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(&uuid_mutex);
1150 mutex_lock(&fs_devices->device_list_mutex);
1151 if (fs_devices->opened) {
1152 fs_devices->opened++;
1155 list_sort(NULL, &fs_devices->devices, devid_cmp);
1156 ret = open_fs_devices(fs_devices, flags, holder);
1158 mutex_unlock(&fs_devices->device_list_mutex);
1159 mutex_unlock(&uuid_mutex);
1164 static void btrfs_release_disk_super(struct page *page)
1170 static int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr,
1172 struct btrfs_super_block **disk_super)
1177 /* make sure our super fits in the device */
1178 if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1181 /* make sure our super fits in the page */
1182 if (sizeof(**disk_super) > PAGE_SIZE)
1185 /* make sure our super doesn't straddle pages on disk */
1186 index = bytenr >> PAGE_SHIFT;
1187 if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index)
1190 /* pull in the page with our super */
1191 *page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
1194 if (IS_ERR_OR_NULL(*page))
1199 /* align our pointer to the offset of the super block */
1200 *disk_super = p + (bytenr & ~PAGE_MASK);
1202 if (btrfs_super_bytenr(*disk_super) != bytenr ||
1203 btrfs_super_magic(*disk_super) != BTRFS_MAGIC) {
1204 btrfs_release_disk_super(*page);
1208 if ((*disk_super)->label[0] &&
1209 (*disk_super)->label[BTRFS_LABEL_SIZE - 1])
1210 (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0';
1216 * Look for a btrfs signature on a device. This may be called out of the mount path
1217 * and we are not allowed to call set_blocksize during the scan. The superblock
1218 * is read via pagecache
1220 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
1221 struct btrfs_fs_devices **fs_devices_ret)
1223 struct btrfs_super_block *disk_super;
1224 struct btrfs_device *device;
1225 struct block_device *bdev;
1231 * we would like to check all the supers, but that would make
1232 * a btrfs mount succeed after a mkfs from a different FS.
1233 * So, we need to add a special mount option to scan for
1234 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1236 bytenr = btrfs_sb_offset(0);
1237 flags |= FMODE_EXCL;
1239 bdev = blkdev_get_by_path(path, flags, holder);
1241 return PTR_ERR(bdev);
1243 if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super)) {
1245 goto error_bdev_put;
1248 mutex_lock(&uuid_mutex);
1249 device = device_list_add(path, disk_super);
1251 ret = PTR_ERR(device);
1253 *fs_devices_ret = device->fs_devices;
1254 mutex_unlock(&uuid_mutex);
1256 btrfs_release_disk_super(page);
1259 blkdev_put(bdev, flags);
1264 /* helper to account the used device space in the range */
1265 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
1266 u64 end, u64 *length)
1268 struct btrfs_key key;
1269 struct btrfs_root *root = device->fs_info->dev_root;
1270 struct btrfs_dev_extent *dev_extent;
1271 struct btrfs_path *path;
1275 struct extent_buffer *l;
1279 if (start >= device->total_bytes ||
1280 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
1283 path = btrfs_alloc_path();
1286 path->reada = READA_FORWARD;
1288 key.objectid = device->devid;
1290 key.type = BTRFS_DEV_EXTENT_KEY;
1292 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1296 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1303 slot = path->slots[0];
1304 if (slot >= btrfs_header_nritems(l)) {
1305 ret = btrfs_next_leaf(root, path);
1313 btrfs_item_key_to_cpu(l, &key, slot);
1315 if (key.objectid < device->devid)
1318 if (key.objectid > device->devid)
1321 if (key.type != BTRFS_DEV_EXTENT_KEY)
1324 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1325 extent_end = key.offset + btrfs_dev_extent_length(l,
1327 if (key.offset <= start && extent_end > end) {
1328 *length = end - start + 1;
1330 } else if (key.offset <= start && extent_end > start)
1331 *length += extent_end - start;
1332 else if (key.offset > start && extent_end <= end)
1333 *length += extent_end - key.offset;
1334 else if (key.offset > start && key.offset <= end) {
1335 *length += end - key.offset + 1;
1337 } else if (key.offset > end)
1345 btrfs_free_path(path);
1349 static int contains_pending_extent(struct btrfs_transaction *transaction,
1350 struct btrfs_device *device,
1351 u64 *start, u64 len)
1353 struct btrfs_fs_info *fs_info = device->fs_info;
1354 struct extent_map *em;
1355 struct list_head *search_list = &fs_info->pinned_chunks;
1357 u64 physical_start = *start;
1360 search_list = &transaction->pending_chunks;
1362 list_for_each_entry(em, search_list, list) {
1363 struct map_lookup *map;
1366 map = em->map_lookup;
1367 for (i = 0; i < map->num_stripes; i++) {
1370 if (map->stripes[i].dev != device)
1372 if (map->stripes[i].physical >= physical_start + len ||
1373 map->stripes[i].physical + em->orig_block_len <=
1377 * Make sure that while processing the pinned list we do
1378 * not override our *start with a lower value, because
1379 * we can have pinned chunks that fall within this
1380 * device hole and that have lower physical addresses
1381 * than the pending chunks we processed before. If we
1382 * do not take this special care we can end up getting
1383 * 2 pending chunks that start at the same physical
1384 * device offsets because the end offset of a pinned
1385 * chunk can be equal to the start offset of some
1388 end = map->stripes[i].physical + em->orig_block_len;
1395 if (search_list != &fs_info->pinned_chunks) {
1396 search_list = &fs_info->pinned_chunks;
1405 * find_free_dev_extent_start - find free space in the specified device
1406 * @device: the device which we search the free space in
1407 * @num_bytes: the size of the free space that we need
1408 * @search_start: the position from which to begin the search
1409 * @start: store the start of the free space.
1410 * @len: the size of the free space. that we find, or the size
1411 * of the max free space if we don't find suitable free space
1413 * this uses a pretty simple search, the expectation is that it is
1414 * called very infrequently and that a given device has a small number
1417 * @start is used to store the start of the free space if we find. But if we
1418 * don't find suitable free space, it will be used to store the start position
1419 * of the max free space.
1421 * @len is used to store the size of the free space that we find.
1422 * But if we don't find suitable free space, it is used to store the size of
1423 * the max free space.
1425 int find_free_dev_extent_start(struct btrfs_transaction *transaction,
1426 struct btrfs_device *device, u64 num_bytes,
1427 u64 search_start, u64 *start, u64 *len)
1429 struct btrfs_fs_info *fs_info = device->fs_info;
1430 struct btrfs_root *root = fs_info->dev_root;
1431 struct btrfs_key key;
1432 struct btrfs_dev_extent *dev_extent;
1433 struct btrfs_path *path;
1438 u64 search_end = device->total_bytes;
1441 struct extent_buffer *l;
1444 * We don't want to overwrite the superblock on the drive nor any area
1445 * used by the boot loader (grub for example), so we make sure to start
1446 * at an offset of at least 1MB.
1448 search_start = max_t(u64, search_start, SZ_1M);
1450 path = btrfs_alloc_path();
1454 max_hole_start = search_start;
1458 if (search_start >= search_end ||
1459 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1464 path->reada = READA_FORWARD;
1465 path->search_commit_root = 1;
1466 path->skip_locking = 1;
1468 key.objectid = device->devid;
1469 key.offset = search_start;
1470 key.type = BTRFS_DEV_EXTENT_KEY;
1472 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1476 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1483 slot = path->slots[0];
1484 if (slot >= btrfs_header_nritems(l)) {
1485 ret = btrfs_next_leaf(root, path);
1493 btrfs_item_key_to_cpu(l, &key, slot);
1495 if (key.objectid < device->devid)
1498 if (key.objectid > device->devid)
1501 if (key.type != BTRFS_DEV_EXTENT_KEY)
1504 if (key.offset > search_start) {
1505 hole_size = key.offset - search_start;
1508 * Have to check before we set max_hole_start, otherwise
1509 * we could end up sending back this offset anyway.
1511 if (contains_pending_extent(transaction, device,
1514 if (key.offset >= search_start) {
1515 hole_size = key.offset - search_start;
1522 if (hole_size > max_hole_size) {
1523 max_hole_start = search_start;
1524 max_hole_size = hole_size;
1528 * If this free space is greater than which we need,
1529 * it must be the max free space that we have found
1530 * until now, so max_hole_start must point to the start
1531 * of this free space and the length of this free space
1532 * is stored in max_hole_size. Thus, we return
1533 * max_hole_start and max_hole_size and go back to the
1536 if (hole_size >= num_bytes) {
1542 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1543 extent_end = key.offset + btrfs_dev_extent_length(l,
1545 if (extent_end > search_start)
1546 search_start = extent_end;
1553 * At this point, search_start should be the end of
1554 * allocated dev extents, and when shrinking the device,
1555 * search_end may be smaller than search_start.
1557 if (search_end > search_start) {
1558 hole_size = search_end - search_start;
1560 if (contains_pending_extent(transaction, device, &search_start,
1562 btrfs_release_path(path);
1566 if (hole_size > max_hole_size) {
1567 max_hole_start = search_start;
1568 max_hole_size = hole_size;
1573 if (max_hole_size < num_bytes)
1579 btrfs_free_path(path);
1580 *start = max_hole_start;
1582 *len = max_hole_size;
1586 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1587 struct btrfs_device *device, u64 num_bytes,
1588 u64 *start, u64 *len)
1590 /* FIXME use last free of some kind */
1591 return find_free_dev_extent_start(trans->transaction, device,
1592 num_bytes, 0, start, len);
1595 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1596 struct btrfs_device *device,
1597 u64 start, u64 *dev_extent_len)
1599 struct btrfs_fs_info *fs_info = device->fs_info;
1600 struct btrfs_root *root = fs_info->dev_root;
1602 struct btrfs_path *path;
1603 struct btrfs_key key;
1604 struct btrfs_key found_key;
1605 struct extent_buffer *leaf = NULL;
1606 struct btrfs_dev_extent *extent = NULL;
1608 path = btrfs_alloc_path();
1612 key.objectid = device->devid;
1614 key.type = BTRFS_DEV_EXTENT_KEY;
1616 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1618 ret = btrfs_previous_item(root, path, key.objectid,
1619 BTRFS_DEV_EXTENT_KEY);
1622 leaf = path->nodes[0];
1623 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1624 extent = btrfs_item_ptr(leaf, path->slots[0],
1625 struct btrfs_dev_extent);
1626 BUG_ON(found_key.offset > start || found_key.offset +
1627 btrfs_dev_extent_length(leaf, extent) < start);
1629 btrfs_release_path(path);
1631 } else if (ret == 0) {
1632 leaf = path->nodes[0];
1633 extent = btrfs_item_ptr(leaf, path->slots[0],
1634 struct btrfs_dev_extent);
1636 btrfs_handle_fs_error(fs_info, ret, "Slot search failed");
1640 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1642 ret = btrfs_del_item(trans, root, path);
1644 btrfs_handle_fs_error(fs_info, ret,
1645 "Failed to remove dev extent item");
1647 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1650 btrfs_free_path(path);
1654 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1655 struct btrfs_device *device,
1656 u64 chunk_offset, u64 start, u64 num_bytes)
1659 struct btrfs_path *path;
1660 struct btrfs_fs_info *fs_info = device->fs_info;
1661 struct btrfs_root *root = fs_info->dev_root;
1662 struct btrfs_dev_extent *extent;
1663 struct extent_buffer *leaf;
1664 struct btrfs_key key;
1666 WARN_ON(!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state));
1667 WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state));
1668 path = btrfs_alloc_path();
1672 key.objectid = device->devid;
1674 key.type = BTRFS_DEV_EXTENT_KEY;
1675 ret = btrfs_insert_empty_item(trans, root, path, &key,
1680 leaf = path->nodes[0];
1681 extent = btrfs_item_ptr(leaf, path->slots[0],
1682 struct btrfs_dev_extent);
1683 btrfs_set_dev_extent_chunk_tree(leaf, extent,
1684 BTRFS_CHUNK_TREE_OBJECTID);
1685 btrfs_set_dev_extent_chunk_objectid(leaf, extent,
1686 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
1687 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1689 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1690 btrfs_mark_buffer_dirty(leaf);
1692 btrfs_free_path(path);
1696 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1698 struct extent_map_tree *em_tree;
1699 struct extent_map *em;
1703 em_tree = &fs_info->mapping_tree.map_tree;
1704 read_lock(&em_tree->lock);
1705 n = rb_last(&em_tree->map);
1707 em = rb_entry(n, struct extent_map, rb_node);
1708 ret = em->start + em->len;
1710 read_unlock(&em_tree->lock);
1715 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1719 struct btrfs_key key;
1720 struct btrfs_key found_key;
1721 struct btrfs_path *path;
1723 path = btrfs_alloc_path();
1727 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1728 key.type = BTRFS_DEV_ITEM_KEY;
1729 key.offset = (u64)-1;
1731 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1735 BUG_ON(ret == 0); /* Corruption */
1737 ret = btrfs_previous_item(fs_info->chunk_root, path,
1738 BTRFS_DEV_ITEMS_OBJECTID,
1739 BTRFS_DEV_ITEM_KEY);
1743 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1745 *devid_ret = found_key.offset + 1;
1749 btrfs_free_path(path);
1754 * the device information is stored in the chunk root
1755 * the btrfs_device struct should be fully filled in
1757 static int btrfs_add_dev_item(struct btrfs_trans_handle *trans,
1758 struct btrfs_fs_info *fs_info,
1759 struct btrfs_device *device)
1761 struct btrfs_root *root = fs_info->chunk_root;
1763 struct btrfs_path *path;
1764 struct btrfs_dev_item *dev_item;
1765 struct extent_buffer *leaf;
1766 struct btrfs_key key;
1769 path = btrfs_alloc_path();
1773 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1774 key.type = BTRFS_DEV_ITEM_KEY;
1775 key.offset = device->devid;
1777 ret = btrfs_insert_empty_item(trans, root, path, &key,
1782 leaf = path->nodes[0];
1783 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1785 btrfs_set_device_id(leaf, dev_item, device->devid);
1786 btrfs_set_device_generation(leaf, dev_item, 0);
1787 btrfs_set_device_type(leaf, dev_item, device->type);
1788 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1789 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1790 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1791 btrfs_set_device_total_bytes(leaf, dev_item,
1792 btrfs_device_get_disk_total_bytes(device));
1793 btrfs_set_device_bytes_used(leaf, dev_item,
1794 btrfs_device_get_bytes_used(device));
1795 btrfs_set_device_group(leaf, dev_item, 0);
1796 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1797 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1798 btrfs_set_device_start_offset(leaf, dev_item, 0);
1800 ptr = btrfs_device_uuid(dev_item);
1801 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1802 ptr = btrfs_device_fsid(dev_item);
1803 write_extent_buffer(leaf, fs_info->fsid, ptr, BTRFS_FSID_SIZE);
1804 btrfs_mark_buffer_dirty(leaf);
1808 btrfs_free_path(path);
1813 * Function to update ctime/mtime for a given device path.
1814 * Mainly used for ctime/mtime based probe like libblkid.
1816 static void update_dev_time(const char *path_name)
1820 filp = filp_open(path_name, O_RDWR, 0);
1823 file_update_time(filp);
1824 filp_close(filp, NULL);
1827 static int btrfs_rm_dev_item(struct btrfs_fs_info *fs_info,
1828 struct btrfs_device *device)
1830 struct btrfs_root *root = fs_info->chunk_root;
1832 struct btrfs_path *path;
1833 struct btrfs_key key;
1834 struct btrfs_trans_handle *trans;
1836 path = btrfs_alloc_path();
1840 trans = btrfs_start_transaction(root, 0);
1841 if (IS_ERR(trans)) {
1842 btrfs_free_path(path);
1843 return PTR_ERR(trans);
1845 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1846 key.type = BTRFS_DEV_ITEM_KEY;
1847 key.offset = device->devid;
1849 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1853 btrfs_abort_transaction(trans, ret);
1854 btrfs_end_transaction(trans);
1858 ret = btrfs_del_item(trans, root, path);
1860 btrfs_abort_transaction(trans, ret);
1861 btrfs_end_transaction(trans);
1865 btrfs_free_path(path);
1867 ret = btrfs_commit_transaction(trans);
1872 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1873 * filesystem. It's up to the caller to adjust that number regarding eg. device
1876 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1884 seq = read_seqbegin(&fs_info->profiles_lock);
1886 all_avail = fs_info->avail_data_alloc_bits |
1887 fs_info->avail_system_alloc_bits |
1888 fs_info->avail_metadata_alloc_bits;
1889 } while (read_seqretry(&fs_info->profiles_lock, seq));
1891 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1892 if (!(all_avail & btrfs_raid_array[i].bg_flag))
1895 if (num_devices < btrfs_raid_array[i].devs_min) {
1896 int ret = btrfs_raid_array[i].mindev_error;
1906 static struct btrfs_device * btrfs_find_next_active_device(
1907 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device)
1909 struct btrfs_device *next_device;
1911 list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1912 if (next_device != device &&
1913 !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state)
1914 && next_device->bdev)
1922 * Helper function to check if the given device is part of s_bdev / latest_bdev
1923 * and replace it with the provided or the next active device, in the context
1924 * where this function called, there should be always be another device (or
1925 * this_dev) which is active.
1927 void btrfs_assign_next_active_device(struct btrfs_fs_info *fs_info,
1928 struct btrfs_device *device, struct btrfs_device *this_dev)
1930 struct btrfs_device *next_device;
1933 next_device = this_dev;
1935 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
1937 ASSERT(next_device);
1939 if (fs_info->sb->s_bdev &&
1940 (fs_info->sb->s_bdev == device->bdev))
1941 fs_info->sb->s_bdev = next_device->bdev;
1943 if (fs_info->fs_devices->latest_bdev == device->bdev)
1944 fs_info->fs_devices->latest_bdev = next_device->bdev;
1947 int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path,
1950 struct btrfs_device *device;
1951 struct btrfs_fs_devices *cur_devices;
1952 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
1956 mutex_lock(&uuid_mutex);
1958 num_devices = fs_devices->num_devices;
1959 btrfs_dev_replace_read_lock(&fs_info->dev_replace);
1960 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
1961 WARN_ON(num_devices < 1);
1964 btrfs_dev_replace_read_unlock(&fs_info->dev_replace);
1966 ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
1970 ret = btrfs_find_device_by_devspec(fs_info, devid, device_path,
1975 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1976 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1980 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
1981 fs_info->fs_devices->rw_devices == 1) {
1982 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1986 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1987 mutex_lock(&fs_info->chunk_mutex);
1988 list_del_init(&device->dev_alloc_list);
1989 device->fs_devices->rw_devices--;
1990 mutex_unlock(&fs_info->chunk_mutex);
1993 mutex_unlock(&uuid_mutex);
1994 ret = btrfs_shrink_device(device, 0);
1995 mutex_lock(&uuid_mutex);
2000 * TODO: the superblock still includes this device in its num_devices
2001 * counter although write_all_supers() is not locked out. This
2002 * could give a filesystem state which requires a degraded mount.
2004 ret = btrfs_rm_dev_item(fs_info, device);
2008 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2009 btrfs_scrub_cancel_dev(fs_info, device);
2012 * the device list mutex makes sure that we don't change
2013 * the device list while someone else is writing out all
2014 * the device supers. Whoever is writing all supers, should
2015 * lock the device list mutex before getting the number of
2016 * devices in the super block (super_copy). Conversely,
2017 * whoever updates the number of devices in the super block
2018 * (super_copy) should hold the device list mutex.
2022 * In normal cases the cur_devices == fs_devices. But in case
2023 * of deleting a seed device, the cur_devices should point to
2024 * its own fs_devices listed under the fs_devices->seed.
2026 cur_devices = device->fs_devices;
2027 mutex_lock(&fs_devices->device_list_mutex);
2028 list_del_rcu(&device->dev_list);
2030 cur_devices->num_devices--;
2031 cur_devices->total_devices--;
2033 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
2034 cur_devices->missing_devices--;
2036 btrfs_assign_next_active_device(fs_info, device, NULL);
2039 cur_devices->open_devices--;
2040 /* remove sysfs entry */
2041 btrfs_sysfs_rm_device_link(fs_devices, device);
2044 num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
2045 btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
2046 mutex_unlock(&fs_devices->device_list_mutex);
2049 * at this point, the device is zero sized and detached from
2050 * the devices list. All that's left is to zero out the old
2051 * supers and free the device.
2053 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2054 btrfs_scratch_superblocks(device->bdev, device->name->str);
2056 btrfs_close_bdev(device);
2057 call_rcu(&device->rcu, free_device_rcu);
2059 if (cur_devices->open_devices == 0) {
2060 while (fs_devices) {
2061 if (fs_devices->seed == cur_devices) {
2062 fs_devices->seed = cur_devices->seed;
2065 fs_devices = fs_devices->seed;
2067 cur_devices->seed = NULL;
2068 close_fs_devices(cur_devices);
2069 free_fs_devices(cur_devices);
2073 mutex_unlock(&uuid_mutex);
2077 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2078 mutex_lock(&fs_info->chunk_mutex);
2079 list_add(&device->dev_alloc_list,
2080 &fs_devices->alloc_list);
2081 device->fs_devices->rw_devices++;
2082 mutex_unlock(&fs_info->chunk_mutex);
2087 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
2088 struct btrfs_device *srcdev)
2090 struct btrfs_fs_devices *fs_devices;
2092 lockdep_assert_held(&fs_info->fs_devices->device_list_mutex);
2095 * in case of fs with no seed, srcdev->fs_devices will point
2096 * to fs_devices of fs_info. However when the dev being replaced is
2097 * a seed dev it will point to the seed's local fs_devices. In short
2098 * srcdev will have its correct fs_devices in both the cases.
2100 fs_devices = srcdev->fs_devices;
2102 list_del_rcu(&srcdev->dev_list);
2103 list_del(&srcdev->dev_alloc_list);
2104 fs_devices->num_devices--;
2105 if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state))
2106 fs_devices->missing_devices--;
2108 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state))
2109 fs_devices->rw_devices--;
2112 fs_devices->open_devices--;
2115 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
2116 struct btrfs_device *srcdev)
2118 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2120 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) {
2121 /* zero out the old super if it is writable */
2122 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
2125 btrfs_close_bdev(srcdev);
2126 call_rcu(&srcdev->rcu, free_device_rcu);
2128 /* if this is no devs we rather delete the fs_devices */
2129 if (!fs_devices->num_devices) {
2130 struct btrfs_fs_devices *tmp_fs_devices;
2133 * On a mounted FS, num_devices can't be zero unless it's a
2134 * seed. In case of a seed device being replaced, the replace
2135 * target added to the sprout FS, so there will be no more
2136 * device left under the seed FS.
2138 ASSERT(fs_devices->seeding);
2140 tmp_fs_devices = fs_info->fs_devices;
2141 while (tmp_fs_devices) {
2142 if (tmp_fs_devices->seed == fs_devices) {
2143 tmp_fs_devices->seed = fs_devices->seed;
2146 tmp_fs_devices = tmp_fs_devices->seed;
2148 fs_devices->seed = NULL;
2149 close_fs_devices(fs_devices);
2150 free_fs_devices(fs_devices);
2154 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2155 struct btrfs_device *tgtdev)
2157 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2160 mutex_lock(&fs_devices->device_list_mutex);
2162 btrfs_sysfs_rm_device_link(fs_devices, tgtdev);
2165 fs_devices->open_devices--;
2167 fs_devices->num_devices--;
2169 btrfs_assign_next_active_device(fs_info, tgtdev, NULL);
2171 list_del_rcu(&tgtdev->dev_list);
2173 mutex_unlock(&fs_devices->device_list_mutex);
2176 * The update_dev_time() with in btrfs_scratch_superblocks()
2177 * may lead to a call to btrfs_show_devname() which will try
2178 * to hold device_list_mutex. And here this device
2179 * is already out of device list, so we don't have to hold
2180 * the device_list_mutex lock.
2182 btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
2184 btrfs_close_bdev(tgtdev);
2185 call_rcu(&tgtdev->rcu, free_device_rcu);
2188 static int btrfs_find_device_by_path(struct btrfs_fs_info *fs_info,
2189 const char *device_path,
2190 struct btrfs_device **device)
2193 struct btrfs_super_block *disk_super;
2196 struct block_device *bdev;
2197 struct buffer_head *bh;
2200 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2201 fs_info->bdev_holder, 0, &bdev, &bh);
2204 disk_super = (struct btrfs_super_block *)bh->b_data;
2205 devid = btrfs_stack_device_id(&disk_super->dev_item);
2206 dev_uuid = disk_super->dev_item.uuid;
2207 *device = btrfs_find_device(fs_info, devid, dev_uuid, disk_super->fsid);
2211 blkdev_put(bdev, FMODE_READ);
2215 int btrfs_find_device_missing_or_by_path(struct btrfs_fs_info *fs_info,
2216 const char *device_path,
2217 struct btrfs_device **device)
2220 if (strcmp(device_path, "missing") == 0) {
2221 struct list_head *devices;
2222 struct btrfs_device *tmp;
2224 devices = &fs_info->fs_devices->devices;
2225 list_for_each_entry(tmp, devices, dev_list) {
2226 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
2227 &tmp->dev_state) && !tmp->bdev) {
2234 return BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2238 return btrfs_find_device_by_path(fs_info, device_path, device);
2243 * Lookup a device given by device id, or the path if the id is 0.
2245 int btrfs_find_device_by_devspec(struct btrfs_fs_info *fs_info, u64 devid,
2246 const char *devpath,
2247 struct btrfs_device **device)
2253 *device = btrfs_find_device(fs_info, devid, NULL, NULL);
2257 if (!devpath || !devpath[0])
2260 ret = btrfs_find_device_missing_or_by_path(fs_info, devpath,
2267 * does all the dirty work required for changing file system's UUID.
2269 static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info)
2271 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2272 struct btrfs_fs_devices *old_devices;
2273 struct btrfs_fs_devices *seed_devices;
2274 struct btrfs_super_block *disk_super = fs_info->super_copy;
2275 struct btrfs_device *device;
2278 lockdep_assert_held(&uuid_mutex);
2279 if (!fs_devices->seeding)
2282 seed_devices = alloc_fs_devices(NULL);
2283 if (IS_ERR(seed_devices))
2284 return PTR_ERR(seed_devices);
2286 old_devices = clone_fs_devices(fs_devices);
2287 if (IS_ERR(old_devices)) {
2288 kfree(seed_devices);
2289 return PTR_ERR(old_devices);
2292 list_add(&old_devices->fs_list, &fs_uuids);
2294 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2295 seed_devices->opened = 1;
2296 INIT_LIST_HEAD(&seed_devices->devices);
2297 INIT_LIST_HEAD(&seed_devices->alloc_list);
2298 mutex_init(&seed_devices->device_list_mutex);
2300 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2301 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2303 list_for_each_entry(device, &seed_devices->devices, dev_list)
2304 device->fs_devices = seed_devices;
2306 mutex_lock(&fs_info->chunk_mutex);
2307 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2308 mutex_unlock(&fs_info->chunk_mutex);
2310 fs_devices->seeding = 0;
2311 fs_devices->num_devices = 0;
2312 fs_devices->open_devices = 0;
2313 fs_devices->missing_devices = 0;
2314 fs_devices->rotating = 0;
2315 fs_devices->seed = seed_devices;
2317 generate_random_uuid(fs_devices->fsid);
2318 memcpy(fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2319 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2320 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2322 super_flags = btrfs_super_flags(disk_super) &
2323 ~BTRFS_SUPER_FLAG_SEEDING;
2324 btrfs_set_super_flags(disk_super, super_flags);
2330 * Store the expected generation for seed devices in device items.
2332 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2333 struct btrfs_fs_info *fs_info)
2335 struct btrfs_root *root = fs_info->chunk_root;
2336 struct btrfs_path *path;
2337 struct extent_buffer *leaf;
2338 struct btrfs_dev_item *dev_item;
2339 struct btrfs_device *device;
2340 struct btrfs_key key;
2341 u8 fs_uuid[BTRFS_FSID_SIZE];
2342 u8 dev_uuid[BTRFS_UUID_SIZE];
2346 path = btrfs_alloc_path();
2350 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2352 key.type = BTRFS_DEV_ITEM_KEY;
2355 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2359 leaf = path->nodes[0];
2361 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2362 ret = btrfs_next_leaf(root, path);
2367 leaf = path->nodes[0];
2368 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2369 btrfs_release_path(path);
2373 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2374 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2375 key.type != BTRFS_DEV_ITEM_KEY)
2378 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2379 struct btrfs_dev_item);
2380 devid = btrfs_device_id(leaf, dev_item);
2381 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2383 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2385 device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
2386 BUG_ON(!device); /* Logic error */
2388 if (device->fs_devices->seeding) {
2389 btrfs_set_device_generation(leaf, dev_item,
2390 device->generation);
2391 btrfs_mark_buffer_dirty(leaf);
2399 btrfs_free_path(path);
2403 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2405 struct btrfs_root *root = fs_info->dev_root;
2406 struct request_queue *q;
2407 struct btrfs_trans_handle *trans;
2408 struct btrfs_device *device;
2409 struct block_device *bdev;
2410 struct list_head *devices;
2411 struct super_block *sb = fs_info->sb;
2412 struct rcu_string *name;
2414 int seeding_dev = 0;
2416 bool unlocked = false;
2418 if (sb_rdonly(sb) && !fs_info->fs_devices->seeding)
2421 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2422 fs_info->bdev_holder);
2424 return PTR_ERR(bdev);
2426 if (fs_info->fs_devices->seeding) {
2428 down_write(&sb->s_umount);
2429 mutex_lock(&uuid_mutex);
2432 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2434 devices = &fs_info->fs_devices->devices;
2436 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2437 list_for_each_entry(device, devices, dev_list) {
2438 if (device->bdev == bdev) {
2441 &fs_info->fs_devices->device_list_mutex);
2445 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2447 device = btrfs_alloc_device(fs_info, NULL, NULL);
2448 if (IS_ERR(device)) {
2449 /* we can safely leave the fs_devices entry around */
2450 ret = PTR_ERR(device);
2454 name = rcu_string_strdup(device_path, GFP_KERNEL);
2457 goto error_free_device;
2459 rcu_assign_pointer(device->name, name);
2461 trans = btrfs_start_transaction(root, 0);
2462 if (IS_ERR(trans)) {
2463 ret = PTR_ERR(trans);
2464 goto error_free_device;
2467 q = bdev_get_queue(bdev);
2468 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
2469 device->generation = trans->transid;
2470 device->io_width = fs_info->sectorsize;
2471 device->io_align = fs_info->sectorsize;
2472 device->sector_size = fs_info->sectorsize;
2473 device->total_bytes = round_down(i_size_read(bdev->bd_inode),
2474 fs_info->sectorsize);
2475 device->disk_total_bytes = device->total_bytes;
2476 device->commit_total_bytes = device->total_bytes;
2477 device->fs_info = fs_info;
2478 device->bdev = bdev;
2479 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2480 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
2481 device->mode = FMODE_EXCL;
2482 device->dev_stats_valid = 1;
2483 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2486 sb->s_flags &= ~SB_RDONLY;
2487 ret = btrfs_prepare_sprout(fs_info);
2489 btrfs_abort_transaction(trans, ret);
2494 device->fs_devices = fs_info->fs_devices;
2496 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2497 mutex_lock(&fs_info->chunk_mutex);
2498 list_add_rcu(&device->dev_list, &fs_info->fs_devices->devices);
2499 list_add(&device->dev_alloc_list,
2500 &fs_info->fs_devices->alloc_list);
2501 fs_info->fs_devices->num_devices++;
2502 fs_info->fs_devices->open_devices++;
2503 fs_info->fs_devices->rw_devices++;
2504 fs_info->fs_devices->total_devices++;
2505 fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2507 atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2509 if (!blk_queue_nonrot(q))
2510 fs_info->fs_devices->rotating = 1;
2512 tmp = btrfs_super_total_bytes(fs_info->super_copy);
2513 btrfs_set_super_total_bytes(fs_info->super_copy,
2514 round_down(tmp + device->total_bytes, fs_info->sectorsize));
2516 tmp = btrfs_super_num_devices(fs_info->super_copy);
2517 btrfs_set_super_num_devices(fs_info->super_copy, tmp + 1);
2519 /* add sysfs device entry */
2520 btrfs_sysfs_add_device_link(fs_info->fs_devices, device);
2523 * we've got more storage, clear any full flags on the space
2526 btrfs_clear_space_info_full(fs_info);
2528 mutex_unlock(&fs_info->chunk_mutex);
2529 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2532 mutex_lock(&fs_info->chunk_mutex);
2533 ret = init_first_rw_device(trans, fs_info);
2534 mutex_unlock(&fs_info->chunk_mutex);
2536 btrfs_abort_transaction(trans, ret);
2541 ret = btrfs_add_dev_item(trans, fs_info, device);
2543 btrfs_abort_transaction(trans, ret);
2548 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2550 ret = btrfs_finish_sprout(trans, fs_info);
2552 btrfs_abort_transaction(trans, ret);
2556 /* Sprouting would change fsid of the mounted root,
2557 * so rename the fsid on the sysfs
2559 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2561 if (kobject_rename(&fs_info->fs_devices->fsid_kobj, fsid_buf))
2563 "sysfs: failed to create fsid for sprout");
2566 ret = btrfs_commit_transaction(trans);
2569 mutex_unlock(&uuid_mutex);
2570 up_write(&sb->s_umount);
2573 if (ret) /* transaction commit */
2576 ret = btrfs_relocate_sys_chunks(fs_info);
2578 btrfs_handle_fs_error(fs_info, ret,
2579 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2580 trans = btrfs_attach_transaction(root);
2581 if (IS_ERR(trans)) {
2582 if (PTR_ERR(trans) == -ENOENT)
2584 ret = PTR_ERR(trans);
2588 ret = btrfs_commit_transaction(trans);
2591 /* Update ctime/mtime for libblkid */
2592 update_dev_time(device_path);
2596 btrfs_sysfs_rm_device_link(fs_info->fs_devices, device);
2599 sb->s_flags |= SB_RDONLY;
2601 btrfs_end_transaction(trans);
2603 btrfs_free_device(device);
2605 blkdev_put(bdev, FMODE_EXCL);
2606 if (seeding_dev && !unlocked) {
2607 mutex_unlock(&uuid_mutex);
2608 up_write(&sb->s_umount);
2613 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2614 struct btrfs_device *device)
2617 struct btrfs_path *path;
2618 struct btrfs_root *root = device->fs_info->chunk_root;
2619 struct btrfs_dev_item *dev_item;
2620 struct extent_buffer *leaf;
2621 struct btrfs_key key;
2623 path = btrfs_alloc_path();
2627 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2628 key.type = BTRFS_DEV_ITEM_KEY;
2629 key.offset = device->devid;
2631 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2640 leaf = path->nodes[0];
2641 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2643 btrfs_set_device_id(leaf, dev_item, device->devid);
2644 btrfs_set_device_type(leaf, dev_item, device->type);
2645 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2646 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2647 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2648 btrfs_set_device_total_bytes(leaf, dev_item,
2649 btrfs_device_get_disk_total_bytes(device));
2650 btrfs_set_device_bytes_used(leaf, dev_item,
2651 btrfs_device_get_bytes_used(device));
2652 btrfs_mark_buffer_dirty(leaf);
2655 btrfs_free_path(path);
2659 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2660 struct btrfs_device *device, u64 new_size)
2662 struct btrfs_fs_info *fs_info = device->fs_info;
2663 struct btrfs_super_block *super_copy = fs_info->super_copy;
2664 struct btrfs_fs_devices *fs_devices;
2668 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2671 new_size = round_down(new_size, fs_info->sectorsize);
2673 mutex_lock(&fs_info->chunk_mutex);
2674 old_total = btrfs_super_total_bytes(super_copy);
2675 diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2677 if (new_size <= device->total_bytes ||
2678 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2679 mutex_unlock(&fs_info->chunk_mutex);
2683 fs_devices = fs_info->fs_devices;
2685 btrfs_set_super_total_bytes(super_copy,
2686 round_down(old_total + diff, fs_info->sectorsize));
2687 device->fs_devices->total_rw_bytes += diff;
2689 btrfs_device_set_total_bytes(device, new_size);
2690 btrfs_device_set_disk_total_bytes(device, new_size);
2691 btrfs_clear_space_info_full(device->fs_info);
2692 if (list_empty(&device->resized_list))
2693 list_add_tail(&device->resized_list,
2694 &fs_devices->resized_devices);
2695 mutex_unlock(&fs_info->chunk_mutex);
2697 return btrfs_update_device(trans, device);
2700 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2701 struct btrfs_fs_info *fs_info, u64 chunk_offset)
2703 struct btrfs_root *root = fs_info->chunk_root;
2705 struct btrfs_path *path;
2706 struct btrfs_key key;
2708 path = btrfs_alloc_path();
2712 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2713 key.offset = chunk_offset;
2714 key.type = BTRFS_CHUNK_ITEM_KEY;
2716 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2719 else if (ret > 0) { /* Logic error or corruption */
2720 btrfs_handle_fs_error(fs_info, -ENOENT,
2721 "Failed lookup while freeing chunk.");
2726 ret = btrfs_del_item(trans, root, path);
2728 btrfs_handle_fs_error(fs_info, ret,
2729 "Failed to delete chunk item.");
2731 btrfs_free_path(path);
2735 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2737 struct btrfs_super_block *super_copy = fs_info->super_copy;
2738 struct btrfs_disk_key *disk_key;
2739 struct btrfs_chunk *chunk;
2746 struct btrfs_key key;
2748 mutex_lock(&fs_info->chunk_mutex);
2749 array_size = btrfs_super_sys_array_size(super_copy);
2751 ptr = super_copy->sys_chunk_array;
2754 while (cur < array_size) {
2755 disk_key = (struct btrfs_disk_key *)ptr;
2756 btrfs_disk_key_to_cpu(&key, disk_key);
2758 len = sizeof(*disk_key);
2760 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2761 chunk = (struct btrfs_chunk *)(ptr + len);
2762 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2763 len += btrfs_chunk_item_size(num_stripes);
2768 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID &&
2769 key.offset == chunk_offset) {
2770 memmove(ptr, ptr + len, array_size - (cur + len));
2772 btrfs_set_super_sys_array_size(super_copy, array_size);
2778 mutex_unlock(&fs_info->chunk_mutex);
2782 static struct extent_map *get_chunk_map(struct btrfs_fs_info *fs_info,
2783 u64 logical, u64 length)
2785 struct extent_map_tree *em_tree;
2786 struct extent_map *em;
2788 em_tree = &fs_info->mapping_tree.map_tree;
2789 read_lock(&em_tree->lock);
2790 em = lookup_extent_mapping(em_tree, logical, length);
2791 read_unlock(&em_tree->lock);
2794 btrfs_crit(fs_info, "unable to find logical %llu length %llu",
2796 return ERR_PTR(-EINVAL);
2799 if (em->start > logical || em->start + em->len < logical) {
2801 "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
2802 logical, length, em->start, em->start + em->len);
2803 free_extent_map(em);
2804 return ERR_PTR(-EINVAL);
2807 /* callers are responsible for dropping em's ref. */
2811 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2812 struct btrfs_fs_info *fs_info, u64 chunk_offset)
2814 struct extent_map *em;
2815 struct map_lookup *map;
2816 u64 dev_extent_len = 0;
2818 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2820 em = get_chunk_map(fs_info, chunk_offset, 1);
2823 * This is a logic error, but we don't want to just rely on the
2824 * user having built with ASSERT enabled, so if ASSERT doesn't
2825 * do anything we still error out.
2830 map = em->map_lookup;
2831 mutex_lock(&fs_info->chunk_mutex);
2832 check_system_chunk(trans, fs_info, map->type);
2833 mutex_unlock(&fs_info->chunk_mutex);
2836 * Take the device list mutex to prevent races with the final phase of
2837 * a device replace operation that replaces the device object associated
2838 * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2840 mutex_lock(&fs_devices->device_list_mutex);
2841 for (i = 0; i < map->num_stripes; i++) {
2842 struct btrfs_device *device = map->stripes[i].dev;
2843 ret = btrfs_free_dev_extent(trans, device,
2844 map->stripes[i].physical,
2847 mutex_unlock(&fs_devices->device_list_mutex);
2848 btrfs_abort_transaction(trans, ret);
2852 if (device->bytes_used > 0) {
2853 mutex_lock(&fs_info->chunk_mutex);
2854 btrfs_device_set_bytes_used(device,
2855 device->bytes_used - dev_extent_len);
2856 atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
2857 btrfs_clear_space_info_full(fs_info);
2858 mutex_unlock(&fs_info->chunk_mutex);
2861 if (map->stripes[i].dev) {
2862 ret = btrfs_update_device(trans, map->stripes[i].dev);
2864 mutex_unlock(&fs_devices->device_list_mutex);
2865 btrfs_abort_transaction(trans, ret);
2870 mutex_unlock(&fs_devices->device_list_mutex);
2872 ret = btrfs_free_chunk(trans, fs_info, chunk_offset);
2874 btrfs_abort_transaction(trans, ret);
2878 trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
2880 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2881 ret = btrfs_del_sys_chunk(fs_info, chunk_offset);
2883 btrfs_abort_transaction(trans, ret);
2888 ret = btrfs_remove_block_group(trans, fs_info, chunk_offset, em);
2890 btrfs_abort_transaction(trans, ret);
2896 free_extent_map(em);
2900 static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2902 struct btrfs_root *root = fs_info->chunk_root;
2903 struct btrfs_trans_handle *trans;
2907 * Prevent races with automatic removal of unused block groups.
2908 * After we relocate and before we remove the chunk with offset
2909 * chunk_offset, automatic removal of the block group can kick in,
2910 * resulting in a failure when calling btrfs_remove_chunk() below.
2912 * Make sure to acquire this mutex before doing a tree search (dev
2913 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2914 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2915 * we release the path used to search the chunk/dev tree and before
2916 * the current task acquires this mutex and calls us.
2918 lockdep_assert_held(&fs_info->delete_unused_bgs_mutex);
2920 ret = btrfs_can_relocate(fs_info, chunk_offset);
2924 /* step one, relocate all the extents inside this chunk */
2925 btrfs_scrub_pause(fs_info);
2926 ret = btrfs_relocate_block_group(fs_info, chunk_offset);
2927 btrfs_scrub_continue(fs_info);
2932 * We add the kobjects here (and after forcing data chunk creation)
2933 * since relocation is the only place we'll create chunks of a new
2934 * type at runtime. The only place where we'll remove the last
2935 * chunk of a type is the call immediately below this one. Even
2936 * so, we're protected against races with the cleaner thread since
2937 * we're covered by the delete_unused_bgs_mutex.
2939 btrfs_add_raid_kobjects(fs_info);
2941 trans = btrfs_start_trans_remove_block_group(root->fs_info,
2943 if (IS_ERR(trans)) {
2944 ret = PTR_ERR(trans);
2945 btrfs_handle_fs_error(root->fs_info, ret, NULL);
2950 * step two, delete the device extents and the
2951 * chunk tree entries
2953 ret = btrfs_remove_chunk(trans, fs_info, chunk_offset);
2954 btrfs_end_transaction(trans);
2958 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
2960 struct btrfs_root *chunk_root = fs_info->chunk_root;
2961 struct btrfs_path *path;
2962 struct extent_buffer *leaf;
2963 struct btrfs_chunk *chunk;
2964 struct btrfs_key key;
2965 struct btrfs_key found_key;
2967 bool retried = false;
2971 path = btrfs_alloc_path();
2976 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2977 key.offset = (u64)-1;
2978 key.type = BTRFS_CHUNK_ITEM_KEY;
2981 mutex_lock(&fs_info->delete_unused_bgs_mutex);
2982 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2984 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
2987 BUG_ON(ret == 0); /* Corruption */
2989 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2992 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
2998 leaf = path->nodes[0];
2999 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3001 chunk = btrfs_item_ptr(leaf, path->slots[0],
3002 struct btrfs_chunk);
3003 chunk_type = btrfs_chunk_type(leaf, chunk);
3004 btrfs_release_path(path);
3006 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3007 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3013 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3015 if (found_key.offset == 0)
3017 key.offset = found_key.offset - 1;
3020 if (failed && !retried) {
3024 } else if (WARN_ON(failed && retried)) {
3028 btrfs_free_path(path);
3033 * return 1 : allocate a data chunk successfully,
3034 * return <0: errors during allocating a data chunk,
3035 * return 0 : no need to allocate a data chunk.
3037 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info,
3040 struct btrfs_block_group_cache *cache;
3044 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3046 chunk_type = cache->flags;
3047 btrfs_put_block_group(cache);
3049 if (chunk_type & BTRFS_BLOCK_GROUP_DATA) {
3050 spin_lock(&fs_info->data_sinfo->lock);
3051 bytes_used = fs_info->data_sinfo->bytes_used;
3052 spin_unlock(&fs_info->data_sinfo->lock);
3055 struct btrfs_trans_handle *trans;
3058 trans = btrfs_join_transaction(fs_info->tree_root);
3060 return PTR_ERR(trans);
3062 ret = btrfs_force_chunk_alloc(trans, fs_info,
3063 BTRFS_BLOCK_GROUP_DATA);
3064 btrfs_end_transaction(trans);
3068 btrfs_add_raid_kobjects(fs_info);
3076 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3077 struct btrfs_balance_control *bctl)
3079 struct btrfs_root *root = fs_info->tree_root;
3080 struct btrfs_trans_handle *trans;
3081 struct btrfs_balance_item *item;
3082 struct btrfs_disk_balance_args disk_bargs;
3083 struct btrfs_path *path;
3084 struct extent_buffer *leaf;
3085 struct btrfs_key key;
3088 path = btrfs_alloc_path();
3092 trans = btrfs_start_transaction(root, 0);
3093 if (IS_ERR(trans)) {
3094 btrfs_free_path(path);
3095 return PTR_ERR(trans);
3098 key.objectid = BTRFS_BALANCE_OBJECTID;
3099 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3102 ret = btrfs_insert_empty_item(trans, root, path, &key,
3107 leaf = path->nodes[0];
3108 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3110 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3112 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3113 btrfs_set_balance_data(leaf, item, &disk_bargs);
3114 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3115 btrfs_set_balance_meta(leaf, item, &disk_bargs);
3116 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3117 btrfs_set_balance_sys(leaf, item, &disk_bargs);
3119 btrfs_set_balance_flags(leaf, item, bctl->flags);
3121 btrfs_mark_buffer_dirty(leaf);
3123 btrfs_free_path(path);
3124 err = btrfs_commit_transaction(trans);
3130 static int del_balance_item(struct btrfs_fs_info *fs_info)
3132 struct btrfs_root *root = fs_info->tree_root;
3133 struct btrfs_trans_handle *trans;
3134 struct btrfs_path *path;
3135 struct btrfs_key key;
3138 path = btrfs_alloc_path();
3142 trans = btrfs_start_transaction(root, 0);
3143 if (IS_ERR(trans)) {
3144 btrfs_free_path(path);
3145 return PTR_ERR(trans);
3148 key.objectid = BTRFS_BALANCE_OBJECTID;
3149 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3152 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3160 ret = btrfs_del_item(trans, root, path);
3162 btrfs_free_path(path);
3163 err = btrfs_commit_transaction(trans);
3170 * This is a heuristic used to reduce the number of chunks balanced on
3171 * resume after balance was interrupted.
3173 static void update_balance_args(struct btrfs_balance_control *bctl)
3176 * Turn on soft mode for chunk types that were being converted.
3178 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3179 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3180 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3181 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3182 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3183 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3186 * Turn on usage filter if is not already used. The idea is
3187 * that chunks that we have already balanced should be
3188 * reasonably full. Don't do it for chunks that are being
3189 * converted - that will keep us from relocating unconverted
3190 * (albeit full) chunks.
3192 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3193 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3194 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3195 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3196 bctl->data.usage = 90;
3198 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3199 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3200 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3201 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3202 bctl->sys.usage = 90;
3204 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3205 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3206 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3207 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3208 bctl->meta.usage = 90;
3213 * Clear the balance status in fs_info and delete the balance item from disk.
3215 static void reset_balance_state(struct btrfs_fs_info *fs_info)
3217 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3220 BUG_ON(!fs_info->balance_ctl);
3222 spin_lock(&fs_info->balance_lock);
3223 fs_info->balance_ctl = NULL;
3224 spin_unlock(&fs_info->balance_lock);
3227 ret = del_balance_item(fs_info);
3229 btrfs_handle_fs_error(fs_info, ret, NULL);
3233 * Balance filters. Return 1 if chunk should be filtered out
3234 * (should not be balanced).
3236 static int chunk_profiles_filter(u64 chunk_type,
3237 struct btrfs_balance_args *bargs)
3239 chunk_type = chunk_to_extended(chunk_type) &
3240 BTRFS_EXTENDED_PROFILE_MASK;
3242 if (bargs->profiles & chunk_type)
3248 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3249 struct btrfs_balance_args *bargs)
3251 struct btrfs_block_group_cache *cache;
3253 u64 user_thresh_min;
3254 u64 user_thresh_max;
3257 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3258 chunk_used = btrfs_block_group_used(&cache->item);
3260 if (bargs->usage_min == 0)
3261 user_thresh_min = 0;
3263 user_thresh_min = div_factor_fine(cache->key.offset,
3266 if (bargs->usage_max == 0)
3267 user_thresh_max = 1;
3268 else if (bargs->usage_max > 100)
3269 user_thresh_max = cache->key.offset;
3271 user_thresh_max = div_factor_fine(cache->key.offset,
3274 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3277 btrfs_put_block_group(cache);
3281 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3282 u64 chunk_offset, struct btrfs_balance_args *bargs)
3284 struct btrfs_block_group_cache *cache;
3285 u64 chunk_used, user_thresh;
3288 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3289 chunk_used = btrfs_block_group_used(&cache->item);
3291 if (bargs->usage_min == 0)
3293 else if (bargs->usage > 100)
3294 user_thresh = cache->key.offset;
3296 user_thresh = div_factor_fine(cache->key.offset,
3299 if (chunk_used < user_thresh)
3302 btrfs_put_block_group(cache);
3306 static int chunk_devid_filter(struct extent_buffer *leaf,
3307 struct btrfs_chunk *chunk,
3308 struct btrfs_balance_args *bargs)
3310 struct btrfs_stripe *stripe;
3311 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3314 for (i = 0; i < num_stripes; i++) {
3315 stripe = btrfs_stripe_nr(chunk, i);
3316 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3323 /* [pstart, pend) */
3324 static int chunk_drange_filter(struct extent_buffer *leaf,
3325 struct btrfs_chunk *chunk,
3326 struct btrfs_balance_args *bargs)
3328 struct btrfs_stripe *stripe;
3329 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3335 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3338 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3339 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3340 factor = num_stripes / 2;
3341 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3342 factor = num_stripes - 1;
3343 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3344 factor = num_stripes - 2;
3346 factor = num_stripes;
3349 for (i = 0; i < num_stripes; i++) {
3350 stripe = btrfs_stripe_nr(chunk, i);
3351 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3354 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3355 stripe_length = btrfs_chunk_length(leaf, chunk);
3356 stripe_length = div_u64(stripe_length, factor);
3358 if (stripe_offset < bargs->pend &&
3359 stripe_offset + stripe_length > bargs->pstart)
3366 /* [vstart, vend) */
3367 static int chunk_vrange_filter(struct extent_buffer *leaf,
3368 struct btrfs_chunk *chunk,
3370 struct btrfs_balance_args *bargs)
3372 if (chunk_offset < bargs->vend &&
3373 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3374 /* at least part of the chunk is inside this vrange */
3380 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3381 struct btrfs_chunk *chunk,
3382 struct btrfs_balance_args *bargs)
3384 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3386 if (bargs->stripes_min <= num_stripes
3387 && num_stripes <= bargs->stripes_max)
3393 static int chunk_soft_convert_filter(u64 chunk_type,
3394 struct btrfs_balance_args *bargs)
3396 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3399 chunk_type = chunk_to_extended(chunk_type) &
3400 BTRFS_EXTENDED_PROFILE_MASK;
3402 if (bargs->target == chunk_type)
3408 static int should_balance_chunk(struct btrfs_fs_info *fs_info,
3409 struct extent_buffer *leaf,
3410 struct btrfs_chunk *chunk, u64 chunk_offset)
3412 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3413 struct btrfs_balance_args *bargs = NULL;
3414 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3417 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3418 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3422 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3423 bargs = &bctl->data;
3424 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3426 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3427 bargs = &bctl->meta;
3429 /* profiles filter */
3430 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3431 chunk_profiles_filter(chunk_type, bargs)) {
3436 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3437 chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3439 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3440 chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3445 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3446 chunk_devid_filter(leaf, chunk, bargs)) {
3450 /* drange filter, makes sense only with devid filter */
3451 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3452 chunk_drange_filter(leaf, chunk, bargs)) {
3457 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3458 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3462 /* stripes filter */
3463 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3464 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3468 /* soft profile changing mode */
3469 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3470 chunk_soft_convert_filter(chunk_type, bargs)) {
3475 * limited by count, must be the last filter
3477 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3478 if (bargs->limit == 0)
3482 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3484 * Same logic as the 'limit' filter; the minimum cannot be
3485 * determined here because we do not have the global information
3486 * about the count of all chunks that satisfy the filters.
3488 if (bargs->limit_max == 0)
3497 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3499 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3500 struct btrfs_root *chunk_root = fs_info->chunk_root;
3501 struct btrfs_root *dev_root = fs_info->dev_root;
3502 struct list_head *devices;
3503 struct btrfs_device *device;
3507 struct btrfs_chunk *chunk;
3508 struct btrfs_path *path = NULL;
3509 struct btrfs_key key;
3510 struct btrfs_key found_key;
3511 struct btrfs_trans_handle *trans;
3512 struct extent_buffer *leaf;
3515 int enospc_errors = 0;
3516 bool counting = true;
3517 /* The single value limit and min/max limits use the same bytes in the */
3518 u64 limit_data = bctl->data.limit;
3519 u64 limit_meta = bctl->meta.limit;
3520 u64 limit_sys = bctl->sys.limit;
3524 int chunk_reserved = 0;
3526 /* step one make some room on all the devices */
3527 devices = &fs_info->fs_devices->devices;
3528 list_for_each_entry(device, devices, dev_list) {
3529 old_size = btrfs_device_get_total_bytes(device);
3530 size_to_free = div_factor(old_size, 1);
3531 size_to_free = min_t(u64, size_to_free, SZ_1M);
3532 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) ||
3533 btrfs_device_get_total_bytes(device) -
3534 btrfs_device_get_bytes_used(device) > size_to_free ||
3535 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
3538 ret = btrfs_shrink_device(device, old_size - size_to_free);
3542 /* btrfs_shrink_device never returns ret > 0 */
3547 trans = btrfs_start_transaction(dev_root, 0);
3548 if (IS_ERR(trans)) {
3549 ret = PTR_ERR(trans);
3550 btrfs_info_in_rcu(fs_info,
3551 "resize: unable to start transaction after shrinking device %s (error %d), old size %llu, new size %llu",
3552 rcu_str_deref(device->name), ret,
3553 old_size, old_size - size_to_free);
3557 ret = btrfs_grow_device(trans, device, old_size);
3559 btrfs_end_transaction(trans);
3560 /* btrfs_grow_device never returns ret > 0 */
3562 btrfs_info_in_rcu(fs_info,
3563 "resize: unable to grow device after shrinking device %s (error %d), old size %llu, new size %llu",
3564 rcu_str_deref(device->name), ret,
3565 old_size, old_size - size_to_free);
3569 btrfs_end_transaction(trans);
3572 /* step two, relocate all the chunks */
3573 path = btrfs_alloc_path();
3579 /* zero out stat counters */
3580 spin_lock(&fs_info->balance_lock);
3581 memset(&bctl->stat, 0, sizeof(bctl->stat));
3582 spin_unlock(&fs_info->balance_lock);
3586 * The single value limit and min/max limits use the same bytes
3589 bctl->data.limit = limit_data;
3590 bctl->meta.limit = limit_meta;
3591 bctl->sys.limit = limit_sys;
3593 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3594 key.offset = (u64)-1;
3595 key.type = BTRFS_CHUNK_ITEM_KEY;
3598 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3599 atomic_read(&fs_info->balance_cancel_req)) {
3604 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3605 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3607 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3612 * this shouldn't happen, it means the last relocate
3616 BUG(); /* FIXME break ? */
3618 ret = btrfs_previous_item(chunk_root, path, 0,
3619 BTRFS_CHUNK_ITEM_KEY);
3621 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3626 leaf = path->nodes[0];
3627 slot = path->slots[0];
3628 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3630 if (found_key.objectid != key.objectid) {
3631 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3635 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3636 chunk_type = btrfs_chunk_type(leaf, chunk);
3639 spin_lock(&fs_info->balance_lock);
3640 bctl->stat.considered++;
3641 spin_unlock(&fs_info->balance_lock);
3644 ret = should_balance_chunk(fs_info, leaf, chunk,
3647 btrfs_release_path(path);
3649 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3654 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3655 spin_lock(&fs_info->balance_lock);
3656 bctl->stat.expected++;
3657 spin_unlock(&fs_info->balance_lock);
3659 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3661 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3663 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3670 * Apply limit_min filter, no need to check if the LIMITS
3671 * filter is used, limit_min is 0 by default
3673 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3674 count_data < bctl->data.limit_min)
3675 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3676 count_meta < bctl->meta.limit_min)
3677 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3678 count_sys < bctl->sys.limit_min)) {
3679 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3683 if (!chunk_reserved) {
3685 * We may be relocating the only data chunk we have,
3686 * which could potentially end up with losing data's
3687 * raid profile, so lets allocate an empty one in
3690 ret = btrfs_may_alloc_data_chunk(fs_info,
3693 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3695 } else if (ret == 1) {
3700 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3701 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3702 if (ret && ret != -ENOSPC)
3704 if (ret == -ENOSPC) {
3707 spin_lock(&fs_info->balance_lock);
3708 bctl->stat.completed++;
3709 spin_unlock(&fs_info->balance_lock);
3712 if (found_key.offset == 0)
3714 key.offset = found_key.offset - 1;
3718 btrfs_release_path(path);
3723 btrfs_free_path(path);
3724 if (enospc_errors) {
3725 btrfs_info(fs_info, "%d enospc errors during balance",
3735 * alloc_profile_is_valid - see if a given profile is valid and reduced
3736 * @flags: profile to validate
3737 * @extended: if true @flags is treated as an extended profile
3739 static int alloc_profile_is_valid(u64 flags, int extended)
3741 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3742 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3744 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3746 /* 1) check that all other bits are zeroed */
3750 /* 2) see if profile is reduced */
3752 return !extended; /* "0" is valid for usual profiles */
3754 /* true if exactly one bit set */
3755 return (flags & (flags - 1)) == 0;
3758 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3760 /* cancel requested || normal exit path */
3761 return atomic_read(&fs_info->balance_cancel_req) ||
3762 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3763 atomic_read(&fs_info->balance_cancel_req) == 0);
3766 /* Non-zero return value signifies invalidity */
3767 static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg,
3770 return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3771 (!alloc_profile_is_valid(bctl_arg->target, 1) ||
3772 (bctl_arg->target & ~allowed)));
3776 * Should be called with balance mutexe held
3778 int btrfs_balance(struct btrfs_fs_info *fs_info,
3779 struct btrfs_balance_control *bctl,
3780 struct btrfs_ioctl_balance_args *bargs)
3782 u64 meta_target, data_target;
3789 if (btrfs_fs_closing(fs_info) ||
3790 atomic_read(&fs_info->balance_pause_req) ||
3791 atomic_read(&fs_info->balance_cancel_req)) {
3796 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3797 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3801 * In case of mixed groups both data and meta should be picked,
3802 * and identical options should be given for both of them.
3804 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3805 if (mixed && (bctl->flags & allowed)) {
3806 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3807 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3808 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3810 "balance: mixed groups data and metadata options must be the same");
3816 num_devices = fs_info->fs_devices->num_devices;
3817 btrfs_dev_replace_read_lock(&fs_info->dev_replace);
3818 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3819 BUG_ON(num_devices < 1);
3822 btrfs_dev_replace_read_unlock(&fs_info->dev_replace);
3823 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE | BTRFS_BLOCK_GROUP_DUP;
3824 if (num_devices > 1)
3825 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3826 if (num_devices > 2)
3827 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3828 if (num_devices > 3)
3829 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3830 BTRFS_BLOCK_GROUP_RAID6);
3831 if (validate_convert_profile(&bctl->data, allowed)) {
3832 int index = btrfs_bg_flags_to_raid_index(bctl->data.target);
3835 "balance: invalid convert data profile %s",
3836 get_raid_name(index));
3840 if (validate_convert_profile(&bctl->meta, allowed)) {
3841 int index = btrfs_bg_flags_to_raid_index(bctl->meta.target);
3844 "balance: invalid convert metadata profile %s",
3845 get_raid_name(index));
3849 if (validate_convert_profile(&bctl->sys, allowed)) {
3850 int index = btrfs_bg_flags_to_raid_index(bctl->sys.target);
3853 "balance: invalid convert system profile %s",
3854 get_raid_name(index));
3859 /* allow to reduce meta or sys integrity only if force set */
3860 allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3861 BTRFS_BLOCK_GROUP_RAID10 |
3862 BTRFS_BLOCK_GROUP_RAID5 |
3863 BTRFS_BLOCK_GROUP_RAID6;
3865 seq = read_seqbegin(&fs_info->profiles_lock);
3867 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3868 (fs_info->avail_system_alloc_bits & allowed) &&
3869 !(bctl->sys.target & allowed)) ||
3870 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3871 (fs_info->avail_metadata_alloc_bits & allowed) &&
3872 !(bctl->meta.target & allowed))) {
3873 if (bctl->flags & BTRFS_BALANCE_FORCE) {
3875 "balance: force reducing metadata integrity");
3878 "balance: reduces metadata integrity, use --force if you want this");
3883 } while (read_seqretry(&fs_info->profiles_lock, seq));
3885 /* if we're not converting, the target field is uninitialized */
3886 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
3887 bctl->meta.target : fs_info->avail_metadata_alloc_bits;
3888 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
3889 bctl->data.target : fs_info->avail_data_alloc_bits;
3890 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <
3891 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {
3892 int meta_index = btrfs_bg_flags_to_raid_index(meta_target);
3893 int data_index = btrfs_bg_flags_to_raid_index(data_target);
3896 "balance: metadata profile %s has lower redundancy than data profile %s",
3897 get_raid_name(meta_index), get_raid_name(data_index));
3900 ret = insert_balance_item(fs_info, bctl);
3901 if (ret && ret != -EEXIST)
3904 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3905 BUG_ON(ret == -EEXIST);
3906 BUG_ON(fs_info->balance_ctl);
3907 spin_lock(&fs_info->balance_lock);
3908 fs_info->balance_ctl = bctl;
3909 spin_unlock(&fs_info->balance_lock);
3911 BUG_ON(ret != -EEXIST);
3912 spin_lock(&fs_info->balance_lock);
3913 update_balance_args(bctl);
3914 spin_unlock(&fs_info->balance_lock);
3917 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
3918 set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
3919 mutex_unlock(&fs_info->balance_mutex);
3921 ret = __btrfs_balance(fs_info);
3923 mutex_lock(&fs_info->balance_mutex);
3924 clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
3927 memset(bargs, 0, sizeof(*bargs));
3928 btrfs_update_ioctl_balance_args(fs_info, bargs);
3931 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3932 balance_need_close(fs_info)) {
3933 reset_balance_state(fs_info);
3934 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3937 wake_up(&fs_info->balance_wait_q);
3941 if (bctl->flags & BTRFS_BALANCE_RESUME)
3942 reset_balance_state(fs_info);
3945 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3950 static int balance_kthread(void *data)
3952 struct btrfs_fs_info *fs_info = data;
3955 mutex_lock(&fs_info->balance_mutex);
3956 if (fs_info->balance_ctl) {
3957 btrfs_info(fs_info, "balance: resuming");
3958 ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL);
3960 mutex_unlock(&fs_info->balance_mutex);
3965 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3967 struct task_struct *tsk;
3969 mutex_lock(&fs_info->balance_mutex);
3970 if (!fs_info->balance_ctl) {
3971 mutex_unlock(&fs_info->balance_mutex);
3974 mutex_unlock(&fs_info->balance_mutex);
3976 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
3977 btrfs_info(fs_info, "balance: resume skipped");
3982 * A ro->rw remount sequence should continue with the paused balance
3983 * regardless of who pauses it, system or the user as of now, so set
3986 spin_lock(&fs_info->balance_lock);
3987 fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME;
3988 spin_unlock(&fs_info->balance_lock);
3990 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3991 return PTR_ERR_OR_ZERO(tsk);
3994 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3996 struct btrfs_balance_control *bctl;
3997 struct btrfs_balance_item *item;
3998 struct btrfs_disk_balance_args disk_bargs;
3999 struct btrfs_path *path;
4000 struct extent_buffer *leaf;
4001 struct btrfs_key key;
4004 path = btrfs_alloc_path();
4008 key.objectid = BTRFS_BALANCE_OBJECTID;
4009 key.type = BTRFS_TEMPORARY_ITEM_KEY;
4012 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4015 if (ret > 0) { /* ret = -ENOENT; */
4020 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4026 leaf = path->nodes[0];
4027 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
4029 bctl->flags = btrfs_balance_flags(leaf, item);
4030 bctl->flags |= BTRFS_BALANCE_RESUME;
4032 btrfs_balance_data(leaf, item, &disk_bargs);
4033 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4034 btrfs_balance_meta(leaf, item, &disk_bargs);
4035 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4036 btrfs_balance_sys(leaf, item, &disk_bargs);
4037 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4040 * This should never happen, as the paused balance state is recovered
4041 * during mount without any chance of other exclusive ops to collide.
4043 * This gives the exclusive op status to balance and keeps in paused
4044 * state until user intervention (cancel or umount). If the ownership
4045 * cannot be assigned, show a message but do not fail. The balance
4046 * is in a paused state and must have fs_info::balance_ctl properly
4049 if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags))
4051 "balance: cannot set exclusive op status, resume manually");
4053 mutex_lock(&fs_info->balance_mutex);
4054 BUG_ON(fs_info->balance_ctl);
4055 spin_lock(&fs_info->balance_lock);
4056 fs_info->balance_ctl = bctl;
4057 spin_unlock(&fs_info->balance_lock);
4058 mutex_unlock(&fs_info->balance_mutex);
4060 btrfs_free_path(path);
4064 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4068 mutex_lock(&fs_info->balance_mutex);
4069 if (!fs_info->balance_ctl) {
4070 mutex_unlock(&fs_info->balance_mutex);
4074 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4075 atomic_inc(&fs_info->balance_pause_req);
4076 mutex_unlock(&fs_info->balance_mutex);
4078 wait_event(fs_info->balance_wait_q,
4079 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4081 mutex_lock(&fs_info->balance_mutex);
4082 /* we are good with balance_ctl ripped off from under us */
4083 BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4084 atomic_dec(&fs_info->balance_pause_req);
4089 mutex_unlock(&fs_info->balance_mutex);
4093 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4095 mutex_lock(&fs_info->balance_mutex);
4096 if (!fs_info->balance_ctl) {
4097 mutex_unlock(&fs_info->balance_mutex);
4102 * A paused balance with the item stored on disk can be resumed at
4103 * mount time if the mount is read-write. Otherwise it's still paused
4104 * and we must not allow cancelling as it deletes the item.
4106 if (sb_rdonly(fs_info->sb)) {
4107 mutex_unlock(&fs_info->balance_mutex);
4111 atomic_inc(&fs_info->balance_cancel_req);
4113 * if we are running just wait and return, balance item is
4114 * deleted in btrfs_balance in this case
4116 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4117 mutex_unlock(&fs_info->balance_mutex);
4118 wait_event(fs_info->balance_wait_q,
4119 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4120 mutex_lock(&fs_info->balance_mutex);
4122 mutex_unlock(&fs_info->balance_mutex);
4124 * Lock released to allow other waiters to continue, we'll
4125 * reexamine the status again.
4127 mutex_lock(&fs_info->balance_mutex);
4129 if (fs_info->balance_ctl) {
4130 reset_balance_state(fs_info);
4131 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4132 btrfs_info(fs_info, "balance: canceled");
4136 BUG_ON(fs_info->balance_ctl ||
4137 test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4138 atomic_dec(&fs_info->balance_cancel_req);
4139 mutex_unlock(&fs_info->balance_mutex);
4143 static int btrfs_uuid_scan_kthread(void *data)
4145 struct btrfs_fs_info *fs_info = data;
4146 struct btrfs_root *root = fs_info->tree_root;
4147 struct btrfs_key key;
4148 struct btrfs_path *path = NULL;
4150 struct extent_buffer *eb;
4152 struct btrfs_root_item root_item;
4154 struct btrfs_trans_handle *trans = NULL;
4156 path = btrfs_alloc_path();
4163 key.type = BTRFS_ROOT_ITEM_KEY;
4167 ret = btrfs_search_forward(root, &key, path,
4168 BTRFS_OLDEST_GENERATION);
4175 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4176 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4177 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4178 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4181 eb = path->nodes[0];
4182 slot = path->slots[0];
4183 item_size = btrfs_item_size_nr(eb, slot);
4184 if (item_size < sizeof(root_item))
4187 read_extent_buffer(eb, &root_item,
4188 btrfs_item_ptr_offset(eb, slot),
4189 (int)sizeof(root_item));
4190 if (btrfs_root_refs(&root_item) == 0)
4193 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4194 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4198 btrfs_release_path(path);
4200 * 1 - subvol uuid item
4201 * 1 - received_subvol uuid item
4203 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4204 if (IS_ERR(trans)) {
4205 ret = PTR_ERR(trans);
4213 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4214 ret = btrfs_uuid_tree_add(trans, root_item.uuid,
4215 BTRFS_UUID_KEY_SUBVOL,
4218 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4224 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4225 ret = btrfs_uuid_tree_add(trans,
4226 root_item.received_uuid,
4227 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4230 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4238 ret = btrfs_end_transaction(trans);
4244 btrfs_release_path(path);
4245 if (key.offset < (u64)-1) {
4247 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4249 key.type = BTRFS_ROOT_ITEM_KEY;
4250 } else if (key.objectid < (u64)-1) {
4252 key.type = BTRFS_ROOT_ITEM_KEY;
4261 btrfs_free_path(path);
4262 if (trans && !IS_ERR(trans))
4263 btrfs_end_transaction(trans);
4265 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4267 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4268 up(&fs_info->uuid_tree_rescan_sem);
4273 * Callback for btrfs_uuid_tree_iterate().
4275 * 0 check succeeded, the entry is not outdated.
4276 * < 0 if an error occurred.
4277 * > 0 if the check failed, which means the caller shall remove the entry.
4279 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
4280 u8 *uuid, u8 type, u64 subid)
4282 struct btrfs_key key;
4284 struct btrfs_root *subvol_root;
4286 if (type != BTRFS_UUID_KEY_SUBVOL &&
4287 type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
4290 key.objectid = subid;
4291 key.type = BTRFS_ROOT_ITEM_KEY;
4292 key.offset = (u64)-1;
4293 subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
4294 if (IS_ERR(subvol_root)) {
4295 ret = PTR_ERR(subvol_root);
4302 case BTRFS_UUID_KEY_SUBVOL:
4303 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
4306 case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
4307 if (memcmp(uuid, subvol_root->root_item.received_uuid,
4317 static int btrfs_uuid_rescan_kthread(void *data)
4319 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4323 * 1st step is to iterate through the existing UUID tree and
4324 * to delete all entries that contain outdated data.
4325 * 2nd step is to add all missing entries to the UUID tree.
4327 ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4329 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4330 up(&fs_info->uuid_tree_rescan_sem);
4333 return btrfs_uuid_scan_kthread(data);
4336 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4338 struct btrfs_trans_handle *trans;
4339 struct btrfs_root *tree_root = fs_info->tree_root;
4340 struct btrfs_root *uuid_root;
4341 struct task_struct *task;
4348 trans = btrfs_start_transaction(tree_root, 2);
4350 return PTR_ERR(trans);
4352 uuid_root = btrfs_create_tree(trans, fs_info,
4353 BTRFS_UUID_TREE_OBJECTID);
4354 if (IS_ERR(uuid_root)) {
4355 ret = PTR_ERR(uuid_root);
4356 btrfs_abort_transaction(trans, ret);
4357 btrfs_end_transaction(trans);
4361 fs_info->uuid_root = uuid_root;
4363 ret = btrfs_commit_transaction(trans);
4367 down(&fs_info->uuid_tree_rescan_sem);
4368 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4370 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4371 btrfs_warn(fs_info, "failed to start uuid_scan task");
4372 up(&fs_info->uuid_tree_rescan_sem);
4373 return PTR_ERR(task);
4379 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4381 struct task_struct *task;
4383 down(&fs_info->uuid_tree_rescan_sem);
4384 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4386 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4387 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4388 up(&fs_info->uuid_tree_rescan_sem);
4389 return PTR_ERR(task);
4396 * shrinking a device means finding all of the device extents past
4397 * the new size, and then following the back refs to the chunks.
4398 * The chunk relocation code actually frees the device extent
4400 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4402 struct btrfs_fs_info *fs_info = device->fs_info;
4403 struct btrfs_root *root = fs_info->dev_root;
4404 struct btrfs_trans_handle *trans;
4405 struct btrfs_dev_extent *dev_extent = NULL;
4406 struct btrfs_path *path;
4412 bool retried = false;
4413 bool checked_pending_chunks = false;
4414 struct extent_buffer *l;
4415 struct btrfs_key key;
4416 struct btrfs_super_block *super_copy = fs_info->super_copy;
4417 u64 old_total = btrfs_super_total_bytes(super_copy);
4418 u64 old_size = btrfs_device_get_total_bytes(device);
4421 new_size = round_down(new_size, fs_info->sectorsize);
4422 diff = round_down(old_size - new_size, fs_info->sectorsize);
4424 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4427 path = btrfs_alloc_path();
4431 path->reada = READA_BACK;
4433 mutex_lock(&fs_info->chunk_mutex);
4435 btrfs_device_set_total_bytes(device, new_size);
4436 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4437 device->fs_devices->total_rw_bytes -= diff;
4438 atomic64_sub(diff, &fs_info->free_chunk_space);
4440 mutex_unlock(&fs_info->chunk_mutex);
4443 key.objectid = device->devid;
4444 key.offset = (u64)-1;
4445 key.type = BTRFS_DEV_EXTENT_KEY;
4448 mutex_lock(&fs_info->delete_unused_bgs_mutex);
4449 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4451 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4455 ret = btrfs_previous_item(root, path, 0, key.type);
4457 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4462 btrfs_release_path(path);
4467 slot = path->slots[0];
4468 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4470 if (key.objectid != device->devid) {
4471 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4472 btrfs_release_path(path);
4476 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4477 length = btrfs_dev_extent_length(l, dev_extent);
4479 if (key.offset + length <= new_size) {
4480 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4481 btrfs_release_path(path);
4485 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4486 btrfs_release_path(path);
4489 * We may be relocating the only data chunk we have,
4490 * which could potentially end up with losing data's
4491 * raid profile, so lets allocate an empty one in
4494 ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset);
4496 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4500 ret = btrfs_relocate_chunk(fs_info, chunk_offset);
4501 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4502 if (ret && ret != -ENOSPC)
4506 } while (key.offset-- > 0);
4508 if (failed && !retried) {
4512 } else if (failed && retried) {
4517 /* Shrinking succeeded, else we would be at "done". */
4518 trans = btrfs_start_transaction(root, 0);
4519 if (IS_ERR(trans)) {
4520 ret = PTR_ERR(trans);
4524 mutex_lock(&fs_info->chunk_mutex);
4527 * We checked in the above loop all device extents that were already in
4528 * the device tree. However before we have updated the device's
4529 * total_bytes to the new size, we might have had chunk allocations that
4530 * have not complete yet (new block groups attached to transaction
4531 * handles), and therefore their device extents were not yet in the
4532 * device tree and we missed them in the loop above. So if we have any
4533 * pending chunk using a device extent that overlaps the device range
4534 * that we can not use anymore, commit the current transaction and
4535 * repeat the search on the device tree - this way we guarantee we will
4536 * not have chunks using device extents that end beyond 'new_size'.
4538 if (!checked_pending_chunks) {
4539 u64 start = new_size;
4540 u64 len = old_size - new_size;
4542 if (contains_pending_extent(trans->transaction, device,
4544 mutex_unlock(&fs_info->chunk_mutex);
4545 checked_pending_chunks = true;
4548 ret = btrfs_commit_transaction(trans);
4555 btrfs_device_set_disk_total_bytes(device, new_size);
4556 if (list_empty(&device->resized_list))
4557 list_add_tail(&device->resized_list,
4558 &fs_info->fs_devices->resized_devices);
4560 WARN_ON(diff > old_total);
4561 btrfs_set_super_total_bytes(super_copy,
4562 round_down(old_total - diff, fs_info->sectorsize));
4563 mutex_unlock(&fs_info->chunk_mutex);
4565 /* Now btrfs_update_device() will change the on-disk size. */
4566 ret = btrfs_update_device(trans, device);
4567 btrfs_end_transaction(trans);
4569 btrfs_free_path(path);
4571 mutex_lock(&fs_info->chunk_mutex);
4572 btrfs_device_set_total_bytes(device, old_size);
4573 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
4574 device->fs_devices->total_rw_bytes += diff;
4575 atomic64_add(diff, &fs_info->free_chunk_space);
4576 mutex_unlock(&fs_info->chunk_mutex);
4581 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
4582 struct btrfs_key *key,
4583 struct btrfs_chunk *chunk, int item_size)
4585 struct btrfs_super_block *super_copy = fs_info->super_copy;
4586 struct btrfs_disk_key disk_key;
4590 mutex_lock(&fs_info->chunk_mutex);
4591 array_size = btrfs_super_sys_array_size(super_copy);
4592 if (array_size + item_size + sizeof(disk_key)
4593 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4594 mutex_unlock(&fs_info->chunk_mutex);
4598 ptr = super_copy->sys_chunk_array + array_size;
4599 btrfs_cpu_key_to_disk(&disk_key, key);
4600 memcpy(ptr, &disk_key, sizeof(disk_key));
4601 ptr += sizeof(disk_key);
4602 memcpy(ptr, chunk, item_size);
4603 item_size += sizeof(disk_key);
4604 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4605 mutex_unlock(&fs_info->chunk_mutex);
4611 * sort the devices in descending order by max_avail, total_avail
4613 static int btrfs_cmp_device_info(const void *a, const void *b)
4615 const struct btrfs_device_info *di_a = a;
4616 const struct btrfs_device_info *di_b = b;
4618 if (di_a->max_avail > di_b->max_avail)
4620 if (di_a->max_avail < di_b->max_avail)
4622 if (di_a->total_avail > di_b->total_avail)
4624 if (di_a->total_avail < di_b->total_avail)
4629 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4631 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4634 btrfs_set_fs_incompat(info, RAID56);
4637 #define BTRFS_MAX_DEVS(info) ((BTRFS_MAX_ITEM_SIZE(info) \
4638 - sizeof(struct btrfs_chunk)) \
4639 / sizeof(struct btrfs_stripe) + 1)
4641 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
4642 - 2 * sizeof(struct btrfs_disk_key) \
4643 - 2 * sizeof(struct btrfs_chunk)) \
4644 / sizeof(struct btrfs_stripe) + 1)
4646 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4647 u64 start, u64 type)
4649 struct btrfs_fs_info *info = trans->fs_info;
4650 struct btrfs_fs_devices *fs_devices = info->fs_devices;
4651 struct btrfs_device *device;
4652 struct map_lookup *map = NULL;
4653 struct extent_map_tree *em_tree;
4654 struct extent_map *em;
4655 struct btrfs_device_info *devices_info = NULL;
4657 int num_stripes; /* total number of stripes to allocate */
4658 int data_stripes; /* number of stripes that count for
4660 int sub_stripes; /* sub_stripes info for map */
4661 int dev_stripes; /* stripes per dev */
4662 int devs_max; /* max devs to use */
4663 int devs_min; /* min devs needed */
4664 int devs_increment; /* ndevs has to be a multiple of this */
4665 int ncopies; /* how many copies to data has */
4667 u64 max_stripe_size;
4676 BUG_ON(!alloc_profile_is_valid(type, 0));
4678 if (list_empty(&fs_devices->alloc_list)) {
4679 if (btrfs_test_opt(info, ENOSPC_DEBUG))
4680 btrfs_debug(info, "%s: no writable device", __func__);
4684 index = btrfs_bg_flags_to_raid_index(type);
4686 sub_stripes = btrfs_raid_array[index].sub_stripes;
4687 dev_stripes = btrfs_raid_array[index].dev_stripes;
4688 devs_max = btrfs_raid_array[index].devs_max;
4689 devs_min = btrfs_raid_array[index].devs_min;
4690 devs_increment = btrfs_raid_array[index].devs_increment;
4691 ncopies = btrfs_raid_array[index].ncopies;
4693 if (type & BTRFS_BLOCK_GROUP_DATA) {
4694 max_stripe_size = SZ_1G;
4695 max_chunk_size = 10 * max_stripe_size;
4697 devs_max = BTRFS_MAX_DEVS(info);
4698 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4699 /* for larger filesystems, use larger metadata chunks */
4700 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
4701 max_stripe_size = SZ_1G;
4703 max_stripe_size = SZ_256M;
4704 max_chunk_size = max_stripe_size;
4706 devs_max = BTRFS_MAX_DEVS(info);
4707 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4708 max_stripe_size = SZ_32M;
4709 max_chunk_size = 2 * max_stripe_size;
4711 devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4713 btrfs_err(info, "invalid chunk type 0x%llx requested",
4718 /* we don't want a chunk larger than 10% of writeable space */
4719 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4722 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4728 * in the first pass through the devices list, we gather information
4729 * about the available holes on each device.
4732 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
4736 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4738 "BTRFS: read-only device in alloc_list\n");
4742 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
4743 &device->dev_state) ||
4744 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4747 if (device->total_bytes > device->bytes_used)
4748 total_avail = device->total_bytes - device->bytes_used;
4752 /* If there is no space on this device, skip it. */
4753 if (total_avail == 0)
4756 ret = find_free_dev_extent(trans, device,
4757 max_stripe_size * dev_stripes,
4758 &dev_offset, &max_avail);
4759 if (ret && ret != -ENOSPC)
4763 max_avail = max_stripe_size * dev_stripes;
4765 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes) {
4766 if (btrfs_test_opt(info, ENOSPC_DEBUG))
4768 "%s: devid %llu has no free space, have=%llu want=%u",
4769 __func__, device->devid, max_avail,
4770 BTRFS_STRIPE_LEN * dev_stripes);
4774 if (ndevs == fs_devices->rw_devices) {
4775 WARN(1, "%s: found more than %llu devices\n",
4776 __func__, fs_devices->rw_devices);
4779 devices_info[ndevs].dev_offset = dev_offset;
4780 devices_info[ndevs].max_avail = max_avail;
4781 devices_info[ndevs].total_avail = total_avail;
4782 devices_info[ndevs].dev = device;
4787 * now sort the devices by hole size / available space
4789 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4790 btrfs_cmp_device_info, NULL);
4792 /* round down to number of usable stripes */
4793 ndevs = round_down(ndevs, devs_increment);
4795 if (ndevs < devs_min) {
4797 if (btrfs_test_opt(info, ENOSPC_DEBUG)) {
4799 "%s: not enough devices with free space: have=%d minimum required=%d",
4800 __func__, ndevs, devs_min);
4805 ndevs = min(ndevs, devs_max);
4808 * The primary goal is to maximize the number of stripes, so use as
4809 * many devices as possible, even if the stripes are not maximum sized.
4811 * The DUP profile stores more than one stripe per device, the
4812 * max_avail is the total size so we have to adjust.
4814 stripe_size = div_u64(devices_info[ndevs - 1].max_avail, dev_stripes);
4815 num_stripes = ndevs * dev_stripes;
4818 * this will have to be fixed for RAID1 and RAID10 over
4821 data_stripes = num_stripes / ncopies;
4823 if (type & BTRFS_BLOCK_GROUP_RAID5)
4824 data_stripes = num_stripes - 1;
4826 if (type & BTRFS_BLOCK_GROUP_RAID6)
4827 data_stripes = num_stripes - 2;
4830 * Use the number of data stripes to figure out how big this chunk
4831 * is really going to be in terms of logical address space,
4832 * and compare that answer with the max chunk size
4834 if (stripe_size * data_stripes > max_chunk_size) {
4835 stripe_size = div_u64(max_chunk_size, data_stripes);
4837 /* bump the answer up to a 16MB boundary */
4838 stripe_size = round_up(stripe_size, SZ_16M);
4841 * But don't go higher than the limits we found while searching
4844 stripe_size = min(devices_info[ndevs - 1].max_avail,
4848 /* align to BTRFS_STRIPE_LEN */
4849 stripe_size = round_down(stripe_size, BTRFS_STRIPE_LEN);
4851 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4856 map->num_stripes = num_stripes;
4858 for (i = 0; i < ndevs; ++i) {
4859 for (j = 0; j < dev_stripes; ++j) {
4860 int s = i * dev_stripes + j;
4861 map->stripes[s].dev = devices_info[i].dev;
4862 map->stripes[s].physical = devices_info[i].dev_offset +
4866 map->stripe_len = BTRFS_STRIPE_LEN;
4867 map->io_align = BTRFS_STRIPE_LEN;
4868 map->io_width = BTRFS_STRIPE_LEN;
4870 map->sub_stripes = sub_stripes;
4872 num_bytes = stripe_size * data_stripes;
4874 trace_btrfs_chunk_alloc(info, map, start, num_bytes);
4876 em = alloc_extent_map();
4882 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4883 em->map_lookup = map;
4885 em->len = num_bytes;
4886 em->block_start = 0;
4887 em->block_len = em->len;
4888 em->orig_block_len = stripe_size;
4890 em_tree = &info->mapping_tree.map_tree;
4891 write_lock(&em_tree->lock);
4892 ret = add_extent_mapping(em_tree, em, 0);
4894 write_unlock(&em_tree->lock);
4895 free_extent_map(em);
4899 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4900 refcount_inc(&em->refs);
4901 write_unlock(&em_tree->lock);
4903 ret = btrfs_make_block_group(trans, info, 0, type, start, num_bytes);
4905 goto error_del_extent;
4907 for (i = 0; i < map->num_stripes; i++) {
4908 num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4909 btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4912 atomic64_sub(stripe_size * map->num_stripes, &info->free_chunk_space);
4914 free_extent_map(em);
4915 check_raid56_incompat_flag(info, type);
4917 kfree(devices_info);
4921 write_lock(&em_tree->lock);
4922 remove_extent_mapping(em_tree, em);
4923 write_unlock(&em_tree->lock);
4925 /* One for our allocation */
4926 free_extent_map(em);
4927 /* One for the tree reference */
4928 free_extent_map(em);
4929 /* One for the pending_chunks list reference */
4930 free_extent_map(em);
4932 kfree(devices_info);
4936 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4937 struct btrfs_fs_info *fs_info,
4938 u64 chunk_offset, u64 chunk_size)
4940 struct btrfs_root *extent_root = fs_info->extent_root;
4941 struct btrfs_root *chunk_root = fs_info->chunk_root;
4942 struct btrfs_key key;
4943 struct btrfs_device *device;
4944 struct btrfs_chunk *chunk;
4945 struct btrfs_stripe *stripe;
4946 struct extent_map *em;
4947 struct map_lookup *map;
4954 em = get_chunk_map(fs_info, chunk_offset, chunk_size);
4958 map = em->map_lookup;
4959 item_size = btrfs_chunk_item_size(map->num_stripes);
4960 stripe_size = em->orig_block_len;
4962 chunk = kzalloc(item_size, GFP_NOFS);
4969 * Take the device list mutex to prevent races with the final phase of
4970 * a device replace operation that replaces the device object associated
4971 * with the map's stripes, because the device object's id can change
4972 * at any time during that final phase of the device replace operation
4973 * (dev-replace.c:btrfs_dev_replace_finishing()).
4975 mutex_lock(&fs_info->fs_devices->device_list_mutex);
4976 for (i = 0; i < map->num_stripes; i++) {
4977 device = map->stripes[i].dev;
4978 dev_offset = map->stripes[i].physical;
4980 ret = btrfs_update_device(trans, device);
4983 ret = btrfs_alloc_dev_extent(trans, device, chunk_offset,
4984 dev_offset, stripe_size);
4989 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4993 stripe = &chunk->stripe;
4994 for (i = 0; i < map->num_stripes; i++) {
4995 device = map->stripes[i].dev;
4996 dev_offset = map->stripes[i].physical;
4998 btrfs_set_stack_stripe_devid(stripe, device->devid);
4999 btrfs_set_stack_stripe_offset(stripe, dev_offset);
5000 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
5003 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
5005 btrfs_set_stack_chunk_length(chunk, chunk_size);
5006 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
5007 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
5008 btrfs_set_stack_chunk_type(chunk, map->type);
5009 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
5010 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
5011 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
5012 btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
5013 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
5015 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
5016 key.type = BTRFS_CHUNK_ITEM_KEY;
5017 key.offset = chunk_offset;
5019 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
5020 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
5022 * TODO: Cleanup of inserted chunk root in case of
5025 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
5030 free_extent_map(em);
5035 * Chunk allocation falls into two parts. The first part does works
5036 * that make the new allocated chunk useable, but not do any operation
5037 * that modifies the chunk tree. The second part does the works that
5038 * require modifying the chunk tree. This division is important for the
5039 * bootstrap process of adding storage to a seed btrfs.
5041 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
5042 struct btrfs_fs_info *fs_info, u64 type)
5046 lockdep_assert_held(&fs_info->chunk_mutex);
5047 chunk_offset = find_next_chunk(fs_info);
5048 return __btrfs_alloc_chunk(trans, chunk_offset, type);
5051 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
5052 struct btrfs_fs_info *fs_info)
5055 u64 sys_chunk_offset;
5059 chunk_offset = find_next_chunk(fs_info);
5060 alloc_profile = btrfs_metadata_alloc_profile(fs_info);
5061 ret = __btrfs_alloc_chunk(trans, chunk_offset, alloc_profile);
5065 sys_chunk_offset = find_next_chunk(fs_info);
5066 alloc_profile = btrfs_system_alloc_profile(fs_info);
5067 ret = __btrfs_alloc_chunk(trans, sys_chunk_offset, alloc_profile);
5071 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
5075 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5076 BTRFS_BLOCK_GROUP_RAID10 |
5077 BTRFS_BLOCK_GROUP_RAID5 |
5078 BTRFS_BLOCK_GROUP_DUP)) {
5080 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
5089 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset)
5091 struct extent_map *em;
5092 struct map_lookup *map;
5097 em = get_chunk_map(fs_info, chunk_offset, 1);
5101 map = em->map_lookup;
5102 for (i = 0; i < map->num_stripes; i++) {
5103 if (test_bit(BTRFS_DEV_STATE_MISSING,
5104 &map->stripes[i].dev->dev_state)) {
5108 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE,
5109 &map->stripes[i].dev->dev_state)) {
5116 * If the number of missing devices is larger than max errors,
5117 * we can not write the data into that chunk successfully, so
5120 if (miss_ndevs > btrfs_chunk_max_errors(map))
5123 free_extent_map(em);
5127 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
5129 extent_map_tree_init(&tree->map_tree);
5132 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
5134 struct extent_map *em;
5137 write_lock(&tree->map_tree.lock);
5138 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
5140 remove_extent_mapping(&tree->map_tree, em);
5141 write_unlock(&tree->map_tree.lock);
5145 free_extent_map(em);
5146 /* once for the tree */
5147 free_extent_map(em);
5151 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5153 struct extent_map *em;
5154 struct map_lookup *map;
5157 em = get_chunk_map(fs_info, logical, len);
5160 * We could return errors for these cases, but that could get
5161 * ugly and we'd probably do the same thing which is just not do
5162 * anything else and exit, so return 1 so the callers don't try
5163 * to use other copies.
5167 map = em->map_lookup;
5168 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
5169 ret = map->num_stripes;
5170 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5171 ret = map->sub_stripes;
5172 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5174 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5176 * There could be two corrupted data stripes, we need
5177 * to loop retry in order to rebuild the correct data.
5179 * Fail a stripe at a time on every retry except the
5180 * stripe under reconstruction.
5182 ret = map->num_stripes;
5185 free_extent_map(em);
5187 btrfs_dev_replace_read_lock(&fs_info->dev_replace);
5188 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) &&
5189 fs_info->dev_replace.tgtdev)
5191 btrfs_dev_replace_read_unlock(&fs_info->dev_replace);
5196 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
5199 struct extent_map *em;
5200 struct map_lookup *map;
5201 unsigned long len = fs_info->sectorsize;
5203 em = get_chunk_map(fs_info, logical, len);
5205 if (!WARN_ON(IS_ERR(em))) {
5206 map = em->map_lookup;
5207 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5208 len = map->stripe_len * nr_data_stripes(map);
5209 free_extent_map(em);
5214 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5216 struct extent_map *em;
5217 struct map_lookup *map;
5220 em = get_chunk_map(fs_info, logical, len);
5222 if(!WARN_ON(IS_ERR(em))) {
5223 map = em->map_lookup;
5224 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5226 free_extent_map(em);
5231 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5232 struct map_lookup *map, int first,
5233 int dev_replace_is_ongoing)
5237 int preferred_mirror;
5239 struct btrfs_device *srcdev;
5242 (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)));
5244 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5245 num_stripes = map->sub_stripes;
5247 num_stripes = map->num_stripes;
5249 preferred_mirror = first + current->pid % num_stripes;
5251 if (dev_replace_is_ongoing &&
5252 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5253 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5254 srcdev = fs_info->dev_replace.srcdev;
5259 * try to avoid the drive that is the source drive for a
5260 * dev-replace procedure, only choose it if no other non-missing
5261 * mirror is available
5263 for (tolerance = 0; tolerance < 2; tolerance++) {
5264 if (map->stripes[preferred_mirror].dev->bdev &&
5265 (tolerance || map->stripes[preferred_mirror].dev != srcdev))
5266 return preferred_mirror;
5267 for (i = first; i < first + num_stripes; i++) {
5268 if (map->stripes[i].dev->bdev &&
5269 (tolerance || map->stripes[i].dev != srcdev))
5274 /* we couldn't find one that doesn't fail. Just return something
5275 * and the io error handling code will clean up eventually
5277 return preferred_mirror;
5280 static inline int parity_smaller(u64 a, u64 b)
5285 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5286 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5288 struct btrfs_bio_stripe s;
5295 for (i = 0; i < num_stripes - 1; i++) {
5296 if (parity_smaller(bbio->raid_map[i],
5297 bbio->raid_map[i+1])) {
5298 s = bbio->stripes[i];
5299 l = bbio->raid_map[i];
5300 bbio->stripes[i] = bbio->stripes[i+1];
5301 bbio->raid_map[i] = bbio->raid_map[i+1];
5302 bbio->stripes[i+1] = s;
5303 bbio->raid_map[i+1] = l;
5311 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5313 struct btrfs_bio *bbio = kzalloc(
5314 /* the size of the btrfs_bio */
5315 sizeof(struct btrfs_bio) +
5316 /* plus the variable array for the stripes */
5317 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5318 /* plus the variable array for the tgt dev */
5319 sizeof(int) * (real_stripes) +
5321 * plus the raid_map, which includes both the tgt dev
5324 sizeof(u64) * (total_stripes),
5325 GFP_NOFS|__GFP_NOFAIL);
5327 atomic_set(&bbio->error, 0);
5328 refcount_set(&bbio->refs, 1);
5333 void btrfs_get_bbio(struct btrfs_bio *bbio)
5335 WARN_ON(!refcount_read(&bbio->refs));
5336 refcount_inc(&bbio->refs);
5339 void btrfs_put_bbio(struct btrfs_bio *bbio)
5343 if (refcount_dec_and_test(&bbio->refs))
5347 /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */
5349 * Please note that, discard won't be sent to target device of device
5352 static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info,
5353 u64 logical, u64 length,
5354 struct btrfs_bio **bbio_ret)
5356 struct extent_map *em;
5357 struct map_lookup *map;
5358 struct btrfs_bio *bbio;
5362 u64 stripe_end_offset;
5369 u32 sub_stripes = 0;
5370 u64 stripes_per_dev = 0;
5371 u32 remaining_stripes = 0;
5372 u32 last_stripe = 0;
5376 /* discard always return a bbio */
5379 em = get_chunk_map(fs_info, logical, length);
5383 map = em->map_lookup;
5384 /* we don't discard raid56 yet */
5385 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5390 offset = logical - em->start;
5391 length = min_t(u64, em->len - offset, length);
5393 stripe_len = map->stripe_len;
5395 * stripe_nr counts the total number of stripes we have to stride
5396 * to get to this block
5398 stripe_nr = div64_u64(offset, stripe_len);
5400 /* stripe_offset is the offset of this block in its stripe */
5401 stripe_offset = offset - stripe_nr * stripe_len;
5403 stripe_nr_end = round_up(offset + length, map->stripe_len);
5404 stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len);
5405 stripe_cnt = stripe_nr_end - stripe_nr;
5406 stripe_end_offset = stripe_nr_end * map->stripe_len -
5409 * after this, stripe_nr is the number of stripes on this
5410 * device we have to walk to find the data, and stripe_index is
5411 * the number of our device in the stripe array
5415 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5416 BTRFS_BLOCK_GROUP_RAID10)) {
5417 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5420 sub_stripes = map->sub_stripes;
5422 factor = map->num_stripes / sub_stripes;
5423 num_stripes = min_t(u64, map->num_stripes,
5424 sub_stripes * stripe_cnt);
5425 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5426 stripe_index *= sub_stripes;
5427 stripes_per_dev = div_u64_rem(stripe_cnt, factor,
5428 &remaining_stripes);
5429 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5430 last_stripe *= sub_stripes;
5431 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5432 BTRFS_BLOCK_GROUP_DUP)) {
5433 num_stripes = map->num_stripes;
5435 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5439 bbio = alloc_btrfs_bio(num_stripes, 0);
5445 for (i = 0; i < num_stripes; i++) {
5446 bbio->stripes[i].physical =
5447 map->stripes[stripe_index].physical +
5448 stripe_offset + stripe_nr * map->stripe_len;
5449 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5451 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5452 BTRFS_BLOCK_GROUP_RAID10)) {
5453 bbio->stripes[i].length = stripes_per_dev *
5456 if (i / sub_stripes < remaining_stripes)
5457 bbio->stripes[i].length +=
5461 * Special for the first stripe and
5464 * |-------|...|-------|
5468 if (i < sub_stripes)
5469 bbio->stripes[i].length -=
5472 if (stripe_index >= last_stripe &&
5473 stripe_index <= (last_stripe +
5475 bbio->stripes[i].length -=
5478 if (i == sub_stripes - 1)
5481 bbio->stripes[i].length = length;
5485 if (stripe_index == map->num_stripes) {
5492 bbio->map_type = map->type;
5493 bbio->num_stripes = num_stripes;
5495 free_extent_map(em);
5500 * In dev-replace case, for repair case (that's the only case where the mirror
5501 * is selected explicitly when calling btrfs_map_block), blocks left of the
5502 * left cursor can also be read from the target drive.
5504 * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the
5506 * For READ, it also needs to be supported using the same mirror number.
5508 * If the requested block is not left of the left cursor, EIO is returned. This
5509 * can happen because btrfs_num_copies() returns one more in the dev-replace
5512 static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info,
5513 u64 logical, u64 length,
5514 u64 srcdev_devid, int *mirror_num,
5517 struct btrfs_bio *bbio = NULL;
5519 int index_srcdev = 0;
5521 u64 physical_of_found = 0;
5525 ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS,
5526 logical, &length, &bbio, 0, 0);
5528 ASSERT(bbio == NULL);
5532 num_stripes = bbio->num_stripes;
5533 if (*mirror_num > num_stripes) {
5535 * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror,
5536 * that means that the requested area is not left of the left
5539 btrfs_put_bbio(bbio);
5544 * process the rest of the function using the mirror_num of the source
5545 * drive. Therefore look it up first. At the end, patch the device
5546 * pointer to the one of the target drive.
5548 for (i = 0; i < num_stripes; i++) {
5549 if (bbio->stripes[i].dev->devid != srcdev_devid)
5553 * In case of DUP, in order to keep it simple, only add the
5554 * mirror with the lowest physical address
5557 physical_of_found <= bbio->stripes[i].physical)
5562 physical_of_found = bbio->stripes[i].physical;
5565 btrfs_put_bbio(bbio);
5571 *mirror_num = index_srcdev + 1;
5572 *physical = physical_of_found;
5576 static void handle_ops_on_dev_replace(enum btrfs_map_op op,
5577 struct btrfs_bio **bbio_ret,
5578 struct btrfs_dev_replace *dev_replace,
5579 int *num_stripes_ret, int *max_errors_ret)
5581 struct btrfs_bio *bbio = *bbio_ret;
5582 u64 srcdev_devid = dev_replace->srcdev->devid;
5583 int tgtdev_indexes = 0;
5584 int num_stripes = *num_stripes_ret;
5585 int max_errors = *max_errors_ret;
5588 if (op == BTRFS_MAP_WRITE) {
5589 int index_where_to_add;
5592 * duplicate the write operations while the dev replace
5593 * procedure is running. Since the copying of the old disk to
5594 * the new disk takes place at run time while the filesystem is
5595 * mounted writable, the regular write operations to the old
5596 * disk have to be duplicated to go to the new disk as well.
5598 * Note that device->missing is handled by the caller, and that
5599 * the write to the old disk is already set up in the stripes
5602 index_where_to_add = num_stripes;
5603 for (i = 0; i < num_stripes; i++) {
5604 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5605 /* write to new disk, too */
5606 struct btrfs_bio_stripe *new =
5607 bbio->stripes + index_where_to_add;
5608 struct btrfs_bio_stripe *old =
5611 new->physical = old->physical;
5612 new->length = old->length;
5613 new->dev = dev_replace->tgtdev;
5614 bbio->tgtdev_map[i] = index_where_to_add;
5615 index_where_to_add++;
5620 num_stripes = index_where_to_add;
5621 } else if (op == BTRFS_MAP_GET_READ_MIRRORS) {
5622 int index_srcdev = 0;
5624 u64 physical_of_found = 0;
5627 * During the dev-replace procedure, the target drive can also
5628 * be used to read data in case it is needed to repair a corrupt
5629 * block elsewhere. This is possible if the requested area is
5630 * left of the left cursor. In this area, the target drive is a
5631 * full copy of the source drive.
5633 for (i = 0; i < num_stripes; i++) {
5634 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5636 * In case of DUP, in order to keep it simple,
5637 * only add the mirror with the lowest physical
5641 physical_of_found <=
5642 bbio->stripes[i].physical)
5646 physical_of_found = bbio->stripes[i].physical;
5650 struct btrfs_bio_stripe *tgtdev_stripe =
5651 bbio->stripes + num_stripes;
5653 tgtdev_stripe->physical = physical_of_found;
5654 tgtdev_stripe->length =
5655 bbio->stripes[index_srcdev].length;
5656 tgtdev_stripe->dev = dev_replace->tgtdev;
5657 bbio->tgtdev_map[index_srcdev] = num_stripes;
5664 *num_stripes_ret = num_stripes;
5665 *max_errors_ret = max_errors;
5666 bbio->num_tgtdevs = tgtdev_indexes;
5670 static bool need_full_stripe(enum btrfs_map_op op)
5672 return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS);
5675 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
5676 enum btrfs_map_op op,
5677 u64 logical, u64 *length,
5678 struct btrfs_bio **bbio_ret,
5679 int mirror_num, int need_raid_map)
5681 struct extent_map *em;
5682 struct map_lookup *map;
5692 int tgtdev_indexes = 0;
5693 struct btrfs_bio *bbio = NULL;
5694 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
5695 int dev_replace_is_ongoing = 0;
5696 int num_alloc_stripes;
5697 int patch_the_first_stripe_for_dev_replace = 0;
5698 u64 physical_to_patch_in_first_stripe = 0;
5699 u64 raid56_full_stripe_start = (u64)-1;
5701 if (op == BTRFS_MAP_DISCARD)
5702 return __btrfs_map_block_for_discard(fs_info, logical,
5705 em = get_chunk_map(fs_info, logical, *length);
5709 map = em->map_lookup;
5710 offset = logical - em->start;
5712 stripe_len = map->stripe_len;
5715 * stripe_nr counts the total number of stripes we have to stride
5716 * to get to this block
5718 stripe_nr = div64_u64(stripe_nr, stripe_len);
5720 stripe_offset = stripe_nr * stripe_len;
5721 if (offset < stripe_offset) {
5723 "stripe math has gone wrong, stripe_offset=%llu, offset=%llu, start=%llu, logical=%llu, stripe_len=%llu",
5724 stripe_offset, offset, em->start, logical,
5726 free_extent_map(em);
5730 /* stripe_offset is the offset of this block in its stripe*/
5731 stripe_offset = offset - stripe_offset;
5733 /* if we're here for raid56, we need to know the stripe aligned start */
5734 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5735 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5736 raid56_full_stripe_start = offset;
5738 /* allow a write of a full stripe, but make sure we don't
5739 * allow straddling of stripes
5741 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5743 raid56_full_stripe_start *= full_stripe_len;
5746 if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5748 /* For writes to RAID[56], allow a full stripeset across all disks.
5749 For other RAID types and for RAID[56] reads, just allow a single
5750 stripe (on a single disk). */
5751 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
5752 (op == BTRFS_MAP_WRITE)) {
5753 max_len = stripe_len * nr_data_stripes(map) -
5754 (offset - raid56_full_stripe_start);
5756 /* we limit the length of each bio to what fits in a stripe */
5757 max_len = stripe_len - stripe_offset;
5759 *length = min_t(u64, em->len - offset, max_len);
5761 *length = em->len - offset;
5764 /* This is for when we're called from btrfs_merge_bio_hook() and all
5765 it cares about is the length */
5769 btrfs_dev_replace_read_lock(dev_replace);
5770 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5771 if (!dev_replace_is_ongoing)
5772 btrfs_dev_replace_read_unlock(dev_replace);
5774 btrfs_dev_replace_set_lock_blocking(dev_replace);
5776 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
5777 !need_full_stripe(op) && dev_replace->tgtdev != NULL) {
5778 ret = get_extra_mirror_from_replace(fs_info, logical, *length,
5779 dev_replace->srcdev->devid,
5781 &physical_to_patch_in_first_stripe);
5785 patch_the_first_stripe_for_dev_replace = 1;
5786 } else if (mirror_num > map->num_stripes) {
5792 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5793 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5795 if (!need_full_stripe(op))
5797 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
5798 if (need_full_stripe(op))
5799 num_stripes = map->num_stripes;
5800 else if (mirror_num)
5801 stripe_index = mirror_num - 1;
5803 stripe_index = find_live_mirror(fs_info, map, 0,
5804 dev_replace_is_ongoing);
5805 mirror_num = stripe_index + 1;
5808 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
5809 if (need_full_stripe(op)) {
5810 num_stripes = map->num_stripes;
5811 } else if (mirror_num) {
5812 stripe_index = mirror_num - 1;
5817 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5818 u32 factor = map->num_stripes / map->sub_stripes;
5820 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5821 stripe_index *= map->sub_stripes;
5823 if (need_full_stripe(op))
5824 num_stripes = map->sub_stripes;
5825 else if (mirror_num)
5826 stripe_index += mirror_num - 1;
5828 int old_stripe_index = stripe_index;
5829 stripe_index = find_live_mirror(fs_info, map,
5831 dev_replace_is_ongoing);
5832 mirror_num = stripe_index - old_stripe_index + 1;
5835 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5836 if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) {
5837 /* push stripe_nr back to the start of the full stripe */
5838 stripe_nr = div64_u64(raid56_full_stripe_start,
5839 stripe_len * nr_data_stripes(map));
5841 /* RAID[56] write or recovery. Return all stripes */
5842 num_stripes = map->num_stripes;
5843 max_errors = nr_parity_stripes(map);
5845 *length = map->stripe_len;
5850 * Mirror #0 or #1 means the original data block.
5851 * Mirror #2 is RAID5 parity block.
5852 * Mirror #3 is RAID6 Q block.
5854 stripe_nr = div_u64_rem(stripe_nr,
5855 nr_data_stripes(map), &stripe_index);
5857 stripe_index = nr_data_stripes(map) +
5860 /* We distribute the parity blocks across stripes */
5861 div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
5863 if (!need_full_stripe(op) && mirror_num <= 1)
5868 * after this, stripe_nr is the number of stripes on this
5869 * device we have to walk to find the data, and stripe_index is
5870 * the number of our device in the stripe array
5872 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5874 mirror_num = stripe_index + 1;
5876 if (stripe_index >= map->num_stripes) {
5878 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
5879 stripe_index, map->num_stripes);
5884 num_alloc_stripes = num_stripes;
5885 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) {
5886 if (op == BTRFS_MAP_WRITE)
5887 num_alloc_stripes <<= 1;
5888 if (op == BTRFS_MAP_GET_READ_MIRRORS)
5889 num_alloc_stripes++;
5890 tgtdev_indexes = num_stripes;
5893 bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
5898 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL)
5899 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
5901 /* build raid_map */
5902 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map &&
5903 (need_full_stripe(op) || mirror_num > 1)) {
5907 bbio->raid_map = (u64 *)((void *)bbio->stripes +
5908 sizeof(struct btrfs_bio_stripe) *
5910 sizeof(int) * tgtdev_indexes);
5912 /* Work out the disk rotation on this stripe-set */
5913 div_u64_rem(stripe_nr, num_stripes, &rot);
5915 /* Fill in the logical address of each stripe */
5916 tmp = stripe_nr * nr_data_stripes(map);
5917 for (i = 0; i < nr_data_stripes(map); i++)
5918 bbio->raid_map[(i+rot) % num_stripes] =
5919 em->start + (tmp + i) * map->stripe_len;
5921 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5922 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5923 bbio->raid_map[(i+rot+1) % num_stripes] =
5928 for (i = 0; i < num_stripes; i++) {
5929 bbio->stripes[i].physical =
5930 map->stripes[stripe_index].physical +
5932 stripe_nr * map->stripe_len;
5933 bbio->stripes[i].dev =
5934 map->stripes[stripe_index].dev;
5938 if (need_full_stripe(op))
5939 max_errors = btrfs_chunk_max_errors(map);
5942 sort_parity_stripes(bbio, num_stripes);
5944 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
5945 need_full_stripe(op)) {
5946 handle_ops_on_dev_replace(op, &bbio, dev_replace, &num_stripes,
5951 bbio->map_type = map->type;
5952 bbio->num_stripes = num_stripes;
5953 bbio->max_errors = max_errors;
5954 bbio->mirror_num = mirror_num;
5957 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5958 * mirror_num == num_stripes + 1 && dev_replace target drive is
5959 * available as a mirror
5961 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5962 WARN_ON(num_stripes > 1);
5963 bbio->stripes[0].dev = dev_replace->tgtdev;
5964 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5965 bbio->mirror_num = map->num_stripes + 1;
5968 if (dev_replace_is_ongoing) {
5969 btrfs_dev_replace_clear_lock_blocking(dev_replace);
5970 btrfs_dev_replace_read_unlock(dev_replace);
5972 free_extent_map(em);
5976 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
5977 u64 logical, u64 *length,
5978 struct btrfs_bio **bbio_ret, int mirror_num)
5980 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret,
5984 /* For Scrub/replace */
5985 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
5986 u64 logical, u64 *length,
5987 struct btrfs_bio **bbio_ret)
5989 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1);
5992 int btrfs_rmap_block(struct btrfs_fs_info *fs_info, u64 chunk_start,
5993 u64 physical, u64 **logical, int *naddrs, int *stripe_len)
5995 struct extent_map *em;
5996 struct map_lookup *map;
6004 em = get_chunk_map(fs_info, chunk_start, 1);
6008 map = em->map_lookup;
6010 rmap_len = map->stripe_len;
6012 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
6013 length = div_u64(length, map->num_stripes / map->sub_stripes);
6014 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
6015 length = div_u64(length, map->num_stripes);
6016 else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6017 length = div_u64(length, nr_data_stripes(map));
6018 rmap_len = map->stripe_len * nr_data_stripes(map);
6021 buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
6022 BUG_ON(!buf); /* -ENOMEM */
6024 for (i = 0; i < map->num_stripes; i++) {
6025 if (map->stripes[i].physical > physical ||
6026 map->stripes[i].physical + length <= physical)
6029 stripe_nr = physical - map->stripes[i].physical;
6030 stripe_nr = div64_u64(stripe_nr, map->stripe_len);
6032 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
6033 stripe_nr = stripe_nr * map->num_stripes + i;
6034 stripe_nr = div_u64(stripe_nr, map->sub_stripes);
6035 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
6036 stripe_nr = stripe_nr * map->num_stripes + i;
6037 } /* else if RAID[56], multiply by nr_data_stripes().
6038 * Alternatively, just use rmap_len below instead of
6039 * map->stripe_len */
6041 bytenr = chunk_start + stripe_nr * rmap_len;
6042 WARN_ON(nr >= map->num_stripes);
6043 for (j = 0; j < nr; j++) {
6044 if (buf[j] == bytenr)
6048 WARN_ON(nr >= map->num_stripes);
6055 *stripe_len = rmap_len;
6057 free_extent_map(em);
6061 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
6063 bio->bi_private = bbio->private;
6064 bio->bi_end_io = bbio->end_io;
6067 btrfs_put_bbio(bbio);
6070 static void btrfs_end_bio(struct bio *bio)
6072 struct btrfs_bio *bbio = bio->bi_private;
6073 int is_orig_bio = 0;
6075 if (bio->bi_status) {
6076 atomic_inc(&bbio->error);
6077 if (bio->bi_status == BLK_STS_IOERR ||
6078 bio->bi_status == BLK_STS_TARGET) {
6079 unsigned int stripe_index =
6080 btrfs_io_bio(bio)->stripe_index;
6081 struct btrfs_device *dev;
6083 BUG_ON(stripe_index >= bbio->num_stripes);
6084 dev = bbio->stripes[stripe_index].dev;
6086 if (bio_op(bio) == REQ_OP_WRITE)
6087 btrfs_dev_stat_inc_and_print(dev,
6088 BTRFS_DEV_STAT_WRITE_ERRS);
6090 btrfs_dev_stat_inc_and_print(dev,
6091 BTRFS_DEV_STAT_READ_ERRS);
6092 if (bio->bi_opf & REQ_PREFLUSH)
6093 btrfs_dev_stat_inc_and_print(dev,
6094 BTRFS_DEV_STAT_FLUSH_ERRS);
6099 if (bio == bbio->orig_bio)
6102 btrfs_bio_counter_dec(bbio->fs_info);
6104 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6107 bio = bbio->orig_bio;
6110 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6111 /* only send an error to the higher layers if it is
6112 * beyond the tolerance of the btrfs bio
6114 if (atomic_read(&bbio->error) > bbio->max_errors) {
6115 bio->bi_status = BLK_STS_IOERR;
6118 * this bio is actually up to date, we didn't
6119 * go over the max number of errors
6121 bio->bi_status = BLK_STS_OK;
6124 btrfs_end_bbio(bbio, bio);
6125 } else if (!is_orig_bio) {
6131 * see run_scheduled_bios for a description of why bios are collected for
6134 * This will add one bio to the pending list for a device and make sure
6135 * the work struct is scheduled.
6137 static noinline void btrfs_schedule_bio(struct btrfs_device *device,
6140 struct btrfs_fs_info *fs_info = device->fs_info;
6141 int should_queue = 1;
6142 struct btrfs_pending_bios *pending_bios;
6144 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state) ||
6150 /* don't bother with additional async steps for reads, right now */
6151 if (bio_op(bio) == REQ_OP_READ) {
6152 btrfsic_submit_bio(bio);
6156 WARN_ON(bio->bi_next);
6157 bio->bi_next = NULL;
6159 spin_lock(&device->io_lock);
6160 if (op_is_sync(bio->bi_opf))
6161 pending_bios = &device->pending_sync_bios;
6163 pending_bios = &device->pending_bios;
6165 if (pending_bios->tail)
6166 pending_bios->tail->bi_next = bio;
6168 pending_bios->tail = bio;
6169 if (!pending_bios->head)
6170 pending_bios->head = bio;
6171 if (device->running_pending)
6174 spin_unlock(&device->io_lock);
6177 btrfs_queue_work(fs_info->submit_workers, &device->work);
6180 static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio,
6181 u64 physical, int dev_nr, int async)
6183 struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
6184 struct btrfs_fs_info *fs_info = bbio->fs_info;
6186 bio->bi_private = bbio;
6187 btrfs_io_bio(bio)->stripe_index = dev_nr;
6188 bio->bi_end_io = btrfs_end_bio;
6189 bio->bi_iter.bi_sector = physical >> 9;
6192 struct rcu_string *name;
6195 name = rcu_dereference(dev->name);
6196 btrfs_debug(fs_info,
6197 "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
6198 bio_op(bio), bio->bi_opf,
6199 (u64)bio->bi_iter.bi_sector,
6200 (u_long)dev->bdev->bd_dev, name->str, dev->devid,
6201 bio->bi_iter.bi_size);
6205 bio_set_dev(bio, dev->bdev);
6207 btrfs_bio_counter_inc_noblocked(fs_info);
6210 btrfs_schedule_bio(dev, bio);
6212 btrfsic_submit_bio(bio);
6215 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6217 atomic_inc(&bbio->error);
6218 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6219 /* Should be the original bio. */
6220 WARN_ON(bio != bbio->orig_bio);
6222 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6223 bio->bi_iter.bi_sector = logical >> 9;
6224 if (atomic_read(&bbio->error) > bbio->max_errors)
6225 bio->bi_status = BLK_STS_IOERR;
6227 bio->bi_status = BLK_STS_OK;
6228 btrfs_end_bbio(bbio, bio);
6232 blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
6233 int mirror_num, int async_submit)
6235 struct btrfs_device *dev;
6236 struct bio *first_bio = bio;
6237 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
6243 struct btrfs_bio *bbio = NULL;
6245 length = bio->bi_iter.bi_size;
6246 map_length = length;
6248 btrfs_bio_counter_inc_blocked(fs_info);
6249 ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical,
6250 &map_length, &bbio, mirror_num, 1);
6252 btrfs_bio_counter_dec(fs_info);
6253 return errno_to_blk_status(ret);
6256 total_devs = bbio->num_stripes;
6257 bbio->orig_bio = first_bio;
6258 bbio->private = first_bio->bi_private;
6259 bbio->end_io = first_bio->bi_end_io;
6260 bbio->fs_info = fs_info;
6261 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6263 if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6264 ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) {
6265 /* In this case, map_length has been set to the length of
6266 a single stripe; not the whole write */
6267 if (bio_op(bio) == REQ_OP_WRITE) {
6268 ret = raid56_parity_write(fs_info, bio, bbio,
6271 ret = raid56_parity_recover(fs_info, bio, bbio,
6272 map_length, mirror_num, 1);
6275 btrfs_bio_counter_dec(fs_info);
6276 return errno_to_blk_status(ret);
6279 if (map_length < length) {
6281 "mapping failed logical %llu bio len %llu len %llu",
6282 logical, length, map_length);
6286 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6287 dev = bbio->stripes[dev_nr].dev;
6288 if (!dev || !dev->bdev ||
6289 (bio_op(first_bio) == REQ_OP_WRITE &&
6290 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) {
6291 bbio_error(bbio, first_bio, logical);
6295 if (dev_nr < total_devs - 1)
6296 bio = btrfs_bio_clone(first_bio);
6300 submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical,
6301 dev_nr, async_submit);
6303 btrfs_bio_counter_dec(fs_info);
6307 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
6310 struct btrfs_device *device;
6311 struct btrfs_fs_devices *cur_devices;
6313 cur_devices = fs_info->fs_devices;
6314 while (cur_devices) {
6316 !memcmp(cur_devices->fsid, fsid, BTRFS_FSID_SIZE)) {
6317 device = find_device(cur_devices, devid, uuid);
6321 cur_devices = cur_devices->seed;
6326 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices,
6327 u64 devid, u8 *dev_uuid)
6329 struct btrfs_device *device;
6331 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6335 list_add(&device->dev_list, &fs_devices->devices);
6336 device->fs_devices = fs_devices;
6337 fs_devices->num_devices++;
6339 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6340 fs_devices->missing_devices++;
6346 * btrfs_alloc_device - allocate struct btrfs_device
6347 * @fs_info: used only for generating a new devid, can be NULL if
6348 * devid is provided (i.e. @devid != NULL).
6349 * @devid: a pointer to devid for this device. If NULL a new devid
6351 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6354 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6355 * on error. Returned struct is not linked onto any lists and must be
6356 * destroyed with btrfs_free_device.
6358 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6362 struct btrfs_device *dev;
6365 if (WARN_ON(!devid && !fs_info))
6366 return ERR_PTR(-EINVAL);
6368 dev = __alloc_device();
6377 ret = find_next_devid(fs_info, &tmp);
6379 btrfs_free_device(dev);
6380 return ERR_PTR(ret);
6386 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6388 generate_random_uuid(dev->uuid);
6390 btrfs_init_work(&dev->work, btrfs_submit_helper,
6391 pending_bios_fn, NULL, NULL);
6396 /* Return -EIO if any error, otherwise return 0. */
6397 static int btrfs_check_chunk_valid(struct btrfs_fs_info *fs_info,
6398 struct extent_buffer *leaf,
6399 struct btrfs_chunk *chunk, u64 logical)
6407 length = btrfs_chunk_length(leaf, chunk);
6408 stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6409 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6410 sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6411 type = btrfs_chunk_type(leaf, chunk);
6414 btrfs_err(fs_info, "invalid chunk num_stripes: %u",
6418 if (!IS_ALIGNED(logical, fs_info->sectorsize)) {
6419 btrfs_err(fs_info, "invalid chunk logical %llu", logical);
6422 if (btrfs_chunk_sector_size(leaf, chunk) != fs_info->sectorsize) {
6423 btrfs_err(fs_info, "invalid chunk sectorsize %u",
6424 btrfs_chunk_sector_size(leaf, chunk));
6427 if (!length || !IS_ALIGNED(length, fs_info->sectorsize)) {
6428 btrfs_err(fs_info, "invalid chunk length %llu", length);
6431 if (!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN) {
6432 btrfs_err(fs_info, "invalid chunk stripe length: %llu",
6436 if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6438 btrfs_err(fs_info, "unrecognized chunk type: %llu",
6439 ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
6440 BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6441 btrfs_chunk_type(leaf, chunk));
6444 if ((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes != 2) ||
6445 (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) ||
6446 (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) ||
6447 (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) ||
6448 (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) ||
6449 ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
6450 num_stripes != 1)) {
6452 "invalid num_stripes:sub_stripes %u:%u for profile %llu",
6453 num_stripes, sub_stripes,
6454 type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
6461 static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info,
6462 u64 devid, u8 *uuid, bool error)
6465 btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing",
6468 btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing",
6472 static int read_one_chunk(struct btrfs_fs_info *fs_info, struct btrfs_key *key,
6473 struct extent_buffer *leaf,
6474 struct btrfs_chunk *chunk)
6476 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
6477 struct map_lookup *map;
6478 struct extent_map *em;
6482 u8 uuid[BTRFS_UUID_SIZE];
6487 logical = key->offset;
6488 length = btrfs_chunk_length(leaf, chunk);
6489 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6491 ret = btrfs_check_chunk_valid(fs_info, leaf, chunk, logical);
6495 read_lock(&map_tree->map_tree.lock);
6496 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
6497 read_unlock(&map_tree->map_tree.lock);
6499 /* already mapped? */
6500 if (em && em->start <= logical && em->start + em->len > logical) {
6501 free_extent_map(em);
6504 free_extent_map(em);
6507 em = alloc_extent_map();
6510 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6512 free_extent_map(em);
6516 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6517 em->map_lookup = map;
6518 em->start = logical;
6521 em->block_start = 0;
6522 em->block_len = em->len;
6524 map->num_stripes = num_stripes;
6525 map->io_width = btrfs_chunk_io_width(leaf, chunk);
6526 map->io_align = btrfs_chunk_io_align(leaf, chunk);
6527 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6528 map->type = btrfs_chunk_type(leaf, chunk);
6529 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6530 for (i = 0; i < num_stripes; i++) {
6531 map->stripes[i].physical =
6532 btrfs_stripe_offset_nr(leaf, chunk, i);
6533 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6534 read_extent_buffer(leaf, uuid, (unsigned long)
6535 btrfs_stripe_dev_uuid_nr(chunk, i),
6537 map->stripes[i].dev = btrfs_find_device(fs_info, devid,
6539 if (!map->stripes[i].dev &&
6540 !btrfs_test_opt(fs_info, DEGRADED)) {
6541 free_extent_map(em);
6542 btrfs_report_missing_device(fs_info, devid, uuid, true);
6545 if (!map->stripes[i].dev) {
6546 map->stripes[i].dev =
6547 add_missing_dev(fs_info->fs_devices, devid,
6549 if (IS_ERR(map->stripes[i].dev)) {
6550 free_extent_map(em);
6552 "failed to init missing dev %llu: %ld",
6553 devid, PTR_ERR(map->stripes[i].dev));
6554 return PTR_ERR(map->stripes[i].dev);
6556 btrfs_report_missing_device(fs_info, devid, uuid, false);
6558 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
6559 &(map->stripes[i].dev->dev_state));
6563 write_lock(&map_tree->map_tree.lock);
6564 ret = add_extent_mapping(&map_tree->map_tree, em, 0);
6565 write_unlock(&map_tree->map_tree.lock);
6566 BUG_ON(ret); /* Tree corruption */
6567 free_extent_map(em);
6572 static void fill_device_from_item(struct extent_buffer *leaf,
6573 struct btrfs_dev_item *dev_item,
6574 struct btrfs_device *device)
6578 device->devid = btrfs_device_id(leaf, dev_item);
6579 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6580 device->total_bytes = device->disk_total_bytes;
6581 device->commit_total_bytes = device->disk_total_bytes;
6582 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6583 device->commit_bytes_used = device->bytes_used;
6584 device->type = btrfs_device_type(leaf, dev_item);
6585 device->io_align = btrfs_device_io_align(leaf, dev_item);
6586 device->io_width = btrfs_device_io_width(leaf, dev_item);
6587 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6588 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6589 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
6591 ptr = btrfs_device_uuid(dev_item);
6592 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6595 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info,
6598 struct btrfs_fs_devices *fs_devices;
6601 lockdep_assert_held(&uuid_mutex);
6604 fs_devices = fs_info->fs_devices->seed;
6605 while (fs_devices) {
6606 if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE))
6609 fs_devices = fs_devices->seed;
6612 fs_devices = find_fsid(fsid);
6614 if (!btrfs_test_opt(fs_info, DEGRADED))
6615 return ERR_PTR(-ENOENT);
6617 fs_devices = alloc_fs_devices(fsid);
6618 if (IS_ERR(fs_devices))
6621 fs_devices->seeding = 1;
6622 fs_devices->opened = 1;
6626 fs_devices = clone_fs_devices(fs_devices);
6627 if (IS_ERR(fs_devices))
6630 ret = open_fs_devices(fs_devices, FMODE_READ, fs_info->bdev_holder);
6632 free_fs_devices(fs_devices);
6633 fs_devices = ERR_PTR(ret);
6637 if (!fs_devices->seeding) {
6638 close_fs_devices(fs_devices);
6639 free_fs_devices(fs_devices);
6640 fs_devices = ERR_PTR(-EINVAL);
6644 fs_devices->seed = fs_info->fs_devices->seed;
6645 fs_info->fs_devices->seed = fs_devices;
6650 static int read_one_dev(struct btrfs_fs_info *fs_info,
6651 struct extent_buffer *leaf,
6652 struct btrfs_dev_item *dev_item)
6654 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6655 struct btrfs_device *device;
6658 u8 fs_uuid[BTRFS_FSID_SIZE];
6659 u8 dev_uuid[BTRFS_UUID_SIZE];
6661 devid = btrfs_device_id(leaf, dev_item);
6662 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6664 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6667 if (memcmp(fs_uuid, fs_info->fsid, BTRFS_FSID_SIZE)) {
6668 fs_devices = open_seed_devices(fs_info, fs_uuid);
6669 if (IS_ERR(fs_devices))
6670 return PTR_ERR(fs_devices);
6673 device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
6675 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6676 btrfs_report_missing_device(fs_info, devid,
6681 device = add_missing_dev(fs_devices, devid, dev_uuid);
6682 if (IS_ERR(device)) {
6684 "failed to add missing dev %llu: %ld",
6685 devid, PTR_ERR(device));
6686 return PTR_ERR(device);
6688 btrfs_report_missing_device(fs_info, devid, dev_uuid, false);
6690 if (!device->bdev) {
6691 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6692 btrfs_report_missing_device(fs_info,
6693 devid, dev_uuid, true);
6696 btrfs_report_missing_device(fs_info, devid,
6700 if (!device->bdev &&
6701 !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
6703 * this happens when a device that was properly setup
6704 * in the device info lists suddenly goes bad.
6705 * device->bdev is NULL, and so we have to set
6706 * device->missing to one here
6708 device->fs_devices->missing_devices++;
6709 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6712 /* Move the device to its own fs_devices */
6713 if (device->fs_devices != fs_devices) {
6714 ASSERT(test_bit(BTRFS_DEV_STATE_MISSING,
6715 &device->dev_state));
6717 list_move(&device->dev_list, &fs_devices->devices);
6718 device->fs_devices->num_devices--;
6719 fs_devices->num_devices++;
6721 device->fs_devices->missing_devices--;
6722 fs_devices->missing_devices++;
6724 device->fs_devices = fs_devices;
6728 if (device->fs_devices != fs_info->fs_devices) {
6729 BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state));
6730 if (device->generation !=
6731 btrfs_device_generation(leaf, dev_item))
6735 fill_device_from_item(leaf, dev_item, device);
6736 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
6737 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
6738 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
6739 device->fs_devices->total_rw_bytes += device->total_bytes;
6740 atomic64_add(device->total_bytes - device->bytes_used,
6741 &fs_info->free_chunk_space);
6747 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
6749 struct btrfs_root *root = fs_info->tree_root;
6750 struct btrfs_super_block *super_copy = fs_info->super_copy;
6751 struct extent_buffer *sb;
6752 struct btrfs_disk_key *disk_key;
6753 struct btrfs_chunk *chunk;
6755 unsigned long sb_array_offset;
6762 struct btrfs_key key;
6764 ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize);
6766 * This will create extent buffer of nodesize, superblock size is
6767 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6768 * overallocate but we can keep it as-is, only the first page is used.
6770 sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET);
6773 set_extent_buffer_uptodate(sb);
6774 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6776 * The sb extent buffer is artificial and just used to read the system array.
6777 * set_extent_buffer_uptodate() call does not properly mark all it's
6778 * pages up-to-date when the page is larger: extent does not cover the
6779 * whole page and consequently check_page_uptodate does not find all
6780 * the page's extents up-to-date (the hole beyond sb),
6781 * write_extent_buffer then triggers a WARN_ON.
6783 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6784 * but sb spans only this function. Add an explicit SetPageUptodate call
6785 * to silence the warning eg. on PowerPC 64.
6787 if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
6788 SetPageUptodate(sb->pages[0]);
6790 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6791 array_size = btrfs_super_sys_array_size(super_copy);
6793 array_ptr = super_copy->sys_chunk_array;
6794 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6797 while (cur_offset < array_size) {
6798 disk_key = (struct btrfs_disk_key *)array_ptr;
6799 len = sizeof(*disk_key);
6800 if (cur_offset + len > array_size)
6801 goto out_short_read;
6803 btrfs_disk_key_to_cpu(&key, disk_key);
6806 sb_array_offset += len;
6809 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6810 chunk = (struct btrfs_chunk *)sb_array_offset;
6812 * At least one btrfs_chunk with one stripe must be
6813 * present, exact stripe count check comes afterwards
6815 len = btrfs_chunk_item_size(1);
6816 if (cur_offset + len > array_size)
6817 goto out_short_read;
6819 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6822 "invalid number of stripes %u in sys_array at offset %u",
6823 num_stripes, cur_offset);
6828 type = btrfs_chunk_type(sb, chunk);
6829 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
6831 "invalid chunk type %llu in sys_array at offset %u",
6837 len = btrfs_chunk_item_size(num_stripes);
6838 if (cur_offset + len > array_size)
6839 goto out_short_read;
6841 ret = read_one_chunk(fs_info, &key, sb, chunk);
6846 "unexpected item type %u in sys_array at offset %u",
6847 (u32)key.type, cur_offset);
6852 sb_array_offset += len;
6855 clear_extent_buffer_uptodate(sb);
6856 free_extent_buffer_stale(sb);
6860 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
6862 clear_extent_buffer_uptodate(sb);
6863 free_extent_buffer_stale(sb);
6868 * Check if all chunks in the fs are OK for read-write degraded mount
6870 * If the @failing_dev is specified, it's accounted as missing.
6872 * Return true if all chunks meet the minimal RW mount requirements.
6873 * Return false if any chunk doesn't meet the minimal RW mount requirements.
6875 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
6876 struct btrfs_device *failing_dev)
6878 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
6879 struct extent_map *em;
6883 read_lock(&map_tree->map_tree.lock);
6884 em = lookup_extent_mapping(&map_tree->map_tree, 0, (u64)-1);
6885 read_unlock(&map_tree->map_tree.lock);
6886 /* No chunk at all? Return false anyway */
6892 struct map_lookup *map;
6897 map = em->map_lookup;
6899 btrfs_get_num_tolerated_disk_barrier_failures(
6901 for (i = 0; i < map->num_stripes; i++) {
6902 struct btrfs_device *dev = map->stripes[i].dev;
6904 if (!dev || !dev->bdev ||
6905 test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) ||
6906 dev->last_flush_error)
6908 else if (failing_dev && failing_dev == dev)
6911 if (missing > max_tolerated) {
6914 "chunk %llu missing %d devices, max tolerance is %d for writeable mount",
6915 em->start, missing, max_tolerated);
6916 free_extent_map(em);
6920 next_start = extent_map_end(em);
6921 free_extent_map(em);
6923 read_lock(&map_tree->map_tree.lock);
6924 em = lookup_extent_mapping(&map_tree->map_tree, next_start,
6925 (u64)(-1) - next_start);
6926 read_unlock(&map_tree->map_tree.lock);
6932 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
6934 struct btrfs_root *root = fs_info->chunk_root;
6935 struct btrfs_path *path;
6936 struct extent_buffer *leaf;
6937 struct btrfs_key key;
6938 struct btrfs_key found_key;
6943 path = btrfs_alloc_path();
6948 * uuid_mutex is needed only if we are mounting a sprout FS
6949 * otherwise we don't need it.
6951 mutex_lock(&uuid_mutex);
6952 mutex_lock(&fs_info->chunk_mutex);
6955 * Read all device items, and then all the chunk items. All
6956 * device items are found before any chunk item (their object id
6957 * is smaller than the lowest possible object id for a chunk
6958 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6960 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6963 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6967 leaf = path->nodes[0];
6968 slot = path->slots[0];
6969 if (slot >= btrfs_header_nritems(leaf)) {
6970 ret = btrfs_next_leaf(root, path);
6977 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6978 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6979 struct btrfs_dev_item *dev_item;
6980 dev_item = btrfs_item_ptr(leaf, slot,
6981 struct btrfs_dev_item);
6982 ret = read_one_dev(fs_info, leaf, dev_item);
6986 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6987 struct btrfs_chunk *chunk;
6988 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6989 ret = read_one_chunk(fs_info, &found_key, leaf, chunk);
6997 * After loading chunk tree, we've got all device information,
6998 * do another round of validation checks.
7000 if (total_dev != fs_info->fs_devices->total_devices) {
7002 "super_num_devices %llu mismatch with num_devices %llu found here",
7003 btrfs_super_num_devices(fs_info->super_copy),
7008 if (btrfs_super_total_bytes(fs_info->super_copy) <
7009 fs_info->fs_devices->total_rw_bytes) {
7011 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
7012 btrfs_super_total_bytes(fs_info->super_copy),
7013 fs_info->fs_devices->total_rw_bytes);
7019 mutex_unlock(&fs_info->chunk_mutex);
7020 mutex_unlock(&uuid_mutex);
7022 btrfs_free_path(path);
7026 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
7028 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7029 struct btrfs_device *device;
7031 while (fs_devices) {
7032 mutex_lock(&fs_devices->device_list_mutex);
7033 list_for_each_entry(device, &fs_devices->devices, dev_list)
7034 device->fs_info = fs_info;
7035 mutex_unlock(&fs_devices->device_list_mutex);
7037 fs_devices = fs_devices->seed;
7041 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
7045 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7046 btrfs_dev_stat_reset(dev, i);
7049 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
7051 struct btrfs_key key;
7052 struct btrfs_key found_key;
7053 struct btrfs_root *dev_root = fs_info->dev_root;
7054 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7055 struct extent_buffer *eb;
7058 struct btrfs_device *device;
7059 struct btrfs_path *path = NULL;
7062 path = btrfs_alloc_path();
7068 mutex_lock(&fs_devices->device_list_mutex);
7069 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7071 struct btrfs_dev_stats_item *ptr;
7073 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7074 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7075 key.offset = device->devid;
7076 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
7078 __btrfs_reset_dev_stats(device);
7079 device->dev_stats_valid = 1;
7080 btrfs_release_path(path);
7083 slot = path->slots[0];
7084 eb = path->nodes[0];
7085 btrfs_item_key_to_cpu(eb, &found_key, slot);
7086 item_size = btrfs_item_size_nr(eb, slot);
7088 ptr = btrfs_item_ptr(eb, slot,
7089 struct btrfs_dev_stats_item);
7091 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7092 if (item_size >= (1 + i) * sizeof(__le64))
7093 btrfs_dev_stat_set(device, i,
7094 btrfs_dev_stats_value(eb, ptr, i));
7096 btrfs_dev_stat_reset(device, i);
7099 device->dev_stats_valid = 1;
7100 btrfs_dev_stat_print_on_load(device);
7101 btrfs_release_path(path);
7103 mutex_unlock(&fs_devices->device_list_mutex);
7106 btrfs_free_path(path);
7107 return ret < 0 ? ret : 0;
7110 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
7111 struct btrfs_fs_info *fs_info,
7112 struct btrfs_device *device)
7114 struct btrfs_root *dev_root = fs_info->dev_root;
7115 struct btrfs_path *path;
7116 struct btrfs_key key;
7117 struct extent_buffer *eb;
7118 struct btrfs_dev_stats_item *ptr;
7122 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7123 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7124 key.offset = device->devid;
7126 path = btrfs_alloc_path();
7129 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
7131 btrfs_warn_in_rcu(fs_info,
7132 "error %d while searching for dev_stats item for device %s",
7133 ret, rcu_str_deref(device->name));
7138 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
7139 /* need to delete old one and insert a new one */
7140 ret = btrfs_del_item(trans, dev_root, path);
7142 btrfs_warn_in_rcu(fs_info,
7143 "delete too small dev_stats item for device %s failed %d",
7144 rcu_str_deref(device->name), ret);
7151 /* need to insert a new item */
7152 btrfs_release_path(path);
7153 ret = btrfs_insert_empty_item(trans, dev_root, path,
7154 &key, sizeof(*ptr));
7156 btrfs_warn_in_rcu(fs_info,
7157 "insert dev_stats item for device %s failed %d",
7158 rcu_str_deref(device->name), ret);
7163 eb = path->nodes[0];
7164 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7165 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7166 btrfs_set_dev_stats_value(eb, ptr, i,
7167 btrfs_dev_stat_read(device, i));
7168 btrfs_mark_buffer_dirty(eb);
7171 btrfs_free_path(path);
7176 * called from commit_transaction. Writes all changed device stats to disk.
7178 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
7179 struct btrfs_fs_info *fs_info)
7181 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7182 struct btrfs_device *device;
7186 mutex_lock(&fs_devices->device_list_mutex);
7187 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7188 stats_cnt = atomic_read(&device->dev_stats_ccnt);
7189 if (!device->dev_stats_valid || stats_cnt == 0)
7194 * There is a LOAD-LOAD control dependency between the value of
7195 * dev_stats_ccnt and updating the on-disk values which requires
7196 * reading the in-memory counters. Such control dependencies
7197 * require explicit read memory barriers.
7199 * This memory barriers pairs with smp_mb__before_atomic in
7200 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full
7201 * barrier implied by atomic_xchg in
7202 * btrfs_dev_stats_read_and_reset
7206 ret = update_dev_stat_item(trans, fs_info, device);
7208 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7210 mutex_unlock(&fs_devices->device_list_mutex);
7215 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7217 btrfs_dev_stat_inc(dev, index);
7218 btrfs_dev_stat_print_on_error(dev);
7221 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
7223 if (!dev->dev_stats_valid)
7225 btrfs_err_rl_in_rcu(dev->fs_info,
7226 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7227 rcu_str_deref(dev->name),
7228 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7229 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7230 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7231 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7232 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7235 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7239 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7240 if (btrfs_dev_stat_read(dev, i) != 0)
7242 if (i == BTRFS_DEV_STAT_VALUES_MAX)
7243 return; /* all values == 0, suppress message */
7245 btrfs_info_in_rcu(dev->fs_info,
7246 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7247 rcu_str_deref(dev->name),
7248 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7249 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7250 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7251 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7252 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7255 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
7256 struct btrfs_ioctl_get_dev_stats *stats)
7258 struct btrfs_device *dev;
7259 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7262 mutex_lock(&fs_devices->device_list_mutex);
7263 dev = btrfs_find_device(fs_info, stats->devid, NULL, NULL);
7264 mutex_unlock(&fs_devices->device_list_mutex);
7267 btrfs_warn(fs_info, "get dev_stats failed, device not found");
7269 } else if (!dev->dev_stats_valid) {
7270 btrfs_warn(fs_info, "get dev_stats failed, not yet valid");
7272 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7273 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7274 if (stats->nr_items > i)
7276 btrfs_dev_stat_read_and_reset(dev, i);
7278 btrfs_dev_stat_reset(dev, i);
7281 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7282 if (stats->nr_items > i)
7283 stats->values[i] = btrfs_dev_stat_read(dev, i);
7285 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7286 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7290 void btrfs_scratch_superblocks(struct block_device *bdev, const char *device_path)
7292 struct buffer_head *bh;
7293 struct btrfs_super_block *disk_super;
7299 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX;
7302 if (btrfs_read_dev_one_super(bdev, copy_num, &bh))
7305 disk_super = (struct btrfs_super_block *)bh->b_data;
7307 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
7308 set_buffer_dirty(bh);
7309 sync_dirty_buffer(bh);
7313 /* Notify udev that device has changed */
7314 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
7316 /* Update ctime/mtime for device path for libblkid */
7317 update_dev_time(device_path);
7321 * Update the size of all devices, which is used for writing out the
7324 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
7326 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7327 struct btrfs_device *curr, *next;
7329 if (list_empty(&fs_devices->resized_devices))
7332 mutex_lock(&fs_devices->device_list_mutex);
7333 mutex_lock(&fs_info->chunk_mutex);
7334 list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
7336 list_del_init(&curr->resized_list);
7337 curr->commit_total_bytes = curr->disk_total_bytes;
7339 mutex_unlock(&fs_info->chunk_mutex);
7340 mutex_unlock(&fs_devices->device_list_mutex);
7343 /* Must be invoked during the transaction commit */
7344 void btrfs_update_commit_device_bytes_used(struct btrfs_transaction *trans)
7346 struct btrfs_fs_info *fs_info = trans->fs_info;
7347 struct extent_map *em;
7348 struct map_lookup *map;
7349 struct btrfs_device *dev;
7352 if (list_empty(&trans->pending_chunks))
7355 /* In order to kick the device replace finish process */
7356 mutex_lock(&fs_info->chunk_mutex);
7357 list_for_each_entry(em, &trans->pending_chunks, list) {
7358 map = em->map_lookup;
7360 for (i = 0; i < map->num_stripes; i++) {
7361 dev = map->stripes[i].dev;
7362 dev->commit_bytes_used = dev->bytes_used;
7365 mutex_unlock(&fs_info->chunk_mutex);
7368 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
7370 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7371 while (fs_devices) {
7372 fs_devices->fs_info = fs_info;
7373 fs_devices = fs_devices->seed;
7377 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
7379 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7380 while (fs_devices) {
7381 fs_devices->fs_info = NULL;
7382 fs_devices = fs_devices->seed;