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);
128 static void btrfs_close_one_device(struct btrfs_device *device);
130 DEFINE_MUTEX(uuid_mutex);
131 static LIST_HEAD(fs_uuids);
132 struct list_head *btrfs_get_fs_uuids(void)
137 static struct btrfs_fs_devices *__alloc_fs_devices(void)
139 struct btrfs_fs_devices *fs_devs;
141 fs_devs = kzalloc(sizeof(*fs_devs), GFP_NOFS);
143 return ERR_PTR(-ENOMEM);
145 mutex_init(&fs_devs->device_list_mutex);
147 INIT_LIST_HEAD(&fs_devs->devices);
148 INIT_LIST_HEAD(&fs_devs->resized_devices);
149 INIT_LIST_HEAD(&fs_devs->alloc_list);
150 INIT_LIST_HEAD(&fs_devs->list);
156 * alloc_fs_devices - allocate struct btrfs_fs_devices
157 * @fsid: a pointer to UUID for this FS. If NULL a new UUID is
160 * Return: a pointer to a new &struct btrfs_fs_devices on success;
161 * ERR_PTR() on error. Returned struct is not linked onto any lists and
162 * can be destroyed with kfree() right away.
164 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
166 struct btrfs_fs_devices *fs_devs;
168 fs_devs = __alloc_fs_devices();
173 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
175 generate_random_uuid(fs_devs->fsid);
180 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
182 struct btrfs_device *device;
183 WARN_ON(fs_devices->opened);
184 while (!list_empty(&fs_devices->devices)) {
185 device = list_entry(fs_devices->devices.next,
186 struct btrfs_device, dev_list);
187 list_del(&device->dev_list);
188 rcu_string_free(device->name);
194 static void btrfs_kobject_uevent(struct block_device *bdev,
195 enum kobject_action action)
199 ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
201 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
203 kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
204 &disk_to_dev(bdev->bd_disk)->kobj);
207 void btrfs_cleanup_fs_uuids(void)
209 struct btrfs_fs_devices *fs_devices;
211 while (!list_empty(&fs_uuids)) {
212 fs_devices = list_entry(fs_uuids.next,
213 struct btrfs_fs_devices, list);
214 list_del(&fs_devices->list);
215 free_fs_devices(fs_devices);
219 static struct btrfs_device *__alloc_device(void)
221 struct btrfs_device *dev;
223 dev = kzalloc(sizeof(*dev), GFP_NOFS);
225 return ERR_PTR(-ENOMEM);
227 INIT_LIST_HEAD(&dev->dev_list);
228 INIT_LIST_HEAD(&dev->dev_alloc_list);
229 INIT_LIST_HEAD(&dev->resized_list);
231 spin_lock_init(&dev->io_lock);
233 spin_lock_init(&dev->reada_lock);
234 atomic_set(&dev->reada_in_flight, 0);
235 atomic_set(&dev->dev_stats_ccnt, 0);
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();
1106 path->reada = READA_FORWARD;
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;
1262 path = btrfs_alloc_path();
1266 max_hole_start = search_start;
1270 if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1275 path->reada = READA_FORWARD;
1276 path->search_commit_root = 1;
1277 path->skip_locking = 1;
1279 key.objectid = device->devid;
1280 key.offset = search_start;
1281 key.type = BTRFS_DEV_EXTENT_KEY;
1283 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1287 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1294 slot = path->slots[0];
1295 if (slot >= btrfs_header_nritems(l)) {
1296 ret = btrfs_next_leaf(root, path);
1304 btrfs_item_key_to_cpu(l, &key, slot);
1306 if (key.objectid < device->devid)
1309 if (key.objectid > device->devid)
1312 if (key.type != BTRFS_DEV_EXTENT_KEY)
1315 if (key.offset > search_start) {
1316 hole_size = key.offset - search_start;
1319 * Have to check before we set max_hole_start, otherwise
1320 * we could end up sending back this offset anyway.
1322 if (contains_pending_extent(transaction, device,
1325 if (key.offset >= search_start) {
1326 hole_size = key.offset - search_start;
1333 if (hole_size > max_hole_size) {
1334 max_hole_start = search_start;
1335 max_hole_size = hole_size;
1339 * If this free space is greater than which we need,
1340 * it must be the max free space that we have found
1341 * until now, so max_hole_start must point to the start
1342 * of this free space and the length of this free space
1343 * is stored in max_hole_size. Thus, we return
1344 * max_hole_start and max_hole_size and go back to the
1347 if (hole_size >= num_bytes) {
1353 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1354 extent_end = key.offset + btrfs_dev_extent_length(l,
1356 if (extent_end > search_start)
1357 search_start = extent_end;
1364 * At this point, search_start should be the end of
1365 * allocated dev extents, and when shrinking the device,
1366 * search_end may be smaller than search_start.
1368 if (search_end > search_start) {
1369 hole_size = search_end - search_start;
1371 if (contains_pending_extent(transaction, device, &search_start,
1373 btrfs_release_path(path);
1377 if (hole_size > max_hole_size) {
1378 max_hole_start = search_start;
1379 max_hole_size = hole_size;
1384 if (max_hole_size < num_bytes)
1390 btrfs_free_path(path);
1391 *start = max_hole_start;
1393 *len = max_hole_size;
1397 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1398 struct btrfs_device *device, u64 num_bytes,
1399 u64 *start, u64 *len)
1401 struct btrfs_root *root = device->dev_root;
1404 /* FIXME use last free of some kind */
1407 * we don't want to overwrite the superblock on the drive,
1408 * so we make sure to start at an offset of at least 1MB
1410 search_start = max_t(u64, root->fs_info->alloc_start, SZ_1M);
1411 return find_free_dev_extent_start(trans->transaction, device,
1412 num_bytes, search_start, 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);
1648 static int btrfs_rm_dev_item(struct btrfs_root *root,
1649 struct btrfs_device *device)
1652 struct btrfs_path *path;
1653 struct btrfs_key key;
1654 struct btrfs_trans_handle *trans;
1656 root = root->fs_info->chunk_root;
1658 path = btrfs_alloc_path();
1662 trans = btrfs_start_transaction(root, 0);
1663 if (IS_ERR(trans)) {
1664 btrfs_free_path(path);
1665 return PTR_ERR(trans);
1667 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1668 key.type = BTRFS_DEV_ITEM_KEY;
1669 key.offset = device->devid;
1671 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1680 ret = btrfs_del_item(trans, root, path);
1684 btrfs_free_path(path);
1685 btrfs_commit_transaction(trans, root);
1689 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1691 struct btrfs_device *device;
1692 struct btrfs_device *next_device;
1693 struct block_device *bdev;
1694 struct buffer_head *bh = NULL;
1695 struct btrfs_super_block *disk_super;
1696 struct btrfs_fs_devices *cur_devices;
1703 bool clear_super = false;
1705 mutex_lock(&uuid_mutex);
1708 seq = read_seqbegin(&root->fs_info->profiles_lock);
1710 all_avail = root->fs_info->avail_data_alloc_bits |
1711 root->fs_info->avail_system_alloc_bits |
1712 root->fs_info->avail_metadata_alloc_bits;
1713 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
1715 num_devices = root->fs_info->fs_devices->num_devices;
1716 btrfs_dev_replace_lock(&root->fs_info->dev_replace);
1717 if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1718 WARN_ON(num_devices < 1);
1721 btrfs_dev_replace_unlock(&root->fs_info->dev_replace);
1723 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && num_devices <= 4) {
1724 ret = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET;
1728 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && num_devices <= 2) {
1729 ret = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET;
1733 if ((all_avail & BTRFS_BLOCK_GROUP_RAID5) &&
1734 root->fs_info->fs_devices->rw_devices <= 2) {
1735 ret = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET;
1738 if ((all_avail & BTRFS_BLOCK_GROUP_RAID6) &&
1739 root->fs_info->fs_devices->rw_devices <= 3) {
1740 ret = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET;
1744 if (strcmp(device_path, "missing") == 0) {
1745 struct list_head *devices;
1746 struct btrfs_device *tmp;
1749 devices = &root->fs_info->fs_devices->devices;
1751 * It is safe to read the devices since the volume_mutex
1754 list_for_each_entry(tmp, devices, dev_list) {
1755 if (tmp->in_fs_metadata &&
1756 !tmp->is_tgtdev_for_dev_replace &&
1766 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
1770 ret = btrfs_get_bdev_and_sb(device_path,
1771 FMODE_WRITE | FMODE_EXCL,
1772 root->fs_info->bdev_holder, 0,
1776 disk_super = (struct btrfs_super_block *)bh->b_data;
1777 devid = btrfs_stack_device_id(&disk_super->dev_item);
1778 dev_uuid = disk_super->dev_item.uuid;
1779 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1787 if (device->is_tgtdev_for_dev_replace) {
1788 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1792 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1793 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1797 if (device->writeable) {
1799 list_del_init(&device->dev_alloc_list);
1800 device->fs_devices->rw_devices--;
1801 unlock_chunks(root);
1805 mutex_unlock(&uuid_mutex);
1806 ret = btrfs_shrink_device(device, 0);
1807 mutex_lock(&uuid_mutex);
1812 * TODO: the superblock still includes this device in its num_devices
1813 * counter although write_all_supers() is not locked out. This
1814 * could give a filesystem state which requires a degraded mount.
1816 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1820 device->in_fs_metadata = 0;
1821 btrfs_scrub_cancel_dev(root->fs_info, device);
1824 * the device list mutex makes sure that we don't change
1825 * the device list while someone else is writing out all
1826 * the device supers. Whoever is writing all supers, should
1827 * lock the device list mutex before getting the number of
1828 * devices in the super block (super_copy). Conversely,
1829 * whoever updates the number of devices in the super block
1830 * (super_copy) should hold the device list mutex.
1833 cur_devices = device->fs_devices;
1834 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1835 list_del_rcu(&device->dev_list);
1837 device->fs_devices->num_devices--;
1838 device->fs_devices->total_devices--;
1840 if (device->missing)
1841 device->fs_devices->missing_devices--;
1843 next_device = list_entry(root->fs_info->fs_devices->devices.next,
1844 struct btrfs_device, dev_list);
1845 if (device->bdev == root->fs_info->sb->s_bdev)
1846 root->fs_info->sb->s_bdev = next_device->bdev;
1847 if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1848 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1851 device->fs_devices->open_devices--;
1852 /* remove sysfs entry */
1853 btrfs_sysfs_rm_device_link(root->fs_info->fs_devices, device);
1856 call_rcu(&device->rcu, free_device);
1858 num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1859 btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1860 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1862 if (cur_devices->open_devices == 0) {
1863 struct btrfs_fs_devices *fs_devices;
1864 fs_devices = root->fs_info->fs_devices;
1865 while (fs_devices) {
1866 if (fs_devices->seed == cur_devices) {
1867 fs_devices->seed = cur_devices->seed;
1870 fs_devices = fs_devices->seed;
1872 cur_devices->seed = NULL;
1873 __btrfs_close_devices(cur_devices);
1874 free_fs_devices(cur_devices);
1877 root->fs_info->num_tolerated_disk_barrier_failures =
1878 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1881 * at this point, the device is zero sized. We want to
1882 * remove it from the devices list and zero out the old super
1884 if (clear_super && disk_super) {
1888 /* make sure this device isn't detected as part of
1891 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1892 set_buffer_dirty(bh);
1893 sync_dirty_buffer(bh);
1895 /* clear the mirror copies of super block on the disk
1896 * being removed, 0th copy is been taken care above and
1897 * the below would take of the rest
1899 for (i = 1; i < BTRFS_SUPER_MIRROR_MAX; i++) {
1900 bytenr = btrfs_sb_offset(i);
1901 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
1902 i_size_read(bdev->bd_inode))
1906 bh = __bread(bdev, bytenr / 4096,
1907 BTRFS_SUPER_INFO_SIZE);
1911 disk_super = (struct btrfs_super_block *)bh->b_data;
1913 if (btrfs_super_bytenr(disk_super) != bytenr ||
1914 btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
1917 memset(&disk_super->magic, 0,
1918 sizeof(disk_super->magic));
1919 set_buffer_dirty(bh);
1920 sync_dirty_buffer(bh);
1927 /* Notify udev that device has changed */
1928 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
1930 /* Update ctime/mtime for device path for libblkid */
1931 update_dev_time(device_path);
1937 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1939 mutex_unlock(&uuid_mutex);
1942 if (device->writeable) {
1944 list_add(&device->dev_alloc_list,
1945 &root->fs_info->fs_devices->alloc_list);
1946 device->fs_devices->rw_devices++;
1947 unlock_chunks(root);
1952 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
1953 struct btrfs_device *srcdev)
1955 struct btrfs_fs_devices *fs_devices;
1957 WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1960 * in case of fs with no seed, srcdev->fs_devices will point
1961 * to fs_devices of fs_info. However when the dev being replaced is
1962 * a seed dev it will point to the seed's local fs_devices. In short
1963 * srcdev will have its correct fs_devices in both the cases.
1965 fs_devices = srcdev->fs_devices;
1967 list_del_rcu(&srcdev->dev_list);
1968 list_del_rcu(&srcdev->dev_alloc_list);
1969 fs_devices->num_devices--;
1970 if (srcdev->missing)
1971 fs_devices->missing_devices--;
1973 if (srcdev->writeable) {
1974 fs_devices->rw_devices--;
1975 /* zero out the old super if it is writable */
1976 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
1980 fs_devices->open_devices--;
1983 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
1984 struct btrfs_device *srcdev)
1986 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
1988 call_rcu(&srcdev->rcu, free_device);
1991 * unless fs_devices is seed fs, num_devices shouldn't go
1994 BUG_ON(!fs_devices->num_devices && !fs_devices->seeding);
1996 /* if this is no devs we rather delete the fs_devices */
1997 if (!fs_devices->num_devices) {
1998 struct btrfs_fs_devices *tmp_fs_devices;
2000 tmp_fs_devices = fs_info->fs_devices;
2001 while (tmp_fs_devices) {
2002 if (tmp_fs_devices->seed == fs_devices) {
2003 tmp_fs_devices->seed = fs_devices->seed;
2006 tmp_fs_devices = tmp_fs_devices->seed;
2008 fs_devices->seed = NULL;
2009 __btrfs_close_devices(fs_devices);
2010 free_fs_devices(fs_devices);
2014 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2015 struct btrfs_device *tgtdev)
2017 struct btrfs_device *next_device;
2019 mutex_lock(&uuid_mutex);
2021 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2023 btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev);
2026 btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
2027 fs_info->fs_devices->open_devices--;
2029 fs_info->fs_devices->num_devices--;
2031 next_device = list_entry(fs_info->fs_devices->devices.next,
2032 struct btrfs_device, dev_list);
2033 if (tgtdev->bdev == fs_info->sb->s_bdev)
2034 fs_info->sb->s_bdev = next_device->bdev;
2035 if (tgtdev->bdev == fs_info->fs_devices->latest_bdev)
2036 fs_info->fs_devices->latest_bdev = next_device->bdev;
2037 list_del_rcu(&tgtdev->dev_list);
2039 call_rcu(&tgtdev->rcu, free_device);
2041 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2042 mutex_unlock(&uuid_mutex);
2045 static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
2046 struct btrfs_device **device)
2049 struct btrfs_super_block *disk_super;
2052 struct block_device *bdev;
2053 struct buffer_head *bh;
2056 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2057 root->fs_info->bdev_holder, 0, &bdev, &bh);
2060 disk_super = (struct btrfs_super_block *)bh->b_data;
2061 devid = btrfs_stack_device_id(&disk_super->dev_item);
2062 dev_uuid = disk_super->dev_item.uuid;
2063 *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2068 blkdev_put(bdev, FMODE_READ);
2072 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
2074 struct btrfs_device **device)
2077 if (strcmp(device_path, "missing") == 0) {
2078 struct list_head *devices;
2079 struct btrfs_device *tmp;
2081 devices = &root->fs_info->fs_devices->devices;
2083 * It is safe to read the devices since the volume_mutex
2084 * is held by the caller.
2086 list_for_each_entry(tmp, devices, dev_list) {
2087 if (tmp->in_fs_metadata && !tmp->bdev) {
2094 return BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2098 return btrfs_find_device_by_path(root, device_path, device);
2103 * does all the dirty work required for changing file system's UUID.
2105 static int btrfs_prepare_sprout(struct btrfs_root *root)
2107 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2108 struct btrfs_fs_devices *old_devices;
2109 struct btrfs_fs_devices *seed_devices;
2110 struct btrfs_super_block *disk_super = root->fs_info->super_copy;
2111 struct btrfs_device *device;
2114 BUG_ON(!mutex_is_locked(&uuid_mutex));
2115 if (!fs_devices->seeding)
2118 seed_devices = __alloc_fs_devices();
2119 if (IS_ERR(seed_devices))
2120 return PTR_ERR(seed_devices);
2122 old_devices = clone_fs_devices(fs_devices);
2123 if (IS_ERR(old_devices)) {
2124 kfree(seed_devices);
2125 return PTR_ERR(old_devices);
2128 list_add(&old_devices->list, &fs_uuids);
2130 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2131 seed_devices->opened = 1;
2132 INIT_LIST_HEAD(&seed_devices->devices);
2133 INIT_LIST_HEAD(&seed_devices->alloc_list);
2134 mutex_init(&seed_devices->device_list_mutex);
2136 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2137 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2139 list_for_each_entry(device, &seed_devices->devices, dev_list)
2140 device->fs_devices = seed_devices;
2143 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2144 unlock_chunks(root);
2146 fs_devices->seeding = 0;
2147 fs_devices->num_devices = 0;
2148 fs_devices->open_devices = 0;
2149 fs_devices->missing_devices = 0;
2150 fs_devices->rotating = 0;
2151 fs_devices->seed = seed_devices;
2153 generate_random_uuid(fs_devices->fsid);
2154 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2155 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2156 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2158 super_flags = btrfs_super_flags(disk_super) &
2159 ~BTRFS_SUPER_FLAG_SEEDING;
2160 btrfs_set_super_flags(disk_super, super_flags);
2166 * strore the expected generation for seed devices in device items.
2168 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2169 struct btrfs_root *root)
2171 struct btrfs_path *path;
2172 struct extent_buffer *leaf;
2173 struct btrfs_dev_item *dev_item;
2174 struct btrfs_device *device;
2175 struct btrfs_key key;
2176 u8 fs_uuid[BTRFS_UUID_SIZE];
2177 u8 dev_uuid[BTRFS_UUID_SIZE];
2181 path = btrfs_alloc_path();
2185 root = root->fs_info->chunk_root;
2186 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2188 key.type = BTRFS_DEV_ITEM_KEY;
2191 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2195 leaf = path->nodes[0];
2197 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2198 ret = btrfs_next_leaf(root, path);
2203 leaf = path->nodes[0];
2204 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2205 btrfs_release_path(path);
2209 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2210 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2211 key.type != BTRFS_DEV_ITEM_KEY)
2214 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2215 struct btrfs_dev_item);
2216 devid = btrfs_device_id(leaf, dev_item);
2217 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2219 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2221 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2223 BUG_ON(!device); /* Logic error */
2225 if (device->fs_devices->seeding) {
2226 btrfs_set_device_generation(leaf, dev_item,
2227 device->generation);
2228 btrfs_mark_buffer_dirty(leaf);
2236 btrfs_free_path(path);
2240 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
2242 struct request_queue *q;
2243 struct btrfs_trans_handle *trans;
2244 struct btrfs_device *device;
2245 struct block_device *bdev;
2246 struct list_head *devices;
2247 struct super_block *sb = root->fs_info->sb;
2248 struct rcu_string *name;
2250 int seeding_dev = 0;
2253 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
2256 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2257 root->fs_info->bdev_holder);
2259 return PTR_ERR(bdev);
2261 if (root->fs_info->fs_devices->seeding) {
2263 down_write(&sb->s_umount);
2264 mutex_lock(&uuid_mutex);
2267 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2269 devices = &root->fs_info->fs_devices->devices;
2271 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2272 list_for_each_entry(device, devices, dev_list) {
2273 if (device->bdev == bdev) {
2276 &root->fs_info->fs_devices->device_list_mutex);
2280 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2282 device = btrfs_alloc_device(root->fs_info, NULL, NULL);
2283 if (IS_ERR(device)) {
2284 /* we can safely leave the fs_devices entry around */
2285 ret = PTR_ERR(device);
2289 name = rcu_string_strdup(device_path, GFP_NOFS);
2295 rcu_assign_pointer(device->name, name);
2297 trans = btrfs_start_transaction(root, 0);
2298 if (IS_ERR(trans)) {
2299 rcu_string_free(device->name);
2301 ret = PTR_ERR(trans);
2305 q = bdev_get_queue(bdev);
2306 if (blk_queue_discard(q))
2307 device->can_discard = 1;
2308 device->writeable = 1;
2309 device->generation = trans->transid;
2310 device->io_width = root->sectorsize;
2311 device->io_align = root->sectorsize;
2312 device->sector_size = root->sectorsize;
2313 device->total_bytes = i_size_read(bdev->bd_inode);
2314 device->disk_total_bytes = device->total_bytes;
2315 device->commit_total_bytes = device->total_bytes;
2316 device->dev_root = root->fs_info->dev_root;
2317 device->bdev = bdev;
2318 device->in_fs_metadata = 1;
2319 device->is_tgtdev_for_dev_replace = 0;
2320 device->mode = FMODE_EXCL;
2321 device->dev_stats_valid = 1;
2322 set_blocksize(device->bdev, 4096);
2325 sb->s_flags &= ~MS_RDONLY;
2326 ret = btrfs_prepare_sprout(root);
2327 BUG_ON(ret); /* -ENOMEM */
2330 device->fs_devices = root->fs_info->fs_devices;
2332 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2334 list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
2335 list_add(&device->dev_alloc_list,
2336 &root->fs_info->fs_devices->alloc_list);
2337 root->fs_info->fs_devices->num_devices++;
2338 root->fs_info->fs_devices->open_devices++;
2339 root->fs_info->fs_devices->rw_devices++;
2340 root->fs_info->fs_devices->total_devices++;
2341 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2343 spin_lock(&root->fs_info->free_chunk_lock);
2344 root->fs_info->free_chunk_space += device->total_bytes;
2345 spin_unlock(&root->fs_info->free_chunk_lock);
2347 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2348 root->fs_info->fs_devices->rotating = 1;
2350 tmp = btrfs_super_total_bytes(root->fs_info->super_copy);
2351 btrfs_set_super_total_bytes(root->fs_info->super_copy,
2352 tmp + device->total_bytes);
2354 tmp = btrfs_super_num_devices(root->fs_info->super_copy);
2355 btrfs_set_super_num_devices(root->fs_info->super_copy,
2358 /* add sysfs device entry */
2359 btrfs_sysfs_add_device_link(root->fs_info->fs_devices, device);
2362 * we've got more storage, clear any full flags on the space
2365 btrfs_clear_space_info_full(root->fs_info);
2367 unlock_chunks(root);
2368 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2372 ret = init_first_rw_device(trans, root, device);
2373 unlock_chunks(root);
2375 btrfs_abort_transaction(trans, root, ret);
2380 ret = btrfs_add_device(trans, root, device);
2382 btrfs_abort_transaction(trans, root, ret);
2387 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2389 ret = btrfs_finish_sprout(trans, root);
2391 btrfs_abort_transaction(trans, root, ret);
2395 /* Sprouting would change fsid of the mounted root,
2396 * so rename the fsid on the sysfs
2398 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2399 root->fs_info->fsid);
2400 if (kobject_rename(&root->fs_info->fs_devices->fsid_kobj,
2402 btrfs_warn(root->fs_info,
2403 "sysfs: failed to create fsid for sprout");
2406 root->fs_info->num_tolerated_disk_barrier_failures =
2407 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
2408 ret = btrfs_commit_transaction(trans, root);
2411 mutex_unlock(&uuid_mutex);
2412 up_write(&sb->s_umount);
2414 if (ret) /* transaction commit */
2417 ret = btrfs_relocate_sys_chunks(root);
2419 btrfs_std_error(root->fs_info, ret,
2420 "Failed to relocate sys chunks after "
2421 "device initialization. This can be fixed "
2422 "using the \"btrfs balance\" command.");
2423 trans = btrfs_attach_transaction(root);
2424 if (IS_ERR(trans)) {
2425 if (PTR_ERR(trans) == -ENOENT)
2427 return PTR_ERR(trans);
2429 ret = btrfs_commit_transaction(trans, root);
2432 /* Update ctime/mtime for libblkid */
2433 update_dev_time(device_path);
2437 btrfs_end_transaction(trans, root);
2438 rcu_string_free(device->name);
2439 btrfs_sysfs_rm_device_link(root->fs_info->fs_devices, device);
2442 blkdev_put(bdev, FMODE_EXCL);
2444 mutex_unlock(&uuid_mutex);
2445 up_write(&sb->s_umount);
2450 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2451 struct btrfs_device *srcdev,
2452 struct btrfs_device **device_out)
2454 struct request_queue *q;
2455 struct btrfs_device *device;
2456 struct block_device *bdev;
2457 struct btrfs_fs_info *fs_info = root->fs_info;
2458 struct list_head *devices;
2459 struct rcu_string *name;
2460 u64 devid = BTRFS_DEV_REPLACE_DEVID;
2464 if (fs_info->fs_devices->seeding) {
2465 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
2469 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2470 fs_info->bdev_holder);
2472 btrfs_err(fs_info, "target device %s is invalid!", device_path);
2473 return PTR_ERR(bdev);
2476 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2478 devices = &fs_info->fs_devices->devices;
2479 list_for_each_entry(device, devices, dev_list) {
2480 if (device->bdev == bdev) {
2481 btrfs_err(fs_info, "target device is in the filesystem!");
2488 if (i_size_read(bdev->bd_inode) <
2489 btrfs_device_get_total_bytes(srcdev)) {
2490 btrfs_err(fs_info, "target device is smaller than source device!");
2496 device = btrfs_alloc_device(NULL, &devid, NULL);
2497 if (IS_ERR(device)) {
2498 ret = PTR_ERR(device);
2502 name = rcu_string_strdup(device_path, GFP_NOFS);
2508 rcu_assign_pointer(device->name, name);
2510 q = bdev_get_queue(bdev);
2511 if (blk_queue_discard(q))
2512 device->can_discard = 1;
2513 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2514 device->writeable = 1;
2515 device->generation = 0;
2516 device->io_width = root->sectorsize;
2517 device->io_align = root->sectorsize;
2518 device->sector_size = root->sectorsize;
2519 device->total_bytes = btrfs_device_get_total_bytes(srcdev);
2520 device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
2521 device->bytes_used = btrfs_device_get_bytes_used(srcdev);
2522 ASSERT(list_empty(&srcdev->resized_list));
2523 device->commit_total_bytes = srcdev->commit_total_bytes;
2524 device->commit_bytes_used = device->bytes_used;
2525 device->dev_root = fs_info->dev_root;
2526 device->bdev = bdev;
2527 device->in_fs_metadata = 1;
2528 device->is_tgtdev_for_dev_replace = 1;
2529 device->mode = FMODE_EXCL;
2530 device->dev_stats_valid = 1;
2531 set_blocksize(device->bdev, 4096);
2532 device->fs_devices = fs_info->fs_devices;
2533 list_add(&device->dev_list, &fs_info->fs_devices->devices);
2534 fs_info->fs_devices->num_devices++;
2535 fs_info->fs_devices->open_devices++;
2536 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2538 *device_out = device;
2542 blkdev_put(bdev, FMODE_EXCL);
2546 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2547 struct btrfs_device *tgtdev)
2549 WARN_ON(fs_info->fs_devices->rw_devices == 0);
2550 tgtdev->io_width = fs_info->dev_root->sectorsize;
2551 tgtdev->io_align = fs_info->dev_root->sectorsize;
2552 tgtdev->sector_size = fs_info->dev_root->sectorsize;
2553 tgtdev->dev_root = fs_info->dev_root;
2554 tgtdev->in_fs_metadata = 1;
2557 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2558 struct btrfs_device *device)
2561 struct btrfs_path *path;
2562 struct btrfs_root *root;
2563 struct btrfs_dev_item *dev_item;
2564 struct extent_buffer *leaf;
2565 struct btrfs_key key;
2567 root = device->dev_root->fs_info->chunk_root;
2569 path = btrfs_alloc_path();
2573 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2574 key.type = BTRFS_DEV_ITEM_KEY;
2575 key.offset = device->devid;
2577 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2586 leaf = path->nodes[0];
2587 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2589 btrfs_set_device_id(leaf, dev_item, device->devid);
2590 btrfs_set_device_type(leaf, dev_item, device->type);
2591 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2592 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2593 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2594 btrfs_set_device_total_bytes(leaf, dev_item,
2595 btrfs_device_get_disk_total_bytes(device));
2596 btrfs_set_device_bytes_used(leaf, dev_item,
2597 btrfs_device_get_bytes_used(device));
2598 btrfs_mark_buffer_dirty(leaf);
2601 btrfs_free_path(path);
2605 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2606 struct btrfs_device *device, u64 new_size)
2608 struct btrfs_super_block *super_copy =
2609 device->dev_root->fs_info->super_copy;
2610 struct btrfs_fs_devices *fs_devices;
2614 if (!device->writeable)
2617 lock_chunks(device->dev_root);
2618 old_total = btrfs_super_total_bytes(super_copy);
2619 diff = new_size - device->total_bytes;
2621 if (new_size <= device->total_bytes ||
2622 device->is_tgtdev_for_dev_replace) {
2623 unlock_chunks(device->dev_root);
2627 fs_devices = device->dev_root->fs_info->fs_devices;
2629 btrfs_set_super_total_bytes(super_copy, old_total + diff);
2630 device->fs_devices->total_rw_bytes += diff;
2632 btrfs_device_set_total_bytes(device, new_size);
2633 btrfs_device_set_disk_total_bytes(device, new_size);
2634 btrfs_clear_space_info_full(device->dev_root->fs_info);
2635 if (list_empty(&device->resized_list))
2636 list_add_tail(&device->resized_list,
2637 &fs_devices->resized_devices);
2638 unlock_chunks(device->dev_root);
2640 return btrfs_update_device(trans, device);
2643 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2644 struct btrfs_root *root, u64 chunk_objectid,
2648 struct btrfs_path *path;
2649 struct btrfs_key key;
2651 root = root->fs_info->chunk_root;
2652 path = btrfs_alloc_path();
2656 key.objectid = chunk_objectid;
2657 key.offset = chunk_offset;
2658 key.type = BTRFS_CHUNK_ITEM_KEY;
2660 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2663 else if (ret > 0) { /* Logic error or corruption */
2664 btrfs_std_error(root->fs_info, -ENOENT,
2665 "Failed lookup while freeing chunk.");
2670 ret = btrfs_del_item(trans, root, path);
2672 btrfs_std_error(root->fs_info, ret,
2673 "Failed to delete chunk item.");
2675 btrfs_free_path(path);
2679 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2682 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2683 struct btrfs_disk_key *disk_key;
2684 struct btrfs_chunk *chunk;
2691 struct btrfs_key key;
2694 array_size = btrfs_super_sys_array_size(super_copy);
2696 ptr = super_copy->sys_chunk_array;
2699 while (cur < array_size) {
2700 disk_key = (struct btrfs_disk_key *)ptr;
2701 btrfs_disk_key_to_cpu(&key, disk_key);
2703 len = sizeof(*disk_key);
2705 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2706 chunk = (struct btrfs_chunk *)(ptr + len);
2707 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2708 len += btrfs_chunk_item_size(num_stripes);
2713 if (key.objectid == chunk_objectid &&
2714 key.offset == chunk_offset) {
2715 memmove(ptr, ptr + len, array_size - (cur + len));
2717 btrfs_set_super_sys_array_size(super_copy, array_size);
2723 unlock_chunks(root);
2727 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2728 struct btrfs_root *root, u64 chunk_offset)
2730 struct extent_map_tree *em_tree;
2731 struct extent_map *em;
2732 struct btrfs_root *extent_root = root->fs_info->extent_root;
2733 struct map_lookup *map;
2734 u64 dev_extent_len = 0;
2735 u64 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2739 root = root->fs_info->chunk_root;
2740 em_tree = &root->fs_info->mapping_tree.map_tree;
2742 read_lock(&em_tree->lock);
2743 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2744 read_unlock(&em_tree->lock);
2746 if (!em || em->start > chunk_offset ||
2747 em->start + em->len < chunk_offset) {
2749 * This is a logic error, but we don't want to just rely on the
2750 * user having built with ASSERT enabled, so if ASSERT doens't
2751 * do anything we still error out.
2755 free_extent_map(em);
2758 map = (struct map_lookup *)em->bdev;
2759 lock_chunks(root->fs_info->chunk_root);
2760 check_system_chunk(trans, extent_root, map->type);
2761 unlock_chunks(root->fs_info->chunk_root);
2763 for (i = 0; i < map->num_stripes; i++) {
2764 struct btrfs_device *device = map->stripes[i].dev;
2765 ret = btrfs_free_dev_extent(trans, device,
2766 map->stripes[i].physical,
2769 btrfs_abort_transaction(trans, root, ret);
2773 if (device->bytes_used > 0) {
2775 btrfs_device_set_bytes_used(device,
2776 device->bytes_used - dev_extent_len);
2777 spin_lock(&root->fs_info->free_chunk_lock);
2778 root->fs_info->free_chunk_space += dev_extent_len;
2779 spin_unlock(&root->fs_info->free_chunk_lock);
2780 btrfs_clear_space_info_full(root->fs_info);
2781 unlock_chunks(root);
2784 if (map->stripes[i].dev) {
2785 ret = btrfs_update_device(trans, map->stripes[i].dev);
2787 btrfs_abort_transaction(trans, root, ret);
2792 ret = btrfs_free_chunk(trans, root, chunk_objectid, chunk_offset);
2794 btrfs_abort_transaction(trans, root, ret);
2798 trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2800 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2801 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2803 btrfs_abort_transaction(trans, root, ret);
2808 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset, em);
2810 btrfs_abort_transaction(trans, extent_root, ret);
2816 free_extent_map(em);
2820 static int btrfs_relocate_chunk(struct btrfs_root *root, u64 chunk_offset)
2822 struct btrfs_root *extent_root;
2823 struct btrfs_trans_handle *trans;
2826 root = root->fs_info->chunk_root;
2827 extent_root = root->fs_info->extent_root;
2830 * Prevent races with automatic removal of unused block groups.
2831 * After we relocate and before we remove the chunk with offset
2832 * chunk_offset, automatic removal of the block group can kick in,
2833 * resulting in a failure when calling btrfs_remove_chunk() below.
2835 * Make sure to acquire this mutex before doing a tree search (dev
2836 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2837 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2838 * we release the path used to search the chunk/dev tree and before
2839 * the current task acquires this mutex and calls us.
2841 ASSERT(mutex_is_locked(&root->fs_info->delete_unused_bgs_mutex));
2843 ret = btrfs_can_relocate(extent_root, chunk_offset);
2847 /* step one, relocate all the extents inside this chunk */
2848 btrfs_scrub_pause(root);
2849 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2850 btrfs_scrub_continue(root);
2854 trans = btrfs_start_trans_remove_block_group(root->fs_info,
2856 if (IS_ERR(trans)) {
2857 ret = PTR_ERR(trans);
2858 btrfs_std_error(root->fs_info, ret, NULL);
2863 * step two, delete the device extents and the
2864 * chunk tree entries
2866 ret = btrfs_remove_chunk(trans, root, chunk_offset);
2867 btrfs_end_transaction(trans, root);
2871 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2873 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2874 struct btrfs_path *path;
2875 struct extent_buffer *leaf;
2876 struct btrfs_chunk *chunk;
2877 struct btrfs_key key;
2878 struct btrfs_key found_key;
2880 bool retried = false;
2884 path = btrfs_alloc_path();
2889 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2890 key.offset = (u64)-1;
2891 key.type = BTRFS_CHUNK_ITEM_KEY;
2894 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
2895 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2897 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2900 BUG_ON(ret == 0); /* Corruption */
2902 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2905 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2911 leaf = path->nodes[0];
2912 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2914 chunk = btrfs_item_ptr(leaf, path->slots[0],
2915 struct btrfs_chunk);
2916 chunk_type = btrfs_chunk_type(leaf, chunk);
2917 btrfs_release_path(path);
2919 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2920 ret = btrfs_relocate_chunk(chunk_root,
2927 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2929 if (found_key.offset == 0)
2931 key.offset = found_key.offset - 1;
2934 if (failed && !retried) {
2938 } else if (WARN_ON(failed && retried)) {
2942 btrfs_free_path(path);
2946 static int insert_balance_item(struct btrfs_root *root,
2947 struct btrfs_balance_control *bctl)
2949 struct btrfs_trans_handle *trans;
2950 struct btrfs_balance_item *item;
2951 struct btrfs_disk_balance_args disk_bargs;
2952 struct btrfs_path *path;
2953 struct extent_buffer *leaf;
2954 struct btrfs_key key;
2957 path = btrfs_alloc_path();
2961 trans = btrfs_start_transaction(root, 0);
2962 if (IS_ERR(trans)) {
2963 btrfs_free_path(path);
2964 return PTR_ERR(trans);
2967 key.objectid = BTRFS_BALANCE_OBJECTID;
2968 key.type = BTRFS_BALANCE_ITEM_KEY;
2971 ret = btrfs_insert_empty_item(trans, root, path, &key,
2976 leaf = path->nodes[0];
2977 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2979 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2981 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2982 btrfs_set_balance_data(leaf, item, &disk_bargs);
2983 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2984 btrfs_set_balance_meta(leaf, item, &disk_bargs);
2985 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2986 btrfs_set_balance_sys(leaf, item, &disk_bargs);
2988 btrfs_set_balance_flags(leaf, item, bctl->flags);
2990 btrfs_mark_buffer_dirty(leaf);
2992 btrfs_free_path(path);
2993 err = btrfs_commit_transaction(trans, root);
2999 static int del_balance_item(struct btrfs_root *root)
3001 struct btrfs_trans_handle *trans;
3002 struct btrfs_path *path;
3003 struct btrfs_key key;
3006 path = btrfs_alloc_path();
3010 trans = btrfs_start_transaction(root, 0);
3011 if (IS_ERR(trans)) {
3012 btrfs_free_path(path);
3013 return PTR_ERR(trans);
3016 key.objectid = BTRFS_BALANCE_OBJECTID;
3017 key.type = BTRFS_BALANCE_ITEM_KEY;
3020 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3028 ret = btrfs_del_item(trans, root, path);
3030 btrfs_free_path(path);
3031 err = btrfs_commit_transaction(trans, root);
3038 * This is a heuristic used to reduce the number of chunks balanced on
3039 * resume after balance was interrupted.
3041 static void update_balance_args(struct btrfs_balance_control *bctl)
3044 * Turn on soft mode for chunk types that were being converted.
3046 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3047 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3048 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3049 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3050 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3051 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3054 * Turn on usage filter if is not already used. The idea is
3055 * that chunks that we have already balanced should be
3056 * reasonably full. Don't do it for chunks that are being
3057 * converted - that will keep us from relocating unconverted
3058 * (albeit full) chunks.
3060 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3061 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3062 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3063 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3064 bctl->data.usage = 90;
3066 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3067 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3068 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3069 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3070 bctl->sys.usage = 90;
3072 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3073 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3074 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3075 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3076 bctl->meta.usage = 90;
3081 * Should be called with both balance and volume mutexes held to
3082 * serialize other volume operations (add_dev/rm_dev/resize) with
3083 * restriper. Same goes for unset_balance_control.
3085 static void set_balance_control(struct btrfs_balance_control *bctl)
3087 struct btrfs_fs_info *fs_info = bctl->fs_info;
3089 BUG_ON(fs_info->balance_ctl);
3091 spin_lock(&fs_info->balance_lock);
3092 fs_info->balance_ctl = bctl;
3093 spin_unlock(&fs_info->balance_lock);
3096 static void unset_balance_control(struct btrfs_fs_info *fs_info)
3098 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3100 BUG_ON(!fs_info->balance_ctl);
3102 spin_lock(&fs_info->balance_lock);
3103 fs_info->balance_ctl = NULL;
3104 spin_unlock(&fs_info->balance_lock);
3110 * Balance filters. Return 1 if chunk should be filtered out
3111 * (should not be balanced).
3113 static int chunk_profiles_filter(u64 chunk_type,
3114 struct btrfs_balance_args *bargs)
3116 chunk_type = chunk_to_extended(chunk_type) &
3117 BTRFS_EXTENDED_PROFILE_MASK;
3119 if (bargs->profiles & chunk_type)
3125 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3126 struct btrfs_balance_args *bargs)
3128 struct btrfs_block_group_cache *cache;
3130 u64 user_thresh_min;
3131 u64 user_thresh_max;
3134 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3135 chunk_used = btrfs_block_group_used(&cache->item);
3137 if (bargs->usage_min == 0)
3138 user_thresh_min = 0;
3140 user_thresh_min = div_factor_fine(cache->key.offset,
3143 if (bargs->usage_max == 0)
3144 user_thresh_max = 1;
3145 else if (bargs->usage_max > 100)
3146 user_thresh_max = cache->key.offset;
3148 user_thresh_max = div_factor_fine(cache->key.offset,
3151 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3154 btrfs_put_block_group(cache);
3158 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3159 u64 chunk_offset, struct btrfs_balance_args *bargs)
3161 struct btrfs_block_group_cache *cache;
3162 u64 chunk_used, user_thresh;
3165 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3166 chunk_used = btrfs_block_group_used(&cache->item);
3168 if (bargs->usage_min == 0)
3170 else if (bargs->usage > 100)
3171 user_thresh = cache->key.offset;
3173 user_thresh = div_factor_fine(cache->key.offset,
3176 if (chunk_used < user_thresh)
3179 btrfs_put_block_group(cache);
3183 static int chunk_devid_filter(struct extent_buffer *leaf,
3184 struct btrfs_chunk *chunk,
3185 struct btrfs_balance_args *bargs)
3187 struct btrfs_stripe *stripe;
3188 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3191 for (i = 0; i < num_stripes; i++) {
3192 stripe = btrfs_stripe_nr(chunk, i);
3193 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3200 /* [pstart, pend) */
3201 static int chunk_drange_filter(struct extent_buffer *leaf,
3202 struct btrfs_chunk *chunk,
3204 struct btrfs_balance_args *bargs)
3206 struct btrfs_stripe *stripe;
3207 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3213 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3216 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3217 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3218 factor = num_stripes / 2;
3219 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3220 factor = num_stripes - 1;
3221 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3222 factor = num_stripes - 2;
3224 factor = num_stripes;
3227 for (i = 0; i < num_stripes; i++) {
3228 stripe = btrfs_stripe_nr(chunk, i);
3229 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3232 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3233 stripe_length = btrfs_chunk_length(leaf, chunk);
3234 stripe_length = div_u64(stripe_length, factor);
3236 if (stripe_offset < bargs->pend &&
3237 stripe_offset + stripe_length > bargs->pstart)
3244 /* [vstart, vend) */
3245 static int chunk_vrange_filter(struct extent_buffer *leaf,
3246 struct btrfs_chunk *chunk,
3248 struct btrfs_balance_args *bargs)
3250 if (chunk_offset < bargs->vend &&
3251 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3252 /* at least part of the chunk is inside this vrange */
3258 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3259 struct btrfs_chunk *chunk,
3260 struct btrfs_balance_args *bargs)
3262 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3264 if (bargs->stripes_min <= num_stripes
3265 && num_stripes <= bargs->stripes_max)
3271 static int chunk_soft_convert_filter(u64 chunk_type,
3272 struct btrfs_balance_args *bargs)
3274 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3277 chunk_type = chunk_to_extended(chunk_type) &
3278 BTRFS_EXTENDED_PROFILE_MASK;
3280 if (bargs->target == chunk_type)
3286 static int should_balance_chunk(struct btrfs_root *root,
3287 struct extent_buffer *leaf,
3288 struct btrfs_chunk *chunk, u64 chunk_offset)
3290 struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
3291 struct btrfs_balance_args *bargs = NULL;
3292 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3295 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3296 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3300 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3301 bargs = &bctl->data;
3302 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3304 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3305 bargs = &bctl->meta;
3307 /* profiles filter */
3308 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3309 chunk_profiles_filter(chunk_type, bargs)) {
3314 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3315 chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
3317 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3318 chunk_usage_range_filter(bctl->fs_info, chunk_offset, bargs)) {
3323 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3324 chunk_devid_filter(leaf, chunk, bargs)) {
3328 /* drange filter, makes sense only with devid filter */
3329 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3330 chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
3335 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3336 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3340 /* stripes filter */
3341 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3342 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3346 /* soft profile changing mode */
3347 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3348 chunk_soft_convert_filter(chunk_type, bargs)) {
3353 * limited by count, must be the last filter
3355 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3356 if (bargs->limit == 0)
3360 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3362 * Same logic as the 'limit' filter; the minimum cannot be
3363 * determined here because we do not have the global informatoin
3364 * about the count of all chunks that satisfy the filters.
3366 if (bargs->limit_max == 0)
3375 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3377 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3378 struct btrfs_root *chunk_root = fs_info->chunk_root;
3379 struct btrfs_root *dev_root = fs_info->dev_root;
3380 struct list_head *devices;
3381 struct btrfs_device *device;
3385 struct btrfs_chunk *chunk;
3386 struct btrfs_path *path;
3387 struct btrfs_key key;
3388 struct btrfs_key found_key;
3389 struct btrfs_trans_handle *trans;
3390 struct extent_buffer *leaf;
3393 int enospc_errors = 0;
3394 bool counting = true;
3395 /* The single value limit and min/max limits use the same bytes in the */
3396 u64 limit_data = bctl->data.limit;
3397 u64 limit_meta = bctl->meta.limit;
3398 u64 limit_sys = bctl->sys.limit;
3402 int chunk_reserved = 0;
3404 /* step one make some room on all the devices */
3405 devices = &fs_info->fs_devices->devices;
3406 list_for_each_entry(device, devices, dev_list) {
3407 old_size = btrfs_device_get_total_bytes(device);
3408 size_to_free = div_factor(old_size, 1);
3409 size_to_free = min_t(u64, size_to_free, SZ_1M);
3410 if (!device->writeable ||
3411 btrfs_device_get_total_bytes(device) -
3412 btrfs_device_get_bytes_used(device) > size_to_free ||
3413 device->is_tgtdev_for_dev_replace)
3416 ret = btrfs_shrink_device(device, old_size - size_to_free);
3421 trans = btrfs_start_transaction(dev_root, 0);
3422 BUG_ON(IS_ERR(trans));
3424 ret = btrfs_grow_device(trans, device, old_size);
3427 btrfs_end_transaction(trans, dev_root);
3430 /* step two, relocate all the chunks */
3431 path = btrfs_alloc_path();
3437 /* zero out stat counters */
3438 spin_lock(&fs_info->balance_lock);
3439 memset(&bctl->stat, 0, sizeof(bctl->stat));
3440 spin_unlock(&fs_info->balance_lock);
3444 * The single value limit and min/max limits use the same bytes
3447 bctl->data.limit = limit_data;
3448 bctl->meta.limit = limit_meta;
3449 bctl->sys.limit = limit_sys;
3451 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3452 key.offset = (u64)-1;
3453 key.type = BTRFS_CHUNK_ITEM_KEY;
3456 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3457 atomic_read(&fs_info->balance_cancel_req)) {
3462 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3463 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3465 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3470 * this shouldn't happen, it means the last relocate
3474 BUG(); /* FIXME break ? */
3476 ret = btrfs_previous_item(chunk_root, path, 0,
3477 BTRFS_CHUNK_ITEM_KEY);
3479 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3484 leaf = path->nodes[0];
3485 slot = path->slots[0];
3486 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3488 if (found_key.objectid != key.objectid) {
3489 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3493 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3494 chunk_type = btrfs_chunk_type(leaf, chunk);
3497 spin_lock(&fs_info->balance_lock);
3498 bctl->stat.considered++;
3499 spin_unlock(&fs_info->balance_lock);
3502 ret = should_balance_chunk(chunk_root, leaf, chunk,
3505 btrfs_release_path(path);
3507 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3512 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3513 spin_lock(&fs_info->balance_lock);
3514 bctl->stat.expected++;
3515 spin_unlock(&fs_info->balance_lock);
3517 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3519 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3521 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3528 * Apply limit_min filter, no need to check if the LIMITS
3529 * filter is used, limit_min is 0 by default
3531 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3532 count_data < bctl->data.limit_min)
3533 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3534 count_meta < bctl->meta.limit_min)
3535 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3536 count_sys < bctl->sys.limit_min)) {
3537 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3541 if ((chunk_type & BTRFS_BLOCK_GROUP_DATA) && !chunk_reserved) {
3542 trans = btrfs_start_transaction(chunk_root, 0);
3543 if (IS_ERR(trans)) {
3544 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3545 ret = PTR_ERR(trans);
3549 ret = btrfs_force_chunk_alloc(trans, chunk_root,
3550 BTRFS_BLOCK_GROUP_DATA);
3551 btrfs_end_transaction(trans, chunk_root);
3553 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3559 ret = btrfs_relocate_chunk(chunk_root,
3561 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3562 if (ret && ret != -ENOSPC)
3564 if (ret == -ENOSPC) {
3567 spin_lock(&fs_info->balance_lock);
3568 bctl->stat.completed++;
3569 spin_unlock(&fs_info->balance_lock);
3572 if (found_key.offset == 0)
3574 key.offset = found_key.offset - 1;
3578 btrfs_release_path(path);
3583 btrfs_free_path(path);
3584 if (enospc_errors) {
3585 btrfs_info(fs_info, "%d enospc errors during balance",
3595 * alloc_profile_is_valid - see if a given profile is valid and reduced
3596 * @flags: profile to validate
3597 * @extended: if true @flags is treated as an extended profile
3599 static int alloc_profile_is_valid(u64 flags, int extended)
3601 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3602 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3604 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3606 /* 1) check that all other bits are zeroed */
3610 /* 2) see if profile is reduced */
3612 return !extended; /* "0" is valid for usual profiles */
3614 /* true if exactly one bit set */
3615 return (flags & (flags - 1)) == 0;
3618 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3620 /* cancel requested || normal exit path */
3621 return atomic_read(&fs_info->balance_cancel_req) ||
3622 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3623 atomic_read(&fs_info->balance_cancel_req) == 0);
3626 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3630 unset_balance_control(fs_info);
3631 ret = del_balance_item(fs_info->tree_root);
3633 btrfs_std_error(fs_info, ret, NULL);
3635 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3638 /* Non-zero return value signifies invalidity */
3639 static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg,
3642 return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3643 (!alloc_profile_is_valid(bctl_arg->target, 1) ||
3644 (bctl_arg->target & ~allowed)));
3648 * Should be called with both balance and volume mutexes held
3650 int btrfs_balance(struct btrfs_balance_control *bctl,
3651 struct btrfs_ioctl_balance_args *bargs)
3653 struct btrfs_fs_info *fs_info = bctl->fs_info;
3660 if (btrfs_fs_closing(fs_info) ||
3661 atomic_read(&fs_info->balance_pause_req) ||
3662 atomic_read(&fs_info->balance_cancel_req)) {
3667 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3668 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3672 * In case of mixed groups both data and meta should be picked,
3673 * and identical options should be given for both of them.
3675 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3676 if (mixed && (bctl->flags & allowed)) {
3677 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3678 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3679 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3680 btrfs_err(fs_info, "with mixed groups data and "
3681 "metadata balance options must be the same");
3687 num_devices = fs_info->fs_devices->num_devices;
3688 btrfs_dev_replace_lock(&fs_info->dev_replace);
3689 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3690 BUG_ON(num_devices < 1);
3693 btrfs_dev_replace_unlock(&fs_info->dev_replace);
3694 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3695 if (num_devices == 1)
3696 allowed |= BTRFS_BLOCK_GROUP_DUP;
3697 else if (num_devices > 1)
3698 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3699 if (num_devices > 2)
3700 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3701 if (num_devices > 3)
3702 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3703 BTRFS_BLOCK_GROUP_RAID6);
3704 if (validate_convert_profile(&bctl->data, allowed)) {
3705 btrfs_err(fs_info, "unable to start balance with target "
3706 "data profile %llu",
3711 if (validate_convert_profile(&bctl->meta, allowed)) {
3713 "unable to start balance with target metadata profile %llu",
3718 if (validate_convert_profile(&bctl->sys, allowed)) {
3720 "unable to start balance with target system profile %llu",
3726 /* allow to reduce meta or sys integrity only if force set */
3727 allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3728 BTRFS_BLOCK_GROUP_RAID10 |
3729 BTRFS_BLOCK_GROUP_RAID5 |
3730 BTRFS_BLOCK_GROUP_RAID6;
3732 seq = read_seqbegin(&fs_info->profiles_lock);
3734 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3735 (fs_info->avail_system_alloc_bits & allowed) &&
3736 !(bctl->sys.target & allowed)) ||
3737 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3738 (fs_info->avail_metadata_alloc_bits & allowed) &&
3739 !(bctl->meta.target & allowed))) {
3740 if (bctl->flags & BTRFS_BALANCE_FORCE) {
3741 btrfs_info(fs_info, "force reducing metadata integrity");
3743 btrfs_err(fs_info, "balance will reduce metadata "
3744 "integrity, use force if you want this");
3749 } while (read_seqretry(&fs_info->profiles_lock, seq));
3751 if (btrfs_get_num_tolerated_disk_barrier_failures(bctl->meta.target) <
3752 btrfs_get_num_tolerated_disk_barrier_failures(bctl->data.target)) {
3754 "metatdata profile 0x%llx has lower redundancy than data profile 0x%llx",
3755 bctl->meta.target, bctl->data.target);
3758 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3759 fs_info->num_tolerated_disk_barrier_failures = min(
3760 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info),
3761 btrfs_get_num_tolerated_disk_barrier_failures(
3765 ret = insert_balance_item(fs_info->tree_root, bctl);
3766 if (ret && ret != -EEXIST)
3769 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3770 BUG_ON(ret == -EEXIST);
3771 set_balance_control(bctl);
3773 BUG_ON(ret != -EEXIST);
3774 spin_lock(&fs_info->balance_lock);
3775 update_balance_args(bctl);
3776 spin_unlock(&fs_info->balance_lock);
3779 atomic_inc(&fs_info->balance_running);
3780 mutex_unlock(&fs_info->balance_mutex);
3782 ret = __btrfs_balance(fs_info);
3784 mutex_lock(&fs_info->balance_mutex);
3785 atomic_dec(&fs_info->balance_running);
3787 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3788 fs_info->num_tolerated_disk_barrier_failures =
3789 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3793 memset(bargs, 0, sizeof(*bargs));
3794 update_ioctl_balance_args(fs_info, 0, bargs);
3797 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3798 balance_need_close(fs_info)) {
3799 __cancel_balance(fs_info);
3802 wake_up(&fs_info->balance_wait_q);
3806 if (bctl->flags & BTRFS_BALANCE_RESUME)
3807 __cancel_balance(fs_info);
3810 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3815 static int balance_kthread(void *data)
3817 struct btrfs_fs_info *fs_info = data;
3820 mutex_lock(&fs_info->volume_mutex);
3821 mutex_lock(&fs_info->balance_mutex);
3823 if (fs_info->balance_ctl) {
3824 btrfs_info(fs_info, "continuing balance");
3825 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3828 mutex_unlock(&fs_info->balance_mutex);
3829 mutex_unlock(&fs_info->volume_mutex);
3834 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3836 struct task_struct *tsk;
3838 spin_lock(&fs_info->balance_lock);
3839 if (!fs_info->balance_ctl) {
3840 spin_unlock(&fs_info->balance_lock);
3843 spin_unlock(&fs_info->balance_lock);
3845 if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
3846 btrfs_info(fs_info, "force skipping balance");
3850 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3851 return PTR_ERR_OR_ZERO(tsk);
3854 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3856 struct btrfs_balance_control *bctl;
3857 struct btrfs_balance_item *item;
3858 struct btrfs_disk_balance_args disk_bargs;
3859 struct btrfs_path *path;
3860 struct extent_buffer *leaf;
3861 struct btrfs_key key;
3864 path = btrfs_alloc_path();
3868 key.objectid = BTRFS_BALANCE_OBJECTID;
3869 key.type = BTRFS_BALANCE_ITEM_KEY;
3872 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3875 if (ret > 0) { /* ret = -ENOENT; */
3880 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3886 leaf = path->nodes[0];
3887 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3889 bctl->fs_info = fs_info;
3890 bctl->flags = btrfs_balance_flags(leaf, item);
3891 bctl->flags |= BTRFS_BALANCE_RESUME;
3893 btrfs_balance_data(leaf, item, &disk_bargs);
3894 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3895 btrfs_balance_meta(leaf, item, &disk_bargs);
3896 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3897 btrfs_balance_sys(leaf, item, &disk_bargs);
3898 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3900 WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
3902 mutex_lock(&fs_info->volume_mutex);
3903 mutex_lock(&fs_info->balance_mutex);
3905 set_balance_control(bctl);
3907 mutex_unlock(&fs_info->balance_mutex);
3908 mutex_unlock(&fs_info->volume_mutex);
3910 btrfs_free_path(path);
3914 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3918 mutex_lock(&fs_info->balance_mutex);
3919 if (!fs_info->balance_ctl) {
3920 mutex_unlock(&fs_info->balance_mutex);
3924 if (atomic_read(&fs_info->balance_running)) {
3925 atomic_inc(&fs_info->balance_pause_req);
3926 mutex_unlock(&fs_info->balance_mutex);
3928 wait_event(fs_info->balance_wait_q,
3929 atomic_read(&fs_info->balance_running) == 0);
3931 mutex_lock(&fs_info->balance_mutex);
3932 /* we are good with balance_ctl ripped off from under us */
3933 BUG_ON(atomic_read(&fs_info->balance_running));
3934 atomic_dec(&fs_info->balance_pause_req);
3939 mutex_unlock(&fs_info->balance_mutex);
3943 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3945 if (fs_info->sb->s_flags & MS_RDONLY)
3948 mutex_lock(&fs_info->balance_mutex);
3949 if (!fs_info->balance_ctl) {
3950 mutex_unlock(&fs_info->balance_mutex);
3954 atomic_inc(&fs_info->balance_cancel_req);
3956 * if we are running just wait and return, balance item is
3957 * deleted in btrfs_balance in this case
3959 if (atomic_read(&fs_info->balance_running)) {
3960 mutex_unlock(&fs_info->balance_mutex);
3961 wait_event(fs_info->balance_wait_q,
3962 atomic_read(&fs_info->balance_running) == 0);
3963 mutex_lock(&fs_info->balance_mutex);
3965 /* __cancel_balance needs volume_mutex */
3966 mutex_unlock(&fs_info->balance_mutex);
3967 mutex_lock(&fs_info->volume_mutex);
3968 mutex_lock(&fs_info->balance_mutex);
3970 if (fs_info->balance_ctl)
3971 __cancel_balance(fs_info);
3973 mutex_unlock(&fs_info->volume_mutex);
3976 BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
3977 atomic_dec(&fs_info->balance_cancel_req);
3978 mutex_unlock(&fs_info->balance_mutex);
3982 static int btrfs_uuid_scan_kthread(void *data)
3984 struct btrfs_fs_info *fs_info = data;
3985 struct btrfs_root *root = fs_info->tree_root;
3986 struct btrfs_key key;
3987 struct btrfs_key max_key;
3988 struct btrfs_path *path = NULL;
3990 struct extent_buffer *eb;
3992 struct btrfs_root_item root_item;
3994 struct btrfs_trans_handle *trans = NULL;
3996 path = btrfs_alloc_path();
4003 key.type = BTRFS_ROOT_ITEM_KEY;
4006 max_key.objectid = (u64)-1;
4007 max_key.type = BTRFS_ROOT_ITEM_KEY;
4008 max_key.offset = (u64)-1;
4011 ret = btrfs_search_forward(root, &key, path, 0);
4018 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4019 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4020 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4021 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4024 eb = path->nodes[0];
4025 slot = path->slots[0];
4026 item_size = btrfs_item_size_nr(eb, slot);
4027 if (item_size < sizeof(root_item))
4030 read_extent_buffer(eb, &root_item,
4031 btrfs_item_ptr_offset(eb, slot),
4032 (int)sizeof(root_item));
4033 if (btrfs_root_refs(&root_item) == 0)
4036 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4037 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4041 btrfs_release_path(path);
4043 * 1 - subvol uuid item
4044 * 1 - received_subvol uuid item
4046 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4047 if (IS_ERR(trans)) {
4048 ret = PTR_ERR(trans);
4056 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4057 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
4059 BTRFS_UUID_KEY_SUBVOL,
4062 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4068 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4069 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
4070 root_item.received_uuid,
4071 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4074 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4082 ret = btrfs_end_transaction(trans, fs_info->uuid_root);
4088 btrfs_release_path(path);
4089 if (key.offset < (u64)-1) {
4091 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4093 key.type = BTRFS_ROOT_ITEM_KEY;
4094 } else if (key.objectid < (u64)-1) {
4096 key.type = BTRFS_ROOT_ITEM_KEY;
4105 btrfs_free_path(path);
4106 if (trans && !IS_ERR(trans))
4107 btrfs_end_transaction(trans, fs_info->uuid_root);
4109 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4111 fs_info->update_uuid_tree_gen = 1;
4112 up(&fs_info->uuid_tree_rescan_sem);
4117 * Callback for btrfs_uuid_tree_iterate().
4119 * 0 check succeeded, the entry is not outdated.
4120 * < 0 if an error occured.
4121 * > 0 if the check failed, which means the caller shall remove the entry.
4123 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
4124 u8 *uuid, u8 type, u64 subid)
4126 struct btrfs_key key;
4128 struct btrfs_root *subvol_root;
4130 if (type != BTRFS_UUID_KEY_SUBVOL &&
4131 type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
4134 key.objectid = subid;
4135 key.type = BTRFS_ROOT_ITEM_KEY;
4136 key.offset = (u64)-1;
4137 subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
4138 if (IS_ERR(subvol_root)) {
4139 ret = PTR_ERR(subvol_root);
4146 case BTRFS_UUID_KEY_SUBVOL:
4147 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
4150 case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
4151 if (memcmp(uuid, subvol_root->root_item.received_uuid,
4161 static int btrfs_uuid_rescan_kthread(void *data)
4163 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4167 * 1st step is to iterate through the existing UUID tree and
4168 * to delete all entries that contain outdated data.
4169 * 2nd step is to add all missing entries to the UUID tree.
4171 ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4173 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4174 up(&fs_info->uuid_tree_rescan_sem);
4177 return btrfs_uuid_scan_kthread(data);
4180 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4182 struct btrfs_trans_handle *trans;
4183 struct btrfs_root *tree_root = fs_info->tree_root;
4184 struct btrfs_root *uuid_root;
4185 struct task_struct *task;
4192 trans = btrfs_start_transaction(tree_root, 2);
4194 return PTR_ERR(trans);
4196 uuid_root = btrfs_create_tree(trans, fs_info,
4197 BTRFS_UUID_TREE_OBJECTID);
4198 if (IS_ERR(uuid_root)) {
4199 ret = PTR_ERR(uuid_root);
4200 btrfs_abort_transaction(trans, tree_root, ret);
4204 fs_info->uuid_root = uuid_root;
4206 ret = btrfs_commit_transaction(trans, tree_root);
4210 down(&fs_info->uuid_tree_rescan_sem);
4211 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4213 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4214 btrfs_warn(fs_info, "failed to start uuid_scan task");
4215 up(&fs_info->uuid_tree_rescan_sem);
4216 return PTR_ERR(task);
4222 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4224 struct task_struct *task;
4226 down(&fs_info->uuid_tree_rescan_sem);
4227 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4229 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4230 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4231 up(&fs_info->uuid_tree_rescan_sem);
4232 return PTR_ERR(task);
4239 * shrinking a device means finding all of the device extents past
4240 * the new size, and then following the back refs to the chunks.
4241 * The chunk relocation code actually frees the device extent
4243 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4245 struct btrfs_trans_handle *trans;
4246 struct btrfs_root *root = device->dev_root;
4247 struct btrfs_dev_extent *dev_extent = NULL;
4248 struct btrfs_path *path;
4254 bool retried = false;
4255 bool checked_pending_chunks = false;
4256 struct extent_buffer *l;
4257 struct btrfs_key key;
4258 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4259 u64 old_total = btrfs_super_total_bytes(super_copy);
4260 u64 old_size = btrfs_device_get_total_bytes(device);
4261 u64 diff = old_size - new_size;
4263 if (device->is_tgtdev_for_dev_replace)
4266 path = btrfs_alloc_path();
4270 path->reada = READA_FORWARD;
4274 btrfs_device_set_total_bytes(device, new_size);
4275 if (device->writeable) {
4276 device->fs_devices->total_rw_bytes -= diff;
4277 spin_lock(&root->fs_info->free_chunk_lock);
4278 root->fs_info->free_chunk_space -= diff;
4279 spin_unlock(&root->fs_info->free_chunk_lock);
4281 unlock_chunks(root);
4284 key.objectid = device->devid;
4285 key.offset = (u64)-1;
4286 key.type = BTRFS_DEV_EXTENT_KEY;
4289 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
4290 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4292 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4296 ret = btrfs_previous_item(root, path, 0, key.type);
4298 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4303 btrfs_release_path(path);
4308 slot = path->slots[0];
4309 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4311 if (key.objectid != device->devid) {
4312 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4313 btrfs_release_path(path);
4317 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4318 length = btrfs_dev_extent_length(l, dev_extent);
4320 if (key.offset + length <= new_size) {
4321 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4322 btrfs_release_path(path);
4326 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4327 btrfs_release_path(path);
4329 ret = btrfs_relocate_chunk(root, chunk_offset);
4330 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4331 if (ret && ret != -ENOSPC)
4335 } while (key.offset-- > 0);
4337 if (failed && !retried) {
4341 } else if (failed && retried) {
4346 /* Shrinking succeeded, else we would be at "done". */
4347 trans = btrfs_start_transaction(root, 0);
4348 if (IS_ERR(trans)) {
4349 ret = PTR_ERR(trans);
4356 * We checked in the above loop all device extents that were already in
4357 * the device tree. However before we have updated the device's
4358 * total_bytes to the new size, we might have had chunk allocations that
4359 * have not complete yet (new block groups attached to transaction
4360 * handles), and therefore their device extents were not yet in the
4361 * device tree and we missed them in the loop above. So if we have any
4362 * pending chunk using a device extent that overlaps the device range
4363 * that we can not use anymore, commit the current transaction and
4364 * repeat the search on the device tree - this way we guarantee we will
4365 * not have chunks using device extents that end beyond 'new_size'.
4367 if (!checked_pending_chunks) {
4368 u64 start = new_size;
4369 u64 len = old_size - new_size;
4371 if (contains_pending_extent(trans->transaction, device,
4373 unlock_chunks(root);
4374 checked_pending_chunks = true;
4377 ret = btrfs_commit_transaction(trans, root);
4384 btrfs_device_set_disk_total_bytes(device, new_size);
4385 if (list_empty(&device->resized_list))
4386 list_add_tail(&device->resized_list,
4387 &root->fs_info->fs_devices->resized_devices);
4389 WARN_ON(diff > old_total);
4390 btrfs_set_super_total_bytes(super_copy, old_total - diff);
4391 unlock_chunks(root);
4393 /* Now btrfs_update_device() will change the on-disk size. */
4394 ret = btrfs_update_device(trans, device);
4395 btrfs_end_transaction(trans, root);
4397 btrfs_free_path(path);
4400 btrfs_device_set_total_bytes(device, old_size);
4401 if (device->writeable)
4402 device->fs_devices->total_rw_bytes += diff;
4403 spin_lock(&root->fs_info->free_chunk_lock);
4404 root->fs_info->free_chunk_space += diff;
4405 spin_unlock(&root->fs_info->free_chunk_lock);
4406 unlock_chunks(root);
4411 static int btrfs_add_system_chunk(struct btrfs_root *root,
4412 struct btrfs_key *key,
4413 struct btrfs_chunk *chunk, int item_size)
4415 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4416 struct btrfs_disk_key disk_key;
4421 array_size = btrfs_super_sys_array_size(super_copy);
4422 if (array_size + item_size + sizeof(disk_key)
4423 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4424 unlock_chunks(root);
4428 ptr = super_copy->sys_chunk_array + array_size;
4429 btrfs_cpu_key_to_disk(&disk_key, key);
4430 memcpy(ptr, &disk_key, sizeof(disk_key));
4431 ptr += sizeof(disk_key);
4432 memcpy(ptr, chunk, item_size);
4433 item_size += sizeof(disk_key);
4434 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4435 unlock_chunks(root);
4441 * sort the devices in descending order by max_avail, total_avail
4443 static int btrfs_cmp_device_info(const void *a, const void *b)
4445 const struct btrfs_device_info *di_a = a;
4446 const struct btrfs_device_info *di_b = b;
4448 if (di_a->max_avail > di_b->max_avail)
4450 if (di_a->max_avail < di_b->max_avail)
4452 if (di_a->total_avail > di_b->total_avail)
4454 if (di_a->total_avail < di_b->total_avail)
4459 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
4461 /* TODO allow them to set a preferred stripe size */
4465 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4467 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4470 btrfs_set_fs_incompat(info, RAID56);
4473 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r) \
4474 - sizeof(struct btrfs_item) \
4475 - sizeof(struct btrfs_chunk)) \
4476 / sizeof(struct btrfs_stripe) + 1)
4478 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
4479 - 2 * sizeof(struct btrfs_disk_key) \
4480 - 2 * sizeof(struct btrfs_chunk)) \
4481 / sizeof(struct btrfs_stripe) + 1)
4483 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4484 struct btrfs_root *extent_root, u64 start,
4487 struct btrfs_fs_info *info = extent_root->fs_info;
4488 struct btrfs_fs_devices *fs_devices = info->fs_devices;
4489 struct list_head *cur;
4490 struct map_lookup *map = NULL;
4491 struct extent_map_tree *em_tree;
4492 struct extent_map *em;
4493 struct btrfs_device_info *devices_info = NULL;
4495 int num_stripes; /* total number of stripes to allocate */
4496 int data_stripes; /* number of stripes that count for
4498 int sub_stripes; /* sub_stripes info for map */
4499 int dev_stripes; /* stripes per dev */
4500 int devs_max; /* max devs to use */
4501 int devs_min; /* min devs needed */
4502 int devs_increment; /* ndevs has to be a multiple of this */
4503 int ncopies; /* how many copies to data has */
4505 u64 max_stripe_size;
4509 u64 raid_stripe_len = BTRFS_STRIPE_LEN;
4515 BUG_ON(!alloc_profile_is_valid(type, 0));
4517 if (list_empty(&fs_devices->alloc_list))
4520 index = __get_raid_index(type);
4522 sub_stripes = btrfs_raid_array[index].sub_stripes;
4523 dev_stripes = btrfs_raid_array[index].dev_stripes;
4524 devs_max = btrfs_raid_array[index].devs_max;
4525 devs_min = btrfs_raid_array[index].devs_min;
4526 devs_increment = btrfs_raid_array[index].devs_increment;
4527 ncopies = btrfs_raid_array[index].ncopies;
4529 if (type & BTRFS_BLOCK_GROUP_DATA) {
4530 max_stripe_size = SZ_1G;
4531 max_chunk_size = 10 * max_stripe_size;
4533 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4534 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4535 /* for larger filesystems, use larger metadata chunks */
4536 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
4537 max_stripe_size = SZ_1G;
4539 max_stripe_size = SZ_256M;
4540 max_chunk_size = max_stripe_size;
4542 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4543 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4544 max_stripe_size = SZ_32M;
4545 max_chunk_size = 2 * max_stripe_size;
4547 devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4549 btrfs_err(info, "invalid chunk type 0x%llx requested",
4554 /* we don't want a chunk larger than 10% of writeable space */
4555 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4558 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4563 cur = fs_devices->alloc_list.next;
4566 * in the first pass through the devices list, we gather information
4567 * about the available holes on each device.
4570 while (cur != &fs_devices->alloc_list) {
4571 struct btrfs_device *device;
4575 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4579 if (!device->writeable) {
4581 "BTRFS: read-only device in alloc_list\n");
4585 if (!device->in_fs_metadata ||
4586 device->is_tgtdev_for_dev_replace)
4589 if (device->total_bytes > device->bytes_used)
4590 total_avail = device->total_bytes - device->bytes_used;
4594 /* If there is no space on this device, skip it. */
4595 if (total_avail == 0)
4598 ret = find_free_dev_extent(trans, device,
4599 max_stripe_size * dev_stripes,
4600 &dev_offset, &max_avail);
4601 if (ret && ret != -ENOSPC)
4605 max_avail = max_stripe_size * dev_stripes;
4607 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4610 if (ndevs == fs_devices->rw_devices) {
4611 WARN(1, "%s: found more than %llu devices\n",
4612 __func__, fs_devices->rw_devices);
4615 devices_info[ndevs].dev_offset = dev_offset;
4616 devices_info[ndevs].max_avail = max_avail;
4617 devices_info[ndevs].total_avail = total_avail;
4618 devices_info[ndevs].dev = device;
4623 * now sort the devices by hole size / available space
4625 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4626 btrfs_cmp_device_info, NULL);
4628 /* round down to number of usable stripes */
4629 ndevs -= ndevs % devs_increment;
4631 if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4636 if (devs_max && ndevs > devs_max)
4639 * the primary goal is to maximize the number of stripes, so use as many
4640 * devices as possible, even if the stripes are not maximum sized.
4642 stripe_size = devices_info[ndevs-1].max_avail;
4643 num_stripes = ndevs * dev_stripes;
4646 * this will have to be fixed for RAID1 and RAID10 over
4649 data_stripes = num_stripes / ncopies;
4651 if (type & BTRFS_BLOCK_GROUP_RAID5) {
4652 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
4653 btrfs_super_stripesize(info->super_copy));
4654 data_stripes = num_stripes - 1;
4656 if (type & BTRFS_BLOCK_GROUP_RAID6) {
4657 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
4658 btrfs_super_stripesize(info->super_copy));
4659 data_stripes = num_stripes - 2;
4663 * Use the number of data stripes to figure out how big this chunk
4664 * is really going to be in terms of logical address space,
4665 * and compare that answer with the max chunk size
4667 if (stripe_size * data_stripes > max_chunk_size) {
4668 u64 mask = (1ULL << 24) - 1;
4670 stripe_size = div_u64(max_chunk_size, data_stripes);
4672 /* bump the answer up to a 16MB boundary */
4673 stripe_size = (stripe_size + mask) & ~mask;
4675 /* but don't go higher than the limits we found
4676 * while searching for free extents
4678 if (stripe_size > devices_info[ndevs-1].max_avail)
4679 stripe_size = devices_info[ndevs-1].max_avail;
4682 stripe_size = div_u64(stripe_size, dev_stripes);
4684 /* align to BTRFS_STRIPE_LEN */
4685 stripe_size = div_u64(stripe_size, raid_stripe_len);
4686 stripe_size *= raid_stripe_len;
4688 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4693 map->num_stripes = num_stripes;
4695 for (i = 0; i < ndevs; ++i) {
4696 for (j = 0; j < dev_stripes; ++j) {
4697 int s = i * dev_stripes + j;
4698 map->stripes[s].dev = devices_info[i].dev;
4699 map->stripes[s].physical = devices_info[i].dev_offset +
4703 map->sector_size = extent_root->sectorsize;
4704 map->stripe_len = raid_stripe_len;
4705 map->io_align = raid_stripe_len;
4706 map->io_width = raid_stripe_len;
4708 map->sub_stripes = sub_stripes;
4710 num_bytes = stripe_size * data_stripes;
4712 trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
4714 em = alloc_extent_map();
4720 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4721 em->bdev = (struct block_device *)map;
4723 em->len = num_bytes;
4724 em->block_start = 0;
4725 em->block_len = em->len;
4726 em->orig_block_len = stripe_size;
4728 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4729 write_lock(&em_tree->lock);
4730 ret = add_extent_mapping(em_tree, em, 0);
4732 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4733 atomic_inc(&em->refs);
4735 write_unlock(&em_tree->lock);
4737 free_extent_map(em);
4741 ret = btrfs_make_block_group(trans, extent_root, 0, type,
4742 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4745 goto error_del_extent;
4747 for (i = 0; i < map->num_stripes; i++) {
4748 num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4749 btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4752 spin_lock(&extent_root->fs_info->free_chunk_lock);
4753 extent_root->fs_info->free_chunk_space -= (stripe_size *
4755 spin_unlock(&extent_root->fs_info->free_chunk_lock);
4757 free_extent_map(em);
4758 check_raid56_incompat_flag(extent_root->fs_info, type);
4760 kfree(devices_info);
4764 write_lock(&em_tree->lock);
4765 remove_extent_mapping(em_tree, em);
4766 write_unlock(&em_tree->lock);
4768 /* One for our allocation */
4769 free_extent_map(em);
4770 /* One for the tree reference */
4771 free_extent_map(em);
4772 /* One for the pending_chunks list reference */
4773 free_extent_map(em);
4775 kfree(devices_info);
4779 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4780 struct btrfs_root *extent_root,
4781 u64 chunk_offset, u64 chunk_size)
4783 struct btrfs_key key;
4784 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4785 struct btrfs_device *device;
4786 struct btrfs_chunk *chunk;
4787 struct btrfs_stripe *stripe;
4788 struct extent_map_tree *em_tree;
4789 struct extent_map *em;
4790 struct map_lookup *map;
4797 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4798 read_lock(&em_tree->lock);
4799 em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size);
4800 read_unlock(&em_tree->lock);
4803 btrfs_crit(extent_root->fs_info, "unable to find logical "
4804 "%Lu len %Lu", chunk_offset, chunk_size);
4808 if (em->start != chunk_offset || em->len != chunk_size) {
4809 btrfs_crit(extent_root->fs_info, "found a bad mapping, wanted"
4810 " %Lu-%Lu, found %Lu-%Lu", chunk_offset,
4811 chunk_size, em->start, em->len);
4812 free_extent_map(em);
4816 map = (struct map_lookup *)em->bdev;
4817 item_size = btrfs_chunk_item_size(map->num_stripes);
4818 stripe_size = em->orig_block_len;
4820 chunk = kzalloc(item_size, GFP_NOFS);
4827 * Take the device list mutex to prevent races with the final phase of
4828 * a device replace operation that replaces the device object associated
4829 * with the map's stripes, because the device object's id can change
4830 * at any time during that final phase of the device replace operation
4831 * (dev-replace.c:btrfs_dev_replace_finishing()).
4833 mutex_lock(&chunk_root->fs_info->fs_devices->device_list_mutex);
4834 for (i = 0; i < map->num_stripes; i++) {
4835 device = map->stripes[i].dev;
4836 dev_offset = map->stripes[i].physical;
4838 ret = btrfs_update_device(trans, device);
4841 ret = btrfs_alloc_dev_extent(trans, device,
4842 chunk_root->root_key.objectid,
4843 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4844 chunk_offset, dev_offset,
4850 mutex_unlock(&chunk_root->fs_info->fs_devices->device_list_mutex);
4854 stripe = &chunk->stripe;
4855 for (i = 0; i < map->num_stripes; i++) {
4856 device = map->stripes[i].dev;
4857 dev_offset = map->stripes[i].physical;
4859 btrfs_set_stack_stripe_devid(stripe, device->devid);
4860 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4861 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4864 mutex_unlock(&chunk_root->fs_info->fs_devices->device_list_mutex);
4866 btrfs_set_stack_chunk_length(chunk, chunk_size);
4867 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4868 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4869 btrfs_set_stack_chunk_type(chunk, map->type);
4870 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4871 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4872 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4873 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4874 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4876 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4877 key.type = BTRFS_CHUNK_ITEM_KEY;
4878 key.offset = chunk_offset;
4880 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4881 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4883 * TODO: Cleanup of inserted chunk root in case of
4886 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
4892 free_extent_map(em);
4897 * Chunk allocation falls into two parts. The first part does works
4898 * that make the new allocated chunk useable, but not do any operation
4899 * that modifies the chunk tree. The second part does the works that
4900 * require modifying the chunk tree. This division is important for the
4901 * bootstrap process of adding storage to a seed btrfs.
4903 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4904 struct btrfs_root *extent_root, u64 type)
4908 ASSERT(mutex_is_locked(&extent_root->fs_info->chunk_mutex));
4909 chunk_offset = find_next_chunk(extent_root->fs_info);
4910 return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type);
4913 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4914 struct btrfs_root *root,
4915 struct btrfs_device *device)
4918 u64 sys_chunk_offset;
4920 struct btrfs_fs_info *fs_info = root->fs_info;
4921 struct btrfs_root *extent_root = fs_info->extent_root;
4924 chunk_offset = find_next_chunk(fs_info);
4925 alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
4926 ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset,
4931 sys_chunk_offset = find_next_chunk(root->fs_info);
4932 alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
4933 ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset,
4938 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
4942 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
4943 BTRFS_BLOCK_GROUP_RAID10 |
4944 BTRFS_BLOCK_GROUP_RAID5 |
4945 BTRFS_BLOCK_GROUP_DUP)) {
4947 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
4956 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
4958 struct extent_map *em;
4959 struct map_lookup *map;
4960 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4965 read_lock(&map_tree->map_tree.lock);
4966 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
4967 read_unlock(&map_tree->map_tree.lock);
4971 map = (struct map_lookup *)em->bdev;
4972 for (i = 0; i < map->num_stripes; i++) {
4973 if (map->stripes[i].dev->missing) {
4978 if (!map->stripes[i].dev->writeable) {
4985 * If the number of missing devices is larger than max errors,
4986 * we can not write the data into that chunk successfully, so
4989 if (miss_ndevs > btrfs_chunk_max_errors(map))
4992 free_extent_map(em);
4996 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
4998 extent_map_tree_init(&tree->map_tree);
5001 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
5003 struct extent_map *em;
5006 write_lock(&tree->map_tree.lock);
5007 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
5009 remove_extent_mapping(&tree->map_tree, em);
5010 write_unlock(&tree->map_tree.lock);
5014 free_extent_map(em);
5015 /* once for the tree */
5016 free_extent_map(em);
5020 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5022 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5023 struct extent_map *em;
5024 struct map_lookup *map;
5025 struct extent_map_tree *em_tree = &map_tree->map_tree;
5028 read_lock(&em_tree->lock);
5029 em = lookup_extent_mapping(em_tree, logical, len);
5030 read_unlock(&em_tree->lock);
5033 * We could return errors for these cases, but that could get ugly and
5034 * we'd probably do the same thing which is just not do anything else
5035 * and exit, so return 1 so the callers don't try to use other copies.
5038 btrfs_crit(fs_info, "No mapping for %Lu-%Lu", logical,
5043 if (em->start > logical || em->start + em->len < logical) {
5044 btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got "
5045 "%Lu-%Lu", logical, logical+len, em->start,
5046 em->start + em->len);
5047 free_extent_map(em);
5051 map = (struct map_lookup *)em->bdev;
5052 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
5053 ret = map->num_stripes;
5054 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5055 ret = map->sub_stripes;
5056 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5058 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5062 free_extent_map(em);
5064 btrfs_dev_replace_lock(&fs_info->dev_replace);
5065 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
5067 btrfs_dev_replace_unlock(&fs_info->dev_replace);
5072 unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
5073 struct btrfs_mapping_tree *map_tree,
5076 struct extent_map *em;
5077 struct map_lookup *map;
5078 struct extent_map_tree *em_tree = &map_tree->map_tree;
5079 unsigned long len = root->sectorsize;
5081 read_lock(&em_tree->lock);
5082 em = lookup_extent_mapping(em_tree, logical, len);
5083 read_unlock(&em_tree->lock);
5086 BUG_ON(em->start > logical || em->start + em->len < logical);
5087 map = (struct map_lookup *)em->bdev;
5088 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5089 len = map->stripe_len * nr_data_stripes(map);
5090 free_extent_map(em);
5094 int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
5095 u64 logical, u64 len, int mirror_num)
5097 struct extent_map *em;
5098 struct map_lookup *map;
5099 struct extent_map_tree *em_tree = &map_tree->map_tree;
5102 read_lock(&em_tree->lock);
5103 em = lookup_extent_mapping(em_tree, logical, len);
5104 read_unlock(&em_tree->lock);
5107 BUG_ON(em->start > logical || em->start + em->len < logical);
5108 map = (struct map_lookup *)em->bdev;
5109 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5111 free_extent_map(em);
5115 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5116 struct map_lookup *map, int first, int num,
5117 int optimal, int dev_replace_is_ongoing)
5121 struct btrfs_device *srcdev;
5123 if (dev_replace_is_ongoing &&
5124 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5125 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5126 srcdev = fs_info->dev_replace.srcdev;
5131 * try to avoid the drive that is the source drive for a
5132 * dev-replace procedure, only choose it if no other non-missing
5133 * mirror is available
5135 for (tolerance = 0; tolerance < 2; tolerance++) {
5136 if (map->stripes[optimal].dev->bdev &&
5137 (tolerance || map->stripes[optimal].dev != srcdev))
5139 for (i = first; i < first + num; i++) {
5140 if (map->stripes[i].dev->bdev &&
5141 (tolerance || map->stripes[i].dev != srcdev))
5146 /* we couldn't find one that doesn't fail. Just return something
5147 * and the io error handling code will clean up eventually
5152 static inline int parity_smaller(u64 a, u64 b)
5157 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5158 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5160 struct btrfs_bio_stripe s;
5167 for (i = 0; i < num_stripes - 1; i++) {
5168 if (parity_smaller(bbio->raid_map[i],
5169 bbio->raid_map[i+1])) {
5170 s = bbio->stripes[i];
5171 l = bbio->raid_map[i];
5172 bbio->stripes[i] = bbio->stripes[i+1];
5173 bbio->raid_map[i] = bbio->raid_map[i+1];
5174 bbio->stripes[i+1] = s;
5175 bbio->raid_map[i+1] = l;
5183 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5185 struct btrfs_bio *bbio = kzalloc(
5186 /* the size of the btrfs_bio */
5187 sizeof(struct btrfs_bio) +
5188 /* plus the variable array for the stripes */
5189 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5190 /* plus the variable array for the tgt dev */
5191 sizeof(int) * (real_stripes) +
5193 * plus the raid_map, which includes both the tgt dev
5196 sizeof(u64) * (total_stripes),
5197 GFP_NOFS|__GFP_NOFAIL);
5199 atomic_set(&bbio->error, 0);
5200 atomic_set(&bbio->refs, 1);
5205 void btrfs_get_bbio(struct btrfs_bio *bbio)
5207 WARN_ON(!atomic_read(&bbio->refs));
5208 atomic_inc(&bbio->refs);
5211 void btrfs_put_bbio(struct btrfs_bio *bbio)
5215 if (atomic_dec_and_test(&bbio->refs))
5219 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5220 u64 logical, u64 *length,
5221 struct btrfs_bio **bbio_ret,
5222 int mirror_num, int need_raid_map)
5224 struct extent_map *em;
5225 struct map_lookup *map;
5226 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5227 struct extent_map_tree *em_tree = &map_tree->map_tree;
5230 u64 stripe_end_offset;
5240 int tgtdev_indexes = 0;
5241 struct btrfs_bio *bbio = NULL;
5242 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
5243 int dev_replace_is_ongoing = 0;
5244 int num_alloc_stripes;
5245 int patch_the_first_stripe_for_dev_replace = 0;
5246 u64 physical_to_patch_in_first_stripe = 0;
5247 u64 raid56_full_stripe_start = (u64)-1;
5249 read_lock(&em_tree->lock);
5250 em = lookup_extent_mapping(em_tree, logical, *length);
5251 read_unlock(&em_tree->lock);
5254 btrfs_crit(fs_info, "unable to find logical %llu len %llu",
5259 if (em->start > logical || em->start + em->len < logical) {
5260 btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, "
5261 "found %Lu-%Lu", logical, em->start,
5262 em->start + em->len);
5263 free_extent_map(em);
5267 map = (struct map_lookup *)em->bdev;
5268 offset = logical - em->start;
5270 stripe_len = map->stripe_len;
5273 * stripe_nr counts the total number of stripes we have to stride
5274 * to get to this block
5276 stripe_nr = div64_u64(stripe_nr, stripe_len);
5278 stripe_offset = stripe_nr * stripe_len;
5279 BUG_ON(offset < stripe_offset);
5281 /* stripe_offset is the offset of this block in its stripe*/
5282 stripe_offset = offset - stripe_offset;
5284 /* if we're here for raid56, we need to know the stripe aligned start */
5285 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5286 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5287 raid56_full_stripe_start = offset;
5289 /* allow a write of a full stripe, but make sure we don't
5290 * allow straddling of stripes
5292 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5294 raid56_full_stripe_start *= full_stripe_len;
5297 if (rw & REQ_DISCARD) {
5298 /* we don't discard raid56 yet */
5299 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5303 *length = min_t(u64, em->len - offset, *length);
5304 } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5306 /* For writes to RAID[56], allow a full stripeset across all disks.
5307 For other RAID types and for RAID[56] reads, just allow a single
5308 stripe (on a single disk). */
5309 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
5311 max_len = stripe_len * nr_data_stripes(map) -
5312 (offset - raid56_full_stripe_start);
5314 /* we limit the length of each bio to what fits in a stripe */
5315 max_len = stripe_len - stripe_offset;
5317 *length = min_t(u64, em->len - offset, max_len);
5319 *length = em->len - offset;
5322 /* This is for when we're called from btrfs_merge_bio_hook() and all
5323 it cares about is the length */
5327 btrfs_dev_replace_lock(dev_replace);
5328 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5329 if (!dev_replace_is_ongoing)
5330 btrfs_dev_replace_unlock(dev_replace);
5332 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
5333 !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) &&
5334 dev_replace->tgtdev != NULL) {
5336 * in dev-replace case, for repair case (that's the only
5337 * case where the mirror is selected explicitly when
5338 * calling btrfs_map_block), blocks left of the left cursor
5339 * can also be read from the target drive.
5340 * For REQ_GET_READ_MIRRORS, the target drive is added as
5341 * the last one to the array of stripes. For READ, it also
5342 * needs to be supported using the same mirror number.
5343 * If the requested block is not left of the left cursor,
5344 * EIO is returned. This can happen because btrfs_num_copies()
5345 * returns one more in the dev-replace case.
5347 u64 tmp_length = *length;
5348 struct btrfs_bio *tmp_bbio = NULL;
5349 int tmp_num_stripes;
5350 u64 srcdev_devid = dev_replace->srcdev->devid;
5351 int index_srcdev = 0;
5353 u64 physical_of_found = 0;
5355 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
5356 logical, &tmp_length, &tmp_bbio, 0, 0);
5358 WARN_ON(tmp_bbio != NULL);
5362 tmp_num_stripes = tmp_bbio->num_stripes;
5363 if (mirror_num > tmp_num_stripes) {
5365 * REQ_GET_READ_MIRRORS does not contain this
5366 * mirror, that means that the requested area
5367 * is not left of the left cursor
5370 btrfs_put_bbio(tmp_bbio);
5375 * process the rest of the function using the mirror_num
5376 * of the source drive. Therefore look it up first.
5377 * At the end, patch the device pointer to the one of the
5380 for (i = 0; i < tmp_num_stripes; i++) {
5381 if (tmp_bbio->stripes[i].dev->devid == srcdev_devid) {
5383 * In case of DUP, in order to keep it
5384 * simple, only add the mirror with the
5385 * lowest physical address
5388 physical_of_found <=
5389 tmp_bbio->stripes[i].physical)
5394 tmp_bbio->stripes[i].physical;
5399 mirror_num = index_srcdev + 1;
5400 patch_the_first_stripe_for_dev_replace = 1;
5401 physical_to_patch_in_first_stripe = physical_of_found;
5405 btrfs_put_bbio(tmp_bbio);
5409 btrfs_put_bbio(tmp_bbio);
5410 } else if (mirror_num > map->num_stripes) {
5416 stripe_nr_orig = stripe_nr;
5417 stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
5418 stripe_nr_end = div_u64(stripe_nr_end, map->stripe_len);
5419 stripe_end_offset = stripe_nr_end * map->stripe_len -
5422 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5423 if (rw & REQ_DISCARD)
5424 num_stripes = min_t(u64, map->num_stripes,
5425 stripe_nr_end - stripe_nr_orig);
5426 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5428 if (!(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)))
5430 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
5431 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))
5432 num_stripes = map->num_stripes;
5433 else if (mirror_num)
5434 stripe_index = mirror_num - 1;
5436 stripe_index = find_live_mirror(fs_info, map, 0,
5438 current->pid % map->num_stripes,
5439 dev_replace_is_ongoing);
5440 mirror_num = stripe_index + 1;
5443 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
5444 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
5445 num_stripes = map->num_stripes;
5446 } else if (mirror_num) {
5447 stripe_index = mirror_num - 1;
5452 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5453 u32 factor = map->num_stripes / map->sub_stripes;
5455 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5456 stripe_index *= map->sub_stripes;
5458 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
5459 num_stripes = map->sub_stripes;
5460 else if (rw & REQ_DISCARD)
5461 num_stripes = min_t(u64, map->sub_stripes *
5462 (stripe_nr_end - stripe_nr_orig),
5464 else if (mirror_num)
5465 stripe_index += mirror_num - 1;
5467 int old_stripe_index = stripe_index;
5468 stripe_index = find_live_mirror(fs_info, map,
5470 map->sub_stripes, stripe_index +
5471 current->pid % map->sub_stripes,
5472 dev_replace_is_ongoing);
5473 mirror_num = stripe_index - old_stripe_index + 1;
5476 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5477 if (need_raid_map &&
5478 ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) ||
5480 /* push stripe_nr back to the start of the full stripe */
5481 stripe_nr = div_u64(raid56_full_stripe_start,
5482 stripe_len * nr_data_stripes(map));
5484 /* RAID[56] write or recovery. Return all stripes */
5485 num_stripes = map->num_stripes;
5486 max_errors = nr_parity_stripes(map);
5488 *length = map->stripe_len;
5493 * Mirror #0 or #1 means the original data block.
5494 * Mirror #2 is RAID5 parity block.
5495 * Mirror #3 is RAID6 Q block.
5497 stripe_nr = div_u64_rem(stripe_nr,
5498 nr_data_stripes(map), &stripe_index);
5500 stripe_index = nr_data_stripes(map) +
5503 /* We distribute the parity blocks across stripes */
5504 div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
5506 if (!(rw & (REQ_WRITE | REQ_DISCARD |
5507 REQ_GET_READ_MIRRORS)) && mirror_num <= 1)
5512 * after this, stripe_nr is the number of stripes on this
5513 * device we have to walk to find the data, and stripe_index is
5514 * the number of our device in the stripe array
5516 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5518 mirror_num = stripe_index + 1;
5520 BUG_ON(stripe_index >= map->num_stripes);
5522 num_alloc_stripes = num_stripes;
5523 if (dev_replace_is_ongoing) {
5524 if (rw & (REQ_WRITE | REQ_DISCARD))
5525 num_alloc_stripes <<= 1;
5526 if (rw & REQ_GET_READ_MIRRORS)
5527 num_alloc_stripes++;
5528 tgtdev_indexes = num_stripes;
5531 bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
5536 if (dev_replace_is_ongoing)
5537 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
5539 /* build raid_map */
5540 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK &&
5541 need_raid_map && ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) ||
5546 bbio->raid_map = (u64 *)((void *)bbio->stripes +
5547 sizeof(struct btrfs_bio_stripe) *
5549 sizeof(int) * tgtdev_indexes);
5551 /* Work out the disk rotation on this stripe-set */
5552 div_u64_rem(stripe_nr, num_stripes, &rot);
5554 /* Fill in the logical address of each stripe */
5555 tmp = stripe_nr * nr_data_stripes(map);
5556 for (i = 0; i < nr_data_stripes(map); i++)
5557 bbio->raid_map[(i+rot) % num_stripes] =
5558 em->start + (tmp + i) * map->stripe_len;
5560 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5561 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5562 bbio->raid_map[(i+rot+1) % num_stripes] =
5566 if (rw & REQ_DISCARD) {
5568 u32 sub_stripes = 0;
5569 u64 stripes_per_dev = 0;
5570 u32 remaining_stripes = 0;
5571 u32 last_stripe = 0;
5574 (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
5575 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5578 sub_stripes = map->sub_stripes;
5580 factor = map->num_stripes / sub_stripes;
5581 stripes_per_dev = div_u64_rem(stripe_nr_end -
5584 &remaining_stripes);
5585 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5586 last_stripe *= sub_stripes;
5589 for (i = 0; i < num_stripes; i++) {
5590 bbio->stripes[i].physical =
5591 map->stripes[stripe_index].physical +
5592 stripe_offset + stripe_nr * map->stripe_len;
5593 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5595 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5596 BTRFS_BLOCK_GROUP_RAID10)) {
5597 bbio->stripes[i].length = stripes_per_dev *
5600 if (i / sub_stripes < remaining_stripes)
5601 bbio->stripes[i].length +=
5605 * Special for the first stripe and
5608 * |-------|...|-------|
5612 if (i < sub_stripes)
5613 bbio->stripes[i].length -=
5616 if (stripe_index >= last_stripe &&
5617 stripe_index <= (last_stripe +
5619 bbio->stripes[i].length -=
5622 if (i == sub_stripes - 1)
5625 bbio->stripes[i].length = *length;
5628 if (stripe_index == map->num_stripes) {
5629 /* This could only happen for RAID0/10 */
5635 for (i = 0; i < num_stripes; i++) {
5636 bbio->stripes[i].physical =
5637 map->stripes[stripe_index].physical +
5639 stripe_nr * map->stripe_len;
5640 bbio->stripes[i].dev =
5641 map->stripes[stripe_index].dev;
5646 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
5647 max_errors = btrfs_chunk_max_errors(map);
5650 sort_parity_stripes(bbio, num_stripes);
5653 if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) &&
5654 dev_replace->tgtdev != NULL) {
5655 int index_where_to_add;
5656 u64 srcdev_devid = dev_replace->srcdev->devid;
5659 * duplicate the write operations while the dev replace
5660 * procedure is running. Since the copying of the old disk
5661 * to the new disk takes place at run time while the
5662 * filesystem is mounted writable, the regular write
5663 * operations to the old disk have to be duplicated to go
5664 * to the new disk as well.
5665 * Note that device->missing is handled by the caller, and
5666 * that the write to the old disk is already set up in the
5669 index_where_to_add = num_stripes;
5670 for (i = 0; i < num_stripes; i++) {
5671 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5672 /* write to new disk, too */
5673 struct btrfs_bio_stripe *new =
5674 bbio->stripes + index_where_to_add;
5675 struct btrfs_bio_stripe *old =
5678 new->physical = old->physical;
5679 new->length = old->length;
5680 new->dev = dev_replace->tgtdev;
5681 bbio->tgtdev_map[i] = index_where_to_add;
5682 index_where_to_add++;
5687 num_stripes = index_where_to_add;
5688 } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) &&
5689 dev_replace->tgtdev != NULL) {
5690 u64 srcdev_devid = dev_replace->srcdev->devid;
5691 int index_srcdev = 0;
5693 u64 physical_of_found = 0;
5696 * During the dev-replace procedure, the target drive can
5697 * also be used to read data in case it is needed to repair
5698 * a corrupt block elsewhere. This is possible if the
5699 * requested area is left of the left cursor. In this area,
5700 * the target drive is a full copy of the source drive.
5702 for (i = 0; i < num_stripes; i++) {
5703 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5705 * In case of DUP, in order to keep it
5706 * simple, only add the mirror with the
5707 * lowest physical address
5710 physical_of_found <=
5711 bbio->stripes[i].physical)
5715 physical_of_found = bbio->stripes[i].physical;
5719 if (physical_of_found + map->stripe_len <=
5720 dev_replace->cursor_left) {
5721 struct btrfs_bio_stripe *tgtdev_stripe =
5722 bbio->stripes + num_stripes;
5724 tgtdev_stripe->physical = physical_of_found;
5725 tgtdev_stripe->length =
5726 bbio->stripes[index_srcdev].length;
5727 tgtdev_stripe->dev = dev_replace->tgtdev;
5728 bbio->tgtdev_map[index_srcdev] = num_stripes;
5737 bbio->map_type = map->type;
5738 bbio->num_stripes = num_stripes;
5739 bbio->max_errors = max_errors;
5740 bbio->mirror_num = mirror_num;
5741 bbio->num_tgtdevs = tgtdev_indexes;
5744 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5745 * mirror_num == num_stripes + 1 && dev_replace target drive is
5746 * available as a mirror
5748 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5749 WARN_ON(num_stripes > 1);
5750 bbio->stripes[0].dev = dev_replace->tgtdev;
5751 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5752 bbio->mirror_num = map->num_stripes + 1;
5755 if (dev_replace_is_ongoing)
5756 btrfs_dev_replace_unlock(dev_replace);
5757 free_extent_map(em);
5761 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5762 u64 logical, u64 *length,
5763 struct btrfs_bio **bbio_ret, int mirror_num)
5765 return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5769 /* For Scrub/replace */
5770 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, int rw,
5771 u64 logical, u64 *length,
5772 struct btrfs_bio **bbio_ret, int mirror_num,
5775 return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5776 mirror_num, need_raid_map);
5779 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
5780 u64 chunk_start, u64 physical, u64 devid,
5781 u64 **logical, int *naddrs, int *stripe_len)
5783 struct extent_map_tree *em_tree = &map_tree->map_tree;
5784 struct extent_map *em;
5785 struct map_lookup *map;
5793 read_lock(&em_tree->lock);
5794 em = lookup_extent_mapping(em_tree, chunk_start, 1);
5795 read_unlock(&em_tree->lock);
5798 printk(KERN_ERR "BTRFS: couldn't find em for chunk %Lu\n",
5803 if (em->start != chunk_start) {
5804 printk(KERN_ERR "BTRFS: bad chunk start, em=%Lu, wanted=%Lu\n",
5805 em->start, chunk_start);
5806 free_extent_map(em);
5809 map = (struct map_lookup *)em->bdev;
5812 rmap_len = map->stripe_len;
5814 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5815 length = div_u64(length, map->num_stripes / map->sub_stripes);
5816 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5817 length = div_u64(length, map->num_stripes);
5818 else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5819 length = div_u64(length, nr_data_stripes(map));
5820 rmap_len = map->stripe_len * nr_data_stripes(map);
5823 buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
5824 BUG_ON(!buf); /* -ENOMEM */
5826 for (i = 0; i < map->num_stripes; i++) {
5827 if (devid && map->stripes[i].dev->devid != devid)
5829 if (map->stripes[i].physical > physical ||
5830 map->stripes[i].physical + length <= physical)
5833 stripe_nr = physical - map->stripes[i].physical;
5834 stripe_nr = div_u64(stripe_nr, map->stripe_len);
5836 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5837 stripe_nr = stripe_nr * map->num_stripes + i;
5838 stripe_nr = div_u64(stripe_nr, map->sub_stripes);
5839 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5840 stripe_nr = stripe_nr * map->num_stripes + i;
5841 } /* else if RAID[56], multiply by nr_data_stripes().
5842 * Alternatively, just use rmap_len below instead of
5843 * map->stripe_len */
5845 bytenr = chunk_start + stripe_nr * rmap_len;
5846 WARN_ON(nr >= map->num_stripes);
5847 for (j = 0; j < nr; j++) {
5848 if (buf[j] == bytenr)
5852 WARN_ON(nr >= map->num_stripes);
5859 *stripe_len = rmap_len;
5861 free_extent_map(em);
5865 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
5867 bio->bi_private = bbio->private;
5868 bio->bi_end_io = bbio->end_io;
5871 btrfs_put_bbio(bbio);
5874 static void btrfs_end_bio(struct bio *bio)
5876 struct btrfs_bio *bbio = bio->bi_private;
5877 int is_orig_bio = 0;
5879 if (bio->bi_error) {
5880 atomic_inc(&bbio->error);
5881 if (bio->bi_error == -EIO || bio->bi_error == -EREMOTEIO) {
5882 unsigned int stripe_index =
5883 btrfs_io_bio(bio)->stripe_index;
5884 struct btrfs_device *dev;
5886 BUG_ON(stripe_index >= bbio->num_stripes);
5887 dev = bbio->stripes[stripe_index].dev;
5889 if (bio->bi_rw & WRITE)
5890 btrfs_dev_stat_inc(dev,
5891 BTRFS_DEV_STAT_WRITE_ERRS);
5893 btrfs_dev_stat_inc(dev,
5894 BTRFS_DEV_STAT_READ_ERRS);
5895 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
5896 btrfs_dev_stat_inc(dev,
5897 BTRFS_DEV_STAT_FLUSH_ERRS);
5898 btrfs_dev_stat_print_on_error(dev);
5903 if (bio == bbio->orig_bio)
5906 btrfs_bio_counter_dec(bbio->fs_info);
5908 if (atomic_dec_and_test(&bbio->stripes_pending)) {
5911 bio = bbio->orig_bio;
5914 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5915 /* only send an error to the higher layers if it is
5916 * beyond the tolerance of the btrfs bio
5918 if (atomic_read(&bbio->error) > bbio->max_errors) {
5919 bio->bi_error = -EIO;
5922 * this bio is actually up to date, we didn't
5923 * go over the max number of errors
5928 btrfs_end_bbio(bbio, bio);
5929 } else if (!is_orig_bio) {
5935 * see run_scheduled_bios for a description of why bios are collected for
5938 * This will add one bio to the pending list for a device and make sure
5939 * the work struct is scheduled.
5941 static noinline void btrfs_schedule_bio(struct btrfs_root *root,
5942 struct btrfs_device *device,
5943 int rw, struct bio *bio)
5945 int should_queue = 1;
5946 struct btrfs_pending_bios *pending_bios;
5948 if (device->missing || !device->bdev) {
5953 /* don't bother with additional async steps for reads, right now */
5954 if (!(rw & REQ_WRITE)) {
5956 btrfsic_submit_bio(rw, bio);
5962 * nr_async_bios allows us to reliably return congestion to the
5963 * higher layers. Otherwise, the async bio makes it appear we have
5964 * made progress against dirty pages when we've really just put it
5965 * on a queue for later
5967 atomic_inc(&root->fs_info->nr_async_bios);
5968 WARN_ON(bio->bi_next);
5969 bio->bi_next = NULL;
5972 spin_lock(&device->io_lock);
5973 if (bio->bi_rw & REQ_SYNC)
5974 pending_bios = &device->pending_sync_bios;
5976 pending_bios = &device->pending_bios;
5978 if (pending_bios->tail)
5979 pending_bios->tail->bi_next = bio;
5981 pending_bios->tail = bio;
5982 if (!pending_bios->head)
5983 pending_bios->head = bio;
5984 if (device->running_pending)
5987 spin_unlock(&device->io_lock);
5990 btrfs_queue_work(root->fs_info->submit_workers,
5994 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5995 struct bio *bio, u64 physical, int dev_nr,
5998 struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
6000 bio->bi_private = bbio;
6001 btrfs_io_bio(bio)->stripe_index = dev_nr;
6002 bio->bi_end_io = btrfs_end_bio;
6003 bio->bi_iter.bi_sector = physical >> 9;
6006 struct rcu_string *name;
6009 name = rcu_dereference(dev->name);
6010 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
6011 "(%s id %llu), size=%u\n", rw,
6012 (u64)bio->bi_iter.bi_sector, (u_long)dev->bdev->bd_dev,
6013 name->str, dev->devid, bio->bi_iter.bi_size);
6017 bio->bi_bdev = dev->bdev;
6019 btrfs_bio_counter_inc_noblocked(root->fs_info);
6022 btrfs_schedule_bio(root, dev, rw, bio);
6024 btrfsic_submit_bio(rw, bio);
6027 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6029 atomic_inc(&bbio->error);
6030 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6031 /* Shoud be the original bio. */
6032 WARN_ON(bio != bbio->orig_bio);
6034 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6035 bio->bi_iter.bi_sector = logical >> 9;
6036 bio->bi_error = -EIO;
6037 btrfs_end_bbio(bbio, bio);
6041 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
6042 int mirror_num, int async_submit)
6044 struct btrfs_device *dev;
6045 struct bio *first_bio = bio;
6046 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
6052 struct btrfs_bio *bbio = NULL;
6054 length = bio->bi_iter.bi_size;
6055 map_length = length;
6057 btrfs_bio_counter_inc_blocked(root->fs_info);
6058 ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
6061 btrfs_bio_counter_dec(root->fs_info);
6065 total_devs = bbio->num_stripes;
6066 bbio->orig_bio = first_bio;
6067 bbio->private = first_bio->bi_private;
6068 bbio->end_io = first_bio->bi_end_io;
6069 bbio->fs_info = root->fs_info;
6070 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6072 if (bbio->raid_map) {
6073 /* In this case, map_length has been set to the length of
6074 a single stripe; not the whole write */
6076 ret = raid56_parity_write(root, bio, bbio, map_length);
6078 ret = raid56_parity_recover(root, bio, bbio, map_length,
6082 btrfs_bio_counter_dec(root->fs_info);
6086 if (map_length < length) {
6087 btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu",
6088 logical, length, map_length);
6092 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6093 dev = bbio->stripes[dev_nr].dev;
6094 if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
6095 bbio_error(bbio, first_bio, logical);
6099 if (dev_nr < total_devs - 1) {
6100 bio = btrfs_bio_clone(first_bio, GFP_NOFS);
6101 BUG_ON(!bio); /* -ENOMEM */
6105 submit_stripe_bio(root, bbio, bio,
6106 bbio->stripes[dev_nr].physical, dev_nr, rw,
6109 btrfs_bio_counter_dec(root->fs_info);
6113 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
6116 struct btrfs_device *device;
6117 struct btrfs_fs_devices *cur_devices;
6119 cur_devices = fs_info->fs_devices;
6120 while (cur_devices) {
6122 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
6123 device = __find_device(&cur_devices->devices,
6128 cur_devices = cur_devices->seed;
6133 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
6134 struct btrfs_fs_devices *fs_devices,
6135 u64 devid, u8 *dev_uuid)
6137 struct btrfs_device *device;
6139 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6143 list_add(&device->dev_list, &fs_devices->devices);
6144 device->fs_devices = fs_devices;
6145 fs_devices->num_devices++;
6147 device->missing = 1;
6148 fs_devices->missing_devices++;
6154 * btrfs_alloc_device - allocate struct btrfs_device
6155 * @fs_info: used only for generating a new devid, can be NULL if
6156 * devid is provided (i.e. @devid != NULL).
6157 * @devid: a pointer to devid for this device. If NULL a new devid
6159 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6162 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6163 * on error. Returned struct is not linked onto any lists and can be
6164 * destroyed with kfree() right away.
6166 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6170 struct btrfs_device *dev;
6173 if (WARN_ON(!devid && !fs_info))
6174 return ERR_PTR(-EINVAL);
6176 dev = __alloc_device();
6185 ret = find_next_devid(fs_info, &tmp);
6188 return ERR_PTR(ret);
6194 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6196 generate_random_uuid(dev->uuid);
6198 btrfs_init_work(&dev->work, btrfs_submit_helper,
6199 pending_bios_fn, NULL, NULL);
6204 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
6205 struct extent_buffer *leaf,
6206 struct btrfs_chunk *chunk)
6208 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
6209 struct map_lookup *map;
6210 struct extent_map *em;
6214 u8 uuid[BTRFS_UUID_SIZE];
6219 logical = key->offset;
6220 length = btrfs_chunk_length(leaf, chunk);
6222 read_lock(&map_tree->map_tree.lock);
6223 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
6224 read_unlock(&map_tree->map_tree.lock);
6226 /* already mapped? */
6227 if (em && em->start <= logical && em->start + em->len > logical) {
6228 free_extent_map(em);
6231 free_extent_map(em);
6234 em = alloc_extent_map();
6237 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6238 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6240 free_extent_map(em);
6244 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6245 em->bdev = (struct block_device *)map;
6246 em->start = logical;
6249 em->block_start = 0;
6250 em->block_len = em->len;
6252 map->num_stripes = num_stripes;
6253 map->io_width = btrfs_chunk_io_width(leaf, chunk);
6254 map->io_align = btrfs_chunk_io_align(leaf, chunk);
6255 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
6256 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6257 map->type = btrfs_chunk_type(leaf, chunk);
6258 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6259 for (i = 0; i < num_stripes; i++) {
6260 map->stripes[i].physical =
6261 btrfs_stripe_offset_nr(leaf, chunk, i);
6262 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6263 read_extent_buffer(leaf, uuid, (unsigned long)
6264 btrfs_stripe_dev_uuid_nr(chunk, i),
6266 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
6268 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
6269 free_extent_map(em);
6272 if (!map->stripes[i].dev) {
6273 map->stripes[i].dev =
6274 add_missing_dev(root, root->fs_info->fs_devices,
6276 if (!map->stripes[i].dev) {
6277 free_extent_map(em);
6280 btrfs_warn(root->fs_info, "devid %llu uuid %pU is missing",
6283 map->stripes[i].dev->in_fs_metadata = 1;
6286 write_lock(&map_tree->map_tree.lock);
6287 ret = add_extent_mapping(&map_tree->map_tree, em, 0);
6288 write_unlock(&map_tree->map_tree.lock);
6289 BUG_ON(ret); /* Tree corruption */
6290 free_extent_map(em);
6295 static void fill_device_from_item(struct extent_buffer *leaf,
6296 struct btrfs_dev_item *dev_item,
6297 struct btrfs_device *device)
6301 device->devid = btrfs_device_id(leaf, dev_item);
6302 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6303 device->total_bytes = device->disk_total_bytes;
6304 device->commit_total_bytes = device->disk_total_bytes;
6305 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6306 device->commit_bytes_used = device->bytes_used;
6307 device->type = btrfs_device_type(leaf, dev_item);
6308 device->io_align = btrfs_device_io_align(leaf, dev_item);
6309 device->io_width = btrfs_device_io_width(leaf, dev_item);
6310 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6311 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6312 device->is_tgtdev_for_dev_replace = 0;
6314 ptr = btrfs_device_uuid(dev_item);
6315 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6318 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_root *root,
6321 struct btrfs_fs_devices *fs_devices;
6324 BUG_ON(!mutex_is_locked(&uuid_mutex));
6326 fs_devices = root->fs_info->fs_devices->seed;
6327 while (fs_devices) {
6328 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE))
6331 fs_devices = fs_devices->seed;
6334 fs_devices = find_fsid(fsid);
6336 if (!btrfs_test_opt(root, DEGRADED))
6337 return ERR_PTR(-ENOENT);
6339 fs_devices = alloc_fs_devices(fsid);
6340 if (IS_ERR(fs_devices))
6343 fs_devices->seeding = 1;
6344 fs_devices->opened = 1;
6348 fs_devices = clone_fs_devices(fs_devices);
6349 if (IS_ERR(fs_devices))
6352 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
6353 root->fs_info->bdev_holder);
6355 free_fs_devices(fs_devices);
6356 fs_devices = ERR_PTR(ret);
6360 if (!fs_devices->seeding) {
6361 __btrfs_close_devices(fs_devices);
6362 free_fs_devices(fs_devices);
6363 fs_devices = ERR_PTR(-EINVAL);
6367 fs_devices->seed = root->fs_info->fs_devices->seed;
6368 root->fs_info->fs_devices->seed = fs_devices;
6373 static int read_one_dev(struct btrfs_root *root,
6374 struct extent_buffer *leaf,
6375 struct btrfs_dev_item *dev_item)
6377 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6378 struct btrfs_device *device;
6381 u8 fs_uuid[BTRFS_UUID_SIZE];
6382 u8 dev_uuid[BTRFS_UUID_SIZE];
6384 devid = btrfs_device_id(leaf, dev_item);
6385 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6387 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6390 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
6391 fs_devices = open_seed_devices(root, fs_uuid);
6392 if (IS_ERR(fs_devices))
6393 return PTR_ERR(fs_devices);
6396 device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
6398 if (!btrfs_test_opt(root, DEGRADED))
6401 device = add_missing_dev(root, fs_devices, devid, dev_uuid);
6404 btrfs_warn(root->fs_info, "devid %llu uuid %pU missing",
6407 if (!device->bdev && !btrfs_test_opt(root, DEGRADED))
6410 if(!device->bdev && !device->missing) {
6412 * this happens when a device that was properly setup
6413 * in the device info lists suddenly goes bad.
6414 * device->bdev is NULL, and so we have to set
6415 * device->missing to one here
6417 device->fs_devices->missing_devices++;
6418 device->missing = 1;
6421 /* Move the device to its own fs_devices */
6422 if (device->fs_devices != fs_devices) {
6423 ASSERT(device->missing);
6425 list_move(&device->dev_list, &fs_devices->devices);
6426 device->fs_devices->num_devices--;
6427 fs_devices->num_devices++;
6429 device->fs_devices->missing_devices--;
6430 fs_devices->missing_devices++;
6432 device->fs_devices = fs_devices;
6436 if (device->fs_devices != root->fs_info->fs_devices) {
6437 BUG_ON(device->writeable);
6438 if (device->generation !=
6439 btrfs_device_generation(leaf, dev_item))
6443 fill_device_from_item(leaf, dev_item, device);
6444 device->in_fs_metadata = 1;
6445 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
6446 device->fs_devices->total_rw_bytes += device->total_bytes;
6447 spin_lock(&root->fs_info->free_chunk_lock);
6448 root->fs_info->free_chunk_space += device->total_bytes -
6450 spin_unlock(&root->fs_info->free_chunk_lock);
6456 int btrfs_read_sys_array(struct btrfs_root *root)
6458 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
6459 struct extent_buffer *sb;
6460 struct btrfs_disk_key *disk_key;
6461 struct btrfs_chunk *chunk;
6463 unsigned long sb_array_offset;
6469 struct btrfs_key key;
6471 ASSERT(BTRFS_SUPER_INFO_SIZE <= root->nodesize);
6473 * This will create extent buffer of nodesize, superblock size is
6474 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6475 * overallocate but we can keep it as-is, only the first page is used.
6477 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET);
6480 set_extent_buffer_uptodate(sb);
6481 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6483 * The sb extent buffer is artifical and just used to read the system array.
6484 * set_extent_buffer_uptodate() call does not properly mark all it's
6485 * pages up-to-date when the page is larger: extent does not cover the
6486 * whole page and consequently check_page_uptodate does not find all
6487 * the page's extents up-to-date (the hole beyond sb),
6488 * write_extent_buffer then triggers a WARN_ON.
6490 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6491 * but sb spans only this function. Add an explicit SetPageUptodate call
6492 * to silence the warning eg. on PowerPC 64.
6494 if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
6495 SetPageUptodate(sb->pages[0]);
6497 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6498 array_size = btrfs_super_sys_array_size(super_copy);
6500 array_ptr = super_copy->sys_chunk_array;
6501 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6504 while (cur_offset < array_size) {
6505 disk_key = (struct btrfs_disk_key *)array_ptr;
6506 len = sizeof(*disk_key);
6507 if (cur_offset + len > array_size)
6508 goto out_short_read;
6510 btrfs_disk_key_to_cpu(&key, disk_key);
6513 sb_array_offset += len;
6516 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6517 chunk = (struct btrfs_chunk *)sb_array_offset;
6519 * At least one btrfs_chunk with one stripe must be
6520 * present, exact stripe count check comes afterwards
6522 len = btrfs_chunk_item_size(1);
6523 if (cur_offset + len > array_size)
6524 goto out_short_read;
6526 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6529 "BTRFS: invalid number of stripes %u in sys_array at offset %u\n",
6530 num_stripes, cur_offset);
6535 len = btrfs_chunk_item_size(num_stripes);
6536 if (cur_offset + len > array_size)
6537 goto out_short_read;
6539 ret = read_one_chunk(root, &key, sb, chunk);
6544 "BTRFS: unexpected item type %u in sys_array at offset %u\n",
6545 (u32)key.type, cur_offset);
6550 sb_array_offset += len;
6553 free_extent_buffer(sb);
6557 printk(KERN_ERR "BTRFS: sys_array too short to read %u bytes at offset %u\n",
6559 free_extent_buffer(sb);
6563 int btrfs_read_chunk_tree(struct btrfs_root *root)
6565 struct btrfs_path *path;
6566 struct extent_buffer *leaf;
6567 struct btrfs_key key;
6568 struct btrfs_key found_key;
6572 root = root->fs_info->chunk_root;
6574 path = btrfs_alloc_path();
6578 mutex_lock(&uuid_mutex);
6582 * Read all device items, and then all the chunk items. All
6583 * device items are found before any chunk item (their object id
6584 * is smaller than the lowest possible object id for a chunk
6585 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6587 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6590 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6594 leaf = path->nodes[0];
6595 slot = path->slots[0];
6596 if (slot >= btrfs_header_nritems(leaf)) {
6597 ret = btrfs_next_leaf(root, path);
6604 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6605 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6606 struct btrfs_dev_item *dev_item;
6607 dev_item = btrfs_item_ptr(leaf, slot,
6608 struct btrfs_dev_item);
6609 ret = read_one_dev(root, leaf, dev_item);
6612 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6613 struct btrfs_chunk *chunk;
6614 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6615 ret = read_one_chunk(root, &found_key, leaf, chunk);
6623 unlock_chunks(root);
6624 mutex_unlock(&uuid_mutex);
6626 btrfs_free_path(path);
6630 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6632 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6633 struct btrfs_device *device;
6635 while (fs_devices) {
6636 mutex_lock(&fs_devices->device_list_mutex);
6637 list_for_each_entry(device, &fs_devices->devices, dev_list)
6638 device->dev_root = fs_info->dev_root;
6639 mutex_unlock(&fs_devices->device_list_mutex);
6641 fs_devices = fs_devices->seed;
6645 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6649 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6650 btrfs_dev_stat_reset(dev, i);
6653 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6655 struct btrfs_key key;
6656 struct btrfs_key found_key;
6657 struct btrfs_root *dev_root = fs_info->dev_root;
6658 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6659 struct extent_buffer *eb;
6662 struct btrfs_device *device;
6663 struct btrfs_path *path = NULL;
6666 path = btrfs_alloc_path();
6672 mutex_lock(&fs_devices->device_list_mutex);
6673 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6675 struct btrfs_dev_stats_item *ptr;
6678 key.type = BTRFS_DEV_STATS_KEY;
6679 key.offset = device->devid;
6680 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6682 __btrfs_reset_dev_stats(device);
6683 device->dev_stats_valid = 1;
6684 btrfs_release_path(path);
6687 slot = path->slots[0];
6688 eb = path->nodes[0];
6689 btrfs_item_key_to_cpu(eb, &found_key, slot);
6690 item_size = btrfs_item_size_nr(eb, slot);
6692 ptr = btrfs_item_ptr(eb, slot,
6693 struct btrfs_dev_stats_item);
6695 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6696 if (item_size >= (1 + i) * sizeof(__le64))
6697 btrfs_dev_stat_set(device, i,
6698 btrfs_dev_stats_value(eb, ptr, i));
6700 btrfs_dev_stat_reset(device, i);
6703 device->dev_stats_valid = 1;
6704 btrfs_dev_stat_print_on_load(device);
6705 btrfs_release_path(path);
6707 mutex_unlock(&fs_devices->device_list_mutex);
6710 btrfs_free_path(path);
6711 return ret < 0 ? ret : 0;
6714 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
6715 struct btrfs_root *dev_root,
6716 struct btrfs_device *device)
6718 struct btrfs_path *path;
6719 struct btrfs_key key;
6720 struct extent_buffer *eb;
6721 struct btrfs_dev_stats_item *ptr;
6726 key.type = BTRFS_DEV_STATS_KEY;
6727 key.offset = device->devid;
6729 path = btrfs_alloc_path();
6731 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
6733 btrfs_warn_in_rcu(dev_root->fs_info,
6734 "error %d while searching for dev_stats item for device %s",
6735 ret, rcu_str_deref(device->name));
6740 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
6741 /* need to delete old one and insert a new one */
6742 ret = btrfs_del_item(trans, dev_root, path);
6744 btrfs_warn_in_rcu(dev_root->fs_info,
6745 "delete too small dev_stats item for device %s failed %d",
6746 rcu_str_deref(device->name), ret);
6753 /* need to insert a new item */
6754 btrfs_release_path(path);
6755 ret = btrfs_insert_empty_item(trans, dev_root, path,
6756 &key, sizeof(*ptr));
6758 btrfs_warn_in_rcu(dev_root->fs_info,
6759 "insert dev_stats item for device %s failed %d",
6760 rcu_str_deref(device->name), ret);
6765 eb = path->nodes[0];
6766 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
6767 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6768 btrfs_set_dev_stats_value(eb, ptr, i,
6769 btrfs_dev_stat_read(device, i));
6770 btrfs_mark_buffer_dirty(eb);
6773 btrfs_free_path(path);
6778 * called from commit_transaction. Writes all changed device stats to disk.
6780 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
6781 struct btrfs_fs_info *fs_info)
6783 struct btrfs_root *dev_root = fs_info->dev_root;
6784 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6785 struct btrfs_device *device;
6789 mutex_lock(&fs_devices->device_list_mutex);
6790 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6791 if (!device->dev_stats_valid || !btrfs_dev_stats_dirty(device))
6794 stats_cnt = atomic_read(&device->dev_stats_ccnt);
6795 ret = update_dev_stat_item(trans, dev_root, device);
6797 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
6799 mutex_unlock(&fs_devices->device_list_mutex);
6804 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
6806 btrfs_dev_stat_inc(dev, index);
6807 btrfs_dev_stat_print_on_error(dev);
6810 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
6812 if (!dev->dev_stats_valid)
6814 btrfs_err_rl_in_rcu(dev->dev_root->fs_info,
6815 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
6816 rcu_str_deref(dev->name),
6817 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6818 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6819 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6820 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6821 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6824 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
6828 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6829 if (btrfs_dev_stat_read(dev, i) != 0)
6831 if (i == BTRFS_DEV_STAT_VALUES_MAX)
6832 return; /* all values == 0, suppress message */
6834 btrfs_info_in_rcu(dev->dev_root->fs_info,
6835 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
6836 rcu_str_deref(dev->name),
6837 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6838 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6839 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6840 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6841 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6844 int btrfs_get_dev_stats(struct btrfs_root *root,
6845 struct btrfs_ioctl_get_dev_stats *stats)
6847 struct btrfs_device *dev;
6848 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6851 mutex_lock(&fs_devices->device_list_mutex);
6852 dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
6853 mutex_unlock(&fs_devices->device_list_mutex);
6856 btrfs_warn(root->fs_info, "get dev_stats failed, device not found");
6858 } else if (!dev->dev_stats_valid) {
6859 btrfs_warn(root->fs_info, "get dev_stats failed, not yet valid");
6861 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
6862 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6863 if (stats->nr_items > i)
6865 btrfs_dev_stat_read_and_reset(dev, i);
6867 btrfs_dev_stat_reset(dev, i);
6870 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6871 if (stats->nr_items > i)
6872 stats->values[i] = btrfs_dev_stat_read(dev, i);
6874 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
6875 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
6879 void btrfs_scratch_superblocks(struct block_device *bdev, char *device_path)
6881 struct buffer_head *bh;
6882 struct btrfs_super_block *disk_super;
6888 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX;
6891 if (btrfs_read_dev_one_super(bdev, copy_num, &bh))
6894 disk_super = (struct btrfs_super_block *)bh->b_data;
6896 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
6897 set_buffer_dirty(bh);
6898 sync_dirty_buffer(bh);
6902 /* Notify udev that device has changed */
6903 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
6905 /* Update ctime/mtime for device path for libblkid */
6906 update_dev_time(device_path);
6910 * Update the size of all devices, which is used for writing out the
6913 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
6915 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6916 struct btrfs_device *curr, *next;
6918 if (list_empty(&fs_devices->resized_devices))
6921 mutex_lock(&fs_devices->device_list_mutex);
6922 lock_chunks(fs_info->dev_root);
6923 list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
6925 list_del_init(&curr->resized_list);
6926 curr->commit_total_bytes = curr->disk_total_bytes;
6928 unlock_chunks(fs_info->dev_root);
6929 mutex_unlock(&fs_devices->device_list_mutex);
6932 /* Must be invoked during the transaction commit */
6933 void btrfs_update_commit_device_bytes_used(struct btrfs_root *root,
6934 struct btrfs_transaction *transaction)
6936 struct extent_map *em;
6937 struct map_lookup *map;
6938 struct btrfs_device *dev;
6941 if (list_empty(&transaction->pending_chunks))
6944 /* In order to kick the device replace finish process */
6946 list_for_each_entry(em, &transaction->pending_chunks, list) {
6947 map = (struct map_lookup *)em->bdev;
6949 for (i = 0; i < map->num_stripes; i++) {
6950 dev = map->stripes[i].dev;
6951 dev->commit_bytes_used = dev->bytes_used;
6954 unlock_chunks(root);
6957 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
6959 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6960 while (fs_devices) {
6961 fs_devices->fs_info = fs_info;
6962 fs_devices = fs_devices->seed;
6966 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
6968 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6969 while (fs_devices) {
6970 fs_devices->fs_info = NULL;
6971 fs_devices = fs_devices->seed;
6975 static void btrfs_close_one_device(struct btrfs_device *device)
6977 struct btrfs_fs_devices *fs_devices = device->fs_devices;
6978 struct btrfs_device *new_device;
6979 struct rcu_string *name;
6982 fs_devices->open_devices--;
6984 if (device->writeable &&
6985 device->devid != BTRFS_DEV_REPLACE_DEVID) {
6986 list_del_init(&device->dev_alloc_list);
6987 fs_devices->rw_devices--;
6990 if (device->missing)
6991 fs_devices->missing_devices--;
6993 new_device = btrfs_alloc_device(NULL, &device->devid,
6995 BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
6997 /* Safe because we are under uuid_mutex */
6999 name = rcu_string_strdup(device->name->str, GFP_NOFS);
7000 BUG_ON(!name); /* -ENOMEM */
7001 rcu_assign_pointer(new_device->name, name);
7004 list_replace_rcu(&device->dev_list, &new_device->dev_list);
7005 new_device->fs_devices = device->fs_devices;
7007 call_rcu(&device->rcu, free_device);