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);
192 bio_put(device->flush_bio);
198 static void btrfs_kobject_uevent(struct block_device *bdev,
199 enum kobject_action action)
203 ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
205 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
207 kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
208 &disk_to_dev(bdev->bd_disk)->kobj);
211 void btrfs_cleanup_fs_uuids(void)
213 struct btrfs_fs_devices *fs_devices;
215 while (!list_empty(&fs_uuids)) {
216 fs_devices = list_entry(fs_uuids.next,
217 struct btrfs_fs_devices, list);
218 list_del(&fs_devices->list);
219 free_fs_devices(fs_devices);
223 static struct btrfs_device *__alloc_device(void)
225 struct btrfs_device *dev;
227 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
229 return ERR_PTR(-ENOMEM);
232 * Preallocate a bio that's always going to be used for flushing device
233 * barriers and matches the device lifespan
235 dev->flush_bio = bio_alloc_bioset(GFP_KERNEL, 0, NULL);
236 if (!dev->flush_bio) {
238 return ERR_PTR(-ENOMEM);
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;
363 unsigned long last_waited = 0;
365 int sync_pending = 0;
366 struct blk_plug plug;
369 * this function runs all the bios we've collected for
370 * a particular device. We don't want to wander off to
371 * another device without first sending all of these down.
372 * So, setup a plug here and finish it off before we return
374 blk_start_plug(&plug);
376 bdi = device->bdev->bd_bdi;
379 spin_lock(&device->io_lock);
384 /* take all the bios off the list at once and process them
385 * later on (without the lock held). But, remember the
386 * tail and other pointers so the bios can be properly reinserted
387 * into the list if we hit congestion
389 if (!force_reg && device->pending_sync_bios.head) {
390 pending_bios = &device->pending_sync_bios;
393 pending_bios = &device->pending_bios;
397 pending = pending_bios->head;
398 tail = pending_bios->tail;
399 WARN_ON(pending && !tail);
402 * if pending was null this time around, no bios need processing
403 * at all and we can stop. Otherwise it'll loop back up again
404 * and do an additional check so no bios are missed.
406 * device->running_pending is used to synchronize with the
409 if (device->pending_sync_bios.head == NULL &&
410 device->pending_bios.head == NULL) {
412 device->running_pending = 0;
415 device->running_pending = 1;
418 pending_bios->head = NULL;
419 pending_bios->tail = NULL;
421 spin_unlock(&device->io_lock);
426 /* we want to work on both lists, but do more bios on the
427 * sync list than the regular list
430 pending_bios != &device->pending_sync_bios &&
431 device->pending_sync_bios.head) ||
432 (num_run > 64 && pending_bios == &device->pending_sync_bios &&
433 device->pending_bios.head)) {
434 spin_lock(&device->io_lock);
435 requeue_list(pending_bios, pending, tail);
440 pending = pending->bi_next;
443 BUG_ON(atomic_read(&cur->__bi_cnt) == 0);
446 * if we're doing the sync list, record that our
447 * plug has some sync requests on it
449 * If we're doing the regular list and there are
450 * sync requests sitting around, unplug before
453 if (pending_bios == &device->pending_sync_bios) {
455 } else if (sync_pending) {
456 blk_finish_plug(&plug);
457 blk_start_plug(&plug);
461 btrfsic_submit_bio(cur);
468 * we made progress, there is more work to do and the bdi
469 * is now congested. Back off and let other work structs
472 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
473 fs_info->fs_devices->open_devices > 1) {
474 struct io_context *ioc;
476 ioc = current->io_context;
479 * the main goal here is that we don't want to
480 * block if we're going to be able to submit
481 * more requests without blocking.
483 * This code does two great things, it pokes into
484 * the elevator code from a filesystem _and_
485 * it makes assumptions about how batching works.
487 if (ioc && ioc->nr_batch_requests > 0 &&
488 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
490 ioc->last_waited == last_waited)) {
492 * we want to go through our batch of
493 * requests and stop. So, we copy out
494 * the ioc->last_waited time and test
495 * against it before looping
497 last_waited = ioc->last_waited;
501 spin_lock(&device->io_lock);
502 requeue_list(pending_bios, pending, tail);
503 device->running_pending = 1;
505 spin_unlock(&device->io_lock);
506 btrfs_queue_work(fs_info->submit_workers,
516 spin_lock(&device->io_lock);
517 if (device->pending_bios.head || device->pending_sync_bios.head)
519 spin_unlock(&device->io_lock);
522 blk_finish_plug(&plug);
525 static void pending_bios_fn(struct btrfs_work *work)
527 struct btrfs_device *device;
529 device = container_of(work, struct btrfs_device, work);
530 run_scheduled_bios(device);
534 static void btrfs_free_stale_device(struct btrfs_device *cur_dev)
536 struct btrfs_fs_devices *fs_devs;
537 struct btrfs_device *dev;
542 list_for_each_entry(fs_devs, &fs_uuids, list) {
547 if (fs_devs->seeding)
550 list_for_each_entry(dev, &fs_devs->devices, dev_list) {
558 * Todo: This won't be enough. What if the same device
559 * comes back (with new uuid and) with its mapper path?
560 * But for now, this does help as mostly an admin will
561 * either use mapper or non mapper path throughout.
564 del = strcmp(rcu_str_deref(dev->name),
565 rcu_str_deref(cur_dev->name));
572 /* delete the stale device */
573 if (fs_devs->num_devices == 1) {
574 btrfs_sysfs_remove_fsid(fs_devs);
575 list_del(&fs_devs->list);
576 free_fs_devices(fs_devs);
578 fs_devs->num_devices--;
579 list_del(&dev->dev_list);
580 rcu_string_free(dev->name);
581 bio_put(dev->flush_bio);
590 * Add new device to list of registered devices
593 * 1 - first time device is seen
594 * 0 - device already known
597 static noinline int device_list_add(const char *path,
598 struct btrfs_super_block *disk_super,
599 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
601 struct btrfs_device *device;
602 struct btrfs_fs_devices *fs_devices;
603 struct rcu_string *name;
605 u64 found_transid = btrfs_super_generation(disk_super);
607 fs_devices = find_fsid(disk_super->fsid);
609 fs_devices = alloc_fs_devices(disk_super->fsid);
610 if (IS_ERR(fs_devices))
611 return PTR_ERR(fs_devices);
613 list_add(&fs_devices->list, &fs_uuids);
617 device = find_device(fs_devices, devid,
618 disk_super->dev_item.uuid);
622 if (fs_devices->opened)
625 device = btrfs_alloc_device(NULL, &devid,
626 disk_super->dev_item.uuid);
627 if (IS_ERR(device)) {
628 /* we can safely leave the fs_devices entry around */
629 return PTR_ERR(device);
632 name = rcu_string_strdup(path, GFP_NOFS);
634 bio_put(device->flush_bio);
638 rcu_assign_pointer(device->name, name);
640 mutex_lock(&fs_devices->device_list_mutex);
641 list_add_rcu(&device->dev_list, &fs_devices->devices);
642 fs_devices->num_devices++;
643 mutex_unlock(&fs_devices->device_list_mutex);
646 device->fs_devices = fs_devices;
647 } else if (!device->name || strcmp(device->name->str, path)) {
649 * When FS is already mounted.
650 * 1. If you are here and if the device->name is NULL that
651 * means this device was missing at time of FS mount.
652 * 2. If you are here and if the device->name is different
653 * from 'path' that means either
654 * a. The same device disappeared and reappeared with
656 * b. The missing-disk-which-was-replaced, has
659 * We must allow 1 and 2a above. But 2b would be a spurious
662 * Further in case of 1 and 2a above, the disk at 'path'
663 * would have missed some transaction when it was away and
664 * in case of 2a the stale bdev has to be updated as well.
665 * 2b must not be allowed at all time.
669 * For now, we do allow update to btrfs_fs_device through the
670 * btrfs dev scan cli after FS has been mounted. We're still
671 * tracking a problem where systems fail mount by subvolume id
672 * when we reject replacement on a mounted FS.
674 if (!fs_devices->opened && found_transid < device->generation) {
676 * That is if the FS is _not_ mounted and if you
677 * are here, that means there is more than one
678 * disk with same uuid and devid.We keep the one
679 * with larger generation number or the last-in if
680 * generation are equal.
685 name = rcu_string_strdup(path, GFP_NOFS);
688 rcu_string_free(device->name);
689 rcu_assign_pointer(device->name, name);
690 if (device->missing) {
691 fs_devices->missing_devices--;
697 * Unmount does not free the btrfs_device struct but would zero
698 * generation along with most of the other members. So just update
699 * it back. We need it to pick the disk with largest generation
702 if (!fs_devices->opened)
703 device->generation = found_transid;
706 * if there is new btrfs on an already registered device,
707 * then remove the stale device entry.
710 btrfs_free_stale_device(device);
712 *fs_devices_ret = fs_devices;
717 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
719 struct btrfs_fs_devices *fs_devices;
720 struct btrfs_device *device;
721 struct btrfs_device *orig_dev;
723 fs_devices = alloc_fs_devices(orig->fsid);
724 if (IS_ERR(fs_devices))
727 mutex_lock(&orig->device_list_mutex);
728 fs_devices->total_devices = orig->total_devices;
730 /* We have held the volume lock, it is safe to get the devices. */
731 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
732 struct rcu_string *name;
734 device = btrfs_alloc_device(NULL, &orig_dev->devid,
740 * This is ok to do without rcu read locked because we hold the
741 * uuid mutex so nothing we touch in here is going to disappear.
743 if (orig_dev->name) {
744 name = rcu_string_strdup(orig_dev->name->str,
747 bio_put(device->flush_bio);
751 rcu_assign_pointer(device->name, name);
754 list_add(&device->dev_list, &fs_devices->devices);
755 device->fs_devices = fs_devices;
756 fs_devices->num_devices++;
758 mutex_unlock(&orig->device_list_mutex);
761 mutex_unlock(&orig->device_list_mutex);
762 free_fs_devices(fs_devices);
763 return ERR_PTR(-ENOMEM);
766 void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step)
768 struct btrfs_device *device, *next;
769 struct btrfs_device *latest_dev = NULL;
771 mutex_lock(&uuid_mutex);
773 /* This is the initialized path, it is safe to release the devices. */
774 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
775 if (device->in_fs_metadata) {
776 if (!device->is_tgtdev_for_dev_replace &&
778 device->generation > latest_dev->generation)) {
784 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
786 * In the first step, keep the device which has
787 * the correct fsid and the devid that is used
788 * for the dev_replace procedure.
789 * In the second step, the dev_replace state is
790 * read from the device tree and it is known
791 * whether the procedure is really active or
792 * not, which means whether this device is
793 * used or whether it should be removed.
795 if (step == 0 || device->is_tgtdev_for_dev_replace) {
800 blkdev_put(device->bdev, device->mode);
802 fs_devices->open_devices--;
804 if (device->writeable) {
805 list_del_init(&device->dev_alloc_list);
806 device->writeable = 0;
807 if (!device->is_tgtdev_for_dev_replace)
808 fs_devices->rw_devices--;
810 list_del_init(&device->dev_list);
811 fs_devices->num_devices--;
812 rcu_string_free(device->name);
813 bio_put(device->flush_bio);
817 if (fs_devices->seed) {
818 fs_devices = fs_devices->seed;
822 fs_devices->latest_bdev = latest_dev->bdev;
824 mutex_unlock(&uuid_mutex);
827 static void free_device(struct rcu_head *head)
829 struct btrfs_device *device;
831 device = container_of(head, struct btrfs_device, rcu);
832 rcu_string_free(device->name);
833 bio_put(device->flush_bio);
837 static void btrfs_close_bdev(struct btrfs_device *device)
839 if (device->bdev && device->writeable) {
840 sync_blockdev(device->bdev);
841 invalidate_bdev(device->bdev);
845 blkdev_put(device->bdev, device->mode);
848 static void btrfs_prepare_close_one_device(struct btrfs_device *device)
850 struct btrfs_fs_devices *fs_devices = device->fs_devices;
851 struct btrfs_device *new_device;
852 struct rcu_string *name;
855 fs_devices->open_devices--;
857 if (device->writeable &&
858 device->devid != BTRFS_DEV_REPLACE_DEVID) {
859 list_del_init(&device->dev_alloc_list);
860 fs_devices->rw_devices--;
864 fs_devices->missing_devices--;
866 new_device = btrfs_alloc_device(NULL, &device->devid,
868 BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
870 /* Safe because we are under uuid_mutex */
872 name = rcu_string_strdup(device->name->str, GFP_NOFS);
873 BUG_ON(!name); /* -ENOMEM */
874 rcu_assign_pointer(new_device->name, name);
877 list_replace_rcu(&device->dev_list, &new_device->dev_list);
878 new_device->fs_devices = device->fs_devices;
881 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
883 struct btrfs_device *device, *tmp;
884 struct list_head pending_put;
886 INIT_LIST_HEAD(&pending_put);
888 if (--fs_devices->opened > 0)
891 mutex_lock(&fs_devices->device_list_mutex);
892 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
893 btrfs_prepare_close_one_device(device);
894 list_add(&device->dev_list, &pending_put);
896 mutex_unlock(&fs_devices->device_list_mutex);
899 * btrfs_show_devname() is using the device_list_mutex,
900 * sometimes call to blkdev_put() leads vfs calling
901 * into this func. So do put outside of device_list_mutex,
904 while (!list_empty(&pending_put)) {
905 device = list_first_entry(&pending_put,
906 struct btrfs_device, dev_list);
907 list_del(&device->dev_list);
908 btrfs_close_bdev(device);
909 call_rcu(&device->rcu, free_device);
912 WARN_ON(fs_devices->open_devices);
913 WARN_ON(fs_devices->rw_devices);
914 fs_devices->opened = 0;
915 fs_devices->seeding = 0;
920 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
922 struct btrfs_fs_devices *seed_devices = NULL;
925 mutex_lock(&uuid_mutex);
926 ret = __btrfs_close_devices(fs_devices);
927 if (!fs_devices->opened) {
928 seed_devices = fs_devices->seed;
929 fs_devices->seed = NULL;
931 mutex_unlock(&uuid_mutex);
933 while (seed_devices) {
934 fs_devices = seed_devices;
935 seed_devices = fs_devices->seed;
936 __btrfs_close_devices(fs_devices);
937 free_fs_devices(fs_devices);
942 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
943 fmode_t flags, void *holder)
945 struct request_queue *q;
946 struct block_device *bdev;
947 struct list_head *head = &fs_devices->devices;
948 struct btrfs_device *device;
949 struct btrfs_device *latest_dev = NULL;
950 struct buffer_head *bh;
951 struct btrfs_super_block *disk_super;
958 list_for_each_entry(device, head, dev_list) {
964 /* Just open everything we can; ignore failures here */
965 if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
969 disk_super = (struct btrfs_super_block *)bh->b_data;
970 devid = btrfs_stack_device_id(&disk_super->dev_item);
971 if (devid != device->devid)
974 if (memcmp(device->uuid, disk_super->dev_item.uuid,
978 device->generation = btrfs_super_generation(disk_super);
980 device->generation > latest_dev->generation)
983 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
984 device->writeable = 0;
986 device->writeable = !bdev_read_only(bdev);
990 q = bdev_get_queue(bdev);
991 if (blk_queue_discard(q))
992 device->can_discard = 1;
993 if (!blk_queue_nonrot(q))
994 fs_devices->rotating = 1;
997 device->in_fs_metadata = 0;
998 device->mode = flags;
1000 fs_devices->open_devices++;
1001 if (device->writeable &&
1002 device->devid != BTRFS_DEV_REPLACE_DEVID) {
1003 fs_devices->rw_devices++;
1004 list_add(&device->dev_alloc_list,
1005 &fs_devices->alloc_list);
1012 blkdev_put(bdev, flags);
1015 if (fs_devices->open_devices == 0) {
1019 fs_devices->seeding = seeding;
1020 fs_devices->opened = 1;
1021 fs_devices->latest_bdev = latest_dev->bdev;
1022 fs_devices->total_rw_bytes = 0;
1027 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1028 fmode_t flags, void *holder)
1032 mutex_lock(&uuid_mutex);
1033 if (fs_devices->opened) {
1034 fs_devices->opened++;
1037 ret = __btrfs_open_devices(fs_devices, flags, holder);
1039 mutex_unlock(&uuid_mutex);
1043 static void btrfs_release_disk_super(struct page *page)
1049 static int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr,
1051 struct btrfs_super_block **disk_super)
1056 /* make sure our super fits in the device */
1057 if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1060 /* make sure our super fits in the page */
1061 if (sizeof(**disk_super) > PAGE_SIZE)
1064 /* make sure our super doesn't straddle pages on disk */
1065 index = bytenr >> PAGE_SHIFT;
1066 if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index)
1069 /* pull in the page with our super */
1070 *page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
1073 if (IS_ERR_OR_NULL(*page))
1078 /* align our pointer to the offset of the super block */
1079 *disk_super = p + (bytenr & ~PAGE_MASK);
1081 if (btrfs_super_bytenr(*disk_super) != bytenr ||
1082 btrfs_super_magic(*disk_super) != BTRFS_MAGIC) {
1083 btrfs_release_disk_super(*page);
1087 if ((*disk_super)->label[0] &&
1088 (*disk_super)->label[BTRFS_LABEL_SIZE - 1])
1089 (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0';
1095 * Look for a btrfs signature on a device. This may be called out of the mount path
1096 * and we are not allowed to call set_blocksize during the scan. The superblock
1097 * is read via pagecache
1099 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
1100 struct btrfs_fs_devices **fs_devices_ret)
1102 struct btrfs_super_block *disk_super;
1103 struct block_device *bdev;
1112 * we would like to check all the supers, but that would make
1113 * a btrfs mount succeed after a mkfs from a different FS.
1114 * So, we need to add a special mount option to scan for
1115 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1117 bytenr = btrfs_sb_offset(0);
1118 flags |= FMODE_EXCL;
1119 mutex_lock(&uuid_mutex);
1121 bdev = blkdev_get_by_path(path, flags, holder);
1123 ret = PTR_ERR(bdev);
1127 if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super))
1128 goto error_bdev_put;
1130 devid = btrfs_stack_device_id(&disk_super->dev_item);
1131 transid = btrfs_super_generation(disk_super);
1132 total_devices = btrfs_super_num_devices(disk_super);
1134 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
1136 if (disk_super->label[0]) {
1137 pr_info("BTRFS: device label %s ", disk_super->label);
1139 pr_info("BTRFS: device fsid %pU ", disk_super->fsid);
1142 pr_cont("devid %llu transid %llu %s\n", devid, transid, path);
1145 if (!ret && fs_devices_ret)
1146 (*fs_devices_ret)->total_devices = total_devices;
1148 btrfs_release_disk_super(page);
1151 blkdev_put(bdev, flags);
1153 mutex_unlock(&uuid_mutex);
1157 /* helper to account the used device space in the range */
1158 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
1159 u64 end, u64 *length)
1161 struct btrfs_key key;
1162 struct btrfs_root *root = device->fs_info->dev_root;
1163 struct btrfs_dev_extent *dev_extent;
1164 struct btrfs_path *path;
1168 struct extent_buffer *l;
1172 if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
1175 path = btrfs_alloc_path();
1178 path->reada = READA_FORWARD;
1180 key.objectid = device->devid;
1182 key.type = BTRFS_DEV_EXTENT_KEY;
1184 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1188 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1195 slot = path->slots[0];
1196 if (slot >= btrfs_header_nritems(l)) {
1197 ret = btrfs_next_leaf(root, path);
1205 btrfs_item_key_to_cpu(l, &key, slot);
1207 if (key.objectid < device->devid)
1210 if (key.objectid > device->devid)
1213 if (key.type != BTRFS_DEV_EXTENT_KEY)
1216 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1217 extent_end = key.offset + btrfs_dev_extent_length(l,
1219 if (key.offset <= start && extent_end > end) {
1220 *length = end - start + 1;
1222 } else if (key.offset <= start && extent_end > start)
1223 *length += extent_end - start;
1224 else if (key.offset > start && extent_end <= end)
1225 *length += extent_end - key.offset;
1226 else if (key.offset > start && key.offset <= end) {
1227 *length += end - key.offset + 1;
1229 } else if (key.offset > end)
1237 btrfs_free_path(path);
1241 static int contains_pending_extent(struct btrfs_transaction *transaction,
1242 struct btrfs_device *device,
1243 u64 *start, u64 len)
1245 struct btrfs_fs_info *fs_info = device->fs_info;
1246 struct extent_map *em;
1247 struct list_head *search_list = &fs_info->pinned_chunks;
1249 u64 physical_start = *start;
1252 search_list = &transaction->pending_chunks;
1254 list_for_each_entry(em, search_list, list) {
1255 struct map_lookup *map;
1258 map = em->map_lookup;
1259 for (i = 0; i < map->num_stripes; i++) {
1262 if (map->stripes[i].dev != device)
1264 if (map->stripes[i].physical >= physical_start + len ||
1265 map->stripes[i].physical + em->orig_block_len <=
1269 * Make sure that while processing the pinned list we do
1270 * not override our *start with a lower value, because
1271 * we can have pinned chunks that fall within this
1272 * device hole and that have lower physical addresses
1273 * than the pending chunks we processed before. If we
1274 * do not take this special care we can end up getting
1275 * 2 pending chunks that start at the same physical
1276 * device offsets because the end offset of a pinned
1277 * chunk can be equal to the start offset of some
1280 end = map->stripes[i].physical + em->orig_block_len;
1287 if (search_list != &fs_info->pinned_chunks) {
1288 search_list = &fs_info->pinned_chunks;
1297 * find_free_dev_extent_start - find free space in the specified device
1298 * @device: the device which we search the free space in
1299 * @num_bytes: the size of the free space that we need
1300 * @search_start: the position from which to begin the search
1301 * @start: store the start of the free space.
1302 * @len: the size of the free space. that we find, or the size
1303 * of the max free space if we don't find suitable free space
1305 * this uses a pretty simple search, the expectation is that it is
1306 * called very infrequently and that a given device has a small number
1309 * @start is used to store the start of the free space if we find. But if we
1310 * don't find suitable free space, it will be used to store the start position
1311 * of the max free space.
1313 * @len is used to store the size of the free space that we find.
1314 * But if we don't find suitable free space, it is used to store the size of
1315 * the max free space.
1317 int find_free_dev_extent_start(struct btrfs_transaction *transaction,
1318 struct btrfs_device *device, u64 num_bytes,
1319 u64 search_start, u64 *start, u64 *len)
1321 struct btrfs_fs_info *fs_info = device->fs_info;
1322 struct btrfs_root *root = fs_info->dev_root;
1323 struct btrfs_key key;
1324 struct btrfs_dev_extent *dev_extent;
1325 struct btrfs_path *path;
1330 u64 search_end = device->total_bytes;
1333 struct extent_buffer *l;
1336 * We don't want to overwrite the superblock on the drive nor any area
1337 * used by the boot loader (grub for example), so we make sure to start
1338 * at an offset of at least 1MB.
1340 search_start = max_t(u64, search_start, SZ_1M);
1342 path = btrfs_alloc_path();
1346 max_hole_start = search_start;
1350 if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1355 path->reada = READA_FORWARD;
1356 path->search_commit_root = 1;
1357 path->skip_locking = 1;
1359 key.objectid = device->devid;
1360 key.offset = search_start;
1361 key.type = BTRFS_DEV_EXTENT_KEY;
1363 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1367 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1374 slot = path->slots[0];
1375 if (slot >= btrfs_header_nritems(l)) {
1376 ret = btrfs_next_leaf(root, path);
1384 btrfs_item_key_to_cpu(l, &key, slot);
1386 if (key.objectid < device->devid)
1389 if (key.objectid > device->devid)
1392 if (key.type != BTRFS_DEV_EXTENT_KEY)
1395 if (key.offset > search_start) {
1396 hole_size = key.offset - search_start;
1399 * Have to check before we set max_hole_start, otherwise
1400 * we could end up sending back this offset anyway.
1402 if (contains_pending_extent(transaction, device,
1405 if (key.offset >= search_start) {
1406 hole_size = key.offset - search_start;
1413 if (hole_size > max_hole_size) {
1414 max_hole_start = search_start;
1415 max_hole_size = hole_size;
1419 * If this free space is greater than which we need,
1420 * it must be the max free space that we have found
1421 * until now, so max_hole_start must point to the start
1422 * of this free space and the length of this free space
1423 * is stored in max_hole_size. Thus, we return
1424 * max_hole_start and max_hole_size and go back to the
1427 if (hole_size >= num_bytes) {
1433 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1434 extent_end = key.offset + btrfs_dev_extent_length(l,
1436 if (extent_end > search_start)
1437 search_start = extent_end;
1444 * At this point, search_start should be the end of
1445 * allocated dev extents, and when shrinking the device,
1446 * search_end may be smaller than search_start.
1448 if (search_end > search_start) {
1449 hole_size = search_end - search_start;
1451 if (contains_pending_extent(transaction, device, &search_start,
1453 btrfs_release_path(path);
1457 if (hole_size > max_hole_size) {
1458 max_hole_start = search_start;
1459 max_hole_size = hole_size;
1464 if (max_hole_size < num_bytes)
1470 btrfs_free_path(path);
1471 *start = max_hole_start;
1473 *len = max_hole_size;
1477 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1478 struct btrfs_device *device, u64 num_bytes,
1479 u64 *start, u64 *len)
1481 /* FIXME use last free of some kind */
1482 return find_free_dev_extent_start(trans->transaction, device,
1483 num_bytes, 0, start, len);
1486 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1487 struct btrfs_device *device,
1488 u64 start, u64 *dev_extent_len)
1490 struct btrfs_fs_info *fs_info = device->fs_info;
1491 struct btrfs_root *root = fs_info->dev_root;
1493 struct btrfs_path *path;
1494 struct btrfs_key key;
1495 struct btrfs_key found_key;
1496 struct extent_buffer *leaf = NULL;
1497 struct btrfs_dev_extent *extent = NULL;
1499 path = btrfs_alloc_path();
1503 key.objectid = device->devid;
1505 key.type = BTRFS_DEV_EXTENT_KEY;
1507 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1509 ret = btrfs_previous_item(root, path, key.objectid,
1510 BTRFS_DEV_EXTENT_KEY);
1513 leaf = path->nodes[0];
1514 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1515 extent = btrfs_item_ptr(leaf, path->slots[0],
1516 struct btrfs_dev_extent);
1517 BUG_ON(found_key.offset > start || found_key.offset +
1518 btrfs_dev_extent_length(leaf, extent) < start);
1520 btrfs_release_path(path);
1522 } else if (ret == 0) {
1523 leaf = path->nodes[0];
1524 extent = btrfs_item_ptr(leaf, path->slots[0],
1525 struct btrfs_dev_extent);
1527 btrfs_handle_fs_error(fs_info, ret, "Slot search failed");
1531 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1533 ret = btrfs_del_item(trans, root, path);
1535 btrfs_handle_fs_error(fs_info, ret,
1536 "Failed to remove dev extent item");
1538 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1541 btrfs_free_path(path);
1545 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1546 struct btrfs_device *device,
1547 u64 chunk_offset, u64 start, u64 num_bytes)
1550 struct btrfs_path *path;
1551 struct btrfs_fs_info *fs_info = device->fs_info;
1552 struct btrfs_root *root = fs_info->dev_root;
1553 struct btrfs_dev_extent *extent;
1554 struct extent_buffer *leaf;
1555 struct btrfs_key key;
1557 WARN_ON(!device->in_fs_metadata);
1558 WARN_ON(device->is_tgtdev_for_dev_replace);
1559 path = btrfs_alloc_path();
1563 key.objectid = device->devid;
1565 key.type = BTRFS_DEV_EXTENT_KEY;
1566 ret = btrfs_insert_empty_item(trans, root, path, &key,
1571 leaf = path->nodes[0];
1572 extent = btrfs_item_ptr(leaf, path->slots[0],
1573 struct btrfs_dev_extent);
1574 btrfs_set_dev_extent_chunk_tree(leaf, extent,
1575 BTRFS_CHUNK_TREE_OBJECTID);
1576 btrfs_set_dev_extent_chunk_objectid(leaf, extent,
1577 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
1578 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1580 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1581 btrfs_mark_buffer_dirty(leaf);
1583 btrfs_free_path(path);
1587 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1589 struct extent_map_tree *em_tree;
1590 struct extent_map *em;
1594 em_tree = &fs_info->mapping_tree.map_tree;
1595 read_lock(&em_tree->lock);
1596 n = rb_last(&em_tree->map);
1598 em = rb_entry(n, struct extent_map, rb_node);
1599 ret = em->start + em->len;
1601 read_unlock(&em_tree->lock);
1606 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1610 struct btrfs_key key;
1611 struct btrfs_key found_key;
1612 struct btrfs_path *path;
1614 path = btrfs_alloc_path();
1618 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1619 key.type = BTRFS_DEV_ITEM_KEY;
1620 key.offset = (u64)-1;
1622 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1626 BUG_ON(ret == 0); /* Corruption */
1628 ret = btrfs_previous_item(fs_info->chunk_root, path,
1629 BTRFS_DEV_ITEMS_OBJECTID,
1630 BTRFS_DEV_ITEM_KEY);
1634 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1636 *devid_ret = found_key.offset + 1;
1640 btrfs_free_path(path);
1645 * the device information is stored in the chunk root
1646 * the btrfs_device struct should be fully filled in
1648 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1649 struct btrfs_fs_info *fs_info,
1650 struct btrfs_device *device)
1652 struct btrfs_root *root = fs_info->chunk_root;
1654 struct btrfs_path *path;
1655 struct btrfs_dev_item *dev_item;
1656 struct extent_buffer *leaf;
1657 struct btrfs_key key;
1660 path = btrfs_alloc_path();
1664 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1665 key.type = BTRFS_DEV_ITEM_KEY;
1666 key.offset = device->devid;
1668 ret = btrfs_insert_empty_item(trans, root, path, &key,
1673 leaf = path->nodes[0];
1674 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1676 btrfs_set_device_id(leaf, dev_item, device->devid);
1677 btrfs_set_device_generation(leaf, dev_item, 0);
1678 btrfs_set_device_type(leaf, dev_item, device->type);
1679 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1680 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1681 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1682 btrfs_set_device_total_bytes(leaf, dev_item,
1683 btrfs_device_get_disk_total_bytes(device));
1684 btrfs_set_device_bytes_used(leaf, dev_item,
1685 btrfs_device_get_bytes_used(device));
1686 btrfs_set_device_group(leaf, dev_item, 0);
1687 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1688 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1689 btrfs_set_device_start_offset(leaf, dev_item, 0);
1691 ptr = btrfs_device_uuid(dev_item);
1692 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1693 ptr = btrfs_device_fsid(dev_item);
1694 write_extent_buffer(leaf, fs_info->fsid, ptr, BTRFS_FSID_SIZE);
1695 btrfs_mark_buffer_dirty(leaf);
1699 btrfs_free_path(path);
1704 * Function to update ctime/mtime for a given device path.
1705 * Mainly used for ctime/mtime based probe like libblkid.
1707 static void update_dev_time(const char *path_name)
1711 filp = filp_open(path_name, O_RDWR, 0);
1714 file_update_time(filp);
1715 filp_close(filp, NULL);
1718 static int btrfs_rm_dev_item(struct btrfs_fs_info *fs_info,
1719 struct btrfs_device *device)
1721 struct btrfs_root *root = fs_info->chunk_root;
1723 struct btrfs_path *path;
1724 struct btrfs_key key;
1725 struct btrfs_trans_handle *trans;
1727 path = btrfs_alloc_path();
1731 trans = btrfs_start_transaction(root, 0);
1732 if (IS_ERR(trans)) {
1733 btrfs_free_path(path);
1734 return PTR_ERR(trans);
1736 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1737 key.type = BTRFS_DEV_ITEM_KEY;
1738 key.offset = device->devid;
1740 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1744 btrfs_abort_transaction(trans, ret);
1745 btrfs_end_transaction(trans);
1749 ret = btrfs_del_item(trans, root, path);
1751 btrfs_abort_transaction(trans, ret);
1752 btrfs_end_transaction(trans);
1756 btrfs_free_path(path);
1758 ret = btrfs_commit_transaction(trans);
1763 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1764 * filesystem. It's up to the caller to adjust that number regarding eg. device
1767 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1775 seq = read_seqbegin(&fs_info->profiles_lock);
1777 all_avail = fs_info->avail_data_alloc_bits |
1778 fs_info->avail_system_alloc_bits |
1779 fs_info->avail_metadata_alloc_bits;
1780 } while (read_seqretry(&fs_info->profiles_lock, seq));
1782 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1783 if (!(all_avail & btrfs_raid_group[i]))
1786 if (num_devices < btrfs_raid_array[i].devs_min) {
1787 int ret = btrfs_raid_mindev_error[i];
1797 static struct btrfs_device * btrfs_find_next_active_device(
1798 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device)
1800 struct btrfs_device *next_device;
1802 list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1803 if (next_device != device &&
1804 !next_device->missing && next_device->bdev)
1812 * Helper function to check if the given device is part of s_bdev / latest_bdev
1813 * and replace it with the provided or the next active device, in the context
1814 * where this function called, there should be always be another device (or
1815 * this_dev) which is active.
1817 void btrfs_assign_next_active_device(struct btrfs_fs_info *fs_info,
1818 struct btrfs_device *device, struct btrfs_device *this_dev)
1820 struct btrfs_device *next_device;
1823 next_device = this_dev;
1825 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
1827 ASSERT(next_device);
1829 if (fs_info->sb->s_bdev &&
1830 (fs_info->sb->s_bdev == device->bdev))
1831 fs_info->sb->s_bdev = next_device->bdev;
1833 if (fs_info->fs_devices->latest_bdev == device->bdev)
1834 fs_info->fs_devices->latest_bdev = next_device->bdev;
1837 int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path,
1840 struct btrfs_device *device;
1841 struct btrfs_fs_devices *cur_devices;
1845 mutex_lock(&fs_info->volume_mutex);
1846 mutex_lock(&uuid_mutex);
1848 num_devices = fs_info->fs_devices->num_devices;
1849 btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
1850 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
1851 WARN_ON(num_devices < 1);
1854 btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
1856 ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
1860 ret = btrfs_find_device_by_devspec(fs_info, devid, device_path,
1865 if (device->is_tgtdev_for_dev_replace) {
1866 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1870 if (device->writeable && fs_info->fs_devices->rw_devices == 1) {
1871 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1875 if (device->writeable) {
1876 mutex_lock(&fs_info->chunk_mutex);
1877 list_del_init(&device->dev_alloc_list);
1878 device->fs_devices->rw_devices--;
1879 mutex_unlock(&fs_info->chunk_mutex);
1882 mutex_unlock(&uuid_mutex);
1883 ret = btrfs_shrink_device(device, 0);
1884 mutex_lock(&uuid_mutex);
1889 * TODO: the superblock still includes this device in its num_devices
1890 * counter although write_all_supers() is not locked out. This
1891 * could give a filesystem state which requires a degraded mount.
1893 ret = btrfs_rm_dev_item(fs_info, device);
1897 device->in_fs_metadata = 0;
1898 btrfs_scrub_cancel_dev(fs_info, device);
1901 * the device list mutex makes sure that we don't change
1902 * the device list while someone else is writing out all
1903 * the device supers. Whoever is writing all supers, should
1904 * lock the device list mutex before getting the number of
1905 * devices in the super block (super_copy). Conversely,
1906 * whoever updates the number of devices in the super block
1907 * (super_copy) should hold the device list mutex.
1910 cur_devices = device->fs_devices;
1911 mutex_lock(&fs_info->fs_devices->device_list_mutex);
1912 list_del_rcu(&device->dev_list);
1914 device->fs_devices->num_devices--;
1915 device->fs_devices->total_devices--;
1917 if (device->missing)
1918 device->fs_devices->missing_devices--;
1920 btrfs_assign_next_active_device(fs_info, device, NULL);
1923 device->fs_devices->open_devices--;
1924 /* remove sysfs entry */
1925 btrfs_sysfs_rm_device_link(fs_info->fs_devices, device);
1928 num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
1929 btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
1930 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1933 * at this point, the device is zero sized and detached from
1934 * the devices list. All that's left is to zero out the old
1935 * supers and free the device.
1937 if (device->writeable)
1938 btrfs_scratch_superblocks(device->bdev, device->name->str);
1940 btrfs_close_bdev(device);
1941 call_rcu(&device->rcu, free_device);
1943 if (cur_devices->open_devices == 0) {
1944 struct btrfs_fs_devices *fs_devices;
1945 fs_devices = fs_info->fs_devices;
1946 while (fs_devices) {
1947 if (fs_devices->seed == cur_devices) {
1948 fs_devices->seed = cur_devices->seed;
1951 fs_devices = fs_devices->seed;
1953 cur_devices->seed = NULL;
1954 __btrfs_close_devices(cur_devices);
1955 free_fs_devices(cur_devices);
1959 mutex_unlock(&uuid_mutex);
1960 mutex_unlock(&fs_info->volume_mutex);
1964 if (device->writeable) {
1965 mutex_lock(&fs_info->chunk_mutex);
1966 list_add(&device->dev_alloc_list,
1967 &fs_info->fs_devices->alloc_list);
1968 device->fs_devices->rw_devices++;
1969 mutex_unlock(&fs_info->chunk_mutex);
1974 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
1975 struct btrfs_device *srcdev)
1977 struct btrfs_fs_devices *fs_devices;
1979 WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1982 * in case of fs with no seed, srcdev->fs_devices will point
1983 * to fs_devices of fs_info. However when the dev being replaced is
1984 * a seed dev it will point to the seed's local fs_devices. In short
1985 * srcdev will have its correct fs_devices in both the cases.
1987 fs_devices = srcdev->fs_devices;
1989 list_del_rcu(&srcdev->dev_list);
1990 list_del(&srcdev->dev_alloc_list);
1991 fs_devices->num_devices--;
1992 if (srcdev->missing)
1993 fs_devices->missing_devices--;
1995 if (srcdev->writeable)
1996 fs_devices->rw_devices--;
1999 fs_devices->open_devices--;
2002 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
2003 struct btrfs_device *srcdev)
2005 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2007 if (srcdev->writeable) {
2008 /* zero out the old super if it is writable */
2009 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
2012 btrfs_close_bdev(srcdev);
2013 call_rcu(&srcdev->rcu, free_device);
2015 /* if this is no devs we rather delete the fs_devices */
2016 if (!fs_devices->num_devices) {
2017 struct btrfs_fs_devices *tmp_fs_devices;
2020 * On a mounted FS, num_devices can't be zero unless it's a
2021 * seed. In case of a seed device being replaced, the replace
2022 * target added to the sprout FS, so there will be no more
2023 * device left under the seed FS.
2025 ASSERT(fs_devices->seeding);
2027 tmp_fs_devices = fs_info->fs_devices;
2028 while (tmp_fs_devices) {
2029 if (tmp_fs_devices->seed == fs_devices) {
2030 tmp_fs_devices->seed = fs_devices->seed;
2033 tmp_fs_devices = tmp_fs_devices->seed;
2035 fs_devices->seed = NULL;
2036 __btrfs_close_devices(fs_devices);
2037 free_fs_devices(fs_devices);
2041 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2042 struct btrfs_device *tgtdev)
2044 mutex_lock(&uuid_mutex);
2046 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2048 btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev);
2051 fs_info->fs_devices->open_devices--;
2053 fs_info->fs_devices->num_devices--;
2055 btrfs_assign_next_active_device(fs_info, tgtdev, NULL);
2057 list_del_rcu(&tgtdev->dev_list);
2059 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2060 mutex_unlock(&uuid_mutex);
2063 * The update_dev_time() with in btrfs_scratch_superblocks()
2064 * may lead to a call to btrfs_show_devname() which will try
2065 * to hold device_list_mutex. And here this device
2066 * is already out of device list, so we don't have to hold
2067 * the device_list_mutex lock.
2069 btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
2071 btrfs_close_bdev(tgtdev);
2072 call_rcu(&tgtdev->rcu, free_device);
2075 static int btrfs_find_device_by_path(struct btrfs_fs_info *fs_info,
2076 const char *device_path,
2077 struct btrfs_device **device)
2080 struct btrfs_super_block *disk_super;
2083 struct block_device *bdev;
2084 struct buffer_head *bh;
2087 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2088 fs_info->bdev_holder, 0, &bdev, &bh);
2091 disk_super = (struct btrfs_super_block *)bh->b_data;
2092 devid = btrfs_stack_device_id(&disk_super->dev_item);
2093 dev_uuid = disk_super->dev_item.uuid;
2094 *device = btrfs_find_device(fs_info, devid, dev_uuid, disk_super->fsid);
2098 blkdev_put(bdev, FMODE_READ);
2102 int btrfs_find_device_missing_or_by_path(struct btrfs_fs_info *fs_info,
2103 const char *device_path,
2104 struct btrfs_device **device)
2107 if (strcmp(device_path, "missing") == 0) {
2108 struct list_head *devices;
2109 struct btrfs_device *tmp;
2111 devices = &fs_info->fs_devices->devices;
2113 * It is safe to read the devices since the volume_mutex
2114 * is held by the caller.
2116 list_for_each_entry(tmp, devices, dev_list) {
2117 if (tmp->in_fs_metadata && !tmp->bdev) {
2124 return BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2128 return btrfs_find_device_by_path(fs_info, device_path, device);
2133 * Lookup a device given by device id, or the path if the id is 0.
2135 int btrfs_find_device_by_devspec(struct btrfs_fs_info *fs_info, u64 devid,
2136 const char *devpath,
2137 struct btrfs_device **device)
2143 *device = btrfs_find_device(fs_info, devid, NULL, NULL);
2147 if (!devpath || !devpath[0])
2150 ret = btrfs_find_device_missing_or_by_path(fs_info, devpath,
2157 * does all the dirty work required for changing file system's UUID.
2159 static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info)
2161 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2162 struct btrfs_fs_devices *old_devices;
2163 struct btrfs_fs_devices *seed_devices;
2164 struct btrfs_super_block *disk_super = fs_info->super_copy;
2165 struct btrfs_device *device;
2168 BUG_ON(!mutex_is_locked(&uuid_mutex));
2169 if (!fs_devices->seeding)
2172 seed_devices = alloc_fs_devices(NULL);
2173 if (IS_ERR(seed_devices))
2174 return PTR_ERR(seed_devices);
2176 old_devices = clone_fs_devices(fs_devices);
2177 if (IS_ERR(old_devices)) {
2178 kfree(seed_devices);
2179 return PTR_ERR(old_devices);
2182 list_add(&old_devices->list, &fs_uuids);
2184 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2185 seed_devices->opened = 1;
2186 INIT_LIST_HEAD(&seed_devices->devices);
2187 INIT_LIST_HEAD(&seed_devices->alloc_list);
2188 mutex_init(&seed_devices->device_list_mutex);
2190 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2191 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2193 list_for_each_entry(device, &seed_devices->devices, dev_list)
2194 device->fs_devices = seed_devices;
2196 mutex_lock(&fs_info->chunk_mutex);
2197 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2198 mutex_unlock(&fs_info->chunk_mutex);
2200 fs_devices->seeding = 0;
2201 fs_devices->num_devices = 0;
2202 fs_devices->open_devices = 0;
2203 fs_devices->missing_devices = 0;
2204 fs_devices->rotating = 0;
2205 fs_devices->seed = seed_devices;
2207 generate_random_uuid(fs_devices->fsid);
2208 memcpy(fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2209 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2210 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2212 super_flags = btrfs_super_flags(disk_super) &
2213 ~BTRFS_SUPER_FLAG_SEEDING;
2214 btrfs_set_super_flags(disk_super, super_flags);
2220 * Store the expected generation for seed devices in device items.
2222 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2223 struct btrfs_fs_info *fs_info)
2225 struct btrfs_root *root = fs_info->chunk_root;
2226 struct btrfs_path *path;
2227 struct extent_buffer *leaf;
2228 struct btrfs_dev_item *dev_item;
2229 struct btrfs_device *device;
2230 struct btrfs_key key;
2231 u8 fs_uuid[BTRFS_FSID_SIZE];
2232 u8 dev_uuid[BTRFS_UUID_SIZE];
2236 path = btrfs_alloc_path();
2240 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2242 key.type = BTRFS_DEV_ITEM_KEY;
2245 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2249 leaf = path->nodes[0];
2251 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2252 ret = btrfs_next_leaf(root, path);
2257 leaf = path->nodes[0];
2258 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2259 btrfs_release_path(path);
2263 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2264 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2265 key.type != BTRFS_DEV_ITEM_KEY)
2268 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2269 struct btrfs_dev_item);
2270 devid = btrfs_device_id(leaf, dev_item);
2271 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2273 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2275 device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
2276 BUG_ON(!device); /* Logic error */
2278 if (device->fs_devices->seeding) {
2279 btrfs_set_device_generation(leaf, dev_item,
2280 device->generation);
2281 btrfs_mark_buffer_dirty(leaf);
2289 btrfs_free_path(path);
2293 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2295 struct btrfs_root *root = fs_info->dev_root;
2296 struct request_queue *q;
2297 struct btrfs_trans_handle *trans;
2298 struct btrfs_device *device;
2299 struct block_device *bdev;
2300 struct list_head *devices;
2301 struct super_block *sb = fs_info->sb;
2302 struct rcu_string *name;
2304 int seeding_dev = 0;
2306 bool unlocked = false;
2308 if (sb_rdonly(sb) && !fs_info->fs_devices->seeding)
2311 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2312 fs_info->bdev_holder);
2314 return PTR_ERR(bdev);
2316 if (fs_info->fs_devices->seeding) {
2318 down_write(&sb->s_umount);
2319 mutex_lock(&uuid_mutex);
2322 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2324 devices = &fs_info->fs_devices->devices;
2326 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2327 list_for_each_entry(device, devices, dev_list) {
2328 if (device->bdev == bdev) {
2331 &fs_info->fs_devices->device_list_mutex);
2335 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2337 device = btrfs_alloc_device(fs_info, NULL, NULL);
2338 if (IS_ERR(device)) {
2339 /* we can safely leave the fs_devices entry around */
2340 ret = PTR_ERR(device);
2344 name = rcu_string_strdup(device_path, GFP_KERNEL);
2346 bio_put(device->flush_bio);
2351 rcu_assign_pointer(device->name, name);
2353 trans = btrfs_start_transaction(root, 0);
2354 if (IS_ERR(trans)) {
2355 rcu_string_free(device->name);
2356 bio_put(device->flush_bio);
2358 ret = PTR_ERR(trans);
2362 q = bdev_get_queue(bdev);
2363 if (blk_queue_discard(q))
2364 device->can_discard = 1;
2365 device->writeable = 1;
2366 device->generation = trans->transid;
2367 device->io_width = fs_info->sectorsize;
2368 device->io_align = fs_info->sectorsize;
2369 device->sector_size = fs_info->sectorsize;
2370 device->total_bytes = round_down(i_size_read(bdev->bd_inode),
2371 fs_info->sectorsize);
2372 device->disk_total_bytes = device->total_bytes;
2373 device->commit_total_bytes = device->total_bytes;
2374 device->fs_info = fs_info;
2375 device->bdev = bdev;
2376 device->in_fs_metadata = 1;
2377 device->is_tgtdev_for_dev_replace = 0;
2378 device->mode = FMODE_EXCL;
2379 device->dev_stats_valid = 1;
2380 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2383 sb->s_flags &= ~SB_RDONLY;
2384 ret = btrfs_prepare_sprout(fs_info);
2386 btrfs_abort_transaction(trans, ret);
2391 device->fs_devices = fs_info->fs_devices;
2393 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2394 mutex_lock(&fs_info->chunk_mutex);
2395 list_add_rcu(&device->dev_list, &fs_info->fs_devices->devices);
2396 list_add(&device->dev_alloc_list,
2397 &fs_info->fs_devices->alloc_list);
2398 fs_info->fs_devices->num_devices++;
2399 fs_info->fs_devices->open_devices++;
2400 fs_info->fs_devices->rw_devices++;
2401 fs_info->fs_devices->total_devices++;
2402 fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2404 atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2406 if (!blk_queue_nonrot(q))
2407 fs_info->fs_devices->rotating = 1;
2409 tmp = btrfs_super_total_bytes(fs_info->super_copy);
2410 btrfs_set_super_total_bytes(fs_info->super_copy,
2411 round_down(tmp + device->total_bytes, fs_info->sectorsize));
2413 tmp = btrfs_super_num_devices(fs_info->super_copy);
2414 btrfs_set_super_num_devices(fs_info->super_copy, tmp + 1);
2416 /* add sysfs device entry */
2417 btrfs_sysfs_add_device_link(fs_info->fs_devices, device);
2420 * we've got more storage, clear any full flags on the space
2423 btrfs_clear_space_info_full(fs_info);
2425 mutex_unlock(&fs_info->chunk_mutex);
2426 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2429 mutex_lock(&fs_info->chunk_mutex);
2430 ret = init_first_rw_device(trans, fs_info);
2431 mutex_unlock(&fs_info->chunk_mutex);
2433 btrfs_abort_transaction(trans, ret);
2438 ret = btrfs_add_device(trans, fs_info, device);
2440 btrfs_abort_transaction(trans, ret);
2445 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2447 ret = btrfs_finish_sprout(trans, fs_info);
2449 btrfs_abort_transaction(trans, ret);
2453 /* Sprouting would change fsid of the mounted root,
2454 * so rename the fsid on the sysfs
2456 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2458 if (kobject_rename(&fs_info->fs_devices->fsid_kobj, fsid_buf))
2460 "sysfs: failed to create fsid for sprout");
2463 ret = btrfs_commit_transaction(trans);
2466 mutex_unlock(&uuid_mutex);
2467 up_write(&sb->s_umount);
2470 if (ret) /* transaction commit */
2473 ret = btrfs_relocate_sys_chunks(fs_info);
2475 btrfs_handle_fs_error(fs_info, ret,
2476 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2477 trans = btrfs_attach_transaction(root);
2478 if (IS_ERR(trans)) {
2479 if (PTR_ERR(trans) == -ENOENT)
2481 ret = PTR_ERR(trans);
2485 ret = btrfs_commit_transaction(trans);
2488 /* Update ctime/mtime for libblkid */
2489 update_dev_time(device_path);
2493 btrfs_sysfs_rm_device_link(fs_info->fs_devices, device);
2496 sb->s_flags |= SB_RDONLY;
2498 btrfs_end_transaction(trans);
2499 rcu_string_free(device->name);
2500 bio_put(device->flush_bio);
2503 blkdev_put(bdev, FMODE_EXCL);
2504 if (seeding_dev && !unlocked) {
2505 mutex_unlock(&uuid_mutex);
2506 up_write(&sb->s_umount);
2511 int btrfs_init_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2512 const char *device_path,
2513 struct btrfs_device *srcdev,
2514 struct btrfs_device **device_out)
2516 struct request_queue *q;
2517 struct btrfs_device *device;
2518 struct block_device *bdev;
2519 struct list_head *devices;
2520 struct rcu_string *name;
2521 u64 devid = BTRFS_DEV_REPLACE_DEVID;
2525 if (fs_info->fs_devices->seeding) {
2526 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
2530 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2531 fs_info->bdev_holder);
2533 btrfs_err(fs_info, "target device %s is invalid!", device_path);
2534 return PTR_ERR(bdev);
2537 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2539 devices = &fs_info->fs_devices->devices;
2540 list_for_each_entry(device, devices, dev_list) {
2541 if (device->bdev == bdev) {
2543 "target device is in the filesystem!");
2550 if (i_size_read(bdev->bd_inode) <
2551 btrfs_device_get_total_bytes(srcdev)) {
2553 "target device is smaller than source device!");
2559 device = btrfs_alloc_device(NULL, &devid, NULL);
2560 if (IS_ERR(device)) {
2561 ret = PTR_ERR(device);
2565 name = rcu_string_strdup(device_path, GFP_KERNEL);
2567 bio_put(device->flush_bio);
2572 rcu_assign_pointer(device->name, name);
2574 q = bdev_get_queue(bdev);
2575 if (blk_queue_discard(q))
2576 device->can_discard = 1;
2577 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2578 device->writeable = 1;
2579 device->generation = 0;
2580 device->io_width = fs_info->sectorsize;
2581 device->io_align = fs_info->sectorsize;
2582 device->sector_size = fs_info->sectorsize;
2583 device->total_bytes = btrfs_device_get_total_bytes(srcdev);
2584 device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
2585 device->bytes_used = btrfs_device_get_bytes_used(srcdev);
2586 ASSERT(list_empty(&srcdev->resized_list));
2587 device->commit_total_bytes = srcdev->commit_total_bytes;
2588 device->commit_bytes_used = device->bytes_used;
2589 device->fs_info = fs_info;
2590 device->bdev = bdev;
2591 device->in_fs_metadata = 1;
2592 device->is_tgtdev_for_dev_replace = 1;
2593 device->mode = FMODE_EXCL;
2594 device->dev_stats_valid = 1;
2595 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2596 device->fs_devices = fs_info->fs_devices;
2597 list_add(&device->dev_list, &fs_info->fs_devices->devices);
2598 fs_info->fs_devices->num_devices++;
2599 fs_info->fs_devices->open_devices++;
2600 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2602 *device_out = device;
2606 blkdev_put(bdev, FMODE_EXCL);
2610 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2611 struct btrfs_device *tgtdev)
2613 u32 sectorsize = fs_info->sectorsize;
2615 WARN_ON(fs_info->fs_devices->rw_devices == 0);
2616 tgtdev->io_width = sectorsize;
2617 tgtdev->io_align = sectorsize;
2618 tgtdev->sector_size = sectorsize;
2619 tgtdev->fs_info = fs_info;
2620 tgtdev->in_fs_metadata = 1;
2623 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2624 struct btrfs_device *device)
2627 struct btrfs_path *path;
2628 struct btrfs_root *root = device->fs_info->chunk_root;
2629 struct btrfs_dev_item *dev_item;
2630 struct extent_buffer *leaf;
2631 struct btrfs_key key;
2633 path = btrfs_alloc_path();
2637 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2638 key.type = BTRFS_DEV_ITEM_KEY;
2639 key.offset = device->devid;
2641 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2650 leaf = path->nodes[0];
2651 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2653 btrfs_set_device_id(leaf, dev_item, device->devid);
2654 btrfs_set_device_type(leaf, dev_item, device->type);
2655 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2656 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2657 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2658 btrfs_set_device_total_bytes(leaf, dev_item,
2659 btrfs_device_get_disk_total_bytes(device));
2660 btrfs_set_device_bytes_used(leaf, dev_item,
2661 btrfs_device_get_bytes_used(device));
2662 btrfs_mark_buffer_dirty(leaf);
2665 btrfs_free_path(path);
2669 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2670 struct btrfs_device *device, u64 new_size)
2672 struct btrfs_fs_info *fs_info = device->fs_info;
2673 struct btrfs_super_block *super_copy = fs_info->super_copy;
2674 struct btrfs_fs_devices *fs_devices;
2678 if (!device->writeable)
2681 new_size = round_down(new_size, fs_info->sectorsize);
2683 mutex_lock(&fs_info->chunk_mutex);
2684 old_total = btrfs_super_total_bytes(super_copy);
2685 diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2687 if (new_size <= device->total_bytes ||
2688 device->is_tgtdev_for_dev_replace) {
2689 mutex_unlock(&fs_info->chunk_mutex);
2693 fs_devices = fs_info->fs_devices;
2695 btrfs_set_super_total_bytes(super_copy,
2696 round_down(old_total + diff, fs_info->sectorsize));
2697 device->fs_devices->total_rw_bytes += diff;
2699 btrfs_device_set_total_bytes(device, new_size);
2700 btrfs_device_set_disk_total_bytes(device, new_size);
2701 btrfs_clear_space_info_full(device->fs_info);
2702 if (list_empty(&device->resized_list))
2703 list_add_tail(&device->resized_list,
2704 &fs_devices->resized_devices);
2705 mutex_unlock(&fs_info->chunk_mutex);
2707 return btrfs_update_device(trans, device);
2710 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2711 struct btrfs_fs_info *fs_info, u64 chunk_offset)
2713 struct btrfs_root *root = fs_info->chunk_root;
2715 struct btrfs_path *path;
2716 struct btrfs_key key;
2718 path = btrfs_alloc_path();
2722 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2723 key.offset = chunk_offset;
2724 key.type = BTRFS_CHUNK_ITEM_KEY;
2726 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2729 else if (ret > 0) { /* Logic error or corruption */
2730 btrfs_handle_fs_error(fs_info, -ENOENT,
2731 "Failed lookup while freeing chunk.");
2736 ret = btrfs_del_item(trans, root, path);
2738 btrfs_handle_fs_error(fs_info, ret,
2739 "Failed to delete chunk item.");
2741 btrfs_free_path(path);
2745 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2747 struct btrfs_super_block *super_copy = fs_info->super_copy;
2748 struct btrfs_disk_key *disk_key;
2749 struct btrfs_chunk *chunk;
2756 struct btrfs_key key;
2758 mutex_lock(&fs_info->chunk_mutex);
2759 array_size = btrfs_super_sys_array_size(super_copy);
2761 ptr = super_copy->sys_chunk_array;
2764 while (cur < array_size) {
2765 disk_key = (struct btrfs_disk_key *)ptr;
2766 btrfs_disk_key_to_cpu(&key, disk_key);
2768 len = sizeof(*disk_key);
2770 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2771 chunk = (struct btrfs_chunk *)(ptr + len);
2772 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2773 len += btrfs_chunk_item_size(num_stripes);
2778 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID &&
2779 key.offset == chunk_offset) {
2780 memmove(ptr, ptr + len, array_size - (cur + len));
2782 btrfs_set_super_sys_array_size(super_copy, array_size);
2788 mutex_unlock(&fs_info->chunk_mutex);
2792 static struct extent_map *get_chunk_map(struct btrfs_fs_info *fs_info,
2793 u64 logical, u64 length)
2795 struct extent_map_tree *em_tree;
2796 struct extent_map *em;
2798 em_tree = &fs_info->mapping_tree.map_tree;
2799 read_lock(&em_tree->lock);
2800 em = lookup_extent_mapping(em_tree, logical, length);
2801 read_unlock(&em_tree->lock);
2804 btrfs_crit(fs_info, "unable to find logical %llu length %llu",
2806 return ERR_PTR(-EINVAL);
2809 if (em->start > logical || em->start + em->len < logical) {
2811 "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
2812 logical, length, em->start, em->start + em->len);
2813 free_extent_map(em);
2814 return ERR_PTR(-EINVAL);
2817 /* callers are responsible for dropping em's ref. */
2821 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2822 struct btrfs_fs_info *fs_info, u64 chunk_offset)
2824 struct extent_map *em;
2825 struct map_lookup *map;
2826 u64 dev_extent_len = 0;
2828 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2830 em = get_chunk_map(fs_info, chunk_offset, 1);
2833 * This is a logic error, but we don't want to just rely on the
2834 * user having built with ASSERT enabled, so if ASSERT doesn't
2835 * do anything we still error out.
2840 map = em->map_lookup;
2841 mutex_lock(&fs_info->chunk_mutex);
2842 check_system_chunk(trans, fs_info, map->type);
2843 mutex_unlock(&fs_info->chunk_mutex);
2846 * Take the device list mutex to prevent races with the final phase of
2847 * a device replace operation that replaces the device object associated
2848 * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2850 mutex_lock(&fs_devices->device_list_mutex);
2851 for (i = 0; i < map->num_stripes; i++) {
2852 struct btrfs_device *device = map->stripes[i].dev;
2853 ret = btrfs_free_dev_extent(trans, device,
2854 map->stripes[i].physical,
2857 mutex_unlock(&fs_devices->device_list_mutex);
2858 btrfs_abort_transaction(trans, ret);
2862 if (device->bytes_used > 0) {
2863 mutex_lock(&fs_info->chunk_mutex);
2864 btrfs_device_set_bytes_used(device,
2865 device->bytes_used - dev_extent_len);
2866 atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
2867 btrfs_clear_space_info_full(fs_info);
2868 mutex_unlock(&fs_info->chunk_mutex);
2871 if (map->stripes[i].dev) {
2872 ret = btrfs_update_device(trans, map->stripes[i].dev);
2874 mutex_unlock(&fs_devices->device_list_mutex);
2875 btrfs_abort_transaction(trans, ret);
2880 mutex_unlock(&fs_devices->device_list_mutex);
2882 ret = btrfs_free_chunk(trans, fs_info, chunk_offset);
2884 btrfs_abort_transaction(trans, ret);
2888 trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
2890 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2891 ret = btrfs_del_sys_chunk(fs_info, chunk_offset);
2893 btrfs_abort_transaction(trans, ret);
2898 ret = btrfs_remove_block_group(trans, fs_info, chunk_offset, em);
2900 btrfs_abort_transaction(trans, ret);
2906 free_extent_map(em);
2910 static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2912 struct btrfs_root *root = fs_info->chunk_root;
2913 struct btrfs_trans_handle *trans;
2917 * Prevent races with automatic removal of unused block groups.
2918 * After we relocate and before we remove the chunk with offset
2919 * chunk_offset, automatic removal of the block group can kick in,
2920 * resulting in a failure when calling btrfs_remove_chunk() below.
2922 * Make sure to acquire this mutex before doing a tree search (dev
2923 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2924 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2925 * we release the path used to search the chunk/dev tree and before
2926 * the current task acquires this mutex and calls us.
2928 ASSERT(mutex_is_locked(&fs_info->delete_unused_bgs_mutex));
2930 ret = btrfs_can_relocate(fs_info, chunk_offset);
2934 /* step one, relocate all the extents inside this chunk */
2935 btrfs_scrub_pause(fs_info);
2936 ret = btrfs_relocate_block_group(fs_info, chunk_offset);
2937 btrfs_scrub_continue(fs_info);
2941 trans = btrfs_start_trans_remove_block_group(root->fs_info,
2943 if (IS_ERR(trans)) {
2944 ret = PTR_ERR(trans);
2945 btrfs_handle_fs_error(root->fs_info, ret, NULL);
2950 * step two, delete the device extents and the
2951 * chunk tree entries
2953 ret = btrfs_remove_chunk(trans, fs_info, chunk_offset);
2954 btrfs_end_transaction(trans);
2958 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
2960 struct btrfs_root *chunk_root = fs_info->chunk_root;
2961 struct btrfs_path *path;
2962 struct extent_buffer *leaf;
2963 struct btrfs_chunk *chunk;
2964 struct btrfs_key key;
2965 struct btrfs_key found_key;
2967 bool retried = false;
2971 path = btrfs_alloc_path();
2976 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2977 key.offset = (u64)-1;
2978 key.type = BTRFS_CHUNK_ITEM_KEY;
2981 mutex_lock(&fs_info->delete_unused_bgs_mutex);
2982 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2984 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
2987 BUG_ON(ret == 0); /* Corruption */
2989 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2992 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
2998 leaf = path->nodes[0];
2999 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3001 chunk = btrfs_item_ptr(leaf, path->slots[0],
3002 struct btrfs_chunk);
3003 chunk_type = btrfs_chunk_type(leaf, chunk);
3004 btrfs_release_path(path);
3006 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3007 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3013 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3015 if (found_key.offset == 0)
3017 key.offset = found_key.offset - 1;
3020 if (failed && !retried) {
3024 } else if (WARN_ON(failed && retried)) {
3028 btrfs_free_path(path);
3032 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3033 struct btrfs_balance_control *bctl)
3035 struct btrfs_root *root = fs_info->tree_root;
3036 struct btrfs_trans_handle *trans;
3037 struct btrfs_balance_item *item;
3038 struct btrfs_disk_balance_args disk_bargs;
3039 struct btrfs_path *path;
3040 struct extent_buffer *leaf;
3041 struct btrfs_key key;
3044 path = btrfs_alloc_path();
3048 trans = btrfs_start_transaction(root, 0);
3049 if (IS_ERR(trans)) {
3050 btrfs_free_path(path);
3051 return PTR_ERR(trans);
3054 key.objectid = BTRFS_BALANCE_OBJECTID;
3055 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3058 ret = btrfs_insert_empty_item(trans, root, path, &key,
3063 leaf = path->nodes[0];
3064 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3066 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3068 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3069 btrfs_set_balance_data(leaf, item, &disk_bargs);
3070 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3071 btrfs_set_balance_meta(leaf, item, &disk_bargs);
3072 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3073 btrfs_set_balance_sys(leaf, item, &disk_bargs);
3075 btrfs_set_balance_flags(leaf, item, bctl->flags);
3077 btrfs_mark_buffer_dirty(leaf);
3079 btrfs_free_path(path);
3080 err = btrfs_commit_transaction(trans);
3086 static int del_balance_item(struct btrfs_fs_info *fs_info)
3088 struct btrfs_root *root = fs_info->tree_root;
3089 struct btrfs_trans_handle *trans;
3090 struct btrfs_path *path;
3091 struct btrfs_key key;
3094 path = btrfs_alloc_path();
3098 trans = btrfs_start_transaction(root, 0);
3099 if (IS_ERR(trans)) {
3100 btrfs_free_path(path);
3101 return PTR_ERR(trans);
3104 key.objectid = BTRFS_BALANCE_OBJECTID;
3105 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3108 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3116 ret = btrfs_del_item(trans, root, path);
3118 btrfs_free_path(path);
3119 err = btrfs_commit_transaction(trans);
3126 * This is a heuristic used to reduce the number of chunks balanced on
3127 * resume after balance was interrupted.
3129 static void update_balance_args(struct btrfs_balance_control *bctl)
3132 * Turn on soft mode for chunk types that were being converted.
3134 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3135 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3136 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3137 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3138 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3139 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3142 * Turn on usage filter if is not already used. The idea is
3143 * that chunks that we have already balanced should be
3144 * reasonably full. Don't do it for chunks that are being
3145 * converted - that will keep us from relocating unconverted
3146 * (albeit full) chunks.
3148 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3149 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3150 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3151 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3152 bctl->data.usage = 90;
3154 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3155 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3156 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3157 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3158 bctl->sys.usage = 90;
3160 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3161 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3162 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3163 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3164 bctl->meta.usage = 90;
3169 * Should be called with both balance and volume mutexes held to
3170 * serialize other volume operations (add_dev/rm_dev/resize) with
3171 * restriper. Same goes for unset_balance_control.
3173 static void set_balance_control(struct btrfs_balance_control *bctl)
3175 struct btrfs_fs_info *fs_info = bctl->fs_info;
3177 BUG_ON(fs_info->balance_ctl);
3179 spin_lock(&fs_info->balance_lock);
3180 fs_info->balance_ctl = bctl;
3181 spin_unlock(&fs_info->balance_lock);
3184 static void unset_balance_control(struct btrfs_fs_info *fs_info)
3186 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3188 BUG_ON(!fs_info->balance_ctl);
3190 spin_lock(&fs_info->balance_lock);
3191 fs_info->balance_ctl = NULL;
3192 spin_unlock(&fs_info->balance_lock);
3198 * Balance filters. Return 1 if chunk should be filtered out
3199 * (should not be balanced).
3201 static int chunk_profiles_filter(u64 chunk_type,
3202 struct btrfs_balance_args *bargs)
3204 chunk_type = chunk_to_extended(chunk_type) &
3205 BTRFS_EXTENDED_PROFILE_MASK;
3207 if (bargs->profiles & chunk_type)
3213 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3214 struct btrfs_balance_args *bargs)
3216 struct btrfs_block_group_cache *cache;
3218 u64 user_thresh_min;
3219 u64 user_thresh_max;
3222 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3223 chunk_used = btrfs_block_group_used(&cache->item);
3225 if (bargs->usage_min == 0)
3226 user_thresh_min = 0;
3228 user_thresh_min = div_factor_fine(cache->key.offset,
3231 if (bargs->usage_max == 0)
3232 user_thresh_max = 1;
3233 else if (bargs->usage_max > 100)
3234 user_thresh_max = cache->key.offset;
3236 user_thresh_max = div_factor_fine(cache->key.offset,
3239 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3242 btrfs_put_block_group(cache);
3246 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3247 u64 chunk_offset, struct btrfs_balance_args *bargs)
3249 struct btrfs_block_group_cache *cache;
3250 u64 chunk_used, user_thresh;
3253 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3254 chunk_used = btrfs_block_group_used(&cache->item);
3256 if (bargs->usage_min == 0)
3258 else if (bargs->usage > 100)
3259 user_thresh = cache->key.offset;
3261 user_thresh = div_factor_fine(cache->key.offset,
3264 if (chunk_used < user_thresh)
3267 btrfs_put_block_group(cache);
3271 static int chunk_devid_filter(struct extent_buffer *leaf,
3272 struct btrfs_chunk *chunk,
3273 struct btrfs_balance_args *bargs)
3275 struct btrfs_stripe *stripe;
3276 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3279 for (i = 0; i < num_stripes; i++) {
3280 stripe = btrfs_stripe_nr(chunk, i);
3281 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3288 /* [pstart, pend) */
3289 static int chunk_drange_filter(struct extent_buffer *leaf,
3290 struct btrfs_chunk *chunk,
3291 struct btrfs_balance_args *bargs)
3293 struct btrfs_stripe *stripe;
3294 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3300 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3303 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3304 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3305 factor = num_stripes / 2;
3306 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3307 factor = num_stripes - 1;
3308 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3309 factor = num_stripes - 2;
3311 factor = num_stripes;
3314 for (i = 0; i < num_stripes; i++) {
3315 stripe = btrfs_stripe_nr(chunk, i);
3316 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3319 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3320 stripe_length = btrfs_chunk_length(leaf, chunk);
3321 stripe_length = div_u64(stripe_length, factor);
3323 if (stripe_offset < bargs->pend &&
3324 stripe_offset + stripe_length > bargs->pstart)
3331 /* [vstart, vend) */
3332 static int chunk_vrange_filter(struct extent_buffer *leaf,
3333 struct btrfs_chunk *chunk,
3335 struct btrfs_balance_args *bargs)
3337 if (chunk_offset < bargs->vend &&
3338 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3339 /* at least part of the chunk is inside this vrange */
3345 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3346 struct btrfs_chunk *chunk,
3347 struct btrfs_balance_args *bargs)
3349 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3351 if (bargs->stripes_min <= num_stripes
3352 && num_stripes <= bargs->stripes_max)
3358 static int chunk_soft_convert_filter(u64 chunk_type,
3359 struct btrfs_balance_args *bargs)
3361 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3364 chunk_type = chunk_to_extended(chunk_type) &
3365 BTRFS_EXTENDED_PROFILE_MASK;
3367 if (bargs->target == chunk_type)
3373 static int should_balance_chunk(struct btrfs_fs_info *fs_info,
3374 struct extent_buffer *leaf,
3375 struct btrfs_chunk *chunk, u64 chunk_offset)
3377 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3378 struct btrfs_balance_args *bargs = NULL;
3379 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3382 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3383 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3387 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3388 bargs = &bctl->data;
3389 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3391 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3392 bargs = &bctl->meta;
3394 /* profiles filter */
3395 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3396 chunk_profiles_filter(chunk_type, bargs)) {
3401 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3402 chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3404 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3405 chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3410 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3411 chunk_devid_filter(leaf, chunk, bargs)) {
3415 /* drange filter, makes sense only with devid filter */
3416 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3417 chunk_drange_filter(leaf, chunk, bargs)) {
3422 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3423 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3427 /* stripes filter */
3428 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3429 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3433 /* soft profile changing mode */
3434 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3435 chunk_soft_convert_filter(chunk_type, bargs)) {
3440 * limited by count, must be the last filter
3442 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3443 if (bargs->limit == 0)
3447 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3449 * Same logic as the 'limit' filter; the minimum cannot be
3450 * determined here because we do not have the global information
3451 * about the count of all chunks that satisfy the filters.
3453 if (bargs->limit_max == 0)
3462 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3464 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3465 struct btrfs_root *chunk_root = fs_info->chunk_root;
3466 struct btrfs_root *dev_root = fs_info->dev_root;
3467 struct list_head *devices;
3468 struct btrfs_device *device;
3472 struct btrfs_chunk *chunk;
3473 struct btrfs_path *path = NULL;
3474 struct btrfs_key key;
3475 struct btrfs_key found_key;
3476 struct btrfs_trans_handle *trans;
3477 struct extent_buffer *leaf;
3480 int enospc_errors = 0;
3481 bool counting = true;
3482 /* The single value limit and min/max limits use the same bytes in the */
3483 u64 limit_data = bctl->data.limit;
3484 u64 limit_meta = bctl->meta.limit;
3485 u64 limit_sys = bctl->sys.limit;
3489 int chunk_reserved = 0;
3492 /* step one make some room on all the devices */
3493 devices = &fs_info->fs_devices->devices;
3494 list_for_each_entry(device, devices, dev_list) {
3495 old_size = btrfs_device_get_total_bytes(device);
3496 size_to_free = div_factor(old_size, 1);
3497 size_to_free = min_t(u64, size_to_free, SZ_1M);
3498 if (!device->writeable ||
3499 btrfs_device_get_total_bytes(device) -
3500 btrfs_device_get_bytes_used(device) > size_to_free ||
3501 device->is_tgtdev_for_dev_replace)
3504 ret = btrfs_shrink_device(device, old_size - size_to_free);
3508 /* btrfs_shrink_device never returns ret > 0 */
3513 trans = btrfs_start_transaction(dev_root, 0);
3514 if (IS_ERR(trans)) {
3515 ret = PTR_ERR(trans);
3516 btrfs_info_in_rcu(fs_info,
3517 "resize: unable to start transaction after shrinking device %s (error %d), old size %llu, new size %llu",
3518 rcu_str_deref(device->name), ret,
3519 old_size, old_size - size_to_free);
3523 ret = btrfs_grow_device(trans, device, old_size);
3525 btrfs_end_transaction(trans);
3526 /* btrfs_grow_device never returns ret > 0 */
3528 btrfs_info_in_rcu(fs_info,
3529 "resize: unable to grow device after shrinking device %s (error %d), old size %llu, new size %llu",
3530 rcu_str_deref(device->name), ret,
3531 old_size, old_size - size_to_free);
3535 btrfs_end_transaction(trans);
3538 /* step two, relocate all the chunks */
3539 path = btrfs_alloc_path();
3545 /* zero out stat counters */
3546 spin_lock(&fs_info->balance_lock);
3547 memset(&bctl->stat, 0, sizeof(bctl->stat));
3548 spin_unlock(&fs_info->balance_lock);
3552 * The single value limit and min/max limits use the same bytes
3555 bctl->data.limit = limit_data;
3556 bctl->meta.limit = limit_meta;
3557 bctl->sys.limit = limit_sys;
3559 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3560 key.offset = (u64)-1;
3561 key.type = BTRFS_CHUNK_ITEM_KEY;
3564 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3565 atomic_read(&fs_info->balance_cancel_req)) {
3570 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3571 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3573 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3578 * this shouldn't happen, it means the last relocate
3582 BUG(); /* FIXME break ? */
3584 ret = btrfs_previous_item(chunk_root, path, 0,
3585 BTRFS_CHUNK_ITEM_KEY);
3587 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3592 leaf = path->nodes[0];
3593 slot = path->slots[0];
3594 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3596 if (found_key.objectid != key.objectid) {
3597 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3601 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3602 chunk_type = btrfs_chunk_type(leaf, chunk);
3605 spin_lock(&fs_info->balance_lock);
3606 bctl->stat.considered++;
3607 spin_unlock(&fs_info->balance_lock);
3610 ret = should_balance_chunk(fs_info, leaf, chunk,
3613 btrfs_release_path(path);
3615 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3620 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3621 spin_lock(&fs_info->balance_lock);
3622 bctl->stat.expected++;
3623 spin_unlock(&fs_info->balance_lock);
3625 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3627 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3629 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3636 * Apply limit_min filter, no need to check if the LIMITS
3637 * filter is used, limit_min is 0 by default
3639 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3640 count_data < bctl->data.limit_min)
3641 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3642 count_meta < bctl->meta.limit_min)
3643 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3644 count_sys < bctl->sys.limit_min)) {
3645 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3649 ASSERT(fs_info->data_sinfo);
3650 spin_lock(&fs_info->data_sinfo->lock);
3651 bytes_used = fs_info->data_sinfo->bytes_used;
3652 spin_unlock(&fs_info->data_sinfo->lock);
3654 if ((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3655 !chunk_reserved && !bytes_used) {
3656 trans = btrfs_start_transaction(chunk_root, 0);
3657 if (IS_ERR(trans)) {
3658 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3659 ret = PTR_ERR(trans);
3663 ret = btrfs_force_chunk_alloc(trans, fs_info,
3664 BTRFS_BLOCK_GROUP_DATA);
3665 btrfs_end_transaction(trans);
3667 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3673 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3674 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3675 if (ret && ret != -ENOSPC)
3677 if (ret == -ENOSPC) {
3680 spin_lock(&fs_info->balance_lock);
3681 bctl->stat.completed++;
3682 spin_unlock(&fs_info->balance_lock);
3685 if (found_key.offset == 0)
3687 key.offset = found_key.offset - 1;
3691 btrfs_release_path(path);
3696 btrfs_free_path(path);
3697 if (enospc_errors) {
3698 btrfs_info(fs_info, "%d enospc errors during balance",
3708 * alloc_profile_is_valid - see if a given profile is valid and reduced
3709 * @flags: profile to validate
3710 * @extended: if true @flags is treated as an extended profile
3712 static int alloc_profile_is_valid(u64 flags, int extended)
3714 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3715 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3717 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3719 /* 1) check that all other bits are zeroed */
3723 /* 2) see if profile is reduced */
3725 return !extended; /* "0" is valid for usual profiles */
3727 /* true if exactly one bit set */
3728 return (flags & (flags - 1)) == 0;
3731 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3733 /* cancel requested || normal exit path */
3734 return atomic_read(&fs_info->balance_cancel_req) ||
3735 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3736 atomic_read(&fs_info->balance_cancel_req) == 0);
3739 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3743 unset_balance_control(fs_info);
3744 ret = del_balance_item(fs_info);
3746 btrfs_handle_fs_error(fs_info, ret, NULL);
3748 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3751 /* Non-zero return value signifies invalidity */
3752 static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg,
3755 return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3756 (!alloc_profile_is_valid(bctl_arg->target, 1) ||
3757 (bctl_arg->target & ~allowed)));
3761 * Should be called with both balance and volume mutexes held
3763 int btrfs_balance(struct btrfs_balance_control *bctl,
3764 struct btrfs_ioctl_balance_args *bargs)
3766 struct btrfs_fs_info *fs_info = bctl->fs_info;
3767 u64 meta_target, data_target;
3774 if (btrfs_fs_closing(fs_info) ||
3775 atomic_read(&fs_info->balance_pause_req) ||
3776 atomic_read(&fs_info->balance_cancel_req)) {
3781 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3782 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3786 * In case of mixed groups both data and meta should be picked,
3787 * and identical options should be given for both of them.
3789 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3790 if (mixed && (bctl->flags & allowed)) {
3791 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3792 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3793 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3795 "with mixed groups data and metadata balance options must be the same");
3801 num_devices = fs_info->fs_devices->num_devices;
3802 btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
3803 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3804 BUG_ON(num_devices < 1);
3807 btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
3808 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE | BTRFS_BLOCK_GROUP_DUP;
3809 if (num_devices > 1)
3810 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3811 if (num_devices > 2)
3812 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3813 if (num_devices > 3)
3814 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3815 BTRFS_BLOCK_GROUP_RAID6);
3816 if (validate_convert_profile(&bctl->data, allowed)) {
3818 "unable to start balance with target data profile %llu",
3823 if (validate_convert_profile(&bctl->meta, allowed)) {
3825 "unable to start balance with target metadata profile %llu",
3830 if (validate_convert_profile(&bctl->sys, allowed)) {
3832 "unable to start balance with target system profile %llu",
3838 /* allow to reduce meta or sys integrity only if force set */
3839 allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3840 BTRFS_BLOCK_GROUP_RAID10 |
3841 BTRFS_BLOCK_GROUP_RAID5 |
3842 BTRFS_BLOCK_GROUP_RAID6;
3844 seq = read_seqbegin(&fs_info->profiles_lock);
3846 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3847 (fs_info->avail_system_alloc_bits & allowed) &&
3848 !(bctl->sys.target & allowed)) ||
3849 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3850 (fs_info->avail_metadata_alloc_bits & allowed) &&
3851 !(bctl->meta.target & allowed))) {
3852 if (bctl->flags & BTRFS_BALANCE_FORCE) {
3854 "force reducing metadata integrity");
3857 "balance will reduce metadata integrity, use force if you want this");
3862 } while (read_seqretry(&fs_info->profiles_lock, seq));
3864 /* if we're not converting, the target field is uninitialized */
3865 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
3866 bctl->meta.target : fs_info->avail_metadata_alloc_bits;
3867 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
3868 bctl->data.target : fs_info->avail_data_alloc_bits;
3869 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <
3870 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {
3872 "metadata profile 0x%llx has lower redundancy than data profile 0x%llx",
3873 meta_target, data_target);
3876 ret = insert_balance_item(fs_info, bctl);
3877 if (ret && ret != -EEXIST)
3880 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3881 BUG_ON(ret == -EEXIST);
3882 set_balance_control(bctl);
3884 BUG_ON(ret != -EEXIST);
3885 spin_lock(&fs_info->balance_lock);
3886 update_balance_args(bctl);
3887 spin_unlock(&fs_info->balance_lock);
3890 atomic_inc(&fs_info->balance_running);
3891 mutex_unlock(&fs_info->balance_mutex);
3893 ret = __btrfs_balance(fs_info);
3895 mutex_lock(&fs_info->balance_mutex);
3896 atomic_dec(&fs_info->balance_running);
3899 memset(bargs, 0, sizeof(*bargs));
3900 update_ioctl_balance_args(fs_info, 0, bargs);
3903 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3904 balance_need_close(fs_info)) {
3905 __cancel_balance(fs_info);
3908 wake_up(&fs_info->balance_wait_q);
3912 if (bctl->flags & BTRFS_BALANCE_RESUME)
3913 __cancel_balance(fs_info);
3916 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3921 static int balance_kthread(void *data)
3923 struct btrfs_fs_info *fs_info = data;
3926 mutex_lock(&fs_info->volume_mutex);
3927 mutex_lock(&fs_info->balance_mutex);
3929 if (fs_info->balance_ctl) {
3930 btrfs_info(fs_info, "continuing balance");
3931 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3934 mutex_unlock(&fs_info->balance_mutex);
3935 mutex_unlock(&fs_info->volume_mutex);
3940 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3942 struct task_struct *tsk;
3944 spin_lock(&fs_info->balance_lock);
3945 if (!fs_info->balance_ctl) {
3946 spin_unlock(&fs_info->balance_lock);
3949 spin_unlock(&fs_info->balance_lock);
3951 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
3952 btrfs_info(fs_info, "force skipping balance");
3956 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3957 return PTR_ERR_OR_ZERO(tsk);
3960 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3962 struct btrfs_balance_control *bctl;
3963 struct btrfs_balance_item *item;
3964 struct btrfs_disk_balance_args disk_bargs;
3965 struct btrfs_path *path;
3966 struct extent_buffer *leaf;
3967 struct btrfs_key key;
3970 path = btrfs_alloc_path();
3974 key.objectid = BTRFS_BALANCE_OBJECTID;
3975 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3978 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3981 if (ret > 0) { /* ret = -ENOENT; */
3986 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3992 leaf = path->nodes[0];
3993 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3995 bctl->fs_info = fs_info;
3996 bctl->flags = btrfs_balance_flags(leaf, item);
3997 bctl->flags |= BTRFS_BALANCE_RESUME;
3999 btrfs_balance_data(leaf, item, &disk_bargs);
4000 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4001 btrfs_balance_meta(leaf, item, &disk_bargs);
4002 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4003 btrfs_balance_sys(leaf, item, &disk_bargs);
4004 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4006 WARN_ON(test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags));
4008 mutex_lock(&fs_info->volume_mutex);
4009 mutex_lock(&fs_info->balance_mutex);
4011 set_balance_control(bctl);
4013 mutex_unlock(&fs_info->balance_mutex);
4014 mutex_unlock(&fs_info->volume_mutex);
4016 btrfs_free_path(path);
4020 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4024 mutex_lock(&fs_info->balance_mutex);
4025 if (!fs_info->balance_ctl) {
4026 mutex_unlock(&fs_info->balance_mutex);
4030 if (atomic_read(&fs_info->balance_running)) {
4031 atomic_inc(&fs_info->balance_pause_req);
4032 mutex_unlock(&fs_info->balance_mutex);
4034 wait_event(fs_info->balance_wait_q,
4035 atomic_read(&fs_info->balance_running) == 0);
4037 mutex_lock(&fs_info->balance_mutex);
4038 /* we are good with balance_ctl ripped off from under us */
4039 BUG_ON(atomic_read(&fs_info->balance_running));
4040 atomic_dec(&fs_info->balance_pause_req);
4045 mutex_unlock(&fs_info->balance_mutex);
4049 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4051 if (sb_rdonly(fs_info->sb))
4054 mutex_lock(&fs_info->balance_mutex);
4055 if (!fs_info->balance_ctl) {
4056 mutex_unlock(&fs_info->balance_mutex);
4060 atomic_inc(&fs_info->balance_cancel_req);
4062 * if we are running just wait and return, balance item is
4063 * deleted in btrfs_balance in this case
4065 if (atomic_read(&fs_info->balance_running)) {
4066 mutex_unlock(&fs_info->balance_mutex);
4067 wait_event(fs_info->balance_wait_q,
4068 atomic_read(&fs_info->balance_running) == 0);
4069 mutex_lock(&fs_info->balance_mutex);
4071 /* __cancel_balance needs volume_mutex */
4072 mutex_unlock(&fs_info->balance_mutex);
4073 mutex_lock(&fs_info->volume_mutex);
4074 mutex_lock(&fs_info->balance_mutex);
4076 if (fs_info->balance_ctl)
4077 __cancel_balance(fs_info);
4079 mutex_unlock(&fs_info->volume_mutex);
4082 BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
4083 atomic_dec(&fs_info->balance_cancel_req);
4084 mutex_unlock(&fs_info->balance_mutex);
4088 static int btrfs_uuid_scan_kthread(void *data)
4090 struct btrfs_fs_info *fs_info = data;
4091 struct btrfs_root *root = fs_info->tree_root;
4092 struct btrfs_key key;
4093 struct btrfs_path *path = NULL;
4095 struct extent_buffer *eb;
4097 struct btrfs_root_item root_item;
4099 struct btrfs_trans_handle *trans = NULL;
4101 path = btrfs_alloc_path();
4108 key.type = BTRFS_ROOT_ITEM_KEY;
4112 ret = btrfs_search_forward(root, &key, path, 0);
4119 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4120 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4121 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4122 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4125 eb = path->nodes[0];
4126 slot = path->slots[0];
4127 item_size = btrfs_item_size_nr(eb, slot);
4128 if (item_size < sizeof(root_item))
4131 read_extent_buffer(eb, &root_item,
4132 btrfs_item_ptr_offset(eb, slot),
4133 (int)sizeof(root_item));
4134 if (btrfs_root_refs(&root_item) == 0)
4137 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4138 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4142 btrfs_release_path(path);
4144 * 1 - subvol uuid item
4145 * 1 - received_subvol uuid item
4147 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4148 if (IS_ERR(trans)) {
4149 ret = PTR_ERR(trans);
4157 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4158 ret = btrfs_uuid_tree_add(trans, fs_info,
4160 BTRFS_UUID_KEY_SUBVOL,
4163 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4169 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4170 ret = btrfs_uuid_tree_add(trans, fs_info,
4171 root_item.received_uuid,
4172 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4175 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4183 ret = btrfs_end_transaction(trans);
4189 btrfs_release_path(path);
4190 if (key.offset < (u64)-1) {
4192 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4194 key.type = BTRFS_ROOT_ITEM_KEY;
4195 } else if (key.objectid < (u64)-1) {
4197 key.type = BTRFS_ROOT_ITEM_KEY;
4206 btrfs_free_path(path);
4207 if (trans && !IS_ERR(trans))
4208 btrfs_end_transaction(trans);
4210 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4212 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4213 up(&fs_info->uuid_tree_rescan_sem);
4218 * Callback for btrfs_uuid_tree_iterate().
4220 * 0 check succeeded, the entry is not outdated.
4221 * < 0 if an error occurred.
4222 * > 0 if the check failed, which means the caller shall remove the entry.
4224 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
4225 u8 *uuid, u8 type, u64 subid)
4227 struct btrfs_key key;
4229 struct btrfs_root *subvol_root;
4231 if (type != BTRFS_UUID_KEY_SUBVOL &&
4232 type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
4235 key.objectid = subid;
4236 key.type = BTRFS_ROOT_ITEM_KEY;
4237 key.offset = (u64)-1;
4238 subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
4239 if (IS_ERR(subvol_root)) {
4240 ret = PTR_ERR(subvol_root);
4247 case BTRFS_UUID_KEY_SUBVOL:
4248 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
4251 case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
4252 if (memcmp(uuid, subvol_root->root_item.received_uuid,
4262 static int btrfs_uuid_rescan_kthread(void *data)
4264 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4268 * 1st step is to iterate through the existing UUID tree and
4269 * to delete all entries that contain outdated data.
4270 * 2nd step is to add all missing entries to the UUID tree.
4272 ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4274 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4275 up(&fs_info->uuid_tree_rescan_sem);
4278 return btrfs_uuid_scan_kthread(data);
4281 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4283 struct btrfs_trans_handle *trans;
4284 struct btrfs_root *tree_root = fs_info->tree_root;
4285 struct btrfs_root *uuid_root;
4286 struct task_struct *task;
4293 trans = btrfs_start_transaction(tree_root, 2);
4295 return PTR_ERR(trans);
4297 uuid_root = btrfs_create_tree(trans, fs_info,
4298 BTRFS_UUID_TREE_OBJECTID);
4299 if (IS_ERR(uuid_root)) {
4300 ret = PTR_ERR(uuid_root);
4301 btrfs_abort_transaction(trans, ret);
4302 btrfs_end_transaction(trans);
4306 fs_info->uuid_root = uuid_root;
4308 ret = btrfs_commit_transaction(trans);
4312 down(&fs_info->uuid_tree_rescan_sem);
4313 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4315 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4316 btrfs_warn(fs_info, "failed to start uuid_scan task");
4317 up(&fs_info->uuid_tree_rescan_sem);
4318 return PTR_ERR(task);
4324 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4326 struct task_struct *task;
4328 down(&fs_info->uuid_tree_rescan_sem);
4329 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4331 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4332 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4333 up(&fs_info->uuid_tree_rescan_sem);
4334 return PTR_ERR(task);
4341 * shrinking a device means finding all of the device extents past
4342 * the new size, and then following the back refs to the chunks.
4343 * The chunk relocation code actually frees the device extent
4345 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4347 struct btrfs_fs_info *fs_info = device->fs_info;
4348 struct btrfs_root *root = fs_info->dev_root;
4349 struct btrfs_trans_handle *trans;
4350 struct btrfs_dev_extent *dev_extent = NULL;
4351 struct btrfs_path *path;
4357 bool retried = false;
4358 bool checked_pending_chunks = false;
4359 struct extent_buffer *l;
4360 struct btrfs_key key;
4361 struct btrfs_super_block *super_copy = fs_info->super_copy;
4362 u64 old_total = btrfs_super_total_bytes(super_copy);
4363 u64 old_size = btrfs_device_get_total_bytes(device);
4366 new_size = round_down(new_size, fs_info->sectorsize);
4367 diff = round_down(old_size - new_size, fs_info->sectorsize);
4369 if (device->is_tgtdev_for_dev_replace)
4372 path = btrfs_alloc_path();
4376 path->reada = READA_FORWARD;
4378 mutex_lock(&fs_info->chunk_mutex);
4380 btrfs_device_set_total_bytes(device, new_size);
4381 if (device->writeable) {
4382 device->fs_devices->total_rw_bytes -= diff;
4383 atomic64_sub(diff, &fs_info->free_chunk_space);
4385 mutex_unlock(&fs_info->chunk_mutex);
4388 key.objectid = device->devid;
4389 key.offset = (u64)-1;
4390 key.type = BTRFS_DEV_EXTENT_KEY;
4393 mutex_lock(&fs_info->delete_unused_bgs_mutex);
4394 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4396 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4400 ret = btrfs_previous_item(root, path, 0, key.type);
4402 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4407 btrfs_release_path(path);
4412 slot = path->slots[0];
4413 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4415 if (key.objectid != device->devid) {
4416 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4417 btrfs_release_path(path);
4421 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4422 length = btrfs_dev_extent_length(l, dev_extent);
4424 if (key.offset + length <= new_size) {
4425 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4426 btrfs_release_path(path);
4430 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4431 btrfs_release_path(path);
4433 ret = btrfs_relocate_chunk(fs_info, chunk_offset);
4434 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4435 if (ret && ret != -ENOSPC)
4439 } while (key.offset-- > 0);
4441 if (failed && !retried) {
4445 } else if (failed && retried) {
4450 /* Shrinking succeeded, else we would be at "done". */
4451 trans = btrfs_start_transaction(root, 0);
4452 if (IS_ERR(trans)) {
4453 ret = PTR_ERR(trans);
4457 mutex_lock(&fs_info->chunk_mutex);
4460 * We checked in the above loop all device extents that were already in
4461 * the device tree. However before we have updated the device's
4462 * total_bytes to the new size, we might have had chunk allocations that
4463 * have not complete yet (new block groups attached to transaction
4464 * handles), and therefore their device extents were not yet in the
4465 * device tree and we missed them in the loop above. So if we have any
4466 * pending chunk using a device extent that overlaps the device range
4467 * that we can not use anymore, commit the current transaction and
4468 * repeat the search on the device tree - this way we guarantee we will
4469 * not have chunks using device extents that end beyond 'new_size'.
4471 if (!checked_pending_chunks) {
4472 u64 start = new_size;
4473 u64 len = old_size - new_size;
4475 if (contains_pending_extent(trans->transaction, device,
4477 mutex_unlock(&fs_info->chunk_mutex);
4478 checked_pending_chunks = true;
4481 ret = btrfs_commit_transaction(trans);
4488 btrfs_device_set_disk_total_bytes(device, new_size);
4489 if (list_empty(&device->resized_list))
4490 list_add_tail(&device->resized_list,
4491 &fs_info->fs_devices->resized_devices);
4493 WARN_ON(diff > old_total);
4494 btrfs_set_super_total_bytes(super_copy,
4495 round_down(old_total - diff, fs_info->sectorsize));
4496 mutex_unlock(&fs_info->chunk_mutex);
4498 /* Now btrfs_update_device() will change the on-disk size. */
4499 ret = btrfs_update_device(trans, device);
4500 btrfs_end_transaction(trans);
4502 btrfs_free_path(path);
4504 mutex_lock(&fs_info->chunk_mutex);
4505 btrfs_device_set_total_bytes(device, old_size);
4506 if (device->writeable)
4507 device->fs_devices->total_rw_bytes += diff;
4508 atomic64_add(diff, &fs_info->free_chunk_space);
4509 mutex_unlock(&fs_info->chunk_mutex);
4514 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
4515 struct btrfs_key *key,
4516 struct btrfs_chunk *chunk, int item_size)
4518 struct btrfs_super_block *super_copy = fs_info->super_copy;
4519 struct btrfs_disk_key disk_key;
4523 mutex_lock(&fs_info->chunk_mutex);
4524 array_size = btrfs_super_sys_array_size(super_copy);
4525 if (array_size + item_size + sizeof(disk_key)
4526 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4527 mutex_unlock(&fs_info->chunk_mutex);
4531 ptr = super_copy->sys_chunk_array + array_size;
4532 btrfs_cpu_key_to_disk(&disk_key, key);
4533 memcpy(ptr, &disk_key, sizeof(disk_key));
4534 ptr += sizeof(disk_key);
4535 memcpy(ptr, chunk, item_size);
4536 item_size += sizeof(disk_key);
4537 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4538 mutex_unlock(&fs_info->chunk_mutex);
4544 * sort the devices in descending order by max_avail, total_avail
4546 static int btrfs_cmp_device_info(const void *a, const void *b)
4548 const struct btrfs_device_info *di_a = a;
4549 const struct btrfs_device_info *di_b = b;
4551 if (di_a->max_avail > di_b->max_avail)
4553 if (di_a->max_avail < di_b->max_avail)
4555 if (di_a->total_avail > di_b->total_avail)
4557 if (di_a->total_avail < di_b->total_avail)
4562 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4564 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4567 btrfs_set_fs_incompat(info, RAID56);
4570 #define BTRFS_MAX_DEVS(r) ((BTRFS_MAX_ITEM_SIZE(r->fs_info) \
4571 - sizeof(struct btrfs_chunk)) \
4572 / sizeof(struct btrfs_stripe) + 1)
4574 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
4575 - 2 * sizeof(struct btrfs_disk_key) \
4576 - 2 * sizeof(struct btrfs_chunk)) \
4577 / sizeof(struct btrfs_stripe) + 1)
4579 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4580 u64 start, u64 type)
4582 struct btrfs_fs_info *info = trans->fs_info;
4583 struct btrfs_fs_devices *fs_devices = info->fs_devices;
4584 struct btrfs_device *device;
4585 struct map_lookup *map = NULL;
4586 struct extent_map_tree *em_tree;
4587 struct extent_map *em;
4588 struct btrfs_device_info *devices_info = NULL;
4590 int num_stripes; /* total number of stripes to allocate */
4591 int data_stripes; /* number of stripes that count for
4593 int sub_stripes; /* sub_stripes info for map */
4594 int dev_stripes; /* stripes per dev */
4595 int devs_max; /* max devs to use */
4596 int devs_min; /* min devs needed */
4597 int devs_increment; /* ndevs has to be a multiple of this */
4598 int ncopies; /* how many copies to data has */
4600 u64 max_stripe_size;
4609 BUG_ON(!alloc_profile_is_valid(type, 0));
4611 if (list_empty(&fs_devices->alloc_list))
4614 index = __get_raid_index(type);
4616 sub_stripes = btrfs_raid_array[index].sub_stripes;
4617 dev_stripes = btrfs_raid_array[index].dev_stripes;
4618 devs_max = btrfs_raid_array[index].devs_max;
4619 devs_min = btrfs_raid_array[index].devs_min;
4620 devs_increment = btrfs_raid_array[index].devs_increment;
4621 ncopies = btrfs_raid_array[index].ncopies;
4623 if (type & BTRFS_BLOCK_GROUP_DATA) {
4624 max_stripe_size = SZ_1G;
4625 max_chunk_size = 10 * max_stripe_size;
4627 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4628 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4629 /* for larger filesystems, use larger metadata chunks */
4630 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
4631 max_stripe_size = SZ_1G;
4633 max_stripe_size = SZ_256M;
4634 max_chunk_size = max_stripe_size;
4636 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4637 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4638 max_stripe_size = SZ_32M;
4639 max_chunk_size = 2 * max_stripe_size;
4641 devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4643 btrfs_err(info, "invalid chunk type 0x%llx requested",
4648 /* we don't want a chunk larger than 10% of writeable space */
4649 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4652 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4658 * in the first pass through the devices list, we gather information
4659 * about the available holes on each device.
4662 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
4666 if (!device->writeable) {
4668 "BTRFS: read-only device in alloc_list\n");
4672 if (!device->in_fs_metadata ||
4673 device->is_tgtdev_for_dev_replace)
4676 if (device->total_bytes > device->bytes_used)
4677 total_avail = device->total_bytes - device->bytes_used;
4681 /* If there is no space on this device, skip it. */
4682 if (total_avail == 0)
4685 ret = find_free_dev_extent(trans, device,
4686 max_stripe_size * dev_stripes,
4687 &dev_offset, &max_avail);
4688 if (ret && ret != -ENOSPC)
4692 max_avail = max_stripe_size * dev_stripes;
4694 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4697 if (ndevs == fs_devices->rw_devices) {
4698 WARN(1, "%s: found more than %llu devices\n",
4699 __func__, fs_devices->rw_devices);
4702 devices_info[ndevs].dev_offset = dev_offset;
4703 devices_info[ndevs].max_avail = max_avail;
4704 devices_info[ndevs].total_avail = total_avail;
4705 devices_info[ndevs].dev = device;
4710 * now sort the devices by hole size / available space
4712 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4713 btrfs_cmp_device_info, NULL);
4715 /* round down to number of usable stripes */
4716 ndevs = round_down(ndevs, devs_increment);
4718 if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4723 ndevs = min(ndevs, devs_max);
4726 * the primary goal is to maximize the number of stripes, so use as many
4727 * devices as possible, even if the stripes are not maximum sized.
4729 stripe_size = devices_info[ndevs-1].max_avail;
4730 num_stripes = ndevs * dev_stripes;
4733 * this will have to be fixed for RAID1 and RAID10 over
4736 data_stripes = num_stripes / ncopies;
4738 if (type & BTRFS_BLOCK_GROUP_RAID5)
4739 data_stripes = num_stripes - 1;
4741 if (type & BTRFS_BLOCK_GROUP_RAID6)
4742 data_stripes = num_stripes - 2;
4745 * Use the number of data stripes to figure out how big this chunk
4746 * is really going to be in terms of logical address space,
4747 * and compare that answer with the max chunk size
4749 if (stripe_size * data_stripes > max_chunk_size) {
4750 u64 mask = (1ULL << 24) - 1;
4752 stripe_size = div_u64(max_chunk_size, data_stripes);
4754 /* bump the answer up to a 16MB boundary */
4755 stripe_size = (stripe_size + mask) & ~mask;
4757 /* but don't go higher than the limits we found
4758 * while searching for free extents
4760 if (stripe_size > devices_info[ndevs-1].max_avail)
4761 stripe_size = devices_info[ndevs-1].max_avail;
4764 stripe_size = div_u64(stripe_size, dev_stripes);
4766 /* align to BTRFS_STRIPE_LEN */
4767 stripe_size = round_down(stripe_size, BTRFS_STRIPE_LEN);
4769 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4774 map->num_stripes = num_stripes;
4776 for (i = 0; i < ndevs; ++i) {
4777 for (j = 0; j < dev_stripes; ++j) {
4778 int s = i * dev_stripes + j;
4779 map->stripes[s].dev = devices_info[i].dev;
4780 map->stripes[s].physical = devices_info[i].dev_offset +
4784 map->stripe_len = BTRFS_STRIPE_LEN;
4785 map->io_align = BTRFS_STRIPE_LEN;
4786 map->io_width = BTRFS_STRIPE_LEN;
4788 map->sub_stripes = sub_stripes;
4790 num_bytes = stripe_size * data_stripes;
4792 trace_btrfs_chunk_alloc(info, map, start, num_bytes);
4794 em = alloc_extent_map();
4800 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4801 em->map_lookup = map;
4803 em->len = num_bytes;
4804 em->block_start = 0;
4805 em->block_len = em->len;
4806 em->orig_block_len = stripe_size;
4808 em_tree = &info->mapping_tree.map_tree;
4809 write_lock(&em_tree->lock);
4810 ret = add_extent_mapping(em_tree, em, 0);
4812 write_unlock(&em_tree->lock);
4813 free_extent_map(em);
4817 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4818 refcount_inc(&em->refs);
4819 write_unlock(&em_tree->lock);
4821 ret = btrfs_make_block_group(trans, info, 0, type, start, num_bytes);
4823 goto error_del_extent;
4825 for (i = 0; i < map->num_stripes; i++) {
4826 num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4827 btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4830 atomic64_sub(stripe_size * map->num_stripes, &info->free_chunk_space);
4832 free_extent_map(em);
4833 check_raid56_incompat_flag(info, type);
4835 kfree(devices_info);
4839 write_lock(&em_tree->lock);
4840 remove_extent_mapping(em_tree, em);
4841 write_unlock(&em_tree->lock);
4843 /* One for our allocation */
4844 free_extent_map(em);
4845 /* One for the tree reference */
4846 free_extent_map(em);
4847 /* One for the pending_chunks list reference */
4848 free_extent_map(em);
4850 kfree(devices_info);
4854 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4855 struct btrfs_fs_info *fs_info,
4856 u64 chunk_offset, u64 chunk_size)
4858 struct btrfs_root *extent_root = fs_info->extent_root;
4859 struct btrfs_root *chunk_root = fs_info->chunk_root;
4860 struct btrfs_key key;
4861 struct btrfs_device *device;
4862 struct btrfs_chunk *chunk;
4863 struct btrfs_stripe *stripe;
4864 struct extent_map *em;
4865 struct map_lookup *map;
4872 em = get_chunk_map(fs_info, chunk_offset, chunk_size);
4876 map = em->map_lookup;
4877 item_size = btrfs_chunk_item_size(map->num_stripes);
4878 stripe_size = em->orig_block_len;
4880 chunk = kzalloc(item_size, GFP_NOFS);
4887 * Take the device list mutex to prevent races with the final phase of
4888 * a device replace operation that replaces the device object associated
4889 * with the map's stripes, because the device object's id can change
4890 * at any time during that final phase of the device replace operation
4891 * (dev-replace.c:btrfs_dev_replace_finishing()).
4893 mutex_lock(&fs_info->fs_devices->device_list_mutex);
4894 for (i = 0; i < map->num_stripes; i++) {
4895 device = map->stripes[i].dev;
4896 dev_offset = map->stripes[i].physical;
4898 ret = btrfs_update_device(trans, device);
4901 ret = btrfs_alloc_dev_extent(trans, device, chunk_offset,
4902 dev_offset, stripe_size);
4907 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4911 stripe = &chunk->stripe;
4912 for (i = 0; i < map->num_stripes; i++) {
4913 device = map->stripes[i].dev;
4914 dev_offset = map->stripes[i].physical;
4916 btrfs_set_stack_stripe_devid(stripe, device->devid);
4917 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4918 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4921 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
4923 btrfs_set_stack_chunk_length(chunk, chunk_size);
4924 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4925 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4926 btrfs_set_stack_chunk_type(chunk, map->type);
4927 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4928 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4929 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4930 btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
4931 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4933 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4934 key.type = BTRFS_CHUNK_ITEM_KEY;
4935 key.offset = chunk_offset;
4937 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4938 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4940 * TODO: Cleanup of inserted chunk root in case of
4943 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
4948 free_extent_map(em);
4953 * Chunk allocation falls into two parts. The first part does works
4954 * that make the new allocated chunk useable, but not do any operation
4955 * that modifies the chunk tree. The second part does the works that
4956 * require modifying the chunk tree. This division is important for the
4957 * bootstrap process of adding storage to a seed btrfs.
4959 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4960 struct btrfs_fs_info *fs_info, u64 type)
4964 ASSERT(mutex_is_locked(&fs_info->chunk_mutex));
4965 chunk_offset = find_next_chunk(fs_info);
4966 return __btrfs_alloc_chunk(trans, chunk_offset, type);
4969 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4970 struct btrfs_fs_info *fs_info)
4973 u64 sys_chunk_offset;
4977 chunk_offset = find_next_chunk(fs_info);
4978 alloc_profile = btrfs_metadata_alloc_profile(fs_info);
4979 ret = __btrfs_alloc_chunk(trans, chunk_offset, alloc_profile);
4983 sys_chunk_offset = find_next_chunk(fs_info);
4984 alloc_profile = btrfs_system_alloc_profile(fs_info);
4985 ret = __btrfs_alloc_chunk(trans, sys_chunk_offset, alloc_profile);
4989 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
4993 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
4994 BTRFS_BLOCK_GROUP_RAID10 |
4995 BTRFS_BLOCK_GROUP_RAID5 |
4996 BTRFS_BLOCK_GROUP_DUP)) {
4998 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
5007 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset)
5009 struct extent_map *em;
5010 struct map_lookup *map;
5015 em = get_chunk_map(fs_info, chunk_offset, 1);
5019 map = em->map_lookup;
5020 for (i = 0; i < map->num_stripes; i++) {
5021 if (map->stripes[i].dev->missing) {
5026 if (!map->stripes[i].dev->writeable) {
5033 * If the number of missing devices is larger than max errors,
5034 * we can not write the data into that chunk successfully, so
5037 if (miss_ndevs > btrfs_chunk_max_errors(map))
5040 free_extent_map(em);
5044 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
5046 extent_map_tree_init(&tree->map_tree);
5049 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
5051 struct extent_map *em;
5054 write_lock(&tree->map_tree.lock);
5055 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
5057 remove_extent_mapping(&tree->map_tree, em);
5058 write_unlock(&tree->map_tree.lock);
5062 free_extent_map(em);
5063 /* once for the tree */
5064 free_extent_map(em);
5068 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5070 struct extent_map *em;
5071 struct map_lookup *map;
5074 em = get_chunk_map(fs_info, logical, len);
5077 * We could return errors for these cases, but that could get
5078 * ugly and we'd probably do the same thing which is just not do
5079 * anything else and exit, so return 1 so the callers don't try
5080 * to use other copies.
5084 map = em->map_lookup;
5085 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
5086 ret = map->num_stripes;
5087 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5088 ret = map->sub_stripes;
5089 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5091 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5095 free_extent_map(em);
5097 btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
5098 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) &&
5099 fs_info->dev_replace.tgtdev)
5101 btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
5106 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
5109 struct extent_map *em;
5110 struct map_lookup *map;
5111 unsigned long len = fs_info->sectorsize;
5113 em = get_chunk_map(fs_info, logical, len);
5115 if (!WARN_ON(IS_ERR(em))) {
5116 map = em->map_lookup;
5117 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5118 len = map->stripe_len * nr_data_stripes(map);
5119 free_extent_map(em);
5124 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5126 struct extent_map *em;
5127 struct map_lookup *map;
5130 em = get_chunk_map(fs_info, logical, len);
5132 if(!WARN_ON(IS_ERR(em))) {
5133 map = em->map_lookup;
5134 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5136 free_extent_map(em);
5141 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5142 struct map_lookup *map, int first, int num,
5143 int optimal, int dev_replace_is_ongoing)
5147 struct btrfs_device *srcdev;
5149 if (dev_replace_is_ongoing &&
5150 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5151 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5152 srcdev = fs_info->dev_replace.srcdev;
5157 * try to avoid the drive that is the source drive for a
5158 * dev-replace procedure, only choose it if no other non-missing
5159 * mirror is available
5161 for (tolerance = 0; tolerance < 2; tolerance++) {
5162 if (map->stripes[optimal].dev->bdev &&
5163 (tolerance || map->stripes[optimal].dev != srcdev))
5165 for (i = first; i < first + num; i++) {
5166 if (map->stripes[i].dev->bdev &&
5167 (tolerance || map->stripes[i].dev != srcdev))
5172 /* we couldn't find one that doesn't fail. Just return something
5173 * and the io error handling code will clean up eventually
5178 static inline int parity_smaller(u64 a, u64 b)
5183 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5184 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5186 struct btrfs_bio_stripe s;
5193 for (i = 0; i < num_stripes - 1; i++) {
5194 if (parity_smaller(bbio->raid_map[i],
5195 bbio->raid_map[i+1])) {
5196 s = bbio->stripes[i];
5197 l = bbio->raid_map[i];
5198 bbio->stripes[i] = bbio->stripes[i+1];
5199 bbio->raid_map[i] = bbio->raid_map[i+1];
5200 bbio->stripes[i+1] = s;
5201 bbio->raid_map[i+1] = l;
5209 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5211 struct btrfs_bio *bbio = kzalloc(
5212 /* the size of the btrfs_bio */
5213 sizeof(struct btrfs_bio) +
5214 /* plus the variable array for the stripes */
5215 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5216 /* plus the variable array for the tgt dev */
5217 sizeof(int) * (real_stripes) +
5219 * plus the raid_map, which includes both the tgt dev
5222 sizeof(u64) * (total_stripes),
5223 GFP_NOFS|__GFP_NOFAIL);
5225 atomic_set(&bbio->error, 0);
5226 refcount_set(&bbio->refs, 1);
5231 void btrfs_get_bbio(struct btrfs_bio *bbio)
5233 WARN_ON(!refcount_read(&bbio->refs));
5234 refcount_inc(&bbio->refs);
5237 void btrfs_put_bbio(struct btrfs_bio *bbio)
5241 if (refcount_dec_and_test(&bbio->refs))
5245 /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */
5247 * Please note that, discard won't be sent to target device of device
5250 static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info,
5251 u64 logical, u64 length,
5252 struct btrfs_bio **bbio_ret)
5254 struct extent_map *em;
5255 struct map_lookup *map;
5256 struct btrfs_bio *bbio;
5260 u64 stripe_end_offset;
5267 u32 sub_stripes = 0;
5268 u64 stripes_per_dev = 0;
5269 u32 remaining_stripes = 0;
5270 u32 last_stripe = 0;
5274 /* discard always return a bbio */
5277 em = get_chunk_map(fs_info, logical, length);
5281 map = em->map_lookup;
5282 /* we don't discard raid56 yet */
5283 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5288 offset = logical - em->start;
5289 length = min_t(u64, em->len - offset, length);
5291 stripe_len = map->stripe_len;
5293 * stripe_nr counts the total number of stripes we have to stride
5294 * to get to this block
5296 stripe_nr = div64_u64(offset, stripe_len);
5298 /* stripe_offset is the offset of this block in its stripe */
5299 stripe_offset = offset - stripe_nr * stripe_len;
5301 stripe_nr_end = round_up(offset + length, map->stripe_len);
5302 stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len);
5303 stripe_cnt = stripe_nr_end - stripe_nr;
5304 stripe_end_offset = stripe_nr_end * map->stripe_len -
5307 * after this, stripe_nr is the number of stripes on this
5308 * device we have to walk to find the data, and stripe_index is
5309 * the number of our device in the stripe array
5313 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5314 BTRFS_BLOCK_GROUP_RAID10)) {
5315 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5318 sub_stripes = map->sub_stripes;
5320 factor = map->num_stripes / sub_stripes;
5321 num_stripes = min_t(u64, map->num_stripes,
5322 sub_stripes * stripe_cnt);
5323 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5324 stripe_index *= sub_stripes;
5325 stripes_per_dev = div_u64_rem(stripe_cnt, factor,
5326 &remaining_stripes);
5327 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5328 last_stripe *= sub_stripes;
5329 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5330 BTRFS_BLOCK_GROUP_DUP)) {
5331 num_stripes = map->num_stripes;
5333 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5337 bbio = alloc_btrfs_bio(num_stripes, 0);
5343 for (i = 0; i < num_stripes; i++) {
5344 bbio->stripes[i].physical =
5345 map->stripes[stripe_index].physical +
5346 stripe_offset + stripe_nr * map->stripe_len;
5347 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5349 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5350 BTRFS_BLOCK_GROUP_RAID10)) {
5351 bbio->stripes[i].length = stripes_per_dev *
5354 if (i / sub_stripes < remaining_stripes)
5355 bbio->stripes[i].length +=
5359 * Special for the first stripe and
5362 * |-------|...|-------|
5366 if (i < sub_stripes)
5367 bbio->stripes[i].length -=
5370 if (stripe_index >= last_stripe &&
5371 stripe_index <= (last_stripe +
5373 bbio->stripes[i].length -=
5376 if (i == sub_stripes - 1)
5379 bbio->stripes[i].length = length;
5383 if (stripe_index == map->num_stripes) {
5390 bbio->map_type = map->type;
5391 bbio->num_stripes = num_stripes;
5393 free_extent_map(em);
5398 * In dev-replace case, for repair case (that's the only case where the mirror
5399 * is selected explicitly when calling btrfs_map_block), blocks left of the
5400 * left cursor can also be read from the target drive.
5402 * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the
5404 * For READ, it also needs to be supported using the same mirror number.
5406 * If the requested block is not left of the left cursor, EIO is returned. This
5407 * can happen because btrfs_num_copies() returns one more in the dev-replace
5410 static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info,
5411 u64 logical, u64 length,
5412 u64 srcdev_devid, int *mirror_num,
5415 struct btrfs_bio *bbio = NULL;
5417 int index_srcdev = 0;
5419 u64 physical_of_found = 0;
5423 ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS,
5424 logical, &length, &bbio, 0, 0);
5426 ASSERT(bbio == NULL);
5430 num_stripes = bbio->num_stripes;
5431 if (*mirror_num > num_stripes) {
5433 * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror,
5434 * that means that the requested area is not left of the left
5437 btrfs_put_bbio(bbio);
5442 * process the rest of the function using the mirror_num of the source
5443 * drive. Therefore look it up first. At the end, patch the device
5444 * pointer to the one of the target drive.
5446 for (i = 0; i < num_stripes; i++) {
5447 if (bbio->stripes[i].dev->devid != srcdev_devid)
5451 * In case of DUP, in order to keep it simple, only add the
5452 * mirror with the lowest physical address
5455 physical_of_found <= bbio->stripes[i].physical)
5460 physical_of_found = bbio->stripes[i].physical;
5463 btrfs_put_bbio(bbio);
5469 *mirror_num = index_srcdev + 1;
5470 *physical = physical_of_found;
5474 static void handle_ops_on_dev_replace(enum btrfs_map_op op,
5475 struct btrfs_bio **bbio_ret,
5476 struct btrfs_dev_replace *dev_replace,
5477 int *num_stripes_ret, int *max_errors_ret)
5479 struct btrfs_bio *bbio = *bbio_ret;
5480 u64 srcdev_devid = dev_replace->srcdev->devid;
5481 int tgtdev_indexes = 0;
5482 int num_stripes = *num_stripes_ret;
5483 int max_errors = *max_errors_ret;
5486 if (op == BTRFS_MAP_WRITE) {
5487 int index_where_to_add;
5490 * duplicate the write operations while the dev replace
5491 * procedure is running. Since the copying of the old disk to
5492 * the new disk takes place at run time while the filesystem is
5493 * mounted writable, the regular write operations to the old
5494 * disk have to be duplicated to go to the new disk as well.
5496 * Note that device->missing is handled by the caller, and that
5497 * the write to the old disk is already set up in the stripes
5500 index_where_to_add = num_stripes;
5501 for (i = 0; i < num_stripes; i++) {
5502 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5503 /* write to new disk, too */
5504 struct btrfs_bio_stripe *new =
5505 bbio->stripes + index_where_to_add;
5506 struct btrfs_bio_stripe *old =
5509 new->physical = old->physical;
5510 new->length = old->length;
5511 new->dev = dev_replace->tgtdev;
5512 bbio->tgtdev_map[i] = index_where_to_add;
5513 index_where_to_add++;
5518 num_stripes = index_where_to_add;
5519 } else if (op == BTRFS_MAP_GET_READ_MIRRORS) {
5520 int index_srcdev = 0;
5522 u64 physical_of_found = 0;
5525 * During the dev-replace procedure, the target drive can also
5526 * be used to read data in case it is needed to repair a corrupt
5527 * block elsewhere. This is possible if the requested area is
5528 * left of the left cursor. In this area, the target drive is a
5529 * full copy of the source drive.
5531 for (i = 0; i < num_stripes; i++) {
5532 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5534 * In case of DUP, in order to keep it simple,
5535 * only add the mirror with the lowest physical
5539 physical_of_found <=
5540 bbio->stripes[i].physical)
5544 physical_of_found = bbio->stripes[i].physical;
5548 struct btrfs_bio_stripe *tgtdev_stripe =
5549 bbio->stripes + num_stripes;
5551 tgtdev_stripe->physical = physical_of_found;
5552 tgtdev_stripe->length =
5553 bbio->stripes[index_srcdev].length;
5554 tgtdev_stripe->dev = dev_replace->tgtdev;
5555 bbio->tgtdev_map[index_srcdev] = num_stripes;
5562 *num_stripes_ret = num_stripes;
5563 *max_errors_ret = max_errors;
5564 bbio->num_tgtdevs = tgtdev_indexes;
5568 static bool need_full_stripe(enum btrfs_map_op op)
5570 return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS);
5573 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
5574 enum btrfs_map_op op,
5575 u64 logical, u64 *length,
5576 struct btrfs_bio **bbio_ret,
5577 int mirror_num, int need_raid_map)
5579 struct extent_map *em;
5580 struct map_lookup *map;
5590 int tgtdev_indexes = 0;
5591 struct btrfs_bio *bbio = NULL;
5592 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
5593 int dev_replace_is_ongoing = 0;
5594 int num_alloc_stripes;
5595 int patch_the_first_stripe_for_dev_replace = 0;
5596 u64 physical_to_patch_in_first_stripe = 0;
5597 u64 raid56_full_stripe_start = (u64)-1;
5599 if (op == BTRFS_MAP_DISCARD)
5600 return __btrfs_map_block_for_discard(fs_info, logical,
5603 em = get_chunk_map(fs_info, logical, *length);
5607 map = em->map_lookup;
5608 offset = logical - em->start;
5610 stripe_len = map->stripe_len;
5613 * stripe_nr counts the total number of stripes we have to stride
5614 * to get to this block
5616 stripe_nr = div64_u64(stripe_nr, stripe_len);
5618 stripe_offset = stripe_nr * stripe_len;
5619 if (offset < stripe_offset) {
5621 "stripe math has gone wrong, stripe_offset=%llu, offset=%llu, start=%llu, logical=%llu, stripe_len=%llu",
5622 stripe_offset, offset, em->start, logical,
5624 free_extent_map(em);
5628 /* stripe_offset is the offset of this block in its stripe*/
5629 stripe_offset = offset - stripe_offset;
5631 /* if we're here for raid56, we need to know the stripe aligned start */
5632 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5633 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5634 raid56_full_stripe_start = offset;
5636 /* allow a write of a full stripe, but make sure we don't
5637 * allow straddling of stripes
5639 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5641 raid56_full_stripe_start *= full_stripe_len;
5644 if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5646 /* For writes to RAID[56], allow a full stripeset across all disks.
5647 For other RAID types and for RAID[56] reads, just allow a single
5648 stripe (on a single disk). */
5649 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
5650 (op == BTRFS_MAP_WRITE)) {
5651 max_len = stripe_len * nr_data_stripes(map) -
5652 (offset - raid56_full_stripe_start);
5654 /* we limit the length of each bio to what fits in a stripe */
5655 max_len = stripe_len - stripe_offset;
5657 *length = min_t(u64, em->len - offset, max_len);
5659 *length = em->len - offset;
5662 /* This is for when we're called from btrfs_merge_bio_hook() and all
5663 it cares about is the length */
5667 btrfs_dev_replace_lock(dev_replace, 0);
5668 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5669 if (!dev_replace_is_ongoing)
5670 btrfs_dev_replace_unlock(dev_replace, 0);
5672 btrfs_dev_replace_set_lock_blocking(dev_replace);
5674 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
5675 !need_full_stripe(op) && dev_replace->tgtdev != NULL) {
5676 ret = get_extra_mirror_from_replace(fs_info, logical, *length,
5677 dev_replace->srcdev->devid,
5679 &physical_to_patch_in_first_stripe);
5683 patch_the_first_stripe_for_dev_replace = 1;
5684 } else if (mirror_num > map->num_stripes) {
5690 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5691 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5693 if (!need_full_stripe(op))
5695 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
5696 if (need_full_stripe(op))
5697 num_stripes = map->num_stripes;
5698 else if (mirror_num)
5699 stripe_index = mirror_num - 1;
5701 stripe_index = find_live_mirror(fs_info, map, 0,
5703 current->pid % map->num_stripes,
5704 dev_replace_is_ongoing);
5705 mirror_num = stripe_index + 1;
5708 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
5709 if (need_full_stripe(op)) {
5710 num_stripes = map->num_stripes;
5711 } else if (mirror_num) {
5712 stripe_index = mirror_num - 1;
5717 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5718 u32 factor = map->num_stripes / map->sub_stripes;
5720 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5721 stripe_index *= map->sub_stripes;
5723 if (need_full_stripe(op))
5724 num_stripes = map->sub_stripes;
5725 else if (mirror_num)
5726 stripe_index += mirror_num - 1;
5728 int old_stripe_index = stripe_index;
5729 stripe_index = find_live_mirror(fs_info, map,
5731 map->sub_stripes, stripe_index +
5732 current->pid % map->sub_stripes,
5733 dev_replace_is_ongoing);
5734 mirror_num = stripe_index - old_stripe_index + 1;
5737 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5738 if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) {
5739 /* push stripe_nr back to the start of the full stripe */
5740 stripe_nr = div64_u64(raid56_full_stripe_start,
5741 stripe_len * nr_data_stripes(map));
5743 /* RAID[56] write or recovery. Return all stripes */
5744 num_stripes = map->num_stripes;
5745 max_errors = nr_parity_stripes(map);
5747 *length = map->stripe_len;
5752 * Mirror #0 or #1 means the original data block.
5753 * Mirror #2 is RAID5 parity block.
5754 * Mirror #3 is RAID6 Q block.
5756 stripe_nr = div_u64_rem(stripe_nr,
5757 nr_data_stripes(map), &stripe_index);
5759 stripe_index = nr_data_stripes(map) +
5762 /* We distribute the parity blocks across stripes */
5763 div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
5765 if (!need_full_stripe(op) && mirror_num <= 1)
5770 * after this, stripe_nr is the number of stripes on this
5771 * device we have to walk to find the data, and stripe_index is
5772 * the number of our device in the stripe array
5774 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5776 mirror_num = stripe_index + 1;
5778 if (stripe_index >= map->num_stripes) {
5780 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
5781 stripe_index, map->num_stripes);
5786 num_alloc_stripes = num_stripes;
5787 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) {
5788 if (op == BTRFS_MAP_WRITE)
5789 num_alloc_stripes <<= 1;
5790 if (op == BTRFS_MAP_GET_READ_MIRRORS)
5791 num_alloc_stripes++;
5792 tgtdev_indexes = num_stripes;
5795 bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
5800 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL)
5801 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
5803 /* build raid_map */
5804 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map &&
5805 (need_full_stripe(op) || mirror_num > 1)) {
5809 bbio->raid_map = (u64 *)((void *)bbio->stripes +
5810 sizeof(struct btrfs_bio_stripe) *
5812 sizeof(int) * tgtdev_indexes);
5814 /* Work out the disk rotation on this stripe-set */
5815 div_u64_rem(stripe_nr, num_stripes, &rot);
5817 /* Fill in the logical address of each stripe */
5818 tmp = stripe_nr * nr_data_stripes(map);
5819 for (i = 0; i < nr_data_stripes(map); i++)
5820 bbio->raid_map[(i+rot) % num_stripes] =
5821 em->start + (tmp + i) * map->stripe_len;
5823 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5824 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5825 bbio->raid_map[(i+rot+1) % num_stripes] =
5830 for (i = 0; i < num_stripes; i++) {
5831 bbio->stripes[i].physical =
5832 map->stripes[stripe_index].physical +
5834 stripe_nr * map->stripe_len;
5835 bbio->stripes[i].dev =
5836 map->stripes[stripe_index].dev;
5840 if (need_full_stripe(op))
5841 max_errors = btrfs_chunk_max_errors(map);
5844 sort_parity_stripes(bbio, num_stripes);
5846 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
5847 need_full_stripe(op)) {
5848 handle_ops_on_dev_replace(op, &bbio, dev_replace, &num_stripes,
5853 bbio->map_type = map->type;
5854 bbio->num_stripes = num_stripes;
5855 bbio->max_errors = max_errors;
5856 bbio->mirror_num = mirror_num;
5859 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5860 * mirror_num == num_stripes + 1 && dev_replace target drive is
5861 * available as a mirror
5863 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5864 WARN_ON(num_stripes > 1);
5865 bbio->stripes[0].dev = dev_replace->tgtdev;
5866 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5867 bbio->mirror_num = map->num_stripes + 1;
5870 if (dev_replace_is_ongoing) {
5871 btrfs_dev_replace_clear_lock_blocking(dev_replace);
5872 btrfs_dev_replace_unlock(dev_replace, 0);
5874 free_extent_map(em);
5878 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
5879 u64 logical, u64 *length,
5880 struct btrfs_bio **bbio_ret, int mirror_num)
5882 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret,
5886 /* For Scrub/replace */
5887 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
5888 u64 logical, u64 *length,
5889 struct btrfs_bio **bbio_ret)
5891 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1);
5894 int btrfs_rmap_block(struct btrfs_fs_info *fs_info,
5895 u64 chunk_start, u64 physical, u64 devid,
5896 u64 **logical, int *naddrs, int *stripe_len)
5898 struct extent_map *em;
5899 struct map_lookup *map;
5907 em = get_chunk_map(fs_info, chunk_start, 1);
5911 map = em->map_lookup;
5913 rmap_len = map->stripe_len;
5915 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5916 length = div_u64(length, map->num_stripes / map->sub_stripes);
5917 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5918 length = div_u64(length, map->num_stripes);
5919 else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5920 length = div_u64(length, nr_data_stripes(map));
5921 rmap_len = map->stripe_len * nr_data_stripes(map);
5924 buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
5925 BUG_ON(!buf); /* -ENOMEM */
5927 for (i = 0; i < map->num_stripes; i++) {
5928 if (devid && map->stripes[i].dev->devid != devid)
5930 if (map->stripes[i].physical > physical ||
5931 map->stripes[i].physical + length <= physical)
5934 stripe_nr = physical - map->stripes[i].physical;
5935 stripe_nr = div64_u64(stripe_nr, map->stripe_len);
5937 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5938 stripe_nr = stripe_nr * map->num_stripes + i;
5939 stripe_nr = div_u64(stripe_nr, map->sub_stripes);
5940 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5941 stripe_nr = stripe_nr * map->num_stripes + i;
5942 } /* else if RAID[56], multiply by nr_data_stripes().
5943 * Alternatively, just use rmap_len below instead of
5944 * map->stripe_len */
5946 bytenr = chunk_start + stripe_nr * rmap_len;
5947 WARN_ON(nr >= map->num_stripes);
5948 for (j = 0; j < nr; j++) {
5949 if (buf[j] == bytenr)
5953 WARN_ON(nr >= map->num_stripes);
5960 *stripe_len = rmap_len;
5962 free_extent_map(em);
5966 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
5968 bio->bi_private = bbio->private;
5969 bio->bi_end_io = bbio->end_io;
5972 btrfs_put_bbio(bbio);
5975 static void btrfs_end_bio(struct bio *bio)
5977 struct btrfs_bio *bbio = bio->bi_private;
5978 int is_orig_bio = 0;
5980 if (bio->bi_status) {
5981 atomic_inc(&bbio->error);
5982 if (bio->bi_status == BLK_STS_IOERR ||
5983 bio->bi_status == BLK_STS_TARGET) {
5984 unsigned int stripe_index =
5985 btrfs_io_bio(bio)->stripe_index;
5986 struct btrfs_device *dev;
5988 BUG_ON(stripe_index >= bbio->num_stripes);
5989 dev = bbio->stripes[stripe_index].dev;
5991 if (bio_op(bio) == REQ_OP_WRITE)
5992 btrfs_dev_stat_inc_and_print(dev,
5993 BTRFS_DEV_STAT_WRITE_ERRS);
5995 btrfs_dev_stat_inc_and_print(dev,
5996 BTRFS_DEV_STAT_READ_ERRS);
5997 if (bio->bi_opf & REQ_PREFLUSH)
5998 btrfs_dev_stat_inc_and_print(dev,
5999 BTRFS_DEV_STAT_FLUSH_ERRS);
6004 if (bio == bbio->orig_bio)
6007 btrfs_bio_counter_dec(bbio->fs_info);
6009 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6012 bio = bbio->orig_bio;
6015 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6016 /* only send an error to the higher layers if it is
6017 * beyond the tolerance of the btrfs bio
6019 if (atomic_read(&bbio->error) > bbio->max_errors) {
6020 bio->bi_status = BLK_STS_IOERR;
6023 * this bio is actually up to date, we didn't
6024 * go over the max number of errors
6026 bio->bi_status = BLK_STS_OK;
6029 btrfs_end_bbio(bbio, bio);
6030 } else if (!is_orig_bio) {
6036 * see run_scheduled_bios for a description of why bios are collected for
6039 * This will add one bio to the pending list for a device and make sure
6040 * the work struct is scheduled.
6042 static noinline void btrfs_schedule_bio(struct btrfs_device *device,
6045 struct btrfs_fs_info *fs_info = device->fs_info;
6046 int should_queue = 1;
6047 struct btrfs_pending_bios *pending_bios;
6049 if (device->missing || !device->bdev) {
6054 /* don't bother with additional async steps for reads, right now */
6055 if (bio_op(bio) == REQ_OP_READ) {
6057 btrfsic_submit_bio(bio);
6062 WARN_ON(bio->bi_next);
6063 bio->bi_next = NULL;
6065 spin_lock(&device->io_lock);
6066 if (op_is_sync(bio->bi_opf))
6067 pending_bios = &device->pending_sync_bios;
6069 pending_bios = &device->pending_bios;
6071 if (pending_bios->tail)
6072 pending_bios->tail->bi_next = bio;
6074 pending_bios->tail = bio;
6075 if (!pending_bios->head)
6076 pending_bios->head = bio;
6077 if (device->running_pending)
6080 spin_unlock(&device->io_lock);
6083 btrfs_queue_work(fs_info->submit_workers, &device->work);
6086 static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio,
6087 u64 physical, int dev_nr, int async)
6089 struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
6090 struct btrfs_fs_info *fs_info = bbio->fs_info;
6092 bio->bi_private = bbio;
6093 btrfs_io_bio(bio)->stripe_index = dev_nr;
6094 bio->bi_end_io = btrfs_end_bio;
6095 bio->bi_iter.bi_sector = physical >> 9;
6098 struct rcu_string *name;
6101 name = rcu_dereference(dev->name);
6102 btrfs_debug(fs_info,
6103 "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
6104 bio_op(bio), bio->bi_opf,
6105 (u64)bio->bi_iter.bi_sector,
6106 (u_long)dev->bdev->bd_dev, name->str, dev->devid,
6107 bio->bi_iter.bi_size);
6111 bio_set_dev(bio, dev->bdev);
6113 btrfs_bio_counter_inc_noblocked(fs_info);
6116 btrfs_schedule_bio(dev, bio);
6118 btrfsic_submit_bio(bio);
6121 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6123 atomic_inc(&bbio->error);
6124 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6125 /* Should be the original bio. */
6126 WARN_ON(bio != bbio->orig_bio);
6128 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6129 bio->bi_iter.bi_sector = logical >> 9;
6130 if (atomic_read(&bbio->error) > bbio->max_errors)
6131 bio->bi_status = BLK_STS_IOERR;
6133 bio->bi_status = BLK_STS_OK;
6134 btrfs_end_bbio(bbio, bio);
6138 blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
6139 int mirror_num, int async_submit)
6141 struct btrfs_device *dev;
6142 struct bio *first_bio = bio;
6143 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
6149 struct btrfs_bio *bbio = NULL;
6151 length = bio->bi_iter.bi_size;
6152 map_length = length;
6154 btrfs_bio_counter_inc_blocked(fs_info);
6155 ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical,
6156 &map_length, &bbio, mirror_num, 1);
6158 btrfs_bio_counter_dec(fs_info);
6159 return errno_to_blk_status(ret);
6162 total_devs = bbio->num_stripes;
6163 bbio->orig_bio = first_bio;
6164 bbio->private = first_bio->bi_private;
6165 bbio->end_io = first_bio->bi_end_io;
6166 bbio->fs_info = fs_info;
6167 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6169 if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6170 ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) {
6171 /* In this case, map_length has been set to the length of
6172 a single stripe; not the whole write */
6173 if (bio_op(bio) == REQ_OP_WRITE) {
6174 ret = raid56_parity_write(fs_info, bio, bbio,
6177 ret = raid56_parity_recover(fs_info, bio, bbio,
6178 map_length, mirror_num, 1);
6181 btrfs_bio_counter_dec(fs_info);
6182 return errno_to_blk_status(ret);
6185 if (map_length < length) {
6187 "mapping failed logical %llu bio len %llu len %llu",
6188 logical, length, map_length);
6192 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6193 dev = bbio->stripes[dev_nr].dev;
6194 if (!dev || !dev->bdev ||
6195 (bio_op(first_bio) == REQ_OP_WRITE && !dev->writeable)) {
6196 bbio_error(bbio, first_bio, logical);
6200 if (dev_nr < total_devs - 1)
6201 bio = btrfs_bio_clone(first_bio);
6205 submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical,
6206 dev_nr, async_submit);
6208 btrfs_bio_counter_dec(fs_info);
6212 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
6215 struct btrfs_device *device;
6216 struct btrfs_fs_devices *cur_devices;
6218 cur_devices = fs_info->fs_devices;
6219 while (cur_devices) {
6221 !memcmp(cur_devices->fsid, fsid, BTRFS_FSID_SIZE)) {
6222 device = find_device(cur_devices, devid, uuid);
6226 cur_devices = cur_devices->seed;
6231 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices,
6232 u64 devid, u8 *dev_uuid)
6234 struct btrfs_device *device;
6236 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6240 list_add(&device->dev_list, &fs_devices->devices);
6241 device->fs_devices = fs_devices;
6242 fs_devices->num_devices++;
6244 device->missing = 1;
6245 fs_devices->missing_devices++;
6251 * btrfs_alloc_device - allocate struct btrfs_device
6252 * @fs_info: used only for generating a new devid, can be NULL if
6253 * devid is provided (i.e. @devid != NULL).
6254 * @devid: a pointer to devid for this device. If NULL a new devid
6256 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6259 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6260 * on error. Returned struct is not linked onto any lists and can be
6261 * destroyed with kfree() right away.
6263 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6267 struct btrfs_device *dev;
6270 if (WARN_ON(!devid && !fs_info))
6271 return ERR_PTR(-EINVAL);
6273 dev = __alloc_device();
6282 ret = find_next_devid(fs_info, &tmp);
6284 bio_put(dev->flush_bio);
6286 return ERR_PTR(ret);
6292 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6294 generate_random_uuid(dev->uuid);
6296 btrfs_init_work(&dev->work, btrfs_submit_helper,
6297 pending_bios_fn, NULL, NULL);
6302 /* Return -EIO if any error, otherwise return 0. */
6303 static int btrfs_check_chunk_valid(struct btrfs_fs_info *fs_info,
6304 struct extent_buffer *leaf,
6305 struct btrfs_chunk *chunk, u64 logical)
6313 length = btrfs_chunk_length(leaf, chunk);
6314 stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6315 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6316 sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6317 type = btrfs_chunk_type(leaf, chunk);
6320 btrfs_err(fs_info, "invalid chunk num_stripes: %u",
6324 if (!IS_ALIGNED(logical, fs_info->sectorsize)) {
6325 btrfs_err(fs_info, "invalid chunk logical %llu", logical);
6328 if (btrfs_chunk_sector_size(leaf, chunk) != fs_info->sectorsize) {
6329 btrfs_err(fs_info, "invalid chunk sectorsize %u",
6330 btrfs_chunk_sector_size(leaf, chunk));
6333 if (!length || !IS_ALIGNED(length, fs_info->sectorsize)) {
6334 btrfs_err(fs_info, "invalid chunk length %llu", length);
6337 if (!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN) {
6338 btrfs_err(fs_info, "invalid chunk stripe length: %llu",
6342 if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6344 btrfs_err(fs_info, "unrecognized chunk type: %llu",
6345 ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
6346 BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6347 btrfs_chunk_type(leaf, chunk));
6350 if ((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes != 2) ||
6351 (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) ||
6352 (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) ||
6353 (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) ||
6354 (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) ||
6355 ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
6356 num_stripes != 1)) {
6358 "invalid num_stripes:sub_stripes %u:%u for profile %llu",
6359 num_stripes, sub_stripes,
6360 type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
6367 static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info,
6368 u64 devid, u8 *uuid, bool error)
6371 btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing",
6374 btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing",
6378 static int read_one_chunk(struct btrfs_fs_info *fs_info, struct btrfs_key *key,
6379 struct extent_buffer *leaf,
6380 struct btrfs_chunk *chunk)
6382 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
6383 struct map_lookup *map;
6384 struct extent_map *em;
6388 u8 uuid[BTRFS_UUID_SIZE];
6393 logical = key->offset;
6394 length = btrfs_chunk_length(leaf, chunk);
6395 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6397 ret = btrfs_check_chunk_valid(fs_info, leaf, chunk, logical);
6401 read_lock(&map_tree->map_tree.lock);
6402 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
6403 read_unlock(&map_tree->map_tree.lock);
6405 /* already mapped? */
6406 if (em && em->start <= logical && em->start + em->len > logical) {
6407 free_extent_map(em);
6410 free_extent_map(em);
6413 em = alloc_extent_map();
6416 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6418 free_extent_map(em);
6422 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6423 em->map_lookup = map;
6424 em->start = logical;
6427 em->block_start = 0;
6428 em->block_len = em->len;
6430 map->num_stripes = num_stripes;
6431 map->io_width = btrfs_chunk_io_width(leaf, chunk);
6432 map->io_align = btrfs_chunk_io_align(leaf, chunk);
6433 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6434 map->type = btrfs_chunk_type(leaf, chunk);
6435 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6436 for (i = 0; i < num_stripes; i++) {
6437 map->stripes[i].physical =
6438 btrfs_stripe_offset_nr(leaf, chunk, i);
6439 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6440 read_extent_buffer(leaf, uuid, (unsigned long)
6441 btrfs_stripe_dev_uuid_nr(chunk, i),
6443 map->stripes[i].dev = btrfs_find_device(fs_info, devid,
6445 if (!map->stripes[i].dev &&
6446 !btrfs_test_opt(fs_info, DEGRADED)) {
6447 free_extent_map(em);
6448 btrfs_report_missing_device(fs_info, devid, uuid, true);
6451 if (!map->stripes[i].dev) {
6452 map->stripes[i].dev =
6453 add_missing_dev(fs_info->fs_devices, devid,
6455 if (IS_ERR(map->stripes[i].dev)) {
6456 free_extent_map(em);
6458 "failed to init missing dev %llu: %ld",
6459 devid, PTR_ERR(map->stripes[i].dev));
6460 return PTR_ERR(map->stripes[i].dev);
6462 btrfs_report_missing_device(fs_info, devid, uuid, false);
6464 map->stripes[i].dev->in_fs_metadata = 1;
6467 write_lock(&map_tree->map_tree.lock);
6468 ret = add_extent_mapping(&map_tree->map_tree, em, 0);
6469 write_unlock(&map_tree->map_tree.lock);
6470 BUG_ON(ret); /* Tree corruption */
6471 free_extent_map(em);
6476 static void fill_device_from_item(struct extent_buffer *leaf,
6477 struct btrfs_dev_item *dev_item,
6478 struct btrfs_device *device)
6482 device->devid = btrfs_device_id(leaf, dev_item);
6483 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6484 device->total_bytes = device->disk_total_bytes;
6485 device->commit_total_bytes = device->disk_total_bytes;
6486 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6487 device->commit_bytes_used = device->bytes_used;
6488 device->type = btrfs_device_type(leaf, dev_item);
6489 device->io_align = btrfs_device_io_align(leaf, dev_item);
6490 device->io_width = btrfs_device_io_width(leaf, dev_item);
6491 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6492 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6493 device->is_tgtdev_for_dev_replace = 0;
6495 ptr = btrfs_device_uuid(dev_item);
6496 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6499 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info,
6502 struct btrfs_fs_devices *fs_devices;
6505 BUG_ON(!mutex_is_locked(&uuid_mutex));
6508 fs_devices = fs_info->fs_devices->seed;
6509 while (fs_devices) {
6510 if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE))
6513 fs_devices = fs_devices->seed;
6516 fs_devices = find_fsid(fsid);
6518 if (!btrfs_test_opt(fs_info, DEGRADED))
6519 return ERR_PTR(-ENOENT);
6521 fs_devices = alloc_fs_devices(fsid);
6522 if (IS_ERR(fs_devices))
6525 fs_devices->seeding = 1;
6526 fs_devices->opened = 1;
6530 fs_devices = clone_fs_devices(fs_devices);
6531 if (IS_ERR(fs_devices))
6534 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
6535 fs_info->bdev_holder);
6537 free_fs_devices(fs_devices);
6538 fs_devices = ERR_PTR(ret);
6542 if (!fs_devices->seeding) {
6543 __btrfs_close_devices(fs_devices);
6544 free_fs_devices(fs_devices);
6545 fs_devices = ERR_PTR(-EINVAL);
6549 fs_devices->seed = fs_info->fs_devices->seed;
6550 fs_info->fs_devices->seed = fs_devices;
6555 static int read_one_dev(struct btrfs_fs_info *fs_info,
6556 struct extent_buffer *leaf,
6557 struct btrfs_dev_item *dev_item)
6559 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6560 struct btrfs_device *device;
6563 u8 fs_uuid[BTRFS_FSID_SIZE];
6564 u8 dev_uuid[BTRFS_UUID_SIZE];
6566 devid = btrfs_device_id(leaf, dev_item);
6567 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6569 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6572 if (memcmp(fs_uuid, fs_info->fsid, BTRFS_FSID_SIZE)) {
6573 fs_devices = open_seed_devices(fs_info, fs_uuid);
6574 if (IS_ERR(fs_devices))
6575 return PTR_ERR(fs_devices);
6578 device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid);
6580 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6581 btrfs_report_missing_device(fs_info, devid,
6586 device = add_missing_dev(fs_devices, devid, dev_uuid);
6587 if (IS_ERR(device)) {
6589 "failed to add missing dev %llu: %ld",
6590 devid, PTR_ERR(device));
6591 return PTR_ERR(device);
6593 btrfs_report_missing_device(fs_info, devid, dev_uuid, false);
6595 if (!device->bdev) {
6596 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6597 btrfs_report_missing_device(fs_info,
6598 devid, dev_uuid, true);
6601 btrfs_report_missing_device(fs_info, devid,
6605 if(!device->bdev && !device->missing) {
6607 * this happens when a device that was properly setup
6608 * in the device info lists suddenly goes bad.
6609 * device->bdev is NULL, and so we have to set
6610 * device->missing to one here
6612 device->fs_devices->missing_devices++;
6613 device->missing = 1;
6616 /* Move the device to its own fs_devices */
6617 if (device->fs_devices != fs_devices) {
6618 ASSERT(device->missing);
6620 list_move(&device->dev_list, &fs_devices->devices);
6621 device->fs_devices->num_devices--;
6622 fs_devices->num_devices++;
6624 device->fs_devices->missing_devices--;
6625 fs_devices->missing_devices++;
6627 device->fs_devices = fs_devices;
6631 if (device->fs_devices != fs_info->fs_devices) {
6632 BUG_ON(device->writeable);
6633 if (device->generation !=
6634 btrfs_device_generation(leaf, dev_item))
6638 fill_device_from_item(leaf, dev_item, device);
6639 device->in_fs_metadata = 1;
6640 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
6641 device->fs_devices->total_rw_bytes += device->total_bytes;
6642 atomic64_add(device->total_bytes - device->bytes_used,
6643 &fs_info->free_chunk_space);
6649 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
6651 struct btrfs_root *root = fs_info->tree_root;
6652 struct btrfs_super_block *super_copy = fs_info->super_copy;
6653 struct extent_buffer *sb;
6654 struct btrfs_disk_key *disk_key;
6655 struct btrfs_chunk *chunk;
6657 unsigned long sb_array_offset;
6664 struct btrfs_key key;
6666 ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize);
6668 * This will create extent buffer of nodesize, superblock size is
6669 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6670 * overallocate but we can keep it as-is, only the first page is used.
6672 sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET);
6675 set_extent_buffer_uptodate(sb);
6676 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6678 * The sb extent buffer is artificial and just used to read the system array.
6679 * set_extent_buffer_uptodate() call does not properly mark all it's
6680 * pages up-to-date when the page is larger: extent does not cover the
6681 * whole page and consequently check_page_uptodate does not find all
6682 * the page's extents up-to-date (the hole beyond sb),
6683 * write_extent_buffer then triggers a WARN_ON.
6685 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6686 * but sb spans only this function. Add an explicit SetPageUptodate call
6687 * to silence the warning eg. on PowerPC 64.
6689 if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
6690 SetPageUptodate(sb->pages[0]);
6692 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6693 array_size = btrfs_super_sys_array_size(super_copy);
6695 array_ptr = super_copy->sys_chunk_array;
6696 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6699 while (cur_offset < array_size) {
6700 disk_key = (struct btrfs_disk_key *)array_ptr;
6701 len = sizeof(*disk_key);
6702 if (cur_offset + len > array_size)
6703 goto out_short_read;
6705 btrfs_disk_key_to_cpu(&key, disk_key);
6708 sb_array_offset += len;
6711 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6712 chunk = (struct btrfs_chunk *)sb_array_offset;
6714 * At least one btrfs_chunk with one stripe must be
6715 * present, exact stripe count check comes afterwards
6717 len = btrfs_chunk_item_size(1);
6718 if (cur_offset + len > array_size)
6719 goto out_short_read;
6721 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6724 "invalid number of stripes %u in sys_array at offset %u",
6725 num_stripes, cur_offset);
6730 type = btrfs_chunk_type(sb, chunk);
6731 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
6733 "invalid chunk type %llu in sys_array at offset %u",
6739 len = btrfs_chunk_item_size(num_stripes);
6740 if (cur_offset + len > array_size)
6741 goto out_short_read;
6743 ret = read_one_chunk(fs_info, &key, sb, chunk);
6748 "unexpected item type %u in sys_array at offset %u",
6749 (u32)key.type, cur_offset);
6754 sb_array_offset += len;
6757 clear_extent_buffer_uptodate(sb);
6758 free_extent_buffer_stale(sb);
6762 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
6764 clear_extent_buffer_uptodate(sb);
6765 free_extent_buffer_stale(sb);
6770 * Check if all chunks in the fs are OK for read-write degraded mount
6772 * Return true if all chunks meet the minimal RW mount requirements.
6773 * Return false if any chunk doesn't meet the minimal RW mount requirements.
6775 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info)
6777 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
6778 struct extent_map *em;
6782 read_lock(&map_tree->map_tree.lock);
6783 em = lookup_extent_mapping(&map_tree->map_tree, 0, (u64)-1);
6784 read_unlock(&map_tree->map_tree.lock);
6785 /* No chunk at all? Return false anyway */
6791 struct map_lookup *map;
6796 map = em->map_lookup;
6798 btrfs_get_num_tolerated_disk_barrier_failures(
6800 for (i = 0; i < map->num_stripes; i++) {
6801 struct btrfs_device *dev = map->stripes[i].dev;
6803 if (!dev || !dev->bdev || dev->missing ||
6804 dev->last_flush_error)
6807 if (missing > max_tolerated) {
6809 "chunk %llu missing %d devices, max tolerance is %d for writeable mount",
6810 em->start, missing, max_tolerated);
6811 free_extent_map(em);
6815 next_start = extent_map_end(em);
6816 free_extent_map(em);
6818 read_lock(&map_tree->map_tree.lock);
6819 em = lookup_extent_mapping(&map_tree->map_tree, next_start,
6820 (u64)(-1) - next_start);
6821 read_unlock(&map_tree->map_tree.lock);
6827 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
6829 struct btrfs_root *root = fs_info->chunk_root;
6830 struct btrfs_path *path;
6831 struct extent_buffer *leaf;
6832 struct btrfs_key key;
6833 struct btrfs_key found_key;
6838 path = btrfs_alloc_path();
6842 mutex_lock(&uuid_mutex);
6843 mutex_lock(&fs_info->chunk_mutex);
6846 * Read all device items, and then all the chunk items. All
6847 * device items are found before any chunk item (their object id
6848 * is smaller than the lowest possible object id for a chunk
6849 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6851 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6854 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6858 leaf = path->nodes[0];
6859 slot = path->slots[0];
6860 if (slot >= btrfs_header_nritems(leaf)) {
6861 ret = btrfs_next_leaf(root, path);
6868 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6869 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6870 struct btrfs_dev_item *dev_item;
6871 dev_item = btrfs_item_ptr(leaf, slot,
6872 struct btrfs_dev_item);
6873 ret = read_one_dev(fs_info, leaf, dev_item);
6877 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6878 struct btrfs_chunk *chunk;
6879 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6880 ret = read_one_chunk(fs_info, &found_key, leaf, chunk);
6888 * After loading chunk tree, we've got all device information,
6889 * do another round of validation checks.
6891 if (total_dev != fs_info->fs_devices->total_devices) {
6893 "super_num_devices %llu mismatch with num_devices %llu found here",
6894 btrfs_super_num_devices(fs_info->super_copy),
6899 if (btrfs_super_total_bytes(fs_info->super_copy) <
6900 fs_info->fs_devices->total_rw_bytes) {
6902 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
6903 btrfs_super_total_bytes(fs_info->super_copy),
6904 fs_info->fs_devices->total_rw_bytes);
6910 mutex_unlock(&fs_info->chunk_mutex);
6911 mutex_unlock(&uuid_mutex);
6913 btrfs_free_path(path);
6917 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6919 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6920 struct btrfs_device *device;
6922 while (fs_devices) {
6923 mutex_lock(&fs_devices->device_list_mutex);
6924 list_for_each_entry(device, &fs_devices->devices, dev_list)
6925 device->fs_info = fs_info;
6926 mutex_unlock(&fs_devices->device_list_mutex);
6928 fs_devices = fs_devices->seed;
6932 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6936 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6937 btrfs_dev_stat_reset(dev, i);
6940 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6942 struct btrfs_key key;
6943 struct btrfs_key found_key;
6944 struct btrfs_root *dev_root = fs_info->dev_root;
6945 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6946 struct extent_buffer *eb;
6949 struct btrfs_device *device;
6950 struct btrfs_path *path = NULL;
6953 path = btrfs_alloc_path();
6959 mutex_lock(&fs_devices->device_list_mutex);
6960 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6962 struct btrfs_dev_stats_item *ptr;
6964 key.objectid = BTRFS_DEV_STATS_OBJECTID;
6965 key.type = BTRFS_PERSISTENT_ITEM_KEY;
6966 key.offset = device->devid;
6967 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6969 __btrfs_reset_dev_stats(device);
6970 device->dev_stats_valid = 1;
6971 btrfs_release_path(path);
6974 slot = path->slots[0];
6975 eb = path->nodes[0];
6976 btrfs_item_key_to_cpu(eb, &found_key, slot);
6977 item_size = btrfs_item_size_nr(eb, slot);
6979 ptr = btrfs_item_ptr(eb, slot,
6980 struct btrfs_dev_stats_item);
6982 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6983 if (item_size >= (1 + i) * sizeof(__le64))
6984 btrfs_dev_stat_set(device, i,
6985 btrfs_dev_stats_value(eb, ptr, i));
6987 btrfs_dev_stat_reset(device, i);
6990 device->dev_stats_valid = 1;
6991 btrfs_dev_stat_print_on_load(device);
6992 btrfs_release_path(path);
6994 mutex_unlock(&fs_devices->device_list_mutex);
6997 btrfs_free_path(path);
6998 return ret < 0 ? ret : 0;
7001 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
7002 struct btrfs_fs_info *fs_info,
7003 struct btrfs_device *device)
7005 struct btrfs_root *dev_root = fs_info->dev_root;
7006 struct btrfs_path *path;
7007 struct btrfs_key key;
7008 struct extent_buffer *eb;
7009 struct btrfs_dev_stats_item *ptr;
7013 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7014 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7015 key.offset = device->devid;
7017 path = btrfs_alloc_path();
7020 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
7022 btrfs_warn_in_rcu(fs_info,
7023 "error %d while searching for dev_stats item for device %s",
7024 ret, rcu_str_deref(device->name));
7029 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
7030 /* need to delete old one and insert a new one */
7031 ret = btrfs_del_item(trans, dev_root, path);
7033 btrfs_warn_in_rcu(fs_info,
7034 "delete too small dev_stats item for device %s failed %d",
7035 rcu_str_deref(device->name), ret);
7042 /* need to insert a new item */
7043 btrfs_release_path(path);
7044 ret = btrfs_insert_empty_item(trans, dev_root, path,
7045 &key, sizeof(*ptr));
7047 btrfs_warn_in_rcu(fs_info,
7048 "insert dev_stats item for device %s failed %d",
7049 rcu_str_deref(device->name), ret);
7054 eb = path->nodes[0];
7055 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7056 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7057 btrfs_set_dev_stats_value(eb, ptr, i,
7058 btrfs_dev_stat_read(device, i));
7059 btrfs_mark_buffer_dirty(eb);
7062 btrfs_free_path(path);
7067 * called from commit_transaction. Writes all changed device stats to disk.
7069 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
7070 struct btrfs_fs_info *fs_info)
7072 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7073 struct btrfs_device *device;
7077 mutex_lock(&fs_devices->device_list_mutex);
7078 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7079 stats_cnt = atomic_read(&device->dev_stats_ccnt);
7080 if (!device->dev_stats_valid || stats_cnt == 0)
7085 * There is a LOAD-LOAD control dependency between the value of
7086 * dev_stats_ccnt and updating the on-disk values which requires
7087 * reading the in-memory counters. Such control dependencies
7088 * require explicit read memory barriers.
7090 * This memory barriers pairs with smp_mb__before_atomic in
7091 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full
7092 * barrier implied by atomic_xchg in
7093 * btrfs_dev_stats_read_and_reset
7097 ret = update_dev_stat_item(trans, fs_info, device);
7099 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7101 mutex_unlock(&fs_devices->device_list_mutex);
7106 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7108 btrfs_dev_stat_inc(dev, index);
7109 btrfs_dev_stat_print_on_error(dev);
7112 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
7114 if (!dev->dev_stats_valid)
7116 btrfs_err_rl_in_rcu(dev->fs_info,
7117 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7118 rcu_str_deref(dev->name),
7119 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7120 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7121 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7122 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7123 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7126 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7130 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7131 if (btrfs_dev_stat_read(dev, i) != 0)
7133 if (i == BTRFS_DEV_STAT_VALUES_MAX)
7134 return; /* all values == 0, suppress message */
7136 btrfs_info_in_rcu(dev->fs_info,
7137 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7138 rcu_str_deref(dev->name),
7139 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7140 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7141 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7142 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7143 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7146 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
7147 struct btrfs_ioctl_get_dev_stats *stats)
7149 struct btrfs_device *dev;
7150 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7153 mutex_lock(&fs_devices->device_list_mutex);
7154 dev = btrfs_find_device(fs_info, stats->devid, NULL, NULL);
7155 mutex_unlock(&fs_devices->device_list_mutex);
7158 btrfs_warn(fs_info, "get dev_stats failed, device not found");
7160 } else if (!dev->dev_stats_valid) {
7161 btrfs_warn(fs_info, "get dev_stats failed, not yet valid");
7163 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7164 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7165 if (stats->nr_items > i)
7167 btrfs_dev_stat_read_and_reset(dev, i);
7169 btrfs_dev_stat_reset(dev, i);
7172 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7173 if (stats->nr_items > i)
7174 stats->values[i] = btrfs_dev_stat_read(dev, i);
7176 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7177 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7181 void btrfs_scratch_superblocks(struct block_device *bdev, const char *device_path)
7183 struct buffer_head *bh;
7184 struct btrfs_super_block *disk_super;
7190 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX;
7193 if (btrfs_read_dev_one_super(bdev, copy_num, &bh))
7196 disk_super = (struct btrfs_super_block *)bh->b_data;
7198 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
7199 set_buffer_dirty(bh);
7200 sync_dirty_buffer(bh);
7204 /* Notify udev that device has changed */
7205 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
7207 /* Update ctime/mtime for device path for libblkid */
7208 update_dev_time(device_path);
7212 * Update the size of all devices, which is used for writing out the
7215 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
7217 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7218 struct btrfs_device *curr, *next;
7220 if (list_empty(&fs_devices->resized_devices))
7223 mutex_lock(&fs_devices->device_list_mutex);
7224 mutex_lock(&fs_info->chunk_mutex);
7225 list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
7227 list_del_init(&curr->resized_list);
7228 curr->commit_total_bytes = curr->disk_total_bytes;
7230 mutex_unlock(&fs_info->chunk_mutex);
7231 mutex_unlock(&fs_devices->device_list_mutex);
7234 /* Must be invoked during the transaction commit */
7235 void btrfs_update_commit_device_bytes_used(struct btrfs_fs_info *fs_info,
7236 struct btrfs_transaction *transaction)
7238 struct extent_map *em;
7239 struct map_lookup *map;
7240 struct btrfs_device *dev;
7243 if (list_empty(&transaction->pending_chunks))
7246 /* In order to kick the device replace finish process */
7247 mutex_lock(&fs_info->chunk_mutex);
7248 list_for_each_entry(em, &transaction->pending_chunks, list) {
7249 map = em->map_lookup;
7251 for (i = 0; i < map->num_stripes; i++) {
7252 dev = map->stripes[i].dev;
7253 dev->commit_bytes_used = dev->bytes_used;
7256 mutex_unlock(&fs_info->chunk_mutex);
7259 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
7261 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7262 while (fs_devices) {
7263 fs_devices->fs_info = fs_info;
7264 fs_devices = fs_devices->seed;
7268 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
7270 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7271 while (fs_devices) {
7272 fs_devices->fs_info = NULL;
7273 fs_devices = fs_devices->seed;