock(&fs_info->chunk_mutex);
4913 btrfs_device_set_total_bytes(device, old_size);
4914 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
4915 device->fs_devices->total_rw_bytes += diff;
4916 atomic64_add(diff, &fs_info->free_chunk_space);
4917 mutex_unlock(&fs_info->chunk_mutex);
4922 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
4923 struct btrfs_key *key,
4924 struct btrfs_chunk *chunk, int item_size)
4926 struct btrfs_super_block *super_copy = fs_info->super_copy;
4927 struct btrfs_disk_key disk_key;
4931 lockdep_assert_held(&fs_info->chunk_mutex);
4933 array_size = btrfs_super_sys_array_size(super_copy);
4934 if (array_size + item_size + sizeof(disk_key)
4935 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
4938 ptr = super_copy->sys_chunk_array + array_size;
4939 btrfs_cpu_key_to_disk(&disk_key, key);
4940 memcpy(ptr, &disk_key, sizeof(disk_key));
4941 ptr += sizeof(disk_key);
4942 memcpy(ptr, chunk, item_size);
4943 item_size += sizeof(disk_key);
4944 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4950 * sort the devices in descending order by max_avail, total_avail
4952 static int btrfs_cmp_device_info(const void *a, const void *b)
4954 const struct btrfs_device_info *di_a = a;
4955 const struct btrfs_device_info *di_b = b;
4957 if (di_a->max_avail > di_b->max_avail)
4959 if (di_a->max_avail < di_b->max_avail)
4961 if (di_a->total_avail > di_b->total_avail)
4963 if (di_a->total_avail < di_b->total_avail)
4968 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4970 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4973 btrfs_set_fs_incompat(info, RAID56);
4976 static void check_raid1c34_incompat_flag(struct btrfs_fs_info *info, u64 type)
4978 if (!(type & (BTRFS_BLOCK_GROUP_RAID1C3 | BTRFS_BLOCK_GROUP_RAID1C4)))
4981 btrfs_set_fs_incompat(info, RAID1C34);
4985 * Structure used internally for __btrfs_alloc_chunk() function.
4986 * Wraps needed parameters.
4988 struct alloc_chunk_ctl {
4991 /* Total number of stripes to allocate */
4993 /* sub_stripes info for map */
4995 /* Stripes per device */
4997 /* Maximum number of devices to use */
4999 /* Minimum number of devices to use */
5001 /* ndevs has to be a multiple of this */
5003 /* Number of copies */
5005 /* Number of stripes worth of bytes to store parity information */
5007 u64 max_stripe_size;
5015 static void init_alloc_chunk_ctl_policy_regular(
5016 struct btrfs_fs_devices *fs_devices,
5017 struct alloc_chunk_ctl *ctl)
5019 u64 type = ctl->type;
5021 if (type & BTRFS_BLOCK_GROUP_DATA) {
5022 ctl->max_stripe_size = SZ_1G;
5023 ctl->max_chunk_size = BTRFS_MAX_DATA_CHUNK_SIZE;
5024 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
5025 /* For larger filesystems, use larger metadata chunks */
5026 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
5027 ctl->max_stripe_size = SZ_1G;
5029 ctl->max_stripe_size = SZ_256M;
5030 ctl->max_chunk_size = ctl->max_stripe_size;
5031 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
5032 ctl->max_stripe_size = SZ_32M;
5033 ctl->max_chunk_size = 2 * ctl->max_stripe_size;
5034 ctl->devs_max = min_t(int, ctl->devs_max,
5035 BTRFS_MAX_DEVS_SYS_CHUNK);
5040 /* We don't want a chunk larger than 10% of writable space */
5041 ctl->max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
5042 ctl->max_chunk_size);
5043 ctl->dev_extent_min = BTRFS_STRIPE_LEN * ctl->dev_stripes;
5046 static void init_alloc_chunk_ctl_policy_zoned(
5047 struct btrfs_fs_devices *fs_devices,
5048 struct alloc_chunk_ctl *ctl)
5050 u64 zone_size = fs_devices->fs_info->zone_size;
5052 int min_num_stripes = ctl->devs_min * ctl->dev_stripes;
5053 int min_data_stripes = (min_num_stripes - ctl->nparity) / ctl->ncopies;
5054 u64 min_chunk_size = min_data_stripes * zone_size;
5055 u64 type = ctl->type;
5057 ctl->max_stripe_size = zone_size;
5058 if (type & BTRFS_BLOCK_GROUP_DATA) {
5059 ctl->max_chunk_size = round_down(BTRFS_MAX_DATA_CHUNK_SIZE,
5061 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
5062 ctl->max_chunk_size = ctl->max_stripe_size;
5063 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
5064 ctl->max_chunk_size = 2 * ctl->max_stripe_size;
5065 ctl->devs_max = min_t(int, ctl->devs_max,
5066 BTRFS_MAX_DEVS_SYS_CHUNK);
5071 /* We don't want a chunk larger than 10% of writable space */
5072 limit = max(round_down(div_factor(fs_devices->total_rw_bytes, 1),
5075 ctl->max_chunk_size = min(limit, ctl->max_chunk_size);
5076 ctl->dev_extent_min = zone_size * ctl->dev_stripes;
5079 static void init_alloc_chunk_ctl(struct btrfs_fs_devices *fs_devices,
5080 struct alloc_chunk_ctl *ctl)
5082 int index = btrfs_bg_flags_to_raid_index(ctl->type);
5084 ctl->sub_stripes = btrfs_raid_array[index].sub_stripes;
5085 ctl->dev_stripes = btrfs_raid_array[index].dev_stripes;
5086 ctl->devs_max = btrfs_raid_array[index].devs_max;
5088 ctl->devs_max = BTRFS_MAX_DEVS(fs_devices->fs_info);
5089 ctl->devs_min = btrfs_raid_array[index].devs_min;
5090 ctl->devs_increment = btrfs_raid_array[index].devs_increment;
5091 ctl->ncopies = btrfs_raid_array[index].ncopies;
5092 ctl->nparity = btrfs_raid_array[index].nparity;
5095 switch (fs_devices->chunk_alloc_policy) {
5096 case BTRFS_CHUNK_ALLOC_REGULAR:
5097 init_alloc_chunk_ctl_policy_regular(fs_devices, ctl);
5099 case BTRFS_CHUNK_ALLOC_ZONED:
5100 init_alloc_chunk_ctl_policy_zoned(fs_devices, ctl);
5107 static int gather_device_info(struct btrfs_fs_devices *fs_devices,
5108 struct alloc_chunk_ctl *ctl,
5109 struct btrfs_device_info *devices_info)
5111 struct btrfs_fs_info *info = fs_devices->fs_info;
5112 struct btrfs_device *device;
5114 u64 dev_extent_want = ctl->max_stripe_size * ctl->dev_stripes;
5121 * in the first pass through the devices list, we gather information
5122 * about the available holes on each device.
5124 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
5125 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
5127 "BTRFS: read-only device in alloc_list\n");
5131 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
5132 &device->dev_state) ||
5133 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
5136 if (device->total_bytes > device->bytes_used)
5137 total_avail = device->total_bytes - device->bytes_used;
5141 /* If there is no space on this device, skip it. */
5142 if (total_avail < ctl->dev_extent_min)
5145 ret = find_free_dev_extent(device, dev_extent_want, &dev_offset,
5147 if (ret && ret != -ENOSPC)
5151 max_avail = dev_extent_want;
5153 if (max_avail < ctl->dev_extent_min) {
5154 if (btrfs_test_opt(info, ENOSPC_DEBUG))
5156 "%s: devid %llu has no free space, have=%llu want=%llu",
5157 __func__, device->devid, max_avail,
5158 ctl->dev_extent_min);
5162 if (ndevs == fs_devices->rw_devices) {
5163 WARN(1, "%s: found more than %llu devices\n",
5164 __func__, fs_devices->rw_devices);
5167 devices_info[ndevs].dev_offset = dev_offset;
5168 devices_info[ndevs].max_avail = max_avail;
5169 devices_info[ndevs].total_avail = total_avail;
5170 devices_info[ndevs].dev = device;
5176 * now sort the devices by hole size / available space
5178 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
5179 btrfs_cmp_device_info, NULL);
5184 static int decide_stripe_size_regular(struct alloc_chunk_ctl *ctl,
5185 struct btrfs_device_info *devices_info)
5187 /* Number of stripes that count for block group size */
5191 * The primary goal is to maximize the number of stripes, so use as
5192 * many devices as possible, even if the stripes are not maximum sized.
5194 * The DUP profile stores more than one stripe per device, the
5195 * max_avail is the total size so we have to adjust.
5197 ctl->stripe_size = div_u64(devices_info[ctl->ndevs - 1].max_avail,
5199 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5201 /* This will have to be fixed for RAID1 and RAID10 over more drives */
5202 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5205 * Use the number of data stripes to figure out how big this chunk is
5206 * really going to be in terms of logical address space, and compare
5207 * that answer with the max chunk size. If it's higher, we try to
5208 * reduce stripe_size.
5210 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) {
5212 * Reduce stripe_size, round it up to a 16MB boundary again and
5213 * then use it, unless it ends up being even bigger than the
5214 * previous value we had already.
5216 ctl->stripe_size = min(round_up(div_u64(ctl->max_chunk_size,
5217 data_stripes), SZ_16M),
5221 /* Align to BTRFS_STRIPE_LEN */
5222 ctl->stripe_size = round_down(ctl->stripe_size, BTRFS_STRIPE_LEN);
5223 ctl->chunk_size = ctl->stripe_size * data_stripes;
5228 static int decide_stripe_size_zoned(struct alloc_chunk_ctl *ctl,
5229 struct btrfs_device_info *devices_info)
5231 u64 zone_size = devices_info[0].dev->zone_info->zone_size;
5232 /* Number of stripes that count for block group size */
5236 * It should hold because:
5237 * dev_extent_min == dev_extent_want == zone_size * dev_stripes
5239 ASSERT(devices_info[ctl->ndevs - 1].max_avail == ctl->dev_extent_min);
5241 ctl->stripe_size = zone_size;
5242 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5243 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5245 /* stripe_size is fixed in zoned filesysmte. Reduce ndevs instead. */
5246 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) {
5247 ctl->ndevs = div_u64(div_u64(ctl->max_chunk_size * ctl->ncopies,
5248 ctl->stripe_size) + ctl->nparity,
5250 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5251 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5252 ASSERT(ctl->stripe_size * data_stripes <= ctl->max_chunk_size);
5255 ctl->chunk_size = ctl->stripe_size * data_stripes;
5260 static int decide_stripe_size(struct btrfs_fs_devices *fs_devices,
5261 struct alloc_chunk_ctl *ctl,
5262 struct btrfs_device_info *devices_info)
5264 struct btrfs_fs_info *info = fs_devices->fs_info;
5267 * Round down to number of usable stripes, devs_increment can be any
5268 * number so we can't use round_down() that requires power of 2, while
5269 * rounddown is safe.
5271 ctl->ndevs = rounddown(ctl->ndevs, ctl->devs_increment);
5273 if (ctl->ndevs < ctl->devs_min) {
5274 if (btrfs_test_opt(info, ENOSPC_DEBUG)) {
5276 "%s: not enough devices with free space: have=%d minimum required=%d",
5277 __func__, ctl->ndevs, ctl->devs_min);
5282 ctl->ndevs = min(ctl->ndevs, ctl->devs_max);
5284 switch (fs_devices->chunk_alloc_policy) {
5285 case BTRFS_CHUNK_ALLOC_REGULAR:
5286 return decide_stripe_size_regular(ctl, devices_info);
5287 case BTRFS_CHUNK_ALLOC_ZONED:
5288 return decide_stripe_size_zoned(ctl, devices_info);
5294 static struct btrfs_block_group *create_chunk(struct btrfs_trans_handle *trans,
5295 struct alloc_chunk_ctl *ctl,
5296 struct btrfs_device_info *devices_info)
5298 struct btrfs_fs_info *info = trans->fs_info;
5299 struct map_lookup *map = NULL;
5300 struct extent_map_tree *em_tree;
5301 struct btrfs_block_group *block_group;
5302 struct extent_map *em;
5303 u64 start = ctl->start;
5304 u64 type = ctl->type;
5309 map = kmalloc(map_lookup_size(ctl->num_stripes), GFP_NOFS);
5311 return ERR_PTR(-ENOMEM);
5312 map->num_stripes = ctl->num_stripes;
5314 for (i = 0; i < ctl->ndevs; ++i) {
5315 for (j = 0; j < ctl->dev_stripes; ++j) {
5316 int s = i * ctl->dev_stripes + j;
5317 map->stripes[s].dev = devices_info[i].dev;
5318 map->stripes[s].physical = devices_info[i].dev_offset +
5319 j * ctl->stripe_size;
5322 map->stripe_len = BTRFS_STRIPE_LEN;
5323 map->io_align = BTRFS_STRIPE_LEN;
5324 map->io_width = BTRFS_STRIPE_LEN;
5326 map->sub_stripes = ctl->sub_stripes;
5328 trace_btrfs_chunk_alloc(info, map, start, ctl->chunk_size);
5330 em = alloc_extent_map();
5333 return ERR_PTR(-ENOMEM);
5335 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
5336 em->map_lookup = map;
5338 em->len = ctl->chunk_size;
5339 em->block_start = 0;
5340 em->block_len = em->len;
5341 em->orig_block_len = ctl->stripe_size;
5343 em_tree = &info->mapping_tree;
5344 write_lock(&em_tree->lock);
5345 ret = add_extent_mapping(em_tree, em, 0);
5347 write_unlock(&em_tree->lock);
5348 free_extent_map(em);
5349 return ERR_PTR(ret);
5351 write_unlock(&em_tree->lock);
5353 block_group = btrfs_make_block_group(trans, 0, type, start, ctl->chunk_size);
5354 if (IS_ERR(block_group))
5355 goto error_del_extent;
5357 for (i = 0; i < map->num_stripes; i++) {
5358 struct btrfs_device *dev = map->stripes[i].dev;
5360 btrfs_device_set_bytes_used(dev,
5361 dev->bytes_used + ctl->stripe_size);
5362 if (list_empty(&dev->post_commit_list))
5363 list_add_tail(&dev->post_commit_list,
5364 &trans->transaction->dev_update_list);
5367 atomic64_sub(ctl->stripe_size * map->num_stripes,
5368 &info->free_chunk_space);
5370 free_extent_map(em);
5371 check_raid56_incompat_flag(info, type);
5372 check_raid1c34_incompat_flag(info, type);
5377 write_lock(&em_tree->lock);
5378 remove_extent_mapping(em_tree, em);
5379 write_unlock(&em_tree->lock);
5381 /* One for our allocation */
5382 free_extent_map(em);
5383 /* One for the tree reference */
5384 free_extent_map(em);
5389 struct btrfs_block_group *btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
5392 struct btrfs_fs_info *info = trans->fs_info;
5393 struct btrfs_fs_devices *fs_devices = info->fs_devices;
5394 struct btrfs_device_info *devices_info = NULL;
5395 struct alloc_chunk_ctl ctl;
5396 struct btrfs_block_group *block_group;
5399 lockdep_assert_held(&info->chunk_mutex);
5401 if (!alloc_profile_is_valid(type, 0)) {
5403 return ERR_PTR(-EINVAL);
5406 if (list_empty(&fs_devices->alloc_list)) {
5407 if (btrfs_test_opt(info, ENOSPC_DEBUG))
5408 btrfs_debug(info, "%s: no writable device", __func__);
5409 return ERR_PTR(-ENOSPC);
5412 if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
5413 btrfs_err(info, "invalid chunk type 0x%llx requested", type);
5415 return ERR_PTR(-EINVAL);
5418 ctl.start = find_next_chunk(info);
5420 init_alloc_chunk_ctl(fs_devices, &ctl);
5422 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
5425 return ERR_PTR(-ENOMEM);
5427 ret = gather_device_info(fs_devices, &ctl, devices_info);
5429 block_group = ERR_PTR(ret);
5433 ret = decide_stripe_size(fs_devices, &ctl, devices_info);
5435 block_group = ERR_PTR(ret);
5439 block_group = create_chunk(trans, &ctl, devices_info);
5442 kfree(devices_info);
5447 * This function, btrfs_chunk_alloc_add_chunk_item(), typically belongs to the
5448 * phase 1 of chunk allocation. It belongs to phase 2 only when allocating system
5451 * See the comment at btrfs_chunk_alloc() for details about the chunk allocation
5454 int btrfs_chunk_alloc_add_chunk_item(struct btrfs_trans_handle *trans,
5455 struct btrfs_block_group *bg)
5457 struct btrfs_fs_info *fs_info = trans->fs_info;
5458 struct btrfs_root *extent_root = fs_info->extent_root;
5459 struct btrfs_root *chunk_root = fs_info->chunk_root;
5460 struct btrfs_key key;
5461 struct btrfs_chunk *chunk;
5462 struct btrfs_stripe *stripe;
5463 struct extent_map *em;
5464 struct map_lookup *map;
5470 * We take the chunk_mutex for 2 reasons:
5472 * 1) Updates and insertions in the chunk btree must be done while holding
5473 * the chunk_mutex, as well as updating the system chunk array in the
5474 * superblock. See the comment on top of btrfs_chunk_alloc() for the
5477 * 2) To prevent races with the final phase of a device replace operation
5478 * that replaces the device object associated with the map's stripes,
5479 * because the device object's id can change at any time during that
5480 * final phase of the device replace operation
5481 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the
5482 * replaced device and then see it with an ID of BTRFS_DEV_REPLACE_DEVID,
5483 * which would cause a failure when updating the device item, which does
5484 * not exists, or persisting a stripe of the chunk item with such ID.
5485 * Here we can't use the device_list_mutex because our caller already
5486 * has locked the chunk_mutex, and the final phase of device replace
5487 * acquires both mutexes - first the device_list_mutex and then the
5488 * chunk_mutex. Using any of those two mutexes protects us from a
5489 * concurrent device replace.
5491 lockdep_assert_held(&fs_info->chunk_mutex);
5493 em = btrfs_get_chunk_map(fs_info, bg->start, bg->length);
5496 btrfs_abort_transaction(trans, ret);
5500 map = em->map_lookup;
5501 item_size = btrfs_chunk_item_size(map->num_stripes);
5503 chunk = kzalloc(item_size, GFP_NOFS);
5506 btrfs_abort_transaction(trans, ret);
5510 for (i = 0; i < map->num_stripes; i++) {
5511 struct btrfs_device *device = map->stripes[i].dev;
5513 ret = btrfs_update_device(trans, device);
5518 stripe = &chunk->stripe;
5519 for (i = 0; i < map->num_stripes; i++) {
5520 struct btrfs_device *device = map->stripes[i].dev;
5521 const u64 dev_offset = map->stripes[i].physical;
5523 btrfs_set_stack_stripe_devid(stripe, device->devid);
5524 btrfs_set_stack_stripe_offset(stripe, dev_offset);
5525 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
5529 btrfs_set_stack_chunk_length(chunk, bg->length);
5530 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
5531 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
5532 btrfs_set_stack_chunk_type(chunk, map->type);
5533 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
5534 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
5535 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
5536 btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
5537 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
5539 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
5540 key.type = BTRFS_CHUNK_ITEM_KEY;
5541 key.offset = bg->start;
5543 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
5547 bg->chunk_item_inserted = 1;
5549 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
5550 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
5557 free_extent_map(em);
5561 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans)
5563 struct btrfs_fs_info *fs_info = trans->fs_info;
5565 struct btrfs_block_group *meta_bg;
5566 struct btrfs_block_group *sys_bg;
5569 * When adding a new device for sprouting, the seed device is read-only
5570 * so we must first allocate a metadata and a system chunk. But before
5571 * adding the block group items to the extent, device and chunk btrees,
5574 * 1) Create both chunks without doing any changes to the btrees, as
5575 * otherwise we would get -ENOSPC since the block groups from the
5576 * seed device are read-only;
5578 * 2) Add the device item for the new sprout device - finishing the setup
5579 * of a new block group requires updating the device item in the chunk
5580 * btree, so it must exist when we attempt to do it. The previous step
5581 * ensures this does not fail with -ENOSPC.
5583 * After that we can add the block group items to their btrees:
5584 * update existing device item in the chunk btree, add a new block group
5585 * item to the extent btree, add a new chunk item to the chunk btree and
5586 * finally add the new device extent items to the devices btree.
5589 alloc_profile = btrfs_metadata_alloc_profile(fs_info);
5590 meta_bg = btrfs_alloc_chunk(trans, alloc_profile);
5591 if (IS_ERR(meta_bg))
5592 return PTR_ERR(meta_bg);
5594 alloc_profile = btrfs_system_alloc_profile(fs_info);
5595 sys_bg = btrfs_alloc_chunk(trans, alloc_profile);
5597 return PTR_ERR(sys_bg);
5602 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
5604 const int index = btrfs_bg_flags_to_raid_index(map->type);
5606 return btrfs_raid_array[index].tolerated_failures;
5609 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset)
5611 struct extent_map *em;
5612 struct map_lookup *map;
5617 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
5621 map = em->map_lookup;
5622 for (i = 0; i < map->num_stripes; i++) {
5623 if (test_bit(BTRFS_DEV_STATE_MISSING,
5624 &map->stripes[i].dev->dev_state)) {
5628 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE,
5629 &map->stripes[i].dev->dev_state)) {
5636 * If the number of missing devices is larger than max errors,
5637 * we can not write the data into that chunk successfully, so
5640 if (miss_ndevs > btrfs_chunk_max_errors(map))
5643 free_extent_map(em);
5647 void btrfs_mapping_tree_free(struct extent_map_tree *tree)
5649 struct extent_map *em;
5652 write_lock(&tree->lock);
5653 em = lookup_extent_mapping(tree, 0, (u64)-1);
5655 remove_extent_mapping(tree, em);
5656 write_unlock(&tree->lock);
5660 free_extent_map(em);
5661 /* once for the tree */
5662 free_extent_map(em);
5666 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5668 struct extent_map *em;
5669 struct map_lookup *map;
5672 em = btrfs_get_chunk_map(fs_info, logical, len);
5675 * We could return errors for these cases, but that could get
5676 * ugly and we'd probably do the same thing which is just not do
5677 * anything else and exit, so return 1 so the callers don't try
5678 * to use other copies.
5682 map = em->map_lookup;
5683 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1_MASK))
5684 ret = map->num_stripes;
5685 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5686 ret = map->sub_stripes;
5687 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5689 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5691 * There could be two corrupted data stripes, we need
5692 * to loop retry in order to rebuild the correct data.
5694 * Fail a stripe at a time on every retry except the
5695 * stripe under reconstruction.
5697 ret = map->num_stripes;
5700 free_extent_map(em);
5702 down_read(&fs_info->dev_replace.rwsem);
5703 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) &&
5704 fs_info->dev_replace.tgtdev)
5706 up_read(&fs_info->dev_replace.rwsem);
5711 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
5714 struct extent_map *em;
5715 struct map_lookup *map;
5716 unsigned long len = fs_info->sectorsize;
5718 em = btrfs_get_chunk_map(fs_info, logical, len);
5720 if (!WARN_ON(IS_ERR(em))) {
5721 map = em->map_lookup;
5722 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5723 len = map->stripe_len * nr_data_stripes(map);
5724 free_extent_map(em);
5729 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5731 struct extent_map *em;
5732 struct map_lookup *map;
5735 em = btrfs_get_chunk_map(fs_info, logical, len);
5737 if(!WARN_ON(IS_ERR(em))) {
5738 map = em->map_lookup;
5739 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5741 free_extent_map(em);
5746 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5747 struct map_lookup *map, int first,
5748 int dev_replace_is_ongoing)
5752 int preferred_mirror;
5754 struct btrfs_device *srcdev;
5757 (BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10)));
5759 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5760 num_stripes = map->sub_stripes;
5762 num_stripes = map->num_stripes;
5764 switch (fs_info->fs_devices->read_policy) {
5766 /* Shouldn't happen, just warn and use pid instead of failing */
5767 btrfs_warn_rl(fs_info,
5768 "unknown read_policy type %u, reset to pid",
5769 fs_info->fs_devices->read_policy);
5770 fs_info->fs_devices->read_policy = BTRFS_READ_POLICY_PID;
5772 case BTRFS_READ_POLICY_PID:
5773 preferred_mirror = first + (current->pid % num_stripes);
5777 if (dev_replace_is_ongoing &&
5778 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5779 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5780 srcdev = fs_info->dev_replace.srcdev;
5785 * try to avoid the drive that is the source drive for a
5786 * dev-replace procedure, only choose it if no other non-missing
5787 * mirror is available
5789 for (tolerance = 0; tolerance < 2; tolerance++) {
5790 if (map->stripes[preferred_mirror].dev->bdev &&
5791 (tolerance || map->stripes[preferred_mirror].dev != srcdev))
5792 return preferred_mirror;
5793 for (i = first; i < first + num_stripes; i++) {
5794 if (map->stripes[i].dev->bdev &&
5795 (tolerance || map->stripes[i].dev != srcdev))
5800 /* we couldn't find one that doesn't fail. Just return something
5801 * and the io error handling code will clean up eventually
5803 return preferred_mirror;
5806 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5807 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5814 for (i = 0; i < num_stripes - 1; i++) {
5815 /* Swap if parity is on a smaller index */
5816 if (bbio->raid_map[i] > bbio->raid_map[i + 1]) {
5817 swap(bbio->stripes[i], bbio->stripes[i + 1]);
5818 swap(bbio->raid_map[i], bbio->raid_map[i + 1]);
5825 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5827 struct btrfs_bio *bbio = kzalloc(
5828 /* the size of the btrfs_bio */
5829 sizeof(struct btrfs_bio) +
5830 /* plus the variable array for the stripes */
5831 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5832 /* plus the variable array for the tgt dev */
5833 sizeof(int) * (real_stripes) +
5835 * plus the raid_map, which includes both the tgt dev
5838 sizeof(u64) * (total_stripes),
5839 GFP_NOFS|__GFP_NOFAIL);
5841 atomic_set(&bbio->error, 0);
5842 refcount_set(&bbio->refs, 1);
5844 bbio->tgtdev_map = (int *)(bbio->stripes + total_stripes);
5845 bbio->raid_map = (u64 *)(bbio->tgtdev_map + real_stripes);
5850 void btrfs_get_bbio(struct btrfs_bio *bbio)
5852 WARN_ON(!refcount_read(&bbio->refs));
5853 refcount_inc(&bbio->refs);
5856 void btrfs_put_bbio(struct btrfs_bio *bbio)
5860 if (refcount_dec_and_test(&bbio->refs))
5864 /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */
5866 * Please note that, discard won't be sent to target device of device
5869 static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info,
5870 u64 logical, u64 *length_ret,
5871 struct btrfs_bio **bbio_ret)
5873 struct extent_map *em;
5874 struct map_lookup *map;
5875 struct btrfs_bio *bbio;
5876 u64 length = *length_ret;
5880 u64 stripe_end_offset;
5887 u32 sub_stripes = 0;
5888 u64 stripes_per_dev = 0;
5889 u32 remaining_stripes = 0;
5890 u32 last_stripe = 0;
5894 /* discard always return a bbio */
5897 em = btrfs_get_chunk_map(fs_info, logical, length);
5901 map = em->map_lookup;
5902 /* we don't discard raid56 yet */
5903 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5908 offset = logical - em->start;
5909 length = min_t(u64, em->start + em->len - logical, length);
5910 *length_ret = length;
5912 stripe_len = map->stripe_len;
5914 * stripe_nr counts the total number of stripes we have to stride
5915 * to get to this block
5917 stripe_nr = div64_u64(offset, stripe_len);
5919 /* stripe_offset is the offset of this block in its stripe */
5920 stripe_offset = offset - stripe_nr * stripe_len;
5922 stripe_nr_end = round_up(offset + length, map->stripe_len);
5923 stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len);
5924 stripe_cnt = stripe_nr_end - stripe_nr;
5925 stripe_end_offset = stripe_nr_end * map->stripe_len -
5928 * after this, stripe_nr is the number of stripes on this
5929 * device we have to walk to find the data, and stripe_index is
5930 * the number of our device in the stripe array
5934 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5935 BTRFS_BLOCK_GROUP_RAID10)) {
5936 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5939 sub_stripes = map->sub_stripes;
5941 factor = map->num_stripes / sub_stripes;
5942 num_stripes = min_t(u64, map->num_stripes,
5943 sub_stripes * stripe_cnt);
5944 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5945 stripe_index *= sub_stripes;
5946 stripes_per_dev = div_u64_rem(stripe_cnt, factor,
5947 &remaining_stripes);
5948 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5949 last_stripe *= sub_stripes;
5950 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1_MASK |
5951 BTRFS_BLOCK_GROUP_DUP)) {
5952 num_stripes = map->num_stripes;
5954 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5958 bbio = alloc_btrfs_bio(num_stripes, 0);
5964 for (i = 0; i < num_stripes; i++) {
5965 bbio->stripes[i].physical =
5966 map->stripes[stripe_index].physical +
5967 stripe_offset + stripe_nr * map->stripe_len;
5968 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5970 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5971 BTRFS_BLOCK_GROUP_RAID10)) {
5972 bbio->stripes[i].length = stripes_per_dev *
5975 if (i / sub_stripes < remaining_stripes)
5976 bbio->stripes[i].length +=
5980 * Special for the first stripe and
5983 * |-------|...|-------|
5987 if (i < sub_stripes)
5988 bbio->stripes[i].length -=
5991 if (stripe_index >= last_stripe &&
5992 stripe_index <= (last_stripe +
5994 bbio->stripes[i].length -=
5997 if (i == sub_stripes - 1)
6000 bbio->stripes[i].length = length;
6004 if (stripe_index == map->num_stripes) {
6011 bbio->map_type = map->type;
6012 bbio->num_stripes = num_stripes;
6014 free_extent_map(em);
6019 * In dev-replace case, for repair case (that's the only case where the mirror
6020 * is selected explicitly when calling btrfs_map_block), blocks left of the
6021 * left cursor can also be read from the target drive.
6023 * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the
6025 * For READ, it also needs to be supported using the same mirror number.
6027 * If the requested block is not left of the left cursor, EIO is returned. This
6028 * can happen because btrfs_num_copies() returns one more in the dev-replace
6031 static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info,
6032 u64 logical, u64 length,
6033 u64 srcdev_devid, int *mirror_num,
6036 struct btrfs_bio *bbio = NULL;
6038 int index_srcdev = 0;
6040 u64 physical_of_found = 0;
6044 ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS,
6045 logical, &length, &bbio, 0, 0);
6047 ASSERT(bbio == NULL);
6051 num_stripes = bbio->num_stripes;
6052 if (*mirror_num > num_stripes) {
6054 * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror,
6055 * that means that the requested area is not left of the left
6058 btrfs_put_bbio(bbio);
6063 * process the rest of the function using the mirror_num of the source
6064 * drive. Therefore look it up first. At the end, patch the device
6065 * pointer to the one of the target drive.
6067 for (i = 0; i < num_stripes; i++) {
6068 if (bbio->stripes[i].dev->devid != srcdev_devid)
6072 * In case of DUP, in order to keep it simple, only add the
6073 * mirror with the lowest physical address
6076 physical_of_found <= bbio->stripes[i].physical)
6081 physical_of_found = bbio->stripes[i].physical;
6084 btrfs_put_bbio(bbio);
6090 *mirror_num = index_srcdev + 1;
6091 *physical = physical_of_found;
6095 static bool is_block_group_to_copy(struct btrfs_fs_info *fs_info, u64 logical)
6097 struct btrfs_block_group *cache;
6100 /* Non zoned filesystem does not use "to_copy" flag */
6101 if (!btrfs_is_zoned(fs_info))
6104 cache = btrfs_lookup_block_group(fs_info, logical);
6106 spin_lock(&cache->lock);
6107 ret = cache->to_copy;
6108 spin_unlock(&cache->lock);
6110 btrfs_put_block_group(cache);
6114 static void handle_ops_on_dev_replace(enum btrfs_map_op op,
6115 struct btrfs_bio **bbio_ret,
6116 struct btrfs_dev_replace *dev_replace,
6118 int *num_stripes_ret, int *max_errors_ret)
6120 struct btrfs_bio *bbio = *bbio_ret;
6121 u64 srcdev_devid = dev_replace->srcdev->devid;
6122 int tgtdev_indexes = 0;
6123 int num_stripes = *num_stripes_ret;
6124 int max_errors = *max_errors_ret;
6127 if (op == BTRFS_MAP_WRITE) {
6128 int index_where_to_add;
6131 * A block group which have "to_copy" set will eventually
6132 * copied by dev-replace process. We can avoid cloning IO here.
6134 if (is_block_group_to_copy(dev_replace->srcdev->fs_info, logical))
6138 * duplicate the write operations while the dev replace
6139 * procedure is running. Since the copying of the old disk to
6140 * the new disk takes place at run time while the filesystem is
6141 * mounted writable, the regular write operations to the old
6142 * disk have to be duplicated to go to the new disk as well.
6144 * Note that device->missing is handled by the caller, and that
6145 * the write to the old disk is already set up in the stripes
6148 index_where_to_add = num_stripes;
6149 for (i = 0; i < num_stripes; i++) {
6150 if (bbio->stripes[i].dev->devid == srcdev_devid) {
6151 /* write to new disk, too */
6152 struct btrfs_bio_stripe *new =
6153 bbio->stripes + index_where_to_add;
6154 struct btrfs_bio_stripe *old =
6157 new->physical = old->physical;
6158 new->length = old->length;
6159 new->dev = dev_replace->tgtdev;
6160 bbio->tgtdev_map[i] = index_where_to_add;
6161 index_where_to_add++;
6166 num_stripes = index_where_to_add;
6167 } else if (op == BTRFS_MAP_GET_READ_MIRRORS) {
6168 int index_srcdev = 0;
6170 u64 physical_of_found = 0;
6173 * During the dev-replace procedure, the target drive can also
6174 * be used to read data in case it is needed to repair a corrupt
6175 * block elsewhere. This is possible if the requested area is
6176 * left of the left cursor. In this area, the target drive is a
6177 * full copy of the source drive.
6179 for (i = 0; i < num_stripes; i++) {
6180 if (bbio->stripes[i].dev->devid == srcdev_devid) {
6182 * In case of DUP, in order to keep it simple,
6183 * only add the mirror with the lowest physical
6187 physical_of_found <=
6188 bbio->stripes[i].physical)
6192 physical_of_found = bbio->stripes[i].physical;
6196 struct btrfs_bio_stripe *tgtdev_stripe =
6197 bbio->stripes + num_stripes;
6199 tgtdev_stripe->physical = physical_of_found;
6200 tgtdev_stripe->length =
6201 bbio->stripes[index_srcdev].length;
6202 tgtdev_stripe->dev = dev_replace->tgtdev;
6203 bbio->tgtdev_map[index_srcdev] = num_stripes;
6210 *num_stripes_ret = num_stripes;
6211 *max_errors_ret = max_errors;
6212 bbio->num_tgtdevs = tgtdev_indexes;
6216 static bool need_full_stripe(enum btrfs_map_op op)
6218 return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS);
6222 * Calculate the geometry of a particular (address, len) tuple. This
6223 * information is used to calculate how big a particular bio can get before it
6224 * straddles a stripe.
6226 * @fs_info: the filesystem
6227 * @em: mapping containing the logical extent
6228 * @op: type of operation - write or read
6229 * @logical: address that we want to figure out the geometry of
6230 * @io_geom: pointer used to return values
6232 * Returns < 0 in case a chunk for the given logical address cannot be found,
6233 * usually shouldn't happen unless @logical is corrupted, 0 otherwise.
6235 int btrfs_get_io_geometry(struct btrfs_fs_info *fs_info, struct extent_map *em,
6236 enum btrfs_map_op op, u64 logical,
6237 struct btrfs_io_geometry *io_geom)
6239 struct map_lookup *map;
6245 u64 raid56_full_stripe_start = (u64)-1;
6248 ASSERT(op != BTRFS_MAP_DISCARD);
6250 map = em->map_lookup;
6251 /* Offset of this logical address in the chunk */
6252 offset = logical - em->start;
6253 /* Len of a stripe in a chunk */
6254 stripe_len = map->stripe_len;
6255 /* Stripe where this block falls in */
6256 stripe_nr = div64_u64(offset, stripe_len);
6257 /* Offset of stripe in the chunk */
6258 stripe_offset = stripe_nr * stripe_len;
6259 if (offset < stripe_offset) {
6261 "stripe math has gone wrong, stripe_offset=%llu offset=%llu start=%llu logical=%llu stripe_len=%llu",
6262 stripe_offset, offset, em->start, logical, stripe_len);
6266 /* stripe_offset is the offset of this block in its stripe */
6267 stripe_offset = offset - stripe_offset;
6268 data_stripes = nr_data_stripes(map);
6270 if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
6271 u64 max_len = stripe_len - stripe_offset;
6274 * In case of raid56, we need to know the stripe aligned start
6276 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6277 unsigned long full_stripe_len = stripe_len * data_stripes;
6278 raid56_full_stripe_start = offset;
6281 * Allow a write of a full stripe, but make sure we
6282 * don't allow straddling of stripes
6284 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
6286 raid56_full_stripe_start *= full_stripe_len;
6289 * For writes to RAID[56], allow a full stripeset across
6290 * all disks. For other RAID types and for RAID[56]
6291 * reads, just allow a single stripe (on a single disk).
6293 if (op == BTRFS_MAP_WRITE) {
6294 max_len = stripe_len * data_stripes -
6295 (offset - raid56_full_stripe_start);
6298 len = min_t(u64, em->len - offset, max_len);
6300 len = em->len - offset;
6304 io_geom->offset = offset;
6305 io_geom->stripe_len = stripe_len;
6306 io_geom->stripe_nr = stripe_nr;
6307 io_geom->stripe_offset = stripe_offset;
6308 io_geom->raid56_stripe_offset = raid56_full_stripe_start;
6313 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
6314 enum btrfs_map_op op,
6315 u64 logical, u64 *length,
6316 struct btrfs_bio **bbio_ret,
6317 int mirror_num, int need_raid_map)
6319 struct extent_map *em;
6320 struct map_lookup *map;
6330 int tgtdev_indexes = 0;
6331 struct btrfs_bio *bbio = NULL;
6332 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
6333 int dev_replace_is_ongoing = 0;
6334 int num_alloc_stripes;
6335 int patch_the_first_stripe_for_dev_replace = 0;
6336 u64 physical_to_patch_in_first_stripe = 0;
6337 u64 raid56_full_stripe_start = (u64)-1;
6338 struct btrfs_io_geometry geom;
6341 ASSERT(op != BTRFS_MAP_DISCARD);
6343 em = btrfs_get_chunk_map(fs_info, logical, *length);
6344 ASSERT(!IS_ERR(em));
6346 ret = btrfs_get_io_geometry(fs_info, em, op, logical, &geom);
6350 map = em->map_lookup;
6353 stripe_len = geom.stripe_len;
6354 stripe_nr = geom.stripe_nr;
6355 stripe_offset = geom.stripe_offset;
6356 raid56_full_stripe_start = geom.raid56_stripe_offset;
6357 data_stripes = nr_data_stripes(map);
6359 down_read(&dev_replace->rwsem);
6360 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
6362 * Hold the semaphore for read during the whole operation, write is
6363 * requested at commit time but must wait.
6365 if (!dev_replace_is_ongoing)
6366 up_read(&dev_replace->rwsem);
6368 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
6369 !need_full_stripe(op) && dev_replace->tgtdev != NULL) {
6370 ret = get_extra_mirror_from_replace(fs_info, logical, *length,
6371 dev_replace->srcdev->devid,
6373 &physical_to_patch_in_first_stripe);
6377 patch_the_first_stripe_for_dev_replace = 1;
6378 } else if (mirror_num > map->num_stripes) {
6384 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
6385 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
6387 if (!need_full_stripe(op))
6389 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1_MASK) {
6390 if (need_full_stripe(op))
6391 num_stripes = map->num_stripes;
6392 else if (mirror_num)
6393 stripe_index = mirror_num - 1;
6395 stripe_index = find_live_mirror(fs_info, map, 0,
6396 dev_replace_is_ongoing);
6397 mirror_num = stripe_index + 1;
6400 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
6401 if (need_full_stripe(op)) {
6402 num_stripes = map->num_stripes;
6403 } else if (mirror_num) {
6404 stripe_index = mirror_num - 1;
6409 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
6410 u32 factor = map->num_stripes / map->sub_stripes;
6412 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
6413 stripe_index *= map->sub_stripes;
6415 if (need_full_stripe(op))
6416 num_stripes = map->sub_stripes;
6417 else if (mirror_num)
6418 stripe_index += mirror_num - 1;
6420 int old_stripe_index = stripe_index;
6421 stripe_index = find_live_mirror(fs_info, map,
6423 dev_replace_is_ongoing);
6424 mirror_num = stripe_index - old_stripe_index + 1;
6427 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6428 if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) {
6429 /* push stripe_nr back to the start of the full stripe */
6430 stripe_nr = div64_u64(raid56_full_stripe_start,
6431 stripe_len * data_stripes);
6433 /* RAID[56] write or recovery. Return all stripes */
6434 num_stripes = map->num_stripes;
6435 max_errors = nr_parity_stripes(map);
6437 *length = map->stripe_len;
6442 * Mirror #0 or #1 means the original data block.
6443 * Mirror #2 is RAID5 parity block.
6444 * Mirror #3 is RAID6 Q block.
6446 stripe_nr = div_u64_rem(stripe_nr,
6447 data_stripes, &stripe_index);
6449 stripe_index = data_stripes + mirror_num - 2;
6451 /* We distribute the parity blocks across stripes */
6452 div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
6454 if (!need_full_stripe(op) && mirror_num <= 1)
6459 * after this, stripe_nr is the number of stripes on this
6460 * device we have to walk to find the data, and stripe_index is
6461 * the number of our device in the stripe array
6463 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
6465 mirror_num = stripe_index + 1;
6467 if (stripe_index >= map->num_stripes) {
6469 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
6470 stripe_index, map->num_stripes);
6475 num_alloc_stripes = num_stripes;
6476 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) {
6477 if (op == BTRFS_MAP_WRITE)
6478 num_alloc_stripes <<= 1;
6479 if (op == BTRFS_MAP_GET_READ_MIRRORS)
6480 num_alloc_stripes++;
6481 tgtdev_indexes = num_stripes;
6484 bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
6490 for (i = 0; i < num_stripes; i++) {
6491 bbio->stripes[i].physical = map->stripes[stripe_index].physical +
6492 stripe_offset + stripe_nr * map->stripe_len;
6493 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
6497 /* build raid_map */
6498 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map &&
6499 (need_full_stripe(op) || mirror_num > 1)) {
6503 /* Work out the disk rotation on this stripe-set */
6504 div_u64_rem(stripe_nr, num_stripes, &rot);
6506 /* Fill in the logical address of each stripe */
6507 tmp = stripe_nr * data_stripes;
6508 for (i = 0; i < data_stripes; i++)
6509 bbio->raid_map[(i+rot) % num_stripes] =
6510 em->start + (tmp + i) * map->stripe_len;
6512 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
6513 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
6514 bbio->raid_map[(i+rot+1) % num_stripes] =
6517 sort_parity_stripes(bbio, num_stripes);
6520 if (need_full_stripe(op))
6521 max_errors = btrfs_chunk_max_errors(map);
6523 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
6524 need_full_stripe(op)) {
6525 handle_ops_on_dev_replace(op, &bbio, dev_replace, logical,
6526 &num_stripes, &max_errors);
6530 bbio->map_type = map->type;
6531 bbio->num_stripes = num_stripes;
6532 bbio->max_errors = max_errors;
6533 bbio->mirror_num = mirror_num;
6536 * this is the case that REQ_READ && dev_replace_is_ongoing &&
6537 * mirror_num == num_stripes + 1 && dev_replace target drive is
6538 * available as a mirror
6540 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
6541 WARN_ON(num_stripes > 1);
6542 bbio->stripes[0].dev = dev_replace->tgtdev;
6543 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
6544 bbio->mirror_num = map->num_stripes + 1;
6547 if (dev_replace_is_ongoing) {
6548 lockdep_assert_held(&dev_replace->rwsem);
6549 /* Unlock and let waiting writers proceed */
6550 up_read(&dev_replace->rwsem);
6552 free_extent_map(em);
6556 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6557 u64 logical, u64 *length,
6558 struct btrfs_bio **bbio_ret, int mirror_num)
6560 if (op == BTRFS_MAP_DISCARD)
6561 return __btrfs_map_block_for_discard(fs_info, logical,
6564 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret,
6568 /* For Scrub/replace */
6569 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6570 u64 logical, u64 *length,
6571 struct btrfs_bio **bbio_ret)
6573 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1);
6576 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
6578 bio->bi_private = bbio->private;
6579 bio->bi_end_io = bbio->end_io;
6582 btrfs_put_bbio(bbio);
6585 static void btrfs_end_bio(struct bio *bio)
6587 struct btrfs_bio *bbio = bio->bi_private;
6588 int is_orig_bio = 0;
6590 if (bio->bi_status) {
6591 atomic_inc(&bbio->error);
6592 if (bio->bi_status == BLK_STS_IOERR ||
6593 bio->bi_status == BLK_STS_TARGET) {
6594 struct btrfs_device *dev = btrfs_io_bio(bio)->device;
6597 if (btrfs_op(bio) == BTRFS_MAP_WRITE)
6598 btrfs_dev_stat_inc_and_print(dev,
6599 BTRFS_DEV_STAT_WRITE_ERRS);
6600 else if (!(bio->bi_opf & REQ_RAHEAD))
6601 btrfs_dev_stat_inc_and_print(dev,
6602 BTRFS_DEV_STAT_READ_ERRS);
6603 if (bio->bi_opf & REQ_PREFLUSH)
6604 btrfs_dev_stat_inc_and_print(dev,
6605 BTRFS_DEV_STAT_FLUSH_ERRS);
6609 if (bio == bbio->orig_bio)
6612 btrfs_bio_counter_dec(bbio->fs_info);
6614 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6617 bio = bbio->orig_bio;
6620 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6621 /* only send an error to the higher layers if it is
6622 * beyond the tolerance of the btrfs bio
6624 if (atomic_read(&bbio->error) > bbio->max_errors) {
6625 bio->bi_status = BLK_STS_IOERR;
6628 * this bio is actually up to date, we didn't
6629 * go over the max number of errors
6631 bio->bi_status = BLK_STS_OK;
6634 btrfs_end_bbio(bbio, bio);
6635 } else if (!is_orig_bio) {
6640 static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio,
6641 u64 physical, struct btrfs_device *dev)
6643 struct btrfs_fs_info *fs_info = bbio->fs_info;
6645 bio->bi_private = bbio;
6646 btrfs_io_bio(bio)->device = dev;
6647 bio->bi_end_io = btrfs_end_bio;
6648 bio->bi_iter.bi_sector = physical >> 9;
6650 * For zone append writing, bi_sector must point the beginning of the
6653 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
6654 if (btrfs_dev_is_sequential(dev, physical)) {
6655 u64 zone_start = round_down(physical, fs_info->zone_size);
6657 bio->bi_iter.bi_sector = zone_start >> SECTOR_SHIFT;
6659 bio->bi_opf &= ~REQ_OP_ZONE_APPEND;
6660 bio->bi_opf |= REQ_OP_WRITE;
6663 btrfs_debug_in_rcu(fs_info,
6664 "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
6665 bio_op(bio), bio->bi_opf, bio->bi_iter.bi_sector,
6666 (unsigned long)dev->bdev->bd_dev, rcu_str_deref(dev->name),
6667 dev->devid, bio->bi_iter.bi_size);
6668 bio_set_dev(bio, dev->bdev);
6670 btrfs_bio_counter_inc_noblocked(fs_info);
6672 btrfsic_submit_bio(bio);
6675 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6677 atomic_inc(&bbio->error);
6678 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6679 /* Should be the original bio. */
6680 WARN_ON(bio != bbio->orig_bio);
6682 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6683 bio->bi_iter.bi_sector = logical >> 9;
6684 if (atomic_read(&bbio->error) > bbio->max_errors)
6685 bio->bi_status = BLK_STS_IOERR;
6687 bio->bi_status = BLK_STS_OK;
6688 btrfs_end_bbio(bbio, bio);
6692 blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
6695 struct btrfs_device *dev;
6696 struct bio *first_bio = bio;
6697 u64 logical = bio->bi_iter.bi_sector << 9;
6703 struct btrfs_bio *bbio = NULL;
6705 length = bio->bi_iter.bi_size;
6706 map_length = length;
6708 btrfs_bio_counter_inc_blocked(fs_info);
6709 ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical,
6710 &map_length, &bbio, mirror_num, 1);
6712 btrfs_bio_counter_dec(fs_info);
6713 return errno_to_blk_status(ret);
6716 total_devs = bbio->num_stripes;
6717 bbio->orig_bio = first_bio;
6718 bbio->private = first_bio->bi_private;
6719 bbio->end_io = first_bio->bi_end_io;
6720 bbio->fs_info = fs_info;
6721 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6723 if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6724 ((btrfs_op(bio) == BTRFS_MAP_WRITE) || (mirror_num > 1))) {
6725 /* In this case, map_length has been set to the length of
6726 a single stripe; not the whole write */
6727 if (btrfs_op(bio) == BTRFS_MAP_WRITE) {
6728 ret = raid56_parity_write(fs_info, bio, bbio,
6731 ret = raid56_parity_recover(fs_info, bio, bbio,
6732 map_length, mirror_num, 1);
6735 btrfs_bio_counter_dec(fs_info);
6736 return errno_to_blk_status(ret);
6739 if (map_length < length) {
6741 "mapping failed logical %llu bio len %llu len %llu",
6742 logical, length, map_length);
6746 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6747 dev = bbio->stripes[dev_nr].dev;
6748 if (!dev || !dev->bdev || test_bit(BTRFS_DEV_STATE_MISSING,
6750 (btrfs_op(first_bio) == BTRFS_MAP_WRITE &&
6751 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) {
6752 bbio_error(bbio, first_bio, logical);
6756 if (dev_nr < total_devs - 1)
6757 bio = btrfs_bio_clone(first_bio);
6761 submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical, dev);
6763 btrfs_bio_counter_dec(fs_info);
6768 * Find a device specified by @devid or @uuid in the list of @fs_devices, or
6771 * If devid and uuid are both specified, the match must be exact, otherwise
6772 * only devid is used.
6774 struct btrfs_device *btrfs_find_device(struct btrfs_fs_devices *fs_devices,
6775 u64 devid, u8 *uuid, u8 *fsid)
6777 struct btrfs_device *device;
6778 struct btrfs_fs_devices *seed_devs;
6780 if (!fsid || !memcmp(fs_devices->metadata_uuid, fsid, BTRFS_FSID_SIZE)) {
6781 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6782 if (device->devid == devid &&
6783 (!uuid || memcmp(device->uuid, uuid,
6784 BTRFS_UUID_SIZE) == 0))
6789 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
6791 !memcmp(seed_devs->metadata_uuid, fsid, BTRFS_FSID_SIZE)) {
6792 list_for_each_entry(device, &seed_devs->devices,
6794 if (device->devid == devid &&
6795 (!uuid || memcmp(device->uuid, uuid,
6796 BTRFS_UUID_SIZE) == 0))
6805 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices,
6806 u64 devid, u8 *dev_uuid)
6808 struct btrfs_device *device;
6809 unsigned int nofs_flag;
6812 * We call this under the chunk_mutex, so we want to use NOFS for this
6813 * allocation, however we don't want to change btrfs_alloc_device() to
6814 * always do NOFS because we use it in a lot of other GFP_KERNEL safe
6817 nofs_flag = memalloc_nofs_save();
6818 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6819 memalloc_nofs_restore(nofs_flag);
6823 list_add(&device->dev_list, &fs_devices->devices);
6824 device->fs_devices = fs_devices;
6825 fs_devices->num_devices++;
6827 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6828 fs_devices->missing_devices++;
6834 * btrfs_alloc_device - allocate struct btrfs_device
6835 * @fs_info: used only for generating a new devid, can be NULL if
6836 * devid is provided (i.e. @devid != NULL).
6837 * @devid: a pointer to devid for this device. If NULL a new devid
6839 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6842 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6843 * on error. Returned struct is not linked onto any lists and must be
6844 * destroyed with btrfs_free_device.
6846 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6850 struct btrfs_device *dev;
6853 if (WARN_ON(!devid && !fs_info))
6854 return ERR_PTR(-EINVAL);
6856 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
6858 return ERR_PTR(-ENOMEM);
6861 * Preallocate a bio that's always going to be used for flushing device
6862 * barriers and matches the device lifespan
6864 dev->flush_bio = bio_kmalloc(GFP_KERNEL, 0);
6865 if (!dev->flush_bio) {
6867 return ERR_PTR(-ENOMEM);
6870 INIT_LIST_HEAD(&dev->dev_list);
6871 INIT_LIST_HEAD(&dev->dev_alloc_list);
6872 INIT_LIST_HEAD(&dev->post_commit_list);
6874 atomic_set(&dev->reada_in_flight, 0);
6875 atomic_set(&dev->dev_stats_ccnt, 0);
6876 btrfs_device_data_ordered_init(dev);
6877 INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
6878 INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
6879 extent_io_tree_init(fs_info, &dev->alloc_state,
6880 IO_TREE_DEVICE_ALLOC_STATE, NULL);
6887 ret = find_next_devid(fs_info, &tmp);
6889 btrfs_free_device(dev);
6890 return ERR_PTR(ret);
6896 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6898 generate_random_uuid(dev->uuid);
6903 static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info,
6904 u64 devid, u8 *uuid, bool error)
6907 btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing",
6910 btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing",
6914 static u64 calc_stripe_length(u64 type, u64 chunk_len, int num_stripes)
6916 const int data_stripes = calc_data_stripes(type, num_stripes);
6918 return div_u64(chunk_len, data_stripes);
6921 #if BITS_PER_LONG == 32
6923 * Due to page cache limit, metadata beyond BTRFS_32BIT_MAX_FILE_SIZE
6924 * can't be accessed on 32bit systems.
6926 * This function do mount time check to reject the fs if it already has
6927 * metadata chunk beyond that limit.
6929 static int check_32bit_meta_chunk(struct btrfs_fs_info *fs_info,
6930 u64 logical, u64 length, u64 type)
6932 if (!(type & BTRFS_BLOCK_GROUP_METADATA))
6935 if (logical + length < MAX_LFS_FILESIZE)
6938 btrfs_err_32bit_limit(fs_info);
6943 * This is to give early warning for any metadata chunk reaching
6944 * BTRFS_32BIT_EARLY_WARN_THRESHOLD.
6945 * Although we can still access the metadata, it's not going to be possible
6946 * once the limit is reached.
6948 static void warn_32bit_meta_chunk(struct btrfs_fs_info *fs_info,
6949 u64 logical, u64 length, u64 type)
6951 if (!(type & BTRFS_BLOCK_GROUP_METADATA))
6954 if (logical + length < BTRFS_32BIT_EARLY_WARN_THRESHOLD)
6957 btrfs_warn_32bit_limit(fs_info);
6961 static int read_one_chunk(struct btrfs_key *key, struct extent_buffer *leaf,
6962 struct btrfs_chunk *chunk)
6964 struct btrfs_fs_info *fs_info = leaf->fs_info;
6965 struct extent_map_tree *map_tree = &fs_info->mapping_tree;
6966 struct map_lookup *map;
6967 struct extent_map *em;
6972 u8 uuid[BTRFS_UUID_SIZE];
6977 logical = key->offset;
6978 length = btrfs_chunk_length(leaf, chunk);
6979 type = btrfs_chunk_type(leaf, chunk);
6980 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6982 #if BITS_PER_LONG == 32
6983 ret = check_32bit_meta_chunk(fs_info, logical, length, type);
6986 warn_32bit_meta_chunk(fs_info, logical, length, type);
6990 * Only need to verify chunk item if we're reading from sys chunk array,
6991 * as chunk item in tree block is already verified by tree-checker.
6993 if (leaf->start == BTRFS_SUPER_INFO_OFFSET) {
6994 ret = btrfs_check_chunk_valid(leaf, chunk, logical);
6999 read_lock(&map_tree->lock);
7000 em = lookup_extent_mapping(map_tree, logical, 1);
7001 read_unlock(&map_tree->lock);
7003 /* already mapped? */
7004 if (em && em->start <= logical && em->start + em->len > logical) {
7005 free_extent_map(em);
7008 free_extent_map(em);
7011 em = alloc_extent_map();
7014 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
7016 free_extent_map(em);
7020 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
7021 em->map_lookup = map;
7022 em->start = logical;
7025 em->block_start = 0;
7026 em->block_len = em->len;
7028 map->num_stripes = num_stripes;
7029 map->io_width = btrfs_chunk_io_width(leaf, chunk);
7030 map->io_align = btrfs_chunk_io_align(leaf, chunk);
7031 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
7033 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
7034 map->verified_stripes = 0;
7035 em->orig_block_len = calc_stripe_length(type, em->len,
7037 for (i = 0; i < num_stripes; i++) {
7038 map->stripes[i].physical =
7039 btrfs_stripe_offset_nr(leaf, chunk, i);
7040 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
7041 read_extent_buffer(leaf, uuid, (unsigned long)
7042 btrfs_stripe_dev_uuid_nr(chunk, i),
7044 map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices,
7046 if (!map->stripes[i].dev &&
7047 !btrfs_test_opt(fs_info, DEGRADED)) {
7048 free_extent_map(em);
7049 btrfs_report_missing_device(fs_info, devid, uuid, true);
7052 if (!map->stripes[i].dev) {
7053 map->stripes[i].dev =
7054 add_missing_dev(fs_info->fs_devices, devid,
7056 if (IS_ERR(map->stripes[i].dev)) {
7057 free_extent_map(em);
7059 "failed to init missing dev %llu: %ld",
7060 devid, PTR_ERR(map->stripes[i].dev));
7061 return PTR_ERR(map->stripes[i].dev);
7063 btrfs_report_missing_device(fs_info, devid, uuid, false);
7065 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
7066 &(map->stripes[i].dev->dev_state));
7070 write_lock(&map_tree->lock);
7071 ret = add_extent_mapping(map_tree, em, 0);
7072 write_unlock(&map_tree->lock);
7075 "failed to add chunk map, start=%llu len=%llu: %d",
7076 em->start, em->len, ret);
7078 free_extent_map(em);
7083 static void fill_device_from_item(struct extent_buffer *leaf,
7084 struct btrfs_dev_item *dev_item,
7085 struct btrfs_device *device)
7089 device->devid = btrfs_device_id(leaf, dev_item);
7090 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
7091 device->total_bytes = device->disk_total_bytes;
7092 device->commit_total_bytes = device->disk_total_bytes;
7093 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
7094 device->commit_bytes_used = device->bytes_used;
7095 device->type = btrfs_device_type(leaf, dev_item);
7096 device->io_align = btrfs_device_io_align(leaf, dev_item);
7097 device->io_width = btrfs_device_io_width(leaf, dev_item);
7098 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
7099 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
7100 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
7102 ptr = btrfs_device_uuid(dev_item);
7103 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
7106 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info,
7109 struct btrfs_fs_devices *fs_devices;
7112 lockdep_assert_held(&uuid_mutex);
7115 /* This will match only for multi-device seed fs */
7116 list_for_each_entry(fs_devices, &fs_info->fs_devices->seed_list, seed_list)
7117 if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE))
7121 fs_devices = find_fsid(fsid, NULL);
7123 if (!btrfs_test_opt(fs_info, DEGRADED))
7124 return ERR_PTR(-ENOENT);
7126 fs_devices = alloc_fs_devices(fsid, NULL);
7127 if (IS_ERR(fs_devices))
7130 fs_devices->seeding = true;
7131 fs_devices->opened = 1;
7136 * Upon first call for a seed fs fsid, just create a private copy of the
7137 * respective fs_devices and anchor it at fs_info->fs_devices->seed_list
7139 fs_devices = clone_fs_devices(fs_devices);
7140 if (IS_ERR(fs_devices))
7143 ret = open_fs_devices(fs_devices, FMODE_READ, fs_info->bdev_holder);
7145 free_fs_devices(fs_devices);
7146 return ERR_PTR(ret);
7149 if (!fs_devices->seeding) {
7150 close_fs_devices(fs_devices);
7151 free_fs_devices(fs_devices);
7152 return ERR_PTR(-EINVAL);
7155 list_add(&fs_devices->seed_list, &fs_info->fs_devices->seed_list);
7160 static int read_one_dev(struct extent_buffer *leaf,
7161 struct btrfs_dev_item *dev_item)
7163 struct btrfs_fs_info *fs_info = leaf->fs_info;
7164 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7165 struct btrfs_device *device;
7168 u8 fs_uuid[BTRFS_FSID_SIZE];
7169 u8 dev_uuid[BTRFS_UUID_SIZE];
7171 devid = btrfs_device_id(leaf, dev_item);
7172 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
7174 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
7177 if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) {
7178 fs_devices = open_seed_devices(fs_info, fs_uuid);
7179 if (IS_ERR(fs_devices))
7180 return PTR_ERR(fs_devices);
7183 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
7186 if (!btrfs_test_opt(fs_info, DEGRADED)) {
7187 btrfs_report_missing_device(fs_info, devid,
7192 device = add_missing_dev(fs_devices, devid, dev_uuid);
7193 if (IS_ERR(device)) {
7195 "failed to add missing dev %llu: %ld",
7196 devid, PTR_ERR(device));
7197 return PTR_ERR(device);
7199 btrfs_report_missing_device(fs_info, devid, dev_uuid, false);
7201 if (!device->bdev) {
7202 if (!btrfs_test_opt(fs_info, DEGRADED)) {
7203 btrfs_report_missing_device(fs_info,
7204 devid, dev_uuid, true);
7207 btrfs_report_missing_device(fs_info, devid,
7211 if (!device->bdev &&
7212 !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
7214 * this happens when a device that was properly setup
7215 * in the device info lists suddenly goes bad.
7216 * device->bdev is NULL, and so we have to set
7217 * device->missing to one here
7219 device->fs_devices->missing_devices++;
7220 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
7223 /* Move the device to its own fs_devices */
7224 if (device->fs_devices != fs_devices) {
7225 ASSERT(test_bit(BTRFS_DEV_STATE_MISSING,
7226 &device->dev_state));
7228 list_move(&device->dev_list, &fs_devices->devices);
7229 device->fs_devices->num_devices--;
7230 fs_devices->num_devices++;
7232 device->fs_devices->missing_devices--;
7233 fs_devices->missing_devices++;
7235 device->fs_devices = fs_devices;
7239 if (device->fs_devices != fs_info->fs_devices) {
7240 BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state));
7241 if (device->generation !=
7242 btrfs_device_generation(leaf, dev_item))
7246 fill_device_from_item(leaf, dev_item, device);
7248 u64 max_total_bytes = i_size_read(device->bdev->bd_inode);
7250 if (device->total_bytes > max_total_bytes) {
7252 "device total_bytes should be at most %llu but found %llu",
7253 max_total_bytes, device->total_bytes);
7257 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
7258 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
7259 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
7260 device->fs_devices->total_rw_bytes += device->total_bytes;
7261 atomic64_add(device->total_bytes - device->bytes_used,
7262 &fs_info->free_chunk_space);
7268 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
7270 struct btrfs_root *root = fs_info->tree_root;
7271 struct btrfs_super_block *super_copy = fs_info->super_copy;
7272 struct extent_buffer *sb;
7273 struct btrfs_disk_key *disk_key;
7274 struct btrfs_chunk *chunk;
7276 unsigned long sb_array_offset;
7283 struct btrfs_key key;
7285 ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize);
7287 * This will create extent buffer of nodesize, superblock size is
7288 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
7289 * overallocate but we can keep it as-is, only the first page is used.
7291 sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET,
7292 root->root_key.objectid, 0);
7295 set_extent_buffer_uptodate(sb);
7297 * The sb extent buffer is artificial and just used to read the system array.
7298 * set_extent_buffer_uptodate() call does not properly mark all it's
7299 * pages up-to-date when the page is larger: extent does not cover the
7300 * whole page and consequently check_page_uptodate does not find all
7301 * the page's extents up-to-date (the hole beyond sb),
7302 * write_extent_buffer then triggers a WARN_ON.
7304 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
7305 * but sb spans only this function. Add an explicit SetPageUptodate call
7306 * to silence the warning eg. on PowerPC 64.
7308 if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
7309 SetPageUptodate(sb->pages[0]);
7311 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
7312 array_size = btrfs_super_sys_array_size(super_copy);
7314 array_ptr = super_copy->sys_chunk_array;
7315 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
7318 while (cur_offset < array_size) {
7319 disk_key = (struct btrfs_disk_key *)array_ptr;
7320 len = sizeof(*disk_key);
7321 if (cur_offset + len > array_size)
7322 goto out_short_read;
7324 btrfs_disk_key_to_cpu(&key, disk_key);
7327 sb_array_offset += len;
7330 if (key.type != BTRFS_CHUNK_ITEM_KEY) {
7332 "unexpected item type %u in sys_array at offset %u",
7333 (u32)key.type, cur_offset);
7338 chunk = (struct btrfs_chunk *)sb_array_offset;
7340 * At least one btrfs_chunk with one stripe must be present,
7341 * exact stripe count check comes afterwards
7343 len = btrfs_chunk_item_size(1);
7344 if (cur_offset + len > array_size)
7345 goto out_short_read;
7347 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
7350 "invalid number of stripes %u in sys_array at offset %u",
7351 num_stripes, cur_offset);
7356 type = btrfs_chunk_type(sb, chunk);
7357 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
7359 "invalid chunk type %llu in sys_array at offset %u",
7365 len = btrfs_chunk_item_size(num_stripes);
7366 if (cur_offset + len > array_size)
7367 goto out_short_read;
7369 ret = read_one_chunk(&key, sb, chunk);
7374 sb_array_offset += len;
7377 clear_extent_buffer_uptodate(sb);
7378 free_extent_buffer_stale(sb);
7382 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
7384 clear_extent_buffer_uptodate(sb);
7385 free_extent_buffer_stale(sb);
7390 * Check if all chunks in the fs are OK for read-write degraded mount
7392 * If the @failing_dev is specified, it's accounted as missing.
7394 * Return true if all chunks meet the minimal RW mount requirements.
7395 * Return false if any chunk doesn't meet the minimal RW mount requirements.
7397 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
7398 struct btrfs_device *failing_dev)
7400 struct extent_map_tree *map_tree = &fs_info->mapping_tree;
7401 struct extent_map *em;
7405 read_lock(&map_tree->lock);
7406 em = lookup_extent_mapping(map_tree, 0, (u64)-1);
7407 read_unlock(&map_tree->lock);
7408 /* No chunk at all? Return false anyway */
7414 struct map_lookup *map;
7419 map = em->map_lookup;
7421 btrfs_get_num_tolerated_disk_barrier_failures(
7423 for (i = 0; i < map->num_stripes; i++) {
7424 struct btrfs_device *dev = map->stripes[i].dev;
7426 if (!dev || !dev->bdev ||
7427 test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) ||
7428 dev->last_flush_error)
7430 else if (failing_dev && failing_dev == dev)
7433 if (missing > max_tolerated) {
7436 "chunk %llu missing %d devices, max tolerance is %d for writable mount",
7437 em->start, missing, max_tolerated);
7438 free_extent_map(em);
7442 next_start = extent_map_end(em);
7443 free_extent_map(em);
7445 read_lock(&map_tree->lock);
7446 em = lookup_extent_mapping(map_tree, next_start,
7447 (u64)(-1) - next_start);
7448 read_unlock(&map_tree->lock);
7454 static void readahead_tree_node_children(struct extent_buffer *node)
7457 const int nr_items = btrfs_header_nritems(node);
7459 for (i = 0; i < nr_items; i++)
7460 btrfs_readahead_node_child(node, i);
7463 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
7465 struct btrfs_root *root = fs_info->chunk_root;
7466 struct btrfs_path *path;
7467 struct extent_buffer *leaf;
7468 struct btrfs_key key;
7469 struct btrfs_key found_key;
7473 u64 last_ra_node = 0;
7475 path = btrfs_alloc_path();
7480 * uuid_mutex is needed only if we are mounting a sprout FS
7481 * otherwise we don't need it.
7483 mutex_lock(&uuid_mutex);
7486 * It is possible for mount and umount to race in such a way that
7487 * we execute this code path, but open_fs_devices failed to clear
7488 * total_rw_bytes. We certainly want it cleared before reading the
7489 * device items, so clear it here.
7491 fs_info->fs_devices->total_rw_bytes = 0;
7494 * Lockdep complains about possible circular locking dependency between
7495 * a disk's open_mutex (struct gendisk.open_mutex), the rw semaphores
7496 * used for freeze procection of a fs (struct super_block.s_writers),
7497 * which we take when starting a transaction, and extent buffers of the
7498 * chunk tree if we call read_one_dev() while holding a lock on an
7499 * extent buffer of the chunk tree. Since we are mounting the filesystem
7500 * and at this point there can't be any concurrent task modifying the
7501 * chunk tree, to keep it simple, just skip locking on the chunk tree.
7503 ASSERT(!test_bit(BTRFS_FS_OPEN, &fs_info->flags));
7504 path->skip_locking = 1;
7507 * Read all device items, and then all the chunk items. All
7508 * device items are found before any chunk item (their object id
7509 * is smaller than the lowest possible object id for a chunk
7510 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
7512 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
7515 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7519 struct extent_buffer *node;
7521 leaf = path->nodes[0];
7522 slot = path->slots[0];
7523 if (slot >= btrfs_header_nritems(leaf)) {
7524 ret = btrfs_next_leaf(root, path);
7531 node = path->nodes[1];
7533 if (last_ra_node != node->start) {
7534 readahead_tree_node_children(node);
7535 last_ra_node = node->start;
7538 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7539 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
7540 struct btrfs_dev_item *dev_item;
7541 dev_item = btrfs_item_ptr(leaf, slot,
7542 struct btrfs_dev_item);
7543 ret = read_one_dev(leaf, dev_item);
7547 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
7548 struct btrfs_chunk *chunk;
7551 * We are only called at mount time, so no need to take
7552 * fs_info->chunk_mutex. Plus, to avoid lockdep warnings,
7553 * we always lock first fs_info->chunk_mutex before
7554 * acquiring any locks on the chunk tree. This is a
7555 * requirement for chunk allocation, see the comment on
7556 * top of btrfs_chunk_alloc() for details.
7558 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
7559 ret = read_one_chunk(&found_key, leaf, chunk);
7567 * After loading chunk tree, we've got all device information,
7568 * do another round of validation checks.
7570 if (total_dev != fs_info->fs_devices->total_devices) {
7572 "super_num_devices %llu mismatch with num_devices %llu found here",
7573 btrfs_super_num_devices(fs_info->super_copy),
7578 if (btrfs_super_total_bytes(fs_info->super_copy) <
7579 fs_info->fs_devices->total_rw_bytes) {
7581 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
7582 btrfs_super_total_bytes(fs_info->super_copy),
7583 fs_info->fs_devices->total_rw_bytes);
7589 mutex_unlock(&uuid_mutex);
7591 btrfs_free_path(path);
7595 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
7597 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
7598 struct btrfs_device *device;
7600 fs_devices->fs_info = fs_info;
7602 mutex_lock(&fs_devices->device_list_mutex);
7603 list_for_each_entry(device, &fs_devices->devices, dev_list)
7604 device->fs_info = fs_info;
7606 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
7607 list_for_each_entry(device, &seed_devs->devices, dev_list)
7608 device->fs_info = fs_info;
7610 seed_devs->fs_info = fs_info;
7612 mutex_unlock(&fs_devices->device_list_mutex);
7615 static u64 btrfs_dev_stats_value(const struct extent_buffer *eb,
7616 const struct btrfs_dev_stats_item *ptr,
7621 read_extent_buffer(eb, &val,
7622 offsetof(struct btrfs_dev_stats_item, values) +
7623 ((unsigned long)ptr) + (index * sizeof(u64)),
7628 static void btrfs_set_dev_stats_value(struct extent_buffer *eb,
7629 struct btrfs_dev_stats_item *ptr,
7632 write_extent_buffer(eb, &val,
7633 offsetof(struct btrfs_dev_stats_item, values) +
7634 ((unsigned long)ptr) + (index * sizeof(u64)),
7638 static int btrfs_device_init_dev_stats(struct btrfs_device *device,
7639 struct btrfs_path *path)
7641 struct btrfs_dev_stats_item *ptr;
7642 struct extent_buffer *eb;
7643 struct btrfs_key key;
7647 if (!device->fs_info->dev_root)
7650 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7651 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7652 key.offset = device->devid;
7653 ret = btrfs_search_slot(NULL, device->fs_info->dev_root, &key, path, 0, 0);
7655 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7656 btrfs_dev_stat_set(device, i, 0);
7657 device->dev_stats_valid = 1;
7658 btrfs_release_path(path);
7659 return ret < 0 ? ret : 0;
7661 slot = path->slots[0];
7662 eb = path->nodes[0];
7663 item_size = btrfs_item_size_nr(eb, slot);
7665 ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_stats_item);
7667 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7668 if (item_size >= (1 + i) * sizeof(__le64))
7669 btrfs_dev_stat_set(device, i,
7670 btrfs_dev_stats_value(eb, ptr, i));
7672 btrfs_dev_stat_set(device, i, 0);
7675 device->dev_stats_valid = 1;
7676 btrfs_dev_stat_print_on_load(device);
7677 btrfs_release_path(path);
7682 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
7684 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
7685 struct btrfs_device *device;
7686 struct btrfs_path *path = NULL;
7689 path = btrfs_alloc_path();
7693 mutex_lock(&fs_devices->device_list_mutex);
7694 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7695 ret = btrfs_device_init_dev_stats(device, path);
7699 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
7700 list_for_each_entry(device, &seed_devs->devices, dev_list) {
7701 ret = btrfs_device_init_dev_stats(device, path);
7707 mutex_unlock(&fs_devices->device_list_mutex);
7709 btrfs_free_path(path);
7713 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
7714 struct btrfs_device *device)
7716 struct btrfs_fs_info *fs_info = trans->fs_info;
7717 struct btrfs_root *dev_root = fs_info->dev_root;
7718 struct btrfs_path *path;
7719 struct btrfs_key key;
7720 struct extent_buffer *eb;
7721 struct btrfs_dev_stats_item *ptr;
7725 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7726 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7727 key.offset = device->devid;
7729 path = btrfs_alloc_path();
7732 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
7734 btrfs_warn_in_rcu(fs_info,
7735 "error %d while searching for dev_stats item for device %s",
7736 ret, rcu_str_deref(device->name));
7741 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
7742 /* need to delete old one and insert a new one */
7743 ret = btrfs_del_item(trans, dev_root, path);
7745 btrfs_warn_in_rcu(fs_info,
7746 "delete too small dev_stats item for device %s failed %d",
7747 rcu_str_deref(device->name), ret);
7754 /* need to insert a new item */
7755 btrfs_release_path(path);
7756 ret = btrfs_insert_empty_item(trans, dev_root, path,
7757 &key, sizeof(*ptr));
7759 btrfs_warn_in_rcu(fs_info,
7760 "insert dev_stats item for device %s failed %d",
7761 rcu_str_deref(device->name), ret);
7766 eb = path->nodes[0];
7767 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7768 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7769 btrfs_set_dev_stats_value(eb, ptr, i,
7770 btrfs_dev_stat_read(device, i));
7771 btrfs_mark_buffer_dirty(eb);
7774 btrfs_free_path(path);
7779 * called from commit_transaction. Writes all changed device stats to disk.
7781 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans)
7783 struct btrfs_fs_info *fs_info = trans->fs_info;
7784 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7785 struct btrfs_device *device;
7789 mutex_lock(&fs_devices->device_list_mutex);
7790 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7791 stats_cnt = atomic_read(&device->dev_stats_ccnt);
7792 if (!device->dev_stats_valid || stats_cnt == 0)
7797 * There is a LOAD-LOAD control dependency between the value of
7798 * dev_stats_ccnt and updating the on-disk values which requires
7799 * reading the in-memory counters. Such control dependencies
7800 * require explicit read memory barriers.
7802 * This memory barriers pairs with smp_mb__before_atomic in
7803 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full
7804 * barrier implied by atomic_xchg in
7805 * btrfs_dev_stats_read_and_reset
7809 ret = update_dev_stat_item(trans, device);
7811 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7813 mutex_unlock(&fs_devices->device_list_mutex);
7818 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7820 btrfs_dev_stat_inc(dev, index);
7821 btrfs_dev_stat_print_on_error(dev);
7824 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
7826 if (!dev->dev_stats_valid)
7828 btrfs_err_rl_in_rcu(dev->fs_info,
7829 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7830 rcu_str_deref(dev->name),
7831 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7832 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7833 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7834 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7835 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7838 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7842 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7843 if (btrfs_dev_stat_read(dev, i) != 0)
7845 if (i == BTRFS_DEV_STAT_VALUES_MAX)
7846 return; /* all values == 0, suppress message */
7848 btrfs_info_in_rcu(dev->fs_info,
7849 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7850 rcu_str_deref(dev->name),
7851 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7852 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7853 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7854 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7855 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7858 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
7859 struct btrfs_ioctl_get_dev_stats *stats)
7861 struct btrfs_device *dev;
7862 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7865 mutex_lock(&fs_devices->device_list_mutex);
7866 dev = btrfs_find_device(fs_info->fs_devices, stats->devid, NULL, NULL);
7867 mutex_unlock(&fs_devices->device_list_mutex);
7870 btrfs_warn(fs_info, "get dev_stats failed, device not found");
7872 } else if (!dev->dev_stats_valid) {
7873 btrfs_warn(fs_info, "get dev_stats failed, not yet valid");
7875 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7876 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7877 if (stats->nr_items > i)
7879 btrfs_dev_stat_read_and_reset(dev, i);
7881 btrfs_dev_stat_set(dev, i, 0);
7883 btrfs_info(fs_info, "device stats zeroed by %s (%d)",
7884 current->comm, task_pid_nr(current));
7886 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7887 if (stats->nr_items > i)
7888 stats->values[i] = btrfs_dev_stat_read(dev, i);
7890 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7891 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7896 * Update the size and bytes used for each device where it changed. This is
7897 * delayed since we would otherwise get errors while writing out the
7900 * Must be invoked during transaction commit.
7902 void btrfs_commit_device_sizes(struct btrfs_transaction *trans)
7904 struct btrfs_device *curr, *next;
7906 ASSERT(trans->state == TRANS_STATE_COMMIT_DOING);
7908 if (list_empty(&trans->dev_update_list))
7912 * We don't need the device_list_mutex here. This list is owned by the
7913 * transaction and the transaction must complete before the device is
7916 mutex_lock(&trans->fs_info->chunk_mutex);
7917 list_for_each_entry_safe(curr, next, &trans->dev_update_list,
7919 list_del_init(&curr->post_commit_list);
7920 curr->commit_total_bytes = curr->disk_total_bytes;
7921 curr->commit_bytes_used = curr->bytes_used;
7923 mutex_unlock(&trans->fs_info->chunk_mutex);
7927 * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10.
7929 int btrfs_bg_type_to_factor(u64 flags)
7931 const int index = btrfs_bg_flags_to_raid_index(flags);
7933 return btrfs_raid_array[index].ncopies;
7938 static int verify_one_dev_extent(struct btrfs_fs_info *fs_info,
7939 u64 chunk_offset, u64 devid,
7940 u64 physical_offset, u64 physical_len)
7942 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
7943 struct extent_map *em;
7944 struct map_lookup *map;
7945 struct btrfs_device *dev;
7951 read_lock(&em_tree->lock);
7952 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
7953 read_unlock(&em_tree->lock);
7957 "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk",
7958 physical_offset, devid);
7963 map = em->map_lookup;
7964 stripe_len = calc_stripe_length(map->type, em->len, map->num_stripes);
7965 if (physical_len != stripe_len) {
7967 "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu",
7968 physical_offset, devid, em->start, physical_len,
7974 for (i = 0; i < map->num_stripes; i++) {
7975 if (map->stripes[i].dev->devid == devid &&
7976 map->stripes[i].physical == physical_offset) {
7978 if (map->verified_stripes >= map->num_stripes) {
7980 "too many dev extents for chunk %llu found",
7985 map->verified_stripes++;
7991 "dev extent physical offset %llu devid %llu has no corresponding chunk",
7992 physical_offset, devid);
7996 /* Make sure no dev extent is beyond device boundary */
7997 dev = btrfs_find_device(fs_info->fs_devices, devid, NULL, NULL);
7999 btrfs_err(fs_info, "failed to find devid %llu", devid);
8004 if (physical_offset + physical_len > dev->disk_total_bytes) {
8006 "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu",
8007 devid, physical_offset, physical_len,
8008 dev->disk_total_bytes);
8013 if (dev->zone_info) {
8014 u64 zone_size = dev->zone_info->zone_size;
8016 if (!IS_ALIGNED(physical_offset, zone_size) ||
8017 !IS_ALIGNED(physical_len, zone_size)) {
8019 "zoned: dev extent devid %llu physical offset %llu len %llu is not aligned to device zone",
8020 devid, physical_offset, physical_len);
8027 free_extent_map(em);
8031 static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info)
8033 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
8034 struct extent_map *em;
8035 struct rb_node *node;
8038 read_lock(&em_tree->lock);
8039 for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) {
8040 em = rb_entry(node, struct extent_map, rb_node);
8041 if (em->map_lookup->num_stripes !=
8042 em->map_lookup->verified_stripes) {
8044 "chunk %llu has missing dev extent, have %d expect %d",
8045 em->start, em->map_lookup->verified_stripes,
8046 em->map_lookup->num_stripes);
8052 read_unlock(&em_tree->lock);
8057 * Ensure that all dev extents are mapped to correct chunk, otherwise
8058 * later chunk allocation/free would cause unexpected behavior.
8060 * NOTE: This will iterate through the whole device tree, which should be of
8061 * the same size level as the chunk tree. This slightly increases mount time.
8063 int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info)
8065 struct btrfs_path *path;
8066 struct btrfs_root *root = fs_info->dev_root;
8067 struct btrfs_key key;
8069 u64 prev_dev_ext_end = 0;
8073 * We don't have a dev_root because we mounted with ignorebadroots and
8074 * failed to load the root, so we want to skip the verification in this
8077 * However if the dev root is fine, but the tree itself is corrupted
8078 * we'd still fail to mount. This verification is only to make sure
8079 * writes can happen safely, so instead just bypass this check
8080 * completely in the case of IGNOREBADROOTS.
8082 if (btrfs_test_opt(fs_info, IGNOREBADROOTS))
8086 key.type = BTRFS_DEV_EXTENT_KEY;
8089 path = btrfs_alloc_path();
8093 path->reada = READA_FORWARD;
8094 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
8098 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
8099 ret = btrfs_next_leaf(root, path);
8102 /* No dev extents at all? Not good */
8109 struct extent_buffer *leaf = path->nodes[0];
8110 struct btrfs_dev_extent *dext;
8111 int slot = path->slots[0];
8113 u64 physical_offset;
8117 btrfs_item_key_to_cpu(leaf, &key, slot);
8118 if (key.type != BTRFS_DEV_EXTENT_KEY)
8120 devid = key.objectid;
8121 physical_offset = key.offset;
8123 dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent);
8124 chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext);
8125 physical_len = btrfs_dev_extent_length(leaf, dext);
8127 /* Check if this dev extent overlaps with the previous one */
8128 if (devid == prev_devid && physical_offset < prev_dev_ext_end) {
8130 "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu",
8131 devid, physical_offset, prev_dev_ext_end);
8136 ret = verify_one_dev_extent(fs_info, chunk_offset, devid,
8137 physical_offset, physical_len);
8141 prev_dev_ext_end = physical_offset + physical_len;
8143 ret = btrfs_next_item(root, path);
8152 /* Ensure all chunks have corresponding dev extents */
8153 ret = verify_chunk_dev_extent_mapping(fs_info);
8155 btrfs_free_path(path);
8160 * Check whether the given block group or device is pinned by any inode being
8161 * used as a swapfile.
8163 bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr)
8165 struct btrfs_swapfile_pin *sp;
8166 struct rb_node *node;
8168 spin_lock(&fs_info->swapfile_pins_lock);
8169 node = fs_info->swapfile_pins.rb_node;
8171 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
8173 node = node->rb_left;
8174 else if (ptr > sp->ptr)
8175 node = node->rb_right;
8179 spin_unlock(&fs_info->swapfile_pins_lock);
8180 return node != NULL;
8183 static int relocating_repair_kthread(void *data)
8185 struct btrfs_block_group *cache = (struct btrfs_block_group *)data;
8186 struct btrfs_fs_info *fs_info = cache->fs_info;
8190 target = cache->start;
8191 btrfs_put_block_group(cache);
8193 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) {
8195 "zoned: skip relocating block group %llu to repair: EBUSY",
8200 mutex_lock(&fs_info->reclaim_bgs_lock);
8202 /* Ensure block group still exists */
8203 cache = btrfs_lookup_block_group(fs_info, target);
8207 if (!cache->relocating_repair)
8210 ret = btrfs_may_alloc_data_chunk(fs_info, target);
8215 "zoned: relocating block group %llu to repair IO failure",
8217 ret = btrfs_relocate_chunk(fs_info, target);
8221 btrfs_put_block_group(cache);
8222 mutex_unlock(&fs_info->reclaim_bgs_lock);
8223 btrfs_exclop_finish(fs_info);
8228 int btrfs_repair_one_zone(struct btrfs_fs_info *fs_info, u64 logical)
8230 struct btrfs_block_group *cache;
8232 /* Do not attempt to repair in degraded state */
8233 if (btrfs_test_opt(fs_info, DEGRADED))
8236 cache = btrfs_lookup_block_group(fs_info, logical);
8240 spin_lock(&cache->lock);
8241 if (cache->relocating_repair) {
8242 spin_unlock(&cache->lock);
8243 btrfs_put_block_group(cache);
8246 cache->relocating_repair = 1;
8247 spin_unlock(&cache->lock);
8249 kthread_run(relocating_repair_kthread, cache,
8250 "btrfs-relocating-repair");