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/iocontext.h>
12 #include <linux/capability.h>
13 #include <linux/ratelimit.h>
14 #include <linux/kthread.h>
15 #include <linux/raid/pq.h>
16 #include <linux/semaphore.h>
17 #include <linux/uuid.h>
18 #include <linux/list_sort.h>
19 #include <asm/div64.h>
21 #include "extent_map.h"
23 #include "transaction.h"
24 #include "print-tree.h"
27 #include "async-thread.h"
28 #include "check-integrity.h"
29 #include "rcu-string.h"
31 #include "dev-replace.h"
34 const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
35 [BTRFS_RAID_RAID10] = {
38 .devs_max = 0, /* 0 == as many as possible */
40 .tolerated_failures = 1,
43 .raid_name = "raid10",
44 .bg_flag = BTRFS_BLOCK_GROUP_RAID10,
45 .mindev_error = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET,
47 [BTRFS_RAID_RAID1] = {
52 .tolerated_failures = 1,
56 .bg_flag = BTRFS_BLOCK_GROUP_RAID1,
57 .mindev_error = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET,
64 .tolerated_failures = 0,
68 .bg_flag = BTRFS_BLOCK_GROUP_DUP,
71 [BTRFS_RAID_RAID0] = {
76 .tolerated_failures = 0,
80 .bg_flag = BTRFS_BLOCK_GROUP_RAID0,
83 [BTRFS_RAID_SINGLE] = {
88 .tolerated_failures = 0,
91 .raid_name = "single",
95 [BTRFS_RAID_RAID5] = {
100 .tolerated_failures = 1,
103 .raid_name = "raid5",
104 .bg_flag = BTRFS_BLOCK_GROUP_RAID5,
105 .mindev_error = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET,
107 [BTRFS_RAID_RAID6] = {
112 .tolerated_failures = 2,
115 .raid_name = "raid6",
116 .bg_flag = BTRFS_BLOCK_GROUP_RAID6,
117 .mindev_error = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET,
121 const char *get_raid_name(enum btrfs_raid_types type)
123 if (type >= BTRFS_NR_RAID_TYPES)
126 return btrfs_raid_array[type].raid_name;
129 static int init_first_rw_device(struct btrfs_trans_handle *trans,
130 struct btrfs_fs_info *fs_info);
131 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info);
132 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
133 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
134 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
135 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
136 enum btrfs_map_op op,
137 u64 logical, u64 *length,
138 struct btrfs_bio **bbio_ret,
139 int mirror_num, int need_raid_map);
145 * There are several mutexes that protect manipulation of devices and low-level
146 * structures like chunks but not block groups, extents or files
148 * uuid_mutex (global lock)
149 * ------------------------
150 * protects the fs_uuids list that tracks all per-fs fs_devices, resulting from
151 * the SCAN_DEV ioctl registration or from mount either implicitly (the first
152 * device) or requested by the device= mount option
154 * the mutex can be very coarse and can cover long-running operations
156 * protects: updates to fs_devices counters like missing devices, rw devices,
157 * seeding, structure cloning, openning/closing devices at mount/umount time
159 * global::fs_devs - add, remove, updates to the global list
161 * does not protect: manipulation of the fs_devices::devices list!
163 * btrfs_device::name - renames (write side), read is RCU
165 * fs_devices::device_list_mutex (per-fs, with RCU)
166 * ------------------------------------------------
167 * protects updates to fs_devices::devices, ie. adding and deleting
169 * simple list traversal with read-only actions can be done with RCU protection
171 * may be used to exclude some operations from running concurrently without any
172 * modifications to the list (see write_all_supers)
176 * protects balance structures (status, state) and context accessed from
177 * several places (internally, ioctl)
181 * protects chunks, adding or removing during allocation, trim or when a new
182 * device is added/removed
186 * a big lock that is held by the cleaner thread and prevents running subvolume
187 * cleaning together with relocation or delayed iputs
200 * Exclusive operations, BTRFS_FS_EXCL_OP
201 * ======================================
203 * Maintains the exclusivity of the following operations that apply to the
204 * whole filesystem and cannot run in parallel.
209 * - Device replace (*)
212 * The device operations (as above) can be in one of the following states:
218 * Only device operations marked with (*) can go into the Paused state for the
221 * - ioctl (only Balance can be Paused through ioctl)
222 * - filesystem remounted as read-only
223 * - filesystem unmounted and mounted as read-only
224 * - system power-cycle and filesystem mounted as read-only
225 * - filesystem or device errors leading to forced read-only
227 * BTRFS_FS_EXCL_OP flag is set and cleared using atomic operations.
228 * During the course of Paused state, the BTRFS_FS_EXCL_OP remains set.
229 * A device operation in Paused or Running state can be canceled or resumed
230 * either by ioctl (Balance only) or when remounted as read-write.
231 * BTRFS_FS_EXCL_OP flag is cleared when the device operation is canceled or
235 DEFINE_MUTEX(uuid_mutex);
236 static LIST_HEAD(fs_uuids);
237 struct list_head *btrfs_get_fs_uuids(void)
243 * alloc_fs_devices - allocate struct btrfs_fs_devices
244 * @fsid: if not NULL, copy the uuid to fs_devices::fsid
246 * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR().
247 * The returned struct is not linked onto any lists and can be destroyed with
248 * kfree() right away.
250 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
252 struct btrfs_fs_devices *fs_devs;
254 fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL);
256 return ERR_PTR(-ENOMEM);
258 mutex_init(&fs_devs->device_list_mutex);
260 INIT_LIST_HEAD(&fs_devs->devices);
261 INIT_LIST_HEAD(&fs_devs->resized_devices);
262 INIT_LIST_HEAD(&fs_devs->alloc_list);
263 INIT_LIST_HEAD(&fs_devs->fs_list);
265 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
270 void btrfs_free_device(struct btrfs_device *device)
272 rcu_string_free(device->name);
273 bio_put(device->flush_bio);
277 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
279 struct btrfs_device *device;
280 WARN_ON(fs_devices->opened);
281 while (!list_empty(&fs_devices->devices)) {
282 device = list_entry(fs_devices->devices.next,
283 struct btrfs_device, dev_list);
284 list_del(&device->dev_list);
285 btrfs_free_device(device);
290 static void btrfs_kobject_uevent(struct block_device *bdev,
291 enum kobject_action action)
295 ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
297 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
299 kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
300 &disk_to_dev(bdev->bd_disk)->kobj);
303 void __exit btrfs_cleanup_fs_uuids(void)
305 struct btrfs_fs_devices *fs_devices;
307 while (!list_empty(&fs_uuids)) {
308 fs_devices = list_entry(fs_uuids.next,
309 struct btrfs_fs_devices, fs_list);
310 list_del(&fs_devices->fs_list);
311 free_fs_devices(fs_devices);
316 * Returns a pointer to a new btrfs_device on success; ERR_PTR() on error.
317 * Returned struct is not linked onto any lists and must be destroyed using
320 static struct btrfs_device *__alloc_device(void)
322 struct btrfs_device *dev;
324 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
326 return ERR_PTR(-ENOMEM);
329 * Preallocate a bio that's always going to be used for flushing device
330 * barriers and matches the device lifespan
332 dev->flush_bio = bio_alloc_bioset(GFP_KERNEL, 0, NULL);
333 if (!dev->flush_bio) {
335 return ERR_PTR(-ENOMEM);
338 INIT_LIST_HEAD(&dev->dev_list);
339 INIT_LIST_HEAD(&dev->dev_alloc_list);
340 INIT_LIST_HEAD(&dev->resized_list);
342 spin_lock_init(&dev->io_lock);
344 atomic_set(&dev->reada_in_flight, 0);
345 atomic_set(&dev->dev_stats_ccnt, 0);
346 btrfs_device_data_ordered_init(dev);
347 INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
348 INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
354 * Find a device specified by @devid or @uuid in the list of @fs_devices, or
357 * If devid and uuid are both specified, the match must be exact, otherwise
358 * only devid is used.
360 static struct btrfs_device *find_device(struct btrfs_fs_devices *fs_devices,
361 u64 devid, const u8 *uuid)
363 struct btrfs_device *dev;
365 list_for_each_entry(dev, &fs_devices->devices, dev_list) {
366 if (dev->devid == devid &&
367 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
374 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
376 struct btrfs_fs_devices *fs_devices;
378 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
379 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
386 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
387 int flush, struct block_device **bdev,
388 struct buffer_head **bh)
392 *bdev = blkdev_get_by_path(device_path, flags, holder);
395 ret = PTR_ERR(*bdev);
400 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
401 ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE);
403 blkdev_put(*bdev, flags);
406 invalidate_bdev(*bdev);
407 *bh = btrfs_read_dev_super(*bdev);
410 blkdev_put(*bdev, flags);
422 static void requeue_list(struct btrfs_pending_bios *pending_bios,
423 struct bio *head, struct bio *tail)
426 struct bio *old_head;
428 old_head = pending_bios->head;
429 pending_bios->head = head;
430 if (pending_bios->tail)
431 tail->bi_next = old_head;
433 pending_bios->tail = tail;
437 * we try to collect pending bios for a device so we don't get a large
438 * number of procs sending bios down to the same device. This greatly
439 * improves the schedulers ability to collect and merge the bios.
441 * But, it also turns into a long list of bios to process and that is sure
442 * to eventually make the worker thread block. The solution here is to
443 * make some progress and then put this work struct back at the end of
444 * the list if the block device is congested. This way, multiple devices
445 * can make progress from a single worker thread.
447 static noinline void run_scheduled_bios(struct btrfs_device *device)
449 struct btrfs_fs_info *fs_info = device->fs_info;
451 struct backing_dev_info *bdi;
452 struct btrfs_pending_bios *pending_bios;
456 unsigned long num_run;
457 unsigned long batch_run = 0;
458 unsigned long last_waited = 0;
460 int sync_pending = 0;
461 struct blk_plug plug;
464 * this function runs all the bios we've collected for
465 * a particular device. We don't want to wander off to
466 * another device without first sending all of these down.
467 * So, setup a plug here and finish it off before we return
469 blk_start_plug(&plug);
471 bdi = device->bdev->bd_bdi;
474 spin_lock(&device->io_lock);
479 /* take all the bios off the list at once and process them
480 * later on (without the lock held). But, remember the
481 * tail and other pointers so the bios can be properly reinserted
482 * into the list if we hit congestion
484 if (!force_reg && device->pending_sync_bios.head) {
485 pending_bios = &device->pending_sync_bios;
488 pending_bios = &device->pending_bios;
492 pending = pending_bios->head;
493 tail = pending_bios->tail;
494 WARN_ON(pending && !tail);
497 * if pending was null this time around, no bios need processing
498 * at all and we can stop. Otherwise it'll loop back up again
499 * and do an additional check so no bios are missed.
501 * device->running_pending is used to synchronize with the
504 if (device->pending_sync_bios.head == NULL &&
505 device->pending_bios.head == NULL) {
507 device->running_pending = 0;
510 device->running_pending = 1;
513 pending_bios->head = NULL;
514 pending_bios->tail = NULL;
516 spin_unlock(&device->io_lock);
521 /* we want to work on both lists, but do more bios on the
522 * sync list than the regular list
525 pending_bios != &device->pending_sync_bios &&
526 device->pending_sync_bios.head) ||
527 (num_run > 64 && pending_bios == &device->pending_sync_bios &&
528 device->pending_bios.head)) {
529 spin_lock(&device->io_lock);
530 requeue_list(pending_bios, pending, tail);
535 pending = pending->bi_next;
538 BUG_ON(atomic_read(&cur->__bi_cnt) == 0);
541 * if we're doing the sync list, record that our
542 * plug has some sync requests on it
544 * If we're doing the regular list and there are
545 * sync requests sitting around, unplug before
548 if (pending_bios == &device->pending_sync_bios) {
550 } else if (sync_pending) {
551 blk_finish_plug(&plug);
552 blk_start_plug(&plug);
556 btrfsic_submit_bio(cur);
563 * we made progress, there is more work to do and the bdi
564 * is now congested. Back off and let other work structs
567 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
568 fs_info->fs_devices->open_devices > 1) {
569 struct io_context *ioc;
571 ioc = current->io_context;
574 * the main goal here is that we don't want to
575 * block if we're going to be able to submit
576 * more requests without blocking.
578 * This code does two great things, it pokes into
579 * the elevator code from a filesystem _and_
580 * it makes assumptions about how batching works.
582 if (ioc && ioc->nr_batch_requests > 0 &&
583 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
585 ioc->last_waited == last_waited)) {
587 * we want to go through our batch of
588 * requests and stop. So, we copy out
589 * the ioc->last_waited time and test
590 * against it before looping
592 last_waited = ioc->last_waited;
596 spin_lock(&device->io_lock);
597 requeue_list(pending_bios, pending, tail);
598 device->running_pending = 1;
600 spin_unlock(&device->io_lock);
601 btrfs_queue_work(fs_info->submit_workers,
611 spin_lock(&device->io_lock);
612 if (device->pending_bios.head || device->pending_sync_bios.head)
614 spin_unlock(&device->io_lock);
617 blk_finish_plug(&plug);
620 static void pending_bios_fn(struct btrfs_work *work)
622 struct btrfs_device *device;
624 device = container_of(work, struct btrfs_device, work);
625 run_scheduled_bios(device);
629 * Search and remove all stale (devices which are not mounted) devices.
630 * When both inputs are NULL, it will search and release all stale devices.
631 * path: Optional. When provided will it release all unmounted devices
632 * matching this path only.
633 * skip_dev: Optional. Will skip this device when searching for the stale
636 static void btrfs_free_stale_devices(const char *path,
637 struct btrfs_device *skip_dev)
639 struct btrfs_fs_devices *fs_devs, *tmp_fs_devs;
640 struct btrfs_device *dev, *tmp_dev;
642 list_for_each_entry_safe(fs_devs, tmp_fs_devs, &fs_uuids, fs_list) {
647 list_for_each_entry_safe(dev, tmp_dev,
648 &fs_devs->devices, dev_list) {
651 if (skip_dev && skip_dev == dev)
653 if (path && !dev->name)
658 not_found = strcmp(rcu_str_deref(dev->name),
664 /* delete the stale device */
665 if (fs_devs->num_devices == 1) {
666 btrfs_sysfs_remove_fsid(fs_devs);
667 list_del(&fs_devs->fs_list);
668 free_fs_devices(fs_devs);
671 fs_devs->num_devices--;
672 list_del(&dev->dev_list);
673 btrfs_free_device(dev);
679 static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices,
680 struct btrfs_device *device, fmode_t flags,
683 struct request_queue *q;
684 struct block_device *bdev;
685 struct buffer_head *bh;
686 struct btrfs_super_block *disk_super;
695 ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
700 disk_super = (struct btrfs_super_block *)bh->b_data;
701 devid = btrfs_stack_device_id(&disk_super->dev_item);
702 if (devid != device->devid)
705 if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE))
708 device->generation = btrfs_super_generation(disk_super);
710 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
711 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
712 fs_devices->seeding = 1;
714 if (bdev_read_only(bdev))
715 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
717 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
720 q = bdev_get_queue(bdev);
721 if (!blk_queue_nonrot(q))
722 fs_devices->rotating = 1;
725 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
726 device->mode = flags;
728 fs_devices->open_devices++;
729 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
730 device->devid != BTRFS_DEV_REPLACE_DEVID) {
731 fs_devices->rw_devices++;
732 list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list);
740 blkdev_put(bdev, flags);
746 * Add new device to list of registered devices
749 * device pointer which was just added or updated when successful
750 * error pointer when failed
752 static noinline struct btrfs_device *device_list_add(const char *path,
753 struct btrfs_super_block *disk_super)
755 struct btrfs_device *device;
756 struct btrfs_fs_devices *fs_devices;
757 struct rcu_string *name;
758 u64 found_transid = btrfs_super_generation(disk_super);
759 u64 devid = btrfs_stack_device_id(&disk_super->dev_item);
761 fs_devices = find_fsid(disk_super->fsid);
763 fs_devices = alloc_fs_devices(disk_super->fsid);
764 if (IS_ERR(fs_devices))
765 return ERR_CAST(fs_devices);
767 list_add(&fs_devices->fs_list, &fs_uuids);
771 device = find_device(fs_devices, devid,
772 disk_super->dev_item.uuid);
776 if (fs_devices->opened)
777 return ERR_PTR(-EBUSY);
779 device = btrfs_alloc_device(NULL, &devid,
780 disk_super->dev_item.uuid);
781 if (IS_ERR(device)) {
782 /* we can safely leave the fs_devices entry around */
786 name = rcu_string_strdup(path, GFP_NOFS);
788 btrfs_free_device(device);
789 return ERR_PTR(-ENOMEM);
791 rcu_assign_pointer(device->name, name);
793 mutex_lock(&fs_devices->device_list_mutex);
794 list_add_rcu(&device->dev_list, &fs_devices->devices);
795 fs_devices->num_devices++;
796 mutex_unlock(&fs_devices->device_list_mutex);
798 device->fs_devices = fs_devices;
799 btrfs_free_stale_devices(path, device);
801 if (disk_super->label[0])
802 pr_info("BTRFS: device label %s devid %llu transid %llu %s\n",
803 disk_super->label, devid, found_transid, path);
805 pr_info("BTRFS: device fsid %pU devid %llu transid %llu %s\n",
806 disk_super->fsid, devid, found_transid, path);
808 } else if (!device->name || strcmp(device->name->str, path)) {
810 * When FS is already mounted.
811 * 1. If you are here and if the device->name is NULL that
812 * means this device was missing at time of FS mount.
813 * 2. If you are here and if the device->name is different
814 * from 'path' that means either
815 * a. The same device disappeared and reappeared with
817 * b. The missing-disk-which-was-replaced, has
820 * We must allow 1 and 2a above. But 2b would be a spurious
823 * Further in case of 1 and 2a above, the disk at 'path'
824 * would have missed some transaction when it was away and
825 * in case of 2a the stale bdev has to be updated as well.
826 * 2b must not be allowed at all time.
830 * For now, we do allow update to btrfs_fs_device through the
831 * btrfs dev scan cli after FS has been mounted. We're still
832 * tracking a problem where systems fail mount by subvolume id
833 * when we reject replacement on a mounted FS.
835 if (!fs_devices->opened && found_transid < device->generation) {
837 * That is if the FS is _not_ mounted and if you
838 * are here, that means there is more than one
839 * disk with same uuid and devid.We keep the one
840 * with larger generation number or the last-in if
841 * generation are equal.
843 return ERR_PTR(-EEXIST);
846 name = rcu_string_strdup(path, GFP_NOFS);
848 return ERR_PTR(-ENOMEM);
849 rcu_string_free(device->name);
850 rcu_assign_pointer(device->name, name);
851 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
852 fs_devices->missing_devices--;
853 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
858 * Unmount does not free the btrfs_device struct but would zero
859 * generation along with most of the other members. So just update
860 * it back. We need it to pick the disk with largest generation
863 if (!fs_devices->opened)
864 device->generation = found_transid;
866 fs_devices->total_devices = btrfs_super_num_devices(disk_super);
871 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
873 struct btrfs_fs_devices *fs_devices;
874 struct btrfs_device *device;
875 struct btrfs_device *orig_dev;
877 fs_devices = alloc_fs_devices(orig->fsid);
878 if (IS_ERR(fs_devices))
881 mutex_lock(&orig->device_list_mutex);
882 fs_devices->total_devices = orig->total_devices;
884 /* We have held the volume lock, it is safe to get the devices. */
885 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
886 struct rcu_string *name;
888 device = btrfs_alloc_device(NULL, &orig_dev->devid,
894 * This is ok to do without rcu read locked because we hold the
895 * uuid mutex so nothing we touch in here is going to disappear.
897 if (orig_dev->name) {
898 name = rcu_string_strdup(orig_dev->name->str,
901 btrfs_free_device(device);
904 rcu_assign_pointer(device->name, name);
907 list_add(&device->dev_list, &fs_devices->devices);
908 device->fs_devices = fs_devices;
909 fs_devices->num_devices++;
911 mutex_unlock(&orig->device_list_mutex);
914 mutex_unlock(&orig->device_list_mutex);
915 free_fs_devices(fs_devices);
916 return ERR_PTR(-ENOMEM);
920 * After we have read the system tree and know devids belonging to
921 * this filesystem, remove the device which does not belong there.
923 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices, int step)
925 struct btrfs_device *device, *next;
926 struct btrfs_device *latest_dev = NULL;
928 mutex_lock(&uuid_mutex);
930 /* This is the initialized path, it is safe to release the devices. */
931 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
932 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
933 &device->dev_state)) {
934 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
935 &device->dev_state) &&
937 device->generation > latest_dev->generation)) {
943 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
945 * In the first step, keep the device which has
946 * the correct fsid and the devid that is used
947 * for the dev_replace procedure.
948 * In the second step, the dev_replace state is
949 * read from the device tree and it is known
950 * whether the procedure is really active or
951 * not, which means whether this device is
952 * used or whether it should be removed.
954 if (step == 0 || test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
955 &device->dev_state)) {
960 blkdev_put(device->bdev, device->mode);
962 fs_devices->open_devices--;
964 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
965 list_del_init(&device->dev_alloc_list);
966 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
967 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
969 fs_devices->rw_devices--;
971 list_del_init(&device->dev_list);
972 fs_devices->num_devices--;
973 btrfs_free_device(device);
976 if (fs_devices->seed) {
977 fs_devices = fs_devices->seed;
981 fs_devices->latest_bdev = latest_dev->bdev;
983 mutex_unlock(&uuid_mutex);
986 static void free_device_rcu(struct rcu_head *head)
988 struct btrfs_device *device;
990 device = container_of(head, struct btrfs_device, rcu);
991 btrfs_free_device(device);
994 static void btrfs_close_bdev(struct btrfs_device *device)
999 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1000 sync_blockdev(device->bdev);
1001 invalidate_bdev(device->bdev);
1004 blkdev_put(device->bdev, device->mode);
1007 static void btrfs_prepare_close_one_device(struct btrfs_device *device)
1009 struct btrfs_fs_devices *fs_devices = device->fs_devices;
1010 struct btrfs_device *new_device;
1011 struct rcu_string *name;
1014 fs_devices->open_devices--;
1016 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
1017 device->devid != BTRFS_DEV_REPLACE_DEVID) {
1018 list_del_init(&device->dev_alloc_list);
1019 fs_devices->rw_devices--;
1022 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
1023 fs_devices->missing_devices--;
1025 new_device = btrfs_alloc_device(NULL, &device->devid,
1027 BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
1029 /* Safe because we are under uuid_mutex */
1031 name = rcu_string_strdup(device->name->str, GFP_NOFS);
1032 BUG_ON(!name); /* -ENOMEM */
1033 rcu_assign_pointer(new_device->name, name);
1036 list_replace_rcu(&device->dev_list, &new_device->dev_list);
1037 new_device->fs_devices = device->fs_devices;
1040 static int close_fs_devices(struct btrfs_fs_devices *fs_devices)
1042 struct btrfs_device *device, *tmp;
1043 struct list_head pending_put;
1045 INIT_LIST_HEAD(&pending_put);
1047 if (--fs_devices->opened > 0)
1050 mutex_lock(&fs_devices->device_list_mutex);
1051 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
1052 btrfs_prepare_close_one_device(device);
1053 list_add(&device->dev_list, &pending_put);
1055 mutex_unlock(&fs_devices->device_list_mutex);
1058 * btrfs_show_devname() is using the device_list_mutex,
1059 * sometimes call to blkdev_put() leads vfs calling
1060 * into this func. So do put outside of device_list_mutex,
1063 while (!list_empty(&pending_put)) {
1064 device = list_first_entry(&pending_put,
1065 struct btrfs_device, dev_list);
1066 list_del(&device->dev_list);
1067 btrfs_close_bdev(device);
1068 call_rcu(&device->rcu, free_device_rcu);
1071 WARN_ON(fs_devices->open_devices);
1072 WARN_ON(fs_devices->rw_devices);
1073 fs_devices->opened = 0;
1074 fs_devices->seeding = 0;
1079 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
1081 struct btrfs_fs_devices *seed_devices = NULL;
1084 mutex_lock(&uuid_mutex);
1085 ret = close_fs_devices(fs_devices);
1086 if (!fs_devices->opened) {
1087 seed_devices = fs_devices->seed;
1088 fs_devices->seed = NULL;
1090 mutex_unlock(&uuid_mutex);
1092 while (seed_devices) {
1093 fs_devices = seed_devices;
1094 seed_devices = fs_devices->seed;
1095 close_fs_devices(fs_devices);
1096 free_fs_devices(fs_devices);
1101 static int open_fs_devices(struct btrfs_fs_devices *fs_devices,
1102 fmode_t flags, void *holder)
1104 struct btrfs_device *device;
1105 struct btrfs_device *latest_dev = NULL;
1108 flags |= FMODE_EXCL;
1110 list_for_each_entry(device, &fs_devices->devices, dev_list) {
1111 /* Just open everything we can; ignore failures here */
1112 if (btrfs_open_one_device(fs_devices, device, flags, holder))
1116 device->generation > latest_dev->generation)
1117 latest_dev = device;
1119 if (fs_devices->open_devices == 0) {
1123 fs_devices->opened = 1;
1124 fs_devices->latest_bdev = latest_dev->bdev;
1125 fs_devices->total_rw_bytes = 0;
1130 static int devid_cmp(void *priv, struct list_head *a, struct list_head *b)
1132 struct btrfs_device *dev1, *dev2;
1134 dev1 = list_entry(a, struct btrfs_device, dev_list);
1135 dev2 = list_entry(b, struct btrfs_device, dev_list);
1137 if (dev1->devid < dev2->devid)
1139 else if (dev1->devid > dev2->devid)
1144 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1145 fmode_t flags, void *holder)
1149 mutex_lock(&fs_devices->device_list_mutex);
1150 if (fs_devices->opened) {
1151 fs_devices->opened++;
1154 list_sort(NULL, &fs_devices->devices, devid_cmp);
1155 ret = open_fs_devices(fs_devices, flags, holder);
1157 mutex_unlock(&fs_devices->device_list_mutex);
1162 static void btrfs_release_disk_super(struct page *page)
1168 static int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr,
1170 struct btrfs_super_block **disk_super)
1175 /* make sure our super fits in the device */
1176 if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1179 /* make sure our super fits in the page */
1180 if (sizeof(**disk_super) > PAGE_SIZE)
1183 /* make sure our super doesn't straddle pages on disk */
1184 index = bytenr >> PAGE_SHIFT;
1185 if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index)
1188 /* pull in the page with our super */
1189 *page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
1192 if (IS_ERR_OR_NULL(*page))
1197 /* align our pointer to the offset of the super block */
1198 *disk_super = p + (bytenr & ~PAGE_MASK);
1200 if (btrfs_super_bytenr(*disk_super) != bytenr ||
1201 btrfs_super_magic(*disk_super) != BTRFS_MAGIC) {
1202 btrfs_release_disk_super(*page);
1206 if ((*disk_super)->label[0] &&
1207 (*disk_super)->label[BTRFS_LABEL_SIZE - 1])
1208 (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0';
1214 * Look for a btrfs signature on a device. This may be called out of the mount path
1215 * and we are not allowed to call set_blocksize during the scan. The superblock
1216 * is read via pagecache
1218 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
1219 struct btrfs_fs_devices **fs_devices_ret)
1221 struct btrfs_super_block *disk_super;
1222 struct btrfs_device *device;
1223 struct block_device *bdev;
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 PTR_ERR(bdev);
1241 if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super)) {
1243 goto error_bdev_put;
1246 mutex_lock(&uuid_mutex);
1247 device = device_list_add(path, disk_super);
1249 ret = PTR_ERR(device);
1251 *fs_devices_ret = device->fs_devices;
1252 mutex_unlock(&uuid_mutex);
1254 btrfs_release_disk_super(page);
1257 blkdev_put(bdev, flags);
1262 /* helper to account the used device space in the range */
1263 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
1264 u64 end, u64 *length)
1266 struct btrfs_key key;
1267 struct btrfs_root *root = device->fs_info->dev_root;
1268 struct btrfs_dev_extent *dev_extent;
1269 struct btrfs_path *path;
1273 struct extent_buffer *l;
1277 if (start >= device->total_bytes ||
1278 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
1281 path = btrfs_alloc_path();
1284 path->reada = READA_FORWARD;
1286 key.objectid = device->devid;
1288 key.type = BTRFS_DEV_EXTENT_KEY;
1290 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1294 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1301 slot = path->slots[0];
1302 if (slot >= btrfs_header_nritems(l)) {
1303 ret = btrfs_next_leaf(root, path);
1311 btrfs_item_key_to_cpu(l, &key, slot);
1313 if (key.objectid < device->devid)
1316 if (key.objectid > device->devid)
1319 if (key.type != BTRFS_DEV_EXTENT_KEY)
1322 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1323 extent_end = key.offset + btrfs_dev_extent_length(l,
1325 if (key.offset <= start && extent_end > end) {
1326 *length = end - start + 1;
1328 } else if (key.offset <= start && extent_end > start)
1329 *length += extent_end - start;
1330 else if (key.offset > start && extent_end <= end)
1331 *length += extent_end - key.offset;
1332 else if (key.offset > start && key.offset <= end) {
1333 *length += end - key.offset + 1;
1335 } else if (key.offset > end)
1343 btrfs_free_path(path);
1347 static int contains_pending_extent(struct btrfs_transaction *transaction,
1348 struct btrfs_device *device,
1349 u64 *start, u64 len)
1351 struct btrfs_fs_info *fs_info = device->fs_info;
1352 struct extent_map *em;
1353 struct list_head *search_list = &fs_info->pinned_chunks;
1355 u64 physical_start = *start;
1358 search_list = &transaction->pending_chunks;
1360 list_for_each_entry(em, search_list, list) {
1361 struct map_lookup *map;
1364 map = em->map_lookup;
1365 for (i = 0; i < map->num_stripes; i++) {
1368 if (map->stripes[i].dev != device)
1370 if (map->stripes[i].physical >= physical_start + len ||
1371 map->stripes[i].physical + em->orig_block_len <=
1375 * Make sure that while processing the pinned list we do
1376 * not override our *start with a lower value, because
1377 * we can have pinned chunks that fall within this
1378 * device hole and that have lower physical addresses
1379 * than the pending chunks we processed before. If we
1380 * do not take this special care we can end up getting
1381 * 2 pending chunks that start at the same physical
1382 * device offsets because the end offset of a pinned
1383 * chunk can be equal to the start offset of some
1386 end = map->stripes[i].physical + em->orig_block_len;
1393 if (search_list != &fs_info->pinned_chunks) {
1394 search_list = &fs_info->pinned_chunks;
1403 * find_free_dev_extent_start - find free space in the specified device
1404 * @device: the device which we search the free space in
1405 * @num_bytes: the size of the free space that we need
1406 * @search_start: the position from which to begin the search
1407 * @start: store the start of the free space.
1408 * @len: the size of the free space. that we find, or the size
1409 * of the max free space if we don't find suitable free space
1411 * this uses a pretty simple search, the expectation is that it is
1412 * called very infrequently and that a given device has a small number
1415 * @start is used to store the start of the free space if we find. But if we
1416 * don't find suitable free space, it will be used to store the start position
1417 * of the max free space.
1419 * @len is used to store the size of the free space that we find.
1420 * But if we don't find suitable free space, it is used to store the size of
1421 * the max free space.
1423 int find_free_dev_extent_start(struct btrfs_transaction *transaction,
1424 struct btrfs_device *device, u64 num_bytes,
1425 u64 search_start, u64 *start, u64 *len)
1427 struct btrfs_fs_info *fs_info = device->fs_info;
1428 struct btrfs_root *root = fs_info->dev_root;
1429 struct btrfs_key key;
1430 struct btrfs_dev_extent *dev_extent;
1431 struct btrfs_path *path;
1436 u64 search_end = device->total_bytes;
1439 struct extent_buffer *l;
1442 * We don't want to overwrite the superblock on the drive nor any area
1443 * used by the boot loader (grub for example), so we make sure to start
1444 * at an offset of at least 1MB.
1446 search_start = max_t(u64, search_start, SZ_1M);
1448 path = btrfs_alloc_path();
1452 max_hole_start = search_start;
1456 if (search_start >= search_end ||
1457 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1462 path->reada = READA_FORWARD;
1463 path->search_commit_root = 1;
1464 path->skip_locking = 1;
1466 key.objectid = device->devid;
1467 key.offset = search_start;
1468 key.type = BTRFS_DEV_EXTENT_KEY;
1470 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1474 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1481 slot = path->slots[0];
1482 if (slot >= btrfs_header_nritems(l)) {
1483 ret = btrfs_next_leaf(root, path);
1491 btrfs_item_key_to_cpu(l, &key, slot);
1493 if (key.objectid < device->devid)
1496 if (key.objectid > device->devid)
1499 if (key.type != BTRFS_DEV_EXTENT_KEY)
1502 if (key.offset > search_start) {
1503 hole_size = key.offset - search_start;
1506 * Have to check before we set max_hole_start, otherwise
1507 * we could end up sending back this offset anyway.
1509 if (contains_pending_extent(transaction, device,
1512 if (key.offset >= search_start) {
1513 hole_size = key.offset - search_start;
1520 if (hole_size > max_hole_size) {
1521 max_hole_start = search_start;
1522 max_hole_size = hole_size;
1526 * If this free space is greater than which we need,
1527 * it must be the max free space that we have found
1528 * until now, so max_hole_start must point to the start
1529 * of this free space and the length of this free space
1530 * is stored in max_hole_size. Thus, we return
1531 * max_hole_start and max_hole_size and go back to the
1534 if (hole_size >= num_bytes) {
1540 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1541 extent_end = key.offset + btrfs_dev_extent_length(l,
1543 if (extent_end > search_start)
1544 search_start = extent_end;
1551 * At this point, search_start should be the end of
1552 * allocated dev extents, and when shrinking the device,
1553 * search_end may be smaller than search_start.
1555 if (search_end > search_start) {
1556 hole_size = search_end - search_start;
1558 if (contains_pending_extent(transaction, device, &search_start,
1560 btrfs_release_path(path);
1564 if (hole_size > max_hole_size) {
1565 max_hole_start = search_start;
1566 max_hole_size = hole_size;
1571 if (max_hole_size < num_bytes)
1577 btrfs_free_path(path);
1578 *start = max_hole_start;
1580 *len = max_hole_size;
1584 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1585 struct btrfs_device *device, u64 num_bytes,
1586 u64 *start, u64 *len)
1588 /* FIXME use last free of some kind */
1589 return find_free_dev_extent_start(trans->transaction, device,
1590 num_bytes, 0, start, len);
1593 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1594 struct btrfs_device *device,
1595 u64 start, u64 *dev_extent_len)
1597 struct btrfs_fs_info *fs_info = device->fs_info;
1598 struct btrfs_root *root = fs_info->dev_root;
1600 struct btrfs_path *path;
1601 struct btrfs_key key;
1602 struct btrfs_key found_key;
1603 struct extent_buffer *leaf = NULL;
1604 struct btrfs_dev_extent *extent = NULL;
1606 path = btrfs_alloc_path();
1610 key.objectid = device->devid;
1612 key.type = BTRFS_DEV_EXTENT_KEY;
1614 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1616 ret = btrfs_previous_item(root, path, key.objectid,
1617 BTRFS_DEV_EXTENT_KEY);
1620 leaf = path->nodes[0];
1621 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1622 extent = btrfs_item_ptr(leaf, path->slots[0],
1623 struct btrfs_dev_extent);
1624 BUG_ON(found_key.offset > start || found_key.offset +
1625 btrfs_dev_extent_length(leaf, extent) < start);
1627 btrfs_release_path(path);
1629 } else if (ret == 0) {
1630 leaf = path->nodes[0];
1631 extent = btrfs_item_ptr(leaf, path->slots[0],
1632 struct btrfs_dev_extent);
1634 btrfs_handle_fs_error(fs_info, ret, "Slot search failed");
1638 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1640 ret = btrfs_del_item(trans, root, path);
1642 btrfs_handle_fs_error(fs_info, ret,
1643 "Failed to remove dev extent item");
1645 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1648 btrfs_free_path(path);
1652 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1653 struct btrfs_device *device,
1654 u64 chunk_offset, u64 start, u64 num_bytes)
1657 struct btrfs_path *path;
1658 struct btrfs_fs_info *fs_info = device->fs_info;
1659 struct btrfs_root *root = fs_info->dev_root;
1660 struct btrfs_dev_extent *extent;
1661 struct extent_buffer *leaf;
1662 struct btrfs_key key;
1664 WARN_ON(!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state));
1665 WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state));
1666 path = btrfs_alloc_path();
1670 key.objectid = device->devid;
1672 key.type = BTRFS_DEV_EXTENT_KEY;
1673 ret = btrfs_insert_empty_item(trans, root, path, &key,
1678 leaf = path->nodes[0];
1679 extent = btrfs_item_ptr(leaf, path->slots[0],
1680 struct btrfs_dev_extent);
1681 btrfs_set_dev_extent_chunk_tree(leaf, extent,
1682 BTRFS_CHUNK_TREE_OBJECTID);
1683 btrfs_set_dev_extent_chunk_objectid(leaf, extent,
1684 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
1685 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1687 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1688 btrfs_mark_buffer_dirty(leaf);
1690 btrfs_free_path(path);
1694 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1696 struct extent_map_tree *em_tree;
1697 struct extent_map *em;
1701 em_tree = &fs_info->mapping_tree.map_tree;
1702 read_lock(&em_tree->lock);
1703 n = rb_last(&em_tree->map);
1705 em = rb_entry(n, struct extent_map, rb_node);
1706 ret = em->start + em->len;
1708 read_unlock(&em_tree->lock);
1713 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1717 struct btrfs_key key;
1718 struct btrfs_key found_key;
1719 struct btrfs_path *path;
1721 path = btrfs_alloc_path();
1725 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1726 key.type = BTRFS_DEV_ITEM_KEY;
1727 key.offset = (u64)-1;
1729 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1733 BUG_ON(ret == 0); /* Corruption */
1735 ret = btrfs_previous_item(fs_info->chunk_root, path,
1736 BTRFS_DEV_ITEMS_OBJECTID,
1737 BTRFS_DEV_ITEM_KEY);
1741 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1743 *devid_ret = found_key.offset + 1;
1747 btrfs_free_path(path);
1752 * the device information is stored in the chunk root
1753 * the btrfs_device struct should be fully filled in
1755 static int btrfs_add_dev_item(struct btrfs_trans_handle *trans,
1756 struct btrfs_fs_info *fs_info,
1757 struct btrfs_device *device)
1759 struct btrfs_root *root = fs_info->chunk_root;
1761 struct btrfs_path *path;
1762 struct btrfs_dev_item *dev_item;
1763 struct extent_buffer *leaf;
1764 struct btrfs_key key;
1767 path = btrfs_alloc_path();
1771 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1772 key.type = BTRFS_DEV_ITEM_KEY;
1773 key.offset = device->devid;
1775 ret = btrfs_insert_empty_item(trans, root, path, &key,
1780 leaf = path->nodes[0];
1781 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1783 btrfs_set_device_id(leaf, dev_item, device->devid);
1784 btrfs_set_device_generation(leaf, dev_item, 0);
1785 btrfs_set_device_type(leaf, dev_item, device->type);
1786 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1787 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1788 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1789 btrfs_set_device_total_bytes(leaf, dev_item,
1790 btrfs_device_get_disk_total_bytes(device));
1791 btrfs_set_device_bytes_used(leaf, dev_item,
1792 btrfs_device_get_bytes_used(device));
1793 btrfs_set_device_group(leaf, dev_item, 0);
1794 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1795 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1796 btrfs_set_device_start_offset(leaf, dev_item, 0);
1798 ptr = btrfs_device_uuid(dev_item);
1799 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1800 ptr = btrfs_device_fsid(dev_item);
1801 write_extent_buffer(leaf, fs_info->fsid, ptr, BTRFS_FSID_SIZE);
1802 btrfs_mark_buffer_dirty(leaf);
1806 btrfs_free_path(path);
1811 * Function to update ctime/mtime for a given device path.
1812 * Mainly used for ctime/mtime based probe like libblkid.
1814 static void update_dev_time(const char *path_name)
1818 filp = filp_open(path_name, O_RDWR, 0);
1821 file_update_time(filp);
1822 filp_close(filp, NULL);
1825 static int btrfs_rm_dev_item(struct btrfs_fs_info *fs_info,
1826 struct btrfs_device *device)
1828 struct btrfs_root *root = fs_info->chunk_root;
1830 struct btrfs_path *path;
1831 struct btrfs_key key;
1832 struct btrfs_trans_handle *trans;
1834 path = btrfs_alloc_path();
1838 trans = btrfs_start_transaction(root, 0);
1839 if (IS_ERR(trans)) {
1840 btrfs_free_path(path);
1841 return PTR_ERR(trans);
1843 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1844 key.type = BTRFS_DEV_ITEM_KEY;
1845 key.offset = device->devid;
1847 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1851 btrfs_abort_transaction(trans, ret);
1852 btrfs_end_transaction(trans);
1856 ret = btrfs_del_item(trans, root, path);
1858 btrfs_abort_transaction(trans, ret);
1859 btrfs_end_transaction(trans);
1863 btrfs_free_path(path);
1865 ret = btrfs_commit_transaction(trans);
1870 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1871 * filesystem. It's up to the caller to adjust that number regarding eg. device
1874 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1882 seq = read_seqbegin(&fs_info->profiles_lock);
1884 all_avail = fs_info->avail_data_alloc_bits |
1885 fs_info->avail_system_alloc_bits |
1886 fs_info->avail_metadata_alloc_bits;
1887 } while (read_seqretry(&fs_info->profiles_lock, seq));
1889 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1890 if (!(all_avail & btrfs_raid_array[i].bg_flag))
1893 if (num_devices < btrfs_raid_array[i].devs_min) {
1894 int ret = btrfs_raid_array[i].mindev_error;
1904 static struct btrfs_device * btrfs_find_next_active_device(
1905 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device)
1907 struct btrfs_device *next_device;
1909 list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1910 if (next_device != device &&
1911 !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state)
1912 && next_device->bdev)
1920 * Helper function to check if the given device is part of s_bdev / latest_bdev
1921 * and replace it with the provided or the next active device, in the context
1922 * where this function called, there should be always be another device (or
1923 * this_dev) which is active.
1925 void btrfs_assign_next_active_device(struct btrfs_fs_info *fs_info,
1926 struct btrfs_device *device, struct btrfs_device *this_dev)
1928 struct btrfs_device *next_device;
1931 next_device = this_dev;
1933 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
1935 ASSERT(next_device);
1937 if (fs_info->sb->s_bdev &&
1938 (fs_info->sb->s_bdev == device->bdev))
1939 fs_info->sb->s_bdev = next_device->bdev;
1941 if (fs_info->fs_devices->latest_bdev == device->bdev)
1942 fs_info->fs_devices->latest_bdev = next_device->bdev;
1945 int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path,
1948 struct btrfs_device *device;
1949 struct btrfs_fs_devices *cur_devices;
1950 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
1954 mutex_lock(&uuid_mutex);
1956 num_devices = fs_devices->num_devices;
1957 btrfs_dev_replace_read_lock(&fs_info->dev_replace);
1958 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
1959 WARN_ON(num_devices < 1);
1962 btrfs_dev_replace_read_unlock(&fs_info->dev_replace);
1964 ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
1968 ret = btrfs_find_device_by_devspec(fs_info, devid, device_path,
1973 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1974 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1978 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
1979 fs_info->fs_devices->rw_devices == 1) {
1980 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1984 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1985 mutex_lock(&fs_info->chunk_mutex);
1986 list_del_init(&device->dev_alloc_list);
1987 device->fs_devices->rw_devices--;
1988 mutex_unlock(&fs_info->chunk_mutex);
1991 mutex_unlock(&uuid_mutex);
1992 ret = btrfs_shrink_device(device, 0);
1993 mutex_lock(&uuid_mutex);
1998 * TODO: the superblock still includes this device in its num_devices
1999 * counter although write_all_supers() is not locked out. This
2000 * could give a filesystem state which requires a degraded mount.
2002 ret = btrfs_rm_dev_item(fs_info, device);
2006 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2007 btrfs_scrub_cancel_dev(fs_info, device);
2010 * the device list mutex makes sure that we don't change
2011 * the device list while someone else is writing out all
2012 * the device supers. Whoever is writing all supers, should
2013 * lock the device list mutex before getting the number of
2014 * devices in the super block (super_copy). Conversely,
2015 * whoever updates the number of devices in the super block
2016 * (super_copy) should hold the device list mutex.
2020 * In normal cases the cur_devices == fs_devices. But in case
2021 * of deleting a seed device, the cur_devices should point to
2022 * its own fs_devices listed under the fs_devices->seed.
2024 cur_devices = device->fs_devices;
2025 mutex_lock(&fs_devices->device_list_mutex);
2026 list_del_rcu(&device->dev_list);
2028 cur_devices->num_devices--;
2029 cur_devices->total_devices--;
2031 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
2032 cur_devices->missing_devices--;
2034 btrfs_assign_next_active_device(fs_info, device, NULL);
2037 cur_devices->open_devices--;
2038 /* remove sysfs entry */
2039 btrfs_sysfs_rm_device_link(fs_devices, device);
2042 num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
2043 btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
2044 mutex_unlock(&fs_devices->device_list_mutex);
2047 * at this point, the device is zero sized and detached from
2048 * the devices list. All that's left is to zero out the old
2049 * supers and free the device.
2051 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2052 btrfs_scratch_superblocks(device->bdev, device->name->str);
2054 btrfs_close_bdev(device);
2055 call_rcu(&device->rcu, free_device_rcu);
2057 if (cur_devices->open_devices == 0) {
2058 while (fs_devices) {
2059 if (fs_devices->seed == cur_devices) {
2060 fs_devices->seed = cur_devices->seed;
2063 fs_devices = fs_devices->seed;
2065 cur_devices->seed = NULL;
2066 close_fs_devices(cur_devices);
2067 free_fs_devices(cur_devices);
2071 mutex_unlock(&uuid_mutex);
2075 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2076 mutex_lock(&fs_info->chunk_mutex);
2077 list_add(&device->dev_alloc_list,
2078 &fs_devices->alloc_list);
2079 device->fs_devices->rw_devices++;
2080 mutex_unlock(&fs_info->chunk_mutex);
2085 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
2086 struct btrfs_device *srcdev)
2088 struct btrfs_fs_devices *fs_devices;
2090 lockdep_assert_held(&fs_info->fs_devices->device_list_mutex);
2093 * in case of fs with no seed, srcdev->fs_devices will point
2094 * to fs_devices of fs_info. However when the dev being replaced is
2095 * a seed dev it will point to the seed's local fs_devices. In short
2096 * srcdev will have its correct fs_devices in both the cases.
2098 fs_devices = srcdev->fs_devices;
2100 list_del_rcu(&srcdev->dev_list);
2101 list_del(&srcdev->dev_alloc_list);
2102 fs_devices->num_devices--;
2103 if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state))
2104 fs_devices->missing_devices--;
2106 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state))
2107 fs_devices->rw_devices--;
2110 fs_devices->open_devices--;
2113 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
2114 struct btrfs_device *srcdev)
2116 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2118 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) {
2119 /* zero out the old super if it is writable */
2120 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
2123 btrfs_close_bdev(srcdev);
2124 call_rcu(&srcdev->rcu, free_device_rcu);
2126 /* if this is no devs we rather delete the fs_devices */
2127 if (!fs_devices->num_devices) {
2128 struct btrfs_fs_devices *tmp_fs_devices;
2131 * On a mounted FS, num_devices can't be zero unless it's a
2132 * seed. In case of a seed device being replaced, the replace
2133 * target added to the sprout FS, so there will be no more
2134 * device left under the seed FS.
2136 ASSERT(fs_devices->seeding);
2138 tmp_fs_devices = fs_info->fs_devices;
2139 while (tmp_fs_devices) {
2140 if (tmp_fs_devices->seed == fs_devices) {
2141 tmp_fs_devices->seed = fs_devices->seed;
2144 tmp_fs_devices = tmp_fs_devices->seed;
2146 fs_devices->seed = NULL;
2147 close_fs_devices(fs_devices);
2148 free_fs_devices(fs_devices);
2152 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2153 struct btrfs_device *tgtdev)
2156 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2158 btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev);
2161 fs_info->fs_devices->open_devices--;
2163 fs_info->fs_devices->num_devices--;
2165 btrfs_assign_next_active_device(fs_info, tgtdev, NULL);
2167 list_del_rcu(&tgtdev->dev_list);
2169 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2172 * The update_dev_time() with in btrfs_scratch_superblocks()
2173 * may lead to a call to btrfs_show_devname() which will try
2174 * to hold device_list_mutex. And here this device
2175 * is already out of device list, so we don't have to hold
2176 * the device_list_mutex lock.
2178 btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
2180 btrfs_close_bdev(tgtdev);
2181 call_rcu(&tgtdev->rcu, free_device_rcu);
2184 static int btrfs_find_device_by_path(struct btrfs_fs_info *fs_info,
2185 const char *device_path,
2186 struct btrfs_device **device)
2189 struct btrfs_super_block *disk_super;
2192 struct block_device *bdev;
2193 struct buffer_head *bh;
2196 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2197 fs_info->bdev_holder, 0, &bdev, &bh);
2200 disk_super = (struct btrfs_super_block *)bh->b_data;
2201 devid = btrfs_stack_device_id(&disk_super->dev_item);
2202 dev_uuid = disk_super->dev_item.uuid;
2203 *device = btrfs_find_device(fs_info, devid, dev_uuid, disk_super->fsid);
2207 blkdev_put(bdev, FMODE_READ);
2211 int btrfs_find_device_missing_or_by_path(struct btrfs_fs_info *fs_info,
2212 const char *device_path,
2213 struct btrfs_device **device)
2216 if (strcmp(device_path, "missing") == 0) {
2217 struct list_head *devices;
2218 struct btrfs_device *tmp;
2220 devices = &fs_info->fs_devices->devices;
2221 list_for_each_entry(tmp, devices, dev_list) {
2222 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
2223 &tmp->dev_state) && !tmp->bdev) {
2230 return BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2234 return btrfs_find_device_by_path(fs_info, device_path, device);
2239 * Lookup a device given by device id, or the path if the id is 0.
2241 int btrfs_find_device_by_devspec(struct btrfs_fs_info *fs_info, u64 devid,
2242 const char *devpath,
2243 struct btrfs_device **device)
2249 *device = btrfs_find_device(fs_info, devid, NULL, NULL);
2253 if (!devpath || !devpath[0])
2256 ret = btrfs_find_device_missing_or_by_path(fs_info, devpath,
2263 * does all the dirty work required for changing file system's UUID.
2265 static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info)
2267 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2268 struct btrfs_fs_devices *old_devices;
2269 struct btrfs_fs_devices *seed_devices;
2270 struct btrfs_super_block *disk_super = fs_info->super_copy;
2271 struct btrfs_device *device;
2274 lockdep_assert_held(&uuid_mutex);
2275 if (!fs_devices->seeding)
2278 seed_devices = alloc_fs_devices(NULL);
2279 if (IS_ERR(seed_devices))
2280 return PTR_ERR(seed_devices);
2282 old_devices = clone_fs_devices(fs_devices);
2283 if (IS_ERR(old_devices)) {
2284 kfree(seed_devices);
2285 return PTR_ERR(old_devices);
2288 list_add(&old_devices->fs_list, &fs_uuids);
2290 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2291 seed_devices->opened = 1;
2292 INIT_LIST_HEAD(&seed_devices->devices);
2293 INIT_LIST_HEAD(&seed_devices->alloc_list);
2294 mutex_init(&seed_devices->device_list_mutex);
2296 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2297 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2299 list_for_each_entry(device, &seed_devices->devices, dev_list)
2300 device->fs_devices = seed_devices;
2302 mutex_lock(&fs_info->chunk_mutex);
2303 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2304 mutex_unlock(&fs_info->chunk_mutex);
2306 fs_devices->seeding = 0;
2307 fs_devices->num_devices = 0;
2308 fs_devices->open_devices = 0;
2309 fs_devices->missing_devices = 0;
2310 fs_devices->rotating = 0;
2311 fs_devices->seed = seed_devices;
2313 generate_random_uuid(fs_devices->fsid);
2314 memcpy(fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2315 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2316 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2318 super_flags = btrfs_super_flags(disk_super) &
2319 ~BTRFS_SUPER_FLAG_SEEDING;
2320 btrfs_set_super_flags(disk_super, super_flags);
2326 * Store the expected generation for seed devices in device items.
2328 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2329 struct btrfs_fs_info *fs_info)
2331 struct btrfs_root *root = fs_info->chunk_root;
2332 struct btrfs_path *path;
2333 struct extent_buffer *leaf;
2334 struct btrfs_dev_item *dev_item;
2335 struct btrfs_device *device;
2336 struct btrfs_key key;
2337 u8 fs_uuid[BTRFS_FSID_SIZE];
2338 u8 dev_uuid[BTRFS_UUID_SIZE];
2342 path = btrfs_alloc_path();
2346 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2348 key.type = BTRFS_DEV_ITEM_KEY;
2351 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2355 leaf = path->nodes[0];
2357 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2358 ret = btrfs_next_leaf(root, path);
2363 leaf = path->nodes[0];
2364 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2365 btrfs_release_path(path);
2369 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2370 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2371 key.type != BTRFS_DEV_ITEM_KEY)
2374 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2375 struct btrfs_dev_item);
2376 devid = btrfs_device_id(leaf, dev_item);
2377 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2379 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2381 device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
2382 BUG_ON(!device); /* Logic error */
2384 if (device->fs_devices->seeding) {
2385 btrfs_set_device_generation(leaf, dev_item,
2386 device->generation);
2387 btrfs_mark_buffer_dirty(leaf);
2395 btrfs_free_path(path);
2399 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2401 struct btrfs_root *root = fs_info->dev_root;
2402 struct request_queue *q;
2403 struct btrfs_trans_handle *trans;
2404 struct btrfs_device *device;
2405 struct block_device *bdev;
2406 struct list_head *devices;
2407 struct super_block *sb = fs_info->sb;
2408 struct rcu_string *name;
2410 int seeding_dev = 0;
2412 bool unlocked = false;
2414 if (sb_rdonly(sb) && !fs_info->fs_devices->seeding)
2417 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2418 fs_info->bdev_holder);
2420 return PTR_ERR(bdev);
2422 if (fs_info->fs_devices->seeding) {
2424 down_write(&sb->s_umount);
2425 mutex_lock(&uuid_mutex);
2428 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2430 devices = &fs_info->fs_devices->devices;
2432 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2433 list_for_each_entry(device, devices, dev_list) {
2434 if (device->bdev == bdev) {
2437 &fs_info->fs_devices->device_list_mutex);
2441 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2443 device = btrfs_alloc_device(fs_info, NULL, NULL);
2444 if (IS_ERR(device)) {
2445 /* we can safely leave the fs_devices entry around */
2446 ret = PTR_ERR(device);
2450 name = rcu_string_strdup(device_path, GFP_KERNEL);
2453 goto error_free_device;
2455 rcu_assign_pointer(device->name, name);
2457 trans = btrfs_start_transaction(root, 0);
2458 if (IS_ERR(trans)) {
2459 ret = PTR_ERR(trans);
2460 goto error_free_device;
2463 q = bdev_get_queue(bdev);
2464 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
2465 device->generation = trans->transid;
2466 device->io_width = fs_info->sectorsize;
2467 device->io_align = fs_info->sectorsize;
2468 device->sector_size = fs_info->sectorsize;
2469 device->total_bytes = round_down(i_size_read(bdev->bd_inode),
2470 fs_info->sectorsize);
2471 device->disk_total_bytes = device->total_bytes;
2472 device->commit_total_bytes = device->total_bytes;
2473 device->fs_info = fs_info;
2474 device->bdev = bdev;
2475 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2476 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
2477 device->mode = FMODE_EXCL;
2478 device->dev_stats_valid = 1;
2479 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2482 sb->s_flags &= ~SB_RDONLY;
2483 ret = btrfs_prepare_sprout(fs_info);
2485 btrfs_abort_transaction(trans, ret);
2490 device->fs_devices = fs_info->fs_devices;
2492 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2493 mutex_lock(&fs_info->chunk_mutex);
2494 list_add_rcu(&device->dev_list, &fs_info->fs_devices->devices);
2495 list_add(&device->dev_alloc_list,
2496 &fs_info->fs_devices->alloc_list);
2497 fs_info->fs_devices->num_devices++;
2498 fs_info->fs_devices->open_devices++;
2499 fs_info->fs_devices->rw_devices++;
2500 fs_info->fs_devices->total_devices++;
2501 fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2503 atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2505 if (!blk_queue_nonrot(q))
2506 fs_info->fs_devices->rotating = 1;
2508 tmp = btrfs_super_total_bytes(fs_info->super_copy);
2509 btrfs_set_super_total_bytes(fs_info->super_copy,
2510 round_down(tmp + device->total_bytes, fs_info->sectorsize));
2512 tmp = btrfs_super_num_devices(fs_info->super_copy);
2513 btrfs_set_super_num_devices(fs_info->super_copy, tmp + 1);
2515 /* add sysfs device entry */
2516 btrfs_sysfs_add_device_link(fs_info->fs_devices, device);
2519 * we've got more storage, clear any full flags on the space
2522 btrfs_clear_space_info_full(fs_info);
2524 mutex_unlock(&fs_info->chunk_mutex);
2525 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2528 mutex_lock(&fs_info->chunk_mutex);
2529 ret = init_first_rw_device(trans, fs_info);
2530 mutex_unlock(&fs_info->chunk_mutex);
2532 btrfs_abort_transaction(trans, ret);
2537 ret = btrfs_add_dev_item(trans, fs_info, device);
2539 btrfs_abort_transaction(trans, ret);
2544 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2546 ret = btrfs_finish_sprout(trans, fs_info);
2548 btrfs_abort_transaction(trans, ret);
2552 /* Sprouting would change fsid of the mounted root,
2553 * so rename the fsid on the sysfs
2555 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2557 if (kobject_rename(&fs_info->fs_devices->fsid_kobj, fsid_buf))
2559 "sysfs: failed to create fsid for sprout");
2562 ret = btrfs_commit_transaction(trans);
2565 mutex_unlock(&uuid_mutex);
2566 up_write(&sb->s_umount);
2569 if (ret) /* transaction commit */
2572 ret = btrfs_relocate_sys_chunks(fs_info);
2574 btrfs_handle_fs_error(fs_info, ret,
2575 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2576 trans = btrfs_attach_transaction(root);
2577 if (IS_ERR(trans)) {
2578 if (PTR_ERR(trans) == -ENOENT)
2580 ret = PTR_ERR(trans);
2584 ret = btrfs_commit_transaction(trans);
2587 /* Update ctime/mtime for libblkid */
2588 update_dev_time(device_path);
2592 btrfs_sysfs_rm_device_link(fs_info->fs_devices, device);
2595 sb->s_flags |= SB_RDONLY;
2597 btrfs_end_transaction(trans);
2599 btrfs_free_device(device);
2601 blkdev_put(bdev, FMODE_EXCL);
2602 if (seeding_dev && !unlocked) {
2603 mutex_unlock(&uuid_mutex);
2604 up_write(&sb->s_umount);
2609 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2610 struct btrfs_device *device)
2613 struct btrfs_path *path;
2614 struct btrfs_root *root = device->fs_info->chunk_root;
2615 struct btrfs_dev_item *dev_item;
2616 struct extent_buffer *leaf;
2617 struct btrfs_key key;
2619 path = btrfs_alloc_path();
2623 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2624 key.type = BTRFS_DEV_ITEM_KEY;
2625 key.offset = device->devid;
2627 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2636 leaf = path->nodes[0];
2637 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2639 btrfs_set_device_id(leaf, dev_item, device->devid);
2640 btrfs_set_device_type(leaf, dev_item, device->type);
2641 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2642 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2643 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2644 btrfs_set_device_total_bytes(leaf, dev_item,
2645 btrfs_device_get_disk_total_bytes(device));
2646 btrfs_set_device_bytes_used(leaf, dev_item,
2647 btrfs_device_get_bytes_used(device));
2648 btrfs_mark_buffer_dirty(leaf);
2651 btrfs_free_path(path);
2655 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2656 struct btrfs_device *device, u64 new_size)
2658 struct btrfs_fs_info *fs_info = device->fs_info;
2659 struct btrfs_super_block *super_copy = fs_info->super_copy;
2660 struct btrfs_fs_devices *fs_devices;
2664 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2667 new_size = round_down(new_size, fs_info->sectorsize);
2669 mutex_lock(&fs_info->chunk_mutex);
2670 old_total = btrfs_super_total_bytes(super_copy);
2671 diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2673 if (new_size <= device->total_bytes ||
2674 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2675 mutex_unlock(&fs_info->chunk_mutex);
2679 fs_devices = fs_info->fs_devices;
2681 btrfs_set_super_total_bytes(super_copy,
2682 round_down(old_total + diff, fs_info->sectorsize));
2683 device->fs_devices->total_rw_bytes += diff;
2685 btrfs_device_set_total_bytes(device, new_size);
2686 btrfs_device_set_disk_total_bytes(device, new_size);
2687 btrfs_clear_space_info_full(device->fs_info);
2688 if (list_empty(&device->resized_list))
2689 list_add_tail(&device->resized_list,
2690 &fs_devices->resized_devices);
2691 mutex_unlock(&fs_info->chunk_mutex);
2693 return btrfs_update_device(trans, device);
2696 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2697 struct btrfs_fs_info *fs_info, u64 chunk_offset)
2699 struct btrfs_root *root = fs_info->chunk_root;
2701 struct btrfs_path *path;
2702 struct btrfs_key key;
2704 path = btrfs_alloc_path();
2708 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2709 key.offset = chunk_offset;
2710 key.type = BTRFS_CHUNK_ITEM_KEY;
2712 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2715 else if (ret > 0) { /* Logic error or corruption */
2716 btrfs_handle_fs_error(fs_info, -ENOENT,
2717 "Failed lookup while freeing chunk.");
2722 ret = btrfs_del_item(trans, root, path);
2724 btrfs_handle_fs_error(fs_info, ret,
2725 "Failed to delete chunk item.");
2727 btrfs_free_path(path);
2731 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2733 struct btrfs_super_block *super_copy = fs_info->super_copy;
2734 struct btrfs_disk_key *disk_key;
2735 struct btrfs_chunk *chunk;
2742 struct btrfs_key key;
2744 mutex_lock(&fs_info->chunk_mutex);
2745 array_size = btrfs_super_sys_array_size(super_copy);
2747 ptr = super_copy->sys_chunk_array;
2750 while (cur < array_size) {
2751 disk_key = (struct btrfs_disk_key *)ptr;
2752 btrfs_disk_key_to_cpu(&key, disk_key);
2754 len = sizeof(*disk_key);
2756 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2757 chunk = (struct btrfs_chunk *)(ptr + len);
2758 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2759 len += btrfs_chunk_item_size(num_stripes);
2764 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID &&
2765 key.offset == chunk_offset) {
2766 memmove(ptr, ptr + len, array_size - (cur + len));
2768 btrfs_set_super_sys_array_size(super_copy, array_size);
2774 mutex_unlock(&fs_info->chunk_mutex);
2778 static struct extent_map *get_chunk_map(struct btrfs_fs_info *fs_info,
2779 u64 logical, u64 length)
2781 struct extent_map_tree *em_tree;
2782 struct extent_map *em;
2784 em_tree = &fs_info->mapping_tree.map_tree;
2785 read_lock(&em_tree->lock);
2786 em = lookup_extent_mapping(em_tree, logical, length);
2787 read_unlock(&em_tree->lock);
2790 btrfs_crit(fs_info, "unable to find logical %llu length %llu",
2792 return ERR_PTR(-EINVAL);
2795 if (em->start > logical || em->start + em->len < logical) {
2797 "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
2798 logical, length, em->start, em->start + em->len);
2799 free_extent_map(em);
2800 return ERR_PTR(-EINVAL);
2803 /* callers are responsible for dropping em's ref. */
2807 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2808 struct btrfs_fs_info *fs_info, u64 chunk_offset)
2810 struct extent_map *em;
2811 struct map_lookup *map;
2812 u64 dev_extent_len = 0;
2814 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2816 em = get_chunk_map(fs_info, chunk_offset, 1);
2819 * This is a logic error, but we don't want to just rely on the
2820 * user having built with ASSERT enabled, so if ASSERT doesn't
2821 * do anything we still error out.
2826 map = em->map_lookup;
2827 mutex_lock(&fs_info->chunk_mutex);
2828 check_system_chunk(trans, fs_info, map->type);
2829 mutex_unlock(&fs_info->chunk_mutex);
2832 * Take the device list mutex to prevent races with the final phase of
2833 * a device replace operation that replaces the device object associated
2834 * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2836 mutex_lock(&fs_devices->device_list_mutex);
2837 for (i = 0; i < map->num_stripes; i++) {
2838 struct btrfs_device *device = map->stripes[i].dev;
2839 ret = btrfs_free_dev_extent(trans, device,
2840 map->stripes[i].physical,
2843 mutex_unlock(&fs_devices->device_list_mutex);
2844 btrfs_abort_transaction(trans, ret);
2848 if (device->bytes_used > 0) {
2849 mutex_lock(&fs_info->chunk_mutex);
2850 btrfs_device_set_bytes_used(device,
2851 device->bytes_used - dev_extent_len);
2852 atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
2853 btrfs_clear_space_info_full(fs_info);
2854 mutex_unlock(&fs_info->chunk_mutex);
2857 if (map->stripes[i].dev) {
2858 ret = btrfs_update_device(trans, map->stripes[i].dev);
2860 mutex_unlock(&fs_devices->device_list_mutex);
2861 btrfs_abort_transaction(trans, ret);
2866 mutex_unlock(&fs_devices->device_list_mutex);
2868 ret = btrfs_free_chunk(trans, fs_info, chunk_offset);
2870 btrfs_abort_transaction(trans, ret);
2874 trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
2876 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2877 ret = btrfs_del_sys_chunk(fs_info, chunk_offset);
2879 btrfs_abort_transaction(trans, ret);
2884 ret = btrfs_remove_block_group(trans, fs_info, chunk_offset, em);
2886 btrfs_abort_transaction(trans, ret);
2892 free_extent_map(em);
2896 static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2898 struct btrfs_root *root = fs_info->chunk_root;
2899 struct btrfs_trans_handle *trans;
2903 * Prevent races with automatic removal of unused block groups.
2904 * After we relocate and before we remove the chunk with offset
2905 * chunk_offset, automatic removal of the block group can kick in,
2906 * resulting in a failure when calling btrfs_remove_chunk() below.
2908 * Make sure to acquire this mutex before doing a tree search (dev
2909 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2910 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2911 * we release the path used to search the chunk/dev tree and before
2912 * the current task acquires this mutex and calls us.
2914 lockdep_assert_held(&fs_info->delete_unused_bgs_mutex);
2916 ret = btrfs_can_relocate(fs_info, chunk_offset);
2920 /* step one, relocate all the extents inside this chunk */
2921 btrfs_scrub_pause(fs_info);
2922 ret = btrfs_relocate_block_group(fs_info, chunk_offset);
2923 btrfs_scrub_continue(fs_info);
2928 * We add the kobjects here (and after forcing data chunk creation)
2929 * since relocation is the only place we'll create chunks of a new
2930 * type at runtime. The only place where we'll remove the last
2931 * chunk of a type is the call immediately below this one. Even
2932 * so, we're protected against races with the cleaner thread since
2933 * we're covered by the delete_unused_bgs_mutex.
2935 btrfs_add_raid_kobjects(fs_info);
2937 trans = btrfs_start_trans_remove_block_group(root->fs_info,
2939 if (IS_ERR(trans)) {
2940 ret = PTR_ERR(trans);
2941 btrfs_handle_fs_error(root->fs_info, ret, NULL);
2946 * step two, delete the device extents and the
2947 * chunk tree entries
2949 ret = btrfs_remove_chunk(trans, fs_info, chunk_offset);
2950 btrfs_end_transaction(trans);
2954 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
2956 struct btrfs_root *chunk_root = fs_info->chunk_root;
2957 struct btrfs_path *path;
2958 struct extent_buffer *leaf;
2959 struct btrfs_chunk *chunk;
2960 struct btrfs_key key;
2961 struct btrfs_key found_key;
2963 bool retried = false;
2967 path = btrfs_alloc_path();
2972 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2973 key.offset = (u64)-1;
2974 key.type = BTRFS_CHUNK_ITEM_KEY;
2977 mutex_lock(&fs_info->delete_unused_bgs_mutex);
2978 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2980 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
2983 BUG_ON(ret == 0); /* Corruption */
2985 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2988 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
2994 leaf = path->nodes[0];
2995 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2997 chunk = btrfs_item_ptr(leaf, path->slots[0],
2998 struct btrfs_chunk);
2999 chunk_type = btrfs_chunk_type(leaf, chunk);
3000 btrfs_release_path(path);
3002 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3003 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3009 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3011 if (found_key.offset == 0)
3013 key.offset = found_key.offset - 1;
3016 if (failed && !retried) {
3020 } else if (WARN_ON(failed && retried)) {
3024 btrfs_free_path(path);
3029 * return 1 : allocate a data chunk successfully,
3030 * return <0: errors during allocating a data chunk,
3031 * return 0 : no need to allocate a data chunk.
3033 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info,
3036 struct btrfs_block_group_cache *cache;
3040 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3042 chunk_type = cache->flags;
3043 btrfs_put_block_group(cache);
3045 if (chunk_type & BTRFS_BLOCK_GROUP_DATA) {
3046 spin_lock(&fs_info->data_sinfo->lock);
3047 bytes_used = fs_info->data_sinfo->bytes_used;
3048 spin_unlock(&fs_info->data_sinfo->lock);
3051 struct btrfs_trans_handle *trans;
3054 trans = btrfs_join_transaction(fs_info->tree_root);
3056 return PTR_ERR(trans);
3058 ret = btrfs_force_chunk_alloc(trans, fs_info,
3059 BTRFS_BLOCK_GROUP_DATA);
3060 btrfs_end_transaction(trans);
3064 btrfs_add_raid_kobjects(fs_info);
3072 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3073 struct btrfs_balance_control *bctl)
3075 struct btrfs_root *root = fs_info->tree_root;
3076 struct btrfs_trans_handle *trans;
3077 struct btrfs_balance_item *item;
3078 struct btrfs_disk_balance_args disk_bargs;
3079 struct btrfs_path *path;
3080 struct extent_buffer *leaf;
3081 struct btrfs_key key;
3084 path = btrfs_alloc_path();
3088 trans = btrfs_start_transaction(root, 0);
3089 if (IS_ERR(trans)) {
3090 btrfs_free_path(path);
3091 return PTR_ERR(trans);
3094 key.objectid = BTRFS_BALANCE_OBJECTID;
3095 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3098 ret = btrfs_insert_empty_item(trans, root, path, &key,
3103 leaf = path->nodes[0];
3104 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3106 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3108 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3109 btrfs_set_balance_data(leaf, item, &disk_bargs);
3110 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3111 btrfs_set_balance_meta(leaf, item, &disk_bargs);
3112 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3113 btrfs_set_balance_sys(leaf, item, &disk_bargs);
3115 btrfs_set_balance_flags(leaf, item, bctl->flags);
3117 btrfs_mark_buffer_dirty(leaf);
3119 btrfs_free_path(path);
3120 err = btrfs_commit_transaction(trans);
3126 static int del_balance_item(struct btrfs_fs_info *fs_info)
3128 struct btrfs_root *root = fs_info->tree_root;
3129 struct btrfs_trans_handle *trans;
3130 struct btrfs_path *path;
3131 struct btrfs_key key;
3134 path = btrfs_alloc_path();
3138 trans = btrfs_start_transaction(root, 0);
3139 if (IS_ERR(trans)) {
3140 btrfs_free_path(path);
3141 return PTR_ERR(trans);
3144 key.objectid = BTRFS_BALANCE_OBJECTID;
3145 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3148 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3156 ret = btrfs_del_item(trans, root, path);
3158 btrfs_free_path(path);
3159 err = btrfs_commit_transaction(trans);
3166 * This is a heuristic used to reduce the number of chunks balanced on
3167 * resume after balance was interrupted.
3169 static void update_balance_args(struct btrfs_balance_control *bctl)
3172 * Turn on soft mode for chunk types that were being converted.
3174 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3175 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3176 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3177 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3178 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3179 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3182 * Turn on usage filter if is not already used. The idea is
3183 * that chunks that we have already balanced should be
3184 * reasonably full. Don't do it for chunks that are being
3185 * converted - that will keep us from relocating unconverted
3186 * (albeit full) chunks.
3188 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3189 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3190 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3191 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3192 bctl->data.usage = 90;
3194 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3195 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3196 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3197 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3198 bctl->sys.usage = 90;
3200 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3201 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3202 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3203 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3204 bctl->meta.usage = 90;
3209 * Clear the balance status in fs_info and delete the balance item from disk.
3211 static void reset_balance_state(struct btrfs_fs_info *fs_info)
3213 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3216 BUG_ON(!fs_info->balance_ctl);
3218 spin_lock(&fs_info->balance_lock);
3219 fs_info->balance_ctl = NULL;
3220 spin_unlock(&fs_info->balance_lock);
3223 ret = del_balance_item(fs_info);
3225 btrfs_handle_fs_error(fs_info, ret, NULL);
3229 * Balance filters. Return 1 if chunk should be filtered out
3230 * (should not be balanced).
3232 static int chunk_profiles_filter(u64 chunk_type,
3233 struct btrfs_balance_args *bargs)
3235 chunk_type = chunk_to_extended(chunk_type) &
3236 BTRFS_EXTENDED_PROFILE_MASK;
3238 if (bargs->profiles & chunk_type)
3244 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3245 struct btrfs_balance_args *bargs)
3247 struct btrfs_block_group_cache *cache;
3249 u64 user_thresh_min;
3250 u64 user_thresh_max;
3253 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3254 chunk_used = btrfs_block_group_used(&cache->item);
3256 if (bargs->usage_min == 0)
3257 user_thresh_min = 0;
3259 user_thresh_min = div_factor_fine(cache->key.offset,
3262 if (bargs->usage_max == 0)
3263 user_thresh_max = 1;
3264 else if (bargs->usage_max > 100)
3265 user_thresh_max = cache->key.offset;
3267 user_thresh_max = div_factor_fine(cache->key.offset,
3270 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3273 btrfs_put_block_group(cache);
3277 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3278 u64 chunk_offset, struct btrfs_balance_args *bargs)
3280 struct btrfs_block_group_cache *cache;
3281 u64 chunk_used, user_thresh;
3284 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3285 chunk_used = btrfs_block_group_used(&cache->item);
3287 if (bargs->usage_min == 0)
3289 else if (bargs->usage > 100)
3290 user_thresh = cache->key.offset;
3292 user_thresh = div_factor_fine(cache->key.offset,
3295 if (chunk_used < user_thresh)
3298 btrfs_put_block_group(cache);
3302 static int chunk_devid_filter(struct extent_buffer *leaf,
3303 struct btrfs_chunk *chunk,
3304 struct btrfs_balance_args *bargs)
3306 struct btrfs_stripe *stripe;
3307 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3310 for (i = 0; i < num_stripes; i++) {
3311 stripe = btrfs_stripe_nr(chunk, i);
3312 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3319 /* [pstart, pend) */
3320 static int chunk_drange_filter(struct extent_buffer *leaf,
3321 struct btrfs_chunk *chunk,
3322 struct btrfs_balance_args *bargs)
3324 struct btrfs_stripe *stripe;
3325 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3331 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3334 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3335 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3336 factor = num_stripes / 2;
3337 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3338 factor = num_stripes - 1;
3339 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3340 factor = num_stripes - 2;
3342 factor = num_stripes;
3345 for (i = 0; i < num_stripes; i++) {
3346 stripe = btrfs_stripe_nr(chunk, i);
3347 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3350 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3351 stripe_length = btrfs_chunk_length(leaf, chunk);
3352 stripe_length = div_u64(stripe_length, factor);
3354 if (stripe_offset < bargs->pend &&
3355 stripe_offset + stripe_length > bargs->pstart)
3362 /* [vstart, vend) */
3363 static int chunk_vrange_filter(struct extent_buffer *leaf,
3364 struct btrfs_chunk *chunk,
3366 struct btrfs_balance_args *bargs)
3368 if (chunk_offset < bargs->vend &&
3369 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3370 /* at least part of the chunk is inside this vrange */
3376 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3377 struct btrfs_chunk *chunk,
3378 struct btrfs_balance_args *bargs)
3380 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3382 if (bargs->stripes_min <= num_stripes
3383 && num_stripes <= bargs->stripes_max)
3389 static int chunk_soft_convert_filter(u64 chunk_type,
3390 struct btrfs_balance_args *bargs)
3392 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3395 chunk_type = chunk_to_extended(chunk_type) &
3396 BTRFS_EXTENDED_PROFILE_MASK;
3398 if (bargs->target == chunk_type)
3404 static int should_balance_chunk(struct btrfs_fs_info *fs_info,
3405 struct extent_buffer *leaf,
3406 struct btrfs_chunk *chunk, u64 chunk_offset)
3408 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3409 struct btrfs_balance_args *bargs = NULL;
3410 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3413 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3414 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3418 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3419 bargs = &bctl->data;
3420 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3422 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3423 bargs = &bctl->meta;
3425 /* profiles filter */
3426 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3427 chunk_profiles_filter(chunk_type, bargs)) {
3432 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3433 chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3435 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3436 chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3441 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3442 chunk_devid_filter(leaf, chunk, bargs)) {
3446 /* drange filter, makes sense only with devid filter */
3447 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3448 chunk_drange_filter(leaf, chunk, bargs)) {
3453 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3454 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3458 /* stripes filter */
3459 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3460 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3464 /* soft profile changing mode */
3465 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3466 chunk_soft_convert_filter(chunk_type, bargs)) {
3471 * limited by count, must be the last filter
3473 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3474 if (bargs->limit == 0)
3478 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3480 * Same logic as the 'limit' filter; the minimum cannot be
3481 * determined here because we do not have the global information
3482 * about the count of all chunks that satisfy the filters.
3484 if (bargs->limit_max == 0)
3493 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3495 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3496 struct btrfs_root *chunk_root = fs_info->chunk_root;
3497 struct btrfs_root *dev_root = fs_info->dev_root;
3498 struct list_head *devices;
3499 struct btrfs_device *device;
3503 struct btrfs_chunk *chunk;
3504 struct btrfs_path *path = NULL;
3505 struct btrfs_key key;
3506 struct btrfs_key found_key;
3507 struct btrfs_trans_handle *trans;
3508 struct extent_buffer *leaf;
3511 int enospc_errors = 0;
3512 bool counting = true;
3513 /* The single value limit and min/max limits use the same bytes in the */
3514 u64 limit_data = bctl->data.limit;
3515 u64 limit_meta = bctl->meta.limit;
3516 u64 limit_sys = bctl->sys.limit;
3520 int chunk_reserved = 0;
3522 /* step one make some room on all the devices */
3523 devices = &fs_info->fs_devices->devices;
3524 list_for_each_entry(device, devices, dev_list) {
3525 old_size = btrfs_device_get_total_bytes(device);
3526 size_to_free = div_factor(old_size, 1);
3527 size_to_free = min_t(u64, size_to_free, SZ_1M);
3528 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) ||
3529 btrfs_device_get_total_bytes(device) -
3530 btrfs_device_get_bytes_used(device) > size_to_free ||
3531 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
3534 ret = btrfs_shrink_device(device, old_size - size_to_free);
3538 /* btrfs_shrink_device never returns ret > 0 */
3543 trans = btrfs_start_transaction(dev_root, 0);
3544 if (IS_ERR(trans)) {
3545 ret = PTR_ERR(trans);
3546 btrfs_info_in_rcu(fs_info,
3547 "resize: unable to start transaction after shrinking device %s (error %d), old size %llu, new size %llu",
3548 rcu_str_deref(device->name), ret,
3549 old_size, old_size - size_to_free);
3553 ret = btrfs_grow_device(trans, device, old_size);
3555 btrfs_end_transaction(trans);
3556 /* btrfs_grow_device never returns ret > 0 */
3558 btrfs_info_in_rcu(fs_info,
3559 "resize: unable to grow device after shrinking device %s (error %d), old size %llu, new size %llu",
3560 rcu_str_deref(device->name), ret,
3561 old_size, old_size - size_to_free);
3565 btrfs_end_transaction(trans);
3568 /* step two, relocate all the chunks */
3569 path = btrfs_alloc_path();
3575 /* zero out stat counters */
3576 spin_lock(&fs_info->balance_lock);
3577 memset(&bctl->stat, 0, sizeof(bctl->stat));
3578 spin_unlock(&fs_info->balance_lock);
3582 * The single value limit and min/max limits use the same bytes
3585 bctl->data.limit = limit_data;
3586 bctl->meta.limit = limit_meta;
3587 bctl->sys.limit = limit_sys;
3589 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3590 key.offset = (u64)-1;
3591 key.type = BTRFS_CHUNK_ITEM_KEY;
3594 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3595 atomic_read(&fs_info->balance_cancel_req)) {
3600 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3601 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3603 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3608 * this shouldn't happen, it means the last relocate
3612 BUG(); /* FIXME break ? */
3614 ret = btrfs_previous_item(chunk_root, path, 0,
3615 BTRFS_CHUNK_ITEM_KEY);
3617 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3622 leaf = path->nodes[0];
3623 slot = path->slots[0];
3624 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3626 if (found_key.objectid != key.objectid) {
3627 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3631 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3632 chunk_type = btrfs_chunk_type(leaf, chunk);
3635 spin_lock(&fs_info->balance_lock);
3636 bctl->stat.considered++;
3637 spin_unlock(&fs_info->balance_lock);
3640 ret = should_balance_chunk(fs_info, leaf, chunk,
3643 btrfs_release_path(path);
3645 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3650 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3651 spin_lock(&fs_info->balance_lock);
3652 bctl->stat.expected++;
3653 spin_unlock(&fs_info->balance_lock);
3655 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3657 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3659 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3666 * Apply limit_min filter, no need to check if the LIMITS
3667 * filter is used, limit_min is 0 by default
3669 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3670 count_data < bctl->data.limit_min)
3671 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3672 count_meta < bctl->meta.limit_min)
3673 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3674 count_sys < bctl->sys.limit_min)) {
3675 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3679 if (!chunk_reserved) {
3681 * We may be relocating the only data chunk we have,
3682 * which could potentially end up with losing data's
3683 * raid profile, so lets allocate an empty one in
3686 ret = btrfs_may_alloc_data_chunk(fs_info,
3689 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3691 } else if (ret == 1) {
3696 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3697 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3698 if (ret && ret != -ENOSPC)
3700 if (ret == -ENOSPC) {
3703 spin_lock(&fs_info->balance_lock);
3704 bctl->stat.completed++;
3705 spin_unlock(&fs_info->balance_lock);
3708 if (found_key.offset == 0)
3710 key.offset = found_key.offset - 1;
3714 btrfs_release_path(path);
3719 btrfs_free_path(path);
3720 if (enospc_errors) {
3721 btrfs_info(fs_info, "%d enospc errors during balance",
3731 * alloc_profile_is_valid - see if a given profile is valid and reduced
3732 * @flags: profile to validate
3733 * @extended: if true @flags is treated as an extended profile
3735 static int alloc_profile_is_valid(u64 flags, int extended)
3737 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3738 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3740 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3742 /* 1) check that all other bits are zeroed */
3746 /* 2) see if profile is reduced */
3748 return !extended; /* "0" is valid for usual profiles */
3750 /* true if exactly one bit set */
3751 return (flags & (flags - 1)) == 0;
3754 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3756 /* cancel requested || normal exit path */
3757 return atomic_read(&fs_info->balance_cancel_req) ||
3758 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3759 atomic_read(&fs_info->balance_cancel_req) == 0);
3762 /* Non-zero return value signifies invalidity */
3763 static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg,
3766 return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3767 (!alloc_profile_is_valid(bctl_arg->target, 1) ||
3768 (bctl_arg->target & ~allowed)));
3772 * Should be called with balance mutexe held
3774 int btrfs_balance(struct btrfs_fs_info *fs_info,
3775 struct btrfs_balance_control *bctl,
3776 struct btrfs_ioctl_balance_args *bargs)
3778 u64 meta_target, data_target;
3785 if (btrfs_fs_closing(fs_info) ||
3786 atomic_read(&fs_info->balance_pause_req) ||
3787 atomic_read(&fs_info->balance_cancel_req)) {
3792 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3793 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3797 * In case of mixed groups both data and meta should be picked,
3798 * and identical options should be given for both of them.
3800 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3801 if (mixed && (bctl->flags & allowed)) {
3802 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3803 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3804 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3806 "with mixed groups data and metadata balance options must be the same");
3812 num_devices = fs_info->fs_devices->num_devices;
3813 btrfs_dev_replace_read_lock(&fs_info->dev_replace);
3814 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3815 BUG_ON(num_devices < 1);
3818 btrfs_dev_replace_read_unlock(&fs_info->dev_replace);
3819 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE | BTRFS_BLOCK_GROUP_DUP;
3820 if (num_devices > 1)
3821 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3822 if (num_devices > 2)
3823 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3824 if (num_devices > 3)
3825 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3826 BTRFS_BLOCK_GROUP_RAID6);
3827 if (validate_convert_profile(&bctl->data, allowed)) {
3829 "unable to start balance with target data profile %llu",
3834 if (validate_convert_profile(&bctl->meta, allowed)) {
3836 "unable to start balance with target metadata profile %llu",
3841 if (validate_convert_profile(&bctl->sys, allowed)) {
3843 "unable to start balance with target system profile %llu",
3849 /* allow to reduce meta or sys integrity only if force set */
3850 allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3851 BTRFS_BLOCK_GROUP_RAID10 |
3852 BTRFS_BLOCK_GROUP_RAID5 |
3853 BTRFS_BLOCK_GROUP_RAID6;
3855 seq = read_seqbegin(&fs_info->profiles_lock);
3857 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3858 (fs_info->avail_system_alloc_bits & allowed) &&
3859 !(bctl->sys.target & allowed)) ||
3860 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3861 (fs_info->avail_metadata_alloc_bits & allowed) &&
3862 !(bctl->meta.target & allowed))) {
3863 if (bctl->flags & BTRFS_BALANCE_FORCE) {
3865 "force reducing metadata integrity");
3868 "balance will reduce metadata integrity, use force if you want this");
3873 } while (read_seqretry(&fs_info->profiles_lock, seq));
3875 /* if we're not converting, the target field is uninitialized */
3876 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
3877 bctl->meta.target : fs_info->avail_metadata_alloc_bits;
3878 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
3879 bctl->data.target : fs_info->avail_data_alloc_bits;
3880 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <
3881 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {
3883 "metadata profile 0x%llx has lower redundancy than data profile 0x%llx",
3884 meta_target, data_target);
3887 ret = insert_balance_item(fs_info, bctl);
3888 if (ret && ret != -EEXIST)
3891 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3892 BUG_ON(ret == -EEXIST);
3893 BUG_ON(fs_info->balance_ctl);
3894 spin_lock(&fs_info->balance_lock);
3895 fs_info->balance_ctl = bctl;
3896 spin_unlock(&fs_info->balance_lock);
3898 BUG_ON(ret != -EEXIST);
3899 spin_lock(&fs_info->balance_lock);
3900 update_balance_args(bctl);
3901 spin_unlock(&fs_info->balance_lock);
3904 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
3905 set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
3906 mutex_unlock(&fs_info->balance_mutex);
3908 ret = __btrfs_balance(fs_info);
3910 mutex_lock(&fs_info->balance_mutex);
3911 clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
3914 memset(bargs, 0, sizeof(*bargs));
3915 btrfs_update_ioctl_balance_args(fs_info, bargs);
3918 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3919 balance_need_close(fs_info)) {
3920 reset_balance_state(fs_info);
3921 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3924 wake_up(&fs_info->balance_wait_q);
3928 if (bctl->flags & BTRFS_BALANCE_RESUME)
3929 reset_balance_state(fs_info);
3932 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3937 static int balance_kthread(void *data)
3939 struct btrfs_fs_info *fs_info = data;
3942 mutex_lock(&fs_info->balance_mutex);
3943 if (fs_info->balance_ctl) {
3944 btrfs_info(fs_info, "continuing balance");
3945 ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL);
3947 mutex_unlock(&fs_info->balance_mutex);
3952 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3954 struct task_struct *tsk;
3956 mutex_lock(&fs_info->balance_mutex);
3957 if (!fs_info->balance_ctl) {
3958 mutex_unlock(&fs_info->balance_mutex);
3961 mutex_unlock(&fs_info->balance_mutex);
3963 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
3964 btrfs_info(fs_info, "force skipping balance");
3969 * A ro->rw remount sequence should continue with the paused balance
3970 * regardless of who pauses it, system or the user as of now, so set
3973 spin_lock(&fs_info->balance_lock);
3974 fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME;
3975 spin_unlock(&fs_info->balance_lock);
3977 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3978 return PTR_ERR_OR_ZERO(tsk);
3981 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3983 struct btrfs_balance_control *bctl;
3984 struct btrfs_balance_item *item;
3985 struct btrfs_disk_balance_args disk_bargs;
3986 struct btrfs_path *path;
3987 struct extent_buffer *leaf;
3988 struct btrfs_key key;
3991 path = btrfs_alloc_path();
3995 key.objectid = BTRFS_BALANCE_OBJECTID;
3996 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3999 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4002 if (ret > 0) { /* ret = -ENOENT; */
4007 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4013 leaf = path->nodes[0];
4014 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
4016 bctl->flags = btrfs_balance_flags(leaf, item);
4017 bctl->flags |= BTRFS_BALANCE_RESUME;
4019 btrfs_balance_data(leaf, item, &disk_bargs);
4020 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4021 btrfs_balance_meta(leaf, item, &disk_bargs);
4022 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4023 btrfs_balance_sys(leaf, item, &disk_bargs);
4024 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4027 * This should never happen, as the paused balance state is recovered
4028 * during mount without any chance of other exclusive ops to collide.
4030 * This gives the exclusive op status to balance and keeps in paused
4031 * state until user intervention (cancel or umount). If the ownership
4032 * cannot be assigned, show a message but do not fail. The balance
4033 * is in a paused state and must have fs_info::balance_ctl properly
4036 if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags))
4038 "cannot set exclusive op status to balance, resume manually");
4040 mutex_lock(&fs_info->balance_mutex);
4041 BUG_ON(fs_info->balance_ctl);
4042 spin_lock(&fs_info->balance_lock);
4043 fs_info->balance_ctl = bctl;
4044 spin_unlock(&fs_info->balance_lock);
4045 mutex_unlock(&fs_info->balance_mutex);
4047 btrfs_free_path(path);
4051 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4055 mutex_lock(&fs_info->balance_mutex);
4056 if (!fs_info->balance_ctl) {
4057 mutex_unlock(&fs_info->balance_mutex);
4061 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4062 atomic_inc(&fs_info->balance_pause_req);
4063 mutex_unlock(&fs_info->balance_mutex);
4065 wait_event(fs_info->balance_wait_q,
4066 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4068 mutex_lock(&fs_info->balance_mutex);
4069 /* we are good with balance_ctl ripped off from under us */
4070 BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4071 atomic_dec(&fs_info->balance_pause_req);
4076 mutex_unlock(&fs_info->balance_mutex);
4080 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4082 mutex_lock(&fs_info->balance_mutex);
4083 if (!fs_info->balance_ctl) {
4084 mutex_unlock(&fs_info->balance_mutex);
4089 * A paused balance with the item stored on disk can be resumed at
4090 * mount time if the mount is read-write. Otherwise it's still paused
4091 * and we must not allow cancelling as it deletes the item.
4093 if (sb_rdonly(fs_info->sb)) {
4094 mutex_unlock(&fs_info->balance_mutex);
4098 atomic_inc(&fs_info->balance_cancel_req);
4100 * if we are running just wait and return, balance item is
4101 * deleted in btrfs_balance in this case
4103 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4104 mutex_unlock(&fs_info->balance_mutex);
4105 wait_event(fs_info->balance_wait_q,
4106 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4107 mutex_lock(&fs_info->balance_mutex);
4109 mutex_unlock(&fs_info->balance_mutex);
4111 * Lock released to allow other waiters to continue, we'll
4112 * reexamine the status again.
4114 mutex_lock(&fs_info->balance_mutex);
4116 if (fs_info->balance_ctl) {
4117 reset_balance_state(fs_info);
4118 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4122 BUG_ON(fs_info->balance_ctl ||
4123 test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4124 atomic_dec(&fs_info->balance_cancel_req);
4125 mutex_unlock(&fs_info->balance_mutex);
4129 static int btrfs_uuid_scan_kthread(void *data)
4131 struct btrfs_fs_info *fs_info = data;
4132 struct btrfs_root *root = fs_info->tree_root;
4133 struct btrfs_key key;
4134 struct btrfs_path *path = NULL;
4136 struct extent_buffer *eb;
4138 struct btrfs_root_item root_item;
4140 struct btrfs_trans_handle *trans = NULL;
4142 path = btrfs_alloc_path();
4149 key.type = BTRFS_ROOT_ITEM_KEY;
4153 ret = btrfs_search_forward(root, &key, path,
4154 BTRFS_OLDEST_GENERATION);
4161 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4162 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4163 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4164 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4167 eb = path->nodes[0];
4168 slot = path->slots[0];
4169 item_size = btrfs_item_size_nr(eb, slot);
4170 if (item_size < sizeof(root_item))
4173 read_extent_buffer(eb, &root_item,
4174 btrfs_item_ptr_offset(eb, slot),
4175 (int)sizeof(root_item));
4176 if (btrfs_root_refs(&root_item) == 0)
4179 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4180 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4184 btrfs_release_path(path);
4186 * 1 - subvol uuid item
4187 * 1 - received_subvol uuid item
4189 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4190 if (IS_ERR(trans)) {
4191 ret = PTR_ERR(trans);
4199 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4200 ret = btrfs_uuid_tree_add(trans, fs_info,
4202 BTRFS_UUID_KEY_SUBVOL,
4205 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4211 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4212 ret = btrfs_uuid_tree_add(trans, fs_info,
4213 root_item.received_uuid,
4214 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4217 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4225 ret = btrfs_end_transaction(trans);
4231 btrfs_release_path(path);
4232 if (key.offset < (u64)-1) {
4234 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4236 key.type = BTRFS_ROOT_ITEM_KEY;
4237 } else if (key.objectid < (u64)-1) {
4239 key.type = BTRFS_ROOT_ITEM_KEY;
4248 btrfs_free_path(path);
4249 if (trans && !IS_ERR(trans))
4250 btrfs_end_transaction(trans);
4252 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4254 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4255 up(&fs_info->uuid_tree_rescan_sem);
4260 * Callback for btrfs_uuid_tree_iterate().
4262 * 0 check succeeded, the entry is not outdated.
4263 * < 0 if an error occurred.
4264 * > 0 if the check failed, which means the caller shall remove the entry.
4266 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
4267 u8 *uuid, u8 type, u64 subid)
4269 struct btrfs_key key;
4271 struct btrfs_root *subvol_root;
4273 if (type != BTRFS_UUID_KEY_SUBVOL &&
4274 type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
4277 key.objectid = subid;
4278 key.type = BTRFS_ROOT_ITEM_KEY;
4279 key.offset = (u64)-1;
4280 subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
4281 if (IS_ERR(subvol_root)) {
4282 ret = PTR_ERR(subvol_root);
4289 case BTRFS_UUID_KEY_SUBVOL:
4290 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
4293 case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
4294 if (memcmp(uuid, subvol_root->root_item.received_uuid,
4304 static int btrfs_uuid_rescan_kthread(void *data)
4306 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4310 * 1st step is to iterate through the existing UUID tree and
4311 * to delete all entries that contain outdated data.
4312 * 2nd step is to add all missing entries to the UUID tree.
4314 ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4316 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4317 up(&fs_info->uuid_tree_rescan_sem);
4320 return btrfs_uuid_scan_kthread(data);
4323 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4325 struct btrfs_trans_handle *trans;
4326 struct btrfs_root *tree_root = fs_info->tree_root;
4327 struct btrfs_root *uuid_root;
4328 struct task_struct *task;
4335 trans = btrfs_start_transaction(tree_root, 2);
4337 return PTR_ERR(trans);
4339 uuid_root = btrfs_create_tree(trans, fs_info,
4340 BTRFS_UUID_TREE_OBJECTID);
4341 if (IS_ERR(uuid_root)) {
4342 ret = PTR_ERR(uuid_root);
4343 btrfs_abort_transaction(trans, ret);
4344 btrfs_end_transaction(trans);
4348 fs_info->uuid_root = uuid_root;
4350 ret = btrfs_commit_transaction(trans);
4354 down(&fs_info->uuid_tree_rescan_sem);
4355 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4357 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4358 btrfs_warn(fs_info, "failed to start uuid_scan task");
4359 up(&fs_info->uuid_tree_rescan_sem);
4360 return PTR_ERR(task);
4366 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4368 struct task_struct *task;
4370 down(&fs_info->uuid_tree_rescan_sem);
4371 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4373 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4374 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4375 up(&fs_info->uuid_tree_rescan_sem);
4376 return PTR_ERR(task);
4383 * shrinking a device means finding all of the device extents past
4384 * the new size, and then following the back refs to the chunks.
4385 * The chunk relocation code actually frees the device extent
4387 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4389 struct btrfs_fs_info *fs_info = device->fs_info;
4390 struct btrfs_root *root = fs_info->dev_root;
4391 struct btrfs_trans_handle *trans;
4392 struct btrfs_dev_extent *dev_extent = NULL;
4393 struct btrfs_path *path;
4399 bool retried = false;
4400 bool checked_pending_chunks = false;
4401 struct extent_buffer *l;
4402 struct btrfs_key key;
4403 struct btrfs_super_block *super_copy = fs_info->super_copy;
4404 u64 old_total = btrfs_super_total_bytes(super_copy);
4405 u64 old_size = btrfs_device_get_total_bytes(device);
4408 new_size = round_down(new_size, fs_info->sectorsize);
4409 diff = round_down(old_size - new_size, fs_info->sectorsize);
4411 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4414 path = btrfs_alloc_path();
4418 path->reada = READA_BACK;
4420 mutex_lock(&fs_info->chunk_mutex);
4422 btrfs_device_set_total_bytes(device, new_size);
4423 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4424 device->fs_devices->total_rw_bytes -= diff;
4425 atomic64_sub(diff, &fs_info->free_chunk_space);
4427 mutex_unlock(&fs_info->chunk_mutex);
4430 key.objectid = device->devid;
4431 key.offset = (u64)-1;
4432 key.type = BTRFS_DEV_EXTENT_KEY;
4435 mutex_lock(&fs_info->delete_unused_bgs_mutex);
4436 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4438 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4442 ret = btrfs_previous_item(root, path, 0, key.type);
4444 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4449 btrfs_release_path(path);
4454 slot = path->slots[0];
4455 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4457 if (key.objectid != device->devid) {
4458 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4459 btrfs_release_path(path);
4463 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4464 length = btrfs_dev_extent_length(l, dev_extent);
4466 if (key.offset + length <= new_size) {
4467 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4468 btrfs_release_path(path);
4472 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4473 btrfs_release_path(path);
4476 * We may be relocating the only data chunk we have,
4477 * which could potentially end up with losing data's
4478 * raid profile, so lets allocate an empty one in
4481 ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset);
4483 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4487 ret = btrfs_relocate_chunk(fs_info, chunk_offset);
4488 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4489 if (ret && ret != -ENOSPC)
4493 } while (key.offset-- > 0);
4495 if (failed && !retried) {
4499 } else if (failed && retried) {
4504 /* Shrinking succeeded, else we would be at "done". */
4505 trans = btrfs_start_transaction(root, 0);
4506 if (IS_ERR(trans)) {
4507 ret = PTR_ERR(trans);
4511 mutex_lock(&fs_info->chunk_mutex);
4514 * We checked in the above loop all device extents that were already in
4515 * the device tree. However before we have updated the device's
4516 * total_bytes to the new size, we might have had chunk allocations that
4517 * have not complete yet (new block groups attached to transaction
4518 * handles), and therefore their device extents were not yet in the
4519 * device tree and we missed them in the loop above. So if we have any
4520 * pending chunk using a device extent that overlaps the device range
4521 * that we can not use anymore, commit the current transaction and
4522 * repeat the search on the device tree - this way we guarantee we will
4523 * not have chunks using device extents that end beyond 'new_size'.
4525 if (!checked_pending_chunks) {
4526 u64 start = new_size;
4527 u64 len = old_size - new_size;
4529 if (contains_pending_extent(trans->transaction, device,
4531 mutex_unlock(&fs_info->chunk_mutex);
4532 checked_pending_chunks = true;
4535 ret = btrfs_commit_transaction(trans);
4542 btrfs_device_set_disk_total_bytes(device, new_size);
4543 if (list_empty(&device->resized_list))
4544 list_add_tail(&device->resized_list,
4545 &fs_info->fs_devices->resized_devices);
4547 WARN_ON(diff > old_total);
4548 btrfs_set_super_total_bytes(super_copy,
4549 round_down(old_total - diff, fs_info->sectorsize));
4550 mutex_unlock(&fs_info->chunk_mutex);
4552 /* Now btrfs_update_device() will change the on-disk size. */
4553 ret = btrfs_update_device(trans, device);
4554 btrfs_end_transaction(trans);
4556 btrfs_free_path(path);
4558 mutex_lock(&fs_info->chunk_mutex);
4559 btrfs_device_set_total_bytes(device, old_size);
4560 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
4561 device->fs_devices->total_rw_bytes += diff;
4562 atomic64_add(diff, &fs_info->free_chunk_space);
4563 mutex_unlock(&fs_info->chunk_mutex);
4568 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
4569 struct btrfs_key *key,
4570 struct btrfs_chunk *chunk, int item_size)
4572 struct btrfs_super_block *super_copy = fs_info->super_copy;
4573 struct btrfs_disk_key disk_key;
4577 mutex_lock(&fs_info->chunk_mutex);
4578 array_size = btrfs_super_sys_array_size(super_copy);
4579 if (array_size + item_size + sizeof(disk_key)
4580 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4581 mutex_unlock(&fs_info->chunk_mutex);
4585 ptr = super_copy->sys_chunk_array + array_size;
4586 btrfs_cpu_key_to_disk(&disk_key, key);
4587 memcpy(ptr, &disk_key, sizeof(disk_key));
4588 ptr += sizeof(disk_key);
4589 memcpy(ptr, chunk, item_size);
4590 item_size += sizeof(disk_key);
4591 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4592 mutex_unlock(&fs_info->chunk_mutex);
4598 * sort the devices in descending order by max_avail, total_avail
4600 static int btrfs_cmp_device_info(const void *a, const void *b)
4602 const struct btrfs_device_info *di_a = a;
4603 const struct btrfs_device_info *di_b = b;
4605 if (di_a->max_avail > di_b->max_avail)
4607 if (di_a->max_avail < di_b->max_avail)
4609 if (di_a->total_avail > di_b->total_avail)
4611 if (di_a->total_avail < di_b->total_avail)
4616 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4618 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4621 btrfs_set_fs_incompat(info, RAID56);
4624 #define BTRFS_MAX_DEVS(info) ((BTRFS_MAX_ITEM_SIZE(info) \
4625 - sizeof(struct btrfs_chunk)) \
4626 / sizeof(struct btrfs_stripe) + 1)
4628 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
4629 - 2 * sizeof(struct btrfs_disk_key) \
4630 - 2 * sizeof(struct btrfs_chunk)) \
4631 / sizeof(struct btrfs_stripe) + 1)
4633 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4634 u64 start, u64 type)
4636 struct btrfs_fs_info *info = trans->fs_info;
4637 struct btrfs_fs_devices *fs_devices = info->fs_devices;
4638 struct btrfs_device *device;
4639 struct map_lookup *map = NULL;
4640 struct extent_map_tree *em_tree;
4641 struct extent_map *em;
4642 struct btrfs_device_info *devices_info = NULL;
4644 int num_stripes; /* total number of stripes to allocate */
4645 int data_stripes; /* number of stripes that count for
4647 int sub_stripes; /* sub_stripes info for map */
4648 int dev_stripes; /* stripes per dev */
4649 int devs_max; /* max devs to use */
4650 int devs_min; /* min devs needed */
4651 int devs_increment; /* ndevs has to be a multiple of this */
4652 int ncopies; /* how many copies to data has */
4654 u64 max_stripe_size;
4663 BUG_ON(!alloc_profile_is_valid(type, 0));
4665 if (list_empty(&fs_devices->alloc_list)) {
4666 if (btrfs_test_opt(info, ENOSPC_DEBUG))
4667 btrfs_debug(info, "%s: no writable device", __func__);
4671 index = btrfs_bg_flags_to_raid_index(type);
4673 sub_stripes = btrfs_raid_array[index].sub_stripes;
4674 dev_stripes = btrfs_raid_array[index].dev_stripes;
4675 devs_max = btrfs_raid_array[index].devs_max;
4676 devs_min = btrfs_raid_array[index].devs_min;
4677 devs_increment = btrfs_raid_array[index].devs_increment;
4678 ncopies = btrfs_raid_array[index].ncopies;
4680 if (type & BTRFS_BLOCK_GROUP_DATA) {
4681 max_stripe_size = SZ_1G;
4682 max_chunk_size = 10 * max_stripe_size;
4684 devs_max = BTRFS_MAX_DEVS(info);
4685 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4686 /* for larger filesystems, use larger metadata chunks */
4687 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
4688 max_stripe_size = SZ_1G;
4690 max_stripe_size = SZ_256M;
4691 max_chunk_size = max_stripe_size;
4693 devs_max = BTRFS_MAX_DEVS(info);
4694 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4695 max_stripe_size = SZ_32M;
4696 max_chunk_size = 2 * max_stripe_size;
4698 devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4700 btrfs_err(info, "invalid chunk type 0x%llx requested",
4705 /* we don't want a chunk larger than 10% of writeable space */
4706 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4709 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4715 * in the first pass through the devices list, we gather information
4716 * about the available holes on each device.
4719 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
4723 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4725 "BTRFS: read-only device in alloc_list\n");
4729 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
4730 &device->dev_state) ||
4731 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4734 if (device->total_bytes > device->bytes_used)
4735 total_avail = device->total_bytes - device->bytes_used;
4739 /* If there is no space on this device, skip it. */
4740 if (total_avail == 0)
4743 ret = find_free_dev_extent(trans, device,
4744 max_stripe_size * dev_stripes,
4745 &dev_offset, &max_avail);
4746 if (ret && ret != -ENOSPC)
4750 max_avail = max_stripe_size * dev_stripes;
4752 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes) {
4753 if (btrfs_test_opt(info, ENOSPC_DEBUG))
4755 "%s: devid %llu has no free space, have=%llu want=%u",
4756 __func__, device->devid, max_avail,
4757 BTRFS_STRIPE_LEN * dev_stripes);
4761 if (ndevs == fs_devices->rw_devices) {
4762 WARN(1, "%s: found more than %llu devices\n",
4763 __func__, fs_devices->rw_devices);
4766 devices_info[ndevs].dev_offset = dev_offset;
4767 devices_info[ndevs].max_avail = max_avail;
4768 devices_info[ndevs].total_avail = total_avail;
4769 devices_info[ndevs].dev = device;
4774 * now sort the devices by hole size / available space
4776 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4777 btrfs_cmp_device_info, NULL);
4779 /* round down to number of usable stripes */
4780 ndevs = round_down(ndevs, devs_increment);
4782 if (ndevs < devs_min) {
4784 if (btrfs_test_opt(info, ENOSPC_DEBUG)) {
4786 "%s: not enough devices with free space: have=%d minimum required=%d",
4787 __func__, ndevs, devs_min);
4792 ndevs = min(ndevs, devs_max);
4795 * The primary goal is to maximize the number of stripes, so use as
4796 * many devices as possible, even if the stripes are not maximum sized.
4798 * The DUP profile stores more than one stripe per device, the
4799 * max_avail is the total size so we have to adjust.
4801 stripe_size = div_u64(devices_info[ndevs - 1].max_avail, dev_stripes);
4802 num_stripes = ndevs * dev_stripes;
4805 * this will have to be fixed for RAID1 and RAID10 over
4808 data_stripes = num_stripes / ncopies;
4810 if (type & BTRFS_BLOCK_GROUP_RAID5)
4811 data_stripes = num_stripes - 1;
4813 if (type & BTRFS_BLOCK_GROUP_RAID6)
4814 data_stripes = num_stripes - 2;
4817 * Use the number of data stripes to figure out how big this chunk
4818 * is really going to be in terms of logical address space,
4819 * and compare that answer with the max chunk size
4821 if (stripe_size * data_stripes > max_chunk_size) {
4822 stripe_size = div_u64(max_chunk_size, data_stripes);
4824 /* bump the answer up to a 16MB boundary */
4825 stripe_size = round_up(stripe_size, SZ_16M);
4828 * But don't go higher than the limits we found while searching
4831 stripe_size = min(devices_info[ndevs - 1].max_avail,
4835 /* align to BTRFS_STRIPE_LEN */
4836 stripe_size = round_down(stripe_size, BTRFS_STRIPE_LEN);
4838 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4843 map->num_stripes = num_stripes;
4845 for (i = 0; i < ndevs; ++i) {
4846 for (j = 0; j < dev_stripes; ++j) {
4847 int s = i * dev_stripes + j;
4848 map->stripes[s].dev = devices_info[i].dev;
4849 map->stripes[s].physical = devices_info[i].dev_offset +
4853 map->stripe_len = BTRFS_STRIPE_LEN;
4854 map->io_align = BTRFS_STRIPE_LEN;
4855 map->io_width = BTRFS_STRIPE_LEN;
4857 map->sub_stripes = sub_stripes;
4859 num_bytes = stripe_size * data_stripes;
4861 trace_btrfs_chunk_alloc(info, map, start, num_bytes);
4863 em = alloc_extent_map();
4869 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4870 em->map_lookup = map;
4872 em->len = num_bytes;
4873 em->block_start = 0;
4874 em->block_len = em->len;
4875 em->orig_block_len = stripe_size;
4877 em_tree = &info->mapping_tree.map_tree;
4878 write_lock(&em_tree->lock);
4879 ret = add_extent_mapping(em_tree, em, 0);
4881 write_unlock(&em_tree->lock);
4882 free_extent_map(em);
4886 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4887 refcount_inc(&em->refs);
4888 write_unlock(&em_tree->lock);
4890 ret = btrfs_make_block_group(trans, info, 0, type, start, num_bytes);
4892 goto error_del_extent;
4894 for (i = 0; i < map->num_stripes; i++) {
4895 num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4896 btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4899 atomic64_sub(stripe_size * map->num_stripes, &info->free_chunk_space);
4901 free_extent_map(em);
4902 check_raid56_incompat_flag(info, type);
4904 kfree(devices_info);
4908 write_lock(&em_tree->lock);
4909 remove_extent_mapping(em_tree, em);
4910 write_unlock(&em_tree->lock);
4912 /* One for our allocation */
4913 free_extent_map(em);
4914 /* One for the tree reference */
4915 free_extent_map(em);
4916 /* One for the pending_chunks list reference */
4917 free_extent_map(em);
4919 kfree(devices_info);
4923 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4924 struct btrfs_fs_info *fs_info,
4925 u64 chunk_offset, u64 chunk_size)
4927 struct btrfs_root *extent_root = fs_info->extent_root;
4928 struct btrfs_root *chunk_root = fs_info->chunk_root;
4929 struct btrfs_key key;
4930 struct btrfs_device *device;
4931 struct btrfs_chunk *chunk;
4932 struct btrfs_stripe *stripe;
4933 struct extent_map *em;
4934 struct map_lookup *map;
4941 em = get_chunk_map(fs_info, chunk_offset, chunk_size);
4945 map = em->map_lookup;
4946 item_size = btrfs_chunk_item_size(map->num_stripes);
4947 stripe_size = em->orig_block_len;
4949 chunk = kzalloc(item_size, GFP_NOFS);
4956 * Take the device list mutex to prevent races with the final phase of
4957 * a device replace operation that replaces the device object associated
4958 * with the map's stripes, because the device object's id can change
4959 * at any time during that final phase of the device replace operation
4960 * (dev-replace.c:btrfs_dev_replace_finishing()).
4962 mutex_lock(&fs_info->fs_devices->device_list_mutex);
4963 for (i = 0; i < map->num_stripes; i++) {
4964 device = map->stripes[i].dev;
4965 dev_offset = map->stripes[i].physical;
4967 ret = btrfs_update_device(trans, device);
4970 ret = btrfs_alloc_dev_extent(trans, device, chunk_offset,
4971 dev_offset, stripe_size);
4976 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4980 stripe = &chunk->stripe;
4981 for (i = 0; i < map->num_stripes; i++) {
4982 device = map->stripes[i].dev;
4983 dev_offset = map->stripes[i].physical;
4985 btrfs_set_stack_stripe_devid(stripe, device->devid);
4986 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4987 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4990 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4992 btrfs_set_stack_chunk_length(chunk, chunk_size);
4993 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4994 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4995 btrfs_set_stack_chunk_type(chunk, map->type);
4996 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4997 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4998 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4999 btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
5000 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
5002 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
5003 key.type = BTRFS_CHUNK_ITEM_KEY;
5004 key.offset = chunk_offset;
5006 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
5007 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
5009 * TODO: Cleanup of inserted chunk root in case of
5012 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
5017 free_extent_map(em);
5022 * Chunk allocation falls into two parts. The first part does works
5023 * that make the new allocated chunk useable, but not do any operation
5024 * that modifies the chunk tree. The second part does the works that
5025 * require modifying the chunk tree. This division is important for the
5026 * bootstrap process of adding storage to a seed btrfs.
5028 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
5029 struct btrfs_fs_info *fs_info, u64 type)
5033 lockdep_assert_held(&fs_info->chunk_mutex);
5034 chunk_offset = find_next_chunk(fs_info);
5035 return __btrfs_alloc_chunk(trans, chunk_offset, type);
5038 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
5039 struct btrfs_fs_info *fs_info)
5042 u64 sys_chunk_offset;
5046 chunk_offset = find_next_chunk(fs_info);
5047 alloc_profile = btrfs_metadata_alloc_profile(fs_info);
5048 ret = __btrfs_alloc_chunk(trans, chunk_offset, alloc_profile);
5052 sys_chunk_offset = find_next_chunk(fs_info);
5053 alloc_profile = btrfs_system_alloc_profile(fs_info);
5054 ret = __btrfs_alloc_chunk(trans, sys_chunk_offset, alloc_profile);
5058 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
5062 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5063 BTRFS_BLOCK_GROUP_RAID10 |
5064 BTRFS_BLOCK_GROUP_RAID5 |
5065 BTRFS_BLOCK_GROUP_DUP)) {
5067 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
5076 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset)
5078 struct extent_map *em;
5079 struct map_lookup *map;
5084 em = get_chunk_map(fs_info, chunk_offset, 1);
5088 map = em->map_lookup;
5089 for (i = 0; i < map->num_stripes; i++) {
5090 if (test_bit(BTRFS_DEV_STATE_MISSING,
5091 &map->stripes[i].dev->dev_state)) {
5095 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE,
5096 &map->stripes[i].dev->dev_state)) {
5103 * If the number of missing devices is larger than max errors,
5104 * we can not write the data into that chunk successfully, so
5107 if (miss_ndevs > btrfs_chunk_max_errors(map))
5110 free_extent_map(em);
5114 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
5116 extent_map_tree_init(&tree->map_tree);
5119 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
5121 struct extent_map *em;
5124 write_lock(&tree->map_tree.lock);
5125 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
5127 remove_extent_mapping(&tree->map_tree, em);
5128 write_unlock(&tree->map_tree.lock);
5132 free_extent_map(em);
5133 /* once for the tree */
5134 free_extent_map(em);
5138 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5140 struct extent_map *em;
5141 struct map_lookup *map;
5144 em = get_chunk_map(fs_info, logical, len);
5147 * We could return errors for these cases, but that could get
5148 * ugly and we'd probably do the same thing which is just not do
5149 * anything else and exit, so return 1 so the callers don't try
5150 * to use other copies.
5154 map = em->map_lookup;
5155 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
5156 ret = map->num_stripes;
5157 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5158 ret = map->sub_stripes;
5159 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5161 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5163 * There could be two corrupted data stripes, we need
5164 * to loop retry in order to rebuild the correct data.
5166 * Fail a stripe at a time on every retry except the
5167 * stripe under reconstruction.
5169 ret = map->num_stripes;
5172 free_extent_map(em);
5174 btrfs_dev_replace_read_lock(&fs_info->dev_replace);
5175 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) &&
5176 fs_info->dev_replace.tgtdev)
5178 btrfs_dev_replace_read_unlock(&fs_info->dev_replace);
5183 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
5186 struct extent_map *em;
5187 struct map_lookup *map;
5188 unsigned long len = fs_info->sectorsize;
5190 em = get_chunk_map(fs_info, logical, len);
5192 if (!WARN_ON(IS_ERR(em))) {
5193 map = em->map_lookup;
5194 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5195 len = map->stripe_len * nr_data_stripes(map);
5196 free_extent_map(em);
5201 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5203 struct extent_map *em;
5204 struct map_lookup *map;
5207 em = get_chunk_map(fs_info, logical, len);
5209 if(!WARN_ON(IS_ERR(em))) {
5210 map = em->map_lookup;
5211 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5213 free_extent_map(em);
5218 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5219 struct map_lookup *map, int first,
5220 int dev_replace_is_ongoing)
5224 int preferred_mirror;
5226 struct btrfs_device *srcdev;
5229 (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)));
5231 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5232 num_stripes = map->sub_stripes;
5234 num_stripes = map->num_stripes;
5236 preferred_mirror = first + current->pid % num_stripes;
5238 if (dev_replace_is_ongoing &&
5239 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5240 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5241 srcdev = fs_info->dev_replace.srcdev;
5246 * try to avoid the drive that is the source drive for a
5247 * dev-replace procedure, only choose it if no other non-missing
5248 * mirror is available
5250 for (tolerance = 0; tolerance < 2; tolerance++) {
5251 if (map->stripes[preferred_mirror].dev->bdev &&
5252 (tolerance || map->stripes[preferred_mirror].dev != srcdev))
5253 return preferred_mirror;
5254 for (i = first; i < first + num_stripes; i++) {
5255 if (map->stripes[i].dev->bdev &&
5256 (tolerance || map->stripes[i].dev != srcdev))
5261 /* we couldn't find one that doesn't fail. Just return something
5262 * and the io error handling code will clean up eventually
5264 return preferred_mirror;
5267 static inline int parity_smaller(u64 a, u64 b)
5272 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5273 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5275 struct btrfs_bio_stripe s;
5282 for (i = 0; i < num_stripes - 1; i++) {
5283 if (parity_smaller(bbio->raid_map[i],
5284 bbio->raid_map[i+1])) {
5285 s = bbio->stripes[i];
5286 l = bbio->raid_map[i];
5287 bbio->stripes[i] = bbio->stripes[i+1];
5288 bbio->raid_map[i] = bbio->raid_map[i+1];
5289 bbio->stripes[i+1] = s;
5290 bbio->raid_map[i+1] = l;
5298 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5300 struct btrfs_bio *bbio = kzalloc(
5301 /* the size of the btrfs_bio */
5302 sizeof(struct btrfs_bio) +
5303 /* plus the variable array for the stripes */
5304 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5305 /* plus the variable array for the tgt dev */
5306 sizeof(int) * (real_stripes) +
5308 * plus the raid_map, which includes both the tgt dev
5311 sizeof(u64) * (total_stripes),
5312 GFP_NOFS|__GFP_NOFAIL);
5314 atomic_set(&bbio->error, 0);
5315 refcount_set(&bbio->refs, 1);
5320 void btrfs_get_bbio(struct btrfs_bio *bbio)
5322 WARN_ON(!refcount_read(&bbio->refs));
5323 refcount_inc(&bbio->refs);
5326 void btrfs_put_bbio(struct btrfs_bio *bbio)
5330 if (refcount_dec_and_test(&bbio->refs))
5334 /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */
5336 * Please note that, discard won't be sent to target device of device
5339 static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info,
5340 u64 logical, u64 length,
5341 struct btrfs_bio **bbio_ret)
5343 struct extent_map *em;
5344 struct map_lookup *map;
5345 struct btrfs_bio *bbio;
5349 u64 stripe_end_offset;
5356 u32 sub_stripes = 0;
5357 u64 stripes_per_dev = 0;
5358 u32 remaining_stripes = 0;
5359 u32 last_stripe = 0;
5363 /* discard always return a bbio */
5366 em = get_chunk_map(fs_info, logical, length);
5370 map = em->map_lookup;
5371 /* we don't discard raid56 yet */
5372 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5377 offset = logical - em->start;
5378 length = min_t(u64, em->len - offset, length);
5380 stripe_len = map->stripe_len;
5382 * stripe_nr counts the total number of stripes we have to stride
5383 * to get to this block
5385 stripe_nr = div64_u64(offset, stripe_len);
5387 /* stripe_offset is the offset of this block in its stripe */
5388 stripe_offset = offset - stripe_nr * stripe_len;
5390 stripe_nr_end = round_up(offset + length, map->stripe_len);
5391 stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len);
5392 stripe_cnt = stripe_nr_end - stripe_nr;
5393 stripe_end_offset = stripe_nr_end * map->stripe_len -
5396 * after this, stripe_nr is the number of stripes on this
5397 * device we have to walk to find the data, and stripe_index is
5398 * the number of our device in the stripe array
5402 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5403 BTRFS_BLOCK_GROUP_RAID10)) {
5404 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5407 sub_stripes = map->sub_stripes;
5409 factor = map->num_stripes / sub_stripes;
5410 num_stripes = min_t(u64, map->num_stripes,
5411 sub_stripes * stripe_cnt);
5412 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5413 stripe_index *= sub_stripes;
5414 stripes_per_dev = div_u64_rem(stripe_cnt, factor,
5415 &remaining_stripes);
5416 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5417 last_stripe *= sub_stripes;
5418 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5419 BTRFS_BLOCK_GROUP_DUP)) {
5420 num_stripes = map->num_stripes;
5422 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5426 bbio = alloc_btrfs_bio(num_stripes, 0);
5432 for (i = 0; i < num_stripes; i++) {
5433 bbio->stripes[i].physical =
5434 map->stripes[stripe_index].physical +
5435 stripe_offset + stripe_nr * map->stripe_len;
5436 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5438 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5439 BTRFS_BLOCK_GROUP_RAID10)) {
5440 bbio->stripes[i].length = stripes_per_dev *
5443 if (i / sub_stripes < remaining_stripes)
5444 bbio->stripes[i].length +=
5448 * Special for the first stripe and
5451 * |-------|...|-------|
5455 if (i < sub_stripes)
5456 bbio->stripes[i].length -=
5459 if (stripe_index >= last_stripe &&
5460 stripe_index <= (last_stripe +
5462 bbio->stripes[i].length -=
5465 if (i == sub_stripes - 1)
5468 bbio->stripes[i].length = length;
5472 if (stripe_index == map->num_stripes) {
5479 bbio->map_type = map->type;
5480 bbio->num_stripes = num_stripes;
5482 free_extent_map(em);
5487 * In dev-replace case, for repair case (that's the only case where the mirror
5488 * is selected explicitly when calling btrfs_map_block), blocks left of the
5489 * left cursor can also be read from the target drive.
5491 * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the
5493 * For READ, it also needs to be supported using the same mirror number.
5495 * If the requested block is not left of the left cursor, EIO is returned. This
5496 * can happen because btrfs_num_copies() returns one more in the dev-replace
5499 static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info,
5500 u64 logical, u64 length,
5501 u64 srcdev_devid, int *mirror_num,
5504 struct btrfs_bio *bbio = NULL;
5506 int index_srcdev = 0;
5508 u64 physical_of_found = 0;
5512 ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS,
5513 logical, &length, &bbio, 0, 0);
5515 ASSERT(bbio == NULL);
5519 num_stripes = bbio->num_stripes;
5520 if (*mirror_num > num_stripes) {
5522 * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror,
5523 * that means that the requested area is not left of the left
5526 btrfs_put_bbio(bbio);
5531 * process the rest of the function using the mirror_num of the source
5532 * drive. Therefore look it up first. At the end, patch the device
5533 * pointer to the one of the target drive.
5535 for (i = 0; i < num_stripes; i++) {
5536 if (bbio->stripes[i].dev->devid != srcdev_devid)
5540 * In case of DUP, in order to keep it simple, only add the
5541 * mirror with the lowest physical address
5544 physical_of_found <= bbio->stripes[i].physical)
5549 physical_of_found = bbio->stripes[i].physical;
5552 btrfs_put_bbio(bbio);
5558 *mirror_num = index_srcdev + 1;
5559 *physical = physical_of_found;
5563 static void handle_ops_on_dev_replace(enum btrfs_map_op op,
5564 struct btrfs_bio **bbio_ret,
5565 struct btrfs_dev_replace *dev_replace,
5566 int *num_stripes_ret, int *max_errors_ret)
5568 struct btrfs_bio *bbio = *bbio_ret;
5569 u64 srcdev_devid = dev_replace->srcdev->devid;
5570 int tgtdev_indexes = 0;
5571 int num_stripes = *num_stripes_ret;
5572 int max_errors = *max_errors_ret;
5575 if (op == BTRFS_MAP_WRITE) {
5576 int index_where_to_add;
5579 * duplicate the write operations while the dev replace
5580 * procedure is running. Since the copying of the old disk to
5581 * the new disk takes place at run time while the filesystem is
5582 * mounted writable, the regular write operations to the old
5583 * disk have to be duplicated to go to the new disk as well.
5585 * Note that device->missing is handled by the caller, and that
5586 * the write to the old disk is already set up in the stripes
5589 index_where_to_add = num_stripes;
5590 for (i = 0; i < num_stripes; i++) {
5591 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5592 /* write to new disk, too */
5593 struct btrfs_bio_stripe *new =
5594 bbio->stripes + index_where_to_add;
5595 struct btrfs_bio_stripe *old =
5598 new->physical = old->physical;
5599 new->length = old->length;
5600 new->dev = dev_replace->tgtdev;
5601 bbio->tgtdev_map[i] = index_where_to_add;
5602 index_where_to_add++;
5607 num_stripes = index_where_to_add;
5608 } else if (op == BTRFS_MAP_GET_READ_MIRRORS) {
5609 int index_srcdev = 0;
5611 u64 physical_of_found = 0;
5614 * During the dev-replace procedure, the target drive can also
5615 * be used to read data in case it is needed to repair a corrupt
5616 * block elsewhere. This is possible if the requested area is
5617 * left of the left cursor. In this area, the target drive is a
5618 * full copy of the source drive.
5620 for (i = 0; i < num_stripes; i++) {
5621 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5623 * In case of DUP, in order to keep it simple,
5624 * only add the mirror with the lowest physical
5628 physical_of_found <=
5629 bbio->stripes[i].physical)
5633 physical_of_found = bbio->stripes[i].physical;
5637 struct btrfs_bio_stripe *tgtdev_stripe =
5638 bbio->stripes + num_stripes;
5640 tgtdev_stripe->physical = physical_of_found;
5641 tgtdev_stripe->length =
5642 bbio->stripes[index_srcdev].length;
5643 tgtdev_stripe->dev = dev_replace->tgtdev;
5644 bbio->tgtdev_map[index_srcdev] = num_stripes;
5651 *num_stripes_ret = num_stripes;
5652 *max_errors_ret = max_errors;
5653 bbio->num_tgtdevs = tgtdev_indexes;
5657 static bool need_full_stripe(enum btrfs_map_op op)
5659 return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS);
5662 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
5663 enum btrfs_map_op op,
5664 u64 logical, u64 *length,
5665 struct btrfs_bio **bbio_ret,
5666 int mirror_num, int need_raid_map)
5668 struct extent_map *em;
5669 struct map_lookup *map;
5679 int tgtdev_indexes = 0;
5680 struct btrfs_bio *bbio = NULL;
5681 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
5682 int dev_replace_is_ongoing = 0;
5683 int num_alloc_stripes;
5684 int patch_the_first_stripe_for_dev_replace = 0;
5685 u64 physical_to_patch_in_first_stripe = 0;
5686 u64 raid56_full_stripe_start = (u64)-1;
5688 if (op == BTRFS_MAP_DISCARD)
5689 return __btrfs_map_block_for_discard(fs_info, logical,
5692 em = get_chunk_map(fs_info, logical, *length);
5696 map = em->map_lookup;
5697 offset = logical - em->start;
5699 stripe_len = map->stripe_len;
5702 * stripe_nr counts the total number of stripes we have to stride
5703 * to get to this block
5705 stripe_nr = div64_u64(stripe_nr, stripe_len);
5707 stripe_offset = stripe_nr * stripe_len;
5708 if (offset < stripe_offset) {
5710 "stripe math has gone wrong, stripe_offset=%llu, offset=%llu, start=%llu, logical=%llu, stripe_len=%llu",
5711 stripe_offset, offset, em->start, logical,
5713 free_extent_map(em);
5717 /* stripe_offset is the offset of this block in its stripe*/
5718 stripe_offset = offset - stripe_offset;
5720 /* if we're here for raid56, we need to know the stripe aligned start */
5721 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5722 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5723 raid56_full_stripe_start = offset;
5725 /* allow a write of a full stripe, but make sure we don't
5726 * allow straddling of stripes
5728 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5730 raid56_full_stripe_start *= full_stripe_len;
5733 if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5735 /* For writes to RAID[56], allow a full stripeset across all disks.
5736 For other RAID types and for RAID[56] reads, just allow a single
5737 stripe (on a single disk). */
5738 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
5739 (op == BTRFS_MAP_WRITE)) {
5740 max_len = stripe_len * nr_data_stripes(map) -
5741 (offset - raid56_full_stripe_start);
5743 /* we limit the length of each bio to what fits in a stripe */
5744 max_len = stripe_len - stripe_offset;
5746 *length = min_t(u64, em->len - offset, max_len);
5748 *length = em->len - offset;
5751 /* This is for when we're called from btrfs_merge_bio_hook() and all
5752 it cares about is the length */
5756 btrfs_dev_replace_read_lock(dev_replace);
5757 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5758 if (!dev_replace_is_ongoing)
5759 btrfs_dev_replace_read_unlock(dev_replace);
5761 btrfs_dev_replace_set_lock_blocking(dev_replace);
5763 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
5764 !need_full_stripe(op) && dev_replace->tgtdev != NULL) {
5765 ret = get_extra_mirror_from_replace(fs_info, logical, *length,
5766 dev_replace->srcdev->devid,
5768 &physical_to_patch_in_first_stripe);
5772 patch_the_first_stripe_for_dev_replace = 1;
5773 } else if (mirror_num > map->num_stripes) {
5779 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5780 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5782 if (!need_full_stripe(op))
5784 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
5785 if (need_full_stripe(op))
5786 num_stripes = map->num_stripes;
5787 else if (mirror_num)
5788 stripe_index = mirror_num - 1;
5790 stripe_index = find_live_mirror(fs_info, map, 0,
5791 dev_replace_is_ongoing);
5792 mirror_num = stripe_index + 1;
5795 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
5796 if (need_full_stripe(op)) {
5797 num_stripes = map->num_stripes;
5798 } else if (mirror_num) {
5799 stripe_index = mirror_num - 1;
5804 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5805 u32 factor = map->num_stripes / map->sub_stripes;
5807 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5808 stripe_index *= map->sub_stripes;
5810 if (need_full_stripe(op))
5811 num_stripes = map->sub_stripes;
5812 else if (mirror_num)
5813 stripe_index += mirror_num - 1;
5815 int old_stripe_index = stripe_index;
5816 stripe_index = find_live_mirror(fs_info, map,
5818 dev_replace_is_ongoing);
5819 mirror_num = stripe_index - old_stripe_index + 1;
5822 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5823 if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) {
5824 /* push stripe_nr back to the start of the full stripe */
5825 stripe_nr = div64_u64(raid56_full_stripe_start,
5826 stripe_len * nr_data_stripes(map));
5828 /* RAID[56] write or recovery. Return all stripes */
5829 num_stripes = map->num_stripes;
5830 max_errors = nr_parity_stripes(map);
5832 *length = map->stripe_len;
5837 * Mirror #0 or #1 means the original data block.
5838 * Mirror #2 is RAID5 parity block.
5839 * Mirror #3 is RAID6 Q block.
5841 stripe_nr = div_u64_rem(stripe_nr,
5842 nr_data_stripes(map), &stripe_index);
5844 stripe_index = nr_data_stripes(map) +
5847 /* We distribute the parity blocks across stripes */
5848 div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
5850 if (!need_full_stripe(op) && mirror_num <= 1)
5855 * after this, stripe_nr is the number of stripes on this
5856 * device we have to walk to find the data, and stripe_index is
5857 * the number of our device in the stripe array
5859 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5861 mirror_num = stripe_index + 1;
5863 if (stripe_index >= map->num_stripes) {
5865 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
5866 stripe_index, map->num_stripes);
5871 num_alloc_stripes = num_stripes;
5872 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) {
5873 if (op == BTRFS_MAP_WRITE)
5874 num_alloc_stripes <<= 1;
5875 if (op == BTRFS_MAP_GET_READ_MIRRORS)
5876 num_alloc_stripes++;
5877 tgtdev_indexes = num_stripes;
5880 bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
5885 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL)
5886 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
5888 /* build raid_map */
5889 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map &&
5890 (need_full_stripe(op) || mirror_num > 1)) {
5894 bbio->raid_map = (u64 *)((void *)bbio->stripes +
5895 sizeof(struct btrfs_bio_stripe) *
5897 sizeof(int) * tgtdev_indexes);
5899 /* Work out the disk rotation on this stripe-set */
5900 div_u64_rem(stripe_nr, num_stripes, &rot);
5902 /* Fill in the logical address of each stripe */
5903 tmp = stripe_nr * nr_data_stripes(map);
5904 for (i = 0; i < nr_data_stripes(map); i++)
5905 bbio->raid_map[(i+rot) % num_stripes] =
5906 em->start + (tmp + i) * map->stripe_len;
5908 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5909 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5910 bbio->raid_map[(i+rot+1) % num_stripes] =
5915 for (i = 0; i < num_stripes; i++) {
5916 bbio->stripes[i].physical =
5917 map->stripes[stripe_index].physical +
5919 stripe_nr * map->stripe_len;
5920 bbio->stripes[i].dev =
5921 map->stripes[stripe_index].dev;
5925 if (need_full_stripe(op))
5926 max_errors = btrfs_chunk_max_errors(map);
5929 sort_parity_stripes(bbio, num_stripes);
5931 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
5932 need_full_stripe(op)) {
5933 handle_ops_on_dev_replace(op, &bbio, dev_replace, &num_stripes,
5938 bbio->map_type = map->type;
5939 bbio->num_stripes = num_stripes;
5940 bbio->max_errors = max_errors;
5941 bbio->mirror_num = mirror_num;
5944 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5945 * mirror_num == num_stripes + 1 && dev_replace target drive is
5946 * available as a mirror
5948 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5949 WARN_ON(num_stripes > 1);
5950 bbio->stripes[0].dev = dev_replace->tgtdev;
5951 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5952 bbio->mirror_num = map->num_stripes + 1;
5955 if (dev_replace_is_ongoing) {
5956 btrfs_dev_replace_clear_lock_blocking(dev_replace);
5957 btrfs_dev_replace_read_unlock(dev_replace);
5959 free_extent_map(em);
5963 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
5964 u64 logical, u64 *length,
5965 struct btrfs_bio **bbio_ret, int mirror_num)
5967 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret,
5971 /* For Scrub/replace */
5972 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
5973 u64 logical, u64 *length,
5974 struct btrfs_bio **bbio_ret)
5976 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1);
5979 int btrfs_rmap_block(struct btrfs_fs_info *fs_info, u64 chunk_start,
5980 u64 physical, u64 **logical, int *naddrs, int *stripe_len)
5982 struct extent_map *em;
5983 struct map_lookup *map;
5991 em = get_chunk_map(fs_info, chunk_start, 1);
5995 map = em->map_lookup;
5997 rmap_len = map->stripe_len;
5999 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
6000 length = div_u64(length, map->num_stripes / map->sub_stripes);
6001 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
6002 length = div_u64(length, map->num_stripes);
6003 else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6004 length = div_u64(length, nr_data_stripes(map));
6005 rmap_len = map->stripe_len * nr_data_stripes(map);
6008 buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
6009 BUG_ON(!buf); /* -ENOMEM */
6011 for (i = 0; i < map->num_stripes; i++) {
6012 if (map->stripes[i].physical > physical ||
6013 map->stripes[i].physical + length <= physical)
6016 stripe_nr = physical - map->stripes[i].physical;
6017 stripe_nr = div64_u64(stripe_nr, map->stripe_len);
6019 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
6020 stripe_nr = stripe_nr * map->num_stripes + i;
6021 stripe_nr = div_u64(stripe_nr, map->sub_stripes);
6022 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
6023 stripe_nr = stripe_nr * map->num_stripes + i;
6024 } /* else if RAID[56], multiply by nr_data_stripes().
6025 * Alternatively, just use rmap_len below instead of
6026 * map->stripe_len */
6028 bytenr = chunk_start + stripe_nr * rmap_len;
6029 WARN_ON(nr >= map->num_stripes);
6030 for (j = 0; j < nr; j++) {
6031 if (buf[j] == bytenr)
6035 WARN_ON(nr >= map->num_stripes);
6042 *stripe_len = rmap_len;
6044 free_extent_map(em);
6048 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
6050 bio->bi_private = bbio->private;
6051 bio->bi_end_io = bbio->end_io;
6054 btrfs_put_bbio(bbio);
6057 static void btrfs_end_bio(struct bio *bio)
6059 struct btrfs_bio *bbio = bio->bi_private;
6060 int is_orig_bio = 0;
6062 if (bio->bi_status) {
6063 atomic_inc(&bbio->error);
6064 if (bio->bi_status == BLK_STS_IOERR ||
6065 bio->bi_status == BLK_STS_TARGET) {
6066 unsigned int stripe_index =
6067 btrfs_io_bio(bio)->stripe_index;
6068 struct btrfs_device *dev;
6070 BUG_ON(stripe_index >= bbio->num_stripes);
6071 dev = bbio->stripes[stripe_index].dev;
6073 if (bio_op(bio) == REQ_OP_WRITE)
6074 btrfs_dev_stat_inc_and_print(dev,
6075 BTRFS_DEV_STAT_WRITE_ERRS);
6077 btrfs_dev_stat_inc_and_print(dev,
6078 BTRFS_DEV_STAT_READ_ERRS);
6079 if (bio->bi_opf & REQ_PREFLUSH)
6080 btrfs_dev_stat_inc_and_print(dev,
6081 BTRFS_DEV_STAT_FLUSH_ERRS);
6086 if (bio == bbio->orig_bio)
6089 btrfs_bio_counter_dec(bbio->fs_info);
6091 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6094 bio = bbio->orig_bio;
6097 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6098 /* only send an error to the higher layers if it is
6099 * beyond the tolerance of the btrfs bio
6101 if (atomic_read(&bbio->error) > bbio->max_errors) {
6102 bio->bi_status = BLK_STS_IOERR;
6105 * this bio is actually up to date, we didn't
6106 * go over the max number of errors
6108 bio->bi_status = BLK_STS_OK;
6111 btrfs_end_bbio(bbio, bio);
6112 } else if (!is_orig_bio) {
6118 * see run_scheduled_bios for a description of why bios are collected for
6121 * This will add one bio to the pending list for a device and make sure
6122 * the work struct is scheduled.
6124 static noinline void btrfs_schedule_bio(struct btrfs_device *device,
6127 struct btrfs_fs_info *fs_info = device->fs_info;
6128 int should_queue = 1;
6129 struct btrfs_pending_bios *pending_bios;
6131 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state) ||
6137 /* don't bother with additional async steps for reads, right now */
6138 if (bio_op(bio) == REQ_OP_READ) {
6139 btrfsic_submit_bio(bio);
6143 WARN_ON(bio->bi_next);
6144 bio->bi_next = NULL;
6146 spin_lock(&device->io_lock);
6147 if (op_is_sync(bio->bi_opf))
6148 pending_bios = &device->pending_sync_bios;
6150 pending_bios = &device->pending_bios;
6152 if (pending_bios->tail)
6153 pending_bios->tail->bi_next = bio;
6155 pending_bios->tail = bio;
6156 if (!pending_bios->head)
6157 pending_bios->head = bio;
6158 if (device->running_pending)
6161 spin_unlock(&device->io_lock);
6164 btrfs_queue_work(fs_info->submit_workers, &device->work);
6167 static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio,
6168 u64 physical, int dev_nr, int async)
6170 struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
6171 struct btrfs_fs_info *fs_info = bbio->fs_info;
6173 bio->bi_private = bbio;
6174 btrfs_io_bio(bio)->stripe_index = dev_nr;
6175 bio->bi_end_io = btrfs_end_bio;
6176 bio->bi_iter.bi_sector = physical >> 9;
6179 struct rcu_string *name;
6182 name = rcu_dereference(dev->name);
6183 btrfs_debug(fs_info,
6184 "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
6185 bio_op(bio), bio->bi_opf,
6186 (u64)bio->bi_iter.bi_sector,
6187 (u_long)dev->bdev->bd_dev, name->str, dev->devid,
6188 bio->bi_iter.bi_size);
6192 bio_set_dev(bio, dev->bdev);
6194 btrfs_bio_counter_inc_noblocked(fs_info);
6197 btrfs_schedule_bio(dev, bio);
6199 btrfsic_submit_bio(bio);
6202 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6204 atomic_inc(&bbio->error);
6205 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6206 /* Should be the original bio. */
6207 WARN_ON(bio != bbio->orig_bio);
6209 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6210 bio->bi_iter.bi_sector = logical >> 9;
6211 if (atomic_read(&bbio->error) > bbio->max_errors)
6212 bio->bi_status = BLK_STS_IOERR;
6214 bio->bi_status = BLK_STS_OK;
6215 btrfs_end_bbio(bbio, bio);
6219 blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
6220 int mirror_num, int async_submit)
6222 struct btrfs_device *dev;
6223 struct bio *first_bio = bio;
6224 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
6230 struct btrfs_bio *bbio = NULL;
6232 length = bio->bi_iter.bi_size;
6233 map_length = length;
6235 btrfs_bio_counter_inc_blocked(fs_info);
6236 ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical,
6237 &map_length, &bbio, mirror_num, 1);
6239 btrfs_bio_counter_dec(fs_info);
6240 return errno_to_blk_status(ret);
6243 total_devs = bbio->num_stripes;
6244 bbio->orig_bio = first_bio;
6245 bbio->private = first_bio->bi_private;
6246 bbio->end_io = first_bio->bi_end_io;
6247 bbio->fs_info = fs_info;
6248 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6250 if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6251 ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) {
6252 /* In this case, map_length has been set to the length of
6253 a single stripe; not the whole write */
6254 if (bio_op(bio) == REQ_OP_WRITE) {
6255 ret = raid56_parity_write(fs_info, bio, bbio,
6258 ret = raid56_parity_recover(fs_info, bio, bbio,
6259 map_length, mirror_num, 1);
6262 btrfs_bio_counter_dec(fs_info);
6263 return errno_to_blk_status(ret);
6266 if (map_length < length) {
6268 "mapping failed logical %llu bio len %llu len %llu",
6269 logical, length, map_length);
6273 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6274 dev = bbio->stripes[dev_nr].dev;
6275 if (!dev || !dev->bdev ||
6276 (bio_op(first_bio) == REQ_OP_WRITE &&
6277 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) {
6278 bbio_error(bbio, first_bio, logical);
6282 if (dev_nr < total_devs - 1)
6283 bio = btrfs_bio_clone(first_bio);
6287 submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical,
6288 dev_nr, async_submit);
6290 btrfs_bio_counter_dec(fs_info);
6294 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
6297 struct btrfs_device *device;
6298 struct btrfs_fs_devices *cur_devices;
6300 cur_devices = fs_info->fs_devices;
6301 while (cur_devices) {
6303 !memcmp(cur_devices->fsid, fsid, BTRFS_FSID_SIZE)) {
6304 device = find_device(cur_devices, devid, uuid);
6308 cur_devices = cur_devices->seed;
6313 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices,
6314 u64 devid, u8 *dev_uuid)
6316 struct btrfs_device *device;
6318 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6322 list_add(&device->dev_list, &fs_devices->devices);
6323 device->fs_devices = fs_devices;
6324 fs_devices->num_devices++;
6326 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6327 fs_devices->missing_devices++;
6333 * btrfs_alloc_device - allocate struct btrfs_device
6334 * @fs_info: used only for generating a new devid, can be NULL if
6335 * devid is provided (i.e. @devid != NULL).
6336 * @devid: a pointer to devid for this device. If NULL a new devid
6338 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6341 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6342 * on error. Returned struct is not linked onto any lists and must be
6343 * destroyed with btrfs_free_device.
6345 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6349 struct btrfs_device *dev;
6352 if (WARN_ON(!devid && !fs_info))
6353 return ERR_PTR(-EINVAL);
6355 dev = __alloc_device();
6364 ret = find_next_devid(fs_info, &tmp);
6366 btrfs_free_device(dev);
6367 return ERR_PTR(ret);
6373 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6375 generate_random_uuid(dev->uuid);
6377 btrfs_init_work(&dev->work, btrfs_submit_helper,
6378 pending_bios_fn, NULL, NULL);
6383 /* Return -EIO if any error, otherwise return 0. */
6384 static int btrfs_check_chunk_valid(struct btrfs_fs_info *fs_info,
6385 struct extent_buffer *leaf,
6386 struct btrfs_chunk *chunk, u64 logical)
6394 length = btrfs_chunk_length(leaf, chunk);
6395 stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6396 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6397 sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6398 type = btrfs_chunk_type(leaf, chunk);
6401 btrfs_err(fs_info, "invalid chunk num_stripes: %u",
6405 if (!IS_ALIGNED(logical, fs_info->sectorsize)) {
6406 btrfs_err(fs_info, "invalid chunk logical %llu", logical);
6409 if (btrfs_chunk_sector_size(leaf, chunk) != fs_info->sectorsize) {
6410 btrfs_err(fs_info, "invalid chunk sectorsize %u",
6411 btrfs_chunk_sector_size(leaf, chunk));
6414 if (!length || !IS_ALIGNED(length, fs_info->sectorsize)) {
6415 btrfs_err(fs_info, "invalid chunk length %llu", length);
6418 if (!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN) {
6419 btrfs_err(fs_info, "invalid chunk stripe length: %llu",
6423 if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6425 btrfs_err(fs_info, "unrecognized chunk type: %llu",
6426 ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
6427 BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6428 btrfs_chunk_type(leaf, chunk));
6431 if ((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes != 2) ||
6432 (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) ||
6433 (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) ||
6434 (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) ||
6435 (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) ||
6436 ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
6437 num_stripes != 1)) {
6439 "invalid num_stripes:sub_stripes %u:%u for profile %llu",
6440 num_stripes, sub_stripes,
6441 type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
6448 static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info,
6449 u64 devid, u8 *uuid, bool error)
6452 btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing",
6455 btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing",
6459 static int read_one_chunk(struct btrfs_fs_info *fs_info, struct btrfs_key *key,
6460 struct extent_buffer *leaf,
6461 struct btrfs_chunk *chunk)
6463 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
6464 struct map_lookup *map;
6465 struct extent_map *em;
6469 u8 uuid[BTRFS_UUID_SIZE];
6474 logical = key->offset;
6475 length = btrfs_chunk_length(leaf, chunk);
6476 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6478 ret = btrfs_check_chunk_valid(fs_info, leaf, chunk, logical);
6482 read_lock(&map_tree->map_tree.lock);
6483 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
6484 read_unlock(&map_tree->map_tree.lock);
6486 /* already mapped? */
6487 if (em && em->start <= logical && em->start + em->len > logical) {
6488 free_extent_map(em);
6491 free_extent_map(em);
6494 em = alloc_extent_map();
6497 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6499 free_extent_map(em);
6503 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6504 em->map_lookup = map;
6505 em->start = logical;
6508 em->block_start = 0;
6509 em->block_len = em->len;
6511 map->num_stripes = num_stripes;
6512 map->io_width = btrfs_chunk_io_width(leaf, chunk);
6513 map->io_align = btrfs_chunk_io_align(leaf, chunk);
6514 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6515 map->type = btrfs_chunk_type(leaf, chunk);
6516 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6517 for (i = 0; i < num_stripes; i++) {
6518 map->stripes[i].physical =
6519 btrfs_stripe_offset_nr(leaf, chunk, i);
6520 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6521 read_extent_buffer(leaf, uuid, (unsigned long)
6522 btrfs_stripe_dev_uuid_nr(chunk, i),
6524 map->stripes[i].dev = btrfs_find_device(fs_info, devid,
6526 if (!map->stripes[i].dev &&
6527 !btrfs_test_opt(fs_info, DEGRADED)) {
6528 free_extent_map(em);
6529 btrfs_report_missing_device(fs_info, devid, uuid, true);
6532 if (!map->stripes[i].dev) {
6533 map->stripes[i].dev =
6534 add_missing_dev(fs_info->fs_devices, devid,
6536 if (IS_ERR(map->stripes[i].dev)) {
6537 free_extent_map(em);
6539 "failed to init missing dev %llu: %ld",
6540 devid, PTR_ERR(map->stripes[i].dev));
6541 return PTR_ERR(map->stripes[i].dev);
6543 btrfs_report_missing_device(fs_info, devid, uuid, false);
6545 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
6546 &(map->stripes[i].dev->dev_state));
6550 write_lock(&map_tree->map_tree.lock);
6551 ret = add_extent_mapping(&map_tree->map_tree, em, 0);
6552 write_unlock(&map_tree->map_tree.lock);
6553 BUG_ON(ret); /* Tree corruption */
6554 free_extent_map(em);
6559 static void fill_device_from_item(struct extent_buffer *leaf,
6560 struct btrfs_dev_item *dev_item,
6561 struct btrfs_device *device)
6565 device->devid = btrfs_device_id(leaf, dev_item);
6566 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6567 device->total_bytes = device->disk_total_bytes;
6568 device->commit_total_bytes = device->disk_total_bytes;
6569 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6570 device->commit_bytes_used = device->bytes_used;
6571 device->type = btrfs_device_type(leaf, dev_item);
6572 device->io_align = btrfs_device_io_align(leaf, dev_item);
6573 device->io_width = btrfs_device_io_width(leaf, dev_item);
6574 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6575 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6576 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
6578 ptr = btrfs_device_uuid(dev_item);
6579 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6582 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info,
6585 struct btrfs_fs_devices *fs_devices;
6588 lockdep_assert_held(&uuid_mutex);
6591 fs_devices = fs_info->fs_devices->seed;
6592 while (fs_devices) {
6593 if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE))
6596 fs_devices = fs_devices->seed;
6599 fs_devices = find_fsid(fsid);
6601 if (!btrfs_test_opt(fs_info, DEGRADED))
6602 return ERR_PTR(-ENOENT);
6604 fs_devices = alloc_fs_devices(fsid);
6605 if (IS_ERR(fs_devices))
6608 fs_devices->seeding = 1;
6609 fs_devices->opened = 1;
6613 fs_devices = clone_fs_devices(fs_devices);
6614 if (IS_ERR(fs_devices))
6617 ret = open_fs_devices(fs_devices, FMODE_READ, fs_info->bdev_holder);
6619 free_fs_devices(fs_devices);
6620 fs_devices = ERR_PTR(ret);
6624 if (!fs_devices->seeding) {
6625 close_fs_devices(fs_devices);
6626 free_fs_devices(fs_devices);
6627 fs_devices = ERR_PTR(-EINVAL);
6631 fs_devices->seed = fs_info->fs_devices->seed;
6632 fs_info->fs_devices->seed = fs_devices;
6637 static int read_one_dev(struct btrfs_fs_info *fs_info,
6638 struct extent_buffer *leaf,
6639 struct btrfs_dev_item *dev_item)
6641 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6642 struct btrfs_device *device;
6645 u8 fs_uuid[BTRFS_FSID_SIZE];
6646 u8 dev_uuid[BTRFS_UUID_SIZE];
6648 devid = btrfs_device_id(leaf, dev_item);
6649 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6651 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6654 if (memcmp(fs_uuid, fs_info->fsid, BTRFS_FSID_SIZE)) {
6655 fs_devices = open_seed_devices(fs_info, fs_uuid);
6656 if (IS_ERR(fs_devices))
6657 return PTR_ERR(fs_devices);
6660 device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
6662 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6663 btrfs_report_missing_device(fs_info, devid,
6668 device = add_missing_dev(fs_devices, devid, dev_uuid);
6669 if (IS_ERR(device)) {
6671 "failed to add missing dev %llu: %ld",
6672 devid, PTR_ERR(device));
6673 return PTR_ERR(device);
6675 btrfs_report_missing_device(fs_info, devid, dev_uuid, false);
6677 if (!device->bdev) {
6678 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6679 btrfs_report_missing_device(fs_info,
6680 devid, dev_uuid, true);
6683 btrfs_report_missing_device(fs_info, devid,
6687 if (!device->bdev &&
6688 !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
6690 * this happens when a device that was properly setup
6691 * in the device info lists suddenly goes bad.
6692 * device->bdev is NULL, and so we have to set
6693 * device->missing to one here
6695 device->fs_devices->missing_devices++;
6696 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6699 /* Move the device to its own fs_devices */
6700 if (device->fs_devices != fs_devices) {
6701 ASSERT(test_bit(BTRFS_DEV_STATE_MISSING,
6702 &device->dev_state));
6704 list_move(&device->dev_list, &fs_devices->devices);
6705 device->fs_devices->num_devices--;
6706 fs_devices->num_devices++;
6708 device->fs_devices->missing_devices--;
6709 fs_devices->missing_devices++;
6711 device->fs_devices = fs_devices;
6715 if (device->fs_devices != fs_info->fs_devices) {
6716 BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state));
6717 if (device->generation !=
6718 btrfs_device_generation(leaf, dev_item))
6722 fill_device_from_item(leaf, dev_item, device);
6723 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
6724 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
6725 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
6726 device->fs_devices->total_rw_bytes += device->total_bytes;
6727 atomic64_add(device->total_bytes - device->bytes_used,
6728 &fs_info->free_chunk_space);
6734 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
6736 struct btrfs_root *root = fs_info->tree_root;
6737 struct btrfs_super_block *super_copy = fs_info->super_copy;
6738 struct extent_buffer *sb;
6739 struct btrfs_disk_key *disk_key;
6740 struct btrfs_chunk *chunk;
6742 unsigned long sb_array_offset;
6749 struct btrfs_key key;
6751 ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize);
6753 * This will create extent buffer of nodesize, superblock size is
6754 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6755 * overallocate but we can keep it as-is, only the first page is used.
6757 sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET);
6760 set_extent_buffer_uptodate(sb);
6761 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6763 * The sb extent buffer is artificial and just used to read the system array.
6764 * set_extent_buffer_uptodate() call does not properly mark all it's
6765 * pages up-to-date when the page is larger: extent does not cover the
6766 * whole page and consequently check_page_uptodate does not find all
6767 * the page's extents up-to-date (the hole beyond sb),
6768 * write_extent_buffer then triggers a WARN_ON.
6770 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6771 * but sb spans only this function. Add an explicit SetPageUptodate call
6772 * to silence the warning eg. on PowerPC 64.
6774 if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
6775 SetPageUptodate(sb->pages[0]);
6777 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6778 array_size = btrfs_super_sys_array_size(super_copy);
6780 array_ptr = super_copy->sys_chunk_array;
6781 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6784 while (cur_offset < array_size) {
6785 disk_key = (struct btrfs_disk_key *)array_ptr;
6786 len = sizeof(*disk_key);
6787 if (cur_offset + len > array_size)
6788 goto out_short_read;
6790 btrfs_disk_key_to_cpu(&key, disk_key);
6793 sb_array_offset += len;
6796 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6797 chunk = (struct btrfs_chunk *)sb_array_offset;
6799 * At least one btrfs_chunk with one stripe must be
6800 * present, exact stripe count check comes afterwards
6802 len = btrfs_chunk_item_size(1);
6803 if (cur_offset + len > array_size)
6804 goto out_short_read;
6806 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6809 "invalid number of stripes %u in sys_array at offset %u",
6810 num_stripes, cur_offset);
6815 type = btrfs_chunk_type(sb, chunk);
6816 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
6818 "invalid chunk type %llu in sys_array at offset %u",
6824 len = btrfs_chunk_item_size(num_stripes);
6825 if (cur_offset + len > array_size)
6826 goto out_short_read;
6828 ret = read_one_chunk(fs_info, &key, sb, chunk);
6833 "unexpected item type %u in sys_array at offset %u",
6834 (u32)key.type, cur_offset);
6839 sb_array_offset += len;
6842 clear_extent_buffer_uptodate(sb);
6843 free_extent_buffer_stale(sb);
6847 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
6849 clear_extent_buffer_uptodate(sb);
6850 free_extent_buffer_stale(sb);
6855 * Check if all chunks in the fs are OK for read-write degraded mount
6857 * If the @failing_dev is specified, it's accounted as missing.
6859 * Return true if all chunks meet the minimal RW mount requirements.
6860 * Return false if any chunk doesn't meet the minimal RW mount requirements.
6862 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
6863 struct btrfs_device *failing_dev)
6865 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
6866 struct extent_map *em;
6870 read_lock(&map_tree->map_tree.lock);
6871 em = lookup_extent_mapping(&map_tree->map_tree, 0, (u64)-1);
6872 read_unlock(&map_tree->map_tree.lock);
6873 /* No chunk at all? Return false anyway */
6879 struct map_lookup *map;
6884 map = em->map_lookup;
6886 btrfs_get_num_tolerated_disk_barrier_failures(
6888 for (i = 0; i < map->num_stripes; i++) {
6889 struct btrfs_device *dev = map->stripes[i].dev;
6891 if (!dev || !dev->bdev ||
6892 test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) ||
6893 dev->last_flush_error)
6895 else if (failing_dev && failing_dev == dev)
6898 if (missing > max_tolerated) {
6901 "chunk %llu missing %d devices, max tolerance is %d for writeable mount",
6902 em->start, missing, max_tolerated);
6903 free_extent_map(em);
6907 next_start = extent_map_end(em);
6908 free_extent_map(em);
6910 read_lock(&map_tree->map_tree.lock);
6911 em = lookup_extent_mapping(&map_tree->map_tree, next_start,
6912 (u64)(-1) - next_start);
6913 read_unlock(&map_tree->map_tree.lock);
6919 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
6921 struct btrfs_root *root = fs_info->chunk_root;
6922 struct btrfs_path *path;
6923 struct extent_buffer *leaf;
6924 struct btrfs_key key;
6925 struct btrfs_key found_key;
6930 path = btrfs_alloc_path();
6935 * uuid_mutex is needed only if we are mounting a sprout FS
6936 * otherwise we don't need it.
6938 mutex_lock(&uuid_mutex);
6939 mutex_lock(&fs_info->chunk_mutex);
6942 * Read all device items, and then all the chunk items. All
6943 * device items are found before any chunk item (their object id
6944 * is smaller than the lowest possible object id for a chunk
6945 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6947 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6950 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6954 leaf = path->nodes[0];
6955 slot = path->slots[0];
6956 if (slot >= btrfs_header_nritems(leaf)) {
6957 ret = btrfs_next_leaf(root, path);
6964 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6965 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6966 struct btrfs_dev_item *dev_item;
6967 dev_item = btrfs_item_ptr(leaf, slot,
6968 struct btrfs_dev_item);
6969 ret = read_one_dev(fs_info, leaf, dev_item);
6973 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6974 struct btrfs_chunk *chunk;
6975 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6976 ret = read_one_chunk(fs_info, &found_key, leaf, chunk);
6984 * After loading chunk tree, we've got all device information,
6985 * do another round of validation checks.
6987 if (total_dev != fs_info->fs_devices->total_devices) {
6989 "super_num_devices %llu mismatch with num_devices %llu found here",
6990 btrfs_super_num_devices(fs_info->super_copy),
6995 if (btrfs_super_total_bytes(fs_info->super_copy) <
6996 fs_info->fs_devices->total_rw_bytes) {
6998 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
6999 btrfs_super_total_bytes(fs_info->super_copy),
7000 fs_info->fs_devices->total_rw_bytes);
7006 mutex_unlock(&fs_info->chunk_mutex);
7007 mutex_unlock(&uuid_mutex);
7009 btrfs_free_path(path);
7013 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
7015 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7016 struct btrfs_device *device;
7018 while (fs_devices) {
7019 mutex_lock(&fs_devices->device_list_mutex);
7020 list_for_each_entry(device, &fs_devices->devices, dev_list)
7021 device->fs_info = fs_info;
7022 mutex_unlock(&fs_devices->device_list_mutex);
7024 fs_devices = fs_devices->seed;
7028 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
7032 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7033 btrfs_dev_stat_reset(dev, i);
7036 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
7038 struct btrfs_key key;
7039 struct btrfs_key found_key;
7040 struct btrfs_root *dev_root = fs_info->dev_root;
7041 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7042 struct extent_buffer *eb;
7045 struct btrfs_device *device;
7046 struct btrfs_path *path = NULL;
7049 path = btrfs_alloc_path();
7055 mutex_lock(&fs_devices->device_list_mutex);
7056 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7058 struct btrfs_dev_stats_item *ptr;
7060 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7061 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7062 key.offset = device->devid;
7063 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
7065 __btrfs_reset_dev_stats(device);
7066 device->dev_stats_valid = 1;
7067 btrfs_release_path(path);
7070 slot = path->slots[0];
7071 eb = path->nodes[0];
7072 btrfs_item_key_to_cpu(eb, &found_key, slot);
7073 item_size = btrfs_item_size_nr(eb, slot);
7075 ptr = btrfs_item_ptr(eb, slot,
7076 struct btrfs_dev_stats_item);
7078 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7079 if (item_size >= (1 + i) * sizeof(__le64))
7080 btrfs_dev_stat_set(device, i,
7081 btrfs_dev_stats_value(eb, ptr, i));
7083 btrfs_dev_stat_reset(device, i);
7086 device->dev_stats_valid = 1;
7087 btrfs_dev_stat_print_on_load(device);
7088 btrfs_release_path(path);
7090 mutex_unlock(&fs_devices->device_list_mutex);
7093 btrfs_free_path(path);
7094 return ret < 0 ? ret : 0;
7097 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
7098 struct btrfs_fs_info *fs_info,
7099 struct btrfs_device *device)
7101 struct btrfs_root *dev_root = fs_info->dev_root;
7102 struct btrfs_path *path;
7103 struct btrfs_key key;
7104 struct extent_buffer *eb;
7105 struct btrfs_dev_stats_item *ptr;
7109 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7110 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7111 key.offset = device->devid;
7113 path = btrfs_alloc_path();
7116 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
7118 btrfs_warn_in_rcu(fs_info,
7119 "error %d while searching for dev_stats item for device %s",
7120 ret, rcu_str_deref(device->name));
7125 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
7126 /* need to delete old one and insert a new one */
7127 ret = btrfs_del_item(trans, dev_root, path);
7129 btrfs_warn_in_rcu(fs_info,
7130 "delete too small dev_stats item for device %s failed %d",
7131 rcu_str_deref(device->name), ret);
7138 /* need to insert a new item */
7139 btrfs_release_path(path);
7140 ret = btrfs_insert_empty_item(trans, dev_root, path,
7141 &key, sizeof(*ptr));
7143 btrfs_warn_in_rcu(fs_info,
7144 "insert dev_stats item for device %s failed %d",
7145 rcu_str_deref(device->name), ret);
7150 eb = path->nodes[0];
7151 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7152 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7153 btrfs_set_dev_stats_value(eb, ptr, i,
7154 btrfs_dev_stat_read(device, i));
7155 btrfs_mark_buffer_dirty(eb);
7158 btrfs_free_path(path);
7163 * called from commit_transaction. Writes all changed device stats to disk.
7165 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
7166 struct btrfs_fs_info *fs_info)
7168 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7169 struct btrfs_device *device;
7173 mutex_lock(&fs_devices->device_list_mutex);
7174 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7175 stats_cnt = atomic_read(&device->dev_stats_ccnt);
7176 if (!device->dev_stats_valid || stats_cnt == 0)
7181 * There is a LOAD-LOAD control dependency between the value of
7182 * dev_stats_ccnt and updating the on-disk values which requires
7183 * reading the in-memory counters. Such control dependencies
7184 * require explicit read memory barriers.
7186 * This memory barriers pairs with smp_mb__before_atomic in
7187 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full
7188 * barrier implied by atomic_xchg in
7189 * btrfs_dev_stats_read_and_reset
7193 ret = update_dev_stat_item(trans, fs_info, device);
7195 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7197 mutex_unlock(&fs_devices->device_list_mutex);
7202 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7204 btrfs_dev_stat_inc(dev, index);
7205 btrfs_dev_stat_print_on_error(dev);
7208 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
7210 if (!dev->dev_stats_valid)
7212 btrfs_err_rl_in_rcu(dev->fs_info,
7213 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7214 rcu_str_deref(dev->name),
7215 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7216 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7217 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7218 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7219 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7222 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7226 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7227 if (btrfs_dev_stat_read(dev, i) != 0)
7229 if (i == BTRFS_DEV_STAT_VALUES_MAX)
7230 return; /* all values == 0, suppress message */
7232 btrfs_info_in_rcu(dev->fs_info,
7233 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7234 rcu_str_deref(dev->name),
7235 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7236 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7237 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7238 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7239 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7242 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
7243 struct btrfs_ioctl_get_dev_stats *stats)
7245 struct btrfs_device *dev;
7246 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7249 mutex_lock(&fs_devices->device_list_mutex);
7250 dev = btrfs_find_device(fs_info, stats->devid, NULL, NULL);
7251 mutex_unlock(&fs_devices->device_list_mutex);
7254 btrfs_warn(fs_info, "get dev_stats failed, device not found");
7256 } else if (!dev->dev_stats_valid) {
7257 btrfs_warn(fs_info, "get dev_stats failed, not yet valid");
7259 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7260 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7261 if (stats->nr_items > i)
7263 btrfs_dev_stat_read_and_reset(dev, i);
7265 btrfs_dev_stat_reset(dev, i);
7268 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7269 if (stats->nr_items > i)
7270 stats->values[i] = btrfs_dev_stat_read(dev, i);
7272 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7273 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7277 void btrfs_scratch_superblocks(struct block_device *bdev, const char *device_path)
7279 struct buffer_head *bh;
7280 struct btrfs_super_block *disk_super;
7286 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX;
7289 if (btrfs_read_dev_one_super(bdev, copy_num, &bh))
7292 disk_super = (struct btrfs_super_block *)bh->b_data;
7294 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
7295 set_buffer_dirty(bh);
7296 sync_dirty_buffer(bh);
7300 /* Notify udev that device has changed */
7301 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
7303 /* Update ctime/mtime for device path for libblkid */
7304 update_dev_time(device_path);
7308 * Update the size of all devices, which is used for writing out the
7311 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
7313 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7314 struct btrfs_device *curr, *next;
7316 if (list_empty(&fs_devices->resized_devices))
7319 mutex_lock(&fs_devices->device_list_mutex);
7320 mutex_lock(&fs_info->chunk_mutex);
7321 list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
7323 list_del_init(&curr->resized_list);
7324 curr->commit_total_bytes = curr->disk_total_bytes;
7326 mutex_unlock(&fs_info->chunk_mutex);
7327 mutex_unlock(&fs_devices->device_list_mutex);
7330 /* Must be invoked during the transaction commit */
7331 void btrfs_update_commit_device_bytes_used(struct btrfs_transaction *trans)
7333 struct btrfs_fs_info *fs_info = trans->fs_info;
7334 struct extent_map *em;
7335 struct map_lookup *map;
7336 struct btrfs_device *dev;
7339 if (list_empty(&trans->pending_chunks))
7342 /* In order to kick the device replace finish process */
7343 mutex_lock(&fs_info->chunk_mutex);
7344 list_for_each_entry(em, &trans->pending_chunks, list) {
7345 map = em->map_lookup;
7347 for (i = 0; i < map->num_stripes; i++) {
7348 dev = map->stripes[i].dev;
7349 dev->commit_bytes_used = dev->bytes_used;
7352 mutex_unlock(&fs_info->chunk_mutex);
7355 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
7357 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7358 while (fs_devices) {
7359 fs_devices->fs_info = fs_info;
7360 fs_devices = fs_devices->seed;
7364 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
7366 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7367 while (fs_devices) {
7368 fs_devices->fs_info = NULL;
7369 fs_devices = fs_devices->seed;