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);
991 void btrfs_release_disk_super(struct page *page)
997 int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr,
998 struct page **page, struct btrfs_super_block **disk_super)
1003 /* make sure our super fits in the device */
1004 if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1007 /* make sure our super fits in the page */
1008 if (sizeof(**disk_super) > PAGE_SIZE)
1011 /* make sure our super doesn't straddle pages on disk */
1012 index = bytenr >> PAGE_SHIFT;
1013 if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index)
1016 /* pull in the page with our super */
1017 *page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
1020 if (IS_ERR_OR_NULL(*page))
1025 /* align our pointer to the offset of the super block */
1026 *disk_super = p + (bytenr & ~PAGE_MASK);
1028 if (btrfs_super_bytenr(*disk_super) != bytenr ||
1029 btrfs_super_magic(*disk_super) != BTRFS_MAGIC) {
1030 btrfs_release_disk_super(*page);
1034 if ((*disk_super)->label[0] &&
1035 (*disk_super)->label[BTRFS_LABEL_SIZE - 1])
1036 (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0';
1042 * Look for a btrfs signature on a device. This may be called out of the mount path
1043 * and we are not allowed to call set_blocksize during the scan. The superblock
1044 * is read via pagecache
1046 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
1047 struct btrfs_fs_devices **fs_devices_ret)
1049 struct btrfs_super_block *disk_super;
1050 struct block_device *bdev;
1059 * we would like to check all the supers, but that would make
1060 * a btrfs mount succeed after a mkfs from a different FS.
1061 * So, we need to add a special mount option to scan for
1062 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1064 bytenr = btrfs_sb_offset(0);
1065 flags |= FMODE_EXCL;
1066 mutex_lock(&uuid_mutex);
1068 bdev = blkdev_get_by_path(path, flags, holder);
1070 ret = PTR_ERR(bdev);
1074 if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super))
1075 goto error_bdev_put;
1077 devid = btrfs_stack_device_id(&disk_super->dev_item);
1078 transid = btrfs_super_generation(disk_super);
1079 total_devices = btrfs_super_num_devices(disk_super);
1081 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
1083 if (disk_super->label[0]) {
1084 printk(KERN_INFO "BTRFS: device label %s ", disk_super->label);
1086 printk(KERN_INFO "BTRFS: device fsid %pU ", disk_super->fsid);
1089 printk(KERN_CONT "devid %llu transid %llu %s\n", devid, transid, path);
1092 if (!ret && fs_devices_ret)
1093 (*fs_devices_ret)->total_devices = total_devices;
1095 btrfs_release_disk_super(page);
1098 blkdev_put(bdev, flags);
1100 mutex_unlock(&uuid_mutex);
1104 /* helper to account the used device space in the range */
1105 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
1106 u64 end, u64 *length)
1108 struct btrfs_key key;
1109 struct btrfs_root *root = device->dev_root;
1110 struct btrfs_dev_extent *dev_extent;
1111 struct btrfs_path *path;
1115 struct extent_buffer *l;
1119 if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
1122 path = btrfs_alloc_path();
1125 path->reada = READA_FORWARD;
1127 key.objectid = device->devid;
1129 key.type = BTRFS_DEV_EXTENT_KEY;
1131 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1135 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1142 slot = path->slots[0];
1143 if (slot >= btrfs_header_nritems(l)) {
1144 ret = btrfs_next_leaf(root, path);
1152 btrfs_item_key_to_cpu(l, &key, slot);
1154 if (key.objectid < device->devid)
1157 if (key.objectid > device->devid)
1160 if (key.type != BTRFS_DEV_EXTENT_KEY)
1163 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1164 extent_end = key.offset + btrfs_dev_extent_length(l,
1166 if (key.offset <= start && extent_end > end) {
1167 *length = end - start + 1;
1169 } else if (key.offset <= start && extent_end > start)
1170 *length += extent_end - start;
1171 else if (key.offset > start && extent_end <= end)
1172 *length += extent_end - key.offset;
1173 else if (key.offset > start && key.offset <= end) {
1174 *length += end - key.offset + 1;
1176 } else if (key.offset > end)
1184 btrfs_free_path(path);
1188 static int contains_pending_extent(struct btrfs_transaction *transaction,
1189 struct btrfs_device *device,
1190 u64 *start, u64 len)
1192 struct btrfs_fs_info *fs_info = device->dev_root->fs_info;
1193 struct extent_map *em;
1194 struct list_head *search_list = &fs_info->pinned_chunks;
1196 u64 physical_start = *start;
1199 search_list = &transaction->pending_chunks;
1201 list_for_each_entry(em, search_list, list) {
1202 struct map_lookup *map;
1205 map = em->map_lookup;
1206 for (i = 0; i < map->num_stripes; i++) {
1209 if (map->stripes[i].dev != device)
1211 if (map->stripes[i].physical >= physical_start + len ||
1212 map->stripes[i].physical + em->orig_block_len <=
1216 * Make sure that while processing the pinned list we do
1217 * not override our *start with a lower value, because
1218 * we can have pinned chunks that fall within this
1219 * device hole and that have lower physical addresses
1220 * than the pending chunks we processed before. If we
1221 * do not take this special care we can end up getting
1222 * 2 pending chunks that start at the same physical
1223 * device offsets because the end offset of a pinned
1224 * chunk can be equal to the start offset of some
1227 end = map->stripes[i].physical + em->orig_block_len;
1234 if (search_list != &fs_info->pinned_chunks) {
1235 search_list = &fs_info->pinned_chunks;
1244 * find_free_dev_extent_start - find free space in the specified device
1245 * @device: the device which we search the free space in
1246 * @num_bytes: the size of the free space that we need
1247 * @search_start: the position from which to begin the search
1248 * @start: store the start of the free space.
1249 * @len: the size of the free space. that we find, or the size
1250 * of the max free space if we don't find suitable free space
1252 * this uses a pretty simple search, the expectation is that it is
1253 * called very infrequently and that a given device has a small number
1256 * @start is used to store the start of the free space if we find. But if we
1257 * don't find suitable free space, it will be used to store the start position
1258 * of the max free space.
1260 * @len is used to store the size of the free space that we find.
1261 * But if we don't find suitable free space, it is used to store the size of
1262 * the max free space.
1264 int find_free_dev_extent_start(struct btrfs_transaction *transaction,
1265 struct btrfs_device *device, u64 num_bytes,
1266 u64 search_start, u64 *start, u64 *len)
1268 struct btrfs_key key;
1269 struct btrfs_root *root = device->dev_root;
1270 struct btrfs_dev_extent *dev_extent;
1271 struct btrfs_path *path;
1276 u64 search_end = device->total_bytes;
1279 struct extent_buffer *l;
1280 u64 min_search_start;
1283 * We don't want to overwrite the superblock on the drive nor any area
1284 * used by the boot loader (grub for example), so we make sure to start
1285 * at an offset of at least 1MB.
1287 min_search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
1288 search_start = max(search_start, min_search_start);
1290 path = btrfs_alloc_path();
1294 max_hole_start = search_start;
1298 if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1303 path->reada = READA_FORWARD;
1304 path->search_commit_root = 1;
1305 path->skip_locking = 1;
1307 key.objectid = device->devid;
1308 key.offset = search_start;
1309 key.type = BTRFS_DEV_EXTENT_KEY;
1311 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1315 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1322 slot = path->slots[0];
1323 if (slot >= btrfs_header_nritems(l)) {
1324 ret = btrfs_next_leaf(root, path);
1332 btrfs_item_key_to_cpu(l, &key, slot);
1334 if (key.objectid < device->devid)
1337 if (key.objectid > device->devid)
1340 if (key.type != BTRFS_DEV_EXTENT_KEY)
1343 if (key.offset > search_start) {
1344 hole_size = key.offset - search_start;
1347 * Have to check before we set max_hole_start, otherwise
1348 * we could end up sending back this offset anyway.
1350 if (contains_pending_extent(transaction, device,
1353 if (key.offset >= search_start) {
1354 hole_size = key.offset - search_start;
1361 if (hole_size > max_hole_size) {
1362 max_hole_start = search_start;
1363 max_hole_size = hole_size;
1367 * If this free space is greater than which we need,
1368 * it must be the max free space that we have found
1369 * until now, so max_hole_start must point to the start
1370 * of this free space and the length of this free space
1371 * is stored in max_hole_size. Thus, we return
1372 * max_hole_start and max_hole_size and go back to the
1375 if (hole_size >= num_bytes) {
1381 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1382 extent_end = key.offset + btrfs_dev_extent_length(l,
1384 if (extent_end > search_start)
1385 search_start = extent_end;
1392 * At this point, search_start should be the end of
1393 * allocated dev extents, and when shrinking the device,
1394 * search_end may be smaller than search_start.
1396 if (search_end > search_start) {
1397 hole_size = search_end - search_start;
1399 if (contains_pending_extent(transaction, device, &search_start,
1401 btrfs_release_path(path);
1405 if (hole_size > max_hole_size) {
1406 max_hole_start = search_start;
1407 max_hole_size = hole_size;
1412 if (max_hole_size < num_bytes)
1418 btrfs_free_path(path);
1419 *start = max_hole_start;
1421 *len = max_hole_size;
1425 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1426 struct btrfs_device *device, u64 num_bytes,
1427 u64 *start, u64 *len)
1429 /* FIXME use last free of some kind */
1430 return find_free_dev_extent_start(trans->transaction, device,
1431 num_bytes, 0, start, len);
1434 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1435 struct btrfs_device *device,
1436 u64 start, u64 *dev_extent_len)
1439 struct btrfs_path *path;
1440 struct btrfs_root *root = device->dev_root;
1441 struct btrfs_key key;
1442 struct btrfs_key found_key;
1443 struct extent_buffer *leaf = NULL;
1444 struct btrfs_dev_extent *extent = NULL;
1446 path = btrfs_alloc_path();
1450 key.objectid = device->devid;
1452 key.type = BTRFS_DEV_EXTENT_KEY;
1454 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1456 ret = btrfs_previous_item(root, path, key.objectid,
1457 BTRFS_DEV_EXTENT_KEY);
1460 leaf = path->nodes[0];
1461 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1462 extent = btrfs_item_ptr(leaf, path->slots[0],
1463 struct btrfs_dev_extent);
1464 BUG_ON(found_key.offset > start || found_key.offset +
1465 btrfs_dev_extent_length(leaf, extent) < start);
1467 btrfs_release_path(path);
1469 } else if (ret == 0) {
1470 leaf = path->nodes[0];
1471 extent = btrfs_item_ptr(leaf, path->slots[0],
1472 struct btrfs_dev_extent);
1474 btrfs_std_error(root->fs_info, ret, "Slot search failed");
1478 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1480 ret = btrfs_del_item(trans, root, path);
1482 btrfs_std_error(root->fs_info, ret,
1483 "Failed to remove dev extent item");
1485 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1488 btrfs_free_path(path);
1492 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1493 struct btrfs_device *device,
1494 u64 chunk_tree, u64 chunk_objectid,
1495 u64 chunk_offset, u64 start, u64 num_bytes)
1498 struct btrfs_path *path;
1499 struct btrfs_root *root = device->dev_root;
1500 struct btrfs_dev_extent *extent;
1501 struct extent_buffer *leaf;
1502 struct btrfs_key key;
1504 WARN_ON(!device->in_fs_metadata);
1505 WARN_ON(device->is_tgtdev_for_dev_replace);
1506 path = btrfs_alloc_path();
1510 key.objectid = device->devid;
1512 key.type = BTRFS_DEV_EXTENT_KEY;
1513 ret = btrfs_insert_empty_item(trans, root, path, &key,
1518 leaf = path->nodes[0];
1519 extent = btrfs_item_ptr(leaf, path->slots[0],
1520 struct btrfs_dev_extent);
1521 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1522 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1523 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1525 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1526 btrfs_dev_extent_chunk_tree_uuid(extent), BTRFS_UUID_SIZE);
1528 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1529 btrfs_mark_buffer_dirty(leaf);
1531 btrfs_free_path(path);
1535 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1537 struct extent_map_tree *em_tree;
1538 struct extent_map *em;
1542 em_tree = &fs_info->mapping_tree.map_tree;
1543 read_lock(&em_tree->lock);
1544 n = rb_last(&em_tree->map);
1546 em = rb_entry(n, struct extent_map, rb_node);
1547 ret = em->start + em->len;
1549 read_unlock(&em_tree->lock);
1554 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1558 struct btrfs_key key;
1559 struct btrfs_key found_key;
1560 struct btrfs_path *path;
1562 path = btrfs_alloc_path();
1566 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1567 key.type = BTRFS_DEV_ITEM_KEY;
1568 key.offset = (u64)-1;
1570 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1574 BUG_ON(ret == 0); /* Corruption */
1576 ret = btrfs_previous_item(fs_info->chunk_root, path,
1577 BTRFS_DEV_ITEMS_OBJECTID,
1578 BTRFS_DEV_ITEM_KEY);
1582 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1584 *devid_ret = found_key.offset + 1;
1588 btrfs_free_path(path);
1593 * the device information is stored in the chunk root
1594 * the btrfs_device struct should be fully filled in
1596 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1597 struct btrfs_root *root,
1598 struct btrfs_device *device)
1601 struct btrfs_path *path;
1602 struct btrfs_dev_item *dev_item;
1603 struct extent_buffer *leaf;
1604 struct btrfs_key key;
1607 root = root->fs_info->chunk_root;
1609 path = btrfs_alloc_path();
1613 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1614 key.type = BTRFS_DEV_ITEM_KEY;
1615 key.offset = device->devid;
1617 ret = btrfs_insert_empty_item(trans, root, path, &key,
1622 leaf = path->nodes[0];
1623 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1625 btrfs_set_device_id(leaf, dev_item, device->devid);
1626 btrfs_set_device_generation(leaf, dev_item, 0);
1627 btrfs_set_device_type(leaf, dev_item, device->type);
1628 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1629 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1630 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1631 btrfs_set_device_total_bytes(leaf, dev_item,
1632 btrfs_device_get_disk_total_bytes(device));
1633 btrfs_set_device_bytes_used(leaf, dev_item,
1634 btrfs_device_get_bytes_used(device));
1635 btrfs_set_device_group(leaf, dev_item, 0);
1636 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1637 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1638 btrfs_set_device_start_offset(leaf, dev_item, 0);
1640 ptr = btrfs_device_uuid(dev_item);
1641 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1642 ptr = btrfs_device_fsid(dev_item);
1643 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1644 btrfs_mark_buffer_dirty(leaf);
1648 btrfs_free_path(path);
1653 * Function to update ctime/mtime for a given device path.
1654 * Mainly used for ctime/mtime based probe like libblkid.
1656 static void update_dev_time(char *path_name)
1660 filp = filp_open(path_name, O_RDWR, 0);
1663 file_update_time(filp);
1664 filp_close(filp, NULL);
1667 static int btrfs_rm_dev_item(struct btrfs_root *root,
1668 struct btrfs_device *device)
1671 struct btrfs_path *path;
1672 struct btrfs_key key;
1673 struct btrfs_trans_handle *trans;
1675 root = root->fs_info->chunk_root;
1677 path = btrfs_alloc_path();
1681 trans = btrfs_start_transaction(root, 0);
1682 if (IS_ERR(trans)) {
1683 btrfs_free_path(path);
1684 return PTR_ERR(trans);
1686 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1687 key.type = BTRFS_DEV_ITEM_KEY;
1688 key.offset = device->devid;
1690 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1699 ret = btrfs_del_item(trans, root, path);
1703 btrfs_free_path(path);
1704 btrfs_commit_transaction(trans, root);
1708 static int __check_raid_min_devices(struct btrfs_fs_info *fs_info)
1714 num_devices = fs_info->fs_devices->num_devices;
1715 btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
1716 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
1717 WARN_ON(num_devices < 1);
1720 btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
1723 seq = read_seqbegin(&fs_info->profiles_lock);
1725 all_avail = fs_info->avail_data_alloc_bits |
1726 fs_info->avail_system_alloc_bits |
1727 fs_info->avail_metadata_alloc_bits;
1728 } while (read_seqretry(&fs_info->profiles_lock, seq));
1730 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && num_devices <= 4) {
1731 return BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET;
1734 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && num_devices <= 2) {
1735 return BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET;
1738 if ((all_avail & BTRFS_BLOCK_GROUP_RAID5) &&
1739 fs_info->fs_devices->rw_devices <= 2) {
1740 return BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET;
1743 if ((all_avail & BTRFS_BLOCK_GROUP_RAID6) &&
1744 fs_info->fs_devices->rw_devices <= 3) {
1745 return BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET;
1751 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1753 struct btrfs_device *device;
1754 struct btrfs_device *next_device;
1755 struct block_device *bdev = NULL;
1756 struct buffer_head *bh = NULL;
1757 struct btrfs_super_block *disk_super = NULL;
1758 struct btrfs_fs_devices *cur_devices;
1761 bool clear_super = false;
1763 mutex_lock(&uuid_mutex);
1765 ret = __check_raid_min_devices(root->fs_info);
1769 ret = btrfs_find_device_by_user_input(root, 0, device_path,
1774 if (device->is_tgtdev_for_dev_replace) {
1775 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1779 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1780 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1784 if (device->writeable) {
1786 list_del_init(&device->dev_alloc_list);
1787 device->fs_devices->rw_devices--;
1788 unlock_chunks(root);
1792 mutex_unlock(&uuid_mutex);
1793 ret = btrfs_shrink_device(device, 0);
1794 mutex_lock(&uuid_mutex);
1799 * TODO: the superblock still includes this device in its num_devices
1800 * counter although write_all_supers() is not locked out. This
1801 * could give a filesystem state which requires a degraded mount.
1803 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1807 device->in_fs_metadata = 0;
1808 btrfs_scrub_cancel_dev(root->fs_info, device);
1811 * the device list mutex makes sure that we don't change
1812 * the device list while someone else is writing out all
1813 * the device supers. Whoever is writing all supers, should
1814 * lock the device list mutex before getting the number of
1815 * devices in the super block (super_copy). Conversely,
1816 * whoever updates the number of devices in the super block
1817 * (super_copy) should hold the device list mutex.
1820 cur_devices = device->fs_devices;
1821 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1822 list_del_rcu(&device->dev_list);
1824 device->fs_devices->num_devices--;
1825 device->fs_devices->total_devices--;
1827 if (device->missing)
1828 device->fs_devices->missing_devices--;
1830 next_device = list_entry(root->fs_info->fs_devices->devices.next,
1831 struct btrfs_device, dev_list);
1832 if (device->bdev == root->fs_info->sb->s_bdev)
1833 root->fs_info->sb->s_bdev = next_device->bdev;
1834 if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1835 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1838 device->fs_devices->open_devices--;
1839 /* remove sysfs entry */
1840 btrfs_sysfs_rm_device_link(root->fs_info->fs_devices, device);
1843 call_rcu(&device->rcu, free_device);
1845 num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1846 btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1847 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1849 if (cur_devices->open_devices == 0) {
1850 struct btrfs_fs_devices *fs_devices;
1851 fs_devices = root->fs_info->fs_devices;
1852 while (fs_devices) {
1853 if (fs_devices->seed == cur_devices) {
1854 fs_devices->seed = cur_devices->seed;
1857 fs_devices = fs_devices->seed;
1859 cur_devices->seed = NULL;
1860 __btrfs_close_devices(cur_devices);
1861 free_fs_devices(cur_devices);
1864 root->fs_info->num_tolerated_disk_barrier_failures =
1865 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1868 * at this point, the device is zero sized. We want to
1869 * remove it from the devices list and zero out the old super
1876 ret = btrfs_get_bdev_and_sb(device_path,
1877 FMODE_WRITE | FMODE_EXCL,
1878 root->fs_info->bdev_holder, 0,
1882 * It could be a failed device ok for clear_super
1883 * to fail. So return success
1889 disk_super = (struct btrfs_super_block *)bh->b_data;
1891 /* make sure this device isn't detected as part of
1894 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1895 set_buffer_dirty(bh);
1896 sync_dirty_buffer(bh);
1899 /* clear the mirror copies of super block on the disk
1900 * being removed, 0th copy is been taken care above and
1901 * the below would take of the rest
1903 for (i = 1; i < BTRFS_SUPER_MIRROR_MAX; i++) {
1904 bytenr = btrfs_sb_offset(i);
1905 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
1906 i_size_read(bdev->bd_inode))
1909 bh = __bread(bdev, bytenr / 4096,
1910 BTRFS_SUPER_INFO_SIZE);
1914 disk_super = (struct btrfs_super_block *)bh->b_data;
1916 if (btrfs_super_bytenr(disk_super) != bytenr ||
1917 btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
1921 memset(&disk_super->magic, 0,
1922 sizeof(disk_super->magic));
1923 set_buffer_dirty(bh);
1924 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);
1934 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1939 mutex_unlock(&uuid_mutex);
1943 if (device->writeable) {
1945 list_add(&device->dev_alloc_list,
1946 &root->fs_info->fs_devices->alloc_list);
1947 device->fs_devices->rw_devices++;
1948 unlock_chunks(root);
1953 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
1954 struct btrfs_device *srcdev)
1956 struct btrfs_fs_devices *fs_devices;
1958 WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1961 * in case of fs with no seed, srcdev->fs_devices will point
1962 * to fs_devices of fs_info. However when the dev being replaced is
1963 * a seed dev it will point to the seed's local fs_devices. In short
1964 * srcdev will have its correct fs_devices in both the cases.
1966 fs_devices = srcdev->fs_devices;
1968 list_del_rcu(&srcdev->dev_list);
1969 list_del_rcu(&srcdev->dev_alloc_list);
1970 fs_devices->num_devices--;
1971 if (srcdev->missing)
1972 fs_devices->missing_devices--;
1974 if (srcdev->writeable) {
1975 fs_devices->rw_devices--;
1976 /* zero out the old super if it is writable */
1977 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
1981 fs_devices->open_devices--;
1984 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
1985 struct btrfs_device *srcdev)
1987 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
1989 call_rcu(&srcdev->rcu, free_device);
1992 * unless fs_devices is seed fs, num_devices shouldn't go
1995 BUG_ON(!fs_devices->num_devices && !fs_devices->seeding);
1997 /* if this is no devs we rather delete the fs_devices */
1998 if (!fs_devices->num_devices) {
1999 struct btrfs_fs_devices *tmp_fs_devices;
2001 tmp_fs_devices = fs_info->fs_devices;
2002 while (tmp_fs_devices) {
2003 if (tmp_fs_devices->seed == fs_devices) {
2004 tmp_fs_devices->seed = fs_devices->seed;
2007 tmp_fs_devices = tmp_fs_devices->seed;
2009 fs_devices->seed = NULL;
2010 __btrfs_close_devices(fs_devices);
2011 free_fs_devices(fs_devices);
2015 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2016 struct btrfs_device *tgtdev)
2018 struct btrfs_device *next_device;
2020 mutex_lock(&uuid_mutex);
2022 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2024 btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev);
2027 btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
2028 fs_info->fs_devices->open_devices--;
2030 fs_info->fs_devices->num_devices--;
2032 next_device = list_entry(fs_info->fs_devices->devices.next,
2033 struct btrfs_device, dev_list);
2034 if (tgtdev->bdev == fs_info->sb->s_bdev)
2035 fs_info->sb->s_bdev = next_device->bdev;
2036 if (tgtdev->bdev == fs_info->fs_devices->latest_bdev)
2037 fs_info->fs_devices->latest_bdev = next_device->bdev;
2038 list_del_rcu(&tgtdev->dev_list);
2040 call_rcu(&tgtdev->rcu, free_device);
2042 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2043 mutex_unlock(&uuid_mutex);
2046 static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
2047 struct btrfs_device **device)
2050 struct btrfs_super_block *disk_super;
2053 struct block_device *bdev;
2054 struct buffer_head *bh;
2057 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2058 root->fs_info->bdev_holder, 0, &bdev, &bh);
2061 disk_super = (struct btrfs_super_block *)bh->b_data;
2062 devid = btrfs_stack_device_id(&disk_super->dev_item);
2063 dev_uuid = disk_super->dev_item.uuid;
2064 *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2069 blkdev_put(bdev, FMODE_READ);
2073 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
2075 struct btrfs_device **device)
2078 if (strcmp(device_path, "missing") == 0) {
2079 struct list_head *devices;
2080 struct btrfs_device *tmp;
2082 devices = &root->fs_info->fs_devices->devices;
2084 * It is safe to read the devices since the volume_mutex
2085 * is held by the caller.
2087 list_for_each_entry(tmp, devices, dev_list) {
2088 if (tmp->in_fs_metadata && !tmp->bdev) {
2095 return BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2099 return btrfs_find_device_by_path(root, device_path, device);
2103 int btrfs_find_device_by_user_input(struct btrfs_root *root, u64 srcdevid,
2105 struct btrfs_device **device)
2111 *device = btrfs_find_device(root->fs_info, srcdevid, NULL,
2116 ret = btrfs_find_device_missing_or_by_path(root, srcdev_name,
2123 * does all the dirty work required for changing file system's UUID.
2125 static int btrfs_prepare_sprout(struct btrfs_root *root)
2127 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2128 struct btrfs_fs_devices *old_devices;
2129 struct btrfs_fs_devices *seed_devices;
2130 struct btrfs_super_block *disk_super = root->fs_info->super_copy;
2131 struct btrfs_device *device;
2134 BUG_ON(!mutex_is_locked(&uuid_mutex));
2135 if (!fs_devices->seeding)
2138 seed_devices = __alloc_fs_devices();
2139 if (IS_ERR(seed_devices))
2140 return PTR_ERR(seed_devices);
2142 old_devices = clone_fs_devices(fs_devices);
2143 if (IS_ERR(old_devices)) {
2144 kfree(seed_devices);
2145 return PTR_ERR(old_devices);
2148 list_add(&old_devices->list, &fs_uuids);
2150 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2151 seed_devices->opened = 1;
2152 INIT_LIST_HEAD(&seed_devices->devices);
2153 INIT_LIST_HEAD(&seed_devices->alloc_list);
2154 mutex_init(&seed_devices->device_list_mutex);
2156 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2157 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2159 list_for_each_entry(device, &seed_devices->devices, dev_list)
2160 device->fs_devices = seed_devices;
2163 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2164 unlock_chunks(root);
2166 fs_devices->seeding = 0;
2167 fs_devices->num_devices = 0;
2168 fs_devices->open_devices = 0;
2169 fs_devices->missing_devices = 0;
2170 fs_devices->rotating = 0;
2171 fs_devices->seed = seed_devices;
2173 generate_random_uuid(fs_devices->fsid);
2174 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2175 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2176 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2178 super_flags = btrfs_super_flags(disk_super) &
2179 ~BTRFS_SUPER_FLAG_SEEDING;
2180 btrfs_set_super_flags(disk_super, super_flags);
2186 * strore the expected generation for seed devices in device items.
2188 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2189 struct btrfs_root *root)
2191 struct btrfs_path *path;
2192 struct extent_buffer *leaf;
2193 struct btrfs_dev_item *dev_item;
2194 struct btrfs_device *device;
2195 struct btrfs_key key;
2196 u8 fs_uuid[BTRFS_UUID_SIZE];
2197 u8 dev_uuid[BTRFS_UUID_SIZE];
2201 path = btrfs_alloc_path();
2205 root = root->fs_info->chunk_root;
2206 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2208 key.type = BTRFS_DEV_ITEM_KEY;
2211 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2215 leaf = path->nodes[0];
2217 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2218 ret = btrfs_next_leaf(root, path);
2223 leaf = path->nodes[0];
2224 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2225 btrfs_release_path(path);
2229 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2230 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2231 key.type != BTRFS_DEV_ITEM_KEY)
2234 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2235 struct btrfs_dev_item);
2236 devid = btrfs_device_id(leaf, dev_item);
2237 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2239 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2241 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2243 BUG_ON(!device); /* Logic error */
2245 if (device->fs_devices->seeding) {
2246 btrfs_set_device_generation(leaf, dev_item,
2247 device->generation);
2248 btrfs_mark_buffer_dirty(leaf);
2256 btrfs_free_path(path);
2260 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
2262 struct request_queue *q;
2263 struct btrfs_trans_handle *trans;
2264 struct btrfs_device *device;
2265 struct block_device *bdev;
2266 struct list_head *devices;
2267 struct super_block *sb = root->fs_info->sb;
2268 struct rcu_string *name;
2270 int seeding_dev = 0;
2273 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
2276 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2277 root->fs_info->bdev_holder);
2279 return PTR_ERR(bdev);
2281 if (root->fs_info->fs_devices->seeding) {
2283 down_write(&sb->s_umount);
2284 mutex_lock(&uuid_mutex);
2287 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2289 devices = &root->fs_info->fs_devices->devices;
2291 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2292 list_for_each_entry(device, devices, dev_list) {
2293 if (device->bdev == bdev) {
2296 &root->fs_info->fs_devices->device_list_mutex);
2300 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2302 device = btrfs_alloc_device(root->fs_info, NULL, NULL);
2303 if (IS_ERR(device)) {
2304 /* we can safely leave the fs_devices entry around */
2305 ret = PTR_ERR(device);
2309 name = rcu_string_strdup(device_path, GFP_KERNEL);
2315 rcu_assign_pointer(device->name, name);
2317 trans = btrfs_start_transaction(root, 0);
2318 if (IS_ERR(trans)) {
2319 rcu_string_free(device->name);
2321 ret = PTR_ERR(trans);
2325 q = bdev_get_queue(bdev);
2326 if (blk_queue_discard(q))
2327 device->can_discard = 1;
2328 device->writeable = 1;
2329 device->generation = trans->transid;
2330 device->io_width = root->sectorsize;
2331 device->io_align = root->sectorsize;
2332 device->sector_size = root->sectorsize;
2333 device->total_bytes = i_size_read(bdev->bd_inode);
2334 device->disk_total_bytes = device->total_bytes;
2335 device->commit_total_bytes = device->total_bytes;
2336 device->dev_root = root->fs_info->dev_root;
2337 device->bdev = bdev;
2338 device->in_fs_metadata = 1;
2339 device->is_tgtdev_for_dev_replace = 0;
2340 device->mode = FMODE_EXCL;
2341 device->dev_stats_valid = 1;
2342 set_blocksize(device->bdev, 4096);
2345 sb->s_flags &= ~MS_RDONLY;
2346 ret = btrfs_prepare_sprout(root);
2347 BUG_ON(ret); /* -ENOMEM */
2350 device->fs_devices = root->fs_info->fs_devices;
2352 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2354 list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
2355 list_add(&device->dev_alloc_list,
2356 &root->fs_info->fs_devices->alloc_list);
2357 root->fs_info->fs_devices->num_devices++;
2358 root->fs_info->fs_devices->open_devices++;
2359 root->fs_info->fs_devices->rw_devices++;
2360 root->fs_info->fs_devices->total_devices++;
2361 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2363 spin_lock(&root->fs_info->free_chunk_lock);
2364 root->fs_info->free_chunk_space += device->total_bytes;
2365 spin_unlock(&root->fs_info->free_chunk_lock);
2367 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2368 root->fs_info->fs_devices->rotating = 1;
2370 tmp = btrfs_super_total_bytes(root->fs_info->super_copy);
2371 btrfs_set_super_total_bytes(root->fs_info->super_copy,
2372 tmp + device->total_bytes);
2374 tmp = btrfs_super_num_devices(root->fs_info->super_copy);
2375 btrfs_set_super_num_devices(root->fs_info->super_copy,
2378 /* add sysfs device entry */
2379 btrfs_sysfs_add_device_link(root->fs_info->fs_devices, device);
2382 * we've got more storage, clear any full flags on the space
2385 btrfs_clear_space_info_full(root->fs_info);
2387 unlock_chunks(root);
2388 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2392 ret = init_first_rw_device(trans, root, device);
2393 unlock_chunks(root);
2395 btrfs_abort_transaction(trans, root, ret);
2400 ret = btrfs_add_device(trans, root, device);
2402 btrfs_abort_transaction(trans, root, ret);
2407 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2409 ret = btrfs_finish_sprout(trans, root);
2411 btrfs_abort_transaction(trans, root, ret);
2415 /* Sprouting would change fsid of the mounted root,
2416 * so rename the fsid on the sysfs
2418 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2419 root->fs_info->fsid);
2420 if (kobject_rename(&root->fs_info->fs_devices->fsid_kobj,
2422 btrfs_warn(root->fs_info,
2423 "sysfs: failed to create fsid for sprout");
2426 root->fs_info->num_tolerated_disk_barrier_failures =
2427 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
2428 ret = btrfs_commit_transaction(trans, root);
2431 mutex_unlock(&uuid_mutex);
2432 up_write(&sb->s_umount);
2434 if (ret) /* transaction commit */
2437 ret = btrfs_relocate_sys_chunks(root);
2439 btrfs_std_error(root->fs_info, ret,
2440 "Failed to relocate sys chunks after "
2441 "device initialization. This can be fixed "
2442 "using the \"btrfs balance\" command.");
2443 trans = btrfs_attach_transaction(root);
2444 if (IS_ERR(trans)) {
2445 if (PTR_ERR(trans) == -ENOENT)
2447 return PTR_ERR(trans);
2449 ret = btrfs_commit_transaction(trans, root);
2452 /* Update ctime/mtime for libblkid */
2453 update_dev_time(device_path);
2457 btrfs_end_transaction(trans, root);
2458 rcu_string_free(device->name);
2459 btrfs_sysfs_rm_device_link(root->fs_info->fs_devices, device);
2462 blkdev_put(bdev, FMODE_EXCL);
2464 mutex_unlock(&uuid_mutex);
2465 up_write(&sb->s_umount);
2470 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2471 struct btrfs_device *srcdev,
2472 struct btrfs_device **device_out)
2474 struct request_queue *q;
2475 struct btrfs_device *device;
2476 struct block_device *bdev;
2477 struct btrfs_fs_info *fs_info = root->fs_info;
2478 struct list_head *devices;
2479 struct rcu_string *name;
2480 u64 devid = BTRFS_DEV_REPLACE_DEVID;
2484 if (fs_info->fs_devices->seeding) {
2485 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
2489 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2490 fs_info->bdev_holder);
2492 btrfs_err(fs_info, "target device %s is invalid!", device_path);
2493 return PTR_ERR(bdev);
2496 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2498 devices = &fs_info->fs_devices->devices;
2499 list_for_each_entry(device, devices, dev_list) {
2500 if (device->bdev == bdev) {
2501 btrfs_err(fs_info, "target device is in the filesystem!");
2508 if (i_size_read(bdev->bd_inode) <
2509 btrfs_device_get_total_bytes(srcdev)) {
2510 btrfs_err(fs_info, "target device is smaller than source device!");
2516 device = btrfs_alloc_device(NULL, &devid, NULL);
2517 if (IS_ERR(device)) {
2518 ret = PTR_ERR(device);
2522 name = rcu_string_strdup(device_path, GFP_NOFS);
2528 rcu_assign_pointer(device->name, name);
2530 q = bdev_get_queue(bdev);
2531 if (blk_queue_discard(q))
2532 device->can_discard = 1;
2533 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2534 device->writeable = 1;
2535 device->generation = 0;
2536 device->io_width = root->sectorsize;
2537 device->io_align = root->sectorsize;
2538 device->sector_size = root->sectorsize;
2539 device->total_bytes = btrfs_device_get_total_bytes(srcdev);
2540 device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
2541 device->bytes_used = btrfs_device_get_bytes_used(srcdev);
2542 ASSERT(list_empty(&srcdev->resized_list));
2543 device->commit_total_bytes = srcdev->commit_total_bytes;
2544 device->commit_bytes_used = device->bytes_used;
2545 device->dev_root = fs_info->dev_root;
2546 device->bdev = bdev;
2547 device->in_fs_metadata = 1;
2548 device->is_tgtdev_for_dev_replace = 1;
2549 device->mode = FMODE_EXCL;
2550 device->dev_stats_valid = 1;
2551 set_blocksize(device->bdev, 4096);
2552 device->fs_devices = fs_info->fs_devices;
2553 list_add(&device->dev_list, &fs_info->fs_devices->devices);
2554 fs_info->fs_devices->num_devices++;
2555 fs_info->fs_devices->open_devices++;
2556 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2558 *device_out = device;
2562 blkdev_put(bdev, FMODE_EXCL);
2566 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2567 struct btrfs_device *tgtdev)
2569 WARN_ON(fs_info->fs_devices->rw_devices == 0);
2570 tgtdev->io_width = fs_info->dev_root->sectorsize;
2571 tgtdev->io_align = fs_info->dev_root->sectorsize;
2572 tgtdev->sector_size = fs_info->dev_root->sectorsize;
2573 tgtdev->dev_root = fs_info->dev_root;
2574 tgtdev->in_fs_metadata = 1;
2577 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2578 struct btrfs_device *device)
2581 struct btrfs_path *path;
2582 struct btrfs_root *root;
2583 struct btrfs_dev_item *dev_item;
2584 struct extent_buffer *leaf;
2585 struct btrfs_key key;
2587 root = device->dev_root->fs_info->chunk_root;
2589 path = btrfs_alloc_path();
2593 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2594 key.type = BTRFS_DEV_ITEM_KEY;
2595 key.offset = device->devid;
2597 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2606 leaf = path->nodes[0];
2607 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2609 btrfs_set_device_id(leaf, dev_item, device->devid);
2610 btrfs_set_device_type(leaf, dev_item, device->type);
2611 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2612 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2613 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2614 btrfs_set_device_total_bytes(leaf, dev_item,
2615 btrfs_device_get_disk_total_bytes(device));
2616 btrfs_set_device_bytes_used(leaf, dev_item,
2617 btrfs_device_get_bytes_used(device));
2618 btrfs_mark_buffer_dirty(leaf);
2621 btrfs_free_path(path);
2625 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2626 struct btrfs_device *device, u64 new_size)
2628 struct btrfs_super_block *super_copy =
2629 device->dev_root->fs_info->super_copy;
2630 struct btrfs_fs_devices *fs_devices;
2634 if (!device->writeable)
2637 lock_chunks(device->dev_root);
2638 old_total = btrfs_super_total_bytes(super_copy);
2639 diff = new_size - device->total_bytes;
2641 if (new_size <= device->total_bytes ||
2642 device->is_tgtdev_for_dev_replace) {
2643 unlock_chunks(device->dev_root);
2647 fs_devices = device->dev_root->fs_info->fs_devices;
2649 btrfs_set_super_total_bytes(super_copy, old_total + diff);
2650 device->fs_devices->total_rw_bytes += diff;
2652 btrfs_device_set_total_bytes(device, new_size);
2653 btrfs_device_set_disk_total_bytes(device, new_size);
2654 btrfs_clear_space_info_full(device->dev_root->fs_info);
2655 if (list_empty(&device->resized_list))
2656 list_add_tail(&device->resized_list,
2657 &fs_devices->resized_devices);
2658 unlock_chunks(device->dev_root);
2660 return btrfs_update_device(trans, device);
2663 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2664 struct btrfs_root *root, u64 chunk_objectid,
2668 struct btrfs_path *path;
2669 struct btrfs_key key;
2671 root = root->fs_info->chunk_root;
2672 path = btrfs_alloc_path();
2676 key.objectid = chunk_objectid;
2677 key.offset = chunk_offset;
2678 key.type = BTRFS_CHUNK_ITEM_KEY;
2680 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2683 else if (ret > 0) { /* Logic error or corruption */
2684 btrfs_std_error(root->fs_info, -ENOENT,
2685 "Failed lookup while freeing chunk.");
2690 ret = btrfs_del_item(trans, root, path);
2692 btrfs_std_error(root->fs_info, ret,
2693 "Failed to delete chunk item.");
2695 btrfs_free_path(path);
2699 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2702 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2703 struct btrfs_disk_key *disk_key;
2704 struct btrfs_chunk *chunk;
2711 struct btrfs_key key;
2714 array_size = btrfs_super_sys_array_size(super_copy);
2716 ptr = super_copy->sys_chunk_array;
2719 while (cur < array_size) {
2720 disk_key = (struct btrfs_disk_key *)ptr;
2721 btrfs_disk_key_to_cpu(&key, disk_key);
2723 len = sizeof(*disk_key);
2725 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2726 chunk = (struct btrfs_chunk *)(ptr + len);
2727 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2728 len += btrfs_chunk_item_size(num_stripes);
2733 if (key.objectid == chunk_objectid &&
2734 key.offset == chunk_offset) {
2735 memmove(ptr, ptr + len, array_size - (cur + len));
2737 btrfs_set_super_sys_array_size(super_copy, array_size);
2743 unlock_chunks(root);
2747 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2748 struct btrfs_root *root, u64 chunk_offset)
2750 struct extent_map_tree *em_tree;
2751 struct extent_map *em;
2752 struct btrfs_root *extent_root = root->fs_info->extent_root;
2753 struct map_lookup *map;
2754 u64 dev_extent_len = 0;
2755 u64 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2759 root = root->fs_info->chunk_root;
2760 em_tree = &root->fs_info->mapping_tree.map_tree;
2762 read_lock(&em_tree->lock);
2763 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2764 read_unlock(&em_tree->lock);
2766 if (!em || em->start > chunk_offset ||
2767 em->start + em->len < chunk_offset) {
2769 * This is a logic error, but we don't want to just rely on the
2770 * user having built with ASSERT enabled, so if ASSERT doesn't
2771 * do anything we still error out.
2775 free_extent_map(em);
2778 map = em->map_lookup;
2779 lock_chunks(root->fs_info->chunk_root);
2780 check_system_chunk(trans, extent_root, map->type);
2781 unlock_chunks(root->fs_info->chunk_root);
2783 for (i = 0; i < map->num_stripes; i++) {
2784 struct btrfs_device *device = map->stripes[i].dev;
2785 ret = btrfs_free_dev_extent(trans, device,
2786 map->stripes[i].physical,
2789 btrfs_abort_transaction(trans, root, ret);
2793 if (device->bytes_used > 0) {
2795 btrfs_device_set_bytes_used(device,
2796 device->bytes_used - dev_extent_len);
2797 spin_lock(&root->fs_info->free_chunk_lock);
2798 root->fs_info->free_chunk_space += dev_extent_len;
2799 spin_unlock(&root->fs_info->free_chunk_lock);
2800 btrfs_clear_space_info_full(root->fs_info);
2801 unlock_chunks(root);
2804 if (map->stripes[i].dev) {
2805 ret = btrfs_update_device(trans, map->stripes[i].dev);
2807 btrfs_abort_transaction(trans, root, ret);
2812 ret = btrfs_free_chunk(trans, root, chunk_objectid, chunk_offset);
2814 btrfs_abort_transaction(trans, root, ret);
2818 trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2820 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2821 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2823 btrfs_abort_transaction(trans, root, ret);
2828 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset, em);
2830 btrfs_abort_transaction(trans, extent_root, ret);
2836 free_extent_map(em);
2840 static int btrfs_relocate_chunk(struct btrfs_root *root, u64 chunk_offset)
2842 struct btrfs_root *extent_root;
2843 struct btrfs_trans_handle *trans;
2846 root = root->fs_info->chunk_root;
2847 extent_root = root->fs_info->extent_root;
2850 * Prevent races with automatic removal of unused block groups.
2851 * After we relocate and before we remove the chunk with offset
2852 * chunk_offset, automatic removal of the block group can kick in,
2853 * resulting in a failure when calling btrfs_remove_chunk() below.
2855 * Make sure to acquire this mutex before doing a tree search (dev
2856 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2857 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2858 * we release the path used to search the chunk/dev tree and before
2859 * the current task acquires this mutex and calls us.
2861 ASSERT(mutex_is_locked(&root->fs_info->delete_unused_bgs_mutex));
2863 ret = btrfs_can_relocate(extent_root, chunk_offset);
2867 /* step one, relocate all the extents inside this chunk */
2868 btrfs_scrub_pause(root);
2869 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2870 btrfs_scrub_continue(root);
2874 trans = btrfs_start_trans_remove_block_group(root->fs_info,
2876 if (IS_ERR(trans)) {
2877 ret = PTR_ERR(trans);
2878 btrfs_std_error(root->fs_info, ret, NULL);
2883 * step two, delete the device extents and the
2884 * chunk tree entries
2886 ret = btrfs_remove_chunk(trans, root, chunk_offset);
2887 btrfs_end_transaction(trans, root);
2891 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2893 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2894 struct btrfs_path *path;
2895 struct extent_buffer *leaf;
2896 struct btrfs_chunk *chunk;
2897 struct btrfs_key key;
2898 struct btrfs_key found_key;
2900 bool retried = false;
2904 path = btrfs_alloc_path();
2909 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2910 key.offset = (u64)-1;
2911 key.type = BTRFS_CHUNK_ITEM_KEY;
2914 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
2915 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2917 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2920 BUG_ON(ret == 0); /* Corruption */
2922 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2925 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2931 leaf = path->nodes[0];
2932 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2934 chunk = btrfs_item_ptr(leaf, path->slots[0],
2935 struct btrfs_chunk);
2936 chunk_type = btrfs_chunk_type(leaf, chunk);
2937 btrfs_release_path(path);
2939 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2940 ret = btrfs_relocate_chunk(chunk_root,
2947 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2949 if (found_key.offset == 0)
2951 key.offset = found_key.offset - 1;
2954 if (failed && !retried) {
2958 } else if (WARN_ON(failed && retried)) {
2962 btrfs_free_path(path);
2966 static int insert_balance_item(struct btrfs_root *root,
2967 struct btrfs_balance_control *bctl)
2969 struct btrfs_trans_handle *trans;
2970 struct btrfs_balance_item *item;
2971 struct btrfs_disk_balance_args disk_bargs;
2972 struct btrfs_path *path;
2973 struct extent_buffer *leaf;
2974 struct btrfs_key key;
2977 path = btrfs_alloc_path();
2981 trans = btrfs_start_transaction(root, 0);
2982 if (IS_ERR(trans)) {
2983 btrfs_free_path(path);
2984 return PTR_ERR(trans);
2987 key.objectid = BTRFS_BALANCE_OBJECTID;
2988 key.type = BTRFS_TEMPORARY_ITEM_KEY;
2991 ret = btrfs_insert_empty_item(trans, root, path, &key,
2996 leaf = path->nodes[0];
2997 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2999 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
3001 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3002 btrfs_set_balance_data(leaf, item, &disk_bargs);
3003 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3004 btrfs_set_balance_meta(leaf, item, &disk_bargs);
3005 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3006 btrfs_set_balance_sys(leaf, item, &disk_bargs);
3008 btrfs_set_balance_flags(leaf, item, bctl->flags);
3010 btrfs_mark_buffer_dirty(leaf);
3012 btrfs_free_path(path);
3013 err = btrfs_commit_transaction(trans, root);
3019 static int del_balance_item(struct btrfs_root *root)
3021 struct btrfs_trans_handle *trans;
3022 struct btrfs_path *path;
3023 struct btrfs_key key;
3026 path = btrfs_alloc_path();
3030 trans = btrfs_start_transaction(root, 0);
3031 if (IS_ERR(trans)) {
3032 btrfs_free_path(path);
3033 return PTR_ERR(trans);
3036 key.objectid = BTRFS_BALANCE_OBJECTID;
3037 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3040 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3048 ret = btrfs_del_item(trans, root, path);
3050 btrfs_free_path(path);
3051 err = btrfs_commit_transaction(trans, root);
3058 * This is a heuristic used to reduce the number of chunks balanced on
3059 * resume after balance was interrupted.
3061 static void update_balance_args(struct btrfs_balance_control *bctl)
3064 * Turn on soft mode for chunk types that were being converted.
3066 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3067 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3068 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3069 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3070 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3071 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3074 * Turn on usage filter if is not already used. The idea is
3075 * that chunks that we have already balanced should be
3076 * reasonably full. Don't do it for chunks that are being
3077 * converted - that will keep us from relocating unconverted
3078 * (albeit full) chunks.
3080 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3081 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3082 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3083 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3084 bctl->data.usage = 90;
3086 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3087 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3088 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3089 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3090 bctl->sys.usage = 90;
3092 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3093 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3094 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3095 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3096 bctl->meta.usage = 90;
3101 * Should be called with both balance and volume mutexes held to
3102 * serialize other volume operations (add_dev/rm_dev/resize) with
3103 * restriper. Same goes for unset_balance_control.
3105 static void set_balance_control(struct btrfs_balance_control *bctl)
3107 struct btrfs_fs_info *fs_info = bctl->fs_info;
3109 BUG_ON(fs_info->balance_ctl);
3111 spin_lock(&fs_info->balance_lock);
3112 fs_info->balance_ctl = bctl;
3113 spin_unlock(&fs_info->balance_lock);
3116 static void unset_balance_control(struct btrfs_fs_info *fs_info)
3118 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3120 BUG_ON(!fs_info->balance_ctl);
3122 spin_lock(&fs_info->balance_lock);
3123 fs_info->balance_ctl = NULL;
3124 spin_unlock(&fs_info->balance_lock);
3130 * Balance filters. Return 1 if chunk should be filtered out
3131 * (should not be balanced).
3133 static int chunk_profiles_filter(u64 chunk_type,
3134 struct btrfs_balance_args *bargs)
3136 chunk_type = chunk_to_extended(chunk_type) &
3137 BTRFS_EXTENDED_PROFILE_MASK;
3139 if (bargs->profiles & chunk_type)
3145 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3146 struct btrfs_balance_args *bargs)
3148 struct btrfs_block_group_cache *cache;
3150 u64 user_thresh_min;
3151 u64 user_thresh_max;
3154 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3155 chunk_used = btrfs_block_group_used(&cache->item);
3157 if (bargs->usage_min == 0)
3158 user_thresh_min = 0;
3160 user_thresh_min = div_factor_fine(cache->key.offset,
3163 if (bargs->usage_max == 0)
3164 user_thresh_max = 1;
3165 else if (bargs->usage_max > 100)
3166 user_thresh_max = cache->key.offset;
3168 user_thresh_max = div_factor_fine(cache->key.offset,
3171 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3174 btrfs_put_block_group(cache);
3178 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3179 u64 chunk_offset, struct btrfs_balance_args *bargs)
3181 struct btrfs_block_group_cache *cache;
3182 u64 chunk_used, user_thresh;
3185 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3186 chunk_used = btrfs_block_group_used(&cache->item);
3188 if (bargs->usage_min == 0)
3190 else if (bargs->usage > 100)
3191 user_thresh = cache->key.offset;
3193 user_thresh = div_factor_fine(cache->key.offset,
3196 if (chunk_used < user_thresh)
3199 btrfs_put_block_group(cache);
3203 static int chunk_devid_filter(struct extent_buffer *leaf,
3204 struct btrfs_chunk *chunk,
3205 struct btrfs_balance_args *bargs)
3207 struct btrfs_stripe *stripe;
3208 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3211 for (i = 0; i < num_stripes; i++) {
3212 stripe = btrfs_stripe_nr(chunk, i);
3213 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3220 /* [pstart, pend) */
3221 static int chunk_drange_filter(struct extent_buffer *leaf,
3222 struct btrfs_chunk *chunk,
3224 struct btrfs_balance_args *bargs)
3226 struct btrfs_stripe *stripe;
3227 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3233 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3236 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3237 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3238 factor = num_stripes / 2;
3239 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3240 factor = num_stripes - 1;
3241 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3242 factor = num_stripes - 2;
3244 factor = num_stripes;
3247 for (i = 0; i < num_stripes; i++) {
3248 stripe = btrfs_stripe_nr(chunk, i);
3249 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3252 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3253 stripe_length = btrfs_chunk_length(leaf, chunk);
3254 stripe_length = div_u64(stripe_length, factor);
3256 if (stripe_offset < bargs->pend &&
3257 stripe_offset + stripe_length > bargs->pstart)
3264 /* [vstart, vend) */
3265 static int chunk_vrange_filter(struct extent_buffer *leaf,
3266 struct btrfs_chunk *chunk,
3268 struct btrfs_balance_args *bargs)
3270 if (chunk_offset < bargs->vend &&
3271 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3272 /* at least part of the chunk is inside this vrange */
3278 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3279 struct btrfs_chunk *chunk,
3280 struct btrfs_balance_args *bargs)
3282 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3284 if (bargs->stripes_min <= num_stripes
3285 && num_stripes <= bargs->stripes_max)
3291 static int chunk_soft_convert_filter(u64 chunk_type,
3292 struct btrfs_balance_args *bargs)
3294 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3297 chunk_type = chunk_to_extended(chunk_type) &
3298 BTRFS_EXTENDED_PROFILE_MASK;
3300 if (bargs->target == chunk_type)
3306 static int should_balance_chunk(struct btrfs_root *root,
3307 struct extent_buffer *leaf,
3308 struct btrfs_chunk *chunk, u64 chunk_offset)
3310 struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
3311 struct btrfs_balance_args *bargs = NULL;
3312 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3315 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3316 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3320 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3321 bargs = &bctl->data;
3322 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3324 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3325 bargs = &bctl->meta;
3327 /* profiles filter */
3328 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3329 chunk_profiles_filter(chunk_type, bargs)) {
3334 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3335 chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
3337 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3338 chunk_usage_range_filter(bctl->fs_info, chunk_offset, bargs)) {
3343 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3344 chunk_devid_filter(leaf, chunk, bargs)) {
3348 /* drange filter, makes sense only with devid filter */
3349 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3350 chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
3355 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3356 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3360 /* stripes filter */
3361 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3362 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3366 /* soft profile changing mode */
3367 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3368 chunk_soft_convert_filter(chunk_type, bargs)) {
3373 * limited by count, must be the last filter
3375 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3376 if (bargs->limit == 0)
3380 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3382 * Same logic as the 'limit' filter; the minimum cannot be
3383 * determined here because we do not have the global informatoin
3384 * about the count of all chunks that satisfy the filters.
3386 if (bargs->limit_max == 0)
3395 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3397 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3398 struct btrfs_root *chunk_root = fs_info->chunk_root;
3399 struct btrfs_root *dev_root = fs_info->dev_root;
3400 struct list_head *devices;
3401 struct btrfs_device *device;
3405 struct btrfs_chunk *chunk;
3406 struct btrfs_path *path;
3407 struct btrfs_key key;
3408 struct btrfs_key found_key;
3409 struct btrfs_trans_handle *trans;
3410 struct extent_buffer *leaf;
3413 int enospc_errors = 0;
3414 bool counting = true;
3415 /* The single value limit and min/max limits use the same bytes in the */
3416 u64 limit_data = bctl->data.limit;
3417 u64 limit_meta = bctl->meta.limit;
3418 u64 limit_sys = bctl->sys.limit;
3422 int chunk_reserved = 0;
3424 /* step one make some room on all the devices */
3425 devices = &fs_info->fs_devices->devices;
3426 list_for_each_entry(device, devices, dev_list) {
3427 old_size = btrfs_device_get_total_bytes(device);
3428 size_to_free = div_factor(old_size, 1);
3429 size_to_free = min_t(u64, size_to_free, SZ_1M);
3430 if (!device->writeable ||
3431 btrfs_device_get_total_bytes(device) -
3432 btrfs_device_get_bytes_used(device) > size_to_free ||
3433 device->is_tgtdev_for_dev_replace)
3436 ret = btrfs_shrink_device(device, old_size - size_to_free);
3441 trans = btrfs_start_transaction(dev_root, 0);
3442 BUG_ON(IS_ERR(trans));
3444 ret = btrfs_grow_device(trans, device, old_size);
3447 btrfs_end_transaction(trans, dev_root);
3450 /* step two, relocate all the chunks */
3451 path = btrfs_alloc_path();
3457 /* zero out stat counters */
3458 spin_lock(&fs_info->balance_lock);
3459 memset(&bctl->stat, 0, sizeof(bctl->stat));
3460 spin_unlock(&fs_info->balance_lock);
3464 * The single value limit and min/max limits use the same bytes
3467 bctl->data.limit = limit_data;
3468 bctl->meta.limit = limit_meta;
3469 bctl->sys.limit = limit_sys;
3471 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3472 key.offset = (u64)-1;
3473 key.type = BTRFS_CHUNK_ITEM_KEY;
3476 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3477 atomic_read(&fs_info->balance_cancel_req)) {
3482 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3483 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3485 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3490 * this shouldn't happen, it means the last relocate
3494 BUG(); /* FIXME break ? */
3496 ret = btrfs_previous_item(chunk_root, path, 0,
3497 BTRFS_CHUNK_ITEM_KEY);
3499 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3504 leaf = path->nodes[0];
3505 slot = path->slots[0];
3506 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3508 if (found_key.objectid != key.objectid) {
3509 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3513 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3514 chunk_type = btrfs_chunk_type(leaf, chunk);
3517 spin_lock(&fs_info->balance_lock);
3518 bctl->stat.considered++;
3519 spin_unlock(&fs_info->balance_lock);
3522 ret = should_balance_chunk(chunk_root, leaf, chunk,
3525 btrfs_release_path(path);
3527 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3532 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3533 spin_lock(&fs_info->balance_lock);
3534 bctl->stat.expected++;
3535 spin_unlock(&fs_info->balance_lock);
3537 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3539 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3541 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3548 * Apply limit_min filter, no need to check if the LIMITS
3549 * filter is used, limit_min is 0 by default
3551 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3552 count_data < bctl->data.limit_min)
3553 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3554 count_meta < bctl->meta.limit_min)
3555 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3556 count_sys < bctl->sys.limit_min)) {
3557 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3561 if ((chunk_type & BTRFS_BLOCK_GROUP_DATA) && !chunk_reserved) {
3562 trans = btrfs_start_transaction(chunk_root, 0);
3563 if (IS_ERR(trans)) {
3564 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3565 ret = PTR_ERR(trans);
3569 ret = btrfs_force_chunk_alloc(trans, chunk_root,
3570 BTRFS_BLOCK_GROUP_DATA);
3571 btrfs_end_transaction(trans, chunk_root);
3573 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3579 ret = btrfs_relocate_chunk(chunk_root,
3581 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3582 if (ret && ret != -ENOSPC)
3584 if (ret == -ENOSPC) {
3587 spin_lock(&fs_info->balance_lock);
3588 bctl->stat.completed++;
3589 spin_unlock(&fs_info->balance_lock);
3592 if (found_key.offset == 0)
3594 key.offset = found_key.offset - 1;
3598 btrfs_release_path(path);
3603 btrfs_free_path(path);
3604 if (enospc_errors) {
3605 btrfs_info(fs_info, "%d enospc errors during balance",
3615 * alloc_profile_is_valid - see if a given profile is valid and reduced
3616 * @flags: profile to validate
3617 * @extended: if true @flags is treated as an extended profile
3619 static int alloc_profile_is_valid(u64 flags, int extended)
3621 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3622 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3624 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3626 /* 1) check that all other bits are zeroed */
3630 /* 2) see if profile is reduced */
3632 return !extended; /* "0" is valid for usual profiles */
3634 /* true if exactly one bit set */
3635 return (flags & (flags - 1)) == 0;
3638 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3640 /* cancel requested || normal exit path */
3641 return atomic_read(&fs_info->balance_cancel_req) ||
3642 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3643 atomic_read(&fs_info->balance_cancel_req) == 0);
3646 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3650 unset_balance_control(fs_info);
3651 ret = del_balance_item(fs_info->tree_root);
3653 btrfs_std_error(fs_info, ret, NULL);
3655 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3658 /* Non-zero return value signifies invalidity */
3659 static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg,
3662 return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3663 (!alloc_profile_is_valid(bctl_arg->target, 1) ||
3664 (bctl_arg->target & ~allowed)));
3668 * Should be called with both balance and volume mutexes held
3670 int btrfs_balance(struct btrfs_balance_control *bctl,
3671 struct btrfs_ioctl_balance_args *bargs)
3673 struct btrfs_fs_info *fs_info = bctl->fs_info;
3680 if (btrfs_fs_closing(fs_info) ||
3681 atomic_read(&fs_info->balance_pause_req) ||
3682 atomic_read(&fs_info->balance_cancel_req)) {
3687 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3688 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3692 * In case of mixed groups both data and meta should be picked,
3693 * and identical options should be given for both of them.
3695 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3696 if (mixed && (bctl->flags & allowed)) {
3697 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3698 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3699 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3700 btrfs_err(fs_info, "with mixed groups data and "
3701 "metadata balance options must be the same");
3707 num_devices = fs_info->fs_devices->num_devices;
3708 btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
3709 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3710 BUG_ON(num_devices < 1);
3713 btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
3714 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3715 if (num_devices == 1)
3716 allowed |= BTRFS_BLOCK_GROUP_DUP;
3717 else if (num_devices > 1)
3718 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3719 if (num_devices > 2)
3720 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3721 if (num_devices > 3)
3722 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3723 BTRFS_BLOCK_GROUP_RAID6);
3724 if (validate_convert_profile(&bctl->data, allowed)) {
3725 btrfs_err(fs_info, "unable to start balance with target "
3726 "data profile %llu",
3731 if (validate_convert_profile(&bctl->meta, allowed)) {
3733 "unable to start balance with target metadata profile %llu",
3738 if (validate_convert_profile(&bctl->sys, allowed)) {
3740 "unable to start balance with target system profile %llu",
3746 /* allow to reduce meta or sys integrity only if force set */
3747 allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3748 BTRFS_BLOCK_GROUP_RAID10 |
3749 BTRFS_BLOCK_GROUP_RAID5 |
3750 BTRFS_BLOCK_GROUP_RAID6;
3752 seq = read_seqbegin(&fs_info->profiles_lock);
3754 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3755 (fs_info->avail_system_alloc_bits & allowed) &&
3756 !(bctl->sys.target & allowed)) ||
3757 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3758 (fs_info->avail_metadata_alloc_bits & allowed) &&
3759 !(bctl->meta.target & allowed))) {
3760 if (bctl->flags & BTRFS_BALANCE_FORCE) {
3761 btrfs_info(fs_info, "force reducing metadata integrity");
3763 btrfs_err(fs_info, "balance will reduce metadata "
3764 "integrity, use force if you want this");
3769 } while (read_seqretry(&fs_info->profiles_lock, seq));
3771 if (btrfs_get_num_tolerated_disk_barrier_failures(bctl->meta.target) <
3772 btrfs_get_num_tolerated_disk_barrier_failures(bctl->data.target)) {
3774 "metadata profile 0x%llx has lower redundancy than data profile 0x%llx",
3775 bctl->meta.target, bctl->data.target);
3778 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3779 fs_info->num_tolerated_disk_barrier_failures = min(
3780 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info),
3781 btrfs_get_num_tolerated_disk_barrier_failures(
3785 ret = insert_balance_item(fs_info->tree_root, bctl);
3786 if (ret && ret != -EEXIST)
3789 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3790 BUG_ON(ret == -EEXIST);
3791 set_balance_control(bctl);
3793 BUG_ON(ret != -EEXIST);
3794 spin_lock(&fs_info->balance_lock);
3795 update_balance_args(bctl);
3796 spin_unlock(&fs_info->balance_lock);
3799 atomic_inc(&fs_info->balance_running);
3800 mutex_unlock(&fs_info->balance_mutex);
3802 ret = __btrfs_balance(fs_info);
3804 mutex_lock(&fs_info->balance_mutex);
3805 atomic_dec(&fs_info->balance_running);
3807 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3808 fs_info->num_tolerated_disk_barrier_failures =
3809 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3813 memset(bargs, 0, sizeof(*bargs));
3814 update_ioctl_balance_args(fs_info, 0, bargs);
3817 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3818 balance_need_close(fs_info)) {
3819 __cancel_balance(fs_info);
3822 wake_up(&fs_info->balance_wait_q);
3826 if (bctl->flags & BTRFS_BALANCE_RESUME)
3827 __cancel_balance(fs_info);
3830 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3835 static int balance_kthread(void *data)
3837 struct btrfs_fs_info *fs_info = data;
3840 mutex_lock(&fs_info->volume_mutex);
3841 mutex_lock(&fs_info->balance_mutex);
3843 if (fs_info->balance_ctl) {
3844 btrfs_info(fs_info, "continuing balance");
3845 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3848 mutex_unlock(&fs_info->balance_mutex);
3849 mutex_unlock(&fs_info->volume_mutex);
3854 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3856 struct task_struct *tsk;
3858 spin_lock(&fs_info->balance_lock);
3859 if (!fs_info->balance_ctl) {
3860 spin_unlock(&fs_info->balance_lock);
3863 spin_unlock(&fs_info->balance_lock);
3865 if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
3866 btrfs_info(fs_info, "force skipping balance");
3870 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3871 return PTR_ERR_OR_ZERO(tsk);
3874 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3876 struct btrfs_balance_control *bctl;
3877 struct btrfs_balance_item *item;
3878 struct btrfs_disk_balance_args disk_bargs;
3879 struct btrfs_path *path;
3880 struct extent_buffer *leaf;
3881 struct btrfs_key key;
3884 path = btrfs_alloc_path();
3888 key.objectid = BTRFS_BALANCE_OBJECTID;
3889 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3892 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3895 if (ret > 0) { /* ret = -ENOENT; */
3900 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3906 leaf = path->nodes[0];
3907 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3909 bctl->fs_info = fs_info;
3910 bctl->flags = btrfs_balance_flags(leaf, item);
3911 bctl->flags |= BTRFS_BALANCE_RESUME;
3913 btrfs_balance_data(leaf, item, &disk_bargs);
3914 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3915 btrfs_balance_meta(leaf, item, &disk_bargs);
3916 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3917 btrfs_balance_sys(leaf, item, &disk_bargs);
3918 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3920 WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
3922 mutex_lock(&fs_info->volume_mutex);
3923 mutex_lock(&fs_info->balance_mutex);
3925 set_balance_control(bctl);
3927 mutex_unlock(&fs_info->balance_mutex);
3928 mutex_unlock(&fs_info->volume_mutex);
3930 btrfs_free_path(path);
3934 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3938 mutex_lock(&fs_info->balance_mutex);
3939 if (!fs_info->balance_ctl) {
3940 mutex_unlock(&fs_info->balance_mutex);
3944 if (atomic_read(&fs_info->balance_running)) {
3945 atomic_inc(&fs_info->balance_pause_req);
3946 mutex_unlock(&fs_info->balance_mutex);
3948 wait_event(fs_info->balance_wait_q,
3949 atomic_read(&fs_info->balance_running) == 0);
3951 mutex_lock(&fs_info->balance_mutex);
3952 /* we are good with balance_ctl ripped off from under us */
3953 BUG_ON(atomic_read(&fs_info->balance_running));
3954 atomic_dec(&fs_info->balance_pause_req);
3959 mutex_unlock(&fs_info->balance_mutex);
3963 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3965 if (fs_info->sb->s_flags & MS_RDONLY)
3968 mutex_lock(&fs_info->balance_mutex);
3969 if (!fs_info->balance_ctl) {
3970 mutex_unlock(&fs_info->balance_mutex);
3974 atomic_inc(&fs_info->balance_cancel_req);
3976 * if we are running just wait and return, balance item is
3977 * deleted in btrfs_balance in this case
3979 if (atomic_read(&fs_info->balance_running)) {
3980 mutex_unlock(&fs_info->balance_mutex);
3981 wait_event(fs_info->balance_wait_q,
3982 atomic_read(&fs_info->balance_running) == 0);
3983 mutex_lock(&fs_info->balance_mutex);
3985 /* __cancel_balance needs volume_mutex */
3986 mutex_unlock(&fs_info->balance_mutex);
3987 mutex_lock(&fs_info->volume_mutex);
3988 mutex_lock(&fs_info->balance_mutex);
3990 if (fs_info->balance_ctl)
3991 __cancel_balance(fs_info);
3993 mutex_unlock(&fs_info->volume_mutex);
3996 BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
3997 atomic_dec(&fs_info->balance_cancel_req);
3998 mutex_unlock(&fs_info->balance_mutex);
4002 static int btrfs_uuid_scan_kthread(void *data)
4004 struct btrfs_fs_info *fs_info = data;
4005 struct btrfs_root *root = fs_info->tree_root;
4006 struct btrfs_key key;
4007 struct btrfs_key max_key;
4008 struct btrfs_path *path = NULL;
4010 struct extent_buffer *eb;
4012 struct btrfs_root_item root_item;
4014 struct btrfs_trans_handle *trans = NULL;
4016 path = btrfs_alloc_path();
4023 key.type = BTRFS_ROOT_ITEM_KEY;
4026 max_key.objectid = (u64)-1;
4027 max_key.type = BTRFS_ROOT_ITEM_KEY;
4028 max_key.offset = (u64)-1;
4031 ret = btrfs_search_forward(root, &key, path, 0);
4038 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4039 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4040 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4041 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4044 eb = path->nodes[0];
4045 slot = path->slots[0];
4046 item_size = btrfs_item_size_nr(eb, slot);
4047 if (item_size < sizeof(root_item))
4050 read_extent_buffer(eb, &root_item,
4051 btrfs_item_ptr_offset(eb, slot),
4052 (int)sizeof(root_item));
4053 if (btrfs_root_refs(&root_item) == 0)
4056 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4057 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4061 btrfs_release_path(path);
4063 * 1 - subvol uuid item
4064 * 1 - received_subvol uuid item
4066 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4067 if (IS_ERR(trans)) {
4068 ret = PTR_ERR(trans);
4076 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4077 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
4079 BTRFS_UUID_KEY_SUBVOL,
4082 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4088 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4089 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
4090 root_item.received_uuid,
4091 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4094 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4102 ret = btrfs_end_transaction(trans, fs_info->uuid_root);
4108 btrfs_release_path(path);
4109 if (key.offset < (u64)-1) {
4111 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4113 key.type = BTRFS_ROOT_ITEM_KEY;
4114 } else if (key.objectid < (u64)-1) {
4116 key.type = BTRFS_ROOT_ITEM_KEY;
4125 btrfs_free_path(path);
4126 if (trans && !IS_ERR(trans))
4127 btrfs_end_transaction(trans, fs_info->uuid_root);
4129 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4131 fs_info->update_uuid_tree_gen = 1;
4132 up(&fs_info->uuid_tree_rescan_sem);
4137 * Callback for btrfs_uuid_tree_iterate().
4139 * 0 check succeeded, the entry is not outdated.
4140 * < 0 if an error occurred.
4141 * > 0 if the check failed, which means the caller shall remove the entry.
4143 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
4144 u8 *uuid, u8 type, u64 subid)
4146 struct btrfs_key key;
4148 struct btrfs_root *subvol_root;
4150 if (type != BTRFS_UUID_KEY_SUBVOL &&
4151 type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
4154 key.objectid = subid;
4155 key.type = BTRFS_ROOT_ITEM_KEY;
4156 key.offset = (u64)-1;
4157 subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
4158 if (IS_ERR(subvol_root)) {
4159 ret = PTR_ERR(subvol_root);
4166 case BTRFS_UUID_KEY_SUBVOL:
4167 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
4170 case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
4171 if (memcmp(uuid, subvol_root->root_item.received_uuid,
4181 static int btrfs_uuid_rescan_kthread(void *data)
4183 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4187 * 1st step is to iterate through the existing UUID tree and
4188 * to delete all entries that contain outdated data.
4189 * 2nd step is to add all missing entries to the UUID tree.
4191 ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4193 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4194 up(&fs_info->uuid_tree_rescan_sem);
4197 return btrfs_uuid_scan_kthread(data);
4200 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4202 struct btrfs_trans_handle *trans;
4203 struct btrfs_root *tree_root = fs_info->tree_root;
4204 struct btrfs_root *uuid_root;
4205 struct task_struct *task;
4212 trans = btrfs_start_transaction(tree_root, 2);
4214 return PTR_ERR(trans);
4216 uuid_root = btrfs_create_tree(trans, fs_info,
4217 BTRFS_UUID_TREE_OBJECTID);
4218 if (IS_ERR(uuid_root)) {
4219 ret = PTR_ERR(uuid_root);
4220 btrfs_abort_transaction(trans, tree_root, ret);
4224 fs_info->uuid_root = uuid_root;
4226 ret = btrfs_commit_transaction(trans, tree_root);
4230 down(&fs_info->uuid_tree_rescan_sem);
4231 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4233 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4234 btrfs_warn(fs_info, "failed to start uuid_scan task");
4235 up(&fs_info->uuid_tree_rescan_sem);
4236 return PTR_ERR(task);
4242 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4244 struct task_struct *task;
4246 down(&fs_info->uuid_tree_rescan_sem);
4247 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4249 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4250 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4251 up(&fs_info->uuid_tree_rescan_sem);
4252 return PTR_ERR(task);
4259 * shrinking a device means finding all of the device extents past
4260 * the new size, and then following the back refs to the chunks.
4261 * The chunk relocation code actually frees the device extent
4263 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4265 struct btrfs_trans_handle *trans;
4266 struct btrfs_root *root = device->dev_root;
4267 struct btrfs_dev_extent *dev_extent = NULL;
4268 struct btrfs_path *path;
4274 bool retried = false;
4275 bool checked_pending_chunks = false;
4276 struct extent_buffer *l;
4277 struct btrfs_key key;
4278 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4279 u64 old_total = btrfs_super_total_bytes(super_copy);
4280 u64 old_size = btrfs_device_get_total_bytes(device);
4281 u64 diff = old_size - new_size;
4283 if (device->is_tgtdev_for_dev_replace)
4286 path = btrfs_alloc_path();
4290 path->reada = READA_FORWARD;
4294 btrfs_device_set_total_bytes(device, new_size);
4295 if (device->writeable) {
4296 device->fs_devices->total_rw_bytes -= diff;
4297 spin_lock(&root->fs_info->free_chunk_lock);
4298 root->fs_info->free_chunk_space -= diff;
4299 spin_unlock(&root->fs_info->free_chunk_lock);
4301 unlock_chunks(root);
4304 key.objectid = device->devid;
4305 key.offset = (u64)-1;
4306 key.type = BTRFS_DEV_EXTENT_KEY;
4309 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
4310 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4312 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4316 ret = btrfs_previous_item(root, path, 0, key.type);
4318 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4323 btrfs_release_path(path);
4328 slot = path->slots[0];
4329 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4331 if (key.objectid != device->devid) {
4332 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4333 btrfs_release_path(path);
4337 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4338 length = btrfs_dev_extent_length(l, dev_extent);
4340 if (key.offset + length <= new_size) {
4341 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4342 btrfs_release_path(path);
4346 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4347 btrfs_release_path(path);
4349 ret = btrfs_relocate_chunk(root, chunk_offset);
4350 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4351 if (ret && ret != -ENOSPC)
4355 } while (key.offset-- > 0);
4357 if (failed && !retried) {
4361 } else if (failed && retried) {
4366 /* Shrinking succeeded, else we would be at "done". */
4367 trans = btrfs_start_transaction(root, 0);
4368 if (IS_ERR(trans)) {
4369 ret = PTR_ERR(trans);
4376 * We checked in the above loop all device extents that were already in
4377 * the device tree. However before we have updated the device's
4378 * total_bytes to the new size, we might have had chunk allocations that
4379 * have not complete yet (new block groups attached to transaction
4380 * handles), and therefore their device extents were not yet in the
4381 * device tree and we missed them in the loop above. So if we have any
4382 * pending chunk using a device extent that overlaps the device range
4383 * that we can not use anymore, commit the current transaction and
4384 * repeat the search on the device tree - this way we guarantee we will
4385 * not have chunks using device extents that end beyond 'new_size'.
4387 if (!checked_pending_chunks) {
4388 u64 start = new_size;
4389 u64 len = old_size - new_size;
4391 if (contains_pending_extent(trans->transaction, device,
4393 unlock_chunks(root);
4394 checked_pending_chunks = true;
4397 ret = btrfs_commit_transaction(trans, root);
4404 btrfs_device_set_disk_total_bytes(device, new_size);
4405 if (list_empty(&device->resized_list))
4406 list_add_tail(&device->resized_list,
4407 &root->fs_info->fs_devices->resized_devices);
4409 WARN_ON(diff > old_total);
4410 btrfs_set_super_total_bytes(super_copy, old_total - diff);
4411 unlock_chunks(root);
4413 /* Now btrfs_update_device() will change the on-disk size. */
4414 ret = btrfs_update_device(trans, device);
4415 btrfs_end_transaction(trans, root);
4417 btrfs_free_path(path);
4420 btrfs_device_set_total_bytes(device, old_size);
4421 if (device->writeable)
4422 device->fs_devices->total_rw_bytes += diff;
4423 spin_lock(&root->fs_info->free_chunk_lock);
4424 root->fs_info->free_chunk_space += diff;
4425 spin_unlock(&root->fs_info->free_chunk_lock);
4426 unlock_chunks(root);
4431 static int btrfs_add_system_chunk(struct btrfs_root *root,
4432 struct btrfs_key *key,
4433 struct btrfs_chunk *chunk, int item_size)
4435 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4436 struct btrfs_disk_key disk_key;
4441 array_size = btrfs_super_sys_array_size(super_copy);
4442 if (array_size + item_size + sizeof(disk_key)
4443 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4444 unlock_chunks(root);
4448 ptr = super_copy->sys_chunk_array + array_size;
4449 btrfs_cpu_key_to_disk(&disk_key, key);
4450 memcpy(ptr, &disk_key, sizeof(disk_key));
4451 ptr += sizeof(disk_key);
4452 memcpy(ptr, chunk, item_size);
4453 item_size += sizeof(disk_key);
4454 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4455 unlock_chunks(root);
4461 * sort the devices in descending order by max_avail, total_avail
4463 static int btrfs_cmp_device_info(const void *a, const void *b)
4465 const struct btrfs_device_info *di_a = a;
4466 const struct btrfs_device_info *di_b = b;
4468 if (di_a->max_avail > di_b->max_avail)
4470 if (di_a->max_avail < di_b->max_avail)
4472 if (di_a->total_avail > di_b->total_avail)
4474 if (di_a->total_avail < di_b->total_avail)
4479 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
4481 /* TODO allow them to set a preferred stripe size */
4485 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4487 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4490 btrfs_set_fs_incompat(info, RAID56);
4493 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r) \
4494 - sizeof(struct btrfs_item) \
4495 - sizeof(struct btrfs_chunk)) \
4496 / sizeof(struct btrfs_stripe) + 1)
4498 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
4499 - 2 * sizeof(struct btrfs_disk_key) \
4500 - 2 * sizeof(struct btrfs_chunk)) \
4501 / sizeof(struct btrfs_stripe) + 1)
4503 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4504 struct btrfs_root *extent_root, u64 start,
4507 struct btrfs_fs_info *info = extent_root->fs_info;
4508 struct btrfs_fs_devices *fs_devices = info->fs_devices;
4509 struct list_head *cur;
4510 struct map_lookup *map = NULL;
4511 struct extent_map_tree *em_tree;
4512 struct extent_map *em;
4513 struct btrfs_device_info *devices_info = NULL;
4515 int num_stripes; /* total number of stripes to allocate */
4516 int data_stripes; /* number of stripes that count for
4518 int sub_stripes; /* sub_stripes info for map */
4519 int dev_stripes; /* stripes per dev */
4520 int devs_max; /* max devs to use */
4521 int devs_min; /* min devs needed */
4522 int devs_increment; /* ndevs has to be a multiple of this */
4523 int ncopies; /* how many copies to data has */
4525 u64 max_stripe_size;
4529 u64 raid_stripe_len = BTRFS_STRIPE_LEN;
4535 BUG_ON(!alloc_profile_is_valid(type, 0));
4537 if (list_empty(&fs_devices->alloc_list))
4540 index = __get_raid_index(type);
4542 sub_stripes = btrfs_raid_array[index].sub_stripes;
4543 dev_stripes = btrfs_raid_array[index].dev_stripes;
4544 devs_max = btrfs_raid_array[index].devs_max;
4545 devs_min = btrfs_raid_array[index].devs_min;
4546 devs_increment = btrfs_raid_array[index].devs_increment;
4547 ncopies = btrfs_raid_array[index].ncopies;
4549 if (type & BTRFS_BLOCK_GROUP_DATA) {
4550 max_stripe_size = SZ_1G;
4551 max_chunk_size = 10 * max_stripe_size;
4553 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4554 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4555 /* for larger filesystems, use larger metadata chunks */
4556 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
4557 max_stripe_size = SZ_1G;
4559 max_stripe_size = SZ_256M;
4560 max_chunk_size = max_stripe_size;
4562 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4563 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4564 max_stripe_size = SZ_32M;
4565 max_chunk_size = 2 * max_stripe_size;
4567 devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4569 btrfs_err(info, "invalid chunk type 0x%llx requested",
4574 /* we don't want a chunk larger than 10% of writeable space */
4575 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4578 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4583 cur = fs_devices->alloc_list.next;
4586 * in the first pass through the devices list, we gather information
4587 * about the available holes on each device.
4590 while (cur != &fs_devices->alloc_list) {
4591 struct btrfs_device *device;
4595 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4599 if (!device->writeable) {
4601 "BTRFS: read-only device in alloc_list\n");
4605 if (!device->in_fs_metadata ||
4606 device->is_tgtdev_for_dev_replace)
4609 if (device->total_bytes > device->bytes_used)
4610 total_avail = device->total_bytes - device->bytes_used;
4614 /* If there is no space on this device, skip it. */
4615 if (total_avail == 0)
4618 ret = find_free_dev_extent(trans, device,
4619 max_stripe_size * dev_stripes,
4620 &dev_offset, &max_avail);
4621 if (ret && ret != -ENOSPC)
4625 max_avail = max_stripe_size * dev_stripes;
4627 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4630 if (ndevs == fs_devices->rw_devices) {
4631 WARN(1, "%s: found more than %llu devices\n",
4632 __func__, fs_devices->rw_devices);
4635 devices_info[ndevs].dev_offset = dev_offset;
4636 devices_info[ndevs].max_avail = max_avail;
4637 devices_info[ndevs].total_avail = total_avail;
4638 devices_info[ndevs].dev = device;
4643 * now sort the devices by hole size / available space
4645 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4646 btrfs_cmp_device_info, NULL);
4648 /* round down to number of usable stripes */
4649 ndevs -= ndevs % devs_increment;
4651 if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4656 if (devs_max && ndevs > devs_max)
4659 * the primary goal is to maximize the number of stripes, so use as many
4660 * devices as possible, even if the stripes are not maximum sized.
4662 stripe_size = devices_info[ndevs-1].max_avail;
4663 num_stripes = ndevs * dev_stripes;
4666 * this will have to be fixed for RAID1 and RAID10 over
4669 data_stripes = num_stripes / ncopies;
4671 if (type & BTRFS_BLOCK_GROUP_RAID5) {
4672 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
4673 btrfs_super_stripesize(info->super_copy));
4674 data_stripes = num_stripes - 1;
4676 if (type & BTRFS_BLOCK_GROUP_RAID6) {
4677 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
4678 btrfs_super_stripesize(info->super_copy));
4679 data_stripes = num_stripes - 2;
4683 * Use the number of data stripes to figure out how big this chunk
4684 * is really going to be in terms of logical address space,
4685 * and compare that answer with the max chunk size
4687 if (stripe_size * data_stripes > max_chunk_size) {
4688 u64 mask = (1ULL << 24) - 1;
4690 stripe_size = div_u64(max_chunk_size, data_stripes);
4692 /* bump the answer up to a 16MB boundary */
4693 stripe_size = (stripe_size + mask) & ~mask;
4695 /* but don't go higher than the limits we found
4696 * while searching for free extents
4698 if (stripe_size > devices_info[ndevs-1].max_avail)
4699 stripe_size = devices_info[ndevs-1].max_avail;
4702 stripe_size = div_u64(stripe_size, dev_stripes);
4704 /* align to BTRFS_STRIPE_LEN */
4705 stripe_size = div_u64(stripe_size, raid_stripe_len);
4706 stripe_size *= raid_stripe_len;
4708 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4713 map->num_stripes = num_stripes;
4715 for (i = 0; i < ndevs; ++i) {
4716 for (j = 0; j < dev_stripes; ++j) {
4717 int s = i * dev_stripes + j;
4718 map->stripes[s].dev = devices_info[i].dev;
4719 map->stripes[s].physical = devices_info[i].dev_offset +
4723 map->sector_size = extent_root->sectorsize;
4724 map->stripe_len = raid_stripe_len;
4725 map->io_align = raid_stripe_len;
4726 map->io_width = raid_stripe_len;
4728 map->sub_stripes = sub_stripes;
4730 num_bytes = stripe_size * data_stripes;
4732 trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
4734 em = alloc_extent_map();
4740 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4741 em->map_lookup = map;
4743 em->len = num_bytes;
4744 em->block_start = 0;
4745 em->block_len = em->len;
4746 em->orig_block_len = stripe_size;
4748 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4749 write_lock(&em_tree->lock);
4750 ret = add_extent_mapping(em_tree, em, 0);
4752 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4753 atomic_inc(&em->refs);
4755 write_unlock(&em_tree->lock);
4757 free_extent_map(em);
4761 ret = btrfs_make_block_group(trans, extent_root, 0, type,
4762 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4765 goto error_del_extent;
4767 for (i = 0; i < map->num_stripes; i++) {
4768 num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4769 btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4772 spin_lock(&extent_root->fs_info->free_chunk_lock);
4773 extent_root->fs_info->free_chunk_space -= (stripe_size *
4775 spin_unlock(&extent_root->fs_info->free_chunk_lock);
4777 free_extent_map(em);
4778 check_raid56_incompat_flag(extent_root->fs_info, type);
4780 kfree(devices_info);
4784 write_lock(&em_tree->lock);
4785 remove_extent_mapping(em_tree, em);
4786 write_unlock(&em_tree->lock);
4788 /* One for our allocation */
4789 free_extent_map(em);
4790 /* One for the tree reference */
4791 free_extent_map(em);
4792 /* One for the pending_chunks list reference */
4793 free_extent_map(em);
4795 kfree(devices_info);
4799 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4800 struct btrfs_root *extent_root,
4801 u64 chunk_offset, u64 chunk_size)
4803 struct btrfs_key key;
4804 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4805 struct btrfs_device *device;
4806 struct btrfs_chunk *chunk;
4807 struct btrfs_stripe *stripe;
4808 struct extent_map_tree *em_tree;
4809 struct extent_map *em;
4810 struct map_lookup *map;
4817 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4818 read_lock(&em_tree->lock);
4819 em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size);
4820 read_unlock(&em_tree->lock);
4823 btrfs_crit(extent_root->fs_info, "unable to find logical "
4824 "%Lu len %Lu", chunk_offset, chunk_size);
4828 if (em->start != chunk_offset || em->len != chunk_size) {
4829 btrfs_crit(extent_root->fs_info, "found a bad mapping, wanted"
4830 " %Lu-%Lu, found %Lu-%Lu", chunk_offset,
4831 chunk_size, em->start, em->len);
4832 free_extent_map(em);
4836 map = em->map_lookup;
4837 item_size = btrfs_chunk_item_size(map->num_stripes);
4838 stripe_size = em->orig_block_len;
4840 chunk = kzalloc(item_size, GFP_NOFS);
4847 * Take the device list mutex to prevent races with the final phase of
4848 * a device replace operation that replaces the device object associated
4849 * with the map's stripes, because the device object's id can change
4850 * at any time during that final phase of the device replace operation
4851 * (dev-replace.c:btrfs_dev_replace_finishing()).
4853 mutex_lock(&chunk_root->fs_info->fs_devices->device_list_mutex);
4854 for (i = 0; i < map->num_stripes; i++) {
4855 device = map->stripes[i].dev;
4856 dev_offset = map->stripes[i].physical;
4858 ret = btrfs_update_device(trans, device);
4861 ret = btrfs_alloc_dev_extent(trans, device,
4862 chunk_root->root_key.objectid,
4863 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4864 chunk_offset, dev_offset,
4870 mutex_unlock(&chunk_root->fs_info->fs_devices->device_list_mutex);
4874 stripe = &chunk->stripe;
4875 for (i = 0; i < map->num_stripes; i++) {
4876 device = map->stripes[i].dev;
4877 dev_offset = map->stripes[i].physical;
4879 btrfs_set_stack_stripe_devid(stripe, device->devid);
4880 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4881 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4884 mutex_unlock(&chunk_root->fs_info->fs_devices->device_list_mutex);
4886 btrfs_set_stack_chunk_length(chunk, chunk_size);
4887 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4888 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4889 btrfs_set_stack_chunk_type(chunk, map->type);
4890 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4891 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4892 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4893 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4894 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4896 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4897 key.type = BTRFS_CHUNK_ITEM_KEY;
4898 key.offset = chunk_offset;
4900 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4901 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4903 * TODO: Cleanup of inserted chunk root in case of
4906 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
4912 free_extent_map(em);
4917 * Chunk allocation falls into two parts. The first part does works
4918 * that make the new allocated chunk useable, but not do any operation
4919 * that modifies the chunk tree. The second part does the works that
4920 * require modifying the chunk tree. This division is important for the
4921 * bootstrap process of adding storage to a seed btrfs.
4923 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4924 struct btrfs_root *extent_root, u64 type)
4928 ASSERT(mutex_is_locked(&extent_root->fs_info->chunk_mutex));
4929 chunk_offset = find_next_chunk(extent_root->fs_info);
4930 return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type);
4933 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4934 struct btrfs_root *root,
4935 struct btrfs_device *device)
4938 u64 sys_chunk_offset;
4940 struct btrfs_fs_info *fs_info = root->fs_info;
4941 struct btrfs_root *extent_root = fs_info->extent_root;
4944 chunk_offset = find_next_chunk(fs_info);
4945 alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
4946 ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset,
4951 sys_chunk_offset = find_next_chunk(root->fs_info);
4952 alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
4953 ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset,
4958 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
4962 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
4963 BTRFS_BLOCK_GROUP_RAID10 |
4964 BTRFS_BLOCK_GROUP_RAID5 |
4965 BTRFS_BLOCK_GROUP_DUP)) {
4967 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
4976 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
4978 struct extent_map *em;
4979 struct map_lookup *map;
4980 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4985 read_lock(&map_tree->map_tree.lock);
4986 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
4987 read_unlock(&map_tree->map_tree.lock);
4991 map = em->map_lookup;
4992 for (i = 0; i < map->num_stripes; i++) {
4993 if (map->stripes[i].dev->missing) {
4998 if (!map->stripes[i].dev->writeable) {
5005 * If the number of missing devices is larger than max errors,
5006 * we can not write the data into that chunk successfully, so
5009 if (miss_ndevs > btrfs_chunk_max_errors(map))
5012 free_extent_map(em);
5016 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
5018 extent_map_tree_init(&tree->map_tree);
5021 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
5023 struct extent_map *em;
5026 write_lock(&tree->map_tree.lock);
5027 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
5029 remove_extent_mapping(&tree->map_tree, em);
5030 write_unlock(&tree->map_tree.lock);
5034 free_extent_map(em);
5035 /* once for the tree */
5036 free_extent_map(em);
5040 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5042 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5043 struct extent_map *em;
5044 struct map_lookup *map;
5045 struct extent_map_tree *em_tree = &map_tree->map_tree;
5048 read_lock(&em_tree->lock);
5049 em = lookup_extent_mapping(em_tree, logical, len);
5050 read_unlock(&em_tree->lock);
5053 * We could return errors for these cases, but that could get ugly and
5054 * we'd probably do the same thing which is just not do anything else
5055 * and exit, so return 1 so the callers don't try to use other copies.
5058 btrfs_crit(fs_info, "No mapping for %Lu-%Lu", logical,
5063 if (em->start > logical || em->start + em->len < logical) {
5064 btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got "
5065 "%Lu-%Lu", logical, logical+len, em->start,
5066 em->start + em->len);
5067 free_extent_map(em);
5071 map = em->map_lookup;
5072 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
5073 ret = map->num_stripes;
5074 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5075 ret = map->sub_stripes;
5076 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5078 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5082 free_extent_map(em);
5084 btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
5085 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
5087 btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
5092 unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
5093 struct btrfs_mapping_tree *map_tree,
5096 struct extent_map *em;
5097 struct map_lookup *map;
5098 struct extent_map_tree *em_tree = &map_tree->map_tree;
5099 unsigned long len = root->sectorsize;
5101 read_lock(&em_tree->lock);
5102 em = lookup_extent_mapping(em_tree, logical, len);
5103 read_unlock(&em_tree->lock);
5106 BUG_ON(em->start > logical || em->start + em->len < logical);
5107 map = em->map_lookup;
5108 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5109 len = map->stripe_len * nr_data_stripes(map);
5110 free_extent_map(em);
5114 int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
5115 u64 logical, u64 len, int mirror_num)
5117 struct extent_map *em;
5118 struct map_lookup *map;
5119 struct extent_map_tree *em_tree = &map_tree->map_tree;
5122 read_lock(&em_tree->lock);
5123 em = lookup_extent_mapping(em_tree, logical, len);
5124 read_unlock(&em_tree->lock);
5127 BUG_ON(em->start > logical || em->start + em->len < logical);
5128 map = em->map_lookup;
5129 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5131 free_extent_map(em);
5135 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5136 struct map_lookup *map, int first, int num,
5137 int optimal, int dev_replace_is_ongoing)
5141 struct btrfs_device *srcdev;
5143 if (dev_replace_is_ongoing &&
5144 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5145 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5146 srcdev = fs_info->dev_replace.srcdev;
5151 * try to avoid the drive that is the source drive for a
5152 * dev-replace procedure, only choose it if no other non-missing
5153 * mirror is available
5155 for (tolerance = 0; tolerance < 2; tolerance++) {
5156 if (map->stripes[optimal].dev->bdev &&
5157 (tolerance || map->stripes[optimal].dev != srcdev))
5159 for (i = first; i < first + num; i++) {
5160 if (map->stripes[i].dev->bdev &&
5161 (tolerance || map->stripes[i].dev != srcdev))
5166 /* we couldn't find one that doesn't fail. Just return something
5167 * and the io error handling code will clean up eventually
5172 static inline int parity_smaller(u64 a, u64 b)
5177 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5178 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5180 struct btrfs_bio_stripe s;
5187 for (i = 0; i < num_stripes - 1; i++) {
5188 if (parity_smaller(bbio->raid_map[i],
5189 bbio->raid_map[i+1])) {
5190 s = bbio->stripes[i];
5191 l = bbio->raid_map[i];
5192 bbio->stripes[i] = bbio->stripes[i+1];
5193 bbio->raid_map[i] = bbio->raid_map[i+1];
5194 bbio->stripes[i+1] = s;
5195 bbio->raid_map[i+1] = l;
5203 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5205 struct btrfs_bio *bbio = kzalloc(
5206 /* the size of the btrfs_bio */
5207 sizeof(struct btrfs_bio) +
5208 /* plus the variable array for the stripes */
5209 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5210 /* plus the variable array for the tgt dev */
5211 sizeof(int) * (real_stripes) +
5213 * plus the raid_map, which includes both the tgt dev
5216 sizeof(u64) * (total_stripes),
5217 GFP_NOFS|__GFP_NOFAIL);
5219 atomic_set(&bbio->error, 0);
5220 atomic_set(&bbio->refs, 1);
5225 void btrfs_get_bbio(struct btrfs_bio *bbio)
5227 WARN_ON(!atomic_read(&bbio->refs));
5228 atomic_inc(&bbio->refs);
5231 void btrfs_put_bbio(struct btrfs_bio *bbio)
5235 if (atomic_dec_and_test(&bbio->refs))
5239 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5240 u64 logical, u64 *length,
5241 struct btrfs_bio **bbio_ret,
5242 int mirror_num, int need_raid_map)
5244 struct extent_map *em;
5245 struct map_lookup *map;
5246 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5247 struct extent_map_tree *em_tree = &map_tree->map_tree;
5250 u64 stripe_end_offset;
5260 int tgtdev_indexes = 0;
5261 struct btrfs_bio *bbio = NULL;
5262 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
5263 int dev_replace_is_ongoing = 0;
5264 int num_alloc_stripes;
5265 int patch_the_first_stripe_for_dev_replace = 0;
5266 u64 physical_to_patch_in_first_stripe = 0;
5267 u64 raid56_full_stripe_start = (u64)-1;
5269 read_lock(&em_tree->lock);
5270 em = lookup_extent_mapping(em_tree, logical, *length);
5271 read_unlock(&em_tree->lock);
5274 btrfs_crit(fs_info, "unable to find logical %llu len %llu",
5279 if (em->start > logical || em->start + em->len < logical) {
5280 btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, "
5281 "found %Lu-%Lu", logical, em->start,
5282 em->start + em->len);
5283 free_extent_map(em);
5287 map = em->map_lookup;
5288 offset = logical - em->start;
5290 stripe_len = map->stripe_len;
5293 * stripe_nr counts the total number of stripes we have to stride
5294 * to get to this block
5296 stripe_nr = div64_u64(stripe_nr, stripe_len);
5298 stripe_offset = stripe_nr * stripe_len;
5299 BUG_ON(offset < stripe_offset);
5301 /* stripe_offset is the offset of this block in its stripe*/
5302 stripe_offset = offset - stripe_offset;
5304 /* if we're here for raid56, we need to know the stripe aligned start */
5305 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5306 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5307 raid56_full_stripe_start = offset;
5309 /* allow a write of a full stripe, but make sure we don't
5310 * allow straddling of stripes
5312 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5314 raid56_full_stripe_start *= full_stripe_len;
5317 if (rw & REQ_DISCARD) {
5318 /* we don't discard raid56 yet */
5319 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5323 *length = min_t(u64, em->len - offset, *length);
5324 } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5326 /* For writes to RAID[56], allow a full stripeset across all disks.
5327 For other RAID types and for RAID[56] reads, just allow a single
5328 stripe (on a single disk). */
5329 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
5331 max_len = stripe_len * nr_data_stripes(map) -
5332 (offset - raid56_full_stripe_start);
5334 /* we limit the length of each bio to what fits in a stripe */
5335 max_len = stripe_len - stripe_offset;
5337 *length = min_t(u64, em->len - offset, max_len);
5339 *length = em->len - offset;
5342 /* This is for when we're called from btrfs_merge_bio_hook() and all
5343 it cares about is the length */
5347 btrfs_dev_replace_lock(dev_replace, 0);
5348 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5349 if (!dev_replace_is_ongoing)
5350 btrfs_dev_replace_unlock(dev_replace, 0);
5352 btrfs_dev_replace_set_lock_blocking(dev_replace);
5354 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
5355 !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) &&
5356 dev_replace->tgtdev != NULL) {
5358 * in dev-replace case, for repair case (that's the only
5359 * case where the mirror is selected explicitly when
5360 * calling btrfs_map_block), blocks left of the left cursor
5361 * can also be read from the target drive.
5362 * For REQ_GET_READ_MIRRORS, the target drive is added as
5363 * the last one to the array of stripes. For READ, it also
5364 * needs to be supported using the same mirror number.
5365 * If the requested block is not left of the left cursor,
5366 * EIO is returned. This can happen because btrfs_num_copies()
5367 * returns one more in the dev-replace case.
5369 u64 tmp_length = *length;
5370 struct btrfs_bio *tmp_bbio = NULL;
5371 int tmp_num_stripes;
5372 u64 srcdev_devid = dev_replace->srcdev->devid;
5373 int index_srcdev = 0;
5375 u64 physical_of_found = 0;
5377 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
5378 logical, &tmp_length, &tmp_bbio, 0, 0);
5380 WARN_ON(tmp_bbio != NULL);
5384 tmp_num_stripes = tmp_bbio->num_stripes;
5385 if (mirror_num > tmp_num_stripes) {
5387 * REQ_GET_READ_MIRRORS does not contain this
5388 * mirror, that means that the requested area
5389 * is not left of the left cursor
5392 btrfs_put_bbio(tmp_bbio);
5397 * process the rest of the function using the mirror_num
5398 * of the source drive. Therefore look it up first.
5399 * At the end, patch the device pointer to the one of the
5402 for (i = 0; i < tmp_num_stripes; i++) {
5403 if (tmp_bbio->stripes[i].dev->devid != srcdev_devid)
5407 * In case of DUP, in order to keep it simple, only add
5408 * the mirror with the lowest physical address
5411 physical_of_found <= tmp_bbio->stripes[i].physical)
5416 physical_of_found = tmp_bbio->stripes[i].physical;
5419 btrfs_put_bbio(tmp_bbio);
5427 mirror_num = index_srcdev + 1;
5428 patch_the_first_stripe_for_dev_replace = 1;
5429 physical_to_patch_in_first_stripe = physical_of_found;
5430 } else if (mirror_num > map->num_stripes) {
5436 stripe_nr_orig = stripe_nr;
5437 stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
5438 stripe_nr_end = div_u64(stripe_nr_end, map->stripe_len);
5439 stripe_end_offset = stripe_nr_end * map->stripe_len -
5442 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5443 if (rw & REQ_DISCARD)
5444 num_stripes = min_t(u64, map->num_stripes,
5445 stripe_nr_end - stripe_nr_orig);
5446 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5448 if (!(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)))
5450 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
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;
5456 stripe_index = find_live_mirror(fs_info, map, 0,
5458 current->pid % map->num_stripes,
5459 dev_replace_is_ongoing);
5460 mirror_num = stripe_index + 1;
5463 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
5464 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
5465 num_stripes = map->num_stripes;
5466 } else if (mirror_num) {
5467 stripe_index = mirror_num - 1;
5472 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5473 u32 factor = map->num_stripes / map->sub_stripes;
5475 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5476 stripe_index *= map->sub_stripes;
5478 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
5479 num_stripes = map->sub_stripes;
5480 else if (rw & REQ_DISCARD)
5481 num_stripes = min_t(u64, map->sub_stripes *
5482 (stripe_nr_end - stripe_nr_orig),
5484 else if (mirror_num)
5485 stripe_index += mirror_num - 1;
5487 int old_stripe_index = stripe_index;
5488 stripe_index = find_live_mirror(fs_info, map,
5490 map->sub_stripes, stripe_index +
5491 current->pid % map->sub_stripes,
5492 dev_replace_is_ongoing);
5493 mirror_num = stripe_index - old_stripe_index + 1;
5496 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5497 if (need_raid_map &&
5498 ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) ||
5500 /* push stripe_nr back to the start of the full stripe */
5501 stripe_nr = div_u64(raid56_full_stripe_start,
5502 stripe_len * nr_data_stripes(map));
5504 /* RAID[56] write or recovery. Return all stripes */
5505 num_stripes = map->num_stripes;
5506 max_errors = nr_parity_stripes(map);
5508 *length = map->stripe_len;
5513 * Mirror #0 or #1 means the original data block.
5514 * Mirror #2 is RAID5 parity block.
5515 * Mirror #3 is RAID6 Q block.
5517 stripe_nr = div_u64_rem(stripe_nr,
5518 nr_data_stripes(map), &stripe_index);
5520 stripe_index = nr_data_stripes(map) +
5523 /* We distribute the parity blocks across stripes */
5524 div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
5526 if (!(rw & (REQ_WRITE | REQ_DISCARD |
5527 REQ_GET_READ_MIRRORS)) && mirror_num <= 1)
5532 * after this, stripe_nr is the number of stripes on this
5533 * device we have to walk to find the data, and stripe_index is
5534 * the number of our device in the stripe array
5536 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5538 mirror_num = stripe_index + 1;
5540 BUG_ON(stripe_index >= map->num_stripes);
5542 num_alloc_stripes = num_stripes;
5543 if (dev_replace_is_ongoing) {
5544 if (rw & (REQ_WRITE | REQ_DISCARD))
5545 num_alloc_stripes <<= 1;
5546 if (rw & REQ_GET_READ_MIRRORS)
5547 num_alloc_stripes++;
5548 tgtdev_indexes = num_stripes;
5551 bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
5556 if (dev_replace_is_ongoing)
5557 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
5559 /* build raid_map */
5560 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK &&
5561 need_raid_map && ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) ||
5566 bbio->raid_map = (u64 *)((void *)bbio->stripes +
5567 sizeof(struct btrfs_bio_stripe) *
5569 sizeof(int) * tgtdev_indexes);
5571 /* Work out the disk rotation on this stripe-set */
5572 div_u64_rem(stripe_nr, num_stripes, &rot);
5574 /* Fill in the logical address of each stripe */
5575 tmp = stripe_nr * nr_data_stripes(map);
5576 for (i = 0; i < nr_data_stripes(map); i++)
5577 bbio->raid_map[(i+rot) % num_stripes] =
5578 em->start + (tmp + i) * map->stripe_len;
5580 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5581 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5582 bbio->raid_map[(i+rot+1) % num_stripes] =
5586 if (rw & REQ_DISCARD) {
5588 u32 sub_stripes = 0;
5589 u64 stripes_per_dev = 0;
5590 u32 remaining_stripes = 0;
5591 u32 last_stripe = 0;
5594 (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
5595 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5598 sub_stripes = map->sub_stripes;
5600 factor = map->num_stripes / sub_stripes;
5601 stripes_per_dev = div_u64_rem(stripe_nr_end -
5604 &remaining_stripes);
5605 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5606 last_stripe *= sub_stripes;
5609 for (i = 0; i < num_stripes; i++) {
5610 bbio->stripes[i].physical =
5611 map->stripes[stripe_index].physical +
5612 stripe_offset + stripe_nr * map->stripe_len;
5613 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5615 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5616 BTRFS_BLOCK_GROUP_RAID10)) {
5617 bbio->stripes[i].length = stripes_per_dev *
5620 if (i / sub_stripes < remaining_stripes)
5621 bbio->stripes[i].length +=
5625 * Special for the first stripe and
5628 * |-------|...|-------|
5632 if (i < sub_stripes)
5633 bbio->stripes[i].length -=
5636 if (stripe_index >= last_stripe &&
5637 stripe_index <= (last_stripe +
5639 bbio->stripes[i].length -=
5642 if (i == sub_stripes - 1)
5645 bbio->stripes[i].length = *length;
5648 if (stripe_index == map->num_stripes) {
5649 /* This could only happen for RAID0/10 */
5655 for (i = 0; i < num_stripes; i++) {
5656 bbio->stripes[i].physical =
5657 map->stripes[stripe_index].physical +
5659 stripe_nr * map->stripe_len;
5660 bbio->stripes[i].dev =
5661 map->stripes[stripe_index].dev;
5666 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
5667 max_errors = btrfs_chunk_max_errors(map);
5670 sort_parity_stripes(bbio, num_stripes);
5673 if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) &&
5674 dev_replace->tgtdev != NULL) {
5675 int index_where_to_add;
5676 u64 srcdev_devid = dev_replace->srcdev->devid;
5679 * duplicate the write operations while the dev replace
5680 * procedure is running. Since the copying of the old disk
5681 * to the new disk takes place at run time while the
5682 * filesystem is mounted writable, the regular write
5683 * operations to the old disk have to be duplicated to go
5684 * to the new disk as well.
5685 * Note that device->missing is handled by the caller, and
5686 * that the write to the old disk is already set up in the
5689 index_where_to_add = num_stripes;
5690 for (i = 0; i < num_stripes; i++) {
5691 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5692 /* write to new disk, too */
5693 struct btrfs_bio_stripe *new =
5694 bbio->stripes + index_where_to_add;
5695 struct btrfs_bio_stripe *old =
5698 new->physical = old->physical;
5699 new->length = old->length;
5700 new->dev = dev_replace->tgtdev;
5701 bbio->tgtdev_map[i] = index_where_to_add;
5702 index_where_to_add++;
5707 num_stripes = index_where_to_add;
5708 } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) &&
5709 dev_replace->tgtdev != NULL) {
5710 u64 srcdev_devid = dev_replace->srcdev->devid;
5711 int index_srcdev = 0;
5713 u64 physical_of_found = 0;
5716 * During the dev-replace procedure, the target drive can
5717 * also be used to read data in case it is needed to repair
5718 * a corrupt block elsewhere. This is possible if the
5719 * requested area is left of the left cursor. In this area,
5720 * the target drive is a full copy of the source drive.
5722 for (i = 0; i < num_stripes; i++) {
5723 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5725 * In case of DUP, in order to keep it
5726 * simple, only add the mirror with the
5727 * lowest physical address
5730 physical_of_found <=
5731 bbio->stripes[i].physical)
5735 physical_of_found = bbio->stripes[i].physical;
5739 if (physical_of_found + map->stripe_len <=
5740 dev_replace->cursor_left) {
5741 struct btrfs_bio_stripe *tgtdev_stripe =
5742 bbio->stripes + num_stripes;
5744 tgtdev_stripe->physical = physical_of_found;
5745 tgtdev_stripe->length =
5746 bbio->stripes[index_srcdev].length;
5747 tgtdev_stripe->dev = dev_replace->tgtdev;
5748 bbio->tgtdev_map[index_srcdev] = num_stripes;
5757 bbio->map_type = map->type;
5758 bbio->num_stripes = num_stripes;
5759 bbio->max_errors = max_errors;
5760 bbio->mirror_num = mirror_num;
5761 bbio->num_tgtdevs = tgtdev_indexes;
5764 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5765 * mirror_num == num_stripes + 1 && dev_replace target drive is
5766 * available as a mirror
5768 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5769 WARN_ON(num_stripes > 1);
5770 bbio->stripes[0].dev = dev_replace->tgtdev;
5771 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5772 bbio->mirror_num = map->num_stripes + 1;
5775 if (dev_replace_is_ongoing) {
5776 btrfs_dev_replace_clear_lock_blocking(dev_replace);
5777 btrfs_dev_replace_unlock(dev_replace, 0);
5779 free_extent_map(em);
5783 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5784 u64 logical, u64 *length,
5785 struct btrfs_bio **bbio_ret, int mirror_num)
5787 return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5791 /* For Scrub/replace */
5792 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, int rw,
5793 u64 logical, u64 *length,
5794 struct btrfs_bio **bbio_ret, int mirror_num,
5797 return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5798 mirror_num, need_raid_map);
5801 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
5802 u64 chunk_start, u64 physical, u64 devid,
5803 u64 **logical, int *naddrs, int *stripe_len)
5805 struct extent_map_tree *em_tree = &map_tree->map_tree;
5806 struct extent_map *em;
5807 struct map_lookup *map;
5815 read_lock(&em_tree->lock);
5816 em = lookup_extent_mapping(em_tree, chunk_start, 1);
5817 read_unlock(&em_tree->lock);
5820 printk(KERN_ERR "BTRFS: couldn't find em for chunk %Lu\n",
5825 if (em->start != chunk_start) {
5826 printk(KERN_ERR "BTRFS: bad chunk start, em=%Lu, wanted=%Lu\n",
5827 em->start, chunk_start);
5828 free_extent_map(em);
5831 map = em->map_lookup;
5834 rmap_len = map->stripe_len;
5836 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5837 length = div_u64(length, map->num_stripes / map->sub_stripes);
5838 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5839 length = div_u64(length, map->num_stripes);
5840 else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5841 length = div_u64(length, nr_data_stripes(map));
5842 rmap_len = map->stripe_len * nr_data_stripes(map);
5845 buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
5846 BUG_ON(!buf); /* -ENOMEM */
5848 for (i = 0; i < map->num_stripes; i++) {
5849 if (devid && map->stripes[i].dev->devid != devid)
5851 if (map->stripes[i].physical > physical ||
5852 map->stripes[i].physical + length <= physical)
5855 stripe_nr = physical - map->stripes[i].physical;
5856 stripe_nr = div_u64(stripe_nr, map->stripe_len);
5858 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5859 stripe_nr = stripe_nr * map->num_stripes + i;
5860 stripe_nr = div_u64(stripe_nr, map->sub_stripes);
5861 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5862 stripe_nr = stripe_nr * map->num_stripes + i;
5863 } /* else if RAID[56], multiply by nr_data_stripes().
5864 * Alternatively, just use rmap_len below instead of
5865 * map->stripe_len */
5867 bytenr = chunk_start + stripe_nr * rmap_len;
5868 WARN_ON(nr >= map->num_stripes);
5869 for (j = 0; j < nr; j++) {
5870 if (buf[j] == bytenr)
5874 WARN_ON(nr >= map->num_stripes);
5881 *stripe_len = rmap_len;
5883 free_extent_map(em);
5887 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
5889 bio->bi_private = bbio->private;
5890 bio->bi_end_io = bbio->end_io;
5893 btrfs_put_bbio(bbio);
5896 static void btrfs_end_bio(struct bio *bio)
5898 struct btrfs_bio *bbio = bio->bi_private;
5899 int is_orig_bio = 0;
5901 if (bio->bi_error) {
5902 atomic_inc(&bbio->error);
5903 if (bio->bi_error == -EIO || bio->bi_error == -EREMOTEIO) {
5904 unsigned int stripe_index =
5905 btrfs_io_bio(bio)->stripe_index;
5906 struct btrfs_device *dev;
5908 BUG_ON(stripe_index >= bbio->num_stripes);
5909 dev = bbio->stripes[stripe_index].dev;
5911 if (bio->bi_rw & WRITE)
5912 btrfs_dev_stat_inc(dev,
5913 BTRFS_DEV_STAT_WRITE_ERRS);
5915 btrfs_dev_stat_inc(dev,
5916 BTRFS_DEV_STAT_READ_ERRS);
5917 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
5918 btrfs_dev_stat_inc(dev,
5919 BTRFS_DEV_STAT_FLUSH_ERRS);
5920 btrfs_dev_stat_print_on_error(dev);
5925 if (bio == bbio->orig_bio)
5928 btrfs_bio_counter_dec(bbio->fs_info);
5930 if (atomic_dec_and_test(&bbio->stripes_pending)) {
5933 bio = bbio->orig_bio;
5936 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5937 /* only send an error to the higher layers if it is
5938 * beyond the tolerance of the btrfs bio
5940 if (atomic_read(&bbio->error) > bbio->max_errors) {
5941 bio->bi_error = -EIO;
5944 * this bio is actually up to date, we didn't
5945 * go over the max number of errors
5950 btrfs_end_bbio(bbio, bio);
5951 } else if (!is_orig_bio) {
5957 * see run_scheduled_bios for a description of why bios are collected for
5960 * This will add one bio to the pending list for a device and make sure
5961 * the work struct is scheduled.
5963 static noinline void btrfs_schedule_bio(struct btrfs_root *root,
5964 struct btrfs_device *device,
5965 int rw, struct bio *bio)
5967 int should_queue = 1;
5968 struct btrfs_pending_bios *pending_bios;
5970 if (device->missing || !device->bdev) {
5975 /* don't bother with additional async steps for reads, right now */
5976 if (!(rw & REQ_WRITE)) {
5978 btrfsic_submit_bio(rw, bio);
5984 * nr_async_bios allows us to reliably return congestion to the
5985 * higher layers. Otherwise, the async bio makes it appear we have
5986 * made progress against dirty pages when we've really just put it
5987 * on a queue for later
5989 atomic_inc(&root->fs_info->nr_async_bios);
5990 WARN_ON(bio->bi_next);
5991 bio->bi_next = NULL;
5994 spin_lock(&device->io_lock);
5995 if (bio->bi_rw & REQ_SYNC)
5996 pending_bios = &device->pending_sync_bios;
5998 pending_bios = &device->pending_bios;
6000 if (pending_bios->tail)
6001 pending_bios->tail->bi_next = bio;
6003 pending_bios->tail = bio;
6004 if (!pending_bios->head)
6005 pending_bios->head = bio;
6006 if (device->running_pending)
6009 spin_unlock(&device->io_lock);
6012 btrfs_queue_work(root->fs_info->submit_workers,
6016 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
6017 struct bio *bio, u64 physical, int dev_nr,
6020 struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
6022 bio->bi_private = bbio;
6023 btrfs_io_bio(bio)->stripe_index = dev_nr;
6024 bio->bi_end_io = btrfs_end_bio;
6025 bio->bi_iter.bi_sector = physical >> 9;
6028 struct rcu_string *name;
6031 name = rcu_dereference(dev->name);
6032 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
6033 "(%s id %llu), size=%u\n", rw,
6034 (u64)bio->bi_iter.bi_sector, (u_long)dev->bdev->bd_dev,
6035 name->str, dev->devid, bio->bi_iter.bi_size);
6039 bio->bi_bdev = dev->bdev;
6041 btrfs_bio_counter_inc_noblocked(root->fs_info);
6044 btrfs_schedule_bio(root, dev, rw, bio);
6046 btrfsic_submit_bio(rw, bio);
6049 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6051 atomic_inc(&bbio->error);
6052 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6053 /* Shoud be the original bio. */
6054 WARN_ON(bio != bbio->orig_bio);
6056 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6057 bio->bi_iter.bi_sector = logical >> 9;
6058 bio->bi_error = -EIO;
6059 btrfs_end_bbio(bbio, bio);
6063 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
6064 int mirror_num, int async_submit)
6066 struct btrfs_device *dev;
6067 struct bio *first_bio = bio;
6068 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
6074 struct btrfs_bio *bbio = NULL;
6076 length = bio->bi_iter.bi_size;
6077 map_length = length;
6079 btrfs_bio_counter_inc_blocked(root->fs_info);
6080 ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
6083 btrfs_bio_counter_dec(root->fs_info);
6087 total_devs = bbio->num_stripes;
6088 bbio->orig_bio = first_bio;
6089 bbio->private = first_bio->bi_private;
6090 bbio->end_io = first_bio->bi_end_io;
6091 bbio->fs_info = root->fs_info;
6092 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6094 if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6095 ((rw & WRITE) || (mirror_num > 1))) {
6096 /* In this case, map_length has been set to the length of
6097 a single stripe; not the whole write */
6099 ret = raid56_parity_write(root, bio, bbio, map_length);
6101 ret = raid56_parity_recover(root, bio, bbio, map_length,
6105 btrfs_bio_counter_dec(root->fs_info);
6109 if (map_length < length) {
6110 btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu",
6111 logical, length, map_length);
6115 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6116 dev = bbio->stripes[dev_nr].dev;
6117 if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
6118 bbio_error(bbio, first_bio, logical);
6122 if (dev_nr < total_devs - 1) {
6123 bio = btrfs_bio_clone(first_bio, GFP_NOFS);
6124 BUG_ON(!bio); /* -ENOMEM */
6128 submit_stripe_bio(root, bbio, bio,
6129 bbio->stripes[dev_nr].physical, dev_nr, rw,
6132 btrfs_bio_counter_dec(root->fs_info);
6136 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
6139 struct btrfs_device *device;
6140 struct btrfs_fs_devices *cur_devices;
6142 cur_devices = fs_info->fs_devices;
6143 while (cur_devices) {
6145 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
6146 device = __find_device(&cur_devices->devices,
6151 cur_devices = cur_devices->seed;
6156 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
6157 struct btrfs_fs_devices *fs_devices,
6158 u64 devid, u8 *dev_uuid)
6160 struct btrfs_device *device;
6162 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6166 list_add(&device->dev_list, &fs_devices->devices);
6167 device->fs_devices = fs_devices;
6168 fs_devices->num_devices++;
6170 device->missing = 1;
6171 fs_devices->missing_devices++;
6177 * btrfs_alloc_device - allocate struct btrfs_device
6178 * @fs_info: used only for generating a new devid, can be NULL if
6179 * devid is provided (i.e. @devid != NULL).
6180 * @devid: a pointer to devid for this device. If NULL a new devid
6182 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6185 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6186 * on error. Returned struct is not linked onto any lists and can be
6187 * destroyed with kfree() right away.
6189 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6193 struct btrfs_device *dev;
6196 if (WARN_ON(!devid && !fs_info))
6197 return ERR_PTR(-EINVAL);
6199 dev = __alloc_device();
6208 ret = find_next_devid(fs_info, &tmp);
6211 return ERR_PTR(ret);
6217 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6219 generate_random_uuid(dev->uuid);
6221 btrfs_init_work(&dev->work, btrfs_submit_helper,
6222 pending_bios_fn, NULL, NULL);
6227 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
6228 struct extent_buffer *leaf,
6229 struct btrfs_chunk *chunk)
6231 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
6232 struct map_lookup *map;
6233 struct extent_map *em;
6238 u8 uuid[BTRFS_UUID_SIZE];
6243 logical = key->offset;
6244 length = btrfs_chunk_length(leaf, chunk);
6245 stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6246 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6247 /* Validation check */
6249 btrfs_err(root->fs_info, "invalid chunk num_stripes: %u",
6253 if (!IS_ALIGNED(logical, root->sectorsize)) {
6254 btrfs_err(root->fs_info,
6255 "invalid chunk logical %llu", logical);
6258 if (!length || !IS_ALIGNED(length, root->sectorsize)) {
6259 btrfs_err(root->fs_info,
6260 "invalid chunk length %llu", length);
6263 if (!is_power_of_2(stripe_len)) {
6264 btrfs_err(root->fs_info, "invalid chunk stripe length: %llu",
6268 if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6269 btrfs_chunk_type(leaf, chunk)) {
6270 btrfs_err(root->fs_info, "unrecognized chunk type: %llu",
6271 ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
6272 BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6273 btrfs_chunk_type(leaf, chunk));
6277 read_lock(&map_tree->map_tree.lock);
6278 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
6279 read_unlock(&map_tree->map_tree.lock);
6281 /* already mapped? */
6282 if (em && em->start <= logical && em->start + em->len > logical) {
6283 free_extent_map(em);
6286 free_extent_map(em);
6289 em = alloc_extent_map();
6292 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6294 free_extent_map(em);
6298 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6299 em->map_lookup = map;
6300 em->start = logical;
6303 em->block_start = 0;
6304 em->block_len = em->len;
6306 map->num_stripes = num_stripes;
6307 map->io_width = btrfs_chunk_io_width(leaf, chunk);
6308 map->io_align = btrfs_chunk_io_align(leaf, chunk);
6309 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
6310 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6311 map->type = btrfs_chunk_type(leaf, chunk);
6312 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6313 for (i = 0; i < num_stripes; i++) {
6314 map->stripes[i].physical =
6315 btrfs_stripe_offset_nr(leaf, chunk, i);
6316 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6317 read_extent_buffer(leaf, uuid, (unsigned long)
6318 btrfs_stripe_dev_uuid_nr(chunk, i),
6320 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
6322 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
6323 free_extent_map(em);
6326 if (!map->stripes[i].dev) {
6327 map->stripes[i].dev =
6328 add_missing_dev(root, root->fs_info->fs_devices,
6330 if (!map->stripes[i].dev) {
6331 free_extent_map(em);
6334 btrfs_warn(root->fs_info, "devid %llu uuid %pU is missing",
6337 map->stripes[i].dev->in_fs_metadata = 1;
6340 write_lock(&map_tree->map_tree.lock);
6341 ret = add_extent_mapping(&map_tree->map_tree, em, 0);
6342 write_unlock(&map_tree->map_tree.lock);
6343 BUG_ON(ret); /* Tree corruption */
6344 free_extent_map(em);
6349 static void fill_device_from_item(struct extent_buffer *leaf,
6350 struct btrfs_dev_item *dev_item,
6351 struct btrfs_device *device)
6355 device->devid = btrfs_device_id(leaf, dev_item);
6356 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6357 device->total_bytes = device->disk_total_bytes;
6358 device->commit_total_bytes = device->disk_total_bytes;
6359 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6360 device->commit_bytes_used = device->bytes_used;
6361 device->type = btrfs_device_type(leaf, dev_item);
6362 device->io_align = btrfs_device_io_align(leaf, dev_item);
6363 device->io_width = btrfs_device_io_width(leaf, dev_item);
6364 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6365 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6366 device->is_tgtdev_for_dev_replace = 0;
6368 ptr = btrfs_device_uuid(dev_item);
6369 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6372 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_root *root,
6375 struct btrfs_fs_devices *fs_devices;
6378 BUG_ON(!mutex_is_locked(&uuid_mutex));
6380 fs_devices = root->fs_info->fs_devices->seed;
6381 while (fs_devices) {
6382 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE))
6385 fs_devices = fs_devices->seed;
6388 fs_devices = find_fsid(fsid);
6390 if (!btrfs_test_opt(root, DEGRADED))
6391 return ERR_PTR(-ENOENT);
6393 fs_devices = alloc_fs_devices(fsid);
6394 if (IS_ERR(fs_devices))
6397 fs_devices->seeding = 1;
6398 fs_devices->opened = 1;
6402 fs_devices = clone_fs_devices(fs_devices);
6403 if (IS_ERR(fs_devices))
6406 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
6407 root->fs_info->bdev_holder);
6409 free_fs_devices(fs_devices);
6410 fs_devices = ERR_PTR(ret);
6414 if (!fs_devices->seeding) {
6415 __btrfs_close_devices(fs_devices);
6416 free_fs_devices(fs_devices);
6417 fs_devices = ERR_PTR(-EINVAL);
6421 fs_devices->seed = root->fs_info->fs_devices->seed;
6422 root->fs_info->fs_devices->seed = fs_devices;
6427 static int read_one_dev(struct btrfs_root *root,
6428 struct extent_buffer *leaf,
6429 struct btrfs_dev_item *dev_item)
6431 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6432 struct btrfs_device *device;
6435 u8 fs_uuid[BTRFS_UUID_SIZE];
6436 u8 dev_uuid[BTRFS_UUID_SIZE];
6438 devid = btrfs_device_id(leaf, dev_item);
6439 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6441 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6444 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
6445 fs_devices = open_seed_devices(root, fs_uuid);
6446 if (IS_ERR(fs_devices))
6447 return PTR_ERR(fs_devices);
6450 device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
6452 if (!btrfs_test_opt(root, DEGRADED))
6455 device = add_missing_dev(root, fs_devices, devid, dev_uuid);
6458 btrfs_warn(root->fs_info, "devid %llu uuid %pU missing",
6461 if (!device->bdev && !btrfs_test_opt(root, DEGRADED))
6464 if(!device->bdev && !device->missing) {
6466 * this happens when a device that was properly setup
6467 * in the device info lists suddenly goes bad.
6468 * device->bdev is NULL, and so we have to set
6469 * device->missing to one here
6471 device->fs_devices->missing_devices++;
6472 device->missing = 1;
6475 /* Move the device to its own fs_devices */
6476 if (device->fs_devices != fs_devices) {
6477 ASSERT(device->missing);
6479 list_move(&device->dev_list, &fs_devices->devices);
6480 device->fs_devices->num_devices--;
6481 fs_devices->num_devices++;
6483 device->fs_devices->missing_devices--;
6484 fs_devices->missing_devices++;
6486 device->fs_devices = fs_devices;
6490 if (device->fs_devices != root->fs_info->fs_devices) {
6491 BUG_ON(device->writeable);
6492 if (device->generation !=
6493 btrfs_device_generation(leaf, dev_item))
6497 fill_device_from_item(leaf, dev_item, device);
6498 device->in_fs_metadata = 1;
6499 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
6500 device->fs_devices->total_rw_bytes += device->total_bytes;
6501 spin_lock(&root->fs_info->free_chunk_lock);
6502 root->fs_info->free_chunk_space += device->total_bytes -
6504 spin_unlock(&root->fs_info->free_chunk_lock);
6510 int btrfs_read_sys_array(struct btrfs_root *root)
6512 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
6513 struct extent_buffer *sb;
6514 struct btrfs_disk_key *disk_key;
6515 struct btrfs_chunk *chunk;
6517 unsigned long sb_array_offset;
6523 struct btrfs_key key;
6525 ASSERT(BTRFS_SUPER_INFO_SIZE <= root->nodesize);
6527 * This will create extent buffer of nodesize, superblock size is
6528 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6529 * overallocate but we can keep it as-is, only the first page is used.
6531 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET);
6534 set_extent_buffer_uptodate(sb);
6535 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6537 * The sb extent buffer is artifical and just used to read the system array.
6538 * set_extent_buffer_uptodate() call does not properly mark all it's
6539 * pages up-to-date when the page is larger: extent does not cover the
6540 * whole page and consequently check_page_uptodate does not find all
6541 * the page's extents up-to-date (the hole beyond sb),
6542 * write_extent_buffer then triggers a WARN_ON.
6544 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6545 * but sb spans only this function. Add an explicit SetPageUptodate call
6546 * to silence the warning eg. on PowerPC 64.
6548 if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
6549 SetPageUptodate(sb->pages[0]);
6551 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6552 array_size = btrfs_super_sys_array_size(super_copy);
6554 array_ptr = super_copy->sys_chunk_array;
6555 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6558 while (cur_offset < array_size) {
6559 disk_key = (struct btrfs_disk_key *)array_ptr;
6560 len = sizeof(*disk_key);
6561 if (cur_offset + len > array_size)
6562 goto out_short_read;
6564 btrfs_disk_key_to_cpu(&key, disk_key);
6567 sb_array_offset += len;
6570 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6571 chunk = (struct btrfs_chunk *)sb_array_offset;
6573 * At least one btrfs_chunk with one stripe must be
6574 * present, exact stripe count check comes afterwards
6576 len = btrfs_chunk_item_size(1);
6577 if (cur_offset + len > array_size)
6578 goto out_short_read;
6580 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6583 "BTRFS: invalid number of stripes %u in sys_array at offset %u\n",
6584 num_stripes, cur_offset);
6589 len = btrfs_chunk_item_size(num_stripes);
6590 if (cur_offset + len > array_size)
6591 goto out_short_read;
6593 ret = read_one_chunk(root, &key, sb, chunk);
6598 "BTRFS: unexpected item type %u in sys_array at offset %u\n",
6599 (u32)key.type, cur_offset);
6604 sb_array_offset += len;
6607 free_extent_buffer(sb);
6611 printk(KERN_ERR "BTRFS: sys_array too short to read %u bytes at offset %u\n",
6613 free_extent_buffer(sb);
6617 int btrfs_read_chunk_tree(struct btrfs_root *root)
6619 struct btrfs_path *path;
6620 struct extent_buffer *leaf;
6621 struct btrfs_key key;
6622 struct btrfs_key found_key;
6626 root = root->fs_info->chunk_root;
6628 path = btrfs_alloc_path();
6632 mutex_lock(&uuid_mutex);
6636 * Read all device items, and then all the chunk items. All
6637 * device items are found before any chunk item (their object id
6638 * is smaller than the lowest possible object id for a chunk
6639 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6641 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6644 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6648 leaf = path->nodes[0];
6649 slot = path->slots[0];
6650 if (slot >= btrfs_header_nritems(leaf)) {
6651 ret = btrfs_next_leaf(root, path);
6658 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6659 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6660 struct btrfs_dev_item *dev_item;
6661 dev_item = btrfs_item_ptr(leaf, slot,
6662 struct btrfs_dev_item);
6663 ret = read_one_dev(root, leaf, dev_item);
6666 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6667 struct btrfs_chunk *chunk;
6668 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6669 ret = read_one_chunk(root, &found_key, leaf, chunk);
6677 unlock_chunks(root);
6678 mutex_unlock(&uuid_mutex);
6680 btrfs_free_path(path);
6684 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6686 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6687 struct btrfs_device *device;
6689 while (fs_devices) {
6690 mutex_lock(&fs_devices->device_list_mutex);
6691 list_for_each_entry(device, &fs_devices->devices, dev_list)
6692 device->dev_root = fs_info->dev_root;
6693 mutex_unlock(&fs_devices->device_list_mutex);
6695 fs_devices = fs_devices->seed;
6699 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6703 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6704 btrfs_dev_stat_reset(dev, i);
6707 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6709 struct btrfs_key key;
6710 struct btrfs_key found_key;
6711 struct btrfs_root *dev_root = fs_info->dev_root;
6712 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6713 struct extent_buffer *eb;
6716 struct btrfs_device *device;
6717 struct btrfs_path *path = NULL;
6720 path = btrfs_alloc_path();
6726 mutex_lock(&fs_devices->device_list_mutex);
6727 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6729 struct btrfs_dev_stats_item *ptr;
6731 key.objectid = BTRFS_DEV_STATS_OBJECTID;
6732 key.type = BTRFS_PERSISTENT_ITEM_KEY;
6733 key.offset = device->devid;
6734 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6736 __btrfs_reset_dev_stats(device);
6737 device->dev_stats_valid = 1;
6738 btrfs_release_path(path);
6741 slot = path->slots[0];
6742 eb = path->nodes[0];
6743 btrfs_item_key_to_cpu(eb, &found_key, slot);
6744 item_size = btrfs_item_size_nr(eb, slot);
6746 ptr = btrfs_item_ptr(eb, slot,
6747 struct btrfs_dev_stats_item);
6749 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6750 if (item_size >= (1 + i) * sizeof(__le64))
6751 btrfs_dev_stat_set(device, i,
6752 btrfs_dev_stats_value(eb, ptr, i));
6754 btrfs_dev_stat_reset(device, i);
6757 device->dev_stats_valid = 1;
6758 btrfs_dev_stat_print_on_load(device);
6759 btrfs_release_path(path);
6761 mutex_unlock(&fs_devices->device_list_mutex);
6764 btrfs_free_path(path);
6765 return ret < 0 ? ret : 0;
6768 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
6769 struct btrfs_root *dev_root,
6770 struct btrfs_device *device)
6772 struct btrfs_path *path;
6773 struct btrfs_key key;
6774 struct extent_buffer *eb;
6775 struct btrfs_dev_stats_item *ptr;
6779 key.objectid = BTRFS_DEV_STATS_OBJECTID;
6780 key.type = BTRFS_PERSISTENT_ITEM_KEY;
6781 key.offset = device->devid;
6783 path = btrfs_alloc_path();
6785 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
6787 btrfs_warn_in_rcu(dev_root->fs_info,
6788 "error %d while searching for dev_stats item for device %s",
6789 ret, rcu_str_deref(device->name));
6794 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
6795 /* need to delete old one and insert a new one */
6796 ret = btrfs_del_item(trans, dev_root, path);
6798 btrfs_warn_in_rcu(dev_root->fs_info,
6799 "delete too small dev_stats item for device %s failed %d",
6800 rcu_str_deref(device->name), ret);
6807 /* need to insert a new item */
6808 btrfs_release_path(path);
6809 ret = btrfs_insert_empty_item(trans, dev_root, path,
6810 &key, sizeof(*ptr));
6812 btrfs_warn_in_rcu(dev_root->fs_info,
6813 "insert dev_stats item for device %s failed %d",
6814 rcu_str_deref(device->name), ret);
6819 eb = path->nodes[0];
6820 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
6821 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6822 btrfs_set_dev_stats_value(eb, ptr, i,
6823 btrfs_dev_stat_read(device, i));
6824 btrfs_mark_buffer_dirty(eb);
6827 btrfs_free_path(path);
6832 * called from commit_transaction. Writes all changed device stats to disk.
6834 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
6835 struct btrfs_fs_info *fs_info)
6837 struct btrfs_root *dev_root = fs_info->dev_root;
6838 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6839 struct btrfs_device *device;
6843 mutex_lock(&fs_devices->device_list_mutex);
6844 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6845 if (!device->dev_stats_valid || !btrfs_dev_stats_dirty(device))
6848 stats_cnt = atomic_read(&device->dev_stats_ccnt);
6849 ret = update_dev_stat_item(trans, dev_root, device);
6851 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
6853 mutex_unlock(&fs_devices->device_list_mutex);
6858 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
6860 btrfs_dev_stat_inc(dev, index);
6861 btrfs_dev_stat_print_on_error(dev);
6864 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
6866 if (!dev->dev_stats_valid)
6868 btrfs_err_rl_in_rcu(dev->dev_root->fs_info,
6869 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
6870 rcu_str_deref(dev->name),
6871 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6872 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6873 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6874 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6875 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6878 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
6882 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6883 if (btrfs_dev_stat_read(dev, i) != 0)
6885 if (i == BTRFS_DEV_STAT_VALUES_MAX)
6886 return; /* all values == 0, suppress message */
6888 btrfs_info_in_rcu(dev->dev_root->fs_info,
6889 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
6890 rcu_str_deref(dev->name),
6891 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6892 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6893 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6894 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6895 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6898 int btrfs_get_dev_stats(struct btrfs_root *root,
6899 struct btrfs_ioctl_get_dev_stats *stats)
6901 struct btrfs_device *dev;
6902 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6905 mutex_lock(&fs_devices->device_list_mutex);
6906 dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
6907 mutex_unlock(&fs_devices->device_list_mutex);
6910 btrfs_warn(root->fs_info, "get dev_stats failed, device not found");
6912 } else if (!dev->dev_stats_valid) {
6913 btrfs_warn(root->fs_info, "get dev_stats failed, not yet valid");
6915 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
6916 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6917 if (stats->nr_items > i)
6919 btrfs_dev_stat_read_and_reset(dev, i);
6921 btrfs_dev_stat_reset(dev, i);
6924 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6925 if (stats->nr_items > i)
6926 stats->values[i] = btrfs_dev_stat_read(dev, i);
6928 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
6929 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
6933 void btrfs_scratch_superblocks(struct block_device *bdev, char *device_path)
6935 struct buffer_head *bh;
6936 struct btrfs_super_block *disk_super;
6942 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX;
6945 if (btrfs_read_dev_one_super(bdev, copy_num, &bh))
6948 disk_super = (struct btrfs_super_block *)bh->b_data;
6950 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
6951 set_buffer_dirty(bh);
6952 sync_dirty_buffer(bh);
6956 /* Notify udev that device has changed */
6957 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
6959 /* Update ctime/mtime for device path for libblkid */
6960 update_dev_time(device_path);
6964 * Update the size of all devices, which is used for writing out the
6967 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
6969 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6970 struct btrfs_device *curr, *next;
6972 if (list_empty(&fs_devices->resized_devices))
6975 mutex_lock(&fs_devices->device_list_mutex);
6976 lock_chunks(fs_info->dev_root);
6977 list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
6979 list_del_init(&curr->resized_list);
6980 curr->commit_total_bytes = curr->disk_total_bytes;
6982 unlock_chunks(fs_info->dev_root);
6983 mutex_unlock(&fs_devices->device_list_mutex);
6986 /* Must be invoked during the transaction commit */
6987 void btrfs_update_commit_device_bytes_used(struct btrfs_root *root,
6988 struct btrfs_transaction *transaction)
6990 struct extent_map *em;
6991 struct map_lookup *map;
6992 struct btrfs_device *dev;
6995 if (list_empty(&transaction->pending_chunks))
6998 /* In order to kick the device replace finish process */
7000 list_for_each_entry(em, &transaction->pending_chunks, list) {
7001 map = em->map_lookup;
7003 for (i = 0; i < map->num_stripes; i++) {
7004 dev = map->stripes[i].dev;
7005 dev->commit_bytes_used = dev->bytes_used;
7008 unlock_chunks(root);
7011 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
7013 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7014 while (fs_devices) {
7015 fs_devices->fs_info = fs_info;
7016 fs_devices = fs_devices->seed;
7020 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
7022 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7023 while (fs_devices) {
7024 fs_devices->fs_info = NULL;
7025 fs_devices = fs_devices->seed;
7029 static void btrfs_close_one_device(struct btrfs_device *device)
7031 struct btrfs_fs_devices *fs_devices = device->fs_devices;
7032 struct btrfs_device *new_device;
7033 struct rcu_string *name;
7036 fs_devices->open_devices--;
7038 if (device->writeable &&
7039 device->devid != BTRFS_DEV_REPLACE_DEVID) {
7040 list_del_init(&device->dev_alloc_list);
7041 fs_devices->rw_devices--;
7044 if (device->missing)
7045 fs_devices->missing_devices--;
7047 new_device = btrfs_alloc_device(NULL, &device->devid,
7049 BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
7051 /* Safe because we are under uuid_mutex */
7053 name = rcu_string_strdup(device->name->str, GFP_NOFS);
7054 BUG_ON(!name); /* -ENOMEM */
7055 rcu_assign_pointer(new_device->name, name);
7058 list_replace_rcu(&device->dev_list, &new_device->dev_list);
7059 new_device->fs_devices = device->fs_devices;
7061 call_rcu(&device->rcu, free_device);