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_KERNEL);
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;
1017 device->bdev = bdev;
1018 device->in_fs_metadata = 0;
1019 device->mode = flags;
1021 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1022 fs_devices->rotating = 1;
1024 fs_devices->open_devices++;
1025 if (device->writeable &&
1026 device->devid != BTRFS_DEV_REPLACE_DEVID) {
1027 fs_devices->rw_devices++;
1028 list_add(&device->dev_alloc_list,
1029 &fs_devices->alloc_list);
1036 blkdev_put(bdev, flags);
1039 if (fs_devices->open_devices == 0) {
1043 fs_devices->seeding = seeding;
1044 fs_devices->opened = 1;
1045 fs_devices->latest_bdev = latest_dev->bdev;
1046 fs_devices->total_rw_bytes = 0;
1051 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1052 fmode_t flags, void *holder)
1056 mutex_lock(&uuid_mutex);
1057 if (fs_devices->opened) {
1058 fs_devices->opened++;
1061 ret = __btrfs_open_devices(fs_devices, flags, holder);
1063 mutex_unlock(&uuid_mutex);
1067 void btrfs_release_disk_super(struct page *page)
1073 int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr,
1074 struct page **page, struct btrfs_super_block **disk_super)
1079 /* make sure our super fits in the device */
1080 if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1083 /* make sure our super fits in the page */
1084 if (sizeof(**disk_super) > PAGE_SIZE)
1087 /* make sure our super doesn't straddle pages on disk */
1088 index = bytenr >> PAGE_SHIFT;
1089 if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index)
1092 /* pull in the page with our super */
1093 *page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
1096 if (IS_ERR_OR_NULL(*page))
1101 /* align our pointer to the offset of the super block */
1102 *disk_super = p + (bytenr & ~PAGE_MASK);
1104 if (btrfs_super_bytenr(*disk_super) != bytenr ||
1105 btrfs_super_magic(*disk_super) != BTRFS_MAGIC) {
1106 btrfs_release_disk_super(*page);
1110 if ((*disk_super)->label[0] &&
1111 (*disk_super)->label[BTRFS_LABEL_SIZE - 1])
1112 (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0';
1118 * Look for a btrfs signature on a device. This may be called out of the mount path
1119 * and we are not allowed to call set_blocksize during the scan. The superblock
1120 * is read via pagecache
1122 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
1123 struct btrfs_fs_devices **fs_devices_ret)
1125 struct btrfs_super_block *disk_super;
1126 struct block_device *bdev;
1135 * we would like to check all the supers, but that would make
1136 * a btrfs mount succeed after a mkfs from a different FS.
1137 * So, we need to add a special mount option to scan for
1138 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1140 bytenr = btrfs_sb_offset(0);
1141 flags |= FMODE_EXCL;
1142 mutex_lock(&uuid_mutex);
1144 bdev = blkdev_get_by_path(path, flags, holder);
1146 ret = PTR_ERR(bdev);
1150 if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super))
1151 goto error_bdev_put;
1153 devid = btrfs_stack_device_id(&disk_super->dev_item);
1154 transid = btrfs_super_generation(disk_super);
1155 total_devices = btrfs_super_num_devices(disk_super);
1157 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
1159 if (disk_super->label[0]) {
1160 pr_info("BTRFS: device label %s ", disk_super->label);
1162 pr_info("BTRFS: device fsid %pU ", disk_super->fsid);
1165 pr_cont("devid %llu transid %llu %s\n", devid, transid, path);
1168 if (!ret && fs_devices_ret)
1169 (*fs_devices_ret)->total_devices = total_devices;
1171 btrfs_release_disk_super(page);
1174 blkdev_put(bdev, flags);
1176 mutex_unlock(&uuid_mutex);
1180 /* helper to account the used device space in the range */
1181 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
1182 u64 end, u64 *length)
1184 struct btrfs_key key;
1185 struct btrfs_root *root = device->fs_info->dev_root;
1186 struct btrfs_dev_extent *dev_extent;
1187 struct btrfs_path *path;
1191 struct extent_buffer *l;
1195 if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
1198 path = btrfs_alloc_path();
1201 path->reada = READA_FORWARD;
1203 key.objectid = device->devid;
1205 key.type = BTRFS_DEV_EXTENT_KEY;
1207 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1211 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1218 slot = path->slots[0];
1219 if (slot >= btrfs_header_nritems(l)) {
1220 ret = btrfs_next_leaf(root, path);
1228 btrfs_item_key_to_cpu(l, &key, slot);
1230 if (key.objectid < device->devid)
1233 if (key.objectid > device->devid)
1236 if (key.type != BTRFS_DEV_EXTENT_KEY)
1239 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1240 extent_end = key.offset + btrfs_dev_extent_length(l,
1242 if (key.offset <= start && extent_end > end) {
1243 *length = end - start + 1;
1245 } else if (key.offset <= start && extent_end > start)
1246 *length += extent_end - start;
1247 else if (key.offset > start && extent_end <= end)
1248 *length += extent_end - key.offset;
1249 else if (key.offset > start && key.offset <= end) {
1250 *length += end - key.offset + 1;
1252 } else if (key.offset > end)
1260 btrfs_free_path(path);
1264 static int contains_pending_extent(struct btrfs_transaction *transaction,
1265 struct btrfs_device *device,
1266 u64 *start, u64 len)
1268 struct btrfs_fs_info *fs_info = device->fs_info;
1269 struct extent_map *em;
1270 struct list_head *search_list = &fs_info->pinned_chunks;
1272 u64 physical_start = *start;
1275 search_list = &transaction->pending_chunks;
1277 list_for_each_entry(em, search_list, list) {
1278 struct map_lookup *map;
1281 map = em->map_lookup;
1282 for (i = 0; i < map->num_stripes; i++) {
1285 if (map->stripes[i].dev != device)
1287 if (map->stripes[i].physical >= physical_start + len ||
1288 map->stripes[i].physical + em->orig_block_len <=
1292 * Make sure that while processing the pinned list we do
1293 * not override our *start with a lower value, because
1294 * we can have pinned chunks that fall within this
1295 * device hole and that have lower physical addresses
1296 * than the pending chunks we processed before. If we
1297 * do not take this special care we can end up getting
1298 * 2 pending chunks that start at the same physical
1299 * device offsets because the end offset of a pinned
1300 * chunk can be equal to the start offset of some
1303 end = map->stripes[i].physical + em->orig_block_len;
1310 if (search_list != &fs_info->pinned_chunks) {
1311 search_list = &fs_info->pinned_chunks;
1320 * find_free_dev_extent_start - find free space in the specified device
1321 * @device: the device which we search the free space in
1322 * @num_bytes: the size of the free space that we need
1323 * @search_start: the position from which to begin the search
1324 * @start: store the start of the free space.
1325 * @len: the size of the free space. that we find, or the size
1326 * of the max free space if we don't find suitable free space
1328 * this uses a pretty simple search, the expectation is that it is
1329 * called very infrequently and that a given device has a small number
1332 * @start is used to store the start of the free space if we find. But if we
1333 * don't find suitable free space, it will be used to store the start position
1334 * of the max free space.
1336 * @len is used to store the size of the free space that we find.
1337 * But if we don't find suitable free space, it is used to store the size of
1338 * the max free space.
1340 int find_free_dev_extent_start(struct btrfs_transaction *transaction,
1341 struct btrfs_device *device, u64 num_bytes,
1342 u64 search_start, u64 *start, u64 *len)
1344 struct btrfs_fs_info *fs_info = device->fs_info;
1345 struct btrfs_root *root = fs_info->dev_root;
1346 struct btrfs_key key;
1347 struct btrfs_dev_extent *dev_extent;
1348 struct btrfs_path *path;
1353 u64 search_end = device->total_bytes;
1356 struct extent_buffer *l;
1357 u64 min_search_start;
1360 * We don't want to overwrite the superblock on the drive nor any area
1361 * used by the boot loader (grub for example), so we make sure to start
1362 * at an offset of at least 1MB.
1364 min_search_start = max(fs_info->alloc_start, 1024ull * 1024);
1365 search_start = max(search_start, min_search_start);
1367 path = btrfs_alloc_path();
1371 max_hole_start = search_start;
1375 if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1380 path->reada = READA_FORWARD;
1381 path->search_commit_root = 1;
1382 path->skip_locking = 1;
1384 key.objectid = device->devid;
1385 key.offset = search_start;
1386 key.type = BTRFS_DEV_EXTENT_KEY;
1388 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1392 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1399 slot = path->slots[0];
1400 if (slot >= btrfs_header_nritems(l)) {
1401 ret = btrfs_next_leaf(root, path);
1409 btrfs_item_key_to_cpu(l, &key, slot);
1411 if (key.objectid < device->devid)
1414 if (key.objectid > device->devid)
1417 if (key.type != BTRFS_DEV_EXTENT_KEY)
1420 if (key.offset > search_start) {
1421 hole_size = key.offset - search_start;
1424 * Have to check before we set max_hole_start, otherwise
1425 * we could end up sending back this offset anyway.
1427 if (contains_pending_extent(transaction, device,
1430 if (key.offset >= search_start) {
1431 hole_size = key.offset - search_start;
1438 if (hole_size > max_hole_size) {
1439 max_hole_start = search_start;
1440 max_hole_size = hole_size;
1444 * If this free space is greater than which we need,
1445 * it must be the max free space that we have found
1446 * until now, so max_hole_start must point to the start
1447 * of this free space and the length of this free space
1448 * is stored in max_hole_size. Thus, we return
1449 * max_hole_start and max_hole_size and go back to the
1452 if (hole_size >= num_bytes) {
1458 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1459 extent_end = key.offset + btrfs_dev_extent_length(l,
1461 if (extent_end > search_start)
1462 search_start = extent_end;
1469 * At this point, search_start should be the end of
1470 * allocated dev extents, and when shrinking the device,
1471 * search_end may be smaller than search_start.
1473 if (search_end > search_start) {
1474 hole_size = search_end - search_start;
1476 if (contains_pending_extent(transaction, device, &search_start,
1478 btrfs_release_path(path);
1482 if (hole_size > max_hole_size) {
1483 max_hole_start = search_start;
1484 max_hole_size = hole_size;
1489 if (max_hole_size < num_bytes)
1495 btrfs_free_path(path);
1496 *start = max_hole_start;
1498 *len = max_hole_size;
1502 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1503 struct btrfs_device *device, u64 num_bytes,
1504 u64 *start, u64 *len)
1506 /* FIXME use last free of some kind */
1507 return find_free_dev_extent_start(trans->transaction, device,
1508 num_bytes, 0, start, len);
1511 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1512 struct btrfs_device *device,
1513 u64 start, u64 *dev_extent_len)
1515 struct btrfs_fs_info *fs_info = device->fs_info;
1516 struct btrfs_root *root = fs_info->dev_root;
1518 struct btrfs_path *path;
1519 struct btrfs_key key;
1520 struct btrfs_key found_key;
1521 struct extent_buffer *leaf = NULL;
1522 struct btrfs_dev_extent *extent = NULL;
1524 path = btrfs_alloc_path();
1528 key.objectid = device->devid;
1530 key.type = BTRFS_DEV_EXTENT_KEY;
1532 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1534 ret = btrfs_previous_item(root, path, key.objectid,
1535 BTRFS_DEV_EXTENT_KEY);
1538 leaf = path->nodes[0];
1539 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1540 extent = btrfs_item_ptr(leaf, path->slots[0],
1541 struct btrfs_dev_extent);
1542 BUG_ON(found_key.offset > start || found_key.offset +
1543 btrfs_dev_extent_length(leaf, extent) < start);
1545 btrfs_release_path(path);
1547 } else if (ret == 0) {
1548 leaf = path->nodes[0];
1549 extent = btrfs_item_ptr(leaf, path->slots[0],
1550 struct btrfs_dev_extent);
1552 btrfs_handle_fs_error(fs_info, ret, "Slot search failed");
1556 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1558 ret = btrfs_del_item(trans, root, path);
1560 btrfs_handle_fs_error(fs_info, ret,
1561 "Failed to remove dev extent item");
1563 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1566 btrfs_free_path(path);
1570 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1571 struct btrfs_device *device,
1572 u64 chunk_tree, u64 chunk_objectid,
1573 u64 chunk_offset, u64 start, u64 num_bytes)
1576 struct btrfs_path *path;
1577 struct btrfs_fs_info *fs_info = device->fs_info;
1578 struct btrfs_root *root = fs_info->dev_root;
1579 struct btrfs_dev_extent *extent;
1580 struct extent_buffer *leaf;
1581 struct btrfs_key key;
1583 WARN_ON(!device->in_fs_metadata);
1584 WARN_ON(device->is_tgtdev_for_dev_replace);
1585 path = btrfs_alloc_path();
1589 key.objectid = device->devid;
1591 key.type = BTRFS_DEV_EXTENT_KEY;
1592 ret = btrfs_insert_empty_item(trans, root, path, &key,
1597 leaf = path->nodes[0];
1598 extent = btrfs_item_ptr(leaf, path->slots[0],
1599 struct btrfs_dev_extent);
1600 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1601 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1602 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1604 write_extent_buffer_chunk_tree_uuid(leaf, fs_info->chunk_tree_uuid);
1606 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1607 btrfs_mark_buffer_dirty(leaf);
1609 btrfs_free_path(path);
1613 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1615 struct extent_map_tree *em_tree;
1616 struct extent_map *em;
1620 em_tree = &fs_info->mapping_tree.map_tree;
1621 read_lock(&em_tree->lock);
1622 n = rb_last(&em_tree->map);
1624 em = rb_entry(n, struct extent_map, rb_node);
1625 ret = em->start + em->len;
1627 read_unlock(&em_tree->lock);
1632 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1636 struct btrfs_key key;
1637 struct btrfs_key found_key;
1638 struct btrfs_path *path;
1640 path = btrfs_alloc_path();
1644 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1645 key.type = BTRFS_DEV_ITEM_KEY;
1646 key.offset = (u64)-1;
1648 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1652 BUG_ON(ret == 0); /* Corruption */
1654 ret = btrfs_previous_item(fs_info->chunk_root, path,
1655 BTRFS_DEV_ITEMS_OBJECTID,
1656 BTRFS_DEV_ITEM_KEY);
1660 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1662 *devid_ret = found_key.offset + 1;
1666 btrfs_free_path(path);
1671 * the device information is stored in the chunk root
1672 * the btrfs_device struct should be fully filled in
1674 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1675 struct btrfs_fs_info *fs_info,
1676 struct btrfs_device *device)
1678 struct btrfs_root *root = fs_info->chunk_root;
1680 struct btrfs_path *path;
1681 struct btrfs_dev_item *dev_item;
1682 struct extent_buffer *leaf;
1683 struct btrfs_key key;
1686 path = btrfs_alloc_path();
1690 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1691 key.type = BTRFS_DEV_ITEM_KEY;
1692 key.offset = device->devid;
1694 ret = btrfs_insert_empty_item(trans, root, path, &key,
1699 leaf = path->nodes[0];
1700 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1702 btrfs_set_device_id(leaf, dev_item, device->devid);
1703 btrfs_set_device_generation(leaf, dev_item, 0);
1704 btrfs_set_device_type(leaf, dev_item, device->type);
1705 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1706 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1707 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1708 btrfs_set_device_total_bytes(leaf, dev_item,
1709 btrfs_device_get_disk_total_bytes(device));
1710 btrfs_set_device_bytes_used(leaf, dev_item,
1711 btrfs_device_get_bytes_used(device));
1712 btrfs_set_device_group(leaf, dev_item, 0);
1713 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1714 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1715 btrfs_set_device_start_offset(leaf, dev_item, 0);
1717 ptr = btrfs_device_uuid(dev_item);
1718 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1719 ptr = btrfs_device_fsid(dev_item);
1720 write_extent_buffer(leaf, fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1721 btrfs_mark_buffer_dirty(leaf);
1725 btrfs_free_path(path);
1730 * Function to update ctime/mtime for a given device path.
1731 * Mainly used for ctime/mtime based probe like libblkid.
1733 static void update_dev_time(const char *path_name)
1737 filp = filp_open(path_name, O_RDWR, 0);
1740 file_update_time(filp);
1741 filp_close(filp, NULL);
1744 static int btrfs_rm_dev_item(struct btrfs_fs_info *fs_info,
1745 struct btrfs_device *device)
1747 struct btrfs_root *root = fs_info->chunk_root;
1749 struct btrfs_path *path;
1750 struct btrfs_key key;
1751 struct btrfs_trans_handle *trans;
1753 path = btrfs_alloc_path();
1757 trans = btrfs_start_transaction(root, 0);
1758 if (IS_ERR(trans)) {
1759 btrfs_free_path(path);
1760 return PTR_ERR(trans);
1762 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1763 key.type = BTRFS_DEV_ITEM_KEY;
1764 key.offset = device->devid;
1766 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1775 ret = btrfs_del_item(trans, root, path);
1779 btrfs_free_path(path);
1780 btrfs_commit_transaction(trans);
1785 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1786 * filesystem. It's up to the caller to adjust that number regarding eg. device
1789 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1797 seq = read_seqbegin(&fs_info->profiles_lock);
1799 all_avail = fs_info->avail_data_alloc_bits |
1800 fs_info->avail_system_alloc_bits |
1801 fs_info->avail_metadata_alloc_bits;
1802 } while (read_seqretry(&fs_info->profiles_lock, seq));
1804 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1805 if (!(all_avail & btrfs_raid_group[i]))
1808 if (num_devices < btrfs_raid_array[i].devs_min) {
1809 int ret = btrfs_raid_mindev_error[i];
1819 struct btrfs_device *btrfs_find_next_active_device(struct btrfs_fs_devices *fs_devs,
1820 struct btrfs_device *device)
1822 struct btrfs_device *next_device;
1824 list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1825 if (next_device != device &&
1826 !next_device->missing && next_device->bdev)
1834 * Helper function to check if the given device is part of s_bdev / latest_bdev
1835 * and replace it with the provided or the next active device, in the context
1836 * where this function called, there should be always be another device (or
1837 * this_dev) which is active.
1839 void btrfs_assign_next_active_device(struct btrfs_fs_info *fs_info,
1840 struct btrfs_device *device, struct btrfs_device *this_dev)
1842 struct btrfs_device *next_device;
1845 next_device = this_dev;
1847 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
1849 ASSERT(next_device);
1851 if (fs_info->sb->s_bdev &&
1852 (fs_info->sb->s_bdev == device->bdev))
1853 fs_info->sb->s_bdev = next_device->bdev;
1855 if (fs_info->fs_devices->latest_bdev == device->bdev)
1856 fs_info->fs_devices->latest_bdev = next_device->bdev;
1859 int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path,
1862 struct btrfs_device *device;
1863 struct btrfs_fs_devices *cur_devices;
1866 bool clear_super = false;
1868 mutex_lock(&uuid_mutex);
1870 num_devices = fs_info->fs_devices->num_devices;
1871 btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
1872 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
1873 WARN_ON(num_devices < 1);
1876 btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
1878 ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
1882 ret = btrfs_find_device_by_devspec(fs_info, devid, device_path,
1887 if (device->is_tgtdev_for_dev_replace) {
1888 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1892 if (device->writeable && fs_info->fs_devices->rw_devices == 1) {
1893 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1897 if (device->writeable) {
1898 mutex_lock(&fs_info->chunk_mutex);
1899 list_del_init(&device->dev_alloc_list);
1900 device->fs_devices->rw_devices--;
1901 mutex_unlock(&fs_info->chunk_mutex);
1905 mutex_unlock(&uuid_mutex);
1906 ret = btrfs_shrink_device(device, 0);
1907 mutex_lock(&uuid_mutex);
1912 * TODO: the superblock still includes this device in its num_devices
1913 * counter although write_all_supers() is not locked out. This
1914 * could give a filesystem state which requires a degraded mount.
1916 ret = btrfs_rm_dev_item(fs_info, device);
1920 device->in_fs_metadata = 0;
1921 btrfs_scrub_cancel_dev(fs_info, device);
1924 * the device list mutex makes sure that we don't change
1925 * the device list while someone else is writing out all
1926 * the device supers. Whoever is writing all supers, should
1927 * lock the device list mutex before getting the number of
1928 * devices in the super block (super_copy). Conversely,
1929 * whoever updates the number of devices in the super block
1930 * (super_copy) should hold the device list mutex.
1933 cur_devices = device->fs_devices;
1934 mutex_lock(&fs_info->fs_devices->device_list_mutex);
1935 list_del_rcu(&device->dev_list);
1937 device->fs_devices->num_devices--;
1938 device->fs_devices->total_devices--;
1940 if (device->missing)
1941 device->fs_devices->missing_devices--;
1943 btrfs_assign_next_active_device(fs_info, device, NULL);
1946 device->fs_devices->open_devices--;
1947 /* remove sysfs entry */
1948 btrfs_sysfs_rm_device_link(fs_info->fs_devices, device);
1951 num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
1952 btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
1953 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1956 * at this point, the device is zero sized and detached from
1957 * the devices list. All that's left is to zero out the old
1958 * supers and free the device.
1960 if (device->writeable)
1961 btrfs_scratch_superblocks(device->bdev, device->name->str);
1963 btrfs_close_bdev(device);
1964 call_rcu(&device->rcu, free_device);
1966 if (cur_devices->open_devices == 0) {
1967 struct btrfs_fs_devices *fs_devices;
1968 fs_devices = fs_info->fs_devices;
1969 while (fs_devices) {
1970 if (fs_devices->seed == cur_devices) {
1971 fs_devices->seed = cur_devices->seed;
1974 fs_devices = fs_devices->seed;
1976 cur_devices->seed = NULL;
1977 __btrfs_close_devices(cur_devices);
1978 free_fs_devices(cur_devices);
1981 fs_info->num_tolerated_disk_barrier_failures =
1982 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
1985 mutex_unlock(&uuid_mutex);
1989 if (device->writeable) {
1990 mutex_lock(&fs_info->chunk_mutex);
1991 list_add(&device->dev_alloc_list,
1992 &fs_info->fs_devices->alloc_list);
1993 device->fs_devices->rw_devices++;
1994 mutex_unlock(&fs_info->chunk_mutex);
1999 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
2000 struct btrfs_device *srcdev)
2002 struct btrfs_fs_devices *fs_devices;
2004 WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
2007 * in case of fs with no seed, srcdev->fs_devices will point
2008 * to fs_devices of fs_info. However when the dev being replaced is
2009 * a seed dev it will point to the seed's local fs_devices. In short
2010 * srcdev will have its correct fs_devices in both the cases.
2012 fs_devices = srcdev->fs_devices;
2014 list_del_rcu(&srcdev->dev_list);
2015 list_del_rcu(&srcdev->dev_alloc_list);
2016 fs_devices->num_devices--;
2017 if (srcdev->missing)
2018 fs_devices->missing_devices--;
2020 if (srcdev->writeable)
2021 fs_devices->rw_devices--;
2024 fs_devices->open_devices--;
2027 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
2028 struct btrfs_device *srcdev)
2030 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2032 if (srcdev->writeable) {
2033 /* zero out the old super if it is writable */
2034 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
2037 btrfs_close_bdev(srcdev);
2039 call_rcu(&srcdev->rcu, free_device);
2042 * unless fs_devices is seed fs, num_devices shouldn't go
2045 BUG_ON(!fs_devices->num_devices && !fs_devices->seeding);
2047 /* if this is no devs we rather delete the fs_devices */
2048 if (!fs_devices->num_devices) {
2049 struct btrfs_fs_devices *tmp_fs_devices;
2051 tmp_fs_devices = fs_info->fs_devices;
2052 while (tmp_fs_devices) {
2053 if (tmp_fs_devices->seed == fs_devices) {
2054 tmp_fs_devices->seed = fs_devices->seed;
2057 tmp_fs_devices = tmp_fs_devices->seed;
2059 fs_devices->seed = NULL;
2060 __btrfs_close_devices(fs_devices);
2061 free_fs_devices(fs_devices);
2065 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2066 struct btrfs_device *tgtdev)
2068 mutex_lock(&uuid_mutex);
2070 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2072 btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev);
2075 fs_info->fs_devices->open_devices--;
2077 fs_info->fs_devices->num_devices--;
2079 btrfs_assign_next_active_device(fs_info, tgtdev, NULL);
2081 list_del_rcu(&tgtdev->dev_list);
2083 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2084 mutex_unlock(&uuid_mutex);
2087 * The update_dev_time() with in btrfs_scratch_superblocks()
2088 * may lead to a call to btrfs_show_devname() which will try
2089 * to hold device_list_mutex. And here this device
2090 * is already out of device list, so we don't have to hold
2091 * the device_list_mutex lock.
2093 btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
2095 btrfs_close_bdev(tgtdev);
2096 call_rcu(&tgtdev->rcu, free_device);
2099 static int btrfs_find_device_by_path(struct btrfs_fs_info *fs_info,
2100 const char *device_path,
2101 struct btrfs_device **device)
2104 struct btrfs_super_block *disk_super;
2107 struct block_device *bdev;
2108 struct buffer_head *bh;
2111 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2112 fs_info->bdev_holder, 0, &bdev, &bh);
2115 disk_super = (struct btrfs_super_block *)bh->b_data;
2116 devid = btrfs_stack_device_id(&disk_super->dev_item);
2117 dev_uuid = disk_super->dev_item.uuid;
2118 *device = btrfs_find_device(fs_info, devid, dev_uuid, disk_super->fsid);
2122 blkdev_put(bdev, FMODE_READ);
2126 int btrfs_find_device_missing_or_by_path(struct btrfs_fs_info *fs_info,
2127 const char *device_path,
2128 struct btrfs_device **device)
2131 if (strcmp(device_path, "missing") == 0) {
2132 struct list_head *devices;
2133 struct btrfs_device *tmp;
2135 devices = &fs_info->fs_devices->devices;
2137 * It is safe to read the devices since the volume_mutex
2138 * is held by the caller.
2140 list_for_each_entry(tmp, devices, dev_list) {
2141 if (tmp->in_fs_metadata && !tmp->bdev) {
2148 return BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2152 return btrfs_find_device_by_path(fs_info, device_path, device);
2157 * Lookup a device given by device id, or the path if the id is 0.
2159 int btrfs_find_device_by_devspec(struct btrfs_fs_info *fs_info, u64 devid,
2160 const char *devpath,
2161 struct btrfs_device **device)
2167 *device = btrfs_find_device(fs_info, devid, NULL, NULL);
2171 if (!devpath || !devpath[0])
2174 ret = btrfs_find_device_missing_or_by_path(fs_info, devpath,
2181 * does all the dirty work required for changing file system's UUID.
2183 static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info)
2185 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2186 struct btrfs_fs_devices *old_devices;
2187 struct btrfs_fs_devices *seed_devices;
2188 struct btrfs_super_block *disk_super = fs_info->super_copy;
2189 struct btrfs_device *device;
2192 BUG_ON(!mutex_is_locked(&uuid_mutex));
2193 if (!fs_devices->seeding)
2196 seed_devices = __alloc_fs_devices();
2197 if (IS_ERR(seed_devices))
2198 return PTR_ERR(seed_devices);
2200 old_devices = clone_fs_devices(fs_devices);
2201 if (IS_ERR(old_devices)) {
2202 kfree(seed_devices);
2203 return PTR_ERR(old_devices);
2206 list_add(&old_devices->list, &fs_uuids);
2208 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2209 seed_devices->opened = 1;
2210 INIT_LIST_HEAD(&seed_devices->devices);
2211 INIT_LIST_HEAD(&seed_devices->alloc_list);
2212 mutex_init(&seed_devices->device_list_mutex);
2214 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2215 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2217 list_for_each_entry(device, &seed_devices->devices, dev_list)
2218 device->fs_devices = seed_devices;
2220 mutex_lock(&fs_info->chunk_mutex);
2221 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2222 mutex_unlock(&fs_info->chunk_mutex);
2224 fs_devices->seeding = 0;
2225 fs_devices->num_devices = 0;
2226 fs_devices->open_devices = 0;
2227 fs_devices->missing_devices = 0;
2228 fs_devices->rotating = 0;
2229 fs_devices->seed = seed_devices;
2231 generate_random_uuid(fs_devices->fsid);
2232 memcpy(fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2233 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2234 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2236 super_flags = btrfs_super_flags(disk_super) &
2237 ~BTRFS_SUPER_FLAG_SEEDING;
2238 btrfs_set_super_flags(disk_super, super_flags);
2244 * Store the expected generation for seed devices in device items.
2246 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2247 struct btrfs_fs_info *fs_info)
2249 struct btrfs_root *root = fs_info->chunk_root;
2250 struct btrfs_path *path;
2251 struct extent_buffer *leaf;
2252 struct btrfs_dev_item *dev_item;
2253 struct btrfs_device *device;
2254 struct btrfs_key key;
2255 u8 fs_uuid[BTRFS_UUID_SIZE];
2256 u8 dev_uuid[BTRFS_UUID_SIZE];
2260 path = btrfs_alloc_path();
2264 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2266 key.type = BTRFS_DEV_ITEM_KEY;
2269 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2273 leaf = path->nodes[0];
2275 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2276 ret = btrfs_next_leaf(root, path);
2281 leaf = path->nodes[0];
2282 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2283 btrfs_release_path(path);
2287 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2288 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2289 key.type != BTRFS_DEV_ITEM_KEY)
2292 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2293 struct btrfs_dev_item);
2294 devid = btrfs_device_id(leaf, dev_item);
2295 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2297 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2299 device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
2300 BUG_ON(!device); /* Logic error */
2302 if (device->fs_devices->seeding) {
2303 btrfs_set_device_generation(leaf, dev_item,
2304 device->generation);
2305 btrfs_mark_buffer_dirty(leaf);
2313 btrfs_free_path(path);
2317 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2319 struct btrfs_root *root = fs_info->dev_root;
2320 struct request_queue *q;
2321 struct btrfs_trans_handle *trans;
2322 struct btrfs_device *device;
2323 struct block_device *bdev;
2324 struct list_head *devices;
2325 struct super_block *sb = fs_info->sb;
2326 struct rcu_string *name;
2328 int seeding_dev = 0;
2331 if ((sb->s_flags & MS_RDONLY) && !fs_info->fs_devices->seeding)
2334 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2335 fs_info->bdev_holder);
2337 return PTR_ERR(bdev);
2339 if (fs_info->fs_devices->seeding) {
2341 down_write(&sb->s_umount);
2342 mutex_lock(&uuid_mutex);
2345 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2347 devices = &fs_info->fs_devices->devices;
2349 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2350 list_for_each_entry(device, devices, dev_list) {
2351 if (device->bdev == bdev) {
2354 &fs_info->fs_devices->device_list_mutex);
2358 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2360 device = btrfs_alloc_device(fs_info, NULL, NULL);
2361 if (IS_ERR(device)) {
2362 /* we can safely leave the fs_devices entry around */
2363 ret = PTR_ERR(device);
2367 name = rcu_string_strdup(device_path, GFP_KERNEL);
2373 rcu_assign_pointer(device->name, name);
2375 trans = btrfs_start_transaction(root, 0);
2376 if (IS_ERR(trans)) {
2377 rcu_string_free(device->name);
2379 ret = PTR_ERR(trans);
2383 q = bdev_get_queue(bdev);
2384 if (blk_queue_discard(q))
2385 device->can_discard = 1;
2386 device->writeable = 1;
2387 device->generation = trans->transid;
2388 device->io_width = fs_info->sectorsize;
2389 device->io_align = fs_info->sectorsize;
2390 device->sector_size = fs_info->sectorsize;
2391 device->total_bytes = i_size_read(bdev->bd_inode);
2392 device->disk_total_bytes = device->total_bytes;
2393 device->commit_total_bytes = device->total_bytes;
2394 device->fs_info = fs_info;
2395 device->bdev = bdev;
2396 device->in_fs_metadata = 1;
2397 device->is_tgtdev_for_dev_replace = 0;
2398 device->mode = FMODE_EXCL;
2399 device->dev_stats_valid = 1;
2400 set_blocksize(device->bdev, 4096);
2403 sb->s_flags &= ~MS_RDONLY;
2404 ret = btrfs_prepare_sprout(fs_info);
2405 BUG_ON(ret); /* -ENOMEM */
2408 device->fs_devices = fs_info->fs_devices;
2410 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2411 mutex_lock(&fs_info->chunk_mutex);
2412 list_add_rcu(&device->dev_list, &fs_info->fs_devices->devices);
2413 list_add(&device->dev_alloc_list,
2414 &fs_info->fs_devices->alloc_list);
2415 fs_info->fs_devices->num_devices++;
2416 fs_info->fs_devices->open_devices++;
2417 fs_info->fs_devices->rw_devices++;
2418 fs_info->fs_devices->total_devices++;
2419 fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2421 spin_lock(&fs_info->free_chunk_lock);
2422 fs_info->free_chunk_space += device->total_bytes;
2423 spin_unlock(&fs_info->free_chunk_lock);
2425 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2426 fs_info->fs_devices->rotating = 1;
2428 tmp = btrfs_super_total_bytes(fs_info->super_copy);
2429 btrfs_set_super_total_bytes(fs_info->super_copy,
2430 tmp + device->total_bytes);
2432 tmp = btrfs_super_num_devices(fs_info->super_copy);
2433 btrfs_set_super_num_devices(fs_info->super_copy, tmp + 1);
2435 /* add sysfs device entry */
2436 btrfs_sysfs_add_device_link(fs_info->fs_devices, device);
2439 * we've got more storage, clear any full flags on the space
2442 btrfs_clear_space_info_full(fs_info);
2444 mutex_unlock(&fs_info->chunk_mutex);
2445 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2448 mutex_lock(&fs_info->chunk_mutex);
2449 ret = init_first_rw_device(trans, fs_info);
2450 mutex_unlock(&fs_info->chunk_mutex);
2452 btrfs_abort_transaction(trans, ret);
2457 ret = btrfs_add_device(trans, fs_info, device);
2459 btrfs_abort_transaction(trans, ret);
2464 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2466 ret = btrfs_finish_sprout(trans, fs_info);
2468 btrfs_abort_transaction(trans, ret);
2472 /* Sprouting would change fsid of the mounted root,
2473 * so rename the fsid on the sysfs
2475 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2477 if (kobject_rename(&fs_info->fs_devices->fsid_kobj, fsid_buf))
2479 "sysfs: failed to create fsid for sprout");
2482 fs_info->num_tolerated_disk_barrier_failures =
2483 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2484 ret = btrfs_commit_transaction(trans);
2487 mutex_unlock(&uuid_mutex);
2488 up_write(&sb->s_umount);
2490 if (ret) /* transaction commit */
2493 ret = btrfs_relocate_sys_chunks(fs_info);
2495 btrfs_handle_fs_error(fs_info, ret,
2496 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2497 trans = btrfs_attach_transaction(root);
2498 if (IS_ERR(trans)) {
2499 if (PTR_ERR(trans) == -ENOENT)
2501 return PTR_ERR(trans);
2503 ret = btrfs_commit_transaction(trans);
2506 /* Update ctime/mtime for libblkid */
2507 update_dev_time(device_path);
2511 btrfs_end_transaction(trans);
2512 rcu_string_free(device->name);
2513 btrfs_sysfs_rm_device_link(fs_info->fs_devices, device);
2516 blkdev_put(bdev, FMODE_EXCL);
2518 mutex_unlock(&uuid_mutex);
2519 up_write(&sb->s_umount);
2524 int btrfs_init_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2525 const char *device_path,
2526 struct btrfs_device *srcdev,
2527 struct btrfs_device **device_out)
2529 struct request_queue *q;
2530 struct btrfs_device *device;
2531 struct block_device *bdev;
2532 struct list_head *devices;
2533 struct rcu_string *name;
2534 u64 devid = BTRFS_DEV_REPLACE_DEVID;
2538 if (fs_info->fs_devices->seeding) {
2539 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
2543 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2544 fs_info->bdev_holder);
2546 btrfs_err(fs_info, "target device %s is invalid!", device_path);
2547 return PTR_ERR(bdev);
2550 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2552 devices = &fs_info->fs_devices->devices;
2553 list_for_each_entry(device, devices, dev_list) {
2554 if (device->bdev == bdev) {
2556 "target device is in the filesystem!");
2563 if (i_size_read(bdev->bd_inode) <
2564 btrfs_device_get_total_bytes(srcdev)) {
2566 "target device is smaller than source device!");
2572 device = btrfs_alloc_device(NULL, &devid, NULL);
2573 if (IS_ERR(device)) {
2574 ret = PTR_ERR(device);
2578 name = rcu_string_strdup(device_path, GFP_NOFS);
2584 rcu_assign_pointer(device->name, name);
2586 q = bdev_get_queue(bdev);
2587 if (blk_queue_discard(q))
2588 device->can_discard = 1;
2589 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2590 device->writeable = 1;
2591 device->generation = 0;
2592 device->io_width = fs_info->sectorsize;
2593 device->io_align = fs_info->sectorsize;
2594 device->sector_size = fs_info->sectorsize;
2595 device->total_bytes = btrfs_device_get_total_bytes(srcdev);
2596 device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
2597 device->bytes_used = btrfs_device_get_bytes_used(srcdev);
2598 ASSERT(list_empty(&srcdev->resized_list));
2599 device->commit_total_bytes = srcdev->commit_total_bytes;
2600 device->commit_bytes_used = device->bytes_used;
2601 device->fs_info = fs_info;
2602 device->bdev = bdev;
2603 device->in_fs_metadata = 1;
2604 device->is_tgtdev_for_dev_replace = 1;
2605 device->mode = FMODE_EXCL;
2606 device->dev_stats_valid = 1;
2607 set_blocksize(device->bdev, 4096);
2608 device->fs_devices = fs_info->fs_devices;
2609 list_add(&device->dev_list, &fs_info->fs_devices->devices);
2610 fs_info->fs_devices->num_devices++;
2611 fs_info->fs_devices->open_devices++;
2612 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2614 *device_out = device;
2618 blkdev_put(bdev, FMODE_EXCL);
2622 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2623 struct btrfs_device *tgtdev)
2625 u32 sectorsize = fs_info->sectorsize;
2627 WARN_ON(fs_info->fs_devices->rw_devices == 0);
2628 tgtdev->io_width = sectorsize;
2629 tgtdev->io_align = sectorsize;
2630 tgtdev->sector_size = sectorsize;
2631 tgtdev->fs_info = fs_info;
2632 tgtdev->in_fs_metadata = 1;
2635 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2636 struct btrfs_device *device)
2639 struct btrfs_path *path;
2640 struct btrfs_root *root = device->fs_info->chunk_root;
2641 struct btrfs_dev_item *dev_item;
2642 struct extent_buffer *leaf;
2643 struct btrfs_key key;
2645 path = btrfs_alloc_path();
2649 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2650 key.type = BTRFS_DEV_ITEM_KEY;
2651 key.offset = device->devid;
2653 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2662 leaf = path->nodes[0];
2663 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2665 btrfs_set_device_id(leaf, dev_item, device->devid);
2666 btrfs_set_device_type(leaf, dev_item, device->type);
2667 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2668 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2669 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2670 btrfs_set_device_total_bytes(leaf, dev_item,
2671 btrfs_device_get_disk_total_bytes(device));
2672 btrfs_set_device_bytes_used(leaf, dev_item,
2673 btrfs_device_get_bytes_used(device));
2674 btrfs_mark_buffer_dirty(leaf);
2677 btrfs_free_path(path);
2681 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2682 struct btrfs_device *device, u64 new_size)
2684 struct btrfs_fs_info *fs_info = device->fs_info;
2685 struct btrfs_super_block *super_copy = fs_info->super_copy;
2686 struct btrfs_fs_devices *fs_devices;
2690 if (!device->writeable)
2693 mutex_lock(&fs_info->chunk_mutex);
2694 old_total = btrfs_super_total_bytes(super_copy);
2695 diff = new_size - device->total_bytes;
2697 if (new_size <= device->total_bytes ||
2698 device->is_tgtdev_for_dev_replace) {
2699 mutex_unlock(&fs_info->chunk_mutex);
2703 fs_devices = fs_info->fs_devices;
2705 btrfs_set_super_total_bytes(super_copy, old_total + diff);
2706 device->fs_devices->total_rw_bytes += diff;
2708 btrfs_device_set_total_bytes(device, new_size);
2709 btrfs_device_set_disk_total_bytes(device, new_size);
2710 btrfs_clear_space_info_full(device->fs_info);
2711 if (list_empty(&device->resized_list))
2712 list_add_tail(&device->resized_list,
2713 &fs_devices->resized_devices);
2714 mutex_unlock(&fs_info->chunk_mutex);
2716 return btrfs_update_device(trans, device);
2719 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2720 struct btrfs_fs_info *fs_info, u64 chunk_objectid,
2723 struct btrfs_root *root = fs_info->chunk_root;
2725 struct btrfs_path *path;
2726 struct btrfs_key key;
2728 path = btrfs_alloc_path();
2732 key.objectid = chunk_objectid;
2733 key.offset = chunk_offset;
2734 key.type = BTRFS_CHUNK_ITEM_KEY;
2736 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2739 else if (ret > 0) { /* Logic error or corruption */
2740 btrfs_handle_fs_error(fs_info, -ENOENT,
2741 "Failed lookup while freeing chunk.");
2746 ret = btrfs_del_item(trans, root, path);
2748 btrfs_handle_fs_error(fs_info, ret,
2749 "Failed to delete chunk item.");
2751 btrfs_free_path(path);
2755 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info,
2756 u64 chunk_objectid, u64 chunk_offset)
2758 struct btrfs_super_block *super_copy = fs_info->super_copy;
2759 struct btrfs_disk_key *disk_key;
2760 struct btrfs_chunk *chunk;
2767 struct btrfs_key key;
2769 mutex_lock(&fs_info->chunk_mutex);
2770 array_size = btrfs_super_sys_array_size(super_copy);
2772 ptr = super_copy->sys_chunk_array;
2775 while (cur < array_size) {
2776 disk_key = (struct btrfs_disk_key *)ptr;
2777 btrfs_disk_key_to_cpu(&key, disk_key);
2779 len = sizeof(*disk_key);
2781 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2782 chunk = (struct btrfs_chunk *)(ptr + len);
2783 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2784 len += btrfs_chunk_item_size(num_stripes);
2789 if (key.objectid == chunk_objectid &&
2790 key.offset == chunk_offset) {
2791 memmove(ptr, ptr + len, array_size - (cur + len));
2793 btrfs_set_super_sys_array_size(super_copy, array_size);
2799 mutex_unlock(&fs_info->chunk_mutex);
2803 static struct extent_map *get_chunk_map(struct btrfs_fs_info *fs_info,
2804 u64 logical, u64 length)
2806 struct extent_map_tree *em_tree;
2807 struct extent_map *em;
2809 em_tree = &fs_info->mapping_tree.map_tree;
2810 read_lock(&em_tree->lock);
2811 em = lookup_extent_mapping(em_tree, logical, length);
2812 read_unlock(&em_tree->lock);
2815 btrfs_crit(fs_info, "unable to find logical %llu length %llu",
2817 return ERR_PTR(-EINVAL);
2820 if (em->start > logical || em->start + em->len < logical) {
2822 "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
2823 logical, length, em->start, em->start + em->len);
2824 free_extent_map(em);
2825 return ERR_PTR(-EINVAL);
2828 /* callers are responsible for dropping em's ref. */
2832 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2833 struct btrfs_fs_info *fs_info, u64 chunk_offset)
2835 struct extent_map *em;
2836 struct map_lookup *map;
2837 u64 dev_extent_len = 0;
2838 u64 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2840 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2842 em = get_chunk_map(fs_info, chunk_offset, 1);
2845 * This is a logic error, but we don't want to just rely on the
2846 * user having built with ASSERT enabled, so if ASSERT doesn't
2847 * do anything we still error out.
2852 map = em->map_lookup;
2853 mutex_lock(&fs_info->chunk_mutex);
2854 check_system_chunk(trans, fs_info, map->type);
2855 mutex_unlock(&fs_info->chunk_mutex);
2858 * Take the device list mutex to prevent races with the final phase of
2859 * a device replace operation that replaces the device object associated
2860 * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2862 mutex_lock(&fs_devices->device_list_mutex);
2863 for (i = 0; i < map->num_stripes; i++) {
2864 struct btrfs_device *device = map->stripes[i].dev;
2865 ret = btrfs_free_dev_extent(trans, device,
2866 map->stripes[i].physical,
2869 mutex_unlock(&fs_devices->device_list_mutex);
2870 btrfs_abort_transaction(trans, ret);
2874 if (device->bytes_used > 0) {
2875 mutex_lock(&fs_info->chunk_mutex);
2876 btrfs_device_set_bytes_used(device,
2877 device->bytes_used - dev_extent_len);
2878 spin_lock(&fs_info->free_chunk_lock);
2879 fs_info->free_chunk_space += dev_extent_len;
2880 spin_unlock(&fs_info->free_chunk_lock);
2881 btrfs_clear_space_info_full(fs_info);
2882 mutex_unlock(&fs_info->chunk_mutex);
2885 if (map->stripes[i].dev) {
2886 ret = btrfs_update_device(trans, map->stripes[i].dev);
2888 mutex_unlock(&fs_devices->device_list_mutex);
2889 btrfs_abort_transaction(trans, ret);
2894 mutex_unlock(&fs_devices->device_list_mutex);
2896 ret = btrfs_free_chunk(trans, fs_info, chunk_objectid, chunk_offset);
2898 btrfs_abort_transaction(trans, ret);
2902 trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
2904 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2905 ret = btrfs_del_sys_chunk(fs_info, chunk_objectid,
2908 btrfs_abort_transaction(trans, ret);
2913 ret = btrfs_remove_block_group(trans, fs_info, chunk_offset, em);
2915 btrfs_abort_transaction(trans, ret);
2921 free_extent_map(em);
2925 static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2927 struct btrfs_root *root = fs_info->chunk_root;
2928 struct btrfs_trans_handle *trans;
2932 * Prevent races with automatic removal of unused block groups.
2933 * After we relocate and before we remove the chunk with offset
2934 * chunk_offset, automatic removal of the block group can kick in,
2935 * resulting in a failure when calling btrfs_remove_chunk() below.
2937 * Make sure to acquire this mutex before doing a tree search (dev
2938 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2939 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2940 * we release the path used to search the chunk/dev tree and before
2941 * the current task acquires this mutex and calls us.
2943 ASSERT(mutex_is_locked(&fs_info->delete_unused_bgs_mutex));
2945 ret = btrfs_can_relocate(fs_info, chunk_offset);
2949 /* step one, relocate all the extents inside this chunk */
2950 btrfs_scrub_pause(fs_info);
2951 ret = btrfs_relocate_block_group(fs_info, chunk_offset);
2952 btrfs_scrub_continue(fs_info);
2956 trans = btrfs_start_trans_remove_block_group(root->fs_info,
2958 if (IS_ERR(trans)) {
2959 ret = PTR_ERR(trans);
2960 btrfs_handle_fs_error(root->fs_info, ret, NULL);
2965 * step two, delete the device extents and the
2966 * chunk tree entries
2968 ret = btrfs_remove_chunk(trans, fs_info, chunk_offset);
2969 btrfs_end_transaction(trans);
2973 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
2975 struct btrfs_root *chunk_root = fs_info->chunk_root;
2976 struct btrfs_path *path;
2977 struct extent_buffer *leaf;
2978 struct btrfs_chunk *chunk;
2979 struct btrfs_key key;
2980 struct btrfs_key found_key;
2982 bool retried = false;
2986 path = btrfs_alloc_path();
2991 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2992 key.offset = (u64)-1;
2993 key.type = BTRFS_CHUNK_ITEM_KEY;
2996 mutex_lock(&fs_info->delete_unused_bgs_mutex);
2997 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2999 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3002 BUG_ON(ret == 0); /* Corruption */
3004 ret = btrfs_previous_item(chunk_root, path, key.objectid,
3007 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3013 leaf = path->nodes[0];
3014 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3016 chunk = btrfs_item_ptr(leaf, path->slots[0],
3017 struct btrfs_chunk);
3018 chunk_type = btrfs_chunk_type(leaf, chunk);
3019 btrfs_release_path(path);
3021 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3022 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3028 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3030 if (found_key.offset == 0)
3032 key.offset = found_key.offset - 1;
3035 if (failed && !retried) {
3039 } else if (WARN_ON(failed && retried)) {
3043 btrfs_free_path(path);
3047 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3048 struct btrfs_balance_control *bctl)
3050 struct btrfs_root *root = fs_info->tree_root;
3051 struct btrfs_trans_handle *trans;
3052 struct btrfs_balance_item *item;
3053 struct btrfs_disk_balance_args disk_bargs;
3054 struct btrfs_path *path;
3055 struct extent_buffer *leaf;
3056 struct btrfs_key key;
3059 path = btrfs_alloc_path();
3063 trans = btrfs_start_transaction(root, 0);
3064 if (IS_ERR(trans)) {
3065 btrfs_free_path(path);
3066 return PTR_ERR(trans);
3069 key.objectid = BTRFS_BALANCE_OBJECTID;
3070 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3073 ret = btrfs_insert_empty_item(trans, root, path, &key,
3078 leaf = path->nodes[0];
3079 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3081 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3083 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3084 btrfs_set_balance_data(leaf, item, &disk_bargs);
3085 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3086 btrfs_set_balance_meta(leaf, item, &disk_bargs);
3087 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3088 btrfs_set_balance_sys(leaf, item, &disk_bargs);
3090 btrfs_set_balance_flags(leaf, item, bctl->flags);
3092 btrfs_mark_buffer_dirty(leaf);
3094 btrfs_free_path(path);
3095 err = btrfs_commit_transaction(trans);
3101 static int del_balance_item(struct btrfs_fs_info *fs_info)
3103 struct btrfs_root *root = fs_info->tree_root;
3104 struct btrfs_trans_handle *trans;
3105 struct btrfs_path *path;
3106 struct btrfs_key key;
3109 path = btrfs_alloc_path();
3113 trans = btrfs_start_transaction(root, 0);
3114 if (IS_ERR(trans)) {
3115 btrfs_free_path(path);
3116 return PTR_ERR(trans);
3119 key.objectid = BTRFS_BALANCE_OBJECTID;
3120 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3123 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3131 ret = btrfs_del_item(trans, root, path);
3133 btrfs_free_path(path);
3134 err = btrfs_commit_transaction(trans);
3141 * This is a heuristic used to reduce the number of chunks balanced on
3142 * resume after balance was interrupted.
3144 static void update_balance_args(struct btrfs_balance_control *bctl)
3147 * Turn on soft mode for chunk types that were being converted.
3149 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3150 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3151 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3152 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3153 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3154 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3157 * Turn on usage filter if is not already used. The idea is
3158 * that chunks that we have already balanced should be
3159 * reasonably full. Don't do it for chunks that are being
3160 * converted - that will keep us from relocating unconverted
3161 * (albeit full) chunks.
3163 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3164 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3165 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3166 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3167 bctl->data.usage = 90;
3169 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3170 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3171 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3172 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3173 bctl->sys.usage = 90;
3175 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3176 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3177 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3178 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3179 bctl->meta.usage = 90;
3184 * Should be called with both balance and volume mutexes held to
3185 * serialize other volume operations (add_dev/rm_dev/resize) with
3186 * restriper. Same goes for unset_balance_control.
3188 static void set_balance_control(struct btrfs_balance_control *bctl)
3190 struct btrfs_fs_info *fs_info = bctl->fs_info;
3192 BUG_ON(fs_info->balance_ctl);
3194 spin_lock(&fs_info->balance_lock);
3195 fs_info->balance_ctl = bctl;
3196 spin_unlock(&fs_info->balance_lock);
3199 static void unset_balance_control(struct btrfs_fs_info *fs_info)
3201 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3203 BUG_ON(!fs_info->balance_ctl);
3205 spin_lock(&fs_info->balance_lock);
3206 fs_info->balance_ctl = NULL;
3207 spin_unlock(&fs_info->balance_lock);
3213 * Balance filters. Return 1 if chunk should be filtered out
3214 * (should not be balanced).
3216 static int chunk_profiles_filter(u64 chunk_type,
3217 struct btrfs_balance_args *bargs)
3219 chunk_type = chunk_to_extended(chunk_type) &
3220 BTRFS_EXTENDED_PROFILE_MASK;
3222 if (bargs->profiles & chunk_type)
3228 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3229 struct btrfs_balance_args *bargs)
3231 struct btrfs_block_group_cache *cache;
3233 u64 user_thresh_min;
3234 u64 user_thresh_max;
3237 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3238 chunk_used = btrfs_block_group_used(&cache->item);
3240 if (bargs->usage_min == 0)
3241 user_thresh_min = 0;
3243 user_thresh_min = div_factor_fine(cache->key.offset,
3246 if (bargs->usage_max == 0)
3247 user_thresh_max = 1;
3248 else if (bargs->usage_max > 100)
3249 user_thresh_max = cache->key.offset;
3251 user_thresh_max = div_factor_fine(cache->key.offset,
3254 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3257 btrfs_put_block_group(cache);
3261 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3262 u64 chunk_offset, struct btrfs_balance_args *bargs)
3264 struct btrfs_block_group_cache *cache;
3265 u64 chunk_used, user_thresh;
3268 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3269 chunk_used = btrfs_block_group_used(&cache->item);
3271 if (bargs->usage_min == 0)
3273 else if (bargs->usage > 100)
3274 user_thresh = cache->key.offset;
3276 user_thresh = div_factor_fine(cache->key.offset,
3279 if (chunk_used < user_thresh)
3282 btrfs_put_block_group(cache);
3286 static int chunk_devid_filter(struct extent_buffer *leaf,
3287 struct btrfs_chunk *chunk,
3288 struct btrfs_balance_args *bargs)
3290 struct btrfs_stripe *stripe;
3291 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3294 for (i = 0; i < num_stripes; i++) {
3295 stripe = btrfs_stripe_nr(chunk, i);
3296 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3303 /* [pstart, pend) */
3304 static int chunk_drange_filter(struct extent_buffer *leaf,
3305 struct btrfs_chunk *chunk,
3307 struct btrfs_balance_args *bargs)
3309 struct btrfs_stripe *stripe;
3310 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3316 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3319 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3320 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3321 factor = num_stripes / 2;
3322 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3323 factor = num_stripes - 1;
3324 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3325 factor = num_stripes - 2;
3327 factor = num_stripes;
3330 for (i = 0; i < num_stripes; i++) {
3331 stripe = btrfs_stripe_nr(chunk, i);
3332 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3335 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3336 stripe_length = btrfs_chunk_length(leaf, chunk);
3337 stripe_length = div_u64(stripe_length, factor);
3339 if (stripe_offset < bargs->pend &&
3340 stripe_offset + stripe_length > bargs->pstart)
3347 /* [vstart, vend) */
3348 static int chunk_vrange_filter(struct extent_buffer *leaf,
3349 struct btrfs_chunk *chunk,
3351 struct btrfs_balance_args *bargs)
3353 if (chunk_offset < bargs->vend &&
3354 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3355 /* at least part of the chunk is inside this vrange */
3361 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3362 struct btrfs_chunk *chunk,
3363 struct btrfs_balance_args *bargs)
3365 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3367 if (bargs->stripes_min <= num_stripes
3368 && num_stripes <= bargs->stripes_max)
3374 static int chunk_soft_convert_filter(u64 chunk_type,
3375 struct btrfs_balance_args *bargs)
3377 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3380 chunk_type = chunk_to_extended(chunk_type) &
3381 BTRFS_EXTENDED_PROFILE_MASK;
3383 if (bargs->target == chunk_type)
3389 static int should_balance_chunk(struct btrfs_fs_info *fs_info,
3390 struct extent_buffer *leaf,
3391 struct btrfs_chunk *chunk, u64 chunk_offset)
3393 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3394 struct btrfs_balance_args *bargs = NULL;
3395 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3398 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3399 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3403 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3404 bargs = &bctl->data;
3405 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3407 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3408 bargs = &bctl->meta;
3410 /* profiles filter */
3411 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3412 chunk_profiles_filter(chunk_type, bargs)) {
3417 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3418 chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3420 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3421 chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3426 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3427 chunk_devid_filter(leaf, chunk, bargs)) {
3431 /* drange filter, makes sense only with devid filter */
3432 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3433 chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
3438 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3439 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3443 /* stripes filter */
3444 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3445 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3449 /* soft profile changing mode */
3450 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3451 chunk_soft_convert_filter(chunk_type, bargs)) {
3456 * limited by count, must be the last filter
3458 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3459 if (bargs->limit == 0)
3463 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3465 * Same logic as the 'limit' filter; the minimum cannot be
3466 * determined here because we do not have the global information
3467 * about the count of all chunks that satisfy the filters.
3469 if (bargs->limit_max == 0)
3478 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3480 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3481 struct btrfs_root *chunk_root = fs_info->chunk_root;
3482 struct btrfs_root *dev_root = fs_info->dev_root;
3483 struct list_head *devices;
3484 struct btrfs_device *device;
3488 struct btrfs_chunk *chunk;
3489 struct btrfs_path *path = NULL;
3490 struct btrfs_key key;
3491 struct btrfs_key found_key;
3492 struct btrfs_trans_handle *trans;
3493 struct extent_buffer *leaf;
3496 int enospc_errors = 0;
3497 bool counting = true;
3498 /* The single value limit and min/max limits use the same bytes in the */
3499 u64 limit_data = bctl->data.limit;
3500 u64 limit_meta = bctl->meta.limit;
3501 u64 limit_sys = bctl->sys.limit;
3505 int chunk_reserved = 0;
3508 /* step one make some room on all the devices */
3509 devices = &fs_info->fs_devices->devices;
3510 list_for_each_entry(device, devices, dev_list) {
3511 old_size = btrfs_device_get_total_bytes(device);
3512 size_to_free = div_factor(old_size, 1);
3513 size_to_free = min_t(u64, size_to_free, SZ_1M);
3514 if (!device->writeable ||
3515 btrfs_device_get_total_bytes(device) -
3516 btrfs_device_get_bytes_used(device) > size_to_free ||
3517 device->is_tgtdev_for_dev_replace)
3520 ret = btrfs_shrink_device(device, old_size - size_to_free);
3524 /* btrfs_shrink_device never returns ret > 0 */
3529 trans = btrfs_start_transaction(dev_root, 0);
3530 if (IS_ERR(trans)) {
3531 ret = PTR_ERR(trans);
3532 btrfs_info_in_rcu(fs_info,
3533 "resize: unable to start transaction after shrinking device %s (error %d), old size %llu, new size %llu",
3534 rcu_str_deref(device->name), ret,
3535 old_size, old_size - size_to_free);
3539 ret = btrfs_grow_device(trans, device, old_size);
3541 btrfs_end_transaction(trans);
3542 /* btrfs_grow_device never returns ret > 0 */
3544 btrfs_info_in_rcu(fs_info,
3545 "resize: unable to grow device after shrinking device %s (error %d), old size %llu, new size %llu",
3546 rcu_str_deref(device->name), ret,
3547 old_size, old_size - size_to_free);
3551 btrfs_end_transaction(trans);
3554 /* step two, relocate all the chunks */
3555 path = btrfs_alloc_path();
3561 /* zero out stat counters */
3562 spin_lock(&fs_info->balance_lock);
3563 memset(&bctl->stat, 0, sizeof(bctl->stat));
3564 spin_unlock(&fs_info->balance_lock);
3568 * The single value limit and min/max limits use the same bytes
3571 bctl->data.limit = limit_data;
3572 bctl->meta.limit = limit_meta;
3573 bctl->sys.limit = limit_sys;
3575 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3576 key.offset = (u64)-1;
3577 key.type = BTRFS_CHUNK_ITEM_KEY;
3580 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3581 atomic_read(&fs_info->balance_cancel_req)) {
3586 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3587 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3589 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3594 * this shouldn't happen, it means the last relocate
3598 BUG(); /* FIXME break ? */
3600 ret = btrfs_previous_item(chunk_root, path, 0,
3601 BTRFS_CHUNK_ITEM_KEY);
3603 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3608 leaf = path->nodes[0];
3609 slot = path->slots[0];
3610 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3612 if (found_key.objectid != key.objectid) {
3613 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3617 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3618 chunk_type = btrfs_chunk_type(leaf, chunk);
3621 spin_lock(&fs_info->balance_lock);
3622 bctl->stat.considered++;
3623 spin_unlock(&fs_info->balance_lock);
3626 ret = should_balance_chunk(fs_info, leaf, chunk,
3629 btrfs_release_path(path);
3631 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3636 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3637 spin_lock(&fs_info->balance_lock);
3638 bctl->stat.expected++;
3639 spin_unlock(&fs_info->balance_lock);
3641 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3643 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3645 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3652 * Apply limit_min filter, no need to check if the LIMITS
3653 * filter is used, limit_min is 0 by default
3655 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3656 count_data < bctl->data.limit_min)
3657 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3658 count_meta < bctl->meta.limit_min)
3659 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3660 count_sys < bctl->sys.limit_min)) {
3661 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3665 ASSERT(fs_info->data_sinfo);
3666 spin_lock(&fs_info->data_sinfo->lock);
3667 bytes_used = fs_info->data_sinfo->bytes_used;
3668 spin_unlock(&fs_info->data_sinfo->lock);
3670 if ((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3671 !chunk_reserved && !bytes_used) {
3672 trans = btrfs_start_transaction(chunk_root, 0);
3673 if (IS_ERR(trans)) {
3674 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3675 ret = PTR_ERR(trans);
3679 ret = btrfs_force_chunk_alloc(trans, fs_info,
3680 BTRFS_BLOCK_GROUP_DATA);
3681 btrfs_end_transaction(trans);
3683 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3689 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3690 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3691 if (ret && ret != -ENOSPC)
3693 if (ret == -ENOSPC) {
3696 spin_lock(&fs_info->balance_lock);
3697 bctl->stat.completed++;
3698 spin_unlock(&fs_info->balance_lock);
3701 if (found_key.offset == 0)
3703 key.offset = found_key.offset - 1;
3707 btrfs_release_path(path);
3712 btrfs_free_path(path);
3713 if (enospc_errors) {
3714 btrfs_info(fs_info, "%d enospc errors during balance",
3724 * alloc_profile_is_valid - see if a given profile is valid and reduced
3725 * @flags: profile to validate
3726 * @extended: if true @flags is treated as an extended profile
3728 static int alloc_profile_is_valid(u64 flags, int extended)
3730 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3731 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3733 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3735 /* 1) check that all other bits are zeroed */
3739 /* 2) see if profile is reduced */
3741 return !extended; /* "0" is valid for usual profiles */
3743 /* true if exactly one bit set */
3744 return (flags & (flags - 1)) == 0;
3747 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3749 /* cancel requested || normal exit path */
3750 return atomic_read(&fs_info->balance_cancel_req) ||
3751 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3752 atomic_read(&fs_info->balance_cancel_req) == 0);
3755 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3759 unset_balance_control(fs_info);
3760 ret = del_balance_item(fs_info);
3762 btrfs_handle_fs_error(fs_info, ret, NULL);
3764 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3767 /* Non-zero return value signifies invalidity */
3768 static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg,
3771 return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3772 (!alloc_profile_is_valid(bctl_arg->target, 1) ||
3773 (bctl_arg->target & ~allowed)));
3777 * Should be called with both balance and volume mutexes held
3779 int btrfs_balance(struct btrfs_balance_control *bctl,
3780 struct btrfs_ioctl_balance_args *bargs)
3782 struct btrfs_fs_info *fs_info = bctl->fs_info;
3783 u64 meta_target, data_target;
3790 if (btrfs_fs_closing(fs_info) ||
3791 atomic_read(&fs_info->balance_pause_req) ||
3792 atomic_read(&fs_info->balance_cancel_req)) {
3797 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3798 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3802 * In case of mixed groups both data and meta should be picked,
3803 * and identical options should be given for both of them.
3805 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3806 if (mixed && (bctl->flags & allowed)) {
3807 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3808 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3809 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3811 "with mixed groups data and metadata balance options must be the same");
3817 num_devices = fs_info->fs_devices->num_devices;
3818 btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
3819 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3820 BUG_ON(num_devices < 1);
3823 btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
3824 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE | BTRFS_BLOCK_GROUP_DUP;
3825 if (num_devices > 1)
3826 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3827 if (num_devices > 2)
3828 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3829 if (num_devices > 3)
3830 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3831 BTRFS_BLOCK_GROUP_RAID6);
3832 if (validate_convert_profile(&bctl->data, allowed)) {
3834 "unable to start balance with target data profile %llu",
3839 if (validate_convert_profile(&bctl->meta, allowed)) {
3841 "unable to start balance with target metadata profile %llu",
3846 if (validate_convert_profile(&bctl->sys, allowed)) {
3848 "unable to start balance with target system profile %llu",
3854 /* allow to reduce meta or sys integrity only if force set */
3855 allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3856 BTRFS_BLOCK_GROUP_RAID10 |
3857 BTRFS_BLOCK_GROUP_RAID5 |
3858 BTRFS_BLOCK_GROUP_RAID6;
3860 seq = read_seqbegin(&fs_info->profiles_lock);
3862 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3863 (fs_info->avail_system_alloc_bits & allowed) &&
3864 !(bctl->sys.target & allowed)) ||
3865 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3866 (fs_info->avail_metadata_alloc_bits & allowed) &&
3867 !(bctl->meta.target & allowed))) {
3868 if (bctl->flags & BTRFS_BALANCE_FORCE) {
3870 "force reducing metadata integrity");
3873 "balance will reduce metadata integrity, use force if you want this");
3878 } while (read_seqretry(&fs_info->profiles_lock, seq));
3880 /* if we're not converting, the target field is uninitialized */
3881 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
3882 bctl->meta.target : fs_info->avail_metadata_alloc_bits;
3883 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
3884 bctl->data.target : fs_info->avail_data_alloc_bits;
3885 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <
3886 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {
3888 "metadata profile 0x%llx has lower redundancy than data profile 0x%llx",
3889 meta_target, data_target);
3892 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3893 fs_info->num_tolerated_disk_barrier_failures = min(
3894 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info),
3895 btrfs_get_num_tolerated_disk_barrier_failures(
3899 ret = insert_balance_item(fs_info, bctl);
3900 if (ret && ret != -EEXIST)
3903 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3904 BUG_ON(ret == -EEXIST);
3905 set_balance_control(bctl);
3907 BUG_ON(ret != -EEXIST);
3908 spin_lock(&fs_info->balance_lock);
3909 update_balance_args(bctl);
3910 spin_unlock(&fs_info->balance_lock);
3913 atomic_inc(&fs_info->balance_running);
3914 mutex_unlock(&fs_info->balance_mutex);
3916 ret = __btrfs_balance(fs_info);
3918 mutex_lock(&fs_info->balance_mutex);
3919 atomic_dec(&fs_info->balance_running);
3921 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3922 fs_info->num_tolerated_disk_barrier_failures =
3923 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3927 memset(bargs, 0, sizeof(*bargs));
3928 update_ioctl_balance_args(fs_info, 0, bargs);
3931 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3932 balance_need_close(fs_info)) {
3933 __cancel_balance(fs_info);
3936 wake_up(&fs_info->balance_wait_q);
3940 if (bctl->flags & BTRFS_BALANCE_RESUME)
3941 __cancel_balance(fs_info);
3944 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3949 static int balance_kthread(void *data)
3951 struct btrfs_fs_info *fs_info = data;
3954 mutex_lock(&fs_info->volume_mutex);
3955 mutex_lock(&fs_info->balance_mutex);
3957 if (fs_info->balance_ctl) {
3958 btrfs_info(fs_info, "continuing balance");
3959 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3962 mutex_unlock(&fs_info->balance_mutex);
3963 mutex_unlock(&fs_info->volume_mutex);
3968 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3970 struct task_struct *tsk;
3972 spin_lock(&fs_info->balance_lock);
3973 if (!fs_info->balance_ctl) {
3974 spin_unlock(&fs_info->balance_lock);
3977 spin_unlock(&fs_info->balance_lock);
3979 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
3980 btrfs_info(fs_info, "force skipping balance");
3984 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3985 return PTR_ERR_OR_ZERO(tsk);
3988 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3990 struct btrfs_balance_control *bctl;
3991 struct btrfs_balance_item *item;
3992 struct btrfs_disk_balance_args disk_bargs;
3993 struct btrfs_path *path;
3994 struct extent_buffer *leaf;
3995 struct btrfs_key key;
3998 path = btrfs_alloc_path();
4002 key.objectid = BTRFS_BALANCE_OBJECTID;
4003 key.type = BTRFS_TEMPORARY_ITEM_KEY;
4006 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4009 if (ret > 0) { /* ret = -ENOENT; */
4014 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4020 leaf = path->nodes[0];
4021 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
4023 bctl->fs_info = fs_info;
4024 bctl->flags = btrfs_balance_flags(leaf, item);
4025 bctl->flags |= BTRFS_BALANCE_RESUME;
4027 btrfs_balance_data(leaf, item, &disk_bargs);
4028 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4029 btrfs_balance_meta(leaf, item, &disk_bargs);
4030 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4031 btrfs_balance_sys(leaf, item, &disk_bargs);
4032 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4034 WARN_ON(test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags));
4036 mutex_lock(&fs_info->volume_mutex);
4037 mutex_lock(&fs_info->balance_mutex);
4039 set_balance_control(bctl);
4041 mutex_unlock(&fs_info->balance_mutex);
4042 mutex_unlock(&fs_info->volume_mutex);
4044 btrfs_free_path(path);
4048 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4052 mutex_lock(&fs_info->balance_mutex);
4053 if (!fs_info->balance_ctl) {
4054 mutex_unlock(&fs_info->balance_mutex);
4058 if (atomic_read(&fs_info->balance_running)) {
4059 atomic_inc(&fs_info->balance_pause_req);
4060 mutex_unlock(&fs_info->balance_mutex);
4062 wait_event(fs_info->balance_wait_q,
4063 atomic_read(&fs_info->balance_running) == 0);
4065 mutex_lock(&fs_info->balance_mutex);
4066 /* we are good with balance_ctl ripped off from under us */
4067 BUG_ON(atomic_read(&fs_info->balance_running));
4068 atomic_dec(&fs_info->balance_pause_req);
4073 mutex_unlock(&fs_info->balance_mutex);
4077 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4079 if (fs_info->sb->s_flags & MS_RDONLY)
4082 mutex_lock(&fs_info->balance_mutex);
4083 if (!fs_info->balance_ctl) {
4084 mutex_unlock(&fs_info->balance_mutex);
4088 atomic_inc(&fs_info->balance_cancel_req);
4090 * if we are running just wait and return, balance item is
4091 * deleted in btrfs_balance in this case
4093 if (atomic_read(&fs_info->balance_running)) {
4094 mutex_unlock(&fs_info->balance_mutex);
4095 wait_event(fs_info->balance_wait_q,
4096 atomic_read(&fs_info->balance_running) == 0);
4097 mutex_lock(&fs_info->balance_mutex);
4099 /* __cancel_balance needs volume_mutex */
4100 mutex_unlock(&fs_info->balance_mutex);
4101 mutex_lock(&fs_info->volume_mutex);
4102 mutex_lock(&fs_info->balance_mutex);
4104 if (fs_info->balance_ctl)
4105 __cancel_balance(fs_info);
4107 mutex_unlock(&fs_info->volume_mutex);
4110 BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
4111 atomic_dec(&fs_info->balance_cancel_req);
4112 mutex_unlock(&fs_info->balance_mutex);
4116 static int btrfs_uuid_scan_kthread(void *data)
4118 struct btrfs_fs_info *fs_info = data;
4119 struct btrfs_root *root = fs_info->tree_root;
4120 struct btrfs_key key;
4121 struct btrfs_key max_key;
4122 struct btrfs_path *path = NULL;
4124 struct extent_buffer *eb;
4126 struct btrfs_root_item root_item;
4128 struct btrfs_trans_handle *trans = NULL;
4130 path = btrfs_alloc_path();
4137 key.type = BTRFS_ROOT_ITEM_KEY;
4140 max_key.objectid = (u64)-1;
4141 max_key.type = BTRFS_ROOT_ITEM_KEY;
4142 max_key.offset = (u64)-1;
4145 ret = btrfs_search_forward(root, &key, path, 0);
4152 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4153 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4154 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4155 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4158 eb = path->nodes[0];
4159 slot = path->slots[0];
4160 item_size = btrfs_item_size_nr(eb, slot);
4161 if (item_size < sizeof(root_item))
4164 read_extent_buffer(eb, &root_item,
4165 btrfs_item_ptr_offset(eb, slot),
4166 (int)sizeof(root_item));
4167 if (btrfs_root_refs(&root_item) == 0)
4170 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4171 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4175 btrfs_release_path(path);
4177 * 1 - subvol uuid item
4178 * 1 - received_subvol uuid item
4180 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4181 if (IS_ERR(trans)) {
4182 ret = PTR_ERR(trans);
4190 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4191 ret = btrfs_uuid_tree_add(trans, fs_info,
4193 BTRFS_UUID_KEY_SUBVOL,
4196 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4202 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4203 ret = btrfs_uuid_tree_add(trans, fs_info,
4204 root_item.received_uuid,
4205 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4208 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4216 ret = btrfs_end_transaction(trans);
4222 btrfs_release_path(path);
4223 if (key.offset < (u64)-1) {
4225 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4227 key.type = BTRFS_ROOT_ITEM_KEY;
4228 } else if (key.objectid < (u64)-1) {
4230 key.type = BTRFS_ROOT_ITEM_KEY;
4239 btrfs_free_path(path);
4240 if (trans && !IS_ERR(trans))
4241 btrfs_end_transaction(trans);
4243 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4245 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4246 up(&fs_info->uuid_tree_rescan_sem);
4251 * Callback for btrfs_uuid_tree_iterate().
4253 * 0 check succeeded, the entry is not outdated.
4254 * < 0 if an error occurred.
4255 * > 0 if the check failed, which means the caller shall remove the entry.
4257 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
4258 u8 *uuid, u8 type, u64 subid)
4260 struct btrfs_key key;
4262 struct btrfs_root *subvol_root;
4264 if (type != BTRFS_UUID_KEY_SUBVOL &&
4265 type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
4268 key.objectid = subid;
4269 key.type = BTRFS_ROOT_ITEM_KEY;
4270 key.offset = (u64)-1;
4271 subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
4272 if (IS_ERR(subvol_root)) {
4273 ret = PTR_ERR(subvol_root);
4280 case BTRFS_UUID_KEY_SUBVOL:
4281 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
4284 case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
4285 if (memcmp(uuid, subvol_root->root_item.received_uuid,
4295 static int btrfs_uuid_rescan_kthread(void *data)
4297 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4301 * 1st step is to iterate through the existing UUID tree and
4302 * to delete all entries that contain outdated data.
4303 * 2nd step is to add all missing entries to the UUID tree.
4305 ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4307 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4308 up(&fs_info->uuid_tree_rescan_sem);
4311 return btrfs_uuid_scan_kthread(data);
4314 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4316 struct btrfs_trans_handle *trans;
4317 struct btrfs_root *tree_root = fs_info->tree_root;
4318 struct btrfs_root *uuid_root;
4319 struct task_struct *task;
4326 trans = btrfs_start_transaction(tree_root, 2);
4328 return PTR_ERR(trans);
4330 uuid_root = btrfs_create_tree(trans, fs_info,
4331 BTRFS_UUID_TREE_OBJECTID);
4332 if (IS_ERR(uuid_root)) {
4333 ret = PTR_ERR(uuid_root);
4334 btrfs_abort_transaction(trans, ret);
4335 btrfs_end_transaction(trans);
4339 fs_info->uuid_root = uuid_root;
4341 ret = btrfs_commit_transaction(trans);
4345 down(&fs_info->uuid_tree_rescan_sem);
4346 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4348 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4349 btrfs_warn(fs_info, "failed to start uuid_scan task");
4350 up(&fs_info->uuid_tree_rescan_sem);
4351 return PTR_ERR(task);
4357 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4359 struct task_struct *task;
4361 down(&fs_info->uuid_tree_rescan_sem);
4362 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4364 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4365 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4366 up(&fs_info->uuid_tree_rescan_sem);
4367 return PTR_ERR(task);
4374 * shrinking a device means finding all of the device extents past
4375 * the new size, and then following the back refs to the chunks.
4376 * The chunk relocation code actually frees the device extent
4378 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4380 struct btrfs_fs_info *fs_info = device->fs_info;
4381 struct btrfs_root *root = fs_info->dev_root;
4382 struct btrfs_trans_handle *trans;
4383 struct btrfs_dev_extent *dev_extent = NULL;
4384 struct btrfs_path *path;
4390 bool retried = false;
4391 bool checked_pending_chunks = false;
4392 struct extent_buffer *l;
4393 struct btrfs_key key;
4394 struct btrfs_super_block *super_copy = fs_info->super_copy;
4395 u64 old_total = btrfs_super_total_bytes(super_copy);
4396 u64 old_size = btrfs_device_get_total_bytes(device);
4397 u64 diff = old_size - new_size;
4399 if (device->is_tgtdev_for_dev_replace)
4402 path = btrfs_alloc_path();
4406 path->reada = READA_FORWARD;
4408 mutex_lock(&fs_info->chunk_mutex);
4410 btrfs_device_set_total_bytes(device, new_size);
4411 if (device->writeable) {
4412 device->fs_devices->total_rw_bytes -= diff;
4413 spin_lock(&fs_info->free_chunk_lock);
4414 fs_info->free_chunk_space -= diff;
4415 spin_unlock(&fs_info->free_chunk_lock);
4417 mutex_unlock(&fs_info->chunk_mutex);
4420 key.objectid = device->devid;
4421 key.offset = (u64)-1;
4422 key.type = BTRFS_DEV_EXTENT_KEY;
4425 mutex_lock(&fs_info->delete_unused_bgs_mutex);
4426 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4428 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4432 ret = btrfs_previous_item(root, path, 0, key.type);
4434 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4439 btrfs_release_path(path);
4444 slot = path->slots[0];
4445 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4447 if (key.objectid != device->devid) {
4448 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4449 btrfs_release_path(path);
4453 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4454 length = btrfs_dev_extent_length(l, dev_extent);
4456 if (key.offset + length <= new_size) {
4457 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4458 btrfs_release_path(path);
4462 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4463 btrfs_release_path(path);
4465 ret = btrfs_relocate_chunk(fs_info, chunk_offset);
4466 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4467 if (ret && ret != -ENOSPC)
4471 } while (key.offset-- > 0);
4473 if (failed && !retried) {
4477 } else if (failed && retried) {
4482 /* Shrinking succeeded, else we would be at "done". */
4483 trans = btrfs_start_transaction(root, 0);
4484 if (IS_ERR(trans)) {
4485 ret = PTR_ERR(trans);
4489 mutex_lock(&fs_info->chunk_mutex);
4492 * We checked in the above loop all device extents that were already in
4493 * the device tree. However before we have updated the device's
4494 * total_bytes to the new size, we might have had chunk allocations that
4495 * have not complete yet (new block groups attached to transaction
4496 * handles), and therefore their device extents were not yet in the
4497 * device tree and we missed them in the loop above. So if we have any
4498 * pending chunk using a device extent that overlaps the device range
4499 * that we can not use anymore, commit the current transaction and
4500 * repeat the search on the device tree - this way we guarantee we will
4501 * not have chunks using device extents that end beyond 'new_size'.
4503 if (!checked_pending_chunks) {
4504 u64 start = new_size;
4505 u64 len = old_size - new_size;
4507 if (contains_pending_extent(trans->transaction, device,
4509 mutex_unlock(&fs_info->chunk_mutex);
4510 checked_pending_chunks = true;
4513 ret = btrfs_commit_transaction(trans);
4520 btrfs_device_set_disk_total_bytes(device, new_size);
4521 if (list_empty(&device->resized_list))
4522 list_add_tail(&device->resized_list,
4523 &fs_info->fs_devices->resized_devices);
4525 WARN_ON(diff > old_total);
4526 btrfs_set_super_total_bytes(super_copy, old_total - diff);
4527 mutex_unlock(&fs_info->chunk_mutex);
4529 /* Now btrfs_update_device() will change the on-disk size. */
4530 ret = btrfs_update_device(trans, device);
4531 btrfs_end_transaction(trans);
4533 btrfs_free_path(path);
4535 mutex_lock(&fs_info->chunk_mutex);
4536 btrfs_device_set_total_bytes(device, old_size);
4537 if (device->writeable)
4538 device->fs_devices->total_rw_bytes += diff;
4539 spin_lock(&fs_info->free_chunk_lock);
4540 fs_info->free_chunk_space += diff;
4541 spin_unlock(&fs_info->free_chunk_lock);
4542 mutex_unlock(&fs_info->chunk_mutex);
4547 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
4548 struct btrfs_key *key,
4549 struct btrfs_chunk *chunk, int item_size)
4551 struct btrfs_super_block *super_copy = fs_info->super_copy;
4552 struct btrfs_disk_key disk_key;
4556 mutex_lock(&fs_info->chunk_mutex);
4557 array_size = btrfs_super_sys_array_size(super_copy);
4558 if (array_size + item_size + sizeof(disk_key)
4559 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4560 mutex_unlock(&fs_info->chunk_mutex);
4564 ptr = super_copy->sys_chunk_array + array_size;
4565 btrfs_cpu_key_to_disk(&disk_key, key);
4566 memcpy(ptr, &disk_key, sizeof(disk_key));
4567 ptr += sizeof(disk_key);
4568 memcpy(ptr, chunk, item_size);
4569 item_size += sizeof(disk_key);
4570 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4571 mutex_unlock(&fs_info->chunk_mutex);
4577 * sort the devices in descending order by max_avail, total_avail
4579 static int btrfs_cmp_device_info(const void *a, const void *b)
4581 const struct btrfs_device_info *di_a = a;
4582 const struct btrfs_device_info *di_b = b;
4584 if (di_a->max_avail > di_b->max_avail)
4586 if (di_a->max_avail < di_b->max_avail)
4588 if (di_a->total_avail > di_b->total_avail)
4590 if (di_a->total_avail < di_b->total_avail)
4595 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
4597 /* TODO allow them to set a preferred stripe size */
4601 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4603 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4606 btrfs_set_fs_incompat(info, RAID56);
4609 #define BTRFS_MAX_DEVS(r) ((BTRFS_MAX_ITEM_SIZE(r->fs_info) \
4610 - sizeof(struct btrfs_chunk)) \
4611 / sizeof(struct btrfs_stripe) + 1)
4613 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
4614 - 2 * sizeof(struct btrfs_disk_key) \
4615 - 2 * sizeof(struct btrfs_chunk)) \
4616 / sizeof(struct btrfs_stripe) + 1)
4618 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4619 u64 start, u64 type)
4621 struct btrfs_fs_info *info = trans->fs_info;
4622 struct btrfs_fs_devices *fs_devices = info->fs_devices;
4623 struct list_head *cur;
4624 struct map_lookup *map = NULL;
4625 struct extent_map_tree *em_tree;
4626 struct extent_map *em;
4627 struct btrfs_device_info *devices_info = NULL;
4629 int num_stripes; /* total number of stripes to allocate */
4630 int data_stripes; /* number of stripes that count for
4632 int sub_stripes; /* sub_stripes info for map */
4633 int dev_stripes; /* stripes per dev */
4634 int devs_max; /* max devs to use */
4635 int devs_min; /* min devs needed */
4636 int devs_increment; /* ndevs has to be a multiple of this */
4637 int ncopies; /* how many copies to data has */
4639 u64 max_stripe_size;
4643 u64 raid_stripe_len = BTRFS_STRIPE_LEN;
4649 BUG_ON(!alloc_profile_is_valid(type, 0));
4651 if (list_empty(&fs_devices->alloc_list))
4654 index = __get_raid_index(type);
4656 sub_stripes = btrfs_raid_array[index].sub_stripes;
4657 dev_stripes = btrfs_raid_array[index].dev_stripes;
4658 devs_max = btrfs_raid_array[index].devs_max;
4659 devs_min = btrfs_raid_array[index].devs_min;
4660 devs_increment = btrfs_raid_array[index].devs_increment;
4661 ncopies = btrfs_raid_array[index].ncopies;
4663 if (type & BTRFS_BLOCK_GROUP_DATA) {
4664 max_stripe_size = SZ_1G;
4665 max_chunk_size = 10 * max_stripe_size;
4667 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4668 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4669 /* for larger filesystems, use larger metadata chunks */
4670 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
4671 max_stripe_size = SZ_1G;
4673 max_stripe_size = SZ_256M;
4674 max_chunk_size = max_stripe_size;
4676 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4677 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4678 max_stripe_size = SZ_32M;
4679 max_chunk_size = 2 * max_stripe_size;
4681 devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4683 btrfs_err(info, "invalid chunk type 0x%llx requested",
4688 /* we don't want a chunk larger than 10% of writeable space */
4689 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4692 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4697 cur = fs_devices->alloc_list.next;
4700 * in the first pass through the devices list, we gather information
4701 * about the available holes on each device.
4704 while (cur != &fs_devices->alloc_list) {
4705 struct btrfs_device *device;
4709 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4713 if (!device->writeable) {
4715 "BTRFS: read-only device in alloc_list\n");
4719 if (!device->in_fs_metadata ||
4720 device->is_tgtdev_for_dev_replace)
4723 if (device->total_bytes > device->bytes_used)
4724 total_avail = device->total_bytes - device->bytes_used;
4728 /* If there is no space on this device, skip it. */
4729 if (total_avail == 0)
4732 ret = find_free_dev_extent(trans, device,
4733 max_stripe_size * dev_stripes,
4734 &dev_offset, &max_avail);
4735 if (ret && ret != -ENOSPC)
4739 max_avail = max_stripe_size * dev_stripes;
4741 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4744 if (ndevs == fs_devices->rw_devices) {
4745 WARN(1, "%s: found more than %llu devices\n",
4746 __func__, fs_devices->rw_devices);
4749 devices_info[ndevs].dev_offset = dev_offset;
4750 devices_info[ndevs].max_avail = max_avail;
4751 devices_info[ndevs].total_avail = total_avail;
4752 devices_info[ndevs].dev = device;
4757 * now sort the devices by hole size / available space
4759 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4760 btrfs_cmp_device_info, NULL);
4762 /* round down to number of usable stripes */
4763 ndevs -= ndevs % devs_increment;
4765 if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4770 if (devs_max && ndevs > devs_max)
4773 * the primary goal is to maximize the number of stripes, so use as many
4774 * devices as possible, even if the stripes are not maximum sized.
4776 stripe_size = devices_info[ndevs-1].max_avail;
4777 num_stripes = ndevs * dev_stripes;
4780 * this will have to be fixed for RAID1 and RAID10 over
4783 data_stripes = num_stripes / ncopies;
4785 if (type & BTRFS_BLOCK_GROUP_RAID5) {
4786 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
4788 data_stripes = num_stripes - 1;
4790 if (type & BTRFS_BLOCK_GROUP_RAID6) {
4791 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
4793 data_stripes = num_stripes - 2;
4797 * Use the number of data stripes to figure out how big this chunk
4798 * is really going to be in terms of logical address space,
4799 * and compare that answer with the max chunk size
4801 if (stripe_size * data_stripes > max_chunk_size) {
4802 u64 mask = (1ULL << 24) - 1;
4804 stripe_size = div_u64(max_chunk_size, data_stripes);
4806 /* bump the answer up to a 16MB boundary */
4807 stripe_size = (stripe_size + mask) & ~mask;
4809 /* but don't go higher than the limits we found
4810 * while searching for free extents
4812 if (stripe_size > devices_info[ndevs-1].max_avail)
4813 stripe_size = devices_info[ndevs-1].max_avail;
4816 stripe_size = div_u64(stripe_size, dev_stripes);
4818 /* align to BTRFS_STRIPE_LEN */
4819 stripe_size = div_u64(stripe_size, raid_stripe_len);
4820 stripe_size *= raid_stripe_len;
4822 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4827 map->num_stripes = num_stripes;
4829 for (i = 0; i < ndevs; ++i) {
4830 for (j = 0; j < dev_stripes; ++j) {
4831 int s = i * dev_stripes + j;
4832 map->stripes[s].dev = devices_info[i].dev;
4833 map->stripes[s].physical = devices_info[i].dev_offset +
4837 map->sector_size = info->sectorsize;
4838 map->stripe_len = raid_stripe_len;
4839 map->io_align = raid_stripe_len;
4840 map->io_width = raid_stripe_len;
4842 map->sub_stripes = sub_stripes;
4844 num_bytes = stripe_size * data_stripes;
4846 trace_btrfs_chunk_alloc(info, map, start, num_bytes);
4848 em = alloc_extent_map();
4854 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4855 em->map_lookup = map;
4857 em->len = num_bytes;
4858 em->block_start = 0;
4859 em->block_len = em->len;
4860 em->orig_block_len = stripe_size;
4862 em_tree = &info->mapping_tree.map_tree;
4863 write_lock(&em_tree->lock);
4864 ret = add_extent_mapping(em_tree, em, 0);
4866 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4867 refcount_inc(&em->refs);
4869 write_unlock(&em_tree->lock);
4871 free_extent_map(em);
4875 ret = btrfs_make_block_group(trans, info, 0, type,
4876 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4879 goto error_del_extent;
4881 for (i = 0; i < map->num_stripes; i++) {
4882 num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4883 btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4886 spin_lock(&info->free_chunk_lock);
4887 info->free_chunk_space -= (stripe_size * map->num_stripes);
4888 spin_unlock(&info->free_chunk_lock);
4890 free_extent_map(em);
4891 check_raid56_incompat_flag(info, type);
4893 kfree(devices_info);
4897 write_lock(&em_tree->lock);
4898 remove_extent_mapping(em_tree, em);
4899 write_unlock(&em_tree->lock);
4901 /* One for our allocation */
4902 free_extent_map(em);
4903 /* One for the tree reference */
4904 free_extent_map(em);
4905 /* One for the pending_chunks list reference */
4906 free_extent_map(em);
4908 kfree(devices_info);
4912 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4913 struct btrfs_fs_info *fs_info,
4914 u64 chunk_offset, u64 chunk_size)
4916 struct btrfs_root *extent_root = fs_info->extent_root;
4917 struct btrfs_root *chunk_root = fs_info->chunk_root;
4918 struct btrfs_key key;
4919 struct btrfs_device *device;
4920 struct btrfs_chunk *chunk;
4921 struct btrfs_stripe *stripe;
4922 struct extent_map *em;
4923 struct map_lookup *map;
4930 em = get_chunk_map(fs_info, chunk_offset, chunk_size);
4934 map = em->map_lookup;
4935 item_size = btrfs_chunk_item_size(map->num_stripes);
4936 stripe_size = em->orig_block_len;
4938 chunk = kzalloc(item_size, GFP_NOFS);
4945 * Take the device list mutex to prevent races with the final phase of
4946 * a device replace operation that replaces the device object associated
4947 * with the map's stripes, because the device object's id can change
4948 * at any time during that final phase of the device replace operation
4949 * (dev-replace.c:btrfs_dev_replace_finishing()).
4951 mutex_lock(&fs_info->fs_devices->device_list_mutex);
4952 for (i = 0; i < map->num_stripes; i++) {
4953 device = map->stripes[i].dev;
4954 dev_offset = map->stripes[i].physical;
4956 ret = btrfs_update_device(trans, device);
4959 ret = btrfs_alloc_dev_extent(trans, device,
4960 chunk_root->root_key.objectid,
4961 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4962 chunk_offset, dev_offset,
4968 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4972 stripe = &chunk->stripe;
4973 for (i = 0; i < map->num_stripes; i++) {
4974 device = map->stripes[i].dev;
4975 dev_offset = map->stripes[i].physical;
4977 btrfs_set_stack_stripe_devid(stripe, device->devid);
4978 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4979 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4982 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4984 btrfs_set_stack_chunk_length(chunk, chunk_size);
4985 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4986 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4987 btrfs_set_stack_chunk_type(chunk, map->type);
4988 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4989 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4990 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4991 btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
4992 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4994 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4995 key.type = BTRFS_CHUNK_ITEM_KEY;
4996 key.offset = chunk_offset;
4998 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4999 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
5001 * TODO: Cleanup of inserted chunk root in case of
5004 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
5009 free_extent_map(em);
5014 * Chunk allocation falls into two parts. The first part does works
5015 * that make the new allocated chunk useable, but not do any operation
5016 * that modifies the chunk tree. The second part does the works that
5017 * require modifying the chunk tree. This division is important for the
5018 * bootstrap process of adding storage to a seed btrfs.
5020 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
5021 struct btrfs_fs_info *fs_info, u64 type)
5025 ASSERT(mutex_is_locked(&fs_info->chunk_mutex));
5026 chunk_offset = find_next_chunk(fs_info);
5027 return __btrfs_alloc_chunk(trans, chunk_offset, type);
5030 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
5031 struct btrfs_fs_info *fs_info)
5033 struct btrfs_root *extent_root = fs_info->extent_root;
5035 u64 sys_chunk_offset;
5039 chunk_offset = find_next_chunk(fs_info);
5040 alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
5041 ret = __btrfs_alloc_chunk(trans, chunk_offset, alloc_profile);
5045 sys_chunk_offset = find_next_chunk(fs_info);
5046 alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
5047 ret = __btrfs_alloc_chunk(trans, sys_chunk_offset, alloc_profile);
5051 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
5055 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5056 BTRFS_BLOCK_GROUP_RAID10 |
5057 BTRFS_BLOCK_GROUP_RAID5 |
5058 BTRFS_BLOCK_GROUP_DUP)) {
5060 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
5069 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset)
5071 struct extent_map *em;
5072 struct map_lookup *map;
5077 em = get_chunk_map(fs_info, chunk_offset, 1);
5081 map = em->map_lookup;
5082 for (i = 0; i < map->num_stripes; i++) {
5083 if (map->stripes[i].dev->missing) {
5088 if (!map->stripes[i].dev->writeable) {
5095 * If the number of missing devices is larger than max errors,
5096 * we can not write the data into that chunk successfully, so
5099 if (miss_ndevs > btrfs_chunk_max_errors(map))
5102 free_extent_map(em);
5106 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
5108 extent_map_tree_init(&tree->map_tree);
5111 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
5113 struct extent_map *em;
5116 write_lock(&tree->map_tree.lock);
5117 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
5119 remove_extent_mapping(&tree->map_tree, em);
5120 write_unlock(&tree->map_tree.lock);
5124 free_extent_map(em);
5125 /* once for the tree */
5126 free_extent_map(em);
5130 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5132 struct extent_map *em;
5133 struct map_lookup *map;
5136 em = get_chunk_map(fs_info, logical, len);
5139 * We could return errors for these cases, but that could get
5140 * ugly and we'd probably do the same thing which is just not do
5141 * anything else and exit, so return 1 so the callers don't try
5142 * to use other copies.
5146 map = em->map_lookup;
5147 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
5148 ret = map->num_stripes;
5149 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5150 ret = map->sub_stripes;
5151 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5153 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5157 free_extent_map(em);
5159 btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
5160 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) &&
5161 fs_info->dev_replace.tgtdev)
5163 btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
5168 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
5169 struct btrfs_mapping_tree *map_tree,
5172 struct extent_map *em;
5173 struct map_lookup *map;
5174 unsigned long len = fs_info->sectorsize;
5176 em = get_chunk_map(fs_info, logical, len);
5177 WARN_ON(IS_ERR(em));
5179 map = em->map_lookup;
5180 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5181 len = map->stripe_len * nr_data_stripes(map);
5182 free_extent_map(em);
5186 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info,
5187 u64 logical, u64 len, int mirror_num)
5189 struct extent_map *em;
5190 struct map_lookup *map;
5193 em = get_chunk_map(fs_info, logical, len);
5194 WARN_ON(IS_ERR(em));
5196 map = em->map_lookup;
5197 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5199 free_extent_map(em);
5203 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5204 struct map_lookup *map, int first, int num,
5205 int optimal, int dev_replace_is_ongoing)
5209 struct btrfs_device *srcdev;
5211 if (dev_replace_is_ongoing &&
5212 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5213 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5214 srcdev = fs_info->dev_replace.srcdev;
5219 * try to avoid the drive that is the source drive for a
5220 * dev-replace procedure, only choose it if no other non-missing
5221 * mirror is available
5223 for (tolerance = 0; tolerance < 2; tolerance++) {
5224 if (map->stripes[optimal].dev->bdev &&
5225 (tolerance || map->stripes[optimal].dev != srcdev))
5227 for (i = first; i < first + num; i++) {
5228 if (map->stripes[i].dev->bdev &&
5229 (tolerance || map->stripes[i].dev != srcdev))
5234 /* we couldn't find one that doesn't fail. Just return something
5235 * and the io error handling code will clean up eventually
5240 static inline int parity_smaller(u64 a, u64 b)
5245 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5246 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5248 struct btrfs_bio_stripe s;
5255 for (i = 0; i < num_stripes - 1; i++) {
5256 if (parity_smaller(bbio->raid_map[i],
5257 bbio->raid_map[i+1])) {
5258 s = bbio->stripes[i];
5259 l = bbio->raid_map[i];
5260 bbio->stripes[i] = bbio->stripes[i+1];
5261 bbio->raid_map[i] = bbio->raid_map[i+1];
5262 bbio->stripes[i+1] = s;
5263 bbio->raid_map[i+1] = l;
5271 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5273 struct btrfs_bio *bbio = kzalloc(
5274 /* the size of the btrfs_bio */
5275 sizeof(struct btrfs_bio) +
5276 /* plus the variable array for the stripes */
5277 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5278 /* plus the variable array for the tgt dev */
5279 sizeof(int) * (real_stripes) +
5281 * plus the raid_map, which includes both the tgt dev
5284 sizeof(u64) * (total_stripes),
5285 GFP_NOFS|__GFP_NOFAIL);
5287 atomic_set(&bbio->error, 0);
5288 refcount_set(&bbio->refs, 1);
5293 void btrfs_get_bbio(struct btrfs_bio *bbio)
5295 WARN_ON(!refcount_read(&bbio->refs));
5296 refcount_inc(&bbio->refs);
5299 void btrfs_put_bbio(struct btrfs_bio *bbio)
5303 if (refcount_dec_and_test(&bbio->refs))
5307 /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */
5309 * Please note that, discard won't be sent to target device of device
5312 static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info,
5313 u64 logical, u64 length,
5314 struct btrfs_bio **bbio_ret)
5316 struct extent_map *em;
5317 struct map_lookup *map;
5318 struct btrfs_bio *bbio;
5322 u64 stripe_end_offset;
5329 u32 sub_stripes = 0;
5330 u64 stripes_per_dev = 0;
5331 u32 remaining_stripes = 0;
5332 u32 last_stripe = 0;
5336 /* discard always return a bbio */
5339 em = get_chunk_map(fs_info, logical, length);
5343 map = em->map_lookup;
5344 /* we don't discard raid56 yet */
5345 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5350 offset = logical - em->start;
5351 length = min_t(u64, em->len - offset, length);
5353 stripe_len = map->stripe_len;
5355 * stripe_nr counts the total number of stripes we have to stride
5356 * to get to this block
5358 stripe_nr = div64_u64(offset, stripe_len);
5360 /* stripe_offset is the offset of this block in its stripe */
5361 stripe_offset = offset - stripe_nr * stripe_len;
5363 stripe_nr_end = round_up(offset + length, map->stripe_len);
5364 stripe_nr_end = div_u64(stripe_nr_end, map->stripe_len);
5365 stripe_cnt = stripe_nr_end - stripe_nr;
5366 stripe_end_offset = stripe_nr_end * map->stripe_len -
5369 * after this, stripe_nr is the number of stripes on this
5370 * device we have to walk to find the data, and stripe_index is
5371 * the number of our device in the stripe array
5375 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5376 BTRFS_BLOCK_GROUP_RAID10)) {
5377 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5380 sub_stripes = map->sub_stripes;
5382 factor = map->num_stripes / sub_stripes;
5383 num_stripes = min_t(u64, map->num_stripes,
5384 sub_stripes * stripe_cnt);
5385 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5386 stripe_index *= sub_stripes;
5387 stripes_per_dev = div_u64_rem(stripe_cnt, factor,
5388 &remaining_stripes);
5389 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5390 last_stripe *= sub_stripes;
5391 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5392 BTRFS_BLOCK_GROUP_DUP)) {
5393 num_stripes = map->num_stripes;
5395 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5399 bbio = alloc_btrfs_bio(num_stripes, 0);
5405 for (i = 0; i < num_stripes; i++) {
5406 bbio->stripes[i].physical =
5407 map->stripes[stripe_index].physical +
5408 stripe_offset + stripe_nr * map->stripe_len;
5409 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5411 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5412 BTRFS_BLOCK_GROUP_RAID10)) {
5413 bbio->stripes[i].length = stripes_per_dev *
5416 if (i / sub_stripes < remaining_stripes)
5417 bbio->stripes[i].length +=
5421 * Special for the first stripe and
5424 * |-------|...|-------|
5428 if (i < sub_stripes)
5429 bbio->stripes[i].length -=
5432 if (stripe_index >= last_stripe &&
5433 stripe_index <= (last_stripe +
5435 bbio->stripes[i].length -=
5438 if (i == sub_stripes - 1)
5441 bbio->stripes[i].length = length;
5445 if (stripe_index == map->num_stripes) {
5452 bbio->map_type = map->type;
5453 bbio->num_stripes = num_stripes;
5455 free_extent_map(em);
5460 * In dev-replace case, for repair case (that's the only case where the mirror
5461 * is selected explicitly when calling btrfs_map_block), blocks left of the
5462 * left cursor can also be read from the target drive.
5464 * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the
5466 * For READ, it also needs to be supported using the same mirror number.
5468 * If the requested block is not left of the left cursor, EIO is returned. This
5469 * can happen because btrfs_num_copies() returns one more in the dev-replace
5472 static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info,
5473 u64 logical, u64 length,
5474 u64 srcdev_devid, int *mirror_num,
5477 struct btrfs_bio *bbio = NULL;
5479 int index_srcdev = 0;
5481 u64 physical_of_found = 0;
5485 ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS,
5486 logical, &length, &bbio, 0, 0);
5488 ASSERT(bbio == NULL);
5492 num_stripes = bbio->num_stripes;
5493 if (*mirror_num > num_stripes) {
5495 * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror,
5496 * that means that the requested area is not left of the left
5499 btrfs_put_bbio(bbio);
5504 * process the rest of the function using the mirror_num of the source
5505 * drive. Therefore look it up first. At the end, patch the device
5506 * pointer to the one of the target drive.
5508 for (i = 0; i < num_stripes; i++) {
5509 if (bbio->stripes[i].dev->devid != srcdev_devid)
5513 * In case of DUP, in order to keep it simple, only add the
5514 * mirror with the lowest physical address
5517 physical_of_found <= bbio->stripes[i].physical)
5522 physical_of_found = bbio->stripes[i].physical;
5525 btrfs_put_bbio(bbio);
5531 *mirror_num = index_srcdev + 1;
5532 *physical = physical_of_found;
5536 static void handle_ops_on_dev_replace(enum btrfs_map_op op,
5537 struct btrfs_bio **bbio_ret,
5538 struct btrfs_dev_replace *dev_replace,
5539 int *num_stripes_ret, int *max_errors_ret)
5541 struct btrfs_bio *bbio = *bbio_ret;
5542 u64 srcdev_devid = dev_replace->srcdev->devid;
5543 int tgtdev_indexes = 0;
5544 int num_stripes = *num_stripes_ret;
5545 int max_errors = *max_errors_ret;
5548 if (op == BTRFS_MAP_WRITE) {
5549 int index_where_to_add;
5552 * duplicate the write operations while the dev replace
5553 * procedure is running. Since the copying of the old disk to
5554 * the new disk takes place at run time while the filesystem is
5555 * mounted writable, the regular write operations to the old
5556 * disk have to be duplicated to go to the new disk as well.
5558 * Note that device->missing is handled by the caller, and that
5559 * the write to the old disk is already set up in the stripes
5562 index_where_to_add = num_stripes;
5563 for (i = 0; i < num_stripes; i++) {
5564 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5565 /* write to new disk, too */
5566 struct btrfs_bio_stripe *new =
5567 bbio->stripes + index_where_to_add;
5568 struct btrfs_bio_stripe *old =
5571 new->physical = old->physical;
5572 new->length = old->length;
5573 new->dev = dev_replace->tgtdev;
5574 bbio->tgtdev_map[i] = index_where_to_add;
5575 index_where_to_add++;
5580 num_stripes = index_where_to_add;
5581 } else if (op == BTRFS_MAP_GET_READ_MIRRORS) {
5582 int index_srcdev = 0;
5584 u64 physical_of_found = 0;
5587 * During the dev-replace procedure, the target drive can also
5588 * be used to read data in case it is needed to repair a corrupt
5589 * block elsewhere. This is possible if the requested area is
5590 * left of the left cursor. In this area, the target drive is a
5591 * full copy of the source drive.
5593 for (i = 0; i < num_stripes; i++) {
5594 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5596 * In case of DUP, in order to keep it simple,
5597 * only add the mirror with the lowest physical
5601 physical_of_found <=
5602 bbio->stripes[i].physical)
5606 physical_of_found = bbio->stripes[i].physical;
5610 struct btrfs_bio_stripe *tgtdev_stripe =
5611 bbio->stripes + num_stripes;
5613 tgtdev_stripe->physical = physical_of_found;
5614 tgtdev_stripe->length =
5615 bbio->stripes[index_srcdev].length;
5616 tgtdev_stripe->dev = dev_replace->tgtdev;
5617 bbio->tgtdev_map[index_srcdev] = num_stripes;
5624 *num_stripes_ret = num_stripes;
5625 *max_errors_ret = max_errors;
5626 bbio->num_tgtdevs = tgtdev_indexes;
5630 static bool need_full_stripe(enum btrfs_map_op op)
5632 return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS);
5635 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
5636 enum btrfs_map_op op,
5637 u64 logical, u64 *length,
5638 struct btrfs_bio **bbio_ret,
5639 int mirror_num, int need_raid_map)
5641 struct extent_map *em;
5642 struct map_lookup *map;
5652 int tgtdev_indexes = 0;
5653 struct btrfs_bio *bbio = NULL;
5654 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
5655 int dev_replace_is_ongoing = 0;
5656 int num_alloc_stripes;
5657 int patch_the_first_stripe_for_dev_replace = 0;
5658 u64 physical_to_patch_in_first_stripe = 0;
5659 u64 raid56_full_stripe_start = (u64)-1;
5661 if (op == BTRFS_MAP_DISCARD)
5662 return __btrfs_map_block_for_discard(fs_info, logical,
5665 em = get_chunk_map(fs_info, logical, *length);
5669 map = em->map_lookup;
5670 offset = logical - em->start;
5672 stripe_len = map->stripe_len;
5675 * stripe_nr counts the total number of stripes we have to stride
5676 * to get to this block
5678 stripe_nr = div64_u64(stripe_nr, stripe_len);
5680 stripe_offset = stripe_nr * stripe_len;
5681 if (offset < stripe_offset) {
5683 "stripe math has gone wrong, stripe_offset=%llu, offset=%llu, start=%llu, logical=%llu, stripe_len=%llu",
5684 stripe_offset, offset, em->start, logical,
5686 free_extent_map(em);
5690 /* stripe_offset is the offset of this block in its stripe*/
5691 stripe_offset = offset - stripe_offset;
5693 /* if we're here for raid56, we need to know the stripe aligned start */
5694 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5695 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5696 raid56_full_stripe_start = offset;
5698 /* allow a write of a full stripe, but make sure we don't
5699 * allow straddling of stripes
5701 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5703 raid56_full_stripe_start *= full_stripe_len;
5706 if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5708 /* For writes to RAID[56], allow a full stripeset across all disks.
5709 For other RAID types and for RAID[56] reads, just allow a single
5710 stripe (on a single disk). */
5711 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
5712 (op == BTRFS_MAP_WRITE)) {
5713 max_len = stripe_len * nr_data_stripes(map) -
5714 (offset - raid56_full_stripe_start);
5716 /* we limit the length of each bio to what fits in a stripe */
5717 max_len = stripe_len - stripe_offset;
5719 *length = min_t(u64, em->len - offset, max_len);
5721 *length = em->len - offset;
5724 /* This is for when we're called from btrfs_merge_bio_hook() and all
5725 it cares about is the length */
5729 btrfs_dev_replace_lock(dev_replace, 0);
5730 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5731 if (!dev_replace_is_ongoing)
5732 btrfs_dev_replace_unlock(dev_replace, 0);
5734 btrfs_dev_replace_set_lock_blocking(dev_replace);
5736 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
5737 !need_full_stripe(op) && dev_replace->tgtdev != NULL) {
5738 ret = get_extra_mirror_from_replace(fs_info, logical, *length,
5739 dev_replace->srcdev->devid,
5741 &physical_to_patch_in_first_stripe);
5745 patch_the_first_stripe_for_dev_replace = 1;
5746 } else if (mirror_num > map->num_stripes) {
5752 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5753 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5755 if (op != BTRFS_MAP_WRITE && op != BTRFS_MAP_GET_READ_MIRRORS)
5757 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
5758 if (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS)
5759 num_stripes = map->num_stripes;
5760 else if (mirror_num)
5761 stripe_index = mirror_num - 1;
5763 stripe_index = find_live_mirror(fs_info, map, 0,
5765 current->pid % map->num_stripes,
5766 dev_replace_is_ongoing);
5767 mirror_num = stripe_index + 1;
5770 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
5771 if (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS) {
5772 num_stripes = map->num_stripes;
5773 } else if (mirror_num) {
5774 stripe_index = mirror_num - 1;
5779 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5780 u32 factor = map->num_stripes / map->sub_stripes;
5782 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5783 stripe_index *= map->sub_stripes;
5785 if (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS)
5786 num_stripes = map->sub_stripes;
5787 else if (mirror_num)
5788 stripe_index += mirror_num - 1;
5790 int old_stripe_index = stripe_index;
5791 stripe_index = find_live_mirror(fs_info, map,
5793 map->sub_stripes, stripe_index +
5794 current->pid % map->sub_stripes,
5795 dev_replace_is_ongoing);
5796 mirror_num = stripe_index - old_stripe_index + 1;
5799 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5800 if (need_raid_map &&
5801 (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS ||
5803 /* push stripe_nr back to the start of the full stripe */
5804 stripe_nr = div_u64(raid56_full_stripe_start,
5805 stripe_len * nr_data_stripes(map));
5807 /* RAID[56] write or recovery. Return all stripes */
5808 num_stripes = map->num_stripes;
5809 max_errors = nr_parity_stripes(map);
5811 *length = map->stripe_len;
5816 * Mirror #0 or #1 means the original data block.
5817 * Mirror #2 is RAID5 parity block.
5818 * Mirror #3 is RAID6 Q block.
5820 stripe_nr = div_u64_rem(stripe_nr,
5821 nr_data_stripes(map), &stripe_index);
5823 stripe_index = nr_data_stripes(map) +
5826 /* We distribute the parity blocks across stripes */
5827 div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
5829 if ((op != BTRFS_MAP_WRITE &&
5830 op != BTRFS_MAP_GET_READ_MIRRORS) &&
5836 * after this, stripe_nr is the number of stripes on this
5837 * device we have to walk to find the data, and stripe_index is
5838 * the number of our device in the stripe array
5840 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5842 mirror_num = stripe_index + 1;
5844 if (stripe_index >= map->num_stripes) {
5846 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
5847 stripe_index, map->num_stripes);
5852 num_alloc_stripes = num_stripes;
5853 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) {
5854 if (op == BTRFS_MAP_WRITE)
5855 num_alloc_stripes <<= 1;
5856 if (op == BTRFS_MAP_GET_READ_MIRRORS)
5857 num_alloc_stripes++;
5858 tgtdev_indexes = num_stripes;
5861 bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
5866 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL)
5867 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
5869 /* build raid_map */
5870 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map &&
5871 (need_full_stripe(op) || mirror_num > 1)) {
5875 bbio->raid_map = (u64 *)((void *)bbio->stripes +
5876 sizeof(struct btrfs_bio_stripe) *
5878 sizeof(int) * tgtdev_indexes);
5880 /* Work out the disk rotation on this stripe-set */
5881 div_u64_rem(stripe_nr, num_stripes, &rot);
5883 /* Fill in the logical address of each stripe */
5884 tmp = stripe_nr * nr_data_stripes(map);
5885 for (i = 0; i < nr_data_stripes(map); i++)
5886 bbio->raid_map[(i+rot) % num_stripes] =
5887 em->start + (tmp + i) * map->stripe_len;
5889 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5890 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5891 bbio->raid_map[(i+rot+1) % num_stripes] =
5896 for (i = 0; i < num_stripes; i++) {
5897 bbio->stripes[i].physical =
5898 map->stripes[stripe_index].physical +
5900 stripe_nr * map->stripe_len;
5901 bbio->stripes[i].dev =
5902 map->stripes[stripe_index].dev;
5906 if (need_full_stripe(op))
5907 max_errors = btrfs_chunk_max_errors(map);
5910 sort_parity_stripes(bbio, num_stripes);
5912 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
5913 need_full_stripe(op)) {
5914 handle_ops_on_dev_replace(op, &bbio, dev_replace, &num_stripes,
5919 bbio->map_type = map->type;
5920 bbio->num_stripes = num_stripes;
5921 bbio->max_errors = max_errors;
5922 bbio->mirror_num = mirror_num;
5925 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5926 * mirror_num == num_stripes + 1 && dev_replace target drive is
5927 * available as a mirror
5929 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5930 WARN_ON(num_stripes > 1);
5931 bbio->stripes[0].dev = dev_replace->tgtdev;
5932 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5933 bbio->mirror_num = map->num_stripes + 1;
5936 if (dev_replace_is_ongoing) {
5937 btrfs_dev_replace_clear_lock_blocking(dev_replace);
5938 btrfs_dev_replace_unlock(dev_replace, 0);
5940 free_extent_map(em);
5944 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
5945 u64 logical, u64 *length,
5946 struct btrfs_bio **bbio_ret, int mirror_num)
5948 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret,
5952 /* For Scrub/replace */
5953 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
5954 u64 logical, u64 *length,
5955 struct btrfs_bio **bbio_ret)
5957 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1);
5960 int btrfs_rmap_block(struct btrfs_fs_info *fs_info,
5961 u64 chunk_start, u64 physical, u64 devid,
5962 u64 **logical, int *naddrs, int *stripe_len)
5964 struct extent_map *em;
5965 struct map_lookup *map;
5973 em = get_chunk_map(fs_info, chunk_start, 1);
5977 map = em->map_lookup;
5979 rmap_len = map->stripe_len;
5981 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5982 length = div_u64(length, map->num_stripes / map->sub_stripes);
5983 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5984 length = div_u64(length, map->num_stripes);
5985 else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5986 length = div_u64(length, nr_data_stripes(map));
5987 rmap_len = map->stripe_len * nr_data_stripes(map);
5990 buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
5991 BUG_ON(!buf); /* -ENOMEM */
5993 for (i = 0; i < map->num_stripes; i++) {
5994 if (devid && map->stripes[i].dev->devid != devid)
5996 if (map->stripes[i].physical > physical ||
5997 map->stripes[i].physical + length <= physical)
6000 stripe_nr = physical - map->stripes[i].physical;
6001 stripe_nr = div_u64(stripe_nr, map->stripe_len);
6003 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
6004 stripe_nr = stripe_nr * map->num_stripes + i;
6005 stripe_nr = div_u64(stripe_nr, map->sub_stripes);
6006 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
6007 stripe_nr = stripe_nr * map->num_stripes + i;
6008 } /* else if RAID[56], multiply by nr_data_stripes().
6009 * Alternatively, just use rmap_len below instead of
6010 * map->stripe_len */
6012 bytenr = chunk_start + stripe_nr * rmap_len;
6013 WARN_ON(nr >= map->num_stripes);
6014 for (j = 0; j < nr; j++) {
6015 if (buf[j] == bytenr)
6019 WARN_ON(nr >= map->num_stripes);
6026 *stripe_len = rmap_len;
6028 free_extent_map(em);
6032 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
6034 bio->bi_private = bbio->private;
6035 bio->bi_end_io = bbio->end_io;
6038 btrfs_put_bbio(bbio);
6041 static void btrfs_end_bio(struct bio *bio)
6043 struct btrfs_bio *bbio = bio->bi_private;
6044 int is_orig_bio = 0;
6046 if (bio->bi_error) {
6047 atomic_inc(&bbio->error);
6048 if (bio->bi_error == -EIO || bio->bi_error == -EREMOTEIO) {
6049 unsigned int stripe_index =
6050 btrfs_io_bio(bio)->stripe_index;
6051 struct btrfs_device *dev;
6053 BUG_ON(stripe_index >= bbio->num_stripes);
6054 dev = bbio->stripes[stripe_index].dev;
6056 if (bio_op(bio) == REQ_OP_WRITE)
6057 btrfs_dev_stat_inc(dev,
6058 BTRFS_DEV_STAT_WRITE_ERRS);
6060 btrfs_dev_stat_inc(dev,
6061 BTRFS_DEV_STAT_READ_ERRS);
6062 if (bio->bi_opf & REQ_PREFLUSH)
6063 btrfs_dev_stat_inc(dev,
6064 BTRFS_DEV_STAT_FLUSH_ERRS);
6065 btrfs_dev_stat_print_on_error(dev);
6070 if (bio == bbio->orig_bio)
6073 btrfs_bio_counter_dec(bbio->fs_info);
6075 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6078 bio = bbio->orig_bio;
6081 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6082 /* only send an error to the higher layers if it is
6083 * beyond the tolerance of the btrfs bio
6085 if (atomic_read(&bbio->error) > bbio->max_errors) {
6086 bio->bi_error = -EIO;
6089 * this bio is actually up to date, we didn't
6090 * go over the max number of errors
6095 btrfs_end_bbio(bbio, bio);
6096 } else if (!is_orig_bio) {
6102 * see run_scheduled_bios for a description of why bios are collected for
6105 * This will add one bio to the pending list for a device and make sure
6106 * the work struct is scheduled.
6108 static noinline void btrfs_schedule_bio(struct btrfs_device *device,
6111 struct btrfs_fs_info *fs_info = device->fs_info;
6112 int should_queue = 1;
6113 struct btrfs_pending_bios *pending_bios;
6115 if (device->missing || !device->bdev) {
6120 /* don't bother with additional async steps for reads, right now */
6121 if (bio_op(bio) == REQ_OP_READ) {
6123 btrfsic_submit_bio(bio);
6129 * nr_async_bios allows us to reliably return congestion to the
6130 * higher layers. Otherwise, the async bio makes it appear we have
6131 * made progress against dirty pages when we've really just put it
6132 * on a queue for later
6134 atomic_inc(&fs_info->nr_async_bios);
6135 WARN_ON(bio->bi_next);
6136 bio->bi_next = NULL;
6138 spin_lock(&device->io_lock);
6139 if (op_is_sync(bio->bi_opf))
6140 pending_bios = &device->pending_sync_bios;
6142 pending_bios = &device->pending_bios;
6144 if (pending_bios->tail)
6145 pending_bios->tail->bi_next = bio;
6147 pending_bios->tail = bio;
6148 if (!pending_bios->head)
6149 pending_bios->head = bio;
6150 if (device->running_pending)
6153 spin_unlock(&device->io_lock);
6156 btrfs_queue_work(fs_info->submit_workers, &device->work);
6159 static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio,
6160 u64 physical, int dev_nr, int async)
6162 struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
6163 struct btrfs_fs_info *fs_info = bbio->fs_info;
6165 bio->bi_private = bbio;
6166 btrfs_io_bio(bio)->stripe_index = dev_nr;
6167 bio->bi_end_io = btrfs_end_bio;
6168 bio->bi_iter.bi_sector = physical >> 9;
6171 struct rcu_string *name;
6174 name = rcu_dereference(dev->name);
6175 btrfs_debug(fs_info,
6176 "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
6177 bio_op(bio), bio->bi_opf,
6178 (u64)bio->bi_iter.bi_sector,
6179 (u_long)dev->bdev->bd_dev, name->str, dev->devid,
6180 bio->bi_iter.bi_size);
6184 bio->bi_bdev = dev->bdev;
6186 btrfs_bio_counter_inc_noblocked(fs_info);
6189 btrfs_schedule_bio(dev, bio);
6191 btrfsic_submit_bio(bio);
6194 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6196 atomic_inc(&bbio->error);
6197 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6198 /* Should be the original bio. */
6199 WARN_ON(bio != bbio->orig_bio);
6201 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6202 bio->bi_iter.bi_sector = logical >> 9;
6203 bio->bi_error = -EIO;
6204 btrfs_end_bbio(bbio, bio);
6208 int btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
6209 int mirror_num, int async_submit)
6211 struct btrfs_device *dev;
6212 struct bio *first_bio = bio;
6213 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
6219 struct btrfs_bio *bbio = NULL;
6221 length = bio->bi_iter.bi_size;
6222 map_length = length;
6224 btrfs_bio_counter_inc_blocked(fs_info);
6225 ret = __btrfs_map_block(fs_info, bio_op(bio), logical,
6226 &map_length, &bbio, mirror_num, 1);
6228 btrfs_bio_counter_dec(fs_info);
6232 total_devs = bbio->num_stripes;
6233 bbio->orig_bio = first_bio;
6234 bbio->private = first_bio->bi_private;
6235 bbio->end_io = first_bio->bi_end_io;
6236 bbio->fs_info = fs_info;
6237 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6239 if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6240 ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) {
6241 /* In this case, map_length has been set to the length of
6242 a single stripe; not the whole write */
6243 if (bio_op(bio) == REQ_OP_WRITE) {
6244 ret = raid56_parity_write(fs_info, bio, bbio,
6247 ret = raid56_parity_recover(fs_info, bio, bbio,
6248 map_length, mirror_num, 1);
6251 btrfs_bio_counter_dec(fs_info);
6255 if (map_length < length) {
6257 "mapping failed logical %llu bio len %llu len %llu",
6258 logical, length, map_length);
6262 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6263 dev = bbio->stripes[dev_nr].dev;
6264 if (!dev || !dev->bdev ||
6265 (bio_op(first_bio) == REQ_OP_WRITE && !dev->writeable)) {
6266 bbio_error(bbio, first_bio, logical);
6270 if (dev_nr < total_devs - 1) {
6271 bio = btrfs_bio_clone(first_bio, GFP_NOFS);
6272 BUG_ON(!bio); /* -ENOMEM */
6276 submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical,
6277 dev_nr, async_submit);
6279 btrfs_bio_counter_dec(fs_info);
6283 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
6286 struct btrfs_device *device;
6287 struct btrfs_fs_devices *cur_devices;
6289 cur_devices = fs_info->fs_devices;
6290 while (cur_devices) {
6292 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
6293 device = __find_device(&cur_devices->devices,
6298 cur_devices = cur_devices->seed;
6303 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices,
6304 u64 devid, u8 *dev_uuid)
6306 struct btrfs_device *device;
6308 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6312 list_add(&device->dev_list, &fs_devices->devices);
6313 device->fs_devices = fs_devices;
6314 fs_devices->num_devices++;
6316 device->missing = 1;
6317 fs_devices->missing_devices++;
6323 * btrfs_alloc_device - allocate struct btrfs_device
6324 * @fs_info: used only for generating a new devid, can be NULL if
6325 * devid is provided (i.e. @devid != NULL).
6326 * @devid: a pointer to devid for this device. If NULL a new devid
6328 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6331 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6332 * on error. Returned struct is not linked onto any lists and can be
6333 * destroyed with kfree() right away.
6335 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6339 struct btrfs_device *dev;
6342 if (WARN_ON(!devid && !fs_info))
6343 return ERR_PTR(-EINVAL);
6345 dev = __alloc_device();
6354 ret = find_next_devid(fs_info, &tmp);
6357 return ERR_PTR(ret);
6363 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6365 generate_random_uuid(dev->uuid);
6367 btrfs_init_work(&dev->work, btrfs_submit_helper,
6368 pending_bios_fn, NULL, NULL);
6373 /* Return -EIO if any error, otherwise return 0. */
6374 static int btrfs_check_chunk_valid(struct btrfs_fs_info *fs_info,
6375 struct extent_buffer *leaf,
6376 struct btrfs_chunk *chunk, u64 logical)
6384 length = btrfs_chunk_length(leaf, chunk);
6385 stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6386 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6387 sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6388 type = btrfs_chunk_type(leaf, chunk);
6391 btrfs_err(fs_info, "invalid chunk num_stripes: %u",
6395 if (!IS_ALIGNED(logical, fs_info->sectorsize)) {
6396 btrfs_err(fs_info, "invalid chunk logical %llu", logical);
6399 if (btrfs_chunk_sector_size(leaf, chunk) != fs_info->sectorsize) {
6400 btrfs_err(fs_info, "invalid chunk sectorsize %u",
6401 btrfs_chunk_sector_size(leaf, chunk));
6404 if (!length || !IS_ALIGNED(length, fs_info->sectorsize)) {
6405 btrfs_err(fs_info, "invalid chunk length %llu", length);
6408 if (!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN) {
6409 btrfs_err(fs_info, "invalid chunk stripe length: %llu",
6413 if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6415 btrfs_err(fs_info, "unrecognized chunk type: %llu",
6416 ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
6417 BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6418 btrfs_chunk_type(leaf, chunk));
6421 if ((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes != 2) ||
6422 (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) ||
6423 (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) ||
6424 (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) ||
6425 (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) ||
6426 ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
6427 num_stripes != 1)) {
6429 "invalid num_stripes:sub_stripes %u:%u for profile %llu",
6430 num_stripes, sub_stripes,
6431 type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
6438 static int read_one_chunk(struct btrfs_fs_info *fs_info, struct btrfs_key *key,
6439 struct extent_buffer *leaf,
6440 struct btrfs_chunk *chunk)
6442 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
6443 struct map_lookup *map;
6444 struct extent_map *em;
6449 u8 uuid[BTRFS_UUID_SIZE];
6454 logical = key->offset;
6455 length = btrfs_chunk_length(leaf, chunk);
6456 stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6457 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6459 ret = btrfs_check_chunk_valid(fs_info, leaf, chunk, logical);
6463 read_lock(&map_tree->map_tree.lock);
6464 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
6465 read_unlock(&map_tree->map_tree.lock);
6467 /* already mapped? */
6468 if (em && em->start <= logical && em->start + em->len > logical) {
6469 free_extent_map(em);
6472 free_extent_map(em);
6475 em = alloc_extent_map();
6478 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6480 free_extent_map(em);
6484 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6485 em->map_lookup = map;
6486 em->start = logical;
6489 em->block_start = 0;
6490 em->block_len = em->len;
6492 map->num_stripes = num_stripes;
6493 map->io_width = btrfs_chunk_io_width(leaf, chunk);
6494 map->io_align = btrfs_chunk_io_align(leaf, chunk);
6495 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
6496 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6497 map->type = btrfs_chunk_type(leaf, chunk);
6498 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6499 for (i = 0; i < num_stripes; i++) {
6500 map->stripes[i].physical =
6501 btrfs_stripe_offset_nr(leaf, chunk, i);
6502 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6503 read_extent_buffer(leaf, uuid, (unsigned long)
6504 btrfs_stripe_dev_uuid_nr(chunk, i),
6506 map->stripes[i].dev = btrfs_find_device(fs_info, devid,
6508 if (!map->stripes[i].dev &&
6509 !btrfs_test_opt(fs_info, DEGRADED)) {
6510 free_extent_map(em);
6513 if (!map->stripes[i].dev) {
6514 map->stripes[i].dev =
6515 add_missing_dev(fs_info->fs_devices, devid,
6517 if (!map->stripes[i].dev) {
6518 free_extent_map(em);
6521 btrfs_warn(fs_info, "devid %llu uuid %pU is missing",
6524 map->stripes[i].dev->in_fs_metadata = 1;
6527 write_lock(&map_tree->map_tree.lock);
6528 ret = add_extent_mapping(&map_tree->map_tree, em, 0);
6529 write_unlock(&map_tree->map_tree.lock);
6530 BUG_ON(ret); /* Tree corruption */
6531 free_extent_map(em);
6536 static void fill_device_from_item(struct extent_buffer *leaf,
6537 struct btrfs_dev_item *dev_item,
6538 struct btrfs_device *device)
6542 device->devid = btrfs_device_id(leaf, dev_item);
6543 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6544 device->total_bytes = device->disk_total_bytes;
6545 device->commit_total_bytes = device->disk_total_bytes;
6546 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6547 device->commit_bytes_used = device->bytes_used;
6548 device->type = btrfs_device_type(leaf, dev_item);
6549 device->io_align = btrfs_device_io_align(leaf, dev_item);
6550 device->io_width = btrfs_device_io_width(leaf, dev_item);
6551 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6552 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6553 device->is_tgtdev_for_dev_replace = 0;
6555 ptr = btrfs_device_uuid(dev_item);
6556 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6559 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info,
6562 struct btrfs_fs_devices *fs_devices;
6565 BUG_ON(!mutex_is_locked(&uuid_mutex));
6567 fs_devices = fs_info->fs_devices->seed;
6568 while (fs_devices) {
6569 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE))
6572 fs_devices = fs_devices->seed;
6575 fs_devices = find_fsid(fsid);
6577 if (!btrfs_test_opt(fs_info, DEGRADED))
6578 return ERR_PTR(-ENOENT);
6580 fs_devices = alloc_fs_devices(fsid);
6581 if (IS_ERR(fs_devices))
6584 fs_devices->seeding = 1;
6585 fs_devices->opened = 1;
6589 fs_devices = clone_fs_devices(fs_devices);
6590 if (IS_ERR(fs_devices))
6593 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
6594 fs_info->bdev_holder);
6596 free_fs_devices(fs_devices);
6597 fs_devices = ERR_PTR(ret);
6601 if (!fs_devices->seeding) {
6602 __btrfs_close_devices(fs_devices);
6603 free_fs_devices(fs_devices);
6604 fs_devices = ERR_PTR(-EINVAL);
6608 fs_devices->seed = fs_info->fs_devices->seed;
6609 fs_info->fs_devices->seed = fs_devices;
6614 static int read_one_dev(struct btrfs_fs_info *fs_info,
6615 struct extent_buffer *leaf,
6616 struct btrfs_dev_item *dev_item)
6618 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6619 struct btrfs_device *device;
6622 u8 fs_uuid[BTRFS_UUID_SIZE];
6623 u8 dev_uuid[BTRFS_UUID_SIZE];
6625 devid = btrfs_device_id(leaf, dev_item);
6626 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6628 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6631 if (memcmp(fs_uuid, fs_info->fsid, BTRFS_UUID_SIZE)) {
6632 fs_devices = open_seed_devices(fs_info, fs_uuid);
6633 if (IS_ERR(fs_devices))
6634 return PTR_ERR(fs_devices);
6637 device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
6639 if (!btrfs_test_opt(fs_info, DEGRADED))
6642 device = add_missing_dev(fs_devices, devid, dev_uuid);
6645 btrfs_warn(fs_info, "devid %llu uuid %pU missing",
6648 if (!device->bdev && !btrfs_test_opt(fs_info, DEGRADED))
6651 if(!device->bdev && !device->missing) {
6653 * this happens when a device that was properly setup
6654 * in the device info lists suddenly goes bad.
6655 * device->bdev is NULL, and so we have to set
6656 * device->missing to one here
6658 device->fs_devices->missing_devices++;
6659 device->missing = 1;
6662 /* Move the device to its own fs_devices */
6663 if (device->fs_devices != fs_devices) {
6664 ASSERT(device->missing);
6666 list_move(&device->dev_list, &fs_devices->devices);
6667 device->fs_devices->num_devices--;
6668 fs_devices->num_devices++;
6670 device->fs_devices->missing_devices--;
6671 fs_devices->missing_devices++;
6673 device->fs_devices = fs_devices;
6677 if (device->fs_devices != fs_info->fs_devices) {
6678 BUG_ON(device->writeable);
6679 if (device->generation !=
6680 btrfs_device_generation(leaf, dev_item))
6684 fill_device_from_item(leaf, dev_item, device);
6685 device->in_fs_metadata = 1;
6686 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
6687 device->fs_devices->total_rw_bytes += device->total_bytes;
6688 spin_lock(&fs_info->free_chunk_lock);
6689 fs_info->free_chunk_space += device->total_bytes -
6691 spin_unlock(&fs_info->free_chunk_lock);
6697 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
6699 struct btrfs_root *root = fs_info->tree_root;
6700 struct btrfs_super_block *super_copy = fs_info->super_copy;
6701 struct extent_buffer *sb;
6702 struct btrfs_disk_key *disk_key;
6703 struct btrfs_chunk *chunk;
6705 unsigned long sb_array_offset;
6712 struct btrfs_key key;
6714 ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize);
6716 * This will create extent buffer of nodesize, superblock size is
6717 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6718 * overallocate but we can keep it as-is, only the first page is used.
6720 sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET);
6723 set_extent_buffer_uptodate(sb);
6724 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6726 * The sb extent buffer is artificial and just used to read the system array.
6727 * set_extent_buffer_uptodate() call does not properly mark all it's
6728 * pages up-to-date when the page is larger: extent does not cover the
6729 * whole page and consequently check_page_uptodate does not find all
6730 * the page's extents up-to-date (the hole beyond sb),
6731 * write_extent_buffer then triggers a WARN_ON.
6733 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6734 * but sb spans only this function. Add an explicit SetPageUptodate call
6735 * to silence the warning eg. on PowerPC 64.
6737 if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
6738 SetPageUptodate(sb->pages[0]);
6740 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6741 array_size = btrfs_super_sys_array_size(super_copy);
6743 array_ptr = super_copy->sys_chunk_array;
6744 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6747 while (cur_offset < array_size) {
6748 disk_key = (struct btrfs_disk_key *)array_ptr;
6749 len = sizeof(*disk_key);
6750 if (cur_offset + len > array_size)
6751 goto out_short_read;
6753 btrfs_disk_key_to_cpu(&key, disk_key);
6756 sb_array_offset += len;
6759 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6760 chunk = (struct btrfs_chunk *)sb_array_offset;
6762 * At least one btrfs_chunk with one stripe must be
6763 * present, exact stripe count check comes afterwards
6765 len = btrfs_chunk_item_size(1);
6766 if (cur_offset + len > array_size)
6767 goto out_short_read;
6769 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6772 "invalid number of stripes %u in sys_array at offset %u",
6773 num_stripes, cur_offset);
6778 type = btrfs_chunk_type(sb, chunk);
6779 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
6781 "invalid chunk type %llu in sys_array at offset %u",
6787 len = btrfs_chunk_item_size(num_stripes);
6788 if (cur_offset + len > array_size)
6789 goto out_short_read;
6791 ret = read_one_chunk(fs_info, &key, sb, chunk);
6796 "unexpected item type %u in sys_array at offset %u",
6797 (u32)key.type, cur_offset);
6802 sb_array_offset += len;
6805 clear_extent_buffer_uptodate(sb);
6806 free_extent_buffer_stale(sb);
6810 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
6812 clear_extent_buffer_uptodate(sb);
6813 free_extent_buffer_stale(sb);
6817 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
6819 struct btrfs_root *root = fs_info->chunk_root;
6820 struct btrfs_path *path;
6821 struct extent_buffer *leaf;
6822 struct btrfs_key key;
6823 struct btrfs_key found_key;
6828 path = btrfs_alloc_path();
6832 mutex_lock(&uuid_mutex);
6833 mutex_lock(&fs_info->chunk_mutex);
6836 * Read all device items, and then all the chunk items. All
6837 * device items are found before any chunk item (their object id
6838 * is smaller than the lowest possible object id for a chunk
6839 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6841 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6844 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6848 leaf = path->nodes[0];
6849 slot = path->slots[0];
6850 if (slot >= btrfs_header_nritems(leaf)) {
6851 ret = btrfs_next_leaf(root, path);
6858 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6859 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6860 struct btrfs_dev_item *dev_item;
6861 dev_item = btrfs_item_ptr(leaf, slot,
6862 struct btrfs_dev_item);
6863 ret = read_one_dev(fs_info, leaf, dev_item);
6867 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6868 struct btrfs_chunk *chunk;
6869 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6870 ret = read_one_chunk(fs_info, &found_key, leaf, chunk);
6878 * After loading chunk tree, we've got all device information,
6879 * do another round of validation checks.
6881 if (total_dev != fs_info->fs_devices->total_devices) {
6883 "super_num_devices %llu mismatch with num_devices %llu found here",
6884 btrfs_super_num_devices(fs_info->super_copy),
6889 if (btrfs_super_total_bytes(fs_info->super_copy) <
6890 fs_info->fs_devices->total_rw_bytes) {
6892 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
6893 btrfs_super_total_bytes(fs_info->super_copy),
6894 fs_info->fs_devices->total_rw_bytes);
6900 mutex_unlock(&fs_info->chunk_mutex);
6901 mutex_unlock(&uuid_mutex);
6903 btrfs_free_path(path);
6907 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6909 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6910 struct btrfs_device *device;
6912 while (fs_devices) {
6913 mutex_lock(&fs_devices->device_list_mutex);
6914 list_for_each_entry(device, &fs_devices->devices, dev_list)
6915 device->fs_info = fs_info;
6916 mutex_unlock(&fs_devices->device_list_mutex);
6918 fs_devices = fs_devices->seed;
6922 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6926 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6927 btrfs_dev_stat_reset(dev, i);
6930 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6932 struct btrfs_key key;
6933 struct btrfs_key found_key;
6934 struct btrfs_root *dev_root = fs_info->dev_root;
6935 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6936 struct extent_buffer *eb;
6939 struct btrfs_device *device;
6940 struct btrfs_path *path = NULL;
6943 path = btrfs_alloc_path();
6949 mutex_lock(&fs_devices->device_list_mutex);
6950 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6952 struct btrfs_dev_stats_item *ptr;
6954 key.objectid = BTRFS_DEV_STATS_OBJECTID;
6955 key.type = BTRFS_PERSISTENT_ITEM_KEY;
6956 key.offset = device->devid;
6957 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6959 __btrfs_reset_dev_stats(device);
6960 device->dev_stats_valid = 1;
6961 btrfs_release_path(path);
6964 slot = path->slots[0];
6965 eb = path->nodes[0];
6966 btrfs_item_key_to_cpu(eb, &found_key, slot);
6967 item_size = btrfs_item_size_nr(eb, slot);
6969 ptr = btrfs_item_ptr(eb, slot,
6970 struct btrfs_dev_stats_item);
6972 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6973 if (item_size >= (1 + i) * sizeof(__le64))
6974 btrfs_dev_stat_set(device, i,
6975 btrfs_dev_stats_value(eb, ptr, i));
6977 btrfs_dev_stat_reset(device, i);
6980 device->dev_stats_valid = 1;
6981 btrfs_dev_stat_print_on_load(device);
6982 btrfs_release_path(path);
6984 mutex_unlock(&fs_devices->device_list_mutex);
6987 btrfs_free_path(path);
6988 return ret < 0 ? ret : 0;
6991 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
6992 struct btrfs_fs_info *fs_info,
6993 struct btrfs_device *device)
6995 struct btrfs_root *dev_root = fs_info->dev_root;
6996 struct btrfs_path *path;
6997 struct btrfs_key key;
6998 struct extent_buffer *eb;
6999 struct btrfs_dev_stats_item *ptr;
7003 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7004 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7005 key.offset = device->devid;
7007 path = btrfs_alloc_path();
7010 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
7012 btrfs_warn_in_rcu(fs_info,
7013 "error %d while searching for dev_stats item for device %s",
7014 ret, rcu_str_deref(device->name));
7019 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
7020 /* need to delete old one and insert a new one */
7021 ret = btrfs_del_item(trans, dev_root, path);
7023 btrfs_warn_in_rcu(fs_info,
7024 "delete too small dev_stats item for device %s failed %d",
7025 rcu_str_deref(device->name), ret);
7032 /* need to insert a new item */
7033 btrfs_release_path(path);
7034 ret = btrfs_insert_empty_item(trans, dev_root, path,
7035 &key, sizeof(*ptr));
7037 btrfs_warn_in_rcu(fs_info,
7038 "insert dev_stats item for device %s failed %d",
7039 rcu_str_deref(device->name), ret);
7044 eb = path->nodes[0];
7045 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7046 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7047 btrfs_set_dev_stats_value(eb, ptr, i,
7048 btrfs_dev_stat_read(device, i));
7049 btrfs_mark_buffer_dirty(eb);
7052 btrfs_free_path(path);
7057 * called from commit_transaction. Writes all changed device stats to disk.
7059 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
7060 struct btrfs_fs_info *fs_info)
7062 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7063 struct btrfs_device *device;
7067 mutex_lock(&fs_devices->device_list_mutex);
7068 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7069 if (!device->dev_stats_valid || !btrfs_dev_stats_dirty(device))
7072 stats_cnt = atomic_read(&device->dev_stats_ccnt);
7073 ret = update_dev_stat_item(trans, fs_info, device);
7075 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7077 mutex_unlock(&fs_devices->device_list_mutex);
7082 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7084 btrfs_dev_stat_inc(dev, index);
7085 btrfs_dev_stat_print_on_error(dev);
7088 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
7090 if (!dev->dev_stats_valid)
7092 btrfs_err_rl_in_rcu(dev->fs_info,
7093 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7094 rcu_str_deref(dev->name),
7095 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7096 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7097 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7098 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7099 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7102 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7106 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7107 if (btrfs_dev_stat_read(dev, i) != 0)
7109 if (i == BTRFS_DEV_STAT_VALUES_MAX)
7110 return; /* all values == 0, suppress message */
7112 btrfs_info_in_rcu(dev->fs_info,
7113 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7114 rcu_str_deref(dev->name),
7115 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7116 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7117 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7118 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7119 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7122 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
7123 struct btrfs_ioctl_get_dev_stats *stats)
7125 struct btrfs_device *dev;
7126 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7129 mutex_lock(&fs_devices->device_list_mutex);
7130 dev = btrfs_find_device(fs_info, stats->devid, NULL, NULL);
7131 mutex_unlock(&fs_devices->device_list_mutex);
7134 btrfs_warn(fs_info, "get dev_stats failed, device not found");
7136 } else if (!dev->dev_stats_valid) {
7137 btrfs_warn(fs_info, "get dev_stats failed, not yet valid");
7139 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7140 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7141 if (stats->nr_items > i)
7143 btrfs_dev_stat_read_and_reset(dev, i);
7145 btrfs_dev_stat_reset(dev, i);
7148 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7149 if (stats->nr_items > i)
7150 stats->values[i] = btrfs_dev_stat_read(dev, i);
7152 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7153 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7157 void btrfs_scratch_superblocks(struct block_device *bdev, const char *device_path)
7159 struct buffer_head *bh;
7160 struct btrfs_super_block *disk_super;
7166 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX;
7169 if (btrfs_read_dev_one_super(bdev, copy_num, &bh))
7172 disk_super = (struct btrfs_super_block *)bh->b_data;
7174 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
7175 set_buffer_dirty(bh);
7176 sync_dirty_buffer(bh);
7180 /* Notify udev that device has changed */
7181 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
7183 /* Update ctime/mtime for device path for libblkid */
7184 update_dev_time(device_path);
7188 * Update the size of all devices, which is used for writing out the
7191 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
7193 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7194 struct btrfs_device *curr, *next;
7196 if (list_empty(&fs_devices->resized_devices))
7199 mutex_lock(&fs_devices->device_list_mutex);
7200 mutex_lock(&fs_info->chunk_mutex);
7201 list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
7203 list_del_init(&curr->resized_list);
7204 curr->commit_total_bytes = curr->disk_total_bytes;
7206 mutex_unlock(&fs_info->chunk_mutex);
7207 mutex_unlock(&fs_devices->device_list_mutex);
7210 /* Must be invoked during the transaction commit */
7211 void btrfs_update_commit_device_bytes_used(struct btrfs_fs_info *fs_info,
7212 struct btrfs_transaction *transaction)
7214 struct extent_map *em;
7215 struct map_lookup *map;
7216 struct btrfs_device *dev;
7219 if (list_empty(&transaction->pending_chunks))
7222 /* In order to kick the device replace finish process */
7223 mutex_lock(&fs_info->chunk_mutex);
7224 list_for_each_entry(em, &transaction->pending_chunks, list) {
7225 map = em->map_lookup;
7227 for (i = 0; i < map->num_stripes; i++) {
7228 dev = map->stripes[i].dev;
7229 dev->commit_bytes_used = dev->bytes_used;
7232 mutex_unlock(&fs_info->chunk_mutex);
7235 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
7237 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7238 while (fs_devices) {
7239 fs_devices->fs_info = fs_info;
7240 fs_devices = fs_devices->seed;
7244 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
7246 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7247 while (fs_devices) {
7248 fs_devices->fs_info = NULL;
7249 fs_devices = fs_devices->seed;