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 const 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);
572 fs_devs->num_devices--;
573 list_del(&dev->dev_list);
574 rcu_string_free(dev->name);
583 * Add new device to list of registered devices
586 * 1 - first time device is seen
587 * 0 - device already known
590 static noinline int device_list_add(const char *path,
591 struct btrfs_super_block *disk_super,
592 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
594 struct btrfs_device *device;
595 struct btrfs_fs_devices *fs_devices;
596 struct rcu_string *name;
598 u64 found_transid = btrfs_super_generation(disk_super);
600 fs_devices = find_fsid(disk_super->fsid);
602 fs_devices = alloc_fs_devices(disk_super->fsid);
603 if (IS_ERR(fs_devices))
604 return PTR_ERR(fs_devices);
606 list_add(&fs_devices->list, &fs_uuids);
610 device = __find_device(&fs_devices->devices, devid,
611 disk_super->dev_item.uuid);
615 if (fs_devices->opened)
618 device = btrfs_alloc_device(NULL, &devid,
619 disk_super->dev_item.uuid);
620 if (IS_ERR(device)) {
621 /* we can safely leave the fs_devices entry around */
622 return PTR_ERR(device);
625 name = rcu_string_strdup(path, GFP_NOFS);
630 rcu_assign_pointer(device->name, name);
632 mutex_lock(&fs_devices->device_list_mutex);
633 list_add_rcu(&device->dev_list, &fs_devices->devices);
634 fs_devices->num_devices++;
635 mutex_unlock(&fs_devices->device_list_mutex);
638 device->fs_devices = fs_devices;
639 } else if (!device->name || strcmp(device->name->str, path)) {
641 * When FS is already mounted.
642 * 1. If you are here and if the device->name is NULL that
643 * means this device was missing at time of FS mount.
644 * 2. If you are here and if the device->name is different
645 * from 'path' that means either
646 * a. The same device disappeared and reappeared with
648 * b. The missing-disk-which-was-replaced, has
651 * We must allow 1 and 2a above. But 2b would be a spurious
654 * Further in case of 1 and 2a above, the disk at 'path'
655 * would have missed some transaction when it was away and
656 * in case of 2a the stale bdev has to be updated as well.
657 * 2b must not be allowed at all time.
661 * For now, we do allow update to btrfs_fs_device through the
662 * btrfs dev scan cli after FS has been mounted. We're still
663 * tracking a problem where systems fail mount by subvolume id
664 * when we reject replacement on a mounted FS.
666 if (!fs_devices->opened && found_transid < device->generation) {
668 * That is if the FS is _not_ mounted and if you
669 * are here, that means there is more than one
670 * disk with same uuid and devid.We keep the one
671 * with larger generation number or the last-in if
672 * generation are equal.
677 name = rcu_string_strdup(path, GFP_NOFS);
680 rcu_string_free(device->name);
681 rcu_assign_pointer(device->name, name);
682 if (device->missing) {
683 fs_devices->missing_devices--;
689 * Unmount does not free the btrfs_device struct but would zero
690 * generation along with most of the other members. So just update
691 * it back. We need it to pick the disk with largest generation
694 if (!fs_devices->opened)
695 device->generation = found_transid;
698 * if there is new btrfs on an already registered device,
699 * then remove the stale device entry.
701 btrfs_free_stale_device(device);
703 *fs_devices_ret = fs_devices;
708 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
710 struct btrfs_fs_devices *fs_devices;
711 struct btrfs_device *device;
712 struct btrfs_device *orig_dev;
714 fs_devices = alloc_fs_devices(orig->fsid);
715 if (IS_ERR(fs_devices))
718 mutex_lock(&orig->device_list_mutex);
719 fs_devices->total_devices = orig->total_devices;
721 /* We have held the volume lock, it is safe to get the devices. */
722 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
723 struct rcu_string *name;
725 device = btrfs_alloc_device(NULL, &orig_dev->devid,
731 * This is ok to do without rcu read locked because we hold the
732 * uuid mutex so nothing we touch in here is going to disappear.
734 if (orig_dev->name) {
735 name = rcu_string_strdup(orig_dev->name->str, GFP_NOFS);
740 rcu_assign_pointer(device->name, name);
743 list_add(&device->dev_list, &fs_devices->devices);
744 device->fs_devices = fs_devices;
745 fs_devices->num_devices++;
747 mutex_unlock(&orig->device_list_mutex);
750 mutex_unlock(&orig->device_list_mutex);
751 free_fs_devices(fs_devices);
752 return ERR_PTR(-ENOMEM);
755 void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step)
757 struct btrfs_device *device, *next;
758 struct btrfs_device *latest_dev = NULL;
760 mutex_lock(&uuid_mutex);
762 /* This is the initialized path, it is safe to release the devices. */
763 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
764 if (device->in_fs_metadata) {
765 if (!device->is_tgtdev_for_dev_replace &&
767 device->generation > latest_dev->generation)) {
773 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
775 * In the first step, keep the device which has
776 * the correct fsid and the devid that is used
777 * for the dev_replace procedure.
778 * In the second step, the dev_replace state is
779 * read from the device tree and it is known
780 * whether the procedure is really active or
781 * not, which means whether this device is
782 * used or whether it should be removed.
784 if (step == 0 || device->is_tgtdev_for_dev_replace) {
789 blkdev_put(device->bdev, device->mode);
791 fs_devices->open_devices--;
793 if (device->writeable) {
794 list_del_init(&device->dev_alloc_list);
795 device->writeable = 0;
796 if (!device->is_tgtdev_for_dev_replace)
797 fs_devices->rw_devices--;
799 list_del_init(&device->dev_list);
800 fs_devices->num_devices--;
801 rcu_string_free(device->name);
805 if (fs_devices->seed) {
806 fs_devices = fs_devices->seed;
810 fs_devices->latest_bdev = latest_dev->bdev;
812 mutex_unlock(&uuid_mutex);
815 static void __free_device(struct work_struct *work)
817 struct btrfs_device *device;
819 device = container_of(work, struct btrfs_device, rcu_work);
822 blkdev_put(device->bdev, device->mode);
824 rcu_string_free(device->name);
828 static void free_device(struct rcu_head *head)
830 struct btrfs_device *device;
832 device = container_of(head, struct btrfs_device, rcu);
834 INIT_WORK(&device->rcu_work, __free_device);
835 schedule_work(&device->rcu_work);
838 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
840 struct btrfs_device *device, *tmp;
842 if (--fs_devices->opened > 0)
845 mutex_lock(&fs_devices->device_list_mutex);
846 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
847 btrfs_close_one_device(device);
849 mutex_unlock(&fs_devices->device_list_mutex);
851 WARN_ON(fs_devices->open_devices);
852 WARN_ON(fs_devices->rw_devices);
853 fs_devices->opened = 0;
854 fs_devices->seeding = 0;
859 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
861 struct btrfs_fs_devices *seed_devices = NULL;
864 mutex_lock(&uuid_mutex);
865 ret = __btrfs_close_devices(fs_devices);
866 if (!fs_devices->opened) {
867 seed_devices = fs_devices->seed;
868 fs_devices->seed = NULL;
870 mutex_unlock(&uuid_mutex);
872 while (seed_devices) {
873 fs_devices = seed_devices;
874 seed_devices = fs_devices->seed;
875 __btrfs_close_devices(fs_devices);
876 free_fs_devices(fs_devices);
879 * Wait for rcu kworkers under __btrfs_close_devices
880 * to finish all blkdev_puts so device is really
881 * free when umount is done.
887 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
888 fmode_t flags, void *holder)
890 struct request_queue *q;
891 struct block_device *bdev;
892 struct list_head *head = &fs_devices->devices;
893 struct btrfs_device *device;
894 struct btrfs_device *latest_dev = NULL;
895 struct buffer_head *bh;
896 struct btrfs_super_block *disk_super;
903 list_for_each_entry(device, head, dev_list) {
909 /* Just open everything we can; ignore failures here */
910 if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
914 disk_super = (struct btrfs_super_block *)bh->b_data;
915 devid = btrfs_stack_device_id(&disk_super->dev_item);
916 if (devid != device->devid)
919 if (memcmp(device->uuid, disk_super->dev_item.uuid,
923 device->generation = btrfs_super_generation(disk_super);
925 device->generation > latest_dev->generation)
928 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
929 device->writeable = 0;
931 device->writeable = !bdev_read_only(bdev);
935 q = bdev_get_queue(bdev);
936 if (blk_queue_discard(q))
937 device->can_discard = 1;
940 device->in_fs_metadata = 0;
941 device->mode = flags;
943 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
944 fs_devices->rotating = 1;
946 fs_devices->open_devices++;
947 if (device->writeable &&
948 device->devid != BTRFS_DEV_REPLACE_DEVID) {
949 fs_devices->rw_devices++;
950 list_add(&device->dev_alloc_list,
951 &fs_devices->alloc_list);
958 blkdev_put(bdev, flags);
961 if (fs_devices->open_devices == 0) {
965 fs_devices->seeding = seeding;
966 fs_devices->opened = 1;
967 fs_devices->latest_bdev = latest_dev->bdev;
968 fs_devices->total_rw_bytes = 0;
973 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
974 fmode_t flags, void *holder)
978 mutex_lock(&uuid_mutex);
979 if (fs_devices->opened) {
980 fs_devices->opened++;
983 ret = __btrfs_open_devices(fs_devices, flags, holder);
985 mutex_unlock(&uuid_mutex);
990 * Look for a btrfs signature on a device. This may be called out of the mount path
991 * and we are not allowed to call set_blocksize during the scan. The superblock
992 * is read via pagecache
994 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
995 struct btrfs_fs_devices **fs_devices_ret)
997 struct btrfs_super_block *disk_super;
998 struct block_device *bdev;
1009 * we would like to check all the supers, but that would make
1010 * a btrfs mount succeed after a mkfs from a different FS.
1011 * So, we need to add a special mount option to scan for
1012 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1014 bytenr = btrfs_sb_offset(0);
1015 flags |= FMODE_EXCL;
1016 mutex_lock(&uuid_mutex);
1018 bdev = blkdev_get_by_path(path, flags, holder);
1021 ret = PTR_ERR(bdev);
1025 /* make sure our super fits in the device */
1026 if (bytenr + PAGE_CACHE_SIZE >= i_size_read(bdev->bd_inode))
1027 goto error_bdev_put;
1029 /* make sure our super fits in the page */
1030 if (sizeof(*disk_super) > PAGE_CACHE_SIZE)
1031 goto error_bdev_put;
1033 /* make sure our super doesn't straddle pages on disk */
1034 index = bytenr >> PAGE_CACHE_SHIFT;
1035 if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_CACHE_SHIFT != index)
1036 goto error_bdev_put;
1038 /* pull in the page with our super */
1039 page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
1042 if (IS_ERR_OR_NULL(page))
1043 goto error_bdev_put;
1047 /* align our pointer to the offset of the super block */
1048 disk_super = p + (bytenr & ~PAGE_CACHE_MASK);
1050 if (btrfs_super_bytenr(disk_super) != bytenr ||
1051 btrfs_super_magic(disk_super) != BTRFS_MAGIC)
1054 devid = btrfs_stack_device_id(&disk_super->dev_item);
1055 transid = btrfs_super_generation(disk_super);
1056 total_devices = btrfs_super_num_devices(disk_super);
1058 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
1060 if (disk_super->label[0]) {
1061 if (disk_super->label[BTRFS_LABEL_SIZE - 1])
1062 disk_super->label[BTRFS_LABEL_SIZE - 1] = '\0';
1063 printk(KERN_INFO "BTRFS: device label %s ", disk_super->label);
1065 printk(KERN_INFO "BTRFS: device fsid %pU ", disk_super->fsid);
1068 printk(KERN_CONT "devid %llu transid %llu %s\n", devid, transid, path);
1071 if (!ret && fs_devices_ret)
1072 (*fs_devices_ret)->total_devices = total_devices;
1076 page_cache_release(page);
1079 blkdev_put(bdev, flags);
1081 mutex_unlock(&uuid_mutex);
1085 /* helper to account the used device space in the range */
1086 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
1087 u64 end, u64 *length)
1089 struct btrfs_key key;
1090 struct btrfs_root *root = device->dev_root;
1091 struct btrfs_dev_extent *dev_extent;
1092 struct btrfs_path *path;
1096 struct extent_buffer *l;
1100 if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
1103 path = btrfs_alloc_path();
1108 key.objectid = device->devid;
1110 key.type = BTRFS_DEV_EXTENT_KEY;
1112 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1116 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1123 slot = path->slots[0];
1124 if (slot >= btrfs_header_nritems(l)) {
1125 ret = btrfs_next_leaf(root, path);
1133 btrfs_item_key_to_cpu(l, &key, slot);
1135 if (key.objectid < device->devid)
1138 if (key.objectid > device->devid)
1141 if (key.type != BTRFS_DEV_EXTENT_KEY)
1144 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1145 extent_end = key.offset + btrfs_dev_extent_length(l,
1147 if (key.offset <= start && extent_end > end) {
1148 *length = end - start + 1;
1150 } else if (key.offset <= start && extent_end > start)
1151 *length += extent_end - start;
1152 else if (key.offset > start && extent_end <= end)
1153 *length += extent_end - key.offset;
1154 else if (key.offset > start && key.offset <= end) {
1155 *length += end - key.offset + 1;
1157 } else if (key.offset > end)
1165 btrfs_free_path(path);
1169 static int contains_pending_extent(struct btrfs_transaction *transaction,
1170 struct btrfs_device *device,
1171 u64 *start, u64 len)
1173 struct btrfs_fs_info *fs_info = device->dev_root->fs_info;
1174 struct extent_map *em;
1175 struct list_head *search_list = &fs_info->pinned_chunks;
1177 u64 physical_start = *start;
1180 search_list = &transaction->pending_chunks;
1182 list_for_each_entry(em, search_list, list) {
1183 struct map_lookup *map;
1186 map = (struct map_lookup *)em->bdev;
1187 for (i = 0; i < map->num_stripes; i++) {
1190 if (map->stripes[i].dev != device)
1192 if (map->stripes[i].physical >= physical_start + len ||
1193 map->stripes[i].physical + em->orig_block_len <=
1197 * Make sure that while processing the pinned list we do
1198 * not override our *start with a lower value, because
1199 * we can have pinned chunks that fall within this
1200 * device hole and that have lower physical addresses
1201 * than the pending chunks we processed before. If we
1202 * do not take this special care we can end up getting
1203 * 2 pending chunks that start at the same physical
1204 * device offsets because the end offset of a pinned
1205 * chunk can be equal to the start offset of some
1208 end = map->stripes[i].physical + em->orig_block_len;
1215 if (search_list != &fs_info->pinned_chunks) {
1216 search_list = &fs_info->pinned_chunks;
1225 * find_free_dev_extent_start - find free space in the specified device
1226 * @device: the device which we search the free space in
1227 * @num_bytes: the size of the free space that we need
1228 * @search_start: the position from which to begin the search
1229 * @start: store the start of the free space.
1230 * @len: the size of the free space. that we find, or the size
1231 * of the max free space if we don't find suitable free space
1233 * this uses a pretty simple search, the expectation is that it is
1234 * called very infrequently and that a given device has a small number
1237 * @start is used to store the start of the free space if we find. But if we
1238 * don't find suitable free space, it will be used to store the start position
1239 * of the max free space.
1241 * @len is used to store the size of the free space that we find.
1242 * But if we don't find suitable free space, it is used to store the size of
1243 * the max free space.
1245 int find_free_dev_extent_start(struct btrfs_transaction *transaction,
1246 struct btrfs_device *device, u64 num_bytes,
1247 u64 search_start, u64 *start, u64 *len)
1249 struct btrfs_key key;
1250 struct btrfs_root *root = device->dev_root;
1251 struct btrfs_dev_extent *dev_extent;
1252 struct btrfs_path *path;
1257 u64 search_end = device->total_bytes;
1260 struct extent_buffer *l;
1261 u64 min_search_start;
1264 * We don't want to overwrite the superblock on the drive nor any area
1265 * used by the boot loader (grub for example), so we make sure to start
1266 * at an offset of at least 1MB.
1268 min_search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
1269 search_start = max(search_start, min_search_start);
1271 path = btrfs_alloc_path();
1275 max_hole_start = search_start;
1279 if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1285 path->search_commit_root = 1;
1286 path->skip_locking = 1;
1288 key.objectid = device->devid;
1289 key.offset = search_start;
1290 key.type = BTRFS_DEV_EXTENT_KEY;
1292 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1296 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1303 slot = path->slots[0];
1304 if (slot >= btrfs_header_nritems(l)) {
1305 ret = btrfs_next_leaf(root, path);
1313 btrfs_item_key_to_cpu(l, &key, slot);
1315 if (key.objectid < device->devid)
1318 if (key.objectid > device->devid)
1321 if (key.type != BTRFS_DEV_EXTENT_KEY)
1324 if (key.offset > search_start) {
1325 hole_size = key.offset - search_start;
1328 * Have to check before we set max_hole_start, otherwise
1329 * we could end up sending back this offset anyway.
1331 if (contains_pending_extent(transaction, device,
1334 if (key.offset >= search_start) {
1335 hole_size = key.offset - search_start;
1342 if (hole_size > max_hole_size) {
1343 max_hole_start = search_start;
1344 max_hole_size = hole_size;
1348 * If this free space is greater than which we need,
1349 * it must be the max free space that we have found
1350 * until now, so max_hole_start must point to the start
1351 * of this free space and the length of this free space
1352 * is stored in max_hole_size. Thus, we return
1353 * max_hole_start and max_hole_size and go back to the
1356 if (hole_size >= num_bytes) {
1362 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1363 extent_end = key.offset + btrfs_dev_extent_length(l,
1365 if (extent_end > search_start)
1366 search_start = extent_end;
1373 * At this point, search_start should be the end of
1374 * allocated dev extents, and when shrinking the device,
1375 * search_end may be smaller than search_start.
1377 if (search_end > search_start) {
1378 hole_size = search_end - search_start;
1380 if (contains_pending_extent(transaction, device, &search_start,
1382 btrfs_release_path(path);
1386 if (hole_size > max_hole_size) {
1387 max_hole_start = search_start;
1388 max_hole_size = hole_size;
1393 if (max_hole_size < num_bytes)
1399 btrfs_free_path(path);
1400 *start = max_hole_start;
1402 *len = max_hole_size;
1406 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1407 struct btrfs_device *device, u64 num_bytes,
1408 u64 *start, u64 *len)
1410 /* FIXME use last free of some kind */
1411 return find_free_dev_extent_start(trans->transaction, device,
1412 num_bytes, 0, start, len);
1415 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1416 struct btrfs_device *device,
1417 u64 start, u64 *dev_extent_len)
1420 struct btrfs_path *path;
1421 struct btrfs_root *root = device->dev_root;
1422 struct btrfs_key key;
1423 struct btrfs_key found_key;
1424 struct extent_buffer *leaf = NULL;
1425 struct btrfs_dev_extent *extent = NULL;
1427 path = btrfs_alloc_path();
1431 key.objectid = device->devid;
1433 key.type = BTRFS_DEV_EXTENT_KEY;
1435 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1437 ret = btrfs_previous_item(root, path, key.objectid,
1438 BTRFS_DEV_EXTENT_KEY);
1441 leaf = path->nodes[0];
1442 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1443 extent = btrfs_item_ptr(leaf, path->slots[0],
1444 struct btrfs_dev_extent);
1445 BUG_ON(found_key.offset > start || found_key.offset +
1446 btrfs_dev_extent_length(leaf, extent) < start);
1448 btrfs_release_path(path);
1450 } else if (ret == 0) {
1451 leaf = path->nodes[0];
1452 extent = btrfs_item_ptr(leaf, path->slots[0],
1453 struct btrfs_dev_extent);
1455 btrfs_std_error(root->fs_info, ret, "Slot search failed");
1459 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1461 ret = btrfs_del_item(trans, root, path);
1463 btrfs_std_error(root->fs_info, ret,
1464 "Failed to remove dev extent item");
1466 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1469 btrfs_free_path(path);
1473 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1474 struct btrfs_device *device,
1475 u64 chunk_tree, u64 chunk_objectid,
1476 u64 chunk_offset, u64 start, u64 num_bytes)
1479 struct btrfs_path *path;
1480 struct btrfs_root *root = device->dev_root;
1481 struct btrfs_dev_extent *extent;
1482 struct extent_buffer *leaf;
1483 struct btrfs_key key;
1485 WARN_ON(!device->in_fs_metadata);
1486 WARN_ON(device->is_tgtdev_for_dev_replace);
1487 path = btrfs_alloc_path();
1491 key.objectid = device->devid;
1493 key.type = BTRFS_DEV_EXTENT_KEY;
1494 ret = btrfs_insert_empty_item(trans, root, path, &key,
1499 leaf = path->nodes[0];
1500 extent = btrfs_item_ptr(leaf, path->slots[0],
1501 struct btrfs_dev_extent);
1502 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1503 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1504 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1506 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1507 btrfs_dev_extent_chunk_tree_uuid(extent), BTRFS_UUID_SIZE);
1509 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1510 btrfs_mark_buffer_dirty(leaf);
1512 btrfs_free_path(path);
1516 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1518 struct extent_map_tree *em_tree;
1519 struct extent_map *em;
1523 em_tree = &fs_info->mapping_tree.map_tree;
1524 read_lock(&em_tree->lock);
1525 n = rb_last(&em_tree->map);
1527 em = rb_entry(n, struct extent_map, rb_node);
1528 ret = em->start + em->len;
1530 read_unlock(&em_tree->lock);
1535 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1539 struct btrfs_key key;
1540 struct btrfs_key found_key;
1541 struct btrfs_path *path;
1543 path = btrfs_alloc_path();
1547 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1548 key.type = BTRFS_DEV_ITEM_KEY;
1549 key.offset = (u64)-1;
1551 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1555 BUG_ON(ret == 0); /* Corruption */
1557 ret = btrfs_previous_item(fs_info->chunk_root, path,
1558 BTRFS_DEV_ITEMS_OBJECTID,
1559 BTRFS_DEV_ITEM_KEY);
1563 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1565 *devid_ret = found_key.offset + 1;
1569 btrfs_free_path(path);
1574 * the device information is stored in the chunk root
1575 * the btrfs_device struct should be fully filled in
1577 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1578 struct btrfs_root *root,
1579 struct btrfs_device *device)
1582 struct btrfs_path *path;
1583 struct btrfs_dev_item *dev_item;
1584 struct extent_buffer *leaf;
1585 struct btrfs_key key;
1588 root = root->fs_info->chunk_root;
1590 path = btrfs_alloc_path();
1594 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1595 key.type = BTRFS_DEV_ITEM_KEY;
1596 key.offset = device->devid;
1598 ret = btrfs_insert_empty_item(trans, root, path, &key,
1603 leaf = path->nodes[0];
1604 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1606 btrfs_set_device_id(leaf, dev_item, device->devid);
1607 btrfs_set_device_generation(leaf, dev_item, 0);
1608 btrfs_set_device_type(leaf, dev_item, device->type);
1609 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1610 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1611 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1612 btrfs_set_device_total_bytes(leaf, dev_item,
1613 btrfs_device_get_disk_total_bytes(device));
1614 btrfs_set_device_bytes_used(leaf, dev_item,
1615 btrfs_device_get_bytes_used(device));
1616 btrfs_set_device_group(leaf, dev_item, 0);
1617 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1618 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1619 btrfs_set_device_start_offset(leaf, dev_item, 0);
1621 ptr = btrfs_device_uuid(dev_item);
1622 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1623 ptr = btrfs_device_fsid(dev_item);
1624 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1625 btrfs_mark_buffer_dirty(leaf);
1629 btrfs_free_path(path);
1634 * Function to update ctime/mtime for a given device path.
1635 * Mainly used for ctime/mtime based probe like libblkid.
1637 static void update_dev_time(char *path_name)
1641 filp = filp_open(path_name, O_RDWR, 0);
1644 file_update_time(filp);
1645 filp_close(filp, NULL);
1649 static int btrfs_rm_dev_item(struct btrfs_root *root,
1650 struct btrfs_device *device)
1653 struct btrfs_path *path;
1654 struct btrfs_key key;
1655 struct btrfs_trans_handle *trans;
1657 root = root->fs_info->chunk_root;
1659 path = btrfs_alloc_path();
1663 trans = btrfs_start_transaction(root, 0);
1664 if (IS_ERR(trans)) {
1665 btrfs_free_path(path);
1666 return PTR_ERR(trans);
1668 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1669 key.type = BTRFS_DEV_ITEM_KEY;
1670 key.offset = device->devid;
1672 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1681 ret = btrfs_del_item(trans, root, path);
1685 btrfs_free_path(path);
1686 btrfs_commit_transaction(trans, root);
1690 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1692 struct btrfs_device *device;
1693 struct btrfs_device *next_device;
1694 struct block_device *bdev;
1695 struct buffer_head *bh = NULL;
1696 struct btrfs_super_block *disk_super;
1697 struct btrfs_fs_devices *cur_devices;
1704 bool clear_super = false;
1706 mutex_lock(&uuid_mutex);
1709 seq = read_seqbegin(&root->fs_info->profiles_lock);
1711 all_avail = root->fs_info->avail_data_alloc_bits |
1712 root->fs_info->avail_system_alloc_bits |
1713 root->fs_info->avail_metadata_alloc_bits;
1714 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
1716 num_devices = root->fs_info->fs_devices->num_devices;
1717 btrfs_dev_replace_lock(&root->fs_info->dev_replace);
1718 if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1719 WARN_ON(num_devices < 1);
1722 btrfs_dev_replace_unlock(&root->fs_info->dev_replace);
1724 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && num_devices <= 4) {
1725 ret = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET;
1729 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && num_devices <= 2) {
1730 ret = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET;
1734 if ((all_avail & BTRFS_BLOCK_GROUP_RAID5) &&
1735 root->fs_info->fs_devices->rw_devices <= 2) {
1736 ret = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET;
1739 if ((all_avail & BTRFS_BLOCK_GROUP_RAID6) &&
1740 root->fs_info->fs_devices->rw_devices <= 3) {
1741 ret = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET;
1745 if (strcmp(device_path, "missing") == 0) {
1746 struct list_head *devices;
1747 struct btrfs_device *tmp;
1750 devices = &root->fs_info->fs_devices->devices;
1752 * It is safe to read the devices since the volume_mutex
1755 list_for_each_entry(tmp, devices, dev_list) {
1756 if (tmp->in_fs_metadata &&
1757 !tmp->is_tgtdev_for_dev_replace &&
1767 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
1771 ret = btrfs_get_bdev_and_sb(device_path,
1772 FMODE_WRITE | FMODE_EXCL,
1773 root->fs_info->bdev_holder, 0,
1777 disk_super = (struct btrfs_super_block *)bh->b_data;
1778 devid = btrfs_stack_device_id(&disk_super->dev_item);
1779 dev_uuid = disk_super->dev_item.uuid;
1780 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1788 if (device->is_tgtdev_for_dev_replace) {
1789 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1793 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1794 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1798 if (device->writeable) {
1800 list_del_init(&device->dev_alloc_list);
1801 device->fs_devices->rw_devices--;
1802 unlock_chunks(root);
1806 mutex_unlock(&uuid_mutex);
1807 ret = btrfs_shrink_device(device, 0);
1808 mutex_lock(&uuid_mutex);
1813 * TODO: the superblock still includes this device in its num_devices
1814 * counter although write_all_supers() is not locked out. This
1815 * could give a filesystem state which requires a degraded mount.
1817 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1821 device->in_fs_metadata = 0;
1822 btrfs_scrub_cancel_dev(root->fs_info, device);
1825 * the device list mutex makes sure that we don't change
1826 * the device list while someone else is writing out all
1827 * the device supers. Whoever is writing all supers, should
1828 * lock the device list mutex before getting the number of
1829 * devices in the super block (super_copy). Conversely,
1830 * whoever updates the number of devices in the super block
1831 * (super_copy) should hold the device list mutex.
1834 cur_devices = device->fs_devices;
1835 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1836 list_del_rcu(&device->dev_list);
1838 device->fs_devices->num_devices--;
1839 device->fs_devices->total_devices--;
1841 if (device->missing)
1842 device->fs_devices->missing_devices--;
1844 next_device = list_entry(root->fs_info->fs_devices->devices.next,
1845 struct btrfs_device, dev_list);
1846 if (device->bdev == root->fs_info->sb->s_bdev)
1847 root->fs_info->sb->s_bdev = next_device->bdev;
1848 if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1849 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1852 device->fs_devices->open_devices--;
1853 /* remove sysfs entry */
1854 btrfs_sysfs_rm_device_link(root->fs_info->fs_devices, device);
1857 call_rcu(&device->rcu, free_device);
1859 num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1860 btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1861 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1863 if (cur_devices->open_devices == 0) {
1864 struct btrfs_fs_devices *fs_devices;
1865 fs_devices = root->fs_info->fs_devices;
1866 while (fs_devices) {
1867 if (fs_devices->seed == cur_devices) {
1868 fs_devices->seed = cur_devices->seed;
1871 fs_devices = fs_devices->seed;
1873 cur_devices->seed = NULL;
1874 __btrfs_close_devices(cur_devices);
1875 free_fs_devices(cur_devices);
1878 root->fs_info->num_tolerated_disk_barrier_failures =
1879 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1882 * at this point, the device is zero sized. We want to
1883 * remove it from the devices list and zero out the old super
1885 if (clear_super && disk_super) {
1889 /* make sure this device isn't detected as part of
1892 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1893 set_buffer_dirty(bh);
1894 sync_dirty_buffer(bh);
1896 /* clear the mirror copies of super block on the disk
1897 * being removed, 0th copy is been taken care above and
1898 * the below would take of the rest
1900 for (i = 1; i < BTRFS_SUPER_MIRROR_MAX; i++) {
1901 bytenr = btrfs_sb_offset(i);
1902 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
1903 i_size_read(bdev->bd_inode))
1907 bh = __bread(bdev, bytenr / 4096,
1908 BTRFS_SUPER_INFO_SIZE);
1912 disk_super = (struct btrfs_super_block *)bh->b_data;
1914 if (btrfs_super_bytenr(disk_super) != bytenr ||
1915 btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
1918 memset(&disk_super->magic, 0,
1919 sizeof(disk_super->magic));
1920 set_buffer_dirty(bh);
1921 sync_dirty_buffer(bh);
1928 /* Notify udev that device has changed */
1929 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
1931 /* Update ctime/mtime for device path for libblkid */
1932 update_dev_time(device_path);
1938 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1940 mutex_unlock(&uuid_mutex);
1943 if (device->writeable) {
1945 list_add(&device->dev_alloc_list,
1946 &root->fs_info->fs_devices->alloc_list);
1947 device->fs_devices->rw_devices++;
1948 unlock_chunks(root);
1953 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
1954 struct btrfs_device *srcdev)
1956 struct btrfs_fs_devices *fs_devices;
1958 WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1961 * in case of fs with no seed, srcdev->fs_devices will point
1962 * to fs_devices of fs_info. However when the dev being replaced is
1963 * a seed dev it will point to the seed's local fs_devices. In short
1964 * srcdev will have its correct fs_devices in both the cases.
1966 fs_devices = srcdev->fs_devices;
1968 list_del_rcu(&srcdev->dev_list);
1969 list_del_rcu(&srcdev->dev_alloc_list);
1970 fs_devices->num_devices--;
1971 if (srcdev->missing)
1972 fs_devices->missing_devices--;
1974 if (srcdev->writeable) {
1975 fs_devices->rw_devices--;
1976 /* zero out the old super if it is writable */
1977 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
1981 fs_devices->open_devices--;
1984 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
1985 struct btrfs_device *srcdev)
1987 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
1989 call_rcu(&srcdev->rcu, free_device);
1992 * unless fs_devices is seed fs, num_devices shouldn't go
1995 BUG_ON(!fs_devices->num_devices && !fs_devices->seeding);
1997 /* if this is no devs we rather delete the fs_devices */
1998 if (!fs_devices->num_devices) {
1999 struct btrfs_fs_devices *tmp_fs_devices;
2001 tmp_fs_devices = fs_info->fs_devices;
2002 while (tmp_fs_devices) {
2003 if (tmp_fs_devices->seed == fs_devices) {
2004 tmp_fs_devices->seed = fs_devices->seed;
2007 tmp_fs_devices = tmp_fs_devices->seed;
2009 fs_devices->seed = NULL;
2010 __btrfs_close_devices(fs_devices);
2011 free_fs_devices(fs_devices);
2015 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2016 struct btrfs_device *tgtdev)
2018 struct btrfs_device *next_device;
2020 mutex_lock(&uuid_mutex);
2022 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2024 btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev);
2027 btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
2028 fs_info->fs_devices->open_devices--;
2030 fs_info->fs_devices->num_devices--;
2032 next_device = list_entry(fs_info->fs_devices->devices.next,
2033 struct btrfs_device, dev_list);
2034 if (tgtdev->bdev == fs_info->sb->s_bdev)
2035 fs_info->sb->s_bdev = next_device->bdev;
2036 if (tgtdev->bdev == fs_info->fs_devices->latest_bdev)
2037 fs_info->fs_devices->latest_bdev = next_device->bdev;
2038 list_del_rcu(&tgtdev->dev_list);
2040 call_rcu(&tgtdev->rcu, free_device);
2042 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2043 mutex_unlock(&uuid_mutex);
2046 static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
2047 struct btrfs_device **device)
2050 struct btrfs_super_block *disk_super;
2053 struct block_device *bdev;
2054 struct buffer_head *bh;
2057 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2058 root->fs_info->bdev_holder, 0, &bdev, &bh);
2061 disk_super = (struct btrfs_super_block *)bh->b_data;
2062 devid = btrfs_stack_device_id(&disk_super->dev_item);
2063 dev_uuid = disk_super->dev_item.uuid;
2064 *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2069 blkdev_put(bdev, FMODE_READ);
2073 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
2075 struct btrfs_device **device)
2078 if (strcmp(device_path, "missing") == 0) {
2079 struct list_head *devices;
2080 struct btrfs_device *tmp;
2082 devices = &root->fs_info->fs_devices->devices;
2084 * It is safe to read the devices since the volume_mutex
2085 * is held by the caller.
2087 list_for_each_entry(tmp, devices, dev_list) {
2088 if (tmp->in_fs_metadata && !tmp->bdev) {
2095 return BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2099 return btrfs_find_device_by_path(root, device_path, device);
2104 * does all the dirty work required for changing file system's UUID.
2106 static int btrfs_prepare_sprout(struct btrfs_root *root)
2108 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2109 struct btrfs_fs_devices *old_devices;
2110 struct btrfs_fs_devices *seed_devices;
2111 struct btrfs_super_block *disk_super = root->fs_info->super_copy;
2112 struct btrfs_device *device;
2115 BUG_ON(!mutex_is_locked(&uuid_mutex));
2116 if (!fs_devices->seeding)
2119 seed_devices = __alloc_fs_devices();
2120 if (IS_ERR(seed_devices))
2121 return PTR_ERR(seed_devices);
2123 old_devices = clone_fs_devices(fs_devices);
2124 if (IS_ERR(old_devices)) {
2125 kfree(seed_devices);
2126 return PTR_ERR(old_devices);
2129 list_add(&old_devices->list, &fs_uuids);
2131 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2132 seed_devices->opened = 1;
2133 INIT_LIST_HEAD(&seed_devices->devices);
2134 INIT_LIST_HEAD(&seed_devices->alloc_list);
2135 mutex_init(&seed_devices->device_list_mutex);
2137 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2138 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2140 list_for_each_entry(device, &seed_devices->devices, dev_list)
2141 device->fs_devices = seed_devices;
2144 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2145 unlock_chunks(root);
2147 fs_devices->seeding = 0;
2148 fs_devices->num_devices = 0;
2149 fs_devices->open_devices = 0;
2150 fs_devices->missing_devices = 0;
2151 fs_devices->rotating = 0;
2152 fs_devices->seed = seed_devices;
2154 generate_random_uuid(fs_devices->fsid);
2155 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2156 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2157 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2159 super_flags = btrfs_super_flags(disk_super) &
2160 ~BTRFS_SUPER_FLAG_SEEDING;
2161 btrfs_set_super_flags(disk_super, super_flags);
2167 * strore the expected generation for seed devices in device items.
2169 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2170 struct btrfs_root *root)
2172 struct btrfs_path *path;
2173 struct extent_buffer *leaf;
2174 struct btrfs_dev_item *dev_item;
2175 struct btrfs_device *device;
2176 struct btrfs_key key;
2177 u8 fs_uuid[BTRFS_UUID_SIZE];
2178 u8 dev_uuid[BTRFS_UUID_SIZE];
2182 path = btrfs_alloc_path();
2186 root = root->fs_info->chunk_root;
2187 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2189 key.type = BTRFS_DEV_ITEM_KEY;
2192 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2196 leaf = path->nodes[0];
2198 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2199 ret = btrfs_next_leaf(root, path);
2204 leaf = path->nodes[0];
2205 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2206 btrfs_release_path(path);
2210 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2211 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2212 key.type != BTRFS_DEV_ITEM_KEY)
2215 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2216 struct btrfs_dev_item);
2217 devid = btrfs_device_id(leaf, dev_item);
2218 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2220 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2222 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2224 BUG_ON(!device); /* Logic error */
2226 if (device->fs_devices->seeding) {
2227 btrfs_set_device_generation(leaf, dev_item,
2228 device->generation);
2229 btrfs_mark_buffer_dirty(leaf);
2237 btrfs_free_path(path);
2241 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
2243 struct request_queue *q;
2244 struct btrfs_trans_handle *trans;
2245 struct btrfs_device *device;
2246 struct block_device *bdev;
2247 struct list_head *devices;
2248 struct super_block *sb = root->fs_info->sb;
2249 struct rcu_string *name;
2251 int seeding_dev = 0;
2254 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
2257 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2258 root->fs_info->bdev_holder);
2260 return PTR_ERR(bdev);
2262 if (root->fs_info->fs_devices->seeding) {
2264 down_write(&sb->s_umount);
2265 mutex_lock(&uuid_mutex);
2268 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2270 devices = &root->fs_info->fs_devices->devices;
2272 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2273 list_for_each_entry(device, devices, dev_list) {
2274 if (device->bdev == bdev) {
2277 &root->fs_info->fs_devices->device_list_mutex);
2281 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2283 device = btrfs_alloc_device(root->fs_info, NULL, NULL);
2284 if (IS_ERR(device)) {
2285 /* we can safely leave the fs_devices entry around */
2286 ret = PTR_ERR(device);
2290 name = rcu_string_strdup(device_path, GFP_NOFS);
2296 rcu_assign_pointer(device->name, name);
2298 trans = btrfs_start_transaction(root, 0);
2299 if (IS_ERR(trans)) {
2300 rcu_string_free(device->name);
2302 ret = PTR_ERR(trans);
2306 q = bdev_get_queue(bdev);
2307 if (blk_queue_discard(q))
2308 device->can_discard = 1;
2309 device->writeable = 1;
2310 device->generation = trans->transid;
2311 device->io_width = root->sectorsize;
2312 device->io_align = root->sectorsize;
2313 device->sector_size = root->sectorsize;
2314 device->total_bytes = i_size_read(bdev->bd_inode);
2315 device->disk_total_bytes = device->total_bytes;
2316 device->commit_total_bytes = device->total_bytes;
2317 device->dev_root = root->fs_info->dev_root;
2318 device->bdev = bdev;
2319 device->in_fs_metadata = 1;
2320 device->is_tgtdev_for_dev_replace = 0;
2321 device->mode = FMODE_EXCL;
2322 device->dev_stats_valid = 1;
2323 set_blocksize(device->bdev, 4096);
2326 sb->s_flags &= ~MS_RDONLY;
2327 ret = btrfs_prepare_sprout(root);
2328 BUG_ON(ret); /* -ENOMEM */
2331 device->fs_devices = root->fs_info->fs_devices;
2333 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2335 list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
2336 list_add(&device->dev_alloc_list,
2337 &root->fs_info->fs_devices->alloc_list);
2338 root->fs_info->fs_devices->num_devices++;
2339 root->fs_info->fs_devices->open_devices++;
2340 root->fs_info->fs_devices->rw_devices++;
2341 root->fs_info->fs_devices->total_devices++;
2342 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2344 spin_lock(&root->fs_info->free_chunk_lock);
2345 root->fs_info->free_chunk_space += device->total_bytes;
2346 spin_unlock(&root->fs_info->free_chunk_lock);
2348 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2349 root->fs_info->fs_devices->rotating = 1;
2351 tmp = btrfs_super_total_bytes(root->fs_info->super_copy);
2352 btrfs_set_super_total_bytes(root->fs_info->super_copy,
2353 tmp + device->total_bytes);
2355 tmp = btrfs_super_num_devices(root->fs_info->super_copy);
2356 btrfs_set_super_num_devices(root->fs_info->super_copy,
2359 /* add sysfs device entry */
2360 btrfs_sysfs_add_device_link(root->fs_info->fs_devices, device);
2363 * we've got more storage, clear any full flags on the space
2366 btrfs_clear_space_info_full(root->fs_info);
2368 unlock_chunks(root);
2369 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2373 ret = init_first_rw_device(trans, root, device);
2374 unlock_chunks(root);
2376 btrfs_abort_transaction(trans, root, ret);
2381 ret = btrfs_add_device(trans, root, device);
2383 btrfs_abort_transaction(trans, root, ret);
2388 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2390 ret = btrfs_finish_sprout(trans, root);
2392 btrfs_abort_transaction(trans, root, ret);
2396 /* Sprouting would change fsid of the mounted root,
2397 * so rename the fsid on the sysfs
2399 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2400 root->fs_info->fsid);
2401 if (kobject_rename(&root->fs_info->fs_devices->fsid_kobj,
2403 btrfs_warn(root->fs_info,
2404 "sysfs: failed to create fsid for sprout");
2407 root->fs_info->num_tolerated_disk_barrier_failures =
2408 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
2409 ret = btrfs_commit_transaction(trans, root);
2412 mutex_unlock(&uuid_mutex);
2413 up_write(&sb->s_umount);
2415 if (ret) /* transaction commit */
2418 ret = btrfs_relocate_sys_chunks(root);
2420 btrfs_std_error(root->fs_info, ret,
2421 "Failed to relocate sys chunks after "
2422 "device initialization. This can be fixed "
2423 "using the \"btrfs balance\" command.");
2424 trans = btrfs_attach_transaction(root);
2425 if (IS_ERR(trans)) {
2426 if (PTR_ERR(trans) == -ENOENT)
2428 return PTR_ERR(trans);
2430 ret = btrfs_commit_transaction(trans, root);
2433 /* Update ctime/mtime for libblkid */
2434 update_dev_time(device_path);
2438 btrfs_end_transaction(trans, root);
2439 rcu_string_free(device->name);
2440 btrfs_sysfs_rm_device_link(root->fs_info->fs_devices, device);
2443 blkdev_put(bdev, FMODE_EXCL);
2445 mutex_unlock(&uuid_mutex);
2446 up_write(&sb->s_umount);
2451 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2452 struct btrfs_device *srcdev,
2453 struct btrfs_device **device_out)
2455 struct request_queue *q;
2456 struct btrfs_device *device;
2457 struct block_device *bdev;
2458 struct btrfs_fs_info *fs_info = root->fs_info;
2459 struct list_head *devices;
2460 struct rcu_string *name;
2461 u64 devid = BTRFS_DEV_REPLACE_DEVID;
2465 if (fs_info->fs_devices->seeding) {
2466 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
2470 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2471 fs_info->bdev_holder);
2473 btrfs_err(fs_info, "target device %s is invalid!", device_path);
2474 return PTR_ERR(bdev);
2477 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2479 devices = &fs_info->fs_devices->devices;
2480 list_for_each_entry(device, devices, dev_list) {
2481 if (device->bdev == bdev) {
2482 btrfs_err(fs_info, "target device is in the filesystem!");
2489 if (i_size_read(bdev->bd_inode) <
2490 btrfs_device_get_total_bytes(srcdev)) {
2491 btrfs_err(fs_info, "target device is smaller than source device!");
2497 device = btrfs_alloc_device(NULL, &devid, NULL);
2498 if (IS_ERR(device)) {
2499 ret = PTR_ERR(device);
2503 name = rcu_string_strdup(device_path, GFP_NOFS);
2509 rcu_assign_pointer(device->name, name);
2511 q = bdev_get_queue(bdev);
2512 if (blk_queue_discard(q))
2513 device->can_discard = 1;
2514 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2515 device->writeable = 1;
2516 device->generation = 0;
2517 device->io_width = root->sectorsize;
2518 device->io_align = root->sectorsize;
2519 device->sector_size = root->sectorsize;
2520 device->total_bytes = btrfs_device_get_total_bytes(srcdev);
2521 device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
2522 device->bytes_used = btrfs_device_get_bytes_used(srcdev);
2523 ASSERT(list_empty(&srcdev->resized_list));
2524 device->commit_total_bytes = srcdev->commit_total_bytes;
2525 device->commit_bytes_used = device->bytes_used;
2526 device->dev_root = fs_info->dev_root;
2527 device->bdev = bdev;
2528 device->in_fs_metadata = 1;
2529 device->is_tgtdev_for_dev_replace = 1;
2530 device->mode = FMODE_EXCL;
2531 device->dev_stats_valid = 1;
2532 set_blocksize(device->bdev, 4096);
2533 device->fs_devices = fs_info->fs_devices;
2534 list_add(&device->dev_list, &fs_info->fs_devices->devices);
2535 fs_info->fs_devices->num_devices++;
2536 fs_info->fs_devices->open_devices++;
2537 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2539 *device_out = device;
2543 blkdev_put(bdev, FMODE_EXCL);
2547 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2548 struct btrfs_device *tgtdev)
2550 WARN_ON(fs_info->fs_devices->rw_devices == 0);
2551 tgtdev->io_width = fs_info->dev_root->sectorsize;
2552 tgtdev->io_align = fs_info->dev_root->sectorsize;
2553 tgtdev->sector_size = fs_info->dev_root->sectorsize;
2554 tgtdev->dev_root = fs_info->dev_root;
2555 tgtdev->in_fs_metadata = 1;
2558 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2559 struct btrfs_device *device)
2562 struct btrfs_path *path;
2563 struct btrfs_root *root;
2564 struct btrfs_dev_item *dev_item;
2565 struct extent_buffer *leaf;
2566 struct btrfs_key key;
2568 root = device->dev_root->fs_info->chunk_root;
2570 path = btrfs_alloc_path();
2574 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2575 key.type = BTRFS_DEV_ITEM_KEY;
2576 key.offset = device->devid;
2578 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2587 leaf = path->nodes[0];
2588 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2590 btrfs_set_device_id(leaf, dev_item, device->devid);
2591 btrfs_set_device_type(leaf, dev_item, device->type);
2592 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2593 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2594 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2595 btrfs_set_device_total_bytes(leaf, dev_item,
2596 btrfs_device_get_disk_total_bytes(device));
2597 btrfs_set_device_bytes_used(leaf, dev_item,
2598 btrfs_device_get_bytes_used(device));
2599 btrfs_mark_buffer_dirty(leaf);
2602 btrfs_free_path(path);
2606 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2607 struct btrfs_device *device, u64 new_size)
2609 struct btrfs_super_block *super_copy =
2610 device->dev_root->fs_info->super_copy;
2611 struct btrfs_fs_devices *fs_devices;
2615 if (!device->writeable)
2618 lock_chunks(device->dev_root);
2619 old_total = btrfs_super_total_bytes(super_copy);
2620 diff = new_size - device->total_bytes;
2622 if (new_size <= device->total_bytes ||
2623 device->is_tgtdev_for_dev_replace) {
2624 unlock_chunks(device->dev_root);
2628 fs_devices = device->dev_root->fs_info->fs_devices;
2630 btrfs_set_super_total_bytes(super_copy, old_total + diff);
2631 device->fs_devices->total_rw_bytes += diff;
2633 btrfs_device_set_total_bytes(device, new_size);
2634 btrfs_device_set_disk_total_bytes(device, new_size);
2635 btrfs_clear_space_info_full(device->dev_root->fs_info);
2636 if (list_empty(&device->resized_list))
2637 list_add_tail(&device->resized_list,
2638 &fs_devices->resized_devices);
2639 unlock_chunks(device->dev_root);
2641 return btrfs_update_device(trans, device);
2644 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2645 struct btrfs_root *root, u64 chunk_objectid,
2649 struct btrfs_path *path;
2650 struct btrfs_key key;
2652 root = root->fs_info->chunk_root;
2653 path = btrfs_alloc_path();
2657 key.objectid = chunk_objectid;
2658 key.offset = chunk_offset;
2659 key.type = BTRFS_CHUNK_ITEM_KEY;
2661 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2664 else if (ret > 0) { /* Logic error or corruption */
2665 btrfs_std_error(root->fs_info, -ENOENT,
2666 "Failed lookup while freeing chunk.");
2671 ret = btrfs_del_item(trans, root, path);
2673 btrfs_std_error(root->fs_info, ret,
2674 "Failed to delete chunk item.");
2676 btrfs_free_path(path);
2680 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2683 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2684 struct btrfs_disk_key *disk_key;
2685 struct btrfs_chunk *chunk;
2692 struct btrfs_key key;
2695 array_size = btrfs_super_sys_array_size(super_copy);
2697 ptr = super_copy->sys_chunk_array;
2700 while (cur < array_size) {
2701 disk_key = (struct btrfs_disk_key *)ptr;
2702 btrfs_disk_key_to_cpu(&key, disk_key);
2704 len = sizeof(*disk_key);
2706 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2707 chunk = (struct btrfs_chunk *)(ptr + len);
2708 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2709 len += btrfs_chunk_item_size(num_stripes);
2714 if (key.objectid == chunk_objectid &&
2715 key.offset == chunk_offset) {
2716 memmove(ptr, ptr + len, array_size - (cur + len));
2718 btrfs_set_super_sys_array_size(super_copy, array_size);
2724 unlock_chunks(root);
2728 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2729 struct btrfs_root *root, u64 chunk_offset)
2731 struct extent_map_tree *em_tree;
2732 struct extent_map *em;
2733 struct btrfs_root *extent_root = root->fs_info->extent_root;
2734 struct map_lookup *map;
2735 u64 dev_extent_len = 0;
2736 u64 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2740 root = root->fs_info->chunk_root;
2741 em_tree = &root->fs_info->mapping_tree.map_tree;
2743 read_lock(&em_tree->lock);
2744 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2745 read_unlock(&em_tree->lock);
2747 if (!em || em->start > chunk_offset ||
2748 em->start + em->len < chunk_offset) {
2750 * This is a logic error, but we don't want to just rely on the
2751 * user having built with ASSERT enabled, so if ASSERT doens't
2752 * do anything we still error out.
2756 free_extent_map(em);
2759 map = (struct map_lookup *)em->bdev;
2760 lock_chunks(root->fs_info->chunk_root);
2761 check_system_chunk(trans, extent_root, map->type);
2762 unlock_chunks(root->fs_info->chunk_root);
2764 for (i = 0; i < map->num_stripes; i++) {
2765 struct btrfs_device *device = map->stripes[i].dev;
2766 ret = btrfs_free_dev_extent(trans, device,
2767 map->stripes[i].physical,
2770 btrfs_abort_transaction(trans, root, ret);
2774 if (device->bytes_used > 0) {
2776 btrfs_device_set_bytes_used(device,
2777 device->bytes_used - dev_extent_len);
2778 spin_lock(&root->fs_info->free_chunk_lock);
2779 root->fs_info->free_chunk_space += dev_extent_len;
2780 spin_unlock(&root->fs_info->free_chunk_lock);
2781 btrfs_clear_space_info_full(root->fs_info);
2782 unlock_chunks(root);
2785 if (map->stripes[i].dev) {
2786 ret = btrfs_update_device(trans, map->stripes[i].dev);
2788 btrfs_abort_transaction(trans, root, ret);
2793 ret = btrfs_free_chunk(trans, root, chunk_objectid, chunk_offset);
2795 btrfs_abort_transaction(trans, root, ret);
2799 trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2801 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2802 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2804 btrfs_abort_transaction(trans, root, ret);
2809 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset, em);
2811 btrfs_abort_transaction(trans, extent_root, ret);
2817 free_extent_map(em);
2821 static int btrfs_relocate_chunk(struct btrfs_root *root, u64 chunk_offset)
2823 struct btrfs_root *extent_root;
2824 struct btrfs_trans_handle *trans;
2827 root = root->fs_info->chunk_root;
2828 extent_root = root->fs_info->extent_root;
2831 * Prevent races with automatic removal of unused block groups.
2832 * After we relocate and before we remove the chunk with offset
2833 * chunk_offset, automatic removal of the block group can kick in,
2834 * resulting in a failure when calling btrfs_remove_chunk() below.
2836 * Make sure to acquire this mutex before doing a tree search (dev
2837 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2838 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2839 * we release the path used to search the chunk/dev tree and before
2840 * the current task acquires this mutex and calls us.
2842 ASSERT(mutex_is_locked(&root->fs_info->delete_unused_bgs_mutex));
2844 ret = btrfs_can_relocate(extent_root, chunk_offset);
2848 /* step one, relocate all the extents inside this chunk */
2849 btrfs_scrub_pause(root);
2850 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2851 btrfs_scrub_continue(root);
2855 trans = btrfs_start_trans_remove_block_group(root->fs_info,
2857 if (IS_ERR(trans)) {
2858 ret = PTR_ERR(trans);
2859 btrfs_std_error(root->fs_info, ret, NULL);
2864 * step two, delete the device extents and the
2865 * chunk tree entries
2867 ret = btrfs_remove_chunk(trans, root, chunk_offset);
2868 btrfs_end_transaction(trans, root);
2872 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2874 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2875 struct btrfs_path *path;
2876 struct extent_buffer *leaf;
2877 struct btrfs_chunk *chunk;
2878 struct btrfs_key key;
2879 struct btrfs_key found_key;
2881 bool retried = false;
2885 path = btrfs_alloc_path();
2890 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2891 key.offset = (u64)-1;
2892 key.type = BTRFS_CHUNK_ITEM_KEY;
2895 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
2896 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2898 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2901 BUG_ON(ret == 0); /* Corruption */
2903 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2906 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2912 leaf = path->nodes[0];
2913 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2915 chunk = btrfs_item_ptr(leaf, path->slots[0],
2916 struct btrfs_chunk);
2917 chunk_type = btrfs_chunk_type(leaf, chunk);
2918 btrfs_release_path(path);
2920 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2921 ret = btrfs_relocate_chunk(chunk_root,
2928 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2930 if (found_key.offset == 0)
2932 key.offset = found_key.offset - 1;
2935 if (failed && !retried) {
2939 } else if (WARN_ON(failed && retried)) {
2943 btrfs_free_path(path);
2947 static int insert_balance_item(struct btrfs_root *root,
2948 struct btrfs_balance_control *bctl)
2950 struct btrfs_trans_handle *trans;
2951 struct btrfs_balance_item *item;
2952 struct btrfs_disk_balance_args disk_bargs;
2953 struct btrfs_path *path;
2954 struct extent_buffer *leaf;
2955 struct btrfs_key key;
2958 path = btrfs_alloc_path();
2962 trans = btrfs_start_transaction(root, 0);
2963 if (IS_ERR(trans)) {
2964 btrfs_free_path(path);
2965 return PTR_ERR(trans);
2968 key.objectid = BTRFS_BALANCE_OBJECTID;
2969 key.type = BTRFS_BALANCE_ITEM_KEY;
2972 ret = btrfs_insert_empty_item(trans, root, path, &key,
2977 leaf = path->nodes[0];
2978 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2980 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2982 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2983 btrfs_set_balance_data(leaf, item, &disk_bargs);
2984 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2985 btrfs_set_balance_meta(leaf, item, &disk_bargs);
2986 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2987 btrfs_set_balance_sys(leaf, item, &disk_bargs);
2989 btrfs_set_balance_flags(leaf, item, bctl->flags);
2991 btrfs_mark_buffer_dirty(leaf);
2993 btrfs_free_path(path);
2994 err = btrfs_commit_transaction(trans, root);
3000 static int del_balance_item(struct btrfs_root *root)
3002 struct btrfs_trans_handle *trans;
3003 struct btrfs_path *path;
3004 struct btrfs_key key;
3007 path = btrfs_alloc_path();
3011 trans = btrfs_start_transaction(root, 0);
3012 if (IS_ERR(trans)) {
3013 btrfs_free_path(path);
3014 return PTR_ERR(trans);
3017 key.objectid = BTRFS_BALANCE_OBJECTID;
3018 key.type = BTRFS_BALANCE_ITEM_KEY;
3021 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3029 ret = btrfs_del_item(trans, root, path);
3031 btrfs_free_path(path);
3032 err = btrfs_commit_transaction(trans, root);
3039 * This is a heuristic used to reduce the number of chunks balanced on
3040 * resume after balance was interrupted.
3042 static void update_balance_args(struct btrfs_balance_control *bctl)
3045 * Turn on soft mode for chunk types that were being converted.
3047 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3048 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3049 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3050 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3051 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3052 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3055 * Turn on usage filter if is not already used. The idea is
3056 * that chunks that we have already balanced should be
3057 * reasonably full. Don't do it for chunks that are being
3058 * converted - that will keep us from relocating unconverted
3059 * (albeit full) chunks.
3061 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3062 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3063 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3064 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3065 bctl->data.usage = 90;
3067 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3068 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3069 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3070 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3071 bctl->sys.usage = 90;
3073 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3074 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3075 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3076 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3077 bctl->meta.usage = 90;
3082 * Should be called with both balance and volume mutexes held to
3083 * serialize other volume operations (add_dev/rm_dev/resize) with
3084 * restriper. Same goes for unset_balance_control.
3086 static void set_balance_control(struct btrfs_balance_control *bctl)
3088 struct btrfs_fs_info *fs_info = bctl->fs_info;
3090 BUG_ON(fs_info->balance_ctl);
3092 spin_lock(&fs_info->balance_lock);
3093 fs_info->balance_ctl = bctl;
3094 spin_unlock(&fs_info->balance_lock);
3097 static void unset_balance_control(struct btrfs_fs_info *fs_info)
3099 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3101 BUG_ON(!fs_info->balance_ctl);
3103 spin_lock(&fs_info->balance_lock);
3104 fs_info->balance_ctl = NULL;
3105 spin_unlock(&fs_info->balance_lock);
3111 * Balance filters. Return 1 if chunk should be filtered out
3112 * (should not be balanced).
3114 static int chunk_profiles_filter(u64 chunk_type,
3115 struct btrfs_balance_args *bargs)
3117 chunk_type = chunk_to_extended(chunk_type) &
3118 BTRFS_EXTENDED_PROFILE_MASK;
3120 if (bargs->profiles & chunk_type)
3126 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3127 struct btrfs_balance_args *bargs)
3129 struct btrfs_block_group_cache *cache;
3131 u64 user_thresh_min;
3132 u64 user_thresh_max;
3135 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3136 chunk_used = btrfs_block_group_used(&cache->item);
3138 if (bargs->usage_min == 0)
3139 user_thresh_min = 0;
3141 user_thresh_min = div_factor_fine(cache->key.offset,
3144 if (bargs->usage_max == 0)
3145 user_thresh_max = 1;
3146 else if (bargs->usage_max > 100)
3147 user_thresh_max = cache->key.offset;
3149 user_thresh_max = div_factor_fine(cache->key.offset,
3152 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3155 btrfs_put_block_group(cache);
3159 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3160 u64 chunk_offset, struct btrfs_balance_args *bargs)
3162 struct btrfs_block_group_cache *cache;
3163 u64 chunk_used, user_thresh;
3166 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3167 chunk_used = btrfs_block_group_used(&cache->item);
3169 if (bargs->usage_min == 0)
3171 else if (bargs->usage > 100)
3172 user_thresh = cache->key.offset;
3174 user_thresh = div_factor_fine(cache->key.offset,
3177 if (chunk_used < user_thresh)
3180 btrfs_put_block_group(cache);
3184 static int chunk_devid_filter(struct extent_buffer *leaf,
3185 struct btrfs_chunk *chunk,
3186 struct btrfs_balance_args *bargs)
3188 struct btrfs_stripe *stripe;
3189 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3192 for (i = 0; i < num_stripes; i++) {
3193 stripe = btrfs_stripe_nr(chunk, i);
3194 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3201 /* [pstart, pend) */
3202 static int chunk_drange_filter(struct extent_buffer *leaf,
3203 struct btrfs_chunk *chunk,
3205 struct btrfs_balance_args *bargs)
3207 struct btrfs_stripe *stripe;
3208 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3214 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3217 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3218 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3219 factor = num_stripes / 2;
3220 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3221 factor = num_stripes - 1;
3222 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3223 factor = num_stripes - 2;
3225 factor = num_stripes;
3228 for (i = 0; i < num_stripes; i++) {
3229 stripe = btrfs_stripe_nr(chunk, i);
3230 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3233 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3234 stripe_length = btrfs_chunk_length(leaf, chunk);
3235 stripe_length = div_u64(stripe_length, factor);
3237 if (stripe_offset < bargs->pend &&
3238 stripe_offset + stripe_length > bargs->pstart)
3245 /* [vstart, vend) */
3246 static int chunk_vrange_filter(struct extent_buffer *leaf,
3247 struct btrfs_chunk *chunk,
3249 struct btrfs_balance_args *bargs)
3251 if (chunk_offset < bargs->vend &&
3252 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3253 /* at least part of the chunk is inside this vrange */
3259 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3260 struct btrfs_chunk *chunk,
3261 struct btrfs_balance_args *bargs)
3263 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3265 if (bargs->stripes_min <= num_stripes
3266 && num_stripes <= bargs->stripes_max)
3272 static int chunk_soft_convert_filter(u64 chunk_type,
3273 struct btrfs_balance_args *bargs)
3275 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3278 chunk_type = chunk_to_extended(chunk_type) &
3279 BTRFS_EXTENDED_PROFILE_MASK;
3281 if (bargs->target == chunk_type)
3287 static int should_balance_chunk(struct btrfs_root *root,
3288 struct extent_buffer *leaf,
3289 struct btrfs_chunk *chunk, u64 chunk_offset)
3291 struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
3292 struct btrfs_balance_args *bargs = NULL;
3293 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3296 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3297 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3301 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3302 bargs = &bctl->data;
3303 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3305 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3306 bargs = &bctl->meta;
3308 /* profiles filter */
3309 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3310 chunk_profiles_filter(chunk_type, bargs)) {
3315 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3316 chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
3318 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3319 chunk_usage_range_filter(bctl->fs_info, chunk_offset, bargs)) {
3324 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3325 chunk_devid_filter(leaf, chunk, bargs)) {
3329 /* drange filter, makes sense only with devid filter */
3330 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3331 chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
3336 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3337 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3341 /* stripes filter */
3342 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3343 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3347 /* soft profile changing mode */
3348 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3349 chunk_soft_convert_filter(chunk_type, bargs)) {
3354 * limited by count, must be the last filter
3356 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3357 if (bargs->limit == 0)
3361 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3363 * Same logic as the 'limit' filter; the minimum cannot be
3364 * determined here because we do not have the global informatoin
3365 * about the count of all chunks that satisfy the filters.
3367 if (bargs->limit_max == 0)
3376 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3378 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3379 struct btrfs_root *chunk_root = fs_info->chunk_root;
3380 struct btrfs_root *dev_root = fs_info->dev_root;
3381 struct list_head *devices;
3382 struct btrfs_device *device;
3386 struct btrfs_chunk *chunk;
3387 struct btrfs_path *path;
3388 struct btrfs_key key;
3389 struct btrfs_key found_key;
3390 struct btrfs_trans_handle *trans;
3391 struct extent_buffer *leaf;
3394 int enospc_errors = 0;
3395 bool counting = true;
3396 /* The single value limit and min/max limits use the same bytes in the */
3397 u64 limit_data = bctl->data.limit;
3398 u64 limit_meta = bctl->meta.limit;
3399 u64 limit_sys = bctl->sys.limit;
3403 int chunk_reserved = 0;
3405 /* step one make some room on all the devices */
3406 devices = &fs_info->fs_devices->devices;
3407 list_for_each_entry(device, devices, dev_list) {
3408 old_size = btrfs_device_get_total_bytes(device);
3409 size_to_free = div_factor(old_size, 1);
3410 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
3411 if (!device->writeable ||
3412 btrfs_device_get_total_bytes(device) -
3413 btrfs_device_get_bytes_used(device) > size_to_free ||
3414 device->is_tgtdev_for_dev_replace)
3417 ret = btrfs_shrink_device(device, old_size - size_to_free);
3422 trans = btrfs_start_transaction(dev_root, 0);
3423 BUG_ON(IS_ERR(trans));
3425 ret = btrfs_grow_device(trans, device, old_size);
3428 btrfs_end_transaction(trans, dev_root);
3431 /* step two, relocate all the chunks */
3432 path = btrfs_alloc_path();
3438 /* zero out stat counters */
3439 spin_lock(&fs_info->balance_lock);
3440 memset(&bctl->stat, 0, sizeof(bctl->stat));
3441 spin_unlock(&fs_info->balance_lock);
3445 * The single value limit and min/max limits use the same bytes
3448 bctl->data.limit = limit_data;
3449 bctl->meta.limit = limit_meta;
3450 bctl->sys.limit = limit_sys;
3452 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3453 key.offset = (u64)-1;
3454 key.type = BTRFS_CHUNK_ITEM_KEY;
3457 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3458 atomic_read(&fs_info->balance_cancel_req)) {
3463 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3464 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3466 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3471 * this shouldn't happen, it means the last relocate
3475 BUG(); /* FIXME break ? */
3477 ret = btrfs_previous_item(chunk_root, path, 0,
3478 BTRFS_CHUNK_ITEM_KEY);
3480 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3485 leaf = path->nodes[0];
3486 slot = path->slots[0];
3487 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3489 if (found_key.objectid != key.objectid) {
3490 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3494 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3495 chunk_type = btrfs_chunk_type(leaf, chunk);
3498 spin_lock(&fs_info->balance_lock);
3499 bctl->stat.considered++;
3500 spin_unlock(&fs_info->balance_lock);
3503 ret = should_balance_chunk(chunk_root, leaf, chunk,
3506 btrfs_release_path(path);
3508 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3513 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3514 spin_lock(&fs_info->balance_lock);
3515 bctl->stat.expected++;
3516 spin_unlock(&fs_info->balance_lock);
3518 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3520 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3522 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3529 * Apply limit_min filter, no need to check if the LIMITS
3530 * filter is used, limit_min is 0 by default
3532 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3533 count_data < bctl->data.limit_min)
3534 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3535 count_meta < bctl->meta.limit_min)
3536 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3537 count_sys < bctl->sys.limit_min)) {
3538 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3542 if ((chunk_type & BTRFS_BLOCK_GROUP_DATA) && !chunk_reserved) {
3543 trans = btrfs_start_transaction(chunk_root, 0);
3544 if (IS_ERR(trans)) {
3545 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3546 ret = PTR_ERR(trans);
3550 ret = btrfs_force_chunk_alloc(trans, chunk_root,
3551 BTRFS_BLOCK_GROUP_DATA);
3552 btrfs_end_transaction(trans, chunk_root);
3554 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3560 ret = btrfs_relocate_chunk(chunk_root,
3562 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3563 if (ret && ret != -ENOSPC)
3565 if (ret == -ENOSPC) {
3568 spin_lock(&fs_info->balance_lock);
3569 bctl->stat.completed++;
3570 spin_unlock(&fs_info->balance_lock);
3573 if (found_key.offset == 0)
3575 key.offset = found_key.offset - 1;
3579 btrfs_release_path(path);
3584 btrfs_free_path(path);
3585 if (enospc_errors) {
3586 btrfs_info(fs_info, "%d enospc errors during balance",
3596 * alloc_profile_is_valid - see if a given profile is valid and reduced
3597 * @flags: profile to validate
3598 * @extended: if true @flags is treated as an extended profile
3600 static int alloc_profile_is_valid(u64 flags, int extended)
3602 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3603 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3605 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3607 /* 1) check that all other bits are zeroed */
3611 /* 2) see if profile is reduced */
3613 return !extended; /* "0" is valid for usual profiles */
3615 /* true if exactly one bit set */
3616 return (flags & (flags - 1)) == 0;
3619 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3621 /* cancel requested || normal exit path */
3622 return atomic_read(&fs_info->balance_cancel_req) ||
3623 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3624 atomic_read(&fs_info->balance_cancel_req) == 0);
3627 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3631 unset_balance_control(fs_info);
3632 ret = del_balance_item(fs_info->tree_root);
3634 btrfs_std_error(fs_info, ret, NULL);
3636 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3639 /* Non-zero return value signifies invalidity */
3640 static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg,
3643 return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3644 (!alloc_profile_is_valid(bctl_arg->target, 1) ||
3645 (bctl_arg->target & ~allowed)));
3649 * Should be called with both balance and volume mutexes held
3651 int btrfs_balance(struct btrfs_balance_control *bctl,
3652 struct btrfs_ioctl_balance_args *bargs)
3654 struct btrfs_fs_info *fs_info = bctl->fs_info;
3661 if (btrfs_fs_closing(fs_info) ||
3662 atomic_read(&fs_info->balance_pause_req) ||
3663 atomic_read(&fs_info->balance_cancel_req)) {
3668 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3669 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3673 * In case of mixed groups both data and meta should be picked,
3674 * and identical options should be given for both of them.
3676 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3677 if (mixed && (bctl->flags & allowed)) {
3678 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3679 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3680 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3681 btrfs_err(fs_info, "with mixed groups data and "
3682 "metadata balance options must be the same");
3688 num_devices = fs_info->fs_devices->num_devices;
3689 btrfs_dev_replace_lock(&fs_info->dev_replace);
3690 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3691 BUG_ON(num_devices < 1);
3694 btrfs_dev_replace_unlock(&fs_info->dev_replace);
3695 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3696 if (num_devices == 1)
3697 allowed |= BTRFS_BLOCK_GROUP_DUP;
3698 else if (num_devices > 1)
3699 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3700 if (num_devices > 2)
3701 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3702 if (num_devices > 3)
3703 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3704 BTRFS_BLOCK_GROUP_RAID6);
3705 if (validate_convert_profile(&bctl->data, allowed)) {
3706 btrfs_err(fs_info, "unable to start balance with target "
3707 "data profile %llu",
3712 if (validate_convert_profile(&bctl->meta, allowed)) {
3714 "unable to start balance with target metadata profile %llu",
3719 if (validate_convert_profile(&bctl->sys, allowed)) {
3721 "unable to start balance with target system profile %llu",
3727 /* allow dup'ed data chunks only in mixed mode */
3728 if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3729 (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) {
3730 btrfs_err(fs_info, "dup for data is not allowed");
3735 /* allow to reduce meta or sys integrity only if force set */
3736 allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3737 BTRFS_BLOCK_GROUP_RAID10 |
3738 BTRFS_BLOCK_GROUP_RAID5 |
3739 BTRFS_BLOCK_GROUP_RAID6;
3741 seq = read_seqbegin(&fs_info->profiles_lock);
3743 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3744 (fs_info->avail_system_alloc_bits & allowed) &&
3745 !(bctl->sys.target & allowed)) ||
3746 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3747 (fs_info->avail_metadata_alloc_bits & allowed) &&
3748 !(bctl->meta.target & allowed))) {
3749 if (bctl->flags & BTRFS_BALANCE_FORCE) {
3750 btrfs_info(fs_info, "force reducing metadata integrity");
3752 btrfs_err(fs_info, "balance will reduce metadata "
3753 "integrity, use force if you want this");
3758 } while (read_seqretry(&fs_info->profiles_lock, seq));
3760 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3761 fs_info->num_tolerated_disk_barrier_failures = min(
3762 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info),
3763 btrfs_get_num_tolerated_disk_barrier_failures(
3767 ret = insert_balance_item(fs_info->tree_root, bctl);
3768 if (ret && ret != -EEXIST)
3771 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3772 BUG_ON(ret == -EEXIST);
3773 set_balance_control(bctl);
3775 BUG_ON(ret != -EEXIST);
3776 spin_lock(&fs_info->balance_lock);
3777 update_balance_args(bctl);
3778 spin_unlock(&fs_info->balance_lock);
3781 atomic_inc(&fs_info->balance_running);
3782 mutex_unlock(&fs_info->balance_mutex);
3784 ret = __btrfs_balance(fs_info);
3786 mutex_lock(&fs_info->balance_mutex);
3787 atomic_dec(&fs_info->balance_running);
3789 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3790 fs_info->num_tolerated_disk_barrier_failures =
3791 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3795 memset(bargs, 0, sizeof(*bargs));
3796 update_ioctl_balance_args(fs_info, 0, bargs);
3799 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3800 balance_need_close(fs_info)) {
3801 __cancel_balance(fs_info);
3804 wake_up(&fs_info->balance_wait_q);
3808 if (bctl->flags & BTRFS_BALANCE_RESUME)
3809 __cancel_balance(fs_info);
3812 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3817 static int balance_kthread(void *data)
3819 struct btrfs_fs_info *fs_info = data;
3822 mutex_lock(&fs_info->volume_mutex);
3823 mutex_lock(&fs_info->balance_mutex);
3825 if (fs_info->balance_ctl) {
3826 btrfs_info(fs_info, "continuing balance");
3827 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3830 mutex_unlock(&fs_info->balance_mutex);
3831 mutex_unlock(&fs_info->volume_mutex);
3836 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3838 struct task_struct *tsk;
3840 spin_lock(&fs_info->balance_lock);
3841 if (!fs_info->balance_ctl) {
3842 spin_unlock(&fs_info->balance_lock);
3845 spin_unlock(&fs_info->balance_lock);
3847 if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
3848 btrfs_info(fs_info, "force skipping balance");
3852 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3853 return PTR_ERR_OR_ZERO(tsk);
3856 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3858 struct btrfs_balance_control *bctl;
3859 struct btrfs_balance_item *item;
3860 struct btrfs_disk_balance_args disk_bargs;
3861 struct btrfs_path *path;
3862 struct extent_buffer *leaf;
3863 struct btrfs_key key;
3866 path = btrfs_alloc_path();
3870 key.objectid = BTRFS_BALANCE_OBJECTID;
3871 key.type = BTRFS_BALANCE_ITEM_KEY;
3874 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3877 if (ret > 0) { /* ret = -ENOENT; */
3882 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3888 leaf = path->nodes[0];
3889 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3891 bctl->fs_info = fs_info;
3892 bctl->flags = btrfs_balance_flags(leaf, item);
3893 bctl->flags |= BTRFS_BALANCE_RESUME;
3895 btrfs_balance_data(leaf, item, &disk_bargs);
3896 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3897 btrfs_balance_meta(leaf, item, &disk_bargs);
3898 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3899 btrfs_balance_sys(leaf, item, &disk_bargs);
3900 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3902 WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
3904 mutex_lock(&fs_info->volume_mutex);
3905 mutex_lock(&fs_info->balance_mutex);
3907 set_balance_control(bctl);
3909 mutex_unlock(&fs_info->balance_mutex);
3910 mutex_unlock(&fs_info->volume_mutex);
3912 btrfs_free_path(path);
3916 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3920 mutex_lock(&fs_info->balance_mutex);
3921 if (!fs_info->balance_ctl) {
3922 mutex_unlock(&fs_info->balance_mutex);
3926 if (atomic_read(&fs_info->balance_running)) {
3927 atomic_inc(&fs_info->balance_pause_req);
3928 mutex_unlock(&fs_info->balance_mutex);
3930 wait_event(fs_info->balance_wait_q,
3931 atomic_read(&fs_info->balance_running) == 0);
3933 mutex_lock(&fs_info->balance_mutex);
3934 /* we are good with balance_ctl ripped off from under us */
3935 BUG_ON(atomic_read(&fs_info->balance_running));
3936 atomic_dec(&fs_info->balance_pause_req);
3941 mutex_unlock(&fs_info->balance_mutex);
3945 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3947 if (fs_info->sb->s_flags & MS_RDONLY)
3950 mutex_lock(&fs_info->balance_mutex);
3951 if (!fs_info->balance_ctl) {
3952 mutex_unlock(&fs_info->balance_mutex);
3956 atomic_inc(&fs_info->balance_cancel_req);
3958 * if we are running just wait and return, balance item is
3959 * deleted in btrfs_balance in this case
3961 if (atomic_read(&fs_info->balance_running)) {
3962 mutex_unlock(&fs_info->balance_mutex);
3963 wait_event(fs_info->balance_wait_q,
3964 atomic_read(&fs_info->balance_running) == 0);
3965 mutex_lock(&fs_info->balance_mutex);
3967 /* __cancel_balance needs volume_mutex */
3968 mutex_unlock(&fs_info->balance_mutex);
3969 mutex_lock(&fs_info->volume_mutex);
3970 mutex_lock(&fs_info->balance_mutex);
3972 if (fs_info->balance_ctl)
3973 __cancel_balance(fs_info);
3975 mutex_unlock(&fs_info->volume_mutex);
3978 BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
3979 atomic_dec(&fs_info->balance_cancel_req);
3980 mutex_unlock(&fs_info->balance_mutex);
3984 static int btrfs_uuid_scan_kthread(void *data)
3986 struct btrfs_fs_info *fs_info = data;
3987 struct btrfs_root *root = fs_info->tree_root;
3988 struct btrfs_key key;
3989 struct btrfs_key max_key;
3990 struct btrfs_path *path = NULL;
3992 struct extent_buffer *eb;
3994 struct btrfs_root_item root_item;
3996 struct btrfs_trans_handle *trans = NULL;
3998 path = btrfs_alloc_path();
4005 key.type = BTRFS_ROOT_ITEM_KEY;
4008 max_key.objectid = (u64)-1;
4009 max_key.type = BTRFS_ROOT_ITEM_KEY;
4010 max_key.offset = (u64)-1;
4013 ret = btrfs_search_forward(root, &key, path, 0);
4020 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4021 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4022 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4023 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4026 eb = path->nodes[0];
4027 slot = path->slots[0];
4028 item_size = btrfs_item_size_nr(eb, slot);
4029 if (item_size < sizeof(root_item))
4032 read_extent_buffer(eb, &root_item,
4033 btrfs_item_ptr_offset(eb, slot),
4034 (int)sizeof(root_item));
4035 if (btrfs_root_refs(&root_item) == 0)
4038 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4039 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4043 btrfs_release_path(path);
4045 * 1 - subvol uuid item
4046 * 1 - received_subvol uuid item
4048 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4049 if (IS_ERR(trans)) {
4050 ret = PTR_ERR(trans);
4058 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4059 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
4061 BTRFS_UUID_KEY_SUBVOL,
4064 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4070 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4071 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
4072 root_item.received_uuid,
4073 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4076 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4084 ret = btrfs_end_transaction(trans, fs_info->uuid_root);
4090 btrfs_release_path(path);
4091 if (key.offset < (u64)-1) {
4093 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4095 key.type = BTRFS_ROOT_ITEM_KEY;
4096 } else if (key.objectid < (u64)-1) {
4098 key.type = BTRFS_ROOT_ITEM_KEY;
4107 btrfs_free_path(path);
4108 if (trans && !IS_ERR(trans))
4109 btrfs_end_transaction(trans, fs_info->uuid_root);
4111 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4113 fs_info->update_uuid_tree_gen = 1;
4114 up(&fs_info->uuid_tree_rescan_sem);
4119 * Callback for btrfs_uuid_tree_iterate().
4121 * 0 check succeeded, the entry is not outdated.
4122 * < 0 if an error occured.
4123 * > 0 if the check failed, which means the caller shall remove the entry.
4125 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
4126 u8 *uuid, u8 type, u64 subid)
4128 struct btrfs_key key;
4130 struct btrfs_root *subvol_root;
4132 if (type != BTRFS_UUID_KEY_SUBVOL &&
4133 type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
4136 key.objectid = subid;
4137 key.type = BTRFS_ROOT_ITEM_KEY;
4138 key.offset = (u64)-1;
4139 subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
4140 if (IS_ERR(subvol_root)) {
4141 ret = PTR_ERR(subvol_root);
4148 case BTRFS_UUID_KEY_SUBVOL:
4149 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
4152 case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
4153 if (memcmp(uuid, subvol_root->root_item.received_uuid,
4163 static int btrfs_uuid_rescan_kthread(void *data)
4165 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4169 * 1st step is to iterate through the existing UUID tree and
4170 * to delete all entries that contain outdated data.
4171 * 2nd step is to add all missing entries to the UUID tree.
4173 ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4175 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4176 up(&fs_info->uuid_tree_rescan_sem);
4179 return btrfs_uuid_scan_kthread(data);
4182 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4184 struct btrfs_trans_handle *trans;
4185 struct btrfs_root *tree_root = fs_info->tree_root;
4186 struct btrfs_root *uuid_root;
4187 struct task_struct *task;
4194 trans = btrfs_start_transaction(tree_root, 2);
4196 return PTR_ERR(trans);
4198 uuid_root = btrfs_create_tree(trans, fs_info,
4199 BTRFS_UUID_TREE_OBJECTID);
4200 if (IS_ERR(uuid_root)) {
4201 ret = PTR_ERR(uuid_root);
4202 btrfs_abort_transaction(trans, tree_root, ret);
4206 fs_info->uuid_root = uuid_root;
4208 ret = btrfs_commit_transaction(trans, tree_root);
4212 down(&fs_info->uuid_tree_rescan_sem);
4213 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4215 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4216 btrfs_warn(fs_info, "failed to start uuid_scan task");
4217 up(&fs_info->uuid_tree_rescan_sem);
4218 return PTR_ERR(task);
4224 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4226 struct task_struct *task;
4228 down(&fs_info->uuid_tree_rescan_sem);
4229 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4231 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4232 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4233 up(&fs_info->uuid_tree_rescan_sem);
4234 return PTR_ERR(task);
4241 * shrinking a device means finding all of the device extents past
4242 * the new size, and then following the back refs to the chunks.
4243 * The chunk relocation code actually frees the device extent
4245 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4247 struct btrfs_trans_handle *trans;
4248 struct btrfs_root *root = device->dev_root;
4249 struct btrfs_dev_extent *dev_extent = NULL;
4250 struct btrfs_path *path;
4256 bool retried = false;
4257 bool checked_pending_chunks = false;
4258 struct extent_buffer *l;
4259 struct btrfs_key key;
4260 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4261 u64 old_total = btrfs_super_total_bytes(super_copy);
4262 u64 old_size = btrfs_device_get_total_bytes(device);
4263 u64 diff = old_size - new_size;
4265 if (device->is_tgtdev_for_dev_replace)
4268 path = btrfs_alloc_path();
4276 btrfs_device_set_total_bytes(device, new_size);
4277 if (device->writeable) {
4278 device->fs_devices->total_rw_bytes -= diff;
4279 spin_lock(&root->fs_info->free_chunk_lock);
4280 root->fs_info->free_chunk_space -= diff;
4281 spin_unlock(&root->fs_info->free_chunk_lock);
4283 unlock_chunks(root);
4286 key.objectid = device->devid;
4287 key.offset = (u64)-1;
4288 key.type = BTRFS_DEV_EXTENT_KEY;
4291 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
4292 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4294 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4298 ret = btrfs_previous_item(root, path, 0, key.type);
4300 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4305 btrfs_release_path(path);
4310 slot = path->slots[0];
4311 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4313 if (key.objectid != device->devid) {
4314 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4315 btrfs_release_path(path);
4319 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4320 length = btrfs_dev_extent_length(l, dev_extent);
4322 if (key.offset + length <= new_size) {
4323 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4324 btrfs_release_path(path);
4328 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4329 btrfs_release_path(path);
4331 ret = btrfs_relocate_chunk(root, chunk_offset);
4332 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4333 if (ret && ret != -ENOSPC)
4337 } while (key.offset-- > 0);
4339 if (failed && !retried) {
4343 } else if (failed && retried) {
4348 /* Shrinking succeeded, else we would be at "done". */
4349 trans = btrfs_start_transaction(root, 0);
4350 if (IS_ERR(trans)) {
4351 ret = PTR_ERR(trans);
4358 * We checked in the above loop all device extents that were already in
4359 * the device tree. However before we have updated the device's
4360 * total_bytes to the new size, we might have had chunk allocations that
4361 * have not complete yet (new block groups attached to transaction
4362 * handles), and therefore their device extents were not yet in the
4363 * device tree and we missed them in the loop above. So if we have any
4364 * pending chunk using a device extent that overlaps the device range
4365 * that we can not use anymore, commit the current transaction and
4366 * repeat the search on the device tree - this way we guarantee we will
4367 * not have chunks using device extents that end beyond 'new_size'.
4369 if (!checked_pending_chunks) {
4370 u64 start = new_size;
4371 u64 len = old_size - new_size;
4373 if (contains_pending_extent(trans->transaction, device,
4375 unlock_chunks(root);
4376 checked_pending_chunks = true;
4379 ret = btrfs_commit_transaction(trans, root);
4386 btrfs_device_set_disk_total_bytes(device, new_size);
4387 if (list_empty(&device->resized_list))
4388 list_add_tail(&device->resized_list,
4389 &root->fs_info->fs_devices->resized_devices);
4391 WARN_ON(diff > old_total);
4392 btrfs_set_super_total_bytes(super_copy, old_total - diff);
4393 unlock_chunks(root);
4395 /* Now btrfs_update_device() will change the on-disk size. */
4396 ret = btrfs_update_device(trans, device);
4397 btrfs_end_transaction(trans, root);
4399 btrfs_free_path(path);
4402 btrfs_device_set_total_bytes(device, old_size);
4403 if (device->writeable)
4404 device->fs_devices->total_rw_bytes += diff;
4405 spin_lock(&root->fs_info->free_chunk_lock);
4406 root->fs_info->free_chunk_space += diff;
4407 spin_unlock(&root->fs_info->free_chunk_lock);
4408 unlock_chunks(root);
4413 static int btrfs_add_system_chunk(struct btrfs_root *root,
4414 struct btrfs_key *key,
4415 struct btrfs_chunk *chunk, int item_size)
4417 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4418 struct btrfs_disk_key disk_key;
4423 array_size = btrfs_super_sys_array_size(super_copy);
4424 if (array_size + item_size + sizeof(disk_key)
4425 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4426 unlock_chunks(root);
4430 ptr = super_copy->sys_chunk_array + array_size;
4431 btrfs_cpu_key_to_disk(&disk_key, key);
4432 memcpy(ptr, &disk_key, sizeof(disk_key));
4433 ptr += sizeof(disk_key);
4434 memcpy(ptr, chunk, item_size);
4435 item_size += sizeof(disk_key);
4436 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4437 unlock_chunks(root);
4443 * sort the devices in descending order by max_avail, total_avail
4445 static int btrfs_cmp_device_info(const void *a, const void *b)
4447 const struct btrfs_device_info *di_a = a;
4448 const struct btrfs_device_info *di_b = b;
4450 if (di_a->max_avail > di_b->max_avail)
4452 if (di_a->max_avail < di_b->max_avail)
4454 if (di_a->total_avail > di_b->total_avail)
4456 if (di_a->total_avail < di_b->total_avail)
4461 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
4463 /* TODO allow them to set a preferred stripe size */
4467 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4469 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4472 btrfs_set_fs_incompat(info, RAID56);
4475 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r) \
4476 - sizeof(struct btrfs_item) \
4477 - sizeof(struct btrfs_chunk)) \
4478 / sizeof(struct btrfs_stripe) + 1)
4480 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
4481 - 2 * sizeof(struct btrfs_disk_key) \
4482 - 2 * sizeof(struct btrfs_chunk)) \
4483 / sizeof(struct btrfs_stripe) + 1)
4485 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4486 struct btrfs_root *extent_root, u64 start,
4489 struct btrfs_fs_info *info = extent_root->fs_info;
4490 struct btrfs_fs_devices *fs_devices = info->fs_devices;
4491 struct list_head *cur;
4492 struct map_lookup *map = NULL;
4493 struct extent_map_tree *em_tree;
4494 struct extent_map *em;
4495 struct btrfs_device_info *devices_info = NULL;
4497 int num_stripes; /* total number of stripes to allocate */
4498 int data_stripes; /* number of stripes that count for
4500 int sub_stripes; /* sub_stripes info for map */
4501 int dev_stripes; /* stripes per dev */
4502 int devs_max; /* max devs to use */
4503 int devs_min; /* min devs needed */
4504 int devs_increment; /* ndevs has to be a multiple of this */
4505 int ncopies; /* how many copies to data has */
4507 u64 max_stripe_size;
4511 u64 raid_stripe_len = BTRFS_STRIPE_LEN;
4517 BUG_ON(!alloc_profile_is_valid(type, 0));
4519 if (list_empty(&fs_devices->alloc_list))
4522 index = __get_raid_index(type);
4524 sub_stripes = btrfs_raid_array[index].sub_stripes;
4525 dev_stripes = btrfs_raid_array[index].dev_stripes;
4526 devs_max = btrfs_raid_array[index].devs_max;
4527 devs_min = btrfs_raid_array[index].devs_min;
4528 devs_increment = btrfs_raid_array[index].devs_increment;
4529 ncopies = btrfs_raid_array[index].ncopies;
4531 if (type & BTRFS_BLOCK_GROUP_DATA) {
4532 max_stripe_size = 1024 * 1024 * 1024;
4533 max_chunk_size = 10 * max_stripe_size;
4535 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4536 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4537 /* for larger filesystems, use larger metadata chunks */
4538 if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024)
4539 max_stripe_size = 1024 * 1024 * 1024;
4541 max_stripe_size = 256 * 1024 * 1024;
4542 max_chunk_size = max_stripe_size;
4544 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4545 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4546 max_stripe_size = 32 * 1024 * 1024;
4547 max_chunk_size = 2 * max_stripe_size;
4549 devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4551 btrfs_err(info, "invalid chunk type 0x%llx requested",
4556 /* we don't want a chunk larger than 10% of writeable space */
4557 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4560 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4565 cur = fs_devices->alloc_list.next;
4568 * in the first pass through the devices list, we gather information
4569 * about the available holes on each device.
4572 while (cur != &fs_devices->alloc_list) {
4573 struct btrfs_device *device;
4577 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4581 if (!device->writeable) {
4583 "BTRFS: read-only device in alloc_list\n");
4587 if (!device->in_fs_metadata ||
4588 device->is_tgtdev_for_dev_replace)
4591 if (device->total_bytes > device->bytes_used)
4592 total_avail = device->total_bytes - device->bytes_used;
4596 /* If there is no space on this device, skip it. */
4597 if (total_avail == 0)
4600 ret = find_free_dev_extent(trans, device,
4601 max_stripe_size * dev_stripes,
4602 &dev_offset, &max_avail);
4603 if (ret && ret != -ENOSPC)
4607 max_avail = max_stripe_size * dev_stripes;
4609 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4612 if (ndevs == fs_devices->rw_devices) {
4613 WARN(1, "%s: found more than %llu devices\n",
4614 __func__, fs_devices->rw_devices);
4617 devices_info[ndevs].dev_offset = dev_offset;
4618 devices_info[ndevs].max_avail = max_avail;
4619 devices_info[ndevs].total_avail = total_avail;
4620 devices_info[ndevs].dev = device;
4625 * now sort the devices by hole size / available space
4627 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4628 btrfs_cmp_device_info, NULL);
4630 /* round down to number of usable stripes */
4631 ndevs -= ndevs % devs_increment;
4633 if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4638 if (devs_max && ndevs > devs_max)
4641 * the primary goal is to maximize the number of stripes, so use as many
4642 * devices as possible, even if the stripes are not maximum sized.
4644 stripe_size = devices_info[ndevs-1].max_avail;
4645 num_stripes = ndevs * dev_stripes;
4648 * this will have to be fixed for RAID1 and RAID10 over
4651 data_stripes = num_stripes / ncopies;
4653 if (type & BTRFS_BLOCK_GROUP_RAID5) {
4654 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
4655 btrfs_super_stripesize(info->super_copy));
4656 data_stripes = num_stripes - 1;
4658 if (type & BTRFS_BLOCK_GROUP_RAID6) {
4659 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
4660 btrfs_super_stripesize(info->super_copy));
4661 data_stripes = num_stripes - 2;
4665 * Use the number of data stripes to figure out how big this chunk
4666 * is really going to be in terms of logical address space,
4667 * and compare that answer with the max chunk size
4669 if (stripe_size * data_stripes > max_chunk_size) {
4670 u64 mask = (1ULL << 24) - 1;
4672 stripe_size = div_u64(max_chunk_size, data_stripes);
4674 /* bump the answer up to a 16MB boundary */
4675 stripe_size = (stripe_size + mask) & ~mask;
4677 /* but don't go higher than the limits we found
4678 * while searching for free extents
4680 if (stripe_size > devices_info[ndevs-1].max_avail)
4681 stripe_size = devices_info[ndevs-1].max_avail;
4684 stripe_size = div_u64(stripe_size, dev_stripes);
4686 /* align to BTRFS_STRIPE_LEN */
4687 stripe_size = div_u64(stripe_size, raid_stripe_len);
4688 stripe_size *= raid_stripe_len;
4690 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4695 map->num_stripes = num_stripes;
4697 for (i = 0; i < ndevs; ++i) {
4698 for (j = 0; j < dev_stripes; ++j) {
4699 int s = i * dev_stripes + j;
4700 map->stripes[s].dev = devices_info[i].dev;
4701 map->stripes[s].physical = devices_info[i].dev_offset +
4705 map->sector_size = extent_root->sectorsize;
4706 map->stripe_len = raid_stripe_len;
4707 map->io_align = raid_stripe_len;
4708 map->io_width = raid_stripe_len;
4710 map->sub_stripes = sub_stripes;
4712 num_bytes = stripe_size * data_stripes;
4714 trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
4716 em = alloc_extent_map();
4722 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4723 em->bdev = (struct block_device *)map;
4725 em->len = num_bytes;
4726 em->block_start = 0;
4727 em->block_len = em->len;
4728 em->orig_block_len = stripe_size;
4730 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4731 write_lock(&em_tree->lock);
4732 ret = add_extent_mapping(em_tree, em, 0);
4734 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4735 atomic_inc(&em->refs);
4737 write_unlock(&em_tree->lock);
4739 free_extent_map(em);
4743 ret = btrfs_make_block_group(trans, extent_root, 0, type,
4744 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4747 goto error_del_extent;
4749 for (i = 0; i < map->num_stripes; i++) {
4750 num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4751 btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4754 spin_lock(&extent_root->fs_info->free_chunk_lock);
4755 extent_root->fs_info->free_chunk_space -= (stripe_size *
4757 spin_unlock(&extent_root->fs_info->free_chunk_lock);
4759 free_extent_map(em);
4760 check_raid56_incompat_flag(extent_root->fs_info, type);
4762 kfree(devices_info);
4766 write_lock(&em_tree->lock);
4767 remove_extent_mapping(em_tree, em);
4768 write_unlock(&em_tree->lock);
4770 /* One for our allocation */
4771 free_extent_map(em);
4772 /* One for the tree reference */
4773 free_extent_map(em);
4774 /* One for the pending_chunks list reference */
4775 free_extent_map(em);
4777 kfree(devices_info);
4781 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4782 struct btrfs_root *extent_root,
4783 u64 chunk_offset, u64 chunk_size)
4785 struct btrfs_key key;
4786 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4787 struct btrfs_device *device;
4788 struct btrfs_chunk *chunk;
4789 struct btrfs_stripe *stripe;
4790 struct extent_map_tree *em_tree;
4791 struct extent_map *em;
4792 struct map_lookup *map;
4799 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4800 read_lock(&em_tree->lock);
4801 em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size);
4802 read_unlock(&em_tree->lock);
4805 btrfs_crit(extent_root->fs_info, "unable to find logical "
4806 "%Lu len %Lu", chunk_offset, chunk_size);
4810 if (em->start != chunk_offset || em->len != chunk_size) {
4811 btrfs_crit(extent_root->fs_info, "found a bad mapping, wanted"
4812 " %Lu-%Lu, found %Lu-%Lu", chunk_offset,
4813 chunk_size, em->start, em->len);
4814 free_extent_map(em);
4818 map = (struct map_lookup *)em->bdev;
4819 item_size = btrfs_chunk_item_size(map->num_stripes);
4820 stripe_size = em->orig_block_len;
4822 chunk = kzalloc(item_size, GFP_NOFS);
4828 for (i = 0; i < map->num_stripes; i++) {
4829 device = map->stripes[i].dev;
4830 dev_offset = map->stripes[i].physical;
4832 ret = btrfs_update_device(trans, device);
4835 ret = btrfs_alloc_dev_extent(trans, device,
4836 chunk_root->root_key.objectid,
4837 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4838 chunk_offset, dev_offset,
4844 stripe = &chunk->stripe;
4845 for (i = 0; i < map->num_stripes; i++) {
4846 device = map->stripes[i].dev;
4847 dev_offset = map->stripes[i].physical;
4849 btrfs_set_stack_stripe_devid(stripe, device->devid);
4850 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4851 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4855 btrfs_set_stack_chunk_length(chunk, chunk_size);
4856 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4857 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4858 btrfs_set_stack_chunk_type(chunk, map->type);
4859 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4860 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4861 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4862 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4863 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4865 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4866 key.type = BTRFS_CHUNK_ITEM_KEY;
4867 key.offset = chunk_offset;
4869 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4870 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4872 * TODO: Cleanup of inserted chunk root in case of
4875 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
4881 free_extent_map(em);
4886 * Chunk allocation falls into two parts. The first part does works
4887 * that make the new allocated chunk useable, but not do any operation
4888 * that modifies the chunk tree. The second part does the works that
4889 * require modifying the chunk tree. This division is important for the
4890 * bootstrap process of adding storage to a seed btrfs.
4892 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4893 struct btrfs_root *extent_root, u64 type)
4897 ASSERT(mutex_is_locked(&extent_root->fs_info->chunk_mutex));
4898 chunk_offset = find_next_chunk(extent_root->fs_info);
4899 return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type);
4902 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4903 struct btrfs_root *root,
4904 struct btrfs_device *device)
4907 u64 sys_chunk_offset;
4909 struct btrfs_fs_info *fs_info = root->fs_info;
4910 struct btrfs_root *extent_root = fs_info->extent_root;
4913 chunk_offset = find_next_chunk(fs_info);
4914 alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
4915 ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset,
4920 sys_chunk_offset = find_next_chunk(root->fs_info);
4921 alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
4922 ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset,
4927 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
4931 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
4932 BTRFS_BLOCK_GROUP_RAID10 |
4933 BTRFS_BLOCK_GROUP_RAID5 |
4934 BTRFS_BLOCK_GROUP_DUP)) {
4936 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
4945 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
4947 struct extent_map *em;
4948 struct map_lookup *map;
4949 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4954 read_lock(&map_tree->map_tree.lock);
4955 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
4956 read_unlock(&map_tree->map_tree.lock);
4960 map = (struct map_lookup *)em->bdev;
4961 for (i = 0; i < map->num_stripes; i++) {
4962 if (map->stripes[i].dev->missing) {
4967 if (!map->stripes[i].dev->writeable) {
4974 * If the number of missing devices is larger than max errors,
4975 * we can not write the data into that chunk successfully, so
4978 if (miss_ndevs > btrfs_chunk_max_errors(map))
4981 free_extent_map(em);
4985 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
4987 extent_map_tree_init(&tree->map_tree);
4990 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
4992 struct extent_map *em;
4995 write_lock(&tree->map_tree.lock);
4996 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
4998 remove_extent_mapping(&tree->map_tree, em);
4999 write_unlock(&tree->map_tree.lock);
5003 free_extent_map(em);
5004 /* once for the tree */
5005 free_extent_map(em);
5009 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5011 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5012 struct extent_map *em;
5013 struct map_lookup *map;
5014 struct extent_map_tree *em_tree = &map_tree->map_tree;
5017 read_lock(&em_tree->lock);
5018 em = lookup_extent_mapping(em_tree, logical, len);
5019 read_unlock(&em_tree->lock);
5022 * We could return errors for these cases, but that could get ugly and
5023 * we'd probably do the same thing which is just not do anything else
5024 * and exit, so return 1 so the callers don't try to use other copies.
5027 btrfs_crit(fs_info, "No mapping for %Lu-%Lu", logical,
5032 if (em->start > logical || em->start + em->len < logical) {
5033 btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got "
5034 "%Lu-%Lu", logical, logical+len, em->start,
5035 em->start + em->len);
5036 free_extent_map(em);
5040 map = (struct map_lookup *)em->bdev;
5041 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
5042 ret = map->num_stripes;
5043 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5044 ret = map->sub_stripes;
5045 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5047 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5051 free_extent_map(em);
5053 btrfs_dev_replace_lock(&fs_info->dev_replace);
5054 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
5056 btrfs_dev_replace_unlock(&fs_info->dev_replace);
5061 unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
5062 struct btrfs_mapping_tree *map_tree,
5065 struct extent_map *em;
5066 struct map_lookup *map;
5067 struct extent_map_tree *em_tree = &map_tree->map_tree;
5068 unsigned long len = root->sectorsize;
5070 read_lock(&em_tree->lock);
5071 em = lookup_extent_mapping(em_tree, logical, len);
5072 read_unlock(&em_tree->lock);
5075 BUG_ON(em->start > logical || em->start + em->len < logical);
5076 map = (struct map_lookup *)em->bdev;
5077 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5078 len = map->stripe_len * nr_data_stripes(map);
5079 free_extent_map(em);
5083 int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
5084 u64 logical, u64 len, int mirror_num)
5086 struct extent_map *em;
5087 struct map_lookup *map;
5088 struct extent_map_tree *em_tree = &map_tree->map_tree;
5091 read_lock(&em_tree->lock);
5092 em = lookup_extent_mapping(em_tree, logical, len);
5093 read_unlock(&em_tree->lock);
5096 BUG_ON(em->start > logical || em->start + em->len < logical);
5097 map = (struct map_lookup *)em->bdev;
5098 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5100 free_extent_map(em);
5104 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5105 struct map_lookup *map, int first, int num,
5106 int optimal, int dev_replace_is_ongoing)
5110 struct btrfs_device *srcdev;
5112 if (dev_replace_is_ongoing &&
5113 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5114 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5115 srcdev = fs_info->dev_replace.srcdev;
5120 * try to avoid the drive that is the source drive for a
5121 * dev-replace procedure, only choose it if no other non-missing
5122 * mirror is available
5124 for (tolerance = 0; tolerance < 2; tolerance++) {
5125 if (map->stripes[optimal].dev->bdev &&
5126 (tolerance || map->stripes[optimal].dev != srcdev))
5128 for (i = first; i < first + num; i++) {
5129 if (map->stripes[i].dev->bdev &&
5130 (tolerance || map->stripes[i].dev != srcdev))
5135 /* we couldn't find one that doesn't fail. Just return something
5136 * and the io error handling code will clean up eventually
5141 static inline int parity_smaller(u64 a, u64 b)
5146 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5147 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5149 struct btrfs_bio_stripe s;
5156 for (i = 0; i < num_stripes - 1; i++) {
5157 if (parity_smaller(bbio->raid_map[i],
5158 bbio->raid_map[i+1])) {
5159 s = bbio->stripes[i];
5160 l = bbio->raid_map[i];
5161 bbio->stripes[i] = bbio->stripes[i+1];
5162 bbio->raid_map[i] = bbio->raid_map[i+1];
5163 bbio->stripes[i+1] = s;
5164 bbio->raid_map[i+1] = l;
5172 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5174 struct btrfs_bio *bbio = kzalloc(
5175 /* the size of the btrfs_bio */
5176 sizeof(struct btrfs_bio) +
5177 /* plus the variable array for the stripes */
5178 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5179 /* plus the variable array for the tgt dev */
5180 sizeof(int) * (real_stripes) +
5182 * plus the raid_map, which includes both the tgt dev
5185 sizeof(u64) * (total_stripes),
5186 GFP_NOFS|__GFP_NOFAIL);
5188 atomic_set(&bbio->error, 0);
5189 atomic_set(&bbio->refs, 1);
5194 void btrfs_get_bbio(struct btrfs_bio *bbio)
5196 WARN_ON(!atomic_read(&bbio->refs));
5197 atomic_inc(&bbio->refs);
5200 void btrfs_put_bbio(struct btrfs_bio *bbio)
5204 if (atomic_dec_and_test(&bbio->refs))
5208 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5209 u64 logical, u64 *length,
5210 struct btrfs_bio **bbio_ret,
5211 int mirror_num, int need_raid_map)
5213 struct extent_map *em;
5214 struct map_lookup *map;
5215 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5216 struct extent_map_tree *em_tree = &map_tree->map_tree;
5219 u64 stripe_end_offset;
5229 int tgtdev_indexes = 0;
5230 struct btrfs_bio *bbio = NULL;
5231 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
5232 int dev_replace_is_ongoing = 0;
5233 int num_alloc_stripes;
5234 int patch_the_first_stripe_for_dev_replace = 0;
5235 u64 physical_to_patch_in_first_stripe = 0;
5236 u64 raid56_full_stripe_start = (u64)-1;
5238 read_lock(&em_tree->lock);
5239 em = lookup_extent_mapping(em_tree, logical, *length);
5240 read_unlock(&em_tree->lock);
5243 btrfs_crit(fs_info, "unable to find logical %llu len %llu",
5248 if (em->start > logical || em->start + em->len < logical) {
5249 btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, "
5250 "found %Lu-%Lu", logical, em->start,
5251 em->start + em->len);
5252 free_extent_map(em);
5256 map = (struct map_lookup *)em->bdev;
5257 offset = logical - em->start;
5259 stripe_len = map->stripe_len;
5262 * stripe_nr counts the total number of stripes we have to stride
5263 * to get to this block
5265 stripe_nr = div64_u64(stripe_nr, stripe_len);
5267 stripe_offset = stripe_nr * stripe_len;
5268 BUG_ON(offset < stripe_offset);
5270 /* stripe_offset is the offset of this block in its stripe*/
5271 stripe_offset = offset - stripe_offset;
5273 /* if we're here for raid56, we need to know the stripe aligned start */
5274 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5275 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5276 raid56_full_stripe_start = offset;
5278 /* allow a write of a full stripe, but make sure we don't
5279 * allow straddling of stripes
5281 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5283 raid56_full_stripe_start *= full_stripe_len;
5286 if (rw & REQ_DISCARD) {
5287 /* we don't discard raid56 yet */
5288 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5292 *length = min_t(u64, em->len - offset, *length);
5293 } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5295 /* For writes to RAID[56], allow a full stripeset across all disks.
5296 For other RAID types and for RAID[56] reads, just allow a single
5297 stripe (on a single disk). */
5298 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
5300 max_len = stripe_len * nr_data_stripes(map) -
5301 (offset - raid56_full_stripe_start);
5303 /* we limit the length of each bio to what fits in a stripe */
5304 max_len = stripe_len - stripe_offset;
5306 *length = min_t(u64, em->len - offset, max_len);
5308 *length = em->len - offset;
5311 /* This is for when we're called from btrfs_merge_bio_hook() and all
5312 it cares about is the length */
5316 btrfs_dev_replace_lock(dev_replace);
5317 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5318 if (!dev_replace_is_ongoing)
5319 btrfs_dev_replace_unlock(dev_replace);
5321 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
5322 !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) &&
5323 dev_replace->tgtdev != NULL) {
5325 * in dev-replace case, for repair case (that's the only
5326 * case where the mirror is selected explicitly when
5327 * calling btrfs_map_block), blocks left of the left cursor
5328 * can also be read from the target drive.
5329 * For REQ_GET_READ_MIRRORS, the target drive is added as
5330 * the last one to the array of stripes. For READ, it also
5331 * needs to be supported using the same mirror number.
5332 * If the requested block is not left of the left cursor,
5333 * EIO is returned. This can happen because btrfs_num_copies()
5334 * returns one more in the dev-replace case.
5336 u64 tmp_length = *length;
5337 struct btrfs_bio *tmp_bbio = NULL;
5338 int tmp_num_stripes;
5339 u64 srcdev_devid = dev_replace->srcdev->devid;
5340 int index_srcdev = 0;
5342 u64 physical_of_found = 0;
5344 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
5345 logical, &tmp_length, &tmp_bbio, 0, 0);
5347 WARN_ON(tmp_bbio != NULL);
5351 tmp_num_stripes = tmp_bbio->num_stripes;
5352 if (mirror_num > tmp_num_stripes) {
5354 * REQ_GET_READ_MIRRORS does not contain this
5355 * mirror, that means that the requested area
5356 * is not left of the left cursor
5359 btrfs_put_bbio(tmp_bbio);
5364 * process the rest of the function using the mirror_num
5365 * of the source drive. Therefore look it up first.
5366 * At the end, patch the device pointer to the one of the
5369 for (i = 0; i < tmp_num_stripes; i++) {
5370 if (tmp_bbio->stripes[i].dev->devid == srcdev_devid) {
5372 * In case of DUP, in order to keep it
5373 * simple, only add the mirror with the
5374 * lowest physical address
5377 physical_of_found <=
5378 tmp_bbio->stripes[i].physical)
5383 tmp_bbio->stripes[i].physical;
5388 mirror_num = index_srcdev + 1;
5389 patch_the_first_stripe_for_dev_replace = 1;
5390 physical_to_patch_in_first_stripe = physical_of_found;
5394 btrfs_put_bbio(tmp_bbio);
5398 btrfs_put_bbio(tmp_bbio);
5399 } else if (mirror_num > map->num_stripes) {
5405 stripe_nr_orig = stripe_nr;
5406 stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
5407 stripe_nr_end = div_u64(stripe_nr_end, map->stripe_len);
5408 stripe_end_offset = stripe_nr_end * map->stripe_len -
5411 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5412 if (rw & REQ_DISCARD)
5413 num_stripes = min_t(u64, map->num_stripes,
5414 stripe_nr_end - stripe_nr_orig);
5415 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5417 if (!(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)))
5419 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
5420 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))
5421 num_stripes = map->num_stripes;
5422 else if (mirror_num)
5423 stripe_index = mirror_num - 1;
5425 stripe_index = find_live_mirror(fs_info, map, 0,
5427 current->pid % map->num_stripes,
5428 dev_replace_is_ongoing);
5429 mirror_num = stripe_index + 1;
5432 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
5433 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
5434 num_stripes = map->num_stripes;
5435 } else if (mirror_num) {
5436 stripe_index = mirror_num - 1;
5441 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5442 u32 factor = map->num_stripes / map->sub_stripes;
5444 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5445 stripe_index *= map->sub_stripes;
5447 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
5448 num_stripes = map->sub_stripes;
5449 else if (rw & REQ_DISCARD)
5450 num_stripes = min_t(u64, map->sub_stripes *
5451 (stripe_nr_end - stripe_nr_orig),
5453 else if (mirror_num)
5454 stripe_index += mirror_num - 1;
5456 int old_stripe_index = stripe_index;
5457 stripe_index = find_live_mirror(fs_info, map,
5459 map->sub_stripes, stripe_index +
5460 current->pid % map->sub_stripes,
5461 dev_replace_is_ongoing);
5462 mirror_num = stripe_index - old_stripe_index + 1;
5465 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5466 if (need_raid_map &&
5467 ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) ||
5469 /* push stripe_nr back to the start of the full stripe */
5470 stripe_nr = div_u64(raid56_full_stripe_start,
5471 stripe_len * nr_data_stripes(map));
5473 /* RAID[56] write or recovery. Return all stripes */
5474 num_stripes = map->num_stripes;
5475 max_errors = nr_parity_stripes(map);
5477 *length = map->stripe_len;
5482 * Mirror #0 or #1 means the original data block.
5483 * Mirror #2 is RAID5 parity block.
5484 * Mirror #3 is RAID6 Q block.
5486 stripe_nr = div_u64_rem(stripe_nr,
5487 nr_data_stripes(map), &stripe_index);
5489 stripe_index = nr_data_stripes(map) +
5492 /* We distribute the parity blocks across stripes */
5493 div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
5495 if (!(rw & (REQ_WRITE | REQ_DISCARD |
5496 REQ_GET_READ_MIRRORS)) && mirror_num <= 1)
5501 * after this, stripe_nr is the number of stripes on this
5502 * device we have to walk to find the data, and stripe_index is
5503 * the number of our device in the stripe array
5505 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5507 mirror_num = stripe_index + 1;
5509 BUG_ON(stripe_index >= map->num_stripes);
5511 num_alloc_stripes = num_stripes;
5512 if (dev_replace_is_ongoing) {
5513 if (rw & (REQ_WRITE | REQ_DISCARD))
5514 num_alloc_stripes <<= 1;
5515 if (rw & REQ_GET_READ_MIRRORS)
5516 num_alloc_stripes++;
5517 tgtdev_indexes = num_stripes;
5520 bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
5525 if (dev_replace_is_ongoing)
5526 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
5528 /* build raid_map */
5529 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK &&
5530 need_raid_map && ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) ||
5535 bbio->raid_map = (u64 *)((void *)bbio->stripes +
5536 sizeof(struct btrfs_bio_stripe) *
5538 sizeof(int) * tgtdev_indexes);
5540 /* Work out the disk rotation on this stripe-set */
5541 div_u64_rem(stripe_nr, num_stripes, &rot);
5543 /* Fill in the logical address of each stripe */
5544 tmp = stripe_nr * nr_data_stripes(map);
5545 for (i = 0; i < nr_data_stripes(map); i++)
5546 bbio->raid_map[(i+rot) % num_stripes] =
5547 em->start + (tmp + i) * map->stripe_len;
5549 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5550 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5551 bbio->raid_map[(i+rot+1) % num_stripes] =
5555 if (rw & REQ_DISCARD) {
5557 u32 sub_stripes = 0;
5558 u64 stripes_per_dev = 0;
5559 u32 remaining_stripes = 0;
5560 u32 last_stripe = 0;
5563 (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
5564 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5567 sub_stripes = map->sub_stripes;
5569 factor = map->num_stripes / sub_stripes;
5570 stripes_per_dev = div_u64_rem(stripe_nr_end -
5573 &remaining_stripes);
5574 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5575 last_stripe *= sub_stripes;
5578 for (i = 0; i < num_stripes; i++) {
5579 bbio->stripes[i].physical =
5580 map->stripes[stripe_index].physical +
5581 stripe_offset + stripe_nr * map->stripe_len;
5582 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5584 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5585 BTRFS_BLOCK_GROUP_RAID10)) {
5586 bbio->stripes[i].length = stripes_per_dev *
5589 if (i / sub_stripes < remaining_stripes)
5590 bbio->stripes[i].length +=
5594 * Special for the first stripe and
5597 * |-------|...|-------|
5601 if (i < sub_stripes)
5602 bbio->stripes[i].length -=
5605 if (stripe_index >= last_stripe &&
5606 stripe_index <= (last_stripe +
5608 bbio->stripes[i].length -=
5611 if (i == sub_stripes - 1)
5614 bbio->stripes[i].length = *length;
5617 if (stripe_index == map->num_stripes) {
5618 /* This could only happen for RAID0/10 */
5624 for (i = 0; i < num_stripes; i++) {
5625 bbio->stripes[i].physical =
5626 map->stripes[stripe_index].physical +
5628 stripe_nr * map->stripe_len;
5629 bbio->stripes[i].dev =
5630 map->stripes[stripe_index].dev;
5635 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
5636 max_errors = btrfs_chunk_max_errors(map);
5639 sort_parity_stripes(bbio, num_stripes);
5642 if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) &&
5643 dev_replace->tgtdev != NULL) {
5644 int index_where_to_add;
5645 u64 srcdev_devid = dev_replace->srcdev->devid;
5648 * duplicate the write operations while the dev replace
5649 * procedure is running. Since the copying of the old disk
5650 * to the new disk takes place at run time while the
5651 * filesystem is mounted writable, the regular write
5652 * operations to the old disk have to be duplicated to go
5653 * to the new disk as well.
5654 * Note that device->missing is handled by the caller, and
5655 * that the write to the old disk is already set up in the
5658 index_where_to_add = num_stripes;
5659 for (i = 0; i < num_stripes; i++) {
5660 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5661 /* write to new disk, too */
5662 struct btrfs_bio_stripe *new =
5663 bbio->stripes + index_where_to_add;
5664 struct btrfs_bio_stripe *old =
5667 new->physical = old->physical;
5668 new->length = old->length;
5669 new->dev = dev_replace->tgtdev;
5670 bbio->tgtdev_map[i] = index_where_to_add;
5671 index_where_to_add++;
5676 num_stripes = index_where_to_add;
5677 } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) &&
5678 dev_replace->tgtdev != NULL) {
5679 u64 srcdev_devid = dev_replace->srcdev->devid;
5680 int index_srcdev = 0;
5682 u64 physical_of_found = 0;
5685 * During the dev-replace procedure, the target drive can
5686 * also be used to read data in case it is needed to repair
5687 * a corrupt block elsewhere. This is possible if the
5688 * requested area is left of the left cursor. In this area,
5689 * the target drive is a full copy of the source drive.
5691 for (i = 0; i < num_stripes; i++) {
5692 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5694 * In case of DUP, in order to keep it
5695 * simple, only add the mirror with the
5696 * lowest physical address
5699 physical_of_found <=
5700 bbio->stripes[i].physical)
5704 physical_of_found = bbio->stripes[i].physical;
5708 if (physical_of_found + map->stripe_len <=
5709 dev_replace->cursor_left) {
5710 struct btrfs_bio_stripe *tgtdev_stripe =
5711 bbio->stripes + num_stripes;
5713 tgtdev_stripe->physical = physical_of_found;
5714 tgtdev_stripe->length =
5715 bbio->stripes[index_srcdev].length;
5716 tgtdev_stripe->dev = dev_replace->tgtdev;
5717 bbio->tgtdev_map[index_srcdev] = num_stripes;
5726 bbio->map_type = map->type;
5727 bbio->num_stripes = num_stripes;
5728 bbio->max_errors = max_errors;
5729 bbio->mirror_num = mirror_num;
5730 bbio->num_tgtdevs = tgtdev_indexes;
5733 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5734 * mirror_num == num_stripes + 1 && dev_replace target drive is
5735 * available as a mirror
5737 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5738 WARN_ON(num_stripes > 1);
5739 bbio->stripes[0].dev = dev_replace->tgtdev;
5740 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5741 bbio->mirror_num = map->num_stripes + 1;
5744 if (dev_replace_is_ongoing)
5745 btrfs_dev_replace_unlock(dev_replace);
5746 free_extent_map(em);
5750 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5751 u64 logical, u64 *length,
5752 struct btrfs_bio **bbio_ret, int mirror_num)
5754 return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5758 /* For Scrub/replace */
5759 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, int rw,
5760 u64 logical, u64 *length,
5761 struct btrfs_bio **bbio_ret, int mirror_num,
5764 return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5765 mirror_num, need_raid_map);
5768 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
5769 u64 chunk_start, u64 physical, u64 devid,
5770 u64 **logical, int *naddrs, int *stripe_len)
5772 struct extent_map_tree *em_tree = &map_tree->map_tree;
5773 struct extent_map *em;
5774 struct map_lookup *map;
5782 read_lock(&em_tree->lock);
5783 em = lookup_extent_mapping(em_tree, chunk_start, 1);
5784 read_unlock(&em_tree->lock);
5787 printk(KERN_ERR "BTRFS: couldn't find em for chunk %Lu\n",
5792 if (em->start != chunk_start) {
5793 printk(KERN_ERR "BTRFS: bad chunk start, em=%Lu, wanted=%Lu\n",
5794 em->start, chunk_start);
5795 free_extent_map(em);
5798 map = (struct map_lookup *)em->bdev;
5801 rmap_len = map->stripe_len;
5803 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5804 length = div_u64(length, map->num_stripes / map->sub_stripes);
5805 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5806 length = div_u64(length, map->num_stripes);
5807 else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5808 length = div_u64(length, nr_data_stripes(map));
5809 rmap_len = map->stripe_len * nr_data_stripes(map);
5812 buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
5813 BUG_ON(!buf); /* -ENOMEM */
5815 for (i = 0; i < map->num_stripes; i++) {
5816 if (devid && map->stripes[i].dev->devid != devid)
5818 if (map->stripes[i].physical > physical ||
5819 map->stripes[i].physical + length <= physical)
5822 stripe_nr = physical - map->stripes[i].physical;
5823 stripe_nr = div_u64(stripe_nr, map->stripe_len);
5825 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5826 stripe_nr = stripe_nr * map->num_stripes + i;
5827 stripe_nr = div_u64(stripe_nr, map->sub_stripes);
5828 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5829 stripe_nr = stripe_nr * map->num_stripes + i;
5830 } /* else if RAID[56], multiply by nr_data_stripes().
5831 * Alternatively, just use rmap_len below instead of
5832 * map->stripe_len */
5834 bytenr = chunk_start + stripe_nr * rmap_len;
5835 WARN_ON(nr >= map->num_stripes);
5836 for (j = 0; j < nr; j++) {
5837 if (buf[j] == bytenr)
5841 WARN_ON(nr >= map->num_stripes);
5848 *stripe_len = rmap_len;
5850 free_extent_map(em);
5854 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
5856 bio->bi_private = bbio->private;
5857 bio->bi_end_io = bbio->end_io;
5860 btrfs_put_bbio(bbio);
5863 static void btrfs_end_bio(struct bio *bio)
5865 struct btrfs_bio *bbio = bio->bi_private;
5866 int is_orig_bio = 0;
5868 if (bio->bi_error) {
5869 atomic_inc(&bbio->error);
5870 if (bio->bi_error == -EIO || bio->bi_error == -EREMOTEIO) {
5871 unsigned int stripe_index =
5872 btrfs_io_bio(bio)->stripe_index;
5873 struct btrfs_device *dev;
5875 BUG_ON(stripe_index >= bbio->num_stripes);
5876 dev = bbio->stripes[stripe_index].dev;
5878 if (bio->bi_rw & WRITE)
5879 btrfs_dev_stat_inc(dev,
5880 BTRFS_DEV_STAT_WRITE_ERRS);
5882 btrfs_dev_stat_inc(dev,
5883 BTRFS_DEV_STAT_READ_ERRS);
5884 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
5885 btrfs_dev_stat_inc(dev,
5886 BTRFS_DEV_STAT_FLUSH_ERRS);
5887 btrfs_dev_stat_print_on_error(dev);
5892 if (bio == bbio->orig_bio)
5895 btrfs_bio_counter_dec(bbio->fs_info);
5897 if (atomic_dec_and_test(&bbio->stripes_pending)) {
5900 bio = bbio->orig_bio;
5903 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5904 /* only send an error to the higher layers if it is
5905 * beyond the tolerance of the btrfs bio
5907 if (atomic_read(&bbio->error) > bbio->max_errors) {
5908 bio->bi_error = -EIO;
5911 * this bio is actually up to date, we didn't
5912 * go over the max number of errors
5917 btrfs_end_bbio(bbio, bio);
5918 } else if (!is_orig_bio) {
5924 * see run_scheduled_bios for a description of why bios are collected for
5927 * This will add one bio to the pending list for a device and make sure
5928 * the work struct is scheduled.
5930 static noinline void btrfs_schedule_bio(struct btrfs_root *root,
5931 struct btrfs_device *device,
5932 int rw, struct bio *bio)
5934 int should_queue = 1;
5935 struct btrfs_pending_bios *pending_bios;
5937 if (device->missing || !device->bdev) {
5942 /* don't bother with additional async steps for reads, right now */
5943 if (!(rw & REQ_WRITE)) {
5945 btrfsic_submit_bio(rw, bio);
5951 * nr_async_bios allows us to reliably return congestion to the
5952 * higher layers. Otherwise, the async bio makes it appear we have
5953 * made progress against dirty pages when we've really just put it
5954 * on a queue for later
5956 atomic_inc(&root->fs_info->nr_async_bios);
5957 WARN_ON(bio->bi_next);
5958 bio->bi_next = NULL;
5961 spin_lock(&device->io_lock);
5962 if (bio->bi_rw & REQ_SYNC)
5963 pending_bios = &device->pending_sync_bios;
5965 pending_bios = &device->pending_bios;
5967 if (pending_bios->tail)
5968 pending_bios->tail->bi_next = bio;
5970 pending_bios->tail = bio;
5971 if (!pending_bios->head)
5972 pending_bios->head = bio;
5973 if (device->running_pending)
5976 spin_unlock(&device->io_lock);
5979 btrfs_queue_work(root->fs_info->submit_workers,
5983 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5984 struct bio *bio, u64 physical, int dev_nr,
5987 struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
5989 bio->bi_private = bbio;
5990 btrfs_io_bio(bio)->stripe_index = dev_nr;
5991 bio->bi_end_io = btrfs_end_bio;
5992 bio->bi_iter.bi_sector = physical >> 9;
5995 struct rcu_string *name;
5998 name = rcu_dereference(dev->name);
5999 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
6000 "(%s id %llu), size=%u\n", rw,
6001 (u64)bio->bi_iter.bi_sector, (u_long)dev->bdev->bd_dev,
6002 name->str, dev->devid, bio->bi_iter.bi_size);
6006 bio->bi_bdev = dev->bdev;
6008 btrfs_bio_counter_inc_noblocked(root->fs_info);
6011 btrfs_schedule_bio(root, dev, rw, bio);
6013 btrfsic_submit_bio(rw, bio);
6016 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6018 atomic_inc(&bbio->error);
6019 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6020 /* Shoud be the original bio. */
6021 WARN_ON(bio != bbio->orig_bio);
6023 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6024 bio->bi_iter.bi_sector = logical >> 9;
6025 bio->bi_error = -EIO;
6026 btrfs_end_bbio(bbio, bio);
6030 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
6031 int mirror_num, int async_submit)
6033 struct btrfs_device *dev;
6034 struct bio *first_bio = bio;
6035 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
6041 struct btrfs_bio *bbio = NULL;
6043 length = bio->bi_iter.bi_size;
6044 map_length = length;
6046 btrfs_bio_counter_inc_blocked(root->fs_info);
6047 ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
6050 btrfs_bio_counter_dec(root->fs_info);
6054 total_devs = bbio->num_stripes;
6055 bbio->orig_bio = first_bio;
6056 bbio->private = first_bio->bi_private;
6057 bbio->end_io = first_bio->bi_end_io;
6058 bbio->fs_info = root->fs_info;
6059 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6061 if (bbio->raid_map) {
6062 /* In this case, map_length has been set to the length of
6063 a single stripe; not the whole write */
6065 ret = raid56_parity_write(root, bio, bbio, map_length);
6067 ret = raid56_parity_recover(root, bio, bbio, map_length,
6071 btrfs_bio_counter_dec(root->fs_info);
6075 if (map_length < length) {
6076 btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu",
6077 logical, length, map_length);
6081 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6082 dev = bbio->stripes[dev_nr].dev;
6083 if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
6084 bbio_error(bbio, first_bio, logical);
6088 if (dev_nr < total_devs - 1) {
6089 bio = btrfs_bio_clone(first_bio, GFP_NOFS);
6090 BUG_ON(!bio); /* -ENOMEM */
6094 submit_stripe_bio(root, bbio, bio,
6095 bbio->stripes[dev_nr].physical, dev_nr, rw,
6098 btrfs_bio_counter_dec(root->fs_info);
6102 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
6105 struct btrfs_device *device;
6106 struct btrfs_fs_devices *cur_devices;
6108 cur_devices = fs_info->fs_devices;
6109 while (cur_devices) {
6111 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
6112 device = __find_device(&cur_devices->devices,
6117 cur_devices = cur_devices->seed;
6122 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
6123 struct btrfs_fs_devices *fs_devices,
6124 u64 devid, u8 *dev_uuid)
6126 struct btrfs_device *device;
6128 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6132 list_add(&device->dev_list, &fs_devices->devices);
6133 device->fs_devices = fs_devices;
6134 fs_devices->num_devices++;
6136 device->missing = 1;
6137 fs_devices->missing_devices++;
6143 * btrfs_alloc_device - allocate struct btrfs_device
6144 * @fs_info: used only for generating a new devid, can be NULL if
6145 * devid is provided (i.e. @devid != NULL).
6146 * @devid: a pointer to devid for this device. If NULL a new devid
6148 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6151 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6152 * on error. Returned struct is not linked onto any lists and can be
6153 * destroyed with kfree() right away.
6155 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6159 struct btrfs_device *dev;
6162 if (WARN_ON(!devid && !fs_info))
6163 return ERR_PTR(-EINVAL);
6165 dev = __alloc_device();
6174 ret = find_next_devid(fs_info, &tmp);
6177 return ERR_PTR(ret);
6183 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6185 generate_random_uuid(dev->uuid);
6187 btrfs_init_work(&dev->work, btrfs_submit_helper,
6188 pending_bios_fn, NULL, NULL);
6193 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
6194 struct extent_buffer *leaf,
6195 struct btrfs_chunk *chunk)
6197 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
6198 struct map_lookup *map;
6199 struct extent_map *em;
6203 u8 uuid[BTRFS_UUID_SIZE];
6208 logical = key->offset;
6209 length = btrfs_chunk_length(leaf, chunk);
6211 read_lock(&map_tree->map_tree.lock);
6212 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
6213 read_unlock(&map_tree->map_tree.lock);
6215 /* already mapped? */
6216 if (em && em->start <= logical && em->start + em->len > logical) {
6217 free_extent_map(em);
6220 free_extent_map(em);
6223 em = alloc_extent_map();
6226 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6227 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6229 free_extent_map(em);
6233 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6234 em->bdev = (struct block_device *)map;
6235 em->start = logical;
6238 em->block_start = 0;
6239 em->block_len = em->len;
6241 map->num_stripes = num_stripes;
6242 map->io_width = btrfs_chunk_io_width(leaf, chunk);
6243 map->io_align = btrfs_chunk_io_align(leaf, chunk);
6244 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
6245 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6246 map->type = btrfs_chunk_type(leaf, chunk);
6247 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6248 for (i = 0; i < num_stripes; i++) {
6249 map->stripes[i].physical =
6250 btrfs_stripe_offset_nr(leaf, chunk, i);
6251 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6252 read_extent_buffer(leaf, uuid, (unsigned long)
6253 btrfs_stripe_dev_uuid_nr(chunk, i),
6255 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
6257 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
6258 free_extent_map(em);
6261 if (!map->stripes[i].dev) {
6262 map->stripes[i].dev =
6263 add_missing_dev(root, root->fs_info->fs_devices,
6265 if (!map->stripes[i].dev) {
6266 free_extent_map(em);
6269 btrfs_warn(root->fs_info, "devid %llu uuid %pU is missing",
6272 map->stripes[i].dev->in_fs_metadata = 1;
6275 write_lock(&map_tree->map_tree.lock);
6276 ret = add_extent_mapping(&map_tree->map_tree, em, 0);
6277 write_unlock(&map_tree->map_tree.lock);
6278 BUG_ON(ret); /* Tree corruption */
6279 free_extent_map(em);
6284 static void fill_device_from_item(struct extent_buffer *leaf,
6285 struct btrfs_dev_item *dev_item,
6286 struct btrfs_device *device)
6290 device->devid = btrfs_device_id(leaf, dev_item);
6291 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6292 device->total_bytes = device->disk_total_bytes;
6293 device->commit_total_bytes = device->disk_total_bytes;
6294 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6295 device->commit_bytes_used = device->bytes_used;
6296 device->type = btrfs_device_type(leaf, dev_item);
6297 device->io_align = btrfs_device_io_align(leaf, dev_item);
6298 device->io_width = btrfs_device_io_width(leaf, dev_item);
6299 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6300 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6301 device->is_tgtdev_for_dev_replace = 0;
6303 ptr = btrfs_device_uuid(dev_item);
6304 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6307 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_root *root,
6310 struct btrfs_fs_devices *fs_devices;
6313 BUG_ON(!mutex_is_locked(&uuid_mutex));
6315 fs_devices = root->fs_info->fs_devices->seed;
6316 while (fs_devices) {
6317 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE))
6320 fs_devices = fs_devices->seed;
6323 fs_devices = find_fsid(fsid);
6325 if (!btrfs_test_opt(root, DEGRADED))
6326 return ERR_PTR(-ENOENT);
6328 fs_devices = alloc_fs_devices(fsid);
6329 if (IS_ERR(fs_devices))
6332 fs_devices->seeding = 1;
6333 fs_devices->opened = 1;
6337 fs_devices = clone_fs_devices(fs_devices);
6338 if (IS_ERR(fs_devices))
6341 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
6342 root->fs_info->bdev_holder);
6344 free_fs_devices(fs_devices);
6345 fs_devices = ERR_PTR(ret);
6349 if (!fs_devices->seeding) {
6350 __btrfs_close_devices(fs_devices);
6351 free_fs_devices(fs_devices);
6352 fs_devices = ERR_PTR(-EINVAL);
6356 fs_devices->seed = root->fs_info->fs_devices->seed;
6357 root->fs_info->fs_devices->seed = fs_devices;
6362 static int read_one_dev(struct btrfs_root *root,
6363 struct extent_buffer *leaf,
6364 struct btrfs_dev_item *dev_item)
6366 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6367 struct btrfs_device *device;
6370 u8 fs_uuid[BTRFS_UUID_SIZE];
6371 u8 dev_uuid[BTRFS_UUID_SIZE];
6373 devid = btrfs_device_id(leaf, dev_item);
6374 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6376 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6379 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
6380 fs_devices = open_seed_devices(root, fs_uuid);
6381 if (IS_ERR(fs_devices))
6382 return PTR_ERR(fs_devices);
6385 device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
6387 if (!btrfs_test_opt(root, DEGRADED))
6390 device = add_missing_dev(root, fs_devices, devid, dev_uuid);
6393 btrfs_warn(root->fs_info, "devid %llu uuid %pU missing",
6396 if (!device->bdev && !btrfs_test_opt(root, DEGRADED))
6399 if(!device->bdev && !device->missing) {
6401 * this happens when a device that was properly setup
6402 * in the device info lists suddenly goes bad.
6403 * device->bdev is NULL, and so we have to set
6404 * device->missing to one here
6406 device->fs_devices->missing_devices++;
6407 device->missing = 1;
6410 /* Move the device to its own fs_devices */
6411 if (device->fs_devices != fs_devices) {
6412 ASSERT(device->missing);
6414 list_move(&device->dev_list, &fs_devices->devices);
6415 device->fs_devices->num_devices--;
6416 fs_devices->num_devices++;
6418 device->fs_devices->missing_devices--;
6419 fs_devices->missing_devices++;
6421 device->fs_devices = fs_devices;
6425 if (device->fs_devices != root->fs_info->fs_devices) {
6426 BUG_ON(device->writeable);
6427 if (device->generation !=
6428 btrfs_device_generation(leaf, dev_item))
6432 fill_device_from_item(leaf, dev_item, device);
6433 device->in_fs_metadata = 1;
6434 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
6435 device->fs_devices->total_rw_bytes += device->total_bytes;
6436 spin_lock(&root->fs_info->free_chunk_lock);
6437 root->fs_info->free_chunk_space += device->total_bytes -
6439 spin_unlock(&root->fs_info->free_chunk_lock);
6445 int btrfs_read_sys_array(struct btrfs_root *root)
6447 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
6448 struct extent_buffer *sb;
6449 struct btrfs_disk_key *disk_key;
6450 struct btrfs_chunk *chunk;
6452 unsigned long sb_array_offset;
6458 struct btrfs_key key;
6460 ASSERT(BTRFS_SUPER_INFO_SIZE <= root->nodesize);
6462 * This will create extent buffer of nodesize, superblock size is
6463 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6464 * overallocate but we can keep it as-is, only the first page is used.
6466 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET);
6469 btrfs_set_buffer_uptodate(sb);
6470 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6472 * The sb extent buffer is artifical and just used to read the system array.
6473 * btrfs_set_buffer_uptodate() call does not properly mark all it's
6474 * pages up-to-date when the page is larger: extent does not cover the
6475 * whole page and consequently check_page_uptodate does not find all
6476 * the page's extents up-to-date (the hole beyond sb),
6477 * write_extent_buffer then triggers a WARN_ON.
6479 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6480 * but sb spans only this function. Add an explicit SetPageUptodate call
6481 * to silence the warning eg. on PowerPC 64.
6483 if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
6484 SetPageUptodate(sb->pages[0]);
6486 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6487 array_size = btrfs_super_sys_array_size(super_copy);
6489 array_ptr = super_copy->sys_chunk_array;
6490 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6493 while (cur_offset < array_size) {
6494 disk_key = (struct btrfs_disk_key *)array_ptr;
6495 len = sizeof(*disk_key);
6496 if (cur_offset + len > array_size)
6497 goto out_short_read;
6499 btrfs_disk_key_to_cpu(&key, disk_key);
6502 sb_array_offset += len;
6505 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6506 chunk = (struct btrfs_chunk *)sb_array_offset;
6508 * At least one btrfs_chunk with one stripe must be
6509 * present, exact stripe count check comes afterwards
6511 len = btrfs_chunk_item_size(1);
6512 if (cur_offset + len > array_size)
6513 goto out_short_read;
6515 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6518 "BTRFS: invalid number of stripes %u in sys_array at offset %u\n",
6519 num_stripes, cur_offset);
6524 len = btrfs_chunk_item_size(num_stripes);
6525 if (cur_offset + len > array_size)
6526 goto out_short_read;
6528 ret = read_one_chunk(root, &key, sb, chunk);
6536 sb_array_offset += len;
6539 free_extent_buffer(sb);
6543 printk(KERN_ERR "BTRFS: sys_array too short to read %u bytes at offset %u\n",
6545 free_extent_buffer(sb);
6549 int btrfs_read_chunk_tree(struct btrfs_root *root)
6551 struct btrfs_path *path;
6552 struct extent_buffer *leaf;
6553 struct btrfs_key key;
6554 struct btrfs_key found_key;
6558 root = root->fs_info->chunk_root;
6560 path = btrfs_alloc_path();
6564 mutex_lock(&uuid_mutex);
6568 * Read all device items, and then all the chunk items. All
6569 * device items are found before any chunk item (their object id
6570 * is smaller than the lowest possible object id for a chunk
6571 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6573 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6576 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6580 leaf = path->nodes[0];
6581 slot = path->slots[0];
6582 if (slot >= btrfs_header_nritems(leaf)) {
6583 ret = btrfs_next_leaf(root, path);
6590 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6591 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6592 struct btrfs_dev_item *dev_item;
6593 dev_item = btrfs_item_ptr(leaf, slot,
6594 struct btrfs_dev_item);
6595 ret = read_one_dev(root, leaf, dev_item);
6598 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6599 struct btrfs_chunk *chunk;
6600 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6601 ret = read_one_chunk(root, &found_key, leaf, chunk);
6609 unlock_chunks(root);
6610 mutex_unlock(&uuid_mutex);
6612 btrfs_free_path(path);
6616 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6618 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6619 struct btrfs_device *device;
6621 while (fs_devices) {
6622 mutex_lock(&fs_devices->device_list_mutex);
6623 list_for_each_entry(device, &fs_devices->devices, dev_list)
6624 device->dev_root = fs_info->dev_root;
6625 mutex_unlock(&fs_devices->device_list_mutex);
6627 fs_devices = fs_devices->seed;
6631 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6635 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6636 btrfs_dev_stat_reset(dev, i);
6639 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6641 struct btrfs_key key;
6642 struct btrfs_key found_key;
6643 struct btrfs_root *dev_root = fs_info->dev_root;
6644 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6645 struct extent_buffer *eb;
6648 struct btrfs_device *device;
6649 struct btrfs_path *path = NULL;
6652 path = btrfs_alloc_path();
6658 mutex_lock(&fs_devices->device_list_mutex);
6659 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6661 struct btrfs_dev_stats_item *ptr;
6664 key.type = BTRFS_DEV_STATS_KEY;
6665 key.offset = device->devid;
6666 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6668 __btrfs_reset_dev_stats(device);
6669 device->dev_stats_valid = 1;
6670 btrfs_release_path(path);
6673 slot = path->slots[0];
6674 eb = path->nodes[0];
6675 btrfs_item_key_to_cpu(eb, &found_key, slot);
6676 item_size = btrfs_item_size_nr(eb, slot);
6678 ptr = btrfs_item_ptr(eb, slot,
6679 struct btrfs_dev_stats_item);
6681 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6682 if (item_size >= (1 + i) * sizeof(__le64))
6683 btrfs_dev_stat_set(device, i,
6684 btrfs_dev_stats_value(eb, ptr, i));
6686 btrfs_dev_stat_reset(device, i);
6689 device->dev_stats_valid = 1;
6690 btrfs_dev_stat_print_on_load(device);
6691 btrfs_release_path(path);
6693 mutex_unlock(&fs_devices->device_list_mutex);
6696 btrfs_free_path(path);
6697 return ret < 0 ? ret : 0;
6700 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
6701 struct btrfs_root *dev_root,
6702 struct btrfs_device *device)
6704 struct btrfs_path *path;
6705 struct btrfs_key key;
6706 struct extent_buffer *eb;
6707 struct btrfs_dev_stats_item *ptr;
6712 key.type = BTRFS_DEV_STATS_KEY;
6713 key.offset = device->devid;
6715 path = btrfs_alloc_path();
6717 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
6719 btrfs_warn_in_rcu(dev_root->fs_info,
6720 "error %d while searching for dev_stats item for device %s",
6721 ret, rcu_str_deref(device->name));
6726 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
6727 /* need to delete old one and insert a new one */
6728 ret = btrfs_del_item(trans, dev_root, path);
6730 btrfs_warn_in_rcu(dev_root->fs_info,
6731 "delete too small dev_stats item for device %s failed %d",
6732 rcu_str_deref(device->name), ret);
6739 /* need to insert a new item */
6740 btrfs_release_path(path);
6741 ret = btrfs_insert_empty_item(trans, dev_root, path,
6742 &key, sizeof(*ptr));
6744 btrfs_warn_in_rcu(dev_root->fs_info,
6745 "insert dev_stats item for device %s failed %d",
6746 rcu_str_deref(device->name), ret);
6751 eb = path->nodes[0];
6752 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
6753 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6754 btrfs_set_dev_stats_value(eb, ptr, i,
6755 btrfs_dev_stat_read(device, i));
6756 btrfs_mark_buffer_dirty(eb);
6759 btrfs_free_path(path);
6764 * called from commit_transaction. Writes all changed device stats to disk.
6766 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
6767 struct btrfs_fs_info *fs_info)
6769 struct btrfs_root *dev_root = fs_info->dev_root;
6770 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6771 struct btrfs_device *device;
6775 mutex_lock(&fs_devices->device_list_mutex);
6776 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6777 if (!device->dev_stats_valid || !btrfs_dev_stats_dirty(device))
6780 stats_cnt = atomic_read(&device->dev_stats_ccnt);
6781 ret = update_dev_stat_item(trans, dev_root, device);
6783 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
6785 mutex_unlock(&fs_devices->device_list_mutex);
6790 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
6792 btrfs_dev_stat_inc(dev, index);
6793 btrfs_dev_stat_print_on_error(dev);
6796 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
6798 if (!dev->dev_stats_valid)
6800 btrfs_err_rl_in_rcu(dev->dev_root->fs_info,
6801 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
6802 rcu_str_deref(dev->name),
6803 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6804 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6805 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6806 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6807 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6810 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
6814 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6815 if (btrfs_dev_stat_read(dev, i) != 0)
6817 if (i == BTRFS_DEV_STAT_VALUES_MAX)
6818 return; /* all values == 0, suppress message */
6820 btrfs_info_in_rcu(dev->dev_root->fs_info,
6821 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
6822 rcu_str_deref(dev->name),
6823 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6824 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6825 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6826 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6827 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6830 int btrfs_get_dev_stats(struct btrfs_root *root,
6831 struct btrfs_ioctl_get_dev_stats *stats)
6833 struct btrfs_device *dev;
6834 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6837 mutex_lock(&fs_devices->device_list_mutex);
6838 dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
6839 mutex_unlock(&fs_devices->device_list_mutex);
6842 btrfs_warn(root->fs_info, "get dev_stats failed, device not found");
6844 } else if (!dev->dev_stats_valid) {
6845 btrfs_warn(root->fs_info, "get dev_stats failed, not yet valid");
6847 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
6848 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6849 if (stats->nr_items > i)
6851 btrfs_dev_stat_read_and_reset(dev, i);
6853 btrfs_dev_stat_reset(dev, i);
6856 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6857 if (stats->nr_items > i)
6858 stats->values[i] = btrfs_dev_stat_read(dev, i);
6860 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
6861 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
6865 void btrfs_scratch_superblocks(struct block_device *bdev, char *device_path)
6867 struct buffer_head *bh;
6868 struct btrfs_super_block *disk_super;
6874 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX;
6877 if (btrfs_read_dev_one_super(bdev, copy_num, &bh))
6880 disk_super = (struct btrfs_super_block *)bh->b_data;
6882 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
6883 set_buffer_dirty(bh);
6884 sync_dirty_buffer(bh);
6888 /* Notify udev that device has changed */
6889 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
6891 /* Update ctime/mtime for device path for libblkid */
6892 update_dev_time(device_path);
6896 * Update the size of all devices, which is used for writing out the
6899 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
6901 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6902 struct btrfs_device *curr, *next;
6904 if (list_empty(&fs_devices->resized_devices))
6907 mutex_lock(&fs_devices->device_list_mutex);
6908 lock_chunks(fs_info->dev_root);
6909 list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
6911 list_del_init(&curr->resized_list);
6912 curr->commit_total_bytes = curr->disk_total_bytes;
6914 unlock_chunks(fs_info->dev_root);
6915 mutex_unlock(&fs_devices->device_list_mutex);
6918 /* Must be invoked during the transaction commit */
6919 void btrfs_update_commit_device_bytes_used(struct btrfs_root *root,
6920 struct btrfs_transaction *transaction)
6922 struct extent_map *em;
6923 struct map_lookup *map;
6924 struct btrfs_device *dev;
6927 if (list_empty(&transaction->pending_chunks))
6930 /* In order to kick the device replace finish process */
6932 list_for_each_entry(em, &transaction->pending_chunks, list) {
6933 map = (struct map_lookup *)em->bdev;
6935 for (i = 0; i < map->num_stripes; i++) {
6936 dev = map->stripes[i].dev;
6937 dev->commit_bytes_used = dev->bytes_used;
6940 unlock_chunks(root);
6943 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
6945 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6946 while (fs_devices) {
6947 fs_devices->fs_info = fs_info;
6948 fs_devices = fs_devices->seed;
6952 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
6954 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6955 while (fs_devices) {
6956 fs_devices->fs_info = NULL;
6957 fs_devices = fs_devices->seed;
6961 void btrfs_close_one_device(struct btrfs_device *device)
6963 struct btrfs_fs_devices *fs_devices = device->fs_devices;
6964 struct btrfs_device *new_device;
6965 struct rcu_string *name;
6968 fs_devices->open_devices--;
6970 if (device->writeable &&
6971 device->devid != BTRFS_DEV_REPLACE_DEVID) {
6972 list_del_init(&device->dev_alloc_list);
6973 fs_devices->rw_devices--;
6976 if (device->missing)
6977 fs_devices->missing_devices--;
6979 new_device = btrfs_alloc_device(NULL, &device->devid,
6981 BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
6983 /* Safe because we are under uuid_mutex */
6985 name = rcu_string_strdup(device->name->str, GFP_NOFS);
6986 BUG_ON(!name); /* -ENOMEM */
6987 rcu_assign_pointer(new_device->name, name);
6990 list_replace_rcu(&device->dev_list, &new_device->dev_list);
6991 new_device->fs_devices = device->fs_devices;
6993 call_rcu(&device->rcu, free_device);