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/random.h>
24 #include <linux/iocontext.h>
25 #include <linux/capability.h>
26 #include <linux/ratelimit.h>
27 #include <linux/kthread.h>
28 #include <linux/raid/pq.h>
29 #include <linux/semaphore.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 static int init_first_rw_device(struct btrfs_trans_handle *trans,
46 struct btrfs_root *root,
47 struct btrfs_device *device);
48 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
49 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
50 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
51 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
53 static DEFINE_MUTEX(uuid_mutex);
54 static LIST_HEAD(fs_uuids);
56 static void lock_chunks(struct btrfs_root *root)
58 mutex_lock(&root->fs_info->chunk_mutex);
61 static void unlock_chunks(struct btrfs_root *root)
63 mutex_unlock(&root->fs_info->chunk_mutex);
66 static struct btrfs_fs_devices *__alloc_fs_devices(void)
68 struct btrfs_fs_devices *fs_devs;
70 fs_devs = kzalloc(sizeof(*fs_devs), GFP_NOFS);
72 return ERR_PTR(-ENOMEM);
74 mutex_init(&fs_devs->device_list_mutex);
76 INIT_LIST_HEAD(&fs_devs->devices);
77 INIT_LIST_HEAD(&fs_devs->resized_devices);
78 INIT_LIST_HEAD(&fs_devs->alloc_list);
79 INIT_LIST_HEAD(&fs_devs->list);
85 * alloc_fs_devices - allocate struct btrfs_fs_devices
86 * @fsid: a pointer to UUID for this FS. If NULL a new UUID is
89 * Return: a pointer to a new &struct btrfs_fs_devices on success;
90 * ERR_PTR() on error. Returned struct is not linked onto any lists and
91 * can be destroyed with kfree() right away.
93 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
95 struct btrfs_fs_devices *fs_devs;
97 fs_devs = __alloc_fs_devices();
102 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
104 generate_random_uuid(fs_devs->fsid);
109 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
111 struct btrfs_device *device;
112 WARN_ON(fs_devices->opened);
113 while (!list_empty(&fs_devices->devices)) {
114 device = list_entry(fs_devices->devices.next,
115 struct btrfs_device, dev_list);
116 list_del(&device->dev_list);
117 rcu_string_free(device->name);
123 static void btrfs_kobject_uevent(struct block_device *bdev,
124 enum kobject_action action)
128 ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
130 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
132 kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
133 &disk_to_dev(bdev->bd_disk)->kobj);
136 void btrfs_cleanup_fs_uuids(void)
138 struct btrfs_fs_devices *fs_devices;
140 while (!list_empty(&fs_uuids)) {
141 fs_devices = list_entry(fs_uuids.next,
142 struct btrfs_fs_devices, list);
143 list_del(&fs_devices->list);
144 free_fs_devices(fs_devices);
148 static struct btrfs_device *__alloc_device(void)
150 struct btrfs_device *dev;
152 dev = kzalloc(sizeof(*dev), GFP_NOFS);
154 return ERR_PTR(-ENOMEM);
156 INIT_LIST_HEAD(&dev->dev_list);
157 INIT_LIST_HEAD(&dev->dev_alloc_list);
158 INIT_LIST_HEAD(&dev->resized_list);
160 spin_lock_init(&dev->io_lock);
162 spin_lock_init(&dev->reada_lock);
163 atomic_set(&dev->reada_in_flight, 0);
164 atomic_set(&dev->dev_stats_ccnt, 0);
165 INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_WAIT);
166 INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_WAIT);
171 static noinline struct btrfs_device *__find_device(struct list_head *head,
174 struct btrfs_device *dev;
176 list_for_each_entry(dev, head, dev_list) {
177 if (dev->devid == devid &&
178 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
185 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
187 struct btrfs_fs_devices *fs_devices;
189 list_for_each_entry(fs_devices, &fs_uuids, list) {
190 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
197 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
198 int flush, struct block_device **bdev,
199 struct buffer_head **bh)
203 *bdev = blkdev_get_by_path(device_path, flags, holder);
206 ret = PTR_ERR(*bdev);
207 printk(KERN_INFO "BTRFS: open %s failed\n", device_path);
212 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
213 ret = set_blocksize(*bdev, 4096);
215 blkdev_put(*bdev, flags);
218 invalidate_bdev(*bdev);
219 *bh = btrfs_read_dev_super(*bdev);
222 blkdev_put(*bdev, flags);
234 static void requeue_list(struct btrfs_pending_bios *pending_bios,
235 struct bio *head, struct bio *tail)
238 struct bio *old_head;
240 old_head = pending_bios->head;
241 pending_bios->head = head;
242 if (pending_bios->tail)
243 tail->bi_next = old_head;
245 pending_bios->tail = tail;
249 * we try to collect pending bios for a device so we don't get a large
250 * number of procs sending bios down to the same device. This greatly
251 * improves the schedulers ability to collect and merge the bios.
253 * But, it also turns into a long list of bios to process and that is sure
254 * to eventually make the worker thread block. The solution here is to
255 * make some progress and then put this work struct back at the end of
256 * the list if the block device is congested. This way, multiple devices
257 * can make progress from a single worker thread.
259 static noinline void run_scheduled_bios(struct btrfs_device *device)
262 struct backing_dev_info *bdi;
263 struct btrfs_fs_info *fs_info;
264 struct btrfs_pending_bios *pending_bios;
268 unsigned long num_run;
269 unsigned long batch_run = 0;
271 unsigned long last_waited = 0;
273 int sync_pending = 0;
274 struct blk_plug plug;
277 * this function runs all the bios we've collected for
278 * a particular device. We don't want to wander off to
279 * another device without first sending all of these down.
280 * So, setup a plug here and finish it off before we return
282 blk_start_plug(&plug);
284 bdi = blk_get_backing_dev_info(device->bdev);
285 fs_info = device->dev_root->fs_info;
286 limit = btrfs_async_submit_limit(fs_info);
287 limit = limit * 2 / 3;
290 spin_lock(&device->io_lock);
295 /* take all the bios off the list at once and process them
296 * later on (without the lock held). But, remember the
297 * tail and other pointers so the bios can be properly reinserted
298 * into the list if we hit congestion
300 if (!force_reg && device->pending_sync_bios.head) {
301 pending_bios = &device->pending_sync_bios;
304 pending_bios = &device->pending_bios;
308 pending = pending_bios->head;
309 tail = pending_bios->tail;
310 WARN_ON(pending && !tail);
313 * if pending was null this time around, no bios need processing
314 * at all and we can stop. Otherwise it'll loop back up again
315 * and do an additional check so no bios are missed.
317 * device->running_pending is used to synchronize with the
320 if (device->pending_sync_bios.head == NULL &&
321 device->pending_bios.head == NULL) {
323 device->running_pending = 0;
326 device->running_pending = 1;
329 pending_bios->head = NULL;
330 pending_bios->tail = NULL;
332 spin_unlock(&device->io_lock);
337 /* we want to work on both lists, but do more bios on the
338 * sync list than the regular list
341 pending_bios != &device->pending_sync_bios &&
342 device->pending_sync_bios.head) ||
343 (num_run > 64 && pending_bios == &device->pending_sync_bios &&
344 device->pending_bios.head)) {
345 spin_lock(&device->io_lock);
346 requeue_list(pending_bios, pending, tail);
351 pending = pending->bi_next;
354 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
355 waitqueue_active(&fs_info->async_submit_wait))
356 wake_up(&fs_info->async_submit_wait);
358 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
361 * if we're doing the sync list, record that our
362 * plug has some sync requests on it
364 * If we're doing the regular list and there are
365 * sync requests sitting around, unplug before
368 if (pending_bios == &device->pending_sync_bios) {
370 } else if (sync_pending) {
371 blk_finish_plug(&plug);
372 blk_start_plug(&plug);
376 btrfsic_submit_bio(cur->bi_rw, cur);
383 * we made progress, there is more work to do and the bdi
384 * is now congested. Back off and let other work structs
387 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
388 fs_info->fs_devices->open_devices > 1) {
389 struct io_context *ioc;
391 ioc = current->io_context;
394 * the main goal here is that we don't want to
395 * block if we're going to be able to submit
396 * more requests without blocking.
398 * This code does two great things, it pokes into
399 * the elevator code from a filesystem _and_
400 * it makes assumptions about how batching works.
402 if (ioc && ioc->nr_batch_requests > 0 &&
403 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
405 ioc->last_waited == last_waited)) {
407 * we want to go through our batch of
408 * requests and stop. So, we copy out
409 * the ioc->last_waited time and test
410 * against it before looping
412 last_waited = ioc->last_waited;
417 spin_lock(&device->io_lock);
418 requeue_list(pending_bios, pending, tail);
419 device->running_pending = 1;
421 spin_unlock(&device->io_lock);
422 btrfs_queue_work(fs_info->submit_workers,
426 /* unplug every 64 requests just for good measure */
427 if (batch_run % 64 == 0) {
428 blk_finish_plug(&plug);
429 blk_start_plug(&plug);
438 spin_lock(&device->io_lock);
439 if (device->pending_bios.head || device->pending_sync_bios.head)
441 spin_unlock(&device->io_lock);
444 blk_finish_plug(&plug);
447 static void pending_bios_fn(struct btrfs_work *work)
449 struct btrfs_device *device;
451 device = container_of(work, struct btrfs_device, work);
452 run_scheduled_bios(device);
456 * Add new device to list of registered devices
459 * 1 - first time device is seen
460 * 0 - device already known
463 static noinline int device_list_add(const char *path,
464 struct btrfs_super_block *disk_super,
465 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
467 struct btrfs_device *device;
468 struct btrfs_fs_devices *fs_devices;
469 struct rcu_string *name;
471 u64 found_transid = btrfs_super_generation(disk_super);
473 fs_devices = find_fsid(disk_super->fsid);
475 fs_devices = alloc_fs_devices(disk_super->fsid);
476 if (IS_ERR(fs_devices))
477 return PTR_ERR(fs_devices);
479 list_add(&fs_devices->list, &fs_uuids);
483 device = __find_device(&fs_devices->devices, devid,
484 disk_super->dev_item.uuid);
488 if (fs_devices->opened)
491 device = btrfs_alloc_device(NULL, &devid,
492 disk_super->dev_item.uuid);
493 if (IS_ERR(device)) {
494 /* we can safely leave the fs_devices entry around */
495 return PTR_ERR(device);
498 name = rcu_string_strdup(path, GFP_NOFS);
503 rcu_assign_pointer(device->name, name);
505 mutex_lock(&fs_devices->device_list_mutex);
506 list_add_rcu(&device->dev_list, &fs_devices->devices);
507 fs_devices->num_devices++;
508 mutex_unlock(&fs_devices->device_list_mutex);
511 device->fs_devices = fs_devices;
512 } else if (!device->name || strcmp(device->name->str, path)) {
514 * When FS is already mounted.
515 * 1. If you are here and if the device->name is NULL that
516 * means this device was missing at time of FS mount.
517 * 2. If you are here and if the device->name is different
518 * from 'path' that means either
519 * a. The same device disappeared and reappeared with
521 * b. The missing-disk-which-was-replaced, has
524 * We must allow 1 and 2a above. But 2b would be a spurious
527 * Further in case of 1 and 2a above, the disk at 'path'
528 * would have missed some transaction when it was away and
529 * in case of 2a the stale bdev has to be updated as well.
530 * 2b must not be allowed at all time.
534 * As of now don't allow update to btrfs_fs_device through
535 * the btrfs dev scan cli, after FS has been mounted.
537 if (fs_devices->opened) {
541 * That is if the FS is _not_ mounted and if you
542 * are here, that means there is more than one
543 * disk with same uuid and devid.We keep the one
544 * with larger generation number or the last-in if
545 * generation are equal.
547 if (found_transid < device->generation)
551 name = rcu_string_strdup(path, GFP_NOFS);
554 rcu_string_free(device->name);
555 rcu_assign_pointer(device->name, name);
556 if (device->missing) {
557 fs_devices->missing_devices--;
563 * Unmount does not free the btrfs_device struct but would zero
564 * generation along with most of the other members. So just update
565 * it back. We need it to pick the disk with largest generation
568 if (!fs_devices->opened)
569 device->generation = found_transid;
571 *fs_devices_ret = fs_devices;
576 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
578 struct btrfs_fs_devices *fs_devices;
579 struct btrfs_device *device;
580 struct btrfs_device *orig_dev;
582 fs_devices = alloc_fs_devices(orig->fsid);
583 if (IS_ERR(fs_devices))
586 mutex_lock(&orig->device_list_mutex);
587 fs_devices->total_devices = orig->total_devices;
589 /* We have held the volume lock, it is safe to get the devices. */
590 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
591 struct rcu_string *name;
593 device = btrfs_alloc_device(NULL, &orig_dev->devid,
599 * This is ok to do without rcu read locked because we hold the
600 * uuid mutex so nothing we touch in here is going to disappear.
602 if (orig_dev->name) {
603 name = rcu_string_strdup(orig_dev->name->str, GFP_NOFS);
608 rcu_assign_pointer(device->name, name);
611 list_add(&device->dev_list, &fs_devices->devices);
612 device->fs_devices = fs_devices;
613 fs_devices->num_devices++;
615 mutex_unlock(&orig->device_list_mutex);
618 mutex_unlock(&orig->device_list_mutex);
619 free_fs_devices(fs_devices);
620 return ERR_PTR(-ENOMEM);
623 void btrfs_close_extra_devices(struct btrfs_fs_info *fs_info,
624 struct btrfs_fs_devices *fs_devices, int step)
626 struct btrfs_device *device, *next;
627 struct btrfs_device *latest_dev = NULL;
629 mutex_lock(&uuid_mutex);
631 /* This is the initialized path, it is safe to release the devices. */
632 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
633 if (device->in_fs_metadata) {
634 if (!device->is_tgtdev_for_dev_replace &&
636 device->generation > latest_dev->generation)) {
642 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
644 * In the first step, keep the device which has
645 * the correct fsid and the devid that is used
646 * for the dev_replace procedure.
647 * In the second step, the dev_replace state is
648 * read from the device tree and it is known
649 * whether the procedure is really active or
650 * not, which means whether this device is
651 * used or whether it should be removed.
653 if (step == 0 || device->is_tgtdev_for_dev_replace) {
658 blkdev_put(device->bdev, device->mode);
660 fs_devices->open_devices--;
662 if (device->writeable) {
663 list_del_init(&device->dev_alloc_list);
664 device->writeable = 0;
665 if (!device->is_tgtdev_for_dev_replace)
666 fs_devices->rw_devices--;
668 list_del_init(&device->dev_list);
669 fs_devices->num_devices--;
670 rcu_string_free(device->name);
674 if (fs_devices->seed) {
675 fs_devices = fs_devices->seed;
679 fs_devices->latest_bdev = latest_dev->bdev;
681 mutex_unlock(&uuid_mutex);
684 static void __free_device(struct work_struct *work)
686 struct btrfs_device *device;
688 device = container_of(work, struct btrfs_device, rcu_work);
691 blkdev_put(device->bdev, device->mode);
693 rcu_string_free(device->name);
697 static void free_device(struct rcu_head *head)
699 struct btrfs_device *device;
701 device = container_of(head, struct btrfs_device, rcu);
703 INIT_WORK(&device->rcu_work, __free_device);
704 schedule_work(&device->rcu_work);
707 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
709 struct btrfs_device *device;
711 if (--fs_devices->opened > 0)
714 mutex_lock(&fs_devices->device_list_mutex);
715 list_for_each_entry(device, &fs_devices->devices, dev_list) {
716 struct btrfs_device *new_device;
717 struct rcu_string *name;
720 fs_devices->open_devices--;
722 if (device->writeable &&
723 device->devid != BTRFS_DEV_REPLACE_DEVID) {
724 list_del_init(&device->dev_alloc_list);
725 fs_devices->rw_devices--;
729 fs_devices->missing_devices--;
731 new_device = btrfs_alloc_device(NULL, &device->devid,
733 BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
735 /* Safe because we are under uuid_mutex */
737 name = rcu_string_strdup(device->name->str, GFP_NOFS);
738 BUG_ON(!name); /* -ENOMEM */
739 rcu_assign_pointer(new_device->name, name);
742 list_replace_rcu(&device->dev_list, &new_device->dev_list);
743 new_device->fs_devices = device->fs_devices;
745 call_rcu(&device->rcu, free_device);
747 mutex_unlock(&fs_devices->device_list_mutex);
749 WARN_ON(fs_devices->open_devices);
750 WARN_ON(fs_devices->rw_devices);
751 fs_devices->opened = 0;
752 fs_devices->seeding = 0;
757 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
759 struct btrfs_fs_devices *seed_devices = NULL;
762 mutex_lock(&uuid_mutex);
763 ret = __btrfs_close_devices(fs_devices);
764 if (!fs_devices->opened) {
765 seed_devices = fs_devices->seed;
766 fs_devices->seed = NULL;
768 mutex_unlock(&uuid_mutex);
770 while (seed_devices) {
771 fs_devices = seed_devices;
772 seed_devices = fs_devices->seed;
773 __btrfs_close_devices(fs_devices);
774 free_fs_devices(fs_devices);
777 * Wait for rcu kworkers under __btrfs_close_devices
778 * to finish all blkdev_puts so device is really
779 * free when umount is done.
785 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
786 fmode_t flags, void *holder)
788 struct request_queue *q;
789 struct block_device *bdev;
790 struct list_head *head = &fs_devices->devices;
791 struct btrfs_device *device;
792 struct btrfs_device *latest_dev = NULL;
793 struct buffer_head *bh;
794 struct btrfs_super_block *disk_super;
801 list_for_each_entry(device, head, dev_list) {
807 /* Just open everything we can; ignore failures here */
808 if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
812 disk_super = (struct btrfs_super_block *)bh->b_data;
813 devid = btrfs_stack_device_id(&disk_super->dev_item);
814 if (devid != device->devid)
817 if (memcmp(device->uuid, disk_super->dev_item.uuid,
821 device->generation = btrfs_super_generation(disk_super);
823 device->generation > latest_dev->generation)
826 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
827 device->writeable = 0;
829 device->writeable = !bdev_read_only(bdev);
833 q = bdev_get_queue(bdev);
834 if (blk_queue_discard(q))
835 device->can_discard = 1;
838 device->in_fs_metadata = 0;
839 device->mode = flags;
841 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
842 fs_devices->rotating = 1;
844 fs_devices->open_devices++;
845 if (device->writeable &&
846 device->devid != BTRFS_DEV_REPLACE_DEVID) {
847 fs_devices->rw_devices++;
848 list_add(&device->dev_alloc_list,
849 &fs_devices->alloc_list);
856 blkdev_put(bdev, flags);
859 if (fs_devices->open_devices == 0) {
863 fs_devices->seeding = seeding;
864 fs_devices->opened = 1;
865 fs_devices->latest_bdev = latest_dev->bdev;
866 fs_devices->total_rw_bytes = 0;
871 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
872 fmode_t flags, void *holder)
876 mutex_lock(&uuid_mutex);
877 if (fs_devices->opened) {
878 fs_devices->opened++;
881 ret = __btrfs_open_devices(fs_devices, flags, holder);
883 mutex_unlock(&uuid_mutex);
888 * Look for a btrfs signature on a device. This may be called out of the mount path
889 * and we are not allowed to call set_blocksize during the scan. The superblock
890 * is read via pagecache
892 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
893 struct btrfs_fs_devices **fs_devices_ret)
895 struct btrfs_super_block *disk_super;
896 struct block_device *bdev;
907 * we would like to check all the supers, but that would make
908 * a btrfs mount succeed after a mkfs from a different FS.
909 * So, we need to add a special mount option to scan for
910 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
912 bytenr = btrfs_sb_offset(0);
914 mutex_lock(&uuid_mutex);
916 bdev = blkdev_get_by_path(path, flags, holder);
923 /* make sure our super fits in the device */
924 if (bytenr + PAGE_CACHE_SIZE >= i_size_read(bdev->bd_inode))
927 /* make sure our super fits in the page */
928 if (sizeof(*disk_super) > PAGE_CACHE_SIZE)
931 /* make sure our super doesn't straddle pages on disk */
932 index = bytenr >> PAGE_CACHE_SHIFT;
933 if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_CACHE_SHIFT != index)
936 /* pull in the page with our super */
937 page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
940 if (IS_ERR_OR_NULL(page))
945 /* align our pointer to the offset of the super block */
946 disk_super = p + (bytenr & ~PAGE_CACHE_MASK);
948 if (btrfs_super_bytenr(disk_super) != bytenr ||
949 btrfs_super_magic(disk_super) != BTRFS_MAGIC)
952 devid = btrfs_stack_device_id(&disk_super->dev_item);
953 transid = btrfs_super_generation(disk_super);
954 total_devices = btrfs_super_num_devices(disk_super);
956 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
958 if (disk_super->label[0]) {
959 if (disk_super->label[BTRFS_LABEL_SIZE - 1])
960 disk_super->label[BTRFS_LABEL_SIZE - 1] = '\0';
961 printk(KERN_INFO "BTRFS: device label %s ", disk_super->label);
963 printk(KERN_INFO "BTRFS: device fsid %pU ", disk_super->fsid);
966 printk(KERN_CONT "devid %llu transid %llu %s\n", devid, transid, path);
969 if (!ret && fs_devices_ret)
970 (*fs_devices_ret)->total_devices = total_devices;
974 page_cache_release(page);
977 blkdev_put(bdev, flags);
979 mutex_unlock(&uuid_mutex);
983 /* helper to account the used device space in the range */
984 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
985 u64 end, u64 *length)
987 struct btrfs_key key;
988 struct btrfs_root *root = device->dev_root;
989 struct btrfs_dev_extent *dev_extent;
990 struct btrfs_path *path;
994 struct extent_buffer *l;
998 if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
1001 path = btrfs_alloc_path();
1006 key.objectid = device->devid;
1008 key.type = BTRFS_DEV_EXTENT_KEY;
1010 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1014 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1021 slot = path->slots[0];
1022 if (slot >= btrfs_header_nritems(l)) {
1023 ret = btrfs_next_leaf(root, path);
1031 btrfs_item_key_to_cpu(l, &key, slot);
1033 if (key.objectid < device->devid)
1036 if (key.objectid > device->devid)
1039 if (key.type != BTRFS_DEV_EXTENT_KEY)
1042 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1043 extent_end = key.offset + btrfs_dev_extent_length(l,
1045 if (key.offset <= start && extent_end > end) {
1046 *length = end - start + 1;
1048 } else if (key.offset <= start && extent_end > start)
1049 *length += extent_end - start;
1050 else if (key.offset > start && extent_end <= end)
1051 *length += extent_end - key.offset;
1052 else if (key.offset > start && key.offset <= end) {
1053 *length += end - key.offset + 1;
1055 } else if (key.offset > end)
1063 btrfs_free_path(path);
1067 static int contains_pending_extent(struct btrfs_trans_handle *trans,
1068 struct btrfs_device *device,
1069 u64 *start, u64 len)
1071 struct extent_map *em;
1074 list_for_each_entry(em, &trans->transaction->pending_chunks, list) {
1075 struct map_lookup *map;
1078 map = (struct map_lookup *)em->bdev;
1079 for (i = 0; i < map->num_stripes; i++) {
1080 if (map->stripes[i].dev != device)
1082 if (map->stripes[i].physical >= *start + len ||
1083 map->stripes[i].physical + em->orig_block_len <=
1086 *start = map->stripes[i].physical +
1097 * find_free_dev_extent - find free space in the specified device
1098 * @device: the device which we search the free space in
1099 * @num_bytes: the size of the free space that we need
1100 * @start: store the start of the free space.
1101 * @len: the size of the free space. that we find, or the size of the max
1102 * free space if we don't find suitable free space
1104 * this uses a pretty simple search, the expectation is that it is
1105 * called very infrequently and that a given device has a small number
1108 * @start is used to store the start of the free space if we find. But if we
1109 * don't find suitable free space, it will be used to store the start position
1110 * of the max free space.
1112 * @len is used to store the size of the free space that we find.
1113 * But if we don't find suitable free space, it is used to store the size of
1114 * the max free space.
1116 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1117 struct btrfs_device *device, u64 num_bytes,
1118 u64 *start, u64 *len)
1120 struct btrfs_key key;
1121 struct btrfs_root *root = device->dev_root;
1122 struct btrfs_dev_extent *dev_extent;
1123 struct btrfs_path *path;
1129 u64 search_end = device->total_bytes;
1132 struct extent_buffer *l;
1134 /* FIXME use last free of some kind */
1136 /* we don't want to overwrite the superblock on the drive,
1137 * so we make sure to start at an offset of at least 1MB
1139 search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
1141 path = btrfs_alloc_path();
1145 max_hole_start = search_start;
1149 if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1155 path->search_commit_root = 1;
1156 path->skip_locking = 1;
1158 key.objectid = device->devid;
1159 key.offset = search_start;
1160 key.type = BTRFS_DEV_EXTENT_KEY;
1162 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1166 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1173 slot = path->slots[0];
1174 if (slot >= btrfs_header_nritems(l)) {
1175 ret = btrfs_next_leaf(root, path);
1183 btrfs_item_key_to_cpu(l, &key, slot);
1185 if (key.objectid < device->devid)
1188 if (key.objectid > device->devid)
1191 if (key.type != BTRFS_DEV_EXTENT_KEY)
1194 if (key.offset > search_start) {
1195 hole_size = key.offset - search_start;
1198 * Have to check before we set max_hole_start, otherwise
1199 * we could end up sending back this offset anyway.
1201 if (contains_pending_extent(trans, device,
1206 if (hole_size > max_hole_size) {
1207 max_hole_start = search_start;
1208 max_hole_size = hole_size;
1212 * If this free space is greater than which we need,
1213 * it must be the max free space that we have found
1214 * until now, so max_hole_start must point to the start
1215 * of this free space and the length of this free space
1216 * is stored in max_hole_size. Thus, we return
1217 * max_hole_start and max_hole_size and go back to the
1220 if (hole_size >= num_bytes) {
1226 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1227 extent_end = key.offset + btrfs_dev_extent_length(l,
1229 if (extent_end > search_start)
1230 search_start = extent_end;
1237 * At this point, search_start should be the end of
1238 * allocated dev extents, and when shrinking the device,
1239 * search_end may be smaller than search_start.
1241 if (search_end > search_start)
1242 hole_size = search_end - search_start;
1244 if (hole_size > max_hole_size) {
1245 max_hole_start = search_start;
1246 max_hole_size = hole_size;
1249 if (contains_pending_extent(trans, device, &search_start, hole_size)) {
1250 btrfs_release_path(path);
1255 if (hole_size < num_bytes)
1261 btrfs_free_path(path);
1262 *start = max_hole_start;
1264 *len = max_hole_size;
1268 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1269 struct btrfs_device *device,
1270 u64 start, u64 *dev_extent_len)
1273 struct btrfs_path *path;
1274 struct btrfs_root *root = device->dev_root;
1275 struct btrfs_key key;
1276 struct btrfs_key found_key;
1277 struct extent_buffer *leaf = NULL;
1278 struct btrfs_dev_extent *extent = NULL;
1280 path = btrfs_alloc_path();
1284 key.objectid = device->devid;
1286 key.type = BTRFS_DEV_EXTENT_KEY;
1288 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1290 ret = btrfs_previous_item(root, path, key.objectid,
1291 BTRFS_DEV_EXTENT_KEY);
1294 leaf = path->nodes[0];
1295 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1296 extent = btrfs_item_ptr(leaf, path->slots[0],
1297 struct btrfs_dev_extent);
1298 BUG_ON(found_key.offset > start || found_key.offset +
1299 btrfs_dev_extent_length(leaf, extent) < start);
1301 btrfs_release_path(path);
1303 } else if (ret == 0) {
1304 leaf = path->nodes[0];
1305 extent = btrfs_item_ptr(leaf, path->slots[0],
1306 struct btrfs_dev_extent);
1308 btrfs_error(root->fs_info, ret, "Slot search failed");
1312 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1314 ret = btrfs_del_item(trans, root, path);
1316 btrfs_error(root->fs_info, ret,
1317 "Failed to remove dev extent item");
1320 btrfs_free_path(path);
1324 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1325 struct btrfs_device *device,
1326 u64 chunk_tree, u64 chunk_objectid,
1327 u64 chunk_offset, u64 start, u64 num_bytes)
1330 struct btrfs_path *path;
1331 struct btrfs_root *root = device->dev_root;
1332 struct btrfs_dev_extent *extent;
1333 struct extent_buffer *leaf;
1334 struct btrfs_key key;
1336 WARN_ON(!device->in_fs_metadata);
1337 WARN_ON(device->is_tgtdev_for_dev_replace);
1338 path = btrfs_alloc_path();
1342 key.objectid = device->devid;
1344 key.type = BTRFS_DEV_EXTENT_KEY;
1345 ret = btrfs_insert_empty_item(trans, root, path, &key,
1350 leaf = path->nodes[0];
1351 extent = btrfs_item_ptr(leaf, path->slots[0],
1352 struct btrfs_dev_extent);
1353 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1354 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1355 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1357 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1358 btrfs_dev_extent_chunk_tree_uuid(extent), BTRFS_UUID_SIZE);
1360 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1361 btrfs_mark_buffer_dirty(leaf);
1363 btrfs_free_path(path);
1367 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1369 struct extent_map_tree *em_tree;
1370 struct extent_map *em;
1374 em_tree = &fs_info->mapping_tree.map_tree;
1375 read_lock(&em_tree->lock);
1376 n = rb_last(&em_tree->map);
1378 em = rb_entry(n, struct extent_map, rb_node);
1379 ret = em->start + em->len;
1381 read_unlock(&em_tree->lock);
1386 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1390 struct btrfs_key key;
1391 struct btrfs_key found_key;
1392 struct btrfs_path *path;
1394 path = btrfs_alloc_path();
1398 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1399 key.type = BTRFS_DEV_ITEM_KEY;
1400 key.offset = (u64)-1;
1402 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1406 BUG_ON(ret == 0); /* Corruption */
1408 ret = btrfs_previous_item(fs_info->chunk_root, path,
1409 BTRFS_DEV_ITEMS_OBJECTID,
1410 BTRFS_DEV_ITEM_KEY);
1414 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1416 *devid_ret = found_key.offset + 1;
1420 btrfs_free_path(path);
1425 * the device information is stored in the chunk root
1426 * the btrfs_device struct should be fully filled in
1428 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1429 struct btrfs_root *root,
1430 struct btrfs_device *device)
1433 struct btrfs_path *path;
1434 struct btrfs_dev_item *dev_item;
1435 struct extent_buffer *leaf;
1436 struct btrfs_key key;
1439 root = root->fs_info->chunk_root;
1441 path = btrfs_alloc_path();
1445 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1446 key.type = BTRFS_DEV_ITEM_KEY;
1447 key.offset = device->devid;
1449 ret = btrfs_insert_empty_item(trans, root, path, &key,
1454 leaf = path->nodes[0];
1455 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1457 btrfs_set_device_id(leaf, dev_item, device->devid);
1458 btrfs_set_device_generation(leaf, dev_item, 0);
1459 btrfs_set_device_type(leaf, dev_item, device->type);
1460 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1461 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1462 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1463 btrfs_set_device_total_bytes(leaf, dev_item,
1464 btrfs_device_get_disk_total_bytes(device));
1465 btrfs_set_device_bytes_used(leaf, dev_item,
1466 btrfs_device_get_bytes_used(device));
1467 btrfs_set_device_group(leaf, dev_item, 0);
1468 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1469 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1470 btrfs_set_device_start_offset(leaf, dev_item, 0);
1472 ptr = btrfs_device_uuid(dev_item);
1473 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1474 ptr = btrfs_device_fsid(dev_item);
1475 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1476 btrfs_mark_buffer_dirty(leaf);
1480 btrfs_free_path(path);
1485 * Function to update ctime/mtime for a given device path.
1486 * Mainly used for ctime/mtime based probe like libblkid.
1488 static void update_dev_time(char *path_name)
1492 filp = filp_open(path_name, O_RDWR, 0);
1495 file_update_time(filp);
1496 filp_close(filp, NULL);
1500 static int btrfs_rm_dev_item(struct btrfs_root *root,
1501 struct btrfs_device *device)
1504 struct btrfs_path *path;
1505 struct btrfs_key key;
1506 struct btrfs_trans_handle *trans;
1508 root = root->fs_info->chunk_root;
1510 path = btrfs_alloc_path();
1514 trans = btrfs_start_transaction(root, 0);
1515 if (IS_ERR(trans)) {
1516 btrfs_free_path(path);
1517 return PTR_ERR(trans);
1519 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1520 key.type = BTRFS_DEV_ITEM_KEY;
1521 key.offset = device->devid;
1523 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1532 ret = btrfs_del_item(trans, root, path);
1536 btrfs_free_path(path);
1537 btrfs_commit_transaction(trans, root);
1541 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1543 struct btrfs_device *device;
1544 struct btrfs_device *next_device;
1545 struct block_device *bdev;
1546 struct buffer_head *bh = NULL;
1547 struct btrfs_super_block *disk_super;
1548 struct btrfs_fs_devices *cur_devices;
1555 bool clear_super = false;
1557 mutex_lock(&uuid_mutex);
1560 seq = read_seqbegin(&root->fs_info->profiles_lock);
1562 all_avail = root->fs_info->avail_data_alloc_bits |
1563 root->fs_info->avail_system_alloc_bits |
1564 root->fs_info->avail_metadata_alloc_bits;
1565 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
1567 num_devices = root->fs_info->fs_devices->num_devices;
1568 btrfs_dev_replace_lock(&root->fs_info->dev_replace);
1569 if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1570 WARN_ON(num_devices < 1);
1573 btrfs_dev_replace_unlock(&root->fs_info->dev_replace);
1575 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && num_devices <= 4) {
1576 ret = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET;
1580 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && num_devices <= 2) {
1581 ret = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET;
1585 if ((all_avail & BTRFS_BLOCK_GROUP_RAID5) &&
1586 root->fs_info->fs_devices->rw_devices <= 2) {
1587 ret = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET;
1590 if ((all_avail & BTRFS_BLOCK_GROUP_RAID6) &&
1591 root->fs_info->fs_devices->rw_devices <= 3) {
1592 ret = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET;
1596 if (strcmp(device_path, "missing") == 0) {
1597 struct list_head *devices;
1598 struct btrfs_device *tmp;
1601 devices = &root->fs_info->fs_devices->devices;
1603 * It is safe to read the devices since the volume_mutex
1606 list_for_each_entry(tmp, devices, dev_list) {
1607 if (tmp->in_fs_metadata &&
1608 !tmp->is_tgtdev_for_dev_replace &&
1618 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
1622 ret = btrfs_get_bdev_and_sb(device_path,
1623 FMODE_WRITE | FMODE_EXCL,
1624 root->fs_info->bdev_holder, 0,
1628 disk_super = (struct btrfs_super_block *)bh->b_data;
1629 devid = btrfs_stack_device_id(&disk_super->dev_item);
1630 dev_uuid = disk_super->dev_item.uuid;
1631 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1639 if (device->is_tgtdev_for_dev_replace) {
1640 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1644 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1645 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1649 if (device->writeable) {
1651 list_del_init(&device->dev_alloc_list);
1652 unlock_chunks(root);
1653 root->fs_info->fs_devices->rw_devices--;
1657 mutex_unlock(&uuid_mutex);
1658 ret = btrfs_shrink_device(device, 0);
1659 mutex_lock(&uuid_mutex);
1664 * TODO: the superblock still includes this device in its num_devices
1665 * counter although write_all_supers() is not locked out. This
1666 * could give a filesystem state which requires a degraded mount.
1668 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1672 device->in_fs_metadata = 0;
1673 btrfs_scrub_cancel_dev(root->fs_info, device);
1676 * the device list mutex makes sure that we don't change
1677 * the device list while someone else is writing out all
1678 * the device supers. Whoever is writing all supers, should
1679 * lock the device list mutex before getting the number of
1680 * devices in the super block (super_copy). Conversely,
1681 * whoever updates the number of devices in the super block
1682 * (super_copy) should hold the device list mutex.
1685 cur_devices = device->fs_devices;
1686 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1687 list_del_rcu(&device->dev_list);
1689 device->fs_devices->num_devices--;
1690 device->fs_devices->total_devices--;
1692 if (device->missing)
1693 device->fs_devices->missing_devices--;
1695 next_device = list_entry(root->fs_info->fs_devices->devices.next,
1696 struct btrfs_device, dev_list);
1697 if (device->bdev == root->fs_info->sb->s_bdev)
1698 root->fs_info->sb->s_bdev = next_device->bdev;
1699 if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1700 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1703 device->fs_devices->open_devices--;
1704 /* remove sysfs entry */
1705 btrfs_kobj_rm_device(root->fs_info, device);
1708 call_rcu(&device->rcu, free_device);
1710 num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1711 btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1712 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1714 if (cur_devices->open_devices == 0) {
1715 struct btrfs_fs_devices *fs_devices;
1716 fs_devices = root->fs_info->fs_devices;
1717 while (fs_devices) {
1718 if (fs_devices->seed == cur_devices) {
1719 fs_devices->seed = cur_devices->seed;
1722 fs_devices = fs_devices->seed;
1724 cur_devices->seed = NULL;
1725 __btrfs_close_devices(cur_devices);
1726 free_fs_devices(cur_devices);
1729 root->fs_info->num_tolerated_disk_barrier_failures =
1730 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1733 * at this point, the device is zero sized. We want to
1734 * remove it from the devices list and zero out the old super
1736 if (clear_super && disk_super) {
1740 /* make sure this device isn't detected as part of
1743 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1744 set_buffer_dirty(bh);
1745 sync_dirty_buffer(bh);
1747 /* clear the mirror copies of super block on the disk
1748 * being removed, 0th copy is been taken care above and
1749 * the below would take of the rest
1751 for (i = 1; i < BTRFS_SUPER_MIRROR_MAX; i++) {
1752 bytenr = btrfs_sb_offset(i);
1753 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
1754 i_size_read(bdev->bd_inode))
1758 bh = __bread(bdev, bytenr / 4096,
1759 BTRFS_SUPER_INFO_SIZE);
1763 disk_super = (struct btrfs_super_block *)bh->b_data;
1765 if (btrfs_super_bytenr(disk_super) != bytenr ||
1766 btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
1769 memset(&disk_super->magic, 0,
1770 sizeof(disk_super->magic));
1771 set_buffer_dirty(bh);
1772 sync_dirty_buffer(bh);
1779 /* Notify udev that device has changed */
1780 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
1782 /* Update ctime/mtime for device path for libblkid */
1783 update_dev_time(device_path);
1789 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1791 mutex_unlock(&uuid_mutex);
1794 if (device->writeable) {
1796 list_add(&device->dev_alloc_list,
1797 &root->fs_info->fs_devices->alloc_list);
1798 unlock_chunks(root);
1799 root->fs_info->fs_devices->rw_devices++;
1804 void btrfs_rm_dev_replace_srcdev(struct btrfs_fs_info *fs_info,
1805 struct btrfs_device *srcdev)
1807 struct btrfs_fs_devices *fs_devices;
1809 WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1812 * in case of fs with no seed, srcdev->fs_devices will point
1813 * to fs_devices of fs_info. However when the dev being replaced is
1814 * a seed dev it will point to the seed's local fs_devices. In short
1815 * srcdev will have its correct fs_devices in both the cases.
1817 fs_devices = srcdev->fs_devices;
1819 list_del_rcu(&srcdev->dev_list);
1820 list_del_rcu(&srcdev->dev_alloc_list);
1821 fs_devices->num_devices--;
1822 if (srcdev->missing) {
1823 fs_devices->missing_devices--;
1824 if (!fs_devices->seeding)
1825 fs_devices->rw_devices++;
1829 fs_devices->open_devices--;
1832 * zero out the old super if it is not writable
1833 * (e.g. seed device)
1835 if (srcdev->writeable)
1836 btrfs_scratch_superblock(srcdev);
1839 call_rcu(&srcdev->rcu, free_device);
1842 * unless fs_devices is seed fs, num_devices shouldn't go
1845 BUG_ON(!fs_devices->num_devices && !fs_devices->seeding);
1847 /* if this is no devs we rather delete the fs_devices */
1848 if (!fs_devices->num_devices) {
1849 struct btrfs_fs_devices *tmp_fs_devices;
1851 tmp_fs_devices = fs_info->fs_devices;
1852 while (tmp_fs_devices) {
1853 if (tmp_fs_devices->seed == fs_devices) {
1854 tmp_fs_devices->seed = fs_devices->seed;
1857 tmp_fs_devices = tmp_fs_devices->seed;
1859 fs_devices->seed = NULL;
1860 __btrfs_close_devices(fs_devices);
1861 free_fs_devices(fs_devices);
1865 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
1866 struct btrfs_device *tgtdev)
1868 struct btrfs_device *next_device;
1871 mutex_lock(&fs_info->fs_devices->device_list_mutex);
1873 btrfs_scratch_superblock(tgtdev);
1874 fs_info->fs_devices->open_devices--;
1876 fs_info->fs_devices->num_devices--;
1878 next_device = list_entry(fs_info->fs_devices->devices.next,
1879 struct btrfs_device, dev_list);
1880 if (tgtdev->bdev == fs_info->sb->s_bdev)
1881 fs_info->sb->s_bdev = next_device->bdev;
1882 if (tgtdev->bdev == fs_info->fs_devices->latest_bdev)
1883 fs_info->fs_devices->latest_bdev = next_device->bdev;
1884 list_del_rcu(&tgtdev->dev_list);
1886 call_rcu(&tgtdev->rcu, free_device);
1888 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1891 static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
1892 struct btrfs_device **device)
1895 struct btrfs_super_block *disk_super;
1898 struct block_device *bdev;
1899 struct buffer_head *bh;
1902 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
1903 root->fs_info->bdev_holder, 0, &bdev, &bh);
1906 disk_super = (struct btrfs_super_block *)bh->b_data;
1907 devid = btrfs_stack_device_id(&disk_super->dev_item);
1908 dev_uuid = disk_super->dev_item.uuid;
1909 *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1914 blkdev_put(bdev, FMODE_READ);
1918 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
1920 struct btrfs_device **device)
1923 if (strcmp(device_path, "missing") == 0) {
1924 struct list_head *devices;
1925 struct btrfs_device *tmp;
1927 devices = &root->fs_info->fs_devices->devices;
1929 * It is safe to read the devices since the volume_mutex
1930 * is held by the caller.
1932 list_for_each_entry(tmp, devices, dev_list) {
1933 if (tmp->in_fs_metadata && !tmp->bdev) {
1940 btrfs_err(root->fs_info, "no missing device found");
1946 return btrfs_find_device_by_path(root, device_path, device);
1951 * does all the dirty work required for changing file system's UUID.
1953 static int btrfs_prepare_sprout(struct btrfs_root *root)
1955 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1956 struct btrfs_fs_devices *old_devices;
1957 struct btrfs_fs_devices *seed_devices;
1958 struct btrfs_super_block *disk_super = root->fs_info->super_copy;
1959 struct btrfs_device *device;
1962 BUG_ON(!mutex_is_locked(&uuid_mutex));
1963 if (!fs_devices->seeding)
1966 seed_devices = __alloc_fs_devices();
1967 if (IS_ERR(seed_devices))
1968 return PTR_ERR(seed_devices);
1970 old_devices = clone_fs_devices(fs_devices);
1971 if (IS_ERR(old_devices)) {
1972 kfree(seed_devices);
1973 return PTR_ERR(old_devices);
1976 list_add(&old_devices->list, &fs_uuids);
1978 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1979 seed_devices->opened = 1;
1980 INIT_LIST_HEAD(&seed_devices->devices);
1981 INIT_LIST_HEAD(&seed_devices->alloc_list);
1982 mutex_init(&seed_devices->device_list_mutex);
1984 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1985 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1987 list_for_each_entry(device, &seed_devices->devices, dev_list)
1988 device->fs_devices = seed_devices;
1991 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1992 unlock_chunks(root);
1994 fs_devices->seeding = 0;
1995 fs_devices->num_devices = 0;
1996 fs_devices->open_devices = 0;
1997 fs_devices->missing_devices = 0;
1998 fs_devices->rotating = 0;
1999 fs_devices->seed = seed_devices;
2001 generate_random_uuid(fs_devices->fsid);
2002 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2003 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2004 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2006 super_flags = btrfs_super_flags(disk_super) &
2007 ~BTRFS_SUPER_FLAG_SEEDING;
2008 btrfs_set_super_flags(disk_super, super_flags);
2014 * strore the expected generation for seed devices in device items.
2016 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2017 struct btrfs_root *root)
2019 struct btrfs_path *path;
2020 struct extent_buffer *leaf;
2021 struct btrfs_dev_item *dev_item;
2022 struct btrfs_device *device;
2023 struct btrfs_key key;
2024 u8 fs_uuid[BTRFS_UUID_SIZE];
2025 u8 dev_uuid[BTRFS_UUID_SIZE];
2029 path = btrfs_alloc_path();
2033 root = root->fs_info->chunk_root;
2034 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2036 key.type = BTRFS_DEV_ITEM_KEY;
2039 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2043 leaf = path->nodes[0];
2045 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2046 ret = btrfs_next_leaf(root, path);
2051 leaf = path->nodes[0];
2052 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2053 btrfs_release_path(path);
2057 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2058 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2059 key.type != BTRFS_DEV_ITEM_KEY)
2062 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2063 struct btrfs_dev_item);
2064 devid = btrfs_device_id(leaf, dev_item);
2065 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2067 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2069 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2071 BUG_ON(!device); /* Logic error */
2073 if (device->fs_devices->seeding) {
2074 btrfs_set_device_generation(leaf, dev_item,
2075 device->generation);
2076 btrfs_mark_buffer_dirty(leaf);
2084 btrfs_free_path(path);
2088 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
2090 struct request_queue *q;
2091 struct btrfs_trans_handle *trans;
2092 struct btrfs_device *device;
2093 struct block_device *bdev;
2094 struct list_head *devices;
2095 struct super_block *sb = root->fs_info->sb;
2096 struct rcu_string *name;
2098 int seeding_dev = 0;
2101 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
2104 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2105 root->fs_info->bdev_holder);
2107 return PTR_ERR(bdev);
2109 if (root->fs_info->fs_devices->seeding) {
2111 down_write(&sb->s_umount);
2112 mutex_lock(&uuid_mutex);
2115 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2117 devices = &root->fs_info->fs_devices->devices;
2119 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2120 list_for_each_entry(device, devices, dev_list) {
2121 if (device->bdev == bdev) {
2124 &root->fs_info->fs_devices->device_list_mutex);
2128 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2130 device = btrfs_alloc_device(root->fs_info, NULL, NULL);
2131 if (IS_ERR(device)) {
2132 /* we can safely leave the fs_devices entry around */
2133 ret = PTR_ERR(device);
2137 name = rcu_string_strdup(device_path, GFP_NOFS);
2143 rcu_assign_pointer(device->name, name);
2145 trans = btrfs_start_transaction(root, 0);
2146 if (IS_ERR(trans)) {
2147 rcu_string_free(device->name);
2149 ret = PTR_ERR(trans);
2153 q = bdev_get_queue(bdev);
2154 if (blk_queue_discard(q))
2155 device->can_discard = 1;
2156 device->writeable = 1;
2157 device->generation = trans->transid;
2158 device->io_width = root->sectorsize;
2159 device->io_align = root->sectorsize;
2160 device->sector_size = root->sectorsize;
2161 device->total_bytes = i_size_read(bdev->bd_inode);
2162 device->disk_total_bytes = device->total_bytes;
2163 device->commit_total_bytes = device->total_bytes;
2164 device->dev_root = root->fs_info->dev_root;
2165 device->bdev = bdev;
2166 device->in_fs_metadata = 1;
2167 device->is_tgtdev_for_dev_replace = 0;
2168 device->mode = FMODE_EXCL;
2169 device->dev_stats_valid = 1;
2170 set_blocksize(device->bdev, 4096);
2173 sb->s_flags &= ~MS_RDONLY;
2174 ret = btrfs_prepare_sprout(root);
2175 BUG_ON(ret); /* -ENOMEM */
2178 device->fs_devices = root->fs_info->fs_devices;
2180 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2182 list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
2183 list_add(&device->dev_alloc_list,
2184 &root->fs_info->fs_devices->alloc_list);
2185 root->fs_info->fs_devices->num_devices++;
2186 root->fs_info->fs_devices->open_devices++;
2187 root->fs_info->fs_devices->rw_devices++;
2188 root->fs_info->fs_devices->total_devices++;
2189 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2191 spin_lock(&root->fs_info->free_chunk_lock);
2192 root->fs_info->free_chunk_space += device->total_bytes;
2193 spin_unlock(&root->fs_info->free_chunk_lock);
2195 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2196 root->fs_info->fs_devices->rotating = 1;
2198 tmp = btrfs_super_total_bytes(root->fs_info->super_copy);
2199 btrfs_set_super_total_bytes(root->fs_info->super_copy,
2200 tmp + device->total_bytes);
2202 tmp = btrfs_super_num_devices(root->fs_info->super_copy);
2203 btrfs_set_super_num_devices(root->fs_info->super_copy,
2206 /* add sysfs device entry */
2207 btrfs_kobj_add_device(root->fs_info, device);
2210 * we've got more storage, clear any full flags on the space
2213 btrfs_clear_space_info_full(root->fs_info);
2215 unlock_chunks(root);
2216 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2220 ret = init_first_rw_device(trans, root, device);
2221 unlock_chunks(root);
2223 btrfs_abort_transaction(trans, root, ret);
2228 ret = btrfs_add_device(trans, root, device);
2230 btrfs_abort_transaction(trans, root, ret);
2235 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2237 ret = btrfs_finish_sprout(trans, root);
2239 btrfs_abort_transaction(trans, root, ret);
2243 /* Sprouting would change fsid of the mounted root,
2244 * so rename the fsid on the sysfs
2246 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2247 root->fs_info->fsid);
2248 if (kobject_rename(&root->fs_info->super_kobj, fsid_buf))
2252 root->fs_info->num_tolerated_disk_barrier_failures =
2253 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
2254 ret = btrfs_commit_transaction(trans, root);
2257 mutex_unlock(&uuid_mutex);
2258 up_write(&sb->s_umount);
2260 if (ret) /* transaction commit */
2263 ret = btrfs_relocate_sys_chunks(root);
2265 btrfs_error(root->fs_info, ret,
2266 "Failed to relocate sys chunks after "
2267 "device initialization. This can be fixed "
2268 "using the \"btrfs balance\" command.");
2269 trans = btrfs_attach_transaction(root);
2270 if (IS_ERR(trans)) {
2271 if (PTR_ERR(trans) == -ENOENT)
2273 return PTR_ERR(trans);
2275 ret = btrfs_commit_transaction(trans, root);
2278 /* Update ctime/mtime for libblkid */
2279 update_dev_time(device_path);
2283 btrfs_end_transaction(trans, root);
2284 rcu_string_free(device->name);
2285 btrfs_kobj_rm_device(root->fs_info, device);
2288 blkdev_put(bdev, FMODE_EXCL);
2290 mutex_unlock(&uuid_mutex);
2291 up_write(&sb->s_umount);
2296 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2297 struct btrfs_device *srcdev,
2298 struct btrfs_device **device_out)
2300 struct request_queue *q;
2301 struct btrfs_device *device;
2302 struct block_device *bdev;
2303 struct btrfs_fs_info *fs_info = root->fs_info;
2304 struct list_head *devices;
2305 struct rcu_string *name;
2306 u64 devid = BTRFS_DEV_REPLACE_DEVID;
2310 if (fs_info->fs_devices->seeding) {
2311 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
2315 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2316 fs_info->bdev_holder);
2318 btrfs_err(fs_info, "target device %s is invalid!", device_path);
2319 return PTR_ERR(bdev);
2322 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2324 devices = &fs_info->fs_devices->devices;
2325 list_for_each_entry(device, devices, dev_list) {
2326 if (device->bdev == bdev) {
2327 btrfs_err(fs_info, "target device is in the filesystem!");
2334 if (i_size_read(bdev->bd_inode) <
2335 btrfs_device_get_total_bytes(srcdev)) {
2336 btrfs_err(fs_info, "target device is smaller than source device!");
2342 device = btrfs_alloc_device(NULL, &devid, NULL);
2343 if (IS_ERR(device)) {
2344 ret = PTR_ERR(device);
2348 name = rcu_string_strdup(device_path, GFP_NOFS);
2354 rcu_assign_pointer(device->name, name);
2356 q = bdev_get_queue(bdev);
2357 if (blk_queue_discard(q))
2358 device->can_discard = 1;
2359 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2360 device->writeable = 1;
2361 device->generation = 0;
2362 device->io_width = root->sectorsize;
2363 device->io_align = root->sectorsize;
2364 device->sector_size = root->sectorsize;
2365 device->total_bytes = btrfs_device_get_total_bytes(srcdev);
2366 device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
2367 device->bytes_used = btrfs_device_get_bytes_used(srcdev);
2368 ASSERT(list_empty(&srcdev->resized_list));
2369 device->commit_total_bytes = srcdev->commit_total_bytes;
2370 device->commit_bytes_used = device->bytes_used;
2371 device->dev_root = fs_info->dev_root;
2372 device->bdev = bdev;
2373 device->in_fs_metadata = 1;
2374 device->is_tgtdev_for_dev_replace = 1;
2375 device->mode = FMODE_EXCL;
2376 device->dev_stats_valid = 1;
2377 set_blocksize(device->bdev, 4096);
2378 device->fs_devices = fs_info->fs_devices;
2379 list_add(&device->dev_list, &fs_info->fs_devices->devices);
2380 fs_info->fs_devices->num_devices++;
2381 fs_info->fs_devices->open_devices++;
2382 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2384 *device_out = device;
2388 blkdev_put(bdev, FMODE_EXCL);
2392 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2393 struct btrfs_device *tgtdev)
2395 WARN_ON(fs_info->fs_devices->rw_devices == 0);
2396 tgtdev->io_width = fs_info->dev_root->sectorsize;
2397 tgtdev->io_align = fs_info->dev_root->sectorsize;
2398 tgtdev->sector_size = fs_info->dev_root->sectorsize;
2399 tgtdev->dev_root = fs_info->dev_root;
2400 tgtdev->in_fs_metadata = 1;
2403 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2404 struct btrfs_device *device)
2407 struct btrfs_path *path;
2408 struct btrfs_root *root;
2409 struct btrfs_dev_item *dev_item;
2410 struct extent_buffer *leaf;
2411 struct btrfs_key key;
2413 root = device->dev_root->fs_info->chunk_root;
2415 path = btrfs_alloc_path();
2419 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2420 key.type = BTRFS_DEV_ITEM_KEY;
2421 key.offset = device->devid;
2423 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2432 leaf = path->nodes[0];
2433 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2435 btrfs_set_device_id(leaf, dev_item, device->devid);
2436 btrfs_set_device_type(leaf, dev_item, device->type);
2437 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2438 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2439 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2440 btrfs_set_device_total_bytes(leaf, dev_item,
2441 btrfs_device_get_disk_total_bytes(device));
2442 btrfs_set_device_bytes_used(leaf, dev_item,
2443 btrfs_device_get_bytes_used(device));
2444 btrfs_mark_buffer_dirty(leaf);
2447 btrfs_free_path(path);
2451 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2452 struct btrfs_device *device, u64 new_size)
2454 struct btrfs_super_block *super_copy =
2455 device->dev_root->fs_info->super_copy;
2456 struct btrfs_fs_devices *fs_devices;
2460 if (!device->writeable)
2463 lock_chunks(device->dev_root);
2464 old_total = btrfs_super_total_bytes(super_copy);
2465 diff = new_size - device->total_bytes;
2467 if (new_size <= device->total_bytes ||
2468 device->is_tgtdev_for_dev_replace) {
2469 unlock_chunks(device->dev_root);
2473 fs_devices = device->dev_root->fs_info->fs_devices;
2475 btrfs_set_super_total_bytes(super_copy, old_total + diff);
2476 device->fs_devices->total_rw_bytes += diff;
2478 btrfs_device_set_total_bytes(device, new_size);
2479 btrfs_device_set_disk_total_bytes(device, new_size);
2480 btrfs_clear_space_info_full(device->dev_root->fs_info);
2481 if (list_empty(&device->resized_list))
2482 list_add_tail(&device->resized_list,
2483 &fs_devices->resized_devices);
2484 unlock_chunks(device->dev_root);
2486 return btrfs_update_device(trans, device);
2489 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2490 struct btrfs_root *root,
2491 u64 chunk_tree, u64 chunk_objectid,
2495 struct btrfs_path *path;
2496 struct btrfs_key key;
2498 root = root->fs_info->chunk_root;
2499 path = btrfs_alloc_path();
2503 key.objectid = chunk_objectid;
2504 key.offset = chunk_offset;
2505 key.type = BTRFS_CHUNK_ITEM_KEY;
2507 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2510 else if (ret > 0) { /* Logic error or corruption */
2511 btrfs_error(root->fs_info, -ENOENT,
2512 "Failed lookup while freeing chunk.");
2517 ret = btrfs_del_item(trans, root, path);
2519 btrfs_error(root->fs_info, ret,
2520 "Failed to delete chunk item.");
2522 btrfs_free_path(path);
2526 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2529 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2530 struct btrfs_disk_key *disk_key;
2531 struct btrfs_chunk *chunk;
2538 struct btrfs_key key;
2541 array_size = btrfs_super_sys_array_size(super_copy);
2543 ptr = super_copy->sys_chunk_array;
2546 while (cur < array_size) {
2547 disk_key = (struct btrfs_disk_key *)ptr;
2548 btrfs_disk_key_to_cpu(&key, disk_key);
2550 len = sizeof(*disk_key);
2552 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2553 chunk = (struct btrfs_chunk *)(ptr + len);
2554 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2555 len += btrfs_chunk_item_size(num_stripes);
2560 if (key.objectid == chunk_objectid &&
2561 key.offset == chunk_offset) {
2562 memmove(ptr, ptr + len, array_size - (cur + len));
2564 btrfs_set_super_sys_array_size(super_copy, array_size);
2570 unlock_chunks(root);
2574 static int btrfs_relocate_chunk(struct btrfs_root *root,
2575 u64 chunk_tree, u64 chunk_objectid,
2578 struct extent_map_tree *em_tree;
2579 struct btrfs_root *extent_root;
2580 struct btrfs_trans_handle *trans;
2581 struct btrfs_device *device;
2582 struct extent_map *em;
2583 struct map_lookup *map;
2584 u64 dev_extent_len = 0;
2588 root = root->fs_info->chunk_root;
2589 extent_root = root->fs_info->extent_root;
2590 em_tree = &root->fs_info->mapping_tree.map_tree;
2592 ret = btrfs_can_relocate(extent_root, chunk_offset);
2596 /* step one, relocate all the extents inside this chunk */
2597 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2601 trans = btrfs_start_transaction(root, 0);
2602 if (IS_ERR(trans)) {
2603 ret = PTR_ERR(trans);
2604 btrfs_std_error(root->fs_info, ret);
2609 * step two, delete the device extents and the
2610 * chunk tree entries
2612 read_lock(&em_tree->lock);
2613 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2614 read_unlock(&em_tree->lock);
2616 BUG_ON(!em || em->start > chunk_offset ||
2617 em->start + em->len < chunk_offset);
2618 map = (struct map_lookup *)em->bdev;
2620 for (i = 0; i < map->num_stripes; i++) {
2621 device = map->stripes[i].dev;
2622 ret = btrfs_free_dev_extent(trans, device,
2623 map->stripes[i].physical,
2627 if (device->bytes_used > 0) {
2629 btrfs_device_set_bytes_used(device,
2630 device->bytes_used - dev_extent_len);
2631 spin_lock(&root->fs_info->free_chunk_lock);
2632 root->fs_info->free_chunk_space += dev_extent_len;
2633 spin_unlock(&root->fs_info->free_chunk_lock);
2634 btrfs_clear_space_info_full(root->fs_info);
2635 unlock_chunks(root);
2638 if (map->stripes[i].dev) {
2639 ret = btrfs_update_device(trans, map->stripes[i].dev);
2643 ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
2648 trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2650 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2651 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2655 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
2658 write_lock(&em_tree->lock);
2659 remove_extent_mapping(em_tree, em);
2660 write_unlock(&em_tree->lock);
2662 /* once for the tree */
2663 free_extent_map(em);
2665 free_extent_map(em);
2667 btrfs_end_transaction(trans, root);
2671 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2673 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2674 struct btrfs_path *path;
2675 struct extent_buffer *leaf;
2676 struct btrfs_chunk *chunk;
2677 struct btrfs_key key;
2678 struct btrfs_key found_key;
2679 u64 chunk_tree = chunk_root->root_key.objectid;
2681 bool retried = false;
2685 path = btrfs_alloc_path();
2690 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2691 key.offset = (u64)-1;
2692 key.type = BTRFS_CHUNK_ITEM_KEY;
2695 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2698 BUG_ON(ret == 0); /* Corruption */
2700 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2707 leaf = path->nodes[0];
2708 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2710 chunk = btrfs_item_ptr(leaf, path->slots[0],
2711 struct btrfs_chunk);
2712 chunk_type = btrfs_chunk_type(leaf, chunk);
2713 btrfs_release_path(path);
2715 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2716 ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
2725 if (found_key.offset == 0)
2727 key.offset = found_key.offset - 1;
2730 if (failed && !retried) {
2734 } else if (WARN_ON(failed && retried)) {
2738 btrfs_free_path(path);
2742 static int insert_balance_item(struct btrfs_root *root,
2743 struct btrfs_balance_control *bctl)
2745 struct btrfs_trans_handle *trans;
2746 struct btrfs_balance_item *item;
2747 struct btrfs_disk_balance_args disk_bargs;
2748 struct btrfs_path *path;
2749 struct extent_buffer *leaf;
2750 struct btrfs_key key;
2753 path = btrfs_alloc_path();
2757 trans = btrfs_start_transaction(root, 0);
2758 if (IS_ERR(trans)) {
2759 btrfs_free_path(path);
2760 return PTR_ERR(trans);
2763 key.objectid = BTRFS_BALANCE_OBJECTID;
2764 key.type = BTRFS_BALANCE_ITEM_KEY;
2767 ret = btrfs_insert_empty_item(trans, root, path, &key,
2772 leaf = path->nodes[0];
2773 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2775 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2777 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2778 btrfs_set_balance_data(leaf, item, &disk_bargs);
2779 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2780 btrfs_set_balance_meta(leaf, item, &disk_bargs);
2781 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2782 btrfs_set_balance_sys(leaf, item, &disk_bargs);
2784 btrfs_set_balance_flags(leaf, item, bctl->flags);
2786 btrfs_mark_buffer_dirty(leaf);
2788 btrfs_free_path(path);
2789 err = btrfs_commit_transaction(trans, root);
2795 static int del_balance_item(struct btrfs_root *root)
2797 struct btrfs_trans_handle *trans;
2798 struct btrfs_path *path;
2799 struct btrfs_key key;
2802 path = btrfs_alloc_path();
2806 trans = btrfs_start_transaction(root, 0);
2807 if (IS_ERR(trans)) {
2808 btrfs_free_path(path);
2809 return PTR_ERR(trans);
2812 key.objectid = BTRFS_BALANCE_OBJECTID;
2813 key.type = BTRFS_BALANCE_ITEM_KEY;
2816 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2824 ret = btrfs_del_item(trans, root, path);
2826 btrfs_free_path(path);
2827 err = btrfs_commit_transaction(trans, root);
2834 * This is a heuristic used to reduce the number of chunks balanced on
2835 * resume after balance was interrupted.
2837 static void update_balance_args(struct btrfs_balance_control *bctl)
2840 * Turn on soft mode for chunk types that were being converted.
2842 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
2843 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
2844 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
2845 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
2846 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
2847 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
2850 * Turn on usage filter if is not already used. The idea is
2851 * that chunks that we have already balanced should be
2852 * reasonably full. Don't do it for chunks that are being
2853 * converted - that will keep us from relocating unconverted
2854 * (albeit full) chunks.
2856 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2857 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2858 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
2859 bctl->data.usage = 90;
2861 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2862 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2863 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
2864 bctl->sys.usage = 90;
2866 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2867 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2868 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
2869 bctl->meta.usage = 90;
2874 * Should be called with both balance and volume mutexes held to
2875 * serialize other volume operations (add_dev/rm_dev/resize) with
2876 * restriper. Same goes for unset_balance_control.
2878 static void set_balance_control(struct btrfs_balance_control *bctl)
2880 struct btrfs_fs_info *fs_info = bctl->fs_info;
2882 BUG_ON(fs_info->balance_ctl);
2884 spin_lock(&fs_info->balance_lock);
2885 fs_info->balance_ctl = bctl;
2886 spin_unlock(&fs_info->balance_lock);
2889 static void unset_balance_control(struct btrfs_fs_info *fs_info)
2891 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2893 BUG_ON(!fs_info->balance_ctl);
2895 spin_lock(&fs_info->balance_lock);
2896 fs_info->balance_ctl = NULL;
2897 spin_unlock(&fs_info->balance_lock);
2903 * Balance filters. Return 1 if chunk should be filtered out
2904 * (should not be balanced).
2906 static int chunk_profiles_filter(u64 chunk_type,
2907 struct btrfs_balance_args *bargs)
2909 chunk_type = chunk_to_extended(chunk_type) &
2910 BTRFS_EXTENDED_PROFILE_MASK;
2912 if (bargs->profiles & chunk_type)
2918 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
2919 struct btrfs_balance_args *bargs)
2921 struct btrfs_block_group_cache *cache;
2922 u64 chunk_used, user_thresh;
2925 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
2926 chunk_used = btrfs_block_group_used(&cache->item);
2928 if (bargs->usage == 0)
2930 else if (bargs->usage > 100)
2931 user_thresh = cache->key.offset;
2933 user_thresh = div_factor_fine(cache->key.offset,
2936 if (chunk_used < user_thresh)
2939 btrfs_put_block_group(cache);
2943 static int chunk_devid_filter(struct extent_buffer *leaf,
2944 struct btrfs_chunk *chunk,
2945 struct btrfs_balance_args *bargs)
2947 struct btrfs_stripe *stripe;
2948 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2951 for (i = 0; i < num_stripes; i++) {
2952 stripe = btrfs_stripe_nr(chunk, i);
2953 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
2960 /* [pstart, pend) */
2961 static int chunk_drange_filter(struct extent_buffer *leaf,
2962 struct btrfs_chunk *chunk,
2964 struct btrfs_balance_args *bargs)
2966 struct btrfs_stripe *stripe;
2967 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2973 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
2976 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
2977 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
2978 factor = num_stripes / 2;
2979 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
2980 factor = num_stripes - 1;
2981 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
2982 factor = num_stripes - 2;
2984 factor = num_stripes;
2987 for (i = 0; i < num_stripes; i++) {
2988 stripe = btrfs_stripe_nr(chunk, i);
2989 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
2992 stripe_offset = btrfs_stripe_offset(leaf, stripe);
2993 stripe_length = btrfs_chunk_length(leaf, chunk);
2994 do_div(stripe_length, factor);
2996 if (stripe_offset < bargs->pend &&
2997 stripe_offset + stripe_length > bargs->pstart)
3004 /* [vstart, vend) */
3005 static int chunk_vrange_filter(struct extent_buffer *leaf,
3006 struct btrfs_chunk *chunk,
3008 struct btrfs_balance_args *bargs)
3010 if (chunk_offset < bargs->vend &&
3011 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3012 /* at least part of the chunk is inside this vrange */
3018 static int chunk_soft_convert_filter(u64 chunk_type,
3019 struct btrfs_balance_args *bargs)
3021 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3024 chunk_type = chunk_to_extended(chunk_type) &
3025 BTRFS_EXTENDED_PROFILE_MASK;
3027 if (bargs->target == chunk_type)
3033 static int should_balance_chunk(struct btrfs_root *root,
3034 struct extent_buffer *leaf,
3035 struct btrfs_chunk *chunk, u64 chunk_offset)
3037 struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
3038 struct btrfs_balance_args *bargs = NULL;
3039 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3042 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3043 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3047 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3048 bargs = &bctl->data;
3049 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3051 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3052 bargs = &bctl->meta;
3054 /* profiles filter */
3055 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3056 chunk_profiles_filter(chunk_type, bargs)) {
3061 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3062 chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
3067 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3068 chunk_devid_filter(leaf, chunk, bargs)) {
3072 /* drange filter, makes sense only with devid filter */
3073 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3074 chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
3079 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3080 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3084 /* soft profile changing mode */
3085 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3086 chunk_soft_convert_filter(chunk_type, bargs)) {
3091 * limited by count, must be the last filter
3093 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3094 if (bargs->limit == 0)
3103 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3105 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3106 struct btrfs_root *chunk_root = fs_info->chunk_root;
3107 struct btrfs_root *dev_root = fs_info->dev_root;
3108 struct list_head *devices;
3109 struct btrfs_device *device;
3112 struct btrfs_chunk *chunk;
3113 struct btrfs_path *path;
3114 struct btrfs_key key;
3115 struct btrfs_key found_key;
3116 struct btrfs_trans_handle *trans;
3117 struct extent_buffer *leaf;
3120 int enospc_errors = 0;
3121 bool counting = true;
3122 u64 limit_data = bctl->data.limit;
3123 u64 limit_meta = bctl->meta.limit;
3124 u64 limit_sys = bctl->sys.limit;
3126 /* step one make some room on all the devices */
3127 devices = &fs_info->fs_devices->devices;
3128 list_for_each_entry(device, devices, dev_list) {
3129 old_size = btrfs_device_get_total_bytes(device);
3130 size_to_free = div_factor(old_size, 1);
3131 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
3132 if (!device->writeable ||
3133 btrfs_device_get_total_bytes(device) -
3134 btrfs_device_get_bytes_used(device) > size_to_free ||
3135 device->is_tgtdev_for_dev_replace)
3138 ret = btrfs_shrink_device(device, old_size - size_to_free);
3143 trans = btrfs_start_transaction(dev_root, 0);
3144 BUG_ON(IS_ERR(trans));
3146 ret = btrfs_grow_device(trans, device, old_size);
3149 btrfs_end_transaction(trans, dev_root);
3152 /* step two, relocate all the chunks */
3153 path = btrfs_alloc_path();
3159 /* zero out stat counters */
3160 spin_lock(&fs_info->balance_lock);
3161 memset(&bctl->stat, 0, sizeof(bctl->stat));
3162 spin_unlock(&fs_info->balance_lock);
3165 bctl->data.limit = limit_data;
3166 bctl->meta.limit = limit_meta;
3167 bctl->sys.limit = limit_sys;
3169 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3170 key.offset = (u64)-1;
3171 key.type = BTRFS_CHUNK_ITEM_KEY;
3174 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3175 atomic_read(&fs_info->balance_cancel_req)) {
3180 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3185 * this shouldn't happen, it means the last relocate
3189 BUG(); /* FIXME break ? */
3191 ret = btrfs_previous_item(chunk_root, path, 0,
3192 BTRFS_CHUNK_ITEM_KEY);
3198 leaf = path->nodes[0];
3199 slot = path->slots[0];
3200 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3202 if (found_key.objectid != key.objectid)
3205 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3208 spin_lock(&fs_info->balance_lock);
3209 bctl->stat.considered++;
3210 spin_unlock(&fs_info->balance_lock);
3213 ret = should_balance_chunk(chunk_root, leaf, chunk,
3215 btrfs_release_path(path);
3220 spin_lock(&fs_info->balance_lock);
3221 bctl->stat.expected++;
3222 spin_unlock(&fs_info->balance_lock);
3226 ret = btrfs_relocate_chunk(chunk_root,
3227 chunk_root->root_key.objectid,
3230 if (ret && ret != -ENOSPC)
3232 if (ret == -ENOSPC) {
3235 spin_lock(&fs_info->balance_lock);
3236 bctl->stat.completed++;
3237 spin_unlock(&fs_info->balance_lock);
3240 if (found_key.offset == 0)
3242 key.offset = found_key.offset - 1;
3246 btrfs_release_path(path);
3251 btrfs_free_path(path);
3252 if (enospc_errors) {
3253 btrfs_info(fs_info, "%d enospc errors during balance",
3263 * alloc_profile_is_valid - see if a given profile is valid and reduced
3264 * @flags: profile to validate
3265 * @extended: if true @flags is treated as an extended profile
3267 static int alloc_profile_is_valid(u64 flags, int extended)
3269 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3270 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3272 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3274 /* 1) check that all other bits are zeroed */
3278 /* 2) see if profile is reduced */
3280 return !extended; /* "0" is valid for usual profiles */
3282 /* true if exactly one bit set */
3283 return (flags & (flags - 1)) == 0;
3286 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3288 /* cancel requested || normal exit path */
3289 return atomic_read(&fs_info->balance_cancel_req) ||
3290 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3291 atomic_read(&fs_info->balance_cancel_req) == 0);
3294 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3298 unset_balance_control(fs_info);
3299 ret = del_balance_item(fs_info->tree_root);
3301 btrfs_std_error(fs_info, ret);
3303 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3307 * Should be called with both balance and volume mutexes held
3309 int btrfs_balance(struct btrfs_balance_control *bctl,
3310 struct btrfs_ioctl_balance_args *bargs)
3312 struct btrfs_fs_info *fs_info = bctl->fs_info;
3319 if (btrfs_fs_closing(fs_info) ||
3320 atomic_read(&fs_info->balance_pause_req) ||
3321 atomic_read(&fs_info->balance_cancel_req)) {
3326 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3327 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3331 * In case of mixed groups both data and meta should be picked,
3332 * and identical options should be given for both of them.
3334 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3335 if (mixed && (bctl->flags & allowed)) {
3336 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3337 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3338 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3339 btrfs_err(fs_info, "with mixed groups data and "
3340 "metadata balance options must be the same");
3346 num_devices = fs_info->fs_devices->num_devices;
3347 btrfs_dev_replace_lock(&fs_info->dev_replace);
3348 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3349 BUG_ON(num_devices < 1);
3352 btrfs_dev_replace_unlock(&fs_info->dev_replace);
3353 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3354 if (num_devices == 1)
3355 allowed |= BTRFS_BLOCK_GROUP_DUP;
3356 else if (num_devices > 1)
3357 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3358 if (num_devices > 2)
3359 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3360 if (num_devices > 3)
3361 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3362 BTRFS_BLOCK_GROUP_RAID6);
3363 if ((bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3364 (!alloc_profile_is_valid(bctl->data.target, 1) ||
3365 (bctl->data.target & ~allowed))) {
3366 btrfs_err(fs_info, "unable to start balance with target "
3367 "data profile %llu",
3372 if ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3373 (!alloc_profile_is_valid(bctl->meta.target, 1) ||
3374 (bctl->meta.target & ~allowed))) {
3376 "unable to start balance with target metadata profile %llu",
3381 if ((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3382 (!alloc_profile_is_valid(bctl->sys.target, 1) ||
3383 (bctl->sys.target & ~allowed))) {
3385 "unable to start balance with target system profile %llu",
3391 /* allow dup'ed data chunks only in mixed mode */
3392 if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3393 (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) {
3394 btrfs_err(fs_info, "dup for data is not allowed");
3399 /* allow to reduce meta or sys integrity only if force set */
3400 allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3401 BTRFS_BLOCK_GROUP_RAID10 |
3402 BTRFS_BLOCK_GROUP_RAID5 |
3403 BTRFS_BLOCK_GROUP_RAID6;
3405 seq = read_seqbegin(&fs_info->profiles_lock);
3407 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3408 (fs_info->avail_system_alloc_bits & allowed) &&
3409 !(bctl->sys.target & allowed)) ||
3410 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3411 (fs_info->avail_metadata_alloc_bits & allowed) &&
3412 !(bctl->meta.target & allowed))) {
3413 if (bctl->flags & BTRFS_BALANCE_FORCE) {
3414 btrfs_info(fs_info, "force reducing metadata integrity");
3416 btrfs_err(fs_info, "balance will reduce metadata "
3417 "integrity, use force if you want this");
3422 } while (read_seqretry(&fs_info->profiles_lock, seq));
3424 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3425 int num_tolerated_disk_barrier_failures;
3426 u64 target = bctl->sys.target;
3428 num_tolerated_disk_barrier_failures =
3429 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3430 if (num_tolerated_disk_barrier_failures > 0 &&
3432 (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3433 BTRFS_AVAIL_ALLOC_BIT_SINGLE)))
3434 num_tolerated_disk_barrier_failures = 0;
3435 else if (num_tolerated_disk_barrier_failures > 1 &&
3437 (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)))
3438 num_tolerated_disk_barrier_failures = 1;
3440 fs_info->num_tolerated_disk_barrier_failures =
3441 num_tolerated_disk_barrier_failures;
3444 ret = insert_balance_item(fs_info->tree_root, bctl);
3445 if (ret && ret != -EEXIST)
3448 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3449 BUG_ON(ret == -EEXIST);
3450 set_balance_control(bctl);
3452 BUG_ON(ret != -EEXIST);
3453 spin_lock(&fs_info->balance_lock);
3454 update_balance_args(bctl);
3455 spin_unlock(&fs_info->balance_lock);
3458 atomic_inc(&fs_info->balance_running);
3459 mutex_unlock(&fs_info->balance_mutex);
3461 ret = __btrfs_balance(fs_info);
3463 mutex_lock(&fs_info->balance_mutex);
3464 atomic_dec(&fs_info->balance_running);
3466 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3467 fs_info->num_tolerated_disk_barrier_failures =
3468 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3472 memset(bargs, 0, sizeof(*bargs));
3473 update_ioctl_balance_args(fs_info, 0, bargs);
3476 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3477 balance_need_close(fs_info)) {
3478 __cancel_balance(fs_info);
3481 wake_up(&fs_info->balance_wait_q);
3485 if (bctl->flags & BTRFS_BALANCE_RESUME)
3486 __cancel_balance(fs_info);
3489 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3494 static int balance_kthread(void *data)
3496 struct btrfs_fs_info *fs_info = data;
3499 mutex_lock(&fs_info->volume_mutex);
3500 mutex_lock(&fs_info->balance_mutex);
3502 if (fs_info->balance_ctl) {
3503 btrfs_info(fs_info, "continuing balance");
3504 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3507 mutex_unlock(&fs_info->balance_mutex);
3508 mutex_unlock(&fs_info->volume_mutex);
3513 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3515 struct task_struct *tsk;
3517 spin_lock(&fs_info->balance_lock);
3518 if (!fs_info->balance_ctl) {
3519 spin_unlock(&fs_info->balance_lock);
3522 spin_unlock(&fs_info->balance_lock);
3524 if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
3525 btrfs_info(fs_info, "force skipping balance");
3529 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3530 return PTR_ERR_OR_ZERO(tsk);
3533 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3535 struct btrfs_balance_control *bctl;
3536 struct btrfs_balance_item *item;
3537 struct btrfs_disk_balance_args disk_bargs;
3538 struct btrfs_path *path;
3539 struct extent_buffer *leaf;
3540 struct btrfs_key key;
3543 path = btrfs_alloc_path();
3547 key.objectid = BTRFS_BALANCE_OBJECTID;
3548 key.type = BTRFS_BALANCE_ITEM_KEY;
3551 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3554 if (ret > 0) { /* ret = -ENOENT; */
3559 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3565 leaf = path->nodes[0];
3566 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3568 bctl->fs_info = fs_info;
3569 bctl->flags = btrfs_balance_flags(leaf, item);
3570 bctl->flags |= BTRFS_BALANCE_RESUME;
3572 btrfs_balance_data(leaf, item, &disk_bargs);
3573 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3574 btrfs_balance_meta(leaf, item, &disk_bargs);
3575 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3576 btrfs_balance_sys(leaf, item, &disk_bargs);
3577 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3579 WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
3581 mutex_lock(&fs_info->volume_mutex);
3582 mutex_lock(&fs_info->balance_mutex);
3584 set_balance_control(bctl);
3586 mutex_unlock(&fs_info->balance_mutex);
3587 mutex_unlock(&fs_info->volume_mutex);
3589 btrfs_free_path(path);
3593 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3597 mutex_lock(&fs_info->balance_mutex);
3598 if (!fs_info->balance_ctl) {
3599 mutex_unlock(&fs_info->balance_mutex);
3603 if (atomic_read(&fs_info->balance_running)) {
3604 atomic_inc(&fs_info->balance_pause_req);
3605 mutex_unlock(&fs_info->balance_mutex);
3607 wait_event(fs_info->balance_wait_q,
3608 atomic_read(&fs_info->balance_running) == 0);
3610 mutex_lock(&fs_info->balance_mutex);
3611 /* we are good with balance_ctl ripped off from under us */
3612 BUG_ON(atomic_read(&fs_info->balance_running));
3613 atomic_dec(&fs_info->balance_pause_req);
3618 mutex_unlock(&fs_info->balance_mutex);
3622 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3624 if (fs_info->sb->s_flags & MS_RDONLY)
3627 mutex_lock(&fs_info->balance_mutex);
3628 if (!fs_info->balance_ctl) {
3629 mutex_unlock(&fs_info->balance_mutex);
3633 atomic_inc(&fs_info->balance_cancel_req);
3635 * if we are running just wait and return, balance item is
3636 * deleted in btrfs_balance in this case
3638 if (atomic_read(&fs_info->balance_running)) {
3639 mutex_unlock(&fs_info->balance_mutex);
3640 wait_event(fs_info->balance_wait_q,
3641 atomic_read(&fs_info->balance_running) == 0);
3642 mutex_lock(&fs_info->balance_mutex);
3644 /* __cancel_balance needs volume_mutex */
3645 mutex_unlock(&fs_info->balance_mutex);
3646 mutex_lock(&fs_info->volume_mutex);
3647 mutex_lock(&fs_info->balance_mutex);
3649 if (fs_info->balance_ctl)
3650 __cancel_balance(fs_info);
3652 mutex_unlock(&fs_info->volume_mutex);
3655 BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
3656 atomic_dec(&fs_info->balance_cancel_req);
3657 mutex_unlock(&fs_info->balance_mutex);
3661 static int btrfs_uuid_scan_kthread(void *data)
3663 struct btrfs_fs_info *fs_info = data;
3664 struct btrfs_root *root = fs_info->tree_root;
3665 struct btrfs_key key;
3666 struct btrfs_key max_key;
3667 struct btrfs_path *path = NULL;
3669 struct extent_buffer *eb;
3671 struct btrfs_root_item root_item;
3673 struct btrfs_trans_handle *trans = NULL;
3675 path = btrfs_alloc_path();
3682 key.type = BTRFS_ROOT_ITEM_KEY;
3685 max_key.objectid = (u64)-1;
3686 max_key.type = BTRFS_ROOT_ITEM_KEY;
3687 max_key.offset = (u64)-1;
3690 ret = btrfs_search_forward(root, &key, path, 0);
3697 if (key.type != BTRFS_ROOT_ITEM_KEY ||
3698 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
3699 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
3700 key.objectid > BTRFS_LAST_FREE_OBJECTID)
3703 eb = path->nodes[0];
3704 slot = path->slots[0];
3705 item_size = btrfs_item_size_nr(eb, slot);
3706 if (item_size < sizeof(root_item))
3709 read_extent_buffer(eb, &root_item,
3710 btrfs_item_ptr_offset(eb, slot),
3711 (int)sizeof(root_item));
3712 if (btrfs_root_refs(&root_item) == 0)
3715 if (!btrfs_is_empty_uuid(root_item.uuid) ||
3716 !btrfs_is_empty_uuid(root_item.received_uuid)) {
3720 btrfs_release_path(path);
3722 * 1 - subvol uuid item
3723 * 1 - received_subvol uuid item
3725 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
3726 if (IS_ERR(trans)) {
3727 ret = PTR_ERR(trans);
3735 if (!btrfs_is_empty_uuid(root_item.uuid)) {
3736 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
3738 BTRFS_UUID_KEY_SUBVOL,
3741 btrfs_warn(fs_info, "uuid_tree_add failed %d",
3747 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
3748 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
3749 root_item.received_uuid,
3750 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
3753 btrfs_warn(fs_info, "uuid_tree_add failed %d",
3761 ret = btrfs_end_transaction(trans, fs_info->uuid_root);
3767 btrfs_release_path(path);
3768 if (key.offset < (u64)-1) {
3770 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
3772 key.type = BTRFS_ROOT_ITEM_KEY;
3773 } else if (key.objectid < (u64)-1) {
3775 key.type = BTRFS_ROOT_ITEM_KEY;
3784 btrfs_free_path(path);
3785 if (trans && !IS_ERR(trans))
3786 btrfs_end_transaction(trans, fs_info->uuid_root);
3788 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
3790 fs_info->update_uuid_tree_gen = 1;
3791 up(&fs_info->uuid_tree_rescan_sem);
3796 * Callback for btrfs_uuid_tree_iterate().
3798 * 0 check succeeded, the entry is not outdated.
3799 * < 0 if an error occured.
3800 * > 0 if the check failed, which means the caller shall remove the entry.
3802 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
3803 u8 *uuid, u8 type, u64 subid)
3805 struct btrfs_key key;
3807 struct btrfs_root *subvol_root;
3809 if (type != BTRFS_UUID_KEY_SUBVOL &&
3810 type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
3813 key.objectid = subid;
3814 key.type = BTRFS_ROOT_ITEM_KEY;
3815 key.offset = (u64)-1;
3816 subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
3817 if (IS_ERR(subvol_root)) {
3818 ret = PTR_ERR(subvol_root);
3825 case BTRFS_UUID_KEY_SUBVOL:
3826 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
3829 case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
3830 if (memcmp(uuid, subvol_root->root_item.received_uuid,
3840 static int btrfs_uuid_rescan_kthread(void *data)
3842 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
3846 * 1st step is to iterate through the existing UUID tree and
3847 * to delete all entries that contain outdated data.
3848 * 2nd step is to add all missing entries to the UUID tree.
3850 ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
3852 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
3853 up(&fs_info->uuid_tree_rescan_sem);
3856 return btrfs_uuid_scan_kthread(data);
3859 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
3861 struct btrfs_trans_handle *trans;
3862 struct btrfs_root *tree_root = fs_info->tree_root;
3863 struct btrfs_root *uuid_root;
3864 struct task_struct *task;
3871 trans = btrfs_start_transaction(tree_root, 2);
3873 return PTR_ERR(trans);
3875 uuid_root = btrfs_create_tree(trans, fs_info,
3876 BTRFS_UUID_TREE_OBJECTID);
3877 if (IS_ERR(uuid_root)) {
3878 btrfs_abort_transaction(trans, tree_root,
3879 PTR_ERR(uuid_root));
3880 return PTR_ERR(uuid_root);
3883 fs_info->uuid_root = uuid_root;
3885 ret = btrfs_commit_transaction(trans, tree_root);
3889 down(&fs_info->uuid_tree_rescan_sem);
3890 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
3892 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3893 btrfs_warn(fs_info, "failed to start uuid_scan task");
3894 up(&fs_info->uuid_tree_rescan_sem);
3895 return PTR_ERR(task);
3901 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
3903 struct task_struct *task;
3905 down(&fs_info->uuid_tree_rescan_sem);
3906 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
3908 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3909 btrfs_warn(fs_info, "failed to start uuid_rescan task");
3910 up(&fs_info->uuid_tree_rescan_sem);
3911 return PTR_ERR(task);
3918 * shrinking a device means finding all of the device extents past
3919 * the new size, and then following the back refs to the chunks.
3920 * The chunk relocation code actually frees the device extent
3922 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
3924 struct btrfs_trans_handle *trans;
3925 struct btrfs_root *root = device->dev_root;
3926 struct btrfs_dev_extent *dev_extent = NULL;
3927 struct btrfs_path *path;
3935 bool retried = false;
3936 struct extent_buffer *l;
3937 struct btrfs_key key;
3938 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3939 u64 old_total = btrfs_super_total_bytes(super_copy);
3940 u64 old_size = btrfs_device_get_total_bytes(device);
3941 u64 diff = old_size - new_size;
3943 if (device->is_tgtdev_for_dev_replace)
3946 path = btrfs_alloc_path();
3954 btrfs_device_set_total_bytes(device, new_size);
3955 if (device->writeable) {
3956 device->fs_devices->total_rw_bytes -= diff;
3957 spin_lock(&root->fs_info->free_chunk_lock);
3958 root->fs_info->free_chunk_space -= diff;
3959 spin_unlock(&root->fs_info->free_chunk_lock);
3961 unlock_chunks(root);
3964 key.objectid = device->devid;
3965 key.offset = (u64)-1;
3966 key.type = BTRFS_DEV_EXTENT_KEY;
3969 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3973 ret = btrfs_previous_item(root, path, 0, key.type);
3978 btrfs_release_path(path);
3983 slot = path->slots[0];
3984 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
3986 if (key.objectid != device->devid) {
3987 btrfs_release_path(path);
3991 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
3992 length = btrfs_dev_extent_length(l, dev_extent);
3994 if (key.offset + length <= new_size) {
3995 btrfs_release_path(path);
3999 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
4000 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
4001 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4002 btrfs_release_path(path);
4004 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
4006 if (ret && ret != -ENOSPC)
4010 } while (key.offset-- > 0);
4012 if (failed && !retried) {
4016 } else if (failed && retried) {
4020 btrfs_device_set_total_bytes(device, old_size);
4021 if (device->writeable)
4022 device->fs_devices->total_rw_bytes += diff;
4023 spin_lock(&root->fs_info->free_chunk_lock);
4024 root->fs_info->free_chunk_space += diff;
4025 spin_unlock(&root->fs_info->free_chunk_lock);
4026 unlock_chunks(root);
4030 /* Shrinking succeeded, else we would be at "done". */
4031 trans = btrfs_start_transaction(root, 0);
4032 if (IS_ERR(trans)) {
4033 ret = PTR_ERR(trans);
4038 btrfs_device_set_disk_total_bytes(device, new_size);
4039 if (list_empty(&device->resized_list))
4040 list_add_tail(&device->resized_list,
4041 &root->fs_info->fs_devices->resized_devices);
4043 WARN_ON(diff > old_total);
4044 btrfs_set_super_total_bytes(super_copy, old_total - diff);
4045 unlock_chunks(root);
4047 /* Now btrfs_update_device() will change the on-disk size. */
4048 ret = btrfs_update_device(trans, device);
4049 btrfs_end_transaction(trans, root);
4051 btrfs_free_path(path);
4055 static int btrfs_add_system_chunk(struct btrfs_root *root,
4056 struct btrfs_key *key,
4057 struct btrfs_chunk *chunk, int item_size)
4059 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4060 struct btrfs_disk_key disk_key;
4065 array_size = btrfs_super_sys_array_size(super_copy);
4066 if (array_size + item_size + sizeof(disk_key)
4067 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4068 unlock_chunks(root);
4072 ptr = super_copy->sys_chunk_array + array_size;
4073 btrfs_cpu_key_to_disk(&disk_key, key);
4074 memcpy(ptr, &disk_key, sizeof(disk_key));
4075 ptr += sizeof(disk_key);
4076 memcpy(ptr, chunk, item_size);
4077 item_size += sizeof(disk_key);
4078 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4079 unlock_chunks(root);
4085 * sort the devices in descending order by max_avail, total_avail
4087 static int btrfs_cmp_device_info(const void *a, const void *b)
4089 const struct btrfs_device_info *di_a = a;
4090 const struct btrfs_device_info *di_b = b;
4092 if (di_a->max_avail > di_b->max_avail)
4094 if (di_a->max_avail < di_b->max_avail)
4096 if (di_a->total_avail > di_b->total_avail)
4098 if (di_a->total_avail < di_b->total_avail)
4103 static struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
4104 [BTRFS_RAID_RAID10] = {
4107 .devs_max = 0, /* 0 == as many as possible */
4109 .devs_increment = 2,
4112 [BTRFS_RAID_RAID1] = {
4117 .devs_increment = 2,
4120 [BTRFS_RAID_DUP] = {
4125 .devs_increment = 1,
4128 [BTRFS_RAID_RAID0] = {
4133 .devs_increment = 1,
4136 [BTRFS_RAID_SINGLE] = {
4141 .devs_increment = 1,
4144 [BTRFS_RAID_RAID5] = {
4149 .devs_increment = 1,
4152 [BTRFS_RAID_RAID6] = {
4157 .devs_increment = 1,
4162 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
4164 /* TODO allow them to set a preferred stripe size */
4168 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4170 if (!(type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)))
4173 btrfs_set_fs_incompat(info, RAID56);
4176 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r) \
4177 - sizeof(struct btrfs_item) \
4178 - sizeof(struct btrfs_chunk)) \
4179 / sizeof(struct btrfs_stripe) + 1)
4181 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
4182 - 2 * sizeof(struct btrfs_disk_key) \
4183 - 2 * sizeof(struct btrfs_chunk)) \
4184 / sizeof(struct btrfs_stripe) + 1)
4186 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4187 struct btrfs_root *extent_root, u64 start,
4190 struct btrfs_fs_info *info = extent_root->fs_info;
4191 struct btrfs_fs_devices *fs_devices = info->fs_devices;
4192 struct list_head *cur;
4193 struct map_lookup *map = NULL;
4194 struct extent_map_tree *em_tree;
4195 struct extent_map *em;
4196 struct btrfs_device_info *devices_info = NULL;
4198 int num_stripes; /* total number of stripes to allocate */
4199 int data_stripes; /* number of stripes that count for
4201 int sub_stripes; /* sub_stripes info for map */
4202 int dev_stripes; /* stripes per dev */
4203 int devs_max; /* max devs to use */
4204 int devs_min; /* min devs needed */
4205 int devs_increment; /* ndevs has to be a multiple of this */
4206 int ncopies; /* how many copies to data has */
4208 u64 max_stripe_size;
4212 u64 raid_stripe_len = BTRFS_STRIPE_LEN;
4218 BUG_ON(!alloc_profile_is_valid(type, 0));
4220 if (list_empty(&fs_devices->alloc_list))
4223 index = __get_raid_index(type);
4225 sub_stripes = btrfs_raid_array[index].sub_stripes;
4226 dev_stripes = btrfs_raid_array[index].dev_stripes;
4227 devs_max = btrfs_raid_array[index].devs_max;
4228 devs_min = btrfs_raid_array[index].devs_min;
4229 devs_increment = btrfs_raid_array[index].devs_increment;
4230 ncopies = btrfs_raid_array[index].ncopies;
4232 if (type & BTRFS_BLOCK_GROUP_DATA) {
4233 max_stripe_size = 1024 * 1024 * 1024;
4234 max_chunk_size = 10 * max_stripe_size;
4236 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4237 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4238 /* for larger filesystems, use larger metadata chunks */
4239 if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024)
4240 max_stripe_size = 1024 * 1024 * 1024;
4242 max_stripe_size = 256 * 1024 * 1024;
4243 max_chunk_size = max_stripe_size;
4245 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4246 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4247 max_stripe_size = 32 * 1024 * 1024;
4248 max_chunk_size = 2 * max_stripe_size;
4250 devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4252 btrfs_err(info, "invalid chunk type 0x%llx requested",
4257 /* we don't want a chunk larger than 10% of writeable space */
4258 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4261 devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
4266 cur = fs_devices->alloc_list.next;
4269 * in the first pass through the devices list, we gather information
4270 * about the available holes on each device.
4273 while (cur != &fs_devices->alloc_list) {
4274 struct btrfs_device *device;
4278 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4282 if (!device->writeable) {
4284 "BTRFS: read-only device in alloc_list\n");
4288 if (!device->in_fs_metadata ||
4289 device->is_tgtdev_for_dev_replace)
4292 if (device->total_bytes > device->bytes_used)
4293 total_avail = device->total_bytes - device->bytes_used;
4297 /* If there is no space on this device, skip it. */
4298 if (total_avail == 0)
4301 ret = find_free_dev_extent(trans, device,
4302 max_stripe_size * dev_stripes,
4303 &dev_offset, &max_avail);
4304 if (ret && ret != -ENOSPC)
4308 max_avail = max_stripe_size * dev_stripes;
4310 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4313 if (ndevs == fs_devices->rw_devices) {
4314 WARN(1, "%s: found more than %llu devices\n",
4315 __func__, fs_devices->rw_devices);
4318 devices_info[ndevs].dev_offset = dev_offset;
4319 devices_info[ndevs].max_avail = max_avail;
4320 devices_info[ndevs].total_avail = total_avail;
4321 devices_info[ndevs].dev = device;
4326 * now sort the devices by hole size / available space
4328 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4329 btrfs_cmp_device_info, NULL);
4331 /* round down to number of usable stripes */
4332 ndevs -= ndevs % devs_increment;
4334 if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4339 if (devs_max && ndevs > devs_max)
4342 * the primary goal is to maximize the number of stripes, so use as many
4343 * devices as possible, even if the stripes are not maximum sized.
4345 stripe_size = devices_info[ndevs-1].max_avail;
4346 num_stripes = ndevs * dev_stripes;
4349 * this will have to be fixed for RAID1 and RAID10 over
4352 data_stripes = num_stripes / ncopies;
4354 if (type & BTRFS_BLOCK_GROUP_RAID5) {
4355 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
4356 btrfs_super_stripesize(info->super_copy));
4357 data_stripes = num_stripes - 1;
4359 if (type & BTRFS_BLOCK_GROUP_RAID6) {
4360 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
4361 btrfs_super_stripesize(info->super_copy));
4362 data_stripes = num_stripes - 2;
4366 * Use the number of data stripes to figure out how big this chunk
4367 * is really going to be in terms of logical address space,
4368 * and compare that answer with the max chunk size
4370 if (stripe_size * data_stripes > max_chunk_size) {
4371 u64 mask = (1ULL << 24) - 1;
4372 stripe_size = max_chunk_size;
4373 do_div(stripe_size, data_stripes);
4375 /* bump the answer up to a 16MB boundary */
4376 stripe_size = (stripe_size + mask) & ~mask;
4378 /* but don't go higher than the limits we found
4379 * while searching for free extents
4381 if (stripe_size > devices_info[ndevs-1].max_avail)
4382 stripe_size = devices_info[ndevs-1].max_avail;
4385 do_div(stripe_size, dev_stripes);
4387 /* align to BTRFS_STRIPE_LEN */
4388 do_div(stripe_size, raid_stripe_len);
4389 stripe_size *= raid_stripe_len;
4391 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4396 map->num_stripes = num_stripes;
4398 for (i = 0; i < ndevs; ++i) {
4399 for (j = 0; j < dev_stripes; ++j) {
4400 int s = i * dev_stripes + j;
4401 map->stripes[s].dev = devices_info[i].dev;
4402 map->stripes[s].physical = devices_info[i].dev_offset +
4406 map->sector_size = extent_root->sectorsize;
4407 map->stripe_len = raid_stripe_len;
4408 map->io_align = raid_stripe_len;
4409 map->io_width = raid_stripe_len;
4411 map->sub_stripes = sub_stripes;
4413 num_bytes = stripe_size * data_stripes;
4415 trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
4417 em = alloc_extent_map();
4423 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4424 em->bdev = (struct block_device *)map;
4426 em->len = num_bytes;
4427 em->block_start = 0;
4428 em->block_len = em->len;
4429 em->orig_block_len = stripe_size;
4431 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4432 write_lock(&em_tree->lock);
4433 ret = add_extent_mapping(em_tree, em, 0);
4435 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4436 atomic_inc(&em->refs);
4438 write_unlock(&em_tree->lock);
4440 free_extent_map(em);
4444 ret = btrfs_make_block_group(trans, extent_root, 0, type,
4445 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4448 goto error_del_extent;
4450 for (i = 0; i < map->num_stripes; i++) {
4451 num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4452 btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4455 spin_lock(&extent_root->fs_info->free_chunk_lock);
4456 extent_root->fs_info->free_chunk_space -= (stripe_size *
4458 spin_unlock(&extent_root->fs_info->free_chunk_lock);
4460 free_extent_map(em);
4461 check_raid56_incompat_flag(extent_root->fs_info, type);
4463 kfree(devices_info);
4467 write_lock(&em_tree->lock);
4468 remove_extent_mapping(em_tree, em);
4469 write_unlock(&em_tree->lock);
4471 /* One for our allocation */
4472 free_extent_map(em);
4473 /* One for the tree reference */
4474 free_extent_map(em);
4476 kfree(devices_info);
4480 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4481 struct btrfs_root *extent_root,
4482 u64 chunk_offset, u64 chunk_size)
4484 struct btrfs_key key;
4485 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4486 struct btrfs_device *device;
4487 struct btrfs_chunk *chunk;
4488 struct btrfs_stripe *stripe;
4489 struct extent_map_tree *em_tree;
4490 struct extent_map *em;
4491 struct map_lookup *map;
4498 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4499 read_lock(&em_tree->lock);
4500 em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size);
4501 read_unlock(&em_tree->lock);
4504 btrfs_crit(extent_root->fs_info, "unable to find logical "
4505 "%Lu len %Lu", chunk_offset, chunk_size);
4509 if (em->start != chunk_offset || em->len != chunk_size) {
4510 btrfs_crit(extent_root->fs_info, "found a bad mapping, wanted"
4511 " %Lu-%Lu, found %Lu-%Lu", chunk_offset,
4512 chunk_size, em->start, em->len);
4513 free_extent_map(em);
4517 map = (struct map_lookup *)em->bdev;
4518 item_size = btrfs_chunk_item_size(map->num_stripes);
4519 stripe_size = em->orig_block_len;
4521 chunk = kzalloc(item_size, GFP_NOFS);
4527 for (i = 0; i < map->num_stripes; i++) {
4528 device = map->stripes[i].dev;
4529 dev_offset = map->stripes[i].physical;
4531 ret = btrfs_update_device(trans, device);
4534 ret = btrfs_alloc_dev_extent(trans, device,
4535 chunk_root->root_key.objectid,
4536 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4537 chunk_offset, dev_offset,
4543 stripe = &chunk->stripe;
4544 for (i = 0; i < map->num_stripes; i++) {
4545 device = map->stripes[i].dev;
4546 dev_offset = map->stripes[i].physical;
4548 btrfs_set_stack_stripe_devid(stripe, device->devid);
4549 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4550 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4554 btrfs_set_stack_chunk_length(chunk, chunk_size);
4555 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4556 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4557 btrfs_set_stack_chunk_type(chunk, map->type);
4558 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4559 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4560 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4561 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4562 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4564 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4565 key.type = BTRFS_CHUNK_ITEM_KEY;
4566 key.offset = chunk_offset;
4568 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4569 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4571 * TODO: Cleanup of inserted chunk root in case of
4574 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
4580 free_extent_map(em);
4585 * Chunk allocation falls into two parts. The first part does works
4586 * that make the new allocated chunk useable, but not do any operation
4587 * that modifies the chunk tree. The second part does the works that
4588 * require modifying the chunk tree. This division is important for the
4589 * bootstrap process of adding storage to a seed btrfs.
4591 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4592 struct btrfs_root *extent_root, u64 type)
4596 chunk_offset = find_next_chunk(extent_root->fs_info);
4597 return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type);
4600 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4601 struct btrfs_root *root,
4602 struct btrfs_device *device)
4605 u64 sys_chunk_offset;
4607 struct btrfs_fs_info *fs_info = root->fs_info;
4608 struct btrfs_root *extent_root = fs_info->extent_root;
4611 chunk_offset = find_next_chunk(fs_info);
4612 alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
4613 ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset,
4618 sys_chunk_offset = find_next_chunk(root->fs_info);
4619 alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
4620 ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset,
4625 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
4629 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
4630 BTRFS_BLOCK_GROUP_RAID10 |
4631 BTRFS_BLOCK_GROUP_RAID5 |
4632 BTRFS_BLOCK_GROUP_DUP)) {
4634 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
4643 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
4645 struct extent_map *em;
4646 struct map_lookup *map;
4647 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4652 read_lock(&map_tree->map_tree.lock);
4653 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
4654 read_unlock(&map_tree->map_tree.lock);
4658 map = (struct map_lookup *)em->bdev;
4659 for (i = 0; i < map->num_stripes; i++) {
4660 if (map->stripes[i].dev->missing) {
4665 if (!map->stripes[i].dev->writeable) {
4672 * If the number of missing devices is larger than max errors,
4673 * we can not write the data into that chunk successfully, so
4676 if (miss_ndevs > btrfs_chunk_max_errors(map))
4679 free_extent_map(em);
4683 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
4685 extent_map_tree_init(&tree->map_tree);
4688 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
4690 struct extent_map *em;
4693 write_lock(&tree->map_tree.lock);
4694 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
4696 remove_extent_mapping(&tree->map_tree, em);
4697 write_unlock(&tree->map_tree.lock);
4701 free_extent_map(em);
4702 /* once for the tree */
4703 free_extent_map(em);
4707 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
4709 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4710 struct extent_map *em;
4711 struct map_lookup *map;
4712 struct extent_map_tree *em_tree = &map_tree->map_tree;
4715 read_lock(&em_tree->lock);
4716 em = lookup_extent_mapping(em_tree, logical, len);
4717 read_unlock(&em_tree->lock);
4720 * We could return errors for these cases, but that could get ugly and
4721 * we'd probably do the same thing which is just not do anything else
4722 * and exit, so return 1 so the callers don't try to use other copies.
4725 btrfs_crit(fs_info, "No mapping for %Lu-%Lu", logical,
4730 if (em->start > logical || em->start + em->len < logical) {
4731 btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got "
4732 "%Lu-%Lu", logical, logical+len, em->start,
4733 em->start + em->len);
4734 free_extent_map(em);
4738 map = (struct map_lookup *)em->bdev;
4739 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
4740 ret = map->num_stripes;
4741 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4742 ret = map->sub_stripes;
4743 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
4745 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
4749 free_extent_map(em);
4751 btrfs_dev_replace_lock(&fs_info->dev_replace);
4752 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
4754 btrfs_dev_replace_unlock(&fs_info->dev_replace);
4759 unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
4760 struct btrfs_mapping_tree *map_tree,
4763 struct extent_map *em;
4764 struct map_lookup *map;
4765 struct extent_map_tree *em_tree = &map_tree->map_tree;
4766 unsigned long len = root->sectorsize;
4768 read_lock(&em_tree->lock);
4769 em = lookup_extent_mapping(em_tree, logical, len);
4770 read_unlock(&em_tree->lock);
4773 BUG_ON(em->start > logical || em->start + em->len < logical);
4774 map = (struct map_lookup *)em->bdev;
4775 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4776 BTRFS_BLOCK_GROUP_RAID6)) {
4777 len = map->stripe_len * nr_data_stripes(map);
4779 free_extent_map(em);
4783 int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
4784 u64 logical, u64 len, int mirror_num)
4786 struct extent_map *em;
4787 struct map_lookup *map;
4788 struct extent_map_tree *em_tree = &map_tree->map_tree;
4791 read_lock(&em_tree->lock);
4792 em = lookup_extent_mapping(em_tree, logical, len);
4793 read_unlock(&em_tree->lock);
4796 BUG_ON(em->start > logical || em->start + em->len < logical);
4797 map = (struct map_lookup *)em->bdev;
4798 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4799 BTRFS_BLOCK_GROUP_RAID6))
4801 free_extent_map(em);
4805 static int find_live_mirror(struct btrfs_fs_info *fs_info,
4806 struct map_lookup *map, int first, int num,
4807 int optimal, int dev_replace_is_ongoing)
4811 struct btrfs_device *srcdev;
4813 if (dev_replace_is_ongoing &&
4814 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
4815 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
4816 srcdev = fs_info->dev_replace.srcdev;
4821 * try to avoid the drive that is the source drive for a
4822 * dev-replace procedure, only choose it if no other non-missing
4823 * mirror is available
4825 for (tolerance = 0; tolerance < 2; tolerance++) {
4826 if (map->stripes[optimal].dev->bdev &&
4827 (tolerance || map->stripes[optimal].dev != srcdev))
4829 for (i = first; i < first + num; i++) {
4830 if (map->stripes[i].dev->bdev &&
4831 (tolerance || map->stripes[i].dev != srcdev))
4836 /* we couldn't find one that doesn't fail. Just return something
4837 * and the io error handling code will clean up eventually
4842 static inline int parity_smaller(u64 a, u64 b)
4847 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
4848 static void sort_parity_stripes(struct btrfs_bio *bbio, u64 *raid_map)
4850 struct btrfs_bio_stripe s;
4857 for (i = 0; i < bbio->num_stripes - 1; i++) {
4858 if (parity_smaller(raid_map[i], raid_map[i+1])) {
4859 s = bbio->stripes[i];
4861 bbio->stripes[i] = bbio->stripes[i+1];
4862 raid_map[i] = raid_map[i+1];
4863 bbio->stripes[i+1] = s;
4871 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
4872 u64 logical, u64 *length,
4873 struct btrfs_bio **bbio_ret,
4874 int mirror_num, u64 **raid_map_ret)
4876 struct extent_map *em;
4877 struct map_lookup *map;
4878 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4879 struct extent_map_tree *em_tree = &map_tree->map_tree;
4882 u64 stripe_end_offset;
4887 u64 *raid_map = NULL;
4893 struct btrfs_bio *bbio = NULL;
4894 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
4895 int dev_replace_is_ongoing = 0;
4896 int num_alloc_stripes;
4897 int patch_the_first_stripe_for_dev_replace = 0;
4898 u64 physical_to_patch_in_first_stripe = 0;
4899 u64 raid56_full_stripe_start = (u64)-1;
4901 read_lock(&em_tree->lock);
4902 em = lookup_extent_mapping(em_tree, logical, *length);
4903 read_unlock(&em_tree->lock);
4906 btrfs_crit(fs_info, "unable to find logical %llu len %llu",
4911 if (em->start > logical || em->start + em->len < logical) {
4912 btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, "
4913 "found %Lu-%Lu", logical, em->start,
4914 em->start + em->len);
4915 free_extent_map(em);
4919 map = (struct map_lookup *)em->bdev;
4920 offset = logical - em->start;
4922 stripe_len = map->stripe_len;
4925 * stripe_nr counts the total number of stripes we have to stride
4926 * to get to this block
4928 do_div(stripe_nr, stripe_len);
4930 stripe_offset = stripe_nr * stripe_len;
4931 BUG_ON(offset < stripe_offset);
4933 /* stripe_offset is the offset of this block in its stripe*/
4934 stripe_offset = offset - stripe_offset;
4936 /* if we're here for raid56, we need to know the stripe aligned start */
4937 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
4938 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
4939 raid56_full_stripe_start = offset;
4941 /* allow a write of a full stripe, but make sure we don't
4942 * allow straddling of stripes
4944 do_div(raid56_full_stripe_start, full_stripe_len);
4945 raid56_full_stripe_start *= full_stripe_len;
4948 if (rw & REQ_DISCARD) {
4949 /* we don't discard raid56 yet */
4951 (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
4955 *length = min_t(u64, em->len - offset, *length);
4956 } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
4958 /* For writes to RAID[56], allow a full stripeset across all disks.
4959 For other RAID types and for RAID[56] reads, just allow a single
4960 stripe (on a single disk). */
4961 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6) &&
4963 max_len = stripe_len * nr_data_stripes(map) -
4964 (offset - raid56_full_stripe_start);
4966 /* we limit the length of each bio to what fits in a stripe */
4967 max_len = stripe_len - stripe_offset;
4969 *length = min_t(u64, em->len - offset, max_len);
4971 *length = em->len - offset;
4974 /* This is for when we're called from btrfs_merge_bio_hook() and all
4975 it cares about is the length */
4979 btrfs_dev_replace_lock(dev_replace);
4980 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
4981 if (!dev_replace_is_ongoing)
4982 btrfs_dev_replace_unlock(dev_replace);
4984 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
4985 !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) &&
4986 dev_replace->tgtdev != NULL) {
4988 * in dev-replace case, for repair case (that's the only
4989 * case where the mirror is selected explicitly when
4990 * calling btrfs_map_block), blocks left of the left cursor
4991 * can also be read from the target drive.
4992 * For REQ_GET_READ_MIRRORS, the target drive is added as
4993 * the last one to the array of stripes. For READ, it also
4994 * needs to be supported using the same mirror number.
4995 * If the requested block is not left of the left cursor,
4996 * EIO is returned. This can happen because btrfs_num_copies()
4997 * returns one more in the dev-replace case.
4999 u64 tmp_length = *length;
5000 struct btrfs_bio *tmp_bbio = NULL;
5001 int tmp_num_stripes;
5002 u64 srcdev_devid = dev_replace->srcdev->devid;
5003 int index_srcdev = 0;
5005 u64 physical_of_found = 0;
5007 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
5008 logical, &tmp_length, &tmp_bbio, 0, NULL);
5010 WARN_ON(tmp_bbio != NULL);
5014 tmp_num_stripes = tmp_bbio->num_stripes;
5015 if (mirror_num > tmp_num_stripes) {
5017 * REQ_GET_READ_MIRRORS does not contain this
5018 * mirror, that means that the requested area
5019 * is not left of the left cursor
5027 * process the rest of the function using the mirror_num
5028 * of the source drive. Therefore look it up first.
5029 * At the end, patch the device pointer to the one of the
5032 for (i = 0; i < tmp_num_stripes; i++) {
5033 if (tmp_bbio->stripes[i].dev->devid == srcdev_devid) {
5035 * In case of DUP, in order to keep it
5036 * simple, only add the mirror with the
5037 * lowest physical address
5040 physical_of_found <=
5041 tmp_bbio->stripes[i].physical)
5046 tmp_bbio->stripes[i].physical;
5051 mirror_num = index_srcdev + 1;
5052 patch_the_first_stripe_for_dev_replace = 1;
5053 physical_to_patch_in_first_stripe = physical_of_found;
5062 } else if (mirror_num > map->num_stripes) {
5068 stripe_nr_orig = stripe_nr;
5069 stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
5070 do_div(stripe_nr_end, map->stripe_len);
5071 stripe_end_offset = stripe_nr_end * map->stripe_len -
5074 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5075 if (rw & REQ_DISCARD)
5076 num_stripes = min_t(u64, map->num_stripes,
5077 stripe_nr_end - stripe_nr_orig);
5078 stripe_index = do_div(stripe_nr, map->num_stripes);
5079 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
5080 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))
5081 num_stripes = map->num_stripes;
5082 else if (mirror_num)
5083 stripe_index = mirror_num - 1;
5085 stripe_index = find_live_mirror(fs_info, map, 0,
5087 current->pid % map->num_stripes,
5088 dev_replace_is_ongoing);
5089 mirror_num = stripe_index + 1;
5092 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
5093 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
5094 num_stripes = map->num_stripes;
5095 } else if (mirror_num) {
5096 stripe_index = mirror_num - 1;
5101 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5102 int factor = map->num_stripes / map->sub_stripes;
5104 stripe_index = do_div(stripe_nr, factor);
5105 stripe_index *= map->sub_stripes;
5107 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
5108 num_stripes = map->sub_stripes;
5109 else if (rw & REQ_DISCARD)
5110 num_stripes = min_t(u64, map->sub_stripes *
5111 (stripe_nr_end - stripe_nr_orig),
5113 else if (mirror_num)
5114 stripe_index += mirror_num - 1;
5116 int old_stripe_index = stripe_index;
5117 stripe_index = find_live_mirror(fs_info, map,
5119 map->sub_stripes, stripe_index +
5120 current->pid % map->sub_stripes,
5121 dev_replace_is_ongoing);
5122 mirror_num = stripe_index - old_stripe_index + 1;
5125 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
5126 BTRFS_BLOCK_GROUP_RAID6)) {
5129 if (bbio_ret && ((rw & REQ_WRITE) || mirror_num > 1)
5133 /* push stripe_nr back to the start of the full stripe */
5134 stripe_nr = raid56_full_stripe_start;
5135 do_div(stripe_nr, stripe_len);
5137 stripe_index = do_div(stripe_nr, nr_data_stripes(map));
5139 /* RAID[56] write or recovery. Return all stripes */
5140 num_stripes = map->num_stripes;
5141 max_errors = nr_parity_stripes(map);
5143 raid_map = kmalloc_array(num_stripes, sizeof(u64),
5150 /* Work out the disk rotation on this stripe-set */
5152 rot = do_div(tmp, num_stripes);
5154 /* Fill in the logical address of each stripe */
5155 tmp = stripe_nr * nr_data_stripes(map);
5156 for (i = 0; i < nr_data_stripes(map); i++)
5157 raid_map[(i+rot) % num_stripes] =
5158 em->start + (tmp + i) * map->stripe_len;
5160 raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5161 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5162 raid_map[(i+rot+1) % num_stripes] =
5165 *length = map->stripe_len;
5170 * Mirror #0 or #1 means the original data block.
5171 * Mirror #2 is RAID5 parity block.
5172 * Mirror #3 is RAID6 Q block.
5174 stripe_index = do_div(stripe_nr, nr_data_stripes(map));
5176 stripe_index = nr_data_stripes(map) +
5179 /* We distribute the parity blocks across stripes */
5180 tmp = stripe_nr + stripe_index;
5181 stripe_index = do_div(tmp, map->num_stripes);
5185 * after this do_div call, stripe_nr is the number of stripes
5186 * on this device we have to walk to find the data, and
5187 * stripe_index is the number of our device in the stripe array
5189 stripe_index = do_div(stripe_nr, map->num_stripes);
5190 mirror_num = stripe_index + 1;
5192 BUG_ON(stripe_index >= map->num_stripes);
5194 num_alloc_stripes = num_stripes;
5195 if (dev_replace_is_ongoing) {
5196 if (rw & (REQ_WRITE | REQ_DISCARD))
5197 num_alloc_stripes <<= 1;
5198 if (rw & REQ_GET_READ_MIRRORS)
5199 num_alloc_stripes++;
5201 bbio = kzalloc(btrfs_bio_size(num_alloc_stripes), GFP_NOFS);
5207 atomic_set(&bbio->error, 0);
5209 if (rw & REQ_DISCARD) {
5211 int sub_stripes = 0;
5212 u64 stripes_per_dev = 0;
5213 u32 remaining_stripes = 0;
5214 u32 last_stripe = 0;
5217 (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
5218 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5221 sub_stripes = map->sub_stripes;
5223 factor = map->num_stripes / sub_stripes;
5224 stripes_per_dev = div_u64_rem(stripe_nr_end -
5227 &remaining_stripes);
5228 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5229 last_stripe *= sub_stripes;
5232 for (i = 0; i < num_stripes; i++) {
5233 bbio->stripes[i].physical =
5234 map->stripes[stripe_index].physical +
5235 stripe_offset + stripe_nr * map->stripe_len;
5236 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5238 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5239 BTRFS_BLOCK_GROUP_RAID10)) {
5240 bbio->stripes[i].length = stripes_per_dev *
5243 if (i / sub_stripes < remaining_stripes)
5244 bbio->stripes[i].length +=
5248 * Special for the first stripe and
5251 * |-------|...|-------|
5255 if (i < sub_stripes)
5256 bbio->stripes[i].length -=
5259 if (stripe_index >= last_stripe &&
5260 stripe_index <= (last_stripe +
5262 bbio->stripes[i].length -=
5265 if (i == sub_stripes - 1)
5268 bbio->stripes[i].length = *length;
5271 if (stripe_index == map->num_stripes) {
5272 /* This could only happen for RAID0/10 */
5278 for (i = 0; i < num_stripes; i++) {
5279 bbio->stripes[i].physical =
5280 map->stripes[stripe_index].physical +
5282 stripe_nr * map->stripe_len;
5283 bbio->stripes[i].dev =
5284 map->stripes[stripe_index].dev;
5289 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
5290 max_errors = btrfs_chunk_max_errors(map);
5292 if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) &&
5293 dev_replace->tgtdev != NULL) {
5294 int index_where_to_add;
5295 u64 srcdev_devid = dev_replace->srcdev->devid;
5298 * duplicate the write operations while the dev replace
5299 * procedure is running. Since the copying of the old disk
5300 * to the new disk takes place at run time while the
5301 * filesystem is mounted writable, the regular write
5302 * operations to the old disk have to be duplicated to go
5303 * to the new disk as well.
5304 * Note that device->missing is handled by the caller, and
5305 * that the write to the old disk is already set up in the
5308 index_where_to_add = num_stripes;
5309 for (i = 0; i < num_stripes; i++) {
5310 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5311 /* write to new disk, too */
5312 struct btrfs_bio_stripe *new =
5313 bbio->stripes + index_where_to_add;
5314 struct btrfs_bio_stripe *old =
5317 new->physical = old->physical;
5318 new->length = old->length;
5319 new->dev = dev_replace->tgtdev;
5320 index_where_to_add++;
5324 num_stripes = index_where_to_add;
5325 } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) &&
5326 dev_replace->tgtdev != NULL) {
5327 u64 srcdev_devid = dev_replace->srcdev->devid;
5328 int index_srcdev = 0;
5330 u64 physical_of_found = 0;
5333 * During the dev-replace procedure, the target drive can
5334 * also be used to read data in case it is needed to repair
5335 * a corrupt block elsewhere. This is possible if the
5336 * requested area is left of the left cursor. In this area,
5337 * the target drive is a full copy of the source drive.
5339 for (i = 0; i < num_stripes; i++) {
5340 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5342 * In case of DUP, in order to keep it
5343 * simple, only add the mirror with the
5344 * lowest physical address
5347 physical_of_found <=
5348 bbio->stripes[i].physical)
5352 physical_of_found = bbio->stripes[i].physical;
5356 u64 length = map->stripe_len;
5358 if (physical_of_found + length <=
5359 dev_replace->cursor_left) {
5360 struct btrfs_bio_stripe *tgtdev_stripe =
5361 bbio->stripes + num_stripes;
5363 tgtdev_stripe->physical = physical_of_found;
5364 tgtdev_stripe->length =
5365 bbio->stripes[index_srcdev].length;
5366 tgtdev_stripe->dev = dev_replace->tgtdev;
5374 bbio->num_stripes = num_stripes;
5375 bbio->max_errors = max_errors;
5376 bbio->mirror_num = mirror_num;
5379 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5380 * mirror_num == num_stripes + 1 && dev_replace target drive is
5381 * available as a mirror
5383 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5384 WARN_ON(num_stripes > 1);
5385 bbio->stripes[0].dev = dev_replace->tgtdev;
5386 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5387 bbio->mirror_num = map->num_stripes + 1;
5390 sort_parity_stripes(bbio, raid_map);
5391 *raid_map_ret = raid_map;
5394 if (dev_replace_is_ongoing)
5395 btrfs_dev_replace_unlock(dev_replace);
5396 free_extent_map(em);
5400 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5401 u64 logical, u64 *length,
5402 struct btrfs_bio **bbio_ret, int mirror_num)
5404 return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5408 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
5409 u64 chunk_start, u64 physical, u64 devid,
5410 u64 **logical, int *naddrs, int *stripe_len)
5412 struct extent_map_tree *em_tree = &map_tree->map_tree;
5413 struct extent_map *em;
5414 struct map_lookup *map;
5422 read_lock(&em_tree->lock);
5423 em = lookup_extent_mapping(em_tree, chunk_start, 1);
5424 read_unlock(&em_tree->lock);
5427 printk(KERN_ERR "BTRFS: couldn't find em for chunk %Lu\n",
5432 if (em->start != chunk_start) {
5433 printk(KERN_ERR "BTRFS: bad chunk start, em=%Lu, wanted=%Lu\n",
5434 em->start, chunk_start);
5435 free_extent_map(em);
5438 map = (struct map_lookup *)em->bdev;
5441 rmap_len = map->stripe_len;
5443 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5444 do_div(length, map->num_stripes / map->sub_stripes);
5445 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5446 do_div(length, map->num_stripes);
5447 else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
5448 BTRFS_BLOCK_GROUP_RAID6)) {
5449 do_div(length, nr_data_stripes(map));
5450 rmap_len = map->stripe_len * nr_data_stripes(map);
5453 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
5454 BUG_ON(!buf); /* -ENOMEM */
5456 for (i = 0; i < map->num_stripes; i++) {
5457 if (devid && map->stripes[i].dev->devid != devid)
5459 if (map->stripes[i].physical > physical ||
5460 map->stripes[i].physical + length <= physical)
5463 stripe_nr = physical - map->stripes[i].physical;
5464 do_div(stripe_nr, map->stripe_len);
5466 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5467 stripe_nr = stripe_nr * map->num_stripes + i;
5468 do_div(stripe_nr, map->sub_stripes);
5469 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5470 stripe_nr = stripe_nr * map->num_stripes + i;
5471 } /* else if RAID[56], multiply by nr_data_stripes().
5472 * Alternatively, just use rmap_len below instead of
5473 * map->stripe_len */
5475 bytenr = chunk_start + stripe_nr * rmap_len;
5476 WARN_ON(nr >= map->num_stripes);
5477 for (j = 0; j < nr; j++) {
5478 if (buf[j] == bytenr)
5482 WARN_ON(nr >= map->num_stripes);
5489 *stripe_len = rmap_len;
5491 free_extent_map(em);
5495 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio, int err)
5497 if (likely(bbio->flags & BTRFS_BIO_ORIG_BIO_SUBMITTED))
5498 bio_endio_nodec(bio, err);
5500 bio_endio(bio, err);
5504 static void btrfs_end_bio(struct bio *bio, int err)
5506 struct btrfs_bio *bbio = bio->bi_private;
5507 struct btrfs_device *dev = bbio->stripes[0].dev;
5508 int is_orig_bio = 0;
5511 atomic_inc(&bbio->error);
5512 if (err == -EIO || err == -EREMOTEIO) {
5513 unsigned int stripe_index =
5514 btrfs_io_bio(bio)->stripe_index;
5516 BUG_ON(stripe_index >= bbio->num_stripes);
5517 dev = bbio->stripes[stripe_index].dev;
5519 if (bio->bi_rw & WRITE)
5520 btrfs_dev_stat_inc(dev,
5521 BTRFS_DEV_STAT_WRITE_ERRS);
5523 btrfs_dev_stat_inc(dev,
5524 BTRFS_DEV_STAT_READ_ERRS);
5525 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
5526 btrfs_dev_stat_inc(dev,
5527 BTRFS_DEV_STAT_FLUSH_ERRS);
5528 btrfs_dev_stat_print_on_error(dev);
5533 if (bio == bbio->orig_bio)
5536 btrfs_bio_counter_dec(bbio->fs_info);
5538 if (atomic_dec_and_test(&bbio->stripes_pending)) {
5541 bio = bbio->orig_bio;
5544 bio->bi_private = bbio->private;
5545 bio->bi_end_io = bbio->end_io;
5546 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5547 /* only send an error to the higher layers if it is
5548 * beyond the tolerance of the btrfs bio
5550 if (atomic_read(&bbio->error) > bbio->max_errors) {
5554 * this bio is actually up to date, we didn't
5555 * go over the max number of errors
5557 set_bit(BIO_UPTODATE, &bio->bi_flags);
5561 btrfs_end_bbio(bbio, bio, err);
5562 } else if (!is_orig_bio) {
5568 * see run_scheduled_bios for a description of why bios are collected for
5571 * This will add one bio to the pending list for a device and make sure
5572 * the work struct is scheduled.
5574 static noinline void btrfs_schedule_bio(struct btrfs_root *root,
5575 struct btrfs_device *device,
5576 int rw, struct bio *bio)
5578 int should_queue = 1;
5579 struct btrfs_pending_bios *pending_bios;
5581 if (device->missing || !device->bdev) {
5582 bio_endio(bio, -EIO);
5586 /* don't bother with additional async steps for reads, right now */
5587 if (!(rw & REQ_WRITE)) {
5589 btrfsic_submit_bio(rw, bio);
5595 * nr_async_bios allows us to reliably return congestion to the
5596 * higher layers. Otherwise, the async bio makes it appear we have
5597 * made progress against dirty pages when we've really just put it
5598 * on a queue for later
5600 atomic_inc(&root->fs_info->nr_async_bios);
5601 WARN_ON(bio->bi_next);
5602 bio->bi_next = NULL;
5605 spin_lock(&device->io_lock);
5606 if (bio->bi_rw & REQ_SYNC)
5607 pending_bios = &device->pending_sync_bios;
5609 pending_bios = &device->pending_bios;
5611 if (pending_bios->tail)
5612 pending_bios->tail->bi_next = bio;
5614 pending_bios->tail = bio;
5615 if (!pending_bios->head)
5616 pending_bios->head = bio;
5617 if (device->running_pending)
5620 spin_unlock(&device->io_lock);
5623 btrfs_queue_work(root->fs_info->submit_workers,
5627 static int bio_size_ok(struct block_device *bdev, struct bio *bio,
5630 struct bio_vec *prev;
5631 struct request_queue *q = bdev_get_queue(bdev);
5632 unsigned int max_sectors = queue_max_sectors(q);
5633 struct bvec_merge_data bvm = {
5635 .bi_sector = sector,
5636 .bi_rw = bio->bi_rw,
5639 if (WARN_ON(bio->bi_vcnt == 0))
5642 prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
5643 if (bio_sectors(bio) > max_sectors)
5646 if (!q->merge_bvec_fn)
5649 bvm.bi_size = bio->bi_iter.bi_size - prev->bv_len;
5650 if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len)
5655 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5656 struct bio *bio, u64 physical, int dev_nr,
5659 struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
5661 bio->bi_private = bbio;
5662 btrfs_io_bio(bio)->stripe_index = dev_nr;
5663 bio->bi_end_io = btrfs_end_bio;
5664 bio->bi_iter.bi_sector = physical >> 9;
5667 struct rcu_string *name;
5670 name = rcu_dereference(dev->name);
5671 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
5672 "(%s id %llu), size=%u\n", rw,
5673 (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev,
5674 name->str, dev->devid, bio->bi_size);
5678 bio->bi_bdev = dev->bdev;
5680 btrfs_bio_counter_inc_noblocked(root->fs_info);
5683 btrfs_schedule_bio(root, dev, rw, bio);
5685 btrfsic_submit_bio(rw, bio);
5688 static int breakup_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5689 struct bio *first_bio, struct btrfs_device *dev,
5690 int dev_nr, int rw, int async)
5692 struct bio_vec *bvec = first_bio->bi_io_vec;
5694 int nr_vecs = bio_get_nr_vecs(dev->bdev);
5695 u64 physical = bbio->stripes[dev_nr].physical;
5698 bio = btrfs_bio_alloc(dev->bdev, physical >> 9, nr_vecs, GFP_NOFS);
5702 while (bvec <= (first_bio->bi_io_vec + first_bio->bi_vcnt - 1)) {
5703 if (bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5704 bvec->bv_offset) < bvec->bv_len) {
5705 u64 len = bio->bi_iter.bi_size;
5707 atomic_inc(&bbio->stripes_pending);
5708 submit_stripe_bio(root, bbio, bio, physical, dev_nr,
5716 submit_stripe_bio(root, bbio, bio, physical, dev_nr, rw, async);
5720 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
5722 atomic_inc(&bbio->error);
5723 if (atomic_dec_and_test(&bbio->stripes_pending)) {
5724 /* Shoud be the original bio. */
5725 WARN_ON(bio != bbio->orig_bio);
5727 bio->bi_private = bbio->private;
5728 bio->bi_end_io = bbio->end_io;
5729 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5730 bio->bi_iter.bi_sector = logical >> 9;
5732 btrfs_end_bbio(bbio, bio, -EIO);
5736 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
5737 int mirror_num, int async_submit)
5739 struct btrfs_device *dev;
5740 struct bio *first_bio = bio;
5741 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
5744 u64 *raid_map = NULL;
5748 struct btrfs_bio *bbio = NULL;
5750 length = bio->bi_iter.bi_size;
5751 map_length = length;
5753 btrfs_bio_counter_inc_blocked(root->fs_info);
5754 ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
5755 mirror_num, &raid_map);
5757 btrfs_bio_counter_dec(root->fs_info);
5761 total_devs = bbio->num_stripes;
5762 bbio->orig_bio = first_bio;
5763 bbio->private = first_bio->bi_private;
5764 bbio->end_io = first_bio->bi_end_io;
5765 bbio->fs_info = root->fs_info;
5766 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
5769 /* In this case, map_length has been set to the length of
5770 a single stripe; not the whole write */
5772 ret = raid56_parity_write(root, bio, bbio,
5773 raid_map, map_length);
5775 ret = raid56_parity_recover(root, bio, bbio,
5776 raid_map, map_length,
5780 * FIXME, replace dosen't support raid56 yet, please fix
5783 btrfs_bio_counter_dec(root->fs_info);
5787 if (map_length < length) {
5788 btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu",
5789 logical, length, map_length);
5793 while (dev_nr < total_devs) {
5794 dev = bbio->stripes[dev_nr].dev;
5795 if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
5796 bbio_error(bbio, first_bio, logical);
5802 * Check and see if we're ok with this bio based on it's size
5803 * and offset with the given device.
5805 if (!bio_size_ok(dev->bdev, first_bio,
5806 bbio->stripes[dev_nr].physical >> 9)) {
5807 ret = breakup_stripe_bio(root, bbio, first_bio, dev,
5808 dev_nr, rw, async_submit);
5814 if (dev_nr < total_devs - 1) {
5815 bio = btrfs_bio_clone(first_bio, GFP_NOFS);
5816 BUG_ON(!bio); /* -ENOMEM */
5819 bbio->flags |= BTRFS_BIO_ORIG_BIO_SUBMITTED;
5822 submit_stripe_bio(root, bbio, bio,
5823 bbio->stripes[dev_nr].physical, dev_nr, rw,
5827 btrfs_bio_counter_dec(root->fs_info);
5831 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
5834 struct btrfs_device *device;
5835 struct btrfs_fs_devices *cur_devices;
5837 cur_devices = fs_info->fs_devices;
5838 while (cur_devices) {
5840 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5841 device = __find_device(&cur_devices->devices,
5846 cur_devices = cur_devices->seed;
5851 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
5852 u64 devid, u8 *dev_uuid)
5854 struct btrfs_device *device;
5855 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
5857 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
5861 list_add(&device->dev_list, &fs_devices->devices);
5862 device->fs_devices = fs_devices;
5863 fs_devices->num_devices++;
5865 device->missing = 1;
5866 fs_devices->missing_devices++;
5872 * btrfs_alloc_device - allocate struct btrfs_device
5873 * @fs_info: used only for generating a new devid, can be NULL if
5874 * devid is provided (i.e. @devid != NULL).
5875 * @devid: a pointer to devid for this device. If NULL a new devid
5877 * @uuid: a pointer to UUID for this device. If NULL a new UUID
5880 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
5881 * on error. Returned struct is not linked onto any lists and can be
5882 * destroyed with kfree() right away.
5884 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
5888 struct btrfs_device *dev;
5891 if (WARN_ON(!devid && !fs_info))
5892 return ERR_PTR(-EINVAL);
5894 dev = __alloc_device();
5903 ret = find_next_devid(fs_info, &tmp);
5906 return ERR_PTR(ret);
5912 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
5914 generate_random_uuid(dev->uuid);
5916 btrfs_init_work(&dev->work, btrfs_submit_helper,
5917 pending_bios_fn, NULL, NULL);
5922 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
5923 struct extent_buffer *leaf,
5924 struct btrfs_chunk *chunk)
5926 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5927 struct map_lookup *map;
5928 struct extent_map *em;
5932 u8 uuid[BTRFS_UUID_SIZE];
5937 logical = key->offset;
5938 length = btrfs_chunk_length(leaf, chunk);
5940 read_lock(&map_tree->map_tree.lock);
5941 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
5942 read_unlock(&map_tree->map_tree.lock);
5944 /* already mapped? */
5945 if (em && em->start <= logical && em->start + em->len > logical) {
5946 free_extent_map(em);
5949 free_extent_map(em);
5952 em = alloc_extent_map();
5955 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
5956 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
5958 free_extent_map(em);
5962 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
5963 em->bdev = (struct block_device *)map;
5964 em->start = logical;
5967 em->block_start = 0;
5968 em->block_len = em->len;
5970 map->num_stripes = num_stripes;
5971 map->io_width = btrfs_chunk_io_width(leaf, chunk);
5972 map->io_align = btrfs_chunk_io_align(leaf, chunk);
5973 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
5974 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
5975 map->type = btrfs_chunk_type(leaf, chunk);
5976 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
5977 for (i = 0; i < num_stripes; i++) {
5978 map->stripes[i].physical =
5979 btrfs_stripe_offset_nr(leaf, chunk, i);
5980 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
5981 read_extent_buffer(leaf, uuid, (unsigned long)
5982 btrfs_stripe_dev_uuid_nr(chunk, i),
5984 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
5986 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
5987 free_extent_map(em);
5990 if (!map->stripes[i].dev) {
5991 map->stripes[i].dev =
5992 add_missing_dev(root, devid, uuid);
5993 if (!map->stripes[i].dev) {
5994 free_extent_map(em);
5998 map->stripes[i].dev->in_fs_metadata = 1;
6001 write_lock(&map_tree->map_tree.lock);
6002 ret = add_extent_mapping(&map_tree->map_tree, em, 0);
6003 write_unlock(&map_tree->map_tree.lock);
6004 BUG_ON(ret); /* Tree corruption */
6005 free_extent_map(em);
6010 static void fill_device_from_item(struct extent_buffer *leaf,
6011 struct btrfs_dev_item *dev_item,
6012 struct btrfs_device *device)
6016 device->devid = btrfs_device_id(leaf, dev_item);
6017 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6018 device->total_bytes = device->disk_total_bytes;
6019 device->commit_total_bytes = device->disk_total_bytes;
6020 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6021 device->commit_bytes_used = device->bytes_used;
6022 device->type = btrfs_device_type(leaf, dev_item);
6023 device->io_align = btrfs_device_io_align(leaf, dev_item);
6024 device->io_width = btrfs_device_io_width(leaf, dev_item);
6025 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6026 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6027 device->is_tgtdev_for_dev_replace = 0;
6029 ptr = btrfs_device_uuid(dev_item);
6030 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6033 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
6035 struct btrfs_fs_devices *fs_devices;
6038 BUG_ON(!mutex_is_locked(&uuid_mutex));
6040 fs_devices = root->fs_info->fs_devices->seed;
6041 while (fs_devices) {
6042 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
6046 fs_devices = fs_devices->seed;
6049 fs_devices = find_fsid(fsid);
6055 fs_devices = clone_fs_devices(fs_devices);
6056 if (IS_ERR(fs_devices)) {
6057 ret = PTR_ERR(fs_devices);
6061 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
6062 root->fs_info->bdev_holder);
6064 free_fs_devices(fs_devices);
6068 if (!fs_devices->seeding) {
6069 __btrfs_close_devices(fs_devices);
6070 free_fs_devices(fs_devices);
6075 fs_devices->seed = root->fs_info->fs_devices->seed;
6076 root->fs_info->fs_devices->seed = fs_devices;
6081 static int read_one_dev(struct btrfs_root *root,
6082 struct extent_buffer *leaf,
6083 struct btrfs_dev_item *dev_item)
6085 struct btrfs_device *device;
6088 u8 fs_uuid[BTRFS_UUID_SIZE];
6089 u8 dev_uuid[BTRFS_UUID_SIZE];
6091 devid = btrfs_device_id(leaf, dev_item);
6092 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6094 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6097 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
6098 ret = open_seed_devices(root, fs_uuid);
6099 if (ret && !btrfs_test_opt(root, DEGRADED))
6103 device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
6104 if (!device || !device->bdev) {
6105 if (!btrfs_test_opt(root, DEGRADED))
6109 btrfs_warn(root->fs_info, "devid %llu missing", devid);
6110 device = add_missing_dev(root, devid, dev_uuid);
6113 } else if (!device->missing) {
6115 * this happens when a device that was properly setup
6116 * in the device info lists suddenly goes bad.
6117 * device->bdev is NULL, and so we have to set
6118 * device->missing to one here
6120 root->fs_info->fs_devices->missing_devices++;
6121 device->missing = 1;
6125 if (device->fs_devices != root->fs_info->fs_devices) {
6126 BUG_ON(device->writeable);
6127 if (device->generation !=
6128 btrfs_device_generation(leaf, dev_item))
6132 fill_device_from_item(leaf, dev_item, device);
6133 device->in_fs_metadata = 1;
6134 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
6135 device->fs_devices->total_rw_bytes += device->total_bytes;
6136 spin_lock(&root->fs_info->free_chunk_lock);
6137 root->fs_info->free_chunk_space += device->total_bytes -
6139 spin_unlock(&root->fs_info->free_chunk_lock);
6145 int btrfs_read_sys_array(struct btrfs_root *root)
6147 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
6148 struct extent_buffer *sb;
6149 struct btrfs_disk_key *disk_key;
6150 struct btrfs_chunk *chunk;
6152 unsigned long sb_ptr;
6158 struct btrfs_key key;
6160 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
6161 BTRFS_SUPER_INFO_SIZE);
6164 btrfs_set_buffer_uptodate(sb);
6165 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6167 * The sb extent buffer is artifical and just used to read the system array.
6168 * btrfs_set_buffer_uptodate() call does not properly mark all it's
6169 * pages up-to-date when the page is larger: extent does not cover the
6170 * whole page and consequently check_page_uptodate does not find all
6171 * the page's extents up-to-date (the hole beyond sb),
6172 * write_extent_buffer then triggers a WARN_ON.
6174 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6175 * but sb spans only this function. Add an explicit SetPageUptodate call
6176 * to silence the warning eg. on PowerPC 64.
6178 if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
6179 SetPageUptodate(sb->pages[0]);
6181 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6182 array_size = btrfs_super_sys_array_size(super_copy);
6184 ptr = super_copy->sys_chunk_array;
6185 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
6188 while (cur < array_size) {
6189 disk_key = (struct btrfs_disk_key *)ptr;
6190 btrfs_disk_key_to_cpu(&key, disk_key);
6192 len = sizeof(*disk_key); ptr += len;
6196 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6197 chunk = (struct btrfs_chunk *)sb_ptr;
6198 ret = read_one_chunk(root, &key, sb, chunk);
6201 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6202 len = btrfs_chunk_item_size(num_stripes);
6211 free_extent_buffer(sb);
6215 int btrfs_read_chunk_tree(struct btrfs_root *root)
6217 struct btrfs_path *path;
6218 struct extent_buffer *leaf;
6219 struct btrfs_key key;
6220 struct btrfs_key found_key;
6224 root = root->fs_info->chunk_root;
6226 path = btrfs_alloc_path();
6230 mutex_lock(&uuid_mutex);
6234 * Read all device items, and then all the chunk items. All
6235 * device items are found before any chunk item (their object id
6236 * is smaller than the lowest possible object id for a chunk
6237 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6239 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6242 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6246 leaf = path->nodes[0];
6247 slot = path->slots[0];
6248 if (slot >= btrfs_header_nritems(leaf)) {
6249 ret = btrfs_next_leaf(root, path);
6256 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6257 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6258 struct btrfs_dev_item *dev_item;
6259 dev_item = btrfs_item_ptr(leaf, slot,
6260 struct btrfs_dev_item);
6261 ret = read_one_dev(root, leaf, dev_item);
6264 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6265 struct btrfs_chunk *chunk;
6266 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6267 ret = read_one_chunk(root, &found_key, leaf, chunk);
6275 unlock_chunks(root);
6276 mutex_unlock(&uuid_mutex);
6278 btrfs_free_path(path);
6282 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6284 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6285 struct btrfs_device *device;
6287 while (fs_devices) {
6288 mutex_lock(&fs_devices->device_list_mutex);
6289 list_for_each_entry(device, &fs_devices->devices, dev_list)
6290 device->dev_root = fs_info->dev_root;
6291 mutex_unlock(&fs_devices->device_list_mutex);
6293 fs_devices = fs_devices->seed;
6297 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6301 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6302 btrfs_dev_stat_reset(dev, i);
6305 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6307 struct btrfs_key key;
6308 struct btrfs_key found_key;
6309 struct btrfs_root *dev_root = fs_info->dev_root;
6310 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6311 struct extent_buffer *eb;
6314 struct btrfs_device *device;
6315 struct btrfs_path *path = NULL;
6318 path = btrfs_alloc_path();
6324 mutex_lock(&fs_devices->device_list_mutex);
6325 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6327 struct btrfs_dev_stats_item *ptr;
6330 key.type = BTRFS_DEV_STATS_KEY;
6331 key.offset = device->devid;
6332 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6334 __btrfs_reset_dev_stats(device);
6335 device->dev_stats_valid = 1;
6336 btrfs_release_path(path);
6339 slot = path->slots[0];
6340 eb = path->nodes[0];
6341 btrfs_item_key_to_cpu(eb, &found_key, slot);
6342 item_size = btrfs_item_size_nr(eb, slot);
6344 ptr = btrfs_item_ptr(eb, slot,
6345 struct btrfs_dev_stats_item);
6347 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6348 if (item_size >= (1 + i) * sizeof(__le64))
6349 btrfs_dev_stat_set(device, i,
6350 btrfs_dev_stats_value(eb, ptr, i));
6352 btrfs_dev_stat_reset(device, i);
6355 device->dev_stats_valid = 1;
6356 btrfs_dev_stat_print_on_load(device);
6357 btrfs_release_path(path);
6359 mutex_unlock(&fs_devices->device_list_mutex);
6362 btrfs_free_path(path);
6363 return ret < 0 ? ret : 0;
6366 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
6367 struct btrfs_root *dev_root,
6368 struct btrfs_device *device)
6370 struct btrfs_path *path;
6371 struct btrfs_key key;
6372 struct extent_buffer *eb;
6373 struct btrfs_dev_stats_item *ptr;
6378 key.type = BTRFS_DEV_STATS_KEY;
6379 key.offset = device->devid;
6381 path = btrfs_alloc_path();
6383 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
6385 printk_in_rcu(KERN_WARNING "BTRFS: "
6386 "error %d while searching for dev_stats item for device %s!\n",
6387 ret, rcu_str_deref(device->name));
6392 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
6393 /* need to delete old one and insert a new one */
6394 ret = btrfs_del_item(trans, dev_root, path);
6396 printk_in_rcu(KERN_WARNING "BTRFS: "
6397 "delete too small dev_stats item for device %s failed %d!\n",
6398 rcu_str_deref(device->name), ret);
6405 /* need to insert a new item */
6406 btrfs_release_path(path);
6407 ret = btrfs_insert_empty_item(trans, dev_root, path,
6408 &key, sizeof(*ptr));
6410 printk_in_rcu(KERN_WARNING "BTRFS: "
6411 "insert dev_stats item for device %s failed %d!\n",
6412 rcu_str_deref(device->name), ret);
6417 eb = path->nodes[0];
6418 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
6419 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6420 btrfs_set_dev_stats_value(eb, ptr, i,
6421 btrfs_dev_stat_read(device, i));
6422 btrfs_mark_buffer_dirty(eb);
6425 btrfs_free_path(path);
6430 * called from commit_transaction. Writes all changed device stats to disk.
6432 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
6433 struct btrfs_fs_info *fs_info)
6435 struct btrfs_root *dev_root = fs_info->dev_root;
6436 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6437 struct btrfs_device *device;
6441 mutex_lock(&fs_devices->device_list_mutex);
6442 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6443 if (!device->dev_stats_valid || !btrfs_dev_stats_dirty(device))
6446 stats_cnt = atomic_read(&device->dev_stats_ccnt);
6447 ret = update_dev_stat_item(trans, dev_root, device);
6449 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
6451 mutex_unlock(&fs_devices->device_list_mutex);
6456 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
6458 btrfs_dev_stat_inc(dev, index);
6459 btrfs_dev_stat_print_on_error(dev);
6462 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
6464 if (!dev->dev_stats_valid)
6466 printk_ratelimited_in_rcu(KERN_ERR "BTRFS: "
6467 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6468 rcu_str_deref(dev->name),
6469 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6470 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6471 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6472 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6473 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6476 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
6480 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6481 if (btrfs_dev_stat_read(dev, i) != 0)
6483 if (i == BTRFS_DEV_STAT_VALUES_MAX)
6484 return; /* all values == 0, suppress message */
6486 printk_in_rcu(KERN_INFO "BTRFS: "
6487 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6488 rcu_str_deref(dev->name),
6489 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6490 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6491 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6492 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6493 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6496 int btrfs_get_dev_stats(struct btrfs_root *root,
6497 struct btrfs_ioctl_get_dev_stats *stats)
6499 struct btrfs_device *dev;
6500 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6503 mutex_lock(&fs_devices->device_list_mutex);
6504 dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
6505 mutex_unlock(&fs_devices->device_list_mutex);
6508 btrfs_warn(root->fs_info, "get dev_stats failed, device not found");
6510 } else if (!dev->dev_stats_valid) {
6511 btrfs_warn(root->fs_info, "get dev_stats failed, not yet valid");
6513 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
6514 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6515 if (stats->nr_items > i)
6517 btrfs_dev_stat_read_and_reset(dev, i);
6519 btrfs_dev_stat_reset(dev, i);
6522 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6523 if (stats->nr_items > i)
6524 stats->values[i] = btrfs_dev_stat_read(dev, i);
6526 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
6527 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
6531 int btrfs_scratch_superblock(struct btrfs_device *device)
6533 struct buffer_head *bh;
6534 struct btrfs_super_block *disk_super;
6536 bh = btrfs_read_dev_super(device->bdev);
6539 disk_super = (struct btrfs_super_block *)bh->b_data;
6541 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
6542 set_buffer_dirty(bh);
6543 sync_dirty_buffer(bh);
6550 * Update the size of all devices, which is used for writing out the
6553 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
6555 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6556 struct btrfs_device *curr, *next;
6558 if (list_empty(&fs_devices->resized_devices))
6561 mutex_lock(&fs_devices->device_list_mutex);
6562 lock_chunks(fs_info->dev_root);
6563 list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
6565 list_del_init(&curr->resized_list);
6566 curr->commit_total_bytes = curr->disk_total_bytes;
6568 unlock_chunks(fs_info->dev_root);
6569 mutex_unlock(&fs_devices->device_list_mutex);
6572 /* Must be invoked during the transaction commit */
6573 void btrfs_update_commit_device_bytes_used(struct btrfs_root *root,
6574 struct btrfs_transaction *transaction)
6576 struct extent_map *em;
6577 struct map_lookup *map;
6578 struct btrfs_device *dev;
6581 if (list_empty(&transaction->pending_chunks))
6584 /* In order to kick the device replace finish process */
6586 list_for_each_entry(em, &transaction->pending_chunks, list) {
6587 map = (struct map_lookup *)em->bdev;
6589 for (i = 0; i < map->num_stripes; i++) {
6590 dev = map->stripes[i].dev;
6591 dev->commit_bytes_used = dev->bytes_used;
6594 unlock_chunks(root);