2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/bio.h>
20 #include <linux/slab.h>
21 #include <linux/buffer_head.h>
22 #include <linux/blkdev.h>
23 #include <linux/random.h>
24 #include <linux/iocontext.h>
25 #include <linux/capability.h>
26 #include <linux/ratelimit.h>
27 #include <linux/kthread.h>
28 #include <linux/raid/pq.h>
29 #include <linux/semaphore.h>
30 #include <asm/div64.h>
32 #include "extent_map.h"
34 #include "transaction.h"
35 #include "print-tree.h"
38 #include "async-thread.h"
39 #include "check-integrity.h"
40 #include "rcu-string.h"
42 #include "dev-replace.h"
45 const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
46 [BTRFS_RAID_RAID10] = {
49 .devs_max = 0, /* 0 == as many as possible */
51 .tolerated_failures = 1,
55 [BTRFS_RAID_RAID1] = {
60 .tolerated_failures = 1,
69 .tolerated_failures = 0,
73 [BTRFS_RAID_RAID0] = {
78 .tolerated_failures = 0,
82 [BTRFS_RAID_SINGLE] = {
87 .tolerated_failures = 0,
91 [BTRFS_RAID_RAID5] = {
96 .tolerated_failures = 1,
100 [BTRFS_RAID_RAID6] = {
105 .tolerated_failures = 2,
111 const u64 btrfs_raid_group[BTRFS_NR_RAID_TYPES] = {
112 [BTRFS_RAID_RAID10] = BTRFS_BLOCK_GROUP_RAID10,
113 [BTRFS_RAID_RAID1] = BTRFS_BLOCK_GROUP_RAID1,
114 [BTRFS_RAID_DUP] = BTRFS_BLOCK_GROUP_DUP,
115 [BTRFS_RAID_RAID0] = BTRFS_BLOCK_GROUP_RAID0,
116 [BTRFS_RAID_SINGLE] = 0,
117 [BTRFS_RAID_RAID5] = BTRFS_BLOCK_GROUP_RAID5,
118 [BTRFS_RAID_RAID6] = BTRFS_BLOCK_GROUP_RAID6,
121 static int init_first_rw_device(struct btrfs_trans_handle *trans,
122 struct btrfs_root *root,
123 struct btrfs_device *device);
124 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
125 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
126 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
127 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
128 static void btrfs_close_one_device(struct btrfs_device *device);
130 DEFINE_MUTEX(uuid_mutex);
131 static LIST_HEAD(fs_uuids);
132 struct list_head *btrfs_get_fs_uuids(void)
137 static struct btrfs_fs_devices *__alloc_fs_devices(void)
139 struct btrfs_fs_devices *fs_devs;
141 fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL);
143 return ERR_PTR(-ENOMEM);
145 mutex_init(&fs_devs->device_list_mutex);
147 INIT_LIST_HEAD(&fs_devs->devices);
148 INIT_LIST_HEAD(&fs_devs->resized_devices);
149 INIT_LIST_HEAD(&fs_devs->alloc_list);
150 INIT_LIST_HEAD(&fs_devs->list);
156 * alloc_fs_devices - allocate struct btrfs_fs_devices
157 * @fsid: a pointer to UUID for this FS. If NULL a new UUID is
160 * Return: a pointer to a new &struct btrfs_fs_devices on success;
161 * ERR_PTR() on error. Returned struct is not linked onto any lists and
162 * can be destroyed with kfree() right away.
164 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
166 struct btrfs_fs_devices *fs_devs;
168 fs_devs = __alloc_fs_devices();
173 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
175 generate_random_uuid(fs_devs->fsid);
180 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
182 struct btrfs_device *device;
183 WARN_ON(fs_devices->opened);
184 while (!list_empty(&fs_devices->devices)) {
185 device = list_entry(fs_devices->devices.next,
186 struct btrfs_device, dev_list);
187 list_del(&device->dev_list);
188 rcu_string_free(device->name);
194 static void btrfs_kobject_uevent(struct block_device *bdev,
195 enum kobject_action action)
199 ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
201 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
203 kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
204 &disk_to_dev(bdev->bd_disk)->kobj);
207 void btrfs_cleanup_fs_uuids(void)
209 struct btrfs_fs_devices *fs_devices;
211 while (!list_empty(&fs_uuids)) {
212 fs_devices = list_entry(fs_uuids.next,
213 struct btrfs_fs_devices, list);
214 list_del(&fs_devices->list);
215 free_fs_devices(fs_devices);
219 static struct btrfs_device *__alloc_device(void)
221 struct btrfs_device *dev;
223 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
225 return ERR_PTR(-ENOMEM);
227 INIT_LIST_HEAD(&dev->dev_list);
228 INIT_LIST_HEAD(&dev->dev_alloc_list);
229 INIT_LIST_HEAD(&dev->resized_list);
231 spin_lock_init(&dev->io_lock);
233 spin_lock_init(&dev->reada_lock);
234 atomic_set(&dev->reada_in_flight, 0);
235 atomic_set(&dev->dev_stats_ccnt, 0);
236 btrfs_device_data_ordered_init(dev);
237 INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
238 INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
243 static noinline struct btrfs_device *__find_device(struct list_head *head,
246 struct btrfs_device *dev;
248 list_for_each_entry(dev, head, dev_list) {
249 if (dev->devid == devid &&
250 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
257 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
259 struct btrfs_fs_devices *fs_devices;
261 list_for_each_entry(fs_devices, &fs_uuids, list) {
262 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
269 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
270 int flush, struct block_device **bdev,
271 struct buffer_head **bh)
275 *bdev = blkdev_get_by_path(device_path, flags, holder);
278 ret = PTR_ERR(*bdev);
283 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
284 ret = set_blocksize(*bdev, 4096);
286 blkdev_put(*bdev, flags);
289 invalidate_bdev(*bdev);
290 *bh = btrfs_read_dev_super(*bdev);
293 blkdev_put(*bdev, flags);
305 static void requeue_list(struct btrfs_pending_bios *pending_bios,
306 struct bio *head, struct bio *tail)
309 struct bio *old_head;
311 old_head = pending_bios->head;
312 pending_bios->head = head;
313 if (pending_bios->tail)
314 tail->bi_next = old_head;
316 pending_bios->tail = tail;
320 * we try to collect pending bios for a device so we don't get a large
321 * number of procs sending bios down to the same device. This greatly
322 * improves the schedulers ability to collect and merge the bios.
324 * But, it also turns into a long list of bios to process and that is sure
325 * to eventually make the worker thread block. The solution here is to
326 * make some progress and then put this work struct back at the end of
327 * the list if the block device is congested. This way, multiple devices
328 * can make progress from a single worker thread.
330 static noinline void run_scheduled_bios(struct btrfs_device *device)
333 struct backing_dev_info *bdi;
334 struct btrfs_fs_info *fs_info;
335 struct btrfs_pending_bios *pending_bios;
339 unsigned long num_run;
340 unsigned long batch_run = 0;
342 unsigned long last_waited = 0;
344 int sync_pending = 0;
345 struct blk_plug plug;
348 * this function runs all the bios we've collected for
349 * a particular device. We don't want to wander off to
350 * another device without first sending all of these down.
351 * So, setup a plug here and finish it off before we return
353 blk_start_plug(&plug);
355 bdi = blk_get_backing_dev_info(device->bdev);
356 fs_info = device->dev_root->fs_info;
357 limit = btrfs_async_submit_limit(fs_info);
358 limit = limit * 2 / 3;
361 spin_lock(&device->io_lock);
366 /* take all the bios off the list at once and process them
367 * later on (without the lock held). But, remember the
368 * tail and other pointers so the bios can be properly reinserted
369 * into the list if we hit congestion
371 if (!force_reg && device->pending_sync_bios.head) {
372 pending_bios = &device->pending_sync_bios;
375 pending_bios = &device->pending_bios;
379 pending = pending_bios->head;
380 tail = pending_bios->tail;
381 WARN_ON(pending && !tail);
384 * if pending was null this time around, no bios need processing
385 * at all and we can stop. Otherwise it'll loop back up again
386 * and do an additional check so no bios are missed.
388 * device->running_pending is used to synchronize with the
391 if (device->pending_sync_bios.head == NULL &&
392 device->pending_bios.head == NULL) {
394 device->running_pending = 0;
397 device->running_pending = 1;
400 pending_bios->head = NULL;
401 pending_bios->tail = NULL;
403 spin_unlock(&device->io_lock);
408 /* we want to work on both lists, but do more bios on the
409 * sync list than the regular list
412 pending_bios != &device->pending_sync_bios &&
413 device->pending_sync_bios.head) ||
414 (num_run > 64 && pending_bios == &device->pending_sync_bios &&
415 device->pending_bios.head)) {
416 spin_lock(&device->io_lock);
417 requeue_list(pending_bios, pending, tail);
422 pending = pending->bi_next;
426 * atomic_dec_return implies a barrier for waitqueue_active
428 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
429 waitqueue_active(&fs_info->async_submit_wait))
430 wake_up(&fs_info->async_submit_wait);
432 BUG_ON(atomic_read(&cur->__bi_cnt) == 0);
435 * if we're doing the sync list, record that our
436 * plug has some sync requests on it
438 * If we're doing the regular list and there are
439 * sync requests sitting around, unplug before
442 if (pending_bios == &device->pending_sync_bios) {
444 } else if (sync_pending) {
445 blk_finish_plug(&plug);
446 blk_start_plug(&plug);
450 btrfsic_submit_bio(cur->bi_rw, cur);
457 * we made progress, there is more work to do and the bdi
458 * is now congested. Back off and let other work structs
461 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
462 fs_info->fs_devices->open_devices > 1) {
463 struct io_context *ioc;
465 ioc = current->io_context;
468 * the main goal here is that we don't want to
469 * block if we're going to be able to submit
470 * more requests without blocking.
472 * This code does two great things, it pokes into
473 * the elevator code from a filesystem _and_
474 * it makes assumptions about how batching works.
476 if (ioc && ioc->nr_batch_requests > 0 &&
477 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
479 ioc->last_waited == last_waited)) {
481 * we want to go through our batch of
482 * requests and stop. So, we copy out
483 * the ioc->last_waited time and test
484 * against it before looping
486 last_waited = ioc->last_waited;
490 spin_lock(&device->io_lock);
491 requeue_list(pending_bios, pending, tail);
492 device->running_pending = 1;
494 spin_unlock(&device->io_lock);
495 btrfs_queue_work(fs_info->submit_workers,
499 /* unplug every 64 requests just for good measure */
500 if (batch_run % 64 == 0) {
501 blk_finish_plug(&plug);
502 blk_start_plug(&plug);
511 spin_lock(&device->io_lock);
512 if (device->pending_bios.head || device->pending_sync_bios.head)
514 spin_unlock(&device->io_lock);
517 blk_finish_plug(&plug);
520 static void pending_bios_fn(struct btrfs_work *work)
522 struct btrfs_device *device;
524 device = container_of(work, struct btrfs_device, work);
525 run_scheduled_bios(device);
529 void btrfs_free_stale_device(struct btrfs_device *cur_dev)
531 struct btrfs_fs_devices *fs_devs;
532 struct btrfs_device *dev;
537 list_for_each_entry(fs_devs, &fs_uuids, list) {
542 if (fs_devs->seeding)
545 list_for_each_entry(dev, &fs_devs->devices, dev_list) {
553 * Todo: This won't be enough. What if the same device
554 * comes back (with new uuid and) with its mapper path?
555 * But for now, this does help as mostly an admin will
556 * either use mapper or non mapper path throughout.
559 del = strcmp(rcu_str_deref(dev->name),
560 rcu_str_deref(cur_dev->name));
567 /* delete the stale device */
568 if (fs_devs->num_devices == 1) {
569 btrfs_sysfs_remove_fsid(fs_devs);
570 list_del(&fs_devs->list);
571 free_fs_devices(fs_devs);
573 fs_devs->num_devices--;
574 list_del(&dev->dev_list);
575 rcu_string_free(dev->name);
584 * Add new device to list of registered devices
587 * 1 - first time device is seen
588 * 0 - device already known
591 static noinline int device_list_add(const char *path,
592 struct btrfs_super_block *disk_super,
593 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
595 struct btrfs_device *device;
596 struct btrfs_fs_devices *fs_devices;
597 struct rcu_string *name;
599 u64 found_transid = btrfs_super_generation(disk_super);
601 fs_devices = find_fsid(disk_super->fsid);
603 fs_devices = alloc_fs_devices(disk_super->fsid);
604 if (IS_ERR(fs_devices))
605 return PTR_ERR(fs_devices);
607 list_add(&fs_devices->list, &fs_uuids);
611 device = __find_device(&fs_devices->devices, devid,
612 disk_super->dev_item.uuid);
616 if (fs_devices->opened)
619 device = btrfs_alloc_device(NULL, &devid,
620 disk_super->dev_item.uuid);
621 if (IS_ERR(device)) {
622 /* we can safely leave the fs_devices entry around */
623 return PTR_ERR(device);
626 name = rcu_string_strdup(path, GFP_NOFS);
631 rcu_assign_pointer(device->name, name);
633 mutex_lock(&fs_devices->device_list_mutex);
634 list_add_rcu(&device->dev_list, &fs_devices->devices);
635 fs_devices->num_devices++;
636 mutex_unlock(&fs_devices->device_list_mutex);
639 device->fs_devices = fs_devices;
640 } else if (!device->name || strcmp(device->name->str, path)) {
642 * When FS is already mounted.
643 * 1. If you are here and if the device->name is NULL that
644 * means this device was missing at time of FS mount.
645 * 2. If you are here and if the device->name is different
646 * from 'path' that means either
647 * a. The same device disappeared and reappeared with
649 * b. The missing-disk-which-was-replaced, has
652 * We must allow 1 and 2a above. But 2b would be a spurious
655 * Further in case of 1 and 2a above, the disk at 'path'
656 * would have missed some transaction when it was away and
657 * in case of 2a the stale bdev has to be updated as well.
658 * 2b must not be allowed at all time.
662 * For now, we do allow update to btrfs_fs_device through the
663 * btrfs dev scan cli after FS has been mounted. We're still
664 * tracking a problem where systems fail mount by subvolume id
665 * when we reject replacement on a mounted FS.
667 if (!fs_devices->opened && found_transid < device->generation) {
669 * That is if the FS is _not_ mounted and if you
670 * are here, that means there is more than one
671 * disk with same uuid and devid.We keep the one
672 * with larger generation number or the last-in if
673 * generation are equal.
678 name = rcu_string_strdup(path, GFP_NOFS);
681 rcu_string_free(device->name);
682 rcu_assign_pointer(device->name, name);
683 if (device->missing) {
684 fs_devices->missing_devices--;
690 * Unmount does not free the btrfs_device struct but would zero
691 * generation along with most of the other members. So just update
692 * it back. We need it to pick the disk with largest generation
695 if (!fs_devices->opened)
696 device->generation = found_transid;
699 * if there is new btrfs on an already registered device,
700 * then remove the stale device entry.
702 btrfs_free_stale_device(device);
704 *fs_devices_ret = fs_devices;
709 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
711 struct btrfs_fs_devices *fs_devices;
712 struct btrfs_device *device;
713 struct btrfs_device *orig_dev;
715 fs_devices = alloc_fs_devices(orig->fsid);
716 if (IS_ERR(fs_devices))
719 mutex_lock(&orig->device_list_mutex);
720 fs_devices->total_devices = orig->total_devices;
722 /* We have held the volume lock, it is safe to get the devices. */
723 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
724 struct rcu_string *name;
726 device = btrfs_alloc_device(NULL, &orig_dev->devid,
732 * This is ok to do without rcu read locked because we hold the
733 * uuid mutex so nothing we touch in here is going to disappear.
735 if (orig_dev->name) {
736 name = rcu_string_strdup(orig_dev->name->str,
742 rcu_assign_pointer(device->name, name);
745 list_add(&device->dev_list, &fs_devices->devices);
746 device->fs_devices = fs_devices;
747 fs_devices->num_devices++;
749 mutex_unlock(&orig->device_list_mutex);
752 mutex_unlock(&orig->device_list_mutex);
753 free_fs_devices(fs_devices);
754 return ERR_PTR(-ENOMEM);
757 void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step)
759 struct btrfs_device *device, *next;
760 struct btrfs_device *latest_dev = NULL;
762 mutex_lock(&uuid_mutex);
764 /* This is the initialized path, it is safe to release the devices. */
765 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
766 if (device->in_fs_metadata) {
767 if (!device->is_tgtdev_for_dev_replace &&
769 device->generation > latest_dev->generation)) {
775 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
777 * In the first step, keep the device which has
778 * the correct fsid and the devid that is used
779 * for the dev_replace procedure.
780 * In the second step, the dev_replace state is
781 * read from the device tree and it is known
782 * whether the procedure is really active or
783 * not, which means whether this device is
784 * used or whether it should be removed.
786 if (step == 0 || device->is_tgtdev_for_dev_replace) {
791 blkdev_put(device->bdev, device->mode);
793 fs_devices->open_devices--;
795 if (device->writeable) {
796 list_del_init(&device->dev_alloc_list);
797 device->writeable = 0;
798 if (!device->is_tgtdev_for_dev_replace)
799 fs_devices->rw_devices--;
801 list_del_init(&device->dev_list);
802 fs_devices->num_devices--;
803 rcu_string_free(device->name);
807 if (fs_devices->seed) {
808 fs_devices = fs_devices->seed;
812 fs_devices->latest_bdev = latest_dev->bdev;
814 mutex_unlock(&uuid_mutex);
817 static void __free_device(struct work_struct *work)
819 struct btrfs_device *device;
821 device = container_of(work, struct btrfs_device, rcu_work);
824 blkdev_put(device->bdev, device->mode);
826 rcu_string_free(device->name);
830 static void free_device(struct rcu_head *head)
832 struct btrfs_device *device;
834 device = container_of(head, struct btrfs_device, rcu);
836 INIT_WORK(&device->rcu_work, __free_device);
837 schedule_work(&device->rcu_work);
840 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
842 struct btrfs_device *device, *tmp;
844 if (--fs_devices->opened > 0)
847 mutex_lock(&fs_devices->device_list_mutex);
848 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
849 btrfs_close_one_device(device);
851 mutex_unlock(&fs_devices->device_list_mutex);
853 WARN_ON(fs_devices->open_devices);
854 WARN_ON(fs_devices->rw_devices);
855 fs_devices->opened = 0;
856 fs_devices->seeding = 0;
861 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
863 struct btrfs_fs_devices *seed_devices = NULL;
866 mutex_lock(&uuid_mutex);
867 ret = __btrfs_close_devices(fs_devices);
868 if (!fs_devices->opened) {
869 seed_devices = fs_devices->seed;
870 fs_devices->seed = NULL;
872 mutex_unlock(&uuid_mutex);
874 while (seed_devices) {
875 fs_devices = seed_devices;
876 seed_devices = fs_devices->seed;
877 __btrfs_close_devices(fs_devices);
878 free_fs_devices(fs_devices);
881 * Wait for rcu kworkers under __btrfs_close_devices
882 * to finish all blkdev_puts so device is really
883 * free when umount is done.
889 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
890 fmode_t flags, void *holder)
892 struct request_queue *q;
893 struct block_device *bdev;
894 struct list_head *head = &fs_devices->devices;
895 struct btrfs_device *device;
896 struct btrfs_device *latest_dev = NULL;
897 struct buffer_head *bh;
898 struct btrfs_super_block *disk_super;
905 list_for_each_entry(device, head, dev_list) {
911 /* Just open everything we can; ignore failures here */
912 if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
916 disk_super = (struct btrfs_super_block *)bh->b_data;
917 devid = btrfs_stack_device_id(&disk_super->dev_item);
918 if (devid != device->devid)
921 if (memcmp(device->uuid, disk_super->dev_item.uuid,
925 device->generation = btrfs_super_generation(disk_super);
927 device->generation > latest_dev->generation)
930 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
931 device->writeable = 0;
933 device->writeable = !bdev_read_only(bdev);
937 q = bdev_get_queue(bdev);
938 if (blk_queue_discard(q))
939 device->can_discard = 1;
942 device->in_fs_metadata = 0;
943 device->mode = flags;
945 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
946 fs_devices->rotating = 1;
948 fs_devices->open_devices++;
949 if (device->writeable &&
950 device->devid != BTRFS_DEV_REPLACE_DEVID) {
951 fs_devices->rw_devices++;
952 list_add(&device->dev_alloc_list,
953 &fs_devices->alloc_list);
960 blkdev_put(bdev, flags);
963 if (fs_devices->open_devices == 0) {
967 fs_devices->seeding = seeding;
968 fs_devices->opened = 1;
969 fs_devices->latest_bdev = latest_dev->bdev;
970 fs_devices->total_rw_bytes = 0;
975 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
976 fmode_t flags, void *holder)
980 mutex_lock(&uuid_mutex);
981 if (fs_devices->opened) {
982 fs_devices->opened++;
985 ret = __btrfs_open_devices(fs_devices, flags, holder);
987 mutex_unlock(&uuid_mutex);
992 * Look for a btrfs signature on a device. This may be called out of the mount path
993 * and we are not allowed to call set_blocksize during the scan. The superblock
994 * is read via pagecache
996 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
997 struct btrfs_fs_devices **fs_devices_ret)
999 struct btrfs_super_block *disk_super;
1000 struct block_device *bdev;
1011 * we would like to check all the supers, but that would make
1012 * a btrfs mount succeed after a mkfs from a different FS.
1013 * So, we need to add a special mount option to scan for
1014 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1016 bytenr = btrfs_sb_offset(0);
1017 flags |= FMODE_EXCL;
1018 mutex_lock(&uuid_mutex);
1020 bdev = blkdev_get_by_path(path, flags, holder);
1023 ret = PTR_ERR(bdev);
1027 /* make sure our super fits in the device */
1028 if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1029 goto error_bdev_put;
1031 /* make sure our super fits in the page */
1032 if (sizeof(*disk_super) > PAGE_SIZE)
1033 goto error_bdev_put;
1035 /* make sure our super doesn't straddle pages on disk */
1036 index = bytenr >> PAGE_SHIFT;
1037 if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_SHIFT != index)
1038 goto error_bdev_put;
1040 /* pull in the page with our super */
1041 page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
1044 if (IS_ERR_OR_NULL(page))
1045 goto error_bdev_put;
1049 /* align our pointer to the offset of the super block */
1050 disk_super = p + (bytenr & ~PAGE_MASK);
1052 if (btrfs_super_bytenr(disk_super) != bytenr ||
1053 btrfs_super_magic(disk_super) != BTRFS_MAGIC)
1056 devid = btrfs_stack_device_id(&disk_super->dev_item);
1057 transid = btrfs_super_generation(disk_super);
1058 total_devices = btrfs_super_num_devices(disk_super);
1060 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
1062 if (disk_super->label[0]) {
1063 if (disk_super->label[BTRFS_LABEL_SIZE - 1])
1064 disk_super->label[BTRFS_LABEL_SIZE - 1] = '\0';
1065 printk(KERN_INFO "BTRFS: device label %s ", disk_super->label);
1067 printk(KERN_INFO "BTRFS: device fsid %pU ", disk_super->fsid);
1070 printk(KERN_CONT "devid %llu transid %llu %s\n", devid, transid, path);
1073 if (!ret && fs_devices_ret)
1074 (*fs_devices_ret)->total_devices = total_devices;
1081 blkdev_put(bdev, flags);
1083 mutex_unlock(&uuid_mutex);
1087 /* helper to account the used device space in the range */
1088 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
1089 u64 end, u64 *length)
1091 struct btrfs_key key;
1092 struct btrfs_root *root = device->dev_root;
1093 struct btrfs_dev_extent *dev_extent;
1094 struct btrfs_path *path;
1098 struct extent_buffer *l;
1102 if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
1105 path = btrfs_alloc_path();
1108 path->reada = READA_FORWARD;
1110 key.objectid = device->devid;
1112 key.type = BTRFS_DEV_EXTENT_KEY;
1114 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1118 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1125 slot = path->slots[0];
1126 if (slot >= btrfs_header_nritems(l)) {
1127 ret = btrfs_next_leaf(root, path);
1135 btrfs_item_key_to_cpu(l, &key, slot);
1137 if (key.objectid < device->devid)
1140 if (key.objectid > device->devid)
1143 if (key.type != BTRFS_DEV_EXTENT_KEY)
1146 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1147 extent_end = key.offset + btrfs_dev_extent_length(l,
1149 if (key.offset <= start && extent_end > end) {
1150 *length = end - start + 1;
1152 } else if (key.offset <= start && extent_end > start)
1153 *length += extent_end - start;
1154 else if (key.offset > start && extent_end <= end)
1155 *length += extent_end - key.offset;
1156 else if (key.offset > start && key.offset <= end) {
1157 *length += end - key.offset + 1;
1159 } else if (key.offset > end)
1167 btrfs_free_path(path);
1171 static int contains_pending_extent(struct btrfs_transaction *transaction,
1172 struct btrfs_device *device,
1173 u64 *start, u64 len)
1175 struct btrfs_fs_info *fs_info = device->dev_root->fs_info;
1176 struct extent_map *em;
1177 struct list_head *search_list = &fs_info->pinned_chunks;
1179 u64 physical_start = *start;
1182 search_list = &transaction->pending_chunks;
1184 list_for_each_entry(em, search_list, list) {
1185 struct map_lookup *map;
1188 map = em->map_lookup;
1189 for (i = 0; i < map->num_stripes; i++) {
1192 if (map->stripes[i].dev != device)
1194 if (map->stripes[i].physical >= physical_start + len ||
1195 map->stripes[i].physical + em->orig_block_len <=
1199 * Make sure that while processing the pinned list we do
1200 * not override our *start with a lower value, because
1201 * we can have pinned chunks that fall within this
1202 * device hole and that have lower physical addresses
1203 * than the pending chunks we processed before. If we
1204 * do not take this special care we can end up getting
1205 * 2 pending chunks that start at the same physical
1206 * device offsets because the end offset of a pinned
1207 * chunk can be equal to the start offset of some
1210 end = map->stripes[i].physical + em->orig_block_len;
1217 if (search_list != &fs_info->pinned_chunks) {
1218 search_list = &fs_info->pinned_chunks;
1227 * find_free_dev_extent_start - find free space in the specified device
1228 * @device: the device which we search the free space in
1229 * @num_bytes: the size of the free space that we need
1230 * @search_start: the position from which to begin the search
1231 * @start: store the start of the free space.
1232 * @len: the size of the free space. that we find, or the size
1233 * of the max free space if we don't find suitable free space
1235 * this uses a pretty simple search, the expectation is that it is
1236 * called very infrequently and that a given device has a small number
1239 * @start is used to store the start of the free space if we find. But if we
1240 * don't find suitable free space, it will be used to store the start position
1241 * of the max free space.
1243 * @len is used to store the size of the free space that we find.
1244 * But if we don't find suitable free space, it is used to store the size of
1245 * the max free space.
1247 int find_free_dev_extent_start(struct btrfs_transaction *transaction,
1248 struct btrfs_device *device, u64 num_bytes,
1249 u64 search_start, u64 *start, u64 *len)
1251 struct btrfs_key key;
1252 struct btrfs_root *root = device->dev_root;
1253 struct btrfs_dev_extent *dev_extent;
1254 struct btrfs_path *path;
1259 u64 search_end = device->total_bytes;
1262 struct extent_buffer *l;
1263 u64 min_search_start;
1266 * We don't want to overwrite the superblock on the drive nor any area
1267 * used by the boot loader (grub for example), so we make sure to start
1268 * at an offset of at least 1MB.
1270 min_search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
1271 search_start = max(search_start, min_search_start);
1273 path = btrfs_alloc_path();
1277 max_hole_start = search_start;
1281 if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1286 path->reada = READA_FORWARD;
1287 path->search_commit_root = 1;
1288 path->skip_locking = 1;
1290 key.objectid = device->devid;
1291 key.offset = search_start;
1292 key.type = BTRFS_DEV_EXTENT_KEY;
1294 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1298 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1305 slot = path->slots[0];
1306 if (slot >= btrfs_header_nritems(l)) {
1307 ret = btrfs_next_leaf(root, path);
1315 btrfs_item_key_to_cpu(l, &key, slot);
1317 if (key.objectid < device->devid)
1320 if (key.objectid > device->devid)
1323 if (key.type != BTRFS_DEV_EXTENT_KEY)
1326 if (key.offset > search_start) {
1327 hole_size = key.offset - search_start;
1330 * Have to check before we set max_hole_start, otherwise
1331 * we could end up sending back this offset anyway.
1333 if (contains_pending_extent(transaction, device,
1336 if (key.offset >= search_start) {
1337 hole_size = key.offset - search_start;
1344 if (hole_size > max_hole_size) {
1345 max_hole_start = search_start;
1346 max_hole_size = hole_size;
1350 * If this free space is greater than which we need,
1351 * it must be the max free space that we have found
1352 * until now, so max_hole_start must point to the start
1353 * of this free space and the length of this free space
1354 * is stored in max_hole_size. Thus, we return
1355 * max_hole_start and max_hole_size and go back to the
1358 if (hole_size >= num_bytes) {
1364 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1365 extent_end = key.offset + btrfs_dev_extent_length(l,
1367 if (extent_end > search_start)
1368 search_start = extent_end;
1375 * At this point, search_start should be the end of
1376 * allocated dev extents, and when shrinking the device,
1377 * search_end may be smaller than search_start.
1379 if (search_end > search_start) {
1380 hole_size = search_end - search_start;
1382 if (contains_pending_extent(transaction, device, &search_start,
1384 btrfs_release_path(path);
1388 if (hole_size > max_hole_size) {
1389 max_hole_start = search_start;
1390 max_hole_size = hole_size;
1395 if (max_hole_size < num_bytes)
1401 btrfs_free_path(path);
1402 *start = max_hole_start;
1404 *len = max_hole_size;
1408 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1409 struct btrfs_device *device, u64 num_bytes,
1410 u64 *start, u64 *len)
1412 /* FIXME use last free of some kind */
1413 return find_free_dev_extent_start(trans->transaction, device,
1414 num_bytes, 0, start, len);
1417 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1418 struct btrfs_device *device,
1419 u64 start, u64 *dev_extent_len)
1422 struct btrfs_path *path;
1423 struct btrfs_root *root = device->dev_root;
1424 struct btrfs_key key;
1425 struct btrfs_key found_key;
1426 struct extent_buffer *leaf = NULL;
1427 struct btrfs_dev_extent *extent = NULL;
1429 path = btrfs_alloc_path();
1433 key.objectid = device->devid;
1435 key.type = BTRFS_DEV_EXTENT_KEY;
1437 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1439 ret = btrfs_previous_item(root, path, key.objectid,
1440 BTRFS_DEV_EXTENT_KEY);
1443 leaf = path->nodes[0];
1444 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1445 extent = btrfs_item_ptr(leaf, path->slots[0],
1446 struct btrfs_dev_extent);
1447 BUG_ON(found_key.offset > start || found_key.offset +
1448 btrfs_dev_extent_length(leaf, extent) < start);
1450 btrfs_release_path(path);
1452 } else if (ret == 0) {
1453 leaf = path->nodes[0];
1454 extent = btrfs_item_ptr(leaf, path->slots[0],
1455 struct btrfs_dev_extent);
1457 btrfs_std_error(root->fs_info, ret, "Slot search failed");
1461 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1463 ret = btrfs_del_item(trans, root, path);
1465 btrfs_std_error(root->fs_info, ret,
1466 "Failed to remove dev extent item");
1468 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1471 btrfs_free_path(path);
1475 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1476 struct btrfs_device *device,
1477 u64 chunk_tree, u64 chunk_objectid,
1478 u64 chunk_offset, u64 start, u64 num_bytes)
1481 struct btrfs_path *path;
1482 struct btrfs_root *root = device->dev_root;
1483 struct btrfs_dev_extent *extent;
1484 struct extent_buffer *leaf;
1485 struct btrfs_key key;
1487 WARN_ON(!device->in_fs_metadata);
1488 WARN_ON(device->is_tgtdev_for_dev_replace);
1489 path = btrfs_alloc_path();
1493 key.objectid = device->devid;
1495 key.type = BTRFS_DEV_EXTENT_KEY;
1496 ret = btrfs_insert_empty_item(trans, root, path, &key,
1501 leaf = path->nodes[0];
1502 extent = btrfs_item_ptr(leaf, path->slots[0],
1503 struct btrfs_dev_extent);
1504 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1505 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1506 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1508 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1509 btrfs_dev_extent_chunk_tree_uuid(extent), BTRFS_UUID_SIZE);
1511 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1512 btrfs_mark_buffer_dirty(leaf);
1514 btrfs_free_path(path);
1518 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1520 struct extent_map_tree *em_tree;
1521 struct extent_map *em;
1525 em_tree = &fs_info->mapping_tree.map_tree;
1526 read_lock(&em_tree->lock);
1527 n = rb_last(&em_tree->map);
1529 em = rb_entry(n, struct extent_map, rb_node);
1530 ret = em->start + em->len;
1532 read_unlock(&em_tree->lock);
1537 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1541 struct btrfs_key key;
1542 struct btrfs_key found_key;
1543 struct btrfs_path *path;
1545 path = btrfs_alloc_path();
1549 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1550 key.type = BTRFS_DEV_ITEM_KEY;
1551 key.offset = (u64)-1;
1553 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1557 BUG_ON(ret == 0); /* Corruption */
1559 ret = btrfs_previous_item(fs_info->chunk_root, path,
1560 BTRFS_DEV_ITEMS_OBJECTID,
1561 BTRFS_DEV_ITEM_KEY);
1565 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1567 *devid_ret = found_key.offset + 1;
1571 btrfs_free_path(path);
1576 * the device information is stored in the chunk root
1577 * the btrfs_device struct should be fully filled in
1579 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1580 struct btrfs_root *root,
1581 struct btrfs_device *device)
1584 struct btrfs_path *path;
1585 struct btrfs_dev_item *dev_item;
1586 struct extent_buffer *leaf;
1587 struct btrfs_key key;
1590 root = root->fs_info->chunk_root;
1592 path = btrfs_alloc_path();
1596 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1597 key.type = BTRFS_DEV_ITEM_KEY;
1598 key.offset = device->devid;
1600 ret = btrfs_insert_empty_item(trans, root, path, &key,
1605 leaf = path->nodes[0];
1606 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1608 btrfs_set_device_id(leaf, dev_item, device->devid);
1609 btrfs_set_device_generation(leaf, dev_item, 0);
1610 btrfs_set_device_type(leaf, dev_item, device->type);
1611 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1612 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1613 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1614 btrfs_set_device_total_bytes(leaf, dev_item,
1615 btrfs_device_get_disk_total_bytes(device));
1616 btrfs_set_device_bytes_used(leaf, dev_item,
1617 btrfs_device_get_bytes_used(device));
1618 btrfs_set_device_group(leaf, dev_item, 0);
1619 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1620 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1621 btrfs_set_device_start_offset(leaf, dev_item, 0);
1623 ptr = btrfs_device_uuid(dev_item);
1624 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1625 ptr = btrfs_device_fsid(dev_item);
1626 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1627 btrfs_mark_buffer_dirty(leaf);
1631 btrfs_free_path(path);
1636 * Function to update ctime/mtime for a given device path.
1637 * Mainly used for ctime/mtime based probe like libblkid.
1639 static void update_dev_time(char *path_name)
1643 filp = filp_open(path_name, O_RDWR, 0);
1646 file_update_time(filp);
1647 filp_close(filp, NULL);
1650 static int btrfs_rm_dev_item(struct btrfs_root *root,
1651 struct btrfs_device *device)
1654 struct btrfs_path *path;
1655 struct btrfs_key key;
1656 struct btrfs_trans_handle *trans;
1658 root = root->fs_info->chunk_root;
1660 path = btrfs_alloc_path();
1664 trans = btrfs_start_transaction(root, 0);
1665 if (IS_ERR(trans)) {
1666 btrfs_free_path(path);
1667 return PTR_ERR(trans);
1669 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1670 key.type = BTRFS_DEV_ITEM_KEY;
1671 key.offset = device->devid;
1673 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1682 ret = btrfs_del_item(trans, root, path);
1686 btrfs_free_path(path);
1687 btrfs_commit_transaction(trans, root);
1691 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1693 struct btrfs_device *device;
1694 struct btrfs_device *next_device;
1695 struct block_device *bdev;
1696 struct buffer_head *bh = NULL;
1697 struct btrfs_super_block *disk_super;
1698 struct btrfs_fs_devices *cur_devices;
1705 bool clear_super = false;
1707 mutex_lock(&uuid_mutex);
1710 seq = read_seqbegin(&root->fs_info->profiles_lock);
1712 all_avail = root->fs_info->avail_data_alloc_bits |
1713 root->fs_info->avail_system_alloc_bits |
1714 root->fs_info->avail_metadata_alloc_bits;
1715 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
1717 num_devices = root->fs_info->fs_devices->num_devices;
1718 btrfs_dev_replace_lock(&root->fs_info->dev_replace, 0);
1719 if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1720 WARN_ON(num_devices < 1);
1723 btrfs_dev_replace_unlock(&root->fs_info->dev_replace, 0);
1725 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && num_devices <= 4) {
1726 ret = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET;
1730 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && num_devices <= 2) {
1731 ret = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET;
1735 if ((all_avail & BTRFS_BLOCK_GROUP_RAID5) &&
1736 root->fs_info->fs_devices->rw_devices <= 2) {
1737 ret = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET;
1740 if ((all_avail & BTRFS_BLOCK_GROUP_RAID6) &&
1741 root->fs_info->fs_devices->rw_devices <= 3) {
1742 ret = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET;
1746 if (strcmp(device_path, "missing") == 0) {
1747 struct list_head *devices;
1748 struct btrfs_device *tmp;
1751 devices = &root->fs_info->fs_devices->devices;
1753 * It is safe to read the devices since the volume_mutex
1756 list_for_each_entry(tmp, devices, dev_list) {
1757 if (tmp->in_fs_metadata &&
1758 !tmp->is_tgtdev_for_dev_replace &&
1768 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
1772 ret = btrfs_get_bdev_and_sb(device_path,
1773 FMODE_WRITE | FMODE_EXCL,
1774 root->fs_info->bdev_holder, 0,
1778 disk_super = (struct btrfs_super_block *)bh->b_data;
1779 devid = btrfs_stack_device_id(&disk_super->dev_item);
1780 dev_uuid = disk_super->dev_item.uuid;
1781 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1789 if (device->is_tgtdev_for_dev_replace) {
1790 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1794 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1795 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1799 if (device->writeable) {
1801 list_del_init(&device->dev_alloc_list);
1802 device->fs_devices->rw_devices--;
1803 unlock_chunks(root);
1807 mutex_unlock(&uuid_mutex);
1808 ret = btrfs_shrink_device(device, 0);
1809 mutex_lock(&uuid_mutex);
1814 * TODO: the superblock still includes this device in its num_devices
1815 * counter although write_all_supers() is not locked out. This
1816 * could give a filesystem state which requires a degraded mount.
1818 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1822 device->in_fs_metadata = 0;
1823 btrfs_scrub_cancel_dev(root->fs_info, device);
1826 * the device list mutex makes sure that we don't change
1827 * the device list while someone else is writing out all
1828 * the device supers. Whoever is writing all supers, should
1829 * lock the device list mutex before getting the number of
1830 * devices in the super block (super_copy). Conversely,
1831 * whoever updates the number of devices in the super block
1832 * (super_copy) should hold the device list mutex.
1835 cur_devices = device->fs_devices;
1836 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1837 list_del_rcu(&device->dev_list);
1839 device->fs_devices->num_devices--;
1840 device->fs_devices->total_devices--;
1842 if (device->missing)
1843 device->fs_devices->missing_devices--;
1845 next_device = list_entry(root->fs_info->fs_devices->devices.next,
1846 struct btrfs_device, dev_list);
1847 if (device->bdev == root->fs_info->sb->s_bdev)
1848 root->fs_info->sb->s_bdev = next_device->bdev;
1849 if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1850 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1853 device->fs_devices->open_devices--;
1854 /* remove sysfs entry */
1855 btrfs_sysfs_rm_device_link(root->fs_info->fs_devices, device);
1858 call_rcu(&device->rcu, free_device);
1860 num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1861 btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1862 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1864 if (cur_devices->open_devices == 0) {
1865 struct btrfs_fs_devices *fs_devices;
1866 fs_devices = root->fs_info->fs_devices;
1867 while (fs_devices) {
1868 if (fs_devices->seed == cur_devices) {
1869 fs_devices->seed = cur_devices->seed;
1872 fs_devices = fs_devices->seed;
1874 cur_devices->seed = NULL;
1875 __btrfs_close_devices(cur_devices);
1876 free_fs_devices(cur_devices);
1879 root->fs_info->num_tolerated_disk_barrier_failures =
1880 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1883 * at this point, the device is zero sized. We want to
1884 * remove it from the devices list and zero out the old super
1886 if (clear_super && disk_super) {
1890 /* make sure this device isn't detected as part of
1893 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1894 set_buffer_dirty(bh);
1895 sync_dirty_buffer(bh);
1897 /* clear the mirror copies of super block on the disk
1898 * being removed, 0th copy is been taken care above and
1899 * the below would take of the rest
1901 for (i = 1; i < BTRFS_SUPER_MIRROR_MAX; i++) {
1902 bytenr = btrfs_sb_offset(i);
1903 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
1904 i_size_read(bdev->bd_inode))
1908 bh = __bread(bdev, bytenr / 4096,
1909 BTRFS_SUPER_INFO_SIZE);
1913 disk_super = (struct btrfs_super_block *)bh->b_data;
1915 if (btrfs_super_bytenr(disk_super) != bytenr ||
1916 btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
1919 memset(&disk_super->magic, 0,
1920 sizeof(disk_super->magic));
1921 set_buffer_dirty(bh);
1922 sync_dirty_buffer(bh);
1929 /* Notify udev that device has changed */
1930 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
1932 /* Update ctime/mtime for device path for libblkid */
1933 update_dev_time(device_path);
1939 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1941 mutex_unlock(&uuid_mutex);
1944 if (device->writeable) {
1946 list_add(&device->dev_alloc_list,
1947 &root->fs_info->fs_devices->alloc_list);
1948 device->fs_devices->rw_devices++;
1949 unlock_chunks(root);
1954 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
1955 struct btrfs_device *srcdev)
1957 struct btrfs_fs_devices *fs_devices;
1959 WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1962 * in case of fs with no seed, srcdev->fs_devices will point
1963 * to fs_devices of fs_info. However when the dev being replaced is
1964 * a seed dev it will point to the seed's local fs_devices. In short
1965 * srcdev will have its correct fs_devices in both the cases.
1967 fs_devices = srcdev->fs_devices;
1969 list_del_rcu(&srcdev->dev_list);
1970 list_del_rcu(&srcdev->dev_alloc_list);
1971 fs_devices->num_devices--;
1972 if (srcdev->missing)
1973 fs_devices->missing_devices--;
1975 if (srcdev->writeable) {
1976 fs_devices->rw_devices--;
1977 /* zero out the old super if it is writable */
1978 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
1982 fs_devices->open_devices--;
1985 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
1986 struct btrfs_device *srcdev)
1988 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
1990 call_rcu(&srcdev->rcu, free_device);
1993 * unless fs_devices is seed fs, num_devices shouldn't go
1996 BUG_ON(!fs_devices->num_devices && !fs_devices->seeding);
1998 /* if this is no devs we rather delete the fs_devices */
1999 if (!fs_devices->num_devices) {
2000 struct btrfs_fs_devices *tmp_fs_devices;
2002 tmp_fs_devices = fs_info->fs_devices;
2003 while (tmp_fs_devices) {
2004 if (tmp_fs_devices->seed == fs_devices) {
2005 tmp_fs_devices->seed = fs_devices->seed;
2008 tmp_fs_devices = tmp_fs_devices->seed;
2010 fs_devices->seed = NULL;
2011 __btrfs_close_devices(fs_devices);
2012 free_fs_devices(fs_devices);
2016 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2017 struct btrfs_device *tgtdev)
2019 struct btrfs_device *next_device;
2021 mutex_lock(&uuid_mutex);
2023 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2025 btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev);
2028 btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
2029 fs_info->fs_devices->open_devices--;
2031 fs_info->fs_devices->num_devices--;
2033 next_device = list_entry(fs_info->fs_devices->devices.next,
2034 struct btrfs_device, dev_list);
2035 if (tgtdev->bdev == fs_info->sb->s_bdev)
2036 fs_info->sb->s_bdev = next_device->bdev;
2037 if (tgtdev->bdev == fs_info->fs_devices->latest_bdev)
2038 fs_info->fs_devices->latest_bdev = next_device->bdev;
2039 list_del_rcu(&tgtdev->dev_list);
2041 call_rcu(&tgtdev->rcu, free_device);
2043 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2044 mutex_unlock(&uuid_mutex);
2047 static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
2048 struct btrfs_device **device)
2051 struct btrfs_super_block *disk_super;
2054 struct block_device *bdev;
2055 struct buffer_head *bh;
2058 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2059 root->fs_info->bdev_holder, 0, &bdev, &bh);
2062 disk_super = (struct btrfs_super_block *)bh->b_data;
2063 devid = btrfs_stack_device_id(&disk_super->dev_item);
2064 dev_uuid = disk_super->dev_item.uuid;
2065 *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2070 blkdev_put(bdev, FMODE_READ);
2074 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
2076 struct btrfs_device **device)
2079 if (strcmp(device_path, "missing") == 0) {
2080 struct list_head *devices;
2081 struct btrfs_device *tmp;
2083 devices = &root->fs_info->fs_devices->devices;
2085 * It is safe to read the devices since the volume_mutex
2086 * is held by the caller.
2088 list_for_each_entry(tmp, devices, dev_list) {
2089 if (tmp->in_fs_metadata && !tmp->bdev) {
2096 return BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2100 return btrfs_find_device_by_path(root, device_path, device);
2105 * does all the dirty work required for changing file system's UUID.
2107 static int btrfs_prepare_sprout(struct btrfs_root *root)
2109 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2110 struct btrfs_fs_devices *old_devices;
2111 struct btrfs_fs_devices *seed_devices;
2112 struct btrfs_super_block *disk_super = root->fs_info->super_copy;
2113 struct btrfs_device *device;
2116 BUG_ON(!mutex_is_locked(&uuid_mutex));
2117 if (!fs_devices->seeding)
2120 seed_devices = __alloc_fs_devices();
2121 if (IS_ERR(seed_devices))
2122 return PTR_ERR(seed_devices);
2124 old_devices = clone_fs_devices(fs_devices);
2125 if (IS_ERR(old_devices)) {
2126 kfree(seed_devices);
2127 return PTR_ERR(old_devices);
2130 list_add(&old_devices->list, &fs_uuids);
2132 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2133 seed_devices->opened = 1;
2134 INIT_LIST_HEAD(&seed_devices->devices);
2135 INIT_LIST_HEAD(&seed_devices->alloc_list);
2136 mutex_init(&seed_devices->device_list_mutex);
2138 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2139 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2141 list_for_each_entry(device, &seed_devices->devices, dev_list)
2142 device->fs_devices = seed_devices;
2145 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2146 unlock_chunks(root);
2148 fs_devices->seeding = 0;
2149 fs_devices->num_devices = 0;
2150 fs_devices->open_devices = 0;
2151 fs_devices->missing_devices = 0;
2152 fs_devices->rotating = 0;
2153 fs_devices->seed = seed_devices;
2155 generate_random_uuid(fs_devices->fsid);
2156 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2157 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2158 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2160 super_flags = btrfs_super_flags(disk_super) &
2161 ~BTRFS_SUPER_FLAG_SEEDING;
2162 btrfs_set_super_flags(disk_super, super_flags);
2168 * strore the expected generation for seed devices in device items.
2170 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2171 struct btrfs_root *root)
2173 struct btrfs_path *path;
2174 struct extent_buffer *leaf;
2175 struct btrfs_dev_item *dev_item;
2176 struct btrfs_device *device;
2177 struct btrfs_key key;
2178 u8 fs_uuid[BTRFS_UUID_SIZE];
2179 u8 dev_uuid[BTRFS_UUID_SIZE];
2183 path = btrfs_alloc_path();
2187 root = root->fs_info->chunk_root;
2188 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2190 key.type = BTRFS_DEV_ITEM_KEY;
2193 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2197 leaf = path->nodes[0];
2199 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2200 ret = btrfs_next_leaf(root, path);
2205 leaf = path->nodes[0];
2206 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2207 btrfs_release_path(path);
2211 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2212 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2213 key.type != BTRFS_DEV_ITEM_KEY)
2216 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2217 struct btrfs_dev_item);
2218 devid = btrfs_device_id(leaf, dev_item);
2219 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2221 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2223 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2225 BUG_ON(!device); /* Logic error */
2227 if (device->fs_devices->seeding) {
2228 btrfs_set_device_generation(leaf, dev_item,
2229 device->generation);
2230 btrfs_mark_buffer_dirty(leaf);
2238 btrfs_free_path(path);
2242 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
2244 struct request_queue *q;
2245 struct btrfs_trans_handle *trans;
2246 struct btrfs_device *device;
2247 struct block_device *bdev;
2248 struct list_head *devices;
2249 struct super_block *sb = root->fs_info->sb;
2250 struct rcu_string *name;
2252 int seeding_dev = 0;
2255 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
2258 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2259 root->fs_info->bdev_holder);
2261 return PTR_ERR(bdev);
2263 if (root->fs_info->fs_devices->seeding) {
2265 down_write(&sb->s_umount);
2266 mutex_lock(&uuid_mutex);
2269 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2271 devices = &root->fs_info->fs_devices->devices;
2273 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2274 list_for_each_entry(device, devices, dev_list) {
2275 if (device->bdev == bdev) {
2278 &root->fs_info->fs_devices->device_list_mutex);
2282 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2284 device = btrfs_alloc_device(root->fs_info, NULL, NULL);
2285 if (IS_ERR(device)) {
2286 /* we can safely leave the fs_devices entry around */
2287 ret = PTR_ERR(device);
2291 name = rcu_string_strdup(device_path, GFP_KERNEL);
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 doesn't
2753 * do anything we still error out.
2757 free_extent_map(em);
2760 map = em->map_lookup;
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_trans_remove_block_group(root->fs_info,
2858 if (IS_ERR(trans)) {
2859 ret = PTR_ERR(trans);
2860 btrfs_std_error(root->fs_info, ret, NULL);
2865 * step two, delete the device extents and the
2866 * chunk tree entries
2868 ret = btrfs_remove_chunk(trans, root, chunk_offset);
2869 btrfs_end_transaction(trans, root);
2873 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2875 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2876 struct btrfs_path *path;
2877 struct extent_buffer *leaf;
2878 struct btrfs_chunk *chunk;
2879 struct btrfs_key key;
2880 struct btrfs_key found_key;
2882 bool retried = false;
2886 path = btrfs_alloc_path();
2891 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2892 key.offset = (u64)-1;
2893 key.type = BTRFS_CHUNK_ITEM_KEY;
2896 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
2897 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2899 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2902 BUG_ON(ret == 0); /* Corruption */
2904 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2907 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2913 leaf = path->nodes[0];
2914 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2916 chunk = btrfs_item_ptr(leaf, path->slots[0],
2917 struct btrfs_chunk);
2918 chunk_type = btrfs_chunk_type(leaf, chunk);
2919 btrfs_release_path(path);
2921 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2922 ret = btrfs_relocate_chunk(chunk_root,
2929 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2931 if (found_key.offset == 0)
2933 key.offset = found_key.offset - 1;
2936 if (failed && !retried) {
2940 } else if (WARN_ON(failed && retried)) {
2944 btrfs_free_path(path);
2948 static int insert_balance_item(struct btrfs_root *root,
2949 struct btrfs_balance_control *bctl)
2951 struct btrfs_trans_handle *trans;
2952 struct btrfs_balance_item *item;
2953 struct btrfs_disk_balance_args disk_bargs;
2954 struct btrfs_path *path;
2955 struct extent_buffer *leaf;
2956 struct btrfs_key key;
2959 path = btrfs_alloc_path();
2963 trans = btrfs_start_transaction(root, 0);
2964 if (IS_ERR(trans)) {
2965 btrfs_free_path(path);
2966 return PTR_ERR(trans);
2969 key.objectid = BTRFS_BALANCE_OBJECTID;
2970 key.type = BTRFS_TEMPORARY_ITEM_KEY;
2973 ret = btrfs_insert_empty_item(trans, root, path, &key,
2978 leaf = path->nodes[0];
2979 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2981 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2983 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2984 btrfs_set_balance_data(leaf, item, &disk_bargs);
2985 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2986 btrfs_set_balance_meta(leaf, item, &disk_bargs);
2987 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2988 btrfs_set_balance_sys(leaf, item, &disk_bargs);
2990 btrfs_set_balance_flags(leaf, item, bctl->flags);
2992 btrfs_mark_buffer_dirty(leaf);
2994 btrfs_free_path(path);
2995 err = btrfs_commit_transaction(trans, root);
3001 static int del_balance_item(struct btrfs_root *root)
3003 struct btrfs_trans_handle *trans;
3004 struct btrfs_path *path;
3005 struct btrfs_key key;
3008 path = btrfs_alloc_path();
3012 trans = btrfs_start_transaction(root, 0);
3013 if (IS_ERR(trans)) {
3014 btrfs_free_path(path);
3015 return PTR_ERR(trans);
3018 key.objectid = BTRFS_BALANCE_OBJECTID;
3019 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3022 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3030 ret = btrfs_del_item(trans, root, path);
3032 btrfs_free_path(path);
3033 err = btrfs_commit_transaction(trans, root);
3040 * This is a heuristic used to reduce the number of chunks balanced on
3041 * resume after balance was interrupted.
3043 static void update_balance_args(struct btrfs_balance_control *bctl)
3046 * Turn on soft mode for chunk types that were being converted.
3048 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3049 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3050 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3051 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3052 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3053 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3056 * Turn on usage filter if is not already used. The idea is
3057 * that chunks that we have already balanced should be
3058 * reasonably full. Don't do it for chunks that are being
3059 * converted - that will keep us from relocating unconverted
3060 * (albeit full) chunks.
3062 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3063 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3064 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3065 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3066 bctl->data.usage = 90;
3068 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3069 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3070 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3071 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3072 bctl->sys.usage = 90;
3074 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3075 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3076 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3077 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3078 bctl->meta.usage = 90;
3083 * Should be called with both balance and volume mutexes held to
3084 * serialize other volume operations (add_dev/rm_dev/resize) with
3085 * restriper. Same goes for unset_balance_control.
3087 static void set_balance_control(struct btrfs_balance_control *bctl)
3089 struct btrfs_fs_info *fs_info = bctl->fs_info;
3091 BUG_ON(fs_info->balance_ctl);
3093 spin_lock(&fs_info->balance_lock);
3094 fs_info->balance_ctl = bctl;
3095 spin_unlock(&fs_info->balance_lock);
3098 static void unset_balance_control(struct btrfs_fs_info *fs_info)
3100 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3102 BUG_ON(!fs_info->balance_ctl);
3104 spin_lock(&fs_info->balance_lock);
3105 fs_info->balance_ctl = NULL;
3106 spin_unlock(&fs_info->balance_lock);
3112 * Balance filters. Return 1 if chunk should be filtered out
3113 * (should not be balanced).
3115 static int chunk_profiles_filter(u64 chunk_type,
3116 struct btrfs_balance_args *bargs)
3118 chunk_type = chunk_to_extended(chunk_type) &
3119 BTRFS_EXTENDED_PROFILE_MASK;
3121 if (bargs->profiles & chunk_type)
3127 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3128 struct btrfs_balance_args *bargs)
3130 struct btrfs_block_group_cache *cache;
3132 u64 user_thresh_min;
3133 u64 user_thresh_max;
3136 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3137 chunk_used = btrfs_block_group_used(&cache->item);
3139 if (bargs->usage_min == 0)
3140 user_thresh_min = 0;
3142 user_thresh_min = div_factor_fine(cache->key.offset,
3145 if (bargs->usage_max == 0)
3146 user_thresh_max = 1;
3147 else if (bargs->usage_max > 100)
3148 user_thresh_max = cache->key.offset;
3150 user_thresh_max = div_factor_fine(cache->key.offset,
3153 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3156 btrfs_put_block_group(cache);
3160 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3161 u64 chunk_offset, struct btrfs_balance_args *bargs)
3163 struct btrfs_block_group_cache *cache;
3164 u64 chunk_used, user_thresh;
3167 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3168 chunk_used = btrfs_block_group_used(&cache->item);
3170 if (bargs->usage_min == 0)
3172 else if (bargs->usage > 100)
3173 user_thresh = cache->key.offset;
3175 user_thresh = div_factor_fine(cache->key.offset,
3178 if (chunk_used < user_thresh)
3181 btrfs_put_block_group(cache);
3185 static int chunk_devid_filter(struct extent_buffer *leaf,
3186 struct btrfs_chunk *chunk,
3187 struct btrfs_balance_args *bargs)
3189 struct btrfs_stripe *stripe;
3190 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3193 for (i = 0; i < num_stripes; i++) {
3194 stripe = btrfs_stripe_nr(chunk, i);
3195 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3202 /* [pstart, pend) */
3203 static int chunk_drange_filter(struct extent_buffer *leaf,
3204 struct btrfs_chunk *chunk,
3206 struct btrfs_balance_args *bargs)
3208 struct btrfs_stripe *stripe;
3209 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3215 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3218 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3219 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3220 factor = num_stripes / 2;
3221 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3222 factor = num_stripes - 1;
3223 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3224 factor = num_stripes - 2;
3226 factor = num_stripes;
3229 for (i = 0; i < num_stripes; i++) {
3230 stripe = btrfs_stripe_nr(chunk, i);
3231 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3234 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3235 stripe_length = btrfs_chunk_length(leaf, chunk);
3236 stripe_length = div_u64(stripe_length, factor);
3238 if (stripe_offset < bargs->pend &&
3239 stripe_offset + stripe_length > bargs->pstart)
3246 /* [vstart, vend) */
3247 static int chunk_vrange_filter(struct extent_buffer *leaf,
3248 struct btrfs_chunk *chunk,
3250 struct btrfs_balance_args *bargs)
3252 if (chunk_offset < bargs->vend &&
3253 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3254 /* at least part of the chunk is inside this vrange */
3260 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3261 struct btrfs_chunk *chunk,
3262 struct btrfs_balance_args *bargs)
3264 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3266 if (bargs->stripes_min <= num_stripes
3267 && num_stripes <= bargs->stripes_max)
3273 static int chunk_soft_convert_filter(u64 chunk_type,
3274 struct btrfs_balance_args *bargs)
3276 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3279 chunk_type = chunk_to_extended(chunk_type) &
3280 BTRFS_EXTENDED_PROFILE_MASK;
3282 if (bargs->target == chunk_type)
3288 static int should_balance_chunk(struct btrfs_root *root,
3289 struct extent_buffer *leaf,
3290 struct btrfs_chunk *chunk, u64 chunk_offset)
3292 struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
3293 struct btrfs_balance_args *bargs = NULL;
3294 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3297 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3298 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3302 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3303 bargs = &bctl->data;
3304 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3306 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3307 bargs = &bctl->meta;
3309 /* profiles filter */
3310 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3311 chunk_profiles_filter(chunk_type, bargs)) {
3316 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3317 chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
3319 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3320 chunk_usage_range_filter(bctl->fs_info, chunk_offset, bargs)) {
3325 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3326 chunk_devid_filter(leaf, chunk, bargs)) {
3330 /* drange filter, makes sense only with devid filter */
3331 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3332 chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
3337 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3338 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3342 /* stripes filter */
3343 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3344 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3348 /* soft profile changing mode */
3349 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3350 chunk_soft_convert_filter(chunk_type, bargs)) {
3355 * limited by count, must be the last filter
3357 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3358 if (bargs->limit == 0)
3362 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3364 * Same logic as the 'limit' filter; the minimum cannot be
3365 * determined here because we do not have the global informatoin
3366 * about the count of all chunks that satisfy the filters.
3368 if (bargs->limit_max == 0)
3377 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3379 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3380 struct btrfs_root *chunk_root = fs_info->chunk_root;
3381 struct btrfs_root *dev_root = fs_info->dev_root;
3382 struct list_head *devices;
3383 struct btrfs_device *device;
3387 struct btrfs_chunk *chunk;
3388 struct btrfs_path *path;
3389 struct btrfs_key key;
3390 struct btrfs_key found_key;
3391 struct btrfs_trans_handle *trans;
3392 struct extent_buffer *leaf;
3395 int enospc_errors = 0;
3396 bool counting = true;
3397 /* The single value limit and min/max limits use the same bytes in the */
3398 u64 limit_data = bctl->data.limit;
3399 u64 limit_meta = bctl->meta.limit;
3400 u64 limit_sys = bctl->sys.limit;
3404 int chunk_reserved = 0;
3407 /* step one make some room on all the devices */
3408 devices = &fs_info->fs_devices->devices;
3409 list_for_each_entry(device, devices, dev_list) {
3410 old_size = btrfs_device_get_total_bytes(device);
3411 size_to_free = div_factor(old_size, 1);
3412 size_to_free = min_t(u64, size_to_free, SZ_1M);
3413 if (!device->writeable ||
3414 btrfs_device_get_total_bytes(device) -
3415 btrfs_device_get_bytes_used(device) > size_to_free ||
3416 device->is_tgtdev_for_dev_replace)
3419 ret = btrfs_shrink_device(device, old_size - size_to_free);
3424 trans = btrfs_start_transaction(dev_root, 0);
3425 BUG_ON(IS_ERR(trans));
3427 ret = btrfs_grow_device(trans, device, old_size);
3430 btrfs_end_transaction(trans, dev_root);
3433 /* step two, relocate all the chunks */
3434 path = btrfs_alloc_path();
3440 /* zero out stat counters */
3441 spin_lock(&fs_info->balance_lock);
3442 memset(&bctl->stat, 0, sizeof(bctl->stat));
3443 spin_unlock(&fs_info->balance_lock);
3447 * The single value limit and min/max limits use the same bytes
3450 bctl->data.limit = limit_data;
3451 bctl->meta.limit = limit_meta;
3452 bctl->sys.limit = limit_sys;
3454 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3455 key.offset = (u64)-1;
3456 key.type = BTRFS_CHUNK_ITEM_KEY;
3459 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3460 atomic_read(&fs_info->balance_cancel_req)) {
3465 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3466 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3468 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3473 * this shouldn't happen, it means the last relocate
3477 BUG(); /* FIXME break ? */
3479 ret = btrfs_previous_item(chunk_root, path, 0,
3480 BTRFS_CHUNK_ITEM_KEY);
3482 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3487 leaf = path->nodes[0];
3488 slot = path->slots[0];
3489 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3491 if (found_key.objectid != key.objectid) {
3492 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3496 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3497 chunk_type = btrfs_chunk_type(leaf, chunk);
3500 spin_lock(&fs_info->balance_lock);
3501 bctl->stat.considered++;
3502 spin_unlock(&fs_info->balance_lock);
3505 ret = should_balance_chunk(chunk_root, leaf, chunk,
3508 btrfs_release_path(path);
3510 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3515 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3516 spin_lock(&fs_info->balance_lock);
3517 bctl->stat.expected++;
3518 spin_unlock(&fs_info->balance_lock);
3520 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3522 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3524 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3531 * Apply limit_min filter, no need to check if the LIMITS
3532 * filter is used, limit_min is 0 by default
3534 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3535 count_data < bctl->data.limit_min)
3536 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3537 count_meta < bctl->meta.limit_min)
3538 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3539 count_sys < bctl->sys.limit_min)) {
3540 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3544 ASSERT(fs_info->data_sinfo);
3545 spin_lock(&fs_info->data_sinfo->lock);
3546 bytes_used = fs_info->data_sinfo->bytes_used;
3547 spin_unlock(&fs_info->data_sinfo->lock);
3549 if ((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3550 !chunk_reserved && !bytes_used) {
3551 trans = btrfs_start_transaction(chunk_root, 0);
3552 if (IS_ERR(trans)) {
3553 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3554 ret = PTR_ERR(trans);
3558 ret = btrfs_force_chunk_alloc(trans, chunk_root,
3559 BTRFS_BLOCK_GROUP_DATA);
3560 btrfs_end_transaction(trans, chunk_root);
3562 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3568 ret = btrfs_relocate_chunk(chunk_root,
3570 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3571 if (ret && ret != -ENOSPC)
3573 if (ret == -ENOSPC) {
3576 spin_lock(&fs_info->balance_lock);
3577 bctl->stat.completed++;
3578 spin_unlock(&fs_info->balance_lock);
3581 if (found_key.offset == 0)
3583 key.offset = found_key.offset - 1;
3587 btrfs_release_path(path);
3592 btrfs_free_path(path);
3593 if (enospc_errors) {
3594 btrfs_info(fs_info, "%d enospc errors during balance",
3604 * alloc_profile_is_valid - see if a given profile is valid and reduced
3605 * @flags: profile to validate
3606 * @extended: if true @flags is treated as an extended profile
3608 static int alloc_profile_is_valid(u64 flags, int extended)
3610 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3611 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3613 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3615 /* 1) check that all other bits are zeroed */
3619 /* 2) see if profile is reduced */
3621 return !extended; /* "0" is valid for usual profiles */
3623 /* true if exactly one bit set */
3624 return (flags & (flags - 1)) == 0;
3627 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3629 /* cancel requested || normal exit path */
3630 return atomic_read(&fs_info->balance_cancel_req) ||
3631 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3632 atomic_read(&fs_info->balance_cancel_req) == 0);
3635 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3639 unset_balance_control(fs_info);
3640 ret = del_balance_item(fs_info->tree_root);
3642 btrfs_std_error(fs_info, ret, NULL);
3644 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3647 /* Non-zero return value signifies invalidity */
3648 static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg,
3651 return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3652 (!alloc_profile_is_valid(bctl_arg->target, 1) ||
3653 (bctl_arg->target & ~allowed)));
3657 * Should be called with both balance and volume mutexes held
3659 int btrfs_balance(struct btrfs_balance_control *bctl,
3660 struct btrfs_ioctl_balance_args *bargs)
3662 struct btrfs_fs_info *fs_info = bctl->fs_info;
3669 if (btrfs_fs_closing(fs_info) ||
3670 atomic_read(&fs_info->balance_pause_req) ||
3671 atomic_read(&fs_info->balance_cancel_req)) {
3676 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3677 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3681 * In case of mixed groups both data and meta should be picked,
3682 * and identical options should be given for both of them.
3684 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3685 if (mixed && (bctl->flags & allowed)) {
3686 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3687 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3688 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3689 btrfs_err(fs_info, "with mixed groups data and "
3690 "metadata balance options must be the same");
3696 num_devices = fs_info->fs_devices->num_devices;
3697 btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
3698 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3699 BUG_ON(num_devices < 1);
3702 btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
3703 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE | BTRFS_BLOCK_GROUP_DUP;
3704 if (num_devices > 1)
3705 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3706 if (num_devices > 2)
3707 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3708 if (num_devices > 3)
3709 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3710 BTRFS_BLOCK_GROUP_RAID6);
3711 if (validate_convert_profile(&bctl->data, allowed)) {
3712 btrfs_err(fs_info, "unable to start balance with target "
3713 "data profile %llu",
3718 if (validate_convert_profile(&bctl->meta, allowed)) {
3720 "unable to start balance with target metadata profile %llu",
3725 if (validate_convert_profile(&bctl->sys, allowed)) {
3727 "unable to start balance with target system profile %llu",
3733 /* allow to reduce meta or sys integrity only if force set */
3734 allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3735 BTRFS_BLOCK_GROUP_RAID10 |
3736 BTRFS_BLOCK_GROUP_RAID5 |
3737 BTRFS_BLOCK_GROUP_RAID6;
3739 seq = read_seqbegin(&fs_info->profiles_lock);
3741 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3742 (fs_info->avail_system_alloc_bits & allowed) &&
3743 !(bctl->sys.target & allowed)) ||
3744 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3745 (fs_info->avail_metadata_alloc_bits & allowed) &&
3746 !(bctl->meta.target & allowed))) {
3747 if (bctl->flags & BTRFS_BALANCE_FORCE) {
3748 btrfs_info(fs_info, "force reducing metadata integrity");
3750 btrfs_err(fs_info, "balance will reduce metadata "
3751 "integrity, use force if you want this");
3756 } while (read_seqretry(&fs_info->profiles_lock, seq));
3758 if (btrfs_get_num_tolerated_disk_barrier_failures(bctl->meta.target) <
3759 btrfs_get_num_tolerated_disk_barrier_failures(bctl->data.target)) {
3761 "metadata profile 0x%llx has lower redundancy than data profile 0x%llx",
3762 bctl->meta.target, bctl->data.target);
3765 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3766 fs_info->num_tolerated_disk_barrier_failures = min(
3767 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info),
3768 btrfs_get_num_tolerated_disk_barrier_failures(
3772 ret = insert_balance_item(fs_info->tree_root, bctl);
3773 if (ret && ret != -EEXIST)
3776 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3777 BUG_ON(ret == -EEXIST);
3778 set_balance_control(bctl);
3780 BUG_ON(ret != -EEXIST);
3781 spin_lock(&fs_info->balance_lock);
3782 update_balance_args(bctl);
3783 spin_unlock(&fs_info->balance_lock);
3786 atomic_inc(&fs_info->balance_running);
3787 mutex_unlock(&fs_info->balance_mutex);
3789 ret = __btrfs_balance(fs_info);
3791 mutex_lock(&fs_info->balance_mutex);
3792 atomic_dec(&fs_info->balance_running);
3794 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3795 fs_info->num_tolerated_disk_barrier_failures =
3796 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3800 memset(bargs, 0, sizeof(*bargs));
3801 update_ioctl_balance_args(fs_info, 0, bargs);
3804 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3805 balance_need_close(fs_info)) {
3806 __cancel_balance(fs_info);
3809 wake_up(&fs_info->balance_wait_q);
3813 if (bctl->flags & BTRFS_BALANCE_RESUME)
3814 __cancel_balance(fs_info);
3817 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3822 static int balance_kthread(void *data)
3824 struct btrfs_fs_info *fs_info = data;
3827 mutex_lock(&fs_info->volume_mutex);
3828 mutex_lock(&fs_info->balance_mutex);
3830 if (fs_info->balance_ctl) {
3831 btrfs_info(fs_info, "continuing balance");
3832 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3835 mutex_unlock(&fs_info->balance_mutex);
3836 mutex_unlock(&fs_info->volume_mutex);
3841 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3843 struct task_struct *tsk;
3845 spin_lock(&fs_info->balance_lock);
3846 if (!fs_info->balance_ctl) {
3847 spin_unlock(&fs_info->balance_lock);
3850 spin_unlock(&fs_info->balance_lock);
3852 if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
3853 btrfs_info(fs_info, "force skipping balance");
3857 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3858 return PTR_ERR_OR_ZERO(tsk);
3861 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3863 struct btrfs_balance_control *bctl;
3864 struct btrfs_balance_item *item;
3865 struct btrfs_disk_balance_args disk_bargs;
3866 struct btrfs_path *path;
3867 struct extent_buffer *leaf;
3868 struct btrfs_key key;
3871 path = btrfs_alloc_path();
3875 key.objectid = BTRFS_BALANCE_OBJECTID;
3876 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3879 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3882 if (ret > 0) { /* ret = -ENOENT; */
3887 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3893 leaf = path->nodes[0];
3894 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3896 bctl->fs_info = fs_info;
3897 bctl->flags = btrfs_balance_flags(leaf, item);
3898 bctl->flags |= BTRFS_BALANCE_RESUME;
3900 btrfs_balance_data(leaf, item, &disk_bargs);
3901 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3902 btrfs_balance_meta(leaf, item, &disk_bargs);
3903 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3904 btrfs_balance_sys(leaf, item, &disk_bargs);
3905 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3907 WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
3909 mutex_lock(&fs_info->volume_mutex);
3910 mutex_lock(&fs_info->balance_mutex);
3912 set_balance_control(bctl);
3914 mutex_unlock(&fs_info->balance_mutex);
3915 mutex_unlock(&fs_info->volume_mutex);
3917 btrfs_free_path(path);
3921 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3925 mutex_lock(&fs_info->balance_mutex);
3926 if (!fs_info->balance_ctl) {
3927 mutex_unlock(&fs_info->balance_mutex);
3931 if (atomic_read(&fs_info->balance_running)) {
3932 atomic_inc(&fs_info->balance_pause_req);
3933 mutex_unlock(&fs_info->balance_mutex);
3935 wait_event(fs_info->balance_wait_q,
3936 atomic_read(&fs_info->balance_running) == 0);
3938 mutex_lock(&fs_info->balance_mutex);
3939 /* we are good with balance_ctl ripped off from under us */
3940 BUG_ON(atomic_read(&fs_info->balance_running));
3941 atomic_dec(&fs_info->balance_pause_req);
3946 mutex_unlock(&fs_info->balance_mutex);
3950 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3952 if (fs_info->sb->s_flags & MS_RDONLY)
3955 mutex_lock(&fs_info->balance_mutex);
3956 if (!fs_info->balance_ctl) {
3957 mutex_unlock(&fs_info->balance_mutex);
3961 atomic_inc(&fs_info->balance_cancel_req);
3963 * if we are running just wait and return, balance item is
3964 * deleted in btrfs_balance in this case
3966 if (atomic_read(&fs_info->balance_running)) {
3967 mutex_unlock(&fs_info->balance_mutex);
3968 wait_event(fs_info->balance_wait_q,
3969 atomic_read(&fs_info->balance_running) == 0);
3970 mutex_lock(&fs_info->balance_mutex);
3972 /* __cancel_balance needs volume_mutex */
3973 mutex_unlock(&fs_info->balance_mutex);
3974 mutex_lock(&fs_info->volume_mutex);
3975 mutex_lock(&fs_info->balance_mutex);
3977 if (fs_info->balance_ctl)
3978 __cancel_balance(fs_info);
3980 mutex_unlock(&fs_info->volume_mutex);
3983 BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
3984 atomic_dec(&fs_info->balance_cancel_req);
3985 mutex_unlock(&fs_info->balance_mutex);
3989 static int btrfs_uuid_scan_kthread(void *data)
3991 struct btrfs_fs_info *fs_info = data;
3992 struct btrfs_root *root = fs_info->tree_root;
3993 struct btrfs_key key;
3994 struct btrfs_key max_key;
3995 struct btrfs_path *path = NULL;
3997 struct extent_buffer *eb;
3999 struct btrfs_root_item root_item;
4001 struct btrfs_trans_handle *trans = NULL;
4003 path = btrfs_alloc_path();
4010 key.type = BTRFS_ROOT_ITEM_KEY;
4013 max_key.objectid = (u64)-1;
4014 max_key.type = BTRFS_ROOT_ITEM_KEY;
4015 max_key.offset = (u64)-1;
4018 ret = btrfs_search_forward(root, &key, path, 0);
4025 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4026 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4027 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4028 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4031 eb = path->nodes[0];
4032 slot = path->slots[0];
4033 item_size = btrfs_item_size_nr(eb, slot);
4034 if (item_size < sizeof(root_item))
4037 read_extent_buffer(eb, &root_item,
4038 btrfs_item_ptr_offset(eb, slot),
4039 (int)sizeof(root_item));
4040 if (btrfs_root_refs(&root_item) == 0)
4043 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4044 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4048 btrfs_release_path(path);
4050 * 1 - subvol uuid item
4051 * 1 - received_subvol uuid item
4053 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4054 if (IS_ERR(trans)) {
4055 ret = PTR_ERR(trans);
4063 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4064 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
4066 BTRFS_UUID_KEY_SUBVOL,
4069 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4075 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4076 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
4077 root_item.received_uuid,
4078 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4081 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4089 ret = btrfs_end_transaction(trans, fs_info->uuid_root);
4095 btrfs_release_path(path);
4096 if (key.offset < (u64)-1) {
4098 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4100 key.type = BTRFS_ROOT_ITEM_KEY;
4101 } else if (key.objectid < (u64)-1) {
4103 key.type = BTRFS_ROOT_ITEM_KEY;
4112 btrfs_free_path(path);
4113 if (trans && !IS_ERR(trans))
4114 btrfs_end_transaction(trans, fs_info->uuid_root);
4116 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4118 fs_info->update_uuid_tree_gen = 1;
4119 up(&fs_info->uuid_tree_rescan_sem);
4124 * Callback for btrfs_uuid_tree_iterate().
4126 * 0 check succeeded, the entry is not outdated.
4127 * < 0 if an error occurred.
4128 * > 0 if the check failed, which means the caller shall remove the entry.
4130 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
4131 u8 *uuid, u8 type, u64 subid)
4133 struct btrfs_key key;
4135 struct btrfs_root *subvol_root;
4137 if (type != BTRFS_UUID_KEY_SUBVOL &&
4138 type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
4141 key.objectid = subid;
4142 key.type = BTRFS_ROOT_ITEM_KEY;
4143 key.offset = (u64)-1;
4144 subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
4145 if (IS_ERR(subvol_root)) {
4146 ret = PTR_ERR(subvol_root);
4153 case BTRFS_UUID_KEY_SUBVOL:
4154 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
4157 case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
4158 if (memcmp(uuid, subvol_root->root_item.received_uuid,
4168 static int btrfs_uuid_rescan_kthread(void *data)
4170 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4174 * 1st step is to iterate through the existing UUID tree and
4175 * to delete all entries that contain outdated data.
4176 * 2nd step is to add all missing entries to the UUID tree.
4178 ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4180 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4181 up(&fs_info->uuid_tree_rescan_sem);
4184 return btrfs_uuid_scan_kthread(data);
4187 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4189 struct btrfs_trans_handle *trans;
4190 struct btrfs_root *tree_root = fs_info->tree_root;
4191 struct btrfs_root *uuid_root;
4192 struct task_struct *task;
4199 trans = btrfs_start_transaction(tree_root, 2);
4201 return PTR_ERR(trans);
4203 uuid_root = btrfs_create_tree(trans, fs_info,
4204 BTRFS_UUID_TREE_OBJECTID);
4205 if (IS_ERR(uuid_root)) {
4206 ret = PTR_ERR(uuid_root);
4207 btrfs_abort_transaction(trans, tree_root, ret);
4211 fs_info->uuid_root = uuid_root;
4213 ret = btrfs_commit_transaction(trans, tree_root);
4217 down(&fs_info->uuid_tree_rescan_sem);
4218 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4220 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4221 btrfs_warn(fs_info, "failed to start uuid_scan task");
4222 up(&fs_info->uuid_tree_rescan_sem);
4223 return PTR_ERR(task);
4229 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4231 struct task_struct *task;
4233 down(&fs_info->uuid_tree_rescan_sem);
4234 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4236 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4237 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4238 up(&fs_info->uuid_tree_rescan_sem);
4239 return PTR_ERR(task);
4246 * shrinking a device means finding all of the device extents past
4247 * the new size, and then following the back refs to the chunks.
4248 * The chunk relocation code actually frees the device extent
4250 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4252 struct btrfs_trans_handle *trans;
4253 struct btrfs_root *root = device->dev_root;
4254 struct btrfs_dev_extent *dev_extent = NULL;
4255 struct btrfs_path *path;
4261 bool retried = false;
4262 bool checked_pending_chunks = false;
4263 struct extent_buffer *l;
4264 struct btrfs_key key;
4265 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4266 u64 old_total = btrfs_super_total_bytes(super_copy);
4267 u64 old_size = btrfs_device_get_total_bytes(device);
4268 u64 diff = old_size - new_size;
4270 if (device->is_tgtdev_for_dev_replace)
4273 path = btrfs_alloc_path();
4277 path->reada = READA_FORWARD;
4281 btrfs_device_set_total_bytes(device, new_size);
4282 if (device->writeable) {
4283 device->fs_devices->total_rw_bytes -= diff;
4284 spin_lock(&root->fs_info->free_chunk_lock);
4285 root->fs_info->free_chunk_space -= diff;
4286 spin_unlock(&root->fs_info->free_chunk_lock);
4288 unlock_chunks(root);
4291 key.objectid = device->devid;
4292 key.offset = (u64)-1;
4293 key.type = BTRFS_DEV_EXTENT_KEY;
4296 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
4297 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4299 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4303 ret = btrfs_previous_item(root, path, 0, key.type);
4305 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4310 btrfs_release_path(path);
4315 slot = path->slots[0];
4316 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4318 if (key.objectid != device->devid) {
4319 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4320 btrfs_release_path(path);
4324 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4325 length = btrfs_dev_extent_length(l, dev_extent);
4327 if (key.offset + length <= new_size) {
4328 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4329 btrfs_release_path(path);
4333 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4334 btrfs_release_path(path);
4336 ret = btrfs_relocate_chunk(root, chunk_offset);
4337 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4338 if (ret && ret != -ENOSPC)
4342 } while (key.offset-- > 0);
4344 if (failed && !retried) {
4348 } else if (failed && retried) {
4353 /* Shrinking succeeded, else we would be at "done". */
4354 trans = btrfs_start_transaction(root, 0);
4355 if (IS_ERR(trans)) {
4356 ret = PTR_ERR(trans);
4363 * We checked in the above loop all device extents that were already in
4364 * the device tree. However before we have updated the device's
4365 * total_bytes to the new size, we might have had chunk allocations that
4366 * have not complete yet (new block groups attached to transaction
4367 * handles), and therefore their device extents were not yet in the
4368 * device tree and we missed them in the loop above. So if we have any
4369 * pending chunk using a device extent that overlaps the device range
4370 * that we can not use anymore, commit the current transaction and
4371 * repeat the search on the device tree - this way we guarantee we will
4372 * not have chunks using device extents that end beyond 'new_size'.
4374 if (!checked_pending_chunks) {
4375 u64 start = new_size;
4376 u64 len = old_size - new_size;
4378 if (contains_pending_extent(trans->transaction, device,
4380 unlock_chunks(root);
4381 checked_pending_chunks = true;
4384 ret = btrfs_commit_transaction(trans, root);
4391 btrfs_device_set_disk_total_bytes(device, new_size);
4392 if (list_empty(&device->resized_list))
4393 list_add_tail(&device->resized_list,
4394 &root->fs_info->fs_devices->resized_devices);
4396 WARN_ON(diff > old_total);
4397 btrfs_set_super_total_bytes(super_copy, old_total - diff);
4398 unlock_chunks(root);
4400 /* Now btrfs_update_device() will change the on-disk size. */
4401 ret = btrfs_update_device(trans, device);
4402 btrfs_end_transaction(trans, root);
4404 btrfs_free_path(path);
4407 btrfs_device_set_total_bytes(device, old_size);
4408 if (device->writeable)
4409 device->fs_devices->total_rw_bytes += diff;
4410 spin_lock(&root->fs_info->free_chunk_lock);
4411 root->fs_info->free_chunk_space += diff;
4412 spin_unlock(&root->fs_info->free_chunk_lock);
4413 unlock_chunks(root);
4418 static int btrfs_add_system_chunk(struct btrfs_root *root,
4419 struct btrfs_key *key,
4420 struct btrfs_chunk *chunk, int item_size)
4422 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4423 struct btrfs_disk_key disk_key;
4428 array_size = btrfs_super_sys_array_size(super_copy);
4429 if (array_size + item_size + sizeof(disk_key)
4430 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4431 unlock_chunks(root);
4435 ptr = super_copy->sys_chunk_array + array_size;
4436 btrfs_cpu_key_to_disk(&disk_key, key);
4437 memcpy(ptr, &disk_key, sizeof(disk_key));
4438 ptr += sizeof(disk_key);
4439 memcpy(ptr, chunk, item_size);
4440 item_size += sizeof(disk_key);
4441 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4442 unlock_chunks(root);
4448 * sort the devices in descending order by max_avail, total_avail
4450 static int btrfs_cmp_device_info(const void *a, const void *b)
4452 const struct btrfs_device_info *di_a = a;
4453 const struct btrfs_device_info *di_b = b;
4455 if (di_a->max_avail > di_b->max_avail)
4457 if (di_a->max_avail < di_b->max_avail)
4459 if (di_a->total_avail > di_b->total_avail)
4461 if (di_a->total_avail < di_b->total_avail)
4466 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
4468 /* TODO allow them to set a preferred stripe size */
4472 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4474 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4477 btrfs_set_fs_incompat(info, RAID56);
4480 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r) \
4481 - sizeof(struct btrfs_item) \
4482 - sizeof(struct btrfs_chunk)) \
4483 / sizeof(struct btrfs_stripe) + 1)
4485 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
4486 - 2 * sizeof(struct btrfs_disk_key) \
4487 - 2 * sizeof(struct btrfs_chunk)) \
4488 / sizeof(struct btrfs_stripe) + 1)
4490 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4491 struct btrfs_root *extent_root, u64 start,
4494 struct btrfs_fs_info *info = extent_root->fs_info;
4495 struct btrfs_fs_devices *fs_devices = info->fs_devices;
4496 struct list_head *cur;
4497 struct map_lookup *map = NULL;
4498 struct extent_map_tree *em_tree;
4499 struct extent_map *em;
4500 struct btrfs_device_info *devices_info = NULL;
4502 int num_stripes; /* total number of stripes to allocate */
4503 int data_stripes; /* number of stripes that count for
4505 int sub_stripes; /* sub_stripes info for map */
4506 int dev_stripes; /* stripes per dev */
4507 int devs_max; /* max devs to use */
4508 int devs_min; /* min devs needed */
4509 int devs_increment; /* ndevs has to be a multiple of this */
4510 int ncopies; /* how many copies to data has */
4512 u64 max_stripe_size;
4516 u64 raid_stripe_len = BTRFS_STRIPE_LEN;
4522 BUG_ON(!alloc_profile_is_valid(type, 0));
4524 if (list_empty(&fs_devices->alloc_list))
4527 index = __get_raid_index(type);
4529 sub_stripes = btrfs_raid_array[index].sub_stripes;
4530 dev_stripes = btrfs_raid_array[index].dev_stripes;
4531 devs_max = btrfs_raid_array[index].devs_max;
4532 devs_min = btrfs_raid_array[index].devs_min;
4533 devs_increment = btrfs_raid_array[index].devs_increment;
4534 ncopies = btrfs_raid_array[index].ncopies;
4536 if (type & BTRFS_BLOCK_GROUP_DATA) {
4537 max_stripe_size = SZ_1G;
4538 max_chunk_size = 10 * max_stripe_size;
4540 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4541 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4542 /* for larger filesystems, use larger metadata chunks */
4543 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
4544 max_stripe_size = SZ_1G;
4546 max_stripe_size = SZ_256M;
4547 max_chunk_size = max_stripe_size;
4549 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4550 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4551 max_stripe_size = SZ_32M;
4552 max_chunk_size = 2 * max_stripe_size;
4554 devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4556 btrfs_err(info, "invalid chunk type 0x%llx requested",
4561 /* we don't want a chunk larger than 10% of writeable space */
4562 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4565 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4570 cur = fs_devices->alloc_list.next;
4573 * in the first pass through the devices list, we gather information
4574 * about the available holes on each device.
4577 while (cur != &fs_devices->alloc_list) {
4578 struct btrfs_device *device;
4582 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4586 if (!device->writeable) {
4588 "BTRFS: read-only device in alloc_list\n");
4592 if (!device->in_fs_metadata ||
4593 device->is_tgtdev_for_dev_replace)
4596 if (device->total_bytes > device->bytes_used)
4597 total_avail = device->total_bytes - device->bytes_used;
4601 /* If there is no space on this device, skip it. */
4602 if (total_avail == 0)
4605 ret = find_free_dev_extent(trans, device,
4606 max_stripe_size * dev_stripes,
4607 &dev_offset, &max_avail);
4608 if (ret && ret != -ENOSPC)
4612 max_avail = max_stripe_size * dev_stripes;
4614 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4617 if (ndevs == fs_devices->rw_devices) {
4618 WARN(1, "%s: found more than %llu devices\n",
4619 __func__, fs_devices->rw_devices);
4622 devices_info[ndevs].dev_offset = dev_offset;
4623 devices_info[ndevs].max_avail = max_avail;
4624 devices_info[ndevs].total_avail = total_avail;
4625 devices_info[ndevs].dev = device;
4630 * now sort the devices by hole size / available space
4632 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4633 btrfs_cmp_device_info, NULL);
4635 /* round down to number of usable stripes */
4636 ndevs -= ndevs % devs_increment;
4638 if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4643 if (devs_max && ndevs > devs_max)
4646 * the primary goal is to maximize the number of stripes, so use as many
4647 * devices as possible, even if the stripes are not maximum sized.
4649 stripe_size = devices_info[ndevs-1].max_avail;
4650 num_stripes = ndevs * dev_stripes;
4653 * this will have to be fixed for RAID1 and RAID10 over
4656 data_stripes = num_stripes / ncopies;
4658 if (type & BTRFS_BLOCK_GROUP_RAID5) {
4659 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
4660 btrfs_super_stripesize(info->super_copy));
4661 data_stripes = num_stripes - 1;
4663 if (type & BTRFS_BLOCK_GROUP_RAID6) {
4664 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
4665 btrfs_super_stripesize(info->super_copy));
4666 data_stripes = num_stripes - 2;
4670 * Use the number of data stripes to figure out how big this chunk
4671 * is really going to be in terms of logical address space,
4672 * and compare that answer with the max chunk size
4674 if (stripe_size * data_stripes > max_chunk_size) {
4675 u64 mask = (1ULL << 24) - 1;
4677 stripe_size = div_u64(max_chunk_size, data_stripes);
4679 /* bump the answer up to a 16MB boundary */
4680 stripe_size = (stripe_size + mask) & ~mask;
4682 /* but don't go higher than the limits we found
4683 * while searching for free extents
4685 if (stripe_size > devices_info[ndevs-1].max_avail)
4686 stripe_size = devices_info[ndevs-1].max_avail;
4689 stripe_size = div_u64(stripe_size, dev_stripes);
4691 /* align to BTRFS_STRIPE_LEN */
4692 stripe_size = div_u64(stripe_size, raid_stripe_len);
4693 stripe_size *= raid_stripe_len;
4695 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4700 map->num_stripes = num_stripes;
4702 for (i = 0; i < ndevs; ++i) {
4703 for (j = 0; j < dev_stripes; ++j) {
4704 int s = i * dev_stripes + j;
4705 map->stripes[s].dev = devices_info[i].dev;
4706 map->stripes[s].physical = devices_info[i].dev_offset +
4710 map->sector_size = extent_root->sectorsize;
4711 map->stripe_len = raid_stripe_len;
4712 map->io_align = raid_stripe_len;
4713 map->io_width = raid_stripe_len;
4715 map->sub_stripes = sub_stripes;
4717 num_bytes = stripe_size * data_stripes;
4719 trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
4721 em = alloc_extent_map();
4727 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4728 em->map_lookup = map;
4730 em->len = num_bytes;
4731 em->block_start = 0;
4732 em->block_len = em->len;
4733 em->orig_block_len = stripe_size;
4735 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4736 write_lock(&em_tree->lock);
4737 ret = add_extent_mapping(em_tree, em, 0);
4739 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4740 atomic_inc(&em->refs);
4742 write_unlock(&em_tree->lock);
4744 free_extent_map(em);
4748 ret = btrfs_make_block_group(trans, extent_root, 0, type,
4749 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4752 goto error_del_extent;
4754 for (i = 0; i < map->num_stripes; i++) {
4755 num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4756 btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4759 spin_lock(&extent_root->fs_info->free_chunk_lock);
4760 extent_root->fs_info->free_chunk_space -= (stripe_size *
4762 spin_unlock(&extent_root->fs_info->free_chunk_lock);
4764 free_extent_map(em);
4765 check_raid56_incompat_flag(extent_root->fs_info, type);
4767 kfree(devices_info);
4771 write_lock(&em_tree->lock);
4772 remove_extent_mapping(em_tree, em);
4773 write_unlock(&em_tree->lock);
4775 /* One for our allocation */
4776 free_extent_map(em);
4777 /* One for the tree reference */
4778 free_extent_map(em);
4779 /* One for the pending_chunks list reference */
4780 free_extent_map(em);
4782 kfree(devices_info);
4786 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4787 struct btrfs_root *extent_root,
4788 u64 chunk_offset, u64 chunk_size)
4790 struct btrfs_key key;
4791 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4792 struct btrfs_device *device;
4793 struct btrfs_chunk *chunk;
4794 struct btrfs_stripe *stripe;
4795 struct extent_map_tree *em_tree;
4796 struct extent_map *em;
4797 struct map_lookup *map;
4804 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4805 read_lock(&em_tree->lock);
4806 em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size);
4807 read_unlock(&em_tree->lock);
4810 btrfs_crit(extent_root->fs_info, "unable to find logical "
4811 "%Lu len %Lu", chunk_offset, chunk_size);
4815 if (em->start != chunk_offset || em->len != chunk_size) {
4816 btrfs_crit(extent_root->fs_info, "found a bad mapping, wanted"
4817 " %Lu-%Lu, found %Lu-%Lu", chunk_offset,
4818 chunk_size, em->start, em->len);
4819 free_extent_map(em);
4823 map = em->map_lookup;
4824 item_size = btrfs_chunk_item_size(map->num_stripes);
4825 stripe_size = em->orig_block_len;
4827 chunk = kzalloc(item_size, GFP_NOFS);
4834 * Take the device list mutex to prevent races with the final phase of
4835 * a device replace operation that replaces the device object associated
4836 * with the map's stripes, because the device object's id can change
4837 * at any time during that final phase of the device replace operation
4838 * (dev-replace.c:btrfs_dev_replace_finishing()).
4840 mutex_lock(&chunk_root->fs_info->fs_devices->device_list_mutex);
4841 for (i = 0; i < map->num_stripes; i++) {
4842 device = map->stripes[i].dev;
4843 dev_offset = map->stripes[i].physical;
4845 ret = btrfs_update_device(trans, device);
4848 ret = btrfs_alloc_dev_extent(trans, device,
4849 chunk_root->root_key.objectid,
4850 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4851 chunk_offset, dev_offset,
4857 mutex_unlock(&chunk_root->fs_info->fs_devices->device_list_mutex);
4861 stripe = &chunk->stripe;
4862 for (i = 0; i < map->num_stripes; i++) {
4863 device = map->stripes[i].dev;
4864 dev_offset = map->stripes[i].physical;
4866 btrfs_set_stack_stripe_devid(stripe, device->devid);
4867 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4868 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4871 mutex_unlock(&chunk_root->fs_info->fs_devices->device_list_mutex);
4873 btrfs_set_stack_chunk_length(chunk, chunk_size);
4874 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4875 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4876 btrfs_set_stack_chunk_type(chunk, map->type);
4877 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4878 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4879 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4880 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4881 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4883 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4884 key.type = BTRFS_CHUNK_ITEM_KEY;
4885 key.offset = chunk_offset;
4887 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4888 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4890 * TODO: Cleanup of inserted chunk root in case of
4893 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
4899 free_extent_map(em);
4904 * Chunk allocation falls into two parts. The first part does works
4905 * that make the new allocated chunk useable, but not do any operation
4906 * that modifies the chunk tree. The second part does the works that
4907 * require modifying the chunk tree. This division is important for the
4908 * bootstrap process of adding storage to a seed btrfs.
4910 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4911 struct btrfs_root *extent_root, u64 type)
4915 ASSERT(mutex_is_locked(&extent_root->fs_info->chunk_mutex));
4916 chunk_offset = find_next_chunk(extent_root->fs_info);
4917 return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type);
4920 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4921 struct btrfs_root *root,
4922 struct btrfs_device *device)
4925 u64 sys_chunk_offset;
4927 struct btrfs_fs_info *fs_info = root->fs_info;
4928 struct btrfs_root *extent_root = fs_info->extent_root;
4931 chunk_offset = find_next_chunk(fs_info);
4932 alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
4933 ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset,
4938 sys_chunk_offset = find_next_chunk(root->fs_info);
4939 alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
4940 ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset,
4945 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
4949 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
4950 BTRFS_BLOCK_GROUP_RAID10 |
4951 BTRFS_BLOCK_GROUP_RAID5 |
4952 BTRFS_BLOCK_GROUP_DUP)) {
4954 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
4963 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
4965 struct extent_map *em;
4966 struct map_lookup *map;
4967 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4972 read_lock(&map_tree->map_tree.lock);
4973 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
4974 read_unlock(&map_tree->map_tree.lock);
4978 map = em->map_lookup;
4979 for (i = 0; i < map->num_stripes; i++) {
4980 if (map->stripes[i].dev->missing) {
4985 if (!map->stripes[i].dev->writeable) {
4992 * If the number of missing devices is larger than max errors,
4993 * we can not write the data into that chunk successfully, so
4996 if (miss_ndevs > btrfs_chunk_max_errors(map))
4999 free_extent_map(em);
5003 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
5005 extent_map_tree_init(&tree->map_tree);
5008 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
5010 struct extent_map *em;
5013 write_lock(&tree->map_tree.lock);
5014 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
5016 remove_extent_mapping(&tree->map_tree, em);
5017 write_unlock(&tree->map_tree.lock);
5021 free_extent_map(em);
5022 /* once for the tree */
5023 free_extent_map(em);
5027 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5029 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5030 struct extent_map *em;
5031 struct map_lookup *map;
5032 struct extent_map_tree *em_tree = &map_tree->map_tree;
5035 read_lock(&em_tree->lock);
5036 em = lookup_extent_mapping(em_tree, logical, len);
5037 read_unlock(&em_tree->lock);
5040 * We could return errors for these cases, but that could get ugly and
5041 * we'd probably do the same thing which is just not do anything else
5042 * and exit, so return 1 so the callers don't try to use other copies.
5045 btrfs_crit(fs_info, "No mapping for %Lu-%Lu", logical,
5050 if (em->start > logical || em->start + em->len < logical) {
5051 btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got "
5052 "%Lu-%Lu", logical, logical+len, em->start,
5053 em->start + em->len);
5054 free_extent_map(em);
5058 map = em->map_lookup;
5059 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
5060 ret = map->num_stripes;
5061 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5062 ret = map->sub_stripes;
5063 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5065 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5069 free_extent_map(em);
5071 btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
5072 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
5074 btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
5079 unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
5080 struct btrfs_mapping_tree *map_tree,
5083 struct extent_map *em;
5084 struct map_lookup *map;
5085 struct extent_map_tree *em_tree = &map_tree->map_tree;
5086 unsigned long len = root->sectorsize;
5088 read_lock(&em_tree->lock);
5089 em = lookup_extent_mapping(em_tree, logical, len);
5090 read_unlock(&em_tree->lock);
5093 BUG_ON(em->start > logical || em->start + em->len < logical);
5094 map = em->map_lookup;
5095 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5096 len = map->stripe_len * nr_data_stripes(map);
5097 free_extent_map(em);
5101 int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
5102 u64 logical, u64 len, int mirror_num)
5104 struct extent_map *em;
5105 struct map_lookup *map;
5106 struct extent_map_tree *em_tree = &map_tree->map_tree;
5109 read_lock(&em_tree->lock);
5110 em = lookup_extent_mapping(em_tree, logical, len);
5111 read_unlock(&em_tree->lock);
5114 BUG_ON(em->start > logical || em->start + em->len < logical);
5115 map = em->map_lookup;
5116 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5118 free_extent_map(em);
5122 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5123 struct map_lookup *map, int first, int num,
5124 int optimal, int dev_replace_is_ongoing)
5128 struct btrfs_device *srcdev;
5130 if (dev_replace_is_ongoing &&
5131 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5132 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5133 srcdev = fs_info->dev_replace.srcdev;
5138 * try to avoid the drive that is the source drive for a
5139 * dev-replace procedure, only choose it if no other non-missing
5140 * mirror is available
5142 for (tolerance = 0; tolerance < 2; tolerance++) {
5143 if (map->stripes[optimal].dev->bdev &&
5144 (tolerance || map->stripes[optimal].dev != srcdev))
5146 for (i = first; i < first + num; i++) {
5147 if (map->stripes[i].dev->bdev &&
5148 (tolerance || map->stripes[i].dev != srcdev))
5153 /* we couldn't find one that doesn't fail. Just return something
5154 * and the io error handling code will clean up eventually
5159 static inline int parity_smaller(u64 a, u64 b)
5164 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5165 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5167 struct btrfs_bio_stripe s;
5174 for (i = 0; i < num_stripes - 1; i++) {
5175 if (parity_smaller(bbio->raid_map[i],
5176 bbio->raid_map[i+1])) {
5177 s = bbio->stripes[i];
5178 l = bbio->raid_map[i];
5179 bbio->stripes[i] = bbio->stripes[i+1];
5180 bbio->raid_map[i] = bbio->raid_map[i+1];
5181 bbio->stripes[i+1] = s;
5182 bbio->raid_map[i+1] = l;
5190 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5192 struct btrfs_bio *bbio = kzalloc(
5193 /* the size of the btrfs_bio */
5194 sizeof(struct btrfs_bio) +
5195 /* plus the variable array for the stripes */
5196 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5197 /* plus the variable array for the tgt dev */
5198 sizeof(int) * (real_stripes) +
5200 * plus the raid_map, which includes both the tgt dev
5203 sizeof(u64) * (total_stripes),
5204 GFP_NOFS|__GFP_NOFAIL);
5206 atomic_set(&bbio->error, 0);
5207 atomic_set(&bbio->refs, 1);
5212 void btrfs_get_bbio(struct btrfs_bio *bbio)
5214 WARN_ON(!atomic_read(&bbio->refs));
5215 atomic_inc(&bbio->refs);
5218 void btrfs_put_bbio(struct btrfs_bio *bbio)
5222 if (atomic_dec_and_test(&bbio->refs))
5226 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5227 u64 logical, u64 *length,
5228 struct btrfs_bio **bbio_ret,
5229 int mirror_num, int need_raid_map)
5231 struct extent_map *em;
5232 struct map_lookup *map;
5233 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5234 struct extent_map_tree *em_tree = &map_tree->map_tree;
5237 u64 stripe_end_offset;
5247 int tgtdev_indexes = 0;
5248 struct btrfs_bio *bbio = NULL;
5249 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
5250 int dev_replace_is_ongoing = 0;
5251 int num_alloc_stripes;
5252 int patch_the_first_stripe_for_dev_replace = 0;
5253 u64 physical_to_patch_in_first_stripe = 0;
5254 u64 raid56_full_stripe_start = (u64)-1;
5256 read_lock(&em_tree->lock);
5257 em = lookup_extent_mapping(em_tree, logical, *length);
5258 read_unlock(&em_tree->lock);
5261 btrfs_crit(fs_info, "unable to find logical %llu len %llu",
5266 if (em->start > logical || em->start + em->len < logical) {
5267 btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, "
5268 "found %Lu-%Lu", logical, em->start,
5269 em->start + em->len);
5270 free_extent_map(em);
5274 map = em->map_lookup;
5275 offset = logical - em->start;
5277 stripe_len = map->stripe_len;
5280 * stripe_nr counts the total number of stripes we have to stride
5281 * to get to this block
5283 stripe_nr = div64_u64(stripe_nr, stripe_len);
5285 stripe_offset = stripe_nr * stripe_len;
5286 BUG_ON(offset < stripe_offset);
5288 /* stripe_offset is the offset of this block in its stripe*/
5289 stripe_offset = offset - stripe_offset;
5291 /* if we're here for raid56, we need to know the stripe aligned start */
5292 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5293 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5294 raid56_full_stripe_start = offset;
5296 /* allow a write of a full stripe, but make sure we don't
5297 * allow straddling of stripes
5299 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5301 raid56_full_stripe_start *= full_stripe_len;
5304 if (rw & REQ_DISCARD) {
5305 /* we don't discard raid56 yet */
5306 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5310 *length = min_t(u64, em->len - offset, *length);
5311 } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5313 /* For writes to RAID[56], allow a full stripeset across all disks.
5314 For other RAID types and for RAID[56] reads, just allow a single
5315 stripe (on a single disk). */
5316 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
5318 max_len = stripe_len * nr_data_stripes(map) -
5319 (offset - raid56_full_stripe_start);
5321 /* we limit the length of each bio to what fits in a stripe */
5322 max_len = stripe_len - stripe_offset;
5324 *length = min_t(u64, em->len - offset, max_len);
5326 *length = em->len - offset;
5329 /* This is for when we're called from btrfs_merge_bio_hook() and all
5330 it cares about is the length */
5334 btrfs_dev_replace_lock(dev_replace, 0);
5335 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5336 if (!dev_replace_is_ongoing)
5337 btrfs_dev_replace_unlock(dev_replace, 0);
5339 btrfs_dev_replace_set_lock_blocking(dev_replace);
5341 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
5342 !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) &&
5343 dev_replace->tgtdev != NULL) {
5345 * in dev-replace case, for repair case (that's the only
5346 * case where the mirror is selected explicitly when
5347 * calling btrfs_map_block), blocks left of the left cursor
5348 * can also be read from the target drive.
5349 * For REQ_GET_READ_MIRRORS, the target drive is added as
5350 * the last one to the array of stripes. For READ, it also
5351 * needs to be supported using the same mirror number.
5352 * If the requested block is not left of the left cursor,
5353 * EIO is returned. This can happen because btrfs_num_copies()
5354 * returns one more in the dev-replace case.
5356 u64 tmp_length = *length;
5357 struct btrfs_bio *tmp_bbio = NULL;
5358 int tmp_num_stripes;
5359 u64 srcdev_devid = dev_replace->srcdev->devid;
5360 int index_srcdev = 0;
5362 u64 physical_of_found = 0;
5364 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
5365 logical, &tmp_length, &tmp_bbio, 0, 0);
5367 WARN_ON(tmp_bbio != NULL);
5371 tmp_num_stripes = tmp_bbio->num_stripes;
5372 if (mirror_num > tmp_num_stripes) {
5374 * REQ_GET_READ_MIRRORS does not contain this
5375 * mirror, that means that the requested area
5376 * is not left of the left cursor
5379 btrfs_put_bbio(tmp_bbio);
5384 * process the rest of the function using the mirror_num
5385 * of the source drive. Therefore look it up first.
5386 * At the end, patch the device pointer to the one of the
5389 for (i = 0; i < tmp_num_stripes; i++) {
5390 if (tmp_bbio->stripes[i].dev->devid != srcdev_devid)
5394 * In case of DUP, in order to keep it simple, only add
5395 * the mirror with the lowest physical address
5398 physical_of_found <= tmp_bbio->stripes[i].physical)
5403 physical_of_found = tmp_bbio->stripes[i].physical;
5406 btrfs_put_bbio(tmp_bbio);
5414 mirror_num = index_srcdev + 1;
5415 patch_the_first_stripe_for_dev_replace = 1;
5416 physical_to_patch_in_first_stripe = physical_of_found;
5417 } else if (mirror_num > map->num_stripes) {
5423 stripe_nr_orig = stripe_nr;
5424 stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
5425 stripe_nr_end = div_u64(stripe_nr_end, map->stripe_len);
5426 stripe_end_offset = stripe_nr_end * map->stripe_len -
5429 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5430 if (rw & REQ_DISCARD)
5431 num_stripes = min_t(u64, map->num_stripes,
5432 stripe_nr_end - stripe_nr_orig);
5433 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5435 if (!(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)))
5437 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
5438 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))
5439 num_stripes = map->num_stripes;
5440 else if (mirror_num)
5441 stripe_index = mirror_num - 1;
5443 stripe_index = find_live_mirror(fs_info, map, 0,
5445 current->pid % map->num_stripes,
5446 dev_replace_is_ongoing);
5447 mirror_num = stripe_index + 1;
5450 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
5451 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
5452 num_stripes = map->num_stripes;
5453 } else if (mirror_num) {
5454 stripe_index = mirror_num - 1;
5459 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5460 u32 factor = map->num_stripes / map->sub_stripes;
5462 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5463 stripe_index *= map->sub_stripes;
5465 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
5466 num_stripes = map->sub_stripes;
5467 else if (rw & REQ_DISCARD)
5468 num_stripes = min_t(u64, map->sub_stripes *
5469 (stripe_nr_end - stripe_nr_orig),
5471 else if (mirror_num)
5472 stripe_index += mirror_num - 1;
5474 int old_stripe_index = stripe_index;
5475 stripe_index = find_live_mirror(fs_info, map,
5477 map->sub_stripes, stripe_index +
5478 current->pid % map->sub_stripes,
5479 dev_replace_is_ongoing);
5480 mirror_num = stripe_index - old_stripe_index + 1;
5483 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5484 if (need_raid_map &&
5485 ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) ||
5487 /* push stripe_nr back to the start of the full stripe */
5488 stripe_nr = div_u64(raid56_full_stripe_start,
5489 stripe_len * nr_data_stripes(map));
5491 /* RAID[56] write or recovery. Return all stripes */
5492 num_stripes = map->num_stripes;
5493 max_errors = nr_parity_stripes(map);
5495 *length = map->stripe_len;
5500 * Mirror #0 or #1 means the original data block.
5501 * Mirror #2 is RAID5 parity block.
5502 * Mirror #3 is RAID6 Q block.
5504 stripe_nr = div_u64_rem(stripe_nr,
5505 nr_data_stripes(map), &stripe_index);
5507 stripe_index = nr_data_stripes(map) +
5510 /* We distribute the parity blocks across stripes */
5511 div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
5513 if (!(rw & (REQ_WRITE | REQ_DISCARD |
5514 REQ_GET_READ_MIRRORS)) && mirror_num <= 1)
5519 * after this, stripe_nr is the number of stripes on this
5520 * device we have to walk to find the data, and stripe_index is
5521 * the number of our device in the stripe array
5523 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5525 mirror_num = stripe_index + 1;
5527 BUG_ON(stripe_index >= map->num_stripes);
5529 num_alloc_stripes = num_stripes;
5530 if (dev_replace_is_ongoing) {
5531 if (rw & (REQ_WRITE | REQ_DISCARD))
5532 num_alloc_stripes <<= 1;
5533 if (rw & REQ_GET_READ_MIRRORS)
5534 num_alloc_stripes++;
5535 tgtdev_indexes = num_stripes;
5538 bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
5543 if (dev_replace_is_ongoing)
5544 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
5546 /* build raid_map */
5547 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK &&
5548 need_raid_map && ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) ||
5553 bbio->raid_map = (u64 *)((void *)bbio->stripes +
5554 sizeof(struct btrfs_bio_stripe) *
5556 sizeof(int) * tgtdev_indexes);
5558 /* Work out the disk rotation on this stripe-set */
5559 div_u64_rem(stripe_nr, num_stripes, &rot);
5561 /* Fill in the logical address of each stripe */
5562 tmp = stripe_nr * nr_data_stripes(map);
5563 for (i = 0; i < nr_data_stripes(map); i++)
5564 bbio->raid_map[(i+rot) % num_stripes] =
5565 em->start + (tmp + i) * map->stripe_len;
5567 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5568 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5569 bbio->raid_map[(i+rot+1) % num_stripes] =
5573 if (rw & REQ_DISCARD) {
5575 u32 sub_stripes = 0;
5576 u64 stripes_per_dev = 0;
5577 u32 remaining_stripes = 0;
5578 u32 last_stripe = 0;
5581 (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
5582 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5585 sub_stripes = map->sub_stripes;
5587 factor = map->num_stripes / sub_stripes;
5588 stripes_per_dev = div_u64_rem(stripe_nr_end -
5591 &remaining_stripes);
5592 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5593 last_stripe *= sub_stripes;
5596 for (i = 0; i < num_stripes; i++) {
5597 bbio->stripes[i].physical =
5598 map->stripes[stripe_index].physical +
5599 stripe_offset + stripe_nr * map->stripe_len;
5600 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5602 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5603 BTRFS_BLOCK_GROUP_RAID10)) {
5604 bbio->stripes[i].length = stripes_per_dev *
5607 if (i / sub_stripes < remaining_stripes)
5608 bbio->stripes[i].length +=
5612 * Special for the first stripe and
5615 * |-------|...|-------|
5619 if (i < sub_stripes)
5620 bbio->stripes[i].length -=
5623 if (stripe_index >= last_stripe &&
5624 stripe_index <= (last_stripe +
5626 bbio->stripes[i].length -=
5629 if (i == sub_stripes - 1)
5632 bbio->stripes[i].length = *length;
5635 if (stripe_index == map->num_stripes) {
5636 /* This could only happen for RAID0/10 */
5642 for (i = 0; i < num_stripes; i++) {
5643 bbio->stripes[i].physical =
5644 map->stripes[stripe_index].physical +
5646 stripe_nr * map->stripe_len;
5647 bbio->stripes[i].dev =
5648 map->stripes[stripe_index].dev;
5653 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
5654 max_errors = btrfs_chunk_max_errors(map);
5657 sort_parity_stripes(bbio, num_stripes);
5660 if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) &&
5661 dev_replace->tgtdev != NULL) {
5662 int index_where_to_add;
5663 u64 srcdev_devid = dev_replace->srcdev->devid;
5666 * duplicate the write operations while the dev replace
5667 * procedure is running. Since the copying of the old disk
5668 * to the new disk takes place at run time while the
5669 * filesystem is mounted writable, the regular write
5670 * operations to the old disk have to be duplicated to go
5671 * to the new disk as well.
5672 * Note that device->missing is handled by the caller, and
5673 * that the write to the old disk is already set up in the
5676 index_where_to_add = num_stripes;
5677 for (i = 0; i < num_stripes; i++) {
5678 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5679 /* write to new disk, too */
5680 struct btrfs_bio_stripe *new =
5681 bbio->stripes + index_where_to_add;
5682 struct btrfs_bio_stripe *old =
5685 new->physical = old->physical;
5686 new->length = old->length;
5687 new->dev = dev_replace->tgtdev;
5688 bbio->tgtdev_map[i] = index_where_to_add;
5689 index_where_to_add++;
5694 num_stripes = index_where_to_add;
5695 } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) &&
5696 dev_replace->tgtdev != NULL) {
5697 u64 srcdev_devid = dev_replace->srcdev->devid;
5698 int index_srcdev = 0;
5700 u64 physical_of_found = 0;
5703 * During the dev-replace procedure, the target drive can
5704 * also be used to read data in case it is needed to repair
5705 * a corrupt block elsewhere. This is possible if the
5706 * requested area is left of the left cursor. In this area,
5707 * the target drive is a full copy of the source drive.
5709 for (i = 0; i < num_stripes; i++) {
5710 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5712 * In case of DUP, in order to keep it
5713 * simple, only add the mirror with the
5714 * lowest physical address
5717 physical_of_found <=
5718 bbio->stripes[i].physical)
5722 physical_of_found = bbio->stripes[i].physical;
5726 if (physical_of_found + map->stripe_len <=
5727 dev_replace->cursor_left) {
5728 struct btrfs_bio_stripe *tgtdev_stripe =
5729 bbio->stripes + num_stripes;
5731 tgtdev_stripe->physical = physical_of_found;
5732 tgtdev_stripe->length =
5733 bbio->stripes[index_srcdev].length;
5734 tgtdev_stripe->dev = dev_replace->tgtdev;
5735 bbio->tgtdev_map[index_srcdev] = num_stripes;
5744 bbio->map_type = map->type;
5745 bbio->num_stripes = num_stripes;
5746 bbio->max_errors = max_errors;
5747 bbio->mirror_num = mirror_num;
5748 bbio->num_tgtdevs = tgtdev_indexes;
5751 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5752 * mirror_num == num_stripes + 1 && dev_replace target drive is
5753 * available as a mirror
5755 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5756 WARN_ON(num_stripes > 1);
5757 bbio->stripes[0].dev = dev_replace->tgtdev;
5758 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5759 bbio->mirror_num = map->num_stripes + 1;
5762 if (dev_replace_is_ongoing) {
5763 btrfs_dev_replace_clear_lock_blocking(dev_replace);
5764 btrfs_dev_replace_unlock(dev_replace, 0);
5766 free_extent_map(em);
5770 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5771 u64 logical, u64 *length,
5772 struct btrfs_bio **bbio_ret, int mirror_num)
5774 return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5778 /* For Scrub/replace */
5779 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, int rw,
5780 u64 logical, u64 *length,
5781 struct btrfs_bio **bbio_ret, int mirror_num,
5784 return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5785 mirror_num, need_raid_map);
5788 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
5789 u64 chunk_start, u64 physical, u64 devid,
5790 u64 **logical, int *naddrs, int *stripe_len)
5792 struct extent_map_tree *em_tree = &map_tree->map_tree;
5793 struct extent_map *em;
5794 struct map_lookup *map;
5802 read_lock(&em_tree->lock);
5803 em = lookup_extent_mapping(em_tree, chunk_start, 1);
5804 read_unlock(&em_tree->lock);
5807 printk(KERN_ERR "BTRFS: couldn't find em for chunk %Lu\n",
5812 if (em->start != chunk_start) {
5813 printk(KERN_ERR "BTRFS: bad chunk start, em=%Lu, wanted=%Lu\n",
5814 em->start, chunk_start);
5815 free_extent_map(em);
5818 map = em->map_lookup;
5821 rmap_len = map->stripe_len;
5823 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5824 length = div_u64(length, map->num_stripes / map->sub_stripes);
5825 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5826 length = div_u64(length, map->num_stripes);
5827 else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5828 length = div_u64(length, nr_data_stripes(map));
5829 rmap_len = map->stripe_len * nr_data_stripes(map);
5832 buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
5833 BUG_ON(!buf); /* -ENOMEM */
5835 for (i = 0; i < map->num_stripes; i++) {
5836 if (devid && map->stripes[i].dev->devid != devid)
5838 if (map->stripes[i].physical > physical ||
5839 map->stripes[i].physical + length <= physical)
5842 stripe_nr = physical - map->stripes[i].physical;
5843 stripe_nr = div_u64(stripe_nr, map->stripe_len);
5845 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5846 stripe_nr = stripe_nr * map->num_stripes + i;
5847 stripe_nr = div_u64(stripe_nr, map->sub_stripes);
5848 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5849 stripe_nr = stripe_nr * map->num_stripes + i;
5850 } /* else if RAID[56], multiply by nr_data_stripes().
5851 * Alternatively, just use rmap_len below instead of
5852 * map->stripe_len */
5854 bytenr = chunk_start + stripe_nr * rmap_len;
5855 WARN_ON(nr >= map->num_stripes);
5856 for (j = 0; j < nr; j++) {
5857 if (buf[j] == bytenr)
5861 WARN_ON(nr >= map->num_stripes);
5868 *stripe_len = rmap_len;
5870 free_extent_map(em);
5874 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
5876 bio->bi_private = bbio->private;
5877 bio->bi_end_io = bbio->end_io;
5880 btrfs_put_bbio(bbio);
5883 static void btrfs_end_bio(struct bio *bio)
5885 struct btrfs_bio *bbio = bio->bi_private;
5886 int is_orig_bio = 0;
5888 if (bio->bi_error) {
5889 atomic_inc(&bbio->error);
5890 if (bio->bi_error == -EIO || bio->bi_error == -EREMOTEIO) {
5891 unsigned int stripe_index =
5892 btrfs_io_bio(bio)->stripe_index;
5893 struct btrfs_device *dev;
5895 BUG_ON(stripe_index >= bbio->num_stripes);
5896 dev = bbio->stripes[stripe_index].dev;
5898 if (bio->bi_rw & WRITE)
5899 btrfs_dev_stat_inc(dev,
5900 BTRFS_DEV_STAT_WRITE_ERRS);
5902 btrfs_dev_stat_inc(dev,
5903 BTRFS_DEV_STAT_READ_ERRS);
5904 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
5905 btrfs_dev_stat_inc(dev,
5906 BTRFS_DEV_STAT_FLUSH_ERRS);
5907 btrfs_dev_stat_print_on_error(dev);
5912 if (bio == bbio->orig_bio)
5915 btrfs_bio_counter_dec(bbio->fs_info);
5917 if (atomic_dec_and_test(&bbio->stripes_pending)) {
5920 bio = bbio->orig_bio;
5923 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5924 /* only send an error to the higher layers if it is
5925 * beyond the tolerance of the btrfs bio
5927 if (atomic_read(&bbio->error) > bbio->max_errors) {
5928 bio->bi_error = -EIO;
5931 * this bio is actually up to date, we didn't
5932 * go over the max number of errors
5937 btrfs_end_bbio(bbio, bio);
5938 } else if (!is_orig_bio) {
5944 * see run_scheduled_bios for a description of why bios are collected for
5947 * This will add one bio to the pending list for a device and make sure
5948 * the work struct is scheduled.
5950 static noinline void btrfs_schedule_bio(struct btrfs_root *root,
5951 struct btrfs_device *device,
5952 int rw, struct bio *bio)
5954 int should_queue = 1;
5955 struct btrfs_pending_bios *pending_bios;
5957 if (device->missing || !device->bdev) {
5962 /* don't bother with additional async steps for reads, right now */
5963 if (!(rw & REQ_WRITE)) {
5965 btrfsic_submit_bio(rw, bio);
5971 * nr_async_bios allows us to reliably return congestion to the
5972 * higher layers. Otherwise, the async bio makes it appear we have
5973 * made progress against dirty pages when we've really just put it
5974 * on a queue for later
5976 atomic_inc(&root->fs_info->nr_async_bios);
5977 WARN_ON(bio->bi_next);
5978 bio->bi_next = NULL;
5981 spin_lock(&device->io_lock);
5982 if (bio->bi_rw & REQ_SYNC)
5983 pending_bios = &device->pending_sync_bios;
5985 pending_bios = &device->pending_bios;
5987 if (pending_bios->tail)
5988 pending_bios->tail->bi_next = bio;
5990 pending_bios->tail = bio;
5991 if (!pending_bios->head)
5992 pending_bios->head = bio;
5993 if (device->running_pending)
5996 spin_unlock(&device->io_lock);
5999 btrfs_queue_work(root->fs_info->submit_workers,
6003 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
6004 struct bio *bio, u64 physical, int dev_nr,
6007 struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
6009 bio->bi_private = bbio;
6010 btrfs_io_bio(bio)->stripe_index = dev_nr;
6011 bio->bi_end_io = btrfs_end_bio;
6012 bio->bi_iter.bi_sector = physical >> 9;
6015 struct rcu_string *name;
6018 name = rcu_dereference(dev->name);
6019 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
6020 "(%s id %llu), size=%u\n", rw,
6021 (u64)bio->bi_iter.bi_sector, (u_long)dev->bdev->bd_dev,
6022 name->str, dev->devid, bio->bi_iter.bi_size);
6026 bio->bi_bdev = dev->bdev;
6028 btrfs_bio_counter_inc_noblocked(root->fs_info);
6031 btrfs_schedule_bio(root, dev, rw, bio);
6033 btrfsic_submit_bio(rw, bio);
6036 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6038 atomic_inc(&bbio->error);
6039 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6040 /* Shoud be the original bio. */
6041 WARN_ON(bio != bbio->orig_bio);
6043 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6044 bio->bi_iter.bi_sector = logical >> 9;
6045 bio->bi_error = -EIO;
6046 btrfs_end_bbio(bbio, bio);
6050 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
6051 int mirror_num, int async_submit)
6053 struct btrfs_device *dev;
6054 struct bio *first_bio = bio;
6055 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
6061 struct btrfs_bio *bbio = NULL;
6063 length = bio->bi_iter.bi_size;
6064 map_length = length;
6066 btrfs_bio_counter_inc_blocked(root->fs_info);
6067 ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
6070 btrfs_bio_counter_dec(root->fs_info);
6074 total_devs = bbio->num_stripes;
6075 bbio->orig_bio = first_bio;
6076 bbio->private = first_bio->bi_private;
6077 bbio->end_io = first_bio->bi_end_io;
6078 bbio->fs_info = root->fs_info;
6079 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6081 if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6082 ((rw & WRITE) || (mirror_num > 1))) {
6083 /* In this case, map_length has been set to the length of
6084 a single stripe; not the whole write */
6086 ret = raid56_parity_write(root, bio, bbio, map_length);
6088 ret = raid56_parity_recover(root, bio, bbio, map_length,
6092 btrfs_bio_counter_dec(root->fs_info);
6096 if (map_length < length) {
6097 btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu",
6098 logical, length, map_length);
6102 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6103 dev = bbio->stripes[dev_nr].dev;
6104 if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
6105 bbio_error(bbio, first_bio, logical);
6109 if (dev_nr < total_devs - 1) {
6110 bio = btrfs_bio_clone(first_bio, GFP_NOFS);
6111 BUG_ON(!bio); /* -ENOMEM */
6115 submit_stripe_bio(root, bbio, bio,
6116 bbio->stripes[dev_nr].physical, dev_nr, rw,
6119 btrfs_bio_counter_dec(root->fs_info);
6123 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
6126 struct btrfs_device *device;
6127 struct btrfs_fs_devices *cur_devices;
6129 cur_devices = fs_info->fs_devices;
6130 while (cur_devices) {
6132 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
6133 device = __find_device(&cur_devices->devices,
6138 cur_devices = cur_devices->seed;
6143 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
6144 struct btrfs_fs_devices *fs_devices,
6145 u64 devid, u8 *dev_uuid)
6147 struct btrfs_device *device;
6149 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6153 list_add(&device->dev_list, &fs_devices->devices);
6154 device->fs_devices = fs_devices;
6155 fs_devices->num_devices++;
6157 device->missing = 1;
6158 fs_devices->missing_devices++;
6164 * btrfs_alloc_device - allocate struct btrfs_device
6165 * @fs_info: used only for generating a new devid, can be NULL if
6166 * devid is provided (i.e. @devid != NULL).
6167 * @devid: a pointer to devid for this device. If NULL a new devid
6169 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6172 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6173 * on error. Returned struct is not linked onto any lists and can be
6174 * destroyed with kfree() right away.
6176 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6180 struct btrfs_device *dev;
6183 if (WARN_ON(!devid && !fs_info))
6184 return ERR_PTR(-EINVAL);
6186 dev = __alloc_device();
6195 ret = find_next_devid(fs_info, &tmp);
6198 return ERR_PTR(ret);
6204 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6206 generate_random_uuid(dev->uuid);
6208 btrfs_init_work(&dev->work, btrfs_submit_helper,
6209 pending_bios_fn, NULL, NULL);
6214 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
6215 struct extent_buffer *leaf,
6216 struct btrfs_chunk *chunk)
6218 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
6219 struct map_lookup *map;
6220 struct extent_map *em;
6225 u8 uuid[BTRFS_UUID_SIZE];
6230 logical = key->offset;
6231 length = btrfs_chunk_length(leaf, chunk);
6232 stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6233 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6234 /* Validation check */
6236 btrfs_err(root->fs_info, "invalid chunk num_stripes: %u",
6240 if (!IS_ALIGNED(logical, root->sectorsize)) {
6241 btrfs_err(root->fs_info,
6242 "invalid chunk logical %llu", logical);
6245 if (!length || !IS_ALIGNED(length, root->sectorsize)) {
6246 btrfs_err(root->fs_info,
6247 "invalid chunk length %llu", length);
6250 if (!is_power_of_2(stripe_len)) {
6251 btrfs_err(root->fs_info, "invalid chunk stripe length: %llu",
6255 if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6256 btrfs_chunk_type(leaf, chunk)) {
6257 btrfs_err(root->fs_info, "unrecognized chunk type: %llu",
6258 ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
6259 BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6260 btrfs_chunk_type(leaf, chunk));
6264 read_lock(&map_tree->map_tree.lock);
6265 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
6266 read_unlock(&map_tree->map_tree.lock);
6268 /* already mapped? */
6269 if (em && em->start <= logical && em->start + em->len > logical) {
6270 free_extent_map(em);
6273 free_extent_map(em);
6276 em = alloc_extent_map();
6279 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6281 free_extent_map(em);
6285 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6286 em->map_lookup = map;
6287 em->start = logical;
6290 em->block_start = 0;
6291 em->block_len = em->len;
6293 map->num_stripes = num_stripes;
6294 map->io_width = btrfs_chunk_io_width(leaf, chunk);
6295 map->io_align = btrfs_chunk_io_align(leaf, chunk);
6296 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
6297 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6298 map->type = btrfs_chunk_type(leaf, chunk);
6299 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6300 for (i = 0; i < num_stripes; i++) {
6301 map->stripes[i].physical =
6302 btrfs_stripe_offset_nr(leaf, chunk, i);
6303 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6304 read_extent_buffer(leaf, uuid, (unsigned long)
6305 btrfs_stripe_dev_uuid_nr(chunk, i),
6307 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
6309 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
6310 free_extent_map(em);
6313 if (!map->stripes[i].dev) {
6314 map->stripes[i].dev =
6315 add_missing_dev(root, root->fs_info->fs_devices,
6317 if (!map->stripes[i].dev) {
6318 free_extent_map(em);
6321 btrfs_warn(root->fs_info, "devid %llu uuid %pU is missing",
6324 map->stripes[i].dev->in_fs_metadata = 1;
6327 write_lock(&map_tree->map_tree.lock);
6328 ret = add_extent_mapping(&map_tree->map_tree, em, 0);
6329 write_unlock(&map_tree->map_tree.lock);
6330 BUG_ON(ret); /* Tree corruption */
6331 free_extent_map(em);
6336 static void fill_device_from_item(struct extent_buffer *leaf,
6337 struct btrfs_dev_item *dev_item,
6338 struct btrfs_device *device)
6342 device->devid = btrfs_device_id(leaf, dev_item);
6343 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6344 device->total_bytes = device->disk_total_bytes;
6345 device->commit_total_bytes = device->disk_total_bytes;
6346 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6347 device->commit_bytes_used = device->bytes_used;
6348 device->type = btrfs_device_type(leaf, dev_item);
6349 device->io_align = btrfs_device_io_align(leaf, dev_item);
6350 device->io_width = btrfs_device_io_width(leaf, dev_item);
6351 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6352 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6353 device->is_tgtdev_for_dev_replace = 0;
6355 ptr = btrfs_device_uuid(dev_item);
6356 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6359 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_root *root,
6362 struct btrfs_fs_devices *fs_devices;
6365 BUG_ON(!mutex_is_locked(&uuid_mutex));
6367 fs_devices = root->fs_info->fs_devices->seed;
6368 while (fs_devices) {
6369 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE))
6372 fs_devices = fs_devices->seed;
6375 fs_devices = find_fsid(fsid);
6377 if (!btrfs_test_opt(root, DEGRADED))
6378 return ERR_PTR(-ENOENT);
6380 fs_devices = alloc_fs_devices(fsid);
6381 if (IS_ERR(fs_devices))
6384 fs_devices->seeding = 1;
6385 fs_devices->opened = 1;
6389 fs_devices = clone_fs_devices(fs_devices);
6390 if (IS_ERR(fs_devices))
6393 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
6394 root->fs_info->bdev_holder);
6396 free_fs_devices(fs_devices);
6397 fs_devices = ERR_PTR(ret);
6401 if (!fs_devices->seeding) {
6402 __btrfs_close_devices(fs_devices);
6403 free_fs_devices(fs_devices);
6404 fs_devices = ERR_PTR(-EINVAL);
6408 fs_devices->seed = root->fs_info->fs_devices->seed;
6409 root->fs_info->fs_devices->seed = fs_devices;
6414 static int read_one_dev(struct btrfs_root *root,
6415 struct extent_buffer *leaf,
6416 struct btrfs_dev_item *dev_item)
6418 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6419 struct btrfs_device *device;
6422 u8 fs_uuid[BTRFS_UUID_SIZE];
6423 u8 dev_uuid[BTRFS_UUID_SIZE];
6425 devid = btrfs_device_id(leaf, dev_item);
6426 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6428 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6431 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
6432 fs_devices = open_seed_devices(root, fs_uuid);
6433 if (IS_ERR(fs_devices))
6434 return PTR_ERR(fs_devices);
6437 device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
6439 if (!btrfs_test_opt(root, DEGRADED))
6442 device = add_missing_dev(root, fs_devices, devid, dev_uuid);
6445 btrfs_warn(root->fs_info, "devid %llu uuid %pU missing",
6448 if (!device->bdev && !btrfs_test_opt(root, DEGRADED))
6451 if(!device->bdev && !device->missing) {
6453 * this happens when a device that was properly setup
6454 * in the device info lists suddenly goes bad.
6455 * device->bdev is NULL, and so we have to set
6456 * device->missing to one here
6458 device->fs_devices->missing_devices++;
6459 device->missing = 1;
6462 /* Move the device to its own fs_devices */
6463 if (device->fs_devices != fs_devices) {
6464 ASSERT(device->missing);
6466 list_move(&device->dev_list, &fs_devices->devices);
6467 device->fs_devices->num_devices--;
6468 fs_devices->num_devices++;
6470 device->fs_devices->missing_devices--;
6471 fs_devices->missing_devices++;
6473 device->fs_devices = fs_devices;
6477 if (device->fs_devices != root->fs_info->fs_devices) {
6478 BUG_ON(device->writeable);
6479 if (device->generation !=
6480 btrfs_device_generation(leaf, dev_item))
6484 fill_device_from_item(leaf, dev_item, device);
6485 device->in_fs_metadata = 1;
6486 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
6487 device->fs_devices->total_rw_bytes += device->total_bytes;
6488 spin_lock(&root->fs_info->free_chunk_lock);
6489 root->fs_info->free_chunk_space += device->total_bytes -
6491 spin_unlock(&root->fs_info->free_chunk_lock);
6497 int btrfs_read_sys_array(struct btrfs_root *root)
6499 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
6500 struct extent_buffer *sb;
6501 struct btrfs_disk_key *disk_key;
6502 struct btrfs_chunk *chunk;
6504 unsigned long sb_array_offset;
6510 struct btrfs_key key;
6512 ASSERT(BTRFS_SUPER_INFO_SIZE <= root->nodesize);
6514 * This will create extent buffer of nodesize, superblock size is
6515 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6516 * overallocate but we can keep it as-is, only the first page is used.
6518 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET);
6521 set_extent_buffer_uptodate(sb);
6522 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6524 * The sb extent buffer is artifical and just used to read the system array.
6525 * set_extent_buffer_uptodate() call does not properly mark all it's
6526 * pages up-to-date when the page is larger: extent does not cover the
6527 * whole page and consequently check_page_uptodate does not find all
6528 * the page's extents up-to-date (the hole beyond sb),
6529 * write_extent_buffer then triggers a WARN_ON.
6531 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6532 * but sb spans only this function. Add an explicit SetPageUptodate call
6533 * to silence the warning eg. on PowerPC 64.
6535 if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
6536 SetPageUptodate(sb->pages[0]);
6538 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6539 array_size = btrfs_super_sys_array_size(super_copy);
6541 array_ptr = super_copy->sys_chunk_array;
6542 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6545 while (cur_offset < array_size) {
6546 disk_key = (struct btrfs_disk_key *)array_ptr;
6547 len = sizeof(*disk_key);
6548 if (cur_offset + len > array_size)
6549 goto out_short_read;
6551 btrfs_disk_key_to_cpu(&key, disk_key);
6554 sb_array_offset += len;
6557 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6558 chunk = (struct btrfs_chunk *)sb_array_offset;
6560 * At least one btrfs_chunk with one stripe must be
6561 * present, exact stripe count check comes afterwards
6563 len = btrfs_chunk_item_size(1);
6564 if (cur_offset + len > array_size)
6565 goto out_short_read;
6567 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6570 "BTRFS: invalid number of stripes %u in sys_array at offset %u\n",
6571 num_stripes, cur_offset);
6576 len = btrfs_chunk_item_size(num_stripes);
6577 if (cur_offset + len > array_size)
6578 goto out_short_read;
6580 ret = read_one_chunk(root, &key, sb, chunk);
6585 "BTRFS: unexpected item type %u in sys_array at offset %u\n",
6586 (u32)key.type, cur_offset);
6591 sb_array_offset += len;
6594 free_extent_buffer(sb);
6598 printk(KERN_ERR "BTRFS: sys_array too short to read %u bytes at offset %u\n",
6600 free_extent_buffer(sb);
6604 int btrfs_read_chunk_tree(struct btrfs_root *root)
6606 struct btrfs_path *path;
6607 struct extent_buffer *leaf;
6608 struct btrfs_key key;
6609 struct btrfs_key found_key;
6613 root = root->fs_info->chunk_root;
6615 path = btrfs_alloc_path();
6619 mutex_lock(&uuid_mutex);
6623 * Read all device items, and then all the chunk items. All
6624 * device items are found before any chunk item (their object id
6625 * is smaller than the lowest possible object id for a chunk
6626 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6628 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6631 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6635 leaf = path->nodes[0];
6636 slot = path->slots[0];
6637 if (slot >= btrfs_header_nritems(leaf)) {
6638 ret = btrfs_next_leaf(root, path);
6645 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6646 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6647 struct btrfs_dev_item *dev_item;
6648 dev_item = btrfs_item_ptr(leaf, slot,
6649 struct btrfs_dev_item);
6650 ret = read_one_dev(root, leaf, dev_item);
6653 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6654 struct btrfs_chunk *chunk;
6655 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6656 ret = read_one_chunk(root, &found_key, leaf, chunk);
6664 unlock_chunks(root);
6665 mutex_unlock(&uuid_mutex);
6667 btrfs_free_path(path);
6671 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6673 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6674 struct btrfs_device *device;
6676 while (fs_devices) {
6677 mutex_lock(&fs_devices->device_list_mutex);
6678 list_for_each_entry(device, &fs_devices->devices, dev_list)
6679 device->dev_root = fs_info->dev_root;
6680 mutex_unlock(&fs_devices->device_list_mutex);
6682 fs_devices = fs_devices->seed;
6686 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6690 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6691 btrfs_dev_stat_reset(dev, i);
6694 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6696 struct btrfs_key key;
6697 struct btrfs_key found_key;
6698 struct btrfs_root *dev_root = fs_info->dev_root;
6699 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6700 struct extent_buffer *eb;
6703 struct btrfs_device *device;
6704 struct btrfs_path *path = NULL;
6707 path = btrfs_alloc_path();
6713 mutex_lock(&fs_devices->device_list_mutex);
6714 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6716 struct btrfs_dev_stats_item *ptr;
6718 key.objectid = BTRFS_DEV_STATS_OBJECTID;
6719 key.type = BTRFS_PERSISTENT_ITEM_KEY;
6720 key.offset = device->devid;
6721 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6723 __btrfs_reset_dev_stats(device);
6724 device->dev_stats_valid = 1;
6725 btrfs_release_path(path);
6728 slot = path->slots[0];
6729 eb = path->nodes[0];
6730 btrfs_item_key_to_cpu(eb, &found_key, slot);
6731 item_size = btrfs_item_size_nr(eb, slot);
6733 ptr = btrfs_item_ptr(eb, slot,
6734 struct btrfs_dev_stats_item);
6736 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6737 if (item_size >= (1 + i) * sizeof(__le64))
6738 btrfs_dev_stat_set(device, i,
6739 btrfs_dev_stats_value(eb, ptr, i));
6741 btrfs_dev_stat_reset(device, i);
6744 device->dev_stats_valid = 1;
6745 btrfs_dev_stat_print_on_load(device);
6746 btrfs_release_path(path);
6748 mutex_unlock(&fs_devices->device_list_mutex);
6751 btrfs_free_path(path);
6752 return ret < 0 ? ret : 0;
6755 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
6756 struct btrfs_root *dev_root,
6757 struct btrfs_device *device)
6759 struct btrfs_path *path;
6760 struct btrfs_key key;
6761 struct extent_buffer *eb;
6762 struct btrfs_dev_stats_item *ptr;
6766 key.objectid = BTRFS_DEV_STATS_OBJECTID;
6767 key.type = BTRFS_PERSISTENT_ITEM_KEY;
6768 key.offset = device->devid;
6770 path = btrfs_alloc_path();
6772 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
6774 btrfs_warn_in_rcu(dev_root->fs_info,
6775 "error %d while searching for dev_stats item for device %s",
6776 ret, rcu_str_deref(device->name));
6781 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
6782 /* need to delete old one and insert a new one */
6783 ret = btrfs_del_item(trans, dev_root, path);
6785 btrfs_warn_in_rcu(dev_root->fs_info,
6786 "delete too small dev_stats item for device %s failed %d",
6787 rcu_str_deref(device->name), ret);
6794 /* need to insert a new item */
6795 btrfs_release_path(path);
6796 ret = btrfs_insert_empty_item(trans, dev_root, path,
6797 &key, sizeof(*ptr));
6799 btrfs_warn_in_rcu(dev_root->fs_info,
6800 "insert dev_stats item for device %s failed %d",
6801 rcu_str_deref(device->name), ret);
6806 eb = path->nodes[0];
6807 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
6808 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6809 btrfs_set_dev_stats_value(eb, ptr, i,
6810 btrfs_dev_stat_read(device, i));
6811 btrfs_mark_buffer_dirty(eb);
6814 btrfs_free_path(path);
6819 * called from commit_transaction. Writes all changed device stats to disk.
6821 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
6822 struct btrfs_fs_info *fs_info)
6824 struct btrfs_root *dev_root = fs_info->dev_root;
6825 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6826 struct btrfs_device *device;
6830 mutex_lock(&fs_devices->device_list_mutex);
6831 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6832 if (!device->dev_stats_valid || !btrfs_dev_stats_dirty(device))
6835 stats_cnt = atomic_read(&device->dev_stats_ccnt);
6836 ret = update_dev_stat_item(trans, dev_root, device);
6838 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
6840 mutex_unlock(&fs_devices->device_list_mutex);
6845 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
6847 btrfs_dev_stat_inc(dev, index);
6848 btrfs_dev_stat_print_on_error(dev);
6851 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
6853 if (!dev->dev_stats_valid)
6855 btrfs_err_rl_in_rcu(dev->dev_root->fs_info,
6856 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
6857 rcu_str_deref(dev->name),
6858 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6859 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6860 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6861 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6862 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6865 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
6869 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6870 if (btrfs_dev_stat_read(dev, i) != 0)
6872 if (i == BTRFS_DEV_STAT_VALUES_MAX)
6873 return; /* all values == 0, suppress message */
6875 btrfs_info_in_rcu(dev->dev_root->fs_info,
6876 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
6877 rcu_str_deref(dev->name),
6878 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6879 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6880 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6881 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6882 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6885 int btrfs_get_dev_stats(struct btrfs_root *root,
6886 struct btrfs_ioctl_get_dev_stats *stats)
6888 struct btrfs_device *dev;
6889 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6892 mutex_lock(&fs_devices->device_list_mutex);
6893 dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
6894 mutex_unlock(&fs_devices->device_list_mutex);
6897 btrfs_warn(root->fs_info, "get dev_stats failed, device not found");
6899 } else if (!dev->dev_stats_valid) {
6900 btrfs_warn(root->fs_info, "get dev_stats failed, not yet valid");
6902 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
6903 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6904 if (stats->nr_items > i)
6906 btrfs_dev_stat_read_and_reset(dev, i);
6908 btrfs_dev_stat_reset(dev, i);
6911 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6912 if (stats->nr_items > i)
6913 stats->values[i] = btrfs_dev_stat_read(dev, i);
6915 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
6916 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
6920 void btrfs_scratch_superblocks(struct block_device *bdev, char *device_path)
6922 struct buffer_head *bh;
6923 struct btrfs_super_block *disk_super;
6929 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX;
6932 if (btrfs_read_dev_one_super(bdev, copy_num, &bh))
6935 disk_super = (struct btrfs_super_block *)bh->b_data;
6937 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
6938 set_buffer_dirty(bh);
6939 sync_dirty_buffer(bh);
6943 /* Notify udev that device has changed */
6944 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
6946 /* Update ctime/mtime for device path for libblkid */
6947 update_dev_time(device_path);
6951 * Update the size of all devices, which is used for writing out the
6954 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
6956 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6957 struct btrfs_device *curr, *next;
6959 if (list_empty(&fs_devices->resized_devices))
6962 mutex_lock(&fs_devices->device_list_mutex);
6963 lock_chunks(fs_info->dev_root);
6964 list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
6966 list_del_init(&curr->resized_list);
6967 curr->commit_total_bytes = curr->disk_total_bytes;
6969 unlock_chunks(fs_info->dev_root);
6970 mutex_unlock(&fs_devices->device_list_mutex);
6973 /* Must be invoked during the transaction commit */
6974 void btrfs_update_commit_device_bytes_used(struct btrfs_root *root,
6975 struct btrfs_transaction *transaction)
6977 struct extent_map *em;
6978 struct map_lookup *map;
6979 struct btrfs_device *dev;
6982 if (list_empty(&transaction->pending_chunks))
6985 /* In order to kick the device replace finish process */
6987 list_for_each_entry(em, &transaction->pending_chunks, list) {
6988 map = em->map_lookup;
6990 for (i = 0; i < map->num_stripes; i++) {
6991 dev = map->stripes[i].dev;
6992 dev->commit_bytes_used = dev->bytes_used;
6995 unlock_chunks(root);
6998 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
7000 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7001 while (fs_devices) {
7002 fs_devices->fs_info = fs_info;
7003 fs_devices = fs_devices->seed;
7007 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
7009 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7010 while (fs_devices) {
7011 fs_devices->fs_info = NULL;
7012 fs_devices = fs_devices->seed;
7016 static void btrfs_close_one_device(struct btrfs_device *device)
7018 struct btrfs_fs_devices *fs_devices = device->fs_devices;
7019 struct btrfs_device *new_device;
7020 struct rcu_string *name;
7023 fs_devices->open_devices--;
7025 if (device->writeable &&
7026 device->devid != BTRFS_DEV_REPLACE_DEVID) {
7027 list_del_init(&device->dev_alloc_list);
7028 fs_devices->rw_devices--;
7031 if (device->missing)
7032 fs_devices->missing_devices--;
7034 new_device = btrfs_alloc_device(NULL, &device->devid,
7036 BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
7038 /* Safe because we are under uuid_mutex */
7040 name = rcu_string_strdup(device->name->str, GFP_NOFS);
7041 BUG_ON(!name); /* -ENOMEM */
7042 rcu_assign_pointer(new_device->name, name);
7045 list_replace_rcu(&device->dev_list, &new_device->dev_list);
7046 new_device->fs_devices = device->fs_devices;
7048 call_rcu(&device->rcu, free_device);