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;
2910 struct btrfs_block_group_cache *block_group;
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);
2941 * step two, flag the chunk as removed and let
2942 * btrfs_delete_unused_bgs() remove it.
2944 block_group = btrfs_lookup_block_group(root->fs_info, chunk_offset);
2945 spin_lock(&block_group->lock);
2946 block_group->removed = 1;
2947 spin_unlock(&block_group->lock);
2948 btrfs_put_block_group(block_group);
2953 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2955 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2956 struct btrfs_path *path;
2957 struct extent_buffer *leaf;
2958 struct btrfs_chunk *chunk;
2959 struct btrfs_key key;
2960 struct btrfs_key found_key;
2962 bool retried = false;
2966 path = btrfs_alloc_path();
2971 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2972 key.offset = (u64)-1;
2973 key.type = BTRFS_CHUNK_ITEM_KEY;
2976 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
2977 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2979 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2982 BUG_ON(ret == 0); /* Corruption */
2984 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2987 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2993 leaf = path->nodes[0];
2994 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2996 chunk = btrfs_item_ptr(leaf, path->slots[0],
2997 struct btrfs_chunk);
2998 chunk_type = btrfs_chunk_type(leaf, chunk);
2999 btrfs_release_path(path);
3001 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3002 ret = btrfs_relocate_chunk(chunk_root,
3009 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
3011 if (found_key.offset == 0)
3013 key.offset = found_key.offset - 1;
3016 if (failed && !retried) {
3020 } else if (WARN_ON(failed && retried)) {
3024 btrfs_free_path(path);
3028 static int insert_balance_item(struct btrfs_root *root,
3029 struct btrfs_balance_control *bctl)
3031 struct btrfs_trans_handle *trans;
3032 struct btrfs_balance_item *item;
3033 struct btrfs_disk_balance_args disk_bargs;
3034 struct btrfs_path *path;
3035 struct extent_buffer *leaf;
3036 struct btrfs_key key;
3039 path = btrfs_alloc_path();
3043 trans = btrfs_start_transaction(root, 0);
3044 if (IS_ERR(trans)) {
3045 btrfs_free_path(path);
3046 return PTR_ERR(trans);
3049 key.objectid = BTRFS_BALANCE_OBJECTID;
3050 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3053 ret = btrfs_insert_empty_item(trans, root, path, &key,
3058 leaf = path->nodes[0];
3059 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3061 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
3063 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3064 btrfs_set_balance_data(leaf, item, &disk_bargs);
3065 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3066 btrfs_set_balance_meta(leaf, item, &disk_bargs);
3067 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3068 btrfs_set_balance_sys(leaf, item, &disk_bargs);
3070 btrfs_set_balance_flags(leaf, item, bctl->flags);
3072 btrfs_mark_buffer_dirty(leaf);
3074 btrfs_free_path(path);
3075 err = btrfs_commit_transaction(trans, root);
3081 static int del_balance_item(struct btrfs_root *root)
3083 struct btrfs_trans_handle *trans;
3084 struct btrfs_path *path;
3085 struct btrfs_key key;
3088 path = btrfs_alloc_path();
3092 trans = btrfs_start_transaction(root, 0);
3093 if (IS_ERR(trans)) {
3094 btrfs_free_path(path);
3095 return PTR_ERR(trans);
3098 key.objectid = BTRFS_BALANCE_OBJECTID;
3099 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3102 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3110 ret = btrfs_del_item(trans, root, path);
3112 btrfs_free_path(path);
3113 err = btrfs_commit_transaction(trans, root);
3120 * This is a heuristic used to reduce the number of chunks balanced on
3121 * resume after balance was interrupted.
3123 static void update_balance_args(struct btrfs_balance_control *bctl)
3126 * Turn on soft mode for chunk types that were being converted.
3128 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3129 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3130 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3131 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3132 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3133 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3136 * Turn on usage filter if is not already used. The idea is
3137 * that chunks that we have already balanced should be
3138 * reasonably full. Don't do it for chunks that are being
3139 * converted - that will keep us from relocating unconverted
3140 * (albeit full) chunks.
3142 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3143 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3144 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3145 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3146 bctl->data.usage = 90;
3148 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3149 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3150 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3151 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3152 bctl->sys.usage = 90;
3154 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3155 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3156 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3157 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3158 bctl->meta.usage = 90;
3163 * Should be called with both balance and volume mutexes held to
3164 * serialize other volume operations (add_dev/rm_dev/resize) with
3165 * restriper. Same goes for unset_balance_control.
3167 static void set_balance_control(struct btrfs_balance_control *bctl)
3169 struct btrfs_fs_info *fs_info = bctl->fs_info;
3171 BUG_ON(fs_info->balance_ctl);
3173 spin_lock(&fs_info->balance_lock);
3174 fs_info->balance_ctl = bctl;
3175 spin_unlock(&fs_info->balance_lock);
3178 static void unset_balance_control(struct btrfs_fs_info *fs_info)
3180 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3182 BUG_ON(!fs_info->balance_ctl);
3184 spin_lock(&fs_info->balance_lock);
3185 fs_info->balance_ctl = NULL;
3186 spin_unlock(&fs_info->balance_lock);
3192 * Balance filters. Return 1 if chunk should be filtered out
3193 * (should not be balanced).
3195 static int chunk_profiles_filter(u64 chunk_type,
3196 struct btrfs_balance_args *bargs)
3198 chunk_type = chunk_to_extended(chunk_type) &
3199 BTRFS_EXTENDED_PROFILE_MASK;
3201 if (bargs->profiles & chunk_type)
3207 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3208 struct btrfs_balance_args *bargs)
3210 struct btrfs_block_group_cache *cache;
3212 u64 user_thresh_min;
3213 u64 user_thresh_max;
3216 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3217 chunk_used = btrfs_block_group_used(&cache->item);
3219 if (bargs->usage_min == 0)
3220 user_thresh_min = 0;
3222 user_thresh_min = div_factor_fine(cache->key.offset,
3225 if (bargs->usage_max == 0)
3226 user_thresh_max = 1;
3227 else if (bargs->usage_max > 100)
3228 user_thresh_max = cache->key.offset;
3230 user_thresh_max = div_factor_fine(cache->key.offset,
3233 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3236 btrfs_put_block_group(cache);
3240 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3241 u64 chunk_offset, struct btrfs_balance_args *bargs)
3243 struct btrfs_block_group_cache *cache;
3244 u64 chunk_used, user_thresh;
3247 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3248 chunk_used = btrfs_block_group_used(&cache->item);
3250 if (bargs->usage_min == 0)
3252 else if (bargs->usage > 100)
3253 user_thresh = cache->key.offset;
3255 user_thresh = div_factor_fine(cache->key.offset,
3258 if (chunk_used < user_thresh)
3261 btrfs_put_block_group(cache);
3265 static int chunk_devid_filter(struct extent_buffer *leaf,
3266 struct btrfs_chunk *chunk,
3267 struct btrfs_balance_args *bargs)
3269 struct btrfs_stripe *stripe;
3270 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3273 for (i = 0; i < num_stripes; i++) {
3274 stripe = btrfs_stripe_nr(chunk, i);
3275 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3282 /* [pstart, pend) */
3283 static int chunk_drange_filter(struct extent_buffer *leaf,
3284 struct btrfs_chunk *chunk,
3286 struct btrfs_balance_args *bargs)
3288 struct btrfs_stripe *stripe;
3289 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3295 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3298 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3299 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3300 factor = num_stripes / 2;
3301 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3302 factor = num_stripes - 1;
3303 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3304 factor = num_stripes - 2;
3306 factor = num_stripes;
3309 for (i = 0; i < num_stripes; i++) {
3310 stripe = btrfs_stripe_nr(chunk, i);
3311 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3314 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3315 stripe_length = btrfs_chunk_length(leaf, chunk);
3316 stripe_length = div_u64(stripe_length, factor);
3318 if (stripe_offset < bargs->pend &&
3319 stripe_offset + stripe_length > bargs->pstart)
3326 /* [vstart, vend) */
3327 static int chunk_vrange_filter(struct extent_buffer *leaf,
3328 struct btrfs_chunk *chunk,
3330 struct btrfs_balance_args *bargs)
3332 if (chunk_offset < bargs->vend &&
3333 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3334 /* at least part of the chunk is inside this vrange */
3340 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3341 struct btrfs_chunk *chunk,
3342 struct btrfs_balance_args *bargs)
3344 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3346 if (bargs->stripes_min <= num_stripes
3347 && num_stripes <= bargs->stripes_max)
3353 static int chunk_soft_convert_filter(u64 chunk_type,
3354 struct btrfs_balance_args *bargs)
3356 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3359 chunk_type = chunk_to_extended(chunk_type) &
3360 BTRFS_EXTENDED_PROFILE_MASK;
3362 if (bargs->target == chunk_type)
3368 static int should_balance_chunk(struct btrfs_root *root,
3369 struct extent_buffer *leaf,
3370 struct btrfs_chunk *chunk, u64 chunk_offset)
3372 struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
3373 struct btrfs_balance_args *bargs = NULL;
3374 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3377 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3378 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3382 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3383 bargs = &bctl->data;
3384 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3386 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3387 bargs = &bctl->meta;
3389 /* profiles filter */
3390 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3391 chunk_profiles_filter(chunk_type, bargs)) {
3396 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3397 chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
3399 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3400 chunk_usage_range_filter(bctl->fs_info, chunk_offset, bargs)) {
3405 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3406 chunk_devid_filter(leaf, chunk, bargs)) {
3410 /* drange filter, makes sense only with devid filter */
3411 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3412 chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
3417 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3418 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3422 /* stripes filter */
3423 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3424 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3428 /* soft profile changing mode */
3429 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3430 chunk_soft_convert_filter(chunk_type, bargs)) {
3435 * limited by count, must be the last filter
3437 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3438 if (bargs->limit == 0)
3442 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3444 * Same logic as the 'limit' filter; the minimum cannot be
3445 * determined here because we do not have the global information
3446 * about the count of all chunks that satisfy the filters.
3448 if (bargs->limit_max == 0)
3457 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3459 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3460 struct btrfs_root *chunk_root = fs_info->chunk_root;
3461 struct btrfs_root *dev_root = fs_info->dev_root;
3462 struct list_head *devices;
3463 struct btrfs_device *device;
3467 struct btrfs_chunk *chunk;
3468 struct btrfs_path *path = NULL;
3469 struct btrfs_key key;
3470 struct btrfs_key found_key;
3471 struct btrfs_trans_handle *trans;
3472 struct extent_buffer *leaf;
3475 int enospc_errors = 0;
3476 bool counting = true;
3477 /* The single value limit and min/max limits use the same bytes in the */
3478 u64 limit_data = bctl->data.limit;
3479 u64 limit_meta = bctl->meta.limit;
3480 u64 limit_sys = bctl->sys.limit;
3484 int chunk_reserved = 0;
3487 /* step one make some room on all the devices */
3488 devices = &fs_info->fs_devices->devices;
3489 list_for_each_entry(device, devices, dev_list) {
3490 old_size = btrfs_device_get_total_bytes(device);
3491 size_to_free = div_factor(old_size, 1);
3492 size_to_free = min_t(u64, size_to_free, SZ_1M);
3493 if (!device->writeable ||
3494 btrfs_device_get_total_bytes(device) -
3495 btrfs_device_get_bytes_used(device) > size_to_free ||
3496 device->is_tgtdev_for_dev_replace)
3499 ret = btrfs_shrink_device(device, old_size - size_to_free);
3503 /* btrfs_shrink_device never returns ret > 0 */
3508 trans = btrfs_start_transaction(dev_root, 0);
3509 if (IS_ERR(trans)) {
3510 ret = PTR_ERR(trans);
3511 btrfs_info_in_rcu(fs_info,
3512 "resize: unable to start transaction after shrinking device %s (error %d), old size %llu, new size %llu",
3513 rcu_str_deref(device->name), ret,
3514 old_size, old_size - size_to_free);
3518 ret = btrfs_grow_device(trans, device, old_size);
3520 btrfs_end_transaction(trans, dev_root);
3521 /* btrfs_grow_device never returns ret > 0 */
3523 btrfs_info_in_rcu(fs_info,
3524 "resize: unable to grow device after shrinking device %s (error %d), old size %llu, new size %llu",
3525 rcu_str_deref(device->name), ret,
3526 old_size, old_size - size_to_free);
3530 btrfs_end_transaction(trans, dev_root);
3533 /* step two, relocate all the chunks */
3534 path = btrfs_alloc_path();
3540 /* zero out stat counters */
3541 spin_lock(&fs_info->balance_lock);
3542 memset(&bctl->stat, 0, sizeof(bctl->stat));
3543 spin_unlock(&fs_info->balance_lock);
3547 * The single value limit and min/max limits use the same bytes
3550 bctl->data.limit = limit_data;
3551 bctl->meta.limit = limit_meta;
3552 bctl->sys.limit = limit_sys;
3554 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3555 key.offset = (u64)-1;
3556 key.type = BTRFS_CHUNK_ITEM_KEY;
3559 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3560 atomic_read(&fs_info->balance_cancel_req)) {
3565 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3566 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3568 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3573 * this shouldn't happen, it means the last relocate
3577 BUG(); /* FIXME break ? */
3579 ret = btrfs_previous_item(chunk_root, path, 0,
3580 BTRFS_CHUNK_ITEM_KEY);
3582 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3587 leaf = path->nodes[0];
3588 slot = path->slots[0];
3589 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3591 if (found_key.objectid != key.objectid) {
3592 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3596 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3597 chunk_type = btrfs_chunk_type(leaf, chunk);
3600 spin_lock(&fs_info->balance_lock);
3601 bctl->stat.considered++;
3602 spin_unlock(&fs_info->balance_lock);
3605 ret = should_balance_chunk(chunk_root, leaf, chunk,
3608 btrfs_release_path(path);
3610 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3615 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3616 spin_lock(&fs_info->balance_lock);
3617 bctl->stat.expected++;
3618 spin_unlock(&fs_info->balance_lock);
3620 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3622 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3624 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3631 * Apply limit_min filter, no need to check if the LIMITS
3632 * filter is used, limit_min is 0 by default
3634 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3635 count_data < bctl->data.limit_min)
3636 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3637 count_meta < bctl->meta.limit_min)
3638 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3639 count_sys < bctl->sys.limit_min)) {
3640 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3644 ASSERT(fs_info->data_sinfo);
3645 spin_lock(&fs_info->data_sinfo->lock);
3646 bytes_used = fs_info->data_sinfo->bytes_used;
3647 spin_unlock(&fs_info->data_sinfo->lock);
3649 if ((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3650 !chunk_reserved && !bytes_used) {
3651 trans = btrfs_start_transaction(chunk_root, 0);
3652 if (IS_ERR(trans)) {
3653 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3654 ret = PTR_ERR(trans);
3658 ret = btrfs_force_chunk_alloc(trans, chunk_root,
3659 BTRFS_BLOCK_GROUP_DATA);
3660 btrfs_end_transaction(trans, chunk_root);
3662 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3668 ret = btrfs_relocate_chunk(chunk_root,
3670 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3671 if (ret && ret != -ENOSPC)
3673 if (ret == -ENOSPC) {
3676 spin_lock(&fs_info->balance_lock);
3677 bctl->stat.completed++;
3678 spin_unlock(&fs_info->balance_lock);
3681 if (found_key.offset == 0)
3683 key.offset = found_key.offset - 1;
3687 btrfs_release_path(path);
3692 btrfs_free_path(path);
3693 if (enospc_errors) {
3694 btrfs_info(fs_info, "%d enospc errors during balance",
3704 * alloc_profile_is_valid - see if a given profile is valid and reduced
3705 * @flags: profile to validate
3706 * @extended: if true @flags is treated as an extended profile
3708 static int alloc_profile_is_valid(u64 flags, int extended)
3710 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3711 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3713 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3715 /* 1) check that all other bits are zeroed */
3719 /* 2) see if profile is reduced */
3721 return !extended; /* "0" is valid for usual profiles */
3723 /* true if exactly one bit set */
3724 return (flags & (flags - 1)) == 0;
3727 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3729 /* cancel requested || normal exit path */
3730 return atomic_read(&fs_info->balance_cancel_req) ||
3731 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3732 atomic_read(&fs_info->balance_cancel_req) == 0);
3735 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3739 unset_balance_control(fs_info);
3740 ret = del_balance_item(fs_info->tree_root);
3742 btrfs_handle_fs_error(fs_info, ret, NULL);
3744 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3747 /* Non-zero return value signifies invalidity */
3748 static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg,
3751 return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3752 (!alloc_profile_is_valid(bctl_arg->target, 1) ||
3753 (bctl_arg->target & ~allowed)));
3757 * Should be called with both balance and volume mutexes held
3759 int btrfs_balance(struct btrfs_balance_control *bctl,
3760 struct btrfs_ioctl_balance_args *bargs)
3762 struct btrfs_fs_info *fs_info = bctl->fs_info;
3769 if (btrfs_fs_closing(fs_info) ||
3770 atomic_read(&fs_info->balance_pause_req) ||
3771 atomic_read(&fs_info->balance_cancel_req)) {
3776 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3777 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3781 * In case of mixed groups both data and meta should be picked,
3782 * and identical options should be given for both of them.
3784 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3785 if (mixed && (bctl->flags & allowed)) {
3786 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3787 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3788 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3790 "with mixed groups data and metadata balance options must be the same");
3796 num_devices = fs_info->fs_devices->num_devices;
3797 btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
3798 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3799 BUG_ON(num_devices < 1);
3802 btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
3803 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE | BTRFS_BLOCK_GROUP_DUP;
3804 if (num_devices > 1)
3805 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3806 if (num_devices > 2)
3807 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3808 if (num_devices > 3)
3809 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3810 BTRFS_BLOCK_GROUP_RAID6);
3811 if (validate_convert_profile(&bctl->data, allowed)) {
3813 "unable to start balance with target data profile %llu",
3818 if (validate_convert_profile(&bctl->meta, allowed)) {
3820 "unable to start balance with target metadata profile %llu",
3825 if (validate_convert_profile(&bctl->sys, allowed)) {
3827 "unable to start balance with target system profile %llu",
3833 /* allow to reduce meta or sys integrity only if force set */
3834 allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3835 BTRFS_BLOCK_GROUP_RAID10 |
3836 BTRFS_BLOCK_GROUP_RAID5 |
3837 BTRFS_BLOCK_GROUP_RAID6;
3839 seq = read_seqbegin(&fs_info->profiles_lock);
3841 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3842 (fs_info->avail_system_alloc_bits & allowed) &&
3843 !(bctl->sys.target & allowed)) ||
3844 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3845 (fs_info->avail_metadata_alloc_bits & allowed) &&
3846 !(bctl->meta.target & allowed))) {
3847 if (bctl->flags & BTRFS_BALANCE_FORCE) {
3849 "force reducing metadata integrity");
3852 "balance will reduce metadata integrity, use force if you want this");
3857 } while (read_seqretry(&fs_info->profiles_lock, seq));
3859 if (btrfs_get_num_tolerated_disk_barrier_failures(bctl->meta.target) <
3860 btrfs_get_num_tolerated_disk_barrier_failures(bctl->data.target)) {
3862 "metadata profile 0x%llx has lower redundancy than data profile 0x%llx",
3863 bctl->meta.target, bctl->data.target);
3866 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3867 fs_info->num_tolerated_disk_barrier_failures = min(
3868 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info),
3869 btrfs_get_num_tolerated_disk_barrier_failures(
3873 ret = insert_balance_item(fs_info->tree_root, bctl);
3874 if (ret && ret != -EEXIST)
3877 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3878 BUG_ON(ret == -EEXIST);
3879 set_balance_control(bctl);
3881 BUG_ON(ret != -EEXIST);
3882 spin_lock(&fs_info->balance_lock);
3883 update_balance_args(bctl);
3884 spin_unlock(&fs_info->balance_lock);
3887 atomic_inc(&fs_info->balance_running);
3888 mutex_unlock(&fs_info->balance_mutex);
3890 ret = __btrfs_balance(fs_info);
3892 mutex_lock(&fs_info->balance_mutex);
3893 atomic_dec(&fs_info->balance_running);
3895 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3896 fs_info->num_tolerated_disk_barrier_failures =
3897 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3901 memset(bargs, 0, sizeof(*bargs));
3902 update_ioctl_balance_args(fs_info, 0, bargs);
3905 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3906 balance_need_close(fs_info)) {
3907 __cancel_balance(fs_info);
3910 wake_up(&fs_info->balance_wait_q);
3914 if (bctl->flags & BTRFS_BALANCE_RESUME)
3915 __cancel_balance(fs_info);
3918 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3923 static int balance_kthread(void *data)
3925 struct btrfs_fs_info *fs_info = data;
3928 mutex_lock(&fs_info->volume_mutex);
3929 mutex_lock(&fs_info->balance_mutex);
3931 if (fs_info->balance_ctl) {
3932 btrfs_info(fs_info, "continuing balance");
3933 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3936 mutex_unlock(&fs_info->balance_mutex);
3937 mutex_unlock(&fs_info->volume_mutex);
3942 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3944 struct task_struct *tsk;
3946 spin_lock(&fs_info->balance_lock);
3947 if (!fs_info->balance_ctl) {
3948 spin_unlock(&fs_info->balance_lock);
3951 spin_unlock(&fs_info->balance_lock);
3953 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
3954 btrfs_info(fs_info, "force skipping balance");
3958 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3959 return PTR_ERR_OR_ZERO(tsk);
3962 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3964 struct btrfs_balance_control *bctl;
3965 struct btrfs_balance_item *item;
3966 struct btrfs_disk_balance_args disk_bargs;
3967 struct btrfs_path *path;
3968 struct extent_buffer *leaf;
3969 struct btrfs_key key;
3972 path = btrfs_alloc_path();
3976 key.objectid = BTRFS_BALANCE_OBJECTID;
3977 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3980 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3983 if (ret > 0) { /* ret = -ENOENT; */
3988 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3994 leaf = path->nodes[0];
3995 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3997 bctl->fs_info = fs_info;
3998 bctl->flags = btrfs_balance_flags(leaf, item);
3999 bctl->flags |= BTRFS_BALANCE_RESUME;
4001 btrfs_balance_data(leaf, item, &disk_bargs);
4002 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4003 btrfs_balance_meta(leaf, item, &disk_bargs);
4004 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4005 btrfs_balance_sys(leaf, item, &disk_bargs);
4006 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4008 WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
4010 mutex_lock(&fs_info->volume_mutex);
4011 mutex_lock(&fs_info->balance_mutex);
4013 set_balance_control(bctl);
4015 mutex_unlock(&fs_info->balance_mutex);
4016 mutex_unlock(&fs_info->volume_mutex);
4018 btrfs_free_path(path);
4022 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4026 mutex_lock(&fs_info->balance_mutex);
4027 if (!fs_info->balance_ctl) {
4028 mutex_unlock(&fs_info->balance_mutex);
4032 if (atomic_read(&fs_info->balance_running)) {
4033 atomic_inc(&fs_info->balance_pause_req);
4034 mutex_unlock(&fs_info->balance_mutex);
4036 wait_event(fs_info->balance_wait_q,
4037 atomic_read(&fs_info->balance_running) == 0);
4039 mutex_lock(&fs_info->balance_mutex);
4040 /* we are good with balance_ctl ripped off from under us */
4041 BUG_ON(atomic_read(&fs_info->balance_running));
4042 atomic_dec(&fs_info->balance_pause_req);
4047 mutex_unlock(&fs_info->balance_mutex);
4051 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4053 if (fs_info->sb->s_flags & MS_RDONLY)
4056 mutex_lock(&fs_info->balance_mutex);
4057 if (!fs_info->balance_ctl) {
4058 mutex_unlock(&fs_info->balance_mutex);
4062 atomic_inc(&fs_info->balance_cancel_req);
4064 * if we are running just wait and return, balance item is
4065 * deleted in btrfs_balance in this case
4067 if (atomic_read(&fs_info->balance_running)) {
4068 mutex_unlock(&fs_info->balance_mutex);
4069 wait_event(fs_info->balance_wait_q,
4070 atomic_read(&fs_info->balance_running) == 0);
4071 mutex_lock(&fs_info->balance_mutex);
4073 /* __cancel_balance needs volume_mutex */
4074 mutex_unlock(&fs_info->balance_mutex);
4075 mutex_lock(&fs_info->volume_mutex);
4076 mutex_lock(&fs_info->balance_mutex);
4078 if (fs_info->balance_ctl)
4079 __cancel_balance(fs_info);
4081 mutex_unlock(&fs_info->volume_mutex);
4084 BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
4085 atomic_dec(&fs_info->balance_cancel_req);
4086 mutex_unlock(&fs_info->balance_mutex);
4090 static int btrfs_uuid_scan_kthread(void *data)
4092 struct btrfs_fs_info *fs_info = data;
4093 struct btrfs_root *root = fs_info->tree_root;
4094 struct btrfs_key key;
4095 struct btrfs_key max_key;
4096 struct btrfs_path *path = NULL;
4098 struct extent_buffer *eb;
4100 struct btrfs_root_item root_item;
4102 struct btrfs_trans_handle *trans = NULL;
4104 path = btrfs_alloc_path();
4111 key.type = BTRFS_ROOT_ITEM_KEY;
4114 max_key.objectid = (u64)-1;
4115 max_key.type = BTRFS_ROOT_ITEM_KEY;
4116 max_key.offset = (u64)-1;
4119 ret = btrfs_search_forward(root, &key, path, 0);
4126 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4127 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4128 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4129 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4132 eb = path->nodes[0];
4133 slot = path->slots[0];
4134 item_size = btrfs_item_size_nr(eb, slot);
4135 if (item_size < sizeof(root_item))
4138 read_extent_buffer(eb, &root_item,
4139 btrfs_item_ptr_offset(eb, slot),
4140 (int)sizeof(root_item));
4141 if (btrfs_root_refs(&root_item) == 0)
4144 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4145 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4149 btrfs_release_path(path);
4151 * 1 - subvol uuid item
4152 * 1 - received_subvol uuid item
4154 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4155 if (IS_ERR(trans)) {
4156 ret = PTR_ERR(trans);
4164 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4165 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
4167 BTRFS_UUID_KEY_SUBVOL,
4170 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4176 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4177 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
4178 root_item.received_uuid,
4179 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4182 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4190 ret = btrfs_end_transaction(trans, fs_info->uuid_root);
4196 btrfs_release_path(path);
4197 if (key.offset < (u64)-1) {
4199 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4201 key.type = BTRFS_ROOT_ITEM_KEY;
4202 } else if (key.objectid < (u64)-1) {
4204 key.type = BTRFS_ROOT_ITEM_KEY;
4213 btrfs_free_path(path);
4214 if (trans && !IS_ERR(trans))
4215 btrfs_end_transaction(trans, fs_info->uuid_root);
4217 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4219 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4220 up(&fs_info->uuid_tree_rescan_sem);
4225 * Callback for btrfs_uuid_tree_iterate().
4227 * 0 check succeeded, the entry is not outdated.
4228 * < 0 if an error occurred.
4229 * > 0 if the check failed, which means the caller shall remove the entry.
4231 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
4232 u8 *uuid, u8 type, u64 subid)
4234 struct btrfs_key key;
4236 struct btrfs_root *subvol_root;
4238 if (type != BTRFS_UUID_KEY_SUBVOL &&
4239 type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
4242 key.objectid = subid;
4243 key.type = BTRFS_ROOT_ITEM_KEY;
4244 key.offset = (u64)-1;
4245 subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
4246 if (IS_ERR(subvol_root)) {
4247 ret = PTR_ERR(subvol_root);
4254 case BTRFS_UUID_KEY_SUBVOL:
4255 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
4258 case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
4259 if (memcmp(uuid, subvol_root->root_item.received_uuid,
4269 static int btrfs_uuid_rescan_kthread(void *data)
4271 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4275 * 1st step is to iterate through the existing UUID tree and
4276 * to delete all entries that contain outdated data.
4277 * 2nd step is to add all missing entries to the UUID tree.
4279 ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4281 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4282 up(&fs_info->uuid_tree_rescan_sem);
4285 return btrfs_uuid_scan_kthread(data);
4288 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4290 struct btrfs_trans_handle *trans;
4291 struct btrfs_root *tree_root = fs_info->tree_root;
4292 struct btrfs_root *uuid_root;
4293 struct task_struct *task;
4300 trans = btrfs_start_transaction(tree_root, 2);
4302 return PTR_ERR(trans);
4304 uuid_root = btrfs_create_tree(trans, fs_info,
4305 BTRFS_UUID_TREE_OBJECTID);
4306 if (IS_ERR(uuid_root)) {
4307 ret = PTR_ERR(uuid_root);
4308 btrfs_abort_transaction(trans, ret);
4309 btrfs_end_transaction(trans, tree_root);
4313 fs_info->uuid_root = uuid_root;
4315 ret = btrfs_commit_transaction(trans, tree_root);
4319 down(&fs_info->uuid_tree_rescan_sem);
4320 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4322 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4323 btrfs_warn(fs_info, "failed to start uuid_scan task");
4324 up(&fs_info->uuid_tree_rescan_sem);
4325 return PTR_ERR(task);
4331 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4333 struct task_struct *task;
4335 down(&fs_info->uuid_tree_rescan_sem);
4336 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4338 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4339 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4340 up(&fs_info->uuid_tree_rescan_sem);
4341 return PTR_ERR(task);
4348 * shrinking a device means finding all of the device extents past
4349 * the new size, and then following the back refs to the chunks.
4350 * The chunk relocation code actually frees the device extent
4352 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4354 struct btrfs_trans_handle *trans;
4355 struct btrfs_root *root = device->dev_root;
4356 struct btrfs_dev_extent *dev_extent = NULL;
4357 struct btrfs_path *path;
4363 bool retried = false;
4364 bool checked_pending_chunks = false;
4365 struct extent_buffer *l;
4366 struct btrfs_key key;
4367 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4368 u64 old_total = btrfs_super_total_bytes(super_copy);
4369 u64 old_size = btrfs_device_get_total_bytes(device);
4370 u64 diff = old_size - new_size;
4372 if (device->is_tgtdev_for_dev_replace)
4375 path = btrfs_alloc_path();
4379 path->reada = READA_FORWARD;
4383 btrfs_device_set_total_bytes(device, new_size);
4384 if (device->writeable) {
4385 device->fs_devices->total_rw_bytes -= diff;
4386 spin_lock(&root->fs_info->free_chunk_lock);
4387 root->fs_info->free_chunk_space -= diff;
4388 spin_unlock(&root->fs_info->free_chunk_lock);
4390 unlock_chunks(root);
4393 key.objectid = device->devid;
4394 key.offset = (u64)-1;
4395 key.type = BTRFS_DEV_EXTENT_KEY;
4398 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
4399 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4401 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4405 ret = btrfs_previous_item(root, path, 0, key.type);
4407 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4412 btrfs_release_path(path);
4417 slot = path->slots[0];
4418 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4420 if (key.objectid != device->devid) {
4421 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4422 btrfs_release_path(path);
4426 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4427 length = btrfs_dev_extent_length(l, dev_extent);
4429 if (key.offset + length <= new_size) {
4430 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4431 btrfs_release_path(path);
4435 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4436 btrfs_release_path(path);
4438 ret = btrfs_relocate_chunk(root, chunk_offset);
4439 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4440 if (ret && ret != -ENOSPC)
4444 } while (key.offset-- > 0);
4446 if (failed && !retried) {
4450 } else if (failed && retried) {
4455 /* Shrinking succeeded, else we would be at "done". */
4456 trans = btrfs_start_transaction(root, 0);
4457 if (IS_ERR(trans)) {
4458 ret = PTR_ERR(trans);
4465 * We checked in the above loop all device extents that were already in
4466 * the device tree. However before we have updated the device's
4467 * total_bytes to the new size, we might have had chunk allocations that
4468 * have not complete yet (new block groups attached to transaction
4469 * handles), and therefore their device extents were not yet in the
4470 * device tree and we missed them in the loop above. So if we have any
4471 * pending chunk using a device extent that overlaps the device range
4472 * that we can not use anymore, commit the current transaction and
4473 * repeat the search on the device tree - this way we guarantee we will
4474 * not have chunks using device extents that end beyond 'new_size'.
4476 if (!checked_pending_chunks) {
4477 u64 start = new_size;
4478 u64 len = old_size - new_size;
4480 if (contains_pending_extent(trans->transaction, device,
4482 unlock_chunks(root);
4483 checked_pending_chunks = true;
4486 ret = btrfs_commit_transaction(trans, root);
4493 btrfs_device_set_disk_total_bytes(device, new_size);
4494 if (list_empty(&device->resized_list))
4495 list_add_tail(&device->resized_list,
4496 &root->fs_info->fs_devices->resized_devices);
4498 WARN_ON(diff > old_total);
4499 btrfs_set_super_total_bytes(super_copy, old_total - diff);
4500 unlock_chunks(root);
4502 /* Now btrfs_update_device() will change the on-disk size. */
4503 ret = btrfs_update_device(trans, device);
4504 btrfs_end_transaction(trans, root);
4506 btrfs_free_path(path);
4509 btrfs_device_set_total_bytes(device, old_size);
4510 if (device->writeable)
4511 device->fs_devices->total_rw_bytes += diff;
4512 spin_lock(&root->fs_info->free_chunk_lock);
4513 root->fs_info->free_chunk_space += diff;
4514 spin_unlock(&root->fs_info->free_chunk_lock);
4515 unlock_chunks(root);
4520 static int btrfs_add_system_chunk(struct btrfs_root *root,
4521 struct btrfs_key *key,
4522 struct btrfs_chunk *chunk, int item_size)
4524 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4525 struct btrfs_disk_key disk_key;
4530 array_size = btrfs_super_sys_array_size(super_copy);
4531 if (array_size + item_size + sizeof(disk_key)
4532 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4533 unlock_chunks(root);
4537 ptr = super_copy->sys_chunk_array + array_size;
4538 btrfs_cpu_key_to_disk(&disk_key, key);
4539 memcpy(ptr, &disk_key, sizeof(disk_key));
4540 ptr += sizeof(disk_key);
4541 memcpy(ptr, chunk, item_size);
4542 item_size += sizeof(disk_key);
4543 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4544 unlock_chunks(root);
4550 * sort the devices in descending order by max_avail, total_avail
4552 static int btrfs_cmp_device_info(const void *a, const void *b)
4554 const struct btrfs_device_info *di_a = a;
4555 const struct btrfs_device_info *di_b = b;
4557 if (di_a->max_avail > di_b->max_avail)
4559 if (di_a->max_avail < di_b->max_avail)
4561 if (di_a->total_avail > di_b->total_avail)
4563 if (di_a->total_avail < di_b->total_avail)
4568 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
4570 /* TODO allow them to set a preferred stripe size */
4574 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4576 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4579 btrfs_set_fs_incompat(info, RAID56);
4582 #define BTRFS_MAX_DEVS(r) ((BTRFS_MAX_ITEM_SIZE(r) \
4583 - sizeof(struct btrfs_chunk)) \
4584 / sizeof(struct btrfs_stripe) + 1)
4586 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
4587 - 2 * sizeof(struct btrfs_disk_key) \
4588 - 2 * sizeof(struct btrfs_chunk)) \
4589 / sizeof(struct btrfs_stripe) + 1)
4591 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4592 struct btrfs_root *extent_root, u64 start,
4595 struct btrfs_fs_info *info = extent_root->fs_info;
4596 struct btrfs_fs_devices *fs_devices = info->fs_devices;
4597 struct list_head *cur;
4598 struct map_lookup *map = NULL;
4599 struct extent_map_tree *em_tree;
4600 struct extent_map *em;
4601 struct btrfs_device_info *devices_info = NULL;
4603 int num_stripes; /* total number of stripes to allocate */
4604 int data_stripes; /* number of stripes that count for
4606 int sub_stripes; /* sub_stripes info for map */
4607 int dev_stripes; /* stripes per dev */
4608 int devs_max; /* max devs to use */
4609 int devs_min; /* min devs needed */
4610 int devs_increment; /* ndevs has to be a multiple of this */
4611 int ncopies; /* how many copies to data has */
4613 u64 max_stripe_size;
4617 u64 raid_stripe_len = BTRFS_STRIPE_LEN;
4623 BUG_ON(!alloc_profile_is_valid(type, 0));
4625 if (list_empty(&fs_devices->alloc_list))
4628 index = __get_raid_index(type);
4630 sub_stripes = btrfs_raid_array[index].sub_stripes;
4631 dev_stripes = btrfs_raid_array[index].dev_stripes;
4632 devs_max = btrfs_raid_array[index].devs_max;
4633 devs_min = btrfs_raid_array[index].devs_min;
4634 devs_increment = btrfs_raid_array[index].devs_increment;
4635 ncopies = btrfs_raid_array[index].ncopies;
4637 if (type & BTRFS_BLOCK_GROUP_DATA) {
4638 max_stripe_size = SZ_1G;
4639 max_chunk_size = 10 * max_stripe_size;
4641 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4642 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4643 /* for larger filesystems, use larger metadata chunks */
4644 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
4645 max_stripe_size = SZ_1G;
4647 max_stripe_size = SZ_256M;
4648 max_chunk_size = max_stripe_size;
4650 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4651 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4652 max_stripe_size = SZ_32M;
4653 max_chunk_size = 2 * max_stripe_size;
4655 devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4657 btrfs_err(info, "invalid chunk type 0x%llx requested",
4662 /* we don't want a chunk larger than 10% of writeable space */
4663 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4666 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4671 cur = fs_devices->alloc_list.next;
4674 * in the first pass through the devices list, we gather information
4675 * about the available holes on each device.
4678 while (cur != &fs_devices->alloc_list) {
4679 struct btrfs_device *device;
4683 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4687 if (!device->writeable) {
4689 "BTRFS: read-only device in alloc_list\n");
4693 if (!device->in_fs_metadata ||
4694 device->is_tgtdev_for_dev_replace)
4697 if (device->total_bytes > device->bytes_used)
4698 total_avail = device->total_bytes - device->bytes_used;
4702 /* If there is no space on this device, skip it. */
4703 if (total_avail == 0)
4706 ret = find_free_dev_extent(trans, device,
4707 max_stripe_size * dev_stripes,
4708 &dev_offset, &max_avail);
4709 if (ret && ret != -ENOSPC)
4713 max_avail = max_stripe_size * dev_stripes;
4715 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4718 if (ndevs == fs_devices->rw_devices) {
4719 WARN(1, "%s: found more than %llu devices\n",
4720 __func__, fs_devices->rw_devices);
4723 devices_info[ndevs].dev_offset = dev_offset;
4724 devices_info[ndevs].max_avail = max_avail;
4725 devices_info[ndevs].total_avail = total_avail;
4726 devices_info[ndevs].dev = device;
4731 * now sort the devices by hole size / available space
4733 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4734 btrfs_cmp_device_info, NULL);
4736 /* round down to number of usable stripes */
4737 ndevs -= ndevs % devs_increment;
4739 if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4744 if (devs_max && ndevs > devs_max)
4747 * the primary goal is to maximize the number of stripes, so use as many
4748 * devices as possible, even if the stripes are not maximum sized.
4750 stripe_size = devices_info[ndevs-1].max_avail;
4751 num_stripes = ndevs * dev_stripes;
4754 * this will have to be fixed for RAID1 and RAID10 over
4757 data_stripes = num_stripes / ncopies;
4759 if (type & BTRFS_BLOCK_GROUP_RAID5) {
4760 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
4761 extent_root->stripesize);
4762 data_stripes = num_stripes - 1;
4764 if (type & BTRFS_BLOCK_GROUP_RAID6) {
4765 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
4766 extent_root->stripesize);
4767 data_stripes = num_stripes - 2;
4771 * Use the number of data stripes to figure out how big this chunk
4772 * is really going to be in terms of logical address space,
4773 * and compare that answer with the max chunk size
4775 if (stripe_size * data_stripes > max_chunk_size) {
4776 u64 mask = (1ULL << 24) - 1;
4778 stripe_size = div_u64(max_chunk_size, data_stripes);
4780 /* bump the answer up to a 16MB boundary */
4781 stripe_size = (stripe_size + mask) & ~mask;
4783 /* but don't go higher than the limits we found
4784 * while searching for free extents
4786 if (stripe_size > devices_info[ndevs-1].max_avail)
4787 stripe_size = devices_info[ndevs-1].max_avail;
4790 stripe_size = div_u64(stripe_size, dev_stripes);
4792 /* align to BTRFS_STRIPE_LEN */
4793 stripe_size = div_u64(stripe_size, raid_stripe_len);
4794 stripe_size *= raid_stripe_len;
4796 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4801 map->num_stripes = num_stripes;
4803 for (i = 0; i < ndevs; ++i) {
4804 for (j = 0; j < dev_stripes; ++j) {
4805 int s = i * dev_stripes + j;
4806 map->stripes[s].dev = devices_info[i].dev;
4807 map->stripes[s].physical = devices_info[i].dev_offset +
4811 map->sector_size = extent_root->sectorsize;
4812 map->stripe_len = raid_stripe_len;
4813 map->io_align = raid_stripe_len;
4814 map->io_width = raid_stripe_len;
4816 map->sub_stripes = sub_stripes;
4818 num_bytes = stripe_size * data_stripes;
4820 trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
4822 em = alloc_extent_map();
4828 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4829 em->map_lookup = map;
4831 em->len = num_bytes;
4832 em->block_start = 0;
4833 em->block_len = em->len;
4834 em->orig_block_len = stripe_size;
4836 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4837 write_lock(&em_tree->lock);
4838 ret = add_extent_mapping(em_tree, em, 0);
4840 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4841 atomic_inc(&em->refs);
4843 write_unlock(&em_tree->lock);
4845 free_extent_map(em);
4849 ret = btrfs_make_block_group(trans, extent_root, 0, type,
4850 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4853 goto error_del_extent;
4855 for (i = 0; i < map->num_stripes; i++) {
4856 num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4857 btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4860 spin_lock(&extent_root->fs_info->free_chunk_lock);
4861 extent_root->fs_info->free_chunk_space -= (stripe_size *
4863 spin_unlock(&extent_root->fs_info->free_chunk_lock);
4865 free_extent_map(em);
4866 check_raid56_incompat_flag(extent_root->fs_info, type);
4868 kfree(devices_info);
4872 write_lock(&em_tree->lock);
4873 remove_extent_mapping(em_tree, em);
4874 write_unlock(&em_tree->lock);
4876 /* One for our allocation */
4877 free_extent_map(em);
4878 /* One for the tree reference */
4879 free_extent_map(em);
4880 /* One for the pending_chunks list reference */
4881 free_extent_map(em);
4883 kfree(devices_info);
4887 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4888 struct btrfs_root *extent_root,
4889 u64 chunk_offset, u64 chunk_size)
4891 struct btrfs_key key;
4892 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4893 struct btrfs_device *device;
4894 struct btrfs_chunk *chunk;
4895 struct btrfs_stripe *stripe;
4896 struct extent_map_tree *em_tree;
4897 struct extent_map *em;
4898 struct map_lookup *map;
4905 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4906 read_lock(&em_tree->lock);
4907 em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size);
4908 read_unlock(&em_tree->lock);
4911 btrfs_crit(extent_root->fs_info,
4912 "unable to find logical %Lu len %Lu",
4913 chunk_offset, chunk_size);
4917 if (em->start != chunk_offset || em->len != chunk_size) {
4918 btrfs_crit(extent_root->fs_info,
4919 "found a bad mapping, wanted %Lu-%Lu, found %Lu-%Lu",
4920 chunk_offset, chunk_size, em->start, em->len);
4921 free_extent_map(em);
4925 map = em->map_lookup;
4926 item_size = btrfs_chunk_item_size(map->num_stripes);
4927 stripe_size = em->orig_block_len;
4929 chunk = kzalloc(item_size, GFP_NOFS);
4936 * Take the device list mutex to prevent races with the final phase of
4937 * a device replace operation that replaces the device object associated
4938 * with the map's stripes, because the device object's id can change
4939 * at any time during that final phase of the device replace operation
4940 * (dev-replace.c:btrfs_dev_replace_finishing()).
4942 mutex_lock(&chunk_root->fs_info->fs_devices->device_list_mutex);
4943 for (i = 0; i < map->num_stripes; i++) {
4944 device = map->stripes[i].dev;
4945 dev_offset = map->stripes[i].physical;
4947 ret = btrfs_update_device(trans, device);
4950 ret = btrfs_alloc_dev_extent(trans, device,
4951 chunk_root->root_key.objectid,
4952 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4953 chunk_offset, dev_offset,
4959 mutex_unlock(&chunk_root->fs_info->fs_devices->device_list_mutex);
4963 stripe = &chunk->stripe;
4964 for (i = 0; i < map->num_stripes; i++) {
4965 device = map->stripes[i].dev;
4966 dev_offset = map->stripes[i].physical;
4968 btrfs_set_stack_stripe_devid(stripe, device->devid);
4969 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4970 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4973 mutex_unlock(&chunk_root->fs_info->fs_devices->device_list_mutex);
4975 btrfs_set_stack_chunk_length(chunk, chunk_size);
4976 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4977 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4978 btrfs_set_stack_chunk_type(chunk, map->type);
4979 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4980 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4981 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4982 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4983 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4985 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4986 key.type = BTRFS_CHUNK_ITEM_KEY;
4987 key.offset = chunk_offset;
4989 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4990 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4992 * TODO: Cleanup of inserted chunk root in case of
4995 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
5001 free_extent_map(em);
5006 * Chunk allocation falls into two parts. The first part does works
5007 * that make the new allocated chunk useable, but not do any operation
5008 * that modifies the chunk tree. The second part does the works that
5009 * require modifying the chunk tree. This division is important for the
5010 * bootstrap process of adding storage to a seed btrfs.
5012 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
5013 struct btrfs_root *extent_root, u64 type)
5017 ASSERT(mutex_is_locked(&extent_root->fs_info->chunk_mutex));
5018 chunk_offset = find_next_chunk(extent_root->fs_info);
5019 return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type);
5022 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
5023 struct btrfs_root *root,
5024 struct btrfs_device *device)
5027 u64 sys_chunk_offset;
5029 struct btrfs_fs_info *fs_info = root->fs_info;
5030 struct btrfs_root *extent_root = fs_info->extent_root;
5033 chunk_offset = find_next_chunk(fs_info);
5034 alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
5035 ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset,
5040 sys_chunk_offset = find_next_chunk(root->fs_info);
5041 alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
5042 ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset,
5047 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
5051 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5052 BTRFS_BLOCK_GROUP_RAID10 |
5053 BTRFS_BLOCK_GROUP_RAID5 |
5054 BTRFS_BLOCK_GROUP_DUP)) {
5056 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
5065 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
5067 struct extent_map *em;
5068 struct map_lookup *map;
5069 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5074 read_lock(&map_tree->map_tree.lock);
5075 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
5076 read_unlock(&map_tree->map_tree.lock);
5080 map = em->map_lookup;
5081 for (i = 0; i < map->num_stripes; i++) {
5082 if (map->stripes[i].dev->missing) {
5087 if (!map->stripes[i].dev->writeable) {
5094 * If the number of missing devices is larger than max errors,
5095 * we can not write the data into that chunk successfully, so
5098 if (miss_ndevs > btrfs_chunk_max_errors(map))
5101 free_extent_map(em);
5105 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
5107 extent_map_tree_init(&tree->map_tree);
5110 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
5112 struct extent_map *em;
5115 write_lock(&tree->map_tree.lock);
5116 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
5118 remove_extent_mapping(&tree->map_tree, em);
5119 write_unlock(&tree->map_tree.lock);
5123 free_extent_map(em);
5124 /* once for the tree */
5125 free_extent_map(em);
5129 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5131 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5132 struct extent_map *em;
5133 struct map_lookup *map;
5134 struct extent_map_tree *em_tree = &map_tree->map_tree;
5137 read_lock(&em_tree->lock);
5138 em = lookup_extent_mapping(em_tree, logical, len);
5139 read_unlock(&em_tree->lock);
5142 * We could return errors for these cases, but that could get ugly and
5143 * we'd probably do the same thing which is just not do anything else
5144 * and exit, so return 1 so the callers don't try to use other copies.
5147 btrfs_crit(fs_info, "No mapping for %Lu-%Lu", logical,
5152 if (em->start > logical || em->start + em->len < logical) {
5153 btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got %Lu-%Lu",
5154 logical, logical+len, em->start,
5155 em->start + em->len);
5156 free_extent_map(em);
5160 map = em->map_lookup;
5161 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
5162 ret = map->num_stripes;
5163 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5164 ret = map->sub_stripes;
5165 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5167 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5171 free_extent_map(em);
5173 btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
5174 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
5176 btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
5181 unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
5182 struct btrfs_mapping_tree *map_tree,
5185 struct extent_map *em;
5186 struct map_lookup *map;
5187 struct extent_map_tree *em_tree = &map_tree->map_tree;
5188 unsigned long len = root->sectorsize;
5190 read_lock(&em_tree->lock);
5191 em = lookup_extent_mapping(em_tree, logical, len);
5192 read_unlock(&em_tree->lock);
5195 BUG_ON(em->start > logical || em->start + em->len < logical);
5196 map = em->map_lookup;
5197 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5198 len = map->stripe_len * nr_data_stripes(map);
5199 free_extent_map(em);
5203 int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
5204 u64 logical, u64 len, int mirror_num)
5206 struct extent_map *em;
5207 struct map_lookup *map;
5208 struct extent_map_tree *em_tree = &map_tree->map_tree;
5211 read_lock(&em_tree->lock);
5212 em = lookup_extent_mapping(em_tree, logical, len);
5213 read_unlock(&em_tree->lock);
5216 BUG_ON(em->start > logical || em->start + em->len < logical);
5217 map = em->map_lookup;
5218 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5220 free_extent_map(em);
5224 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5225 struct map_lookup *map, int first, int num,
5226 int optimal, int dev_replace_is_ongoing)
5230 struct btrfs_device *srcdev;
5232 if (dev_replace_is_ongoing &&
5233 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5234 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5235 srcdev = fs_info->dev_replace.srcdev;
5240 * try to avoid the drive that is the source drive for a
5241 * dev-replace procedure, only choose it if no other non-missing
5242 * mirror is available
5244 for (tolerance = 0; tolerance < 2; tolerance++) {
5245 if (map->stripes[optimal].dev->bdev &&
5246 (tolerance || map->stripes[optimal].dev != srcdev))
5248 for (i = first; i < first + num; i++) {
5249 if (map->stripes[i].dev->bdev &&
5250 (tolerance || map->stripes[i].dev != srcdev))
5255 /* we couldn't find one that doesn't fail. Just return something
5256 * and the io error handling code will clean up eventually
5261 static inline int parity_smaller(u64 a, u64 b)
5266 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5267 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5269 struct btrfs_bio_stripe s;
5276 for (i = 0; i < num_stripes - 1; i++) {
5277 if (parity_smaller(bbio->raid_map[i],
5278 bbio->raid_map[i+1])) {
5279 s = bbio->stripes[i];
5280 l = bbio->raid_map[i];
5281 bbio->stripes[i] = bbio->stripes[i+1];
5282 bbio->raid_map[i] = bbio->raid_map[i+1];
5283 bbio->stripes[i+1] = s;
5284 bbio->raid_map[i+1] = l;
5292 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5294 struct btrfs_bio *bbio = kzalloc(
5295 /* the size of the btrfs_bio */
5296 sizeof(struct btrfs_bio) +
5297 /* plus the variable array for the stripes */
5298 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5299 /* plus the variable array for the tgt dev */
5300 sizeof(int) * (real_stripes) +
5302 * plus the raid_map, which includes both the tgt dev
5305 sizeof(u64) * (total_stripes),
5306 GFP_NOFS|__GFP_NOFAIL);
5308 atomic_set(&bbio->error, 0);
5309 atomic_set(&bbio->refs, 1);
5314 void btrfs_get_bbio(struct btrfs_bio *bbio)
5316 WARN_ON(!atomic_read(&bbio->refs));
5317 atomic_inc(&bbio->refs);
5320 void btrfs_put_bbio(struct btrfs_bio *bbio)
5324 if (atomic_dec_and_test(&bbio->refs))
5328 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int op,
5329 u64 logical, u64 *length,
5330 struct btrfs_bio **bbio_ret,
5331 int mirror_num, int need_raid_map)
5333 struct extent_map *em;
5334 struct map_lookup *map;
5335 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5336 struct extent_map_tree *em_tree = &map_tree->map_tree;
5339 u64 stripe_end_offset;
5349 int tgtdev_indexes = 0;
5350 struct btrfs_bio *bbio = NULL;
5351 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
5352 int dev_replace_is_ongoing = 0;
5353 int num_alloc_stripes;
5354 int patch_the_first_stripe_for_dev_replace = 0;
5355 u64 physical_to_patch_in_first_stripe = 0;
5356 u64 raid56_full_stripe_start = (u64)-1;
5358 read_lock(&em_tree->lock);
5359 em = lookup_extent_mapping(em_tree, logical, *length);
5360 read_unlock(&em_tree->lock);
5363 btrfs_crit(fs_info, "unable to find logical %llu len %llu",
5368 if (em->start > logical || em->start + em->len < logical) {
5370 "found a bad mapping, wanted %Lu, found %Lu-%Lu",
5371 logical, em->start, em->start + em->len);
5372 free_extent_map(em);
5376 map = em->map_lookup;
5377 offset = logical - em->start;
5379 stripe_len = map->stripe_len;
5382 * stripe_nr counts the total number of stripes we have to stride
5383 * to get to this block
5385 stripe_nr = div64_u64(stripe_nr, stripe_len);
5387 stripe_offset = stripe_nr * stripe_len;
5388 if (offset < stripe_offset) {
5390 "stripe math has gone wrong, stripe_offset=%llu, offset=%llu, start=%llu, logical=%llu, stripe_len=%llu",
5391 stripe_offset, offset, em->start, logical,
5393 free_extent_map(em);
5397 /* stripe_offset is the offset of this block in its stripe*/
5398 stripe_offset = offset - stripe_offset;
5400 /* if we're here for raid56, we need to know the stripe aligned start */
5401 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5402 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5403 raid56_full_stripe_start = offset;
5405 /* allow a write of a full stripe, but make sure we don't
5406 * allow straddling of stripes
5408 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5410 raid56_full_stripe_start *= full_stripe_len;
5413 if (op == REQ_OP_DISCARD) {
5414 /* we don't discard raid56 yet */
5415 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5419 *length = min_t(u64, em->len - offset, *length);
5420 } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5422 /* For writes to RAID[56], allow a full stripeset across all disks.
5423 For other RAID types and for RAID[56] reads, just allow a single
5424 stripe (on a single disk). */
5425 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
5426 (op == REQ_OP_WRITE)) {
5427 max_len = stripe_len * nr_data_stripes(map) -
5428 (offset - raid56_full_stripe_start);
5430 /* we limit the length of each bio to what fits in a stripe */
5431 max_len = stripe_len - stripe_offset;
5433 *length = min_t(u64, em->len - offset, max_len);
5435 *length = em->len - offset;
5438 /* This is for when we're called from btrfs_merge_bio_hook() and all
5439 it cares about is the length */
5443 btrfs_dev_replace_lock(dev_replace, 0);
5444 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5445 if (!dev_replace_is_ongoing)
5446 btrfs_dev_replace_unlock(dev_replace, 0);
5448 btrfs_dev_replace_set_lock_blocking(dev_replace);
5450 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
5451 op != REQ_OP_WRITE && op != REQ_OP_DISCARD &&
5452 op != REQ_GET_READ_MIRRORS && dev_replace->tgtdev != NULL) {
5454 * in dev-replace case, for repair case (that's the only
5455 * case where the mirror is selected explicitly when
5456 * calling btrfs_map_block), blocks left of the left cursor
5457 * can also be read from the target drive.
5458 * For REQ_GET_READ_MIRRORS, the target drive is added as
5459 * the last one to the array of stripes. For READ, it also
5460 * needs to be supported using the same mirror number.
5461 * If the requested block is not left of the left cursor,
5462 * EIO is returned. This can happen because btrfs_num_copies()
5463 * returns one more in the dev-replace case.
5465 u64 tmp_length = *length;
5466 struct btrfs_bio *tmp_bbio = NULL;
5467 int tmp_num_stripes;
5468 u64 srcdev_devid = dev_replace->srcdev->devid;
5469 int index_srcdev = 0;
5471 u64 physical_of_found = 0;
5473 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
5474 logical, &tmp_length, &tmp_bbio, 0, 0);
5476 WARN_ON(tmp_bbio != NULL);
5480 tmp_num_stripes = tmp_bbio->num_stripes;
5481 if (mirror_num > tmp_num_stripes) {
5483 * REQ_GET_READ_MIRRORS does not contain this
5484 * mirror, that means that the requested area
5485 * is not left of the left cursor
5488 btrfs_put_bbio(tmp_bbio);
5493 * process the rest of the function using the mirror_num
5494 * of the source drive. Therefore look it up first.
5495 * At the end, patch the device pointer to the one of the
5498 for (i = 0; i < tmp_num_stripes; i++) {
5499 if (tmp_bbio->stripes[i].dev->devid != srcdev_devid)
5503 * In case of DUP, in order to keep it simple, only add
5504 * the mirror with the lowest physical address
5507 physical_of_found <= tmp_bbio->stripes[i].physical)
5512 physical_of_found = tmp_bbio->stripes[i].physical;
5515 btrfs_put_bbio(tmp_bbio);
5523 mirror_num = index_srcdev + 1;
5524 patch_the_first_stripe_for_dev_replace = 1;
5525 physical_to_patch_in_first_stripe = physical_of_found;
5526 } else if (mirror_num > map->num_stripes) {
5532 stripe_nr_orig = stripe_nr;
5533 stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
5534 stripe_nr_end = div_u64(stripe_nr_end, map->stripe_len);
5535 stripe_end_offset = stripe_nr_end * map->stripe_len -
5538 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5539 if (op == REQ_OP_DISCARD)
5540 num_stripes = min_t(u64, map->num_stripes,
5541 stripe_nr_end - stripe_nr_orig);
5542 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5544 if (op != REQ_OP_WRITE && op != REQ_OP_DISCARD &&
5545 op != REQ_GET_READ_MIRRORS)
5547 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
5548 if (op == REQ_OP_WRITE || op == REQ_OP_DISCARD ||
5549 op == REQ_GET_READ_MIRRORS)
5550 num_stripes = map->num_stripes;
5551 else if (mirror_num)
5552 stripe_index = mirror_num - 1;
5554 stripe_index = find_live_mirror(fs_info, map, 0,
5556 current->pid % map->num_stripes,
5557 dev_replace_is_ongoing);
5558 mirror_num = stripe_index + 1;
5561 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
5562 if (op == REQ_OP_WRITE || op == REQ_OP_DISCARD ||
5563 op == REQ_GET_READ_MIRRORS) {
5564 num_stripes = map->num_stripes;
5565 } else if (mirror_num) {
5566 stripe_index = mirror_num - 1;
5571 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5572 u32 factor = map->num_stripes / map->sub_stripes;
5574 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5575 stripe_index *= map->sub_stripes;
5577 if (op == REQ_OP_WRITE || op == REQ_GET_READ_MIRRORS)
5578 num_stripes = map->sub_stripes;
5579 else if (op == REQ_OP_DISCARD)
5580 num_stripes = min_t(u64, map->sub_stripes *
5581 (stripe_nr_end - stripe_nr_orig),
5583 else if (mirror_num)
5584 stripe_index += mirror_num - 1;
5586 int old_stripe_index = stripe_index;
5587 stripe_index = find_live_mirror(fs_info, map,
5589 map->sub_stripes, stripe_index +
5590 current->pid % map->sub_stripes,
5591 dev_replace_is_ongoing);
5592 mirror_num = stripe_index - old_stripe_index + 1;
5595 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5596 if (need_raid_map &&
5597 (op == REQ_OP_WRITE || op == REQ_GET_READ_MIRRORS ||
5599 /* push stripe_nr back to the start of the full stripe */
5600 stripe_nr = div_u64(raid56_full_stripe_start,
5601 stripe_len * nr_data_stripes(map));
5603 /* RAID[56] write or recovery. Return all stripes */
5604 num_stripes = map->num_stripes;
5605 max_errors = nr_parity_stripes(map);
5607 *length = map->stripe_len;
5612 * Mirror #0 or #1 means the original data block.
5613 * Mirror #2 is RAID5 parity block.
5614 * Mirror #3 is RAID6 Q block.
5616 stripe_nr = div_u64_rem(stripe_nr,
5617 nr_data_stripes(map), &stripe_index);
5619 stripe_index = nr_data_stripes(map) +
5622 /* We distribute the parity blocks across stripes */
5623 div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
5625 if ((op != REQ_OP_WRITE && op != REQ_OP_DISCARD &&
5626 op != REQ_GET_READ_MIRRORS) && mirror_num <= 1)
5631 * after this, stripe_nr is the number of stripes on this
5632 * device we have to walk to find the data, and stripe_index is
5633 * the number of our device in the stripe array
5635 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5637 mirror_num = stripe_index + 1;
5639 if (stripe_index >= map->num_stripes) {
5641 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
5642 stripe_index, map->num_stripes);
5647 num_alloc_stripes = num_stripes;
5648 if (dev_replace_is_ongoing) {
5649 if (op == REQ_OP_WRITE || op == REQ_OP_DISCARD)
5650 num_alloc_stripes <<= 1;
5651 if (op == REQ_GET_READ_MIRRORS)
5652 num_alloc_stripes++;
5653 tgtdev_indexes = num_stripes;
5656 bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
5661 if (dev_replace_is_ongoing)
5662 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
5664 /* build raid_map */
5665 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK &&
5667 ((op == REQ_OP_WRITE || op == REQ_GET_READ_MIRRORS) ||
5672 bbio->raid_map = (u64 *)((void *)bbio->stripes +
5673 sizeof(struct btrfs_bio_stripe) *
5675 sizeof(int) * tgtdev_indexes);
5677 /* Work out the disk rotation on this stripe-set */
5678 div_u64_rem(stripe_nr, num_stripes, &rot);
5680 /* Fill in the logical address of each stripe */
5681 tmp = stripe_nr * nr_data_stripes(map);
5682 for (i = 0; i < nr_data_stripes(map); i++)
5683 bbio->raid_map[(i+rot) % num_stripes] =
5684 em->start + (tmp + i) * map->stripe_len;
5686 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5687 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5688 bbio->raid_map[(i+rot+1) % num_stripes] =
5692 if (op == REQ_OP_DISCARD) {
5694 u32 sub_stripes = 0;
5695 u64 stripes_per_dev = 0;
5696 u32 remaining_stripes = 0;
5697 u32 last_stripe = 0;
5700 (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
5701 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5704 sub_stripes = map->sub_stripes;
5706 factor = map->num_stripes / sub_stripes;
5707 stripes_per_dev = div_u64_rem(stripe_nr_end -
5710 &remaining_stripes);
5711 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5712 last_stripe *= sub_stripes;
5715 for (i = 0; i < num_stripes; i++) {
5716 bbio->stripes[i].physical =
5717 map->stripes[stripe_index].physical +
5718 stripe_offset + stripe_nr * map->stripe_len;
5719 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5721 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5722 BTRFS_BLOCK_GROUP_RAID10)) {
5723 bbio->stripes[i].length = stripes_per_dev *
5726 if (i / sub_stripes < remaining_stripes)
5727 bbio->stripes[i].length +=
5731 * Special for the first stripe and
5734 * |-------|...|-------|
5738 if (i < sub_stripes)
5739 bbio->stripes[i].length -=
5742 if (stripe_index >= last_stripe &&
5743 stripe_index <= (last_stripe +
5745 bbio->stripes[i].length -=
5748 if (i == sub_stripes - 1)
5751 bbio->stripes[i].length = *length;
5754 if (stripe_index == map->num_stripes) {
5755 /* This could only happen for RAID0/10 */
5761 for (i = 0; i < num_stripes; i++) {
5762 bbio->stripes[i].physical =
5763 map->stripes[stripe_index].physical +
5765 stripe_nr * map->stripe_len;
5766 bbio->stripes[i].dev =
5767 map->stripes[stripe_index].dev;
5772 if (op == REQ_OP_WRITE || op == REQ_GET_READ_MIRRORS)
5773 max_errors = btrfs_chunk_max_errors(map);
5776 sort_parity_stripes(bbio, num_stripes);
5779 if (dev_replace_is_ongoing &&
5780 (op == REQ_OP_WRITE || op == REQ_OP_DISCARD) &&
5781 dev_replace->tgtdev != NULL) {
5782 int index_where_to_add;
5783 u64 srcdev_devid = dev_replace->srcdev->devid;
5786 * duplicate the write operations while the dev replace
5787 * procedure is running. Since the copying of the old disk
5788 * to the new disk takes place at run time while the
5789 * filesystem is mounted writable, the regular write
5790 * operations to the old disk have to be duplicated to go
5791 * to the new disk as well.
5792 * Note that device->missing is handled by the caller, and
5793 * that the write to the old disk is already set up in the
5796 index_where_to_add = num_stripes;
5797 for (i = 0; i < num_stripes; i++) {
5798 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5799 /* write to new disk, too */
5800 struct btrfs_bio_stripe *new =
5801 bbio->stripes + index_where_to_add;
5802 struct btrfs_bio_stripe *old =
5805 new->physical = old->physical;
5806 new->length = old->length;
5807 new->dev = dev_replace->tgtdev;
5808 bbio->tgtdev_map[i] = index_where_to_add;
5809 index_where_to_add++;
5814 num_stripes = index_where_to_add;
5815 } else if (dev_replace_is_ongoing && (op == REQ_GET_READ_MIRRORS) &&
5816 dev_replace->tgtdev != NULL) {
5817 u64 srcdev_devid = dev_replace->srcdev->devid;
5818 int index_srcdev = 0;
5820 u64 physical_of_found = 0;
5823 * During the dev-replace procedure, the target drive can
5824 * also be used to read data in case it is needed to repair
5825 * a corrupt block elsewhere. This is possible if the
5826 * requested area is left of the left cursor. In this area,
5827 * the target drive is a full copy of the source drive.
5829 for (i = 0; i < num_stripes; i++) {
5830 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5832 * In case of DUP, in order to keep it
5833 * simple, only add the mirror with the
5834 * lowest physical address
5837 physical_of_found <=
5838 bbio->stripes[i].physical)
5842 physical_of_found = bbio->stripes[i].physical;
5846 struct btrfs_bio_stripe *tgtdev_stripe =
5847 bbio->stripes + num_stripes;
5849 tgtdev_stripe->physical = physical_of_found;
5850 tgtdev_stripe->length =
5851 bbio->stripes[index_srcdev].length;
5852 tgtdev_stripe->dev = dev_replace->tgtdev;
5853 bbio->tgtdev_map[index_srcdev] = num_stripes;
5861 bbio->map_type = map->type;
5862 bbio->num_stripes = num_stripes;
5863 bbio->max_errors = max_errors;
5864 bbio->mirror_num = mirror_num;
5865 bbio->num_tgtdevs = tgtdev_indexes;
5868 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5869 * mirror_num == num_stripes + 1 && dev_replace target drive is
5870 * available as a mirror
5872 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5873 WARN_ON(num_stripes > 1);
5874 bbio->stripes[0].dev = dev_replace->tgtdev;
5875 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5876 bbio->mirror_num = map->num_stripes + 1;
5879 if (dev_replace_is_ongoing) {
5880 btrfs_dev_replace_clear_lock_blocking(dev_replace);
5881 btrfs_dev_replace_unlock(dev_replace, 0);
5883 free_extent_map(em);
5887 int btrfs_map_block(struct btrfs_fs_info *fs_info, int op,
5888 u64 logical, u64 *length,
5889 struct btrfs_bio **bbio_ret, int mirror_num)
5891 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret,
5895 /* For Scrub/replace */
5896 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, int op,
5897 u64 logical, u64 *length,
5898 struct btrfs_bio **bbio_ret, int mirror_num,
5901 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret,
5902 mirror_num, need_raid_map);
5905 int btrfs_rmap_block(struct btrfs_fs_info *fs_info,
5906 u64 chunk_start, u64 physical, u64 devid,
5907 u64 **logical, int *naddrs, int *stripe_len)
5909 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5910 struct extent_map_tree *em_tree = &map_tree->map_tree;
5911 struct extent_map *em;
5912 struct map_lookup *map;
5920 read_lock(&em_tree->lock);
5921 em = lookup_extent_mapping(em_tree, chunk_start, 1);
5922 read_unlock(&em_tree->lock);
5925 btrfs_err(fs_info, "couldn't find em for chunk %Lu",
5930 if (em->start != chunk_start) {
5931 btrfs_err(fs_info, "bad chunk start, em=%Lu, wanted=%Lu",
5932 em->start, chunk_start);
5933 free_extent_map(em);
5936 map = em->map_lookup;
5939 rmap_len = map->stripe_len;
5941 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5942 length = div_u64(length, map->num_stripes / map->sub_stripes);
5943 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5944 length = div_u64(length, map->num_stripes);
5945 else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5946 length = div_u64(length, nr_data_stripes(map));
5947 rmap_len = map->stripe_len * nr_data_stripes(map);
5950 buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
5951 BUG_ON(!buf); /* -ENOMEM */
5953 for (i = 0; i < map->num_stripes; i++) {
5954 if (devid && map->stripes[i].dev->devid != devid)
5956 if (map->stripes[i].physical > physical ||
5957 map->stripes[i].physical + length <= physical)
5960 stripe_nr = physical - map->stripes[i].physical;
5961 stripe_nr = div_u64(stripe_nr, map->stripe_len);
5963 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5964 stripe_nr = stripe_nr * map->num_stripes + i;
5965 stripe_nr = div_u64(stripe_nr, map->sub_stripes);
5966 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5967 stripe_nr = stripe_nr * map->num_stripes + i;
5968 } /* else if RAID[56], multiply by nr_data_stripes().
5969 * Alternatively, just use rmap_len below instead of
5970 * map->stripe_len */
5972 bytenr = chunk_start + stripe_nr * rmap_len;
5973 WARN_ON(nr >= map->num_stripes);
5974 for (j = 0; j < nr; j++) {
5975 if (buf[j] == bytenr)
5979 WARN_ON(nr >= map->num_stripes);
5986 *stripe_len = rmap_len;
5988 free_extent_map(em);
5992 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
5994 bio->bi_private = bbio->private;
5995 bio->bi_end_io = bbio->end_io;
5998 btrfs_put_bbio(bbio);
6001 static void btrfs_end_bio(struct bio *bio)
6003 struct btrfs_bio *bbio = bio->bi_private;
6004 int is_orig_bio = 0;
6006 if (bio->bi_error) {
6007 atomic_inc(&bbio->error);
6008 if (bio->bi_error == -EIO || bio->bi_error == -EREMOTEIO) {
6009 unsigned int stripe_index =
6010 btrfs_io_bio(bio)->stripe_index;
6011 struct btrfs_device *dev;
6013 BUG_ON(stripe_index >= bbio->num_stripes);
6014 dev = bbio->stripes[stripe_index].dev;
6016 if (bio_op(bio) == REQ_OP_WRITE)
6017 btrfs_dev_stat_inc(dev,
6018 BTRFS_DEV_STAT_WRITE_ERRS);
6020 btrfs_dev_stat_inc(dev,
6021 BTRFS_DEV_STAT_READ_ERRS);
6022 if ((bio->bi_opf & WRITE_FLUSH) == WRITE_FLUSH)
6023 btrfs_dev_stat_inc(dev,
6024 BTRFS_DEV_STAT_FLUSH_ERRS);
6025 btrfs_dev_stat_print_on_error(dev);
6030 if (bio == bbio->orig_bio)
6033 btrfs_bio_counter_dec(bbio->fs_info);
6035 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6038 bio = bbio->orig_bio;
6041 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6042 /* only send an error to the higher layers if it is
6043 * beyond the tolerance of the btrfs bio
6045 if (atomic_read(&bbio->error) > bbio->max_errors) {
6046 bio->bi_error = -EIO;
6049 * this bio is actually up to date, we didn't
6050 * go over the max number of errors
6055 btrfs_end_bbio(bbio, bio);
6056 } else if (!is_orig_bio) {
6062 * see run_scheduled_bios for a description of why bios are collected for
6065 * This will add one bio to the pending list for a device and make sure
6066 * the work struct is scheduled.
6068 static noinline void btrfs_schedule_bio(struct btrfs_root *root,
6069 struct btrfs_device *device,
6072 int should_queue = 1;
6073 struct btrfs_pending_bios *pending_bios;
6075 if (device->missing || !device->bdev) {
6080 /* don't bother with additional async steps for reads, right now */
6081 if (bio_op(bio) == REQ_OP_READ) {
6083 btrfsic_submit_bio(bio);
6089 * nr_async_bios allows us to reliably return congestion to the
6090 * higher layers. Otherwise, the async bio makes it appear we have
6091 * made progress against dirty pages when we've really just put it
6092 * on a queue for later
6094 atomic_inc(&root->fs_info->nr_async_bios);
6095 WARN_ON(bio->bi_next);
6096 bio->bi_next = NULL;
6098 spin_lock(&device->io_lock);
6099 if (bio->bi_opf & REQ_SYNC)
6100 pending_bios = &device->pending_sync_bios;
6102 pending_bios = &device->pending_bios;
6104 if (pending_bios->tail)
6105 pending_bios->tail->bi_next = bio;
6107 pending_bios->tail = bio;
6108 if (!pending_bios->head)
6109 pending_bios->head = bio;
6110 if (device->running_pending)
6113 spin_unlock(&device->io_lock);
6116 btrfs_queue_work(root->fs_info->submit_workers,
6120 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
6121 struct bio *bio, u64 physical, int dev_nr,
6124 struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
6126 bio->bi_private = bbio;
6127 btrfs_io_bio(bio)->stripe_index = dev_nr;
6128 bio->bi_end_io = btrfs_end_bio;
6129 bio->bi_iter.bi_sector = physical >> 9;
6132 struct rcu_string *name;
6135 name = rcu_dereference(dev->name);
6136 btrfs_debug(fs_info,
6137 "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
6138 bio_op(bio), bio->bi_opf,
6139 (u64)bio->bi_iter.bi_sector,
6140 (u_long)dev->bdev->bd_dev, name->str, dev->devid,
6141 bio->bi_iter.bi_size);
6145 bio->bi_bdev = dev->bdev;
6147 btrfs_bio_counter_inc_noblocked(root->fs_info);
6150 btrfs_schedule_bio(root, dev, bio);
6152 btrfsic_submit_bio(bio);
6155 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6157 atomic_inc(&bbio->error);
6158 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6159 /* Should be the original bio. */
6160 WARN_ON(bio != bbio->orig_bio);
6162 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6163 bio->bi_iter.bi_sector = logical >> 9;
6164 bio->bi_error = -EIO;
6165 btrfs_end_bbio(bbio, bio);
6169 int btrfs_map_bio(struct btrfs_root *root, struct bio *bio,
6170 int mirror_num, int async_submit)
6172 struct btrfs_device *dev;
6173 struct bio *first_bio = bio;
6174 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
6180 struct btrfs_bio *bbio = NULL;
6182 length = bio->bi_iter.bi_size;
6183 map_length = length;
6185 btrfs_bio_counter_inc_blocked(root->fs_info);
6186 ret = __btrfs_map_block(root->fs_info, bio_op(bio), logical,
6187 &map_length, &bbio, mirror_num, 1);
6189 btrfs_bio_counter_dec(root->fs_info);
6193 total_devs = bbio->num_stripes;
6194 bbio->orig_bio = first_bio;
6195 bbio->private = first_bio->bi_private;
6196 bbio->end_io = first_bio->bi_end_io;
6197 bbio->fs_info = root->fs_info;
6198 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6200 if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6201 ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) {
6202 /* In this case, map_length has been set to the length of
6203 a single stripe; not the whole write */
6204 if (bio_op(bio) == REQ_OP_WRITE) {
6205 ret = raid56_parity_write(root, bio, bbio, map_length);
6207 ret = raid56_parity_recover(root, bio, bbio, map_length,
6211 btrfs_bio_counter_dec(root->fs_info);
6215 if (map_length < length) {
6216 btrfs_crit(root->fs_info,
6217 "mapping failed logical %llu bio len %llu len %llu",
6218 logical, length, map_length);
6222 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6223 dev = bbio->stripes[dev_nr].dev;
6224 if (!dev || !dev->bdev ||
6225 (bio_op(bio) == REQ_OP_WRITE && !dev->writeable)) {
6226 bbio_error(bbio, first_bio, logical);
6230 if (dev_nr < total_devs - 1) {
6231 bio = btrfs_bio_clone(first_bio, GFP_NOFS);
6232 BUG_ON(!bio); /* -ENOMEM */
6236 submit_stripe_bio(root, bbio, bio,
6237 bbio->stripes[dev_nr].physical, dev_nr,
6240 btrfs_bio_counter_dec(root->fs_info);
6244 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
6247 struct btrfs_device *device;
6248 struct btrfs_fs_devices *cur_devices;
6250 cur_devices = fs_info->fs_devices;
6251 while (cur_devices) {
6253 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
6254 device = __find_device(&cur_devices->devices,
6259 cur_devices = cur_devices->seed;
6264 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
6265 struct btrfs_fs_devices *fs_devices,
6266 u64 devid, u8 *dev_uuid)
6268 struct btrfs_device *device;
6270 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6274 list_add(&device->dev_list, &fs_devices->devices);
6275 device->fs_devices = fs_devices;
6276 fs_devices->num_devices++;
6278 device->missing = 1;
6279 fs_devices->missing_devices++;
6285 * btrfs_alloc_device - allocate struct btrfs_device
6286 * @fs_info: used only for generating a new devid, can be NULL if
6287 * devid is provided (i.e. @devid != NULL).
6288 * @devid: a pointer to devid for this device. If NULL a new devid
6290 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6293 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6294 * on error. Returned struct is not linked onto any lists and can be
6295 * destroyed with kfree() right away.
6297 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6301 struct btrfs_device *dev;
6304 if (WARN_ON(!devid && !fs_info))
6305 return ERR_PTR(-EINVAL);
6307 dev = __alloc_device();
6316 ret = find_next_devid(fs_info, &tmp);
6319 return ERR_PTR(ret);
6325 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6327 generate_random_uuid(dev->uuid);
6329 btrfs_init_work(&dev->work, btrfs_submit_helper,
6330 pending_bios_fn, NULL, NULL);
6335 /* Return -EIO if any error, otherwise return 0. */
6336 static int btrfs_check_chunk_valid(struct btrfs_root *root,
6337 struct extent_buffer *leaf,
6338 struct btrfs_chunk *chunk, u64 logical)
6346 length = btrfs_chunk_length(leaf, chunk);
6347 stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6348 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6349 sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6350 type = btrfs_chunk_type(leaf, chunk);
6353 btrfs_err(root->fs_info, "invalid chunk num_stripes: %u",
6357 if (!IS_ALIGNED(logical, root->sectorsize)) {
6358 btrfs_err(root->fs_info,
6359 "invalid chunk logical %llu", logical);
6362 if (btrfs_chunk_sector_size(leaf, chunk) != root->sectorsize) {
6363 btrfs_err(root->fs_info, "invalid chunk sectorsize %u",
6364 btrfs_chunk_sector_size(leaf, chunk));
6367 if (!length || !IS_ALIGNED(length, root->sectorsize)) {
6368 btrfs_err(root->fs_info,
6369 "invalid chunk length %llu", length);
6372 if (!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN) {
6373 btrfs_err(root->fs_info, "invalid chunk stripe length: %llu",
6377 if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6379 btrfs_err(root->fs_info, "unrecognized chunk type: %llu",
6380 ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
6381 BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6382 btrfs_chunk_type(leaf, chunk));
6385 if ((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes != 2) ||
6386 (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) ||
6387 (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) ||
6388 (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) ||
6389 (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) ||
6390 ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
6391 num_stripes != 1)) {
6392 btrfs_err(root->fs_info,
6393 "invalid num_stripes:sub_stripes %u:%u for profile %llu",
6394 num_stripes, sub_stripes,
6395 type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
6402 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
6403 struct extent_buffer *leaf,
6404 struct btrfs_chunk *chunk)
6406 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
6407 struct map_lookup *map;
6408 struct extent_map *em;
6413 u8 uuid[BTRFS_UUID_SIZE];
6418 logical = key->offset;
6419 length = btrfs_chunk_length(leaf, chunk);
6420 stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6421 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6423 ret = btrfs_check_chunk_valid(root, leaf, chunk, logical);
6427 read_lock(&map_tree->map_tree.lock);
6428 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
6429 read_unlock(&map_tree->map_tree.lock);
6431 /* already mapped? */
6432 if (em && em->start <= logical && em->start + em->len > logical) {
6433 free_extent_map(em);
6436 free_extent_map(em);
6439 em = alloc_extent_map();
6442 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6444 free_extent_map(em);
6448 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6449 em->map_lookup = map;
6450 em->start = logical;
6453 em->block_start = 0;
6454 em->block_len = em->len;
6456 map->num_stripes = num_stripes;
6457 map->io_width = btrfs_chunk_io_width(leaf, chunk);
6458 map->io_align = btrfs_chunk_io_align(leaf, chunk);
6459 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
6460 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6461 map->type = btrfs_chunk_type(leaf, chunk);
6462 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6463 for (i = 0; i < num_stripes; i++) {
6464 map->stripes[i].physical =
6465 btrfs_stripe_offset_nr(leaf, chunk, i);
6466 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6467 read_extent_buffer(leaf, uuid, (unsigned long)
6468 btrfs_stripe_dev_uuid_nr(chunk, i),
6470 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
6472 if (!map->stripes[i].dev &&
6473 !btrfs_test_opt(root->fs_info, DEGRADED)) {
6474 free_extent_map(em);
6477 if (!map->stripes[i].dev) {
6478 map->stripes[i].dev =
6479 add_missing_dev(root, root->fs_info->fs_devices,
6481 if (!map->stripes[i].dev) {
6482 free_extent_map(em);
6485 btrfs_warn(root->fs_info,
6486 "devid %llu uuid %pU is missing",
6489 map->stripes[i].dev->in_fs_metadata = 1;
6492 write_lock(&map_tree->map_tree.lock);
6493 ret = add_extent_mapping(&map_tree->map_tree, em, 0);
6494 write_unlock(&map_tree->map_tree.lock);
6495 BUG_ON(ret); /* Tree corruption */
6496 free_extent_map(em);
6501 static void fill_device_from_item(struct extent_buffer *leaf,
6502 struct btrfs_dev_item *dev_item,
6503 struct btrfs_device *device)
6507 device->devid = btrfs_device_id(leaf, dev_item);
6508 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6509 device->total_bytes = device->disk_total_bytes;
6510 device->commit_total_bytes = device->disk_total_bytes;
6511 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6512 device->commit_bytes_used = device->bytes_used;
6513 device->type = btrfs_device_type(leaf, dev_item);
6514 device->io_align = btrfs_device_io_align(leaf, dev_item);
6515 device->io_width = btrfs_device_io_width(leaf, dev_item);
6516 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6517 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6518 device->is_tgtdev_for_dev_replace = 0;
6520 ptr = btrfs_device_uuid(dev_item);
6521 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6524 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_root *root,
6527 struct btrfs_fs_devices *fs_devices;
6530 BUG_ON(!mutex_is_locked(&uuid_mutex));
6532 fs_devices = root->fs_info->fs_devices->seed;
6533 while (fs_devices) {
6534 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE))
6537 fs_devices = fs_devices->seed;
6540 fs_devices = find_fsid(fsid);
6542 if (!btrfs_test_opt(root->fs_info, DEGRADED))
6543 return ERR_PTR(-ENOENT);
6545 fs_devices = alloc_fs_devices(fsid);
6546 if (IS_ERR(fs_devices))
6549 fs_devices->seeding = 1;
6550 fs_devices->opened = 1;
6554 fs_devices = clone_fs_devices(fs_devices);
6555 if (IS_ERR(fs_devices))
6558 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
6559 root->fs_info->bdev_holder);
6561 free_fs_devices(fs_devices);
6562 fs_devices = ERR_PTR(ret);
6566 if (!fs_devices->seeding) {
6567 __btrfs_close_devices(fs_devices);
6568 free_fs_devices(fs_devices);
6569 fs_devices = ERR_PTR(-EINVAL);
6573 fs_devices->seed = root->fs_info->fs_devices->seed;
6574 root->fs_info->fs_devices->seed = fs_devices;
6579 static int read_one_dev(struct btrfs_root *root,
6580 struct extent_buffer *leaf,
6581 struct btrfs_dev_item *dev_item)
6583 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6584 struct btrfs_device *device;
6587 u8 fs_uuid[BTRFS_UUID_SIZE];
6588 u8 dev_uuid[BTRFS_UUID_SIZE];
6590 devid = btrfs_device_id(leaf, dev_item);
6591 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6593 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6596 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
6597 fs_devices = open_seed_devices(root, fs_uuid);
6598 if (IS_ERR(fs_devices))
6599 return PTR_ERR(fs_devices);
6602 device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
6604 if (!btrfs_test_opt(root->fs_info, DEGRADED))
6607 device = add_missing_dev(root, fs_devices, devid, dev_uuid);
6610 btrfs_warn(root->fs_info, "devid %llu uuid %pU missing",
6613 if (!device->bdev && !btrfs_test_opt(root->fs_info, DEGRADED))
6616 if(!device->bdev && !device->missing) {
6618 * this happens when a device that was properly setup
6619 * in the device info lists suddenly goes bad.
6620 * device->bdev is NULL, and so we have to set
6621 * device->missing to one here
6623 device->fs_devices->missing_devices++;
6624 device->missing = 1;
6627 /* Move the device to its own fs_devices */
6628 if (device->fs_devices != fs_devices) {
6629 ASSERT(device->missing);
6631 list_move(&device->dev_list, &fs_devices->devices);
6632 device->fs_devices->num_devices--;
6633 fs_devices->num_devices++;
6635 device->fs_devices->missing_devices--;
6636 fs_devices->missing_devices++;
6638 device->fs_devices = fs_devices;
6642 if (device->fs_devices != root->fs_info->fs_devices) {
6643 BUG_ON(device->writeable);
6644 if (device->generation !=
6645 btrfs_device_generation(leaf, dev_item))
6649 fill_device_from_item(leaf, dev_item, device);
6650 device->in_fs_metadata = 1;
6651 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
6652 device->fs_devices->total_rw_bytes += device->total_bytes;
6653 spin_lock(&root->fs_info->free_chunk_lock);
6654 root->fs_info->free_chunk_space += device->total_bytes -
6656 spin_unlock(&root->fs_info->free_chunk_lock);
6662 int btrfs_read_sys_array(struct btrfs_root *root)
6664 struct btrfs_fs_info *fs_info = root->fs_info;
6665 struct btrfs_super_block *super_copy = fs_info->super_copy;
6666 struct extent_buffer *sb;
6667 struct btrfs_disk_key *disk_key;
6668 struct btrfs_chunk *chunk;
6670 unsigned long sb_array_offset;
6677 struct btrfs_key key;
6679 ASSERT(BTRFS_SUPER_INFO_SIZE <= root->nodesize);
6681 * This will create extent buffer of nodesize, superblock size is
6682 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6683 * overallocate but we can keep it as-is, only the first page is used.
6685 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET);
6688 set_extent_buffer_uptodate(sb);
6689 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6691 * The sb extent buffer is artificial and just used to read the system array.
6692 * set_extent_buffer_uptodate() call does not properly mark all it's
6693 * pages up-to-date when the page is larger: extent does not cover the
6694 * whole page and consequently check_page_uptodate does not find all
6695 * the page's extents up-to-date (the hole beyond sb),
6696 * write_extent_buffer then triggers a WARN_ON.
6698 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6699 * but sb spans only this function. Add an explicit SetPageUptodate call
6700 * to silence the warning eg. on PowerPC 64.
6702 if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
6703 SetPageUptodate(sb->pages[0]);
6705 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6706 array_size = btrfs_super_sys_array_size(super_copy);
6708 array_ptr = super_copy->sys_chunk_array;
6709 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6712 while (cur_offset < array_size) {
6713 disk_key = (struct btrfs_disk_key *)array_ptr;
6714 len = sizeof(*disk_key);
6715 if (cur_offset + len > array_size)
6716 goto out_short_read;
6718 btrfs_disk_key_to_cpu(&key, disk_key);
6721 sb_array_offset += len;
6724 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6725 chunk = (struct btrfs_chunk *)sb_array_offset;
6727 * At least one btrfs_chunk with one stripe must be
6728 * present, exact stripe count check comes afterwards
6730 len = btrfs_chunk_item_size(1);
6731 if (cur_offset + len > array_size)
6732 goto out_short_read;
6734 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6737 "invalid number of stripes %u in sys_array at offset %u",
6738 num_stripes, cur_offset);
6743 type = btrfs_chunk_type(sb, chunk);
6744 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
6746 "invalid chunk type %llu in sys_array at offset %u",
6752 len = btrfs_chunk_item_size(num_stripes);
6753 if (cur_offset + len > array_size)
6754 goto out_short_read;
6756 ret = read_one_chunk(root, &key, sb, chunk);
6761 "unexpected item type %u in sys_array at offset %u",
6762 (u32)key.type, cur_offset);
6767 sb_array_offset += len;
6770 clear_extent_buffer_uptodate(sb);
6771 free_extent_buffer_stale(sb);
6775 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
6777 clear_extent_buffer_uptodate(sb);
6778 free_extent_buffer_stale(sb);
6782 int btrfs_read_chunk_tree(struct btrfs_root *root)
6784 struct btrfs_path *path;
6785 struct extent_buffer *leaf;
6786 struct btrfs_key key;
6787 struct btrfs_key found_key;
6792 root = root->fs_info->chunk_root;
6794 path = btrfs_alloc_path();
6798 mutex_lock(&uuid_mutex);
6802 * Read all device items, and then all the chunk items. All
6803 * device items are found before any chunk item (their object id
6804 * is smaller than the lowest possible object id for a chunk
6805 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6807 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6810 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6814 leaf = path->nodes[0];
6815 slot = path->slots[0];
6816 if (slot >= btrfs_header_nritems(leaf)) {
6817 ret = btrfs_next_leaf(root, path);
6824 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6825 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6826 struct btrfs_dev_item *dev_item;
6827 dev_item = btrfs_item_ptr(leaf, slot,
6828 struct btrfs_dev_item);
6829 ret = read_one_dev(root, leaf, dev_item);
6833 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6834 struct btrfs_chunk *chunk;
6835 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6836 ret = read_one_chunk(root, &found_key, leaf, chunk);
6844 * After loading chunk tree, we've got all device information,
6845 * do another round of validation checks.
6847 if (total_dev != root->fs_info->fs_devices->total_devices) {
6848 btrfs_err(root->fs_info,
6849 "super_num_devices %llu mismatch with num_devices %llu found here",
6850 btrfs_super_num_devices(root->fs_info->super_copy),
6855 if (btrfs_super_total_bytes(root->fs_info->super_copy) <
6856 root->fs_info->fs_devices->total_rw_bytes) {
6857 btrfs_err(root->fs_info,
6858 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
6859 btrfs_super_total_bytes(root->fs_info->super_copy),
6860 root->fs_info->fs_devices->total_rw_bytes);
6866 unlock_chunks(root);
6867 mutex_unlock(&uuid_mutex);
6869 btrfs_free_path(path);
6873 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6875 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6876 struct btrfs_device *device;
6878 while (fs_devices) {
6879 mutex_lock(&fs_devices->device_list_mutex);
6880 list_for_each_entry(device, &fs_devices->devices, dev_list)
6881 device->dev_root = fs_info->dev_root;
6882 mutex_unlock(&fs_devices->device_list_mutex);
6884 fs_devices = fs_devices->seed;
6888 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6892 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6893 btrfs_dev_stat_reset(dev, i);
6896 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6898 struct btrfs_key key;
6899 struct btrfs_key found_key;
6900 struct btrfs_root *dev_root = fs_info->dev_root;
6901 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6902 struct extent_buffer *eb;
6905 struct btrfs_device *device;
6906 struct btrfs_path *path = NULL;
6909 path = btrfs_alloc_path();
6915 mutex_lock(&fs_devices->device_list_mutex);
6916 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6918 struct btrfs_dev_stats_item *ptr;
6920 key.objectid = BTRFS_DEV_STATS_OBJECTID;
6921 key.type = BTRFS_PERSISTENT_ITEM_KEY;
6922 key.offset = device->devid;
6923 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6925 __btrfs_reset_dev_stats(device);
6926 device->dev_stats_valid = 1;
6927 btrfs_release_path(path);
6930 slot = path->slots[0];
6931 eb = path->nodes[0];
6932 btrfs_item_key_to_cpu(eb, &found_key, slot);
6933 item_size = btrfs_item_size_nr(eb, slot);
6935 ptr = btrfs_item_ptr(eb, slot,
6936 struct btrfs_dev_stats_item);
6938 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6939 if (item_size >= (1 + i) * sizeof(__le64))
6940 btrfs_dev_stat_set(device, i,
6941 btrfs_dev_stats_value(eb, ptr, i));
6943 btrfs_dev_stat_reset(device, i);
6946 device->dev_stats_valid = 1;
6947 btrfs_dev_stat_print_on_load(device);
6948 btrfs_release_path(path);
6950 mutex_unlock(&fs_devices->device_list_mutex);
6953 btrfs_free_path(path);
6954 return ret < 0 ? ret : 0;
6957 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
6958 struct btrfs_root *dev_root,
6959 struct btrfs_device *device)
6961 struct btrfs_path *path;
6962 struct btrfs_key key;
6963 struct extent_buffer *eb;
6964 struct btrfs_dev_stats_item *ptr;
6968 key.objectid = BTRFS_DEV_STATS_OBJECTID;
6969 key.type = BTRFS_PERSISTENT_ITEM_KEY;
6970 key.offset = device->devid;
6972 path = btrfs_alloc_path();
6974 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
6976 btrfs_warn_in_rcu(dev_root->fs_info,
6977 "error %d while searching for dev_stats item for device %s",
6978 ret, rcu_str_deref(device->name));
6983 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
6984 /* need to delete old one and insert a new one */
6985 ret = btrfs_del_item(trans, dev_root, path);
6987 btrfs_warn_in_rcu(dev_root->fs_info,
6988 "delete too small dev_stats item for device %s failed %d",
6989 rcu_str_deref(device->name), ret);
6996 /* need to insert a new item */
6997 btrfs_release_path(path);
6998 ret = btrfs_insert_empty_item(trans, dev_root, path,
6999 &key, sizeof(*ptr));
7001 btrfs_warn_in_rcu(dev_root->fs_info,
7002 "insert dev_stats item for device %s failed %d",
7003 rcu_str_deref(device->name), ret);
7008 eb = path->nodes[0];
7009 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7010 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7011 btrfs_set_dev_stats_value(eb, ptr, i,
7012 btrfs_dev_stat_read(device, i));
7013 btrfs_mark_buffer_dirty(eb);
7016 btrfs_free_path(path);
7021 * called from commit_transaction. Writes all changed device stats to disk.
7023 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
7024 struct btrfs_fs_info *fs_info)
7026 struct btrfs_root *dev_root = fs_info->dev_root;
7027 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7028 struct btrfs_device *device;
7032 mutex_lock(&fs_devices->device_list_mutex);
7033 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7034 if (!device->dev_stats_valid || !btrfs_dev_stats_dirty(device))
7037 stats_cnt = atomic_read(&device->dev_stats_ccnt);
7038 ret = update_dev_stat_item(trans, dev_root, device);
7040 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7042 mutex_unlock(&fs_devices->device_list_mutex);
7047 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7049 btrfs_dev_stat_inc(dev, index);
7050 btrfs_dev_stat_print_on_error(dev);
7053 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
7055 if (!dev->dev_stats_valid)
7057 btrfs_err_rl_in_rcu(dev->dev_root->fs_info,
7058 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7059 rcu_str_deref(dev->name),
7060 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7061 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7062 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7063 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7064 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7067 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7071 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7072 if (btrfs_dev_stat_read(dev, i) != 0)
7074 if (i == BTRFS_DEV_STAT_VALUES_MAX)
7075 return; /* all values == 0, suppress message */
7077 btrfs_info_in_rcu(dev->dev_root->fs_info,
7078 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7079 rcu_str_deref(dev->name),
7080 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7081 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7082 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7083 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7084 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7087 int btrfs_get_dev_stats(struct btrfs_root *root,
7088 struct btrfs_ioctl_get_dev_stats *stats)
7090 struct btrfs_device *dev;
7091 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7094 mutex_lock(&fs_devices->device_list_mutex);
7095 dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
7096 mutex_unlock(&fs_devices->device_list_mutex);
7099 btrfs_warn(root->fs_info,
7100 "get dev_stats failed, device not found");
7102 } else if (!dev->dev_stats_valid) {
7103 btrfs_warn(root->fs_info,
7104 "get dev_stats failed, not yet valid");
7106 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7107 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7108 if (stats->nr_items > i)
7110 btrfs_dev_stat_read_and_reset(dev, i);
7112 btrfs_dev_stat_reset(dev, i);
7115 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7116 if (stats->nr_items > i)
7117 stats->values[i] = btrfs_dev_stat_read(dev, i);
7119 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7120 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7124 void btrfs_scratch_superblocks(struct block_device *bdev, char *device_path)
7126 struct buffer_head *bh;
7127 struct btrfs_super_block *disk_super;
7133 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX;
7136 if (btrfs_read_dev_one_super(bdev, copy_num, &bh))
7139 disk_super = (struct btrfs_super_block *)bh->b_data;
7141 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
7142 set_buffer_dirty(bh);
7143 sync_dirty_buffer(bh);
7147 /* Notify udev that device has changed */
7148 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
7150 /* Update ctime/mtime for device path for libblkid */
7151 update_dev_time(device_path);
7155 * Update the size of all devices, which is used for writing out the
7158 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
7160 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7161 struct btrfs_device *curr, *next;
7163 if (list_empty(&fs_devices->resized_devices))
7166 mutex_lock(&fs_devices->device_list_mutex);
7167 lock_chunks(fs_info->dev_root);
7168 list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
7170 list_del_init(&curr->resized_list);
7171 curr->commit_total_bytes = curr->disk_total_bytes;
7173 unlock_chunks(fs_info->dev_root);
7174 mutex_unlock(&fs_devices->device_list_mutex);
7177 /* Must be invoked during the transaction commit */
7178 void btrfs_update_commit_device_bytes_used(struct btrfs_root *root,
7179 struct btrfs_transaction *transaction)
7181 struct extent_map *em;
7182 struct map_lookup *map;
7183 struct btrfs_device *dev;
7186 if (list_empty(&transaction->pending_chunks))
7189 /* In order to kick the device replace finish process */
7191 list_for_each_entry(em, &transaction->pending_chunks, list) {
7192 map = em->map_lookup;
7194 for (i = 0; i < map->num_stripes; i++) {
7195 dev = map->stripes[i].dev;
7196 dev->commit_bytes_used = dev->bytes_used;
7199 unlock_chunks(root);
7202 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
7204 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7205 while (fs_devices) {
7206 fs_devices->fs_info = fs_info;
7207 fs_devices = fs_devices->seed;
7211 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
7213 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7214 while (fs_devices) {
7215 fs_devices->fs_info = NULL;
7216 fs_devices = fs_devices->seed;