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 #define _XOPEN_SOURCE 600
22 #include <sys/types.h>
24 #include <uuid/uuid.h>
29 #include "transaction.h"
30 #include "print-tree.h"
34 struct btrfs_device *dev;
39 struct cache_extent ce;
47 struct btrfs_bio_stripe stripes[];
50 #define map_lookup_size(n) (sizeof(struct map_lookup) + \
51 (sizeof(struct btrfs_bio_stripe) * (n)))
53 static LIST_HEAD(fs_uuids);
55 static struct btrfs_device *__find_device(struct list_head *head, u64 devid,
58 struct btrfs_device *dev;
59 struct list_head *cur;
61 list_for_each(cur, head) {
62 dev = list_entry(cur, struct btrfs_device, dev_list);
63 if (dev->devid == devid &&
64 !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE)) {
71 static struct btrfs_fs_devices *find_fsid(u8 *fsid)
73 struct list_head *cur;
74 struct btrfs_fs_devices *fs_devices;
76 list_for_each(cur, &fs_uuids) {
77 fs_devices = list_entry(cur, struct btrfs_fs_devices, list);
78 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
84 static int device_list_add(const char *path,
85 struct btrfs_super_block *disk_super,
86 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
88 struct btrfs_device *device;
89 struct btrfs_fs_devices *fs_devices;
90 u64 found_transid = btrfs_super_generation(disk_super);
92 fs_devices = find_fsid(disk_super->fsid);
94 fs_devices = kmalloc(sizeof(*fs_devices), GFP_NOFS);
97 INIT_LIST_HEAD(&fs_devices->devices);
98 list_add(&fs_devices->list, &fs_uuids);
99 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
100 fs_devices->latest_devid = devid;
101 fs_devices->latest_trans = found_transid;
102 fs_devices->lowest_devid = (u64)-1;
105 device = __find_device(&fs_devices->devices, devid,
106 disk_super->dev_item.uuid);
109 device = kzalloc(sizeof(*device), GFP_NOFS);
111 /* we can safely leave the fs_devices entry around */
114 device->devid = devid;
115 memcpy(device->uuid, disk_super->dev_item.uuid,
117 device->name = kstrdup(path, GFP_NOFS);
122 list_add(&device->dev_list, &fs_devices->devices);
125 if (found_transid > fs_devices->latest_trans) {
126 fs_devices->latest_devid = devid;
127 fs_devices->latest_trans = found_transid;
129 if (fs_devices->lowest_devid > devid) {
130 fs_devices->lowest_devid = devid;
132 *fs_devices_ret = fs_devices;
136 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
138 struct list_head *head = &fs_devices->devices;
139 struct list_head *cur;
140 struct btrfs_device *device;
142 list_for_each(cur, head) {
143 device = list_entry(cur, struct btrfs_device, dev_list);
149 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, int flags)
152 struct list_head *head = &fs_devices->devices;
153 struct list_head *cur;
154 struct btrfs_device *device;
157 list_for_each(cur, head) {
158 device = list_entry(cur, struct btrfs_device, dev_list);
159 fd = open(device->name, flags);
160 printk("opening %s devid %llu fd %d\n", device->name,
161 (unsigned long long)device->devid, fd);
166 if (device->devid == fs_devices->latest_devid)
167 fs_devices->latest_bdev = fd;
168 if (device->devid == fs_devices->lowest_devid)
169 fs_devices->lowest_bdev = fd;
174 btrfs_close_devices(fs_devices);
178 int btrfs_scan_one_device(int fd, const char *path,
179 struct btrfs_fs_devices **fs_devices_ret,
180 u64 *total_devs, u64 super_offset)
182 struct btrfs_super_block *disk_super;
193 ret = pread(fd, buf, 4096, super_offset);
198 disk_super = (struct btrfs_super_block *)buf;
199 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
200 sizeof(disk_super->magic))) {
204 devid = le64_to_cpu(disk_super->dev_item.devid);
205 *total_devs = btrfs_super_num_devices(disk_super);
206 uuid_unparse(disk_super->fsid, uuidbuf);
209 if (disk_super->label[0])
210 printf("label %s ", disk_super->label);
212 printf("fsuuid %s ", uuidbuf);
213 printf("devid %llu %s\n", (unsigned long long)devid, path);
214 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
223 * this uses a pretty simple search, the expectation is that it is
224 * called very infrequently and that a given device has a small number
227 static int find_free_dev_extent(struct btrfs_trans_handle *trans,
228 struct btrfs_device *device,
229 struct btrfs_path *path,
230 u64 num_bytes, u64 *start)
232 struct btrfs_key key;
233 struct btrfs_root *root = device->dev_root;
234 struct btrfs_dev_extent *dev_extent = NULL;
237 u64 search_start = 0;
238 u64 search_end = device->total_bytes;
242 struct extent_buffer *l;
247 /* FIXME use last free of some kind */
249 /* we don't want to overwrite the superblock on the drive,
250 * so we make sure to start at an offset of at least 1MB
252 search_start = max((u64)1024 * 1024, search_start);
253 key.objectid = device->devid;
254 key.offset = search_start;
255 key.type = BTRFS_DEV_EXTENT_KEY;
256 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
259 ret = btrfs_previous_item(root, path, 0, key.type);
263 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
266 slot = path->slots[0];
267 if (slot >= btrfs_header_nritems(l)) {
268 ret = btrfs_next_leaf(root, path);
275 if (search_start >= search_end) {
279 *start = search_start;
283 *start = last_byte > search_start ?
284 last_byte : search_start;
285 if (search_end <= *start) {
291 btrfs_item_key_to_cpu(l, &key, slot);
293 if (key.objectid < device->devid)
296 if (key.objectid > device->devid)
299 if (key.offset >= search_start && key.offset > last_byte &&
301 if (last_byte < search_start)
302 last_byte = search_start;
303 hole_size = key.offset - last_byte;
304 if (key.offset > last_byte &&
305 hole_size >= num_bytes) {
310 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) {
315 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
316 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
322 /* we have to make sure we didn't find an extent that has already
323 * been allocated by the map tree or the original allocation
325 btrfs_release_path(root, path);
326 BUG_ON(*start < search_start);
328 if (*start + num_bytes > search_end) {
332 /* check for pending inserts here */
336 btrfs_release_path(root, path);
340 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
341 struct btrfs_device *device,
342 u64 chunk_tree, u64 chunk_objectid,
344 u64 num_bytes, u64 *start)
347 struct btrfs_path *path;
348 struct btrfs_root *root = device->dev_root;
349 struct btrfs_dev_extent *extent;
350 struct extent_buffer *leaf;
351 struct btrfs_key key;
353 path = btrfs_alloc_path();
357 ret = find_free_dev_extent(trans, device, path, num_bytes, start);
362 key.objectid = device->devid;
364 key.type = BTRFS_DEV_EXTENT_KEY;
365 ret = btrfs_insert_empty_item(trans, root, path, &key,
369 leaf = path->nodes[0];
370 extent = btrfs_item_ptr(leaf, path->slots[0],
371 struct btrfs_dev_extent);
372 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
373 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
374 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
376 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
377 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
380 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
381 btrfs_mark_buffer_dirty(leaf);
383 btrfs_free_path(path);
387 static int find_next_chunk(struct btrfs_root *root, u64 objectid, u64 *offset)
389 struct btrfs_path *path;
391 struct btrfs_key key;
392 struct btrfs_chunk *chunk;
393 struct btrfs_key found_key;
395 path = btrfs_alloc_path();
398 key.objectid = objectid;
399 key.offset = (u64)-1;
400 key.type = BTRFS_CHUNK_ITEM_KEY;
402 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
408 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
412 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
414 if (found_key.objectid != objectid)
417 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
419 *offset = found_key.offset +
420 btrfs_chunk_length(path->nodes[0], chunk);
425 btrfs_free_path(path);
429 static int find_next_devid(struct btrfs_root *root, struct btrfs_path *path,
433 struct btrfs_key key;
434 struct btrfs_key found_key;
436 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
437 key.type = BTRFS_DEV_ITEM_KEY;
438 key.offset = (u64)-1;
440 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
446 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
451 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
453 *objectid = found_key.offset + 1;
457 btrfs_release_path(root, path);
462 * the device information is stored in the chunk root
463 * the btrfs_device struct should be fully filled in
465 int btrfs_add_device(struct btrfs_trans_handle *trans,
466 struct btrfs_root *root,
467 struct btrfs_device *device)
470 struct btrfs_path *path;
471 struct btrfs_dev_item *dev_item;
472 struct extent_buffer *leaf;
473 struct btrfs_key key;
477 root = root->fs_info->chunk_root;
479 path = btrfs_alloc_path();
483 ret = find_next_devid(root, path, &free_devid);
487 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
488 key.type = BTRFS_DEV_ITEM_KEY;
489 key.offset = free_devid;
491 ret = btrfs_insert_empty_item(trans, root, path, &key,
496 leaf = path->nodes[0];
497 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
499 device->devid = free_devid;
500 btrfs_set_device_id(leaf, dev_item, device->devid);
501 btrfs_set_device_type(leaf, dev_item, device->type);
502 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
503 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
504 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
505 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
506 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
507 btrfs_set_device_group(leaf, dev_item, 0);
508 btrfs_set_device_seek_speed(leaf, dev_item, 0);
509 btrfs_set_device_bandwidth(leaf, dev_item, 0);
511 ptr = (unsigned long)btrfs_device_uuid(dev_item);
512 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
513 btrfs_mark_buffer_dirty(leaf);
517 btrfs_free_path(path);
521 int btrfs_update_device(struct btrfs_trans_handle *trans,
522 struct btrfs_device *device)
525 struct btrfs_path *path;
526 struct btrfs_root *root;
527 struct btrfs_dev_item *dev_item;
528 struct extent_buffer *leaf;
529 struct btrfs_key key;
531 root = device->dev_root->fs_info->chunk_root;
533 path = btrfs_alloc_path();
537 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
538 key.type = BTRFS_DEV_ITEM_KEY;
539 key.offset = device->devid;
541 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
550 leaf = path->nodes[0];
551 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
553 btrfs_set_device_id(leaf, dev_item, device->devid);
554 btrfs_set_device_type(leaf, dev_item, device->type);
555 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
556 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
557 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
558 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
559 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
560 btrfs_mark_buffer_dirty(leaf);
563 btrfs_free_path(path);
567 int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
568 struct btrfs_root *root,
569 struct btrfs_key *key,
570 struct btrfs_chunk *chunk, int item_size)
572 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
573 struct btrfs_disk_key disk_key;
577 array_size = btrfs_super_sys_array_size(super_copy);
578 if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
581 ptr = super_copy->sys_chunk_array + array_size;
582 btrfs_cpu_key_to_disk(&disk_key, key);
583 memcpy(ptr, &disk_key, sizeof(disk_key));
584 ptr += sizeof(disk_key);
585 memcpy(ptr, chunk, item_size);
586 item_size += sizeof(disk_key);
587 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
591 static u64 div_factor(u64 num, int factor)
599 static u64 chunk_bytes_by_type(u64 type, u64 calc_size, int num_stripes,
602 if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
604 else if (type & BTRFS_BLOCK_GROUP_RAID10)
605 return calc_size * (num_stripes / sub_stripes);
607 return calc_size * num_stripes;
611 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
612 struct btrfs_root *extent_root, u64 *start,
613 u64 *num_bytes, u64 type)
616 struct btrfs_fs_info *info = extent_root->fs_info;
617 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
618 struct btrfs_stripe *stripes;
619 struct btrfs_device *device = NULL;
620 struct btrfs_chunk *chunk;
621 struct list_head private_devs;
622 struct list_head *dev_list = &extent_root->fs_info->fs_devices->devices;
623 struct list_head *cur;
624 struct map_lookup *map;
625 int min_stripe_size = 1 * 1024 * 1024;
627 u64 calc_size = 8 * 1024 * 1024;
629 u64 max_chunk_size = 4 * calc_size;
639 int stripe_len = 64 * 1024;
640 struct btrfs_key key;
642 if (list_empty(dev_list)) {
646 if (type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
647 BTRFS_BLOCK_GROUP_RAID10 |
648 BTRFS_BLOCK_GROUP_DUP)) {
649 if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
650 calc_size = 8 * 1024 * 1024;
651 max_chunk_size = calc_size * 2;
652 min_stripe_size = 1 * 1024 * 1024;
653 } else if (type & BTRFS_BLOCK_GROUP_DATA) {
654 calc_size = 1024 * 1024 * 1024;
655 max_chunk_size = 10 * calc_size;
656 min_stripe_size = 64 * 1024 * 1024;
657 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
658 calc_size = 1024 * 1024 * 1024;
659 max_chunk_size = 4 * calc_size;
660 min_stripe_size = 32 * 1024 * 1024;
663 if (type & BTRFS_BLOCK_GROUP_RAID1) {
664 num_stripes = min_t(u64, 2,
665 btrfs_super_num_devices(&info->super_copy));
670 if (type & BTRFS_BLOCK_GROUP_DUP) {
674 if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
675 num_stripes = btrfs_super_num_devices(&info->super_copy);
678 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
679 num_stripes = btrfs_super_num_devices(&info->super_copy);
682 num_stripes &= ~(u32)1;
687 /* we don't want a chunk larger than 10% of the FS */
688 percent_max = div_factor(btrfs_super_total_bytes(&info->super_copy), 1);
689 max_chunk_size = min(percent_max, max_chunk_size);
692 if (chunk_bytes_by_type(type, calc_size, num_stripes, sub_stripes) >
694 calc_size = max_chunk_size;
695 calc_size /= num_stripes;
696 calc_size /= stripe_len;
697 calc_size *= stripe_len;
699 /* we don't want tiny stripes */
700 calc_size = max_t(u64, calc_size, min_stripe_size);
702 calc_size /= stripe_len;
703 calc_size *= stripe_len;
704 INIT_LIST_HEAD(&private_devs);
705 cur = dev_list->next;
708 if (type & BTRFS_BLOCK_GROUP_DUP)
709 min_free = calc_size * 2;
711 min_free = calc_size;
713 /* build a private list of devices we will allocate from */
714 while(index < num_stripes) {
715 device = list_entry(cur, struct btrfs_device, dev_list);
716 avail = device->total_bytes - device->bytes_used;
718 if (avail >= min_free) {
719 list_move_tail(&device->dev_list, &private_devs);
721 if (type & BTRFS_BLOCK_GROUP_DUP)
723 } else if (avail > max_avail)
728 if (index < num_stripes) {
729 list_splice(&private_devs, dev_list);
730 if (index >= min_stripes) {
732 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
733 num_stripes /= sub_stripes;
734 num_stripes *= sub_stripes;
739 if (!looped && max_avail > 0) {
741 calc_size = max_avail;
746 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
747 key.type = BTRFS_CHUNK_ITEM_KEY;
748 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
753 chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
757 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
763 stripes = &chunk->stripe;
764 *num_bytes = chunk_bytes_by_type(type, calc_size,
765 num_stripes, sub_stripes);
767 printk("new chunk type %Lu start %Lu size %Lu\n", type, key.offset, *num_bytes);
768 while(index < num_stripes) {
769 struct btrfs_stripe *stripe;
770 BUG_ON(list_empty(&private_devs));
771 cur = private_devs.next;
772 device = list_entry(cur, struct btrfs_device, dev_list);
774 /* loop over this device again if we're doing a dup group */
775 if (!(type & BTRFS_BLOCK_GROUP_DUP) ||
776 (index == num_stripes - 1))
777 list_move_tail(&device->dev_list, dev_list);
779 ret = btrfs_alloc_dev_extent(trans, device,
780 info->chunk_root->root_key.objectid,
781 BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
782 calc_size, &dev_offset);
784 printk("\talloc chunk size %llu from dev %llu phys %llu\n",
785 (unsigned long long)calc_size,
786 (unsigned long long)device->devid,
787 (unsigned long long)dev_offset);
788 device->bytes_used += calc_size;
789 ret = btrfs_update_device(trans, device);
792 map->stripes[index].dev = device;
793 map->stripes[index].physical = dev_offset;
794 stripe = stripes + index;
795 btrfs_set_stack_stripe_devid(stripe, device->devid);
796 btrfs_set_stack_stripe_offset(stripe, dev_offset);
797 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
798 physical = dev_offset;
801 BUG_ON(!list_empty(&private_devs));
803 /* key was set above */
804 btrfs_set_stack_chunk_length(chunk, *num_bytes);
805 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
806 btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
807 btrfs_set_stack_chunk_type(chunk, type);
808 btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
809 btrfs_set_stack_chunk_io_align(chunk, stripe_len);
810 btrfs_set_stack_chunk_io_width(chunk, stripe_len);
811 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
812 btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
813 map->sector_size = extent_root->sectorsize;
814 map->stripe_len = stripe_len;
815 map->io_align = stripe_len;
816 map->io_width = stripe_len;
818 map->num_stripes = num_stripes;
819 map->sub_stripes = sub_stripes;
821 ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
822 btrfs_chunk_item_size(num_stripes));
824 *start = key.offset;;
826 map->ce.start = key.offset;
827 map->ce.size = *num_bytes;
829 ret = insert_existing_cache_extent(
830 &extent_root->fs_info->mapping_tree.cache_tree,
834 if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
835 ret = btrfs_add_system_chunk(trans, chunk_root, &key,
836 chunk, btrfs_chunk_item_size(num_stripes));
844 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
846 cache_tree_init(&tree->cache_tree);
849 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
851 struct cache_extent *ce;
852 struct map_lookup *map;
856 ce = find_first_cache_extent(&map_tree->cache_tree, logical);
858 BUG_ON(ce->start > logical || ce->start + ce->size < logical);
859 map = container_of(ce, struct map_lookup, ce);
861 offset = logical - ce->start;
862 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
863 ret = map->num_stripes;
864 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
865 ret = map->sub_stripes;
871 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
872 u64 logical, u64 *length,
873 struct btrfs_multi_bio **multi_ret, int mirror_num)
875 struct cache_extent *ce;
876 struct map_lookup *map;
880 int stripes_allocated = 8;
881 int stripes_required = 1;
884 struct btrfs_multi_bio *multi = NULL;
886 if (multi_ret && rw == READ) {
887 stripes_allocated = 1;
891 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
897 ce = find_first_cache_extent(&map_tree->cache_tree, logical);
899 BUG_ON(ce->start > logical || ce->start + ce->size < logical);
900 map = container_of(ce, struct map_lookup, ce);
901 offset = logical - ce->start;
904 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
905 BTRFS_BLOCK_GROUP_DUP)) {
906 stripes_required = map->num_stripes;
907 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
908 stripes_required = map->sub_stripes;
911 /* if our multi bio struct is too small, back off and try again */
912 if (multi_ret && rw == WRITE &&
913 stripes_allocated < stripes_required) {
914 stripes_allocated = map->num_stripes;
920 * stripe_nr counts the total number of stripes we have to stride
921 * to get to this block
923 stripe_nr = stripe_nr / map->stripe_len;
925 stripe_offset = stripe_nr * map->stripe_len;
926 BUG_ON(offset < stripe_offset);
928 /* stripe_offset is the offset of this block in its stripe*/
929 stripe_offset = offset - stripe_offset;
931 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
932 BTRFS_BLOCK_GROUP_RAID10 |
933 BTRFS_BLOCK_GROUP_DUP)) {
934 /* we limit the length of each bio to what fits in a stripe */
935 *length = min_t(u64, ce->size - offset,
936 map->stripe_len - stripe_offset);
938 *length = ce->size - offset;
944 multi->num_stripes = 1;
946 if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
948 multi->num_stripes = map->num_stripes;
950 stripe_index = mirror_num - 1;
952 stripe_index = stripe_nr % map->num_stripes;
953 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
954 int factor = map->num_stripes / map->sub_stripes;
956 stripe_index = stripe_nr % factor;
957 stripe_index *= map->sub_stripes;
960 multi->num_stripes = map->sub_stripes;
962 stripe_index += mirror_num - 1;
964 stripe_index = stripe_nr % map->sub_stripes;
966 stripe_nr = stripe_nr / factor;
967 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
969 multi->num_stripes = map->num_stripes;
971 stripe_index = mirror_num - 1;
974 * after this do_div call, stripe_nr is the number of stripes
975 * on this device we have to walk to find the data, and
976 * stripe_index is the number of our device in the stripe array
978 stripe_index = stripe_nr % map->num_stripes;
979 stripe_nr = stripe_nr / map->num_stripes;
981 BUG_ON(stripe_index >= map->num_stripes);
983 BUG_ON(stripe_index != 0 && multi->num_stripes > 1);
984 for (i = 0; i < multi->num_stripes; i++) {
985 multi->stripes[i].physical =
986 map->stripes[stripe_index].physical + stripe_offset +
987 stripe_nr * map->stripe_len;
988 multi->stripes[i].dev = map->stripes[stripe_index].dev;
996 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
999 struct list_head *head = &root->fs_info->fs_devices->devices;
1001 return __find_device(head, devid, uuid);
1004 int btrfs_bootstrap_super_map(struct btrfs_mapping_tree *map_tree,
1005 struct btrfs_fs_devices *fs_devices)
1007 struct map_lookup *map;
1008 u64 logical = BTRFS_SUPER_INFO_OFFSET;
1009 u64 length = BTRFS_SUPER_INFO_SIZE;
1010 int num_stripes = 0;
1011 int sub_stripes = 0;
1014 struct list_head *cur;
1016 list_for_each(cur, &fs_devices->devices) {
1019 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
1023 map->ce.start = logical;
1024 map->ce.size = length;
1025 map->num_stripes = num_stripes;
1026 map->sub_stripes = sub_stripes;
1027 map->io_width = length;
1028 map->io_align = length;
1029 map->sector_size = length;
1030 map->stripe_len = length;
1031 map->type = BTRFS_BLOCK_GROUP_RAID1;
1034 list_for_each(cur, &fs_devices->devices) {
1035 struct btrfs_device *device = list_entry(cur,
1036 struct btrfs_device,
1038 map->stripes[i].physical = logical;
1039 map->stripes[i].dev = device;
1042 ret = insert_existing_cache_extent(&map_tree->cache_tree, &map->ce);
1043 if (ret == -EEXIST) {
1044 struct cache_extent *old;
1045 struct map_lookup *old_map;
1046 old = find_cache_extent(&map_tree->cache_tree, logical, length);
1047 old_map = container_of(old, struct map_lookup, ce);
1048 remove_cache_extent(&map_tree->cache_tree, old);
1050 ret = insert_existing_cache_extent(&map_tree->cache_tree,
1057 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
1058 struct extent_buffer *leaf,
1059 struct btrfs_chunk *chunk)
1061 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
1062 struct map_lookup *map;
1063 struct cache_extent *ce;
1067 u64 super_offset_diff = 0;
1068 u8 uuid[BTRFS_UUID_SIZE];
1073 logical = key->offset;
1074 length = btrfs_chunk_length(leaf, chunk);
1076 if (logical < BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE) {
1077 super_offset_diff = BTRFS_SUPER_INFO_OFFSET +
1078 BTRFS_SUPER_INFO_SIZE - logical;
1079 logical = BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE;
1082 ce = find_first_cache_extent(&map_tree->cache_tree, logical);
1084 /* already mapped? */
1085 if (ce && ce->start <= logical && ce->start + ce->size > logical) {
1089 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
1090 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
1094 map->ce.start = logical;
1095 map->ce.size = length - super_offset_diff;
1096 map->num_stripes = num_stripes;
1097 map->io_width = btrfs_chunk_io_width(leaf, chunk);
1098 map->io_align = btrfs_chunk_io_align(leaf, chunk);
1099 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
1100 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
1101 map->type = btrfs_chunk_type(leaf, chunk);
1102 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
1104 for (i = 0; i < num_stripes; i++) {
1105 map->stripes[i].physical =
1106 btrfs_stripe_offset_nr(leaf, chunk, i) +
1108 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
1109 read_extent_buffer(leaf, uuid, (unsigned long)
1110 btrfs_stripe_dev_uuid_nr(chunk, i),
1112 map->stripes[i].dev = btrfs_find_device(root, devid, uuid);
1113 if (!map->stripes[i].dev) {
1119 ret = insert_existing_cache_extent(&map_tree->cache_tree, &map->ce);
1125 static int fill_device_from_item(struct extent_buffer *leaf,
1126 struct btrfs_dev_item *dev_item,
1127 struct btrfs_device *device)
1131 device->devid = btrfs_device_id(leaf, dev_item);
1132 device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
1133 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
1134 device->type = btrfs_device_type(leaf, dev_item);
1135 device->io_align = btrfs_device_io_align(leaf, dev_item);
1136 device->io_width = btrfs_device_io_width(leaf, dev_item);
1137 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
1139 ptr = (unsigned long)btrfs_device_uuid(dev_item);
1140 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1145 static int read_one_dev(struct btrfs_root *root,
1146 struct extent_buffer *leaf,
1147 struct btrfs_dev_item *dev_item)
1149 struct btrfs_device *device;
1152 u8 dev_uuid[BTRFS_UUID_SIZE];
1154 devid = btrfs_device_id(leaf, dev_item);
1155 read_extent_buffer(leaf, dev_uuid,
1156 (unsigned long)btrfs_device_uuid(dev_item),
1158 device = btrfs_find_device(root, devid, dev_uuid);
1160 printk("warning devid %llu not found already\n",
1161 (unsigned long long)devid);
1162 device = kmalloc(sizeof(*device), GFP_NOFS);
1165 device->total_ios = 0;
1166 list_add(&device->dev_list,
1167 &root->fs_info->fs_devices->devices);
1170 fill_device_from_item(leaf, dev_item, device);
1171 device->dev_root = root->fs_info->dev_root;
1175 int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
1177 struct btrfs_dev_item *dev_item;
1179 dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
1181 return read_one_dev(root, buf, dev_item);
1184 int btrfs_read_sys_array(struct btrfs_root *root)
1186 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1187 struct extent_buffer *sb = root->fs_info->sb_buffer;
1188 struct btrfs_disk_key *disk_key;
1189 struct btrfs_chunk *chunk;
1190 struct btrfs_key key;
1195 unsigned long sb_ptr;
1199 array_size = btrfs_super_sys_array_size(super_copy);
1202 * we do this loop twice, once for the device items and
1203 * once for all of the chunks. This way there are device
1204 * structs filled in for every chunk
1206 ptr = super_copy->sys_chunk_array;
1207 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
1210 while (cur < array_size) {
1211 disk_key = (struct btrfs_disk_key *)ptr;
1212 btrfs_disk_key_to_cpu(&key, disk_key);
1214 len = sizeof(*disk_key);
1219 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1220 chunk = (struct btrfs_chunk *)sb_ptr;
1221 ret = read_one_chunk(root, &key, sb, chunk);
1223 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
1224 len = btrfs_chunk_item_size(num_stripes);
1235 int btrfs_read_chunk_tree(struct btrfs_root *root)
1237 struct btrfs_path *path;
1238 struct extent_buffer *leaf;
1239 struct btrfs_key key;
1240 struct btrfs_key found_key;
1244 root = root->fs_info->chunk_root;
1246 path = btrfs_alloc_path();
1250 /* first we search for all of the device items, and then we
1251 * read in all of the chunk items. This way we can create chunk
1252 * mappings that reference all of the devices that are afound
1254 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1258 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1260 leaf = path->nodes[0];
1261 slot = path->slots[0];
1262 if (slot >= btrfs_header_nritems(leaf)) {
1263 ret = btrfs_next_leaf(root, path);
1270 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1271 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
1272 if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
1274 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
1275 struct btrfs_dev_item *dev_item;
1276 dev_item = btrfs_item_ptr(leaf, slot,
1277 struct btrfs_dev_item);
1278 ret = read_one_dev(root, leaf, dev_item);
1281 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
1282 struct btrfs_chunk *chunk;
1283 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
1284 ret = read_one_chunk(root, &found_key, leaf, chunk);
1288 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
1290 btrfs_release_path(root, path);
1294 btrfs_free_path(path);