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)
156 * alloc_fs_devices - allocate struct btrfs_fs_devices
157 * @fsid: if not NULL, copy the uuid to fs_devices::fsid
159 * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR().
160 * The returned struct is not linked onto any lists and can be destroyed with
161 * kfree() right away.
163 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
165 struct btrfs_fs_devices *fs_devs;
167 fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL);
169 return ERR_PTR(-ENOMEM);
171 mutex_init(&fs_devs->device_list_mutex);
173 INIT_LIST_HEAD(&fs_devs->devices);
174 INIT_LIST_HEAD(&fs_devs->resized_devices);
175 INIT_LIST_HEAD(&fs_devs->alloc_list);
176 INIT_LIST_HEAD(&fs_devs->list);
178 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
183 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
185 struct btrfs_device *device;
186 WARN_ON(fs_devices->opened);
187 while (!list_empty(&fs_devices->devices)) {
188 device = list_entry(fs_devices->devices.next,
189 struct btrfs_device, dev_list);
190 list_del(&device->dev_list);
191 rcu_string_free(device->name);
197 static void btrfs_kobject_uevent(struct block_device *bdev,
198 enum kobject_action action)
202 ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
204 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
206 kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
207 &disk_to_dev(bdev->bd_disk)->kobj);
210 void btrfs_cleanup_fs_uuids(void)
212 struct btrfs_fs_devices *fs_devices;
214 while (!list_empty(&fs_uuids)) {
215 fs_devices = list_entry(fs_uuids.next,
216 struct btrfs_fs_devices, list);
217 list_del(&fs_devices->list);
218 free_fs_devices(fs_devices);
222 static struct btrfs_device *__alloc_device(void)
224 struct btrfs_device *dev;
226 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
228 return ERR_PTR(-ENOMEM);
231 * Preallocate a bio that's always going to be used for flushing device
232 * barriers and matches the device lifespan
234 dev->flush_bio = bio_alloc_bioset(GFP_KERNEL, 0, NULL);
235 if (!dev->flush_bio) {
237 return ERR_PTR(-ENOMEM);
240 INIT_LIST_HEAD(&dev->dev_list);
241 INIT_LIST_HEAD(&dev->dev_alloc_list);
242 INIT_LIST_HEAD(&dev->resized_list);
244 spin_lock_init(&dev->io_lock);
246 spin_lock_init(&dev->reada_lock);
247 atomic_set(&dev->reada_in_flight, 0);
248 atomic_set(&dev->dev_stats_ccnt, 0);
249 btrfs_device_data_ordered_init(dev);
250 INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
251 INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
257 * Find a device specified by @devid or @uuid in the list of @fs_devices, or
260 * If devid and uuid are both specified, the match must be exact, otherwise
261 * only devid is used.
263 static struct btrfs_device *find_device(struct btrfs_fs_devices *fs_devices,
264 u64 devid, const u8 *uuid)
266 struct list_head *head = &fs_devices->devices;
267 struct btrfs_device *dev;
269 list_for_each_entry(dev, head, dev_list) {
270 if (dev->devid == devid &&
271 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
278 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
280 struct btrfs_fs_devices *fs_devices;
282 list_for_each_entry(fs_devices, &fs_uuids, list) {
283 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
290 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
291 int flush, struct block_device **bdev,
292 struct buffer_head **bh)
296 *bdev = blkdev_get_by_path(device_path, flags, holder);
299 ret = PTR_ERR(*bdev);
304 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
305 ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE);
307 blkdev_put(*bdev, flags);
310 invalidate_bdev(*bdev);
311 *bh = btrfs_read_dev_super(*bdev);
314 blkdev_put(*bdev, flags);
326 static void requeue_list(struct btrfs_pending_bios *pending_bios,
327 struct bio *head, struct bio *tail)
330 struct bio *old_head;
332 old_head = pending_bios->head;
333 pending_bios->head = head;
334 if (pending_bios->tail)
335 tail->bi_next = old_head;
337 pending_bios->tail = tail;
341 * we try to collect pending bios for a device so we don't get a large
342 * number of procs sending bios down to the same device. This greatly
343 * improves the schedulers ability to collect and merge the bios.
345 * But, it also turns into a long list of bios to process and that is sure
346 * to eventually make the worker thread block. The solution here is to
347 * make some progress and then put this work struct back at the end of
348 * the list if the block device is congested. This way, multiple devices
349 * can make progress from a single worker thread.
351 static noinline void run_scheduled_bios(struct btrfs_device *device)
353 struct btrfs_fs_info *fs_info = device->fs_info;
355 struct backing_dev_info *bdi;
356 struct btrfs_pending_bios *pending_bios;
360 unsigned long num_run;
361 unsigned long batch_run = 0;
363 unsigned long last_waited = 0;
365 int sync_pending = 0;
366 struct blk_plug plug;
369 * this function runs all the bios we've collected for
370 * a particular device. We don't want to wander off to
371 * another device without first sending all of these down.
372 * So, setup a plug here and finish it off before we return
374 blk_start_plug(&plug);
376 bdi = device->bdev->bd_bdi;
377 limit = btrfs_async_submit_limit(fs_info);
378 limit = limit * 2 / 3;
381 spin_lock(&device->io_lock);
386 /* take all the bios off the list at once and process them
387 * later on (without the lock held). But, remember the
388 * tail and other pointers so the bios can be properly reinserted
389 * into the list if we hit congestion
391 if (!force_reg && device->pending_sync_bios.head) {
392 pending_bios = &device->pending_sync_bios;
395 pending_bios = &device->pending_bios;
399 pending = pending_bios->head;
400 tail = pending_bios->tail;
401 WARN_ON(pending && !tail);
404 * if pending was null this time around, no bios need processing
405 * at all and we can stop. Otherwise it'll loop back up again
406 * and do an additional check so no bios are missed.
408 * device->running_pending is used to synchronize with the
411 if (device->pending_sync_bios.head == NULL &&
412 device->pending_bios.head == NULL) {
414 device->running_pending = 0;
417 device->running_pending = 1;
420 pending_bios->head = NULL;
421 pending_bios->tail = NULL;
423 spin_unlock(&device->io_lock);
428 /* we want to work on both lists, but do more bios on the
429 * sync list than the regular list
432 pending_bios != &device->pending_sync_bios &&
433 device->pending_sync_bios.head) ||
434 (num_run > 64 && pending_bios == &device->pending_sync_bios &&
435 device->pending_bios.head)) {
436 spin_lock(&device->io_lock);
437 requeue_list(pending_bios, pending, tail);
442 pending = pending->bi_next;
446 * atomic_dec_return implies a barrier for waitqueue_active
448 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
449 waitqueue_active(&fs_info->async_submit_wait))
450 wake_up(&fs_info->async_submit_wait);
452 BUG_ON(atomic_read(&cur->__bi_cnt) == 0);
455 * if we're doing the sync list, record that our
456 * plug has some sync requests on it
458 * If we're doing the regular list and there are
459 * sync requests sitting around, unplug before
462 if (pending_bios == &device->pending_sync_bios) {
464 } else if (sync_pending) {
465 blk_finish_plug(&plug);
466 blk_start_plug(&plug);
470 btrfsic_submit_bio(cur);
477 * we made progress, there is more work to do and the bdi
478 * is now congested. Back off and let other work structs
481 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
482 fs_info->fs_devices->open_devices > 1) {
483 struct io_context *ioc;
485 ioc = current->io_context;
488 * the main goal here is that we don't want to
489 * block if we're going to be able to submit
490 * more requests without blocking.
492 * This code does two great things, it pokes into
493 * the elevator code from a filesystem _and_
494 * it makes assumptions about how batching works.
496 if (ioc && ioc->nr_batch_requests > 0 &&
497 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
499 ioc->last_waited == last_waited)) {
501 * we want to go through our batch of
502 * requests and stop. So, we copy out
503 * the ioc->last_waited time and test
504 * against it before looping
506 last_waited = ioc->last_waited;
510 spin_lock(&device->io_lock);
511 requeue_list(pending_bios, pending, tail);
512 device->running_pending = 1;
514 spin_unlock(&device->io_lock);
515 btrfs_queue_work(fs_info->submit_workers,
519 /* unplug every 64 requests just for good measure */
520 if (batch_run % 64 == 0) {
521 blk_finish_plug(&plug);
522 blk_start_plug(&plug);
531 spin_lock(&device->io_lock);
532 if (device->pending_bios.head || device->pending_sync_bios.head)
534 spin_unlock(&device->io_lock);
537 blk_finish_plug(&plug);
540 static void pending_bios_fn(struct btrfs_work *work)
542 struct btrfs_device *device;
544 device = container_of(work, struct btrfs_device, work);
545 run_scheduled_bios(device);
549 void btrfs_free_stale_device(struct btrfs_device *cur_dev)
551 struct btrfs_fs_devices *fs_devs;
552 struct btrfs_device *dev;
557 list_for_each_entry(fs_devs, &fs_uuids, list) {
562 if (fs_devs->seeding)
565 list_for_each_entry(dev, &fs_devs->devices, dev_list) {
573 * Todo: This won't be enough. What if the same device
574 * comes back (with new uuid and) with its mapper path?
575 * But for now, this does help as mostly an admin will
576 * either use mapper or non mapper path throughout.
579 del = strcmp(rcu_str_deref(dev->name),
580 rcu_str_deref(cur_dev->name));
587 /* delete the stale device */
588 if (fs_devs->num_devices == 1) {
589 btrfs_sysfs_remove_fsid(fs_devs);
590 list_del(&fs_devs->list);
591 free_fs_devices(fs_devs);
594 fs_devs->num_devices--;
595 list_del(&dev->dev_list);
596 rcu_string_free(dev->name);
605 * Add new device to list of registered devices
608 * 1 - first time device is seen
609 * 0 - device already known
612 static noinline int device_list_add(const char *path,
613 struct btrfs_super_block *disk_super,
614 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
616 struct btrfs_device *device;
617 struct btrfs_fs_devices *fs_devices;
618 struct rcu_string *name;
620 u64 found_transid = btrfs_super_generation(disk_super);
622 fs_devices = find_fsid(disk_super->fsid);
624 fs_devices = alloc_fs_devices(disk_super->fsid);
625 if (IS_ERR(fs_devices))
626 return PTR_ERR(fs_devices);
628 list_add(&fs_devices->list, &fs_uuids);
632 device = find_device(fs_devices, devid,
633 disk_super->dev_item.uuid);
637 if (fs_devices->opened)
640 device = btrfs_alloc_device(NULL, &devid,
641 disk_super->dev_item.uuid);
642 if (IS_ERR(device)) {
643 /* we can safely leave the fs_devices entry around */
644 return PTR_ERR(device);
647 name = rcu_string_strdup(path, GFP_NOFS);
652 rcu_assign_pointer(device->name, name);
654 mutex_lock(&fs_devices->device_list_mutex);
655 list_add_rcu(&device->dev_list, &fs_devices->devices);
656 fs_devices->num_devices++;
657 mutex_unlock(&fs_devices->device_list_mutex);
660 device->fs_devices = fs_devices;
661 } else if (!device->name || strcmp(device->name->str, path)) {
663 * When FS is already mounted.
664 * 1. If you are here and if the device->name is NULL that
665 * means this device was missing at time of FS mount.
666 * 2. If you are here and if the device->name is different
667 * from 'path' that means either
668 * a. The same device disappeared and reappeared with
670 * b. The missing-disk-which-was-replaced, has
673 * We must allow 1 and 2a above. But 2b would be a spurious
676 * Further in case of 1 and 2a above, the disk at 'path'
677 * would have missed some transaction when it was away and
678 * in case of 2a the stale bdev has to be updated as well.
679 * 2b must not be allowed at all time.
683 * For now, we do allow update to btrfs_fs_device through the
684 * btrfs dev scan cli after FS has been mounted. We're still
685 * tracking a problem where systems fail mount by subvolume id
686 * when we reject replacement on a mounted FS.
688 if (!fs_devices->opened && found_transid < device->generation) {
690 * That is if the FS is _not_ mounted and if you
691 * are here, that means there is more than one
692 * disk with same uuid and devid.We keep the one
693 * with larger generation number or the last-in if
694 * generation are equal.
699 name = rcu_string_strdup(path, GFP_NOFS);
702 rcu_string_free(device->name);
703 rcu_assign_pointer(device->name, name);
704 if (device->missing) {
705 fs_devices->missing_devices--;
711 * Unmount does not free the btrfs_device struct but would zero
712 * generation along with most of the other members. So just update
713 * it back. We need it to pick the disk with largest generation
716 if (!fs_devices->opened)
717 device->generation = found_transid;
720 * if there is new btrfs on an already registered device,
721 * then remove the stale device entry.
724 btrfs_free_stale_device(device);
726 *fs_devices_ret = fs_devices;
731 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
733 struct btrfs_fs_devices *fs_devices;
734 struct btrfs_device *device;
735 struct btrfs_device *orig_dev;
737 fs_devices = alloc_fs_devices(orig->fsid);
738 if (IS_ERR(fs_devices))
741 mutex_lock(&orig->device_list_mutex);
742 fs_devices->total_devices = orig->total_devices;
744 /* We have held the volume lock, it is safe to get the devices. */
745 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
746 struct rcu_string *name;
748 device = btrfs_alloc_device(NULL, &orig_dev->devid,
754 * This is ok to do without rcu read locked because we hold the
755 * uuid mutex so nothing we touch in here is going to disappear.
757 if (orig_dev->name) {
758 name = rcu_string_strdup(orig_dev->name->str,
764 rcu_assign_pointer(device->name, name);
767 list_add(&device->dev_list, &fs_devices->devices);
768 device->fs_devices = fs_devices;
769 fs_devices->num_devices++;
771 mutex_unlock(&orig->device_list_mutex);
774 mutex_unlock(&orig->device_list_mutex);
775 free_fs_devices(fs_devices);
776 return ERR_PTR(-ENOMEM);
779 void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step)
781 struct btrfs_device *device, *next;
782 struct btrfs_device *latest_dev = NULL;
784 mutex_lock(&uuid_mutex);
786 /* This is the initialized path, it is safe to release the devices. */
787 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
788 if (device->in_fs_metadata) {
789 if (!device->is_tgtdev_for_dev_replace &&
791 device->generation > latest_dev->generation)) {
797 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
799 * In the first step, keep the device which has
800 * the correct fsid and the devid that is used
801 * for the dev_replace procedure.
802 * In the second step, the dev_replace state is
803 * read from the device tree and it is known
804 * whether the procedure is really active or
805 * not, which means whether this device is
806 * used or whether it should be removed.
808 if (step == 0 || device->is_tgtdev_for_dev_replace) {
813 blkdev_put(device->bdev, device->mode);
815 fs_devices->open_devices--;
817 if (device->writeable) {
818 list_del_init(&device->dev_alloc_list);
819 device->writeable = 0;
820 if (!device->is_tgtdev_for_dev_replace)
821 fs_devices->rw_devices--;
823 list_del_init(&device->dev_list);
824 fs_devices->num_devices--;
825 rcu_string_free(device->name);
829 if (fs_devices->seed) {
830 fs_devices = fs_devices->seed;
834 fs_devices->latest_bdev = latest_dev->bdev;
836 mutex_unlock(&uuid_mutex);
839 static void __free_device(struct work_struct *work)
841 struct btrfs_device *device;
843 device = container_of(work, struct btrfs_device, rcu_work);
844 rcu_string_free(device->name);
845 bio_put(device->flush_bio);
849 static void free_device(struct rcu_head *head)
851 struct btrfs_device *device;
853 device = container_of(head, struct btrfs_device, rcu);
855 INIT_WORK(&device->rcu_work, __free_device);
856 schedule_work(&device->rcu_work);
859 static void btrfs_close_bdev(struct btrfs_device *device)
861 if (device->bdev && device->writeable) {
862 sync_blockdev(device->bdev);
863 invalidate_bdev(device->bdev);
867 blkdev_put(device->bdev, device->mode);
870 static void btrfs_prepare_close_one_device(struct btrfs_device *device)
872 struct btrfs_fs_devices *fs_devices = device->fs_devices;
873 struct btrfs_device *new_device;
874 struct rcu_string *name;
877 fs_devices->open_devices--;
879 if (device->writeable &&
880 device->devid != BTRFS_DEV_REPLACE_DEVID) {
881 list_del_init(&device->dev_alloc_list);
882 fs_devices->rw_devices--;
886 fs_devices->missing_devices--;
888 new_device = btrfs_alloc_device(NULL, &device->devid,
890 BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
892 /* Safe because we are under uuid_mutex */
894 name = rcu_string_strdup(device->name->str, GFP_NOFS);
895 BUG_ON(!name); /* -ENOMEM */
896 rcu_assign_pointer(new_device->name, name);
899 list_replace_rcu(&device->dev_list, &new_device->dev_list);
900 new_device->fs_devices = device->fs_devices;
903 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
905 struct btrfs_device *device, *tmp;
906 struct list_head pending_put;
908 INIT_LIST_HEAD(&pending_put);
910 if (--fs_devices->opened > 0)
913 mutex_lock(&fs_devices->device_list_mutex);
914 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
915 btrfs_prepare_close_one_device(device);
916 list_add(&device->dev_list, &pending_put);
918 mutex_unlock(&fs_devices->device_list_mutex);
921 * btrfs_show_devname() is using the device_list_mutex,
922 * sometimes call to blkdev_put() leads vfs calling
923 * into this func. So do put outside of device_list_mutex,
926 while (!list_empty(&pending_put)) {
927 device = list_first_entry(&pending_put,
928 struct btrfs_device, dev_list);
929 list_del(&device->dev_list);
930 btrfs_close_bdev(device);
931 call_rcu(&device->rcu, free_device);
934 WARN_ON(fs_devices->open_devices);
935 WARN_ON(fs_devices->rw_devices);
936 fs_devices->opened = 0;
937 fs_devices->seeding = 0;
942 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
944 struct btrfs_fs_devices *seed_devices = NULL;
947 mutex_lock(&uuid_mutex);
948 ret = __btrfs_close_devices(fs_devices);
949 if (!fs_devices->opened) {
950 seed_devices = fs_devices->seed;
951 fs_devices->seed = NULL;
953 mutex_unlock(&uuid_mutex);
955 while (seed_devices) {
956 fs_devices = seed_devices;
957 seed_devices = fs_devices->seed;
958 __btrfs_close_devices(fs_devices);
959 free_fs_devices(fs_devices);
962 * Wait for rcu kworkers under __btrfs_close_devices
963 * to finish all blkdev_puts so device is really
964 * free when umount is done.
970 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
971 fmode_t flags, void *holder)
973 struct request_queue *q;
974 struct block_device *bdev;
975 struct list_head *head = &fs_devices->devices;
976 struct btrfs_device *device;
977 struct btrfs_device *latest_dev = NULL;
978 struct buffer_head *bh;
979 struct btrfs_super_block *disk_super;
986 list_for_each_entry(device, head, dev_list) {
992 /* Just open everything we can; ignore failures here */
993 if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
997 disk_super = (struct btrfs_super_block *)bh->b_data;
998 devid = btrfs_stack_device_id(&disk_super->dev_item);
999 if (devid != device->devid)
1002 if (memcmp(device->uuid, disk_super->dev_item.uuid,
1006 device->generation = btrfs_super_generation(disk_super);
1008 device->generation > latest_dev->generation)
1009 latest_dev = device;
1011 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
1012 device->writeable = 0;
1014 device->writeable = !bdev_read_only(bdev);
1018 q = bdev_get_queue(bdev);
1019 if (blk_queue_discard(q))
1020 device->can_discard = 1;
1021 if (!blk_queue_nonrot(q))
1022 fs_devices->rotating = 1;
1024 device->bdev = bdev;
1025 device->in_fs_metadata = 0;
1026 device->mode = flags;
1028 fs_devices->open_devices++;
1029 if (device->writeable &&
1030 device->devid != BTRFS_DEV_REPLACE_DEVID) {
1031 fs_devices->rw_devices++;
1032 list_add(&device->dev_alloc_list,
1033 &fs_devices->alloc_list);
1040 blkdev_put(bdev, flags);
1043 if (fs_devices->open_devices == 0) {
1047 fs_devices->seeding = seeding;
1048 fs_devices->opened = 1;
1049 fs_devices->latest_bdev = latest_dev->bdev;
1050 fs_devices->total_rw_bytes = 0;
1055 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1056 fmode_t flags, void *holder)
1060 mutex_lock(&uuid_mutex);
1061 if (fs_devices->opened) {
1062 fs_devices->opened++;
1065 ret = __btrfs_open_devices(fs_devices, flags, holder);
1067 mutex_unlock(&uuid_mutex);
1071 void btrfs_release_disk_super(struct page *page)
1077 int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr,
1078 struct page **page, struct btrfs_super_block **disk_super)
1083 /* make sure our super fits in the device */
1084 if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1087 /* make sure our super fits in the page */
1088 if (sizeof(**disk_super) > PAGE_SIZE)
1091 /* make sure our super doesn't straddle pages on disk */
1092 index = bytenr >> PAGE_SHIFT;
1093 if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index)
1096 /* pull in the page with our super */
1097 *page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
1100 if (IS_ERR_OR_NULL(*page))
1105 /* align our pointer to the offset of the super block */
1106 *disk_super = p + (bytenr & ~PAGE_MASK);
1108 if (btrfs_super_bytenr(*disk_super) != bytenr ||
1109 btrfs_super_magic(*disk_super) != BTRFS_MAGIC) {
1110 btrfs_release_disk_super(*page);
1114 if ((*disk_super)->label[0] &&
1115 (*disk_super)->label[BTRFS_LABEL_SIZE - 1])
1116 (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0';
1122 * Look for a btrfs signature on a device. This may be called out of the mount path
1123 * and we are not allowed to call set_blocksize during the scan. The superblock
1124 * is read via pagecache
1126 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
1127 struct btrfs_fs_devices **fs_devices_ret)
1129 struct btrfs_super_block *disk_super;
1130 struct block_device *bdev;
1139 * we would like to check all the supers, but that would make
1140 * a btrfs mount succeed after a mkfs from a different FS.
1141 * So, we need to add a special mount option to scan for
1142 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1144 bytenr = btrfs_sb_offset(0);
1145 flags |= FMODE_EXCL;
1146 mutex_lock(&uuid_mutex);
1148 bdev = blkdev_get_by_path(path, flags, holder);
1150 ret = PTR_ERR(bdev);
1154 if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super))
1155 goto error_bdev_put;
1157 devid = btrfs_stack_device_id(&disk_super->dev_item);
1158 transid = btrfs_super_generation(disk_super);
1159 total_devices = btrfs_super_num_devices(disk_super);
1161 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
1163 if (disk_super->label[0]) {
1164 pr_info("BTRFS: device label %s ", disk_super->label);
1166 pr_info("BTRFS: device fsid %pU ", disk_super->fsid);
1169 pr_cont("devid %llu transid %llu %s\n", devid, transid, path);
1172 if (!ret && fs_devices_ret)
1173 (*fs_devices_ret)->total_devices = total_devices;
1175 btrfs_release_disk_super(page);
1178 blkdev_put(bdev, flags);
1180 mutex_unlock(&uuid_mutex);
1184 /* helper to account the used device space in the range */
1185 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
1186 u64 end, u64 *length)
1188 struct btrfs_key key;
1189 struct btrfs_root *root = device->fs_info->dev_root;
1190 struct btrfs_dev_extent *dev_extent;
1191 struct btrfs_path *path;
1195 struct extent_buffer *l;
1199 if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
1202 path = btrfs_alloc_path();
1205 path->reada = READA_FORWARD;
1207 key.objectid = device->devid;
1209 key.type = BTRFS_DEV_EXTENT_KEY;
1211 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1215 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1222 slot = path->slots[0];
1223 if (slot >= btrfs_header_nritems(l)) {
1224 ret = btrfs_next_leaf(root, path);
1232 btrfs_item_key_to_cpu(l, &key, slot);
1234 if (key.objectid < device->devid)
1237 if (key.objectid > device->devid)
1240 if (key.type != BTRFS_DEV_EXTENT_KEY)
1243 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1244 extent_end = key.offset + btrfs_dev_extent_length(l,
1246 if (key.offset <= start && extent_end > end) {
1247 *length = end - start + 1;
1249 } else if (key.offset <= start && extent_end > start)
1250 *length += extent_end - start;
1251 else if (key.offset > start && extent_end <= end)
1252 *length += extent_end - key.offset;
1253 else if (key.offset > start && key.offset <= end) {
1254 *length += end - key.offset + 1;
1256 } else if (key.offset > end)
1264 btrfs_free_path(path);
1268 static int contains_pending_extent(struct btrfs_transaction *transaction,
1269 struct btrfs_device *device,
1270 u64 *start, u64 len)
1272 struct btrfs_fs_info *fs_info = device->fs_info;
1273 struct extent_map *em;
1274 struct list_head *search_list = &fs_info->pinned_chunks;
1276 u64 physical_start = *start;
1279 search_list = &transaction->pending_chunks;
1281 list_for_each_entry(em, search_list, list) {
1282 struct map_lookup *map;
1285 map = em->map_lookup;
1286 for (i = 0; i < map->num_stripes; i++) {
1289 if (map->stripes[i].dev != device)
1291 if (map->stripes[i].physical >= physical_start + len ||
1292 map->stripes[i].physical + em->orig_block_len <=
1296 * Make sure that while processing the pinned list we do
1297 * not override our *start with a lower value, because
1298 * we can have pinned chunks that fall within this
1299 * device hole and that have lower physical addresses
1300 * than the pending chunks we processed before. If we
1301 * do not take this special care we can end up getting
1302 * 2 pending chunks that start at the same physical
1303 * device offsets because the end offset of a pinned
1304 * chunk can be equal to the start offset of some
1307 end = map->stripes[i].physical + em->orig_block_len;
1314 if (search_list != &fs_info->pinned_chunks) {
1315 search_list = &fs_info->pinned_chunks;
1324 * find_free_dev_extent_start - find free space in the specified device
1325 * @device: the device which we search the free space in
1326 * @num_bytes: the size of the free space that we need
1327 * @search_start: the position from which to begin the search
1328 * @start: store the start of the free space.
1329 * @len: the size of the free space. that we find, or the size
1330 * of the max free space if we don't find suitable free space
1332 * this uses a pretty simple search, the expectation is that it is
1333 * called very infrequently and that a given device has a small number
1336 * @start is used to store the start of the free space if we find. But if we
1337 * don't find suitable free space, it will be used to store the start position
1338 * of the max free space.
1340 * @len is used to store the size of the free space that we find.
1341 * But if we don't find suitable free space, it is used to store the size of
1342 * the max free space.
1344 int find_free_dev_extent_start(struct btrfs_transaction *transaction,
1345 struct btrfs_device *device, u64 num_bytes,
1346 u64 search_start, u64 *start, u64 *len)
1348 struct btrfs_fs_info *fs_info = device->fs_info;
1349 struct btrfs_root *root = fs_info->dev_root;
1350 struct btrfs_key key;
1351 struct btrfs_dev_extent *dev_extent;
1352 struct btrfs_path *path;
1357 u64 search_end = device->total_bytes;
1360 struct extent_buffer *l;
1363 * We don't want to overwrite the superblock on the drive nor any area
1364 * used by the boot loader (grub for example), so we make sure to start
1365 * at an offset of at least 1MB.
1367 search_start = max_t(u64, search_start, SZ_1M);
1369 path = btrfs_alloc_path();
1373 max_hole_start = search_start;
1377 if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1382 path->reada = READA_FORWARD;
1383 path->search_commit_root = 1;
1384 path->skip_locking = 1;
1386 key.objectid = device->devid;
1387 key.offset = search_start;
1388 key.type = BTRFS_DEV_EXTENT_KEY;
1390 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1394 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1401 slot = path->slots[0];
1402 if (slot >= btrfs_header_nritems(l)) {
1403 ret = btrfs_next_leaf(root, path);
1411 btrfs_item_key_to_cpu(l, &key, slot);
1413 if (key.objectid < device->devid)
1416 if (key.objectid > device->devid)
1419 if (key.type != BTRFS_DEV_EXTENT_KEY)
1422 if (key.offset > search_start) {
1423 hole_size = key.offset - search_start;
1426 * Have to check before we set max_hole_start, otherwise
1427 * we could end up sending back this offset anyway.
1429 if (contains_pending_extent(transaction, device,
1432 if (key.offset >= search_start) {
1433 hole_size = key.offset - search_start;
1440 if (hole_size > max_hole_size) {
1441 max_hole_start = search_start;
1442 max_hole_size = hole_size;
1446 * If this free space is greater than which we need,
1447 * it must be the max free space that we have found
1448 * until now, so max_hole_start must point to the start
1449 * of this free space and the length of this free space
1450 * is stored in max_hole_size. Thus, we return
1451 * max_hole_start and max_hole_size and go back to the
1454 if (hole_size >= num_bytes) {
1460 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1461 extent_end = key.offset + btrfs_dev_extent_length(l,
1463 if (extent_end > search_start)
1464 search_start = extent_end;
1471 * At this point, search_start should be the end of
1472 * allocated dev extents, and when shrinking the device,
1473 * search_end may be smaller than search_start.
1475 if (search_end > search_start) {
1476 hole_size = search_end - search_start;
1478 if (contains_pending_extent(transaction, device, &search_start,
1480 btrfs_release_path(path);
1484 if (hole_size > max_hole_size) {
1485 max_hole_start = search_start;
1486 max_hole_size = hole_size;
1491 if (max_hole_size < num_bytes)
1497 btrfs_free_path(path);
1498 *start = max_hole_start;
1500 *len = max_hole_size;
1504 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1505 struct btrfs_device *device, u64 num_bytes,
1506 u64 *start, u64 *len)
1508 /* FIXME use last free of some kind */
1509 return find_free_dev_extent_start(trans->transaction, device,
1510 num_bytes, 0, start, len);
1513 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1514 struct btrfs_device *device,
1515 u64 start, u64 *dev_extent_len)
1517 struct btrfs_fs_info *fs_info = device->fs_info;
1518 struct btrfs_root *root = fs_info->dev_root;
1520 struct btrfs_path *path;
1521 struct btrfs_key key;
1522 struct btrfs_key found_key;
1523 struct extent_buffer *leaf = NULL;
1524 struct btrfs_dev_extent *extent = NULL;
1526 path = btrfs_alloc_path();
1530 key.objectid = device->devid;
1532 key.type = BTRFS_DEV_EXTENT_KEY;
1534 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1536 ret = btrfs_previous_item(root, path, key.objectid,
1537 BTRFS_DEV_EXTENT_KEY);
1540 leaf = path->nodes[0];
1541 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1542 extent = btrfs_item_ptr(leaf, path->slots[0],
1543 struct btrfs_dev_extent);
1544 BUG_ON(found_key.offset > start || found_key.offset +
1545 btrfs_dev_extent_length(leaf, extent) < start);
1547 btrfs_release_path(path);
1549 } else if (ret == 0) {
1550 leaf = path->nodes[0];
1551 extent = btrfs_item_ptr(leaf, path->slots[0],
1552 struct btrfs_dev_extent);
1554 btrfs_handle_fs_error(fs_info, ret, "Slot search failed");
1558 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1560 ret = btrfs_del_item(trans, root, path);
1562 btrfs_handle_fs_error(fs_info, ret,
1563 "Failed to remove dev extent item");
1565 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1568 btrfs_free_path(path);
1572 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1573 struct btrfs_device *device,
1574 u64 chunk_offset, u64 start, u64 num_bytes)
1577 struct btrfs_path *path;
1578 struct btrfs_fs_info *fs_info = device->fs_info;
1579 struct btrfs_root *root = fs_info->dev_root;
1580 struct btrfs_dev_extent *extent;
1581 struct extent_buffer *leaf;
1582 struct btrfs_key key;
1584 WARN_ON(!device->in_fs_metadata);
1585 WARN_ON(device->is_tgtdev_for_dev_replace);
1586 path = btrfs_alloc_path();
1590 key.objectid = device->devid;
1592 key.type = BTRFS_DEV_EXTENT_KEY;
1593 ret = btrfs_insert_empty_item(trans, root, path, &key,
1598 leaf = path->nodes[0];
1599 extent = btrfs_item_ptr(leaf, path->slots[0],
1600 struct btrfs_dev_extent);
1601 btrfs_set_dev_extent_chunk_tree(leaf, extent,
1602 BTRFS_CHUNK_TREE_OBJECTID);
1603 btrfs_set_dev_extent_chunk_objectid(leaf, extent,
1604 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
1605 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1607 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1608 btrfs_mark_buffer_dirty(leaf);
1610 btrfs_free_path(path);
1614 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1616 struct extent_map_tree *em_tree;
1617 struct extent_map *em;
1621 em_tree = &fs_info->mapping_tree.map_tree;
1622 read_lock(&em_tree->lock);
1623 n = rb_last(&em_tree->map);
1625 em = rb_entry(n, struct extent_map, rb_node);
1626 ret = em->start + em->len;
1628 read_unlock(&em_tree->lock);
1633 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1637 struct btrfs_key key;
1638 struct btrfs_key found_key;
1639 struct btrfs_path *path;
1641 path = btrfs_alloc_path();
1645 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1646 key.type = BTRFS_DEV_ITEM_KEY;
1647 key.offset = (u64)-1;
1649 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1653 BUG_ON(ret == 0); /* Corruption */
1655 ret = btrfs_previous_item(fs_info->chunk_root, path,
1656 BTRFS_DEV_ITEMS_OBJECTID,
1657 BTRFS_DEV_ITEM_KEY);
1661 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1663 *devid_ret = found_key.offset + 1;
1667 btrfs_free_path(path);
1672 * the device information is stored in the chunk root
1673 * the btrfs_device struct should be fully filled in
1675 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1676 struct btrfs_fs_info *fs_info,
1677 struct btrfs_device *device)
1679 struct btrfs_root *root = fs_info->chunk_root;
1681 struct btrfs_path *path;
1682 struct btrfs_dev_item *dev_item;
1683 struct extent_buffer *leaf;
1684 struct btrfs_key key;
1687 path = btrfs_alloc_path();
1691 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1692 key.type = BTRFS_DEV_ITEM_KEY;
1693 key.offset = device->devid;
1695 ret = btrfs_insert_empty_item(trans, root, path, &key,
1700 leaf = path->nodes[0];
1701 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1703 btrfs_set_device_id(leaf, dev_item, device->devid);
1704 btrfs_set_device_generation(leaf, dev_item, 0);
1705 btrfs_set_device_type(leaf, dev_item, device->type);
1706 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1707 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1708 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1709 btrfs_set_device_total_bytes(leaf, dev_item,
1710 btrfs_device_get_disk_total_bytes(device));
1711 btrfs_set_device_bytes_used(leaf, dev_item,
1712 btrfs_device_get_bytes_used(device));
1713 btrfs_set_device_group(leaf, dev_item, 0);
1714 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1715 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1716 btrfs_set_device_start_offset(leaf, dev_item, 0);
1718 ptr = btrfs_device_uuid(dev_item);
1719 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1720 ptr = btrfs_device_fsid(dev_item);
1721 write_extent_buffer(leaf, fs_info->fsid, ptr, BTRFS_FSID_SIZE);
1722 btrfs_mark_buffer_dirty(leaf);
1726 btrfs_free_path(path);
1731 * Function to update ctime/mtime for a given device path.
1732 * Mainly used for ctime/mtime based probe like libblkid.
1734 static void update_dev_time(const char *path_name)
1738 filp = filp_open(path_name, O_RDWR, 0);
1741 file_update_time(filp);
1742 filp_close(filp, NULL);
1745 static int btrfs_rm_dev_item(struct btrfs_fs_info *fs_info,
1746 struct btrfs_device *device)
1748 struct btrfs_root *root = fs_info->chunk_root;
1750 struct btrfs_path *path;
1751 struct btrfs_key key;
1752 struct btrfs_trans_handle *trans;
1754 path = btrfs_alloc_path();
1758 trans = btrfs_start_transaction(root, 0);
1759 if (IS_ERR(trans)) {
1760 btrfs_free_path(path);
1761 return PTR_ERR(trans);
1763 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1764 key.type = BTRFS_DEV_ITEM_KEY;
1765 key.offset = device->devid;
1767 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1771 btrfs_abort_transaction(trans, ret);
1772 btrfs_end_transaction(trans);
1776 ret = btrfs_del_item(trans, root, path);
1778 btrfs_abort_transaction(trans, ret);
1779 btrfs_end_transaction(trans);
1783 btrfs_free_path(path);
1785 ret = btrfs_commit_transaction(trans);
1790 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1791 * filesystem. It's up to the caller to adjust that number regarding eg. device
1794 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1802 seq = read_seqbegin(&fs_info->profiles_lock);
1804 all_avail = fs_info->avail_data_alloc_bits |
1805 fs_info->avail_system_alloc_bits |
1806 fs_info->avail_metadata_alloc_bits;
1807 } while (read_seqretry(&fs_info->profiles_lock, seq));
1809 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1810 if (!(all_avail & btrfs_raid_group[i]))
1813 if (num_devices < btrfs_raid_array[i].devs_min) {
1814 int ret = btrfs_raid_mindev_error[i];
1824 struct btrfs_device *btrfs_find_next_active_device(struct btrfs_fs_devices *fs_devs,
1825 struct btrfs_device *device)
1827 struct btrfs_device *next_device;
1829 list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1830 if (next_device != device &&
1831 !next_device->missing && next_device->bdev)
1839 * Helper function to check if the given device is part of s_bdev / latest_bdev
1840 * and replace it with the provided or the next active device, in the context
1841 * where this function called, there should be always be another device (or
1842 * this_dev) which is active.
1844 void btrfs_assign_next_active_device(struct btrfs_fs_info *fs_info,
1845 struct btrfs_device *device, struct btrfs_device *this_dev)
1847 struct btrfs_device *next_device;
1850 next_device = this_dev;
1852 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
1854 ASSERT(next_device);
1856 if (fs_info->sb->s_bdev &&
1857 (fs_info->sb->s_bdev == device->bdev))
1858 fs_info->sb->s_bdev = next_device->bdev;
1860 if (fs_info->fs_devices->latest_bdev == device->bdev)
1861 fs_info->fs_devices->latest_bdev = next_device->bdev;
1864 int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path,
1867 struct btrfs_device *device;
1868 struct btrfs_fs_devices *cur_devices;
1872 mutex_lock(&uuid_mutex);
1874 num_devices = fs_info->fs_devices->num_devices;
1875 btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
1876 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
1877 WARN_ON(num_devices < 1);
1880 btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
1882 ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
1886 ret = btrfs_find_device_by_devspec(fs_info, devid, device_path,
1891 if (device->is_tgtdev_for_dev_replace) {
1892 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1896 if (device->writeable && fs_info->fs_devices->rw_devices == 1) {
1897 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1901 if (device->writeable) {
1902 mutex_lock(&fs_info->chunk_mutex);
1903 list_del_init(&device->dev_alloc_list);
1904 device->fs_devices->rw_devices--;
1905 mutex_unlock(&fs_info->chunk_mutex);
1908 mutex_unlock(&uuid_mutex);
1909 ret = btrfs_shrink_device(device, 0);
1910 mutex_lock(&uuid_mutex);
1915 * TODO: the superblock still includes this device in its num_devices
1916 * counter although write_all_supers() is not locked out. This
1917 * could give a filesystem state which requires a degraded mount.
1919 ret = btrfs_rm_dev_item(fs_info, device);
1923 device->in_fs_metadata = 0;
1924 btrfs_scrub_cancel_dev(fs_info, device);
1927 * the device list mutex makes sure that we don't change
1928 * the device list while someone else is writing out all
1929 * the device supers. Whoever is writing all supers, should
1930 * lock the device list mutex before getting the number of
1931 * devices in the super block (super_copy). Conversely,
1932 * whoever updates the number of devices in the super block
1933 * (super_copy) should hold the device list mutex.
1936 cur_devices = device->fs_devices;
1937 mutex_lock(&fs_info->fs_devices->device_list_mutex);
1938 list_del_rcu(&device->dev_list);
1940 device->fs_devices->num_devices--;
1941 device->fs_devices->total_devices--;
1943 if (device->missing)
1944 device->fs_devices->missing_devices--;
1946 btrfs_assign_next_active_device(fs_info, device, NULL);
1949 device->fs_devices->open_devices--;
1950 /* remove sysfs entry */
1951 btrfs_sysfs_rm_device_link(fs_info->fs_devices, device);
1954 num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
1955 btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
1956 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1959 * at this point, the device is zero sized and detached from
1960 * the devices list. All that's left is to zero out the old
1961 * supers and free the device.
1963 if (device->writeable)
1964 btrfs_scratch_superblocks(device->bdev, device->name->str);
1966 btrfs_close_bdev(device);
1967 call_rcu(&device->rcu, free_device);
1969 if (cur_devices->open_devices == 0) {
1970 struct btrfs_fs_devices *fs_devices;
1971 fs_devices = fs_info->fs_devices;
1972 while (fs_devices) {
1973 if (fs_devices->seed == cur_devices) {
1974 fs_devices->seed = cur_devices->seed;
1977 fs_devices = fs_devices->seed;
1979 cur_devices->seed = NULL;
1980 __btrfs_close_devices(cur_devices);
1981 free_fs_devices(cur_devices);
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(NULL);
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_FSID_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_rdonly(sb) && !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 = round_down(i_size_read(bdev->bd_inode),
2392 fs_info->sectorsize);
2393 device->disk_total_bytes = device->total_bytes;
2394 device->commit_total_bytes = device->total_bytes;
2395 device->fs_info = fs_info;
2396 device->bdev = bdev;
2397 device->in_fs_metadata = 1;
2398 device->is_tgtdev_for_dev_replace = 0;
2399 device->mode = FMODE_EXCL;
2400 device->dev_stats_valid = 1;
2401 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2404 sb->s_flags &= ~MS_RDONLY;
2405 ret = btrfs_prepare_sprout(fs_info);
2406 BUG_ON(ret); /* -ENOMEM */
2409 device->fs_devices = fs_info->fs_devices;
2411 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2412 mutex_lock(&fs_info->chunk_mutex);
2413 list_add_rcu(&device->dev_list, &fs_info->fs_devices->devices);
2414 list_add(&device->dev_alloc_list,
2415 &fs_info->fs_devices->alloc_list);
2416 fs_info->fs_devices->num_devices++;
2417 fs_info->fs_devices->open_devices++;
2418 fs_info->fs_devices->rw_devices++;
2419 fs_info->fs_devices->total_devices++;
2420 fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2422 atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2424 if (!blk_queue_nonrot(q))
2425 fs_info->fs_devices->rotating = 1;
2427 tmp = btrfs_super_total_bytes(fs_info->super_copy);
2428 btrfs_set_super_total_bytes(fs_info->super_copy,
2429 round_down(tmp + device->total_bytes, fs_info->sectorsize));
2431 tmp = btrfs_super_num_devices(fs_info->super_copy);
2432 btrfs_set_super_num_devices(fs_info->super_copy, tmp + 1);
2434 /* add sysfs device entry */
2435 btrfs_sysfs_add_device_link(fs_info->fs_devices, device);
2438 * we've got more storage, clear any full flags on the space
2441 btrfs_clear_space_info_full(fs_info);
2443 mutex_unlock(&fs_info->chunk_mutex);
2444 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2447 mutex_lock(&fs_info->chunk_mutex);
2448 ret = init_first_rw_device(trans, fs_info);
2449 mutex_unlock(&fs_info->chunk_mutex);
2451 btrfs_abort_transaction(trans, ret);
2456 ret = btrfs_add_device(trans, fs_info, device);
2458 btrfs_abort_transaction(trans, ret);
2463 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2465 ret = btrfs_finish_sprout(trans, fs_info);
2467 btrfs_abort_transaction(trans, ret);
2471 /* Sprouting would change fsid of the mounted root,
2472 * so rename the fsid on the sysfs
2474 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2476 if (kobject_rename(&fs_info->fs_devices->fsid_kobj, fsid_buf))
2478 "sysfs: failed to create fsid for sprout");
2481 ret = btrfs_commit_transaction(trans);
2484 mutex_unlock(&uuid_mutex);
2485 up_write(&sb->s_umount);
2487 if (ret) /* transaction commit */
2490 ret = btrfs_relocate_sys_chunks(fs_info);
2492 btrfs_handle_fs_error(fs_info, ret,
2493 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2494 trans = btrfs_attach_transaction(root);
2495 if (IS_ERR(trans)) {
2496 if (PTR_ERR(trans) == -ENOENT)
2498 return PTR_ERR(trans);
2500 ret = btrfs_commit_transaction(trans);
2503 /* Update ctime/mtime for libblkid */
2504 update_dev_time(device_path);
2509 sb->s_flags |= MS_RDONLY;
2510 btrfs_end_transaction(trans);
2511 rcu_string_free(device->name);
2512 btrfs_sysfs_rm_device_link(fs_info->fs_devices, device);
2515 blkdev_put(bdev, FMODE_EXCL);
2517 mutex_unlock(&uuid_mutex);
2518 up_write(&sb->s_umount);
2523 int btrfs_init_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2524 const char *device_path,
2525 struct btrfs_device *srcdev,
2526 struct btrfs_device **device_out)
2528 struct request_queue *q;
2529 struct btrfs_device *device;
2530 struct block_device *bdev;
2531 struct list_head *devices;
2532 struct rcu_string *name;
2533 u64 devid = BTRFS_DEV_REPLACE_DEVID;
2537 if (fs_info->fs_devices->seeding) {
2538 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
2542 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2543 fs_info->bdev_holder);
2545 btrfs_err(fs_info, "target device %s is invalid!", device_path);
2546 return PTR_ERR(bdev);
2549 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2551 devices = &fs_info->fs_devices->devices;
2552 list_for_each_entry(device, devices, dev_list) {
2553 if (device->bdev == bdev) {
2555 "target device is in the filesystem!");
2562 if (i_size_read(bdev->bd_inode) <
2563 btrfs_device_get_total_bytes(srcdev)) {
2565 "target device is smaller than source device!");
2571 device = btrfs_alloc_device(NULL, &devid, NULL);
2572 if (IS_ERR(device)) {
2573 ret = PTR_ERR(device);
2577 name = rcu_string_strdup(device_path, GFP_KERNEL);
2583 rcu_assign_pointer(device->name, name);
2585 q = bdev_get_queue(bdev);
2586 if (blk_queue_discard(q))
2587 device->can_discard = 1;
2588 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2589 device->writeable = 1;
2590 device->generation = 0;
2591 device->io_width = fs_info->sectorsize;
2592 device->io_align = fs_info->sectorsize;
2593 device->sector_size = fs_info->sectorsize;
2594 device->total_bytes = btrfs_device_get_total_bytes(srcdev);
2595 device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
2596 device->bytes_used = btrfs_device_get_bytes_used(srcdev);
2597 ASSERT(list_empty(&srcdev->resized_list));
2598 device->commit_total_bytes = srcdev->commit_total_bytes;
2599 device->commit_bytes_used = device->bytes_used;
2600 device->fs_info = fs_info;
2601 device->bdev = bdev;
2602 device->in_fs_metadata = 1;
2603 device->is_tgtdev_for_dev_replace = 1;
2604 device->mode = FMODE_EXCL;
2605 device->dev_stats_valid = 1;
2606 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2607 device->fs_devices = fs_info->fs_devices;
2608 list_add(&device->dev_list, &fs_info->fs_devices->devices);
2609 fs_info->fs_devices->num_devices++;
2610 fs_info->fs_devices->open_devices++;
2611 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2613 *device_out = device;
2617 blkdev_put(bdev, FMODE_EXCL);
2621 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2622 struct btrfs_device *tgtdev)
2624 u32 sectorsize = fs_info->sectorsize;
2626 WARN_ON(fs_info->fs_devices->rw_devices == 0);
2627 tgtdev->io_width = sectorsize;
2628 tgtdev->io_align = sectorsize;
2629 tgtdev->sector_size = sectorsize;
2630 tgtdev->fs_info = fs_info;
2631 tgtdev->in_fs_metadata = 1;
2634 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2635 struct btrfs_device *device)
2638 struct btrfs_path *path;
2639 struct btrfs_root *root = device->fs_info->chunk_root;
2640 struct btrfs_dev_item *dev_item;
2641 struct extent_buffer *leaf;
2642 struct btrfs_key key;
2644 path = btrfs_alloc_path();
2648 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2649 key.type = BTRFS_DEV_ITEM_KEY;
2650 key.offset = device->devid;
2652 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2661 leaf = path->nodes[0];
2662 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2664 btrfs_set_device_id(leaf, dev_item, device->devid);
2665 btrfs_set_device_type(leaf, dev_item, device->type);
2666 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2667 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2668 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2669 btrfs_set_device_total_bytes(leaf, dev_item,
2670 btrfs_device_get_disk_total_bytes(device));
2671 btrfs_set_device_bytes_used(leaf, dev_item,
2672 btrfs_device_get_bytes_used(device));
2673 btrfs_mark_buffer_dirty(leaf);
2676 btrfs_free_path(path);
2680 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2681 struct btrfs_device *device, u64 new_size)
2683 struct btrfs_fs_info *fs_info = device->fs_info;
2684 struct btrfs_super_block *super_copy = fs_info->super_copy;
2685 struct btrfs_fs_devices *fs_devices;
2689 if (!device->writeable)
2692 new_size = round_down(new_size, fs_info->sectorsize);
2694 mutex_lock(&fs_info->chunk_mutex);
2695 old_total = btrfs_super_total_bytes(super_copy);
2696 diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2698 if (new_size <= device->total_bytes ||
2699 device->is_tgtdev_for_dev_replace) {
2700 mutex_unlock(&fs_info->chunk_mutex);
2704 fs_devices = fs_info->fs_devices;
2706 btrfs_set_super_total_bytes(super_copy,
2707 round_down(old_total + diff, fs_info->sectorsize));
2708 device->fs_devices->total_rw_bytes += diff;
2710 btrfs_device_set_total_bytes(device, new_size);
2711 btrfs_device_set_disk_total_bytes(device, new_size);
2712 btrfs_clear_space_info_full(device->fs_info);
2713 if (list_empty(&device->resized_list))
2714 list_add_tail(&device->resized_list,
2715 &fs_devices->resized_devices);
2716 mutex_unlock(&fs_info->chunk_mutex);
2718 return btrfs_update_device(trans, device);
2721 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2722 struct btrfs_fs_info *fs_info, u64 chunk_offset)
2724 struct btrfs_root *root = fs_info->chunk_root;
2726 struct btrfs_path *path;
2727 struct btrfs_key key;
2729 path = btrfs_alloc_path();
2733 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2734 key.offset = chunk_offset;
2735 key.type = BTRFS_CHUNK_ITEM_KEY;
2737 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2740 else if (ret > 0) { /* Logic error or corruption */
2741 btrfs_handle_fs_error(fs_info, -ENOENT,
2742 "Failed lookup while freeing chunk.");
2747 ret = btrfs_del_item(trans, root, path);
2749 btrfs_handle_fs_error(fs_info, ret,
2750 "Failed to delete chunk item.");
2752 btrfs_free_path(path);
2756 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, 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 == BTRFS_FIRST_CHUNK_TREE_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;
2839 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2841 em = get_chunk_map(fs_info, chunk_offset, 1);
2844 * This is a logic error, but we don't want to just rely on the
2845 * user having built with ASSERT enabled, so if ASSERT doesn't
2846 * do anything we still error out.
2851 map = em->map_lookup;
2852 mutex_lock(&fs_info->chunk_mutex);
2853 check_system_chunk(trans, fs_info, map->type);
2854 mutex_unlock(&fs_info->chunk_mutex);
2857 * Take the device list mutex to prevent races with the final phase of
2858 * a device replace operation that replaces the device object associated
2859 * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2861 mutex_lock(&fs_devices->device_list_mutex);
2862 for (i = 0; i < map->num_stripes; i++) {
2863 struct btrfs_device *device = map->stripes[i].dev;
2864 ret = btrfs_free_dev_extent(trans, device,
2865 map->stripes[i].physical,
2868 mutex_unlock(&fs_devices->device_list_mutex);
2869 btrfs_abort_transaction(trans, ret);
2873 if (device->bytes_used > 0) {
2874 mutex_lock(&fs_info->chunk_mutex);
2875 btrfs_device_set_bytes_used(device,
2876 device->bytes_used - dev_extent_len);
2877 atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
2878 btrfs_clear_space_info_full(fs_info);
2879 mutex_unlock(&fs_info->chunk_mutex);
2882 if (map->stripes[i].dev) {
2883 ret = btrfs_update_device(trans, map->stripes[i].dev);
2885 mutex_unlock(&fs_devices->device_list_mutex);
2886 btrfs_abort_transaction(trans, ret);
2891 mutex_unlock(&fs_devices->device_list_mutex);
2893 ret = btrfs_free_chunk(trans, fs_info, chunk_offset);
2895 btrfs_abort_transaction(trans, ret);
2899 trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
2901 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2902 ret = btrfs_del_sys_chunk(fs_info, chunk_offset);
2904 btrfs_abort_transaction(trans, ret);
2909 ret = btrfs_remove_block_group(trans, fs_info, chunk_offset, em);
2911 btrfs_abort_transaction(trans, ret);
2917 free_extent_map(em);
2921 static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2923 struct btrfs_root *root = fs_info->chunk_root;
2924 struct btrfs_trans_handle *trans;
2928 * Prevent races with automatic removal of unused block groups.
2929 * After we relocate and before we remove the chunk with offset
2930 * chunk_offset, automatic removal of the block group can kick in,
2931 * resulting in a failure when calling btrfs_remove_chunk() below.
2933 * Make sure to acquire this mutex before doing a tree search (dev
2934 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2935 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2936 * we release the path used to search the chunk/dev tree and before
2937 * the current task acquires this mutex and calls us.
2939 ASSERT(mutex_is_locked(&fs_info->delete_unused_bgs_mutex));
2941 ret = btrfs_can_relocate(fs_info, chunk_offset);
2945 /* step one, relocate all the extents inside this chunk */
2946 btrfs_scrub_pause(fs_info);
2947 ret = btrfs_relocate_block_group(fs_info, chunk_offset);
2948 btrfs_scrub_continue(fs_info);
2952 trans = btrfs_start_trans_remove_block_group(root->fs_info,
2954 if (IS_ERR(trans)) {
2955 ret = PTR_ERR(trans);
2956 btrfs_handle_fs_error(root->fs_info, ret, NULL);
2961 * step two, delete the device extents and the
2962 * chunk tree entries
2964 ret = btrfs_remove_chunk(trans, fs_info, chunk_offset);
2965 btrfs_end_transaction(trans);
2969 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
2971 struct btrfs_root *chunk_root = fs_info->chunk_root;
2972 struct btrfs_path *path;
2973 struct extent_buffer *leaf;
2974 struct btrfs_chunk *chunk;
2975 struct btrfs_key key;
2976 struct btrfs_key found_key;
2978 bool retried = false;
2982 path = btrfs_alloc_path();
2987 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2988 key.offset = (u64)-1;
2989 key.type = BTRFS_CHUNK_ITEM_KEY;
2992 mutex_lock(&fs_info->delete_unused_bgs_mutex);
2993 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2995 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
2998 BUG_ON(ret == 0); /* Corruption */
3000 ret = btrfs_previous_item(chunk_root, path, key.objectid,
3003 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3009 leaf = path->nodes[0];
3010 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3012 chunk = btrfs_item_ptr(leaf, path->slots[0],
3013 struct btrfs_chunk);
3014 chunk_type = btrfs_chunk_type(leaf, chunk);
3015 btrfs_release_path(path);
3017 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3018 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3024 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3026 if (found_key.offset == 0)
3028 key.offset = found_key.offset - 1;
3031 if (failed && !retried) {
3035 } else if (WARN_ON(failed && retried)) {
3039 btrfs_free_path(path);
3043 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3044 struct btrfs_balance_control *bctl)
3046 struct btrfs_root *root = fs_info->tree_root;
3047 struct btrfs_trans_handle *trans;
3048 struct btrfs_balance_item *item;
3049 struct btrfs_disk_balance_args disk_bargs;
3050 struct btrfs_path *path;
3051 struct extent_buffer *leaf;
3052 struct btrfs_key key;
3055 path = btrfs_alloc_path();
3059 trans = btrfs_start_transaction(root, 0);
3060 if (IS_ERR(trans)) {
3061 btrfs_free_path(path);
3062 return PTR_ERR(trans);
3065 key.objectid = BTRFS_BALANCE_OBJECTID;
3066 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3069 ret = btrfs_insert_empty_item(trans, root, path, &key,
3074 leaf = path->nodes[0];
3075 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3077 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3079 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3080 btrfs_set_balance_data(leaf, item, &disk_bargs);
3081 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3082 btrfs_set_balance_meta(leaf, item, &disk_bargs);
3083 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3084 btrfs_set_balance_sys(leaf, item, &disk_bargs);
3086 btrfs_set_balance_flags(leaf, item, bctl->flags);
3088 btrfs_mark_buffer_dirty(leaf);
3090 btrfs_free_path(path);
3091 err = btrfs_commit_transaction(trans);
3097 static int del_balance_item(struct btrfs_fs_info *fs_info)
3099 struct btrfs_root *root = fs_info->tree_root;
3100 struct btrfs_trans_handle *trans;
3101 struct btrfs_path *path;
3102 struct btrfs_key key;
3105 path = btrfs_alloc_path();
3109 trans = btrfs_start_transaction(root, 0);
3110 if (IS_ERR(trans)) {
3111 btrfs_free_path(path);
3112 return PTR_ERR(trans);
3115 key.objectid = BTRFS_BALANCE_OBJECTID;
3116 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3119 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3127 ret = btrfs_del_item(trans, root, path);
3129 btrfs_free_path(path);
3130 err = btrfs_commit_transaction(trans);
3137 * This is a heuristic used to reduce the number of chunks balanced on
3138 * resume after balance was interrupted.
3140 static void update_balance_args(struct btrfs_balance_control *bctl)
3143 * Turn on soft mode for chunk types that were being converted.
3145 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3146 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3147 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3148 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3149 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3150 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3153 * Turn on usage filter if is not already used. The idea is
3154 * that chunks that we have already balanced should be
3155 * reasonably full. Don't do it for chunks that are being
3156 * converted - that will keep us from relocating unconverted
3157 * (albeit full) chunks.
3159 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3160 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3161 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3162 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3163 bctl->data.usage = 90;
3165 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3166 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3167 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3168 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3169 bctl->sys.usage = 90;
3171 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3172 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3173 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3174 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3175 bctl->meta.usage = 90;
3180 * Should be called with both balance and volume mutexes held to
3181 * serialize other volume operations (add_dev/rm_dev/resize) with
3182 * restriper. Same goes for unset_balance_control.
3184 static void set_balance_control(struct btrfs_balance_control *bctl)
3186 struct btrfs_fs_info *fs_info = bctl->fs_info;
3188 BUG_ON(fs_info->balance_ctl);
3190 spin_lock(&fs_info->balance_lock);
3191 fs_info->balance_ctl = bctl;
3192 spin_unlock(&fs_info->balance_lock);
3195 static void unset_balance_control(struct btrfs_fs_info *fs_info)
3197 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3199 BUG_ON(!fs_info->balance_ctl);
3201 spin_lock(&fs_info->balance_lock);
3202 fs_info->balance_ctl = NULL;
3203 spin_unlock(&fs_info->balance_lock);
3209 * Balance filters. Return 1 if chunk should be filtered out
3210 * (should not be balanced).
3212 static int chunk_profiles_filter(u64 chunk_type,
3213 struct btrfs_balance_args *bargs)
3215 chunk_type = chunk_to_extended(chunk_type) &
3216 BTRFS_EXTENDED_PROFILE_MASK;
3218 if (bargs->profiles & chunk_type)
3224 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3225 struct btrfs_balance_args *bargs)
3227 struct btrfs_block_group_cache *cache;
3229 u64 user_thresh_min;
3230 u64 user_thresh_max;
3233 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3234 chunk_used = btrfs_block_group_used(&cache->item);
3236 if (bargs->usage_min == 0)
3237 user_thresh_min = 0;
3239 user_thresh_min = div_factor_fine(cache->key.offset,
3242 if (bargs->usage_max == 0)
3243 user_thresh_max = 1;
3244 else if (bargs->usage_max > 100)
3245 user_thresh_max = cache->key.offset;
3247 user_thresh_max = div_factor_fine(cache->key.offset,
3250 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3253 btrfs_put_block_group(cache);
3257 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3258 u64 chunk_offset, struct btrfs_balance_args *bargs)
3260 struct btrfs_block_group_cache *cache;
3261 u64 chunk_used, user_thresh;
3264 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3265 chunk_used = btrfs_block_group_used(&cache->item);
3267 if (bargs->usage_min == 0)
3269 else if (bargs->usage > 100)
3270 user_thresh = cache->key.offset;
3272 user_thresh = div_factor_fine(cache->key.offset,
3275 if (chunk_used < user_thresh)
3278 btrfs_put_block_group(cache);
3282 static int chunk_devid_filter(struct extent_buffer *leaf,
3283 struct btrfs_chunk *chunk,
3284 struct btrfs_balance_args *bargs)
3286 struct btrfs_stripe *stripe;
3287 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3290 for (i = 0; i < num_stripes; i++) {
3291 stripe = btrfs_stripe_nr(chunk, i);
3292 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3299 /* [pstart, pend) */
3300 static int chunk_drange_filter(struct extent_buffer *leaf,
3301 struct btrfs_chunk *chunk,
3302 struct btrfs_balance_args *bargs)
3304 struct btrfs_stripe *stripe;
3305 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3311 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3314 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3315 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3316 factor = num_stripes / 2;
3317 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3318 factor = num_stripes - 1;
3319 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3320 factor = num_stripes - 2;
3322 factor = num_stripes;
3325 for (i = 0; i < num_stripes; i++) {
3326 stripe = btrfs_stripe_nr(chunk, i);
3327 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3330 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3331 stripe_length = btrfs_chunk_length(leaf, chunk);
3332 stripe_length = div_u64(stripe_length, factor);
3334 if (stripe_offset < bargs->pend &&
3335 stripe_offset + stripe_length > bargs->pstart)
3342 /* [vstart, vend) */
3343 static int chunk_vrange_filter(struct extent_buffer *leaf,
3344 struct btrfs_chunk *chunk,
3346 struct btrfs_balance_args *bargs)
3348 if (chunk_offset < bargs->vend &&
3349 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3350 /* at least part of the chunk is inside this vrange */
3356 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3357 struct btrfs_chunk *chunk,
3358 struct btrfs_balance_args *bargs)
3360 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3362 if (bargs->stripes_min <= num_stripes
3363 && num_stripes <= bargs->stripes_max)
3369 static int chunk_soft_convert_filter(u64 chunk_type,
3370 struct btrfs_balance_args *bargs)
3372 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3375 chunk_type = chunk_to_extended(chunk_type) &
3376 BTRFS_EXTENDED_PROFILE_MASK;
3378 if (bargs->target == chunk_type)
3384 static int should_balance_chunk(struct btrfs_fs_info *fs_info,
3385 struct extent_buffer *leaf,
3386 struct btrfs_chunk *chunk, u64 chunk_offset)
3388 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3389 struct btrfs_balance_args *bargs = NULL;
3390 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3393 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3394 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3398 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3399 bargs = &bctl->data;
3400 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3402 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3403 bargs = &bctl->meta;
3405 /* profiles filter */
3406 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3407 chunk_profiles_filter(chunk_type, bargs)) {
3412 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3413 chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3415 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3416 chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3421 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3422 chunk_devid_filter(leaf, chunk, bargs)) {
3426 /* drange filter, makes sense only with devid filter */
3427 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3428 chunk_drange_filter(leaf, chunk, bargs)) {
3433 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3434 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3438 /* stripes filter */
3439 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3440 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3444 /* soft profile changing mode */
3445 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3446 chunk_soft_convert_filter(chunk_type, bargs)) {
3451 * limited by count, must be the last filter
3453 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3454 if (bargs->limit == 0)
3458 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3460 * Same logic as the 'limit' filter; the minimum cannot be
3461 * determined here because we do not have the global information
3462 * about the count of all chunks that satisfy the filters.
3464 if (bargs->limit_max == 0)
3473 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3475 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3476 struct btrfs_root *chunk_root = fs_info->chunk_root;
3477 struct btrfs_root *dev_root = fs_info->dev_root;
3478 struct list_head *devices;
3479 struct btrfs_device *device;
3483 struct btrfs_chunk *chunk;
3484 struct btrfs_path *path = NULL;
3485 struct btrfs_key key;
3486 struct btrfs_key found_key;
3487 struct btrfs_trans_handle *trans;
3488 struct extent_buffer *leaf;
3491 int enospc_errors = 0;
3492 bool counting = true;
3493 /* The single value limit and min/max limits use the same bytes in the */
3494 u64 limit_data = bctl->data.limit;
3495 u64 limit_meta = bctl->meta.limit;
3496 u64 limit_sys = bctl->sys.limit;
3500 int chunk_reserved = 0;
3503 /* step one make some room on all the devices */
3504 devices = &fs_info->fs_devices->devices;
3505 list_for_each_entry(device, devices, dev_list) {
3506 old_size = btrfs_device_get_total_bytes(device);
3507 size_to_free = div_factor(old_size, 1);
3508 size_to_free = min_t(u64, size_to_free, SZ_1M);
3509 if (!device->writeable ||
3510 btrfs_device_get_total_bytes(device) -
3511 btrfs_device_get_bytes_used(device) > size_to_free ||
3512 device->is_tgtdev_for_dev_replace)
3515 ret = btrfs_shrink_device(device, old_size - size_to_free);
3519 /* btrfs_shrink_device never returns ret > 0 */
3524 trans = btrfs_start_transaction(dev_root, 0);
3525 if (IS_ERR(trans)) {
3526 ret = PTR_ERR(trans);
3527 btrfs_info_in_rcu(fs_info,
3528 "resize: unable to start transaction after shrinking device %s (error %d), old size %llu, new size %llu",
3529 rcu_str_deref(device->name), ret,
3530 old_size, old_size - size_to_free);
3534 ret = btrfs_grow_device(trans, device, old_size);
3536 btrfs_end_transaction(trans);
3537 /* btrfs_grow_device never returns ret > 0 */
3539 btrfs_info_in_rcu(fs_info,
3540 "resize: unable to grow device after shrinking device %s (error %d), old size %llu, new size %llu",
3541 rcu_str_deref(device->name), ret,
3542 old_size, old_size - size_to_free);
3546 btrfs_end_transaction(trans);
3549 /* step two, relocate all the chunks */
3550 path = btrfs_alloc_path();
3556 /* zero out stat counters */
3557 spin_lock(&fs_info->balance_lock);
3558 memset(&bctl->stat, 0, sizeof(bctl->stat));
3559 spin_unlock(&fs_info->balance_lock);
3563 * The single value limit and min/max limits use the same bytes
3566 bctl->data.limit = limit_data;
3567 bctl->meta.limit = limit_meta;
3568 bctl->sys.limit = limit_sys;
3570 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3571 key.offset = (u64)-1;
3572 key.type = BTRFS_CHUNK_ITEM_KEY;
3575 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3576 atomic_read(&fs_info->balance_cancel_req)) {
3581 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3582 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3584 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3589 * this shouldn't happen, it means the last relocate
3593 BUG(); /* FIXME break ? */
3595 ret = btrfs_previous_item(chunk_root, path, 0,
3596 BTRFS_CHUNK_ITEM_KEY);
3598 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3603 leaf = path->nodes[0];
3604 slot = path->slots[0];
3605 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3607 if (found_key.objectid != key.objectid) {
3608 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3612 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3613 chunk_type = btrfs_chunk_type(leaf, chunk);
3616 spin_lock(&fs_info->balance_lock);
3617 bctl->stat.considered++;
3618 spin_unlock(&fs_info->balance_lock);
3621 ret = should_balance_chunk(fs_info, leaf, chunk,
3624 btrfs_release_path(path);
3626 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3631 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3632 spin_lock(&fs_info->balance_lock);
3633 bctl->stat.expected++;
3634 spin_unlock(&fs_info->balance_lock);
3636 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3638 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3640 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3647 * Apply limit_min filter, no need to check if the LIMITS
3648 * filter is used, limit_min is 0 by default
3650 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3651 count_data < bctl->data.limit_min)
3652 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3653 count_meta < bctl->meta.limit_min)
3654 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3655 count_sys < bctl->sys.limit_min)) {
3656 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3660 ASSERT(fs_info->data_sinfo);
3661 spin_lock(&fs_info->data_sinfo->lock);
3662 bytes_used = fs_info->data_sinfo->bytes_used;
3663 spin_unlock(&fs_info->data_sinfo->lock);
3665 if ((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3666 !chunk_reserved && !bytes_used) {
3667 trans = btrfs_start_transaction(chunk_root, 0);
3668 if (IS_ERR(trans)) {
3669 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3670 ret = PTR_ERR(trans);
3674 ret = btrfs_force_chunk_alloc(trans, fs_info,
3675 BTRFS_BLOCK_GROUP_DATA);
3676 btrfs_end_transaction(trans);
3678 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3684 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3685 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3686 if (ret && ret != -ENOSPC)
3688 if (ret == -ENOSPC) {
3691 spin_lock(&fs_info->balance_lock);
3692 bctl->stat.completed++;
3693 spin_unlock(&fs_info->balance_lock);
3696 if (found_key.offset == 0)
3698 key.offset = found_key.offset - 1;
3702 btrfs_release_path(path);
3707 btrfs_free_path(path);
3708 if (enospc_errors) {
3709 btrfs_info(fs_info, "%d enospc errors during balance",
3719 * alloc_profile_is_valid - see if a given profile is valid and reduced
3720 * @flags: profile to validate
3721 * @extended: if true @flags is treated as an extended profile
3723 static int alloc_profile_is_valid(u64 flags, int extended)
3725 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3726 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3728 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3730 /* 1) check that all other bits are zeroed */
3734 /* 2) see if profile is reduced */
3736 return !extended; /* "0" is valid for usual profiles */
3738 /* true if exactly one bit set */
3739 return (flags & (flags - 1)) == 0;
3742 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3744 /* cancel requested || normal exit path */
3745 return atomic_read(&fs_info->balance_cancel_req) ||
3746 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3747 atomic_read(&fs_info->balance_cancel_req) == 0);
3750 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3754 unset_balance_control(fs_info);
3755 ret = del_balance_item(fs_info);
3757 btrfs_handle_fs_error(fs_info, ret, NULL);
3759 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3762 /* Non-zero return value signifies invalidity */
3763 static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg,
3766 return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3767 (!alloc_profile_is_valid(bctl_arg->target, 1) ||
3768 (bctl_arg->target & ~allowed)));
3772 * Should be called with both balance and volume mutexes held
3774 int btrfs_balance(struct btrfs_balance_control *bctl,
3775 struct btrfs_ioctl_balance_args *bargs)
3777 struct btrfs_fs_info *fs_info = bctl->fs_info;
3778 u64 meta_target, data_target;
3785 if (btrfs_fs_closing(fs_info) ||
3786 atomic_read(&fs_info->balance_pause_req) ||
3787 atomic_read(&fs_info->balance_cancel_req)) {
3792 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3793 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3797 * In case of mixed groups both data and meta should be picked,
3798 * and identical options should be given for both of them.
3800 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3801 if (mixed && (bctl->flags & allowed)) {
3802 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3803 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3804 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3806 "with mixed groups data and metadata balance options must be the same");
3812 num_devices = fs_info->fs_devices->num_devices;
3813 btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
3814 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3815 BUG_ON(num_devices < 1);
3818 btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
3819 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE | BTRFS_BLOCK_GROUP_DUP;
3820 if (num_devices > 1)
3821 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3822 if (num_devices > 2)
3823 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3824 if (num_devices > 3)
3825 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3826 BTRFS_BLOCK_GROUP_RAID6);
3827 if (validate_convert_profile(&bctl->data, allowed)) {
3829 "unable to start balance with target data profile %llu",
3834 if (validate_convert_profile(&bctl->meta, allowed)) {
3836 "unable to start balance with target metadata profile %llu",
3841 if (validate_convert_profile(&bctl->sys, allowed)) {
3843 "unable to start balance with target system profile %llu",
3849 /* allow to reduce meta or sys integrity only if force set */
3850 allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3851 BTRFS_BLOCK_GROUP_RAID10 |
3852 BTRFS_BLOCK_GROUP_RAID5 |
3853 BTRFS_BLOCK_GROUP_RAID6;
3855 seq = read_seqbegin(&fs_info->profiles_lock);
3857 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3858 (fs_info->avail_system_alloc_bits & allowed) &&
3859 !(bctl->sys.target & allowed)) ||
3860 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3861 (fs_info->avail_metadata_alloc_bits & allowed) &&
3862 !(bctl->meta.target & allowed))) {
3863 if (bctl->flags & BTRFS_BALANCE_FORCE) {
3865 "force reducing metadata integrity");
3868 "balance will reduce metadata integrity, use force if you want this");
3873 } while (read_seqretry(&fs_info->profiles_lock, seq));
3875 /* if we're not converting, the target field is uninitialized */
3876 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
3877 bctl->meta.target : fs_info->avail_metadata_alloc_bits;
3878 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
3879 bctl->data.target : fs_info->avail_data_alloc_bits;
3880 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <
3881 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {
3883 "metadata profile 0x%llx has lower redundancy than data profile 0x%llx",
3884 meta_target, data_target);
3887 ret = insert_balance_item(fs_info, bctl);
3888 if (ret && ret != -EEXIST)
3891 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3892 BUG_ON(ret == -EEXIST);
3893 set_balance_control(bctl);
3895 BUG_ON(ret != -EEXIST);
3896 spin_lock(&fs_info->balance_lock);
3897 update_balance_args(bctl);
3898 spin_unlock(&fs_info->balance_lock);
3901 atomic_inc(&fs_info->balance_running);
3902 mutex_unlock(&fs_info->balance_mutex);
3904 ret = __btrfs_balance(fs_info);
3906 mutex_lock(&fs_info->balance_mutex);
3907 atomic_dec(&fs_info->balance_running);
3910 memset(bargs, 0, sizeof(*bargs));
3911 update_ioctl_balance_args(fs_info, 0, bargs);
3914 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3915 balance_need_close(fs_info)) {
3916 __cancel_balance(fs_info);
3919 wake_up(&fs_info->balance_wait_q);
3923 if (bctl->flags & BTRFS_BALANCE_RESUME)
3924 __cancel_balance(fs_info);
3927 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3932 static int balance_kthread(void *data)
3934 struct btrfs_fs_info *fs_info = data;
3937 mutex_lock(&fs_info->volume_mutex);
3938 mutex_lock(&fs_info->balance_mutex);
3940 if (fs_info->balance_ctl) {
3941 btrfs_info(fs_info, "continuing balance");
3942 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3945 mutex_unlock(&fs_info->balance_mutex);
3946 mutex_unlock(&fs_info->volume_mutex);
3951 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3953 struct task_struct *tsk;
3955 spin_lock(&fs_info->balance_lock);
3956 if (!fs_info->balance_ctl) {
3957 spin_unlock(&fs_info->balance_lock);
3960 spin_unlock(&fs_info->balance_lock);
3962 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
3963 btrfs_info(fs_info, "force skipping balance");
3968 * A ro->rw remount sequence should continue with the paused balance
3969 * regardless of who pauses it, system or the user as of now, so set
3972 spin_lock(&fs_info->balance_lock);
3973 fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME;
3974 spin_unlock(&fs_info->balance_lock);
3976 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3977 return PTR_ERR_OR_ZERO(tsk);
3980 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3982 struct btrfs_balance_control *bctl;
3983 struct btrfs_balance_item *item;
3984 struct btrfs_disk_balance_args disk_bargs;
3985 struct btrfs_path *path;
3986 struct extent_buffer *leaf;
3987 struct btrfs_key key;
3990 path = btrfs_alloc_path();
3994 key.objectid = BTRFS_BALANCE_OBJECTID;
3995 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3998 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4001 if (ret > 0) { /* ret = -ENOENT; */
4006 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4012 leaf = path->nodes[0];
4013 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
4015 bctl->fs_info = fs_info;
4016 bctl->flags = btrfs_balance_flags(leaf, item);
4017 bctl->flags |= BTRFS_BALANCE_RESUME;
4019 btrfs_balance_data(leaf, item, &disk_bargs);
4020 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4021 btrfs_balance_meta(leaf, item, &disk_bargs);
4022 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4023 btrfs_balance_sys(leaf, item, &disk_bargs);
4024 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4026 WARN_ON(test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags));
4028 mutex_lock(&fs_info->volume_mutex);
4029 mutex_lock(&fs_info->balance_mutex);
4031 set_balance_control(bctl);
4033 mutex_unlock(&fs_info->balance_mutex);
4034 mutex_unlock(&fs_info->volume_mutex);
4036 btrfs_free_path(path);
4040 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4044 mutex_lock(&fs_info->balance_mutex);
4045 if (!fs_info->balance_ctl) {
4046 mutex_unlock(&fs_info->balance_mutex);
4050 if (atomic_read(&fs_info->balance_running)) {
4051 atomic_inc(&fs_info->balance_pause_req);
4052 mutex_unlock(&fs_info->balance_mutex);
4054 wait_event(fs_info->balance_wait_q,
4055 atomic_read(&fs_info->balance_running) == 0);
4057 mutex_lock(&fs_info->balance_mutex);
4058 /* we are good with balance_ctl ripped off from under us */
4059 BUG_ON(atomic_read(&fs_info->balance_running));
4060 atomic_dec(&fs_info->balance_pause_req);
4065 mutex_unlock(&fs_info->balance_mutex);
4069 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4071 if (sb_rdonly(fs_info->sb))
4074 mutex_lock(&fs_info->balance_mutex);
4075 if (!fs_info->balance_ctl) {
4076 mutex_unlock(&fs_info->balance_mutex);
4080 atomic_inc(&fs_info->balance_cancel_req);
4082 * if we are running just wait and return, balance item is
4083 * deleted in btrfs_balance in this case
4085 if (atomic_read(&fs_info->balance_running)) {
4086 mutex_unlock(&fs_info->balance_mutex);
4087 wait_event(fs_info->balance_wait_q,
4088 atomic_read(&fs_info->balance_running) == 0);
4089 mutex_lock(&fs_info->balance_mutex);
4091 /* __cancel_balance needs volume_mutex */
4092 mutex_unlock(&fs_info->balance_mutex);
4093 mutex_lock(&fs_info->volume_mutex);
4094 mutex_lock(&fs_info->balance_mutex);
4096 if (fs_info->balance_ctl)
4097 __cancel_balance(fs_info);
4099 mutex_unlock(&fs_info->volume_mutex);
4102 BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
4103 atomic_dec(&fs_info->balance_cancel_req);
4104 mutex_unlock(&fs_info->balance_mutex);
4108 static int btrfs_uuid_scan_kthread(void *data)
4110 struct btrfs_fs_info *fs_info = data;
4111 struct btrfs_root *root = fs_info->tree_root;
4112 struct btrfs_key key;
4113 struct btrfs_path *path = NULL;
4115 struct extent_buffer *eb;
4117 struct btrfs_root_item root_item;
4119 struct btrfs_trans_handle *trans = NULL;
4121 path = btrfs_alloc_path();
4128 key.type = BTRFS_ROOT_ITEM_KEY;
4132 ret = btrfs_search_forward(root, &key, path, 0);
4139 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4140 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4141 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4142 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4145 eb = path->nodes[0];
4146 slot = path->slots[0];
4147 item_size = btrfs_item_size_nr(eb, slot);
4148 if (item_size < sizeof(root_item))
4151 read_extent_buffer(eb, &root_item,
4152 btrfs_item_ptr_offset(eb, slot),
4153 (int)sizeof(root_item));
4154 if (btrfs_root_refs(&root_item) == 0)
4157 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4158 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4162 btrfs_release_path(path);
4164 * 1 - subvol uuid item
4165 * 1 - received_subvol uuid item
4167 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4168 if (IS_ERR(trans)) {
4169 ret = PTR_ERR(trans);
4177 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4178 ret = btrfs_uuid_tree_add(trans, fs_info,
4180 BTRFS_UUID_KEY_SUBVOL,
4183 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4189 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4190 ret = btrfs_uuid_tree_add(trans, fs_info,
4191 root_item.received_uuid,
4192 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4195 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4203 ret = btrfs_end_transaction(trans);
4209 btrfs_release_path(path);
4210 if (key.offset < (u64)-1) {
4212 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4214 key.type = BTRFS_ROOT_ITEM_KEY;
4215 } else if (key.objectid < (u64)-1) {
4217 key.type = BTRFS_ROOT_ITEM_KEY;
4226 btrfs_free_path(path);
4227 if (trans && !IS_ERR(trans))
4228 btrfs_end_transaction(trans);
4230 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4232 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4233 up(&fs_info->uuid_tree_rescan_sem);
4238 * Callback for btrfs_uuid_tree_iterate().
4240 * 0 check succeeded, the entry is not outdated.
4241 * < 0 if an error occurred.
4242 * > 0 if the check failed, which means the caller shall remove the entry.
4244 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
4245 u8 *uuid, u8 type, u64 subid)
4247 struct btrfs_key key;
4249 struct btrfs_root *subvol_root;
4251 if (type != BTRFS_UUID_KEY_SUBVOL &&
4252 type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
4255 key.objectid = subid;
4256 key.type = BTRFS_ROOT_ITEM_KEY;
4257 key.offset = (u64)-1;
4258 subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
4259 if (IS_ERR(subvol_root)) {
4260 ret = PTR_ERR(subvol_root);
4267 case BTRFS_UUID_KEY_SUBVOL:
4268 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
4271 case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
4272 if (memcmp(uuid, subvol_root->root_item.received_uuid,
4282 static int btrfs_uuid_rescan_kthread(void *data)
4284 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4288 * 1st step is to iterate through the existing UUID tree and
4289 * to delete all entries that contain outdated data.
4290 * 2nd step is to add all missing entries to the UUID tree.
4292 ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4294 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4295 up(&fs_info->uuid_tree_rescan_sem);
4298 return btrfs_uuid_scan_kthread(data);
4301 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4303 struct btrfs_trans_handle *trans;
4304 struct btrfs_root *tree_root = fs_info->tree_root;
4305 struct btrfs_root *uuid_root;
4306 struct task_struct *task;
4313 trans = btrfs_start_transaction(tree_root, 2);
4315 return PTR_ERR(trans);
4317 uuid_root = btrfs_create_tree(trans, fs_info,
4318 BTRFS_UUID_TREE_OBJECTID);
4319 if (IS_ERR(uuid_root)) {
4320 ret = PTR_ERR(uuid_root);
4321 btrfs_abort_transaction(trans, ret);
4322 btrfs_end_transaction(trans);
4326 fs_info->uuid_root = uuid_root;
4328 ret = btrfs_commit_transaction(trans);
4332 down(&fs_info->uuid_tree_rescan_sem);
4333 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4335 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4336 btrfs_warn(fs_info, "failed to start uuid_scan task");
4337 up(&fs_info->uuid_tree_rescan_sem);
4338 return PTR_ERR(task);
4344 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4346 struct task_struct *task;
4348 down(&fs_info->uuid_tree_rescan_sem);
4349 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4351 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4352 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4353 up(&fs_info->uuid_tree_rescan_sem);
4354 return PTR_ERR(task);
4361 * shrinking a device means finding all of the device extents past
4362 * the new size, and then following the back refs to the chunks.
4363 * The chunk relocation code actually frees the device extent
4365 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4367 struct btrfs_fs_info *fs_info = device->fs_info;
4368 struct btrfs_root *root = fs_info->dev_root;
4369 struct btrfs_trans_handle *trans;
4370 struct btrfs_dev_extent *dev_extent = NULL;
4371 struct btrfs_path *path;
4377 bool retried = false;
4378 bool checked_pending_chunks = false;
4379 struct extent_buffer *l;
4380 struct btrfs_key key;
4381 struct btrfs_super_block *super_copy = fs_info->super_copy;
4382 u64 old_total = btrfs_super_total_bytes(super_copy);
4383 u64 old_size = btrfs_device_get_total_bytes(device);
4386 new_size = round_down(new_size, fs_info->sectorsize);
4387 diff = round_down(old_size - new_size, fs_info->sectorsize);
4389 if (device->is_tgtdev_for_dev_replace)
4392 path = btrfs_alloc_path();
4396 path->reada = READA_FORWARD;
4398 mutex_lock(&fs_info->chunk_mutex);
4400 btrfs_device_set_total_bytes(device, new_size);
4401 if (device->writeable) {
4402 device->fs_devices->total_rw_bytes -= diff;
4403 atomic64_sub(diff, &fs_info->free_chunk_space);
4405 mutex_unlock(&fs_info->chunk_mutex);
4408 key.objectid = device->devid;
4409 key.offset = (u64)-1;
4410 key.type = BTRFS_DEV_EXTENT_KEY;
4413 mutex_lock(&fs_info->delete_unused_bgs_mutex);
4414 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4416 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4420 ret = btrfs_previous_item(root, path, 0, key.type);
4422 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4427 btrfs_release_path(path);
4432 slot = path->slots[0];
4433 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4435 if (key.objectid != device->devid) {
4436 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4437 btrfs_release_path(path);
4441 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4442 length = btrfs_dev_extent_length(l, dev_extent);
4444 if (key.offset + length <= new_size) {
4445 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4446 btrfs_release_path(path);
4450 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4451 btrfs_release_path(path);
4453 ret = btrfs_relocate_chunk(fs_info, chunk_offset);
4454 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4455 if (ret && ret != -ENOSPC)
4459 } while (key.offset-- > 0);
4461 if (failed && !retried) {
4465 } else if (failed && retried) {
4470 /* Shrinking succeeded, else we would be at "done". */
4471 trans = btrfs_start_transaction(root, 0);
4472 if (IS_ERR(trans)) {
4473 ret = PTR_ERR(trans);
4477 mutex_lock(&fs_info->chunk_mutex);
4480 * We checked in the above loop all device extents that were already in
4481 * the device tree. However before we have updated the device's
4482 * total_bytes to the new size, we might have had chunk allocations that
4483 * have not complete yet (new block groups attached to transaction
4484 * handles), and therefore their device extents were not yet in the
4485 * device tree and we missed them in the loop above. So if we have any
4486 * pending chunk using a device extent that overlaps the device range
4487 * that we can not use anymore, commit the current transaction and
4488 * repeat the search on the device tree - this way we guarantee we will
4489 * not have chunks using device extents that end beyond 'new_size'.
4491 if (!checked_pending_chunks) {
4492 u64 start = new_size;
4493 u64 len = old_size - new_size;
4495 if (contains_pending_extent(trans->transaction, device,
4497 mutex_unlock(&fs_info->chunk_mutex);
4498 checked_pending_chunks = true;
4501 ret = btrfs_commit_transaction(trans);
4508 btrfs_device_set_disk_total_bytes(device, new_size);
4509 if (list_empty(&device->resized_list))
4510 list_add_tail(&device->resized_list,
4511 &fs_info->fs_devices->resized_devices);
4513 WARN_ON(diff > old_total);
4514 btrfs_set_super_total_bytes(super_copy,
4515 round_down(old_total - diff, fs_info->sectorsize));
4516 mutex_unlock(&fs_info->chunk_mutex);
4518 /* Now btrfs_update_device() will change the on-disk size. */
4519 ret = btrfs_update_device(trans, device);
4521 btrfs_abort_transaction(trans, ret);
4522 btrfs_end_transaction(trans);
4524 ret = btrfs_commit_transaction(trans);
4527 btrfs_free_path(path);
4529 mutex_lock(&fs_info->chunk_mutex);
4530 btrfs_device_set_total_bytes(device, old_size);
4531 if (device->writeable)
4532 device->fs_devices->total_rw_bytes += diff;
4533 atomic64_add(diff, &fs_info->free_chunk_space);
4534 mutex_unlock(&fs_info->chunk_mutex);
4539 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
4540 struct btrfs_key *key,
4541 struct btrfs_chunk *chunk, int item_size)
4543 struct btrfs_super_block *super_copy = fs_info->super_copy;
4544 struct btrfs_disk_key disk_key;
4548 mutex_lock(&fs_info->chunk_mutex);
4549 array_size = btrfs_super_sys_array_size(super_copy);
4550 if (array_size + item_size + sizeof(disk_key)
4551 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4552 mutex_unlock(&fs_info->chunk_mutex);
4556 ptr = super_copy->sys_chunk_array + array_size;
4557 btrfs_cpu_key_to_disk(&disk_key, key);
4558 memcpy(ptr, &disk_key, sizeof(disk_key));
4559 ptr += sizeof(disk_key);
4560 memcpy(ptr, chunk, item_size);
4561 item_size += sizeof(disk_key);
4562 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4563 mutex_unlock(&fs_info->chunk_mutex);
4569 * sort the devices in descending order by max_avail, total_avail
4571 static int btrfs_cmp_device_info(const void *a, const void *b)
4573 const struct btrfs_device_info *di_a = a;
4574 const struct btrfs_device_info *di_b = b;
4576 if (di_a->max_avail > di_b->max_avail)
4578 if (di_a->max_avail < di_b->max_avail)
4580 if (di_a->total_avail > di_b->total_avail)
4582 if (di_a->total_avail < di_b->total_avail)
4587 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4589 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4592 btrfs_set_fs_incompat(info, RAID56);
4595 #define BTRFS_MAX_DEVS(r) ((BTRFS_MAX_ITEM_SIZE(r->fs_info) \
4596 - sizeof(struct btrfs_chunk)) \
4597 / sizeof(struct btrfs_stripe) + 1)
4599 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
4600 - 2 * sizeof(struct btrfs_disk_key) \
4601 - 2 * sizeof(struct btrfs_chunk)) \
4602 / sizeof(struct btrfs_stripe) + 1)
4604 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4605 u64 start, u64 type)
4607 struct btrfs_fs_info *info = trans->fs_info;
4608 struct btrfs_fs_devices *fs_devices = info->fs_devices;
4609 struct btrfs_device *device;
4610 struct map_lookup *map = NULL;
4611 struct extent_map_tree *em_tree;
4612 struct extent_map *em;
4613 struct btrfs_device_info *devices_info = NULL;
4615 int num_stripes; /* total number of stripes to allocate */
4616 int data_stripes; /* number of stripes that count for
4618 int sub_stripes; /* sub_stripes info for map */
4619 int dev_stripes; /* stripes per dev */
4620 int devs_max; /* max devs to use */
4621 int devs_min; /* min devs needed */
4622 int devs_increment; /* ndevs has to be a multiple of this */
4623 int ncopies; /* how many copies to data has */
4625 u64 max_stripe_size;
4634 BUG_ON(!alloc_profile_is_valid(type, 0));
4636 if (list_empty(&fs_devices->alloc_list))
4639 index = __get_raid_index(type);
4641 sub_stripes = btrfs_raid_array[index].sub_stripes;
4642 dev_stripes = btrfs_raid_array[index].dev_stripes;
4643 devs_max = btrfs_raid_array[index].devs_max;
4644 devs_min = btrfs_raid_array[index].devs_min;
4645 devs_increment = btrfs_raid_array[index].devs_increment;
4646 ncopies = btrfs_raid_array[index].ncopies;
4648 if (type & BTRFS_BLOCK_GROUP_DATA) {
4649 max_stripe_size = SZ_1G;
4650 max_chunk_size = 10 * max_stripe_size;
4652 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4653 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4654 /* for larger filesystems, use larger metadata chunks */
4655 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
4656 max_stripe_size = SZ_1G;
4658 max_stripe_size = SZ_256M;
4659 max_chunk_size = max_stripe_size;
4661 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4662 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4663 max_stripe_size = SZ_32M;
4664 max_chunk_size = 2 * max_stripe_size;
4666 devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4668 btrfs_err(info, "invalid chunk type 0x%llx requested",
4673 /* we don't want a chunk larger than 10% of writeable space */
4674 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4677 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4683 * in the first pass through the devices list, we gather information
4684 * about the available holes on each device.
4687 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
4691 if (!device->writeable) {
4693 "BTRFS: read-only device in alloc_list\n");
4697 if (!device->in_fs_metadata ||
4698 device->is_tgtdev_for_dev_replace)
4701 if (device->total_bytes > device->bytes_used)
4702 total_avail = device->total_bytes - device->bytes_used;
4706 /* If there is no space on this device, skip it. */
4707 if (total_avail == 0)
4710 ret = find_free_dev_extent(trans, device,
4711 max_stripe_size * dev_stripes,
4712 &dev_offset, &max_avail);
4713 if (ret && ret != -ENOSPC)
4717 max_avail = max_stripe_size * dev_stripes;
4719 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4722 if (ndevs == fs_devices->rw_devices) {
4723 WARN(1, "%s: found more than %llu devices\n",
4724 __func__, fs_devices->rw_devices);
4727 devices_info[ndevs].dev_offset = dev_offset;
4728 devices_info[ndevs].max_avail = max_avail;
4729 devices_info[ndevs].total_avail = total_avail;
4730 devices_info[ndevs].dev = device;
4735 * now sort the devices by hole size / available space
4737 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4738 btrfs_cmp_device_info, NULL);
4740 /* round down to number of usable stripes */
4741 ndevs = round_down(ndevs, devs_increment);
4743 if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4748 ndevs = min(ndevs, devs_max);
4751 * The primary goal is to maximize the number of stripes, so use as
4752 * many devices as possible, even if the stripes are not maximum sized.
4754 * The DUP profile stores more than one stripe per device, the
4755 * max_avail is the total size so we have to adjust.
4757 stripe_size = div_u64(devices_info[ndevs - 1].max_avail, dev_stripes);
4758 num_stripes = ndevs * dev_stripes;
4761 * this will have to be fixed for RAID1 and RAID10 over
4764 data_stripes = num_stripes / ncopies;
4766 if (type & BTRFS_BLOCK_GROUP_RAID5)
4767 data_stripes = num_stripes - 1;
4769 if (type & BTRFS_BLOCK_GROUP_RAID6)
4770 data_stripes = num_stripes - 2;
4773 * Use the number of data stripes to figure out how big this chunk
4774 * is really going to be in terms of logical address space,
4775 * and compare that answer with the max chunk size
4777 if (stripe_size * data_stripes > max_chunk_size) {
4778 u64 mask = (1ULL << 24) - 1;
4780 stripe_size = div_u64(max_chunk_size, data_stripes);
4782 /* bump the answer up to a 16MB boundary */
4783 stripe_size = (stripe_size + mask) & ~mask;
4785 /* but don't go higher than the limits we found
4786 * while searching for free extents
4788 if (stripe_size > devices_info[ndevs-1].max_avail)
4789 stripe_size = devices_info[ndevs-1].max_avail;
4792 /* align to BTRFS_STRIPE_LEN */
4793 stripe_size = round_down(stripe_size, BTRFS_STRIPE_LEN);
4795 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4800 map->num_stripes = num_stripes;
4802 for (i = 0; i < ndevs; ++i) {
4803 for (j = 0; j < dev_stripes; ++j) {
4804 int s = i * dev_stripes + j;
4805 map->stripes[s].dev = devices_info[i].dev;
4806 map->stripes[s].physical = devices_info[i].dev_offset +
4810 map->stripe_len = BTRFS_STRIPE_LEN;
4811 map->io_align = BTRFS_STRIPE_LEN;
4812 map->io_width = BTRFS_STRIPE_LEN;
4814 map->sub_stripes = sub_stripes;
4816 num_bytes = stripe_size * data_stripes;
4818 trace_btrfs_chunk_alloc(info, map, start, num_bytes);
4820 em = alloc_extent_map();
4826 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4827 em->map_lookup = map;
4829 em->len = num_bytes;
4830 em->block_start = 0;
4831 em->block_len = em->len;
4832 em->orig_block_len = stripe_size;
4834 em_tree = &info->mapping_tree.map_tree;
4835 write_lock(&em_tree->lock);
4836 ret = add_extent_mapping(em_tree, em, 0);
4838 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4839 refcount_inc(&em->refs);
4841 write_unlock(&em_tree->lock);
4843 free_extent_map(em);
4847 ret = btrfs_make_block_group(trans, info, 0, type, start, num_bytes);
4849 goto error_del_extent;
4851 for (i = 0; i < map->num_stripes; i++) {
4852 num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4853 btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4856 atomic64_sub(stripe_size * map->num_stripes, &info->free_chunk_space);
4858 free_extent_map(em);
4859 check_raid56_incompat_flag(info, type);
4861 kfree(devices_info);
4865 write_lock(&em_tree->lock);
4866 remove_extent_mapping(em_tree, em);
4867 write_unlock(&em_tree->lock);
4869 /* One for our allocation */
4870 free_extent_map(em);
4871 /* One for the tree reference */
4872 free_extent_map(em);
4873 /* One for the pending_chunks list reference */
4874 free_extent_map(em);
4876 kfree(devices_info);
4880 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4881 struct btrfs_fs_info *fs_info,
4882 u64 chunk_offset, u64 chunk_size)
4884 struct btrfs_root *extent_root = fs_info->extent_root;
4885 struct btrfs_root *chunk_root = fs_info->chunk_root;
4886 struct btrfs_key key;
4887 struct btrfs_device *device;
4888 struct btrfs_chunk *chunk;
4889 struct btrfs_stripe *stripe;
4890 struct extent_map *em;
4891 struct map_lookup *map;
4898 em = get_chunk_map(fs_info, chunk_offset, chunk_size);
4902 map = em->map_lookup;
4903 item_size = btrfs_chunk_item_size(map->num_stripes);
4904 stripe_size = em->orig_block_len;
4906 chunk = kzalloc(item_size, GFP_NOFS);
4913 * Take the device list mutex to prevent races with the final phase of
4914 * a device replace operation that replaces the device object associated
4915 * with the map's stripes, because the device object's id can change
4916 * at any time during that final phase of the device replace operation
4917 * (dev-replace.c:btrfs_dev_replace_finishing()).
4919 mutex_lock(&fs_info->fs_devices->device_list_mutex);
4920 for (i = 0; i < map->num_stripes; i++) {
4921 device = map->stripes[i].dev;
4922 dev_offset = map->stripes[i].physical;
4924 ret = btrfs_update_device(trans, device);
4927 ret = btrfs_alloc_dev_extent(trans, device, chunk_offset,
4928 dev_offset, stripe_size);
4933 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4937 stripe = &chunk->stripe;
4938 for (i = 0; i < map->num_stripes; i++) {
4939 device = map->stripes[i].dev;
4940 dev_offset = map->stripes[i].physical;
4942 btrfs_set_stack_stripe_devid(stripe, device->devid);
4943 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4944 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4947 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4949 btrfs_set_stack_chunk_length(chunk, chunk_size);
4950 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4951 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4952 btrfs_set_stack_chunk_type(chunk, map->type);
4953 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4954 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4955 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4956 btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
4957 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4959 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4960 key.type = BTRFS_CHUNK_ITEM_KEY;
4961 key.offset = chunk_offset;
4963 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4964 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4966 * TODO: Cleanup of inserted chunk root in case of
4969 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
4974 free_extent_map(em);
4979 * Chunk allocation falls into two parts. The first part does works
4980 * that make the new allocated chunk useable, but not do any operation
4981 * that modifies the chunk tree. The second part does the works that
4982 * require modifying the chunk tree. This division is important for the
4983 * bootstrap process of adding storage to a seed btrfs.
4985 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4986 struct btrfs_fs_info *fs_info, u64 type)
4990 ASSERT(mutex_is_locked(&fs_info->chunk_mutex));
4991 chunk_offset = find_next_chunk(fs_info);
4992 return __btrfs_alloc_chunk(trans, chunk_offset, type);
4995 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4996 struct btrfs_fs_info *fs_info)
4999 u64 sys_chunk_offset;
5003 chunk_offset = find_next_chunk(fs_info);
5004 alloc_profile = btrfs_metadata_alloc_profile(fs_info);
5005 ret = __btrfs_alloc_chunk(trans, chunk_offset, alloc_profile);
5009 sys_chunk_offset = find_next_chunk(fs_info);
5010 alloc_profile = btrfs_system_alloc_profile(fs_info);
5011 ret = __btrfs_alloc_chunk(trans, sys_chunk_offset, alloc_profile);
5015 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
5019 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5020 BTRFS_BLOCK_GROUP_RAID10 |
5021 BTRFS_BLOCK_GROUP_RAID5 |
5022 BTRFS_BLOCK_GROUP_DUP)) {
5024 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
5033 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset)
5035 struct extent_map *em;
5036 struct map_lookup *map;
5041 em = get_chunk_map(fs_info, chunk_offset, 1);
5045 map = em->map_lookup;
5046 for (i = 0; i < map->num_stripes; i++) {
5047 if (map->stripes[i].dev->missing) {
5052 if (!map->stripes[i].dev->writeable) {
5059 * If the number of missing devices is larger than max errors,
5060 * we can not write the data into that chunk successfully, so
5063 if (miss_ndevs > btrfs_chunk_max_errors(map))
5066 free_extent_map(em);
5070 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
5072 extent_map_tree_init(&tree->map_tree);
5075 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
5077 struct extent_map *em;
5080 write_lock(&tree->map_tree.lock);
5081 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
5083 remove_extent_mapping(&tree->map_tree, em);
5084 write_unlock(&tree->map_tree.lock);
5088 free_extent_map(em);
5089 /* once for the tree */
5090 free_extent_map(em);
5094 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5096 struct extent_map *em;
5097 struct map_lookup *map;
5100 em = get_chunk_map(fs_info, logical, len);
5103 * We could return errors for these cases, but that could get
5104 * ugly and we'd probably do the same thing which is just not do
5105 * anything else and exit, so return 1 so the callers don't try
5106 * to use other copies.
5110 map = em->map_lookup;
5111 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
5112 ret = map->num_stripes;
5113 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5114 ret = map->sub_stripes;
5115 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5117 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5119 * There could be two corrupted data stripes, we need
5120 * to loop retry in order to rebuild the correct data.
5122 * Fail a stripe at a time on every retry except the
5123 * stripe under reconstruction.
5125 ret = map->num_stripes;
5128 free_extent_map(em);
5130 btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
5131 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) &&
5132 fs_info->dev_replace.tgtdev)
5134 btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
5139 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
5142 struct extent_map *em;
5143 struct map_lookup *map;
5144 unsigned long len = fs_info->sectorsize;
5146 em = get_chunk_map(fs_info, logical, len);
5148 if (!WARN_ON(IS_ERR(em))) {
5149 map = em->map_lookup;
5150 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5151 len = map->stripe_len * nr_data_stripes(map);
5152 free_extent_map(em);
5157 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5159 struct extent_map *em;
5160 struct map_lookup *map;
5163 em = get_chunk_map(fs_info, logical, len);
5165 if(!WARN_ON(IS_ERR(em))) {
5166 map = em->map_lookup;
5167 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5169 free_extent_map(em);
5174 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5175 struct map_lookup *map, int first, int num,
5176 int optimal, int dev_replace_is_ongoing)
5180 struct btrfs_device *srcdev;
5182 if (dev_replace_is_ongoing &&
5183 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5184 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5185 srcdev = fs_info->dev_replace.srcdev;
5190 * try to avoid the drive that is the source drive for a
5191 * dev-replace procedure, only choose it if no other non-missing
5192 * mirror is available
5194 for (tolerance = 0; tolerance < 2; tolerance++) {
5195 if (map->stripes[optimal].dev->bdev &&
5196 (tolerance || map->stripes[optimal].dev != srcdev))
5198 for (i = first; i < first + num; i++) {
5199 if (map->stripes[i].dev->bdev &&
5200 (tolerance || map->stripes[i].dev != srcdev))
5205 /* we couldn't find one that doesn't fail. Just return something
5206 * and the io error handling code will clean up eventually
5211 static inline int parity_smaller(u64 a, u64 b)
5216 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5217 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5219 struct btrfs_bio_stripe s;
5226 for (i = 0; i < num_stripes - 1; i++) {
5227 if (parity_smaller(bbio->raid_map[i],
5228 bbio->raid_map[i+1])) {
5229 s = bbio->stripes[i];
5230 l = bbio->raid_map[i];
5231 bbio->stripes[i] = bbio->stripes[i+1];
5232 bbio->raid_map[i] = bbio->raid_map[i+1];
5233 bbio->stripes[i+1] = s;
5234 bbio->raid_map[i+1] = l;
5242 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5244 struct btrfs_bio *bbio = kzalloc(
5245 /* the size of the btrfs_bio */
5246 sizeof(struct btrfs_bio) +
5247 /* plus the variable array for the stripes */
5248 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5249 /* plus the variable array for the tgt dev */
5250 sizeof(int) * (real_stripes) +
5252 * plus the raid_map, which includes both the tgt dev
5255 sizeof(u64) * (total_stripes),
5256 GFP_NOFS|__GFP_NOFAIL);
5258 atomic_set(&bbio->error, 0);
5259 refcount_set(&bbio->refs, 1);
5264 void btrfs_get_bbio(struct btrfs_bio *bbio)
5266 WARN_ON(!refcount_read(&bbio->refs));
5267 refcount_inc(&bbio->refs);
5270 void btrfs_put_bbio(struct btrfs_bio *bbio)
5274 if (refcount_dec_and_test(&bbio->refs))
5278 /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */
5280 * Please note that, discard won't be sent to target device of device
5283 static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info,
5284 u64 logical, u64 length,
5285 struct btrfs_bio **bbio_ret)
5287 struct extent_map *em;
5288 struct map_lookup *map;
5289 struct btrfs_bio *bbio;
5293 u64 stripe_end_offset;
5300 u32 sub_stripes = 0;
5301 u64 stripes_per_dev = 0;
5302 u32 remaining_stripes = 0;
5303 u32 last_stripe = 0;
5307 /* discard always return a bbio */
5310 em = get_chunk_map(fs_info, logical, length);
5314 map = em->map_lookup;
5315 /* we don't discard raid56 yet */
5316 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5321 offset = logical - em->start;
5322 length = min_t(u64, em->len - offset, length);
5324 stripe_len = map->stripe_len;
5326 * stripe_nr counts the total number of stripes we have to stride
5327 * to get to this block
5329 stripe_nr = div64_u64(offset, stripe_len);
5331 /* stripe_offset is the offset of this block in its stripe */
5332 stripe_offset = offset - stripe_nr * stripe_len;
5334 stripe_nr_end = round_up(offset + length, map->stripe_len);
5335 stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len);
5336 stripe_cnt = stripe_nr_end - stripe_nr;
5337 stripe_end_offset = stripe_nr_end * map->stripe_len -
5340 * after this, stripe_nr is the number of stripes on this
5341 * device we have to walk to find the data, and stripe_index is
5342 * the number of our device in the stripe array
5346 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5347 BTRFS_BLOCK_GROUP_RAID10)) {
5348 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5351 sub_stripes = map->sub_stripes;
5353 factor = map->num_stripes / sub_stripes;
5354 num_stripes = min_t(u64, map->num_stripes,
5355 sub_stripes * stripe_cnt);
5356 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5357 stripe_index *= sub_stripes;
5358 stripes_per_dev = div_u64_rem(stripe_cnt, factor,
5359 &remaining_stripes);
5360 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5361 last_stripe *= sub_stripes;
5362 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5363 BTRFS_BLOCK_GROUP_DUP)) {
5364 num_stripes = map->num_stripes;
5366 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5370 bbio = alloc_btrfs_bio(num_stripes, 0);
5376 for (i = 0; i < num_stripes; i++) {
5377 bbio->stripes[i].physical =
5378 map->stripes[stripe_index].physical +
5379 stripe_offset + stripe_nr * map->stripe_len;
5380 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5382 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5383 BTRFS_BLOCK_GROUP_RAID10)) {
5384 bbio->stripes[i].length = stripes_per_dev *
5387 if (i / sub_stripes < remaining_stripes)
5388 bbio->stripes[i].length +=
5392 * Special for the first stripe and
5395 * |-------|...|-------|
5399 if (i < sub_stripes)
5400 bbio->stripes[i].length -=
5403 if (stripe_index >= last_stripe &&
5404 stripe_index <= (last_stripe +
5406 bbio->stripes[i].length -=
5409 if (i == sub_stripes - 1)
5412 bbio->stripes[i].length = length;
5416 if (stripe_index == map->num_stripes) {
5423 bbio->map_type = map->type;
5424 bbio->num_stripes = num_stripes;
5426 free_extent_map(em);
5431 * In dev-replace case, for repair case (that's the only case where the mirror
5432 * is selected explicitly when calling btrfs_map_block), blocks left of the
5433 * left cursor can also be read from the target drive.
5435 * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the
5437 * For READ, it also needs to be supported using the same mirror number.
5439 * If the requested block is not left of the left cursor, EIO is returned. This
5440 * can happen because btrfs_num_copies() returns one more in the dev-replace
5443 static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info,
5444 u64 logical, u64 length,
5445 u64 srcdev_devid, int *mirror_num,
5448 struct btrfs_bio *bbio = NULL;
5450 int index_srcdev = 0;
5452 u64 physical_of_found = 0;
5456 ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS,
5457 logical, &length, &bbio, 0, 0);
5459 ASSERT(bbio == NULL);
5463 num_stripes = bbio->num_stripes;
5464 if (*mirror_num > num_stripes) {
5466 * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror,
5467 * that means that the requested area is not left of the left
5470 btrfs_put_bbio(bbio);
5475 * process the rest of the function using the mirror_num of the source
5476 * drive. Therefore look it up first. At the end, patch the device
5477 * pointer to the one of the target drive.
5479 for (i = 0; i < num_stripes; i++) {
5480 if (bbio->stripes[i].dev->devid != srcdev_devid)
5484 * In case of DUP, in order to keep it simple, only add the
5485 * mirror with the lowest physical address
5488 physical_of_found <= bbio->stripes[i].physical)
5493 physical_of_found = bbio->stripes[i].physical;
5496 btrfs_put_bbio(bbio);
5502 *mirror_num = index_srcdev + 1;
5503 *physical = physical_of_found;
5507 static void handle_ops_on_dev_replace(enum btrfs_map_op op,
5508 struct btrfs_bio **bbio_ret,
5509 struct btrfs_dev_replace *dev_replace,
5510 int *num_stripes_ret, int *max_errors_ret)
5512 struct btrfs_bio *bbio = *bbio_ret;
5513 u64 srcdev_devid = dev_replace->srcdev->devid;
5514 int tgtdev_indexes = 0;
5515 int num_stripes = *num_stripes_ret;
5516 int max_errors = *max_errors_ret;
5519 if (op == BTRFS_MAP_WRITE) {
5520 int index_where_to_add;
5523 * duplicate the write operations while the dev replace
5524 * procedure is running. Since the copying of the old disk to
5525 * the new disk takes place at run time while the filesystem is
5526 * mounted writable, the regular write operations to the old
5527 * disk have to be duplicated to go to the new disk as well.
5529 * Note that device->missing is handled by the caller, and that
5530 * the write to the old disk is already set up in the stripes
5533 index_where_to_add = num_stripes;
5534 for (i = 0; i < num_stripes; i++) {
5535 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5536 /* write to new disk, too */
5537 struct btrfs_bio_stripe *new =
5538 bbio->stripes + index_where_to_add;
5539 struct btrfs_bio_stripe *old =
5542 new->physical = old->physical;
5543 new->length = old->length;
5544 new->dev = dev_replace->tgtdev;
5545 bbio->tgtdev_map[i] = index_where_to_add;
5546 index_where_to_add++;
5551 num_stripes = index_where_to_add;
5552 } else if (op == BTRFS_MAP_GET_READ_MIRRORS) {
5553 int index_srcdev = 0;
5555 u64 physical_of_found = 0;
5558 * During the dev-replace procedure, the target drive can also
5559 * be used to read data in case it is needed to repair a corrupt
5560 * block elsewhere. This is possible if the requested area is
5561 * left of the left cursor. In this area, the target drive is a
5562 * full copy of the source drive.
5564 for (i = 0; i < num_stripes; i++) {
5565 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5567 * In case of DUP, in order to keep it simple,
5568 * only add the mirror with the lowest physical
5572 physical_of_found <=
5573 bbio->stripes[i].physical)
5577 physical_of_found = bbio->stripes[i].physical;
5581 struct btrfs_bio_stripe *tgtdev_stripe =
5582 bbio->stripes + num_stripes;
5584 tgtdev_stripe->physical = physical_of_found;
5585 tgtdev_stripe->length =
5586 bbio->stripes[index_srcdev].length;
5587 tgtdev_stripe->dev = dev_replace->tgtdev;
5588 bbio->tgtdev_map[index_srcdev] = num_stripes;
5595 *num_stripes_ret = num_stripes;
5596 *max_errors_ret = max_errors;
5597 bbio->num_tgtdevs = tgtdev_indexes;
5601 static bool need_full_stripe(enum btrfs_map_op op)
5603 return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS);
5606 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
5607 enum btrfs_map_op op,
5608 u64 logical, u64 *length,
5609 struct btrfs_bio **bbio_ret,
5610 int mirror_num, int need_raid_map)
5612 struct extent_map *em;
5613 struct map_lookup *map;
5623 int tgtdev_indexes = 0;
5624 struct btrfs_bio *bbio = NULL;
5625 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
5626 int dev_replace_is_ongoing = 0;
5627 int num_alloc_stripes;
5628 int patch_the_first_stripe_for_dev_replace = 0;
5629 u64 physical_to_patch_in_first_stripe = 0;
5630 u64 raid56_full_stripe_start = (u64)-1;
5632 if (op == BTRFS_MAP_DISCARD)
5633 return __btrfs_map_block_for_discard(fs_info, logical,
5636 em = get_chunk_map(fs_info, logical, *length);
5640 map = em->map_lookup;
5641 offset = logical - em->start;
5643 stripe_len = map->stripe_len;
5646 * stripe_nr counts the total number of stripes we have to stride
5647 * to get to this block
5649 stripe_nr = div64_u64(stripe_nr, stripe_len);
5651 stripe_offset = stripe_nr * stripe_len;
5652 if (offset < stripe_offset) {
5654 "stripe math has gone wrong, stripe_offset=%llu, offset=%llu, start=%llu, logical=%llu, stripe_len=%llu",
5655 stripe_offset, offset, em->start, logical,
5657 free_extent_map(em);
5661 /* stripe_offset is the offset of this block in its stripe*/
5662 stripe_offset = offset - stripe_offset;
5664 /* if we're here for raid56, we need to know the stripe aligned start */
5665 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5666 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5667 raid56_full_stripe_start = offset;
5669 /* allow a write of a full stripe, but make sure we don't
5670 * allow straddling of stripes
5672 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5674 raid56_full_stripe_start *= full_stripe_len;
5677 if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5679 /* For writes to RAID[56], allow a full stripeset across all disks.
5680 For other RAID types and for RAID[56] reads, just allow a single
5681 stripe (on a single disk). */
5682 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
5683 (op == BTRFS_MAP_WRITE)) {
5684 max_len = stripe_len * nr_data_stripes(map) -
5685 (offset - raid56_full_stripe_start);
5687 /* we limit the length of each bio to what fits in a stripe */
5688 max_len = stripe_len - stripe_offset;
5690 *length = min_t(u64, em->len - offset, max_len);
5692 *length = em->len - offset;
5695 /* This is for when we're called from btrfs_merge_bio_hook() and all
5696 it cares about is the length */
5700 btrfs_dev_replace_lock(dev_replace, 0);
5701 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5702 if (!dev_replace_is_ongoing)
5703 btrfs_dev_replace_unlock(dev_replace, 0);
5705 btrfs_dev_replace_set_lock_blocking(dev_replace);
5707 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
5708 !need_full_stripe(op) && dev_replace->tgtdev != NULL) {
5709 ret = get_extra_mirror_from_replace(fs_info, logical, *length,
5710 dev_replace->srcdev->devid,
5712 &physical_to_patch_in_first_stripe);
5716 patch_the_first_stripe_for_dev_replace = 1;
5717 } else if (mirror_num > map->num_stripes) {
5723 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5724 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5726 if (op != BTRFS_MAP_WRITE && op != BTRFS_MAP_GET_READ_MIRRORS)
5728 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
5729 if (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS)
5730 num_stripes = map->num_stripes;
5731 else if (mirror_num)
5732 stripe_index = mirror_num - 1;
5734 stripe_index = find_live_mirror(fs_info, map, 0,
5736 current->pid % map->num_stripes,
5737 dev_replace_is_ongoing);
5738 mirror_num = stripe_index + 1;
5741 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
5742 if (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS) {
5743 num_stripes = map->num_stripes;
5744 } else if (mirror_num) {
5745 stripe_index = mirror_num - 1;
5750 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5751 u32 factor = map->num_stripes / map->sub_stripes;
5753 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5754 stripe_index *= map->sub_stripes;
5756 if (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS)
5757 num_stripes = map->sub_stripes;
5758 else if (mirror_num)
5759 stripe_index += mirror_num - 1;
5761 int old_stripe_index = stripe_index;
5762 stripe_index = find_live_mirror(fs_info, map,
5764 map->sub_stripes, stripe_index +
5765 current->pid % map->sub_stripes,
5766 dev_replace_is_ongoing);
5767 mirror_num = stripe_index - old_stripe_index + 1;
5770 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5771 if (need_raid_map &&
5772 (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS ||
5774 /* push stripe_nr back to the start of the full stripe */
5775 stripe_nr = div64_u64(raid56_full_stripe_start,
5776 stripe_len * nr_data_stripes(map));
5778 /* RAID[56] write or recovery. Return all stripes */
5779 num_stripes = map->num_stripes;
5780 max_errors = nr_parity_stripes(map);
5782 *length = map->stripe_len;
5787 * Mirror #0 or #1 means the original data block.
5788 * Mirror #2 is RAID5 parity block.
5789 * Mirror #3 is RAID6 Q block.
5791 stripe_nr = div_u64_rem(stripe_nr,
5792 nr_data_stripes(map), &stripe_index);
5794 stripe_index = nr_data_stripes(map) +
5797 /* We distribute the parity blocks across stripes */
5798 div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
5800 if ((op != BTRFS_MAP_WRITE &&
5801 op != BTRFS_MAP_GET_READ_MIRRORS) &&
5807 * after this, stripe_nr is the number of stripes on this
5808 * device we have to walk to find the data, and stripe_index is
5809 * the number of our device in the stripe array
5811 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5813 mirror_num = stripe_index + 1;
5815 if (stripe_index >= map->num_stripes) {
5817 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
5818 stripe_index, map->num_stripes);
5823 num_alloc_stripes = num_stripes;
5824 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) {
5825 if (op == BTRFS_MAP_WRITE)
5826 num_alloc_stripes <<= 1;
5827 if (op == BTRFS_MAP_GET_READ_MIRRORS)
5828 num_alloc_stripes++;
5829 tgtdev_indexes = num_stripes;
5832 bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
5837 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL)
5838 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
5840 /* build raid_map */
5841 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map &&
5842 (need_full_stripe(op) || mirror_num > 1)) {
5846 bbio->raid_map = (u64 *)((void *)bbio->stripes +
5847 sizeof(struct btrfs_bio_stripe) *
5849 sizeof(int) * tgtdev_indexes);
5851 /* Work out the disk rotation on this stripe-set */
5852 div_u64_rem(stripe_nr, num_stripes, &rot);
5854 /* Fill in the logical address of each stripe */
5855 tmp = stripe_nr * nr_data_stripes(map);
5856 for (i = 0; i < nr_data_stripes(map); i++)
5857 bbio->raid_map[(i+rot) % num_stripes] =
5858 em->start + (tmp + i) * map->stripe_len;
5860 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5861 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5862 bbio->raid_map[(i+rot+1) % num_stripes] =
5867 for (i = 0; i < num_stripes; i++) {
5868 bbio->stripes[i].physical =
5869 map->stripes[stripe_index].physical +
5871 stripe_nr * map->stripe_len;
5872 bbio->stripes[i].dev =
5873 map->stripes[stripe_index].dev;
5877 if (need_full_stripe(op))
5878 max_errors = btrfs_chunk_max_errors(map);
5881 sort_parity_stripes(bbio, num_stripes);
5883 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
5884 need_full_stripe(op)) {
5885 handle_ops_on_dev_replace(op, &bbio, dev_replace, &num_stripes,
5890 bbio->map_type = map->type;
5891 bbio->num_stripes = num_stripes;
5892 bbio->max_errors = max_errors;
5893 bbio->mirror_num = mirror_num;
5896 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5897 * mirror_num == num_stripes + 1 && dev_replace target drive is
5898 * available as a mirror
5900 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5901 WARN_ON(num_stripes > 1);
5902 bbio->stripes[0].dev = dev_replace->tgtdev;
5903 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5904 bbio->mirror_num = map->num_stripes + 1;
5907 if (dev_replace_is_ongoing) {
5908 btrfs_dev_replace_clear_lock_blocking(dev_replace);
5909 btrfs_dev_replace_unlock(dev_replace, 0);
5911 free_extent_map(em);
5915 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
5916 u64 logical, u64 *length,
5917 struct btrfs_bio **bbio_ret, int mirror_num)
5919 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret,
5923 /* For Scrub/replace */
5924 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
5925 u64 logical, u64 *length,
5926 struct btrfs_bio **bbio_ret)
5928 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1);
5931 int btrfs_rmap_block(struct btrfs_fs_info *fs_info,
5932 u64 chunk_start, u64 physical, u64 devid,
5933 u64 **logical, int *naddrs, int *stripe_len)
5935 struct extent_map *em;
5936 struct map_lookup *map;
5944 em = get_chunk_map(fs_info, chunk_start, 1);
5948 map = em->map_lookup;
5950 rmap_len = map->stripe_len;
5952 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5953 length = div_u64(length, map->num_stripes / map->sub_stripes);
5954 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5955 length = div_u64(length, map->num_stripes);
5956 else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5957 length = div_u64(length, nr_data_stripes(map));
5958 rmap_len = map->stripe_len * nr_data_stripes(map);
5961 buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
5962 BUG_ON(!buf); /* -ENOMEM */
5964 for (i = 0; i < map->num_stripes; i++) {
5965 if (devid && map->stripes[i].dev->devid != devid)
5967 if (map->stripes[i].physical > physical ||
5968 map->stripes[i].physical + length <= physical)
5971 stripe_nr = physical - map->stripes[i].physical;
5972 stripe_nr = div64_u64(stripe_nr, map->stripe_len);
5974 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5975 stripe_nr = stripe_nr * map->num_stripes + i;
5976 stripe_nr = div_u64(stripe_nr, map->sub_stripes);
5977 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5978 stripe_nr = stripe_nr * map->num_stripes + i;
5979 } /* else if RAID[56], multiply by nr_data_stripes().
5980 * Alternatively, just use rmap_len below instead of
5981 * map->stripe_len */
5983 bytenr = chunk_start + stripe_nr * rmap_len;
5984 WARN_ON(nr >= map->num_stripes);
5985 for (j = 0; j < nr; j++) {
5986 if (buf[j] == bytenr)
5990 WARN_ON(nr >= map->num_stripes);
5997 *stripe_len = rmap_len;
5999 free_extent_map(em);
6003 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
6005 bio->bi_private = bbio->private;
6006 bio->bi_end_io = bbio->end_io;
6009 btrfs_put_bbio(bbio);
6012 static void btrfs_end_bio(struct bio *bio)
6014 struct btrfs_bio *bbio = bio->bi_private;
6015 int is_orig_bio = 0;
6017 if (bio->bi_status) {
6018 atomic_inc(&bbio->error);
6019 if (bio->bi_status == BLK_STS_IOERR ||
6020 bio->bi_status == BLK_STS_TARGET) {
6021 unsigned int stripe_index =
6022 btrfs_io_bio(bio)->stripe_index;
6023 struct btrfs_device *dev;
6025 BUG_ON(stripe_index >= bbio->num_stripes);
6026 dev = bbio->stripes[stripe_index].dev;
6028 if (bio_op(bio) == REQ_OP_WRITE)
6029 btrfs_dev_stat_inc(dev,
6030 BTRFS_DEV_STAT_WRITE_ERRS);
6032 btrfs_dev_stat_inc(dev,
6033 BTRFS_DEV_STAT_READ_ERRS);
6034 if (bio->bi_opf & REQ_PREFLUSH)
6035 btrfs_dev_stat_inc(dev,
6036 BTRFS_DEV_STAT_FLUSH_ERRS);
6037 btrfs_dev_stat_print_on_error(dev);
6042 if (bio == bbio->orig_bio)
6045 btrfs_bio_counter_dec(bbio->fs_info);
6047 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6050 bio = bbio->orig_bio;
6053 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6054 /* only send an error to the higher layers if it is
6055 * beyond the tolerance of the btrfs bio
6057 if (atomic_read(&bbio->error) > bbio->max_errors) {
6058 bio->bi_status = BLK_STS_IOERR;
6061 * this bio is actually up to date, we didn't
6062 * go over the max number of errors
6067 btrfs_end_bbio(bbio, bio);
6068 } else if (!is_orig_bio) {
6074 * see run_scheduled_bios for a description of why bios are collected for
6077 * This will add one bio to the pending list for a device and make sure
6078 * the work struct is scheduled.
6080 static noinline void btrfs_schedule_bio(struct btrfs_device *device,
6083 struct btrfs_fs_info *fs_info = device->fs_info;
6084 int should_queue = 1;
6085 struct btrfs_pending_bios *pending_bios;
6087 if (device->missing || !device->bdev) {
6092 /* don't bother with additional async steps for reads, right now */
6093 if (bio_op(bio) == REQ_OP_READ) {
6095 btrfsic_submit_bio(bio);
6101 * nr_async_bios allows us to reliably return congestion to the
6102 * higher layers. Otherwise, the async bio makes it appear we have
6103 * made progress against dirty pages when we've really just put it
6104 * on a queue for later
6106 atomic_inc(&fs_info->nr_async_bios);
6107 WARN_ON(bio->bi_next);
6108 bio->bi_next = NULL;
6110 spin_lock(&device->io_lock);
6111 if (op_is_sync(bio->bi_opf))
6112 pending_bios = &device->pending_sync_bios;
6114 pending_bios = &device->pending_bios;
6116 if (pending_bios->tail)
6117 pending_bios->tail->bi_next = bio;
6119 pending_bios->tail = bio;
6120 if (!pending_bios->head)
6121 pending_bios->head = bio;
6122 if (device->running_pending)
6125 spin_unlock(&device->io_lock);
6128 btrfs_queue_work(fs_info->submit_workers, &device->work);
6131 static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio,
6132 u64 physical, int dev_nr, int async)
6134 struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
6135 struct btrfs_fs_info *fs_info = bbio->fs_info;
6137 bio->bi_private = bbio;
6138 btrfs_io_bio(bio)->stripe_index = dev_nr;
6139 bio->bi_end_io = btrfs_end_bio;
6140 bio->bi_iter.bi_sector = physical >> 9;
6143 struct rcu_string *name;
6146 name = rcu_dereference(dev->name);
6147 btrfs_debug(fs_info,
6148 "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
6149 bio_op(bio), bio->bi_opf,
6150 (u64)bio->bi_iter.bi_sector,
6151 (u_long)dev->bdev->bd_dev, name->str, dev->devid,
6152 bio->bi_iter.bi_size);
6156 bio_set_dev(bio, dev->bdev);
6158 btrfs_bio_counter_inc_noblocked(fs_info);
6161 btrfs_schedule_bio(dev, bio);
6163 btrfsic_submit_bio(bio);
6166 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6168 atomic_inc(&bbio->error);
6169 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6170 /* Should be the original bio. */
6171 WARN_ON(bio != bbio->orig_bio);
6173 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6174 bio->bi_iter.bi_sector = logical >> 9;
6175 if (atomic_read(&bbio->error) > bbio->max_errors)
6176 bio->bi_status = BLK_STS_IOERR;
6178 bio->bi_status = BLK_STS_OK;
6179 btrfs_end_bbio(bbio, bio);
6183 blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
6184 int mirror_num, int async_submit)
6186 struct btrfs_device *dev;
6187 struct bio *first_bio = bio;
6188 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
6194 struct btrfs_bio *bbio = NULL;
6196 length = bio->bi_iter.bi_size;
6197 map_length = length;
6199 btrfs_bio_counter_inc_blocked(fs_info);
6200 ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical,
6201 &map_length, &bbio, mirror_num, 1);
6203 btrfs_bio_counter_dec(fs_info);
6204 return errno_to_blk_status(ret);
6207 total_devs = bbio->num_stripes;
6208 bbio->orig_bio = first_bio;
6209 bbio->private = first_bio->bi_private;
6210 bbio->end_io = first_bio->bi_end_io;
6211 bbio->fs_info = fs_info;
6212 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6214 if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6215 ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) {
6216 /* In this case, map_length has been set to the length of
6217 a single stripe; not the whole write */
6218 if (bio_op(bio) == REQ_OP_WRITE) {
6219 ret = raid56_parity_write(fs_info, bio, bbio,
6222 ret = raid56_parity_recover(fs_info, bio, bbio,
6223 map_length, mirror_num, 1);
6226 btrfs_bio_counter_dec(fs_info);
6227 return errno_to_blk_status(ret);
6230 if (map_length < length) {
6232 "mapping failed logical %llu bio len %llu len %llu",
6233 logical, length, map_length);
6237 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6238 dev = bbio->stripes[dev_nr].dev;
6239 if (!dev || !dev->bdev ||
6240 (bio_op(first_bio) == REQ_OP_WRITE && !dev->writeable)) {
6241 bbio_error(bbio, first_bio, logical);
6245 if (dev_nr < total_devs - 1)
6246 bio = btrfs_bio_clone(first_bio);
6250 submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical,
6251 dev_nr, async_submit);
6253 btrfs_bio_counter_dec(fs_info);
6257 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
6260 struct btrfs_device *device;
6261 struct btrfs_fs_devices *cur_devices;
6263 cur_devices = fs_info->fs_devices;
6264 while (cur_devices) {
6266 !memcmp(cur_devices->fsid, fsid, BTRFS_FSID_SIZE)) {
6267 device = find_device(cur_devices, devid, uuid);
6271 cur_devices = cur_devices->seed;
6276 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices,
6277 u64 devid, u8 *dev_uuid)
6279 struct btrfs_device *device;
6281 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6285 list_add(&device->dev_list, &fs_devices->devices);
6286 device->fs_devices = fs_devices;
6287 fs_devices->num_devices++;
6289 device->missing = 1;
6290 fs_devices->missing_devices++;
6296 * btrfs_alloc_device - allocate struct btrfs_device
6297 * @fs_info: used only for generating a new devid, can be NULL if
6298 * devid is provided (i.e. @devid != NULL).
6299 * @devid: a pointer to devid for this device. If NULL a new devid
6301 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6304 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6305 * on error. Returned struct is not linked onto any lists and can be
6306 * destroyed with kfree() right away.
6308 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6312 struct btrfs_device *dev;
6315 if (WARN_ON(!devid && !fs_info))
6316 return ERR_PTR(-EINVAL);
6318 dev = __alloc_device();
6327 ret = find_next_devid(fs_info, &tmp);
6330 return ERR_PTR(ret);
6336 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6338 generate_random_uuid(dev->uuid);
6340 btrfs_init_work(&dev->work, btrfs_submit_helper,
6341 pending_bios_fn, NULL, NULL);
6346 /* Return -EIO if any error, otherwise return 0. */
6347 static int btrfs_check_chunk_valid(struct btrfs_fs_info *fs_info,
6348 struct extent_buffer *leaf,
6349 struct btrfs_chunk *chunk, u64 logical)
6357 length = btrfs_chunk_length(leaf, chunk);
6358 stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6359 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6360 sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6361 type = btrfs_chunk_type(leaf, chunk);
6364 btrfs_err(fs_info, "invalid chunk num_stripes: %u",
6368 if (!IS_ALIGNED(logical, fs_info->sectorsize)) {
6369 btrfs_err(fs_info, "invalid chunk logical %llu", logical);
6372 if (btrfs_chunk_sector_size(leaf, chunk) != fs_info->sectorsize) {
6373 btrfs_err(fs_info, "invalid chunk sectorsize %u",
6374 btrfs_chunk_sector_size(leaf, chunk));
6377 if (!length || !IS_ALIGNED(length, fs_info->sectorsize)) {
6378 btrfs_err(fs_info, "invalid chunk length %llu", length);
6381 if (!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN) {
6382 btrfs_err(fs_info, "invalid chunk stripe length: %llu",
6386 if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6388 btrfs_err(fs_info, "unrecognized chunk type: %llu",
6389 ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
6390 BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6391 btrfs_chunk_type(leaf, chunk));
6394 if ((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes != 2) ||
6395 (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) ||
6396 (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) ||
6397 (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) ||
6398 (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) ||
6399 ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
6400 num_stripes != 1)) {
6402 "invalid num_stripes:sub_stripes %u:%u for profile %llu",
6403 num_stripes, sub_stripes,
6404 type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
6411 static int read_one_chunk(struct btrfs_fs_info *fs_info, struct btrfs_key *key,
6412 struct extent_buffer *leaf,
6413 struct btrfs_chunk *chunk)
6415 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
6416 struct map_lookup *map;
6417 struct extent_map *em;
6421 u8 uuid[BTRFS_UUID_SIZE];
6426 logical = key->offset;
6427 length = btrfs_chunk_length(leaf, chunk);
6428 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6430 ret = btrfs_check_chunk_valid(fs_info, leaf, chunk, logical);
6434 read_lock(&map_tree->map_tree.lock);
6435 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
6436 read_unlock(&map_tree->map_tree.lock);
6438 /* already mapped? */
6439 if (em && em->start <= logical && em->start + em->len > logical) {
6440 free_extent_map(em);
6443 free_extent_map(em);
6446 em = alloc_extent_map();
6449 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6451 free_extent_map(em);
6455 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6456 em->map_lookup = map;
6457 em->start = logical;
6460 em->block_start = 0;
6461 em->block_len = em->len;
6463 map->num_stripes = num_stripes;
6464 map->io_width = btrfs_chunk_io_width(leaf, chunk);
6465 map->io_align = btrfs_chunk_io_align(leaf, chunk);
6466 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6467 map->type = btrfs_chunk_type(leaf, chunk);
6468 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6469 for (i = 0; i < num_stripes; i++) {
6470 map->stripes[i].physical =
6471 btrfs_stripe_offset_nr(leaf, chunk, i);
6472 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6473 read_extent_buffer(leaf, uuid, (unsigned long)
6474 btrfs_stripe_dev_uuid_nr(chunk, i),
6476 map->stripes[i].dev = btrfs_find_device(fs_info, devid,
6478 if (!map->stripes[i].dev &&
6479 !btrfs_test_opt(fs_info, DEGRADED)) {
6480 free_extent_map(em);
6481 btrfs_report_missing_device(fs_info, devid, uuid);
6484 if (!map->stripes[i].dev) {
6485 map->stripes[i].dev =
6486 add_missing_dev(fs_info->fs_devices, devid,
6488 if (!map->stripes[i].dev) {
6489 free_extent_map(em);
6492 btrfs_report_missing_device(fs_info, devid, uuid);
6494 map->stripes[i].dev->in_fs_metadata = 1;
6497 write_lock(&map_tree->map_tree.lock);
6498 ret = add_extent_mapping(&map_tree->map_tree, em, 0);
6499 write_unlock(&map_tree->map_tree.lock);
6502 "failed to add chunk map, start=%llu len=%llu: %d",
6503 em->start, em->len, ret);
6505 free_extent_map(em);
6510 static void fill_device_from_item(struct extent_buffer *leaf,
6511 struct btrfs_dev_item *dev_item,
6512 struct btrfs_device *device)
6516 device->devid = btrfs_device_id(leaf, dev_item);
6517 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6518 device->total_bytes = device->disk_total_bytes;
6519 device->commit_total_bytes = device->disk_total_bytes;
6520 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6521 device->commit_bytes_used = device->bytes_used;
6522 device->type = btrfs_device_type(leaf, dev_item);
6523 device->io_align = btrfs_device_io_align(leaf, dev_item);
6524 device->io_width = btrfs_device_io_width(leaf, dev_item);
6525 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6526 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6527 device->is_tgtdev_for_dev_replace = 0;
6529 ptr = btrfs_device_uuid(dev_item);
6530 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6533 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info,
6536 struct btrfs_fs_devices *fs_devices;
6539 BUG_ON(!mutex_is_locked(&uuid_mutex));
6542 fs_devices = fs_info->fs_devices->seed;
6543 while (fs_devices) {
6544 if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE))
6547 fs_devices = fs_devices->seed;
6550 fs_devices = find_fsid(fsid);
6552 if (!btrfs_test_opt(fs_info, DEGRADED))
6553 return ERR_PTR(-ENOENT);
6555 fs_devices = alloc_fs_devices(fsid);
6556 if (IS_ERR(fs_devices))
6559 fs_devices->seeding = 1;
6560 fs_devices->opened = 1;
6564 fs_devices = clone_fs_devices(fs_devices);
6565 if (IS_ERR(fs_devices))
6568 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
6569 fs_info->bdev_holder);
6571 free_fs_devices(fs_devices);
6572 fs_devices = ERR_PTR(ret);
6576 if (!fs_devices->seeding) {
6577 __btrfs_close_devices(fs_devices);
6578 free_fs_devices(fs_devices);
6579 fs_devices = ERR_PTR(-EINVAL);
6583 fs_devices->seed = fs_info->fs_devices->seed;
6584 fs_info->fs_devices->seed = fs_devices;
6589 static int read_one_dev(struct btrfs_fs_info *fs_info,
6590 struct extent_buffer *leaf,
6591 struct btrfs_dev_item *dev_item)
6593 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6594 struct btrfs_device *device;
6597 u8 fs_uuid[BTRFS_FSID_SIZE];
6598 u8 dev_uuid[BTRFS_UUID_SIZE];
6600 devid = btrfs_device_id(leaf, dev_item);
6601 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6603 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6606 if (memcmp(fs_uuid, fs_info->fsid, BTRFS_FSID_SIZE)) {
6607 fs_devices = open_seed_devices(fs_info, fs_uuid);
6608 if (IS_ERR(fs_devices))
6609 return PTR_ERR(fs_devices);
6612 device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
6614 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6615 btrfs_report_missing_device(fs_info, devid, dev_uuid);
6619 device = add_missing_dev(fs_devices, devid, dev_uuid);
6622 btrfs_report_missing_device(fs_info, devid, dev_uuid);
6624 if (!device->bdev) {
6625 btrfs_report_missing_device(fs_info, devid, dev_uuid);
6626 if (!btrfs_test_opt(fs_info, DEGRADED))
6630 if(!device->bdev && !device->missing) {
6632 * this happens when a device that was properly setup
6633 * in the device info lists suddenly goes bad.
6634 * device->bdev is NULL, and so we have to set
6635 * device->missing to one here
6637 device->fs_devices->missing_devices++;
6638 device->missing = 1;
6641 /* Move the device to its own fs_devices */
6642 if (device->fs_devices != fs_devices) {
6643 ASSERT(device->missing);
6645 list_move(&device->dev_list, &fs_devices->devices);
6646 device->fs_devices->num_devices--;
6647 fs_devices->num_devices++;
6649 device->fs_devices->missing_devices--;
6650 fs_devices->missing_devices++;
6652 device->fs_devices = fs_devices;
6656 if (device->fs_devices != fs_info->fs_devices) {
6657 BUG_ON(device->writeable);
6658 if (device->generation !=
6659 btrfs_device_generation(leaf, dev_item))
6663 fill_device_from_item(leaf, dev_item, device);
6664 device->in_fs_metadata = 1;
6665 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
6666 device->fs_devices->total_rw_bytes += device->total_bytes;
6667 atomic64_add(device->total_bytes - device->bytes_used,
6668 &fs_info->free_chunk_space);
6674 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
6676 struct btrfs_root *root = fs_info->tree_root;
6677 struct btrfs_super_block *super_copy = fs_info->super_copy;
6678 struct extent_buffer *sb;
6679 struct btrfs_disk_key *disk_key;
6680 struct btrfs_chunk *chunk;
6682 unsigned long sb_array_offset;
6689 struct btrfs_key key;
6691 ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize);
6693 * This will create extent buffer of nodesize, superblock size is
6694 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6695 * overallocate but we can keep it as-is, only the first page is used.
6697 sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET);
6700 set_extent_buffer_uptodate(sb);
6701 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6703 * The sb extent buffer is artificial and just used to read the system array.
6704 * set_extent_buffer_uptodate() call does not properly mark all it's
6705 * pages up-to-date when the page is larger: extent does not cover the
6706 * whole page and consequently check_page_uptodate does not find all
6707 * the page's extents up-to-date (the hole beyond sb),
6708 * write_extent_buffer then triggers a WARN_ON.
6710 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6711 * but sb spans only this function. Add an explicit SetPageUptodate call
6712 * to silence the warning eg. on PowerPC 64.
6714 if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
6715 SetPageUptodate(sb->pages[0]);
6717 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6718 array_size = btrfs_super_sys_array_size(super_copy);
6720 array_ptr = super_copy->sys_chunk_array;
6721 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6724 while (cur_offset < array_size) {
6725 disk_key = (struct btrfs_disk_key *)array_ptr;
6726 len = sizeof(*disk_key);
6727 if (cur_offset + len > array_size)
6728 goto out_short_read;
6730 btrfs_disk_key_to_cpu(&key, disk_key);
6733 sb_array_offset += len;
6736 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6737 chunk = (struct btrfs_chunk *)sb_array_offset;
6739 * At least one btrfs_chunk with one stripe must be
6740 * present, exact stripe count check comes afterwards
6742 len = btrfs_chunk_item_size(1);
6743 if (cur_offset + len > array_size)
6744 goto out_short_read;
6746 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6749 "invalid number of stripes %u in sys_array at offset %u",
6750 num_stripes, cur_offset);
6755 type = btrfs_chunk_type(sb, chunk);
6756 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
6758 "invalid chunk type %llu in sys_array at offset %u",
6764 len = btrfs_chunk_item_size(num_stripes);
6765 if (cur_offset + len > array_size)
6766 goto out_short_read;
6768 ret = read_one_chunk(fs_info, &key, sb, chunk);
6773 "unexpected item type %u in sys_array at offset %u",
6774 (u32)key.type, cur_offset);
6779 sb_array_offset += len;
6782 clear_extent_buffer_uptodate(sb);
6783 free_extent_buffer_stale(sb);
6787 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
6789 clear_extent_buffer_uptodate(sb);
6790 free_extent_buffer_stale(sb);
6794 void btrfs_report_missing_device(struct btrfs_fs_info *fs_info, u64 devid,
6797 btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing", devid, uuid);
6801 * Check if all chunks in the fs are OK for read-write degraded mount
6803 * Return true if all chunks meet the minimal RW mount requirements.
6804 * Return false if any chunk doesn't meet the minimal RW mount requirements.
6806 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info)
6808 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
6809 struct extent_map *em;
6813 read_lock(&map_tree->map_tree.lock);
6814 em = lookup_extent_mapping(&map_tree->map_tree, 0, (u64)-1);
6815 read_unlock(&map_tree->map_tree.lock);
6816 /* No chunk at all? Return false anyway */
6822 struct map_lookup *map;
6827 map = em->map_lookup;
6829 btrfs_get_num_tolerated_disk_barrier_failures(
6831 for (i = 0; i < map->num_stripes; i++) {
6832 struct btrfs_device *dev = map->stripes[i].dev;
6834 if (!dev || !dev->bdev || dev->missing ||
6835 dev->last_flush_error)
6838 if (missing > max_tolerated) {
6840 "chunk %llu missing %d devices, max tolerance is %d for writeable mount",
6841 em->start, missing, max_tolerated);
6842 free_extent_map(em);
6846 next_start = extent_map_end(em);
6847 free_extent_map(em);
6849 read_lock(&map_tree->map_tree.lock);
6850 em = lookup_extent_mapping(&map_tree->map_tree, next_start,
6851 (u64)(-1) - next_start);
6852 read_unlock(&map_tree->map_tree.lock);
6858 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
6860 struct btrfs_root *root = fs_info->chunk_root;
6861 struct btrfs_path *path;
6862 struct extent_buffer *leaf;
6863 struct btrfs_key key;
6864 struct btrfs_key found_key;
6869 path = btrfs_alloc_path();
6873 mutex_lock(&uuid_mutex);
6874 mutex_lock(&fs_info->chunk_mutex);
6877 * Read all device items, and then all the chunk items. All
6878 * device items are found before any chunk item (their object id
6879 * is smaller than the lowest possible object id for a chunk
6880 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6882 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6885 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6889 leaf = path->nodes[0];
6890 slot = path->slots[0];
6891 if (slot >= btrfs_header_nritems(leaf)) {
6892 ret = btrfs_next_leaf(root, path);
6899 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6900 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6901 struct btrfs_dev_item *dev_item;
6902 dev_item = btrfs_item_ptr(leaf, slot,
6903 struct btrfs_dev_item);
6904 ret = read_one_dev(fs_info, leaf, dev_item);
6908 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6909 struct btrfs_chunk *chunk;
6910 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6911 ret = read_one_chunk(fs_info, &found_key, leaf, chunk);
6919 * After loading chunk tree, we've got all device information,
6920 * do another round of validation checks.
6922 if (total_dev != fs_info->fs_devices->total_devices) {
6924 "super_num_devices %llu mismatch with num_devices %llu found here",
6925 btrfs_super_num_devices(fs_info->super_copy),
6930 if (btrfs_super_total_bytes(fs_info->super_copy) <
6931 fs_info->fs_devices->total_rw_bytes) {
6933 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
6934 btrfs_super_total_bytes(fs_info->super_copy),
6935 fs_info->fs_devices->total_rw_bytes);
6941 mutex_unlock(&fs_info->chunk_mutex);
6942 mutex_unlock(&uuid_mutex);
6944 btrfs_free_path(path);
6948 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6950 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6951 struct btrfs_device *device;
6953 while (fs_devices) {
6954 mutex_lock(&fs_devices->device_list_mutex);
6955 list_for_each_entry(device, &fs_devices->devices, dev_list)
6956 device->fs_info = fs_info;
6957 mutex_unlock(&fs_devices->device_list_mutex);
6959 fs_devices = fs_devices->seed;
6963 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6967 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6968 btrfs_dev_stat_reset(dev, i);
6971 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6973 struct btrfs_key key;
6974 struct btrfs_key found_key;
6975 struct btrfs_root *dev_root = fs_info->dev_root;
6976 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6977 struct extent_buffer *eb;
6980 struct btrfs_device *device;
6981 struct btrfs_path *path = NULL;
6984 path = btrfs_alloc_path();
6990 mutex_lock(&fs_devices->device_list_mutex);
6991 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6993 struct btrfs_dev_stats_item *ptr;
6995 key.objectid = BTRFS_DEV_STATS_OBJECTID;
6996 key.type = BTRFS_PERSISTENT_ITEM_KEY;
6997 key.offset = device->devid;
6998 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
7000 __btrfs_reset_dev_stats(device);
7001 device->dev_stats_valid = 1;
7002 btrfs_release_path(path);
7005 slot = path->slots[0];
7006 eb = path->nodes[0];
7007 btrfs_item_key_to_cpu(eb, &found_key, slot);
7008 item_size = btrfs_item_size_nr(eb, slot);
7010 ptr = btrfs_item_ptr(eb, slot,
7011 struct btrfs_dev_stats_item);
7013 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7014 if (item_size >= (1 + i) * sizeof(__le64))
7015 btrfs_dev_stat_set(device, i,
7016 btrfs_dev_stats_value(eb, ptr, i));
7018 btrfs_dev_stat_reset(device, i);
7021 device->dev_stats_valid = 1;
7022 btrfs_dev_stat_print_on_load(device);
7023 btrfs_release_path(path);
7025 mutex_unlock(&fs_devices->device_list_mutex);
7028 btrfs_free_path(path);
7029 return ret < 0 ? ret : 0;
7032 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
7033 struct btrfs_fs_info *fs_info,
7034 struct btrfs_device *device)
7036 struct btrfs_root *dev_root = fs_info->dev_root;
7037 struct btrfs_path *path;
7038 struct btrfs_key key;
7039 struct extent_buffer *eb;
7040 struct btrfs_dev_stats_item *ptr;
7044 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7045 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7046 key.offset = device->devid;
7048 path = btrfs_alloc_path();
7051 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
7053 btrfs_warn_in_rcu(fs_info,
7054 "error %d while searching for dev_stats item for device %s",
7055 ret, rcu_str_deref(device->name));
7060 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
7061 /* need to delete old one and insert a new one */
7062 ret = btrfs_del_item(trans, dev_root, path);
7064 btrfs_warn_in_rcu(fs_info,
7065 "delete too small dev_stats item for device %s failed %d",
7066 rcu_str_deref(device->name), ret);
7073 /* need to insert a new item */
7074 btrfs_release_path(path);
7075 ret = btrfs_insert_empty_item(trans, dev_root, path,
7076 &key, sizeof(*ptr));
7078 btrfs_warn_in_rcu(fs_info,
7079 "insert dev_stats item for device %s failed %d",
7080 rcu_str_deref(device->name), ret);
7085 eb = path->nodes[0];
7086 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7087 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7088 btrfs_set_dev_stats_value(eb, ptr, i,
7089 btrfs_dev_stat_read(device, i));
7090 btrfs_mark_buffer_dirty(eb);
7093 btrfs_free_path(path);
7098 * called from commit_transaction. Writes all changed device stats to disk.
7100 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
7101 struct btrfs_fs_info *fs_info)
7103 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7104 struct btrfs_device *device;
7108 mutex_lock(&fs_devices->device_list_mutex);
7109 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7110 stats_cnt = atomic_read(&device->dev_stats_ccnt);
7111 if (!device->dev_stats_valid || stats_cnt == 0)
7116 * There is a LOAD-LOAD control dependency between the value of
7117 * dev_stats_ccnt and updating the on-disk values which requires
7118 * reading the in-memory counters. Such control dependencies
7119 * require explicit read memory barriers.
7121 * This memory barriers pairs with smp_mb__before_atomic in
7122 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full
7123 * barrier implied by atomic_xchg in
7124 * btrfs_dev_stats_read_and_reset
7128 ret = update_dev_stat_item(trans, fs_info, device);
7130 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7132 mutex_unlock(&fs_devices->device_list_mutex);
7137 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7139 btrfs_dev_stat_inc(dev, index);
7140 btrfs_dev_stat_print_on_error(dev);
7143 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
7145 if (!dev->dev_stats_valid)
7147 btrfs_err_rl_in_rcu(dev->fs_info,
7148 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7149 rcu_str_deref(dev->name),
7150 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7151 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7152 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7153 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7154 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7157 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7161 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7162 if (btrfs_dev_stat_read(dev, i) != 0)
7164 if (i == BTRFS_DEV_STAT_VALUES_MAX)
7165 return; /* all values == 0, suppress message */
7167 btrfs_info_in_rcu(dev->fs_info,
7168 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7169 rcu_str_deref(dev->name),
7170 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7171 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7172 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7173 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7174 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7177 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
7178 struct btrfs_ioctl_get_dev_stats *stats)
7180 struct btrfs_device *dev;
7181 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7184 mutex_lock(&fs_devices->device_list_mutex);
7185 dev = btrfs_find_device(fs_info, stats->devid, NULL, NULL);
7186 mutex_unlock(&fs_devices->device_list_mutex);
7189 btrfs_warn(fs_info, "get dev_stats failed, device not found");
7191 } else if (!dev->dev_stats_valid) {
7192 btrfs_warn(fs_info, "get dev_stats failed, not yet valid");
7194 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7195 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7196 if (stats->nr_items > i)
7198 btrfs_dev_stat_read_and_reset(dev, i);
7200 btrfs_dev_stat_reset(dev, i);
7203 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7204 if (stats->nr_items > i)
7205 stats->values[i] = btrfs_dev_stat_read(dev, i);
7207 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7208 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7212 void btrfs_scratch_superblocks(struct block_device *bdev, const char *device_path)
7214 struct buffer_head *bh;
7215 struct btrfs_super_block *disk_super;
7221 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX;
7224 if (btrfs_read_dev_one_super(bdev, copy_num, &bh))
7227 disk_super = (struct btrfs_super_block *)bh->b_data;
7229 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
7230 set_buffer_dirty(bh);
7231 sync_dirty_buffer(bh);
7235 /* Notify udev that device has changed */
7236 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
7238 /* Update ctime/mtime for device path for libblkid */
7239 update_dev_time(device_path);
7243 * Update the size of all devices, which is used for writing out the
7246 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
7248 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7249 struct btrfs_device *curr, *next;
7251 if (list_empty(&fs_devices->resized_devices))
7254 mutex_lock(&fs_devices->device_list_mutex);
7255 mutex_lock(&fs_info->chunk_mutex);
7256 list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
7258 list_del_init(&curr->resized_list);
7259 curr->commit_total_bytes = curr->disk_total_bytes;
7261 mutex_unlock(&fs_info->chunk_mutex);
7262 mutex_unlock(&fs_devices->device_list_mutex);
7265 /* Must be invoked during the transaction commit */
7266 void btrfs_update_commit_device_bytes_used(struct btrfs_fs_info *fs_info,
7267 struct btrfs_transaction *transaction)
7269 struct extent_map *em;
7270 struct map_lookup *map;
7271 struct btrfs_device *dev;
7274 if (list_empty(&transaction->pending_chunks))
7277 /* In order to kick the device replace finish process */
7278 mutex_lock(&fs_info->chunk_mutex);
7279 list_for_each_entry(em, &transaction->pending_chunks, list) {
7280 map = em->map_lookup;
7282 for (i = 0; i < map->num_stripes; i++) {
7283 dev = map->stripes[i].dev;
7284 dev->commit_bytes_used = dev->bytes_used;
7287 mutex_unlock(&fs_info->chunk_mutex);
7290 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
7292 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7293 while (fs_devices) {
7294 fs_devices->fs_info = fs_info;
7295 fs_devices = fs_devices->seed;
7299 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
7301 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7302 while (fs_devices) {
7303 fs_devices->fs_info = NULL;
7304 fs_devices = fs_devices->seed;