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);
2639 device->fs_devices = fs_devices;
2641 mutex_lock(&fs_devices->device_list_mutex);
2642 mutex_lock(&fs_info->chunk_mutex);
2643 list_add_rcu(&device->dev_list, &fs_devices->devices);
2644 list_add(&device->dev_alloc_list, &fs_devices->alloc_list);
2645 fs_devices->num_devices++;
2646 fs_devices->open_devices++;
2647 fs_devices->rw_devices++;
2648 fs_devices->total_devices++;
2649 fs_devices->total_rw_bytes += device->total_bytes;
2651 atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2653 if (!blk_queue_nonrot(q))
2654 fs_devices->rotating = true;
2656 orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
2657 btrfs_set_super_total_bytes(fs_info->super_copy,
2658 round_down(orig_super_total_bytes + device->total_bytes,
2659 fs_info->sectorsize));
2661 orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy);
2662 btrfs_set_super_num_devices(fs_info->super_copy,
2663 orig_super_num_devices + 1);
2666 * we've got more storage, clear any full flags on the space
2669 btrfs_clear_space_info_full(fs_info);
2671 mutex_unlock(&fs_info->chunk_mutex);
2673 /* Add sysfs device entry */
2674 btrfs_sysfs_add_device(device);
2676 mutex_unlock(&fs_devices->device_list_mutex);
2679 mutex_lock(&fs_info->chunk_mutex);
2680 ret = init_first_rw_device(trans);
2681 mutex_unlock(&fs_info->chunk_mutex);
2683 btrfs_abort_transaction(trans, ret);
2688 ret = btrfs_add_dev_item(trans, device);
2690 btrfs_abort_transaction(trans, ret);
2695 ret = btrfs_finish_sprout(trans);
2697 btrfs_abort_transaction(trans, ret);
2702 * fs_devices now represents the newly sprouted filesystem and
2703 * its fsid has been changed by btrfs_prepare_sprout
2705 btrfs_sysfs_update_sprout_fsid(fs_devices);
2708 ret = btrfs_commit_transaction(trans);
2711 mutex_unlock(&uuid_mutex);
2712 up_write(&sb->s_umount);
2715 if (ret) /* transaction commit */
2718 ret = btrfs_relocate_sys_chunks(fs_info);
2720 btrfs_handle_fs_error(fs_info, ret,
2721 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2722 trans = btrfs_attach_transaction(root);
2723 if (IS_ERR(trans)) {
2724 if (PTR_ERR(trans) == -ENOENT)
2726 ret = PTR_ERR(trans);
2730 ret = btrfs_commit_transaction(trans);
2734 * Now that we have written a new super block to this device, check all
2735 * other fs_devices list if device_path alienates any other scanned
2737 * We can ignore the return value as it typically returns -EINVAL and
2738 * only succeeds if the device was an alien.
2740 btrfs_forget_devices(device_path);
2742 /* Update ctime/mtime for blkid or udev */
2743 update_dev_time(device_path);
2748 btrfs_sysfs_remove_device(device);
2749 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2750 mutex_lock(&fs_info->chunk_mutex);
2751 list_del_rcu(&device->dev_list);
2752 list_del(&device->dev_alloc_list);
2753 fs_info->fs_devices->num_devices--;
2754 fs_info->fs_devices->open_devices--;
2755 fs_info->fs_devices->rw_devices--;
2756 fs_info->fs_devices->total_devices--;
2757 fs_info->fs_devices->total_rw_bytes -= device->total_bytes;
2758 atomic64_sub(device->total_bytes, &fs_info->free_chunk_space);
2759 btrfs_set_super_total_bytes(fs_info->super_copy,
2760 orig_super_total_bytes);
2761 btrfs_set_super_num_devices(fs_info->super_copy,
2762 orig_super_num_devices);
2763 mutex_unlock(&fs_info->chunk_mutex);
2764 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2767 btrfs_set_sb_rdonly(sb);
2769 btrfs_end_transaction(trans);
2771 btrfs_destroy_dev_zone_info(device);
2773 btrfs_free_device(device);
2775 blkdev_put(bdev, FMODE_EXCL);
2777 mutex_unlock(&uuid_mutex);
2778 up_write(&sb->s_umount);
2783 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2784 struct btrfs_device *device)
2787 struct btrfs_path *path;
2788 struct btrfs_root *root = device->fs_info->chunk_root;
2789 struct btrfs_dev_item *dev_item;
2790 struct extent_buffer *leaf;
2791 struct btrfs_key key;
2793 path = btrfs_alloc_path();
2797 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2798 key.type = BTRFS_DEV_ITEM_KEY;
2799 key.offset = device->devid;
2801 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2810 leaf = path->nodes[0];
2811 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2813 btrfs_set_device_id(leaf, dev_item, device->devid);
2814 btrfs_set_device_type(leaf, dev_item, device->type);
2815 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2816 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2817 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2818 btrfs_set_device_total_bytes(leaf, dev_item,
2819 btrfs_device_get_disk_total_bytes(device));
2820 btrfs_set_device_bytes_used(leaf, dev_item,
2821 btrfs_device_get_bytes_used(device));
2822 btrfs_mark_buffer_dirty(leaf);
2825 btrfs_free_path(path);
2829 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2830 struct btrfs_device *device, u64 new_size)
2832 struct btrfs_fs_info *fs_info = device->fs_info;
2833 struct btrfs_super_block *super_copy = fs_info->super_copy;
2837 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2840 new_size = round_down(new_size, fs_info->sectorsize);
2842 mutex_lock(&fs_info->chunk_mutex);
2843 old_total = btrfs_super_total_bytes(super_copy);
2844 diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2846 if (new_size <= device->total_bytes ||
2847 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2848 mutex_unlock(&fs_info->chunk_mutex);
2852 btrfs_set_super_total_bytes(super_copy,
2853 round_down(old_total + diff, fs_info->sectorsize));
2854 device->fs_devices->total_rw_bytes += diff;
2856 btrfs_device_set_total_bytes(device, new_size);
2857 btrfs_device_set_disk_total_bytes(device, new_size);
2858 btrfs_clear_space_info_full(device->fs_info);
2859 if (list_empty(&device->post_commit_list))
2860 list_add_tail(&device->post_commit_list,
2861 &trans->transaction->dev_update_list);
2862 mutex_unlock(&fs_info->chunk_mutex);
2864 return btrfs_update_device(trans, device);
2867 static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
2869 struct btrfs_fs_info *fs_info = trans->fs_info;
2870 struct btrfs_root *root = fs_info->chunk_root;
2872 struct btrfs_path *path;
2873 struct btrfs_key key;
2875 path = btrfs_alloc_path();
2879 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2880 key.offset = chunk_offset;
2881 key.type = BTRFS_CHUNK_ITEM_KEY;
2883 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2886 else if (ret > 0) { /* Logic error or corruption */
2887 btrfs_handle_fs_error(fs_info, -ENOENT,
2888 "Failed lookup while freeing chunk.");
2893 ret = btrfs_del_item(trans, root, path);
2895 btrfs_handle_fs_error(fs_info, ret,
2896 "Failed to delete chunk item.");
2898 btrfs_free_path(path);
2902 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2904 struct btrfs_super_block *super_copy = fs_info->super_copy;
2905 struct btrfs_disk_key *disk_key;
2906 struct btrfs_chunk *chunk;
2913 struct btrfs_key key;
2915 lockdep_assert_held(&fs_info->chunk_mutex);
2916 array_size = btrfs_super_sys_array_size(super_copy);
2918 ptr = super_copy->sys_chunk_array;
2921 while (cur < array_size) {
2922 disk_key = (struct btrfs_disk_key *)ptr;
2923 btrfs_disk_key_to_cpu(&key, disk_key);
2925 len = sizeof(*disk_key);
2927 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2928 chunk = (struct btrfs_chunk *)(ptr + len);
2929 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2930 len += btrfs_chunk_item_size(num_stripes);
2935 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID &&
2936 key.offset == chunk_offset) {
2937 memmove(ptr, ptr + len, array_size - (cur + len));
2939 btrfs_set_super_sys_array_size(super_copy, array_size);
2949 * btrfs_get_chunk_map() - Find the mapping containing the given logical extent.
2950 * @logical: Logical block offset in bytes.
2951 * @length: Length of extent in bytes.
2953 * Return: Chunk mapping or ERR_PTR.
2955 struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info,
2956 u64 logical, u64 length)
2958 struct extent_map_tree *em_tree;
2959 struct extent_map *em;
2961 em_tree = &fs_info->mapping_tree;
2962 read_lock(&em_tree->lock);
2963 em = lookup_extent_mapping(em_tree, logical, length);
2964 read_unlock(&em_tree->lock);
2967 btrfs_crit(fs_info, "unable to find logical %llu length %llu",
2969 return ERR_PTR(-EINVAL);
2972 if (em->start > logical || em->start + em->len < logical) {
2974 "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
2975 logical, length, em->start, em->start + em->len);
2976 free_extent_map(em);
2977 return ERR_PTR(-EINVAL);
2980 /* callers are responsible for dropping em's ref. */
2984 static int remove_chunk_item(struct btrfs_trans_handle *trans,
2985 struct map_lookup *map, u64 chunk_offset)
2990 * Removing chunk items and updating the device items in the chunks btree
2991 * requires holding the chunk_mutex.
2992 * See the comment at btrfs_chunk_alloc() for the details.
2994 lockdep_assert_held(&trans->fs_info->chunk_mutex);
2996 for (i = 0; i < map->num_stripes; i++) {
2999 ret = btrfs_update_device(trans, map->stripes[i].dev);
3004 return btrfs_free_chunk(trans, chunk_offset);
3007 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
3009 struct btrfs_fs_info *fs_info = trans->fs_info;
3010 struct extent_map *em;
3011 struct map_lookup *map;
3012 u64 dev_extent_len = 0;
3014 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
3016 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
3019 * This is a logic error, but we don't want to just rely on the
3020 * user having built with ASSERT enabled, so if ASSERT doesn't
3021 * do anything we still error out.
3026 map = em->map_lookup;
3029 * First delete the device extent items from the devices btree.
3030 * We take the device_list_mutex to avoid racing with the finishing phase
3031 * of a device replace operation. See the comment below before acquiring
3032 * fs_info->chunk_mutex. Note that here we do not acquire the chunk_mutex
3033 * because that can result in a deadlock when deleting the device extent
3034 * items from the devices btree - COWing an extent buffer from the btree
3035 * may result in allocating a new metadata chunk, which would attempt to
3036 * lock again fs_info->chunk_mutex.
3038 mutex_lock(&fs_devices->device_list_mutex);
3039 for (i = 0; i < map->num_stripes; i++) {
3040 struct btrfs_device *device = map->stripes[i].dev;
3041 ret = btrfs_free_dev_extent(trans, device,
3042 map->stripes[i].physical,
3045 mutex_unlock(&fs_devices->device_list_mutex);
3046 btrfs_abort_transaction(trans, ret);
3050 if (device->bytes_used > 0) {
3051 mutex_lock(&fs_info->chunk_mutex);
3052 btrfs_device_set_bytes_used(device,
3053 device->bytes_used - dev_extent_len);
3054 atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
3055 btrfs_clear_space_info_full(fs_info);
3056 mutex_unlock(&fs_info->chunk_mutex);
3059 mutex_unlock(&fs_devices->device_list_mutex);
3062 * We acquire fs_info->chunk_mutex for 2 reasons:
3064 * 1) Just like with the first phase of the chunk allocation, we must
3065 * reserve system space, do all chunk btree updates and deletions, and
3066 * update the system chunk array in the superblock while holding this
3067 * mutex. This is for similar reasons as explained on the comment at
3068 * the top of btrfs_chunk_alloc();
3070 * 2) Prevent races with the final phase of a device replace operation
3071 * that replaces the device object associated with the map's stripes,
3072 * because the device object's id can change at any time during that
3073 * final phase of the device replace operation
3074 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the
3075 * replaced device and then see it with an ID of
3076 * BTRFS_DEV_REPLACE_DEVID, which would cause a failure when updating
3077 * the device item, which does not exists on the chunk btree.
3078 * The finishing phase of device replace acquires both the
3079 * device_list_mutex and the chunk_mutex, in that order, so we are
3080 * safe by just acquiring the chunk_mutex.
3082 trans->removing_chunk = true;
3083 mutex_lock(&fs_info->chunk_mutex);
3085 check_system_chunk(trans, map->type);
3087 ret = remove_chunk_item(trans, map, chunk_offset);
3089 * Normally we should not get -ENOSPC since we reserved space before
3090 * through the call to check_system_chunk().
3092 * Despite our system space_info having enough free space, we may not
3093 * be able to allocate extents from its block groups, because all have
3094 * an incompatible profile, which will force us to allocate a new system
3095 * block group with the right profile, or right after we called
3096 * check_system_space() above, a scrub turned the only system block group
3097 * with enough free space into RO mode.
3098 * This is explained with more detail at do_chunk_alloc().
3100 * So if we get -ENOSPC, allocate a new system chunk and retry once.
3102 if (ret == -ENOSPC) {
3103 const u64 sys_flags = btrfs_system_alloc_profile(fs_info);
3104 struct btrfs_block_group *sys_bg;
3106 sys_bg = btrfs_alloc_chunk(trans, sys_flags);
3107 if (IS_ERR(sys_bg)) {
3108 ret = PTR_ERR(sys_bg);
3109 btrfs_abort_transaction(trans, ret);
3113 ret = btrfs_chunk_alloc_add_chunk_item(trans, sys_bg);
3115 btrfs_abort_transaction(trans, ret);
3119 ret = remove_chunk_item(trans, map, chunk_offset);
3121 btrfs_abort_transaction(trans, ret);
3125 btrfs_abort_transaction(trans, ret);
3129 trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
3131 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
3132 ret = btrfs_del_sys_chunk(fs_info, chunk_offset);
3134 btrfs_abort_transaction(trans, ret);
3139 mutex_unlock(&fs_info->chunk_mutex);
3140 trans->removing_chunk = false;
3143 * We are done with chunk btree updates and deletions, so release the
3144 * system space we previously reserved (with check_system_chunk()).
3146 btrfs_trans_release_chunk_metadata(trans);
3148 ret = btrfs_remove_block_group(trans, chunk_offset, em);
3150 btrfs_abort_transaction(trans, ret);
3155 if (trans->removing_chunk) {
3156 mutex_unlock(&fs_info->chunk_mutex);
3157 trans->removing_chunk = false;
3160 free_extent_map(em);
3164 int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
3166 struct btrfs_root *root = fs_info->chunk_root;
3167 struct btrfs_trans_handle *trans;
3168 struct btrfs_block_group *block_group;
3173 * Prevent races with automatic removal of unused block groups.
3174 * After we relocate and before we remove the chunk with offset
3175 * chunk_offset, automatic removal of the block group can kick in,
3176 * resulting in a failure when calling btrfs_remove_chunk() below.
3178 * Make sure to acquire this mutex before doing a tree search (dev
3179 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
3180 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
3181 * we release the path used to search the chunk/dev tree and before
3182 * the current task acquires this mutex and calls us.
3184 lockdep_assert_held(&fs_info->reclaim_bgs_lock);
3186 /* step one, relocate all the extents inside this chunk */
3187 btrfs_scrub_pause(fs_info);
3188 ret = btrfs_relocate_block_group(fs_info, chunk_offset);
3189 btrfs_scrub_continue(fs_info);
3193 block_group = btrfs_lookup_block_group(fs_info, chunk_offset);
3196 btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group);
3197 length = block_group->length;
3198 btrfs_put_block_group(block_group);
3201 * On a zoned file system, discard the whole block group, this will
3202 * trigger a REQ_OP_ZONE_RESET operation on the device zone. If
3203 * resetting the zone fails, don't treat it as a fatal problem from the
3204 * filesystem's point of view.
3206 if (btrfs_is_zoned(fs_info)) {
3207 ret = btrfs_discard_extent(fs_info, chunk_offset, length, NULL);
3210 "failed to reset zone %llu after relocation",
3214 trans = btrfs_start_trans_remove_block_group(root->fs_info,
3216 if (IS_ERR(trans)) {
3217 ret = PTR_ERR(trans);
3218 btrfs_handle_fs_error(root->fs_info, ret, NULL);
3223 * step two, delete the device extents and the
3224 * chunk tree entries
3226 ret = btrfs_remove_chunk(trans, chunk_offset);
3227 btrfs_end_transaction(trans);
3231 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
3233 struct btrfs_root *chunk_root = fs_info->chunk_root;
3234 struct btrfs_path *path;
3235 struct extent_buffer *leaf;
3236 struct btrfs_chunk *chunk;
3237 struct btrfs_key key;
3238 struct btrfs_key found_key;
3240 bool retried = false;
3244 path = btrfs_alloc_path();
3249 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3250 key.offset = (u64)-1;
3251 key.type = BTRFS_CHUNK_ITEM_KEY;
3254 mutex_lock(&fs_info->reclaim_bgs_lock);
3255 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3257 mutex_unlock(&fs_info->reclaim_bgs_lock);
3260 BUG_ON(ret == 0); /* Corruption */
3262 ret = btrfs_previous_item(chunk_root, path, key.objectid,
3265 mutex_unlock(&fs_info->reclaim_bgs_lock);
3271 leaf = path->nodes[0];
3272 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3274 chunk = btrfs_item_ptr(leaf, path->slots[0],
3275 struct btrfs_chunk);
3276 chunk_type = btrfs_chunk_type(leaf, chunk);
3277 btrfs_release_path(path);
3279 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3280 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3286 mutex_unlock(&fs_info->reclaim_bgs_lock);
3288 if (found_key.offset == 0)
3290 key.offset = found_key.offset - 1;
3293 if (failed && !retried) {
3297 } else if (WARN_ON(failed && retried)) {
3301 btrfs_free_path(path);
3306 * return 1 : allocate a data chunk successfully,
3307 * return <0: errors during allocating a data chunk,
3308 * return 0 : no need to allocate a data chunk.
3310 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info,
3313 struct btrfs_block_group *cache;
3317 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3319 chunk_type = cache->flags;
3320 btrfs_put_block_group(cache);
3322 if (!(chunk_type & BTRFS_BLOCK_GROUP_DATA))
3325 spin_lock(&fs_info->data_sinfo->lock);
3326 bytes_used = fs_info->data_sinfo->bytes_used;
3327 spin_unlock(&fs_info->data_sinfo->lock);
3330 struct btrfs_trans_handle *trans;
3333 trans = btrfs_join_transaction(fs_info->tree_root);
3335 return PTR_ERR(trans);
3337 ret = btrfs_force_chunk_alloc(trans, BTRFS_BLOCK_GROUP_DATA);
3338 btrfs_end_transaction(trans);
3347 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3348 struct btrfs_balance_control *bctl)
3350 struct btrfs_root *root = fs_info->tree_root;
3351 struct btrfs_trans_handle *trans;
3352 struct btrfs_balance_item *item;
3353 struct btrfs_disk_balance_args disk_bargs;
3354 struct btrfs_path *path;
3355 struct extent_buffer *leaf;
3356 struct btrfs_key key;
3359 path = btrfs_alloc_path();
3363 trans = btrfs_start_transaction(root, 0);
3364 if (IS_ERR(trans)) {
3365 btrfs_free_path(path);
3366 return PTR_ERR(trans);
3369 key.objectid = BTRFS_BALANCE_OBJECTID;
3370 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3373 ret = btrfs_insert_empty_item(trans, root, path, &key,
3378 leaf = path->nodes[0];
3379 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3381 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3383 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3384 btrfs_set_balance_data(leaf, item, &disk_bargs);
3385 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3386 btrfs_set_balance_meta(leaf, item, &disk_bargs);
3387 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3388 btrfs_set_balance_sys(leaf, item, &disk_bargs);
3390 btrfs_set_balance_flags(leaf, item, bctl->flags);
3392 btrfs_mark_buffer_dirty(leaf);
3394 btrfs_free_path(path);
3395 err = btrfs_commit_transaction(trans);
3401 static int del_balance_item(struct btrfs_fs_info *fs_info)
3403 struct btrfs_root *root = fs_info->tree_root;
3404 struct btrfs_trans_handle *trans;
3405 struct btrfs_path *path;
3406 struct btrfs_key key;
3409 path = btrfs_alloc_path();
3413 trans = btrfs_start_transaction_fallback_global_rsv(root, 0);
3414 if (IS_ERR(trans)) {
3415 btrfs_free_path(path);
3416 return PTR_ERR(trans);
3419 key.objectid = BTRFS_BALANCE_OBJECTID;
3420 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3423 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3431 ret = btrfs_del_item(trans, root, path);
3433 btrfs_free_path(path);
3434 err = btrfs_commit_transaction(trans);
3441 * This is a heuristic used to reduce the number of chunks balanced on
3442 * resume after balance was interrupted.
3444 static void update_balance_args(struct btrfs_balance_control *bctl)
3447 * Turn on soft mode for chunk types that were being converted.
3449 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3450 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3451 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3452 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3453 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3454 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3457 * Turn on usage filter if is not already used. The idea is
3458 * that chunks that we have already balanced should be
3459 * reasonably full. Don't do it for chunks that are being
3460 * converted - that will keep us from relocating unconverted
3461 * (albeit full) chunks.
3463 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3464 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3465 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3466 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3467 bctl->data.usage = 90;
3469 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3470 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3471 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3472 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3473 bctl->sys.usage = 90;
3475 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3476 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3477 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3478 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3479 bctl->meta.usage = 90;
3484 * Clear the balance status in fs_info and delete the balance item from disk.
3486 static void reset_balance_state(struct btrfs_fs_info *fs_info)
3488 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3491 BUG_ON(!fs_info->balance_ctl);
3493 spin_lock(&fs_info->balance_lock);
3494 fs_info->balance_ctl = NULL;
3495 spin_unlock(&fs_info->balance_lock);
3498 ret = del_balance_item(fs_info);
3500 btrfs_handle_fs_error(fs_info, ret, NULL);
3504 * Balance filters. Return 1 if chunk should be filtered out
3505 * (should not be balanced).
3507 static int chunk_profiles_filter(u64 chunk_type,
3508 struct btrfs_balance_args *bargs)
3510 chunk_type = chunk_to_extended(chunk_type) &
3511 BTRFS_EXTENDED_PROFILE_MASK;
3513 if (bargs->profiles & chunk_type)
3519 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3520 struct btrfs_balance_args *bargs)
3522 struct btrfs_block_group *cache;
3524 u64 user_thresh_min;
3525 u64 user_thresh_max;
3528 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3529 chunk_used = cache->used;
3531 if (bargs->usage_min == 0)
3532 user_thresh_min = 0;
3534 user_thresh_min = div_factor_fine(cache->length,
3537 if (bargs->usage_max == 0)
3538 user_thresh_max = 1;
3539 else if (bargs->usage_max > 100)
3540 user_thresh_max = cache->length;
3542 user_thresh_max = div_factor_fine(cache->length,
3545 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3548 btrfs_put_block_group(cache);
3552 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3553 u64 chunk_offset, struct btrfs_balance_args *bargs)
3555 struct btrfs_block_group *cache;
3556 u64 chunk_used, user_thresh;
3559 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3560 chunk_used = cache->used;
3562 if (bargs->usage_min == 0)
3564 else if (bargs->usage > 100)
3565 user_thresh = cache->length;
3567 user_thresh = div_factor_fine(cache->length, bargs->usage);
3569 if (chunk_used < user_thresh)
3572 btrfs_put_block_group(cache);
3576 static int chunk_devid_filter(struct extent_buffer *leaf,
3577 struct btrfs_chunk *chunk,
3578 struct btrfs_balance_args *bargs)
3580 struct btrfs_stripe *stripe;
3581 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3584 for (i = 0; i < num_stripes; i++) {
3585 stripe = btrfs_stripe_nr(chunk, i);
3586 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3593 static u64 calc_data_stripes(u64 type, int num_stripes)
3595 const int index = btrfs_bg_flags_to_raid_index(type);
3596 const int ncopies = btrfs_raid_array[index].ncopies;
3597 const int nparity = btrfs_raid_array[index].nparity;
3599 return (num_stripes - nparity) / ncopies;
3602 /* [pstart, pend) */
3603 static int chunk_drange_filter(struct extent_buffer *leaf,
3604 struct btrfs_chunk *chunk,
3605 struct btrfs_balance_args *bargs)
3607 struct btrfs_stripe *stripe;
3608 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3615 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3618 type = btrfs_chunk_type(leaf, chunk);
3619 factor = calc_data_stripes(type, num_stripes);
3621 for (i = 0; i < num_stripes; i++) {
3622 stripe = btrfs_stripe_nr(chunk, i);
3623 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3626 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3627 stripe_length = btrfs_chunk_length(leaf, chunk);
3628 stripe_length = div_u64(stripe_length, factor);
3630 if (stripe_offset < bargs->pend &&
3631 stripe_offset + stripe_length > bargs->pstart)
3638 /* [vstart, vend) */
3639 static int chunk_vrange_filter(struct extent_buffer *leaf,
3640 struct btrfs_chunk *chunk,
3642 struct btrfs_balance_args *bargs)
3644 if (chunk_offset < bargs->vend &&
3645 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3646 /* at least part of the chunk is inside this vrange */
3652 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3653 struct btrfs_chunk *chunk,
3654 struct btrfs_balance_args *bargs)
3656 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3658 if (bargs->stripes_min <= num_stripes
3659 && num_stripes <= bargs->stripes_max)
3665 static int chunk_soft_convert_filter(u64 chunk_type,
3666 struct btrfs_balance_args *bargs)
3668 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3671 chunk_type = chunk_to_extended(chunk_type) &
3672 BTRFS_EXTENDED_PROFILE_MASK;
3674 if (bargs->target == chunk_type)
3680 static int should_balance_chunk(struct extent_buffer *leaf,
3681 struct btrfs_chunk *chunk, u64 chunk_offset)
3683 struct btrfs_fs_info *fs_info = leaf->fs_info;
3684 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3685 struct btrfs_balance_args *bargs = NULL;
3686 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3689 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3690 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3694 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3695 bargs = &bctl->data;
3696 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3698 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3699 bargs = &bctl->meta;
3701 /* profiles filter */
3702 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3703 chunk_profiles_filter(chunk_type, bargs)) {
3708 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3709 chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3711 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3712 chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3717 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3718 chunk_devid_filter(leaf, chunk, bargs)) {
3722 /* drange filter, makes sense only with devid filter */
3723 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3724 chunk_drange_filter(leaf, chunk, bargs)) {
3729 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3730 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3734 /* stripes filter */
3735 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3736 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3740 /* soft profile changing mode */
3741 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3742 chunk_soft_convert_filter(chunk_type, bargs)) {
3747 * limited by count, must be the last filter
3749 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3750 if (bargs->limit == 0)
3754 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3756 * Same logic as the 'limit' filter; the minimum cannot be
3757 * determined here because we do not have the global information
3758 * about the count of all chunks that satisfy the filters.
3760 if (bargs->limit_max == 0)
3769 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3771 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3772 struct btrfs_root *chunk_root = fs_info->chunk_root;
3774 struct btrfs_chunk *chunk;
3775 struct btrfs_path *path = NULL;
3776 struct btrfs_key key;
3777 struct btrfs_key found_key;
3778 struct extent_buffer *leaf;
3781 int enospc_errors = 0;
3782 bool counting = true;
3783 /* The single value limit and min/max limits use the same bytes in the */
3784 u64 limit_data = bctl->data.limit;
3785 u64 limit_meta = bctl->meta.limit;
3786 u64 limit_sys = bctl->sys.limit;
3790 int chunk_reserved = 0;
3792 path = btrfs_alloc_path();
3798 /* zero out stat counters */
3799 spin_lock(&fs_info->balance_lock);
3800 memset(&bctl->stat, 0, sizeof(bctl->stat));
3801 spin_unlock(&fs_info->balance_lock);
3805 * The single value limit and min/max limits use the same bytes
3808 bctl->data.limit = limit_data;
3809 bctl->meta.limit = limit_meta;
3810 bctl->sys.limit = limit_sys;
3812 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3813 key.offset = (u64)-1;
3814 key.type = BTRFS_CHUNK_ITEM_KEY;
3817 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3818 atomic_read(&fs_info->balance_cancel_req)) {
3823 mutex_lock(&fs_info->reclaim_bgs_lock);
3824 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3826 mutex_unlock(&fs_info->reclaim_bgs_lock);
3831 * this shouldn't happen, it means the last relocate
3835 BUG(); /* FIXME break ? */
3837 ret = btrfs_previous_item(chunk_root, path, 0,
3838 BTRFS_CHUNK_ITEM_KEY);
3840 mutex_unlock(&fs_info->reclaim_bgs_lock);
3845 leaf = path->nodes[0];
3846 slot = path->slots[0];
3847 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3849 if (found_key.objectid != key.objectid) {
3850 mutex_unlock(&fs_info->reclaim_bgs_lock);
3854 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3855 chunk_type = btrfs_chunk_type(leaf, chunk);
3858 spin_lock(&fs_info->balance_lock);
3859 bctl->stat.considered++;
3860 spin_unlock(&fs_info->balance_lock);
3863 ret = should_balance_chunk(leaf, chunk, found_key.offset);
3865 btrfs_release_path(path);
3867 mutex_unlock(&fs_info->reclaim_bgs_lock);
3872 mutex_unlock(&fs_info->reclaim_bgs_lock);
3873 spin_lock(&fs_info->balance_lock);
3874 bctl->stat.expected++;
3875 spin_unlock(&fs_info->balance_lock);
3877 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3879 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3881 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3888 * Apply limit_min filter, no need to check if the LIMITS
3889 * filter is used, limit_min is 0 by default
3891 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3892 count_data < bctl->data.limit_min)
3893 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3894 count_meta < bctl->meta.limit_min)
3895 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3896 count_sys < bctl->sys.limit_min)) {
3897 mutex_unlock(&fs_info->reclaim_bgs_lock);
3901 if (!chunk_reserved) {
3903 * We may be relocating the only data chunk we have,
3904 * which could potentially end up with losing data's
3905 * raid profile, so lets allocate an empty one in
3908 ret = btrfs_may_alloc_data_chunk(fs_info,
3911 mutex_unlock(&fs_info->reclaim_bgs_lock);
3913 } else if (ret == 1) {
3918 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3919 mutex_unlock(&fs_info->reclaim_bgs_lock);
3920 if (ret == -ENOSPC) {
3922 } else if (ret == -ETXTBSY) {
3924 "skipping relocation of block group %llu due to active swapfile",
3930 spin_lock(&fs_info->balance_lock);
3931 bctl->stat.completed++;
3932 spin_unlock(&fs_info->balance_lock);
3935 if (found_key.offset == 0)
3937 key.offset = found_key.offset - 1;
3941 btrfs_release_path(path);
3946 btrfs_free_path(path);
3947 if (enospc_errors) {
3948 btrfs_info(fs_info, "%d enospc errors during balance",
3958 * alloc_profile_is_valid - see if a given profile is valid and reduced
3959 * @flags: profile to validate
3960 * @extended: if true @flags is treated as an extended profile
3962 static int alloc_profile_is_valid(u64 flags, int extended)
3964 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3965 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3967 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3969 /* 1) check that all other bits are zeroed */
3973 /* 2) see if profile is reduced */
3975 return !extended; /* "0" is valid for usual profiles */
3977 return has_single_bit_set(flags);
3980 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3982 /* cancel requested || normal exit path */
3983 return atomic_read(&fs_info->balance_cancel_req) ||
3984 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3985 atomic_read(&fs_info->balance_cancel_req) == 0);
3989 * Validate target profile against allowed profiles and return true if it's OK.
3990 * Otherwise print the error message and return false.
3992 static inline int validate_convert_profile(struct btrfs_fs_info *fs_info,
3993 const struct btrfs_balance_args *bargs,
3994 u64 allowed, const char *type)
3996 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3999 if (fs_info->sectorsize < PAGE_SIZE &&
4000 bargs->target & BTRFS_BLOCK_GROUP_RAID56_MASK) {
4002 "RAID56 is not yet supported for sectorsize %u with page size %lu",
4003 fs_info->sectorsize, PAGE_SIZE);
4006 /* Profile is valid and does not have bits outside of the allowed set */
4007 if (alloc_profile_is_valid(bargs->target, 1) &&
4008 (bargs->target & ~allowed) == 0)
4011 btrfs_err(fs_info, "balance: invalid convert %s profile %s",
4012 type, btrfs_bg_type_to_raid_name(bargs->target));
4017 * Fill @buf with textual description of balance filter flags @bargs, up to
4018 * @size_buf including the terminating null. The output may be trimmed if it
4019 * does not fit into the provided buffer.
4021 static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf,
4025 u32 size_bp = size_buf;
4027 u64 flags = bargs->flags;
4028 char tmp_buf[128] = {'\0'};
4033 #define CHECK_APPEND_NOARG(a) \
4035 ret = snprintf(bp, size_bp, (a)); \
4036 if (ret < 0 || ret >= size_bp) \
4037 goto out_overflow; \
4042 #define CHECK_APPEND_1ARG(a, v1) \
4044 ret = snprintf(bp, size_bp, (a), (v1)); \
4045 if (ret < 0 || ret >= size_bp) \
4046 goto out_overflow; \
4051 #define CHECK_APPEND_2ARG(a, v1, v2) \
4053 ret = snprintf(bp, size_bp, (a), (v1), (v2)); \
4054 if (ret < 0 || ret >= size_bp) \
4055 goto out_overflow; \
4060 if (flags & BTRFS_BALANCE_ARGS_CONVERT)
4061 CHECK_APPEND_1ARG("convert=%s,",
4062 btrfs_bg_type_to_raid_name(bargs->target));
4064 if (flags & BTRFS_BALANCE_ARGS_SOFT)
4065 CHECK_APPEND_NOARG("soft,");
4067 if (flags & BTRFS_BALANCE_ARGS_PROFILES) {
4068 btrfs_describe_block_groups(bargs->profiles, tmp_buf,
4070 CHECK_APPEND_1ARG("profiles=%s,", tmp_buf);
4073 if (flags & BTRFS_BALANCE_ARGS_USAGE)
4074 CHECK_APPEND_1ARG("usage=%llu,", bargs->usage);
4076 if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE)
4077 CHECK_APPEND_2ARG("usage=%u..%u,",
4078 bargs->usage_min, bargs->usage_max);
4080 if (flags & BTRFS_BALANCE_ARGS_DEVID)
4081 CHECK_APPEND_1ARG("devid=%llu,", bargs->devid);
4083 if (flags & BTRFS_BALANCE_ARGS_DRANGE)
4084 CHECK_APPEND_2ARG("drange=%llu..%llu,",
4085 bargs->pstart, bargs->pend);
4087 if (flags & BTRFS_BALANCE_ARGS_VRANGE)
4088 CHECK_APPEND_2ARG("vrange=%llu..%llu,",
4089 bargs->vstart, bargs->vend);
4091 if (flags & BTRFS_BALANCE_ARGS_LIMIT)
4092 CHECK_APPEND_1ARG("limit=%llu,", bargs->limit);
4094 if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)
4095 CHECK_APPEND_2ARG("limit=%u..%u,",
4096 bargs->limit_min, bargs->limit_max);
4098 if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE)
4099 CHECK_APPEND_2ARG("stripes=%u..%u,",
4100 bargs->stripes_min, bargs->stripes_max);
4102 #undef CHECK_APPEND_2ARG
4103 #undef CHECK_APPEND_1ARG
4104 #undef CHECK_APPEND_NOARG
4108 if (size_bp < size_buf)
4109 buf[size_buf - size_bp - 1] = '\0'; /* remove last , */
4114 static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info)
4116 u32 size_buf = 1024;
4117 char tmp_buf[192] = {'\0'};
4120 u32 size_bp = size_buf;
4122 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4124 buf = kzalloc(size_buf, GFP_KERNEL);
4130 #define CHECK_APPEND_1ARG(a, v1) \
4132 ret = snprintf(bp, size_bp, (a), (v1)); \
4133 if (ret < 0 || ret >= size_bp) \
4134 goto out_overflow; \
4139 if (bctl->flags & BTRFS_BALANCE_FORCE)
4140 CHECK_APPEND_1ARG("%s", "-f ");
4142 if (bctl->flags & BTRFS_BALANCE_DATA) {
4143 describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf));
4144 CHECK_APPEND_1ARG("-d%s ", tmp_buf);
4147 if (bctl->flags & BTRFS_BALANCE_METADATA) {
4148 describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf));
4149 CHECK_APPEND_1ARG("-m%s ", tmp_buf);
4152 if (bctl->flags & BTRFS_BALANCE_SYSTEM) {
4153 describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf));
4154 CHECK_APPEND_1ARG("-s%s ", tmp_buf);
4157 #undef CHECK_APPEND_1ARG
4161 if (size_bp < size_buf)
4162 buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */
4163 btrfs_info(fs_info, "balance: %s %s",
4164 (bctl->flags & BTRFS_BALANCE_RESUME) ?
4165 "resume" : "start", buf);
4171 * Should be called with balance mutexe held
4173 int btrfs_balance(struct btrfs_fs_info *fs_info,
4174 struct btrfs_balance_control *bctl,
4175 struct btrfs_ioctl_balance_args *bargs)
4177 u64 meta_target, data_target;
4183 bool reducing_redundancy;
4186 if (btrfs_fs_closing(fs_info) ||
4187 atomic_read(&fs_info->balance_pause_req) ||
4188 btrfs_should_cancel_balance(fs_info)) {
4193 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
4194 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
4198 * In case of mixed groups both data and meta should be picked,
4199 * and identical options should be given for both of them.
4201 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
4202 if (mixed && (bctl->flags & allowed)) {
4203 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
4204 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
4205 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
4207 "balance: mixed groups data and metadata options must be the same");
4214 * rw_devices will not change at the moment, device add/delete/replace
4217 num_devices = fs_info->fs_devices->rw_devices;
4220 * SINGLE profile on-disk has no profile bit, but in-memory we have a
4221 * special bit for it, to make it easier to distinguish. Thus we need
4222 * to set it manually, or balance would refuse the profile.
4224 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
4225 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++)
4226 if (num_devices >= btrfs_raid_array[i].devs_min)
4227 allowed |= btrfs_raid_array[i].bg_flag;
4229 if (!validate_convert_profile(fs_info, &bctl->data, allowed, "data") ||
4230 !validate_convert_profile(fs_info, &bctl->meta, allowed, "metadata") ||
4231 !validate_convert_profile(fs_info, &bctl->sys, allowed, "system")) {
4237 * Allow to reduce metadata or system integrity only if force set for
4238 * profiles with redundancy (copies, parity)
4241 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) {
4242 if (btrfs_raid_array[i].ncopies >= 2 ||
4243 btrfs_raid_array[i].tolerated_failures >= 1)
4244 allowed |= btrfs_raid_array[i].bg_flag;
4247 seq = read_seqbegin(&fs_info->profiles_lock);
4249 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4250 (fs_info->avail_system_alloc_bits & allowed) &&
4251 !(bctl->sys.target & allowed)) ||
4252 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4253 (fs_info->avail_metadata_alloc_bits & allowed) &&
4254 !(bctl->meta.target & allowed)))
4255 reducing_redundancy = true;
4257 reducing_redundancy = false;
4259 /* if we're not converting, the target field is uninitialized */
4260 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4261 bctl->meta.target : fs_info->avail_metadata_alloc_bits;
4262 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4263 bctl->data.target : fs_info->avail_data_alloc_bits;
4264 } while (read_seqretry(&fs_info->profiles_lock, seq));
4266 if (reducing_redundancy) {
4267 if (bctl->flags & BTRFS_BALANCE_FORCE) {
4269 "balance: force reducing metadata redundancy");
4272 "balance: reduces metadata redundancy, use --force if you want this");
4278 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <
4279 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {
4281 "balance: metadata profile %s has lower redundancy than data profile %s",
4282 btrfs_bg_type_to_raid_name(meta_target),
4283 btrfs_bg_type_to_raid_name(data_target));
4286 ret = insert_balance_item(fs_info, bctl);
4287 if (ret && ret != -EEXIST)
4290 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
4291 BUG_ON(ret == -EEXIST);
4292 BUG_ON(fs_info->balance_ctl);
4293 spin_lock(&fs_info->balance_lock);
4294 fs_info->balance_ctl = bctl;
4295 spin_unlock(&fs_info->balance_lock);
4297 BUG_ON(ret != -EEXIST);
4298 spin_lock(&fs_info->balance_lock);
4299 update_balance_args(bctl);
4300 spin_unlock(&fs_info->balance_lock);
4303 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4304 set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4305 describe_balance_start_or_resume(fs_info);
4306 mutex_unlock(&fs_info->balance_mutex);
4308 ret = __btrfs_balance(fs_info);
4310 mutex_lock(&fs_info->balance_mutex);
4311 if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req))
4312 btrfs_info(fs_info, "balance: paused");
4314 * Balance can be canceled by:
4316 * - Regular cancel request
4317 * Then ret == -ECANCELED and balance_cancel_req > 0
4319 * - Fatal signal to "btrfs" process
4320 * Either the signal caught by wait_reserve_ticket() and callers
4321 * got -EINTR, or caught by btrfs_should_cancel_balance() and
4323 * Either way, in this case balance_cancel_req = 0, and
4324 * ret == -EINTR or ret == -ECANCELED.
4326 * So here we only check the return value to catch canceled balance.
4328 else if (ret == -ECANCELED || ret == -EINTR)
4329 btrfs_info(fs_info, "balance: canceled");
4331 btrfs_info(fs_info, "balance: ended with status: %d", ret);
4333 clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4336 memset(bargs, 0, sizeof(*bargs));
4337 btrfs_update_ioctl_balance_args(fs_info, bargs);
4340 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
4341 balance_need_close(fs_info)) {
4342 reset_balance_state(fs_info);
4343 btrfs_exclop_finish(fs_info);
4346 wake_up(&fs_info->balance_wait_q);
4350 if (bctl->flags & BTRFS_BALANCE_RESUME)
4351 reset_balance_state(fs_info);
4354 btrfs_exclop_finish(fs_info);
4359 static int balance_kthread(void *data)
4361 struct btrfs_fs_info *fs_info = data;
4364 mutex_lock(&fs_info->balance_mutex);
4365 if (fs_info->balance_ctl)
4366 ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL);
4367 mutex_unlock(&fs_info->balance_mutex);
4372 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
4374 struct task_struct *tsk;
4376 mutex_lock(&fs_info->balance_mutex);
4377 if (!fs_info->balance_ctl) {
4378 mutex_unlock(&fs_info->balance_mutex);
4381 mutex_unlock(&fs_info->balance_mutex);
4383 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
4384 btrfs_info(fs_info, "balance: resume skipped");
4389 * A ro->rw remount sequence should continue with the paused balance
4390 * regardless of who pauses it, system or the user as of now, so set
4393 spin_lock(&fs_info->balance_lock);
4394 fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME;
4395 spin_unlock(&fs_info->balance_lock);
4397 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
4398 return PTR_ERR_OR_ZERO(tsk);
4401 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
4403 struct btrfs_balance_control *bctl;
4404 struct btrfs_balance_item *item;
4405 struct btrfs_disk_balance_args disk_bargs;
4406 struct btrfs_path *path;
4407 struct extent_buffer *leaf;
4408 struct btrfs_key key;
4411 path = btrfs_alloc_path();
4415 key.objectid = BTRFS_BALANCE_OBJECTID;
4416 key.type = BTRFS_TEMPORARY_ITEM_KEY;
4419 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4422 if (ret > 0) { /* ret = -ENOENT; */
4427 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4433 leaf = path->nodes[0];
4434 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
4436 bctl->flags = btrfs_balance_flags(leaf, item);
4437 bctl->flags |= BTRFS_BALANCE_RESUME;
4439 btrfs_balance_data(leaf, item, &disk_bargs);
4440 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4441 btrfs_balance_meta(leaf, item, &disk_bargs);
4442 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4443 btrfs_balance_sys(leaf, item, &disk_bargs);
4444 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4447 * This should never happen, as the paused balance state is recovered
4448 * during mount without any chance of other exclusive ops to collide.
4450 * This gives the exclusive op status to balance and keeps in paused
4451 * state until user intervention (cancel or umount). If the ownership
4452 * cannot be assigned, show a message but do not fail. The balance
4453 * is in a paused state and must have fs_info::balance_ctl properly
4456 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE))
4458 "balance: cannot set exclusive op status, resume manually");
4460 btrfs_release_path(path);
4462 mutex_lock(&fs_info->balance_mutex);
4463 BUG_ON(fs_info->balance_ctl);
4464 spin_lock(&fs_info->balance_lock);
4465 fs_info->balance_ctl = bctl;
4466 spin_unlock(&fs_info->balance_lock);
4467 mutex_unlock(&fs_info->balance_mutex);
4469 btrfs_free_path(path);
4473 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4477 mutex_lock(&fs_info->balance_mutex);
4478 if (!fs_info->balance_ctl) {
4479 mutex_unlock(&fs_info->balance_mutex);
4483 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4484 atomic_inc(&fs_info->balance_pause_req);
4485 mutex_unlock(&fs_info->balance_mutex);
4487 wait_event(fs_info->balance_wait_q,
4488 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4490 mutex_lock(&fs_info->balance_mutex);
4491 /* we are good with balance_ctl ripped off from under us */
4492 BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4493 atomic_dec(&fs_info->balance_pause_req);
4498 mutex_unlock(&fs_info->balance_mutex);
4502 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4504 mutex_lock(&fs_info->balance_mutex);
4505 if (!fs_info->balance_ctl) {
4506 mutex_unlock(&fs_info->balance_mutex);
4511 * A paused balance with the item stored on disk can be resumed at
4512 * mount time if the mount is read-write. Otherwise it's still paused
4513 * and we must not allow cancelling as it deletes the item.
4515 if (sb_rdonly(fs_info->sb)) {
4516 mutex_unlock(&fs_info->balance_mutex);
4520 atomic_inc(&fs_info->balance_cancel_req);
4522 * if we are running just wait and return, balance item is
4523 * deleted in btrfs_balance in this case
4525 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4526 mutex_unlock(&fs_info->balance_mutex);
4527 wait_event(fs_info->balance_wait_q,
4528 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4529 mutex_lock(&fs_info->balance_mutex);
4531 mutex_unlock(&fs_info->balance_mutex);
4533 * Lock released to allow other waiters to continue, we'll
4534 * reexamine the status again.
4536 mutex_lock(&fs_info->balance_mutex);
4538 if (fs_info->balance_ctl) {
4539 reset_balance_state(fs_info);
4540 btrfs_exclop_finish(fs_info);
4541 btrfs_info(fs_info, "balance: canceled");
4545 BUG_ON(fs_info->balance_ctl ||
4546 test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4547 atomic_dec(&fs_info->balance_cancel_req);
4548 mutex_unlock(&fs_info->balance_mutex);
4552 int btrfs_uuid_scan_kthread(void *data)
4554 struct btrfs_fs_info *fs_info = data;
4555 struct btrfs_root *root = fs_info->tree_root;
4556 struct btrfs_key key;
4557 struct btrfs_path *path = NULL;
4559 struct extent_buffer *eb;
4561 struct btrfs_root_item root_item;
4563 struct btrfs_trans_handle *trans = NULL;
4564 bool closing = false;
4566 path = btrfs_alloc_path();
4573 key.type = BTRFS_ROOT_ITEM_KEY;
4577 if (btrfs_fs_closing(fs_info)) {
4581 ret = btrfs_search_forward(root, &key, path,
4582 BTRFS_OLDEST_GENERATION);
4589 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4590 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4591 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4592 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4595 eb = path->nodes[0];
4596 slot = path->slots[0];
4597 item_size = btrfs_item_size_nr(eb, slot);
4598 if (item_size < sizeof(root_item))
4601 read_extent_buffer(eb, &root_item,
4602 btrfs_item_ptr_offset(eb, slot),
4603 (int)sizeof(root_item));
4604 if (btrfs_root_refs(&root_item) == 0)
4607 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4608 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4612 btrfs_release_path(path);
4614 * 1 - subvol uuid item
4615 * 1 - received_subvol uuid item
4617 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4618 if (IS_ERR(trans)) {
4619 ret = PTR_ERR(trans);
4627 btrfs_release_path(path);
4628 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4629 ret = btrfs_uuid_tree_add(trans, root_item.uuid,
4630 BTRFS_UUID_KEY_SUBVOL,
4633 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4639 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4640 ret = btrfs_uuid_tree_add(trans,
4641 root_item.received_uuid,
4642 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4645 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4652 btrfs_release_path(path);
4654 ret = btrfs_end_transaction(trans);
4660 if (key.offset < (u64)-1) {
4662 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4664 key.type = BTRFS_ROOT_ITEM_KEY;
4665 } else if (key.objectid < (u64)-1) {
4667 key.type = BTRFS_ROOT_ITEM_KEY;
4676 btrfs_free_path(path);
4677 if (trans && !IS_ERR(trans))
4678 btrfs_end_transaction(trans);
4680 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4682 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4683 up(&fs_info->uuid_tree_rescan_sem);
4687 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4689 struct btrfs_trans_handle *trans;
4690 struct btrfs_root *tree_root = fs_info->tree_root;
4691 struct btrfs_root *uuid_root;
4692 struct task_struct *task;
4699 trans = btrfs_start_transaction(tree_root, 2);
4701 return PTR_ERR(trans);
4703 uuid_root = btrfs_create_tree(trans, BTRFS_UUID_TREE_OBJECTID);
4704 if (IS_ERR(uuid_root)) {
4705 ret = PTR_ERR(uuid_root);
4706 btrfs_abort_transaction(trans, ret);
4707 btrfs_end_transaction(trans);
4711 fs_info->uuid_root = uuid_root;
4713 ret = btrfs_commit_transaction(trans);
4717 down(&fs_info->uuid_tree_rescan_sem);
4718 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4720 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4721 btrfs_warn(fs_info, "failed to start uuid_scan task");
4722 up(&fs_info->uuid_tree_rescan_sem);
4723 return PTR_ERR(task);
4730 * shrinking a device means finding all of the device extents past
4731 * the new size, and then following the back refs to the chunks.
4732 * The chunk relocation code actually frees the device extent
4734 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4736 struct btrfs_fs_info *fs_info = device->fs_info;
4737 struct btrfs_root *root = fs_info->dev_root;
4738 struct btrfs_trans_handle *trans;
4739 struct btrfs_dev_extent *dev_extent = NULL;
4740 struct btrfs_path *path;
4746 bool retried = false;
4747 struct extent_buffer *l;
4748 struct btrfs_key key;
4749 struct btrfs_super_block *super_copy = fs_info->super_copy;
4750 u64 old_total = btrfs_super_total_bytes(super_copy);
4751 u64 old_size = btrfs_device_get_total_bytes(device);
4755 new_size = round_down(new_size, fs_info->sectorsize);
4757 diff = round_down(old_size - new_size, fs_info->sectorsize);
4759 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4762 path = btrfs_alloc_path();
4766 path->reada = READA_BACK;
4768 trans = btrfs_start_transaction(root, 0);
4769 if (IS_ERR(trans)) {
4770 btrfs_free_path(path);
4771 return PTR_ERR(trans);
4774 mutex_lock(&fs_info->chunk_mutex);
4776 btrfs_device_set_total_bytes(device, new_size);
4777 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4778 device->fs_devices->total_rw_bytes -= diff;
4779 atomic64_sub(diff, &fs_info->free_chunk_space);
4783 * Once the device's size has been set to the new size, ensure all
4784 * in-memory chunks are synced to disk so that the loop below sees them
4785 * and relocates them accordingly.
4787 if (contains_pending_extent(device, &start, diff)) {
4788 mutex_unlock(&fs_info->chunk_mutex);
4789 ret = btrfs_commit_transaction(trans);
4793 mutex_unlock(&fs_info->chunk_mutex);
4794 btrfs_end_transaction(trans);
4798 key.objectid = device->devid;
4799 key.offset = (u64)-1;
4800 key.type = BTRFS_DEV_EXTENT_KEY;
4803 mutex_lock(&fs_info->reclaim_bgs_lock);
4804 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4806 mutex_unlock(&fs_info->reclaim_bgs_lock);
4810 ret = btrfs_previous_item(root, path, 0, key.type);
4812 mutex_unlock(&fs_info->reclaim_bgs_lock);
4816 btrfs_release_path(path);
4821 slot = path->slots[0];
4822 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4824 if (key.objectid != device->devid) {
4825 mutex_unlock(&fs_info->reclaim_bgs_lock);
4826 btrfs_release_path(path);
4830 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4831 length = btrfs_dev_extent_length(l, dev_extent);
4833 if (key.offset + length <= new_size) {
4834 mutex_unlock(&fs_info->reclaim_bgs_lock);
4835 btrfs_release_path(path);
4839 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4840 btrfs_release_path(path);
4843 * We may be relocating the only data chunk we have,
4844 * which could potentially end up with losing data's
4845 * raid profile, so lets allocate an empty one in
4848 ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset);
4850 mutex_unlock(&fs_info->reclaim_bgs_lock);
4854 ret = btrfs_relocate_chunk(fs_info, chunk_offset);
4855 mutex_unlock(&fs_info->reclaim_bgs_lock);
4856 if (ret == -ENOSPC) {
4859 if (ret == -ETXTBSY) {
4861 "could not shrink block group %llu due to active swapfile",
4866 } while (key.offset-- > 0);
4868 if (failed && !retried) {
4872 } else if (failed && retried) {
4877 /* Shrinking succeeded, else we would be at "done". */
4878 trans = btrfs_start_transaction(root, 0);
4879 if (IS_ERR(trans)) {
4880 ret = PTR_ERR(trans);
4884 mutex_lock(&fs_info->chunk_mutex);
4885 /* Clear all state bits beyond the shrunk device size */
4886 clear_extent_bits(&device->alloc_state, new_size, (u64)-1,
4889 btrfs_device_set_disk_total_bytes(device, new_size);
4890 if (list_empty(&device->post_commit_list))
4891 list_add_tail(&device->post_commit_list,
4892 &trans->transaction->dev_update_list);
4894 WARN_ON(diff > old_total);
4895 btrfs_set_super_total_bytes(super_copy,
4896 round_down(old_total - diff, fs_info->sectorsize));
4897 mutex_unlock(&fs_info->chunk_mutex);
4899 /* Now btrfs_update_device() will change the on-disk size. */
4900 ret = btrfs_update_device(trans, device);
4902 btrfs_abort_transaction(trans, ret);
4903 btrfs_end_transaction(trans);
4905 ret = btrfs_commit_transaction(trans);
4908 btrfs_free_path(path);
4910 mutex_lock(&fs_info->chunk_mutex);
4911 btrfs_device_set_total_bytes(device, old_size);
4912 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
4913 device->fs_devices->total_rw_bytes += diff;
4914 atomic64_add(diff, &fs_info->free_chunk_space);
4915 mutex_unlock(&fs_info->chunk_mutex);
4920 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
4921 struct btrfs_key *key,
4922 struct btrfs_chunk *chunk, int item_size)
4924 struct btrfs_super_block *super_copy = fs_info->super_copy;
4925 struct btrfs_disk_key disk_key;
4929 lockdep_assert_held(&fs_info->chunk_mutex);
4931 array_size = btrfs_super_sys_array_size(super_copy);
4932 if (array_size + item_size + sizeof(disk_key)
4933 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
4936 ptr = super_copy->sys_chunk_array + array_size;
4937 btrfs_cpu_key_to_disk(&disk_key, key);
4938 memcpy(ptr, &disk_key, sizeof(disk_key));
4939 ptr += sizeof(disk_key);
4940 memcpy(ptr, chunk, item_size);
4941 item_size += sizeof(disk_key);
4942 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4948 * sort the devices in descending order by max_avail, total_avail
4950 static int btrfs_cmp_device_info(const void *a, const void *b)
4952 const struct btrfs_device_info *di_a = a;
4953 const struct btrfs_device_info *di_b = b;
4955 if (di_a->max_avail > di_b->max_avail)
4957 if (di_a->max_avail < di_b->max_avail)
4959 if (di_a->total_avail > di_b->total_avail)
4961 if (di_a->total_avail < di_b->total_avail)
4966 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4968 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4971 btrfs_set_fs_incompat(info, RAID56);
4974 static void check_raid1c34_incompat_flag(struct btrfs_fs_info *info, u64 type)
4976 if (!(type & (BTRFS_BLOCK_GROUP_RAID1C3 | BTRFS_BLOCK_GROUP_RAID1C4)))
4979 btrfs_set_fs_incompat(info, RAID1C34);
4983 * Structure used internally for __btrfs_alloc_chunk() function.
4984 * Wraps needed parameters.
4986 struct alloc_chunk_ctl {
4989 /* Total number of stripes to allocate */
4991 /* sub_stripes info for map */
4993 /* Stripes per device */
4995 /* Maximum number of devices to use */
4997 /* Minimum number of devices to use */
4999 /* ndevs has to be a multiple of this */
5001 /* Number of copies */
5003 /* Number of stripes worth of bytes to store parity information */
5005 u64 max_stripe_size;
5013 static void init_alloc_chunk_ctl_policy_regular(
5014 struct btrfs_fs_devices *fs_devices,
5015 struct alloc_chunk_ctl *ctl)
5017 u64 type = ctl->type;
5019 if (type & BTRFS_BLOCK_GROUP_DATA) {
5020 ctl->max_stripe_size = SZ_1G;
5021 ctl->max_chunk_size = BTRFS_MAX_DATA_CHUNK_SIZE;
5022 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
5023 /* For larger filesystems, use larger metadata chunks */
5024 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
5025 ctl->max_stripe_size = SZ_1G;
5027 ctl->max_stripe_size = SZ_256M;
5028 ctl->max_chunk_size = ctl->max_stripe_size;
5029 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
5030 ctl->max_stripe_size = SZ_32M;
5031 ctl->max_chunk_size = 2 * ctl->max_stripe_size;
5032 ctl->devs_max = min_t(int, ctl->devs_max,
5033 BTRFS_MAX_DEVS_SYS_CHUNK);
5038 /* We don't want a chunk larger than 10% of writable space */
5039 ctl->max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
5040 ctl->max_chunk_size);
5041 ctl->dev_extent_min = BTRFS_STRIPE_LEN * ctl->dev_stripes;
5044 static void init_alloc_chunk_ctl_policy_zoned(
5045 struct btrfs_fs_devices *fs_devices,
5046 struct alloc_chunk_ctl *ctl)
5048 u64 zone_size = fs_devices->fs_info->zone_size;
5050 int min_num_stripes = ctl->devs_min * ctl->dev_stripes;
5051 int min_data_stripes = (min_num_stripes - ctl->nparity) / ctl->ncopies;
5052 u64 min_chunk_size = min_data_stripes * zone_size;
5053 u64 type = ctl->type;
5055 ctl->max_stripe_size = zone_size;
5056 if (type & BTRFS_BLOCK_GROUP_DATA) {
5057 ctl->max_chunk_size = round_down(BTRFS_MAX_DATA_CHUNK_SIZE,
5059 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
5060 ctl->max_chunk_size = ctl->max_stripe_size;
5061 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
5062 ctl->max_chunk_size = 2 * ctl->max_stripe_size;
5063 ctl->devs_max = min_t(int, ctl->devs_max,
5064 BTRFS_MAX_DEVS_SYS_CHUNK);
5069 /* We don't want a chunk larger than 10% of writable space */
5070 limit = max(round_down(div_factor(fs_devices->total_rw_bytes, 1),
5073 ctl->max_chunk_size = min(limit, ctl->max_chunk_size);
5074 ctl->dev_extent_min = zone_size * ctl->dev_stripes;
5077 static void init_alloc_chunk_ctl(struct btrfs_fs_devices *fs_devices,
5078 struct alloc_chunk_ctl *ctl)
5080 int index = btrfs_bg_flags_to_raid_index(ctl->type);
5082 ctl->sub_stripes = btrfs_raid_array[index].sub_stripes;
5083 ctl->dev_stripes = btrfs_raid_array[index].dev_stripes;
5084 ctl->devs_max = btrfs_raid_array[index].devs_max;
5086 ctl->devs_max = BTRFS_MAX_DEVS(fs_devices->fs_info);
5087 ctl->devs_min = btrfs_raid_array[index].devs_min;
5088 ctl->devs_increment = btrfs_raid_array[index].devs_increment;
5089 ctl->ncopies = btrfs_raid_array[index].ncopies;
5090 ctl->nparity = btrfs_raid_array[index].nparity;
5093 switch (fs_devices->chunk_alloc_policy) {
5094 case BTRFS_CHUNK_ALLOC_REGULAR:
5095 init_alloc_chunk_ctl_policy_regular(fs_devices, ctl);
5097 case BTRFS_CHUNK_ALLOC_ZONED:
5098 init_alloc_chunk_ctl_policy_zoned(fs_devices, ctl);
5105 static int gather_device_info(struct btrfs_fs_devices *fs_devices,
5106 struct alloc_chunk_ctl *ctl,
5107 struct btrfs_device_info *devices_info)
5109 struct btrfs_fs_info *info = fs_devices->fs_info;
5110 struct btrfs_device *device;
5112 u64 dev_extent_want = ctl->max_stripe_size * ctl->dev_stripes;
5119 * in the first pass through the devices list, we gather information
5120 * about the available holes on each device.
5122 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
5123 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
5125 "BTRFS: read-only device in alloc_list\n");
5129 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
5130 &device->dev_state) ||
5131 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
5134 if (device->total_bytes > device->bytes_used)
5135 total_avail = device->total_bytes - device->bytes_used;
5139 /* If there is no space on this device, skip it. */
5140 if (total_avail < ctl->dev_extent_min)
5143 ret = find_free_dev_extent(device, dev_extent_want, &dev_offset,
5145 if (ret && ret != -ENOSPC)
5149 max_avail = dev_extent_want;
5151 if (max_avail < ctl->dev_extent_min) {
5152 if (btrfs_test_opt(info, ENOSPC_DEBUG))
5154 "%s: devid %llu has no free space, have=%llu want=%llu",
5155 __func__, device->devid, max_avail,
5156 ctl->dev_extent_min);
5160 if (ndevs == fs_devices->rw_devices) {
5161 WARN(1, "%s: found more than %llu devices\n",
5162 __func__, fs_devices->rw_devices);
5165 devices_info[ndevs].dev_offset = dev_offset;
5166 devices_info[ndevs].max_avail = max_avail;
5167 devices_info[ndevs].total_avail = total_avail;
5168 devices_info[ndevs].dev = device;
5174 * now sort the devices by hole size / available space
5176 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
5177 btrfs_cmp_device_info, NULL);
5182 static int decide_stripe_size_regular(struct alloc_chunk_ctl *ctl,
5183 struct btrfs_device_info *devices_info)
5185 /* Number of stripes that count for block group size */
5189 * The primary goal is to maximize the number of stripes, so use as
5190 * many devices as possible, even if the stripes are not maximum sized.
5192 * The DUP profile stores more than one stripe per device, the
5193 * max_avail is the total size so we have to adjust.
5195 ctl->stripe_size = div_u64(devices_info[ctl->ndevs - 1].max_avail,
5197 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5199 /* This will have to be fixed for RAID1 and RAID10 over more drives */
5200 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5203 * Use the number of data stripes to figure out how big this chunk is
5204 * really going to be in terms of logical address space, and compare
5205 * that answer with the max chunk size. If it's higher, we try to
5206 * reduce stripe_size.
5208 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) {
5210 * Reduce stripe_size, round it up to a 16MB boundary again and
5211 * then use it, unless it ends up being even bigger than the
5212 * previous value we had already.
5214 ctl->stripe_size = min(round_up(div_u64(ctl->max_chunk_size,
5215 data_stripes), SZ_16M),
5219 /* Align to BTRFS_STRIPE_LEN */
5220 ctl->stripe_size = round_down(ctl->stripe_size, BTRFS_STRIPE_LEN);
5221 ctl->chunk_size = ctl->stripe_size * data_stripes;
5226 static int decide_stripe_size_zoned(struct alloc_chunk_ctl *ctl,
5227 struct btrfs_device_info *devices_info)
5229 u64 zone_size = devices_info[0].dev->zone_info->zone_size;
5230 /* Number of stripes that count for block group size */
5234 * It should hold because:
5235 * dev_extent_min == dev_extent_want == zone_size * dev_stripes
5237 ASSERT(devices_info[ctl->ndevs - 1].max_avail == ctl->dev_extent_min);
5239 ctl->stripe_size = zone_size;
5240 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5241 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5243 /* stripe_size is fixed in zoned filesysmte. Reduce ndevs instead. */
5244 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) {
5245 ctl->ndevs = div_u64(div_u64(ctl->max_chunk_size * ctl->ncopies,
5246 ctl->stripe_size) + ctl->nparity,
5248 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5249 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5250 ASSERT(ctl->stripe_size * data_stripes <= ctl->max_chunk_size);
5253 ctl->chunk_size = ctl->stripe_size * data_stripes;
5258 static int decide_stripe_size(struct btrfs_fs_devices *fs_devices,
5259 struct alloc_chunk_ctl *ctl,
5260 struct btrfs_device_info *devices_info)
5262 struct btrfs_fs_info *info = fs_devices->fs_info;
5265 * Round down to number of usable stripes, devs_increment can be any
5266 * number so we can't use round_down() that requires power of 2, while
5267 * rounddown is safe.
5269 ctl->ndevs = rounddown(ctl->ndevs, ctl->devs_increment);
5271 if (ctl->ndevs < ctl->devs_min) {
5272 if (btrfs_test_opt(info, ENOSPC_DEBUG)) {
5274 "%s: not enough devices with free space: have=%d minimum required=%d",
5275 __func__, ctl->ndevs, ctl->devs_min);
5280 ctl->ndevs = min(ctl->ndevs, ctl->devs_max);
5282 switch (fs_devices->chunk_alloc_policy) {
5283 case BTRFS_CHUNK_ALLOC_REGULAR:
5284 return decide_stripe_size_regular(ctl, devices_info);
5285 case BTRFS_CHUNK_ALLOC_ZONED:
5286 return decide_stripe_size_zoned(ctl, devices_info);
5292 static struct btrfs_block_group *create_chunk(struct btrfs_trans_handle *trans,
5293 struct alloc_chunk_ctl *ctl,
5294 struct btrfs_device_info *devices_info)
5296 struct btrfs_fs_info *info = trans->fs_info;
5297 struct map_lookup *map = NULL;
5298 struct extent_map_tree *em_tree;
5299 struct btrfs_block_group *block_group;
5300 struct extent_map *em;
5301 u64 start = ctl->start;
5302 u64 type = ctl->type;
5307 map = kmalloc(map_lookup_size(ctl->num_stripes), GFP_NOFS);
5309 return ERR_PTR(-ENOMEM);
5310 map->num_stripes = ctl->num_stripes;
5312 for (i = 0; i < ctl->ndevs; ++i) {
5313 for (j = 0; j < ctl->dev_stripes; ++j) {
5314 int s = i * ctl->dev_stripes + j;
5315 map->stripes[s].dev = devices_info[i].dev;
5316 map->stripes[s].physical = devices_info[i].dev_offset +
5317 j * ctl->stripe_size;
5320 map->stripe_len = BTRFS_STRIPE_LEN;
5321 map->io_align = BTRFS_STRIPE_LEN;
5322 map->io_width = BTRFS_STRIPE_LEN;
5324 map->sub_stripes = ctl->sub_stripes;
5326 trace_btrfs_chunk_alloc(info, map, start, ctl->chunk_size);
5328 em = alloc_extent_map();
5331 return ERR_PTR(-ENOMEM);
5333 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
5334 em->map_lookup = map;
5336 em->len = ctl->chunk_size;
5337 em->block_start = 0;
5338 em->block_len = em->len;
5339 em->orig_block_len = ctl->stripe_size;
5341 em_tree = &info->mapping_tree;
5342 write_lock(&em_tree->lock);
5343 ret = add_extent_mapping(em_tree, em, 0);
5345 write_unlock(&em_tree->lock);
5346 free_extent_map(em);
5347 return ERR_PTR(ret);
5349 write_unlock(&em_tree->lock);
5351 block_group = btrfs_make_block_group(trans, 0, type, start, ctl->chunk_size);
5352 if (IS_ERR(block_group))
5353 goto error_del_extent;
5355 for (i = 0; i < map->num_stripes; i++) {
5356 struct btrfs_device *dev = map->stripes[i].dev;
5358 btrfs_device_set_bytes_used(dev,
5359 dev->bytes_used + ctl->stripe_size);
5360 if (list_empty(&dev->post_commit_list))
5361 list_add_tail(&dev->post_commit_list,
5362 &trans->transaction->dev_update_list);
5365 atomic64_sub(ctl->stripe_size * map->num_stripes,
5366 &info->free_chunk_space);
5368 free_extent_map(em);
5369 check_raid56_incompat_flag(info, type);
5370 check_raid1c34_incompat_flag(info, type);
5375 write_lock(&em_tree->lock);
5376 remove_extent_mapping(em_tree, em);
5377 write_unlock(&em_tree->lock);
5379 /* One for our allocation */
5380 free_extent_map(em);
5381 /* One for the tree reference */
5382 free_extent_map(em);
5387 struct btrfs_block_group *btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
5390 struct btrfs_fs_info *info = trans->fs_info;
5391 struct btrfs_fs_devices *fs_devices = info->fs_devices;
5392 struct btrfs_device_info *devices_info = NULL;
5393 struct alloc_chunk_ctl ctl;
5394 struct btrfs_block_group *block_group;
5397 lockdep_assert_held(&info->chunk_mutex);
5399 if (!alloc_profile_is_valid(type, 0)) {
5401 return ERR_PTR(-EINVAL);
5404 if (list_empty(&fs_devices->alloc_list)) {
5405 if (btrfs_test_opt(info, ENOSPC_DEBUG))
5406 btrfs_debug(info, "%s: no writable device", __func__);
5407 return ERR_PTR(-ENOSPC);
5410 if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
5411 btrfs_err(info, "invalid chunk type 0x%llx requested", type);
5413 return ERR_PTR(-EINVAL);
5416 ctl.start = find_next_chunk(info);
5418 init_alloc_chunk_ctl(fs_devices, &ctl);
5420 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
5423 return ERR_PTR(-ENOMEM);
5425 ret = gather_device_info(fs_devices, &ctl, devices_info);
5427 block_group = ERR_PTR(ret);
5431 ret = decide_stripe_size(fs_devices, &ctl, devices_info);
5433 block_group = ERR_PTR(ret);
5437 block_group = create_chunk(trans, &ctl, devices_info);
5440 kfree(devices_info);
5445 * This function, btrfs_chunk_alloc_add_chunk_item(), typically belongs to the
5446 * phase 1 of chunk allocation. It belongs to phase 2 only when allocating system
5449 * See the comment at btrfs_chunk_alloc() for details about the chunk allocation
5452 int btrfs_chunk_alloc_add_chunk_item(struct btrfs_trans_handle *trans,
5453 struct btrfs_block_group *bg)
5455 struct btrfs_fs_info *fs_info = trans->fs_info;
5456 struct btrfs_root *extent_root = fs_info->extent_root;
5457 struct btrfs_root *chunk_root = fs_info->chunk_root;
5458 struct btrfs_key key;
5459 struct btrfs_chunk *chunk;
5460 struct btrfs_stripe *stripe;
5461 struct extent_map *em;
5462 struct map_lookup *map;
5468 * We take the chunk_mutex for 2 reasons:
5470 * 1) Updates and insertions in the chunk btree must be done while holding
5471 * the chunk_mutex, as well as updating the system chunk array in the
5472 * superblock. See the comment on top of btrfs_chunk_alloc() for the
5475 * 2) To prevent races with the final phase of a device replace operation
5476 * that replaces the device object associated with the map's stripes,
5477 * because the device object's id can change at any time during that
5478 * final phase of the device replace operation
5479 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the
5480 * replaced device and then see it with an ID of BTRFS_DEV_REPLACE_DEVID,
5481 * which would cause a failure when updating the device item, which does
5482 * not exists, or persisting a stripe of the chunk item with such ID.
5483 * Here we can't use the device_list_mutex because our caller already
5484 * has locked the chunk_mutex, and the final phase of device replace
5485 * acquires both mutexes - first the device_list_mutex and then the
5486 * chunk_mutex. Using any of those two mutexes protects us from a
5487 * concurrent device replace.
5489 lockdep_assert_held(&fs_info->chunk_mutex);
5491 em = btrfs_get_chunk_map(fs_info, bg->start, bg->length);
5494 btrfs_abort_transaction(trans, ret);
5498 map = em->map_lookup;
5499 item_size = btrfs_chunk_item_size(map->num_stripes);
5501 chunk = kzalloc(item_size, GFP_NOFS);
5504 btrfs_abort_transaction(trans, ret);
5508 for (i = 0; i < map->num_stripes; i++) {
5509 struct btrfs_device *device = map->stripes[i].dev;
5511 ret = btrfs_update_device(trans, device);
5516 stripe = &chunk->stripe;
5517 for (i = 0; i < map->num_stripes; i++) {
5518 struct btrfs_device *device = map->stripes[i].dev;
5519 const u64 dev_offset = map->stripes[i].physical;
5521 btrfs_set_stack_stripe_devid(stripe, device->devid);
5522 btrfs_set_stack_stripe_offset(stripe, dev_offset);
5523 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
5527 btrfs_set_stack_chunk_length(chunk, bg->length);
5528 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
5529 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
5530 btrfs_set_stack_chunk_type(chunk, map->type);
5531 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
5532 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
5533 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
5534 btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
5535 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
5537 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
5538 key.type = BTRFS_CHUNK_ITEM_KEY;
5539 key.offset = bg->start;
5541 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
5545 bg->chunk_item_inserted = 1;
5547 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
5548 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
5555 free_extent_map(em);
5559 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans)
5561 struct btrfs_fs_info *fs_info = trans->fs_info;
5563 struct btrfs_block_group *meta_bg;
5564 struct btrfs_block_group *sys_bg;
5567 * When adding a new device for sprouting, the seed device is read-only
5568 * so we must first allocate a metadata and a system chunk. But before
5569 * adding the block group items to the extent, device and chunk btrees,
5572 * 1) Create both chunks without doing any changes to the btrees, as
5573 * otherwise we would get -ENOSPC since the block groups from the
5574 * seed device are read-only;
5576 * 2) Add the device item for the new sprout device - finishing the setup
5577 * of a new block group requires updating the device item in the chunk
5578 * btree, so it must exist when we attempt to do it. The previous step
5579 * ensures this does not fail with -ENOSPC.
5581 * After that we can add the block group items to their btrees:
5582 * update existing device item in the chunk btree, add a new block group
5583 * item to the extent btree, add a new chunk item to the chunk btree and
5584 * finally add the new device extent items to the devices btree.
5587 alloc_profile = btrfs_metadata_alloc_profile(fs_info);
5588 meta_bg = btrfs_alloc_chunk(trans, alloc_profile);
5589 if (IS_ERR(meta_bg))
5590 return PTR_ERR(meta_bg);
5592 alloc_profile = btrfs_system_alloc_profile(fs_info);
5593 sys_bg = btrfs_alloc_chunk(trans, alloc_profile);
5595 return PTR_ERR(sys_bg);
5600 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
5602 const int index = btrfs_bg_flags_to_raid_index(map->type);
5604 return btrfs_raid_array[index].tolerated_failures;
5607 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset)
5609 struct extent_map *em;
5610 struct map_lookup *map;
5615 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
5619 map = em->map_lookup;
5620 for (i = 0; i < map->num_stripes; i++) {
5621 if (test_bit(BTRFS_DEV_STATE_MISSING,
5622 &map->stripes[i].dev->dev_state)) {
5626 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE,
5627 &map->stripes[i].dev->dev_state)) {
5634 * If the number of missing devices is larger than max errors,
5635 * we can not write the data into that chunk successfully, so
5638 if (miss_ndevs > btrfs_chunk_max_errors(map))
5641 free_extent_map(em);
5645 void btrfs_mapping_tree_free(struct extent_map_tree *tree)
5647 struct extent_map *em;
5650 write_lock(&tree->lock);
5651 em = lookup_extent_mapping(tree, 0, (u64)-1);
5653 remove_extent_mapping(tree, em);
5654 write_unlock(&tree->lock);
5658 free_extent_map(em);
5659 /* once for the tree */
5660 free_extent_map(em);
5664 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5666 struct extent_map *em;
5667 struct map_lookup *map;
5670 em = btrfs_get_chunk_map(fs_info, logical, len);
5673 * We could return errors for these cases, but that could get
5674 * ugly and we'd probably do the same thing which is just not do
5675 * anything else and exit, so return 1 so the callers don't try
5676 * to use other copies.
5680 map = em->map_lookup;
5681 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1_MASK))
5682 ret = map->num_stripes;
5683 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5684 ret = map->sub_stripes;
5685 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5687 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5689 * There could be two corrupted data stripes, we need
5690 * to loop retry in order to rebuild the correct data.
5692 * Fail a stripe at a time on every retry except the
5693 * stripe under reconstruction.
5695 ret = map->num_stripes;
5698 free_extent_map(em);
5700 down_read(&fs_info->dev_replace.rwsem);
5701 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) &&
5702 fs_info->dev_replace.tgtdev)
5704 up_read(&fs_info->dev_replace.rwsem);
5709 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
5712 struct extent_map *em;
5713 struct map_lookup *map;
5714 unsigned long len = fs_info->sectorsize;
5716 em = btrfs_get_chunk_map(fs_info, logical, len);
5718 if (!WARN_ON(IS_ERR(em))) {
5719 map = em->map_lookup;
5720 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5721 len = map->stripe_len * nr_data_stripes(map);
5722 free_extent_map(em);
5727 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5729 struct extent_map *em;
5730 struct map_lookup *map;
5733 em = btrfs_get_chunk_map(fs_info, logical, len);
5735 if(!WARN_ON(IS_ERR(em))) {
5736 map = em->map_lookup;
5737 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5739 free_extent_map(em);
5744 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5745 struct map_lookup *map, int first,
5746 int dev_replace_is_ongoing)
5750 int preferred_mirror;
5752 struct btrfs_device *srcdev;
5755 (BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10)));
5757 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5758 num_stripes = map->sub_stripes;
5760 num_stripes = map->num_stripes;
5762 switch (fs_info->fs_devices->read_policy) {
5764 /* Shouldn't happen, just warn and use pid instead of failing */
5765 btrfs_warn_rl(fs_info,
5766 "unknown read_policy type %u, reset to pid",
5767 fs_info->fs_devices->read_policy);
5768 fs_info->fs_devices->read_policy = BTRFS_READ_POLICY_PID;
5770 case BTRFS_READ_POLICY_PID:
5771 preferred_mirror = first + (current->pid % num_stripes);
5775 if (dev_replace_is_ongoing &&
5776 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5777 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5778 srcdev = fs_info->dev_replace.srcdev;
5783 * try to avoid the drive that is the source drive for a
5784 * dev-replace procedure, only choose it if no other non-missing
5785 * mirror is available
5787 for (tolerance = 0; tolerance < 2; tolerance++) {
5788 if (map->stripes[preferred_mirror].dev->bdev &&
5789 (tolerance || map->stripes[preferred_mirror].dev != srcdev))
5790 return preferred_mirror;
5791 for (i = first; i < first + num_stripes; i++) {
5792 if (map->stripes[i].dev->bdev &&
5793 (tolerance || map->stripes[i].dev != srcdev))
5798 /* we couldn't find one that doesn't fail. Just return something
5799 * and the io error handling code will clean up eventually
5801 return preferred_mirror;
5804 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5805 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5812 for (i = 0; i < num_stripes - 1; i++) {
5813 /* Swap if parity is on a smaller index */
5814 if (bbio->raid_map[i] > bbio->raid_map[i + 1]) {
5815 swap(bbio->stripes[i], bbio->stripes[i + 1]);
5816 swap(bbio->raid_map[i], bbio->raid_map[i + 1]);
5823 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5825 struct btrfs_bio *bbio = kzalloc(
5826 /* the size of the btrfs_bio */
5827 sizeof(struct btrfs_bio) +
5828 /* plus the variable array for the stripes */
5829 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5830 /* plus the variable array for the tgt dev */
5831 sizeof(int) * (real_stripes) +
5833 * plus the raid_map, which includes both the tgt dev
5836 sizeof(u64) * (total_stripes),
5837 GFP_NOFS|__GFP_NOFAIL);
5839 atomic_set(&bbio->error, 0);
5840 refcount_set(&bbio->refs, 1);
5842 bbio->tgtdev_map = (int *)(bbio->stripes + total_stripes);
5843 bbio->raid_map = (u64 *)(bbio->tgtdev_map + real_stripes);
5848 void btrfs_get_bbio(struct btrfs_bio *bbio)
5850 WARN_ON(!refcount_read(&bbio->refs));
5851 refcount_inc(&bbio->refs);
5854 void btrfs_put_bbio(struct btrfs_bio *bbio)
5858 if (refcount_dec_and_test(&bbio->refs))
5862 /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */
5864 * Please note that, discard won't be sent to target device of device
5867 static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info,
5868 u64 logical, u64 *length_ret,
5869 struct btrfs_bio **bbio_ret)
5871 struct extent_map *em;
5872 struct map_lookup *map;
5873 struct btrfs_bio *bbio;
5874 u64 length = *length_ret;
5878 u64 stripe_end_offset;
5885 u32 sub_stripes = 0;
5886 u64 stripes_per_dev = 0;
5887 u32 remaining_stripes = 0;
5888 u32 last_stripe = 0;
5892 /* discard always return a bbio */
5895 em = btrfs_get_chunk_map(fs_info, logical, length);
5899 map = em->map_lookup;
5900 /* we don't discard raid56 yet */
5901 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5906 offset = logical - em->start;
5907 length = min_t(u64, em->start + em->len - logical, length);
5908 *length_ret = length;
5910 stripe_len = map->stripe_len;
5912 * stripe_nr counts the total number of stripes we have to stride
5913 * to get to this block
5915 stripe_nr = div64_u64(offset, stripe_len);
5917 /* stripe_offset is the offset of this block in its stripe */
5918 stripe_offset = offset - stripe_nr * stripe_len;
5920 stripe_nr_end = round_up(offset + length, map->stripe_len);
5921 stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len);
5922 stripe_cnt = stripe_nr_end - stripe_nr;
5923 stripe_end_offset = stripe_nr_end * map->stripe_len -
5926 * after this, stripe_nr is the number of stripes on this
5927 * device we have to walk to find the data, and stripe_index is
5928 * the number of our device in the stripe array
5932 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5933 BTRFS_BLOCK_GROUP_RAID10)) {
5934 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5937 sub_stripes = map->sub_stripes;
5939 factor = map->num_stripes / sub_stripes;
5940 num_stripes = min_t(u64, map->num_stripes,
5941 sub_stripes * stripe_cnt);
5942 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5943 stripe_index *= sub_stripes;
5944 stripes_per_dev = div_u64_rem(stripe_cnt, factor,
5945 &remaining_stripes);
5946 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5947 last_stripe *= sub_stripes;
5948 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1_MASK |
5949 BTRFS_BLOCK_GROUP_DUP)) {
5950 num_stripes = map->num_stripes;
5952 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5956 bbio = alloc_btrfs_bio(num_stripes, 0);
5962 for (i = 0; i < num_stripes; i++) {
5963 bbio->stripes[i].physical =
5964 map->stripes[stripe_index].physical +
5965 stripe_offset + stripe_nr * map->stripe_len;
5966 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5968 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5969 BTRFS_BLOCK_GROUP_RAID10)) {
5970 bbio->stripes[i].length = stripes_per_dev *
5973 if (i / sub_stripes < remaining_stripes)
5974 bbio->stripes[i].length +=
5978 * Special for the first stripe and
5981 * |-------|...|-------|
5985 if (i < sub_stripes)
5986 bbio->stripes[i].length -=
5989 if (stripe_index >= last_stripe &&
5990 stripe_index <= (last_stripe +
5992 bbio->stripes[i].length -=
5995 if (i == sub_stripes - 1)
5998 bbio->stripes[i].length = length;
6002 if (stripe_index == map->num_stripes) {
6009 bbio->map_type = map->type;
6010 bbio->num_stripes = num_stripes;
6012 free_extent_map(em);
6017 * In dev-replace case, for repair case (that's the only case where the mirror
6018 * is selected explicitly when calling btrfs_map_block), blocks left of the
6019 * left cursor can also be read from the target drive.
6021 * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the
6023 * For READ, it also needs to be supported using the same mirror number.
6025 * If the requested block is not left of the left cursor, EIO is returned. This
6026 * can happen because btrfs_num_copies() returns one more in the dev-replace
6029 static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info,
6030 u64 logical, u64 length,
6031 u64 srcdev_devid, int *mirror_num,
6034 struct btrfs_bio *bbio = NULL;
6036 int index_srcdev = 0;
6038 u64 physical_of_found = 0;
6042 ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS,
6043 logical, &length, &bbio, 0, 0);
6045 ASSERT(bbio == NULL);
6049 num_stripes = bbio->num_stripes;
6050 if (*mirror_num > num_stripes) {
6052 * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror,
6053 * that means that the requested area is not left of the left
6056 btrfs_put_bbio(bbio);
6061 * process the rest of the function using the mirror_num of the source
6062 * drive. Therefore look it up first. At the end, patch the device
6063 * pointer to the one of the target drive.
6065 for (i = 0; i < num_stripes; i++) {
6066 if (bbio->stripes[i].dev->devid != srcdev_devid)
6070 * In case of DUP, in order to keep it simple, only add the
6071 * mirror with the lowest physical address
6074 physical_of_found <= bbio->stripes[i].physical)
6079 physical_of_found = bbio->stripes[i].physical;
6082 btrfs_put_bbio(bbio);
6088 *mirror_num = index_srcdev + 1;
6089 *physical = physical_of_found;
6093 static bool is_block_group_to_copy(struct btrfs_fs_info *fs_info, u64 logical)
6095 struct btrfs_block_group *cache;
6098 /* Non zoned filesystem does not use "to_copy" flag */
6099 if (!btrfs_is_zoned(fs_info))
6102 cache = btrfs_lookup_block_group(fs_info, logical);
6104 spin_lock(&cache->lock);
6105 ret = cache->to_copy;
6106 spin_unlock(&cache->lock);
6108 btrfs_put_block_group(cache);
6112 static void handle_ops_on_dev_replace(enum btrfs_map_op op,
6113 struct btrfs_bio **bbio_ret,
6114 struct btrfs_dev_replace *dev_replace,
6116 int *num_stripes_ret, int *max_errors_ret)
6118 struct btrfs_bio *bbio = *bbio_ret;
6119 u64 srcdev_devid = dev_replace->srcdev->devid;
6120 int tgtdev_indexes = 0;
6121 int num_stripes = *num_stripes_ret;
6122 int max_errors = *max_errors_ret;
6125 if (op == BTRFS_MAP_WRITE) {
6126 int index_where_to_add;
6129 * A block group which have "to_copy" set will eventually
6130 * copied by dev-replace process. We can avoid cloning IO here.
6132 if (is_block_group_to_copy(dev_replace->srcdev->fs_info, logical))
6136 * duplicate the write operations while the dev replace
6137 * procedure is running. Since the copying of the old disk to
6138 * the new disk takes place at run time while the filesystem is
6139 * mounted writable, the regular write operations to the old
6140 * disk have to be duplicated to go to the new disk as well.
6142 * Note that device->missing is handled by the caller, and that
6143 * the write to the old disk is already set up in the stripes
6146 index_where_to_add = num_stripes;
6147 for (i = 0; i < num_stripes; i++) {
6148 if (bbio->stripes[i].dev->devid == srcdev_devid) {
6149 /* write to new disk, too */
6150 struct btrfs_bio_stripe *new =
6151 bbio->stripes + index_where_to_add;
6152 struct btrfs_bio_stripe *old =
6155 new->physical = old->physical;
6156 new->length = old->length;
6157 new->dev = dev_replace->tgtdev;
6158 bbio->tgtdev_map[i] = index_where_to_add;
6159 index_where_to_add++;
6164 num_stripes = index_where_to_add;
6165 } else if (op == BTRFS_MAP_GET_READ_MIRRORS) {
6166 int index_srcdev = 0;
6168 u64 physical_of_found = 0;
6171 * During the dev-replace procedure, the target drive can also
6172 * be used to read data in case it is needed to repair a corrupt
6173 * block elsewhere. This is possible if the requested area is
6174 * left of the left cursor. In this area, the target drive is a
6175 * full copy of the source drive.
6177 for (i = 0; i < num_stripes; i++) {
6178 if (bbio->stripes[i].dev->devid == srcdev_devid) {
6180 * In case of DUP, in order to keep it simple,
6181 * only add the mirror with the lowest physical
6185 physical_of_found <=
6186 bbio->stripes[i].physical)
6190 physical_of_found = bbio->stripes[i].physical;
6194 struct btrfs_bio_stripe *tgtdev_stripe =
6195 bbio->stripes + num_stripes;
6197 tgtdev_stripe->physical = physical_of_found;
6198 tgtdev_stripe->length =
6199 bbio->stripes[index_srcdev].length;
6200 tgtdev_stripe->dev = dev_replace->tgtdev;
6201 bbio->tgtdev_map[index_srcdev] = num_stripes;
6208 *num_stripes_ret = num_stripes;
6209 *max_errors_ret = max_errors;
6210 bbio->num_tgtdevs = tgtdev_indexes;
6214 static bool need_full_stripe(enum btrfs_map_op op)
6216 return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS);
6220 * Calculate the geometry of a particular (address, len) tuple. This
6221 * information is used to calculate how big a particular bio can get before it
6222 * straddles a stripe.
6224 * @fs_info: the filesystem
6225 * @em: mapping containing the logical extent
6226 * @op: type of operation - write or read
6227 * @logical: address that we want to figure out the geometry of
6228 * @io_geom: pointer used to return values
6230 * Returns < 0 in case a chunk for the given logical address cannot be found,
6231 * usually shouldn't happen unless @logical is corrupted, 0 otherwise.
6233 int btrfs_get_io_geometry(struct btrfs_fs_info *fs_info, struct extent_map *em,
6234 enum btrfs_map_op op, u64 logical,
6235 struct btrfs_io_geometry *io_geom)
6237 struct map_lookup *map;
6243 u64 raid56_full_stripe_start = (u64)-1;
6246 ASSERT(op != BTRFS_MAP_DISCARD);
6248 map = em->map_lookup;
6249 /* Offset of this logical address in the chunk */
6250 offset = logical - em->start;
6251 /* Len of a stripe in a chunk */
6252 stripe_len = map->stripe_len;
6253 /* Stripe where this block falls in */
6254 stripe_nr = div64_u64(offset, stripe_len);
6255 /* Offset of stripe in the chunk */
6256 stripe_offset = stripe_nr * stripe_len;
6257 if (offset < stripe_offset) {
6259 "stripe math has gone wrong, stripe_offset=%llu offset=%llu start=%llu logical=%llu stripe_len=%llu",
6260 stripe_offset, offset, em->start, logical, stripe_len);
6264 /* stripe_offset is the offset of this block in its stripe */
6265 stripe_offset = offset - stripe_offset;
6266 data_stripes = nr_data_stripes(map);
6268 if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
6269 u64 max_len = stripe_len - stripe_offset;
6272 * In case of raid56, we need to know the stripe aligned start
6274 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6275 unsigned long full_stripe_len = stripe_len * data_stripes;
6276 raid56_full_stripe_start = offset;
6279 * Allow a write of a full stripe, but make sure we
6280 * don't allow straddling of stripes
6282 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
6284 raid56_full_stripe_start *= full_stripe_len;
6287 * For writes to RAID[56], allow a full stripeset across
6288 * all disks. For other RAID types and for RAID[56]
6289 * reads, just allow a single stripe (on a single disk).
6291 if (op == BTRFS_MAP_WRITE) {
6292 max_len = stripe_len * data_stripes -
6293 (offset - raid56_full_stripe_start);
6296 len = min_t(u64, em->len - offset, max_len);
6298 len = em->len - offset;
6302 io_geom->offset = offset;
6303 io_geom->stripe_len = stripe_len;
6304 io_geom->stripe_nr = stripe_nr;
6305 io_geom->stripe_offset = stripe_offset;
6306 io_geom->raid56_stripe_offset = raid56_full_stripe_start;
6311 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
6312 enum btrfs_map_op op,
6313 u64 logical, u64 *length,
6314 struct btrfs_bio **bbio_ret,
6315 int mirror_num, int need_raid_map)
6317 struct extent_map *em;
6318 struct map_lookup *map;
6328 int tgtdev_indexes = 0;
6329 struct btrfs_bio *bbio = NULL;
6330 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
6331 int dev_replace_is_ongoing = 0;
6332 int num_alloc_stripes;
6333 int patch_the_first_stripe_for_dev_replace = 0;
6334 u64 physical_to_patch_in_first_stripe = 0;
6335 u64 raid56_full_stripe_start = (u64)-1;
6336 struct btrfs_io_geometry geom;
6339 ASSERT(op != BTRFS_MAP_DISCARD);
6341 em = btrfs_get_chunk_map(fs_info, logical, *length);
6342 ASSERT(!IS_ERR(em));
6344 ret = btrfs_get_io_geometry(fs_info, em, op, logical, &geom);
6348 map = em->map_lookup;
6351 stripe_len = geom.stripe_len;
6352 stripe_nr = geom.stripe_nr;
6353 stripe_offset = geom.stripe_offset;
6354 raid56_full_stripe_start = geom.raid56_stripe_offset;
6355 data_stripes = nr_data_stripes(map);
6357 down_read(&dev_replace->rwsem);
6358 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
6360 * Hold the semaphore for read during the whole operation, write is
6361 * requested at commit time but must wait.
6363 if (!dev_replace_is_ongoing)
6364 up_read(&dev_replace->rwsem);
6366 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
6367 !need_full_stripe(op) && dev_replace->tgtdev != NULL) {
6368 ret = get_extra_mirror_from_replace(fs_info, logical, *length,
6369 dev_replace->srcdev->devid,
6371 &physical_to_patch_in_first_stripe);
6375 patch_the_first_stripe_for_dev_replace = 1;
6376 } else if (mirror_num > map->num_stripes) {
6382 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
6383 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
6385 if (!need_full_stripe(op))
6387 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1_MASK) {
6388 if (need_full_stripe(op))
6389 num_stripes = map->num_stripes;
6390 else if (mirror_num)
6391 stripe_index = mirror_num - 1;
6393 stripe_index = find_live_mirror(fs_info, map, 0,
6394 dev_replace_is_ongoing);
6395 mirror_num = stripe_index + 1;
6398 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
6399 if (need_full_stripe(op)) {
6400 num_stripes = map->num_stripes;
6401 } else if (mirror_num) {
6402 stripe_index = mirror_num - 1;
6407 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
6408 u32 factor = map->num_stripes / map->sub_stripes;
6410 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
6411 stripe_index *= map->sub_stripes;
6413 if (need_full_stripe(op))
6414 num_stripes = map->sub_stripes;
6415 else if (mirror_num)
6416 stripe_index += mirror_num - 1;
6418 int old_stripe_index = stripe_index;
6419 stripe_index = find_live_mirror(fs_info, map,
6421 dev_replace_is_ongoing);
6422 mirror_num = stripe_index - old_stripe_index + 1;
6425 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6426 if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) {
6427 /* push stripe_nr back to the start of the full stripe */
6428 stripe_nr = div64_u64(raid56_full_stripe_start,
6429 stripe_len * data_stripes);
6431 /* RAID[56] write or recovery. Return all stripes */
6432 num_stripes = map->num_stripes;
6433 max_errors = nr_parity_stripes(map);
6435 *length = map->stripe_len;
6440 * Mirror #0 or #1 means the original data block.
6441 * Mirror #2 is RAID5 parity block.
6442 * Mirror #3 is RAID6 Q block.
6444 stripe_nr = div_u64_rem(stripe_nr,
6445 data_stripes, &stripe_index);
6447 stripe_index = data_stripes + mirror_num - 2;
6449 /* We distribute the parity blocks across stripes */
6450 div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
6452 if (!need_full_stripe(op) && mirror_num <= 1)
6457 * after this, stripe_nr is the number of stripes on this
6458 * device we have to walk to find the data, and stripe_index is
6459 * the number of our device in the stripe array
6461 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
6463 mirror_num = stripe_index + 1;
6465 if (stripe_index >= map->num_stripes) {
6467 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
6468 stripe_index, map->num_stripes);
6473 num_alloc_stripes = num_stripes;
6474 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) {
6475 if (op == BTRFS_MAP_WRITE)
6476 num_alloc_stripes <<= 1;
6477 if (op == BTRFS_MAP_GET_READ_MIRRORS)
6478 num_alloc_stripes++;
6479 tgtdev_indexes = num_stripes;
6482 bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
6488 for (i = 0; i < num_stripes; i++) {
6489 bbio->stripes[i].physical = map->stripes[stripe_index].physical +
6490 stripe_offset + stripe_nr * map->stripe_len;
6491 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
6495 /* build raid_map */
6496 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map &&
6497 (need_full_stripe(op) || mirror_num > 1)) {
6501 /* Work out the disk rotation on this stripe-set */
6502 div_u64_rem(stripe_nr, num_stripes, &rot);
6504 /* Fill in the logical address of each stripe */
6505 tmp = stripe_nr * data_stripes;
6506 for (i = 0; i < data_stripes; i++)
6507 bbio->raid_map[(i+rot) % num_stripes] =
6508 em->start + (tmp + i) * map->stripe_len;
6510 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
6511 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
6512 bbio->raid_map[(i+rot+1) % num_stripes] =
6515 sort_parity_stripes(bbio, num_stripes);
6518 if (need_full_stripe(op))
6519 max_errors = btrfs_chunk_max_errors(map);
6521 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
6522 need_full_stripe(op)) {
6523 handle_ops_on_dev_replace(op, &bbio, dev_replace, logical,
6524 &num_stripes, &max_errors);
6528 bbio->map_type = map->type;
6529 bbio->num_stripes = num_stripes;
6530 bbio->max_errors = max_errors;
6531 bbio->mirror_num = mirror_num;
6534 * this is the case that REQ_READ && dev_replace_is_ongoing &&
6535 * mirror_num == num_stripes + 1 && dev_replace target drive is
6536 * available as a mirror
6538 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
6539 WARN_ON(num_stripes > 1);
6540 bbio->stripes[0].dev = dev_replace->tgtdev;
6541 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
6542 bbio->mirror_num = map->num_stripes + 1;
6545 if (dev_replace_is_ongoing) {
6546 lockdep_assert_held(&dev_replace->rwsem);
6547 /* Unlock and let waiting writers proceed */
6548 up_read(&dev_replace->rwsem);
6550 free_extent_map(em);
6554 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6555 u64 logical, u64 *length,
6556 struct btrfs_bio **bbio_ret, int mirror_num)
6558 if (op == BTRFS_MAP_DISCARD)
6559 return __btrfs_map_block_for_discard(fs_info, logical,
6562 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret,
6566 /* For Scrub/replace */
6567 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6568 u64 logical, u64 *length,
6569 struct btrfs_bio **bbio_ret)
6571 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1);
6574 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
6576 bio->bi_private = bbio->private;
6577 bio->bi_end_io = bbio->end_io;
6580 btrfs_put_bbio(bbio);
6583 static void btrfs_end_bio(struct bio *bio)
6585 struct btrfs_bio *bbio = bio->bi_private;
6586 int is_orig_bio = 0;
6588 if (bio->bi_status) {
6589 atomic_inc(&bbio->error);
6590 if (bio->bi_status == BLK_STS_IOERR ||
6591 bio->bi_status == BLK_STS_TARGET) {
6592 struct btrfs_device *dev = btrfs_io_bio(bio)->device;
6595 if (btrfs_op(bio) == BTRFS_MAP_WRITE)
6596 btrfs_dev_stat_inc_and_print(dev,
6597 BTRFS_DEV_STAT_WRITE_ERRS);
6598 else if (!(bio->bi_opf & REQ_RAHEAD))
6599 btrfs_dev_stat_inc_and_print(dev,
6600 BTRFS_DEV_STAT_READ_ERRS);
6601 if (bio->bi_opf & REQ_PREFLUSH)
6602 btrfs_dev_stat_inc_and_print(dev,
6603 BTRFS_DEV_STAT_FLUSH_ERRS);
6607 if (bio == bbio->orig_bio)
6610 btrfs_bio_counter_dec(bbio->fs_info);
6612 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6615 bio = bbio->orig_bio;
6618 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6619 /* only send an error to the higher layers if it is
6620 * beyond the tolerance of the btrfs bio
6622 if (atomic_read(&bbio->error) > bbio->max_errors) {
6623 bio->bi_status = BLK_STS_IOERR;
6626 * this bio is actually up to date, we didn't
6627 * go over the max number of errors
6629 bio->bi_status = BLK_STS_OK;
6632 btrfs_end_bbio(bbio, bio);
6633 } else if (!is_orig_bio) {
6638 static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio,
6639 u64 physical, struct btrfs_device *dev)
6641 struct btrfs_fs_info *fs_info = bbio->fs_info;
6643 bio->bi_private = bbio;
6644 btrfs_io_bio(bio)->device = dev;
6645 bio->bi_end_io = btrfs_end_bio;
6646 bio->bi_iter.bi_sector = physical >> 9;
6648 * For zone append writing, bi_sector must point the beginning of the
6651 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
6652 if (btrfs_dev_is_sequential(dev, physical)) {
6653 u64 zone_start = round_down(physical, fs_info->zone_size);
6655 bio->bi_iter.bi_sector = zone_start >> SECTOR_SHIFT;
6657 bio->bi_opf &= ~REQ_OP_ZONE_APPEND;
6658 bio->bi_opf |= REQ_OP_WRITE;
6661 btrfs_debug_in_rcu(fs_info,
6662 "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
6663 bio_op(bio), bio->bi_opf, bio->bi_iter.bi_sector,
6664 (unsigned long)dev->bdev->bd_dev, rcu_str_deref(dev->name),
6665 dev->devid, bio->bi_iter.bi_size);
6666 bio_set_dev(bio, dev->bdev);
6668 btrfs_bio_counter_inc_noblocked(fs_info);
6670 btrfsic_submit_bio(bio);
6673 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6675 atomic_inc(&bbio->error);
6676 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6677 /* Should be the original bio. */
6678 WARN_ON(bio != bbio->orig_bio);
6680 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6681 bio->bi_iter.bi_sector = logical >> 9;
6682 if (atomic_read(&bbio->error) > bbio->max_errors)
6683 bio->bi_status = BLK_STS_IOERR;
6685 bio->bi_status = BLK_STS_OK;
6686 btrfs_end_bbio(bbio, bio);
6690 blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
6693 struct btrfs_device *dev;
6694 struct bio *first_bio = bio;
6695 u64 logical = bio->bi_iter.bi_sector << 9;
6701 struct btrfs_bio *bbio = NULL;
6703 length = bio->bi_iter.bi_size;
6704 map_length = length;
6706 btrfs_bio_counter_inc_blocked(fs_info);
6707 ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical,
6708 &map_length, &bbio, mirror_num, 1);
6710 btrfs_bio_counter_dec(fs_info);
6711 return errno_to_blk_status(ret);
6714 total_devs = bbio->num_stripes;
6715 bbio->orig_bio = first_bio;
6716 bbio->private = first_bio->bi_private;
6717 bbio->end_io = first_bio->bi_end_io;
6718 bbio->fs_info = fs_info;
6719 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6721 if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6722 ((btrfs_op(bio) == BTRFS_MAP_WRITE) || (mirror_num > 1))) {
6723 /* In this case, map_length has been set to the length of
6724 a single stripe; not the whole write */
6725 if (btrfs_op(bio) == BTRFS_MAP_WRITE) {
6726 ret = raid56_parity_write(fs_info, bio, bbio,
6729 ret = raid56_parity_recover(fs_info, bio, bbio,
6730 map_length, mirror_num, 1);
6733 btrfs_bio_counter_dec(fs_info);
6734 return errno_to_blk_status(ret);
6737 if (map_length < length) {
6739 "mapping failed logical %llu bio len %llu len %llu",
6740 logical, length, map_length);
6744 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6745 dev = bbio->stripes[dev_nr].dev;
6746 if (!dev || !dev->bdev || test_bit(BTRFS_DEV_STATE_MISSING,
6748 (btrfs_op(first_bio) == BTRFS_MAP_WRITE &&
6749 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) {
6750 bbio_error(bbio, first_bio, logical);
6754 if (dev_nr < total_devs - 1)
6755 bio = btrfs_bio_clone(first_bio);
6759 submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical, dev);
6761 btrfs_bio_counter_dec(fs_info);
6766 * Find a device specified by @devid or @uuid in the list of @fs_devices, or
6769 * If devid and uuid are both specified, the match must be exact, otherwise
6770 * only devid is used.
6772 struct btrfs_device *btrfs_find_device(struct btrfs_fs_devices *fs_devices,
6773 u64 devid, u8 *uuid, u8 *fsid)
6775 struct btrfs_device *device;
6776 struct btrfs_fs_devices *seed_devs;
6778 if (!fsid || !memcmp(fs_devices->metadata_uuid, fsid, BTRFS_FSID_SIZE)) {
6779 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6780 if (device->devid == devid &&
6781 (!uuid || memcmp(device->uuid, uuid,
6782 BTRFS_UUID_SIZE) == 0))
6787 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
6789 !memcmp(seed_devs->metadata_uuid, fsid, BTRFS_FSID_SIZE)) {
6790 list_for_each_entry(device, &seed_devs->devices,
6792 if (device->devid == devid &&
6793 (!uuid || memcmp(device->uuid, uuid,
6794 BTRFS_UUID_SIZE) == 0))
6803 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices,
6804 u64 devid, u8 *dev_uuid)
6806 struct btrfs_device *device;
6807 unsigned int nofs_flag;
6810 * We call this under the chunk_mutex, so we want to use NOFS for this
6811 * allocation, however we don't want to change btrfs_alloc_device() to
6812 * always do NOFS because we use it in a lot of other GFP_KERNEL safe
6815 nofs_flag = memalloc_nofs_save();
6816 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6817 memalloc_nofs_restore(nofs_flag);
6821 list_add(&device->dev_list, &fs_devices->devices);
6822 device->fs_devices = fs_devices;
6823 fs_devices->num_devices++;
6825 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6826 fs_devices->missing_devices++;
6832 * btrfs_alloc_device - allocate struct btrfs_device
6833 * @fs_info: used only for generating a new devid, can be NULL if
6834 * devid is provided (i.e. @devid != NULL).
6835 * @devid: a pointer to devid for this device. If NULL a new devid
6837 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6840 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6841 * on error. Returned struct is not linked onto any lists and must be
6842 * destroyed with btrfs_free_device.
6844 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6848 struct btrfs_device *dev;
6851 if (WARN_ON(!devid && !fs_info))
6852 return ERR_PTR(-EINVAL);
6854 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
6856 return ERR_PTR(-ENOMEM);
6859 * Preallocate a bio that's always going to be used for flushing device
6860 * barriers and matches the device lifespan
6862 dev->flush_bio = bio_kmalloc(GFP_KERNEL, 0);
6863 if (!dev->flush_bio) {
6865 return ERR_PTR(-ENOMEM);
6868 INIT_LIST_HEAD(&dev->dev_list);
6869 INIT_LIST_HEAD(&dev->dev_alloc_list);
6870 INIT_LIST_HEAD(&dev->post_commit_list);
6872 atomic_set(&dev->reada_in_flight, 0);
6873 atomic_set(&dev->dev_stats_ccnt, 0);
6874 btrfs_device_data_ordered_init(dev);
6875 INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
6876 INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
6877 extent_io_tree_init(fs_info, &dev->alloc_state,
6878 IO_TREE_DEVICE_ALLOC_STATE, NULL);
6885 ret = find_next_devid(fs_info, &tmp);
6887 btrfs_free_device(dev);
6888 return ERR_PTR(ret);
6894 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6896 generate_random_uuid(dev->uuid);
6901 static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info,
6902 u64 devid, u8 *uuid, bool error)
6905 btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing",
6908 btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing",
6912 static u64 calc_stripe_length(u64 type, u64 chunk_len, int num_stripes)
6914 const int data_stripes = calc_data_stripes(type, num_stripes);
6916 return div_u64(chunk_len, data_stripes);
6919 #if BITS_PER_LONG == 32
6921 * Due to page cache limit, metadata beyond BTRFS_32BIT_MAX_FILE_SIZE
6922 * can't be accessed on 32bit systems.
6924 * This function do mount time check to reject the fs if it already has
6925 * metadata chunk beyond that limit.
6927 static int check_32bit_meta_chunk(struct btrfs_fs_info *fs_info,
6928 u64 logical, u64 length, u64 type)
6930 if (!(type & BTRFS_BLOCK_GROUP_METADATA))
6933 if (logical + length < MAX_LFS_FILESIZE)
6936 btrfs_err_32bit_limit(fs_info);
6941 * This is to give early warning for any metadata chunk reaching
6942 * BTRFS_32BIT_EARLY_WARN_THRESHOLD.
6943 * Although we can still access the metadata, it's not going to be possible
6944 * once the limit is reached.
6946 static void warn_32bit_meta_chunk(struct btrfs_fs_info *fs_info,
6947 u64 logical, u64 length, u64 type)
6949 if (!(type & BTRFS_BLOCK_GROUP_METADATA))
6952 if (logical + length < BTRFS_32BIT_EARLY_WARN_THRESHOLD)
6955 btrfs_warn_32bit_limit(fs_info);
6959 static int read_one_chunk(struct btrfs_key *key, struct extent_buffer *leaf,
6960 struct btrfs_chunk *chunk)
6962 struct btrfs_fs_info *fs_info = leaf->fs_info;
6963 struct extent_map_tree *map_tree = &fs_info->mapping_tree;
6964 struct map_lookup *map;
6965 struct extent_map *em;
6970 u8 uuid[BTRFS_UUID_SIZE];
6975 logical = key->offset;
6976 length = btrfs_chunk_length(leaf, chunk);
6977 type = btrfs_chunk_type(leaf, chunk);
6978 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6980 #if BITS_PER_LONG == 32
6981 ret = check_32bit_meta_chunk(fs_info, logical, length, type);
6984 warn_32bit_meta_chunk(fs_info, logical, length, type);
6988 * Only need to verify chunk item if we're reading from sys chunk array,
6989 * as chunk item in tree block is already verified by tree-checker.
6991 if (leaf->start == BTRFS_SUPER_INFO_OFFSET) {
6992 ret = btrfs_check_chunk_valid(leaf, chunk, logical);
6997 read_lock(&map_tree->lock);
6998 em = lookup_extent_mapping(map_tree, logical, 1);
6999 read_unlock(&map_tree->lock);
7001 /* already mapped? */
7002 if (em && em->start <= logical && em->start + em->len > logical) {
7003 free_extent_map(em);
7006 free_extent_map(em);
7009 em = alloc_extent_map();
7012 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
7014 free_extent_map(em);
7018 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
7019 em->map_lookup = map;
7020 em->start = logical;
7023 em->block_start = 0;
7024 em->block_len = em->len;
7026 map->num_stripes = num_stripes;
7027 map->io_width = btrfs_chunk_io_width(leaf, chunk);
7028 map->io_align = btrfs_chunk_io_align(leaf, chunk);
7029 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
7031 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
7032 map->verified_stripes = 0;
7033 em->orig_block_len = calc_stripe_length(type, em->len,
7035 for (i = 0; i < num_stripes; i++) {
7036 map->stripes[i].physical =
7037 btrfs_stripe_offset_nr(leaf, chunk, i);
7038 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
7039 read_extent_buffer(leaf, uuid, (unsigned long)
7040 btrfs_stripe_dev_uuid_nr(chunk, i),
7042 map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices,
7044 if (!map->stripes[i].dev &&
7045 !btrfs_test_opt(fs_info, DEGRADED)) {
7046 free_extent_map(em);
7047 btrfs_report_missing_device(fs_info, devid, uuid, true);
7050 if (!map->stripes[i].dev) {
7051 map->stripes[i].dev =
7052 add_missing_dev(fs_info->fs_devices, devid,
7054 if (IS_ERR(map->stripes[i].dev)) {
7055 free_extent_map(em);
7057 "failed to init missing dev %llu: %ld",
7058 devid, PTR_ERR(map->stripes[i].dev));
7059 return PTR_ERR(map->stripes[i].dev);
7061 btrfs_report_missing_device(fs_info, devid, uuid, false);
7063 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
7064 &(map->stripes[i].dev->dev_state));
7068 write_lock(&map_tree->lock);
7069 ret = add_extent_mapping(map_tree, em, 0);
7070 write_unlock(&map_tree->lock);
7073 "failed to add chunk map, start=%llu len=%llu: %d",
7074 em->start, em->len, ret);
7076 free_extent_map(em);
7081 static void fill_device_from_item(struct extent_buffer *leaf,
7082 struct btrfs_dev_item *dev_item,
7083 struct btrfs_device *device)
7087 device->devid = btrfs_device_id(leaf, dev_item);
7088 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
7089 device->total_bytes = device->disk_total_bytes;
7090 device->commit_total_bytes = device->disk_total_bytes;
7091 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
7092 device->commit_bytes_used = device->bytes_used;
7093 device->type = btrfs_device_type(leaf, dev_item);
7094 device->io_align = btrfs_device_io_align(leaf, dev_item);
7095 device->io_width = btrfs_device_io_width(leaf, dev_item);
7096 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
7097 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
7098 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
7100 ptr = btrfs_device_uuid(dev_item);
7101 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
7104 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info,
7107 struct btrfs_fs_devices *fs_devices;
7110 lockdep_assert_held(&uuid_mutex);
7113 /* This will match only for multi-device seed fs */
7114 list_for_each_entry(fs_devices, &fs_info->fs_devices->seed_list, seed_list)
7115 if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE))
7119 fs_devices = find_fsid(fsid, NULL);
7121 if (!btrfs_test_opt(fs_info, DEGRADED))
7122 return ERR_PTR(-ENOENT);
7124 fs_devices = alloc_fs_devices(fsid, NULL);
7125 if (IS_ERR(fs_devices))
7128 fs_devices->seeding = true;
7129 fs_devices->opened = 1;
7134 * Upon first call for a seed fs fsid, just create a private copy of the
7135 * respective fs_devices and anchor it at fs_info->fs_devices->seed_list
7137 fs_devices = clone_fs_devices(fs_devices);
7138 if (IS_ERR(fs_devices))
7141 ret = open_fs_devices(fs_devices, FMODE_READ, fs_info->bdev_holder);
7143 free_fs_devices(fs_devices);
7144 return ERR_PTR(ret);
7147 if (!fs_devices->seeding) {
7148 close_fs_devices(fs_devices);
7149 free_fs_devices(fs_devices);
7150 return ERR_PTR(-EINVAL);
7153 list_add(&fs_devices->seed_list, &fs_info->fs_devices->seed_list);
7158 static int read_one_dev(struct extent_buffer *leaf,
7159 struct btrfs_dev_item *dev_item)
7161 struct btrfs_fs_info *fs_info = leaf->fs_info;
7162 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7163 struct btrfs_device *device;
7166 u8 fs_uuid[BTRFS_FSID_SIZE];
7167 u8 dev_uuid[BTRFS_UUID_SIZE];
7169 devid = btrfs_device_id(leaf, dev_item);
7170 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
7172 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
7175 if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) {
7176 fs_devices = open_seed_devices(fs_info, fs_uuid);
7177 if (IS_ERR(fs_devices))
7178 return PTR_ERR(fs_devices);
7181 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
7184 if (!btrfs_test_opt(fs_info, DEGRADED)) {
7185 btrfs_report_missing_device(fs_info, devid,
7190 device = add_missing_dev(fs_devices, devid, dev_uuid);
7191 if (IS_ERR(device)) {
7193 "failed to add missing dev %llu: %ld",
7194 devid, PTR_ERR(device));
7195 return PTR_ERR(device);
7197 btrfs_report_missing_device(fs_info, devid, dev_uuid, false);
7199 if (!device->bdev) {
7200 if (!btrfs_test_opt(fs_info, DEGRADED)) {
7201 btrfs_report_missing_device(fs_info,
7202 devid, dev_uuid, true);
7205 btrfs_report_missing_device(fs_info, devid,
7209 if (!device->bdev &&
7210 !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
7212 * this happens when a device that was properly setup
7213 * in the device info lists suddenly goes bad.
7214 * device->bdev is NULL, and so we have to set
7215 * device->missing to one here
7217 device->fs_devices->missing_devices++;
7218 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
7221 /* Move the device to its own fs_devices */
7222 if (device->fs_devices != fs_devices) {
7223 ASSERT(test_bit(BTRFS_DEV_STATE_MISSING,
7224 &device->dev_state));
7226 list_move(&device->dev_list, &fs_devices->devices);
7227 device->fs_devices->num_devices--;
7228 fs_devices->num_devices++;
7230 device->fs_devices->missing_devices--;
7231 fs_devices->missing_devices++;
7233 device->fs_devices = fs_devices;
7237 if (device->fs_devices != fs_info->fs_devices) {
7238 BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state));
7239 if (device->generation !=
7240 btrfs_device_generation(leaf, dev_item))
7244 fill_device_from_item(leaf, dev_item, device);
7246 u64 max_total_bytes = i_size_read(device->bdev->bd_inode);
7248 if (device->total_bytes > max_total_bytes) {
7250 "device total_bytes should be at most %llu but found %llu",
7251 max_total_bytes, device->total_bytes);
7255 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
7256 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
7257 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
7258 device->fs_devices->total_rw_bytes += device->total_bytes;
7259 atomic64_add(device->total_bytes - device->bytes_used,
7260 &fs_info->free_chunk_space);
7266 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
7268 struct btrfs_root *root = fs_info->tree_root;
7269 struct btrfs_super_block *super_copy = fs_info->super_copy;
7270 struct extent_buffer *sb;
7271 struct btrfs_disk_key *disk_key;
7272 struct btrfs_chunk *chunk;
7274 unsigned long sb_array_offset;
7281 struct btrfs_key key;
7283 ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize);
7285 * This will create extent buffer of nodesize, superblock size is
7286 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
7287 * overallocate but we can keep it as-is, only the first page is used.
7289 sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET,
7290 root->root_key.objectid, 0);
7293 set_extent_buffer_uptodate(sb);
7295 * The sb extent buffer is artificial and just used to read the system array.
7296 * set_extent_buffer_uptodate() call does not properly mark all it's
7297 * pages up-to-date when the page is larger: extent does not cover the
7298 * whole page and consequently check_page_uptodate does not find all
7299 * the page's extents up-to-date (the hole beyond sb),
7300 * write_extent_buffer then triggers a WARN_ON.
7302 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
7303 * but sb spans only this function. Add an explicit SetPageUptodate call
7304 * to silence the warning eg. on PowerPC 64.
7306 if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
7307 SetPageUptodate(sb->pages[0]);
7309 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
7310 array_size = btrfs_super_sys_array_size(super_copy);
7312 array_ptr = super_copy->sys_chunk_array;
7313 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
7316 while (cur_offset < array_size) {
7317 disk_key = (struct btrfs_disk_key *)array_ptr;
7318 len = sizeof(*disk_key);
7319 if (cur_offset + len > array_size)
7320 goto out_short_read;
7322 btrfs_disk_key_to_cpu(&key, disk_key);
7325 sb_array_offset += len;
7328 if (key.type != BTRFS_CHUNK_ITEM_KEY) {
7330 "unexpected item type %u in sys_array at offset %u",
7331 (u32)key.type, cur_offset);
7336 chunk = (struct btrfs_chunk *)sb_array_offset;
7338 * At least one btrfs_chunk with one stripe must be present,
7339 * exact stripe count check comes afterwards
7341 len = btrfs_chunk_item_size(1);
7342 if (cur_offset + len > array_size)
7343 goto out_short_read;
7345 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
7348 "invalid number of stripes %u in sys_array at offset %u",
7349 num_stripes, cur_offset);
7354 type = btrfs_chunk_type(sb, chunk);
7355 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
7357 "invalid chunk type %llu in sys_array at offset %u",
7363 len = btrfs_chunk_item_size(num_stripes);
7364 if (cur_offset + len > array_size)
7365 goto out_short_read;
7367 ret = read_one_chunk(&key, sb, chunk);
7372 sb_array_offset += len;
7375 clear_extent_buffer_uptodate(sb);
7376 free_extent_buffer_stale(sb);
7380 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
7382 clear_extent_buffer_uptodate(sb);
7383 free_extent_buffer_stale(sb);
7388 * Check if all chunks in the fs are OK for read-write degraded mount
7390 * If the @failing_dev is specified, it's accounted as missing.
7392 * Return true if all chunks meet the minimal RW mount requirements.
7393 * Return false if any chunk doesn't meet the minimal RW mount requirements.
7395 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
7396 struct btrfs_device *failing_dev)
7398 struct extent_map_tree *map_tree = &fs_info->mapping_tree;
7399 struct extent_map *em;
7403 read_lock(&map_tree->lock);
7404 em = lookup_extent_mapping(map_tree, 0, (u64)-1);
7405 read_unlock(&map_tree->lock);
7406 /* No chunk at all? Return false anyway */
7412 struct map_lookup *map;
7417 map = em->map_lookup;
7419 btrfs_get_num_tolerated_disk_barrier_failures(
7421 for (i = 0; i < map->num_stripes; i++) {
7422 struct btrfs_device *dev = map->stripes[i].dev;
7424 if (!dev || !dev->bdev ||
7425 test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) ||
7426 dev->last_flush_error)
7428 else if (failing_dev && failing_dev == dev)
7431 if (missing > max_tolerated) {
7434 "chunk %llu missing %d devices, max tolerance is %d for writable mount",
7435 em->start, missing, max_tolerated);
7436 free_extent_map(em);
7440 next_start = extent_map_end(em);
7441 free_extent_map(em);
7443 read_lock(&map_tree->lock);
7444 em = lookup_extent_mapping(map_tree, next_start,
7445 (u64)(-1) - next_start);
7446 read_unlock(&map_tree->lock);
7452 static void readahead_tree_node_children(struct extent_buffer *node)
7455 const int nr_items = btrfs_header_nritems(node);
7457 for (i = 0; i < nr_items; i++)
7458 btrfs_readahead_node_child(node, i);
7461 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
7463 struct btrfs_root *root = fs_info->chunk_root;
7464 struct btrfs_path *path;
7465 struct extent_buffer *leaf;
7466 struct btrfs_key key;
7467 struct btrfs_key found_key;
7471 u64 last_ra_node = 0;
7473 path = btrfs_alloc_path();
7478 * uuid_mutex is needed only if we are mounting a sprout FS
7479 * otherwise we don't need it.
7481 mutex_lock(&uuid_mutex);
7484 * It is possible for mount and umount to race in such a way that
7485 * we execute this code path, but open_fs_devices failed to clear
7486 * total_rw_bytes. We certainly want it cleared before reading the
7487 * device items, so clear it here.
7489 fs_info->fs_devices->total_rw_bytes = 0;
7492 * Lockdep complains about possible circular locking dependency between
7493 * a disk's open_mutex (struct gendisk.open_mutex), the rw semaphores
7494 * used for freeze procection of a fs (struct super_block.s_writers),
7495 * which we take when starting a transaction, and extent buffers of the
7496 * chunk tree if we call read_one_dev() while holding a lock on an
7497 * extent buffer of the chunk tree. Since we are mounting the filesystem
7498 * and at this point there can't be any concurrent task modifying the
7499 * chunk tree, to keep it simple, just skip locking on the chunk tree.
7501 ASSERT(!test_bit(BTRFS_FS_OPEN, &fs_info->flags));
7502 path->skip_locking = 1;
7505 * Read all device items, and then all the chunk items. All
7506 * device items are found before any chunk item (their object id
7507 * is smaller than the lowest possible object id for a chunk
7508 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
7510 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
7513 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7517 struct extent_buffer *node;
7519 leaf = path->nodes[0];
7520 slot = path->slots[0];
7521 if (slot >= btrfs_header_nritems(leaf)) {
7522 ret = btrfs_next_leaf(root, path);
7529 node = path->nodes[1];
7531 if (last_ra_node != node->start) {
7532 readahead_tree_node_children(node);
7533 last_ra_node = node->start;
7536 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7537 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
7538 struct btrfs_dev_item *dev_item;
7539 dev_item = btrfs_item_ptr(leaf, slot,
7540 struct btrfs_dev_item);
7541 ret = read_one_dev(leaf, dev_item);
7545 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
7546 struct btrfs_chunk *chunk;
7549 * We are only called at mount time, so no need to take
7550 * fs_info->chunk_mutex. Plus, to avoid lockdep warnings,
7551 * we always lock first fs_info->chunk_mutex before
7552 * acquiring any locks on the chunk tree. This is a
7553 * requirement for chunk allocation, see the comment on
7554 * top of btrfs_chunk_alloc() for details.
7556 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
7557 ret = read_one_chunk(&found_key, leaf, chunk);
7565 * After loading chunk tree, we've got all device information,
7566 * do another round of validation checks.
7568 if (total_dev != fs_info->fs_devices->total_devices) {
7570 "super_num_devices %llu mismatch with num_devices %llu found here",
7571 btrfs_super_num_devices(fs_info->super_copy),
7576 if (btrfs_super_total_bytes(fs_info->super_copy) <
7577 fs_info->fs_devices->total_rw_bytes) {
7579 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
7580 btrfs_super_total_bytes(fs_info->super_copy),
7581 fs_info->fs_devices->total_rw_bytes);
7587 mutex_unlock(&uuid_mutex);
7589 btrfs_free_path(path);
7593 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
7595 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
7596 struct btrfs_device *device;
7598 fs_devices->fs_info = fs_info;
7600 mutex_lock(&fs_devices->device_list_mutex);
7601 list_for_each_entry(device, &fs_devices->devices, dev_list)
7602 device->fs_info = fs_info;
7604 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
7605 list_for_each_entry(device, &seed_devs->devices, dev_list)
7606 device->fs_info = fs_info;
7608 seed_devs->fs_info = fs_info;
7610 mutex_unlock(&fs_devices->device_list_mutex);
7613 static u64 btrfs_dev_stats_value(const struct extent_buffer *eb,
7614 const struct btrfs_dev_stats_item *ptr,
7619 read_extent_buffer(eb, &val,
7620 offsetof(struct btrfs_dev_stats_item, values) +
7621 ((unsigned long)ptr) + (index * sizeof(u64)),
7626 static void btrfs_set_dev_stats_value(struct extent_buffer *eb,
7627 struct btrfs_dev_stats_item *ptr,
7630 write_extent_buffer(eb, &val,
7631 offsetof(struct btrfs_dev_stats_item, values) +
7632 ((unsigned long)ptr) + (index * sizeof(u64)),
7636 static int btrfs_device_init_dev_stats(struct btrfs_device *device,
7637 struct btrfs_path *path)
7639 struct btrfs_dev_stats_item *ptr;
7640 struct extent_buffer *eb;
7641 struct btrfs_key key;
7645 if (!device->fs_info->dev_root)
7648 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7649 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7650 key.offset = device->devid;
7651 ret = btrfs_search_slot(NULL, device->fs_info->dev_root, &key, path, 0, 0);
7653 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7654 btrfs_dev_stat_set(device, i, 0);
7655 device->dev_stats_valid = 1;
7656 btrfs_release_path(path);
7657 return ret < 0 ? ret : 0;
7659 slot = path->slots[0];
7660 eb = path->nodes[0];
7661 item_size = btrfs_item_size_nr(eb, slot);
7663 ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_stats_item);
7665 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7666 if (item_size >= (1 + i) * sizeof(__le64))
7667 btrfs_dev_stat_set(device, i,
7668 btrfs_dev_stats_value(eb, ptr, i));
7670 btrfs_dev_stat_set(device, i, 0);
7673 device->dev_stats_valid = 1;
7674 btrfs_dev_stat_print_on_load(device);
7675 btrfs_release_path(path);
7680 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
7682 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
7683 struct btrfs_device *device;
7684 struct btrfs_path *path = NULL;
7687 path = btrfs_alloc_path();
7691 mutex_lock(&fs_devices->device_list_mutex);
7692 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7693 ret = btrfs_device_init_dev_stats(device, path);
7697 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
7698 list_for_each_entry(device, &seed_devs->devices, dev_list) {
7699 ret = btrfs_device_init_dev_stats(device, path);
7705 mutex_unlock(&fs_devices->device_list_mutex);
7707 btrfs_free_path(path);
7711 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
7712 struct btrfs_device *device)
7714 struct btrfs_fs_info *fs_info = trans->fs_info;
7715 struct btrfs_root *dev_root = fs_info->dev_root;
7716 struct btrfs_path *path;
7717 struct btrfs_key key;
7718 struct extent_buffer *eb;
7719 struct btrfs_dev_stats_item *ptr;
7723 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7724 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7725 key.offset = device->devid;
7727 path = btrfs_alloc_path();
7730 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
7732 btrfs_warn_in_rcu(fs_info,
7733 "error %d while searching for dev_stats item for device %s",
7734 ret, rcu_str_deref(device->name));
7739 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
7740 /* need to delete old one and insert a new one */
7741 ret = btrfs_del_item(trans, dev_root, path);
7743 btrfs_warn_in_rcu(fs_info,
7744 "delete too small dev_stats item for device %s failed %d",
7745 rcu_str_deref(device->name), ret);
7752 /* need to insert a new item */
7753 btrfs_release_path(path);
7754 ret = btrfs_insert_empty_item(trans, dev_root, path,
7755 &key, sizeof(*ptr));
7757 btrfs_warn_in_rcu(fs_info,
7758 "insert dev_stats item for device %s failed %d",
7759 rcu_str_deref(device->name), ret);
7764 eb = path->nodes[0];
7765 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7766 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7767 btrfs_set_dev_stats_value(eb, ptr, i,
7768 btrfs_dev_stat_read(device, i));
7769 btrfs_mark_buffer_dirty(eb);
7772 btrfs_free_path(path);
7777 * called from commit_transaction. Writes all changed device stats to disk.
7779 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans)
7781 struct btrfs_fs_info *fs_info = trans->fs_info;
7782 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7783 struct btrfs_device *device;
7787 mutex_lock(&fs_devices->device_list_mutex);
7788 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7789 stats_cnt = atomic_read(&device->dev_stats_ccnt);
7790 if (!device->dev_stats_valid || stats_cnt == 0)
7795 * There is a LOAD-LOAD control dependency between the value of
7796 * dev_stats_ccnt and updating the on-disk values which requires
7797 * reading the in-memory counters. Such control dependencies
7798 * require explicit read memory barriers.
7800 * This memory barriers pairs with smp_mb__before_atomic in
7801 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full
7802 * barrier implied by atomic_xchg in
7803 * btrfs_dev_stats_read_and_reset
7807 ret = update_dev_stat_item(trans, device);
7809 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7811 mutex_unlock(&fs_devices->device_list_mutex);
7816 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7818 btrfs_dev_stat_inc(dev, index);
7819 btrfs_dev_stat_print_on_error(dev);
7822 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
7824 if (!dev->dev_stats_valid)
7826 btrfs_err_rl_in_rcu(dev->fs_info,
7827 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7828 rcu_str_deref(dev->name),
7829 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7830 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7831 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7832 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7833 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7836 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7840 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7841 if (btrfs_dev_stat_read(dev, i) != 0)
7843 if (i == BTRFS_DEV_STAT_VALUES_MAX)
7844 return; /* all values == 0, suppress message */
7846 btrfs_info_in_rcu(dev->fs_info,
7847 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7848 rcu_str_deref(dev->name),
7849 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7850 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7851 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7852 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7853 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7856 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
7857 struct btrfs_ioctl_get_dev_stats *stats)
7859 struct btrfs_device *dev;
7860 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7863 mutex_lock(&fs_devices->device_list_mutex);
7864 dev = btrfs_find_device(fs_info->fs_devices, stats->devid, NULL, NULL);
7865 mutex_unlock(&fs_devices->device_list_mutex);
7868 btrfs_warn(fs_info, "get dev_stats failed, device not found");
7870 } else if (!dev->dev_stats_valid) {
7871 btrfs_warn(fs_info, "get dev_stats failed, not yet valid");
7873 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7874 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7875 if (stats->nr_items > i)
7877 btrfs_dev_stat_read_and_reset(dev, i);
7879 btrfs_dev_stat_set(dev, i, 0);
7881 btrfs_info(fs_info, "device stats zeroed by %s (%d)",
7882 current->comm, task_pid_nr(current));
7884 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7885 if (stats->nr_items > i)
7886 stats->values[i] = btrfs_dev_stat_read(dev, i);
7888 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7889 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7894 * Update the size and bytes used for each device where it changed. This is
7895 * delayed since we would otherwise get errors while writing out the
7898 * Must be invoked during transaction commit.
7900 void btrfs_commit_device_sizes(struct btrfs_transaction *trans)
7902 struct btrfs_device *curr, *next;
7904 ASSERT(trans->state == TRANS_STATE_COMMIT_DOING);
7906 if (list_empty(&trans->dev_update_list))
7910 * We don't need the device_list_mutex here. This list is owned by the
7911 * transaction and the transaction must complete before the device is
7914 mutex_lock(&trans->fs_info->chunk_mutex);
7915 list_for_each_entry_safe(curr, next, &trans->dev_update_list,
7917 list_del_init(&curr->post_commit_list);
7918 curr->commit_total_bytes = curr->disk_total_bytes;
7919 curr->commit_bytes_used = curr->bytes_used;
7921 mutex_unlock(&trans->fs_info->chunk_mutex);
7925 * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10.
7927 int btrfs_bg_type_to_factor(u64 flags)
7929 const int index = btrfs_bg_flags_to_raid_index(flags);
7931 return btrfs_raid_array[index].ncopies;
7936 static int verify_one_dev_extent(struct btrfs_fs_info *fs_info,
7937 u64 chunk_offset, u64 devid,
7938 u64 physical_offset, u64 physical_len)
7940 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
7941 struct extent_map *em;
7942 struct map_lookup *map;
7943 struct btrfs_device *dev;
7949 read_lock(&em_tree->lock);
7950 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
7951 read_unlock(&em_tree->lock);
7955 "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk",
7956 physical_offset, devid);
7961 map = em->map_lookup;
7962 stripe_len = calc_stripe_length(map->type, em->len, map->num_stripes);
7963 if (physical_len != stripe_len) {
7965 "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu",
7966 physical_offset, devid, em->start, physical_len,
7972 for (i = 0; i < map->num_stripes; i++) {
7973 if (map->stripes[i].dev->devid == devid &&
7974 map->stripes[i].physical == physical_offset) {
7976 if (map->verified_stripes >= map->num_stripes) {
7978 "too many dev extents for chunk %llu found",
7983 map->verified_stripes++;
7989 "dev extent physical offset %llu devid %llu has no corresponding chunk",
7990 physical_offset, devid);
7994 /* Make sure no dev extent is beyond device boundary */
7995 dev = btrfs_find_device(fs_info->fs_devices, devid, NULL, NULL);
7997 btrfs_err(fs_info, "failed to find devid %llu", devid);
8002 if (physical_offset + physical_len > dev->disk_total_bytes) {
8004 "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu",
8005 devid, physical_offset, physical_len,
8006 dev->disk_total_bytes);
8011 if (dev->zone_info) {
8012 u64 zone_size = dev->zone_info->zone_size;
8014 if (!IS_ALIGNED(physical_offset, zone_size) ||
8015 !IS_ALIGNED(physical_len, zone_size)) {
8017 "zoned: dev extent devid %llu physical offset %llu len %llu is not aligned to device zone",
8018 devid, physical_offset, physical_len);
8025 free_extent_map(em);
8029 static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info)
8031 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
8032 struct extent_map *em;
8033 struct rb_node *node;
8036 read_lock(&em_tree->lock);
8037 for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) {
8038 em = rb_entry(node, struct extent_map, rb_node);
8039 if (em->map_lookup->num_stripes !=
8040 em->map_lookup->verified_stripes) {
8042 "chunk %llu has missing dev extent, have %d expect %d",
8043 em->start, em->map_lookup->verified_stripes,
8044 em->map_lookup->num_stripes);
8050 read_unlock(&em_tree->lock);
8055 * Ensure that all dev extents are mapped to correct chunk, otherwise
8056 * later chunk allocation/free would cause unexpected behavior.
8058 * NOTE: This will iterate through the whole device tree, which should be of
8059 * the same size level as the chunk tree. This slightly increases mount time.
8061 int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info)
8063 struct btrfs_path *path;
8064 struct btrfs_root *root = fs_info->dev_root;
8065 struct btrfs_key key;
8067 u64 prev_dev_ext_end = 0;
8071 * We don't have a dev_root because we mounted with ignorebadroots and
8072 * failed to load the root, so we want to skip the verification in this
8075 * However if the dev root is fine, but the tree itself is corrupted
8076 * we'd still fail to mount. This verification is only to make sure
8077 * writes can happen safely, so instead just bypass this check
8078 * completely in the case of IGNOREBADROOTS.
8080 if (btrfs_test_opt(fs_info, IGNOREBADROOTS))
8084 key.type = BTRFS_DEV_EXTENT_KEY;
8087 path = btrfs_alloc_path();
8091 path->reada = READA_FORWARD;
8092 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
8096 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
8097 ret = btrfs_next_leaf(root, path);
8100 /* No dev extents at all? Not good */
8107 struct extent_buffer *leaf = path->nodes[0];
8108 struct btrfs_dev_extent *dext;
8109 int slot = path->slots[0];
8111 u64 physical_offset;
8115 btrfs_item_key_to_cpu(leaf, &key, slot);
8116 if (key.type != BTRFS_DEV_EXTENT_KEY)
8118 devid = key.objectid;
8119 physical_offset = key.offset;
8121 dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent);
8122 chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext);
8123 physical_len = btrfs_dev_extent_length(leaf, dext);
8125 /* Check if this dev extent overlaps with the previous one */
8126 if (devid == prev_devid && physical_offset < prev_dev_ext_end) {
8128 "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu",
8129 devid, physical_offset, prev_dev_ext_end);
8134 ret = verify_one_dev_extent(fs_info, chunk_offset, devid,
8135 physical_offset, physical_len);
8139 prev_dev_ext_end = physical_offset + physical_len;
8141 ret = btrfs_next_item(root, path);
8150 /* Ensure all chunks have corresponding dev extents */
8151 ret = verify_chunk_dev_extent_mapping(fs_info);
8153 btrfs_free_path(path);
8158 * Check whether the given block group or device is pinned by any inode being
8159 * used as a swapfile.
8161 bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr)
8163 struct btrfs_swapfile_pin *sp;
8164 struct rb_node *node;
8166 spin_lock(&fs_info->swapfile_pins_lock);
8167 node = fs_info->swapfile_pins.rb_node;
8169 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
8171 node = node->rb_left;
8172 else if (ptr > sp->ptr)
8173 node = node->rb_right;
8177 spin_unlock(&fs_info->swapfile_pins_lock);
8178 return node != NULL;
8181 static int relocating_repair_kthread(void *data)
8183 struct btrfs_block_group *cache = (struct btrfs_block_group *)data;
8184 struct btrfs_fs_info *fs_info = cache->fs_info;
8188 target = cache->start;
8189 btrfs_put_block_group(cache);
8191 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) {
8193 "zoned: skip relocating block group %llu to repair: EBUSY",
8198 mutex_lock(&fs_info->reclaim_bgs_lock);
8200 /* Ensure block group still exists */
8201 cache = btrfs_lookup_block_group(fs_info, target);
8205 if (!cache->relocating_repair)
8208 ret = btrfs_may_alloc_data_chunk(fs_info, target);
8213 "zoned: relocating block group %llu to repair IO failure",
8215 ret = btrfs_relocate_chunk(fs_info, target);
8219 btrfs_put_block_group(cache);
8220 mutex_unlock(&fs_info->reclaim_bgs_lock);
8221 btrfs_exclop_finish(fs_info);
8226 int btrfs_repair_one_zone(struct btrfs_fs_info *fs_info, u64 logical)
8228 struct btrfs_block_group *cache;
8230 /* Do not attempt to repair in degraded state */
8231 if (btrfs_test_opt(fs_info, DEGRADED))
8234 cache = btrfs_lookup_block_group(fs_info, logical);
8238 spin_lock(&cache->lock);
8239 if (cache->relocating_repair) {
8240 spin_unlock(&cache->lock);
8241 btrfs_put_block_group(cache);
8244 cache->relocating_repair = 1;
8245 spin_unlock(&cache->lock);
8247 kthread_run(relocating_repair_kthread, cache,
8248 "btrfs-relocating-repair");