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/random.h>
24 #include <linux/iocontext.h>
25 #include <linux/capability.h>
26 #include <linux/ratelimit.h>
27 #include <linux/kthread.h>
28 #include <linux/raid/pq.h>
29 #include <linux/semaphore.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,
121 static int init_first_rw_device(struct btrfs_trans_handle *trans,
122 struct btrfs_root *root,
123 struct btrfs_device *device);
124 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
125 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
126 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
127 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
129 DEFINE_MUTEX(uuid_mutex);
130 static LIST_HEAD(fs_uuids);
131 struct list_head *btrfs_get_fs_uuids(void)
136 static struct btrfs_fs_devices *__alloc_fs_devices(void)
138 struct btrfs_fs_devices *fs_devs;
140 fs_devs = kzalloc(sizeof(*fs_devs), GFP_NOFS);
142 return ERR_PTR(-ENOMEM);
144 mutex_init(&fs_devs->device_list_mutex);
146 INIT_LIST_HEAD(&fs_devs->devices);
147 INIT_LIST_HEAD(&fs_devs->resized_devices);
148 INIT_LIST_HEAD(&fs_devs->alloc_list);
149 INIT_LIST_HEAD(&fs_devs->list);
155 * alloc_fs_devices - allocate struct btrfs_fs_devices
156 * @fsid: a pointer to UUID for this FS. If NULL a new UUID is
159 * Return: a pointer to a new &struct btrfs_fs_devices on success;
160 * ERR_PTR() on error. Returned struct is not linked onto any lists and
161 * can be destroyed with kfree() right away.
163 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
165 struct btrfs_fs_devices *fs_devs;
167 fs_devs = __alloc_fs_devices();
172 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
174 generate_random_uuid(fs_devs->fsid);
179 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
181 struct btrfs_device *device;
182 WARN_ON(fs_devices->opened);
183 while (!list_empty(&fs_devices->devices)) {
184 device = list_entry(fs_devices->devices.next,
185 struct btrfs_device, dev_list);
186 list_del(&device->dev_list);
187 rcu_string_free(device->name);
193 static void btrfs_kobject_uevent(struct block_device *bdev,
194 enum kobject_action action)
198 ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
200 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
202 kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
203 &disk_to_dev(bdev->bd_disk)->kobj);
206 void btrfs_cleanup_fs_uuids(void)
208 struct btrfs_fs_devices *fs_devices;
210 while (!list_empty(&fs_uuids)) {
211 fs_devices = list_entry(fs_uuids.next,
212 struct btrfs_fs_devices, list);
213 list_del(&fs_devices->list);
214 free_fs_devices(fs_devices);
218 static struct btrfs_device *__alloc_device(void)
220 struct btrfs_device *dev;
222 dev = kzalloc(sizeof(*dev), GFP_NOFS);
224 return ERR_PTR(-ENOMEM);
226 INIT_LIST_HEAD(&dev->dev_list);
227 INIT_LIST_HEAD(&dev->dev_alloc_list);
228 INIT_LIST_HEAD(&dev->resized_list);
230 spin_lock_init(&dev->io_lock);
232 spin_lock_init(&dev->reada_lock);
233 atomic_set(&dev->reada_in_flight, 0);
234 atomic_set(&dev->dev_stats_ccnt, 0);
235 btrfs_device_data_ordered_init(dev);
236 INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
237 INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
242 static noinline struct btrfs_device *__find_device(struct list_head *head,
245 struct btrfs_device *dev;
247 list_for_each_entry(dev, head, dev_list) {
248 if (dev->devid == devid &&
249 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
256 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
258 struct btrfs_fs_devices *fs_devices;
260 list_for_each_entry(fs_devices, &fs_uuids, list) {
261 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
268 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
269 int flush, struct block_device **bdev,
270 struct buffer_head **bh)
274 *bdev = blkdev_get_by_path(device_path, flags, holder);
277 ret = PTR_ERR(*bdev);
282 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
283 ret = set_blocksize(*bdev, 4096);
285 blkdev_put(*bdev, flags);
288 invalidate_bdev(*bdev);
289 *bh = btrfs_read_dev_super(*bdev);
292 blkdev_put(*bdev, flags);
304 static void requeue_list(struct btrfs_pending_bios *pending_bios,
305 struct bio *head, struct bio *tail)
308 struct bio *old_head;
310 old_head = pending_bios->head;
311 pending_bios->head = head;
312 if (pending_bios->tail)
313 tail->bi_next = old_head;
315 pending_bios->tail = tail;
319 * we try to collect pending bios for a device so we don't get a large
320 * number of procs sending bios down to the same device. This greatly
321 * improves the schedulers ability to collect and merge the bios.
323 * But, it also turns into a long list of bios to process and that is sure
324 * to eventually make the worker thread block. The solution here is to
325 * make some progress and then put this work struct back at the end of
326 * the list if the block device is congested. This way, multiple devices
327 * can make progress from a single worker thread.
329 static noinline void run_scheduled_bios(struct btrfs_device *device)
332 struct backing_dev_info *bdi;
333 struct btrfs_fs_info *fs_info;
334 struct btrfs_pending_bios *pending_bios;
338 unsigned long num_run;
339 unsigned long batch_run = 0;
341 unsigned long last_waited = 0;
343 int sync_pending = 0;
344 struct blk_plug plug;
347 * this function runs all the bios we've collected for
348 * a particular device. We don't want to wander off to
349 * another device without first sending all of these down.
350 * So, setup a plug here and finish it off before we return
352 blk_start_plug(&plug);
354 bdi = blk_get_backing_dev_info(device->bdev);
355 fs_info = device->dev_root->fs_info;
356 limit = btrfs_async_submit_limit(fs_info);
357 limit = limit * 2 / 3;
360 spin_lock(&device->io_lock);
365 /* take all the bios off the list at once and process them
366 * later on (without the lock held). But, remember the
367 * tail and other pointers so the bios can be properly reinserted
368 * into the list if we hit congestion
370 if (!force_reg && device->pending_sync_bios.head) {
371 pending_bios = &device->pending_sync_bios;
374 pending_bios = &device->pending_bios;
378 pending = pending_bios->head;
379 tail = pending_bios->tail;
380 WARN_ON(pending && !tail);
383 * if pending was null this time around, no bios need processing
384 * at all and we can stop. Otherwise it'll loop back up again
385 * and do an additional check so no bios are missed.
387 * device->running_pending is used to synchronize with the
390 if (device->pending_sync_bios.head == NULL &&
391 device->pending_bios.head == NULL) {
393 device->running_pending = 0;
396 device->running_pending = 1;
399 pending_bios->head = NULL;
400 pending_bios->tail = NULL;
402 spin_unlock(&device->io_lock);
407 /* we want to work on both lists, but do more bios on the
408 * sync list than the regular list
411 pending_bios != &device->pending_sync_bios &&
412 device->pending_sync_bios.head) ||
413 (num_run > 64 && pending_bios == &device->pending_sync_bios &&
414 device->pending_bios.head)) {
415 spin_lock(&device->io_lock);
416 requeue_list(pending_bios, pending, tail);
421 pending = pending->bi_next;
425 * atomic_dec_return implies a barrier for waitqueue_active
427 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
428 waitqueue_active(&fs_info->async_submit_wait))
429 wake_up(&fs_info->async_submit_wait);
431 BUG_ON(atomic_read(&cur->__bi_cnt) == 0);
434 * if we're doing the sync list, record that our
435 * plug has some sync requests on it
437 * If we're doing the regular list and there are
438 * sync requests sitting around, unplug before
441 if (pending_bios == &device->pending_sync_bios) {
443 } else if (sync_pending) {
444 blk_finish_plug(&plug);
445 blk_start_plug(&plug);
449 btrfsic_submit_bio(cur->bi_rw, cur);
456 * we made progress, there is more work to do and the bdi
457 * is now congested. Back off and let other work structs
460 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
461 fs_info->fs_devices->open_devices > 1) {
462 struct io_context *ioc;
464 ioc = current->io_context;
467 * the main goal here is that we don't want to
468 * block if we're going to be able to submit
469 * more requests without blocking.
471 * This code does two great things, it pokes into
472 * the elevator code from a filesystem _and_
473 * it makes assumptions about how batching works.
475 if (ioc && ioc->nr_batch_requests > 0 &&
476 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
478 ioc->last_waited == last_waited)) {
480 * we want to go through our batch of
481 * requests and stop. So, we copy out
482 * the ioc->last_waited time and test
483 * against it before looping
485 last_waited = ioc->last_waited;
489 spin_lock(&device->io_lock);
490 requeue_list(pending_bios, pending, tail);
491 device->running_pending = 1;
493 spin_unlock(&device->io_lock);
494 btrfs_queue_work(fs_info->submit_workers,
498 /* unplug every 64 requests just for good measure */
499 if (batch_run % 64 == 0) {
500 blk_finish_plug(&plug);
501 blk_start_plug(&plug);
510 spin_lock(&device->io_lock);
511 if (device->pending_bios.head || device->pending_sync_bios.head)
513 spin_unlock(&device->io_lock);
516 blk_finish_plug(&plug);
519 static void pending_bios_fn(struct btrfs_work *work)
521 struct btrfs_device *device;
523 device = container_of(work, struct btrfs_device, work);
524 run_scheduled_bios(device);
528 void btrfs_free_stale_device(struct btrfs_device *cur_dev)
530 struct btrfs_fs_devices *fs_devs;
531 struct btrfs_device *dev;
536 list_for_each_entry(fs_devs, &fs_uuids, list) {
541 if (fs_devs->seeding)
544 list_for_each_entry(dev, &fs_devs->devices, dev_list) {
552 * Todo: This won't be enough. What if the same device
553 * comes back (with new uuid and) with its mapper path?
554 * But for now, this does help as mostly an admin will
555 * either use mapper or non mapper path throughout.
558 del = strcmp(rcu_str_deref(dev->name),
559 rcu_str_deref(cur_dev->name));
566 /* delete the stale device */
567 if (fs_devs->num_devices == 1) {
568 btrfs_sysfs_remove_fsid(fs_devs);
569 list_del(&fs_devs->list);
570 free_fs_devices(fs_devs);
573 fs_devs->num_devices--;
574 list_del(&dev->dev_list);
575 rcu_string_free(dev->name);
584 * Add new device to list of registered devices
587 * 1 - first time device is seen
588 * 0 - device already known
591 static noinline int device_list_add(const char *path,
592 struct btrfs_super_block *disk_super,
593 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
595 struct btrfs_device *device;
596 struct btrfs_fs_devices *fs_devices;
597 struct rcu_string *name;
599 u64 found_transid = btrfs_super_generation(disk_super);
601 fs_devices = find_fsid(disk_super->fsid);
603 fs_devices = alloc_fs_devices(disk_super->fsid);
604 if (IS_ERR(fs_devices))
605 return PTR_ERR(fs_devices);
607 list_add(&fs_devices->list, &fs_uuids);
611 device = __find_device(&fs_devices->devices, devid,
612 disk_super->dev_item.uuid);
616 if (fs_devices->opened)
619 device = btrfs_alloc_device(NULL, &devid,
620 disk_super->dev_item.uuid);
621 if (IS_ERR(device)) {
622 /* we can safely leave the fs_devices entry around */
623 return PTR_ERR(device);
626 name = rcu_string_strdup(path, GFP_NOFS);
631 rcu_assign_pointer(device->name, name);
633 mutex_lock(&fs_devices->device_list_mutex);
634 list_add_rcu(&device->dev_list, &fs_devices->devices);
635 fs_devices->num_devices++;
636 mutex_unlock(&fs_devices->device_list_mutex);
639 device->fs_devices = fs_devices;
640 } else if (!device->name || strcmp(device->name->str, path)) {
642 * When FS is already mounted.
643 * 1. If you are here and if the device->name is NULL that
644 * means this device was missing at time of FS mount.
645 * 2. If you are here and if the device->name is different
646 * from 'path' that means either
647 * a. The same device disappeared and reappeared with
649 * b. The missing-disk-which-was-replaced, has
652 * We must allow 1 and 2a above. But 2b would be a spurious
655 * Further in case of 1 and 2a above, the disk at 'path'
656 * would have missed some transaction when it was away and
657 * in case of 2a the stale bdev has to be updated as well.
658 * 2b must not be allowed at all time.
662 * For now, we do allow update to btrfs_fs_device through the
663 * btrfs dev scan cli after FS has been mounted. We're still
664 * tracking a problem where systems fail mount by subvolume id
665 * when we reject replacement on a mounted FS.
667 if (!fs_devices->opened && found_transid < device->generation) {
669 * That is if the FS is _not_ mounted and if you
670 * are here, that means there is more than one
671 * disk with same uuid and devid.We keep the one
672 * with larger generation number or the last-in if
673 * generation are equal.
678 name = rcu_string_strdup(path, GFP_NOFS);
681 rcu_string_free(device->name);
682 rcu_assign_pointer(device->name, name);
683 if (device->missing) {
684 fs_devices->missing_devices--;
690 * Unmount does not free the btrfs_device struct but would zero
691 * generation along with most of the other members. So just update
692 * it back. We need it to pick the disk with largest generation
695 if (!fs_devices->opened)
696 device->generation = found_transid;
699 * if there is new btrfs on an already registered device,
700 * then remove the stale device entry.
702 btrfs_free_stale_device(device);
704 *fs_devices_ret = fs_devices;
709 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
711 struct btrfs_fs_devices *fs_devices;
712 struct btrfs_device *device;
713 struct btrfs_device *orig_dev;
715 fs_devices = alloc_fs_devices(orig->fsid);
716 if (IS_ERR(fs_devices))
719 mutex_lock(&orig->device_list_mutex);
720 fs_devices->total_devices = orig->total_devices;
722 /* We have held the volume lock, it is safe to get the devices. */
723 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
724 struct rcu_string *name;
726 device = btrfs_alloc_device(NULL, &orig_dev->devid,
732 * This is ok to do without rcu read locked because we hold the
733 * uuid mutex so nothing we touch in here is going to disappear.
735 if (orig_dev->name) {
736 name = rcu_string_strdup(orig_dev->name->str, GFP_NOFS);
741 rcu_assign_pointer(device->name, name);
744 list_add(&device->dev_list, &fs_devices->devices);
745 device->fs_devices = fs_devices;
746 fs_devices->num_devices++;
748 mutex_unlock(&orig->device_list_mutex);
751 mutex_unlock(&orig->device_list_mutex);
752 free_fs_devices(fs_devices);
753 return ERR_PTR(-ENOMEM);
756 void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step)
758 struct btrfs_device *device, *next;
759 struct btrfs_device *latest_dev = NULL;
761 mutex_lock(&uuid_mutex);
763 /* This is the initialized path, it is safe to release the devices. */
764 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
765 if (device->in_fs_metadata) {
766 if (!device->is_tgtdev_for_dev_replace &&
768 device->generation > latest_dev->generation)) {
774 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
776 * In the first step, keep the device which has
777 * the correct fsid and the devid that is used
778 * for the dev_replace procedure.
779 * In the second step, the dev_replace state is
780 * read from the device tree and it is known
781 * whether the procedure is really active or
782 * not, which means whether this device is
783 * used or whether it should be removed.
785 if (step == 0 || device->is_tgtdev_for_dev_replace) {
790 blkdev_put(device->bdev, device->mode);
792 fs_devices->open_devices--;
794 if (device->writeable) {
795 list_del_init(&device->dev_alloc_list);
796 device->writeable = 0;
797 if (!device->is_tgtdev_for_dev_replace)
798 fs_devices->rw_devices--;
800 list_del_init(&device->dev_list);
801 fs_devices->num_devices--;
802 rcu_string_free(device->name);
806 if (fs_devices->seed) {
807 fs_devices = fs_devices->seed;
811 fs_devices->latest_bdev = latest_dev->bdev;
813 mutex_unlock(&uuid_mutex);
816 static void __free_device(struct work_struct *work)
818 struct btrfs_device *device;
820 device = container_of(work, struct btrfs_device, rcu_work);
823 blkdev_put(device->bdev, device->mode);
825 rcu_string_free(device->name);
829 static void free_device(struct rcu_head *head)
831 struct btrfs_device *device;
833 device = container_of(head, struct btrfs_device, rcu);
835 INIT_WORK(&device->rcu_work, __free_device);
836 schedule_work(&device->rcu_work);
839 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
841 struct btrfs_device *device, *tmp;
843 if (--fs_devices->opened > 0)
846 mutex_lock(&fs_devices->device_list_mutex);
847 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
848 btrfs_close_one_device(device);
850 mutex_unlock(&fs_devices->device_list_mutex);
852 WARN_ON(fs_devices->open_devices);
853 WARN_ON(fs_devices->rw_devices);
854 fs_devices->opened = 0;
855 fs_devices->seeding = 0;
860 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
862 struct btrfs_fs_devices *seed_devices = NULL;
865 mutex_lock(&uuid_mutex);
866 ret = __btrfs_close_devices(fs_devices);
867 if (!fs_devices->opened) {
868 seed_devices = fs_devices->seed;
869 fs_devices->seed = NULL;
871 mutex_unlock(&uuid_mutex);
873 while (seed_devices) {
874 fs_devices = seed_devices;
875 seed_devices = fs_devices->seed;
876 __btrfs_close_devices(fs_devices);
877 free_fs_devices(fs_devices);
880 * Wait for rcu kworkers under __btrfs_close_devices
881 * to finish all blkdev_puts so device is really
882 * free when umount is done.
888 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
889 fmode_t flags, void *holder)
891 struct request_queue *q;
892 struct block_device *bdev;
893 struct list_head *head = &fs_devices->devices;
894 struct btrfs_device *device;
895 struct btrfs_device *latest_dev = NULL;
896 struct buffer_head *bh;
897 struct btrfs_super_block *disk_super;
904 list_for_each_entry(device, head, dev_list) {
910 /* Just open everything we can; ignore failures here */
911 if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
915 disk_super = (struct btrfs_super_block *)bh->b_data;
916 devid = btrfs_stack_device_id(&disk_super->dev_item);
917 if (devid != device->devid)
920 if (memcmp(device->uuid, disk_super->dev_item.uuid,
924 device->generation = btrfs_super_generation(disk_super);
926 device->generation > latest_dev->generation)
929 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
930 device->writeable = 0;
932 device->writeable = !bdev_read_only(bdev);
936 q = bdev_get_queue(bdev);
937 if (blk_queue_discard(q))
938 device->can_discard = 1;
941 device->in_fs_metadata = 0;
942 device->mode = flags;
944 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
945 fs_devices->rotating = 1;
947 fs_devices->open_devices++;
948 if (device->writeable &&
949 device->devid != BTRFS_DEV_REPLACE_DEVID) {
950 fs_devices->rw_devices++;
951 list_add(&device->dev_alloc_list,
952 &fs_devices->alloc_list);
959 blkdev_put(bdev, flags);
962 if (fs_devices->open_devices == 0) {
966 fs_devices->seeding = seeding;
967 fs_devices->opened = 1;
968 fs_devices->latest_bdev = latest_dev->bdev;
969 fs_devices->total_rw_bytes = 0;
974 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
975 fmode_t flags, void *holder)
979 mutex_lock(&uuid_mutex);
980 if (fs_devices->opened) {
981 fs_devices->opened++;
984 ret = __btrfs_open_devices(fs_devices, flags, holder);
986 mutex_unlock(&uuid_mutex);
991 * Look for a btrfs signature on a device. This may be called out of the mount path
992 * and we are not allowed to call set_blocksize during the scan. The superblock
993 * is read via pagecache
995 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
996 struct btrfs_fs_devices **fs_devices_ret)
998 struct btrfs_super_block *disk_super;
999 struct block_device *bdev;
1010 * we would like to check all the supers, but that would make
1011 * a btrfs mount succeed after a mkfs from a different FS.
1012 * So, we need to add a special mount option to scan for
1013 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1015 bytenr = btrfs_sb_offset(0);
1016 flags |= FMODE_EXCL;
1017 mutex_lock(&uuid_mutex);
1019 bdev = blkdev_get_by_path(path, flags, holder);
1022 ret = PTR_ERR(bdev);
1026 /* make sure our super fits in the device */
1027 if (bytenr + PAGE_CACHE_SIZE >= i_size_read(bdev->bd_inode))
1028 goto error_bdev_put;
1030 /* make sure our super fits in the page */
1031 if (sizeof(*disk_super) > PAGE_CACHE_SIZE)
1032 goto error_bdev_put;
1034 /* make sure our super doesn't straddle pages on disk */
1035 index = bytenr >> PAGE_CACHE_SHIFT;
1036 if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_CACHE_SHIFT != index)
1037 goto error_bdev_put;
1039 /* pull in the page with our super */
1040 page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
1043 if (IS_ERR_OR_NULL(page))
1044 goto error_bdev_put;
1048 /* align our pointer to the offset of the super block */
1049 disk_super = p + (bytenr & ~PAGE_CACHE_MASK);
1051 if (btrfs_super_bytenr(disk_super) != bytenr ||
1052 btrfs_super_magic(disk_super) != BTRFS_MAGIC)
1055 devid = btrfs_stack_device_id(&disk_super->dev_item);
1056 transid = btrfs_super_generation(disk_super);
1057 total_devices = btrfs_super_num_devices(disk_super);
1059 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
1061 if (disk_super->label[0]) {
1062 if (disk_super->label[BTRFS_LABEL_SIZE - 1])
1063 disk_super->label[BTRFS_LABEL_SIZE - 1] = '\0';
1064 printk(KERN_INFO "BTRFS: device label %s ", disk_super->label);
1066 printk(KERN_INFO "BTRFS: device fsid %pU ", disk_super->fsid);
1069 printk(KERN_CONT "devid %llu transid %llu %s\n", devid, transid, path);
1072 if (!ret && fs_devices_ret)
1073 (*fs_devices_ret)->total_devices = total_devices;
1077 page_cache_release(page);
1080 blkdev_put(bdev, flags);
1082 mutex_unlock(&uuid_mutex);
1086 /* helper to account the used device space in the range */
1087 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
1088 u64 end, u64 *length)
1090 struct btrfs_key key;
1091 struct btrfs_root *root = device->dev_root;
1092 struct btrfs_dev_extent *dev_extent;
1093 struct btrfs_path *path;
1097 struct extent_buffer *l;
1101 if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
1104 path = btrfs_alloc_path();
1109 key.objectid = device->devid;
1111 key.type = BTRFS_DEV_EXTENT_KEY;
1113 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1117 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1124 slot = path->slots[0];
1125 if (slot >= btrfs_header_nritems(l)) {
1126 ret = btrfs_next_leaf(root, path);
1134 btrfs_item_key_to_cpu(l, &key, slot);
1136 if (key.objectid < device->devid)
1139 if (key.objectid > device->devid)
1142 if (key.type != BTRFS_DEV_EXTENT_KEY)
1145 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1146 extent_end = key.offset + btrfs_dev_extent_length(l,
1148 if (key.offset <= start && extent_end > end) {
1149 *length = end - start + 1;
1151 } else if (key.offset <= start && extent_end > start)
1152 *length += extent_end - start;
1153 else if (key.offset > start && extent_end <= end)
1154 *length += extent_end - key.offset;
1155 else if (key.offset > start && key.offset <= end) {
1156 *length += end - key.offset + 1;
1158 } else if (key.offset > end)
1166 btrfs_free_path(path);
1170 static int contains_pending_extent(struct btrfs_transaction *transaction,
1171 struct btrfs_device *device,
1172 u64 *start, u64 len)
1174 struct btrfs_fs_info *fs_info = device->dev_root->fs_info;
1175 struct extent_map *em;
1176 struct list_head *search_list = &fs_info->pinned_chunks;
1178 u64 physical_start = *start;
1181 search_list = &transaction->pending_chunks;
1183 list_for_each_entry(em, search_list, list) {
1184 struct map_lookup *map;
1187 map = em->map_lookup;
1188 for (i = 0; i < map->num_stripes; i++) {
1191 if (map->stripes[i].dev != device)
1193 if (map->stripes[i].physical >= physical_start + len ||
1194 map->stripes[i].physical + em->orig_block_len <=
1198 * Make sure that while processing the pinned list we do
1199 * not override our *start with a lower value, because
1200 * we can have pinned chunks that fall within this
1201 * device hole and that have lower physical addresses
1202 * than the pending chunks we processed before. If we
1203 * do not take this special care we can end up getting
1204 * 2 pending chunks that start at the same physical
1205 * device offsets because the end offset of a pinned
1206 * chunk can be equal to the start offset of some
1209 end = map->stripes[i].physical + em->orig_block_len;
1216 if (search_list != &fs_info->pinned_chunks) {
1217 search_list = &fs_info->pinned_chunks;
1226 * find_free_dev_extent_start - find free space in the specified device
1227 * @device: the device which we search the free space in
1228 * @num_bytes: the size of the free space that we need
1229 * @search_start: the position from which to begin the search
1230 * @start: store the start of the free space.
1231 * @len: the size of the free space. that we find, or the size
1232 * of the max free space if we don't find suitable free space
1234 * this uses a pretty simple search, the expectation is that it is
1235 * called very infrequently and that a given device has a small number
1238 * @start is used to store the start of the free space if we find. But if we
1239 * don't find suitable free space, it will be used to store the start position
1240 * of the max free space.
1242 * @len is used to store the size of the free space that we find.
1243 * But if we don't find suitable free space, it is used to store the size of
1244 * the max free space.
1246 int find_free_dev_extent_start(struct btrfs_transaction *transaction,
1247 struct btrfs_device *device, u64 num_bytes,
1248 u64 search_start, u64 *start, u64 *len)
1250 struct btrfs_key key;
1251 struct btrfs_root *root = device->dev_root;
1252 struct btrfs_dev_extent *dev_extent;
1253 struct btrfs_path *path;
1258 u64 search_end = device->total_bytes;
1261 struct extent_buffer *l;
1262 u64 min_search_start;
1265 * We don't want to overwrite the superblock on the drive nor any area
1266 * used by the boot loader (grub for example), so we make sure to start
1267 * at an offset of at least 1MB.
1269 min_search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
1270 search_start = max(search_start, min_search_start);
1272 path = btrfs_alloc_path();
1276 max_hole_start = search_start;
1280 if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1286 path->search_commit_root = 1;
1287 path->skip_locking = 1;
1289 key.objectid = device->devid;
1290 key.offset = search_start;
1291 key.type = BTRFS_DEV_EXTENT_KEY;
1293 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1297 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1304 slot = path->slots[0];
1305 if (slot >= btrfs_header_nritems(l)) {
1306 ret = btrfs_next_leaf(root, path);
1314 btrfs_item_key_to_cpu(l, &key, slot);
1316 if (key.objectid < device->devid)
1319 if (key.objectid > device->devid)
1322 if (key.type != BTRFS_DEV_EXTENT_KEY)
1325 if (key.offset > search_start) {
1326 hole_size = key.offset - search_start;
1329 * Have to check before we set max_hole_start, otherwise
1330 * we could end up sending back this offset anyway.
1332 if (contains_pending_extent(transaction, device,
1335 if (key.offset >= search_start) {
1336 hole_size = key.offset - search_start;
1343 if (hole_size > max_hole_size) {
1344 max_hole_start = search_start;
1345 max_hole_size = hole_size;
1349 * If this free space is greater than which we need,
1350 * it must be the max free space that we have found
1351 * until now, so max_hole_start must point to the start
1352 * of this free space and the length of this free space
1353 * is stored in max_hole_size. Thus, we return
1354 * max_hole_start and max_hole_size and go back to the
1357 if (hole_size >= num_bytes) {
1363 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1364 extent_end = key.offset + btrfs_dev_extent_length(l,
1366 if (extent_end > search_start)
1367 search_start = extent_end;
1374 * At this point, search_start should be the end of
1375 * allocated dev extents, and when shrinking the device,
1376 * search_end may be smaller than search_start.
1378 if (search_end > search_start) {
1379 hole_size = search_end - search_start;
1381 if (contains_pending_extent(transaction, device, &search_start,
1383 btrfs_release_path(path);
1387 if (hole_size > max_hole_size) {
1388 max_hole_start = search_start;
1389 max_hole_size = hole_size;
1394 if (max_hole_size < num_bytes)
1400 btrfs_free_path(path);
1401 *start = max_hole_start;
1403 *len = max_hole_size;
1407 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1408 struct btrfs_device *device, u64 num_bytes,
1409 u64 *start, u64 *len)
1411 /* FIXME use last free of some kind */
1412 return find_free_dev_extent_start(trans->transaction, device,
1413 num_bytes, 0, start, len);
1416 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1417 struct btrfs_device *device,
1418 u64 start, u64 *dev_extent_len)
1421 struct btrfs_path *path;
1422 struct btrfs_root *root = device->dev_root;
1423 struct btrfs_key key;
1424 struct btrfs_key found_key;
1425 struct extent_buffer *leaf = NULL;
1426 struct btrfs_dev_extent *extent = NULL;
1428 path = btrfs_alloc_path();
1432 key.objectid = device->devid;
1434 key.type = BTRFS_DEV_EXTENT_KEY;
1436 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1438 ret = btrfs_previous_item(root, path, key.objectid,
1439 BTRFS_DEV_EXTENT_KEY);
1442 leaf = path->nodes[0];
1443 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1444 extent = btrfs_item_ptr(leaf, path->slots[0],
1445 struct btrfs_dev_extent);
1446 BUG_ON(found_key.offset > start || found_key.offset +
1447 btrfs_dev_extent_length(leaf, extent) < start);
1449 btrfs_release_path(path);
1451 } else if (ret == 0) {
1452 leaf = path->nodes[0];
1453 extent = btrfs_item_ptr(leaf, path->slots[0],
1454 struct btrfs_dev_extent);
1456 btrfs_std_error(root->fs_info, ret, "Slot search failed");
1460 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1462 ret = btrfs_del_item(trans, root, path);
1464 btrfs_std_error(root->fs_info, ret,
1465 "Failed to remove dev extent item");
1467 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1470 btrfs_free_path(path);
1474 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1475 struct btrfs_device *device,
1476 u64 chunk_tree, u64 chunk_objectid,
1477 u64 chunk_offset, u64 start, u64 num_bytes)
1480 struct btrfs_path *path;
1481 struct btrfs_root *root = device->dev_root;
1482 struct btrfs_dev_extent *extent;
1483 struct extent_buffer *leaf;
1484 struct btrfs_key key;
1486 WARN_ON(!device->in_fs_metadata);
1487 WARN_ON(device->is_tgtdev_for_dev_replace);
1488 path = btrfs_alloc_path();
1492 key.objectid = device->devid;
1494 key.type = BTRFS_DEV_EXTENT_KEY;
1495 ret = btrfs_insert_empty_item(trans, root, path, &key,
1500 leaf = path->nodes[0];
1501 extent = btrfs_item_ptr(leaf, path->slots[0],
1502 struct btrfs_dev_extent);
1503 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1504 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1505 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1507 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1508 btrfs_dev_extent_chunk_tree_uuid(extent), BTRFS_UUID_SIZE);
1510 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1511 btrfs_mark_buffer_dirty(leaf);
1513 btrfs_free_path(path);
1517 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1519 struct extent_map_tree *em_tree;
1520 struct extent_map *em;
1524 em_tree = &fs_info->mapping_tree.map_tree;
1525 read_lock(&em_tree->lock);
1526 n = rb_last(&em_tree->map);
1528 em = rb_entry(n, struct extent_map, rb_node);
1529 ret = em->start + em->len;
1531 read_unlock(&em_tree->lock);
1536 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1540 struct btrfs_key key;
1541 struct btrfs_key found_key;
1542 struct btrfs_path *path;
1544 path = btrfs_alloc_path();
1548 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1549 key.type = BTRFS_DEV_ITEM_KEY;
1550 key.offset = (u64)-1;
1552 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1556 BUG_ON(ret == 0); /* Corruption */
1558 ret = btrfs_previous_item(fs_info->chunk_root, path,
1559 BTRFS_DEV_ITEMS_OBJECTID,
1560 BTRFS_DEV_ITEM_KEY);
1564 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1566 *devid_ret = found_key.offset + 1;
1570 btrfs_free_path(path);
1575 * the device information is stored in the chunk root
1576 * the btrfs_device struct should be fully filled in
1578 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1579 struct btrfs_root *root,
1580 struct btrfs_device *device)
1583 struct btrfs_path *path;
1584 struct btrfs_dev_item *dev_item;
1585 struct extent_buffer *leaf;
1586 struct btrfs_key key;
1589 root = root->fs_info->chunk_root;
1591 path = btrfs_alloc_path();
1595 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1596 key.type = BTRFS_DEV_ITEM_KEY;
1597 key.offset = device->devid;
1599 ret = btrfs_insert_empty_item(trans, root, path, &key,
1604 leaf = path->nodes[0];
1605 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1607 btrfs_set_device_id(leaf, dev_item, device->devid);
1608 btrfs_set_device_generation(leaf, dev_item, 0);
1609 btrfs_set_device_type(leaf, dev_item, device->type);
1610 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1611 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1612 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1613 btrfs_set_device_total_bytes(leaf, dev_item,
1614 btrfs_device_get_disk_total_bytes(device));
1615 btrfs_set_device_bytes_used(leaf, dev_item,
1616 btrfs_device_get_bytes_used(device));
1617 btrfs_set_device_group(leaf, dev_item, 0);
1618 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1619 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1620 btrfs_set_device_start_offset(leaf, dev_item, 0);
1622 ptr = btrfs_device_uuid(dev_item);
1623 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1624 ptr = btrfs_device_fsid(dev_item);
1625 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1626 btrfs_mark_buffer_dirty(leaf);
1630 btrfs_free_path(path);
1635 * Function to update ctime/mtime for a given device path.
1636 * Mainly used for ctime/mtime based probe like libblkid.
1638 static void update_dev_time(char *path_name)
1642 filp = filp_open(path_name, O_RDWR, 0);
1645 file_update_time(filp);
1646 filp_close(filp, NULL);
1650 static int btrfs_rm_dev_item(struct btrfs_root *root,
1651 struct btrfs_device *device)
1654 struct btrfs_path *path;
1655 struct btrfs_key key;
1656 struct btrfs_trans_handle *trans;
1658 root = root->fs_info->chunk_root;
1660 path = btrfs_alloc_path();
1664 trans = btrfs_start_transaction(root, 0);
1665 if (IS_ERR(trans)) {
1666 btrfs_free_path(path);
1667 return PTR_ERR(trans);
1669 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1670 key.type = BTRFS_DEV_ITEM_KEY;
1671 key.offset = device->devid;
1673 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1682 ret = btrfs_del_item(trans, root, path);
1686 btrfs_free_path(path);
1687 btrfs_commit_transaction(trans, root);
1691 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1693 struct btrfs_device *device;
1694 struct btrfs_device *next_device;
1695 struct block_device *bdev;
1696 struct buffer_head *bh = NULL;
1697 struct btrfs_super_block *disk_super;
1698 struct btrfs_fs_devices *cur_devices;
1705 bool clear_super = false;
1707 mutex_lock(&uuid_mutex);
1710 seq = read_seqbegin(&root->fs_info->profiles_lock);
1712 all_avail = root->fs_info->avail_data_alloc_bits |
1713 root->fs_info->avail_system_alloc_bits |
1714 root->fs_info->avail_metadata_alloc_bits;
1715 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
1717 num_devices = root->fs_info->fs_devices->num_devices;
1718 btrfs_dev_replace_lock(&root->fs_info->dev_replace);
1719 if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1720 WARN_ON(num_devices < 1);
1723 btrfs_dev_replace_unlock(&root->fs_info->dev_replace);
1725 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && num_devices <= 4) {
1726 ret = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET;
1730 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && num_devices <= 2) {
1731 ret = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET;
1735 if ((all_avail & BTRFS_BLOCK_GROUP_RAID5) &&
1736 root->fs_info->fs_devices->rw_devices <= 2) {
1737 ret = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET;
1740 if ((all_avail & BTRFS_BLOCK_GROUP_RAID6) &&
1741 root->fs_info->fs_devices->rw_devices <= 3) {
1742 ret = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET;
1746 if (strcmp(device_path, "missing") == 0) {
1747 struct list_head *devices;
1748 struct btrfs_device *tmp;
1751 devices = &root->fs_info->fs_devices->devices;
1753 * It is safe to read the devices since the volume_mutex
1756 list_for_each_entry(tmp, devices, dev_list) {
1757 if (tmp->in_fs_metadata &&
1758 !tmp->is_tgtdev_for_dev_replace &&
1768 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
1772 ret = btrfs_get_bdev_and_sb(device_path,
1773 FMODE_WRITE | FMODE_EXCL,
1774 root->fs_info->bdev_holder, 0,
1778 disk_super = (struct btrfs_super_block *)bh->b_data;
1779 devid = btrfs_stack_device_id(&disk_super->dev_item);
1780 dev_uuid = disk_super->dev_item.uuid;
1781 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1789 if (device->is_tgtdev_for_dev_replace) {
1790 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1794 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1795 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1799 if (device->writeable) {
1801 list_del_init(&device->dev_alloc_list);
1802 device->fs_devices->rw_devices--;
1803 unlock_chunks(root);
1807 mutex_unlock(&uuid_mutex);
1808 ret = btrfs_shrink_device(device, 0);
1809 mutex_lock(&uuid_mutex);
1814 * TODO: the superblock still includes this device in its num_devices
1815 * counter although write_all_supers() is not locked out. This
1816 * could give a filesystem state which requires a degraded mount.
1818 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1822 device->in_fs_metadata = 0;
1823 btrfs_scrub_cancel_dev(root->fs_info, device);
1826 * the device list mutex makes sure that we don't change
1827 * the device list while someone else is writing out all
1828 * the device supers. Whoever is writing all supers, should
1829 * lock the device list mutex before getting the number of
1830 * devices in the super block (super_copy). Conversely,
1831 * whoever updates the number of devices in the super block
1832 * (super_copy) should hold the device list mutex.
1835 cur_devices = device->fs_devices;
1836 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1837 list_del_rcu(&device->dev_list);
1839 device->fs_devices->num_devices--;
1840 device->fs_devices->total_devices--;
1842 if (device->missing)
1843 device->fs_devices->missing_devices--;
1845 next_device = list_entry(root->fs_info->fs_devices->devices.next,
1846 struct btrfs_device, dev_list);
1847 if (device->bdev == root->fs_info->sb->s_bdev)
1848 root->fs_info->sb->s_bdev = next_device->bdev;
1849 if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1850 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1853 device->fs_devices->open_devices--;
1854 /* remove sysfs entry */
1855 btrfs_sysfs_rm_device_link(root->fs_info->fs_devices, device);
1858 call_rcu(&device->rcu, free_device);
1860 num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1861 btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1862 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1864 if (cur_devices->open_devices == 0) {
1865 struct btrfs_fs_devices *fs_devices;
1866 fs_devices = root->fs_info->fs_devices;
1867 while (fs_devices) {
1868 if (fs_devices->seed == cur_devices) {
1869 fs_devices->seed = cur_devices->seed;
1872 fs_devices = fs_devices->seed;
1874 cur_devices->seed = NULL;
1875 __btrfs_close_devices(cur_devices);
1876 free_fs_devices(cur_devices);
1879 root->fs_info->num_tolerated_disk_barrier_failures =
1880 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1883 * at this point, the device is zero sized. We want to
1884 * remove it from the devices list and zero out the old super
1886 if (clear_super && disk_super) {
1890 /* make sure this device isn't detected as part of
1893 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1894 set_buffer_dirty(bh);
1895 sync_dirty_buffer(bh);
1897 /* clear the mirror copies of super block on the disk
1898 * being removed, 0th copy is been taken care above and
1899 * the below would take of the rest
1901 for (i = 1; i < BTRFS_SUPER_MIRROR_MAX; i++) {
1902 bytenr = btrfs_sb_offset(i);
1903 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
1904 i_size_read(bdev->bd_inode))
1908 bh = __bread(bdev, bytenr / 4096,
1909 BTRFS_SUPER_INFO_SIZE);
1913 disk_super = (struct btrfs_super_block *)bh->b_data;
1915 if (btrfs_super_bytenr(disk_super) != bytenr ||
1916 btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
1919 memset(&disk_super->magic, 0,
1920 sizeof(disk_super->magic));
1921 set_buffer_dirty(bh);
1922 sync_dirty_buffer(bh);
1929 /* Notify udev that device has changed */
1930 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
1932 /* Update ctime/mtime for device path for libblkid */
1933 update_dev_time(device_path);
1939 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1941 mutex_unlock(&uuid_mutex);
1944 if (device->writeable) {
1946 list_add(&device->dev_alloc_list,
1947 &root->fs_info->fs_devices->alloc_list);
1948 device->fs_devices->rw_devices++;
1949 unlock_chunks(root);
1954 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
1955 struct btrfs_device *srcdev)
1957 struct btrfs_fs_devices *fs_devices;
1959 WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1962 * in case of fs with no seed, srcdev->fs_devices will point
1963 * to fs_devices of fs_info. However when the dev being replaced is
1964 * a seed dev it will point to the seed's local fs_devices. In short
1965 * srcdev will have its correct fs_devices in both the cases.
1967 fs_devices = srcdev->fs_devices;
1969 list_del_rcu(&srcdev->dev_list);
1970 list_del_rcu(&srcdev->dev_alloc_list);
1971 fs_devices->num_devices--;
1972 if (srcdev->missing)
1973 fs_devices->missing_devices--;
1975 if (srcdev->writeable) {
1976 fs_devices->rw_devices--;
1977 /* zero out the old super if it is writable */
1978 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
1982 fs_devices->open_devices--;
1985 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
1986 struct btrfs_device *srcdev)
1988 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
1990 call_rcu(&srcdev->rcu, free_device);
1993 * unless fs_devices is seed fs, num_devices shouldn't go
1996 BUG_ON(!fs_devices->num_devices && !fs_devices->seeding);
1998 /* if this is no devs we rather delete the fs_devices */
1999 if (!fs_devices->num_devices) {
2000 struct btrfs_fs_devices *tmp_fs_devices;
2002 tmp_fs_devices = fs_info->fs_devices;
2003 while (tmp_fs_devices) {
2004 if (tmp_fs_devices->seed == fs_devices) {
2005 tmp_fs_devices->seed = fs_devices->seed;
2008 tmp_fs_devices = tmp_fs_devices->seed;
2010 fs_devices->seed = NULL;
2011 __btrfs_close_devices(fs_devices);
2012 free_fs_devices(fs_devices);
2016 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2017 struct btrfs_device *tgtdev)
2019 struct btrfs_device *next_device;
2021 mutex_lock(&uuid_mutex);
2023 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2025 btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev);
2028 btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
2029 fs_info->fs_devices->open_devices--;
2031 fs_info->fs_devices->num_devices--;
2033 next_device = list_entry(fs_info->fs_devices->devices.next,
2034 struct btrfs_device, dev_list);
2035 if (tgtdev->bdev == fs_info->sb->s_bdev)
2036 fs_info->sb->s_bdev = next_device->bdev;
2037 if (tgtdev->bdev == fs_info->fs_devices->latest_bdev)
2038 fs_info->fs_devices->latest_bdev = next_device->bdev;
2039 list_del_rcu(&tgtdev->dev_list);
2041 call_rcu(&tgtdev->rcu, free_device);
2043 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2044 mutex_unlock(&uuid_mutex);
2047 static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
2048 struct btrfs_device **device)
2051 struct btrfs_super_block *disk_super;
2054 struct block_device *bdev;
2055 struct buffer_head *bh;
2058 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2059 root->fs_info->bdev_holder, 0, &bdev, &bh);
2062 disk_super = (struct btrfs_super_block *)bh->b_data;
2063 devid = btrfs_stack_device_id(&disk_super->dev_item);
2064 dev_uuid = disk_super->dev_item.uuid;
2065 *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2070 blkdev_put(bdev, FMODE_READ);
2074 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
2076 struct btrfs_device **device)
2079 if (strcmp(device_path, "missing") == 0) {
2080 struct list_head *devices;
2081 struct btrfs_device *tmp;
2083 devices = &root->fs_info->fs_devices->devices;
2085 * It is safe to read the devices since the volume_mutex
2086 * is held by the caller.
2088 list_for_each_entry(tmp, devices, dev_list) {
2089 if (tmp->in_fs_metadata && !tmp->bdev) {
2096 return BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2100 return btrfs_find_device_by_path(root, device_path, device);
2105 * does all the dirty work required for changing file system's UUID.
2107 static int btrfs_prepare_sprout(struct btrfs_root *root)
2109 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2110 struct btrfs_fs_devices *old_devices;
2111 struct btrfs_fs_devices *seed_devices;
2112 struct btrfs_super_block *disk_super = root->fs_info->super_copy;
2113 struct btrfs_device *device;
2116 BUG_ON(!mutex_is_locked(&uuid_mutex));
2117 if (!fs_devices->seeding)
2120 seed_devices = __alloc_fs_devices();
2121 if (IS_ERR(seed_devices))
2122 return PTR_ERR(seed_devices);
2124 old_devices = clone_fs_devices(fs_devices);
2125 if (IS_ERR(old_devices)) {
2126 kfree(seed_devices);
2127 return PTR_ERR(old_devices);
2130 list_add(&old_devices->list, &fs_uuids);
2132 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2133 seed_devices->opened = 1;
2134 INIT_LIST_HEAD(&seed_devices->devices);
2135 INIT_LIST_HEAD(&seed_devices->alloc_list);
2136 mutex_init(&seed_devices->device_list_mutex);
2138 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2139 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2141 list_for_each_entry(device, &seed_devices->devices, dev_list)
2142 device->fs_devices = seed_devices;
2145 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2146 unlock_chunks(root);
2148 fs_devices->seeding = 0;
2149 fs_devices->num_devices = 0;
2150 fs_devices->open_devices = 0;
2151 fs_devices->missing_devices = 0;
2152 fs_devices->rotating = 0;
2153 fs_devices->seed = seed_devices;
2155 generate_random_uuid(fs_devices->fsid);
2156 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2157 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2158 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2160 super_flags = btrfs_super_flags(disk_super) &
2161 ~BTRFS_SUPER_FLAG_SEEDING;
2162 btrfs_set_super_flags(disk_super, super_flags);
2168 * strore the expected generation for seed devices in device items.
2170 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2171 struct btrfs_root *root)
2173 struct btrfs_path *path;
2174 struct extent_buffer *leaf;
2175 struct btrfs_dev_item *dev_item;
2176 struct btrfs_device *device;
2177 struct btrfs_key key;
2178 u8 fs_uuid[BTRFS_UUID_SIZE];
2179 u8 dev_uuid[BTRFS_UUID_SIZE];
2183 path = btrfs_alloc_path();
2187 root = root->fs_info->chunk_root;
2188 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2190 key.type = BTRFS_DEV_ITEM_KEY;
2193 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2197 leaf = path->nodes[0];
2199 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2200 ret = btrfs_next_leaf(root, path);
2205 leaf = path->nodes[0];
2206 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2207 btrfs_release_path(path);
2211 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2212 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2213 key.type != BTRFS_DEV_ITEM_KEY)
2216 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2217 struct btrfs_dev_item);
2218 devid = btrfs_device_id(leaf, dev_item);
2219 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2221 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2223 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2225 BUG_ON(!device); /* Logic error */
2227 if (device->fs_devices->seeding) {
2228 btrfs_set_device_generation(leaf, dev_item,
2229 device->generation);
2230 btrfs_mark_buffer_dirty(leaf);
2238 btrfs_free_path(path);
2242 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
2244 struct request_queue *q;
2245 struct btrfs_trans_handle *trans;
2246 struct btrfs_device *device;
2247 struct block_device *bdev;
2248 struct list_head *devices;
2249 struct super_block *sb = root->fs_info->sb;
2250 struct rcu_string *name;
2252 int seeding_dev = 0;
2255 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
2258 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2259 root->fs_info->bdev_holder);
2261 return PTR_ERR(bdev);
2263 if (root->fs_info->fs_devices->seeding) {
2265 down_write(&sb->s_umount);
2266 mutex_lock(&uuid_mutex);
2269 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2271 devices = &root->fs_info->fs_devices->devices;
2273 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2274 list_for_each_entry(device, devices, dev_list) {
2275 if (device->bdev == bdev) {
2278 &root->fs_info->fs_devices->device_list_mutex);
2282 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2284 device = btrfs_alloc_device(root->fs_info, NULL, NULL);
2285 if (IS_ERR(device)) {
2286 /* we can safely leave the fs_devices entry around */
2287 ret = PTR_ERR(device);
2291 name = rcu_string_strdup(device_path, GFP_NOFS);
2297 rcu_assign_pointer(device->name, name);
2299 trans = btrfs_start_transaction(root, 0);
2300 if (IS_ERR(trans)) {
2301 rcu_string_free(device->name);
2303 ret = PTR_ERR(trans);
2307 q = bdev_get_queue(bdev);
2308 if (blk_queue_discard(q))
2309 device->can_discard = 1;
2310 device->writeable = 1;
2311 device->generation = trans->transid;
2312 device->io_width = root->sectorsize;
2313 device->io_align = root->sectorsize;
2314 device->sector_size = root->sectorsize;
2315 device->total_bytes = i_size_read(bdev->bd_inode);
2316 device->disk_total_bytes = device->total_bytes;
2317 device->commit_total_bytes = device->total_bytes;
2318 device->dev_root = root->fs_info->dev_root;
2319 device->bdev = bdev;
2320 device->in_fs_metadata = 1;
2321 device->is_tgtdev_for_dev_replace = 0;
2322 device->mode = FMODE_EXCL;
2323 device->dev_stats_valid = 1;
2324 set_blocksize(device->bdev, 4096);
2327 sb->s_flags &= ~MS_RDONLY;
2328 ret = btrfs_prepare_sprout(root);
2329 BUG_ON(ret); /* -ENOMEM */
2332 device->fs_devices = root->fs_info->fs_devices;
2334 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2336 list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
2337 list_add(&device->dev_alloc_list,
2338 &root->fs_info->fs_devices->alloc_list);
2339 root->fs_info->fs_devices->num_devices++;
2340 root->fs_info->fs_devices->open_devices++;
2341 root->fs_info->fs_devices->rw_devices++;
2342 root->fs_info->fs_devices->total_devices++;
2343 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2345 spin_lock(&root->fs_info->free_chunk_lock);
2346 root->fs_info->free_chunk_space += device->total_bytes;
2347 spin_unlock(&root->fs_info->free_chunk_lock);
2349 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2350 root->fs_info->fs_devices->rotating = 1;
2352 tmp = btrfs_super_total_bytes(root->fs_info->super_copy);
2353 btrfs_set_super_total_bytes(root->fs_info->super_copy,
2354 tmp + device->total_bytes);
2356 tmp = btrfs_super_num_devices(root->fs_info->super_copy);
2357 btrfs_set_super_num_devices(root->fs_info->super_copy,
2361 * we've got more storage, clear any full flags on the space
2364 btrfs_clear_space_info_full(root->fs_info);
2366 unlock_chunks(root);
2368 /* add sysfs device entry */
2369 btrfs_sysfs_add_device_link(root->fs_info->fs_devices, device);
2371 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2375 ret = init_first_rw_device(trans, root, device);
2376 unlock_chunks(root);
2378 btrfs_abort_transaction(trans, root, ret);
2383 ret = btrfs_add_device(trans, root, device);
2385 btrfs_abort_transaction(trans, root, ret);
2390 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2392 ret = btrfs_finish_sprout(trans, root);
2394 btrfs_abort_transaction(trans, root, ret);
2398 /* Sprouting would change fsid of the mounted root,
2399 * so rename the fsid on the sysfs
2401 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2402 root->fs_info->fsid);
2403 if (kobject_rename(&root->fs_info->fs_devices->fsid_kobj,
2405 btrfs_warn(root->fs_info,
2406 "sysfs: failed to create fsid for sprout");
2409 root->fs_info->num_tolerated_disk_barrier_failures =
2410 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
2411 ret = btrfs_commit_transaction(trans, root);
2414 mutex_unlock(&uuid_mutex);
2415 up_write(&sb->s_umount);
2417 if (ret) /* transaction commit */
2420 ret = btrfs_relocate_sys_chunks(root);
2422 btrfs_std_error(root->fs_info, ret,
2423 "Failed to relocate sys chunks after "
2424 "device initialization. This can be fixed "
2425 "using the \"btrfs balance\" command.");
2426 trans = btrfs_attach_transaction(root);
2427 if (IS_ERR(trans)) {
2428 if (PTR_ERR(trans) == -ENOENT)
2430 return PTR_ERR(trans);
2432 ret = btrfs_commit_transaction(trans, root);
2435 /* Update ctime/mtime for libblkid */
2436 update_dev_time(device_path);
2440 btrfs_end_transaction(trans, root);
2441 rcu_string_free(device->name);
2442 btrfs_sysfs_rm_device_link(root->fs_info->fs_devices, device);
2445 blkdev_put(bdev, FMODE_EXCL);
2447 mutex_unlock(&uuid_mutex);
2448 up_write(&sb->s_umount);
2453 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2454 struct btrfs_device *srcdev,
2455 struct btrfs_device **device_out)
2457 struct request_queue *q;
2458 struct btrfs_device *device;
2459 struct block_device *bdev;
2460 struct btrfs_fs_info *fs_info = root->fs_info;
2461 struct list_head *devices;
2462 struct rcu_string *name;
2463 u64 devid = BTRFS_DEV_REPLACE_DEVID;
2467 if (fs_info->fs_devices->seeding) {
2468 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
2472 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2473 fs_info->bdev_holder);
2475 btrfs_err(fs_info, "target device %s is invalid!", device_path);
2476 return PTR_ERR(bdev);
2479 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2481 devices = &fs_info->fs_devices->devices;
2482 list_for_each_entry(device, devices, dev_list) {
2483 if (device->bdev == bdev) {
2484 btrfs_err(fs_info, "target device is in the filesystem!");
2491 if (i_size_read(bdev->bd_inode) <
2492 btrfs_device_get_total_bytes(srcdev)) {
2493 btrfs_err(fs_info, "target device is smaller than source device!");
2499 device = btrfs_alloc_device(NULL, &devid, NULL);
2500 if (IS_ERR(device)) {
2501 ret = PTR_ERR(device);
2505 name = rcu_string_strdup(device_path, GFP_NOFS);
2511 rcu_assign_pointer(device->name, name);
2513 q = bdev_get_queue(bdev);
2514 if (blk_queue_discard(q))
2515 device->can_discard = 1;
2516 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2517 device->writeable = 1;
2518 device->generation = 0;
2519 device->io_width = root->sectorsize;
2520 device->io_align = root->sectorsize;
2521 device->sector_size = root->sectorsize;
2522 device->total_bytes = btrfs_device_get_total_bytes(srcdev);
2523 device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
2524 device->bytes_used = btrfs_device_get_bytes_used(srcdev);
2525 ASSERT(list_empty(&srcdev->resized_list));
2526 device->commit_total_bytes = srcdev->commit_total_bytes;
2527 device->commit_bytes_used = device->bytes_used;
2528 device->dev_root = fs_info->dev_root;
2529 device->bdev = bdev;
2530 device->in_fs_metadata = 1;
2531 device->is_tgtdev_for_dev_replace = 1;
2532 device->mode = FMODE_EXCL;
2533 device->dev_stats_valid = 1;
2534 set_blocksize(device->bdev, 4096);
2535 device->fs_devices = fs_info->fs_devices;
2536 list_add(&device->dev_list, &fs_info->fs_devices->devices);
2537 fs_info->fs_devices->num_devices++;
2538 fs_info->fs_devices->open_devices++;
2539 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2541 *device_out = device;
2545 blkdev_put(bdev, FMODE_EXCL);
2549 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2550 struct btrfs_device *tgtdev)
2552 WARN_ON(fs_info->fs_devices->rw_devices == 0);
2553 tgtdev->io_width = fs_info->dev_root->sectorsize;
2554 tgtdev->io_align = fs_info->dev_root->sectorsize;
2555 tgtdev->sector_size = fs_info->dev_root->sectorsize;
2556 tgtdev->dev_root = fs_info->dev_root;
2557 tgtdev->in_fs_metadata = 1;
2560 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2561 struct btrfs_device *device)
2564 struct btrfs_path *path;
2565 struct btrfs_root *root;
2566 struct btrfs_dev_item *dev_item;
2567 struct extent_buffer *leaf;
2568 struct btrfs_key key;
2570 root = device->dev_root->fs_info->chunk_root;
2572 path = btrfs_alloc_path();
2576 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2577 key.type = BTRFS_DEV_ITEM_KEY;
2578 key.offset = device->devid;
2580 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2589 leaf = path->nodes[0];
2590 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2592 btrfs_set_device_id(leaf, dev_item, device->devid);
2593 btrfs_set_device_type(leaf, dev_item, device->type);
2594 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2595 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2596 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2597 btrfs_set_device_total_bytes(leaf, dev_item,
2598 btrfs_device_get_disk_total_bytes(device));
2599 btrfs_set_device_bytes_used(leaf, dev_item,
2600 btrfs_device_get_bytes_used(device));
2601 btrfs_mark_buffer_dirty(leaf);
2604 btrfs_free_path(path);
2608 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2609 struct btrfs_device *device, u64 new_size)
2611 struct btrfs_super_block *super_copy =
2612 device->dev_root->fs_info->super_copy;
2613 struct btrfs_fs_devices *fs_devices;
2617 if (!device->writeable)
2620 lock_chunks(device->dev_root);
2621 old_total = btrfs_super_total_bytes(super_copy);
2622 diff = new_size - device->total_bytes;
2624 if (new_size <= device->total_bytes ||
2625 device->is_tgtdev_for_dev_replace) {
2626 unlock_chunks(device->dev_root);
2630 fs_devices = device->dev_root->fs_info->fs_devices;
2632 btrfs_set_super_total_bytes(super_copy, old_total + diff);
2633 device->fs_devices->total_rw_bytes += diff;
2635 btrfs_device_set_total_bytes(device, new_size);
2636 btrfs_device_set_disk_total_bytes(device, new_size);
2637 btrfs_clear_space_info_full(device->dev_root->fs_info);
2638 if (list_empty(&device->resized_list))
2639 list_add_tail(&device->resized_list,
2640 &fs_devices->resized_devices);
2641 unlock_chunks(device->dev_root);
2643 return btrfs_update_device(trans, device);
2646 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2647 struct btrfs_root *root, u64 chunk_objectid,
2651 struct btrfs_path *path;
2652 struct btrfs_key key;
2654 root = root->fs_info->chunk_root;
2655 path = btrfs_alloc_path();
2659 key.objectid = chunk_objectid;
2660 key.offset = chunk_offset;
2661 key.type = BTRFS_CHUNK_ITEM_KEY;
2663 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2666 else if (ret > 0) { /* Logic error or corruption */
2667 btrfs_std_error(root->fs_info, -ENOENT,
2668 "Failed lookup while freeing chunk.");
2673 ret = btrfs_del_item(trans, root, path);
2675 btrfs_std_error(root->fs_info, ret,
2676 "Failed to delete chunk item.");
2678 btrfs_free_path(path);
2682 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2685 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2686 struct btrfs_disk_key *disk_key;
2687 struct btrfs_chunk *chunk;
2694 struct btrfs_key key;
2697 array_size = btrfs_super_sys_array_size(super_copy);
2699 ptr = super_copy->sys_chunk_array;
2702 while (cur < array_size) {
2703 disk_key = (struct btrfs_disk_key *)ptr;
2704 btrfs_disk_key_to_cpu(&key, disk_key);
2706 len = sizeof(*disk_key);
2708 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2709 chunk = (struct btrfs_chunk *)(ptr + len);
2710 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2711 len += btrfs_chunk_item_size(num_stripes);
2716 if (key.objectid == chunk_objectid &&
2717 key.offset == chunk_offset) {
2718 memmove(ptr, ptr + len, array_size - (cur + len));
2720 btrfs_set_super_sys_array_size(super_copy, array_size);
2726 unlock_chunks(root);
2730 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2731 struct btrfs_root *root, u64 chunk_offset)
2733 struct extent_map_tree *em_tree;
2734 struct extent_map *em;
2735 struct btrfs_root *extent_root = root->fs_info->extent_root;
2736 struct map_lookup *map;
2737 u64 dev_extent_len = 0;
2738 u64 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2742 root = root->fs_info->chunk_root;
2743 em_tree = &root->fs_info->mapping_tree.map_tree;
2745 read_lock(&em_tree->lock);
2746 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2747 read_unlock(&em_tree->lock);
2749 if (!em || em->start > chunk_offset ||
2750 em->start + em->len < chunk_offset) {
2752 * This is a logic error, but we don't want to just rely on the
2753 * user having built with ASSERT enabled, so if ASSERT doens't
2754 * do anything we still error out.
2758 free_extent_map(em);
2761 map = em->map_lookup;
2762 lock_chunks(root->fs_info->chunk_root);
2763 check_system_chunk(trans, extent_root, map->type);
2764 unlock_chunks(root->fs_info->chunk_root);
2766 for (i = 0; i < map->num_stripes; i++) {
2767 struct btrfs_device *device = map->stripes[i].dev;
2768 ret = btrfs_free_dev_extent(trans, device,
2769 map->stripes[i].physical,
2772 btrfs_abort_transaction(trans, root, ret);
2776 if (device->bytes_used > 0) {
2778 btrfs_device_set_bytes_used(device,
2779 device->bytes_used - dev_extent_len);
2780 spin_lock(&root->fs_info->free_chunk_lock);
2781 root->fs_info->free_chunk_space += dev_extent_len;
2782 spin_unlock(&root->fs_info->free_chunk_lock);
2783 btrfs_clear_space_info_full(root->fs_info);
2784 unlock_chunks(root);
2787 if (map->stripes[i].dev) {
2788 ret = btrfs_update_device(trans, map->stripes[i].dev);
2790 btrfs_abort_transaction(trans, root, ret);
2795 ret = btrfs_free_chunk(trans, root, chunk_objectid, chunk_offset);
2797 btrfs_abort_transaction(trans, root, ret);
2801 trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2803 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2804 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2806 btrfs_abort_transaction(trans, root, ret);
2811 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset, em);
2813 btrfs_abort_transaction(trans, extent_root, ret);
2819 free_extent_map(em);
2823 static int btrfs_relocate_chunk(struct btrfs_root *root, u64 chunk_offset)
2825 struct btrfs_root *extent_root;
2826 struct btrfs_trans_handle *trans;
2829 root = root->fs_info->chunk_root;
2830 extent_root = root->fs_info->extent_root;
2833 * Prevent races with automatic removal of unused block groups.
2834 * After we relocate and before we remove the chunk with offset
2835 * chunk_offset, automatic removal of the block group can kick in,
2836 * resulting in a failure when calling btrfs_remove_chunk() below.
2838 * Make sure to acquire this mutex before doing a tree search (dev
2839 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2840 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2841 * we release the path used to search the chunk/dev tree and before
2842 * the current task acquires this mutex and calls us.
2844 ASSERT(mutex_is_locked(&root->fs_info->delete_unused_bgs_mutex));
2846 ret = btrfs_can_relocate(extent_root, chunk_offset);
2850 /* step one, relocate all the extents inside this chunk */
2851 btrfs_scrub_pause(root);
2852 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2853 btrfs_scrub_continue(root);
2857 trans = btrfs_start_trans_remove_block_group(root->fs_info,
2859 if (IS_ERR(trans)) {
2860 ret = PTR_ERR(trans);
2861 btrfs_std_error(root->fs_info, ret, NULL);
2866 * step two, delete the device extents and the
2867 * chunk tree entries
2869 ret = btrfs_remove_chunk(trans, root, chunk_offset);
2870 btrfs_end_transaction(trans, root);
2874 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2876 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2877 struct btrfs_path *path;
2878 struct extent_buffer *leaf;
2879 struct btrfs_chunk *chunk;
2880 struct btrfs_key key;
2881 struct btrfs_key found_key;
2883 bool retried = false;
2887 path = btrfs_alloc_path();
2892 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2893 key.offset = (u64)-1;
2894 key.type = BTRFS_CHUNK_ITEM_KEY;
2897 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
2898 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2900 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2903 BUG_ON(ret == 0); /* Corruption */
2905 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2908 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2914 leaf = path->nodes[0];
2915 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2917 chunk = btrfs_item_ptr(leaf, path->slots[0],
2918 struct btrfs_chunk);
2919 chunk_type = btrfs_chunk_type(leaf, chunk);
2920 btrfs_release_path(path);
2922 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2923 ret = btrfs_relocate_chunk(chunk_root,
2930 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2932 if (found_key.offset == 0)
2934 key.offset = found_key.offset - 1;
2937 if (failed && !retried) {
2941 } else if (WARN_ON(failed && retried)) {
2945 btrfs_free_path(path);
2949 static int insert_balance_item(struct btrfs_root *root,
2950 struct btrfs_balance_control *bctl)
2952 struct btrfs_trans_handle *trans;
2953 struct btrfs_balance_item *item;
2954 struct btrfs_disk_balance_args disk_bargs;
2955 struct btrfs_path *path;
2956 struct extent_buffer *leaf;
2957 struct btrfs_key key;
2960 path = btrfs_alloc_path();
2964 trans = btrfs_start_transaction(root, 0);
2965 if (IS_ERR(trans)) {
2966 btrfs_free_path(path);
2967 return PTR_ERR(trans);
2970 key.objectid = BTRFS_BALANCE_OBJECTID;
2971 key.type = BTRFS_BALANCE_ITEM_KEY;
2974 ret = btrfs_insert_empty_item(trans, root, path, &key,
2979 leaf = path->nodes[0];
2980 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2982 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2984 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2985 btrfs_set_balance_data(leaf, item, &disk_bargs);
2986 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2987 btrfs_set_balance_meta(leaf, item, &disk_bargs);
2988 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2989 btrfs_set_balance_sys(leaf, item, &disk_bargs);
2991 btrfs_set_balance_flags(leaf, item, bctl->flags);
2993 btrfs_mark_buffer_dirty(leaf);
2995 btrfs_free_path(path);
2996 err = btrfs_commit_transaction(trans, root);
3002 static int del_balance_item(struct btrfs_root *root)
3004 struct btrfs_trans_handle *trans;
3005 struct btrfs_path *path;
3006 struct btrfs_key key;
3009 path = btrfs_alloc_path();
3013 trans = btrfs_start_transaction(root, 0);
3014 if (IS_ERR(trans)) {
3015 btrfs_free_path(path);
3016 return PTR_ERR(trans);
3019 key.objectid = BTRFS_BALANCE_OBJECTID;
3020 key.type = BTRFS_BALANCE_ITEM_KEY;
3023 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3031 ret = btrfs_del_item(trans, root, path);
3033 btrfs_free_path(path);
3034 err = btrfs_commit_transaction(trans, root);
3041 * This is a heuristic used to reduce the number of chunks balanced on
3042 * resume after balance was interrupted.
3044 static void update_balance_args(struct btrfs_balance_control *bctl)
3047 * Turn on soft mode for chunk types that were being converted.
3049 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3050 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3051 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3052 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3053 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3054 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3057 * Turn on usage filter if is not already used. The idea is
3058 * that chunks that we have already balanced should be
3059 * reasonably full. Don't do it for chunks that are being
3060 * converted - that will keep us from relocating unconverted
3061 * (albeit full) chunks.
3063 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3064 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3065 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3066 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3067 bctl->data.usage = 90;
3069 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3070 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3071 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3072 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3073 bctl->sys.usage = 90;
3075 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3076 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3077 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3078 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3079 bctl->meta.usage = 90;
3084 * Should be called with both balance and volume mutexes held to
3085 * serialize other volume operations (add_dev/rm_dev/resize) with
3086 * restriper. Same goes for unset_balance_control.
3088 static void set_balance_control(struct btrfs_balance_control *bctl)
3090 struct btrfs_fs_info *fs_info = bctl->fs_info;
3092 BUG_ON(fs_info->balance_ctl);
3094 spin_lock(&fs_info->balance_lock);
3095 fs_info->balance_ctl = bctl;
3096 spin_unlock(&fs_info->balance_lock);
3099 static void unset_balance_control(struct btrfs_fs_info *fs_info)
3101 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3103 BUG_ON(!fs_info->balance_ctl);
3105 spin_lock(&fs_info->balance_lock);
3106 fs_info->balance_ctl = NULL;
3107 spin_unlock(&fs_info->balance_lock);
3113 * Balance filters. Return 1 if chunk should be filtered out
3114 * (should not be balanced).
3116 static int chunk_profiles_filter(u64 chunk_type,
3117 struct btrfs_balance_args *bargs)
3119 chunk_type = chunk_to_extended(chunk_type) &
3120 BTRFS_EXTENDED_PROFILE_MASK;
3122 if (bargs->profiles & chunk_type)
3128 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3129 struct btrfs_balance_args *bargs)
3131 struct btrfs_block_group_cache *cache;
3133 u64 user_thresh_min;
3134 u64 user_thresh_max;
3137 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3138 chunk_used = btrfs_block_group_used(&cache->item);
3140 if (bargs->usage_min == 0)
3141 user_thresh_min = 0;
3143 user_thresh_min = div_factor_fine(cache->key.offset,
3146 if (bargs->usage_max == 0)
3147 user_thresh_max = 1;
3148 else if (bargs->usage_max > 100)
3149 user_thresh_max = cache->key.offset;
3151 user_thresh_max = div_factor_fine(cache->key.offset,
3154 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3157 btrfs_put_block_group(cache);
3161 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3162 u64 chunk_offset, struct btrfs_balance_args *bargs)
3164 struct btrfs_block_group_cache *cache;
3165 u64 chunk_used, user_thresh;
3168 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3169 chunk_used = btrfs_block_group_used(&cache->item);
3171 if (bargs->usage_min == 0)
3173 else if (bargs->usage > 100)
3174 user_thresh = cache->key.offset;
3176 user_thresh = div_factor_fine(cache->key.offset,
3179 if (chunk_used < user_thresh)
3182 btrfs_put_block_group(cache);
3186 static int chunk_devid_filter(struct extent_buffer *leaf,
3187 struct btrfs_chunk *chunk,
3188 struct btrfs_balance_args *bargs)
3190 struct btrfs_stripe *stripe;
3191 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3194 for (i = 0; i < num_stripes; i++) {
3195 stripe = btrfs_stripe_nr(chunk, i);
3196 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3203 /* [pstart, pend) */
3204 static int chunk_drange_filter(struct extent_buffer *leaf,
3205 struct btrfs_chunk *chunk,
3207 struct btrfs_balance_args *bargs)
3209 struct btrfs_stripe *stripe;
3210 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3216 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3219 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3220 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3221 factor = num_stripes / 2;
3222 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3223 factor = num_stripes - 1;
3224 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3225 factor = num_stripes - 2;
3227 factor = num_stripes;
3230 for (i = 0; i < num_stripes; i++) {
3231 stripe = btrfs_stripe_nr(chunk, i);
3232 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3235 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3236 stripe_length = btrfs_chunk_length(leaf, chunk);
3237 stripe_length = div_u64(stripe_length, factor);
3239 if (stripe_offset < bargs->pend &&
3240 stripe_offset + stripe_length > bargs->pstart)
3247 /* [vstart, vend) */
3248 static int chunk_vrange_filter(struct extent_buffer *leaf,
3249 struct btrfs_chunk *chunk,
3251 struct btrfs_balance_args *bargs)
3253 if (chunk_offset < bargs->vend &&
3254 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3255 /* at least part of the chunk is inside this vrange */
3261 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3262 struct btrfs_chunk *chunk,
3263 struct btrfs_balance_args *bargs)
3265 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3267 if (bargs->stripes_min <= num_stripes
3268 && num_stripes <= bargs->stripes_max)
3274 static int chunk_soft_convert_filter(u64 chunk_type,
3275 struct btrfs_balance_args *bargs)
3277 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3280 chunk_type = chunk_to_extended(chunk_type) &
3281 BTRFS_EXTENDED_PROFILE_MASK;
3283 if (bargs->target == chunk_type)
3289 static int should_balance_chunk(struct btrfs_root *root,
3290 struct extent_buffer *leaf,
3291 struct btrfs_chunk *chunk, u64 chunk_offset)
3293 struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
3294 struct btrfs_balance_args *bargs = NULL;
3295 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3298 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3299 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3303 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3304 bargs = &bctl->data;
3305 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3307 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3308 bargs = &bctl->meta;
3310 /* profiles filter */
3311 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3312 chunk_profiles_filter(chunk_type, bargs)) {
3317 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3318 chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
3320 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3321 chunk_usage_range_filter(bctl->fs_info, chunk_offset, bargs)) {
3326 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3327 chunk_devid_filter(leaf, chunk, bargs)) {
3331 /* drange filter, makes sense only with devid filter */
3332 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3333 chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
3338 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3339 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3343 /* stripes filter */
3344 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3345 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3349 /* soft profile changing mode */
3350 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3351 chunk_soft_convert_filter(chunk_type, bargs)) {
3356 * limited by count, must be the last filter
3358 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3359 if (bargs->limit == 0)
3363 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3365 * Same logic as the 'limit' filter; the minimum cannot be
3366 * determined here because we do not have the global informatoin
3367 * about the count of all chunks that satisfy the filters.
3369 if (bargs->limit_max == 0)
3378 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3380 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3381 struct btrfs_root *chunk_root = fs_info->chunk_root;
3382 struct btrfs_root *dev_root = fs_info->dev_root;
3383 struct list_head *devices;
3384 struct btrfs_device *device;
3388 struct btrfs_chunk *chunk;
3389 struct btrfs_path *path;
3390 struct btrfs_key key;
3391 struct btrfs_key found_key;
3392 struct btrfs_trans_handle *trans;
3393 struct extent_buffer *leaf;
3396 int enospc_errors = 0;
3397 bool counting = true;
3398 /* The single value limit and min/max limits use the same bytes in the */
3399 u64 limit_data = bctl->data.limit;
3400 u64 limit_meta = bctl->meta.limit;
3401 u64 limit_sys = bctl->sys.limit;
3405 int chunk_reserved = 0;
3407 /* step one make some room on all the devices */
3408 devices = &fs_info->fs_devices->devices;
3409 list_for_each_entry(device, devices, dev_list) {
3410 old_size = btrfs_device_get_total_bytes(device);
3411 size_to_free = div_factor(old_size, 1);
3412 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
3413 if (!device->writeable ||
3414 btrfs_device_get_total_bytes(device) -
3415 btrfs_device_get_bytes_used(device) > size_to_free ||
3416 device->is_tgtdev_for_dev_replace)
3419 ret = btrfs_shrink_device(device, old_size - size_to_free);
3424 trans = btrfs_start_transaction(dev_root, 0);
3425 BUG_ON(IS_ERR(trans));
3427 ret = btrfs_grow_device(trans, device, old_size);
3430 btrfs_end_transaction(trans, dev_root);
3433 /* step two, relocate all the chunks */
3434 path = btrfs_alloc_path();
3440 /* zero out stat counters */
3441 spin_lock(&fs_info->balance_lock);
3442 memset(&bctl->stat, 0, sizeof(bctl->stat));
3443 spin_unlock(&fs_info->balance_lock);
3447 * The single value limit and min/max limits use the same bytes
3450 bctl->data.limit = limit_data;
3451 bctl->meta.limit = limit_meta;
3452 bctl->sys.limit = limit_sys;
3454 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3455 key.offset = (u64)-1;
3456 key.type = BTRFS_CHUNK_ITEM_KEY;
3459 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3460 atomic_read(&fs_info->balance_cancel_req)) {
3465 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3466 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3468 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3473 * this shouldn't happen, it means the last relocate
3477 BUG(); /* FIXME break ? */
3479 ret = btrfs_previous_item(chunk_root, path, 0,
3480 BTRFS_CHUNK_ITEM_KEY);
3482 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3487 leaf = path->nodes[0];
3488 slot = path->slots[0];
3489 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3491 if (found_key.objectid != key.objectid) {
3492 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3496 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3497 chunk_type = btrfs_chunk_type(leaf, chunk);
3500 spin_lock(&fs_info->balance_lock);
3501 bctl->stat.considered++;
3502 spin_unlock(&fs_info->balance_lock);
3505 ret = should_balance_chunk(chunk_root, leaf, chunk,
3508 btrfs_release_path(path);
3510 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3515 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3516 spin_lock(&fs_info->balance_lock);
3517 bctl->stat.expected++;
3518 spin_unlock(&fs_info->balance_lock);
3520 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3522 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3524 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3531 * Apply limit_min filter, no need to check if the LIMITS
3532 * filter is used, limit_min is 0 by default
3534 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3535 count_data < bctl->data.limit_min)
3536 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3537 count_meta < bctl->meta.limit_min)
3538 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3539 count_sys < bctl->sys.limit_min)) {
3540 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3544 if ((chunk_type & BTRFS_BLOCK_GROUP_DATA) && !chunk_reserved) {
3545 trans = btrfs_start_transaction(chunk_root, 0);
3546 if (IS_ERR(trans)) {
3547 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3548 ret = PTR_ERR(trans);
3552 ret = btrfs_force_chunk_alloc(trans, chunk_root,
3553 BTRFS_BLOCK_GROUP_DATA);
3554 btrfs_end_transaction(trans, chunk_root);
3556 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3562 ret = btrfs_relocate_chunk(chunk_root,
3564 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3565 if (ret && ret != -ENOSPC)
3567 if (ret == -ENOSPC) {
3570 spin_lock(&fs_info->balance_lock);
3571 bctl->stat.completed++;
3572 spin_unlock(&fs_info->balance_lock);
3575 if (found_key.offset == 0)
3577 key.offset = found_key.offset - 1;
3581 btrfs_release_path(path);
3586 btrfs_free_path(path);
3587 if (enospc_errors) {
3588 btrfs_info(fs_info, "%d enospc errors during balance",
3598 * alloc_profile_is_valid - see if a given profile is valid and reduced
3599 * @flags: profile to validate
3600 * @extended: if true @flags is treated as an extended profile
3602 static int alloc_profile_is_valid(u64 flags, int extended)
3604 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3605 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3607 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3609 /* 1) check that all other bits are zeroed */
3613 /* 2) see if profile is reduced */
3615 return !extended; /* "0" is valid for usual profiles */
3617 /* true if exactly one bit set */
3618 return (flags & (flags - 1)) == 0;
3621 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3623 /* cancel requested || normal exit path */
3624 return atomic_read(&fs_info->balance_cancel_req) ||
3625 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3626 atomic_read(&fs_info->balance_cancel_req) == 0);
3629 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3633 unset_balance_control(fs_info);
3634 ret = del_balance_item(fs_info->tree_root);
3636 btrfs_std_error(fs_info, ret, NULL);
3638 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3641 /* Non-zero return value signifies invalidity */
3642 static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg,
3645 return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3646 (!alloc_profile_is_valid(bctl_arg->target, 1) ||
3647 (bctl_arg->target & ~allowed)));
3651 * Should be called with both balance and volume mutexes held
3653 int btrfs_balance(struct btrfs_balance_control *bctl,
3654 struct btrfs_ioctl_balance_args *bargs)
3656 struct btrfs_fs_info *fs_info = bctl->fs_info;
3663 if (btrfs_fs_closing(fs_info) ||
3664 atomic_read(&fs_info->balance_pause_req) ||
3665 atomic_read(&fs_info->balance_cancel_req)) {
3670 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3671 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3675 * In case of mixed groups both data and meta should be picked,
3676 * and identical options should be given for both of them.
3678 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3679 if (mixed && (bctl->flags & allowed)) {
3680 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3681 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3682 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3683 btrfs_err(fs_info, "with mixed groups data and "
3684 "metadata balance options must be the same");
3690 num_devices = fs_info->fs_devices->num_devices;
3691 btrfs_dev_replace_lock(&fs_info->dev_replace);
3692 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3693 BUG_ON(num_devices < 1);
3696 btrfs_dev_replace_unlock(&fs_info->dev_replace);
3697 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3698 if (num_devices == 1)
3699 allowed |= BTRFS_BLOCK_GROUP_DUP;
3700 else if (num_devices > 1)
3701 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3702 if (num_devices > 2)
3703 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3704 if (num_devices > 3)
3705 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3706 BTRFS_BLOCK_GROUP_RAID6);
3707 if (validate_convert_profile(&bctl->data, allowed)) {
3708 btrfs_err(fs_info, "unable to start balance with target "
3709 "data profile %llu",
3714 if (validate_convert_profile(&bctl->meta, allowed)) {
3716 "unable to start balance with target metadata profile %llu",
3721 if (validate_convert_profile(&bctl->sys, allowed)) {
3723 "unable to start balance with target system profile %llu",
3729 /* allow dup'ed data chunks only in mixed mode */
3730 if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3731 (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) {
3732 btrfs_err(fs_info, "dup for data is not allowed");
3737 /* allow to reduce meta or sys integrity only if force set */
3738 allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3739 BTRFS_BLOCK_GROUP_RAID10 |
3740 BTRFS_BLOCK_GROUP_RAID5 |
3741 BTRFS_BLOCK_GROUP_RAID6;
3743 seq = read_seqbegin(&fs_info->profiles_lock);
3745 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3746 (fs_info->avail_system_alloc_bits & allowed) &&
3747 !(bctl->sys.target & allowed)) ||
3748 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3749 (fs_info->avail_metadata_alloc_bits & allowed) &&
3750 !(bctl->meta.target & allowed))) {
3751 if (bctl->flags & BTRFS_BALANCE_FORCE) {
3752 btrfs_info(fs_info, "force reducing metadata integrity");
3754 btrfs_err(fs_info, "balance will reduce metadata "
3755 "integrity, use force if you want this");
3760 } while (read_seqretry(&fs_info->profiles_lock, seq));
3762 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3763 fs_info->num_tolerated_disk_barrier_failures = min(
3764 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info),
3765 btrfs_get_num_tolerated_disk_barrier_failures(
3769 ret = insert_balance_item(fs_info->tree_root, bctl);
3770 if (ret && ret != -EEXIST)
3773 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3774 BUG_ON(ret == -EEXIST);
3775 set_balance_control(bctl);
3777 BUG_ON(ret != -EEXIST);
3778 spin_lock(&fs_info->balance_lock);
3779 update_balance_args(bctl);
3780 spin_unlock(&fs_info->balance_lock);
3783 atomic_inc(&fs_info->balance_running);
3784 mutex_unlock(&fs_info->balance_mutex);
3786 ret = __btrfs_balance(fs_info);
3788 mutex_lock(&fs_info->balance_mutex);
3789 atomic_dec(&fs_info->balance_running);
3791 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3792 fs_info->num_tolerated_disk_barrier_failures =
3793 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3797 memset(bargs, 0, sizeof(*bargs));
3798 update_ioctl_balance_args(fs_info, 0, bargs);
3801 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3802 balance_need_close(fs_info)) {
3803 __cancel_balance(fs_info);
3806 wake_up(&fs_info->balance_wait_q);
3810 if (bctl->flags & BTRFS_BALANCE_RESUME)
3811 __cancel_balance(fs_info);
3814 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3819 static int balance_kthread(void *data)
3821 struct btrfs_fs_info *fs_info = data;
3824 mutex_lock(&fs_info->volume_mutex);
3825 mutex_lock(&fs_info->balance_mutex);
3827 if (fs_info->balance_ctl) {
3828 btrfs_info(fs_info, "continuing balance");
3829 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3832 mutex_unlock(&fs_info->balance_mutex);
3833 mutex_unlock(&fs_info->volume_mutex);
3838 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3840 struct task_struct *tsk;
3842 spin_lock(&fs_info->balance_lock);
3843 if (!fs_info->balance_ctl) {
3844 spin_unlock(&fs_info->balance_lock);
3847 spin_unlock(&fs_info->balance_lock);
3849 if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
3850 btrfs_info(fs_info, "force skipping balance");
3855 * A ro->rw remount sequence should continue with the paused balance
3856 * regardless of who pauses it, system or the user as of now, so set
3859 spin_lock(&fs_info->balance_lock);
3860 fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME;
3861 spin_unlock(&fs_info->balance_lock);
3863 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3864 return PTR_ERR_OR_ZERO(tsk);
3867 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3869 struct btrfs_balance_control *bctl;
3870 struct btrfs_balance_item *item;
3871 struct btrfs_disk_balance_args disk_bargs;
3872 struct btrfs_path *path;
3873 struct extent_buffer *leaf;
3874 struct btrfs_key key;
3877 path = btrfs_alloc_path();
3881 key.objectid = BTRFS_BALANCE_OBJECTID;
3882 key.type = BTRFS_BALANCE_ITEM_KEY;
3885 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3888 if (ret > 0) { /* ret = -ENOENT; */
3893 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3899 leaf = path->nodes[0];
3900 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3902 bctl->fs_info = fs_info;
3903 bctl->flags = btrfs_balance_flags(leaf, item);
3904 bctl->flags |= BTRFS_BALANCE_RESUME;
3906 btrfs_balance_data(leaf, item, &disk_bargs);
3907 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3908 btrfs_balance_meta(leaf, item, &disk_bargs);
3909 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3910 btrfs_balance_sys(leaf, item, &disk_bargs);
3911 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3913 WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
3915 mutex_lock(&fs_info->volume_mutex);
3916 mutex_lock(&fs_info->balance_mutex);
3918 set_balance_control(bctl);
3920 mutex_unlock(&fs_info->balance_mutex);
3921 mutex_unlock(&fs_info->volume_mutex);
3923 btrfs_free_path(path);
3927 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3931 mutex_lock(&fs_info->balance_mutex);
3932 if (!fs_info->balance_ctl) {
3933 mutex_unlock(&fs_info->balance_mutex);
3937 if (atomic_read(&fs_info->balance_running)) {
3938 atomic_inc(&fs_info->balance_pause_req);
3939 mutex_unlock(&fs_info->balance_mutex);
3941 wait_event(fs_info->balance_wait_q,
3942 atomic_read(&fs_info->balance_running) == 0);
3944 mutex_lock(&fs_info->balance_mutex);
3945 /* we are good with balance_ctl ripped off from under us */
3946 BUG_ON(atomic_read(&fs_info->balance_running));
3947 atomic_dec(&fs_info->balance_pause_req);
3952 mutex_unlock(&fs_info->balance_mutex);
3956 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3958 if (fs_info->sb->s_flags & MS_RDONLY)
3961 mutex_lock(&fs_info->balance_mutex);
3962 if (!fs_info->balance_ctl) {
3963 mutex_unlock(&fs_info->balance_mutex);
3967 atomic_inc(&fs_info->balance_cancel_req);
3969 * if we are running just wait and return, balance item is
3970 * deleted in btrfs_balance in this case
3972 if (atomic_read(&fs_info->balance_running)) {
3973 mutex_unlock(&fs_info->balance_mutex);
3974 wait_event(fs_info->balance_wait_q,
3975 atomic_read(&fs_info->balance_running) == 0);
3976 mutex_lock(&fs_info->balance_mutex);
3978 /* __cancel_balance needs volume_mutex */
3979 mutex_unlock(&fs_info->balance_mutex);
3980 mutex_lock(&fs_info->volume_mutex);
3981 mutex_lock(&fs_info->balance_mutex);
3983 if (fs_info->balance_ctl)
3984 __cancel_balance(fs_info);
3986 mutex_unlock(&fs_info->volume_mutex);
3989 BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
3990 atomic_dec(&fs_info->balance_cancel_req);
3991 mutex_unlock(&fs_info->balance_mutex);
3995 static int btrfs_uuid_scan_kthread(void *data)
3997 struct btrfs_fs_info *fs_info = data;
3998 struct btrfs_root *root = fs_info->tree_root;
3999 struct btrfs_key key;
4000 struct btrfs_key max_key;
4001 struct btrfs_path *path = NULL;
4003 struct extent_buffer *eb;
4005 struct btrfs_root_item root_item;
4007 struct btrfs_trans_handle *trans = NULL;
4009 path = btrfs_alloc_path();
4016 key.type = BTRFS_ROOT_ITEM_KEY;
4019 max_key.objectid = (u64)-1;
4020 max_key.type = BTRFS_ROOT_ITEM_KEY;
4021 max_key.offset = (u64)-1;
4024 ret = btrfs_search_forward(root, &key, path, 0);
4031 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4032 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4033 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4034 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4037 eb = path->nodes[0];
4038 slot = path->slots[0];
4039 item_size = btrfs_item_size_nr(eb, slot);
4040 if (item_size < sizeof(root_item))
4043 read_extent_buffer(eb, &root_item,
4044 btrfs_item_ptr_offset(eb, slot),
4045 (int)sizeof(root_item));
4046 if (btrfs_root_refs(&root_item) == 0)
4049 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4050 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4054 btrfs_release_path(path);
4056 * 1 - subvol uuid item
4057 * 1 - received_subvol uuid item
4059 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4060 if (IS_ERR(trans)) {
4061 ret = PTR_ERR(trans);
4069 btrfs_release_path(path);
4070 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4071 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
4073 BTRFS_UUID_KEY_SUBVOL,
4076 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4082 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4083 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
4084 root_item.received_uuid,
4085 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4088 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4095 btrfs_release_path(path);
4097 ret = btrfs_end_transaction(trans, fs_info->uuid_root);
4103 if (key.offset < (u64)-1) {
4105 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4107 key.type = BTRFS_ROOT_ITEM_KEY;
4108 } else if (key.objectid < (u64)-1) {
4110 key.type = BTRFS_ROOT_ITEM_KEY;
4119 btrfs_free_path(path);
4120 if (trans && !IS_ERR(trans))
4121 btrfs_end_transaction(trans, fs_info->uuid_root);
4123 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4125 fs_info->update_uuid_tree_gen = 1;
4126 up(&fs_info->uuid_tree_rescan_sem);
4131 * Callback for btrfs_uuid_tree_iterate().
4133 * 0 check succeeded, the entry is not outdated.
4134 * < 0 if an error occured.
4135 * > 0 if the check failed, which means the caller shall remove the entry.
4137 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
4138 u8 *uuid, u8 type, u64 subid)
4140 struct btrfs_key key;
4142 struct btrfs_root *subvol_root;
4144 if (type != BTRFS_UUID_KEY_SUBVOL &&
4145 type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
4148 key.objectid = subid;
4149 key.type = BTRFS_ROOT_ITEM_KEY;
4150 key.offset = (u64)-1;
4151 subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
4152 if (IS_ERR(subvol_root)) {
4153 ret = PTR_ERR(subvol_root);
4160 case BTRFS_UUID_KEY_SUBVOL:
4161 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
4164 case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
4165 if (memcmp(uuid, subvol_root->root_item.received_uuid,
4175 static int btrfs_uuid_rescan_kthread(void *data)
4177 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4181 * 1st step is to iterate through the existing UUID tree and
4182 * to delete all entries that contain outdated data.
4183 * 2nd step is to add all missing entries to the UUID tree.
4185 ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4187 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4188 up(&fs_info->uuid_tree_rescan_sem);
4191 return btrfs_uuid_scan_kthread(data);
4194 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4196 struct btrfs_trans_handle *trans;
4197 struct btrfs_root *tree_root = fs_info->tree_root;
4198 struct btrfs_root *uuid_root;
4199 struct task_struct *task;
4206 trans = btrfs_start_transaction(tree_root, 2);
4208 return PTR_ERR(trans);
4210 uuid_root = btrfs_create_tree(trans, fs_info,
4211 BTRFS_UUID_TREE_OBJECTID);
4212 if (IS_ERR(uuid_root)) {
4213 ret = PTR_ERR(uuid_root);
4214 btrfs_abort_transaction(trans, tree_root, ret);
4218 fs_info->uuid_root = uuid_root;
4220 ret = btrfs_commit_transaction(trans, tree_root);
4224 down(&fs_info->uuid_tree_rescan_sem);
4225 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4227 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4228 btrfs_warn(fs_info, "failed to start uuid_scan task");
4229 up(&fs_info->uuid_tree_rescan_sem);
4230 return PTR_ERR(task);
4236 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4238 struct task_struct *task;
4240 down(&fs_info->uuid_tree_rescan_sem);
4241 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4243 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4244 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4245 up(&fs_info->uuid_tree_rescan_sem);
4246 return PTR_ERR(task);
4253 * shrinking a device means finding all of the device extents past
4254 * the new size, and then following the back refs to the chunks.
4255 * The chunk relocation code actually frees the device extent
4257 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4259 struct btrfs_trans_handle *trans;
4260 struct btrfs_root *root = device->dev_root;
4261 struct btrfs_dev_extent *dev_extent = NULL;
4262 struct btrfs_path *path;
4268 bool retried = false;
4269 bool checked_pending_chunks = false;
4270 struct extent_buffer *l;
4271 struct btrfs_key key;
4272 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4273 u64 old_total = btrfs_super_total_bytes(super_copy);
4274 u64 old_size = btrfs_device_get_total_bytes(device);
4275 u64 diff = old_size - new_size;
4277 if (device->is_tgtdev_for_dev_replace)
4280 path = btrfs_alloc_path();
4288 btrfs_device_set_total_bytes(device, new_size);
4289 if (device->writeable) {
4290 device->fs_devices->total_rw_bytes -= diff;
4291 spin_lock(&root->fs_info->free_chunk_lock);
4292 root->fs_info->free_chunk_space -= diff;
4293 spin_unlock(&root->fs_info->free_chunk_lock);
4295 unlock_chunks(root);
4298 key.objectid = device->devid;
4299 key.offset = (u64)-1;
4300 key.type = BTRFS_DEV_EXTENT_KEY;
4303 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
4304 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4306 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4310 ret = btrfs_previous_item(root, path, 0, key.type);
4312 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4317 btrfs_release_path(path);
4322 slot = path->slots[0];
4323 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4325 if (key.objectid != device->devid) {
4326 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4327 btrfs_release_path(path);
4331 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4332 length = btrfs_dev_extent_length(l, dev_extent);
4334 if (key.offset + length <= new_size) {
4335 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4336 btrfs_release_path(path);
4340 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4341 btrfs_release_path(path);
4343 ret = btrfs_relocate_chunk(root, chunk_offset);
4344 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4345 if (ret && ret != -ENOSPC)
4349 } while (key.offset-- > 0);
4351 if (failed && !retried) {
4355 } else if (failed && retried) {
4360 /* Shrinking succeeded, else we would be at "done". */
4361 trans = btrfs_start_transaction(root, 0);
4362 if (IS_ERR(trans)) {
4363 ret = PTR_ERR(trans);
4370 * We checked in the above loop all device extents that were already in
4371 * the device tree. However before we have updated the device's
4372 * total_bytes to the new size, we might have had chunk allocations that
4373 * have not complete yet (new block groups attached to transaction
4374 * handles), and therefore their device extents were not yet in the
4375 * device tree and we missed them in the loop above. So if we have any
4376 * pending chunk using a device extent that overlaps the device range
4377 * that we can not use anymore, commit the current transaction and
4378 * repeat the search on the device tree - this way we guarantee we will
4379 * not have chunks using device extents that end beyond 'new_size'.
4381 if (!checked_pending_chunks) {
4382 u64 start = new_size;
4383 u64 len = old_size - new_size;
4385 if (contains_pending_extent(trans->transaction, device,
4387 unlock_chunks(root);
4388 checked_pending_chunks = true;
4391 ret = btrfs_commit_transaction(trans, root);
4398 btrfs_device_set_disk_total_bytes(device, new_size);
4399 if (list_empty(&device->resized_list))
4400 list_add_tail(&device->resized_list,
4401 &root->fs_info->fs_devices->resized_devices);
4403 WARN_ON(diff > old_total);
4404 btrfs_set_super_total_bytes(super_copy, old_total - diff);
4405 unlock_chunks(root);
4407 /* Now btrfs_update_device() will change the on-disk size. */
4408 ret = btrfs_update_device(trans, device);
4409 btrfs_end_transaction(trans, root);
4411 btrfs_free_path(path);
4414 btrfs_device_set_total_bytes(device, old_size);
4415 if (device->writeable)
4416 device->fs_devices->total_rw_bytes += diff;
4417 spin_lock(&root->fs_info->free_chunk_lock);
4418 root->fs_info->free_chunk_space += diff;
4419 spin_unlock(&root->fs_info->free_chunk_lock);
4420 unlock_chunks(root);
4425 static int btrfs_add_system_chunk(struct btrfs_root *root,
4426 struct btrfs_key *key,
4427 struct btrfs_chunk *chunk, int item_size)
4429 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4430 struct btrfs_disk_key disk_key;
4435 array_size = btrfs_super_sys_array_size(super_copy);
4436 if (array_size + item_size + sizeof(disk_key)
4437 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4438 unlock_chunks(root);
4442 ptr = super_copy->sys_chunk_array + array_size;
4443 btrfs_cpu_key_to_disk(&disk_key, key);
4444 memcpy(ptr, &disk_key, sizeof(disk_key));
4445 ptr += sizeof(disk_key);
4446 memcpy(ptr, chunk, item_size);
4447 item_size += sizeof(disk_key);
4448 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4449 unlock_chunks(root);
4455 * sort the devices in descending order by max_avail, total_avail
4457 static int btrfs_cmp_device_info(const void *a, const void *b)
4459 const struct btrfs_device_info *di_a = a;
4460 const struct btrfs_device_info *di_b = b;
4462 if (di_a->max_avail > di_b->max_avail)
4464 if (di_a->max_avail < di_b->max_avail)
4466 if (di_a->total_avail > di_b->total_avail)
4468 if (di_a->total_avail < di_b->total_avail)
4473 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
4475 /* TODO allow them to set a preferred stripe size */
4479 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4481 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4484 btrfs_set_fs_incompat(info, RAID56);
4487 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r) \
4488 - sizeof(struct btrfs_item) \
4489 - sizeof(struct btrfs_chunk)) \
4490 / sizeof(struct btrfs_stripe) + 1)
4492 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
4493 - 2 * sizeof(struct btrfs_disk_key) \
4494 - 2 * sizeof(struct btrfs_chunk)) \
4495 / sizeof(struct btrfs_stripe) + 1)
4497 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4498 struct btrfs_root *extent_root, u64 start,
4501 struct btrfs_fs_info *info = extent_root->fs_info;
4502 struct btrfs_fs_devices *fs_devices = info->fs_devices;
4503 struct list_head *cur;
4504 struct map_lookup *map = NULL;
4505 struct extent_map_tree *em_tree;
4506 struct extent_map *em;
4507 struct btrfs_device_info *devices_info = NULL;
4509 int num_stripes; /* total number of stripes to allocate */
4510 int data_stripes; /* number of stripes that count for
4512 int sub_stripes; /* sub_stripes info for map */
4513 int dev_stripes; /* stripes per dev */
4514 int devs_max; /* max devs to use */
4515 int devs_min; /* min devs needed */
4516 int devs_increment; /* ndevs has to be a multiple of this */
4517 int ncopies; /* how many copies to data has */
4519 u64 max_stripe_size;
4523 u64 raid_stripe_len = BTRFS_STRIPE_LEN;
4529 BUG_ON(!alloc_profile_is_valid(type, 0));
4531 if (list_empty(&fs_devices->alloc_list))
4534 index = __get_raid_index(type);
4536 sub_stripes = btrfs_raid_array[index].sub_stripes;
4537 dev_stripes = btrfs_raid_array[index].dev_stripes;
4538 devs_max = btrfs_raid_array[index].devs_max;
4539 devs_min = btrfs_raid_array[index].devs_min;
4540 devs_increment = btrfs_raid_array[index].devs_increment;
4541 ncopies = btrfs_raid_array[index].ncopies;
4543 if (type & BTRFS_BLOCK_GROUP_DATA) {
4544 max_stripe_size = 1024 * 1024 * 1024;
4545 max_chunk_size = BTRFS_MAX_DATA_CHUNK_SIZE;
4547 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4548 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4549 /* for larger filesystems, use larger metadata chunks */
4550 if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024)
4551 max_stripe_size = 1024 * 1024 * 1024;
4553 max_stripe_size = 256 * 1024 * 1024;
4554 max_chunk_size = max_stripe_size;
4556 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4557 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4558 max_stripe_size = 32 * 1024 * 1024;
4559 max_chunk_size = 2 * max_stripe_size;
4561 devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4563 btrfs_err(info, "invalid chunk type 0x%llx requested",
4568 /* we don't want a chunk larger than 10% of writeable space */
4569 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4572 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4577 cur = fs_devices->alloc_list.next;
4580 * in the first pass through the devices list, we gather information
4581 * about the available holes on each device.
4584 while (cur != &fs_devices->alloc_list) {
4585 struct btrfs_device *device;
4589 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4593 if (!device->writeable) {
4595 "BTRFS: read-only device in alloc_list\n");
4599 if (!device->in_fs_metadata ||
4600 device->is_tgtdev_for_dev_replace)
4603 if (device->total_bytes > device->bytes_used)
4604 total_avail = device->total_bytes - device->bytes_used;
4608 /* If there is no space on this device, skip it. */
4609 if (total_avail == 0)
4612 ret = find_free_dev_extent(trans, device,
4613 max_stripe_size * dev_stripes,
4614 &dev_offset, &max_avail);
4615 if (ret && ret != -ENOSPC)
4619 max_avail = max_stripe_size * dev_stripes;
4621 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4624 if (ndevs == fs_devices->rw_devices) {
4625 WARN(1, "%s: found more than %llu devices\n",
4626 __func__, fs_devices->rw_devices);
4629 devices_info[ndevs].dev_offset = dev_offset;
4630 devices_info[ndevs].max_avail = max_avail;
4631 devices_info[ndevs].total_avail = total_avail;
4632 devices_info[ndevs].dev = device;
4637 * now sort the devices by hole size / available space
4639 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4640 btrfs_cmp_device_info, NULL);
4642 /* round down to number of usable stripes */
4643 ndevs -= ndevs % devs_increment;
4645 if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4650 if (devs_max && ndevs > devs_max)
4653 * The primary goal is to maximize the number of stripes, so use as
4654 * many devices as possible, even if the stripes are not maximum sized.
4656 * The DUP profile stores more than one stripe per device, the
4657 * max_avail is the total size so we have to adjust.
4659 stripe_size = div_u64(devices_info[ndevs - 1].max_avail, dev_stripes);
4660 num_stripes = ndevs * dev_stripes;
4663 * this will have to be fixed for RAID1 and RAID10 over
4666 data_stripes = num_stripes / ncopies;
4668 if (type & BTRFS_BLOCK_GROUP_RAID5) {
4669 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
4670 btrfs_super_stripesize(info->super_copy));
4671 data_stripes = num_stripes - 1;
4673 if (type & BTRFS_BLOCK_GROUP_RAID6) {
4674 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
4675 btrfs_super_stripesize(info->super_copy));
4676 data_stripes = num_stripes - 2;
4680 * Use the number of data stripes to figure out how big this chunk
4681 * is really going to be in terms of logical address space,
4682 * and compare that answer with the max chunk size
4684 if (stripe_size * data_stripes > max_chunk_size) {
4685 u64 mask = (1ULL << 24) - 1;
4687 stripe_size = div_u64(max_chunk_size, data_stripes);
4689 /* bump the answer up to a 16MB boundary */
4690 stripe_size = (stripe_size + mask) & ~mask;
4692 /* but don't go higher than the limits we found
4693 * while searching for free extents
4695 if (stripe_size > devices_info[ndevs-1].max_avail)
4696 stripe_size = devices_info[ndevs-1].max_avail;
4699 /* align to BTRFS_STRIPE_LEN */
4700 stripe_size = div_u64(stripe_size, raid_stripe_len);
4701 stripe_size *= raid_stripe_len;
4703 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4708 map->num_stripes = num_stripes;
4710 for (i = 0; i < ndevs; ++i) {
4711 for (j = 0; j < dev_stripes; ++j) {
4712 int s = i * dev_stripes + j;
4713 map->stripes[s].dev = devices_info[i].dev;
4714 map->stripes[s].physical = devices_info[i].dev_offset +
4718 map->sector_size = extent_root->sectorsize;
4719 map->stripe_len = raid_stripe_len;
4720 map->io_align = raid_stripe_len;
4721 map->io_width = raid_stripe_len;
4723 map->sub_stripes = sub_stripes;
4725 num_bytes = stripe_size * data_stripes;
4727 trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
4729 em = alloc_extent_map();
4735 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4736 em->map_lookup = map;
4738 em->len = num_bytes;
4739 em->block_start = 0;
4740 em->block_len = em->len;
4741 em->orig_block_len = stripe_size;
4743 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4744 write_lock(&em_tree->lock);
4745 ret = add_extent_mapping(em_tree, em, 0);
4747 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4748 atomic_inc(&em->refs);
4750 write_unlock(&em_tree->lock);
4752 free_extent_map(em);
4756 ret = btrfs_make_block_group(trans, extent_root, 0, type,
4757 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4760 goto error_del_extent;
4762 for (i = 0; i < map->num_stripes; i++) {
4763 num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4764 btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4765 map->stripes[i].dev->has_pending_chunks = true;
4768 spin_lock(&extent_root->fs_info->free_chunk_lock);
4769 extent_root->fs_info->free_chunk_space -= (stripe_size *
4771 spin_unlock(&extent_root->fs_info->free_chunk_lock);
4773 free_extent_map(em);
4774 check_raid56_incompat_flag(extent_root->fs_info, type);
4776 kfree(devices_info);
4780 write_lock(&em_tree->lock);
4781 remove_extent_mapping(em_tree, em);
4782 write_unlock(&em_tree->lock);
4784 /* One for our allocation */
4785 free_extent_map(em);
4786 /* One for the tree reference */
4787 free_extent_map(em);
4788 /* One for the pending_chunks list reference */
4789 free_extent_map(em);
4791 kfree(devices_info);
4795 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4796 struct btrfs_root *extent_root,
4797 u64 chunk_offset, u64 chunk_size)
4799 struct btrfs_key key;
4800 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4801 struct btrfs_device *device;
4802 struct btrfs_chunk *chunk;
4803 struct btrfs_stripe *stripe;
4804 struct extent_map_tree *em_tree;
4805 struct extent_map *em;
4806 struct map_lookup *map;
4813 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4814 read_lock(&em_tree->lock);
4815 em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size);
4816 read_unlock(&em_tree->lock);
4819 btrfs_crit(extent_root->fs_info, "unable to find logical "
4820 "%Lu len %Lu", chunk_offset, chunk_size);
4824 if (em->start != chunk_offset || em->len != chunk_size) {
4825 btrfs_crit(extent_root->fs_info, "found a bad mapping, wanted"
4826 " %Lu-%Lu, found %Lu-%Lu", chunk_offset,
4827 chunk_size, em->start, em->len);
4828 free_extent_map(em);
4832 map = em->map_lookup;
4833 item_size = btrfs_chunk_item_size(map->num_stripes);
4834 stripe_size = em->orig_block_len;
4836 chunk = kzalloc(item_size, GFP_NOFS);
4842 for (i = 0; i < map->num_stripes; i++) {
4843 device = map->stripes[i].dev;
4844 dev_offset = map->stripes[i].physical;
4846 ret = btrfs_update_device(trans, device);
4849 ret = btrfs_alloc_dev_extent(trans, device,
4850 chunk_root->root_key.objectid,
4851 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4852 chunk_offset, dev_offset,
4858 stripe = &chunk->stripe;
4859 for (i = 0; i < map->num_stripes; i++) {
4860 device = map->stripes[i].dev;
4861 dev_offset = map->stripes[i].physical;
4863 btrfs_set_stack_stripe_devid(stripe, device->devid);
4864 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4865 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4869 btrfs_set_stack_chunk_length(chunk, chunk_size);
4870 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4871 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4872 btrfs_set_stack_chunk_type(chunk, map->type);
4873 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4874 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4875 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4876 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4877 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4879 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4880 key.type = BTRFS_CHUNK_ITEM_KEY;
4881 key.offset = chunk_offset;
4883 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4884 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4886 * TODO: Cleanup of inserted chunk root in case of
4889 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
4895 free_extent_map(em);
4900 * Chunk allocation falls into two parts. The first part does works
4901 * that make the new allocated chunk useable, but not do any operation
4902 * that modifies the chunk tree. The second part does the works that
4903 * require modifying the chunk tree. This division is important for the
4904 * bootstrap process of adding storage to a seed btrfs.
4906 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4907 struct btrfs_root *extent_root, u64 type)
4911 ASSERT(mutex_is_locked(&extent_root->fs_info->chunk_mutex));
4912 chunk_offset = find_next_chunk(extent_root->fs_info);
4913 return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type);
4916 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4917 struct btrfs_root *root,
4918 struct btrfs_device *device)
4921 u64 sys_chunk_offset;
4923 struct btrfs_fs_info *fs_info = root->fs_info;
4924 struct btrfs_root *extent_root = fs_info->extent_root;
4927 chunk_offset = find_next_chunk(fs_info);
4928 alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
4929 ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset,
4934 sys_chunk_offset = find_next_chunk(root->fs_info);
4935 alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
4936 ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset,
4941 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
4945 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
4946 BTRFS_BLOCK_GROUP_RAID10 |
4947 BTRFS_BLOCK_GROUP_RAID5)) {
4949 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
4958 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
4960 struct extent_map *em;
4961 struct map_lookup *map;
4962 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4967 read_lock(&map_tree->map_tree.lock);
4968 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
4969 read_unlock(&map_tree->map_tree.lock);
4973 map = em->map_lookup;
4974 for (i = 0; i < map->num_stripes; i++) {
4975 if (map->stripes[i].dev->missing) {
4980 if (!map->stripes[i].dev->writeable) {
4987 * If the number of missing devices is larger than max errors,
4988 * we can not write the data into that chunk successfully, so
4991 if (miss_ndevs > btrfs_chunk_max_errors(map))
4994 free_extent_map(em);
4998 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
5000 extent_map_tree_init(&tree->map_tree);
5003 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
5005 struct extent_map *em;
5008 write_lock(&tree->map_tree.lock);
5009 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
5011 remove_extent_mapping(&tree->map_tree, em);
5012 write_unlock(&tree->map_tree.lock);
5016 free_extent_map(em);
5017 /* once for the tree */
5018 free_extent_map(em);
5022 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5024 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5025 struct extent_map *em;
5026 struct map_lookup *map;
5027 struct extent_map_tree *em_tree = &map_tree->map_tree;
5030 read_lock(&em_tree->lock);
5031 em = lookup_extent_mapping(em_tree, logical, len);
5032 read_unlock(&em_tree->lock);
5035 * We could return errors for these cases, but that could get ugly and
5036 * we'd probably do the same thing which is just not do anything else
5037 * and exit, so return 1 so the callers don't try to use other copies.
5040 btrfs_crit(fs_info, "No mapping for %Lu-%Lu", logical,
5045 if (em->start > logical || em->start + em->len < logical) {
5046 btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got "
5047 "%Lu-%Lu", logical, logical+len, em->start,
5048 em->start + em->len);
5049 free_extent_map(em);
5053 map = em->map_lookup;
5054 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
5055 ret = map->num_stripes;
5056 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5057 ret = map->sub_stripes;
5058 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5060 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5062 * There could be two corrupted data stripes, we need
5063 * to loop retry in order to rebuild the correct data.
5065 * Fail a stripe at a time on every retry except the
5066 * stripe under reconstruction.
5068 ret = map->num_stripes;
5071 free_extent_map(em);
5073 btrfs_dev_replace_lock(&fs_info->dev_replace);
5074 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
5076 btrfs_dev_replace_unlock(&fs_info->dev_replace);
5081 unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
5082 struct btrfs_mapping_tree *map_tree,
5085 struct extent_map *em;
5086 struct map_lookup *map;
5087 struct extent_map_tree *em_tree = &map_tree->map_tree;
5088 unsigned long len = root->sectorsize;
5090 read_lock(&em_tree->lock);
5091 em = lookup_extent_mapping(em_tree, logical, len);
5092 read_unlock(&em_tree->lock);
5095 BUG_ON(em->start > logical || em->start + em->len < logical);
5096 map = em->map_lookup;
5097 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5098 len = map->stripe_len * nr_data_stripes(map);
5099 free_extent_map(em);
5103 int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
5104 u64 logical, u64 len, int mirror_num)
5106 struct extent_map *em;
5107 struct map_lookup *map;
5108 struct extent_map_tree *em_tree = &map_tree->map_tree;
5111 read_lock(&em_tree->lock);
5112 em = lookup_extent_mapping(em_tree, logical, len);
5113 read_unlock(&em_tree->lock);
5116 BUG_ON(em->start > logical || em->start + em->len < logical);
5117 map = em->map_lookup;
5118 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5120 free_extent_map(em);
5124 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5125 struct map_lookup *map, int first, int num,
5126 int optimal, int dev_replace_is_ongoing)
5130 struct btrfs_device *srcdev;
5132 if (dev_replace_is_ongoing &&
5133 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5134 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5135 srcdev = fs_info->dev_replace.srcdev;
5140 * try to avoid the drive that is the source drive for a
5141 * dev-replace procedure, only choose it if no other non-missing
5142 * mirror is available
5144 for (tolerance = 0; tolerance < 2; tolerance++) {
5145 if (map->stripes[optimal].dev->bdev &&
5146 (tolerance || map->stripes[optimal].dev != srcdev))
5148 for (i = first; i < first + num; i++) {
5149 if (map->stripes[i].dev->bdev &&
5150 (tolerance || map->stripes[i].dev != srcdev))
5155 /* we couldn't find one that doesn't fail. Just return something
5156 * and the io error handling code will clean up eventually
5161 static inline int parity_smaller(u64 a, u64 b)
5166 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5167 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5169 struct btrfs_bio_stripe s;
5176 for (i = 0; i < num_stripes - 1; i++) {
5177 if (parity_smaller(bbio->raid_map[i],
5178 bbio->raid_map[i+1])) {
5179 s = bbio->stripes[i];
5180 l = bbio->raid_map[i];
5181 bbio->stripes[i] = bbio->stripes[i+1];
5182 bbio->raid_map[i] = bbio->raid_map[i+1];
5183 bbio->stripes[i+1] = s;
5184 bbio->raid_map[i+1] = l;
5192 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5194 struct btrfs_bio *bbio = kzalloc(
5195 /* the size of the btrfs_bio */
5196 sizeof(struct btrfs_bio) +
5197 /* plus the variable array for the stripes */
5198 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5199 /* plus the variable array for the tgt dev */
5200 sizeof(int) * (real_stripes) +
5202 * plus the raid_map, which includes both the tgt dev
5205 sizeof(u64) * (total_stripes),
5206 GFP_NOFS|__GFP_NOFAIL);
5208 atomic_set(&bbio->error, 0);
5209 atomic_set(&bbio->refs, 1);
5214 void btrfs_get_bbio(struct btrfs_bio *bbio)
5216 WARN_ON(!atomic_read(&bbio->refs));
5217 atomic_inc(&bbio->refs);
5220 void btrfs_put_bbio(struct btrfs_bio *bbio)
5224 if (atomic_dec_and_test(&bbio->refs))
5228 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5229 u64 logical, u64 *length,
5230 struct btrfs_bio **bbio_ret,
5231 int mirror_num, int need_raid_map)
5233 struct extent_map *em;
5234 struct map_lookup *map;
5235 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5236 struct extent_map_tree *em_tree = &map_tree->map_tree;
5239 u64 stripe_end_offset;
5249 int tgtdev_indexes = 0;
5250 struct btrfs_bio *bbio = NULL;
5251 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
5252 int dev_replace_is_ongoing = 0;
5253 int num_alloc_stripes;
5254 int patch_the_first_stripe_for_dev_replace = 0;
5255 u64 physical_to_patch_in_first_stripe = 0;
5256 u64 raid56_full_stripe_start = (u64)-1;
5258 read_lock(&em_tree->lock);
5259 em = lookup_extent_mapping(em_tree, logical, *length);
5260 read_unlock(&em_tree->lock);
5263 btrfs_crit(fs_info, "unable to find logical %llu len %llu",
5268 if (em->start > logical || em->start + em->len < logical) {
5269 btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, "
5270 "found %Lu-%Lu", logical, em->start,
5271 em->start + em->len);
5272 free_extent_map(em);
5276 map = em->map_lookup;
5277 offset = logical - em->start;
5279 stripe_len = map->stripe_len;
5282 * stripe_nr counts the total number of stripes we have to stride
5283 * to get to this block
5285 stripe_nr = div64_u64(stripe_nr, stripe_len);
5287 stripe_offset = stripe_nr * stripe_len;
5288 BUG_ON(offset < stripe_offset);
5290 /* stripe_offset is the offset of this block in its stripe*/
5291 stripe_offset = offset - stripe_offset;
5293 /* if we're here for raid56, we need to know the stripe aligned start */
5294 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5295 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5296 raid56_full_stripe_start = offset;
5298 /* allow a write of a full stripe, but make sure we don't
5299 * allow straddling of stripes
5301 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5303 raid56_full_stripe_start *= full_stripe_len;
5306 if (rw & REQ_DISCARD) {
5307 /* we don't discard raid56 yet */
5308 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5312 *length = min_t(u64, em->len - offset, *length);
5313 } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5315 /* For writes to RAID[56], allow a full stripeset across all disks.
5316 For other RAID types and for RAID[56] reads, just allow a single
5317 stripe (on a single disk). */
5318 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
5320 max_len = stripe_len * nr_data_stripes(map) -
5321 (offset - raid56_full_stripe_start);
5323 /* we limit the length of each bio to what fits in a stripe */
5324 max_len = stripe_len - stripe_offset;
5326 *length = min_t(u64, em->len - offset, max_len);
5328 *length = em->len - offset;
5331 /* This is for when we're called from btrfs_merge_bio_hook() and all
5332 it cares about is the length */
5336 btrfs_dev_replace_lock(dev_replace);
5337 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5338 if (!dev_replace_is_ongoing)
5339 btrfs_dev_replace_unlock(dev_replace);
5341 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
5342 !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) &&
5343 dev_replace->tgtdev != NULL) {
5345 * in dev-replace case, for repair case (that's the only
5346 * case where the mirror is selected explicitly when
5347 * calling btrfs_map_block), blocks left of the left cursor
5348 * can also be read from the target drive.
5349 * For REQ_GET_READ_MIRRORS, the target drive is added as
5350 * the last one to the array of stripes. For READ, it also
5351 * needs to be supported using the same mirror number.
5352 * If the requested block is not left of the left cursor,
5353 * EIO is returned. This can happen because btrfs_num_copies()
5354 * returns one more in the dev-replace case.
5356 u64 tmp_length = *length;
5357 struct btrfs_bio *tmp_bbio = NULL;
5358 int tmp_num_stripes;
5359 u64 srcdev_devid = dev_replace->srcdev->devid;
5360 int index_srcdev = 0;
5362 u64 physical_of_found = 0;
5364 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
5365 logical, &tmp_length, &tmp_bbio, 0, 0);
5367 WARN_ON(tmp_bbio != NULL);
5371 tmp_num_stripes = tmp_bbio->num_stripes;
5372 if (mirror_num > tmp_num_stripes) {
5374 * REQ_GET_READ_MIRRORS does not contain this
5375 * mirror, that means that the requested area
5376 * is not left of the left cursor
5379 btrfs_put_bbio(tmp_bbio);
5384 * process the rest of the function using the mirror_num
5385 * of the source drive. Therefore look it up first.
5386 * At the end, patch the device pointer to the one of the
5389 for (i = 0; i < tmp_num_stripes; i++) {
5390 if (tmp_bbio->stripes[i].dev->devid == srcdev_devid) {
5392 * In case of DUP, in order to keep it
5393 * simple, only add the mirror with the
5394 * lowest physical address
5397 physical_of_found <=
5398 tmp_bbio->stripes[i].physical)
5403 tmp_bbio->stripes[i].physical;
5408 mirror_num = index_srcdev + 1;
5409 patch_the_first_stripe_for_dev_replace = 1;
5410 physical_to_patch_in_first_stripe = physical_of_found;
5414 btrfs_put_bbio(tmp_bbio);
5418 btrfs_put_bbio(tmp_bbio);
5419 } else if (mirror_num > map->num_stripes) {
5425 stripe_nr_orig = stripe_nr;
5426 stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
5427 stripe_nr_end = div_u64(stripe_nr_end, map->stripe_len);
5428 stripe_end_offset = stripe_nr_end * map->stripe_len -
5431 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5432 if (rw & REQ_DISCARD)
5433 num_stripes = min_t(u64, map->num_stripes,
5434 stripe_nr_end - stripe_nr_orig);
5435 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5437 if (!(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)))
5439 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
5440 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))
5441 num_stripes = map->num_stripes;
5442 else if (mirror_num)
5443 stripe_index = mirror_num - 1;
5445 stripe_index = find_live_mirror(fs_info, map, 0,
5447 current->pid % map->num_stripes,
5448 dev_replace_is_ongoing);
5449 mirror_num = stripe_index + 1;
5452 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
5453 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
5454 num_stripes = map->num_stripes;
5455 } else if (mirror_num) {
5456 stripe_index = mirror_num - 1;
5461 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5462 u32 factor = map->num_stripes / map->sub_stripes;
5464 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5465 stripe_index *= map->sub_stripes;
5467 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
5468 num_stripes = map->sub_stripes;
5469 else if (rw & REQ_DISCARD)
5470 num_stripes = min_t(u64, map->sub_stripes *
5471 (stripe_nr_end - stripe_nr_orig),
5473 else if (mirror_num)
5474 stripe_index += mirror_num - 1;
5476 int old_stripe_index = stripe_index;
5477 stripe_index = find_live_mirror(fs_info, map,
5479 map->sub_stripes, stripe_index +
5480 current->pid % map->sub_stripes,
5481 dev_replace_is_ongoing);
5482 mirror_num = stripe_index - old_stripe_index + 1;
5485 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5486 if (need_raid_map &&
5487 ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) ||
5489 /* push stripe_nr back to the start of the full stripe */
5490 stripe_nr = div_u64(raid56_full_stripe_start,
5491 stripe_len * nr_data_stripes(map));
5493 /* RAID[56] write or recovery. Return all stripes */
5494 num_stripes = map->num_stripes;
5495 max_errors = nr_parity_stripes(map);
5497 *length = map->stripe_len;
5502 * Mirror #0 or #1 means the original data block.
5503 * Mirror #2 is RAID5 parity block.
5504 * Mirror #3 is RAID6 Q block.
5506 stripe_nr = div_u64_rem(stripe_nr,
5507 nr_data_stripes(map), &stripe_index);
5509 stripe_index = nr_data_stripes(map) +
5512 /* We distribute the parity blocks across stripes */
5513 div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
5515 if (!(rw & (REQ_WRITE | REQ_DISCARD |
5516 REQ_GET_READ_MIRRORS)) && mirror_num <= 1)
5521 * after this, stripe_nr is the number of stripes on this
5522 * device we have to walk to find the data, and stripe_index is
5523 * the number of our device in the stripe array
5525 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5527 mirror_num = stripe_index + 1;
5529 BUG_ON(stripe_index >= map->num_stripes);
5531 num_alloc_stripes = num_stripes;
5532 if (dev_replace_is_ongoing) {
5533 if (rw & (REQ_WRITE | REQ_DISCARD))
5534 num_alloc_stripes <<= 1;
5535 if (rw & REQ_GET_READ_MIRRORS)
5536 num_alloc_stripes++;
5537 tgtdev_indexes = num_stripes;
5540 bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
5545 if (dev_replace_is_ongoing)
5546 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
5548 /* build raid_map */
5549 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK &&
5550 need_raid_map && ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) ||
5555 bbio->raid_map = (u64 *)((void *)bbio->stripes +
5556 sizeof(struct btrfs_bio_stripe) *
5558 sizeof(int) * tgtdev_indexes);
5560 /* Work out the disk rotation on this stripe-set */
5561 div_u64_rem(stripe_nr, num_stripes, &rot);
5563 /* Fill in the logical address of each stripe */
5564 tmp = stripe_nr * nr_data_stripes(map);
5565 for (i = 0; i < nr_data_stripes(map); i++)
5566 bbio->raid_map[(i+rot) % num_stripes] =
5567 em->start + (tmp + i) * map->stripe_len;
5569 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5570 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5571 bbio->raid_map[(i+rot+1) % num_stripes] =
5575 if (rw & REQ_DISCARD) {
5577 u32 sub_stripes = 0;
5578 u64 stripes_per_dev = 0;
5579 u32 remaining_stripes = 0;
5580 u32 last_stripe = 0;
5583 (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
5584 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5587 sub_stripes = map->sub_stripes;
5589 factor = map->num_stripes / sub_stripes;
5590 stripes_per_dev = div_u64_rem(stripe_nr_end -
5593 &remaining_stripes);
5594 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5595 last_stripe *= sub_stripes;
5598 for (i = 0; i < num_stripes; i++) {
5599 bbio->stripes[i].physical =
5600 map->stripes[stripe_index].physical +
5601 stripe_offset + stripe_nr * map->stripe_len;
5602 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5604 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5605 BTRFS_BLOCK_GROUP_RAID10)) {
5606 bbio->stripes[i].length = stripes_per_dev *
5609 if (i / sub_stripes < remaining_stripes)
5610 bbio->stripes[i].length +=
5614 * Special for the first stripe and
5617 * |-------|...|-------|
5621 if (i < sub_stripes)
5622 bbio->stripes[i].length -=
5625 if (stripe_index >= last_stripe &&
5626 stripe_index <= (last_stripe +
5628 bbio->stripes[i].length -=
5631 if (i == sub_stripes - 1)
5634 bbio->stripes[i].length = *length;
5637 if (stripe_index == map->num_stripes) {
5638 /* This could only happen for RAID0/10 */
5644 for (i = 0; i < num_stripes; i++) {
5645 bbio->stripes[i].physical =
5646 map->stripes[stripe_index].physical +
5648 stripe_nr * map->stripe_len;
5649 bbio->stripes[i].dev =
5650 map->stripes[stripe_index].dev;
5655 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
5656 max_errors = btrfs_chunk_max_errors(map);
5659 sort_parity_stripes(bbio, num_stripes);
5662 if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) &&
5663 dev_replace->tgtdev != NULL) {
5664 int index_where_to_add;
5665 u64 srcdev_devid = dev_replace->srcdev->devid;
5668 * duplicate the write operations while the dev replace
5669 * procedure is running. Since the copying of the old disk
5670 * to the new disk takes place at run time while the
5671 * filesystem is mounted writable, the regular write
5672 * operations to the old disk have to be duplicated to go
5673 * to the new disk as well.
5674 * Note that device->missing is handled by the caller, and
5675 * that the write to the old disk is already set up in the
5678 index_where_to_add = num_stripes;
5679 for (i = 0; i < num_stripes; i++) {
5680 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5681 /* write to new disk, too */
5682 struct btrfs_bio_stripe *new =
5683 bbio->stripes + index_where_to_add;
5684 struct btrfs_bio_stripe *old =
5687 new->physical = old->physical;
5688 new->length = old->length;
5689 new->dev = dev_replace->tgtdev;
5690 bbio->tgtdev_map[i] = index_where_to_add;
5691 index_where_to_add++;
5696 num_stripes = index_where_to_add;
5697 } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) &&
5698 dev_replace->tgtdev != NULL) {
5699 u64 srcdev_devid = dev_replace->srcdev->devid;
5700 int index_srcdev = 0;
5702 u64 physical_of_found = 0;
5705 * During the dev-replace procedure, the target drive can
5706 * also be used to read data in case it is needed to repair
5707 * a corrupt block elsewhere. This is possible if the
5708 * requested area is left of the left cursor. In this area,
5709 * the target drive is a full copy of the source drive.
5711 for (i = 0; i < num_stripes; i++) {
5712 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5714 * In case of DUP, in order to keep it
5715 * simple, only add the mirror with the
5716 * lowest physical address
5719 physical_of_found <=
5720 bbio->stripes[i].physical)
5724 physical_of_found = bbio->stripes[i].physical;
5728 if (physical_of_found + map->stripe_len <=
5729 dev_replace->cursor_left) {
5730 struct btrfs_bio_stripe *tgtdev_stripe =
5731 bbio->stripes + num_stripes;
5733 tgtdev_stripe->physical = physical_of_found;
5734 tgtdev_stripe->length =
5735 bbio->stripes[index_srcdev].length;
5736 tgtdev_stripe->dev = dev_replace->tgtdev;
5737 bbio->tgtdev_map[index_srcdev] = num_stripes;
5746 bbio->map_type = map->type;
5747 bbio->num_stripes = num_stripes;
5748 bbio->max_errors = max_errors;
5749 bbio->mirror_num = mirror_num;
5750 bbio->num_tgtdevs = tgtdev_indexes;
5753 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5754 * mirror_num == num_stripes + 1 && dev_replace target drive is
5755 * available as a mirror
5757 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5758 WARN_ON(num_stripes > 1);
5759 bbio->stripes[0].dev = dev_replace->tgtdev;
5760 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5761 bbio->mirror_num = map->num_stripes + 1;
5764 if (dev_replace_is_ongoing)
5765 btrfs_dev_replace_unlock(dev_replace);
5766 free_extent_map(em);
5770 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5771 u64 logical, u64 *length,
5772 struct btrfs_bio **bbio_ret, int mirror_num)
5774 return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5778 /* For Scrub/replace */
5779 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, int rw,
5780 u64 logical, u64 *length,
5781 struct btrfs_bio **bbio_ret, int mirror_num,
5784 return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5785 mirror_num, need_raid_map);
5788 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
5789 u64 chunk_start, u64 physical, u64 devid,
5790 u64 **logical, int *naddrs, int *stripe_len)
5792 struct extent_map_tree *em_tree = &map_tree->map_tree;
5793 struct extent_map *em;
5794 struct map_lookup *map;
5802 read_lock(&em_tree->lock);
5803 em = lookup_extent_mapping(em_tree, chunk_start, 1);
5804 read_unlock(&em_tree->lock);
5807 printk(KERN_ERR "BTRFS: couldn't find em for chunk %Lu\n",
5812 if (em->start != chunk_start) {
5813 printk(KERN_ERR "BTRFS: bad chunk start, em=%Lu, wanted=%Lu\n",
5814 em->start, chunk_start);
5815 free_extent_map(em);
5818 map = em->map_lookup;
5821 rmap_len = map->stripe_len;
5823 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5824 length = div_u64(length, map->num_stripes / map->sub_stripes);
5825 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5826 length = div_u64(length, map->num_stripes);
5827 else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5828 length = div_u64(length, nr_data_stripes(map));
5829 rmap_len = map->stripe_len * nr_data_stripes(map);
5832 buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
5833 BUG_ON(!buf); /* -ENOMEM */
5835 for (i = 0; i < map->num_stripes; i++) {
5836 if (devid && map->stripes[i].dev->devid != devid)
5838 if (map->stripes[i].physical > physical ||
5839 map->stripes[i].physical + length <= physical)
5842 stripe_nr = physical - map->stripes[i].physical;
5843 stripe_nr = div_u64(stripe_nr, map->stripe_len);
5845 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5846 stripe_nr = stripe_nr * map->num_stripes + i;
5847 stripe_nr = div_u64(stripe_nr, map->sub_stripes);
5848 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5849 stripe_nr = stripe_nr * map->num_stripes + i;
5850 } /* else if RAID[56], multiply by nr_data_stripes().
5851 * Alternatively, just use rmap_len below instead of
5852 * map->stripe_len */
5854 bytenr = chunk_start + stripe_nr * rmap_len;
5855 WARN_ON(nr >= map->num_stripes);
5856 for (j = 0; j < nr; j++) {
5857 if (buf[j] == bytenr)
5861 WARN_ON(nr >= map->num_stripes);
5868 *stripe_len = rmap_len;
5870 free_extent_map(em);
5874 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
5876 bio->bi_private = bbio->private;
5877 bio->bi_end_io = bbio->end_io;
5880 btrfs_put_bbio(bbio);
5883 static void btrfs_end_bio(struct bio *bio)
5885 struct btrfs_bio *bbio = bio->bi_private;
5886 int is_orig_bio = 0;
5888 if (bio->bi_error) {
5889 atomic_inc(&bbio->error);
5890 if (bio->bi_error == -EIO || bio->bi_error == -EREMOTEIO) {
5891 unsigned int stripe_index =
5892 btrfs_io_bio(bio)->stripe_index;
5893 struct btrfs_device *dev;
5895 BUG_ON(stripe_index >= bbio->num_stripes);
5896 dev = bbio->stripes[stripe_index].dev;
5898 if (bio->bi_rw & WRITE)
5899 btrfs_dev_stat_inc(dev,
5900 BTRFS_DEV_STAT_WRITE_ERRS);
5902 btrfs_dev_stat_inc(dev,
5903 BTRFS_DEV_STAT_READ_ERRS);
5904 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
5905 btrfs_dev_stat_inc(dev,
5906 BTRFS_DEV_STAT_FLUSH_ERRS);
5907 btrfs_dev_stat_print_on_error(dev);
5912 if (bio == bbio->orig_bio)
5915 btrfs_bio_counter_dec(bbio->fs_info);
5917 if (atomic_dec_and_test(&bbio->stripes_pending)) {
5920 bio = bbio->orig_bio;
5923 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5924 /* only send an error to the higher layers if it is
5925 * beyond the tolerance of the btrfs bio
5927 if (atomic_read(&bbio->error) > bbio->max_errors) {
5928 bio->bi_error = -EIO;
5931 * this bio is actually up to date, we didn't
5932 * go over the max number of errors
5937 btrfs_end_bbio(bbio, bio);
5938 } else if (!is_orig_bio) {
5944 * see run_scheduled_bios for a description of why bios are collected for
5947 * This will add one bio to the pending list for a device and make sure
5948 * the work struct is scheduled.
5950 static noinline void btrfs_schedule_bio(struct btrfs_root *root,
5951 struct btrfs_device *device,
5952 int rw, struct bio *bio)
5954 int should_queue = 1;
5955 struct btrfs_pending_bios *pending_bios;
5957 if (device->missing || !device->bdev) {
5962 /* don't bother with additional async steps for reads, right now */
5963 if (!(rw & REQ_WRITE)) {
5965 btrfsic_submit_bio(rw, bio);
5971 * nr_async_bios allows us to reliably return congestion to the
5972 * higher layers. Otherwise, the async bio makes it appear we have
5973 * made progress against dirty pages when we've really just put it
5974 * on a queue for later
5976 atomic_inc(&root->fs_info->nr_async_bios);
5977 WARN_ON(bio->bi_next);
5978 bio->bi_next = NULL;
5981 spin_lock(&device->io_lock);
5982 if (bio->bi_rw & REQ_SYNC)
5983 pending_bios = &device->pending_sync_bios;
5985 pending_bios = &device->pending_bios;
5987 if (pending_bios->tail)
5988 pending_bios->tail->bi_next = bio;
5990 pending_bios->tail = bio;
5991 if (!pending_bios->head)
5992 pending_bios->head = bio;
5993 if (device->running_pending)
5996 spin_unlock(&device->io_lock);
5999 btrfs_queue_work(root->fs_info->submit_workers,
6003 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
6004 struct bio *bio, u64 physical, int dev_nr,
6007 struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
6009 bio->bi_private = bbio;
6010 btrfs_io_bio(bio)->stripe_index = dev_nr;
6011 bio->bi_end_io = btrfs_end_bio;
6012 bio->bi_iter.bi_sector = physical >> 9;
6015 struct rcu_string *name;
6018 name = rcu_dereference(dev->name);
6019 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
6020 "(%s id %llu), size=%u\n", rw,
6021 (u64)bio->bi_iter.bi_sector, (u_long)dev->bdev->bd_dev,
6022 name->str, dev->devid, bio->bi_iter.bi_size);
6026 bio->bi_bdev = dev->bdev;
6028 btrfs_bio_counter_inc_noblocked(root->fs_info);
6031 btrfs_schedule_bio(root, dev, rw, bio);
6033 btrfsic_submit_bio(rw, bio);
6036 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6038 atomic_inc(&bbio->error);
6039 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6040 /* Shoud be the original bio. */
6041 WARN_ON(bio != bbio->orig_bio);
6043 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6044 bio->bi_iter.bi_sector = logical >> 9;
6045 bio->bi_error = -EIO;
6046 btrfs_end_bbio(bbio, bio);
6050 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
6051 int mirror_num, int async_submit)
6053 struct btrfs_device *dev;
6054 struct bio *first_bio = bio;
6055 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
6061 struct btrfs_bio *bbio = NULL;
6063 length = bio->bi_iter.bi_size;
6064 map_length = length;
6066 btrfs_bio_counter_inc_blocked(root->fs_info);
6067 ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
6070 btrfs_bio_counter_dec(root->fs_info);
6074 total_devs = bbio->num_stripes;
6075 bbio->orig_bio = first_bio;
6076 bbio->private = first_bio->bi_private;
6077 bbio->end_io = first_bio->bi_end_io;
6078 bbio->fs_info = root->fs_info;
6079 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6081 if (bbio->raid_map) {
6082 /* In this case, map_length has been set to the length of
6083 a single stripe; not the whole write */
6085 ret = raid56_parity_write(root, bio, bbio, map_length);
6087 ret = raid56_parity_recover(root, bio, bbio, map_length,
6091 btrfs_bio_counter_dec(root->fs_info);
6095 if (map_length < length) {
6096 btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu",
6097 logical, length, map_length);
6101 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6102 dev = bbio->stripes[dev_nr].dev;
6103 if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
6104 bbio_error(bbio, first_bio, logical);
6108 if (dev_nr < total_devs - 1) {
6109 bio = btrfs_bio_clone(first_bio, GFP_NOFS);
6110 BUG_ON(!bio); /* -ENOMEM */
6114 submit_stripe_bio(root, bbio, bio,
6115 bbio->stripes[dev_nr].physical, dev_nr, rw,
6118 btrfs_bio_counter_dec(root->fs_info);
6122 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
6125 struct btrfs_device *device;
6126 struct btrfs_fs_devices *cur_devices;
6128 cur_devices = fs_info->fs_devices;
6129 while (cur_devices) {
6131 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
6132 device = __find_device(&cur_devices->devices,
6137 cur_devices = cur_devices->seed;
6142 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
6143 struct btrfs_fs_devices *fs_devices,
6144 u64 devid, u8 *dev_uuid)
6146 struct btrfs_device *device;
6148 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6152 list_add(&device->dev_list, &fs_devices->devices);
6153 device->fs_devices = fs_devices;
6154 fs_devices->num_devices++;
6156 device->missing = 1;
6157 fs_devices->missing_devices++;
6163 * btrfs_alloc_device - allocate struct btrfs_device
6164 * @fs_info: used only for generating a new devid, can be NULL if
6165 * devid is provided (i.e. @devid != NULL).
6166 * @devid: a pointer to devid for this device. If NULL a new devid
6168 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6171 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6172 * on error. Returned struct is not linked onto any lists and can be
6173 * destroyed with kfree() right away.
6175 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6179 struct btrfs_device *dev;
6182 if (WARN_ON(!devid && !fs_info))
6183 return ERR_PTR(-EINVAL);
6185 dev = __alloc_device();
6194 ret = find_next_devid(fs_info, &tmp);
6197 return ERR_PTR(ret);
6203 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6205 generate_random_uuid(dev->uuid);
6207 btrfs_init_work(&dev->work, btrfs_submit_helper,
6208 pending_bios_fn, NULL, NULL);
6213 /* Return -EIO if any error, otherwise return 0. */
6214 static int btrfs_check_chunk_valid(struct btrfs_root *root,
6215 struct extent_buffer *leaf,
6216 struct btrfs_chunk *chunk, u64 logical)
6226 length = btrfs_chunk_length(leaf, chunk);
6227 stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6228 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6229 sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6230 type = btrfs_chunk_type(leaf, chunk);
6233 btrfs_err(root->fs_info, "invalid chunk num_stripes: %u",
6237 if (!IS_ALIGNED(logical, root->sectorsize)) {
6238 btrfs_err(root->fs_info,
6239 "invalid chunk logical %llu", logical);
6242 if (btrfs_chunk_sector_size(leaf, chunk) != root->sectorsize) {
6243 btrfs_err(root->fs_info, "invalid chunk sectorsize %u",
6244 btrfs_chunk_sector_size(leaf, chunk));
6247 if (!length || !IS_ALIGNED(length, root->sectorsize)) {
6248 btrfs_err(root->fs_info,
6249 "invalid chunk length %llu", length);
6252 if (!is_power_of_2(stripe_len)) {
6253 btrfs_err(root->fs_info, "invalid chunk stripe length: %llu",
6257 if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6259 btrfs_err(root->fs_info, "unrecognized chunk type: %llu",
6260 ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
6261 BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6262 btrfs_chunk_type(leaf, chunk));
6266 if (!is_power_of_2(type & BTRFS_BLOCK_GROUP_PROFILE_MASK) &&
6267 (type & BTRFS_BLOCK_GROUP_PROFILE_MASK) != 0) {
6268 btrfs_err(root->fs_info,
6269 "invalid chunk profile flag: 0x%llx, expect 0 or 1 bit set",
6270 type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
6273 if ((type & BTRFS_BLOCK_GROUP_TYPE_MASK) == 0) {
6274 btrfs_err(root->fs_info, "missing chunk type flag: 0x%llx", type);
6278 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) &&
6279 (type & (BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA))) {
6280 btrfs_err(root->fs_info,
6281 "system chunk with data or metadata type: 0x%llx", type);
6285 features = btrfs_super_incompat_flags(root->fs_info->super_copy);
6286 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
6290 if ((type & BTRFS_BLOCK_GROUP_METADATA) &&
6291 (type & BTRFS_BLOCK_GROUP_DATA)) {
6292 btrfs_err(root->fs_info,
6293 "mixed chunk type in non-mixed mode: 0x%llx", type);
6298 if ((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes != 2) ||
6299 (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes != 2) ||
6300 (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) ||
6301 (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) ||
6302 (type & BTRFS_BLOCK_GROUP_DUP && num_stripes != 2) ||
6303 ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
6304 num_stripes != 1)) {
6305 btrfs_err(root->fs_info,
6306 "invalid num_stripes:sub_stripes %u:%u for profile %llu",
6307 num_stripes, sub_stripes,
6308 type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
6315 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
6316 struct extent_buffer *leaf,
6317 struct btrfs_chunk *chunk)
6319 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
6320 struct map_lookup *map;
6321 struct extent_map *em;
6326 u8 uuid[BTRFS_UUID_SIZE];
6331 logical = key->offset;
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);
6336 ret = btrfs_check_chunk_valid(root, leaf, chunk, logical);
6340 read_lock(&map_tree->map_tree.lock);
6341 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
6342 read_unlock(&map_tree->map_tree.lock);
6344 /* already mapped? */
6345 if (em && em->start <= logical && em->start + em->len > logical) {
6346 free_extent_map(em);
6349 free_extent_map(em);
6352 em = alloc_extent_map();
6355 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6357 free_extent_map(em);
6361 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6362 em->map_lookup = map;
6363 em->start = logical;
6366 em->block_start = 0;
6367 em->block_len = em->len;
6369 map->num_stripes = num_stripes;
6370 map->io_width = btrfs_chunk_io_width(leaf, chunk);
6371 map->io_align = btrfs_chunk_io_align(leaf, chunk);
6372 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
6373 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6374 map->type = btrfs_chunk_type(leaf, chunk);
6375 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6376 for (i = 0; i < num_stripes; i++) {
6377 map->stripes[i].physical =
6378 btrfs_stripe_offset_nr(leaf, chunk, i);
6379 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6380 read_extent_buffer(leaf, uuid, (unsigned long)
6381 btrfs_stripe_dev_uuid_nr(chunk, i),
6383 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
6385 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
6386 free_extent_map(em);
6389 if (!map->stripes[i].dev) {
6390 map->stripes[i].dev =
6391 add_missing_dev(root, root->fs_info->fs_devices,
6393 if (!map->stripes[i].dev) {
6394 free_extent_map(em);
6397 btrfs_warn(root->fs_info, "devid %llu uuid %pU is missing",
6400 map->stripes[i].dev->in_fs_metadata = 1;
6403 write_lock(&map_tree->map_tree.lock);
6404 ret = add_extent_mapping(&map_tree->map_tree, em, 0);
6405 write_unlock(&map_tree->map_tree.lock);
6406 BUG_ON(ret); /* Tree corruption */
6407 free_extent_map(em);
6412 static void fill_device_from_item(struct extent_buffer *leaf,
6413 struct btrfs_dev_item *dev_item,
6414 struct btrfs_device *device)
6418 device->devid = btrfs_device_id(leaf, dev_item);
6419 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6420 device->total_bytes = device->disk_total_bytes;
6421 device->commit_total_bytes = device->disk_total_bytes;
6422 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6423 device->commit_bytes_used = device->bytes_used;
6424 device->type = btrfs_device_type(leaf, dev_item);
6425 device->io_align = btrfs_device_io_align(leaf, dev_item);
6426 device->io_width = btrfs_device_io_width(leaf, dev_item);
6427 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6428 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6429 device->is_tgtdev_for_dev_replace = 0;
6431 ptr = btrfs_device_uuid(dev_item);
6432 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6435 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_root *root,
6438 struct btrfs_fs_devices *fs_devices;
6441 BUG_ON(!mutex_is_locked(&uuid_mutex));
6443 fs_devices = root->fs_info->fs_devices->seed;
6444 while (fs_devices) {
6445 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE))
6448 fs_devices = fs_devices->seed;
6451 fs_devices = find_fsid(fsid);
6453 if (!btrfs_test_opt(root, DEGRADED))
6454 return ERR_PTR(-ENOENT);
6456 fs_devices = alloc_fs_devices(fsid);
6457 if (IS_ERR(fs_devices))
6460 fs_devices->seeding = 1;
6461 fs_devices->opened = 1;
6465 fs_devices = clone_fs_devices(fs_devices);
6466 if (IS_ERR(fs_devices))
6469 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
6470 root->fs_info->bdev_holder);
6472 free_fs_devices(fs_devices);
6473 fs_devices = ERR_PTR(ret);
6477 if (!fs_devices->seeding) {
6478 __btrfs_close_devices(fs_devices);
6479 free_fs_devices(fs_devices);
6480 fs_devices = ERR_PTR(-EINVAL);
6484 fs_devices->seed = root->fs_info->fs_devices->seed;
6485 root->fs_info->fs_devices->seed = fs_devices;
6490 static int read_one_dev(struct btrfs_root *root,
6491 struct extent_buffer *leaf,
6492 struct btrfs_dev_item *dev_item)
6494 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6495 struct btrfs_device *device;
6498 u8 fs_uuid[BTRFS_UUID_SIZE];
6499 u8 dev_uuid[BTRFS_UUID_SIZE];
6501 devid = btrfs_device_id(leaf, dev_item);
6502 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6504 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6507 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
6508 fs_devices = open_seed_devices(root, fs_uuid);
6509 if (IS_ERR(fs_devices))
6510 return PTR_ERR(fs_devices);
6513 device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
6515 if (!btrfs_test_opt(root, DEGRADED))
6518 device = add_missing_dev(root, fs_devices, devid, dev_uuid);
6521 btrfs_warn(root->fs_info, "devid %llu uuid %pU missing",
6524 if (!device->bdev && !btrfs_test_opt(root, DEGRADED))
6527 if(!device->bdev && !device->missing) {
6529 * this happens when a device that was properly setup
6530 * in the device info lists suddenly goes bad.
6531 * device->bdev is NULL, and so we have to set
6532 * device->missing to one here
6534 device->fs_devices->missing_devices++;
6535 device->missing = 1;
6538 /* Move the device to its own fs_devices */
6539 if (device->fs_devices != fs_devices) {
6540 ASSERT(device->missing);
6542 list_move(&device->dev_list, &fs_devices->devices);
6543 device->fs_devices->num_devices--;
6544 fs_devices->num_devices++;
6546 device->fs_devices->missing_devices--;
6547 fs_devices->missing_devices++;
6549 device->fs_devices = fs_devices;
6553 if (device->fs_devices != root->fs_info->fs_devices) {
6554 BUG_ON(device->writeable);
6555 if (device->generation !=
6556 btrfs_device_generation(leaf, dev_item))
6560 fill_device_from_item(leaf, dev_item, device);
6561 device->in_fs_metadata = 1;
6562 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
6563 device->fs_devices->total_rw_bytes += device->total_bytes;
6564 spin_lock(&root->fs_info->free_chunk_lock);
6565 root->fs_info->free_chunk_space += device->total_bytes -
6567 spin_unlock(&root->fs_info->free_chunk_lock);
6573 int btrfs_read_sys_array(struct btrfs_root *root)
6575 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
6576 struct extent_buffer *sb;
6577 struct btrfs_disk_key *disk_key;
6578 struct btrfs_chunk *chunk;
6580 unsigned long sb_array_offset;
6587 struct btrfs_key key;
6589 ASSERT(BTRFS_SUPER_INFO_SIZE <= root->nodesize);
6591 * This will create extent buffer of nodesize, superblock size is
6592 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6593 * overallocate but we can keep it as-is, only the first page is used.
6595 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET);
6598 btrfs_set_buffer_uptodate(sb);
6599 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6601 * The sb extent buffer is artifical and just used to read the system array.
6602 * btrfs_set_buffer_uptodate() call does not properly mark all it's
6603 * pages up-to-date when the page is larger: extent does not cover the
6604 * whole page and consequently check_page_uptodate does not find all
6605 * the page's extents up-to-date (the hole beyond sb),
6606 * write_extent_buffer then triggers a WARN_ON.
6608 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6609 * but sb spans only this function. Add an explicit SetPageUptodate call
6610 * to silence the warning eg. on PowerPC 64.
6612 if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
6613 SetPageUptodate(sb->pages[0]);
6615 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6616 array_size = btrfs_super_sys_array_size(super_copy);
6618 array_ptr = super_copy->sys_chunk_array;
6619 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6622 while (cur_offset < array_size) {
6623 disk_key = (struct btrfs_disk_key *)array_ptr;
6624 len = sizeof(*disk_key);
6625 if (cur_offset + len > array_size)
6626 goto out_short_read;
6628 btrfs_disk_key_to_cpu(&key, disk_key);
6631 sb_array_offset += len;
6634 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6635 chunk = (struct btrfs_chunk *)sb_array_offset;
6637 * At least one btrfs_chunk with one stripe must be
6638 * present, exact stripe count check comes afterwards
6640 len = btrfs_chunk_item_size(1);
6641 if (cur_offset + len > array_size)
6642 goto out_short_read;
6644 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6647 "BTRFS: invalid number of stripes %u in sys_array at offset %u\n",
6648 num_stripes, cur_offset);
6653 type = btrfs_chunk_type(sb, chunk);
6654 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
6655 btrfs_err(root->fs_info,
6656 "invalid chunk type %llu in sys_array at offset %u",
6662 len = btrfs_chunk_item_size(num_stripes);
6663 if (cur_offset + len > array_size)
6664 goto out_short_read;
6666 ret = read_one_chunk(root, &key, sb, chunk);
6674 sb_array_offset += len;
6677 free_extent_buffer(sb);
6681 printk(KERN_ERR "BTRFS: sys_array too short to read %u bytes at offset %u\n",
6683 free_extent_buffer(sb);
6687 int btrfs_read_chunk_tree(struct btrfs_root *root)
6689 struct btrfs_path *path;
6690 struct extent_buffer *leaf;
6691 struct btrfs_key key;
6692 struct btrfs_key found_key;
6696 root = root->fs_info->chunk_root;
6698 path = btrfs_alloc_path();
6702 mutex_lock(&uuid_mutex);
6706 * It is possible for mount and umount to race in such a way that
6707 * we execute this code path, but open_fs_devices failed to clear
6708 * total_rw_bytes. We certainly want it cleared before reading the
6709 * device items, so clear it here.
6711 root->fs_info->fs_devices->total_rw_bytes = 0;
6714 * Read all device items, and then all the chunk items. All
6715 * device items are found before any chunk item (their object id
6716 * is smaller than the lowest possible object id for a chunk
6717 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6719 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6722 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6726 leaf = path->nodes[0];
6727 slot = path->slots[0];
6728 if (slot >= btrfs_header_nritems(leaf)) {
6729 ret = btrfs_next_leaf(root, path);
6736 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6737 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6738 struct btrfs_dev_item *dev_item;
6739 dev_item = btrfs_item_ptr(leaf, slot,
6740 struct btrfs_dev_item);
6741 ret = read_one_dev(root, leaf, dev_item);
6744 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6745 struct btrfs_chunk *chunk;
6746 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6747 ret = read_one_chunk(root, &found_key, leaf, chunk);
6755 unlock_chunks(root);
6756 mutex_unlock(&uuid_mutex);
6758 btrfs_free_path(path);
6762 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6764 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6765 struct btrfs_device *device;
6767 while (fs_devices) {
6768 mutex_lock(&fs_devices->device_list_mutex);
6769 list_for_each_entry(device, &fs_devices->devices, dev_list)
6770 device->dev_root = fs_info->dev_root;
6771 mutex_unlock(&fs_devices->device_list_mutex);
6773 fs_devices = fs_devices->seed;
6777 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6781 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6782 btrfs_dev_stat_reset(dev, i);
6785 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6787 struct btrfs_key key;
6788 struct btrfs_key found_key;
6789 struct btrfs_root *dev_root = fs_info->dev_root;
6790 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6791 struct extent_buffer *eb;
6794 struct btrfs_device *device;
6795 struct btrfs_path *path = NULL;
6798 path = btrfs_alloc_path();
6804 mutex_lock(&fs_devices->device_list_mutex);
6805 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6807 struct btrfs_dev_stats_item *ptr;
6810 key.type = BTRFS_DEV_STATS_KEY;
6811 key.offset = device->devid;
6812 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6814 __btrfs_reset_dev_stats(device);
6815 device->dev_stats_valid = 1;
6816 btrfs_release_path(path);
6819 slot = path->slots[0];
6820 eb = path->nodes[0];
6821 btrfs_item_key_to_cpu(eb, &found_key, slot);
6822 item_size = btrfs_item_size_nr(eb, slot);
6824 ptr = btrfs_item_ptr(eb, slot,
6825 struct btrfs_dev_stats_item);
6827 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6828 if (item_size >= (1 + i) * sizeof(__le64))
6829 btrfs_dev_stat_set(device, i,
6830 btrfs_dev_stats_value(eb, ptr, i));
6832 btrfs_dev_stat_reset(device, i);
6835 device->dev_stats_valid = 1;
6836 btrfs_dev_stat_print_on_load(device);
6837 btrfs_release_path(path);
6839 mutex_unlock(&fs_devices->device_list_mutex);
6842 btrfs_free_path(path);
6843 return ret < 0 ? ret : 0;
6846 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
6847 struct btrfs_root *dev_root,
6848 struct btrfs_device *device)
6850 struct btrfs_path *path;
6851 struct btrfs_key key;
6852 struct extent_buffer *eb;
6853 struct btrfs_dev_stats_item *ptr;
6858 key.type = BTRFS_DEV_STATS_KEY;
6859 key.offset = device->devid;
6861 path = btrfs_alloc_path();
6863 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
6865 btrfs_warn_in_rcu(dev_root->fs_info,
6866 "error %d while searching for dev_stats item for device %s",
6867 ret, rcu_str_deref(device->name));
6872 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
6873 /* need to delete old one and insert a new one */
6874 ret = btrfs_del_item(trans, dev_root, path);
6876 btrfs_warn_in_rcu(dev_root->fs_info,
6877 "delete too small dev_stats item for device %s failed %d",
6878 rcu_str_deref(device->name), ret);
6885 /* need to insert a new item */
6886 btrfs_release_path(path);
6887 ret = btrfs_insert_empty_item(trans, dev_root, path,
6888 &key, sizeof(*ptr));
6890 btrfs_warn_in_rcu(dev_root->fs_info,
6891 "insert dev_stats item for device %s failed %d",
6892 rcu_str_deref(device->name), ret);
6897 eb = path->nodes[0];
6898 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
6899 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6900 btrfs_set_dev_stats_value(eb, ptr, i,
6901 btrfs_dev_stat_read(device, i));
6902 btrfs_mark_buffer_dirty(eb);
6905 btrfs_free_path(path);
6910 * called from commit_transaction. Writes all changed device stats to disk.
6912 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
6913 struct btrfs_fs_info *fs_info)
6915 struct btrfs_root *dev_root = fs_info->dev_root;
6916 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6917 struct btrfs_device *device;
6921 mutex_lock(&fs_devices->device_list_mutex);
6922 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6923 if (!device->dev_stats_valid || !btrfs_dev_stats_dirty(device))
6926 stats_cnt = atomic_read(&device->dev_stats_ccnt);
6927 ret = update_dev_stat_item(trans, dev_root, device);
6929 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
6931 mutex_unlock(&fs_devices->device_list_mutex);
6936 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
6938 btrfs_dev_stat_inc(dev, index);
6939 btrfs_dev_stat_print_on_error(dev);
6942 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
6944 if (!dev->dev_stats_valid)
6946 btrfs_err_rl_in_rcu(dev->dev_root->fs_info,
6947 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
6948 rcu_str_deref(dev->name),
6949 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6950 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6951 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6952 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6953 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6956 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
6960 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6961 if (btrfs_dev_stat_read(dev, i) != 0)
6963 if (i == BTRFS_DEV_STAT_VALUES_MAX)
6964 return; /* all values == 0, suppress message */
6966 btrfs_info_in_rcu(dev->dev_root->fs_info,
6967 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
6968 rcu_str_deref(dev->name),
6969 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6970 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6971 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6972 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6973 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6976 int btrfs_get_dev_stats(struct btrfs_root *root,
6977 struct btrfs_ioctl_get_dev_stats *stats)
6979 struct btrfs_device *dev;
6980 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6983 mutex_lock(&fs_devices->device_list_mutex);
6984 dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
6985 mutex_unlock(&fs_devices->device_list_mutex);
6988 btrfs_warn(root->fs_info, "get dev_stats failed, device not found");
6990 } else if (!dev->dev_stats_valid) {
6991 btrfs_warn(root->fs_info, "get dev_stats failed, not yet valid");
6993 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
6994 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6995 if (stats->nr_items > i)
6997 btrfs_dev_stat_read_and_reset(dev, i);
6999 btrfs_dev_stat_reset(dev, i);
7002 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7003 if (stats->nr_items > i)
7004 stats->values[i] = btrfs_dev_stat_read(dev, i);
7006 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7007 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7011 void btrfs_scratch_superblocks(struct block_device *bdev, char *device_path)
7013 struct buffer_head *bh;
7014 struct btrfs_super_block *disk_super;
7020 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX;
7023 if (btrfs_read_dev_one_super(bdev, copy_num, &bh))
7026 disk_super = (struct btrfs_super_block *)bh->b_data;
7028 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
7029 set_buffer_dirty(bh);
7030 sync_dirty_buffer(bh);
7034 /* Notify udev that device has changed */
7035 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
7037 /* Update ctime/mtime for device path for libblkid */
7038 update_dev_time(device_path);
7042 * Update the size of all devices, which is used for writing out the
7045 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
7047 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7048 struct btrfs_device *curr, *next;
7050 if (list_empty(&fs_devices->resized_devices))
7053 mutex_lock(&fs_devices->device_list_mutex);
7054 lock_chunks(fs_info->dev_root);
7055 list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
7057 list_del_init(&curr->resized_list);
7058 curr->commit_total_bytes = curr->disk_total_bytes;
7060 unlock_chunks(fs_info->dev_root);
7061 mutex_unlock(&fs_devices->device_list_mutex);
7064 /* Must be invoked during the transaction commit */
7065 void btrfs_update_commit_device_bytes_used(struct btrfs_root *root,
7066 struct btrfs_transaction *transaction)
7068 struct extent_map *em;
7069 struct map_lookup *map;
7070 struct btrfs_device *dev;
7073 if (list_empty(&transaction->pending_chunks))
7076 /* In order to kick the device replace finish process */
7078 list_for_each_entry(em, &transaction->pending_chunks, list) {
7079 map = em->map_lookup;
7081 for (i = 0; i < map->num_stripes; i++) {
7082 dev = map->stripes[i].dev;
7083 dev->commit_bytes_used = dev->bytes_used;
7084 dev->has_pending_chunks = false;
7087 unlock_chunks(root);
7090 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
7092 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7093 while (fs_devices) {
7094 fs_devices->fs_info = fs_info;
7095 fs_devices = fs_devices->seed;
7099 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
7101 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7102 while (fs_devices) {
7103 fs_devices->fs_info = NULL;
7104 fs_devices = fs_devices->seed;
7108 void btrfs_close_one_device(struct btrfs_device *device)
7110 struct btrfs_fs_devices *fs_devices = device->fs_devices;
7111 struct btrfs_device *new_device;
7112 struct rcu_string *name;
7115 fs_devices->open_devices--;
7117 if (device->writeable &&
7118 device->devid != BTRFS_DEV_REPLACE_DEVID) {
7119 list_del_init(&device->dev_alloc_list);
7120 fs_devices->rw_devices--;
7123 if (device->missing)
7124 fs_devices->missing_devices--;
7126 new_device = btrfs_alloc_device(NULL, &device->devid,
7128 BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
7130 /* Safe because we are under uuid_mutex */
7132 name = rcu_string_strdup(device->name->str, GFP_NOFS);
7133 BUG_ON(!name); /* -ENOMEM */
7134 rcu_assign_pointer(new_device->name, name);
7137 list_replace_rcu(&device->dev_list, &new_device->dev_list);
7138 new_device->fs_devices = device->fs_devices;
7140 call_rcu(&device->rcu, free_device);