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 <asm/div64.h>
29 #include "extent_map.h"
31 #include "transaction.h"
32 #include "print-tree.h"
34 #include "async-thread.h"
36 static int init_first_rw_device(struct btrfs_trans_handle *trans,
37 struct btrfs_root *root,
38 struct btrfs_device *device);
39 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
41 static DEFINE_MUTEX(uuid_mutex);
42 static LIST_HEAD(fs_uuids);
44 static void lock_chunks(struct btrfs_root *root)
46 mutex_lock(&root->fs_info->chunk_mutex);
49 static void unlock_chunks(struct btrfs_root *root)
51 mutex_unlock(&root->fs_info->chunk_mutex);
54 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
56 struct btrfs_device *device;
57 WARN_ON(fs_devices->opened);
58 while (!list_empty(&fs_devices->devices)) {
59 device = list_entry(fs_devices->devices.next,
60 struct btrfs_device, dev_list);
61 list_del(&device->dev_list);
68 int btrfs_cleanup_fs_uuids(void)
70 struct btrfs_fs_devices *fs_devices;
72 while (!list_empty(&fs_uuids)) {
73 fs_devices = list_entry(fs_uuids.next,
74 struct btrfs_fs_devices, list);
75 list_del(&fs_devices->list);
76 free_fs_devices(fs_devices);
81 static noinline struct btrfs_device *__find_device(struct list_head *head,
84 struct btrfs_device *dev;
86 list_for_each_entry(dev, head, dev_list) {
87 if (dev->devid == devid &&
88 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
95 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
97 struct btrfs_fs_devices *fs_devices;
99 list_for_each_entry(fs_devices, &fs_uuids, list) {
100 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
106 static void requeue_list(struct btrfs_pending_bios *pending_bios,
107 struct bio *head, struct bio *tail)
110 struct bio *old_head;
112 old_head = pending_bios->head;
113 pending_bios->head = head;
114 if (pending_bios->tail)
115 tail->bi_next = old_head;
117 pending_bios->tail = tail;
121 * we try to collect pending bios for a device so we don't get a large
122 * number of procs sending bios down to the same device. This greatly
123 * improves the schedulers ability to collect and merge the bios.
125 * But, it also turns into a long list of bios to process and that is sure
126 * to eventually make the worker thread block. The solution here is to
127 * make some progress and then put this work struct back at the end of
128 * the list if the block device is congested. This way, multiple devices
129 * can make progress from a single worker thread.
131 static noinline int run_scheduled_bios(struct btrfs_device *device)
134 struct backing_dev_info *bdi;
135 struct btrfs_fs_info *fs_info;
136 struct btrfs_pending_bios *pending_bios;
140 unsigned long num_run;
141 unsigned long batch_run = 0;
143 unsigned long last_waited = 0;
145 struct blk_plug plug;
148 * this function runs all the bios we've collected for
149 * a particular device. We don't want to wander off to
150 * another device without first sending all of these down.
151 * So, setup a plug here and finish it off before we return
153 blk_start_plug(&plug);
155 bdi = blk_get_backing_dev_info(device->bdev);
156 fs_info = device->dev_root->fs_info;
157 limit = btrfs_async_submit_limit(fs_info);
158 limit = limit * 2 / 3;
161 spin_lock(&device->io_lock);
166 /* take all the bios off the list at once and process them
167 * later on (without the lock held). But, remember the
168 * tail and other pointers so the bios can be properly reinserted
169 * into the list if we hit congestion
171 if (!force_reg && device->pending_sync_bios.head) {
172 pending_bios = &device->pending_sync_bios;
175 pending_bios = &device->pending_bios;
179 pending = pending_bios->head;
180 tail = pending_bios->tail;
181 WARN_ON(pending && !tail);
184 * if pending was null this time around, no bios need processing
185 * at all and we can stop. Otherwise it'll loop back up again
186 * and do an additional check so no bios are missed.
188 * device->running_pending is used to synchronize with the
191 if (device->pending_sync_bios.head == NULL &&
192 device->pending_bios.head == NULL) {
194 device->running_pending = 0;
197 device->running_pending = 1;
200 pending_bios->head = NULL;
201 pending_bios->tail = NULL;
203 spin_unlock(&device->io_lock);
208 /* we want to work on both lists, but do more bios on the
209 * sync list than the regular list
212 pending_bios != &device->pending_sync_bios &&
213 device->pending_sync_bios.head) ||
214 (num_run > 64 && pending_bios == &device->pending_sync_bios &&
215 device->pending_bios.head)) {
216 spin_lock(&device->io_lock);
217 requeue_list(pending_bios, pending, tail);
222 pending = pending->bi_next;
224 atomic_dec(&fs_info->nr_async_bios);
226 if (atomic_read(&fs_info->nr_async_bios) < limit &&
227 waitqueue_active(&fs_info->async_submit_wait))
228 wake_up(&fs_info->async_submit_wait);
230 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
232 submit_bio(cur->bi_rw, cur);
239 * we made progress, there is more work to do and the bdi
240 * is now congested. Back off and let other work structs
243 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
244 fs_info->fs_devices->open_devices > 1) {
245 struct io_context *ioc;
247 ioc = current->io_context;
250 * the main goal here is that we don't want to
251 * block if we're going to be able to submit
252 * more requests without blocking.
254 * This code does two great things, it pokes into
255 * the elevator code from a filesystem _and_
256 * it makes assumptions about how batching works.
258 if (ioc && ioc->nr_batch_requests > 0 &&
259 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
261 ioc->last_waited == last_waited)) {
263 * we want to go through our batch of
264 * requests and stop. So, we copy out
265 * the ioc->last_waited time and test
266 * against it before looping
268 last_waited = ioc->last_waited;
273 spin_lock(&device->io_lock);
274 requeue_list(pending_bios, pending, tail);
275 device->running_pending = 1;
277 spin_unlock(&device->io_lock);
278 btrfs_requeue_work(&device->work);
287 spin_lock(&device->io_lock);
288 if (device->pending_bios.head || device->pending_sync_bios.head)
290 spin_unlock(&device->io_lock);
293 blk_finish_plug(&plug);
297 static void pending_bios_fn(struct btrfs_work *work)
299 struct btrfs_device *device;
301 device = container_of(work, struct btrfs_device, work);
302 run_scheduled_bios(device);
305 static noinline int device_list_add(const char *path,
306 struct btrfs_super_block *disk_super,
307 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
309 struct btrfs_device *device;
310 struct btrfs_fs_devices *fs_devices;
311 u64 found_transid = btrfs_super_generation(disk_super);
314 fs_devices = find_fsid(disk_super->fsid);
316 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
319 INIT_LIST_HEAD(&fs_devices->devices);
320 INIT_LIST_HEAD(&fs_devices->alloc_list);
321 list_add(&fs_devices->list, &fs_uuids);
322 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
323 fs_devices->latest_devid = devid;
324 fs_devices->latest_trans = found_transid;
325 mutex_init(&fs_devices->device_list_mutex);
328 device = __find_device(&fs_devices->devices, devid,
329 disk_super->dev_item.uuid);
332 if (fs_devices->opened)
335 device = kzalloc(sizeof(*device), GFP_NOFS);
337 /* we can safely leave the fs_devices entry around */
340 device->devid = devid;
341 device->work.func = pending_bios_fn;
342 memcpy(device->uuid, disk_super->dev_item.uuid,
344 spin_lock_init(&device->io_lock);
345 device->name = kstrdup(path, GFP_NOFS);
350 INIT_LIST_HEAD(&device->dev_alloc_list);
352 mutex_lock(&fs_devices->device_list_mutex);
353 list_add_rcu(&device->dev_list, &fs_devices->devices);
354 mutex_unlock(&fs_devices->device_list_mutex);
356 device->fs_devices = fs_devices;
357 fs_devices->num_devices++;
358 } else if (!device->name || strcmp(device->name, path)) {
359 name = kstrdup(path, GFP_NOFS);
364 if (device->missing) {
365 fs_devices->missing_devices--;
370 if (found_transid > fs_devices->latest_trans) {
371 fs_devices->latest_devid = devid;
372 fs_devices->latest_trans = found_transid;
374 *fs_devices_ret = fs_devices;
378 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
380 struct btrfs_fs_devices *fs_devices;
381 struct btrfs_device *device;
382 struct btrfs_device *orig_dev;
384 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
386 return ERR_PTR(-ENOMEM);
388 INIT_LIST_HEAD(&fs_devices->devices);
389 INIT_LIST_HEAD(&fs_devices->alloc_list);
390 INIT_LIST_HEAD(&fs_devices->list);
391 mutex_init(&fs_devices->device_list_mutex);
392 fs_devices->latest_devid = orig->latest_devid;
393 fs_devices->latest_trans = orig->latest_trans;
394 memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
396 /* We have held the volume lock, it is safe to get the devices. */
397 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
398 device = kzalloc(sizeof(*device), GFP_NOFS);
402 device->name = kstrdup(orig_dev->name, GFP_NOFS);
408 device->devid = orig_dev->devid;
409 device->work.func = pending_bios_fn;
410 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
411 spin_lock_init(&device->io_lock);
412 INIT_LIST_HEAD(&device->dev_list);
413 INIT_LIST_HEAD(&device->dev_alloc_list);
415 list_add(&device->dev_list, &fs_devices->devices);
416 device->fs_devices = fs_devices;
417 fs_devices->num_devices++;
421 free_fs_devices(fs_devices);
422 return ERR_PTR(-ENOMEM);
425 int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
427 struct btrfs_device *device, *next;
429 mutex_lock(&uuid_mutex);
431 /* This is the initialized path, it is safe to release the devices. */
432 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
433 if (device->in_fs_metadata)
437 blkdev_put(device->bdev, device->mode);
439 fs_devices->open_devices--;
441 if (device->writeable) {
442 list_del_init(&device->dev_alloc_list);
443 device->writeable = 0;
444 fs_devices->rw_devices--;
446 list_del_init(&device->dev_list);
447 fs_devices->num_devices--;
452 if (fs_devices->seed) {
453 fs_devices = fs_devices->seed;
457 mutex_unlock(&uuid_mutex);
461 static void __free_device(struct work_struct *work)
463 struct btrfs_device *device;
465 device = container_of(work, struct btrfs_device, rcu_work);
468 blkdev_put(device->bdev, device->mode);
474 static void free_device(struct rcu_head *head)
476 struct btrfs_device *device;
478 device = container_of(head, struct btrfs_device, rcu);
480 INIT_WORK(&device->rcu_work, __free_device);
481 schedule_work(&device->rcu_work);
484 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
486 struct btrfs_device *device;
488 if (--fs_devices->opened > 0)
491 mutex_lock(&fs_devices->device_list_mutex);
492 list_for_each_entry(device, &fs_devices->devices, dev_list) {
493 struct btrfs_device *new_device;
496 fs_devices->open_devices--;
498 if (device->writeable) {
499 list_del_init(&device->dev_alloc_list);
500 fs_devices->rw_devices--;
503 new_device = kmalloc(sizeof(*new_device), GFP_NOFS);
505 memcpy(new_device, device, sizeof(*new_device));
506 new_device->name = kstrdup(device->name, GFP_NOFS);
507 BUG_ON(device->name && !new_device->name);
508 new_device->bdev = NULL;
509 new_device->writeable = 0;
510 new_device->in_fs_metadata = 0;
511 list_replace_rcu(&device->dev_list, &new_device->dev_list);
513 call_rcu(&device->rcu, free_device);
515 mutex_unlock(&fs_devices->device_list_mutex);
517 WARN_ON(fs_devices->open_devices);
518 WARN_ON(fs_devices->rw_devices);
519 fs_devices->opened = 0;
520 fs_devices->seeding = 0;
525 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
527 struct btrfs_fs_devices *seed_devices = NULL;
530 mutex_lock(&uuid_mutex);
531 ret = __btrfs_close_devices(fs_devices);
532 if (!fs_devices->opened) {
533 seed_devices = fs_devices->seed;
534 fs_devices->seed = NULL;
536 mutex_unlock(&uuid_mutex);
538 while (seed_devices) {
539 fs_devices = seed_devices;
540 seed_devices = fs_devices->seed;
541 __btrfs_close_devices(fs_devices);
542 free_fs_devices(fs_devices);
547 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
548 fmode_t flags, void *holder)
550 struct block_device *bdev;
551 struct list_head *head = &fs_devices->devices;
552 struct btrfs_device *device;
553 struct block_device *latest_bdev = NULL;
554 struct buffer_head *bh;
555 struct btrfs_super_block *disk_super;
556 u64 latest_devid = 0;
557 u64 latest_transid = 0;
564 list_for_each_entry(device, head, dev_list) {
570 bdev = blkdev_get_by_path(device->name, flags, holder);
572 printk(KERN_INFO "open %s failed\n", device->name);
575 set_blocksize(bdev, 4096);
577 bh = btrfs_read_dev_super(bdev);
583 disk_super = (struct btrfs_super_block *)bh->b_data;
584 devid = btrfs_stack_device_id(&disk_super->dev_item);
585 if (devid != device->devid)
588 if (memcmp(device->uuid, disk_super->dev_item.uuid,
592 device->generation = btrfs_super_generation(disk_super);
593 if (!latest_transid || device->generation > latest_transid) {
594 latest_devid = devid;
595 latest_transid = device->generation;
599 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
600 device->writeable = 0;
602 device->writeable = !bdev_read_only(bdev);
607 device->in_fs_metadata = 0;
608 device->mode = flags;
610 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
611 fs_devices->rotating = 1;
613 fs_devices->open_devices++;
614 if (device->writeable) {
615 fs_devices->rw_devices++;
616 list_add(&device->dev_alloc_list,
617 &fs_devices->alloc_list);
625 blkdev_put(bdev, flags);
629 if (fs_devices->open_devices == 0) {
633 fs_devices->seeding = seeding;
634 fs_devices->opened = 1;
635 fs_devices->latest_bdev = latest_bdev;
636 fs_devices->latest_devid = latest_devid;
637 fs_devices->latest_trans = latest_transid;
638 fs_devices->total_rw_bytes = 0;
643 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
644 fmode_t flags, void *holder)
648 mutex_lock(&uuid_mutex);
649 if (fs_devices->opened) {
650 fs_devices->opened++;
653 ret = __btrfs_open_devices(fs_devices, flags, holder);
655 mutex_unlock(&uuid_mutex);
659 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
660 struct btrfs_fs_devices **fs_devices_ret)
662 struct btrfs_super_block *disk_super;
663 struct block_device *bdev;
664 struct buffer_head *bh;
669 mutex_lock(&uuid_mutex);
672 bdev = blkdev_get_by_path(path, flags, holder);
679 ret = set_blocksize(bdev, 4096);
682 bh = btrfs_read_dev_super(bdev);
687 disk_super = (struct btrfs_super_block *)bh->b_data;
688 devid = btrfs_stack_device_id(&disk_super->dev_item);
689 transid = btrfs_super_generation(disk_super);
690 if (disk_super->label[0])
691 printk(KERN_INFO "device label %s ", disk_super->label);
693 printk(KERN_INFO "device fsid %pU ", disk_super->fsid);
694 printk(KERN_CONT "devid %llu transid %llu %s\n",
695 (unsigned long long)devid, (unsigned long long)transid, path);
696 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
700 blkdev_put(bdev, flags);
702 mutex_unlock(&uuid_mutex);
706 /* helper to account the used device space in the range */
707 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
708 u64 end, u64 *length)
710 struct btrfs_key key;
711 struct btrfs_root *root = device->dev_root;
712 struct btrfs_dev_extent *dev_extent;
713 struct btrfs_path *path;
717 struct extent_buffer *l;
721 if (start >= device->total_bytes)
724 path = btrfs_alloc_path();
729 key.objectid = device->devid;
731 key.type = BTRFS_DEV_EXTENT_KEY;
733 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
737 ret = btrfs_previous_item(root, path, key.objectid, key.type);
744 slot = path->slots[0];
745 if (slot >= btrfs_header_nritems(l)) {
746 ret = btrfs_next_leaf(root, path);
754 btrfs_item_key_to_cpu(l, &key, slot);
756 if (key.objectid < device->devid)
759 if (key.objectid > device->devid)
762 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
765 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
766 extent_end = key.offset + btrfs_dev_extent_length(l,
768 if (key.offset <= start && extent_end > end) {
769 *length = end - start + 1;
771 } else if (key.offset <= start && extent_end > start)
772 *length += extent_end - start;
773 else if (key.offset > start && extent_end <= end)
774 *length += extent_end - key.offset;
775 else if (key.offset > start && key.offset <= end) {
776 *length += end - key.offset + 1;
778 } else if (key.offset > end)
786 btrfs_free_path(path);
791 * find_free_dev_extent - find free space in the specified device
792 * @trans: transaction handler
793 * @device: the device which we search the free space in
794 * @num_bytes: the size of the free space that we need
795 * @start: store the start of the free space.
796 * @len: the size of the free space. that we find, or the size of the max
797 * free space if we don't find suitable free space
799 * this uses a pretty simple search, the expectation is that it is
800 * called very infrequently and that a given device has a small number
803 * @start is used to store the start of the free space if we find. But if we
804 * don't find suitable free space, it will be used to store the start position
805 * of the max free space.
807 * @len is used to store the size of the free space that we find.
808 * But if we don't find suitable free space, it is used to store the size of
809 * the max free space.
811 int find_free_dev_extent(struct btrfs_trans_handle *trans,
812 struct btrfs_device *device, u64 num_bytes,
813 u64 *start, u64 *len)
815 struct btrfs_key key;
816 struct btrfs_root *root = device->dev_root;
817 struct btrfs_dev_extent *dev_extent;
818 struct btrfs_path *path;
824 u64 search_end = device->total_bytes;
827 struct extent_buffer *l;
829 /* FIXME use last free of some kind */
831 /* we don't want to overwrite the superblock on the drive,
832 * so we make sure to start at an offset of at least 1MB
834 search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
836 max_hole_start = search_start;
839 if (search_start >= search_end) {
844 path = btrfs_alloc_path();
851 key.objectid = device->devid;
852 key.offset = search_start;
853 key.type = BTRFS_DEV_EXTENT_KEY;
855 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
859 ret = btrfs_previous_item(root, path, key.objectid, key.type);
866 slot = path->slots[0];
867 if (slot >= btrfs_header_nritems(l)) {
868 ret = btrfs_next_leaf(root, path);
876 btrfs_item_key_to_cpu(l, &key, slot);
878 if (key.objectid < device->devid)
881 if (key.objectid > device->devid)
884 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
887 if (key.offset > search_start) {
888 hole_size = key.offset - search_start;
890 if (hole_size > max_hole_size) {
891 max_hole_start = search_start;
892 max_hole_size = hole_size;
896 * If this free space is greater than which we need,
897 * it must be the max free space that we have found
898 * until now, so max_hole_start must point to the start
899 * of this free space and the length of this free space
900 * is stored in max_hole_size. Thus, we return
901 * max_hole_start and max_hole_size and go back to the
904 if (hole_size >= num_bytes) {
910 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
911 extent_end = key.offset + btrfs_dev_extent_length(l,
913 if (extent_end > search_start)
914 search_start = extent_end;
920 hole_size = search_end- search_start;
921 if (hole_size > max_hole_size) {
922 max_hole_start = search_start;
923 max_hole_size = hole_size;
927 if (hole_size < num_bytes)
933 btrfs_free_path(path);
935 *start = max_hole_start;
937 *len = max_hole_size;
941 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
942 struct btrfs_device *device,
946 struct btrfs_path *path;
947 struct btrfs_root *root = device->dev_root;
948 struct btrfs_key key;
949 struct btrfs_key found_key;
950 struct extent_buffer *leaf = NULL;
951 struct btrfs_dev_extent *extent = NULL;
953 path = btrfs_alloc_path();
957 key.objectid = device->devid;
959 key.type = BTRFS_DEV_EXTENT_KEY;
961 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
963 ret = btrfs_previous_item(root, path, key.objectid,
964 BTRFS_DEV_EXTENT_KEY);
967 leaf = path->nodes[0];
968 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
969 extent = btrfs_item_ptr(leaf, path->slots[0],
970 struct btrfs_dev_extent);
971 BUG_ON(found_key.offset > start || found_key.offset +
972 btrfs_dev_extent_length(leaf, extent) < start);
973 } else if (ret == 0) {
974 leaf = path->nodes[0];
975 extent = btrfs_item_ptr(leaf, path->slots[0],
976 struct btrfs_dev_extent);
980 if (device->bytes_used > 0)
981 device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
982 ret = btrfs_del_item(trans, root, path);
985 btrfs_free_path(path);
989 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
990 struct btrfs_device *device,
991 u64 chunk_tree, u64 chunk_objectid,
992 u64 chunk_offset, u64 start, u64 num_bytes)
995 struct btrfs_path *path;
996 struct btrfs_root *root = device->dev_root;
997 struct btrfs_dev_extent *extent;
998 struct extent_buffer *leaf;
999 struct btrfs_key key;
1001 WARN_ON(!device->in_fs_metadata);
1002 path = btrfs_alloc_path();
1006 key.objectid = device->devid;
1008 key.type = BTRFS_DEV_EXTENT_KEY;
1009 ret = btrfs_insert_empty_item(trans, root, path, &key,
1013 leaf = path->nodes[0];
1014 extent = btrfs_item_ptr(leaf, path->slots[0],
1015 struct btrfs_dev_extent);
1016 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1017 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1018 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1020 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1021 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
1024 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1025 btrfs_mark_buffer_dirty(leaf);
1026 btrfs_free_path(path);
1030 static noinline int find_next_chunk(struct btrfs_root *root,
1031 u64 objectid, u64 *offset)
1033 struct btrfs_path *path;
1035 struct btrfs_key key;
1036 struct btrfs_chunk *chunk;
1037 struct btrfs_key found_key;
1039 path = btrfs_alloc_path();
1042 key.objectid = objectid;
1043 key.offset = (u64)-1;
1044 key.type = BTRFS_CHUNK_ITEM_KEY;
1046 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1052 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
1056 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1058 if (found_key.objectid != objectid)
1061 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
1062 struct btrfs_chunk);
1063 *offset = found_key.offset +
1064 btrfs_chunk_length(path->nodes[0], chunk);
1069 btrfs_free_path(path);
1073 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
1076 struct btrfs_key key;
1077 struct btrfs_key found_key;
1078 struct btrfs_path *path;
1080 root = root->fs_info->chunk_root;
1082 path = btrfs_alloc_path();
1086 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1087 key.type = BTRFS_DEV_ITEM_KEY;
1088 key.offset = (u64)-1;
1090 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1096 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1097 BTRFS_DEV_ITEM_KEY);
1101 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1103 *objectid = found_key.offset + 1;
1107 btrfs_free_path(path);
1112 * the device information is stored in the chunk root
1113 * the btrfs_device struct should be fully filled in
1115 int btrfs_add_device(struct btrfs_trans_handle *trans,
1116 struct btrfs_root *root,
1117 struct btrfs_device *device)
1120 struct btrfs_path *path;
1121 struct btrfs_dev_item *dev_item;
1122 struct extent_buffer *leaf;
1123 struct btrfs_key key;
1126 root = root->fs_info->chunk_root;
1128 path = btrfs_alloc_path();
1132 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1133 key.type = BTRFS_DEV_ITEM_KEY;
1134 key.offset = device->devid;
1136 ret = btrfs_insert_empty_item(trans, root, path, &key,
1141 leaf = path->nodes[0];
1142 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1144 btrfs_set_device_id(leaf, dev_item, device->devid);
1145 btrfs_set_device_generation(leaf, dev_item, 0);
1146 btrfs_set_device_type(leaf, dev_item, device->type);
1147 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1148 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1149 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1150 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1151 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1152 btrfs_set_device_group(leaf, dev_item, 0);
1153 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1154 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1155 btrfs_set_device_start_offset(leaf, dev_item, 0);
1157 ptr = (unsigned long)btrfs_device_uuid(dev_item);
1158 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1159 ptr = (unsigned long)btrfs_device_fsid(dev_item);
1160 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1161 btrfs_mark_buffer_dirty(leaf);
1165 btrfs_free_path(path);
1169 static int btrfs_rm_dev_item(struct btrfs_root *root,
1170 struct btrfs_device *device)
1173 struct btrfs_path *path;
1174 struct btrfs_key key;
1175 struct btrfs_trans_handle *trans;
1177 root = root->fs_info->chunk_root;
1179 path = btrfs_alloc_path();
1183 trans = btrfs_start_transaction(root, 0);
1184 if (IS_ERR(trans)) {
1185 btrfs_free_path(path);
1186 return PTR_ERR(trans);
1188 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1189 key.type = BTRFS_DEV_ITEM_KEY;
1190 key.offset = device->devid;
1193 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1202 ret = btrfs_del_item(trans, root, path);
1206 btrfs_free_path(path);
1207 unlock_chunks(root);
1208 btrfs_commit_transaction(trans, root);
1212 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1214 struct btrfs_device *device;
1215 struct btrfs_device *next_device;
1216 struct block_device *bdev;
1217 struct buffer_head *bh = NULL;
1218 struct btrfs_super_block *disk_super;
1219 struct btrfs_fs_devices *cur_devices;
1225 bool clear_super = false;
1227 mutex_lock(&uuid_mutex);
1228 mutex_lock(&root->fs_info->volume_mutex);
1230 all_avail = root->fs_info->avail_data_alloc_bits |
1231 root->fs_info->avail_system_alloc_bits |
1232 root->fs_info->avail_metadata_alloc_bits;
1234 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
1235 root->fs_info->fs_devices->num_devices <= 4) {
1236 printk(KERN_ERR "btrfs: unable to go below four devices "
1242 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
1243 root->fs_info->fs_devices->num_devices <= 2) {
1244 printk(KERN_ERR "btrfs: unable to go below two "
1245 "devices on raid1\n");
1250 if (strcmp(device_path, "missing") == 0) {
1251 struct list_head *devices;
1252 struct btrfs_device *tmp;
1255 devices = &root->fs_info->fs_devices->devices;
1257 * It is safe to read the devices since the volume_mutex
1260 list_for_each_entry(tmp, devices, dev_list) {
1261 if (tmp->in_fs_metadata && !tmp->bdev) {
1270 printk(KERN_ERR "btrfs: no missing devices found to "
1275 bdev = blkdev_get_by_path(device_path, FMODE_READ | FMODE_EXCL,
1276 root->fs_info->bdev_holder);
1278 ret = PTR_ERR(bdev);
1282 set_blocksize(bdev, 4096);
1283 bh = btrfs_read_dev_super(bdev);
1288 disk_super = (struct btrfs_super_block *)bh->b_data;
1289 devid = btrfs_stack_device_id(&disk_super->dev_item);
1290 dev_uuid = disk_super->dev_item.uuid;
1291 device = btrfs_find_device(root, devid, dev_uuid,
1299 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1300 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1306 if (device->writeable) {
1308 list_del_init(&device->dev_alloc_list);
1309 unlock_chunks(root);
1310 root->fs_info->fs_devices->rw_devices--;
1314 ret = btrfs_shrink_device(device, 0);
1318 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1322 device->in_fs_metadata = 0;
1323 btrfs_scrub_cancel_dev(root, device);
1326 * the device list mutex makes sure that we don't change
1327 * the device list while someone else is writing out all
1328 * the device supers.
1331 cur_devices = device->fs_devices;
1332 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1333 list_del_rcu(&device->dev_list);
1335 device->fs_devices->num_devices--;
1337 if (device->missing)
1338 root->fs_info->fs_devices->missing_devices--;
1340 next_device = list_entry(root->fs_info->fs_devices->devices.next,
1341 struct btrfs_device, dev_list);
1342 if (device->bdev == root->fs_info->sb->s_bdev)
1343 root->fs_info->sb->s_bdev = next_device->bdev;
1344 if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1345 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1348 device->fs_devices->open_devices--;
1350 call_rcu(&device->rcu, free_device);
1351 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1353 num_devices = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1354 btrfs_set_super_num_devices(&root->fs_info->super_copy, num_devices);
1356 if (cur_devices->open_devices == 0) {
1357 struct btrfs_fs_devices *fs_devices;
1358 fs_devices = root->fs_info->fs_devices;
1359 while (fs_devices) {
1360 if (fs_devices->seed == cur_devices)
1362 fs_devices = fs_devices->seed;
1364 fs_devices->seed = cur_devices->seed;
1365 cur_devices->seed = NULL;
1367 __btrfs_close_devices(cur_devices);
1368 unlock_chunks(root);
1369 free_fs_devices(cur_devices);
1373 * at this point, the device is zero sized. We want to
1374 * remove it from the devices list and zero out the old super
1377 /* make sure this device isn't detected as part of
1380 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1381 set_buffer_dirty(bh);
1382 sync_dirty_buffer(bh);
1391 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1393 mutex_unlock(&root->fs_info->volume_mutex);
1394 mutex_unlock(&uuid_mutex);
1397 if (device->writeable) {
1399 list_add(&device->dev_alloc_list,
1400 &root->fs_info->fs_devices->alloc_list);
1401 unlock_chunks(root);
1402 root->fs_info->fs_devices->rw_devices++;
1408 * does all the dirty work required for changing file system's UUID.
1410 static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
1411 struct btrfs_root *root)
1413 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1414 struct btrfs_fs_devices *old_devices;
1415 struct btrfs_fs_devices *seed_devices;
1416 struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
1417 struct btrfs_device *device;
1420 BUG_ON(!mutex_is_locked(&uuid_mutex));
1421 if (!fs_devices->seeding)
1424 seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1428 old_devices = clone_fs_devices(fs_devices);
1429 if (IS_ERR(old_devices)) {
1430 kfree(seed_devices);
1431 return PTR_ERR(old_devices);
1434 list_add(&old_devices->list, &fs_uuids);
1436 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1437 seed_devices->opened = 1;
1438 INIT_LIST_HEAD(&seed_devices->devices);
1439 INIT_LIST_HEAD(&seed_devices->alloc_list);
1440 mutex_init(&seed_devices->device_list_mutex);
1442 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1443 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1445 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1447 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1448 list_for_each_entry(device, &seed_devices->devices, dev_list) {
1449 device->fs_devices = seed_devices;
1452 fs_devices->seeding = 0;
1453 fs_devices->num_devices = 0;
1454 fs_devices->open_devices = 0;
1455 fs_devices->seed = seed_devices;
1457 generate_random_uuid(fs_devices->fsid);
1458 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1459 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1460 super_flags = btrfs_super_flags(disk_super) &
1461 ~BTRFS_SUPER_FLAG_SEEDING;
1462 btrfs_set_super_flags(disk_super, super_flags);
1468 * strore the expected generation for seed devices in device items.
1470 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1471 struct btrfs_root *root)
1473 struct btrfs_path *path;
1474 struct extent_buffer *leaf;
1475 struct btrfs_dev_item *dev_item;
1476 struct btrfs_device *device;
1477 struct btrfs_key key;
1478 u8 fs_uuid[BTRFS_UUID_SIZE];
1479 u8 dev_uuid[BTRFS_UUID_SIZE];
1483 path = btrfs_alloc_path();
1487 root = root->fs_info->chunk_root;
1488 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1490 key.type = BTRFS_DEV_ITEM_KEY;
1493 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1497 leaf = path->nodes[0];
1499 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1500 ret = btrfs_next_leaf(root, path);
1505 leaf = path->nodes[0];
1506 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1507 btrfs_release_path(path);
1511 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1512 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1513 key.type != BTRFS_DEV_ITEM_KEY)
1516 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1517 struct btrfs_dev_item);
1518 devid = btrfs_device_id(leaf, dev_item);
1519 read_extent_buffer(leaf, dev_uuid,
1520 (unsigned long)btrfs_device_uuid(dev_item),
1522 read_extent_buffer(leaf, fs_uuid,
1523 (unsigned long)btrfs_device_fsid(dev_item),
1525 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1528 if (device->fs_devices->seeding) {
1529 btrfs_set_device_generation(leaf, dev_item,
1530 device->generation);
1531 btrfs_mark_buffer_dirty(leaf);
1539 btrfs_free_path(path);
1543 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1545 struct btrfs_trans_handle *trans;
1546 struct btrfs_device *device;
1547 struct block_device *bdev;
1548 struct list_head *devices;
1549 struct super_block *sb = root->fs_info->sb;
1551 int seeding_dev = 0;
1554 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1557 bdev = blkdev_get_by_path(device_path, FMODE_EXCL,
1558 root->fs_info->bdev_holder);
1560 return PTR_ERR(bdev);
1562 if (root->fs_info->fs_devices->seeding) {
1564 down_write(&sb->s_umount);
1565 mutex_lock(&uuid_mutex);
1568 filemap_write_and_wait(bdev->bd_inode->i_mapping);
1569 mutex_lock(&root->fs_info->volume_mutex);
1571 devices = &root->fs_info->fs_devices->devices;
1573 * we have the volume lock, so we don't need the extra
1574 * device list mutex while reading the list here.
1576 list_for_each_entry(device, devices, dev_list) {
1577 if (device->bdev == bdev) {
1583 device = kzalloc(sizeof(*device), GFP_NOFS);
1585 /* we can safely leave the fs_devices entry around */
1590 device->name = kstrdup(device_path, GFP_NOFS);
1591 if (!device->name) {
1597 ret = find_next_devid(root, &device->devid);
1599 kfree(device->name);
1604 trans = btrfs_start_transaction(root, 0);
1605 if (IS_ERR(trans)) {
1606 kfree(device->name);
1608 ret = PTR_ERR(trans);
1614 device->writeable = 1;
1615 device->work.func = pending_bios_fn;
1616 generate_random_uuid(device->uuid);
1617 spin_lock_init(&device->io_lock);
1618 device->generation = trans->transid;
1619 device->io_width = root->sectorsize;
1620 device->io_align = root->sectorsize;
1621 device->sector_size = root->sectorsize;
1622 device->total_bytes = i_size_read(bdev->bd_inode);
1623 device->disk_total_bytes = device->total_bytes;
1624 device->dev_root = root->fs_info->dev_root;
1625 device->bdev = bdev;
1626 device->in_fs_metadata = 1;
1627 device->mode = FMODE_EXCL;
1628 set_blocksize(device->bdev, 4096);
1631 sb->s_flags &= ~MS_RDONLY;
1632 ret = btrfs_prepare_sprout(trans, root);
1636 device->fs_devices = root->fs_info->fs_devices;
1639 * we don't want write_supers to jump in here with our device
1642 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1643 list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
1644 list_add(&device->dev_alloc_list,
1645 &root->fs_info->fs_devices->alloc_list);
1646 root->fs_info->fs_devices->num_devices++;
1647 root->fs_info->fs_devices->open_devices++;
1648 root->fs_info->fs_devices->rw_devices++;
1649 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1651 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1652 root->fs_info->fs_devices->rotating = 1;
1654 total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
1655 btrfs_set_super_total_bytes(&root->fs_info->super_copy,
1656 total_bytes + device->total_bytes);
1658 total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
1659 btrfs_set_super_num_devices(&root->fs_info->super_copy,
1661 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1664 ret = init_first_rw_device(trans, root, device);
1666 ret = btrfs_finish_sprout(trans, root);
1669 ret = btrfs_add_device(trans, root, device);
1673 * we've got more storage, clear any full flags on the space
1676 btrfs_clear_space_info_full(root->fs_info);
1678 unlock_chunks(root);
1679 btrfs_commit_transaction(trans, root);
1682 mutex_unlock(&uuid_mutex);
1683 up_write(&sb->s_umount);
1685 ret = btrfs_relocate_sys_chunks(root);
1689 mutex_unlock(&root->fs_info->volume_mutex);
1692 blkdev_put(bdev, FMODE_EXCL);
1694 mutex_unlock(&uuid_mutex);
1695 up_write(&sb->s_umount);
1700 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1701 struct btrfs_device *device)
1704 struct btrfs_path *path;
1705 struct btrfs_root *root;
1706 struct btrfs_dev_item *dev_item;
1707 struct extent_buffer *leaf;
1708 struct btrfs_key key;
1710 root = device->dev_root->fs_info->chunk_root;
1712 path = btrfs_alloc_path();
1716 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1717 key.type = BTRFS_DEV_ITEM_KEY;
1718 key.offset = device->devid;
1720 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1729 leaf = path->nodes[0];
1730 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1732 btrfs_set_device_id(leaf, dev_item, device->devid);
1733 btrfs_set_device_type(leaf, dev_item, device->type);
1734 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1735 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1736 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1737 btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
1738 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1739 btrfs_mark_buffer_dirty(leaf);
1742 btrfs_free_path(path);
1746 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1747 struct btrfs_device *device, u64 new_size)
1749 struct btrfs_super_block *super_copy =
1750 &device->dev_root->fs_info->super_copy;
1751 u64 old_total = btrfs_super_total_bytes(super_copy);
1752 u64 diff = new_size - device->total_bytes;
1754 if (!device->writeable)
1756 if (new_size <= device->total_bytes)
1759 btrfs_set_super_total_bytes(super_copy, old_total + diff);
1760 device->fs_devices->total_rw_bytes += diff;
1762 device->total_bytes = new_size;
1763 device->disk_total_bytes = new_size;
1764 btrfs_clear_space_info_full(device->dev_root->fs_info);
1766 return btrfs_update_device(trans, device);
1769 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1770 struct btrfs_device *device, u64 new_size)
1773 lock_chunks(device->dev_root);
1774 ret = __btrfs_grow_device(trans, device, new_size);
1775 unlock_chunks(device->dev_root);
1779 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1780 struct btrfs_root *root,
1781 u64 chunk_tree, u64 chunk_objectid,
1785 struct btrfs_path *path;
1786 struct btrfs_key key;
1788 root = root->fs_info->chunk_root;
1789 path = btrfs_alloc_path();
1793 key.objectid = chunk_objectid;
1794 key.offset = chunk_offset;
1795 key.type = BTRFS_CHUNK_ITEM_KEY;
1797 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1800 ret = btrfs_del_item(trans, root, path);
1802 btrfs_free_path(path);
1806 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1809 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1810 struct btrfs_disk_key *disk_key;
1811 struct btrfs_chunk *chunk;
1818 struct btrfs_key key;
1820 array_size = btrfs_super_sys_array_size(super_copy);
1822 ptr = super_copy->sys_chunk_array;
1825 while (cur < array_size) {
1826 disk_key = (struct btrfs_disk_key *)ptr;
1827 btrfs_disk_key_to_cpu(&key, disk_key);
1829 len = sizeof(*disk_key);
1831 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1832 chunk = (struct btrfs_chunk *)(ptr + len);
1833 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1834 len += btrfs_chunk_item_size(num_stripes);
1839 if (key.objectid == chunk_objectid &&
1840 key.offset == chunk_offset) {
1841 memmove(ptr, ptr + len, array_size - (cur + len));
1843 btrfs_set_super_sys_array_size(super_copy, array_size);
1852 static int btrfs_relocate_chunk(struct btrfs_root *root,
1853 u64 chunk_tree, u64 chunk_objectid,
1856 struct extent_map_tree *em_tree;
1857 struct btrfs_root *extent_root;
1858 struct btrfs_trans_handle *trans;
1859 struct extent_map *em;
1860 struct map_lookup *map;
1864 root = root->fs_info->chunk_root;
1865 extent_root = root->fs_info->extent_root;
1866 em_tree = &root->fs_info->mapping_tree.map_tree;
1868 ret = btrfs_can_relocate(extent_root, chunk_offset);
1872 /* step one, relocate all the extents inside this chunk */
1873 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
1877 trans = btrfs_start_transaction(root, 0);
1878 BUG_ON(IS_ERR(trans));
1883 * step two, delete the device extents and the
1884 * chunk tree entries
1886 read_lock(&em_tree->lock);
1887 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1888 read_unlock(&em_tree->lock);
1890 BUG_ON(em->start > chunk_offset ||
1891 em->start + em->len < chunk_offset);
1892 map = (struct map_lookup *)em->bdev;
1894 for (i = 0; i < map->num_stripes; i++) {
1895 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1896 map->stripes[i].physical);
1899 if (map->stripes[i].dev) {
1900 ret = btrfs_update_device(trans, map->stripes[i].dev);
1904 ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1909 trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
1911 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1912 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1916 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1919 write_lock(&em_tree->lock);
1920 remove_extent_mapping(em_tree, em);
1921 write_unlock(&em_tree->lock);
1926 /* once for the tree */
1927 free_extent_map(em);
1929 free_extent_map(em);
1931 unlock_chunks(root);
1932 btrfs_end_transaction(trans, root);
1936 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
1938 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
1939 struct btrfs_path *path;
1940 struct extent_buffer *leaf;
1941 struct btrfs_chunk *chunk;
1942 struct btrfs_key key;
1943 struct btrfs_key found_key;
1944 u64 chunk_tree = chunk_root->root_key.objectid;
1946 bool retried = false;
1950 path = btrfs_alloc_path();
1955 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1956 key.offset = (u64)-1;
1957 key.type = BTRFS_CHUNK_ITEM_KEY;
1960 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1965 ret = btrfs_previous_item(chunk_root, path, key.objectid,
1972 leaf = path->nodes[0];
1973 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1975 chunk = btrfs_item_ptr(leaf, path->slots[0],
1976 struct btrfs_chunk);
1977 chunk_type = btrfs_chunk_type(leaf, chunk);
1978 btrfs_release_path(path);
1980 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
1981 ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
1990 if (found_key.offset == 0)
1992 key.offset = found_key.offset - 1;
1995 if (failed && !retried) {
1999 } else if (failed && retried) {
2004 btrfs_free_path(path);
2008 static u64 div_factor(u64 num, int factor)
2017 int btrfs_balance(struct btrfs_root *dev_root)
2020 struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
2021 struct btrfs_device *device;
2024 struct btrfs_path *path;
2025 struct btrfs_key key;
2026 struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
2027 struct btrfs_trans_handle *trans;
2028 struct btrfs_key found_key;
2030 if (dev_root->fs_info->sb->s_flags & MS_RDONLY)
2033 if (!capable(CAP_SYS_ADMIN))
2036 mutex_lock(&dev_root->fs_info->volume_mutex);
2037 dev_root = dev_root->fs_info->dev_root;
2039 /* step one make some room on all the devices */
2040 list_for_each_entry(device, devices, dev_list) {
2041 old_size = device->total_bytes;
2042 size_to_free = div_factor(old_size, 1);
2043 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2044 if (!device->writeable ||
2045 device->total_bytes - device->bytes_used > size_to_free)
2048 ret = btrfs_shrink_device(device, old_size - size_to_free);
2053 trans = btrfs_start_transaction(dev_root, 0);
2054 BUG_ON(IS_ERR(trans));
2056 ret = btrfs_grow_device(trans, device, old_size);
2059 btrfs_end_transaction(trans, dev_root);
2062 /* step two, relocate all the chunks */
2063 path = btrfs_alloc_path();
2066 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2067 key.offset = (u64)-1;
2068 key.type = BTRFS_CHUNK_ITEM_KEY;
2071 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2076 * this shouldn't happen, it means the last relocate
2082 ret = btrfs_previous_item(chunk_root, path, 0,
2083 BTRFS_CHUNK_ITEM_KEY);
2087 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2089 if (found_key.objectid != key.objectid)
2092 /* chunk zero is special */
2093 if (found_key.offset == 0)
2096 btrfs_release_path(path);
2097 ret = btrfs_relocate_chunk(chunk_root,
2098 chunk_root->root_key.objectid,
2101 BUG_ON(ret && ret != -ENOSPC);
2102 key.offset = found_key.offset - 1;
2106 btrfs_free_path(path);
2107 mutex_unlock(&dev_root->fs_info->volume_mutex);
2112 * shrinking a device means finding all of the device extents past
2113 * the new size, and then following the back refs to the chunks.
2114 * The chunk relocation code actually frees the device extent
2116 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
2118 struct btrfs_trans_handle *trans;
2119 struct btrfs_root *root = device->dev_root;
2120 struct btrfs_dev_extent *dev_extent = NULL;
2121 struct btrfs_path *path;
2129 bool retried = false;
2130 struct extent_buffer *l;
2131 struct btrfs_key key;
2132 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2133 u64 old_total = btrfs_super_total_bytes(super_copy);
2134 u64 old_size = device->total_bytes;
2135 u64 diff = device->total_bytes - new_size;
2137 if (new_size >= device->total_bytes)
2140 path = btrfs_alloc_path();
2148 device->total_bytes = new_size;
2149 if (device->writeable)
2150 device->fs_devices->total_rw_bytes -= diff;
2151 unlock_chunks(root);
2154 key.objectid = device->devid;
2155 key.offset = (u64)-1;
2156 key.type = BTRFS_DEV_EXTENT_KEY;
2159 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2163 ret = btrfs_previous_item(root, path, 0, key.type);
2168 btrfs_release_path(path);
2173 slot = path->slots[0];
2174 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
2176 if (key.objectid != device->devid) {
2177 btrfs_release_path(path);
2181 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
2182 length = btrfs_dev_extent_length(l, dev_extent);
2184 if (key.offset + length <= new_size) {
2185 btrfs_release_path(path);
2189 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
2190 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
2191 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
2192 btrfs_release_path(path);
2194 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
2196 if (ret && ret != -ENOSPC)
2203 if (failed && !retried) {
2207 } else if (failed && retried) {
2211 device->total_bytes = old_size;
2212 if (device->writeable)
2213 device->fs_devices->total_rw_bytes += diff;
2214 unlock_chunks(root);
2218 /* Shrinking succeeded, else we would be at "done". */
2219 trans = btrfs_start_transaction(root, 0);
2220 if (IS_ERR(trans)) {
2221 ret = PTR_ERR(trans);
2227 device->disk_total_bytes = new_size;
2228 /* Now btrfs_update_device() will change the on-disk size. */
2229 ret = btrfs_update_device(trans, device);
2231 unlock_chunks(root);
2232 btrfs_end_transaction(trans, root);
2235 WARN_ON(diff > old_total);
2236 btrfs_set_super_total_bytes(super_copy, old_total - diff);
2237 unlock_chunks(root);
2238 btrfs_end_transaction(trans, root);
2240 btrfs_free_path(path);
2244 static int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
2245 struct btrfs_root *root,
2246 struct btrfs_key *key,
2247 struct btrfs_chunk *chunk, int item_size)
2249 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2250 struct btrfs_disk_key disk_key;
2254 array_size = btrfs_super_sys_array_size(super_copy);
2255 if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
2258 ptr = super_copy->sys_chunk_array + array_size;
2259 btrfs_cpu_key_to_disk(&disk_key, key);
2260 memcpy(ptr, &disk_key, sizeof(disk_key));
2261 ptr += sizeof(disk_key);
2262 memcpy(ptr, chunk, item_size);
2263 item_size += sizeof(disk_key);
2264 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
2269 * sort the devices in descending order by max_avail, total_avail
2271 static int btrfs_cmp_device_info(const void *a, const void *b)
2273 const struct btrfs_device_info *di_a = a;
2274 const struct btrfs_device_info *di_b = b;
2276 if (di_a->max_avail > di_b->max_avail)
2278 if (di_a->max_avail < di_b->max_avail)
2280 if (di_a->total_avail > di_b->total_avail)
2282 if (di_a->total_avail < di_b->total_avail)
2287 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2288 struct btrfs_root *extent_root,
2289 struct map_lookup **map_ret,
2290 u64 *num_bytes_out, u64 *stripe_size_out,
2291 u64 start, u64 type)
2293 struct btrfs_fs_info *info = extent_root->fs_info;
2294 struct btrfs_fs_devices *fs_devices = info->fs_devices;
2295 struct list_head *cur;
2296 struct map_lookup *map = NULL;
2297 struct extent_map_tree *em_tree;
2298 struct extent_map *em;
2299 struct btrfs_device_info *devices_info = NULL;
2301 int num_stripes; /* total number of stripes to allocate */
2302 int sub_stripes; /* sub_stripes info for map */
2303 int dev_stripes; /* stripes per dev */
2304 int devs_max; /* max devs to use */
2305 int devs_min; /* min devs needed */
2306 int devs_increment; /* ndevs has to be a multiple of this */
2307 int ncopies; /* how many copies to data has */
2309 u64 max_stripe_size;
2317 if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
2318 (type & BTRFS_BLOCK_GROUP_DUP)) {
2320 type &= ~BTRFS_BLOCK_GROUP_DUP;
2323 if (list_empty(&fs_devices->alloc_list))
2330 devs_max = 0; /* 0 == as many as possible */
2334 * define the properties of each RAID type.
2335 * FIXME: move this to a global table and use it in all RAID
2338 if (type & (BTRFS_BLOCK_GROUP_DUP)) {
2342 } else if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
2344 } else if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
2349 } else if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2358 if (type & BTRFS_BLOCK_GROUP_DATA) {
2359 max_stripe_size = 1024 * 1024 * 1024;
2360 max_chunk_size = 10 * max_stripe_size;
2361 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
2362 max_stripe_size = 256 * 1024 * 1024;
2363 max_chunk_size = max_stripe_size;
2364 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
2365 max_stripe_size = 8 * 1024 * 1024;
2366 max_chunk_size = 2 * max_stripe_size;
2368 printk(KERN_ERR "btrfs: invalid chunk type 0x%llx requested\n",
2373 /* we don't want a chunk larger than 10% of writeable space */
2374 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
2377 devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
2382 cur = fs_devices->alloc_list.next;
2385 * in the first pass through the devices list, we gather information
2386 * about the available holes on each device.
2389 while (cur != &fs_devices->alloc_list) {
2390 struct btrfs_device *device;
2394 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
2398 if (!device->writeable) {
2400 "btrfs: read-only device in alloc_list\n");
2405 if (!device->in_fs_metadata)
2408 if (device->total_bytes > device->bytes_used)
2409 total_avail = device->total_bytes - device->bytes_used;
2412 /* avail is off by max(alloc_start, 1MB), but that is the same
2413 * for all devices, so it doesn't hurt the sorting later on
2416 ret = find_free_dev_extent(trans, device,
2417 max_stripe_size * dev_stripes,
2418 &dev_offset, &max_avail);
2419 if (ret && ret != -ENOSPC)
2423 max_avail = max_stripe_size * dev_stripes;
2425 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
2428 devices_info[ndevs].dev_offset = dev_offset;
2429 devices_info[ndevs].max_avail = max_avail;
2430 devices_info[ndevs].total_avail = total_avail;
2431 devices_info[ndevs].dev = device;
2436 * now sort the devices by hole size / available space
2438 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
2439 btrfs_cmp_device_info, NULL);
2441 /* round down to number of usable stripes */
2442 ndevs -= ndevs % devs_increment;
2444 if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
2449 if (devs_max && ndevs > devs_max)
2452 * the primary goal is to maximize the number of stripes, so use as many
2453 * devices as possible, even if the stripes are not maximum sized.
2455 stripe_size = devices_info[ndevs-1].max_avail;
2456 num_stripes = ndevs * dev_stripes;
2458 if (stripe_size * num_stripes > max_chunk_size * ncopies) {
2459 stripe_size = max_chunk_size * ncopies;
2460 do_div(stripe_size, num_stripes);
2463 do_div(stripe_size, dev_stripes);
2464 do_div(stripe_size, BTRFS_STRIPE_LEN);
2465 stripe_size *= BTRFS_STRIPE_LEN;
2467 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2472 map->num_stripes = num_stripes;
2474 for (i = 0; i < ndevs; ++i) {
2475 for (j = 0; j < dev_stripes; ++j) {
2476 int s = i * dev_stripes + j;
2477 map->stripes[s].dev = devices_info[i].dev;
2478 map->stripes[s].physical = devices_info[i].dev_offset +
2482 map->sector_size = extent_root->sectorsize;
2483 map->stripe_len = BTRFS_STRIPE_LEN;
2484 map->io_align = BTRFS_STRIPE_LEN;
2485 map->io_width = BTRFS_STRIPE_LEN;
2487 map->sub_stripes = sub_stripes;
2490 num_bytes = stripe_size * (num_stripes / ncopies);
2492 *stripe_size_out = stripe_size;
2493 *num_bytes_out = num_bytes;
2495 trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
2497 em = alloc_extent_map();
2502 em->bdev = (struct block_device *)map;
2504 em->len = num_bytes;
2505 em->block_start = 0;
2506 em->block_len = em->len;
2508 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
2509 write_lock(&em_tree->lock);
2510 ret = add_extent_mapping(em_tree, em);
2511 write_unlock(&em_tree->lock);
2513 free_extent_map(em);
2515 ret = btrfs_make_block_group(trans, extent_root, 0, type,
2516 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2520 for (i = 0; i < map->num_stripes; ++i) {
2521 struct btrfs_device *device;
2524 device = map->stripes[i].dev;
2525 dev_offset = map->stripes[i].physical;
2527 ret = btrfs_alloc_dev_extent(trans, device,
2528 info->chunk_root->root_key.objectid,
2529 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2530 start, dev_offset, stripe_size);
2534 kfree(devices_info);
2539 kfree(devices_info);
2543 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
2544 struct btrfs_root *extent_root,
2545 struct map_lookup *map, u64 chunk_offset,
2546 u64 chunk_size, u64 stripe_size)
2549 struct btrfs_key key;
2550 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2551 struct btrfs_device *device;
2552 struct btrfs_chunk *chunk;
2553 struct btrfs_stripe *stripe;
2554 size_t item_size = btrfs_chunk_item_size(map->num_stripes);
2558 chunk = kzalloc(item_size, GFP_NOFS);
2563 while (index < map->num_stripes) {
2564 device = map->stripes[index].dev;
2565 device->bytes_used += stripe_size;
2566 ret = btrfs_update_device(trans, device);
2572 stripe = &chunk->stripe;
2573 while (index < map->num_stripes) {
2574 device = map->stripes[index].dev;
2575 dev_offset = map->stripes[index].physical;
2577 btrfs_set_stack_stripe_devid(stripe, device->devid);
2578 btrfs_set_stack_stripe_offset(stripe, dev_offset);
2579 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
2584 btrfs_set_stack_chunk_length(chunk, chunk_size);
2585 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
2586 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
2587 btrfs_set_stack_chunk_type(chunk, map->type);
2588 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
2589 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
2590 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
2591 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
2592 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
2594 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2595 key.type = BTRFS_CHUNK_ITEM_KEY;
2596 key.offset = chunk_offset;
2598 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
2601 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2602 ret = btrfs_add_system_chunk(trans, chunk_root, &key, chunk,
2612 * Chunk allocation falls into two parts. The first part does works
2613 * that make the new allocated chunk useable, but not do any operation
2614 * that modifies the chunk tree. The second part does the works that
2615 * require modifying the chunk tree. This division is important for the
2616 * bootstrap process of adding storage to a seed btrfs.
2618 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2619 struct btrfs_root *extent_root, u64 type)
2624 struct map_lookup *map;
2625 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2628 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2633 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2634 &stripe_size, chunk_offset, type);
2638 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2639 chunk_size, stripe_size);
2644 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
2645 struct btrfs_root *root,
2646 struct btrfs_device *device)
2649 u64 sys_chunk_offset;
2653 u64 sys_stripe_size;
2655 struct map_lookup *map;
2656 struct map_lookup *sys_map;
2657 struct btrfs_fs_info *fs_info = root->fs_info;
2658 struct btrfs_root *extent_root = fs_info->extent_root;
2661 ret = find_next_chunk(fs_info->chunk_root,
2662 BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
2665 alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
2666 (fs_info->metadata_alloc_profile &
2667 fs_info->avail_metadata_alloc_bits);
2668 alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2670 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2671 &stripe_size, chunk_offset, alloc_profile);
2674 sys_chunk_offset = chunk_offset + chunk_size;
2676 alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
2677 (fs_info->system_alloc_profile &
2678 fs_info->avail_system_alloc_bits);
2679 alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2681 ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
2682 &sys_chunk_size, &sys_stripe_size,
2683 sys_chunk_offset, alloc_profile);
2686 ret = btrfs_add_device(trans, fs_info->chunk_root, device);
2690 * Modifying chunk tree needs allocating new blocks from both
2691 * system block group and metadata block group. So we only can
2692 * do operations require modifying the chunk tree after both
2693 * block groups were created.
2695 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2696 chunk_size, stripe_size);
2699 ret = __finish_chunk_alloc(trans, extent_root, sys_map,
2700 sys_chunk_offset, sys_chunk_size,
2706 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
2708 struct extent_map *em;
2709 struct map_lookup *map;
2710 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2714 read_lock(&map_tree->map_tree.lock);
2715 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
2716 read_unlock(&map_tree->map_tree.lock);
2720 if (btrfs_test_opt(root, DEGRADED)) {
2721 free_extent_map(em);
2725 map = (struct map_lookup *)em->bdev;
2726 for (i = 0; i < map->num_stripes; i++) {
2727 if (!map->stripes[i].dev->writeable) {
2732 free_extent_map(em);
2736 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
2738 extent_map_tree_init(&tree->map_tree);
2741 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
2743 struct extent_map *em;
2746 write_lock(&tree->map_tree.lock);
2747 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
2749 remove_extent_mapping(&tree->map_tree, em);
2750 write_unlock(&tree->map_tree.lock);
2755 free_extent_map(em);
2756 /* once for the tree */
2757 free_extent_map(em);
2761 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
2763 struct extent_map *em;
2764 struct map_lookup *map;
2765 struct extent_map_tree *em_tree = &map_tree->map_tree;
2768 read_lock(&em_tree->lock);
2769 em = lookup_extent_mapping(em_tree, logical, len);
2770 read_unlock(&em_tree->lock);
2773 BUG_ON(em->start > logical || em->start + em->len < logical);
2774 map = (struct map_lookup *)em->bdev;
2775 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
2776 ret = map->num_stripes;
2777 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2778 ret = map->sub_stripes;
2781 free_extent_map(em);
2785 static int find_live_mirror(struct map_lookup *map, int first, int num,
2789 if (map->stripes[optimal].dev->bdev)
2791 for (i = first; i < first + num; i++) {
2792 if (map->stripes[i].dev->bdev)
2795 /* we couldn't find one that doesn't fail. Just return something
2796 * and the io error handling code will clean up eventually
2801 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2802 u64 logical, u64 *length,
2803 struct btrfs_multi_bio **multi_ret,
2806 struct extent_map *em;
2807 struct map_lookup *map;
2808 struct extent_map_tree *em_tree = &map_tree->map_tree;
2811 u64 stripe_end_offset;
2815 int stripes_allocated = 8;
2816 int stripes_required = 1;
2821 struct btrfs_multi_bio *multi = NULL;
2823 if (multi_ret && !(rw & (REQ_WRITE | REQ_DISCARD)))
2824 stripes_allocated = 1;
2827 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
2832 atomic_set(&multi->error, 0);
2835 read_lock(&em_tree->lock);
2836 em = lookup_extent_mapping(em_tree, logical, *length);
2837 read_unlock(&em_tree->lock);
2840 printk(KERN_CRIT "unable to find logical %llu len %llu\n",
2841 (unsigned long long)logical,
2842 (unsigned long long)*length);
2846 BUG_ON(em->start > logical || em->start + em->len < logical);
2847 map = (struct map_lookup *)em->bdev;
2848 offset = logical - em->start;
2850 if (mirror_num > map->num_stripes)
2853 /* if our multi bio struct is too small, back off and try again */
2854 if (rw & REQ_WRITE) {
2855 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
2856 BTRFS_BLOCK_GROUP_DUP)) {
2857 stripes_required = map->num_stripes;
2859 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2860 stripes_required = map->sub_stripes;
2864 if (rw & REQ_DISCARD) {
2865 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
2866 BTRFS_BLOCK_GROUP_RAID1 |
2867 BTRFS_BLOCK_GROUP_DUP |
2868 BTRFS_BLOCK_GROUP_RAID10)) {
2869 stripes_required = map->num_stripes;
2872 if (multi_ret && (rw & (REQ_WRITE | REQ_DISCARD)) &&
2873 stripes_allocated < stripes_required) {
2874 stripes_allocated = map->num_stripes;
2875 free_extent_map(em);
2881 * stripe_nr counts the total number of stripes we have to stride
2882 * to get to this block
2884 do_div(stripe_nr, map->stripe_len);
2886 stripe_offset = stripe_nr * map->stripe_len;
2887 BUG_ON(offset < stripe_offset);
2889 /* stripe_offset is the offset of this block in its stripe*/
2890 stripe_offset = offset - stripe_offset;
2892 if (rw & REQ_DISCARD)
2893 *length = min_t(u64, em->len - offset, *length);
2894 else if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
2895 BTRFS_BLOCK_GROUP_RAID1 |
2896 BTRFS_BLOCK_GROUP_RAID10 |
2897 BTRFS_BLOCK_GROUP_DUP)) {
2898 /* we limit the length of each bio to what fits in a stripe */
2899 *length = min_t(u64, em->len - offset,
2900 map->stripe_len - stripe_offset);
2902 *length = em->len - offset;
2910 stripe_nr_orig = stripe_nr;
2911 stripe_nr_end = (offset + *length + map->stripe_len - 1) &
2912 (~(map->stripe_len - 1));
2913 do_div(stripe_nr_end, map->stripe_len);
2914 stripe_end_offset = stripe_nr_end * map->stripe_len -
2916 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
2917 if (rw & REQ_DISCARD)
2918 num_stripes = min_t(u64, map->num_stripes,
2919 stripe_nr_end - stripe_nr_orig);
2920 stripe_index = do_div(stripe_nr, map->num_stripes);
2921 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
2922 if (rw & (REQ_WRITE | REQ_DISCARD))
2923 num_stripes = map->num_stripes;
2924 else if (mirror_num)
2925 stripe_index = mirror_num - 1;
2927 stripe_index = find_live_mirror(map, 0,
2929 current->pid % map->num_stripes);
2932 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
2933 if (rw & (REQ_WRITE | REQ_DISCARD))
2934 num_stripes = map->num_stripes;
2935 else if (mirror_num)
2936 stripe_index = mirror_num - 1;
2938 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2939 int factor = map->num_stripes / map->sub_stripes;
2941 stripe_index = do_div(stripe_nr, factor);
2942 stripe_index *= map->sub_stripes;
2945 num_stripes = map->sub_stripes;
2946 else if (rw & REQ_DISCARD)
2947 num_stripes = min_t(u64, map->sub_stripes *
2948 (stripe_nr_end - stripe_nr_orig),
2950 else if (mirror_num)
2951 stripe_index += mirror_num - 1;
2953 stripe_index = find_live_mirror(map, stripe_index,
2954 map->sub_stripes, stripe_index +
2955 current->pid % map->sub_stripes);
2959 * after this do_div call, stripe_nr is the number of stripes
2960 * on this device we have to walk to find the data, and
2961 * stripe_index is the number of our device in the stripe array
2963 stripe_index = do_div(stripe_nr, map->num_stripes);
2965 BUG_ON(stripe_index >= map->num_stripes);
2967 if (rw & REQ_DISCARD) {
2968 for (i = 0; i < num_stripes; i++) {
2969 multi->stripes[i].physical =
2970 map->stripes[stripe_index].physical +
2971 stripe_offset + stripe_nr * map->stripe_len;
2972 multi->stripes[i].dev = map->stripes[stripe_index].dev;
2974 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
2976 u32 last_stripe = 0;
2979 div_u64_rem(stripe_nr_end - 1,
2983 for (j = 0; j < map->num_stripes; j++) {
2986 div_u64_rem(stripe_nr_end - 1 - j,
2987 map->num_stripes, &test);
2988 if (test == stripe_index)
2991 stripes = stripe_nr_end - 1 - j;
2992 do_div(stripes, map->num_stripes);
2993 multi->stripes[i].length = map->stripe_len *
2994 (stripes - stripe_nr + 1);
2997 multi->stripes[i].length -=
3001 if (stripe_index == last_stripe)
3002 multi->stripes[i].length -=
3004 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3007 int factor = map->num_stripes /
3009 u32 last_stripe = 0;
3011 div_u64_rem(stripe_nr_end - 1,
3012 factor, &last_stripe);
3013 last_stripe *= map->sub_stripes;
3015 for (j = 0; j < factor; j++) {
3018 div_u64_rem(stripe_nr_end - 1 - j,
3022 stripe_index / map->sub_stripes)
3025 stripes = stripe_nr_end - 1 - j;
3026 do_div(stripes, factor);
3027 multi->stripes[i].length = map->stripe_len *
3028 (stripes - stripe_nr + 1);
3030 if (i < map->sub_stripes) {
3031 multi->stripes[i].length -=
3033 if (i == map->sub_stripes - 1)
3036 if (stripe_index >= last_stripe &&
3037 stripe_index <= (last_stripe +
3038 map->sub_stripes - 1)) {
3039 multi->stripes[i].length -=
3043 multi->stripes[i].length = *length;
3046 if (stripe_index == map->num_stripes) {
3047 /* This could only happen for RAID0/10 */
3053 for (i = 0; i < num_stripes; i++) {
3054 multi->stripes[i].physical =
3055 map->stripes[stripe_index].physical +
3057 stripe_nr * map->stripe_len;
3058 multi->stripes[i].dev =
3059 map->stripes[stripe_index].dev;
3065 multi->num_stripes = num_stripes;
3066 multi->max_errors = max_errors;
3069 free_extent_map(em);
3073 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
3074 u64 logical, u64 *length,
3075 struct btrfs_multi_bio **multi_ret, int mirror_num)
3077 return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
3081 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
3082 u64 chunk_start, u64 physical, u64 devid,
3083 u64 **logical, int *naddrs, int *stripe_len)
3085 struct extent_map_tree *em_tree = &map_tree->map_tree;
3086 struct extent_map *em;
3087 struct map_lookup *map;
3094 read_lock(&em_tree->lock);
3095 em = lookup_extent_mapping(em_tree, chunk_start, 1);
3096 read_unlock(&em_tree->lock);
3098 BUG_ON(!em || em->start != chunk_start);
3099 map = (struct map_lookup *)em->bdev;
3102 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
3103 do_div(length, map->num_stripes / map->sub_stripes);
3104 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
3105 do_div(length, map->num_stripes);
3107 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
3110 for (i = 0; i < map->num_stripes; i++) {
3111 if (devid && map->stripes[i].dev->devid != devid)
3113 if (map->stripes[i].physical > physical ||
3114 map->stripes[i].physical + length <= physical)
3117 stripe_nr = physical - map->stripes[i].physical;
3118 do_div(stripe_nr, map->stripe_len);
3120 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3121 stripe_nr = stripe_nr * map->num_stripes + i;
3122 do_div(stripe_nr, map->sub_stripes);
3123 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3124 stripe_nr = stripe_nr * map->num_stripes + i;
3126 bytenr = chunk_start + stripe_nr * map->stripe_len;
3127 WARN_ON(nr >= map->num_stripes);
3128 for (j = 0; j < nr; j++) {
3129 if (buf[j] == bytenr)
3133 WARN_ON(nr >= map->num_stripes);
3140 *stripe_len = map->stripe_len;
3142 free_extent_map(em);
3146 static void end_bio_multi_stripe(struct bio *bio, int err)
3148 struct btrfs_multi_bio *multi = bio->bi_private;
3149 int is_orig_bio = 0;
3152 atomic_inc(&multi->error);
3154 if (bio == multi->orig_bio)
3157 if (atomic_dec_and_test(&multi->stripes_pending)) {
3160 bio = multi->orig_bio;
3162 bio->bi_private = multi->private;
3163 bio->bi_end_io = multi->end_io;
3164 /* only send an error to the higher layers if it is
3165 * beyond the tolerance of the multi-bio
3167 if (atomic_read(&multi->error) > multi->max_errors) {
3171 * this bio is actually up to date, we didn't
3172 * go over the max number of errors
3174 set_bit(BIO_UPTODATE, &bio->bi_flags);
3179 bio_endio(bio, err);
3180 } else if (!is_orig_bio) {
3185 struct async_sched {
3188 struct btrfs_fs_info *info;
3189 struct btrfs_work work;
3193 * see run_scheduled_bios for a description of why bios are collected for
3196 * This will add one bio to the pending list for a device and make sure
3197 * the work struct is scheduled.
3199 static noinline int schedule_bio(struct btrfs_root *root,
3200 struct btrfs_device *device,
3201 int rw, struct bio *bio)
3203 int should_queue = 1;
3204 struct btrfs_pending_bios *pending_bios;
3206 /* don't bother with additional async steps for reads, right now */
3207 if (!(rw & REQ_WRITE)) {
3209 submit_bio(rw, bio);
3215 * nr_async_bios allows us to reliably return congestion to the
3216 * higher layers. Otherwise, the async bio makes it appear we have
3217 * made progress against dirty pages when we've really just put it
3218 * on a queue for later
3220 atomic_inc(&root->fs_info->nr_async_bios);
3221 WARN_ON(bio->bi_next);
3222 bio->bi_next = NULL;
3225 spin_lock(&device->io_lock);
3226 if (bio->bi_rw & REQ_SYNC)
3227 pending_bios = &device->pending_sync_bios;
3229 pending_bios = &device->pending_bios;
3231 if (pending_bios->tail)
3232 pending_bios->tail->bi_next = bio;
3234 pending_bios->tail = bio;
3235 if (!pending_bios->head)
3236 pending_bios->head = bio;
3237 if (device->running_pending)
3240 spin_unlock(&device->io_lock);
3243 btrfs_queue_worker(&root->fs_info->submit_workers,
3248 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
3249 int mirror_num, int async_submit)
3251 struct btrfs_mapping_tree *map_tree;
3252 struct btrfs_device *dev;
3253 struct bio *first_bio = bio;
3254 u64 logical = (u64)bio->bi_sector << 9;
3257 struct btrfs_multi_bio *multi = NULL;
3262 length = bio->bi_size;
3263 map_tree = &root->fs_info->mapping_tree;
3264 map_length = length;
3266 ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
3270 total_devs = multi->num_stripes;
3271 if (map_length < length) {
3272 printk(KERN_CRIT "mapping failed logical %llu bio len %llu "
3273 "len %llu\n", (unsigned long long)logical,
3274 (unsigned long long)length,
3275 (unsigned long long)map_length);
3278 multi->end_io = first_bio->bi_end_io;
3279 multi->private = first_bio->bi_private;
3280 multi->orig_bio = first_bio;
3281 atomic_set(&multi->stripes_pending, multi->num_stripes);
3283 while (dev_nr < total_devs) {
3284 if (total_devs > 1) {
3285 if (dev_nr < total_devs - 1) {
3286 bio = bio_clone(first_bio, GFP_NOFS);
3291 bio->bi_private = multi;
3292 bio->bi_end_io = end_bio_multi_stripe;
3294 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
3295 dev = multi->stripes[dev_nr].dev;
3296 if (dev && dev->bdev && (rw != WRITE || dev->writeable)) {
3297 bio->bi_bdev = dev->bdev;
3299 schedule_bio(root, dev, rw, bio);
3301 submit_bio(rw, bio);
3303 bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
3304 bio->bi_sector = logical >> 9;
3305 bio_endio(bio, -EIO);
3309 if (total_devs == 1)
3314 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
3317 struct btrfs_device *device;
3318 struct btrfs_fs_devices *cur_devices;
3320 cur_devices = root->fs_info->fs_devices;
3321 while (cur_devices) {
3323 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3324 device = __find_device(&cur_devices->devices,
3329 cur_devices = cur_devices->seed;
3334 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
3335 u64 devid, u8 *dev_uuid)
3337 struct btrfs_device *device;
3338 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
3340 device = kzalloc(sizeof(*device), GFP_NOFS);
3343 list_add(&device->dev_list,
3344 &fs_devices->devices);
3345 device->dev_root = root->fs_info->dev_root;
3346 device->devid = devid;
3347 device->work.func = pending_bios_fn;
3348 device->fs_devices = fs_devices;
3349 device->missing = 1;
3350 fs_devices->num_devices++;
3351 fs_devices->missing_devices++;
3352 spin_lock_init(&device->io_lock);
3353 INIT_LIST_HEAD(&device->dev_alloc_list);
3354 memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
3358 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
3359 struct extent_buffer *leaf,
3360 struct btrfs_chunk *chunk)
3362 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
3363 struct map_lookup *map;
3364 struct extent_map *em;
3368 u8 uuid[BTRFS_UUID_SIZE];
3373 logical = key->offset;
3374 length = btrfs_chunk_length(leaf, chunk);
3376 read_lock(&map_tree->map_tree.lock);
3377 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
3378 read_unlock(&map_tree->map_tree.lock);
3380 /* already mapped? */
3381 if (em && em->start <= logical && em->start + em->len > logical) {
3382 free_extent_map(em);
3385 free_extent_map(em);
3388 em = alloc_extent_map();
3391 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3392 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3394 free_extent_map(em);
3398 em->bdev = (struct block_device *)map;
3399 em->start = logical;
3401 em->block_start = 0;
3402 em->block_len = em->len;
3404 map->num_stripes = num_stripes;
3405 map->io_width = btrfs_chunk_io_width(leaf, chunk);
3406 map->io_align = btrfs_chunk_io_align(leaf, chunk);
3407 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
3408 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
3409 map->type = btrfs_chunk_type(leaf, chunk);
3410 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
3411 for (i = 0; i < num_stripes; i++) {
3412 map->stripes[i].physical =
3413 btrfs_stripe_offset_nr(leaf, chunk, i);
3414 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
3415 read_extent_buffer(leaf, uuid, (unsigned long)
3416 btrfs_stripe_dev_uuid_nr(chunk, i),
3418 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
3420 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
3422 free_extent_map(em);
3425 if (!map->stripes[i].dev) {
3426 map->stripes[i].dev =
3427 add_missing_dev(root, devid, uuid);
3428 if (!map->stripes[i].dev) {
3430 free_extent_map(em);
3434 map->stripes[i].dev->in_fs_metadata = 1;
3437 write_lock(&map_tree->map_tree.lock);
3438 ret = add_extent_mapping(&map_tree->map_tree, em);
3439 write_unlock(&map_tree->map_tree.lock);
3441 free_extent_map(em);
3446 static int fill_device_from_item(struct extent_buffer *leaf,
3447 struct btrfs_dev_item *dev_item,
3448 struct btrfs_device *device)
3452 device->devid = btrfs_device_id(leaf, dev_item);
3453 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
3454 device->total_bytes = device->disk_total_bytes;
3455 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
3456 device->type = btrfs_device_type(leaf, dev_item);
3457 device->io_align = btrfs_device_io_align(leaf, dev_item);
3458 device->io_width = btrfs_device_io_width(leaf, dev_item);
3459 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
3461 ptr = (unsigned long)btrfs_device_uuid(dev_item);
3462 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
3467 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
3469 struct btrfs_fs_devices *fs_devices;
3472 mutex_lock(&uuid_mutex);
3474 fs_devices = root->fs_info->fs_devices->seed;
3475 while (fs_devices) {
3476 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3480 fs_devices = fs_devices->seed;
3483 fs_devices = find_fsid(fsid);
3489 fs_devices = clone_fs_devices(fs_devices);
3490 if (IS_ERR(fs_devices)) {
3491 ret = PTR_ERR(fs_devices);
3495 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
3496 root->fs_info->bdev_holder);
3500 if (!fs_devices->seeding) {
3501 __btrfs_close_devices(fs_devices);
3502 free_fs_devices(fs_devices);
3507 fs_devices->seed = root->fs_info->fs_devices->seed;
3508 root->fs_info->fs_devices->seed = fs_devices;
3510 mutex_unlock(&uuid_mutex);
3514 static int read_one_dev(struct btrfs_root *root,
3515 struct extent_buffer *leaf,
3516 struct btrfs_dev_item *dev_item)
3518 struct btrfs_device *device;
3521 u8 fs_uuid[BTRFS_UUID_SIZE];
3522 u8 dev_uuid[BTRFS_UUID_SIZE];
3524 devid = btrfs_device_id(leaf, dev_item);
3525 read_extent_buffer(leaf, dev_uuid,
3526 (unsigned long)btrfs_device_uuid(dev_item),
3528 read_extent_buffer(leaf, fs_uuid,
3529 (unsigned long)btrfs_device_fsid(dev_item),
3532 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
3533 ret = open_seed_devices(root, fs_uuid);
3534 if (ret && !btrfs_test_opt(root, DEGRADED))
3538 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
3539 if (!device || !device->bdev) {
3540 if (!btrfs_test_opt(root, DEGRADED))
3544 printk(KERN_WARNING "warning devid %llu missing\n",
3545 (unsigned long long)devid);
3546 device = add_missing_dev(root, devid, dev_uuid);
3549 } else if (!device->missing) {
3551 * this happens when a device that was properly setup
3552 * in the device info lists suddenly goes bad.
3553 * device->bdev is NULL, and so we have to set
3554 * device->missing to one here
3556 root->fs_info->fs_devices->missing_devices++;
3557 device->missing = 1;
3561 if (device->fs_devices != root->fs_info->fs_devices) {
3562 BUG_ON(device->writeable);
3563 if (device->generation !=
3564 btrfs_device_generation(leaf, dev_item))
3568 fill_device_from_item(leaf, dev_item, device);
3569 device->dev_root = root->fs_info->dev_root;
3570 device->in_fs_metadata = 1;
3571 if (device->writeable)
3572 device->fs_devices->total_rw_bytes += device->total_bytes;
3577 int btrfs_read_sys_array(struct btrfs_root *root)
3579 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
3580 struct extent_buffer *sb;
3581 struct btrfs_disk_key *disk_key;
3582 struct btrfs_chunk *chunk;
3584 unsigned long sb_ptr;
3590 struct btrfs_key key;
3592 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
3593 BTRFS_SUPER_INFO_SIZE);
3596 btrfs_set_buffer_uptodate(sb);
3597 btrfs_set_buffer_lockdep_class(sb, 0);
3599 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
3600 array_size = btrfs_super_sys_array_size(super_copy);
3602 ptr = super_copy->sys_chunk_array;
3603 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
3606 while (cur < array_size) {
3607 disk_key = (struct btrfs_disk_key *)ptr;
3608 btrfs_disk_key_to_cpu(&key, disk_key);
3610 len = sizeof(*disk_key); ptr += len;
3614 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
3615 chunk = (struct btrfs_chunk *)sb_ptr;
3616 ret = read_one_chunk(root, &key, sb, chunk);
3619 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
3620 len = btrfs_chunk_item_size(num_stripes);
3629 free_extent_buffer(sb);
3633 int btrfs_read_chunk_tree(struct btrfs_root *root)
3635 struct btrfs_path *path;
3636 struct extent_buffer *leaf;
3637 struct btrfs_key key;
3638 struct btrfs_key found_key;
3642 root = root->fs_info->chunk_root;
3644 path = btrfs_alloc_path();
3648 /* first we search for all of the device items, and then we
3649 * read in all of the chunk items. This way we can create chunk
3650 * mappings that reference all of the devices that are afound
3652 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
3656 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3660 leaf = path->nodes[0];
3661 slot = path->slots[0];
3662 if (slot >= btrfs_header_nritems(leaf)) {
3663 ret = btrfs_next_leaf(root, path);
3670 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3671 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3672 if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
3674 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
3675 struct btrfs_dev_item *dev_item;
3676 dev_item = btrfs_item_ptr(leaf, slot,
3677 struct btrfs_dev_item);
3678 ret = read_one_dev(root, leaf, dev_item);
3682 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
3683 struct btrfs_chunk *chunk;
3684 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3685 ret = read_one_chunk(root, &found_key, leaf, chunk);
3691 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3693 btrfs_release_path(path);
3698 btrfs_free_path(path);