1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
6 #include <linux/sched.h>
8 #include <linux/slab.h>
9 #include <linux/buffer_head.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>
18 #include "extent_map.h"
20 #include "transaction.h"
21 #include "print-tree.h"
24 #include "async-thread.h"
25 #include "check-integrity.h"
26 #include "rcu-string.h"
28 #include "dev-replace.h"
31 const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
32 [BTRFS_RAID_RAID10] = {
35 .devs_max = 0, /* 0 == as many as possible */
37 .tolerated_failures = 1,
40 .raid_name = "raid10",
41 .bg_flag = BTRFS_BLOCK_GROUP_RAID10,
42 .mindev_error = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET,
44 [BTRFS_RAID_RAID1] = {
49 .tolerated_failures = 1,
53 .bg_flag = BTRFS_BLOCK_GROUP_RAID1,
54 .mindev_error = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET,
61 .tolerated_failures = 0,
65 .bg_flag = BTRFS_BLOCK_GROUP_DUP,
68 [BTRFS_RAID_RAID0] = {
73 .tolerated_failures = 0,
77 .bg_flag = BTRFS_BLOCK_GROUP_RAID0,
80 [BTRFS_RAID_SINGLE] = {
85 .tolerated_failures = 0,
88 .raid_name = "single",
92 [BTRFS_RAID_RAID5] = {
97 .tolerated_failures = 1,
100 .raid_name = "raid5",
101 .bg_flag = BTRFS_BLOCK_GROUP_RAID5,
102 .mindev_error = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET,
104 [BTRFS_RAID_RAID6] = {
109 .tolerated_failures = 2,
112 .raid_name = "raid6",
113 .bg_flag = BTRFS_BLOCK_GROUP_RAID6,
114 .mindev_error = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET,
118 const char *get_raid_name(enum btrfs_raid_types type)
120 if (type >= BTRFS_NR_RAID_TYPES)
123 return btrfs_raid_array[type].raid_name;
126 static int init_first_rw_device(struct btrfs_trans_handle *trans,
127 struct btrfs_fs_info *fs_info);
128 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info);
129 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
130 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
131 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
132 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
133 enum btrfs_map_op op,
134 u64 logical, u64 *length,
135 struct btrfs_bio **bbio_ret,
136 int mirror_num, int need_raid_map);
142 * There are several mutexes that protect manipulation of devices and low-level
143 * structures like chunks but not block groups, extents or files
145 * uuid_mutex (global lock)
146 * ------------------------
147 * protects the fs_uuids list that tracks all per-fs fs_devices, resulting from
148 * the SCAN_DEV ioctl registration or from mount either implicitly (the first
149 * device) or requested by the device= mount option
151 * the mutex can be very coarse and can cover long-running operations
153 * protects: updates to fs_devices counters like missing devices, rw devices,
154 * seeding, structure cloning, openning/closing devices at mount/umount time
156 * global::fs_devs - add, remove, updates to the global list
158 * does not protect: manipulation of the fs_devices::devices list!
160 * btrfs_device::name - renames (write side), read is RCU
162 * fs_devices::device_list_mutex (per-fs, with RCU)
163 * ------------------------------------------------
164 * protects updates to fs_devices::devices, ie. adding and deleting
166 * simple list traversal with read-only actions can be done with RCU protection
168 * may be used to exclude some operations from running concurrently without any
169 * modifications to the list (see write_all_supers)
173 * protects balance structures (status, state) and context accessed from
174 * several places (internally, ioctl)
178 * protects chunks, adding or removing during allocation, trim or when a new
179 * device is added/removed
183 * a big lock that is held by the cleaner thread and prevents running subvolume
184 * cleaning together with relocation or delayed iputs
197 * Exclusive operations, BTRFS_FS_EXCL_OP
198 * ======================================
200 * Maintains the exclusivity of the following operations that apply to the
201 * whole filesystem and cannot run in parallel.
206 * - Device replace (*)
209 * The device operations (as above) can be in one of the following states:
215 * Only device operations marked with (*) can go into the Paused state for the
218 * - ioctl (only Balance can be Paused through ioctl)
219 * - filesystem remounted as read-only
220 * - filesystem unmounted and mounted as read-only
221 * - system power-cycle and filesystem mounted as read-only
222 * - filesystem or device errors leading to forced read-only
224 * BTRFS_FS_EXCL_OP flag is set and cleared using atomic operations.
225 * During the course of Paused state, the BTRFS_FS_EXCL_OP remains set.
226 * A device operation in Paused or Running state can be canceled or resumed
227 * either by ioctl (Balance only) or when remounted as read-write.
228 * BTRFS_FS_EXCL_OP flag is cleared when the device operation is canceled or
232 DEFINE_MUTEX(uuid_mutex);
233 static LIST_HEAD(fs_uuids);
234 struct list_head *btrfs_get_fs_uuids(void)
240 * alloc_fs_devices - allocate struct btrfs_fs_devices
241 * @fsid: if not NULL, copy the uuid to fs_devices::fsid
243 * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR().
244 * The returned struct is not linked onto any lists and can be destroyed with
245 * kfree() right away.
247 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
249 struct btrfs_fs_devices *fs_devs;
251 fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL);
253 return ERR_PTR(-ENOMEM);
255 mutex_init(&fs_devs->device_list_mutex);
257 INIT_LIST_HEAD(&fs_devs->devices);
258 INIT_LIST_HEAD(&fs_devs->resized_devices);
259 INIT_LIST_HEAD(&fs_devs->alloc_list);
260 INIT_LIST_HEAD(&fs_devs->fs_list);
262 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
267 void btrfs_free_device(struct btrfs_device *device)
269 rcu_string_free(device->name);
270 bio_put(device->flush_bio);
274 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
276 struct btrfs_device *device;
277 WARN_ON(fs_devices->opened);
278 while (!list_empty(&fs_devices->devices)) {
279 device = list_entry(fs_devices->devices.next,
280 struct btrfs_device, dev_list);
281 list_del(&device->dev_list);
282 btrfs_free_device(device);
287 static void btrfs_kobject_uevent(struct block_device *bdev,
288 enum kobject_action action)
292 ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
294 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
296 kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
297 &disk_to_dev(bdev->bd_disk)->kobj);
300 void __exit btrfs_cleanup_fs_uuids(void)
302 struct btrfs_fs_devices *fs_devices;
304 while (!list_empty(&fs_uuids)) {
305 fs_devices = list_entry(fs_uuids.next,
306 struct btrfs_fs_devices, fs_list);
307 list_del(&fs_devices->fs_list);
308 free_fs_devices(fs_devices);
313 * Returns a pointer to a new btrfs_device on success; ERR_PTR() on error.
314 * Returned struct is not linked onto any lists and must be destroyed using
317 static struct btrfs_device *__alloc_device(void)
319 struct btrfs_device *dev;
321 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
323 return ERR_PTR(-ENOMEM);
326 * Preallocate a bio that's always going to be used for flushing device
327 * barriers and matches the device lifespan
329 dev->flush_bio = bio_alloc_bioset(GFP_KERNEL, 0, NULL);
330 if (!dev->flush_bio) {
332 return ERR_PTR(-ENOMEM);
335 INIT_LIST_HEAD(&dev->dev_list);
336 INIT_LIST_HEAD(&dev->dev_alloc_list);
337 INIT_LIST_HEAD(&dev->resized_list);
339 spin_lock_init(&dev->io_lock);
341 atomic_set(&dev->reada_in_flight, 0);
342 atomic_set(&dev->dev_stats_ccnt, 0);
343 btrfs_device_data_ordered_init(dev);
344 INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
345 INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
351 * Find a device specified by @devid or @uuid in the list of @fs_devices, or
354 * If devid and uuid are both specified, the match must be exact, otherwise
355 * only devid is used.
357 static struct btrfs_device *find_device(struct btrfs_fs_devices *fs_devices,
358 u64 devid, const u8 *uuid)
360 struct btrfs_device *dev;
362 list_for_each_entry(dev, &fs_devices->devices, dev_list) {
363 if (dev->devid == devid &&
364 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
371 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
373 struct btrfs_fs_devices *fs_devices;
375 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
376 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
383 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
384 int flush, struct block_device **bdev,
385 struct buffer_head **bh)
389 *bdev = blkdev_get_by_path(device_path, flags, holder);
392 ret = PTR_ERR(*bdev);
397 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
398 ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE);
400 blkdev_put(*bdev, flags);
403 invalidate_bdev(*bdev);
404 *bh = btrfs_read_dev_super(*bdev);
407 blkdev_put(*bdev, flags);
419 static void requeue_list(struct btrfs_pending_bios *pending_bios,
420 struct bio *head, struct bio *tail)
423 struct bio *old_head;
425 old_head = pending_bios->head;
426 pending_bios->head = head;
427 if (pending_bios->tail)
428 tail->bi_next = old_head;
430 pending_bios->tail = tail;
434 * we try to collect pending bios for a device so we don't get a large
435 * number of procs sending bios down to the same device. This greatly
436 * improves the schedulers ability to collect and merge the bios.
438 * But, it also turns into a long list of bios to process and that is sure
439 * to eventually make the worker thread block. The solution here is to
440 * make some progress and then put this work struct back at the end of
441 * the list if the block device is congested. This way, multiple devices
442 * can make progress from a single worker thread.
444 static noinline void run_scheduled_bios(struct btrfs_device *device)
446 struct btrfs_fs_info *fs_info = device->fs_info;
448 struct backing_dev_info *bdi;
449 struct btrfs_pending_bios *pending_bios;
453 unsigned long num_run;
454 unsigned long batch_run = 0;
455 unsigned long last_waited = 0;
457 int sync_pending = 0;
458 struct blk_plug plug;
461 * this function runs all the bios we've collected for
462 * a particular device. We don't want to wander off to
463 * another device without first sending all of these down.
464 * So, setup a plug here and finish it off before we return
466 blk_start_plug(&plug);
468 bdi = device->bdev->bd_bdi;
471 spin_lock(&device->io_lock);
476 /* take all the bios off the list at once and process them
477 * later on (without the lock held). But, remember the
478 * tail and other pointers so the bios can be properly reinserted
479 * into the list if we hit congestion
481 if (!force_reg && device->pending_sync_bios.head) {
482 pending_bios = &device->pending_sync_bios;
485 pending_bios = &device->pending_bios;
489 pending = pending_bios->head;
490 tail = pending_bios->tail;
491 WARN_ON(pending && !tail);
494 * if pending was null this time around, no bios need processing
495 * at all and we can stop. Otherwise it'll loop back up again
496 * and do an additional check so no bios are missed.
498 * device->running_pending is used to synchronize with the
501 if (device->pending_sync_bios.head == NULL &&
502 device->pending_bios.head == NULL) {
504 device->running_pending = 0;
507 device->running_pending = 1;
510 pending_bios->head = NULL;
511 pending_bios->tail = NULL;
513 spin_unlock(&device->io_lock);
518 /* we want to work on both lists, but do more bios on the
519 * sync list than the regular list
522 pending_bios != &device->pending_sync_bios &&
523 device->pending_sync_bios.head) ||
524 (num_run > 64 && pending_bios == &device->pending_sync_bios &&
525 device->pending_bios.head)) {
526 spin_lock(&device->io_lock);
527 requeue_list(pending_bios, pending, tail);
532 pending = pending->bi_next;
535 BUG_ON(atomic_read(&cur->__bi_cnt) == 0);
538 * if we're doing the sync list, record that our
539 * plug has some sync requests on it
541 * If we're doing the regular list and there are
542 * sync requests sitting around, unplug before
545 if (pending_bios == &device->pending_sync_bios) {
547 } else if (sync_pending) {
548 blk_finish_plug(&plug);
549 blk_start_plug(&plug);
553 btrfsic_submit_bio(cur);
560 * we made progress, there is more work to do and the bdi
561 * is now congested. Back off and let other work structs
564 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
565 fs_info->fs_devices->open_devices > 1) {
566 struct io_context *ioc;
568 ioc = current->io_context;
571 * the main goal here is that we don't want to
572 * block if we're going to be able to submit
573 * more requests without blocking.
575 * This code does two great things, it pokes into
576 * the elevator code from a filesystem _and_
577 * it makes assumptions about how batching works.
579 if (ioc && ioc->nr_batch_requests > 0 &&
580 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
582 ioc->last_waited == last_waited)) {
584 * we want to go through our batch of
585 * requests and stop. So, we copy out
586 * the ioc->last_waited time and test
587 * against it before looping
589 last_waited = ioc->last_waited;
593 spin_lock(&device->io_lock);
594 requeue_list(pending_bios, pending, tail);
595 device->running_pending = 1;
597 spin_unlock(&device->io_lock);
598 btrfs_queue_work(fs_info->submit_workers,
608 spin_lock(&device->io_lock);
609 if (device->pending_bios.head || device->pending_sync_bios.head)
611 spin_unlock(&device->io_lock);
614 blk_finish_plug(&plug);
617 static void pending_bios_fn(struct btrfs_work *work)
619 struct btrfs_device *device;
621 device = container_of(work, struct btrfs_device, work);
622 run_scheduled_bios(device);
626 * Search and remove all stale (devices which are not mounted) devices.
627 * When both inputs are NULL, it will search and release all stale devices.
628 * path: Optional. When provided will it release all unmounted devices
629 * matching this path only.
630 * skip_dev: Optional. Will skip this device when searching for the stale
633 static void btrfs_free_stale_devices(const char *path,
634 struct btrfs_device *skip_device)
636 struct btrfs_fs_devices *fs_devices, *tmp_fs_devices;
637 struct btrfs_device *device, *tmp_device;
639 list_for_each_entry_safe(fs_devices, tmp_fs_devices, &fs_uuids, fs_list) {
640 mutex_lock(&fs_devices->device_list_mutex);
641 if (fs_devices->opened) {
642 mutex_unlock(&fs_devices->device_list_mutex);
646 list_for_each_entry_safe(device, tmp_device,
647 &fs_devices->devices, dev_list) {
650 if (skip_device && skip_device == device)
652 if (path && !device->name)
657 not_found = strcmp(rcu_str_deref(device->name),
663 /* delete the stale device */
664 fs_devices->num_devices--;
665 list_del(&device->dev_list);
666 btrfs_free_device(device);
668 if (fs_devices->num_devices == 0)
671 mutex_unlock(&fs_devices->device_list_mutex);
672 if (fs_devices->num_devices == 0) {
673 btrfs_sysfs_remove_fsid(fs_devices);
674 list_del(&fs_devices->fs_list);
675 free_fs_devices(fs_devices);
680 static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices,
681 struct btrfs_device *device, fmode_t flags,
684 struct request_queue *q;
685 struct block_device *bdev;
686 struct buffer_head *bh;
687 struct btrfs_super_block *disk_super;
696 ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
701 disk_super = (struct btrfs_super_block *)bh->b_data;
702 devid = btrfs_stack_device_id(&disk_super->dev_item);
703 if (devid != device->devid)
706 if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE))
709 device->generation = btrfs_super_generation(disk_super);
711 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
712 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
713 fs_devices->seeding = 1;
715 if (bdev_read_only(bdev))
716 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
718 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
721 q = bdev_get_queue(bdev);
722 if (!blk_queue_nonrot(q))
723 fs_devices->rotating = 1;
726 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
727 device->mode = flags;
729 fs_devices->open_devices++;
730 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
731 device->devid != BTRFS_DEV_REPLACE_DEVID) {
732 fs_devices->rw_devices++;
733 list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list);
741 blkdev_put(bdev, flags);
747 * Add new device to list of registered devices
750 * device pointer which was just added or updated when successful
751 * error pointer when failed
753 static noinline struct btrfs_device *device_list_add(const char *path,
754 struct btrfs_super_block *disk_super,
755 bool *new_device_added)
757 struct btrfs_device *device;
758 struct btrfs_fs_devices *fs_devices;
759 struct rcu_string *name;
760 u64 found_transid = btrfs_super_generation(disk_super);
761 u64 devid = btrfs_stack_device_id(&disk_super->dev_item);
763 fs_devices = find_fsid(disk_super->fsid);
765 fs_devices = alloc_fs_devices(disk_super->fsid);
766 if (IS_ERR(fs_devices))
767 return ERR_CAST(fs_devices);
769 mutex_lock(&fs_devices->device_list_mutex);
770 list_add(&fs_devices->fs_list, &fs_uuids);
774 mutex_lock(&fs_devices->device_list_mutex);
775 device = find_device(fs_devices, devid,
776 disk_super->dev_item.uuid);
780 if (fs_devices->opened) {
781 mutex_unlock(&fs_devices->device_list_mutex);
782 return ERR_PTR(-EBUSY);
785 device = btrfs_alloc_device(NULL, &devid,
786 disk_super->dev_item.uuid);
787 if (IS_ERR(device)) {
788 mutex_unlock(&fs_devices->device_list_mutex);
789 /* we can safely leave the fs_devices entry around */
793 name = rcu_string_strdup(path, GFP_NOFS);
795 btrfs_free_device(device);
796 mutex_unlock(&fs_devices->device_list_mutex);
797 return ERR_PTR(-ENOMEM);
799 rcu_assign_pointer(device->name, name);
801 list_add_rcu(&device->dev_list, &fs_devices->devices);
802 fs_devices->num_devices++;
804 device->fs_devices = fs_devices;
805 *new_device_added = true;
807 if (disk_super->label[0])
808 pr_info("BTRFS: device label %s devid %llu transid %llu %s\n",
809 disk_super->label, devid, found_transid, path);
811 pr_info("BTRFS: device fsid %pU devid %llu transid %llu %s\n",
812 disk_super->fsid, devid, found_transid, path);
814 } else if (!device->name || strcmp(device->name->str, path)) {
816 * When FS is already mounted.
817 * 1. If you are here and if the device->name is NULL that
818 * means this device was missing at time of FS mount.
819 * 2. If you are here and if the device->name is different
820 * from 'path' that means either
821 * a. The same device disappeared and reappeared with
823 * b. The missing-disk-which-was-replaced, has
826 * We must allow 1 and 2a above. But 2b would be a spurious
829 * Further in case of 1 and 2a above, the disk at 'path'
830 * would have missed some transaction when it was away and
831 * in case of 2a the stale bdev has to be updated as well.
832 * 2b must not be allowed at all time.
836 * For now, we do allow update to btrfs_fs_device through the
837 * btrfs dev scan cli after FS has been mounted. We're still
838 * tracking a problem where systems fail mount by subvolume id
839 * when we reject replacement on a mounted FS.
841 if (!fs_devices->opened && found_transid < device->generation) {
843 * That is if the FS is _not_ mounted and if you
844 * are here, that means there is more than one
845 * disk with same uuid and devid.We keep the one
846 * with larger generation number or the last-in if
847 * generation are equal.
849 mutex_unlock(&fs_devices->device_list_mutex);
850 return ERR_PTR(-EEXIST);
853 name = rcu_string_strdup(path, GFP_NOFS);
855 mutex_unlock(&fs_devices->device_list_mutex);
856 return ERR_PTR(-ENOMEM);
858 rcu_string_free(device->name);
859 rcu_assign_pointer(device->name, name);
860 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
861 fs_devices->missing_devices--;
862 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
867 * Unmount does not free the btrfs_device struct but would zero
868 * generation along with most of the other members. So just update
869 * it back. We need it to pick the disk with largest generation
872 if (!fs_devices->opened)
873 device->generation = found_transid;
875 fs_devices->total_devices = btrfs_super_num_devices(disk_super);
877 mutex_unlock(&fs_devices->device_list_mutex);
881 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
883 struct btrfs_fs_devices *fs_devices;
884 struct btrfs_device *device;
885 struct btrfs_device *orig_dev;
887 fs_devices = alloc_fs_devices(orig->fsid);
888 if (IS_ERR(fs_devices))
891 mutex_lock(&orig->device_list_mutex);
892 fs_devices->total_devices = orig->total_devices;
894 /* We have held the volume lock, it is safe to get the devices. */
895 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
896 struct rcu_string *name;
898 device = btrfs_alloc_device(NULL, &orig_dev->devid,
904 * This is ok to do without rcu read locked because we hold the
905 * uuid mutex so nothing we touch in here is going to disappear.
907 if (orig_dev->name) {
908 name = rcu_string_strdup(orig_dev->name->str,
911 btrfs_free_device(device);
914 rcu_assign_pointer(device->name, name);
917 list_add(&device->dev_list, &fs_devices->devices);
918 device->fs_devices = fs_devices;
919 fs_devices->num_devices++;
921 mutex_unlock(&orig->device_list_mutex);
924 mutex_unlock(&orig->device_list_mutex);
925 free_fs_devices(fs_devices);
926 return ERR_PTR(-ENOMEM);
930 * After we have read the system tree and know devids belonging to
931 * this filesystem, remove the device which does not belong there.
933 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices, int step)
935 struct btrfs_device *device, *next;
936 struct btrfs_device *latest_dev = NULL;
938 mutex_lock(&uuid_mutex);
940 /* This is the initialized path, it is safe to release the devices. */
941 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
942 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
943 &device->dev_state)) {
944 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
945 &device->dev_state) &&
947 device->generation > latest_dev->generation)) {
953 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
955 * In the first step, keep the device which has
956 * the correct fsid and the devid that is used
957 * for the dev_replace procedure.
958 * In the second step, the dev_replace state is
959 * read from the device tree and it is known
960 * whether the procedure is really active or
961 * not, which means whether this device is
962 * used or whether it should be removed.
964 if (step == 0 || test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
965 &device->dev_state)) {
970 blkdev_put(device->bdev, device->mode);
972 fs_devices->open_devices--;
974 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
975 list_del_init(&device->dev_alloc_list);
976 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
977 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
979 fs_devices->rw_devices--;
981 list_del_init(&device->dev_list);
982 fs_devices->num_devices--;
983 btrfs_free_device(device);
986 if (fs_devices->seed) {
987 fs_devices = fs_devices->seed;
991 fs_devices->latest_bdev = latest_dev->bdev;
993 mutex_unlock(&uuid_mutex);
996 static void free_device_rcu(struct rcu_head *head)
998 struct btrfs_device *device;
1000 device = container_of(head, struct btrfs_device, rcu);
1001 btrfs_free_device(device);
1004 static void btrfs_close_bdev(struct btrfs_device *device)
1009 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1010 sync_blockdev(device->bdev);
1011 invalidate_bdev(device->bdev);
1014 blkdev_put(device->bdev, device->mode);
1017 static void btrfs_close_one_device(struct btrfs_device *device)
1019 struct btrfs_fs_devices *fs_devices = device->fs_devices;
1020 struct btrfs_device *new_device;
1021 struct rcu_string *name;
1024 fs_devices->open_devices--;
1026 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
1027 device->devid != BTRFS_DEV_REPLACE_DEVID) {
1028 list_del_init(&device->dev_alloc_list);
1029 fs_devices->rw_devices--;
1032 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
1033 fs_devices->missing_devices--;
1035 btrfs_close_bdev(device);
1037 new_device = btrfs_alloc_device(NULL, &device->devid,
1039 BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
1041 /* Safe because we are under uuid_mutex */
1043 name = rcu_string_strdup(device->name->str, GFP_NOFS);
1044 BUG_ON(!name); /* -ENOMEM */
1045 rcu_assign_pointer(new_device->name, name);
1048 list_replace_rcu(&device->dev_list, &new_device->dev_list);
1049 new_device->fs_devices = device->fs_devices;
1051 call_rcu(&device->rcu, free_device_rcu);
1054 static int close_fs_devices(struct btrfs_fs_devices *fs_devices)
1056 struct btrfs_device *device, *tmp;
1058 if (--fs_devices->opened > 0)
1061 mutex_lock(&fs_devices->device_list_mutex);
1062 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
1063 btrfs_close_one_device(device);
1065 mutex_unlock(&fs_devices->device_list_mutex);
1067 WARN_ON(fs_devices->open_devices);
1068 WARN_ON(fs_devices->rw_devices);
1069 fs_devices->opened = 0;
1070 fs_devices->seeding = 0;
1075 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
1077 struct btrfs_fs_devices *seed_devices = NULL;
1080 mutex_lock(&uuid_mutex);
1081 ret = close_fs_devices(fs_devices);
1082 if (!fs_devices->opened) {
1083 seed_devices = fs_devices->seed;
1084 fs_devices->seed = NULL;
1086 mutex_unlock(&uuid_mutex);
1088 while (seed_devices) {
1089 fs_devices = seed_devices;
1090 seed_devices = fs_devices->seed;
1091 close_fs_devices(fs_devices);
1092 free_fs_devices(fs_devices);
1097 static int open_fs_devices(struct btrfs_fs_devices *fs_devices,
1098 fmode_t flags, void *holder)
1100 struct btrfs_device *device;
1101 struct btrfs_device *latest_dev = NULL;
1104 flags |= FMODE_EXCL;
1106 list_for_each_entry(device, &fs_devices->devices, dev_list) {
1107 /* Just open everything we can; ignore failures here */
1108 if (btrfs_open_one_device(fs_devices, device, flags, holder))
1112 device->generation > latest_dev->generation)
1113 latest_dev = device;
1115 if (fs_devices->open_devices == 0) {
1119 fs_devices->opened = 1;
1120 fs_devices->latest_bdev = latest_dev->bdev;
1121 fs_devices->total_rw_bytes = 0;
1126 static int devid_cmp(void *priv, struct list_head *a, struct list_head *b)
1128 struct btrfs_device *dev1, *dev2;
1130 dev1 = list_entry(a, struct btrfs_device, dev_list);
1131 dev2 = list_entry(b, struct btrfs_device, dev_list);
1133 if (dev1->devid < dev2->devid)
1135 else if (dev1->devid > dev2->devid)
1140 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1141 fmode_t flags, void *holder)
1145 lockdep_assert_held(&uuid_mutex);
1147 mutex_lock(&fs_devices->device_list_mutex);
1148 if (fs_devices->opened) {
1149 fs_devices->opened++;
1152 list_sort(NULL, &fs_devices->devices, devid_cmp);
1153 ret = open_fs_devices(fs_devices, flags, holder);
1155 mutex_unlock(&fs_devices->device_list_mutex);
1160 static void btrfs_release_disk_super(struct page *page)
1166 static int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr,
1168 struct btrfs_super_block **disk_super)
1173 /* make sure our super fits in the device */
1174 if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1177 /* make sure our super fits in the page */
1178 if (sizeof(**disk_super) > PAGE_SIZE)
1181 /* make sure our super doesn't straddle pages on disk */
1182 index = bytenr >> PAGE_SHIFT;
1183 if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index)
1186 /* pull in the page with our super */
1187 *page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
1190 if (IS_ERR_OR_NULL(*page))
1195 /* align our pointer to the offset of the super block */
1196 *disk_super = p + (bytenr & ~PAGE_MASK);
1198 if (btrfs_super_bytenr(*disk_super) != bytenr ||
1199 btrfs_super_magic(*disk_super) != BTRFS_MAGIC) {
1200 btrfs_release_disk_super(*page);
1204 if ((*disk_super)->label[0] &&
1205 (*disk_super)->label[BTRFS_LABEL_SIZE - 1])
1206 (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0';
1212 * Look for a btrfs signature on a device. This may be called out of the mount path
1213 * and we are not allowed to call set_blocksize during the scan. The superblock
1214 * is read via pagecache
1216 struct btrfs_device *btrfs_scan_one_device(const char *path, fmode_t flags,
1219 struct btrfs_super_block *disk_super;
1220 bool new_device_added = false;
1221 struct btrfs_device *device = NULL;
1222 struct block_device *bdev;
1226 lockdep_assert_held(&uuid_mutex);
1229 * we would like to check all the supers, but that would make
1230 * a btrfs mount succeed after a mkfs from a different FS.
1231 * So, we need to add a special mount option to scan for
1232 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1234 bytenr = btrfs_sb_offset(0);
1235 flags |= FMODE_EXCL;
1237 bdev = blkdev_get_by_path(path, flags, holder);
1239 return ERR_CAST(bdev);
1241 if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super)) {
1242 device = ERR_PTR(-EINVAL);
1243 goto error_bdev_put;
1246 device = device_list_add(path, disk_super, &new_device_added);
1247 if (!IS_ERR(device)) {
1248 if (new_device_added)
1249 btrfs_free_stale_devices(path, device);
1252 btrfs_release_disk_super(page);
1255 blkdev_put(bdev, flags);
1260 static int contains_pending_extent(struct btrfs_transaction *transaction,
1261 struct btrfs_device *device,
1262 u64 *start, u64 len)
1264 struct btrfs_fs_info *fs_info = device->fs_info;
1265 struct extent_map *em;
1266 struct list_head *search_list = &fs_info->pinned_chunks;
1268 u64 physical_start = *start;
1271 search_list = &transaction->pending_chunks;
1273 list_for_each_entry(em, search_list, list) {
1274 struct map_lookup *map;
1277 map = em->map_lookup;
1278 for (i = 0; i < map->num_stripes; i++) {
1281 if (map->stripes[i].dev != device)
1283 if (map->stripes[i].physical >= physical_start + len ||
1284 map->stripes[i].physical + em->orig_block_len <=
1288 * Make sure that while processing the pinned list we do
1289 * not override our *start with a lower value, because
1290 * we can have pinned chunks that fall within this
1291 * device hole and that have lower physical addresses
1292 * than the pending chunks we processed before. If we
1293 * do not take this special care we can end up getting
1294 * 2 pending chunks that start at the same physical
1295 * device offsets because the end offset of a pinned
1296 * chunk can be equal to the start offset of some
1299 end = map->stripes[i].physical + em->orig_block_len;
1306 if (search_list != &fs_info->pinned_chunks) {
1307 search_list = &fs_info->pinned_chunks;
1316 * find_free_dev_extent_start - find free space in the specified device
1317 * @device: the device which we search the free space in
1318 * @num_bytes: the size of the free space that we need
1319 * @search_start: the position from which to begin the search
1320 * @start: store the start of the free space.
1321 * @len: the size of the free space. that we find, or the size
1322 * of the max free space if we don't find suitable free space
1324 * this uses a pretty simple search, the expectation is that it is
1325 * called very infrequently and that a given device has a small number
1328 * @start is used to store the start of the free space if we find. But if we
1329 * don't find suitable free space, it will be used to store the start position
1330 * of the max free space.
1332 * @len is used to store the size of the free space that we find.
1333 * But if we don't find suitable free space, it is used to store the size of
1334 * the max free space.
1336 int find_free_dev_extent_start(struct btrfs_transaction *transaction,
1337 struct btrfs_device *device, u64 num_bytes,
1338 u64 search_start, u64 *start, u64 *len)
1340 struct btrfs_fs_info *fs_info = device->fs_info;
1341 struct btrfs_root *root = fs_info->dev_root;
1342 struct btrfs_key key;
1343 struct btrfs_dev_extent *dev_extent;
1344 struct btrfs_path *path;
1349 u64 search_end = device->total_bytes;
1352 struct extent_buffer *l;
1355 * We don't want to overwrite the superblock on the drive nor any area
1356 * used by the boot loader (grub for example), so we make sure to start
1357 * at an offset of at least 1MB.
1359 search_start = max_t(u64, search_start, SZ_1M);
1361 path = btrfs_alloc_path();
1365 max_hole_start = search_start;
1369 if (search_start >= search_end ||
1370 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1375 path->reada = READA_FORWARD;
1376 path->search_commit_root = 1;
1377 path->skip_locking = 1;
1379 key.objectid = device->devid;
1380 key.offset = search_start;
1381 key.type = BTRFS_DEV_EXTENT_KEY;
1383 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1387 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1394 slot = path->slots[0];
1395 if (slot >= btrfs_header_nritems(l)) {
1396 ret = btrfs_next_leaf(root, path);
1404 btrfs_item_key_to_cpu(l, &key, slot);
1406 if (key.objectid < device->devid)
1409 if (key.objectid > device->devid)
1412 if (key.type != BTRFS_DEV_EXTENT_KEY)
1415 if (key.offset > search_start) {
1416 hole_size = key.offset - search_start;
1419 * Have to check before we set max_hole_start, otherwise
1420 * we could end up sending back this offset anyway.
1422 if (contains_pending_extent(transaction, device,
1425 if (key.offset >= search_start) {
1426 hole_size = key.offset - search_start;
1433 if (hole_size > max_hole_size) {
1434 max_hole_start = search_start;
1435 max_hole_size = hole_size;
1439 * If this free space is greater than which we need,
1440 * it must be the max free space that we have found
1441 * until now, so max_hole_start must point to the start
1442 * of this free space and the length of this free space
1443 * is stored in max_hole_size. Thus, we return
1444 * max_hole_start and max_hole_size and go back to the
1447 if (hole_size >= num_bytes) {
1453 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1454 extent_end = key.offset + btrfs_dev_extent_length(l,
1456 if (extent_end > search_start)
1457 search_start = extent_end;
1464 * At this point, search_start should be the end of
1465 * allocated dev extents, and when shrinking the device,
1466 * search_end may be smaller than search_start.
1468 if (search_end > search_start) {
1469 hole_size = search_end - search_start;
1471 if (contains_pending_extent(transaction, device, &search_start,
1473 btrfs_release_path(path);
1477 if (hole_size > max_hole_size) {
1478 max_hole_start = search_start;
1479 max_hole_size = hole_size;
1484 if (max_hole_size < num_bytes)
1490 btrfs_free_path(path);
1491 *start = max_hole_start;
1493 *len = max_hole_size;
1497 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1498 struct btrfs_device *device, u64 num_bytes,
1499 u64 *start, u64 *len)
1501 /* FIXME use last free of some kind */
1502 return find_free_dev_extent_start(trans->transaction, device,
1503 num_bytes, 0, start, len);
1506 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1507 struct btrfs_device *device,
1508 u64 start, u64 *dev_extent_len)
1510 struct btrfs_fs_info *fs_info = device->fs_info;
1511 struct btrfs_root *root = fs_info->dev_root;
1513 struct btrfs_path *path;
1514 struct btrfs_key key;
1515 struct btrfs_key found_key;
1516 struct extent_buffer *leaf = NULL;
1517 struct btrfs_dev_extent *extent = NULL;
1519 path = btrfs_alloc_path();
1523 key.objectid = device->devid;
1525 key.type = BTRFS_DEV_EXTENT_KEY;
1527 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1529 ret = btrfs_previous_item(root, path, key.objectid,
1530 BTRFS_DEV_EXTENT_KEY);
1533 leaf = path->nodes[0];
1534 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1535 extent = btrfs_item_ptr(leaf, path->slots[0],
1536 struct btrfs_dev_extent);
1537 BUG_ON(found_key.offset > start || found_key.offset +
1538 btrfs_dev_extent_length(leaf, extent) < start);
1540 btrfs_release_path(path);
1542 } else if (ret == 0) {
1543 leaf = path->nodes[0];
1544 extent = btrfs_item_ptr(leaf, path->slots[0],
1545 struct btrfs_dev_extent);
1547 btrfs_handle_fs_error(fs_info, ret, "Slot search failed");
1551 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1553 ret = btrfs_del_item(trans, root, path);
1555 btrfs_handle_fs_error(fs_info, ret,
1556 "Failed to remove dev extent item");
1558 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1561 btrfs_free_path(path);
1565 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1566 struct btrfs_device *device,
1567 u64 chunk_offset, u64 start, u64 num_bytes)
1570 struct btrfs_path *path;
1571 struct btrfs_fs_info *fs_info = device->fs_info;
1572 struct btrfs_root *root = fs_info->dev_root;
1573 struct btrfs_dev_extent *extent;
1574 struct extent_buffer *leaf;
1575 struct btrfs_key key;
1577 WARN_ON(!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state));
1578 WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state));
1579 path = btrfs_alloc_path();
1583 key.objectid = device->devid;
1585 key.type = BTRFS_DEV_EXTENT_KEY;
1586 ret = btrfs_insert_empty_item(trans, root, path, &key,
1591 leaf = path->nodes[0];
1592 extent = btrfs_item_ptr(leaf, path->slots[0],
1593 struct btrfs_dev_extent);
1594 btrfs_set_dev_extent_chunk_tree(leaf, extent,
1595 BTRFS_CHUNK_TREE_OBJECTID);
1596 btrfs_set_dev_extent_chunk_objectid(leaf, extent,
1597 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
1598 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1600 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1601 btrfs_mark_buffer_dirty(leaf);
1603 btrfs_free_path(path);
1607 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1609 struct extent_map_tree *em_tree;
1610 struct extent_map *em;
1614 em_tree = &fs_info->mapping_tree.map_tree;
1615 read_lock(&em_tree->lock);
1616 n = rb_last(&em_tree->map);
1618 em = rb_entry(n, struct extent_map, rb_node);
1619 ret = em->start + em->len;
1621 read_unlock(&em_tree->lock);
1626 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1630 struct btrfs_key key;
1631 struct btrfs_key found_key;
1632 struct btrfs_path *path;
1634 path = btrfs_alloc_path();
1638 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1639 key.type = BTRFS_DEV_ITEM_KEY;
1640 key.offset = (u64)-1;
1642 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1646 BUG_ON(ret == 0); /* Corruption */
1648 ret = btrfs_previous_item(fs_info->chunk_root, path,
1649 BTRFS_DEV_ITEMS_OBJECTID,
1650 BTRFS_DEV_ITEM_KEY);
1654 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1656 *devid_ret = found_key.offset + 1;
1660 btrfs_free_path(path);
1665 * the device information is stored in the chunk root
1666 * the btrfs_device struct should be fully filled in
1668 static int btrfs_add_dev_item(struct btrfs_trans_handle *trans,
1669 struct btrfs_device *device)
1672 struct btrfs_path *path;
1673 struct btrfs_dev_item *dev_item;
1674 struct extent_buffer *leaf;
1675 struct btrfs_key key;
1678 path = btrfs_alloc_path();
1682 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1683 key.type = BTRFS_DEV_ITEM_KEY;
1684 key.offset = device->devid;
1686 ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path,
1687 &key, sizeof(*dev_item));
1691 leaf = path->nodes[0];
1692 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1694 btrfs_set_device_id(leaf, dev_item, device->devid);
1695 btrfs_set_device_generation(leaf, dev_item, 0);
1696 btrfs_set_device_type(leaf, dev_item, device->type);
1697 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1698 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1699 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1700 btrfs_set_device_total_bytes(leaf, dev_item,
1701 btrfs_device_get_disk_total_bytes(device));
1702 btrfs_set_device_bytes_used(leaf, dev_item,
1703 btrfs_device_get_bytes_used(device));
1704 btrfs_set_device_group(leaf, dev_item, 0);
1705 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1706 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1707 btrfs_set_device_start_offset(leaf, dev_item, 0);
1709 ptr = btrfs_device_uuid(dev_item);
1710 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1711 ptr = btrfs_device_fsid(dev_item);
1712 write_extent_buffer(leaf, trans->fs_info->fsid, ptr, BTRFS_FSID_SIZE);
1713 btrfs_mark_buffer_dirty(leaf);
1717 btrfs_free_path(path);
1722 * Function to update ctime/mtime for a given device path.
1723 * Mainly used for ctime/mtime based probe like libblkid.
1725 static void update_dev_time(const char *path_name)
1729 filp = filp_open(path_name, O_RDWR, 0);
1732 file_update_time(filp);
1733 filp_close(filp, NULL);
1736 static int btrfs_rm_dev_item(struct btrfs_fs_info *fs_info,
1737 struct btrfs_device *device)
1739 struct btrfs_root *root = fs_info->chunk_root;
1741 struct btrfs_path *path;
1742 struct btrfs_key key;
1743 struct btrfs_trans_handle *trans;
1745 path = btrfs_alloc_path();
1749 trans = btrfs_start_transaction(root, 0);
1750 if (IS_ERR(trans)) {
1751 btrfs_free_path(path);
1752 return PTR_ERR(trans);
1754 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1755 key.type = BTRFS_DEV_ITEM_KEY;
1756 key.offset = device->devid;
1758 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1762 btrfs_abort_transaction(trans, ret);
1763 btrfs_end_transaction(trans);
1767 ret = btrfs_del_item(trans, root, path);
1769 btrfs_abort_transaction(trans, ret);
1770 btrfs_end_transaction(trans);
1774 btrfs_free_path(path);
1776 ret = btrfs_commit_transaction(trans);
1781 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1782 * filesystem. It's up to the caller to adjust that number regarding eg. device
1785 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1793 seq = read_seqbegin(&fs_info->profiles_lock);
1795 all_avail = fs_info->avail_data_alloc_bits |
1796 fs_info->avail_system_alloc_bits |
1797 fs_info->avail_metadata_alloc_bits;
1798 } while (read_seqretry(&fs_info->profiles_lock, seq));
1800 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1801 if (!(all_avail & btrfs_raid_array[i].bg_flag))
1804 if (num_devices < btrfs_raid_array[i].devs_min) {
1805 int ret = btrfs_raid_array[i].mindev_error;
1815 static struct btrfs_device * btrfs_find_next_active_device(
1816 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device)
1818 struct btrfs_device *next_device;
1820 list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1821 if (next_device != device &&
1822 !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state)
1823 && next_device->bdev)
1831 * Helper function to check if the given device is part of s_bdev / latest_bdev
1832 * and replace it with the provided or the next active device, in the context
1833 * where this function called, there should be always be another device (or
1834 * this_dev) which is active.
1836 void btrfs_assign_next_active_device(struct btrfs_device *device,
1837 struct btrfs_device *this_dev)
1839 struct btrfs_fs_info *fs_info = device->fs_info;
1840 struct btrfs_device *next_device;
1843 next_device = this_dev;
1845 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
1847 ASSERT(next_device);
1849 if (fs_info->sb->s_bdev &&
1850 (fs_info->sb->s_bdev == device->bdev))
1851 fs_info->sb->s_bdev = next_device->bdev;
1853 if (fs_info->fs_devices->latest_bdev == device->bdev)
1854 fs_info->fs_devices->latest_bdev = next_device->bdev;
1857 int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path,
1860 struct btrfs_device *device;
1861 struct btrfs_fs_devices *cur_devices;
1862 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
1866 mutex_lock(&uuid_mutex);
1868 num_devices = fs_devices->num_devices;
1869 btrfs_dev_replace_read_lock(&fs_info->dev_replace);
1870 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
1871 WARN_ON(num_devices < 1);
1874 btrfs_dev_replace_read_unlock(&fs_info->dev_replace);
1876 ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
1880 ret = btrfs_find_device_by_devspec(fs_info, devid, device_path,
1885 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1886 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1890 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
1891 fs_info->fs_devices->rw_devices == 1) {
1892 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1896 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1897 mutex_lock(&fs_info->chunk_mutex);
1898 list_del_init(&device->dev_alloc_list);
1899 device->fs_devices->rw_devices--;
1900 mutex_unlock(&fs_info->chunk_mutex);
1903 mutex_unlock(&uuid_mutex);
1904 ret = btrfs_shrink_device(device, 0);
1905 mutex_lock(&uuid_mutex);
1910 * TODO: the superblock still includes this device in its num_devices
1911 * counter although write_all_supers() is not locked out. This
1912 * could give a filesystem state which requires a degraded mount.
1914 ret = btrfs_rm_dev_item(fs_info, device);
1918 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
1919 btrfs_scrub_cancel_dev(fs_info, device);
1922 * the device list mutex makes sure that we don't change
1923 * the device list while someone else is writing out all
1924 * the device supers. Whoever is writing all supers, should
1925 * lock the device list mutex before getting the number of
1926 * devices in the super block (super_copy). Conversely,
1927 * whoever updates the number of devices in the super block
1928 * (super_copy) should hold the device list mutex.
1932 * In normal cases the cur_devices == fs_devices. But in case
1933 * of deleting a seed device, the cur_devices should point to
1934 * its own fs_devices listed under the fs_devices->seed.
1936 cur_devices = device->fs_devices;
1937 mutex_lock(&fs_devices->device_list_mutex);
1938 list_del_rcu(&device->dev_list);
1940 cur_devices->num_devices--;
1941 cur_devices->total_devices--;
1942 /* Update total_devices of the parent fs_devices if it's seed */
1943 if (cur_devices != fs_devices)
1944 fs_devices->total_devices--;
1946 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
1947 cur_devices->missing_devices--;
1949 btrfs_assign_next_active_device(device, NULL);
1952 cur_devices->open_devices--;
1953 /* remove sysfs entry */
1954 btrfs_sysfs_rm_device_link(fs_devices, device);
1957 num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
1958 btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
1959 mutex_unlock(&fs_devices->device_list_mutex);
1962 * at this point, the device is zero sized and detached from
1963 * the devices list. All that's left is to zero out the old
1964 * supers and free the device.
1966 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
1967 btrfs_scratch_superblocks(device->bdev, device->name->str);
1969 btrfs_close_bdev(device);
1970 call_rcu(&device->rcu, free_device_rcu);
1972 if (cur_devices->open_devices == 0) {
1973 while (fs_devices) {
1974 if (fs_devices->seed == cur_devices) {
1975 fs_devices->seed = cur_devices->seed;
1978 fs_devices = fs_devices->seed;
1980 cur_devices->seed = NULL;
1981 close_fs_devices(cur_devices);
1982 free_fs_devices(cur_devices);
1986 mutex_unlock(&uuid_mutex);
1990 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1991 mutex_lock(&fs_info->chunk_mutex);
1992 list_add(&device->dev_alloc_list,
1993 &fs_devices->alloc_list);
1994 device->fs_devices->rw_devices++;
1995 mutex_unlock(&fs_info->chunk_mutex);
2000 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev)
2002 struct btrfs_fs_devices *fs_devices;
2004 lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex);
2007 * in case of fs with no seed, srcdev->fs_devices will point
2008 * to fs_devices of fs_info. However when the dev being replaced is
2009 * a seed dev it will point to the seed's local fs_devices. In short
2010 * srcdev will have its correct fs_devices in both the cases.
2012 fs_devices = srcdev->fs_devices;
2014 list_del_rcu(&srcdev->dev_list);
2015 list_del(&srcdev->dev_alloc_list);
2016 fs_devices->num_devices--;
2017 if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state))
2018 fs_devices->missing_devices--;
2020 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state))
2021 fs_devices->rw_devices--;
2024 fs_devices->open_devices--;
2027 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
2028 struct btrfs_device *srcdev)
2030 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2032 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) {
2033 /* zero out the old super if it is writable */
2034 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
2037 btrfs_close_bdev(srcdev);
2038 call_rcu(&srcdev->rcu, free_device_rcu);
2040 /* if this is no devs we rather delete the fs_devices */
2041 if (!fs_devices->num_devices) {
2042 struct btrfs_fs_devices *tmp_fs_devices;
2045 * On a mounted FS, num_devices can't be zero unless it's a
2046 * seed. In case of a seed device being replaced, the replace
2047 * target added to the sprout FS, so there will be no more
2048 * device left under the seed FS.
2050 ASSERT(fs_devices->seeding);
2052 tmp_fs_devices = fs_info->fs_devices;
2053 while (tmp_fs_devices) {
2054 if (tmp_fs_devices->seed == fs_devices) {
2055 tmp_fs_devices->seed = fs_devices->seed;
2058 tmp_fs_devices = tmp_fs_devices->seed;
2060 fs_devices->seed = NULL;
2061 close_fs_devices(fs_devices);
2062 free_fs_devices(fs_devices);
2066 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev)
2068 struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices;
2071 mutex_lock(&fs_devices->device_list_mutex);
2073 btrfs_sysfs_rm_device_link(fs_devices, tgtdev);
2076 fs_devices->open_devices--;
2078 fs_devices->num_devices--;
2080 btrfs_assign_next_active_device(tgtdev, NULL);
2082 list_del_rcu(&tgtdev->dev_list);
2084 mutex_unlock(&fs_devices->device_list_mutex);
2087 * The update_dev_time() with in btrfs_scratch_superblocks()
2088 * may lead to a call to btrfs_show_devname() which will try
2089 * to hold device_list_mutex. And here this device
2090 * is already out of device list, so we don't have to hold
2091 * the device_list_mutex lock.
2093 btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
2095 btrfs_close_bdev(tgtdev);
2096 call_rcu(&tgtdev->rcu, free_device_rcu);
2099 static int btrfs_find_device_by_path(struct btrfs_fs_info *fs_info,
2100 const char *device_path,
2101 struct btrfs_device **device)
2104 struct btrfs_super_block *disk_super;
2107 struct block_device *bdev;
2108 struct buffer_head *bh;
2111 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2112 fs_info->bdev_holder, 0, &bdev, &bh);
2115 disk_super = (struct btrfs_super_block *)bh->b_data;
2116 devid = btrfs_stack_device_id(&disk_super->dev_item);
2117 dev_uuid = disk_super->dev_item.uuid;
2118 *device = btrfs_find_device(fs_info, devid, dev_uuid, disk_super->fsid);
2122 blkdev_put(bdev, FMODE_READ);
2126 int btrfs_find_device_missing_or_by_path(struct btrfs_fs_info *fs_info,
2127 const char *device_path,
2128 struct btrfs_device **device)
2131 if (strcmp(device_path, "missing") == 0) {
2132 struct list_head *devices;
2133 struct btrfs_device *tmp;
2135 devices = &fs_info->fs_devices->devices;
2136 list_for_each_entry(tmp, devices, dev_list) {
2137 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
2138 &tmp->dev_state) && !tmp->bdev) {
2145 return BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2149 return btrfs_find_device_by_path(fs_info, device_path, device);
2154 * Lookup a device given by device id, or the path if the id is 0.
2156 int btrfs_find_device_by_devspec(struct btrfs_fs_info *fs_info, u64 devid,
2157 const char *devpath,
2158 struct btrfs_device **device)
2164 *device = btrfs_find_device(fs_info, devid, NULL, NULL);
2168 if (!devpath || !devpath[0])
2171 ret = btrfs_find_device_missing_or_by_path(fs_info, devpath,
2178 * does all the dirty work required for changing file system's UUID.
2180 static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info)
2182 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2183 struct btrfs_fs_devices *old_devices;
2184 struct btrfs_fs_devices *seed_devices;
2185 struct btrfs_super_block *disk_super = fs_info->super_copy;
2186 struct btrfs_device *device;
2189 lockdep_assert_held(&uuid_mutex);
2190 if (!fs_devices->seeding)
2193 seed_devices = alloc_fs_devices(NULL);
2194 if (IS_ERR(seed_devices))
2195 return PTR_ERR(seed_devices);
2197 old_devices = clone_fs_devices(fs_devices);
2198 if (IS_ERR(old_devices)) {
2199 kfree(seed_devices);
2200 return PTR_ERR(old_devices);
2203 list_add(&old_devices->fs_list, &fs_uuids);
2205 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2206 seed_devices->opened = 1;
2207 INIT_LIST_HEAD(&seed_devices->devices);
2208 INIT_LIST_HEAD(&seed_devices->alloc_list);
2209 mutex_init(&seed_devices->device_list_mutex);
2211 mutex_lock(&fs_devices->device_list_mutex);
2212 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2214 list_for_each_entry(device, &seed_devices->devices, dev_list)
2215 device->fs_devices = seed_devices;
2217 mutex_lock(&fs_info->chunk_mutex);
2218 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2219 mutex_unlock(&fs_info->chunk_mutex);
2221 fs_devices->seeding = 0;
2222 fs_devices->num_devices = 0;
2223 fs_devices->open_devices = 0;
2224 fs_devices->missing_devices = 0;
2225 fs_devices->rotating = 0;
2226 fs_devices->seed = seed_devices;
2228 generate_random_uuid(fs_devices->fsid);
2229 memcpy(fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2230 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2231 mutex_unlock(&fs_devices->device_list_mutex);
2233 super_flags = btrfs_super_flags(disk_super) &
2234 ~BTRFS_SUPER_FLAG_SEEDING;
2235 btrfs_set_super_flags(disk_super, super_flags);
2241 * Store the expected generation for seed devices in device items.
2243 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2244 struct btrfs_fs_info *fs_info)
2246 struct btrfs_root *root = fs_info->chunk_root;
2247 struct btrfs_path *path;
2248 struct extent_buffer *leaf;
2249 struct btrfs_dev_item *dev_item;
2250 struct btrfs_device *device;
2251 struct btrfs_key key;
2252 u8 fs_uuid[BTRFS_FSID_SIZE];
2253 u8 dev_uuid[BTRFS_UUID_SIZE];
2257 path = btrfs_alloc_path();
2261 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2263 key.type = BTRFS_DEV_ITEM_KEY;
2266 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2270 leaf = path->nodes[0];
2272 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2273 ret = btrfs_next_leaf(root, path);
2278 leaf = path->nodes[0];
2279 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2280 btrfs_release_path(path);
2284 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2285 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2286 key.type != BTRFS_DEV_ITEM_KEY)
2289 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2290 struct btrfs_dev_item);
2291 devid = btrfs_device_id(leaf, dev_item);
2292 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2294 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2296 device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
2297 BUG_ON(!device); /* Logic error */
2299 if (device->fs_devices->seeding) {
2300 btrfs_set_device_generation(leaf, dev_item,
2301 device->generation);
2302 btrfs_mark_buffer_dirty(leaf);
2310 btrfs_free_path(path);
2314 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2316 struct btrfs_root *root = fs_info->dev_root;
2317 struct request_queue *q;
2318 struct btrfs_trans_handle *trans;
2319 struct btrfs_device *device;
2320 struct block_device *bdev;
2321 struct super_block *sb = fs_info->sb;
2322 struct rcu_string *name;
2323 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2325 int seeding_dev = 0;
2327 bool unlocked = false;
2329 if (sb_rdonly(sb) && !fs_devices->seeding)
2332 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2333 fs_info->bdev_holder);
2335 return PTR_ERR(bdev);
2337 if (fs_devices->seeding) {
2339 down_write(&sb->s_umount);
2340 mutex_lock(&uuid_mutex);
2343 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2345 mutex_lock(&fs_devices->device_list_mutex);
2346 list_for_each_entry(device, &fs_devices->devices, dev_list) {
2347 if (device->bdev == bdev) {
2350 &fs_devices->device_list_mutex);
2354 mutex_unlock(&fs_devices->device_list_mutex);
2356 device = btrfs_alloc_device(fs_info, NULL, NULL);
2357 if (IS_ERR(device)) {
2358 /* we can safely leave the fs_devices entry around */
2359 ret = PTR_ERR(device);
2363 name = rcu_string_strdup(device_path, GFP_KERNEL);
2366 goto error_free_device;
2368 rcu_assign_pointer(device->name, name);
2370 trans = btrfs_start_transaction(root, 0);
2371 if (IS_ERR(trans)) {
2372 ret = PTR_ERR(trans);
2373 goto error_free_device;
2376 q = bdev_get_queue(bdev);
2377 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
2378 device->generation = trans->transid;
2379 device->io_width = fs_info->sectorsize;
2380 device->io_align = fs_info->sectorsize;
2381 device->sector_size = fs_info->sectorsize;
2382 device->total_bytes = round_down(i_size_read(bdev->bd_inode),
2383 fs_info->sectorsize);
2384 device->disk_total_bytes = device->total_bytes;
2385 device->commit_total_bytes = device->total_bytes;
2386 device->fs_info = fs_info;
2387 device->bdev = bdev;
2388 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2389 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
2390 device->mode = FMODE_EXCL;
2391 device->dev_stats_valid = 1;
2392 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2395 sb->s_flags &= ~SB_RDONLY;
2396 ret = btrfs_prepare_sprout(fs_info);
2398 btrfs_abort_transaction(trans, ret);
2403 device->fs_devices = fs_devices;
2405 mutex_lock(&fs_devices->device_list_mutex);
2406 mutex_lock(&fs_info->chunk_mutex);
2407 list_add_rcu(&device->dev_list, &fs_devices->devices);
2408 list_add(&device->dev_alloc_list, &fs_devices->alloc_list);
2409 fs_devices->num_devices++;
2410 fs_devices->open_devices++;
2411 fs_devices->rw_devices++;
2412 fs_devices->total_devices++;
2413 fs_devices->total_rw_bytes += device->total_bytes;
2415 atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2417 if (!blk_queue_nonrot(q))
2418 fs_devices->rotating = 1;
2420 tmp = btrfs_super_total_bytes(fs_info->super_copy);
2421 btrfs_set_super_total_bytes(fs_info->super_copy,
2422 round_down(tmp + device->total_bytes, fs_info->sectorsize));
2424 tmp = btrfs_super_num_devices(fs_info->super_copy);
2425 btrfs_set_super_num_devices(fs_info->super_copy, tmp + 1);
2427 /* add sysfs device entry */
2428 btrfs_sysfs_add_device_link(fs_devices, device);
2431 * we've got more storage, clear any full flags on the space
2434 btrfs_clear_space_info_full(fs_info);
2436 mutex_unlock(&fs_info->chunk_mutex);
2437 mutex_unlock(&fs_devices->device_list_mutex);
2440 mutex_lock(&fs_info->chunk_mutex);
2441 ret = init_first_rw_device(trans, fs_info);
2442 mutex_unlock(&fs_info->chunk_mutex);
2444 btrfs_abort_transaction(trans, ret);
2449 ret = btrfs_add_dev_item(trans, device);
2451 btrfs_abort_transaction(trans, ret);
2456 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2458 ret = btrfs_finish_sprout(trans, fs_info);
2460 btrfs_abort_transaction(trans, ret);
2464 /* Sprouting would change fsid of the mounted root,
2465 * so rename the fsid on the sysfs
2467 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2469 if (kobject_rename(&fs_devices->fsid_kobj, fsid_buf))
2471 "sysfs: failed to create fsid for sprout");
2474 ret = btrfs_commit_transaction(trans);
2477 mutex_unlock(&uuid_mutex);
2478 up_write(&sb->s_umount);
2481 if (ret) /* transaction commit */
2484 ret = btrfs_relocate_sys_chunks(fs_info);
2486 btrfs_handle_fs_error(fs_info, ret,
2487 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2488 trans = btrfs_attach_transaction(root);
2489 if (IS_ERR(trans)) {
2490 if (PTR_ERR(trans) == -ENOENT)
2492 ret = PTR_ERR(trans);
2496 ret = btrfs_commit_transaction(trans);
2499 /* Update ctime/mtime for libblkid */
2500 update_dev_time(device_path);
2504 btrfs_sysfs_rm_device_link(fs_devices, device);
2507 sb->s_flags |= SB_RDONLY;
2509 btrfs_end_transaction(trans);
2511 btrfs_free_device(device);
2513 blkdev_put(bdev, FMODE_EXCL);
2514 if (seeding_dev && !unlocked) {
2515 mutex_unlock(&uuid_mutex);
2516 up_write(&sb->s_umount);
2521 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2522 struct btrfs_device *device)
2525 struct btrfs_path *path;
2526 struct btrfs_root *root = device->fs_info->chunk_root;
2527 struct btrfs_dev_item *dev_item;
2528 struct extent_buffer *leaf;
2529 struct btrfs_key key;
2531 path = btrfs_alloc_path();
2535 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2536 key.type = BTRFS_DEV_ITEM_KEY;
2537 key.offset = device->devid;
2539 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2548 leaf = path->nodes[0];
2549 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2551 btrfs_set_device_id(leaf, dev_item, device->devid);
2552 btrfs_set_device_type(leaf, dev_item, device->type);
2553 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2554 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2555 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2556 btrfs_set_device_total_bytes(leaf, dev_item,
2557 btrfs_device_get_disk_total_bytes(device));
2558 btrfs_set_device_bytes_used(leaf, dev_item,
2559 btrfs_device_get_bytes_used(device));
2560 btrfs_mark_buffer_dirty(leaf);
2563 btrfs_free_path(path);
2567 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2568 struct btrfs_device *device, u64 new_size)
2570 struct btrfs_fs_info *fs_info = device->fs_info;
2571 struct btrfs_super_block *super_copy = fs_info->super_copy;
2572 struct btrfs_fs_devices *fs_devices;
2576 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2579 new_size = round_down(new_size, fs_info->sectorsize);
2581 mutex_lock(&fs_info->chunk_mutex);
2582 old_total = btrfs_super_total_bytes(super_copy);
2583 diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2585 if (new_size <= device->total_bytes ||
2586 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2587 mutex_unlock(&fs_info->chunk_mutex);
2591 fs_devices = fs_info->fs_devices;
2593 btrfs_set_super_total_bytes(super_copy,
2594 round_down(old_total + diff, fs_info->sectorsize));
2595 device->fs_devices->total_rw_bytes += diff;
2597 btrfs_device_set_total_bytes(device, new_size);
2598 btrfs_device_set_disk_total_bytes(device, new_size);
2599 btrfs_clear_space_info_full(device->fs_info);
2600 if (list_empty(&device->resized_list))
2601 list_add_tail(&device->resized_list,
2602 &fs_devices->resized_devices);
2603 mutex_unlock(&fs_info->chunk_mutex);
2605 return btrfs_update_device(trans, device);
2608 static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
2610 struct btrfs_fs_info *fs_info = trans->fs_info;
2611 struct btrfs_root *root = fs_info->chunk_root;
2613 struct btrfs_path *path;
2614 struct btrfs_key key;
2616 path = btrfs_alloc_path();
2620 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2621 key.offset = chunk_offset;
2622 key.type = BTRFS_CHUNK_ITEM_KEY;
2624 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2627 else if (ret > 0) { /* Logic error or corruption */
2628 btrfs_handle_fs_error(fs_info, -ENOENT,
2629 "Failed lookup while freeing chunk.");
2634 ret = btrfs_del_item(trans, root, path);
2636 btrfs_handle_fs_error(fs_info, ret,
2637 "Failed to delete chunk item.");
2639 btrfs_free_path(path);
2643 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2645 struct btrfs_super_block *super_copy = fs_info->super_copy;
2646 struct btrfs_disk_key *disk_key;
2647 struct btrfs_chunk *chunk;
2654 struct btrfs_key key;
2656 mutex_lock(&fs_info->chunk_mutex);
2657 array_size = btrfs_super_sys_array_size(super_copy);
2659 ptr = super_copy->sys_chunk_array;
2662 while (cur < array_size) {
2663 disk_key = (struct btrfs_disk_key *)ptr;
2664 btrfs_disk_key_to_cpu(&key, disk_key);
2666 len = sizeof(*disk_key);
2668 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2669 chunk = (struct btrfs_chunk *)(ptr + len);
2670 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2671 len += btrfs_chunk_item_size(num_stripes);
2676 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID &&
2677 key.offset == chunk_offset) {
2678 memmove(ptr, ptr + len, array_size - (cur + len));
2680 btrfs_set_super_sys_array_size(super_copy, array_size);
2686 mutex_unlock(&fs_info->chunk_mutex);
2690 static struct extent_map *get_chunk_map(struct btrfs_fs_info *fs_info,
2691 u64 logical, u64 length)
2693 struct extent_map_tree *em_tree;
2694 struct extent_map *em;
2696 em_tree = &fs_info->mapping_tree.map_tree;
2697 read_lock(&em_tree->lock);
2698 em = lookup_extent_mapping(em_tree, logical, length);
2699 read_unlock(&em_tree->lock);
2702 btrfs_crit(fs_info, "unable to find logical %llu length %llu",
2704 return ERR_PTR(-EINVAL);
2707 if (em->start > logical || em->start + em->len < logical) {
2709 "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
2710 logical, length, em->start, em->start + em->len);
2711 free_extent_map(em);
2712 return ERR_PTR(-EINVAL);
2715 /* callers are responsible for dropping em's ref. */
2719 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2720 struct btrfs_fs_info *fs_info, u64 chunk_offset)
2722 struct extent_map *em;
2723 struct map_lookup *map;
2724 u64 dev_extent_len = 0;
2726 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2728 em = get_chunk_map(fs_info, chunk_offset, 1);
2731 * This is a logic error, but we don't want to just rely on the
2732 * user having built with ASSERT enabled, so if ASSERT doesn't
2733 * do anything we still error out.
2738 map = em->map_lookup;
2739 mutex_lock(&fs_info->chunk_mutex);
2740 check_system_chunk(trans, map->type);
2741 mutex_unlock(&fs_info->chunk_mutex);
2744 * Take the device list mutex to prevent races with the final phase of
2745 * a device replace operation that replaces the device object associated
2746 * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2748 mutex_lock(&fs_devices->device_list_mutex);
2749 for (i = 0; i < map->num_stripes; i++) {
2750 struct btrfs_device *device = map->stripes[i].dev;
2751 ret = btrfs_free_dev_extent(trans, device,
2752 map->stripes[i].physical,
2755 mutex_unlock(&fs_devices->device_list_mutex);
2756 btrfs_abort_transaction(trans, ret);
2760 if (device->bytes_used > 0) {
2761 mutex_lock(&fs_info->chunk_mutex);
2762 btrfs_device_set_bytes_used(device,
2763 device->bytes_used - dev_extent_len);
2764 atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
2765 btrfs_clear_space_info_full(fs_info);
2766 mutex_unlock(&fs_info->chunk_mutex);
2769 if (map->stripes[i].dev) {
2770 ret = btrfs_update_device(trans, map->stripes[i].dev);
2772 mutex_unlock(&fs_devices->device_list_mutex);
2773 btrfs_abort_transaction(trans, ret);
2778 mutex_unlock(&fs_devices->device_list_mutex);
2780 ret = btrfs_free_chunk(trans, chunk_offset);
2782 btrfs_abort_transaction(trans, ret);
2786 trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
2788 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2789 ret = btrfs_del_sys_chunk(fs_info, chunk_offset);
2791 btrfs_abort_transaction(trans, ret);
2796 ret = btrfs_remove_block_group(trans, chunk_offset, em);
2798 btrfs_abort_transaction(trans, ret);
2804 free_extent_map(em);
2808 static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2810 struct btrfs_root *root = fs_info->chunk_root;
2811 struct btrfs_trans_handle *trans;
2815 * Prevent races with automatic removal of unused block groups.
2816 * After we relocate and before we remove the chunk with offset
2817 * chunk_offset, automatic removal of the block group can kick in,
2818 * resulting in a failure when calling btrfs_remove_chunk() below.
2820 * Make sure to acquire this mutex before doing a tree search (dev
2821 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2822 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2823 * we release the path used to search the chunk/dev tree and before
2824 * the current task acquires this mutex and calls us.
2826 lockdep_assert_held(&fs_info->delete_unused_bgs_mutex);
2828 ret = btrfs_can_relocate(fs_info, chunk_offset);
2832 /* step one, relocate all the extents inside this chunk */
2833 btrfs_scrub_pause(fs_info);
2834 ret = btrfs_relocate_block_group(fs_info, chunk_offset);
2835 btrfs_scrub_continue(fs_info);
2840 * We add the kobjects here (and after forcing data chunk creation)
2841 * since relocation is the only place we'll create chunks of a new
2842 * type at runtime. The only place where we'll remove the last
2843 * chunk of a type is the call immediately below this one. Even
2844 * so, we're protected against races with the cleaner thread since
2845 * we're covered by the delete_unused_bgs_mutex.
2847 btrfs_add_raid_kobjects(fs_info);
2849 trans = btrfs_start_trans_remove_block_group(root->fs_info,
2851 if (IS_ERR(trans)) {
2852 ret = PTR_ERR(trans);
2853 btrfs_handle_fs_error(root->fs_info, ret, NULL);
2858 * step two, delete the device extents and the
2859 * chunk tree entries
2861 ret = btrfs_remove_chunk(trans, fs_info, chunk_offset);
2862 btrfs_end_transaction(trans);
2866 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
2868 struct btrfs_root *chunk_root = fs_info->chunk_root;
2869 struct btrfs_path *path;
2870 struct extent_buffer *leaf;
2871 struct btrfs_chunk *chunk;
2872 struct btrfs_key key;
2873 struct btrfs_key found_key;
2875 bool retried = false;
2879 path = btrfs_alloc_path();
2884 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2885 key.offset = (u64)-1;
2886 key.type = BTRFS_CHUNK_ITEM_KEY;
2889 mutex_lock(&fs_info->delete_unused_bgs_mutex);
2890 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2892 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
2895 BUG_ON(ret == 0); /* Corruption */
2897 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2900 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
2906 leaf = path->nodes[0];
2907 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2909 chunk = btrfs_item_ptr(leaf, path->slots[0],
2910 struct btrfs_chunk);
2911 chunk_type = btrfs_chunk_type(leaf, chunk);
2912 btrfs_release_path(path);
2914 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2915 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
2921 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
2923 if (found_key.offset == 0)
2925 key.offset = found_key.offset - 1;
2928 if (failed && !retried) {
2932 } else if (WARN_ON(failed && retried)) {
2936 btrfs_free_path(path);
2941 * return 1 : allocate a data chunk successfully,
2942 * return <0: errors during allocating a data chunk,
2943 * return 0 : no need to allocate a data chunk.
2945 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info,
2948 struct btrfs_block_group_cache *cache;
2952 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
2954 chunk_type = cache->flags;
2955 btrfs_put_block_group(cache);
2957 if (chunk_type & BTRFS_BLOCK_GROUP_DATA) {
2958 spin_lock(&fs_info->data_sinfo->lock);
2959 bytes_used = fs_info->data_sinfo->bytes_used;
2960 spin_unlock(&fs_info->data_sinfo->lock);
2963 struct btrfs_trans_handle *trans;
2966 trans = btrfs_join_transaction(fs_info->tree_root);
2968 return PTR_ERR(trans);
2970 ret = btrfs_force_chunk_alloc(trans,
2971 BTRFS_BLOCK_GROUP_DATA);
2972 btrfs_end_transaction(trans);
2976 btrfs_add_raid_kobjects(fs_info);
2984 static int insert_balance_item(struct btrfs_fs_info *fs_info,
2985 struct btrfs_balance_control *bctl)
2987 struct btrfs_root *root = fs_info->tree_root;
2988 struct btrfs_trans_handle *trans;
2989 struct btrfs_balance_item *item;
2990 struct btrfs_disk_balance_args disk_bargs;
2991 struct btrfs_path *path;
2992 struct extent_buffer *leaf;
2993 struct btrfs_key key;
2996 path = btrfs_alloc_path();
3000 trans = btrfs_start_transaction(root, 0);
3001 if (IS_ERR(trans)) {
3002 btrfs_free_path(path);
3003 return PTR_ERR(trans);
3006 key.objectid = BTRFS_BALANCE_OBJECTID;
3007 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3010 ret = btrfs_insert_empty_item(trans, root, path, &key,
3015 leaf = path->nodes[0];
3016 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3018 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3020 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3021 btrfs_set_balance_data(leaf, item, &disk_bargs);
3022 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3023 btrfs_set_balance_meta(leaf, item, &disk_bargs);
3024 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3025 btrfs_set_balance_sys(leaf, item, &disk_bargs);
3027 btrfs_set_balance_flags(leaf, item, bctl->flags);
3029 btrfs_mark_buffer_dirty(leaf);
3031 btrfs_free_path(path);
3032 err = btrfs_commit_transaction(trans);
3038 static int del_balance_item(struct btrfs_fs_info *fs_info)
3040 struct btrfs_root *root = fs_info->tree_root;
3041 struct btrfs_trans_handle *trans;
3042 struct btrfs_path *path;
3043 struct btrfs_key key;
3046 path = btrfs_alloc_path();
3050 trans = btrfs_start_transaction(root, 0);
3051 if (IS_ERR(trans)) {
3052 btrfs_free_path(path);
3053 return PTR_ERR(trans);
3056 key.objectid = BTRFS_BALANCE_OBJECTID;
3057 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3060 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3068 ret = btrfs_del_item(trans, root, path);
3070 btrfs_free_path(path);
3071 err = btrfs_commit_transaction(trans);
3078 * This is a heuristic used to reduce the number of chunks balanced on
3079 * resume after balance was interrupted.
3081 static void update_balance_args(struct btrfs_balance_control *bctl)
3084 * Turn on soft mode for chunk types that were being converted.
3086 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3087 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3088 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3089 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3090 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3091 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3094 * Turn on usage filter if is not already used. The idea is
3095 * that chunks that we have already balanced should be
3096 * reasonably full. Don't do it for chunks that are being
3097 * converted - that will keep us from relocating unconverted
3098 * (albeit full) chunks.
3100 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3101 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3102 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3103 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3104 bctl->data.usage = 90;
3106 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3107 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3108 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3109 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3110 bctl->sys.usage = 90;
3112 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3113 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3114 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3115 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3116 bctl->meta.usage = 90;
3121 * Clear the balance status in fs_info and delete the balance item from disk.
3123 static void reset_balance_state(struct btrfs_fs_info *fs_info)
3125 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3128 BUG_ON(!fs_info->balance_ctl);
3130 spin_lock(&fs_info->balance_lock);
3131 fs_info->balance_ctl = NULL;
3132 spin_unlock(&fs_info->balance_lock);
3135 ret = del_balance_item(fs_info);
3137 btrfs_handle_fs_error(fs_info, ret, NULL);
3141 * Balance filters. Return 1 if chunk should be filtered out
3142 * (should not be balanced).
3144 static int chunk_profiles_filter(u64 chunk_type,
3145 struct btrfs_balance_args *bargs)
3147 chunk_type = chunk_to_extended(chunk_type) &
3148 BTRFS_EXTENDED_PROFILE_MASK;
3150 if (bargs->profiles & chunk_type)
3156 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3157 struct btrfs_balance_args *bargs)
3159 struct btrfs_block_group_cache *cache;
3161 u64 user_thresh_min;
3162 u64 user_thresh_max;
3165 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3166 chunk_used = btrfs_block_group_used(&cache->item);
3168 if (bargs->usage_min == 0)
3169 user_thresh_min = 0;
3171 user_thresh_min = div_factor_fine(cache->key.offset,
3174 if (bargs->usage_max == 0)
3175 user_thresh_max = 1;
3176 else if (bargs->usage_max > 100)
3177 user_thresh_max = cache->key.offset;
3179 user_thresh_max = div_factor_fine(cache->key.offset,
3182 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3185 btrfs_put_block_group(cache);
3189 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3190 u64 chunk_offset, struct btrfs_balance_args *bargs)
3192 struct btrfs_block_group_cache *cache;
3193 u64 chunk_used, user_thresh;
3196 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3197 chunk_used = btrfs_block_group_used(&cache->item);
3199 if (bargs->usage_min == 0)
3201 else if (bargs->usage > 100)
3202 user_thresh = cache->key.offset;
3204 user_thresh = div_factor_fine(cache->key.offset,
3207 if (chunk_used < user_thresh)
3210 btrfs_put_block_group(cache);
3214 static int chunk_devid_filter(struct extent_buffer *leaf,
3215 struct btrfs_chunk *chunk,
3216 struct btrfs_balance_args *bargs)
3218 struct btrfs_stripe *stripe;
3219 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3222 for (i = 0; i < num_stripes; i++) {
3223 stripe = btrfs_stripe_nr(chunk, i);
3224 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3231 /* [pstart, pend) */
3232 static int chunk_drange_filter(struct extent_buffer *leaf,
3233 struct btrfs_chunk *chunk,
3234 struct btrfs_balance_args *bargs)
3236 struct btrfs_stripe *stripe;
3237 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3243 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3246 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3247 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3248 factor = num_stripes / 2;
3249 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3250 factor = num_stripes - 1;
3251 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3252 factor = num_stripes - 2;
3254 factor = num_stripes;
3257 for (i = 0; i < num_stripes; i++) {
3258 stripe = btrfs_stripe_nr(chunk, i);
3259 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3262 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3263 stripe_length = btrfs_chunk_length(leaf, chunk);
3264 stripe_length = div_u64(stripe_length, factor);
3266 if (stripe_offset < bargs->pend &&
3267 stripe_offset + stripe_length > bargs->pstart)
3274 /* [vstart, vend) */
3275 static int chunk_vrange_filter(struct extent_buffer *leaf,
3276 struct btrfs_chunk *chunk,
3278 struct btrfs_balance_args *bargs)
3280 if (chunk_offset < bargs->vend &&
3281 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3282 /* at least part of the chunk is inside this vrange */
3288 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3289 struct btrfs_chunk *chunk,
3290 struct btrfs_balance_args *bargs)
3292 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3294 if (bargs->stripes_min <= num_stripes
3295 && num_stripes <= bargs->stripes_max)
3301 static int chunk_soft_convert_filter(u64 chunk_type,
3302 struct btrfs_balance_args *bargs)
3304 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3307 chunk_type = chunk_to_extended(chunk_type) &
3308 BTRFS_EXTENDED_PROFILE_MASK;
3310 if (bargs->target == chunk_type)
3316 static int should_balance_chunk(struct btrfs_fs_info *fs_info,
3317 struct extent_buffer *leaf,
3318 struct btrfs_chunk *chunk, u64 chunk_offset)
3320 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3321 struct btrfs_balance_args *bargs = NULL;
3322 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3325 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3326 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3330 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3331 bargs = &bctl->data;
3332 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3334 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3335 bargs = &bctl->meta;
3337 /* profiles filter */
3338 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3339 chunk_profiles_filter(chunk_type, bargs)) {
3344 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3345 chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3347 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3348 chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3353 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3354 chunk_devid_filter(leaf, chunk, bargs)) {
3358 /* drange filter, makes sense only with devid filter */
3359 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3360 chunk_drange_filter(leaf, chunk, bargs)) {
3365 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3366 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3370 /* stripes filter */
3371 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3372 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3376 /* soft profile changing mode */
3377 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3378 chunk_soft_convert_filter(chunk_type, bargs)) {
3383 * limited by count, must be the last filter
3385 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3386 if (bargs->limit == 0)
3390 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3392 * Same logic as the 'limit' filter; the minimum cannot be
3393 * determined here because we do not have the global information
3394 * about the count of all chunks that satisfy the filters.
3396 if (bargs->limit_max == 0)
3405 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3407 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3408 struct btrfs_root *chunk_root = fs_info->chunk_root;
3409 struct btrfs_root *dev_root = fs_info->dev_root;
3410 struct list_head *devices;
3411 struct btrfs_device *device;
3415 struct btrfs_chunk *chunk;
3416 struct btrfs_path *path = NULL;
3417 struct btrfs_key key;
3418 struct btrfs_key found_key;
3419 struct btrfs_trans_handle *trans;
3420 struct extent_buffer *leaf;
3423 int enospc_errors = 0;
3424 bool counting = true;
3425 /* The single value limit and min/max limits use the same bytes in the */
3426 u64 limit_data = bctl->data.limit;
3427 u64 limit_meta = bctl->meta.limit;
3428 u64 limit_sys = bctl->sys.limit;
3432 int chunk_reserved = 0;
3434 /* step one make some room on all the devices */
3435 devices = &fs_info->fs_devices->devices;
3436 list_for_each_entry(device, devices, dev_list) {
3437 old_size = btrfs_device_get_total_bytes(device);
3438 size_to_free = div_factor(old_size, 1);
3439 size_to_free = min_t(u64, size_to_free, SZ_1M);
3440 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) ||
3441 btrfs_device_get_total_bytes(device) -
3442 btrfs_device_get_bytes_used(device) > size_to_free ||
3443 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
3446 ret = btrfs_shrink_device(device, old_size - size_to_free);
3450 /* btrfs_shrink_device never returns ret > 0 */
3455 trans = btrfs_start_transaction(dev_root, 0);
3456 if (IS_ERR(trans)) {
3457 ret = PTR_ERR(trans);
3458 btrfs_info_in_rcu(fs_info,
3459 "resize: unable to start transaction after shrinking device %s (error %d), old size %llu, new size %llu",
3460 rcu_str_deref(device->name), ret,
3461 old_size, old_size - size_to_free);
3465 ret = btrfs_grow_device(trans, device, old_size);
3467 btrfs_end_transaction(trans);
3468 /* btrfs_grow_device never returns ret > 0 */
3470 btrfs_info_in_rcu(fs_info,
3471 "resize: unable to grow device after shrinking device %s (error %d), old size %llu, new size %llu",
3472 rcu_str_deref(device->name), ret,
3473 old_size, old_size - size_to_free);
3477 btrfs_end_transaction(trans);
3480 /* step two, relocate all the chunks */
3481 path = btrfs_alloc_path();
3487 /* zero out stat counters */
3488 spin_lock(&fs_info->balance_lock);
3489 memset(&bctl->stat, 0, sizeof(bctl->stat));
3490 spin_unlock(&fs_info->balance_lock);
3494 * The single value limit and min/max limits use the same bytes
3497 bctl->data.limit = limit_data;
3498 bctl->meta.limit = limit_meta;
3499 bctl->sys.limit = limit_sys;
3501 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3502 key.offset = (u64)-1;
3503 key.type = BTRFS_CHUNK_ITEM_KEY;
3506 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3507 atomic_read(&fs_info->balance_cancel_req)) {
3512 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3513 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3515 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3520 * this shouldn't happen, it means the last relocate
3524 BUG(); /* FIXME break ? */
3526 ret = btrfs_previous_item(chunk_root, path, 0,
3527 BTRFS_CHUNK_ITEM_KEY);
3529 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3534 leaf = path->nodes[0];
3535 slot = path->slots[0];
3536 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3538 if (found_key.objectid != key.objectid) {
3539 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3543 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3544 chunk_type = btrfs_chunk_type(leaf, chunk);
3547 spin_lock(&fs_info->balance_lock);
3548 bctl->stat.considered++;
3549 spin_unlock(&fs_info->balance_lock);
3552 ret = should_balance_chunk(fs_info, leaf, chunk,
3555 btrfs_release_path(path);
3557 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3562 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3563 spin_lock(&fs_info->balance_lock);
3564 bctl->stat.expected++;
3565 spin_unlock(&fs_info->balance_lock);
3567 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3569 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3571 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3578 * Apply limit_min filter, no need to check if the LIMITS
3579 * filter is used, limit_min is 0 by default
3581 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3582 count_data < bctl->data.limit_min)
3583 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3584 count_meta < bctl->meta.limit_min)
3585 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3586 count_sys < bctl->sys.limit_min)) {
3587 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3591 if (!chunk_reserved) {
3593 * We may be relocating the only data chunk we have,
3594 * which could potentially end up with losing data's
3595 * raid profile, so lets allocate an empty one in
3598 ret = btrfs_may_alloc_data_chunk(fs_info,
3601 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3603 } else if (ret == 1) {
3608 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3609 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3610 if (ret && ret != -ENOSPC)
3612 if (ret == -ENOSPC) {
3615 spin_lock(&fs_info->balance_lock);
3616 bctl->stat.completed++;
3617 spin_unlock(&fs_info->balance_lock);
3620 if (found_key.offset == 0)
3622 key.offset = found_key.offset - 1;
3626 btrfs_release_path(path);
3631 btrfs_free_path(path);
3632 if (enospc_errors) {
3633 btrfs_info(fs_info, "%d enospc errors during balance",
3643 * alloc_profile_is_valid - see if a given profile is valid and reduced
3644 * @flags: profile to validate
3645 * @extended: if true @flags is treated as an extended profile
3647 static int alloc_profile_is_valid(u64 flags, int extended)
3649 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3650 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3652 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3654 /* 1) check that all other bits are zeroed */
3658 /* 2) see if profile is reduced */
3660 return !extended; /* "0" is valid for usual profiles */
3662 /* true if exactly one bit set */
3663 return (flags & (flags - 1)) == 0;
3666 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3668 /* cancel requested || normal exit path */
3669 return atomic_read(&fs_info->balance_cancel_req) ||
3670 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3671 atomic_read(&fs_info->balance_cancel_req) == 0);
3674 /* Non-zero return value signifies invalidity */
3675 static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg,
3678 return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3679 (!alloc_profile_is_valid(bctl_arg->target, 1) ||
3680 (bctl_arg->target & ~allowed)));
3684 * Should be called with balance mutexe held
3686 int btrfs_balance(struct btrfs_fs_info *fs_info,
3687 struct btrfs_balance_control *bctl,
3688 struct btrfs_ioctl_balance_args *bargs)
3690 u64 meta_target, data_target;
3697 if (btrfs_fs_closing(fs_info) ||
3698 atomic_read(&fs_info->balance_pause_req) ||
3699 atomic_read(&fs_info->balance_cancel_req)) {
3704 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3705 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3709 * In case of mixed groups both data and meta should be picked,
3710 * and identical options should be given for both of them.
3712 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3713 if (mixed && (bctl->flags & allowed)) {
3714 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3715 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3716 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3718 "balance: mixed groups data and metadata options must be the same");
3724 num_devices = fs_info->fs_devices->num_devices;
3725 btrfs_dev_replace_read_lock(&fs_info->dev_replace);
3726 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3727 BUG_ON(num_devices < 1);
3730 btrfs_dev_replace_read_unlock(&fs_info->dev_replace);
3731 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE | BTRFS_BLOCK_GROUP_DUP;
3732 if (num_devices > 1)
3733 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3734 if (num_devices > 2)
3735 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3736 if (num_devices > 3)
3737 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3738 BTRFS_BLOCK_GROUP_RAID6);
3739 if (validate_convert_profile(&bctl->data, allowed)) {
3740 int index = btrfs_bg_flags_to_raid_index(bctl->data.target);
3743 "balance: invalid convert data profile %s",
3744 get_raid_name(index));
3748 if (validate_convert_profile(&bctl->meta, allowed)) {
3749 int index = btrfs_bg_flags_to_raid_index(bctl->meta.target);
3752 "balance: invalid convert metadata profile %s",
3753 get_raid_name(index));
3757 if (validate_convert_profile(&bctl->sys, allowed)) {
3758 int index = btrfs_bg_flags_to_raid_index(bctl->sys.target);
3761 "balance: invalid convert system profile %s",
3762 get_raid_name(index));
3767 /* allow to reduce meta or sys integrity only if force set */
3768 allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3769 BTRFS_BLOCK_GROUP_RAID10 |
3770 BTRFS_BLOCK_GROUP_RAID5 |
3771 BTRFS_BLOCK_GROUP_RAID6;
3773 seq = read_seqbegin(&fs_info->profiles_lock);
3775 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3776 (fs_info->avail_system_alloc_bits & allowed) &&
3777 !(bctl->sys.target & allowed)) ||
3778 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3779 (fs_info->avail_metadata_alloc_bits & allowed) &&
3780 !(bctl->meta.target & allowed))) {
3781 if (bctl->flags & BTRFS_BALANCE_FORCE) {
3783 "balance: force reducing metadata integrity");
3786 "balance: reduces metadata integrity, use --force if you want this");
3791 } while (read_seqretry(&fs_info->profiles_lock, seq));
3793 /* if we're not converting, the target field is uninitialized */
3794 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
3795 bctl->meta.target : fs_info->avail_metadata_alloc_bits;
3796 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
3797 bctl->data.target : fs_info->avail_data_alloc_bits;
3798 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <
3799 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {
3800 int meta_index = btrfs_bg_flags_to_raid_index(meta_target);
3801 int data_index = btrfs_bg_flags_to_raid_index(data_target);
3804 "balance: metadata profile %s has lower redundancy than data profile %s",
3805 get_raid_name(meta_index), get_raid_name(data_index));
3808 ret = insert_balance_item(fs_info, bctl);
3809 if (ret && ret != -EEXIST)
3812 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3813 BUG_ON(ret == -EEXIST);
3814 BUG_ON(fs_info->balance_ctl);
3815 spin_lock(&fs_info->balance_lock);
3816 fs_info->balance_ctl = bctl;
3817 spin_unlock(&fs_info->balance_lock);
3819 BUG_ON(ret != -EEXIST);
3820 spin_lock(&fs_info->balance_lock);
3821 update_balance_args(bctl);
3822 spin_unlock(&fs_info->balance_lock);
3825 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
3826 set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
3827 mutex_unlock(&fs_info->balance_mutex);
3829 ret = __btrfs_balance(fs_info);
3831 mutex_lock(&fs_info->balance_mutex);
3832 clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
3835 memset(bargs, 0, sizeof(*bargs));
3836 btrfs_update_ioctl_balance_args(fs_info, bargs);
3839 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3840 balance_need_close(fs_info)) {
3841 reset_balance_state(fs_info);
3842 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3845 wake_up(&fs_info->balance_wait_q);
3849 if (bctl->flags & BTRFS_BALANCE_RESUME)
3850 reset_balance_state(fs_info);
3853 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3858 static int balance_kthread(void *data)
3860 struct btrfs_fs_info *fs_info = data;
3863 mutex_lock(&fs_info->balance_mutex);
3864 if (fs_info->balance_ctl) {
3865 btrfs_info(fs_info, "balance: resuming");
3866 ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL);
3868 mutex_unlock(&fs_info->balance_mutex);
3873 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3875 struct task_struct *tsk;
3877 mutex_lock(&fs_info->balance_mutex);
3878 if (!fs_info->balance_ctl) {
3879 mutex_unlock(&fs_info->balance_mutex);
3882 mutex_unlock(&fs_info->balance_mutex);
3884 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
3885 btrfs_info(fs_info, "balance: resume skipped");
3890 * A ro->rw remount sequence should continue with the paused balance
3891 * regardless of who pauses it, system or the user as of now, so set
3894 spin_lock(&fs_info->balance_lock);
3895 fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME;
3896 spin_unlock(&fs_info->balance_lock);
3898 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3899 return PTR_ERR_OR_ZERO(tsk);
3902 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3904 struct btrfs_balance_control *bctl;
3905 struct btrfs_balance_item *item;
3906 struct btrfs_disk_balance_args disk_bargs;
3907 struct btrfs_path *path;
3908 struct extent_buffer *leaf;
3909 struct btrfs_key key;
3912 path = btrfs_alloc_path();
3916 key.objectid = BTRFS_BALANCE_OBJECTID;
3917 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3920 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3923 if (ret > 0) { /* ret = -ENOENT; */
3928 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3934 leaf = path->nodes[0];
3935 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3937 bctl->flags = btrfs_balance_flags(leaf, item);
3938 bctl->flags |= BTRFS_BALANCE_RESUME;
3940 btrfs_balance_data(leaf, item, &disk_bargs);
3941 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3942 btrfs_balance_meta(leaf, item, &disk_bargs);
3943 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3944 btrfs_balance_sys(leaf, item, &disk_bargs);
3945 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3948 * This should never happen, as the paused balance state is recovered
3949 * during mount without any chance of other exclusive ops to collide.
3951 * This gives the exclusive op status to balance and keeps in paused
3952 * state until user intervention (cancel or umount). If the ownership
3953 * cannot be assigned, show a message but do not fail. The balance
3954 * is in a paused state and must have fs_info::balance_ctl properly
3957 if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags))
3959 "balance: cannot set exclusive op status, resume manually");
3961 mutex_lock(&fs_info->balance_mutex);
3962 BUG_ON(fs_info->balance_ctl);
3963 spin_lock(&fs_info->balance_lock);
3964 fs_info->balance_ctl = bctl;
3965 spin_unlock(&fs_info->balance_lock);
3966 mutex_unlock(&fs_info->balance_mutex);
3968 btrfs_free_path(path);
3972 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3976 mutex_lock(&fs_info->balance_mutex);
3977 if (!fs_info->balance_ctl) {
3978 mutex_unlock(&fs_info->balance_mutex);
3982 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
3983 atomic_inc(&fs_info->balance_pause_req);
3984 mutex_unlock(&fs_info->balance_mutex);
3986 wait_event(fs_info->balance_wait_q,
3987 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
3989 mutex_lock(&fs_info->balance_mutex);
3990 /* we are good with balance_ctl ripped off from under us */
3991 BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
3992 atomic_dec(&fs_info->balance_pause_req);
3997 mutex_unlock(&fs_info->balance_mutex);
4001 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4003 mutex_lock(&fs_info->balance_mutex);
4004 if (!fs_info->balance_ctl) {
4005 mutex_unlock(&fs_info->balance_mutex);
4010 * A paused balance with the item stored on disk can be resumed at
4011 * mount time if the mount is read-write. Otherwise it's still paused
4012 * and we must not allow cancelling as it deletes the item.
4014 if (sb_rdonly(fs_info->sb)) {
4015 mutex_unlock(&fs_info->balance_mutex);
4019 atomic_inc(&fs_info->balance_cancel_req);
4021 * if we are running just wait and return, balance item is
4022 * deleted in btrfs_balance in this case
4024 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4025 mutex_unlock(&fs_info->balance_mutex);
4026 wait_event(fs_info->balance_wait_q,
4027 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4028 mutex_lock(&fs_info->balance_mutex);
4030 mutex_unlock(&fs_info->balance_mutex);
4032 * Lock released to allow other waiters to continue, we'll
4033 * reexamine the status again.
4035 mutex_lock(&fs_info->balance_mutex);
4037 if (fs_info->balance_ctl) {
4038 reset_balance_state(fs_info);
4039 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4040 btrfs_info(fs_info, "balance: canceled");
4044 BUG_ON(fs_info->balance_ctl ||
4045 test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4046 atomic_dec(&fs_info->balance_cancel_req);
4047 mutex_unlock(&fs_info->balance_mutex);
4051 static int btrfs_uuid_scan_kthread(void *data)
4053 struct btrfs_fs_info *fs_info = data;
4054 struct btrfs_root *root = fs_info->tree_root;
4055 struct btrfs_key key;
4056 struct btrfs_path *path = NULL;
4058 struct extent_buffer *eb;
4060 struct btrfs_root_item root_item;
4062 struct btrfs_trans_handle *trans = NULL;
4064 path = btrfs_alloc_path();
4071 key.type = BTRFS_ROOT_ITEM_KEY;
4075 ret = btrfs_search_forward(root, &key, path,
4076 BTRFS_OLDEST_GENERATION);
4083 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4084 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4085 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4086 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4089 eb = path->nodes[0];
4090 slot = path->slots[0];
4091 item_size = btrfs_item_size_nr(eb, slot);
4092 if (item_size < sizeof(root_item))
4095 read_extent_buffer(eb, &root_item,
4096 btrfs_item_ptr_offset(eb, slot),
4097 (int)sizeof(root_item));
4098 if (btrfs_root_refs(&root_item) == 0)
4101 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4102 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4106 btrfs_release_path(path);
4108 * 1 - subvol uuid item
4109 * 1 - received_subvol uuid item
4111 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4112 if (IS_ERR(trans)) {
4113 ret = PTR_ERR(trans);
4121 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4122 ret = btrfs_uuid_tree_add(trans, root_item.uuid,
4123 BTRFS_UUID_KEY_SUBVOL,
4126 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4132 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4133 ret = btrfs_uuid_tree_add(trans,
4134 root_item.received_uuid,
4135 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4138 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4146 ret = btrfs_end_transaction(trans);
4152 btrfs_release_path(path);
4153 if (key.offset < (u64)-1) {
4155 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4157 key.type = BTRFS_ROOT_ITEM_KEY;
4158 } else if (key.objectid < (u64)-1) {
4160 key.type = BTRFS_ROOT_ITEM_KEY;
4169 btrfs_free_path(path);
4170 if (trans && !IS_ERR(trans))
4171 btrfs_end_transaction(trans);
4173 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4175 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4176 up(&fs_info->uuid_tree_rescan_sem);
4181 * Callback for btrfs_uuid_tree_iterate().
4183 * 0 check succeeded, the entry is not outdated.
4184 * < 0 if an error occurred.
4185 * > 0 if the check failed, which means the caller shall remove the entry.
4187 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
4188 u8 *uuid, u8 type, u64 subid)
4190 struct btrfs_key key;
4192 struct btrfs_root *subvol_root;
4194 if (type != BTRFS_UUID_KEY_SUBVOL &&
4195 type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
4198 key.objectid = subid;
4199 key.type = BTRFS_ROOT_ITEM_KEY;
4200 key.offset = (u64)-1;
4201 subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
4202 if (IS_ERR(subvol_root)) {
4203 ret = PTR_ERR(subvol_root);
4210 case BTRFS_UUID_KEY_SUBVOL:
4211 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
4214 case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
4215 if (memcmp(uuid, subvol_root->root_item.received_uuid,
4225 static int btrfs_uuid_rescan_kthread(void *data)
4227 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4231 * 1st step is to iterate through the existing UUID tree and
4232 * to delete all entries that contain outdated data.
4233 * 2nd step is to add all missing entries to the UUID tree.
4235 ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4237 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4238 up(&fs_info->uuid_tree_rescan_sem);
4241 return btrfs_uuid_scan_kthread(data);
4244 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4246 struct btrfs_trans_handle *trans;
4247 struct btrfs_root *tree_root = fs_info->tree_root;
4248 struct btrfs_root *uuid_root;
4249 struct task_struct *task;
4256 trans = btrfs_start_transaction(tree_root, 2);
4258 return PTR_ERR(trans);
4260 uuid_root = btrfs_create_tree(trans, fs_info,
4261 BTRFS_UUID_TREE_OBJECTID);
4262 if (IS_ERR(uuid_root)) {
4263 ret = PTR_ERR(uuid_root);
4264 btrfs_abort_transaction(trans, ret);
4265 btrfs_end_transaction(trans);
4269 fs_info->uuid_root = uuid_root;
4271 ret = btrfs_commit_transaction(trans);
4275 down(&fs_info->uuid_tree_rescan_sem);
4276 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4278 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4279 btrfs_warn(fs_info, "failed to start uuid_scan task");
4280 up(&fs_info->uuid_tree_rescan_sem);
4281 return PTR_ERR(task);
4287 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4289 struct task_struct *task;
4291 down(&fs_info->uuid_tree_rescan_sem);
4292 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4294 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4295 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4296 up(&fs_info->uuid_tree_rescan_sem);
4297 return PTR_ERR(task);
4304 * shrinking a device means finding all of the device extents past
4305 * the new size, and then following the back refs to the chunks.
4306 * The chunk relocation code actually frees the device extent
4308 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4310 struct btrfs_fs_info *fs_info = device->fs_info;
4311 struct btrfs_root *root = fs_info->dev_root;
4312 struct btrfs_trans_handle *trans;
4313 struct btrfs_dev_extent *dev_extent = NULL;
4314 struct btrfs_path *path;
4320 bool retried = false;
4321 bool checked_pending_chunks = false;
4322 struct extent_buffer *l;
4323 struct btrfs_key key;
4324 struct btrfs_super_block *super_copy = fs_info->super_copy;
4325 u64 old_total = btrfs_super_total_bytes(super_copy);
4326 u64 old_size = btrfs_device_get_total_bytes(device);
4329 new_size = round_down(new_size, fs_info->sectorsize);
4330 diff = round_down(old_size - new_size, fs_info->sectorsize);
4332 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4335 path = btrfs_alloc_path();
4339 path->reada = READA_BACK;
4341 mutex_lock(&fs_info->chunk_mutex);
4343 btrfs_device_set_total_bytes(device, new_size);
4344 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4345 device->fs_devices->total_rw_bytes -= diff;
4346 atomic64_sub(diff, &fs_info->free_chunk_space);
4348 mutex_unlock(&fs_info->chunk_mutex);
4351 key.objectid = device->devid;
4352 key.offset = (u64)-1;
4353 key.type = BTRFS_DEV_EXTENT_KEY;
4356 mutex_lock(&fs_info->delete_unused_bgs_mutex);
4357 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4359 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4363 ret = btrfs_previous_item(root, path, 0, key.type);
4365 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4370 btrfs_release_path(path);
4375 slot = path->slots[0];
4376 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4378 if (key.objectid != device->devid) {
4379 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4380 btrfs_release_path(path);
4384 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4385 length = btrfs_dev_extent_length(l, dev_extent);
4387 if (key.offset + length <= new_size) {
4388 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4389 btrfs_release_path(path);
4393 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4394 btrfs_release_path(path);
4397 * We may be relocating the only data chunk we have,
4398 * which could potentially end up with losing data's
4399 * raid profile, so lets allocate an empty one in
4402 ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset);
4404 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4408 ret = btrfs_relocate_chunk(fs_info, chunk_offset);
4409 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4410 if (ret && ret != -ENOSPC)
4414 } while (key.offset-- > 0);
4416 if (failed && !retried) {
4420 } else if (failed && retried) {
4425 /* Shrinking succeeded, else we would be at "done". */
4426 trans = btrfs_start_transaction(root, 0);
4427 if (IS_ERR(trans)) {
4428 ret = PTR_ERR(trans);
4432 mutex_lock(&fs_info->chunk_mutex);
4435 * We checked in the above loop all device extents that were already in
4436 * the device tree. However before we have updated the device's
4437 * total_bytes to the new size, we might have had chunk allocations that
4438 * have not complete yet (new block groups attached to transaction
4439 * handles), and therefore their device extents were not yet in the
4440 * device tree and we missed them in the loop above. So if we have any
4441 * pending chunk using a device extent that overlaps the device range
4442 * that we can not use anymore, commit the current transaction and
4443 * repeat the search on the device tree - this way we guarantee we will
4444 * not have chunks using device extents that end beyond 'new_size'.
4446 if (!checked_pending_chunks) {
4447 u64 start = new_size;
4448 u64 len = old_size - new_size;
4450 if (contains_pending_extent(trans->transaction, device,
4452 mutex_unlock(&fs_info->chunk_mutex);
4453 checked_pending_chunks = true;
4456 ret = btrfs_commit_transaction(trans);
4463 btrfs_device_set_disk_total_bytes(device, new_size);
4464 if (list_empty(&device->resized_list))
4465 list_add_tail(&device->resized_list,
4466 &fs_info->fs_devices->resized_devices);
4468 WARN_ON(diff > old_total);
4469 btrfs_set_super_total_bytes(super_copy,
4470 round_down(old_total - diff, fs_info->sectorsize));
4471 mutex_unlock(&fs_info->chunk_mutex);
4473 /* Now btrfs_update_device() will change the on-disk size. */
4474 ret = btrfs_update_device(trans, device);
4475 btrfs_end_transaction(trans);
4477 btrfs_free_path(path);
4479 mutex_lock(&fs_info->chunk_mutex);
4480 btrfs_device_set_total_bytes(device, old_size);
4481 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
4482 device->fs_devices->total_rw_bytes += diff;
4483 atomic64_add(diff, &fs_info->free_chunk_space);
4484 mutex_unlock(&fs_info->chunk_mutex);
4489 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
4490 struct btrfs_key *key,
4491 struct btrfs_chunk *chunk, int item_size)
4493 struct btrfs_super_block *super_copy = fs_info->super_copy;
4494 struct btrfs_disk_key disk_key;
4498 mutex_lock(&fs_info->chunk_mutex);
4499 array_size = btrfs_super_sys_array_size(super_copy);
4500 if (array_size + item_size + sizeof(disk_key)
4501 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4502 mutex_unlock(&fs_info->chunk_mutex);
4506 ptr = super_copy->sys_chunk_array + array_size;
4507 btrfs_cpu_key_to_disk(&disk_key, key);
4508 memcpy(ptr, &disk_key, sizeof(disk_key));
4509 ptr += sizeof(disk_key);
4510 memcpy(ptr, chunk, item_size);
4511 item_size += sizeof(disk_key);
4512 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4513 mutex_unlock(&fs_info->chunk_mutex);
4519 * sort the devices in descending order by max_avail, total_avail
4521 static int btrfs_cmp_device_info(const void *a, const void *b)
4523 const struct btrfs_device_info *di_a = a;
4524 const struct btrfs_device_info *di_b = b;
4526 if (di_a->max_avail > di_b->max_avail)
4528 if (di_a->max_avail < di_b->max_avail)
4530 if (di_a->total_avail > di_b->total_avail)
4532 if (di_a->total_avail < di_b->total_avail)
4537 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4539 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4542 btrfs_set_fs_incompat(info, RAID56);
4545 #define BTRFS_MAX_DEVS(info) ((BTRFS_MAX_ITEM_SIZE(info) \
4546 - sizeof(struct btrfs_chunk)) \
4547 / sizeof(struct btrfs_stripe) + 1)
4549 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
4550 - 2 * sizeof(struct btrfs_disk_key) \
4551 - 2 * sizeof(struct btrfs_chunk)) \
4552 / sizeof(struct btrfs_stripe) + 1)
4554 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4555 u64 start, u64 type)
4557 struct btrfs_fs_info *info = trans->fs_info;
4558 struct btrfs_fs_devices *fs_devices = info->fs_devices;
4559 struct btrfs_device *device;
4560 struct map_lookup *map = NULL;
4561 struct extent_map_tree *em_tree;
4562 struct extent_map *em;
4563 struct btrfs_device_info *devices_info = NULL;
4565 int num_stripes; /* total number of stripes to allocate */
4566 int data_stripes; /* number of stripes that count for
4568 int sub_stripes; /* sub_stripes info for map */
4569 int dev_stripes; /* stripes per dev */
4570 int devs_max; /* max devs to use */
4571 int devs_min; /* min devs needed */
4572 int devs_increment; /* ndevs has to be a multiple of this */
4573 int ncopies; /* how many copies to data has */
4575 u64 max_stripe_size;
4584 BUG_ON(!alloc_profile_is_valid(type, 0));
4586 if (list_empty(&fs_devices->alloc_list)) {
4587 if (btrfs_test_opt(info, ENOSPC_DEBUG))
4588 btrfs_debug(info, "%s: no writable device", __func__);
4592 index = btrfs_bg_flags_to_raid_index(type);
4594 sub_stripes = btrfs_raid_array[index].sub_stripes;
4595 dev_stripes = btrfs_raid_array[index].dev_stripes;
4596 devs_max = btrfs_raid_array[index].devs_max;
4597 devs_min = btrfs_raid_array[index].devs_min;
4598 devs_increment = btrfs_raid_array[index].devs_increment;
4599 ncopies = btrfs_raid_array[index].ncopies;
4601 if (type & BTRFS_BLOCK_GROUP_DATA) {
4602 max_stripe_size = SZ_1G;
4603 max_chunk_size = BTRFS_MAX_DATA_CHUNK_SIZE;
4605 devs_max = BTRFS_MAX_DEVS(info);
4606 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4607 /* for larger filesystems, use larger metadata chunks */
4608 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
4609 max_stripe_size = SZ_1G;
4611 max_stripe_size = SZ_256M;
4612 max_chunk_size = max_stripe_size;
4614 devs_max = BTRFS_MAX_DEVS(info);
4615 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4616 max_stripe_size = SZ_32M;
4617 max_chunk_size = 2 * max_stripe_size;
4619 devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4621 btrfs_err(info, "invalid chunk type 0x%llx requested",
4626 /* we don't want a chunk larger than 10% of writeable space */
4627 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4630 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4636 * in the first pass through the devices list, we gather information
4637 * about the available holes on each device.
4640 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
4644 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4646 "BTRFS: read-only device in alloc_list\n");
4650 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
4651 &device->dev_state) ||
4652 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4655 if (device->total_bytes > device->bytes_used)
4656 total_avail = device->total_bytes - device->bytes_used;
4660 /* If there is no space on this device, skip it. */
4661 if (total_avail == 0)
4664 ret = find_free_dev_extent(trans, device,
4665 max_stripe_size * dev_stripes,
4666 &dev_offset, &max_avail);
4667 if (ret && ret != -ENOSPC)
4671 max_avail = max_stripe_size * dev_stripes;
4673 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes) {
4674 if (btrfs_test_opt(info, ENOSPC_DEBUG))
4676 "%s: devid %llu has no free space, have=%llu want=%u",
4677 __func__, device->devid, max_avail,
4678 BTRFS_STRIPE_LEN * dev_stripes);
4682 if (ndevs == fs_devices->rw_devices) {
4683 WARN(1, "%s: found more than %llu devices\n",
4684 __func__, fs_devices->rw_devices);
4687 devices_info[ndevs].dev_offset = dev_offset;
4688 devices_info[ndevs].max_avail = max_avail;
4689 devices_info[ndevs].total_avail = total_avail;
4690 devices_info[ndevs].dev = device;
4695 * now sort the devices by hole size / available space
4697 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4698 btrfs_cmp_device_info, NULL);
4700 /* round down to number of usable stripes */
4701 ndevs = round_down(ndevs, devs_increment);
4703 if (ndevs < devs_min) {
4705 if (btrfs_test_opt(info, ENOSPC_DEBUG)) {
4707 "%s: not enough devices with free space: have=%d minimum required=%d",
4708 __func__, ndevs, devs_min);
4713 ndevs = min(ndevs, devs_max);
4716 * The primary goal is to maximize the number of stripes, so use as
4717 * many devices as possible, even if the stripes are not maximum sized.
4719 * The DUP profile stores more than one stripe per device, the
4720 * max_avail is the total size so we have to adjust.
4722 stripe_size = div_u64(devices_info[ndevs - 1].max_avail, dev_stripes);
4723 num_stripes = ndevs * dev_stripes;
4726 * this will have to be fixed for RAID1 and RAID10 over
4729 data_stripes = num_stripes / ncopies;
4731 if (type & BTRFS_BLOCK_GROUP_RAID5)
4732 data_stripes = num_stripes - 1;
4734 if (type & BTRFS_BLOCK_GROUP_RAID6)
4735 data_stripes = num_stripes - 2;
4738 * Use the number of data stripes to figure out how big this chunk
4739 * is really going to be in terms of logical address space,
4740 * and compare that answer with the max chunk size
4742 if (stripe_size * data_stripes > max_chunk_size) {
4743 stripe_size = div_u64(max_chunk_size, data_stripes);
4745 /* bump the answer up to a 16MB boundary */
4746 stripe_size = round_up(stripe_size, SZ_16M);
4749 * But don't go higher than the limits we found while searching
4752 stripe_size = min(devices_info[ndevs - 1].max_avail,
4756 /* align to BTRFS_STRIPE_LEN */
4757 stripe_size = round_down(stripe_size, BTRFS_STRIPE_LEN);
4759 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4764 map->num_stripes = num_stripes;
4766 for (i = 0; i < ndevs; ++i) {
4767 for (j = 0; j < dev_stripes; ++j) {
4768 int s = i * dev_stripes + j;
4769 map->stripes[s].dev = devices_info[i].dev;
4770 map->stripes[s].physical = devices_info[i].dev_offset +
4774 map->stripe_len = BTRFS_STRIPE_LEN;
4775 map->io_align = BTRFS_STRIPE_LEN;
4776 map->io_width = BTRFS_STRIPE_LEN;
4778 map->sub_stripes = sub_stripes;
4780 num_bytes = stripe_size * data_stripes;
4782 trace_btrfs_chunk_alloc(info, map, start, num_bytes);
4784 em = alloc_extent_map();
4790 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4791 em->map_lookup = map;
4793 em->len = num_bytes;
4794 em->block_start = 0;
4795 em->block_len = em->len;
4796 em->orig_block_len = stripe_size;
4798 em_tree = &info->mapping_tree.map_tree;
4799 write_lock(&em_tree->lock);
4800 ret = add_extent_mapping(em_tree, em, 0);
4802 write_unlock(&em_tree->lock);
4803 free_extent_map(em);
4807 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4808 refcount_inc(&em->refs);
4809 write_unlock(&em_tree->lock);
4811 ret = btrfs_make_block_group(trans, 0, type, start, num_bytes);
4813 goto error_del_extent;
4815 for (i = 0; i < map->num_stripes; i++) {
4816 num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4817 btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4820 atomic64_sub(stripe_size * map->num_stripes, &info->free_chunk_space);
4822 free_extent_map(em);
4823 check_raid56_incompat_flag(info, type);
4825 kfree(devices_info);
4829 write_lock(&em_tree->lock);
4830 remove_extent_mapping(em_tree, em);
4831 write_unlock(&em_tree->lock);
4833 /* One for our allocation */
4834 free_extent_map(em);
4835 /* One for the tree reference */
4836 free_extent_map(em);
4837 /* One for the pending_chunks list reference */
4838 free_extent_map(em);
4840 kfree(devices_info);
4844 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4845 struct btrfs_fs_info *fs_info,
4846 u64 chunk_offset, u64 chunk_size)
4848 struct btrfs_root *extent_root = fs_info->extent_root;
4849 struct btrfs_root *chunk_root = fs_info->chunk_root;
4850 struct btrfs_key key;
4851 struct btrfs_device *device;
4852 struct btrfs_chunk *chunk;
4853 struct btrfs_stripe *stripe;
4854 struct extent_map *em;
4855 struct map_lookup *map;
4862 em = get_chunk_map(fs_info, chunk_offset, chunk_size);
4866 map = em->map_lookup;
4867 item_size = btrfs_chunk_item_size(map->num_stripes);
4868 stripe_size = em->orig_block_len;
4870 chunk = kzalloc(item_size, GFP_NOFS);
4877 * Take the device list mutex to prevent races with the final phase of
4878 * a device replace operation that replaces the device object associated
4879 * with the map's stripes, because the device object's id can change
4880 * at any time during that final phase of the device replace operation
4881 * (dev-replace.c:btrfs_dev_replace_finishing()).
4883 mutex_lock(&fs_info->fs_devices->device_list_mutex);
4884 for (i = 0; i < map->num_stripes; i++) {
4885 device = map->stripes[i].dev;
4886 dev_offset = map->stripes[i].physical;
4888 ret = btrfs_update_device(trans, device);
4891 ret = btrfs_alloc_dev_extent(trans, device, chunk_offset,
4892 dev_offset, stripe_size);
4897 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4901 stripe = &chunk->stripe;
4902 for (i = 0; i < map->num_stripes; i++) {
4903 device = map->stripes[i].dev;
4904 dev_offset = map->stripes[i].physical;
4906 btrfs_set_stack_stripe_devid(stripe, device->devid);
4907 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4908 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4911 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4913 btrfs_set_stack_chunk_length(chunk, chunk_size);
4914 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4915 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4916 btrfs_set_stack_chunk_type(chunk, map->type);
4917 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4918 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4919 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4920 btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
4921 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4923 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4924 key.type = BTRFS_CHUNK_ITEM_KEY;
4925 key.offset = chunk_offset;
4927 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4928 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4930 * TODO: Cleanup of inserted chunk root in case of
4933 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
4938 free_extent_map(em);
4943 * Chunk allocation falls into two parts. The first part does works
4944 * that make the new allocated chunk useable, but not do any operation
4945 * that modifies the chunk tree. The second part does the works that
4946 * require modifying the chunk tree. This division is important for the
4947 * bootstrap process of adding storage to a seed btrfs.
4949 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, u64 type)
4953 lockdep_assert_held(&trans->fs_info->chunk_mutex);
4954 chunk_offset = find_next_chunk(trans->fs_info);
4955 return __btrfs_alloc_chunk(trans, chunk_offset, type);
4958 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4959 struct btrfs_fs_info *fs_info)
4962 u64 sys_chunk_offset;
4966 chunk_offset = find_next_chunk(fs_info);
4967 alloc_profile = btrfs_metadata_alloc_profile(fs_info);
4968 ret = __btrfs_alloc_chunk(trans, chunk_offset, alloc_profile);
4972 sys_chunk_offset = find_next_chunk(fs_info);
4973 alloc_profile = btrfs_system_alloc_profile(fs_info);
4974 ret = __btrfs_alloc_chunk(trans, sys_chunk_offset, alloc_profile);
4978 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
4982 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
4983 BTRFS_BLOCK_GROUP_RAID10 |
4984 BTRFS_BLOCK_GROUP_RAID5 |
4985 BTRFS_BLOCK_GROUP_DUP)) {
4987 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
4996 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset)
4998 struct extent_map *em;
4999 struct map_lookup *map;
5004 em = get_chunk_map(fs_info, chunk_offset, 1);
5008 map = em->map_lookup;
5009 for (i = 0; i < map->num_stripes; i++) {
5010 if (test_bit(BTRFS_DEV_STATE_MISSING,
5011 &map->stripes[i].dev->dev_state)) {
5015 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE,
5016 &map->stripes[i].dev->dev_state)) {
5023 * If the number of missing devices is larger than max errors,
5024 * we can not write the data into that chunk successfully, so
5027 if (miss_ndevs > btrfs_chunk_max_errors(map))
5030 free_extent_map(em);
5034 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
5036 extent_map_tree_init(&tree->map_tree);
5039 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
5041 struct extent_map *em;
5044 write_lock(&tree->map_tree.lock);
5045 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
5047 remove_extent_mapping(&tree->map_tree, em);
5048 write_unlock(&tree->map_tree.lock);
5052 free_extent_map(em);
5053 /* once for the tree */
5054 free_extent_map(em);
5058 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5060 struct extent_map *em;
5061 struct map_lookup *map;
5064 em = get_chunk_map(fs_info, logical, len);
5067 * We could return errors for these cases, but that could get
5068 * ugly and we'd probably do the same thing which is just not do
5069 * anything else and exit, so return 1 so the callers don't try
5070 * to use other copies.
5074 map = em->map_lookup;
5075 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
5076 ret = map->num_stripes;
5077 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5078 ret = map->sub_stripes;
5079 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5081 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5083 * There could be two corrupted data stripes, we need
5084 * to loop retry in order to rebuild the correct data.
5086 * Fail a stripe at a time on every retry except the
5087 * stripe under reconstruction.
5089 ret = map->num_stripes;
5092 free_extent_map(em);
5094 btrfs_dev_replace_read_lock(&fs_info->dev_replace);
5095 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) &&
5096 fs_info->dev_replace.tgtdev)
5098 btrfs_dev_replace_read_unlock(&fs_info->dev_replace);
5103 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
5106 struct extent_map *em;
5107 struct map_lookup *map;
5108 unsigned long len = fs_info->sectorsize;
5110 em = get_chunk_map(fs_info, logical, len);
5112 if (!WARN_ON(IS_ERR(em))) {
5113 map = em->map_lookup;
5114 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5115 len = map->stripe_len * nr_data_stripes(map);
5116 free_extent_map(em);
5121 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5123 struct extent_map *em;
5124 struct map_lookup *map;
5127 em = get_chunk_map(fs_info, logical, len);
5129 if(!WARN_ON(IS_ERR(em))) {
5130 map = em->map_lookup;
5131 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5133 free_extent_map(em);
5138 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5139 struct map_lookup *map, int first,
5140 int dev_replace_is_ongoing)
5144 int preferred_mirror;
5146 struct btrfs_device *srcdev;
5149 (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)));
5151 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5152 num_stripes = map->sub_stripes;
5154 num_stripes = map->num_stripes;
5156 preferred_mirror = first + current->pid % num_stripes;
5158 if (dev_replace_is_ongoing &&
5159 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5160 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5161 srcdev = fs_info->dev_replace.srcdev;
5166 * try to avoid the drive that is the source drive for a
5167 * dev-replace procedure, only choose it if no other non-missing
5168 * mirror is available
5170 for (tolerance = 0; tolerance < 2; tolerance++) {
5171 if (map->stripes[preferred_mirror].dev->bdev &&
5172 (tolerance || map->stripes[preferred_mirror].dev != srcdev))
5173 return preferred_mirror;
5174 for (i = first; i < first + num_stripes; i++) {
5175 if (map->stripes[i].dev->bdev &&
5176 (tolerance || map->stripes[i].dev != srcdev))
5181 /* we couldn't find one that doesn't fail. Just return something
5182 * and the io error handling code will clean up eventually
5184 return preferred_mirror;
5187 static inline int parity_smaller(u64 a, u64 b)
5192 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5193 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5195 struct btrfs_bio_stripe s;
5202 for (i = 0; i < num_stripes - 1; i++) {
5203 if (parity_smaller(bbio->raid_map[i],
5204 bbio->raid_map[i+1])) {
5205 s = bbio->stripes[i];
5206 l = bbio->raid_map[i];
5207 bbio->stripes[i] = bbio->stripes[i+1];
5208 bbio->raid_map[i] = bbio->raid_map[i+1];
5209 bbio->stripes[i+1] = s;
5210 bbio->raid_map[i+1] = l;
5218 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5220 struct btrfs_bio *bbio = kzalloc(
5221 /* the size of the btrfs_bio */
5222 sizeof(struct btrfs_bio) +
5223 /* plus the variable array for the stripes */
5224 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5225 /* plus the variable array for the tgt dev */
5226 sizeof(int) * (real_stripes) +
5228 * plus the raid_map, which includes both the tgt dev
5231 sizeof(u64) * (total_stripes),
5232 GFP_NOFS|__GFP_NOFAIL);
5234 atomic_set(&bbio->error, 0);
5235 refcount_set(&bbio->refs, 1);
5240 void btrfs_get_bbio(struct btrfs_bio *bbio)
5242 WARN_ON(!refcount_read(&bbio->refs));
5243 refcount_inc(&bbio->refs);
5246 void btrfs_put_bbio(struct btrfs_bio *bbio)
5250 if (refcount_dec_and_test(&bbio->refs))
5254 /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */
5256 * Please note that, discard won't be sent to target device of device
5259 static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info,
5260 u64 logical, u64 length,
5261 struct btrfs_bio **bbio_ret)
5263 struct extent_map *em;
5264 struct map_lookup *map;
5265 struct btrfs_bio *bbio;
5269 u64 stripe_end_offset;
5276 u32 sub_stripes = 0;
5277 u64 stripes_per_dev = 0;
5278 u32 remaining_stripes = 0;
5279 u32 last_stripe = 0;
5283 /* discard always return a bbio */
5286 em = get_chunk_map(fs_info, logical, length);
5290 map = em->map_lookup;
5291 /* we don't discard raid56 yet */
5292 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5297 offset = logical - em->start;
5298 length = min_t(u64, em->len - offset, length);
5300 stripe_len = map->stripe_len;
5302 * stripe_nr counts the total number of stripes we have to stride
5303 * to get to this block
5305 stripe_nr = div64_u64(offset, stripe_len);
5307 /* stripe_offset is the offset of this block in its stripe */
5308 stripe_offset = offset - stripe_nr * stripe_len;
5310 stripe_nr_end = round_up(offset + length, map->stripe_len);
5311 stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len);
5312 stripe_cnt = stripe_nr_end - stripe_nr;
5313 stripe_end_offset = stripe_nr_end * map->stripe_len -
5316 * after this, stripe_nr is the number of stripes on this
5317 * device we have to walk to find the data, and stripe_index is
5318 * the number of our device in the stripe array
5322 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5323 BTRFS_BLOCK_GROUP_RAID10)) {
5324 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5327 sub_stripes = map->sub_stripes;
5329 factor = map->num_stripes / sub_stripes;
5330 num_stripes = min_t(u64, map->num_stripes,
5331 sub_stripes * stripe_cnt);
5332 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5333 stripe_index *= sub_stripes;
5334 stripes_per_dev = div_u64_rem(stripe_cnt, factor,
5335 &remaining_stripes);
5336 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5337 last_stripe *= sub_stripes;
5338 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5339 BTRFS_BLOCK_GROUP_DUP)) {
5340 num_stripes = map->num_stripes;
5342 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5346 bbio = alloc_btrfs_bio(num_stripes, 0);
5352 for (i = 0; i < num_stripes; i++) {
5353 bbio->stripes[i].physical =
5354 map->stripes[stripe_index].physical +
5355 stripe_offset + stripe_nr * map->stripe_len;
5356 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5358 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5359 BTRFS_BLOCK_GROUP_RAID10)) {
5360 bbio->stripes[i].length = stripes_per_dev *
5363 if (i / sub_stripes < remaining_stripes)
5364 bbio->stripes[i].length +=
5368 * Special for the first stripe and
5371 * |-------|...|-------|
5375 if (i < sub_stripes)
5376 bbio->stripes[i].length -=
5379 if (stripe_index >= last_stripe &&
5380 stripe_index <= (last_stripe +
5382 bbio->stripes[i].length -=
5385 if (i == sub_stripes - 1)
5388 bbio->stripes[i].length = length;
5392 if (stripe_index == map->num_stripes) {
5399 bbio->map_type = map->type;
5400 bbio->num_stripes = num_stripes;
5402 free_extent_map(em);
5407 * In dev-replace case, for repair case (that's the only case where the mirror
5408 * is selected explicitly when calling btrfs_map_block), blocks left of the
5409 * left cursor can also be read from the target drive.
5411 * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the
5413 * For READ, it also needs to be supported using the same mirror number.
5415 * If the requested block is not left of the left cursor, EIO is returned. This
5416 * can happen because btrfs_num_copies() returns one more in the dev-replace
5419 static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info,
5420 u64 logical, u64 length,
5421 u64 srcdev_devid, int *mirror_num,
5424 struct btrfs_bio *bbio = NULL;
5426 int index_srcdev = 0;
5428 u64 physical_of_found = 0;
5432 ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS,
5433 logical, &length, &bbio, 0, 0);
5435 ASSERT(bbio == NULL);
5439 num_stripes = bbio->num_stripes;
5440 if (*mirror_num > num_stripes) {
5442 * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror,
5443 * that means that the requested area is not left of the left
5446 btrfs_put_bbio(bbio);
5451 * process the rest of the function using the mirror_num of the source
5452 * drive. Therefore look it up first. At the end, patch the device
5453 * pointer to the one of the target drive.
5455 for (i = 0; i < num_stripes; i++) {
5456 if (bbio->stripes[i].dev->devid != srcdev_devid)
5460 * In case of DUP, in order to keep it simple, only add the
5461 * mirror with the lowest physical address
5464 physical_of_found <= bbio->stripes[i].physical)
5469 physical_of_found = bbio->stripes[i].physical;
5472 btrfs_put_bbio(bbio);
5478 *mirror_num = index_srcdev + 1;
5479 *physical = physical_of_found;
5483 static void handle_ops_on_dev_replace(enum btrfs_map_op op,
5484 struct btrfs_bio **bbio_ret,
5485 struct btrfs_dev_replace *dev_replace,
5486 int *num_stripes_ret, int *max_errors_ret)
5488 struct btrfs_bio *bbio = *bbio_ret;
5489 u64 srcdev_devid = dev_replace->srcdev->devid;
5490 int tgtdev_indexes = 0;
5491 int num_stripes = *num_stripes_ret;
5492 int max_errors = *max_errors_ret;
5495 if (op == BTRFS_MAP_WRITE) {
5496 int index_where_to_add;
5499 * duplicate the write operations while the dev replace
5500 * procedure is running. Since the copying of the old disk to
5501 * the new disk takes place at run time while the filesystem is
5502 * mounted writable, the regular write operations to the old
5503 * disk have to be duplicated to go to the new disk as well.
5505 * Note that device->missing is handled by the caller, and that
5506 * the write to the old disk is already set up in the stripes
5509 index_where_to_add = num_stripes;
5510 for (i = 0; i < num_stripes; i++) {
5511 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5512 /* write to new disk, too */
5513 struct btrfs_bio_stripe *new =
5514 bbio->stripes + index_where_to_add;
5515 struct btrfs_bio_stripe *old =
5518 new->physical = old->physical;
5519 new->length = old->length;
5520 new->dev = dev_replace->tgtdev;
5521 bbio->tgtdev_map[i] = index_where_to_add;
5522 index_where_to_add++;
5527 num_stripes = index_where_to_add;
5528 } else if (op == BTRFS_MAP_GET_READ_MIRRORS) {
5529 int index_srcdev = 0;
5531 u64 physical_of_found = 0;
5534 * During the dev-replace procedure, the target drive can also
5535 * be used to read data in case it is needed to repair a corrupt
5536 * block elsewhere. This is possible if the requested area is
5537 * left of the left cursor. In this area, the target drive is a
5538 * full copy of the source drive.
5540 for (i = 0; i < num_stripes; i++) {
5541 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5543 * In case of DUP, in order to keep it simple,
5544 * only add the mirror with the lowest physical
5548 physical_of_found <=
5549 bbio->stripes[i].physical)
5553 physical_of_found = bbio->stripes[i].physical;
5557 struct btrfs_bio_stripe *tgtdev_stripe =
5558 bbio->stripes + num_stripes;
5560 tgtdev_stripe->physical = physical_of_found;
5561 tgtdev_stripe->length =
5562 bbio->stripes[index_srcdev].length;
5563 tgtdev_stripe->dev = dev_replace->tgtdev;
5564 bbio->tgtdev_map[index_srcdev] = num_stripes;
5571 *num_stripes_ret = num_stripes;
5572 *max_errors_ret = max_errors;
5573 bbio->num_tgtdevs = tgtdev_indexes;
5577 static bool need_full_stripe(enum btrfs_map_op op)
5579 return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS);
5582 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
5583 enum btrfs_map_op op,
5584 u64 logical, u64 *length,
5585 struct btrfs_bio **bbio_ret,
5586 int mirror_num, int need_raid_map)
5588 struct extent_map *em;
5589 struct map_lookup *map;
5599 int tgtdev_indexes = 0;
5600 struct btrfs_bio *bbio = NULL;
5601 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
5602 int dev_replace_is_ongoing = 0;
5603 int num_alloc_stripes;
5604 int patch_the_first_stripe_for_dev_replace = 0;
5605 u64 physical_to_patch_in_first_stripe = 0;
5606 u64 raid56_full_stripe_start = (u64)-1;
5608 if (op == BTRFS_MAP_DISCARD)
5609 return __btrfs_map_block_for_discard(fs_info, logical,
5612 em = get_chunk_map(fs_info, logical, *length);
5616 map = em->map_lookup;
5617 offset = logical - em->start;
5619 stripe_len = map->stripe_len;
5622 * stripe_nr counts the total number of stripes we have to stride
5623 * to get to this block
5625 stripe_nr = div64_u64(stripe_nr, stripe_len);
5627 stripe_offset = stripe_nr * stripe_len;
5628 if (offset < stripe_offset) {
5630 "stripe math has gone wrong, stripe_offset=%llu, offset=%llu, start=%llu, logical=%llu, stripe_len=%llu",
5631 stripe_offset, offset, em->start, logical,
5633 free_extent_map(em);
5637 /* stripe_offset is the offset of this block in its stripe*/
5638 stripe_offset = offset - stripe_offset;
5640 /* if we're here for raid56, we need to know the stripe aligned start */
5641 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5642 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5643 raid56_full_stripe_start = offset;
5645 /* allow a write of a full stripe, but make sure we don't
5646 * allow straddling of stripes
5648 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5650 raid56_full_stripe_start *= full_stripe_len;
5653 if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5655 /* For writes to RAID[56], allow a full stripeset across all disks.
5656 For other RAID types and for RAID[56] reads, just allow a single
5657 stripe (on a single disk). */
5658 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
5659 (op == BTRFS_MAP_WRITE)) {
5660 max_len = stripe_len * nr_data_stripes(map) -
5661 (offset - raid56_full_stripe_start);
5663 /* we limit the length of each bio to what fits in a stripe */
5664 max_len = stripe_len - stripe_offset;
5666 *length = min_t(u64, em->len - offset, max_len);
5668 *length = em->len - offset;
5671 /* This is for when we're called from btrfs_merge_bio_hook() and all
5672 it cares about is the length */
5676 btrfs_dev_replace_read_lock(dev_replace);
5677 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5678 if (!dev_replace_is_ongoing)
5679 btrfs_dev_replace_read_unlock(dev_replace);
5681 btrfs_dev_replace_set_lock_blocking(dev_replace);
5683 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
5684 !need_full_stripe(op) && dev_replace->tgtdev != NULL) {
5685 ret = get_extra_mirror_from_replace(fs_info, logical, *length,
5686 dev_replace->srcdev->devid,
5688 &physical_to_patch_in_first_stripe);
5692 patch_the_first_stripe_for_dev_replace = 1;
5693 } else if (mirror_num > map->num_stripes) {
5699 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5700 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5702 if (!need_full_stripe(op))
5704 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
5705 if (need_full_stripe(op))
5706 num_stripes = map->num_stripes;
5707 else if (mirror_num)
5708 stripe_index = mirror_num - 1;
5710 stripe_index = find_live_mirror(fs_info, map, 0,
5711 dev_replace_is_ongoing);
5712 mirror_num = stripe_index + 1;
5715 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
5716 if (need_full_stripe(op)) {
5717 num_stripes = map->num_stripes;
5718 } else if (mirror_num) {
5719 stripe_index = mirror_num - 1;
5724 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5725 u32 factor = map->num_stripes / map->sub_stripes;
5727 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5728 stripe_index *= map->sub_stripes;
5730 if (need_full_stripe(op))
5731 num_stripes = map->sub_stripes;
5732 else if (mirror_num)
5733 stripe_index += mirror_num - 1;
5735 int old_stripe_index = stripe_index;
5736 stripe_index = find_live_mirror(fs_info, map,
5738 dev_replace_is_ongoing);
5739 mirror_num = stripe_index - old_stripe_index + 1;
5742 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5743 if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) {
5744 /* push stripe_nr back to the start of the full stripe */
5745 stripe_nr = div64_u64(raid56_full_stripe_start,
5746 stripe_len * nr_data_stripes(map));
5748 /* RAID[56] write or recovery. Return all stripes */
5749 num_stripes = map->num_stripes;
5750 max_errors = nr_parity_stripes(map);
5752 *length = map->stripe_len;
5757 * Mirror #0 or #1 means the original data block.
5758 * Mirror #2 is RAID5 parity block.
5759 * Mirror #3 is RAID6 Q block.
5761 stripe_nr = div_u64_rem(stripe_nr,
5762 nr_data_stripes(map), &stripe_index);
5764 stripe_index = nr_data_stripes(map) +
5767 /* We distribute the parity blocks across stripes */
5768 div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
5770 if (!need_full_stripe(op) && mirror_num <= 1)
5775 * after this, stripe_nr is the number of stripes on this
5776 * device we have to walk to find the data, and stripe_index is
5777 * the number of our device in the stripe array
5779 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5781 mirror_num = stripe_index + 1;
5783 if (stripe_index >= map->num_stripes) {
5785 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
5786 stripe_index, map->num_stripes);
5791 num_alloc_stripes = num_stripes;
5792 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) {
5793 if (op == BTRFS_MAP_WRITE)
5794 num_alloc_stripes <<= 1;
5795 if (op == BTRFS_MAP_GET_READ_MIRRORS)
5796 num_alloc_stripes++;
5797 tgtdev_indexes = num_stripes;
5800 bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
5805 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL)
5806 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
5808 /* build raid_map */
5809 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map &&
5810 (need_full_stripe(op) || mirror_num > 1)) {
5814 bbio->raid_map = (u64 *)((void *)bbio->stripes +
5815 sizeof(struct btrfs_bio_stripe) *
5817 sizeof(int) * tgtdev_indexes);
5819 /* Work out the disk rotation on this stripe-set */
5820 div_u64_rem(stripe_nr, num_stripes, &rot);
5822 /* Fill in the logical address of each stripe */
5823 tmp = stripe_nr * nr_data_stripes(map);
5824 for (i = 0; i < nr_data_stripes(map); i++)
5825 bbio->raid_map[(i+rot) % num_stripes] =
5826 em->start + (tmp + i) * map->stripe_len;
5828 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5829 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5830 bbio->raid_map[(i+rot+1) % num_stripes] =
5835 for (i = 0; i < num_stripes; i++) {
5836 bbio->stripes[i].physical =
5837 map->stripes[stripe_index].physical +
5839 stripe_nr * map->stripe_len;
5840 bbio->stripes[i].dev =
5841 map->stripes[stripe_index].dev;
5845 if (need_full_stripe(op))
5846 max_errors = btrfs_chunk_max_errors(map);
5849 sort_parity_stripes(bbio, num_stripes);
5851 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
5852 need_full_stripe(op)) {
5853 handle_ops_on_dev_replace(op, &bbio, dev_replace, &num_stripes,
5858 bbio->map_type = map->type;
5859 bbio->num_stripes = num_stripes;
5860 bbio->max_errors = max_errors;
5861 bbio->mirror_num = mirror_num;
5864 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5865 * mirror_num == num_stripes + 1 && dev_replace target drive is
5866 * available as a mirror
5868 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5869 WARN_ON(num_stripes > 1);
5870 bbio->stripes[0].dev = dev_replace->tgtdev;
5871 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5872 bbio->mirror_num = map->num_stripes + 1;
5875 if (dev_replace_is_ongoing) {
5876 btrfs_dev_replace_clear_lock_blocking(dev_replace);
5877 btrfs_dev_replace_read_unlock(dev_replace);
5879 free_extent_map(em);
5883 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
5884 u64 logical, u64 *length,
5885 struct btrfs_bio **bbio_ret, int mirror_num)
5887 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret,
5891 /* For Scrub/replace */
5892 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
5893 u64 logical, u64 *length,
5894 struct btrfs_bio **bbio_ret)
5896 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1);
5899 int btrfs_rmap_block(struct btrfs_fs_info *fs_info, u64 chunk_start,
5900 u64 physical, u64 **logical, int *naddrs, int *stripe_len)
5902 struct extent_map *em;
5903 struct map_lookup *map;
5911 em = get_chunk_map(fs_info, chunk_start, 1);
5915 map = em->map_lookup;
5917 rmap_len = map->stripe_len;
5919 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5920 length = div_u64(length, map->num_stripes / map->sub_stripes);
5921 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5922 length = div_u64(length, map->num_stripes);
5923 else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5924 length = div_u64(length, nr_data_stripes(map));
5925 rmap_len = map->stripe_len * nr_data_stripes(map);
5928 buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
5929 BUG_ON(!buf); /* -ENOMEM */
5931 for (i = 0; i < map->num_stripes; i++) {
5932 if (map->stripes[i].physical > physical ||
5933 map->stripes[i].physical + length <= physical)
5936 stripe_nr = physical - map->stripes[i].physical;
5937 stripe_nr = div64_u64(stripe_nr, map->stripe_len);
5939 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5940 stripe_nr = stripe_nr * map->num_stripes + i;
5941 stripe_nr = div_u64(stripe_nr, map->sub_stripes);
5942 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5943 stripe_nr = stripe_nr * map->num_stripes + i;
5944 } /* else if RAID[56], multiply by nr_data_stripes().
5945 * Alternatively, just use rmap_len below instead of
5946 * map->stripe_len */
5948 bytenr = chunk_start + stripe_nr * rmap_len;
5949 WARN_ON(nr >= map->num_stripes);
5950 for (j = 0; j < nr; j++) {
5951 if (buf[j] == bytenr)
5955 WARN_ON(nr >= map->num_stripes);
5962 *stripe_len = rmap_len;
5964 free_extent_map(em);
5968 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
5970 bio->bi_private = bbio->private;
5971 bio->bi_end_io = bbio->end_io;
5974 btrfs_put_bbio(bbio);
5977 static void btrfs_end_bio(struct bio *bio)
5979 struct btrfs_bio *bbio = bio->bi_private;
5980 int is_orig_bio = 0;
5982 if (bio->bi_status) {
5983 atomic_inc(&bbio->error);
5984 if (bio->bi_status == BLK_STS_IOERR ||
5985 bio->bi_status == BLK_STS_TARGET) {
5986 unsigned int stripe_index =
5987 btrfs_io_bio(bio)->stripe_index;
5988 struct btrfs_device *dev;
5990 BUG_ON(stripe_index >= bbio->num_stripes);
5991 dev = bbio->stripes[stripe_index].dev;
5993 if (bio_op(bio) == REQ_OP_WRITE)
5994 btrfs_dev_stat_inc_and_print(dev,
5995 BTRFS_DEV_STAT_WRITE_ERRS);
5997 btrfs_dev_stat_inc_and_print(dev,
5998 BTRFS_DEV_STAT_READ_ERRS);
5999 if (bio->bi_opf & REQ_PREFLUSH)
6000 btrfs_dev_stat_inc_and_print(dev,
6001 BTRFS_DEV_STAT_FLUSH_ERRS);
6006 if (bio == bbio->orig_bio)
6009 btrfs_bio_counter_dec(bbio->fs_info);
6011 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6014 bio = bbio->orig_bio;
6017 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6018 /* only send an error to the higher layers if it is
6019 * beyond the tolerance of the btrfs bio
6021 if (atomic_read(&bbio->error) > bbio->max_errors) {
6022 bio->bi_status = BLK_STS_IOERR;
6025 * this bio is actually up to date, we didn't
6026 * go over the max number of errors
6028 bio->bi_status = BLK_STS_OK;
6031 btrfs_end_bbio(bbio, bio);
6032 } else if (!is_orig_bio) {
6038 * see run_scheduled_bios for a description of why bios are collected for
6041 * This will add one bio to the pending list for a device and make sure
6042 * the work struct is scheduled.
6044 static noinline void btrfs_schedule_bio(struct btrfs_device *device,
6047 struct btrfs_fs_info *fs_info = device->fs_info;
6048 int should_queue = 1;
6049 struct btrfs_pending_bios *pending_bios;
6051 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state) ||
6057 /* don't bother with additional async steps for reads, right now */
6058 if (bio_op(bio) == REQ_OP_READ) {
6059 btrfsic_submit_bio(bio);
6063 WARN_ON(bio->bi_next);
6064 bio->bi_next = NULL;
6066 spin_lock(&device->io_lock);
6067 if (op_is_sync(bio->bi_opf))
6068 pending_bios = &device->pending_sync_bios;
6070 pending_bios = &device->pending_bios;
6072 if (pending_bios->tail)
6073 pending_bios->tail->bi_next = bio;
6075 pending_bios->tail = bio;
6076 if (!pending_bios->head)
6077 pending_bios->head = bio;
6078 if (device->running_pending)
6081 spin_unlock(&device->io_lock);
6084 btrfs_queue_work(fs_info->submit_workers, &device->work);
6087 static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio,
6088 u64 physical, int dev_nr, int async)
6090 struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
6091 struct btrfs_fs_info *fs_info = bbio->fs_info;
6093 bio->bi_private = bbio;
6094 btrfs_io_bio(bio)->stripe_index = dev_nr;
6095 bio->bi_end_io = btrfs_end_bio;
6096 bio->bi_iter.bi_sector = physical >> 9;
6099 struct rcu_string *name;
6102 name = rcu_dereference(dev->name);
6103 btrfs_debug(fs_info,
6104 "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
6105 bio_op(bio), bio->bi_opf,
6106 (u64)bio->bi_iter.bi_sector,
6107 (u_long)dev->bdev->bd_dev, name->str, dev->devid,
6108 bio->bi_iter.bi_size);
6112 bio_set_dev(bio, dev->bdev);
6114 btrfs_bio_counter_inc_noblocked(fs_info);
6117 btrfs_schedule_bio(dev, bio);
6119 btrfsic_submit_bio(bio);
6122 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6124 atomic_inc(&bbio->error);
6125 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6126 /* Should be the original bio. */
6127 WARN_ON(bio != bbio->orig_bio);
6129 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6130 bio->bi_iter.bi_sector = logical >> 9;
6131 if (atomic_read(&bbio->error) > bbio->max_errors)
6132 bio->bi_status = BLK_STS_IOERR;
6134 bio->bi_status = BLK_STS_OK;
6135 btrfs_end_bbio(bbio, bio);
6139 blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
6140 int mirror_num, int async_submit)
6142 struct btrfs_device *dev;
6143 struct bio *first_bio = bio;
6144 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
6150 struct btrfs_bio *bbio = NULL;
6152 length = bio->bi_iter.bi_size;
6153 map_length = length;
6155 btrfs_bio_counter_inc_blocked(fs_info);
6156 ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical,
6157 &map_length, &bbio, mirror_num, 1);
6159 btrfs_bio_counter_dec(fs_info);
6160 return errno_to_blk_status(ret);
6163 total_devs = bbio->num_stripes;
6164 bbio->orig_bio = first_bio;
6165 bbio->private = first_bio->bi_private;
6166 bbio->end_io = first_bio->bi_end_io;
6167 bbio->fs_info = fs_info;
6168 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6170 if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6171 ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) {
6172 /* In this case, map_length has been set to the length of
6173 a single stripe; not the whole write */
6174 if (bio_op(bio) == REQ_OP_WRITE) {
6175 ret = raid56_parity_write(fs_info, bio, bbio,
6178 ret = raid56_parity_recover(fs_info, bio, bbio,
6179 map_length, mirror_num, 1);
6182 btrfs_bio_counter_dec(fs_info);
6183 return errno_to_blk_status(ret);
6186 if (map_length < length) {
6188 "mapping failed logical %llu bio len %llu len %llu",
6189 logical, length, map_length);
6193 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6194 dev = bbio->stripes[dev_nr].dev;
6195 if (!dev || !dev->bdev ||
6196 (bio_op(first_bio) == REQ_OP_WRITE &&
6197 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) {
6198 bbio_error(bbio, first_bio, logical);
6202 if (dev_nr < total_devs - 1)
6203 bio = btrfs_bio_clone(first_bio);
6207 submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical,
6208 dev_nr, async_submit);
6210 btrfs_bio_counter_dec(fs_info);
6214 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
6217 struct btrfs_device *device;
6218 struct btrfs_fs_devices *cur_devices;
6220 cur_devices = fs_info->fs_devices;
6221 while (cur_devices) {
6223 !memcmp(cur_devices->fsid, fsid, BTRFS_FSID_SIZE)) {
6224 device = find_device(cur_devices, devid, uuid);
6228 cur_devices = cur_devices->seed;
6233 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices,
6234 u64 devid, u8 *dev_uuid)
6236 struct btrfs_device *device;
6238 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6242 list_add(&device->dev_list, &fs_devices->devices);
6243 device->fs_devices = fs_devices;
6244 fs_devices->num_devices++;
6246 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6247 fs_devices->missing_devices++;
6253 * btrfs_alloc_device - allocate struct btrfs_device
6254 * @fs_info: used only for generating a new devid, can be NULL if
6255 * devid is provided (i.e. @devid != NULL).
6256 * @devid: a pointer to devid for this device. If NULL a new devid
6258 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6261 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6262 * on error. Returned struct is not linked onto any lists and must be
6263 * destroyed with btrfs_free_device.
6265 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6269 struct btrfs_device *dev;
6272 if (WARN_ON(!devid && !fs_info))
6273 return ERR_PTR(-EINVAL);
6275 dev = __alloc_device();
6284 ret = find_next_devid(fs_info, &tmp);
6286 btrfs_free_device(dev);
6287 return ERR_PTR(ret);
6293 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6295 generate_random_uuid(dev->uuid);
6297 btrfs_init_work(&dev->work, btrfs_submit_helper,
6298 pending_bios_fn, NULL, NULL);
6303 /* Return -EIO if any error, otherwise return 0. */
6304 static int btrfs_check_chunk_valid(struct btrfs_fs_info *fs_info,
6305 struct extent_buffer *leaf,
6306 struct btrfs_chunk *chunk, u64 logical)
6316 length = btrfs_chunk_length(leaf, chunk);
6317 stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6318 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6319 sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6320 type = btrfs_chunk_type(leaf, chunk);
6323 btrfs_err(fs_info, "invalid chunk num_stripes: %u",
6327 if (!IS_ALIGNED(logical, fs_info->sectorsize)) {
6328 btrfs_err(fs_info, "invalid chunk logical %llu", logical);
6331 if (btrfs_chunk_sector_size(leaf, chunk) != fs_info->sectorsize) {
6332 btrfs_err(fs_info, "invalid chunk sectorsize %u",
6333 btrfs_chunk_sector_size(leaf, chunk));
6336 if (!length || !IS_ALIGNED(length, fs_info->sectorsize)) {
6337 btrfs_err(fs_info, "invalid chunk length %llu", length);
6340 if (!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN) {
6341 btrfs_err(fs_info, "invalid chunk stripe length: %llu",
6345 if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6347 btrfs_err(fs_info, "unrecognized chunk type: %llu",
6348 ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
6349 BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6350 btrfs_chunk_type(leaf, chunk));
6354 if ((type & BTRFS_BLOCK_GROUP_TYPE_MASK) == 0) {
6355 btrfs_err(fs_info, "missing chunk type flag: 0x%llx", type);
6359 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) &&
6360 (type & (BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA))) {
6362 "system chunk with data or metadata type: 0x%llx", type);
6366 features = btrfs_super_incompat_flags(fs_info->super_copy);
6367 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
6371 if ((type & BTRFS_BLOCK_GROUP_METADATA) &&
6372 (type & BTRFS_BLOCK_GROUP_DATA)) {
6374 "mixed chunk type in non-mixed mode: 0x%llx", type);
6379 if ((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes != 2) ||
6380 (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) ||
6381 (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) ||
6382 (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) ||
6383 (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) ||
6384 ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
6385 num_stripes != 1)) {
6387 "invalid num_stripes:sub_stripes %u:%u for profile %llu",
6388 num_stripes, sub_stripes,
6389 type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
6396 static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info,
6397 u64 devid, u8 *uuid, bool error)
6400 btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing",
6403 btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing",
6407 static int read_one_chunk(struct btrfs_fs_info *fs_info, struct btrfs_key *key,
6408 struct extent_buffer *leaf,
6409 struct btrfs_chunk *chunk)
6411 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
6412 struct map_lookup *map;
6413 struct extent_map *em;
6417 u8 uuid[BTRFS_UUID_SIZE];
6422 logical = key->offset;
6423 length = btrfs_chunk_length(leaf, chunk);
6424 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6426 ret = btrfs_check_chunk_valid(fs_info, leaf, chunk, logical);
6430 read_lock(&map_tree->map_tree.lock);
6431 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
6432 read_unlock(&map_tree->map_tree.lock);
6434 /* already mapped? */
6435 if (em && em->start <= logical && em->start + em->len > logical) {
6436 free_extent_map(em);
6439 free_extent_map(em);
6442 em = alloc_extent_map();
6445 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6447 free_extent_map(em);
6451 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6452 em->map_lookup = map;
6453 em->start = logical;
6456 em->block_start = 0;
6457 em->block_len = em->len;
6459 map->num_stripes = num_stripes;
6460 map->io_width = btrfs_chunk_io_width(leaf, chunk);
6461 map->io_align = btrfs_chunk_io_align(leaf, chunk);
6462 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6463 map->type = btrfs_chunk_type(leaf, chunk);
6464 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6465 for (i = 0; i < num_stripes; i++) {
6466 map->stripes[i].physical =
6467 btrfs_stripe_offset_nr(leaf, chunk, i);
6468 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6469 read_extent_buffer(leaf, uuid, (unsigned long)
6470 btrfs_stripe_dev_uuid_nr(chunk, i),
6472 map->stripes[i].dev = btrfs_find_device(fs_info, devid,
6474 if (!map->stripes[i].dev &&
6475 !btrfs_test_opt(fs_info, DEGRADED)) {
6476 free_extent_map(em);
6477 btrfs_report_missing_device(fs_info, devid, uuid, true);
6480 if (!map->stripes[i].dev) {
6481 map->stripes[i].dev =
6482 add_missing_dev(fs_info->fs_devices, devid,
6484 if (IS_ERR(map->stripes[i].dev)) {
6485 free_extent_map(em);
6487 "failed to init missing dev %llu: %ld",
6488 devid, PTR_ERR(map->stripes[i].dev));
6489 return PTR_ERR(map->stripes[i].dev);
6491 btrfs_report_missing_device(fs_info, devid, uuid, false);
6493 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
6494 &(map->stripes[i].dev->dev_state));
6498 write_lock(&map_tree->map_tree.lock);
6499 ret = add_extent_mapping(&map_tree->map_tree, em, 0);
6500 write_unlock(&map_tree->map_tree.lock);
6501 BUG_ON(ret); /* Tree corruption */
6502 free_extent_map(em);
6507 static void fill_device_from_item(struct extent_buffer *leaf,
6508 struct btrfs_dev_item *dev_item,
6509 struct btrfs_device *device)
6513 device->devid = btrfs_device_id(leaf, dev_item);
6514 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6515 device->total_bytes = device->disk_total_bytes;
6516 device->commit_total_bytes = device->disk_total_bytes;
6517 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6518 device->commit_bytes_used = device->bytes_used;
6519 device->type = btrfs_device_type(leaf, dev_item);
6520 device->io_align = btrfs_device_io_align(leaf, dev_item);
6521 device->io_width = btrfs_device_io_width(leaf, dev_item);
6522 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6523 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6524 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
6526 ptr = btrfs_device_uuid(dev_item);
6527 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6530 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info,
6533 struct btrfs_fs_devices *fs_devices;
6536 lockdep_assert_held(&uuid_mutex);
6539 fs_devices = fs_info->fs_devices->seed;
6540 while (fs_devices) {
6541 if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE))
6544 fs_devices = fs_devices->seed;
6547 fs_devices = find_fsid(fsid);
6549 if (!btrfs_test_opt(fs_info, DEGRADED))
6550 return ERR_PTR(-ENOENT);
6552 fs_devices = alloc_fs_devices(fsid);
6553 if (IS_ERR(fs_devices))
6556 fs_devices->seeding = 1;
6557 fs_devices->opened = 1;
6561 fs_devices = clone_fs_devices(fs_devices);
6562 if (IS_ERR(fs_devices))
6565 ret = open_fs_devices(fs_devices, FMODE_READ, fs_info->bdev_holder);
6567 free_fs_devices(fs_devices);
6568 fs_devices = ERR_PTR(ret);
6572 if (!fs_devices->seeding) {
6573 close_fs_devices(fs_devices);
6574 free_fs_devices(fs_devices);
6575 fs_devices = ERR_PTR(-EINVAL);
6579 fs_devices->seed = fs_info->fs_devices->seed;
6580 fs_info->fs_devices->seed = fs_devices;
6585 static int read_one_dev(struct btrfs_fs_info *fs_info,
6586 struct extent_buffer *leaf,
6587 struct btrfs_dev_item *dev_item)
6589 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6590 struct btrfs_device *device;
6593 u8 fs_uuid[BTRFS_FSID_SIZE];
6594 u8 dev_uuid[BTRFS_UUID_SIZE];
6596 devid = btrfs_device_id(leaf, dev_item);
6597 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6599 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6602 if (memcmp(fs_uuid, fs_info->fsid, BTRFS_FSID_SIZE)) {
6603 fs_devices = open_seed_devices(fs_info, fs_uuid);
6604 if (IS_ERR(fs_devices))
6605 return PTR_ERR(fs_devices);
6608 device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
6610 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6611 btrfs_report_missing_device(fs_info, devid,
6616 device = add_missing_dev(fs_devices, devid, dev_uuid);
6617 if (IS_ERR(device)) {
6619 "failed to add missing dev %llu: %ld",
6620 devid, PTR_ERR(device));
6621 return PTR_ERR(device);
6623 btrfs_report_missing_device(fs_info, devid, dev_uuid, false);
6625 if (!device->bdev) {
6626 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6627 btrfs_report_missing_device(fs_info,
6628 devid, dev_uuid, true);
6631 btrfs_report_missing_device(fs_info, devid,
6635 if (!device->bdev &&
6636 !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
6638 * this happens when a device that was properly setup
6639 * in the device info lists suddenly goes bad.
6640 * device->bdev is NULL, and so we have to set
6641 * device->missing to one here
6643 device->fs_devices->missing_devices++;
6644 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6647 /* Move the device to its own fs_devices */
6648 if (device->fs_devices != fs_devices) {
6649 ASSERT(test_bit(BTRFS_DEV_STATE_MISSING,
6650 &device->dev_state));
6652 list_move(&device->dev_list, &fs_devices->devices);
6653 device->fs_devices->num_devices--;
6654 fs_devices->num_devices++;
6656 device->fs_devices->missing_devices--;
6657 fs_devices->missing_devices++;
6659 device->fs_devices = fs_devices;
6663 if (device->fs_devices != fs_info->fs_devices) {
6664 BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state));
6665 if (device->generation !=
6666 btrfs_device_generation(leaf, dev_item))
6670 fill_device_from_item(leaf, dev_item, device);
6671 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
6672 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
6673 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
6674 device->fs_devices->total_rw_bytes += device->total_bytes;
6675 atomic64_add(device->total_bytes - device->bytes_used,
6676 &fs_info->free_chunk_space);
6682 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
6684 struct btrfs_root *root = fs_info->tree_root;
6685 struct btrfs_super_block *super_copy = fs_info->super_copy;
6686 struct extent_buffer *sb;
6687 struct btrfs_disk_key *disk_key;
6688 struct btrfs_chunk *chunk;
6690 unsigned long sb_array_offset;
6697 struct btrfs_key key;
6699 ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize);
6701 * This will create extent buffer of nodesize, superblock size is
6702 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6703 * overallocate but we can keep it as-is, only the first page is used.
6705 sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET);
6708 set_extent_buffer_uptodate(sb);
6709 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6711 * The sb extent buffer is artificial and just used to read the system array.
6712 * set_extent_buffer_uptodate() call does not properly mark all it's
6713 * pages up-to-date when the page is larger: extent does not cover the
6714 * whole page and consequently check_page_uptodate does not find all
6715 * the page's extents up-to-date (the hole beyond sb),
6716 * write_extent_buffer then triggers a WARN_ON.
6718 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6719 * but sb spans only this function. Add an explicit SetPageUptodate call
6720 * to silence the warning eg. on PowerPC 64.
6722 if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
6723 SetPageUptodate(sb->pages[0]);
6725 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6726 array_size = btrfs_super_sys_array_size(super_copy);
6728 array_ptr = super_copy->sys_chunk_array;
6729 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6732 while (cur_offset < array_size) {
6733 disk_key = (struct btrfs_disk_key *)array_ptr;
6734 len = sizeof(*disk_key);
6735 if (cur_offset + len > array_size)
6736 goto out_short_read;
6738 btrfs_disk_key_to_cpu(&key, disk_key);
6741 sb_array_offset += len;
6744 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6745 chunk = (struct btrfs_chunk *)sb_array_offset;
6747 * At least one btrfs_chunk with one stripe must be
6748 * present, exact stripe count check comes afterwards
6750 len = btrfs_chunk_item_size(1);
6751 if (cur_offset + len > array_size)
6752 goto out_short_read;
6754 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6757 "invalid number of stripes %u in sys_array at offset %u",
6758 num_stripes, cur_offset);
6763 type = btrfs_chunk_type(sb, chunk);
6764 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
6766 "invalid chunk type %llu in sys_array at offset %u",
6772 len = btrfs_chunk_item_size(num_stripes);
6773 if (cur_offset + len > array_size)
6774 goto out_short_read;
6776 ret = read_one_chunk(fs_info, &key, sb, chunk);
6781 "unexpected item type %u in sys_array at offset %u",
6782 (u32)key.type, cur_offset);
6787 sb_array_offset += len;
6790 clear_extent_buffer_uptodate(sb);
6791 free_extent_buffer_stale(sb);
6795 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
6797 clear_extent_buffer_uptodate(sb);
6798 free_extent_buffer_stale(sb);
6803 * Check if all chunks in the fs are OK for read-write degraded mount
6805 * If the @failing_dev is specified, it's accounted as missing.
6807 * Return true if all chunks meet the minimal RW mount requirements.
6808 * Return false if any chunk doesn't meet the minimal RW mount requirements.
6810 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
6811 struct btrfs_device *failing_dev)
6813 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
6814 struct extent_map *em;
6818 read_lock(&map_tree->map_tree.lock);
6819 em = lookup_extent_mapping(&map_tree->map_tree, 0, (u64)-1);
6820 read_unlock(&map_tree->map_tree.lock);
6821 /* No chunk at all? Return false anyway */
6827 struct map_lookup *map;
6832 map = em->map_lookup;
6834 btrfs_get_num_tolerated_disk_barrier_failures(
6836 for (i = 0; i < map->num_stripes; i++) {
6837 struct btrfs_device *dev = map->stripes[i].dev;
6839 if (!dev || !dev->bdev ||
6840 test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) ||
6841 dev->last_flush_error)
6843 else if (failing_dev && failing_dev == dev)
6846 if (missing > max_tolerated) {
6849 "chunk %llu missing %d devices, max tolerance is %d for writeable mount",
6850 em->start, missing, max_tolerated);
6851 free_extent_map(em);
6855 next_start = extent_map_end(em);
6856 free_extent_map(em);
6858 read_lock(&map_tree->map_tree.lock);
6859 em = lookup_extent_mapping(&map_tree->map_tree, next_start,
6860 (u64)(-1) - next_start);
6861 read_unlock(&map_tree->map_tree.lock);
6867 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
6869 struct btrfs_root *root = fs_info->chunk_root;
6870 struct btrfs_path *path;
6871 struct extent_buffer *leaf;
6872 struct btrfs_key key;
6873 struct btrfs_key found_key;
6878 path = btrfs_alloc_path();
6883 * uuid_mutex is needed only if we are mounting a sprout FS
6884 * otherwise we don't need it.
6886 mutex_lock(&uuid_mutex);
6887 mutex_lock(&fs_info->chunk_mutex);
6890 * Read all device items, and then all the chunk items. All
6891 * device items are found before any chunk item (their object id
6892 * is smaller than the lowest possible object id for a chunk
6893 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6895 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6898 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6902 leaf = path->nodes[0];
6903 slot = path->slots[0];
6904 if (slot >= btrfs_header_nritems(leaf)) {
6905 ret = btrfs_next_leaf(root, path);
6912 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6913 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6914 struct btrfs_dev_item *dev_item;
6915 dev_item = btrfs_item_ptr(leaf, slot,
6916 struct btrfs_dev_item);
6917 ret = read_one_dev(fs_info, leaf, dev_item);
6921 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6922 struct btrfs_chunk *chunk;
6923 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6924 ret = read_one_chunk(fs_info, &found_key, leaf, chunk);
6932 * After loading chunk tree, we've got all device information,
6933 * do another round of validation checks.
6935 if (total_dev != fs_info->fs_devices->total_devices) {
6937 "super_num_devices %llu mismatch with num_devices %llu found here",
6938 btrfs_super_num_devices(fs_info->super_copy),
6943 if (btrfs_super_total_bytes(fs_info->super_copy) <
6944 fs_info->fs_devices->total_rw_bytes) {
6946 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
6947 btrfs_super_total_bytes(fs_info->super_copy),
6948 fs_info->fs_devices->total_rw_bytes);
6954 mutex_unlock(&fs_info->chunk_mutex);
6955 mutex_unlock(&uuid_mutex);
6957 btrfs_free_path(path);
6961 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6963 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6964 struct btrfs_device *device;
6966 while (fs_devices) {
6967 mutex_lock(&fs_devices->device_list_mutex);
6968 list_for_each_entry(device, &fs_devices->devices, dev_list)
6969 device->fs_info = fs_info;
6970 mutex_unlock(&fs_devices->device_list_mutex);
6972 fs_devices = fs_devices->seed;
6976 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6980 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6981 btrfs_dev_stat_reset(dev, i);
6984 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6986 struct btrfs_key key;
6987 struct btrfs_key found_key;
6988 struct btrfs_root *dev_root = fs_info->dev_root;
6989 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6990 struct extent_buffer *eb;
6993 struct btrfs_device *device;
6994 struct btrfs_path *path = NULL;
6997 path = btrfs_alloc_path();
7003 mutex_lock(&fs_devices->device_list_mutex);
7004 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7006 struct btrfs_dev_stats_item *ptr;
7008 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7009 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7010 key.offset = device->devid;
7011 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
7013 __btrfs_reset_dev_stats(device);
7014 device->dev_stats_valid = 1;
7015 btrfs_release_path(path);
7018 slot = path->slots[0];
7019 eb = path->nodes[0];
7020 btrfs_item_key_to_cpu(eb, &found_key, slot);
7021 item_size = btrfs_item_size_nr(eb, slot);
7023 ptr = btrfs_item_ptr(eb, slot,
7024 struct btrfs_dev_stats_item);
7026 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7027 if (item_size >= (1 + i) * sizeof(__le64))
7028 btrfs_dev_stat_set(device, i,
7029 btrfs_dev_stats_value(eb, ptr, i));
7031 btrfs_dev_stat_reset(device, i);
7034 device->dev_stats_valid = 1;
7035 btrfs_dev_stat_print_on_load(device);
7036 btrfs_release_path(path);
7038 mutex_unlock(&fs_devices->device_list_mutex);
7041 btrfs_free_path(path);
7042 return ret < 0 ? ret : 0;
7045 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
7046 struct btrfs_device *device)
7048 struct btrfs_fs_info *fs_info = trans->fs_info;
7049 struct btrfs_root *dev_root = fs_info->dev_root;
7050 struct btrfs_path *path;
7051 struct btrfs_key key;
7052 struct extent_buffer *eb;
7053 struct btrfs_dev_stats_item *ptr;
7057 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7058 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7059 key.offset = device->devid;
7061 path = btrfs_alloc_path();
7064 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
7066 btrfs_warn_in_rcu(fs_info,
7067 "error %d while searching for dev_stats item for device %s",
7068 ret, rcu_str_deref(device->name));
7073 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
7074 /* need to delete old one and insert a new one */
7075 ret = btrfs_del_item(trans, dev_root, path);
7077 btrfs_warn_in_rcu(fs_info,
7078 "delete too small dev_stats item for device %s failed %d",
7079 rcu_str_deref(device->name), ret);
7086 /* need to insert a new item */
7087 btrfs_release_path(path);
7088 ret = btrfs_insert_empty_item(trans, dev_root, path,
7089 &key, sizeof(*ptr));
7091 btrfs_warn_in_rcu(fs_info,
7092 "insert dev_stats item for device %s failed %d",
7093 rcu_str_deref(device->name), ret);
7098 eb = path->nodes[0];
7099 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7100 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7101 btrfs_set_dev_stats_value(eb, ptr, i,
7102 btrfs_dev_stat_read(device, i));
7103 btrfs_mark_buffer_dirty(eb);
7106 btrfs_free_path(path);
7111 * called from commit_transaction. Writes all changed device stats to disk.
7113 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
7114 struct btrfs_fs_info *fs_info)
7116 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7117 struct btrfs_device *device;
7121 mutex_lock(&fs_devices->device_list_mutex);
7122 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7123 stats_cnt = atomic_read(&device->dev_stats_ccnt);
7124 if (!device->dev_stats_valid || stats_cnt == 0)
7129 * There is a LOAD-LOAD control dependency between the value of
7130 * dev_stats_ccnt and updating the on-disk values which requires
7131 * reading the in-memory counters. Such control dependencies
7132 * require explicit read memory barriers.
7134 * This memory barriers pairs with smp_mb__before_atomic in
7135 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full
7136 * barrier implied by atomic_xchg in
7137 * btrfs_dev_stats_read_and_reset
7141 ret = update_dev_stat_item(trans, device);
7143 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7145 mutex_unlock(&fs_devices->device_list_mutex);
7150 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7152 btrfs_dev_stat_inc(dev, index);
7153 btrfs_dev_stat_print_on_error(dev);
7156 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
7158 if (!dev->dev_stats_valid)
7160 btrfs_err_rl_in_rcu(dev->fs_info,
7161 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7162 rcu_str_deref(dev->name),
7163 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7164 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7165 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7166 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7167 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7170 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7174 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7175 if (btrfs_dev_stat_read(dev, i) != 0)
7177 if (i == BTRFS_DEV_STAT_VALUES_MAX)
7178 return; /* all values == 0, suppress message */
7180 btrfs_info_in_rcu(dev->fs_info,
7181 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7182 rcu_str_deref(dev->name),
7183 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7184 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7185 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7186 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7187 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7190 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
7191 struct btrfs_ioctl_get_dev_stats *stats)
7193 struct btrfs_device *dev;
7194 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7197 mutex_lock(&fs_devices->device_list_mutex);
7198 dev = btrfs_find_device(fs_info, stats->devid, NULL, NULL);
7199 mutex_unlock(&fs_devices->device_list_mutex);
7202 btrfs_warn(fs_info, "get dev_stats failed, device not found");
7204 } else if (!dev->dev_stats_valid) {
7205 btrfs_warn(fs_info, "get dev_stats failed, not yet valid");
7207 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7208 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7209 if (stats->nr_items > i)
7211 btrfs_dev_stat_read_and_reset(dev, i);
7213 btrfs_dev_stat_reset(dev, i);
7216 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7217 if (stats->nr_items > i)
7218 stats->values[i] = btrfs_dev_stat_read(dev, i);
7220 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7221 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7225 void btrfs_scratch_superblocks(struct block_device *bdev, const char *device_path)
7227 struct buffer_head *bh;
7228 struct btrfs_super_block *disk_super;
7234 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX;
7237 if (btrfs_read_dev_one_super(bdev, copy_num, &bh))
7240 disk_super = (struct btrfs_super_block *)bh->b_data;
7242 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
7243 set_buffer_dirty(bh);
7244 sync_dirty_buffer(bh);
7248 /* Notify udev that device has changed */
7249 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
7251 /* Update ctime/mtime for device path for libblkid */
7252 update_dev_time(device_path);
7256 * Update the size of all devices, which is used for writing out the
7259 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
7261 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7262 struct btrfs_device *curr, *next;
7264 if (list_empty(&fs_devices->resized_devices))
7267 mutex_lock(&fs_devices->device_list_mutex);
7268 mutex_lock(&fs_info->chunk_mutex);
7269 list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
7271 list_del_init(&curr->resized_list);
7272 curr->commit_total_bytes = curr->disk_total_bytes;
7274 mutex_unlock(&fs_info->chunk_mutex);
7275 mutex_unlock(&fs_devices->device_list_mutex);
7278 /* Must be invoked during the transaction commit */
7279 void btrfs_update_commit_device_bytes_used(struct btrfs_transaction *trans)
7281 struct btrfs_fs_info *fs_info = trans->fs_info;
7282 struct extent_map *em;
7283 struct map_lookup *map;
7284 struct btrfs_device *dev;
7287 if (list_empty(&trans->pending_chunks))
7290 /* In order to kick the device replace finish process */
7291 mutex_lock(&fs_info->chunk_mutex);
7292 list_for_each_entry(em, &trans->pending_chunks, list) {
7293 map = em->map_lookup;
7295 for (i = 0; i < map->num_stripes; i++) {
7296 dev = map->stripes[i].dev;
7297 dev->commit_bytes_used = dev->bytes_used;
7300 mutex_unlock(&fs_info->chunk_mutex);
7303 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
7305 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7306 while (fs_devices) {
7307 fs_devices->fs_info = fs_info;
7308 fs_devices = fs_devices->seed;
7312 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
7314 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7315 while (fs_devices) {
7316 fs_devices->fs_info = NULL;
7317 fs_devices = fs_devices->seed;
7322 * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10.
7324 int btrfs_bg_type_to_factor(u64 flags)
7326 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
7327 BTRFS_BLOCK_GROUP_RAID10))