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->bi_rw, 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);
839 blkdev_put(device->bdev, device->mode);
841 rcu_string_free(device->name);
845 static void free_device(struct rcu_head *head)
847 struct btrfs_device *device;
849 device = container_of(head, struct btrfs_device, rcu);
851 INIT_WORK(&device->rcu_work, __free_device);
852 schedule_work(&device->rcu_work);
855 static void btrfs_close_one_device(struct btrfs_device *device)
857 struct btrfs_fs_devices *fs_devices = device->fs_devices;
858 struct btrfs_device *new_device;
859 struct rcu_string *name;
862 fs_devices->open_devices--;
864 if (device->writeable &&
865 device->devid != BTRFS_DEV_REPLACE_DEVID) {
866 list_del_init(&device->dev_alloc_list);
867 fs_devices->rw_devices--;
871 fs_devices->missing_devices--;
873 if (device->bdev && device->writeable) {
874 sync_blockdev(device->bdev);
875 invalidate_bdev(device->bdev);
878 new_device = btrfs_alloc_device(NULL, &device->devid,
880 BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
882 /* Safe because we are under uuid_mutex */
884 name = rcu_string_strdup(device->name->str, GFP_NOFS);
885 BUG_ON(!name); /* -ENOMEM */
886 rcu_assign_pointer(new_device->name, name);
889 list_replace_rcu(&device->dev_list, &new_device->dev_list);
890 new_device->fs_devices = device->fs_devices;
892 call_rcu(&device->rcu, free_device);
895 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
897 struct btrfs_device *device, *tmp;
899 if (--fs_devices->opened > 0)
902 mutex_lock(&fs_devices->device_list_mutex);
903 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
904 btrfs_close_one_device(device);
906 mutex_unlock(&fs_devices->device_list_mutex);
908 WARN_ON(fs_devices->open_devices);
909 WARN_ON(fs_devices->rw_devices);
910 fs_devices->opened = 0;
911 fs_devices->seeding = 0;
916 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
918 struct btrfs_fs_devices *seed_devices = NULL;
921 mutex_lock(&uuid_mutex);
922 ret = __btrfs_close_devices(fs_devices);
923 if (!fs_devices->opened) {
924 seed_devices = fs_devices->seed;
925 fs_devices->seed = NULL;
927 mutex_unlock(&uuid_mutex);
929 while (seed_devices) {
930 fs_devices = seed_devices;
931 seed_devices = fs_devices->seed;
932 __btrfs_close_devices(fs_devices);
933 free_fs_devices(fs_devices);
936 * Wait for rcu kworkers under __btrfs_close_devices
937 * to finish all blkdev_puts so device is really
938 * free when umount is done.
944 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
945 fmode_t flags, void *holder)
947 struct request_queue *q;
948 struct block_device *bdev;
949 struct list_head *head = &fs_devices->devices;
950 struct btrfs_device *device;
951 struct btrfs_device *latest_dev = NULL;
952 struct buffer_head *bh;
953 struct btrfs_super_block *disk_super;
960 list_for_each_entry(device, head, dev_list) {
966 /* Just open everything we can; ignore failures here */
967 if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
971 disk_super = (struct btrfs_super_block *)bh->b_data;
972 devid = btrfs_stack_device_id(&disk_super->dev_item);
973 if (devid != device->devid)
976 if (memcmp(device->uuid, disk_super->dev_item.uuid,
980 device->generation = btrfs_super_generation(disk_super);
982 device->generation > latest_dev->generation)
985 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
986 device->writeable = 0;
988 device->writeable = !bdev_read_only(bdev);
992 q = bdev_get_queue(bdev);
993 if (blk_queue_discard(q))
994 device->can_discard = 1;
997 device->in_fs_metadata = 0;
998 device->mode = flags;
1000 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1001 fs_devices->rotating = 1;
1003 fs_devices->open_devices++;
1004 if (device->writeable &&
1005 device->devid != BTRFS_DEV_REPLACE_DEVID) {
1006 fs_devices->rw_devices++;
1007 list_add(&device->dev_alloc_list,
1008 &fs_devices->alloc_list);
1015 blkdev_put(bdev, flags);
1018 if (fs_devices->open_devices == 0) {
1022 fs_devices->seeding = seeding;
1023 fs_devices->opened = 1;
1024 fs_devices->latest_bdev = latest_dev->bdev;
1025 fs_devices->total_rw_bytes = 0;
1030 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1031 fmode_t flags, void *holder)
1035 mutex_lock(&uuid_mutex);
1036 if (fs_devices->opened) {
1037 fs_devices->opened++;
1040 ret = __btrfs_open_devices(fs_devices, flags, holder);
1042 mutex_unlock(&uuid_mutex);
1046 void btrfs_release_disk_super(struct page *page)
1052 int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr,
1053 struct page **page, struct btrfs_super_block **disk_super)
1058 /* make sure our super fits in the device */
1059 if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1062 /* make sure our super fits in the page */
1063 if (sizeof(**disk_super) > PAGE_SIZE)
1066 /* make sure our super doesn't straddle pages on disk */
1067 index = bytenr >> PAGE_SHIFT;
1068 if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index)
1071 /* pull in the page with our super */
1072 *page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
1075 if (IS_ERR_OR_NULL(*page))
1080 /* align our pointer to the offset of the super block */
1081 *disk_super = p + (bytenr & ~PAGE_MASK);
1083 if (btrfs_super_bytenr(*disk_super) != bytenr ||
1084 btrfs_super_magic(*disk_super) != BTRFS_MAGIC) {
1085 btrfs_release_disk_super(*page);
1089 if ((*disk_super)->label[0] &&
1090 (*disk_super)->label[BTRFS_LABEL_SIZE - 1])
1091 (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0';
1097 * Look for a btrfs signature on a device. This may be called out of the mount path
1098 * and we are not allowed to call set_blocksize during the scan. The superblock
1099 * is read via pagecache
1101 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
1102 struct btrfs_fs_devices **fs_devices_ret)
1104 struct btrfs_super_block *disk_super;
1105 struct block_device *bdev;
1114 * we would like to check all the supers, but that would make
1115 * a btrfs mount succeed after a mkfs from a different FS.
1116 * So, we need to add a special mount option to scan for
1117 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1119 bytenr = btrfs_sb_offset(0);
1120 flags |= FMODE_EXCL;
1121 mutex_lock(&uuid_mutex);
1123 bdev = blkdev_get_by_path(path, flags, holder);
1125 ret = PTR_ERR(bdev);
1129 if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super))
1130 goto error_bdev_put;
1132 devid = btrfs_stack_device_id(&disk_super->dev_item);
1133 transid = btrfs_super_generation(disk_super);
1134 total_devices = btrfs_super_num_devices(disk_super);
1136 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
1138 if (disk_super->label[0]) {
1139 printk(KERN_INFO "BTRFS: device label %s ", disk_super->label);
1141 printk(KERN_INFO "BTRFS: device fsid %pU ", disk_super->fsid);
1144 printk(KERN_CONT "devid %llu transid %llu %s\n", devid, transid, path);
1147 if (!ret && fs_devices_ret)
1148 (*fs_devices_ret)->total_devices = total_devices;
1150 btrfs_release_disk_super(page);
1153 blkdev_put(bdev, flags);
1155 mutex_unlock(&uuid_mutex);
1159 /* helper to account the used device space in the range */
1160 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
1161 u64 end, u64 *length)
1163 struct btrfs_key key;
1164 struct btrfs_root *root = device->dev_root;
1165 struct btrfs_dev_extent *dev_extent;
1166 struct btrfs_path *path;
1170 struct extent_buffer *l;
1174 if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
1177 path = btrfs_alloc_path();
1180 path->reada = READA_FORWARD;
1182 key.objectid = device->devid;
1184 key.type = BTRFS_DEV_EXTENT_KEY;
1186 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1190 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1197 slot = path->slots[0];
1198 if (slot >= btrfs_header_nritems(l)) {
1199 ret = btrfs_next_leaf(root, path);
1207 btrfs_item_key_to_cpu(l, &key, slot);
1209 if (key.objectid < device->devid)
1212 if (key.objectid > device->devid)
1215 if (key.type != BTRFS_DEV_EXTENT_KEY)
1218 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1219 extent_end = key.offset + btrfs_dev_extent_length(l,
1221 if (key.offset <= start && extent_end > end) {
1222 *length = end - start + 1;
1224 } else if (key.offset <= start && extent_end > start)
1225 *length += extent_end - start;
1226 else if (key.offset > start && extent_end <= end)
1227 *length += extent_end - key.offset;
1228 else if (key.offset > start && key.offset <= end) {
1229 *length += end - key.offset + 1;
1231 } else if (key.offset > end)
1239 btrfs_free_path(path);
1243 static int contains_pending_extent(struct btrfs_transaction *transaction,
1244 struct btrfs_device *device,
1245 u64 *start, u64 len)
1247 struct btrfs_fs_info *fs_info = device->dev_root->fs_info;
1248 struct extent_map *em;
1249 struct list_head *search_list = &fs_info->pinned_chunks;
1251 u64 physical_start = *start;
1254 search_list = &transaction->pending_chunks;
1256 list_for_each_entry(em, search_list, list) {
1257 struct map_lookup *map;
1260 map = em->map_lookup;
1261 for (i = 0; i < map->num_stripes; i++) {
1264 if (map->stripes[i].dev != device)
1266 if (map->stripes[i].physical >= physical_start + len ||
1267 map->stripes[i].physical + em->orig_block_len <=
1271 * Make sure that while processing the pinned list we do
1272 * not override our *start with a lower value, because
1273 * we can have pinned chunks that fall within this
1274 * device hole and that have lower physical addresses
1275 * than the pending chunks we processed before. If we
1276 * do not take this special care we can end up getting
1277 * 2 pending chunks that start at the same physical
1278 * device offsets because the end offset of a pinned
1279 * chunk can be equal to the start offset of some
1282 end = map->stripes[i].physical + em->orig_block_len;
1289 if (search_list != &fs_info->pinned_chunks) {
1290 search_list = &fs_info->pinned_chunks;
1299 * find_free_dev_extent_start - find free space in the specified device
1300 * @device: the device which we search the free space in
1301 * @num_bytes: the size of the free space that we need
1302 * @search_start: the position from which to begin the search
1303 * @start: store the start of the free space.
1304 * @len: the size of the free space. that we find, or the size
1305 * of the max free space if we don't find suitable free space
1307 * this uses a pretty simple search, the expectation is that it is
1308 * called very infrequently and that a given device has a small number
1311 * @start is used to store the start of the free space if we find. But if we
1312 * don't find suitable free space, it will be used to store the start position
1313 * of the max free space.
1315 * @len is used to store the size of the free space that we find.
1316 * But if we don't find suitable free space, it is used to store the size of
1317 * the max free space.
1319 int find_free_dev_extent_start(struct btrfs_transaction *transaction,
1320 struct btrfs_device *device, u64 num_bytes,
1321 u64 search_start, u64 *start, u64 *len)
1323 struct btrfs_key key;
1324 struct btrfs_root *root = device->dev_root;
1325 struct btrfs_dev_extent *dev_extent;
1326 struct btrfs_path *path;
1331 u64 search_end = device->total_bytes;
1334 struct extent_buffer *l;
1335 u64 min_search_start;
1338 * We don't want to overwrite the superblock on the drive nor any area
1339 * used by the boot loader (grub for example), so we make sure to start
1340 * at an offset of at least 1MB.
1342 min_search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
1343 search_start = max(search_start, min_search_start);
1345 path = btrfs_alloc_path();
1349 max_hole_start = search_start;
1353 if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1358 path->reada = READA_FORWARD;
1359 path->search_commit_root = 1;
1360 path->skip_locking = 1;
1362 key.objectid = device->devid;
1363 key.offset = search_start;
1364 key.type = BTRFS_DEV_EXTENT_KEY;
1366 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1370 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1377 slot = path->slots[0];
1378 if (slot >= btrfs_header_nritems(l)) {
1379 ret = btrfs_next_leaf(root, path);
1387 btrfs_item_key_to_cpu(l, &key, slot);
1389 if (key.objectid < device->devid)
1392 if (key.objectid > device->devid)
1395 if (key.type != BTRFS_DEV_EXTENT_KEY)
1398 if (key.offset > search_start) {
1399 hole_size = key.offset - search_start;
1402 * Have to check before we set max_hole_start, otherwise
1403 * we could end up sending back this offset anyway.
1405 if (contains_pending_extent(transaction, device,
1408 if (key.offset >= search_start) {
1409 hole_size = key.offset - search_start;
1416 if (hole_size > max_hole_size) {
1417 max_hole_start = search_start;
1418 max_hole_size = hole_size;
1422 * If this free space is greater than which we need,
1423 * it must be the max free space that we have found
1424 * until now, so max_hole_start must point to the start
1425 * of this free space and the length of this free space
1426 * is stored in max_hole_size. Thus, we return
1427 * max_hole_start and max_hole_size and go back to the
1430 if (hole_size >= num_bytes) {
1436 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1437 extent_end = key.offset + btrfs_dev_extent_length(l,
1439 if (extent_end > search_start)
1440 search_start = extent_end;
1447 * At this point, search_start should be the end of
1448 * allocated dev extents, and when shrinking the device,
1449 * search_end may be smaller than search_start.
1451 if (search_end > search_start) {
1452 hole_size = search_end - search_start;
1454 if (contains_pending_extent(transaction, device, &search_start,
1456 btrfs_release_path(path);
1460 if (hole_size > max_hole_size) {
1461 max_hole_start = search_start;
1462 max_hole_size = hole_size;
1467 if (max_hole_size < num_bytes)
1473 btrfs_free_path(path);
1474 *start = max_hole_start;
1476 *len = max_hole_size;
1480 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1481 struct btrfs_device *device, u64 num_bytes,
1482 u64 *start, u64 *len)
1484 /* FIXME use last free of some kind */
1485 return find_free_dev_extent_start(trans->transaction, device,
1486 num_bytes, 0, start, len);
1489 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1490 struct btrfs_device *device,
1491 u64 start, u64 *dev_extent_len)
1494 struct btrfs_path *path;
1495 struct btrfs_root *root = device->dev_root;
1496 struct btrfs_key key;
1497 struct btrfs_key found_key;
1498 struct extent_buffer *leaf = NULL;
1499 struct btrfs_dev_extent *extent = NULL;
1501 path = btrfs_alloc_path();
1505 key.objectid = device->devid;
1507 key.type = BTRFS_DEV_EXTENT_KEY;
1509 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1511 ret = btrfs_previous_item(root, path, key.objectid,
1512 BTRFS_DEV_EXTENT_KEY);
1515 leaf = path->nodes[0];
1516 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1517 extent = btrfs_item_ptr(leaf, path->slots[0],
1518 struct btrfs_dev_extent);
1519 BUG_ON(found_key.offset > start || found_key.offset +
1520 btrfs_dev_extent_length(leaf, extent) < start);
1522 btrfs_release_path(path);
1524 } else if (ret == 0) {
1525 leaf = path->nodes[0];
1526 extent = btrfs_item_ptr(leaf, path->slots[0],
1527 struct btrfs_dev_extent);
1529 btrfs_handle_fs_error(root->fs_info, ret, "Slot search failed");
1533 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1535 ret = btrfs_del_item(trans, root, path);
1537 btrfs_handle_fs_error(root->fs_info, ret,
1538 "Failed to remove dev extent item");
1540 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1543 btrfs_free_path(path);
1547 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1548 struct btrfs_device *device,
1549 u64 chunk_tree, u64 chunk_objectid,
1550 u64 chunk_offset, u64 start, u64 num_bytes)
1553 struct btrfs_path *path;
1554 struct btrfs_root *root = device->dev_root;
1555 struct btrfs_dev_extent *extent;
1556 struct extent_buffer *leaf;
1557 struct btrfs_key key;
1559 WARN_ON(!device->in_fs_metadata);
1560 WARN_ON(device->is_tgtdev_for_dev_replace);
1561 path = btrfs_alloc_path();
1565 key.objectid = device->devid;
1567 key.type = BTRFS_DEV_EXTENT_KEY;
1568 ret = btrfs_insert_empty_item(trans, root, path, &key,
1573 leaf = path->nodes[0];
1574 extent = btrfs_item_ptr(leaf, path->slots[0],
1575 struct btrfs_dev_extent);
1576 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1577 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1578 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1580 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1581 btrfs_dev_extent_chunk_tree_uuid(extent), BTRFS_UUID_SIZE);
1583 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1584 btrfs_mark_buffer_dirty(leaf);
1586 btrfs_free_path(path);
1590 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1592 struct extent_map_tree *em_tree;
1593 struct extent_map *em;
1597 em_tree = &fs_info->mapping_tree.map_tree;
1598 read_lock(&em_tree->lock);
1599 n = rb_last(&em_tree->map);
1601 em = rb_entry(n, struct extent_map, rb_node);
1602 ret = em->start + em->len;
1604 read_unlock(&em_tree->lock);
1609 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1613 struct btrfs_key key;
1614 struct btrfs_key found_key;
1615 struct btrfs_path *path;
1617 path = btrfs_alloc_path();
1621 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1622 key.type = BTRFS_DEV_ITEM_KEY;
1623 key.offset = (u64)-1;
1625 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1629 BUG_ON(ret == 0); /* Corruption */
1631 ret = btrfs_previous_item(fs_info->chunk_root, path,
1632 BTRFS_DEV_ITEMS_OBJECTID,
1633 BTRFS_DEV_ITEM_KEY);
1637 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1639 *devid_ret = found_key.offset + 1;
1643 btrfs_free_path(path);
1648 * the device information is stored in the chunk root
1649 * the btrfs_device struct should be fully filled in
1651 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1652 struct btrfs_root *root,
1653 struct btrfs_device *device)
1656 struct btrfs_path *path;
1657 struct btrfs_dev_item *dev_item;
1658 struct extent_buffer *leaf;
1659 struct btrfs_key key;
1662 root = root->fs_info->chunk_root;
1664 path = btrfs_alloc_path();
1668 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1669 key.type = BTRFS_DEV_ITEM_KEY;
1670 key.offset = device->devid;
1672 ret = btrfs_insert_empty_item(trans, root, path, &key,
1677 leaf = path->nodes[0];
1678 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1680 btrfs_set_device_id(leaf, dev_item, device->devid);
1681 btrfs_set_device_generation(leaf, dev_item, 0);
1682 btrfs_set_device_type(leaf, dev_item, device->type);
1683 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1684 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1685 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1686 btrfs_set_device_total_bytes(leaf, dev_item,
1687 btrfs_device_get_disk_total_bytes(device));
1688 btrfs_set_device_bytes_used(leaf, dev_item,
1689 btrfs_device_get_bytes_used(device));
1690 btrfs_set_device_group(leaf, dev_item, 0);
1691 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1692 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1693 btrfs_set_device_start_offset(leaf, dev_item, 0);
1695 ptr = btrfs_device_uuid(dev_item);
1696 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1697 ptr = btrfs_device_fsid(dev_item);
1698 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1699 btrfs_mark_buffer_dirty(leaf);
1703 btrfs_free_path(path);
1708 * Function to update ctime/mtime for a given device path.
1709 * Mainly used for ctime/mtime based probe like libblkid.
1711 static void update_dev_time(char *path_name)
1715 filp = filp_open(path_name, O_RDWR, 0);
1718 file_update_time(filp);
1719 filp_close(filp, NULL);
1722 static int btrfs_rm_dev_item(struct btrfs_root *root,
1723 struct btrfs_device *device)
1726 struct btrfs_path *path;
1727 struct btrfs_key key;
1728 struct btrfs_trans_handle *trans;
1730 root = root->fs_info->chunk_root;
1732 path = btrfs_alloc_path();
1736 trans = btrfs_start_transaction(root, 0);
1737 if (IS_ERR(trans)) {
1738 btrfs_free_path(path);
1739 return PTR_ERR(trans);
1741 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1742 key.type = BTRFS_DEV_ITEM_KEY;
1743 key.offset = device->devid;
1745 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1754 ret = btrfs_del_item(trans, root, path);
1758 btrfs_free_path(path);
1759 btrfs_commit_transaction(trans, root);
1764 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1765 * filesystem. It's up to the caller to adjust that number regarding eg. device
1768 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1776 seq = read_seqbegin(&fs_info->profiles_lock);
1778 all_avail = fs_info->avail_data_alloc_bits |
1779 fs_info->avail_system_alloc_bits |
1780 fs_info->avail_metadata_alloc_bits;
1781 } while (read_seqretry(&fs_info->profiles_lock, seq));
1783 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1784 if (!(all_avail & btrfs_raid_group[i]))
1787 if (num_devices < btrfs_raid_array[i].devs_min) {
1788 int ret = btrfs_raid_mindev_error[i];
1798 struct btrfs_device *btrfs_find_next_active_device(struct btrfs_fs_devices *fs_devs,
1799 struct btrfs_device *device)
1801 struct btrfs_device *next_device;
1803 list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1804 if (next_device != device &&
1805 !next_device->missing && next_device->bdev)
1813 * Helper function to check if the given device is part of s_bdev / latest_bdev
1814 * and replace it with the provided or the next active device, in the context
1815 * where this function called, there should be always be another device (or
1816 * this_dev) which is active.
1818 void btrfs_assign_next_active_device(struct btrfs_fs_info *fs_info,
1819 struct btrfs_device *device, struct btrfs_device *this_dev)
1821 struct btrfs_device *next_device;
1824 next_device = this_dev;
1826 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
1828 ASSERT(next_device);
1830 if (fs_info->sb->s_bdev &&
1831 (fs_info->sb->s_bdev == device->bdev))
1832 fs_info->sb->s_bdev = next_device->bdev;
1834 if (fs_info->fs_devices->latest_bdev == device->bdev)
1835 fs_info->fs_devices->latest_bdev = next_device->bdev;
1838 int btrfs_rm_device(struct btrfs_root *root, char *device_path, u64 devid)
1840 struct btrfs_device *device;
1841 struct btrfs_fs_devices *cur_devices;
1844 bool clear_super = false;
1845 char *dev_name = NULL;
1847 mutex_lock(&uuid_mutex);
1849 num_devices = root->fs_info->fs_devices->num_devices;
1850 btrfs_dev_replace_lock(&root->fs_info->dev_replace, 0);
1851 if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1852 WARN_ON(num_devices < 1);
1855 btrfs_dev_replace_unlock(&root->fs_info->dev_replace, 0);
1857 ret = btrfs_check_raid_min_devices(root->fs_info, num_devices - 1);
1861 ret = btrfs_find_device_by_devspec(root, devid, device_path,
1866 if (device->is_tgtdev_for_dev_replace) {
1867 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1871 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1872 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1876 if (device->writeable) {
1878 list_del_init(&device->dev_alloc_list);
1879 device->fs_devices->rw_devices--;
1880 unlock_chunks(root);
1881 dev_name = kstrdup(device->name->str, GFP_KERNEL);
1889 mutex_unlock(&uuid_mutex);
1890 ret = btrfs_shrink_device(device, 0);
1891 mutex_lock(&uuid_mutex);
1896 * TODO: the superblock still includes this device in its num_devices
1897 * counter although write_all_supers() is not locked out. This
1898 * could give a filesystem state which requires a degraded mount.
1900 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1904 device->in_fs_metadata = 0;
1905 btrfs_scrub_cancel_dev(root->fs_info, device);
1908 * the device list mutex makes sure that we don't change
1909 * the device list while someone else is writing out all
1910 * the device supers. Whoever is writing all supers, should
1911 * lock the device list mutex before getting the number of
1912 * devices in the super block (super_copy). Conversely,
1913 * whoever updates the number of devices in the super block
1914 * (super_copy) should hold the device list mutex.
1917 cur_devices = device->fs_devices;
1918 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1919 list_del_rcu(&device->dev_list);
1921 device->fs_devices->num_devices--;
1922 device->fs_devices->total_devices--;
1924 if (device->missing)
1925 device->fs_devices->missing_devices--;
1927 btrfs_assign_next_active_device(root->fs_info, device, NULL);
1930 device->fs_devices->open_devices--;
1931 /* remove sysfs entry */
1932 btrfs_sysfs_rm_device_link(root->fs_info->fs_devices, device);
1935 call_rcu(&device->rcu, free_device);
1937 num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1938 btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1939 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1941 if (cur_devices->open_devices == 0) {
1942 struct btrfs_fs_devices *fs_devices;
1943 fs_devices = root->fs_info->fs_devices;
1944 while (fs_devices) {
1945 if (fs_devices->seed == cur_devices) {
1946 fs_devices->seed = cur_devices->seed;
1949 fs_devices = fs_devices->seed;
1951 cur_devices->seed = NULL;
1952 __btrfs_close_devices(cur_devices);
1953 free_fs_devices(cur_devices);
1956 root->fs_info->num_tolerated_disk_barrier_failures =
1957 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1960 * at this point, the device is zero sized. We want to
1961 * remove it from the devices list and zero out the old super
1964 struct block_device *bdev;
1966 bdev = blkdev_get_by_path(dev_name, FMODE_READ | FMODE_EXCL,
1967 root->fs_info->bdev_holder);
1968 if (!IS_ERR(bdev)) {
1969 btrfs_scratch_superblocks(bdev, dev_name);
1970 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1977 mutex_unlock(&uuid_mutex);
1981 if (device->writeable) {
1983 list_add(&device->dev_alloc_list,
1984 &root->fs_info->fs_devices->alloc_list);
1985 device->fs_devices->rw_devices++;
1986 unlock_chunks(root);
1991 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
1992 struct btrfs_device *srcdev)
1994 struct btrfs_fs_devices *fs_devices;
1996 WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1999 * in case of fs with no seed, srcdev->fs_devices will point
2000 * to fs_devices of fs_info. However when the dev being replaced is
2001 * a seed dev it will point to the seed's local fs_devices. In short
2002 * srcdev will have its correct fs_devices in both the cases.
2004 fs_devices = srcdev->fs_devices;
2006 list_del_rcu(&srcdev->dev_list);
2007 list_del_rcu(&srcdev->dev_alloc_list);
2008 fs_devices->num_devices--;
2009 if (srcdev->missing)
2010 fs_devices->missing_devices--;
2012 if (srcdev->writeable)
2013 fs_devices->rw_devices--;
2016 fs_devices->open_devices--;
2019 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
2020 struct btrfs_device *srcdev)
2022 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2024 if (srcdev->writeable) {
2025 /* zero out the old super if it is writable */
2026 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
2028 call_rcu(&srcdev->rcu, free_device);
2031 * unless fs_devices is seed fs, num_devices shouldn't go
2034 BUG_ON(!fs_devices->num_devices && !fs_devices->seeding);
2036 /* if this is no devs we rather delete the fs_devices */
2037 if (!fs_devices->num_devices) {
2038 struct btrfs_fs_devices *tmp_fs_devices;
2040 tmp_fs_devices = fs_info->fs_devices;
2041 while (tmp_fs_devices) {
2042 if (tmp_fs_devices->seed == fs_devices) {
2043 tmp_fs_devices->seed = fs_devices->seed;
2046 tmp_fs_devices = tmp_fs_devices->seed;
2048 fs_devices->seed = NULL;
2049 __btrfs_close_devices(fs_devices);
2050 free_fs_devices(fs_devices);
2054 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2055 struct btrfs_device *tgtdev)
2057 mutex_lock(&uuid_mutex);
2059 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2061 btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev);
2064 fs_info->fs_devices->open_devices--;
2066 fs_info->fs_devices->num_devices--;
2068 btrfs_assign_next_active_device(fs_info, tgtdev, NULL);
2070 list_del_rcu(&tgtdev->dev_list);
2072 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2073 mutex_unlock(&uuid_mutex);
2076 * The update_dev_time() with in btrfs_scratch_superblocks()
2077 * may lead to a call to btrfs_show_devname() which will try
2078 * to hold device_list_mutex. And here this device
2079 * is already out of device list, so we don't have to hold
2080 * the device_list_mutex lock.
2082 btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
2083 call_rcu(&tgtdev->rcu, free_device);
2086 static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
2087 struct btrfs_device **device)
2090 struct btrfs_super_block *disk_super;
2093 struct block_device *bdev;
2094 struct buffer_head *bh;
2097 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2098 root->fs_info->bdev_holder, 0, &bdev, &bh);
2101 disk_super = (struct btrfs_super_block *)bh->b_data;
2102 devid = btrfs_stack_device_id(&disk_super->dev_item);
2103 dev_uuid = disk_super->dev_item.uuid;
2104 *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2109 blkdev_put(bdev, FMODE_READ);
2113 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
2115 struct btrfs_device **device)
2118 if (strcmp(device_path, "missing") == 0) {
2119 struct list_head *devices;
2120 struct btrfs_device *tmp;
2122 devices = &root->fs_info->fs_devices->devices;
2124 * It is safe to read the devices since the volume_mutex
2125 * is held by the caller.
2127 list_for_each_entry(tmp, devices, dev_list) {
2128 if (tmp->in_fs_metadata && !tmp->bdev) {
2135 return BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2139 return btrfs_find_device_by_path(root, device_path, device);
2144 * Lookup a device given by device id, or the path if the id is 0.
2146 int btrfs_find_device_by_devspec(struct btrfs_root *root, u64 devid,
2148 struct btrfs_device **device)
2154 *device = btrfs_find_device(root->fs_info, devid, NULL,
2159 if (!devpath || !devpath[0])
2162 ret = btrfs_find_device_missing_or_by_path(root, devpath,
2169 * does all the dirty work required for changing file system's UUID.
2171 static int btrfs_prepare_sprout(struct btrfs_root *root)
2173 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2174 struct btrfs_fs_devices *old_devices;
2175 struct btrfs_fs_devices *seed_devices;
2176 struct btrfs_super_block *disk_super = root->fs_info->super_copy;
2177 struct btrfs_device *device;
2180 BUG_ON(!mutex_is_locked(&uuid_mutex));
2181 if (!fs_devices->seeding)
2184 seed_devices = __alloc_fs_devices();
2185 if (IS_ERR(seed_devices))
2186 return PTR_ERR(seed_devices);
2188 old_devices = clone_fs_devices(fs_devices);
2189 if (IS_ERR(old_devices)) {
2190 kfree(seed_devices);
2191 return PTR_ERR(old_devices);
2194 list_add(&old_devices->list, &fs_uuids);
2196 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2197 seed_devices->opened = 1;
2198 INIT_LIST_HEAD(&seed_devices->devices);
2199 INIT_LIST_HEAD(&seed_devices->alloc_list);
2200 mutex_init(&seed_devices->device_list_mutex);
2202 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2203 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2205 list_for_each_entry(device, &seed_devices->devices, dev_list)
2206 device->fs_devices = seed_devices;
2209 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2210 unlock_chunks(root);
2212 fs_devices->seeding = 0;
2213 fs_devices->num_devices = 0;
2214 fs_devices->open_devices = 0;
2215 fs_devices->missing_devices = 0;
2216 fs_devices->rotating = 0;
2217 fs_devices->seed = seed_devices;
2219 generate_random_uuid(fs_devices->fsid);
2220 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2221 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2222 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2224 super_flags = btrfs_super_flags(disk_super) &
2225 ~BTRFS_SUPER_FLAG_SEEDING;
2226 btrfs_set_super_flags(disk_super, super_flags);
2232 * Store the expected generation for seed devices in device items.
2234 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2235 struct btrfs_root *root)
2237 struct btrfs_path *path;
2238 struct extent_buffer *leaf;
2239 struct btrfs_dev_item *dev_item;
2240 struct btrfs_device *device;
2241 struct btrfs_key key;
2242 u8 fs_uuid[BTRFS_UUID_SIZE];
2243 u8 dev_uuid[BTRFS_UUID_SIZE];
2247 path = btrfs_alloc_path();
2251 root = root->fs_info->chunk_root;
2252 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2254 key.type = BTRFS_DEV_ITEM_KEY;
2257 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2261 leaf = path->nodes[0];
2263 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2264 ret = btrfs_next_leaf(root, path);
2269 leaf = path->nodes[0];
2270 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2271 btrfs_release_path(path);
2275 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2276 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2277 key.type != BTRFS_DEV_ITEM_KEY)
2280 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2281 struct btrfs_dev_item);
2282 devid = btrfs_device_id(leaf, dev_item);
2283 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2285 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2287 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2289 BUG_ON(!device); /* Logic error */
2291 if (device->fs_devices->seeding) {
2292 btrfs_set_device_generation(leaf, dev_item,
2293 device->generation);
2294 btrfs_mark_buffer_dirty(leaf);
2302 btrfs_free_path(path);
2306 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
2308 struct request_queue *q;
2309 struct btrfs_trans_handle *trans;
2310 struct btrfs_device *device;
2311 struct block_device *bdev;
2312 struct list_head *devices;
2313 struct super_block *sb = root->fs_info->sb;
2314 struct rcu_string *name;
2316 int seeding_dev = 0;
2319 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
2322 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2323 root->fs_info->bdev_holder);
2325 return PTR_ERR(bdev);
2327 if (root->fs_info->fs_devices->seeding) {
2329 down_write(&sb->s_umount);
2330 mutex_lock(&uuid_mutex);
2333 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2335 devices = &root->fs_info->fs_devices->devices;
2337 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2338 list_for_each_entry(device, devices, dev_list) {
2339 if (device->bdev == bdev) {
2342 &root->fs_info->fs_devices->device_list_mutex);
2346 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2348 device = btrfs_alloc_device(root->fs_info, NULL, NULL);
2349 if (IS_ERR(device)) {
2350 /* we can safely leave the fs_devices entry around */
2351 ret = PTR_ERR(device);
2355 name = rcu_string_strdup(device_path, GFP_KERNEL);
2361 rcu_assign_pointer(device->name, name);
2363 trans = btrfs_start_transaction(root, 0);
2364 if (IS_ERR(trans)) {
2365 rcu_string_free(device->name);
2367 ret = PTR_ERR(trans);
2371 q = bdev_get_queue(bdev);
2372 if (blk_queue_discard(q))
2373 device->can_discard = 1;
2374 device->writeable = 1;
2375 device->generation = trans->transid;
2376 device->io_width = root->sectorsize;
2377 device->io_align = root->sectorsize;
2378 device->sector_size = root->sectorsize;
2379 device->total_bytes = i_size_read(bdev->bd_inode);
2380 device->disk_total_bytes = device->total_bytes;
2381 device->commit_total_bytes = device->total_bytes;
2382 device->dev_root = root->fs_info->dev_root;
2383 device->bdev = bdev;
2384 device->in_fs_metadata = 1;
2385 device->is_tgtdev_for_dev_replace = 0;
2386 device->mode = FMODE_EXCL;
2387 device->dev_stats_valid = 1;
2388 set_blocksize(device->bdev, 4096);
2391 sb->s_flags &= ~MS_RDONLY;
2392 ret = btrfs_prepare_sprout(root);
2393 BUG_ON(ret); /* -ENOMEM */
2396 device->fs_devices = root->fs_info->fs_devices;
2398 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2400 list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
2401 list_add(&device->dev_alloc_list,
2402 &root->fs_info->fs_devices->alloc_list);
2403 root->fs_info->fs_devices->num_devices++;
2404 root->fs_info->fs_devices->open_devices++;
2405 root->fs_info->fs_devices->rw_devices++;
2406 root->fs_info->fs_devices->total_devices++;
2407 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2409 spin_lock(&root->fs_info->free_chunk_lock);
2410 root->fs_info->free_chunk_space += device->total_bytes;
2411 spin_unlock(&root->fs_info->free_chunk_lock);
2413 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2414 root->fs_info->fs_devices->rotating = 1;
2416 tmp = btrfs_super_total_bytes(root->fs_info->super_copy);
2417 btrfs_set_super_total_bytes(root->fs_info->super_copy,
2418 tmp + device->total_bytes);
2420 tmp = btrfs_super_num_devices(root->fs_info->super_copy);
2421 btrfs_set_super_num_devices(root->fs_info->super_copy,
2424 /* add sysfs device entry */
2425 btrfs_sysfs_add_device_link(root->fs_info->fs_devices, device);
2428 * we've got more storage, clear any full flags on the space
2431 btrfs_clear_space_info_full(root->fs_info);
2433 unlock_chunks(root);
2434 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2438 ret = init_first_rw_device(trans, root, device);
2439 unlock_chunks(root);
2441 btrfs_abort_transaction(trans, root, ret);
2446 ret = btrfs_add_device(trans, root, device);
2448 btrfs_abort_transaction(trans, root, ret);
2453 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2455 ret = btrfs_finish_sprout(trans, root);
2457 btrfs_abort_transaction(trans, root, ret);
2461 /* Sprouting would change fsid of the mounted root,
2462 * so rename the fsid on the sysfs
2464 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2465 root->fs_info->fsid);
2466 if (kobject_rename(&root->fs_info->fs_devices->fsid_kobj,
2468 btrfs_warn(root->fs_info,
2469 "sysfs: failed to create fsid for sprout");
2472 root->fs_info->num_tolerated_disk_barrier_failures =
2473 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
2474 ret = btrfs_commit_transaction(trans, root);
2477 mutex_unlock(&uuid_mutex);
2478 up_write(&sb->s_umount);
2480 if (ret) /* transaction commit */
2483 ret = btrfs_relocate_sys_chunks(root);
2485 btrfs_handle_fs_error(root->fs_info, ret,
2486 "Failed to relocate sys chunks after "
2487 "device initialization. This can be fixed "
2488 "using the \"btrfs balance\" command.");
2489 trans = btrfs_attach_transaction(root);
2490 if (IS_ERR(trans)) {
2491 if (PTR_ERR(trans) == -ENOENT)
2493 return PTR_ERR(trans);
2495 ret = btrfs_commit_transaction(trans, root);
2498 /* Update ctime/mtime for libblkid */
2499 update_dev_time(device_path);
2503 btrfs_end_transaction(trans, root);
2504 rcu_string_free(device->name);
2505 btrfs_sysfs_rm_device_link(root->fs_info->fs_devices, device);
2508 blkdev_put(bdev, FMODE_EXCL);
2510 mutex_unlock(&uuid_mutex);
2511 up_write(&sb->s_umount);
2516 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2517 struct btrfs_device *srcdev,
2518 struct btrfs_device **device_out)
2520 struct request_queue *q;
2521 struct btrfs_device *device;
2522 struct block_device *bdev;
2523 struct btrfs_fs_info *fs_info = root->fs_info;
2524 struct list_head *devices;
2525 struct rcu_string *name;
2526 u64 devid = BTRFS_DEV_REPLACE_DEVID;
2530 if (fs_info->fs_devices->seeding) {
2531 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
2535 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2536 fs_info->bdev_holder);
2538 btrfs_err(fs_info, "target device %s is invalid!", device_path);
2539 return PTR_ERR(bdev);
2542 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2544 devices = &fs_info->fs_devices->devices;
2545 list_for_each_entry(device, devices, dev_list) {
2546 if (device->bdev == bdev) {
2547 btrfs_err(fs_info, "target device is in the filesystem!");
2554 if (i_size_read(bdev->bd_inode) <
2555 btrfs_device_get_total_bytes(srcdev)) {
2556 btrfs_err(fs_info, "target device is smaller than source device!");
2562 device = btrfs_alloc_device(NULL, &devid, NULL);
2563 if (IS_ERR(device)) {
2564 ret = PTR_ERR(device);
2568 name = rcu_string_strdup(device_path, GFP_NOFS);
2574 rcu_assign_pointer(device->name, name);
2576 q = bdev_get_queue(bdev);
2577 if (blk_queue_discard(q))
2578 device->can_discard = 1;
2579 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2580 device->writeable = 1;
2581 device->generation = 0;
2582 device->io_width = root->sectorsize;
2583 device->io_align = root->sectorsize;
2584 device->sector_size = root->sectorsize;
2585 device->total_bytes = btrfs_device_get_total_bytes(srcdev);
2586 device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
2587 device->bytes_used = btrfs_device_get_bytes_used(srcdev);
2588 ASSERT(list_empty(&srcdev->resized_list));
2589 device->commit_total_bytes = srcdev->commit_total_bytes;
2590 device->commit_bytes_used = device->bytes_used;
2591 device->dev_root = fs_info->dev_root;
2592 device->bdev = bdev;
2593 device->in_fs_metadata = 1;
2594 device->is_tgtdev_for_dev_replace = 1;
2595 device->mode = FMODE_EXCL;
2596 device->dev_stats_valid = 1;
2597 set_blocksize(device->bdev, 4096);
2598 device->fs_devices = fs_info->fs_devices;
2599 list_add(&device->dev_list, &fs_info->fs_devices->devices);
2600 fs_info->fs_devices->num_devices++;
2601 fs_info->fs_devices->open_devices++;
2602 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2604 *device_out = device;
2608 blkdev_put(bdev, FMODE_EXCL);
2612 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2613 struct btrfs_device *tgtdev)
2615 WARN_ON(fs_info->fs_devices->rw_devices == 0);
2616 tgtdev->io_width = fs_info->dev_root->sectorsize;
2617 tgtdev->io_align = fs_info->dev_root->sectorsize;
2618 tgtdev->sector_size = fs_info->dev_root->sectorsize;
2619 tgtdev->dev_root = fs_info->dev_root;
2620 tgtdev->in_fs_metadata = 1;
2623 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2624 struct btrfs_device *device)
2627 struct btrfs_path *path;
2628 struct btrfs_root *root;
2629 struct btrfs_dev_item *dev_item;
2630 struct extent_buffer *leaf;
2631 struct btrfs_key key;
2633 root = device->dev_root->fs_info->chunk_root;
2635 path = btrfs_alloc_path();
2639 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2640 key.type = BTRFS_DEV_ITEM_KEY;
2641 key.offset = device->devid;
2643 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2652 leaf = path->nodes[0];
2653 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2655 btrfs_set_device_id(leaf, dev_item, device->devid);
2656 btrfs_set_device_type(leaf, dev_item, device->type);
2657 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2658 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2659 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2660 btrfs_set_device_total_bytes(leaf, dev_item,
2661 btrfs_device_get_disk_total_bytes(device));
2662 btrfs_set_device_bytes_used(leaf, dev_item,
2663 btrfs_device_get_bytes_used(device));
2664 btrfs_mark_buffer_dirty(leaf);
2667 btrfs_free_path(path);
2671 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2672 struct btrfs_device *device, u64 new_size)
2674 struct btrfs_super_block *super_copy =
2675 device->dev_root->fs_info->super_copy;
2676 struct btrfs_fs_devices *fs_devices;
2680 if (!device->writeable)
2683 lock_chunks(device->dev_root);
2684 old_total = btrfs_super_total_bytes(super_copy);
2685 diff = new_size - device->total_bytes;
2687 if (new_size <= device->total_bytes ||
2688 device->is_tgtdev_for_dev_replace) {
2689 unlock_chunks(device->dev_root);
2693 fs_devices = device->dev_root->fs_info->fs_devices;
2695 btrfs_set_super_total_bytes(super_copy, old_total + diff);
2696 device->fs_devices->total_rw_bytes += diff;
2698 btrfs_device_set_total_bytes(device, new_size);
2699 btrfs_device_set_disk_total_bytes(device, new_size);
2700 btrfs_clear_space_info_full(device->dev_root->fs_info);
2701 if (list_empty(&device->resized_list))
2702 list_add_tail(&device->resized_list,
2703 &fs_devices->resized_devices);
2704 unlock_chunks(device->dev_root);
2706 return btrfs_update_device(trans, device);
2709 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2710 struct btrfs_root *root, u64 chunk_objectid,
2714 struct btrfs_path *path;
2715 struct btrfs_key key;
2717 root = root->fs_info->chunk_root;
2718 path = btrfs_alloc_path();
2722 key.objectid = chunk_objectid;
2723 key.offset = chunk_offset;
2724 key.type = BTRFS_CHUNK_ITEM_KEY;
2726 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2729 else if (ret > 0) { /* Logic error or corruption */
2730 btrfs_handle_fs_error(root->fs_info, -ENOENT,
2731 "Failed lookup while freeing chunk.");
2736 ret = btrfs_del_item(trans, root, path);
2738 btrfs_handle_fs_error(root->fs_info, ret,
2739 "Failed to delete chunk item.");
2741 btrfs_free_path(path);
2745 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2748 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2749 struct btrfs_disk_key *disk_key;
2750 struct btrfs_chunk *chunk;
2757 struct btrfs_key key;
2760 array_size = btrfs_super_sys_array_size(super_copy);
2762 ptr = super_copy->sys_chunk_array;
2765 while (cur < array_size) {
2766 disk_key = (struct btrfs_disk_key *)ptr;
2767 btrfs_disk_key_to_cpu(&key, disk_key);
2769 len = sizeof(*disk_key);
2771 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2772 chunk = (struct btrfs_chunk *)(ptr + len);
2773 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2774 len += btrfs_chunk_item_size(num_stripes);
2779 if (key.objectid == chunk_objectid &&
2780 key.offset == chunk_offset) {
2781 memmove(ptr, ptr + len, array_size - (cur + len));
2783 btrfs_set_super_sys_array_size(super_copy, array_size);
2789 unlock_chunks(root);
2793 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2794 struct btrfs_root *root, u64 chunk_offset)
2796 struct extent_map_tree *em_tree;
2797 struct extent_map *em;
2798 struct btrfs_root *extent_root = root->fs_info->extent_root;
2799 struct map_lookup *map;
2800 u64 dev_extent_len = 0;
2801 u64 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2803 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2806 root = root->fs_info->chunk_root;
2807 em_tree = &root->fs_info->mapping_tree.map_tree;
2809 read_lock(&em_tree->lock);
2810 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2811 read_unlock(&em_tree->lock);
2813 if (!em || em->start > chunk_offset ||
2814 em->start + em->len < chunk_offset) {
2816 * This is a logic error, but we don't want to just rely on the
2817 * user having built with ASSERT enabled, so if ASSERT doesn't
2818 * do anything we still error out.
2822 free_extent_map(em);
2825 map = em->map_lookup;
2826 lock_chunks(root->fs_info->chunk_root);
2827 check_system_chunk(trans, extent_root, map->type);
2828 unlock_chunks(root->fs_info->chunk_root);
2831 * Take the device list mutex to prevent races with the final phase of
2832 * a device replace operation that replaces the device object associated
2833 * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2835 mutex_lock(&fs_devices->device_list_mutex);
2836 for (i = 0; i < map->num_stripes; i++) {
2837 struct btrfs_device *device = map->stripes[i].dev;
2838 ret = btrfs_free_dev_extent(trans, device,
2839 map->stripes[i].physical,
2842 mutex_unlock(&fs_devices->device_list_mutex);
2843 btrfs_abort_transaction(trans, root, ret);
2847 if (device->bytes_used > 0) {
2849 btrfs_device_set_bytes_used(device,
2850 device->bytes_used - dev_extent_len);
2851 spin_lock(&root->fs_info->free_chunk_lock);
2852 root->fs_info->free_chunk_space += dev_extent_len;
2853 spin_unlock(&root->fs_info->free_chunk_lock);
2854 btrfs_clear_space_info_full(root->fs_info);
2855 unlock_chunks(root);
2858 if (map->stripes[i].dev) {
2859 ret = btrfs_update_device(trans, map->stripes[i].dev);
2861 mutex_unlock(&fs_devices->device_list_mutex);
2862 btrfs_abort_transaction(trans, root, ret);
2867 mutex_unlock(&fs_devices->device_list_mutex);
2869 ret = btrfs_free_chunk(trans, root, chunk_objectid, chunk_offset);
2871 btrfs_abort_transaction(trans, root, ret);
2875 trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2877 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2878 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2880 btrfs_abort_transaction(trans, root, ret);
2885 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset, em);
2887 btrfs_abort_transaction(trans, extent_root, ret);
2893 free_extent_map(em);
2897 static int btrfs_relocate_chunk(struct btrfs_root *root, u64 chunk_offset)
2899 struct btrfs_root *extent_root;
2900 struct btrfs_trans_handle *trans;
2903 root = root->fs_info->chunk_root;
2904 extent_root = root->fs_info->extent_root;
2907 * Prevent races with automatic removal of unused block groups.
2908 * After we relocate and before we remove the chunk with offset
2909 * chunk_offset, automatic removal of the block group can kick in,
2910 * resulting in a failure when calling btrfs_remove_chunk() below.
2912 * Make sure to acquire this mutex before doing a tree search (dev
2913 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2914 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2915 * we release the path used to search the chunk/dev tree and before
2916 * the current task acquires this mutex and calls us.
2918 ASSERT(mutex_is_locked(&root->fs_info->delete_unused_bgs_mutex));
2920 ret = btrfs_can_relocate(extent_root, chunk_offset);
2924 /* step one, relocate all the extents inside this chunk */
2925 btrfs_scrub_pause(root);
2926 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2927 btrfs_scrub_continue(root);
2931 trans = btrfs_start_trans_remove_block_group(root->fs_info,
2933 if (IS_ERR(trans)) {
2934 ret = PTR_ERR(trans);
2935 btrfs_handle_fs_error(root->fs_info, ret, NULL);
2940 * step two, delete the device extents and the
2941 * chunk tree entries
2943 ret = btrfs_remove_chunk(trans, root, chunk_offset);
2944 btrfs_end_transaction(trans, root);
2948 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2950 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2951 struct btrfs_path *path;
2952 struct extent_buffer *leaf;
2953 struct btrfs_chunk *chunk;
2954 struct btrfs_key key;
2955 struct btrfs_key found_key;
2957 bool retried = false;
2961 path = btrfs_alloc_path();
2966 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2967 key.offset = (u64)-1;
2968 key.type = BTRFS_CHUNK_ITEM_KEY;
2971 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
2972 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2974 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2977 BUG_ON(ret == 0); /* Corruption */
2979 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2982 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2988 leaf = path->nodes[0];
2989 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2991 chunk = btrfs_item_ptr(leaf, path->slots[0],
2992 struct btrfs_chunk);
2993 chunk_type = btrfs_chunk_type(leaf, chunk);
2994 btrfs_release_path(path);
2996 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2997 ret = btrfs_relocate_chunk(chunk_root,
3004 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
3006 if (found_key.offset == 0)
3008 key.offset = found_key.offset - 1;
3011 if (failed && !retried) {
3015 } else if (WARN_ON(failed && retried)) {
3019 btrfs_free_path(path);
3023 static int insert_balance_item(struct btrfs_root *root,
3024 struct btrfs_balance_control *bctl)
3026 struct btrfs_trans_handle *trans;
3027 struct btrfs_balance_item *item;
3028 struct btrfs_disk_balance_args disk_bargs;
3029 struct btrfs_path *path;
3030 struct extent_buffer *leaf;
3031 struct btrfs_key key;
3034 path = btrfs_alloc_path();
3038 trans = btrfs_start_transaction(root, 0);
3039 if (IS_ERR(trans)) {
3040 btrfs_free_path(path);
3041 return PTR_ERR(trans);
3044 key.objectid = BTRFS_BALANCE_OBJECTID;
3045 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3048 ret = btrfs_insert_empty_item(trans, root, path, &key,
3053 leaf = path->nodes[0];
3054 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3056 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
3058 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3059 btrfs_set_balance_data(leaf, item, &disk_bargs);
3060 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3061 btrfs_set_balance_meta(leaf, item, &disk_bargs);
3062 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3063 btrfs_set_balance_sys(leaf, item, &disk_bargs);
3065 btrfs_set_balance_flags(leaf, item, bctl->flags);
3067 btrfs_mark_buffer_dirty(leaf);
3069 btrfs_free_path(path);
3070 err = btrfs_commit_transaction(trans, root);
3076 static int del_balance_item(struct btrfs_root *root)
3078 struct btrfs_trans_handle *trans;
3079 struct btrfs_path *path;
3080 struct btrfs_key key;
3083 path = btrfs_alloc_path();
3087 trans = btrfs_start_transaction(root, 0);
3088 if (IS_ERR(trans)) {
3089 btrfs_free_path(path);
3090 return PTR_ERR(trans);
3093 key.objectid = BTRFS_BALANCE_OBJECTID;
3094 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3097 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3105 ret = btrfs_del_item(trans, root, path);
3107 btrfs_free_path(path);
3108 err = btrfs_commit_transaction(trans, root);
3115 * This is a heuristic used to reduce the number of chunks balanced on
3116 * resume after balance was interrupted.
3118 static void update_balance_args(struct btrfs_balance_control *bctl)
3121 * Turn on soft mode for chunk types that were being converted.
3123 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3124 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3125 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3126 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3127 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3128 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3131 * Turn on usage filter if is not already used. The idea is
3132 * that chunks that we have already balanced should be
3133 * reasonably full. Don't do it for chunks that are being
3134 * converted - that will keep us from relocating unconverted
3135 * (albeit full) chunks.
3137 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3138 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3139 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3140 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3141 bctl->data.usage = 90;
3143 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3144 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3145 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3146 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3147 bctl->sys.usage = 90;
3149 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3150 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3151 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3152 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3153 bctl->meta.usage = 90;
3158 * Should be called with both balance and volume mutexes held to
3159 * serialize other volume operations (add_dev/rm_dev/resize) with
3160 * restriper. Same goes for unset_balance_control.
3162 static void set_balance_control(struct btrfs_balance_control *bctl)
3164 struct btrfs_fs_info *fs_info = bctl->fs_info;
3166 BUG_ON(fs_info->balance_ctl);
3168 spin_lock(&fs_info->balance_lock);
3169 fs_info->balance_ctl = bctl;
3170 spin_unlock(&fs_info->balance_lock);
3173 static void unset_balance_control(struct btrfs_fs_info *fs_info)
3175 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3177 BUG_ON(!fs_info->balance_ctl);
3179 spin_lock(&fs_info->balance_lock);
3180 fs_info->balance_ctl = NULL;
3181 spin_unlock(&fs_info->balance_lock);
3187 * Balance filters. Return 1 if chunk should be filtered out
3188 * (should not be balanced).
3190 static int chunk_profiles_filter(u64 chunk_type,
3191 struct btrfs_balance_args *bargs)
3193 chunk_type = chunk_to_extended(chunk_type) &
3194 BTRFS_EXTENDED_PROFILE_MASK;
3196 if (bargs->profiles & chunk_type)
3202 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3203 struct btrfs_balance_args *bargs)
3205 struct btrfs_block_group_cache *cache;
3207 u64 user_thresh_min;
3208 u64 user_thresh_max;
3211 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3212 chunk_used = btrfs_block_group_used(&cache->item);
3214 if (bargs->usage_min == 0)
3215 user_thresh_min = 0;
3217 user_thresh_min = div_factor_fine(cache->key.offset,
3220 if (bargs->usage_max == 0)
3221 user_thresh_max = 1;
3222 else if (bargs->usage_max > 100)
3223 user_thresh_max = cache->key.offset;
3225 user_thresh_max = div_factor_fine(cache->key.offset,
3228 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3231 btrfs_put_block_group(cache);
3235 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3236 u64 chunk_offset, struct btrfs_balance_args *bargs)
3238 struct btrfs_block_group_cache *cache;
3239 u64 chunk_used, user_thresh;
3242 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3243 chunk_used = btrfs_block_group_used(&cache->item);
3245 if (bargs->usage_min == 0)
3247 else if (bargs->usage > 100)
3248 user_thresh = cache->key.offset;
3250 user_thresh = div_factor_fine(cache->key.offset,
3253 if (chunk_used < user_thresh)
3256 btrfs_put_block_group(cache);
3260 static int chunk_devid_filter(struct extent_buffer *leaf,
3261 struct btrfs_chunk *chunk,
3262 struct btrfs_balance_args *bargs)
3264 struct btrfs_stripe *stripe;
3265 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3268 for (i = 0; i < num_stripes; i++) {
3269 stripe = btrfs_stripe_nr(chunk, i);
3270 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3277 /* [pstart, pend) */
3278 static int chunk_drange_filter(struct extent_buffer *leaf,
3279 struct btrfs_chunk *chunk,
3281 struct btrfs_balance_args *bargs)
3283 struct btrfs_stripe *stripe;
3284 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3290 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3293 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3294 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3295 factor = num_stripes / 2;
3296 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3297 factor = num_stripes - 1;
3298 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3299 factor = num_stripes - 2;
3301 factor = num_stripes;
3304 for (i = 0; i < num_stripes; i++) {
3305 stripe = btrfs_stripe_nr(chunk, i);
3306 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3309 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3310 stripe_length = btrfs_chunk_length(leaf, chunk);
3311 stripe_length = div_u64(stripe_length, factor);
3313 if (stripe_offset < bargs->pend &&
3314 stripe_offset + stripe_length > bargs->pstart)
3321 /* [vstart, vend) */
3322 static int chunk_vrange_filter(struct extent_buffer *leaf,
3323 struct btrfs_chunk *chunk,
3325 struct btrfs_balance_args *bargs)
3327 if (chunk_offset < bargs->vend &&
3328 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3329 /* at least part of the chunk is inside this vrange */
3335 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3336 struct btrfs_chunk *chunk,
3337 struct btrfs_balance_args *bargs)
3339 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3341 if (bargs->stripes_min <= num_stripes
3342 && num_stripes <= bargs->stripes_max)
3348 static int chunk_soft_convert_filter(u64 chunk_type,
3349 struct btrfs_balance_args *bargs)
3351 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3354 chunk_type = chunk_to_extended(chunk_type) &
3355 BTRFS_EXTENDED_PROFILE_MASK;
3357 if (bargs->target == chunk_type)
3363 static int should_balance_chunk(struct btrfs_root *root,
3364 struct extent_buffer *leaf,
3365 struct btrfs_chunk *chunk, u64 chunk_offset)
3367 struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
3368 struct btrfs_balance_args *bargs = NULL;
3369 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3372 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3373 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3377 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3378 bargs = &bctl->data;
3379 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3381 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3382 bargs = &bctl->meta;
3384 /* profiles filter */
3385 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3386 chunk_profiles_filter(chunk_type, bargs)) {
3391 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3392 chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
3394 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3395 chunk_usage_range_filter(bctl->fs_info, chunk_offset, bargs)) {
3400 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3401 chunk_devid_filter(leaf, chunk, bargs)) {
3405 /* drange filter, makes sense only with devid filter */
3406 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3407 chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
3412 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3413 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3417 /* stripes filter */
3418 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3419 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3423 /* soft profile changing mode */
3424 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3425 chunk_soft_convert_filter(chunk_type, bargs)) {
3430 * limited by count, must be the last filter
3432 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3433 if (bargs->limit == 0)
3437 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3439 * Same logic as the 'limit' filter; the minimum cannot be
3440 * determined here because we do not have the global information
3441 * about the count of all chunks that satisfy the filters.
3443 if (bargs->limit_max == 0)
3452 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3454 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3455 struct btrfs_root *chunk_root = fs_info->chunk_root;
3456 struct btrfs_root *dev_root = fs_info->dev_root;
3457 struct list_head *devices;
3458 struct btrfs_device *device;
3462 struct btrfs_chunk *chunk;
3463 struct btrfs_path *path = NULL;
3464 struct btrfs_key key;
3465 struct btrfs_key found_key;
3466 struct btrfs_trans_handle *trans;
3467 struct extent_buffer *leaf;
3470 int enospc_errors = 0;
3471 bool counting = true;
3472 /* The single value limit and min/max limits use the same bytes in the */
3473 u64 limit_data = bctl->data.limit;
3474 u64 limit_meta = bctl->meta.limit;
3475 u64 limit_sys = bctl->sys.limit;
3479 int chunk_reserved = 0;
3482 /* step one make some room on all the devices */
3483 devices = &fs_info->fs_devices->devices;
3484 list_for_each_entry(device, devices, dev_list) {
3485 old_size = btrfs_device_get_total_bytes(device);
3486 size_to_free = div_factor(old_size, 1);
3487 size_to_free = min_t(u64, size_to_free, SZ_1M);
3488 if (!device->writeable ||
3489 btrfs_device_get_total_bytes(device) -
3490 btrfs_device_get_bytes_used(device) > size_to_free ||
3491 device->is_tgtdev_for_dev_replace)
3494 ret = btrfs_shrink_device(device, old_size - size_to_free);
3498 /* btrfs_shrink_device never returns ret > 0 */
3503 trans = btrfs_start_transaction(dev_root, 0);
3504 if (IS_ERR(trans)) {
3505 ret = PTR_ERR(trans);
3506 btrfs_info_in_rcu(fs_info,
3507 "resize: unable to start transaction after shrinking device %s (error %d), old size %llu, new size %llu",
3508 rcu_str_deref(device->name), ret,
3509 old_size, old_size - size_to_free);
3513 ret = btrfs_grow_device(trans, device, old_size);
3515 btrfs_end_transaction(trans, dev_root);
3516 /* btrfs_grow_device never returns ret > 0 */
3518 btrfs_info_in_rcu(fs_info,
3519 "resize: unable to grow device after shrinking device %s (error %d), old size %llu, new size %llu",
3520 rcu_str_deref(device->name), ret,
3521 old_size, old_size - size_to_free);
3525 btrfs_end_transaction(trans, dev_root);
3528 /* step two, relocate all the chunks */
3529 path = btrfs_alloc_path();
3535 /* zero out stat counters */
3536 spin_lock(&fs_info->balance_lock);
3537 memset(&bctl->stat, 0, sizeof(bctl->stat));
3538 spin_unlock(&fs_info->balance_lock);
3542 * The single value limit and min/max limits use the same bytes
3545 bctl->data.limit = limit_data;
3546 bctl->meta.limit = limit_meta;
3547 bctl->sys.limit = limit_sys;
3549 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3550 key.offset = (u64)-1;
3551 key.type = BTRFS_CHUNK_ITEM_KEY;
3554 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3555 atomic_read(&fs_info->balance_cancel_req)) {
3560 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3561 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3563 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3568 * this shouldn't happen, it means the last relocate
3572 BUG(); /* FIXME break ? */
3574 ret = btrfs_previous_item(chunk_root, path, 0,
3575 BTRFS_CHUNK_ITEM_KEY);
3577 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3582 leaf = path->nodes[0];
3583 slot = path->slots[0];
3584 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3586 if (found_key.objectid != key.objectid) {
3587 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3591 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3592 chunk_type = btrfs_chunk_type(leaf, chunk);
3595 spin_lock(&fs_info->balance_lock);
3596 bctl->stat.considered++;
3597 spin_unlock(&fs_info->balance_lock);
3600 ret = should_balance_chunk(chunk_root, leaf, chunk,
3603 btrfs_release_path(path);
3605 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3610 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3611 spin_lock(&fs_info->balance_lock);
3612 bctl->stat.expected++;
3613 spin_unlock(&fs_info->balance_lock);
3615 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3617 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3619 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3626 * Apply limit_min filter, no need to check if the LIMITS
3627 * filter is used, limit_min is 0 by default
3629 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3630 count_data < bctl->data.limit_min)
3631 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3632 count_meta < bctl->meta.limit_min)
3633 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3634 count_sys < bctl->sys.limit_min)) {
3635 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3639 ASSERT(fs_info->data_sinfo);
3640 spin_lock(&fs_info->data_sinfo->lock);
3641 bytes_used = fs_info->data_sinfo->bytes_used;
3642 spin_unlock(&fs_info->data_sinfo->lock);
3644 if ((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3645 !chunk_reserved && !bytes_used) {
3646 trans = btrfs_start_transaction(chunk_root, 0);
3647 if (IS_ERR(trans)) {
3648 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3649 ret = PTR_ERR(trans);
3653 ret = btrfs_force_chunk_alloc(trans, chunk_root,
3654 BTRFS_BLOCK_GROUP_DATA);
3655 btrfs_end_transaction(trans, chunk_root);
3657 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3663 ret = btrfs_relocate_chunk(chunk_root,
3665 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3666 if (ret && ret != -ENOSPC)
3668 if (ret == -ENOSPC) {
3671 spin_lock(&fs_info->balance_lock);
3672 bctl->stat.completed++;
3673 spin_unlock(&fs_info->balance_lock);
3676 if (found_key.offset == 0)
3678 key.offset = found_key.offset - 1;
3682 btrfs_release_path(path);
3687 btrfs_free_path(path);
3688 if (enospc_errors) {
3689 btrfs_info(fs_info, "%d enospc errors during balance",
3699 * alloc_profile_is_valid - see if a given profile is valid and reduced
3700 * @flags: profile to validate
3701 * @extended: if true @flags is treated as an extended profile
3703 static int alloc_profile_is_valid(u64 flags, int extended)
3705 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3706 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3708 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3710 /* 1) check that all other bits are zeroed */
3714 /* 2) see if profile is reduced */
3716 return !extended; /* "0" is valid for usual profiles */
3718 /* true if exactly one bit set */
3719 return (flags & (flags - 1)) == 0;
3722 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3724 /* cancel requested || normal exit path */
3725 return atomic_read(&fs_info->balance_cancel_req) ||
3726 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3727 atomic_read(&fs_info->balance_cancel_req) == 0);
3730 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3734 unset_balance_control(fs_info);
3735 ret = del_balance_item(fs_info->tree_root);
3737 btrfs_handle_fs_error(fs_info, ret, NULL);
3739 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3742 /* Non-zero return value signifies invalidity */
3743 static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg,
3746 return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3747 (!alloc_profile_is_valid(bctl_arg->target, 1) ||
3748 (bctl_arg->target & ~allowed)));
3752 * Should be called with both balance and volume mutexes held
3754 int btrfs_balance(struct btrfs_balance_control *bctl,
3755 struct btrfs_ioctl_balance_args *bargs)
3757 struct btrfs_fs_info *fs_info = bctl->fs_info;
3764 if (btrfs_fs_closing(fs_info) ||
3765 atomic_read(&fs_info->balance_pause_req) ||
3766 atomic_read(&fs_info->balance_cancel_req)) {
3771 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3772 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3776 * In case of mixed groups both data and meta should be picked,
3777 * and identical options should be given for both of them.
3779 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3780 if (mixed && (bctl->flags & allowed)) {
3781 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3782 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3783 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3784 btrfs_err(fs_info, "with mixed groups data and "
3785 "metadata balance options must be the same");
3791 num_devices = fs_info->fs_devices->num_devices;
3792 btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
3793 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3794 BUG_ON(num_devices < 1);
3797 btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
3798 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE | BTRFS_BLOCK_GROUP_DUP;
3799 if (num_devices > 1)
3800 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3801 if (num_devices > 2)
3802 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3803 if (num_devices > 3)
3804 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3805 BTRFS_BLOCK_GROUP_RAID6);
3806 if (validate_convert_profile(&bctl->data, allowed)) {
3807 btrfs_err(fs_info, "unable to start balance with target "
3808 "data profile %llu",
3813 if (validate_convert_profile(&bctl->meta, allowed)) {
3815 "unable to start balance with target metadata profile %llu",
3820 if (validate_convert_profile(&bctl->sys, allowed)) {
3822 "unable to start balance with target system profile %llu",
3828 /* allow to reduce meta or sys integrity only if force set */
3829 allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3830 BTRFS_BLOCK_GROUP_RAID10 |
3831 BTRFS_BLOCK_GROUP_RAID5 |
3832 BTRFS_BLOCK_GROUP_RAID6;
3834 seq = read_seqbegin(&fs_info->profiles_lock);
3836 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3837 (fs_info->avail_system_alloc_bits & allowed) &&
3838 !(bctl->sys.target & allowed)) ||
3839 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3840 (fs_info->avail_metadata_alloc_bits & allowed) &&
3841 !(bctl->meta.target & allowed))) {
3842 if (bctl->flags & BTRFS_BALANCE_FORCE) {
3843 btrfs_info(fs_info, "force reducing metadata integrity");
3845 btrfs_err(fs_info, "balance will reduce metadata "
3846 "integrity, use force if you want this");
3851 } while (read_seqretry(&fs_info->profiles_lock, seq));
3853 if (btrfs_get_num_tolerated_disk_barrier_failures(bctl->meta.target) <
3854 btrfs_get_num_tolerated_disk_barrier_failures(bctl->data.target)) {
3856 "metadata profile 0x%llx has lower redundancy than data profile 0x%llx",
3857 bctl->meta.target, bctl->data.target);
3860 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3861 fs_info->num_tolerated_disk_barrier_failures = min(
3862 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info),
3863 btrfs_get_num_tolerated_disk_barrier_failures(
3867 ret = insert_balance_item(fs_info->tree_root, bctl);
3868 if (ret && ret != -EEXIST)
3871 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3872 BUG_ON(ret == -EEXIST);
3873 set_balance_control(bctl);
3875 BUG_ON(ret != -EEXIST);
3876 spin_lock(&fs_info->balance_lock);
3877 update_balance_args(bctl);
3878 spin_unlock(&fs_info->balance_lock);
3881 atomic_inc(&fs_info->balance_running);
3882 mutex_unlock(&fs_info->balance_mutex);
3884 ret = __btrfs_balance(fs_info);
3886 mutex_lock(&fs_info->balance_mutex);
3887 atomic_dec(&fs_info->balance_running);
3889 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3890 fs_info->num_tolerated_disk_barrier_failures =
3891 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3895 memset(bargs, 0, sizeof(*bargs));
3896 update_ioctl_balance_args(fs_info, 0, bargs);
3899 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3900 balance_need_close(fs_info)) {
3901 __cancel_balance(fs_info);
3904 wake_up(&fs_info->balance_wait_q);
3908 if (bctl->flags & BTRFS_BALANCE_RESUME)
3909 __cancel_balance(fs_info);
3912 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3917 static int balance_kthread(void *data)
3919 struct btrfs_fs_info *fs_info = data;
3922 mutex_lock(&fs_info->volume_mutex);
3923 mutex_lock(&fs_info->balance_mutex);
3925 if (fs_info->balance_ctl) {
3926 btrfs_info(fs_info, "continuing balance");
3927 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3930 mutex_unlock(&fs_info->balance_mutex);
3931 mutex_unlock(&fs_info->volume_mutex);
3936 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3938 struct task_struct *tsk;
3940 spin_lock(&fs_info->balance_lock);
3941 if (!fs_info->balance_ctl) {
3942 spin_unlock(&fs_info->balance_lock);
3945 spin_unlock(&fs_info->balance_lock);
3947 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
3948 btrfs_info(fs_info, "force skipping balance");
3952 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3953 return PTR_ERR_OR_ZERO(tsk);
3956 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3958 struct btrfs_balance_control *bctl;
3959 struct btrfs_balance_item *item;
3960 struct btrfs_disk_balance_args disk_bargs;
3961 struct btrfs_path *path;
3962 struct extent_buffer *leaf;
3963 struct btrfs_key key;
3966 path = btrfs_alloc_path();
3970 key.objectid = BTRFS_BALANCE_OBJECTID;
3971 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3974 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3977 if (ret > 0) { /* ret = -ENOENT; */
3982 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3988 leaf = path->nodes[0];
3989 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3991 bctl->fs_info = fs_info;
3992 bctl->flags = btrfs_balance_flags(leaf, item);
3993 bctl->flags |= BTRFS_BALANCE_RESUME;
3995 btrfs_balance_data(leaf, item, &disk_bargs);
3996 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3997 btrfs_balance_meta(leaf, item, &disk_bargs);
3998 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3999 btrfs_balance_sys(leaf, item, &disk_bargs);
4000 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4002 WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
4004 mutex_lock(&fs_info->volume_mutex);
4005 mutex_lock(&fs_info->balance_mutex);
4007 set_balance_control(bctl);
4009 mutex_unlock(&fs_info->balance_mutex);
4010 mutex_unlock(&fs_info->volume_mutex);
4012 btrfs_free_path(path);
4016 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4020 mutex_lock(&fs_info->balance_mutex);
4021 if (!fs_info->balance_ctl) {
4022 mutex_unlock(&fs_info->balance_mutex);
4026 if (atomic_read(&fs_info->balance_running)) {
4027 atomic_inc(&fs_info->balance_pause_req);
4028 mutex_unlock(&fs_info->balance_mutex);
4030 wait_event(fs_info->balance_wait_q,
4031 atomic_read(&fs_info->balance_running) == 0);
4033 mutex_lock(&fs_info->balance_mutex);
4034 /* we are good with balance_ctl ripped off from under us */
4035 BUG_ON(atomic_read(&fs_info->balance_running));
4036 atomic_dec(&fs_info->balance_pause_req);
4041 mutex_unlock(&fs_info->balance_mutex);
4045 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4047 if (fs_info->sb->s_flags & MS_RDONLY)
4050 mutex_lock(&fs_info->balance_mutex);
4051 if (!fs_info->balance_ctl) {
4052 mutex_unlock(&fs_info->balance_mutex);
4056 atomic_inc(&fs_info->balance_cancel_req);
4058 * if we are running just wait and return, balance item is
4059 * deleted in btrfs_balance in this case
4061 if (atomic_read(&fs_info->balance_running)) {
4062 mutex_unlock(&fs_info->balance_mutex);
4063 wait_event(fs_info->balance_wait_q,
4064 atomic_read(&fs_info->balance_running) == 0);
4065 mutex_lock(&fs_info->balance_mutex);
4067 /* __cancel_balance needs volume_mutex */
4068 mutex_unlock(&fs_info->balance_mutex);
4069 mutex_lock(&fs_info->volume_mutex);
4070 mutex_lock(&fs_info->balance_mutex);
4072 if (fs_info->balance_ctl)
4073 __cancel_balance(fs_info);
4075 mutex_unlock(&fs_info->volume_mutex);
4078 BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
4079 atomic_dec(&fs_info->balance_cancel_req);
4080 mutex_unlock(&fs_info->balance_mutex);
4084 static int btrfs_uuid_scan_kthread(void *data)
4086 struct btrfs_fs_info *fs_info = data;
4087 struct btrfs_root *root = fs_info->tree_root;
4088 struct btrfs_key key;
4089 struct btrfs_key max_key;
4090 struct btrfs_path *path = NULL;
4092 struct extent_buffer *eb;
4094 struct btrfs_root_item root_item;
4096 struct btrfs_trans_handle *trans = NULL;
4098 path = btrfs_alloc_path();
4105 key.type = BTRFS_ROOT_ITEM_KEY;
4108 max_key.objectid = (u64)-1;
4109 max_key.type = BTRFS_ROOT_ITEM_KEY;
4110 max_key.offset = (u64)-1;
4113 ret = btrfs_search_forward(root, &key, path, 0);
4120 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4121 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4122 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4123 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4126 eb = path->nodes[0];
4127 slot = path->slots[0];
4128 item_size = btrfs_item_size_nr(eb, slot);
4129 if (item_size < sizeof(root_item))
4132 read_extent_buffer(eb, &root_item,
4133 btrfs_item_ptr_offset(eb, slot),
4134 (int)sizeof(root_item));
4135 if (btrfs_root_refs(&root_item) == 0)
4138 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4139 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4143 btrfs_release_path(path);
4145 * 1 - subvol uuid item
4146 * 1 - received_subvol uuid item
4148 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4149 if (IS_ERR(trans)) {
4150 ret = PTR_ERR(trans);
4158 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4159 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
4161 BTRFS_UUID_KEY_SUBVOL,
4164 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4170 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4171 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
4172 root_item.received_uuid,
4173 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4176 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4184 ret = btrfs_end_transaction(trans, fs_info->uuid_root);
4190 btrfs_release_path(path);
4191 if (key.offset < (u64)-1) {
4193 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4195 key.type = BTRFS_ROOT_ITEM_KEY;
4196 } else if (key.objectid < (u64)-1) {
4198 key.type = BTRFS_ROOT_ITEM_KEY;
4207 btrfs_free_path(path);
4208 if (trans && !IS_ERR(trans))
4209 btrfs_end_transaction(trans, fs_info->uuid_root);
4211 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4213 fs_info->update_uuid_tree_gen = 1;
4214 up(&fs_info->uuid_tree_rescan_sem);
4219 * Callback for btrfs_uuid_tree_iterate().
4221 * 0 check succeeded, the entry is not outdated.
4222 * < 0 if an error occurred.
4223 * > 0 if the check failed, which means the caller shall remove the entry.
4225 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
4226 u8 *uuid, u8 type, u64 subid)
4228 struct btrfs_key key;
4230 struct btrfs_root *subvol_root;
4232 if (type != BTRFS_UUID_KEY_SUBVOL &&
4233 type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
4236 key.objectid = subid;
4237 key.type = BTRFS_ROOT_ITEM_KEY;
4238 key.offset = (u64)-1;
4239 subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
4240 if (IS_ERR(subvol_root)) {
4241 ret = PTR_ERR(subvol_root);
4248 case BTRFS_UUID_KEY_SUBVOL:
4249 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
4252 case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
4253 if (memcmp(uuid, subvol_root->root_item.received_uuid,
4263 static int btrfs_uuid_rescan_kthread(void *data)
4265 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4269 * 1st step is to iterate through the existing UUID tree and
4270 * to delete all entries that contain outdated data.
4271 * 2nd step is to add all missing entries to the UUID tree.
4273 ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4275 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4276 up(&fs_info->uuid_tree_rescan_sem);
4279 return btrfs_uuid_scan_kthread(data);
4282 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4284 struct btrfs_trans_handle *trans;
4285 struct btrfs_root *tree_root = fs_info->tree_root;
4286 struct btrfs_root *uuid_root;
4287 struct task_struct *task;
4294 trans = btrfs_start_transaction(tree_root, 2);
4296 return PTR_ERR(trans);
4298 uuid_root = btrfs_create_tree(trans, fs_info,
4299 BTRFS_UUID_TREE_OBJECTID);
4300 if (IS_ERR(uuid_root)) {
4301 ret = PTR_ERR(uuid_root);
4302 btrfs_abort_transaction(trans, tree_root, ret);
4303 btrfs_end_transaction(trans, tree_root);
4307 fs_info->uuid_root = uuid_root;
4309 ret = btrfs_commit_transaction(trans, tree_root);
4313 down(&fs_info->uuid_tree_rescan_sem);
4314 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4316 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4317 btrfs_warn(fs_info, "failed to start uuid_scan task");
4318 up(&fs_info->uuid_tree_rescan_sem);
4319 return PTR_ERR(task);
4325 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4327 struct task_struct *task;
4329 down(&fs_info->uuid_tree_rescan_sem);
4330 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4332 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4333 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4334 up(&fs_info->uuid_tree_rescan_sem);
4335 return PTR_ERR(task);
4342 * shrinking a device means finding all of the device extents past
4343 * the new size, and then following the back refs to the chunks.
4344 * The chunk relocation code actually frees the device extent
4346 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4348 struct btrfs_trans_handle *trans;
4349 struct btrfs_root *root = device->dev_root;
4350 struct btrfs_dev_extent *dev_extent = NULL;
4351 struct btrfs_path *path;
4357 bool retried = false;
4358 bool checked_pending_chunks = false;
4359 struct extent_buffer *l;
4360 struct btrfs_key key;
4361 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4362 u64 old_total = btrfs_super_total_bytes(super_copy);
4363 u64 old_size = btrfs_device_get_total_bytes(device);
4364 u64 diff = old_size - new_size;
4366 if (device->is_tgtdev_for_dev_replace)
4369 path = btrfs_alloc_path();
4373 path->reada = READA_FORWARD;
4377 btrfs_device_set_total_bytes(device, new_size);
4378 if (device->writeable) {
4379 device->fs_devices->total_rw_bytes -= diff;
4380 spin_lock(&root->fs_info->free_chunk_lock);
4381 root->fs_info->free_chunk_space -= diff;
4382 spin_unlock(&root->fs_info->free_chunk_lock);
4384 unlock_chunks(root);
4387 key.objectid = device->devid;
4388 key.offset = (u64)-1;
4389 key.type = BTRFS_DEV_EXTENT_KEY;
4392 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
4393 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4395 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4399 ret = btrfs_previous_item(root, path, 0, key.type);
4401 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4406 btrfs_release_path(path);
4411 slot = path->slots[0];
4412 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4414 if (key.objectid != device->devid) {
4415 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4416 btrfs_release_path(path);
4420 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4421 length = btrfs_dev_extent_length(l, dev_extent);
4423 if (key.offset + length <= new_size) {
4424 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4425 btrfs_release_path(path);
4429 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4430 btrfs_release_path(path);
4432 ret = btrfs_relocate_chunk(root, chunk_offset);
4433 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4434 if (ret && ret != -ENOSPC)
4438 } while (key.offset-- > 0);
4440 if (failed && !retried) {
4444 } else if (failed && retried) {
4449 /* Shrinking succeeded, else we would be at "done". */
4450 trans = btrfs_start_transaction(root, 0);
4451 if (IS_ERR(trans)) {
4452 ret = PTR_ERR(trans);
4459 * We checked in the above loop all device extents that were already in
4460 * the device tree. However before we have updated the device's
4461 * total_bytes to the new size, we might have had chunk allocations that
4462 * have not complete yet (new block groups attached to transaction
4463 * handles), and therefore their device extents were not yet in the
4464 * device tree and we missed them in the loop above. So if we have any
4465 * pending chunk using a device extent that overlaps the device range
4466 * that we can not use anymore, commit the current transaction and
4467 * repeat the search on the device tree - this way we guarantee we will
4468 * not have chunks using device extents that end beyond 'new_size'.
4470 if (!checked_pending_chunks) {
4471 u64 start = new_size;
4472 u64 len = old_size - new_size;
4474 if (contains_pending_extent(trans->transaction, device,
4476 unlock_chunks(root);
4477 checked_pending_chunks = true;
4480 ret = btrfs_commit_transaction(trans, root);
4487 btrfs_device_set_disk_total_bytes(device, new_size);
4488 if (list_empty(&device->resized_list))
4489 list_add_tail(&device->resized_list,
4490 &root->fs_info->fs_devices->resized_devices);
4492 WARN_ON(diff > old_total);
4493 btrfs_set_super_total_bytes(super_copy, old_total - diff);
4494 unlock_chunks(root);
4496 /* Now btrfs_update_device() will change the on-disk size. */
4497 ret = btrfs_update_device(trans, device);
4498 btrfs_end_transaction(trans, root);
4500 btrfs_free_path(path);
4503 btrfs_device_set_total_bytes(device, old_size);
4504 if (device->writeable)
4505 device->fs_devices->total_rw_bytes += diff;
4506 spin_lock(&root->fs_info->free_chunk_lock);
4507 root->fs_info->free_chunk_space += diff;
4508 spin_unlock(&root->fs_info->free_chunk_lock);
4509 unlock_chunks(root);
4514 static int btrfs_add_system_chunk(struct btrfs_root *root,
4515 struct btrfs_key *key,
4516 struct btrfs_chunk *chunk, int item_size)
4518 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4519 struct btrfs_disk_key disk_key;
4524 array_size = btrfs_super_sys_array_size(super_copy);
4525 if (array_size + item_size + sizeof(disk_key)
4526 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4527 unlock_chunks(root);
4531 ptr = super_copy->sys_chunk_array + array_size;
4532 btrfs_cpu_key_to_disk(&disk_key, key);
4533 memcpy(ptr, &disk_key, sizeof(disk_key));
4534 ptr += sizeof(disk_key);
4535 memcpy(ptr, chunk, item_size);
4536 item_size += sizeof(disk_key);
4537 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4538 unlock_chunks(root);
4544 * sort the devices in descending order by max_avail, total_avail
4546 static int btrfs_cmp_device_info(const void *a, const void *b)
4548 const struct btrfs_device_info *di_a = a;
4549 const struct btrfs_device_info *di_b = b;
4551 if (di_a->max_avail > di_b->max_avail)
4553 if (di_a->max_avail < di_b->max_avail)
4555 if (di_a->total_avail > di_b->total_avail)
4557 if (di_a->total_avail < di_b->total_avail)
4562 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
4564 /* TODO allow them to set a preferred stripe size */
4568 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4570 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4573 btrfs_set_fs_incompat(info, RAID56);
4576 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r) \
4577 - sizeof(struct btrfs_item) \
4578 - sizeof(struct btrfs_chunk)) \
4579 / sizeof(struct btrfs_stripe) + 1)
4581 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
4582 - 2 * sizeof(struct btrfs_disk_key) \
4583 - 2 * sizeof(struct btrfs_chunk)) \
4584 / sizeof(struct btrfs_stripe) + 1)
4586 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4587 struct btrfs_root *extent_root, u64 start,
4590 struct btrfs_fs_info *info = extent_root->fs_info;
4591 struct btrfs_fs_devices *fs_devices = info->fs_devices;
4592 struct list_head *cur;
4593 struct map_lookup *map = NULL;
4594 struct extent_map_tree *em_tree;
4595 struct extent_map *em;
4596 struct btrfs_device_info *devices_info = NULL;
4598 int num_stripes; /* total number of stripes to allocate */
4599 int data_stripes; /* number of stripes that count for
4601 int sub_stripes; /* sub_stripes info for map */
4602 int dev_stripes; /* stripes per dev */
4603 int devs_max; /* max devs to use */
4604 int devs_min; /* min devs needed */
4605 int devs_increment; /* ndevs has to be a multiple of this */
4606 int ncopies; /* how many copies to data has */
4608 u64 max_stripe_size;
4612 u64 raid_stripe_len = BTRFS_STRIPE_LEN;
4618 BUG_ON(!alloc_profile_is_valid(type, 0));
4620 if (list_empty(&fs_devices->alloc_list))
4623 index = __get_raid_index(type);
4625 sub_stripes = btrfs_raid_array[index].sub_stripes;
4626 dev_stripes = btrfs_raid_array[index].dev_stripes;
4627 devs_max = btrfs_raid_array[index].devs_max;
4628 devs_min = btrfs_raid_array[index].devs_min;
4629 devs_increment = btrfs_raid_array[index].devs_increment;
4630 ncopies = btrfs_raid_array[index].ncopies;
4632 if (type & BTRFS_BLOCK_GROUP_DATA) {
4633 max_stripe_size = SZ_1G;
4634 max_chunk_size = 10 * max_stripe_size;
4636 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4637 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4638 /* for larger filesystems, use larger metadata chunks */
4639 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
4640 max_stripe_size = SZ_1G;
4642 max_stripe_size = SZ_256M;
4643 max_chunk_size = max_stripe_size;
4645 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4646 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4647 max_stripe_size = SZ_32M;
4648 max_chunk_size = 2 * max_stripe_size;
4650 devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4652 btrfs_err(info, "invalid chunk type 0x%llx requested",
4657 /* we don't want a chunk larger than 10% of writeable space */
4658 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4661 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4666 cur = fs_devices->alloc_list.next;
4669 * in the first pass through the devices list, we gather information
4670 * about the available holes on each device.
4673 while (cur != &fs_devices->alloc_list) {
4674 struct btrfs_device *device;
4678 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4682 if (!device->writeable) {
4684 "BTRFS: read-only device in alloc_list\n");
4688 if (!device->in_fs_metadata ||
4689 device->is_tgtdev_for_dev_replace)
4692 if (device->total_bytes > device->bytes_used)
4693 total_avail = device->total_bytes - device->bytes_used;
4697 /* If there is no space on this device, skip it. */
4698 if (total_avail == 0)
4701 ret = find_free_dev_extent(trans, device,
4702 max_stripe_size * dev_stripes,
4703 &dev_offset, &max_avail);
4704 if (ret && ret != -ENOSPC)
4708 max_avail = max_stripe_size * dev_stripes;
4710 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4713 if (ndevs == fs_devices->rw_devices) {
4714 WARN(1, "%s: found more than %llu devices\n",
4715 __func__, fs_devices->rw_devices);
4718 devices_info[ndevs].dev_offset = dev_offset;
4719 devices_info[ndevs].max_avail = max_avail;
4720 devices_info[ndevs].total_avail = total_avail;
4721 devices_info[ndevs].dev = device;
4726 * now sort the devices by hole size / available space
4728 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4729 btrfs_cmp_device_info, NULL);
4731 /* round down to number of usable stripes */
4732 ndevs -= ndevs % devs_increment;
4734 if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4739 if (devs_max && ndevs > devs_max)
4742 * the primary goal is to maximize the number of stripes, so use as many
4743 * devices as possible, even if the stripes are not maximum sized.
4745 stripe_size = devices_info[ndevs-1].max_avail;
4746 num_stripes = ndevs * dev_stripes;
4749 * this will have to be fixed for RAID1 and RAID10 over
4752 data_stripes = num_stripes / ncopies;
4754 if (type & BTRFS_BLOCK_GROUP_RAID5) {
4755 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
4756 extent_root->stripesize);
4757 data_stripes = num_stripes - 1;
4759 if (type & BTRFS_BLOCK_GROUP_RAID6) {
4760 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
4761 extent_root->stripesize);
4762 data_stripes = num_stripes - 2;
4766 * Use the number of data stripes to figure out how big this chunk
4767 * is really going to be in terms of logical address space,
4768 * and compare that answer with the max chunk size
4770 if (stripe_size * data_stripes > max_chunk_size) {
4771 u64 mask = (1ULL << 24) - 1;
4773 stripe_size = div_u64(max_chunk_size, data_stripes);
4775 /* bump the answer up to a 16MB boundary */
4776 stripe_size = (stripe_size + mask) & ~mask;
4778 /* but don't go higher than the limits we found
4779 * while searching for free extents
4781 if (stripe_size > devices_info[ndevs-1].max_avail)
4782 stripe_size = devices_info[ndevs-1].max_avail;
4785 stripe_size = div_u64(stripe_size, dev_stripes);
4787 /* align to BTRFS_STRIPE_LEN */
4788 stripe_size = div_u64(stripe_size, raid_stripe_len);
4789 stripe_size *= raid_stripe_len;
4791 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4796 map->num_stripes = num_stripes;
4798 for (i = 0; i < ndevs; ++i) {
4799 for (j = 0; j < dev_stripes; ++j) {
4800 int s = i * dev_stripes + j;
4801 map->stripes[s].dev = devices_info[i].dev;
4802 map->stripes[s].physical = devices_info[i].dev_offset +
4806 map->sector_size = extent_root->sectorsize;
4807 map->stripe_len = raid_stripe_len;
4808 map->io_align = raid_stripe_len;
4809 map->io_width = raid_stripe_len;
4811 map->sub_stripes = sub_stripes;
4813 num_bytes = stripe_size * data_stripes;
4815 trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
4817 em = alloc_extent_map();
4823 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4824 em->map_lookup = map;
4826 em->len = num_bytes;
4827 em->block_start = 0;
4828 em->block_len = em->len;
4829 em->orig_block_len = stripe_size;
4831 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4832 write_lock(&em_tree->lock);
4833 ret = add_extent_mapping(em_tree, em, 0);
4835 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4836 atomic_inc(&em->refs);
4838 write_unlock(&em_tree->lock);
4840 free_extent_map(em);
4844 ret = btrfs_make_block_group(trans, extent_root, 0, type,
4845 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4848 goto error_del_extent;
4850 for (i = 0; i < map->num_stripes; i++) {
4851 num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4852 btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4855 spin_lock(&extent_root->fs_info->free_chunk_lock);
4856 extent_root->fs_info->free_chunk_space -= (stripe_size *
4858 spin_unlock(&extent_root->fs_info->free_chunk_lock);
4860 free_extent_map(em);
4861 check_raid56_incompat_flag(extent_root->fs_info, type);
4863 kfree(devices_info);
4867 write_lock(&em_tree->lock);
4868 remove_extent_mapping(em_tree, em);
4869 write_unlock(&em_tree->lock);
4871 /* One for our allocation */
4872 free_extent_map(em);
4873 /* One for the tree reference */
4874 free_extent_map(em);
4875 /* One for the pending_chunks list reference */
4876 free_extent_map(em);
4878 kfree(devices_info);
4882 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4883 struct btrfs_root *extent_root,
4884 u64 chunk_offset, u64 chunk_size)
4886 struct btrfs_key key;
4887 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4888 struct btrfs_device *device;
4889 struct btrfs_chunk *chunk;
4890 struct btrfs_stripe *stripe;
4891 struct extent_map_tree *em_tree;
4892 struct extent_map *em;
4893 struct map_lookup *map;
4900 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4901 read_lock(&em_tree->lock);
4902 em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size);
4903 read_unlock(&em_tree->lock);
4906 btrfs_crit(extent_root->fs_info, "unable to find logical "
4907 "%Lu len %Lu", chunk_offset, chunk_size);
4911 if (em->start != chunk_offset || em->len != chunk_size) {
4912 btrfs_crit(extent_root->fs_info, "found a bad mapping, wanted"
4913 " %Lu-%Lu, found %Lu-%Lu", chunk_offset,
4914 chunk_size, em->start, em->len);
4915 free_extent_map(em);
4919 map = em->map_lookup;
4920 item_size = btrfs_chunk_item_size(map->num_stripes);
4921 stripe_size = em->orig_block_len;
4923 chunk = kzalloc(item_size, GFP_NOFS);
4930 * Take the device list mutex to prevent races with the final phase of
4931 * a device replace operation that replaces the device object associated
4932 * with the map's stripes, because the device object's id can change
4933 * at any time during that final phase of the device replace operation
4934 * (dev-replace.c:btrfs_dev_replace_finishing()).
4936 mutex_lock(&chunk_root->fs_info->fs_devices->device_list_mutex);
4937 for (i = 0; i < map->num_stripes; i++) {
4938 device = map->stripes[i].dev;
4939 dev_offset = map->stripes[i].physical;
4941 ret = btrfs_update_device(trans, device);
4944 ret = btrfs_alloc_dev_extent(trans, device,
4945 chunk_root->root_key.objectid,
4946 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4947 chunk_offset, dev_offset,
4953 mutex_unlock(&chunk_root->fs_info->fs_devices->device_list_mutex);
4957 stripe = &chunk->stripe;
4958 for (i = 0; i < map->num_stripes; i++) {
4959 device = map->stripes[i].dev;
4960 dev_offset = map->stripes[i].physical;
4962 btrfs_set_stack_stripe_devid(stripe, device->devid);
4963 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4964 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4967 mutex_unlock(&chunk_root->fs_info->fs_devices->device_list_mutex);
4969 btrfs_set_stack_chunk_length(chunk, chunk_size);
4970 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4971 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4972 btrfs_set_stack_chunk_type(chunk, map->type);
4973 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4974 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4975 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4976 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4977 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4979 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4980 key.type = BTRFS_CHUNK_ITEM_KEY;
4981 key.offset = chunk_offset;
4983 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4984 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4986 * TODO: Cleanup of inserted chunk root in case of
4989 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
4995 free_extent_map(em);
5000 * Chunk allocation falls into two parts. The first part does works
5001 * that make the new allocated chunk useable, but not do any operation
5002 * that modifies the chunk tree. The second part does the works that
5003 * require modifying the chunk tree. This division is important for the
5004 * bootstrap process of adding storage to a seed btrfs.
5006 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
5007 struct btrfs_root *extent_root, u64 type)
5011 ASSERT(mutex_is_locked(&extent_root->fs_info->chunk_mutex));
5012 chunk_offset = find_next_chunk(extent_root->fs_info);
5013 return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type);
5016 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
5017 struct btrfs_root *root,
5018 struct btrfs_device *device)
5021 u64 sys_chunk_offset;
5023 struct btrfs_fs_info *fs_info = root->fs_info;
5024 struct btrfs_root *extent_root = fs_info->extent_root;
5027 chunk_offset = find_next_chunk(fs_info);
5028 alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
5029 ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset,
5034 sys_chunk_offset = find_next_chunk(root->fs_info);
5035 alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
5036 ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset,
5041 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
5045 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5046 BTRFS_BLOCK_GROUP_RAID10 |
5047 BTRFS_BLOCK_GROUP_RAID5 |
5048 BTRFS_BLOCK_GROUP_DUP)) {
5050 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
5059 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
5061 struct extent_map *em;
5062 struct map_lookup *map;
5063 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5068 read_lock(&map_tree->map_tree.lock);
5069 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
5070 read_unlock(&map_tree->map_tree.lock);
5074 map = em->map_lookup;
5075 for (i = 0; i < map->num_stripes; i++) {
5076 if (map->stripes[i].dev->missing) {
5081 if (!map->stripes[i].dev->writeable) {
5088 * If the number of missing devices is larger than max errors,
5089 * we can not write the data into that chunk successfully, so
5092 if (miss_ndevs > btrfs_chunk_max_errors(map))
5095 free_extent_map(em);
5099 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
5101 extent_map_tree_init(&tree->map_tree);
5104 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
5106 struct extent_map *em;
5109 write_lock(&tree->map_tree.lock);
5110 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
5112 remove_extent_mapping(&tree->map_tree, em);
5113 write_unlock(&tree->map_tree.lock);
5117 free_extent_map(em);
5118 /* once for the tree */
5119 free_extent_map(em);
5123 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5125 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5126 struct extent_map *em;
5127 struct map_lookup *map;
5128 struct extent_map_tree *em_tree = &map_tree->map_tree;
5131 read_lock(&em_tree->lock);
5132 em = lookup_extent_mapping(em_tree, logical, len);
5133 read_unlock(&em_tree->lock);
5136 * We could return errors for these cases, but that could get ugly and
5137 * we'd probably do the same thing which is just not do anything else
5138 * and exit, so return 1 so the callers don't try to use other copies.
5141 btrfs_crit(fs_info, "No mapping for %Lu-%Lu", logical,
5146 if (em->start > logical || em->start + em->len < logical) {
5147 btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got "
5148 "%Lu-%Lu", logical, logical+len, em->start,
5149 em->start + em->len);
5150 free_extent_map(em);
5154 map = em->map_lookup;
5155 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
5156 ret = map->num_stripes;
5157 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5158 ret = map->sub_stripes;
5159 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5161 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5165 free_extent_map(em);
5167 btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
5168 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
5170 btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
5175 unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
5176 struct btrfs_mapping_tree *map_tree,
5179 struct extent_map *em;
5180 struct map_lookup *map;
5181 struct extent_map_tree *em_tree = &map_tree->map_tree;
5182 unsigned long len = root->sectorsize;
5184 read_lock(&em_tree->lock);
5185 em = lookup_extent_mapping(em_tree, logical, len);
5186 read_unlock(&em_tree->lock);
5189 BUG_ON(em->start > logical || em->start + em->len < logical);
5190 map = em->map_lookup;
5191 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5192 len = map->stripe_len * nr_data_stripes(map);
5193 free_extent_map(em);
5197 int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
5198 u64 logical, u64 len, int mirror_num)
5200 struct extent_map *em;
5201 struct map_lookup *map;
5202 struct extent_map_tree *em_tree = &map_tree->map_tree;
5205 read_lock(&em_tree->lock);
5206 em = lookup_extent_mapping(em_tree, logical, len);
5207 read_unlock(&em_tree->lock);
5210 BUG_ON(em->start > logical || em->start + em->len < logical);
5211 map = em->map_lookup;
5212 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5214 free_extent_map(em);
5218 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5219 struct map_lookup *map, int first, int num,
5220 int optimal, int dev_replace_is_ongoing)
5224 struct btrfs_device *srcdev;
5226 if (dev_replace_is_ongoing &&
5227 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5228 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5229 srcdev = fs_info->dev_replace.srcdev;
5234 * try to avoid the drive that is the source drive for a
5235 * dev-replace procedure, only choose it if no other non-missing
5236 * mirror is available
5238 for (tolerance = 0; tolerance < 2; tolerance++) {
5239 if (map->stripes[optimal].dev->bdev &&
5240 (tolerance || map->stripes[optimal].dev != srcdev))
5242 for (i = first; i < first + num; i++) {
5243 if (map->stripes[i].dev->bdev &&
5244 (tolerance || map->stripes[i].dev != srcdev))
5249 /* we couldn't find one that doesn't fail. Just return something
5250 * and the io error handling code will clean up eventually
5255 static inline int parity_smaller(u64 a, u64 b)
5260 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5261 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5263 struct btrfs_bio_stripe s;
5270 for (i = 0; i < num_stripes - 1; i++) {
5271 if (parity_smaller(bbio->raid_map[i],
5272 bbio->raid_map[i+1])) {
5273 s = bbio->stripes[i];
5274 l = bbio->raid_map[i];
5275 bbio->stripes[i] = bbio->stripes[i+1];
5276 bbio->raid_map[i] = bbio->raid_map[i+1];
5277 bbio->stripes[i+1] = s;
5278 bbio->raid_map[i+1] = l;
5286 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5288 struct btrfs_bio *bbio = kzalloc(
5289 /* the size of the btrfs_bio */
5290 sizeof(struct btrfs_bio) +
5291 /* plus the variable array for the stripes */
5292 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5293 /* plus the variable array for the tgt dev */
5294 sizeof(int) * (real_stripes) +
5296 * plus the raid_map, which includes both the tgt dev
5299 sizeof(u64) * (total_stripes),
5300 GFP_NOFS|__GFP_NOFAIL);
5302 atomic_set(&bbio->error, 0);
5303 atomic_set(&bbio->refs, 1);
5308 void btrfs_get_bbio(struct btrfs_bio *bbio)
5310 WARN_ON(!atomic_read(&bbio->refs));
5311 atomic_inc(&bbio->refs);
5314 void btrfs_put_bbio(struct btrfs_bio *bbio)
5318 if (atomic_dec_and_test(&bbio->refs))
5322 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5323 u64 logical, u64 *length,
5324 struct btrfs_bio **bbio_ret,
5325 int mirror_num, int need_raid_map)
5327 struct extent_map *em;
5328 struct map_lookup *map;
5329 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5330 struct extent_map_tree *em_tree = &map_tree->map_tree;
5333 u64 stripe_end_offset;
5343 int tgtdev_indexes = 0;
5344 struct btrfs_bio *bbio = NULL;
5345 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
5346 int dev_replace_is_ongoing = 0;
5347 int num_alloc_stripes;
5348 int patch_the_first_stripe_for_dev_replace = 0;
5349 u64 physical_to_patch_in_first_stripe = 0;
5350 u64 raid56_full_stripe_start = (u64)-1;
5352 read_lock(&em_tree->lock);
5353 em = lookup_extent_mapping(em_tree, logical, *length);
5354 read_unlock(&em_tree->lock);
5357 btrfs_crit(fs_info, "unable to find logical %llu len %llu",
5362 if (em->start > logical || em->start + em->len < logical) {
5363 btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, "
5364 "found %Lu-%Lu", logical, em->start,
5365 em->start + em->len);
5366 free_extent_map(em);
5370 map = em->map_lookup;
5371 offset = logical - em->start;
5373 stripe_len = map->stripe_len;
5376 * stripe_nr counts the total number of stripes we have to stride
5377 * to get to this block
5379 stripe_nr = div64_u64(stripe_nr, stripe_len);
5381 stripe_offset = stripe_nr * stripe_len;
5382 if (offset < stripe_offset) {
5383 btrfs_crit(fs_info, "stripe math has gone wrong, "
5384 "stripe_offset=%llu, offset=%llu, start=%llu, "
5385 "logical=%llu, stripe_len=%llu",
5386 stripe_offset, offset, em->start, logical,
5388 free_extent_map(em);
5392 /* stripe_offset is the offset of this block in its stripe*/
5393 stripe_offset = offset - stripe_offset;
5395 /* if we're here for raid56, we need to know the stripe aligned start */
5396 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5397 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5398 raid56_full_stripe_start = offset;
5400 /* allow a write of a full stripe, but make sure we don't
5401 * allow straddling of stripes
5403 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5405 raid56_full_stripe_start *= full_stripe_len;
5408 if (rw & REQ_DISCARD) {
5409 /* we don't discard raid56 yet */
5410 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5414 *length = min_t(u64, em->len - offset, *length);
5415 } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5417 /* For writes to RAID[56], allow a full stripeset across all disks.
5418 For other RAID types and for RAID[56] reads, just allow a single
5419 stripe (on a single disk). */
5420 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
5422 max_len = stripe_len * nr_data_stripes(map) -
5423 (offset - raid56_full_stripe_start);
5425 /* we limit the length of each bio to what fits in a stripe */
5426 max_len = stripe_len - stripe_offset;
5428 *length = min_t(u64, em->len - offset, max_len);
5430 *length = em->len - offset;
5433 /* This is for when we're called from btrfs_merge_bio_hook() and all
5434 it cares about is the length */
5438 btrfs_dev_replace_lock(dev_replace, 0);
5439 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5440 if (!dev_replace_is_ongoing)
5441 btrfs_dev_replace_unlock(dev_replace, 0);
5443 btrfs_dev_replace_set_lock_blocking(dev_replace);
5445 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
5446 !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) &&
5447 dev_replace->tgtdev != NULL) {
5449 * in dev-replace case, for repair case (that's the only
5450 * case where the mirror is selected explicitly when
5451 * calling btrfs_map_block), blocks left of the left cursor
5452 * can also be read from the target drive.
5453 * For REQ_GET_READ_MIRRORS, the target drive is added as
5454 * the last one to the array of stripes. For READ, it also
5455 * needs to be supported using the same mirror number.
5456 * If the requested block is not left of the left cursor,
5457 * EIO is returned. This can happen because btrfs_num_copies()
5458 * returns one more in the dev-replace case.
5460 u64 tmp_length = *length;
5461 struct btrfs_bio *tmp_bbio = NULL;
5462 int tmp_num_stripes;
5463 u64 srcdev_devid = dev_replace->srcdev->devid;
5464 int index_srcdev = 0;
5466 u64 physical_of_found = 0;
5468 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
5469 logical, &tmp_length, &tmp_bbio, 0, 0);
5471 WARN_ON(tmp_bbio != NULL);
5475 tmp_num_stripes = tmp_bbio->num_stripes;
5476 if (mirror_num > tmp_num_stripes) {
5478 * REQ_GET_READ_MIRRORS does not contain this
5479 * mirror, that means that the requested area
5480 * is not left of the left cursor
5483 btrfs_put_bbio(tmp_bbio);
5488 * process the rest of the function using the mirror_num
5489 * of the source drive. Therefore look it up first.
5490 * At the end, patch the device pointer to the one of the
5493 for (i = 0; i < tmp_num_stripes; i++) {
5494 if (tmp_bbio->stripes[i].dev->devid != srcdev_devid)
5498 * In case of DUP, in order to keep it simple, only add
5499 * the mirror with the lowest physical address
5502 physical_of_found <= tmp_bbio->stripes[i].physical)
5507 physical_of_found = tmp_bbio->stripes[i].physical;
5510 btrfs_put_bbio(tmp_bbio);
5518 mirror_num = index_srcdev + 1;
5519 patch_the_first_stripe_for_dev_replace = 1;
5520 physical_to_patch_in_first_stripe = physical_of_found;
5521 } else if (mirror_num > map->num_stripes) {
5527 stripe_nr_orig = stripe_nr;
5528 stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
5529 stripe_nr_end = div_u64(stripe_nr_end, map->stripe_len);
5530 stripe_end_offset = stripe_nr_end * map->stripe_len -
5533 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5534 if (rw & REQ_DISCARD)
5535 num_stripes = min_t(u64, map->num_stripes,
5536 stripe_nr_end - stripe_nr_orig);
5537 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5539 if (!(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)))
5541 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
5542 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))
5543 num_stripes = map->num_stripes;
5544 else if (mirror_num)
5545 stripe_index = mirror_num - 1;
5547 stripe_index = find_live_mirror(fs_info, map, 0,
5549 current->pid % map->num_stripes,
5550 dev_replace_is_ongoing);
5551 mirror_num = stripe_index + 1;
5554 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
5555 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
5556 num_stripes = map->num_stripes;
5557 } else if (mirror_num) {
5558 stripe_index = mirror_num - 1;
5563 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5564 u32 factor = map->num_stripes / map->sub_stripes;
5566 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5567 stripe_index *= map->sub_stripes;
5569 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
5570 num_stripes = map->sub_stripes;
5571 else if (rw & REQ_DISCARD)
5572 num_stripes = min_t(u64, map->sub_stripes *
5573 (stripe_nr_end - stripe_nr_orig),
5575 else if (mirror_num)
5576 stripe_index += mirror_num - 1;
5578 int old_stripe_index = stripe_index;
5579 stripe_index = find_live_mirror(fs_info, map,
5581 map->sub_stripes, stripe_index +
5582 current->pid % map->sub_stripes,
5583 dev_replace_is_ongoing);
5584 mirror_num = stripe_index - old_stripe_index + 1;
5587 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5588 if (need_raid_map &&
5589 ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) ||
5591 /* push stripe_nr back to the start of the full stripe */
5592 stripe_nr = div_u64(raid56_full_stripe_start,
5593 stripe_len * nr_data_stripes(map));
5595 /* RAID[56] write or recovery. Return all stripes */
5596 num_stripes = map->num_stripes;
5597 max_errors = nr_parity_stripes(map);
5599 *length = map->stripe_len;
5604 * Mirror #0 or #1 means the original data block.
5605 * Mirror #2 is RAID5 parity block.
5606 * Mirror #3 is RAID6 Q block.
5608 stripe_nr = div_u64_rem(stripe_nr,
5609 nr_data_stripes(map), &stripe_index);
5611 stripe_index = nr_data_stripes(map) +
5614 /* We distribute the parity blocks across stripes */
5615 div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
5617 if (!(rw & (REQ_WRITE | REQ_DISCARD |
5618 REQ_GET_READ_MIRRORS)) && mirror_num <= 1)
5623 * after this, stripe_nr is the number of stripes on this
5624 * device we have to walk to find the data, and stripe_index is
5625 * the number of our device in the stripe array
5627 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5629 mirror_num = stripe_index + 1;
5631 if (stripe_index >= map->num_stripes) {
5632 btrfs_crit(fs_info, "stripe index math went horribly wrong, "
5633 "got stripe_index=%u, num_stripes=%u",
5634 stripe_index, map->num_stripes);
5639 num_alloc_stripes = num_stripes;
5640 if (dev_replace_is_ongoing) {
5641 if (rw & (REQ_WRITE | REQ_DISCARD))
5642 num_alloc_stripes <<= 1;
5643 if (rw & REQ_GET_READ_MIRRORS)
5644 num_alloc_stripes++;
5645 tgtdev_indexes = num_stripes;
5648 bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
5653 if (dev_replace_is_ongoing)
5654 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
5656 /* build raid_map */
5657 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK &&
5658 need_raid_map && ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) ||
5663 bbio->raid_map = (u64 *)((void *)bbio->stripes +
5664 sizeof(struct btrfs_bio_stripe) *
5666 sizeof(int) * tgtdev_indexes);
5668 /* Work out the disk rotation on this stripe-set */
5669 div_u64_rem(stripe_nr, num_stripes, &rot);
5671 /* Fill in the logical address of each stripe */
5672 tmp = stripe_nr * nr_data_stripes(map);
5673 for (i = 0; i < nr_data_stripes(map); i++)
5674 bbio->raid_map[(i+rot) % num_stripes] =
5675 em->start + (tmp + i) * map->stripe_len;
5677 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5678 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5679 bbio->raid_map[(i+rot+1) % num_stripes] =
5683 if (rw & REQ_DISCARD) {
5685 u32 sub_stripes = 0;
5686 u64 stripes_per_dev = 0;
5687 u32 remaining_stripes = 0;
5688 u32 last_stripe = 0;
5691 (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
5692 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5695 sub_stripes = map->sub_stripes;
5697 factor = map->num_stripes / sub_stripes;
5698 stripes_per_dev = div_u64_rem(stripe_nr_end -
5701 &remaining_stripes);
5702 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5703 last_stripe *= sub_stripes;
5706 for (i = 0; i < num_stripes; i++) {
5707 bbio->stripes[i].physical =
5708 map->stripes[stripe_index].physical +
5709 stripe_offset + stripe_nr * map->stripe_len;
5710 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5712 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5713 BTRFS_BLOCK_GROUP_RAID10)) {
5714 bbio->stripes[i].length = stripes_per_dev *
5717 if (i / sub_stripes < remaining_stripes)
5718 bbio->stripes[i].length +=
5722 * Special for the first stripe and
5725 * |-------|...|-------|
5729 if (i < sub_stripes)
5730 bbio->stripes[i].length -=
5733 if (stripe_index >= last_stripe &&
5734 stripe_index <= (last_stripe +
5736 bbio->stripes[i].length -=
5739 if (i == sub_stripes - 1)
5742 bbio->stripes[i].length = *length;
5745 if (stripe_index == map->num_stripes) {
5746 /* This could only happen for RAID0/10 */
5752 for (i = 0; i < num_stripes; i++) {
5753 bbio->stripes[i].physical =
5754 map->stripes[stripe_index].physical +
5756 stripe_nr * map->stripe_len;
5757 bbio->stripes[i].dev =
5758 map->stripes[stripe_index].dev;
5763 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
5764 max_errors = btrfs_chunk_max_errors(map);
5767 sort_parity_stripes(bbio, num_stripes);
5770 if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) &&
5771 dev_replace->tgtdev != NULL) {
5772 int index_where_to_add;
5773 u64 srcdev_devid = dev_replace->srcdev->devid;
5776 * duplicate the write operations while the dev replace
5777 * procedure is running. Since the copying of the old disk
5778 * to the new disk takes place at run time while the
5779 * filesystem is mounted writable, the regular write
5780 * operations to the old disk have to be duplicated to go
5781 * to the new disk as well.
5782 * Note that device->missing is handled by the caller, and
5783 * that the write to the old disk is already set up in the
5786 index_where_to_add = num_stripes;
5787 for (i = 0; i < num_stripes; i++) {
5788 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5789 /* write to new disk, too */
5790 struct btrfs_bio_stripe *new =
5791 bbio->stripes + index_where_to_add;
5792 struct btrfs_bio_stripe *old =
5795 new->physical = old->physical;
5796 new->length = old->length;
5797 new->dev = dev_replace->tgtdev;
5798 bbio->tgtdev_map[i] = index_where_to_add;
5799 index_where_to_add++;
5804 num_stripes = index_where_to_add;
5805 } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) &&
5806 dev_replace->tgtdev != NULL) {
5807 u64 srcdev_devid = dev_replace->srcdev->devid;
5808 int index_srcdev = 0;
5810 u64 physical_of_found = 0;
5813 * During the dev-replace procedure, the target drive can
5814 * also be used to read data in case it is needed to repair
5815 * a corrupt block elsewhere. This is possible if the
5816 * requested area is left of the left cursor. In this area,
5817 * the target drive is a full copy of the source drive.
5819 for (i = 0; i < num_stripes; i++) {
5820 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5822 * In case of DUP, in order to keep it
5823 * simple, only add the mirror with the
5824 * lowest physical address
5827 physical_of_found <=
5828 bbio->stripes[i].physical)
5832 physical_of_found = bbio->stripes[i].physical;
5836 struct btrfs_bio_stripe *tgtdev_stripe =
5837 bbio->stripes + num_stripes;
5839 tgtdev_stripe->physical = physical_of_found;
5840 tgtdev_stripe->length =
5841 bbio->stripes[index_srcdev].length;
5842 tgtdev_stripe->dev = dev_replace->tgtdev;
5843 bbio->tgtdev_map[index_srcdev] = num_stripes;
5851 bbio->map_type = map->type;
5852 bbio->num_stripes = num_stripes;
5853 bbio->max_errors = max_errors;
5854 bbio->mirror_num = mirror_num;
5855 bbio->num_tgtdevs = tgtdev_indexes;
5858 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5859 * mirror_num == num_stripes + 1 && dev_replace target drive is
5860 * available as a mirror
5862 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5863 WARN_ON(num_stripes > 1);
5864 bbio->stripes[0].dev = dev_replace->tgtdev;
5865 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5866 bbio->mirror_num = map->num_stripes + 1;
5869 if (dev_replace_is_ongoing) {
5870 btrfs_dev_replace_clear_lock_blocking(dev_replace);
5871 btrfs_dev_replace_unlock(dev_replace, 0);
5873 free_extent_map(em);
5877 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5878 u64 logical, u64 *length,
5879 struct btrfs_bio **bbio_ret, int mirror_num)
5881 return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5885 /* For Scrub/replace */
5886 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, int rw,
5887 u64 logical, u64 *length,
5888 struct btrfs_bio **bbio_ret, int mirror_num,
5891 return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5892 mirror_num, need_raid_map);
5895 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
5896 u64 chunk_start, u64 physical, u64 devid,
5897 u64 **logical, int *naddrs, int *stripe_len)
5899 struct extent_map_tree *em_tree = &map_tree->map_tree;
5900 struct extent_map *em;
5901 struct map_lookup *map;
5909 read_lock(&em_tree->lock);
5910 em = lookup_extent_mapping(em_tree, chunk_start, 1);
5911 read_unlock(&em_tree->lock);
5914 printk(KERN_ERR "BTRFS: couldn't find em for chunk %Lu\n",
5919 if (em->start != chunk_start) {
5920 printk(KERN_ERR "BTRFS: bad chunk start, em=%Lu, wanted=%Lu\n",
5921 em->start, chunk_start);
5922 free_extent_map(em);
5925 map = em->map_lookup;
5928 rmap_len = map->stripe_len;
5930 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5931 length = div_u64(length, map->num_stripes / map->sub_stripes);
5932 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5933 length = div_u64(length, map->num_stripes);
5934 else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5935 length = div_u64(length, nr_data_stripes(map));
5936 rmap_len = map->stripe_len * nr_data_stripes(map);
5939 buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
5940 BUG_ON(!buf); /* -ENOMEM */
5942 for (i = 0; i < map->num_stripes; i++) {
5943 if (devid && map->stripes[i].dev->devid != devid)
5945 if (map->stripes[i].physical > physical ||
5946 map->stripes[i].physical + length <= physical)
5949 stripe_nr = physical - map->stripes[i].physical;
5950 stripe_nr = div_u64(stripe_nr, map->stripe_len);
5952 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5953 stripe_nr = stripe_nr * map->num_stripes + i;
5954 stripe_nr = div_u64(stripe_nr, map->sub_stripes);
5955 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5956 stripe_nr = stripe_nr * map->num_stripes + i;
5957 } /* else if RAID[56], multiply by nr_data_stripes().
5958 * Alternatively, just use rmap_len below instead of
5959 * map->stripe_len */
5961 bytenr = chunk_start + stripe_nr * rmap_len;
5962 WARN_ON(nr >= map->num_stripes);
5963 for (j = 0; j < nr; j++) {
5964 if (buf[j] == bytenr)
5968 WARN_ON(nr >= map->num_stripes);
5975 *stripe_len = rmap_len;
5977 free_extent_map(em);
5981 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
5983 bio->bi_private = bbio->private;
5984 bio->bi_end_io = bbio->end_io;
5987 btrfs_put_bbio(bbio);
5990 static void btrfs_end_bio(struct bio *bio)
5992 struct btrfs_bio *bbio = bio->bi_private;
5993 int is_orig_bio = 0;
5995 if (bio->bi_error) {
5996 atomic_inc(&bbio->error);
5997 if (bio->bi_error == -EIO || bio->bi_error == -EREMOTEIO) {
5998 unsigned int stripe_index =
5999 btrfs_io_bio(bio)->stripe_index;
6000 struct btrfs_device *dev;
6002 BUG_ON(stripe_index >= bbio->num_stripes);
6003 dev = bbio->stripes[stripe_index].dev;
6005 if (bio->bi_rw & WRITE)
6006 btrfs_dev_stat_inc(dev,
6007 BTRFS_DEV_STAT_WRITE_ERRS);
6009 btrfs_dev_stat_inc(dev,
6010 BTRFS_DEV_STAT_READ_ERRS);
6011 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
6012 btrfs_dev_stat_inc(dev,
6013 BTRFS_DEV_STAT_FLUSH_ERRS);
6014 btrfs_dev_stat_print_on_error(dev);
6019 if (bio == bbio->orig_bio)
6022 btrfs_bio_counter_dec(bbio->fs_info);
6024 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6027 bio = bbio->orig_bio;
6030 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6031 /* only send an error to the higher layers if it is
6032 * beyond the tolerance of the btrfs bio
6034 if (atomic_read(&bbio->error) > bbio->max_errors) {
6035 bio->bi_error = -EIO;
6038 * this bio is actually up to date, we didn't
6039 * go over the max number of errors
6044 btrfs_end_bbio(bbio, bio);
6045 } else if (!is_orig_bio) {
6051 * see run_scheduled_bios for a description of why bios are collected for
6054 * This will add one bio to the pending list for a device and make sure
6055 * the work struct is scheduled.
6057 static noinline void btrfs_schedule_bio(struct btrfs_root *root,
6058 struct btrfs_device *device,
6059 int rw, struct bio *bio)
6061 int should_queue = 1;
6062 struct btrfs_pending_bios *pending_bios;
6064 if (device->missing || !device->bdev) {
6069 /* don't bother with additional async steps for reads, right now */
6070 if (!(rw & REQ_WRITE)) {
6072 btrfsic_submit_bio(rw, bio);
6078 * nr_async_bios allows us to reliably return congestion to the
6079 * higher layers. Otherwise, the async bio makes it appear we have
6080 * made progress against dirty pages when we've really just put it
6081 * on a queue for later
6083 atomic_inc(&root->fs_info->nr_async_bios);
6084 WARN_ON(bio->bi_next);
6085 bio->bi_next = NULL;
6088 spin_lock(&device->io_lock);
6089 if (bio->bi_rw & REQ_SYNC)
6090 pending_bios = &device->pending_sync_bios;
6092 pending_bios = &device->pending_bios;
6094 if (pending_bios->tail)
6095 pending_bios->tail->bi_next = bio;
6097 pending_bios->tail = bio;
6098 if (!pending_bios->head)
6099 pending_bios->head = bio;
6100 if (device->running_pending)
6103 spin_unlock(&device->io_lock);
6106 btrfs_queue_work(root->fs_info->submit_workers,
6110 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
6111 struct bio *bio, u64 physical, int dev_nr,
6114 struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
6116 bio->bi_private = bbio;
6117 btrfs_io_bio(bio)->stripe_index = dev_nr;
6118 bio->bi_end_io = btrfs_end_bio;
6119 bio->bi_iter.bi_sector = physical >> 9;
6122 struct rcu_string *name;
6125 name = rcu_dereference(dev->name);
6126 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
6127 "(%s id %llu), size=%u\n", rw,
6128 (u64)bio->bi_iter.bi_sector, (u_long)dev->bdev->bd_dev,
6129 name->str, dev->devid, bio->bi_iter.bi_size);
6133 bio->bi_bdev = dev->bdev;
6135 btrfs_bio_counter_inc_noblocked(root->fs_info);
6138 btrfs_schedule_bio(root, dev, rw, bio);
6140 btrfsic_submit_bio(rw, bio);
6143 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6145 atomic_inc(&bbio->error);
6146 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6147 /* Should be the original bio. */
6148 WARN_ON(bio != bbio->orig_bio);
6150 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6151 bio->bi_iter.bi_sector = logical >> 9;
6152 bio->bi_error = -EIO;
6153 btrfs_end_bbio(bbio, bio);
6157 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
6158 int mirror_num, int async_submit)
6160 struct btrfs_device *dev;
6161 struct bio *first_bio = bio;
6162 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
6168 struct btrfs_bio *bbio = NULL;
6170 length = bio->bi_iter.bi_size;
6171 map_length = length;
6173 btrfs_bio_counter_inc_blocked(root->fs_info);
6174 ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
6177 btrfs_bio_counter_dec(root->fs_info);
6181 total_devs = bbio->num_stripes;
6182 bbio->orig_bio = first_bio;
6183 bbio->private = first_bio->bi_private;
6184 bbio->end_io = first_bio->bi_end_io;
6185 bbio->fs_info = root->fs_info;
6186 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6188 if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6189 ((rw & WRITE) || (mirror_num > 1))) {
6190 /* In this case, map_length has been set to the length of
6191 a single stripe; not the whole write */
6193 ret = raid56_parity_write(root, bio, bbio, map_length);
6195 ret = raid56_parity_recover(root, bio, bbio, map_length,
6199 btrfs_bio_counter_dec(root->fs_info);
6203 if (map_length < length) {
6204 btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu",
6205 logical, length, map_length);
6209 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6210 dev = bbio->stripes[dev_nr].dev;
6211 if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
6212 bbio_error(bbio, first_bio, logical);
6216 if (dev_nr < total_devs - 1) {
6217 bio = btrfs_bio_clone(first_bio, GFP_NOFS);
6218 BUG_ON(!bio); /* -ENOMEM */
6222 submit_stripe_bio(root, bbio, bio,
6223 bbio->stripes[dev_nr].physical, dev_nr, rw,
6226 btrfs_bio_counter_dec(root->fs_info);
6230 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
6233 struct btrfs_device *device;
6234 struct btrfs_fs_devices *cur_devices;
6236 cur_devices = fs_info->fs_devices;
6237 while (cur_devices) {
6239 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
6240 device = __find_device(&cur_devices->devices,
6245 cur_devices = cur_devices->seed;
6250 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
6251 struct btrfs_fs_devices *fs_devices,
6252 u64 devid, u8 *dev_uuid)
6254 struct btrfs_device *device;
6256 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6260 list_add(&device->dev_list, &fs_devices->devices);
6261 device->fs_devices = fs_devices;
6262 fs_devices->num_devices++;
6264 device->missing = 1;
6265 fs_devices->missing_devices++;
6271 * btrfs_alloc_device - allocate struct btrfs_device
6272 * @fs_info: used only for generating a new devid, can be NULL if
6273 * devid is provided (i.e. @devid != NULL).
6274 * @devid: a pointer to devid for this device. If NULL a new devid
6276 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6279 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6280 * on error. Returned struct is not linked onto any lists and can be
6281 * destroyed with kfree() right away.
6283 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6287 struct btrfs_device *dev;
6290 if (WARN_ON(!devid && !fs_info))
6291 return ERR_PTR(-EINVAL);
6293 dev = __alloc_device();
6302 ret = find_next_devid(fs_info, &tmp);
6305 return ERR_PTR(ret);
6311 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6313 generate_random_uuid(dev->uuid);
6315 btrfs_init_work(&dev->work, btrfs_submit_helper,
6316 pending_bios_fn, NULL, NULL);
6321 /* Return -EIO if any error, otherwise return 0. */
6322 static int btrfs_check_chunk_valid(struct btrfs_root *root,
6323 struct extent_buffer *leaf,
6324 struct btrfs_chunk *chunk, u64 logical)
6332 length = btrfs_chunk_length(leaf, chunk);
6333 stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6334 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6335 sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6336 type = btrfs_chunk_type(leaf, chunk);
6339 btrfs_err(root->fs_info, "invalid chunk num_stripes: %u",
6343 if (!IS_ALIGNED(logical, root->sectorsize)) {
6344 btrfs_err(root->fs_info,
6345 "invalid chunk logical %llu", logical);
6348 if (btrfs_chunk_sector_size(leaf, chunk) != root->sectorsize) {
6349 btrfs_err(root->fs_info, "invalid chunk sectorsize %u",
6350 btrfs_chunk_sector_size(leaf, chunk));
6353 if (!length || !IS_ALIGNED(length, root->sectorsize)) {
6354 btrfs_err(root->fs_info,
6355 "invalid chunk length %llu", length);
6358 if (!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN) {
6359 btrfs_err(root->fs_info, "invalid chunk stripe length: %llu",
6363 if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6365 btrfs_err(root->fs_info, "unrecognized chunk type: %llu",
6366 ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
6367 BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6368 btrfs_chunk_type(leaf, chunk));
6371 if ((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes != 2) ||
6372 (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) ||
6373 (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) ||
6374 (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) ||
6375 (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) ||
6376 ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
6377 num_stripes != 1)) {
6378 btrfs_err(root->fs_info,
6379 "invalid num_stripes:sub_stripes %u:%u for profile %llu",
6380 num_stripes, sub_stripes,
6381 type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
6388 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
6389 struct extent_buffer *leaf,
6390 struct btrfs_chunk *chunk)
6392 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
6393 struct map_lookup *map;
6394 struct extent_map *em;
6399 u8 uuid[BTRFS_UUID_SIZE];
6404 logical = key->offset;
6405 length = btrfs_chunk_length(leaf, chunk);
6406 stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6407 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6409 ret = btrfs_check_chunk_valid(root, leaf, chunk, logical);
6413 read_lock(&map_tree->map_tree.lock);
6414 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
6415 read_unlock(&map_tree->map_tree.lock);
6417 /* already mapped? */
6418 if (em && em->start <= logical && em->start + em->len > logical) {
6419 free_extent_map(em);
6422 free_extent_map(em);
6425 em = alloc_extent_map();
6428 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6430 free_extent_map(em);
6434 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6435 em->map_lookup = map;
6436 em->start = logical;
6439 em->block_start = 0;
6440 em->block_len = em->len;
6442 map->num_stripes = num_stripes;
6443 map->io_width = btrfs_chunk_io_width(leaf, chunk);
6444 map->io_align = btrfs_chunk_io_align(leaf, chunk);
6445 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
6446 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6447 map->type = btrfs_chunk_type(leaf, chunk);
6448 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6449 for (i = 0; i < num_stripes; i++) {
6450 map->stripes[i].physical =
6451 btrfs_stripe_offset_nr(leaf, chunk, i);
6452 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6453 read_extent_buffer(leaf, uuid, (unsigned long)
6454 btrfs_stripe_dev_uuid_nr(chunk, i),
6456 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
6458 if (!map->stripes[i].dev &&
6459 !btrfs_test_opt(root->fs_info, DEGRADED)) {
6460 free_extent_map(em);
6463 if (!map->stripes[i].dev) {
6464 map->stripes[i].dev =
6465 add_missing_dev(root, root->fs_info->fs_devices,
6467 if (!map->stripes[i].dev) {
6468 free_extent_map(em);
6471 btrfs_warn(root->fs_info, "devid %llu uuid %pU is missing",
6474 map->stripes[i].dev->in_fs_metadata = 1;
6477 write_lock(&map_tree->map_tree.lock);
6478 ret = add_extent_mapping(&map_tree->map_tree, em, 0);
6479 write_unlock(&map_tree->map_tree.lock);
6480 BUG_ON(ret); /* Tree corruption */
6481 free_extent_map(em);
6486 static void fill_device_from_item(struct extent_buffer *leaf,
6487 struct btrfs_dev_item *dev_item,
6488 struct btrfs_device *device)
6492 device->devid = btrfs_device_id(leaf, dev_item);
6493 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6494 device->total_bytes = device->disk_total_bytes;
6495 device->commit_total_bytes = device->disk_total_bytes;
6496 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6497 device->commit_bytes_used = device->bytes_used;
6498 device->type = btrfs_device_type(leaf, dev_item);
6499 device->io_align = btrfs_device_io_align(leaf, dev_item);
6500 device->io_width = btrfs_device_io_width(leaf, dev_item);
6501 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6502 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6503 device->is_tgtdev_for_dev_replace = 0;
6505 ptr = btrfs_device_uuid(dev_item);
6506 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6509 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_root *root,
6512 struct btrfs_fs_devices *fs_devices;
6515 BUG_ON(!mutex_is_locked(&uuid_mutex));
6517 fs_devices = root->fs_info->fs_devices->seed;
6518 while (fs_devices) {
6519 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE))
6522 fs_devices = fs_devices->seed;
6525 fs_devices = find_fsid(fsid);
6527 if (!btrfs_test_opt(root->fs_info, DEGRADED))
6528 return ERR_PTR(-ENOENT);
6530 fs_devices = alloc_fs_devices(fsid);
6531 if (IS_ERR(fs_devices))
6534 fs_devices->seeding = 1;
6535 fs_devices->opened = 1;
6539 fs_devices = clone_fs_devices(fs_devices);
6540 if (IS_ERR(fs_devices))
6543 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
6544 root->fs_info->bdev_holder);
6546 free_fs_devices(fs_devices);
6547 fs_devices = ERR_PTR(ret);
6551 if (!fs_devices->seeding) {
6552 __btrfs_close_devices(fs_devices);
6553 free_fs_devices(fs_devices);
6554 fs_devices = ERR_PTR(-EINVAL);
6558 fs_devices->seed = root->fs_info->fs_devices->seed;
6559 root->fs_info->fs_devices->seed = fs_devices;
6564 static int read_one_dev(struct btrfs_root *root,
6565 struct extent_buffer *leaf,
6566 struct btrfs_dev_item *dev_item)
6568 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6569 struct btrfs_device *device;
6572 u8 fs_uuid[BTRFS_UUID_SIZE];
6573 u8 dev_uuid[BTRFS_UUID_SIZE];
6575 devid = btrfs_device_id(leaf, dev_item);
6576 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6578 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6581 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
6582 fs_devices = open_seed_devices(root, fs_uuid);
6583 if (IS_ERR(fs_devices))
6584 return PTR_ERR(fs_devices);
6587 device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
6589 if (!btrfs_test_opt(root->fs_info, DEGRADED))
6592 device = add_missing_dev(root, fs_devices, devid, dev_uuid);
6595 btrfs_warn(root->fs_info, "devid %llu uuid %pU missing",
6598 if (!device->bdev && !btrfs_test_opt(root->fs_info, DEGRADED))
6601 if(!device->bdev && !device->missing) {
6603 * this happens when a device that was properly setup
6604 * in the device info lists suddenly goes bad.
6605 * device->bdev is NULL, and so we have to set
6606 * device->missing to one here
6608 device->fs_devices->missing_devices++;
6609 device->missing = 1;
6612 /* Move the device to its own fs_devices */
6613 if (device->fs_devices != fs_devices) {
6614 ASSERT(device->missing);
6616 list_move(&device->dev_list, &fs_devices->devices);
6617 device->fs_devices->num_devices--;
6618 fs_devices->num_devices++;
6620 device->fs_devices->missing_devices--;
6621 fs_devices->missing_devices++;
6623 device->fs_devices = fs_devices;
6627 if (device->fs_devices != root->fs_info->fs_devices) {
6628 BUG_ON(device->writeable);
6629 if (device->generation !=
6630 btrfs_device_generation(leaf, dev_item))
6634 fill_device_from_item(leaf, dev_item, device);
6635 device->in_fs_metadata = 1;
6636 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
6637 device->fs_devices->total_rw_bytes += device->total_bytes;
6638 spin_lock(&root->fs_info->free_chunk_lock);
6639 root->fs_info->free_chunk_space += device->total_bytes -
6641 spin_unlock(&root->fs_info->free_chunk_lock);
6647 int btrfs_read_sys_array(struct btrfs_root *root)
6649 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
6650 struct extent_buffer *sb;
6651 struct btrfs_disk_key *disk_key;
6652 struct btrfs_chunk *chunk;
6654 unsigned long sb_array_offset;
6661 struct btrfs_key key;
6663 ASSERT(BTRFS_SUPER_INFO_SIZE <= root->nodesize);
6665 * This will create extent buffer of nodesize, superblock size is
6666 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6667 * overallocate but we can keep it as-is, only the first page is used.
6669 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET);
6672 set_extent_buffer_uptodate(sb);
6673 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6675 * The sb extent buffer is artificial and just used to read the system array.
6676 * set_extent_buffer_uptodate() call does not properly mark all it's
6677 * pages up-to-date when the page is larger: extent does not cover the
6678 * whole page and consequently check_page_uptodate does not find all
6679 * the page's extents up-to-date (the hole beyond sb),
6680 * write_extent_buffer then triggers a WARN_ON.
6682 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6683 * but sb spans only this function. Add an explicit SetPageUptodate call
6684 * to silence the warning eg. on PowerPC 64.
6686 if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
6687 SetPageUptodate(sb->pages[0]);
6689 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6690 array_size = btrfs_super_sys_array_size(super_copy);
6692 array_ptr = super_copy->sys_chunk_array;
6693 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6696 while (cur_offset < array_size) {
6697 disk_key = (struct btrfs_disk_key *)array_ptr;
6698 len = sizeof(*disk_key);
6699 if (cur_offset + len > array_size)
6700 goto out_short_read;
6702 btrfs_disk_key_to_cpu(&key, disk_key);
6705 sb_array_offset += len;
6708 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6709 chunk = (struct btrfs_chunk *)sb_array_offset;
6711 * At least one btrfs_chunk with one stripe must be
6712 * present, exact stripe count check comes afterwards
6714 len = btrfs_chunk_item_size(1);
6715 if (cur_offset + len > array_size)
6716 goto out_short_read;
6718 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6721 "BTRFS: invalid number of stripes %u in sys_array at offset %u\n",
6722 num_stripes, cur_offset);
6727 type = btrfs_chunk_type(sb, chunk);
6728 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
6729 btrfs_err(root->fs_info,
6730 "invalid chunk type %llu in sys_array at offset %u",
6736 len = btrfs_chunk_item_size(num_stripes);
6737 if (cur_offset + len > array_size)
6738 goto out_short_read;
6740 ret = read_one_chunk(root, &key, sb, chunk);
6745 "BTRFS: unexpected item type %u in sys_array at offset %u\n",
6746 (u32)key.type, cur_offset);
6751 sb_array_offset += len;
6754 clear_extent_buffer_uptodate(sb);
6755 free_extent_buffer_stale(sb);
6759 printk(KERN_ERR "BTRFS: sys_array too short to read %u bytes at offset %u\n",
6761 clear_extent_buffer_uptodate(sb);
6762 free_extent_buffer_stale(sb);
6766 int btrfs_read_chunk_tree(struct btrfs_root *root)
6768 struct btrfs_path *path;
6769 struct extent_buffer *leaf;
6770 struct btrfs_key key;
6771 struct btrfs_key found_key;
6776 root = root->fs_info->chunk_root;
6778 path = btrfs_alloc_path();
6782 mutex_lock(&uuid_mutex);
6786 * Read all device items, and then all the chunk items. All
6787 * device items are found before any chunk item (their object id
6788 * is smaller than the lowest possible object id for a chunk
6789 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6791 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6794 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6798 leaf = path->nodes[0];
6799 slot = path->slots[0];
6800 if (slot >= btrfs_header_nritems(leaf)) {
6801 ret = btrfs_next_leaf(root, path);
6808 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6809 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6810 struct btrfs_dev_item *dev_item;
6811 dev_item = btrfs_item_ptr(leaf, slot,
6812 struct btrfs_dev_item);
6813 ret = read_one_dev(root, leaf, dev_item);
6817 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6818 struct btrfs_chunk *chunk;
6819 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6820 ret = read_one_chunk(root, &found_key, leaf, chunk);
6828 * After loading chunk tree, we've got all device information,
6829 * do another round of validation checks.
6831 if (total_dev != root->fs_info->fs_devices->total_devices) {
6832 btrfs_err(root->fs_info,
6833 "super_num_devices %llu mismatch with num_devices %llu found here",
6834 btrfs_super_num_devices(root->fs_info->super_copy),
6839 if (btrfs_super_total_bytes(root->fs_info->super_copy) <
6840 root->fs_info->fs_devices->total_rw_bytes) {
6841 btrfs_err(root->fs_info,
6842 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
6843 btrfs_super_total_bytes(root->fs_info->super_copy),
6844 root->fs_info->fs_devices->total_rw_bytes);
6850 unlock_chunks(root);
6851 mutex_unlock(&uuid_mutex);
6853 btrfs_free_path(path);
6857 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6859 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6860 struct btrfs_device *device;
6862 while (fs_devices) {
6863 mutex_lock(&fs_devices->device_list_mutex);
6864 list_for_each_entry(device, &fs_devices->devices, dev_list)
6865 device->dev_root = fs_info->dev_root;
6866 mutex_unlock(&fs_devices->device_list_mutex);
6868 fs_devices = fs_devices->seed;
6872 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6876 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6877 btrfs_dev_stat_reset(dev, i);
6880 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6882 struct btrfs_key key;
6883 struct btrfs_key found_key;
6884 struct btrfs_root *dev_root = fs_info->dev_root;
6885 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6886 struct extent_buffer *eb;
6889 struct btrfs_device *device;
6890 struct btrfs_path *path = NULL;
6893 path = btrfs_alloc_path();
6899 mutex_lock(&fs_devices->device_list_mutex);
6900 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6902 struct btrfs_dev_stats_item *ptr;
6904 key.objectid = BTRFS_DEV_STATS_OBJECTID;
6905 key.type = BTRFS_PERSISTENT_ITEM_KEY;
6906 key.offset = device->devid;
6907 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6909 __btrfs_reset_dev_stats(device);
6910 device->dev_stats_valid = 1;
6911 btrfs_release_path(path);
6914 slot = path->slots[0];
6915 eb = path->nodes[0];
6916 btrfs_item_key_to_cpu(eb, &found_key, slot);
6917 item_size = btrfs_item_size_nr(eb, slot);
6919 ptr = btrfs_item_ptr(eb, slot,
6920 struct btrfs_dev_stats_item);
6922 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6923 if (item_size >= (1 + i) * sizeof(__le64))
6924 btrfs_dev_stat_set(device, i,
6925 btrfs_dev_stats_value(eb, ptr, i));
6927 btrfs_dev_stat_reset(device, i);
6930 device->dev_stats_valid = 1;
6931 btrfs_dev_stat_print_on_load(device);
6932 btrfs_release_path(path);
6934 mutex_unlock(&fs_devices->device_list_mutex);
6937 btrfs_free_path(path);
6938 return ret < 0 ? ret : 0;
6941 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
6942 struct btrfs_root *dev_root,
6943 struct btrfs_device *device)
6945 struct btrfs_path *path;
6946 struct btrfs_key key;
6947 struct extent_buffer *eb;
6948 struct btrfs_dev_stats_item *ptr;
6952 key.objectid = BTRFS_DEV_STATS_OBJECTID;
6953 key.type = BTRFS_PERSISTENT_ITEM_KEY;
6954 key.offset = device->devid;
6956 path = btrfs_alloc_path();
6958 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
6960 btrfs_warn_in_rcu(dev_root->fs_info,
6961 "error %d while searching for dev_stats item for device %s",
6962 ret, rcu_str_deref(device->name));
6967 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
6968 /* need to delete old one and insert a new one */
6969 ret = btrfs_del_item(trans, dev_root, path);
6971 btrfs_warn_in_rcu(dev_root->fs_info,
6972 "delete too small dev_stats item for device %s failed %d",
6973 rcu_str_deref(device->name), ret);
6980 /* need to insert a new item */
6981 btrfs_release_path(path);
6982 ret = btrfs_insert_empty_item(trans, dev_root, path,
6983 &key, sizeof(*ptr));
6985 btrfs_warn_in_rcu(dev_root->fs_info,
6986 "insert dev_stats item for device %s failed %d",
6987 rcu_str_deref(device->name), ret);
6992 eb = path->nodes[0];
6993 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
6994 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6995 btrfs_set_dev_stats_value(eb, ptr, i,
6996 btrfs_dev_stat_read(device, i));
6997 btrfs_mark_buffer_dirty(eb);
7000 btrfs_free_path(path);
7005 * called from commit_transaction. Writes all changed device stats to disk.
7007 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
7008 struct btrfs_fs_info *fs_info)
7010 struct btrfs_root *dev_root = fs_info->dev_root;
7011 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7012 struct btrfs_device *device;
7016 mutex_lock(&fs_devices->device_list_mutex);
7017 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7018 if (!device->dev_stats_valid || !btrfs_dev_stats_dirty(device))
7021 stats_cnt = atomic_read(&device->dev_stats_ccnt);
7022 ret = update_dev_stat_item(trans, dev_root, device);
7024 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7026 mutex_unlock(&fs_devices->device_list_mutex);
7031 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7033 btrfs_dev_stat_inc(dev, index);
7034 btrfs_dev_stat_print_on_error(dev);
7037 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
7039 if (!dev->dev_stats_valid)
7041 btrfs_err_rl_in_rcu(dev->dev_root->fs_info,
7042 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7043 rcu_str_deref(dev->name),
7044 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7045 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7046 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7047 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7048 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7051 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7055 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7056 if (btrfs_dev_stat_read(dev, i) != 0)
7058 if (i == BTRFS_DEV_STAT_VALUES_MAX)
7059 return; /* all values == 0, suppress message */
7061 btrfs_info_in_rcu(dev->dev_root->fs_info,
7062 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7063 rcu_str_deref(dev->name),
7064 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7065 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7066 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7067 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7068 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7071 int btrfs_get_dev_stats(struct btrfs_root *root,
7072 struct btrfs_ioctl_get_dev_stats *stats)
7074 struct btrfs_device *dev;
7075 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7078 mutex_lock(&fs_devices->device_list_mutex);
7079 dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
7080 mutex_unlock(&fs_devices->device_list_mutex);
7083 btrfs_warn(root->fs_info, "get dev_stats failed, device not found");
7085 } else if (!dev->dev_stats_valid) {
7086 btrfs_warn(root->fs_info, "get dev_stats failed, not yet valid");
7088 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7089 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7090 if (stats->nr_items > i)
7092 btrfs_dev_stat_read_and_reset(dev, i);
7094 btrfs_dev_stat_reset(dev, i);
7097 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7098 if (stats->nr_items > i)
7099 stats->values[i] = btrfs_dev_stat_read(dev, i);
7101 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7102 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7106 void btrfs_scratch_superblocks(struct block_device *bdev, char *device_path)
7108 struct buffer_head *bh;
7109 struct btrfs_super_block *disk_super;
7115 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX;
7118 if (btrfs_read_dev_one_super(bdev, copy_num, &bh))
7121 disk_super = (struct btrfs_super_block *)bh->b_data;
7123 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
7124 set_buffer_dirty(bh);
7125 sync_dirty_buffer(bh);
7129 /* Notify udev that device has changed */
7130 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
7132 /* Update ctime/mtime for device path for libblkid */
7133 update_dev_time(device_path);
7137 * Update the size of all devices, which is used for writing out the
7140 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
7142 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7143 struct btrfs_device *curr, *next;
7145 if (list_empty(&fs_devices->resized_devices))
7148 mutex_lock(&fs_devices->device_list_mutex);
7149 lock_chunks(fs_info->dev_root);
7150 list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
7152 list_del_init(&curr->resized_list);
7153 curr->commit_total_bytes = curr->disk_total_bytes;
7155 unlock_chunks(fs_info->dev_root);
7156 mutex_unlock(&fs_devices->device_list_mutex);
7159 /* Must be invoked during the transaction commit */
7160 void btrfs_update_commit_device_bytes_used(struct btrfs_root *root,
7161 struct btrfs_transaction *transaction)
7163 struct extent_map *em;
7164 struct map_lookup *map;
7165 struct btrfs_device *dev;
7168 if (list_empty(&transaction->pending_chunks))
7171 /* In order to kick the device replace finish process */
7173 list_for_each_entry(em, &transaction->pending_chunks, list) {
7174 map = em->map_lookup;
7176 for (i = 0; i < map->num_stripes; i++) {
7177 dev = map->stripes[i].dev;
7178 dev->commit_bytes_used = dev->bytes_used;
7181 unlock_chunks(root);
7184 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
7186 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7187 while (fs_devices) {
7188 fs_devices->fs_info = fs_info;
7189 fs_devices = fs_devices->seed;
7193 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
7195 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7196 while (fs_devices) {
7197 fs_devices->fs_info = NULL;
7198 fs_devices = fs_devices->seed;