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/iocontext.h>
24 #include <linux/capability.h>
25 #include <linux/ratelimit.h>
26 #include <linux/kthread.h>
27 #include <linux/raid/pq.h>
28 #include <linux/semaphore.h>
29 #include <linux/uuid.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,
122 * Table to convert BTRFS_RAID_* to the error code if minimum number of devices
123 * condition is not met. Zero means there's no corresponding
124 * BTRFS_ERROR_DEV_*_NOT_MET value.
126 const int btrfs_raid_mindev_error[BTRFS_NR_RAID_TYPES] = {
127 [BTRFS_RAID_RAID10] = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET,
128 [BTRFS_RAID_RAID1] = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET,
129 [BTRFS_RAID_DUP] = 0,
130 [BTRFS_RAID_RAID0] = 0,
131 [BTRFS_RAID_SINGLE] = 0,
132 [BTRFS_RAID_RAID5] = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET,
133 [BTRFS_RAID_RAID6] = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET,
136 static int init_first_rw_device(struct btrfs_trans_handle *trans,
137 struct btrfs_root *root,
138 struct btrfs_device *device);
139 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
140 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
141 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
142 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
143 static void btrfs_close_one_device(struct btrfs_device *device);
145 DEFINE_MUTEX(uuid_mutex);
146 static LIST_HEAD(fs_uuids);
147 struct list_head *btrfs_get_fs_uuids(void)
152 static struct btrfs_fs_devices *__alloc_fs_devices(void)
154 struct btrfs_fs_devices *fs_devs;
156 fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL);
158 return ERR_PTR(-ENOMEM);
160 mutex_init(&fs_devs->device_list_mutex);
162 INIT_LIST_HEAD(&fs_devs->devices);
163 INIT_LIST_HEAD(&fs_devs->resized_devices);
164 INIT_LIST_HEAD(&fs_devs->alloc_list);
165 INIT_LIST_HEAD(&fs_devs->list);
171 * alloc_fs_devices - allocate struct btrfs_fs_devices
172 * @fsid: a pointer to UUID for this FS. If NULL a new UUID is
175 * Return: a pointer to a new &struct btrfs_fs_devices on success;
176 * ERR_PTR() on error. Returned struct is not linked onto any lists and
177 * can be destroyed with kfree() right away.
179 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
181 struct btrfs_fs_devices *fs_devs;
183 fs_devs = __alloc_fs_devices();
188 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
190 generate_random_uuid(fs_devs->fsid);
195 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
197 struct btrfs_device *device;
198 WARN_ON(fs_devices->opened);
199 while (!list_empty(&fs_devices->devices)) {
200 device = list_entry(fs_devices->devices.next,
201 struct btrfs_device, dev_list);
202 list_del(&device->dev_list);
203 rcu_string_free(device->name);
209 static void btrfs_kobject_uevent(struct block_device *bdev,
210 enum kobject_action action)
214 ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
216 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
218 kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
219 &disk_to_dev(bdev->bd_disk)->kobj);
222 void btrfs_cleanup_fs_uuids(void)
224 struct btrfs_fs_devices *fs_devices;
226 while (!list_empty(&fs_uuids)) {
227 fs_devices = list_entry(fs_uuids.next,
228 struct btrfs_fs_devices, list);
229 list_del(&fs_devices->list);
230 free_fs_devices(fs_devices);
234 static struct btrfs_device *__alloc_device(void)
236 struct btrfs_device *dev;
238 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
240 return ERR_PTR(-ENOMEM);
242 INIT_LIST_HEAD(&dev->dev_list);
243 INIT_LIST_HEAD(&dev->dev_alloc_list);
244 INIT_LIST_HEAD(&dev->resized_list);
246 spin_lock_init(&dev->io_lock);
248 spin_lock_init(&dev->reada_lock);
249 atomic_set(&dev->reada_in_flight, 0);
250 atomic_set(&dev->dev_stats_ccnt, 0);
251 btrfs_device_data_ordered_init(dev);
252 INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
253 INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
258 static noinline struct btrfs_device *__find_device(struct list_head *head,
261 struct btrfs_device *dev;
263 list_for_each_entry(dev, head, dev_list) {
264 if (dev->devid == devid &&
265 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
272 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
274 struct btrfs_fs_devices *fs_devices;
276 list_for_each_entry(fs_devices, &fs_uuids, list) {
277 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
284 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
285 int flush, struct block_device **bdev,
286 struct buffer_head **bh)
290 *bdev = blkdev_get_by_path(device_path, flags, holder);
293 ret = PTR_ERR(*bdev);
298 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
299 ret = set_blocksize(*bdev, 4096);
301 blkdev_put(*bdev, flags);
304 invalidate_bdev(*bdev);
305 *bh = btrfs_read_dev_super(*bdev);
308 blkdev_put(*bdev, flags);
320 static void requeue_list(struct btrfs_pending_bios *pending_bios,
321 struct bio *head, struct bio *tail)
324 struct bio *old_head;
326 old_head = pending_bios->head;
327 pending_bios->head = head;
328 if (pending_bios->tail)
329 tail->bi_next = old_head;
331 pending_bios->tail = tail;
335 * we try to collect pending bios for a device so we don't get a large
336 * number of procs sending bios down to the same device. This greatly
337 * improves the schedulers ability to collect and merge the bios.
339 * But, it also turns into a long list of bios to process and that is sure
340 * to eventually make the worker thread block. The solution here is to
341 * make some progress and then put this work struct back at the end of
342 * the list if the block device is congested. This way, multiple devices
343 * can make progress from a single worker thread.
345 static noinline void run_scheduled_bios(struct btrfs_device *device)
348 struct backing_dev_info *bdi;
349 struct btrfs_fs_info *fs_info;
350 struct btrfs_pending_bios *pending_bios;
354 unsigned long num_run;
355 unsigned long batch_run = 0;
357 unsigned long last_waited = 0;
359 int sync_pending = 0;
360 struct blk_plug plug;
363 * this function runs all the bios we've collected for
364 * a particular device. We don't want to wander off to
365 * another device without first sending all of these down.
366 * So, setup a plug here and finish it off before we return
368 blk_start_plug(&plug);
370 bdi = blk_get_backing_dev_info(device->bdev);
371 fs_info = device->dev_root->fs_info;
372 limit = btrfs_async_submit_limit(fs_info);
373 limit = limit * 2 / 3;
376 spin_lock(&device->io_lock);
381 /* take all the bios off the list at once and process them
382 * later on (without the lock held). But, remember the
383 * tail and other pointers so the bios can be properly reinserted
384 * into the list if we hit congestion
386 if (!force_reg && device->pending_sync_bios.head) {
387 pending_bios = &device->pending_sync_bios;
390 pending_bios = &device->pending_bios;
394 pending = pending_bios->head;
395 tail = pending_bios->tail;
396 WARN_ON(pending && !tail);
399 * if pending was null this time around, no bios need processing
400 * at all and we can stop. Otherwise it'll loop back up again
401 * and do an additional check so no bios are missed.
403 * device->running_pending is used to synchronize with the
406 if (device->pending_sync_bios.head == NULL &&
407 device->pending_bios.head == NULL) {
409 device->running_pending = 0;
412 device->running_pending = 1;
415 pending_bios->head = NULL;
416 pending_bios->tail = NULL;
418 spin_unlock(&device->io_lock);
423 /* we want to work on both lists, but do more bios on the
424 * sync list than the regular list
427 pending_bios != &device->pending_sync_bios &&
428 device->pending_sync_bios.head) ||
429 (num_run > 64 && pending_bios == &device->pending_sync_bios &&
430 device->pending_bios.head)) {
431 spin_lock(&device->io_lock);
432 requeue_list(pending_bios, pending, tail);
437 pending = pending->bi_next;
441 * atomic_dec_return implies a barrier for waitqueue_active
443 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
444 waitqueue_active(&fs_info->async_submit_wait))
445 wake_up(&fs_info->async_submit_wait);
447 BUG_ON(atomic_read(&cur->__bi_cnt) == 0);
450 * if we're doing the sync list, record that our
451 * plug has some sync requests on it
453 * If we're doing the regular list and there are
454 * sync requests sitting around, unplug before
457 if (pending_bios == &device->pending_sync_bios) {
459 } else if (sync_pending) {
460 blk_finish_plug(&plug);
461 blk_start_plug(&plug);
465 btrfsic_submit_bio(cur->bi_rw, cur);
472 * we made progress, there is more work to do and the bdi
473 * is now congested. Back off and let other work structs
476 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
477 fs_info->fs_devices->open_devices > 1) {
478 struct io_context *ioc;
480 ioc = current->io_context;
483 * the main goal here is that we don't want to
484 * block if we're going to be able to submit
485 * more requests without blocking.
487 * This code does two great things, it pokes into
488 * the elevator code from a filesystem _and_
489 * it makes assumptions about how batching works.
491 if (ioc && ioc->nr_batch_requests > 0 &&
492 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
494 ioc->last_waited == last_waited)) {
496 * we want to go through our batch of
497 * requests and stop. So, we copy out
498 * the ioc->last_waited time and test
499 * against it before looping
501 last_waited = ioc->last_waited;
505 spin_lock(&device->io_lock);
506 requeue_list(pending_bios, pending, tail);
507 device->running_pending = 1;
509 spin_unlock(&device->io_lock);
510 btrfs_queue_work(fs_info->submit_workers,
514 /* unplug every 64 requests just for good measure */
515 if (batch_run % 64 == 0) {
516 blk_finish_plug(&plug);
517 blk_start_plug(&plug);
526 spin_lock(&device->io_lock);
527 if (device->pending_bios.head || device->pending_sync_bios.head)
529 spin_unlock(&device->io_lock);
532 blk_finish_plug(&plug);
535 static void pending_bios_fn(struct btrfs_work *work)
537 struct btrfs_device *device;
539 device = container_of(work, struct btrfs_device, work);
540 run_scheduled_bios(device);
544 void btrfs_free_stale_device(struct btrfs_device *cur_dev)
546 struct btrfs_fs_devices *fs_devs;
547 struct btrfs_device *dev;
552 list_for_each_entry(fs_devs, &fs_uuids, list) {
557 if (fs_devs->seeding)
560 list_for_each_entry(dev, &fs_devs->devices, dev_list) {
568 * Todo: This won't be enough. What if the same device
569 * comes back (with new uuid and) with its mapper path?
570 * But for now, this does help as mostly an admin will
571 * either use mapper or non mapper path throughout.
574 del = strcmp(rcu_str_deref(dev->name),
575 rcu_str_deref(cur_dev->name));
582 /* delete the stale device */
583 if (fs_devs->num_devices == 1) {
584 btrfs_sysfs_remove_fsid(fs_devs);
585 list_del(&fs_devs->list);
586 free_fs_devices(fs_devs);
588 fs_devs->num_devices--;
589 list_del(&dev->dev_list);
590 rcu_string_free(dev->name);
599 * Add new device to list of registered devices
602 * 1 - first time device is seen
603 * 0 - device already known
606 static noinline int device_list_add(const char *path,
607 struct btrfs_super_block *disk_super,
608 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
610 struct btrfs_device *device;
611 struct btrfs_fs_devices *fs_devices;
612 struct rcu_string *name;
614 u64 found_transid = btrfs_super_generation(disk_super);
616 fs_devices = find_fsid(disk_super->fsid);
618 fs_devices = alloc_fs_devices(disk_super->fsid);
619 if (IS_ERR(fs_devices))
620 return PTR_ERR(fs_devices);
622 list_add(&fs_devices->list, &fs_uuids);
626 device = __find_device(&fs_devices->devices, devid,
627 disk_super->dev_item.uuid);
631 if (fs_devices->opened)
634 device = btrfs_alloc_device(NULL, &devid,
635 disk_super->dev_item.uuid);
636 if (IS_ERR(device)) {
637 /* we can safely leave the fs_devices entry around */
638 return PTR_ERR(device);
641 name = rcu_string_strdup(path, GFP_NOFS);
646 rcu_assign_pointer(device->name, name);
648 mutex_lock(&fs_devices->device_list_mutex);
649 list_add_rcu(&device->dev_list, &fs_devices->devices);
650 fs_devices->num_devices++;
651 mutex_unlock(&fs_devices->device_list_mutex);
654 device->fs_devices = fs_devices;
655 } else if (!device->name || strcmp(device->name->str, path)) {
657 * When FS is already mounted.
658 * 1. If you are here and if the device->name is NULL that
659 * means this device was missing at time of FS mount.
660 * 2. If you are here and if the device->name is different
661 * from 'path' that means either
662 * a. The same device disappeared and reappeared with
664 * b. The missing-disk-which-was-replaced, has
667 * We must allow 1 and 2a above. But 2b would be a spurious
670 * Further in case of 1 and 2a above, the disk at 'path'
671 * would have missed some transaction when it was away and
672 * in case of 2a the stale bdev has to be updated as well.
673 * 2b must not be allowed at all time.
677 * For now, we do allow update to btrfs_fs_device through the
678 * btrfs dev scan cli after FS has been mounted. We're still
679 * tracking a problem where systems fail mount by subvolume id
680 * when we reject replacement on a mounted FS.
682 if (!fs_devices->opened && found_transid < device->generation) {
684 * That is if the FS is _not_ mounted and if you
685 * are here, that means there is more than one
686 * disk with same uuid and devid.We keep the one
687 * with larger generation number or the last-in if
688 * generation are equal.
693 name = rcu_string_strdup(path, GFP_NOFS);
696 rcu_string_free(device->name);
697 rcu_assign_pointer(device->name, name);
698 if (device->missing) {
699 fs_devices->missing_devices--;
705 * Unmount does not free the btrfs_device struct but would zero
706 * generation along with most of the other members. So just update
707 * it back. We need it to pick the disk with largest generation
710 if (!fs_devices->opened)
711 device->generation = found_transid;
714 * if there is new btrfs on an already registered device,
715 * then remove the stale device entry.
718 btrfs_free_stale_device(device);
720 *fs_devices_ret = fs_devices;
725 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
727 struct btrfs_fs_devices *fs_devices;
728 struct btrfs_device *device;
729 struct btrfs_device *orig_dev;
731 fs_devices = alloc_fs_devices(orig->fsid);
732 if (IS_ERR(fs_devices))
735 mutex_lock(&orig->device_list_mutex);
736 fs_devices->total_devices = orig->total_devices;
738 /* We have held the volume lock, it is safe to get the devices. */
739 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
740 struct rcu_string *name;
742 device = btrfs_alloc_device(NULL, &orig_dev->devid,
748 * This is ok to do without rcu read locked because we hold the
749 * uuid mutex so nothing we touch in here is going to disappear.
751 if (orig_dev->name) {
752 name = rcu_string_strdup(orig_dev->name->str,
758 rcu_assign_pointer(device->name, name);
761 list_add(&device->dev_list, &fs_devices->devices);
762 device->fs_devices = fs_devices;
763 fs_devices->num_devices++;
765 mutex_unlock(&orig->device_list_mutex);
768 mutex_unlock(&orig->device_list_mutex);
769 free_fs_devices(fs_devices);
770 return ERR_PTR(-ENOMEM);
773 void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step)
775 struct btrfs_device *device, *next;
776 struct btrfs_device *latest_dev = NULL;
778 mutex_lock(&uuid_mutex);
780 /* This is the initialized path, it is safe to release the devices. */
781 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
782 if (device->in_fs_metadata) {
783 if (!device->is_tgtdev_for_dev_replace &&
785 device->generation > latest_dev->generation)) {
791 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
793 * In the first step, keep the device which has
794 * the correct fsid and the devid that is used
795 * for the dev_replace procedure.
796 * In the second step, the dev_replace state is
797 * read from the device tree and it is known
798 * whether the procedure is really active or
799 * not, which means whether this device is
800 * used or whether it should be removed.
802 if (step == 0 || device->is_tgtdev_for_dev_replace) {
807 blkdev_put(device->bdev, device->mode);
809 fs_devices->open_devices--;
811 if (device->writeable) {
812 list_del_init(&device->dev_alloc_list);
813 device->writeable = 0;
814 if (!device->is_tgtdev_for_dev_replace)
815 fs_devices->rw_devices--;
817 list_del_init(&device->dev_list);
818 fs_devices->num_devices--;
819 rcu_string_free(device->name);
823 if (fs_devices->seed) {
824 fs_devices = fs_devices->seed;
828 fs_devices->latest_bdev = latest_dev->bdev;
830 mutex_unlock(&uuid_mutex);
833 static void __free_device(struct work_struct *work)
835 struct btrfs_device *device;
837 device = container_of(work, struct btrfs_device, rcu_work);
840 blkdev_put(device->bdev, device->mode);
842 rcu_string_free(device->name);
846 static void free_device(struct rcu_head *head)
848 struct btrfs_device *device;
850 device = container_of(head, struct btrfs_device, rcu);
852 INIT_WORK(&device->rcu_work, __free_device);
853 schedule_work(&device->rcu_work);
856 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
858 struct btrfs_device *device, *tmp;
860 if (--fs_devices->opened > 0)
863 mutex_lock(&fs_devices->device_list_mutex);
864 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
865 btrfs_close_one_device(device);
867 mutex_unlock(&fs_devices->device_list_mutex);
869 WARN_ON(fs_devices->open_devices);
870 WARN_ON(fs_devices->rw_devices);
871 fs_devices->opened = 0;
872 fs_devices->seeding = 0;
877 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
879 struct btrfs_fs_devices *seed_devices = NULL;
882 mutex_lock(&uuid_mutex);
883 ret = __btrfs_close_devices(fs_devices);
884 if (!fs_devices->opened) {
885 seed_devices = fs_devices->seed;
886 fs_devices->seed = NULL;
888 mutex_unlock(&uuid_mutex);
890 while (seed_devices) {
891 fs_devices = seed_devices;
892 seed_devices = fs_devices->seed;
893 __btrfs_close_devices(fs_devices);
894 free_fs_devices(fs_devices);
897 * Wait for rcu kworkers under __btrfs_close_devices
898 * to finish all blkdev_puts so device is really
899 * free when umount is done.
905 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
906 fmode_t flags, void *holder)
908 struct request_queue *q;
909 struct block_device *bdev;
910 struct list_head *head = &fs_devices->devices;
911 struct btrfs_device *device;
912 struct btrfs_device *latest_dev = NULL;
913 struct buffer_head *bh;
914 struct btrfs_super_block *disk_super;
921 list_for_each_entry(device, head, dev_list) {
927 /* Just open everything we can; ignore failures here */
928 if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
932 disk_super = (struct btrfs_super_block *)bh->b_data;
933 devid = btrfs_stack_device_id(&disk_super->dev_item);
934 if (devid != device->devid)
937 if (memcmp(device->uuid, disk_super->dev_item.uuid,
941 device->generation = btrfs_super_generation(disk_super);
943 device->generation > latest_dev->generation)
946 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
947 device->writeable = 0;
949 device->writeable = !bdev_read_only(bdev);
953 q = bdev_get_queue(bdev);
954 if (blk_queue_discard(q))
955 device->can_discard = 1;
958 device->in_fs_metadata = 0;
959 device->mode = flags;
961 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
962 fs_devices->rotating = 1;
964 fs_devices->open_devices++;
965 if (device->writeable &&
966 device->devid != BTRFS_DEV_REPLACE_DEVID) {
967 fs_devices->rw_devices++;
968 list_add(&device->dev_alloc_list,
969 &fs_devices->alloc_list);
976 blkdev_put(bdev, flags);
979 if (fs_devices->open_devices == 0) {
983 fs_devices->seeding = seeding;
984 fs_devices->opened = 1;
985 fs_devices->latest_bdev = latest_dev->bdev;
986 fs_devices->total_rw_bytes = 0;
991 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
992 fmode_t flags, void *holder)
996 mutex_lock(&uuid_mutex);
997 if (fs_devices->opened) {
998 fs_devices->opened++;
1001 ret = __btrfs_open_devices(fs_devices, flags, holder);
1003 mutex_unlock(&uuid_mutex);
1007 void btrfs_release_disk_super(struct page *page)
1013 int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr,
1014 struct page **page, struct btrfs_super_block **disk_super)
1019 /* make sure our super fits in the device */
1020 if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1023 /* make sure our super fits in the page */
1024 if (sizeof(**disk_super) > PAGE_SIZE)
1027 /* make sure our super doesn't straddle pages on disk */
1028 index = bytenr >> PAGE_SHIFT;
1029 if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index)
1032 /* pull in the page with our super */
1033 *page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
1036 if (IS_ERR_OR_NULL(*page))
1041 /* align our pointer to the offset of the super block */
1042 *disk_super = p + (bytenr & ~PAGE_MASK);
1044 if (btrfs_super_bytenr(*disk_super) != bytenr ||
1045 btrfs_super_magic(*disk_super) != BTRFS_MAGIC) {
1046 btrfs_release_disk_super(*page);
1050 if ((*disk_super)->label[0] &&
1051 (*disk_super)->label[BTRFS_LABEL_SIZE - 1])
1052 (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0';
1058 * Look for a btrfs signature on a device. This may be called out of the mount path
1059 * and we are not allowed to call set_blocksize during the scan. The superblock
1060 * is read via pagecache
1062 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
1063 struct btrfs_fs_devices **fs_devices_ret)
1065 struct btrfs_super_block *disk_super;
1066 struct block_device *bdev;
1075 * we would like to check all the supers, but that would make
1076 * a btrfs mount succeed after a mkfs from a different FS.
1077 * So, we need to add a special mount option to scan for
1078 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1080 bytenr = btrfs_sb_offset(0);
1081 flags |= FMODE_EXCL;
1082 mutex_lock(&uuid_mutex);
1084 bdev = blkdev_get_by_path(path, flags, holder);
1086 ret = PTR_ERR(bdev);
1090 if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super))
1091 goto error_bdev_put;
1093 devid = btrfs_stack_device_id(&disk_super->dev_item);
1094 transid = btrfs_super_generation(disk_super);
1095 total_devices = btrfs_super_num_devices(disk_super);
1097 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
1099 if (disk_super->label[0]) {
1100 printk(KERN_INFO "BTRFS: device label %s ", disk_super->label);
1102 printk(KERN_INFO "BTRFS: device fsid %pU ", disk_super->fsid);
1105 printk(KERN_CONT "devid %llu transid %llu %s\n", devid, transid, path);
1108 if (!ret && fs_devices_ret)
1109 (*fs_devices_ret)->total_devices = total_devices;
1111 btrfs_release_disk_super(page);
1114 blkdev_put(bdev, flags);
1116 mutex_unlock(&uuid_mutex);
1120 /* helper to account the used device space in the range */
1121 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
1122 u64 end, u64 *length)
1124 struct btrfs_key key;
1125 struct btrfs_root *root = device->dev_root;
1126 struct btrfs_dev_extent *dev_extent;
1127 struct btrfs_path *path;
1131 struct extent_buffer *l;
1135 if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
1138 path = btrfs_alloc_path();
1141 path->reada = READA_FORWARD;
1143 key.objectid = device->devid;
1145 key.type = BTRFS_DEV_EXTENT_KEY;
1147 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1151 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1158 slot = path->slots[0];
1159 if (slot >= btrfs_header_nritems(l)) {
1160 ret = btrfs_next_leaf(root, path);
1168 btrfs_item_key_to_cpu(l, &key, slot);
1170 if (key.objectid < device->devid)
1173 if (key.objectid > device->devid)
1176 if (key.type != BTRFS_DEV_EXTENT_KEY)
1179 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1180 extent_end = key.offset + btrfs_dev_extent_length(l,
1182 if (key.offset <= start && extent_end > end) {
1183 *length = end - start + 1;
1185 } else if (key.offset <= start && extent_end > start)
1186 *length += extent_end - start;
1187 else if (key.offset > start && extent_end <= end)
1188 *length += extent_end - key.offset;
1189 else if (key.offset > start && key.offset <= end) {
1190 *length += end - key.offset + 1;
1192 } else if (key.offset > end)
1200 btrfs_free_path(path);
1204 static int contains_pending_extent(struct btrfs_transaction *transaction,
1205 struct btrfs_device *device,
1206 u64 *start, u64 len)
1208 struct btrfs_fs_info *fs_info = device->dev_root->fs_info;
1209 struct extent_map *em;
1210 struct list_head *search_list = &fs_info->pinned_chunks;
1212 u64 physical_start = *start;
1215 search_list = &transaction->pending_chunks;
1217 list_for_each_entry(em, search_list, list) {
1218 struct map_lookup *map;
1221 map = em->map_lookup;
1222 for (i = 0; i < map->num_stripes; i++) {
1225 if (map->stripes[i].dev != device)
1227 if (map->stripes[i].physical >= physical_start + len ||
1228 map->stripes[i].physical + em->orig_block_len <=
1232 * Make sure that while processing the pinned list we do
1233 * not override our *start with a lower value, because
1234 * we can have pinned chunks that fall within this
1235 * device hole and that have lower physical addresses
1236 * than the pending chunks we processed before. If we
1237 * do not take this special care we can end up getting
1238 * 2 pending chunks that start at the same physical
1239 * device offsets because the end offset of a pinned
1240 * chunk can be equal to the start offset of some
1243 end = map->stripes[i].physical + em->orig_block_len;
1250 if (search_list != &fs_info->pinned_chunks) {
1251 search_list = &fs_info->pinned_chunks;
1260 * find_free_dev_extent_start - find free space in the specified device
1261 * @device: the device which we search the free space in
1262 * @num_bytes: the size of the free space that we need
1263 * @search_start: the position from which to begin the search
1264 * @start: store the start of the free space.
1265 * @len: the size of the free space. that we find, or the size
1266 * of the max free space if we don't find suitable free space
1268 * this uses a pretty simple search, the expectation is that it is
1269 * called very infrequently and that a given device has a small number
1272 * @start is used to store the start of the free space if we find. But if we
1273 * don't find suitable free space, it will be used to store the start position
1274 * of the max free space.
1276 * @len is used to store the size of the free space that we find.
1277 * But if we don't find suitable free space, it is used to store the size of
1278 * the max free space.
1280 int find_free_dev_extent_start(struct btrfs_transaction *transaction,
1281 struct btrfs_device *device, u64 num_bytes,
1282 u64 search_start, u64 *start, u64 *len)
1284 struct btrfs_key key;
1285 struct btrfs_root *root = device->dev_root;
1286 struct btrfs_dev_extent *dev_extent;
1287 struct btrfs_path *path;
1292 u64 search_end = device->total_bytes;
1295 struct extent_buffer *l;
1296 u64 min_search_start;
1299 * We don't want to overwrite the superblock on the drive nor any area
1300 * used by the boot loader (grub for example), so we make sure to start
1301 * at an offset of at least 1MB.
1303 min_search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
1304 search_start = max(search_start, min_search_start);
1306 path = btrfs_alloc_path();
1310 max_hole_start = search_start;
1314 if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1319 path->reada = READA_FORWARD;
1320 path->search_commit_root = 1;
1321 path->skip_locking = 1;
1323 key.objectid = device->devid;
1324 key.offset = search_start;
1325 key.type = BTRFS_DEV_EXTENT_KEY;
1327 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1331 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1338 slot = path->slots[0];
1339 if (slot >= btrfs_header_nritems(l)) {
1340 ret = btrfs_next_leaf(root, path);
1348 btrfs_item_key_to_cpu(l, &key, slot);
1350 if (key.objectid < device->devid)
1353 if (key.objectid > device->devid)
1356 if (key.type != BTRFS_DEV_EXTENT_KEY)
1359 if (key.offset > search_start) {
1360 hole_size = key.offset - search_start;
1363 * Have to check before we set max_hole_start, otherwise
1364 * we could end up sending back this offset anyway.
1366 if (contains_pending_extent(transaction, device,
1369 if (key.offset >= search_start) {
1370 hole_size = key.offset - search_start;
1377 if (hole_size > max_hole_size) {
1378 max_hole_start = search_start;
1379 max_hole_size = hole_size;
1383 * If this free space is greater than which we need,
1384 * it must be the max free space that we have found
1385 * until now, so max_hole_start must point to the start
1386 * of this free space and the length of this free space
1387 * is stored in max_hole_size. Thus, we return
1388 * max_hole_start and max_hole_size and go back to the
1391 if (hole_size >= num_bytes) {
1397 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1398 extent_end = key.offset + btrfs_dev_extent_length(l,
1400 if (extent_end > search_start)
1401 search_start = extent_end;
1408 * At this point, search_start should be the end of
1409 * allocated dev extents, and when shrinking the device,
1410 * search_end may be smaller than search_start.
1412 if (search_end > search_start) {
1413 hole_size = search_end - search_start;
1415 if (contains_pending_extent(transaction, device, &search_start,
1417 btrfs_release_path(path);
1421 if (hole_size > max_hole_size) {
1422 max_hole_start = search_start;
1423 max_hole_size = hole_size;
1428 if (max_hole_size < num_bytes)
1434 btrfs_free_path(path);
1435 *start = max_hole_start;
1437 *len = max_hole_size;
1441 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1442 struct btrfs_device *device, u64 num_bytes,
1443 u64 *start, u64 *len)
1445 /* FIXME use last free of some kind */
1446 return find_free_dev_extent_start(trans->transaction, device,
1447 num_bytes, 0, start, len);
1450 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1451 struct btrfs_device *device,
1452 u64 start, u64 *dev_extent_len)
1455 struct btrfs_path *path;
1456 struct btrfs_root *root = device->dev_root;
1457 struct btrfs_key key;
1458 struct btrfs_key found_key;
1459 struct extent_buffer *leaf = NULL;
1460 struct btrfs_dev_extent *extent = NULL;
1462 path = btrfs_alloc_path();
1466 key.objectid = device->devid;
1468 key.type = BTRFS_DEV_EXTENT_KEY;
1470 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1472 ret = btrfs_previous_item(root, path, key.objectid,
1473 BTRFS_DEV_EXTENT_KEY);
1476 leaf = path->nodes[0];
1477 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1478 extent = btrfs_item_ptr(leaf, path->slots[0],
1479 struct btrfs_dev_extent);
1480 BUG_ON(found_key.offset > start || found_key.offset +
1481 btrfs_dev_extent_length(leaf, extent) < start);
1483 btrfs_release_path(path);
1485 } else if (ret == 0) {
1486 leaf = path->nodes[0];
1487 extent = btrfs_item_ptr(leaf, path->slots[0],
1488 struct btrfs_dev_extent);
1490 btrfs_handle_fs_error(root->fs_info, ret, "Slot search failed");
1494 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1496 ret = btrfs_del_item(trans, root, path);
1498 btrfs_handle_fs_error(root->fs_info, ret,
1499 "Failed to remove dev extent item");
1501 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1504 btrfs_free_path(path);
1508 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1509 struct btrfs_device *device,
1510 u64 chunk_tree, u64 chunk_objectid,
1511 u64 chunk_offset, u64 start, u64 num_bytes)
1514 struct btrfs_path *path;
1515 struct btrfs_root *root = device->dev_root;
1516 struct btrfs_dev_extent *extent;
1517 struct extent_buffer *leaf;
1518 struct btrfs_key key;
1520 WARN_ON(!device->in_fs_metadata);
1521 WARN_ON(device->is_tgtdev_for_dev_replace);
1522 path = btrfs_alloc_path();
1526 key.objectid = device->devid;
1528 key.type = BTRFS_DEV_EXTENT_KEY;
1529 ret = btrfs_insert_empty_item(trans, root, path, &key,
1534 leaf = path->nodes[0];
1535 extent = btrfs_item_ptr(leaf, path->slots[0],
1536 struct btrfs_dev_extent);
1537 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1538 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1539 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1541 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1542 btrfs_dev_extent_chunk_tree_uuid(extent), BTRFS_UUID_SIZE);
1544 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1545 btrfs_mark_buffer_dirty(leaf);
1547 btrfs_free_path(path);
1551 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1553 struct extent_map_tree *em_tree;
1554 struct extent_map *em;
1558 em_tree = &fs_info->mapping_tree.map_tree;
1559 read_lock(&em_tree->lock);
1560 n = rb_last(&em_tree->map);
1562 em = rb_entry(n, struct extent_map, rb_node);
1563 ret = em->start + em->len;
1565 read_unlock(&em_tree->lock);
1570 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1574 struct btrfs_key key;
1575 struct btrfs_key found_key;
1576 struct btrfs_path *path;
1578 path = btrfs_alloc_path();
1582 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1583 key.type = BTRFS_DEV_ITEM_KEY;
1584 key.offset = (u64)-1;
1586 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1590 BUG_ON(ret == 0); /* Corruption */
1592 ret = btrfs_previous_item(fs_info->chunk_root, path,
1593 BTRFS_DEV_ITEMS_OBJECTID,
1594 BTRFS_DEV_ITEM_KEY);
1598 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1600 *devid_ret = found_key.offset + 1;
1604 btrfs_free_path(path);
1609 * the device information is stored in the chunk root
1610 * the btrfs_device struct should be fully filled in
1612 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1613 struct btrfs_root *root,
1614 struct btrfs_device *device)
1617 struct btrfs_path *path;
1618 struct btrfs_dev_item *dev_item;
1619 struct extent_buffer *leaf;
1620 struct btrfs_key key;
1623 root = root->fs_info->chunk_root;
1625 path = btrfs_alloc_path();
1629 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1630 key.type = BTRFS_DEV_ITEM_KEY;
1631 key.offset = device->devid;
1633 ret = btrfs_insert_empty_item(trans, root, path, &key,
1638 leaf = path->nodes[0];
1639 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1641 btrfs_set_device_id(leaf, dev_item, device->devid);
1642 btrfs_set_device_generation(leaf, dev_item, 0);
1643 btrfs_set_device_type(leaf, dev_item, device->type);
1644 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1645 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1646 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1647 btrfs_set_device_total_bytes(leaf, dev_item,
1648 btrfs_device_get_disk_total_bytes(device));
1649 btrfs_set_device_bytes_used(leaf, dev_item,
1650 btrfs_device_get_bytes_used(device));
1651 btrfs_set_device_group(leaf, dev_item, 0);
1652 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1653 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1654 btrfs_set_device_start_offset(leaf, dev_item, 0);
1656 ptr = btrfs_device_uuid(dev_item);
1657 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1658 ptr = btrfs_device_fsid(dev_item);
1659 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1660 btrfs_mark_buffer_dirty(leaf);
1664 btrfs_free_path(path);
1669 * Function to update ctime/mtime for a given device path.
1670 * Mainly used for ctime/mtime based probe like libblkid.
1672 static void update_dev_time(char *path_name)
1676 filp = filp_open(path_name, O_RDWR, 0);
1679 file_update_time(filp);
1680 filp_close(filp, NULL);
1683 static int btrfs_rm_dev_item(struct btrfs_root *root,
1684 struct btrfs_device *device)
1687 struct btrfs_path *path;
1688 struct btrfs_key key;
1689 struct btrfs_trans_handle *trans;
1691 root = root->fs_info->chunk_root;
1693 path = btrfs_alloc_path();
1697 trans = btrfs_start_transaction(root, 0);
1698 if (IS_ERR(trans)) {
1699 btrfs_free_path(path);
1700 return PTR_ERR(trans);
1702 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1703 key.type = BTRFS_DEV_ITEM_KEY;
1704 key.offset = device->devid;
1706 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1715 ret = btrfs_del_item(trans, root, path);
1719 btrfs_free_path(path);
1720 btrfs_commit_transaction(trans, root);
1725 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1726 * filesystem. It's up to the caller to adjust that number regarding eg. device
1729 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1737 seq = read_seqbegin(&fs_info->profiles_lock);
1739 all_avail = fs_info->avail_data_alloc_bits |
1740 fs_info->avail_system_alloc_bits |
1741 fs_info->avail_metadata_alloc_bits;
1742 } while (read_seqretry(&fs_info->profiles_lock, seq));
1744 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1745 if (!(all_avail & btrfs_raid_group[i]))
1748 if (num_devices < btrfs_raid_array[i].devs_min) {
1749 int ret = btrfs_raid_mindev_error[i];
1759 struct btrfs_device *btrfs_find_next_active_device(struct btrfs_fs_devices *fs_devs,
1760 struct btrfs_device *device)
1762 struct btrfs_device *next_device;
1764 list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1765 if (next_device != device &&
1766 !next_device->missing && next_device->bdev)
1774 * Helper function to check if the given device is part of s_bdev / latest_bdev
1775 * and replace it with the provided or the next active device, in the context
1776 * where this function called, there should be always be another device (or
1777 * this_dev) which is active.
1779 void btrfs_assign_next_active_device(struct btrfs_fs_info *fs_info,
1780 struct btrfs_device *device, struct btrfs_device *this_dev)
1782 struct btrfs_device *next_device;
1785 next_device = this_dev;
1787 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
1789 ASSERT(next_device);
1791 if (fs_info->sb->s_bdev &&
1792 (fs_info->sb->s_bdev == device->bdev))
1793 fs_info->sb->s_bdev = next_device->bdev;
1795 if (fs_info->fs_devices->latest_bdev == device->bdev)
1796 fs_info->fs_devices->latest_bdev = next_device->bdev;
1799 int btrfs_rm_device(struct btrfs_root *root, char *device_path, u64 devid)
1801 struct btrfs_device *device;
1802 struct btrfs_fs_devices *cur_devices;
1805 bool clear_super = false;
1806 char *dev_name = NULL;
1808 mutex_lock(&uuid_mutex);
1810 num_devices = root->fs_info->fs_devices->num_devices;
1811 btrfs_dev_replace_lock(&root->fs_info->dev_replace, 0);
1812 if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1813 WARN_ON(num_devices < 1);
1816 btrfs_dev_replace_unlock(&root->fs_info->dev_replace, 0);
1818 ret = btrfs_check_raid_min_devices(root->fs_info, num_devices - 1);
1822 ret = btrfs_find_device_by_devspec(root, devid, device_path,
1827 if (device->is_tgtdev_for_dev_replace) {
1828 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1832 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1833 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1837 if (device->writeable) {
1839 list_del_init(&device->dev_alloc_list);
1840 device->fs_devices->rw_devices--;
1841 unlock_chunks(root);
1842 dev_name = kstrdup(device->name->str, GFP_KERNEL);
1850 mutex_unlock(&uuid_mutex);
1851 ret = btrfs_shrink_device(device, 0);
1852 mutex_lock(&uuid_mutex);
1857 * TODO: the superblock still includes this device in its num_devices
1858 * counter although write_all_supers() is not locked out. This
1859 * could give a filesystem state which requires a degraded mount.
1861 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1865 device->in_fs_metadata = 0;
1866 btrfs_scrub_cancel_dev(root->fs_info, device);
1869 * the device list mutex makes sure that we don't change
1870 * the device list while someone else is writing out all
1871 * the device supers. Whoever is writing all supers, should
1872 * lock the device list mutex before getting the number of
1873 * devices in the super block (super_copy). Conversely,
1874 * whoever updates the number of devices in the super block
1875 * (super_copy) should hold the device list mutex.
1878 cur_devices = device->fs_devices;
1879 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1880 list_del_rcu(&device->dev_list);
1882 device->fs_devices->num_devices--;
1883 device->fs_devices->total_devices--;
1885 if (device->missing)
1886 device->fs_devices->missing_devices--;
1888 btrfs_assign_next_active_device(root->fs_info, device, NULL);
1891 device->fs_devices->open_devices--;
1892 /* remove sysfs entry */
1893 btrfs_sysfs_rm_device_link(root->fs_info->fs_devices, device);
1896 call_rcu(&device->rcu, free_device);
1898 num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1899 btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1900 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1902 if (cur_devices->open_devices == 0) {
1903 struct btrfs_fs_devices *fs_devices;
1904 fs_devices = root->fs_info->fs_devices;
1905 while (fs_devices) {
1906 if (fs_devices->seed == cur_devices) {
1907 fs_devices->seed = cur_devices->seed;
1910 fs_devices = fs_devices->seed;
1912 cur_devices->seed = NULL;
1913 __btrfs_close_devices(cur_devices);
1914 free_fs_devices(cur_devices);
1917 root->fs_info->num_tolerated_disk_barrier_failures =
1918 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1921 * at this point, the device is zero sized. We want to
1922 * remove it from the devices list and zero out the old super
1925 struct block_device *bdev;
1927 bdev = blkdev_get_by_path(dev_name, FMODE_READ | FMODE_EXCL,
1928 root->fs_info->bdev_holder);
1929 if (!IS_ERR(bdev)) {
1930 btrfs_scratch_superblocks(bdev, dev_name);
1931 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1938 mutex_unlock(&uuid_mutex);
1942 if (device->writeable) {
1944 list_add(&device->dev_alloc_list,
1945 &root->fs_info->fs_devices->alloc_list);
1946 device->fs_devices->rw_devices++;
1947 unlock_chunks(root);
1952 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
1953 struct btrfs_device *srcdev)
1955 struct btrfs_fs_devices *fs_devices;
1957 WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1960 * in case of fs with no seed, srcdev->fs_devices will point
1961 * to fs_devices of fs_info. However when the dev being replaced is
1962 * a seed dev it will point to the seed's local fs_devices. In short
1963 * srcdev will have its correct fs_devices in both the cases.
1965 fs_devices = srcdev->fs_devices;
1967 list_del_rcu(&srcdev->dev_list);
1968 list_del_rcu(&srcdev->dev_alloc_list);
1969 fs_devices->num_devices--;
1970 if (srcdev->missing)
1971 fs_devices->missing_devices--;
1973 if (srcdev->writeable)
1974 fs_devices->rw_devices--;
1977 fs_devices->open_devices--;
1980 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
1981 struct btrfs_device *srcdev)
1983 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
1985 if (srcdev->writeable) {
1986 /* zero out the old super if it is writable */
1987 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
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 mutex_lock(&uuid_mutex);
2020 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2022 btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev);
2025 fs_info->fs_devices->open_devices--;
2027 fs_info->fs_devices->num_devices--;
2029 btrfs_assign_next_active_device(fs_info, tgtdev, NULL);
2031 list_del_rcu(&tgtdev->dev_list);
2033 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2034 mutex_unlock(&uuid_mutex);
2037 * The update_dev_time() with in btrfs_scratch_superblocks()
2038 * may lead to a call to btrfs_show_devname() which will try
2039 * to hold device_list_mutex. And here this device
2040 * is already out of device list, so we don't have to hold
2041 * the device_list_mutex lock.
2043 btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
2044 call_rcu(&tgtdev->rcu, free_device);
2047 static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
2048 struct btrfs_device **device)
2051 struct btrfs_super_block *disk_super;
2054 struct block_device *bdev;
2055 struct buffer_head *bh;
2058 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2059 root->fs_info->bdev_holder, 0, &bdev, &bh);
2062 disk_super = (struct btrfs_super_block *)bh->b_data;
2063 devid = btrfs_stack_device_id(&disk_super->dev_item);
2064 dev_uuid = disk_super->dev_item.uuid;
2065 *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2070 blkdev_put(bdev, FMODE_READ);
2074 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
2076 struct btrfs_device **device)
2079 if (strcmp(device_path, "missing") == 0) {
2080 struct list_head *devices;
2081 struct btrfs_device *tmp;
2083 devices = &root->fs_info->fs_devices->devices;
2085 * It is safe to read the devices since the volume_mutex
2086 * is held by the caller.
2088 list_for_each_entry(tmp, devices, dev_list) {
2089 if (tmp->in_fs_metadata && !tmp->bdev) {
2096 return BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2100 return btrfs_find_device_by_path(root, device_path, device);
2105 * Lookup a device given by device id, or the path if the id is 0.
2107 int btrfs_find_device_by_devspec(struct btrfs_root *root, u64 devid,
2109 struct btrfs_device **device)
2115 *device = btrfs_find_device(root->fs_info, devid, NULL,
2120 if (!devpath || !devpath[0])
2123 ret = btrfs_find_device_missing_or_by_path(root, devpath,
2130 * does all the dirty work required for changing file system's UUID.
2132 static int btrfs_prepare_sprout(struct btrfs_root *root)
2134 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2135 struct btrfs_fs_devices *old_devices;
2136 struct btrfs_fs_devices *seed_devices;
2137 struct btrfs_super_block *disk_super = root->fs_info->super_copy;
2138 struct btrfs_device *device;
2141 BUG_ON(!mutex_is_locked(&uuid_mutex));
2142 if (!fs_devices->seeding)
2145 seed_devices = __alloc_fs_devices();
2146 if (IS_ERR(seed_devices))
2147 return PTR_ERR(seed_devices);
2149 old_devices = clone_fs_devices(fs_devices);
2150 if (IS_ERR(old_devices)) {
2151 kfree(seed_devices);
2152 return PTR_ERR(old_devices);
2155 list_add(&old_devices->list, &fs_uuids);
2157 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2158 seed_devices->opened = 1;
2159 INIT_LIST_HEAD(&seed_devices->devices);
2160 INIT_LIST_HEAD(&seed_devices->alloc_list);
2161 mutex_init(&seed_devices->device_list_mutex);
2163 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2164 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2166 list_for_each_entry(device, &seed_devices->devices, dev_list)
2167 device->fs_devices = seed_devices;
2170 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2171 unlock_chunks(root);
2173 fs_devices->seeding = 0;
2174 fs_devices->num_devices = 0;
2175 fs_devices->open_devices = 0;
2176 fs_devices->missing_devices = 0;
2177 fs_devices->rotating = 0;
2178 fs_devices->seed = seed_devices;
2180 generate_random_uuid(fs_devices->fsid);
2181 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2182 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2183 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2185 super_flags = btrfs_super_flags(disk_super) &
2186 ~BTRFS_SUPER_FLAG_SEEDING;
2187 btrfs_set_super_flags(disk_super, super_flags);
2193 * Store the expected generation for seed devices in device items.
2195 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2196 struct btrfs_root *root)
2198 struct btrfs_path *path;
2199 struct extent_buffer *leaf;
2200 struct btrfs_dev_item *dev_item;
2201 struct btrfs_device *device;
2202 struct btrfs_key key;
2203 u8 fs_uuid[BTRFS_UUID_SIZE];
2204 u8 dev_uuid[BTRFS_UUID_SIZE];
2208 path = btrfs_alloc_path();
2212 root = root->fs_info->chunk_root;
2213 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2215 key.type = BTRFS_DEV_ITEM_KEY;
2218 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2222 leaf = path->nodes[0];
2224 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2225 ret = btrfs_next_leaf(root, path);
2230 leaf = path->nodes[0];
2231 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2232 btrfs_release_path(path);
2236 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2237 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2238 key.type != BTRFS_DEV_ITEM_KEY)
2241 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2242 struct btrfs_dev_item);
2243 devid = btrfs_device_id(leaf, dev_item);
2244 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2246 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2248 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2250 BUG_ON(!device); /* Logic error */
2252 if (device->fs_devices->seeding) {
2253 btrfs_set_device_generation(leaf, dev_item,
2254 device->generation);
2255 btrfs_mark_buffer_dirty(leaf);
2263 btrfs_free_path(path);
2267 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
2269 struct request_queue *q;
2270 struct btrfs_trans_handle *trans;
2271 struct btrfs_device *device;
2272 struct block_device *bdev;
2273 struct list_head *devices;
2274 struct super_block *sb = root->fs_info->sb;
2275 struct rcu_string *name;
2277 int seeding_dev = 0;
2280 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
2283 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2284 root->fs_info->bdev_holder);
2286 return PTR_ERR(bdev);
2288 if (root->fs_info->fs_devices->seeding) {
2290 down_write(&sb->s_umount);
2291 mutex_lock(&uuid_mutex);
2294 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2296 devices = &root->fs_info->fs_devices->devices;
2298 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2299 list_for_each_entry(device, devices, dev_list) {
2300 if (device->bdev == bdev) {
2303 &root->fs_info->fs_devices->device_list_mutex);
2307 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2309 device = btrfs_alloc_device(root->fs_info, NULL, NULL);
2310 if (IS_ERR(device)) {
2311 /* we can safely leave the fs_devices entry around */
2312 ret = PTR_ERR(device);
2316 name = rcu_string_strdup(device_path, GFP_KERNEL);
2322 rcu_assign_pointer(device->name, name);
2324 trans = btrfs_start_transaction(root, 0);
2325 if (IS_ERR(trans)) {
2326 rcu_string_free(device->name);
2328 ret = PTR_ERR(trans);
2332 q = bdev_get_queue(bdev);
2333 if (blk_queue_discard(q))
2334 device->can_discard = 1;
2335 device->writeable = 1;
2336 device->generation = trans->transid;
2337 device->io_width = root->sectorsize;
2338 device->io_align = root->sectorsize;
2339 device->sector_size = root->sectorsize;
2340 device->total_bytes = i_size_read(bdev->bd_inode);
2341 device->disk_total_bytes = device->total_bytes;
2342 device->commit_total_bytes = device->total_bytes;
2343 device->dev_root = root->fs_info->dev_root;
2344 device->bdev = bdev;
2345 device->in_fs_metadata = 1;
2346 device->is_tgtdev_for_dev_replace = 0;
2347 device->mode = FMODE_EXCL;
2348 device->dev_stats_valid = 1;
2349 set_blocksize(device->bdev, 4096);
2352 sb->s_flags &= ~MS_RDONLY;
2353 ret = btrfs_prepare_sprout(root);
2354 BUG_ON(ret); /* -ENOMEM */
2357 device->fs_devices = root->fs_info->fs_devices;
2359 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2361 list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
2362 list_add(&device->dev_alloc_list,
2363 &root->fs_info->fs_devices->alloc_list);
2364 root->fs_info->fs_devices->num_devices++;
2365 root->fs_info->fs_devices->open_devices++;
2366 root->fs_info->fs_devices->rw_devices++;
2367 root->fs_info->fs_devices->total_devices++;
2368 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2370 spin_lock(&root->fs_info->free_chunk_lock);
2371 root->fs_info->free_chunk_space += device->total_bytes;
2372 spin_unlock(&root->fs_info->free_chunk_lock);
2374 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2375 root->fs_info->fs_devices->rotating = 1;
2377 tmp = btrfs_super_total_bytes(root->fs_info->super_copy);
2378 btrfs_set_super_total_bytes(root->fs_info->super_copy,
2379 tmp + device->total_bytes);
2381 tmp = btrfs_super_num_devices(root->fs_info->super_copy);
2382 btrfs_set_super_num_devices(root->fs_info->super_copy,
2385 /* add sysfs device entry */
2386 btrfs_sysfs_add_device_link(root->fs_info->fs_devices, device);
2389 * we've got more storage, clear any full flags on the space
2392 btrfs_clear_space_info_full(root->fs_info);
2394 unlock_chunks(root);
2395 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2399 ret = init_first_rw_device(trans, root, device);
2400 unlock_chunks(root);
2402 btrfs_abort_transaction(trans, root, ret);
2407 ret = btrfs_add_device(trans, root, device);
2409 btrfs_abort_transaction(trans, root, ret);
2414 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2416 ret = btrfs_finish_sprout(trans, root);
2418 btrfs_abort_transaction(trans, root, ret);
2422 /* Sprouting would change fsid of the mounted root,
2423 * so rename the fsid on the sysfs
2425 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2426 root->fs_info->fsid);
2427 if (kobject_rename(&root->fs_info->fs_devices->fsid_kobj,
2429 btrfs_warn(root->fs_info,
2430 "sysfs: failed to create fsid for sprout");
2433 root->fs_info->num_tolerated_disk_barrier_failures =
2434 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
2435 ret = btrfs_commit_transaction(trans, root);
2438 mutex_unlock(&uuid_mutex);
2439 up_write(&sb->s_umount);
2441 if (ret) /* transaction commit */
2444 ret = btrfs_relocate_sys_chunks(root);
2446 btrfs_handle_fs_error(root->fs_info, ret,
2447 "Failed to relocate sys chunks after "
2448 "device initialization. This can be fixed "
2449 "using the \"btrfs balance\" command.");
2450 trans = btrfs_attach_transaction(root);
2451 if (IS_ERR(trans)) {
2452 if (PTR_ERR(trans) == -ENOENT)
2454 return PTR_ERR(trans);
2456 ret = btrfs_commit_transaction(trans, root);
2459 /* Update ctime/mtime for libblkid */
2460 update_dev_time(device_path);
2464 btrfs_end_transaction(trans, root);
2465 rcu_string_free(device->name);
2466 btrfs_sysfs_rm_device_link(root->fs_info->fs_devices, device);
2469 blkdev_put(bdev, FMODE_EXCL);
2471 mutex_unlock(&uuid_mutex);
2472 up_write(&sb->s_umount);
2477 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2478 struct btrfs_device *srcdev,
2479 struct btrfs_device **device_out)
2481 struct request_queue *q;
2482 struct btrfs_device *device;
2483 struct block_device *bdev;
2484 struct btrfs_fs_info *fs_info = root->fs_info;
2485 struct list_head *devices;
2486 struct rcu_string *name;
2487 u64 devid = BTRFS_DEV_REPLACE_DEVID;
2491 if (fs_info->fs_devices->seeding) {
2492 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
2496 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2497 fs_info->bdev_holder);
2499 btrfs_err(fs_info, "target device %s is invalid!", device_path);
2500 return PTR_ERR(bdev);
2503 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2505 devices = &fs_info->fs_devices->devices;
2506 list_for_each_entry(device, devices, dev_list) {
2507 if (device->bdev == bdev) {
2508 btrfs_err(fs_info, "target device is in the filesystem!");
2515 if (i_size_read(bdev->bd_inode) <
2516 btrfs_device_get_total_bytes(srcdev)) {
2517 btrfs_err(fs_info, "target device is smaller than source device!");
2523 device = btrfs_alloc_device(NULL, &devid, NULL);
2524 if (IS_ERR(device)) {
2525 ret = PTR_ERR(device);
2529 name = rcu_string_strdup(device_path, GFP_NOFS);
2535 rcu_assign_pointer(device->name, name);
2537 q = bdev_get_queue(bdev);
2538 if (blk_queue_discard(q))
2539 device->can_discard = 1;
2540 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2541 device->writeable = 1;
2542 device->generation = 0;
2543 device->io_width = root->sectorsize;
2544 device->io_align = root->sectorsize;
2545 device->sector_size = root->sectorsize;
2546 device->total_bytes = btrfs_device_get_total_bytes(srcdev);
2547 device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
2548 device->bytes_used = btrfs_device_get_bytes_used(srcdev);
2549 ASSERT(list_empty(&srcdev->resized_list));
2550 device->commit_total_bytes = srcdev->commit_total_bytes;
2551 device->commit_bytes_used = device->bytes_used;
2552 device->dev_root = fs_info->dev_root;
2553 device->bdev = bdev;
2554 device->in_fs_metadata = 1;
2555 device->is_tgtdev_for_dev_replace = 1;
2556 device->mode = FMODE_EXCL;
2557 device->dev_stats_valid = 1;
2558 set_blocksize(device->bdev, 4096);
2559 device->fs_devices = fs_info->fs_devices;
2560 list_add(&device->dev_list, &fs_info->fs_devices->devices);
2561 fs_info->fs_devices->num_devices++;
2562 fs_info->fs_devices->open_devices++;
2563 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2565 *device_out = device;
2569 blkdev_put(bdev, FMODE_EXCL);
2573 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2574 struct btrfs_device *tgtdev)
2576 WARN_ON(fs_info->fs_devices->rw_devices == 0);
2577 tgtdev->io_width = fs_info->dev_root->sectorsize;
2578 tgtdev->io_align = fs_info->dev_root->sectorsize;
2579 tgtdev->sector_size = fs_info->dev_root->sectorsize;
2580 tgtdev->dev_root = fs_info->dev_root;
2581 tgtdev->in_fs_metadata = 1;
2584 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2585 struct btrfs_device *device)
2588 struct btrfs_path *path;
2589 struct btrfs_root *root;
2590 struct btrfs_dev_item *dev_item;
2591 struct extent_buffer *leaf;
2592 struct btrfs_key key;
2594 root = device->dev_root->fs_info->chunk_root;
2596 path = btrfs_alloc_path();
2600 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2601 key.type = BTRFS_DEV_ITEM_KEY;
2602 key.offset = device->devid;
2604 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2613 leaf = path->nodes[0];
2614 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2616 btrfs_set_device_id(leaf, dev_item, device->devid);
2617 btrfs_set_device_type(leaf, dev_item, device->type);
2618 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2619 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2620 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2621 btrfs_set_device_total_bytes(leaf, dev_item,
2622 btrfs_device_get_disk_total_bytes(device));
2623 btrfs_set_device_bytes_used(leaf, dev_item,
2624 btrfs_device_get_bytes_used(device));
2625 btrfs_mark_buffer_dirty(leaf);
2628 btrfs_free_path(path);
2632 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2633 struct btrfs_device *device, u64 new_size)
2635 struct btrfs_super_block *super_copy =
2636 device->dev_root->fs_info->super_copy;
2637 struct btrfs_fs_devices *fs_devices;
2641 if (!device->writeable)
2644 lock_chunks(device->dev_root);
2645 old_total = btrfs_super_total_bytes(super_copy);
2646 diff = new_size - device->total_bytes;
2648 if (new_size <= device->total_bytes ||
2649 device->is_tgtdev_for_dev_replace) {
2650 unlock_chunks(device->dev_root);
2654 fs_devices = device->dev_root->fs_info->fs_devices;
2656 btrfs_set_super_total_bytes(super_copy, old_total + diff);
2657 device->fs_devices->total_rw_bytes += diff;
2659 btrfs_device_set_total_bytes(device, new_size);
2660 btrfs_device_set_disk_total_bytes(device, new_size);
2661 btrfs_clear_space_info_full(device->dev_root->fs_info);
2662 if (list_empty(&device->resized_list))
2663 list_add_tail(&device->resized_list,
2664 &fs_devices->resized_devices);
2665 unlock_chunks(device->dev_root);
2667 return btrfs_update_device(trans, device);
2670 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2671 struct btrfs_root *root, u64 chunk_objectid,
2675 struct btrfs_path *path;
2676 struct btrfs_key key;
2678 root = root->fs_info->chunk_root;
2679 path = btrfs_alloc_path();
2683 key.objectid = chunk_objectid;
2684 key.offset = chunk_offset;
2685 key.type = BTRFS_CHUNK_ITEM_KEY;
2687 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2690 else if (ret > 0) { /* Logic error or corruption */
2691 btrfs_handle_fs_error(root->fs_info, -ENOENT,
2692 "Failed lookup while freeing chunk.");
2697 ret = btrfs_del_item(trans, root, path);
2699 btrfs_handle_fs_error(root->fs_info, ret,
2700 "Failed to delete chunk item.");
2702 btrfs_free_path(path);
2706 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2709 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2710 struct btrfs_disk_key *disk_key;
2711 struct btrfs_chunk *chunk;
2718 struct btrfs_key key;
2721 array_size = btrfs_super_sys_array_size(super_copy);
2723 ptr = super_copy->sys_chunk_array;
2726 while (cur < array_size) {
2727 disk_key = (struct btrfs_disk_key *)ptr;
2728 btrfs_disk_key_to_cpu(&key, disk_key);
2730 len = sizeof(*disk_key);
2732 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2733 chunk = (struct btrfs_chunk *)(ptr + len);
2734 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2735 len += btrfs_chunk_item_size(num_stripes);
2740 if (key.objectid == chunk_objectid &&
2741 key.offset == chunk_offset) {
2742 memmove(ptr, ptr + len, array_size - (cur + len));
2744 btrfs_set_super_sys_array_size(super_copy, array_size);
2750 unlock_chunks(root);
2754 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2755 struct btrfs_root *root, u64 chunk_offset)
2757 struct extent_map_tree *em_tree;
2758 struct extent_map *em;
2759 struct btrfs_root *extent_root = root->fs_info->extent_root;
2760 struct map_lookup *map;
2761 u64 dev_extent_len = 0;
2762 u64 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2764 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2767 root = root->fs_info->chunk_root;
2768 em_tree = &root->fs_info->mapping_tree.map_tree;
2770 read_lock(&em_tree->lock);
2771 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2772 read_unlock(&em_tree->lock);
2774 if (!em || em->start > chunk_offset ||
2775 em->start + em->len < chunk_offset) {
2777 * This is a logic error, but we don't want to just rely on the
2778 * user having built with ASSERT enabled, so if ASSERT doesn't
2779 * do anything we still error out.
2783 free_extent_map(em);
2786 map = em->map_lookup;
2787 lock_chunks(root->fs_info->chunk_root);
2788 check_system_chunk(trans, extent_root, map->type);
2789 unlock_chunks(root->fs_info->chunk_root);
2792 * Take the device list mutex to prevent races with the final phase of
2793 * a device replace operation that replaces the device object associated
2794 * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2796 mutex_lock(&fs_devices->device_list_mutex);
2797 for (i = 0; i < map->num_stripes; i++) {
2798 struct btrfs_device *device = map->stripes[i].dev;
2799 ret = btrfs_free_dev_extent(trans, device,
2800 map->stripes[i].physical,
2803 mutex_unlock(&fs_devices->device_list_mutex);
2804 btrfs_abort_transaction(trans, root, ret);
2808 if (device->bytes_used > 0) {
2810 btrfs_device_set_bytes_used(device,
2811 device->bytes_used - dev_extent_len);
2812 spin_lock(&root->fs_info->free_chunk_lock);
2813 root->fs_info->free_chunk_space += dev_extent_len;
2814 spin_unlock(&root->fs_info->free_chunk_lock);
2815 btrfs_clear_space_info_full(root->fs_info);
2816 unlock_chunks(root);
2819 if (map->stripes[i].dev) {
2820 ret = btrfs_update_device(trans, map->stripes[i].dev);
2822 mutex_unlock(&fs_devices->device_list_mutex);
2823 btrfs_abort_transaction(trans, root, ret);
2828 mutex_unlock(&fs_devices->device_list_mutex);
2830 ret = btrfs_free_chunk(trans, root, chunk_objectid, chunk_offset);
2832 btrfs_abort_transaction(trans, root, ret);
2836 trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2838 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2839 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2841 btrfs_abort_transaction(trans, root, ret);
2846 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset, em);
2848 btrfs_abort_transaction(trans, extent_root, ret);
2854 free_extent_map(em);
2858 static int btrfs_relocate_chunk(struct btrfs_root *root, u64 chunk_offset)
2860 struct btrfs_root *extent_root;
2861 struct btrfs_trans_handle *trans;
2864 root = root->fs_info->chunk_root;
2865 extent_root = root->fs_info->extent_root;
2868 * Prevent races with automatic removal of unused block groups.
2869 * After we relocate and before we remove the chunk with offset
2870 * chunk_offset, automatic removal of the block group can kick in,
2871 * resulting in a failure when calling btrfs_remove_chunk() below.
2873 * Make sure to acquire this mutex before doing a tree search (dev
2874 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2875 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2876 * we release the path used to search the chunk/dev tree and before
2877 * the current task acquires this mutex and calls us.
2879 ASSERT(mutex_is_locked(&root->fs_info->delete_unused_bgs_mutex));
2881 ret = btrfs_can_relocate(extent_root, chunk_offset);
2885 /* step one, relocate all the extents inside this chunk */
2886 btrfs_scrub_pause(root);
2887 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2888 btrfs_scrub_continue(root);
2892 trans = btrfs_start_trans_remove_block_group(root->fs_info,
2894 if (IS_ERR(trans)) {
2895 ret = PTR_ERR(trans);
2896 btrfs_handle_fs_error(root->fs_info, ret, NULL);
2901 * step two, delete the device extents and the
2902 * chunk tree entries
2904 ret = btrfs_remove_chunk(trans, root, chunk_offset);
2905 btrfs_end_transaction(trans, root);
2909 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2911 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2912 struct btrfs_path *path;
2913 struct extent_buffer *leaf;
2914 struct btrfs_chunk *chunk;
2915 struct btrfs_key key;
2916 struct btrfs_key found_key;
2918 bool retried = false;
2922 path = btrfs_alloc_path();
2927 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2928 key.offset = (u64)-1;
2929 key.type = BTRFS_CHUNK_ITEM_KEY;
2932 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
2933 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2935 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2938 BUG_ON(ret == 0); /* Corruption */
2940 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2943 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2949 leaf = path->nodes[0];
2950 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2952 chunk = btrfs_item_ptr(leaf, path->slots[0],
2953 struct btrfs_chunk);
2954 chunk_type = btrfs_chunk_type(leaf, chunk);
2955 btrfs_release_path(path);
2957 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2958 ret = btrfs_relocate_chunk(chunk_root,
2965 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2967 if (found_key.offset == 0)
2969 key.offset = found_key.offset - 1;
2972 if (failed && !retried) {
2976 } else if (WARN_ON(failed && retried)) {
2980 btrfs_free_path(path);
2984 static int insert_balance_item(struct btrfs_root *root,
2985 struct btrfs_balance_control *bctl)
2987 struct btrfs_trans_handle *trans;
2988 struct btrfs_balance_item *item;
2989 struct btrfs_disk_balance_args disk_bargs;
2990 struct btrfs_path *path;
2991 struct extent_buffer *leaf;
2992 struct btrfs_key key;
2995 path = btrfs_alloc_path();
2999 trans = btrfs_start_transaction(root, 0);
3000 if (IS_ERR(trans)) {
3001 btrfs_free_path(path);
3002 return PTR_ERR(trans);
3005 key.objectid = BTRFS_BALANCE_OBJECTID;
3006 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3009 ret = btrfs_insert_empty_item(trans, root, path, &key,
3014 leaf = path->nodes[0];
3015 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3017 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
3019 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3020 btrfs_set_balance_data(leaf, item, &disk_bargs);
3021 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3022 btrfs_set_balance_meta(leaf, item, &disk_bargs);
3023 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3024 btrfs_set_balance_sys(leaf, item, &disk_bargs);
3026 btrfs_set_balance_flags(leaf, item, bctl->flags);
3028 btrfs_mark_buffer_dirty(leaf);
3030 btrfs_free_path(path);
3031 err = btrfs_commit_transaction(trans, root);
3037 static int del_balance_item(struct btrfs_root *root)
3039 struct btrfs_trans_handle *trans;
3040 struct btrfs_path *path;
3041 struct btrfs_key key;
3044 path = btrfs_alloc_path();
3048 trans = btrfs_start_transaction(root, 0);
3049 if (IS_ERR(trans)) {
3050 btrfs_free_path(path);
3051 return PTR_ERR(trans);
3054 key.objectid = BTRFS_BALANCE_OBJECTID;
3055 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3058 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3066 ret = btrfs_del_item(trans, root, path);
3068 btrfs_free_path(path);
3069 err = btrfs_commit_transaction(trans, root);
3076 * This is a heuristic used to reduce the number of chunks balanced on
3077 * resume after balance was interrupted.
3079 static void update_balance_args(struct btrfs_balance_control *bctl)
3082 * Turn on soft mode for chunk types that were being converted.
3084 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3085 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3086 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3087 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3088 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3089 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3092 * Turn on usage filter if is not already used. The idea is
3093 * that chunks that we have already balanced should be
3094 * reasonably full. Don't do it for chunks that are being
3095 * converted - that will keep us from relocating unconverted
3096 * (albeit full) chunks.
3098 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3099 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3100 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3101 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3102 bctl->data.usage = 90;
3104 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3105 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3106 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3107 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3108 bctl->sys.usage = 90;
3110 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3111 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3112 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3113 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3114 bctl->meta.usage = 90;
3119 * Should be called with both balance and volume mutexes held to
3120 * serialize other volume operations (add_dev/rm_dev/resize) with
3121 * restriper. Same goes for unset_balance_control.
3123 static void set_balance_control(struct btrfs_balance_control *bctl)
3125 struct btrfs_fs_info *fs_info = bctl->fs_info;
3127 BUG_ON(fs_info->balance_ctl);
3129 spin_lock(&fs_info->balance_lock);
3130 fs_info->balance_ctl = bctl;
3131 spin_unlock(&fs_info->balance_lock);
3134 static void unset_balance_control(struct btrfs_fs_info *fs_info)
3136 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3138 BUG_ON(!fs_info->balance_ctl);
3140 spin_lock(&fs_info->balance_lock);
3141 fs_info->balance_ctl = NULL;
3142 spin_unlock(&fs_info->balance_lock);
3148 * Balance filters. Return 1 if chunk should be filtered out
3149 * (should not be balanced).
3151 static int chunk_profiles_filter(u64 chunk_type,
3152 struct btrfs_balance_args *bargs)
3154 chunk_type = chunk_to_extended(chunk_type) &
3155 BTRFS_EXTENDED_PROFILE_MASK;
3157 if (bargs->profiles & chunk_type)
3163 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3164 struct btrfs_balance_args *bargs)
3166 struct btrfs_block_group_cache *cache;
3168 u64 user_thresh_min;
3169 u64 user_thresh_max;
3172 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3173 chunk_used = btrfs_block_group_used(&cache->item);
3175 if (bargs->usage_min == 0)
3176 user_thresh_min = 0;
3178 user_thresh_min = div_factor_fine(cache->key.offset,
3181 if (bargs->usage_max == 0)
3182 user_thresh_max = 1;
3183 else if (bargs->usage_max > 100)
3184 user_thresh_max = cache->key.offset;
3186 user_thresh_max = div_factor_fine(cache->key.offset,
3189 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3192 btrfs_put_block_group(cache);
3196 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3197 u64 chunk_offset, struct btrfs_balance_args *bargs)
3199 struct btrfs_block_group_cache *cache;
3200 u64 chunk_used, user_thresh;
3203 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3204 chunk_used = btrfs_block_group_used(&cache->item);
3206 if (bargs->usage_min == 0)
3208 else if (bargs->usage > 100)
3209 user_thresh = cache->key.offset;
3211 user_thresh = div_factor_fine(cache->key.offset,
3214 if (chunk_used < user_thresh)
3217 btrfs_put_block_group(cache);
3221 static int chunk_devid_filter(struct extent_buffer *leaf,
3222 struct btrfs_chunk *chunk,
3223 struct btrfs_balance_args *bargs)
3225 struct btrfs_stripe *stripe;
3226 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3229 for (i = 0; i < num_stripes; i++) {
3230 stripe = btrfs_stripe_nr(chunk, i);
3231 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3238 /* [pstart, pend) */
3239 static int chunk_drange_filter(struct extent_buffer *leaf,
3240 struct btrfs_chunk *chunk,
3242 struct btrfs_balance_args *bargs)
3244 struct btrfs_stripe *stripe;
3245 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3251 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3254 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3255 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3256 factor = num_stripes / 2;
3257 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3258 factor = num_stripes - 1;
3259 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3260 factor = num_stripes - 2;
3262 factor = num_stripes;
3265 for (i = 0; i < num_stripes; i++) {
3266 stripe = btrfs_stripe_nr(chunk, i);
3267 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3270 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3271 stripe_length = btrfs_chunk_length(leaf, chunk);
3272 stripe_length = div_u64(stripe_length, factor);
3274 if (stripe_offset < bargs->pend &&
3275 stripe_offset + stripe_length > bargs->pstart)
3282 /* [vstart, vend) */
3283 static int chunk_vrange_filter(struct extent_buffer *leaf,
3284 struct btrfs_chunk *chunk,
3286 struct btrfs_balance_args *bargs)
3288 if (chunk_offset < bargs->vend &&
3289 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3290 /* at least part of the chunk is inside this vrange */
3296 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3297 struct btrfs_chunk *chunk,
3298 struct btrfs_balance_args *bargs)
3300 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3302 if (bargs->stripes_min <= num_stripes
3303 && num_stripes <= bargs->stripes_max)
3309 static int chunk_soft_convert_filter(u64 chunk_type,
3310 struct btrfs_balance_args *bargs)
3312 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3315 chunk_type = chunk_to_extended(chunk_type) &
3316 BTRFS_EXTENDED_PROFILE_MASK;
3318 if (bargs->target == chunk_type)
3324 static int should_balance_chunk(struct btrfs_root *root,
3325 struct extent_buffer *leaf,
3326 struct btrfs_chunk *chunk, u64 chunk_offset)
3328 struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
3329 struct btrfs_balance_args *bargs = NULL;
3330 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3333 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3334 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3338 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3339 bargs = &bctl->data;
3340 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3342 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3343 bargs = &bctl->meta;
3345 /* profiles filter */
3346 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3347 chunk_profiles_filter(chunk_type, bargs)) {
3352 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3353 chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
3355 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3356 chunk_usage_range_filter(bctl->fs_info, chunk_offset, bargs)) {
3361 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3362 chunk_devid_filter(leaf, chunk, bargs)) {
3366 /* drange filter, makes sense only with devid filter */
3367 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3368 chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
3373 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3374 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3378 /* stripes filter */
3379 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3380 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3384 /* soft profile changing mode */
3385 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3386 chunk_soft_convert_filter(chunk_type, bargs)) {
3391 * limited by count, must be the last filter
3393 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3394 if (bargs->limit == 0)
3398 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3400 * Same logic as the 'limit' filter; the minimum cannot be
3401 * determined here because we do not have the global information
3402 * about the count of all chunks that satisfy the filters.
3404 if (bargs->limit_max == 0)
3413 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3415 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3416 struct btrfs_root *chunk_root = fs_info->chunk_root;
3417 struct btrfs_root *dev_root = fs_info->dev_root;
3418 struct list_head *devices;
3419 struct btrfs_device *device;
3423 struct btrfs_chunk *chunk;
3424 struct btrfs_path *path;
3425 struct btrfs_key key;
3426 struct btrfs_key found_key;
3427 struct btrfs_trans_handle *trans;
3428 struct extent_buffer *leaf;
3431 int enospc_errors = 0;
3432 bool counting = true;
3433 /* The single value limit and min/max limits use the same bytes in the */
3434 u64 limit_data = bctl->data.limit;
3435 u64 limit_meta = bctl->meta.limit;
3436 u64 limit_sys = bctl->sys.limit;
3440 int chunk_reserved = 0;
3443 /* step one make some room on all the devices */
3444 devices = &fs_info->fs_devices->devices;
3445 list_for_each_entry(device, devices, dev_list) {
3446 old_size = btrfs_device_get_total_bytes(device);
3447 size_to_free = div_factor(old_size, 1);
3448 size_to_free = min_t(u64, size_to_free, SZ_1M);
3449 if (!device->writeable ||
3450 btrfs_device_get_total_bytes(device) -
3451 btrfs_device_get_bytes_used(device) > size_to_free ||
3452 device->is_tgtdev_for_dev_replace)
3455 ret = btrfs_shrink_device(device, old_size - size_to_free);
3460 trans = btrfs_start_transaction(dev_root, 0);
3461 BUG_ON(IS_ERR(trans));
3463 ret = btrfs_grow_device(trans, device, old_size);
3466 btrfs_end_transaction(trans, dev_root);
3469 /* step two, relocate all the chunks */
3470 path = btrfs_alloc_path();
3476 /* zero out stat counters */
3477 spin_lock(&fs_info->balance_lock);
3478 memset(&bctl->stat, 0, sizeof(bctl->stat));
3479 spin_unlock(&fs_info->balance_lock);
3483 * The single value limit and min/max limits use the same bytes
3486 bctl->data.limit = limit_data;
3487 bctl->meta.limit = limit_meta;
3488 bctl->sys.limit = limit_sys;
3490 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3491 key.offset = (u64)-1;
3492 key.type = BTRFS_CHUNK_ITEM_KEY;
3495 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3496 atomic_read(&fs_info->balance_cancel_req)) {
3501 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3502 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3504 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3509 * this shouldn't happen, it means the last relocate
3513 BUG(); /* FIXME break ? */
3515 ret = btrfs_previous_item(chunk_root, path, 0,
3516 BTRFS_CHUNK_ITEM_KEY);
3518 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3523 leaf = path->nodes[0];
3524 slot = path->slots[0];
3525 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3527 if (found_key.objectid != key.objectid) {
3528 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3532 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3533 chunk_type = btrfs_chunk_type(leaf, chunk);
3536 spin_lock(&fs_info->balance_lock);
3537 bctl->stat.considered++;
3538 spin_unlock(&fs_info->balance_lock);
3541 ret = should_balance_chunk(chunk_root, leaf, chunk,
3544 btrfs_release_path(path);
3546 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3551 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3552 spin_lock(&fs_info->balance_lock);
3553 bctl->stat.expected++;
3554 spin_unlock(&fs_info->balance_lock);
3556 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3558 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3560 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3567 * Apply limit_min filter, no need to check if the LIMITS
3568 * filter is used, limit_min is 0 by default
3570 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3571 count_data < bctl->data.limit_min)
3572 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3573 count_meta < bctl->meta.limit_min)
3574 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3575 count_sys < bctl->sys.limit_min)) {
3576 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3580 ASSERT(fs_info->data_sinfo);
3581 spin_lock(&fs_info->data_sinfo->lock);
3582 bytes_used = fs_info->data_sinfo->bytes_used;
3583 spin_unlock(&fs_info->data_sinfo->lock);
3585 if ((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3586 !chunk_reserved && !bytes_used) {
3587 trans = btrfs_start_transaction(chunk_root, 0);
3588 if (IS_ERR(trans)) {
3589 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3590 ret = PTR_ERR(trans);
3594 ret = btrfs_force_chunk_alloc(trans, chunk_root,
3595 BTRFS_BLOCK_GROUP_DATA);
3596 btrfs_end_transaction(trans, chunk_root);
3598 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3604 ret = btrfs_relocate_chunk(chunk_root,
3606 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3607 if (ret && ret != -ENOSPC)
3609 if (ret == -ENOSPC) {
3612 spin_lock(&fs_info->balance_lock);
3613 bctl->stat.completed++;
3614 spin_unlock(&fs_info->balance_lock);
3617 if (found_key.offset == 0)
3619 key.offset = found_key.offset - 1;
3623 btrfs_release_path(path);
3628 btrfs_free_path(path);
3629 if (enospc_errors) {
3630 btrfs_info(fs_info, "%d enospc errors during balance",
3640 * alloc_profile_is_valid - see if a given profile is valid and reduced
3641 * @flags: profile to validate
3642 * @extended: if true @flags is treated as an extended profile
3644 static int alloc_profile_is_valid(u64 flags, int extended)
3646 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3647 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3649 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3651 /* 1) check that all other bits are zeroed */
3655 /* 2) see if profile is reduced */
3657 return !extended; /* "0" is valid for usual profiles */
3659 /* true if exactly one bit set */
3660 return (flags & (flags - 1)) == 0;
3663 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3665 /* cancel requested || normal exit path */
3666 return atomic_read(&fs_info->balance_cancel_req) ||
3667 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3668 atomic_read(&fs_info->balance_cancel_req) == 0);
3671 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3675 unset_balance_control(fs_info);
3676 ret = del_balance_item(fs_info->tree_root);
3678 btrfs_handle_fs_error(fs_info, ret, NULL);
3680 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3683 /* Non-zero return value signifies invalidity */
3684 static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg,
3687 return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3688 (!alloc_profile_is_valid(bctl_arg->target, 1) ||
3689 (bctl_arg->target & ~allowed)));
3693 * Should be called with both balance and volume mutexes held
3695 int btrfs_balance(struct btrfs_balance_control *bctl,
3696 struct btrfs_ioctl_balance_args *bargs)
3698 struct btrfs_fs_info *fs_info = bctl->fs_info;
3705 if (btrfs_fs_closing(fs_info) ||
3706 atomic_read(&fs_info->balance_pause_req) ||
3707 atomic_read(&fs_info->balance_cancel_req)) {
3712 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3713 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3717 * In case of mixed groups both data and meta should be picked,
3718 * and identical options should be given for both of them.
3720 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3721 if (mixed && (bctl->flags & allowed)) {
3722 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3723 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3724 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3725 btrfs_err(fs_info, "with mixed groups data and "
3726 "metadata balance options must be the same");
3732 num_devices = fs_info->fs_devices->num_devices;
3733 btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
3734 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3735 BUG_ON(num_devices < 1);
3738 btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
3739 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE | BTRFS_BLOCK_GROUP_DUP;
3740 if (num_devices > 1)
3741 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3742 if (num_devices > 2)
3743 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3744 if (num_devices > 3)
3745 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3746 BTRFS_BLOCK_GROUP_RAID6);
3747 if (validate_convert_profile(&bctl->data, allowed)) {
3748 btrfs_err(fs_info, "unable to start balance with target "
3749 "data profile %llu",
3754 if (validate_convert_profile(&bctl->meta, allowed)) {
3756 "unable to start balance with target metadata profile %llu",
3761 if (validate_convert_profile(&bctl->sys, allowed)) {
3763 "unable to start balance with target system profile %llu",
3769 /* allow to reduce meta or sys integrity only if force set */
3770 allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3771 BTRFS_BLOCK_GROUP_RAID10 |
3772 BTRFS_BLOCK_GROUP_RAID5 |
3773 BTRFS_BLOCK_GROUP_RAID6;
3775 seq = read_seqbegin(&fs_info->profiles_lock);
3777 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3778 (fs_info->avail_system_alloc_bits & allowed) &&
3779 !(bctl->sys.target & allowed)) ||
3780 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3781 (fs_info->avail_metadata_alloc_bits & allowed) &&
3782 !(bctl->meta.target & allowed))) {
3783 if (bctl->flags & BTRFS_BALANCE_FORCE) {
3784 btrfs_info(fs_info, "force reducing metadata integrity");
3786 btrfs_err(fs_info, "balance will reduce metadata "
3787 "integrity, use force if you want this");
3792 } while (read_seqretry(&fs_info->profiles_lock, seq));
3794 if (btrfs_get_num_tolerated_disk_barrier_failures(bctl->meta.target) <
3795 btrfs_get_num_tolerated_disk_barrier_failures(bctl->data.target)) {
3797 "metadata profile 0x%llx has lower redundancy than data profile 0x%llx",
3798 bctl->meta.target, bctl->data.target);
3801 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3802 fs_info->num_tolerated_disk_barrier_failures = min(
3803 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info),
3804 btrfs_get_num_tolerated_disk_barrier_failures(
3808 ret = insert_balance_item(fs_info->tree_root, bctl);
3809 if (ret && ret != -EEXIST)
3812 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3813 BUG_ON(ret == -EEXIST);
3814 set_balance_control(bctl);
3816 BUG_ON(ret != -EEXIST);
3817 spin_lock(&fs_info->balance_lock);
3818 update_balance_args(bctl);
3819 spin_unlock(&fs_info->balance_lock);
3822 atomic_inc(&fs_info->balance_running);
3823 mutex_unlock(&fs_info->balance_mutex);
3825 ret = __btrfs_balance(fs_info);
3827 mutex_lock(&fs_info->balance_mutex);
3828 atomic_dec(&fs_info->balance_running);
3830 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3831 fs_info->num_tolerated_disk_barrier_failures =
3832 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3836 memset(bargs, 0, sizeof(*bargs));
3837 update_ioctl_balance_args(fs_info, 0, bargs);
3840 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3841 balance_need_close(fs_info)) {
3842 __cancel_balance(fs_info);
3845 wake_up(&fs_info->balance_wait_q);
3849 if (bctl->flags & BTRFS_BALANCE_RESUME)
3850 __cancel_balance(fs_info);
3853 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3858 static int balance_kthread(void *data)
3860 struct btrfs_fs_info *fs_info = data;
3863 mutex_lock(&fs_info->volume_mutex);
3864 mutex_lock(&fs_info->balance_mutex);
3866 if (fs_info->balance_ctl) {
3867 btrfs_info(fs_info, "continuing balance");
3868 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3871 mutex_unlock(&fs_info->balance_mutex);
3872 mutex_unlock(&fs_info->volume_mutex);
3877 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3879 struct task_struct *tsk;
3881 spin_lock(&fs_info->balance_lock);
3882 if (!fs_info->balance_ctl) {
3883 spin_unlock(&fs_info->balance_lock);
3886 spin_unlock(&fs_info->balance_lock);
3888 if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
3889 btrfs_info(fs_info, "force skipping balance");
3893 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3894 return PTR_ERR_OR_ZERO(tsk);
3897 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3899 struct btrfs_balance_control *bctl;
3900 struct btrfs_balance_item *item;
3901 struct btrfs_disk_balance_args disk_bargs;
3902 struct btrfs_path *path;
3903 struct extent_buffer *leaf;
3904 struct btrfs_key key;
3907 path = btrfs_alloc_path();
3911 key.objectid = BTRFS_BALANCE_OBJECTID;
3912 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3915 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3918 if (ret > 0) { /* ret = -ENOENT; */
3923 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3929 leaf = path->nodes[0];
3930 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3932 bctl->fs_info = fs_info;
3933 bctl->flags = btrfs_balance_flags(leaf, item);
3934 bctl->flags |= BTRFS_BALANCE_RESUME;
3936 btrfs_balance_data(leaf, item, &disk_bargs);
3937 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3938 btrfs_balance_meta(leaf, item, &disk_bargs);
3939 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3940 btrfs_balance_sys(leaf, item, &disk_bargs);
3941 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3943 WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
3945 mutex_lock(&fs_info->volume_mutex);
3946 mutex_lock(&fs_info->balance_mutex);
3948 set_balance_control(bctl);
3950 mutex_unlock(&fs_info->balance_mutex);
3951 mutex_unlock(&fs_info->volume_mutex);
3953 btrfs_free_path(path);
3957 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3961 mutex_lock(&fs_info->balance_mutex);
3962 if (!fs_info->balance_ctl) {
3963 mutex_unlock(&fs_info->balance_mutex);
3967 if (atomic_read(&fs_info->balance_running)) {
3968 atomic_inc(&fs_info->balance_pause_req);
3969 mutex_unlock(&fs_info->balance_mutex);
3971 wait_event(fs_info->balance_wait_q,
3972 atomic_read(&fs_info->balance_running) == 0);
3974 mutex_lock(&fs_info->balance_mutex);
3975 /* we are good with balance_ctl ripped off from under us */
3976 BUG_ON(atomic_read(&fs_info->balance_running));
3977 atomic_dec(&fs_info->balance_pause_req);
3982 mutex_unlock(&fs_info->balance_mutex);
3986 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3988 if (fs_info->sb->s_flags & MS_RDONLY)
3991 mutex_lock(&fs_info->balance_mutex);
3992 if (!fs_info->balance_ctl) {
3993 mutex_unlock(&fs_info->balance_mutex);
3997 atomic_inc(&fs_info->balance_cancel_req);
3999 * if we are running just wait and return, balance item is
4000 * deleted in btrfs_balance in this case
4002 if (atomic_read(&fs_info->balance_running)) {
4003 mutex_unlock(&fs_info->balance_mutex);
4004 wait_event(fs_info->balance_wait_q,
4005 atomic_read(&fs_info->balance_running) == 0);
4006 mutex_lock(&fs_info->balance_mutex);
4008 /* __cancel_balance needs volume_mutex */
4009 mutex_unlock(&fs_info->balance_mutex);
4010 mutex_lock(&fs_info->volume_mutex);
4011 mutex_lock(&fs_info->balance_mutex);
4013 if (fs_info->balance_ctl)
4014 __cancel_balance(fs_info);
4016 mutex_unlock(&fs_info->volume_mutex);
4019 BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
4020 atomic_dec(&fs_info->balance_cancel_req);
4021 mutex_unlock(&fs_info->balance_mutex);
4025 static int btrfs_uuid_scan_kthread(void *data)
4027 struct btrfs_fs_info *fs_info = data;
4028 struct btrfs_root *root = fs_info->tree_root;
4029 struct btrfs_key key;
4030 struct btrfs_key max_key;
4031 struct btrfs_path *path = NULL;
4033 struct extent_buffer *eb;
4035 struct btrfs_root_item root_item;
4037 struct btrfs_trans_handle *trans = NULL;
4039 path = btrfs_alloc_path();
4046 key.type = BTRFS_ROOT_ITEM_KEY;
4049 max_key.objectid = (u64)-1;
4050 max_key.type = BTRFS_ROOT_ITEM_KEY;
4051 max_key.offset = (u64)-1;
4054 ret = btrfs_search_forward(root, &key, path, 0);
4061 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4062 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4063 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4064 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4067 eb = path->nodes[0];
4068 slot = path->slots[0];
4069 item_size = btrfs_item_size_nr(eb, slot);
4070 if (item_size < sizeof(root_item))
4073 read_extent_buffer(eb, &root_item,
4074 btrfs_item_ptr_offset(eb, slot),
4075 (int)sizeof(root_item));
4076 if (btrfs_root_refs(&root_item) == 0)
4079 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4080 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4084 btrfs_release_path(path);
4086 * 1 - subvol uuid item
4087 * 1 - received_subvol uuid item
4089 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4090 if (IS_ERR(trans)) {
4091 ret = PTR_ERR(trans);
4099 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4100 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
4102 BTRFS_UUID_KEY_SUBVOL,
4105 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4111 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4112 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
4113 root_item.received_uuid,
4114 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4117 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4125 ret = btrfs_end_transaction(trans, fs_info->uuid_root);
4131 btrfs_release_path(path);
4132 if (key.offset < (u64)-1) {
4134 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4136 key.type = BTRFS_ROOT_ITEM_KEY;
4137 } else if (key.objectid < (u64)-1) {
4139 key.type = BTRFS_ROOT_ITEM_KEY;
4148 btrfs_free_path(path);
4149 if (trans && !IS_ERR(trans))
4150 btrfs_end_transaction(trans, fs_info->uuid_root);
4152 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4154 fs_info->update_uuid_tree_gen = 1;
4155 up(&fs_info->uuid_tree_rescan_sem);
4160 * Callback for btrfs_uuid_tree_iterate().
4162 * 0 check succeeded, the entry is not outdated.
4163 * < 0 if an error occurred.
4164 * > 0 if the check failed, which means the caller shall remove the entry.
4166 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
4167 u8 *uuid, u8 type, u64 subid)
4169 struct btrfs_key key;
4171 struct btrfs_root *subvol_root;
4173 if (type != BTRFS_UUID_KEY_SUBVOL &&
4174 type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
4177 key.objectid = subid;
4178 key.type = BTRFS_ROOT_ITEM_KEY;
4179 key.offset = (u64)-1;
4180 subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
4181 if (IS_ERR(subvol_root)) {
4182 ret = PTR_ERR(subvol_root);
4189 case BTRFS_UUID_KEY_SUBVOL:
4190 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
4193 case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
4194 if (memcmp(uuid, subvol_root->root_item.received_uuid,
4204 static int btrfs_uuid_rescan_kthread(void *data)
4206 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4210 * 1st step is to iterate through the existing UUID tree and
4211 * to delete all entries that contain outdated data.
4212 * 2nd step is to add all missing entries to the UUID tree.
4214 ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4216 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4217 up(&fs_info->uuid_tree_rescan_sem);
4220 return btrfs_uuid_scan_kthread(data);
4223 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4225 struct btrfs_trans_handle *trans;
4226 struct btrfs_root *tree_root = fs_info->tree_root;
4227 struct btrfs_root *uuid_root;
4228 struct task_struct *task;
4235 trans = btrfs_start_transaction(tree_root, 2);
4237 return PTR_ERR(trans);
4239 uuid_root = btrfs_create_tree(trans, fs_info,
4240 BTRFS_UUID_TREE_OBJECTID);
4241 if (IS_ERR(uuid_root)) {
4242 ret = PTR_ERR(uuid_root);
4243 btrfs_abort_transaction(trans, tree_root, ret);
4244 btrfs_end_transaction(trans, tree_root);
4248 fs_info->uuid_root = uuid_root;
4250 ret = btrfs_commit_transaction(trans, tree_root);
4254 down(&fs_info->uuid_tree_rescan_sem);
4255 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4257 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4258 btrfs_warn(fs_info, "failed to start uuid_scan task");
4259 up(&fs_info->uuid_tree_rescan_sem);
4260 return PTR_ERR(task);
4266 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4268 struct task_struct *task;
4270 down(&fs_info->uuid_tree_rescan_sem);
4271 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4273 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4274 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4275 up(&fs_info->uuid_tree_rescan_sem);
4276 return PTR_ERR(task);
4283 * shrinking a device means finding all of the device extents past
4284 * the new size, and then following the back refs to the chunks.
4285 * The chunk relocation code actually frees the device extent
4287 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4289 struct btrfs_trans_handle *trans;
4290 struct btrfs_root *root = device->dev_root;
4291 struct btrfs_dev_extent *dev_extent = NULL;
4292 struct btrfs_path *path;
4298 bool retried = false;
4299 bool checked_pending_chunks = false;
4300 struct extent_buffer *l;
4301 struct btrfs_key key;
4302 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4303 u64 old_total = btrfs_super_total_bytes(super_copy);
4304 u64 old_size = btrfs_device_get_total_bytes(device);
4305 u64 diff = old_size - new_size;
4307 if (device->is_tgtdev_for_dev_replace)
4310 path = btrfs_alloc_path();
4314 path->reada = READA_FORWARD;
4318 btrfs_device_set_total_bytes(device, new_size);
4319 if (device->writeable) {
4320 device->fs_devices->total_rw_bytes -= diff;
4321 spin_lock(&root->fs_info->free_chunk_lock);
4322 root->fs_info->free_chunk_space -= diff;
4323 spin_unlock(&root->fs_info->free_chunk_lock);
4325 unlock_chunks(root);
4328 key.objectid = device->devid;
4329 key.offset = (u64)-1;
4330 key.type = BTRFS_DEV_EXTENT_KEY;
4333 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
4334 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4336 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4340 ret = btrfs_previous_item(root, path, 0, key.type);
4342 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4347 btrfs_release_path(path);
4352 slot = path->slots[0];
4353 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4355 if (key.objectid != device->devid) {
4356 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4357 btrfs_release_path(path);
4361 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4362 length = btrfs_dev_extent_length(l, dev_extent);
4364 if (key.offset + length <= new_size) {
4365 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4366 btrfs_release_path(path);
4370 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4371 btrfs_release_path(path);
4373 ret = btrfs_relocate_chunk(root, chunk_offset);
4374 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4375 if (ret && ret != -ENOSPC)
4379 } while (key.offset-- > 0);
4381 if (failed && !retried) {
4385 } else if (failed && retried) {
4390 /* Shrinking succeeded, else we would be at "done". */
4391 trans = btrfs_start_transaction(root, 0);
4392 if (IS_ERR(trans)) {
4393 ret = PTR_ERR(trans);
4400 * We checked in the above loop all device extents that were already in
4401 * the device tree. However before we have updated the device's
4402 * total_bytes to the new size, we might have had chunk allocations that
4403 * have not complete yet (new block groups attached to transaction
4404 * handles), and therefore their device extents were not yet in the
4405 * device tree and we missed them in the loop above. So if we have any
4406 * pending chunk using a device extent that overlaps the device range
4407 * that we can not use anymore, commit the current transaction and
4408 * repeat the search on the device tree - this way we guarantee we will
4409 * not have chunks using device extents that end beyond 'new_size'.
4411 if (!checked_pending_chunks) {
4412 u64 start = new_size;
4413 u64 len = old_size - new_size;
4415 if (contains_pending_extent(trans->transaction, device,
4417 unlock_chunks(root);
4418 checked_pending_chunks = true;
4421 ret = btrfs_commit_transaction(trans, root);
4428 btrfs_device_set_disk_total_bytes(device, new_size);
4429 if (list_empty(&device->resized_list))
4430 list_add_tail(&device->resized_list,
4431 &root->fs_info->fs_devices->resized_devices);
4433 WARN_ON(diff > old_total);
4434 btrfs_set_super_total_bytes(super_copy, old_total - diff);
4435 unlock_chunks(root);
4437 /* Now btrfs_update_device() will change the on-disk size. */
4438 ret = btrfs_update_device(trans, device);
4439 btrfs_end_transaction(trans, root);
4441 btrfs_free_path(path);
4444 btrfs_device_set_total_bytes(device, old_size);
4445 if (device->writeable)
4446 device->fs_devices->total_rw_bytes += diff;
4447 spin_lock(&root->fs_info->free_chunk_lock);
4448 root->fs_info->free_chunk_space += diff;
4449 spin_unlock(&root->fs_info->free_chunk_lock);
4450 unlock_chunks(root);
4455 static int btrfs_add_system_chunk(struct btrfs_root *root,
4456 struct btrfs_key *key,
4457 struct btrfs_chunk *chunk, int item_size)
4459 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4460 struct btrfs_disk_key disk_key;
4465 array_size = btrfs_super_sys_array_size(super_copy);
4466 if (array_size + item_size + sizeof(disk_key)
4467 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4468 unlock_chunks(root);
4472 ptr = super_copy->sys_chunk_array + array_size;
4473 btrfs_cpu_key_to_disk(&disk_key, key);
4474 memcpy(ptr, &disk_key, sizeof(disk_key));
4475 ptr += sizeof(disk_key);
4476 memcpy(ptr, chunk, item_size);
4477 item_size += sizeof(disk_key);
4478 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4479 unlock_chunks(root);
4485 * sort the devices in descending order by max_avail, total_avail
4487 static int btrfs_cmp_device_info(const void *a, const void *b)
4489 const struct btrfs_device_info *di_a = a;
4490 const struct btrfs_device_info *di_b = b;
4492 if (di_a->max_avail > di_b->max_avail)
4494 if (di_a->max_avail < di_b->max_avail)
4496 if (di_a->total_avail > di_b->total_avail)
4498 if (di_a->total_avail < di_b->total_avail)
4503 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
4505 /* TODO allow them to set a preferred stripe size */
4509 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4511 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4514 btrfs_set_fs_incompat(info, RAID56);
4517 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r) \
4518 - sizeof(struct btrfs_item) \
4519 - sizeof(struct btrfs_chunk)) \
4520 / sizeof(struct btrfs_stripe) + 1)
4522 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
4523 - 2 * sizeof(struct btrfs_disk_key) \
4524 - 2 * sizeof(struct btrfs_chunk)) \
4525 / sizeof(struct btrfs_stripe) + 1)
4527 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4528 struct btrfs_root *extent_root, u64 start,
4531 struct btrfs_fs_info *info = extent_root->fs_info;
4532 struct btrfs_fs_devices *fs_devices = info->fs_devices;
4533 struct list_head *cur;
4534 struct map_lookup *map = NULL;
4535 struct extent_map_tree *em_tree;
4536 struct extent_map *em;
4537 struct btrfs_device_info *devices_info = NULL;
4539 int num_stripes; /* total number of stripes to allocate */
4540 int data_stripes; /* number of stripes that count for
4542 int sub_stripes; /* sub_stripes info for map */
4543 int dev_stripes; /* stripes per dev */
4544 int devs_max; /* max devs to use */
4545 int devs_min; /* min devs needed */
4546 int devs_increment; /* ndevs has to be a multiple of this */
4547 int ncopies; /* how many copies to data has */
4549 u64 max_stripe_size;
4553 u64 raid_stripe_len = BTRFS_STRIPE_LEN;
4559 BUG_ON(!alloc_profile_is_valid(type, 0));
4561 if (list_empty(&fs_devices->alloc_list))
4564 index = __get_raid_index(type);
4566 sub_stripes = btrfs_raid_array[index].sub_stripes;
4567 dev_stripes = btrfs_raid_array[index].dev_stripes;
4568 devs_max = btrfs_raid_array[index].devs_max;
4569 devs_min = btrfs_raid_array[index].devs_min;
4570 devs_increment = btrfs_raid_array[index].devs_increment;
4571 ncopies = btrfs_raid_array[index].ncopies;
4573 if (type & BTRFS_BLOCK_GROUP_DATA) {
4574 max_stripe_size = SZ_1G;
4575 max_chunk_size = 10 * max_stripe_size;
4577 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4578 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4579 /* for larger filesystems, use larger metadata chunks */
4580 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
4581 max_stripe_size = SZ_1G;
4583 max_stripe_size = SZ_256M;
4584 max_chunk_size = max_stripe_size;
4586 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4587 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4588 max_stripe_size = SZ_32M;
4589 max_chunk_size = 2 * max_stripe_size;
4591 devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4593 btrfs_err(info, "invalid chunk type 0x%llx requested",
4598 /* we don't want a chunk larger than 10% of writeable space */
4599 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4602 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4607 cur = fs_devices->alloc_list.next;
4610 * in the first pass through the devices list, we gather information
4611 * about the available holes on each device.
4614 while (cur != &fs_devices->alloc_list) {
4615 struct btrfs_device *device;
4619 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4623 if (!device->writeable) {
4625 "BTRFS: read-only device in alloc_list\n");
4629 if (!device->in_fs_metadata ||
4630 device->is_tgtdev_for_dev_replace)
4633 if (device->total_bytes > device->bytes_used)
4634 total_avail = device->total_bytes - device->bytes_used;
4638 /* If there is no space on this device, skip it. */
4639 if (total_avail == 0)
4642 ret = find_free_dev_extent(trans, device,
4643 max_stripe_size * dev_stripes,
4644 &dev_offset, &max_avail);
4645 if (ret && ret != -ENOSPC)
4649 max_avail = max_stripe_size * dev_stripes;
4651 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4654 if (ndevs == fs_devices->rw_devices) {
4655 WARN(1, "%s: found more than %llu devices\n",
4656 __func__, fs_devices->rw_devices);
4659 devices_info[ndevs].dev_offset = dev_offset;
4660 devices_info[ndevs].max_avail = max_avail;
4661 devices_info[ndevs].total_avail = total_avail;
4662 devices_info[ndevs].dev = device;
4667 * now sort the devices by hole size / available space
4669 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4670 btrfs_cmp_device_info, NULL);
4672 /* round down to number of usable stripes */
4673 ndevs -= ndevs % devs_increment;
4675 if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4680 if (devs_max && ndevs > devs_max)
4683 * the primary goal is to maximize the number of stripes, so use as many
4684 * devices as possible, even if the stripes are not maximum sized.
4686 stripe_size = devices_info[ndevs-1].max_avail;
4687 num_stripes = ndevs * dev_stripes;
4690 * this will have to be fixed for RAID1 and RAID10 over
4693 data_stripes = num_stripes / ncopies;
4695 if (type & BTRFS_BLOCK_GROUP_RAID5) {
4696 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
4697 extent_root->stripesize);
4698 data_stripes = num_stripes - 1;
4700 if (type & BTRFS_BLOCK_GROUP_RAID6) {
4701 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
4702 extent_root->stripesize);
4703 data_stripes = num_stripes - 2;
4707 * Use the number of data stripes to figure out how big this chunk
4708 * is really going to be in terms of logical address space,
4709 * and compare that answer with the max chunk size
4711 if (stripe_size * data_stripes > max_chunk_size) {
4712 u64 mask = (1ULL << 24) - 1;
4714 stripe_size = div_u64(max_chunk_size, data_stripes);
4716 /* bump the answer up to a 16MB boundary */
4717 stripe_size = (stripe_size + mask) & ~mask;
4719 /* but don't go higher than the limits we found
4720 * while searching for free extents
4722 if (stripe_size > devices_info[ndevs-1].max_avail)
4723 stripe_size = devices_info[ndevs-1].max_avail;
4726 stripe_size = div_u64(stripe_size, dev_stripes);
4728 /* align to BTRFS_STRIPE_LEN */
4729 stripe_size = div_u64(stripe_size, raid_stripe_len);
4730 stripe_size *= raid_stripe_len;
4732 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4737 map->num_stripes = num_stripes;
4739 for (i = 0; i < ndevs; ++i) {
4740 for (j = 0; j < dev_stripes; ++j) {
4741 int s = i * dev_stripes + j;
4742 map->stripes[s].dev = devices_info[i].dev;
4743 map->stripes[s].physical = devices_info[i].dev_offset +
4747 map->sector_size = extent_root->sectorsize;
4748 map->stripe_len = raid_stripe_len;
4749 map->io_align = raid_stripe_len;
4750 map->io_width = raid_stripe_len;
4752 map->sub_stripes = sub_stripes;
4754 num_bytes = stripe_size * data_stripes;
4756 trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
4758 em = alloc_extent_map();
4764 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4765 em->map_lookup = map;
4767 em->len = num_bytes;
4768 em->block_start = 0;
4769 em->block_len = em->len;
4770 em->orig_block_len = stripe_size;
4772 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4773 write_lock(&em_tree->lock);
4774 ret = add_extent_mapping(em_tree, em, 0);
4776 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4777 atomic_inc(&em->refs);
4779 write_unlock(&em_tree->lock);
4781 free_extent_map(em);
4785 ret = btrfs_make_block_group(trans, extent_root, 0, type,
4786 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4789 goto error_del_extent;
4791 for (i = 0; i < map->num_stripes; i++) {
4792 num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4793 btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4796 spin_lock(&extent_root->fs_info->free_chunk_lock);
4797 extent_root->fs_info->free_chunk_space -= (stripe_size *
4799 spin_unlock(&extent_root->fs_info->free_chunk_lock);
4801 free_extent_map(em);
4802 check_raid56_incompat_flag(extent_root->fs_info, type);
4804 kfree(devices_info);
4808 write_lock(&em_tree->lock);
4809 remove_extent_mapping(em_tree, em);
4810 write_unlock(&em_tree->lock);
4812 /* One for our allocation */
4813 free_extent_map(em);
4814 /* One for the tree reference */
4815 free_extent_map(em);
4816 /* One for the pending_chunks list reference */
4817 free_extent_map(em);
4819 kfree(devices_info);
4823 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4824 struct btrfs_root *extent_root,
4825 u64 chunk_offset, u64 chunk_size)
4827 struct btrfs_key key;
4828 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4829 struct btrfs_device *device;
4830 struct btrfs_chunk *chunk;
4831 struct btrfs_stripe *stripe;
4832 struct extent_map_tree *em_tree;
4833 struct extent_map *em;
4834 struct map_lookup *map;
4841 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4842 read_lock(&em_tree->lock);
4843 em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size);
4844 read_unlock(&em_tree->lock);
4847 btrfs_crit(extent_root->fs_info, "unable to find logical "
4848 "%Lu len %Lu", chunk_offset, chunk_size);
4852 if (em->start != chunk_offset || em->len != chunk_size) {
4853 btrfs_crit(extent_root->fs_info, "found a bad mapping, wanted"
4854 " %Lu-%Lu, found %Lu-%Lu", chunk_offset,
4855 chunk_size, em->start, em->len);
4856 free_extent_map(em);
4860 map = em->map_lookup;
4861 item_size = btrfs_chunk_item_size(map->num_stripes);
4862 stripe_size = em->orig_block_len;
4864 chunk = kzalloc(item_size, GFP_NOFS);
4871 * Take the device list mutex to prevent races with the final phase of
4872 * a device replace operation that replaces the device object associated
4873 * with the map's stripes, because the device object's id can change
4874 * at any time during that final phase of the device replace operation
4875 * (dev-replace.c:btrfs_dev_replace_finishing()).
4877 mutex_lock(&chunk_root->fs_info->fs_devices->device_list_mutex);
4878 for (i = 0; i < map->num_stripes; i++) {
4879 device = map->stripes[i].dev;
4880 dev_offset = map->stripes[i].physical;
4882 ret = btrfs_update_device(trans, device);
4885 ret = btrfs_alloc_dev_extent(trans, device,
4886 chunk_root->root_key.objectid,
4887 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4888 chunk_offset, dev_offset,
4894 mutex_unlock(&chunk_root->fs_info->fs_devices->device_list_mutex);
4898 stripe = &chunk->stripe;
4899 for (i = 0; i < map->num_stripes; i++) {
4900 device = map->stripes[i].dev;
4901 dev_offset = map->stripes[i].physical;
4903 btrfs_set_stack_stripe_devid(stripe, device->devid);
4904 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4905 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4908 mutex_unlock(&chunk_root->fs_info->fs_devices->device_list_mutex);
4910 btrfs_set_stack_chunk_length(chunk, chunk_size);
4911 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4912 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4913 btrfs_set_stack_chunk_type(chunk, map->type);
4914 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4915 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4916 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4917 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4918 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4920 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4921 key.type = BTRFS_CHUNK_ITEM_KEY;
4922 key.offset = chunk_offset;
4924 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4925 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4927 * TODO: Cleanup of inserted chunk root in case of
4930 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
4936 free_extent_map(em);
4941 * Chunk allocation falls into two parts. The first part does works
4942 * that make the new allocated chunk useable, but not do any operation
4943 * that modifies the chunk tree. The second part does the works that
4944 * require modifying the chunk tree. This division is important for the
4945 * bootstrap process of adding storage to a seed btrfs.
4947 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4948 struct btrfs_root *extent_root, u64 type)
4952 ASSERT(mutex_is_locked(&extent_root->fs_info->chunk_mutex));
4953 chunk_offset = find_next_chunk(extent_root->fs_info);
4954 return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type);
4957 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4958 struct btrfs_root *root,
4959 struct btrfs_device *device)
4962 u64 sys_chunk_offset;
4964 struct btrfs_fs_info *fs_info = root->fs_info;
4965 struct btrfs_root *extent_root = fs_info->extent_root;
4968 chunk_offset = find_next_chunk(fs_info);
4969 alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
4970 ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset,
4975 sys_chunk_offset = find_next_chunk(root->fs_info);
4976 alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
4977 ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset,
4982 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
4986 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
4987 BTRFS_BLOCK_GROUP_RAID10 |
4988 BTRFS_BLOCK_GROUP_RAID5 |
4989 BTRFS_BLOCK_GROUP_DUP)) {
4991 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
5000 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
5002 struct extent_map *em;
5003 struct map_lookup *map;
5004 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5009 read_lock(&map_tree->map_tree.lock);
5010 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
5011 read_unlock(&map_tree->map_tree.lock);
5015 map = em->map_lookup;
5016 for (i = 0; i < map->num_stripes; i++) {
5017 if (map->stripes[i].dev->missing) {
5022 if (!map->stripes[i].dev->writeable) {
5029 * If the number of missing devices is larger than max errors,
5030 * we can not write the data into that chunk successfully, so
5033 if (miss_ndevs > btrfs_chunk_max_errors(map))
5036 free_extent_map(em);
5040 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
5042 extent_map_tree_init(&tree->map_tree);
5045 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
5047 struct extent_map *em;
5050 write_lock(&tree->map_tree.lock);
5051 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
5053 remove_extent_mapping(&tree->map_tree, em);
5054 write_unlock(&tree->map_tree.lock);
5058 free_extent_map(em);
5059 /* once for the tree */
5060 free_extent_map(em);
5064 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5066 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5067 struct extent_map *em;
5068 struct map_lookup *map;
5069 struct extent_map_tree *em_tree = &map_tree->map_tree;
5072 read_lock(&em_tree->lock);
5073 em = lookup_extent_mapping(em_tree, logical, len);
5074 read_unlock(&em_tree->lock);
5077 * We could return errors for these cases, but that could get ugly and
5078 * we'd probably do the same thing which is just not do anything else
5079 * and exit, so return 1 so the callers don't try to use other copies.
5082 btrfs_crit(fs_info, "No mapping for %Lu-%Lu", logical,
5087 if (em->start > logical || em->start + em->len < logical) {
5088 btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got "
5089 "%Lu-%Lu", logical, logical+len, em->start,
5090 em->start + em->len);
5091 free_extent_map(em);
5095 map = em->map_lookup;
5096 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
5097 ret = map->num_stripes;
5098 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5099 ret = map->sub_stripes;
5100 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5102 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5106 free_extent_map(em);
5108 btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
5109 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
5111 btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
5116 unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
5117 struct btrfs_mapping_tree *map_tree,
5120 struct extent_map *em;
5121 struct map_lookup *map;
5122 struct extent_map_tree *em_tree = &map_tree->map_tree;
5123 unsigned long len = root->sectorsize;
5125 read_lock(&em_tree->lock);
5126 em = lookup_extent_mapping(em_tree, logical, len);
5127 read_unlock(&em_tree->lock);
5130 BUG_ON(em->start > logical || em->start + em->len < logical);
5131 map = em->map_lookup;
5132 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5133 len = map->stripe_len * nr_data_stripes(map);
5134 free_extent_map(em);
5138 int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
5139 u64 logical, u64 len, int mirror_num)
5141 struct extent_map *em;
5142 struct map_lookup *map;
5143 struct extent_map_tree *em_tree = &map_tree->map_tree;
5146 read_lock(&em_tree->lock);
5147 em = lookup_extent_mapping(em_tree, logical, len);
5148 read_unlock(&em_tree->lock);
5151 BUG_ON(em->start > logical || em->start + em->len < logical);
5152 map = em->map_lookup;
5153 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5155 free_extent_map(em);
5159 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5160 struct map_lookup *map, int first, int num,
5161 int optimal, int dev_replace_is_ongoing)
5165 struct btrfs_device *srcdev;
5167 if (dev_replace_is_ongoing &&
5168 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5169 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5170 srcdev = fs_info->dev_replace.srcdev;
5175 * try to avoid the drive that is the source drive for a
5176 * dev-replace procedure, only choose it if no other non-missing
5177 * mirror is available
5179 for (tolerance = 0; tolerance < 2; tolerance++) {
5180 if (map->stripes[optimal].dev->bdev &&
5181 (tolerance || map->stripes[optimal].dev != srcdev))
5183 for (i = first; i < first + num; i++) {
5184 if (map->stripes[i].dev->bdev &&
5185 (tolerance || map->stripes[i].dev != srcdev))
5190 /* we couldn't find one that doesn't fail. Just return something
5191 * and the io error handling code will clean up eventually
5196 static inline int parity_smaller(u64 a, u64 b)
5201 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5202 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5204 struct btrfs_bio_stripe s;
5211 for (i = 0; i < num_stripes - 1; i++) {
5212 if (parity_smaller(bbio->raid_map[i],
5213 bbio->raid_map[i+1])) {
5214 s = bbio->stripes[i];
5215 l = bbio->raid_map[i];
5216 bbio->stripes[i] = bbio->stripes[i+1];
5217 bbio->raid_map[i] = bbio->raid_map[i+1];
5218 bbio->stripes[i+1] = s;
5219 bbio->raid_map[i+1] = l;
5227 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5229 struct btrfs_bio *bbio = kzalloc(
5230 /* the size of the btrfs_bio */
5231 sizeof(struct btrfs_bio) +
5232 /* plus the variable array for the stripes */
5233 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5234 /* plus the variable array for the tgt dev */
5235 sizeof(int) * (real_stripes) +
5237 * plus the raid_map, which includes both the tgt dev
5240 sizeof(u64) * (total_stripes),
5241 GFP_NOFS|__GFP_NOFAIL);
5243 atomic_set(&bbio->error, 0);
5244 atomic_set(&bbio->refs, 1);
5249 void btrfs_get_bbio(struct btrfs_bio *bbio)
5251 WARN_ON(!atomic_read(&bbio->refs));
5252 atomic_inc(&bbio->refs);
5255 void btrfs_put_bbio(struct btrfs_bio *bbio)
5259 if (atomic_dec_and_test(&bbio->refs))
5263 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5264 u64 logical, u64 *length,
5265 struct btrfs_bio **bbio_ret,
5266 int mirror_num, int need_raid_map)
5268 struct extent_map *em;
5269 struct map_lookup *map;
5270 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5271 struct extent_map_tree *em_tree = &map_tree->map_tree;
5274 u64 stripe_end_offset;
5284 int tgtdev_indexes = 0;
5285 struct btrfs_bio *bbio = NULL;
5286 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
5287 int dev_replace_is_ongoing = 0;
5288 int num_alloc_stripes;
5289 int patch_the_first_stripe_for_dev_replace = 0;
5290 u64 physical_to_patch_in_first_stripe = 0;
5291 u64 raid56_full_stripe_start = (u64)-1;
5293 read_lock(&em_tree->lock);
5294 em = lookup_extent_mapping(em_tree, logical, *length);
5295 read_unlock(&em_tree->lock);
5298 btrfs_crit(fs_info, "unable to find logical %llu len %llu",
5303 if (em->start > logical || em->start + em->len < logical) {
5304 btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, "
5305 "found %Lu-%Lu", logical, em->start,
5306 em->start + em->len);
5307 free_extent_map(em);
5311 map = em->map_lookup;
5312 offset = logical - em->start;
5314 stripe_len = map->stripe_len;
5317 * stripe_nr counts the total number of stripes we have to stride
5318 * to get to this block
5320 stripe_nr = div64_u64(stripe_nr, stripe_len);
5322 stripe_offset = stripe_nr * stripe_len;
5323 if (offset < stripe_offset) {
5324 btrfs_crit(fs_info, "stripe math has gone wrong, "
5325 "stripe_offset=%llu, offset=%llu, start=%llu, "
5326 "logical=%llu, stripe_len=%llu",
5327 stripe_offset, offset, em->start, logical,
5329 free_extent_map(em);
5333 /* stripe_offset is the offset of this block in its stripe*/
5334 stripe_offset = offset - stripe_offset;
5336 /* if we're here for raid56, we need to know the stripe aligned start */
5337 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5338 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5339 raid56_full_stripe_start = offset;
5341 /* allow a write of a full stripe, but make sure we don't
5342 * allow straddling of stripes
5344 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5346 raid56_full_stripe_start *= full_stripe_len;
5349 if (rw & REQ_DISCARD) {
5350 /* we don't discard raid56 yet */
5351 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5355 *length = min_t(u64, em->len - offset, *length);
5356 } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5358 /* For writes to RAID[56], allow a full stripeset across all disks.
5359 For other RAID types and for RAID[56] reads, just allow a single
5360 stripe (on a single disk). */
5361 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
5363 max_len = stripe_len * nr_data_stripes(map) -
5364 (offset - raid56_full_stripe_start);
5366 /* we limit the length of each bio to what fits in a stripe */
5367 max_len = stripe_len - stripe_offset;
5369 *length = min_t(u64, em->len - offset, max_len);
5371 *length = em->len - offset;
5374 /* This is for when we're called from btrfs_merge_bio_hook() and all
5375 it cares about is the length */
5379 btrfs_dev_replace_lock(dev_replace, 0);
5380 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5381 if (!dev_replace_is_ongoing)
5382 btrfs_dev_replace_unlock(dev_replace, 0);
5384 btrfs_dev_replace_set_lock_blocking(dev_replace);
5386 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
5387 !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) &&
5388 dev_replace->tgtdev != NULL) {
5390 * in dev-replace case, for repair case (that's the only
5391 * case where the mirror is selected explicitly when
5392 * calling btrfs_map_block), blocks left of the left cursor
5393 * can also be read from the target drive.
5394 * For REQ_GET_READ_MIRRORS, the target drive is added as
5395 * the last one to the array of stripes. For READ, it also
5396 * needs to be supported using the same mirror number.
5397 * If the requested block is not left of the left cursor,
5398 * EIO is returned. This can happen because btrfs_num_copies()
5399 * returns one more in the dev-replace case.
5401 u64 tmp_length = *length;
5402 struct btrfs_bio *tmp_bbio = NULL;
5403 int tmp_num_stripes;
5404 u64 srcdev_devid = dev_replace->srcdev->devid;
5405 int index_srcdev = 0;
5407 u64 physical_of_found = 0;
5409 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
5410 logical, &tmp_length, &tmp_bbio, 0, 0);
5412 WARN_ON(tmp_bbio != NULL);
5416 tmp_num_stripes = tmp_bbio->num_stripes;
5417 if (mirror_num > tmp_num_stripes) {
5419 * REQ_GET_READ_MIRRORS does not contain this
5420 * mirror, that means that the requested area
5421 * is not left of the left cursor
5424 btrfs_put_bbio(tmp_bbio);
5429 * process the rest of the function using the mirror_num
5430 * of the source drive. Therefore look it up first.
5431 * At the end, patch the device pointer to the one of the
5434 for (i = 0; i < tmp_num_stripes; i++) {
5435 if (tmp_bbio->stripes[i].dev->devid != srcdev_devid)
5439 * In case of DUP, in order to keep it simple, only add
5440 * the mirror with the lowest physical address
5443 physical_of_found <= tmp_bbio->stripes[i].physical)
5448 physical_of_found = tmp_bbio->stripes[i].physical;
5451 btrfs_put_bbio(tmp_bbio);
5459 mirror_num = index_srcdev + 1;
5460 patch_the_first_stripe_for_dev_replace = 1;
5461 physical_to_patch_in_first_stripe = physical_of_found;
5462 } else if (mirror_num > map->num_stripes) {
5468 stripe_nr_orig = stripe_nr;
5469 stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
5470 stripe_nr_end = div_u64(stripe_nr_end, map->stripe_len);
5471 stripe_end_offset = stripe_nr_end * map->stripe_len -
5474 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5475 if (rw & REQ_DISCARD)
5476 num_stripes = min_t(u64, map->num_stripes,
5477 stripe_nr_end - stripe_nr_orig);
5478 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5480 if (!(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)))
5482 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
5483 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))
5484 num_stripes = map->num_stripes;
5485 else if (mirror_num)
5486 stripe_index = mirror_num - 1;
5488 stripe_index = find_live_mirror(fs_info, map, 0,
5490 current->pid % map->num_stripes,
5491 dev_replace_is_ongoing);
5492 mirror_num = stripe_index + 1;
5495 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
5496 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
5497 num_stripes = map->num_stripes;
5498 } else if (mirror_num) {
5499 stripe_index = mirror_num - 1;
5504 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5505 u32 factor = map->num_stripes / map->sub_stripes;
5507 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5508 stripe_index *= map->sub_stripes;
5510 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
5511 num_stripes = map->sub_stripes;
5512 else if (rw & REQ_DISCARD)
5513 num_stripes = min_t(u64, map->sub_stripes *
5514 (stripe_nr_end - stripe_nr_orig),
5516 else if (mirror_num)
5517 stripe_index += mirror_num - 1;
5519 int old_stripe_index = stripe_index;
5520 stripe_index = find_live_mirror(fs_info, map,
5522 map->sub_stripes, stripe_index +
5523 current->pid % map->sub_stripes,
5524 dev_replace_is_ongoing);
5525 mirror_num = stripe_index - old_stripe_index + 1;
5528 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5529 if (need_raid_map &&
5530 ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) ||
5532 /* push stripe_nr back to the start of the full stripe */
5533 stripe_nr = div_u64(raid56_full_stripe_start,
5534 stripe_len * nr_data_stripes(map));
5536 /* RAID[56] write or recovery. Return all stripes */
5537 num_stripes = map->num_stripes;
5538 max_errors = nr_parity_stripes(map);
5540 *length = map->stripe_len;
5545 * Mirror #0 or #1 means the original data block.
5546 * Mirror #2 is RAID5 parity block.
5547 * Mirror #3 is RAID6 Q block.
5549 stripe_nr = div_u64_rem(stripe_nr,
5550 nr_data_stripes(map), &stripe_index);
5552 stripe_index = nr_data_stripes(map) +
5555 /* We distribute the parity blocks across stripes */
5556 div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
5558 if (!(rw & (REQ_WRITE | REQ_DISCARD |
5559 REQ_GET_READ_MIRRORS)) && mirror_num <= 1)
5564 * after this, stripe_nr is the number of stripes on this
5565 * device we have to walk to find the data, and stripe_index is
5566 * the number of our device in the stripe array
5568 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5570 mirror_num = stripe_index + 1;
5572 if (stripe_index >= map->num_stripes) {
5573 btrfs_crit(fs_info, "stripe index math went horribly wrong, "
5574 "got stripe_index=%u, num_stripes=%u",
5575 stripe_index, map->num_stripes);
5580 num_alloc_stripes = num_stripes;
5581 if (dev_replace_is_ongoing) {
5582 if (rw & (REQ_WRITE | REQ_DISCARD))
5583 num_alloc_stripes <<= 1;
5584 if (rw & REQ_GET_READ_MIRRORS)
5585 num_alloc_stripes++;
5586 tgtdev_indexes = num_stripes;
5589 bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
5594 if (dev_replace_is_ongoing)
5595 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
5597 /* build raid_map */
5598 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK &&
5599 need_raid_map && ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) ||
5604 bbio->raid_map = (u64 *)((void *)bbio->stripes +
5605 sizeof(struct btrfs_bio_stripe) *
5607 sizeof(int) * tgtdev_indexes);
5609 /* Work out the disk rotation on this stripe-set */
5610 div_u64_rem(stripe_nr, num_stripes, &rot);
5612 /* Fill in the logical address of each stripe */
5613 tmp = stripe_nr * nr_data_stripes(map);
5614 for (i = 0; i < nr_data_stripes(map); i++)
5615 bbio->raid_map[(i+rot) % num_stripes] =
5616 em->start + (tmp + i) * map->stripe_len;
5618 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5619 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5620 bbio->raid_map[(i+rot+1) % num_stripes] =
5624 if (rw & REQ_DISCARD) {
5626 u32 sub_stripes = 0;
5627 u64 stripes_per_dev = 0;
5628 u32 remaining_stripes = 0;
5629 u32 last_stripe = 0;
5632 (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
5633 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5636 sub_stripes = map->sub_stripes;
5638 factor = map->num_stripes / sub_stripes;
5639 stripes_per_dev = div_u64_rem(stripe_nr_end -
5642 &remaining_stripes);
5643 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5644 last_stripe *= sub_stripes;
5647 for (i = 0; i < num_stripes; i++) {
5648 bbio->stripes[i].physical =
5649 map->stripes[stripe_index].physical +
5650 stripe_offset + stripe_nr * map->stripe_len;
5651 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5653 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5654 BTRFS_BLOCK_GROUP_RAID10)) {
5655 bbio->stripes[i].length = stripes_per_dev *
5658 if (i / sub_stripes < remaining_stripes)
5659 bbio->stripes[i].length +=
5663 * Special for the first stripe and
5666 * |-------|...|-------|
5670 if (i < sub_stripes)
5671 bbio->stripes[i].length -=
5674 if (stripe_index >= last_stripe &&
5675 stripe_index <= (last_stripe +
5677 bbio->stripes[i].length -=
5680 if (i == sub_stripes - 1)
5683 bbio->stripes[i].length = *length;
5686 if (stripe_index == map->num_stripes) {
5687 /* This could only happen for RAID0/10 */
5693 for (i = 0; i < num_stripes; i++) {
5694 bbio->stripes[i].physical =
5695 map->stripes[stripe_index].physical +
5697 stripe_nr * map->stripe_len;
5698 bbio->stripes[i].dev =
5699 map->stripes[stripe_index].dev;
5704 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
5705 max_errors = btrfs_chunk_max_errors(map);
5708 sort_parity_stripes(bbio, num_stripes);
5711 if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) &&
5712 dev_replace->tgtdev != NULL) {
5713 int index_where_to_add;
5714 u64 srcdev_devid = dev_replace->srcdev->devid;
5717 * duplicate the write operations while the dev replace
5718 * procedure is running. Since the copying of the old disk
5719 * to the new disk takes place at run time while the
5720 * filesystem is mounted writable, the regular write
5721 * operations to the old disk have to be duplicated to go
5722 * to the new disk as well.
5723 * Note that device->missing is handled by the caller, and
5724 * that the write to the old disk is already set up in the
5727 index_where_to_add = num_stripes;
5728 for (i = 0; i < num_stripes; i++) {
5729 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5730 /* write to new disk, too */
5731 struct btrfs_bio_stripe *new =
5732 bbio->stripes + index_where_to_add;
5733 struct btrfs_bio_stripe *old =
5736 new->physical = old->physical;
5737 new->length = old->length;
5738 new->dev = dev_replace->tgtdev;
5739 bbio->tgtdev_map[i] = index_where_to_add;
5740 index_where_to_add++;
5745 num_stripes = index_where_to_add;
5746 } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) &&
5747 dev_replace->tgtdev != NULL) {
5748 u64 srcdev_devid = dev_replace->srcdev->devid;
5749 int index_srcdev = 0;
5751 u64 physical_of_found = 0;
5754 * During the dev-replace procedure, the target drive can
5755 * also be used to read data in case it is needed to repair
5756 * a corrupt block elsewhere. This is possible if the
5757 * requested area is left of the left cursor. In this area,
5758 * the target drive is a full copy of the source drive.
5760 for (i = 0; i < num_stripes; i++) {
5761 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5763 * In case of DUP, in order to keep it
5764 * simple, only add the mirror with the
5765 * lowest physical address
5768 physical_of_found <=
5769 bbio->stripes[i].physical)
5773 physical_of_found = bbio->stripes[i].physical;
5777 struct btrfs_bio_stripe *tgtdev_stripe =
5778 bbio->stripes + num_stripes;
5780 tgtdev_stripe->physical = physical_of_found;
5781 tgtdev_stripe->length =
5782 bbio->stripes[index_srcdev].length;
5783 tgtdev_stripe->dev = dev_replace->tgtdev;
5784 bbio->tgtdev_map[index_srcdev] = num_stripes;
5792 bbio->map_type = map->type;
5793 bbio->num_stripes = num_stripes;
5794 bbio->max_errors = max_errors;
5795 bbio->mirror_num = mirror_num;
5796 bbio->num_tgtdevs = tgtdev_indexes;
5799 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5800 * mirror_num == num_stripes + 1 && dev_replace target drive is
5801 * available as a mirror
5803 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5804 WARN_ON(num_stripes > 1);
5805 bbio->stripes[0].dev = dev_replace->tgtdev;
5806 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5807 bbio->mirror_num = map->num_stripes + 1;
5810 if (dev_replace_is_ongoing) {
5811 btrfs_dev_replace_clear_lock_blocking(dev_replace);
5812 btrfs_dev_replace_unlock(dev_replace, 0);
5814 free_extent_map(em);
5818 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5819 u64 logical, u64 *length,
5820 struct btrfs_bio **bbio_ret, int mirror_num)
5822 return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5826 /* For Scrub/replace */
5827 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, int rw,
5828 u64 logical, u64 *length,
5829 struct btrfs_bio **bbio_ret, int mirror_num,
5832 return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5833 mirror_num, need_raid_map);
5836 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
5837 u64 chunk_start, u64 physical, u64 devid,
5838 u64 **logical, int *naddrs, int *stripe_len)
5840 struct extent_map_tree *em_tree = &map_tree->map_tree;
5841 struct extent_map *em;
5842 struct map_lookup *map;
5850 read_lock(&em_tree->lock);
5851 em = lookup_extent_mapping(em_tree, chunk_start, 1);
5852 read_unlock(&em_tree->lock);
5855 printk(KERN_ERR "BTRFS: couldn't find em for chunk %Lu\n",
5860 if (em->start != chunk_start) {
5861 printk(KERN_ERR "BTRFS: bad chunk start, em=%Lu, wanted=%Lu\n",
5862 em->start, chunk_start);
5863 free_extent_map(em);
5866 map = em->map_lookup;
5869 rmap_len = map->stripe_len;
5871 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5872 length = div_u64(length, map->num_stripes / map->sub_stripes);
5873 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5874 length = div_u64(length, map->num_stripes);
5875 else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5876 length = div_u64(length, nr_data_stripes(map));
5877 rmap_len = map->stripe_len * nr_data_stripes(map);
5880 buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
5881 BUG_ON(!buf); /* -ENOMEM */
5883 for (i = 0; i < map->num_stripes; i++) {
5884 if (devid && map->stripes[i].dev->devid != devid)
5886 if (map->stripes[i].physical > physical ||
5887 map->stripes[i].physical + length <= physical)
5890 stripe_nr = physical - map->stripes[i].physical;
5891 stripe_nr = div_u64(stripe_nr, map->stripe_len);
5893 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5894 stripe_nr = stripe_nr * map->num_stripes + i;
5895 stripe_nr = div_u64(stripe_nr, map->sub_stripes);
5896 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5897 stripe_nr = stripe_nr * map->num_stripes + i;
5898 } /* else if RAID[56], multiply by nr_data_stripes().
5899 * Alternatively, just use rmap_len below instead of
5900 * map->stripe_len */
5902 bytenr = chunk_start + stripe_nr * rmap_len;
5903 WARN_ON(nr >= map->num_stripes);
5904 for (j = 0; j < nr; j++) {
5905 if (buf[j] == bytenr)
5909 WARN_ON(nr >= map->num_stripes);
5916 *stripe_len = rmap_len;
5918 free_extent_map(em);
5922 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
5924 bio->bi_private = bbio->private;
5925 bio->bi_end_io = bbio->end_io;
5928 btrfs_put_bbio(bbio);
5931 static void btrfs_end_bio(struct bio *bio)
5933 struct btrfs_bio *bbio = bio->bi_private;
5934 int is_orig_bio = 0;
5936 if (bio->bi_error) {
5937 atomic_inc(&bbio->error);
5938 if (bio->bi_error == -EIO || bio->bi_error == -EREMOTEIO) {
5939 unsigned int stripe_index =
5940 btrfs_io_bio(bio)->stripe_index;
5941 struct btrfs_device *dev;
5943 BUG_ON(stripe_index >= bbio->num_stripes);
5944 dev = bbio->stripes[stripe_index].dev;
5946 if (bio->bi_rw & WRITE)
5947 btrfs_dev_stat_inc(dev,
5948 BTRFS_DEV_STAT_WRITE_ERRS);
5950 btrfs_dev_stat_inc(dev,
5951 BTRFS_DEV_STAT_READ_ERRS);
5952 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
5953 btrfs_dev_stat_inc(dev,
5954 BTRFS_DEV_STAT_FLUSH_ERRS);
5955 btrfs_dev_stat_print_on_error(dev);
5960 if (bio == bbio->orig_bio)
5963 btrfs_bio_counter_dec(bbio->fs_info);
5965 if (atomic_dec_and_test(&bbio->stripes_pending)) {
5968 bio = bbio->orig_bio;
5971 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5972 /* only send an error to the higher layers if it is
5973 * beyond the tolerance of the btrfs bio
5975 if (atomic_read(&bbio->error) > bbio->max_errors) {
5976 bio->bi_error = -EIO;
5979 * this bio is actually up to date, we didn't
5980 * go over the max number of errors
5985 btrfs_end_bbio(bbio, bio);
5986 } else if (!is_orig_bio) {
5992 * see run_scheduled_bios for a description of why bios are collected for
5995 * This will add one bio to the pending list for a device and make sure
5996 * the work struct is scheduled.
5998 static noinline void btrfs_schedule_bio(struct btrfs_root *root,
5999 struct btrfs_device *device,
6000 int rw, struct bio *bio)
6002 int should_queue = 1;
6003 struct btrfs_pending_bios *pending_bios;
6005 if (device->missing || !device->bdev) {
6010 /* don't bother with additional async steps for reads, right now */
6011 if (!(rw & REQ_WRITE)) {
6013 btrfsic_submit_bio(rw, bio);
6019 * nr_async_bios allows us to reliably return congestion to the
6020 * higher layers. Otherwise, the async bio makes it appear we have
6021 * made progress against dirty pages when we've really just put it
6022 * on a queue for later
6024 atomic_inc(&root->fs_info->nr_async_bios);
6025 WARN_ON(bio->bi_next);
6026 bio->bi_next = NULL;
6029 spin_lock(&device->io_lock);
6030 if (bio->bi_rw & REQ_SYNC)
6031 pending_bios = &device->pending_sync_bios;
6033 pending_bios = &device->pending_bios;
6035 if (pending_bios->tail)
6036 pending_bios->tail->bi_next = bio;
6038 pending_bios->tail = bio;
6039 if (!pending_bios->head)
6040 pending_bios->head = bio;
6041 if (device->running_pending)
6044 spin_unlock(&device->io_lock);
6047 btrfs_queue_work(root->fs_info->submit_workers,
6051 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
6052 struct bio *bio, u64 physical, int dev_nr,
6055 struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
6057 bio->bi_private = bbio;
6058 btrfs_io_bio(bio)->stripe_index = dev_nr;
6059 bio->bi_end_io = btrfs_end_bio;
6060 bio->bi_iter.bi_sector = physical >> 9;
6063 struct rcu_string *name;
6066 name = rcu_dereference(dev->name);
6067 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
6068 "(%s id %llu), size=%u\n", rw,
6069 (u64)bio->bi_iter.bi_sector, (u_long)dev->bdev->bd_dev,
6070 name->str, dev->devid, bio->bi_iter.bi_size);
6074 bio->bi_bdev = dev->bdev;
6076 btrfs_bio_counter_inc_noblocked(root->fs_info);
6079 btrfs_schedule_bio(root, dev, rw, bio);
6081 btrfsic_submit_bio(rw, bio);
6084 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6086 atomic_inc(&bbio->error);
6087 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6088 /* Should be the original bio. */
6089 WARN_ON(bio != bbio->orig_bio);
6091 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6092 bio->bi_iter.bi_sector = logical >> 9;
6093 bio->bi_error = -EIO;
6094 btrfs_end_bbio(bbio, bio);
6098 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
6099 int mirror_num, int async_submit)
6101 struct btrfs_device *dev;
6102 struct bio *first_bio = bio;
6103 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
6109 struct btrfs_bio *bbio = NULL;
6111 length = bio->bi_iter.bi_size;
6112 map_length = length;
6114 btrfs_bio_counter_inc_blocked(root->fs_info);
6115 ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
6118 btrfs_bio_counter_dec(root->fs_info);
6122 total_devs = bbio->num_stripes;
6123 bbio->orig_bio = first_bio;
6124 bbio->private = first_bio->bi_private;
6125 bbio->end_io = first_bio->bi_end_io;
6126 bbio->fs_info = root->fs_info;
6127 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6129 if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6130 ((rw & WRITE) || (mirror_num > 1))) {
6131 /* In this case, map_length has been set to the length of
6132 a single stripe; not the whole write */
6134 ret = raid56_parity_write(root, bio, bbio, map_length);
6136 ret = raid56_parity_recover(root, bio, bbio, map_length,
6140 btrfs_bio_counter_dec(root->fs_info);
6144 if (map_length < length) {
6145 btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu",
6146 logical, length, map_length);
6150 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6151 dev = bbio->stripes[dev_nr].dev;
6152 if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
6153 bbio_error(bbio, first_bio, logical);
6157 if (dev_nr < total_devs - 1) {
6158 bio = btrfs_bio_clone(first_bio, GFP_NOFS);
6159 BUG_ON(!bio); /* -ENOMEM */
6163 submit_stripe_bio(root, bbio, bio,
6164 bbio->stripes[dev_nr].physical, dev_nr, rw,
6167 btrfs_bio_counter_dec(root->fs_info);
6171 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
6174 struct btrfs_device *device;
6175 struct btrfs_fs_devices *cur_devices;
6177 cur_devices = fs_info->fs_devices;
6178 while (cur_devices) {
6180 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
6181 device = __find_device(&cur_devices->devices,
6186 cur_devices = cur_devices->seed;
6191 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
6192 struct btrfs_fs_devices *fs_devices,
6193 u64 devid, u8 *dev_uuid)
6195 struct btrfs_device *device;
6197 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6201 list_add(&device->dev_list, &fs_devices->devices);
6202 device->fs_devices = fs_devices;
6203 fs_devices->num_devices++;
6205 device->missing = 1;
6206 fs_devices->missing_devices++;
6212 * btrfs_alloc_device - allocate struct btrfs_device
6213 * @fs_info: used only for generating a new devid, can be NULL if
6214 * devid is provided (i.e. @devid != NULL).
6215 * @devid: a pointer to devid for this device. If NULL a new devid
6217 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6220 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6221 * on error. Returned struct is not linked onto any lists and can be
6222 * destroyed with kfree() right away.
6224 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6228 struct btrfs_device *dev;
6231 if (WARN_ON(!devid && !fs_info))
6232 return ERR_PTR(-EINVAL);
6234 dev = __alloc_device();
6243 ret = find_next_devid(fs_info, &tmp);
6246 return ERR_PTR(ret);
6252 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6254 generate_random_uuid(dev->uuid);
6256 btrfs_init_work(&dev->work, btrfs_submit_helper,
6257 pending_bios_fn, NULL, NULL);
6262 /* Return -EIO if any error, otherwise return 0. */
6263 static int btrfs_check_chunk_valid(struct btrfs_root *root,
6264 struct extent_buffer *leaf,
6265 struct btrfs_chunk *chunk, u64 logical)
6273 length = btrfs_chunk_length(leaf, chunk);
6274 stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6275 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6276 sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6277 type = btrfs_chunk_type(leaf, chunk);
6280 btrfs_err(root->fs_info, "invalid chunk num_stripes: %u",
6284 if (!IS_ALIGNED(logical, root->sectorsize)) {
6285 btrfs_err(root->fs_info,
6286 "invalid chunk logical %llu", logical);
6289 if (btrfs_chunk_sector_size(leaf, chunk) != root->sectorsize) {
6290 btrfs_err(root->fs_info, "invalid chunk sectorsize %u",
6291 btrfs_chunk_sector_size(leaf, chunk));
6294 if (!length || !IS_ALIGNED(length, root->sectorsize)) {
6295 btrfs_err(root->fs_info,
6296 "invalid chunk length %llu", length);
6299 if (!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN) {
6300 btrfs_err(root->fs_info, "invalid chunk stripe length: %llu",
6304 if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6306 btrfs_err(root->fs_info, "unrecognized chunk type: %llu",
6307 ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
6308 BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6309 btrfs_chunk_type(leaf, chunk));
6312 if ((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes != 2) ||
6313 (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) ||
6314 (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) ||
6315 (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) ||
6316 (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) ||
6317 ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
6318 num_stripes != 1)) {
6319 btrfs_err(root->fs_info,
6320 "invalid num_stripes:sub_stripes %u:%u for profile %llu",
6321 num_stripes, sub_stripes,
6322 type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
6329 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
6330 struct extent_buffer *leaf,
6331 struct btrfs_chunk *chunk)
6333 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
6334 struct map_lookup *map;
6335 struct extent_map *em;
6340 u8 uuid[BTRFS_UUID_SIZE];
6345 logical = key->offset;
6346 length = btrfs_chunk_length(leaf, chunk);
6347 stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6348 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6350 ret = btrfs_check_chunk_valid(root, leaf, chunk, logical);
6354 read_lock(&map_tree->map_tree.lock);
6355 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
6356 read_unlock(&map_tree->map_tree.lock);
6358 /* already mapped? */
6359 if (em && em->start <= logical && em->start + em->len > logical) {
6360 free_extent_map(em);
6363 free_extent_map(em);
6366 em = alloc_extent_map();
6369 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6371 free_extent_map(em);
6375 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6376 em->map_lookup = map;
6377 em->start = logical;
6380 em->block_start = 0;
6381 em->block_len = em->len;
6383 map->num_stripes = num_stripes;
6384 map->io_width = btrfs_chunk_io_width(leaf, chunk);
6385 map->io_align = btrfs_chunk_io_align(leaf, chunk);
6386 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
6387 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6388 map->type = btrfs_chunk_type(leaf, chunk);
6389 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6390 for (i = 0; i < num_stripes; i++) {
6391 map->stripes[i].physical =
6392 btrfs_stripe_offset_nr(leaf, chunk, i);
6393 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6394 read_extent_buffer(leaf, uuid, (unsigned long)
6395 btrfs_stripe_dev_uuid_nr(chunk, i),
6397 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
6399 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
6400 free_extent_map(em);
6403 if (!map->stripes[i].dev) {
6404 map->stripes[i].dev =
6405 add_missing_dev(root, root->fs_info->fs_devices,
6407 if (!map->stripes[i].dev) {
6408 free_extent_map(em);
6411 btrfs_warn(root->fs_info, "devid %llu uuid %pU is missing",
6414 map->stripes[i].dev->in_fs_metadata = 1;
6417 write_lock(&map_tree->map_tree.lock);
6418 ret = add_extent_mapping(&map_tree->map_tree, em, 0);
6419 write_unlock(&map_tree->map_tree.lock);
6420 BUG_ON(ret); /* Tree corruption */
6421 free_extent_map(em);
6426 static void fill_device_from_item(struct extent_buffer *leaf,
6427 struct btrfs_dev_item *dev_item,
6428 struct btrfs_device *device)
6432 device->devid = btrfs_device_id(leaf, dev_item);
6433 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6434 device->total_bytes = device->disk_total_bytes;
6435 device->commit_total_bytes = device->disk_total_bytes;
6436 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6437 device->commit_bytes_used = device->bytes_used;
6438 device->type = btrfs_device_type(leaf, dev_item);
6439 device->io_align = btrfs_device_io_align(leaf, dev_item);
6440 device->io_width = btrfs_device_io_width(leaf, dev_item);
6441 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6442 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6443 device->is_tgtdev_for_dev_replace = 0;
6445 ptr = btrfs_device_uuid(dev_item);
6446 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6449 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_root *root,
6452 struct btrfs_fs_devices *fs_devices;
6455 BUG_ON(!mutex_is_locked(&uuid_mutex));
6457 fs_devices = root->fs_info->fs_devices->seed;
6458 while (fs_devices) {
6459 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE))
6462 fs_devices = fs_devices->seed;
6465 fs_devices = find_fsid(fsid);
6467 if (!btrfs_test_opt(root, DEGRADED))
6468 return ERR_PTR(-ENOENT);
6470 fs_devices = alloc_fs_devices(fsid);
6471 if (IS_ERR(fs_devices))
6474 fs_devices->seeding = 1;
6475 fs_devices->opened = 1;
6479 fs_devices = clone_fs_devices(fs_devices);
6480 if (IS_ERR(fs_devices))
6483 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
6484 root->fs_info->bdev_holder);
6486 free_fs_devices(fs_devices);
6487 fs_devices = ERR_PTR(ret);
6491 if (!fs_devices->seeding) {
6492 __btrfs_close_devices(fs_devices);
6493 free_fs_devices(fs_devices);
6494 fs_devices = ERR_PTR(-EINVAL);
6498 fs_devices->seed = root->fs_info->fs_devices->seed;
6499 root->fs_info->fs_devices->seed = fs_devices;
6504 static int read_one_dev(struct btrfs_root *root,
6505 struct extent_buffer *leaf,
6506 struct btrfs_dev_item *dev_item)
6508 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6509 struct btrfs_device *device;
6512 u8 fs_uuid[BTRFS_UUID_SIZE];
6513 u8 dev_uuid[BTRFS_UUID_SIZE];
6515 devid = btrfs_device_id(leaf, dev_item);
6516 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6518 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6521 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
6522 fs_devices = open_seed_devices(root, fs_uuid);
6523 if (IS_ERR(fs_devices))
6524 return PTR_ERR(fs_devices);
6527 device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
6529 if (!btrfs_test_opt(root, DEGRADED))
6532 device = add_missing_dev(root, fs_devices, devid, dev_uuid);
6535 btrfs_warn(root->fs_info, "devid %llu uuid %pU missing",
6538 if (!device->bdev && !btrfs_test_opt(root, DEGRADED))
6541 if(!device->bdev && !device->missing) {
6543 * this happens when a device that was properly setup
6544 * in the device info lists suddenly goes bad.
6545 * device->bdev is NULL, and so we have to set
6546 * device->missing to one here
6548 device->fs_devices->missing_devices++;
6549 device->missing = 1;
6552 /* Move the device to its own fs_devices */
6553 if (device->fs_devices != fs_devices) {
6554 ASSERT(device->missing);
6556 list_move(&device->dev_list, &fs_devices->devices);
6557 device->fs_devices->num_devices--;
6558 fs_devices->num_devices++;
6560 device->fs_devices->missing_devices--;
6561 fs_devices->missing_devices++;
6563 device->fs_devices = fs_devices;
6567 if (device->fs_devices != root->fs_info->fs_devices) {
6568 BUG_ON(device->writeable);
6569 if (device->generation !=
6570 btrfs_device_generation(leaf, dev_item))
6574 fill_device_from_item(leaf, dev_item, device);
6575 device->in_fs_metadata = 1;
6576 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
6577 device->fs_devices->total_rw_bytes += device->total_bytes;
6578 spin_lock(&root->fs_info->free_chunk_lock);
6579 root->fs_info->free_chunk_space += device->total_bytes -
6581 spin_unlock(&root->fs_info->free_chunk_lock);
6587 int btrfs_read_sys_array(struct btrfs_root *root)
6589 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
6590 struct extent_buffer *sb;
6591 struct btrfs_disk_key *disk_key;
6592 struct btrfs_chunk *chunk;
6594 unsigned long sb_array_offset;
6601 struct btrfs_key key;
6603 ASSERT(BTRFS_SUPER_INFO_SIZE <= root->nodesize);
6605 * This will create extent buffer of nodesize, superblock size is
6606 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6607 * overallocate but we can keep it as-is, only the first page is used.
6609 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET);
6612 set_extent_buffer_uptodate(sb);
6613 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6615 * The sb extent buffer is artificial and just used to read the system array.
6616 * set_extent_buffer_uptodate() call does not properly mark all it's
6617 * pages up-to-date when the page is larger: extent does not cover the
6618 * whole page and consequently check_page_uptodate does not find all
6619 * the page's extents up-to-date (the hole beyond sb),
6620 * write_extent_buffer then triggers a WARN_ON.
6622 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6623 * but sb spans only this function. Add an explicit SetPageUptodate call
6624 * to silence the warning eg. on PowerPC 64.
6626 if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
6627 SetPageUptodate(sb->pages[0]);
6629 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6630 array_size = btrfs_super_sys_array_size(super_copy);
6632 array_ptr = super_copy->sys_chunk_array;
6633 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6636 while (cur_offset < array_size) {
6637 disk_key = (struct btrfs_disk_key *)array_ptr;
6638 len = sizeof(*disk_key);
6639 if (cur_offset + len > array_size)
6640 goto out_short_read;
6642 btrfs_disk_key_to_cpu(&key, disk_key);
6645 sb_array_offset += len;
6648 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6649 chunk = (struct btrfs_chunk *)sb_array_offset;
6651 * At least one btrfs_chunk with one stripe must be
6652 * present, exact stripe count check comes afterwards
6654 len = btrfs_chunk_item_size(1);
6655 if (cur_offset + len > array_size)
6656 goto out_short_read;
6658 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6661 "BTRFS: invalid number of stripes %u in sys_array at offset %u\n",
6662 num_stripes, cur_offset);
6667 type = btrfs_chunk_type(sb, chunk);
6668 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
6669 btrfs_err(root->fs_info,
6670 "invalid chunk type %llu in sys_array at offset %u",
6676 len = btrfs_chunk_item_size(num_stripes);
6677 if (cur_offset + len > array_size)
6678 goto out_short_read;
6680 ret = read_one_chunk(root, &key, sb, chunk);
6685 "BTRFS: unexpected item type %u in sys_array at offset %u\n",
6686 (u32)key.type, cur_offset);
6691 sb_array_offset += len;
6694 clear_extent_buffer_uptodate(sb);
6695 free_extent_buffer_stale(sb);
6699 printk(KERN_ERR "BTRFS: sys_array too short to read %u bytes at offset %u\n",
6701 clear_extent_buffer_uptodate(sb);
6702 free_extent_buffer_stale(sb);
6706 int btrfs_read_chunk_tree(struct btrfs_root *root)
6708 struct btrfs_path *path;
6709 struct extent_buffer *leaf;
6710 struct btrfs_key key;
6711 struct btrfs_key found_key;
6716 root = root->fs_info->chunk_root;
6718 path = btrfs_alloc_path();
6722 mutex_lock(&uuid_mutex);
6726 * Read all device items, and then all the chunk items. All
6727 * device items are found before any chunk item (their object id
6728 * is smaller than the lowest possible object id for a chunk
6729 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6731 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6734 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6738 leaf = path->nodes[0];
6739 slot = path->slots[0];
6740 if (slot >= btrfs_header_nritems(leaf)) {
6741 ret = btrfs_next_leaf(root, path);
6748 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6749 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6750 struct btrfs_dev_item *dev_item;
6751 dev_item = btrfs_item_ptr(leaf, slot,
6752 struct btrfs_dev_item);
6753 ret = read_one_dev(root, leaf, dev_item);
6757 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6758 struct btrfs_chunk *chunk;
6759 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6760 ret = read_one_chunk(root, &found_key, leaf, chunk);
6768 * After loading chunk tree, we've got all device information,
6769 * do another round of validation checks.
6771 if (total_dev != root->fs_info->fs_devices->total_devices) {
6772 btrfs_err(root->fs_info,
6773 "super_num_devices %llu mismatch with num_devices %llu found here",
6774 btrfs_super_num_devices(root->fs_info->super_copy),
6779 if (btrfs_super_total_bytes(root->fs_info->super_copy) <
6780 root->fs_info->fs_devices->total_rw_bytes) {
6781 btrfs_err(root->fs_info,
6782 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
6783 btrfs_super_total_bytes(root->fs_info->super_copy),
6784 root->fs_info->fs_devices->total_rw_bytes);
6790 unlock_chunks(root);
6791 mutex_unlock(&uuid_mutex);
6793 btrfs_free_path(path);
6797 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6799 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6800 struct btrfs_device *device;
6802 while (fs_devices) {
6803 mutex_lock(&fs_devices->device_list_mutex);
6804 list_for_each_entry(device, &fs_devices->devices, dev_list)
6805 device->dev_root = fs_info->dev_root;
6806 mutex_unlock(&fs_devices->device_list_mutex);
6808 fs_devices = fs_devices->seed;
6812 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6816 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6817 btrfs_dev_stat_reset(dev, i);
6820 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6822 struct btrfs_key key;
6823 struct btrfs_key found_key;
6824 struct btrfs_root *dev_root = fs_info->dev_root;
6825 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6826 struct extent_buffer *eb;
6829 struct btrfs_device *device;
6830 struct btrfs_path *path = NULL;
6833 path = btrfs_alloc_path();
6839 mutex_lock(&fs_devices->device_list_mutex);
6840 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6842 struct btrfs_dev_stats_item *ptr;
6844 key.objectid = BTRFS_DEV_STATS_OBJECTID;
6845 key.type = BTRFS_PERSISTENT_ITEM_KEY;
6846 key.offset = device->devid;
6847 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6849 __btrfs_reset_dev_stats(device);
6850 device->dev_stats_valid = 1;
6851 btrfs_release_path(path);
6854 slot = path->slots[0];
6855 eb = path->nodes[0];
6856 btrfs_item_key_to_cpu(eb, &found_key, slot);
6857 item_size = btrfs_item_size_nr(eb, slot);
6859 ptr = btrfs_item_ptr(eb, slot,
6860 struct btrfs_dev_stats_item);
6862 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6863 if (item_size >= (1 + i) * sizeof(__le64))
6864 btrfs_dev_stat_set(device, i,
6865 btrfs_dev_stats_value(eb, ptr, i));
6867 btrfs_dev_stat_reset(device, i);
6870 device->dev_stats_valid = 1;
6871 btrfs_dev_stat_print_on_load(device);
6872 btrfs_release_path(path);
6874 mutex_unlock(&fs_devices->device_list_mutex);
6877 btrfs_free_path(path);
6878 return ret < 0 ? ret : 0;
6881 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
6882 struct btrfs_root *dev_root,
6883 struct btrfs_device *device)
6885 struct btrfs_path *path;
6886 struct btrfs_key key;
6887 struct extent_buffer *eb;
6888 struct btrfs_dev_stats_item *ptr;
6892 key.objectid = BTRFS_DEV_STATS_OBJECTID;
6893 key.type = BTRFS_PERSISTENT_ITEM_KEY;
6894 key.offset = device->devid;
6896 path = btrfs_alloc_path();
6898 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
6900 btrfs_warn_in_rcu(dev_root->fs_info,
6901 "error %d while searching for dev_stats item for device %s",
6902 ret, rcu_str_deref(device->name));
6907 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
6908 /* need to delete old one and insert a new one */
6909 ret = btrfs_del_item(trans, dev_root, path);
6911 btrfs_warn_in_rcu(dev_root->fs_info,
6912 "delete too small dev_stats item for device %s failed %d",
6913 rcu_str_deref(device->name), ret);
6920 /* need to insert a new item */
6921 btrfs_release_path(path);
6922 ret = btrfs_insert_empty_item(trans, dev_root, path,
6923 &key, sizeof(*ptr));
6925 btrfs_warn_in_rcu(dev_root->fs_info,
6926 "insert dev_stats item for device %s failed %d",
6927 rcu_str_deref(device->name), ret);
6932 eb = path->nodes[0];
6933 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
6934 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6935 btrfs_set_dev_stats_value(eb, ptr, i,
6936 btrfs_dev_stat_read(device, i));
6937 btrfs_mark_buffer_dirty(eb);
6940 btrfs_free_path(path);
6945 * called from commit_transaction. Writes all changed device stats to disk.
6947 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
6948 struct btrfs_fs_info *fs_info)
6950 struct btrfs_root *dev_root = fs_info->dev_root;
6951 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6952 struct btrfs_device *device;
6956 mutex_lock(&fs_devices->device_list_mutex);
6957 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6958 if (!device->dev_stats_valid || !btrfs_dev_stats_dirty(device))
6961 stats_cnt = atomic_read(&device->dev_stats_ccnt);
6962 ret = update_dev_stat_item(trans, dev_root, device);
6964 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
6966 mutex_unlock(&fs_devices->device_list_mutex);
6971 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
6973 btrfs_dev_stat_inc(dev, index);
6974 btrfs_dev_stat_print_on_error(dev);
6977 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
6979 if (!dev->dev_stats_valid)
6981 btrfs_err_rl_in_rcu(dev->dev_root->fs_info,
6982 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
6983 rcu_str_deref(dev->name),
6984 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6985 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6986 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6987 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6988 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6991 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
6995 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6996 if (btrfs_dev_stat_read(dev, i) != 0)
6998 if (i == BTRFS_DEV_STAT_VALUES_MAX)
6999 return; /* all values == 0, suppress message */
7001 btrfs_info_in_rcu(dev->dev_root->fs_info,
7002 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7003 rcu_str_deref(dev->name),
7004 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7005 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7006 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7007 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7008 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7011 int btrfs_get_dev_stats(struct btrfs_root *root,
7012 struct btrfs_ioctl_get_dev_stats *stats)
7014 struct btrfs_device *dev;
7015 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7018 mutex_lock(&fs_devices->device_list_mutex);
7019 dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
7020 mutex_unlock(&fs_devices->device_list_mutex);
7023 btrfs_warn(root->fs_info, "get dev_stats failed, device not found");
7025 } else if (!dev->dev_stats_valid) {
7026 btrfs_warn(root->fs_info, "get dev_stats failed, not yet valid");
7028 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7029 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7030 if (stats->nr_items > i)
7032 btrfs_dev_stat_read_and_reset(dev, i);
7034 btrfs_dev_stat_reset(dev, i);
7037 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7038 if (stats->nr_items > i)
7039 stats->values[i] = btrfs_dev_stat_read(dev, i);
7041 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7042 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7046 void btrfs_scratch_superblocks(struct block_device *bdev, char *device_path)
7048 struct buffer_head *bh;
7049 struct btrfs_super_block *disk_super;
7055 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX;
7058 if (btrfs_read_dev_one_super(bdev, copy_num, &bh))
7061 disk_super = (struct btrfs_super_block *)bh->b_data;
7063 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
7064 set_buffer_dirty(bh);
7065 sync_dirty_buffer(bh);
7069 /* Notify udev that device has changed */
7070 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
7072 /* Update ctime/mtime for device path for libblkid */
7073 update_dev_time(device_path);
7077 * Update the size of all devices, which is used for writing out the
7080 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
7082 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7083 struct btrfs_device *curr, *next;
7085 if (list_empty(&fs_devices->resized_devices))
7088 mutex_lock(&fs_devices->device_list_mutex);
7089 lock_chunks(fs_info->dev_root);
7090 list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
7092 list_del_init(&curr->resized_list);
7093 curr->commit_total_bytes = curr->disk_total_bytes;
7095 unlock_chunks(fs_info->dev_root);
7096 mutex_unlock(&fs_devices->device_list_mutex);
7099 /* Must be invoked during the transaction commit */
7100 void btrfs_update_commit_device_bytes_used(struct btrfs_root *root,
7101 struct btrfs_transaction *transaction)
7103 struct extent_map *em;
7104 struct map_lookup *map;
7105 struct btrfs_device *dev;
7108 if (list_empty(&transaction->pending_chunks))
7111 /* In order to kick the device replace finish process */
7113 list_for_each_entry(em, &transaction->pending_chunks, list) {
7114 map = em->map_lookup;
7116 for (i = 0; i < map->num_stripes; i++) {
7117 dev = map->stripes[i].dev;
7118 dev->commit_bytes_used = dev->bytes_used;
7121 unlock_chunks(root);
7124 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
7126 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7127 while (fs_devices) {
7128 fs_devices->fs_info = fs_info;
7129 fs_devices = fs_devices->seed;
7133 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
7135 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7136 while (fs_devices) {
7137 fs_devices->fs_info = NULL;
7138 fs_devices = fs_devices->seed;
7142 static void btrfs_close_one_device(struct btrfs_device *device)
7144 struct btrfs_fs_devices *fs_devices = device->fs_devices;
7145 struct btrfs_device *new_device;
7146 struct rcu_string *name;
7149 fs_devices->open_devices--;
7151 if (device->writeable &&
7152 device->devid != BTRFS_DEV_REPLACE_DEVID) {
7153 list_del_init(&device->dev_alloc_list);
7154 fs_devices->rw_devices--;
7157 if (device->missing)
7158 fs_devices->missing_devices--;
7160 new_device = btrfs_alloc_device(NULL, &device->devid,
7162 BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
7164 /* Safe because we are under uuid_mutex */
7166 name = rcu_string_strdup(device->name->str, GFP_NOFS);
7167 BUG_ON(!name); /* -ENOMEM */
7168 rcu_assign_pointer(new_device->name, name);
7171 list_replace_rcu(&device->dev_list, &new_device->dev_list);
7172 new_device->fs_devices = device->fs_devices;
7174 call_rcu(&device->rcu, free_device);