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);
239 bio_get(dev->flush_bio);
241 INIT_LIST_HEAD(&dev->dev_list);
242 INIT_LIST_HEAD(&dev->dev_alloc_list);
243 INIT_LIST_HEAD(&dev->resized_list);
245 spin_lock_init(&dev->io_lock);
247 spin_lock_init(&dev->reada_lock);
248 atomic_set(&dev->reada_in_flight, 0);
249 atomic_set(&dev->dev_stats_ccnt, 0);
250 btrfs_device_data_ordered_init(dev);
251 INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
252 INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
258 * Find a device specified by @devid or @uuid in the list of @fs_devices, or
261 * If devid and uuid are both specified, the match must be exact, otherwise
262 * only devid is used.
264 static struct btrfs_device *find_device(struct btrfs_fs_devices *fs_devices,
265 u64 devid, const u8 *uuid)
267 struct list_head *head = &fs_devices->devices;
268 struct btrfs_device *dev;
270 list_for_each_entry(dev, head, dev_list) {
271 if (dev->devid == devid &&
272 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
279 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
281 struct btrfs_fs_devices *fs_devices;
283 list_for_each_entry(fs_devices, &fs_uuids, list) {
284 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
291 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
292 int flush, struct block_device **bdev,
293 struct buffer_head **bh)
297 *bdev = blkdev_get_by_path(device_path, flags, holder);
300 ret = PTR_ERR(*bdev);
305 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
306 ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE);
308 blkdev_put(*bdev, flags);
311 invalidate_bdev(*bdev);
312 *bh = btrfs_read_dev_super(*bdev);
315 blkdev_put(*bdev, flags);
327 static void requeue_list(struct btrfs_pending_bios *pending_bios,
328 struct bio *head, struct bio *tail)
331 struct bio *old_head;
333 old_head = pending_bios->head;
334 pending_bios->head = head;
335 if (pending_bios->tail)
336 tail->bi_next = old_head;
338 pending_bios->tail = tail;
342 * we try to collect pending bios for a device so we don't get a large
343 * number of procs sending bios down to the same device. This greatly
344 * improves the schedulers ability to collect and merge the bios.
346 * But, it also turns into a long list of bios to process and that is sure
347 * to eventually make the worker thread block. The solution here is to
348 * make some progress and then put this work struct back at the end of
349 * the list if the block device is congested. This way, multiple devices
350 * can make progress from a single worker thread.
352 static noinline void run_scheduled_bios(struct btrfs_device *device)
354 struct btrfs_fs_info *fs_info = device->fs_info;
356 struct backing_dev_info *bdi;
357 struct btrfs_pending_bios *pending_bios;
361 unsigned long num_run;
362 unsigned long batch_run = 0;
364 unsigned long last_waited = 0;
366 int sync_pending = 0;
367 struct blk_plug plug;
370 * this function runs all the bios we've collected for
371 * a particular device. We don't want to wander off to
372 * another device without first sending all of these down.
373 * So, setup a plug here and finish it off before we return
375 blk_start_plug(&plug);
377 bdi = device->bdev->bd_bdi;
378 limit = btrfs_async_submit_limit(fs_info);
379 limit = limit * 2 / 3;
382 spin_lock(&device->io_lock);
387 /* take all the bios off the list at once and process them
388 * later on (without the lock held). But, remember the
389 * tail and other pointers so the bios can be properly reinserted
390 * into the list if we hit congestion
392 if (!force_reg && device->pending_sync_bios.head) {
393 pending_bios = &device->pending_sync_bios;
396 pending_bios = &device->pending_bios;
400 pending = pending_bios->head;
401 tail = pending_bios->tail;
402 WARN_ON(pending && !tail);
405 * if pending was null this time around, no bios need processing
406 * at all and we can stop. Otherwise it'll loop back up again
407 * and do an additional check so no bios are missed.
409 * device->running_pending is used to synchronize with the
412 if (device->pending_sync_bios.head == NULL &&
413 device->pending_bios.head == NULL) {
415 device->running_pending = 0;
418 device->running_pending = 1;
421 pending_bios->head = NULL;
422 pending_bios->tail = NULL;
424 spin_unlock(&device->io_lock);
429 /* we want to work on both lists, but do more bios on the
430 * sync list than the regular list
433 pending_bios != &device->pending_sync_bios &&
434 device->pending_sync_bios.head) ||
435 (num_run > 64 && pending_bios == &device->pending_sync_bios &&
436 device->pending_bios.head)) {
437 spin_lock(&device->io_lock);
438 requeue_list(pending_bios, pending, tail);
443 pending = pending->bi_next;
447 * atomic_dec_return implies a barrier for waitqueue_active
449 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
450 waitqueue_active(&fs_info->async_submit_wait))
451 wake_up(&fs_info->async_submit_wait);
453 BUG_ON(atomic_read(&cur->__bi_cnt) == 0);
456 * if we're doing the sync list, record that our
457 * plug has some sync requests on it
459 * If we're doing the regular list and there are
460 * sync requests sitting around, unplug before
463 if (pending_bios == &device->pending_sync_bios) {
465 } else if (sync_pending) {
466 blk_finish_plug(&plug);
467 blk_start_plug(&plug);
471 btrfsic_submit_bio(cur);
478 * we made progress, there is more work to do and the bdi
479 * is now congested. Back off and let other work structs
482 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
483 fs_info->fs_devices->open_devices > 1) {
484 struct io_context *ioc;
486 ioc = current->io_context;
489 * the main goal here is that we don't want to
490 * block if we're going to be able to submit
491 * more requests without blocking.
493 * This code does two great things, it pokes into
494 * the elevator code from a filesystem _and_
495 * it makes assumptions about how batching works.
497 if (ioc && ioc->nr_batch_requests > 0 &&
498 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
500 ioc->last_waited == last_waited)) {
502 * we want to go through our batch of
503 * requests and stop. So, we copy out
504 * the ioc->last_waited time and test
505 * against it before looping
507 last_waited = ioc->last_waited;
511 spin_lock(&device->io_lock);
512 requeue_list(pending_bios, pending, tail);
513 device->running_pending = 1;
515 spin_unlock(&device->io_lock);
516 btrfs_queue_work(fs_info->submit_workers,
520 /* unplug every 64 requests just for good measure */
521 if (batch_run % 64 == 0) {
522 blk_finish_plug(&plug);
523 blk_start_plug(&plug);
532 spin_lock(&device->io_lock);
533 if (device->pending_bios.head || device->pending_sync_bios.head)
535 spin_unlock(&device->io_lock);
538 blk_finish_plug(&plug);
541 static void pending_bios_fn(struct btrfs_work *work)
543 struct btrfs_device *device;
545 device = container_of(work, struct btrfs_device, work);
546 run_scheduled_bios(device);
550 void btrfs_free_stale_device(struct btrfs_device *cur_dev)
552 struct btrfs_fs_devices *fs_devs;
553 struct btrfs_device *dev;
558 list_for_each_entry(fs_devs, &fs_uuids, list) {
563 if (fs_devs->seeding)
566 list_for_each_entry(dev, &fs_devs->devices, dev_list) {
574 * Todo: This won't be enough. What if the same device
575 * comes back (with new uuid and) with its mapper path?
576 * But for now, this does help as mostly an admin will
577 * either use mapper or non mapper path throughout.
580 del = strcmp(rcu_str_deref(dev->name),
581 rcu_str_deref(cur_dev->name));
588 /* delete the stale device */
589 if (fs_devs->num_devices == 1) {
590 btrfs_sysfs_remove_fsid(fs_devs);
591 list_del(&fs_devs->list);
592 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_tree, u64 chunk_objectid,
1575 u64 chunk_offset, u64 start, u64 num_bytes)
1578 struct btrfs_path *path;
1579 struct btrfs_fs_info *fs_info = device->fs_info;
1580 struct btrfs_root *root = fs_info->dev_root;
1581 struct btrfs_dev_extent *extent;
1582 struct extent_buffer *leaf;
1583 struct btrfs_key key;
1585 WARN_ON(!device->in_fs_metadata);
1586 WARN_ON(device->is_tgtdev_for_dev_replace);
1587 path = btrfs_alloc_path();
1591 key.objectid = device->devid;
1593 key.type = BTRFS_DEV_EXTENT_KEY;
1594 ret = btrfs_insert_empty_item(trans, root, path, &key,
1599 leaf = path->nodes[0];
1600 extent = btrfs_item_ptr(leaf, path->slots[0],
1601 struct btrfs_dev_extent);
1602 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1603 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1604 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1606 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1607 btrfs_mark_buffer_dirty(leaf);
1609 btrfs_free_path(path);
1613 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1615 struct extent_map_tree *em_tree;
1616 struct extent_map *em;
1620 em_tree = &fs_info->mapping_tree.map_tree;
1621 read_lock(&em_tree->lock);
1622 n = rb_last(&em_tree->map);
1624 em = rb_entry(n, struct extent_map, rb_node);
1625 ret = em->start + em->len;
1627 read_unlock(&em_tree->lock);
1632 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1636 struct btrfs_key key;
1637 struct btrfs_key found_key;
1638 struct btrfs_path *path;
1640 path = btrfs_alloc_path();
1644 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1645 key.type = BTRFS_DEV_ITEM_KEY;
1646 key.offset = (u64)-1;
1648 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1652 BUG_ON(ret == 0); /* Corruption */
1654 ret = btrfs_previous_item(fs_info->chunk_root, path,
1655 BTRFS_DEV_ITEMS_OBJECTID,
1656 BTRFS_DEV_ITEM_KEY);
1660 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1662 *devid_ret = found_key.offset + 1;
1666 btrfs_free_path(path);
1671 * the device information is stored in the chunk root
1672 * the btrfs_device struct should be fully filled in
1674 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1675 struct btrfs_fs_info *fs_info,
1676 struct btrfs_device *device)
1678 struct btrfs_root *root = fs_info->chunk_root;
1680 struct btrfs_path *path;
1681 struct btrfs_dev_item *dev_item;
1682 struct extent_buffer *leaf;
1683 struct btrfs_key key;
1686 path = btrfs_alloc_path();
1690 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1691 key.type = BTRFS_DEV_ITEM_KEY;
1692 key.offset = device->devid;
1694 ret = btrfs_insert_empty_item(trans, root, path, &key,
1699 leaf = path->nodes[0];
1700 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1702 btrfs_set_device_id(leaf, dev_item, device->devid);
1703 btrfs_set_device_generation(leaf, dev_item, 0);
1704 btrfs_set_device_type(leaf, dev_item, device->type);
1705 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1706 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1707 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1708 btrfs_set_device_total_bytes(leaf, dev_item,
1709 btrfs_device_get_disk_total_bytes(device));
1710 btrfs_set_device_bytes_used(leaf, dev_item,
1711 btrfs_device_get_bytes_used(device));
1712 btrfs_set_device_group(leaf, dev_item, 0);
1713 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1714 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1715 btrfs_set_device_start_offset(leaf, dev_item, 0);
1717 ptr = btrfs_device_uuid(dev_item);
1718 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1719 ptr = btrfs_device_fsid(dev_item);
1720 write_extent_buffer(leaf, fs_info->fsid, ptr, BTRFS_FSID_SIZE);
1721 btrfs_mark_buffer_dirty(leaf);
1725 btrfs_free_path(path);
1730 * Function to update ctime/mtime for a given device path.
1731 * Mainly used for ctime/mtime based probe like libblkid.
1733 static void update_dev_time(const char *path_name)
1737 filp = filp_open(path_name, O_RDWR, 0);
1740 file_update_time(filp);
1741 filp_close(filp, NULL);
1744 static int btrfs_rm_dev_item(struct btrfs_fs_info *fs_info,
1745 struct btrfs_device *device)
1747 struct btrfs_root *root = fs_info->chunk_root;
1749 struct btrfs_path *path;
1750 struct btrfs_key key;
1751 struct btrfs_trans_handle *trans;
1753 path = btrfs_alloc_path();
1757 trans = btrfs_start_transaction(root, 0);
1758 if (IS_ERR(trans)) {
1759 btrfs_free_path(path);
1760 return PTR_ERR(trans);
1762 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1763 key.type = BTRFS_DEV_ITEM_KEY;
1764 key.offset = device->devid;
1766 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1775 ret = btrfs_del_item(trans, root, path);
1779 btrfs_free_path(path);
1780 btrfs_commit_transaction(trans);
1785 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1786 * filesystem. It's up to the caller to adjust that number regarding eg. device
1789 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1797 seq = read_seqbegin(&fs_info->profiles_lock);
1799 all_avail = fs_info->avail_data_alloc_bits |
1800 fs_info->avail_system_alloc_bits |
1801 fs_info->avail_metadata_alloc_bits;
1802 } while (read_seqretry(&fs_info->profiles_lock, seq));
1804 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1805 if (!(all_avail & btrfs_raid_group[i]))
1808 if (num_devices < btrfs_raid_array[i].devs_min) {
1809 int ret = btrfs_raid_mindev_error[i];
1819 struct btrfs_device *btrfs_find_next_active_device(struct btrfs_fs_devices *fs_devs,
1820 struct btrfs_device *device)
1822 struct btrfs_device *next_device;
1824 list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1825 if (next_device != device &&
1826 !next_device->missing && next_device->bdev)
1834 * Helper function to check if the given device is part of s_bdev / latest_bdev
1835 * and replace it with the provided or the next active device, in the context
1836 * where this function called, there should be always be another device (or
1837 * this_dev) which is active.
1839 void btrfs_assign_next_active_device(struct btrfs_fs_info *fs_info,
1840 struct btrfs_device *device, struct btrfs_device *this_dev)
1842 struct btrfs_device *next_device;
1845 next_device = this_dev;
1847 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
1849 ASSERT(next_device);
1851 if (fs_info->sb->s_bdev &&
1852 (fs_info->sb->s_bdev == device->bdev))
1853 fs_info->sb->s_bdev = next_device->bdev;
1855 if (fs_info->fs_devices->latest_bdev == device->bdev)
1856 fs_info->fs_devices->latest_bdev = next_device->bdev;
1859 int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path,
1862 struct btrfs_device *device;
1863 struct btrfs_fs_devices *cur_devices;
1867 mutex_lock(&uuid_mutex);
1869 num_devices = fs_info->fs_devices->num_devices;
1870 btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
1871 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
1872 WARN_ON(num_devices < 1);
1875 btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
1877 ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
1881 ret = btrfs_find_device_by_devspec(fs_info, devid, device_path,
1886 if (device->is_tgtdev_for_dev_replace) {
1887 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1891 if (device->writeable && fs_info->fs_devices->rw_devices == 1) {
1892 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1896 if (device->writeable) {
1897 mutex_lock(&fs_info->chunk_mutex);
1898 list_del_init(&device->dev_alloc_list);
1899 device->fs_devices->rw_devices--;
1900 mutex_unlock(&fs_info->chunk_mutex);
1903 mutex_unlock(&uuid_mutex);
1904 ret = btrfs_shrink_device(device, 0);
1905 mutex_lock(&uuid_mutex);
1910 * TODO: the superblock still includes this device in its num_devices
1911 * counter although write_all_supers() is not locked out. This
1912 * could give a filesystem state which requires a degraded mount.
1914 ret = btrfs_rm_dev_item(fs_info, device);
1918 device->in_fs_metadata = 0;
1919 btrfs_scrub_cancel_dev(fs_info, device);
1922 * the device list mutex makes sure that we don't change
1923 * the device list while someone else is writing out all
1924 * the device supers. Whoever is writing all supers, should
1925 * lock the device list mutex before getting the number of
1926 * devices in the super block (super_copy). Conversely,
1927 * whoever updates the number of devices in the super block
1928 * (super_copy) should hold the device list mutex.
1931 cur_devices = device->fs_devices;
1932 mutex_lock(&fs_info->fs_devices->device_list_mutex);
1933 list_del_rcu(&device->dev_list);
1935 device->fs_devices->num_devices--;
1936 device->fs_devices->total_devices--;
1938 if (device->missing)
1939 device->fs_devices->missing_devices--;
1941 btrfs_assign_next_active_device(fs_info, device, NULL);
1944 device->fs_devices->open_devices--;
1945 /* remove sysfs entry */
1946 btrfs_sysfs_rm_device_link(fs_info->fs_devices, device);
1949 num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
1950 btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
1951 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1954 * at this point, the device is zero sized and detached from
1955 * the devices list. All that's left is to zero out the old
1956 * supers and free the device.
1958 if (device->writeable)
1959 btrfs_scratch_superblocks(device->bdev, device->name->str);
1961 btrfs_close_bdev(device);
1962 call_rcu(&device->rcu, free_device);
1964 if (cur_devices->open_devices == 0) {
1965 struct btrfs_fs_devices *fs_devices;
1966 fs_devices = fs_info->fs_devices;
1967 while (fs_devices) {
1968 if (fs_devices->seed == cur_devices) {
1969 fs_devices->seed = cur_devices->seed;
1972 fs_devices = fs_devices->seed;
1974 cur_devices->seed = NULL;
1975 __btrfs_close_devices(cur_devices);
1976 free_fs_devices(cur_devices);
1980 mutex_unlock(&uuid_mutex);
1984 if (device->writeable) {
1985 mutex_lock(&fs_info->chunk_mutex);
1986 list_add(&device->dev_alloc_list,
1987 &fs_info->fs_devices->alloc_list);
1988 device->fs_devices->rw_devices++;
1989 mutex_unlock(&fs_info->chunk_mutex);
1994 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
1995 struct btrfs_device *srcdev)
1997 struct btrfs_fs_devices *fs_devices;
1999 WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
2002 * in case of fs with no seed, srcdev->fs_devices will point
2003 * to fs_devices of fs_info. However when the dev being replaced is
2004 * a seed dev it will point to the seed's local fs_devices. In short
2005 * srcdev will have its correct fs_devices in both the cases.
2007 fs_devices = srcdev->fs_devices;
2009 list_del_rcu(&srcdev->dev_list);
2010 list_del_rcu(&srcdev->dev_alloc_list);
2011 fs_devices->num_devices--;
2012 if (srcdev->missing)
2013 fs_devices->missing_devices--;
2015 if (srcdev->writeable)
2016 fs_devices->rw_devices--;
2019 fs_devices->open_devices--;
2022 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
2023 struct btrfs_device *srcdev)
2025 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2027 if (srcdev->writeable) {
2028 /* zero out the old super if it is writable */
2029 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
2032 btrfs_close_bdev(srcdev);
2034 call_rcu(&srcdev->rcu, free_device);
2037 * unless fs_devices is seed fs, num_devices shouldn't go
2040 BUG_ON(!fs_devices->num_devices && !fs_devices->seeding);
2042 /* if this is no devs we rather delete the fs_devices */
2043 if (!fs_devices->num_devices) {
2044 struct btrfs_fs_devices *tmp_fs_devices;
2046 tmp_fs_devices = fs_info->fs_devices;
2047 while (tmp_fs_devices) {
2048 if (tmp_fs_devices->seed == fs_devices) {
2049 tmp_fs_devices->seed = fs_devices->seed;
2052 tmp_fs_devices = tmp_fs_devices->seed;
2054 fs_devices->seed = NULL;
2055 __btrfs_close_devices(fs_devices);
2056 free_fs_devices(fs_devices);
2060 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2061 struct btrfs_device *tgtdev)
2063 mutex_lock(&uuid_mutex);
2065 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2067 btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev);
2070 fs_info->fs_devices->open_devices--;
2072 fs_info->fs_devices->num_devices--;
2074 btrfs_assign_next_active_device(fs_info, tgtdev, NULL);
2076 list_del_rcu(&tgtdev->dev_list);
2078 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2079 mutex_unlock(&uuid_mutex);
2082 * The update_dev_time() with in btrfs_scratch_superblocks()
2083 * may lead to a call to btrfs_show_devname() which will try
2084 * to hold device_list_mutex. And here this device
2085 * is already out of device list, so we don't have to hold
2086 * the device_list_mutex lock.
2088 btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
2090 btrfs_close_bdev(tgtdev);
2091 call_rcu(&tgtdev->rcu, free_device);
2094 static int btrfs_find_device_by_path(struct btrfs_fs_info *fs_info,
2095 const char *device_path,
2096 struct btrfs_device **device)
2099 struct btrfs_super_block *disk_super;
2102 struct block_device *bdev;
2103 struct buffer_head *bh;
2106 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2107 fs_info->bdev_holder, 0, &bdev, &bh);
2110 disk_super = (struct btrfs_super_block *)bh->b_data;
2111 devid = btrfs_stack_device_id(&disk_super->dev_item);
2112 dev_uuid = disk_super->dev_item.uuid;
2113 *device = btrfs_find_device(fs_info, devid, dev_uuid, disk_super->fsid);
2117 blkdev_put(bdev, FMODE_READ);
2121 int btrfs_find_device_missing_or_by_path(struct btrfs_fs_info *fs_info,
2122 const char *device_path,
2123 struct btrfs_device **device)
2126 if (strcmp(device_path, "missing") == 0) {
2127 struct list_head *devices;
2128 struct btrfs_device *tmp;
2130 devices = &fs_info->fs_devices->devices;
2132 * It is safe to read the devices since the volume_mutex
2133 * is held by the caller.
2135 list_for_each_entry(tmp, devices, dev_list) {
2136 if (tmp->in_fs_metadata && !tmp->bdev) {
2143 return BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2147 return btrfs_find_device_by_path(fs_info, device_path, device);
2152 * Lookup a device given by device id, or the path if the id is 0.
2154 int btrfs_find_device_by_devspec(struct btrfs_fs_info *fs_info, u64 devid,
2155 const char *devpath,
2156 struct btrfs_device **device)
2162 *device = btrfs_find_device(fs_info, devid, NULL, NULL);
2166 if (!devpath || !devpath[0])
2169 ret = btrfs_find_device_missing_or_by_path(fs_info, devpath,
2176 * does all the dirty work required for changing file system's UUID.
2178 static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info)
2180 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2181 struct btrfs_fs_devices *old_devices;
2182 struct btrfs_fs_devices *seed_devices;
2183 struct btrfs_super_block *disk_super = fs_info->super_copy;
2184 struct btrfs_device *device;
2187 BUG_ON(!mutex_is_locked(&uuid_mutex));
2188 if (!fs_devices->seeding)
2191 seed_devices = alloc_fs_devices(NULL);
2192 if (IS_ERR(seed_devices))
2193 return PTR_ERR(seed_devices);
2195 old_devices = clone_fs_devices(fs_devices);
2196 if (IS_ERR(old_devices)) {
2197 kfree(seed_devices);
2198 return PTR_ERR(old_devices);
2201 list_add(&old_devices->list, &fs_uuids);
2203 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2204 seed_devices->opened = 1;
2205 INIT_LIST_HEAD(&seed_devices->devices);
2206 INIT_LIST_HEAD(&seed_devices->alloc_list);
2207 mutex_init(&seed_devices->device_list_mutex);
2209 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2210 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2212 list_for_each_entry(device, &seed_devices->devices, dev_list)
2213 device->fs_devices = seed_devices;
2215 mutex_lock(&fs_info->chunk_mutex);
2216 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2217 mutex_unlock(&fs_info->chunk_mutex);
2219 fs_devices->seeding = 0;
2220 fs_devices->num_devices = 0;
2221 fs_devices->open_devices = 0;
2222 fs_devices->missing_devices = 0;
2223 fs_devices->rotating = 0;
2224 fs_devices->seed = seed_devices;
2226 generate_random_uuid(fs_devices->fsid);
2227 memcpy(fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2228 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2229 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2231 super_flags = btrfs_super_flags(disk_super) &
2232 ~BTRFS_SUPER_FLAG_SEEDING;
2233 btrfs_set_super_flags(disk_super, super_flags);
2239 * Store the expected generation for seed devices in device items.
2241 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2242 struct btrfs_fs_info *fs_info)
2244 struct btrfs_root *root = fs_info->chunk_root;
2245 struct btrfs_path *path;
2246 struct extent_buffer *leaf;
2247 struct btrfs_dev_item *dev_item;
2248 struct btrfs_device *device;
2249 struct btrfs_key key;
2250 u8 fs_uuid[BTRFS_FSID_SIZE];
2251 u8 dev_uuid[BTRFS_UUID_SIZE];
2255 path = btrfs_alloc_path();
2259 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2261 key.type = BTRFS_DEV_ITEM_KEY;
2264 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2268 leaf = path->nodes[0];
2270 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2271 ret = btrfs_next_leaf(root, path);
2276 leaf = path->nodes[0];
2277 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2278 btrfs_release_path(path);
2282 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2283 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2284 key.type != BTRFS_DEV_ITEM_KEY)
2287 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2288 struct btrfs_dev_item);
2289 devid = btrfs_device_id(leaf, dev_item);
2290 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2292 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2294 device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
2295 BUG_ON(!device); /* Logic error */
2297 if (device->fs_devices->seeding) {
2298 btrfs_set_device_generation(leaf, dev_item,
2299 device->generation);
2300 btrfs_mark_buffer_dirty(leaf);
2308 btrfs_free_path(path);
2312 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2314 struct btrfs_root *root = fs_info->dev_root;
2315 struct request_queue *q;
2316 struct btrfs_trans_handle *trans;
2317 struct btrfs_device *device;
2318 struct block_device *bdev;
2319 struct list_head *devices;
2320 struct super_block *sb = fs_info->sb;
2321 struct rcu_string *name;
2323 int seeding_dev = 0;
2326 if ((sb->s_flags & MS_RDONLY) && !fs_info->fs_devices->seeding)
2329 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2330 fs_info->bdev_holder);
2332 return PTR_ERR(bdev);
2334 if (fs_info->fs_devices->seeding) {
2336 down_write(&sb->s_umount);
2337 mutex_lock(&uuid_mutex);
2340 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2342 devices = &fs_info->fs_devices->devices;
2344 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2345 list_for_each_entry(device, devices, dev_list) {
2346 if (device->bdev == bdev) {
2349 &fs_info->fs_devices->device_list_mutex);
2353 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2355 device = btrfs_alloc_device(fs_info, NULL, NULL);
2356 if (IS_ERR(device)) {
2357 /* we can safely leave the fs_devices entry around */
2358 ret = PTR_ERR(device);
2362 name = rcu_string_strdup(device_path, GFP_KERNEL);
2368 rcu_assign_pointer(device->name, name);
2370 trans = btrfs_start_transaction(root, 0);
2371 if (IS_ERR(trans)) {
2372 rcu_string_free(device->name);
2374 ret = PTR_ERR(trans);
2378 q = bdev_get_queue(bdev);
2379 if (blk_queue_discard(q))
2380 device->can_discard = 1;
2381 device->writeable = 1;
2382 device->generation = trans->transid;
2383 device->io_width = fs_info->sectorsize;
2384 device->io_align = fs_info->sectorsize;
2385 device->sector_size = fs_info->sectorsize;
2386 device->total_bytes = round_down(i_size_read(bdev->bd_inode),
2387 fs_info->sectorsize);
2388 device->disk_total_bytes = device->total_bytes;
2389 device->commit_total_bytes = device->total_bytes;
2390 device->fs_info = fs_info;
2391 device->bdev = bdev;
2392 device->in_fs_metadata = 1;
2393 device->is_tgtdev_for_dev_replace = 0;
2394 device->mode = FMODE_EXCL;
2395 device->dev_stats_valid = 1;
2396 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2399 sb->s_flags &= ~MS_RDONLY;
2400 ret = btrfs_prepare_sprout(fs_info);
2401 BUG_ON(ret); /* -ENOMEM */
2404 device->fs_devices = fs_info->fs_devices;
2406 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2407 mutex_lock(&fs_info->chunk_mutex);
2408 list_add_rcu(&device->dev_list, &fs_info->fs_devices->devices);
2409 list_add(&device->dev_alloc_list,
2410 &fs_info->fs_devices->alloc_list);
2411 fs_info->fs_devices->num_devices++;
2412 fs_info->fs_devices->open_devices++;
2413 fs_info->fs_devices->rw_devices++;
2414 fs_info->fs_devices->total_devices++;
2415 fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2417 atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2419 if (!blk_queue_nonrot(q))
2420 fs_info->fs_devices->rotating = 1;
2422 tmp = btrfs_super_total_bytes(fs_info->super_copy);
2423 btrfs_set_super_total_bytes(fs_info->super_copy,
2424 round_down(tmp + device->total_bytes, fs_info->sectorsize));
2426 tmp = btrfs_super_num_devices(fs_info->super_copy);
2427 btrfs_set_super_num_devices(fs_info->super_copy, tmp + 1);
2429 /* add sysfs device entry */
2430 btrfs_sysfs_add_device_link(fs_info->fs_devices, device);
2433 * we've got more storage, clear any full flags on the space
2436 btrfs_clear_space_info_full(fs_info);
2438 mutex_unlock(&fs_info->chunk_mutex);
2439 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2442 mutex_lock(&fs_info->chunk_mutex);
2443 ret = init_first_rw_device(trans, fs_info);
2444 mutex_unlock(&fs_info->chunk_mutex);
2446 btrfs_abort_transaction(trans, ret);
2451 ret = btrfs_add_device(trans, fs_info, device);
2453 btrfs_abort_transaction(trans, ret);
2458 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2460 ret = btrfs_finish_sprout(trans, fs_info);
2462 btrfs_abort_transaction(trans, ret);
2466 /* Sprouting would change fsid of the mounted root,
2467 * so rename the fsid on the sysfs
2469 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2471 if (kobject_rename(&fs_info->fs_devices->fsid_kobj, fsid_buf))
2473 "sysfs: failed to create fsid for sprout");
2476 ret = btrfs_commit_transaction(trans);
2479 mutex_unlock(&uuid_mutex);
2480 up_write(&sb->s_umount);
2482 if (ret) /* transaction commit */
2485 ret = btrfs_relocate_sys_chunks(fs_info);
2487 btrfs_handle_fs_error(fs_info, ret,
2488 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2489 trans = btrfs_attach_transaction(root);
2490 if (IS_ERR(trans)) {
2491 if (PTR_ERR(trans) == -ENOENT)
2493 return PTR_ERR(trans);
2495 ret = btrfs_commit_transaction(trans);
2498 /* Update ctime/mtime for libblkid */
2499 update_dev_time(device_path);
2503 btrfs_end_transaction(trans);
2504 rcu_string_free(device->name);
2505 btrfs_sysfs_rm_device_link(fs_info->fs_devices, device);
2508 blkdev_put(bdev, FMODE_EXCL);
2510 mutex_unlock(&uuid_mutex);
2511 up_write(&sb->s_umount);
2516 int btrfs_init_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2517 const char *device_path,
2518 struct btrfs_device *srcdev,
2519 struct btrfs_device **device_out)
2521 struct request_queue *q;
2522 struct btrfs_device *device;
2523 struct block_device *bdev;
2524 struct list_head *devices;
2525 struct rcu_string *name;
2526 u64 devid = BTRFS_DEV_REPLACE_DEVID;
2530 if (fs_info->fs_devices->seeding) {
2531 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
2535 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2536 fs_info->bdev_holder);
2538 btrfs_err(fs_info, "target device %s is invalid!", device_path);
2539 return PTR_ERR(bdev);
2542 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2544 devices = &fs_info->fs_devices->devices;
2545 list_for_each_entry(device, devices, dev_list) {
2546 if (device->bdev == bdev) {
2548 "target device is in the filesystem!");
2555 if (i_size_read(bdev->bd_inode) <
2556 btrfs_device_get_total_bytes(srcdev)) {
2558 "target device is smaller than source device!");
2564 device = btrfs_alloc_device(NULL, &devid, NULL);
2565 if (IS_ERR(device)) {
2566 ret = PTR_ERR(device);
2570 name = rcu_string_strdup(device_path, GFP_KERNEL);
2576 rcu_assign_pointer(device->name, name);
2578 q = bdev_get_queue(bdev);
2579 if (blk_queue_discard(q))
2580 device->can_discard = 1;
2581 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2582 device->writeable = 1;
2583 device->generation = 0;
2584 device->io_width = fs_info->sectorsize;
2585 device->io_align = fs_info->sectorsize;
2586 device->sector_size = fs_info->sectorsize;
2587 device->total_bytes = btrfs_device_get_total_bytes(srcdev);
2588 device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
2589 device->bytes_used = btrfs_device_get_bytes_used(srcdev);
2590 ASSERT(list_empty(&srcdev->resized_list));
2591 device->commit_total_bytes = srcdev->commit_total_bytes;
2592 device->commit_bytes_used = device->bytes_used;
2593 device->fs_info = fs_info;
2594 device->bdev = bdev;
2595 device->in_fs_metadata = 1;
2596 device->is_tgtdev_for_dev_replace = 1;
2597 device->mode = FMODE_EXCL;
2598 device->dev_stats_valid = 1;
2599 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2600 device->fs_devices = fs_info->fs_devices;
2601 list_add(&device->dev_list, &fs_info->fs_devices->devices);
2602 fs_info->fs_devices->num_devices++;
2603 fs_info->fs_devices->open_devices++;
2604 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2606 *device_out = device;
2610 blkdev_put(bdev, FMODE_EXCL);
2614 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2615 struct btrfs_device *tgtdev)
2617 u32 sectorsize = fs_info->sectorsize;
2619 WARN_ON(fs_info->fs_devices->rw_devices == 0);
2620 tgtdev->io_width = sectorsize;
2621 tgtdev->io_align = sectorsize;
2622 tgtdev->sector_size = sectorsize;
2623 tgtdev->fs_info = fs_info;
2624 tgtdev->in_fs_metadata = 1;
2627 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2628 struct btrfs_device *device)
2631 struct btrfs_path *path;
2632 struct btrfs_root *root = device->fs_info->chunk_root;
2633 struct btrfs_dev_item *dev_item;
2634 struct extent_buffer *leaf;
2635 struct btrfs_key key;
2637 path = btrfs_alloc_path();
2641 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2642 key.type = BTRFS_DEV_ITEM_KEY;
2643 key.offset = device->devid;
2645 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2654 leaf = path->nodes[0];
2655 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2657 btrfs_set_device_id(leaf, dev_item, device->devid);
2658 btrfs_set_device_type(leaf, dev_item, device->type);
2659 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2660 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2661 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2662 btrfs_set_device_total_bytes(leaf, dev_item,
2663 btrfs_device_get_disk_total_bytes(device));
2664 btrfs_set_device_bytes_used(leaf, dev_item,
2665 btrfs_device_get_bytes_used(device));
2666 btrfs_mark_buffer_dirty(leaf);
2669 btrfs_free_path(path);
2673 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2674 struct btrfs_device *device, u64 new_size)
2676 struct btrfs_fs_info *fs_info = device->fs_info;
2677 struct btrfs_super_block *super_copy = fs_info->super_copy;
2678 struct btrfs_fs_devices *fs_devices;
2682 if (!device->writeable)
2685 new_size = round_down(new_size, fs_info->sectorsize);
2687 mutex_lock(&fs_info->chunk_mutex);
2688 old_total = btrfs_super_total_bytes(super_copy);
2689 diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2691 if (new_size <= device->total_bytes ||
2692 device->is_tgtdev_for_dev_replace) {
2693 mutex_unlock(&fs_info->chunk_mutex);
2697 fs_devices = fs_info->fs_devices;
2699 btrfs_set_super_total_bytes(super_copy,
2700 round_down(old_total + diff, fs_info->sectorsize));
2701 device->fs_devices->total_rw_bytes += diff;
2703 btrfs_device_set_total_bytes(device, new_size);
2704 btrfs_device_set_disk_total_bytes(device, new_size);
2705 btrfs_clear_space_info_full(device->fs_info);
2706 if (list_empty(&device->resized_list))
2707 list_add_tail(&device->resized_list,
2708 &fs_devices->resized_devices);
2709 mutex_unlock(&fs_info->chunk_mutex);
2711 return btrfs_update_device(trans, device);
2714 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2715 struct btrfs_fs_info *fs_info, u64 chunk_objectid,
2718 struct btrfs_root *root = fs_info->chunk_root;
2720 struct btrfs_path *path;
2721 struct btrfs_key key;
2723 path = btrfs_alloc_path();
2727 key.objectid = chunk_objectid;
2728 key.offset = chunk_offset;
2729 key.type = BTRFS_CHUNK_ITEM_KEY;
2731 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2734 else if (ret > 0) { /* Logic error or corruption */
2735 btrfs_handle_fs_error(fs_info, -ENOENT,
2736 "Failed lookup while freeing chunk.");
2741 ret = btrfs_del_item(trans, root, path);
2743 btrfs_handle_fs_error(fs_info, ret,
2744 "Failed to delete chunk item.");
2746 btrfs_free_path(path);
2750 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info,
2751 u64 chunk_objectid, u64 chunk_offset)
2753 struct btrfs_super_block *super_copy = fs_info->super_copy;
2754 struct btrfs_disk_key *disk_key;
2755 struct btrfs_chunk *chunk;
2762 struct btrfs_key key;
2764 mutex_lock(&fs_info->chunk_mutex);
2765 array_size = btrfs_super_sys_array_size(super_copy);
2767 ptr = super_copy->sys_chunk_array;
2770 while (cur < array_size) {
2771 disk_key = (struct btrfs_disk_key *)ptr;
2772 btrfs_disk_key_to_cpu(&key, disk_key);
2774 len = sizeof(*disk_key);
2776 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2777 chunk = (struct btrfs_chunk *)(ptr + len);
2778 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2779 len += btrfs_chunk_item_size(num_stripes);
2784 if (key.objectid == chunk_objectid &&
2785 key.offset == chunk_offset) {
2786 memmove(ptr, ptr + len, array_size - (cur + len));
2788 btrfs_set_super_sys_array_size(super_copy, array_size);
2794 mutex_unlock(&fs_info->chunk_mutex);
2798 static struct extent_map *get_chunk_map(struct btrfs_fs_info *fs_info,
2799 u64 logical, u64 length)
2801 struct extent_map_tree *em_tree;
2802 struct extent_map *em;
2804 em_tree = &fs_info->mapping_tree.map_tree;
2805 read_lock(&em_tree->lock);
2806 em = lookup_extent_mapping(em_tree, logical, length);
2807 read_unlock(&em_tree->lock);
2810 btrfs_crit(fs_info, "unable to find logical %llu length %llu",
2812 return ERR_PTR(-EINVAL);
2815 if (em->start > logical || em->start + em->len < logical) {
2817 "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
2818 logical, length, em->start, em->start + em->len);
2819 free_extent_map(em);
2820 return ERR_PTR(-EINVAL);
2823 /* callers are responsible for dropping em's ref. */
2827 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2828 struct btrfs_fs_info *fs_info, u64 chunk_offset)
2830 struct extent_map *em;
2831 struct map_lookup *map;
2832 u64 dev_extent_len = 0;
2833 u64 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2835 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2837 em = get_chunk_map(fs_info, chunk_offset, 1);
2840 * This is a logic error, but we don't want to just rely on the
2841 * user having built with ASSERT enabled, so if ASSERT doesn't
2842 * do anything we still error out.
2847 map = em->map_lookup;
2848 mutex_lock(&fs_info->chunk_mutex);
2849 check_system_chunk(trans, fs_info, map->type);
2850 mutex_unlock(&fs_info->chunk_mutex);
2853 * Take the device list mutex to prevent races with the final phase of
2854 * a device replace operation that replaces the device object associated
2855 * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2857 mutex_lock(&fs_devices->device_list_mutex);
2858 for (i = 0; i < map->num_stripes; i++) {
2859 struct btrfs_device *device = map->stripes[i].dev;
2860 ret = btrfs_free_dev_extent(trans, device,
2861 map->stripes[i].physical,
2864 mutex_unlock(&fs_devices->device_list_mutex);
2865 btrfs_abort_transaction(trans, ret);
2869 if (device->bytes_used > 0) {
2870 mutex_lock(&fs_info->chunk_mutex);
2871 btrfs_device_set_bytes_used(device,
2872 device->bytes_used - dev_extent_len);
2873 atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
2874 btrfs_clear_space_info_full(fs_info);
2875 mutex_unlock(&fs_info->chunk_mutex);
2878 if (map->stripes[i].dev) {
2879 ret = btrfs_update_device(trans, map->stripes[i].dev);
2881 mutex_unlock(&fs_devices->device_list_mutex);
2882 btrfs_abort_transaction(trans, ret);
2887 mutex_unlock(&fs_devices->device_list_mutex);
2889 ret = btrfs_free_chunk(trans, fs_info, chunk_objectid, chunk_offset);
2891 btrfs_abort_transaction(trans, ret);
2895 trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
2897 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2898 ret = btrfs_del_sys_chunk(fs_info, chunk_objectid,
2901 btrfs_abort_transaction(trans, ret);
2906 ret = btrfs_remove_block_group(trans, fs_info, chunk_offset, em);
2908 btrfs_abort_transaction(trans, ret);
2914 free_extent_map(em);
2918 static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2920 struct btrfs_root *root = fs_info->chunk_root;
2921 struct btrfs_trans_handle *trans;
2925 * Prevent races with automatic removal of unused block groups.
2926 * After we relocate and before we remove the chunk with offset
2927 * chunk_offset, automatic removal of the block group can kick in,
2928 * resulting in a failure when calling btrfs_remove_chunk() below.
2930 * Make sure to acquire this mutex before doing a tree search (dev
2931 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2932 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2933 * we release the path used to search the chunk/dev tree and before
2934 * the current task acquires this mutex and calls us.
2936 ASSERT(mutex_is_locked(&fs_info->delete_unused_bgs_mutex));
2938 ret = btrfs_can_relocate(fs_info, chunk_offset);
2942 /* step one, relocate all the extents inside this chunk */
2943 btrfs_scrub_pause(fs_info);
2944 ret = btrfs_relocate_block_group(fs_info, chunk_offset);
2945 btrfs_scrub_continue(fs_info);
2949 trans = btrfs_start_trans_remove_block_group(root->fs_info,
2951 if (IS_ERR(trans)) {
2952 ret = PTR_ERR(trans);
2953 btrfs_handle_fs_error(root->fs_info, ret, NULL);
2958 * step two, delete the device extents and the
2959 * chunk tree entries
2961 ret = btrfs_remove_chunk(trans, fs_info, chunk_offset);
2962 btrfs_end_transaction(trans);
2966 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
2968 struct btrfs_root *chunk_root = fs_info->chunk_root;
2969 struct btrfs_path *path;
2970 struct extent_buffer *leaf;
2971 struct btrfs_chunk *chunk;
2972 struct btrfs_key key;
2973 struct btrfs_key found_key;
2975 bool retried = false;
2979 path = btrfs_alloc_path();
2984 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2985 key.offset = (u64)-1;
2986 key.type = BTRFS_CHUNK_ITEM_KEY;
2989 mutex_lock(&fs_info->delete_unused_bgs_mutex);
2990 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2992 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
2995 BUG_ON(ret == 0); /* Corruption */
2997 ret = btrfs_previous_item(chunk_root, path, key.objectid,
3000 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3006 leaf = path->nodes[0];
3007 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3009 chunk = btrfs_item_ptr(leaf, path->slots[0],
3010 struct btrfs_chunk);
3011 chunk_type = btrfs_chunk_type(leaf, chunk);
3012 btrfs_release_path(path);
3014 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3015 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3021 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3023 if (found_key.offset == 0)
3025 key.offset = found_key.offset - 1;
3028 if (failed && !retried) {
3032 } else if (WARN_ON(failed && retried)) {
3036 btrfs_free_path(path);
3040 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3041 struct btrfs_balance_control *bctl)
3043 struct btrfs_root *root = fs_info->tree_root;
3044 struct btrfs_trans_handle *trans;
3045 struct btrfs_balance_item *item;
3046 struct btrfs_disk_balance_args disk_bargs;
3047 struct btrfs_path *path;
3048 struct extent_buffer *leaf;
3049 struct btrfs_key key;
3052 path = btrfs_alloc_path();
3056 trans = btrfs_start_transaction(root, 0);
3057 if (IS_ERR(trans)) {
3058 btrfs_free_path(path);
3059 return PTR_ERR(trans);
3062 key.objectid = BTRFS_BALANCE_OBJECTID;
3063 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3066 ret = btrfs_insert_empty_item(trans, root, path, &key,
3071 leaf = path->nodes[0];
3072 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3074 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3076 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3077 btrfs_set_balance_data(leaf, item, &disk_bargs);
3078 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3079 btrfs_set_balance_meta(leaf, item, &disk_bargs);
3080 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3081 btrfs_set_balance_sys(leaf, item, &disk_bargs);
3083 btrfs_set_balance_flags(leaf, item, bctl->flags);
3085 btrfs_mark_buffer_dirty(leaf);
3087 btrfs_free_path(path);
3088 err = btrfs_commit_transaction(trans);
3094 static int del_balance_item(struct btrfs_fs_info *fs_info)
3096 struct btrfs_root *root = fs_info->tree_root;
3097 struct btrfs_trans_handle *trans;
3098 struct btrfs_path *path;
3099 struct btrfs_key key;
3102 path = btrfs_alloc_path();
3106 trans = btrfs_start_transaction(root, 0);
3107 if (IS_ERR(trans)) {
3108 btrfs_free_path(path);
3109 return PTR_ERR(trans);
3112 key.objectid = BTRFS_BALANCE_OBJECTID;
3113 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3116 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3124 ret = btrfs_del_item(trans, root, path);
3126 btrfs_free_path(path);
3127 err = btrfs_commit_transaction(trans);
3134 * This is a heuristic used to reduce the number of chunks balanced on
3135 * resume after balance was interrupted.
3137 static void update_balance_args(struct btrfs_balance_control *bctl)
3140 * Turn on soft mode for chunk types that were being converted.
3142 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3143 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3144 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3145 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3146 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3147 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3150 * Turn on usage filter if is not already used. The idea is
3151 * that chunks that we have already balanced should be
3152 * reasonably full. Don't do it for chunks that are being
3153 * converted - that will keep us from relocating unconverted
3154 * (albeit full) chunks.
3156 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3157 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3158 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3159 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3160 bctl->data.usage = 90;
3162 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3163 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3164 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3165 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3166 bctl->sys.usage = 90;
3168 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3169 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3170 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3171 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3172 bctl->meta.usage = 90;
3177 * Should be called with both balance and volume mutexes held to
3178 * serialize other volume operations (add_dev/rm_dev/resize) with
3179 * restriper. Same goes for unset_balance_control.
3181 static void set_balance_control(struct btrfs_balance_control *bctl)
3183 struct btrfs_fs_info *fs_info = bctl->fs_info;
3185 BUG_ON(fs_info->balance_ctl);
3187 spin_lock(&fs_info->balance_lock);
3188 fs_info->balance_ctl = bctl;
3189 spin_unlock(&fs_info->balance_lock);
3192 static void unset_balance_control(struct btrfs_fs_info *fs_info)
3194 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3196 BUG_ON(!fs_info->balance_ctl);
3198 spin_lock(&fs_info->balance_lock);
3199 fs_info->balance_ctl = NULL;
3200 spin_unlock(&fs_info->balance_lock);
3206 * Balance filters. Return 1 if chunk should be filtered out
3207 * (should not be balanced).
3209 static int chunk_profiles_filter(u64 chunk_type,
3210 struct btrfs_balance_args *bargs)
3212 chunk_type = chunk_to_extended(chunk_type) &
3213 BTRFS_EXTENDED_PROFILE_MASK;
3215 if (bargs->profiles & chunk_type)
3221 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3222 struct btrfs_balance_args *bargs)
3224 struct btrfs_block_group_cache *cache;
3226 u64 user_thresh_min;
3227 u64 user_thresh_max;
3230 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3231 chunk_used = btrfs_block_group_used(&cache->item);
3233 if (bargs->usage_min == 0)
3234 user_thresh_min = 0;
3236 user_thresh_min = div_factor_fine(cache->key.offset,
3239 if (bargs->usage_max == 0)
3240 user_thresh_max = 1;
3241 else if (bargs->usage_max > 100)
3242 user_thresh_max = cache->key.offset;
3244 user_thresh_max = div_factor_fine(cache->key.offset,
3247 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3250 btrfs_put_block_group(cache);
3254 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3255 u64 chunk_offset, struct btrfs_balance_args *bargs)
3257 struct btrfs_block_group_cache *cache;
3258 u64 chunk_used, user_thresh;
3261 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3262 chunk_used = btrfs_block_group_used(&cache->item);
3264 if (bargs->usage_min == 0)
3266 else if (bargs->usage > 100)
3267 user_thresh = cache->key.offset;
3269 user_thresh = div_factor_fine(cache->key.offset,
3272 if (chunk_used < user_thresh)
3275 btrfs_put_block_group(cache);
3279 static int chunk_devid_filter(struct extent_buffer *leaf,
3280 struct btrfs_chunk *chunk,
3281 struct btrfs_balance_args *bargs)
3283 struct btrfs_stripe *stripe;
3284 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3287 for (i = 0; i < num_stripes; i++) {
3288 stripe = btrfs_stripe_nr(chunk, i);
3289 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3296 /* [pstart, pend) */
3297 static int chunk_drange_filter(struct extent_buffer *leaf,
3298 struct btrfs_chunk *chunk,
3299 struct btrfs_balance_args *bargs)
3301 struct btrfs_stripe *stripe;
3302 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3308 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3311 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3312 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3313 factor = num_stripes / 2;
3314 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3315 factor = num_stripes - 1;
3316 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3317 factor = num_stripes - 2;
3319 factor = num_stripes;
3322 for (i = 0; i < num_stripes; i++) {
3323 stripe = btrfs_stripe_nr(chunk, i);
3324 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3327 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3328 stripe_length = btrfs_chunk_length(leaf, chunk);
3329 stripe_length = div_u64(stripe_length, factor);
3331 if (stripe_offset < bargs->pend &&
3332 stripe_offset + stripe_length > bargs->pstart)
3339 /* [vstart, vend) */
3340 static int chunk_vrange_filter(struct extent_buffer *leaf,
3341 struct btrfs_chunk *chunk,
3343 struct btrfs_balance_args *bargs)
3345 if (chunk_offset < bargs->vend &&
3346 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3347 /* at least part of the chunk is inside this vrange */
3353 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3354 struct btrfs_chunk *chunk,
3355 struct btrfs_balance_args *bargs)
3357 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3359 if (bargs->stripes_min <= num_stripes
3360 && num_stripes <= bargs->stripes_max)
3366 static int chunk_soft_convert_filter(u64 chunk_type,
3367 struct btrfs_balance_args *bargs)
3369 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3372 chunk_type = chunk_to_extended(chunk_type) &
3373 BTRFS_EXTENDED_PROFILE_MASK;
3375 if (bargs->target == chunk_type)
3381 static int should_balance_chunk(struct btrfs_fs_info *fs_info,
3382 struct extent_buffer *leaf,
3383 struct btrfs_chunk *chunk, u64 chunk_offset)
3385 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3386 struct btrfs_balance_args *bargs = NULL;
3387 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3390 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3391 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3395 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3396 bargs = &bctl->data;
3397 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3399 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3400 bargs = &bctl->meta;
3402 /* profiles filter */
3403 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3404 chunk_profiles_filter(chunk_type, bargs)) {
3409 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3410 chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3412 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3413 chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3418 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3419 chunk_devid_filter(leaf, chunk, bargs)) {
3423 /* drange filter, makes sense only with devid filter */
3424 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3425 chunk_drange_filter(leaf, chunk, bargs)) {
3430 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3431 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3435 /* stripes filter */
3436 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3437 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3441 /* soft profile changing mode */
3442 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3443 chunk_soft_convert_filter(chunk_type, bargs)) {
3448 * limited by count, must be the last filter
3450 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3451 if (bargs->limit == 0)
3455 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3457 * Same logic as the 'limit' filter; the minimum cannot be
3458 * determined here because we do not have the global information
3459 * about the count of all chunks that satisfy the filters.
3461 if (bargs->limit_max == 0)
3470 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3472 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3473 struct btrfs_root *chunk_root = fs_info->chunk_root;
3474 struct btrfs_root *dev_root = fs_info->dev_root;
3475 struct list_head *devices;
3476 struct btrfs_device *device;
3480 struct btrfs_chunk *chunk;
3481 struct btrfs_path *path = NULL;
3482 struct btrfs_key key;
3483 struct btrfs_key found_key;
3484 struct btrfs_trans_handle *trans;
3485 struct extent_buffer *leaf;
3488 int enospc_errors = 0;
3489 bool counting = true;
3490 /* The single value limit and min/max limits use the same bytes in the */
3491 u64 limit_data = bctl->data.limit;
3492 u64 limit_meta = bctl->meta.limit;
3493 u64 limit_sys = bctl->sys.limit;
3497 int chunk_reserved = 0;
3500 /* step one make some room on all the devices */
3501 devices = &fs_info->fs_devices->devices;
3502 list_for_each_entry(device, devices, dev_list) {
3503 old_size = btrfs_device_get_total_bytes(device);
3504 size_to_free = div_factor(old_size, 1);
3505 size_to_free = min_t(u64, size_to_free, SZ_1M);
3506 if (!device->writeable ||
3507 btrfs_device_get_total_bytes(device) -
3508 btrfs_device_get_bytes_used(device) > size_to_free ||
3509 device->is_tgtdev_for_dev_replace)
3512 ret = btrfs_shrink_device(device, old_size - size_to_free);
3516 /* btrfs_shrink_device never returns ret > 0 */
3521 trans = btrfs_start_transaction(dev_root, 0);
3522 if (IS_ERR(trans)) {
3523 ret = PTR_ERR(trans);
3524 btrfs_info_in_rcu(fs_info,
3525 "resize: unable to start transaction after shrinking device %s (error %d), old size %llu, new size %llu",
3526 rcu_str_deref(device->name), ret,
3527 old_size, old_size - size_to_free);
3531 ret = btrfs_grow_device(trans, device, old_size);
3533 btrfs_end_transaction(trans);
3534 /* btrfs_grow_device never returns ret > 0 */
3536 btrfs_info_in_rcu(fs_info,
3537 "resize: unable to grow device after shrinking device %s (error %d), old size %llu, new size %llu",
3538 rcu_str_deref(device->name), ret,
3539 old_size, old_size - size_to_free);
3543 btrfs_end_transaction(trans);
3546 /* step two, relocate all the chunks */
3547 path = btrfs_alloc_path();
3553 /* zero out stat counters */
3554 spin_lock(&fs_info->balance_lock);
3555 memset(&bctl->stat, 0, sizeof(bctl->stat));
3556 spin_unlock(&fs_info->balance_lock);
3560 * The single value limit and min/max limits use the same bytes
3563 bctl->data.limit = limit_data;
3564 bctl->meta.limit = limit_meta;
3565 bctl->sys.limit = limit_sys;
3567 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3568 key.offset = (u64)-1;
3569 key.type = BTRFS_CHUNK_ITEM_KEY;
3572 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3573 atomic_read(&fs_info->balance_cancel_req)) {
3578 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3579 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3581 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3586 * this shouldn't happen, it means the last relocate
3590 BUG(); /* FIXME break ? */
3592 ret = btrfs_previous_item(chunk_root, path, 0,
3593 BTRFS_CHUNK_ITEM_KEY);
3595 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3600 leaf = path->nodes[0];
3601 slot = path->slots[0];
3602 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3604 if (found_key.objectid != key.objectid) {
3605 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3609 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3610 chunk_type = btrfs_chunk_type(leaf, chunk);
3613 spin_lock(&fs_info->balance_lock);
3614 bctl->stat.considered++;
3615 spin_unlock(&fs_info->balance_lock);
3618 ret = should_balance_chunk(fs_info, leaf, chunk,
3621 btrfs_release_path(path);
3623 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3628 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3629 spin_lock(&fs_info->balance_lock);
3630 bctl->stat.expected++;
3631 spin_unlock(&fs_info->balance_lock);
3633 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3635 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3637 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3644 * Apply limit_min filter, no need to check if the LIMITS
3645 * filter is used, limit_min is 0 by default
3647 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3648 count_data < bctl->data.limit_min)
3649 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3650 count_meta < bctl->meta.limit_min)
3651 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3652 count_sys < bctl->sys.limit_min)) {
3653 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3657 ASSERT(fs_info->data_sinfo);
3658 spin_lock(&fs_info->data_sinfo->lock);
3659 bytes_used = fs_info->data_sinfo->bytes_used;
3660 spin_unlock(&fs_info->data_sinfo->lock);
3662 if ((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3663 !chunk_reserved && !bytes_used) {
3664 trans = btrfs_start_transaction(chunk_root, 0);
3665 if (IS_ERR(trans)) {
3666 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3667 ret = PTR_ERR(trans);
3671 ret = btrfs_force_chunk_alloc(trans, fs_info,
3672 BTRFS_BLOCK_GROUP_DATA);
3673 btrfs_end_transaction(trans);
3675 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3681 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3682 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3683 if (ret && ret != -ENOSPC)
3685 if (ret == -ENOSPC) {
3688 spin_lock(&fs_info->balance_lock);
3689 bctl->stat.completed++;
3690 spin_unlock(&fs_info->balance_lock);
3693 if (found_key.offset == 0)
3695 key.offset = found_key.offset - 1;
3699 btrfs_release_path(path);
3704 btrfs_free_path(path);
3705 if (enospc_errors) {
3706 btrfs_info(fs_info, "%d enospc errors during balance",
3716 * alloc_profile_is_valid - see if a given profile is valid and reduced
3717 * @flags: profile to validate
3718 * @extended: if true @flags is treated as an extended profile
3720 static int alloc_profile_is_valid(u64 flags, int extended)
3722 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3723 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3725 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3727 /* 1) check that all other bits are zeroed */
3731 /* 2) see if profile is reduced */
3733 return !extended; /* "0" is valid for usual profiles */
3735 /* true if exactly one bit set */
3736 return (flags & (flags - 1)) == 0;
3739 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3741 /* cancel requested || normal exit path */
3742 return atomic_read(&fs_info->balance_cancel_req) ||
3743 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3744 atomic_read(&fs_info->balance_cancel_req) == 0);
3747 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3751 unset_balance_control(fs_info);
3752 ret = del_balance_item(fs_info);
3754 btrfs_handle_fs_error(fs_info, ret, NULL);
3756 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3759 /* Non-zero return value signifies invalidity */
3760 static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg,
3763 return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3764 (!alloc_profile_is_valid(bctl_arg->target, 1) ||
3765 (bctl_arg->target & ~allowed)));
3769 * Should be called with both balance and volume mutexes held
3771 int btrfs_balance(struct btrfs_balance_control *bctl,
3772 struct btrfs_ioctl_balance_args *bargs)
3774 struct btrfs_fs_info *fs_info = bctl->fs_info;
3775 u64 meta_target, data_target;
3782 if (btrfs_fs_closing(fs_info) ||
3783 atomic_read(&fs_info->balance_pause_req) ||
3784 atomic_read(&fs_info->balance_cancel_req)) {
3789 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3790 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3794 * In case of mixed groups both data and meta should be picked,
3795 * and identical options should be given for both of them.
3797 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3798 if (mixed && (bctl->flags & allowed)) {
3799 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3800 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3801 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3803 "with mixed groups data and metadata balance options must be the same");
3809 num_devices = fs_info->fs_devices->num_devices;
3810 btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
3811 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3812 BUG_ON(num_devices < 1);
3815 btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
3816 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE | BTRFS_BLOCK_GROUP_DUP;
3817 if (num_devices > 1)
3818 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3819 if (num_devices > 2)
3820 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3821 if (num_devices > 3)
3822 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3823 BTRFS_BLOCK_GROUP_RAID6);
3824 if (validate_convert_profile(&bctl->data, allowed)) {
3826 "unable to start balance with target data profile %llu",
3831 if (validate_convert_profile(&bctl->meta, allowed)) {
3833 "unable to start balance with target metadata profile %llu",
3838 if (validate_convert_profile(&bctl->sys, allowed)) {
3840 "unable to start balance with target system profile %llu",
3846 /* allow to reduce meta or sys integrity only if force set */
3847 allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3848 BTRFS_BLOCK_GROUP_RAID10 |
3849 BTRFS_BLOCK_GROUP_RAID5 |
3850 BTRFS_BLOCK_GROUP_RAID6;
3852 seq = read_seqbegin(&fs_info->profiles_lock);
3854 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3855 (fs_info->avail_system_alloc_bits & allowed) &&
3856 !(bctl->sys.target & allowed)) ||
3857 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3858 (fs_info->avail_metadata_alloc_bits & allowed) &&
3859 !(bctl->meta.target & allowed))) {
3860 if (bctl->flags & BTRFS_BALANCE_FORCE) {
3862 "force reducing metadata integrity");
3865 "balance will reduce metadata integrity, use force if you want this");
3870 } while (read_seqretry(&fs_info->profiles_lock, seq));
3872 /* if we're not converting, the target field is uninitialized */
3873 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
3874 bctl->meta.target : fs_info->avail_metadata_alloc_bits;
3875 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
3876 bctl->data.target : fs_info->avail_data_alloc_bits;
3877 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <
3878 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {
3880 "metadata profile 0x%llx has lower redundancy than data profile 0x%llx",
3881 meta_target, data_target);
3884 ret = insert_balance_item(fs_info, bctl);
3885 if (ret && ret != -EEXIST)
3888 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3889 BUG_ON(ret == -EEXIST);
3890 set_balance_control(bctl);
3892 BUG_ON(ret != -EEXIST);
3893 spin_lock(&fs_info->balance_lock);
3894 update_balance_args(bctl);
3895 spin_unlock(&fs_info->balance_lock);
3898 atomic_inc(&fs_info->balance_running);
3899 mutex_unlock(&fs_info->balance_mutex);
3901 ret = __btrfs_balance(fs_info);
3903 mutex_lock(&fs_info->balance_mutex);
3904 atomic_dec(&fs_info->balance_running);
3907 memset(bargs, 0, sizeof(*bargs));
3908 update_ioctl_balance_args(fs_info, 0, bargs);
3911 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3912 balance_need_close(fs_info)) {
3913 __cancel_balance(fs_info);
3916 wake_up(&fs_info->balance_wait_q);
3920 if (bctl->flags & BTRFS_BALANCE_RESUME)
3921 __cancel_balance(fs_info);
3924 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3929 static int balance_kthread(void *data)
3931 struct btrfs_fs_info *fs_info = data;
3934 mutex_lock(&fs_info->volume_mutex);
3935 mutex_lock(&fs_info->balance_mutex);
3937 if (fs_info->balance_ctl) {
3938 btrfs_info(fs_info, "continuing balance");
3939 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3942 mutex_unlock(&fs_info->balance_mutex);
3943 mutex_unlock(&fs_info->volume_mutex);
3948 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3950 struct task_struct *tsk;
3952 spin_lock(&fs_info->balance_lock);
3953 if (!fs_info->balance_ctl) {
3954 spin_unlock(&fs_info->balance_lock);
3957 spin_unlock(&fs_info->balance_lock);
3959 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
3960 btrfs_info(fs_info, "force skipping balance");
3964 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3965 return PTR_ERR_OR_ZERO(tsk);
3968 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3970 struct btrfs_balance_control *bctl;
3971 struct btrfs_balance_item *item;
3972 struct btrfs_disk_balance_args disk_bargs;
3973 struct btrfs_path *path;
3974 struct extent_buffer *leaf;
3975 struct btrfs_key key;
3978 path = btrfs_alloc_path();
3982 key.objectid = BTRFS_BALANCE_OBJECTID;
3983 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3986 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3989 if (ret > 0) { /* ret = -ENOENT; */
3994 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4000 leaf = path->nodes[0];
4001 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
4003 bctl->fs_info = fs_info;
4004 bctl->flags = btrfs_balance_flags(leaf, item);
4005 bctl->flags |= BTRFS_BALANCE_RESUME;
4007 btrfs_balance_data(leaf, item, &disk_bargs);
4008 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4009 btrfs_balance_meta(leaf, item, &disk_bargs);
4010 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4011 btrfs_balance_sys(leaf, item, &disk_bargs);
4012 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4014 WARN_ON(test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags));
4016 mutex_lock(&fs_info->volume_mutex);
4017 mutex_lock(&fs_info->balance_mutex);
4019 set_balance_control(bctl);
4021 mutex_unlock(&fs_info->balance_mutex);
4022 mutex_unlock(&fs_info->volume_mutex);
4024 btrfs_free_path(path);
4028 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4032 mutex_lock(&fs_info->balance_mutex);
4033 if (!fs_info->balance_ctl) {
4034 mutex_unlock(&fs_info->balance_mutex);
4038 if (atomic_read(&fs_info->balance_running)) {
4039 atomic_inc(&fs_info->balance_pause_req);
4040 mutex_unlock(&fs_info->balance_mutex);
4042 wait_event(fs_info->balance_wait_q,
4043 atomic_read(&fs_info->balance_running) == 0);
4045 mutex_lock(&fs_info->balance_mutex);
4046 /* we are good with balance_ctl ripped off from under us */
4047 BUG_ON(atomic_read(&fs_info->balance_running));
4048 atomic_dec(&fs_info->balance_pause_req);
4053 mutex_unlock(&fs_info->balance_mutex);
4057 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4059 if (fs_info->sb->s_flags & MS_RDONLY)
4062 mutex_lock(&fs_info->balance_mutex);
4063 if (!fs_info->balance_ctl) {
4064 mutex_unlock(&fs_info->balance_mutex);
4068 atomic_inc(&fs_info->balance_cancel_req);
4070 * if we are running just wait and return, balance item is
4071 * deleted in btrfs_balance in this case
4073 if (atomic_read(&fs_info->balance_running)) {
4074 mutex_unlock(&fs_info->balance_mutex);
4075 wait_event(fs_info->balance_wait_q,
4076 atomic_read(&fs_info->balance_running) == 0);
4077 mutex_lock(&fs_info->balance_mutex);
4079 /* __cancel_balance needs volume_mutex */
4080 mutex_unlock(&fs_info->balance_mutex);
4081 mutex_lock(&fs_info->volume_mutex);
4082 mutex_lock(&fs_info->balance_mutex);
4084 if (fs_info->balance_ctl)
4085 __cancel_balance(fs_info);
4087 mutex_unlock(&fs_info->volume_mutex);
4090 BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
4091 atomic_dec(&fs_info->balance_cancel_req);
4092 mutex_unlock(&fs_info->balance_mutex);
4096 static int btrfs_uuid_scan_kthread(void *data)
4098 struct btrfs_fs_info *fs_info = data;
4099 struct btrfs_root *root = fs_info->tree_root;
4100 struct btrfs_key key;
4101 struct btrfs_path *path = NULL;
4103 struct extent_buffer *eb;
4105 struct btrfs_root_item root_item;
4107 struct btrfs_trans_handle *trans = NULL;
4109 path = btrfs_alloc_path();
4116 key.type = BTRFS_ROOT_ITEM_KEY;
4120 ret = btrfs_search_forward(root, &key, path, 0);
4127 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4128 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4129 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4130 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4133 eb = path->nodes[0];
4134 slot = path->slots[0];
4135 item_size = btrfs_item_size_nr(eb, slot);
4136 if (item_size < sizeof(root_item))
4139 read_extent_buffer(eb, &root_item,
4140 btrfs_item_ptr_offset(eb, slot),
4141 (int)sizeof(root_item));
4142 if (btrfs_root_refs(&root_item) == 0)
4145 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4146 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4150 btrfs_release_path(path);
4152 * 1 - subvol uuid item
4153 * 1 - received_subvol uuid item
4155 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4156 if (IS_ERR(trans)) {
4157 ret = PTR_ERR(trans);
4165 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4166 ret = btrfs_uuid_tree_add(trans, fs_info,
4168 BTRFS_UUID_KEY_SUBVOL,
4171 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4177 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4178 ret = btrfs_uuid_tree_add(trans, fs_info,
4179 root_item.received_uuid,
4180 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4183 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4191 ret = btrfs_end_transaction(trans);
4197 btrfs_release_path(path);
4198 if (key.offset < (u64)-1) {
4200 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4202 key.type = BTRFS_ROOT_ITEM_KEY;
4203 } else if (key.objectid < (u64)-1) {
4205 key.type = BTRFS_ROOT_ITEM_KEY;
4214 btrfs_free_path(path);
4215 if (trans && !IS_ERR(trans))
4216 btrfs_end_transaction(trans);
4218 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4220 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4221 up(&fs_info->uuid_tree_rescan_sem);
4226 * Callback for btrfs_uuid_tree_iterate().
4228 * 0 check succeeded, the entry is not outdated.
4229 * < 0 if an error occurred.
4230 * > 0 if the check failed, which means the caller shall remove the entry.
4232 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
4233 u8 *uuid, u8 type, u64 subid)
4235 struct btrfs_key key;
4237 struct btrfs_root *subvol_root;
4239 if (type != BTRFS_UUID_KEY_SUBVOL &&
4240 type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
4243 key.objectid = subid;
4244 key.type = BTRFS_ROOT_ITEM_KEY;
4245 key.offset = (u64)-1;
4246 subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
4247 if (IS_ERR(subvol_root)) {
4248 ret = PTR_ERR(subvol_root);
4255 case BTRFS_UUID_KEY_SUBVOL:
4256 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
4259 case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
4260 if (memcmp(uuid, subvol_root->root_item.received_uuid,
4270 static int btrfs_uuid_rescan_kthread(void *data)
4272 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4276 * 1st step is to iterate through the existing UUID tree and
4277 * to delete all entries that contain outdated data.
4278 * 2nd step is to add all missing entries to the UUID tree.
4280 ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4282 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4283 up(&fs_info->uuid_tree_rescan_sem);
4286 return btrfs_uuid_scan_kthread(data);
4289 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4291 struct btrfs_trans_handle *trans;
4292 struct btrfs_root *tree_root = fs_info->tree_root;
4293 struct btrfs_root *uuid_root;
4294 struct task_struct *task;
4301 trans = btrfs_start_transaction(tree_root, 2);
4303 return PTR_ERR(trans);
4305 uuid_root = btrfs_create_tree(trans, fs_info,
4306 BTRFS_UUID_TREE_OBJECTID);
4307 if (IS_ERR(uuid_root)) {
4308 ret = PTR_ERR(uuid_root);
4309 btrfs_abort_transaction(trans, ret);
4310 btrfs_end_transaction(trans);
4314 fs_info->uuid_root = uuid_root;
4316 ret = btrfs_commit_transaction(trans);
4320 down(&fs_info->uuid_tree_rescan_sem);
4321 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4323 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4324 btrfs_warn(fs_info, "failed to start uuid_scan task");
4325 up(&fs_info->uuid_tree_rescan_sem);
4326 return PTR_ERR(task);
4332 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4334 struct task_struct *task;
4336 down(&fs_info->uuid_tree_rescan_sem);
4337 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4339 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4340 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4341 up(&fs_info->uuid_tree_rescan_sem);
4342 return PTR_ERR(task);
4349 * shrinking a device means finding all of the device extents past
4350 * the new size, and then following the back refs to the chunks.
4351 * The chunk relocation code actually frees the device extent
4353 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4355 struct btrfs_fs_info *fs_info = device->fs_info;
4356 struct btrfs_root *root = fs_info->dev_root;
4357 struct btrfs_trans_handle *trans;
4358 struct btrfs_dev_extent *dev_extent = NULL;
4359 struct btrfs_path *path;
4365 bool retried = false;
4366 bool checked_pending_chunks = false;
4367 struct extent_buffer *l;
4368 struct btrfs_key key;
4369 struct btrfs_super_block *super_copy = fs_info->super_copy;
4370 u64 old_total = btrfs_super_total_bytes(super_copy);
4371 u64 old_size = btrfs_device_get_total_bytes(device);
4374 new_size = round_down(new_size, fs_info->sectorsize);
4375 diff = round_down(old_size - new_size, fs_info->sectorsize);
4377 if (device->is_tgtdev_for_dev_replace)
4380 path = btrfs_alloc_path();
4384 path->reada = READA_FORWARD;
4386 mutex_lock(&fs_info->chunk_mutex);
4388 btrfs_device_set_total_bytes(device, new_size);
4389 if (device->writeable) {
4390 device->fs_devices->total_rw_bytes -= diff;
4391 atomic64_sub(diff, &fs_info->free_chunk_space);
4393 mutex_unlock(&fs_info->chunk_mutex);
4396 key.objectid = device->devid;
4397 key.offset = (u64)-1;
4398 key.type = BTRFS_DEV_EXTENT_KEY;
4401 mutex_lock(&fs_info->delete_unused_bgs_mutex);
4402 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4404 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4408 ret = btrfs_previous_item(root, path, 0, key.type);
4410 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4415 btrfs_release_path(path);
4420 slot = path->slots[0];
4421 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4423 if (key.objectid != device->devid) {
4424 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4425 btrfs_release_path(path);
4429 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4430 length = btrfs_dev_extent_length(l, dev_extent);
4432 if (key.offset + length <= new_size) {
4433 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4434 btrfs_release_path(path);
4438 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4439 btrfs_release_path(path);
4441 ret = btrfs_relocate_chunk(fs_info, chunk_offset);
4442 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4443 if (ret && ret != -ENOSPC)
4447 } while (key.offset-- > 0);
4449 if (failed && !retried) {
4453 } else if (failed && retried) {
4458 /* Shrinking succeeded, else we would be at "done". */
4459 trans = btrfs_start_transaction(root, 0);
4460 if (IS_ERR(trans)) {
4461 ret = PTR_ERR(trans);
4465 mutex_lock(&fs_info->chunk_mutex);
4468 * We checked in the above loop all device extents that were already in
4469 * the device tree. However before we have updated the device's
4470 * total_bytes to the new size, we might have had chunk allocations that
4471 * have not complete yet (new block groups attached to transaction
4472 * handles), and therefore their device extents were not yet in the
4473 * device tree and we missed them in the loop above. So if we have any
4474 * pending chunk using a device extent that overlaps the device range
4475 * that we can not use anymore, commit the current transaction and
4476 * repeat the search on the device tree - this way we guarantee we will
4477 * not have chunks using device extents that end beyond 'new_size'.
4479 if (!checked_pending_chunks) {
4480 u64 start = new_size;
4481 u64 len = old_size - new_size;
4483 if (contains_pending_extent(trans->transaction, device,
4485 mutex_unlock(&fs_info->chunk_mutex);
4486 checked_pending_chunks = true;
4489 ret = btrfs_commit_transaction(trans);
4496 btrfs_device_set_disk_total_bytes(device, new_size);
4497 if (list_empty(&device->resized_list))
4498 list_add_tail(&device->resized_list,
4499 &fs_info->fs_devices->resized_devices);
4501 WARN_ON(diff > old_total);
4502 btrfs_set_super_total_bytes(super_copy,
4503 round_down(old_total - diff, fs_info->sectorsize));
4504 mutex_unlock(&fs_info->chunk_mutex);
4506 /* Now btrfs_update_device() will change the on-disk size. */
4507 ret = btrfs_update_device(trans, device);
4508 btrfs_end_transaction(trans);
4510 btrfs_free_path(path);
4512 mutex_lock(&fs_info->chunk_mutex);
4513 btrfs_device_set_total_bytes(device, old_size);
4514 if (device->writeable)
4515 device->fs_devices->total_rw_bytes += diff;
4516 atomic64_add(diff, &fs_info->free_chunk_space);
4517 mutex_unlock(&fs_info->chunk_mutex);
4522 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
4523 struct btrfs_key *key,
4524 struct btrfs_chunk *chunk, int item_size)
4526 struct btrfs_super_block *super_copy = fs_info->super_copy;
4527 struct btrfs_disk_key disk_key;
4531 mutex_lock(&fs_info->chunk_mutex);
4532 array_size = btrfs_super_sys_array_size(super_copy);
4533 if (array_size + item_size + sizeof(disk_key)
4534 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4535 mutex_unlock(&fs_info->chunk_mutex);
4539 ptr = super_copy->sys_chunk_array + array_size;
4540 btrfs_cpu_key_to_disk(&disk_key, key);
4541 memcpy(ptr, &disk_key, sizeof(disk_key));
4542 ptr += sizeof(disk_key);
4543 memcpy(ptr, chunk, item_size);
4544 item_size += sizeof(disk_key);
4545 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4546 mutex_unlock(&fs_info->chunk_mutex);
4552 * sort the devices in descending order by max_avail, total_avail
4554 static int btrfs_cmp_device_info(const void *a, const void *b)
4556 const struct btrfs_device_info *di_a = a;
4557 const struct btrfs_device_info *di_b = b;
4559 if (di_a->max_avail > di_b->max_avail)
4561 if (di_a->max_avail < di_b->max_avail)
4563 if (di_a->total_avail > di_b->total_avail)
4565 if (di_a->total_avail < di_b->total_avail)
4570 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4572 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4575 btrfs_set_fs_incompat(info, RAID56);
4578 #define BTRFS_MAX_DEVS(r) ((BTRFS_MAX_ITEM_SIZE(r->fs_info) \
4579 - sizeof(struct btrfs_chunk)) \
4580 / sizeof(struct btrfs_stripe) + 1)
4582 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
4583 - 2 * sizeof(struct btrfs_disk_key) \
4584 - 2 * sizeof(struct btrfs_chunk)) \
4585 / sizeof(struct btrfs_stripe) + 1)
4587 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4588 u64 start, u64 type)
4590 struct btrfs_fs_info *info = trans->fs_info;
4591 struct btrfs_fs_devices *fs_devices = info->fs_devices;
4592 struct btrfs_device *device;
4593 struct map_lookup *map = NULL;
4594 struct extent_map_tree *em_tree;
4595 struct extent_map *em;
4596 struct btrfs_device_info *devices_info = NULL;
4598 int num_stripes; /* total number of stripes to allocate */
4599 int data_stripes; /* number of stripes that count for
4601 int sub_stripes; /* sub_stripes info for map */
4602 int dev_stripes; /* stripes per dev */
4603 int devs_max; /* max devs to use */
4604 int devs_min; /* min devs needed */
4605 int devs_increment; /* ndevs has to be a multiple of this */
4606 int ncopies; /* how many copies to data has */
4608 u64 max_stripe_size;
4617 BUG_ON(!alloc_profile_is_valid(type, 0));
4619 if (list_empty(&fs_devices->alloc_list))
4622 index = __get_raid_index(type);
4624 sub_stripes = btrfs_raid_array[index].sub_stripes;
4625 dev_stripes = btrfs_raid_array[index].dev_stripes;
4626 devs_max = btrfs_raid_array[index].devs_max;
4627 devs_min = btrfs_raid_array[index].devs_min;
4628 devs_increment = btrfs_raid_array[index].devs_increment;
4629 ncopies = btrfs_raid_array[index].ncopies;
4631 if (type & BTRFS_BLOCK_GROUP_DATA) {
4632 max_stripe_size = SZ_1G;
4633 max_chunk_size = 10 * max_stripe_size;
4635 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4636 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4637 /* for larger filesystems, use larger metadata chunks */
4638 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
4639 max_stripe_size = SZ_1G;
4641 max_stripe_size = SZ_256M;
4642 max_chunk_size = max_stripe_size;
4644 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4645 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4646 max_stripe_size = SZ_32M;
4647 max_chunk_size = 2 * max_stripe_size;
4649 devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4651 btrfs_err(info, "invalid chunk type 0x%llx requested",
4656 /* we don't want a chunk larger than 10% of writeable space */
4657 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4660 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4666 * in the first pass through the devices list, we gather information
4667 * about the available holes on each device.
4670 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
4674 if (!device->writeable) {
4676 "BTRFS: read-only device in alloc_list\n");
4680 if (!device->in_fs_metadata ||
4681 device->is_tgtdev_for_dev_replace)
4684 if (device->total_bytes > device->bytes_used)
4685 total_avail = device->total_bytes - device->bytes_used;
4689 /* If there is no space on this device, skip it. */
4690 if (total_avail == 0)
4693 ret = find_free_dev_extent(trans, device,
4694 max_stripe_size * dev_stripes,
4695 &dev_offset, &max_avail);
4696 if (ret && ret != -ENOSPC)
4700 max_avail = max_stripe_size * dev_stripes;
4702 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4705 if (ndevs == fs_devices->rw_devices) {
4706 WARN(1, "%s: found more than %llu devices\n",
4707 __func__, fs_devices->rw_devices);
4710 devices_info[ndevs].dev_offset = dev_offset;
4711 devices_info[ndevs].max_avail = max_avail;
4712 devices_info[ndevs].total_avail = total_avail;
4713 devices_info[ndevs].dev = device;
4718 * now sort the devices by hole size / available space
4720 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4721 btrfs_cmp_device_info, NULL);
4723 /* round down to number of usable stripes */
4724 ndevs = round_down(ndevs, devs_increment);
4726 if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4731 ndevs = min(ndevs, devs_max);
4734 * the primary goal is to maximize the number of stripes, so use as many
4735 * devices as possible, even if the stripes are not maximum sized.
4737 stripe_size = devices_info[ndevs-1].max_avail;
4738 num_stripes = ndevs * dev_stripes;
4741 * this will have to be fixed for RAID1 and RAID10 over
4744 data_stripes = num_stripes / ncopies;
4746 if (type & BTRFS_BLOCK_GROUP_RAID5)
4747 data_stripes = num_stripes - 1;
4749 if (type & BTRFS_BLOCK_GROUP_RAID6)
4750 data_stripes = num_stripes - 2;
4753 * Use the number of data stripes to figure out how big this chunk
4754 * is really going to be in terms of logical address space,
4755 * and compare that answer with the max chunk size
4757 if (stripe_size * data_stripes > max_chunk_size) {
4758 u64 mask = (1ULL << 24) - 1;
4760 stripe_size = div_u64(max_chunk_size, data_stripes);
4762 /* bump the answer up to a 16MB boundary */
4763 stripe_size = (stripe_size + mask) & ~mask;
4765 /* but don't go higher than the limits we found
4766 * while searching for free extents
4768 if (stripe_size > devices_info[ndevs-1].max_avail)
4769 stripe_size = devices_info[ndevs-1].max_avail;
4772 stripe_size = div_u64(stripe_size, dev_stripes);
4774 /* align to BTRFS_STRIPE_LEN */
4775 stripe_size = round_down(stripe_size, BTRFS_STRIPE_LEN);
4777 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4782 map->num_stripes = num_stripes;
4784 for (i = 0; i < ndevs; ++i) {
4785 for (j = 0; j < dev_stripes; ++j) {
4786 int s = i * dev_stripes + j;
4787 map->stripes[s].dev = devices_info[i].dev;
4788 map->stripes[s].physical = devices_info[i].dev_offset +
4792 map->stripe_len = BTRFS_STRIPE_LEN;
4793 map->io_align = BTRFS_STRIPE_LEN;
4794 map->io_width = BTRFS_STRIPE_LEN;
4796 map->sub_stripes = sub_stripes;
4798 num_bytes = stripe_size * data_stripes;
4800 trace_btrfs_chunk_alloc(info, map, start, num_bytes);
4802 em = alloc_extent_map();
4808 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4809 em->map_lookup = map;
4811 em->len = num_bytes;
4812 em->block_start = 0;
4813 em->block_len = em->len;
4814 em->orig_block_len = stripe_size;
4816 em_tree = &info->mapping_tree.map_tree;
4817 write_lock(&em_tree->lock);
4818 ret = add_extent_mapping(em_tree, em, 0);
4820 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4821 refcount_inc(&em->refs);
4823 write_unlock(&em_tree->lock);
4825 free_extent_map(em);
4829 ret = btrfs_make_block_group(trans, info, 0, type, start, num_bytes);
4831 goto error_del_extent;
4833 for (i = 0; i < map->num_stripes; i++) {
4834 num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4835 btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4838 atomic64_sub(stripe_size * map->num_stripes, &info->free_chunk_space);
4840 free_extent_map(em);
4841 check_raid56_incompat_flag(info, type);
4843 kfree(devices_info);
4847 write_lock(&em_tree->lock);
4848 remove_extent_mapping(em_tree, em);
4849 write_unlock(&em_tree->lock);
4851 /* One for our allocation */
4852 free_extent_map(em);
4853 /* One for the tree reference */
4854 free_extent_map(em);
4855 /* One for the pending_chunks list reference */
4856 free_extent_map(em);
4858 kfree(devices_info);
4862 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4863 struct btrfs_fs_info *fs_info,
4864 u64 chunk_offset, u64 chunk_size)
4866 struct btrfs_root *extent_root = fs_info->extent_root;
4867 struct btrfs_root *chunk_root = fs_info->chunk_root;
4868 struct btrfs_key key;
4869 struct btrfs_device *device;
4870 struct btrfs_chunk *chunk;
4871 struct btrfs_stripe *stripe;
4872 struct extent_map *em;
4873 struct map_lookup *map;
4880 em = get_chunk_map(fs_info, chunk_offset, chunk_size);
4884 map = em->map_lookup;
4885 item_size = btrfs_chunk_item_size(map->num_stripes);
4886 stripe_size = em->orig_block_len;
4888 chunk = kzalloc(item_size, GFP_NOFS);
4895 * Take the device list mutex to prevent races with the final phase of
4896 * a device replace operation that replaces the device object associated
4897 * with the map's stripes, because the device object's id can change
4898 * at any time during that final phase of the device replace operation
4899 * (dev-replace.c:btrfs_dev_replace_finishing()).
4901 mutex_lock(&fs_info->fs_devices->device_list_mutex);
4902 for (i = 0; i < map->num_stripes; i++) {
4903 device = map->stripes[i].dev;
4904 dev_offset = map->stripes[i].physical;
4906 ret = btrfs_update_device(trans, device);
4909 ret = btrfs_alloc_dev_extent(trans, device,
4910 chunk_root->root_key.objectid,
4911 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4912 chunk_offset, dev_offset,
4918 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4922 stripe = &chunk->stripe;
4923 for (i = 0; i < map->num_stripes; i++) {
4924 device = map->stripes[i].dev;
4925 dev_offset = map->stripes[i].physical;
4927 btrfs_set_stack_stripe_devid(stripe, device->devid);
4928 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4929 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4932 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4934 btrfs_set_stack_chunk_length(chunk, chunk_size);
4935 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4936 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4937 btrfs_set_stack_chunk_type(chunk, map->type);
4938 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4939 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4940 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4941 btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
4942 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4944 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4945 key.type = BTRFS_CHUNK_ITEM_KEY;
4946 key.offset = chunk_offset;
4948 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4949 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4951 * TODO: Cleanup of inserted chunk root in case of
4954 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
4959 free_extent_map(em);
4964 * Chunk allocation falls into two parts. The first part does works
4965 * that make the new allocated chunk useable, but not do any operation
4966 * that modifies the chunk tree. The second part does the works that
4967 * require modifying the chunk tree. This division is important for the
4968 * bootstrap process of adding storage to a seed btrfs.
4970 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4971 struct btrfs_fs_info *fs_info, u64 type)
4975 ASSERT(mutex_is_locked(&fs_info->chunk_mutex));
4976 chunk_offset = find_next_chunk(fs_info);
4977 return __btrfs_alloc_chunk(trans, chunk_offset, type);
4980 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4981 struct btrfs_fs_info *fs_info)
4984 u64 sys_chunk_offset;
4988 chunk_offset = find_next_chunk(fs_info);
4989 alloc_profile = btrfs_metadata_alloc_profile(fs_info);
4990 ret = __btrfs_alloc_chunk(trans, chunk_offset, alloc_profile);
4994 sys_chunk_offset = find_next_chunk(fs_info);
4995 alloc_profile = btrfs_system_alloc_profile(fs_info);
4996 ret = __btrfs_alloc_chunk(trans, sys_chunk_offset, alloc_profile);
5000 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
5004 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5005 BTRFS_BLOCK_GROUP_RAID10 |
5006 BTRFS_BLOCK_GROUP_RAID5 |
5007 BTRFS_BLOCK_GROUP_DUP)) {
5009 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
5018 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset)
5020 struct extent_map *em;
5021 struct map_lookup *map;
5026 em = get_chunk_map(fs_info, chunk_offset, 1);
5030 map = em->map_lookup;
5031 for (i = 0; i < map->num_stripes; i++) {
5032 if (map->stripes[i].dev->missing) {
5037 if (!map->stripes[i].dev->writeable) {
5044 * If the number of missing devices is larger than max errors,
5045 * we can not write the data into that chunk successfully, so
5048 if (miss_ndevs > btrfs_chunk_max_errors(map))
5051 free_extent_map(em);
5055 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
5057 extent_map_tree_init(&tree->map_tree);
5060 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
5062 struct extent_map *em;
5065 write_lock(&tree->map_tree.lock);
5066 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
5068 remove_extent_mapping(&tree->map_tree, em);
5069 write_unlock(&tree->map_tree.lock);
5073 free_extent_map(em);
5074 /* once for the tree */
5075 free_extent_map(em);
5079 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5081 struct extent_map *em;
5082 struct map_lookup *map;
5085 em = get_chunk_map(fs_info, logical, len);
5088 * We could return errors for these cases, but that could get
5089 * ugly and we'd probably do the same thing which is just not do
5090 * anything else and exit, so return 1 so the callers don't try
5091 * to use other copies.
5095 map = em->map_lookup;
5096 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
5097 ret = map->num_stripes;
5098 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5099 ret = map->sub_stripes;
5100 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5102 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5106 free_extent_map(em);
5108 btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
5109 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) &&
5110 fs_info->dev_replace.tgtdev)
5112 btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
5117 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
5120 struct extent_map *em;
5121 struct map_lookup *map;
5122 unsigned long len = fs_info->sectorsize;
5124 em = get_chunk_map(fs_info, logical, len);
5126 if (!WARN_ON(IS_ERR(em))) {
5127 map = em->map_lookup;
5128 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5129 len = map->stripe_len * nr_data_stripes(map);
5130 free_extent_map(em);
5135 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5137 struct extent_map *em;
5138 struct map_lookup *map;
5141 em = get_chunk_map(fs_info, logical, len);
5143 if(!WARN_ON(IS_ERR(em))) {
5144 map = em->map_lookup;
5145 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5147 free_extent_map(em);
5152 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5153 struct map_lookup *map, int first, int num,
5154 int optimal, int dev_replace_is_ongoing)
5158 struct btrfs_device *srcdev;
5160 if (dev_replace_is_ongoing &&
5161 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5162 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5163 srcdev = fs_info->dev_replace.srcdev;
5168 * try to avoid the drive that is the source drive for a
5169 * dev-replace procedure, only choose it if no other non-missing
5170 * mirror is available
5172 for (tolerance = 0; tolerance < 2; tolerance++) {
5173 if (map->stripes[optimal].dev->bdev &&
5174 (tolerance || map->stripes[optimal].dev != srcdev))
5176 for (i = first; i < first + num; i++) {
5177 if (map->stripes[i].dev->bdev &&
5178 (tolerance || map->stripes[i].dev != srcdev))
5183 /* we couldn't find one that doesn't fail. Just return something
5184 * and the io error handling code will clean up eventually
5189 static inline int parity_smaller(u64 a, u64 b)
5194 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5195 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5197 struct btrfs_bio_stripe s;
5204 for (i = 0; i < num_stripes - 1; i++) {
5205 if (parity_smaller(bbio->raid_map[i],
5206 bbio->raid_map[i+1])) {
5207 s = bbio->stripes[i];
5208 l = bbio->raid_map[i];
5209 bbio->stripes[i] = bbio->stripes[i+1];
5210 bbio->raid_map[i] = bbio->raid_map[i+1];
5211 bbio->stripes[i+1] = s;
5212 bbio->raid_map[i+1] = l;
5220 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5222 struct btrfs_bio *bbio = kzalloc(
5223 /* the size of the btrfs_bio */
5224 sizeof(struct btrfs_bio) +
5225 /* plus the variable array for the stripes */
5226 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5227 /* plus the variable array for the tgt dev */
5228 sizeof(int) * (real_stripes) +
5230 * plus the raid_map, which includes both the tgt dev
5233 sizeof(u64) * (total_stripes),
5234 GFP_NOFS|__GFP_NOFAIL);
5236 atomic_set(&bbio->error, 0);
5237 refcount_set(&bbio->refs, 1);
5242 void btrfs_get_bbio(struct btrfs_bio *bbio)
5244 WARN_ON(!refcount_read(&bbio->refs));
5245 refcount_inc(&bbio->refs);
5248 void btrfs_put_bbio(struct btrfs_bio *bbio)
5252 if (refcount_dec_and_test(&bbio->refs))
5256 /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */
5258 * Please note that, discard won't be sent to target device of device
5261 static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info,
5262 u64 logical, u64 length,
5263 struct btrfs_bio **bbio_ret)
5265 struct extent_map *em;
5266 struct map_lookup *map;
5267 struct btrfs_bio *bbio;
5271 u64 stripe_end_offset;
5278 u32 sub_stripes = 0;
5279 u64 stripes_per_dev = 0;
5280 u32 remaining_stripes = 0;
5281 u32 last_stripe = 0;
5285 /* discard always return a bbio */
5288 em = get_chunk_map(fs_info, logical, length);
5292 map = em->map_lookup;
5293 /* we don't discard raid56 yet */
5294 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5299 offset = logical - em->start;
5300 length = min_t(u64, em->len - offset, length);
5302 stripe_len = map->stripe_len;
5304 * stripe_nr counts the total number of stripes we have to stride
5305 * to get to this block
5307 stripe_nr = div64_u64(offset, stripe_len);
5309 /* stripe_offset is the offset of this block in its stripe */
5310 stripe_offset = offset - stripe_nr * stripe_len;
5312 stripe_nr_end = round_up(offset + length, map->stripe_len);
5313 stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len);
5314 stripe_cnt = stripe_nr_end - stripe_nr;
5315 stripe_end_offset = stripe_nr_end * map->stripe_len -
5318 * after this, stripe_nr is the number of stripes on this
5319 * device we have to walk to find the data, and stripe_index is
5320 * the number of our device in the stripe array
5324 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5325 BTRFS_BLOCK_GROUP_RAID10)) {
5326 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5329 sub_stripes = map->sub_stripes;
5331 factor = map->num_stripes / sub_stripes;
5332 num_stripes = min_t(u64, map->num_stripes,
5333 sub_stripes * stripe_cnt);
5334 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5335 stripe_index *= sub_stripes;
5336 stripes_per_dev = div_u64_rem(stripe_cnt, factor,
5337 &remaining_stripes);
5338 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5339 last_stripe *= sub_stripes;
5340 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5341 BTRFS_BLOCK_GROUP_DUP)) {
5342 num_stripes = map->num_stripes;
5344 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5348 bbio = alloc_btrfs_bio(num_stripes, 0);
5354 for (i = 0; i < num_stripes; i++) {
5355 bbio->stripes[i].physical =
5356 map->stripes[stripe_index].physical +
5357 stripe_offset + stripe_nr * map->stripe_len;
5358 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5360 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5361 BTRFS_BLOCK_GROUP_RAID10)) {
5362 bbio->stripes[i].length = stripes_per_dev *
5365 if (i / sub_stripes < remaining_stripes)
5366 bbio->stripes[i].length +=
5370 * Special for the first stripe and
5373 * |-------|...|-------|
5377 if (i < sub_stripes)
5378 bbio->stripes[i].length -=
5381 if (stripe_index >= last_stripe &&
5382 stripe_index <= (last_stripe +
5384 bbio->stripes[i].length -=
5387 if (i == sub_stripes - 1)
5390 bbio->stripes[i].length = length;
5394 if (stripe_index == map->num_stripes) {
5401 bbio->map_type = map->type;
5402 bbio->num_stripes = num_stripes;
5404 free_extent_map(em);
5409 * In dev-replace case, for repair case (that's the only case where the mirror
5410 * is selected explicitly when calling btrfs_map_block), blocks left of the
5411 * left cursor can also be read from the target drive.
5413 * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the
5415 * For READ, it also needs to be supported using the same mirror number.
5417 * If the requested block is not left of the left cursor, EIO is returned. This
5418 * can happen because btrfs_num_copies() returns one more in the dev-replace
5421 static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info,
5422 u64 logical, u64 length,
5423 u64 srcdev_devid, int *mirror_num,
5426 struct btrfs_bio *bbio = NULL;
5428 int index_srcdev = 0;
5430 u64 physical_of_found = 0;
5434 ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS,
5435 logical, &length, &bbio, 0, 0);
5437 ASSERT(bbio == NULL);
5441 num_stripes = bbio->num_stripes;
5442 if (*mirror_num > num_stripes) {
5444 * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror,
5445 * that means that the requested area is not left of the left
5448 btrfs_put_bbio(bbio);
5453 * process the rest of the function using the mirror_num of the source
5454 * drive. Therefore look it up first. At the end, patch the device
5455 * pointer to the one of the target drive.
5457 for (i = 0; i < num_stripes; i++) {
5458 if (bbio->stripes[i].dev->devid != srcdev_devid)
5462 * In case of DUP, in order to keep it simple, only add the
5463 * mirror with the lowest physical address
5466 physical_of_found <= bbio->stripes[i].physical)
5471 physical_of_found = bbio->stripes[i].physical;
5474 btrfs_put_bbio(bbio);
5480 *mirror_num = index_srcdev + 1;
5481 *physical = physical_of_found;
5485 static void handle_ops_on_dev_replace(enum btrfs_map_op op,
5486 struct btrfs_bio **bbio_ret,
5487 struct btrfs_dev_replace *dev_replace,
5488 int *num_stripes_ret, int *max_errors_ret)
5490 struct btrfs_bio *bbio = *bbio_ret;
5491 u64 srcdev_devid = dev_replace->srcdev->devid;
5492 int tgtdev_indexes = 0;
5493 int num_stripes = *num_stripes_ret;
5494 int max_errors = *max_errors_ret;
5497 if (op == BTRFS_MAP_WRITE) {
5498 int index_where_to_add;
5501 * duplicate the write operations while the dev replace
5502 * procedure is running. Since the copying of the old disk to
5503 * the new disk takes place at run time while the filesystem is
5504 * mounted writable, the regular write operations to the old
5505 * disk have to be duplicated to go to the new disk as well.
5507 * Note that device->missing is handled by the caller, and that
5508 * the write to the old disk is already set up in the stripes
5511 index_where_to_add = num_stripes;
5512 for (i = 0; i < num_stripes; i++) {
5513 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5514 /* write to new disk, too */
5515 struct btrfs_bio_stripe *new =
5516 bbio->stripes + index_where_to_add;
5517 struct btrfs_bio_stripe *old =
5520 new->physical = old->physical;
5521 new->length = old->length;
5522 new->dev = dev_replace->tgtdev;
5523 bbio->tgtdev_map[i] = index_where_to_add;
5524 index_where_to_add++;
5529 num_stripes = index_where_to_add;
5530 } else if (op == BTRFS_MAP_GET_READ_MIRRORS) {
5531 int index_srcdev = 0;
5533 u64 physical_of_found = 0;
5536 * During the dev-replace procedure, the target drive can also
5537 * be used to read data in case it is needed to repair a corrupt
5538 * block elsewhere. This is possible if the requested area is
5539 * left of the left cursor. In this area, the target drive is a
5540 * full copy of the source drive.
5542 for (i = 0; i < num_stripes; i++) {
5543 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5545 * In case of DUP, in order to keep it simple,
5546 * only add the mirror with the lowest physical
5550 physical_of_found <=
5551 bbio->stripes[i].physical)
5555 physical_of_found = bbio->stripes[i].physical;
5559 struct btrfs_bio_stripe *tgtdev_stripe =
5560 bbio->stripes + num_stripes;
5562 tgtdev_stripe->physical = physical_of_found;
5563 tgtdev_stripe->length =
5564 bbio->stripes[index_srcdev].length;
5565 tgtdev_stripe->dev = dev_replace->tgtdev;
5566 bbio->tgtdev_map[index_srcdev] = num_stripes;
5573 *num_stripes_ret = num_stripes;
5574 *max_errors_ret = max_errors;
5575 bbio->num_tgtdevs = tgtdev_indexes;
5579 static bool need_full_stripe(enum btrfs_map_op op)
5581 return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS);
5584 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
5585 enum btrfs_map_op op,
5586 u64 logical, u64 *length,
5587 struct btrfs_bio **bbio_ret,
5588 int mirror_num, int need_raid_map)
5590 struct extent_map *em;
5591 struct map_lookup *map;
5601 int tgtdev_indexes = 0;
5602 struct btrfs_bio *bbio = NULL;
5603 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
5604 int dev_replace_is_ongoing = 0;
5605 int num_alloc_stripes;
5606 int patch_the_first_stripe_for_dev_replace = 0;
5607 u64 physical_to_patch_in_first_stripe = 0;
5608 u64 raid56_full_stripe_start = (u64)-1;
5610 if (op == BTRFS_MAP_DISCARD)
5611 return __btrfs_map_block_for_discard(fs_info, logical,
5614 em = get_chunk_map(fs_info, logical, *length);
5618 map = em->map_lookup;
5619 offset = logical - em->start;
5621 stripe_len = map->stripe_len;
5624 * stripe_nr counts the total number of stripes we have to stride
5625 * to get to this block
5627 stripe_nr = div64_u64(stripe_nr, stripe_len);
5629 stripe_offset = stripe_nr * stripe_len;
5630 if (offset < stripe_offset) {
5632 "stripe math has gone wrong, stripe_offset=%llu, offset=%llu, start=%llu, logical=%llu, stripe_len=%llu",
5633 stripe_offset, offset, em->start, logical,
5635 free_extent_map(em);
5639 /* stripe_offset is the offset of this block in its stripe*/
5640 stripe_offset = offset - stripe_offset;
5642 /* if we're here for raid56, we need to know the stripe aligned start */
5643 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5644 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5645 raid56_full_stripe_start = offset;
5647 /* allow a write of a full stripe, but make sure we don't
5648 * allow straddling of stripes
5650 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5652 raid56_full_stripe_start *= full_stripe_len;
5655 if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5657 /* For writes to RAID[56], allow a full stripeset across all disks.
5658 For other RAID types and for RAID[56] reads, just allow a single
5659 stripe (on a single disk). */
5660 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
5661 (op == BTRFS_MAP_WRITE)) {
5662 max_len = stripe_len * nr_data_stripes(map) -
5663 (offset - raid56_full_stripe_start);
5665 /* we limit the length of each bio to what fits in a stripe */
5666 max_len = stripe_len - stripe_offset;
5668 *length = min_t(u64, em->len - offset, max_len);
5670 *length = em->len - offset;
5673 /* This is for when we're called from btrfs_merge_bio_hook() and all
5674 it cares about is the length */
5678 btrfs_dev_replace_lock(dev_replace, 0);
5679 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5680 if (!dev_replace_is_ongoing)
5681 btrfs_dev_replace_unlock(dev_replace, 0);
5683 btrfs_dev_replace_set_lock_blocking(dev_replace);
5685 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
5686 !need_full_stripe(op) && dev_replace->tgtdev != NULL) {
5687 ret = get_extra_mirror_from_replace(fs_info, logical, *length,
5688 dev_replace->srcdev->devid,
5690 &physical_to_patch_in_first_stripe);
5694 patch_the_first_stripe_for_dev_replace = 1;
5695 } else if (mirror_num > map->num_stripes) {
5701 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5702 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5704 if (op != BTRFS_MAP_WRITE && op != BTRFS_MAP_GET_READ_MIRRORS)
5706 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
5707 if (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS)
5708 num_stripes = map->num_stripes;
5709 else if (mirror_num)
5710 stripe_index = mirror_num - 1;
5712 stripe_index = find_live_mirror(fs_info, map, 0,
5714 current->pid % map->num_stripes,
5715 dev_replace_is_ongoing);
5716 mirror_num = stripe_index + 1;
5719 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
5720 if (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS) {
5721 num_stripes = map->num_stripes;
5722 } else if (mirror_num) {
5723 stripe_index = mirror_num - 1;
5728 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5729 u32 factor = map->num_stripes / map->sub_stripes;
5731 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5732 stripe_index *= map->sub_stripes;
5734 if (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS)
5735 num_stripes = map->sub_stripes;
5736 else if (mirror_num)
5737 stripe_index += mirror_num - 1;
5739 int old_stripe_index = stripe_index;
5740 stripe_index = find_live_mirror(fs_info, map,
5742 map->sub_stripes, stripe_index +
5743 current->pid % map->sub_stripes,
5744 dev_replace_is_ongoing);
5745 mirror_num = stripe_index - old_stripe_index + 1;
5748 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5749 if (need_raid_map &&
5750 (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS ||
5752 /* push stripe_nr back to the start of the full stripe */
5753 stripe_nr = div64_u64(raid56_full_stripe_start,
5754 stripe_len * nr_data_stripes(map));
5756 /* RAID[56] write or recovery. Return all stripes */
5757 num_stripes = map->num_stripes;
5758 max_errors = nr_parity_stripes(map);
5760 *length = map->stripe_len;
5765 * Mirror #0 or #1 means the original data block.
5766 * Mirror #2 is RAID5 parity block.
5767 * Mirror #3 is RAID6 Q block.
5769 stripe_nr = div_u64_rem(stripe_nr,
5770 nr_data_stripes(map), &stripe_index);
5772 stripe_index = nr_data_stripes(map) +
5775 /* We distribute the parity blocks across stripes */
5776 div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
5778 if ((op != BTRFS_MAP_WRITE &&
5779 op != BTRFS_MAP_GET_READ_MIRRORS) &&
5785 * after this, stripe_nr is the number of stripes on this
5786 * device we have to walk to find the data, and stripe_index is
5787 * the number of our device in the stripe array
5789 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5791 mirror_num = stripe_index + 1;
5793 if (stripe_index >= map->num_stripes) {
5795 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
5796 stripe_index, map->num_stripes);
5801 num_alloc_stripes = num_stripes;
5802 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) {
5803 if (op == BTRFS_MAP_WRITE)
5804 num_alloc_stripes <<= 1;
5805 if (op == BTRFS_MAP_GET_READ_MIRRORS)
5806 num_alloc_stripes++;
5807 tgtdev_indexes = num_stripes;
5810 bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
5815 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL)
5816 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
5818 /* build raid_map */
5819 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map &&
5820 (need_full_stripe(op) || mirror_num > 1)) {
5824 bbio->raid_map = (u64 *)((void *)bbio->stripes +
5825 sizeof(struct btrfs_bio_stripe) *
5827 sizeof(int) * tgtdev_indexes);
5829 /* Work out the disk rotation on this stripe-set */
5830 div_u64_rem(stripe_nr, num_stripes, &rot);
5832 /* Fill in the logical address of each stripe */
5833 tmp = stripe_nr * nr_data_stripes(map);
5834 for (i = 0; i < nr_data_stripes(map); i++)
5835 bbio->raid_map[(i+rot) % num_stripes] =
5836 em->start + (tmp + i) * map->stripe_len;
5838 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5839 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5840 bbio->raid_map[(i+rot+1) % num_stripes] =
5845 for (i = 0; i < num_stripes; i++) {
5846 bbio->stripes[i].physical =
5847 map->stripes[stripe_index].physical +
5849 stripe_nr * map->stripe_len;
5850 bbio->stripes[i].dev =
5851 map->stripes[stripe_index].dev;
5855 if (need_full_stripe(op))
5856 max_errors = btrfs_chunk_max_errors(map);
5859 sort_parity_stripes(bbio, num_stripes);
5861 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
5862 need_full_stripe(op)) {
5863 handle_ops_on_dev_replace(op, &bbio, dev_replace, &num_stripes,
5868 bbio->map_type = map->type;
5869 bbio->num_stripes = num_stripes;
5870 bbio->max_errors = max_errors;
5871 bbio->mirror_num = mirror_num;
5874 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5875 * mirror_num == num_stripes + 1 && dev_replace target drive is
5876 * available as a mirror
5878 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5879 WARN_ON(num_stripes > 1);
5880 bbio->stripes[0].dev = dev_replace->tgtdev;
5881 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5882 bbio->mirror_num = map->num_stripes + 1;
5885 if (dev_replace_is_ongoing) {
5886 btrfs_dev_replace_clear_lock_blocking(dev_replace);
5887 btrfs_dev_replace_unlock(dev_replace, 0);
5889 free_extent_map(em);
5893 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
5894 u64 logical, u64 *length,
5895 struct btrfs_bio **bbio_ret, int mirror_num)
5897 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret,
5901 /* For Scrub/replace */
5902 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
5903 u64 logical, u64 *length,
5904 struct btrfs_bio **bbio_ret)
5906 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1);
5909 int btrfs_rmap_block(struct btrfs_fs_info *fs_info,
5910 u64 chunk_start, u64 physical, u64 devid,
5911 u64 **logical, int *naddrs, int *stripe_len)
5913 struct extent_map *em;
5914 struct map_lookup *map;
5922 em = get_chunk_map(fs_info, chunk_start, 1);
5926 map = em->map_lookup;
5928 rmap_len = map->stripe_len;
5930 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5931 length = div_u64(length, map->num_stripes / map->sub_stripes);
5932 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5933 length = div_u64(length, map->num_stripes);
5934 else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5935 length = div_u64(length, nr_data_stripes(map));
5936 rmap_len = map->stripe_len * nr_data_stripes(map);
5939 buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
5940 BUG_ON(!buf); /* -ENOMEM */
5942 for (i = 0; i < map->num_stripes; i++) {
5943 if (devid && map->stripes[i].dev->devid != devid)
5945 if (map->stripes[i].physical > physical ||
5946 map->stripes[i].physical + length <= physical)
5949 stripe_nr = physical - map->stripes[i].physical;
5950 stripe_nr = div64_u64(stripe_nr, map->stripe_len);
5952 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5953 stripe_nr = stripe_nr * map->num_stripes + i;
5954 stripe_nr = div_u64(stripe_nr, map->sub_stripes);
5955 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5956 stripe_nr = stripe_nr * map->num_stripes + i;
5957 } /* else if RAID[56], multiply by nr_data_stripes().
5958 * Alternatively, just use rmap_len below instead of
5959 * map->stripe_len */
5961 bytenr = chunk_start + stripe_nr * rmap_len;
5962 WARN_ON(nr >= map->num_stripes);
5963 for (j = 0; j < nr; j++) {
5964 if (buf[j] == bytenr)
5968 WARN_ON(nr >= map->num_stripes);
5975 *stripe_len = rmap_len;
5977 free_extent_map(em);
5981 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
5983 bio->bi_private = bbio->private;
5984 bio->bi_end_io = bbio->end_io;
5987 btrfs_put_bbio(bbio);
5990 static void btrfs_end_bio(struct bio *bio)
5992 struct btrfs_bio *bbio = bio->bi_private;
5993 int is_orig_bio = 0;
5995 if (bio->bi_status) {
5996 atomic_inc(&bbio->error);
5997 if (bio->bi_status == BLK_STS_IOERR ||
5998 bio->bi_status == BLK_STS_TARGET) {
5999 unsigned int stripe_index =
6000 btrfs_io_bio(bio)->stripe_index;
6001 struct btrfs_device *dev;
6003 BUG_ON(stripe_index >= bbio->num_stripes);
6004 dev = bbio->stripes[stripe_index].dev;
6006 if (bio_op(bio) == REQ_OP_WRITE)
6007 btrfs_dev_stat_inc(dev,
6008 BTRFS_DEV_STAT_WRITE_ERRS);
6010 btrfs_dev_stat_inc(dev,
6011 BTRFS_DEV_STAT_READ_ERRS);
6012 if (bio->bi_opf & REQ_PREFLUSH)
6013 btrfs_dev_stat_inc(dev,
6014 BTRFS_DEV_STAT_FLUSH_ERRS);
6015 btrfs_dev_stat_print_on_error(dev);
6020 if (bio == bbio->orig_bio)
6023 btrfs_bio_counter_dec(bbio->fs_info);
6025 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6028 bio = bbio->orig_bio;
6031 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6032 /* only send an error to the higher layers if it is
6033 * beyond the tolerance of the btrfs bio
6035 if (atomic_read(&bbio->error) > bbio->max_errors) {
6036 bio->bi_status = BLK_STS_IOERR;
6039 * this bio is actually up to date, we didn't
6040 * go over the max number of errors
6045 btrfs_end_bbio(bbio, bio);
6046 } else if (!is_orig_bio) {
6052 * see run_scheduled_bios for a description of why bios are collected for
6055 * This will add one bio to the pending list for a device and make sure
6056 * the work struct is scheduled.
6058 static noinline void btrfs_schedule_bio(struct btrfs_device *device,
6061 struct btrfs_fs_info *fs_info = device->fs_info;
6062 int should_queue = 1;
6063 struct btrfs_pending_bios *pending_bios;
6065 if (device->missing || !device->bdev) {
6070 /* don't bother with additional async steps for reads, right now */
6071 if (bio_op(bio) == REQ_OP_READ) {
6073 btrfsic_submit_bio(bio);
6079 * nr_async_bios allows us to reliably return congestion to the
6080 * higher layers. Otherwise, the async bio makes it appear we have
6081 * made progress against dirty pages when we've really just put it
6082 * on a queue for later
6084 atomic_inc(&fs_info->nr_async_bios);
6085 WARN_ON(bio->bi_next);
6086 bio->bi_next = NULL;
6088 spin_lock(&device->io_lock);
6089 if (op_is_sync(bio->bi_opf))
6090 pending_bios = &device->pending_sync_bios;
6092 pending_bios = &device->pending_bios;
6094 if (pending_bios->tail)
6095 pending_bios->tail->bi_next = bio;
6097 pending_bios->tail = bio;
6098 if (!pending_bios->head)
6099 pending_bios->head = bio;
6100 if (device->running_pending)
6103 spin_unlock(&device->io_lock);
6106 btrfs_queue_work(fs_info->submit_workers, &device->work);
6109 static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio,
6110 u64 physical, int dev_nr, int async)
6112 struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
6113 struct btrfs_fs_info *fs_info = bbio->fs_info;
6115 bio->bi_private = bbio;
6116 btrfs_io_bio(bio)->stripe_index = dev_nr;
6117 bio->bi_end_io = btrfs_end_bio;
6118 bio->bi_iter.bi_sector = physical >> 9;
6121 struct rcu_string *name;
6124 name = rcu_dereference(dev->name);
6125 btrfs_debug(fs_info,
6126 "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
6127 bio_op(bio), bio->bi_opf,
6128 (u64)bio->bi_iter.bi_sector,
6129 (u_long)dev->bdev->bd_dev, name->str, dev->devid,
6130 bio->bi_iter.bi_size);
6134 bio->bi_bdev = dev->bdev;
6136 btrfs_bio_counter_inc_noblocked(fs_info);
6139 btrfs_schedule_bio(dev, bio);
6141 btrfsic_submit_bio(bio);
6144 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6146 atomic_inc(&bbio->error);
6147 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6148 /* Should be the original bio. */
6149 WARN_ON(bio != bbio->orig_bio);
6151 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6152 bio->bi_iter.bi_sector = logical >> 9;
6153 bio->bi_status = BLK_STS_IOERR;
6154 btrfs_end_bbio(bbio, bio);
6158 int btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
6159 int mirror_num, int async_submit)
6161 struct btrfs_device *dev;
6162 struct bio *first_bio = bio;
6163 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
6169 struct btrfs_bio *bbio = NULL;
6171 length = bio->bi_iter.bi_size;
6172 map_length = length;
6174 btrfs_bio_counter_inc_blocked(fs_info);
6175 ret = __btrfs_map_block(fs_info, bio_op(bio), logical,
6176 &map_length, &bbio, mirror_num, 1);
6178 btrfs_bio_counter_dec(fs_info);
6182 total_devs = bbio->num_stripes;
6183 bbio->orig_bio = first_bio;
6184 bbio->private = first_bio->bi_private;
6185 bbio->end_io = first_bio->bi_end_io;
6186 bbio->fs_info = fs_info;
6187 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6189 if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6190 ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) {
6191 /* In this case, map_length has been set to the length of
6192 a single stripe; not the whole write */
6193 if (bio_op(bio) == REQ_OP_WRITE) {
6194 ret = raid56_parity_write(fs_info, bio, bbio,
6197 ret = raid56_parity_recover(fs_info, bio, bbio,
6198 map_length, mirror_num, 1);
6201 btrfs_bio_counter_dec(fs_info);
6205 if (map_length < length) {
6207 "mapping failed logical %llu bio len %llu len %llu",
6208 logical, length, map_length);
6212 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6213 dev = bbio->stripes[dev_nr].dev;
6214 if (!dev || !dev->bdev ||
6215 (bio_op(first_bio) == REQ_OP_WRITE && !dev->writeable)) {
6216 bbio_error(bbio, first_bio, logical);
6220 if (dev_nr < total_devs - 1)
6221 bio = btrfs_bio_clone(first_bio);
6225 submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical,
6226 dev_nr, async_submit);
6228 btrfs_bio_counter_dec(fs_info);
6232 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
6235 struct btrfs_device *device;
6236 struct btrfs_fs_devices *cur_devices;
6238 cur_devices = fs_info->fs_devices;
6239 while (cur_devices) {
6241 !memcmp(cur_devices->fsid, fsid, BTRFS_FSID_SIZE)) {
6242 device = find_device(cur_devices, devid, uuid);
6246 cur_devices = cur_devices->seed;
6251 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices,
6252 u64 devid, u8 *dev_uuid)
6254 struct btrfs_device *device;
6256 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6260 list_add(&device->dev_list, &fs_devices->devices);
6261 device->fs_devices = fs_devices;
6262 fs_devices->num_devices++;
6264 device->missing = 1;
6265 fs_devices->missing_devices++;
6271 * btrfs_alloc_device - allocate struct btrfs_device
6272 * @fs_info: used only for generating a new devid, can be NULL if
6273 * devid is provided (i.e. @devid != NULL).
6274 * @devid: a pointer to devid for this device. If NULL a new devid
6276 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6279 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6280 * on error. Returned struct is not linked onto any lists and can be
6281 * destroyed with kfree() right away.
6283 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6287 struct btrfs_device *dev;
6290 if (WARN_ON(!devid && !fs_info))
6291 return ERR_PTR(-EINVAL);
6293 dev = __alloc_device();
6302 ret = find_next_devid(fs_info, &tmp);
6305 return ERR_PTR(ret);
6311 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6313 generate_random_uuid(dev->uuid);
6315 btrfs_init_work(&dev->work, btrfs_submit_helper,
6316 pending_bios_fn, NULL, NULL);
6321 /* Return -EIO if any error, otherwise return 0. */
6322 static int btrfs_check_chunk_valid(struct btrfs_fs_info *fs_info,
6323 struct extent_buffer *leaf,
6324 struct btrfs_chunk *chunk, u64 logical)
6332 length = btrfs_chunk_length(leaf, chunk);
6333 stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6334 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6335 sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6336 type = btrfs_chunk_type(leaf, chunk);
6339 btrfs_err(fs_info, "invalid chunk num_stripes: %u",
6343 if (!IS_ALIGNED(logical, fs_info->sectorsize)) {
6344 btrfs_err(fs_info, "invalid chunk logical %llu", logical);
6347 if (btrfs_chunk_sector_size(leaf, chunk) != fs_info->sectorsize) {
6348 btrfs_err(fs_info, "invalid chunk sectorsize %u",
6349 btrfs_chunk_sector_size(leaf, chunk));
6352 if (!length || !IS_ALIGNED(length, fs_info->sectorsize)) {
6353 btrfs_err(fs_info, "invalid chunk length %llu", length);
6356 if (!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN) {
6357 btrfs_err(fs_info, "invalid chunk stripe length: %llu",
6361 if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6363 btrfs_err(fs_info, "unrecognized chunk type: %llu",
6364 ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
6365 BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6366 btrfs_chunk_type(leaf, chunk));
6369 if ((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes != 2) ||
6370 (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) ||
6371 (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) ||
6372 (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) ||
6373 (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) ||
6374 ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
6375 num_stripes != 1)) {
6377 "invalid num_stripes:sub_stripes %u:%u for profile %llu",
6378 num_stripes, sub_stripes,
6379 type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
6386 static int read_one_chunk(struct btrfs_fs_info *fs_info, struct btrfs_key *key,
6387 struct extent_buffer *leaf,
6388 struct btrfs_chunk *chunk)
6390 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
6391 struct map_lookup *map;
6392 struct extent_map *em;
6396 u8 uuid[BTRFS_UUID_SIZE];
6401 logical = key->offset;
6402 length = btrfs_chunk_length(leaf, chunk);
6403 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6405 ret = btrfs_check_chunk_valid(fs_info, leaf, chunk, logical);
6409 read_lock(&map_tree->map_tree.lock);
6410 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
6411 read_unlock(&map_tree->map_tree.lock);
6413 /* already mapped? */
6414 if (em && em->start <= logical && em->start + em->len > logical) {
6415 free_extent_map(em);
6418 free_extent_map(em);
6421 em = alloc_extent_map();
6424 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6426 free_extent_map(em);
6430 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6431 em->map_lookup = map;
6432 em->start = logical;
6435 em->block_start = 0;
6436 em->block_len = em->len;
6438 map->num_stripes = num_stripes;
6439 map->io_width = btrfs_chunk_io_width(leaf, chunk);
6440 map->io_align = btrfs_chunk_io_align(leaf, chunk);
6441 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6442 map->type = btrfs_chunk_type(leaf, chunk);
6443 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6444 for (i = 0; i < num_stripes; i++) {
6445 map->stripes[i].physical =
6446 btrfs_stripe_offset_nr(leaf, chunk, i);
6447 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6448 read_extent_buffer(leaf, uuid, (unsigned long)
6449 btrfs_stripe_dev_uuid_nr(chunk, i),
6451 map->stripes[i].dev = btrfs_find_device(fs_info, devid,
6453 if (!map->stripes[i].dev &&
6454 !btrfs_test_opt(fs_info, DEGRADED)) {
6455 free_extent_map(em);
6456 btrfs_report_missing_device(fs_info, devid, uuid);
6459 if (!map->stripes[i].dev) {
6460 map->stripes[i].dev =
6461 add_missing_dev(fs_info->fs_devices, devid,
6463 if (!map->stripes[i].dev) {
6464 free_extent_map(em);
6467 btrfs_report_missing_device(fs_info, devid, uuid);
6469 map->stripes[i].dev->in_fs_metadata = 1;
6472 write_lock(&map_tree->map_tree.lock);
6473 ret = add_extent_mapping(&map_tree->map_tree, em, 0);
6474 write_unlock(&map_tree->map_tree.lock);
6475 BUG_ON(ret); /* Tree corruption */
6476 free_extent_map(em);
6481 static void fill_device_from_item(struct extent_buffer *leaf,
6482 struct btrfs_dev_item *dev_item,
6483 struct btrfs_device *device)
6487 device->devid = btrfs_device_id(leaf, dev_item);
6488 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6489 device->total_bytes = device->disk_total_bytes;
6490 device->commit_total_bytes = device->disk_total_bytes;
6491 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6492 device->commit_bytes_used = device->bytes_used;
6493 device->type = btrfs_device_type(leaf, dev_item);
6494 device->io_align = btrfs_device_io_align(leaf, dev_item);
6495 device->io_width = btrfs_device_io_width(leaf, dev_item);
6496 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6497 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6498 device->is_tgtdev_for_dev_replace = 0;
6500 ptr = btrfs_device_uuid(dev_item);
6501 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6504 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info,
6507 struct btrfs_fs_devices *fs_devices;
6510 BUG_ON(!mutex_is_locked(&uuid_mutex));
6513 fs_devices = fs_info->fs_devices->seed;
6514 while (fs_devices) {
6515 if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE))
6518 fs_devices = fs_devices->seed;
6521 fs_devices = find_fsid(fsid);
6523 if (!btrfs_test_opt(fs_info, DEGRADED))
6524 return ERR_PTR(-ENOENT);
6526 fs_devices = alloc_fs_devices(fsid);
6527 if (IS_ERR(fs_devices))
6530 fs_devices->seeding = 1;
6531 fs_devices->opened = 1;
6535 fs_devices = clone_fs_devices(fs_devices);
6536 if (IS_ERR(fs_devices))
6539 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
6540 fs_info->bdev_holder);
6542 free_fs_devices(fs_devices);
6543 fs_devices = ERR_PTR(ret);
6547 if (!fs_devices->seeding) {
6548 __btrfs_close_devices(fs_devices);
6549 free_fs_devices(fs_devices);
6550 fs_devices = ERR_PTR(-EINVAL);
6554 fs_devices->seed = fs_info->fs_devices->seed;
6555 fs_info->fs_devices->seed = fs_devices;
6560 static int read_one_dev(struct btrfs_fs_info *fs_info,
6561 struct extent_buffer *leaf,
6562 struct btrfs_dev_item *dev_item)
6564 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6565 struct btrfs_device *device;
6568 u8 fs_uuid[BTRFS_FSID_SIZE];
6569 u8 dev_uuid[BTRFS_UUID_SIZE];
6571 devid = btrfs_device_id(leaf, dev_item);
6572 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6574 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6577 if (memcmp(fs_uuid, fs_info->fsid, BTRFS_FSID_SIZE)) {
6578 fs_devices = open_seed_devices(fs_info, fs_uuid);
6579 if (IS_ERR(fs_devices))
6580 return PTR_ERR(fs_devices);
6583 device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
6585 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6586 btrfs_report_missing_device(fs_info, devid, dev_uuid);
6590 device = add_missing_dev(fs_devices, devid, dev_uuid);
6593 btrfs_report_missing_device(fs_info, devid, dev_uuid);
6595 if (!device->bdev) {
6596 btrfs_report_missing_device(fs_info, devid, dev_uuid);
6597 if (!btrfs_test_opt(fs_info, DEGRADED))
6601 if(!device->bdev && !device->missing) {
6603 * this happens when a device that was properly setup
6604 * in the device info lists suddenly goes bad.
6605 * device->bdev is NULL, and so we have to set
6606 * device->missing to one here
6608 device->fs_devices->missing_devices++;
6609 device->missing = 1;
6612 /* Move the device to its own fs_devices */
6613 if (device->fs_devices != fs_devices) {
6614 ASSERT(device->missing);
6616 list_move(&device->dev_list, &fs_devices->devices);
6617 device->fs_devices->num_devices--;
6618 fs_devices->num_devices++;
6620 device->fs_devices->missing_devices--;
6621 fs_devices->missing_devices++;
6623 device->fs_devices = fs_devices;
6627 if (device->fs_devices != fs_info->fs_devices) {
6628 BUG_ON(device->writeable);
6629 if (device->generation !=
6630 btrfs_device_generation(leaf, dev_item))
6634 fill_device_from_item(leaf, dev_item, device);
6635 device->in_fs_metadata = 1;
6636 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
6637 device->fs_devices->total_rw_bytes += device->total_bytes;
6638 atomic64_add(device->total_bytes - device->bytes_used,
6639 &fs_info->free_chunk_space);
6645 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
6647 struct btrfs_root *root = fs_info->tree_root;
6648 struct btrfs_super_block *super_copy = fs_info->super_copy;
6649 struct extent_buffer *sb;
6650 struct btrfs_disk_key *disk_key;
6651 struct btrfs_chunk *chunk;
6653 unsigned long sb_array_offset;
6660 struct btrfs_key key;
6662 ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize);
6664 * This will create extent buffer of nodesize, superblock size is
6665 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6666 * overallocate but we can keep it as-is, only the first page is used.
6668 sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET);
6671 set_extent_buffer_uptodate(sb);
6672 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6674 * The sb extent buffer is artificial and just used to read the system array.
6675 * set_extent_buffer_uptodate() call does not properly mark all it's
6676 * pages up-to-date when the page is larger: extent does not cover the
6677 * whole page and consequently check_page_uptodate does not find all
6678 * the page's extents up-to-date (the hole beyond sb),
6679 * write_extent_buffer then triggers a WARN_ON.
6681 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6682 * but sb spans only this function. Add an explicit SetPageUptodate call
6683 * to silence the warning eg. on PowerPC 64.
6685 if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
6686 SetPageUptodate(sb->pages[0]);
6688 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6689 array_size = btrfs_super_sys_array_size(super_copy);
6691 array_ptr = super_copy->sys_chunk_array;
6692 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6695 while (cur_offset < array_size) {
6696 disk_key = (struct btrfs_disk_key *)array_ptr;
6697 len = sizeof(*disk_key);
6698 if (cur_offset + len > array_size)
6699 goto out_short_read;
6701 btrfs_disk_key_to_cpu(&key, disk_key);
6704 sb_array_offset += len;
6707 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6708 chunk = (struct btrfs_chunk *)sb_array_offset;
6710 * At least one btrfs_chunk with one stripe must be
6711 * present, exact stripe count check comes afterwards
6713 len = btrfs_chunk_item_size(1);
6714 if (cur_offset + len > array_size)
6715 goto out_short_read;
6717 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6720 "invalid number of stripes %u in sys_array at offset %u",
6721 num_stripes, cur_offset);
6726 type = btrfs_chunk_type(sb, chunk);
6727 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
6729 "invalid chunk type %llu in sys_array at offset %u",
6735 len = btrfs_chunk_item_size(num_stripes);
6736 if (cur_offset + len > array_size)
6737 goto out_short_read;
6739 ret = read_one_chunk(fs_info, &key, sb, chunk);
6744 "unexpected item type %u in sys_array at offset %u",
6745 (u32)key.type, cur_offset);
6750 sb_array_offset += len;
6753 clear_extent_buffer_uptodate(sb);
6754 free_extent_buffer_stale(sb);
6758 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
6760 clear_extent_buffer_uptodate(sb);
6761 free_extent_buffer_stale(sb);
6765 void btrfs_report_missing_device(struct btrfs_fs_info *fs_info, u64 devid,
6768 btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing", devid, uuid);
6772 * Check if all chunks in the fs are OK for read-write degraded mount
6774 * Return true if all chunks meet the minimal RW mount requirements.
6775 * Return false if any chunk doesn't meet the minimal RW mount requirements.
6777 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info)
6779 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
6780 struct extent_map *em;
6784 read_lock(&map_tree->map_tree.lock);
6785 em = lookup_extent_mapping(&map_tree->map_tree, 0, (u64)-1);
6786 read_unlock(&map_tree->map_tree.lock);
6787 /* No chunk at all? Return false anyway */
6793 struct map_lookup *map;
6798 map = em->map_lookup;
6800 btrfs_get_num_tolerated_disk_barrier_failures(
6802 for (i = 0; i < map->num_stripes; i++) {
6803 struct btrfs_device *dev = map->stripes[i].dev;
6805 if (!dev || !dev->bdev || dev->missing ||
6806 dev->last_flush_error)
6809 if (missing > max_tolerated) {
6811 "chunk %llu missing %d devices, max tolerance is %d for writeable mount",
6812 em->start, missing, max_tolerated);
6813 free_extent_map(em);
6817 next_start = extent_map_end(em);
6818 free_extent_map(em);
6820 read_lock(&map_tree->map_tree.lock);
6821 em = lookup_extent_mapping(&map_tree->map_tree, next_start,
6822 (u64)(-1) - next_start);
6823 read_unlock(&map_tree->map_tree.lock);
6829 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
6831 struct btrfs_root *root = fs_info->chunk_root;
6832 struct btrfs_path *path;
6833 struct extent_buffer *leaf;
6834 struct btrfs_key key;
6835 struct btrfs_key found_key;
6840 path = btrfs_alloc_path();
6844 mutex_lock(&uuid_mutex);
6845 mutex_lock(&fs_info->chunk_mutex);
6848 * Read all device items, and then all the chunk items. All
6849 * device items are found before any chunk item (their object id
6850 * is smaller than the lowest possible object id for a chunk
6851 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6853 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6856 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6860 leaf = path->nodes[0];
6861 slot = path->slots[0];
6862 if (slot >= btrfs_header_nritems(leaf)) {
6863 ret = btrfs_next_leaf(root, path);
6870 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6871 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6872 struct btrfs_dev_item *dev_item;
6873 dev_item = btrfs_item_ptr(leaf, slot,
6874 struct btrfs_dev_item);
6875 ret = read_one_dev(fs_info, leaf, dev_item);
6879 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6880 struct btrfs_chunk *chunk;
6881 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6882 ret = read_one_chunk(fs_info, &found_key, leaf, chunk);
6890 * After loading chunk tree, we've got all device information,
6891 * do another round of validation checks.
6893 if (total_dev != fs_info->fs_devices->total_devices) {
6895 "super_num_devices %llu mismatch with num_devices %llu found here",
6896 btrfs_super_num_devices(fs_info->super_copy),
6901 if (btrfs_super_total_bytes(fs_info->super_copy) <
6902 fs_info->fs_devices->total_rw_bytes) {
6904 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
6905 btrfs_super_total_bytes(fs_info->super_copy),
6906 fs_info->fs_devices->total_rw_bytes);
6912 mutex_unlock(&fs_info->chunk_mutex);
6913 mutex_unlock(&uuid_mutex);
6915 btrfs_free_path(path);
6919 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6921 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6922 struct btrfs_device *device;
6924 while (fs_devices) {
6925 mutex_lock(&fs_devices->device_list_mutex);
6926 list_for_each_entry(device, &fs_devices->devices, dev_list)
6927 device->fs_info = fs_info;
6928 mutex_unlock(&fs_devices->device_list_mutex);
6930 fs_devices = fs_devices->seed;
6934 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6938 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6939 btrfs_dev_stat_reset(dev, i);
6942 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6944 struct btrfs_key key;
6945 struct btrfs_key found_key;
6946 struct btrfs_root *dev_root = fs_info->dev_root;
6947 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6948 struct extent_buffer *eb;
6951 struct btrfs_device *device;
6952 struct btrfs_path *path = NULL;
6955 path = btrfs_alloc_path();
6961 mutex_lock(&fs_devices->device_list_mutex);
6962 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6964 struct btrfs_dev_stats_item *ptr;
6966 key.objectid = BTRFS_DEV_STATS_OBJECTID;
6967 key.type = BTRFS_PERSISTENT_ITEM_KEY;
6968 key.offset = device->devid;
6969 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6971 __btrfs_reset_dev_stats(device);
6972 device->dev_stats_valid = 1;
6973 btrfs_release_path(path);
6976 slot = path->slots[0];
6977 eb = path->nodes[0];
6978 btrfs_item_key_to_cpu(eb, &found_key, slot);
6979 item_size = btrfs_item_size_nr(eb, slot);
6981 ptr = btrfs_item_ptr(eb, slot,
6982 struct btrfs_dev_stats_item);
6984 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6985 if (item_size >= (1 + i) * sizeof(__le64))
6986 btrfs_dev_stat_set(device, i,
6987 btrfs_dev_stats_value(eb, ptr, i));
6989 btrfs_dev_stat_reset(device, i);
6992 device->dev_stats_valid = 1;
6993 btrfs_dev_stat_print_on_load(device);
6994 btrfs_release_path(path);
6996 mutex_unlock(&fs_devices->device_list_mutex);
6999 btrfs_free_path(path);
7000 return ret < 0 ? ret : 0;
7003 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
7004 struct btrfs_fs_info *fs_info,
7005 struct btrfs_device *device)
7007 struct btrfs_root *dev_root = fs_info->dev_root;
7008 struct btrfs_path *path;
7009 struct btrfs_key key;
7010 struct extent_buffer *eb;
7011 struct btrfs_dev_stats_item *ptr;
7015 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7016 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7017 key.offset = device->devid;
7019 path = btrfs_alloc_path();
7022 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
7024 btrfs_warn_in_rcu(fs_info,
7025 "error %d while searching for dev_stats item for device %s",
7026 ret, rcu_str_deref(device->name));
7031 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
7032 /* need to delete old one and insert a new one */
7033 ret = btrfs_del_item(trans, dev_root, path);
7035 btrfs_warn_in_rcu(fs_info,
7036 "delete too small dev_stats item for device %s failed %d",
7037 rcu_str_deref(device->name), ret);
7044 /* need to insert a new item */
7045 btrfs_release_path(path);
7046 ret = btrfs_insert_empty_item(trans, dev_root, path,
7047 &key, sizeof(*ptr));
7049 btrfs_warn_in_rcu(fs_info,
7050 "insert dev_stats item for device %s failed %d",
7051 rcu_str_deref(device->name), ret);
7056 eb = path->nodes[0];
7057 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7058 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7059 btrfs_set_dev_stats_value(eb, ptr, i,
7060 btrfs_dev_stat_read(device, i));
7061 btrfs_mark_buffer_dirty(eb);
7064 btrfs_free_path(path);
7069 * called from commit_transaction. Writes all changed device stats to disk.
7071 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
7072 struct btrfs_fs_info *fs_info)
7074 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7075 struct btrfs_device *device;
7079 mutex_lock(&fs_devices->device_list_mutex);
7080 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7081 if (!device->dev_stats_valid || !btrfs_dev_stats_dirty(device))
7084 stats_cnt = atomic_read(&device->dev_stats_ccnt);
7085 ret = update_dev_stat_item(trans, fs_info, device);
7087 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7089 mutex_unlock(&fs_devices->device_list_mutex);
7094 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7096 btrfs_dev_stat_inc(dev, index);
7097 btrfs_dev_stat_print_on_error(dev);
7100 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
7102 if (!dev->dev_stats_valid)
7104 btrfs_err_rl_in_rcu(dev->fs_info,
7105 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7106 rcu_str_deref(dev->name),
7107 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7108 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7109 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7110 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7111 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7114 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7118 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7119 if (btrfs_dev_stat_read(dev, i) != 0)
7121 if (i == BTRFS_DEV_STAT_VALUES_MAX)
7122 return; /* all values == 0, suppress message */
7124 btrfs_info_in_rcu(dev->fs_info,
7125 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7126 rcu_str_deref(dev->name),
7127 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7128 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7129 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7130 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7131 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7134 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
7135 struct btrfs_ioctl_get_dev_stats *stats)
7137 struct btrfs_device *dev;
7138 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7141 mutex_lock(&fs_devices->device_list_mutex);
7142 dev = btrfs_find_device(fs_info, stats->devid, NULL, NULL);
7143 mutex_unlock(&fs_devices->device_list_mutex);
7146 btrfs_warn(fs_info, "get dev_stats failed, device not found");
7148 } else if (!dev->dev_stats_valid) {
7149 btrfs_warn(fs_info, "get dev_stats failed, not yet valid");
7151 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7152 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7153 if (stats->nr_items > i)
7155 btrfs_dev_stat_read_and_reset(dev, i);
7157 btrfs_dev_stat_reset(dev, i);
7160 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7161 if (stats->nr_items > i)
7162 stats->values[i] = btrfs_dev_stat_read(dev, i);
7164 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7165 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7169 void btrfs_scratch_superblocks(struct block_device *bdev, const char *device_path)
7171 struct buffer_head *bh;
7172 struct btrfs_super_block *disk_super;
7178 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX;
7181 if (btrfs_read_dev_one_super(bdev, copy_num, &bh))
7184 disk_super = (struct btrfs_super_block *)bh->b_data;
7186 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
7187 set_buffer_dirty(bh);
7188 sync_dirty_buffer(bh);
7192 /* Notify udev that device has changed */
7193 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
7195 /* Update ctime/mtime for device path for libblkid */
7196 update_dev_time(device_path);
7200 * Update the size of all devices, which is used for writing out the
7203 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
7205 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7206 struct btrfs_device *curr, *next;
7208 if (list_empty(&fs_devices->resized_devices))
7211 mutex_lock(&fs_devices->device_list_mutex);
7212 mutex_lock(&fs_info->chunk_mutex);
7213 list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
7215 list_del_init(&curr->resized_list);
7216 curr->commit_total_bytes = curr->disk_total_bytes;
7218 mutex_unlock(&fs_info->chunk_mutex);
7219 mutex_unlock(&fs_devices->device_list_mutex);
7222 /* Must be invoked during the transaction commit */
7223 void btrfs_update_commit_device_bytes_used(struct btrfs_fs_info *fs_info,
7224 struct btrfs_transaction *transaction)
7226 struct extent_map *em;
7227 struct map_lookup *map;
7228 struct btrfs_device *dev;
7231 if (list_empty(&transaction->pending_chunks))
7234 /* In order to kick the device replace finish process */
7235 mutex_lock(&fs_info->chunk_mutex);
7236 list_for_each_entry(em, &transaction->pending_chunks, list) {
7237 map = em->map_lookup;
7239 for (i = 0; i < map->num_stripes; i++) {
7240 dev = map->stripes[i].dev;
7241 dev->commit_bytes_used = dev->bytes_used;
7244 mutex_unlock(&fs_info->chunk_mutex);
7247 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
7249 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7250 while (fs_devices) {
7251 fs_devices->fs_info = fs_info;
7252 fs_devices = fs_devices->seed;
7256 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
7258 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7259 while (fs_devices) {
7260 fs_devices->fs_info = NULL;
7261 fs_devices = fs_devices->seed;