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_fs_info *fs_info);
138 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info);
139 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
140 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
141 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
142 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
143 enum btrfs_map_op op,
144 u64 logical, u64 *length,
145 struct btrfs_bio **bbio_ret,
146 int mirror_num, int need_raid_map);
148 DEFINE_MUTEX(uuid_mutex);
149 static LIST_HEAD(fs_uuids);
150 struct list_head *btrfs_get_fs_uuids(void)
155 static struct btrfs_fs_devices *__alloc_fs_devices(void)
157 struct btrfs_fs_devices *fs_devs;
159 fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL);
161 return ERR_PTR(-ENOMEM);
163 mutex_init(&fs_devs->device_list_mutex);
165 INIT_LIST_HEAD(&fs_devs->devices);
166 INIT_LIST_HEAD(&fs_devs->resized_devices);
167 INIT_LIST_HEAD(&fs_devs->alloc_list);
168 INIT_LIST_HEAD(&fs_devs->list);
174 * alloc_fs_devices - allocate struct btrfs_fs_devices
175 * @fsid: a pointer to UUID for this FS. If NULL a new UUID is
178 * Return: a pointer to a new &struct btrfs_fs_devices on success;
179 * ERR_PTR() on error. Returned struct is not linked onto any lists and
180 * can be destroyed with kfree() right away.
182 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
184 struct btrfs_fs_devices *fs_devs;
186 fs_devs = __alloc_fs_devices();
191 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
193 generate_random_uuid(fs_devs->fsid);
198 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
200 struct btrfs_device *device;
201 WARN_ON(fs_devices->opened);
202 while (!list_empty(&fs_devices->devices)) {
203 device = list_entry(fs_devices->devices.next,
204 struct btrfs_device, dev_list);
205 list_del(&device->dev_list);
206 rcu_string_free(device->name);
212 static void btrfs_kobject_uevent(struct block_device *bdev,
213 enum kobject_action action)
217 ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
219 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
221 kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
222 &disk_to_dev(bdev->bd_disk)->kobj);
225 void btrfs_cleanup_fs_uuids(void)
227 struct btrfs_fs_devices *fs_devices;
229 while (!list_empty(&fs_uuids)) {
230 fs_devices = list_entry(fs_uuids.next,
231 struct btrfs_fs_devices, list);
232 list_del(&fs_devices->list);
233 free_fs_devices(fs_devices);
237 static struct btrfs_device *__alloc_device(void)
239 struct btrfs_device *dev;
241 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
243 return ERR_PTR(-ENOMEM);
245 INIT_LIST_HEAD(&dev->dev_list);
246 INIT_LIST_HEAD(&dev->dev_alloc_list);
247 INIT_LIST_HEAD(&dev->resized_list);
249 spin_lock_init(&dev->io_lock);
251 spin_lock_init(&dev->reada_lock);
252 atomic_set(&dev->reada_in_flight, 0);
253 atomic_set(&dev->dev_stats_ccnt, 0);
254 btrfs_device_data_ordered_init(dev);
255 INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
256 INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
261 static noinline struct btrfs_device *__find_device(struct list_head *head,
264 struct btrfs_device *dev;
266 list_for_each_entry(dev, head, dev_list) {
267 if (dev->devid == devid &&
268 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
275 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
277 struct btrfs_fs_devices *fs_devices;
279 list_for_each_entry(fs_devices, &fs_uuids, list) {
280 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
287 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
288 int flush, struct block_device **bdev,
289 struct buffer_head **bh)
293 *bdev = blkdev_get_by_path(device_path, flags, holder);
296 ret = PTR_ERR(*bdev);
301 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
302 ret = set_blocksize(*bdev, 4096);
304 blkdev_put(*bdev, flags);
307 invalidate_bdev(*bdev);
308 *bh = btrfs_read_dev_super(*bdev);
311 blkdev_put(*bdev, flags);
323 static void requeue_list(struct btrfs_pending_bios *pending_bios,
324 struct bio *head, struct bio *tail)
327 struct bio *old_head;
329 old_head = pending_bios->head;
330 pending_bios->head = head;
331 if (pending_bios->tail)
332 tail->bi_next = old_head;
334 pending_bios->tail = tail;
338 * we try to collect pending bios for a device so we don't get a large
339 * number of procs sending bios down to the same device. This greatly
340 * improves the schedulers ability to collect and merge the bios.
342 * But, it also turns into a long list of bios to process and that is sure
343 * to eventually make the worker thread block. The solution here is to
344 * make some progress and then put this work struct back at the end of
345 * the list if the block device is congested. This way, multiple devices
346 * can make progress from a single worker thread.
348 static noinline void run_scheduled_bios(struct btrfs_device *device)
350 struct btrfs_fs_info *fs_info = device->fs_info;
352 struct backing_dev_info *bdi;
353 struct btrfs_pending_bios *pending_bios;
357 unsigned long num_run;
358 unsigned long batch_run = 0;
360 unsigned long last_waited = 0;
362 int sync_pending = 0;
363 struct blk_plug plug;
366 * this function runs all the bios we've collected for
367 * a particular device. We don't want to wander off to
368 * another device without first sending all of these down.
369 * So, setup a plug here and finish it off before we return
371 blk_start_plug(&plug);
373 bdi = device->bdev->bd_bdi;
374 limit = btrfs_async_submit_limit(fs_info);
375 limit = limit * 2 / 3;
378 spin_lock(&device->io_lock);
383 /* take all the bios off the list at once and process them
384 * later on (without the lock held). But, remember the
385 * tail and other pointers so the bios can be properly reinserted
386 * into the list if we hit congestion
388 if (!force_reg && device->pending_sync_bios.head) {
389 pending_bios = &device->pending_sync_bios;
392 pending_bios = &device->pending_bios;
396 pending = pending_bios->head;
397 tail = pending_bios->tail;
398 WARN_ON(pending && !tail);
401 * if pending was null this time around, no bios need processing
402 * at all and we can stop. Otherwise it'll loop back up again
403 * and do an additional check so no bios are missed.
405 * device->running_pending is used to synchronize with the
408 if (device->pending_sync_bios.head == NULL &&
409 device->pending_bios.head == NULL) {
411 device->running_pending = 0;
414 device->running_pending = 1;
417 pending_bios->head = NULL;
418 pending_bios->tail = NULL;
420 spin_unlock(&device->io_lock);
425 /* we want to work on both lists, but do more bios on the
426 * sync list than the regular list
429 pending_bios != &device->pending_sync_bios &&
430 device->pending_sync_bios.head) ||
431 (num_run > 64 && pending_bios == &device->pending_sync_bios &&
432 device->pending_bios.head)) {
433 spin_lock(&device->io_lock);
434 requeue_list(pending_bios, pending, tail);
439 pending = pending->bi_next;
443 * atomic_dec_return implies a barrier for waitqueue_active
445 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
446 waitqueue_active(&fs_info->async_submit_wait))
447 wake_up(&fs_info->async_submit_wait);
449 BUG_ON(atomic_read(&cur->__bi_cnt) == 0);
452 * if we're doing the sync list, record that our
453 * plug has some sync requests on it
455 * If we're doing the regular list and there are
456 * sync requests sitting around, unplug before
459 if (pending_bios == &device->pending_sync_bios) {
461 } else if (sync_pending) {
462 blk_finish_plug(&plug);
463 blk_start_plug(&plug);
467 btrfsic_submit_bio(cur);
474 * we made progress, there is more work to do and the bdi
475 * is now congested. Back off and let other work structs
478 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
479 fs_info->fs_devices->open_devices > 1) {
480 struct io_context *ioc;
482 ioc = current->io_context;
485 * the main goal here is that we don't want to
486 * block if we're going to be able to submit
487 * more requests without blocking.
489 * This code does two great things, it pokes into
490 * the elevator code from a filesystem _and_
491 * it makes assumptions about how batching works.
493 if (ioc && ioc->nr_batch_requests > 0 &&
494 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
496 ioc->last_waited == last_waited)) {
498 * we want to go through our batch of
499 * requests and stop. So, we copy out
500 * the ioc->last_waited time and test
501 * against it before looping
503 last_waited = ioc->last_waited;
507 spin_lock(&device->io_lock);
508 requeue_list(pending_bios, pending, tail);
509 device->running_pending = 1;
511 spin_unlock(&device->io_lock);
512 btrfs_queue_work(fs_info->submit_workers,
516 /* unplug every 64 requests just for good measure */
517 if (batch_run % 64 == 0) {
518 blk_finish_plug(&plug);
519 blk_start_plug(&plug);
528 spin_lock(&device->io_lock);
529 if (device->pending_bios.head || device->pending_sync_bios.head)
531 spin_unlock(&device->io_lock);
534 blk_finish_plug(&plug);
537 static void pending_bios_fn(struct btrfs_work *work)
539 struct btrfs_device *device;
541 device = container_of(work, struct btrfs_device, work);
542 run_scheduled_bios(device);
546 void btrfs_free_stale_device(struct btrfs_device *cur_dev)
548 struct btrfs_fs_devices *fs_devs;
549 struct btrfs_device *dev;
554 list_for_each_entry(fs_devs, &fs_uuids, list) {
559 if (fs_devs->seeding)
562 list_for_each_entry(dev, &fs_devs->devices, dev_list) {
570 * Todo: This won't be enough. What if the same device
571 * comes back (with new uuid and) with its mapper path?
572 * But for now, this does help as mostly an admin will
573 * either use mapper or non mapper path throughout.
576 del = strcmp(rcu_str_deref(dev->name),
577 rcu_str_deref(cur_dev->name));
584 /* delete the stale device */
585 if (fs_devs->num_devices == 1) {
586 btrfs_sysfs_remove_fsid(fs_devs);
587 list_del(&fs_devs->list);
588 free_fs_devices(fs_devs);
590 fs_devs->num_devices--;
591 list_del(&dev->dev_list);
592 rcu_string_free(dev->name);
601 * Add new device to list of registered devices
604 * 1 - first time device is seen
605 * 0 - device already known
608 static noinline int device_list_add(const char *path,
609 struct btrfs_super_block *disk_super,
610 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
612 struct btrfs_device *device;
613 struct btrfs_fs_devices *fs_devices;
614 struct rcu_string *name;
616 u64 found_transid = btrfs_super_generation(disk_super);
618 fs_devices = find_fsid(disk_super->fsid);
620 fs_devices = alloc_fs_devices(disk_super->fsid);
621 if (IS_ERR(fs_devices))
622 return PTR_ERR(fs_devices);
624 list_add(&fs_devices->list, &fs_uuids);
628 device = __find_device(&fs_devices->devices, devid,
629 disk_super->dev_item.uuid);
633 if (fs_devices->opened)
636 device = btrfs_alloc_device(NULL, &devid,
637 disk_super->dev_item.uuid);
638 if (IS_ERR(device)) {
639 /* we can safely leave the fs_devices entry around */
640 return PTR_ERR(device);
643 name = rcu_string_strdup(path, GFP_NOFS);
648 rcu_assign_pointer(device->name, name);
650 mutex_lock(&fs_devices->device_list_mutex);
651 list_add_rcu(&device->dev_list, &fs_devices->devices);
652 fs_devices->num_devices++;
653 mutex_unlock(&fs_devices->device_list_mutex);
656 device->fs_devices = fs_devices;
657 } else if (!device->name || strcmp(device->name->str, path)) {
659 * When FS is already mounted.
660 * 1. If you are here and if the device->name is NULL that
661 * means this device was missing at time of FS mount.
662 * 2. If you are here and if the device->name is different
663 * from 'path' that means either
664 * a. The same device disappeared and reappeared with
666 * b. The missing-disk-which-was-replaced, has
669 * We must allow 1 and 2a above. But 2b would be a spurious
672 * Further in case of 1 and 2a above, the disk at 'path'
673 * would have missed some transaction when it was away and
674 * in case of 2a the stale bdev has to be updated as well.
675 * 2b must not be allowed at all time.
679 * For now, we do allow update to btrfs_fs_device through the
680 * btrfs dev scan cli after FS has been mounted. We're still
681 * tracking a problem where systems fail mount by subvolume id
682 * when we reject replacement on a mounted FS.
684 if (!fs_devices->opened && found_transid < device->generation) {
686 * That is if the FS is _not_ mounted and if you
687 * are here, that means there is more than one
688 * disk with same uuid and devid.We keep the one
689 * with larger generation number or the last-in if
690 * generation are equal.
695 name = rcu_string_strdup(path, GFP_NOFS);
698 rcu_string_free(device->name);
699 rcu_assign_pointer(device->name, name);
700 if (device->missing) {
701 fs_devices->missing_devices--;
707 * Unmount does not free the btrfs_device struct but would zero
708 * generation along with most of the other members. So just update
709 * it back. We need it to pick the disk with largest generation
712 if (!fs_devices->opened)
713 device->generation = found_transid;
716 * if there is new btrfs on an already registered device,
717 * then remove the stale device entry.
720 btrfs_free_stale_device(device);
722 *fs_devices_ret = fs_devices;
727 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
729 struct btrfs_fs_devices *fs_devices;
730 struct btrfs_device *device;
731 struct btrfs_device *orig_dev;
733 fs_devices = alloc_fs_devices(orig->fsid);
734 if (IS_ERR(fs_devices))
737 mutex_lock(&orig->device_list_mutex);
738 fs_devices->total_devices = orig->total_devices;
740 /* We have held the volume lock, it is safe to get the devices. */
741 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
742 struct rcu_string *name;
744 device = btrfs_alloc_device(NULL, &orig_dev->devid,
750 * This is ok to do without rcu read locked because we hold the
751 * uuid mutex so nothing we touch in here is going to disappear.
753 if (orig_dev->name) {
754 name = rcu_string_strdup(orig_dev->name->str,
760 rcu_assign_pointer(device->name, name);
763 list_add(&device->dev_list, &fs_devices->devices);
764 device->fs_devices = fs_devices;
765 fs_devices->num_devices++;
767 mutex_unlock(&orig->device_list_mutex);
770 mutex_unlock(&orig->device_list_mutex);
771 free_fs_devices(fs_devices);
772 return ERR_PTR(-ENOMEM);
775 void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step)
777 struct btrfs_device *device, *next;
778 struct btrfs_device *latest_dev = NULL;
780 mutex_lock(&uuid_mutex);
782 /* This is the initialized path, it is safe to release the devices. */
783 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
784 if (device->in_fs_metadata) {
785 if (!device->is_tgtdev_for_dev_replace &&
787 device->generation > latest_dev->generation)) {
793 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
795 * In the first step, keep the device which has
796 * the correct fsid and the devid that is used
797 * for the dev_replace procedure.
798 * In the second step, the dev_replace state is
799 * read from the device tree and it is known
800 * whether the procedure is really active or
801 * not, which means whether this device is
802 * used or whether it should be removed.
804 if (step == 0 || device->is_tgtdev_for_dev_replace) {
809 blkdev_put(device->bdev, device->mode);
811 fs_devices->open_devices--;
813 if (device->writeable) {
814 list_del_init(&device->dev_alloc_list);
815 device->writeable = 0;
816 if (!device->is_tgtdev_for_dev_replace)
817 fs_devices->rw_devices--;
819 list_del_init(&device->dev_list);
820 fs_devices->num_devices--;
821 rcu_string_free(device->name);
825 if (fs_devices->seed) {
826 fs_devices = fs_devices->seed;
830 fs_devices->latest_bdev = latest_dev->bdev;
832 mutex_unlock(&uuid_mutex);
835 static void __free_device(struct work_struct *work)
837 struct btrfs_device *device;
839 device = container_of(work, struct btrfs_device, rcu_work);
840 rcu_string_free(device->name);
844 static void free_device(struct rcu_head *head)
846 struct btrfs_device *device;
848 device = container_of(head, struct btrfs_device, rcu);
850 INIT_WORK(&device->rcu_work, __free_device);
851 schedule_work(&device->rcu_work);
854 static void btrfs_close_bdev(struct btrfs_device *device)
856 if (device->bdev && device->writeable) {
857 sync_blockdev(device->bdev);
858 invalidate_bdev(device->bdev);
862 blkdev_put(device->bdev, device->mode);
865 static void btrfs_prepare_close_one_device(struct btrfs_device *device)
867 struct btrfs_fs_devices *fs_devices = device->fs_devices;
868 struct btrfs_device *new_device;
869 struct rcu_string *name;
872 fs_devices->open_devices--;
874 if (device->writeable &&
875 device->devid != BTRFS_DEV_REPLACE_DEVID) {
876 list_del_init(&device->dev_alloc_list);
877 fs_devices->rw_devices--;
881 fs_devices->missing_devices--;
883 new_device = btrfs_alloc_device(NULL, &device->devid,
885 BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
887 /* Safe because we are under uuid_mutex */
889 name = rcu_string_strdup(device->name->str, GFP_NOFS);
890 BUG_ON(!name); /* -ENOMEM */
891 rcu_assign_pointer(new_device->name, name);
894 list_replace_rcu(&device->dev_list, &new_device->dev_list);
895 new_device->fs_devices = device->fs_devices;
898 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
900 struct btrfs_device *device, *tmp;
901 struct list_head pending_put;
903 INIT_LIST_HEAD(&pending_put);
905 if (--fs_devices->opened > 0)
908 mutex_lock(&fs_devices->device_list_mutex);
909 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
910 btrfs_prepare_close_one_device(device);
911 list_add(&device->dev_list, &pending_put);
913 mutex_unlock(&fs_devices->device_list_mutex);
916 * btrfs_show_devname() is using the device_list_mutex,
917 * sometimes call to blkdev_put() leads vfs calling
918 * into this func. So do put outside of device_list_mutex,
921 while (!list_empty(&pending_put)) {
922 device = list_first_entry(&pending_put,
923 struct btrfs_device, dev_list);
924 list_del(&device->dev_list);
925 btrfs_close_bdev(device);
926 call_rcu(&device->rcu, free_device);
929 WARN_ON(fs_devices->open_devices);
930 WARN_ON(fs_devices->rw_devices);
931 fs_devices->opened = 0;
932 fs_devices->seeding = 0;
937 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
939 struct btrfs_fs_devices *seed_devices = NULL;
942 mutex_lock(&uuid_mutex);
943 ret = __btrfs_close_devices(fs_devices);
944 if (!fs_devices->opened) {
945 seed_devices = fs_devices->seed;
946 fs_devices->seed = NULL;
948 mutex_unlock(&uuid_mutex);
950 while (seed_devices) {
951 fs_devices = seed_devices;
952 seed_devices = fs_devices->seed;
953 __btrfs_close_devices(fs_devices);
954 free_fs_devices(fs_devices);
957 * Wait for rcu kworkers under __btrfs_close_devices
958 * to finish all blkdev_puts so device is really
959 * free when umount is done.
965 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
966 fmode_t flags, void *holder)
968 struct request_queue *q;
969 struct block_device *bdev;
970 struct list_head *head = &fs_devices->devices;
971 struct btrfs_device *device;
972 struct btrfs_device *latest_dev = NULL;
973 struct buffer_head *bh;
974 struct btrfs_super_block *disk_super;
981 list_for_each_entry(device, head, dev_list) {
987 /* Just open everything we can; ignore failures here */
988 if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
992 disk_super = (struct btrfs_super_block *)bh->b_data;
993 devid = btrfs_stack_device_id(&disk_super->dev_item);
994 if (devid != device->devid)
997 if (memcmp(device->uuid, disk_super->dev_item.uuid,
1001 device->generation = btrfs_super_generation(disk_super);
1003 device->generation > latest_dev->generation)
1004 latest_dev = device;
1006 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
1007 device->writeable = 0;
1009 device->writeable = !bdev_read_only(bdev);
1013 q = bdev_get_queue(bdev);
1014 if (blk_queue_discard(q))
1015 device->can_discard = 1;
1016 if (!blk_queue_nonrot(q))
1017 fs_devices->rotating = 1;
1019 device->bdev = bdev;
1020 device->in_fs_metadata = 0;
1021 device->mode = flags;
1023 fs_devices->open_devices++;
1024 if (device->writeable &&
1025 device->devid != BTRFS_DEV_REPLACE_DEVID) {
1026 fs_devices->rw_devices++;
1027 list_add(&device->dev_alloc_list,
1028 &fs_devices->alloc_list);
1035 blkdev_put(bdev, flags);
1038 if (fs_devices->open_devices == 0) {
1042 fs_devices->seeding = seeding;
1043 fs_devices->opened = 1;
1044 fs_devices->latest_bdev = latest_dev->bdev;
1045 fs_devices->total_rw_bytes = 0;
1050 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1051 fmode_t flags, void *holder)
1055 mutex_lock(&uuid_mutex);
1056 if (fs_devices->opened) {
1057 fs_devices->opened++;
1060 ret = __btrfs_open_devices(fs_devices, flags, holder);
1062 mutex_unlock(&uuid_mutex);
1066 void btrfs_release_disk_super(struct page *page)
1072 int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr,
1073 struct page **page, struct btrfs_super_block **disk_super)
1078 /* make sure our super fits in the device */
1079 if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1082 /* make sure our super fits in the page */
1083 if (sizeof(**disk_super) > PAGE_SIZE)
1086 /* make sure our super doesn't straddle pages on disk */
1087 index = bytenr >> PAGE_SHIFT;
1088 if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index)
1091 /* pull in the page with our super */
1092 *page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
1095 if (IS_ERR_OR_NULL(*page))
1100 /* align our pointer to the offset of the super block */
1101 *disk_super = p + (bytenr & ~PAGE_MASK);
1103 if (btrfs_super_bytenr(*disk_super) != bytenr ||
1104 btrfs_super_magic(*disk_super) != BTRFS_MAGIC) {
1105 btrfs_release_disk_super(*page);
1109 if ((*disk_super)->label[0] &&
1110 (*disk_super)->label[BTRFS_LABEL_SIZE - 1])
1111 (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0';
1117 * Look for a btrfs signature on a device. This may be called out of the mount path
1118 * and we are not allowed to call set_blocksize during the scan. The superblock
1119 * is read via pagecache
1121 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
1122 struct btrfs_fs_devices **fs_devices_ret)
1124 struct btrfs_super_block *disk_super;
1125 struct block_device *bdev;
1134 * we would like to check all the supers, but that would make
1135 * a btrfs mount succeed after a mkfs from a different FS.
1136 * So, we need to add a special mount option to scan for
1137 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1139 bytenr = btrfs_sb_offset(0);
1140 flags |= FMODE_EXCL;
1141 mutex_lock(&uuid_mutex);
1143 bdev = blkdev_get_by_path(path, flags, holder);
1145 ret = PTR_ERR(bdev);
1149 if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super))
1150 goto error_bdev_put;
1152 devid = btrfs_stack_device_id(&disk_super->dev_item);
1153 transid = btrfs_super_generation(disk_super);
1154 total_devices = btrfs_super_num_devices(disk_super);
1156 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
1158 if (disk_super->label[0]) {
1159 pr_info("BTRFS: device label %s ", disk_super->label);
1161 pr_info("BTRFS: device fsid %pU ", disk_super->fsid);
1164 pr_cont("devid %llu transid %llu %s\n", devid, transid, path);
1167 if (!ret && fs_devices_ret)
1168 (*fs_devices_ret)->total_devices = total_devices;
1170 btrfs_release_disk_super(page);
1173 blkdev_put(bdev, flags);
1175 mutex_unlock(&uuid_mutex);
1179 /* helper to account the used device space in the range */
1180 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
1181 u64 end, u64 *length)
1183 struct btrfs_key key;
1184 struct btrfs_root *root = device->fs_info->dev_root;
1185 struct btrfs_dev_extent *dev_extent;
1186 struct btrfs_path *path;
1190 struct extent_buffer *l;
1194 if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
1197 path = btrfs_alloc_path();
1200 path->reada = READA_FORWARD;
1202 key.objectid = device->devid;
1204 key.type = BTRFS_DEV_EXTENT_KEY;
1206 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1210 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1217 slot = path->slots[0];
1218 if (slot >= btrfs_header_nritems(l)) {
1219 ret = btrfs_next_leaf(root, path);
1227 btrfs_item_key_to_cpu(l, &key, slot);
1229 if (key.objectid < device->devid)
1232 if (key.objectid > device->devid)
1235 if (key.type != BTRFS_DEV_EXTENT_KEY)
1238 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1239 extent_end = key.offset + btrfs_dev_extent_length(l,
1241 if (key.offset <= start && extent_end > end) {
1242 *length = end - start + 1;
1244 } else if (key.offset <= start && extent_end > start)
1245 *length += extent_end - start;
1246 else if (key.offset > start && extent_end <= end)
1247 *length += extent_end - key.offset;
1248 else if (key.offset > start && key.offset <= end) {
1249 *length += end - key.offset + 1;
1251 } else if (key.offset > end)
1259 btrfs_free_path(path);
1263 static int contains_pending_extent(struct btrfs_transaction *transaction,
1264 struct btrfs_device *device,
1265 u64 *start, u64 len)
1267 struct btrfs_fs_info *fs_info = device->fs_info;
1268 struct extent_map *em;
1269 struct list_head *search_list = &fs_info->pinned_chunks;
1271 u64 physical_start = *start;
1274 search_list = &transaction->pending_chunks;
1276 list_for_each_entry(em, search_list, list) {
1277 struct map_lookup *map;
1280 map = em->map_lookup;
1281 for (i = 0; i < map->num_stripes; i++) {
1284 if (map->stripes[i].dev != device)
1286 if (map->stripes[i].physical >= physical_start + len ||
1287 map->stripes[i].physical + em->orig_block_len <=
1291 * Make sure that while processing the pinned list we do
1292 * not override our *start with a lower value, because
1293 * we can have pinned chunks that fall within this
1294 * device hole and that have lower physical addresses
1295 * than the pending chunks we processed before. If we
1296 * do not take this special care we can end up getting
1297 * 2 pending chunks that start at the same physical
1298 * device offsets because the end offset of a pinned
1299 * chunk can be equal to the start offset of some
1302 end = map->stripes[i].physical + em->orig_block_len;
1309 if (search_list != &fs_info->pinned_chunks) {
1310 search_list = &fs_info->pinned_chunks;
1319 * find_free_dev_extent_start - find free space in the specified device
1320 * @device: the device which we search the free space in
1321 * @num_bytes: the size of the free space that we need
1322 * @search_start: the position from which to begin the search
1323 * @start: store the start of the free space.
1324 * @len: the size of the free space. that we find, or the size
1325 * of the max free space if we don't find suitable free space
1327 * this uses a pretty simple search, the expectation is that it is
1328 * called very infrequently and that a given device has a small number
1331 * @start is used to store the start of the free space if we find. But if we
1332 * don't find suitable free space, it will be used to store the start position
1333 * of the max free space.
1335 * @len is used to store the size of the free space that we find.
1336 * But if we don't find suitable free space, it is used to store the size of
1337 * the max free space.
1339 int find_free_dev_extent_start(struct btrfs_transaction *transaction,
1340 struct btrfs_device *device, u64 num_bytes,
1341 u64 search_start, u64 *start, u64 *len)
1343 struct btrfs_fs_info *fs_info = device->fs_info;
1344 struct btrfs_root *root = fs_info->dev_root;
1345 struct btrfs_key key;
1346 struct btrfs_dev_extent *dev_extent;
1347 struct btrfs_path *path;
1352 u64 search_end = device->total_bytes;
1355 struct extent_buffer *l;
1356 u64 min_search_start;
1359 * We don't want to overwrite the superblock on the drive nor any area
1360 * used by the boot loader (grub for example), so we make sure to start
1361 * at an offset of at least 1MB.
1363 min_search_start = max(fs_info->alloc_start, 1024ull * 1024);
1364 search_start = max(search_start, min_search_start);
1366 path = btrfs_alloc_path();
1370 max_hole_start = search_start;
1374 if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1379 path->reada = READA_FORWARD;
1380 path->search_commit_root = 1;
1381 path->skip_locking = 1;
1383 key.objectid = device->devid;
1384 key.offset = search_start;
1385 key.type = BTRFS_DEV_EXTENT_KEY;
1387 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1391 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1398 slot = path->slots[0];
1399 if (slot >= btrfs_header_nritems(l)) {
1400 ret = btrfs_next_leaf(root, path);
1408 btrfs_item_key_to_cpu(l, &key, slot);
1410 if (key.objectid < device->devid)
1413 if (key.objectid > device->devid)
1416 if (key.type != BTRFS_DEV_EXTENT_KEY)
1419 if (key.offset > search_start) {
1420 hole_size = key.offset - search_start;
1423 * Have to check before we set max_hole_start, otherwise
1424 * we could end up sending back this offset anyway.
1426 if (contains_pending_extent(transaction, device,
1429 if (key.offset >= search_start) {
1430 hole_size = key.offset - search_start;
1437 if (hole_size > max_hole_size) {
1438 max_hole_start = search_start;
1439 max_hole_size = hole_size;
1443 * If this free space is greater than which we need,
1444 * it must be the max free space that we have found
1445 * until now, so max_hole_start must point to the start
1446 * of this free space and the length of this free space
1447 * is stored in max_hole_size. Thus, we return
1448 * max_hole_start and max_hole_size and go back to the
1451 if (hole_size >= num_bytes) {
1457 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1458 extent_end = key.offset + btrfs_dev_extent_length(l,
1460 if (extent_end > search_start)
1461 search_start = extent_end;
1468 * At this point, search_start should be the end of
1469 * allocated dev extents, and when shrinking the device,
1470 * search_end may be smaller than search_start.
1472 if (search_end > search_start) {
1473 hole_size = search_end - search_start;
1475 if (contains_pending_extent(transaction, device, &search_start,
1477 btrfs_release_path(path);
1481 if (hole_size > max_hole_size) {
1482 max_hole_start = search_start;
1483 max_hole_size = hole_size;
1488 if (max_hole_size < num_bytes)
1494 btrfs_free_path(path);
1495 *start = max_hole_start;
1497 *len = max_hole_size;
1501 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1502 struct btrfs_device *device, u64 num_bytes,
1503 u64 *start, u64 *len)
1505 /* FIXME use last free of some kind */
1506 return find_free_dev_extent_start(trans->transaction, device,
1507 num_bytes, 0, start, len);
1510 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1511 struct btrfs_device *device,
1512 u64 start, u64 *dev_extent_len)
1514 struct btrfs_fs_info *fs_info = device->fs_info;
1515 struct btrfs_root *root = fs_info->dev_root;
1517 struct btrfs_path *path;
1518 struct btrfs_key key;
1519 struct btrfs_key found_key;
1520 struct extent_buffer *leaf = NULL;
1521 struct btrfs_dev_extent *extent = NULL;
1523 path = btrfs_alloc_path();
1527 key.objectid = device->devid;
1529 key.type = BTRFS_DEV_EXTENT_KEY;
1531 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1533 ret = btrfs_previous_item(root, path, key.objectid,
1534 BTRFS_DEV_EXTENT_KEY);
1537 leaf = path->nodes[0];
1538 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1539 extent = btrfs_item_ptr(leaf, path->slots[0],
1540 struct btrfs_dev_extent);
1541 BUG_ON(found_key.offset > start || found_key.offset +
1542 btrfs_dev_extent_length(leaf, extent) < start);
1544 btrfs_release_path(path);
1546 } else if (ret == 0) {
1547 leaf = path->nodes[0];
1548 extent = btrfs_item_ptr(leaf, path->slots[0],
1549 struct btrfs_dev_extent);
1551 btrfs_handle_fs_error(fs_info, ret, "Slot search failed");
1555 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1557 ret = btrfs_del_item(trans, root, path);
1559 btrfs_handle_fs_error(fs_info, ret,
1560 "Failed to remove dev extent item");
1562 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1565 btrfs_free_path(path);
1569 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1570 struct btrfs_device *device,
1571 u64 chunk_tree, u64 chunk_objectid,
1572 u64 chunk_offset, u64 start, u64 num_bytes)
1575 struct btrfs_path *path;
1576 struct btrfs_fs_info *fs_info = device->fs_info;
1577 struct btrfs_root *root = fs_info->dev_root;
1578 struct btrfs_dev_extent *extent;
1579 struct extent_buffer *leaf;
1580 struct btrfs_key key;
1582 WARN_ON(!device->in_fs_metadata);
1583 WARN_ON(device->is_tgtdev_for_dev_replace);
1584 path = btrfs_alloc_path();
1588 key.objectid = device->devid;
1590 key.type = BTRFS_DEV_EXTENT_KEY;
1591 ret = btrfs_insert_empty_item(trans, root, path, &key,
1596 leaf = path->nodes[0];
1597 extent = btrfs_item_ptr(leaf, path->slots[0],
1598 struct btrfs_dev_extent);
1599 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1600 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1601 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1603 write_extent_buffer_chunk_tree_uuid(leaf, fs_info->chunk_tree_uuid);
1605 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1606 btrfs_mark_buffer_dirty(leaf);
1608 btrfs_free_path(path);
1612 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1614 struct extent_map_tree *em_tree;
1615 struct extent_map *em;
1619 em_tree = &fs_info->mapping_tree.map_tree;
1620 read_lock(&em_tree->lock);
1621 n = rb_last(&em_tree->map);
1623 em = rb_entry(n, struct extent_map, rb_node);
1624 ret = em->start + em->len;
1626 read_unlock(&em_tree->lock);
1631 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1635 struct btrfs_key key;
1636 struct btrfs_key found_key;
1637 struct btrfs_path *path;
1639 path = btrfs_alloc_path();
1643 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1644 key.type = BTRFS_DEV_ITEM_KEY;
1645 key.offset = (u64)-1;
1647 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1651 BUG_ON(ret == 0); /* Corruption */
1653 ret = btrfs_previous_item(fs_info->chunk_root, path,
1654 BTRFS_DEV_ITEMS_OBJECTID,
1655 BTRFS_DEV_ITEM_KEY);
1659 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1661 *devid_ret = found_key.offset + 1;
1665 btrfs_free_path(path);
1670 * the device information is stored in the chunk root
1671 * the btrfs_device struct should be fully filled in
1673 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1674 struct btrfs_fs_info *fs_info,
1675 struct btrfs_device *device)
1677 struct btrfs_root *root = fs_info->chunk_root;
1679 struct btrfs_path *path;
1680 struct btrfs_dev_item *dev_item;
1681 struct extent_buffer *leaf;
1682 struct btrfs_key key;
1685 path = btrfs_alloc_path();
1689 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1690 key.type = BTRFS_DEV_ITEM_KEY;
1691 key.offset = device->devid;
1693 ret = btrfs_insert_empty_item(trans, root, path, &key,
1698 leaf = path->nodes[0];
1699 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1701 btrfs_set_device_id(leaf, dev_item, device->devid);
1702 btrfs_set_device_generation(leaf, dev_item, 0);
1703 btrfs_set_device_type(leaf, dev_item, device->type);
1704 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1705 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1706 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1707 btrfs_set_device_total_bytes(leaf, dev_item,
1708 btrfs_device_get_disk_total_bytes(device));
1709 btrfs_set_device_bytes_used(leaf, dev_item,
1710 btrfs_device_get_bytes_used(device));
1711 btrfs_set_device_group(leaf, dev_item, 0);
1712 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1713 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1714 btrfs_set_device_start_offset(leaf, dev_item, 0);
1716 ptr = btrfs_device_uuid(dev_item);
1717 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1718 ptr = btrfs_device_fsid(dev_item);
1719 write_extent_buffer(leaf, fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1720 btrfs_mark_buffer_dirty(leaf);
1724 btrfs_free_path(path);
1729 * Function to update ctime/mtime for a given device path.
1730 * Mainly used for ctime/mtime based probe like libblkid.
1732 static void update_dev_time(const char *path_name)
1736 filp = filp_open(path_name, O_RDWR, 0);
1739 file_update_time(filp);
1740 filp_close(filp, NULL);
1743 static int btrfs_rm_dev_item(struct btrfs_fs_info *fs_info,
1744 struct btrfs_device *device)
1746 struct btrfs_root *root = fs_info->chunk_root;
1748 struct btrfs_path *path;
1749 struct btrfs_key key;
1750 struct btrfs_trans_handle *trans;
1752 path = btrfs_alloc_path();
1756 trans = btrfs_start_transaction(root, 0);
1757 if (IS_ERR(trans)) {
1758 btrfs_free_path(path);
1759 return PTR_ERR(trans);
1761 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1762 key.type = BTRFS_DEV_ITEM_KEY;
1763 key.offset = device->devid;
1765 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1774 ret = btrfs_del_item(trans, root, path);
1778 btrfs_free_path(path);
1779 btrfs_commit_transaction(trans);
1784 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1785 * filesystem. It's up to the caller to adjust that number regarding eg. device
1788 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1796 seq = read_seqbegin(&fs_info->profiles_lock);
1798 all_avail = fs_info->avail_data_alloc_bits |
1799 fs_info->avail_system_alloc_bits |
1800 fs_info->avail_metadata_alloc_bits;
1801 } while (read_seqretry(&fs_info->profiles_lock, seq));
1803 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1804 if (!(all_avail & btrfs_raid_group[i]))
1807 if (num_devices < btrfs_raid_array[i].devs_min) {
1808 int ret = btrfs_raid_mindev_error[i];
1818 struct btrfs_device *btrfs_find_next_active_device(struct btrfs_fs_devices *fs_devs,
1819 struct btrfs_device *device)
1821 struct btrfs_device *next_device;
1823 list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1824 if (next_device != device &&
1825 !next_device->missing && next_device->bdev)
1833 * Helper function to check if the given device is part of s_bdev / latest_bdev
1834 * and replace it with the provided or the next active device, in the context
1835 * where this function called, there should be always be another device (or
1836 * this_dev) which is active.
1838 void btrfs_assign_next_active_device(struct btrfs_fs_info *fs_info,
1839 struct btrfs_device *device, struct btrfs_device *this_dev)
1841 struct btrfs_device *next_device;
1844 next_device = this_dev;
1846 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
1848 ASSERT(next_device);
1850 if (fs_info->sb->s_bdev &&
1851 (fs_info->sb->s_bdev == device->bdev))
1852 fs_info->sb->s_bdev = next_device->bdev;
1854 if (fs_info->fs_devices->latest_bdev == device->bdev)
1855 fs_info->fs_devices->latest_bdev = next_device->bdev;
1858 int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path,
1861 struct btrfs_device *device;
1862 struct btrfs_fs_devices *cur_devices;
1865 bool clear_super = false;
1867 mutex_lock(&uuid_mutex);
1869 num_devices = fs_info->fs_devices->num_devices;
1870 btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
1871 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
1872 WARN_ON(num_devices < 1);
1875 btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
1877 ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
1881 ret = btrfs_find_device_by_devspec(fs_info, devid, device_path,
1886 if (device->is_tgtdev_for_dev_replace) {
1887 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1891 if (device->writeable && fs_info->fs_devices->rw_devices == 1) {
1892 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1896 if (device->writeable) {
1897 mutex_lock(&fs_info->chunk_mutex);
1898 list_del_init(&device->dev_alloc_list);
1899 device->fs_devices->rw_devices--;
1900 mutex_unlock(&fs_info->chunk_mutex);
1904 mutex_unlock(&uuid_mutex);
1905 ret = btrfs_shrink_device(device, 0);
1906 mutex_lock(&uuid_mutex);
1911 * TODO: the superblock still includes this device in its num_devices
1912 * counter although write_all_supers() is not locked out. This
1913 * could give a filesystem state which requires a degraded mount.
1915 ret = btrfs_rm_dev_item(fs_info, device);
1919 device->in_fs_metadata = 0;
1920 btrfs_scrub_cancel_dev(fs_info, device);
1923 * the device list mutex makes sure that we don't change
1924 * the device list while someone else is writing out all
1925 * the device supers. Whoever is writing all supers, should
1926 * lock the device list mutex before getting the number of
1927 * devices in the super block (super_copy). Conversely,
1928 * whoever updates the number of devices in the super block
1929 * (super_copy) should hold the device list mutex.
1932 cur_devices = device->fs_devices;
1933 mutex_lock(&fs_info->fs_devices->device_list_mutex);
1934 list_del_rcu(&device->dev_list);
1936 device->fs_devices->num_devices--;
1937 device->fs_devices->total_devices--;
1939 if (device->missing)
1940 device->fs_devices->missing_devices--;
1942 btrfs_assign_next_active_device(fs_info, device, NULL);
1945 device->fs_devices->open_devices--;
1946 /* remove sysfs entry */
1947 btrfs_sysfs_rm_device_link(fs_info->fs_devices, device);
1950 num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
1951 btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
1952 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1955 * at this point, the device is zero sized and detached from
1956 * the devices list. All that's left is to zero out the old
1957 * supers and free the device.
1959 if (device->writeable)
1960 btrfs_scratch_superblocks(device->bdev, device->name->str);
1962 btrfs_close_bdev(device);
1963 call_rcu(&device->rcu, free_device);
1965 if (cur_devices->open_devices == 0) {
1966 struct btrfs_fs_devices *fs_devices;
1967 fs_devices = fs_info->fs_devices;
1968 while (fs_devices) {
1969 if (fs_devices->seed == cur_devices) {
1970 fs_devices->seed = cur_devices->seed;
1973 fs_devices = fs_devices->seed;
1975 cur_devices->seed = NULL;
1976 __btrfs_close_devices(cur_devices);
1977 free_fs_devices(cur_devices);
1980 fs_info->num_tolerated_disk_barrier_failures =
1981 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
1984 mutex_unlock(&uuid_mutex);
1988 if (device->writeable) {
1989 mutex_lock(&fs_info->chunk_mutex);
1990 list_add(&device->dev_alloc_list,
1991 &fs_info->fs_devices->alloc_list);
1992 device->fs_devices->rw_devices++;
1993 mutex_unlock(&fs_info->chunk_mutex);
1998 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
1999 struct btrfs_device *srcdev)
2001 struct btrfs_fs_devices *fs_devices;
2003 WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
2006 * in case of fs with no seed, srcdev->fs_devices will point
2007 * to fs_devices of fs_info. However when the dev being replaced is
2008 * a seed dev it will point to the seed's local fs_devices. In short
2009 * srcdev will have its correct fs_devices in both the cases.
2011 fs_devices = srcdev->fs_devices;
2013 list_del_rcu(&srcdev->dev_list);
2014 list_del_rcu(&srcdev->dev_alloc_list);
2015 fs_devices->num_devices--;
2016 if (srcdev->missing)
2017 fs_devices->missing_devices--;
2019 if (srcdev->writeable)
2020 fs_devices->rw_devices--;
2023 fs_devices->open_devices--;
2026 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
2027 struct btrfs_device *srcdev)
2029 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2031 if (srcdev->writeable) {
2032 /* zero out the old super if it is writable */
2033 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
2036 btrfs_close_bdev(srcdev);
2038 call_rcu(&srcdev->rcu, free_device);
2041 * unless fs_devices is seed fs, num_devices shouldn't go
2044 BUG_ON(!fs_devices->num_devices && !fs_devices->seeding);
2046 /* if this is no devs we rather delete the fs_devices */
2047 if (!fs_devices->num_devices) {
2048 struct btrfs_fs_devices *tmp_fs_devices;
2050 tmp_fs_devices = fs_info->fs_devices;
2051 while (tmp_fs_devices) {
2052 if (tmp_fs_devices->seed == fs_devices) {
2053 tmp_fs_devices->seed = fs_devices->seed;
2056 tmp_fs_devices = tmp_fs_devices->seed;
2058 fs_devices->seed = NULL;
2059 __btrfs_close_devices(fs_devices);
2060 free_fs_devices(fs_devices);
2064 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2065 struct btrfs_device *tgtdev)
2067 mutex_lock(&uuid_mutex);
2069 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2071 btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev);
2074 fs_info->fs_devices->open_devices--;
2076 fs_info->fs_devices->num_devices--;
2078 btrfs_assign_next_active_device(fs_info, tgtdev, NULL);
2080 list_del_rcu(&tgtdev->dev_list);
2082 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2083 mutex_unlock(&uuid_mutex);
2086 * The update_dev_time() with in btrfs_scratch_superblocks()
2087 * may lead to a call to btrfs_show_devname() which will try
2088 * to hold device_list_mutex. And here this device
2089 * is already out of device list, so we don't have to hold
2090 * the device_list_mutex lock.
2092 btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
2094 btrfs_close_bdev(tgtdev);
2095 call_rcu(&tgtdev->rcu, free_device);
2098 static int btrfs_find_device_by_path(struct btrfs_fs_info *fs_info,
2099 const char *device_path,
2100 struct btrfs_device **device)
2103 struct btrfs_super_block *disk_super;
2106 struct block_device *bdev;
2107 struct buffer_head *bh;
2110 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2111 fs_info->bdev_holder, 0, &bdev, &bh);
2114 disk_super = (struct btrfs_super_block *)bh->b_data;
2115 devid = btrfs_stack_device_id(&disk_super->dev_item);
2116 dev_uuid = disk_super->dev_item.uuid;
2117 *device = btrfs_find_device(fs_info, devid, dev_uuid, disk_super->fsid);
2121 blkdev_put(bdev, FMODE_READ);
2125 int btrfs_find_device_missing_or_by_path(struct btrfs_fs_info *fs_info,
2126 const char *device_path,
2127 struct btrfs_device **device)
2130 if (strcmp(device_path, "missing") == 0) {
2131 struct list_head *devices;
2132 struct btrfs_device *tmp;
2134 devices = &fs_info->fs_devices->devices;
2136 * It is safe to read the devices since the volume_mutex
2137 * is held by the caller.
2139 list_for_each_entry(tmp, devices, dev_list) {
2140 if (tmp->in_fs_metadata && !tmp->bdev) {
2147 return BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2151 return btrfs_find_device_by_path(fs_info, device_path, device);
2156 * Lookup a device given by device id, or the path if the id is 0.
2158 int btrfs_find_device_by_devspec(struct btrfs_fs_info *fs_info, u64 devid,
2159 const char *devpath,
2160 struct btrfs_device **device)
2166 *device = btrfs_find_device(fs_info, devid, NULL, NULL);
2170 if (!devpath || !devpath[0])
2173 ret = btrfs_find_device_missing_or_by_path(fs_info, devpath,
2180 * does all the dirty work required for changing file system's UUID.
2182 static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info)
2184 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2185 struct btrfs_fs_devices *old_devices;
2186 struct btrfs_fs_devices *seed_devices;
2187 struct btrfs_super_block *disk_super = fs_info->super_copy;
2188 struct btrfs_device *device;
2191 BUG_ON(!mutex_is_locked(&uuid_mutex));
2192 if (!fs_devices->seeding)
2195 seed_devices = __alloc_fs_devices();
2196 if (IS_ERR(seed_devices))
2197 return PTR_ERR(seed_devices);
2199 old_devices = clone_fs_devices(fs_devices);
2200 if (IS_ERR(old_devices)) {
2201 kfree(seed_devices);
2202 return PTR_ERR(old_devices);
2205 list_add(&old_devices->list, &fs_uuids);
2207 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2208 seed_devices->opened = 1;
2209 INIT_LIST_HEAD(&seed_devices->devices);
2210 INIT_LIST_HEAD(&seed_devices->alloc_list);
2211 mutex_init(&seed_devices->device_list_mutex);
2213 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2214 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2216 list_for_each_entry(device, &seed_devices->devices, dev_list)
2217 device->fs_devices = seed_devices;
2219 mutex_lock(&fs_info->chunk_mutex);
2220 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2221 mutex_unlock(&fs_info->chunk_mutex);
2223 fs_devices->seeding = 0;
2224 fs_devices->num_devices = 0;
2225 fs_devices->open_devices = 0;
2226 fs_devices->missing_devices = 0;
2227 fs_devices->rotating = 0;
2228 fs_devices->seed = seed_devices;
2230 generate_random_uuid(fs_devices->fsid);
2231 memcpy(fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2232 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2233 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2235 super_flags = btrfs_super_flags(disk_super) &
2236 ~BTRFS_SUPER_FLAG_SEEDING;
2237 btrfs_set_super_flags(disk_super, super_flags);
2243 * Store the expected generation for seed devices in device items.
2245 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2246 struct btrfs_fs_info *fs_info)
2248 struct btrfs_root *root = fs_info->chunk_root;
2249 struct btrfs_path *path;
2250 struct extent_buffer *leaf;
2251 struct btrfs_dev_item *dev_item;
2252 struct btrfs_device *device;
2253 struct btrfs_key key;
2254 u8 fs_uuid[BTRFS_UUID_SIZE];
2255 u8 dev_uuid[BTRFS_UUID_SIZE];
2259 path = btrfs_alloc_path();
2263 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2265 key.type = BTRFS_DEV_ITEM_KEY;
2268 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2272 leaf = path->nodes[0];
2274 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2275 ret = btrfs_next_leaf(root, path);
2280 leaf = path->nodes[0];
2281 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2282 btrfs_release_path(path);
2286 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2287 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2288 key.type != BTRFS_DEV_ITEM_KEY)
2291 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2292 struct btrfs_dev_item);
2293 devid = btrfs_device_id(leaf, dev_item);
2294 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2296 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2298 device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
2299 BUG_ON(!device); /* Logic error */
2301 if (device->fs_devices->seeding) {
2302 btrfs_set_device_generation(leaf, dev_item,
2303 device->generation);
2304 btrfs_mark_buffer_dirty(leaf);
2312 btrfs_free_path(path);
2316 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2318 struct btrfs_root *root = fs_info->dev_root;
2319 struct request_queue *q;
2320 struct btrfs_trans_handle *trans;
2321 struct btrfs_device *device;
2322 struct block_device *bdev;
2323 struct list_head *devices;
2324 struct super_block *sb = fs_info->sb;
2325 struct rcu_string *name;
2327 int seeding_dev = 0;
2330 if ((sb->s_flags & MS_RDONLY) && !fs_info->fs_devices->seeding)
2333 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2334 fs_info->bdev_holder);
2336 return PTR_ERR(bdev);
2338 if (fs_info->fs_devices->seeding) {
2340 down_write(&sb->s_umount);
2341 mutex_lock(&uuid_mutex);
2344 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2346 devices = &fs_info->fs_devices->devices;
2348 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2349 list_for_each_entry(device, devices, dev_list) {
2350 if (device->bdev == bdev) {
2353 &fs_info->fs_devices->device_list_mutex);
2357 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2359 device = btrfs_alloc_device(fs_info, NULL, NULL);
2360 if (IS_ERR(device)) {
2361 /* we can safely leave the fs_devices entry around */
2362 ret = PTR_ERR(device);
2366 name = rcu_string_strdup(device_path, GFP_KERNEL);
2372 rcu_assign_pointer(device->name, name);
2374 trans = btrfs_start_transaction(root, 0);
2375 if (IS_ERR(trans)) {
2376 rcu_string_free(device->name);
2378 ret = PTR_ERR(trans);
2382 q = bdev_get_queue(bdev);
2383 if (blk_queue_discard(q))
2384 device->can_discard = 1;
2385 device->writeable = 1;
2386 device->generation = trans->transid;
2387 device->io_width = fs_info->sectorsize;
2388 device->io_align = fs_info->sectorsize;
2389 device->sector_size = fs_info->sectorsize;
2390 device->total_bytes = i_size_read(bdev->bd_inode);
2391 device->disk_total_bytes = device->total_bytes;
2392 device->commit_total_bytes = device->total_bytes;
2393 device->fs_info = fs_info;
2394 device->bdev = bdev;
2395 device->in_fs_metadata = 1;
2396 device->is_tgtdev_for_dev_replace = 0;
2397 device->mode = FMODE_EXCL;
2398 device->dev_stats_valid = 1;
2399 set_blocksize(device->bdev, 4096);
2402 sb->s_flags &= ~MS_RDONLY;
2403 ret = btrfs_prepare_sprout(fs_info);
2404 BUG_ON(ret); /* -ENOMEM */
2407 device->fs_devices = fs_info->fs_devices;
2409 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2410 mutex_lock(&fs_info->chunk_mutex);
2411 list_add_rcu(&device->dev_list, &fs_info->fs_devices->devices);
2412 list_add(&device->dev_alloc_list,
2413 &fs_info->fs_devices->alloc_list);
2414 fs_info->fs_devices->num_devices++;
2415 fs_info->fs_devices->open_devices++;
2416 fs_info->fs_devices->rw_devices++;
2417 fs_info->fs_devices->total_devices++;
2418 fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2420 atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2422 if (!blk_queue_nonrot(q))
2423 fs_info->fs_devices->rotating = 1;
2425 tmp = btrfs_super_total_bytes(fs_info->super_copy);
2426 btrfs_set_super_total_bytes(fs_info->super_copy,
2427 tmp + device->total_bytes);
2429 tmp = btrfs_super_num_devices(fs_info->super_copy);
2430 btrfs_set_super_num_devices(fs_info->super_copy, tmp + 1);
2432 /* add sysfs device entry */
2433 btrfs_sysfs_add_device_link(fs_info->fs_devices, device);
2436 * we've got more storage, clear any full flags on the space
2439 btrfs_clear_space_info_full(fs_info);
2441 mutex_unlock(&fs_info->chunk_mutex);
2442 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2445 mutex_lock(&fs_info->chunk_mutex);
2446 ret = init_first_rw_device(trans, fs_info);
2447 mutex_unlock(&fs_info->chunk_mutex);
2449 btrfs_abort_transaction(trans, ret);
2454 ret = btrfs_add_device(trans, fs_info, device);
2456 btrfs_abort_transaction(trans, ret);
2461 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2463 ret = btrfs_finish_sprout(trans, fs_info);
2465 btrfs_abort_transaction(trans, ret);
2469 /* Sprouting would change fsid of the mounted root,
2470 * so rename the fsid on the sysfs
2472 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2474 if (kobject_rename(&fs_info->fs_devices->fsid_kobj, fsid_buf))
2476 "sysfs: failed to create fsid for sprout");
2479 fs_info->num_tolerated_disk_barrier_failures =
2480 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2481 ret = btrfs_commit_transaction(trans);
2484 mutex_unlock(&uuid_mutex);
2485 up_write(&sb->s_umount);
2487 if (ret) /* transaction commit */
2490 ret = btrfs_relocate_sys_chunks(fs_info);
2492 btrfs_handle_fs_error(fs_info, ret,
2493 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2494 trans = btrfs_attach_transaction(root);
2495 if (IS_ERR(trans)) {
2496 if (PTR_ERR(trans) == -ENOENT)
2498 return PTR_ERR(trans);
2500 ret = btrfs_commit_transaction(trans);
2503 /* Update ctime/mtime for libblkid */
2504 update_dev_time(device_path);
2508 btrfs_end_transaction(trans);
2509 rcu_string_free(device->name);
2510 btrfs_sysfs_rm_device_link(fs_info->fs_devices, device);
2513 blkdev_put(bdev, FMODE_EXCL);
2515 mutex_unlock(&uuid_mutex);
2516 up_write(&sb->s_umount);
2521 int btrfs_init_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2522 const char *device_path,
2523 struct btrfs_device *srcdev,
2524 struct btrfs_device **device_out)
2526 struct request_queue *q;
2527 struct btrfs_device *device;
2528 struct block_device *bdev;
2529 struct list_head *devices;
2530 struct rcu_string *name;
2531 u64 devid = BTRFS_DEV_REPLACE_DEVID;
2535 if (fs_info->fs_devices->seeding) {
2536 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
2540 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2541 fs_info->bdev_holder);
2543 btrfs_err(fs_info, "target device %s is invalid!", device_path);
2544 return PTR_ERR(bdev);
2547 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2549 devices = &fs_info->fs_devices->devices;
2550 list_for_each_entry(device, devices, dev_list) {
2551 if (device->bdev == bdev) {
2553 "target device is in the filesystem!");
2560 if (i_size_read(bdev->bd_inode) <
2561 btrfs_device_get_total_bytes(srcdev)) {
2563 "target device is smaller than source device!");
2569 device = btrfs_alloc_device(NULL, &devid, NULL);
2570 if (IS_ERR(device)) {
2571 ret = PTR_ERR(device);
2575 name = rcu_string_strdup(device_path, GFP_NOFS);
2581 rcu_assign_pointer(device->name, name);
2583 q = bdev_get_queue(bdev);
2584 if (blk_queue_discard(q))
2585 device->can_discard = 1;
2586 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2587 device->writeable = 1;
2588 device->generation = 0;
2589 device->io_width = fs_info->sectorsize;
2590 device->io_align = fs_info->sectorsize;
2591 device->sector_size = fs_info->sectorsize;
2592 device->total_bytes = btrfs_device_get_total_bytes(srcdev);
2593 device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
2594 device->bytes_used = btrfs_device_get_bytes_used(srcdev);
2595 ASSERT(list_empty(&srcdev->resized_list));
2596 device->commit_total_bytes = srcdev->commit_total_bytes;
2597 device->commit_bytes_used = device->bytes_used;
2598 device->fs_info = fs_info;
2599 device->bdev = bdev;
2600 device->in_fs_metadata = 1;
2601 device->is_tgtdev_for_dev_replace = 1;
2602 device->mode = FMODE_EXCL;
2603 device->dev_stats_valid = 1;
2604 set_blocksize(device->bdev, 4096);
2605 device->fs_devices = fs_info->fs_devices;
2606 list_add(&device->dev_list, &fs_info->fs_devices->devices);
2607 fs_info->fs_devices->num_devices++;
2608 fs_info->fs_devices->open_devices++;
2609 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2611 *device_out = device;
2615 blkdev_put(bdev, FMODE_EXCL);
2619 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2620 struct btrfs_device *tgtdev)
2622 u32 sectorsize = fs_info->sectorsize;
2624 WARN_ON(fs_info->fs_devices->rw_devices == 0);
2625 tgtdev->io_width = sectorsize;
2626 tgtdev->io_align = sectorsize;
2627 tgtdev->sector_size = sectorsize;
2628 tgtdev->fs_info = fs_info;
2629 tgtdev->in_fs_metadata = 1;
2632 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2633 struct btrfs_device *device)
2636 struct btrfs_path *path;
2637 struct btrfs_root *root = device->fs_info->chunk_root;
2638 struct btrfs_dev_item *dev_item;
2639 struct extent_buffer *leaf;
2640 struct btrfs_key key;
2642 path = btrfs_alloc_path();
2646 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2647 key.type = BTRFS_DEV_ITEM_KEY;
2648 key.offset = device->devid;
2650 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2659 leaf = path->nodes[0];
2660 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2662 btrfs_set_device_id(leaf, dev_item, device->devid);
2663 btrfs_set_device_type(leaf, dev_item, device->type);
2664 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2665 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2666 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2667 btrfs_set_device_total_bytes(leaf, dev_item,
2668 btrfs_device_get_disk_total_bytes(device));
2669 btrfs_set_device_bytes_used(leaf, dev_item,
2670 btrfs_device_get_bytes_used(device));
2671 btrfs_mark_buffer_dirty(leaf);
2674 btrfs_free_path(path);
2678 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2679 struct btrfs_device *device, u64 new_size)
2681 struct btrfs_fs_info *fs_info = device->fs_info;
2682 struct btrfs_super_block *super_copy = fs_info->super_copy;
2683 struct btrfs_fs_devices *fs_devices;
2687 if (!device->writeable)
2690 mutex_lock(&fs_info->chunk_mutex);
2691 old_total = btrfs_super_total_bytes(super_copy);
2692 diff = new_size - device->total_bytes;
2694 if (new_size <= device->total_bytes ||
2695 device->is_tgtdev_for_dev_replace) {
2696 mutex_unlock(&fs_info->chunk_mutex);
2700 fs_devices = fs_info->fs_devices;
2702 btrfs_set_super_total_bytes(super_copy, old_total + diff);
2703 device->fs_devices->total_rw_bytes += diff;
2705 btrfs_device_set_total_bytes(device, new_size);
2706 btrfs_device_set_disk_total_bytes(device, new_size);
2707 btrfs_clear_space_info_full(device->fs_info);
2708 if (list_empty(&device->resized_list))
2709 list_add_tail(&device->resized_list,
2710 &fs_devices->resized_devices);
2711 mutex_unlock(&fs_info->chunk_mutex);
2713 return btrfs_update_device(trans, device);
2716 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2717 struct btrfs_fs_info *fs_info, u64 chunk_objectid,
2720 struct btrfs_root *root = fs_info->chunk_root;
2722 struct btrfs_path *path;
2723 struct btrfs_key key;
2725 path = btrfs_alloc_path();
2729 key.objectid = chunk_objectid;
2730 key.offset = chunk_offset;
2731 key.type = BTRFS_CHUNK_ITEM_KEY;
2733 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2736 else if (ret > 0) { /* Logic error or corruption */
2737 btrfs_handle_fs_error(fs_info, -ENOENT,
2738 "Failed lookup while freeing chunk.");
2743 ret = btrfs_del_item(trans, root, path);
2745 btrfs_handle_fs_error(fs_info, ret,
2746 "Failed to delete chunk item.");
2748 btrfs_free_path(path);
2752 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info,
2753 u64 chunk_objectid, u64 chunk_offset)
2755 struct btrfs_super_block *super_copy = fs_info->super_copy;
2756 struct btrfs_disk_key *disk_key;
2757 struct btrfs_chunk *chunk;
2764 struct btrfs_key key;
2766 mutex_lock(&fs_info->chunk_mutex);
2767 array_size = btrfs_super_sys_array_size(super_copy);
2769 ptr = super_copy->sys_chunk_array;
2772 while (cur < array_size) {
2773 disk_key = (struct btrfs_disk_key *)ptr;
2774 btrfs_disk_key_to_cpu(&key, disk_key);
2776 len = sizeof(*disk_key);
2778 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2779 chunk = (struct btrfs_chunk *)(ptr + len);
2780 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2781 len += btrfs_chunk_item_size(num_stripes);
2786 if (key.objectid == chunk_objectid &&
2787 key.offset == chunk_offset) {
2788 memmove(ptr, ptr + len, array_size - (cur + len));
2790 btrfs_set_super_sys_array_size(super_copy, array_size);
2796 mutex_unlock(&fs_info->chunk_mutex);
2800 static struct extent_map *get_chunk_map(struct btrfs_fs_info *fs_info,
2801 u64 logical, u64 length)
2803 struct extent_map_tree *em_tree;
2804 struct extent_map *em;
2806 em_tree = &fs_info->mapping_tree.map_tree;
2807 read_lock(&em_tree->lock);
2808 em = lookup_extent_mapping(em_tree, logical, length);
2809 read_unlock(&em_tree->lock);
2812 btrfs_crit(fs_info, "unable to find logical %llu length %llu",
2814 return ERR_PTR(-EINVAL);
2817 if (em->start > logical || em->start + em->len < logical) {
2819 "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
2820 logical, length, em->start, em->start + em->len);
2821 free_extent_map(em);
2822 return ERR_PTR(-EINVAL);
2825 /* callers are responsible for dropping em's ref. */
2829 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2830 struct btrfs_fs_info *fs_info, u64 chunk_offset)
2832 struct extent_map *em;
2833 struct map_lookup *map;
2834 u64 dev_extent_len = 0;
2835 u64 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2837 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2839 em = get_chunk_map(fs_info, chunk_offset, 1);
2842 * This is a logic error, but we don't want to just rely on the
2843 * user having built with ASSERT enabled, so if ASSERT doesn't
2844 * do anything we still error out.
2849 map = em->map_lookup;
2850 mutex_lock(&fs_info->chunk_mutex);
2851 check_system_chunk(trans, fs_info, map->type);
2852 mutex_unlock(&fs_info->chunk_mutex);
2855 * Take the device list mutex to prevent races with the final phase of
2856 * a device replace operation that replaces the device object associated
2857 * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2859 mutex_lock(&fs_devices->device_list_mutex);
2860 for (i = 0; i < map->num_stripes; i++) {
2861 struct btrfs_device *device = map->stripes[i].dev;
2862 ret = btrfs_free_dev_extent(trans, device,
2863 map->stripes[i].physical,
2866 mutex_unlock(&fs_devices->device_list_mutex);
2867 btrfs_abort_transaction(trans, ret);
2871 if (device->bytes_used > 0) {
2872 mutex_lock(&fs_info->chunk_mutex);
2873 btrfs_device_set_bytes_used(device,
2874 device->bytes_used - dev_extent_len);
2875 atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
2876 btrfs_clear_space_info_full(fs_info);
2877 mutex_unlock(&fs_info->chunk_mutex);
2880 if (map->stripes[i].dev) {
2881 ret = btrfs_update_device(trans, map->stripes[i].dev);
2883 mutex_unlock(&fs_devices->device_list_mutex);
2884 btrfs_abort_transaction(trans, ret);
2889 mutex_unlock(&fs_devices->device_list_mutex);
2891 ret = btrfs_free_chunk(trans, fs_info, chunk_objectid, chunk_offset);
2893 btrfs_abort_transaction(trans, ret);
2897 trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
2899 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2900 ret = btrfs_del_sys_chunk(fs_info, chunk_objectid,
2903 btrfs_abort_transaction(trans, ret);
2908 ret = btrfs_remove_block_group(trans, fs_info, chunk_offset, em);
2910 btrfs_abort_transaction(trans, ret);
2916 free_extent_map(em);
2920 static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2922 struct btrfs_root *root = fs_info->chunk_root;
2923 struct btrfs_trans_handle *trans;
2927 * Prevent races with automatic removal of unused block groups.
2928 * After we relocate and before we remove the chunk with offset
2929 * chunk_offset, automatic removal of the block group can kick in,
2930 * resulting in a failure when calling btrfs_remove_chunk() below.
2932 * Make sure to acquire this mutex before doing a tree search (dev
2933 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2934 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2935 * we release the path used to search the chunk/dev tree and before
2936 * the current task acquires this mutex and calls us.
2938 ASSERT(mutex_is_locked(&fs_info->delete_unused_bgs_mutex));
2940 ret = btrfs_can_relocate(fs_info, chunk_offset);
2944 /* step one, relocate all the extents inside this chunk */
2945 btrfs_scrub_pause(fs_info);
2946 ret = btrfs_relocate_block_group(fs_info, chunk_offset);
2947 btrfs_scrub_continue(fs_info);
2951 trans = btrfs_start_trans_remove_block_group(root->fs_info,
2953 if (IS_ERR(trans)) {
2954 ret = PTR_ERR(trans);
2955 btrfs_handle_fs_error(root->fs_info, ret, NULL);
2960 * step two, delete the device extents and the
2961 * chunk tree entries
2963 ret = btrfs_remove_chunk(trans, fs_info, chunk_offset);
2964 btrfs_end_transaction(trans);
2968 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
2970 struct btrfs_root *chunk_root = fs_info->chunk_root;
2971 struct btrfs_path *path;
2972 struct extent_buffer *leaf;
2973 struct btrfs_chunk *chunk;
2974 struct btrfs_key key;
2975 struct btrfs_key found_key;
2977 bool retried = false;
2981 path = btrfs_alloc_path();
2986 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2987 key.offset = (u64)-1;
2988 key.type = BTRFS_CHUNK_ITEM_KEY;
2991 mutex_lock(&fs_info->delete_unused_bgs_mutex);
2992 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2994 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
2997 BUG_ON(ret == 0); /* Corruption */
2999 ret = btrfs_previous_item(chunk_root, path, key.objectid,
3002 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3008 leaf = path->nodes[0];
3009 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3011 chunk = btrfs_item_ptr(leaf, path->slots[0],
3012 struct btrfs_chunk);
3013 chunk_type = btrfs_chunk_type(leaf, chunk);
3014 btrfs_release_path(path);
3016 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3017 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3023 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3025 if (found_key.offset == 0)
3027 key.offset = found_key.offset - 1;
3030 if (failed && !retried) {
3034 } else if (WARN_ON(failed && retried)) {
3038 btrfs_free_path(path);
3042 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3043 struct btrfs_balance_control *bctl)
3045 struct btrfs_root *root = fs_info->tree_root;
3046 struct btrfs_trans_handle *trans;
3047 struct btrfs_balance_item *item;
3048 struct btrfs_disk_balance_args disk_bargs;
3049 struct btrfs_path *path;
3050 struct extent_buffer *leaf;
3051 struct btrfs_key key;
3054 path = btrfs_alloc_path();
3058 trans = btrfs_start_transaction(root, 0);
3059 if (IS_ERR(trans)) {
3060 btrfs_free_path(path);
3061 return PTR_ERR(trans);
3064 key.objectid = BTRFS_BALANCE_OBJECTID;
3065 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3068 ret = btrfs_insert_empty_item(trans, root, path, &key,
3073 leaf = path->nodes[0];
3074 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3076 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3078 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3079 btrfs_set_balance_data(leaf, item, &disk_bargs);
3080 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3081 btrfs_set_balance_meta(leaf, item, &disk_bargs);
3082 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3083 btrfs_set_balance_sys(leaf, item, &disk_bargs);
3085 btrfs_set_balance_flags(leaf, item, bctl->flags);
3087 btrfs_mark_buffer_dirty(leaf);
3089 btrfs_free_path(path);
3090 err = btrfs_commit_transaction(trans);
3096 static int del_balance_item(struct btrfs_fs_info *fs_info)
3098 struct btrfs_root *root = fs_info->tree_root;
3099 struct btrfs_trans_handle *trans;
3100 struct btrfs_path *path;
3101 struct btrfs_key key;
3104 path = btrfs_alloc_path();
3108 trans = btrfs_start_transaction(root, 0);
3109 if (IS_ERR(trans)) {
3110 btrfs_free_path(path);
3111 return PTR_ERR(trans);
3114 key.objectid = BTRFS_BALANCE_OBJECTID;
3115 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3118 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3126 ret = btrfs_del_item(trans, root, path);
3128 btrfs_free_path(path);
3129 err = btrfs_commit_transaction(trans);
3136 * This is a heuristic used to reduce the number of chunks balanced on
3137 * resume after balance was interrupted.
3139 static void update_balance_args(struct btrfs_balance_control *bctl)
3142 * Turn on soft mode for chunk types that were being converted.
3144 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3145 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3146 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3147 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3148 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3149 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3152 * Turn on usage filter if is not already used. The idea is
3153 * that chunks that we have already balanced should be
3154 * reasonably full. Don't do it for chunks that are being
3155 * converted - that will keep us from relocating unconverted
3156 * (albeit full) chunks.
3158 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3159 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3160 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3161 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3162 bctl->data.usage = 90;
3164 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3165 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3166 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3167 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3168 bctl->sys.usage = 90;
3170 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3171 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3172 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3173 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3174 bctl->meta.usage = 90;
3179 * Should be called with both balance and volume mutexes held to
3180 * serialize other volume operations (add_dev/rm_dev/resize) with
3181 * restriper. Same goes for unset_balance_control.
3183 static void set_balance_control(struct btrfs_balance_control *bctl)
3185 struct btrfs_fs_info *fs_info = bctl->fs_info;
3187 BUG_ON(fs_info->balance_ctl);
3189 spin_lock(&fs_info->balance_lock);
3190 fs_info->balance_ctl = bctl;
3191 spin_unlock(&fs_info->balance_lock);
3194 static void unset_balance_control(struct btrfs_fs_info *fs_info)
3196 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3198 BUG_ON(!fs_info->balance_ctl);
3200 spin_lock(&fs_info->balance_lock);
3201 fs_info->balance_ctl = NULL;
3202 spin_unlock(&fs_info->balance_lock);
3208 * Balance filters. Return 1 if chunk should be filtered out
3209 * (should not be balanced).
3211 static int chunk_profiles_filter(u64 chunk_type,
3212 struct btrfs_balance_args *bargs)
3214 chunk_type = chunk_to_extended(chunk_type) &
3215 BTRFS_EXTENDED_PROFILE_MASK;
3217 if (bargs->profiles & chunk_type)
3223 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3224 struct btrfs_balance_args *bargs)
3226 struct btrfs_block_group_cache *cache;
3228 u64 user_thresh_min;
3229 u64 user_thresh_max;
3232 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3233 chunk_used = btrfs_block_group_used(&cache->item);
3235 if (bargs->usage_min == 0)
3236 user_thresh_min = 0;
3238 user_thresh_min = div_factor_fine(cache->key.offset,
3241 if (bargs->usage_max == 0)
3242 user_thresh_max = 1;
3243 else if (bargs->usage_max > 100)
3244 user_thresh_max = cache->key.offset;
3246 user_thresh_max = div_factor_fine(cache->key.offset,
3249 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3252 btrfs_put_block_group(cache);
3256 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3257 u64 chunk_offset, struct btrfs_balance_args *bargs)
3259 struct btrfs_block_group_cache *cache;
3260 u64 chunk_used, user_thresh;
3263 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3264 chunk_used = btrfs_block_group_used(&cache->item);
3266 if (bargs->usage_min == 0)
3268 else if (bargs->usage > 100)
3269 user_thresh = cache->key.offset;
3271 user_thresh = div_factor_fine(cache->key.offset,
3274 if (chunk_used < user_thresh)
3277 btrfs_put_block_group(cache);
3281 static int chunk_devid_filter(struct extent_buffer *leaf,
3282 struct btrfs_chunk *chunk,
3283 struct btrfs_balance_args *bargs)
3285 struct btrfs_stripe *stripe;
3286 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3289 for (i = 0; i < num_stripes; i++) {
3290 stripe = btrfs_stripe_nr(chunk, i);
3291 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3298 /* [pstart, pend) */
3299 static int chunk_drange_filter(struct extent_buffer *leaf,
3300 struct btrfs_chunk *chunk,
3302 struct btrfs_balance_args *bargs)
3304 struct btrfs_stripe *stripe;
3305 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3311 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3314 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3315 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3316 factor = num_stripes / 2;
3317 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3318 factor = num_stripes - 1;
3319 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3320 factor = num_stripes - 2;
3322 factor = num_stripes;
3325 for (i = 0; i < num_stripes; i++) {
3326 stripe = btrfs_stripe_nr(chunk, i);
3327 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3330 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3331 stripe_length = btrfs_chunk_length(leaf, chunk);
3332 stripe_length = div_u64(stripe_length, factor);
3334 if (stripe_offset < bargs->pend &&
3335 stripe_offset + stripe_length > bargs->pstart)
3342 /* [vstart, vend) */
3343 static int chunk_vrange_filter(struct extent_buffer *leaf,
3344 struct btrfs_chunk *chunk,
3346 struct btrfs_balance_args *bargs)
3348 if (chunk_offset < bargs->vend &&
3349 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3350 /* at least part of the chunk is inside this vrange */
3356 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3357 struct btrfs_chunk *chunk,
3358 struct btrfs_balance_args *bargs)
3360 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3362 if (bargs->stripes_min <= num_stripes
3363 && num_stripes <= bargs->stripes_max)
3369 static int chunk_soft_convert_filter(u64 chunk_type,
3370 struct btrfs_balance_args *bargs)
3372 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3375 chunk_type = chunk_to_extended(chunk_type) &
3376 BTRFS_EXTENDED_PROFILE_MASK;
3378 if (bargs->target == chunk_type)
3384 static int should_balance_chunk(struct btrfs_fs_info *fs_info,
3385 struct extent_buffer *leaf,
3386 struct btrfs_chunk *chunk, u64 chunk_offset)
3388 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3389 struct btrfs_balance_args *bargs = NULL;
3390 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3393 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3394 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3398 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3399 bargs = &bctl->data;
3400 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3402 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3403 bargs = &bctl->meta;
3405 /* profiles filter */
3406 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3407 chunk_profiles_filter(chunk_type, bargs)) {
3412 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3413 chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3415 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3416 chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3421 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3422 chunk_devid_filter(leaf, chunk, bargs)) {
3426 /* drange filter, makes sense only with devid filter */
3427 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3428 chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
3433 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3434 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3438 /* stripes filter */
3439 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3440 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3444 /* soft profile changing mode */
3445 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3446 chunk_soft_convert_filter(chunk_type, bargs)) {
3451 * limited by count, must be the last filter
3453 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3454 if (bargs->limit == 0)
3458 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3460 * Same logic as the 'limit' filter; the minimum cannot be
3461 * determined here because we do not have the global information
3462 * about the count of all chunks that satisfy the filters.
3464 if (bargs->limit_max == 0)
3473 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3475 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3476 struct btrfs_root *chunk_root = fs_info->chunk_root;
3477 struct btrfs_root *dev_root = fs_info->dev_root;
3478 struct list_head *devices;
3479 struct btrfs_device *device;
3483 struct btrfs_chunk *chunk;
3484 struct btrfs_path *path = NULL;
3485 struct btrfs_key key;
3486 struct btrfs_key found_key;
3487 struct btrfs_trans_handle *trans;
3488 struct extent_buffer *leaf;
3491 int enospc_errors = 0;
3492 bool counting = true;
3493 /* The single value limit and min/max limits use the same bytes in the */
3494 u64 limit_data = bctl->data.limit;
3495 u64 limit_meta = bctl->meta.limit;
3496 u64 limit_sys = bctl->sys.limit;
3500 int chunk_reserved = 0;
3503 /* step one make some room on all the devices */
3504 devices = &fs_info->fs_devices->devices;
3505 list_for_each_entry(device, devices, dev_list) {
3506 old_size = btrfs_device_get_total_bytes(device);
3507 size_to_free = div_factor(old_size, 1);
3508 size_to_free = min_t(u64, size_to_free, SZ_1M);
3509 if (!device->writeable ||
3510 btrfs_device_get_total_bytes(device) -
3511 btrfs_device_get_bytes_used(device) > size_to_free ||
3512 device->is_tgtdev_for_dev_replace)
3515 ret = btrfs_shrink_device(device, old_size - size_to_free);
3519 /* btrfs_shrink_device never returns ret > 0 */
3524 trans = btrfs_start_transaction(dev_root, 0);
3525 if (IS_ERR(trans)) {
3526 ret = PTR_ERR(trans);
3527 btrfs_info_in_rcu(fs_info,
3528 "resize: unable to start transaction after shrinking device %s (error %d), old size %llu, new size %llu",
3529 rcu_str_deref(device->name), ret,
3530 old_size, old_size - size_to_free);
3534 ret = btrfs_grow_device(trans, device, old_size);
3536 btrfs_end_transaction(trans);
3537 /* btrfs_grow_device never returns ret > 0 */
3539 btrfs_info_in_rcu(fs_info,
3540 "resize: unable to grow device after shrinking device %s (error %d), old size %llu, new size %llu",
3541 rcu_str_deref(device->name), ret,
3542 old_size, old_size - size_to_free);
3546 btrfs_end_transaction(trans);
3549 /* step two, relocate all the chunks */
3550 path = btrfs_alloc_path();
3556 /* zero out stat counters */
3557 spin_lock(&fs_info->balance_lock);
3558 memset(&bctl->stat, 0, sizeof(bctl->stat));
3559 spin_unlock(&fs_info->balance_lock);
3563 * The single value limit and min/max limits use the same bytes
3566 bctl->data.limit = limit_data;
3567 bctl->meta.limit = limit_meta;
3568 bctl->sys.limit = limit_sys;
3570 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3571 key.offset = (u64)-1;
3572 key.type = BTRFS_CHUNK_ITEM_KEY;
3575 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3576 atomic_read(&fs_info->balance_cancel_req)) {
3581 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3582 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3584 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3589 * this shouldn't happen, it means the last relocate
3593 BUG(); /* FIXME break ? */
3595 ret = btrfs_previous_item(chunk_root, path, 0,
3596 BTRFS_CHUNK_ITEM_KEY);
3598 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3603 leaf = path->nodes[0];
3604 slot = path->slots[0];
3605 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3607 if (found_key.objectid != key.objectid) {
3608 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3612 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3613 chunk_type = btrfs_chunk_type(leaf, chunk);
3616 spin_lock(&fs_info->balance_lock);
3617 bctl->stat.considered++;
3618 spin_unlock(&fs_info->balance_lock);
3621 ret = should_balance_chunk(fs_info, leaf, chunk,
3624 btrfs_release_path(path);
3626 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3631 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3632 spin_lock(&fs_info->balance_lock);
3633 bctl->stat.expected++;
3634 spin_unlock(&fs_info->balance_lock);
3636 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3638 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3640 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3647 * Apply limit_min filter, no need to check if the LIMITS
3648 * filter is used, limit_min is 0 by default
3650 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3651 count_data < bctl->data.limit_min)
3652 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3653 count_meta < bctl->meta.limit_min)
3654 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3655 count_sys < bctl->sys.limit_min)) {
3656 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3660 ASSERT(fs_info->data_sinfo);
3661 spin_lock(&fs_info->data_sinfo->lock);
3662 bytes_used = fs_info->data_sinfo->bytes_used;
3663 spin_unlock(&fs_info->data_sinfo->lock);
3665 if ((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3666 !chunk_reserved && !bytes_used) {
3667 trans = btrfs_start_transaction(chunk_root, 0);
3668 if (IS_ERR(trans)) {
3669 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3670 ret = PTR_ERR(trans);
3674 ret = btrfs_force_chunk_alloc(trans, fs_info,
3675 BTRFS_BLOCK_GROUP_DATA);
3676 btrfs_end_transaction(trans);
3678 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3684 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3685 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3686 if (ret && ret != -ENOSPC)
3688 if (ret == -ENOSPC) {
3691 spin_lock(&fs_info->balance_lock);
3692 bctl->stat.completed++;
3693 spin_unlock(&fs_info->balance_lock);
3696 if (found_key.offset == 0)
3698 key.offset = found_key.offset - 1;
3702 btrfs_release_path(path);
3707 btrfs_free_path(path);
3708 if (enospc_errors) {
3709 btrfs_info(fs_info, "%d enospc errors during balance",
3719 * alloc_profile_is_valid - see if a given profile is valid and reduced
3720 * @flags: profile to validate
3721 * @extended: if true @flags is treated as an extended profile
3723 static int alloc_profile_is_valid(u64 flags, int extended)
3725 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3726 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3728 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3730 /* 1) check that all other bits are zeroed */
3734 /* 2) see if profile is reduced */
3736 return !extended; /* "0" is valid for usual profiles */
3738 /* true if exactly one bit set */
3739 return (flags & (flags - 1)) == 0;
3742 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3744 /* cancel requested || normal exit path */
3745 return atomic_read(&fs_info->balance_cancel_req) ||
3746 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3747 atomic_read(&fs_info->balance_cancel_req) == 0);
3750 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3754 unset_balance_control(fs_info);
3755 ret = del_balance_item(fs_info);
3757 btrfs_handle_fs_error(fs_info, ret, NULL);
3759 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3762 /* Non-zero return value signifies invalidity */
3763 static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg,
3766 return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3767 (!alloc_profile_is_valid(bctl_arg->target, 1) ||
3768 (bctl_arg->target & ~allowed)));
3772 * Should be called with both balance and volume mutexes held
3774 int btrfs_balance(struct btrfs_balance_control *bctl,
3775 struct btrfs_ioctl_balance_args *bargs)
3777 struct btrfs_fs_info *fs_info = bctl->fs_info;
3778 u64 meta_target, data_target;
3785 if (btrfs_fs_closing(fs_info) ||
3786 atomic_read(&fs_info->balance_pause_req) ||
3787 atomic_read(&fs_info->balance_cancel_req)) {
3792 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3793 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3797 * In case of mixed groups both data and meta should be picked,
3798 * and identical options should be given for both of them.
3800 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3801 if (mixed && (bctl->flags & allowed)) {
3802 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3803 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3804 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3806 "with mixed groups data and metadata balance options must be the same");
3812 num_devices = fs_info->fs_devices->num_devices;
3813 btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
3814 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3815 BUG_ON(num_devices < 1);
3818 btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
3819 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE | BTRFS_BLOCK_GROUP_DUP;
3820 if (num_devices > 1)
3821 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3822 if (num_devices > 2)
3823 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3824 if (num_devices > 3)
3825 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3826 BTRFS_BLOCK_GROUP_RAID6);
3827 if (validate_convert_profile(&bctl->data, allowed)) {
3829 "unable to start balance with target data profile %llu",
3834 if (validate_convert_profile(&bctl->meta, allowed)) {
3836 "unable to start balance with target metadata profile %llu",
3841 if (validate_convert_profile(&bctl->sys, allowed)) {
3843 "unable to start balance with target system profile %llu",
3849 /* allow to reduce meta or sys integrity only if force set */
3850 allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3851 BTRFS_BLOCK_GROUP_RAID10 |
3852 BTRFS_BLOCK_GROUP_RAID5 |
3853 BTRFS_BLOCK_GROUP_RAID6;
3855 seq = read_seqbegin(&fs_info->profiles_lock);
3857 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3858 (fs_info->avail_system_alloc_bits & allowed) &&
3859 !(bctl->sys.target & allowed)) ||
3860 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3861 (fs_info->avail_metadata_alloc_bits & allowed) &&
3862 !(bctl->meta.target & allowed))) {
3863 if (bctl->flags & BTRFS_BALANCE_FORCE) {
3865 "force reducing metadata integrity");
3868 "balance will reduce metadata integrity, use force if you want this");
3873 } while (read_seqretry(&fs_info->profiles_lock, seq));
3875 /* if we're not converting, the target field is uninitialized */
3876 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
3877 bctl->meta.target : fs_info->avail_metadata_alloc_bits;
3878 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
3879 bctl->data.target : fs_info->avail_data_alloc_bits;
3880 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <
3881 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {
3883 "metadata profile 0x%llx has lower redundancy than data profile 0x%llx",
3884 meta_target, data_target);
3887 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3888 fs_info->num_tolerated_disk_barrier_failures = min(
3889 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info),
3890 btrfs_get_num_tolerated_disk_barrier_failures(
3894 ret = insert_balance_item(fs_info, bctl);
3895 if (ret && ret != -EEXIST)
3898 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3899 BUG_ON(ret == -EEXIST);
3900 set_balance_control(bctl);
3902 BUG_ON(ret != -EEXIST);
3903 spin_lock(&fs_info->balance_lock);
3904 update_balance_args(bctl);
3905 spin_unlock(&fs_info->balance_lock);
3908 atomic_inc(&fs_info->balance_running);
3909 mutex_unlock(&fs_info->balance_mutex);
3911 ret = __btrfs_balance(fs_info);
3913 mutex_lock(&fs_info->balance_mutex);
3914 atomic_dec(&fs_info->balance_running);
3916 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3917 fs_info->num_tolerated_disk_barrier_failures =
3918 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3922 memset(bargs, 0, sizeof(*bargs));
3923 update_ioctl_balance_args(fs_info, 0, bargs);
3926 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3927 balance_need_close(fs_info)) {
3928 __cancel_balance(fs_info);
3931 wake_up(&fs_info->balance_wait_q);
3935 if (bctl->flags & BTRFS_BALANCE_RESUME)
3936 __cancel_balance(fs_info);
3939 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3944 static int balance_kthread(void *data)
3946 struct btrfs_fs_info *fs_info = data;
3949 mutex_lock(&fs_info->volume_mutex);
3950 mutex_lock(&fs_info->balance_mutex);
3952 if (fs_info->balance_ctl) {
3953 btrfs_info(fs_info, "continuing balance");
3954 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3957 mutex_unlock(&fs_info->balance_mutex);
3958 mutex_unlock(&fs_info->volume_mutex);
3963 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3965 struct task_struct *tsk;
3967 spin_lock(&fs_info->balance_lock);
3968 if (!fs_info->balance_ctl) {
3969 spin_unlock(&fs_info->balance_lock);
3972 spin_unlock(&fs_info->balance_lock);
3974 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
3975 btrfs_info(fs_info, "force skipping balance");
3979 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3980 return PTR_ERR_OR_ZERO(tsk);
3983 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3985 struct btrfs_balance_control *bctl;
3986 struct btrfs_balance_item *item;
3987 struct btrfs_disk_balance_args disk_bargs;
3988 struct btrfs_path *path;
3989 struct extent_buffer *leaf;
3990 struct btrfs_key key;
3993 path = btrfs_alloc_path();
3997 key.objectid = BTRFS_BALANCE_OBJECTID;
3998 key.type = BTRFS_TEMPORARY_ITEM_KEY;
4001 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4004 if (ret > 0) { /* ret = -ENOENT; */
4009 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4015 leaf = path->nodes[0];
4016 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
4018 bctl->fs_info = fs_info;
4019 bctl->flags = btrfs_balance_flags(leaf, item);
4020 bctl->flags |= BTRFS_BALANCE_RESUME;
4022 btrfs_balance_data(leaf, item, &disk_bargs);
4023 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4024 btrfs_balance_meta(leaf, item, &disk_bargs);
4025 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4026 btrfs_balance_sys(leaf, item, &disk_bargs);
4027 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4029 WARN_ON(test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags));
4031 mutex_lock(&fs_info->volume_mutex);
4032 mutex_lock(&fs_info->balance_mutex);
4034 set_balance_control(bctl);
4036 mutex_unlock(&fs_info->balance_mutex);
4037 mutex_unlock(&fs_info->volume_mutex);
4039 btrfs_free_path(path);
4043 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4047 mutex_lock(&fs_info->balance_mutex);
4048 if (!fs_info->balance_ctl) {
4049 mutex_unlock(&fs_info->balance_mutex);
4053 if (atomic_read(&fs_info->balance_running)) {
4054 atomic_inc(&fs_info->balance_pause_req);
4055 mutex_unlock(&fs_info->balance_mutex);
4057 wait_event(fs_info->balance_wait_q,
4058 atomic_read(&fs_info->balance_running) == 0);
4060 mutex_lock(&fs_info->balance_mutex);
4061 /* we are good with balance_ctl ripped off from under us */
4062 BUG_ON(atomic_read(&fs_info->balance_running));
4063 atomic_dec(&fs_info->balance_pause_req);
4068 mutex_unlock(&fs_info->balance_mutex);
4072 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4074 if (fs_info->sb->s_flags & MS_RDONLY)
4077 mutex_lock(&fs_info->balance_mutex);
4078 if (!fs_info->balance_ctl) {
4079 mutex_unlock(&fs_info->balance_mutex);
4083 atomic_inc(&fs_info->balance_cancel_req);
4085 * if we are running just wait and return, balance item is
4086 * deleted in btrfs_balance in this case
4088 if (atomic_read(&fs_info->balance_running)) {
4089 mutex_unlock(&fs_info->balance_mutex);
4090 wait_event(fs_info->balance_wait_q,
4091 atomic_read(&fs_info->balance_running) == 0);
4092 mutex_lock(&fs_info->balance_mutex);
4094 /* __cancel_balance needs volume_mutex */
4095 mutex_unlock(&fs_info->balance_mutex);
4096 mutex_lock(&fs_info->volume_mutex);
4097 mutex_lock(&fs_info->balance_mutex);
4099 if (fs_info->balance_ctl)
4100 __cancel_balance(fs_info);
4102 mutex_unlock(&fs_info->volume_mutex);
4105 BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
4106 atomic_dec(&fs_info->balance_cancel_req);
4107 mutex_unlock(&fs_info->balance_mutex);
4111 static int btrfs_uuid_scan_kthread(void *data)
4113 struct btrfs_fs_info *fs_info = data;
4114 struct btrfs_root *root = fs_info->tree_root;
4115 struct btrfs_key key;
4116 struct btrfs_key max_key;
4117 struct btrfs_path *path = NULL;
4119 struct extent_buffer *eb;
4121 struct btrfs_root_item root_item;
4123 struct btrfs_trans_handle *trans = NULL;
4125 path = btrfs_alloc_path();
4132 key.type = BTRFS_ROOT_ITEM_KEY;
4135 max_key.objectid = (u64)-1;
4136 max_key.type = BTRFS_ROOT_ITEM_KEY;
4137 max_key.offset = (u64)-1;
4140 ret = btrfs_search_forward(root, &key, path, 0);
4147 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4148 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4149 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4150 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4153 eb = path->nodes[0];
4154 slot = path->slots[0];
4155 item_size = btrfs_item_size_nr(eb, slot);
4156 if (item_size < sizeof(root_item))
4159 read_extent_buffer(eb, &root_item,
4160 btrfs_item_ptr_offset(eb, slot),
4161 (int)sizeof(root_item));
4162 if (btrfs_root_refs(&root_item) == 0)
4165 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4166 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4170 btrfs_release_path(path);
4172 * 1 - subvol uuid item
4173 * 1 - received_subvol uuid item
4175 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4176 if (IS_ERR(trans)) {
4177 ret = PTR_ERR(trans);
4185 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4186 ret = btrfs_uuid_tree_add(trans, fs_info,
4188 BTRFS_UUID_KEY_SUBVOL,
4191 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4197 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4198 ret = btrfs_uuid_tree_add(trans, fs_info,
4199 root_item.received_uuid,
4200 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4203 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4211 ret = btrfs_end_transaction(trans);
4217 btrfs_release_path(path);
4218 if (key.offset < (u64)-1) {
4220 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4222 key.type = BTRFS_ROOT_ITEM_KEY;
4223 } else if (key.objectid < (u64)-1) {
4225 key.type = BTRFS_ROOT_ITEM_KEY;
4234 btrfs_free_path(path);
4235 if (trans && !IS_ERR(trans))
4236 btrfs_end_transaction(trans);
4238 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4240 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4241 up(&fs_info->uuid_tree_rescan_sem);
4246 * Callback for btrfs_uuid_tree_iterate().
4248 * 0 check succeeded, the entry is not outdated.
4249 * < 0 if an error occurred.
4250 * > 0 if the check failed, which means the caller shall remove the entry.
4252 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
4253 u8 *uuid, u8 type, u64 subid)
4255 struct btrfs_key key;
4257 struct btrfs_root *subvol_root;
4259 if (type != BTRFS_UUID_KEY_SUBVOL &&
4260 type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
4263 key.objectid = subid;
4264 key.type = BTRFS_ROOT_ITEM_KEY;
4265 key.offset = (u64)-1;
4266 subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
4267 if (IS_ERR(subvol_root)) {
4268 ret = PTR_ERR(subvol_root);
4275 case BTRFS_UUID_KEY_SUBVOL:
4276 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
4279 case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
4280 if (memcmp(uuid, subvol_root->root_item.received_uuid,
4290 static int btrfs_uuid_rescan_kthread(void *data)
4292 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4296 * 1st step is to iterate through the existing UUID tree and
4297 * to delete all entries that contain outdated data.
4298 * 2nd step is to add all missing entries to the UUID tree.
4300 ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4302 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4303 up(&fs_info->uuid_tree_rescan_sem);
4306 return btrfs_uuid_scan_kthread(data);
4309 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4311 struct btrfs_trans_handle *trans;
4312 struct btrfs_root *tree_root = fs_info->tree_root;
4313 struct btrfs_root *uuid_root;
4314 struct task_struct *task;
4321 trans = btrfs_start_transaction(tree_root, 2);
4323 return PTR_ERR(trans);
4325 uuid_root = btrfs_create_tree(trans, fs_info,
4326 BTRFS_UUID_TREE_OBJECTID);
4327 if (IS_ERR(uuid_root)) {
4328 ret = PTR_ERR(uuid_root);
4329 btrfs_abort_transaction(trans, ret);
4330 btrfs_end_transaction(trans);
4334 fs_info->uuid_root = uuid_root;
4336 ret = btrfs_commit_transaction(trans);
4340 down(&fs_info->uuid_tree_rescan_sem);
4341 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4343 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4344 btrfs_warn(fs_info, "failed to start uuid_scan task");
4345 up(&fs_info->uuid_tree_rescan_sem);
4346 return PTR_ERR(task);
4352 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4354 struct task_struct *task;
4356 down(&fs_info->uuid_tree_rescan_sem);
4357 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4359 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4360 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4361 up(&fs_info->uuid_tree_rescan_sem);
4362 return PTR_ERR(task);
4369 * shrinking a device means finding all of the device extents past
4370 * the new size, and then following the back refs to the chunks.
4371 * The chunk relocation code actually frees the device extent
4373 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4375 struct btrfs_fs_info *fs_info = device->fs_info;
4376 struct btrfs_root *root = fs_info->dev_root;
4377 struct btrfs_trans_handle *trans;
4378 struct btrfs_dev_extent *dev_extent = NULL;
4379 struct btrfs_path *path;
4385 bool retried = false;
4386 bool checked_pending_chunks = false;
4387 struct extent_buffer *l;
4388 struct btrfs_key key;
4389 struct btrfs_super_block *super_copy = fs_info->super_copy;
4390 u64 old_total = btrfs_super_total_bytes(super_copy);
4391 u64 old_size = btrfs_device_get_total_bytes(device);
4392 u64 diff = old_size - new_size;
4394 if (device->is_tgtdev_for_dev_replace)
4397 path = btrfs_alloc_path();
4401 path->reada = READA_FORWARD;
4403 mutex_lock(&fs_info->chunk_mutex);
4405 btrfs_device_set_total_bytes(device, new_size);
4406 if (device->writeable) {
4407 device->fs_devices->total_rw_bytes -= diff;
4408 atomic64_sub(diff, &fs_info->free_chunk_space);
4410 mutex_unlock(&fs_info->chunk_mutex);
4413 key.objectid = device->devid;
4414 key.offset = (u64)-1;
4415 key.type = BTRFS_DEV_EXTENT_KEY;
4418 mutex_lock(&fs_info->delete_unused_bgs_mutex);
4419 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4421 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4425 ret = btrfs_previous_item(root, path, 0, key.type);
4427 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4432 btrfs_release_path(path);
4437 slot = path->slots[0];
4438 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4440 if (key.objectid != device->devid) {
4441 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4442 btrfs_release_path(path);
4446 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4447 length = btrfs_dev_extent_length(l, dev_extent);
4449 if (key.offset + length <= new_size) {
4450 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4451 btrfs_release_path(path);
4455 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4456 btrfs_release_path(path);
4458 ret = btrfs_relocate_chunk(fs_info, chunk_offset);
4459 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4460 if (ret && ret != -ENOSPC)
4464 } while (key.offset-- > 0);
4466 if (failed && !retried) {
4470 } else if (failed && retried) {
4475 /* Shrinking succeeded, else we would be at "done". */
4476 trans = btrfs_start_transaction(root, 0);
4477 if (IS_ERR(trans)) {
4478 ret = PTR_ERR(trans);
4482 mutex_lock(&fs_info->chunk_mutex);
4485 * We checked in the above loop all device extents that were already in
4486 * the device tree. However before we have updated the device's
4487 * total_bytes to the new size, we might have had chunk allocations that
4488 * have not complete yet (new block groups attached to transaction
4489 * handles), and therefore their device extents were not yet in the
4490 * device tree and we missed them in the loop above. So if we have any
4491 * pending chunk using a device extent that overlaps the device range
4492 * that we can not use anymore, commit the current transaction and
4493 * repeat the search on the device tree - this way we guarantee we will
4494 * not have chunks using device extents that end beyond 'new_size'.
4496 if (!checked_pending_chunks) {
4497 u64 start = new_size;
4498 u64 len = old_size - new_size;
4500 if (contains_pending_extent(trans->transaction, device,
4502 mutex_unlock(&fs_info->chunk_mutex);
4503 checked_pending_chunks = true;
4506 ret = btrfs_commit_transaction(trans);
4513 btrfs_device_set_disk_total_bytes(device, new_size);
4514 if (list_empty(&device->resized_list))
4515 list_add_tail(&device->resized_list,
4516 &fs_info->fs_devices->resized_devices);
4518 WARN_ON(diff > old_total);
4519 btrfs_set_super_total_bytes(super_copy, old_total - diff);
4520 mutex_unlock(&fs_info->chunk_mutex);
4522 /* Now btrfs_update_device() will change the on-disk size. */
4523 ret = btrfs_update_device(trans, device);
4524 btrfs_end_transaction(trans);
4526 btrfs_free_path(path);
4528 mutex_lock(&fs_info->chunk_mutex);
4529 btrfs_device_set_total_bytes(device, old_size);
4530 if (device->writeable)
4531 device->fs_devices->total_rw_bytes += diff;
4532 atomic64_add(diff, &fs_info->free_chunk_space);
4533 mutex_unlock(&fs_info->chunk_mutex);
4538 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
4539 struct btrfs_key *key,
4540 struct btrfs_chunk *chunk, int item_size)
4542 struct btrfs_super_block *super_copy = fs_info->super_copy;
4543 struct btrfs_disk_key disk_key;
4547 mutex_lock(&fs_info->chunk_mutex);
4548 array_size = btrfs_super_sys_array_size(super_copy);
4549 if (array_size + item_size + sizeof(disk_key)
4550 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4551 mutex_unlock(&fs_info->chunk_mutex);
4555 ptr = super_copy->sys_chunk_array + array_size;
4556 btrfs_cpu_key_to_disk(&disk_key, key);
4557 memcpy(ptr, &disk_key, sizeof(disk_key));
4558 ptr += sizeof(disk_key);
4559 memcpy(ptr, chunk, item_size);
4560 item_size += sizeof(disk_key);
4561 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4562 mutex_unlock(&fs_info->chunk_mutex);
4568 * sort the devices in descending order by max_avail, total_avail
4570 static int btrfs_cmp_device_info(const void *a, const void *b)
4572 const struct btrfs_device_info *di_a = a;
4573 const struct btrfs_device_info *di_b = b;
4575 if (di_a->max_avail > di_b->max_avail)
4577 if (di_a->max_avail < di_b->max_avail)
4579 if (di_a->total_avail > di_b->total_avail)
4581 if (di_a->total_avail < di_b->total_avail)
4586 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
4588 /* TODO allow them to set a preferred stripe size */
4592 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4594 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4597 btrfs_set_fs_incompat(info, RAID56);
4600 #define BTRFS_MAX_DEVS(r) ((BTRFS_MAX_ITEM_SIZE(r->fs_info) \
4601 - sizeof(struct btrfs_chunk)) \
4602 / sizeof(struct btrfs_stripe) + 1)
4604 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
4605 - 2 * sizeof(struct btrfs_disk_key) \
4606 - 2 * sizeof(struct btrfs_chunk)) \
4607 / sizeof(struct btrfs_stripe) + 1)
4609 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4610 u64 start, u64 type)
4612 struct btrfs_fs_info *info = trans->fs_info;
4613 struct btrfs_fs_devices *fs_devices = info->fs_devices;
4614 struct list_head *cur;
4615 struct map_lookup *map = NULL;
4616 struct extent_map_tree *em_tree;
4617 struct extent_map *em;
4618 struct btrfs_device_info *devices_info = NULL;
4620 int num_stripes; /* total number of stripes to allocate */
4621 int data_stripes; /* number of stripes that count for
4623 int sub_stripes; /* sub_stripes info for map */
4624 int dev_stripes; /* stripes per dev */
4625 int devs_max; /* max devs to use */
4626 int devs_min; /* min devs needed */
4627 int devs_increment; /* ndevs has to be a multiple of this */
4628 int ncopies; /* how many copies to data has */
4630 u64 max_stripe_size;
4634 u64 raid_stripe_len = BTRFS_STRIPE_LEN;
4640 BUG_ON(!alloc_profile_is_valid(type, 0));
4642 if (list_empty(&fs_devices->alloc_list))
4645 index = __get_raid_index(type);
4647 sub_stripes = btrfs_raid_array[index].sub_stripes;
4648 dev_stripes = btrfs_raid_array[index].dev_stripes;
4649 devs_max = btrfs_raid_array[index].devs_max;
4650 devs_min = btrfs_raid_array[index].devs_min;
4651 devs_increment = btrfs_raid_array[index].devs_increment;
4652 ncopies = btrfs_raid_array[index].ncopies;
4654 if (type & BTRFS_BLOCK_GROUP_DATA) {
4655 max_stripe_size = SZ_1G;
4656 max_chunk_size = 10 * max_stripe_size;
4658 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4659 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4660 /* for larger filesystems, use larger metadata chunks */
4661 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
4662 max_stripe_size = SZ_1G;
4664 max_stripe_size = SZ_256M;
4665 max_chunk_size = max_stripe_size;
4667 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4668 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4669 max_stripe_size = SZ_32M;
4670 max_chunk_size = 2 * max_stripe_size;
4672 devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4674 btrfs_err(info, "invalid chunk type 0x%llx requested",
4679 /* we don't want a chunk larger than 10% of writeable space */
4680 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4683 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4688 cur = fs_devices->alloc_list.next;
4691 * in the first pass through the devices list, we gather information
4692 * about the available holes on each device.
4695 while (cur != &fs_devices->alloc_list) {
4696 struct btrfs_device *device;
4700 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4704 if (!device->writeable) {
4706 "BTRFS: read-only device in alloc_list\n");
4710 if (!device->in_fs_metadata ||
4711 device->is_tgtdev_for_dev_replace)
4714 if (device->total_bytes > device->bytes_used)
4715 total_avail = device->total_bytes - device->bytes_used;
4719 /* If there is no space on this device, skip it. */
4720 if (total_avail == 0)
4723 ret = find_free_dev_extent(trans, device,
4724 max_stripe_size * dev_stripes,
4725 &dev_offset, &max_avail);
4726 if (ret && ret != -ENOSPC)
4730 max_avail = max_stripe_size * dev_stripes;
4732 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4735 if (ndevs == fs_devices->rw_devices) {
4736 WARN(1, "%s: found more than %llu devices\n",
4737 __func__, fs_devices->rw_devices);
4740 devices_info[ndevs].dev_offset = dev_offset;
4741 devices_info[ndevs].max_avail = max_avail;
4742 devices_info[ndevs].total_avail = total_avail;
4743 devices_info[ndevs].dev = device;
4748 * now sort the devices by hole size / available space
4750 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4751 btrfs_cmp_device_info, NULL);
4753 /* round down to number of usable stripes */
4754 ndevs -= ndevs % devs_increment;
4756 if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4761 if (devs_max && ndevs > devs_max)
4764 * the primary goal is to maximize the number of stripes, so use as many
4765 * devices as possible, even if the stripes are not maximum sized.
4767 stripe_size = devices_info[ndevs-1].max_avail;
4768 num_stripes = ndevs * dev_stripes;
4771 * this will have to be fixed for RAID1 and RAID10 over
4774 data_stripes = num_stripes / ncopies;
4776 if (type & BTRFS_BLOCK_GROUP_RAID5) {
4777 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
4779 data_stripes = num_stripes - 1;
4781 if (type & BTRFS_BLOCK_GROUP_RAID6) {
4782 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
4784 data_stripes = num_stripes - 2;
4788 * Use the number of data stripes to figure out how big this chunk
4789 * is really going to be in terms of logical address space,
4790 * and compare that answer with the max chunk size
4792 if (stripe_size * data_stripes > max_chunk_size) {
4793 u64 mask = (1ULL << 24) - 1;
4795 stripe_size = div_u64(max_chunk_size, data_stripes);
4797 /* bump the answer up to a 16MB boundary */
4798 stripe_size = (stripe_size + mask) & ~mask;
4800 /* but don't go higher than the limits we found
4801 * while searching for free extents
4803 if (stripe_size > devices_info[ndevs-1].max_avail)
4804 stripe_size = devices_info[ndevs-1].max_avail;
4807 stripe_size = div_u64(stripe_size, dev_stripes);
4809 /* align to BTRFS_STRIPE_LEN */
4810 stripe_size = div64_u64(stripe_size, raid_stripe_len);
4811 stripe_size *= raid_stripe_len;
4813 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4818 map->num_stripes = num_stripes;
4820 for (i = 0; i < ndevs; ++i) {
4821 for (j = 0; j < dev_stripes; ++j) {
4822 int s = i * dev_stripes + j;
4823 map->stripes[s].dev = devices_info[i].dev;
4824 map->stripes[s].physical = devices_info[i].dev_offset +
4828 map->sector_size = info->sectorsize;
4829 map->stripe_len = raid_stripe_len;
4830 map->io_align = raid_stripe_len;
4831 map->io_width = raid_stripe_len;
4833 map->sub_stripes = sub_stripes;
4835 num_bytes = stripe_size * data_stripes;
4837 trace_btrfs_chunk_alloc(info, map, start, num_bytes);
4839 em = alloc_extent_map();
4845 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4846 em->map_lookup = map;
4848 em->len = num_bytes;
4849 em->block_start = 0;
4850 em->block_len = em->len;
4851 em->orig_block_len = stripe_size;
4853 em_tree = &info->mapping_tree.map_tree;
4854 write_lock(&em_tree->lock);
4855 ret = add_extent_mapping(em_tree, em, 0);
4857 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4858 refcount_inc(&em->refs);
4860 write_unlock(&em_tree->lock);
4862 free_extent_map(em);
4866 ret = btrfs_make_block_group(trans, info, 0, type,
4867 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4870 goto error_del_extent;
4872 for (i = 0; i < map->num_stripes; i++) {
4873 num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4874 btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4877 atomic64_sub(stripe_size * map->num_stripes, &info->free_chunk_space);
4879 free_extent_map(em);
4880 check_raid56_incompat_flag(info, type);
4882 kfree(devices_info);
4886 write_lock(&em_tree->lock);
4887 remove_extent_mapping(em_tree, em);
4888 write_unlock(&em_tree->lock);
4890 /* One for our allocation */
4891 free_extent_map(em);
4892 /* One for the tree reference */
4893 free_extent_map(em);
4894 /* One for the pending_chunks list reference */
4895 free_extent_map(em);
4897 kfree(devices_info);
4901 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4902 struct btrfs_fs_info *fs_info,
4903 u64 chunk_offset, u64 chunk_size)
4905 struct btrfs_root *extent_root = fs_info->extent_root;
4906 struct btrfs_root *chunk_root = fs_info->chunk_root;
4907 struct btrfs_key key;
4908 struct btrfs_device *device;
4909 struct btrfs_chunk *chunk;
4910 struct btrfs_stripe *stripe;
4911 struct extent_map *em;
4912 struct map_lookup *map;
4919 em = get_chunk_map(fs_info, chunk_offset, chunk_size);
4923 map = em->map_lookup;
4924 item_size = btrfs_chunk_item_size(map->num_stripes);
4925 stripe_size = em->orig_block_len;
4927 chunk = kzalloc(item_size, GFP_NOFS);
4934 * Take the device list mutex to prevent races with the final phase of
4935 * a device replace operation that replaces the device object associated
4936 * with the map's stripes, because the device object's id can change
4937 * at any time during that final phase of the device replace operation
4938 * (dev-replace.c:btrfs_dev_replace_finishing()).
4940 mutex_lock(&fs_info->fs_devices->device_list_mutex);
4941 for (i = 0; i < map->num_stripes; i++) {
4942 device = map->stripes[i].dev;
4943 dev_offset = map->stripes[i].physical;
4945 ret = btrfs_update_device(trans, device);
4948 ret = btrfs_alloc_dev_extent(trans, device,
4949 chunk_root->root_key.objectid,
4950 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4951 chunk_offset, dev_offset,
4957 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4961 stripe = &chunk->stripe;
4962 for (i = 0; i < map->num_stripes; i++) {
4963 device = map->stripes[i].dev;
4964 dev_offset = map->stripes[i].physical;
4966 btrfs_set_stack_stripe_devid(stripe, device->devid);
4967 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4968 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4971 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4973 btrfs_set_stack_chunk_length(chunk, chunk_size);
4974 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4975 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4976 btrfs_set_stack_chunk_type(chunk, map->type);
4977 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4978 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4979 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4980 btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
4981 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4983 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4984 key.type = BTRFS_CHUNK_ITEM_KEY;
4985 key.offset = chunk_offset;
4987 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4988 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4990 * TODO: Cleanup of inserted chunk root in case of
4993 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
4998 free_extent_map(em);
5003 * Chunk allocation falls into two parts. The first part does works
5004 * that make the new allocated chunk useable, but not do any operation
5005 * that modifies the chunk tree. The second part does the works that
5006 * require modifying the chunk tree. This division is important for the
5007 * bootstrap process of adding storage to a seed btrfs.
5009 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
5010 struct btrfs_fs_info *fs_info, u64 type)
5014 ASSERT(mutex_is_locked(&fs_info->chunk_mutex));
5015 chunk_offset = find_next_chunk(fs_info);
5016 return __btrfs_alloc_chunk(trans, chunk_offset, type);
5019 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
5020 struct btrfs_fs_info *fs_info)
5023 u64 sys_chunk_offset;
5027 chunk_offset = find_next_chunk(fs_info);
5028 alloc_profile = btrfs_metadata_alloc_profile(fs_info);
5029 ret = __btrfs_alloc_chunk(trans, chunk_offset, alloc_profile);
5033 sys_chunk_offset = find_next_chunk(fs_info);
5034 alloc_profile = btrfs_system_alloc_profile(fs_info);
5035 ret = __btrfs_alloc_chunk(trans, sys_chunk_offset, alloc_profile);
5039 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
5043 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5044 BTRFS_BLOCK_GROUP_RAID10 |
5045 BTRFS_BLOCK_GROUP_RAID5 |
5046 BTRFS_BLOCK_GROUP_DUP)) {
5048 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
5057 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset)
5059 struct extent_map *em;
5060 struct map_lookup *map;
5065 em = get_chunk_map(fs_info, chunk_offset, 1);
5069 map = em->map_lookup;
5070 for (i = 0; i < map->num_stripes; i++) {
5071 if (map->stripes[i].dev->missing) {
5076 if (!map->stripes[i].dev->writeable) {
5083 * If the number of missing devices is larger than max errors,
5084 * we can not write the data into that chunk successfully, so
5087 if (miss_ndevs > btrfs_chunk_max_errors(map))
5090 free_extent_map(em);
5094 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
5096 extent_map_tree_init(&tree->map_tree);
5099 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
5101 struct extent_map *em;
5104 write_lock(&tree->map_tree.lock);
5105 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
5107 remove_extent_mapping(&tree->map_tree, em);
5108 write_unlock(&tree->map_tree.lock);
5112 free_extent_map(em);
5113 /* once for the tree */
5114 free_extent_map(em);
5118 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5120 struct extent_map *em;
5121 struct map_lookup *map;
5124 em = get_chunk_map(fs_info, logical, len);
5127 * We could return errors for these cases, but that could get
5128 * ugly and we'd probably do the same thing which is just not do
5129 * anything else and exit, so return 1 so the callers don't try
5130 * to use other copies.
5134 map = em->map_lookup;
5135 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
5136 ret = map->num_stripes;
5137 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5138 ret = map->sub_stripes;
5139 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5141 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5145 free_extent_map(em);
5147 btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
5148 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) &&
5149 fs_info->dev_replace.tgtdev)
5151 btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
5156 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
5157 struct btrfs_mapping_tree *map_tree,
5160 struct extent_map *em;
5161 struct map_lookup *map;
5162 unsigned long len = fs_info->sectorsize;
5164 em = get_chunk_map(fs_info, logical, len);
5165 WARN_ON(IS_ERR(em));
5167 map = em->map_lookup;
5168 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5169 len = map->stripe_len * nr_data_stripes(map);
5170 free_extent_map(em);
5174 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info,
5175 u64 logical, u64 len, int mirror_num)
5177 struct extent_map *em;
5178 struct map_lookup *map;
5181 em = get_chunk_map(fs_info, logical, len);
5182 WARN_ON(IS_ERR(em));
5184 map = em->map_lookup;
5185 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5187 free_extent_map(em);
5191 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5192 struct map_lookup *map, int first, int num,
5193 int optimal, int dev_replace_is_ongoing)
5197 struct btrfs_device *srcdev;
5199 if (dev_replace_is_ongoing &&
5200 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5201 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5202 srcdev = fs_info->dev_replace.srcdev;
5207 * try to avoid the drive that is the source drive for a
5208 * dev-replace procedure, only choose it if no other non-missing
5209 * mirror is available
5211 for (tolerance = 0; tolerance < 2; tolerance++) {
5212 if (map->stripes[optimal].dev->bdev &&
5213 (tolerance || map->stripes[optimal].dev != srcdev))
5215 for (i = first; i < first + num; i++) {
5216 if (map->stripes[i].dev->bdev &&
5217 (tolerance || map->stripes[i].dev != srcdev))
5222 /* we couldn't find one that doesn't fail. Just return something
5223 * and the io error handling code will clean up eventually
5228 static inline int parity_smaller(u64 a, u64 b)
5233 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5234 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5236 struct btrfs_bio_stripe s;
5243 for (i = 0; i < num_stripes - 1; i++) {
5244 if (parity_smaller(bbio->raid_map[i],
5245 bbio->raid_map[i+1])) {
5246 s = bbio->stripes[i];
5247 l = bbio->raid_map[i];
5248 bbio->stripes[i] = bbio->stripes[i+1];
5249 bbio->raid_map[i] = bbio->raid_map[i+1];
5250 bbio->stripes[i+1] = s;
5251 bbio->raid_map[i+1] = l;
5259 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5261 struct btrfs_bio *bbio = kzalloc(
5262 /* the size of the btrfs_bio */
5263 sizeof(struct btrfs_bio) +
5264 /* plus the variable array for the stripes */
5265 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5266 /* plus the variable array for the tgt dev */
5267 sizeof(int) * (real_stripes) +
5269 * plus the raid_map, which includes both the tgt dev
5272 sizeof(u64) * (total_stripes),
5273 GFP_NOFS|__GFP_NOFAIL);
5275 atomic_set(&bbio->error, 0);
5276 refcount_set(&bbio->refs, 1);
5281 void btrfs_get_bbio(struct btrfs_bio *bbio)
5283 WARN_ON(!refcount_read(&bbio->refs));
5284 refcount_inc(&bbio->refs);
5287 void btrfs_put_bbio(struct btrfs_bio *bbio)
5291 if (refcount_dec_and_test(&bbio->refs))
5295 /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */
5297 * Please note that, discard won't be sent to target device of device
5300 static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info,
5301 u64 logical, u64 length,
5302 struct btrfs_bio **bbio_ret)
5304 struct extent_map *em;
5305 struct map_lookup *map;
5306 struct btrfs_bio *bbio;
5310 u64 stripe_end_offset;
5317 u32 sub_stripes = 0;
5318 u64 stripes_per_dev = 0;
5319 u32 remaining_stripes = 0;
5320 u32 last_stripe = 0;
5324 /* discard always return a bbio */
5327 em = get_chunk_map(fs_info, logical, length);
5331 map = em->map_lookup;
5332 /* we don't discard raid56 yet */
5333 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5338 offset = logical - em->start;
5339 length = min_t(u64, em->len - offset, length);
5341 stripe_len = map->stripe_len;
5343 * stripe_nr counts the total number of stripes we have to stride
5344 * to get to this block
5346 stripe_nr = div64_u64(offset, stripe_len);
5348 /* stripe_offset is the offset of this block in its stripe */
5349 stripe_offset = offset - stripe_nr * stripe_len;
5351 stripe_nr_end = round_up(offset + length, map->stripe_len);
5352 stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len);
5353 stripe_cnt = stripe_nr_end - stripe_nr;
5354 stripe_end_offset = stripe_nr_end * map->stripe_len -
5357 * after this, stripe_nr is the number of stripes on this
5358 * device we have to walk to find the data, and stripe_index is
5359 * the number of our device in the stripe array
5363 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5364 BTRFS_BLOCK_GROUP_RAID10)) {
5365 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5368 sub_stripes = map->sub_stripes;
5370 factor = map->num_stripes / sub_stripes;
5371 num_stripes = min_t(u64, map->num_stripes,
5372 sub_stripes * stripe_cnt);
5373 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5374 stripe_index *= sub_stripes;
5375 stripes_per_dev = div_u64_rem(stripe_cnt, factor,
5376 &remaining_stripes);
5377 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5378 last_stripe *= sub_stripes;
5379 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5380 BTRFS_BLOCK_GROUP_DUP)) {
5381 num_stripes = map->num_stripes;
5383 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5387 bbio = alloc_btrfs_bio(num_stripes, 0);
5393 for (i = 0; i < num_stripes; i++) {
5394 bbio->stripes[i].physical =
5395 map->stripes[stripe_index].physical +
5396 stripe_offset + stripe_nr * map->stripe_len;
5397 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5399 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5400 BTRFS_BLOCK_GROUP_RAID10)) {
5401 bbio->stripes[i].length = stripes_per_dev *
5404 if (i / sub_stripes < remaining_stripes)
5405 bbio->stripes[i].length +=
5409 * Special for the first stripe and
5412 * |-------|...|-------|
5416 if (i < sub_stripes)
5417 bbio->stripes[i].length -=
5420 if (stripe_index >= last_stripe &&
5421 stripe_index <= (last_stripe +
5423 bbio->stripes[i].length -=
5426 if (i == sub_stripes - 1)
5429 bbio->stripes[i].length = length;
5433 if (stripe_index == map->num_stripes) {
5440 bbio->map_type = map->type;
5441 bbio->num_stripes = num_stripes;
5443 free_extent_map(em);
5448 * In dev-replace case, for repair case (that's the only case where the mirror
5449 * is selected explicitly when calling btrfs_map_block), blocks left of the
5450 * left cursor can also be read from the target drive.
5452 * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the
5454 * For READ, it also needs to be supported using the same mirror number.
5456 * If the requested block is not left of the left cursor, EIO is returned. This
5457 * can happen because btrfs_num_copies() returns one more in the dev-replace
5460 static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info,
5461 u64 logical, u64 length,
5462 u64 srcdev_devid, int *mirror_num,
5465 struct btrfs_bio *bbio = NULL;
5467 int index_srcdev = 0;
5469 u64 physical_of_found = 0;
5473 ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS,
5474 logical, &length, &bbio, 0, 0);
5476 ASSERT(bbio == NULL);
5480 num_stripes = bbio->num_stripes;
5481 if (*mirror_num > num_stripes) {
5483 * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror,
5484 * that means that the requested area is not left of the left
5487 btrfs_put_bbio(bbio);
5492 * process the rest of the function using the mirror_num of the source
5493 * drive. Therefore look it up first. At the end, patch the device
5494 * pointer to the one of the target drive.
5496 for (i = 0; i < num_stripes; i++) {
5497 if (bbio->stripes[i].dev->devid != srcdev_devid)
5501 * In case of DUP, in order to keep it simple, only add the
5502 * mirror with the lowest physical address
5505 physical_of_found <= bbio->stripes[i].physical)
5510 physical_of_found = bbio->stripes[i].physical;
5513 btrfs_put_bbio(bbio);
5519 *mirror_num = index_srcdev + 1;
5520 *physical = physical_of_found;
5524 static void handle_ops_on_dev_replace(enum btrfs_map_op op,
5525 struct btrfs_bio **bbio_ret,
5526 struct btrfs_dev_replace *dev_replace,
5527 int *num_stripes_ret, int *max_errors_ret)
5529 struct btrfs_bio *bbio = *bbio_ret;
5530 u64 srcdev_devid = dev_replace->srcdev->devid;
5531 int tgtdev_indexes = 0;
5532 int num_stripes = *num_stripes_ret;
5533 int max_errors = *max_errors_ret;
5536 if (op == BTRFS_MAP_WRITE) {
5537 int index_where_to_add;
5540 * duplicate the write operations while the dev replace
5541 * procedure is running. Since the copying of the old disk to
5542 * the new disk takes place at run time while the filesystem is
5543 * mounted writable, the regular write operations to the old
5544 * disk have to be duplicated to go to the new disk as well.
5546 * Note that device->missing is handled by the caller, and that
5547 * the write to the old disk is already set up in the stripes
5550 index_where_to_add = num_stripes;
5551 for (i = 0; i < num_stripes; i++) {
5552 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5553 /* write to new disk, too */
5554 struct btrfs_bio_stripe *new =
5555 bbio->stripes + index_where_to_add;
5556 struct btrfs_bio_stripe *old =
5559 new->physical = old->physical;
5560 new->length = old->length;
5561 new->dev = dev_replace->tgtdev;
5562 bbio->tgtdev_map[i] = index_where_to_add;
5563 index_where_to_add++;
5568 num_stripes = index_where_to_add;
5569 } else if (op == BTRFS_MAP_GET_READ_MIRRORS) {
5570 int index_srcdev = 0;
5572 u64 physical_of_found = 0;
5575 * During the dev-replace procedure, the target drive can also
5576 * be used to read data in case it is needed to repair a corrupt
5577 * block elsewhere. This is possible if the requested area is
5578 * left of the left cursor. In this area, the target drive is a
5579 * full copy of the source drive.
5581 for (i = 0; i < num_stripes; i++) {
5582 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5584 * In case of DUP, in order to keep it simple,
5585 * only add the mirror with the lowest physical
5589 physical_of_found <=
5590 bbio->stripes[i].physical)
5594 physical_of_found = bbio->stripes[i].physical;
5598 struct btrfs_bio_stripe *tgtdev_stripe =
5599 bbio->stripes + num_stripes;
5601 tgtdev_stripe->physical = physical_of_found;
5602 tgtdev_stripe->length =
5603 bbio->stripes[index_srcdev].length;
5604 tgtdev_stripe->dev = dev_replace->tgtdev;
5605 bbio->tgtdev_map[index_srcdev] = num_stripes;
5612 *num_stripes_ret = num_stripes;
5613 *max_errors_ret = max_errors;
5614 bbio->num_tgtdevs = tgtdev_indexes;
5618 static bool need_full_stripe(enum btrfs_map_op op)
5620 return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS);
5623 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
5624 enum btrfs_map_op op,
5625 u64 logical, u64 *length,
5626 struct btrfs_bio **bbio_ret,
5627 int mirror_num, int need_raid_map)
5629 struct extent_map *em;
5630 struct map_lookup *map;
5640 int tgtdev_indexes = 0;
5641 struct btrfs_bio *bbio = NULL;
5642 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
5643 int dev_replace_is_ongoing = 0;
5644 int num_alloc_stripes;
5645 int patch_the_first_stripe_for_dev_replace = 0;
5646 u64 physical_to_patch_in_first_stripe = 0;
5647 u64 raid56_full_stripe_start = (u64)-1;
5649 if (op == BTRFS_MAP_DISCARD)
5650 return __btrfs_map_block_for_discard(fs_info, logical,
5653 em = get_chunk_map(fs_info, logical, *length);
5657 map = em->map_lookup;
5658 offset = logical - em->start;
5660 stripe_len = map->stripe_len;
5663 * stripe_nr counts the total number of stripes we have to stride
5664 * to get to this block
5666 stripe_nr = div64_u64(stripe_nr, stripe_len);
5668 stripe_offset = stripe_nr * stripe_len;
5669 if (offset < stripe_offset) {
5671 "stripe math has gone wrong, stripe_offset=%llu, offset=%llu, start=%llu, logical=%llu, stripe_len=%llu",
5672 stripe_offset, offset, em->start, logical,
5674 free_extent_map(em);
5678 /* stripe_offset is the offset of this block in its stripe*/
5679 stripe_offset = offset - stripe_offset;
5681 /* if we're here for raid56, we need to know the stripe aligned start */
5682 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5683 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5684 raid56_full_stripe_start = offset;
5686 /* allow a write of a full stripe, but make sure we don't
5687 * allow straddling of stripes
5689 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5691 raid56_full_stripe_start *= full_stripe_len;
5694 if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5696 /* For writes to RAID[56], allow a full stripeset across all disks.
5697 For other RAID types and for RAID[56] reads, just allow a single
5698 stripe (on a single disk). */
5699 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
5700 (op == BTRFS_MAP_WRITE)) {
5701 max_len = stripe_len * nr_data_stripes(map) -
5702 (offset - raid56_full_stripe_start);
5704 /* we limit the length of each bio to what fits in a stripe */
5705 max_len = stripe_len - stripe_offset;
5707 *length = min_t(u64, em->len - offset, max_len);
5709 *length = em->len - offset;
5712 /* This is for when we're called from btrfs_merge_bio_hook() and all
5713 it cares about is the length */
5717 btrfs_dev_replace_lock(dev_replace, 0);
5718 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5719 if (!dev_replace_is_ongoing)
5720 btrfs_dev_replace_unlock(dev_replace, 0);
5722 btrfs_dev_replace_set_lock_blocking(dev_replace);
5724 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
5725 !need_full_stripe(op) && dev_replace->tgtdev != NULL) {
5726 ret = get_extra_mirror_from_replace(fs_info, logical, *length,
5727 dev_replace->srcdev->devid,
5729 &physical_to_patch_in_first_stripe);
5733 patch_the_first_stripe_for_dev_replace = 1;
5734 } else if (mirror_num > map->num_stripes) {
5740 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5741 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5743 if (op != BTRFS_MAP_WRITE && op != BTRFS_MAP_GET_READ_MIRRORS)
5745 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
5746 if (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS)
5747 num_stripes = map->num_stripes;
5748 else if (mirror_num)
5749 stripe_index = mirror_num - 1;
5751 stripe_index = find_live_mirror(fs_info, map, 0,
5753 current->pid % map->num_stripes,
5754 dev_replace_is_ongoing);
5755 mirror_num = stripe_index + 1;
5758 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
5759 if (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS) {
5760 num_stripes = map->num_stripes;
5761 } else if (mirror_num) {
5762 stripe_index = mirror_num - 1;
5767 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5768 u32 factor = map->num_stripes / map->sub_stripes;
5770 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5771 stripe_index *= map->sub_stripes;
5773 if (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS)
5774 num_stripes = map->sub_stripes;
5775 else if (mirror_num)
5776 stripe_index += mirror_num - 1;
5778 int old_stripe_index = stripe_index;
5779 stripe_index = find_live_mirror(fs_info, map,
5781 map->sub_stripes, stripe_index +
5782 current->pid % map->sub_stripes,
5783 dev_replace_is_ongoing);
5784 mirror_num = stripe_index - old_stripe_index + 1;
5787 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5788 if (need_raid_map &&
5789 (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS ||
5791 /* push stripe_nr back to the start of the full stripe */
5792 stripe_nr = div64_u64(raid56_full_stripe_start,
5793 stripe_len * nr_data_stripes(map));
5795 /* RAID[56] write or recovery. Return all stripes */
5796 num_stripes = map->num_stripes;
5797 max_errors = nr_parity_stripes(map);
5799 *length = map->stripe_len;
5804 * Mirror #0 or #1 means the original data block.
5805 * Mirror #2 is RAID5 parity block.
5806 * Mirror #3 is RAID6 Q block.
5808 stripe_nr = div_u64_rem(stripe_nr,
5809 nr_data_stripes(map), &stripe_index);
5811 stripe_index = nr_data_stripes(map) +
5814 /* We distribute the parity blocks across stripes */
5815 div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
5817 if ((op != BTRFS_MAP_WRITE &&
5818 op != BTRFS_MAP_GET_READ_MIRRORS) &&
5824 * after this, stripe_nr is the number of stripes on this
5825 * device we have to walk to find the data, and stripe_index is
5826 * the number of our device in the stripe array
5828 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5830 mirror_num = stripe_index + 1;
5832 if (stripe_index >= map->num_stripes) {
5834 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
5835 stripe_index, map->num_stripes);
5840 num_alloc_stripes = num_stripes;
5841 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) {
5842 if (op == BTRFS_MAP_WRITE)
5843 num_alloc_stripes <<= 1;
5844 if (op == BTRFS_MAP_GET_READ_MIRRORS)
5845 num_alloc_stripes++;
5846 tgtdev_indexes = num_stripes;
5849 bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
5854 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL)
5855 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
5857 /* build raid_map */
5858 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map &&
5859 (need_full_stripe(op) || mirror_num > 1)) {
5863 bbio->raid_map = (u64 *)((void *)bbio->stripes +
5864 sizeof(struct btrfs_bio_stripe) *
5866 sizeof(int) * tgtdev_indexes);
5868 /* Work out the disk rotation on this stripe-set */
5869 div_u64_rem(stripe_nr, num_stripes, &rot);
5871 /* Fill in the logical address of each stripe */
5872 tmp = stripe_nr * nr_data_stripes(map);
5873 for (i = 0; i < nr_data_stripes(map); i++)
5874 bbio->raid_map[(i+rot) % num_stripes] =
5875 em->start + (tmp + i) * map->stripe_len;
5877 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5878 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5879 bbio->raid_map[(i+rot+1) % num_stripes] =
5884 for (i = 0; i < num_stripes; i++) {
5885 bbio->stripes[i].physical =
5886 map->stripes[stripe_index].physical +
5888 stripe_nr * map->stripe_len;
5889 bbio->stripes[i].dev =
5890 map->stripes[stripe_index].dev;
5894 if (need_full_stripe(op))
5895 max_errors = btrfs_chunk_max_errors(map);
5898 sort_parity_stripes(bbio, num_stripes);
5900 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
5901 need_full_stripe(op)) {
5902 handle_ops_on_dev_replace(op, &bbio, dev_replace, &num_stripes,
5907 bbio->map_type = map->type;
5908 bbio->num_stripes = num_stripes;
5909 bbio->max_errors = max_errors;
5910 bbio->mirror_num = mirror_num;
5913 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5914 * mirror_num == num_stripes + 1 && dev_replace target drive is
5915 * available as a mirror
5917 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5918 WARN_ON(num_stripes > 1);
5919 bbio->stripes[0].dev = dev_replace->tgtdev;
5920 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5921 bbio->mirror_num = map->num_stripes + 1;
5924 if (dev_replace_is_ongoing) {
5925 btrfs_dev_replace_clear_lock_blocking(dev_replace);
5926 btrfs_dev_replace_unlock(dev_replace, 0);
5928 free_extent_map(em);
5932 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
5933 u64 logical, u64 *length,
5934 struct btrfs_bio **bbio_ret, int mirror_num)
5936 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret,
5940 /* For Scrub/replace */
5941 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
5942 u64 logical, u64 *length,
5943 struct btrfs_bio **bbio_ret)
5945 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1);
5948 int btrfs_rmap_block(struct btrfs_fs_info *fs_info,
5949 u64 chunk_start, u64 physical, u64 devid,
5950 u64 **logical, int *naddrs, int *stripe_len)
5952 struct extent_map *em;
5953 struct map_lookup *map;
5961 em = get_chunk_map(fs_info, chunk_start, 1);
5965 map = em->map_lookup;
5967 rmap_len = map->stripe_len;
5969 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5970 length = div_u64(length, map->num_stripes / map->sub_stripes);
5971 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5972 length = div_u64(length, map->num_stripes);
5973 else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5974 length = div_u64(length, nr_data_stripes(map));
5975 rmap_len = map->stripe_len * nr_data_stripes(map);
5978 buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
5979 BUG_ON(!buf); /* -ENOMEM */
5981 for (i = 0; i < map->num_stripes; i++) {
5982 if (devid && map->stripes[i].dev->devid != devid)
5984 if (map->stripes[i].physical > physical ||
5985 map->stripes[i].physical + length <= physical)
5988 stripe_nr = physical - map->stripes[i].physical;
5989 stripe_nr = div64_u64(stripe_nr, map->stripe_len);
5991 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5992 stripe_nr = stripe_nr * map->num_stripes + i;
5993 stripe_nr = div_u64(stripe_nr, map->sub_stripes);
5994 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5995 stripe_nr = stripe_nr * map->num_stripes + i;
5996 } /* else if RAID[56], multiply by nr_data_stripes().
5997 * Alternatively, just use rmap_len below instead of
5998 * map->stripe_len */
6000 bytenr = chunk_start + stripe_nr * rmap_len;
6001 WARN_ON(nr >= map->num_stripes);
6002 for (j = 0; j < nr; j++) {
6003 if (buf[j] == bytenr)
6007 WARN_ON(nr >= map->num_stripes);
6014 *stripe_len = rmap_len;
6016 free_extent_map(em);
6020 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
6022 bio->bi_private = bbio->private;
6023 bio->bi_end_io = bbio->end_io;
6026 btrfs_put_bbio(bbio);
6029 static void btrfs_end_bio(struct bio *bio)
6031 struct btrfs_bio *bbio = bio->bi_private;
6032 int is_orig_bio = 0;
6034 if (bio->bi_error) {
6035 atomic_inc(&bbio->error);
6036 if (bio->bi_error == -EIO || bio->bi_error == -EREMOTEIO) {
6037 unsigned int stripe_index =
6038 btrfs_io_bio(bio)->stripe_index;
6039 struct btrfs_device *dev;
6041 BUG_ON(stripe_index >= bbio->num_stripes);
6042 dev = bbio->stripes[stripe_index].dev;
6044 if (bio_op(bio) == REQ_OP_WRITE)
6045 btrfs_dev_stat_inc(dev,
6046 BTRFS_DEV_STAT_WRITE_ERRS);
6048 btrfs_dev_stat_inc(dev,
6049 BTRFS_DEV_STAT_READ_ERRS);
6050 if (bio->bi_opf & REQ_PREFLUSH)
6051 btrfs_dev_stat_inc(dev,
6052 BTRFS_DEV_STAT_FLUSH_ERRS);
6053 btrfs_dev_stat_print_on_error(dev);
6058 if (bio == bbio->orig_bio)
6061 btrfs_bio_counter_dec(bbio->fs_info);
6063 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6066 bio = bbio->orig_bio;
6069 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6070 /* only send an error to the higher layers if it is
6071 * beyond the tolerance of the btrfs bio
6073 if (atomic_read(&bbio->error) > bbio->max_errors) {
6074 bio->bi_error = -EIO;
6077 * this bio is actually up to date, we didn't
6078 * go over the max number of errors
6083 btrfs_end_bbio(bbio, bio);
6084 } else if (!is_orig_bio) {
6090 * see run_scheduled_bios for a description of why bios are collected for
6093 * This will add one bio to the pending list for a device and make sure
6094 * the work struct is scheduled.
6096 static noinline void btrfs_schedule_bio(struct btrfs_device *device,
6099 struct btrfs_fs_info *fs_info = device->fs_info;
6100 int should_queue = 1;
6101 struct btrfs_pending_bios *pending_bios;
6103 if (device->missing || !device->bdev) {
6108 /* don't bother with additional async steps for reads, right now */
6109 if (bio_op(bio) == REQ_OP_READ) {
6111 btrfsic_submit_bio(bio);
6117 * nr_async_bios allows us to reliably return congestion to the
6118 * higher layers. Otherwise, the async bio makes it appear we have
6119 * made progress against dirty pages when we've really just put it
6120 * on a queue for later
6122 atomic_inc(&fs_info->nr_async_bios);
6123 WARN_ON(bio->bi_next);
6124 bio->bi_next = NULL;
6126 spin_lock(&device->io_lock);
6127 if (op_is_sync(bio->bi_opf))
6128 pending_bios = &device->pending_sync_bios;
6130 pending_bios = &device->pending_bios;
6132 if (pending_bios->tail)
6133 pending_bios->tail->bi_next = bio;
6135 pending_bios->tail = bio;
6136 if (!pending_bios->head)
6137 pending_bios->head = bio;
6138 if (device->running_pending)
6141 spin_unlock(&device->io_lock);
6144 btrfs_queue_work(fs_info->submit_workers, &device->work);
6147 static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio,
6148 u64 physical, int dev_nr, int async)
6150 struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
6151 struct btrfs_fs_info *fs_info = bbio->fs_info;
6153 bio->bi_private = bbio;
6154 btrfs_io_bio(bio)->stripe_index = dev_nr;
6155 bio->bi_end_io = btrfs_end_bio;
6156 bio->bi_iter.bi_sector = physical >> 9;
6159 struct rcu_string *name;
6162 name = rcu_dereference(dev->name);
6163 btrfs_debug(fs_info,
6164 "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
6165 bio_op(bio), bio->bi_opf,
6166 (u64)bio->bi_iter.bi_sector,
6167 (u_long)dev->bdev->bd_dev, name->str, dev->devid,
6168 bio->bi_iter.bi_size);
6172 bio->bi_bdev = dev->bdev;
6174 btrfs_bio_counter_inc_noblocked(fs_info);
6177 btrfs_schedule_bio(dev, bio);
6179 btrfsic_submit_bio(bio);
6182 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6184 atomic_inc(&bbio->error);
6185 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6186 /* Should be the original bio. */
6187 WARN_ON(bio != bbio->orig_bio);
6189 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6190 bio->bi_iter.bi_sector = logical >> 9;
6191 bio->bi_error = -EIO;
6192 btrfs_end_bbio(bbio, bio);
6196 int btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
6197 int mirror_num, int async_submit)
6199 struct btrfs_device *dev;
6200 struct bio *first_bio = bio;
6201 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
6207 struct btrfs_bio *bbio = NULL;
6209 length = bio->bi_iter.bi_size;
6210 map_length = length;
6212 btrfs_bio_counter_inc_blocked(fs_info);
6213 ret = __btrfs_map_block(fs_info, bio_op(bio), logical,
6214 &map_length, &bbio, mirror_num, 1);
6216 btrfs_bio_counter_dec(fs_info);
6220 total_devs = bbio->num_stripes;
6221 bbio->orig_bio = first_bio;
6222 bbio->private = first_bio->bi_private;
6223 bbio->end_io = first_bio->bi_end_io;
6224 bbio->fs_info = fs_info;
6225 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6227 if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6228 ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) {
6229 /* In this case, map_length has been set to the length of
6230 a single stripe; not the whole write */
6231 if (bio_op(bio) == REQ_OP_WRITE) {
6232 ret = raid56_parity_write(fs_info, bio, bbio,
6235 ret = raid56_parity_recover(fs_info, bio, bbio,
6236 map_length, mirror_num, 1);
6239 btrfs_bio_counter_dec(fs_info);
6243 if (map_length < length) {
6245 "mapping failed logical %llu bio len %llu len %llu",
6246 logical, length, map_length);
6250 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6251 dev = bbio->stripes[dev_nr].dev;
6252 if (!dev || !dev->bdev ||
6253 (bio_op(first_bio) == REQ_OP_WRITE && !dev->writeable)) {
6254 bbio_error(bbio, first_bio, logical);
6258 if (dev_nr < total_devs - 1)
6259 bio = btrfs_bio_clone(first_bio);
6263 submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical,
6264 dev_nr, async_submit);
6266 btrfs_bio_counter_dec(fs_info);
6270 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
6273 struct btrfs_device *device;
6274 struct btrfs_fs_devices *cur_devices;
6276 cur_devices = fs_info->fs_devices;
6277 while (cur_devices) {
6279 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
6280 device = __find_device(&cur_devices->devices,
6285 cur_devices = cur_devices->seed;
6290 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices,
6291 u64 devid, u8 *dev_uuid)
6293 struct btrfs_device *device;
6295 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6299 list_add(&device->dev_list, &fs_devices->devices);
6300 device->fs_devices = fs_devices;
6301 fs_devices->num_devices++;
6303 device->missing = 1;
6304 fs_devices->missing_devices++;
6310 * btrfs_alloc_device - allocate struct btrfs_device
6311 * @fs_info: used only for generating a new devid, can be NULL if
6312 * devid is provided (i.e. @devid != NULL).
6313 * @devid: a pointer to devid for this device. If NULL a new devid
6315 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6318 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6319 * on error. Returned struct is not linked onto any lists and can be
6320 * destroyed with kfree() right away.
6322 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6326 struct btrfs_device *dev;
6329 if (WARN_ON(!devid && !fs_info))
6330 return ERR_PTR(-EINVAL);
6332 dev = __alloc_device();
6341 ret = find_next_devid(fs_info, &tmp);
6344 return ERR_PTR(ret);
6350 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6352 generate_random_uuid(dev->uuid);
6354 btrfs_init_work(&dev->work, btrfs_submit_helper,
6355 pending_bios_fn, NULL, NULL);
6360 /* Return -EIO if any error, otherwise return 0. */
6361 static int btrfs_check_chunk_valid(struct btrfs_fs_info *fs_info,
6362 struct extent_buffer *leaf,
6363 struct btrfs_chunk *chunk, u64 logical)
6371 length = btrfs_chunk_length(leaf, chunk);
6372 stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6373 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6374 sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6375 type = btrfs_chunk_type(leaf, chunk);
6378 btrfs_err(fs_info, "invalid chunk num_stripes: %u",
6382 if (!IS_ALIGNED(logical, fs_info->sectorsize)) {
6383 btrfs_err(fs_info, "invalid chunk logical %llu", logical);
6386 if (btrfs_chunk_sector_size(leaf, chunk) != fs_info->sectorsize) {
6387 btrfs_err(fs_info, "invalid chunk sectorsize %u",
6388 btrfs_chunk_sector_size(leaf, chunk));
6391 if (!length || !IS_ALIGNED(length, fs_info->sectorsize)) {
6392 btrfs_err(fs_info, "invalid chunk length %llu", length);
6395 if (!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN) {
6396 btrfs_err(fs_info, "invalid chunk stripe length: %llu",
6400 if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6402 btrfs_err(fs_info, "unrecognized chunk type: %llu",
6403 ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
6404 BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6405 btrfs_chunk_type(leaf, chunk));
6408 if ((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes != 2) ||
6409 (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) ||
6410 (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) ||
6411 (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) ||
6412 (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) ||
6413 ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
6414 num_stripes != 1)) {
6416 "invalid num_stripes:sub_stripes %u:%u for profile %llu",
6417 num_stripes, sub_stripes,
6418 type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
6425 static int read_one_chunk(struct btrfs_fs_info *fs_info, struct btrfs_key *key,
6426 struct extent_buffer *leaf,
6427 struct btrfs_chunk *chunk)
6429 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
6430 struct map_lookup *map;
6431 struct extent_map *em;
6436 u8 uuid[BTRFS_UUID_SIZE];
6441 logical = key->offset;
6442 length = btrfs_chunk_length(leaf, chunk);
6443 stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6444 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6446 ret = btrfs_check_chunk_valid(fs_info, leaf, chunk, logical);
6450 read_lock(&map_tree->map_tree.lock);
6451 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
6452 read_unlock(&map_tree->map_tree.lock);
6454 /* already mapped? */
6455 if (em && em->start <= logical && em->start + em->len > logical) {
6456 free_extent_map(em);
6459 free_extent_map(em);
6462 em = alloc_extent_map();
6465 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6467 free_extent_map(em);
6471 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6472 em->map_lookup = map;
6473 em->start = logical;
6476 em->block_start = 0;
6477 em->block_len = em->len;
6479 map->num_stripes = num_stripes;
6480 map->io_width = btrfs_chunk_io_width(leaf, chunk);
6481 map->io_align = btrfs_chunk_io_align(leaf, chunk);
6482 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
6483 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6484 map->type = btrfs_chunk_type(leaf, chunk);
6485 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6486 for (i = 0; i < num_stripes; i++) {
6487 map->stripes[i].physical =
6488 btrfs_stripe_offset_nr(leaf, chunk, i);
6489 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6490 read_extent_buffer(leaf, uuid, (unsigned long)
6491 btrfs_stripe_dev_uuid_nr(chunk, i),
6493 map->stripes[i].dev = btrfs_find_device(fs_info, devid,
6495 if (!map->stripes[i].dev &&
6496 !btrfs_test_opt(fs_info, DEGRADED)) {
6497 free_extent_map(em);
6500 if (!map->stripes[i].dev) {
6501 map->stripes[i].dev =
6502 add_missing_dev(fs_info->fs_devices, devid,
6504 if (!map->stripes[i].dev) {
6505 free_extent_map(em);
6508 btrfs_warn(fs_info, "devid %llu uuid %pU is missing",
6511 map->stripes[i].dev->in_fs_metadata = 1;
6514 write_lock(&map_tree->map_tree.lock);
6515 ret = add_extent_mapping(&map_tree->map_tree, em, 0);
6516 write_unlock(&map_tree->map_tree.lock);
6517 BUG_ON(ret); /* Tree corruption */
6518 free_extent_map(em);
6523 static void fill_device_from_item(struct extent_buffer *leaf,
6524 struct btrfs_dev_item *dev_item,
6525 struct btrfs_device *device)
6529 device->devid = btrfs_device_id(leaf, dev_item);
6530 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6531 device->total_bytes = device->disk_total_bytes;
6532 device->commit_total_bytes = device->disk_total_bytes;
6533 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6534 device->commit_bytes_used = device->bytes_used;
6535 device->type = btrfs_device_type(leaf, dev_item);
6536 device->io_align = btrfs_device_io_align(leaf, dev_item);
6537 device->io_width = btrfs_device_io_width(leaf, dev_item);
6538 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6539 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6540 device->is_tgtdev_for_dev_replace = 0;
6542 ptr = btrfs_device_uuid(dev_item);
6543 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6546 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info,
6549 struct btrfs_fs_devices *fs_devices;
6552 BUG_ON(!mutex_is_locked(&uuid_mutex));
6554 fs_devices = fs_info->fs_devices->seed;
6555 while (fs_devices) {
6556 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE))
6559 fs_devices = fs_devices->seed;
6562 fs_devices = find_fsid(fsid);
6564 if (!btrfs_test_opt(fs_info, DEGRADED))
6565 return ERR_PTR(-ENOENT);
6567 fs_devices = alloc_fs_devices(fsid);
6568 if (IS_ERR(fs_devices))
6571 fs_devices->seeding = 1;
6572 fs_devices->opened = 1;
6576 fs_devices = clone_fs_devices(fs_devices);
6577 if (IS_ERR(fs_devices))
6580 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
6581 fs_info->bdev_holder);
6583 free_fs_devices(fs_devices);
6584 fs_devices = ERR_PTR(ret);
6588 if (!fs_devices->seeding) {
6589 __btrfs_close_devices(fs_devices);
6590 free_fs_devices(fs_devices);
6591 fs_devices = ERR_PTR(-EINVAL);
6595 fs_devices->seed = fs_info->fs_devices->seed;
6596 fs_info->fs_devices->seed = fs_devices;
6601 static int read_one_dev(struct btrfs_fs_info *fs_info,
6602 struct extent_buffer *leaf,
6603 struct btrfs_dev_item *dev_item)
6605 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6606 struct btrfs_device *device;
6609 u8 fs_uuid[BTRFS_UUID_SIZE];
6610 u8 dev_uuid[BTRFS_UUID_SIZE];
6612 devid = btrfs_device_id(leaf, dev_item);
6613 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6615 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6618 if (memcmp(fs_uuid, fs_info->fsid, BTRFS_UUID_SIZE)) {
6619 fs_devices = open_seed_devices(fs_info, fs_uuid);
6620 if (IS_ERR(fs_devices))
6621 return PTR_ERR(fs_devices);
6624 device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
6626 if (!btrfs_test_opt(fs_info, DEGRADED))
6629 device = add_missing_dev(fs_devices, devid, dev_uuid);
6632 btrfs_warn(fs_info, "devid %llu uuid %pU missing",
6635 if (!device->bdev && !btrfs_test_opt(fs_info, DEGRADED))
6638 if(!device->bdev && !device->missing) {
6640 * this happens when a device that was properly setup
6641 * in the device info lists suddenly goes bad.
6642 * device->bdev is NULL, and so we have to set
6643 * device->missing to one here
6645 device->fs_devices->missing_devices++;
6646 device->missing = 1;
6649 /* Move the device to its own fs_devices */
6650 if (device->fs_devices != fs_devices) {
6651 ASSERT(device->missing);
6653 list_move(&device->dev_list, &fs_devices->devices);
6654 device->fs_devices->num_devices--;
6655 fs_devices->num_devices++;
6657 device->fs_devices->missing_devices--;
6658 fs_devices->missing_devices++;
6660 device->fs_devices = fs_devices;
6664 if (device->fs_devices != fs_info->fs_devices) {
6665 BUG_ON(device->writeable);
6666 if (device->generation !=
6667 btrfs_device_generation(leaf, dev_item))
6671 fill_device_from_item(leaf, dev_item, device);
6672 device->in_fs_metadata = 1;
6673 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
6674 device->fs_devices->total_rw_bytes += device->total_bytes;
6675 atomic64_add(device->total_bytes - device->bytes_used,
6676 &fs_info->free_chunk_space);
6682 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
6684 struct btrfs_root *root = fs_info->tree_root;
6685 struct btrfs_super_block *super_copy = fs_info->super_copy;
6686 struct extent_buffer *sb;
6687 struct btrfs_disk_key *disk_key;
6688 struct btrfs_chunk *chunk;
6690 unsigned long sb_array_offset;
6697 struct btrfs_key key;
6699 ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize);
6701 * This will create extent buffer of nodesize, superblock size is
6702 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6703 * overallocate but we can keep it as-is, only the first page is used.
6705 sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET);
6708 set_extent_buffer_uptodate(sb);
6709 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6711 * The sb extent buffer is artificial and just used to read the system array.
6712 * set_extent_buffer_uptodate() call does not properly mark all it's
6713 * pages up-to-date when the page is larger: extent does not cover the
6714 * whole page and consequently check_page_uptodate does not find all
6715 * the page's extents up-to-date (the hole beyond sb),
6716 * write_extent_buffer then triggers a WARN_ON.
6718 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6719 * but sb spans only this function. Add an explicit SetPageUptodate call
6720 * to silence the warning eg. on PowerPC 64.
6722 if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
6723 SetPageUptodate(sb->pages[0]);
6725 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6726 array_size = btrfs_super_sys_array_size(super_copy);
6728 array_ptr = super_copy->sys_chunk_array;
6729 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6732 while (cur_offset < array_size) {
6733 disk_key = (struct btrfs_disk_key *)array_ptr;
6734 len = sizeof(*disk_key);
6735 if (cur_offset + len > array_size)
6736 goto out_short_read;
6738 btrfs_disk_key_to_cpu(&key, disk_key);
6741 sb_array_offset += len;
6744 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6745 chunk = (struct btrfs_chunk *)sb_array_offset;
6747 * At least one btrfs_chunk with one stripe must be
6748 * present, exact stripe count check comes afterwards
6750 len = btrfs_chunk_item_size(1);
6751 if (cur_offset + len > array_size)
6752 goto out_short_read;
6754 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6757 "invalid number of stripes %u in sys_array at offset %u",
6758 num_stripes, cur_offset);
6763 type = btrfs_chunk_type(sb, chunk);
6764 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
6766 "invalid chunk type %llu in sys_array at offset %u",
6772 len = btrfs_chunk_item_size(num_stripes);
6773 if (cur_offset + len > array_size)
6774 goto out_short_read;
6776 ret = read_one_chunk(fs_info, &key, sb, chunk);
6781 "unexpected item type %u in sys_array at offset %u",
6782 (u32)key.type, cur_offset);
6787 sb_array_offset += len;
6790 clear_extent_buffer_uptodate(sb);
6791 free_extent_buffer_stale(sb);
6795 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
6797 clear_extent_buffer_uptodate(sb);
6798 free_extent_buffer_stale(sb);
6802 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
6804 struct btrfs_root *root = fs_info->chunk_root;
6805 struct btrfs_path *path;
6806 struct extent_buffer *leaf;
6807 struct btrfs_key key;
6808 struct btrfs_key found_key;
6813 path = btrfs_alloc_path();
6817 mutex_lock(&uuid_mutex);
6818 mutex_lock(&fs_info->chunk_mutex);
6821 * Read all device items, and then all the chunk items. All
6822 * device items are found before any chunk item (their object id
6823 * is smaller than the lowest possible object id for a chunk
6824 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6826 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6829 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6833 leaf = path->nodes[0];
6834 slot = path->slots[0];
6835 if (slot >= btrfs_header_nritems(leaf)) {
6836 ret = btrfs_next_leaf(root, path);
6843 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6844 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6845 struct btrfs_dev_item *dev_item;
6846 dev_item = btrfs_item_ptr(leaf, slot,
6847 struct btrfs_dev_item);
6848 ret = read_one_dev(fs_info, leaf, dev_item);
6852 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6853 struct btrfs_chunk *chunk;
6854 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6855 ret = read_one_chunk(fs_info, &found_key, leaf, chunk);
6863 * After loading chunk tree, we've got all device information,
6864 * do another round of validation checks.
6866 if (total_dev != fs_info->fs_devices->total_devices) {
6868 "super_num_devices %llu mismatch with num_devices %llu found here",
6869 btrfs_super_num_devices(fs_info->super_copy),
6874 if (btrfs_super_total_bytes(fs_info->super_copy) <
6875 fs_info->fs_devices->total_rw_bytes) {
6877 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
6878 btrfs_super_total_bytes(fs_info->super_copy),
6879 fs_info->fs_devices->total_rw_bytes);
6885 mutex_unlock(&fs_info->chunk_mutex);
6886 mutex_unlock(&uuid_mutex);
6888 btrfs_free_path(path);
6892 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6894 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6895 struct btrfs_device *device;
6897 while (fs_devices) {
6898 mutex_lock(&fs_devices->device_list_mutex);
6899 list_for_each_entry(device, &fs_devices->devices, dev_list)
6900 device->fs_info = fs_info;
6901 mutex_unlock(&fs_devices->device_list_mutex);
6903 fs_devices = fs_devices->seed;
6907 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6911 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6912 btrfs_dev_stat_reset(dev, i);
6915 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6917 struct btrfs_key key;
6918 struct btrfs_key found_key;
6919 struct btrfs_root *dev_root = fs_info->dev_root;
6920 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6921 struct extent_buffer *eb;
6924 struct btrfs_device *device;
6925 struct btrfs_path *path = NULL;
6928 path = btrfs_alloc_path();
6934 mutex_lock(&fs_devices->device_list_mutex);
6935 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6937 struct btrfs_dev_stats_item *ptr;
6939 key.objectid = BTRFS_DEV_STATS_OBJECTID;
6940 key.type = BTRFS_PERSISTENT_ITEM_KEY;
6941 key.offset = device->devid;
6942 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6944 __btrfs_reset_dev_stats(device);
6945 device->dev_stats_valid = 1;
6946 btrfs_release_path(path);
6949 slot = path->slots[0];
6950 eb = path->nodes[0];
6951 btrfs_item_key_to_cpu(eb, &found_key, slot);
6952 item_size = btrfs_item_size_nr(eb, slot);
6954 ptr = btrfs_item_ptr(eb, slot,
6955 struct btrfs_dev_stats_item);
6957 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6958 if (item_size >= (1 + i) * sizeof(__le64))
6959 btrfs_dev_stat_set(device, i,
6960 btrfs_dev_stats_value(eb, ptr, i));
6962 btrfs_dev_stat_reset(device, i);
6965 device->dev_stats_valid = 1;
6966 btrfs_dev_stat_print_on_load(device);
6967 btrfs_release_path(path);
6969 mutex_unlock(&fs_devices->device_list_mutex);
6972 btrfs_free_path(path);
6973 return ret < 0 ? ret : 0;
6976 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
6977 struct btrfs_fs_info *fs_info,
6978 struct btrfs_device *device)
6980 struct btrfs_root *dev_root = fs_info->dev_root;
6981 struct btrfs_path *path;
6982 struct btrfs_key key;
6983 struct extent_buffer *eb;
6984 struct btrfs_dev_stats_item *ptr;
6988 key.objectid = BTRFS_DEV_STATS_OBJECTID;
6989 key.type = BTRFS_PERSISTENT_ITEM_KEY;
6990 key.offset = device->devid;
6992 path = btrfs_alloc_path();
6995 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
6997 btrfs_warn_in_rcu(fs_info,
6998 "error %d while searching for dev_stats item for device %s",
6999 ret, rcu_str_deref(device->name));
7004 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
7005 /* need to delete old one and insert a new one */
7006 ret = btrfs_del_item(trans, dev_root, path);
7008 btrfs_warn_in_rcu(fs_info,
7009 "delete too small dev_stats item for device %s failed %d",
7010 rcu_str_deref(device->name), ret);
7017 /* need to insert a new item */
7018 btrfs_release_path(path);
7019 ret = btrfs_insert_empty_item(trans, dev_root, path,
7020 &key, sizeof(*ptr));
7022 btrfs_warn_in_rcu(fs_info,
7023 "insert dev_stats item for device %s failed %d",
7024 rcu_str_deref(device->name), ret);
7029 eb = path->nodes[0];
7030 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7031 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7032 btrfs_set_dev_stats_value(eb, ptr, i,
7033 btrfs_dev_stat_read(device, i));
7034 btrfs_mark_buffer_dirty(eb);
7037 btrfs_free_path(path);
7042 * called from commit_transaction. Writes all changed device stats to disk.
7044 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
7045 struct btrfs_fs_info *fs_info)
7047 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7048 struct btrfs_device *device;
7052 mutex_lock(&fs_devices->device_list_mutex);
7053 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7054 if (!device->dev_stats_valid || !btrfs_dev_stats_dirty(device))
7057 stats_cnt = atomic_read(&device->dev_stats_ccnt);
7058 ret = update_dev_stat_item(trans, fs_info, device);
7060 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7062 mutex_unlock(&fs_devices->device_list_mutex);
7067 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7069 btrfs_dev_stat_inc(dev, index);
7070 btrfs_dev_stat_print_on_error(dev);
7073 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
7075 if (!dev->dev_stats_valid)
7077 btrfs_err_rl_in_rcu(dev->fs_info,
7078 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7079 rcu_str_deref(dev->name),
7080 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7081 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7082 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7083 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7084 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7087 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7091 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7092 if (btrfs_dev_stat_read(dev, i) != 0)
7094 if (i == BTRFS_DEV_STAT_VALUES_MAX)
7095 return; /* all values == 0, suppress message */
7097 btrfs_info_in_rcu(dev->fs_info,
7098 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7099 rcu_str_deref(dev->name),
7100 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7101 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7102 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7103 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7104 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7107 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
7108 struct btrfs_ioctl_get_dev_stats *stats)
7110 struct btrfs_device *dev;
7111 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7114 mutex_lock(&fs_devices->device_list_mutex);
7115 dev = btrfs_find_device(fs_info, stats->devid, NULL, NULL);
7116 mutex_unlock(&fs_devices->device_list_mutex);
7119 btrfs_warn(fs_info, "get dev_stats failed, device not found");
7121 } else if (!dev->dev_stats_valid) {
7122 btrfs_warn(fs_info, "get dev_stats failed, not yet valid");
7124 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7125 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7126 if (stats->nr_items > i)
7128 btrfs_dev_stat_read_and_reset(dev, i);
7130 btrfs_dev_stat_reset(dev, i);
7133 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7134 if (stats->nr_items > i)
7135 stats->values[i] = btrfs_dev_stat_read(dev, i);
7137 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7138 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7142 void btrfs_scratch_superblocks(struct block_device *bdev, const char *device_path)
7144 struct buffer_head *bh;
7145 struct btrfs_super_block *disk_super;
7151 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX;
7154 if (btrfs_read_dev_one_super(bdev, copy_num, &bh))
7157 disk_super = (struct btrfs_super_block *)bh->b_data;
7159 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
7160 set_buffer_dirty(bh);
7161 sync_dirty_buffer(bh);
7165 /* Notify udev that device has changed */
7166 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
7168 /* Update ctime/mtime for device path for libblkid */
7169 update_dev_time(device_path);
7173 * Update the size of all devices, which is used for writing out the
7176 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
7178 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7179 struct btrfs_device *curr, *next;
7181 if (list_empty(&fs_devices->resized_devices))
7184 mutex_lock(&fs_devices->device_list_mutex);
7185 mutex_lock(&fs_info->chunk_mutex);
7186 list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
7188 list_del_init(&curr->resized_list);
7189 curr->commit_total_bytes = curr->disk_total_bytes;
7191 mutex_unlock(&fs_info->chunk_mutex);
7192 mutex_unlock(&fs_devices->device_list_mutex);
7195 /* Must be invoked during the transaction commit */
7196 void btrfs_update_commit_device_bytes_used(struct btrfs_fs_info *fs_info,
7197 struct btrfs_transaction *transaction)
7199 struct extent_map *em;
7200 struct map_lookup *map;
7201 struct btrfs_device *dev;
7204 if (list_empty(&transaction->pending_chunks))
7207 /* In order to kick the device replace finish process */
7208 mutex_lock(&fs_info->chunk_mutex);
7209 list_for_each_entry(em, &transaction->pending_chunks, list) {
7210 map = em->map_lookup;
7212 for (i = 0; i < map->num_stripes; i++) {
7213 dev = map->stripes[i].dev;
7214 dev->commit_bytes_used = dev->bytes_used;
7217 mutex_unlock(&fs_info->chunk_mutex);
7220 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
7222 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7223 while (fs_devices) {
7224 fs_devices->fs_info = fs_info;
7225 fs_devices = fs_devices->seed;
7229 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
7231 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7232 while (fs_devices) {
7233 fs_devices->fs_info = NULL;
7234 fs_devices = fs_devices->seed;