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,
44 [BTRFS_RAID_RAID1] = {
49 .tolerated_failures = 1,
58 .tolerated_failures = 0,
62 [BTRFS_RAID_RAID0] = {
67 .tolerated_failures = 0,
71 [BTRFS_RAID_SINGLE] = {
76 .tolerated_failures = 0,
80 [BTRFS_RAID_RAID5] = {
85 .tolerated_failures = 1,
89 [BTRFS_RAID_RAID6] = {
94 .tolerated_failures = 2,
100 const u64 btrfs_raid_group[BTRFS_NR_RAID_TYPES] = {
101 [BTRFS_RAID_RAID10] = BTRFS_BLOCK_GROUP_RAID10,
102 [BTRFS_RAID_RAID1] = BTRFS_BLOCK_GROUP_RAID1,
103 [BTRFS_RAID_DUP] = BTRFS_BLOCK_GROUP_DUP,
104 [BTRFS_RAID_RAID0] = BTRFS_BLOCK_GROUP_RAID0,
105 [BTRFS_RAID_SINGLE] = 0,
106 [BTRFS_RAID_RAID5] = BTRFS_BLOCK_GROUP_RAID5,
107 [BTRFS_RAID_RAID6] = BTRFS_BLOCK_GROUP_RAID6,
111 * Table to convert BTRFS_RAID_* to the error code if minimum number of devices
112 * condition is not met. Zero means there's no corresponding
113 * BTRFS_ERROR_DEV_*_NOT_MET value.
115 const int btrfs_raid_mindev_error[BTRFS_NR_RAID_TYPES] = {
116 [BTRFS_RAID_RAID10] = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET,
117 [BTRFS_RAID_RAID1] = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET,
118 [BTRFS_RAID_DUP] = 0,
119 [BTRFS_RAID_RAID0] = 0,
120 [BTRFS_RAID_SINGLE] = 0,
121 [BTRFS_RAID_RAID5] = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET,
122 [BTRFS_RAID_RAID6] = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET,
125 static int init_first_rw_device(struct btrfs_trans_handle *trans,
126 struct btrfs_fs_info *fs_info);
127 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info);
128 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
129 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
130 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
131 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
132 enum btrfs_map_op op,
133 u64 logical, u64 *length,
134 struct btrfs_bio **bbio_ret,
135 int mirror_num, int need_raid_map);
141 * There are several mutexes that protect manipulation of devices and low-level
142 * structures like chunks but not block groups, extents or files
144 * uuid_mutex (global lock)
145 * ------------------------
146 * protects the fs_uuids list that tracks all per-fs fs_devices, resulting from
147 * the SCAN_DEV ioctl registration or from mount either implicitly (the first
148 * device) or requested by the device= mount option
150 * the mutex can be very coarse and can cover long-running operations
152 * protects: updates to fs_devices counters like missing devices, rw devices,
153 * seeding, structure cloning, openning/closing devices at mount/umount time
155 * global::fs_devs - add, remove, updates to the global list
157 * does not protect: manipulation of the fs_devices::devices list!
159 * btrfs_device::name - renames (write side), read is RCU
161 * fs_devices::device_list_mutex (per-fs, with RCU)
162 * ------------------------------------------------
163 * protects updates to fs_devices::devices, ie. adding and deleting
165 * simple list traversal with read-only actions can be done with RCU protection
167 * may be used to exclude some operations from running concurrently without any
168 * modifications to the list (see write_all_supers)
172 * protects balance structures (status, state) and context accessed from
173 * several places (internally, ioctl)
177 * protects chunks, adding or removing during allocation, trim or when a new
178 * device is added/removed
182 * a big lock that is held by the cleaner thread and prevents running subvolume
183 * cleaning together with relocation or delayed iputs
196 * Exclusive operations, BTRFS_FS_EXCL_OP
197 * ======================================
199 * Maintains the exclusivity of the following operations that apply to the
200 * whole filesystem and cannot run in parallel.
205 * - Device replace (*)
208 * The device operations (as above) can be in one of the following states:
214 * Only device operations marked with (*) can go into the Paused state for the
217 * - ioctl (only Balance can be Paused through ioctl)
218 * - filesystem remounted as read-only
219 * - filesystem unmounted and mounted as read-only
220 * - system power-cycle and filesystem mounted as read-only
221 * - filesystem or device errors leading to forced read-only
223 * BTRFS_FS_EXCL_OP flag is set and cleared using atomic operations.
224 * During the course of Paused state, the BTRFS_FS_EXCL_OP remains set.
225 * A device operation in Paused or Running state can be canceled or resumed
226 * either by ioctl (Balance only) or when remounted as read-write.
227 * BTRFS_FS_EXCL_OP flag is cleared when the device operation is canceled or
231 DEFINE_MUTEX(uuid_mutex);
232 static LIST_HEAD(fs_uuids);
233 struct list_head *btrfs_get_fs_uuids(void)
239 * alloc_fs_devices - allocate struct btrfs_fs_devices
240 * @fsid: if not NULL, copy the uuid to fs_devices::fsid
242 * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR().
243 * The returned struct is not linked onto any lists and can be destroyed with
244 * kfree() right away.
246 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
248 struct btrfs_fs_devices *fs_devs;
250 fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL);
252 return ERR_PTR(-ENOMEM);
254 mutex_init(&fs_devs->device_list_mutex);
256 INIT_LIST_HEAD(&fs_devs->devices);
257 INIT_LIST_HEAD(&fs_devs->resized_devices);
258 INIT_LIST_HEAD(&fs_devs->alloc_list);
259 INIT_LIST_HEAD(&fs_devs->fs_list);
261 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
266 void btrfs_free_device(struct btrfs_device *device)
268 rcu_string_free(device->name);
269 bio_put(device->flush_bio);
273 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
275 struct btrfs_device *device;
276 WARN_ON(fs_devices->opened);
277 while (!list_empty(&fs_devices->devices)) {
278 device = list_entry(fs_devices->devices.next,
279 struct btrfs_device, dev_list);
280 list_del(&device->dev_list);
281 btrfs_free_device(device);
286 static void btrfs_kobject_uevent(struct block_device *bdev,
287 enum kobject_action action)
291 ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
293 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
295 kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
296 &disk_to_dev(bdev->bd_disk)->kobj);
299 void __exit btrfs_cleanup_fs_uuids(void)
301 struct btrfs_fs_devices *fs_devices;
303 while (!list_empty(&fs_uuids)) {
304 fs_devices = list_entry(fs_uuids.next,
305 struct btrfs_fs_devices, fs_list);
306 list_del(&fs_devices->fs_list);
307 free_fs_devices(fs_devices);
312 * Returns a pointer to a new btrfs_device on success; ERR_PTR() on error.
313 * Returned struct is not linked onto any lists and must be destroyed using
316 static struct btrfs_device *__alloc_device(void)
318 struct btrfs_device *dev;
320 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
322 return ERR_PTR(-ENOMEM);
325 * Preallocate a bio that's always going to be used for flushing device
326 * barriers and matches the device lifespan
328 dev->flush_bio = bio_alloc_bioset(GFP_KERNEL, 0, NULL);
329 if (!dev->flush_bio) {
331 return ERR_PTR(-ENOMEM);
334 INIT_LIST_HEAD(&dev->dev_list);
335 INIT_LIST_HEAD(&dev->dev_alloc_list);
336 INIT_LIST_HEAD(&dev->resized_list);
338 spin_lock_init(&dev->io_lock);
340 atomic_set(&dev->reada_in_flight, 0);
341 atomic_set(&dev->dev_stats_ccnt, 0);
342 btrfs_device_data_ordered_init(dev);
343 INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
344 INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
350 * Find a device specified by @devid or @uuid in the list of @fs_devices, or
353 * If devid and uuid are both specified, the match must be exact, otherwise
354 * only devid is used.
356 static struct btrfs_device *find_device(struct btrfs_fs_devices *fs_devices,
357 u64 devid, const u8 *uuid)
359 struct btrfs_device *dev;
361 list_for_each_entry(dev, &fs_devices->devices, dev_list) {
362 if (dev->devid == devid &&
363 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
370 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
372 struct btrfs_fs_devices *fs_devices;
374 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
375 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
382 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
383 int flush, struct block_device **bdev,
384 struct buffer_head **bh)
388 *bdev = blkdev_get_by_path(device_path, flags, holder);
391 ret = PTR_ERR(*bdev);
396 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
397 ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE);
399 blkdev_put(*bdev, flags);
402 invalidate_bdev(*bdev);
403 *bh = btrfs_read_dev_super(*bdev);
406 blkdev_put(*bdev, flags);
418 static void requeue_list(struct btrfs_pending_bios *pending_bios,
419 struct bio *head, struct bio *tail)
422 struct bio *old_head;
424 old_head = pending_bios->head;
425 pending_bios->head = head;
426 if (pending_bios->tail)
427 tail->bi_next = old_head;
429 pending_bios->tail = tail;
433 * we try to collect pending bios for a device so we don't get a large
434 * number of procs sending bios down to the same device. This greatly
435 * improves the schedulers ability to collect and merge the bios.
437 * But, it also turns into a long list of bios to process and that is sure
438 * to eventually make the worker thread block. The solution here is to
439 * make some progress and then put this work struct back at the end of
440 * the list if the block device is congested. This way, multiple devices
441 * can make progress from a single worker thread.
443 static noinline void run_scheduled_bios(struct btrfs_device *device)
445 struct btrfs_fs_info *fs_info = device->fs_info;
447 struct backing_dev_info *bdi;
448 struct btrfs_pending_bios *pending_bios;
452 unsigned long num_run;
453 unsigned long batch_run = 0;
454 unsigned long last_waited = 0;
456 int sync_pending = 0;
457 struct blk_plug plug;
460 * this function runs all the bios we've collected for
461 * a particular device. We don't want to wander off to
462 * another device without first sending all of these down.
463 * So, setup a plug here and finish it off before we return
465 blk_start_plug(&plug);
467 bdi = device->bdev->bd_bdi;
470 spin_lock(&device->io_lock);
475 /* take all the bios off the list at once and process them
476 * later on (without the lock held). But, remember the
477 * tail and other pointers so the bios can be properly reinserted
478 * into the list if we hit congestion
480 if (!force_reg && device->pending_sync_bios.head) {
481 pending_bios = &device->pending_sync_bios;
484 pending_bios = &device->pending_bios;
488 pending = pending_bios->head;
489 tail = pending_bios->tail;
490 WARN_ON(pending && !tail);
493 * if pending was null this time around, no bios need processing
494 * at all and we can stop. Otherwise it'll loop back up again
495 * and do an additional check so no bios are missed.
497 * device->running_pending is used to synchronize with the
500 if (device->pending_sync_bios.head == NULL &&
501 device->pending_bios.head == NULL) {
503 device->running_pending = 0;
506 device->running_pending = 1;
509 pending_bios->head = NULL;
510 pending_bios->tail = NULL;
512 spin_unlock(&device->io_lock);
517 /* we want to work on both lists, but do more bios on the
518 * sync list than the regular list
521 pending_bios != &device->pending_sync_bios &&
522 device->pending_sync_bios.head) ||
523 (num_run > 64 && pending_bios == &device->pending_sync_bios &&
524 device->pending_bios.head)) {
525 spin_lock(&device->io_lock);
526 requeue_list(pending_bios, pending, tail);
531 pending = pending->bi_next;
534 BUG_ON(atomic_read(&cur->__bi_cnt) == 0);
537 * if we're doing the sync list, record that our
538 * plug has some sync requests on it
540 * If we're doing the regular list and there are
541 * sync requests sitting around, unplug before
544 if (pending_bios == &device->pending_sync_bios) {
546 } else if (sync_pending) {
547 blk_finish_plug(&plug);
548 blk_start_plug(&plug);
552 btrfsic_submit_bio(cur);
559 * we made progress, there is more work to do and the bdi
560 * is now congested. Back off and let other work structs
563 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
564 fs_info->fs_devices->open_devices > 1) {
565 struct io_context *ioc;
567 ioc = current->io_context;
570 * the main goal here is that we don't want to
571 * block if we're going to be able to submit
572 * more requests without blocking.
574 * This code does two great things, it pokes into
575 * the elevator code from a filesystem _and_
576 * it makes assumptions about how batching works.
578 if (ioc && ioc->nr_batch_requests > 0 &&
579 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
581 ioc->last_waited == last_waited)) {
583 * we want to go through our batch of
584 * requests and stop. So, we copy out
585 * the ioc->last_waited time and test
586 * against it before looping
588 last_waited = ioc->last_waited;
592 spin_lock(&device->io_lock);
593 requeue_list(pending_bios, pending, tail);
594 device->running_pending = 1;
596 spin_unlock(&device->io_lock);
597 btrfs_queue_work(fs_info->submit_workers,
607 spin_lock(&device->io_lock);
608 if (device->pending_bios.head || device->pending_sync_bios.head)
610 spin_unlock(&device->io_lock);
613 blk_finish_plug(&plug);
616 static void pending_bios_fn(struct btrfs_work *work)
618 struct btrfs_device *device;
620 device = container_of(work, struct btrfs_device, work);
621 run_scheduled_bios(device);
625 * Search and remove all stale (devices which are not mounted) devices.
626 * When both inputs are NULL, it will search and release all stale devices.
627 * path: Optional. When provided will it release all unmounted devices
628 * matching this path only.
629 * skip_dev: Optional. Will skip this device when searching for the stale
632 static void btrfs_free_stale_devices(const char *path,
633 struct btrfs_device *skip_dev)
635 struct btrfs_fs_devices *fs_devs, *tmp_fs_devs;
636 struct btrfs_device *dev, *tmp_dev;
638 list_for_each_entry_safe(fs_devs, tmp_fs_devs, &fs_uuids, fs_list) {
643 list_for_each_entry_safe(dev, tmp_dev,
644 &fs_devs->devices, dev_list) {
647 if (skip_dev && skip_dev == dev)
649 if (path && !dev->name)
654 not_found = strcmp(rcu_str_deref(dev->name),
660 /* delete the stale device */
661 if (fs_devs->num_devices == 1) {
662 btrfs_sysfs_remove_fsid(fs_devs);
663 list_del(&fs_devs->fs_list);
664 free_fs_devices(fs_devs);
667 fs_devs->num_devices--;
668 list_del(&dev->dev_list);
669 btrfs_free_device(dev);
675 static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices,
676 struct btrfs_device *device, fmode_t flags,
679 struct request_queue *q;
680 struct block_device *bdev;
681 struct buffer_head *bh;
682 struct btrfs_super_block *disk_super;
691 ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
696 disk_super = (struct btrfs_super_block *)bh->b_data;
697 devid = btrfs_stack_device_id(&disk_super->dev_item);
698 if (devid != device->devid)
701 if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE))
704 device->generation = btrfs_super_generation(disk_super);
706 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
707 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
708 fs_devices->seeding = 1;
710 if (bdev_read_only(bdev))
711 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
713 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
716 q = bdev_get_queue(bdev);
717 if (!blk_queue_nonrot(q))
718 fs_devices->rotating = 1;
721 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
722 device->mode = flags;
724 fs_devices->open_devices++;
725 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
726 device->devid != BTRFS_DEV_REPLACE_DEVID) {
727 fs_devices->rw_devices++;
728 list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list);
736 blkdev_put(bdev, flags);
742 * Add new device to list of registered devices
745 * device pointer which was just added or updated when successful
746 * error pointer when failed
748 static noinline struct btrfs_device *device_list_add(const char *path,
749 struct btrfs_super_block *disk_super)
751 struct btrfs_device *device;
752 struct btrfs_fs_devices *fs_devices;
753 struct rcu_string *name;
754 u64 found_transid = btrfs_super_generation(disk_super);
755 u64 devid = btrfs_stack_device_id(&disk_super->dev_item);
757 fs_devices = find_fsid(disk_super->fsid);
759 fs_devices = alloc_fs_devices(disk_super->fsid);
760 if (IS_ERR(fs_devices))
761 return ERR_CAST(fs_devices);
763 list_add(&fs_devices->fs_list, &fs_uuids);
767 device = find_device(fs_devices, devid,
768 disk_super->dev_item.uuid);
772 if (fs_devices->opened)
773 return ERR_PTR(-EBUSY);
775 device = btrfs_alloc_device(NULL, &devid,
776 disk_super->dev_item.uuid);
777 if (IS_ERR(device)) {
778 /* we can safely leave the fs_devices entry around */
782 name = rcu_string_strdup(path, GFP_NOFS);
784 btrfs_free_device(device);
785 return ERR_PTR(-ENOMEM);
787 rcu_assign_pointer(device->name, name);
789 mutex_lock(&fs_devices->device_list_mutex);
790 list_add_rcu(&device->dev_list, &fs_devices->devices);
791 fs_devices->num_devices++;
792 mutex_unlock(&fs_devices->device_list_mutex);
794 device->fs_devices = fs_devices;
795 btrfs_free_stale_devices(path, device);
797 if (disk_super->label[0])
798 pr_info("BTRFS: device label %s devid %llu transid %llu %s\n",
799 disk_super->label, devid, found_transid, path);
801 pr_info("BTRFS: device fsid %pU devid %llu transid %llu %s\n",
802 disk_super->fsid, devid, found_transid, path);
804 } else if (!device->name || strcmp(device->name->str, path)) {
806 * When FS is already mounted.
807 * 1. If you are here and if the device->name is NULL that
808 * means this device was missing at time of FS mount.
809 * 2. If you are here and if the device->name is different
810 * from 'path' that means either
811 * a. The same device disappeared and reappeared with
813 * b. The missing-disk-which-was-replaced, has
816 * We must allow 1 and 2a above. But 2b would be a spurious
819 * Further in case of 1 and 2a above, the disk at 'path'
820 * would have missed some transaction when it was away and
821 * in case of 2a the stale bdev has to be updated as well.
822 * 2b must not be allowed at all time.
826 * For now, we do allow update to btrfs_fs_device through the
827 * btrfs dev scan cli after FS has been mounted. We're still
828 * tracking a problem where systems fail mount by subvolume id
829 * when we reject replacement on a mounted FS.
831 if (!fs_devices->opened && found_transid < device->generation) {
833 * That is if the FS is _not_ mounted and if you
834 * are here, that means there is more than one
835 * disk with same uuid and devid.We keep the one
836 * with larger generation number or the last-in if
837 * generation are equal.
839 return ERR_PTR(-EEXIST);
842 name = rcu_string_strdup(path, GFP_NOFS);
844 return ERR_PTR(-ENOMEM);
845 rcu_string_free(device->name);
846 rcu_assign_pointer(device->name, name);
847 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
848 fs_devices->missing_devices--;
849 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
854 * Unmount does not free the btrfs_device struct but would zero
855 * generation along with most of the other members. So just update
856 * it back. We need it to pick the disk with largest generation
859 if (!fs_devices->opened)
860 device->generation = found_transid;
862 fs_devices->total_devices = btrfs_super_num_devices(disk_super);
867 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
869 struct btrfs_fs_devices *fs_devices;
870 struct btrfs_device *device;
871 struct btrfs_device *orig_dev;
873 fs_devices = alloc_fs_devices(orig->fsid);
874 if (IS_ERR(fs_devices))
877 mutex_lock(&orig->device_list_mutex);
878 fs_devices->total_devices = orig->total_devices;
880 /* We have held the volume lock, it is safe to get the devices. */
881 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
882 struct rcu_string *name;
884 device = btrfs_alloc_device(NULL, &orig_dev->devid,
890 * This is ok to do without rcu read locked because we hold the
891 * uuid mutex so nothing we touch in here is going to disappear.
893 if (orig_dev->name) {
894 name = rcu_string_strdup(orig_dev->name->str,
897 btrfs_free_device(device);
900 rcu_assign_pointer(device->name, name);
903 list_add(&device->dev_list, &fs_devices->devices);
904 device->fs_devices = fs_devices;
905 fs_devices->num_devices++;
907 mutex_unlock(&orig->device_list_mutex);
910 mutex_unlock(&orig->device_list_mutex);
911 free_fs_devices(fs_devices);
912 return ERR_PTR(-ENOMEM);
916 * After we have read the system tree and know devids belonging to
917 * this filesystem, remove the device which does not belong there.
919 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices, int step)
921 struct btrfs_device *device, *next;
922 struct btrfs_device *latest_dev = NULL;
924 mutex_lock(&uuid_mutex);
926 /* This is the initialized path, it is safe to release the devices. */
927 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
928 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
929 &device->dev_state)) {
930 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
931 &device->dev_state) &&
933 device->generation > latest_dev->generation)) {
939 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
941 * In the first step, keep the device which has
942 * the correct fsid and the devid that is used
943 * for the dev_replace procedure.
944 * In the second step, the dev_replace state is
945 * read from the device tree and it is known
946 * whether the procedure is really active or
947 * not, which means whether this device is
948 * used or whether it should be removed.
950 if (step == 0 || test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
951 &device->dev_state)) {
956 blkdev_put(device->bdev, device->mode);
958 fs_devices->open_devices--;
960 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
961 list_del_init(&device->dev_alloc_list);
962 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
963 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
965 fs_devices->rw_devices--;
967 list_del_init(&device->dev_list);
968 fs_devices->num_devices--;
969 btrfs_free_device(device);
972 if (fs_devices->seed) {
973 fs_devices = fs_devices->seed;
977 fs_devices->latest_bdev = latest_dev->bdev;
979 mutex_unlock(&uuid_mutex);
982 static void free_device_rcu(struct rcu_head *head)
984 struct btrfs_device *device;
986 device = container_of(head, struct btrfs_device, rcu);
987 btrfs_free_device(device);
990 static void btrfs_close_bdev(struct btrfs_device *device)
995 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
996 sync_blockdev(device->bdev);
997 invalidate_bdev(device->bdev);
1000 blkdev_put(device->bdev, device->mode);
1003 static void btrfs_prepare_close_one_device(struct btrfs_device *device)
1005 struct btrfs_fs_devices *fs_devices = device->fs_devices;
1006 struct btrfs_device *new_device;
1007 struct rcu_string *name;
1010 fs_devices->open_devices--;
1012 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
1013 device->devid != BTRFS_DEV_REPLACE_DEVID) {
1014 list_del_init(&device->dev_alloc_list);
1015 fs_devices->rw_devices--;
1018 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
1019 fs_devices->missing_devices--;
1021 new_device = btrfs_alloc_device(NULL, &device->devid,
1023 BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
1025 /* Safe because we are under uuid_mutex */
1027 name = rcu_string_strdup(device->name->str, GFP_NOFS);
1028 BUG_ON(!name); /* -ENOMEM */
1029 rcu_assign_pointer(new_device->name, name);
1032 list_replace_rcu(&device->dev_list, &new_device->dev_list);
1033 new_device->fs_devices = device->fs_devices;
1036 static int close_fs_devices(struct btrfs_fs_devices *fs_devices)
1038 struct btrfs_device *device, *tmp;
1039 struct list_head pending_put;
1041 INIT_LIST_HEAD(&pending_put);
1043 if (--fs_devices->opened > 0)
1046 mutex_lock(&fs_devices->device_list_mutex);
1047 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
1048 btrfs_prepare_close_one_device(device);
1049 list_add(&device->dev_list, &pending_put);
1051 mutex_unlock(&fs_devices->device_list_mutex);
1054 * btrfs_show_devname() is using the device_list_mutex,
1055 * sometimes call to blkdev_put() leads vfs calling
1056 * into this func. So do put outside of device_list_mutex,
1059 while (!list_empty(&pending_put)) {
1060 device = list_first_entry(&pending_put,
1061 struct btrfs_device, dev_list);
1062 list_del(&device->dev_list);
1063 btrfs_close_bdev(device);
1064 call_rcu(&device->rcu, free_device_rcu);
1067 WARN_ON(fs_devices->open_devices);
1068 WARN_ON(fs_devices->rw_devices);
1069 fs_devices->opened = 0;
1070 fs_devices->seeding = 0;
1075 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
1077 struct btrfs_fs_devices *seed_devices = NULL;
1080 mutex_lock(&uuid_mutex);
1081 ret = close_fs_devices(fs_devices);
1082 if (!fs_devices->opened) {
1083 seed_devices = fs_devices->seed;
1084 fs_devices->seed = NULL;
1086 mutex_unlock(&uuid_mutex);
1088 while (seed_devices) {
1089 fs_devices = seed_devices;
1090 seed_devices = fs_devices->seed;
1091 close_fs_devices(fs_devices);
1092 free_fs_devices(fs_devices);
1097 static int open_fs_devices(struct btrfs_fs_devices *fs_devices,
1098 fmode_t flags, void *holder)
1100 struct btrfs_device *device;
1101 struct btrfs_device *latest_dev = NULL;
1104 flags |= FMODE_EXCL;
1106 list_for_each_entry(device, &fs_devices->devices, dev_list) {
1107 /* Just open everything we can; ignore failures here */
1108 if (btrfs_open_one_device(fs_devices, device, flags, holder))
1112 device->generation > latest_dev->generation)
1113 latest_dev = device;
1115 if (fs_devices->open_devices == 0) {
1119 fs_devices->opened = 1;
1120 fs_devices->latest_bdev = latest_dev->bdev;
1121 fs_devices->total_rw_bytes = 0;
1126 static int devid_cmp(void *priv, struct list_head *a, struct list_head *b)
1128 struct btrfs_device *dev1, *dev2;
1130 dev1 = list_entry(a, struct btrfs_device, dev_list);
1131 dev2 = list_entry(b, struct btrfs_device, dev_list);
1133 if (dev1->devid < dev2->devid)
1135 else if (dev1->devid > dev2->devid)
1140 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1141 fmode_t flags, void *holder)
1145 mutex_lock(&uuid_mutex);
1146 if (fs_devices->opened) {
1147 fs_devices->opened++;
1150 list_sort(NULL, &fs_devices->devices, devid_cmp);
1151 ret = open_fs_devices(fs_devices, flags, holder);
1153 mutex_unlock(&uuid_mutex);
1157 static void btrfs_release_disk_super(struct page *page)
1163 static int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr,
1165 struct btrfs_super_block **disk_super)
1170 /* make sure our super fits in the device */
1171 if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1174 /* make sure our super fits in the page */
1175 if (sizeof(**disk_super) > PAGE_SIZE)
1178 /* make sure our super doesn't straddle pages on disk */
1179 index = bytenr >> PAGE_SHIFT;
1180 if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index)
1183 /* pull in the page with our super */
1184 *page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
1187 if (IS_ERR_OR_NULL(*page))
1192 /* align our pointer to the offset of the super block */
1193 *disk_super = p + (bytenr & ~PAGE_MASK);
1195 if (btrfs_super_bytenr(*disk_super) != bytenr ||
1196 btrfs_super_magic(*disk_super) != BTRFS_MAGIC) {
1197 btrfs_release_disk_super(*page);
1201 if ((*disk_super)->label[0] &&
1202 (*disk_super)->label[BTRFS_LABEL_SIZE - 1])
1203 (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0';
1209 * Look for a btrfs signature on a device. This may be called out of the mount path
1210 * and we are not allowed to call set_blocksize during the scan. The superblock
1211 * is read via pagecache
1213 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
1214 struct btrfs_fs_devices **fs_devices_ret)
1216 struct btrfs_super_block *disk_super;
1217 struct btrfs_device *device;
1218 struct block_device *bdev;
1224 * we would like to check all the supers, but that would make
1225 * a btrfs mount succeed after a mkfs from a different FS.
1226 * So, we need to add a special mount option to scan for
1227 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1229 bytenr = btrfs_sb_offset(0);
1230 flags |= FMODE_EXCL;
1231 mutex_lock(&uuid_mutex);
1233 bdev = blkdev_get_by_path(path, flags, holder);
1235 ret = PTR_ERR(bdev);
1239 if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super)) {
1241 goto error_bdev_put;
1244 device = device_list_add(path, disk_super);
1246 ret = PTR_ERR(device);
1248 *fs_devices_ret = device->fs_devices;
1250 btrfs_release_disk_super(page);
1253 blkdev_put(bdev, flags);
1255 mutex_unlock(&uuid_mutex);
1259 /* helper to account the used device space in the range */
1260 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
1261 u64 end, u64 *length)
1263 struct btrfs_key key;
1264 struct btrfs_root *root = device->fs_info->dev_root;
1265 struct btrfs_dev_extent *dev_extent;
1266 struct btrfs_path *path;
1270 struct extent_buffer *l;
1274 if (start >= device->total_bytes ||
1275 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
1278 path = btrfs_alloc_path();
1281 path->reada = READA_FORWARD;
1283 key.objectid = device->devid;
1285 key.type = BTRFS_DEV_EXTENT_KEY;
1287 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1291 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1298 slot = path->slots[0];
1299 if (slot >= btrfs_header_nritems(l)) {
1300 ret = btrfs_next_leaf(root, path);
1308 btrfs_item_key_to_cpu(l, &key, slot);
1310 if (key.objectid < device->devid)
1313 if (key.objectid > device->devid)
1316 if (key.type != BTRFS_DEV_EXTENT_KEY)
1319 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1320 extent_end = key.offset + btrfs_dev_extent_length(l,
1322 if (key.offset <= start && extent_end > end) {
1323 *length = end - start + 1;
1325 } else if (key.offset <= start && extent_end > start)
1326 *length += extent_end - start;
1327 else if (key.offset > start && extent_end <= end)
1328 *length += extent_end - key.offset;
1329 else if (key.offset > start && key.offset <= end) {
1330 *length += end - key.offset + 1;
1332 } else if (key.offset > end)
1340 btrfs_free_path(path);
1344 static int contains_pending_extent(struct btrfs_transaction *transaction,
1345 struct btrfs_device *device,
1346 u64 *start, u64 len)
1348 struct btrfs_fs_info *fs_info = device->fs_info;
1349 struct extent_map *em;
1350 struct list_head *search_list = &fs_info->pinned_chunks;
1352 u64 physical_start = *start;
1355 search_list = &transaction->pending_chunks;
1357 list_for_each_entry(em, search_list, list) {
1358 struct map_lookup *map;
1361 map = em->map_lookup;
1362 for (i = 0; i < map->num_stripes; i++) {
1365 if (map->stripes[i].dev != device)
1367 if (map->stripes[i].physical >= physical_start + len ||
1368 map->stripes[i].physical + em->orig_block_len <=
1372 * Make sure that while processing the pinned list we do
1373 * not override our *start with a lower value, because
1374 * we can have pinned chunks that fall within this
1375 * device hole and that have lower physical addresses
1376 * than the pending chunks we processed before. If we
1377 * do not take this special care we can end up getting
1378 * 2 pending chunks that start at the same physical
1379 * device offsets because the end offset of a pinned
1380 * chunk can be equal to the start offset of some
1383 end = map->stripes[i].physical + em->orig_block_len;
1390 if (search_list != &fs_info->pinned_chunks) {
1391 search_list = &fs_info->pinned_chunks;
1400 * find_free_dev_extent_start - find free space in the specified device
1401 * @device: the device which we search the free space in
1402 * @num_bytes: the size of the free space that we need
1403 * @search_start: the position from which to begin the search
1404 * @start: store the start of the free space.
1405 * @len: the size of the free space. that we find, or the size
1406 * of the max free space if we don't find suitable free space
1408 * this uses a pretty simple search, the expectation is that it is
1409 * called very infrequently and that a given device has a small number
1412 * @start is used to store the start of the free space if we find. But if we
1413 * don't find suitable free space, it will be used to store the start position
1414 * of the max free space.
1416 * @len is used to store the size of the free space that we find.
1417 * But if we don't find suitable free space, it is used to store the size of
1418 * the max free space.
1420 int find_free_dev_extent_start(struct btrfs_transaction *transaction,
1421 struct btrfs_device *device, u64 num_bytes,
1422 u64 search_start, u64 *start, u64 *len)
1424 struct btrfs_fs_info *fs_info = device->fs_info;
1425 struct btrfs_root *root = fs_info->dev_root;
1426 struct btrfs_key key;
1427 struct btrfs_dev_extent *dev_extent;
1428 struct btrfs_path *path;
1433 u64 search_end = device->total_bytes;
1436 struct extent_buffer *l;
1439 * We don't want to overwrite the superblock on the drive nor any area
1440 * used by the boot loader (grub for example), so we make sure to start
1441 * at an offset of at least 1MB.
1443 search_start = max_t(u64, search_start, SZ_1M);
1445 path = btrfs_alloc_path();
1449 max_hole_start = search_start;
1453 if (search_start >= search_end ||
1454 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1459 path->reada = READA_FORWARD;
1460 path->search_commit_root = 1;
1461 path->skip_locking = 1;
1463 key.objectid = device->devid;
1464 key.offset = search_start;
1465 key.type = BTRFS_DEV_EXTENT_KEY;
1467 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1471 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1478 slot = path->slots[0];
1479 if (slot >= btrfs_header_nritems(l)) {
1480 ret = btrfs_next_leaf(root, path);
1488 btrfs_item_key_to_cpu(l, &key, slot);
1490 if (key.objectid < device->devid)
1493 if (key.objectid > device->devid)
1496 if (key.type != BTRFS_DEV_EXTENT_KEY)
1499 if (key.offset > search_start) {
1500 hole_size = key.offset - search_start;
1503 * Have to check before we set max_hole_start, otherwise
1504 * we could end up sending back this offset anyway.
1506 if (contains_pending_extent(transaction, device,
1509 if (key.offset >= search_start) {
1510 hole_size = key.offset - search_start;
1517 if (hole_size > max_hole_size) {
1518 max_hole_start = search_start;
1519 max_hole_size = hole_size;
1523 * If this free space is greater than which we need,
1524 * it must be the max free space that we have found
1525 * until now, so max_hole_start must point to the start
1526 * of this free space and the length of this free space
1527 * is stored in max_hole_size. Thus, we return
1528 * max_hole_start and max_hole_size and go back to the
1531 if (hole_size >= num_bytes) {
1537 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1538 extent_end = key.offset + btrfs_dev_extent_length(l,
1540 if (extent_end > search_start)
1541 search_start = extent_end;
1548 * At this point, search_start should be the end of
1549 * allocated dev extents, and when shrinking the device,
1550 * search_end may be smaller than search_start.
1552 if (search_end > search_start) {
1553 hole_size = search_end - search_start;
1555 if (contains_pending_extent(transaction, device, &search_start,
1557 btrfs_release_path(path);
1561 if (hole_size > max_hole_size) {
1562 max_hole_start = search_start;
1563 max_hole_size = hole_size;
1568 if (max_hole_size < num_bytes)
1574 btrfs_free_path(path);
1575 *start = max_hole_start;
1577 *len = max_hole_size;
1581 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1582 struct btrfs_device *device, u64 num_bytes,
1583 u64 *start, u64 *len)
1585 /* FIXME use last free of some kind */
1586 return find_free_dev_extent_start(trans->transaction, device,
1587 num_bytes, 0, start, len);
1590 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1591 struct btrfs_device *device,
1592 u64 start, u64 *dev_extent_len)
1594 struct btrfs_fs_info *fs_info = device->fs_info;
1595 struct btrfs_root *root = fs_info->dev_root;
1597 struct btrfs_path *path;
1598 struct btrfs_key key;
1599 struct btrfs_key found_key;
1600 struct extent_buffer *leaf = NULL;
1601 struct btrfs_dev_extent *extent = NULL;
1603 path = btrfs_alloc_path();
1607 key.objectid = device->devid;
1609 key.type = BTRFS_DEV_EXTENT_KEY;
1611 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1613 ret = btrfs_previous_item(root, path, key.objectid,
1614 BTRFS_DEV_EXTENT_KEY);
1617 leaf = path->nodes[0];
1618 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1619 extent = btrfs_item_ptr(leaf, path->slots[0],
1620 struct btrfs_dev_extent);
1621 BUG_ON(found_key.offset > start || found_key.offset +
1622 btrfs_dev_extent_length(leaf, extent) < start);
1624 btrfs_release_path(path);
1626 } else if (ret == 0) {
1627 leaf = path->nodes[0];
1628 extent = btrfs_item_ptr(leaf, path->slots[0],
1629 struct btrfs_dev_extent);
1631 btrfs_handle_fs_error(fs_info, ret, "Slot search failed");
1635 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1637 ret = btrfs_del_item(trans, root, path);
1639 btrfs_handle_fs_error(fs_info, ret,
1640 "Failed to remove dev extent item");
1642 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1645 btrfs_free_path(path);
1649 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1650 struct btrfs_device *device,
1651 u64 chunk_offset, u64 start, u64 num_bytes)
1654 struct btrfs_path *path;
1655 struct btrfs_fs_info *fs_info = device->fs_info;
1656 struct btrfs_root *root = fs_info->dev_root;
1657 struct btrfs_dev_extent *extent;
1658 struct extent_buffer *leaf;
1659 struct btrfs_key key;
1661 WARN_ON(!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state));
1662 WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state));
1663 path = btrfs_alloc_path();
1667 key.objectid = device->devid;
1669 key.type = BTRFS_DEV_EXTENT_KEY;
1670 ret = btrfs_insert_empty_item(trans, root, path, &key,
1675 leaf = path->nodes[0];
1676 extent = btrfs_item_ptr(leaf, path->slots[0],
1677 struct btrfs_dev_extent);
1678 btrfs_set_dev_extent_chunk_tree(leaf, extent,
1679 BTRFS_CHUNK_TREE_OBJECTID);
1680 btrfs_set_dev_extent_chunk_objectid(leaf, extent,
1681 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
1682 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1684 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1685 btrfs_mark_buffer_dirty(leaf);
1687 btrfs_free_path(path);
1691 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1693 struct extent_map_tree *em_tree;
1694 struct extent_map *em;
1698 em_tree = &fs_info->mapping_tree.map_tree;
1699 read_lock(&em_tree->lock);
1700 n = rb_last(&em_tree->map);
1702 em = rb_entry(n, struct extent_map, rb_node);
1703 ret = em->start + em->len;
1705 read_unlock(&em_tree->lock);
1710 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1714 struct btrfs_key key;
1715 struct btrfs_key found_key;
1716 struct btrfs_path *path;
1718 path = btrfs_alloc_path();
1722 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1723 key.type = BTRFS_DEV_ITEM_KEY;
1724 key.offset = (u64)-1;
1726 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1730 BUG_ON(ret == 0); /* Corruption */
1732 ret = btrfs_previous_item(fs_info->chunk_root, path,
1733 BTRFS_DEV_ITEMS_OBJECTID,
1734 BTRFS_DEV_ITEM_KEY);
1738 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1740 *devid_ret = found_key.offset + 1;
1744 btrfs_free_path(path);
1749 * the device information is stored in the chunk root
1750 * the btrfs_device struct should be fully filled in
1752 static int btrfs_add_dev_item(struct btrfs_trans_handle *trans,
1753 struct btrfs_fs_info *fs_info,
1754 struct btrfs_device *device)
1756 struct btrfs_root *root = fs_info->chunk_root;
1758 struct btrfs_path *path;
1759 struct btrfs_dev_item *dev_item;
1760 struct extent_buffer *leaf;
1761 struct btrfs_key key;
1764 path = btrfs_alloc_path();
1768 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1769 key.type = BTRFS_DEV_ITEM_KEY;
1770 key.offset = device->devid;
1772 ret = btrfs_insert_empty_item(trans, root, path, &key,
1777 leaf = path->nodes[0];
1778 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1780 btrfs_set_device_id(leaf, dev_item, device->devid);
1781 btrfs_set_device_generation(leaf, dev_item, 0);
1782 btrfs_set_device_type(leaf, dev_item, device->type);
1783 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1784 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1785 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1786 btrfs_set_device_total_bytes(leaf, dev_item,
1787 btrfs_device_get_disk_total_bytes(device));
1788 btrfs_set_device_bytes_used(leaf, dev_item,
1789 btrfs_device_get_bytes_used(device));
1790 btrfs_set_device_group(leaf, dev_item, 0);
1791 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1792 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1793 btrfs_set_device_start_offset(leaf, dev_item, 0);
1795 ptr = btrfs_device_uuid(dev_item);
1796 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1797 ptr = btrfs_device_fsid(dev_item);
1798 write_extent_buffer(leaf, fs_info->fsid, ptr, BTRFS_FSID_SIZE);
1799 btrfs_mark_buffer_dirty(leaf);
1803 btrfs_free_path(path);
1808 * Function to update ctime/mtime for a given device path.
1809 * Mainly used for ctime/mtime based probe like libblkid.
1811 static void update_dev_time(const char *path_name)
1815 filp = filp_open(path_name, O_RDWR, 0);
1818 file_update_time(filp);
1819 filp_close(filp, NULL);
1822 static int btrfs_rm_dev_item(struct btrfs_fs_info *fs_info,
1823 struct btrfs_device *device)
1825 struct btrfs_root *root = fs_info->chunk_root;
1827 struct btrfs_path *path;
1828 struct btrfs_key key;
1829 struct btrfs_trans_handle *trans;
1831 path = btrfs_alloc_path();
1835 trans = btrfs_start_transaction(root, 0);
1836 if (IS_ERR(trans)) {
1837 btrfs_free_path(path);
1838 return PTR_ERR(trans);
1840 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1841 key.type = BTRFS_DEV_ITEM_KEY;
1842 key.offset = device->devid;
1844 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1848 btrfs_abort_transaction(trans, ret);
1849 btrfs_end_transaction(trans);
1853 ret = btrfs_del_item(trans, root, path);
1855 btrfs_abort_transaction(trans, ret);
1856 btrfs_end_transaction(trans);
1860 btrfs_free_path(path);
1862 ret = btrfs_commit_transaction(trans);
1867 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1868 * filesystem. It's up to the caller to adjust that number regarding eg. device
1871 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1879 seq = read_seqbegin(&fs_info->profiles_lock);
1881 all_avail = fs_info->avail_data_alloc_bits |
1882 fs_info->avail_system_alloc_bits |
1883 fs_info->avail_metadata_alloc_bits;
1884 } while (read_seqretry(&fs_info->profiles_lock, seq));
1886 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1887 if (!(all_avail & btrfs_raid_group[i]))
1890 if (num_devices < btrfs_raid_array[i].devs_min) {
1891 int ret = btrfs_raid_mindev_error[i];
1901 static struct btrfs_device * btrfs_find_next_active_device(
1902 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device)
1904 struct btrfs_device *next_device;
1906 list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1907 if (next_device != device &&
1908 !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state)
1909 && next_device->bdev)
1917 * Helper function to check if the given device is part of s_bdev / latest_bdev
1918 * and replace it with the provided or the next active device, in the context
1919 * where this function called, there should be always be another device (or
1920 * this_dev) which is active.
1922 void btrfs_assign_next_active_device(struct btrfs_fs_info *fs_info,
1923 struct btrfs_device *device, struct btrfs_device *this_dev)
1925 struct btrfs_device *next_device;
1928 next_device = this_dev;
1930 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
1932 ASSERT(next_device);
1934 if (fs_info->sb->s_bdev &&
1935 (fs_info->sb->s_bdev == device->bdev))
1936 fs_info->sb->s_bdev = next_device->bdev;
1938 if (fs_info->fs_devices->latest_bdev == device->bdev)
1939 fs_info->fs_devices->latest_bdev = next_device->bdev;
1942 int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path,
1945 struct btrfs_device *device;
1946 struct btrfs_fs_devices *cur_devices;
1947 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
1951 mutex_lock(&uuid_mutex);
1953 num_devices = fs_devices->num_devices;
1954 btrfs_dev_replace_read_lock(&fs_info->dev_replace);
1955 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
1956 WARN_ON(num_devices < 1);
1959 btrfs_dev_replace_read_unlock(&fs_info->dev_replace);
1961 ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
1965 ret = btrfs_find_device_by_devspec(fs_info, devid, device_path,
1970 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1971 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1975 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
1976 fs_info->fs_devices->rw_devices == 1) {
1977 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1981 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1982 mutex_lock(&fs_info->chunk_mutex);
1983 list_del_init(&device->dev_alloc_list);
1984 device->fs_devices->rw_devices--;
1985 mutex_unlock(&fs_info->chunk_mutex);
1988 mutex_unlock(&uuid_mutex);
1989 ret = btrfs_shrink_device(device, 0);
1990 mutex_lock(&uuid_mutex);
1995 * TODO: the superblock still includes this device in its num_devices
1996 * counter although write_all_supers() is not locked out. This
1997 * could give a filesystem state which requires a degraded mount.
1999 ret = btrfs_rm_dev_item(fs_info, device);
2003 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2004 btrfs_scrub_cancel_dev(fs_info, device);
2007 * the device list mutex makes sure that we don't change
2008 * the device list while someone else is writing out all
2009 * the device supers. Whoever is writing all supers, should
2010 * lock the device list mutex before getting the number of
2011 * devices in the super block (super_copy). Conversely,
2012 * whoever updates the number of devices in the super block
2013 * (super_copy) should hold the device list mutex.
2016 cur_devices = device->fs_devices;
2017 mutex_lock(&fs_devices->device_list_mutex);
2018 list_del_rcu(&device->dev_list);
2020 device->fs_devices->num_devices--;
2021 device->fs_devices->total_devices--;
2023 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
2024 device->fs_devices->missing_devices--;
2026 btrfs_assign_next_active_device(fs_info, device, NULL);
2029 device->fs_devices->open_devices--;
2030 /* remove sysfs entry */
2031 btrfs_sysfs_rm_device_link(fs_devices, device);
2034 num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
2035 btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
2036 mutex_unlock(&fs_devices->device_list_mutex);
2039 * at this point, the device is zero sized and detached from
2040 * the devices list. All that's left is to zero out the old
2041 * supers and free the device.
2043 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2044 btrfs_scratch_superblocks(device->bdev, device->name->str);
2046 btrfs_close_bdev(device);
2047 call_rcu(&device->rcu, free_device_rcu);
2049 if (cur_devices->open_devices == 0) {
2050 while (fs_devices) {
2051 if (fs_devices->seed == cur_devices) {
2052 fs_devices->seed = cur_devices->seed;
2055 fs_devices = fs_devices->seed;
2057 cur_devices->seed = NULL;
2058 close_fs_devices(cur_devices);
2059 free_fs_devices(cur_devices);
2063 mutex_unlock(&uuid_mutex);
2067 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2068 mutex_lock(&fs_info->chunk_mutex);
2069 list_add(&device->dev_alloc_list,
2070 &fs_devices->alloc_list);
2071 device->fs_devices->rw_devices++;
2072 mutex_unlock(&fs_info->chunk_mutex);
2077 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
2078 struct btrfs_device *srcdev)
2080 struct btrfs_fs_devices *fs_devices;
2082 lockdep_assert_held(&fs_info->fs_devices->device_list_mutex);
2085 * in case of fs with no seed, srcdev->fs_devices will point
2086 * to fs_devices of fs_info. However when the dev being replaced is
2087 * a seed dev it will point to the seed's local fs_devices. In short
2088 * srcdev will have its correct fs_devices in both the cases.
2090 fs_devices = srcdev->fs_devices;
2092 list_del_rcu(&srcdev->dev_list);
2093 list_del(&srcdev->dev_alloc_list);
2094 fs_devices->num_devices--;
2095 if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state))
2096 fs_devices->missing_devices--;
2098 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state))
2099 fs_devices->rw_devices--;
2102 fs_devices->open_devices--;
2105 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
2106 struct btrfs_device *srcdev)
2108 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2110 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) {
2111 /* zero out the old super if it is writable */
2112 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
2115 btrfs_close_bdev(srcdev);
2116 call_rcu(&srcdev->rcu, free_device_rcu);
2118 /* if this is no devs we rather delete the fs_devices */
2119 if (!fs_devices->num_devices) {
2120 struct btrfs_fs_devices *tmp_fs_devices;
2123 * On a mounted FS, num_devices can't be zero unless it's a
2124 * seed. In case of a seed device being replaced, the replace
2125 * target added to the sprout FS, so there will be no more
2126 * device left under the seed FS.
2128 ASSERT(fs_devices->seeding);
2130 tmp_fs_devices = fs_info->fs_devices;
2131 while (tmp_fs_devices) {
2132 if (tmp_fs_devices->seed == fs_devices) {
2133 tmp_fs_devices->seed = fs_devices->seed;
2136 tmp_fs_devices = tmp_fs_devices->seed;
2138 fs_devices->seed = NULL;
2139 close_fs_devices(fs_devices);
2140 free_fs_devices(fs_devices);
2144 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2145 struct btrfs_device *tgtdev)
2147 mutex_lock(&uuid_mutex);
2149 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2151 btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev);
2154 fs_info->fs_devices->open_devices--;
2156 fs_info->fs_devices->num_devices--;
2158 btrfs_assign_next_active_device(fs_info, tgtdev, NULL);
2160 list_del_rcu(&tgtdev->dev_list);
2162 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2163 mutex_unlock(&uuid_mutex);
2166 * The update_dev_time() with in btrfs_scratch_superblocks()
2167 * may lead to a call to btrfs_show_devname() which will try
2168 * to hold device_list_mutex. And here this device
2169 * is already out of device list, so we don't have to hold
2170 * the device_list_mutex lock.
2172 btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
2174 btrfs_close_bdev(tgtdev);
2175 call_rcu(&tgtdev->rcu, free_device_rcu);
2178 static int btrfs_find_device_by_path(struct btrfs_fs_info *fs_info,
2179 const char *device_path,
2180 struct btrfs_device **device)
2183 struct btrfs_super_block *disk_super;
2186 struct block_device *bdev;
2187 struct buffer_head *bh;
2190 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2191 fs_info->bdev_holder, 0, &bdev, &bh);
2194 disk_super = (struct btrfs_super_block *)bh->b_data;
2195 devid = btrfs_stack_device_id(&disk_super->dev_item);
2196 dev_uuid = disk_super->dev_item.uuid;
2197 *device = btrfs_find_device(fs_info, devid, dev_uuid, disk_super->fsid);
2201 blkdev_put(bdev, FMODE_READ);
2205 int btrfs_find_device_missing_or_by_path(struct btrfs_fs_info *fs_info,
2206 const char *device_path,
2207 struct btrfs_device **device)
2210 if (strcmp(device_path, "missing") == 0) {
2211 struct list_head *devices;
2212 struct btrfs_device *tmp;
2214 devices = &fs_info->fs_devices->devices;
2215 list_for_each_entry(tmp, devices, dev_list) {
2216 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
2217 &tmp->dev_state) && !tmp->bdev) {
2224 return BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2228 return btrfs_find_device_by_path(fs_info, device_path, device);
2233 * Lookup a device given by device id, or the path if the id is 0.
2235 int btrfs_find_device_by_devspec(struct btrfs_fs_info *fs_info, u64 devid,
2236 const char *devpath,
2237 struct btrfs_device **device)
2243 *device = btrfs_find_device(fs_info, devid, NULL, NULL);
2247 if (!devpath || !devpath[0])
2250 ret = btrfs_find_device_missing_or_by_path(fs_info, devpath,
2257 * does all the dirty work required for changing file system's UUID.
2259 static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info)
2261 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2262 struct btrfs_fs_devices *old_devices;
2263 struct btrfs_fs_devices *seed_devices;
2264 struct btrfs_super_block *disk_super = fs_info->super_copy;
2265 struct btrfs_device *device;
2268 lockdep_assert_held(&uuid_mutex);
2269 if (!fs_devices->seeding)
2272 seed_devices = alloc_fs_devices(NULL);
2273 if (IS_ERR(seed_devices))
2274 return PTR_ERR(seed_devices);
2276 old_devices = clone_fs_devices(fs_devices);
2277 if (IS_ERR(old_devices)) {
2278 kfree(seed_devices);
2279 return PTR_ERR(old_devices);
2282 list_add(&old_devices->fs_list, &fs_uuids);
2284 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2285 seed_devices->opened = 1;
2286 INIT_LIST_HEAD(&seed_devices->devices);
2287 INIT_LIST_HEAD(&seed_devices->alloc_list);
2288 mutex_init(&seed_devices->device_list_mutex);
2290 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2291 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2293 list_for_each_entry(device, &seed_devices->devices, dev_list)
2294 device->fs_devices = seed_devices;
2296 mutex_lock(&fs_info->chunk_mutex);
2297 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2298 mutex_unlock(&fs_info->chunk_mutex);
2300 fs_devices->seeding = 0;
2301 fs_devices->num_devices = 0;
2302 fs_devices->open_devices = 0;
2303 fs_devices->missing_devices = 0;
2304 fs_devices->rotating = 0;
2305 fs_devices->seed = seed_devices;
2307 generate_random_uuid(fs_devices->fsid);
2308 memcpy(fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2309 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2310 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2312 super_flags = btrfs_super_flags(disk_super) &
2313 ~BTRFS_SUPER_FLAG_SEEDING;
2314 btrfs_set_super_flags(disk_super, super_flags);
2320 * Store the expected generation for seed devices in device items.
2322 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2323 struct btrfs_fs_info *fs_info)
2325 struct btrfs_root *root = fs_info->chunk_root;
2326 struct btrfs_path *path;
2327 struct extent_buffer *leaf;
2328 struct btrfs_dev_item *dev_item;
2329 struct btrfs_device *device;
2330 struct btrfs_key key;
2331 u8 fs_uuid[BTRFS_FSID_SIZE];
2332 u8 dev_uuid[BTRFS_UUID_SIZE];
2336 path = btrfs_alloc_path();
2340 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2342 key.type = BTRFS_DEV_ITEM_KEY;
2345 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2349 leaf = path->nodes[0];
2351 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2352 ret = btrfs_next_leaf(root, path);
2357 leaf = path->nodes[0];
2358 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2359 btrfs_release_path(path);
2363 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2364 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2365 key.type != BTRFS_DEV_ITEM_KEY)
2368 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2369 struct btrfs_dev_item);
2370 devid = btrfs_device_id(leaf, dev_item);
2371 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2373 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2375 device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
2376 BUG_ON(!device); /* Logic error */
2378 if (device->fs_devices->seeding) {
2379 btrfs_set_device_generation(leaf, dev_item,
2380 device->generation);
2381 btrfs_mark_buffer_dirty(leaf);
2389 btrfs_free_path(path);
2393 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2395 struct btrfs_root *root = fs_info->dev_root;
2396 struct request_queue *q;
2397 struct btrfs_trans_handle *trans;
2398 struct btrfs_device *device;
2399 struct block_device *bdev;
2400 struct list_head *devices;
2401 struct super_block *sb = fs_info->sb;
2402 struct rcu_string *name;
2404 int seeding_dev = 0;
2406 bool unlocked = false;
2408 if (sb_rdonly(sb) && !fs_info->fs_devices->seeding)
2411 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2412 fs_info->bdev_holder);
2414 return PTR_ERR(bdev);
2416 if (fs_info->fs_devices->seeding) {
2418 down_write(&sb->s_umount);
2419 mutex_lock(&uuid_mutex);
2422 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2424 devices = &fs_info->fs_devices->devices;
2426 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2427 list_for_each_entry(device, devices, dev_list) {
2428 if (device->bdev == bdev) {
2431 &fs_info->fs_devices->device_list_mutex);
2435 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2437 device = btrfs_alloc_device(fs_info, NULL, NULL);
2438 if (IS_ERR(device)) {
2439 /* we can safely leave the fs_devices entry around */
2440 ret = PTR_ERR(device);
2444 name = rcu_string_strdup(device_path, GFP_KERNEL);
2447 goto error_free_device;
2449 rcu_assign_pointer(device->name, name);
2451 trans = btrfs_start_transaction(root, 0);
2452 if (IS_ERR(trans)) {
2453 ret = PTR_ERR(trans);
2454 goto error_free_device;
2457 q = bdev_get_queue(bdev);
2458 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
2459 device->generation = trans->transid;
2460 device->io_width = fs_info->sectorsize;
2461 device->io_align = fs_info->sectorsize;
2462 device->sector_size = fs_info->sectorsize;
2463 device->total_bytes = round_down(i_size_read(bdev->bd_inode),
2464 fs_info->sectorsize);
2465 device->disk_total_bytes = device->total_bytes;
2466 device->commit_total_bytes = device->total_bytes;
2467 device->fs_info = fs_info;
2468 device->bdev = bdev;
2469 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2470 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
2471 device->mode = FMODE_EXCL;
2472 device->dev_stats_valid = 1;
2473 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2476 sb->s_flags &= ~SB_RDONLY;
2477 ret = btrfs_prepare_sprout(fs_info);
2479 btrfs_abort_transaction(trans, ret);
2484 device->fs_devices = fs_info->fs_devices;
2486 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2487 mutex_lock(&fs_info->chunk_mutex);
2488 list_add_rcu(&device->dev_list, &fs_info->fs_devices->devices);
2489 list_add(&device->dev_alloc_list,
2490 &fs_info->fs_devices->alloc_list);
2491 fs_info->fs_devices->num_devices++;
2492 fs_info->fs_devices->open_devices++;
2493 fs_info->fs_devices->rw_devices++;
2494 fs_info->fs_devices->total_devices++;
2495 fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2497 atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2499 if (!blk_queue_nonrot(q))
2500 fs_info->fs_devices->rotating = 1;
2502 tmp = btrfs_super_total_bytes(fs_info->super_copy);
2503 btrfs_set_super_total_bytes(fs_info->super_copy,
2504 round_down(tmp + device->total_bytes, fs_info->sectorsize));
2506 tmp = btrfs_super_num_devices(fs_info->super_copy);
2507 btrfs_set_super_num_devices(fs_info->super_copy, tmp + 1);
2509 /* add sysfs device entry */
2510 btrfs_sysfs_add_device_link(fs_info->fs_devices, device);
2513 * we've got more storage, clear any full flags on the space
2516 btrfs_clear_space_info_full(fs_info);
2518 mutex_unlock(&fs_info->chunk_mutex);
2519 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2522 mutex_lock(&fs_info->chunk_mutex);
2523 ret = init_first_rw_device(trans, fs_info);
2524 mutex_unlock(&fs_info->chunk_mutex);
2526 btrfs_abort_transaction(trans, ret);
2531 ret = btrfs_add_dev_item(trans, fs_info, device);
2533 btrfs_abort_transaction(trans, ret);
2538 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2540 ret = btrfs_finish_sprout(trans, fs_info);
2542 btrfs_abort_transaction(trans, ret);
2546 /* Sprouting would change fsid of the mounted root,
2547 * so rename the fsid on the sysfs
2549 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2551 if (kobject_rename(&fs_info->fs_devices->fsid_kobj, fsid_buf))
2553 "sysfs: failed to create fsid for sprout");
2556 ret = btrfs_commit_transaction(trans);
2559 mutex_unlock(&uuid_mutex);
2560 up_write(&sb->s_umount);
2563 if (ret) /* transaction commit */
2566 ret = btrfs_relocate_sys_chunks(fs_info);
2568 btrfs_handle_fs_error(fs_info, ret,
2569 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2570 trans = btrfs_attach_transaction(root);
2571 if (IS_ERR(trans)) {
2572 if (PTR_ERR(trans) == -ENOENT)
2574 ret = PTR_ERR(trans);
2578 ret = btrfs_commit_transaction(trans);
2581 /* Update ctime/mtime for libblkid */
2582 update_dev_time(device_path);
2586 btrfs_sysfs_rm_device_link(fs_info->fs_devices, device);
2589 sb->s_flags |= SB_RDONLY;
2591 btrfs_end_transaction(trans);
2593 btrfs_free_device(device);
2595 blkdev_put(bdev, FMODE_EXCL);
2596 if (seeding_dev && !unlocked) {
2597 mutex_unlock(&uuid_mutex);
2598 up_write(&sb->s_umount);
2603 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2604 struct btrfs_device *device)
2607 struct btrfs_path *path;
2608 struct btrfs_root *root = device->fs_info->chunk_root;
2609 struct btrfs_dev_item *dev_item;
2610 struct extent_buffer *leaf;
2611 struct btrfs_key key;
2613 path = btrfs_alloc_path();
2617 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2618 key.type = BTRFS_DEV_ITEM_KEY;
2619 key.offset = device->devid;
2621 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2630 leaf = path->nodes[0];
2631 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2633 btrfs_set_device_id(leaf, dev_item, device->devid);
2634 btrfs_set_device_type(leaf, dev_item, device->type);
2635 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2636 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2637 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2638 btrfs_set_device_total_bytes(leaf, dev_item,
2639 btrfs_device_get_disk_total_bytes(device));
2640 btrfs_set_device_bytes_used(leaf, dev_item,
2641 btrfs_device_get_bytes_used(device));
2642 btrfs_mark_buffer_dirty(leaf);
2645 btrfs_free_path(path);
2649 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2650 struct btrfs_device *device, u64 new_size)
2652 struct btrfs_fs_info *fs_info = device->fs_info;
2653 struct btrfs_super_block *super_copy = fs_info->super_copy;
2654 struct btrfs_fs_devices *fs_devices;
2658 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2661 new_size = round_down(new_size, fs_info->sectorsize);
2663 mutex_lock(&fs_info->chunk_mutex);
2664 old_total = btrfs_super_total_bytes(super_copy);
2665 diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2667 if (new_size <= device->total_bytes ||
2668 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2669 mutex_unlock(&fs_info->chunk_mutex);
2673 fs_devices = fs_info->fs_devices;
2675 btrfs_set_super_total_bytes(super_copy,
2676 round_down(old_total + diff, fs_info->sectorsize));
2677 device->fs_devices->total_rw_bytes += diff;
2679 btrfs_device_set_total_bytes(device, new_size);
2680 btrfs_device_set_disk_total_bytes(device, new_size);
2681 btrfs_clear_space_info_full(device->fs_info);
2682 if (list_empty(&device->resized_list))
2683 list_add_tail(&device->resized_list,
2684 &fs_devices->resized_devices);
2685 mutex_unlock(&fs_info->chunk_mutex);
2687 return btrfs_update_device(trans, device);
2690 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2691 struct btrfs_fs_info *fs_info, u64 chunk_offset)
2693 struct btrfs_root *root = fs_info->chunk_root;
2695 struct btrfs_path *path;
2696 struct btrfs_key key;
2698 path = btrfs_alloc_path();
2702 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2703 key.offset = chunk_offset;
2704 key.type = BTRFS_CHUNK_ITEM_KEY;
2706 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2709 else if (ret > 0) { /* Logic error or corruption */
2710 btrfs_handle_fs_error(fs_info, -ENOENT,
2711 "Failed lookup while freeing chunk.");
2716 ret = btrfs_del_item(trans, root, path);
2718 btrfs_handle_fs_error(fs_info, ret,
2719 "Failed to delete chunk item.");
2721 btrfs_free_path(path);
2725 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2727 struct btrfs_super_block *super_copy = fs_info->super_copy;
2728 struct btrfs_disk_key *disk_key;
2729 struct btrfs_chunk *chunk;
2736 struct btrfs_key key;
2738 mutex_lock(&fs_info->chunk_mutex);
2739 array_size = btrfs_super_sys_array_size(super_copy);
2741 ptr = super_copy->sys_chunk_array;
2744 while (cur < array_size) {
2745 disk_key = (struct btrfs_disk_key *)ptr;
2746 btrfs_disk_key_to_cpu(&key, disk_key);
2748 len = sizeof(*disk_key);
2750 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2751 chunk = (struct btrfs_chunk *)(ptr + len);
2752 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2753 len += btrfs_chunk_item_size(num_stripes);
2758 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID &&
2759 key.offset == chunk_offset) {
2760 memmove(ptr, ptr + len, array_size - (cur + len));
2762 btrfs_set_super_sys_array_size(super_copy, array_size);
2768 mutex_unlock(&fs_info->chunk_mutex);
2772 static struct extent_map *get_chunk_map(struct btrfs_fs_info *fs_info,
2773 u64 logical, u64 length)
2775 struct extent_map_tree *em_tree;
2776 struct extent_map *em;
2778 em_tree = &fs_info->mapping_tree.map_tree;
2779 read_lock(&em_tree->lock);
2780 em = lookup_extent_mapping(em_tree, logical, length);
2781 read_unlock(&em_tree->lock);
2784 btrfs_crit(fs_info, "unable to find logical %llu length %llu",
2786 return ERR_PTR(-EINVAL);
2789 if (em->start > logical || em->start + em->len < logical) {
2791 "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
2792 logical, length, em->start, em->start + em->len);
2793 free_extent_map(em);
2794 return ERR_PTR(-EINVAL);
2797 /* callers are responsible for dropping em's ref. */
2801 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2802 struct btrfs_fs_info *fs_info, u64 chunk_offset)
2804 struct extent_map *em;
2805 struct map_lookup *map;
2806 u64 dev_extent_len = 0;
2808 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2810 em = get_chunk_map(fs_info, chunk_offset, 1);
2813 * This is a logic error, but we don't want to just rely on the
2814 * user having built with ASSERT enabled, so if ASSERT doesn't
2815 * do anything we still error out.
2820 map = em->map_lookup;
2821 mutex_lock(&fs_info->chunk_mutex);
2822 check_system_chunk(trans, fs_info, map->type);
2823 mutex_unlock(&fs_info->chunk_mutex);
2826 * Take the device list mutex to prevent races with the final phase of
2827 * a device replace operation that replaces the device object associated
2828 * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2830 mutex_lock(&fs_devices->device_list_mutex);
2831 for (i = 0; i < map->num_stripes; i++) {
2832 struct btrfs_device *device = map->stripes[i].dev;
2833 ret = btrfs_free_dev_extent(trans, device,
2834 map->stripes[i].physical,
2837 mutex_unlock(&fs_devices->device_list_mutex);
2838 btrfs_abort_transaction(trans, ret);
2842 if (device->bytes_used > 0) {
2843 mutex_lock(&fs_info->chunk_mutex);
2844 btrfs_device_set_bytes_used(device,
2845 device->bytes_used - dev_extent_len);
2846 atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
2847 btrfs_clear_space_info_full(fs_info);
2848 mutex_unlock(&fs_info->chunk_mutex);
2851 if (map->stripes[i].dev) {
2852 ret = btrfs_update_device(trans, map->stripes[i].dev);
2854 mutex_unlock(&fs_devices->device_list_mutex);
2855 btrfs_abort_transaction(trans, ret);
2860 mutex_unlock(&fs_devices->device_list_mutex);
2862 ret = btrfs_free_chunk(trans, fs_info, chunk_offset);
2864 btrfs_abort_transaction(trans, ret);
2868 trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
2870 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2871 ret = btrfs_del_sys_chunk(fs_info, chunk_offset);
2873 btrfs_abort_transaction(trans, ret);
2878 ret = btrfs_remove_block_group(trans, fs_info, chunk_offset, em);
2880 btrfs_abort_transaction(trans, ret);
2886 free_extent_map(em);
2890 static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2892 struct btrfs_root *root = fs_info->chunk_root;
2893 struct btrfs_trans_handle *trans;
2897 * Prevent races with automatic removal of unused block groups.
2898 * After we relocate and before we remove the chunk with offset
2899 * chunk_offset, automatic removal of the block group can kick in,
2900 * resulting in a failure when calling btrfs_remove_chunk() below.
2902 * Make sure to acquire this mutex before doing a tree search (dev
2903 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2904 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2905 * we release the path used to search the chunk/dev tree and before
2906 * the current task acquires this mutex and calls us.
2908 lockdep_assert_held(&fs_info->delete_unused_bgs_mutex);
2910 ret = btrfs_can_relocate(fs_info, chunk_offset);
2914 /* step one, relocate all the extents inside this chunk */
2915 btrfs_scrub_pause(fs_info);
2916 ret = btrfs_relocate_block_group(fs_info, chunk_offset);
2917 btrfs_scrub_continue(fs_info);
2922 * We add the kobjects here (and after forcing data chunk creation)
2923 * since relocation is the only place we'll create chunks of a new
2924 * type at runtime. The only place where we'll remove the last
2925 * chunk of a type is the call immediately below this one. Even
2926 * so, we're protected against races with the cleaner thread since
2927 * we're covered by the delete_unused_bgs_mutex.
2929 btrfs_add_raid_kobjects(fs_info);
2931 trans = btrfs_start_trans_remove_block_group(root->fs_info,
2933 if (IS_ERR(trans)) {
2934 ret = PTR_ERR(trans);
2935 btrfs_handle_fs_error(root->fs_info, ret, NULL);
2940 * step two, delete the device extents and the
2941 * chunk tree entries
2943 ret = btrfs_remove_chunk(trans, fs_info, chunk_offset);
2944 btrfs_end_transaction(trans);
2948 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
2950 struct btrfs_root *chunk_root = fs_info->chunk_root;
2951 struct btrfs_path *path;
2952 struct extent_buffer *leaf;
2953 struct btrfs_chunk *chunk;
2954 struct btrfs_key key;
2955 struct btrfs_key found_key;
2957 bool retried = false;
2961 path = btrfs_alloc_path();
2966 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2967 key.offset = (u64)-1;
2968 key.type = BTRFS_CHUNK_ITEM_KEY;
2971 mutex_lock(&fs_info->delete_unused_bgs_mutex);
2972 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2974 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
2977 BUG_ON(ret == 0); /* Corruption */
2979 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2982 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
2988 leaf = path->nodes[0];
2989 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2991 chunk = btrfs_item_ptr(leaf, path->slots[0],
2992 struct btrfs_chunk);
2993 chunk_type = btrfs_chunk_type(leaf, chunk);
2994 btrfs_release_path(path);
2996 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2997 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3003 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3005 if (found_key.offset == 0)
3007 key.offset = found_key.offset - 1;
3010 if (failed && !retried) {
3014 } else if (WARN_ON(failed && retried)) {
3018 btrfs_free_path(path);
3023 * return 1 : allocate a data chunk successfully,
3024 * return <0: errors during allocating a data chunk,
3025 * return 0 : no need to allocate a data chunk.
3027 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info,
3030 struct btrfs_block_group_cache *cache;
3034 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3036 chunk_type = cache->flags;
3037 btrfs_put_block_group(cache);
3039 if (chunk_type & BTRFS_BLOCK_GROUP_DATA) {
3040 spin_lock(&fs_info->data_sinfo->lock);
3041 bytes_used = fs_info->data_sinfo->bytes_used;
3042 spin_unlock(&fs_info->data_sinfo->lock);
3045 struct btrfs_trans_handle *trans;
3048 trans = btrfs_join_transaction(fs_info->tree_root);
3050 return PTR_ERR(trans);
3052 ret = btrfs_force_chunk_alloc(trans, fs_info,
3053 BTRFS_BLOCK_GROUP_DATA);
3054 btrfs_end_transaction(trans);
3058 btrfs_add_raid_kobjects(fs_info);
3066 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3067 struct btrfs_balance_control *bctl)
3069 struct btrfs_root *root = fs_info->tree_root;
3070 struct btrfs_trans_handle *trans;
3071 struct btrfs_balance_item *item;
3072 struct btrfs_disk_balance_args disk_bargs;
3073 struct btrfs_path *path;
3074 struct extent_buffer *leaf;
3075 struct btrfs_key key;
3078 path = btrfs_alloc_path();
3082 trans = btrfs_start_transaction(root, 0);
3083 if (IS_ERR(trans)) {
3084 btrfs_free_path(path);
3085 return PTR_ERR(trans);
3088 key.objectid = BTRFS_BALANCE_OBJECTID;
3089 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3092 ret = btrfs_insert_empty_item(trans, root, path, &key,
3097 leaf = path->nodes[0];
3098 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3100 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3102 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3103 btrfs_set_balance_data(leaf, item, &disk_bargs);
3104 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3105 btrfs_set_balance_meta(leaf, item, &disk_bargs);
3106 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3107 btrfs_set_balance_sys(leaf, item, &disk_bargs);
3109 btrfs_set_balance_flags(leaf, item, bctl->flags);
3111 btrfs_mark_buffer_dirty(leaf);
3113 btrfs_free_path(path);
3114 err = btrfs_commit_transaction(trans);
3120 static int del_balance_item(struct btrfs_fs_info *fs_info)
3122 struct btrfs_root *root = fs_info->tree_root;
3123 struct btrfs_trans_handle *trans;
3124 struct btrfs_path *path;
3125 struct btrfs_key key;
3128 path = btrfs_alloc_path();
3132 trans = btrfs_start_transaction(root, 0);
3133 if (IS_ERR(trans)) {
3134 btrfs_free_path(path);
3135 return PTR_ERR(trans);
3138 key.objectid = BTRFS_BALANCE_OBJECTID;
3139 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3142 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3150 ret = btrfs_del_item(trans, root, path);
3152 btrfs_free_path(path);
3153 err = btrfs_commit_transaction(trans);
3160 * This is a heuristic used to reduce the number of chunks balanced on
3161 * resume after balance was interrupted.
3163 static void update_balance_args(struct btrfs_balance_control *bctl)
3166 * Turn on soft mode for chunk types that were being converted.
3168 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3169 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3170 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3171 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3172 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3173 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3176 * Turn on usage filter if is not already used. The idea is
3177 * that chunks that we have already balanced should be
3178 * reasonably full. Don't do it for chunks that are being
3179 * converted - that will keep us from relocating unconverted
3180 * (albeit full) chunks.
3182 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3183 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3184 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3185 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3186 bctl->data.usage = 90;
3188 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3189 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3190 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3191 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3192 bctl->sys.usage = 90;
3194 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3195 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3196 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3197 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3198 bctl->meta.usage = 90;
3203 * Should be called with balance mutex held to protect against checking the
3204 * balance status or progress. Same goes for reset_balance_state.
3206 static void set_balance_control(struct btrfs_balance_control *bctl)
3208 struct btrfs_fs_info *fs_info = bctl->fs_info;
3210 BUG_ON(fs_info->balance_ctl);
3212 spin_lock(&fs_info->balance_lock);
3213 fs_info->balance_ctl = bctl;
3214 spin_unlock(&fs_info->balance_lock);
3218 * Clear the balance status in fs_info and delete the balance item from disk.
3220 static void reset_balance_state(struct btrfs_fs_info *fs_info)
3222 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3225 BUG_ON(!fs_info->balance_ctl);
3227 spin_lock(&fs_info->balance_lock);
3228 fs_info->balance_ctl = NULL;
3229 spin_unlock(&fs_info->balance_lock);
3232 ret = del_balance_item(fs_info);
3234 btrfs_handle_fs_error(fs_info, ret, NULL);
3238 * Balance filters. Return 1 if chunk should be filtered out
3239 * (should not be balanced).
3241 static int chunk_profiles_filter(u64 chunk_type,
3242 struct btrfs_balance_args *bargs)
3244 chunk_type = chunk_to_extended(chunk_type) &
3245 BTRFS_EXTENDED_PROFILE_MASK;
3247 if (bargs->profiles & chunk_type)
3253 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3254 struct btrfs_balance_args *bargs)
3256 struct btrfs_block_group_cache *cache;
3258 u64 user_thresh_min;
3259 u64 user_thresh_max;
3262 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3263 chunk_used = btrfs_block_group_used(&cache->item);
3265 if (bargs->usage_min == 0)
3266 user_thresh_min = 0;
3268 user_thresh_min = div_factor_fine(cache->key.offset,
3271 if (bargs->usage_max == 0)
3272 user_thresh_max = 1;
3273 else if (bargs->usage_max > 100)
3274 user_thresh_max = cache->key.offset;
3276 user_thresh_max = div_factor_fine(cache->key.offset,
3279 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3282 btrfs_put_block_group(cache);
3286 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3287 u64 chunk_offset, struct btrfs_balance_args *bargs)
3289 struct btrfs_block_group_cache *cache;
3290 u64 chunk_used, user_thresh;
3293 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3294 chunk_used = btrfs_block_group_used(&cache->item);
3296 if (bargs->usage_min == 0)
3298 else if (bargs->usage > 100)
3299 user_thresh = cache->key.offset;
3301 user_thresh = div_factor_fine(cache->key.offset,
3304 if (chunk_used < user_thresh)
3307 btrfs_put_block_group(cache);
3311 static int chunk_devid_filter(struct extent_buffer *leaf,
3312 struct btrfs_chunk *chunk,
3313 struct btrfs_balance_args *bargs)
3315 struct btrfs_stripe *stripe;
3316 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3319 for (i = 0; i < num_stripes; i++) {
3320 stripe = btrfs_stripe_nr(chunk, i);
3321 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3328 /* [pstart, pend) */
3329 static int chunk_drange_filter(struct extent_buffer *leaf,
3330 struct btrfs_chunk *chunk,
3331 struct btrfs_balance_args *bargs)
3333 struct btrfs_stripe *stripe;
3334 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3340 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3343 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3344 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3345 factor = num_stripes / 2;
3346 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3347 factor = num_stripes - 1;
3348 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3349 factor = num_stripes - 2;
3351 factor = num_stripes;
3354 for (i = 0; i < num_stripes; i++) {
3355 stripe = btrfs_stripe_nr(chunk, i);
3356 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3359 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3360 stripe_length = btrfs_chunk_length(leaf, chunk);
3361 stripe_length = div_u64(stripe_length, factor);
3363 if (stripe_offset < bargs->pend &&
3364 stripe_offset + stripe_length > bargs->pstart)
3371 /* [vstart, vend) */
3372 static int chunk_vrange_filter(struct extent_buffer *leaf,
3373 struct btrfs_chunk *chunk,
3375 struct btrfs_balance_args *bargs)
3377 if (chunk_offset < bargs->vend &&
3378 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3379 /* at least part of the chunk is inside this vrange */
3385 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3386 struct btrfs_chunk *chunk,
3387 struct btrfs_balance_args *bargs)
3389 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3391 if (bargs->stripes_min <= num_stripes
3392 && num_stripes <= bargs->stripes_max)
3398 static int chunk_soft_convert_filter(u64 chunk_type,
3399 struct btrfs_balance_args *bargs)
3401 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3404 chunk_type = chunk_to_extended(chunk_type) &
3405 BTRFS_EXTENDED_PROFILE_MASK;
3407 if (bargs->target == chunk_type)
3413 static int should_balance_chunk(struct btrfs_fs_info *fs_info,
3414 struct extent_buffer *leaf,
3415 struct btrfs_chunk *chunk, u64 chunk_offset)
3417 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3418 struct btrfs_balance_args *bargs = NULL;
3419 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3422 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3423 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3427 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3428 bargs = &bctl->data;
3429 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3431 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3432 bargs = &bctl->meta;
3434 /* profiles filter */
3435 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3436 chunk_profiles_filter(chunk_type, bargs)) {
3441 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3442 chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3444 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3445 chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3450 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3451 chunk_devid_filter(leaf, chunk, bargs)) {
3455 /* drange filter, makes sense only with devid filter */
3456 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3457 chunk_drange_filter(leaf, chunk, bargs)) {
3462 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3463 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3467 /* stripes filter */
3468 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3469 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3473 /* soft profile changing mode */
3474 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3475 chunk_soft_convert_filter(chunk_type, bargs)) {
3480 * limited by count, must be the last filter
3482 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3483 if (bargs->limit == 0)
3487 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3489 * Same logic as the 'limit' filter; the minimum cannot be
3490 * determined here because we do not have the global information
3491 * about the count of all chunks that satisfy the filters.
3493 if (bargs->limit_max == 0)
3502 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3504 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3505 struct btrfs_root *chunk_root = fs_info->chunk_root;
3506 struct btrfs_root *dev_root = fs_info->dev_root;
3507 struct list_head *devices;
3508 struct btrfs_device *device;
3512 struct btrfs_chunk *chunk;
3513 struct btrfs_path *path = NULL;
3514 struct btrfs_key key;
3515 struct btrfs_key found_key;
3516 struct btrfs_trans_handle *trans;
3517 struct extent_buffer *leaf;
3520 int enospc_errors = 0;
3521 bool counting = true;
3522 /* The single value limit and min/max limits use the same bytes in the */
3523 u64 limit_data = bctl->data.limit;
3524 u64 limit_meta = bctl->meta.limit;
3525 u64 limit_sys = bctl->sys.limit;
3529 int chunk_reserved = 0;
3531 /* step one make some room on all the devices */
3532 devices = &fs_info->fs_devices->devices;
3533 list_for_each_entry(device, devices, dev_list) {
3534 old_size = btrfs_device_get_total_bytes(device);
3535 size_to_free = div_factor(old_size, 1);
3536 size_to_free = min_t(u64, size_to_free, SZ_1M);
3537 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) ||
3538 btrfs_device_get_total_bytes(device) -
3539 btrfs_device_get_bytes_used(device) > size_to_free ||
3540 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
3543 ret = btrfs_shrink_device(device, old_size - size_to_free);
3547 /* btrfs_shrink_device never returns ret > 0 */
3552 trans = btrfs_start_transaction(dev_root, 0);
3553 if (IS_ERR(trans)) {
3554 ret = PTR_ERR(trans);
3555 btrfs_info_in_rcu(fs_info,
3556 "resize: unable to start transaction after shrinking device %s (error %d), old size %llu, new size %llu",
3557 rcu_str_deref(device->name), ret,
3558 old_size, old_size - size_to_free);
3562 ret = btrfs_grow_device(trans, device, old_size);
3564 btrfs_end_transaction(trans);
3565 /* btrfs_grow_device never returns ret > 0 */
3567 btrfs_info_in_rcu(fs_info,
3568 "resize: unable to grow device after shrinking device %s (error %d), old size %llu, new size %llu",
3569 rcu_str_deref(device->name), ret,
3570 old_size, old_size - size_to_free);
3574 btrfs_end_transaction(trans);
3577 /* step two, relocate all the chunks */
3578 path = btrfs_alloc_path();
3584 /* zero out stat counters */
3585 spin_lock(&fs_info->balance_lock);
3586 memset(&bctl->stat, 0, sizeof(bctl->stat));
3587 spin_unlock(&fs_info->balance_lock);
3591 * The single value limit and min/max limits use the same bytes
3594 bctl->data.limit = limit_data;
3595 bctl->meta.limit = limit_meta;
3596 bctl->sys.limit = limit_sys;
3598 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3599 key.offset = (u64)-1;
3600 key.type = BTRFS_CHUNK_ITEM_KEY;
3603 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3604 atomic_read(&fs_info->balance_cancel_req)) {
3609 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3610 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3612 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3617 * this shouldn't happen, it means the last relocate
3621 BUG(); /* FIXME break ? */
3623 ret = btrfs_previous_item(chunk_root, path, 0,
3624 BTRFS_CHUNK_ITEM_KEY);
3626 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3631 leaf = path->nodes[0];
3632 slot = path->slots[0];
3633 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3635 if (found_key.objectid != key.objectid) {
3636 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3640 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3641 chunk_type = btrfs_chunk_type(leaf, chunk);
3644 spin_lock(&fs_info->balance_lock);
3645 bctl->stat.considered++;
3646 spin_unlock(&fs_info->balance_lock);
3649 ret = should_balance_chunk(fs_info, leaf, chunk,
3652 btrfs_release_path(path);
3654 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3659 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3660 spin_lock(&fs_info->balance_lock);
3661 bctl->stat.expected++;
3662 spin_unlock(&fs_info->balance_lock);
3664 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3666 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3668 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3675 * Apply limit_min filter, no need to check if the LIMITS
3676 * filter is used, limit_min is 0 by default
3678 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3679 count_data < bctl->data.limit_min)
3680 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3681 count_meta < bctl->meta.limit_min)
3682 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3683 count_sys < bctl->sys.limit_min)) {
3684 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3688 if (!chunk_reserved) {
3690 * We may be relocating the only data chunk we have,
3691 * which could potentially end up with losing data's
3692 * raid profile, so lets allocate an empty one in
3695 ret = btrfs_may_alloc_data_chunk(fs_info,
3698 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3700 } else if (ret == 1) {
3705 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3706 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3707 if (ret && ret != -ENOSPC)
3709 if (ret == -ENOSPC) {
3712 spin_lock(&fs_info->balance_lock);
3713 bctl->stat.completed++;
3714 spin_unlock(&fs_info->balance_lock);
3717 if (found_key.offset == 0)
3719 key.offset = found_key.offset - 1;
3723 btrfs_release_path(path);
3728 btrfs_free_path(path);
3729 if (enospc_errors) {
3730 btrfs_info(fs_info, "%d enospc errors during balance",
3740 * alloc_profile_is_valid - see if a given profile is valid and reduced
3741 * @flags: profile to validate
3742 * @extended: if true @flags is treated as an extended profile
3744 static int alloc_profile_is_valid(u64 flags, int extended)
3746 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3747 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3749 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3751 /* 1) check that all other bits are zeroed */
3755 /* 2) see if profile is reduced */
3757 return !extended; /* "0" is valid for usual profiles */
3759 /* true if exactly one bit set */
3760 return (flags & (flags - 1)) == 0;
3763 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3765 /* cancel requested || normal exit path */
3766 return atomic_read(&fs_info->balance_cancel_req) ||
3767 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3768 atomic_read(&fs_info->balance_cancel_req) == 0);
3771 /* Non-zero return value signifies invalidity */
3772 static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg,
3775 return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3776 (!alloc_profile_is_valid(bctl_arg->target, 1) ||
3777 (bctl_arg->target & ~allowed)));
3781 * Should be called with balance mutexe held
3783 int btrfs_balance(struct btrfs_balance_control *bctl,
3784 struct btrfs_ioctl_balance_args *bargs)
3786 struct btrfs_fs_info *fs_info = bctl->fs_info;
3787 u64 meta_target, data_target;
3794 if (btrfs_fs_closing(fs_info) ||
3795 atomic_read(&fs_info->balance_pause_req) ||
3796 atomic_read(&fs_info->balance_cancel_req)) {
3801 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3802 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3806 * In case of mixed groups both data and meta should be picked,
3807 * and identical options should be given for both of them.
3809 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3810 if (mixed && (bctl->flags & allowed)) {
3811 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3812 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3813 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3815 "with mixed groups data and metadata balance options must be the same");
3821 num_devices = fs_info->fs_devices->num_devices;
3822 btrfs_dev_replace_read_lock(&fs_info->dev_replace);
3823 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3824 BUG_ON(num_devices < 1);
3827 btrfs_dev_replace_read_unlock(&fs_info->dev_replace);
3828 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE | BTRFS_BLOCK_GROUP_DUP;
3829 if (num_devices > 1)
3830 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3831 if (num_devices > 2)
3832 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3833 if (num_devices > 3)
3834 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3835 BTRFS_BLOCK_GROUP_RAID6);
3836 if (validate_convert_profile(&bctl->data, allowed)) {
3838 "unable to start balance with target data profile %llu",
3843 if (validate_convert_profile(&bctl->meta, allowed)) {
3845 "unable to start balance with target metadata profile %llu",
3850 if (validate_convert_profile(&bctl->sys, allowed)) {
3852 "unable to start balance with target system profile %llu",
3858 /* allow to reduce meta or sys integrity only if force set */
3859 allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3860 BTRFS_BLOCK_GROUP_RAID10 |
3861 BTRFS_BLOCK_GROUP_RAID5 |
3862 BTRFS_BLOCK_GROUP_RAID6;
3864 seq = read_seqbegin(&fs_info->profiles_lock);
3866 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3867 (fs_info->avail_system_alloc_bits & allowed) &&
3868 !(bctl->sys.target & allowed)) ||
3869 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3870 (fs_info->avail_metadata_alloc_bits & allowed) &&
3871 !(bctl->meta.target & allowed))) {
3872 if (bctl->flags & BTRFS_BALANCE_FORCE) {
3874 "force reducing metadata integrity");
3877 "balance will reduce metadata integrity, use force if you want this");
3882 } while (read_seqretry(&fs_info->profiles_lock, seq));
3884 /* if we're not converting, the target field is uninitialized */
3885 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
3886 bctl->meta.target : fs_info->avail_metadata_alloc_bits;
3887 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
3888 bctl->data.target : fs_info->avail_data_alloc_bits;
3889 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <
3890 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {
3892 "metadata profile 0x%llx has lower redundancy than data profile 0x%llx",
3893 meta_target, data_target);
3896 ret = insert_balance_item(fs_info, bctl);
3897 if (ret && ret != -EEXIST)
3900 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3901 BUG_ON(ret == -EEXIST);
3902 set_balance_control(bctl);
3904 BUG_ON(ret != -EEXIST);
3905 spin_lock(&fs_info->balance_lock);
3906 update_balance_args(bctl);
3907 spin_unlock(&fs_info->balance_lock);
3910 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
3911 set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
3912 mutex_unlock(&fs_info->balance_mutex);
3914 ret = __btrfs_balance(fs_info);
3916 mutex_lock(&fs_info->balance_mutex);
3917 clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
3920 memset(bargs, 0, sizeof(*bargs));
3921 btrfs_update_ioctl_balance_args(fs_info, bargs);
3924 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3925 balance_need_close(fs_info)) {
3926 reset_balance_state(fs_info);
3927 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3930 wake_up(&fs_info->balance_wait_q);
3934 if (bctl->flags & BTRFS_BALANCE_RESUME)
3935 reset_balance_state(fs_info);
3938 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3943 static int balance_kthread(void *data)
3945 struct btrfs_fs_info *fs_info = data;
3948 mutex_lock(&fs_info->balance_mutex);
3949 if (fs_info->balance_ctl) {
3950 btrfs_info(fs_info, "continuing balance");
3951 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3953 mutex_unlock(&fs_info->balance_mutex);
3958 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3960 struct task_struct *tsk;
3962 mutex_lock(&fs_info->balance_mutex);
3963 if (!fs_info->balance_ctl) {
3964 mutex_unlock(&fs_info->balance_mutex);
3967 mutex_unlock(&fs_info->balance_mutex);
3969 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
3970 btrfs_info(fs_info, "force skipping balance");
3975 * A ro->rw remount sequence should continue with the paused balance
3976 * regardless of who pauses it, system or the user as of now, so set
3979 spin_lock(&fs_info->balance_lock);
3980 fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME;
3981 spin_unlock(&fs_info->balance_lock);
3983 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3984 return PTR_ERR_OR_ZERO(tsk);
3987 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3989 struct btrfs_balance_control *bctl;
3990 struct btrfs_balance_item *item;
3991 struct btrfs_disk_balance_args disk_bargs;
3992 struct btrfs_path *path;
3993 struct extent_buffer *leaf;
3994 struct btrfs_key key;
3997 path = btrfs_alloc_path();
4001 key.objectid = BTRFS_BALANCE_OBJECTID;
4002 key.type = BTRFS_TEMPORARY_ITEM_KEY;
4005 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4008 if (ret > 0) { /* ret = -ENOENT; */
4013 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4019 leaf = path->nodes[0];
4020 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
4022 bctl->fs_info = fs_info;
4023 bctl->flags = btrfs_balance_flags(leaf, item);
4024 bctl->flags |= BTRFS_BALANCE_RESUME;
4026 btrfs_balance_data(leaf, item, &disk_bargs);
4027 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4028 btrfs_balance_meta(leaf, item, &disk_bargs);
4029 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4030 btrfs_balance_sys(leaf, item, &disk_bargs);
4031 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4034 * This should never happen, as the paused balance state is recovered
4035 * during mount without any chance of other exclusive ops to collide.
4037 * This gives the exclusive op status to balance and keeps in paused
4038 * state until user intervention (cancel or umount). If the ownership
4039 * cannot be assigned, show a message but do not fail. The balance
4040 * is in a paused state and must have fs_info::balance_ctl properly
4043 if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags))
4045 "cannot set exclusive op status to balance, resume manually");
4047 mutex_lock(&fs_info->balance_mutex);
4048 set_balance_control(bctl);
4049 mutex_unlock(&fs_info->balance_mutex);
4051 btrfs_free_path(path);
4055 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4059 mutex_lock(&fs_info->balance_mutex);
4060 if (!fs_info->balance_ctl) {
4061 mutex_unlock(&fs_info->balance_mutex);
4065 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4066 atomic_inc(&fs_info->balance_pause_req);
4067 mutex_unlock(&fs_info->balance_mutex);
4069 wait_event(fs_info->balance_wait_q,
4070 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4072 mutex_lock(&fs_info->balance_mutex);
4073 /* we are good with balance_ctl ripped off from under us */
4074 BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4075 atomic_dec(&fs_info->balance_pause_req);
4080 mutex_unlock(&fs_info->balance_mutex);
4084 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4086 mutex_lock(&fs_info->balance_mutex);
4087 if (!fs_info->balance_ctl) {
4088 mutex_unlock(&fs_info->balance_mutex);
4093 * A paused balance with the item stored on disk can be resumed at
4094 * mount time if the mount is read-write. Otherwise it's still paused
4095 * and we must not allow cancelling as it deletes the item.
4097 if (sb_rdonly(fs_info->sb)) {
4098 mutex_unlock(&fs_info->balance_mutex);
4102 atomic_inc(&fs_info->balance_cancel_req);
4104 * if we are running just wait and return, balance item is
4105 * deleted in btrfs_balance in this case
4107 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4108 mutex_unlock(&fs_info->balance_mutex);
4109 wait_event(fs_info->balance_wait_q,
4110 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4111 mutex_lock(&fs_info->balance_mutex);
4113 mutex_unlock(&fs_info->balance_mutex);
4115 * Lock released to allow other waiters to continue, we'll
4116 * reexamine the status again.
4118 mutex_lock(&fs_info->balance_mutex);
4120 if (fs_info->balance_ctl) {
4121 reset_balance_state(fs_info);
4122 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4126 BUG_ON(fs_info->balance_ctl ||
4127 test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4128 atomic_dec(&fs_info->balance_cancel_req);
4129 mutex_unlock(&fs_info->balance_mutex);
4133 static int btrfs_uuid_scan_kthread(void *data)
4135 struct btrfs_fs_info *fs_info = data;
4136 struct btrfs_root *root = fs_info->tree_root;
4137 struct btrfs_key key;
4138 struct btrfs_path *path = NULL;
4140 struct extent_buffer *eb;
4142 struct btrfs_root_item root_item;
4144 struct btrfs_trans_handle *trans = NULL;
4146 path = btrfs_alloc_path();
4153 key.type = BTRFS_ROOT_ITEM_KEY;
4157 ret = btrfs_search_forward(root, &key, path,
4158 BTRFS_OLDEST_GENERATION);
4165 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4166 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4167 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4168 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4171 eb = path->nodes[0];
4172 slot = path->slots[0];
4173 item_size = btrfs_item_size_nr(eb, slot);
4174 if (item_size < sizeof(root_item))
4177 read_extent_buffer(eb, &root_item,
4178 btrfs_item_ptr_offset(eb, slot),
4179 (int)sizeof(root_item));
4180 if (btrfs_root_refs(&root_item) == 0)
4183 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4184 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4188 btrfs_release_path(path);
4190 * 1 - subvol uuid item
4191 * 1 - received_subvol uuid item
4193 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4194 if (IS_ERR(trans)) {
4195 ret = PTR_ERR(trans);
4203 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4204 ret = btrfs_uuid_tree_add(trans, fs_info,
4206 BTRFS_UUID_KEY_SUBVOL,
4209 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4215 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4216 ret = btrfs_uuid_tree_add(trans, fs_info,
4217 root_item.received_uuid,
4218 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4221 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4229 ret = btrfs_end_transaction(trans);
4235 btrfs_release_path(path);
4236 if (key.offset < (u64)-1) {
4238 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4240 key.type = BTRFS_ROOT_ITEM_KEY;
4241 } else if (key.objectid < (u64)-1) {
4243 key.type = BTRFS_ROOT_ITEM_KEY;
4252 btrfs_free_path(path);
4253 if (trans && !IS_ERR(trans))
4254 btrfs_end_transaction(trans);
4256 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4258 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4259 up(&fs_info->uuid_tree_rescan_sem);
4264 * Callback for btrfs_uuid_tree_iterate().
4266 * 0 check succeeded, the entry is not outdated.
4267 * < 0 if an error occurred.
4268 * > 0 if the check failed, which means the caller shall remove the entry.
4270 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
4271 u8 *uuid, u8 type, u64 subid)
4273 struct btrfs_key key;
4275 struct btrfs_root *subvol_root;
4277 if (type != BTRFS_UUID_KEY_SUBVOL &&
4278 type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
4281 key.objectid = subid;
4282 key.type = BTRFS_ROOT_ITEM_KEY;
4283 key.offset = (u64)-1;
4284 subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
4285 if (IS_ERR(subvol_root)) {
4286 ret = PTR_ERR(subvol_root);
4293 case BTRFS_UUID_KEY_SUBVOL:
4294 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
4297 case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
4298 if (memcmp(uuid, subvol_root->root_item.received_uuid,
4308 static int btrfs_uuid_rescan_kthread(void *data)
4310 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4314 * 1st step is to iterate through the existing UUID tree and
4315 * to delete all entries that contain outdated data.
4316 * 2nd step is to add all missing entries to the UUID tree.
4318 ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4320 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4321 up(&fs_info->uuid_tree_rescan_sem);
4324 return btrfs_uuid_scan_kthread(data);
4327 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4329 struct btrfs_trans_handle *trans;
4330 struct btrfs_root *tree_root = fs_info->tree_root;
4331 struct btrfs_root *uuid_root;
4332 struct task_struct *task;
4339 trans = btrfs_start_transaction(tree_root, 2);
4341 return PTR_ERR(trans);
4343 uuid_root = btrfs_create_tree(trans, fs_info,
4344 BTRFS_UUID_TREE_OBJECTID);
4345 if (IS_ERR(uuid_root)) {
4346 ret = PTR_ERR(uuid_root);
4347 btrfs_abort_transaction(trans, ret);
4348 btrfs_end_transaction(trans);
4352 fs_info->uuid_root = uuid_root;
4354 ret = btrfs_commit_transaction(trans);
4358 down(&fs_info->uuid_tree_rescan_sem);
4359 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4361 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4362 btrfs_warn(fs_info, "failed to start uuid_scan task");
4363 up(&fs_info->uuid_tree_rescan_sem);
4364 return PTR_ERR(task);
4370 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4372 struct task_struct *task;
4374 down(&fs_info->uuid_tree_rescan_sem);
4375 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4377 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4378 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4379 up(&fs_info->uuid_tree_rescan_sem);
4380 return PTR_ERR(task);
4387 * shrinking a device means finding all of the device extents past
4388 * the new size, and then following the back refs to the chunks.
4389 * The chunk relocation code actually frees the device extent
4391 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4393 struct btrfs_fs_info *fs_info = device->fs_info;
4394 struct btrfs_root *root = fs_info->dev_root;
4395 struct btrfs_trans_handle *trans;
4396 struct btrfs_dev_extent *dev_extent = NULL;
4397 struct btrfs_path *path;
4403 bool retried = false;
4404 bool checked_pending_chunks = false;
4405 struct extent_buffer *l;
4406 struct btrfs_key key;
4407 struct btrfs_super_block *super_copy = fs_info->super_copy;
4408 u64 old_total = btrfs_super_total_bytes(super_copy);
4409 u64 old_size = btrfs_device_get_total_bytes(device);
4412 new_size = round_down(new_size, fs_info->sectorsize);
4413 diff = round_down(old_size - new_size, fs_info->sectorsize);
4415 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4418 path = btrfs_alloc_path();
4422 path->reada = READA_FORWARD;
4424 mutex_lock(&fs_info->chunk_mutex);
4426 btrfs_device_set_total_bytes(device, new_size);
4427 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4428 device->fs_devices->total_rw_bytes -= diff;
4429 atomic64_sub(diff, &fs_info->free_chunk_space);
4431 mutex_unlock(&fs_info->chunk_mutex);
4434 key.objectid = device->devid;
4435 key.offset = (u64)-1;
4436 key.type = BTRFS_DEV_EXTENT_KEY;
4439 mutex_lock(&fs_info->delete_unused_bgs_mutex);
4440 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4442 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4446 ret = btrfs_previous_item(root, path, 0, key.type);
4448 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4453 btrfs_release_path(path);
4458 slot = path->slots[0];
4459 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4461 if (key.objectid != device->devid) {
4462 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4463 btrfs_release_path(path);
4467 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4468 length = btrfs_dev_extent_length(l, dev_extent);
4470 if (key.offset + length <= new_size) {
4471 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4472 btrfs_release_path(path);
4476 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4477 btrfs_release_path(path);
4480 * We may be relocating the only data chunk we have,
4481 * which could potentially end up with losing data's
4482 * raid profile, so lets allocate an empty one in
4485 ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset);
4487 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4491 ret = btrfs_relocate_chunk(fs_info, chunk_offset);
4492 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4493 if (ret && ret != -ENOSPC)
4497 } while (key.offset-- > 0);
4499 if (failed && !retried) {
4503 } else if (failed && retried) {
4508 /* Shrinking succeeded, else we would be at "done". */
4509 trans = btrfs_start_transaction(root, 0);
4510 if (IS_ERR(trans)) {
4511 ret = PTR_ERR(trans);
4515 mutex_lock(&fs_info->chunk_mutex);
4518 * We checked in the above loop all device extents that were already in
4519 * the device tree. However before we have updated the device's
4520 * total_bytes to the new size, we might have had chunk allocations that
4521 * have not complete yet (new block groups attached to transaction
4522 * handles), and therefore their device extents were not yet in the
4523 * device tree and we missed them in the loop above. So if we have any
4524 * pending chunk using a device extent that overlaps the device range
4525 * that we can not use anymore, commit the current transaction and
4526 * repeat the search on the device tree - this way we guarantee we will
4527 * not have chunks using device extents that end beyond 'new_size'.
4529 if (!checked_pending_chunks) {
4530 u64 start = new_size;
4531 u64 len = old_size - new_size;
4533 if (contains_pending_extent(trans->transaction, device,
4535 mutex_unlock(&fs_info->chunk_mutex);
4536 checked_pending_chunks = true;
4539 ret = btrfs_commit_transaction(trans);
4546 btrfs_device_set_disk_total_bytes(device, new_size);
4547 if (list_empty(&device->resized_list))
4548 list_add_tail(&device->resized_list,
4549 &fs_info->fs_devices->resized_devices);
4551 WARN_ON(diff > old_total);
4552 btrfs_set_super_total_bytes(super_copy,
4553 round_down(old_total - diff, fs_info->sectorsize));
4554 mutex_unlock(&fs_info->chunk_mutex);
4556 /* Now btrfs_update_device() will change the on-disk size. */
4557 ret = btrfs_update_device(trans, device);
4558 btrfs_end_transaction(trans);
4560 btrfs_free_path(path);
4562 mutex_lock(&fs_info->chunk_mutex);
4563 btrfs_device_set_total_bytes(device, old_size);
4564 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
4565 device->fs_devices->total_rw_bytes += diff;
4566 atomic64_add(diff, &fs_info->free_chunk_space);
4567 mutex_unlock(&fs_info->chunk_mutex);
4572 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
4573 struct btrfs_key *key,
4574 struct btrfs_chunk *chunk, int item_size)
4576 struct btrfs_super_block *super_copy = fs_info->super_copy;
4577 struct btrfs_disk_key disk_key;
4581 mutex_lock(&fs_info->chunk_mutex);
4582 array_size = btrfs_super_sys_array_size(super_copy);
4583 if (array_size + item_size + sizeof(disk_key)
4584 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4585 mutex_unlock(&fs_info->chunk_mutex);
4589 ptr = super_copy->sys_chunk_array + array_size;
4590 btrfs_cpu_key_to_disk(&disk_key, key);
4591 memcpy(ptr, &disk_key, sizeof(disk_key));
4592 ptr += sizeof(disk_key);
4593 memcpy(ptr, chunk, item_size);
4594 item_size += sizeof(disk_key);
4595 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4596 mutex_unlock(&fs_info->chunk_mutex);
4602 * sort the devices in descending order by max_avail, total_avail
4604 static int btrfs_cmp_device_info(const void *a, const void *b)
4606 const struct btrfs_device_info *di_a = a;
4607 const struct btrfs_device_info *di_b = b;
4609 if (di_a->max_avail > di_b->max_avail)
4611 if (di_a->max_avail < di_b->max_avail)
4613 if (di_a->total_avail > di_b->total_avail)
4615 if (di_a->total_avail < di_b->total_avail)
4620 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4622 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4625 btrfs_set_fs_incompat(info, RAID56);
4628 #define BTRFS_MAX_DEVS(info) ((BTRFS_MAX_ITEM_SIZE(info) \
4629 - sizeof(struct btrfs_chunk)) \
4630 / sizeof(struct btrfs_stripe) + 1)
4632 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
4633 - 2 * sizeof(struct btrfs_disk_key) \
4634 - 2 * sizeof(struct btrfs_chunk)) \
4635 / sizeof(struct btrfs_stripe) + 1)
4637 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4638 u64 start, u64 type)
4640 struct btrfs_fs_info *info = trans->fs_info;
4641 struct btrfs_fs_devices *fs_devices = info->fs_devices;
4642 struct btrfs_device *device;
4643 struct map_lookup *map = NULL;
4644 struct extent_map_tree *em_tree;
4645 struct extent_map *em;
4646 struct btrfs_device_info *devices_info = NULL;
4648 int num_stripes; /* total number of stripes to allocate */
4649 int data_stripes; /* number of stripes that count for
4651 int sub_stripes; /* sub_stripes info for map */
4652 int dev_stripes; /* stripes per dev */
4653 int devs_max; /* max devs to use */
4654 int devs_min; /* min devs needed */
4655 int devs_increment; /* ndevs has to be a multiple of this */
4656 int ncopies; /* how many copies to data has */
4658 u64 max_stripe_size;
4667 BUG_ON(!alloc_profile_is_valid(type, 0));
4669 if (list_empty(&fs_devices->alloc_list)) {
4670 if (btrfs_test_opt(info, ENOSPC_DEBUG))
4671 btrfs_debug(info, "%s: no writable device", __func__);
4675 index = btrfs_bg_flags_to_raid_index(type);
4677 sub_stripes = btrfs_raid_array[index].sub_stripes;
4678 dev_stripes = btrfs_raid_array[index].dev_stripes;
4679 devs_max = btrfs_raid_array[index].devs_max;
4680 devs_min = btrfs_raid_array[index].devs_min;
4681 devs_increment = btrfs_raid_array[index].devs_increment;
4682 ncopies = btrfs_raid_array[index].ncopies;
4684 if (type & BTRFS_BLOCK_GROUP_DATA) {
4685 max_stripe_size = SZ_1G;
4686 max_chunk_size = 10 * max_stripe_size;
4688 devs_max = BTRFS_MAX_DEVS(info);
4689 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4690 /* for larger filesystems, use larger metadata chunks */
4691 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
4692 max_stripe_size = SZ_1G;
4694 max_stripe_size = SZ_256M;
4695 max_chunk_size = max_stripe_size;
4697 devs_max = BTRFS_MAX_DEVS(info);
4698 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4699 max_stripe_size = SZ_32M;
4700 max_chunk_size = 2 * max_stripe_size;
4702 devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4704 btrfs_err(info, "invalid chunk type 0x%llx requested",
4709 /* we don't want a chunk larger than 10% of writeable space */
4710 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4713 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4719 * in the first pass through the devices list, we gather information
4720 * about the available holes on each device.
4723 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
4727 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4729 "BTRFS: read-only device in alloc_list\n");
4733 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
4734 &device->dev_state) ||
4735 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4738 if (device->total_bytes > device->bytes_used)
4739 total_avail = device->total_bytes - device->bytes_used;
4743 /* If there is no space on this device, skip it. */
4744 if (total_avail == 0)
4747 ret = find_free_dev_extent(trans, device,
4748 max_stripe_size * dev_stripes,
4749 &dev_offset, &max_avail);
4750 if (ret && ret != -ENOSPC)
4754 max_avail = max_stripe_size * dev_stripes;
4756 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes) {
4757 if (btrfs_test_opt(info, ENOSPC_DEBUG))
4759 "%s: devid %llu has no free space, have=%llu want=%u",
4760 __func__, device->devid, max_avail,
4761 BTRFS_STRIPE_LEN * dev_stripes);
4765 if (ndevs == fs_devices->rw_devices) {
4766 WARN(1, "%s: found more than %llu devices\n",
4767 __func__, fs_devices->rw_devices);
4770 devices_info[ndevs].dev_offset = dev_offset;
4771 devices_info[ndevs].max_avail = max_avail;
4772 devices_info[ndevs].total_avail = total_avail;
4773 devices_info[ndevs].dev = device;
4778 * now sort the devices by hole size / available space
4780 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4781 btrfs_cmp_device_info, NULL);
4783 /* round down to number of usable stripes */
4784 ndevs = round_down(ndevs, devs_increment);
4786 if (ndevs < devs_min) {
4788 if (btrfs_test_opt(info, ENOSPC_DEBUG)) {
4790 "%s: not enough devices with free space: have=%d minimum required=%d",
4791 __func__, ndevs, devs_min);
4796 ndevs = min(ndevs, devs_max);
4799 * The primary goal is to maximize the number of stripes, so use as
4800 * many devices as possible, even if the stripes are not maximum sized.
4802 * The DUP profile stores more than one stripe per device, the
4803 * max_avail is the total size so we have to adjust.
4805 stripe_size = div_u64(devices_info[ndevs - 1].max_avail, dev_stripes);
4806 num_stripes = ndevs * dev_stripes;
4809 * this will have to be fixed for RAID1 and RAID10 over
4812 data_stripes = num_stripes / ncopies;
4814 if (type & BTRFS_BLOCK_GROUP_RAID5)
4815 data_stripes = num_stripes - 1;
4817 if (type & BTRFS_BLOCK_GROUP_RAID6)
4818 data_stripes = num_stripes - 2;
4821 * Use the number of data stripes to figure out how big this chunk
4822 * is really going to be in terms of logical address space,
4823 * and compare that answer with the max chunk size
4825 if (stripe_size * data_stripes > max_chunk_size) {
4826 stripe_size = div_u64(max_chunk_size, data_stripes);
4828 /* bump the answer up to a 16MB boundary */
4829 stripe_size = round_up(stripe_size, SZ_16M);
4832 * But don't go higher than the limits we found while searching
4835 stripe_size = min(devices_info[ndevs - 1].max_avail,
4839 /* align to BTRFS_STRIPE_LEN */
4840 stripe_size = round_down(stripe_size, BTRFS_STRIPE_LEN);
4842 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4847 map->num_stripes = num_stripes;
4849 for (i = 0; i < ndevs; ++i) {
4850 for (j = 0; j < dev_stripes; ++j) {
4851 int s = i * dev_stripes + j;
4852 map->stripes[s].dev = devices_info[i].dev;
4853 map->stripes[s].physical = devices_info[i].dev_offset +
4857 map->stripe_len = BTRFS_STRIPE_LEN;
4858 map->io_align = BTRFS_STRIPE_LEN;
4859 map->io_width = BTRFS_STRIPE_LEN;
4861 map->sub_stripes = sub_stripes;
4863 num_bytes = stripe_size * data_stripes;
4865 trace_btrfs_chunk_alloc(info, map, start, num_bytes);
4867 em = alloc_extent_map();
4873 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4874 em->map_lookup = map;
4876 em->len = num_bytes;
4877 em->block_start = 0;
4878 em->block_len = em->len;
4879 em->orig_block_len = stripe_size;
4881 em_tree = &info->mapping_tree.map_tree;
4882 write_lock(&em_tree->lock);
4883 ret = add_extent_mapping(em_tree, em, 0);
4885 write_unlock(&em_tree->lock);
4886 free_extent_map(em);
4890 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4891 refcount_inc(&em->refs);
4892 write_unlock(&em_tree->lock);
4894 ret = btrfs_make_block_group(trans, info, 0, type, start, num_bytes);
4896 goto error_del_extent;
4898 for (i = 0; i < map->num_stripes; i++) {
4899 num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4900 btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4903 atomic64_sub(stripe_size * map->num_stripes, &info->free_chunk_space);
4905 free_extent_map(em);
4906 check_raid56_incompat_flag(info, type);
4908 kfree(devices_info);
4912 write_lock(&em_tree->lock);
4913 remove_extent_mapping(em_tree, em);
4914 write_unlock(&em_tree->lock);
4916 /* One for our allocation */
4917 free_extent_map(em);
4918 /* One for the tree reference */
4919 free_extent_map(em);
4920 /* One for the pending_chunks list reference */
4921 free_extent_map(em);
4923 kfree(devices_info);
4927 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4928 struct btrfs_fs_info *fs_info,
4929 u64 chunk_offset, u64 chunk_size)
4931 struct btrfs_root *extent_root = fs_info->extent_root;
4932 struct btrfs_root *chunk_root = fs_info->chunk_root;
4933 struct btrfs_key key;
4934 struct btrfs_device *device;
4935 struct btrfs_chunk *chunk;
4936 struct btrfs_stripe *stripe;
4937 struct extent_map *em;
4938 struct map_lookup *map;
4945 em = get_chunk_map(fs_info, chunk_offset, chunk_size);
4949 map = em->map_lookup;
4950 item_size = btrfs_chunk_item_size(map->num_stripes);
4951 stripe_size = em->orig_block_len;
4953 chunk = kzalloc(item_size, GFP_NOFS);
4960 * Take the device list mutex to prevent races with the final phase of
4961 * a device replace operation that replaces the device object associated
4962 * with the map's stripes, because the device object's id can change
4963 * at any time during that final phase of the device replace operation
4964 * (dev-replace.c:btrfs_dev_replace_finishing()).
4966 mutex_lock(&fs_info->fs_devices->device_list_mutex);
4967 for (i = 0; i < map->num_stripes; i++) {
4968 device = map->stripes[i].dev;
4969 dev_offset = map->stripes[i].physical;
4971 ret = btrfs_update_device(trans, device);
4974 ret = btrfs_alloc_dev_extent(trans, device, chunk_offset,
4975 dev_offset, stripe_size);
4980 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4984 stripe = &chunk->stripe;
4985 for (i = 0; i < map->num_stripes; i++) {
4986 device = map->stripes[i].dev;
4987 dev_offset = map->stripes[i].physical;
4989 btrfs_set_stack_stripe_devid(stripe, device->devid);
4990 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4991 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4994 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4996 btrfs_set_stack_chunk_length(chunk, chunk_size);
4997 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4998 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4999 btrfs_set_stack_chunk_type(chunk, map->type);
5000 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
5001 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
5002 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
5003 btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
5004 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
5006 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
5007 key.type = BTRFS_CHUNK_ITEM_KEY;
5008 key.offset = chunk_offset;
5010 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
5011 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
5013 * TODO: Cleanup of inserted chunk root in case of
5016 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
5021 free_extent_map(em);
5026 * Chunk allocation falls into two parts. The first part does works
5027 * that make the new allocated chunk useable, but not do any operation
5028 * that modifies the chunk tree. The second part does the works that
5029 * require modifying the chunk tree. This division is important for the
5030 * bootstrap process of adding storage to a seed btrfs.
5032 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
5033 struct btrfs_fs_info *fs_info, u64 type)
5037 lockdep_assert_held(&fs_info->chunk_mutex);
5038 chunk_offset = find_next_chunk(fs_info);
5039 return __btrfs_alloc_chunk(trans, chunk_offset, type);
5042 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
5043 struct btrfs_fs_info *fs_info)
5046 u64 sys_chunk_offset;
5050 chunk_offset = find_next_chunk(fs_info);
5051 alloc_profile = btrfs_metadata_alloc_profile(fs_info);
5052 ret = __btrfs_alloc_chunk(trans, chunk_offset, alloc_profile);
5056 sys_chunk_offset = find_next_chunk(fs_info);
5057 alloc_profile = btrfs_system_alloc_profile(fs_info);
5058 ret = __btrfs_alloc_chunk(trans, sys_chunk_offset, alloc_profile);
5062 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
5066 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5067 BTRFS_BLOCK_GROUP_RAID10 |
5068 BTRFS_BLOCK_GROUP_RAID5 |
5069 BTRFS_BLOCK_GROUP_DUP)) {
5071 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
5080 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset)
5082 struct extent_map *em;
5083 struct map_lookup *map;
5088 em = get_chunk_map(fs_info, chunk_offset, 1);
5092 map = em->map_lookup;
5093 for (i = 0; i < map->num_stripes; i++) {
5094 if (test_bit(BTRFS_DEV_STATE_MISSING,
5095 &map->stripes[i].dev->dev_state)) {
5099 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE,
5100 &map->stripes[i].dev->dev_state)) {
5107 * If the number of missing devices is larger than max errors,
5108 * we can not write the data into that chunk successfully, so
5111 if (miss_ndevs > btrfs_chunk_max_errors(map))
5114 free_extent_map(em);
5118 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
5120 extent_map_tree_init(&tree->map_tree);
5123 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
5125 struct extent_map *em;
5128 write_lock(&tree->map_tree.lock);
5129 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
5131 remove_extent_mapping(&tree->map_tree, em);
5132 write_unlock(&tree->map_tree.lock);
5136 free_extent_map(em);
5137 /* once for the tree */
5138 free_extent_map(em);
5142 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5144 struct extent_map *em;
5145 struct map_lookup *map;
5148 em = get_chunk_map(fs_info, logical, len);
5151 * We could return errors for these cases, but that could get
5152 * ugly and we'd probably do the same thing which is just not do
5153 * anything else and exit, so return 1 so the callers don't try
5154 * to use other copies.
5158 map = em->map_lookup;
5159 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
5160 ret = map->num_stripes;
5161 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5162 ret = map->sub_stripes;
5163 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5165 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5167 * There could be two corrupted data stripes, we need
5168 * to loop retry in order to rebuild the correct data.
5170 * Fail a stripe at a time on every retry except the
5171 * stripe under reconstruction.
5173 ret = map->num_stripes;
5176 free_extent_map(em);
5178 btrfs_dev_replace_read_lock(&fs_info->dev_replace);
5179 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) &&
5180 fs_info->dev_replace.tgtdev)
5182 btrfs_dev_replace_read_unlock(&fs_info->dev_replace);
5187 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
5190 struct extent_map *em;
5191 struct map_lookup *map;
5192 unsigned long len = fs_info->sectorsize;
5194 em = get_chunk_map(fs_info, logical, len);
5196 if (!WARN_ON(IS_ERR(em))) {
5197 map = em->map_lookup;
5198 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5199 len = map->stripe_len * nr_data_stripes(map);
5200 free_extent_map(em);
5205 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5207 struct extent_map *em;
5208 struct map_lookup *map;
5211 em = get_chunk_map(fs_info, logical, len);
5213 if(!WARN_ON(IS_ERR(em))) {
5214 map = em->map_lookup;
5215 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5217 free_extent_map(em);
5222 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5223 struct map_lookup *map, int first,
5224 int dev_replace_is_ongoing)
5228 int preferred_mirror;
5230 struct btrfs_device *srcdev;
5233 (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)));
5235 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5236 num_stripes = map->sub_stripes;
5238 num_stripes = map->num_stripes;
5240 preferred_mirror = first + current->pid % num_stripes;
5242 if (dev_replace_is_ongoing &&
5243 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5244 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5245 srcdev = fs_info->dev_replace.srcdev;
5250 * try to avoid the drive that is the source drive for a
5251 * dev-replace procedure, only choose it if no other non-missing
5252 * mirror is available
5254 for (tolerance = 0; tolerance < 2; tolerance++) {
5255 if (map->stripes[preferred_mirror].dev->bdev &&
5256 (tolerance || map->stripes[preferred_mirror].dev != srcdev))
5257 return preferred_mirror;
5258 for (i = first; i < first + num_stripes; i++) {
5259 if (map->stripes[i].dev->bdev &&
5260 (tolerance || map->stripes[i].dev != srcdev))
5265 /* we couldn't find one that doesn't fail. Just return something
5266 * and the io error handling code will clean up eventually
5268 return preferred_mirror;
5271 static inline int parity_smaller(u64 a, u64 b)
5276 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5277 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5279 struct btrfs_bio_stripe s;
5286 for (i = 0; i < num_stripes - 1; i++) {
5287 if (parity_smaller(bbio->raid_map[i],
5288 bbio->raid_map[i+1])) {
5289 s = bbio->stripes[i];
5290 l = bbio->raid_map[i];
5291 bbio->stripes[i] = bbio->stripes[i+1];
5292 bbio->raid_map[i] = bbio->raid_map[i+1];
5293 bbio->stripes[i+1] = s;
5294 bbio->raid_map[i+1] = l;
5302 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5304 struct btrfs_bio *bbio = kzalloc(
5305 /* the size of the btrfs_bio */
5306 sizeof(struct btrfs_bio) +
5307 /* plus the variable array for the stripes */
5308 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5309 /* plus the variable array for the tgt dev */
5310 sizeof(int) * (real_stripes) +
5312 * plus the raid_map, which includes both the tgt dev
5315 sizeof(u64) * (total_stripes),
5316 GFP_NOFS|__GFP_NOFAIL);
5318 atomic_set(&bbio->error, 0);
5319 refcount_set(&bbio->refs, 1);
5324 void btrfs_get_bbio(struct btrfs_bio *bbio)
5326 WARN_ON(!refcount_read(&bbio->refs));
5327 refcount_inc(&bbio->refs);
5330 void btrfs_put_bbio(struct btrfs_bio *bbio)
5334 if (refcount_dec_and_test(&bbio->refs))
5338 /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */
5340 * Please note that, discard won't be sent to target device of device
5343 static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info,
5344 u64 logical, u64 length,
5345 struct btrfs_bio **bbio_ret)
5347 struct extent_map *em;
5348 struct map_lookup *map;
5349 struct btrfs_bio *bbio;
5353 u64 stripe_end_offset;
5360 u32 sub_stripes = 0;
5361 u64 stripes_per_dev = 0;
5362 u32 remaining_stripes = 0;
5363 u32 last_stripe = 0;
5367 /* discard always return a bbio */
5370 em = get_chunk_map(fs_info, logical, length);
5374 map = em->map_lookup;
5375 /* we don't discard raid56 yet */
5376 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5381 offset = logical - em->start;
5382 length = min_t(u64, em->len - offset, length);
5384 stripe_len = map->stripe_len;
5386 * stripe_nr counts the total number of stripes we have to stride
5387 * to get to this block
5389 stripe_nr = div64_u64(offset, stripe_len);
5391 /* stripe_offset is the offset of this block in its stripe */
5392 stripe_offset = offset - stripe_nr * stripe_len;
5394 stripe_nr_end = round_up(offset + length, map->stripe_len);
5395 stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len);
5396 stripe_cnt = stripe_nr_end - stripe_nr;
5397 stripe_end_offset = stripe_nr_end * map->stripe_len -
5400 * after this, stripe_nr is the number of stripes on this
5401 * device we have to walk to find the data, and stripe_index is
5402 * the number of our device in the stripe array
5406 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5407 BTRFS_BLOCK_GROUP_RAID10)) {
5408 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5411 sub_stripes = map->sub_stripes;
5413 factor = map->num_stripes / sub_stripes;
5414 num_stripes = min_t(u64, map->num_stripes,
5415 sub_stripes * stripe_cnt);
5416 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5417 stripe_index *= sub_stripes;
5418 stripes_per_dev = div_u64_rem(stripe_cnt, factor,
5419 &remaining_stripes);
5420 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5421 last_stripe *= sub_stripes;
5422 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5423 BTRFS_BLOCK_GROUP_DUP)) {
5424 num_stripes = map->num_stripes;
5426 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5430 bbio = alloc_btrfs_bio(num_stripes, 0);
5436 for (i = 0; i < num_stripes; i++) {
5437 bbio->stripes[i].physical =
5438 map->stripes[stripe_index].physical +
5439 stripe_offset + stripe_nr * map->stripe_len;
5440 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5442 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5443 BTRFS_BLOCK_GROUP_RAID10)) {
5444 bbio->stripes[i].length = stripes_per_dev *
5447 if (i / sub_stripes < remaining_stripes)
5448 bbio->stripes[i].length +=
5452 * Special for the first stripe and
5455 * |-------|...|-------|
5459 if (i < sub_stripes)
5460 bbio->stripes[i].length -=
5463 if (stripe_index >= last_stripe &&
5464 stripe_index <= (last_stripe +
5466 bbio->stripes[i].length -=
5469 if (i == sub_stripes - 1)
5472 bbio->stripes[i].length = length;
5476 if (stripe_index == map->num_stripes) {
5483 bbio->map_type = map->type;
5484 bbio->num_stripes = num_stripes;
5486 free_extent_map(em);
5491 * In dev-replace case, for repair case (that's the only case where the mirror
5492 * is selected explicitly when calling btrfs_map_block), blocks left of the
5493 * left cursor can also be read from the target drive.
5495 * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the
5497 * For READ, it also needs to be supported using the same mirror number.
5499 * If the requested block is not left of the left cursor, EIO is returned. This
5500 * can happen because btrfs_num_copies() returns one more in the dev-replace
5503 static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info,
5504 u64 logical, u64 length,
5505 u64 srcdev_devid, int *mirror_num,
5508 struct btrfs_bio *bbio = NULL;
5510 int index_srcdev = 0;
5512 u64 physical_of_found = 0;
5516 ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS,
5517 logical, &length, &bbio, 0, 0);
5519 ASSERT(bbio == NULL);
5523 num_stripes = bbio->num_stripes;
5524 if (*mirror_num > num_stripes) {
5526 * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror,
5527 * that means that the requested area is not left of the left
5530 btrfs_put_bbio(bbio);
5535 * process the rest of the function using the mirror_num of the source
5536 * drive. Therefore look it up first. At the end, patch the device
5537 * pointer to the one of the target drive.
5539 for (i = 0; i < num_stripes; i++) {
5540 if (bbio->stripes[i].dev->devid != srcdev_devid)
5544 * In case of DUP, in order to keep it simple, only add the
5545 * mirror with the lowest physical address
5548 physical_of_found <= bbio->stripes[i].physical)
5553 physical_of_found = bbio->stripes[i].physical;
5556 btrfs_put_bbio(bbio);
5562 *mirror_num = index_srcdev + 1;
5563 *physical = physical_of_found;
5567 static void handle_ops_on_dev_replace(enum btrfs_map_op op,
5568 struct btrfs_bio **bbio_ret,
5569 struct btrfs_dev_replace *dev_replace,
5570 int *num_stripes_ret, int *max_errors_ret)
5572 struct btrfs_bio *bbio = *bbio_ret;
5573 u64 srcdev_devid = dev_replace->srcdev->devid;
5574 int tgtdev_indexes = 0;
5575 int num_stripes = *num_stripes_ret;
5576 int max_errors = *max_errors_ret;
5579 if (op == BTRFS_MAP_WRITE) {
5580 int index_where_to_add;
5583 * duplicate the write operations while the dev replace
5584 * procedure is running. Since the copying of the old disk to
5585 * the new disk takes place at run time while the filesystem is
5586 * mounted writable, the regular write operations to the old
5587 * disk have to be duplicated to go to the new disk as well.
5589 * Note that device->missing is handled by the caller, and that
5590 * the write to the old disk is already set up in the stripes
5593 index_where_to_add = num_stripes;
5594 for (i = 0; i < num_stripes; i++) {
5595 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5596 /* write to new disk, too */
5597 struct btrfs_bio_stripe *new =
5598 bbio->stripes + index_where_to_add;
5599 struct btrfs_bio_stripe *old =
5602 new->physical = old->physical;
5603 new->length = old->length;
5604 new->dev = dev_replace->tgtdev;
5605 bbio->tgtdev_map[i] = index_where_to_add;
5606 index_where_to_add++;
5611 num_stripes = index_where_to_add;
5612 } else if (op == BTRFS_MAP_GET_READ_MIRRORS) {
5613 int index_srcdev = 0;
5615 u64 physical_of_found = 0;
5618 * During the dev-replace procedure, the target drive can also
5619 * be used to read data in case it is needed to repair a corrupt
5620 * block elsewhere. This is possible if the requested area is
5621 * left of the left cursor. In this area, the target drive is a
5622 * full copy of the source drive.
5624 for (i = 0; i < num_stripes; i++) {
5625 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5627 * In case of DUP, in order to keep it simple,
5628 * only add the mirror with the lowest physical
5632 physical_of_found <=
5633 bbio->stripes[i].physical)
5637 physical_of_found = bbio->stripes[i].physical;
5641 struct btrfs_bio_stripe *tgtdev_stripe =
5642 bbio->stripes + num_stripes;
5644 tgtdev_stripe->physical = physical_of_found;
5645 tgtdev_stripe->length =
5646 bbio->stripes[index_srcdev].length;
5647 tgtdev_stripe->dev = dev_replace->tgtdev;
5648 bbio->tgtdev_map[index_srcdev] = num_stripes;
5655 *num_stripes_ret = num_stripes;
5656 *max_errors_ret = max_errors;
5657 bbio->num_tgtdevs = tgtdev_indexes;
5661 static bool need_full_stripe(enum btrfs_map_op op)
5663 return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS);
5666 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
5667 enum btrfs_map_op op,
5668 u64 logical, u64 *length,
5669 struct btrfs_bio **bbio_ret,
5670 int mirror_num, int need_raid_map)
5672 struct extent_map *em;
5673 struct map_lookup *map;
5683 int tgtdev_indexes = 0;
5684 struct btrfs_bio *bbio = NULL;
5685 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
5686 int dev_replace_is_ongoing = 0;
5687 int num_alloc_stripes;
5688 int patch_the_first_stripe_for_dev_replace = 0;
5689 u64 physical_to_patch_in_first_stripe = 0;
5690 u64 raid56_full_stripe_start = (u64)-1;
5692 if (op == BTRFS_MAP_DISCARD)
5693 return __btrfs_map_block_for_discard(fs_info, logical,
5696 em = get_chunk_map(fs_info, logical, *length);
5700 map = em->map_lookup;
5701 offset = logical - em->start;
5703 stripe_len = map->stripe_len;
5706 * stripe_nr counts the total number of stripes we have to stride
5707 * to get to this block
5709 stripe_nr = div64_u64(stripe_nr, stripe_len);
5711 stripe_offset = stripe_nr * stripe_len;
5712 if (offset < stripe_offset) {
5714 "stripe math has gone wrong, stripe_offset=%llu, offset=%llu, start=%llu, logical=%llu, stripe_len=%llu",
5715 stripe_offset, offset, em->start, logical,
5717 free_extent_map(em);
5721 /* stripe_offset is the offset of this block in its stripe*/
5722 stripe_offset = offset - stripe_offset;
5724 /* if we're here for raid56, we need to know the stripe aligned start */
5725 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5726 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5727 raid56_full_stripe_start = offset;
5729 /* allow a write of a full stripe, but make sure we don't
5730 * allow straddling of stripes
5732 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5734 raid56_full_stripe_start *= full_stripe_len;
5737 if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5739 /* For writes to RAID[56], allow a full stripeset across all disks.
5740 For other RAID types and for RAID[56] reads, just allow a single
5741 stripe (on a single disk). */
5742 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
5743 (op == BTRFS_MAP_WRITE)) {
5744 max_len = stripe_len * nr_data_stripes(map) -
5745 (offset - raid56_full_stripe_start);
5747 /* we limit the length of each bio to what fits in a stripe */
5748 max_len = stripe_len - stripe_offset;
5750 *length = min_t(u64, em->len - offset, max_len);
5752 *length = em->len - offset;
5755 /* This is for when we're called from btrfs_merge_bio_hook() and all
5756 it cares about is the length */
5760 btrfs_dev_replace_read_lock(dev_replace);
5761 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5762 if (!dev_replace_is_ongoing)
5763 btrfs_dev_replace_read_unlock(dev_replace);
5765 btrfs_dev_replace_set_lock_blocking(dev_replace);
5767 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
5768 !need_full_stripe(op) && dev_replace->tgtdev != NULL) {
5769 ret = get_extra_mirror_from_replace(fs_info, logical, *length,
5770 dev_replace->srcdev->devid,
5772 &physical_to_patch_in_first_stripe);
5776 patch_the_first_stripe_for_dev_replace = 1;
5777 } else if (mirror_num > map->num_stripes) {
5783 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5784 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5786 if (!need_full_stripe(op))
5788 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
5789 if (need_full_stripe(op))
5790 num_stripes = map->num_stripes;
5791 else if (mirror_num)
5792 stripe_index = mirror_num - 1;
5794 stripe_index = find_live_mirror(fs_info, map, 0,
5795 dev_replace_is_ongoing);
5796 mirror_num = stripe_index + 1;
5799 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
5800 if (need_full_stripe(op)) {
5801 num_stripes = map->num_stripes;
5802 } else if (mirror_num) {
5803 stripe_index = mirror_num - 1;
5808 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5809 u32 factor = map->num_stripes / map->sub_stripes;
5811 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5812 stripe_index *= map->sub_stripes;
5814 if (need_full_stripe(op))
5815 num_stripes = map->sub_stripes;
5816 else if (mirror_num)
5817 stripe_index += mirror_num - 1;
5819 int old_stripe_index = stripe_index;
5820 stripe_index = find_live_mirror(fs_info, map,
5822 dev_replace_is_ongoing);
5823 mirror_num = stripe_index - old_stripe_index + 1;
5826 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5827 if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) {
5828 /* push stripe_nr back to the start of the full stripe */
5829 stripe_nr = div64_u64(raid56_full_stripe_start,
5830 stripe_len * nr_data_stripes(map));
5832 /* RAID[56] write or recovery. Return all stripes */
5833 num_stripes = map->num_stripes;
5834 max_errors = nr_parity_stripes(map);
5836 *length = map->stripe_len;
5841 * Mirror #0 or #1 means the original data block.
5842 * Mirror #2 is RAID5 parity block.
5843 * Mirror #3 is RAID6 Q block.
5845 stripe_nr = div_u64_rem(stripe_nr,
5846 nr_data_stripes(map), &stripe_index);
5848 stripe_index = nr_data_stripes(map) +
5851 /* We distribute the parity blocks across stripes */
5852 div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
5854 if (!need_full_stripe(op) && mirror_num <= 1)
5859 * after this, stripe_nr is the number of stripes on this
5860 * device we have to walk to find the data, and stripe_index is
5861 * the number of our device in the stripe array
5863 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5865 mirror_num = stripe_index + 1;
5867 if (stripe_index >= map->num_stripes) {
5869 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
5870 stripe_index, map->num_stripes);
5875 num_alloc_stripes = num_stripes;
5876 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) {
5877 if (op == BTRFS_MAP_WRITE)
5878 num_alloc_stripes <<= 1;
5879 if (op == BTRFS_MAP_GET_READ_MIRRORS)
5880 num_alloc_stripes++;
5881 tgtdev_indexes = num_stripes;
5884 bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
5889 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL)
5890 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
5892 /* build raid_map */
5893 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map &&
5894 (need_full_stripe(op) || mirror_num > 1)) {
5898 bbio->raid_map = (u64 *)((void *)bbio->stripes +
5899 sizeof(struct btrfs_bio_stripe) *
5901 sizeof(int) * tgtdev_indexes);
5903 /* Work out the disk rotation on this stripe-set */
5904 div_u64_rem(stripe_nr, num_stripes, &rot);
5906 /* Fill in the logical address of each stripe */
5907 tmp = stripe_nr * nr_data_stripes(map);
5908 for (i = 0; i < nr_data_stripes(map); i++)
5909 bbio->raid_map[(i+rot) % num_stripes] =
5910 em->start + (tmp + i) * map->stripe_len;
5912 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5913 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5914 bbio->raid_map[(i+rot+1) % num_stripes] =
5919 for (i = 0; i < num_stripes; i++) {
5920 bbio->stripes[i].physical =
5921 map->stripes[stripe_index].physical +
5923 stripe_nr * map->stripe_len;
5924 bbio->stripes[i].dev =
5925 map->stripes[stripe_index].dev;
5929 if (need_full_stripe(op))
5930 max_errors = btrfs_chunk_max_errors(map);
5933 sort_parity_stripes(bbio, num_stripes);
5935 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
5936 need_full_stripe(op)) {
5937 handle_ops_on_dev_replace(op, &bbio, dev_replace, &num_stripes,
5942 bbio->map_type = map->type;
5943 bbio->num_stripes = num_stripes;
5944 bbio->max_errors = max_errors;
5945 bbio->mirror_num = mirror_num;
5948 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5949 * mirror_num == num_stripes + 1 && dev_replace target drive is
5950 * available as a mirror
5952 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5953 WARN_ON(num_stripes > 1);
5954 bbio->stripes[0].dev = dev_replace->tgtdev;
5955 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5956 bbio->mirror_num = map->num_stripes + 1;
5959 if (dev_replace_is_ongoing) {
5960 btrfs_dev_replace_clear_lock_blocking(dev_replace);
5961 btrfs_dev_replace_read_unlock(dev_replace);
5963 free_extent_map(em);
5967 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
5968 u64 logical, u64 *length,
5969 struct btrfs_bio **bbio_ret, int mirror_num)
5971 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret,
5975 /* For Scrub/replace */
5976 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
5977 u64 logical, u64 *length,
5978 struct btrfs_bio **bbio_ret)
5980 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1);
5983 int btrfs_rmap_block(struct btrfs_fs_info *fs_info,
5984 u64 chunk_start, u64 physical, u64 devid,
5985 u64 **logical, int *naddrs, int *stripe_len)
5987 struct extent_map *em;
5988 struct map_lookup *map;
5996 em = get_chunk_map(fs_info, chunk_start, 1);
6000 map = em->map_lookup;
6002 rmap_len = map->stripe_len;
6004 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
6005 length = div_u64(length, map->num_stripes / map->sub_stripes);
6006 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
6007 length = div_u64(length, map->num_stripes);
6008 else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6009 length = div_u64(length, nr_data_stripes(map));
6010 rmap_len = map->stripe_len * nr_data_stripes(map);
6013 buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
6014 BUG_ON(!buf); /* -ENOMEM */
6016 for (i = 0; i < map->num_stripes; i++) {
6017 if (devid && map->stripes[i].dev->devid != devid)
6019 if (map->stripes[i].physical > physical ||
6020 map->stripes[i].physical + length <= physical)
6023 stripe_nr = physical - map->stripes[i].physical;
6024 stripe_nr = div64_u64(stripe_nr, map->stripe_len);
6026 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
6027 stripe_nr = stripe_nr * map->num_stripes + i;
6028 stripe_nr = div_u64(stripe_nr, map->sub_stripes);
6029 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
6030 stripe_nr = stripe_nr * map->num_stripes + i;
6031 } /* else if RAID[56], multiply by nr_data_stripes().
6032 * Alternatively, just use rmap_len below instead of
6033 * map->stripe_len */
6035 bytenr = chunk_start + stripe_nr * rmap_len;
6036 WARN_ON(nr >= map->num_stripes);
6037 for (j = 0; j < nr; j++) {
6038 if (buf[j] == bytenr)
6042 WARN_ON(nr >= map->num_stripes);
6049 *stripe_len = rmap_len;
6051 free_extent_map(em);
6055 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
6057 bio->bi_private = bbio->private;
6058 bio->bi_end_io = bbio->end_io;
6061 btrfs_put_bbio(bbio);
6064 static void btrfs_end_bio(struct bio *bio)
6066 struct btrfs_bio *bbio = bio->bi_private;
6067 int is_orig_bio = 0;
6069 if (bio->bi_status) {
6070 atomic_inc(&bbio->error);
6071 if (bio->bi_status == BLK_STS_IOERR ||
6072 bio->bi_status == BLK_STS_TARGET) {
6073 unsigned int stripe_index =
6074 btrfs_io_bio(bio)->stripe_index;
6075 struct btrfs_device *dev;
6077 BUG_ON(stripe_index >= bbio->num_stripes);
6078 dev = bbio->stripes[stripe_index].dev;
6080 if (bio_op(bio) == REQ_OP_WRITE)
6081 btrfs_dev_stat_inc_and_print(dev,
6082 BTRFS_DEV_STAT_WRITE_ERRS);
6084 btrfs_dev_stat_inc_and_print(dev,
6085 BTRFS_DEV_STAT_READ_ERRS);
6086 if (bio->bi_opf & REQ_PREFLUSH)
6087 btrfs_dev_stat_inc_and_print(dev,
6088 BTRFS_DEV_STAT_FLUSH_ERRS);
6093 if (bio == bbio->orig_bio)
6096 btrfs_bio_counter_dec(bbio->fs_info);
6098 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6101 bio = bbio->orig_bio;
6104 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6105 /* only send an error to the higher layers if it is
6106 * beyond the tolerance of the btrfs bio
6108 if (atomic_read(&bbio->error) > bbio->max_errors) {
6109 bio->bi_status = BLK_STS_IOERR;
6112 * this bio is actually up to date, we didn't
6113 * go over the max number of errors
6115 bio->bi_status = BLK_STS_OK;
6118 btrfs_end_bbio(bbio, bio);
6119 } else if (!is_orig_bio) {
6125 * see run_scheduled_bios for a description of why bios are collected for
6128 * This will add one bio to the pending list for a device and make sure
6129 * the work struct is scheduled.
6131 static noinline void btrfs_schedule_bio(struct btrfs_device *device,
6134 struct btrfs_fs_info *fs_info = device->fs_info;
6135 int should_queue = 1;
6136 struct btrfs_pending_bios *pending_bios;
6138 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state) ||
6144 /* don't bother with additional async steps for reads, right now */
6145 if (bio_op(bio) == REQ_OP_READ) {
6146 btrfsic_submit_bio(bio);
6150 WARN_ON(bio->bi_next);
6151 bio->bi_next = NULL;
6153 spin_lock(&device->io_lock);
6154 if (op_is_sync(bio->bi_opf))
6155 pending_bios = &device->pending_sync_bios;
6157 pending_bios = &device->pending_bios;
6159 if (pending_bios->tail)
6160 pending_bios->tail->bi_next = bio;
6162 pending_bios->tail = bio;
6163 if (!pending_bios->head)
6164 pending_bios->head = bio;
6165 if (device->running_pending)
6168 spin_unlock(&device->io_lock);
6171 btrfs_queue_work(fs_info->submit_workers, &device->work);
6174 static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio,
6175 u64 physical, int dev_nr, int async)
6177 struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
6178 struct btrfs_fs_info *fs_info = bbio->fs_info;
6180 bio->bi_private = bbio;
6181 btrfs_io_bio(bio)->stripe_index = dev_nr;
6182 bio->bi_end_io = btrfs_end_bio;
6183 bio->bi_iter.bi_sector = physical >> 9;
6186 struct rcu_string *name;
6189 name = rcu_dereference(dev->name);
6190 btrfs_debug(fs_info,
6191 "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
6192 bio_op(bio), bio->bi_opf,
6193 (u64)bio->bi_iter.bi_sector,
6194 (u_long)dev->bdev->bd_dev, name->str, dev->devid,
6195 bio->bi_iter.bi_size);
6199 bio_set_dev(bio, dev->bdev);
6201 btrfs_bio_counter_inc_noblocked(fs_info);
6204 btrfs_schedule_bio(dev, bio);
6206 btrfsic_submit_bio(bio);
6209 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6211 atomic_inc(&bbio->error);
6212 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6213 /* Should be the original bio. */
6214 WARN_ON(bio != bbio->orig_bio);
6216 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6217 bio->bi_iter.bi_sector = logical >> 9;
6218 if (atomic_read(&bbio->error) > bbio->max_errors)
6219 bio->bi_status = BLK_STS_IOERR;
6221 bio->bi_status = BLK_STS_OK;
6222 btrfs_end_bbio(bbio, bio);
6226 blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
6227 int mirror_num, int async_submit)
6229 struct btrfs_device *dev;
6230 struct bio *first_bio = bio;
6231 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
6237 struct btrfs_bio *bbio = NULL;
6239 length = bio->bi_iter.bi_size;
6240 map_length = length;
6242 btrfs_bio_counter_inc_blocked(fs_info);
6243 ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical,
6244 &map_length, &bbio, mirror_num, 1);
6246 btrfs_bio_counter_dec(fs_info);
6247 return errno_to_blk_status(ret);
6250 total_devs = bbio->num_stripes;
6251 bbio->orig_bio = first_bio;
6252 bbio->private = first_bio->bi_private;
6253 bbio->end_io = first_bio->bi_end_io;
6254 bbio->fs_info = fs_info;
6255 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6257 if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6258 ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) {
6259 /* In this case, map_length has been set to the length of
6260 a single stripe; not the whole write */
6261 if (bio_op(bio) == REQ_OP_WRITE) {
6262 ret = raid56_parity_write(fs_info, bio, bbio,
6265 ret = raid56_parity_recover(fs_info, bio, bbio,
6266 map_length, mirror_num, 1);
6269 btrfs_bio_counter_dec(fs_info);
6270 return errno_to_blk_status(ret);
6273 if (map_length < length) {
6275 "mapping failed logical %llu bio len %llu len %llu",
6276 logical, length, map_length);
6280 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6281 dev = bbio->stripes[dev_nr].dev;
6282 if (!dev || !dev->bdev ||
6283 (bio_op(first_bio) == REQ_OP_WRITE &&
6284 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) {
6285 bbio_error(bbio, first_bio, logical);
6289 if (dev_nr < total_devs - 1)
6290 bio = btrfs_bio_clone(first_bio);
6294 submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical,
6295 dev_nr, async_submit);
6297 btrfs_bio_counter_dec(fs_info);
6301 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
6304 struct btrfs_device *device;
6305 struct btrfs_fs_devices *cur_devices;
6307 cur_devices = fs_info->fs_devices;
6308 while (cur_devices) {
6310 !memcmp(cur_devices->fsid, fsid, BTRFS_FSID_SIZE)) {
6311 device = find_device(cur_devices, devid, uuid);
6315 cur_devices = cur_devices->seed;
6320 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices,
6321 u64 devid, u8 *dev_uuid)
6323 struct btrfs_device *device;
6325 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6329 list_add(&device->dev_list, &fs_devices->devices);
6330 device->fs_devices = fs_devices;
6331 fs_devices->num_devices++;
6333 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6334 fs_devices->missing_devices++;
6340 * btrfs_alloc_device - allocate struct btrfs_device
6341 * @fs_info: used only for generating a new devid, can be NULL if
6342 * devid is provided (i.e. @devid != NULL).
6343 * @devid: a pointer to devid for this device. If NULL a new devid
6345 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6348 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6349 * on error. Returned struct is not linked onto any lists and must be
6350 * destroyed with btrfs_free_device.
6352 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6356 struct btrfs_device *dev;
6359 if (WARN_ON(!devid && !fs_info))
6360 return ERR_PTR(-EINVAL);
6362 dev = __alloc_device();
6371 ret = find_next_devid(fs_info, &tmp);
6373 btrfs_free_device(dev);
6374 return ERR_PTR(ret);
6380 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6382 generate_random_uuid(dev->uuid);
6384 btrfs_init_work(&dev->work, btrfs_submit_helper,
6385 pending_bios_fn, NULL, NULL);
6390 /* Return -EIO if any error, otherwise return 0. */
6391 static int btrfs_check_chunk_valid(struct btrfs_fs_info *fs_info,
6392 struct extent_buffer *leaf,
6393 struct btrfs_chunk *chunk, u64 logical)
6401 length = btrfs_chunk_length(leaf, chunk);
6402 stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6403 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6404 sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6405 type = btrfs_chunk_type(leaf, chunk);
6408 btrfs_err(fs_info, "invalid chunk num_stripes: %u",
6412 if (!IS_ALIGNED(logical, fs_info->sectorsize)) {
6413 btrfs_err(fs_info, "invalid chunk logical %llu", logical);
6416 if (btrfs_chunk_sector_size(leaf, chunk) != fs_info->sectorsize) {
6417 btrfs_err(fs_info, "invalid chunk sectorsize %u",
6418 btrfs_chunk_sector_size(leaf, chunk));
6421 if (!length || !IS_ALIGNED(length, fs_info->sectorsize)) {
6422 btrfs_err(fs_info, "invalid chunk length %llu", length);
6425 if (!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN) {
6426 btrfs_err(fs_info, "invalid chunk stripe length: %llu",
6430 if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6432 btrfs_err(fs_info, "unrecognized chunk type: %llu",
6433 ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
6434 BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6435 btrfs_chunk_type(leaf, chunk));
6438 if ((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes != 2) ||
6439 (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) ||
6440 (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) ||
6441 (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) ||
6442 (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) ||
6443 ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
6444 num_stripes != 1)) {
6446 "invalid num_stripes:sub_stripes %u:%u for profile %llu",
6447 num_stripes, sub_stripes,
6448 type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
6455 static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info,
6456 u64 devid, u8 *uuid, bool error)
6459 btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing",
6462 btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing",
6466 static int read_one_chunk(struct btrfs_fs_info *fs_info, struct btrfs_key *key,
6467 struct extent_buffer *leaf,
6468 struct btrfs_chunk *chunk)
6470 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
6471 struct map_lookup *map;
6472 struct extent_map *em;
6476 u8 uuid[BTRFS_UUID_SIZE];
6481 logical = key->offset;
6482 length = btrfs_chunk_length(leaf, chunk);
6483 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6485 ret = btrfs_check_chunk_valid(fs_info, leaf, chunk, logical);
6489 read_lock(&map_tree->map_tree.lock);
6490 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
6491 read_unlock(&map_tree->map_tree.lock);
6493 /* already mapped? */
6494 if (em && em->start <= logical && em->start + em->len > logical) {
6495 free_extent_map(em);
6498 free_extent_map(em);
6501 em = alloc_extent_map();
6504 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6506 free_extent_map(em);
6510 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6511 em->map_lookup = map;
6512 em->start = logical;
6515 em->block_start = 0;
6516 em->block_len = em->len;
6518 map->num_stripes = num_stripes;
6519 map->io_width = btrfs_chunk_io_width(leaf, chunk);
6520 map->io_align = btrfs_chunk_io_align(leaf, chunk);
6521 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6522 map->type = btrfs_chunk_type(leaf, chunk);
6523 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6524 for (i = 0; i < num_stripes; i++) {
6525 map->stripes[i].physical =
6526 btrfs_stripe_offset_nr(leaf, chunk, i);
6527 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6528 read_extent_buffer(leaf, uuid, (unsigned long)
6529 btrfs_stripe_dev_uuid_nr(chunk, i),
6531 map->stripes[i].dev = btrfs_find_device(fs_info, devid,
6533 if (!map->stripes[i].dev &&
6534 !btrfs_test_opt(fs_info, DEGRADED)) {
6535 free_extent_map(em);
6536 btrfs_report_missing_device(fs_info, devid, uuid, true);
6539 if (!map->stripes[i].dev) {
6540 map->stripes[i].dev =
6541 add_missing_dev(fs_info->fs_devices, devid,
6543 if (IS_ERR(map->stripes[i].dev)) {
6544 free_extent_map(em);
6546 "failed to init missing dev %llu: %ld",
6547 devid, PTR_ERR(map->stripes[i].dev));
6548 return PTR_ERR(map->stripes[i].dev);
6550 btrfs_report_missing_device(fs_info, devid, uuid, false);
6552 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
6553 &(map->stripes[i].dev->dev_state));
6557 write_lock(&map_tree->map_tree.lock);
6558 ret = add_extent_mapping(&map_tree->map_tree, em, 0);
6559 write_unlock(&map_tree->map_tree.lock);
6560 BUG_ON(ret); /* Tree corruption */
6561 free_extent_map(em);
6566 static void fill_device_from_item(struct extent_buffer *leaf,
6567 struct btrfs_dev_item *dev_item,
6568 struct btrfs_device *device)
6572 device->devid = btrfs_device_id(leaf, dev_item);
6573 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6574 device->total_bytes = device->disk_total_bytes;
6575 device->commit_total_bytes = device->disk_total_bytes;
6576 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6577 device->commit_bytes_used = device->bytes_used;
6578 device->type = btrfs_device_type(leaf, dev_item);
6579 device->io_align = btrfs_device_io_align(leaf, dev_item);
6580 device->io_width = btrfs_device_io_width(leaf, dev_item);
6581 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6582 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6583 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
6585 ptr = btrfs_device_uuid(dev_item);
6586 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6589 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info,
6592 struct btrfs_fs_devices *fs_devices;
6595 lockdep_assert_held(&uuid_mutex);
6598 fs_devices = fs_info->fs_devices->seed;
6599 while (fs_devices) {
6600 if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE))
6603 fs_devices = fs_devices->seed;
6606 fs_devices = find_fsid(fsid);
6608 if (!btrfs_test_opt(fs_info, DEGRADED))
6609 return ERR_PTR(-ENOENT);
6611 fs_devices = alloc_fs_devices(fsid);
6612 if (IS_ERR(fs_devices))
6615 fs_devices->seeding = 1;
6616 fs_devices->opened = 1;
6620 fs_devices = clone_fs_devices(fs_devices);
6621 if (IS_ERR(fs_devices))
6624 ret = open_fs_devices(fs_devices, FMODE_READ, fs_info->bdev_holder);
6626 free_fs_devices(fs_devices);
6627 fs_devices = ERR_PTR(ret);
6631 if (!fs_devices->seeding) {
6632 close_fs_devices(fs_devices);
6633 free_fs_devices(fs_devices);
6634 fs_devices = ERR_PTR(-EINVAL);
6638 fs_devices->seed = fs_info->fs_devices->seed;
6639 fs_info->fs_devices->seed = fs_devices;
6644 static int read_one_dev(struct btrfs_fs_info *fs_info,
6645 struct extent_buffer *leaf,
6646 struct btrfs_dev_item *dev_item)
6648 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6649 struct btrfs_device *device;
6652 u8 fs_uuid[BTRFS_FSID_SIZE];
6653 u8 dev_uuid[BTRFS_UUID_SIZE];
6655 devid = btrfs_device_id(leaf, dev_item);
6656 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6658 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6661 if (memcmp(fs_uuid, fs_info->fsid, BTRFS_FSID_SIZE)) {
6662 fs_devices = open_seed_devices(fs_info, fs_uuid);
6663 if (IS_ERR(fs_devices))
6664 return PTR_ERR(fs_devices);
6667 device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
6669 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6670 btrfs_report_missing_device(fs_info, devid,
6675 device = add_missing_dev(fs_devices, devid, dev_uuid);
6676 if (IS_ERR(device)) {
6678 "failed to add missing dev %llu: %ld",
6679 devid, PTR_ERR(device));
6680 return PTR_ERR(device);
6682 btrfs_report_missing_device(fs_info, devid, dev_uuid, false);
6684 if (!device->bdev) {
6685 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6686 btrfs_report_missing_device(fs_info,
6687 devid, dev_uuid, true);
6690 btrfs_report_missing_device(fs_info, devid,
6694 if (!device->bdev &&
6695 !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
6697 * this happens when a device that was properly setup
6698 * in the device info lists suddenly goes bad.
6699 * device->bdev is NULL, and so we have to set
6700 * device->missing to one here
6702 device->fs_devices->missing_devices++;
6703 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6706 /* Move the device to its own fs_devices */
6707 if (device->fs_devices != fs_devices) {
6708 ASSERT(test_bit(BTRFS_DEV_STATE_MISSING,
6709 &device->dev_state));
6711 list_move(&device->dev_list, &fs_devices->devices);
6712 device->fs_devices->num_devices--;
6713 fs_devices->num_devices++;
6715 device->fs_devices->missing_devices--;
6716 fs_devices->missing_devices++;
6718 device->fs_devices = fs_devices;
6722 if (device->fs_devices != fs_info->fs_devices) {
6723 BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state));
6724 if (device->generation !=
6725 btrfs_device_generation(leaf, dev_item))
6729 fill_device_from_item(leaf, dev_item, device);
6730 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
6731 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
6732 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
6733 device->fs_devices->total_rw_bytes += device->total_bytes;
6734 atomic64_add(device->total_bytes - device->bytes_used,
6735 &fs_info->free_chunk_space);
6741 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
6743 struct btrfs_root *root = fs_info->tree_root;
6744 struct btrfs_super_block *super_copy = fs_info->super_copy;
6745 struct extent_buffer *sb;
6746 struct btrfs_disk_key *disk_key;
6747 struct btrfs_chunk *chunk;
6749 unsigned long sb_array_offset;
6756 struct btrfs_key key;
6758 ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize);
6760 * This will create extent buffer of nodesize, superblock size is
6761 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6762 * overallocate but we can keep it as-is, only the first page is used.
6764 sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET);
6767 set_extent_buffer_uptodate(sb);
6768 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6770 * The sb extent buffer is artificial and just used to read the system array.
6771 * set_extent_buffer_uptodate() call does not properly mark all it's
6772 * pages up-to-date when the page is larger: extent does not cover the
6773 * whole page and consequently check_page_uptodate does not find all
6774 * the page's extents up-to-date (the hole beyond sb),
6775 * write_extent_buffer then triggers a WARN_ON.
6777 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6778 * but sb spans only this function. Add an explicit SetPageUptodate call
6779 * to silence the warning eg. on PowerPC 64.
6781 if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
6782 SetPageUptodate(sb->pages[0]);
6784 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6785 array_size = btrfs_super_sys_array_size(super_copy);
6787 array_ptr = super_copy->sys_chunk_array;
6788 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6791 while (cur_offset < array_size) {
6792 disk_key = (struct btrfs_disk_key *)array_ptr;
6793 len = sizeof(*disk_key);
6794 if (cur_offset + len > array_size)
6795 goto out_short_read;
6797 btrfs_disk_key_to_cpu(&key, disk_key);
6800 sb_array_offset += len;
6803 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6804 chunk = (struct btrfs_chunk *)sb_array_offset;
6806 * At least one btrfs_chunk with one stripe must be
6807 * present, exact stripe count check comes afterwards
6809 len = btrfs_chunk_item_size(1);
6810 if (cur_offset + len > array_size)
6811 goto out_short_read;
6813 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6816 "invalid number of stripes %u in sys_array at offset %u",
6817 num_stripes, cur_offset);
6822 type = btrfs_chunk_type(sb, chunk);
6823 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
6825 "invalid chunk type %llu in sys_array at offset %u",
6831 len = btrfs_chunk_item_size(num_stripes);
6832 if (cur_offset + len > array_size)
6833 goto out_short_read;
6835 ret = read_one_chunk(fs_info, &key, sb, chunk);
6840 "unexpected item type %u in sys_array at offset %u",
6841 (u32)key.type, cur_offset);
6846 sb_array_offset += len;
6849 clear_extent_buffer_uptodate(sb);
6850 free_extent_buffer_stale(sb);
6854 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
6856 clear_extent_buffer_uptodate(sb);
6857 free_extent_buffer_stale(sb);
6862 * Check if all chunks in the fs are OK for read-write degraded mount
6864 * If the @failing_dev is specified, it's accounted as missing.
6866 * Return true if all chunks meet the minimal RW mount requirements.
6867 * Return false if any chunk doesn't meet the minimal RW mount requirements.
6869 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
6870 struct btrfs_device *failing_dev)
6872 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
6873 struct extent_map *em;
6877 read_lock(&map_tree->map_tree.lock);
6878 em = lookup_extent_mapping(&map_tree->map_tree, 0, (u64)-1);
6879 read_unlock(&map_tree->map_tree.lock);
6880 /* No chunk at all? Return false anyway */
6886 struct map_lookup *map;
6891 map = em->map_lookup;
6893 btrfs_get_num_tolerated_disk_barrier_failures(
6895 for (i = 0; i < map->num_stripes; i++) {
6896 struct btrfs_device *dev = map->stripes[i].dev;
6898 if (!dev || !dev->bdev ||
6899 test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) ||
6900 dev->last_flush_error)
6902 else if (failing_dev && failing_dev == dev)
6905 if (missing > max_tolerated) {
6908 "chunk %llu missing %d devices, max tolerance is %d for writeable mount",
6909 em->start, missing, max_tolerated);
6910 free_extent_map(em);
6914 next_start = extent_map_end(em);
6915 free_extent_map(em);
6917 read_lock(&map_tree->map_tree.lock);
6918 em = lookup_extent_mapping(&map_tree->map_tree, next_start,
6919 (u64)(-1) - next_start);
6920 read_unlock(&map_tree->map_tree.lock);
6926 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
6928 struct btrfs_root *root = fs_info->chunk_root;
6929 struct btrfs_path *path;
6930 struct extent_buffer *leaf;
6931 struct btrfs_key key;
6932 struct btrfs_key found_key;
6937 path = btrfs_alloc_path();
6941 mutex_lock(&uuid_mutex);
6942 mutex_lock(&fs_info->chunk_mutex);
6945 * Read all device items, and then all the chunk items. All
6946 * device items are found before any chunk item (their object id
6947 * is smaller than the lowest possible object id for a chunk
6948 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6950 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6953 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6957 leaf = path->nodes[0];
6958 slot = path->slots[0];
6959 if (slot >= btrfs_header_nritems(leaf)) {
6960 ret = btrfs_next_leaf(root, path);
6967 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6968 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6969 struct btrfs_dev_item *dev_item;
6970 dev_item = btrfs_item_ptr(leaf, slot,
6971 struct btrfs_dev_item);
6972 ret = read_one_dev(fs_info, leaf, dev_item);
6976 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6977 struct btrfs_chunk *chunk;
6978 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6979 ret = read_one_chunk(fs_info, &found_key, leaf, chunk);
6987 * After loading chunk tree, we've got all device information,
6988 * do another round of validation checks.
6990 if (total_dev != fs_info->fs_devices->total_devices) {
6992 "super_num_devices %llu mismatch with num_devices %llu found here",
6993 btrfs_super_num_devices(fs_info->super_copy),
6998 if (btrfs_super_total_bytes(fs_info->super_copy) <
6999 fs_info->fs_devices->total_rw_bytes) {
7001 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
7002 btrfs_super_total_bytes(fs_info->super_copy),
7003 fs_info->fs_devices->total_rw_bytes);
7009 mutex_unlock(&fs_info->chunk_mutex);
7010 mutex_unlock(&uuid_mutex);
7012 btrfs_free_path(path);
7016 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
7018 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7019 struct btrfs_device *device;
7021 while (fs_devices) {
7022 mutex_lock(&fs_devices->device_list_mutex);
7023 list_for_each_entry(device, &fs_devices->devices, dev_list)
7024 device->fs_info = fs_info;
7025 mutex_unlock(&fs_devices->device_list_mutex);
7027 fs_devices = fs_devices->seed;
7031 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
7035 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7036 btrfs_dev_stat_reset(dev, i);
7039 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
7041 struct btrfs_key key;
7042 struct btrfs_key found_key;
7043 struct btrfs_root *dev_root = fs_info->dev_root;
7044 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7045 struct extent_buffer *eb;
7048 struct btrfs_device *device;
7049 struct btrfs_path *path = NULL;
7052 path = btrfs_alloc_path();
7058 mutex_lock(&fs_devices->device_list_mutex);
7059 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7061 struct btrfs_dev_stats_item *ptr;
7063 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7064 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7065 key.offset = device->devid;
7066 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
7068 __btrfs_reset_dev_stats(device);
7069 device->dev_stats_valid = 1;
7070 btrfs_release_path(path);
7073 slot = path->slots[0];
7074 eb = path->nodes[0];
7075 btrfs_item_key_to_cpu(eb, &found_key, slot);
7076 item_size = btrfs_item_size_nr(eb, slot);
7078 ptr = btrfs_item_ptr(eb, slot,
7079 struct btrfs_dev_stats_item);
7081 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7082 if (item_size >= (1 + i) * sizeof(__le64))
7083 btrfs_dev_stat_set(device, i,
7084 btrfs_dev_stats_value(eb, ptr, i));
7086 btrfs_dev_stat_reset(device, i);
7089 device->dev_stats_valid = 1;
7090 btrfs_dev_stat_print_on_load(device);
7091 btrfs_release_path(path);
7093 mutex_unlock(&fs_devices->device_list_mutex);
7096 btrfs_free_path(path);
7097 return ret < 0 ? ret : 0;
7100 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
7101 struct btrfs_fs_info *fs_info,
7102 struct btrfs_device *device)
7104 struct btrfs_root *dev_root = fs_info->dev_root;
7105 struct btrfs_path *path;
7106 struct btrfs_key key;
7107 struct extent_buffer *eb;
7108 struct btrfs_dev_stats_item *ptr;
7112 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7113 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7114 key.offset = device->devid;
7116 path = btrfs_alloc_path();
7119 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
7121 btrfs_warn_in_rcu(fs_info,
7122 "error %d while searching for dev_stats item for device %s",
7123 ret, rcu_str_deref(device->name));
7128 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
7129 /* need to delete old one and insert a new one */
7130 ret = btrfs_del_item(trans, dev_root, path);
7132 btrfs_warn_in_rcu(fs_info,
7133 "delete too small dev_stats item for device %s failed %d",
7134 rcu_str_deref(device->name), ret);
7141 /* need to insert a new item */
7142 btrfs_release_path(path);
7143 ret = btrfs_insert_empty_item(trans, dev_root, path,
7144 &key, sizeof(*ptr));
7146 btrfs_warn_in_rcu(fs_info,
7147 "insert dev_stats item for device %s failed %d",
7148 rcu_str_deref(device->name), ret);
7153 eb = path->nodes[0];
7154 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7155 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7156 btrfs_set_dev_stats_value(eb, ptr, i,
7157 btrfs_dev_stat_read(device, i));
7158 btrfs_mark_buffer_dirty(eb);
7161 btrfs_free_path(path);
7166 * called from commit_transaction. Writes all changed device stats to disk.
7168 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
7169 struct btrfs_fs_info *fs_info)
7171 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7172 struct btrfs_device *device;
7176 mutex_lock(&fs_devices->device_list_mutex);
7177 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7178 stats_cnt = atomic_read(&device->dev_stats_ccnt);
7179 if (!device->dev_stats_valid || stats_cnt == 0)
7184 * There is a LOAD-LOAD control dependency between the value of
7185 * dev_stats_ccnt and updating the on-disk values which requires
7186 * reading the in-memory counters. Such control dependencies
7187 * require explicit read memory barriers.
7189 * This memory barriers pairs with smp_mb__before_atomic in
7190 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full
7191 * barrier implied by atomic_xchg in
7192 * btrfs_dev_stats_read_and_reset
7196 ret = update_dev_stat_item(trans, fs_info, device);
7198 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7200 mutex_unlock(&fs_devices->device_list_mutex);
7205 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7207 btrfs_dev_stat_inc(dev, index);
7208 btrfs_dev_stat_print_on_error(dev);
7211 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
7213 if (!dev->dev_stats_valid)
7215 btrfs_err_rl_in_rcu(dev->fs_info,
7216 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7217 rcu_str_deref(dev->name),
7218 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7219 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7220 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7221 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7222 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7225 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7229 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7230 if (btrfs_dev_stat_read(dev, i) != 0)
7232 if (i == BTRFS_DEV_STAT_VALUES_MAX)
7233 return; /* all values == 0, suppress message */
7235 btrfs_info_in_rcu(dev->fs_info,
7236 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7237 rcu_str_deref(dev->name),
7238 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7239 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7240 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7241 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7242 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7245 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
7246 struct btrfs_ioctl_get_dev_stats *stats)
7248 struct btrfs_device *dev;
7249 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7252 mutex_lock(&fs_devices->device_list_mutex);
7253 dev = btrfs_find_device(fs_info, stats->devid, NULL, NULL);
7254 mutex_unlock(&fs_devices->device_list_mutex);
7257 btrfs_warn(fs_info, "get dev_stats failed, device not found");
7259 } else if (!dev->dev_stats_valid) {
7260 btrfs_warn(fs_info, "get dev_stats failed, not yet valid");
7262 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7263 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7264 if (stats->nr_items > i)
7266 btrfs_dev_stat_read_and_reset(dev, i);
7268 btrfs_dev_stat_reset(dev, i);
7271 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7272 if (stats->nr_items > i)
7273 stats->values[i] = btrfs_dev_stat_read(dev, i);
7275 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7276 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7280 void btrfs_scratch_superblocks(struct block_device *bdev, const char *device_path)
7282 struct buffer_head *bh;
7283 struct btrfs_super_block *disk_super;
7289 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX;
7292 if (btrfs_read_dev_one_super(bdev, copy_num, &bh))
7295 disk_super = (struct btrfs_super_block *)bh->b_data;
7297 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
7298 set_buffer_dirty(bh);
7299 sync_dirty_buffer(bh);
7303 /* Notify udev that device has changed */
7304 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
7306 /* Update ctime/mtime for device path for libblkid */
7307 update_dev_time(device_path);
7311 * Update the size of all devices, which is used for writing out the
7314 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
7316 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7317 struct btrfs_device *curr, *next;
7319 if (list_empty(&fs_devices->resized_devices))
7322 mutex_lock(&fs_devices->device_list_mutex);
7323 mutex_lock(&fs_info->chunk_mutex);
7324 list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
7326 list_del_init(&curr->resized_list);
7327 curr->commit_total_bytes = curr->disk_total_bytes;
7329 mutex_unlock(&fs_info->chunk_mutex);
7330 mutex_unlock(&fs_devices->device_list_mutex);
7333 /* Must be invoked during the transaction commit */
7334 void btrfs_update_commit_device_bytes_used(struct btrfs_transaction *trans)
7336 struct btrfs_fs_info *fs_info = trans->fs_info;
7337 struct extent_map *em;
7338 struct map_lookup *map;
7339 struct btrfs_device *dev;
7342 if (list_empty(&trans->pending_chunks))
7345 /* In order to kick the device replace finish process */
7346 mutex_lock(&fs_info->chunk_mutex);
7347 list_for_each_entry(em, &trans->pending_chunks, list) {
7348 map = em->map_lookup;
7350 for (i = 0; i < map->num_stripes; i++) {
7351 dev = map->stripes[i].dev;
7352 dev->commit_bytes_used = dev->bytes_used;
7355 mutex_unlock(&fs_info->chunk_mutex);
7358 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
7360 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7361 while (fs_devices) {
7362 fs_devices->fs_info = fs_info;
7363 fs_devices = fs_devices->seed;
7367 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
7369 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7370 while (fs_devices) {
7371 fs_devices->fs_info = NULL;
7372 fs_devices = fs_devices->seed;