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/buffer_head.h>
21 #include <linux/blkdev.h>
22 #include <asm/div64.h>
24 #include "extent_map.h"
26 #include "transaction.h"
27 #include "print-tree.h"
38 struct btrfs_bio_stripe stripes[];
41 #define map_lookup_size(n) (sizeof(struct map_lookup) + \
42 (sizeof(struct btrfs_bio_stripe) * (n)))
44 static DEFINE_MUTEX(uuid_mutex);
45 static LIST_HEAD(fs_uuids);
47 int btrfs_cleanup_fs_uuids(void)
49 struct btrfs_fs_devices *fs_devices;
50 struct list_head *uuid_cur;
51 struct list_head *devices_cur;
52 struct btrfs_device *dev;
54 list_for_each(uuid_cur, &fs_uuids) {
55 fs_devices = list_entry(uuid_cur, struct btrfs_fs_devices,
57 while(!list_empty(&fs_devices->devices)) {
58 devices_cur = fs_devices->devices.next;
59 dev = list_entry(devices_cur, struct btrfs_device,
62 close_bdev_excl(dev->bdev);
64 list_del(&dev->dev_list);
71 static struct btrfs_device *__find_device(struct list_head *head, u64 devid,
74 struct btrfs_device *dev;
75 struct list_head *cur;
77 list_for_each(cur, head) {
78 dev = list_entry(cur, struct btrfs_device, dev_list);
79 if (dev->devid == devid &&
80 !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE)) {
87 static struct btrfs_fs_devices *find_fsid(u8 *fsid)
89 struct list_head *cur;
90 struct btrfs_fs_devices *fs_devices;
92 list_for_each(cur, &fs_uuids) {
93 fs_devices = list_entry(cur, struct btrfs_fs_devices, list);
94 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
100 static int device_list_add(const char *path,
101 struct btrfs_super_block *disk_super,
102 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
104 struct btrfs_device *device;
105 struct btrfs_fs_devices *fs_devices;
106 u64 found_transid = btrfs_super_generation(disk_super);
108 fs_devices = find_fsid(disk_super->fsid);
110 fs_devices = kmalloc(sizeof(*fs_devices), GFP_NOFS);
113 INIT_LIST_HEAD(&fs_devices->devices);
114 list_add(&fs_devices->list, &fs_uuids);
115 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
116 fs_devices->latest_devid = devid;
117 fs_devices->latest_trans = found_transid;
118 fs_devices->lowest_devid = (u64)-1;
119 fs_devices->num_devices = 0;
122 device = __find_device(&fs_devices->devices, devid,
123 disk_super->dev_item.uuid);
126 device = kzalloc(sizeof(*device), GFP_NOFS);
128 /* we can safely leave the fs_devices entry around */
131 device->devid = devid;
132 memcpy(device->uuid, disk_super->dev_item.uuid,
134 device->barriers = 1;
135 spin_lock_init(&device->io_lock);
136 device->name = kstrdup(path, GFP_NOFS);
141 list_add(&device->dev_list, &fs_devices->devices);
142 fs_devices->num_devices++;
145 if (found_transid > fs_devices->latest_trans) {
146 fs_devices->latest_devid = devid;
147 fs_devices->latest_trans = found_transid;
149 if (fs_devices->lowest_devid > devid) {
150 fs_devices->lowest_devid = devid;
152 *fs_devices_ret = fs_devices;
156 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
158 struct list_head *head = &fs_devices->devices;
159 struct list_head *cur;
160 struct btrfs_device *device;
162 mutex_lock(&uuid_mutex);
163 list_for_each(cur, head) {
164 device = list_entry(cur, struct btrfs_device, dev_list);
166 close_bdev_excl(device->bdev);
170 mutex_unlock(&uuid_mutex);
174 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
175 int flags, void *holder)
177 struct block_device *bdev;
178 struct list_head *head = &fs_devices->devices;
179 struct list_head *cur;
180 struct btrfs_device *device;
183 mutex_lock(&uuid_mutex);
184 list_for_each(cur, head) {
185 device = list_entry(cur, struct btrfs_device, dev_list);
186 bdev = open_bdev_excl(device->name, flags, holder);
189 printk("open %s failed\n", device->name);
193 if (device->devid == fs_devices->latest_devid)
194 fs_devices->latest_bdev = bdev;
195 if (device->devid == fs_devices->lowest_devid) {
196 fs_devices->lowest_bdev = bdev;
200 mutex_unlock(&uuid_mutex);
203 mutex_unlock(&uuid_mutex);
204 btrfs_close_devices(fs_devices);
208 int btrfs_scan_one_device(const char *path, int flags, void *holder,
209 struct btrfs_fs_devices **fs_devices_ret)
211 struct btrfs_super_block *disk_super;
212 struct block_device *bdev;
213 struct buffer_head *bh;
218 mutex_lock(&uuid_mutex);
220 bdev = open_bdev_excl(path, flags, holder);
227 ret = set_blocksize(bdev, 4096);
230 bh = __bread(bdev, BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
235 disk_super = (struct btrfs_super_block *)bh->b_data;
236 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
237 sizeof(disk_super->magic))) {
241 devid = le64_to_cpu(disk_super->dev_item.devid);
242 transid = btrfs_super_generation(disk_super);
243 if (disk_super->label[0])
244 printk("device label %s ", disk_super->label);
246 /* FIXME, make a readl uuid parser */
247 printk("device fsid %llx-%llx ",
248 *(unsigned long long *)disk_super->fsid,
249 *(unsigned long long *)(disk_super->fsid + 8));
251 printk("devid %Lu transid %Lu %s\n", devid, transid, path);
252 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
257 close_bdev_excl(bdev);
259 mutex_unlock(&uuid_mutex);
264 * this uses a pretty simple search, the expectation is that it is
265 * called very infrequently and that a given device has a small number
268 static int find_free_dev_extent(struct btrfs_trans_handle *trans,
269 struct btrfs_device *device,
270 struct btrfs_path *path,
271 u64 num_bytes, u64 *start)
273 struct btrfs_key key;
274 struct btrfs_root *root = device->dev_root;
275 struct btrfs_dev_extent *dev_extent = NULL;
278 u64 search_start = 0;
279 u64 search_end = device->total_bytes;
283 struct extent_buffer *l;
288 /* FIXME use last free of some kind */
290 /* we don't want to overwrite the superblock on the drive,
291 * so we make sure to start at an offset of at least 1MB
293 search_start = max((u64)1024 * 1024, search_start);
294 key.objectid = device->devid;
295 key.offset = search_start;
296 key.type = BTRFS_DEV_EXTENT_KEY;
297 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
300 ret = btrfs_previous_item(root, path, 0, key.type);
304 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
307 slot = path->slots[0];
308 if (slot >= btrfs_header_nritems(l)) {
309 ret = btrfs_next_leaf(root, path);
316 if (search_start >= search_end) {
320 *start = search_start;
324 *start = last_byte > search_start ?
325 last_byte : search_start;
326 if (search_end <= *start) {
332 btrfs_item_key_to_cpu(l, &key, slot);
334 if (key.objectid < device->devid)
337 if (key.objectid > device->devid)
340 if (key.offset >= search_start && key.offset > last_byte &&
342 if (last_byte < search_start)
343 last_byte = search_start;
344 hole_size = key.offset - last_byte;
345 if (key.offset > last_byte &&
346 hole_size >= num_bytes) {
351 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) {
356 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
357 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
363 /* we have to make sure we didn't find an extent that has already
364 * been allocated by the map tree or the original allocation
366 btrfs_release_path(root, path);
367 BUG_ON(*start < search_start);
369 if (*start + num_bytes > search_end) {
373 /* check for pending inserts here */
377 btrfs_release_path(root, path);
381 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
382 struct btrfs_device *device,
383 u64 chunk_tree, u64 chunk_objectid,
385 u64 num_bytes, u64 *start)
388 struct btrfs_path *path;
389 struct btrfs_root *root = device->dev_root;
390 struct btrfs_dev_extent *extent;
391 struct extent_buffer *leaf;
392 struct btrfs_key key;
394 path = btrfs_alloc_path();
398 ret = find_free_dev_extent(trans, device, path, num_bytes, start);
403 key.objectid = device->devid;
405 key.type = BTRFS_DEV_EXTENT_KEY;
406 ret = btrfs_insert_empty_item(trans, root, path, &key,
410 leaf = path->nodes[0];
411 extent = btrfs_item_ptr(leaf, path->slots[0],
412 struct btrfs_dev_extent);
413 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
414 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
415 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
417 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
418 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
421 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
422 btrfs_mark_buffer_dirty(leaf);
424 btrfs_free_path(path);
428 static int find_next_chunk(struct btrfs_root *root, u64 objectid, u64 *offset)
430 struct btrfs_path *path;
432 struct btrfs_key key;
433 struct btrfs_chunk *chunk;
434 struct btrfs_key found_key;
436 path = btrfs_alloc_path();
439 key.objectid = objectid;
440 key.offset = (u64)-1;
441 key.type = BTRFS_CHUNK_ITEM_KEY;
443 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
449 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
453 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
455 if (found_key.objectid != objectid)
458 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
460 *offset = found_key.offset +
461 btrfs_chunk_length(path->nodes[0], chunk);
466 btrfs_free_path(path);
470 static int find_next_devid(struct btrfs_root *root, struct btrfs_path *path,
474 struct btrfs_key key;
475 struct btrfs_key found_key;
477 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
478 key.type = BTRFS_DEV_ITEM_KEY;
479 key.offset = (u64)-1;
481 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
487 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
492 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
494 *objectid = found_key.offset + 1;
498 btrfs_release_path(root, path);
503 * the device information is stored in the chunk root
504 * the btrfs_device struct should be fully filled in
506 int btrfs_add_device(struct btrfs_trans_handle *trans,
507 struct btrfs_root *root,
508 struct btrfs_device *device)
511 struct btrfs_path *path;
512 struct btrfs_dev_item *dev_item;
513 struct extent_buffer *leaf;
514 struct btrfs_key key;
518 root = root->fs_info->chunk_root;
520 path = btrfs_alloc_path();
524 ret = find_next_devid(root, path, &free_devid);
528 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
529 key.type = BTRFS_DEV_ITEM_KEY;
530 key.offset = free_devid;
532 ret = btrfs_insert_empty_item(trans, root, path, &key,
537 leaf = path->nodes[0];
538 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
540 device->devid = free_devid;
541 btrfs_set_device_id(leaf, dev_item, device->devid);
542 btrfs_set_device_type(leaf, dev_item, device->type);
543 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
544 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
545 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
546 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
547 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
548 btrfs_set_device_group(leaf, dev_item, 0);
549 btrfs_set_device_seek_speed(leaf, dev_item, 0);
550 btrfs_set_device_bandwidth(leaf, dev_item, 0);
552 ptr = (unsigned long)btrfs_device_uuid(dev_item);
553 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
554 btrfs_mark_buffer_dirty(leaf);
558 btrfs_free_path(path);
561 int btrfs_update_device(struct btrfs_trans_handle *trans,
562 struct btrfs_device *device)
565 struct btrfs_path *path;
566 struct btrfs_root *root;
567 struct btrfs_dev_item *dev_item;
568 struct extent_buffer *leaf;
569 struct btrfs_key key;
571 root = device->dev_root->fs_info->chunk_root;
573 path = btrfs_alloc_path();
577 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
578 key.type = BTRFS_DEV_ITEM_KEY;
579 key.offset = device->devid;
581 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
590 leaf = path->nodes[0];
591 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
593 btrfs_set_device_id(leaf, dev_item, device->devid);
594 btrfs_set_device_type(leaf, dev_item, device->type);
595 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
596 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
597 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
598 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
599 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
600 btrfs_mark_buffer_dirty(leaf);
603 btrfs_free_path(path);
607 int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
608 struct btrfs_root *root,
609 struct btrfs_key *key,
610 struct btrfs_chunk *chunk, int item_size)
612 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
613 struct btrfs_disk_key disk_key;
617 array_size = btrfs_super_sys_array_size(super_copy);
618 if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
621 ptr = super_copy->sys_chunk_array + array_size;
622 btrfs_cpu_key_to_disk(&disk_key, key);
623 memcpy(ptr, &disk_key, sizeof(disk_key));
624 ptr += sizeof(disk_key);
625 memcpy(ptr, chunk, item_size);
626 item_size += sizeof(disk_key);
627 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
631 static u64 div_factor(u64 num, int factor)
640 static u64 chunk_bytes_by_type(u64 type, u64 calc_size, int num_stripes,
643 if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
645 else if (type & BTRFS_BLOCK_GROUP_RAID10)
646 return calc_size * (num_stripes / sub_stripes);
648 return calc_size * num_stripes;
652 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
653 struct btrfs_root *extent_root, u64 *start,
654 u64 *num_bytes, u64 type)
657 struct btrfs_fs_info *info = extent_root->fs_info;
658 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
659 struct btrfs_stripe *stripes;
660 struct btrfs_device *device = NULL;
661 struct btrfs_chunk *chunk;
662 struct list_head private_devs;
663 struct list_head *dev_list = &extent_root->fs_info->fs_devices->devices;
664 struct list_head *cur;
665 struct extent_map_tree *em_tree;
666 struct map_lookup *map;
667 struct extent_map *em;
668 int min_stripe_size = 1 * 1024 * 1024;
670 u64 calc_size = 1024 * 1024 * 1024;
671 u64 max_chunk_size = calc_size;
682 int stripe_len = 64 * 1024;
683 struct btrfs_key key;
685 if (list_empty(dev_list))
688 if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
689 num_stripes = btrfs_super_num_devices(&info->super_copy);
692 if (type & (BTRFS_BLOCK_GROUP_DUP)) {
696 if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
697 num_stripes = min_t(u64, 2,
698 btrfs_super_num_devices(&info->super_copy));
703 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
704 num_stripes = btrfs_super_num_devices(&info->super_copy);
707 num_stripes &= ~(u32)1;
712 if (type & BTRFS_BLOCK_GROUP_DATA) {
713 max_chunk_size = 10 * calc_size;
714 min_stripe_size = 64 * 1024 * 1024;
715 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
716 max_chunk_size = 4 * calc_size;
717 min_stripe_size = 32 * 1024 * 1024;
718 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
719 calc_size = 8 * 1024 * 1024;
720 max_chunk_size = calc_size * 2;
721 min_stripe_size = 1 * 1024 * 1024;
724 /* we don't want a chunk larger than 10% of the FS */
725 percent_max = div_factor(btrfs_super_total_bytes(&info->super_copy), 1);
726 max_chunk_size = min(percent_max, max_chunk_size);
729 if (calc_size * num_stripes > max_chunk_size) {
730 calc_size = max_chunk_size;
731 do_div(calc_size, num_stripes);
732 do_div(calc_size, stripe_len);
733 calc_size *= stripe_len;
735 /* we don't want tiny stripes */
736 calc_size = max_t(u64, min_stripe_size, calc_size);
738 do_div(calc_size, stripe_len);
739 calc_size *= stripe_len;
741 INIT_LIST_HEAD(&private_devs);
742 cur = dev_list->next;
745 if (type & BTRFS_BLOCK_GROUP_DUP)
746 min_free = calc_size * 2;
748 min_free = calc_size;
750 /* we add 1MB because we never use the first 1MB of the device */
751 min_free += 1024 * 1024;
753 /* build a private list of devices we will allocate from */
754 while(index < num_stripes) {
755 device = list_entry(cur, struct btrfs_device, dev_list);
757 avail = device->total_bytes - device->bytes_used;
759 if (avail >= min_free) {
760 list_move_tail(&device->dev_list, &private_devs);
762 if (type & BTRFS_BLOCK_GROUP_DUP)
764 } else if (avail > max_avail)
769 if (index < num_stripes) {
770 list_splice(&private_devs, dev_list);
771 if (index >= min_stripes) {
773 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
774 num_stripes /= sub_stripes;
775 num_stripes *= sub_stripes;
780 if (!looped && max_avail > 0) {
782 calc_size = max_avail;
787 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
788 key.type = BTRFS_CHUNK_ITEM_KEY;
789 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
794 chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
798 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
804 stripes = &chunk->stripe;
805 *num_bytes = chunk_bytes_by_type(type, calc_size,
806 num_stripes, sub_stripes);
810 printk("new chunk type %Lu start %Lu size %Lu\n", type, key.offset, *num_bytes);
811 while(index < num_stripes) {
812 struct btrfs_stripe *stripe;
813 BUG_ON(list_empty(&private_devs));
814 cur = private_devs.next;
815 device = list_entry(cur, struct btrfs_device, dev_list);
817 /* loop over this device again if we're doing a dup group */
818 if (!(type & BTRFS_BLOCK_GROUP_DUP) ||
819 (index == num_stripes - 1))
820 list_move_tail(&device->dev_list, dev_list);
822 ret = btrfs_alloc_dev_extent(trans, device,
823 info->chunk_root->root_key.objectid,
824 BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
825 calc_size, &dev_offset);
827 printk("alloc chunk start %Lu size %Lu from dev %Lu type %Lu\n", key.offset, calc_size, device->devid, type);
828 device->bytes_used += calc_size;
829 ret = btrfs_update_device(trans, device);
832 map->stripes[index].dev = device;
833 map->stripes[index].physical = dev_offset;
834 stripe = stripes + index;
835 btrfs_set_stack_stripe_devid(stripe, device->devid);
836 btrfs_set_stack_stripe_offset(stripe, dev_offset);
837 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
838 physical = dev_offset;
841 BUG_ON(!list_empty(&private_devs));
843 /* key was set above */
844 btrfs_set_stack_chunk_length(chunk, *num_bytes);
845 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
846 btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
847 btrfs_set_stack_chunk_type(chunk, type);
848 btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
849 btrfs_set_stack_chunk_io_align(chunk, stripe_len);
850 btrfs_set_stack_chunk_io_width(chunk, stripe_len);
851 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
852 btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
853 map->sector_size = extent_root->sectorsize;
854 map->stripe_len = stripe_len;
855 map->io_align = stripe_len;
856 map->io_width = stripe_len;
858 map->num_stripes = num_stripes;
859 map->sub_stripes = sub_stripes;
861 ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
862 btrfs_chunk_item_size(num_stripes));
864 *start = key.offset;;
866 em = alloc_extent_map(GFP_NOFS);
869 em->bdev = (struct block_device *)map;
870 em->start = key.offset;
871 em->len = *num_bytes;
876 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
877 spin_lock(&em_tree->lock);
878 ret = add_extent_mapping(em_tree, em);
879 spin_unlock(&em_tree->lock);
885 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
887 extent_map_tree_init(&tree->map_tree, GFP_NOFS);
890 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
892 struct extent_map *em;
895 spin_lock(&tree->map_tree.lock);
896 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
898 remove_extent_mapping(&tree->map_tree, em);
899 spin_unlock(&tree->map_tree.lock);
905 /* once for the tree */
910 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
912 struct extent_map *em;
913 struct map_lookup *map;
914 struct extent_map_tree *em_tree = &map_tree->map_tree;
917 spin_lock(&em_tree->lock);
918 em = lookup_extent_mapping(em_tree, logical, len);
919 spin_unlock(&em_tree->lock);
922 BUG_ON(em->start > logical || em->start + em->len < logical);
923 map = (struct map_lookup *)em->bdev;
924 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
925 ret = map->num_stripes;
926 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
927 ret = map->sub_stripes;
934 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
935 u64 logical, u64 *length,
936 struct btrfs_multi_bio **multi_ret,
937 int mirror_num, struct page *unplug_page)
939 struct extent_map *em;
940 struct map_lookup *map;
941 struct extent_map_tree *em_tree = &map_tree->map_tree;
945 int stripes_allocated = 8;
946 int stripes_required = 1;
950 struct btrfs_multi_bio *multi = NULL;
952 if (multi_ret && !(rw & (1 << BIO_RW))) {
953 stripes_allocated = 1;
957 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
963 spin_lock(&em_tree->lock);
964 em = lookup_extent_mapping(em_tree, logical, *length);
965 spin_unlock(&em_tree->lock);
967 if (!em && unplug_page)
971 printk("unable to find logical %Lu\n", logical);
975 BUG_ON(em->start > logical || em->start + em->len < logical);
976 map = (struct map_lookup *)em->bdev;
977 offset = logical - em->start;
979 if (mirror_num > map->num_stripes)
982 /* if our multi bio struct is too small, back off and try again */
983 if (rw & (1 << BIO_RW)) {
984 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
985 BTRFS_BLOCK_GROUP_DUP)) {
986 stripes_required = map->num_stripes;
987 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
988 stripes_required = map->sub_stripes;
991 if (multi_ret && rw == WRITE &&
992 stripes_allocated < stripes_required) {
993 stripes_allocated = map->num_stripes;
1000 * stripe_nr counts the total number of stripes we have to stride
1001 * to get to this block
1003 do_div(stripe_nr, map->stripe_len);
1005 stripe_offset = stripe_nr * map->stripe_len;
1006 BUG_ON(offset < stripe_offset);
1008 /* stripe_offset is the offset of this block in its stripe*/
1009 stripe_offset = offset - stripe_offset;
1011 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
1012 BTRFS_BLOCK_GROUP_RAID10 |
1013 BTRFS_BLOCK_GROUP_DUP)) {
1014 /* we limit the length of each bio to what fits in a stripe */
1015 *length = min_t(u64, em->len - offset,
1016 map->stripe_len - stripe_offset);
1018 *length = em->len - offset;
1021 if (!multi_ret && !unplug_page)
1026 if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
1027 if (unplug_page || (rw & (1 << BIO_RW)))
1028 num_stripes = map->num_stripes;
1029 else if (mirror_num) {
1030 stripe_index = mirror_num - 1;
1033 u64 least = (u64)-1;
1034 struct btrfs_device *cur;
1036 for (i = 0; i < map->num_stripes; i++) {
1037 cur = map->stripes[i].dev;
1038 spin_lock(&cur->io_lock);
1039 if (cur->total_ios < least) {
1040 least = cur->total_ios;
1043 spin_unlock(&cur->io_lock);
1046 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
1047 if (rw & (1 << BIO_RW))
1048 num_stripes = map->num_stripes;
1049 else if (mirror_num)
1050 stripe_index = mirror_num - 1;
1051 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1052 int factor = map->num_stripes / map->sub_stripes;
1053 int orig_stripe_nr = stripe_nr;
1055 stripe_index = do_div(stripe_nr, factor);
1056 stripe_index *= map->sub_stripes;
1058 if (unplug_page || (rw & (1 << BIO_RW)))
1059 num_stripes = map->sub_stripes;
1060 else if (mirror_num)
1061 stripe_index += mirror_num - 1;
1063 stripe_index += orig_stripe_nr % map->sub_stripes;
1066 * after this do_div call, stripe_nr is the number of stripes
1067 * on this device we have to walk to find the data, and
1068 * stripe_index is the number of our device in the stripe array
1070 stripe_index = do_div(stripe_nr, map->num_stripes);
1072 BUG_ON(stripe_index >= map->num_stripes);
1074 for (i = 0; i < num_stripes; i++) {
1076 struct btrfs_device *device;
1077 struct backing_dev_info *bdi;
1079 device = map->stripes[stripe_index].dev;
1080 bdi = blk_get_backing_dev_info(device->bdev);
1081 if (bdi->unplug_io_fn) {
1082 bdi->unplug_io_fn(bdi, unplug_page);
1085 multi->stripes[i].physical =
1086 map->stripes[stripe_index].physical +
1087 stripe_offset + stripe_nr * map->stripe_len;
1088 multi->stripes[i].dev = map->stripes[stripe_index].dev;
1094 multi->num_stripes = num_stripes;
1097 free_extent_map(em);
1101 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
1102 u64 logical, u64 *length,
1103 struct btrfs_multi_bio **multi_ret, int mirror_num)
1105 return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
1109 int btrfs_unplug_page(struct btrfs_mapping_tree *map_tree,
1110 u64 logical, struct page *page)
1112 u64 length = PAGE_CACHE_SIZE;
1113 return __btrfs_map_block(map_tree, READ, logical, &length,
1118 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1119 static void end_bio_multi_stripe(struct bio *bio, int err)
1121 static int end_bio_multi_stripe(struct bio *bio,
1122 unsigned int bytes_done, int err)
1125 struct btrfs_multi_bio *multi = bio->bi_private;
1127 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1134 if (atomic_dec_and_test(&multi->stripes_pending)) {
1135 bio->bi_private = multi->private;
1136 bio->bi_end_io = multi->end_io;
1138 if (!err && multi->error)
1142 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1143 bio_endio(bio, bio->bi_size, err);
1145 bio_endio(bio, err);
1150 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1155 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
1158 struct btrfs_mapping_tree *map_tree;
1159 struct btrfs_device *dev;
1160 struct bio *first_bio = bio;
1161 u64 logical = bio->bi_sector << 9;
1164 struct btrfs_multi_bio *multi = NULL;
1169 length = bio->bi_size;
1171 map_tree = &root->fs_info->mapping_tree;
1172 map_length = length;
1174 ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
1178 total_devs = multi->num_stripes;
1179 if (map_length < length) {
1180 printk("mapping failed logical %Lu bio len %Lu "
1181 "len %Lu\n", logical, length, map_length);
1184 multi->end_io = first_bio->bi_end_io;
1185 multi->private = first_bio->bi_private;
1186 atomic_set(&multi->stripes_pending, multi->num_stripes);
1188 while(dev_nr < total_devs) {
1189 if (total_devs > 1) {
1190 if (dev_nr < total_devs - 1) {
1191 bio = bio_clone(first_bio, GFP_NOFS);
1196 bio->bi_private = multi;
1197 bio->bi_end_io = end_bio_multi_stripe;
1199 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
1200 dev = multi->stripes[dev_nr].dev;
1201 bio->bi_bdev = dev->bdev;
1202 spin_lock(&dev->io_lock);
1204 spin_unlock(&dev->io_lock);
1205 submit_bio(rw, bio);
1208 if (total_devs == 1)
1213 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
1216 struct list_head *head = &root->fs_info->fs_devices->devices;
1218 return __find_device(head, devid, uuid);
1221 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
1222 struct extent_buffer *leaf,
1223 struct btrfs_chunk *chunk)
1225 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
1226 struct map_lookup *map;
1227 struct extent_map *em;
1231 u8 uuid[BTRFS_UUID_SIZE];
1236 logical = key->offset;
1237 length = btrfs_chunk_length(leaf, chunk);
1238 spin_lock(&map_tree->map_tree.lock);
1239 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
1240 spin_unlock(&map_tree->map_tree.lock);
1242 /* already mapped? */
1243 if (em && em->start <= logical && em->start + em->len > logical) {
1244 free_extent_map(em);
1247 free_extent_map(em);
1250 map = kzalloc(sizeof(*map), GFP_NOFS);
1254 em = alloc_extent_map(GFP_NOFS);
1257 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
1258 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
1260 free_extent_map(em);
1264 em->bdev = (struct block_device *)map;
1265 em->start = logical;
1267 em->block_start = 0;
1269 map->num_stripes = num_stripes;
1270 map->io_width = btrfs_chunk_io_width(leaf, chunk);
1271 map->io_align = btrfs_chunk_io_align(leaf, chunk);
1272 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
1273 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
1274 map->type = btrfs_chunk_type(leaf, chunk);
1275 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
1276 for (i = 0; i < num_stripes; i++) {
1277 map->stripes[i].physical =
1278 btrfs_stripe_offset_nr(leaf, chunk, i);
1279 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
1280 read_extent_buffer(leaf, uuid, (unsigned long)
1281 btrfs_stripe_dev_uuid_nr(chunk, i),
1283 map->stripes[i].dev = btrfs_find_device(root, devid, uuid);
1284 if (!map->stripes[i].dev) {
1286 free_extent_map(em);
1291 spin_lock(&map_tree->map_tree.lock);
1292 ret = add_extent_mapping(&map_tree->map_tree, em);
1293 spin_unlock(&map_tree->map_tree.lock);
1295 free_extent_map(em);
1300 static int fill_device_from_item(struct extent_buffer *leaf,
1301 struct btrfs_dev_item *dev_item,
1302 struct btrfs_device *device)
1306 device->devid = btrfs_device_id(leaf, dev_item);
1307 device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
1308 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
1309 device->type = btrfs_device_type(leaf, dev_item);
1310 device->io_align = btrfs_device_io_align(leaf, dev_item);
1311 device->io_width = btrfs_device_io_width(leaf, dev_item);
1312 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
1314 ptr = (unsigned long)btrfs_device_uuid(dev_item);
1315 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1320 static int read_one_dev(struct btrfs_root *root,
1321 struct extent_buffer *leaf,
1322 struct btrfs_dev_item *dev_item)
1324 struct btrfs_device *device;
1327 u8 dev_uuid[BTRFS_UUID_SIZE];
1329 devid = btrfs_device_id(leaf, dev_item);
1330 read_extent_buffer(leaf, dev_uuid,
1331 (unsigned long)btrfs_device_uuid(dev_item),
1333 device = btrfs_find_device(root, devid, dev_uuid);
1335 printk("warning devid %Lu not found already\n", devid);
1336 device = kzalloc(sizeof(*device), GFP_NOFS);
1339 list_add(&device->dev_list,
1340 &root->fs_info->fs_devices->devices);
1341 device->barriers = 1;
1342 spin_lock_init(&device->io_lock);
1345 fill_device_from_item(leaf, dev_item, device);
1346 device->dev_root = root->fs_info->dev_root;
1349 ret = btrfs_open_device(device);
1357 int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
1359 struct btrfs_dev_item *dev_item;
1361 dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
1363 return read_one_dev(root, buf, dev_item);
1366 int btrfs_read_sys_array(struct btrfs_root *root)
1368 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1369 struct extent_buffer *sb = root->fs_info->sb_buffer;
1370 struct btrfs_disk_key *disk_key;
1371 struct btrfs_chunk *chunk;
1372 struct btrfs_key key;
1377 unsigned long sb_ptr;
1381 array_size = btrfs_super_sys_array_size(super_copy);
1384 * we do this loop twice, once for the device items and
1385 * once for all of the chunks. This way there are device
1386 * structs filled in for every chunk
1388 ptr = super_copy->sys_chunk_array;
1389 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
1392 while (cur < array_size) {
1393 disk_key = (struct btrfs_disk_key *)ptr;
1394 btrfs_disk_key_to_cpu(&key, disk_key);
1396 len = sizeof(*disk_key);
1401 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1402 chunk = (struct btrfs_chunk *)sb_ptr;
1403 ret = read_one_chunk(root, &key, sb, chunk);
1405 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
1406 len = btrfs_chunk_item_size(num_stripes);
1417 int btrfs_read_chunk_tree(struct btrfs_root *root)
1419 struct btrfs_path *path;
1420 struct extent_buffer *leaf;
1421 struct btrfs_key key;
1422 struct btrfs_key found_key;
1426 root = root->fs_info->chunk_root;
1428 path = btrfs_alloc_path();
1432 /* first we search for all of the device items, and then we
1433 * read in all of the chunk items. This way we can create chunk
1434 * mappings that reference all of the devices that are afound
1436 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1440 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1442 leaf = path->nodes[0];
1443 slot = path->slots[0];
1444 if (slot >= btrfs_header_nritems(leaf)) {
1445 ret = btrfs_next_leaf(root, path);
1452 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1453 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
1454 if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
1456 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
1457 struct btrfs_dev_item *dev_item;
1458 dev_item = btrfs_item_ptr(leaf, slot,
1459 struct btrfs_dev_item);
1460 ret = read_one_dev(root, leaf, dev_item);
1463 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
1464 struct btrfs_chunk *chunk;
1465 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
1466 ret = read_one_chunk(root, &found_key, leaf, chunk);
1470 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
1472 btrfs_release_path(root, path);
1476 btrfs_free_path(path);