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_prepare_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 new_device = btrfs_alloc_device(NULL, &device->devid,
882 BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
884 /* Safe because we are under uuid_mutex */
886 name = rcu_string_strdup(device->name->str, GFP_NOFS);
887 BUG_ON(!name); /* -ENOMEM */
888 rcu_assign_pointer(new_device->name, name);
891 list_replace_rcu(&device->dev_list, &new_device->dev_list);
892 new_device->fs_devices = device->fs_devices;
895 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
897 struct btrfs_device *device, *tmp;
898 struct list_head pending_put;
900 INIT_LIST_HEAD(&pending_put);
902 if (--fs_devices->opened > 0)
905 mutex_lock(&fs_devices->device_list_mutex);
906 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
907 btrfs_prepare_close_one_device(device);
908 list_add(&device->dev_list, &pending_put);
910 mutex_unlock(&fs_devices->device_list_mutex);
913 * btrfs_show_devname() is using the device_list_mutex,
914 * sometimes call to blkdev_put() leads vfs calling
915 * into this func. So do put outside of device_list_mutex,
918 while (!list_empty(&pending_put)) {
919 device = list_first_entry(&pending_put,
920 struct btrfs_device, dev_list);
921 list_del(&device->dev_list);
922 btrfs_close_bdev(device);
923 call_rcu(&device->rcu, free_device);
926 WARN_ON(fs_devices->open_devices);
927 WARN_ON(fs_devices->rw_devices);
928 fs_devices->opened = 0;
929 fs_devices->seeding = 0;
934 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
936 struct btrfs_fs_devices *seed_devices = NULL;
939 mutex_lock(&uuid_mutex);
940 ret = __btrfs_close_devices(fs_devices);
941 if (!fs_devices->opened) {
942 seed_devices = fs_devices->seed;
943 fs_devices->seed = NULL;
945 mutex_unlock(&uuid_mutex);
947 while (seed_devices) {
948 fs_devices = seed_devices;
949 seed_devices = fs_devices->seed;
950 __btrfs_close_devices(fs_devices);
951 free_fs_devices(fs_devices);
954 * Wait for rcu kworkers under __btrfs_close_devices
955 * to finish all blkdev_puts so device is really
956 * free when umount is done.
962 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
963 fmode_t flags, void *holder)
965 struct request_queue *q;
966 struct block_device *bdev;
967 struct list_head *head = &fs_devices->devices;
968 struct btrfs_device *device;
969 struct btrfs_device *latest_dev = NULL;
970 struct buffer_head *bh;
971 struct btrfs_super_block *disk_super;
978 list_for_each_entry(device, head, dev_list) {
984 /* Just open everything we can; ignore failures here */
985 if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
989 disk_super = (struct btrfs_super_block *)bh->b_data;
990 devid = btrfs_stack_device_id(&disk_super->dev_item);
991 if (devid != device->devid)
994 if (memcmp(device->uuid, disk_super->dev_item.uuid,
998 device->generation = btrfs_super_generation(disk_super);
1000 device->generation > latest_dev->generation)
1001 latest_dev = device;
1003 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
1004 device->writeable = 0;
1006 device->writeable = !bdev_read_only(bdev);
1010 q = bdev_get_queue(bdev);
1011 if (blk_queue_discard(q))
1012 device->can_discard = 1;
1014 device->bdev = bdev;
1015 device->in_fs_metadata = 0;
1016 device->mode = flags;
1018 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1019 fs_devices->rotating = 1;
1021 fs_devices->open_devices++;
1022 if (device->writeable &&
1023 device->devid != BTRFS_DEV_REPLACE_DEVID) {
1024 fs_devices->rw_devices++;
1025 list_add(&device->dev_alloc_list,
1026 &fs_devices->alloc_list);
1033 blkdev_put(bdev, flags);
1036 if (fs_devices->open_devices == 0) {
1040 fs_devices->seeding = seeding;
1041 fs_devices->opened = 1;
1042 fs_devices->latest_bdev = latest_dev->bdev;
1043 fs_devices->total_rw_bytes = 0;
1048 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1049 fmode_t flags, void *holder)
1053 mutex_lock(&uuid_mutex);
1054 if (fs_devices->opened) {
1055 fs_devices->opened++;
1058 ret = __btrfs_open_devices(fs_devices, flags, holder);
1060 mutex_unlock(&uuid_mutex);
1064 void btrfs_release_disk_super(struct page *page)
1070 int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr,
1071 struct page **page, struct btrfs_super_block **disk_super)
1076 /* make sure our super fits in the device */
1077 if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1080 /* make sure our super fits in the page */
1081 if (sizeof(**disk_super) > PAGE_SIZE)
1084 /* make sure our super doesn't straddle pages on disk */
1085 index = bytenr >> PAGE_SHIFT;
1086 if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index)
1089 /* pull in the page with our super */
1090 *page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
1093 if (IS_ERR_OR_NULL(*page))
1098 /* align our pointer to the offset of the super block */
1099 *disk_super = p + (bytenr & ~PAGE_MASK);
1101 if (btrfs_super_bytenr(*disk_super) != bytenr ||
1102 btrfs_super_magic(*disk_super) != BTRFS_MAGIC) {
1103 btrfs_release_disk_super(*page);
1107 if ((*disk_super)->label[0] &&
1108 (*disk_super)->label[BTRFS_LABEL_SIZE - 1])
1109 (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0';
1115 * Look for a btrfs signature on a device. This may be called out of the mount path
1116 * and we are not allowed to call set_blocksize during the scan. The superblock
1117 * is read via pagecache
1119 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
1120 struct btrfs_fs_devices **fs_devices_ret)
1122 struct btrfs_super_block *disk_super;
1123 struct block_device *bdev;
1132 * we would like to check all the supers, but that would make
1133 * a btrfs mount succeed after a mkfs from a different FS.
1134 * So, we need to add a special mount option to scan for
1135 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1137 bytenr = btrfs_sb_offset(0);
1138 flags |= FMODE_EXCL;
1139 mutex_lock(&uuid_mutex);
1141 bdev = blkdev_get_by_path(path, flags, holder);
1143 ret = PTR_ERR(bdev);
1147 if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super))
1148 goto error_bdev_put;
1150 devid = btrfs_stack_device_id(&disk_super->dev_item);
1151 transid = btrfs_super_generation(disk_super);
1152 total_devices = btrfs_super_num_devices(disk_super);
1154 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
1156 if (disk_super->label[0]) {
1157 pr_info("BTRFS: device label %s ", disk_super->label);
1159 pr_info("BTRFS: device fsid %pU ", disk_super->fsid);
1162 pr_cont("devid %llu transid %llu %s\n", devid, transid, path);
1165 if (!ret && fs_devices_ret)
1166 (*fs_devices_ret)->total_devices = total_devices;
1168 btrfs_release_disk_super(page);
1171 blkdev_put(bdev, flags);
1173 mutex_unlock(&uuid_mutex);
1177 /* helper to account the used device space in the range */
1178 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
1179 u64 end, u64 *length)
1181 struct btrfs_key key;
1182 struct btrfs_root *root = device->dev_root;
1183 struct btrfs_dev_extent *dev_extent;
1184 struct btrfs_path *path;
1188 struct extent_buffer *l;
1192 if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
1195 path = btrfs_alloc_path();
1198 path->reada = READA_FORWARD;
1200 key.objectid = device->devid;
1202 key.type = BTRFS_DEV_EXTENT_KEY;
1204 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1208 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1215 slot = path->slots[0];
1216 if (slot >= btrfs_header_nritems(l)) {
1217 ret = btrfs_next_leaf(root, path);
1225 btrfs_item_key_to_cpu(l, &key, slot);
1227 if (key.objectid < device->devid)
1230 if (key.objectid > device->devid)
1233 if (key.type != BTRFS_DEV_EXTENT_KEY)
1236 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1237 extent_end = key.offset + btrfs_dev_extent_length(l,
1239 if (key.offset <= start && extent_end > end) {
1240 *length = end - start + 1;
1242 } else if (key.offset <= start && extent_end > start)
1243 *length += extent_end - start;
1244 else if (key.offset > start && extent_end <= end)
1245 *length += extent_end - key.offset;
1246 else if (key.offset > start && key.offset <= end) {
1247 *length += end - key.offset + 1;
1249 } else if (key.offset > end)
1257 btrfs_free_path(path);
1261 static int contains_pending_extent(struct btrfs_transaction *transaction,
1262 struct btrfs_device *device,
1263 u64 *start, u64 len)
1265 struct btrfs_fs_info *fs_info = device->dev_root->fs_info;
1266 struct extent_map *em;
1267 struct list_head *search_list = &fs_info->pinned_chunks;
1269 u64 physical_start = *start;
1272 search_list = &transaction->pending_chunks;
1274 list_for_each_entry(em, search_list, list) {
1275 struct map_lookup *map;
1278 map = em->map_lookup;
1279 for (i = 0; i < map->num_stripes; i++) {
1282 if (map->stripes[i].dev != device)
1284 if (map->stripes[i].physical >= physical_start + len ||
1285 map->stripes[i].physical + em->orig_block_len <=
1289 * Make sure that while processing the pinned list we do
1290 * not override our *start with a lower value, because
1291 * we can have pinned chunks that fall within this
1292 * device hole and that have lower physical addresses
1293 * than the pending chunks we processed before. If we
1294 * do not take this special care we can end up getting
1295 * 2 pending chunks that start at the same physical
1296 * device offsets because the end offset of a pinned
1297 * chunk can be equal to the start offset of some
1300 end = map->stripes[i].physical + em->orig_block_len;
1307 if (search_list != &fs_info->pinned_chunks) {
1308 search_list = &fs_info->pinned_chunks;
1317 * find_free_dev_extent_start - find free space in the specified device
1318 * @device: the device which we search the free space in
1319 * @num_bytes: the size of the free space that we need
1320 * @search_start: the position from which to begin the search
1321 * @start: store the start of the free space.
1322 * @len: the size of the free space. that we find, or the size
1323 * of the max free space if we don't find suitable free space
1325 * this uses a pretty simple search, the expectation is that it is
1326 * called very infrequently and that a given device has a small number
1329 * @start is used to store the start of the free space if we find. But if we
1330 * don't find suitable free space, it will be used to store the start position
1331 * of the max free space.
1333 * @len is used to store the size of the free space that we find.
1334 * But if we don't find suitable free space, it is used to store the size of
1335 * the max free space.
1337 int find_free_dev_extent_start(struct btrfs_transaction *transaction,
1338 struct btrfs_device *device, u64 num_bytes,
1339 u64 search_start, u64 *start, u64 *len)
1341 struct btrfs_key key;
1342 struct btrfs_root *root = device->dev_root;
1343 struct btrfs_dev_extent *dev_extent;
1344 struct btrfs_path *path;
1349 u64 search_end = device->total_bytes;
1352 struct extent_buffer *l;
1353 u64 min_search_start;
1356 * We don't want to overwrite the superblock on the drive nor any area
1357 * used by the boot loader (grub for example), so we make sure to start
1358 * at an offset of at least 1MB.
1360 min_search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
1361 search_start = max(search_start, min_search_start);
1363 path = btrfs_alloc_path();
1367 max_hole_start = search_start;
1371 if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1376 path->reada = READA_FORWARD;
1377 path->search_commit_root = 1;
1378 path->skip_locking = 1;
1380 key.objectid = device->devid;
1381 key.offset = search_start;
1382 key.type = BTRFS_DEV_EXTENT_KEY;
1384 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1388 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1395 slot = path->slots[0];
1396 if (slot >= btrfs_header_nritems(l)) {
1397 ret = btrfs_next_leaf(root, path);
1405 btrfs_item_key_to_cpu(l, &key, slot);
1407 if (key.objectid < device->devid)
1410 if (key.objectid > device->devid)
1413 if (key.type != BTRFS_DEV_EXTENT_KEY)
1416 if (key.offset > search_start) {
1417 hole_size = key.offset - search_start;
1420 * Have to check before we set max_hole_start, otherwise
1421 * we could end up sending back this offset anyway.
1423 if (contains_pending_extent(transaction, device,
1426 if (key.offset >= search_start) {
1427 hole_size = key.offset - search_start;
1434 if (hole_size > max_hole_size) {
1435 max_hole_start = search_start;
1436 max_hole_size = hole_size;
1440 * If this free space is greater than which we need,
1441 * it must be the max free space that we have found
1442 * until now, so max_hole_start must point to the start
1443 * of this free space and the length of this free space
1444 * is stored in max_hole_size. Thus, we return
1445 * max_hole_start and max_hole_size and go back to the
1448 if (hole_size >= num_bytes) {
1454 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1455 extent_end = key.offset + btrfs_dev_extent_length(l,
1457 if (extent_end > search_start)
1458 search_start = extent_end;
1465 * At this point, search_start should be the end of
1466 * allocated dev extents, and when shrinking the device,
1467 * search_end may be smaller than search_start.
1469 if (search_end > search_start) {
1470 hole_size = search_end - search_start;
1472 if (contains_pending_extent(transaction, device, &search_start,
1474 btrfs_release_path(path);
1478 if (hole_size > max_hole_size) {
1479 max_hole_start = search_start;
1480 max_hole_size = hole_size;
1485 if (max_hole_size < num_bytes)
1491 btrfs_free_path(path);
1492 *start = max_hole_start;
1494 *len = max_hole_size;
1498 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1499 struct btrfs_device *device, u64 num_bytes,
1500 u64 *start, u64 *len)
1502 /* FIXME use last free of some kind */
1503 return find_free_dev_extent_start(trans->transaction, device,
1504 num_bytes, 0, start, len);
1507 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1508 struct btrfs_device *device,
1509 u64 start, u64 *dev_extent_len)
1512 struct btrfs_path *path;
1513 struct btrfs_root *root = device->dev_root;
1514 struct btrfs_key key;
1515 struct btrfs_key found_key;
1516 struct extent_buffer *leaf = NULL;
1517 struct btrfs_dev_extent *extent = NULL;
1519 path = btrfs_alloc_path();
1523 key.objectid = device->devid;
1525 key.type = BTRFS_DEV_EXTENT_KEY;
1527 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1529 ret = btrfs_previous_item(root, path, key.objectid,
1530 BTRFS_DEV_EXTENT_KEY);
1533 leaf = path->nodes[0];
1534 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1535 extent = btrfs_item_ptr(leaf, path->slots[0],
1536 struct btrfs_dev_extent);
1537 BUG_ON(found_key.offset > start || found_key.offset +
1538 btrfs_dev_extent_length(leaf, extent) < start);
1540 btrfs_release_path(path);
1542 } else if (ret == 0) {
1543 leaf = path->nodes[0];
1544 extent = btrfs_item_ptr(leaf, path->slots[0],
1545 struct btrfs_dev_extent);
1547 btrfs_handle_fs_error(root->fs_info, ret, "Slot search failed");
1551 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1553 ret = btrfs_del_item(trans, root, path);
1555 btrfs_handle_fs_error(root->fs_info, ret,
1556 "Failed to remove dev extent item");
1558 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1561 btrfs_free_path(path);
1565 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1566 struct btrfs_device *device,
1567 u64 chunk_tree, u64 chunk_objectid,
1568 u64 chunk_offset, u64 start, u64 num_bytes)
1571 struct btrfs_path *path;
1572 struct btrfs_root *root = device->dev_root;
1573 struct btrfs_dev_extent *extent;
1574 struct extent_buffer *leaf;
1575 struct btrfs_key key;
1577 WARN_ON(!device->in_fs_metadata);
1578 WARN_ON(device->is_tgtdev_for_dev_replace);
1579 path = btrfs_alloc_path();
1583 key.objectid = device->devid;
1585 key.type = BTRFS_DEV_EXTENT_KEY;
1586 ret = btrfs_insert_empty_item(trans, root, path, &key,
1591 leaf = path->nodes[0];
1592 extent = btrfs_item_ptr(leaf, path->slots[0],
1593 struct btrfs_dev_extent);
1594 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1595 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1596 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1598 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1599 btrfs_dev_extent_chunk_tree_uuid(extent), BTRFS_UUID_SIZE);
1601 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1602 btrfs_mark_buffer_dirty(leaf);
1604 btrfs_free_path(path);
1608 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1610 struct extent_map_tree *em_tree;
1611 struct extent_map *em;
1615 em_tree = &fs_info->mapping_tree.map_tree;
1616 read_lock(&em_tree->lock);
1617 n = rb_last(&em_tree->map);
1619 em = rb_entry(n, struct extent_map, rb_node);
1620 ret = em->start + em->len;
1622 read_unlock(&em_tree->lock);
1627 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1631 struct btrfs_key key;
1632 struct btrfs_key found_key;
1633 struct btrfs_path *path;
1635 path = btrfs_alloc_path();
1639 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1640 key.type = BTRFS_DEV_ITEM_KEY;
1641 key.offset = (u64)-1;
1643 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1647 BUG_ON(ret == 0); /* Corruption */
1649 ret = btrfs_previous_item(fs_info->chunk_root, path,
1650 BTRFS_DEV_ITEMS_OBJECTID,
1651 BTRFS_DEV_ITEM_KEY);
1655 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1657 *devid_ret = found_key.offset + 1;
1661 btrfs_free_path(path);
1666 * the device information is stored in the chunk root
1667 * the btrfs_device struct should be fully filled in
1669 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1670 struct btrfs_root *root,
1671 struct btrfs_device *device)
1674 struct btrfs_path *path;
1675 struct btrfs_dev_item *dev_item;
1676 struct extent_buffer *leaf;
1677 struct btrfs_key key;
1680 root = root->fs_info->chunk_root;
1682 path = btrfs_alloc_path();
1686 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1687 key.type = BTRFS_DEV_ITEM_KEY;
1688 key.offset = device->devid;
1690 ret = btrfs_insert_empty_item(trans, root, path, &key,
1695 leaf = path->nodes[0];
1696 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1698 btrfs_set_device_id(leaf, dev_item, device->devid);
1699 btrfs_set_device_generation(leaf, dev_item, 0);
1700 btrfs_set_device_type(leaf, dev_item, device->type);
1701 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1702 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1703 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1704 btrfs_set_device_total_bytes(leaf, dev_item,
1705 btrfs_device_get_disk_total_bytes(device));
1706 btrfs_set_device_bytes_used(leaf, dev_item,
1707 btrfs_device_get_bytes_used(device));
1708 btrfs_set_device_group(leaf, dev_item, 0);
1709 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1710 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1711 btrfs_set_device_start_offset(leaf, dev_item, 0);
1713 ptr = btrfs_device_uuid(dev_item);
1714 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1715 ptr = btrfs_device_fsid(dev_item);
1716 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1717 btrfs_mark_buffer_dirty(leaf);
1721 btrfs_free_path(path);
1726 * Function to update ctime/mtime for a given device path.
1727 * Mainly used for ctime/mtime based probe like libblkid.
1729 static void update_dev_time(char *path_name)
1733 filp = filp_open(path_name, O_RDWR, 0);
1736 file_update_time(filp);
1737 filp_close(filp, NULL);
1740 static int btrfs_rm_dev_item(struct btrfs_root *root,
1741 struct btrfs_device *device)
1744 struct btrfs_path *path;
1745 struct btrfs_key key;
1746 struct btrfs_trans_handle *trans;
1748 root = root->fs_info->chunk_root;
1750 path = btrfs_alloc_path();
1754 trans = btrfs_start_transaction(root, 0);
1755 if (IS_ERR(trans)) {
1756 btrfs_free_path(path);
1757 return PTR_ERR(trans);
1759 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1760 key.type = BTRFS_DEV_ITEM_KEY;
1761 key.offset = device->devid;
1763 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1772 ret = btrfs_del_item(trans, root, path);
1776 btrfs_free_path(path);
1777 btrfs_commit_transaction(trans, root);
1782 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1783 * filesystem. It's up to the caller to adjust that number regarding eg. device
1786 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1794 seq = read_seqbegin(&fs_info->profiles_lock);
1796 all_avail = fs_info->avail_data_alloc_bits |
1797 fs_info->avail_system_alloc_bits |
1798 fs_info->avail_metadata_alloc_bits;
1799 } while (read_seqretry(&fs_info->profiles_lock, seq));
1801 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1802 if (!(all_avail & btrfs_raid_group[i]))
1805 if (num_devices < btrfs_raid_array[i].devs_min) {
1806 int ret = btrfs_raid_mindev_error[i];
1816 struct btrfs_device *btrfs_find_next_active_device(struct btrfs_fs_devices *fs_devs,
1817 struct btrfs_device *device)
1819 struct btrfs_device *next_device;
1821 list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1822 if (next_device != device &&
1823 !next_device->missing && next_device->bdev)
1831 * Helper function to check if the given device is part of s_bdev / latest_bdev
1832 * and replace it with the provided or the next active device, in the context
1833 * where this function called, there should be always be another device (or
1834 * this_dev) which is active.
1836 void btrfs_assign_next_active_device(struct btrfs_fs_info *fs_info,
1837 struct btrfs_device *device, struct btrfs_device *this_dev)
1839 struct btrfs_device *next_device;
1842 next_device = this_dev;
1844 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
1846 ASSERT(next_device);
1848 if (fs_info->sb->s_bdev &&
1849 (fs_info->sb->s_bdev == device->bdev))
1850 fs_info->sb->s_bdev = next_device->bdev;
1852 if (fs_info->fs_devices->latest_bdev == device->bdev)
1853 fs_info->fs_devices->latest_bdev = next_device->bdev;
1856 int btrfs_rm_device(struct btrfs_root *root, char *device_path, u64 devid)
1858 struct btrfs_device *device;
1859 struct btrfs_fs_devices *cur_devices;
1862 bool clear_super = false;
1864 mutex_lock(&uuid_mutex);
1866 num_devices = root->fs_info->fs_devices->num_devices;
1867 btrfs_dev_replace_lock(&root->fs_info->dev_replace, 0);
1868 if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1869 WARN_ON(num_devices < 1);
1872 btrfs_dev_replace_unlock(&root->fs_info->dev_replace, 0);
1874 ret = btrfs_check_raid_min_devices(root->fs_info, num_devices - 1);
1878 ret = btrfs_find_device_by_devspec(root, devid, device_path,
1883 if (device->is_tgtdev_for_dev_replace) {
1884 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1888 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1889 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1893 if (device->writeable) {
1895 list_del_init(&device->dev_alloc_list);
1896 device->fs_devices->rw_devices--;
1897 unlock_chunks(root);
1901 mutex_unlock(&uuid_mutex);
1902 ret = btrfs_shrink_device(device, 0);
1903 mutex_lock(&uuid_mutex);
1908 * TODO: the superblock still includes this device in its num_devices
1909 * counter although write_all_supers() is not locked out. This
1910 * could give a filesystem state which requires a degraded mount.
1912 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1916 device->in_fs_metadata = 0;
1917 btrfs_scrub_cancel_dev(root->fs_info, device);
1920 * the device list mutex makes sure that we don't change
1921 * the device list while someone else is writing out all
1922 * the device supers. Whoever is writing all supers, should
1923 * lock the device list mutex before getting the number of
1924 * devices in the super block (super_copy). Conversely,
1925 * whoever updates the number of devices in the super block
1926 * (super_copy) should hold the device list mutex.
1929 cur_devices = device->fs_devices;
1930 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1931 list_del_rcu(&device->dev_list);
1933 device->fs_devices->num_devices--;
1934 device->fs_devices->total_devices--;
1936 if (device->missing)
1937 device->fs_devices->missing_devices--;
1939 btrfs_assign_next_active_device(root->fs_info, device, NULL);
1942 device->fs_devices->open_devices--;
1943 /* remove sysfs entry */
1944 btrfs_sysfs_rm_device_link(root->fs_info->fs_devices, device);
1947 num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1948 btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1949 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1952 * at this point, the device is zero sized and detached from
1953 * the devices list. All that's left is to zero out the old
1954 * supers and free the device.
1956 if (device->writeable)
1957 btrfs_scratch_superblocks(device->bdev, device->name->str);
1959 btrfs_close_bdev(device);
1960 call_rcu(&device->rcu, free_device);
1962 if (cur_devices->open_devices == 0) {
1963 struct btrfs_fs_devices *fs_devices;
1964 fs_devices = root->fs_info->fs_devices;
1965 while (fs_devices) {
1966 if (fs_devices->seed == cur_devices) {
1967 fs_devices->seed = cur_devices->seed;
1970 fs_devices = fs_devices->seed;
1972 cur_devices->seed = NULL;
1973 __btrfs_close_devices(cur_devices);
1974 free_fs_devices(cur_devices);
1977 root->fs_info->num_tolerated_disk_barrier_failures =
1978 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1981 mutex_unlock(&uuid_mutex);
1985 if (device->writeable) {
1987 list_add(&device->dev_alloc_list,
1988 &root->fs_info->fs_devices->alloc_list);
1989 device->fs_devices->rw_devices++;
1990 unlock_chunks(root);
1995 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
1996 struct btrfs_device *srcdev)
1998 struct btrfs_fs_devices *fs_devices;
2000 WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
2003 * in case of fs with no seed, srcdev->fs_devices will point
2004 * to fs_devices of fs_info. However when the dev being replaced is
2005 * a seed dev it will point to the seed's local fs_devices. In short
2006 * srcdev will have its correct fs_devices in both the cases.
2008 fs_devices = srcdev->fs_devices;
2010 list_del_rcu(&srcdev->dev_list);
2011 list_del_rcu(&srcdev->dev_alloc_list);
2012 fs_devices->num_devices--;
2013 if (srcdev->missing)
2014 fs_devices->missing_devices--;
2016 if (srcdev->writeable)
2017 fs_devices->rw_devices--;
2020 fs_devices->open_devices--;
2023 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
2024 struct btrfs_device *srcdev)
2026 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2028 if (srcdev->writeable) {
2029 /* zero out the old super if it is writable */
2030 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
2033 btrfs_close_bdev(srcdev);
2035 call_rcu(&srcdev->rcu, free_device);
2038 * unless fs_devices is seed fs, num_devices shouldn't go
2041 BUG_ON(!fs_devices->num_devices && !fs_devices->seeding);
2043 /* if this is no devs we rather delete the fs_devices */
2044 if (!fs_devices->num_devices) {
2045 struct btrfs_fs_devices *tmp_fs_devices;
2047 tmp_fs_devices = fs_info->fs_devices;
2048 while (tmp_fs_devices) {
2049 if (tmp_fs_devices->seed == fs_devices) {
2050 tmp_fs_devices->seed = fs_devices->seed;
2053 tmp_fs_devices = tmp_fs_devices->seed;
2055 fs_devices->seed = NULL;
2056 __btrfs_close_devices(fs_devices);
2057 free_fs_devices(fs_devices);
2061 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2062 struct btrfs_device *tgtdev)
2064 mutex_lock(&uuid_mutex);
2066 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2068 btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev);
2071 fs_info->fs_devices->open_devices--;
2073 fs_info->fs_devices->num_devices--;
2075 btrfs_assign_next_active_device(fs_info, tgtdev, NULL);
2077 list_del_rcu(&tgtdev->dev_list);
2079 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2080 mutex_unlock(&uuid_mutex);
2083 * The update_dev_time() with in btrfs_scratch_superblocks()
2084 * may lead to a call to btrfs_show_devname() which will try
2085 * to hold device_list_mutex. And here this device
2086 * is already out of device list, so we don't have to hold
2087 * the device_list_mutex lock.
2089 btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
2091 btrfs_close_bdev(tgtdev);
2092 call_rcu(&tgtdev->rcu, free_device);
2095 static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
2096 struct btrfs_device **device)
2099 struct btrfs_super_block *disk_super;
2102 struct block_device *bdev;
2103 struct buffer_head *bh;
2106 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2107 root->fs_info->bdev_holder, 0, &bdev, &bh);
2110 disk_super = (struct btrfs_super_block *)bh->b_data;
2111 devid = btrfs_stack_device_id(&disk_super->dev_item);
2112 dev_uuid = disk_super->dev_item.uuid;
2113 *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2118 blkdev_put(bdev, FMODE_READ);
2122 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
2124 struct btrfs_device **device)
2127 if (strcmp(device_path, "missing") == 0) {
2128 struct list_head *devices;
2129 struct btrfs_device *tmp;
2131 devices = &root->fs_info->fs_devices->devices;
2133 * It is safe to read the devices since the volume_mutex
2134 * is held by the caller.
2136 list_for_each_entry(tmp, devices, dev_list) {
2137 if (tmp->in_fs_metadata && !tmp->bdev) {
2144 return BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2148 return btrfs_find_device_by_path(root, device_path, device);
2153 * Lookup a device given by device id, or the path if the id is 0.
2155 int btrfs_find_device_by_devspec(struct btrfs_root *root, u64 devid,
2157 struct btrfs_device **device)
2163 *device = btrfs_find_device(root->fs_info, devid, NULL,
2168 if (!devpath || !devpath[0])
2171 ret = btrfs_find_device_missing_or_by_path(root, devpath,
2178 * does all the dirty work required for changing file system's UUID.
2180 static int btrfs_prepare_sprout(struct btrfs_root *root)
2182 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2183 struct btrfs_fs_devices *old_devices;
2184 struct btrfs_fs_devices *seed_devices;
2185 struct btrfs_super_block *disk_super = root->fs_info->super_copy;
2186 struct btrfs_device *device;
2189 BUG_ON(!mutex_is_locked(&uuid_mutex));
2190 if (!fs_devices->seeding)
2193 seed_devices = __alloc_fs_devices();
2194 if (IS_ERR(seed_devices))
2195 return PTR_ERR(seed_devices);
2197 old_devices = clone_fs_devices(fs_devices);
2198 if (IS_ERR(old_devices)) {
2199 kfree(seed_devices);
2200 return PTR_ERR(old_devices);
2203 list_add(&old_devices->list, &fs_uuids);
2205 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2206 seed_devices->opened = 1;
2207 INIT_LIST_HEAD(&seed_devices->devices);
2208 INIT_LIST_HEAD(&seed_devices->alloc_list);
2209 mutex_init(&seed_devices->device_list_mutex);
2211 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2212 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2214 list_for_each_entry(device, &seed_devices->devices, dev_list)
2215 device->fs_devices = seed_devices;
2218 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2219 unlock_chunks(root);
2221 fs_devices->seeding = 0;
2222 fs_devices->num_devices = 0;
2223 fs_devices->open_devices = 0;
2224 fs_devices->missing_devices = 0;
2225 fs_devices->rotating = 0;
2226 fs_devices->seed = seed_devices;
2228 generate_random_uuid(fs_devices->fsid);
2229 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2230 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2231 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2233 super_flags = btrfs_super_flags(disk_super) &
2234 ~BTRFS_SUPER_FLAG_SEEDING;
2235 btrfs_set_super_flags(disk_super, super_flags);
2241 * Store the expected generation for seed devices in device items.
2243 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2244 struct btrfs_root *root)
2246 struct btrfs_path *path;
2247 struct extent_buffer *leaf;
2248 struct btrfs_dev_item *dev_item;
2249 struct btrfs_device *device;
2250 struct btrfs_key key;
2251 u8 fs_uuid[BTRFS_UUID_SIZE];
2252 u8 dev_uuid[BTRFS_UUID_SIZE];
2256 path = btrfs_alloc_path();
2260 root = root->fs_info->chunk_root;
2261 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2263 key.type = BTRFS_DEV_ITEM_KEY;
2266 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2270 leaf = path->nodes[0];
2272 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2273 ret = btrfs_next_leaf(root, path);
2278 leaf = path->nodes[0];
2279 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2280 btrfs_release_path(path);
2284 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2285 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2286 key.type != BTRFS_DEV_ITEM_KEY)
2289 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2290 struct btrfs_dev_item);
2291 devid = btrfs_device_id(leaf, dev_item);
2292 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2294 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2296 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2298 BUG_ON(!device); /* Logic error */
2300 if (device->fs_devices->seeding) {
2301 btrfs_set_device_generation(leaf, dev_item,
2302 device->generation);
2303 btrfs_mark_buffer_dirty(leaf);
2311 btrfs_free_path(path);
2315 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
2317 struct request_queue *q;
2318 struct btrfs_trans_handle *trans;
2319 struct btrfs_device *device;
2320 struct block_device *bdev;
2321 struct list_head *devices;
2322 struct super_block *sb = root->fs_info->sb;
2323 struct rcu_string *name;
2325 int seeding_dev = 0;
2328 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
2331 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2332 root->fs_info->bdev_holder);
2334 return PTR_ERR(bdev);
2336 if (root->fs_info->fs_devices->seeding) {
2338 down_write(&sb->s_umount);
2339 mutex_lock(&uuid_mutex);
2342 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2344 devices = &root->fs_info->fs_devices->devices;
2346 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2347 list_for_each_entry(device, devices, dev_list) {
2348 if (device->bdev == bdev) {
2351 &root->fs_info->fs_devices->device_list_mutex);
2355 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2357 device = btrfs_alloc_device(root->fs_info, NULL, NULL);
2358 if (IS_ERR(device)) {
2359 /* we can safely leave the fs_devices entry around */
2360 ret = PTR_ERR(device);
2364 name = rcu_string_strdup(device_path, GFP_KERNEL);
2370 rcu_assign_pointer(device->name, name);
2372 trans = btrfs_start_transaction(root, 0);
2373 if (IS_ERR(trans)) {
2374 rcu_string_free(device->name);
2376 ret = PTR_ERR(trans);
2380 q = bdev_get_queue(bdev);
2381 if (blk_queue_discard(q))
2382 device->can_discard = 1;
2383 device->writeable = 1;
2384 device->generation = trans->transid;
2385 device->io_width = root->sectorsize;
2386 device->io_align = root->sectorsize;
2387 device->sector_size = root->sectorsize;
2388 device->total_bytes = i_size_read(bdev->bd_inode);
2389 device->disk_total_bytes = device->total_bytes;
2390 device->commit_total_bytes = device->total_bytes;
2391 device->dev_root = root->fs_info->dev_root;
2392 device->bdev = bdev;
2393 device->in_fs_metadata = 1;
2394 device->is_tgtdev_for_dev_replace = 0;
2395 device->mode = FMODE_EXCL;
2396 device->dev_stats_valid = 1;
2397 set_blocksize(device->bdev, 4096);
2400 sb->s_flags &= ~MS_RDONLY;
2401 ret = btrfs_prepare_sprout(root);
2402 BUG_ON(ret); /* -ENOMEM */
2405 device->fs_devices = root->fs_info->fs_devices;
2407 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2409 list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
2410 list_add(&device->dev_alloc_list,
2411 &root->fs_info->fs_devices->alloc_list);
2412 root->fs_info->fs_devices->num_devices++;
2413 root->fs_info->fs_devices->open_devices++;
2414 root->fs_info->fs_devices->rw_devices++;
2415 root->fs_info->fs_devices->total_devices++;
2416 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2418 spin_lock(&root->fs_info->free_chunk_lock);
2419 root->fs_info->free_chunk_space += device->total_bytes;
2420 spin_unlock(&root->fs_info->free_chunk_lock);
2422 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2423 root->fs_info->fs_devices->rotating = 1;
2425 tmp = btrfs_super_total_bytes(root->fs_info->super_copy);
2426 btrfs_set_super_total_bytes(root->fs_info->super_copy,
2427 tmp + device->total_bytes);
2429 tmp = btrfs_super_num_devices(root->fs_info->super_copy);
2430 btrfs_set_super_num_devices(root->fs_info->super_copy,
2433 /* add sysfs device entry */
2434 btrfs_sysfs_add_device_link(root->fs_info->fs_devices, device);
2437 * we've got more storage, clear any full flags on the space
2440 btrfs_clear_space_info_full(root->fs_info);
2442 unlock_chunks(root);
2443 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2447 ret = init_first_rw_device(trans, root, device);
2448 unlock_chunks(root);
2450 btrfs_abort_transaction(trans, ret);
2455 ret = btrfs_add_device(trans, root, device);
2457 btrfs_abort_transaction(trans, ret);
2462 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2464 ret = btrfs_finish_sprout(trans, root);
2466 btrfs_abort_transaction(trans, ret);
2470 /* Sprouting would change fsid of the mounted root,
2471 * so rename the fsid on the sysfs
2473 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2474 root->fs_info->fsid);
2475 if (kobject_rename(&root->fs_info->fs_devices->fsid_kobj,
2477 btrfs_warn(root->fs_info,
2478 "sysfs: failed to create fsid for sprout");
2481 root->fs_info->num_tolerated_disk_barrier_failures =
2482 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
2483 ret = btrfs_commit_transaction(trans, root);
2486 mutex_unlock(&uuid_mutex);
2487 up_write(&sb->s_umount);
2489 if (ret) /* transaction commit */
2492 ret = btrfs_relocate_sys_chunks(root);
2494 btrfs_handle_fs_error(root->fs_info, ret,
2495 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2496 trans = btrfs_attach_transaction(root);
2497 if (IS_ERR(trans)) {
2498 if (PTR_ERR(trans) == -ENOENT)
2500 return PTR_ERR(trans);
2502 ret = btrfs_commit_transaction(trans, root);
2505 /* Update ctime/mtime for libblkid */
2506 update_dev_time(device_path);
2510 btrfs_end_transaction(trans, root);
2511 rcu_string_free(device->name);
2512 btrfs_sysfs_rm_device_link(root->fs_info->fs_devices, device);
2515 blkdev_put(bdev, FMODE_EXCL);
2517 mutex_unlock(&uuid_mutex);
2518 up_write(&sb->s_umount);
2523 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2524 struct btrfs_device *srcdev,
2525 struct btrfs_device **device_out)
2527 struct request_queue *q;
2528 struct btrfs_device *device;
2529 struct block_device *bdev;
2530 struct btrfs_fs_info *fs_info = root->fs_info;
2531 struct list_head *devices;
2532 struct rcu_string *name;
2533 u64 devid = BTRFS_DEV_REPLACE_DEVID;
2537 if (fs_info->fs_devices->seeding) {
2538 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
2542 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2543 fs_info->bdev_holder);
2545 btrfs_err(fs_info, "target device %s is invalid!", device_path);
2546 return PTR_ERR(bdev);
2549 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2551 devices = &fs_info->fs_devices->devices;
2552 list_for_each_entry(device, devices, dev_list) {
2553 if (device->bdev == bdev) {
2555 "target device is in the filesystem!");
2562 if (i_size_read(bdev->bd_inode) <
2563 btrfs_device_get_total_bytes(srcdev)) {
2565 "target device is smaller than source device!");
2571 device = btrfs_alloc_device(NULL, &devid, NULL);
2572 if (IS_ERR(device)) {
2573 ret = PTR_ERR(device);
2577 name = rcu_string_strdup(device_path, GFP_NOFS);
2583 rcu_assign_pointer(device->name, name);
2585 q = bdev_get_queue(bdev);
2586 if (blk_queue_discard(q))
2587 device->can_discard = 1;
2588 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2589 device->writeable = 1;
2590 device->generation = 0;
2591 device->io_width = root->sectorsize;
2592 device->io_align = root->sectorsize;
2593 device->sector_size = root->sectorsize;
2594 device->total_bytes = btrfs_device_get_total_bytes(srcdev);
2595 device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
2596 device->bytes_used = btrfs_device_get_bytes_used(srcdev);
2597 ASSERT(list_empty(&srcdev->resized_list));
2598 device->commit_total_bytes = srcdev->commit_total_bytes;
2599 device->commit_bytes_used = device->bytes_used;
2600 device->dev_root = fs_info->dev_root;
2601 device->bdev = bdev;
2602 device->in_fs_metadata = 1;
2603 device->is_tgtdev_for_dev_replace = 1;
2604 device->mode = FMODE_EXCL;
2605 device->dev_stats_valid = 1;
2606 set_blocksize(device->bdev, 4096);
2607 device->fs_devices = fs_info->fs_devices;
2608 list_add(&device->dev_list, &fs_info->fs_devices->devices);
2609 fs_info->fs_devices->num_devices++;
2610 fs_info->fs_devices->open_devices++;
2611 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2613 *device_out = device;
2617 blkdev_put(bdev, FMODE_EXCL);
2621 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2622 struct btrfs_device *tgtdev)
2624 WARN_ON(fs_info->fs_devices->rw_devices == 0);
2625 tgtdev->io_width = fs_info->dev_root->sectorsize;
2626 tgtdev->io_align = fs_info->dev_root->sectorsize;
2627 tgtdev->sector_size = fs_info->dev_root->sectorsize;
2628 tgtdev->dev_root = fs_info->dev_root;
2629 tgtdev->in_fs_metadata = 1;
2632 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2633 struct btrfs_device *device)
2636 struct btrfs_path *path;
2637 struct btrfs_root *root;
2638 struct btrfs_dev_item *dev_item;
2639 struct extent_buffer *leaf;
2640 struct btrfs_key key;
2642 root = device->dev_root->fs_info->chunk_root;
2644 path = btrfs_alloc_path();
2648 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2649 key.type = BTRFS_DEV_ITEM_KEY;
2650 key.offset = device->devid;
2652 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2661 leaf = path->nodes[0];
2662 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2664 btrfs_set_device_id(leaf, dev_item, device->devid);
2665 btrfs_set_device_type(leaf, dev_item, device->type);
2666 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2667 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2668 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2669 btrfs_set_device_total_bytes(leaf, dev_item,
2670 btrfs_device_get_disk_total_bytes(device));
2671 btrfs_set_device_bytes_used(leaf, dev_item,
2672 btrfs_device_get_bytes_used(device));
2673 btrfs_mark_buffer_dirty(leaf);
2676 btrfs_free_path(path);
2680 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2681 struct btrfs_device *device, u64 new_size)
2683 struct btrfs_super_block *super_copy =
2684 device->dev_root->fs_info->super_copy;
2685 struct btrfs_fs_devices *fs_devices;
2689 if (!device->writeable)
2692 lock_chunks(device->dev_root);
2693 old_total = btrfs_super_total_bytes(super_copy);
2694 diff = new_size - device->total_bytes;
2696 if (new_size <= device->total_bytes ||
2697 device->is_tgtdev_for_dev_replace) {
2698 unlock_chunks(device->dev_root);
2702 fs_devices = device->dev_root->fs_info->fs_devices;
2704 btrfs_set_super_total_bytes(super_copy, old_total + diff);
2705 device->fs_devices->total_rw_bytes += diff;
2707 btrfs_device_set_total_bytes(device, new_size);
2708 btrfs_device_set_disk_total_bytes(device, new_size);
2709 btrfs_clear_space_info_full(device->dev_root->fs_info);
2710 if (list_empty(&device->resized_list))
2711 list_add_tail(&device->resized_list,
2712 &fs_devices->resized_devices);
2713 unlock_chunks(device->dev_root);
2715 return btrfs_update_device(trans, device);
2718 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2719 struct btrfs_root *root, u64 chunk_objectid,
2723 struct btrfs_path *path;
2724 struct btrfs_key key;
2726 root = root->fs_info->chunk_root;
2727 path = btrfs_alloc_path();
2731 key.objectid = chunk_objectid;
2732 key.offset = chunk_offset;
2733 key.type = BTRFS_CHUNK_ITEM_KEY;
2735 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2738 else if (ret > 0) { /* Logic error or corruption */
2739 btrfs_handle_fs_error(root->fs_info, -ENOENT,
2740 "Failed lookup while freeing chunk.");
2745 ret = btrfs_del_item(trans, root, path);
2747 btrfs_handle_fs_error(root->fs_info, ret,
2748 "Failed to delete chunk item.");
2750 btrfs_free_path(path);
2754 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2757 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2758 struct btrfs_disk_key *disk_key;
2759 struct btrfs_chunk *chunk;
2766 struct btrfs_key key;
2769 array_size = btrfs_super_sys_array_size(super_copy);
2771 ptr = super_copy->sys_chunk_array;
2774 while (cur < array_size) {
2775 disk_key = (struct btrfs_disk_key *)ptr;
2776 btrfs_disk_key_to_cpu(&key, disk_key);
2778 len = sizeof(*disk_key);
2780 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2781 chunk = (struct btrfs_chunk *)(ptr + len);
2782 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2783 len += btrfs_chunk_item_size(num_stripes);
2788 if (key.objectid == chunk_objectid &&
2789 key.offset == chunk_offset) {
2790 memmove(ptr, ptr + len, array_size - (cur + len));
2792 btrfs_set_super_sys_array_size(super_copy, array_size);
2798 unlock_chunks(root);
2802 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2803 struct btrfs_root *root, u64 chunk_offset)
2805 struct extent_map_tree *em_tree;
2806 struct extent_map *em;
2807 struct btrfs_root *extent_root = root->fs_info->extent_root;
2808 struct map_lookup *map;
2809 u64 dev_extent_len = 0;
2810 u64 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2812 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2815 root = root->fs_info->chunk_root;
2816 em_tree = &root->fs_info->mapping_tree.map_tree;
2818 read_lock(&em_tree->lock);
2819 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2820 read_unlock(&em_tree->lock);
2822 if (!em || em->start > chunk_offset ||
2823 em->start + em->len < chunk_offset) {
2825 * This is a logic error, but we don't want to just rely on the
2826 * user having built with ASSERT enabled, so if ASSERT doesn't
2827 * do anything we still error out.
2831 free_extent_map(em);
2834 map = em->map_lookup;
2835 lock_chunks(root->fs_info->chunk_root);
2836 check_system_chunk(trans, extent_root, map->type);
2837 unlock_chunks(root->fs_info->chunk_root);
2840 * Take the device list mutex to prevent races with the final phase of
2841 * a device replace operation that replaces the device object associated
2842 * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2844 mutex_lock(&fs_devices->device_list_mutex);
2845 for (i = 0; i < map->num_stripes; i++) {
2846 struct btrfs_device *device = map->stripes[i].dev;
2847 ret = btrfs_free_dev_extent(trans, device,
2848 map->stripes[i].physical,
2851 mutex_unlock(&fs_devices->device_list_mutex);
2852 btrfs_abort_transaction(trans, ret);
2856 if (device->bytes_used > 0) {
2858 btrfs_device_set_bytes_used(device,
2859 device->bytes_used - dev_extent_len);
2860 spin_lock(&root->fs_info->free_chunk_lock);
2861 root->fs_info->free_chunk_space += dev_extent_len;
2862 spin_unlock(&root->fs_info->free_chunk_lock);
2863 btrfs_clear_space_info_full(root->fs_info);
2864 unlock_chunks(root);
2867 if (map->stripes[i].dev) {
2868 ret = btrfs_update_device(trans, map->stripes[i].dev);
2870 mutex_unlock(&fs_devices->device_list_mutex);
2871 btrfs_abort_transaction(trans, ret);
2876 mutex_unlock(&fs_devices->device_list_mutex);
2878 ret = btrfs_free_chunk(trans, root, chunk_objectid, chunk_offset);
2880 btrfs_abort_transaction(trans, ret);
2884 trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2886 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2887 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2889 btrfs_abort_transaction(trans, ret);
2894 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset, em);
2896 btrfs_abort_transaction(trans, ret);
2902 free_extent_map(em);
2906 static int btrfs_relocate_chunk(struct btrfs_root *root, u64 chunk_offset)
2908 struct btrfs_root *extent_root;
2909 struct btrfs_trans_handle *trans;
2912 root = root->fs_info->chunk_root;
2913 extent_root = root->fs_info->extent_root;
2916 * Prevent races with automatic removal of unused block groups.
2917 * After we relocate and before we remove the chunk with offset
2918 * chunk_offset, automatic removal of the block group can kick in,
2919 * resulting in a failure when calling btrfs_remove_chunk() below.
2921 * Make sure to acquire this mutex before doing a tree search (dev
2922 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2923 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2924 * we release the path used to search the chunk/dev tree and before
2925 * the current task acquires this mutex and calls us.
2927 ASSERT(mutex_is_locked(&root->fs_info->delete_unused_bgs_mutex));
2929 ret = btrfs_can_relocate(extent_root, chunk_offset);
2933 /* step one, relocate all the extents inside this chunk */
2934 btrfs_scrub_pause(root);
2935 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2936 btrfs_scrub_continue(root);
2940 trans = btrfs_start_trans_remove_block_group(root->fs_info,
2942 if (IS_ERR(trans)) {
2943 ret = PTR_ERR(trans);
2944 btrfs_handle_fs_error(root->fs_info, ret, NULL);
2949 * step two, delete the device extents and the
2950 * chunk tree entries
2952 ret = btrfs_remove_chunk(trans, root, chunk_offset);
2953 btrfs_end_transaction(trans, extent_root);
2957 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2959 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2960 struct btrfs_path *path;
2961 struct extent_buffer *leaf;
2962 struct btrfs_chunk *chunk;
2963 struct btrfs_key key;
2964 struct btrfs_key found_key;
2966 bool retried = false;
2970 path = btrfs_alloc_path();
2975 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2976 key.offset = (u64)-1;
2977 key.type = BTRFS_CHUNK_ITEM_KEY;
2980 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
2981 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2983 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2986 BUG_ON(ret == 0); /* Corruption */
2988 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2991 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2997 leaf = path->nodes[0];
2998 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3000 chunk = btrfs_item_ptr(leaf, path->slots[0],
3001 struct btrfs_chunk);
3002 chunk_type = btrfs_chunk_type(leaf, chunk);
3003 btrfs_release_path(path);
3005 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3006 ret = btrfs_relocate_chunk(chunk_root,
3013 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
3015 if (found_key.offset == 0)
3017 key.offset = found_key.offset - 1;
3020 if (failed && !retried) {
3024 } else if (WARN_ON(failed && retried)) {
3028 btrfs_free_path(path);
3032 static int insert_balance_item(struct btrfs_root *root,
3033 struct btrfs_balance_control *bctl)
3035 struct btrfs_trans_handle *trans;
3036 struct btrfs_balance_item *item;
3037 struct btrfs_disk_balance_args disk_bargs;
3038 struct btrfs_path *path;
3039 struct extent_buffer *leaf;
3040 struct btrfs_key key;
3043 path = btrfs_alloc_path();
3047 trans = btrfs_start_transaction(root, 0);
3048 if (IS_ERR(trans)) {
3049 btrfs_free_path(path);
3050 return PTR_ERR(trans);
3053 key.objectid = BTRFS_BALANCE_OBJECTID;
3054 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3057 ret = btrfs_insert_empty_item(trans, root, path, &key,
3062 leaf = path->nodes[0];
3063 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3065 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
3067 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3068 btrfs_set_balance_data(leaf, item, &disk_bargs);
3069 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3070 btrfs_set_balance_meta(leaf, item, &disk_bargs);
3071 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3072 btrfs_set_balance_sys(leaf, item, &disk_bargs);
3074 btrfs_set_balance_flags(leaf, item, bctl->flags);
3076 btrfs_mark_buffer_dirty(leaf);
3078 btrfs_free_path(path);
3079 err = btrfs_commit_transaction(trans, root);
3085 static int del_balance_item(struct btrfs_root *root)
3087 struct btrfs_trans_handle *trans;
3088 struct btrfs_path *path;
3089 struct btrfs_key key;
3092 path = btrfs_alloc_path();
3096 trans = btrfs_start_transaction(root, 0);
3097 if (IS_ERR(trans)) {
3098 btrfs_free_path(path);
3099 return PTR_ERR(trans);
3102 key.objectid = BTRFS_BALANCE_OBJECTID;
3103 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3106 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3114 ret = btrfs_del_item(trans, root, path);
3116 btrfs_free_path(path);
3117 err = btrfs_commit_transaction(trans, root);
3124 * This is a heuristic used to reduce the number of chunks balanced on
3125 * resume after balance was interrupted.
3127 static void update_balance_args(struct btrfs_balance_control *bctl)
3130 * Turn on soft mode for chunk types that were being converted.
3132 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3133 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3134 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3135 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3136 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3137 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3140 * Turn on usage filter if is not already used. The idea is
3141 * that chunks that we have already balanced should be
3142 * reasonably full. Don't do it for chunks that are being
3143 * converted - that will keep us from relocating unconverted
3144 * (albeit full) chunks.
3146 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3147 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3148 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3149 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3150 bctl->data.usage = 90;
3152 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3153 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3154 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3155 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3156 bctl->sys.usage = 90;
3158 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3159 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3160 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3161 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3162 bctl->meta.usage = 90;
3167 * Should be called with both balance and volume mutexes held to
3168 * serialize other volume operations (add_dev/rm_dev/resize) with
3169 * restriper. Same goes for unset_balance_control.
3171 static void set_balance_control(struct btrfs_balance_control *bctl)
3173 struct btrfs_fs_info *fs_info = bctl->fs_info;
3175 BUG_ON(fs_info->balance_ctl);
3177 spin_lock(&fs_info->balance_lock);
3178 fs_info->balance_ctl = bctl;
3179 spin_unlock(&fs_info->balance_lock);
3182 static void unset_balance_control(struct btrfs_fs_info *fs_info)
3184 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3186 BUG_ON(!fs_info->balance_ctl);
3188 spin_lock(&fs_info->balance_lock);
3189 fs_info->balance_ctl = NULL;
3190 spin_unlock(&fs_info->balance_lock);
3196 * Balance filters. Return 1 if chunk should be filtered out
3197 * (should not be balanced).
3199 static int chunk_profiles_filter(u64 chunk_type,
3200 struct btrfs_balance_args *bargs)
3202 chunk_type = chunk_to_extended(chunk_type) &
3203 BTRFS_EXTENDED_PROFILE_MASK;
3205 if (bargs->profiles & chunk_type)
3211 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3212 struct btrfs_balance_args *bargs)
3214 struct btrfs_block_group_cache *cache;
3216 u64 user_thresh_min;
3217 u64 user_thresh_max;
3220 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3221 chunk_used = btrfs_block_group_used(&cache->item);
3223 if (bargs->usage_min == 0)
3224 user_thresh_min = 0;
3226 user_thresh_min = div_factor_fine(cache->key.offset,
3229 if (bargs->usage_max == 0)
3230 user_thresh_max = 1;
3231 else if (bargs->usage_max > 100)
3232 user_thresh_max = cache->key.offset;
3234 user_thresh_max = div_factor_fine(cache->key.offset,
3237 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3240 btrfs_put_block_group(cache);
3244 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3245 u64 chunk_offset, struct btrfs_balance_args *bargs)
3247 struct btrfs_block_group_cache *cache;
3248 u64 chunk_used, user_thresh;
3251 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3252 chunk_used = btrfs_block_group_used(&cache->item);
3254 if (bargs->usage_min == 0)
3256 else if (bargs->usage > 100)
3257 user_thresh = cache->key.offset;
3259 user_thresh = div_factor_fine(cache->key.offset,
3262 if (chunk_used < user_thresh)
3265 btrfs_put_block_group(cache);
3269 static int chunk_devid_filter(struct extent_buffer *leaf,
3270 struct btrfs_chunk *chunk,
3271 struct btrfs_balance_args *bargs)
3273 struct btrfs_stripe *stripe;
3274 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3277 for (i = 0; i < num_stripes; i++) {
3278 stripe = btrfs_stripe_nr(chunk, i);
3279 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3286 /* [pstart, pend) */
3287 static int chunk_drange_filter(struct extent_buffer *leaf,
3288 struct btrfs_chunk *chunk,
3290 struct btrfs_balance_args *bargs)
3292 struct btrfs_stripe *stripe;
3293 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3299 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3302 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3303 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3304 factor = num_stripes / 2;
3305 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3306 factor = num_stripes - 1;
3307 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3308 factor = num_stripes - 2;
3310 factor = num_stripes;
3313 for (i = 0; i < num_stripes; i++) {
3314 stripe = btrfs_stripe_nr(chunk, i);
3315 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3318 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3319 stripe_length = btrfs_chunk_length(leaf, chunk);
3320 stripe_length = div_u64(stripe_length, factor);
3322 if (stripe_offset < bargs->pend &&
3323 stripe_offset + stripe_length > bargs->pstart)
3330 /* [vstart, vend) */
3331 static int chunk_vrange_filter(struct extent_buffer *leaf,
3332 struct btrfs_chunk *chunk,
3334 struct btrfs_balance_args *bargs)
3336 if (chunk_offset < bargs->vend &&
3337 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3338 /* at least part of the chunk is inside this vrange */
3344 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3345 struct btrfs_chunk *chunk,
3346 struct btrfs_balance_args *bargs)
3348 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3350 if (bargs->stripes_min <= num_stripes
3351 && num_stripes <= bargs->stripes_max)
3357 static int chunk_soft_convert_filter(u64 chunk_type,
3358 struct btrfs_balance_args *bargs)
3360 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3363 chunk_type = chunk_to_extended(chunk_type) &
3364 BTRFS_EXTENDED_PROFILE_MASK;
3366 if (bargs->target == chunk_type)
3372 static int should_balance_chunk(struct btrfs_root *root,
3373 struct extent_buffer *leaf,
3374 struct btrfs_chunk *chunk, u64 chunk_offset)
3376 struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
3377 struct btrfs_balance_args *bargs = NULL;
3378 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3381 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3382 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3386 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3387 bargs = &bctl->data;
3388 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3390 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3391 bargs = &bctl->meta;
3393 /* profiles filter */
3394 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3395 chunk_profiles_filter(chunk_type, bargs)) {
3400 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3401 chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
3403 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3404 chunk_usage_range_filter(bctl->fs_info, chunk_offset, bargs)) {
3409 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3410 chunk_devid_filter(leaf, chunk, bargs)) {
3414 /* drange filter, makes sense only with devid filter */
3415 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3416 chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
3421 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3422 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3426 /* stripes filter */
3427 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3428 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3432 /* soft profile changing mode */
3433 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3434 chunk_soft_convert_filter(chunk_type, bargs)) {
3439 * limited by count, must be the last filter
3441 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3442 if (bargs->limit == 0)
3446 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3448 * Same logic as the 'limit' filter; the minimum cannot be
3449 * determined here because we do not have the global information
3450 * about the count of all chunks that satisfy the filters.
3452 if (bargs->limit_max == 0)
3461 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3463 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3464 struct btrfs_root *chunk_root = fs_info->chunk_root;
3465 struct btrfs_root *dev_root = fs_info->dev_root;
3466 struct list_head *devices;
3467 struct btrfs_device *device;
3471 struct btrfs_chunk *chunk;
3472 struct btrfs_path *path = NULL;
3473 struct btrfs_key key;
3474 struct btrfs_key found_key;
3475 struct btrfs_trans_handle *trans;
3476 struct extent_buffer *leaf;
3479 int enospc_errors = 0;
3480 bool counting = true;
3481 /* The single value limit and min/max limits use the same bytes in the */
3482 u64 limit_data = bctl->data.limit;
3483 u64 limit_meta = bctl->meta.limit;
3484 u64 limit_sys = bctl->sys.limit;
3488 int chunk_reserved = 0;
3491 /* step one make some room on all the devices */
3492 devices = &fs_info->fs_devices->devices;
3493 list_for_each_entry(device, devices, dev_list) {
3494 old_size = btrfs_device_get_total_bytes(device);
3495 size_to_free = div_factor(old_size, 1);
3496 size_to_free = min_t(u64, size_to_free, SZ_1M);
3497 if (!device->writeable ||
3498 btrfs_device_get_total_bytes(device) -
3499 btrfs_device_get_bytes_used(device) > size_to_free ||
3500 device->is_tgtdev_for_dev_replace)
3503 ret = btrfs_shrink_device(device, old_size - size_to_free);
3507 /* btrfs_shrink_device never returns ret > 0 */
3512 trans = btrfs_start_transaction(dev_root, 0);
3513 if (IS_ERR(trans)) {
3514 ret = PTR_ERR(trans);
3515 btrfs_info_in_rcu(fs_info,
3516 "resize: unable to start transaction after shrinking device %s (error %d), old size %llu, new size %llu",
3517 rcu_str_deref(device->name), ret,
3518 old_size, old_size - size_to_free);
3522 ret = btrfs_grow_device(trans, device, old_size);
3524 btrfs_end_transaction(trans, dev_root);
3525 /* btrfs_grow_device never returns ret > 0 */
3527 btrfs_info_in_rcu(fs_info,
3528 "resize: unable to grow device after shrinking device %s (error %d), old size %llu, new size %llu",
3529 rcu_str_deref(device->name), ret,
3530 old_size, old_size - size_to_free);
3534 btrfs_end_transaction(trans, dev_root);
3537 /* step two, relocate all the chunks */
3538 path = btrfs_alloc_path();
3544 /* zero out stat counters */
3545 spin_lock(&fs_info->balance_lock);
3546 memset(&bctl->stat, 0, sizeof(bctl->stat));
3547 spin_unlock(&fs_info->balance_lock);
3551 * The single value limit and min/max limits use the same bytes
3554 bctl->data.limit = limit_data;
3555 bctl->meta.limit = limit_meta;
3556 bctl->sys.limit = limit_sys;
3558 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3559 key.offset = (u64)-1;
3560 key.type = BTRFS_CHUNK_ITEM_KEY;
3563 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3564 atomic_read(&fs_info->balance_cancel_req)) {
3569 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3570 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3572 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3577 * this shouldn't happen, it means the last relocate
3581 BUG(); /* FIXME break ? */
3583 ret = btrfs_previous_item(chunk_root, path, 0,
3584 BTRFS_CHUNK_ITEM_KEY);
3586 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3591 leaf = path->nodes[0];
3592 slot = path->slots[0];
3593 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3595 if (found_key.objectid != key.objectid) {
3596 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3600 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3601 chunk_type = btrfs_chunk_type(leaf, chunk);
3604 spin_lock(&fs_info->balance_lock);
3605 bctl->stat.considered++;
3606 spin_unlock(&fs_info->balance_lock);
3609 ret = should_balance_chunk(chunk_root, leaf, chunk,
3612 btrfs_release_path(path);
3614 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3619 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3620 spin_lock(&fs_info->balance_lock);
3621 bctl->stat.expected++;
3622 spin_unlock(&fs_info->balance_lock);
3624 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3626 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3628 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3635 * Apply limit_min filter, no need to check if the LIMITS
3636 * filter is used, limit_min is 0 by default
3638 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3639 count_data < bctl->data.limit_min)
3640 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3641 count_meta < bctl->meta.limit_min)
3642 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3643 count_sys < bctl->sys.limit_min)) {
3644 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3648 ASSERT(fs_info->data_sinfo);
3649 spin_lock(&fs_info->data_sinfo->lock);
3650 bytes_used = fs_info->data_sinfo->bytes_used;
3651 spin_unlock(&fs_info->data_sinfo->lock);
3653 if ((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3654 !chunk_reserved && !bytes_used) {
3655 trans = btrfs_start_transaction(chunk_root, 0);
3656 if (IS_ERR(trans)) {
3657 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3658 ret = PTR_ERR(trans);
3662 ret = btrfs_force_chunk_alloc(trans, chunk_root,
3663 BTRFS_BLOCK_GROUP_DATA);
3664 btrfs_end_transaction(trans, chunk_root);
3666 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3672 ret = btrfs_relocate_chunk(chunk_root,
3674 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3675 if (ret && ret != -ENOSPC)
3677 if (ret == -ENOSPC) {
3680 spin_lock(&fs_info->balance_lock);
3681 bctl->stat.completed++;
3682 spin_unlock(&fs_info->balance_lock);
3685 if (found_key.offset == 0)
3687 key.offset = found_key.offset - 1;
3691 btrfs_release_path(path);
3696 btrfs_free_path(path);
3697 if (enospc_errors) {
3698 btrfs_info(fs_info, "%d enospc errors during balance",
3708 * alloc_profile_is_valid - see if a given profile is valid and reduced
3709 * @flags: profile to validate
3710 * @extended: if true @flags is treated as an extended profile
3712 static int alloc_profile_is_valid(u64 flags, int extended)
3714 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3715 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3717 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3719 /* 1) check that all other bits are zeroed */
3723 /* 2) see if profile is reduced */
3725 return !extended; /* "0" is valid for usual profiles */
3727 /* true if exactly one bit set */
3728 return (flags & (flags - 1)) == 0;
3731 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3733 /* cancel requested || normal exit path */
3734 return atomic_read(&fs_info->balance_cancel_req) ||
3735 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3736 atomic_read(&fs_info->balance_cancel_req) == 0);
3739 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3743 unset_balance_control(fs_info);
3744 ret = del_balance_item(fs_info->tree_root);
3746 btrfs_handle_fs_error(fs_info, ret, NULL);
3748 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3751 /* Non-zero return value signifies invalidity */
3752 static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg,
3755 return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3756 (!alloc_profile_is_valid(bctl_arg->target, 1) ||
3757 (bctl_arg->target & ~allowed)));
3761 * Should be called with both balance and volume mutexes held
3763 int btrfs_balance(struct btrfs_balance_control *bctl,
3764 struct btrfs_ioctl_balance_args *bargs)
3766 struct btrfs_fs_info *fs_info = bctl->fs_info;
3773 if (btrfs_fs_closing(fs_info) ||
3774 atomic_read(&fs_info->balance_pause_req) ||
3775 atomic_read(&fs_info->balance_cancel_req)) {
3780 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3781 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3785 * In case of mixed groups both data and meta should be picked,
3786 * and identical options should be given for both of them.
3788 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3789 if (mixed && (bctl->flags & allowed)) {
3790 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3791 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3792 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3794 "with mixed groups data and metadata balance options must be the same");
3800 num_devices = fs_info->fs_devices->num_devices;
3801 btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
3802 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3803 BUG_ON(num_devices < 1);
3806 btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
3807 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE | BTRFS_BLOCK_GROUP_DUP;
3808 if (num_devices > 1)
3809 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3810 if (num_devices > 2)
3811 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3812 if (num_devices > 3)
3813 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3814 BTRFS_BLOCK_GROUP_RAID6);
3815 if (validate_convert_profile(&bctl->data, allowed)) {
3817 "unable to start balance with target data profile %llu",
3822 if (validate_convert_profile(&bctl->meta, allowed)) {
3824 "unable to start balance with target metadata profile %llu",
3829 if (validate_convert_profile(&bctl->sys, allowed)) {
3831 "unable to start balance with target system profile %llu",
3837 /* allow to reduce meta or sys integrity only if force set */
3838 allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3839 BTRFS_BLOCK_GROUP_RAID10 |
3840 BTRFS_BLOCK_GROUP_RAID5 |
3841 BTRFS_BLOCK_GROUP_RAID6;
3843 seq = read_seqbegin(&fs_info->profiles_lock);
3845 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3846 (fs_info->avail_system_alloc_bits & allowed) &&
3847 !(bctl->sys.target & allowed)) ||
3848 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3849 (fs_info->avail_metadata_alloc_bits & allowed) &&
3850 !(bctl->meta.target & allowed))) {
3851 if (bctl->flags & BTRFS_BALANCE_FORCE) {
3853 "force reducing metadata integrity");
3856 "balance will reduce metadata integrity, use force if you want this");
3861 } while (read_seqretry(&fs_info->profiles_lock, seq));
3863 if (btrfs_get_num_tolerated_disk_barrier_failures(bctl->meta.target) <
3864 btrfs_get_num_tolerated_disk_barrier_failures(bctl->data.target)) {
3866 "metadata profile 0x%llx has lower redundancy than data profile 0x%llx",
3867 bctl->meta.target, bctl->data.target);
3870 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3871 fs_info->num_tolerated_disk_barrier_failures = min(
3872 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info),
3873 btrfs_get_num_tolerated_disk_barrier_failures(
3877 ret = insert_balance_item(fs_info->tree_root, bctl);
3878 if (ret && ret != -EEXIST)
3881 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3882 BUG_ON(ret == -EEXIST);
3883 set_balance_control(bctl);
3885 BUG_ON(ret != -EEXIST);
3886 spin_lock(&fs_info->balance_lock);
3887 update_balance_args(bctl);
3888 spin_unlock(&fs_info->balance_lock);
3891 atomic_inc(&fs_info->balance_running);
3892 mutex_unlock(&fs_info->balance_mutex);
3894 ret = __btrfs_balance(fs_info);
3896 mutex_lock(&fs_info->balance_mutex);
3897 atomic_dec(&fs_info->balance_running);
3899 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3900 fs_info->num_tolerated_disk_barrier_failures =
3901 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3905 memset(bargs, 0, sizeof(*bargs));
3906 update_ioctl_balance_args(fs_info, 0, bargs);
3909 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3910 balance_need_close(fs_info)) {
3911 __cancel_balance(fs_info);
3914 wake_up(&fs_info->balance_wait_q);
3918 if (bctl->flags & BTRFS_BALANCE_RESUME)
3919 __cancel_balance(fs_info);
3922 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3927 static int balance_kthread(void *data)
3929 struct btrfs_fs_info *fs_info = data;
3932 mutex_lock(&fs_info->volume_mutex);
3933 mutex_lock(&fs_info->balance_mutex);
3935 if (fs_info->balance_ctl) {
3936 btrfs_info(fs_info, "continuing balance");
3937 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3940 mutex_unlock(&fs_info->balance_mutex);
3941 mutex_unlock(&fs_info->volume_mutex);
3946 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3948 struct task_struct *tsk;
3950 spin_lock(&fs_info->balance_lock);
3951 if (!fs_info->balance_ctl) {
3952 spin_unlock(&fs_info->balance_lock);
3955 spin_unlock(&fs_info->balance_lock);
3957 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
3958 btrfs_info(fs_info, "force skipping balance");
3962 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3963 return PTR_ERR_OR_ZERO(tsk);
3966 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3968 struct btrfs_balance_control *bctl;
3969 struct btrfs_balance_item *item;
3970 struct btrfs_disk_balance_args disk_bargs;
3971 struct btrfs_path *path;
3972 struct extent_buffer *leaf;
3973 struct btrfs_key key;
3976 path = btrfs_alloc_path();
3980 key.objectid = BTRFS_BALANCE_OBJECTID;
3981 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3984 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3987 if (ret > 0) { /* ret = -ENOENT; */
3992 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3998 leaf = path->nodes[0];
3999 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
4001 bctl->fs_info = fs_info;
4002 bctl->flags = btrfs_balance_flags(leaf, item);
4003 bctl->flags |= BTRFS_BALANCE_RESUME;
4005 btrfs_balance_data(leaf, item, &disk_bargs);
4006 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4007 btrfs_balance_meta(leaf, item, &disk_bargs);
4008 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4009 btrfs_balance_sys(leaf, item, &disk_bargs);
4010 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4012 WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
4014 mutex_lock(&fs_info->volume_mutex);
4015 mutex_lock(&fs_info->balance_mutex);
4017 set_balance_control(bctl);
4019 mutex_unlock(&fs_info->balance_mutex);
4020 mutex_unlock(&fs_info->volume_mutex);
4022 btrfs_free_path(path);
4026 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4030 mutex_lock(&fs_info->balance_mutex);
4031 if (!fs_info->balance_ctl) {
4032 mutex_unlock(&fs_info->balance_mutex);
4036 if (atomic_read(&fs_info->balance_running)) {
4037 atomic_inc(&fs_info->balance_pause_req);
4038 mutex_unlock(&fs_info->balance_mutex);
4040 wait_event(fs_info->balance_wait_q,
4041 atomic_read(&fs_info->balance_running) == 0);
4043 mutex_lock(&fs_info->balance_mutex);
4044 /* we are good with balance_ctl ripped off from under us */
4045 BUG_ON(atomic_read(&fs_info->balance_running));
4046 atomic_dec(&fs_info->balance_pause_req);
4051 mutex_unlock(&fs_info->balance_mutex);
4055 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4057 if (fs_info->sb->s_flags & MS_RDONLY)
4060 mutex_lock(&fs_info->balance_mutex);
4061 if (!fs_info->balance_ctl) {
4062 mutex_unlock(&fs_info->balance_mutex);
4066 atomic_inc(&fs_info->balance_cancel_req);
4068 * if we are running just wait and return, balance item is
4069 * deleted in btrfs_balance in this case
4071 if (atomic_read(&fs_info->balance_running)) {
4072 mutex_unlock(&fs_info->balance_mutex);
4073 wait_event(fs_info->balance_wait_q,
4074 atomic_read(&fs_info->balance_running) == 0);
4075 mutex_lock(&fs_info->balance_mutex);
4077 /* __cancel_balance needs volume_mutex */
4078 mutex_unlock(&fs_info->balance_mutex);
4079 mutex_lock(&fs_info->volume_mutex);
4080 mutex_lock(&fs_info->balance_mutex);
4082 if (fs_info->balance_ctl)
4083 __cancel_balance(fs_info);
4085 mutex_unlock(&fs_info->volume_mutex);
4088 BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
4089 atomic_dec(&fs_info->balance_cancel_req);
4090 mutex_unlock(&fs_info->balance_mutex);
4094 static int btrfs_uuid_scan_kthread(void *data)
4096 struct btrfs_fs_info *fs_info = data;
4097 struct btrfs_root *root = fs_info->tree_root;
4098 struct btrfs_key key;
4099 struct btrfs_key max_key;
4100 struct btrfs_path *path = NULL;
4102 struct extent_buffer *eb;
4104 struct btrfs_root_item root_item;
4106 struct btrfs_trans_handle *trans = NULL;
4108 path = btrfs_alloc_path();
4115 key.type = BTRFS_ROOT_ITEM_KEY;
4118 max_key.objectid = (u64)-1;
4119 max_key.type = BTRFS_ROOT_ITEM_KEY;
4120 max_key.offset = (u64)-1;
4123 ret = btrfs_search_forward(root, &key, path, 0);
4130 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4131 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4132 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4133 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4136 eb = path->nodes[0];
4137 slot = path->slots[0];
4138 item_size = btrfs_item_size_nr(eb, slot);
4139 if (item_size < sizeof(root_item))
4142 read_extent_buffer(eb, &root_item,
4143 btrfs_item_ptr_offset(eb, slot),
4144 (int)sizeof(root_item));
4145 if (btrfs_root_refs(&root_item) == 0)
4148 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4149 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4153 btrfs_release_path(path);
4155 * 1 - subvol uuid item
4156 * 1 - received_subvol uuid item
4158 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4159 if (IS_ERR(trans)) {
4160 ret = PTR_ERR(trans);
4168 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4169 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
4171 BTRFS_UUID_KEY_SUBVOL,
4174 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4180 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4181 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
4182 root_item.received_uuid,
4183 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4186 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4194 ret = btrfs_end_transaction(trans, fs_info->uuid_root);
4200 btrfs_release_path(path);
4201 if (key.offset < (u64)-1) {
4203 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4205 key.type = BTRFS_ROOT_ITEM_KEY;
4206 } else if (key.objectid < (u64)-1) {
4208 key.type = BTRFS_ROOT_ITEM_KEY;
4217 btrfs_free_path(path);
4218 if (trans && !IS_ERR(trans))
4219 btrfs_end_transaction(trans, fs_info->uuid_root);
4221 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4223 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4224 up(&fs_info->uuid_tree_rescan_sem);
4229 * Callback for btrfs_uuid_tree_iterate().
4231 * 0 check succeeded, the entry is not outdated.
4232 * < 0 if an error occurred.
4233 * > 0 if the check failed, which means the caller shall remove the entry.
4235 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
4236 u8 *uuid, u8 type, u64 subid)
4238 struct btrfs_key key;
4240 struct btrfs_root *subvol_root;
4242 if (type != BTRFS_UUID_KEY_SUBVOL &&
4243 type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
4246 key.objectid = subid;
4247 key.type = BTRFS_ROOT_ITEM_KEY;
4248 key.offset = (u64)-1;
4249 subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
4250 if (IS_ERR(subvol_root)) {
4251 ret = PTR_ERR(subvol_root);
4258 case BTRFS_UUID_KEY_SUBVOL:
4259 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
4262 case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
4263 if (memcmp(uuid, subvol_root->root_item.received_uuid,
4273 static int btrfs_uuid_rescan_kthread(void *data)
4275 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4279 * 1st step is to iterate through the existing UUID tree and
4280 * to delete all entries that contain outdated data.
4281 * 2nd step is to add all missing entries to the UUID tree.
4283 ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4285 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4286 up(&fs_info->uuid_tree_rescan_sem);
4289 return btrfs_uuid_scan_kthread(data);
4292 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4294 struct btrfs_trans_handle *trans;
4295 struct btrfs_root *tree_root = fs_info->tree_root;
4296 struct btrfs_root *uuid_root;
4297 struct task_struct *task;
4304 trans = btrfs_start_transaction(tree_root, 2);
4306 return PTR_ERR(trans);
4308 uuid_root = btrfs_create_tree(trans, fs_info,
4309 BTRFS_UUID_TREE_OBJECTID);
4310 if (IS_ERR(uuid_root)) {
4311 ret = PTR_ERR(uuid_root);
4312 btrfs_abort_transaction(trans, ret);
4313 btrfs_end_transaction(trans, tree_root);
4317 fs_info->uuid_root = uuid_root;
4319 ret = btrfs_commit_transaction(trans, tree_root);
4323 down(&fs_info->uuid_tree_rescan_sem);
4324 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4326 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4327 btrfs_warn(fs_info, "failed to start uuid_scan task");
4328 up(&fs_info->uuid_tree_rescan_sem);
4329 return PTR_ERR(task);
4335 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4337 struct task_struct *task;
4339 down(&fs_info->uuid_tree_rescan_sem);
4340 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4342 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4343 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4344 up(&fs_info->uuid_tree_rescan_sem);
4345 return PTR_ERR(task);
4352 * shrinking a device means finding all of the device extents past
4353 * the new size, and then following the back refs to the chunks.
4354 * The chunk relocation code actually frees the device extent
4356 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4358 struct btrfs_trans_handle *trans;
4359 struct btrfs_root *root = device->dev_root;
4360 struct btrfs_dev_extent *dev_extent = NULL;
4361 struct btrfs_path *path;
4367 bool retried = false;
4368 bool checked_pending_chunks = false;
4369 struct extent_buffer *l;
4370 struct btrfs_key key;
4371 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4372 u64 old_total = btrfs_super_total_bytes(super_copy);
4373 u64 old_size = btrfs_device_get_total_bytes(device);
4374 u64 diff = old_size - new_size;
4376 if (device->is_tgtdev_for_dev_replace)
4379 path = btrfs_alloc_path();
4383 path->reada = READA_FORWARD;
4387 btrfs_device_set_total_bytes(device, new_size);
4388 if (device->writeable) {
4389 device->fs_devices->total_rw_bytes -= diff;
4390 spin_lock(&root->fs_info->free_chunk_lock);
4391 root->fs_info->free_chunk_space -= diff;
4392 spin_unlock(&root->fs_info->free_chunk_lock);
4394 unlock_chunks(root);
4397 key.objectid = device->devid;
4398 key.offset = (u64)-1;
4399 key.type = BTRFS_DEV_EXTENT_KEY;
4402 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
4403 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4405 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4409 ret = btrfs_previous_item(root, path, 0, key.type);
4411 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4416 btrfs_release_path(path);
4421 slot = path->slots[0];
4422 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4424 if (key.objectid != device->devid) {
4425 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4426 btrfs_release_path(path);
4430 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4431 length = btrfs_dev_extent_length(l, dev_extent);
4433 if (key.offset + length <= new_size) {
4434 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4435 btrfs_release_path(path);
4439 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4440 btrfs_release_path(path);
4442 ret = btrfs_relocate_chunk(root, chunk_offset);
4443 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4444 if (ret && ret != -ENOSPC)
4448 } while (key.offset-- > 0);
4450 if (failed && !retried) {
4454 } else if (failed && retried) {
4459 /* Shrinking succeeded, else we would be at "done". */
4460 trans = btrfs_start_transaction(root, 0);
4461 if (IS_ERR(trans)) {
4462 ret = PTR_ERR(trans);
4469 * We checked in the above loop all device extents that were already in
4470 * the device tree. However before we have updated the device's
4471 * total_bytes to the new size, we might have had chunk allocations that
4472 * have not complete yet (new block groups attached to transaction
4473 * handles), and therefore their device extents were not yet in the
4474 * device tree and we missed them in the loop above. So if we have any
4475 * pending chunk using a device extent that overlaps the device range
4476 * that we can not use anymore, commit the current transaction and
4477 * repeat the search on the device tree - this way we guarantee we will
4478 * not have chunks using device extents that end beyond 'new_size'.
4480 if (!checked_pending_chunks) {
4481 u64 start = new_size;
4482 u64 len = old_size - new_size;
4484 if (contains_pending_extent(trans->transaction, device,
4486 unlock_chunks(root);
4487 checked_pending_chunks = true;
4490 ret = btrfs_commit_transaction(trans, root);
4497 btrfs_device_set_disk_total_bytes(device, new_size);
4498 if (list_empty(&device->resized_list))
4499 list_add_tail(&device->resized_list,
4500 &root->fs_info->fs_devices->resized_devices);
4502 WARN_ON(diff > old_total);
4503 btrfs_set_super_total_bytes(super_copy, old_total - diff);
4504 unlock_chunks(root);
4506 /* Now btrfs_update_device() will change the on-disk size. */
4507 ret = btrfs_update_device(trans, device);
4508 btrfs_end_transaction(trans, root);
4510 btrfs_free_path(path);
4513 btrfs_device_set_total_bytes(device, old_size);
4514 if (device->writeable)
4515 device->fs_devices->total_rw_bytes += diff;
4516 spin_lock(&root->fs_info->free_chunk_lock);
4517 root->fs_info->free_chunk_space += diff;
4518 spin_unlock(&root->fs_info->free_chunk_lock);
4519 unlock_chunks(root);
4524 static int btrfs_add_system_chunk(struct btrfs_root *root,
4525 struct btrfs_key *key,
4526 struct btrfs_chunk *chunk, int item_size)
4528 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4529 struct btrfs_disk_key disk_key;
4534 array_size = btrfs_super_sys_array_size(super_copy);
4535 if (array_size + item_size + sizeof(disk_key)
4536 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4537 unlock_chunks(root);
4541 ptr = super_copy->sys_chunk_array + array_size;
4542 btrfs_cpu_key_to_disk(&disk_key, key);
4543 memcpy(ptr, &disk_key, sizeof(disk_key));
4544 ptr += sizeof(disk_key);
4545 memcpy(ptr, chunk, item_size);
4546 item_size += sizeof(disk_key);
4547 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4548 unlock_chunks(root);
4554 * sort the devices in descending order by max_avail, total_avail
4556 static int btrfs_cmp_device_info(const void *a, const void *b)
4558 const struct btrfs_device_info *di_a = a;
4559 const struct btrfs_device_info *di_b = b;
4561 if (di_a->max_avail > di_b->max_avail)
4563 if (di_a->max_avail < di_b->max_avail)
4565 if (di_a->total_avail > di_b->total_avail)
4567 if (di_a->total_avail < di_b->total_avail)
4572 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
4574 /* TODO allow them to set a preferred stripe size */
4578 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4580 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4583 btrfs_set_fs_incompat(info, RAID56);
4586 #define BTRFS_MAX_DEVS(r) ((BTRFS_MAX_ITEM_SIZE(r) \
4587 - sizeof(struct btrfs_chunk)) \
4588 / sizeof(struct btrfs_stripe) + 1)
4590 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
4591 - 2 * sizeof(struct btrfs_disk_key) \
4592 - 2 * sizeof(struct btrfs_chunk)) \
4593 / sizeof(struct btrfs_stripe) + 1)
4595 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4596 struct btrfs_root *extent_root, u64 start,
4599 struct btrfs_fs_info *info = extent_root->fs_info;
4600 struct btrfs_fs_devices *fs_devices = info->fs_devices;
4601 struct list_head *cur;
4602 struct map_lookup *map = NULL;
4603 struct extent_map_tree *em_tree;
4604 struct extent_map *em;
4605 struct btrfs_device_info *devices_info = NULL;
4607 int num_stripes; /* total number of stripes to allocate */
4608 int data_stripes; /* number of stripes that count for
4610 int sub_stripes; /* sub_stripes info for map */
4611 int dev_stripes; /* stripes per dev */
4612 int devs_max; /* max devs to use */
4613 int devs_min; /* min devs needed */
4614 int devs_increment; /* ndevs has to be a multiple of this */
4615 int ncopies; /* how many copies to data has */
4617 u64 max_stripe_size;
4621 u64 raid_stripe_len = BTRFS_STRIPE_LEN;
4627 BUG_ON(!alloc_profile_is_valid(type, 0));
4629 if (list_empty(&fs_devices->alloc_list))
4632 index = __get_raid_index(type);
4634 sub_stripes = btrfs_raid_array[index].sub_stripes;
4635 dev_stripes = btrfs_raid_array[index].dev_stripes;
4636 devs_max = btrfs_raid_array[index].devs_max;
4637 devs_min = btrfs_raid_array[index].devs_min;
4638 devs_increment = btrfs_raid_array[index].devs_increment;
4639 ncopies = btrfs_raid_array[index].ncopies;
4641 if (type & BTRFS_BLOCK_GROUP_DATA) {
4642 max_stripe_size = SZ_1G;
4643 max_chunk_size = 10 * max_stripe_size;
4645 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4646 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4647 /* for larger filesystems, use larger metadata chunks */
4648 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
4649 max_stripe_size = SZ_1G;
4651 max_stripe_size = SZ_256M;
4652 max_chunk_size = max_stripe_size;
4654 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4655 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4656 max_stripe_size = SZ_32M;
4657 max_chunk_size = 2 * max_stripe_size;
4659 devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4661 btrfs_err(info, "invalid chunk type 0x%llx requested",
4666 /* we don't want a chunk larger than 10% of writeable space */
4667 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4670 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4675 cur = fs_devices->alloc_list.next;
4678 * in the first pass through the devices list, we gather information
4679 * about the available holes on each device.
4682 while (cur != &fs_devices->alloc_list) {
4683 struct btrfs_device *device;
4687 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4691 if (!device->writeable) {
4693 "BTRFS: read-only device in alloc_list\n");
4697 if (!device->in_fs_metadata ||
4698 device->is_tgtdev_for_dev_replace)
4701 if (device->total_bytes > device->bytes_used)
4702 total_avail = device->total_bytes - device->bytes_used;
4706 /* If there is no space on this device, skip it. */
4707 if (total_avail == 0)
4710 ret = find_free_dev_extent(trans, device,
4711 max_stripe_size * dev_stripes,
4712 &dev_offset, &max_avail);
4713 if (ret && ret != -ENOSPC)
4717 max_avail = max_stripe_size * dev_stripes;
4719 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4722 if (ndevs == fs_devices->rw_devices) {
4723 WARN(1, "%s: found more than %llu devices\n",
4724 __func__, fs_devices->rw_devices);
4727 devices_info[ndevs].dev_offset = dev_offset;
4728 devices_info[ndevs].max_avail = max_avail;
4729 devices_info[ndevs].total_avail = total_avail;
4730 devices_info[ndevs].dev = device;
4735 * now sort the devices by hole size / available space
4737 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4738 btrfs_cmp_device_info, NULL);
4740 /* round down to number of usable stripes */
4741 ndevs -= ndevs % devs_increment;
4743 if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4748 if (devs_max && ndevs > devs_max)
4751 * the primary goal is to maximize the number of stripes, so use as many
4752 * devices as possible, even if the stripes are not maximum sized.
4754 stripe_size = devices_info[ndevs-1].max_avail;
4755 num_stripes = ndevs * dev_stripes;
4758 * this will have to be fixed for RAID1 and RAID10 over
4761 data_stripes = num_stripes / ncopies;
4763 if (type & BTRFS_BLOCK_GROUP_RAID5) {
4764 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
4765 extent_root->stripesize);
4766 data_stripes = num_stripes - 1;
4768 if (type & BTRFS_BLOCK_GROUP_RAID6) {
4769 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
4770 extent_root->stripesize);
4771 data_stripes = num_stripes - 2;
4775 * Use the number of data stripes to figure out how big this chunk
4776 * is really going to be in terms of logical address space,
4777 * and compare that answer with the max chunk size
4779 if (stripe_size * data_stripes > max_chunk_size) {
4780 u64 mask = (1ULL << 24) - 1;
4782 stripe_size = div_u64(max_chunk_size, data_stripes);
4784 /* bump the answer up to a 16MB boundary */
4785 stripe_size = (stripe_size + mask) & ~mask;
4787 /* but don't go higher than the limits we found
4788 * while searching for free extents
4790 if (stripe_size > devices_info[ndevs-1].max_avail)
4791 stripe_size = devices_info[ndevs-1].max_avail;
4794 stripe_size = div_u64(stripe_size, dev_stripes);
4796 /* align to BTRFS_STRIPE_LEN */
4797 stripe_size = div_u64(stripe_size, raid_stripe_len);
4798 stripe_size *= raid_stripe_len;
4800 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4805 map->num_stripes = num_stripes;
4807 for (i = 0; i < ndevs; ++i) {
4808 for (j = 0; j < dev_stripes; ++j) {
4809 int s = i * dev_stripes + j;
4810 map->stripes[s].dev = devices_info[i].dev;
4811 map->stripes[s].physical = devices_info[i].dev_offset +
4815 map->sector_size = extent_root->sectorsize;
4816 map->stripe_len = raid_stripe_len;
4817 map->io_align = raid_stripe_len;
4818 map->io_width = raid_stripe_len;
4820 map->sub_stripes = sub_stripes;
4822 num_bytes = stripe_size * data_stripes;
4824 trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
4826 em = alloc_extent_map();
4832 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4833 em->map_lookup = map;
4835 em->len = num_bytes;
4836 em->block_start = 0;
4837 em->block_len = em->len;
4838 em->orig_block_len = stripe_size;
4840 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4841 write_lock(&em_tree->lock);
4842 ret = add_extent_mapping(em_tree, em, 0);
4844 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4845 atomic_inc(&em->refs);
4847 write_unlock(&em_tree->lock);
4849 free_extent_map(em);
4853 ret = btrfs_make_block_group(trans, extent_root, 0, type,
4854 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4857 goto error_del_extent;
4859 for (i = 0; i < map->num_stripes; i++) {
4860 num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4861 btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4864 spin_lock(&extent_root->fs_info->free_chunk_lock);
4865 extent_root->fs_info->free_chunk_space -= (stripe_size *
4867 spin_unlock(&extent_root->fs_info->free_chunk_lock);
4869 free_extent_map(em);
4870 check_raid56_incompat_flag(extent_root->fs_info, type);
4872 kfree(devices_info);
4876 write_lock(&em_tree->lock);
4877 remove_extent_mapping(em_tree, em);
4878 write_unlock(&em_tree->lock);
4880 /* One for our allocation */
4881 free_extent_map(em);
4882 /* One for the tree reference */
4883 free_extent_map(em);
4884 /* One for the pending_chunks list reference */
4885 free_extent_map(em);
4887 kfree(devices_info);
4891 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4892 struct btrfs_root *extent_root,
4893 u64 chunk_offset, u64 chunk_size)
4895 struct btrfs_key key;
4896 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4897 struct btrfs_device *device;
4898 struct btrfs_chunk *chunk;
4899 struct btrfs_stripe *stripe;
4900 struct extent_map_tree *em_tree;
4901 struct extent_map *em;
4902 struct map_lookup *map;
4909 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4910 read_lock(&em_tree->lock);
4911 em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size);
4912 read_unlock(&em_tree->lock);
4915 btrfs_crit(extent_root->fs_info,
4916 "unable to find logical %Lu len %Lu",
4917 chunk_offset, chunk_size);
4921 if (em->start != chunk_offset || em->len != chunk_size) {
4922 btrfs_crit(extent_root->fs_info,
4923 "found a bad mapping, wanted %Lu-%Lu, found %Lu-%Lu",
4924 chunk_offset, 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 %Lu-%Lu",
5158 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) {
5374 "found a bad mapping, wanted %Lu, found %Lu-%Lu",
5375 logical, em->start, 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) {
5394 "stripe math has gone wrong, stripe_offset=%llu, offset=%llu, start=%llu, logical=%llu, stripe_len=%llu",
5395 stripe_offset, offset, em->start, logical,
5397 free_extent_map(em);
5401 /* stripe_offset is the offset of this block in its stripe*/
5402 stripe_offset = offset - stripe_offset;
5404 /* if we're here for raid56, we need to know the stripe aligned start */
5405 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5406 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5407 raid56_full_stripe_start = offset;
5409 /* allow a write of a full stripe, but make sure we don't
5410 * allow straddling of stripes
5412 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5414 raid56_full_stripe_start *= full_stripe_len;
5417 if (op == REQ_OP_DISCARD) {
5418 /* we don't discard raid56 yet */
5419 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5423 *length = min_t(u64, em->len - offset, *length);
5424 } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5426 /* For writes to RAID[56], allow a full stripeset across all disks.
5427 For other RAID types and for RAID[56] reads, just allow a single
5428 stripe (on a single disk). */
5429 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
5430 (op == REQ_OP_WRITE)) {
5431 max_len = stripe_len * nr_data_stripes(map) -
5432 (offset - raid56_full_stripe_start);
5434 /* we limit the length of each bio to what fits in a stripe */
5435 max_len = stripe_len - stripe_offset;
5437 *length = min_t(u64, em->len - offset, max_len);
5439 *length = em->len - offset;
5442 /* This is for when we're called from btrfs_merge_bio_hook() and all
5443 it cares about is the length */
5447 btrfs_dev_replace_lock(dev_replace, 0);
5448 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5449 if (!dev_replace_is_ongoing)
5450 btrfs_dev_replace_unlock(dev_replace, 0);
5452 btrfs_dev_replace_set_lock_blocking(dev_replace);
5454 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
5455 op != REQ_OP_WRITE && op != REQ_OP_DISCARD &&
5456 op != REQ_GET_READ_MIRRORS && dev_replace->tgtdev != NULL) {
5458 * in dev-replace case, for repair case (that's the only
5459 * case where the mirror is selected explicitly when
5460 * calling btrfs_map_block), blocks left of the left cursor
5461 * can also be read from the target drive.
5462 * For REQ_GET_READ_MIRRORS, the target drive is added as
5463 * the last one to the array of stripes. For READ, it also
5464 * needs to be supported using the same mirror number.
5465 * If the requested block is not left of the left cursor,
5466 * EIO is returned. This can happen because btrfs_num_copies()
5467 * returns one more in the dev-replace case.
5469 u64 tmp_length = *length;
5470 struct btrfs_bio *tmp_bbio = NULL;
5471 int tmp_num_stripes;
5472 u64 srcdev_devid = dev_replace->srcdev->devid;
5473 int index_srcdev = 0;
5475 u64 physical_of_found = 0;
5477 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
5478 logical, &tmp_length, &tmp_bbio, 0, 0);
5480 WARN_ON(tmp_bbio != NULL);
5484 tmp_num_stripes = tmp_bbio->num_stripes;
5485 if (mirror_num > tmp_num_stripes) {
5487 * REQ_GET_READ_MIRRORS does not contain this
5488 * mirror, that means that the requested area
5489 * is not left of the left cursor
5492 btrfs_put_bbio(tmp_bbio);
5497 * process the rest of the function using the mirror_num
5498 * of the source drive. Therefore look it up first.
5499 * At the end, patch the device pointer to the one of the
5502 for (i = 0; i < tmp_num_stripes; i++) {
5503 if (tmp_bbio->stripes[i].dev->devid != srcdev_devid)
5507 * In case of DUP, in order to keep it simple, only add
5508 * the mirror with the lowest physical address
5511 physical_of_found <= tmp_bbio->stripes[i].physical)
5516 physical_of_found = tmp_bbio->stripes[i].physical;
5519 btrfs_put_bbio(tmp_bbio);
5527 mirror_num = index_srcdev + 1;
5528 patch_the_first_stripe_for_dev_replace = 1;
5529 physical_to_patch_in_first_stripe = physical_of_found;
5530 } else if (mirror_num > map->num_stripes) {
5536 stripe_nr_orig = stripe_nr;
5537 stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
5538 stripe_nr_end = div_u64(stripe_nr_end, map->stripe_len);
5539 stripe_end_offset = stripe_nr_end * map->stripe_len -
5542 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5543 if (op == REQ_OP_DISCARD)
5544 num_stripes = min_t(u64, map->num_stripes,
5545 stripe_nr_end - stripe_nr_orig);
5546 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5548 if (op != REQ_OP_WRITE && op != REQ_OP_DISCARD &&
5549 op != REQ_GET_READ_MIRRORS)
5551 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
5552 if (op == REQ_OP_WRITE || op == REQ_OP_DISCARD ||
5553 op == REQ_GET_READ_MIRRORS)
5554 num_stripes = map->num_stripes;
5555 else if (mirror_num)
5556 stripe_index = mirror_num - 1;
5558 stripe_index = find_live_mirror(fs_info, map, 0,
5560 current->pid % map->num_stripes,
5561 dev_replace_is_ongoing);
5562 mirror_num = stripe_index + 1;
5565 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
5566 if (op == REQ_OP_WRITE || op == REQ_OP_DISCARD ||
5567 op == REQ_GET_READ_MIRRORS) {
5568 num_stripes = map->num_stripes;
5569 } else if (mirror_num) {
5570 stripe_index = mirror_num - 1;
5575 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5576 u32 factor = map->num_stripes / map->sub_stripes;
5578 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5579 stripe_index *= map->sub_stripes;
5581 if (op == REQ_OP_WRITE || op == REQ_GET_READ_MIRRORS)
5582 num_stripes = map->sub_stripes;
5583 else if (op == REQ_OP_DISCARD)
5584 num_stripes = min_t(u64, map->sub_stripes *
5585 (stripe_nr_end - stripe_nr_orig),
5587 else if (mirror_num)
5588 stripe_index += mirror_num - 1;
5590 int old_stripe_index = stripe_index;
5591 stripe_index = find_live_mirror(fs_info, map,
5593 map->sub_stripes, stripe_index +
5594 current->pid % map->sub_stripes,
5595 dev_replace_is_ongoing);
5596 mirror_num = stripe_index - old_stripe_index + 1;
5599 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5600 if (need_raid_map &&
5601 (op == REQ_OP_WRITE || op == REQ_GET_READ_MIRRORS ||
5603 /* push stripe_nr back to the start of the full stripe */
5604 stripe_nr = div_u64(raid56_full_stripe_start,
5605 stripe_len * nr_data_stripes(map));
5607 /* RAID[56] write or recovery. Return all stripes */
5608 num_stripes = map->num_stripes;
5609 max_errors = nr_parity_stripes(map);
5611 *length = map->stripe_len;
5616 * Mirror #0 or #1 means the original data block.
5617 * Mirror #2 is RAID5 parity block.
5618 * Mirror #3 is RAID6 Q block.
5620 stripe_nr = div_u64_rem(stripe_nr,
5621 nr_data_stripes(map), &stripe_index);
5623 stripe_index = nr_data_stripes(map) +
5626 /* We distribute the parity blocks across stripes */
5627 div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
5629 if ((op != REQ_OP_WRITE && op != REQ_OP_DISCARD &&
5630 op != REQ_GET_READ_MIRRORS) && mirror_num <= 1)
5635 * after this, stripe_nr is the number of stripes on this
5636 * device we have to walk to find the data, and stripe_index is
5637 * the number of our device in the stripe array
5639 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5641 mirror_num = stripe_index + 1;
5643 if (stripe_index >= map->num_stripes) {
5645 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
5646 stripe_index, map->num_stripes);
5651 num_alloc_stripes = num_stripes;
5652 if (dev_replace_is_ongoing) {
5653 if (op == REQ_OP_WRITE || op == REQ_OP_DISCARD)
5654 num_alloc_stripes <<= 1;
5655 if (op == REQ_GET_READ_MIRRORS)
5656 num_alloc_stripes++;
5657 tgtdev_indexes = num_stripes;
5660 bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
5665 if (dev_replace_is_ongoing)
5666 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
5668 /* build raid_map */
5669 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK &&
5671 ((op == REQ_OP_WRITE || op == REQ_GET_READ_MIRRORS) ||
5676 bbio->raid_map = (u64 *)((void *)bbio->stripes +
5677 sizeof(struct btrfs_bio_stripe) *
5679 sizeof(int) * tgtdev_indexes);
5681 /* Work out the disk rotation on this stripe-set */
5682 div_u64_rem(stripe_nr, num_stripes, &rot);
5684 /* Fill in the logical address of each stripe */
5685 tmp = stripe_nr * nr_data_stripes(map);
5686 for (i = 0; i < nr_data_stripes(map); i++)
5687 bbio->raid_map[(i+rot) % num_stripes] =
5688 em->start + (tmp + i) * map->stripe_len;
5690 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5691 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5692 bbio->raid_map[(i+rot+1) % num_stripes] =
5696 if (op == REQ_OP_DISCARD) {
5698 u32 sub_stripes = 0;
5699 u64 stripes_per_dev = 0;
5700 u32 remaining_stripes = 0;
5701 u32 last_stripe = 0;
5704 (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
5705 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5708 sub_stripes = map->sub_stripes;
5710 factor = map->num_stripes / sub_stripes;
5711 stripes_per_dev = div_u64_rem(stripe_nr_end -
5714 &remaining_stripes);
5715 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5716 last_stripe *= sub_stripes;
5719 for (i = 0; i < num_stripes; i++) {
5720 bbio->stripes[i].physical =
5721 map->stripes[stripe_index].physical +
5722 stripe_offset + stripe_nr * map->stripe_len;
5723 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5725 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5726 BTRFS_BLOCK_GROUP_RAID10)) {
5727 bbio->stripes[i].length = stripes_per_dev *
5730 if (i / sub_stripes < remaining_stripes)
5731 bbio->stripes[i].length +=
5735 * Special for the first stripe and
5738 * |-------|...|-------|
5742 if (i < sub_stripes)
5743 bbio->stripes[i].length -=
5746 if (stripe_index >= last_stripe &&
5747 stripe_index <= (last_stripe +
5749 bbio->stripes[i].length -=
5752 if (i == sub_stripes - 1)
5755 bbio->stripes[i].length = *length;
5758 if (stripe_index == map->num_stripes) {
5759 /* This could only happen for RAID0/10 */
5765 for (i = 0; i < num_stripes; i++) {
5766 bbio->stripes[i].physical =
5767 map->stripes[stripe_index].physical +
5769 stripe_nr * map->stripe_len;
5770 bbio->stripes[i].dev =
5771 map->stripes[stripe_index].dev;
5776 if (op == REQ_OP_WRITE || op == REQ_GET_READ_MIRRORS)
5777 max_errors = btrfs_chunk_max_errors(map);
5780 sort_parity_stripes(bbio, num_stripes);
5783 if (dev_replace_is_ongoing &&
5784 (op == REQ_OP_WRITE || op == REQ_OP_DISCARD) &&
5785 dev_replace->tgtdev != NULL) {
5786 int index_where_to_add;
5787 u64 srcdev_devid = dev_replace->srcdev->devid;
5790 * duplicate the write operations while the dev replace
5791 * procedure is running. Since the copying of the old disk
5792 * to the new disk takes place at run time while the
5793 * filesystem is mounted writable, the regular write
5794 * operations to the old disk have to be duplicated to go
5795 * to the new disk as well.
5796 * Note that device->missing is handled by the caller, and
5797 * that the write to the old disk is already set up in the
5800 index_where_to_add = num_stripes;
5801 for (i = 0; i < num_stripes; i++) {
5802 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5803 /* write to new disk, too */
5804 struct btrfs_bio_stripe *new =
5805 bbio->stripes + index_where_to_add;
5806 struct btrfs_bio_stripe *old =
5809 new->physical = old->physical;
5810 new->length = old->length;
5811 new->dev = dev_replace->tgtdev;
5812 bbio->tgtdev_map[i] = index_where_to_add;
5813 index_where_to_add++;
5818 num_stripes = index_where_to_add;
5819 } else if (dev_replace_is_ongoing && (op == REQ_GET_READ_MIRRORS) &&
5820 dev_replace->tgtdev != NULL) {
5821 u64 srcdev_devid = dev_replace->srcdev->devid;
5822 int index_srcdev = 0;
5824 u64 physical_of_found = 0;
5827 * During the dev-replace procedure, the target drive can
5828 * also be used to read data in case it is needed to repair
5829 * a corrupt block elsewhere. This is possible if the
5830 * requested area is left of the left cursor. In this area,
5831 * the target drive is a full copy of the source drive.
5833 for (i = 0; i < num_stripes; i++) {
5834 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5836 * In case of DUP, in order to keep it
5837 * simple, only add the mirror with the
5838 * lowest physical address
5841 physical_of_found <=
5842 bbio->stripes[i].physical)
5846 physical_of_found = bbio->stripes[i].physical;
5850 struct btrfs_bio_stripe *tgtdev_stripe =
5851 bbio->stripes + num_stripes;
5853 tgtdev_stripe->physical = physical_of_found;
5854 tgtdev_stripe->length =
5855 bbio->stripes[index_srcdev].length;
5856 tgtdev_stripe->dev = dev_replace->tgtdev;
5857 bbio->tgtdev_map[index_srcdev] = num_stripes;
5865 bbio->map_type = map->type;
5866 bbio->num_stripes = num_stripes;
5867 bbio->max_errors = max_errors;
5868 bbio->mirror_num = mirror_num;
5869 bbio->num_tgtdevs = tgtdev_indexes;
5872 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5873 * mirror_num == num_stripes + 1 && dev_replace target drive is
5874 * available as a mirror
5876 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5877 WARN_ON(num_stripes > 1);
5878 bbio->stripes[0].dev = dev_replace->tgtdev;
5879 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5880 bbio->mirror_num = map->num_stripes + 1;
5883 if (dev_replace_is_ongoing) {
5884 btrfs_dev_replace_clear_lock_blocking(dev_replace);
5885 btrfs_dev_replace_unlock(dev_replace, 0);
5887 free_extent_map(em);
5891 int btrfs_map_block(struct btrfs_fs_info *fs_info, int op,
5892 u64 logical, u64 *length,
5893 struct btrfs_bio **bbio_ret, int mirror_num)
5895 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret,
5899 /* For Scrub/replace */
5900 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, int op,
5901 u64 logical, u64 *length,
5902 struct btrfs_bio **bbio_ret, int mirror_num,
5905 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret,
5906 mirror_num, need_raid_map);
5909 int btrfs_rmap_block(struct btrfs_fs_info *fs_info,
5910 u64 chunk_start, u64 physical, u64 devid,
5911 u64 **logical, int *naddrs, int *stripe_len)
5913 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
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 btrfs_err(fs_info, "couldn't find em for chunk %Lu",
5934 if (em->start != chunk_start) {
5935 btrfs_err(fs_info, "bad chunk start, em=%Lu, wanted=%Lu",
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 & REQ_PREFLUSH)
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 (op_is_sync(bio->bi_opf))
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 btrfs_debug(fs_info,
6141 "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
6142 bio_op(bio), bio->bi_opf,
6143 (u64)bio->bi_iter.bi_sector,
6144 (u_long)dev->bdev->bd_dev, name->str, dev->devid,
6145 bio->bi_iter.bi_size);
6149 bio->bi_bdev = dev->bdev;
6151 btrfs_bio_counter_inc_noblocked(root->fs_info);
6154 btrfs_schedule_bio(root, dev, bio);
6156 btrfsic_submit_bio(bio);
6159 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6161 atomic_inc(&bbio->error);
6162 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6163 /* Should be the original bio. */
6164 WARN_ON(bio != bbio->orig_bio);
6166 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6167 bio->bi_iter.bi_sector = logical >> 9;
6168 bio->bi_error = -EIO;
6169 btrfs_end_bbio(bbio, bio);
6173 int btrfs_map_bio(struct btrfs_root *root, struct bio *bio,
6174 int mirror_num, int async_submit)
6176 struct btrfs_device *dev;
6177 struct bio *first_bio = bio;
6178 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
6184 struct btrfs_bio *bbio = NULL;
6186 length = bio->bi_iter.bi_size;
6187 map_length = length;
6189 btrfs_bio_counter_inc_blocked(root->fs_info);
6190 ret = __btrfs_map_block(root->fs_info, bio_op(bio), logical,
6191 &map_length, &bbio, mirror_num, 1);
6193 btrfs_bio_counter_dec(root->fs_info);
6197 total_devs = bbio->num_stripes;
6198 bbio->orig_bio = first_bio;
6199 bbio->private = first_bio->bi_private;
6200 bbio->end_io = first_bio->bi_end_io;
6201 bbio->fs_info = root->fs_info;
6202 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6204 if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6205 ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) {
6206 /* In this case, map_length has been set to the length of
6207 a single stripe; not the whole write */
6208 if (bio_op(bio) == REQ_OP_WRITE) {
6209 ret = raid56_parity_write(root, bio, bbio, map_length);
6211 ret = raid56_parity_recover(root, bio, bbio, map_length,
6215 btrfs_bio_counter_dec(root->fs_info);
6219 if (map_length < length) {
6220 btrfs_crit(root->fs_info,
6221 "mapping failed logical %llu bio len %llu len %llu",
6222 logical, length, map_length);
6226 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6227 dev = bbio->stripes[dev_nr].dev;
6228 if (!dev || !dev->bdev ||
6229 (bio_op(bio) == REQ_OP_WRITE && !dev->writeable)) {
6230 bbio_error(bbio, first_bio, logical);
6234 if (dev_nr < total_devs - 1) {
6235 bio = btrfs_bio_clone(first_bio, GFP_NOFS);
6236 BUG_ON(!bio); /* -ENOMEM */
6240 submit_stripe_bio(root, bbio, bio,
6241 bbio->stripes[dev_nr].physical, dev_nr,
6244 btrfs_bio_counter_dec(root->fs_info);
6248 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
6251 struct btrfs_device *device;
6252 struct btrfs_fs_devices *cur_devices;
6254 cur_devices = fs_info->fs_devices;
6255 while (cur_devices) {
6257 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
6258 device = __find_device(&cur_devices->devices,
6263 cur_devices = cur_devices->seed;
6268 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
6269 struct btrfs_fs_devices *fs_devices,
6270 u64 devid, u8 *dev_uuid)
6272 struct btrfs_device *device;
6274 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6278 list_add(&device->dev_list, &fs_devices->devices);
6279 device->fs_devices = fs_devices;
6280 fs_devices->num_devices++;
6282 device->missing = 1;
6283 fs_devices->missing_devices++;
6289 * btrfs_alloc_device - allocate struct btrfs_device
6290 * @fs_info: used only for generating a new devid, can be NULL if
6291 * devid is provided (i.e. @devid != NULL).
6292 * @devid: a pointer to devid for this device. If NULL a new devid
6294 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6297 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6298 * on error. Returned struct is not linked onto any lists and can be
6299 * destroyed with kfree() right away.
6301 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6305 struct btrfs_device *dev;
6308 if (WARN_ON(!devid && !fs_info))
6309 return ERR_PTR(-EINVAL);
6311 dev = __alloc_device();
6320 ret = find_next_devid(fs_info, &tmp);
6323 return ERR_PTR(ret);
6329 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6331 generate_random_uuid(dev->uuid);
6333 btrfs_init_work(&dev->work, btrfs_submit_helper,
6334 pending_bios_fn, NULL, NULL);
6339 /* Return -EIO if any error, otherwise return 0. */
6340 static int btrfs_check_chunk_valid(struct btrfs_root *root,
6341 struct extent_buffer *leaf,
6342 struct btrfs_chunk *chunk, u64 logical)
6350 length = btrfs_chunk_length(leaf, chunk);
6351 stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6352 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6353 sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6354 type = btrfs_chunk_type(leaf, chunk);
6357 btrfs_err(root->fs_info, "invalid chunk num_stripes: %u",
6361 if (!IS_ALIGNED(logical, root->sectorsize)) {
6362 btrfs_err(root->fs_info,
6363 "invalid chunk logical %llu", logical);
6366 if (btrfs_chunk_sector_size(leaf, chunk) != root->sectorsize) {
6367 btrfs_err(root->fs_info, "invalid chunk sectorsize %u",
6368 btrfs_chunk_sector_size(leaf, chunk));
6371 if (!length || !IS_ALIGNED(length, root->sectorsize)) {
6372 btrfs_err(root->fs_info,
6373 "invalid chunk length %llu", length);
6376 if (!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN) {
6377 btrfs_err(root->fs_info, "invalid chunk stripe length: %llu",
6381 if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6383 btrfs_err(root->fs_info, "unrecognized chunk type: %llu",
6384 ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
6385 BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6386 btrfs_chunk_type(leaf, chunk));
6389 if ((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes != 2) ||
6390 (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) ||
6391 (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) ||
6392 (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) ||
6393 (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) ||
6394 ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
6395 num_stripes != 1)) {
6396 btrfs_err(root->fs_info,
6397 "invalid num_stripes:sub_stripes %u:%u for profile %llu",
6398 num_stripes, sub_stripes,
6399 type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
6406 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
6407 struct extent_buffer *leaf,
6408 struct btrfs_chunk *chunk)
6410 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
6411 struct map_lookup *map;
6412 struct extent_map *em;
6417 u8 uuid[BTRFS_UUID_SIZE];
6422 logical = key->offset;
6423 length = btrfs_chunk_length(leaf, chunk);
6424 stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6425 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6427 ret = btrfs_check_chunk_valid(root, leaf, chunk, logical);
6431 read_lock(&map_tree->map_tree.lock);
6432 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
6433 read_unlock(&map_tree->map_tree.lock);
6435 /* already mapped? */
6436 if (em && em->start <= logical && em->start + em->len > logical) {
6437 free_extent_map(em);
6440 free_extent_map(em);
6443 em = alloc_extent_map();
6446 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6448 free_extent_map(em);
6452 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6453 em->map_lookup = map;
6454 em->start = logical;
6457 em->block_start = 0;
6458 em->block_len = em->len;
6460 map->num_stripes = num_stripes;
6461 map->io_width = btrfs_chunk_io_width(leaf, chunk);
6462 map->io_align = btrfs_chunk_io_align(leaf, chunk);
6463 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
6464 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6465 map->type = btrfs_chunk_type(leaf, chunk);
6466 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6467 for (i = 0; i < num_stripes; i++) {
6468 map->stripes[i].physical =
6469 btrfs_stripe_offset_nr(leaf, chunk, i);
6470 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6471 read_extent_buffer(leaf, uuid, (unsigned long)
6472 btrfs_stripe_dev_uuid_nr(chunk, i),
6474 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
6476 if (!map->stripes[i].dev &&
6477 !btrfs_test_opt(root->fs_info, DEGRADED)) {
6478 free_extent_map(em);
6481 if (!map->stripes[i].dev) {
6482 map->stripes[i].dev =
6483 add_missing_dev(root, root->fs_info->fs_devices,
6485 if (!map->stripes[i].dev) {
6486 free_extent_map(em);
6489 btrfs_warn(root->fs_info,
6490 "devid %llu uuid %pU is missing",
6493 map->stripes[i].dev->in_fs_metadata = 1;
6496 write_lock(&map_tree->map_tree.lock);
6497 ret = add_extent_mapping(&map_tree->map_tree, em, 0);
6498 write_unlock(&map_tree->map_tree.lock);
6499 BUG_ON(ret); /* Tree corruption */
6500 free_extent_map(em);
6505 static void fill_device_from_item(struct extent_buffer *leaf,
6506 struct btrfs_dev_item *dev_item,
6507 struct btrfs_device *device)
6511 device->devid = btrfs_device_id(leaf, dev_item);
6512 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6513 device->total_bytes = device->disk_total_bytes;
6514 device->commit_total_bytes = device->disk_total_bytes;
6515 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6516 device->commit_bytes_used = device->bytes_used;
6517 device->type = btrfs_device_type(leaf, dev_item);
6518 device->io_align = btrfs_device_io_align(leaf, dev_item);
6519 device->io_width = btrfs_device_io_width(leaf, dev_item);
6520 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6521 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6522 device->is_tgtdev_for_dev_replace = 0;
6524 ptr = btrfs_device_uuid(dev_item);
6525 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6528 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_root *root,
6531 struct btrfs_fs_devices *fs_devices;
6534 BUG_ON(!mutex_is_locked(&uuid_mutex));
6536 fs_devices = root->fs_info->fs_devices->seed;
6537 while (fs_devices) {
6538 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE))
6541 fs_devices = fs_devices->seed;
6544 fs_devices = find_fsid(fsid);
6546 if (!btrfs_test_opt(root->fs_info, DEGRADED))
6547 return ERR_PTR(-ENOENT);
6549 fs_devices = alloc_fs_devices(fsid);
6550 if (IS_ERR(fs_devices))
6553 fs_devices->seeding = 1;
6554 fs_devices->opened = 1;
6558 fs_devices = clone_fs_devices(fs_devices);
6559 if (IS_ERR(fs_devices))
6562 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
6563 root->fs_info->bdev_holder);
6565 free_fs_devices(fs_devices);
6566 fs_devices = ERR_PTR(ret);
6570 if (!fs_devices->seeding) {
6571 __btrfs_close_devices(fs_devices);
6572 free_fs_devices(fs_devices);
6573 fs_devices = ERR_PTR(-EINVAL);
6577 fs_devices->seed = root->fs_info->fs_devices->seed;
6578 root->fs_info->fs_devices->seed = fs_devices;
6583 static int read_one_dev(struct btrfs_root *root,
6584 struct extent_buffer *leaf,
6585 struct btrfs_dev_item *dev_item)
6587 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6588 struct btrfs_device *device;
6591 u8 fs_uuid[BTRFS_UUID_SIZE];
6592 u8 dev_uuid[BTRFS_UUID_SIZE];
6594 devid = btrfs_device_id(leaf, dev_item);
6595 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6597 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6600 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
6601 fs_devices = open_seed_devices(root, fs_uuid);
6602 if (IS_ERR(fs_devices))
6603 return PTR_ERR(fs_devices);
6606 device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
6608 if (!btrfs_test_opt(root->fs_info, DEGRADED))
6611 device = add_missing_dev(root, fs_devices, devid, dev_uuid);
6614 btrfs_warn(root->fs_info, "devid %llu uuid %pU missing",
6617 if (!device->bdev && !btrfs_test_opt(root->fs_info, DEGRADED))
6620 if(!device->bdev && !device->missing) {
6622 * this happens when a device that was properly setup
6623 * in the device info lists suddenly goes bad.
6624 * device->bdev is NULL, and so we have to set
6625 * device->missing to one here
6627 device->fs_devices->missing_devices++;
6628 device->missing = 1;
6631 /* Move the device to its own fs_devices */
6632 if (device->fs_devices != fs_devices) {
6633 ASSERT(device->missing);
6635 list_move(&device->dev_list, &fs_devices->devices);
6636 device->fs_devices->num_devices--;
6637 fs_devices->num_devices++;
6639 device->fs_devices->missing_devices--;
6640 fs_devices->missing_devices++;
6642 device->fs_devices = fs_devices;
6646 if (device->fs_devices != root->fs_info->fs_devices) {
6647 BUG_ON(device->writeable);
6648 if (device->generation !=
6649 btrfs_device_generation(leaf, dev_item))
6653 fill_device_from_item(leaf, dev_item, device);
6654 device->in_fs_metadata = 1;
6655 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
6656 device->fs_devices->total_rw_bytes += device->total_bytes;
6657 spin_lock(&root->fs_info->free_chunk_lock);
6658 root->fs_info->free_chunk_space += device->total_bytes -
6660 spin_unlock(&root->fs_info->free_chunk_lock);
6666 int btrfs_read_sys_array(struct btrfs_root *root)
6668 struct btrfs_fs_info *fs_info = root->fs_info;
6669 struct btrfs_super_block *super_copy = fs_info->super_copy;
6670 struct extent_buffer *sb;
6671 struct btrfs_disk_key *disk_key;
6672 struct btrfs_chunk *chunk;
6674 unsigned long sb_array_offset;
6681 struct btrfs_key key;
6683 ASSERT(BTRFS_SUPER_INFO_SIZE <= root->nodesize);
6685 * This will create extent buffer of nodesize, superblock size is
6686 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6687 * overallocate but we can keep it as-is, only the first page is used.
6689 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET);
6692 set_extent_buffer_uptodate(sb);
6693 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6695 * The sb extent buffer is artificial and just used to read the system array.
6696 * set_extent_buffer_uptodate() call does not properly mark all it's
6697 * pages up-to-date when the page is larger: extent does not cover the
6698 * whole page and consequently check_page_uptodate does not find all
6699 * the page's extents up-to-date (the hole beyond sb),
6700 * write_extent_buffer then triggers a WARN_ON.
6702 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6703 * but sb spans only this function. Add an explicit SetPageUptodate call
6704 * to silence the warning eg. on PowerPC 64.
6706 if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
6707 SetPageUptodate(sb->pages[0]);
6709 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6710 array_size = btrfs_super_sys_array_size(super_copy);
6712 array_ptr = super_copy->sys_chunk_array;
6713 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6716 while (cur_offset < array_size) {
6717 disk_key = (struct btrfs_disk_key *)array_ptr;
6718 len = sizeof(*disk_key);
6719 if (cur_offset + len > array_size)
6720 goto out_short_read;
6722 btrfs_disk_key_to_cpu(&key, disk_key);
6725 sb_array_offset += len;
6728 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6729 chunk = (struct btrfs_chunk *)sb_array_offset;
6731 * At least one btrfs_chunk with one stripe must be
6732 * present, exact stripe count check comes afterwards
6734 len = btrfs_chunk_item_size(1);
6735 if (cur_offset + len > array_size)
6736 goto out_short_read;
6738 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6741 "invalid number of stripes %u in sys_array at offset %u",
6742 num_stripes, cur_offset);
6747 type = btrfs_chunk_type(sb, chunk);
6748 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
6750 "invalid chunk type %llu in sys_array at offset %u",
6756 len = btrfs_chunk_item_size(num_stripes);
6757 if (cur_offset + len > array_size)
6758 goto out_short_read;
6760 ret = read_one_chunk(root, &key, sb, chunk);
6765 "unexpected item type %u in sys_array at offset %u",
6766 (u32)key.type, cur_offset);
6771 sb_array_offset += len;
6774 clear_extent_buffer_uptodate(sb);
6775 free_extent_buffer_stale(sb);
6779 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
6781 clear_extent_buffer_uptodate(sb);
6782 free_extent_buffer_stale(sb);
6786 int btrfs_read_chunk_tree(struct btrfs_root *root)
6788 struct btrfs_path *path;
6789 struct extent_buffer *leaf;
6790 struct btrfs_key key;
6791 struct btrfs_key found_key;
6796 root = root->fs_info->chunk_root;
6798 path = btrfs_alloc_path();
6802 mutex_lock(&uuid_mutex);
6806 * Read all device items, and then all the chunk items. All
6807 * device items are found before any chunk item (their object id
6808 * is smaller than the lowest possible object id for a chunk
6809 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6811 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6814 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6818 leaf = path->nodes[0];
6819 slot = path->slots[0];
6820 if (slot >= btrfs_header_nritems(leaf)) {
6821 ret = btrfs_next_leaf(root, path);
6828 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6829 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6830 struct btrfs_dev_item *dev_item;
6831 dev_item = btrfs_item_ptr(leaf, slot,
6832 struct btrfs_dev_item);
6833 ret = read_one_dev(root, leaf, dev_item);
6837 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6838 struct btrfs_chunk *chunk;
6839 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6840 ret = read_one_chunk(root, &found_key, leaf, chunk);
6848 * After loading chunk tree, we've got all device information,
6849 * do another round of validation checks.
6851 if (total_dev != root->fs_info->fs_devices->total_devices) {
6852 btrfs_err(root->fs_info,
6853 "super_num_devices %llu mismatch with num_devices %llu found here",
6854 btrfs_super_num_devices(root->fs_info->super_copy),
6859 if (btrfs_super_total_bytes(root->fs_info->super_copy) <
6860 root->fs_info->fs_devices->total_rw_bytes) {
6861 btrfs_err(root->fs_info,
6862 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
6863 btrfs_super_total_bytes(root->fs_info->super_copy),
6864 root->fs_info->fs_devices->total_rw_bytes);
6870 unlock_chunks(root);
6871 mutex_unlock(&uuid_mutex);
6873 btrfs_free_path(path);
6877 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6879 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6880 struct btrfs_device *device;
6882 while (fs_devices) {
6883 mutex_lock(&fs_devices->device_list_mutex);
6884 list_for_each_entry(device, &fs_devices->devices, dev_list)
6885 device->dev_root = fs_info->dev_root;
6886 mutex_unlock(&fs_devices->device_list_mutex);
6888 fs_devices = fs_devices->seed;
6892 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6896 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6897 btrfs_dev_stat_reset(dev, i);
6900 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6902 struct btrfs_key key;
6903 struct btrfs_key found_key;
6904 struct btrfs_root *dev_root = fs_info->dev_root;
6905 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6906 struct extent_buffer *eb;
6909 struct btrfs_device *device;
6910 struct btrfs_path *path = NULL;
6913 path = btrfs_alloc_path();
6919 mutex_lock(&fs_devices->device_list_mutex);
6920 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6922 struct btrfs_dev_stats_item *ptr;
6924 key.objectid = BTRFS_DEV_STATS_OBJECTID;
6925 key.type = BTRFS_PERSISTENT_ITEM_KEY;
6926 key.offset = device->devid;
6927 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6929 __btrfs_reset_dev_stats(device);
6930 device->dev_stats_valid = 1;
6931 btrfs_release_path(path);
6934 slot = path->slots[0];
6935 eb = path->nodes[0];
6936 btrfs_item_key_to_cpu(eb, &found_key, slot);
6937 item_size = btrfs_item_size_nr(eb, slot);
6939 ptr = btrfs_item_ptr(eb, slot,
6940 struct btrfs_dev_stats_item);
6942 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6943 if (item_size >= (1 + i) * sizeof(__le64))
6944 btrfs_dev_stat_set(device, i,
6945 btrfs_dev_stats_value(eb, ptr, i));
6947 btrfs_dev_stat_reset(device, i);
6950 device->dev_stats_valid = 1;
6951 btrfs_dev_stat_print_on_load(device);
6952 btrfs_release_path(path);
6954 mutex_unlock(&fs_devices->device_list_mutex);
6957 btrfs_free_path(path);
6958 return ret < 0 ? ret : 0;
6961 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
6962 struct btrfs_root *dev_root,
6963 struct btrfs_device *device)
6965 struct btrfs_path *path;
6966 struct btrfs_key key;
6967 struct extent_buffer *eb;
6968 struct btrfs_dev_stats_item *ptr;
6972 key.objectid = BTRFS_DEV_STATS_OBJECTID;
6973 key.type = BTRFS_PERSISTENT_ITEM_KEY;
6974 key.offset = device->devid;
6976 path = btrfs_alloc_path();
6978 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
6980 btrfs_warn_in_rcu(dev_root->fs_info,
6981 "error %d while searching for dev_stats item for device %s",
6982 ret, rcu_str_deref(device->name));
6987 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
6988 /* need to delete old one and insert a new one */
6989 ret = btrfs_del_item(trans, dev_root, path);
6991 btrfs_warn_in_rcu(dev_root->fs_info,
6992 "delete too small dev_stats item for device %s failed %d",
6993 rcu_str_deref(device->name), ret);
7000 /* need to insert a new item */
7001 btrfs_release_path(path);
7002 ret = btrfs_insert_empty_item(trans, dev_root, path,
7003 &key, sizeof(*ptr));
7005 btrfs_warn_in_rcu(dev_root->fs_info,
7006 "insert dev_stats item for device %s failed %d",
7007 rcu_str_deref(device->name), ret);
7012 eb = path->nodes[0];
7013 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7014 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7015 btrfs_set_dev_stats_value(eb, ptr, i,
7016 btrfs_dev_stat_read(device, i));
7017 btrfs_mark_buffer_dirty(eb);
7020 btrfs_free_path(path);
7025 * called from commit_transaction. Writes all changed device stats to disk.
7027 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
7028 struct btrfs_fs_info *fs_info)
7030 struct btrfs_root *dev_root = fs_info->dev_root;
7031 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7032 struct btrfs_device *device;
7036 mutex_lock(&fs_devices->device_list_mutex);
7037 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7038 if (!device->dev_stats_valid || !btrfs_dev_stats_dirty(device))
7041 stats_cnt = atomic_read(&device->dev_stats_ccnt);
7042 ret = update_dev_stat_item(trans, dev_root, device);
7044 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7046 mutex_unlock(&fs_devices->device_list_mutex);
7051 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7053 btrfs_dev_stat_inc(dev, index);
7054 btrfs_dev_stat_print_on_error(dev);
7057 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
7059 if (!dev->dev_stats_valid)
7061 btrfs_err_rl_in_rcu(dev->dev_root->fs_info,
7062 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7063 rcu_str_deref(dev->name),
7064 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7065 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7066 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7067 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7068 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7071 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7075 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7076 if (btrfs_dev_stat_read(dev, i) != 0)
7078 if (i == BTRFS_DEV_STAT_VALUES_MAX)
7079 return; /* all values == 0, suppress message */
7081 btrfs_info_in_rcu(dev->dev_root->fs_info,
7082 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7083 rcu_str_deref(dev->name),
7084 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7085 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7086 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7087 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7088 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7091 int btrfs_get_dev_stats(struct btrfs_root *root,
7092 struct btrfs_ioctl_get_dev_stats *stats)
7094 struct btrfs_device *dev;
7095 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7098 mutex_lock(&fs_devices->device_list_mutex);
7099 dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
7100 mutex_unlock(&fs_devices->device_list_mutex);
7103 btrfs_warn(root->fs_info,
7104 "get dev_stats failed, device not found");
7106 } else if (!dev->dev_stats_valid) {
7107 btrfs_warn(root->fs_info,
7108 "get dev_stats failed, not yet valid");
7110 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7111 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7112 if (stats->nr_items > i)
7114 btrfs_dev_stat_read_and_reset(dev, i);
7116 btrfs_dev_stat_reset(dev, i);
7119 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7120 if (stats->nr_items > i)
7121 stats->values[i] = btrfs_dev_stat_read(dev, i);
7123 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7124 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7128 void btrfs_scratch_superblocks(struct block_device *bdev, char *device_path)
7130 struct buffer_head *bh;
7131 struct btrfs_super_block *disk_super;
7137 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX;
7140 if (btrfs_read_dev_one_super(bdev, copy_num, &bh))
7143 disk_super = (struct btrfs_super_block *)bh->b_data;
7145 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
7146 set_buffer_dirty(bh);
7147 sync_dirty_buffer(bh);
7151 /* Notify udev that device has changed */
7152 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
7154 /* Update ctime/mtime for device path for libblkid */
7155 update_dev_time(device_path);
7159 * Update the size of all devices, which is used for writing out the
7162 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
7164 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7165 struct btrfs_device *curr, *next;
7167 if (list_empty(&fs_devices->resized_devices))
7170 mutex_lock(&fs_devices->device_list_mutex);
7171 lock_chunks(fs_info->dev_root);
7172 list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
7174 list_del_init(&curr->resized_list);
7175 curr->commit_total_bytes = curr->disk_total_bytes;
7177 unlock_chunks(fs_info->dev_root);
7178 mutex_unlock(&fs_devices->device_list_mutex);
7181 /* Must be invoked during the transaction commit */
7182 void btrfs_update_commit_device_bytes_used(struct btrfs_root *root,
7183 struct btrfs_transaction *transaction)
7185 struct extent_map *em;
7186 struct map_lookup *map;
7187 struct btrfs_device *dev;
7190 if (list_empty(&transaction->pending_chunks))
7193 /* In order to kick the device replace finish process */
7195 list_for_each_entry(em, &transaction->pending_chunks, list) {
7196 map = em->map_lookup;
7198 for (i = 0; i < map->num_stripes; i++) {
7199 dev = map->stripes[i].dev;
7200 dev->commit_bytes_used = dev->bytes_used;
7203 unlock_chunks(root);
7206 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
7208 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7209 while (fs_devices) {
7210 fs_devices->fs_info = fs_info;
7211 fs_devices = fs_devices->seed;
7215 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
7217 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7218 while (fs_devices) {
7219 fs_devices->fs_info = NULL;
7220 fs_devices = fs_devices->seed;