2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/bio.h>
20 #include <linux/slab.h>
21 #include <linux/buffer_head.h>
22 #include <linux/blkdev.h>
23 #include <linux/random.h>
24 #include <linux/iocontext.h>
25 #include <linux/capability.h>
26 #include <linux/ratelimit.h>
27 #include <linux/kthread.h>
28 #include <linux/raid/pq.h>
29 #include <linux/semaphore.h>
30 #include <asm/div64.h>
32 #include "extent_map.h"
34 #include "transaction.h"
35 #include "print-tree.h"
38 #include "async-thread.h"
39 #include "check-integrity.h"
40 #include "rcu-string.h"
42 #include "dev-replace.h"
45 const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
46 [BTRFS_RAID_RAID10] = {
49 .devs_max = 0, /* 0 == as many as possible */
51 .tolerated_failures = 1,
55 [BTRFS_RAID_RAID1] = {
60 .tolerated_failures = 1,
69 .tolerated_failures = 0,
73 [BTRFS_RAID_RAID0] = {
78 .tolerated_failures = 0,
82 [BTRFS_RAID_SINGLE] = {
87 .tolerated_failures = 0,
91 [BTRFS_RAID_RAID5] = {
96 .tolerated_failures = 1,
100 [BTRFS_RAID_RAID6] = {
105 .tolerated_failures = 2,
111 const u64 const btrfs_raid_group[BTRFS_NR_RAID_TYPES] = {
112 [BTRFS_RAID_RAID10] = BTRFS_BLOCK_GROUP_RAID10,
113 [BTRFS_RAID_RAID1] = BTRFS_BLOCK_GROUP_RAID1,
114 [BTRFS_RAID_DUP] = BTRFS_BLOCK_GROUP_DUP,
115 [BTRFS_RAID_RAID0] = BTRFS_BLOCK_GROUP_RAID0,
116 [BTRFS_RAID_SINGLE] = 0,
117 [BTRFS_RAID_RAID5] = BTRFS_BLOCK_GROUP_RAID5,
118 [BTRFS_RAID_RAID6] = BTRFS_BLOCK_GROUP_RAID6,
121 static int init_first_rw_device(struct btrfs_trans_handle *trans,
122 struct btrfs_root *root,
123 struct btrfs_device *device);
124 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
125 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
126 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
127 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
129 DEFINE_MUTEX(uuid_mutex);
130 static LIST_HEAD(fs_uuids);
131 struct list_head *btrfs_get_fs_uuids(void)
136 static struct btrfs_fs_devices *__alloc_fs_devices(void)
138 struct btrfs_fs_devices *fs_devs;
140 fs_devs = kzalloc(sizeof(*fs_devs), GFP_NOFS);
142 return ERR_PTR(-ENOMEM);
144 mutex_init(&fs_devs->device_list_mutex);
146 INIT_LIST_HEAD(&fs_devs->devices);
147 INIT_LIST_HEAD(&fs_devs->resized_devices);
148 INIT_LIST_HEAD(&fs_devs->alloc_list);
149 INIT_LIST_HEAD(&fs_devs->list);
155 * alloc_fs_devices - allocate struct btrfs_fs_devices
156 * @fsid: a pointer to UUID for this FS. If NULL a new UUID is
159 * Return: a pointer to a new &struct btrfs_fs_devices on success;
160 * ERR_PTR() on error. Returned struct is not linked onto any lists and
161 * can be destroyed with kfree() right away.
163 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
165 struct btrfs_fs_devices *fs_devs;
167 fs_devs = __alloc_fs_devices();
172 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
174 generate_random_uuid(fs_devs->fsid);
179 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
181 struct btrfs_device *device;
182 WARN_ON(fs_devices->opened);
183 while (!list_empty(&fs_devices->devices)) {
184 device = list_entry(fs_devices->devices.next,
185 struct btrfs_device, dev_list);
186 list_del(&device->dev_list);
187 rcu_string_free(device->name);
193 static void btrfs_kobject_uevent(struct block_device *bdev,
194 enum kobject_action action)
198 ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
200 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
202 kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
203 &disk_to_dev(bdev->bd_disk)->kobj);
206 void btrfs_cleanup_fs_uuids(void)
208 struct btrfs_fs_devices *fs_devices;
210 while (!list_empty(&fs_uuids)) {
211 fs_devices = list_entry(fs_uuids.next,
212 struct btrfs_fs_devices, list);
213 list_del(&fs_devices->list);
214 free_fs_devices(fs_devices);
218 static struct btrfs_device *__alloc_device(void)
220 struct btrfs_device *dev;
222 dev = kzalloc(sizeof(*dev), GFP_NOFS);
224 return ERR_PTR(-ENOMEM);
226 INIT_LIST_HEAD(&dev->dev_list);
227 INIT_LIST_HEAD(&dev->dev_alloc_list);
228 INIT_LIST_HEAD(&dev->resized_list);
230 spin_lock_init(&dev->io_lock);
232 spin_lock_init(&dev->reada_lock);
233 atomic_set(&dev->reada_in_flight, 0);
234 atomic_set(&dev->dev_stats_ccnt, 0);
235 INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_WAIT);
236 INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_WAIT);
241 static noinline struct btrfs_device *__find_device(struct list_head *head,
244 struct btrfs_device *dev;
246 list_for_each_entry(dev, head, dev_list) {
247 if (dev->devid == devid &&
248 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
255 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
257 struct btrfs_fs_devices *fs_devices;
259 list_for_each_entry(fs_devices, &fs_uuids, list) {
260 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
267 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
268 int flush, struct block_device **bdev,
269 struct buffer_head **bh)
273 *bdev = blkdev_get_by_path(device_path, flags, holder);
276 ret = PTR_ERR(*bdev);
281 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
282 ret = set_blocksize(*bdev, 4096);
284 blkdev_put(*bdev, flags);
287 invalidate_bdev(*bdev);
288 *bh = btrfs_read_dev_super(*bdev);
291 blkdev_put(*bdev, flags);
303 static void requeue_list(struct btrfs_pending_bios *pending_bios,
304 struct bio *head, struct bio *tail)
307 struct bio *old_head;
309 old_head = pending_bios->head;
310 pending_bios->head = head;
311 if (pending_bios->tail)
312 tail->bi_next = old_head;
314 pending_bios->tail = tail;
318 * we try to collect pending bios for a device so we don't get a large
319 * number of procs sending bios down to the same device. This greatly
320 * improves the schedulers ability to collect and merge the bios.
322 * But, it also turns into a long list of bios to process and that is sure
323 * to eventually make the worker thread block. The solution here is to
324 * make some progress and then put this work struct back at the end of
325 * the list if the block device is congested. This way, multiple devices
326 * can make progress from a single worker thread.
328 static noinline void run_scheduled_bios(struct btrfs_device *device)
331 struct backing_dev_info *bdi;
332 struct btrfs_fs_info *fs_info;
333 struct btrfs_pending_bios *pending_bios;
337 unsigned long num_run;
338 unsigned long batch_run = 0;
340 unsigned long last_waited = 0;
342 int sync_pending = 0;
343 struct blk_plug plug;
346 * this function runs all the bios we've collected for
347 * a particular device. We don't want to wander off to
348 * another device without first sending all of these down.
349 * So, setup a plug here and finish it off before we return
351 blk_start_plug(&plug);
353 bdi = blk_get_backing_dev_info(device->bdev);
354 fs_info = device->dev_root->fs_info;
355 limit = btrfs_async_submit_limit(fs_info);
356 limit = limit * 2 / 3;
359 spin_lock(&device->io_lock);
364 /* take all the bios off the list at once and process them
365 * later on (without the lock held). But, remember the
366 * tail and other pointers so the bios can be properly reinserted
367 * into the list if we hit congestion
369 if (!force_reg && device->pending_sync_bios.head) {
370 pending_bios = &device->pending_sync_bios;
373 pending_bios = &device->pending_bios;
377 pending = pending_bios->head;
378 tail = pending_bios->tail;
379 WARN_ON(pending && !tail);
382 * if pending was null this time around, no bios need processing
383 * at all and we can stop. Otherwise it'll loop back up again
384 * and do an additional check so no bios are missed.
386 * device->running_pending is used to synchronize with the
389 if (device->pending_sync_bios.head == NULL &&
390 device->pending_bios.head == NULL) {
392 device->running_pending = 0;
395 device->running_pending = 1;
398 pending_bios->head = NULL;
399 pending_bios->tail = NULL;
401 spin_unlock(&device->io_lock);
406 /* we want to work on both lists, but do more bios on the
407 * sync list than the regular list
410 pending_bios != &device->pending_sync_bios &&
411 device->pending_sync_bios.head) ||
412 (num_run > 64 && pending_bios == &device->pending_sync_bios &&
413 device->pending_bios.head)) {
414 spin_lock(&device->io_lock);
415 requeue_list(pending_bios, pending, tail);
420 pending = pending->bi_next;
424 * atomic_dec_return implies a barrier for waitqueue_active
426 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
427 waitqueue_active(&fs_info->async_submit_wait))
428 wake_up(&fs_info->async_submit_wait);
430 BUG_ON(atomic_read(&cur->__bi_cnt) == 0);
433 * if we're doing the sync list, record that our
434 * plug has some sync requests on it
436 * If we're doing the regular list and there are
437 * sync requests sitting around, unplug before
440 if (pending_bios == &device->pending_sync_bios) {
442 } else if (sync_pending) {
443 blk_finish_plug(&plug);
444 blk_start_plug(&plug);
448 btrfsic_submit_bio(cur->bi_rw, cur);
455 * we made progress, there is more work to do and the bdi
456 * is now congested. Back off and let other work structs
459 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
460 fs_info->fs_devices->open_devices > 1) {
461 struct io_context *ioc;
463 ioc = current->io_context;
466 * the main goal here is that we don't want to
467 * block if we're going to be able to submit
468 * more requests without blocking.
470 * This code does two great things, it pokes into
471 * the elevator code from a filesystem _and_
472 * it makes assumptions about how batching works.
474 if (ioc && ioc->nr_batch_requests > 0 &&
475 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
477 ioc->last_waited == last_waited)) {
479 * we want to go through our batch of
480 * requests and stop. So, we copy out
481 * the ioc->last_waited time and test
482 * against it before looping
484 last_waited = ioc->last_waited;
488 spin_lock(&device->io_lock);
489 requeue_list(pending_bios, pending, tail);
490 device->running_pending = 1;
492 spin_unlock(&device->io_lock);
493 btrfs_queue_work(fs_info->submit_workers,
497 /* unplug every 64 requests just for good measure */
498 if (batch_run % 64 == 0) {
499 blk_finish_plug(&plug);
500 blk_start_plug(&plug);
509 spin_lock(&device->io_lock);
510 if (device->pending_bios.head || device->pending_sync_bios.head)
512 spin_unlock(&device->io_lock);
515 blk_finish_plug(&plug);
518 static void pending_bios_fn(struct btrfs_work *work)
520 struct btrfs_device *device;
522 device = container_of(work, struct btrfs_device, work);
523 run_scheduled_bios(device);
527 void btrfs_free_stale_device(struct btrfs_device *cur_dev)
529 struct btrfs_fs_devices *fs_devs;
530 struct btrfs_device *dev;
535 list_for_each_entry(fs_devs, &fs_uuids, list) {
540 if (fs_devs->seeding)
543 list_for_each_entry(dev, &fs_devs->devices, dev_list) {
551 * Todo: This won't be enough. What if the same device
552 * comes back (with new uuid and) with its mapper path?
553 * But for now, this does help as mostly an admin will
554 * either use mapper or non mapper path throughout.
557 del = strcmp(rcu_str_deref(dev->name),
558 rcu_str_deref(cur_dev->name));
565 /* delete the stale device */
566 if (fs_devs->num_devices == 1) {
567 btrfs_sysfs_remove_fsid(fs_devs);
568 list_del(&fs_devs->list);
569 free_fs_devices(fs_devs);
571 fs_devs->num_devices--;
572 list_del(&dev->dev_list);
573 rcu_string_free(dev->name);
582 * Add new device to list of registered devices
585 * 1 - first time device is seen
586 * 0 - device already known
589 static noinline int device_list_add(const char *path,
590 struct btrfs_super_block *disk_super,
591 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
593 struct btrfs_device *device;
594 struct btrfs_fs_devices *fs_devices;
595 struct rcu_string *name;
597 u64 found_transid = btrfs_super_generation(disk_super);
599 fs_devices = find_fsid(disk_super->fsid);
601 fs_devices = alloc_fs_devices(disk_super->fsid);
602 if (IS_ERR(fs_devices))
603 return PTR_ERR(fs_devices);
605 list_add(&fs_devices->list, &fs_uuids);
609 device = __find_device(&fs_devices->devices, devid,
610 disk_super->dev_item.uuid);
614 if (fs_devices->opened)
617 device = btrfs_alloc_device(NULL, &devid,
618 disk_super->dev_item.uuid);
619 if (IS_ERR(device)) {
620 /* we can safely leave the fs_devices entry around */
621 return PTR_ERR(device);
624 name = rcu_string_strdup(path, GFP_NOFS);
629 rcu_assign_pointer(device->name, name);
631 mutex_lock(&fs_devices->device_list_mutex);
632 list_add_rcu(&device->dev_list, &fs_devices->devices);
633 fs_devices->num_devices++;
634 mutex_unlock(&fs_devices->device_list_mutex);
637 device->fs_devices = fs_devices;
638 } else if (!device->name || strcmp(device->name->str, path)) {
640 * When FS is already mounted.
641 * 1. If you are here and if the device->name is NULL that
642 * means this device was missing at time of FS mount.
643 * 2. If you are here and if the device->name is different
644 * from 'path' that means either
645 * a. The same device disappeared and reappeared with
647 * b. The missing-disk-which-was-replaced, has
650 * We must allow 1 and 2a above. But 2b would be a spurious
653 * Further in case of 1 and 2a above, the disk at 'path'
654 * would have missed some transaction when it was away and
655 * in case of 2a the stale bdev has to be updated as well.
656 * 2b must not be allowed at all time.
660 * For now, we do allow update to btrfs_fs_device through the
661 * btrfs dev scan cli after FS has been mounted. We're still
662 * tracking a problem where systems fail mount by subvolume id
663 * when we reject replacement on a mounted FS.
665 if (!fs_devices->opened && found_transid < device->generation) {
667 * That is if the FS is _not_ mounted and if you
668 * are here, that means there is more than one
669 * disk with same uuid and devid.We keep the one
670 * with larger generation number or the last-in if
671 * generation are equal.
676 name = rcu_string_strdup(path, GFP_NOFS);
679 rcu_string_free(device->name);
680 rcu_assign_pointer(device->name, name);
681 if (device->missing) {
682 fs_devices->missing_devices--;
688 * Unmount does not free the btrfs_device struct but would zero
689 * generation along with most of the other members. So just update
690 * it back. We need it to pick the disk with largest generation
693 if (!fs_devices->opened)
694 device->generation = found_transid;
697 * if there is new btrfs on an already registered device,
698 * then remove the stale device entry.
700 btrfs_free_stale_device(device);
702 *fs_devices_ret = fs_devices;
707 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
709 struct btrfs_fs_devices *fs_devices;
710 struct btrfs_device *device;
711 struct btrfs_device *orig_dev;
713 fs_devices = alloc_fs_devices(orig->fsid);
714 if (IS_ERR(fs_devices))
717 mutex_lock(&orig->device_list_mutex);
718 fs_devices->total_devices = orig->total_devices;
720 /* We have held the volume lock, it is safe to get the devices. */
721 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
722 struct rcu_string *name;
724 device = btrfs_alloc_device(NULL, &orig_dev->devid,
730 * This is ok to do without rcu read locked because we hold the
731 * uuid mutex so nothing we touch in here is going to disappear.
733 if (orig_dev->name) {
734 name = rcu_string_strdup(orig_dev->name->str, GFP_NOFS);
739 rcu_assign_pointer(device->name, name);
742 list_add(&device->dev_list, &fs_devices->devices);
743 device->fs_devices = fs_devices;
744 fs_devices->num_devices++;
746 mutex_unlock(&orig->device_list_mutex);
749 mutex_unlock(&orig->device_list_mutex);
750 free_fs_devices(fs_devices);
751 return ERR_PTR(-ENOMEM);
754 void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step)
756 struct btrfs_device *device, *next;
757 struct btrfs_device *latest_dev = NULL;
759 mutex_lock(&uuid_mutex);
761 /* This is the initialized path, it is safe to release the devices. */
762 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
763 if (device->in_fs_metadata) {
764 if (!device->is_tgtdev_for_dev_replace &&
766 device->generation > latest_dev->generation)) {
772 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
774 * In the first step, keep the device which has
775 * the correct fsid and the devid that is used
776 * for the dev_replace procedure.
777 * In the second step, the dev_replace state is
778 * read from the device tree and it is known
779 * whether the procedure is really active or
780 * not, which means whether this device is
781 * used or whether it should be removed.
783 if (step == 0 || device->is_tgtdev_for_dev_replace) {
788 blkdev_put(device->bdev, device->mode);
790 fs_devices->open_devices--;
792 if (device->writeable) {
793 list_del_init(&device->dev_alloc_list);
794 device->writeable = 0;
795 if (!device->is_tgtdev_for_dev_replace)
796 fs_devices->rw_devices--;
798 list_del_init(&device->dev_list);
799 fs_devices->num_devices--;
800 rcu_string_free(device->name);
804 if (fs_devices->seed) {
805 fs_devices = fs_devices->seed;
809 fs_devices->latest_bdev = latest_dev->bdev;
811 mutex_unlock(&uuid_mutex);
814 static void __free_device(struct work_struct *work)
816 struct btrfs_device *device;
818 device = container_of(work, struct btrfs_device, rcu_work);
821 blkdev_put(device->bdev, device->mode);
823 rcu_string_free(device->name);
827 static void free_device(struct rcu_head *head)
829 struct btrfs_device *device;
831 device = container_of(head, struct btrfs_device, rcu);
833 INIT_WORK(&device->rcu_work, __free_device);
834 schedule_work(&device->rcu_work);
837 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
839 struct btrfs_device *device, *tmp;
841 if (--fs_devices->opened > 0)
844 mutex_lock(&fs_devices->device_list_mutex);
845 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
846 btrfs_close_one_device(device);
848 mutex_unlock(&fs_devices->device_list_mutex);
850 WARN_ON(fs_devices->open_devices);
851 WARN_ON(fs_devices->rw_devices);
852 fs_devices->opened = 0;
853 fs_devices->seeding = 0;
858 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
860 struct btrfs_fs_devices *seed_devices = NULL;
863 mutex_lock(&uuid_mutex);
864 ret = __btrfs_close_devices(fs_devices);
865 if (!fs_devices->opened) {
866 seed_devices = fs_devices->seed;
867 fs_devices->seed = NULL;
869 mutex_unlock(&uuid_mutex);
871 while (seed_devices) {
872 fs_devices = seed_devices;
873 seed_devices = fs_devices->seed;
874 __btrfs_close_devices(fs_devices);
875 free_fs_devices(fs_devices);
878 * Wait for rcu kworkers under __btrfs_close_devices
879 * to finish all blkdev_puts so device is really
880 * free when umount is done.
886 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
887 fmode_t flags, void *holder)
889 struct request_queue *q;
890 struct block_device *bdev;
891 struct list_head *head = &fs_devices->devices;
892 struct btrfs_device *device;
893 struct btrfs_device *latest_dev = NULL;
894 struct buffer_head *bh;
895 struct btrfs_super_block *disk_super;
902 list_for_each_entry(device, head, dev_list) {
908 /* Just open everything we can; ignore failures here */
909 if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
913 disk_super = (struct btrfs_super_block *)bh->b_data;
914 devid = btrfs_stack_device_id(&disk_super->dev_item);
915 if (devid != device->devid)
918 if (memcmp(device->uuid, disk_super->dev_item.uuid,
922 device->generation = btrfs_super_generation(disk_super);
924 device->generation > latest_dev->generation)
927 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
928 device->writeable = 0;
930 device->writeable = !bdev_read_only(bdev);
934 q = bdev_get_queue(bdev);
935 if (blk_queue_discard(q))
936 device->can_discard = 1;
939 device->in_fs_metadata = 0;
940 device->mode = flags;
942 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
943 fs_devices->rotating = 1;
945 fs_devices->open_devices++;
946 if (device->writeable &&
947 device->devid != BTRFS_DEV_REPLACE_DEVID) {
948 fs_devices->rw_devices++;
949 list_add(&device->dev_alloc_list,
950 &fs_devices->alloc_list);
957 blkdev_put(bdev, flags);
960 if (fs_devices->open_devices == 0) {
964 fs_devices->seeding = seeding;
965 fs_devices->opened = 1;
966 fs_devices->latest_bdev = latest_dev->bdev;
967 fs_devices->total_rw_bytes = 0;
972 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
973 fmode_t flags, void *holder)
977 mutex_lock(&uuid_mutex);
978 if (fs_devices->opened) {
979 fs_devices->opened++;
982 ret = __btrfs_open_devices(fs_devices, flags, holder);
984 mutex_unlock(&uuid_mutex);
989 * Look for a btrfs signature on a device. This may be called out of the mount path
990 * and we are not allowed to call set_blocksize during the scan. The superblock
991 * is read via pagecache
993 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
994 struct btrfs_fs_devices **fs_devices_ret)
996 struct btrfs_super_block *disk_super;
997 struct block_device *bdev;
1008 * we would like to check all the supers, but that would make
1009 * a btrfs mount succeed after a mkfs from a different FS.
1010 * So, we need to add a special mount option to scan for
1011 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1013 bytenr = btrfs_sb_offset(0);
1014 flags |= FMODE_EXCL;
1015 mutex_lock(&uuid_mutex);
1017 bdev = blkdev_get_by_path(path, flags, holder);
1020 ret = PTR_ERR(bdev);
1024 /* make sure our super fits in the device */
1025 if (bytenr + PAGE_CACHE_SIZE >= i_size_read(bdev->bd_inode))
1026 goto error_bdev_put;
1028 /* make sure our super fits in the page */
1029 if (sizeof(*disk_super) > PAGE_CACHE_SIZE)
1030 goto error_bdev_put;
1032 /* make sure our super doesn't straddle pages on disk */
1033 index = bytenr >> PAGE_CACHE_SHIFT;
1034 if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_CACHE_SHIFT != index)
1035 goto error_bdev_put;
1037 /* pull in the page with our super */
1038 page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
1041 if (IS_ERR_OR_NULL(page))
1042 goto error_bdev_put;
1046 /* align our pointer to the offset of the super block */
1047 disk_super = p + (bytenr & ~PAGE_CACHE_MASK);
1049 if (btrfs_super_bytenr(disk_super) != bytenr ||
1050 btrfs_super_magic(disk_super) != BTRFS_MAGIC)
1053 devid = btrfs_stack_device_id(&disk_super->dev_item);
1054 transid = btrfs_super_generation(disk_super);
1055 total_devices = btrfs_super_num_devices(disk_super);
1057 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
1059 if (disk_super->label[0]) {
1060 if (disk_super->label[BTRFS_LABEL_SIZE - 1])
1061 disk_super->label[BTRFS_LABEL_SIZE - 1] = '\0';
1062 printk(KERN_INFO "BTRFS: device label %s ", disk_super->label);
1064 printk(KERN_INFO "BTRFS: device fsid %pU ", disk_super->fsid);
1067 printk(KERN_CONT "devid %llu transid %llu %s\n", devid, transid, path);
1070 if (!ret && fs_devices_ret)
1071 (*fs_devices_ret)->total_devices = total_devices;
1075 page_cache_release(page);
1078 blkdev_put(bdev, flags);
1080 mutex_unlock(&uuid_mutex);
1084 /* helper to account the used device space in the range */
1085 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
1086 u64 end, u64 *length)
1088 struct btrfs_key key;
1089 struct btrfs_root *root = device->dev_root;
1090 struct btrfs_dev_extent *dev_extent;
1091 struct btrfs_path *path;
1095 struct extent_buffer *l;
1099 if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
1102 path = btrfs_alloc_path();
1107 key.objectid = device->devid;
1109 key.type = BTRFS_DEV_EXTENT_KEY;
1111 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1115 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1122 slot = path->slots[0];
1123 if (slot >= btrfs_header_nritems(l)) {
1124 ret = btrfs_next_leaf(root, path);
1132 btrfs_item_key_to_cpu(l, &key, slot);
1134 if (key.objectid < device->devid)
1137 if (key.objectid > device->devid)
1140 if (key.type != BTRFS_DEV_EXTENT_KEY)
1143 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1144 extent_end = key.offset + btrfs_dev_extent_length(l,
1146 if (key.offset <= start && extent_end > end) {
1147 *length = end - start + 1;
1149 } else if (key.offset <= start && extent_end > start)
1150 *length += extent_end - start;
1151 else if (key.offset > start && extent_end <= end)
1152 *length += extent_end - key.offset;
1153 else if (key.offset > start && key.offset <= end) {
1154 *length += end - key.offset + 1;
1156 } else if (key.offset > end)
1164 btrfs_free_path(path);
1168 static int contains_pending_extent(struct btrfs_transaction *transaction,
1169 struct btrfs_device *device,
1170 u64 *start, u64 len)
1172 struct btrfs_fs_info *fs_info = device->dev_root->fs_info;
1173 struct extent_map *em;
1174 struct list_head *search_list = &fs_info->pinned_chunks;
1176 u64 physical_start = *start;
1179 search_list = &transaction->pending_chunks;
1181 list_for_each_entry(em, search_list, list) {
1182 struct map_lookup *map;
1185 map = (struct map_lookup *)em->bdev;
1186 for (i = 0; i < map->num_stripes; i++) {
1189 if (map->stripes[i].dev != device)
1191 if (map->stripes[i].physical >= physical_start + len ||
1192 map->stripes[i].physical + em->orig_block_len <=
1196 * Make sure that while processing the pinned list we do
1197 * not override our *start with a lower value, because
1198 * we can have pinned chunks that fall within this
1199 * device hole and that have lower physical addresses
1200 * than the pending chunks we processed before. If we
1201 * do not take this special care we can end up getting
1202 * 2 pending chunks that start at the same physical
1203 * device offsets because the end offset of a pinned
1204 * chunk can be equal to the start offset of some
1207 end = map->stripes[i].physical + em->orig_block_len;
1214 if (search_list != &fs_info->pinned_chunks) {
1215 search_list = &fs_info->pinned_chunks;
1224 * find_free_dev_extent_start - find free space in the specified device
1225 * @device: the device which we search the free space in
1226 * @num_bytes: the size of the free space that we need
1227 * @search_start: the position from which to begin the search
1228 * @start: store the start of the free space.
1229 * @len: the size of the free space. that we find, or the size
1230 * of the max free space if we don't find suitable free space
1232 * this uses a pretty simple search, the expectation is that it is
1233 * called very infrequently and that a given device has a small number
1236 * @start is used to store the start of the free space if we find. But if we
1237 * don't find suitable free space, it will be used to store the start position
1238 * of the max free space.
1240 * @len is used to store the size of the free space that we find.
1241 * But if we don't find suitable free space, it is used to store the size of
1242 * the max free space.
1244 int find_free_dev_extent_start(struct btrfs_transaction *transaction,
1245 struct btrfs_device *device, u64 num_bytes,
1246 u64 search_start, u64 *start, u64 *len)
1248 struct btrfs_key key;
1249 struct btrfs_root *root = device->dev_root;
1250 struct btrfs_dev_extent *dev_extent;
1251 struct btrfs_path *path;
1256 u64 search_end = device->total_bytes;
1259 struct extent_buffer *l;
1261 path = btrfs_alloc_path();
1265 max_hole_start = search_start;
1269 if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1275 path->search_commit_root = 1;
1276 path->skip_locking = 1;
1278 key.objectid = device->devid;
1279 key.offset = search_start;
1280 key.type = BTRFS_DEV_EXTENT_KEY;
1282 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1286 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1293 slot = path->slots[0];
1294 if (slot >= btrfs_header_nritems(l)) {
1295 ret = btrfs_next_leaf(root, path);
1303 btrfs_item_key_to_cpu(l, &key, slot);
1305 if (key.objectid < device->devid)
1308 if (key.objectid > device->devid)
1311 if (key.type != BTRFS_DEV_EXTENT_KEY)
1314 if (key.offset > search_start) {
1315 hole_size = key.offset - search_start;
1318 * Have to check before we set max_hole_start, otherwise
1319 * we could end up sending back this offset anyway.
1321 if (contains_pending_extent(transaction, device,
1324 if (key.offset >= search_start) {
1325 hole_size = key.offset - search_start;
1332 if (hole_size > max_hole_size) {
1333 max_hole_start = search_start;
1334 max_hole_size = hole_size;
1338 * If this free space is greater than which we need,
1339 * it must be the max free space that we have found
1340 * until now, so max_hole_start must point to the start
1341 * of this free space and the length of this free space
1342 * is stored in max_hole_size. Thus, we return
1343 * max_hole_start and max_hole_size and go back to the
1346 if (hole_size >= num_bytes) {
1352 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1353 extent_end = key.offset + btrfs_dev_extent_length(l,
1355 if (extent_end > search_start)
1356 search_start = extent_end;
1363 * At this point, search_start should be the end of
1364 * allocated dev extents, and when shrinking the device,
1365 * search_end may be smaller than search_start.
1367 if (search_end > search_start) {
1368 hole_size = search_end - search_start;
1370 if (contains_pending_extent(transaction, device, &search_start,
1372 btrfs_release_path(path);
1376 if (hole_size > max_hole_size) {
1377 max_hole_start = search_start;
1378 max_hole_size = hole_size;
1383 if (max_hole_size < num_bytes)
1389 btrfs_free_path(path);
1390 *start = max_hole_start;
1392 *len = max_hole_size;
1396 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1397 struct btrfs_device *device, u64 num_bytes,
1398 u64 *start, u64 *len)
1400 struct btrfs_root *root = device->dev_root;
1403 /* FIXME use last free of some kind */
1406 * we don't want to overwrite the superblock on the drive,
1407 * so we make sure to start at an offset of at least 1MB
1409 search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
1410 return find_free_dev_extent_start(trans->transaction, device,
1411 num_bytes, search_start, start, len);
1414 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1415 struct btrfs_device *device,
1416 u64 start, u64 *dev_extent_len)
1419 struct btrfs_path *path;
1420 struct btrfs_root *root = device->dev_root;
1421 struct btrfs_key key;
1422 struct btrfs_key found_key;
1423 struct extent_buffer *leaf = NULL;
1424 struct btrfs_dev_extent *extent = NULL;
1426 path = btrfs_alloc_path();
1430 key.objectid = device->devid;
1432 key.type = BTRFS_DEV_EXTENT_KEY;
1434 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1436 ret = btrfs_previous_item(root, path, key.objectid,
1437 BTRFS_DEV_EXTENT_KEY);
1440 leaf = path->nodes[0];
1441 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1442 extent = btrfs_item_ptr(leaf, path->slots[0],
1443 struct btrfs_dev_extent);
1444 BUG_ON(found_key.offset > start || found_key.offset +
1445 btrfs_dev_extent_length(leaf, extent) < start);
1447 btrfs_release_path(path);
1449 } else if (ret == 0) {
1450 leaf = path->nodes[0];
1451 extent = btrfs_item_ptr(leaf, path->slots[0],
1452 struct btrfs_dev_extent);
1454 btrfs_std_error(root->fs_info, ret, "Slot search failed");
1458 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1460 ret = btrfs_del_item(trans, root, path);
1462 btrfs_std_error(root->fs_info, ret,
1463 "Failed to remove dev extent item");
1465 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1468 btrfs_free_path(path);
1472 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1473 struct btrfs_device *device,
1474 u64 chunk_tree, u64 chunk_objectid,
1475 u64 chunk_offset, u64 start, u64 num_bytes)
1478 struct btrfs_path *path;
1479 struct btrfs_root *root = device->dev_root;
1480 struct btrfs_dev_extent *extent;
1481 struct extent_buffer *leaf;
1482 struct btrfs_key key;
1484 WARN_ON(!device->in_fs_metadata);
1485 WARN_ON(device->is_tgtdev_for_dev_replace);
1486 path = btrfs_alloc_path();
1490 key.objectid = device->devid;
1492 key.type = BTRFS_DEV_EXTENT_KEY;
1493 ret = btrfs_insert_empty_item(trans, root, path, &key,
1498 leaf = path->nodes[0];
1499 extent = btrfs_item_ptr(leaf, path->slots[0],
1500 struct btrfs_dev_extent);
1501 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1502 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1503 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1505 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1506 btrfs_dev_extent_chunk_tree_uuid(extent), BTRFS_UUID_SIZE);
1508 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1509 btrfs_mark_buffer_dirty(leaf);
1511 btrfs_free_path(path);
1515 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1517 struct extent_map_tree *em_tree;
1518 struct extent_map *em;
1522 em_tree = &fs_info->mapping_tree.map_tree;
1523 read_lock(&em_tree->lock);
1524 n = rb_last(&em_tree->map);
1526 em = rb_entry(n, struct extent_map, rb_node);
1527 ret = em->start + em->len;
1529 read_unlock(&em_tree->lock);
1534 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1538 struct btrfs_key key;
1539 struct btrfs_key found_key;
1540 struct btrfs_path *path;
1542 path = btrfs_alloc_path();
1546 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1547 key.type = BTRFS_DEV_ITEM_KEY;
1548 key.offset = (u64)-1;
1550 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1554 BUG_ON(ret == 0); /* Corruption */
1556 ret = btrfs_previous_item(fs_info->chunk_root, path,
1557 BTRFS_DEV_ITEMS_OBJECTID,
1558 BTRFS_DEV_ITEM_KEY);
1562 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1564 *devid_ret = found_key.offset + 1;
1568 btrfs_free_path(path);
1573 * the device information is stored in the chunk root
1574 * the btrfs_device struct should be fully filled in
1576 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1577 struct btrfs_root *root,
1578 struct btrfs_device *device)
1581 struct btrfs_path *path;
1582 struct btrfs_dev_item *dev_item;
1583 struct extent_buffer *leaf;
1584 struct btrfs_key key;
1587 root = root->fs_info->chunk_root;
1589 path = btrfs_alloc_path();
1593 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1594 key.type = BTRFS_DEV_ITEM_KEY;
1595 key.offset = device->devid;
1597 ret = btrfs_insert_empty_item(trans, root, path, &key,
1602 leaf = path->nodes[0];
1603 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1605 btrfs_set_device_id(leaf, dev_item, device->devid);
1606 btrfs_set_device_generation(leaf, dev_item, 0);
1607 btrfs_set_device_type(leaf, dev_item, device->type);
1608 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1609 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1610 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1611 btrfs_set_device_total_bytes(leaf, dev_item,
1612 btrfs_device_get_disk_total_bytes(device));
1613 btrfs_set_device_bytes_used(leaf, dev_item,
1614 btrfs_device_get_bytes_used(device));
1615 btrfs_set_device_group(leaf, dev_item, 0);
1616 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1617 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1618 btrfs_set_device_start_offset(leaf, dev_item, 0);
1620 ptr = btrfs_device_uuid(dev_item);
1621 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1622 ptr = btrfs_device_fsid(dev_item);
1623 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1624 btrfs_mark_buffer_dirty(leaf);
1628 btrfs_free_path(path);
1633 * Function to update ctime/mtime for a given device path.
1634 * Mainly used for ctime/mtime based probe like libblkid.
1636 static void update_dev_time(char *path_name)
1640 filp = filp_open(path_name, O_RDWR, 0);
1643 file_update_time(filp);
1644 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,
1977 rcu_str_deref(srcdev->name));
1981 fs_devices->open_devices--;
1984 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
1985 struct btrfs_device *srcdev)
1987 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
1989 call_rcu(&srcdev->rcu, free_device);
1992 * unless fs_devices is seed fs, num_devices shouldn't go
1995 BUG_ON(!fs_devices->num_devices && !fs_devices->seeding);
1997 /* if this is no devs we rather delete the fs_devices */
1998 if (!fs_devices->num_devices) {
1999 struct btrfs_fs_devices *tmp_fs_devices;
2001 tmp_fs_devices = fs_info->fs_devices;
2002 while (tmp_fs_devices) {
2003 if (tmp_fs_devices->seed == fs_devices) {
2004 tmp_fs_devices->seed = fs_devices->seed;
2007 tmp_fs_devices = tmp_fs_devices->seed;
2009 fs_devices->seed = NULL;
2010 __btrfs_close_devices(fs_devices);
2011 free_fs_devices(fs_devices);
2015 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2016 struct btrfs_device *tgtdev)
2018 struct btrfs_device *next_device;
2020 mutex_lock(&uuid_mutex);
2022 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2024 btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev);
2027 btrfs_scratch_superblocks(tgtdev->bdev,
2028 rcu_str_deref(tgtdev->name));
2029 fs_info->fs_devices->open_devices--;
2031 fs_info->fs_devices->num_devices--;
2033 next_device = list_entry(fs_info->fs_devices->devices.next,
2034 struct btrfs_device, dev_list);
2035 if (tgtdev->bdev == fs_info->sb->s_bdev)
2036 fs_info->sb->s_bdev = next_device->bdev;
2037 if (tgtdev->bdev == fs_info->fs_devices->latest_bdev)
2038 fs_info->fs_devices->latest_bdev = next_device->bdev;
2039 list_del_rcu(&tgtdev->dev_list);
2041 call_rcu(&tgtdev->rcu, free_device);
2043 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2044 mutex_unlock(&uuid_mutex);
2047 static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
2048 struct btrfs_device **device)
2051 struct btrfs_super_block *disk_super;
2054 struct block_device *bdev;
2055 struct buffer_head *bh;
2058 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2059 root->fs_info->bdev_holder, 0, &bdev, &bh);
2062 disk_super = (struct btrfs_super_block *)bh->b_data;
2063 devid = btrfs_stack_device_id(&disk_super->dev_item);
2064 dev_uuid = disk_super->dev_item.uuid;
2065 *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2070 blkdev_put(bdev, FMODE_READ);
2074 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
2076 struct btrfs_device **device)
2079 if (strcmp(device_path, "missing") == 0) {
2080 struct list_head *devices;
2081 struct btrfs_device *tmp;
2083 devices = &root->fs_info->fs_devices->devices;
2085 * It is safe to read the devices since the volume_mutex
2086 * is held by the caller.
2088 list_for_each_entry(tmp, devices, dev_list) {
2089 if (tmp->in_fs_metadata && !tmp->bdev) {
2096 return BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2100 return btrfs_find_device_by_path(root, device_path, device);
2105 * does all the dirty work required for changing file system's UUID.
2107 static int btrfs_prepare_sprout(struct btrfs_root *root)
2109 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2110 struct btrfs_fs_devices *old_devices;
2111 struct btrfs_fs_devices *seed_devices;
2112 struct btrfs_super_block *disk_super = root->fs_info->super_copy;
2113 struct btrfs_device *device;
2116 BUG_ON(!mutex_is_locked(&uuid_mutex));
2117 if (!fs_devices->seeding)
2120 seed_devices = __alloc_fs_devices();
2121 if (IS_ERR(seed_devices))
2122 return PTR_ERR(seed_devices);
2124 old_devices = clone_fs_devices(fs_devices);
2125 if (IS_ERR(old_devices)) {
2126 kfree(seed_devices);
2127 return PTR_ERR(old_devices);
2130 list_add(&old_devices->list, &fs_uuids);
2132 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2133 seed_devices->opened = 1;
2134 INIT_LIST_HEAD(&seed_devices->devices);
2135 INIT_LIST_HEAD(&seed_devices->alloc_list);
2136 mutex_init(&seed_devices->device_list_mutex);
2138 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2139 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2141 list_for_each_entry(device, &seed_devices->devices, dev_list)
2142 device->fs_devices = seed_devices;
2145 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2146 unlock_chunks(root);
2148 fs_devices->seeding = 0;
2149 fs_devices->num_devices = 0;
2150 fs_devices->open_devices = 0;
2151 fs_devices->missing_devices = 0;
2152 fs_devices->rotating = 0;
2153 fs_devices->seed = seed_devices;
2155 generate_random_uuid(fs_devices->fsid);
2156 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2157 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2158 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2160 super_flags = btrfs_super_flags(disk_super) &
2161 ~BTRFS_SUPER_FLAG_SEEDING;
2162 btrfs_set_super_flags(disk_super, super_flags);
2168 * strore the expected generation for seed devices in device items.
2170 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2171 struct btrfs_root *root)
2173 struct btrfs_path *path;
2174 struct extent_buffer *leaf;
2175 struct btrfs_dev_item *dev_item;
2176 struct btrfs_device *device;
2177 struct btrfs_key key;
2178 u8 fs_uuid[BTRFS_UUID_SIZE];
2179 u8 dev_uuid[BTRFS_UUID_SIZE];
2183 path = btrfs_alloc_path();
2187 root = root->fs_info->chunk_root;
2188 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2190 key.type = BTRFS_DEV_ITEM_KEY;
2193 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2197 leaf = path->nodes[0];
2199 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2200 ret = btrfs_next_leaf(root, path);
2205 leaf = path->nodes[0];
2206 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2207 btrfs_release_path(path);
2211 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2212 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2213 key.type != BTRFS_DEV_ITEM_KEY)
2216 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2217 struct btrfs_dev_item);
2218 devid = btrfs_device_id(leaf, dev_item);
2219 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2221 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2223 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2225 BUG_ON(!device); /* Logic error */
2227 if (device->fs_devices->seeding) {
2228 btrfs_set_device_generation(leaf, dev_item,
2229 device->generation);
2230 btrfs_mark_buffer_dirty(leaf);
2238 btrfs_free_path(path);
2242 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
2244 struct request_queue *q;
2245 struct btrfs_trans_handle *trans;
2246 struct btrfs_device *device;
2247 struct block_device *bdev;
2248 struct list_head *devices;
2249 struct super_block *sb = root->fs_info->sb;
2250 struct rcu_string *name;
2252 int seeding_dev = 0;
2255 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
2258 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2259 root->fs_info->bdev_holder);
2261 return PTR_ERR(bdev);
2263 if (root->fs_info->fs_devices->seeding) {
2265 down_write(&sb->s_umount);
2266 mutex_lock(&uuid_mutex);
2269 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2271 devices = &root->fs_info->fs_devices->devices;
2273 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2274 list_for_each_entry(device, devices, dev_list) {
2275 if (device->bdev == bdev) {
2278 &root->fs_info->fs_devices->device_list_mutex);
2282 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2284 device = btrfs_alloc_device(root->fs_info, NULL, NULL);
2285 if (IS_ERR(device)) {
2286 /* we can safely leave the fs_devices entry around */
2287 ret = PTR_ERR(device);
2291 name = rcu_string_strdup(device_path, GFP_NOFS);
2297 rcu_assign_pointer(device->name, name);
2299 trans = btrfs_start_transaction(root, 0);
2300 if (IS_ERR(trans)) {
2301 rcu_string_free(device->name);
2303 ret = PTR_ERR(trans);
2307 q = bdev_get_queue(bdev);
2308 if (blk_queue_discard(q))
2309 device->can_discard = 1;
2310 device->writeable = 1;
2311 device->generation = trans->transid;
2312 device->io_width = root->sectorsize;
2313 device->io_align = root->sectorsize;
2314 device->sector_size = root->sectorsize;
2315 device->total_bytes = i_size_read(bdev->bd_inode);
2316 device->disk_total_bytes = device->total_bytes;
2317 device->commit_total_bytes = device->total_bytes;
2318 device->dev_root = root->fs_info->dev_root;
2319 device->bdev = bdev;
2320 device->in_fs_metadata = 1;
2321 device->is_tgtdev_for_dev_replace = 0;
2322 device->mode = FMODE_EXCL;
2323 device->dev_stats_valid = 1;
2324 set_blocksize(device->bdev, 4096);
2327 sb->s_flags &= ~MS_RDONLY;
2328 ret = btrfs_prepare_sprout(root);
2329 BUG_ON(ret); /* -ENOMEM */
2332 device->fs_devices = root->fs_info->fs_devices;
2334 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2336 list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
2337 list_add(&device->dev_alloc_list,
2338 &root->fs_info->fs_devices->alloc_list);
2339 root->fs_info->fs_devices->num_devices++;
2340 root->fs_info->fs_devices->open_devices++;
2341 root->fs_info->fs_devices->rw_devices++;
2342 root->fs_info->fs_devices->total_devices++;
2343 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2345 spin_lock(&root->fs_info->free_chunk_lock);
2346 root->fs_info->free_chunk_space += device->total_bytes;
2347 spin_unlock(&root->fs_info->free_chunk_lock);
2349 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2350 root->fs_info->fs_devices->rotating = 1;
2352 tmp = btrfs_super_total_bytes(root->fs_info->super_copy);
2353 btrfs_set_super_total_bytes(root->fs_info->super_copy,
2354 tmp + device->total_bytes);
2356 tmp = btrfs_super_num_devices(root->fs_info->super_copy);
2357 btrfs_set_super_num_devices(root->fs_info->super_copy,
2360 /* add sysfs device entry */
2361 btrfs_sysfs_add_device_link(root->fs_info->fs_devices, device);
2364 * we've got more storage, clear any full flags on the space
2367 btrfs_clear_space_info_full(root->fs_info);
2369 unlock_chunks(root);
2370 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2374 ret = init_first_rw_device(trans, root, device);
2375 unlock_chunks(root);
2377 btrfs_abort_transaction(trans, root, ret);
2382 ret = btrfs_add_device(trans, root, device);
2384 btrfs_abort_transaction(trans, root, ret);
2389 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2391 ret = btrfs_finish_sprout(trans, root);
2393 btrfs_abort_transaction(trans, root, ret);
2397 /* Sprouting would change fsid of the mounted root,
2398 * so rename the fsid on the sysfs
2400 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2401 root->fs_info->fsid);
2402 if (kobject_rename(&root->fs_info->fs_devices->fsid_kobj,
2404 btrfs_warn(root->fs_info,
2405 "sysfs: failed to create fsid for sprout");
2408 root->fs_info->num_tolerated_disk_barrier_failures =
2409 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
2410 ret = btrfs_commit_transaction(trans, root);
2413 mutex_unlock(&uuid_mutex);
2414 up_write(&sb->s_umount);
2416 if (ret) /* transaction commit */
2419 ret = btrfs_relocate_sys_chunks(root);
2421 btrfs_std_error(root->fs_info, ret,
2422 "Failed to relocate sys chunks after "
2423 "device initialization. This can be fixed "
2424 "using the \"btrfs balance\" command.");
2425 trans = btrfs_attach_transaction(root);
2426 if (IS_ERR(trans)) {
2427 if (PTR_ERR(trans) == -ENOENT)
2429 return PTR_ERR(trans);
2431 ret = btrfs_commit_transaction(trans, root);
2434 /* Update ctime/mtime for libblkid */
2435 update_dev_time(device_path);
2439 btrfs_end_transaction(trans, root);
2440 rcu_string_free(device->name);
2441 btrfs_sysfs_rm_device_link(root->fs_info->fs_devices, device);
2444 blkdev_put(bdev, FMODE_EXCL);
2446 mutex_unlock(&uuid_mutex);
2447 up_write(&sb->s_umount);
2452 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2453 struct btrfs_device *srcdev,
2454 struct btrfs_device **device_out)
2456 struct request_queue *q;
2457 struct btrfs_device *device;
2458 struct block_device *bdev;
2459 struct btrfs_fs_info *fs_info = root->fs_info;
2460 struct list_head *devices;
2461 struct rcu_string *name;
2462 u64 devid = BTRFS_DEV_REPLACE_DEVID;
2466 if (fs_info->fs_devices->seeding) {
2467 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
2471 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2472 fs_info->bdev_holder);
2474 btrfs_err(fs_info, "target device %s is invalid!", device_path);
2475 return PTR_ERR(bdev);
2478 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2480 devices = &fs_info->fs_devices->devices;
2481 list_for_each_entry(device, devices, dev_list) {
2482 if (device->bdev == bdev) {
2483 btrfs_err(fs_info, "target device is in the filesystem!");
2490 if (i_size_read(bdev->bd_inode) <
2491 btrfs_device_get_total_bytes(srcdev)) {
2492 btrfs_err(fs_info, "target device is smaller than source device!");
2498 device = btrfs_alloc_device(NULL, &devid, NULL);
2499 if (IS_ERR(device)) {
2500 ret = PTR_ERR(device);
2504 name = rcu_string_strdup(device_path, GFP_NOFS);
2510 rcu_assign_pointer(device->name, name);
2512 q = bdev_get_queue(bdev);
2513 if (blk_queue_discard(q))
2514 device->can_discard = 1;
2515 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2516 device->writeable = 1;
2517 device->generation = 0;
2518 device->io_width = root->sectorsize;
2519 device->io_align = root->sectorsize;
2520 device->sector_size = root->sectorsize;
2521 device->total_bytes = btrfs_device_get_total_bytes(srcdev);
2522 device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
2523 device->bytes_used = btrfs_device_get_bytes_used(srcdev);
2524 ASSERT(list_empty(&srcdev->resized_list));
2525 device->commit_total_bytes = srcdev->commit_total_bytes;
2526 device->commit_bytes_used = device->bytes_used;
2527 device->dev_root = fs_info->dev_root;
2528 device->bdev = bdev;
2529 device->in_fs_metadata = 1;
2530 device->is_tgtdev_for_dev_replace = 1;
2531 device->mode = FMODE_EXCL;
2532 device->dev_stats_valid = 1;
2533 set_blocksize(device->bdev, 4096);
2534 device->fs_devices = fs_info->fs_devices;
2535 list_add(&device->dev_list, &fs_info->fs_devices->devices);
2536 fs_info->fs_devices->num_devices++;
2537 fs_info->fs_devices->open_devices++;
2538 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2540 *device_out = device;
2544 blkdev_put(bdev, FMODE_EXCL);
2548 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2549 struct btrfs_device *tgtdev)
2551 WARN_ON(fs_info->fs_devices->rw_devices == 0);
2552 tgtdev->io_width = fs_info->dev_root->sectorsize;
2553 tgtdev->io_align = fs_info->dev_root->sectorsize;
2554 tgtdev->sector_size = fs_info->dev_root->sectorsize;
2555 tgtdev->dev_root = fs_info->dev_root;
2556 tgtdev->in_fs_metadata = 1;
2559 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2560 struct btrfs_device *device)
2563 struct btrfs_path *path;
2564 struct btrfs_root *root;
2565 struct btrfs_dev_item *dev_item;
2566 struct extent_buffer *leaf;
2567 struct btrfs_key key;
2569 root = device->dev_root->fs_info->chunk_root;
2571 path = btrfs_alloc_path();
2575 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2576 key.type = BTRFS_DEV_ITEM_KEY;
2577 key.offset = device->devid;
2579 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2588 leaf = path->nodes[0];
2589 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2591 btrfs_set_device_id(leaf, dev_item, device->devid);
2592 btrfs_set_device_type(leaf, dev_item, device->type);
2593 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2594 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2595 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2596 btrfs_set_device_total_bytes(leaf, dev_item,
2597 btrfs_device_get_disk_total_bytes(device));
2598 btrfs_set_device_bytes_used(leaf, dev_item,
2599 btrfs_device_get_bytes_used(device));
2600 btrfs_mark_buffer_dirty(leaf);
2603 btrfs_free_path(path);
2607 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2608 struct btrfs_device *device, u64 new_size)
2610 struct btrfs_super_block *super_copy =
2611 device->dev_root->fs_info->super_copy;
2612 struct btrfs_fs_devices *fs_devices;
2616 if (!device->writeable)
2619 lock_chunks(device->dev_root);
2620 old_total = btrfs_super_total_bytes(super_copy);
2621 diff = new_size - device->total_bytes;
2623 if (new_size <= device->total_bytes ||
2624 device->is_tgtdev_for_dev_replace) {
2625 unlock_chunks(device->dev_root);
2629 fs_devices = device->dev_root->fs_info->fs_devices;
2631 btrfs_set_super_total_bytes(super_copy, old_total + diff);
2632 device->fs_devices->total_rw_bytes += diff;
2634 btrfs_device_set_total_bytes(device, new_size);
2635 btrfs_device_set_disk_total_bytes(device, new_size);
2636 btrfs_clear_space_info_full(device->dev_root->fs_info);
2637 if (list_empty(&device->resized_list))
2638 list_add_tail(&device->resized_list,
2639 &fs_devices->resized_devices);
2640 unlock_chunks(device->dev_root);
2642 return btrfs_update_device(trans, device);
2645 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2646 struct btrfs_root *root, u64 chunk_objectid,
2650 struct btrfs_path *path;
2651 struct btrfs_key key;
2653 root = root->fs_info->chunk_root;
2654 path = btrfs_alloc_path();
2658 key.objectid = chunk_objectid;
2659 key.offset = chunk_offset;
2660 key.type = BTRFS_CHUNK_ITEM_KEY;
2662 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2665 else if (ret > 0) { /* Logic error or corruption */
2666 btrfs_std_error(root->fs_info, -ENOENT,
2667 "Failed lookup while freeing chunk.");
2672 ret = btrfs_del_item(trans, root, path);
2674 btrfs_std_error(root->fs_info, ret,
2675 "Failed to delete chunk item.");
2677 btrfs_free_path(path);
2681 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2684 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2685 struct btrfs_disk_key *disk_key;
2686 struct btrfs_chunk *chunk;
2693 struct btrfs_key key;
2696 array_size = btrfs_super_sys_array_size(super_copy);
2698 ptr = super_copy->sys_chunk_array;
2701 while (cur < array_size) {
2702 disk_key = (struct btrfs_disk_key *)ptr;
2703 btrfs_disk_key_to_cpu(&key, disk_key);
2705 len = sizeof(*disk_key);
2707 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2708 chunk = (struct btrfs_chunk *)(ptr + len);
2709 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2710 len += btrfs_chunk_item_size(num_stripes);
2715 if (key.objectid == chunk_objectid &&
2716 key.offset == chunk_offset) {
2717 memmove(ptr, ptr + len, array_size - (cur + len));
2719 btrfs_set_super_sys_array_size(super_copy, array_size);
2725 unlock_chunks(root);
2729 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2730 struct btrfs_root *root, u64 chunk_offset)
2732 struct extent_map_tree *em_tree;
2733 struct extent_map *em;
2734 struct btrfs_root *extent_root = root->fs_info->extent_root;
2735 struct map_lookup *map;
2736 u64 dev_extent_len = 0;
2737 u64 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2741 root = root->fs_info->chunk_root;
2742 em_tree = &root->fs_info->mapping_tree.map_tree;
2744 read_lock(&em_tree->lock);
2745 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2746 read_unlock(&em_tree->lock);
2748 if (!em || em->start > chunk_offset ||
2749 em->start + em->len < chunk_offset) {
2751 * This is a logic error, but we don't want to just rely on the
2752 * user having built with ASSERT enabled, so if ASSERT doens't
2753 * do anything we still error out.
2757 free_extent_map(em);
2760 map = (struct map_lookup *)em->bdev;
2761 lock_chunks(root->fs_info->chunk_root);
2762 check_system_chunk(trans, extent_root, map->type);
2763 unlock_chunks(root->fs_info->chunk_root);
2765 for (i = 0; i < map->num_stripes; i++) {
2766 struct btrfs_device *device = map->stripes[i].dev;
2767 ret = btrfs_free_dev_extent(trans, device,
2768 map->stripes[i].physical,
2771 btrfs_abort_transaction(trans, root, ret);
2775 if (device->bytes_used > 0) {
2777 btrfs_device_set_bytes_used(device,
2778 device->bytes_used - dev_extent_len);
2779 spin_lock(&root->fs_info->free_chunk_lock);
2780 root->fs_info->free_chunk_space += dev_extent_len;
2781 spin_unlock(&root->fs_info->free_chunk_lock);
2782 btrfs_clear_space_info_full(root->fs_info);
2783 unlock_chunks(root);
2786 if (map->stripes[i].dev) {
2787 ret = btrfs_update_device(trans, map->stripes[i].dev);
2789 btrfs_abort_transaction(trans, root, ret);
2794 ret = btrfs_free_chunk(trans, root, chunk_objectid, chunk_offset);
2796 btrfs_abort_transaction(trans, root, ret);
2800 trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2802 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2803 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2805 btrfs_abort_transaction(trans, root, ret);
2810 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset, em);
2812 btrfs_abort_transaction(trans, extent_root, ret);
2818 free_extent_map(em);
2822 static int btrfs_relocate_chunk(struct btrfs_root *root, u64 chunk_offset)
2824 struct btrfs_root *extent_root;
2825 struct btrfs_trans_handle *trans;
2828 root = root->fs_info->chunk_root;
2829 extent_root = root->fs_info->extent_root;
2832 * Prevent races with automatic removal of unused block groups.
2833 * After we relocate and before we remove the chunk with offset
2834 * chunk_offset, automatic removal of the block group can kick in,
2835 * resulting in a failure when calling btrfs_remove_chunk() below.
2837 * Make sure to acquire this mutex before doing a tree search (dev
2838 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2839 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2840 * we release the path used to search the chunk/dev tree and before
2841 * the current task acquires this mutex and calls us.
2843 ASSERT(mutex_is_locked(&root->fs_info->delete_unused_bgs_mutex));
2845 ret = btrfs_can_relocate(extent_root, chunk_offset);
2849 /* step one, relocate all the extents inside this chunk */
2850 btrfs_scrub_pause(root);
2851 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2852 btrfs_scrub_continue(root);
2856 trans = btrfs_start_transaction(root, 0);
2857 if (IS_ERR(trans)) {
2858 ret = PTR_ERR(trans);
2859 btrfs_std_error(root->fs_info, ret, NULL);
2864 * step two, delete the device extents and the
2865 * chunk tree entries
2867 ret = btrfs_remove_chunk(trans, root, chunk_offset);
2868 btrfs_end_transaction(trans, root);
2872 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2874 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2875 struct btrfs_path *path;
2876 struct extent_buffer *leaf;
2877 struct btrfs_chunk *chunk;
2878 struct btrfs_key key;
2879 struct btrfs_key found_key;
2881 bool retried = false;
2885 path = btrfs_alloc_path();
2890 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2891 key.offset = (u64)-1;
2892 key.type = BTRFS_CHUNK_ITEM_KEY;
2895 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
2896 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2898 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2901 BUG_ON(ret == 0); /* Corruption */
2903 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2906 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2912 leaf = path->nodes[0];
2913 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2915 chunk = btrfs_item_ptr(leaf, path->slots[0],
2916 struct btrfs_chunk);
2917 chunk_type = btrfs_chunk_type(leaf, chunk);
2918 btrfs_release_path(path);
2920 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2921 ret = btrfs_relocate_chunk(chunk_root,
2928 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2930 if (found_key.offset == 0)
2932 key.offset = found_key.offset - 1;
2935 if (failed && !retried) {
2939 } else if (WARN_ON(failed && retried)) {
2943 btrfs_free_path(path);
2947 static int insert_balance_item(struct btrfs_root *root,
2948 struct btrfs_balance_control *bctl)
2950 struct btrfs_trans_handle *trans;
2951 struct btrfs_balance_item *item;
2952 struct btrfs_disk_balance_args disk_bargs;
2953 struct btrfs_path *path;
2954 struct extent_buffer *leaf;
2955 struct btrfs_key key;
2958 path = btrfs_alloc_path();
2962 trans = btrfs_start_transaction(root, 0);
2963 if (IS_ERR(trans)) {
2964 btrfs_free_path(path);
2965 return PTR_ERR(trans);
2968 key.objectid = BTRFS_BALANCE_OBJECTID;
2969 key.type = BTRFS_BALANCE_ITEM_KEY;
2972 ret = btrfs_insert_empty_item(trans, root, path, &key,
2977 leaf = path->nodes[0];
2978 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2980 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2982 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2983 btrfs_set_balance_data(leaf, item, &disk_bargs);
2984 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2985 btrfs_set_balance_meta(leaf, item, &disk_bargs);
2986 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2987 btrfs_set_balance_sys(leaf, item, &disk_bargs);
2989 btrfs_set_balance_flags(leaf, item, bctl->flags);
2991 btrfs_mark_buffer_dirty(leaf);
2993 btrfs_free_path(path);
2994 err = btrfs_commit_transaction(trans, root);
3000 static int del_balance_item(struct btrfs_root *root)
3002 struct btrfs_trans_handle *trans;
3003 struct btrfs_path *path;
3004 struct btrfs_key key;
3007 path = btrfs_alloc_path();
3011 trans = btrfs_start_transaction(root, 0);
3012 if (IS_ERR(trans)) {
3013 btrfs_free_path(path);
3014 return PTR_ERR(trans);
3017 key.objectid = BTRFS_BALANCE_OBJECTID;
3018 key.type = BTRFS_BALANCE_ITEM_KEY;
3021 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3029 ret = btrfs_del_item(trans, root, path);
3031 btrfs_free_path(path);
3032 err = btrfs_commit_transaction(trans, root);
3039 * This is a heuristic used to reduce the number of chunks balanced on
3040 * resume after balance was interrupted.
3042 static void update_balance_args(struct btrfs_balance_control *bctl)
3045 * Turn on soft mode for chunk types that were being converted.
3047 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3048 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3049 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3050 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3051 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3052 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3055 * Turn on usage filter if is not already used. The idea is
3056 * that chunks that we have already balanced should be
3057 * reasonably full. Don't do it for chunks that are being
3058 * converted - that will keep us from relocating unconverted
3059 * (albeit full) chunks.
3061 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3062 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3063 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3064 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3065 bctl->data.usage = 90;
3067 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3068 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3069 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3070 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3071 bctl->sys.usage = 90;
3073 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3074 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3075 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3076 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3077 bctl->meta.usage = 90;
3082 * Should be called with both balance and volume mutexes held to
3083 * serialize other volume operations (add_dev/rm_dev/resize) with
3084 * restriper. Same goes for unset_balance_control.
3086 static void set_balance_control(struct btrfs_balance_control *bctl)
3088 struct btrfs_fs_info *fs_info = bctl->fs_info;
3090 BUG_ON(fs_info->balance_ctl);
3092 spin_lock(&fs_info->balance_lock);
3093 fs_info->balance_ctl = bctl;
3094 spin_unlock(&fs_info->balance_lock);
3097 static void unset_balance_control(struct btrfs_fs_info *fs_info)
3099 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3101 BUG_ON(!fs_info->balance_ctl);
3103 spin_lock(&fs_info->balance_lock);
3104 fs_info->balance_ctl = NULL;
3105 spin_unlock(&fs_info->balance_lock);
3111 * Balance filters. Return 1 if chunk should be filtered out
3112 * (should not be balanced).
3114 static int chunk_profiles_filter(u64 chunk_type,
3115 struct btrfs_balance_args *bargs)
3117 chunk_type = chunk_to_extended(chunk_type) &
3118 BTRFS_EXTENDED_PROFILE_MASK;
3120 if (bargs->profiles & chunk_type)
3126 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3127 struct btrfs_balance_args *bargs)
3129 struct btrfs_block_group_cache *cache;
3131 u64 user_thresh_min;
3132 u64 user_thresh_max;
3135 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3136 chunk_used = btrfs_block_group_used(&cache->item);
3138 if (bargs->usage_min == 0)
3139 user_thresh_min = 0;
3141 user_thresh_min = div_factor_fine(cache->key.offset,
3144 if (bargs->usage_max == 0)
3145 user_thresh_max = 1;
3146 else if (bargs->usage_max > 100)
3147 user_thresh_max = cache->key.offset;
3149 user_thresh_max = div_factor_fine(cache->key.offset,
3152 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3155 btrfs_put_block_group(cache);
3159 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info,
3160 u64 chunk_offset, struct btrfs_balance_args *bargs)
3162 struct btrfs_block_group_cache *cache;
3163 u64 chunk_used, user_thresh;
3166 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3167 chunk_used = btrfs_block_group_used(&cache->item);
3169 if (bargs->usage_min == 0)
3171 else if (bargs->usage > 100)
3172 user_thresh = cache->key.offset;
3174 user_thresh = div_factor_fine(cache->key.offset,
3177 if (chunk_used < user_thresh)
3180 btrfs_put_block_group(cache);
3184 static int chunk_devid_filter(struct extent_buffer *leaf,
3185 struct btrfs_chunk *chunk,
3186 struct btrfs_balance_args *bargs)
3188 struct btrfs_stripe *stripe;
3189 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3192 for (i = 0; i < num_stripes; i++) {
3193 stripe = btrfs_stripe_nr(chunk, i);
3194 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3201 /* [pstart, pend) */
3202 static int chunk_drange_filter(struct extent_buffer *leaf,
3203 struct btrfs_chunk *chunk,
3205 struct btrfs_balance_args *bargs)
3207 struct btrfs_stripe *stripe;
3208 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3214 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3217 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3218 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3219 factor = num_stripes / 2;
3220 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3221 factor = num_stripes - 1;
3222 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3223 factor = num_stripes - 2;
3225 factor = num_stripes;
3228 for (i = 0; i < num_stripes; i++) {
3229 stripe = btrfs_stripe_nr(chunk, i);
3230 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3233 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3234 stripe_length = btrfs_chunk_length(leaf, chunk);
3235 stripe_length = div_u64(stripe_length, factor);
3237 if (stripe_offset < bargs->pend &&
3238 stripe_offset + stripe_length > bargs->pstart)
3245 /* [vstart, vend) */
3246 static int chunk_vrange_filter(struct extent_buffer *leaf,
3247 struct btrfs_chunk *chunk,
3249 struct btrfs_balance_args *bargs)
3251 if (chunk_offset < bargs->vend &&
3252 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3253 /* at least part of the chunk is inside this vrange */
3259 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3260 struct btrfs_chunk *chunk,
3261 struct btrfs_balance_args *bargs)
3263 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3265 if (bargs->stripes_min <= num_stripes
3266 && num_stripes <= bargs->stripes_max)
3272 static int chunk_soft_convert_filter(u64 chunk_type,
3273 struct btrfs_balance_args *bargs)
3275 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3278 chunk_type = chunk_to_extended(chunk_type) &
3279 BTRFS_EXTENDED_PROFILE_MASK;
3281 if (bargs->target == chunk_type)
3287 static int should_balance_chunk(struct btrfs_root *root,
3288 struct extent_buffer *leaf,
3289 struct btrfs_chunk *chunk, u64 chunk_offset)
3291 struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
3292 struct btrfs_balance_args *bargs = NULL;
3293 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3296 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3297 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3301 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3302 bargs = &bctl->data;
3303 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3305 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3306 bargs = &bctl->meta;
3308 /* profiles filter */
3309 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3310 chunk_profiles_filter(chunk_type, bargs)) {
3315 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3316 chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
3318 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3319 chunk_usage_range_filter(bctl->fs_info, chunk_offset, bargs)) {
3324 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3325 chunk_devid_filter(leaf, chunk, bargs)) {
3329 /* drange filter, makes sense only with devid filter */
3330 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3331 chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
3336 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3337 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3341 /* stripes filter */
3342 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3343 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3347 /* soft profile changing mode */
3348 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3349 chunk_soft_convert_filter(chunk_type, bargs)) {
3354 * limited by count, must be the last filter
3356 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3357 if (bargs->limit == 0)
3361 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3363 * Same logic as the 'limit' filter; the minimum cannot be
3364 * determined here because we do not have the global informatoin
3365 * about the count of all chunks that satisfy the filters.
3367 if (bargs->limit_max == 0)
3376 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3378 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3379 struct btrfs_root *chunk_root = fs_info->chunk_root;
3380 struct btrfs_root *dev_root = fs_info->dev_root;
3381 struct list_head *devices;
3382 struct btrfs_device *device;
3386 struct btrfs_chunk *chunk;
3387 struct btrfs_path *path;
3388 struct btrfs_key key;
3389 struct btrfs_key found_key;
3390 struct btrfs_trans_handle *trans;
3391 struct extent_buffer *leaf;
3394 int enospc_errors = 0;
3395 bool counting = true;
3396 /* The single value limit and min/max limits use the same bytes in the */
3397 u64 limit_data = bctl->data.limit;
3398 u64 limit_meta = bctl->meta.limit;
3399 u64 limit_sys = bctl->sys.limit;
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(size_to_free, (u64)1 * 1024 * 1024);
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,
3504 btrfs_release_path(path);
3506 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3511 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3512 spin_lock(&fs_info->balance_lock);
3513 bctl->stat.expected++;
3514 spin_unlock(&fs_info->balance_lock);
3516 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3518 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3520 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3527 * Apply limit_min filter, no need to check if the LIMITS
3528 * filter is used, limit_min is 0 by default
3530 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3531 count_data < bctl->data.limit_min)
3532 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3533 count_meta < bctl->meta.limit_min)
3534 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3535 count_sys < bctl->sys.limit_min)) {
3536 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3540 ret = btrfs_relocate_chunk(chunk_root,
3542 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3543 if (ret && ret != -ENOSPC)
3545 if (ret == -ENOSPC) {
3548 spin_lock(&fs_info->balance_lock);
3549 bctl->stat.completed++;
3550 spin_unlock(&fs_info->balance_lock);
3553 if (found_key.offset == 0)
3555 key.offset = found_key.offset - 1;
3559 btrfs_release_path(path);
3564 btrfs_free_path(path);
3565 if (enospc_errors) {
3566 btrfs_info(fs_info, "%d enospc errors during balance",
3576 * alloc_profile_is_valid - see if a given profile is valid and reduced
3577 * @flags: profile to validate
3578 * @extended: if true @flags is treated as an extended profile
3580 static int alloc_profile_is_valid(u64 flags, int extended)
3582 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3583 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3585 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3587 /* 1) check that all other bits are zeroed */
3591 /* 2) see if profile is reduced */
3593 return !extended; /* "0" is valid for usual profiles */
3595 /* true if exactly one bit set */
3596 return (flags & (flags - 1)) == 0;
3599 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3601 /* cancel requested || normal exit path */
3602 return atomic_read(&fs_info->balance_cancel_req) ||
3603 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3604 atomic_read(&fs_info->balance_cancel_req) == 0);
3607 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3611 unset_balance_control(fs_info);
3612 ret = del_balance_item(fs_info->tree_root);
3614 btrfs_std_error(fs_info, ret, NULL);
3616 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3619 /* Non-zero return value signifies invalidity */
3620 static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg,
3623 return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3624 (!alloc_profile_is_valid(bctl_arg->target, 1) ||
3625 (bctl_arg->target & ~allowed)));
3629 * Should be called with both balance and volume mutexes held
3631 int btrfs_balance(struct btrfs_balance_control *bctl,
3632 struct btrfs_ioctl_balance_args *bargs)
3634 struct btrfs_fs_info *fs_info = bctl->fs_info;
3641 if (btrfs_fs_closing(fs_info) ||
3642 atomic_read(&fs_info->balance_pause_req) ||
3643 atomic_read(&fs_info->balance_cancel_req)) {
3648 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3649 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3653 * In case of mixed groups both data and meta should be picked,
3654 * and identical options should be given for both of them.
3656 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3657 if (mixed && (bctl->flags & allowed)) {
3658 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3659 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3660 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3661 btrfs_err(fs_info, "with mixed groups data and "
3662 "metadata balance options must be the same");
3668 num_devices = fs_info->fs_devices->num_devices;
3669 btrfs_dev_replace_lock(&fs_info->dev_replace);
3670 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3671 BUG_ON(num_devices < 1);
3674 btrfs_dev_replace_unlock(&fs_info->dev_replace);
3675 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3676 if (num_devices == 1)
3677 allowed |= BTRFS_BLOCK_GROUP_DUP;
3678 else if (num_devices > 1)
3679 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3680 if (num_devices > 2)
3681 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3682 if (num_devices > 3)
3683 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3684 BTRFS_BLOCK_GROUP_RAID6);
3685 if (validate_convert_profile(&bctl->data, allowed)) {
3686 btrfs_err(fs_info, "unable to start balance with target "
3687 "data profile %llu",
3692 if (validate_convert_profile(&bctl->meta, allowed)) {
3694 "unable to start balance with target metadata profile %llu",
3699 if (validate_convert_profile(&bctl->sys, allowed)) {
3701 "unable to start balance with target system profile %llu",
3707 /* allow dup'ed data chunks only in mixed mode */
3708 if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3709 (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) {
3710 btrfs_err(fs_info, "dup for data is not allowed");
3715 /* allow to reduce meta or sys integrity only if force set */
3716 allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3717 BTRFS_BLOCK_GROUP_RAID10 |
3718 BTRFS_BLOCK_GROUP_RAID5 |
3719 BTRFS_BLOCK_GROUP_RAID6;
3721 seq = read_seqbegin(&fs_info->profiles_lock);
3723 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3724 (fs_info->avail_system_alloc_bits & allowed) &&
3725 !(bctl->sys.target & allowed)) ||
3726 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3727 (fs_info->avail_metadata_alloc_bits & allowed) &&
3728 !(bctl->meta.target & allowed))) {
3729 if (bctl->flags & BTRFS_BALANCE_FORCE) {
3730 btrfs_info(fs_info, "force reducing metadata integrity");
3732 btrfs_err(fs_info, "balance will reduce metadata "
3733 "integrity, use force if you want this");
3738 } while (read_seqretry(&fs_info->profiles_lock, seq));
3740 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3741 fs_info->num_tolerated_disk_barrier_failures = min(
3742 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info),
3743 btrfs_get_num_tolerated_disk_barrier_failures(
3747 ret = insert_balance_item(fs_info->tree_root, bctl);
3748 if (ret && ret != -EEXIST)
3751 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3752 BUG_ON(ret == -EEXIST);
3753 set_balance_control(bctl);
3755 BUG_ON(ret != -EEXIST);
3756 spin_lock(&fs_info->balance_lock);
3757 update_balance_args(bctl);
3758 spin_unlock(&fs_info->balance_lock);
3761 atomic_inc(&fs_info->balance_running);
3762 mutex_unlock(&fs_info->balance_mutex);
3764 ret = __btrfs_balance(fs_info);
3766 mutex_lock(&fs_info->balance_mutex);
3767 atomic_dec(&fs_info->balance_running);
3769 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3770 fs_info->num_tolerated_disk_barrier_failures =
3771 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3775 memset(bargs, 0, sizeof(*bargs));
3776 update_ioctl_balance_args(fs_info, 0, bargs);
3779 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3780 balance_need_close(fs_info)) {
3781 __cancel_balance(fs_info);
3784 wake_up(&fs_info->balance_wait_q);
3788 if (bctl->flags & BTRFS_BALANCE_RESUME)
3789 __cancel_balance(fs_info);
3792 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3797 static int balance_kthread(void *data)
3799 struct btrfs_fs_info *fs_info = data;
3802 mutex_lock(&fs_info->volume_mutex);
3803 mutex_lock(&fs_info->balance_mutex);
3805 if (fs_info->balance_ctl) {
3806 btrfs_info(fs_info, "continuing balance");
3807 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3810 mutex_unlock(&fs_info->balance_mutex);
3811 mutex_unlock(&fs_info->volume_mutex);
3816 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3818 struct task_struct *tsk;
3820 spin_lock(&fs_info->balance_lock);
3821 if (!fs_info->balance_ctl) {
3822 spin_unlock(&fs_info->balance_lock);
3825 spin_unlock(&fs_info->balance_lock);
3827 if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
3828 btrfs_info(fs_info, "force skipping balance");
3832 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3833 return PTR_ERR_OR_ZERO(tsk);
3836 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3838 struct btrfs_balance_control *bctl;
3839 struct btrfs_balance_item *item;
3840 struct btrfs_disk_balance_args disk_bargs;
3841 struct btrfs_path *path;
3842 struct extent_buffer *leaf;
3843 struct btrfs_key key;
3846 path = btrfs_alloc_path();
3850 key.objectid = BTRFS_BALANCE_OBJECTID;
3851 key.type = BTRFS_BALANCE_ITEM_KEY;
3854 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3857 if (ret > 0) { /* ret = -ENOENT; */
3862 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3868 leaf = path->nodes[0];
3869 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3871 bctl->fs_info = fs_info;
3872 bctl->flags = btrfs_balance_flags(leaf, item);
3873 bctl->flags |= BTRFS_BALANCE_RESUME;
3875 btrfs_balance_data(leaf, item, &disk_bargs);
3876 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3877 btrfs_balance_meta(leaf, item, &disk_bargs);
3878 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3879 btrfs_balance_sys(leaf, item, &disk_bargs);
3880 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3882 WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
3884 mutex_lock(&fs_info->volume_mutex);
3885 mutex_lock(&fs_info->balance_mutex);
3887 set_balance_control(bctl);
3889 mutex_unlock(&fs_info->balance_mutex);
3890 mutex_unlock(&fs_info->volume_mutex);
3892 btrfs_free_path(path);
3896 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3900 mutex_lock(&fs_info->balance_mutex);
3901 if (!fs_info->balance_ctl) {
3902 mutex_unlock(&fs_info->balance_mutex);
3906 if (atomic_read(&fs_info->balance_running)) {
3907 atomic_inc(&fs_info->balance_pause_req);
3908 mutex_unlock(&fs_info->balance_mutex);
3910 wait_event(fs_info->balance_wait_q,
3911 atomic_read(&fs_info->balance_running) == 0);
3913 mutex_lock(&fs_info->balance_mutex);
3914 /* we are good with balance_ctl ripped off from under us */
3915 BUG_ON(atomic_read(&fs_info->balance_running));
3916 atomic_dec(&fs_info->balance_pause_req);
3921 mutex_unlock(&fs_info->balance_mutex);
3925 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3927 if (fs_info->sb->s_flags & MS_RDONLY)
3930 mutex_lock(&fs_info->balance_mutex);
3931 if (!fs_info->balance_ctl) {
3932 mutex_unlock(&fs_info->balance_mutex);
3936 atomic_inc(&fs_info->balance_cancel_req);
3938 * if we are running just wait and return, balance item is
3939 * deleted in btrfs_balance in this case
3941 if (atomic_read(&fs_info->balance_running)) {
3942 mutex_unlock(&fs_info->balance_mutex);
3943 wait_event(fs_info->balance_wait_q,
3944 atomic_read(&fs_info->balance_running) == 0);
3945 mutex_lock(&fs_info->balance_mutex);
3947 /* __cancel_balance needs volume_mutex */
3948 mutex_unlock(&fs_info->balance_mutex);
3949 mutex_lock(&fs_info->volume_mutex);
3950 mutex_lock(&fs_info->balance_mutex);
3952 if (fs_info->balance_ctl)
3953 __cancel_balance(fs_info);
3955 mutex_unlock(&fs_info->volume_mutex);
3958 BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
3959 atomic_dec(&fs_info->balance_cancel_req);
3960 mutex_unlock(&fs_info->balance_mutex);
3964 static int btrfs_uuid_scan_kthread(void *data)
3966 struct btrfs_fs_info *fs_info = data;
3967 struct btrfs_root *root = fs_info->tree_root;
3968 struct btrfs_key key;
3969 struct btrfs_key max_key;
3970 struct btrfs_path *path = NULL;
3972 struct extent_buffer *eb;
3974 struct btrfs_root_item root_item;
3976 struct btrfs_trans_handle *trans = NULL;
3978 path = btrfs_alloc_path();
3985 key.type = BTRFS_ROOT_ITEM_KEY;
3988 max_key.objectid = (u64)-1;
3989 max_key.type = BTRFS_ROOT_ITEM_KEY;
3990 max_key.offset = (u64)-1;
3993 ret = btrfs_search_forward(root, &key, path, 0);
4000 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4001 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4002 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4003 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4006 eb = path->nodes[0];
4007 slot = path->slots[0];
4008 item_size = btrfs_item_size_nr(eb, slot);
4009 if (item_size < sizeof(root_item))
4012 read_extent_buffer(eb, &root_item,
4013 btrfs_item_ptr_offset(eb, slot),
4014 (int)sizeof(root_item));
4015 if (btrfs_root_refs(&root_item) == 0)
4018 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4019 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4023 btrfs_release_path(path);
4025 * 1 - subvol uuid item
4026 * 1 - received_subvol uuid item
4028 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4029 if (IS_ERR(trans)) {
4030 ret = PTR_ERR(trans);
4038 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4039 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
4041 BTRFS_UUID_KEY_SUBVOL,
4044 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4050 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4051 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
4052 root_item.received_uuid,
4053 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4056 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4064 ret = btrfs_end_transaction(trans, fs_info->uuid_root);
4070 btrfs_release_path(path);
4071 if (key.offset < (u64)-1) {
4073 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4075 key.type = BTRFS_ROOT_ITEM_KEY;
4076 } else if (key.objectid < (u64)-1) {
4078 key.type = BTRFS_ROOT_ITEM_KEY;
4087 btrfs_free_path(path);
4088 if (trans && !IS_ERR(trans))
4089 btrfs_end_transaction(trans, fs_info->uuid_root);
4091 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4093 fs_info->update_uuid_tree_gen = 1;
4094 up(&fs_info->uuid_tree_rescan_sem);
4099 * Callback for btrfs_uuid_tree_iterate().
4101 * 0 check succeeded, the entry is not outdated.
4102 * < 0 if an error occured.
4103 * > 0 if the check failed, which means the caller shall remove the entry.
4105 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
4106 u8 *uuid, u8 type, u64 subid)
4108 struct btrfs_key key;
4110 struct btrfs_root *subvol_root;
4112 if (type != BTRFS_UUID_KEY_SUBVOL &&
4113 type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
4116 key.objectid = subid;
4117 key.type = BTRFS_ROOT_ITEM_KEY;
4118 key.offset = (u64)-1;
4119 subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
4120 if (IS_ERR(subvol_root)) {
4121 ret = PTR_ERR(subvol_root);
4128 case BTRFS_UUID_KEY_SUBVOL:
4129 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
4132 case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
4133 if (memcmp(uuid, subvol_root->root_item.received_uuid,
4143 static int btrfs_uuid_rescan_kthread(void *data)
4145 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4149 * 1st step is to iterate through the existing UUID tree and
4150 * to delete all entries that contain outdated data.
4151 * 2nd step is to add all missing entries to the UUID tree.
4153 ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4155 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4156 up(&fs_info->uuid_tree_rescan_sem);
4159 return btrfs_uuid_scan_kthread(data);
4162 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4164 struct btrfs_trans_handle *trans;
4165 struct btrfs_root *tree_root = fs_info->tree_root;
4166 struct btrfs_root *uuid_root;
4167 struct task_struct *task;
4174 trans = btrfs_start_transaction(tree_root, 2);
4176 return PTR_ERR(trans);
4178 uuid_root = btrfs_create_tree(trans, fs_info,
4179 BTRFS_UUID_TREE_OBJECTID);
4180 if (IS_ERR(uuid_root)) {
4181 ret = PTR_ERR(uuid_root);
4182 btrfs_abort_transaction(trans, tree_root, ret);
4186 fs_info->uuid_root = uuid_root;
4188 ret = btrfs_commit_transaction(trans, tree_root);
4192 down(&fs_info->uuid_tree_rescan_sem);
4193 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4195 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4196 btrfs_warn(fs_info, "failed to start uuid_scan task");
4197 up(&fs_info->uuid_tree_rescan_sem);
4198 return PTR_ERR(task);
4204 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4206 struct task_struct *task;
4208 down(&fs_info->uuid_tree_rescan_sem);
4209 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4211 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4212 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4213 up(&fs_info->uuid_tree_rescan_sem);
4214 return PTR_ERR(task);
4221 * shrinking a device means finding all of the device extents past
4222 * the new size, and then following the back refs to the chunks.
4223 * The chunk relocation code actually frees the device extent
4225 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4227 struct btrfs_trans_handle *trans;
4228 struct btrfs_root *root = device->dev_root;
4229 struct btrfs_dev_extent *dev_extent = NULL;
4230 struct btrfs_path *path;
4236 bool retried = false;
4237 bool checked_pending_chunks = false;
4238 struct extent_buffer *l;
4239 struct btrfs_key key;
4240 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4241 u64 old_total = btrfs_super_total_bytes(super_copy);
4242 u64 old_size = btrfs_device_get_total_bytes(device);
4243 u64 diff = old_size - new_size;
4245 if (device->is_tgtdev_for_dev_replace)
4248 path = btrfs_alloc_path();
4256 btrfs_device_set_total_bytes(device, new_size);
4257 if (device->writeable) {
4258 device->fs_devices->total_rw_bytes -= diff;
4259 spin_lock(&root->fs_info->free_chunk_lock);
4260 root->fs_info->free_chunk_space -= diff;
4261 spin_unlock(&root->fs_info->free_chunk_lock);
4263 unlock_chunks(root);
4266 key.objectid = device->devid;
4267 key.offset = (u64)-1;
4268 key.type = BTRFS_DEV_EXTENT_KEY;
4271 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
4272 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4274 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4278 ret = btrfs_previous_item(root, path, 0, key.type);
4280 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4285 btrfs_release_path(path);
4290 slot = path->slots[0];
4291 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4293 if (key.objectid != device->devid) {
4294 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4295 btrfs_release_path(path);
4299 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4300 length = btrfs_dev_extent_length(l, dev_extent);
4302 if (key.offset + length <= new_size) {
4303 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4304 btrfs_release_path(path);
4308 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4309 btrfs_release_path(path);
4311 ret = btrfs_relocate_chunk(root, chunk_offset);
4312 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4313 if (ret && ret != -ENOSPC)
4317 } while (key.offset-- > 0);
4319 if (failed && !retried) {
4323 } else if (failed && retried) {
4328 /* Shrinking succeeded, else we would be at "done". */
4329 trans = btrfs_start_transaction(root, 0);
4330 if (IS_ERR(trans)) {
4331 ret = PTR_ERR(trans);
4338 * We checked in the above loop all device extents that were already in
4339 * the device tree. However before we have updated the device's
4340 * total_bytes to the new size, we might have had chunk allocations that
4341 * have not complete yet (new block groups attached to transaction
4342 * handles), and therefore their device extents were not yet in the
4343 * device tree and we missed them in the loop above. So if we have any
4344 * pending chunk using a device extent that overlaps the device range
4345 * that we can not use anymore, commit the current transaction and
4346 * repeat the search on the device tree - this way we guarantee we will
4347 * not have chunks using device extents that end beyond 'new_size'.
4349 if (!checked_pending_chunks) {
4350 u64 start = new_size;
4351 u64 len = old_size - new_size;
4353 if (contains_pending_extent(trans->transaction, device,
4355 unlock_chunks(root);
4356 checked_pending_chunks = true;
4359 ret = btrfs_commit_transaction(trans, root);
4366 btrfs_device_set_disk_total_bytes(device, new_size);
4367 if (list_empty(&device->resized_list))
4368 list_add_tail(&device->resized_list,
4369 &root->fs_info->fs_devices->resized_devices);
4371 WARN_ON(diff > old_total);
4372 btrfs_set_super_total_bytes(super_copy, old_total - diff);
4373 unlock_chunks(root);
4375 /* Now btrfs_update_device() will change the on-disk size. */
4376 ret = btrfs_update_device(trans, device);
4377 btrfs_end_transaction(trans, root);
4379 btrfs_free_path(path);
4382 btrfs_device_set_total_bytes(device, old_size);
4383 if (device->writeable)
4384 device->fs_devices->total_rw_bytes += diff;
4385 spin_lock(&root->fs_info->free_chunk_lock);
4386 root->fs_info->free_chunk_space += diff;
4387 spin_unlock(&root->fs_info->free_chunk_lock);
4388 unlock_chunks(root);
4393 static int btrfs_add_system_chunk(struct btrfs_root *root,
4394 struct btrfs_key *key,
4395 struct btrfs_chunk *chunk, int item_size)
4397 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4398 struct btrfs_disk_key disk_key;
4403 array_size = btrfs_super_sys_array_size(super_copy);
4404 if (array_size + item_size + sizeof(disk_key)
4405 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4406 unlock_chunks(root);
4410 ptr = super_copy->sys_chunk_array + array_size;
4411 btrfs_cpu_key_to_disk(&disk_key, key);
4412 memcpy(ptr, &disk_key, sizeof(disk_key));
4413 ptr += sizeof(disk_key);
4414 memcpy(ptr, chunk, item_size);
4415 item_size += sizeof(disk_key);
4416 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4417 unlock_chunks(root);
4423 * sort the devices in descending order by max_avail, total_avail
4425 static int btrfs_cmp_device_info(const void *a, const void *b)
4427 const struct btrfs_device_info *di_a = a;
4428 const struct btrfs_device_info *di_b = b;
4430 if (di_a->max_avail > di_b->max_avail)
4432 if (di_a->max_avail < di_b->max_avail)
4434 if (di_a->total_avail > di_b->total_avail)
4436 if (di_a->total_avail < di_b->total_avail)
4441 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
4443 /* TODO allow them to set a preferred stripe size */
4447 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4449 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4452 btrfs_set_fs_incompat(info, RAID56);
4455 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r) \
4456 - sizeof(struct btrfs_item) \
4457 - sizeof(struct btrfs_chunk)) \
4458 / sizeof(struct btrfs_stripe) + 1)
4460 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
4461 - 2 * sizeof(struct btrfs_disk_key) \
4462 - 2 * sizeof(struct btrfs_chunk)) \
4463 / sizeof(struct btrfs_stripe) + 1)
4465 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4466 struct btrfs_root *extent_root, u64 start,
4469 struct btrfs_fs_info *info = extent_root->fs_info;
4470 struct btrfs_fs_devices *fs_devices = info->fs_devices;
4471 struct list_head *cur;
4472 struct map_lookup *map = NULL;
4473 struct extent_map_tree *em_tree;
4474 struct extent_map *em;
4475 struct btrfs_device_info *devices_info = NULL;
4477 int num_stripes; /* total number of stripes to allocate */
4478 int data_stripes; /* number of stripes that count for
4480 int sub_stripes; /* sub_stripes info for map */
4481 int dev_stripes; /* stripes per dev */
4482 int devs_max; /* max devs to use */
4483 int devs_min; /* min devs needed */
4484 int devs_increment; /* ndevs has to be a multiple of this */
4485 int ncopies; /* how many copies to data has */
4487 u64 max_stripe_size;
4491 u64 raid_stripe_len = BTRFS_STRIPE_LEN;
4497 BUG_ON(!alloc_profile_is_valid(type, 0));
4499 if (list_empty(&fs_devices->alloc_list))
4502 index = __get_raid_index(type);
4504 sub_stripes = btrfs_raid_array[index].sub_stripes;
4505 dev_stripes = btrfs_raid_array[index].dev_stripes;
4506 devs_max = btrfs_raid_array[index].devs_max;
4507 devs_min = btrfs_raid_array[index].devs_min;
4508 devs_increment = btrfs_raid_array[index].devs_increment;
4509 ncopies = btrfs_raid_array[index].ncopies;
4511 if (type & BTRFS_BLOCK_GROUP_DATA) {
4512 max_stripe_size = 1024 * 1024 * 1024;
4513 max_chunk_size = 10 * max_stripe_size;
4515 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4516 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4517 /* for larger filesystems, use larger metadata chunks */
4518 if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024)
4519 max_stripe_size = 1024 * 1024 * 1024;
4521 max_stripe_size = 256 * 1024 * 1024;
4522 max_chunk_size = max_stripe_size;
4524 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4525 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4526 max_stripe_size = 32 * 1024 * 1024;
4527 max_chunk_size = 2 * max_stripe_size;
4529 devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4531 btrfs_err(info, "invalid chunk type 0x%llx requested",
4536 /* we don't want a chunk larger than 10% of writeable space */
4537 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4540 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4545 cur = fs_devices->alloc_list.next;
4548 * in the first pass through the devices list, we gather information
4549 * about the available holes on each device.
4552 while (cur != &fs_devices->alloc_list) {
4553 struct btrfs_device *device;
4557 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4561 if (!device->writeable) {
4563 "BTRFS: read-only device in alloc_list\n");
4567 if (!device->in_fs_metadata ||
4568 device->is_tgtdev_for_dev_replace)
4571 if (device->total_bytes > device->bytes_used)
4572 total_avail = device->total_bytes - device->bytes_used;
4576 /* If there is no space on this device, skip it. */
4577 if (total_avail == 0)
4580 ret = find_free_dev_extent(trans, device,
4581 max_stripe_size * dev_stripes,
4582 &dev_offset, &max_avail);
4583 if (ret && ret != -ENOSPC)
4587 max_avail = max_stripe_size * dev_stripes;
4589 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4592 if (ndevs == fs_devices->rw_devices) {
4593 WARN(1, "%s: found more than %llu devices\n",
4594 __func__, fs_devices->rw_devices);
4597 devices_info[ndevs].dev_offset = dev_offset;
4598 devices_info[ndevs].max_avail = max_avail;
4599 devices_info[ndevs].total_avail = total_avail;
4600 devices_info[ndevs].dev = device;
4605 * now sort the devices by hole size / available space
4607 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4608 btrfs_cmp_device_info, NULL);
4610 /* round down to number of usable stripes */
4611 ndevs -= ndevs % devs_increment;
4613 if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4618 if (devs_max && ndevs > devs_max)
4621 * the primary goal is to maximize the number of stripes, so use as many
4622 * devices as possible, even if the stripes are not maximum sized.
4624 stripe_size = devices_info[ndevs-1].max_avail;
4625 num_stripes = ndevs * dev_stripes;
4628 * this will have to be fixed for RAID1 and RAID10 over
4631 data_stripes = num_stripes / ncopies;
4633 if (type & BTRFS_BLOCK_GROUP_RAID5) {
4634 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
4635 btrfs_super_stripesize(info->super_copy));
4636 data_stripes = num_stripes - 1;
4638 if (type & BTRFS_BLOCK_GROUP_RAID6) {
4639 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
4640 btrfs_super_stripesize(info->super_copy));
4641 data_stripes = num_stripes - 2;
4645 * Use the number of data stripes to figure out how big this chunk
4646 * is really going to be in terms of logical address space,
4647 * and compare that answer with the max chunk size
4649 if (stripe_size * data_stripes > max_chunk_size) {
4650 u64 mask = (1ULL << 24) - 1;
4652 stripe_size = div_u64(max_chunk_size, data_stripes);
4654 /* bump the answer up to a 16MB boundary */
4655 stripe_size = (stripe_size + mask) & ~mask;
4657 /* but don't go higher than the limits we found
4658 * while searching for free extents
4660 if (stripe_size > devices_info[ndevs-1].max_avail)
4661 stripe_size = devices_info[ndevs-1].max_avail;
4664 stripe_size = div_u64(stripe_size, dev_stripes);
4666 /* align to BTRFS_STRIPE_LEN */
4667 stripe_size = div_u64(stripe_size, raid_stripe_len);
4668 stripe_size *= raid_stripe_len;
4670 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4675 map->num_stripes = num_stripes;
4677 for (i = 0; i < ndevs; ++i) {
4678 for (j = 0; j < dev_stripes; ++j) {
4679 int s = i * dev_stripes + j;
4680 map->stripes[s].dev = devices_info[i].dev;
4681 map->stripes[s].physical = devices_info[i].dev_offset +
4685 map->sector_size = extent_root->sectorsize;
4686 map->stripe_len = raid_stripe_len;
4687 map->io_align = raid_stripe_len;
4688 map->io_width = raid_stripe_len;
4690 map->sub_stripes = sub_stripes;
4692 num_bytes = stripe_size * data_stripes;
4694 trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
4696 em = alloc_extent_map();
4702 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4703 em->bdev = (struct block_device *)map;
4705 em->len = num_bytes;
4706 em->block_start = 0;
4707 em->block_len = em->len;
4708 em->orig_block_len = stripe_size;
4710 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4711 write_lock(&em_tree->lock);
4712 ret = add_extent_mapping(em_tree, em, 0);
4714 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4715 atomic_inc(&em->refs);
4717 write_unlock(&em_tree->lock);
4719 free_extent_map(em);
4723 ret = btrfs_make_block_group(trans, extent_root, 0, type,
4724 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4727 goto error_del_extent;
4729 for (i = 0; i < map->num_stripes; i++) {
4730 num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4731 btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4734 spin_lock(&extent_root->fs_info->free_chunk_lock);
4735 extent_root->fs_info->free_chunk_space -= (stripe_size *
4737 spin_unlock(&extent_root->fs_info->free_chunk_lock);
4739 free_extent_map(em);
4740 check_raid56_incompat_flag(extent_root->fs_info, type);
4742 kfree(devices_info);
4746 write_lock(&em_tree->lock);
4747 remove_extent_mapping(em_tree, em);
4748 write_unlock(&em_tree->lock);
4750 /* One for our allocation */
4751 free_extent_map(em);
4752 /* One for the tree reference */
4753 free_extent_map(em);
4754 /* One for the pending_chunks list reference */
4755 free_extent_map(em);
4757 kfree(devices_info);
4761 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4762 struct btrfs_root *extent_root,
4763 u64 chunk_offset, u64 chunk_size)
4765 struct btrfs_key key;
4766 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4767 struct btrfs_device *device;
4768 struct btrfs_chunk *chunk;
4769 struct btrfs_stripe *stripe;
4770 struct extent_map_tree *em_tree;
4771 struct extent_map *em;
4772 struct map_lookup *map;
4779 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4780 read_lock(&em_tree->lock);
4781 em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size);
4782 read_unlock(&em_tree->lock);
4785 btrfs_crit(extent_root->fs_info, "unable to find logical "
4786 "%Lu len %Lu", chunk_offset, chunk_size);
4790 if (em->start != chunk_offset || em->len != chunk_size) {
4791 btrfs_crit(extent_root->fs_info, "found a bad mapping, wanted"
4792 " %Lu-%Lu, found %Lu-%Lu", chunk_offset,
4793 chunk_size, em->start, em->len);
4794 free_extent_map(em);
4798 map = (struct map_lookup *)em->bdev;
4799 item_size = btrfs_chunk_item_size(map->num_stripes);
4800 stripe_size = em->orig_block_len;
4802 chunk = kzalloc(item_size, GFP_NOFS);
4808 for (i = 0; i < map->num_stripes; i++) {
4809 device = map->stripes[i].dev;
4810 dev_offset = map->stripes[i].physical;
4812 ret = btrfs_update_device(trans, device);
4815 ret = btrfs_alloc_dev_extent(trans, device,
4816 chunk_root->root_key.objectid,
4817 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4818 chunk_offset, dev_offset,
4824 stripe = &chunk->stripe;
4825 for (i = 0; i < map->num_stripes; i++) {
4826 device = map->stripes[i].dev;
4827 dev_offset = map->stripes[i].physical;
4829 btrfs_set_stack_stripe_devid(stripe, device->devid);
4830 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4831 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4835 btrfs_set_stack_chunk_length(chunk, chunk_size);
4836 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4837 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4838 btrfs_set_stack_chunk_type(chunk, map->type);
4839 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4840 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4841 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4842 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4843 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4845 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4846 key.type = BTRFS_CHUNK_ITEM_KEY;
4847 key.offset = chunk_offset;
4849 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4850 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4852 * TODO: Cleanup of inserted chunk root in case of
4855 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
4861 free_extent_map(em);
4866 * Chunk allocation falls into two parts. The first part does works
4867 * that make the new allocated chunk useable, but not do any operation
4868 * that modifies the chunk tree. The second part does the works that
4869 * require modifying the chunk tree. This division is important for the
4870 * bootstrap process of adding storage to a seed btrfs.
4872 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4873 struct btrfs_root *extent_root, u64 type)
4877 ASSERT(mutex_is_locked(&extent_root->fs_info->chunk_mutex));
4878 chunk_offset = find_next_chunk(extent_root->fs_info);
4879 return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type);
4882 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4883 struct btrfs_root *root,
4884 struct btrfs_device *device)
4887 u64 sys_chunk_offset;
4889 struct btrfs_fs_info *fs_info = root->fs_info;
4890 struct btrfs_root *extent_root = fs_info->extent_root;
4893 chunk_offset = find_next_chunk(fs_info);
4894 alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
4895 ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset,
4900 sys_chunk_offset = find_next_chunk(root->fs_info);
4901 alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
4902 ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset,
4907 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
4911 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
4912 BTRFS_BLOCK_GROUP_RAID10 |
4913 BTRFS_BLOCK_GROUP_RAID5 |
4914 BTRFS_BLOCK_GROUP_DUP)) {
4916 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
4925 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
4927 struct extent_map *em;
4928 struct map_lookup *map;
4929 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4934 read_lock(&map_tree->map_tree.lock);
4935 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
4936 read_unlock(&map_tree->map_tree.lock);
4940 map = (struct map_lookup *)em->bdev;
4941 for (i = 0; i < map->num_stripes; i++) {
4942 if (map->stripes[i].dev->missing) {
4947 if (!map->stripes[i].dev->writeable) {
4954 * If the number of missing devices is larger than max errors,
4955 * we can not write the data into that chunk successfully, so
4958 if (miss_ndevs > btrfs_chunk_max_errors(map))
4961 free_extent_map(em);
4965 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
4967 extent_map_tree_init(&tree->map_tree);
4970 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
4972 struct extent_map *em;
4975 write_lock(&tree->map_tree.lock);
4976 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
4978 remove_extent_mapping(&tree->map_tree, em);
4979 write_unlock(&tree->map_tree.lock);
4983 free_extent_map(em);
4984 /* once for the tree */
4985 free_extent_map(em);
4989 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
4991 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4992 struct extent_map *em;
4993 struct map_lookup *map;
4994 struct extent_map_tree *em_tree = &map_tree->map_tree;
4997 read_lock(&em_tree->lock);
4998 em = lookup_extent_mapping(em_tree, logical, len);
4999 read_unlock(&em_tree->lock);
5002 * We could return errors for these cases, but that could get ugly and
5003 * we'd probably do the same thing which is just not do anything else
5004 * and exit, so return 1 so the callers don't try to use other copies.
5007 btrfs_crit(fs_info, "No mapping for %Lu-%Lu", logical,
5012 if (em->start > logical || em->start + em->len < logical) {
5013 btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got "
5014 "%Lu-%Lu", logical, logical+len, em->start,
5015 em->start + em->len);
5016 free_extent_map(em);
5020 map = (struct map_lookup *)em->bdev;
5021 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
5022 ret = map->num_stripes;
5023 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5024 ret = map->sub_stripes;
5025 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5027 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5031 free_extent_map(em);
5033 btrfs_dev_replace_lock(&fs_info->dev_replace);
5034 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
5036 btrfs_dev_replace_unlock(&fs_info->dev_replace);
5041 unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
5042 struct btrfs_mapping_tree *map_tree,
5045 struct extent_map *em;
5046 struct map_lookup *map;
5047 struct extent_map_tree *em_tree = &map_tree->map_tree;
5048 unsigned long len = root->sectorsize;
5050 read_lock(&em_tree->lock);
5051 em = lookup_extent_mapping(em_tree, logical, len);
5052 read_unlock(&em_tree->lock);
5055 BUG_ON(em->start > logical || em->start + em->len < logical);
5056 map = (struct map_lookup *)em->bdev;
5057 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5058 len = map->stripe_len * nr_data_stripes(map);
5059 free_extent_map(em);
5063 int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
5064 u64 logical, u64 len, int mirror_num)
5066 struct extent_map *em;
5067 struct map_lookup *map;
5068 struct extent_map_tree *em_tree = &map_tree->map_tree;
5071 read_lock(&em_tree->lock);
5072 em = lookup_extent_mapping(em_tree, logical, len);
5073 read_unlock(&em_tree->lock);
5076 BUG_ON(em->start > logical || em->start + em->len < logical);
5077 map = (struct map_lookup *)em->bdev;
5078 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5080 free_extent_map(em);
5084 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5085 struct map_lookup *map, int first, int num,
5086 int optimal, int dev_replace_is_ongoing)
5090 struct btrfs_device *srcdev;
5092 if (dev_replace_is_ongoing &&
5093 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5094 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5095 srcdev = fs_info->dev_replace.srcdev;
5100 * try to avoid the drive that is the source drive for a
5101 * dev-replace procedure, only choose it if no other non-missing
5102 * mirror is available
5104 for (tolerance = 0; tolerance < 2; tolerance++) {
5105 if (map->stripes[optimal].dev->bdev &&
5106 (tolerance || map->stripes[optimal].dev != srcdev))
5108 for (i = first; i < first + num; i++) {
5109 if (map->stripes[i].dev->bdev &&
5110 (tolerance || map->stripes[i].dev != srcdev))
5115 /* we couldn't find one that doesn't fail. Just return something
5116 * and the io error handling code will clean up eventually
5121 static inline int parity_smaller(u64 a, u64 b)
5126 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5127 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5129 struct btrfs_bio_stripe s;
5136 for (i = 0; i < num_stripes - 1; i++) {
5137 if (parity_smaller(bbio->raid_map[i],
5138 bbio->raid_map[i+1])) {
5139 s = bbio->stripes[i];
5140 l = bbio->raid_map[i];
5141 bbio->stripes[i] = bbio->stripes[i+1];
5142 bbio->raid_map[i] = bbio->raid_map[i+1];
5143 bbio->stripes[i+1] = s;
5144 bbio->raid_map[i+1] = l;
5152 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5154 struct btrfs_bio *bbio = kzalloc(
5155 /* the size of the btrfs_bio */
5156 sizeof(struct btrfs_bio) +
5157 /* plus the variable array for the stripes */
5158 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5159 /* plus the variable array for the tgt dev */
5160 sizeof(int) * (real_stripes) +
5162 * plus the raid_map, which includes both the tgt dev
5165 sizeof(u64) * (total_stripes),
5166 GFP_NOFS|__GFP_NOFAIL);
5168 atomic_set(&bbio->error, 0);
5169 atomic_set(&bbio->refs, 1);
5174 void btrfs_get_bbio(struct btrfs_bio *bbio)
5176 WARN_ON(!atomic_read(&bbio->refs));
5177 atomic_inc(&bbio->refs);
5180 void btrfs_put_bbio(struct btrfs_bio *bbio)
5184 if (atomic_dec_and_test(&bbio->refs))
5188 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5189 u64 logical, u64 *length,
5190 struct btrfs_bio **bbio_ret,
5191 int mirror_num, int need_raid_map)
5193 struct extent_map *em;
5194 struct map_lookup *map;
5195 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5196 struct extent_map_tree *em_tree = &map_tree->map_tree;
5199 u64 stripe_end_offset;
5209 int tgtdev_indexes = 0;
5210 struct btrfs_bio *bbio = NULL;
5211 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
5212 int dev_replace_is_ongoing = 0;
5213 int num_alloc_stripes;
5214 int patch_the_first_stripe_for_dev_replace = 0;
5215 u64 physical_to_patch_in_first_stripe = 0;
5216 u64 raid56_full_stripe_start = (u64)-1;
5218 read_lock(&em_tree->lock);
5219 em = lookup_extent_mapping(em_tree, logical, *length);
5220 read_unlock(&em_tree->lock);
5223 btrfs_crit(fs_info, "unable to find logical %llu len %llu",
5228 if (em->start > logical || em->start + em->len < logical) {
5229 btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, "
5230 "found %Lu-%Lu", logical, em->start,
5231 em->start + em->len);
5232 free_extent_map(em);
5236 map = (struct map_lookup *)em->bdev;
5237 offset = logical - em->start;
5239 stripe_len = map->stripe_len;
5242 * stripe_nr counts the total number of stripes we have to stride
5243 * to get to this block
5245 stripe_nr = div64_u64(stripe_nr, stripe_len);
5247 stripe_offset = stripe_nr * stripe_len;
5248 BUG_ON(offset < stripe_offset);
5250 /* stripe_offset is the offset of this block in its stripe*/
5251 stripe_offset = offset - stripe_offset;
5253 /* if we're here for raid56, we need to know the stripe aligned start */
5254 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5255 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5256 raid56_full_stripe_start = offset;
5258 /* allow a write of a full stripe, but make sure we don't
5259 * allow straddling of stripes
5261 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5263 raid56_full_stripe_start *= full_stripe_len;
5266 if (rw & REQ_DISCARD) {
5267 /* we don't discard raid56 yet */
5268 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5272 *length = min_t(u64, em->len - offset, *length);
5273 } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5275 /* For writes to RAID[56], allow a full stripeset across all disks.
5276 For other RAID types and for RAID[56] reads, just allow a single
5277 stripe (on a single disk). */
5278 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
5280 max_len = stripe_len * nr_data_stripes(map) -
5281 (offset - raid56_full_stripe_start);
5283 /* we limit the length of each bio to what fits in a stripe */
5284 max_len = stripe_len - stripe_offset;
5286 *length = min_t(u64, em->len - offset, max_len);
5288 *length = em->len - offset;
5291 /* This is for when we're called from btrfs_merge_bio_hook() and all
5292 it cares about is the length */
5296 btrfs_dev_replace_lock(dev_replace);
5297 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5298 if (!dev_replace_is_ongoing)
5299 btrfs_dev_replace_unlock(dev_replace);
5301 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
5302 !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) &&
5303 dev_replace->tgtdev != NULL) {
5305 * in dev-replace case, for repair case (that's the only
5306 * case where the mirror is selected explicitly when
5307 * calling btrfs_map_block), blocks left of the left cursor
5308 * can also be read from the target drive.
5309 * For REQ_GET_READ_MIRRORS, the target drive is added as
5310 * the last one to the array of stripes. For READ, it also
5311 * needs to be supported using the same mirror number.
5312 * If the requested block is not left of the left cursor,
5313 * EIO is returned. This can happen because btrfs_num_copies()
5314 * returns one more in the dev-replace case.
5316 u64 tmp_length = *length;
5317 struct btrfs_bio *tmp_bbio = NULL;
5318 int tmp_num_stripes;
5319 u64 srcdev_devid = dev_replace->srcdev->devid;
5320 int index_srcdev = 0;
5322 u64 physical_of_found = 0;
5324 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
5325 logical, &tmp_length, &tmp_bbio, 0, 0);
5327 WARN_ON(tmp_bbio != NULL);
5331 tmp_num_stripes = tmp_bbio->num_stripes;
5332 if (mirror_num > tmp_num_stripes) {
5334 * REQ_GET_READ_MIRRORS does not contain this
5335 * mirror, that means that the requested area
5336 * is not left of the left cursor
5339 btrfs_put_bbio(tmp_bbio);
5344 * process the rest of the function using the mirror_num
5345 * of the source drive. Therefore look it up first.
5346 * At the end, patch the device pointer to the one of the
5349 for (i = 0; i < tmp_num_stripes; i++) {
5350 if (tmp_bbio->stripes[i].dev->devid == srcdev_devid) {
5352 * In case of DUP, in order to keep it
5353 * simple, only add the mirror with the
5354 * lowest physical address
5357 physical_of_found <=
5358 tmp_bbio->stripes[i].physical)
5363 tmp_bbio->stripes[i].physical;
5368 mirror_num = index_srcdev + 1;
5369 patch_the_first_stripe_for_dev_replace = 1;
5370 physical_to_patch_in_first_stripe = physical_of_found;
5374 btrfs_put_bbio(tmp_bbio);
5378 btrfs_put_bbio(tmp_bbio);
5379 } else if (mirror_num > map->num_stripes) {
5385 stripe_nr_orig = stripe_nr;
5386 stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
5387 stripe_nr_end = div_u64(stripe_nr_end, map->stripe_len);
5388 stripe_end_offset = stripe_nr_end * map->stripe_len -
5391 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5392 if (rw & REQ_DISCARD)
5393 num_stripes = min_t(u64, map->num_stripes,
5394 stripe_nr_end - stripe_nr_orig);
5395 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5397 if (!(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)))
5399 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
5400 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))
5401 num_stripes = map->num_stripes;
5402 else if (mirror_num)
5403 stripe_index = mirror_num - 1;
5405 stripe_index = find_live_mirror(fs_info, map, 0,
5407 current->pid % map->num_stripes,
5408 dev_replace_is_ongoing);
5409 mirror_num = stripe_index + 1;
5412 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
5413 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
5414 num_stripes = map->num_stripes;
5415 } else if (mirror_num) {
5416 stripe_index = mirror_num - 1;
5421 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5422 u32 factor = map->num_stripes / map->sub_stripes;
5424 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5425 stripe_index *= map->sub_stripes;
5427 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
5428 num_stripes = map->sub_stripes;
5429 else if (rw & REQ_DISCARD)
5430 num_stripes = min_t(u64, map->sub_stripes *
5431 (stripe_nr_end - stripe_nr_orig),
5433 else if (mirror_num)
5434 stripe_index += mirror_num - 1;
5436 int old_stripe_index = stripe_index;
5437 stripe_index = find_live_mirror(fs_info, map,
5439 map->sub_stripes, stripe_index +
5440 current->pid % map->sub_stripes,
5441 dev_replace_is_ongoing);
5442 mirror_num = stripe_index - old_stripe_index + 1;
5445 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5446 if (need_raid_map &&
5447 ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) ||
5449 /* push stripe_nr back to the start of the full stripe */
5450 stripe_nr = div_u64(raid56_full_stripe_start,
5451 stripe_len * nr_data_stripes(map));
5453 /* RAID[56] write or recovery. Return all stripes */
5454 num_stripes = map->num_stripes;
5455 max_errors = nr_parity_stripes(map);
5457 *length = map->stripe_len;
5462 * Mirror #0 or #1 means the original data block.
5463 * Mirror #2 is RAID5 parity block.
5464 * Mirror #3 is RAID6 Q block.
5466 stripe_nr = div_u64_rem(stripe_nr,
5467 nr_data_stripes(map), &stripe_index);
5469 stripe_index = nr_data_stripes(map) +
5472 /* We distribute the parity blocks across stripes */
5473 div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
5475 if (!(rw & (REQ_WRITE | REQ_DISCARD |
5476 REQ_GET_READ_MIRRORS)) && mirror_num <= 1)
5481 * after this, stripe_nr is the number of stripes on this
5482 * device we have to walk to find the data, and stripe_index is
5483 * the number of our device in the stripe array
5485 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5487 mirror_num = stripe_index + 1;
5489 BUG_ON(stripe_index >= map->num_stripes);
5491 num_alloc_stripes = num_stripes;
5492 if (dev_replace_is_ongoing) {
5493 if (rw & (REQ_WRITE | REQ_DISCARD))
5494 num_alloc_stripes <<= 1;
5495 if (rw & REQ_GET_READ_MIRRORS)
5496 num_alloc_stripes++;
5497 tgtdev_indexes = num_stripes;
5500 bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
5505 if (dev_replace_is_ongoing)
5506 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
5508 /* build raid_map */
5509 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK &&
5510 need_raid_map && ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) ||
5515 bbio->raid_map = (u64 *)((void *)bbio->stripes +
5516 sizeof(struct btrfs_bio_stripe) *
5518 sizeof(int) * tgtdev_indexes);
5520 /* Work out the disk rotation on this stripe-set */
5521 div_u64_rem(stripe_nr, num_stripes, &rot);
5523 /* Fill in the logical address of each stripe */
5524 tmp = stripe_nr * nr_data_stripes(map);
5525 for (i = 0; i < nr_data_stripes(map); i++)
5526 bbio->raid_map[(i+rot) % num_stripes] =
5527 em->start + (tmp + i) * map->stripe_len;
5529 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5530 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5531 bbio->raid_map[(i+rot+1) % num_stripes] =
5535 if (rw & REQ_DISCARD) {
5537 u32 sub_stripes = 0;
5538 u64 stripes_per_dev = 0;
5539 u32 remaining_stripes = 0;
5540 u32 last_stripe = 0;
5543 (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
5544 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5547 sub_stripes = map->sub_stripes;
5549 factor = map->num_stripes / sub_stripes;
5550 stripes_per_dev = div_u64_rem(stripe_nr_end -
5553 &remaining_stripes);
5554 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5555 last_stripe *= sub_stripes;
5558 for (i = 0; i < num_stripes; i++) {
5559 bbio->stripes[i].physical =
5560 map->stripes[stripe_index].physical +
5561 stripe_offset + stripe_nr * map->stripe_len;
5562 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5564 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5565 BTRFS_BLOCK_GROUP_RAID10)) {
5566 bbio->stripes[i].length = stripes_per_dev *
5569 if (i / sub_stripes < remaining_stripes)
5570 bbio->stripes[i].length +=
5574 * Special for the first stripe and
5577 * |-------|...|-------|
5581 if (i < sub_stripes)
5582 bbio->stripes[i].length -=
5585 if (stripe_index >= last_stripe &&
5586 stripe_index <= (last_stripe +
5588 bbio->stripes[i].length -=
5591 if (i == sub_stripes - 1)
5594 bbio->stripes[i].length = *length;
5597 if (stripe_index == map->num_stripes) {
5598 /* This could only happen for RAID0/10 */
5604 for (i = 0; i < num_stripes; i++) {
5605 bbio->stripes[i].physical =
5606 map->stripes[stripe_index].physical +
5608 stripe_nr * map->stripe_len;
5609 bbio->stripes[i].dev =
5610 map->stripes[stripe_index].dev;
5615 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
5616 max_errors = btrfs_chunk_max_errors(map);
5619 sort_parity_stripes(bbio, num_stripes);
5622 if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) &&
5623 dev_replace->tgtdev != NULL) {
5624 int index_where_to_add;
5625 u64 srcdev_devid = dev_replace->srcdev->devid;
5628 * duplicate the write operations while the dev replace
5629 * procedure is running. Since the copying of the old disk
5630 * to the new disk takes place at run time while the
5631 * filesystem is mounted writable, the regular write
5632 * operations to the old disk have to be duplicated to go
5633 * to the new disk as well.
5634 * Note that device->missing is handled by the caller, and
5635 * that the write to the old disk is already set up in the
5638 index_where_to_add = num_stripes;
5639 for (i = 0; i < num_stripes; i++) {
5640 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5641 /* write to new disk, too */
5642 struct btrfs_bio_stripe *new =
5643 bbio->stripes + index_where_to_add;
5644 struct btrfs_bio_stripe *old =
5647 new->physical = old->physical;
5648 new->length = old->length;
5649 new->dev = dev_replace->tgtdev;
5650 bbio->tgtdev_map[i] = index_where_to_add;
5651 index_where_to_add++;
5656 num_stripes = index_where_to_add;
5657 } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) &&
5658 dev_replace->tgtdev != NULL) {
5659 u64 srcdev_devid = dev_replace->srcdev->devid;
5660 int index_srcdev = 0;
5662 u64 physical_of_found = 0;
5665 * During the dev-replace procedure, the target drive can
5666 * also be used to read data in case it is needed to repair
5667 * a corrupt block elsewhere. This is possible if the
5668 * requested area is left of the left cursor. In this area,
5669 * the target drive is a full copy of the source drive.
5671 for (i = 0; i < num_stripes; i++) {
5672 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5674 * In case of DUP, in order to keep it
5675 * simple, only add the mirror with the
5676 * lowest physical address
5679 physical_of_found <=
5680 bbio->stripes[i].physical)
5684 physical_of_found = bbio->stripes[i].physical;
5688 if (physical_of_found + map->stripe_len <=
5689 dev_replace->cursor_left) {
5690 struct btrfs_bio_stripe *tgtdev_stripe =
5691 bbio->stripes + num_stripes;
5693 tgtdev_stripe->physical = physical_of_found;
5694 tgtdev_stripe->length =
5695 bbio->stripes[index_srcdev].length;
5696 tgtdev_stripe->dev = dev_replace->tgtdev;
5697 bbio->tgtdev_map[index_srcdev] = num_stripes;
5706 bbio->map_type = map->type;
5707 bbio->num_stripes = num_stripes;
5708 bbio->max_errors = max_errors;
5709 bbio->mirror_num = mirror_num;
5710 bbio->num_tgtdevs = tgtdev_indexes;
5713 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5714 * mirror_num == num_stripes + 1 && dev_replace target drive is
5715 * available as a mirror
5717 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5718 WARN_ON(num_stripes > 1);
5719 bbio->stripes[0].dev = dev_replace->tgtdev;
5720 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5721 bbio->mirror_num = map->num_stripes + 1;
5724 if (dev_replace_is_ongoing)
5725 btrfs_dev_replace_unlock(dev_replace);
5726 free_extent_map(em);
5730 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5731 u64 logical, u64 *length,
5732 struct btrfs_bio **bbio_ret, int mirror_num)
5734 return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5738 /* For Scrub/replace */
5739 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, int rw,
5740 u64 logical, u64 *length,
5741 struct btrfs_bio **bbio_ret, int mirror_num,
5744 return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5745 mirror_num, need_raid_map);
5748 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
5749 u64 chunk_start, u64 physical, u64 devid,
5750 u64 **logical, int *naddrs, int *stripe_len)
5752 struct extent_map_tree *em_tree = &map_tree->map_tree;
5753 struct extent_map *em;
5754 struct map_lookup *map;
5762 read_lock(&em_tree->lock);
5763 em = lookup_extent_mapping(em_tree, chunk_start, 1);
5764 read_unlock(&em_tree->lock);
5767 printk(KERN_ERR "BTRFS: couldn't find em for chunk %Lu\n",
5772 if (em->start != chunk_start) {
5773 printk(KERN_ERR "BTRFS: bad chunk start, em=%Lu, wanted=%Lu\n",
5774 em->start, chunk_start);
5775 free_extent_map(em);
5778 map = (struct map_lookup *)em->bdev;
5781 rmap_len = map->stripe_len;
5783 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5784 length = div_u64(length, map->num_stripes / map->sub_stripes);
5785 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5786 length = div_u64(length, map->num_stripes);
5787 else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5788 length = div_u64(length, nr_data_stripes(map));
5789 rmap_len = map->stripe_len * nr_data_stripes(map);
5792 buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
5793 BUG_ON(!buf); /* -ENOMEM */
5795 for (i = 0; i < map->num_stripes; i++) {
5796 if (devid && map->stripes[i].dev->devid != devid)
5798 if (map->stripes[i].physical > physical ||
5799 map->stripes[i].physical + length <= physical)
5802 stripe_nr = physical - map->stripes[i].physical;
5803 stripe_nr = div_u64(stripe_nr, map->stripe_len);
5805 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5806 stripe_nr = stripe_nr * map->num_stripes + i;
5807 stripe_nr = div_u64(stripe_nr, map->sub_stripes);
5808 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5809 stripe_nr = stripe_nr * map->num_stripes + i;
5810 } /* else if RAID[56], multiply by nr_data_stripes().
5811 * Alternatively, just use rmap_len below instead of
5812 * map->stripe_len */
5814 bytenr = chunk_start + stripe_nr * rmap_len;
5815 WARN_ON(nr >= map->num_stripes);
5816 for (j = 0; j < nr; j++) {
5817 if (buf[j] == bytenr)
5821 WARN_ON(nr >= map->num_stripes);
5828 *stripe_len = rmap_len;
5830 free_extent_map(em);
5834 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
5836 bio->bi_private = bbio->private;
5837 bio->bi_end_io = bbio->end_io;
5840 btrfs_put_bbio(bbio);
5843 static void btrfs_end_bio(struct bio *bio)
5845 struct btrfs_bio *bbio = bio->bi_private;
5846 int is_orig_bio = 0;
5848 if (bio->bi_error) {
5849 atomic_inc(&bbio->error);
5850 if (bio->bi_error == -EIO || bio->bi_error == -EREMOTEIO) {
5851 unsigned int stripe_index =
5852 btrfs_io_bio(bio)->stripe_index;
5853 struct btrfs_device *dev;
5855 BUG_ON(stripe_index >= bbio->num_stripes);
5856 dev = bbio->stripes[stripe_index].dev;
5858 if (bio->bi_rw & WRITE)
5859 btrfs_dev_stat_inc(dev,
5860 BTRFS_DEV_STAT_WRITE_ERRS);
5862 btrfs_dev_stat_inc(dev,
5863 BTRFS_DEV_STAT_READ_ERRS);
5864 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
5865 btrfs_dev_stat_inc(dev,
5866 BTRFS_DEV_STAT_FLUSH_ERRS);
5867 btrfs_dev_stat_print_on_error(dev);
5872 if (bio == bbio->orig_bio)
5875 btrfs_bio_counter_dec(bbio->fs_info);
5877 if (atomic_dec_and_test(&bbio->stripes_pending)) {
5880 bio = bbio->orig_bio;
5883 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5884 /* only send an error to the higher layers if it is
5885 * beyond the tolerance of the btrfs bio
5887 if (atomic_read(&bbio->error) > bbio->max_errors) {
5888 bio->bi_error = -EIO;
5891 * this bio is actually up to date, we didn't
5892 * go over the max number of errors
5897 btrfs_end_bbio(bbio, bio);
5898 } else if (!is_orig_bio) {
5904 * see run_scheduled_bios for a description of why bios are collected for
5907 * This will add one bio to the pending list for a device and make sure
5908 * the work struct is scheduled.
5910 static noinline void btrfs_schedule_bio(struct btrfs_root *root,
5911 struct btrfs_device *device,
5912 int rw, struct bio *bio)
5914 int should_queue = 1;
5915 struct btrfs_pending_bios *pending_bios;
5917 if (device->missing || !device->bdev) {
5922 /* don't bother with additional async steps for reads, right now */
5923 if (!(rw & REQ_WRITE)) {
5925 btrfsic_submit_bio(rw, bio);
5931 * nr_async_bios allows us to reliably return congestion to the
5932 * higher layers. Otherwise, the async bio makes it appear we have
5933 * made progress against dirty pages when we've really just put it
5934 * on a queue for later
5936 atomic_inc(&root->fs_info->nr_async_bios);
5937 WARN_ON(bio->bi_next);
5938 bio->bi_next = NULL;
5941 spin_lock(&device->io_lock);
5942 if (bio->bi_rw & REQ_SYNC)
5943 pending_bios = &device->pending_sync_bios;
5945 pending_bios = &device->pending_bios;
5947 if (pending_bios->tail)
5948 pending_bios->tail->bi_next = bio;
5950 pending_bios->tail = bio;
5951 if (!pending_bios->head)
5952 pending_bios->head = bio;
5953 if (device->running_pending)
5956 spin_unlock(&device->io_lock);
5959 btrfs_queue_work(root->fs_info->submit_workers,
5963 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5964 struct bio *bio, u64 physical, int dev_nr,
5967 struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
5969 bio->bi_private = bbio;
5970 btrfs_io_bio(bio)->stripe_index = dev_nr;
5971 bio->bi_end_io = btrfs_end_bio;
5972 bio->bi_iter.bi_sector = physical >> 9;
5975 struct rcu_string *name;
5978 name = rcu_dereference(dev->name);
5979 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
5980 "(%s id %llu), size=%u\n", rw,
5981 (u64)bio->bi_iter.bi_sector, (u_long)dev->bdev->bd_dev,
5982 name->str, dev->devid, bio->bi_iter.bi_size);
5986 bio->bi_bdev = dev->bdev;
5988 btrfs_bio_counter_inc_noblocked(root->fs_info);
5991 btrfs_schedule_bio(root, dev, rw, bio);
5993 btrfsic_submit_bio(rw, bio);
5996 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
5998 atomic_inc(&bbio->error);
5999 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6000 /* Shoud be the original bio. */
6001 WARN_ON(bio != bbio->orig_bio);
6003 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6004 bio->bi_iter.bi_sector = logical >> 9;
6005 bio->bi_error = -EIO;
6006 btrfs_end_bbio(bbio, bio);
6010 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
6011 int mirror_num, int async_submit)
6013 struct btrfs_device *dev;
6014 struct bio *first_bio = bio;
6015 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
6021 struct btrfs_bio *bbio = NULL;
6023 length = bio->bi_iter.bi_size;
6024 map_length = length;
6026 btrfs_bio_counter_inc_blocked(root->fs_info);
6027 ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
6030 btrfs_bio_counter_dec(root->fs_info);
6034 total_devs = bbio->num_stripes;
6035 bbio->orig_bio = first_bio;
6036 bbio->private = first_bio->bi_private;
6037 bbio->end_io = first_bio->bi_end_io;
6038 bbio->fs_info = root->fs_info;
6039 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6041 if (bbio->raid_map) {
6042 /* In this case, map_length has been set to the length of
6043 a single stripe; not the whole write */
6045 ret = raid56_parity_write(root, bio, bbio, map_length);
6047 ret = raid56_parity_recover(root, bio, bbio, map_length,
6051 btrfs_bio_counter_dec(root->fs_info);
6055 if (map_length < length) {
6056 btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu",
6057 logical, length, map_length);
6061 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6062 dev = bbio->stripes[dev_nr].dev;
6063 if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
6064 bbio_error(bbio, first_bio, logical);
6068 if (dev_nr < total_devs - 1) {
6069 bio = btrfs_bio_clone(first_bio, GFP_NOFS);
6070 BUG_ON(!bio); /* -ENOMEM */
6074 submit_stripe_bio(root, bbio, bio,
6075 bbio->stripes[dev_nr].physical, dev_nr, rw,
6078 btrfs_bio_counter_dec(root->fs_info);
6082 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
6085 struct btrfs_device *device;
6086 struct btrfs_fs_devices *cur_devices;
6088 cur_devices = fs_info->fs_devices;
6089 while (cur_devices) {
6091 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
6092 device = __find_device(&cur_devices->devices,
6097 cur_devices = cur_devices->seed;
6102 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
6103 struct btrfs_fs_devices *fs_devices,
6104 u64 devid, u8 *dev_uuid)
6106 struct btrfs_device *device;
6108 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6112 list_add(&device->dev_list, &fs_devices->devices);
6113 device->fs_devices = fs_devices;
6114 fs_devices->num_devices++;
6116 device->missing = 1;
6117 fs_devices->missing_devices++;
6123 * btrfs_alloc_device - allocate struct btrfs_device
6124 * @fs_info: used only for generating a new devid, can be NULL if
6125 * devid is provided (i.e. @devid != NULL).
6126 * @devid: a pointer to devid for this device. If NULL a new devid
6128 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6131 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6132 * on error. Returned struct is not linked onto any lists and can be
6133 * destroyed with kfree() right away.
6135 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6139 struct btrfs_device *dev;
6142 if (WARN_ON(!devid && !fs_info))
6143 return ERR_PTR(-EINVAL);
6145 dev = __alloc_device();
6154 ret = find_next_devid(fs_info, &tmp);
6157 return ERR_PTR(ret);
6163 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6165 generate_random_uuid(dev->uuid);
6167 btrfs_init_work(&dev->work, btrfs_submit_helper,
6168 pending_bios_fn, NULL, NULL);
6173 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
6174 struct extent_buffer *leaf,
6175 struct btrfs_chunk *chunk)
6177 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
6178 struct map_lookup *map;
6179 struct extent_map *em;
6183 u8 uuid[BTRFS_UUID_SIZE];
6188 logical = key->offset;
6189 length = btrfs_chunk_length(leaf, chunk);
6191 read_lock(&map_tree->map_tree.lock);
6192 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
6193 read_unlock(&map_tree->map_tree.lock);
6195 /* already mapped? */
6196 if (em && em->start <= logical && em->start + em->len > logical) {
6197 free_extent_map(em);
6200 free_extent_map(em);
6203 em = alloc_extent_map();
6206 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6207 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6209 free_extent_map(em);
6213 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6214 em->bdev = (struct block_device *)map;
6215 em->start = logical;
6218 em->block_start = 0;
6219 em->block_len = em->len;
6221 map->num_stripes = num_stripes;
6222 map->io_width = btrfs_chunk_io_width(leaf, chunk);
6223 map->io_align = btrfs_chunk_io_align(leaf, chunk);
6224 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
6225 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6226 map->type = btrfs_chunk_type(leaf, chunk);
6227 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6228 for (i = 0; i < num_stripes; i++) {
6229 map->stripes[i].physical =
6230 btrfs_stripe_offset_nr(leaf, chunk, i);
6231 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6232 read_extent_buffer(leaf, uuid, (unsigned long)
6233 btrfs_stripe_dev_uuid_nr(chunk, i),
6235 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
6237 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
6238 free_extent_map(em);
6241 if (!map->stripes[i].dev) {
6242 map->stripes[i].dev =
6243 add_missing_dev(root, root->fs_info->fs_devices,
6245 if (!map->stripes[i].dev) {
6246 free_extent_map(em);
6249 btrfs_warn(root->fs_info, "devid %llu uuid %pU is missing",
6252 map->stripes[i].dev->in_fs_metadata = 1;
6255 write_lock(&map_tree->map_tree.lock);
6256 ret = add_extent_mapping(&map_tree->map_tree, em, 0);
6257 write_unlock(&map_tree->map_tree.lock);
6258 BUG_ON(ret); /* Tree corruption */
6259 free_extent_map(em);
6264 static void fill_device_from_item(struct extent_buffer *leaf,
6265 struct btrfs_dev_item *dev_item,
6266 struct btrfs_device *device)
6270 device->devid = btrfs_device_id(leaf, dev_item);
6271 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6272 device->total_bytes = device->disk_total_bytes;
6273 device->commit_total_bytes = device->disk_total_bytes;
6274 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6275 device->commit_bytes_used = device->bytes_used;
6276 device->type = btrfs_device_type(leaf, dev_item);
6277 device->io_align = btrfs_device_io_align(leaf, dev_item);
6278 device->io_width = btrfs_device_io_width(leaf, dev_item);
6279 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6280 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6281 device->is_tgtdev_for_dev_replace = 0;
6283 ptr = btrfs_device_uuid(dev_item);
6284 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6287 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_root *root,
6290 struct btrfs_fs_devices *fs_devices;
6293 BUG_ON(!mutex_is_locked(&uuid_mutex));
6295 fs_devices = root->fs_info->fs_devices->seed;
6296 while (fs_devices) {
6297 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE))
6300 fs_devices = fs_devices->seed;
6303 fs_devices = find_fsid(fsid);
6305 if (!btrfs_test_opt(root, DEGRADED))
6306 return ERR_PTR(-ENOENT);
6308 fs_devices = alloc_fs_devices(fsid);
6309 if (IS_ERR(fs_devices))
6312 fs_devices->seeding = 1;
6313 fs_devices->opened = 1;
6317 fs_devices = clone_fs_devices(fs_devices);
6318 if (IS_ERR(fs_devices))
6321 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
6322 root->fs_info->bdev_holder);
6324 free_fs_devices(fs_devices);
6325 fs_devices = ERR_PTR(ret);
6329 if (!fs_devices->seeding) {
6330 __btrfs_close_devices(fs_devices);
6331 free_fs_devices(fs_devices);
6332 fs_devices = ERR_PTR(-EINVAL);
6336 fs_devices->seed = root->fs_info->fs_devices->seed;
6337 root->fs_info->fs_devices->seed = fs_devices;
6342 static int read_one_dev(struct btrfs_root *root,
6343 struct extent_buffer *leaf,
6344 struct btrfs_dev_item *dev_item)
6346 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6347 struct btrfs_device *device;
6350 u8 fs_uuid[BTRFS_UUID_SIZE];
6351 u8 dev_uuid[BTRFS_UUID_SIZE];
6353 devid = btrfs_device_id(leaf, dev_item);
6354 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6356 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6359 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
6360 fs_devices = open_seed_devices(root, fs_uuid);
6361 if (IS_ERR(fs_devices))
6362 return PTR_ERR(fs_devices);
6365 device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
6367 if (!btrfs_test_opt(root, DEGRADED))
6370 device = add_missing_dev(root, fs_devices, devid, dev_uuid);
6373 btrfs_warn(root->fs_info, "devid %llu uuid %pU missing",
6376 if (!device->bdev && !btrfs_test_opt(root, DEGRADED))
6379 if(!device->bdev && !device->missing) {
6381 * this happens when a device that was properly setup
6382 * in the device info lists suddenly goes bad.
6383 * device->bdev is NULL, and so we have to set
6384 * device->missing to one here
6386 device->fs_devices->missing_devices++;
6387 device->missing = 1;
6390 /* Move the device to its own fs_devices */
6391 if (device->fs_devices != fs_devices) {
6392 ASSERT(device->missing);
6394 list_move(&device->dev_list, &fs_devices->devices);
6395 device->fs_devices->num_devices--;
6396 fs_devices->num_devices++;
6398 device->fs_devices->missing_devices--;
6399 fs_devices->missing_devices++;
6401 device->fs_devices = fs_devices;
6405 if (device->fs_devices != root->fs_info->fs_devices) {
6406 BUG_ON(device->writeable);
6407 if (device->generation !=
6408 btrfs_device_generation(leaf, dev_item))
6412 fill_device_from_item(leaf, dev_item, device);
6413 device->in_fs_metadata = 1;
6414 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
6415 device->fs_devices->total_rw_bytes += device->total_bytes;
6416 spin_lock(&root->fs_info->free_chunk_lock);
6417 root->fs_info->free_chunk_space += device->total_bytes -
6419 spin_unlock(&root->fs_info->free_chunk_lock);
6425 int btrfs_read_sys_array(struct btrfs_root *root)
6427 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
6428 struct extent_buffer *sb;
6429 struct btrfs_disk_key *disk_key;
6430 struct btrfs_chunk *chunk;
6432 unsigned long sb_array_offset;
6438 struct btrfs_key key;
6440 ASSERT(BTRFS_SUPER_INFO_SIZE <= root->nodesize);
6442 * This will create extent buffer of nodesize, superblock size is
6443 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6444 * overallocate but we can keep it as-is, only the first page is used.
6446 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET);
6449 btrfs_set_buffer_uptodate(sb);
6450 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6452 * The sb extent buffer is artifical and just used to read the system array.
6453 * btrfs_set_buffer_uptodate() call does not properly mark all it's
6454 * pages up-to-date when the page is larger: extent does not cover the
6455 * whole page and consequently check_page_uptodate does not find all
6456 * the page's extents up-to-date (the hole beyond sb),
6457 * write_extent_buffer then triggers a WARN_ON.
6459 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6460 * but sb spans only this function. Add an explicit SetPageUptodate call
6461 * to silence the warning eg. on PowerPC 64.
6463 if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
6464 SetPageUptodate(sb->pages[0]);
6466 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6467 array_size = btrfs_super_sys_array_size(super_copy);
6469 array_ptr = super_copy->sys_chunk_array;
6470 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6473 while (cur_offset < array_size) {
6474 disk_key = (struct btrfs_disk_key *)array_ptr;
6475 len = sizeof(*disk_key);
6476 if (cur_offset + len > array_size)
6477 goto out_short_read;
6479 btrfs_disk_key_to_cpu(&key, disk_key);
6482 sb_array_offset += len;
6485 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6486 chunk = (struct btrfs_chunk *)sb_array_offset;
6488 * At least one btrfs_chunk with one stripe must be
6489 * present, exact stripe count check comes afterwards
6491 len = btrfs_chunk_item_size(1);
6492 if (cur_offset + len > array_size)
6493 goto out_short_read;
6495 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6496 len = btrfs_chunk_item_size(num_stripes);
6497 if (cur_offset + len > array_size)
6498 goto out_short_read;
6500 ret = read_one_chunk(root, &key, sb, chunk);
6508 sb_array_offset += len;
6511 free_extent_buffer(sb);
6515 printk(KERN_ERR "BTRFS: sys_array too short to read %u bytes at offset %u\n",
6517 free_extent_buffer(sb);
6521 int btrfs_read_chunk_tree(struct btrfs_root *root)
6523 struct btrfs_path *path;
6524 struct extent_buffer *leaf;
6525 struct btrfs_key key;
6526 struct btrfs_key found_key;
6530 root = root->fs_info->chunk_root;
6532 path = btrfs_alloc_path();
6536 mutex_lock(&uuid_mutex);
6540 * Read all device items, and then all the chunk items. All
6541 * device items are found before any chunk item (their object id
6542 * is smaller than the lowest possible object id for a chunk
6543 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6545 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6548 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6552 leaf = path->nodes[0];
6553 slot = path->slots[0];
6554 if (slot >= btrfs_header_nritems(leaf)) {
6555 ret = btrfs_next_leaf(root, path);
6562 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6563 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6564 struct btrfs_dev_item *dev_item;
6565 dev_item = btrfs_item_ptr(leaf, slot,
6566 struct btrfs_dev_item);
6567 ret = read_one_dev(root, leaf, dev_item);
6570 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6571 struct btrfs_chunk *chunk;
6572 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6573 ret = read_one_chunk(root, &found_key, leaf, chunk);
6581 unlock_chunks(root);
6582 mutex_unlock(&uuid_mutex);
6584 btrfs_free_path(path);
6588 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6590 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6591 struct btrfs_device *device;
6593 while (fs_devices) {
6594 mutex_lock(&fs_devices->device_list_mutex);
6595 list_for_each_entry(device, &fs_devices->devices, dev_list)
6596 device->dev_root = fs_info->dev_root;
6597 mutex_unlock(&fs_devices->device_list_mutex);
6599 fs_devices = fs_devices->seed;
6603 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6607 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6608 btrfs_dev_stat_reset(dev, i);
6611 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6613 struct btrfs_key key;
6614 struct btrfs_key found_key;
6615 struct btrfs_root *dev_root = fs_info->dev_root;
6616 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6617 struct extent_buffer *eb;
6620 struct btrfs_device *device;
6621 struct btrfs_path *path = NULL;
6624 path = btrfs_alloc_path();
6630 mutex_lock(&fs_devices->device_list_mutex);
6631 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6633 struct btrfs_dev_stats_item *ptr;
6636 key.type = BTRFS_DEV_STATS_KEY;
6637 key.offset = device->devid;
6638 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6640 __btrfs_reset_dev_stats(device);
6641 device->dev_stats_valid = 1;
6642 btrfs_release_path(path);
6645 slot = path->slots[0];
6646 eb = path->nodes[0];
6647 btrfs_item_key_to_cpu(eb, &found_key, slot);
6648 item_size = btrfs_item_size_nr(eb, slot);
6650 ptr = btrfs_item_ptr(eb, slot,
6651 struct btrfs_dev_stats_item);
6653 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6654 if (item_size >= (1 + i) * sizeof(__le64))
6655 btrfs_dev_stat_set(device, i,
6656 btrfs_dev_stats_value(eb, ptr, i));
6658 btrfs_dev_stat_reset(device, i);
6661 device->dev_stats_valid = 1;
6662 btrfs_dev_stat_print_on_load(device);
6663 btrfs_release_path(path);
6665 mutex_unlock(&fs_devices->device_list_mutex);
6668 btrfs_free_path(path);
6669 return ret < 0 ? ret : 0;
6672 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
6673 struct btrfs_root *dev_root,
6674 struct btrfs_device *device)
6676 struct btrfs_path *path;
6677 struct btrfs_key key;
6678 struct extent_buffer *eb;
6679 struct btrfs_dev_stats_item *ptr;
6684 key.type = BTRFS_DEV_STATS_KEY;
6685 key.offset = device->devid;
6687 path = btrfs_alloc_path();
6689 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
6691 btrfs_warn_in_rcu(dev_root->fs_info,
6692 "error %d while searching for dev_stats item for device %s",
6693 ret, rcu_str_deref(device->name));
6698 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
6699 /* need to delete old one and insert a new one */
6700 ret = btrfs_del_item(trans, dev_root, path);
6702 btrfs_warn_in_rcu(dev_root->fs_info,
6703 "delete too small dev_stats item for device %s failed %d",
6704 rcu_str_deref(device->name), ret);
6711 /* need to insert a new item */
6712 btrfs_release_path(path);
6713 ret = btrfs_insert_empty_item(trans, dev_root, path,
6714 &key, sizeof(*ptr));
6716 btrfs_warn_in_rcu(dev_root->fs_info,
6717 "insert dev_stats item for device %s failed %d",
6718 rcu_str_deref(device->name), ret);
6723 eb = path->nodes[0];
6724 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
6725 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6726 btrfs_set_dev_stats_value(eb, ptr, i,
6727 btrfs_dev_stat_read(device, i));
6728 btrfs_mark_buffer_dirty(eb);
6731 btrfs_free_path(path);
6736 * called from commit_transaction. Writes all changed device stats to disk.
6738 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
6739 struct btrfs_fs_info *fs_info)
6741 struct btrfs_root *dev_root = fs_info->dev_root;
6742 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6743 struct btrfs_device *device;
6747 mutex_lock(&fs_devices->device_list_mutex);
6748 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6749 if (!device->dev_stats_valid || !btrfs_dev_stats_dirty(device))
6752 stats_cnt = atomic_read(&device->dev_stats_ccnt);
6753 ret = update_dev_stat_item(trans, dev_root, device);
6755 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
6757 mutex_unlock(&fs_devices->device_list_mutex);
6762 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
6764 btrfs_dev_stat_inc(dev, index);
6765 btrfs_dev_stat_print_on_error(dev);
6768 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
6770 if (!dev->dev_stats_valid)
6772 btrfs_err_rl_in_rcu(dev->dev_root->fs_info,
6773 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
6774 rcu_str_deref(dev->name),
6775 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6776 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6777 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6778 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6779 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6782 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
6786 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6787 if (btrfs_dev_stat_read(dev, i) != 0)
6789 if (i == BTRFS_DEV_STAT_VALUES_MAX)
6790 return; /* all values == 0, suppress message */
6792 btrfs_info_in_rcu(dev->dev_root->fs_info,
6793 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
6794 rcu_str_deref(dev->name),
6795 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6796 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6797 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6798 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6799 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6802 int btrfs_get_dev_stats(struct btrfs_root *root,
6803 struct btrfs_ioctl_get_dev_stats *stats)
6805 struct btrfs_device *dev;
6806 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6809 mutex_lock(&fs_devices->device_list_mutex);
6810 dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
6811 mutex_unlock(&fs_devices->device_list_mutex);
6814 btrfs_warn(root->fs_info, "get dev_stats failed, device not found");
6816 } else if (!dev->dev_stats_valid) {
6817 btrfs_warn(root->fs_info, "get dev_stats failed, not yet valid");
6819 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
6820 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6821 if (stats->nr_items > i)
6823 btrfs_dev_stat_read_and_reset(dev, i);
6825 btrfs_dev_stat_reset(dev, i);
6828 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6829 if (stats->nr_items > i)
6830 stats->values[i] = btrfs_dev_stat_read(dev, i);
6832 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
6833 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
6837 void btrfs_scratch_superblocks(struct block_device *bdev, char *device_path)
6839 struct buffer_head *bh;
6840 struct btrfs_super_block *disk_super;
6846 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX;
6849 if (btrfs_read_dev_one_super(bdev, copy_num, &bh))
6852 disk_super = (struct btrfs_super_block *)bh->b_data;
6854 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
6855 set_buffer_dirty(bh);
6856 sync_dirty_buffer(bh);
6860 /* Notify udev that device has changed */
6861 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
6863 /* Update ctime/mtime for device path for libblkid */
6864 update_dev_time(device_path);
6868 * Update the size of all devices, which is used for writing out the
6871 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
6873 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6874 struct btrfs_device *curr, *next;
6876 if (list_empty(&fs_devices->resized_devices))
6879 mutex_lock(&fs_devices->device_list_mutex);
6880 lock_chunks(fs_info->dev_root);
6881 list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
6883 list_del_init(&curr->resized_list);
6884 curr->commit_total_bytes = curr->disk_total_bytes;
6886 unlock_chunks(fs_info->dev_root);
6887 mutex_unlock(&fs_devices->device_list_mutex);
6890 /* Must be invoked during the transaction commit */
6891 void btrfs_update_commit_device_bytes_used(struct btrfs_root *root,
6892 struct btrfs_transaction *transaction)
6894 struct extent_map *em;
6895 struct map_lookup *map;
6896 struct btrfs_device *dev;
6899 if (list_empty(&transaction->pending_chunks))
6902 /* In order to kick the device replace finish process */
6904 list_for_each_entry(em, &transaction->pending_chunks, list) {
6905 map = (struct map_lookup *)em->bdev;
6907 for (i = 0; i < map->num_stripes; i++) {
6908 dev = map->stripes[i].dev;
6909 dev->commit_bytes_used = dev->bytes_used;
6912 unlock_chunks(root);
6915 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
6917 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6918 while (fs_devices) {
6919 fs_devices->fs_info = fs_info;
6920 fs_devices = fs_devices->seed;
6924 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
6926 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6927 while (fs_devices) {
6928 fs_devices->fs_info = NULL;
6929 fs_devices = fs_devices->seed;
6933 void btrfs_close_one_device(struct btrfs_device *device)
6935 struct btrfs_fs_devices *fs_devices = device->fs_devices;
6936 struct btrfs_device *new_device;
6937 struct rcu_string *name;
6940 fs_devices->open_devices--;
6942 if (device->writeable &&
6943 device->devid != BTRFS_DEV_REPLACE_DEVID) {
6944 list_del_init(&device->dev_alloc_list);
6945 fs_devices->rw_devices--;
6948 if (device->missing)
6949 fs_devices->missing_devices--;
6951 new_device = btrfs_alloc_device(NULL, &device->devid,
6953 BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
6955 /* Safe because we are under uuid_mutex */
6957 name = rcu_string_strdup(device->name->str, GFP_NOFS);
6958 BUG_ON(!name); /* -ENOMEM */
6959 rcu_assign_pointer(new_device->name, name);
6962 list_replace_rcu(&device->dev_list, &new_device->dev_list);
6963 new_device->fs_devices = device->fs_devices;
6965 call_rcu(&device->rcu, free_device);