2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/bio.h>
20 #include <linux/slab.h>
21 #include <linux/buffer_head.h>
22 #include <linux/blkdev.h>
23 #include <linux/iocontext.h>
24 #include <linux/capability.h>
25 #include <linux/ratelimit.h>
26 #include <linux/kthread.h>
27 #include <linux/raid/pq.h>
28 #include <linux/semaphore.h>
29 #include <linux/uuid.h>
30 #include <asm/div64.h>
32 #include "extent_map.h"
34 #include "transaction.h"
35 #include "print-tree.h"
38 #include "async-thread.h"
39 #include "check-integrity.h"
40 #include "rcu-string.h"
42 #include "dev-replace.h"
45 const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
46 [BTRFS_RAID_RAID10] = {
49 .devs_max = 0, /* 0 == as many as possible */
51 .tolerated_failures = 1,
55 [BTRFS_RAID_RAID1] = {
60 .tolerated_failures = 1,
69 .tolerated_failures = 0,
73 [BTRFS_RAID_RAID0] = {
78 .tolerated_failures = 0,
82 [BTRFS_RAID_SINGLE] = {
87 .tolerated_failures = 0,
91 [BTRFS_RAID_RAID5] = {
96 .tolerated_failures = 1,
100 [BTRFS_RAID_RAID6] = {
105 .tolerated_failures = 2,
111 const u64 btrfs_raid_group[BTRFS_NR_RAID_TYPES] = {
112 [BTRFS_RAID_RAID10] = BTRFS_BLOCK_GROUP_RAID10,
113 [BTRFS_RAID_RAID1] = BTRFS_BLOCK_GROUP_RAID1,
114 [BTRFS_RAID_DUP] = BTRFS_BLOCK_GROUP_DUP,
115 [BTRFS_RAID_RAID0] = BTRFS_BLOCK_GROUP_RAID0,
116 [BTRFS_RAID_SINGLE] = 0,
117 [BTRFS_RAID_RAID5] = BTRFS_BLOCK_GROUP_RAID5,
118 [BTRFS_RAID_RAID6] = BTRFS_BLOCK_GROUP_RAID6,
122 * Table to convert BTRFS_RAID_* to the error code if minimum number of devices
123 * condition is not met. Zero means there's no corresponding
124 * BTRFS_ERROR_DEV_*_NOT_MET value.
126 const int btrfs_raid_mindev_error[BTRFS_NR_RAID_TYPES] = {
127 [BTRFS_RAID_RAID10] = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET,
128 [BTRFS_RAID_RAID1] = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET,
129 [BTRFS_RAID_DUP] = 0,
130 [BTRFS_RAID_RAID0] = 0,
131 [BTRFS_RAID_SINGLE] = 0,
132 [BTRFS_RAID_RAID5] = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET,
133 [BTRFS_RAID_RAID6] = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET,
136 static int init_first_rw_device(struct btrfs_trans_handle *trans,
137 struct btrfs_root *root,
138 struct btrfs_device *device);
139 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
140 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
141 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
142 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
144 DEFINE_MUTEX(uuid_mutex);
145 static LIST_HEAD(fs_uuids);
146 struct list_head *btrfs_get_fs_uuids(void)
151 static struct btrfs_fs_devices *__alloc_fs_devices(void)
153 struct btrfs_fs_devices *fs_devs;
155 fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL);
157 return ERR_PTR(-ENOMEM);
159 mutex_init(&fs_devs->device_list_mutex);
161 INIT_LIST_HEAD(&fs_devs->devices);
162 INIT_LIST_HEAD(&fs_devs->resized_devices);
163 INIT_LIST_HEAD(&fs_devs->alloc_list);
164 INIT_LIST_HEAD(&fs_devs->list);
170 * alloc_fs_devices - allocate struct btrfs_fs_devices
171 * @fsid: a pointer to UUID for this FS. If NULL a new UUID is
174 * Return: a pointer to a new &struct btrfs_fs_devices on success;
175 * ERR_PTR() on error. Returned struct is not linked onto any lists and
176 * can be destroyed with kfree() right away.
178 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
180 struct btrfs_fs_devices *fs_devs;
182 fs_devs = __alloc_fs_devices();
187 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
189 generate_random_uuid(fs_devs->fsid);
194 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
196 struct btrfs_device *device;
197 WARN_ON(fs_devices->opened);
198 while (!list_empty(&fs_devices->devices)) {
199 device = list_entry(fs_devices->devices.next,
200 struct btrfs_device, dev_list);
201 list_del(&device->dev_list);
202 rcu_string_free(device->name);
208 static void btrfs_kobject_uevent(struct block_device *bdev,
209 enum kobject_action action)
213 ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
215 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
217 kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
218 &disk_to_dev(bdev->bd_disk)->kobj);
221 void btrfs_cleanup_fs_uuids(void)
223 struct btrfs_fs_devices *fs_devices;
225 while (!list_empty(&fs_uuids)) {
226 fs_devices = list_entry(fs_uuids.next,
227 struct btrfs_fs_devices, list);
228 list_del(&fs_devices->list);
229 free_fs_devices(fs_devices);
233 static struct btrfs_device *__alloc_device(void)
235 struct btrfs_device *dev;
237 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
239 return ERR_PTR(-ENOMEM);
241 INIT_LIST_HEAD(&dev->dev_list);
242 INIT_LIST_HEAD(&dev->dev_alloc_list);
243 INIT_LIST_HEAD(&dev->resized_list);
245 spin_lock_init(&dev->io_lock);
247 spin_lock_init(&dev->reada_lock);
248 atomic_set(&dev->reada_in_flight, 0);
249 atomic_set(&dev->dev_stats_ccnt, 0);
250 btrfs_device_data_ordered_init(dev);
251 INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
252 INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
257 static noinline struct btrfs_device *__find_device(struct list_head *head,
260 struct btrfs_device *dev;
262 list_for_each_entry(dev, head, dev_list) {
263 if (dev->devid == devid &&
264 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
271 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
273 struct btrfs_fs_devices *fs_devices;
275 list_for_each_entry(fs_devices, &fs_uuids, list) {
276 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
283 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
284 int flush, struct block_device **bdev,
285 struct buffer_head **bh)
289 *bdev = blkdev_get_by_path(device_path, flags, holder);
292 ret = PTR_ERR(*bdev);
297 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
298 ret = set_blocksize(*bdev, 4096);
300 blkdev_put(*bdev, flags);
303 invalidate_bdev(*bdev);
304 *bh = btrfs_read_dev_super(*bdev);
307 blkdev_put(*bdev, flags);
319 static void requeue_list(struct btrfs_pending_bios *pending_bios,
320 struct bio *head, struct bio *tail)
323 struct bio *old_head;
325 old_head = pending_bios->head;
326 pending_bios->head = head;
327 if (pending_bios->tail)
328 tail->bi_next = old_head;
330 pending_bios->tail = tail;
334 * we try to collect pending bios for a device so we don't get a large
335 * number of procs sending bios down to the same device. This greatly
336 * improves the schedulers ability to collect and merge the bios.
338 * But, it also turns into a long list of bios to process and that is sure
339 * to eventually make the worker thread block. The solution here is to
340 * make some progress and then put this work struct back at the end of
341 * the list if the block device is congested. This way, multiple devices
342 * can make progress from a single worker thread.
344 static noinline void run_scheduled_bios(struct btrfs_device *device)
347 struct backing_dev_info *bdi;
348 struct btrfs_fs_info *fs_info;
349 struct btrfs_pending_bios *pending_bios;
353 unsigned long num_run;
354 unsigned long batch_run = 0;
356 unsigned long last_waited = 0;
358 int sync_pending = 0;
359 struct blk_plug plug;
362 * this function runs all the bios we've collected for
363 * a particular device. We don't want to wander off to
364 * another device without first sending all of these down.
365 * So, setup a plug here and finish it off before we return
367 blk_start_plug(&plug);
369 bdi = blk_get_backing_dev_info(device->bdev);
370 fs_info = device->dev_root->fs_info;
371 limit = btrfs_async_submit_limit(fs_info);
372 limit = limit * 2 / 3;
375 spin_lock(&device->io_lock);
380 /* take all the bios off the list at once and process them
381 * later on (without the lock held). But, remember the
382 * tail and other pointers so the bios can be properly reinserted
383 * into the list if we hit congestion
385 if (!force_reg && device->pending_sync_bios.head) {
386 pending_bios = &device->pending_sync_bios;
389 pending_bios = &device->pending_bios;
393 pending = pending_bios->head;
394 tail = pending_bios->tail;
395 WARN_ON(pending && !tail);
398 * if pending was null this time around, no bios need processing
399 * at all and we can stop. Otherwise it'll loop back up again
400 * and do an additional check so no bios are missed.
402 * device->running_pending is used to synchronize with the
405 if (device->pending_sync_bios.head == NULL &&
406 device->pending_bios.head == NULL) {
408 device->running_pending = 0;
411 device->running_pending = 1;
414 pending_bios->head = NULL;
415 pending_bios->tail = NULL;
417 spin_unlock(&device->io_lock);
422 /* we want to work on both lists, but do more bios on the
423 * sync list than the regular list
426 pending_bios != &device->pending_sync_bios &&
427 device->pending_sync_bios.head) ||
428 (num_run > 64 && pending_bios == &device->pending_sync_bios &&
429 device->pending_bios.head)) {
430 spin_lock(&device->io_lock);
431 requeue_list(pending_bios, pending, tail);
436 pending = pending->bi_next;
440 * atomic_dec_return implies a barrier for waitqueue_active
442 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
443 waitqueue_active(&fs_info->async_submit_wait))
444 wake_up(&fs_info->async_submit_wait);
446 BUG_ON(atomic_read(&cur->__bi_cnt) == 0);
449 * if we're doing the sync list, record that our
450 * plug has some sync requests on it
452 * If we're doing the regular list and there are
453 * sync requests sitting around, unplug before
456 if (pending_bios == &device->pending_sync_bios) {
458 } else if (sync_pending) {
459 blk_finish_plug(&plug);
460 blk_start_plug(&plug);
464 btrfsic_submit_bio(cur);
471 * we made progress, there is more work to do and the bdi
472 * is now congested. Back off and let other work structs
475 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
476 fs_info->fs_devices->open_devices > 1) {
477 struct io_context *ioc;
479 ioc = current->io_context;
482 * the main goal here is that we don't want to
483 * block if we're going to be able to submit
484 * more requests without blocking.
486 * This code does two great things, it pokes into
487 * the elevator code from a filesystem _and_
488 * it makes assumptions about how batching works.
490 if (ioc && ioc->nr_batch_requests > 0 &&
491 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
493 ioc->last_waited == last_waited)) {
495 * we want to go through our batch of
496 * requests and stop. So, we copy out
497 * the ioc->last_waited time and test
498 * against it before looping
500 last_waited = ioc->last_waited;
504 spin_lock(&device->io_lock);
505 requeue_list(pending_bios, pending, tail);
506 device->running_pending = 1;
508 spin_unlock(&device->io_lock);
509 btrfs_queue_work(fs_info->submit_workers,
513 /* unplug every 64 requests just for good measure */
514 if (batch_run % 64 == 0) {
515 blk_finish_plug(&plug);
516 blk_start_plug(&plug);
525 spin_lock(&device->io_lock);
526 if (device->pending_bios.head || device->pending_sync_bios.head)
528 spin_unlock(&device->io_lock);
531 blk_finish_plug(&plug);
534 static void pending_bios_fn(struct btrfs_work *work)
536 struct btrfs_device *device;
538 device = container_of(work, struct btrfs_device, work);
539 run_scheduled_bios(device);
543 void btrfs_free_stale_device(struct btrfs_device *cur_dev)
545 struct btrfs_fs_devices *fs_devs;
546 struct btrfs_device *dev;
551 list_for_each_entry(fs_devs, &fs_uuids, list) {
556 if (fs_devs->seeding)
559 list_for_each_entry(dev, &fs_devs->devices, dev_list) {
567 * Todo: This won't be enough. What if the same device
568 * comes back (with new uuid and) with its mapper path?
569 * But for now, this does help as mostly an admin will
570 * either use mapper or non mapper path throughout.
573 del = strcmp(rcu_str_deref(dev->name),
574 rcu_str_deref(cur_dev->name));
581 /* delete the stale device */
582 if (fs_devs->num_devices == 1) {
583 btrfs_sysfs_remove_fsid(fs_devs);
584 list_del(&fs_devs->list);
585 free_fs_devices(fs_devs);
587 fs_devs->num_devices--;
588 list_del(&dev->dev_list);
589 rcu_string_free(dev->name);
598 * Add new device to list of registered devices
601 * 1 - first time device is seen
602 * 0 - device already known
605 static noinline int device_list_add(const char *path,
606 struct btrfs_super_block *disk_super,
607 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
609 struct btrfs_device *device;
610 struct btrfs_fs_devices *fs_devices;
611 struct rcu_string *name;
613 u64 found_transid = btrfs_super_generation(disk_super);
615 fs_devices = find_fsid(disk_super->fsid);
617 fs_devices = alloc_fs_devices(disk_super->fsid);
618 if (IS_ERR(fs_devices))
619 return PTR_ERR(fs_devices);
621 list_add(&fs_devices->list, &fs_uuids);
625 device = __find_device(&fs_devices->devices, devid,
626 disk_super->dev_item.uuid);
630 if (fs_devices->opened)
633 device = btrfs_alloc_device(NULL, &devid,
634 disk_super->dev_item.uuid);
635 if (IS_ERR(device)) {
636 /* we can safely leave the fs_devices entry around */
637 return PTR_ERR(device);
640 name = rcu_string_strdup(path, GFP_NOFS);
645 rcu_assign_pointer(device->name, name);
647 mutex_lock(&fs_devices->device_list_mutex);
648 list_add_rcu(&device->dev_list, &fs_devices->devices);
649 fs_devices->num_devices++;
650 mutex_unlock(&fs_devices->device_list_mutex);
653 device->fs_devices = fs_devices;
654 } else if (!device->name || strcmp(device->name->str, path)) {
656 * When FS is already mounted.
657 * 1. If you are here and if the device->name is NULL that
658 * means this device was missing at time of FS mount.
659 * 2. If you are here and if the device->name is different
660 * from 'path' that means either
661 * a. The same device disappeared and reappeared with
663 * b. The missing-disk-which-was-replaced, has
666 * We must allow 1 and 2a above. But 2b would be a spurious
669 * Further in case of 1 and 2a above, the disk at 'path'
670 * would have missed some transaction when it was away and
671 * in case of 2a the stale bdev has to be updated as well.
672 * 2b must not be allowed at all time.
676 * For now, we do allow update to btrfs_fs_device through the
677 * btrfs dev scan cli after FS has been mounted. We're still
678 * tracking a problem where systems fail mount by subvolume id
679 * when we reject replacement on a mounted FS.
681 if (!fs_devices->opened && found_transid < device->generation) {
683 * That is if the FS is _not_ mounted and if you
684 * are here, that means there is more than one
685 * disk with same uuid and devid.We keep the one
686 * with larger generation number or the last-in if
687 * generation are equal.
692 name = rcu_string_strdup(path, GFP_NOFS);
695 rcu_string_free(device->name);
696 rcu_assign_pointer(device->name, name);
697 if (device->missing) {
698 fs_devices->missing_devices--;
704 * Unmount does not free the btrfs_device struct but would zero
705 * generation along with most of the other members. So just update
706 * it back. We need it to pick the disk with largest generation
709 if (!fs_devices->opened)
710 device->generation = found_transid;
713 * if there is new btrfs on an already registered device,
714 * then remove the stale device entry.
717 btrfs_free_stale_device(device);
719 *fs_devices_ret = fs_devices;
724 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
726 struct btrfs_fs_devices *fs_devices;
727 struct btrfs_device *device;
728 struct btrfs_device *orig_dev;
730 fs_devices = alloc_fs_devices(orig->fsid);
731 if (IS_ERR(fs_devices))
734 mutex_lock(&orig->device_list_mutex);
735 fs_devices->total_devices = orig->total_devices;
737 /* We have held the volume lock, it is safe to get the devices. */
738 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
739 struct rcu_string *name;
741 device = btrfs_alloc_device(NULL, &orig_dev->devid,
747 * This is ok to do without rcu read locked because we hold the
748 * uuid mutex so nothing we touch in here is going to disappear.
750 if (orig_dev->name) {
751 name = rcu_string_strdup(orig_dev->name->str,
757 rcu_assign_pointer(device->name, name);
760 list_add(&device->dev_list, &fs_devices->devices);
761 device->fs_devices = fs_devices;
762 fs_devices->num_devices++;
764 mutex_unlock(&orig->device_list_mutex);
767 mutex_unlock(&orig->device_list_mutex);
768 free_fs_devices(fs_devices);
769 return ERR_PTR(-ENOMEM);
772 void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step)
774 struct btrfs_device *device, *next;
775 struct btrfs_device *latest_dev = NULL;
777 mutex_lock(&uuid_mutex);
779 /* This is the initialized path, it is safe to release the devices. */
780 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
781 if (device->in_fs_metadata) {
782 if (!device->is_tgtdev_for_dev_replace &&
784 device->generation > latest_dev->generation)) {
790 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
792 * In the first step, keep the device which has
793 * the correct fsid and the devid that is used
794 * for the dev_replace procedure.
795 * In the second step, the dev_replace state is
796 * read from the device tree and it is known
797 * whether the procedure is really active or
798 * not, which means whether this device is
799 * used or whether it should be removed.
801 if (step == 0 || device->is_tgtdev_for_dev_replace) {
806 blkdev_put(device->bdev, device->mode);
808 fs_devices->open_devices--;
810 if (device->writeable) {
811 list_del_init(&device->dev_alloc_list);
812 device->writeable = 0;
813 if (!device->is_tgtdev_for_dev_replace)
814 fs_devices->rw_devices--;
816 list_del_init(&device->dev_list);
817 fs_devices->num_devices--;
818 rcu_string_free(device->name);
822 if (fs_devices->seed) {
823 fs_devices = fs_devices->seed;
827 fs_devices->latest_bdev = latest_dev->bdev;
829 mutex_unlock(&uuid_mutex);
832 static void __free_device(struct work_struct *work)
834 struct btrfs_device *device;
836 device = container_of(work, struct btrfs_device, rcu_work);
837 rcu_string_free(device->name);
841 static void free_device(struct rcu_head *head)
843 struct btrfs_device *device;
845 device = container_of(head, struct btrfs_device, rcu);
847 INIT_WORK(&device->rcu_work, __free_device);
848 schedule_work(&device->rcu_work);
851 static void btrfs_close_bdev(struct btrfs_device *device)
853 if (device->bdev && device->writeable) {
854 sync_blockdev(device->bdev);
855 invalidate_bdev(device->bdev);
859 blkdev_put(device->bdev, device->mode);
862 static void btrfs_close_one_device(struct btrfs_device *device)
864 struct btrfs_fs_devices *fs_devices = device->fs_devices;
865 struct btrfs_device *new_device;
866 struct rcu_string *name;
869 fs_devices->open_devices--;
871 if (device->writeable &&
872 device->devid != BTRFS_DEV_REPLACE_DEVID) {
873 list_del_init(&device->dev_alloc_list);
874 fs_devices->rw_devices--;
878 fs_devices->missing_devices--;
880 btrfs_close_bdev(device);
882 new_device = btrfs_alloc_device(NULL, &device->devid,
884 BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
886 /* Safe because we are under uuid_mutex */
888 name = rcu_string_strdup(device->name->str, GFP_NOFS);
889 BUG_ON(!name); /* -ENOMEM */
890 rcu_assign_pointer(new_device->name, name);
893 list_replace_rcu(&device->dev_list, &new_device->dev_list);
894 new_device->fs_devices = device->fs_devices;
896 call_rcu(&device->rcu, free_device);
899 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
901 struct btrfs_device *device, *tmp;
903 if (--fs_devices->opened > 0)
906 mutex_lock(&fs_devices->device_list_mutex);
907 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
908 btrfs_close_one_device(device);
910 mutex_unlock(&fs_devices->device_list_mutex);
912 WARN_ON(fs_devices->open_devices);
913 WARN_ON(fs_devices->rw_devices);
914 fs_devices->opened = 0;
915 fs_devices->seeding = 0;
920 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
922 struct btrfs_fs_devices *seed_devices = NULL;
925 mutex_lock(&uuid_mutex);
926 ret = __btrfs_close_devices(fs_devices);
927 if (!fs_devices->opened) {
928 seed_devices = fs_devices->seed;
929 fs_devices->seed = NULL;
931 mutex_unlock(&uuid_mutex);
933 while (seed_devices) {
934 fs_devices = seed_devices;
935 seed_devices = fs_devices->seed;
936 __btrfs_close_devices(fs_devices);
937 free_fs_devices(fs_devices);
940 * Wait for rcu kworkers under __btrfs_close_devices
941 * to finish all blkdev_puts so device is really
942 * free when umount is done.
948 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
949 fmode_t flags, void *holder)
951 struct request_queue *q;
952 struct block_device *bdev;
953 struct list_head *head = &fs_devices->devices;
954 struct btrfs_device *device;
955 struct btrfs_device *latest_dev = NULL;
956 struct buffer_head *bh;
957 struct btrfs_super_block *disk_super;
964 list_for_each_entry(device, head, dev_list) {
970 /* Just open everything we can; ignore failures here */
971 if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
975 disk_super = (struct btrfs_super_block *)bh->b_data;
976 devid = btrfs_stack_device_id(&disk_super->dev_item);
977 if (devid != device->devid)
980 if (memcmp(device->uuid, disk_super->dev_item.uuid,
984 device->generation = btrfs_super_generation(disk_super);
986 device->generation > latest_dev->generation)
989 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
990 device->writeable = 0;
992 device->writeable = !bdev_read_only(bdev);
996 q = bdev_get_queue(bdev);
997 if (blk_queue_discard(q))
998 device->can_discard = 1;
1000 device->bdev = bdev;
1001 device->in_fs_metadata = 0;
1002 device->mode = flags;
1004 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1005 fs_devices->rotating = 1;
1007 fs_devices->open_devices++;
1008 if (device->writeable &&
1009 device->devid != BTRFS_DEV_REPLACE_DEVID) {
1010 fs_devices->rw_devices++;
1011 list_add(&device->dev_alloc_list,
1012 &fs_devices->alloc_list);
1019 blkdev_put(bdev, flags);
1022 if (fs_devices->open_devices == 0) {
1026 fs_devices->seeding = seeding;
1027 fs_devices->opened = 1;
1028 fs_devices->latest_bdev = latest_dev->bdev;
1029 fs_devices->total_rw_bytes = 0;
1034 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1035 fmode_t flags, void *holder)
1039 mutex_lock(&uuid_mutex);
1040 if (fs_devices->opened) {
1041 fs_devices->opened++;
1044 ret = __btrfs_open_devices(fs_devices, flags, holder);
1046 mutex_unlock(&uuid_mutex);
1050 void btrfs_release_disk_super(struct page *page)
1056 int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr,
1057 struct page **page, struct btrfs_super_block **disk_super)
1062 /* make sure our super fits in the device */
1063 if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1066 /* make sure our super fits in the page */
1067 if (sizeof(**disk_super) > PAGE_SIZE)
1070 /* make sure our super doesn't straddle pages on disk */
1071 index = bytenr >> PAGE_SHIFT;
1072 if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index)
1075 /* pull in the page with our super */
1076 *page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
1079 if (IS_ERR_OR_NULL(*page))
1084 /* align our pointer to the offset of the super block */
1085 *disk_super = p + (bytenr & ~PAGE_MASK);
1087 if (btrfs_super_bytenr(*disk_super) != bytenr ||
1088 btrfs_super_magic(*disk_super) != BTRFS_MAGIC) {
1089 btrfs_release_disk_super(*page);
1093 if ((*disk_super)->label[0] &&
1094 (*disk_super)->label[BTRFS_LABEL_SIZE - 1])
1095 (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0';
1101 * Look for a btrfs signature on a device. This may be called out of the mount path
1102 * and we are not allowed to call set_blocksize during the scan. The superblock
1103 * is read via pagecache
1105 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
1106 struct btrfs_fs_devices **fs_devices_ret)
1108 struct btrfs_super_block *disk_super;
1109 struct block_device *bdev;
1118 * we would like to check all the supers, but that would make
1119 * a btrfs mount succeed after a mkfs from a different FS.
1120 * So, we need to add a special mount option to scan for
1121 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1123 bytenr = btrfs_sb_offset(0);
1124 flags |= FMODE_EXCL;
1125 mutex_lock(&uuid_mutex);
1127 bdev = blkdev_get_by_path(path, flags, holder);
1129 ret = PTR_ERR(bdev);
1133 if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super))
1134 goto error_bdev_put;
1136 devid = btrfs_stack_device_id(&disk_super->dev_item);
1137 transid = btrfs_super_generation(disk_super);
1138 total_devices = btrfs_super_num_devices(disk_super);
1140 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
1142 if (disk_super->label[0]) {
1143 printk(KERN_INFO "BTRFS: device label %s ", disk_super->label);
1145 printk(KERN_INFO "BTRFS: device fsid %pU ", disk_super->fsid);
1148 printk(KERN_CONT "devid %llu transid %llu %s\n", devid, transid, path);
1151 if (!ret && fs_devices_ret)
1152 (*fs_devices_ret)->total_devices = total_devices;
1154 btrfs_release_disk_super(page);
1157 blkdev_put(bdev, flags);
1159 mutex_unlock(&uuid_mutex);
1163 /* helper to account the used device space in the range */
1164 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
1165 u64 end, u64 *length)
1167 struct btrfs_key key;
1168 struct btrfs_root *root = device->dev_root;
1169 struct btrfs_dev_extent *dev_extent;
1170 struct btrfs_path *path;
1174 struct extent_buffer *l;
1178 if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
1181 path = btrfs_alloc_path();
1184 path->reada = READA_FORWARD;
1186 key.objectid = device->devid;
1188 key.type = BTRFS_DEV_EXTENT_KEY;
1190 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1194 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1201 slot = path->slots[0];
1202 if (slot >= btrfs_header_nritems(l)) {
1203 ret = btrfs_next_leaf(root, path);
1211 btrfs_item_key_to_cpu(l, &key, slot);
1213 if (key.objectid < device->devid)
1216 if (key.objectid > device->devid)
1219 if (key.type != BTRFS_DEV_EXTENT_KEY)
1222 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1223 extent_end = key.offset + btrfs_dev_extent_length(l,
1225 if (key.offset <= start && extent_end > end) {
1226 *length = end - start + 1;
1228 } else if (key.offset <= start && extent_end > start)
1229 *length += extent_end - start;
1230 else if (key.offset > start && extent_end <= end)
1231 *length += extent_end - key.offset;
1232 else if (key.offset > start && key.offset <= end) {
1233 *length += end - key.offset + 1;
1235 } else if (key.offset > end)
1243 btrfs_free_path(path);
1247 static int contains_pending_extent(struct btrfs_transaction *transaction,
1248 struct btrfs_device *device,
1249 u64 *start, u64 len)
1251 struct btrfs_fs_info *fs_info = device->dev_root->fs_info;
1252 struct extent_map *em;
1253 struct list_head *search_list = &fs_info->pinned_chunks;
1255 u64 physical_start = *start;
1258 search_list = &transaction->pending_chunks;
1260 list_for_each_entry(em, search_list, list) {
1261 struct map_lookup *map;
1264 map = em->map_lookup;
1265 for (i = 0; i < map->num_stripes; i++) {
1268 if (map->stripes[i].dev != device)
1270 if (map->stripes[i].physical >= physical_start + len ||
1271 map->stripes[i].physical + em->orig_block_len <=
1275 * Make sure that while processing the pinned list we do
1276 * not override our *start with a lower value, because
1277 * we can have pinned chunks that fall within this
1278 * device hole and that have lower physical addresses
1279 * than the pending chunks we processed before. If we
1280 * do not take this special care we can end up getting
1281 * 2 pending chunks that start at the same physical
1282 * device offsets because the end offset of a pinned
1283 * chunk can be equal to the start offset of some
1286 end = map->stripes[i].physical + em->orig_block_len;
1293 if (search_list != &fs_info->pinned_chunks) {
1294 search_list = &fs_info->pinned_chunks;
1303 * find_free_dev_extent_start - find free space in the specified device
1304 * @device: the device which we search the free space in
1305 * @num_bytes: the size of the free space that we need
1306 * @search_start: the position from which to begin the search
1307 * @start: store the start of the free space.
1308 * @len: the size of the free space. that we find, or the size
1309 * of the max free space if we don't find suitable free space
1311 * this uses a pretty simple search, the expectation is that it is
1312 * called very infrequently and that a given device has a small number
1315 * @start is used to store the start of the free space if we find. But if we
1316 * don't find suitable free space, it will be used to store the start position
1317 * of the max free space.
1319 * @len is used to store the size of the free space that we find.
1320 * But if we don't find suitable free space, it is used to store the size of
1321 * the max free space.
1323 int find_free_dev_extent_start(struct btrfs_transaction *transaction,
1324 struct btrfs_device *device, u64 num_bytes,
1325 u64 search_start, u64 *start, u64 *len)
1327 struct btrfs_key key;
1328 struct btrfs_root *root = device->dev_root;
1329 struct btrfs_dev_extent *dev_extent;
1330 struct btrfs_path *path;
1335 u64 search_end = device->total_bytes;
1338 struct extent_buffer *l;
1339 u64 min_search_start;
1342 * We don't want to overwrite the superblock on the drive nor any area
1343 * used by the boot loader (grub for example), so we make sure to start
1344 * at an offset of at least 1MB.
1346 min_search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
1347 search_start = max(search_start, min_search_start);
1349 path = btrfs_alloc_path();
1353 max_hole_start = search_start;
1357 if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1362 path->reada = READA_FORWARD;
1363 path->search_commit_root = 1;
1364 path->skip_locking = 1;
1366 key.objectid = device->devid;
1367 key.offset = search_start;
1368 key.type = BTRFS_DEV_EXTENT_KEY;
1370 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1374 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1381 slot = path->slots[0];
1382 if (slot >= btrfs_header_nritems(l)) {
1383 ret = btrfs_next_leaf(root, path);
1391 btrfs_item_key_to_cpu(l, &key, slot);
1393 if (key.objectid < device->devid)
1396 if (key.objectid > device->devid)
1399 if (key.type != BTRFS_DEV_EXTENT_KEY)
1402 if (key.offset > search_start) {
1403 hole_size = key.offset - search_start;
1406 * Have to check before we set max_hole_start, otherwise
1407 * we could end up sending back this offset anyway.
1409 if (contains_pending_extent(transaction, device,
1412 if (key.offset >= search_start) {
1413 hole_size = key.offset - search_start;
1420 if (hole_size > max_hole_size) {
1421 max_hole_start = search_start;
1422 max_hole_size = hole_size;
1426 * If this free space is greater than which we need,
1427 * it must be the max free space that we have found
1428 * until now, so max_hole_start must point to the start
1429 * of this free space and the length of this free space
1430 * is stored in max_hole_size. Thus, we return
1431 * max_hole_start and max_hole_size and go back to the
1434 if (hole_size >= num_bytes) {
1440 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1441 extent_end = key.offset + btrfs_dev_extent_length(l,
1443 if (extent_end > search_start)
1444 search_start = extent_end;
1451 * At this point, search_start should be the end of
1452 * allocated dev extents, and when shrinking the device,
1453 * search_end may be smaller than search_start.
1455 if (search_end > search_start) {
1456 hole_size = search_end - search_start;
1458 if (contains_pending_extent(transaction, device, &search_start,
1460 btrfs_release_path(path);
1464 if (hole_size > max_hole_size) {
1465 max_hole_start = search_start;
1466 max_hole_size = hole_size;
1471 if (max_hole_size < num_bytes)
1477 btrfs_free_path(path);
1478 *start = max_hole_start;
1480 *len = max_hole_size;
1484 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1485 struct btrfs_device *device, u64 num_bytes,
1486 u64 *start, u64 *len)
1488 /* FIXME use last free of some kind */
1489 return find_free_dev_extent_start(trans->transaction, device,
1490 num_bytes, 0, start, len);
1493 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1494 struct btrfs_device *device,
1495 u64 start, u64 *dev_extent_len)
1498 struct btrfs_path *path;
1499 struct btrfs_root *root = device->dev_root;
1500 struct btrfs_key key;
1501 struct btrfs_key found_key;
1502 struct extent_buffer *leaf = NULL;
1503 struct btrfs_dev_extent *extent = NULL;
1505 path = btrfs_alloc_path();
1509 key.objectid = device->devid;
1511 key.type = BTRFS_DEV_EXTENT_KEY;
1513 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1515 ret = btrfs_previous_item(root, path, key.objectid,
1516 BTRFS_DEV_EXTENT_KEY);
1519 leaf = path->nodes[0];
1520 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1521 extent = btrfs_item_ptr(leaf, path->slots[0],
1522 struct btrfs_dev_extent);
1523 BUG_ON(found_key.offset > start || found_key.offset +
1524 btrfs_dev_extent_length(leaf, extent) < start);
1526 btrfs_release_path(path);
1528 } else if (ret == 0) {
1529 leaf = path->nodes[0];
1530 extent = btrfs_item_ptr(leaf, path->slots[0],
1531 struct btrfs_dev_extent);
1533 btrfs_handle_fs_error(root->fs_info, ret, "Slot search failed");
1537 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1539 ret = btrfs_del_item(trans, root, path);
1541 btrfs_handle_fs_error(root->fs_info, ret,
1542 "Failed to remove dev extent item");
1544 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1547 btrfs_free_path(path);
1551 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1552 struct btrfs_device *device,
1553 u64 chunk_tree, u64 chunk_objectid,
1554 u64 chunk_offset, u64 start, u64 num_bytes)
1557 struct btrfs_path *path;
1558 struct btrfs_root *root = device->dev_root;
1559 struct btrfs_dev_extent *extent;
1560 struct extent_buffer *leaf;
1561 struct btrfs_key key;
1563 WARN_ON(!device->in_fs_metadata);
1564 WARN_ON(device->is_tgtdev_for_dev_replace);
1565 path = btrfs_alloc_path();
1569 key.objectid = device->devid;
1571 key.type = BTRFS_DEV_EXTENT_KEY;
1572 ret = btrfs_insert_empty_item(trans, root, path, &key,
1577 leaf = path->nodes[0];
1578 extent = btrfs_item_ptr(leaf, path->slots[0],
1579 struct btrfs_dev_extent);
1580 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1581 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1582 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1584 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1585 btrfs_dev_extent_chunk_tree_uuid(extent), BTRFS_UUID_SIZE);
1587 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1588 btrfs_mark_buffer_dirty(leaf);
1590 btrfs_free_path(path);
1594 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1596 struct extent_map_tree *em_tree;
1597 struct extent_map *em;
1601 em_tree = &fs_info->mapping_tree.map_tree;
1602 read_lock(&em_tree->lock);
1603 n = rb_last(&em_tree->map);
1605 em = rb_entry(n, struct extent_map, rb_node);
1606 ret = em->start + em->len;
1608 read_unlock(&em_tree->lock);
1613 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1617 struct btrfs_key key;
1618 struct btrfs_key found_key;
1619 struct btrfs_path *path;
1621 path = btrfs_alloc_path();
1625 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1626 key.type = BTRFS_DEV_ITEM_KEY;
1627 key.offset = (u64)-1;
1629 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1633 BUG_ON(ret == 0); /* Corruption */
1635 ret = btrfs_previous_item(fs_info->chunk_root, path,
1636 BTRFS_DEV_ITEMS_OBJECTID,
1637 BTRFS_DEV_ITEM_KEY);
1641 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1643 *devid_ret = found_key.offset + 1;
1647 btrfs_free_path(path);
1652 * the device information is stored in the chunk root
1653 * the btrfs_device struct should be fully filled in
1655 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1656 struct btrfs_root *root,
1657 struct btrfs_device *device)
1660 struct btrfs_path *path;
1661 struct btrfs_dev_item *dev_item;
1662 struct extent_buffer *leaf;
1663 struct btrfs_key key;
1666 root = root->fs_info->chunk_root;
1668 path = btrfs_alloc_path();
1672 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1673 key.type = BTRFS_DEV_ITEM_KEY;
1674 key.offset = device->devid;
1676 ret = btrfs_insert_empty_item(trans, root, path, &key,
1681 leaf = path->nodes[0];
1682 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1684 btrfs_set_device_id(leaf, dev_item, device->devid);
1685 btrfs_set_device_generation(leaf, dev_item, 0);
1686 btrfs_set_device_type(leaf, dev_item, device->type);
1687 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1688 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1689 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1690 btrfs_set_device_total_bytes(leaf, dev_item,
1691 btrfs_device_get_disk_total_bytes(device));
1692 btrfs_set_device_bytes_used(leaf, dev_item,
1693 btrfs_device_get_bytes_used(device));
1694 btrfs_set_device_group(leaf, dev_item, 0);
1695 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1696 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1697 btrfs_set_device_start_offset(leaf, dev_item, 0);
1699 ptr = btrfs_device_uuid(dev_item);
1700 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1701 ptr = btrfs_device_fsid(dev_item);
1702 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1703 btrfs_mark_buffer_dirty(leaf);
1707 btrfs_free_path(path);
1712 * Function to update ctime/mtime for a given device path.
1713 * Mainly used for ctime/mtime based probe like libblkid.
1715 static void update_dev_time(char *path_name)
1719 filp = filp_open(path_name, O_RDWR, 0);
1722 file_update_time(filp);
1723 filp_close(filp, NULL);
1726 static int btrfs_rm_dev_item(struct btrfs_root *root,
1727 struct btrfs_device *device)
1730 struct btrfs_path *path;
1731 struct btrfs_key key;
1732 struct btrfs_trans_handle *trans;
1734 root = root->fs_info->chunk_root;
1736 path = btrfs_alloc_path();
1740 trans = btrfs_start_transaction(root, 0);
1741 if (IS_ERR(trans)) {
1742 btrfs_free_path(path);
1743 return PTR_ERR(trans);
1745 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1746 key.type = BTRFS_DEV_ITEM_KEY;
1747 key.offset = device->devid;
1749 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1758 ret = btrfs_del_item(trans, root, path);
1762 btrfs_free_path(path);
1763 btrfs_commit_transaction(trans, root);
1768 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1769 * filesystem. It's up to the caller to adjust that number regarding eg. device
1772 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1780 seq = read_seqbegin(&fs_info->profiles_lock);
1782 all_avail = fs_info->avail_data_alloc_bits |
1783 fs_info->avail_system_alloc_bits |
1784 fs_info->avail_metadata_alloc_bits;
1785 } while (read_seqretry(&fs_info->profiles_lock, seq));
1787 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1788 if (!(all_avail & btrfs_raid_group[i]))
1791 if (num_devices < btrfs_raid_array[i].devs_min) {
1792 int ret = btrfs_raid_mindev_error[i];
1802 struct btrfs_device *btrfs_find_next_active_device(struct btrfs_fs_devices *fs_devs,
1803 struct btrfs_device *device)
1805 struct btrfs_device *next_device;
1807 list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1808 if (next_device != device &&
1809 !next_device->missing && next_device->bdev)
1817 * Helper function to check if the given device is part of s_bdev / latest_bdev
1818 * and replace it with the provided or the next active device, in the context
1819 * where this function called, there should be always be another device (or
1820 * this_dev) which is active.
1822 void btrfs_assign_next_active_device(struct btrfs_fs_info *fs_info,
1823 struct btrfs_device *device, struct btrfs_device *this_dev)
1825 struct btrfs_device *next_device;
1828 next_device = this_dev;
1830 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
1832 ASSERT(next_device);
1834 if (fs_info->sb->s_bdev &&
1835 (fs_info->sb->s_bdev == device->bdev))
1836 fs_info->sb->s_bdev = next_device->bdev;
1838 if (fs_info->fs_devices->latest_bdev == device->bdev)
1839 fs_info->fs_devices->latest_bdev = next_device->bdev;
1842 int btrfs_rm_device(struct btrfs_root *root, char *device_path, u64 devid)
1844 struct btrfs_device *device;
1845 struct btrfs_fs_devices *cur_devices;
1848 bool clear_super = false;
1849 char *dev_name = NULL;
1851 mutex_lock(&uuid_mutex);
1853 num_devices = root->fs_info->fs_devices->num_devices;
1854 btrfs_dev_replace_lock(&root->fs_info->dev_replace, 0);
1855 if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1856 WARN_ON(num_devices < 1);
1859 btrfs_dev_replace_unlock(&root->fs_info->dev_replace, 0);
1861 ret = btrfs_check_raid_min_devices(root->fs_info, num_devices - 1);
1865 ret = btrfs_find_device_by_devspec(root, devid, device_path,
1870 if (device->is_tgtdev_for_dev_replace) {
1871 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1875 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1876 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1880 if (device->writeable) {
1882 list_del_init(&device->dev_alloc_list);
1883 device->fs_devices->rw_devices--;
1884 unlock_chunks(root);
1885 dev_name = kstrdup(device->name->str, GFP_KERNEL);
1893 mutex_unlock(&uuid_mutex);
1894 ret = btrfs_shrink_device(device, 0);
1895 mutex_lock(&uuid_mutex);
1900 * TODO: the superblock still includes this device in its num_devices
1901 * counter although write_all_supers() is not locked out. This
1902 * could give a filesystem state which requires a degraded mount.
1904 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1908 device->in_fs_metadata = 0;
1909 btrfs_scrub_cancel_dev(root->fs_info, device);
1912 * the device list mutex makes sure that we don't change
1913 * the device list while someone else is writing out all
1914 * the device supers. Whoever is writing all supers, should
1915 * lock the device list mutex before getting the number of
1916 * devices in the super block (super_copy). Conversely,
1917 * whoever updates the number of devices in the super block
1918 * (super_copy) should hold the device list mutex.
1921 cur_devices = device->fs_devices;
1922 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1923 list_del_rcu(&device->dev_list);
1925 device->fs_devices->num_devices--;
1926 device->fs_devices->total_devices--;
1928 if (device->missing)
1929 device->fs_devices->missing_devices--;
1931 btrfs_assign_next_active_device(root->fs_info, device, NULL);
1934 device->fs_devices->open_devices--;
1935 /* remove sysfs entry */
1936 btrfs_sysfs_rm_device_link(root->fs_info->fs_devices, device);
1939 btrfs_close_bdev(device);
1941 call_rcu(&device->rcu, free_device);
1943 num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1944 btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1945 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1947 if (cur_devices->open_devices == 0) {
1948 struct btrfs_fs_devices *fs_devices;
1949 fs_devices = root->fs_info->fs_devices;
1950 while (fs_devices) {
1951 if (fs_devices->seed == cur_devices) {
1952 fs_devices->seed = cur_devices->seed;
1955 fs_devices = fs_devices->seed;
1957 cur_devices->seed = NULL;
1958 __btrfs_close_devices(cur_devices);
1959 free_fs_devices(cur_devices);
1962 root->fs_info->num_tolerated_disk_barrier_failures =
1963 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1966 * at this point, the device is zero sized. We want to
1967 * remove it from the devices list and zero out the old super
1970 struct block_device *bdev;
1972 bdev = blkdev_get_by_path(dev_name, FMODE_READ | FMODE_EXCL,
1973 root->fs_info->bdev_holder);
1974 if (!IS_ERR(bdev)) {
1975 btrfs_scratch_superblocks(bdev, dev_name);
1976 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1983 mutex_unlock(&uuid_mutex);
1987 if (device->writeable) {
1989 list_add(&device->dev_alloc_list,
1990 &root->fs_info->fs_devices->alloc_list);
1991 device->fs_devices->rw_devices++;
1992 unlock_chunks(root);
1997 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
1998 struct btrfs_device *srcdev)
2000 struct btrfs_fs_devices *fs_devices;
2002 WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
2005 * in case of fs with no seed, srcdev->fs_devices will point
2006 * to fs_devices of fs_info. However when the dev being replaced is
2007 * a seed dev it will point to the seed's local fs_devices. In short
2008 * srcdev will have its correct fs_devices in both the cases.
2010 fs_devices = srcdev->fs_devices;
2012 list_del_rcu(&srcdev->dev_list);
2013 list_del_rcu(&srcdev->dev_alloc_list);
2014 fs_devices->num_devices--;
2015 if (srcdev->missing)
2016 fs_devices->missing_devices--;
2018 if (srcdev->writeable)
2019 fs_devices->rw_devices--;
2022 fs_devices->open_devices--;
2025 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
2026 struct btrfs_device *srcdev)
2028 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2030 if (srcdev->writeable) {
2031 /* zero out the old super if it is writable */
2032 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
2035 btrfs_close_bdev(srcdev);
2037 call_rcu(&srcdev->rcu, free_device);
2040 * unless fs_devices is seed fs, num_devices shouldn't go
2043 BUG_ON(!fs_devices->num_devices && !fs_devices->seeding);
2045 /* if this is no devs we rather delete the fs_devices */
2046 if (!fs_devices->num_devices) {
2047 struct btrfs_fs_devices *tmp_fs_devices;
2049 tmp_fs_devices = fs_info->fs_devices;
2050 while (tmp_fs_devices) {
2051 if (tmp_fs_devices->seed == fs_devices) {
2052 tmp_fs_devices->seed = fs_devices->seed;
2055 tmp_fs_devices = tmp_fs_devices->seed;
2057 fs_devices->seed = NULL;
2058 __btrfs_close_devices(fs_devices);
2059 free_fs_devices(fs_devices);
2063 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2064 struct btrfs_device *tgtdev)
2066 mutex_lock(&uuid_mutex);
2068 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2070 btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev);
2073 fs_info->fs_devices->open_devices--;
2075 fs_info->fs_devices->num_devices--;
2077 btrfs_assign_next_active_device(fs_info, tgtdev, NULL);
2079 list_del_rcu(&tgtdev->dev_list);
2081 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2082 mutex_unlock(&uuid_mutex);
2085 * The update_dev_time() with in btrfs_scratch_superblocks()
2086 * may lead to a call to btrfs_show_devname() which will try
2087 * to hold device_list_mutex. And here this device
2088 * is already out of device list, so we don't have to hold
2089 * the device_list_mutex lock.
2091 btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
2093 btrfs_close_bdev(tgtdev);
2094 call_rcu(&tgtdev->rcu, free_device);
2097 static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
2098 struct btrfs_device **device)
2101 struct btrfs_super_block *disk_super;
2104 struct block_device *bdev;
2105 struct buffer_head *bh;
2108 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2109 root->fs_info->bdev_holder, 0, &bdev, &bh);
2112 disk_super = (struct btrfs_super_block *)bh->b_data;
2113 devid = btrfs_stack_device_id(&disk_super->dev_item);
2114 dev_uuid = disk_super->dev_item.uuid;
2115 *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2120 blkdev_put(bdev, FMODE_READ);
2124 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
2126 struct btrfs_device **device)
2129 if (strcmp(device_path, "missing") == 0) {
2130 struct list_head *devices;
2131 struct btrfs_device *tmp;
2133 devices = &root->fs_info->fs_devices->devices;
2135 * It is safe to read the devices since the volume_mutex
2136 * is held by the caller.
2138 list_for_each_entry(tmp, devices, dev_list) {
2139 if (tmp->in_fs_metadata && !tmp->bdev) {
2146 return BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2150 return btrfs_find_device_by_path(root, device_path, device);
2155 * Lookup a device given by device id, or the path if the id is 0.
2157 int btrfs_find_device_by_devspec(struct btrfs_root *root, u64 devid,
2159 struct btrfs_device **device)
2165 *device = btrfs_find_device(root->fs_info, devid, NULL,
2170 if (!devpath || !devpath[0])
2173 ret = btrfs_find_device_missing_or_by_path(root, devpath,
2180 * does all the dirty work required for changing file system's UUID.
2182 static int btrfs_prepare_sprout(struct btrfs_root *root)
2184 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2185 struct btrfs_fs_devices *old_devices;
2186 struct btrfs_fs_devices *seed_devices;
2187 struct btrfs_super_block *disk_super = root->fs_info->super_copy;
2188 struct btrfs_device *device;
2191 BUG_ON(!mutex_is_locked(&uuid_mutex));
2192 if (!fs_devices->seeding)
2195 seed_devices = __alloc_fs_devices();
2196 if (IS_ERR(seed_devices))
2197 return PTR_ERR(seed_devices);
2199 old_devices = clone_fs_devices(fs_devices);
2200 if (IS_ERR(old_devices)) {
2201 kfree(seed_devices);
2202 return PTR_ERR(old_devices);
2205 list_add(&old_devices->list, &fs_uuids);
2207 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2208 seed_devices->opened = 1;
2209 INIT_LIST_HEAD(&seed_devices->devices);
2210 INIT_LIST_HEAD(&seed_devices->alloc_list);
2211 mutex_init(&seed_devices->device_list_mutex);
2213 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2214 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2216 list_for_each_entry(device, &seed_devices->devices, dev_list)
2217 device->fs_devices = seed_devices;
2220 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2221 unlock_chunks(root);
2223 fs_devices->seeding = 0;
2224 fs_devices->num_devices = 0;
2225 fs_devices->open_devices = 0;
2226 fs_devices->missing_devices = 0;
2227 fs_devices->rotating = 0;
2228 fs_devices->seed = seed_devices;
2230 generate_random_uuid(fs_devices->fsid);
2231 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2232 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2233 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2235 super_flags = btrfs_super_flags(disk_super) &
2236 ~BTRFS_SUPER_FLAG_SEEDING;
2237 btrfs_set_super_flags(disk_super, super_flags);
2243 * Store the expected generation for seed devices in device items.
2245 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2246 struct btrfs_root *root)
2248 struct btrfs_path *path;
2249 struct extent_buffer *leaf;
2250 struct btrfs_dev_item *dev_item;
2251 struct btrfs_device *device;
2252 struct btrfs_key key;
2253 u8 fs_uuid[BTRFS_UUID_SIZE];
2254 u8 dev_uuid[BTRFS_UUID_SIZE];
2258 path = btrfs_alloc_path();
2262 root = root->fs_info->chunk_root;
2263 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2265 key.type = BTRFS_DEV_ITEM_KEY;
2268 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2272 leaf = path->nodes[0];
2274 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2275 ret = btrfs_next_leaf(root, path);
2280 leaf = path->nodes[0];
2281 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2282 btrfs_release_path(path);
2286 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2287 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2288 key.type != BTRFS_DEV_ITEM_KEY)
2291 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2292 struct btrfs_dev_item);
2293 devid = btrfs_device_id(leaf, dev_item);
2294 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2296 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2298 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2300 BUG_ON(!device); /* Logic error */
2302 if (device->fs_devices->seeding) {
2303 btrfs_set_device_generation(leaf, dev_item,
2304 device->generation);
2305 btrfs_mark_buffer_dirty(leaf);
2313 btrfs_free_path(path);
2317 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
2319 struct request_queue *q;
2320 struct btrfs_trans_handle *trans;
2321 struct btrfs_device *device;
2322 struct block_device *bdev;
2323 struct list_head *devices;
2324 struct super_block *sb = root->fs_info->sb;
2325 struct rcu_string *name;
2327 int seeding_dev = 0;
2330 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
2333 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2334 root->fs_info->bdev_holder);
2336 return PTR_ERR(bdev);
2338 if (root->fs_info->fs_devices->seeding) {
2340 down_write(&sb->s_umount);
2341 mutex_lock(&uuid_mutex);
2344 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2346 devices = &root->fs_info->fs_devices->devices;
2348 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2349 list_for_each_entry(device, devices, dev_list) {
2350 if (device->bdev == bdev) {
2353 &root->fs_info->fs_devices->device_list_mutex);
2357 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2359 device = btrfs_alloc_device(root->fs_info, NULL, NULL);
2360 if (IS_ERR(device)) {
2361 /* we can safely leave the fs_devices entry around */
2362 ret = PTR_ERR(device);
2366 name = rcu_string_strdup(device_path, GFP_KERNEL);
2372 rcu_assign_pointer(device->name, name);
2374 trans = btrfs_start_transaction(root, 0);
2375 if (IS_ERR(trans)) {
2376 rcu_string_free(device->name);
2378 ret = PTR_ERR(trans);
2382 q = bdev_get_queue(bdev);
2383 if (blk_queue_discard(q))
2384 device->can_discard = 1;
2385 device->writeable = 1;
2386 device->generation = trans->transid;
2387 device->io_width = root->sectorsize;
2388 device->io_align = root->sectorsize;
2389 device->sector_size = root->sectorsize;
2390 device->total_bytes = i_size_read(bdev->bd_inode);
2391 device->disk_total_bytes = device->total_bytes;
2392 device->commit_total_bytes = device->total_bytes;
2393 device->dev_root = root->fs_info->dev_root;
2394 device->bdev = bdev;
2395 device->in_fs_metadata = 1;
2396 device->is_tgtdev_for_dev_replace = 0;
2397 device->mode = FMODE_EXCL;
2398 device->dev_stats_valid = 1;
2399 set_blocksize(device->bdev, 4096);
2402 sb->s_flags &= ~MS_RDONLY;
2403 ret = btrfs_prepare_sprout(root);
2404 BUG_ON(ret); /* -ENOMEM */
2407 device->fs_devices = root->fs_info->fs_devices;
2409 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2411 list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
2412 list_add(&device->dev_alloc_list,
2413 &root->fs_info->fs_devices->alloc_list);
2414 root->fs_info->fs_devices->num_devices++;
2415 root->fs_info->fs_devices->open_devices++;
2416 root->fs_info->fs_devices->rw_devices++;
2417 root->fs_info->fs_devices->total_devices++;
2418 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2420 spin_lock(&root->fs_info->free_chunk_lock);
2421 root->fs_info->free_chunk_space += device->total_bytes;
2422 spin_unlock(&root->fs_info->free_chunk_lock);
2424 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2425 root->fs_info->fs_devices->rotating = 1;
2427 tmp = btrfs_super_total_bytes(root->fs_info->super_copy);
2428 btrfs_set_super_total_bytes(root->fs_info->super_copy,
2429 tmp + device->total_bytes);
2431 tmp = btrfs_super_num_devices(root->fs_info->super_copy);
2432 btrfs_set_super_num_devices(root->fs_info->super_copy,
2435 /* add sysfs device entry */
2436 btrfs_sysfs_add_device_link(root->fs_info->fs_devices, device);
2439 * we've got more storage, clear any full flags on the space
2442 btrfs_clear_space_info_full(root->fs_info);
2444 unlock_chunks(root);
2445 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2449 ret = init_first_rw_device(trans, root, device);
2450 unlock_chunks(root);
2452 btrfs_abort_transaction(trans, ret);
2457 ret = btrfs_add_device(trans, root, device);
2459 btrfs_abort_transaction(trans, ret);
2464 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2466 ret = btrfs_finish_sprout(trans, root);
2468 btrfs_abort_transaction(trans, ret);
2472 /* Sprouting would change fsid of the mounted root,
2473 * so rename the fsid on the sysfs
2475 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2476 root->fs_info->fsid);
2477 if (kobject_rename(&root->fs_info->fs_devices->fsid_kobj,
2479 btrfs_warn(root->fs_info,
2480 "sysfs: failed to create fsid for sprout");
2483 root->fs_info->num_tolerated_disk_barrier_failures =
2484 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
2485 ret = btrfs_commit_transaction(trans, root);
2488 mutex_unlock(&uuid_mutex);
2489 up_write(&sb->s_umount);
2491 if (ret) /* transaction commit */
2494 ret = btrfs_relocate_sys_chunks(root);
2496 btrfs_handle_fs_error(root->fs_info, ret,
2497 "Failed to relocate sys chunks after "
2498 "device initialization. This can be fixed "
2499 "using the \"btrfs balance\" command.");
2500 trans = btrfs_attach_transaction(root);
2501 if (IS_ERR(trans)) {
2502 if (PTR_ERR(trans) == -ENOENT)
2504 return PTR_ERR(trans);
2506 ret = btrfs_commit_transaction(trans, root);
2509 /* Update ctime/mtime for libblkid */
2510 update_dev_time(device_path);
2514 btrfs_end_transaction(trans, root);
2515 rcu_string_free(device->name);
2516 btrfs_sysfs_rm_device_link(root->fs_info->fs_devices, device);
2519 blkdev_put(bdev, FMODE_EXCL);
2521 mutex_unlock(&uuid_mutex);
2522 up_write(&sb->s_umount);
2527 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2528 struct btrfs_device *srcdev,
2529 struct btrfs_device **device_out)
2531 struct request_queue *q;
2532 struct btrfs_device *device;
2533 struct block_device *bdev;
2534 struct btrfs_fs_info *fs_info = root->fs_info;
2535 struct list_head *devices;
2536 struct rcu_string *name;
2537 u64 devid = BTRFS_DEV_REPLACE_DEVID;
2541 if (fs_info->fs_devices->seeding) {
2542 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
2546 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2547 fs_info->bdev_holder);
2549 btrfs_err(fs_info, "target device %s is invalid!", device_path);
2550 return PTR_ERR(bdev);
2553 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2555 devices = &fs_info->fs_devices->devices;
2556 list_for_each_entry(device, devices, dev_list) {
2557 if (device->bdev == bdev) {
2558 btrfs_err(fs_info, "target device is in the filesystem!");
2565 if (i_size_read(bdev->bd_inode) <
2566 btrfs_device_get_total_bytes(srcdev)) {
2567 btrfs_err(fs_info, "target device is smaller than source device!");
2573 device = btrfs_alloc_device(NULL, &devid, NULL);
2574 if (IS_ERR(device)) {
2575 ret = PTR_ERR(device);
2579 name = rcu_string_strdup(device_path, GFP_NOFS);
2585 rcu_assign_pointer(device->name, name);
2587 q = bdev_get_queue(bdev);
2588 if (blk_queue_discard(q))
2589 device->can_discard = 1;
2590 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2591 device->writeable = 1;
2592 device->generation = 0;
2593 device->io_width = root->sectorsize;
2594 device->io_align = root->sectorsize;
2595 device->sector_size = root->sectorsize;
2596 device->total_bytes = btrfs_device_get_total_bytes(srcdev);
2597 device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
2598 device->bytes_used = btrfs_device_get_bytes_used(srcdev);
2599 ASSERT(list_empty(&srcdev->resized_list));
2600 device->commit_total_bytes = srcdev->commit_total_bytes;
2601 device->commit_bytes_used = device->bytes_used;
2602 device->dev_root = fs_info->dev_root;
2603 device->bdev = bdev;
2604 device->in_fs_metadata = 1;
2605 device->is_tgtdev_for_dev_replace = 1;
2606 device->mode = FMODE_EXCL;
2607 device->dev_stats_valid = 1;
2608 set_blocksize(device->bdev, 4096);
2609 device->fs_devices = fs_info->fs_devices;
2610 list_add(&device->dev_list, &fs_info->fs_devices->devices);
2611 fs_info->fs_devices->num_devices++;
2612 fs_info->fs_devices->open_devices++;
2613 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2615 *device_out = device;
2619 blkdev_put(bdev, FMODE_EXCL);
2623 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2624 struct btrfs_device *tgtdev)
2626 WARN_ON(fs_info->fs_devices->rw_devices == 0);
2627 tgtdev->io_width = fs_info->dev_root->sectorsize;
2628 tgtdev->io_align = fs_info->dev_root->sectorsize;
2629 tgtdev->sector_size = fs_info->dev_root->sectorsize;
2630 tgtdev->dev_root = fs_info->dev_root;
2631 tgtdev->in_fs_metadata = 1;
2634 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2635 struct btrfs_device *device)
2638 struct btrfs_path *path;
2639 struct btrfs_root *root;
2640 struct btrfs_dev_item *dev_item;
2641 struct extent_buffer *leaf;
2642 struct btrfs_key key;
2644 root = device->dev_root->fs_info->chunk_root;
2646 path = btrfs_alloc_path();
2650 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2651 key.type = BTRFS_DEV_ITEM_KEY;
2652 key.offset = device->devid;
2654 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2663 leaf = path->nodes[0];
2664 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2666 btrfs_set_device_id(leaf, dev_item, device->devid);
2667 btrfs_set_device_type(leaf, dev_item, device->type);
2668 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2669 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2670 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2671 btrfs_set_device_total_bytes(leaf, dev_item,
2672 btrfs_device_get_disk_total_bytes(device));
2673 btrfs_set_device_bytes_used(leaf, dev_item,
2674 btrfs_device_get_bytes_used(device));
2675 btrfs_mark_buffer_dirty(leaf);
2678 btrfs_free_path(path);
2682 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2683 struct btrfs_device *device, u64 new_size)
2685 struct btrfs_super_block *super_copy =
2686 device->dev_root->fs_info->super_copy;
2687 struct btrfs_fs_devices *fs_devices;
2691 if (!device->writeable)
2694 lock_chunks(device->dev_root);
2695 old_total = btrfs_super_total_bytes(super_copy);
2696 diff = new_size - device->total_bytes;
2698 if (new_size <= device->total_bytes ||
2699 device->is_tgtdev_for_dev_replace) {
2700 unlock_chunks(device->dev_root);
2704 fs_devices = device->dev_root->fs_info->fs_devices;
2706 btrfs_set_super_total_bytes(super_copy, old_total + diff);
2707 device->fs_devices->total_rw_bytes += diff;
2709 btrfs_device_set_total_bytes(device, new_size);
2710 btrfs_device_set_disk_total_bytes(device, new_size);
2711 btrfs_clear_space_info_full(device->dev_root->fs_info);
2712 if (list_empty(&device->resized_list))
2713 list_add_tail(&device->resized_list,
2714 &fs_devices->resized_devices);
2715 unlock_chunks(device->dev_root);
2717 return btrfs_update_device(trans, device);
2720 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2721 struct btrfs_root *root, u64 chunk_objectid,
2725 struct btrfs_path *path;
2726 struct btrfs_key key;
2728 root = root->fs_info->chunk_root;
2729 path = btrfs_alloc_path();
2733 key.objectid = chunk_objectid;
2734 key.offset = chunk_offset;
2735 key.type = BTRFS_CHUNK_ITEM_KEY;
2737 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2740 else if (ret > 0) { /* Logic error or corruption */
2741 btrfs_handle_fs_error(root->fs_info, -ENOENT,
2742 "Failed lookup while freeing chunk.");
2747 ret = btrfs_del_item(trans, root, path);
2749 btrfs_handle_fs_error(root->fs_info, ret,
2750 "Failed to delete chunk item.");
2752 btrfs_free_path(path);
2756 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2759 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2760 struct btrfs_disk_key *disk_key;
2761 struct btrfs_chunk *chunk;
2768 struct btrfs_key key;
2771 array_size = btrfs_super_sys_array_size(super_copy);
2773 ptr = super_copy->sys_chunk_array;
2776 while (cur < array_size) {
2777 disk_key = (struct btrfs_disk_key *)ptr;
2778 btrfs_disk_key_to_cpu(&key, disk_key);
2780 len = sizeof(*disk_key);
2782 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2783 chunk = (struct btrfs_chunk *)(ptr + len);
2784 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2785 len += btrfs_chunk_item_size(num_stripes);
2790 if (key.objectid == chunk_objectid &&
2791 key.offset == chunk_offset) {
2792 memmove(ptr, ptr + len, array_size - (cur + len));
2794 btrfs_set_super_sys_array_size(super_copy, array_size);
2800 unlock_chunks(root);
2804 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2805 struct btrfs_root *root, u64 chunk_offset)
2807 struct extent_map_tree *em_tree;
2808 struct extent_map *em;
2809 struct btrfs_root *extent_root = root->fs_info->extent_root;
2810 struct map_lookup *map;
2811 u64 dev_extent_len = 0;
2812 u64 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2814 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2817 root = root->fs_info->chunk_root;
2818 em_tree = &root->fs_info->mapping_tree.map_tree;
2820 read_lock(&em_tree->lock);
2821 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2822 read_unlock(&em_tree->lock);
2824 if (!em || em->start > chunk_offset ||
2825 em->start + em->len < chunk_offset) {
2827 * This is a logic error, but we don't want to just rely on the
2828 * user having built with ASSERT enabled, so if ASSERT doesn't
2829 * do anything we still error out.
2833 free_extent_map(em);
2836 map = em->map_lookup;
2837 lock_chunks(root->fs_info->chunk_root);
2838 check_system_chunk(trans, extent_root, map->type);
2839 unlock_chunks(root->fs_info->chunk_root);
2842 * Take the device list mutex to prevent races with the final phase of
2843 * a device replace operation that replaces the device object associated
2844 * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2846 mutex_lock(&fs_devices->device_list_mutex);
2847 for (i = 0; i < map->num_stripes; i++) {
2848 struct btrfs_device *device = map->stripes[i].dev;
2849 ret = btrfs_free_dev_extent(trans, device,
2850 map->stripes[i].physical,
2853 mutex_unlock(&fs_devices->device_list_mutex);
2854 btrfs_abort_transaction(trans, ret);
2858 if (device->bytes_used > 0) {
2860 btrfs_device_set_bytes_used(device,
2861 device->bytes_used - dev_extent_len);
2862 spin_lock(&root->fs_info->free_chunk_lock);
2863 root->fs_info->free_chunk_space += dev_extent_len;
2864 spin_unlock(&root->fs_info->free_chunk_lock);
2865 btrfs_clear_space_info_full(root->fs_info);
2866 unlock_chunks(root);
2869 if (map->stripes[i].dev) {
2870 ret = btrfs_update_device(trans, map->stripes[i].dev);
2872 mutex_unlock(&fs_devices->device_list_mutex);
2873 btrfs_abort_transaction(trans, ret);
2878 mutex_unlock(&fs_devices->device_list_mutex);
2880 ret = btrfs_free_chunk(trans, root, chunk_objectid, chunk_offset);
2882 btrfs_abort_transaction(trans, ret);
2886 trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2888 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2889 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2891 btrfs_abort_transaction(trans, ret);
2896 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset, em);
2898 btrfs_abort_transaction(trans, ret);
2904 free_extent_map(em);
2908 static int btrfs_relocate_chunk(struct btrfs_root *root, u64 chunk_offset)
2910 struct btrfs_root *extent_root;
2911 struct btrfs_trans_handle *trans;
2914 root = root->fs_info->chunk_root;
2915 extent_root = root->fs_info->extent_root;
2918 * Prevent races with automatic removal of unused block groups.
2919 * After we relocate and before we remove the chunk with offset
2920 * chunk_offset, automatic removal of the block group can kick in,
2921 * resulting in a failure when calling btrfs_remove_chunk() below.
2923 * Make sure to acquire this mutex before doing a tree search (dev
2924 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2925 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2926 * we release the path used to search the chunk/dev tree and before
2927 * the current task acquires this mutex and calls us.
2929 ASSERT(mutex_is_locked(&root->fs_info->delete_unused_bgs_mutex));
2931 ret = btrfs_can_relocate(extent_root, chunk_offset);
2935 /* step one, relocate all the extents inside this chunk */
2936 btrfs_scrub_pause(root);
2937 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2938 btrfs_scrub_continue(root);
2942 trans = btrfs_start_trans_remove_block_group(root->fs_info,
2944 if (IS_ERR(trans)) {
2945 ret = PTR_ERR(trans);
2946 btrfs_handle_fs_error(root->fs_info, ret, NULL);
2951 * step two, delete the device extents and the
2952 * chunk tree entries
2954 ret = btrfs_remove_chunk(trans, root, chunk_offset);
2955 btrfs_end_transaction(trans, extent_root);
2959 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2961 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2962 struct btrfs_path *path;
2963 struct extent_buffer *leaf;
2964 struct btrfs_chunk *chunk;
2965 struct btrfs_key key;
2966 struct btrfs_key found_key;
2968 bool retried = false;
2972 path = btrfs_alloc_path();
2977 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2978 key.offset = (u64)-1;
2979 key.type = BTRFS_CHUNK_ITEM_KEY;
2982 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
2983 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2985 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2988 BUG_ON(ret == 0); /* Corruption */
2990 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2993 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2999 leaf = path->nodes[0];
3000 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3002 chunk = btrfs_item_ptr(leaf, path->slots[0],
3003 struct btrfs_chunk);
3004 chunk_type = btrfs_chunk_type(leaf, chunk);
3005 btrfs_release_path(path);
3007 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3008 ret = btrfs_relocate_chunk(chunk_root,
3015 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
3017 if (found_key.offset == 0)
3019 key.offset = found_key.offset - 1;
3022 if (failed && !retried) {
3026 } else if (WARN_ON(failed && retried)) {
3030 btrfs_free_path(path);
3034 static int insert_balance_item(struct btrfs_root *root,
3035 struct btrfs_balance_control *bctl)
3037 struct btrfs_trans_handle *trans;
3038 struct btrfs_balance_item *item;
3039 struct btrfs_disk_balance_args disk_bargs;
3040 struct btrfs_path *path;
3041 struct extent_buffer *leaf;
3042 struct btrfs_key key;
3045 path = btrfs_alloc_path();
3049 trans = btrfs_start_transaction(root, 0);
3050 if (IS_ERR(trans)) {
3051 btrfs_free_path(path);
3052 return PTR_ERR(trans);
3055 key.objectid = BTRFS_BALANCE_OBJECTID;
3056 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3059 ret = btrfs_insert_empty_item(trans, root, path, &key,
3064 leaf = path->nodes[0];
3065 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3067 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
3069 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3070 btrfs_set_balance_data(leaf, item, &disk_bargs);
3071 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3072 btrfs_set_balance_meta(leaf, item, &disk_bargs);
3073 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3074 btrfs_set_balance_sys(leaf, item, &disk_bargs);
3076 btrfs_set_balance_flags(leaf, item, bctl->flags);
3078 btrfs_mark_buffer_dirty(leaf);
3080 btrfs_free_path(path);
3081 err = btrfs_commit_transaction(trans, root);
3087 static int del_balance_item(struct btrfs_root *root)
3089 struct btrfs_trans_handle *trans;
3090 struct btrfs_path *path;
3091 struct btrfs_key key;
3094 path = btrfs_alloc_path();
3098 trans = btrfs_start_transaction(root, 0);
3099 if (IS_ERR(trans)) {
3100 btrfs_free_path(path);
3101 return PTR_ERR(trans);
3104 key.objectid = BTRFS_BALANCE_OBJECTID;
3105 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3108 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3116 ret = btrfs_del_item(trans, root, path);
3118 btrfs_free_path(path);
3119 err = btrfs_commit_transaction(trans, root);
3126 * This is a heuristic used to reduce the number of chunks balanced on
3127 * resume after balance was interrupted.
3129 static void update_balance_args(struct btrfs_balance_control *bctl)
3132 * Turn on soft mode for chunk types that were being converted.
3134 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3135 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3136 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3137 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3138 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3139 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3142 * Turn on usage filter if is not already used. The idea is
3143 * that chunks that we have already balanced should be
3144 * reasonably full. Don't do it for chunks that are being
3145 * converted - that will keep us from relocating unconverted
3146 * (albeit full) chunks.
3148 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3149 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3150 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3151 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3152 bctl->data.usage = 90;
3154 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3155 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3156 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3157 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3158 bctl->sys.usage = 90;
3160 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3161 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3162 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3163 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3164 bctl->meta.usage = 90;
3169 * Should be called with both balance and volume mutexes held to
3170 * serialize other volume operations (add_dev/rm_dev/resize) with
3171 * restriper. Same goes for unset_balance_control.
3173 static void set_balance_control(struct btrfs_balance_control *bctl)
3175 struct btrfs_fs_info *fs_info = bctl->fs_info;
3177 BUG_ON(fs_info->balance_ctl);
3179 spin_lock(&fs_info->balance_lock);
3180 fs_info->balance_ctl = bctl;
3181 spin_unlock(&fs_info->balance_lock);
3184 static void unset_balance_control(struct btrfs_fs_info *fs_info)
3186 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3188 BUG_ON(!fs_info->balance_ctl);
3190 spin_lock(&fs_info->balance_lock);
3191 fs_info->balance_ctl = NULL;
3192 spin_unlock(&fs_info->balance_lock);
3198 * Balance filters. Return 1 if chunk should be filtered out
3199 * (should not be balanced).
3201 static int chunk_profiles_filter(u64 chunk_type,
3202 struct btrfs_balance_args *bargs)
3204 chunk_type = chunk_to_extended(chunk_type) &
3205 BTRFS_EXTENDED_PROFILE_MASK;
3207 if (bargs->profiles & chunk_type)
3213 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3214 struct btrfs_balance_args *bargs)
3216 struct btrfs_block_group_cache *cache;
3218 u64 user_thresh_min;
3219 u64 user_thresh_max;
3222 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3223 chunk_used = btrfs_block_group_used(&cache->item);
3225 if (bargs->usage_min == 0)
3226 user_thresh_min = 0;
3228 user_thresh_min = div_factor_fine(cache->key.offset,
3231 if (bargs->usage_max == 0)
3232 user_thresh_max = 1;
3233 else if (bargs->usage_max > 100)
3234 user_thresh_max = cache->key.offset;
3236 user_thresh_max = div_factor_fine(cache->key.offset,
3239 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3242 btrfs_put_block_group(cache);
3246 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3247 u64 chunk_offset, struct btrfs_balance_args *bargs)
3249 struct btrfs_block_group_cache *cache;
3250 u64 chunk_used, user_thresh;
3253 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3254 chunk_used = btrfs_block_group_used(&cache->item);
3256 if (bargs->usage_min == 0)
3258 else if (bargs->usage > 100)
3259 user_thresh = cache->key.offset;
3261 user_thresh = div_factor_fine(cache->key.offset,
3264 if (chunk_used < user_thresh)
3267 btrfs_put_block_group(cache);
3271 static int chunk_devid_filter(struct extent_buffer *leaf,
3272 struct btrfs_chunk *chunk,
3273 struct btrfs_balance_args *bargs)
3275 struct btrfs_stripe *stripe;
3276 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3279 for (i = 0; i < num_stripes; i++) {
3280 stripe = btrfs_stripe_nr(chunk, i);
3281 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3288 /* [pstart, pend) */
3289 static int chunk_drange_filter(struct extent_buffer *leaf,
3290 struct btrfs_chunk *chunk,
3292 struct btrfs_balance_args *bargs)
3294 struct btrfs_stripe *stripe;
3295 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3301 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3304 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3305 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3306 factor = num_stripes / 2;
3307 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3308 factor = num_stripes - 1;
3309 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3310 factor = num_stripes - 2;
3312 factor = num_stripes;
3315 for (i = 0; i < num_stripes; i++) {
3316 stripe = btrfs_stripe_nr(chunk, i);
3317 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3320 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3321 stripe_length = btrfs_chunk_length(leaf, chunk);
3322 stripe_length = div_u64(stripe_length, factor);
3324 if (stripe_offset < bargs->pend &&
3325 stripe_offset + stripe_length > bargs->pstart)
3332 /* [vstart, vend) */
3333 static int chunk_vrange_filter(struct extent_buffer *leaf,
3334 struct btrfs_chunk *chunk,
3336 struct btrfs_balance_args *bargs)
3338 if (chunk_offset < bargs->vend &&
3339 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3340 /* at least part of the chunk is inside this vrange */
3346 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3347 struct btrfs_chunk *chunk,
3348 struct btrfs_balance_args *bargs)
3350 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3352 if (bargs->stripes_min <= num_stripes
3353 && num_stripes <= bargs->stripes_max)
3359 static int chunk_soft_convert_filter(u64 chunk_type,
3360 struct btrfs_balance_args *bargs)
3362 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3365 chunk_type = chunk_to_extended(chunk_type) &
3366 BTRFS_EXTENDED_PROFILE_MASK;
3368 if (bargs->target == chunk_type)
3374 static int should_balance_chunk(struct btrfs_root *root,
3375 struct extent_buffer *leaf,
3376 struct btrfs_chunk *chunk, u64 chunk_offset)
3378 struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
3379 struct btrfs_balance_args *bargs = NULL;
3380 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3383 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3384 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3388 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3389 bargs = &bctl->data;
3390 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3392 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3393 bargs = &bctl->meta;
3395 /* profiles filter */
3396 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3397 chunk_profiles_filter(chunk_type, bargs)) {
3402 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3403 chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
3405 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3406 chunk_usage_range_filter(bctl->fs_info, chunk_offset, bargs)) {
3411 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3412 chunk_devid_filter(leaf, chunk, bargs)) {
3416 /* drange filter, makes sense only with devid filter */
3417 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3418 chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
3423 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3424 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3428 /* stripes filter */
3429 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3430 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3434 /* soft profile changing mode */
3435 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3436 chunk_soft_convert_filter(chunk_type, bargs)) {
3441 * limited by count, must be the last filter
3443 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3444 if (bargs->limit == 0)
3448 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3450 * Same logic as the 'limit' filter; the minimum cannot be
3451 * determined here because we do not have the global information
3452 * about the count of all chunks that satisfy the filters.
3454 if (bargs->limit_max == 0)
3463 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3465 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3466 struct btrfs_root *chunk_root = fs_info->chunk_root;
3467 struct btrfs_root *dev_root = fs_info->dev_root;
3468 struct list_head *devices;
3469 struct btrfs_device *device;
3473 struct btrfs_chunk *chunk;
3474 struct btrfs_path *path = NULL;
3475 struct btrfs_key key;
3476 struct btrfs_key found_key;
3477 struct btrfs_trans_handle *trans;
3478 struct extent_buffer *leaf;
3481 int enospc_errors = 0;
3482 bool counting = true;
3483 /* The single value limit and min/max limits use the same bytes in the */
3484 u64 limit_data = bctl->data.limit;
3485 u64 limit_meta = bctl->meta.limit;
3486 u64 limit_sys = bctl->sys.limit;
3490 int chunk_reserved = 0;
3493 /* step one make some room on all the devices */
3494 devices = &fs_info->fs_devices->devices;
3495 list_for_each_entry(device, devices, dev_list) {
3496 old_size = btrfs_device_get_total_bytes(device);
3497 size_to_free = div_factor(old_size, 1);
3498 size_to_free = min_t(u64, size_to_free, SZ_1M);
3499 if (!device->writeable ||
3500 btrfs_device_get_total_bytes(device) -
3501 btrfs_device_get_bytes_used(device) > size_to_free ||
3502 device->is_tgtdev_for_dev_replace)
3505 ret = btrfs_shrink_device(device, old_size - size_to_free);
3509 /* btrfs_shrink_device never returns ret > 0 */
3514 trans = btrfs_start_transaction(dev_root, 0);
3515 if (IS_ERR(trans)) {
3516 ret = PTR_ERR(trans);
3517 btrfs_info_in_rcu(fs_info,
3518 "resize: unable to start transaction after shrinking device %s (error %d), old size %llu, new size %llu",
3519 rcu_str_deref(device->name), ret,
3520 old_size, old_size - size_to_free);
3524 ret = btrfs_grow_device(trans, device, old_size);
3526 btrfs_end_transaction(trans, dev_root);
3527 /* btrfs_grow_device never returns ret > 0 */
3529 btrfs_info_in_rcu(fs_info,
3530 "resize: unable to grow device after shrinking device %s (error %d), old size %llu, new size %llu",
3531 rcu_str_deref(device->name), ret,
3532 old_size, old_size - size_to_free);
3536 btrfs_end_transaction(trans, dev_root);
3539 /* step two, relocate all the chunks */
3540 path = btrfs_alloc_path();
3546 /* zero out stat counters */
3547 spin_lock(&fs_info->balance_lock);
3548 memset(&bctl->stat, 0, sizeof(bctl->stat));
3549 spin_unlock(&fs_info->balance_lock);
3553 * The single value limit and min/max limits use the same bytes
3556 bctl->data.limit = limit_data;
3557 bctl->meta.limit = limit_meta;
3558 bctl->sys.limit = limit_sys;
3560 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3561 key.offset = (u64)-1;
3562 key.type = BTRFS_CHUNK_ITEM_KEY;
3565 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3566 atomic_read(&fs_info->balance_cancel_req)) {
3571 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3572 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3574 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3579 * this shouldn't happen, it means the last relocate
3583 BUG(); /* FIXME break ? */
3585 ret = btrfs_previous_item(chunk_root, path, 0,
3586 BTRFS_CHUNK_ITEM_KEY);
3588 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3593 leaf = path->nodes[0];
3594 slot = path->slots[0];
3595 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3597 if (found_key.objectid != key.objectid) {
3598 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3602 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3603 chunk_type = btrfs_chunk_type(leaf, chunk);
3606 spin_lock(&fs_info->balance_lock);
3607 bctl->stat.considered++;
3608 spin_unlock(&fs_info->balance_lock);
3611 ret = should_balance_chunk(chunk_root, leaf, chunk,
3614 btrfs_release_path(path);
3616 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3621 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3622 spin_lock(&fs_info->balance_lock);
3623 bctl->stat.expected++;
3624 spin_unlock(&fs_info->balance_lock);
3626 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3628 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3630 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3637 * Apply limit_min filter, no need to check if the LIMITS
3638 * filter is used, limit_min is 0 by default
3640 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3641 count_data < bctl->data.limit_min)
3642 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3643 count_meta < bctl->meta.limit_min)
3644 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3645 count_sys < bctl->sys.limit_min)) {
3646 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3650 ASSERT(fs_info->data_sinfo);
3651 spin_lock(&fs_info->data_sinfo->lock);
3652 bytes_used = fs_info->data_sinfo->bytes_used;
3653 spin_unlock(&fs_info->data_sinfo->lock);
3655 if ((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3656 !chunk_reserved && !bytes_used) {
3657 trans = btrfs_start_transaction(chunk_root, 0);
3658 if (IS_ERR(trans)) {
3659 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3660 ret = PTR_ERR(trans);
3664 ret = btrfs_force_chunk_alloc(trans, chunk_root,
3665 BTRFS_BLOCK_GROUP_DATA);
3666 btrfs_end_transaction(trans, chunk_root);
3668 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3674 ret = btrfs_relocate_chunk(chunk_root,
3676 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3677 if (ret && ret != -ENOSPC)
3679 if (ret == -ENOSPC) {
3682 spin_lock(&fs_info->balance_lock);
3683 bctl->stat.completed++;
3684 spin_unlock(&fs_info->balance_lock);
3687 if (found_key.offset == 0)
3689 key.offset = found_key.offset - 1;
3693 btrfs_release_path(path);
3698 btrfs_free_path(path);
3699 if (enospc_errors) {
3700 btrfs_info(fs_info, "%d enospc errors during balance",
3710 * alloc_profile_is_valid - see if a given profile is valid and reduced
3711 * @flags: profile to validate
3712 * @extended: if true @flags is treated as an extended profile
3714 static int alloc_profile_is_valid(u64 flags, int extended)
3716 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3717 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3719 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3721 /* 1) check that all other bits are zeroed */
3725 /* 2) see if profile is reduced */
3727 return !extended; /* "0" is valid for usual profiles */
3729 /* true if exactly one bit set */
3730 return (flags & (flags - 1)) == 0;
3733 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3735 /* cancel requested || normal exit path */
3736 return atomic_read(&fs_info->balance_cancel_req) ||
3737 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3738 atomic_read(&fs_info->balance_cancel_req) == 0);
3741 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3745 unset_balance_control(fs_info);
3746 ret = del_balance_item(fs_info->tree_root);
3748 btrfs_handle_fs_error(fs_info, ret, NULL);
3750 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3753 /* Non-zero return value signifies invalidity */
3754 static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg,
3757 return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3758 (!alloc_profile_is_valid(bctl_arg->target, 1) ||
3759 (bctl_arg->target & ~allowed)));
3763 * Should be called with both balance and volume mutexes held
3765 int btrfs_balance(struct btrfs_balance_control *bctl,
3766 struct btrfs_ioctl_balance_args *bargs)
3768 struct btrfs_fs_info *fs_info = bctl->fs_info;
3775 if (btrfs_fs_closing(fs_info) ||
3776 atomic_read(&fs_info->balance_pause_req) ||
3777 atomic_read(&fs_info->balance_cancel_req)) {
3782 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3783 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3787 * In case of mixed groups both data and meta should be picked,
3788 * and identical options should be given for both of them.
3790 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3791 if (mixed && (bctl->flags & allowed)) {
3792 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3793 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3794 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3795 btrfs_err(fs_info, "with mixed groups data and "
3796 "metadata balance options must be the same");
3802 num_devices = fs_info->fs_devices->num_devices;
3803 btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
3804 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3805 BUG_ON(num_devices < 1);
3808 btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
3809 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE | BTRFS_BLOCK_GROUP_DUP;
3810 if (num_devices > 1)
3811 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3812 if (num_devices > 2)
3813 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3814 if (num_devices > 3)
3815 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3816 BTRFS_BLOCK_GROUP_RAID6);
3817 if (validate_convert_profile(&bctl->data, allowed)) {
3818 btrfs_err(fs_info, "unable to start balance with target "
3819 "data profile %llu",
3824 if (validate_convert_profile(&bctl->meta, allowed)) {
3826 "unable to start balance with target metadata profile %llu",
3831 if (validate_convert_profile(&bctl->sys, allowed)) {
3833 "unable to start balance with target system profile %llu",
3839 /* allow to reduce meta or sys integrity only if force set */
3840 allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3841 BTRFS_BLOCK_GROUP_RAID10 |
3842 BTRFS_BLOCK_GROUP_RAID5 |
3843 BTRFS_BLOCK_GROUP_RAID6;
3845 seq = read_seqbegin(&fs_info->profiles_lock);
3847 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3848 (fs_info->avail_system_alloc_bits & allowed) &&
3849 !(bctl->sys.target & allowed)) ||
3850 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3851 (fs_info->avail_metadata_alloc_bits & allowed) &&
3852 !(bctl->meta.target & allowed))) {
3853 if (bctl->flags & BTRFS_BALANCE_FORCE) {
3854 btrfs_info(fs_info, "force reducing metadata integrity");
3856 btrfs_err(fs_info, "balance will reduce metadata "
3857 "integrity, use force if you want this");
3862 } while (read_seqretry(&fs_info->profiles_lock, seq));
3864 if (btrfs_get_num_tolerated_disk_barrier_failures(bctl->meta.target) <
3865 btrfs_get_num_tolerated_disk_barrier_failures(bctl->data.target)) {
3867 "metadata profile 0x%llx has lower redundancy than data profile 0x%llx",
3868 bctl->meta.target, bctl->data.target);
3871 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3872 fs_info->num_tolerated_disk_barrier_failures = min(
3873 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info),
3874 btrfs_get_num_tolerated_disk_barrier_failures(
3878 ret = insert_balance_item(fs_info->tree_root, bctl);
3879 if (ret && ret != -EEXIST)
3882 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3883 BUG_ON(ret == -EEXIST);
3884 set_balance_control(bctl);
3886 BUG_ON(ret != -EEXIST);
3887 spin_lock(&fs_info->balance_lock);
3888 update_balance_args(bctl);
3889 spin_unlock(&fs_info->balance_lock);
3892 atomic_inc(&fs_info->balance_running);
3893 mutex_unlock(&fs_info->balance_mutex);
3895 ret = __btrfs_balance(fs_info);
3897 mutex_lock(&fs_info->balance_mutex);
3898 atomic_dec(&fs_info->balance_running);
3900 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3901 fs_info->num_tolerated_disk_barrier_failures =
3902 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3906 memset(bargs, 0, sizeof(*bargs));
3907 update_ioctl_balance_args(fs_info, 0, bargs);
3910 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3911 balance_need_close(fs_info)) {
3912 __cancel_balance(fs_info);
3915 wake_up(&fs_info->balance_wait_q);
3919 if (bctl->flags & BTRFS_BALANCE_RESUME)
3920 __cancel_balance(fs_info);
3923 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3928 static int balance_kthread(void *data)
3930 struct btrfs_fs_info *fs_info = data;
3933 mutex_lock(&fs_info->volume_mutex);
3934 mutex_lock(&fs_info->balance_mutex);
3936 if (fs_info->balance_ctl) {
3937 btrfs_info(fs_info, "continuing balance");
3938 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3941 mutex_unlock(&fs_info->balance_mutex);
3942 mutex_unlock(&fs_info->volume_mutex);
3947 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3949 struct task_struct *tsk;
3951 spin_lock(&fs_info->balance_lock);
3952 if (!fs_info->balance_ctl) {
3953 spin_unlock(&fs_info->balance_lock);
3956 spin_unlock(&fs_info->balance_lock);
3958 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
3959 btrfs_info(fs_info, "force skipping balance");
3963 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3964 return PTR_ERR_OR_ZERO(tsk);
3967 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3969 struct btrfs_balance_control *bctl;
3970 struct btrfs_balance_item *item;
3971 struct btrfs_disk_balance_args disk_bargs;
3972 struct btrfs_path *path;
3973 struct extent_buffer *leaf;
3974 struct btrfs_key key;
3977 path = btrfs_alloc_path();
3981 key.objectid = BTRFS_BALANCE_OBJECTID;
3982 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3985 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3988 if (ret > 0) { /* ret = -ENOENT; */
3993 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3999 leaf = path->nodes[0];
4000 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
4002 bctl->fs_info = fs_info;
4003 bctl->flags = btrfs_balance_flags(leaf, item);
4004 bctl->flags |= BTRFS_BALANCE_RESUME;
4006 btrfs_balance_data(leaf, item, &disk_bargs);
4007 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4008 btrfs_balance_meta(leaf, item, &disk_bargs);
4009 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4010 btrfs_balance_sys(leaf, item, &disk_bargs);
4011 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4013 WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
4015 mutex_lock(&fs_info->volume_mutex);
4016 mutex_lock(&fs_info->balance_mutex);
4018 set_balance_control(bctl);
4020 mutex_unlock(&fs_info->balance_mutex);
4021 mutex_unlock(&fs_info->volume_mutex);
4023 btrfs_free_path(path);
4027 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4031 mutex_lock(&fs_info->balance_mutex);
4032 if (!fs_info->balance_ctl) {
4033 mutex_unlock(&fs_info->balance_mutex);
4037 if (atomic_read(&fs_info->balance_running)) {
4038 atomic_inc(&fs_info->balance_pause_req);
4039 mutex_unlock(&fs_info->balance_mutex);
4041 wait_event(fs_info->balance_wait_q,
4042 atomic_read(&fs_info->balance_running) == 0);
4044 mutex_lock(&fs_info->balance_mutex);
4045 /* we are good with balance_ctl ripped off from under us */
4046 BUG_ON(atomic_read(&fs_info->balance_running));
4047 atomic_dec(&fs_info->balance_pause_req);
4052 mutex_unlock(&fs_info->balance_mutex);
4056 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4058 if (fs_info->sb->s_flags & MS_RDONLY)
4061 mutex_lock(&fs_info->balance_mutex);
4062 if (!fs_info->balance_ctl) {
4063 mutex_unlock(&fs_info->balance_mutex);
4067 atomic_inc(&fs_info->balance_cancel_req);
4069 * if we are running just wait and return, balance item is
4070 * deleted in btrfs_balance in this case
4072 if (atomic_read(&fs_info->balance_running)) {
4073 mutex_unlock(&fs_info->balance_mutex);
4074 wait_event(fs_info->balance_wait_q,
4075 atomic_read(&fs_info->balance_running) == 0);
4076 mutex_lock(&fs_info->balance_mutex);
4078 /* __cancel_balance needs volume_mutex */
4079 mutex_unlock(&fs_info->balance_mutex);
4080 mutex_lock(&fs_info->volume_mutex);
4081 mutex_lock(&fs_info->balance_mutex);
4083 if (fs_info->balance_ctl)
4084 __cancel_balance(fs_info);
4086 mutex_unlock(&fs_info->volume_mutex);
4089 BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
4090 atomic_dec(&fs_info->balance_cancel_req);
4091 mutex_unlock(&fs_info->balance_mutex);
4095 static int btrfs_uuid_scan_kthread(void *data)
4097 struct btrfs_fs_info *fs_info = data;
4098 struct btrfs_root *root = fs_info->tree_root;
4099 struct btrfs_key key;
4100 struct btrfs_key max_key;
4101 struct btrfs_path *path = NULL;
4103 struct extent_buffer *eb;
4105 struct btrfs_root_item root_item;
4107 struct btrfs_trans_handle *trans = NULL;
4109 path = btrfs_alloc_path();
4116 key.type = BTRFS_ROOT_ITEM_KEY;
4119 max_key.objectid = (u64)-1;
4120 max_key.type = BTRFS_ROOT_ITEM_KEY;
4121 max_key.offset = (u64)-1;
4124 ret = btrfs_search_forward(root, &key, path, 0);
4131 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4132 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4133 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4134 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4137 eb = path->nodes[0];
4138 slot = path->slots[0];
4139 item_size = btrfs_item_size_nr(eb, slot);
4140 if (item_size < sizeof(root_item))
4143 read_extent_buffer(eb, &root_item,
4144 btrfs_item_ptr_offset(eb, slot),
4145 (int)sizeof(root_item));
4146 if (btrfs_root_refs(&root_item) == 0)
4149 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4150 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4154 btrfs_release_path(path);
4156 * 1 - subvol uuid item
4157 * 1 - received_subvol uuid item
4159 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4160 if (IS_ERR(trans)) {
4161 ret = PTR_ERR(trans);
4169 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4170 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
4172 BTRFS_UUID_KEY_SUBVOL,
4175 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4181 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4182 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
4183 root_item.received_uuid,
4184 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4187 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4195 ret = btrfs_end_transaction(trans, fs_info->uuid_root);
4201 btrfs_release_path(path);
4202 if (key.offset < (u64)-1) {
4204 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4206 key.type = BTRFS_ROOT_ITEM_KEY;
4207 } else if (key.objectid < (u64)-1) {
4209 key.type = BTRFS_ROOT_ITEM_KEY;
4218 btrfs_free_path(path);
4219 if (trans && !IS_ERR(trans))
4220 btrfs_end_transaction(trans, fs_info->uuid_root);
4222 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4224 fs_info->update_uuid_tree_gen = 1;
4225 up(&fs_info->uuid_tree_rescan_sem);
4230 * Callback for btrfs_uuid_tree_iterate().
4232 * 0 check succeeded, the entry is not outdated.
4233 * < 0 if an error occurred.
4234 * > 0 if the check failed, which means the caller shall remove the entry.
4236 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
4237 u8 *uuid, u8 type, u64 subid)
4239 struct btrfs_key key;
4241 struct btrfs_root *subvol_root;
4243 if (type != BTRFS_UUID_KEY_SUBVOL &&
4244 type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
4247 key.objectid = subid;
4248 key.type = BTRFS_ROOT_ITEM_KEY;
4249 key.offset = (u64)-1;
4250 subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
4251 if (IS_ERR(subvol_root)) {
4252 ret = PTR_ERR(subvol_root);
4259 case BTRFS_UUID_KEY_SUBVOL:
4260 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
4263 case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
4264 if (memcmp(uuid, subvol_root->root_item.received_uuid,
4274 static int btrfs_uuid_rescan_kthread(void *data)
4276 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4280 * 1st step is to iterate through the existing UUID tree and
4281 * to delete all entries that contain outdated data.
4282 * 2nd step is to add all missing entries to the UUID tree.
4284 ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4286 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4287 up(&fs_info->uuid_tree_rescan_sem);
4290 return btrfs_uuid_scan_kthread(data);
4293 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4295 struct btrfs_trans_handle *trans;
4296 struct btrfs_root *tree_root = fs_info->tree_root;
4297 struct btrfs_root *uuid_root;
4298 struct task_struct *task;
4305 trans = btrfs_start_transaction(tree_root, 2);
4307 return PTR_ERR(trans);
4309 uuid_root = btrfs_create_tree(trans, fs_info,
4310 BTRFS_UUID_TREE_OBJECTID);
4311 if (IS_ERR(uuid_root)) {
4312 ret = PTR_ERR(uuid_root);
4313 btrfs_abort_transaction(trans, ret);
4314 btrfs_end_transaction(trans, tree_root);
4318 fs_info->uuid_root = uuid_root;
4320 ret = btrfs_commit_transaction(trans, tree_root);
4324 down(&fs_info->uuid_tree_rescan_sem);
4325 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4327 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4328 btrfs_warn(fs_info, "failed to start uuid_scan task");
4329 up(&fs_info->uuid_tree_rescan_sem);
4330 return PTR_ERR(task);
4336 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4338 struct task_struct *task;
4340 down(&fs_info->uuid_tree_rescan_sem);
4341 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4343 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4344 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4345 up(&fs_info->uuid_tree_rescan_sem);
4346 return PTR_ERR(task);
4353 * shrinking a device means finding all of the device extents past
4354 * the new size, and then following the back refs to the chunks.
4355 * The chunk relocation code actually frees the device extent
4357 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4359 struct btrfs_trans_handle *trans;
4360 struct btrfs_root *root = device->dev_root;
4361 struct btrfs_dev_extent *dev_extent = NULL;
4362 struct btrfs_path *path;
4368 bool retried = false;
4369 bool checked_pending_chunks = false;
4370 struct extent_buffer *l;
4371 struct btrfs_key key;
4372 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4373 u64 old_total = btrfs_super_total_bytes(super_copy);
4374 u64 old_size = btrfs_device_get_total_bytes(device);
4375 u64 diff = old_size - new_size;
4377 if (device->is_tgtdev_for_dev_replace)
4380 path = btrfs_alloc_path();
4384 path->reada = READA_FORWARD;
4388 btrfs_device_set_total_bytes(device, new_size);
4389 if (device->writeable) {
4390 device->fs_devices->total_rw_bytes -= diff;
4391 spin_lock(&root->fs_info->free_chunk_lock);
4392 root->fs_info->free_chunk_space -= diff;
4393 spin_unlock(&root->fs_info->free_chunk_lock);
4395 unlock_chunks(root);
4398 key.objectid = device->devid;
4399 key.offset = (u64)-1;
4400 key.type = BTRFS_DEV_EXTENT_KEY;
4403 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
4404 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4406 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4410 ret = btrfs_previous_item(root, path, 0, key.type);
4412 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4417 btrfs_release_path(path);
4422 slot = path->slots[0];
4423 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4425 if (key.objectid != device->devid) {
4426 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4427 btrfs_release_path(path);
4431 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4432 length = btrfs_dev_extent_length(l, dev_extent);
4434 if (key.offset + length <= new_size) {
4435 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4436 btrfs_release_path(path);
4440 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4441 btrfs_release_path(path);
4443 ret = btrfs_relocate_chunk(root, chunk_offset);
4444 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4445 if (ret && ret != -ENOSPC)
4449 } while (key.offset-- > 0);
4451 if (failed && !retried) {
4455 } else if (failed && retried) {
4460 /* Shrinking succeeded, else we would be at "done". */
4461 trans = btrfs_start_transaction(root, 0);
4462 if (IS_ERR(trans)) {
4463 ret = PTR_ERR(trans);
4470 * We checked in the above loop all device extents that were already in
4471 * the device tree. However before we have updated the device's
4472 * total_bytes to the new size, we might have had chunk allocations that
4473 * have not complete yet (new block groups attached to transaction
4474 * handles), and therefore their device extents were not yet in the
4475 * device tree and we missed them in the loop above. So if we have any
4476 * pending chunk using a device extent that overlaps the device range
4477 * that we can not use anymore, commit the current transaction and
4478 * repeat the search on the device tree - this way we guarantee we will
4479 * not have chunks using device extents that end beyond 'new_size'.
4481 if (!checked_pending_chunks) {
4482 u64 start = new_size;
4483 u64 len = old_size - new_size;
4485 if (contains_pending_extent(trans->transaction, device,
4487 unlock_chunks(root);
4488 checked_pending_chunks = true;
4491 ret = btrfs_commit_transaction(trans, root);
4498 btrfs_device_set_disk_total_bytes(device, new_size);
4499 if (list_empty(&device->resized_list))
4500 list_add_tail(&device->resized_list,
4501 &root->fs_info->fs_devices->resized_devices);
4503 WARN_ON(diff > old_total);
4504 btrfs_set_super_total_bytes(super_copy, old_total - diff);
4505 unlock_chunks(root);
4507 /* Now btrfs_update_device() will change the on-disk size. */
4508 ret = btrfs_update_device(trans, device);
4509 btrfs_end_transaction(trans, root);
4511 btrfs_free_path(path);
4514 btrfs_device_set_total_bytes(device, old_size);
4515 if (device->writeable)
4516 device->fs_devices->total_rw_bytes += diff;
4517 spin_lock(&root->fs_info->free_chunk_lock);
4518 root->fs_info->free_chunk_space += diff;
4519 spin_unlock(&root->fs_info->free_chunk_lock);
4520 unlock_chunks(root);
4525 static int btrfs_add_system_chunk(struct btrfs_root *root,
4526 struct btrfs_key *key,
4527 struct btrfs_chunk *chunk, int item_size)
4529 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4530 struct btrfs_disk_key disk_key;
4535 array_size = btrfs_super_sys_array_size(super_copy);
4536 if (array_size + item_size + sizeof(disk_key)
4537 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4538 unlock_chunks(root);
4542 ptr = super_copy->sys_chunk_array + array_size;
4543 btrfs_cpu_key_to_disk(&disk_key, key);
4544 memcpy(ptr, &disk_key, sizeof(disk_key));
4545 ptr += sizeof(disk_key);
4546 memcpy(ptr, chunk, item_size);
4547 item_size += sizeof(disk_key);
4548 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4549 unlock_chunks(root);
4555 * sort the devices in descending order by max_avail, total_avail
4557 static int btrfs_cmp_device_info(const void *a, const void *b)
4559 const struct btrfs_device_info *di_a = a;
4560 const struct btrfs_device_info *di_b = b;
4562 if (di_a->max_avail > di_b->max_avail)
4564 if (di_a->max_avail < di_b->max_avail)
4566 if (di_a->total_avail > di_b->total_avail)
4568 if (di_a->total_avail < di_b->total_avail)
4573 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
4575 /* TODO allow them to set a preferred stripe size */
4579 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4581 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4584 btrfs_set_fs_incompat(info, RAID56);
4587 #define BTRFS_MAX_DEVS(r) ((BTRFS_MAX_ITEM_SIZE(r) \
4588 - sizeof(struct btrfs_chunk)) \
4589 / sizeof(struct btrfs_stripe) + 1)
4591 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
4592 - 2 * sizeof(struct btrfs_disk_key) \
4593 - 2 * sizeof(struct btrfs_chunk)) \
4594 / sizeof(struct btrfs_stripe) + 1)
4596 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4597 struct btrfs_root *extent_root, u64 start,
4600 struct btrfs_fs_info *info = extent_root->fs_info;
4601 struct btrfs_fs_devices *fs_devices = info->fs_devices;
4602 struct list_head *cur;
4603 struct map_lookup *map = NULL;
4604 struct extent_map_tree *em_tree;
4605 struct extent_map *em;
4606 struct btrfs_device_info *devices_info = NULL;
4608 int num_stripes; /* total number of stripes to allocate */
4609 int data_stripes; /* number of stripes that count for
4611 int sub_stripes; /* sub_stripes info for map */
4612 int dev_stripes; /* stripes per dev */
4613 int devs_max; /* max devs to use */
4614 int devs_min; /* min devs needed */
4615 int devs_increment; /* ndevs has to be a multiple of this */
4616 int ncopies; /* how many copies to data has */
4618 u64 max_stripe_size;
4622 u64 raid_stripe_len = BTRFS_STRIPE_LEN;
4628 BUG_ON(!alloc_profile_is_valid(type, 0));
4630 if (list_empty(&fs_devices->alloc_list))
4633 index = __get_raid_index(type);
4635 sub_stripes = btrfs_raid_array[index].sub_stripes;
4636 dev_stripes = btrfs_raid_array[index].dev_stripes;
4637 devs_max = btrfs_raid_array[index].devs_max;
4638 devs_min = btrfs_raid_array[index].devs_min;
4639 devs_increment = btrfs_raid_array[index].devs_increment;
4640 ncopies = btrfs_raid_array[index].ncopies;
4642 if (type & BTRFS_BLOCK_GROUP_DATA) {
4643 max_stripe_size = SZ_1G;
4644 max_chunk_size = 10 * max_stripe_size;
4646 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4647 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4648 /* for larger filesystems, use larger metadata chunks */
4649 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
4650 max_stripe_size = SZ_1G;
4652 max_stripe_size = SZ_256M;
4653 max_chunk_size = max_stripe_size;
4655 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4656 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4657 max_stripe_size = SZ_32M;
4658 max_chunk_size = 2 * max_stripe_size;
4660 devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4662 btrfs_err(info, "invalid chunk type 0x%llx requested",
4667 /* we don't want a chunk larger than 10% of writeable space */
4668 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4671 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4676 cur = fs_devices->alloc_list.next;
4679 * in the first pass through the devices list, we gather information
4680 * about the available holes on each device.
4683 while (cur != &fs_devices->alloc_list) {
4684 struct btrfs_device *device;
4688 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4692 if (!device->writeable) {
4694 "BTRFS: read-only device in alloc_list\n");
4698 if (!device->in_fs_metadata ||
4699 device->is_tgtdev_for_dev_replace)
4702 if (device->total_bytes > device->bytes_used)
4703 total_avail = device->total_bytes - device->bytes_used;
4707 /* If there is no space on this device, skip it. */
4708 if (total_avail == 0)
4711 ret = find_free_dev_extent(trans, device,
4712 max_stripe_size * dev_stripes,
4713 &dev_offset, &max_avail);
4714 if (ret && ret != -ENOSPC)
4718 max_avail = max_stripe_size * dev_stripes;
4720 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4723 if (ndevs == fs_devices->rw_devices) {
4724 WARN(1, "%s: found more than %llu devices\n",
4725 __func__, fs_devices->rw_devices);
4728 devices_info[ndevs].dev_offset = dev_offset;
4729 devices_info[ndevs].max_avail = max_avail;
4730 devices_info[ndevs].total_avail = total_avail;
4731 devices_info[ndevs].dev = device;
4736 * now sort the devices by hole size / available space
4738 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4739 btrfs_cmp_device_info, NULL);
4741 /* round down to number of usable stripes */
4742 ndevs -= ndevs % devs_increment;
4744 if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4749 if (devs_max && ndevs > devs_max)
4752 * the primary goal is to maximize the number of stripes, so use as many
4753 * devices as possible, even if the stripes are not maximum sized.
4755 stripe_size = devices_info[ndevs-1].max_avail;
4756 num_stripes = ndevs * dev_stripes;
4759 * this will have to be fixed for RAID1 and RAID10 over
4762 data_stripes = num_stripes / ncopies;
4764 if (type & BTRFS_BLOCK_GROUP_RAID5) {
4765 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
4766 extent_root->stripesize);
4767 data_stripes = num_stripes - 1;
4769 if (type & BTRFS_BLOCK_GROUP_RAID6) {
4770 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
4771 extent_root->stripesize);
4772 data_stripes = num_stripes - 2;
4776 * Use the number of data stripes to figure out how big this chunk
4777 * is really going to be in terms of logical address space,
4778 * and compare that answer with the max chunk size
4780 if (stripe_size * data_stripes > max_chunk_size) {
4781 u64 mask = (1ULL << 24) - 1;
4783 stripe_size = div_u64(max_chunk_size, data_stripes);
4785 /* bump the answer up to a 16MB boundary */
4786 stripe_size = (stripe_size + mask) & ~mask;
4788 /* but don't go higher than the limits we found
4789 * while searching for free extents
4791 if (stripe_size > devices_info[ndevs-1].max_avail)
4792 stripe_size = devices_info[ndevs-1].max_avail;
4795 stripe_size = div_u64(stripe_size, dev_stripes);
4797 /* align to BTRFS_STRIPE_LEN */
4798 stripe_size = div_u64(stripe_size, raid_stripe_len);
4799 stripe_size *= raid_stripe_len;
4801 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4806 map->num_stripes = num_stripes;
4808 for (i = 0; i < ndevs; ++i) {
4809 for (j = 0; j < dev_stripes; ++j) {
4810 int s = i * dev_stripes + j;
4811 map->stripes[s].dev = devices_info[i].dev;
4812 map->stripes[s].physical = devices_info[i].dev_offset +
4816 map->sector_size = extent_root->sectorsize;
4817 map->stripe_len = raid_stripe_len;
4818 map->io_align = raid_stripe_len;
4819 map->io_width = raid_stripe_len;
4821 map->sub_stripes = sub_stripes;
4823 num_bytes = stripe_size * data_stripes;
4825 trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
4827 em = alloc_extent_map();
4833 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4834 em->map_lookup = map;
4836 em->len = num_bytes;
4837 em->block_start = 0;
4838 em->block_len = em->len;
4839 em->orig_block_len = stripe_size;
4841 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4842 write_lock(&em_tree->lock);
4843 ret = add_extent_mapping(em_tree, em, 0);
4845 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4846 atomic_inc(&em->refs);
4848 write_unlock(&em_tree->lock);
4850 free_extent_map(em);
4854 ret = btrfs_make_block_group(trans, extent_root, 0, type,
4855 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4858 goto error_del_extent;
4860 for (i = 0; i < map->num_stripes; i++) {
4861 num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4862 btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4865 spin_lock(&extent_root->fs_info->free_chunk_lock);
4866 extent_root->fs_info->free_chunk_space -= (stripe_size *
4868 spin_unlock(&extent_root->fs_info->free_chunk_lock);
4870 free_extent_map(em);
4871 check_raid56_incompat_flag(extent_root->fs_info, type);
4873 kfree(devices_info);
4877 write_lock(&em_tree->lock);
4878 remove_extent_mapping(em_tree, em);
4879 write_unlock(&em_tree->lock);
4881 /* One for our allocation */
4882 free_extent_map(em);
4883 /* One for the tree reference */
4884 free_extent_map(em);
4885 /* One for the pending_chunks list reference */
4886 free_extent_map(em);
4888 kfree(devices_info);
4892 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4893 struct btrfs_root *extent_root,
4894 u64 chunk_offset, u64 chunk_size)
4896 struct btrfs_key key;
4897 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4898 struct btrfs_device *device;
4899 struct btrfs_chunk *chunk;
4900 struct btrfs_stripe *stripe;
4901 struct extent_map_tree *em_tree;
4902 struct extent_map *em;
4903 struct map_lookup *map;
4910 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4911 read_lock(&em_tree->lock);
4912 em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size);
4913 read_unlock(&em_tree->lock);
4916 btrfs_crit(extent_root->fs_info, "unable to find logical "
4917 "%Lu len %Lu", chunk_offset, chunk_size);
4921 if (em->start != chunk_offset || em->len != chunk_size) {
4922 btrfs_crit(extent_root->fs_info, "found a bad mapping, wanted"
4923 " %Lu-%Lu, found %Lu-%Lu", chunk_offset,
4924 chunk_size, em->start, em->len);
4925 free_extent_map(em);
4929 map = em->map_lookup;
4930 item_size = btrfs_chunk_item_size(map->num_stripes);
4931 stripe_size = em->orig_block_len;
4933 chunk = kzalloc(item_size, GFP_NOFS);
4940 * Take the device list mutex to prevent races with the final phase of
4941 * a device replace operation that replaces the device object associated
4942 * with the map's stripes, because the device object's id can change
4943 * at any time during that final phase of the device replace operation
4944 * (dev-replace.c:btrfs_dev_replace_finishing()).
4946 mutex_lock(&chunk_root->fs_info->fs_devices->device_list_mutex);
4947 for (i = 0; i < map->num_stripes; i++) {
4948 device = map->stripes[i].dev;
4949 dev_offset = map->stripes[i].physical;
4951 ret = btrfs_update_device(trans, device);
4954 ret = btrfs_alloc_dev_extent(trans, device,
4955 chunk_root->root_key.objectid,
4956 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4957 chunk_offset, dev_offset,
4963 mutex_unlock(&chunk_root->fs_info->fs_devices->device_list_mutex);
4967 stripe = &chunk->stripe;
4968 for (i = 0; i < map->num_stripes; i++) {
4969 device = map->stripes[i].dev;
4970 dev_offset = map->stripes[i].physical;
4972 btrfs_set_stack_stripe_devid(stripe, device->devid);
4973 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4974 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4977 mutex_unlock(&chunk_root->fs_info->fs_devices->device_list_mutex);
4979 btrfs_set_stack_chunk_length(chunk, chunk_size);
4980 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4981 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4982 btrfs_set_stack_chunk_type(chunk, map->type);
4983 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4984 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4985 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4986 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4987 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4989 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4990 key.type = BTRFS_CHUNK_ITEM_KEY;
4991 key.offset = chunk_offset;
4993 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4994 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4996 * TODO: Cleanup of inserted chunk root in case of
4999 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
5005 free_extent_map(em);
5010 * Chunk allocation falls into two parts. The first part does works
5011 * that make the new allocated chunk useable, but not do any operation
5012 * that modifies the chunk tree. The second part does the works that
5013 * require modifying the chunk tree. This division is important for the
5014 * bootstrap process of adding storage to a seed btrfs.
5016 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
5017 struct btrfs_root *extent_root, u64 type)
5021 ASSERT(mutex_is_locked(&extent_root->fs_info->chunk_mutex));
5022 chunk_offset = find_next_chunk(extent_root->fs_info);
5023 return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type);
5026 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
5027 struct btrfs_root *root,
5028 struct btrfs_device *device)
5031 u64 sys_chunk_offset;
5033 struct btrfs_fs_info *fs_info = root->fs_info;
5034 struct btrfs_root *extent_root = fs_info->extent_root;
5037 chunk_offset = find_next_chunk(fs_info);
5038 alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
5039 ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset,
5044 sys_chunk_offset = find_next_chunk(root->fs_info);
5045 alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
5046 ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset,
5051 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
5055 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5056 BTRFS_BLOCK_GROUP_RAID10 |
5057 BTRFS_BLOCK_GROUP_RAID5 |
5058 BTRFS_BLOCK_GROUP_DUP)) {
5060 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
5069 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
5071 struct extent_map *em;
5072 struct map_lookup *map;
5073 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5078 read_lock(&map_tree->map_tree.lock);
5079 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
5080 read_unlock(&map_tree->map_tree.lock);
5084 map = em->map_lookup;
5085 for (i = 0; i < map->num_stripes; i++) {
5086 if (map->stripes[i].dev->missing) {
5091 if (!map->stripes[i].dev->writeable) {
5098 * If the number of missing devices is larger than max errors,
5099 * we can not write the data into that chunk successfully, so
5102 if (miss_ndevs > btrfs_chunk_max_errors(map))
5105 free_extent_map(em);
5109 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
5111 extent_map_tree_init(&tree->map_tree);
5114 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
5116 struct extent_map *em;
5119 write_lock(&tree->map_tree.lock);
5120 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
5122 remove_extent_mapping(&tree->map_tree, em);
5123 write_unlock(&tree->map_tree.lock);
5127 free_extent_map(em);
5128 /* once for the tree */
5129 free_extent_map(em);
5133 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5135 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5136 struct extent_map *em;
5137 struct map_lookup *map;
5138 struct extent_map_tree *em_tree = &map_tree->map_tree;
5141 read_lock(&em_tree->lock);
5142 em = lookup_extent_mapping(em_tree, logical, len);
5143 read_unlock(&em_tree->lock);
5146 * We could return errors for these cases, but that could get ugly and
5147 * we'd probably do the same thing which is just not do anything else
5148 * and exit, so return 1 so the callers don't try to use other copies.
5151 btrfs_crit(fs_info, "No mapping for %Lu-%Lu", logical,
5156 if (em->start > logical || em->start + em->len < logical) {
5157 btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got "
5158 "%Lu-%Lu", logical, logical+len, em->start,
5159 em->start + em->len);
5160 free_extent_map(em);
5164 map = em->map_lookup;
5165 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
5166 ret = map->num_stripes;
5167 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5168 ret = map->sub_stripes;
5169 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5171 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5175 free_extent_map(em);
5177 btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
5178 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
5180 btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
5185 unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
5186 struct btrfs_mapping_tree *map_tree,
5189 struct extent_map *em;
5190 struct map_lookup *map;
5191 struct extent_map_tree *em_tree = &map_tree->map_tree;
5192 unsigned long len = root->sectorsize;
5194 read_lock(&em_tree->lock);
5195 em = lookup_extent_mapping(em_tree, logical, len);
5196 read_unlock(&em_tree->lock);
5199 BUG_ON(em->start > logical || em->start + em->len < logical);
5200 map = em->map_lookup;
5201 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5202 len = map->stripe_len * nr_data_stripes(map);
5203 free_extent_map(em);
5207 int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
5208 u64 logical, u64 len, int mirror_num)
5210 struct extent_map *em;
5211 struct map_lookup *map;
5212 struct extent_map_tree *em_tree = &map_tree->map_tree;
5215 read_lock(&em_tree->lock);
5216 em = lookup_extent_mapping(em_tree, logical, len);
5217 read_unlock(&em_tree->lock);
5220 BUG_ON(em->start > logical || em->start + em->len < logical);
5221 map = em->map_lookup;
5222 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5224 free_extent_map(em);
5228 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5229 struct map_lookup *map, int first, int num,
5230 int optimal, int dev_replace_is_ongoing)
5234 struct btrfs_device *srcdev;
5236 if (dev_replace_is_ongoing &&
5237 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5238 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5239 srcdev = fs_info->dev_replace.srcdev;
5244 * try to avoid the drive that is the source drive for a
5245 * dev-replace procedure, only choose it if no other non-missing
5246 * mirror is available
5248 for (tolerance = 0; tolerance < 2; tolerance++) {
5249 if (map->stripes[optimal].dev->bdev &&
5250 (tolerance || map->stripes[optimal].dev != srcdev))
5252 for (i = first; i < first + num; i++) {
5253 if (map->stripes[i].dev->bdev &&
5254 (tolerance || map->stripes[i].dev != srcdev))
5259 /* we couldn't find one that doesn't fail. Just return something
5260 * and the io error handling code will clean up eventually
5265 static inline int parity_smaller(u64 a, u64 b)
5270 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5271 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5273 struct btrfs_bio_stripe s;
5280 for (i = 0; i < num_stripes - 1; i++) {
5281 if (parity_smaller(bbio->raid_map[i],
5282 bbio->raid_map[i+1])) {
5283 s = bbio->stripes[i];
5284 l = bbio->raid_map[i];
5285 bbio->stripes[i] = bbio->stripes[i+1];
5286 bbio->raid_map[i] = bbio->raid_map[i+1];
5287 bbio->stripes[i+1] = s;
5288 bbio->raid_map[i+1] = l;
5296 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5298 struct btrfs_bio *bbio = kzalloc(
5299 /* the size of the btrfs_bio */
5300 sizeof(struct btrfs_bio) +
5301 /* plus the variable array for the stripes */
5302 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5303 /* plus the variable array for the tgt dev */
5304 sizeof(int) * (real_stripes) +
5306 * plus the raid_map, which includes both the tgt dev
5309 sizeof(u64) * (total_stripes),
5310 GFP_NOFS|__GFP_NOFAIL);
5312 atomic_set(&bbio->error, 0);
5313 atomic_set(&bbio->refs, 1);
5318 void btrfs_get_bbio(struct btrfs_bio *bbio)
5320 WARN_ON(!atomic_read(&bbio->refs));
5321 atomic_inc(&bbio->refs);
5324 void btrfs_put_bbio(struct btrfs_bio *bbio)
5328 if (atomic_dec_and_test(&bbio->refs))
5332 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int op,
5333 u64 logical, u64 *length,
5334 struct btrfs_bio **bbio_ret,
5335 int mirror_num, int need_raid_map)
5337 struct extent_map *em;
5338 struct map_lookup *map;
5339 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5340 struct extent_map_tree *em_tree = &map_tree->map_tree;
5343 u64 stripe_end_offset;
5353 int tgtdev_indexes = 0;
5354 struct btrfs_bio *bbio = NULL;
5355 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
5356 int dev_replace_is_ongoing = 0;
5357 int num_alloc_stripes;
5358 int patch_the_first_stripe_for_dev_replace = 0;
5359 u64 physical_to_patch_in_first_stripe = 0;
5360 u64 raid56_full_stripe_start = (u64)-1;
5362 read_lock(&em_tree->lock);
5363 em = lookup_extent_mapping(em_tree, logical, *length);
5364 read_unlock(&em_tree->lock);
5367 btrfs_crit(fs_info, "unable to find logical %llu len %llu",
5372 if (em->start > logical || em->start + em->len < logical) {
5373 btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, "
5374 "found %Lu-%Lu", logical, em->start,
5375 em->start + em->len);
5376 free_extent_map(em);
5380 map = em->map_lookup;
5381 offset = logical - em->start;
5383 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(stripe_nr, stripe_len);
5391 stripe_offset = stripe_nr * stripe_len;
5392 if (offset < stripe_offset) {
5393 btrfs_crit(fs_info, "stripe math has gone wrong, "
5394 "stripe_offset=%llu, offset=%llu, start=%llu, "
5395 "logical=%llu, stripe_len=%llu",
5396 stripe_offset, offset, em->start, logical,
5398 free_extent_map(em);
5402 /* stripe_offset is the offset of this block in its stripe*/
5403 stripe_offset = offset - stripe_offset;
5405 /* if we're here for raid56, we need to know the stripe aligned start */
5406 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5407 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5408 raid56_full_stripe_start = offset;
5410 /* allow a write of a full stripe, but make sure we don't
5411 * allow straddling of stripes
5413 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5415 raid56_full_stripe_start *= full_stripe_len;
5418 if (op == REQ_OP_DISCARD) {
5419 /* we don't discard raid56 yet */
5420 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5424 *length = min_t(u64, em->len - offset, *length);
5425 } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5427 /* For writes to RAID[56], allow a full stripeset across all disks.
5428 For other RAID types and for RAID[56] reads, just allow a single
5429 stripe (on a single disk). */
5430 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
5431 (op == REQ_OP_WRITE)) {
5432 max_len = stripe_len * nr_data_stripes(map) -
5433 (offset - raid56_full_stripe_start);
5435 /* we limit the length of each bio to what fits in a stripe */
5436 max_len = stripe_len - stripe_offset;
5438 *length = min_t(u64, em->len - offset, max_len);
5440 *length = em->len - offset;
5443 /* This is for when we're called from btrfs_merge_bio_hook() and all
5444 it cares about is the length */
5448 btrfs_dev_replace_lock(dev_replace, 0);
5449 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5450 if (!dev_replace_is_ongoing)
5451 btrfs_dev_replace_unlock(dev_replace, 0);
5453 btrfs_dev_replace_set_lock_blocking(dev_replace);
5455 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
5456 op != REQ_OP_WRITE && op != REQ_OP_DISCARD &&
5457 op != REQ_GET_READ_MIRRORS && dev_replace->tgtdev != NULL) {
5459 * in dev-replace case, for repair case (that's the only
5460 * case where the mirror is selected explicitly when
5461 * calling btrfs_map_block), blocks left of the left cursor
5462 * can also be read from the target drive.
5463 * For REQ_GET_READ_MIRRORS, the target drive is added as
5464 * the last one to the array of stripes. For READ, it also
5465 * needs to be supported using the same mirror number.
5466 * If the requested block is not left of the left cursor,
5467 * EIO is returned. This can happen because btrfs_num_copies()
5468 * returns one more in the dev-replace case.
5470 u64 tmp_length = *length;
5471 struct btrfs_bio *tmp_bbio = NULL;
5472 int tmp_num_stripes;
5473 u64 srcdev_devid = dev_replace->srcdev->devid;
5474 int index_srcdev = 0;
5476 u64 physical_of_found = 0;
5478 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
5479 logical, &tmp_length, &tmp_bbio, 0, 0);
5481 WARN_ON(tmp_bbio != NULL);
5485 tmp_num_stripes = tmp_bbio->num_stripes;
5486 if (mirror_num > tmp_num_stripes) {
5488 * REQ_GET_READ_MIRRORS does not contain this
5489 * mirror, that means that the requested area
5490 * is not left of the left cursor
5493 btrfs_put_bbio(tmp_bbio);
5498 * process the rest of the function using the mirror_num
5499 * of the source drive. Therefore look it up first.
5500 * At the end, patch the device pointer to the one of the
5503 for (i = 0; i < tmp_num_stripes; i++) {
5504 if (tmp_bbio->stripes[i].dev->devid != srcdev_devid)
5508 * In case of DUP, in order to keep it simple, only add
5509 * the mirror with the lowest physical address
5512 physical_of_found <= tmp_bbio->stripes[i].physical)
5517 physical_of_found = tmp_bbio->stripes[i].physical;
5520 btrfs_put_bbio(tmp_bbio);
5528 mirror_num = index_srcdev + 1;
5529 patch_the_first_stripe_for_dev_replace = 1;
5530 physical_to_patch_in_first_stripe = physical_of_found;
5531 } else if (mirror_num > map->num_stripes) {
5537 stripe_nr_orig = stripe_nr;
5538 stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
5539 stripe_nr_end = div_u64(stripe_nr_end, map->stripe_len);
5540 stripe_end_offset = stripe_nr_end * map->stripe_len -
5543 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5544 if (op == REQ_OP_DISCARD)
5545 num_stripes = min_t(u64, map->num_stripes,
5546 stripe_nr_end - stripe_nr_orig);
5547 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5549 if (op != REQ_OP_WRITE && op != REQ_OP_DISCARD &&
5550 op != REQ_GET_READ_MIRRORS)
5552 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
5553 if (op == REQ_OP_WRITE || op == REQ_OP_DISCARD ||
5554 op == REQ_GET_READ_MIRRORS)
5555 num_stripes = map->num_stripes;
5556 else if (mirror_num)
5557 stripe_index = mirror_num - 1;
5559 stripe_index = find_live_mirror(fs_info, map, 0,
5561 current->pid % map->num_stripes,
5562 dev_replace_is_ongoing);
5563 mirror_num = stripe_index + 1;
5566 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
5567 if (op == REQ_OP_WRITE || op == REQ_OP_DISCARD ||
5568 op == REQ_GET_READ_MIRRORS) {
5569 num_stripes = map->num_stripes;
5570 } else if (mirror_num) {
5571 stripe_index = mirror_num - 1;
5576 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5577 u32 factor = map->num_stripes / map->sub_stripes;
5579 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5580 stripe_index *= map->sub_stripes;
5582 if (op == REQ_OP_WRITE || op == REQ_GET_READ_MIRRORS)
5583 num_stripes = map->sub_stripes;
5584 else if (op == REQ_OP_DISCARD)
5585 num_stripes = min_t(u64, map->sub_stripes *
5586 (stripe_nr_end - stripe_nr_orig),
5588 else if (mirror_num)
5589 stripe_index += mirror_num - 1;
5591 int old_stripe_index = stripe_index;
5592 stripe_index = find_live_mirror(fs_info, map,
5594 map->sub_stripes, stripe_index +
5595 current->pid % map->sub_stripes,
5596 dev_replace_is_ongoing);
5597 mirror_num = stripe_index - old_stripe_index + 1;
5600 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5601 if (need_raid_map &&
5602 (op == REQ_OP_WRITE || op == REQ_GET_READ_MIRRORS ||
5604 /* push stripe_nr back to the start of the full stripe */
5605 stripe_nr = div_u64(raid56_full_stripe_start,
5606 stripe_len * nr_data_stripes(map));
5608 /* RAID[56] write or recovery. Return all stripes */
5609 num_stripes = map->num_stripes;
5610 max_errors = nr_parity_stripes(map);
5612 *length = map->stripe_len;
5617 * Mirror #0 or #1 means the original data block.
5618 * Mirror #2 is RAID5 parity block.
5619 * Mirror #3 is RAID6 Q block.
5621 stripe_nr = div_u64_rem(stripe_nr,
5622 nr_data_stripes(map), &stripe_index);
5624 stripe_index = nr_data_stripes(map) +
5627 /* We distribute the parity blocks across stripes */
5628 div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
5630 if ((op != REQ_OP_WRITE && op != REQ_OP_DISCARD &&
5631 op != REQ_GET_READ_MIRRORS) && mirror_num <= 1)
5636 * after this, stripe_nr is the number of stripes on this
5637 * device we have to walk to find the data, and stripe_index is
5638 * the number of our device in the stripe array
5640 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5642 mirror_num = stripe_index + 1;
5644 if (stripe_index >= map->num_stripes) {
5645 btrfs_crit(fs_info, "stripe index math went horribly wrong, "
5646 "got stripe_index=%u, num_stripes=%u",
5647 stripe_index, map->num_stripes);
5652 num_alloc_stripes = num_stripes;
5653 if (dev_replace_is_ongoing) {
5654 if (op == REQ_OP_WRITE || op == REQ_OP_DISCARD)
5655 num_alloc_stripes <<= 1;
5656 if (op == REQ_GET_READ_MIRRORS)
5657 num_alloc_stripes++;
5658 tgtdev_indexes = num_stripes;
5661 bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
5666 if (dev_replace_is_ongoing)
5667 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
5669 /* build raid_map */
5670 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK &&
5672 ((op == REQ_OP_WRITE || op == REQ_GET_READ_MIRRORS) ||
5677 bbio->raid_map = (u64 *)((void *)bbio->stripes +
5678 sizeof(struct btrfs_bio_stripe) *
5680 sizeof(int) * tgtdev_indexes);
5682 /* Work out the disk rotation on this stripe-set */
5683 div_u64_rem(stripe_nr, num_stripes, &rot);
5685 /* Fill in the logical address of each stripe */
5686 tmp = stripe_nr * nr_data_stripes(map);
5687 for (i = 0; i < nr_data_stripes(map); i++)
5688 bbio->raid_map[(i+rot) % num_stripes] =
5689 em->start + (tmp + i) * map->stripe_len;
5691 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5692 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5693 bbio->raid_map[(i+rot+1) % num_stripes] =
5697 if (op == REQ_OP_DISCARD) {
5699 u32 sub_stripes = 0;
5700 u64 stripes_per_dev = 0;
5701 u32 remaining_stripes = 0;
5702 u32 last_stripe = 0;
5705 (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
5706 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5709 sub_stripes = map->sub_stripes;
5711 factor = map->num_stripes / sub_stripes;
5712 stripes_per_dev = div_u64_rem(stripe_nr_end -
5715 &remaining_stripes);
5716 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5717 last_stripe *= sub_stripes;
5720 for (i = 0; i < num_stripes; i++) {
5721 bbio->stripes[i].physical =
5722 map->stripes[stripe_index].physical +
5723 stripe_offset + stripe_nr * map->stripe_len;
5724 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5726 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5727 BTRFS_BLOCK_GROUP_RAID10)) {
5728 bbio->stripes[i].length = stripes_per_dev *
5731 if (i / sub_stripes < remaining_stripes)
5732 bbio->stripes[i].length +=
5736 * Special for the first stripe and
5739 * |-------|...|-------|
5743 if (i < sub_stripes)
5744 bbio->stripes[i].length -=
5747 if (stripe_index >= last_stripe &&
5748 stripe_index <= (last_stripe +
5750 bbio->stripes[i].length -=
5753 if (i == sub_stripes - 1)
5756 bbio->stripes[i].length = *length;
5759 if (stripe_index == map->num_stripes) {
5760 /* This could only happen for RAID0/10 */
5766 for (i = 0; i < num_stripes; i++) {
5767 bbio->stripes[i].physical =
5768 map->stripes[stripe_index].physical +
5770 stripe_nr * map->stripe_len;
5771 bbio->stripes[i].dev =
5772 map->stripes[stripe_index].dev;
5777 if (op == REQ_OP_WRITE || op == REQ_GET_READ_MIRRORS)
5778 max_errors = btrfs_chunk_max_errors(map);
5781 sort_parity_stripes(bbio, num_stripes);
5784 if (dev_replace_is_ongoing &&
5785 (op == REQ_OP_WRITE || op == REQ_OP_DISCARD) &&
5786 dev_replace->tgtdev != NULL) {
5787 int index_where_to_add;
5788 u64 srcdev_devid = dev_replace->srcdev->devid;
5791 * duplicate the write operations while the dev replace
5792 * procedure is running. Since the copying of the old disk
5793 * to the new disk takes place at run time while the
5794 * filesystem is mounted writable, the regular write
5795 * operations to the old disk have to be duplicated to go
5796 * to the new disk as well.
5797 * Note that device->missing is handled by the caller, and
5798 * that the write to the old disk is already set up in the
5801 index_where_to_add = num_stripes;
5802 for (i = 0; i < num_stripes; i++) {
5803 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5804 /* write to new disk, too */
5805 struct btrfs_bio_stripe *new =
5806 bbio->stripes + index_where_to_add;
5807 struct btrfs_bio_stripe *old =
5810 new->physical = old->physical;
5811 new->length = old->length;
5812 new->dev = dev_replace->tgtdev;
5813 bbio->tgtdev_map[i] = index_where_to_add;
5814 index_where_to_add++;
5819 num_stripes = index_where_to_add;
5820 } else if (dev_replace_is_ongoing && (op == REQ_GET_READ_MIRRORS) &&
5821 dev_replace->tgtdev != NULL) {
5822 u64 srcdev_devid = dev_replace->srcdev->devid;
5823 int index_srcdev = 0;
5825 u64 physical_of_found = 0;
5828 * During the dev-replace procedure, the target drive can
5829 * also be used to read data in case it is needed to repair
5830 * a corrupt block elsewhere. This is possible if the
5831 * requested area is left of the left cursor. In this area,
5832 * the target drive is a full copy of the source drive.
5834 for (i = 0; i < num_stripes; i++) {
5835 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5837 * In case of DUP, in order to keep it
5838 * simple, only add the mirror with the
5839 * lowest physical address
5842 physical_of_found <=
5843 bbio->stripes[i].physical)
5847 physical_of_found = bbio->stripes[i].physical;
5851 struct btrfs_bio_stripe *tgtdev_stripe =
5852 bbio->stripes + num_stripes;
5854 tgtdev_stripe->physical = physical_of_found;
5855 tgtdev_stripe->length =
5856 bbio->stripes[index_srcdev].length;
5857 tgtdev_stripe->dev = dev_replace->tgtdev;
5858 bbio->tgtdev_map[index_srcdev] = num_stripes;
5866 bbio->map_type = map->type;
5867 bbio->num_stripes = num_stripes;
5868 bbio->max_errors = max_errors;
5869 bbio->mirror_num = mirror_num;
5870 bbio->num_tgtdevs = tgtdev_indexes;
5873 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5874 * mirror_num == num_stripes + 1 && dev_replace target drive is
5875 * available as a mirror
5877 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5878 WARN_ON(num_stripes > 1);
5879 bbio->stripes[0].dev = dev_replace->tgtdev;
5880 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5881 bbio->mirror_num = map->num_stripes + 1;
5884 if (dev_replace_is_ongoing) {
5885 btrfs_dev_replace_clear_lock_blocking(dev_replace);
5886 btrfs_dev_replace_unlock(dev_replace, 0);
5888 free_extent_map(em);
5892 int btrfs_map_block(struct btrfs_fs_info *fs_info, int op,
5893 u64 logical, u64 *length,
5894 struct btrfs_bio **bbio_ret, int mirror_num)
5896 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret,
5900 /* For Scrub/replace */
5901 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, int op,
5902 u64 logical, u64 *length,
5903 struct btrfs_bio **bbio_ret, int mirror_num,
5906 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret,
5907 mirror_num, need_raid_map);
5910 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
5911 u64 chunk_start, u64 physical, u64 devid,
5912 u64 **logical, int *naddrs, int *stripe_len)
5914 struct extent_map_tree *em_tree = &map_tree->map_tree;
5915 struct extent_map *em;
5916 struct map_lookup *map;
5924 read_lock(&em_tree->lock);
5925 em = lookup_extent_mapping(em_tree, chunk_start, 1);
5926 read_unlock(&em_tree->lock);
5929 printk(KERN_ERR "BTRFS: couldn't find em for chunk %Lu\n",
5934 if (em->start != chunk_start) {
5935 printk(KERN_ERR "BTRFS: bad chunk start, em=%Lu, wanted=%Lu\n",
5936 em->start, chunk_start);
5937 free_extent_map(em);
5940 map = em->map_lookup;
5943 rmap_len = map->stripe_len;
5945 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5946 length = div_u64(length, map->num_stripes / map->sub_stripes);
5947 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5948 length = div_u64(length, map->num_stripes);
5949 else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5950 length = div_u64(length, nr_data_stripes(map));
5951 rmap_len = map->stripe_len * nr_data_stripes(map);
5954 buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
5955 BUG_ON(!buf); /* -ENOMEM */
5957 for (i = 0; i < map->num_stripes; i++) {
5958 if (devid && map->stripes[i].dev->devid != devid)
5960 if (map->stripes[i].physical > physical ||
5961 map->stripes[i].physical + length <= physical)
5964 stripe_nr = physical - map->stripes[i].physical;
5965 stripe_nr = div_u64(stripe_nr, map->stripe_len);
5967 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5968 stripe_nr = stripe_nr * map->num_stripes + i;
5969 stripe_nr = div_u64(stripe_nr, map->sub_stripes);
5970 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5971 stripe_nr = stripe_nr * map->num_stripes + i;
5972 } /* else if RAID[56], multiply by nr_data_stripes().
5973 * Alternatively, just use rmap_len below instead of
5974 * map->stripe_len */
5976 bytenr = chunk_start + stripe_nr * rmap_len;
5977 WARN_ON(nr >= map->num_stripes);
5978 for (j = 0; j < nr; j++) {
5979 if (buf[j] == bytenr)
5983 WARN_ON(nr >= map->num_stripes);
5990 *stripe_len = rmap_len;
5992 free_extent_map(em);
5996 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
5998 bio->bi_private = bbio->private;
5999 bio->bi_end_io = bbio->end_io;
6002 btrfs_put_bbio(bbio);
6005 static void btrfs_end_bio(struct bio *bio)
6007 struct btrfs_bio *bbio = bio->bi_private;
6008 int is_orig_bio = 0;
6010 if (bio->bi_error) {
6011 atomic_inc(&bbio->error);
6012 if (bio->bi_error == -EIO || bio->bi_error == -EREMOTEIO) {
6013 unsigned int stripe_index =
6014 btrfs_io_bio(bio)->stripe_index;
6015 struct btrfs_device *dev;
6017 BUG_ON(stripe_index >= bbio->num_stripes);
6018 dev = bbio->stripes[stripe_index].dev;
6020 if (bio_op(bio) == REQ_OP_WRITE)
6021 btrfs_dev_stat_inc(dev,
6022 BTRFS_DEV_STAT_WRITE_ERRS);
6024 btrfs_dev_stat_inc(dev,
6025 BTRFS_DEV_STAT_READ_ERRS);
6026 if ((bio->bi_opf & WRITE_FLUSH) == WRITE_FLUSH)
6027 btrfs_dev_stat_inc(dev,
6028 BTRFS_DEV_STAT_FLUSH_ERRS);
6029 btrfs_dev_stat_print_on_error(dev);
6034 if (bio == bbio->orig_bio)
6037 btrfs_bio_counter_dec(bbio->fs_info);
6039 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6042 bio = bbio->orig_bio;
6045 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6046 /* only send an error to the higher layers if it is
6047 * beyond the tolerance of the btrfs bio
6049 if (atomic_read(&bbio->error) > bbio->max_errors) {
6050 bio->bi_error = -EIO;
6053 * this bio is actually up to date, we didn't
6054 * go over the max number of errors
6059 btrfs_end_bbio(bbio, bio);
6060 } else if (!is_orig_bio) {
6066 * see run_scheduled_bios for a description of why bios are collected for
6069 * This will add one bio to the pending list for a device and make sure
6070 * the work struct is scheduled.
6072 static noinline void btrfs_schedule_bio(struct btrfs_root *root,
6073 struct btrfs_device *device,
6076 int should_queue = 1;
6077 struct btrfs_pending_bios *pending_bios;
6079 if (device->missing || !device->bdev) {
6084 /* don't bother with additional async steps for reads, right now */
6085 if (bio_op(bio) == REQ_OP_READ) {
6087 btrfsic_submit_bio(bio);
6093 * nr_async_bios allows us to reliably return congestion to the
6094 * higher layers. Otherwise, the async bio makes it appear we have
6095 * made progress against dirty pages when we've really just put it
6096 * on a queue for later
6098 atomic_inc(&root->fs_info->nr_async_bios);
6099 WARN_ON(bio->bi_next);
6100 bio->bi_next = NULL;
6102 spin_lock(&device->io_lock);
6103 if (bio->bi_opf & REQ_SYNC)
6104 pending_bios = &device->pending_sync_bios;
6106 pending_bios = &device->pending_bios;
6108 if (pending_bios->tail)
6109 pending_bios->tail->bi_next = bio;
6111 pending_bios->tail = bio;
6112 if (!pending_bios->head)
6113 pending_bios->head = bio;
6114 if (device->running_pending)
6117 spin_unlock(&device->io_lock);
6120 btrfs_queue_work(root->fs_info->submit_workers,
6124 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
6125 struct bio *bio, u64 physical, int dev_nr,
6128 struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
6130 bio->bi_private = bbio;
6131 btrfs_io_bio(bio)->stripe_index = dev_nr;
6132 bio->bi_end_io = btrfs_end_bio;
6133 bio->bi_iter.bi_sector = physical >> 9;
6136 struct rcu_string *name;
6139 name = rcu_dereference(dev->name);
6140 pr_debug("btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu "
6141 "(%s id %llu), size=%u\n", bio_op(bio), bio->bi_opf,
6142 (u64)bio->bi_iter.bi_sector, (u_long)dev->bdev->bd_dev,
6143 name->str, dev->devid, bio->bi_iter.bi_size);
6147 bio->bi_bdev = dev->bdev;
6149 btrfs_bio_counter_inc_noblocked(root->fs_info);
6152 btrfs_schedule_bio(root, dev, bio);
6154 btrfsic_submit_bio(bio);
6157 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6159 atomic_inc(&bbio->error);
6160 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6161 /* Should be the original bio. */
6162 WARN_ON(bio != bbio->orig_bio);
6164 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6165 bio->bi_iter.bi_sector = logical >> 9;
6166 bio->bi_error = -EIO;
6167 btrfs_end_bbio(bbio, bio);
6171 int btrfs_map_bio(struct btrfs_root *root, struct bio *bio,
6172 int mirror_num, int async_submit)
6174 struct btrfs_device *dev;
6175 struct bio *first_bio = bio;
6176 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
6182 struct btrfs_bio *bbio = NULL;
6184 length = bio->bi_iter.bi_size;
6185 map_length = length;
6187 btrfs_bio_counter_inc_blocked(root->fs_info);
6188 ret = __btrfs_map_block(root->fs_info, bio_op(bio), logical,
6189 &map_length, &bbio, mirror_num, 1);
6191 btrfs_bio_counter_dec(root->fs_info);
6195 total_devs = bbio->num_stripes;
6196 bbio->orig_bio = first_bio;
6197 bbio->private = first_bio->bi_private;
6198 bbio->end_io = first_bio->bi_end_io;
6199 bbio->fs_info = root->fs_info;
6200 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6202 if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6203 ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) {
6204 /* In this case, map_length has been set to the length of
6205 a single stripe; not the whole write */
6206 if (bio_op(bio) == REQ_OP_WRITE) {
6207 ret = raid56_parity_write(root, bio, bbio, map_length);
6209 ret = raid56_parity_recover(root, bio, bbio, map_length,
6213 btrfs_bio_counter_dec(root->fs_info);
6217 if (map_length < length) {
6218 btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu",
6219 logical, length, map_length);
6223 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6224 dev = bbio->stripes[dev_nr].dev;
6225 if (!dev || !dev->bdev ||
6226 (bio_op(bio) == REQ_OP_WRITE && !dev->writeable)) {
6227 bbio_error(bbio, first_bio, logical);
6231 if (dev_nr < total_devs - 1) {
6232 bio = btrfs_bio_clone(first_bio, GFP_NOFS);
6233 BUG_ON(!bio); /* -ENOMEM */
6237 submit_stripe_bio(root, bbio, bio,
6238 bbio->stripes[dev_nr].physical, dev_nr,
6241 btrfs_bio_counter_dec(root->fs_info);
6245 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
6248 struct btrfs_device *device;
6249 struct btrfs_fs_devices *cur_devices;
6251 cur_devices = fs_info->fs_devices;
6252 while (cur_devices) {
6254 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
6255 device = __find_device(&cur_devices->devices,
6260 cur_devices = cur_devices->seed;
6265 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
6266 struct btrfs_fs_devices *fs_devices,
6267 u64 devid, u8 *dev_uuid)
6269 struct btrfs_device *device;
6271 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6275 list_add(&device->dev_list, &fs_devices->devices);
6276 device->fs_devices = fs_devices;
6277 fs_devices->num_devices++;
6279 device->missing = 1;
6280 fs_devices->missing_devices++;
6286 * btrfs_alloc_device - allocate struct btrfs_device
6287 * @fs_info: used only for generating a new devid, can be NULL if
6288 * devid is provided (i.e. @devid != NULL).
6289 * @devid: a pointer to devid for this device. If NULL a new devid
6291 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6294 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6295 * on error. Returned struct is not linked onto any lists and can be
6296 * destroyed with kfree() right away.
6298 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6302 struct btrfs_device *dev;
6305 if (WARN_ON(!devid && !fs_info))
6306 return ERR_PTR(-EINVAL);
6308 dev = __alloc_device();
6317 ret = find_next_devid(fs_info, &tmp);
6320 return ERR_PTR(ret);
6326 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6328 generate_random_uuid(dev->uuid);
6330 btrfs_init_work(&dev->work, btrfs_submit_helper,
6331 pending_bios_fn, NULL, NULL);
6336 /* Return -EIO if any error, otherwise return 0. */
6337 static int btrfs_check_chunk_valid(struct btrfs_root *root,
6338 struct extent_buffer *leaf,
6339 struct btrfs_chunk *chunk, u64 logical)
6347 length = btrfs_chunk_length(leaf, chunk);
6348 stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6349 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6350 sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6351 type = btrfs_chunk_type(leaf, chunk);
6354 btrfs_err(root->fs_info, "invalid chunk num_stripes: %u",
6358 if (!IS_ALIGNED(logical, root->sectorsize)) {
6359 btrfs_err(root->fs_info,
6360 "invalid chunk logical %llu", logical);
6363 if (btrfs_chunk_sector_size(leaf, chunk) != root->sectorsize) {
6364 btrfs_err(root->fs_info, "invalid chunk sectorsize %u",
6365 btrfs_chunk_sector_size(leaf, chunk));
6368 if (!length || !IS_ALIGNED(length, root->sectorsize)) {
6369 btrfs_err(root->fs_info,
6370 "invalid chunk length %llu", length);
6373 if (!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN) {
6374 btrfs_err(root->fs_info, "invalid chunk stripe length: %llu",
6378 if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6380 btrfs_err(root->fs_info, "unrecognized chunk type: %llu",
6381 ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
6382 BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6383 btrfs_chunk_type(leaf, chunk));
6386 if ((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes != 2) ||
6387 (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) ||
6388 (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) ||
6389 (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) ||
6390 (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) ||
6391 ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
6392 num_stripes != 1)) {
6393 btrfs_err(root->fs_info,
6394 "invalid num_stripes:sub_stripes %u:%u for profile %llu",
6395 num_stripes, sub_stripes,
6396 type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
6403 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
6404 struct extent_buffer *leaf,
6405 struct btrfs_chunk *chunk)
6407 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
6408 struct map_lookup *map;
6409 struct extent_map *em;
6414 u8 uuid[BTRFS_UUID_SIZE];
6419 logical = key->offset;
6420 length = btrfs_chunk_length(leaf, chunk);
6421 stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6422 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6424 ret = btrfs_check_chunk_valid(root, leaf, chunk, logical);
6428 read_lock(&map_tree->map_tree.lock);
6429 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
6430 read_unlock(&map_tree->map_tree.lock);
6432 /* already mapped? */
6433 if (em && em->start <= logical && em->start + em->len > logical) {
6434 free_extent_map(em);
6437 free_extent_map(em);
6440 em = alloc_extent_map();
6443 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6445 free_extent_map(em);
6449 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6450 em->map_lookup = map;
6451 em->start = logical;
6454 em->block_start = 0;
6455 em->block_len = em->len;
6457 map->num_stripes = num_stripes;
6458 map->io_width = btrfs_chunk_io_width(leaf, chunk);
6459 map->io_align = btrfs_chunk_io_align(leaf, chunk);
6460 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
6461 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6462 map->type = btrfs_chunk_type(leaf, chunk);
6463 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6464 for (i = 0; i < num_stripes; i++) {
6465 map->stripes[i].physical =
6466 btrfs_stripe_offset_nr(leaf, chunk, i);
6467 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6468 read_extent_buffer(leaf, uuid, (unsigned long)
6469 btrfs_stripe_dev_uuid_nr(chunk, i),
6471 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
6473 if (!map->stripes[i].dev &&
6474 !btrfs_test_opt(root->fs_info, DEGRADED)) {
6475 free_extent_map(em);
6478 if (!map->stripes[i].dev) {
6479 map->stripes[i].dev =
6480 add_missing_dev(root, root->fs_info->fs_devices,
6482 if (!map->stripes[i].dev) {
6483 free_extent_map(em);
6486 btrfs_warn(root->fs_info, "devid %llu uuid %pU is missing",
6489 map->stripes[i].dev->in_fs_metadata = 1;
6492 write_lock(&map_tree->map_tree.lock);
6493 ret = add_extent_mapping(&map_tree->map_tree, em, 0);
6494 write_unlock(&map_tree->map_tree.lock);
6495 BUG_ON(ret); /* Tree corruption */
6496 free_extent_map(em);
6501 static void fill_device_from_item(struct extent_buffer *leaf,
6502 struct btrfs_dev_item *dev_item,
6503 struct btrfs_device *device)
6507 device->devid = btrfs_device_id(leaf, dev_item);
6508 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6509 device->total_bytes = device->disk_total_bytes;
6510 device->commit_total_bytes = device->disk_total_bytes;
6511 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6512 device->commit_bytes_used = device->bytes_used;
6513 device->type = btrfs_device_type(leaf, dev_item);
6514 device->io_align = btrfs_device_io_align(leaf, dev_item);
6515 device->io_width = btrfs_device_io_width(leaf, dev_item);
6516 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6517 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6518 device->is_tgtdev_for_dev_replace = 0;
6520 ptr = btrfs_device_uuid(dev_item);
6521 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6524 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_root *root,
6527 struct btrfs_fs_devices *fs_devices;
6530 BUG_ON(!mutex_is_locked(&uuid_mutex));
6532 fs_devices = root->fs_info->fs_devices->seed;
6533 while (fs_devices) {
6534 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE))
6537 fs_devices = fs_devices->seed;
6540 fs_devices = find_fsid(fsid);
6542 if (!btrfs_test_opt(root->fs_info, DEGRADED))
6543 return ERR_PTR(-ENOENT);
6545 fs_devices = alloc_fs_devices(fsid);
6546 if (IS_ERR(fs_devices))
6549 fs_devices->seeding = 1;
6550 fs_devices->opened = 1;
6554 fs_devices = clone_fs_devices(fs_devices);
6555 if (IS_ERR(fs_devices))
6558 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
6559 root->fs_info->bdev_holder);
6561 free_fs_devices(fs_devices);
6562 fs_devices = ERR_PTR(ret);
6566 if (!fs_devices->seeding) {
6567 __btrfs_close_devices(fs_devices);
6568 free_fs_devices(fs_devices);
6569 fs_devices = ERR_PTR(-EINVAL);
6573 fs_devices->seed = root->fs_info->fs_devices->seed;
6574 root->fs_info->fs_devices->seed = fs_devices;
6579 static int read_one_dev(struct btrfs_root *root,
6580 struct extent_buffer *leaf,
6581 struct btrfs_dev_item *dev_item)
6583 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6584 struct btrfs_device *device;
6587 u8 fs_uuid[BTRFS_UUID_SIZE];
6588 u8 dev_uuid[BTRFS_UUID_SIZE];
6590 devid = btrfs_device_id(leaf, dev_item);
6591 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6593 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6596 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
6597 fs_devices = open_seed_devices(root, fs_uuid);
6598 if (IS_ERR(fs_devices))
6599 return PTR_ERR(fs_devices);
6602 device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
6604 if (!btrfs_test_opt(root->fs_info, DEGRADED))
6607 device = add_missing_dev(root, fs_devices, devid, dev_uuid);
6610 btrfs_warn(root->fs_info, "devid %llu uuid %pU missing",
6613 if (!device->bdev && !btrfs_test_opt(root->fs_info, DEGRADED))
6616 if(!device->bdev && !device->missing) {
6618 * this happens when a device that was properly setup
6619 * in the device info lists suddenly goes bad.
6620 * device->bdev is NULL, and so we have to set
6621 * device->missing to one here
6623 device->fs_devices->missing_devices++;
6624 device->missing = 1;
6627 /* Move the device to its own fs_devices */
6628 if (device->fs_devices != fs_devices) {
6629 ASSERT(device->missing);
6631 list_move(&device->dev_list, &fs_devices->devices);
6632 device->fs_devices->num_devices--;
6633 fs_devices->num_devices++;
6635 device->fs_devices->missing_devices--;
6636 fs_devices->missing_devices++;
6638 device->fs_devices = fs_devices;
6642 if (device->fs_devices != root->fs_info->fs_devices) {
6643 BUG_ON(device->writeable);
6644 if (device->generation !=
6645 btrfs_device_generation(leaf, dev_item))
6649 fill_device_from_item(leaf, dev_item, device);
6650 device->in_fs_metadata = 1;
6651 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
6652 device->fs_devices->total_rw_bytes += device->total_bytes;
6653 spin_lock(&root->fs_info->free_chunk_lock);
6654 root->fs_info->free_chunk_space += device->total_bytes -
6656 spin_unlock(&root->fs_info->free_chunk_lock);
6662 int btrfs_read_sys_array(struct btrfs_root *root)
6664 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
6665 struct extent_buffer *sb;
6666 struct btrfs_disk_key *disk_key;
6667 struct btrfs_chunk *chunk;
6669 unsigned long sb_array_offset;
6676 struct btrfs_key key;
6678 ASSERT(BTRFS_SUPER_INFO_SIZE <= root->nodesize);
6680 * This will create extent buffer of nodesize, superblock size is
6681 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6682 * overallocate but we can keep it as-is, only the first page is used.
6684 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET);
6687 set_extent_buffer_uptodate(sb);
6688 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6690 * The sb extent buffer is artificial and just used to read the system array.
6691 * set_extent_buffer_uptodate() call does not properly mark all it's
6692 * pages up-to-date when the page is larger: extent does not cover the
6693 * whole page and consequently check_page_uptodate does not find all
6694 * the page's extents up-to-date (the hole beyond sb),
6695 * write_extent_buffer then triggers a WARN_ON.
6697 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6698 * but sb spans only this function. Add an explicit SetPageUptodate call
6699 * to silence the warning eg. on PowerPC 64.
6701 if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
6702 SetPageUptodate(sb->pages[0]);
6704 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6705 array_size = btrfs_super_sys_array_size(super_copy);
6707 array_ptr = super_copy->sys_chunk_array;
6708 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6711 while (cur_offset < array_size) {
6712 disk_key = (struct btrfs_disk_key *)array_ptr;
6713 len = sizeof(*disk_key);
6714 if (cur_offset + len > array_size)
6715 goto out_short_read;
6717 btrfs_disk_key_to_cpu(&key, disk_key);
6720 sb_array_offset += len;
6723 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6724 chunk = (struct btrfs_chunk *)sb_array_offset;
6726 * At least one btrfs_chunk with one stripe must be
6727 * present, exact stripe count check comes afterwards
6729 len = btrfs_chunk_item_size(1);
6730 if (cur_offset + len > array_size)
6731 goto out_short_read;
6733 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6736 "BTRFS: invalid number of stripes %u in sys_array at offset %u\n",
6737 num_stripes, cur_offset);
6742 type = btrfs_chunk_type(sb, chunk);
6743 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
6744 btrfs_err(root->fs_info,
6745 "invalid chunk type %llu in sys_array at offset %u",
6751 len = btrfs_chunk_item_size(num_stripes);
6752 if (cur_offset + len > array_size)
6753 goto out_short_read;
6755 ret = read_one_chunk(root, &key, sb, chunk);
6760 "BTRFS: unexpected item type %u in sys_array at offset %u\n",
6761 (u32)key.type, cur_offset);
6766 sb_array_offset += len;
6769 clear_extent_buffer_uptodate(sb);
6770 free_extent_buffer_stale(sb);
6774 printk(KERN_ERR "BTRFS: sys_array too short to read %u bytes at offset %u\n",
6776 clear_extent_buffer_uptodate(sb);
6777 free_extent_buffer_stale(sb);
6781 int btrfs_read_chunk_tree(struct btrfs_root *root)
6783 struct btrfs_path *path;
6784 struct extent_buffer *leaf;
6785 struct btrfs_key key;
6786 struct btrfs_key found_key;
6791 root = root->fs_info->chunk_root;
6793 path = btrfs_alloc_path();
6797 mutex_lock(&uuid_mutex);
6801 * Read all device items, and then all the chunk items. All
6802 * device items are found before any chunk item (their object id
6803 * is smaller than the lowest possible object id for a chunk
6804 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6806 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6809 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6813 leaf = path->nodes[0];
6814 slot = path->slots[0];
6815 if (slot >= btrfs_header_nritems(leaf)) {
6816 ret = btrfs_next_leaf(root, path);
6823 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6824 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6825 struct btrfs_dev_item *dev_item;
6826 dev_item = btrfs_item_ptr(leaf, slot,
6827 struct btrfs_dev_item);
6828 ret = read_one_dev(root, leaf, dev_item);
6832 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6833 struct btrfs_chunk *chunk;
6834 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6835 ret = read_one_chunk(root, &found_key, leaf, chunk);
6843 * After loading chunk tree, we've got all device information,
6844 * do another round of validation checks.
6846 if (total_dev != root->fs_info->fs_devices->total_devices) {
6847 btrfs_err(root->fs_info,
6848 "super_num_devices %llu mismatch with num_devices %llu found here",
6849 btrfs_super_num_devices(root->fs_info->super_copy),
6854 if (btrfs_super_total_bytes(root->fs_info->super_copy) <
6855 root->fs_info->fs_devices->total_rw_bytes) {
6856 btrfs_err(root->fs_info,
6857 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
6858 btrfs_super_total_bytes(root->fs_info->super_copy),
6859 root->fs_info->fs_devices->total_rw_bytes);
6865 unlock_chunks(root);
6866 mutex_unlock(&uuid_mutex);
6868 btrfs_free_path(path);
6872 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6874 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6875 struct btrfs_device *device;
6877 while (fs_devices) {
6878 mutex_lock(&fs_devices->device_list_mutex);
6879 list_for_each_entry(device, &fs_devices->devices, dev_list)
6880 device->dev_root = fs_info->dev_root;
6881 mutex_unlock(&fs_devices->device_list_mutex);
6883 fs_devices = fs_devices->seed;
6887 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6891 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6892 btrfs_dev_stat_reset(dev, i);
6895 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6897 struct btrfs_key key;
6898 struct btrfs_key found_key;
6899 struct btrfs_root *dev_root = fs_info->dev_root;
6900 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6901 struct extent_buffer *eb;
6904 struct btrfs_device *device;
6905 struct btrfs_path *path = NULL;
6908 path = btrfs_alloc_path();
6914 mutex_lock(&fs_devices->device_list_mutex);
6915 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6917 struct btrfs_dev_stats_item *ptr;
6919 key.objectid = BTRFS_DEV_STATS_OBJECTID;
6920 key.type = BTRFS_PERSISTENT_ITEM_KEY;
6921 key.offset = device->devid;
6922 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6924 __btrfs_reset_dev_stats(device);
6925 device->dev_stats_valid = 1;
6926 btrfs_release_path(path);
6929 slot = path->slots[0];
6930 eb = path->nodes[0];
6931 btrfs_item_key_to_cpu(eb, &found_key, slot);
6932 item_size = btrfs_item_size_nr(eb, slot);
6934 ptr = btrfs_item_ptr(eb, slot,
6935 struct btrfs_dev_stats_item);
6937 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6938 if (item_size >= (1 + i) * sizeof(__le64))
6939 btrfs_dev_stat_set(device, i,
6940 btrfs_dev_stats_value(eb, ptr, i));
6942 btrfs_dev_stat_reset(device, i);
6945 device->dev_stats_valid = 1;
6946 btrfs_dev_stat_print_on_load(device);
6947 btrfs_release_path(path);
6949 mutex_unlock(&fs_devices->device_list_mutex);
6952 btrfs_free_path(path);
6953 return ret < 0 ? ret : 0;
6956 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
6957 struct btrfs_root *dev_root,
6958 struct btrfs_device *device)
6960 struct btrfs_path *path;
6961 struct btrfs_key key;
6962 struct extent_buffer *eb;
6963 struct btrfs_dev_stats_item *ptr;
6967 key.objectid = BTRFS_DEV_STATS_OBJECTID;
6968 key.type = BTRFS_PERSISTENT_ITEM_KEY;
6969 key.offset = device->devid;
6971 path = btrfs_alloc_path();
6973 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
6975 btrfs_warn_in_rcu(dev_root->fs_info,
6976 "error %d while searching for dev_stats item for device %s",
6977 ret, rcu_str_deref(device->name));
6982 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
6983 /* need to delete old one and insert a new one */
6984 ret = btrfs_del_item(trans, dev_root, path);
6986 btrfs_warn_in_rcu(dev_root->fs_info,
6987 "delete too small dev_stats item for device %s failed %d",
6988 rcu_str_deref(device->name), ret);
6995 /* need to insert a new item */
6996 btrfs_release_path(path);
6997 ret = btrfs_insert_empty_item(trans, dev_root, path,
6998 &key, sizeof(*ptr));
7000 btrfs_warn_in_rcu(dev_root->fs_info,
7001 "insert dev_stats item for device %s failed %d",
7002 rcu_str_deref(device->name), ret);
7007 eb = path->nodes[0];
7008 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7009 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7010 btrfs_set_dev_stats_value(eb, ptr, i,
7011 btrfs_dev_stat_read(device, i));
7012 btrfs_mark_buffer_dirty(eb);
7015 btrfs_free_path(path);
7020 * called from commit_transaction. Writes all changed device stats to disk.
7022 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
7023 struct btrfs_fs_info *fs_info)
7025 struct btrfs_root *dev_root = fs_info->dev_root;
7026 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7027 struct btrfs_device *device;
7031 mutex_lock(&fs_devices->device_list_mutex);
7032 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7033 if (!device->dev_stats_valid || !btrfs_dev_stats_dirty(device))
7036 stats_cnt = atomic_read(&device->dev_stats_ccnt);
7037 ret = update_dev_stat_item(trans, dev_root, device);
7039 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7041 mutex_unlock(&fs_devices->device_list_mutex);
7046 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7048 btrfs_dev_stat_inc(dev, index);
7049 btrfs_dev_stat_print_on_error(dev);
7052 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
7054 if (!dev->dev_stats_valid)
7056 btrfs_err_rl_in_rcu(dev->dev_root->fs_info,
7057 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7058 rcu_str_deref(dev->name),
7059 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7060 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7061 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7062 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7063 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7066 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7070 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7071 if (btrfs_dev_stat_read(dev, i) != 0)
7073 if (i == BTRFS_DEV_STAT_VALUES_MAX)
7074 return; /* all values == 0, suppress message */
7076 btrfs_info_in_rcu(dev->dev_root->fs_info,
7077 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7078 rcu_str_deref(dev->name),
7079 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7080 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7081 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7082 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7083 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7086 int btrfs_get_dev_stats(struct btrfs_root *root,
7087 struct btrfs_ioctl_get_dev_stats *stats)
7089 struct btrfs_device *dev;
7090 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7093 mutex_lock(&fs_devices->device_list_mutex);
7094 dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
7095 mutex_unlock(&fs_devices->device_list_mutex);
7098 btrfs_warn(root->fs_info, "get dev_stats failed, device not found");
7100 } else if (!dev->dev_stats_valid) {
7101 btrfs_warn(root->fs_info, "get dev_stats failed, not yet valid");
7103 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7104 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7105 if (stats->nr_items > i)
7107 btrfs_dev_stat_read_and_reset(dev, i);
7109 btrfs_dev_stat_reset(dev, i);
7112 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7113 if (stats->nr_items > i)
7114 stats->values[i] = btrfs_dev_stat_read(dev, i);
7116 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7117 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7121 void btrfs_scratch_superblocks(struct block_device *bdev, char *device_path)
7123 struct buffer_head *bh;
7124 struct btrfs_super_block *disk_super;
7130 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX;
7133 if (btrfs_read_dev_one_super(bdev, copy_num, &bh))
7136 disk_super = (struct btrfs_super_block *)bh->b_data;
7138 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
7139 set_buffer_dirty(bh);
7140 sync_dirty_buffer(bh);
7144 /* Notify udev that device has changed */
7145 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
7147 /* Update ctime/mtime for device path for libblkid */
7148 update_dev_time(device_path);
7152 * Update the size of all devices, which is used for writing out the
7155 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
7157 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7158 struct btrfs_device *curr, *next;
7160 if (list_empty(&fs_devices->resized_devices))
7163 mutex_lock(&fs_devices->device_list_mutex);
7164 lock_chunks(fs_info->dev_root);
7165 list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
7167 list_del_init(&curr->resized_list);
7168 curr->commit_total_bytes = curr->disk_total_bytes;
7170 unlock_chunks(fs_info->dev_root);
7171 mutex_unlock(&fs_devices->device_list_mutex);
7174 /* Must be invoked during the transaction commit */
7175 void btrfs_update_commit_device_bytes_used(struct btrfs_root *root,
7176 struct btrfs_transaction *transaction)
7178 struct extent_map *em;
7179 struct map_lookup *map;
7180 struct btrfs_device *dev;
7183 if (list_empty(&transaction->pending_chunks))
7186 /* In order to kick the device replace finish process */
7188 list_for_each_entry(em, &transaction->pending_chunks, list) {
7189 map = em->map_lookup;
7191 for (i = 0; i < map->num_stripes; i++) {
7192 dev = map->stripes[i].dev;
7193 dev->commit_bytes_used = dev->bytes_used;
7196 unlock_chunks(root);
7199 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
7201 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7202 while (fs_devices) {
7203 fs_devices->fs_info = fs_info;
7204 fs_devices = fs_devices->seed;
7208 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
7210 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7211 while (fs_devices) {
7212 fs_devices->fs_info = NULL;
7213 fs_devices = fs_devices->seed;