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 = kzalloc(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 device->label = kstrdup(disk_super->label, GFP_NOFS);
123 device->total_devs = btrfs_super_num_devices(disk_super);
124 device->super_bytes_used = btrfs_super_bytes_used(disk_super);
125 device->total_bytes =
126 btrfs_stack_device_total_bytes(&disk_super->dev_item);
128 btrfs_stack_device_bytes_used(&disk_super->dev_item);
129 list_add(&device->dev_list, &fs_devices->devices);
130 device->fs_devices = fs_devices;
133 if (found_transid > fs_devices->latest_trans) {
134 fs_devices->latest_devid = devid;
135 fs_devices->latest_trans = found_transid;
137 if (fs_devices->lowest_devid > devid) {
138 fs_devices->lowest_devid = devid;
140 *fs_devices_ret = fs_devices;
144 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
146 struct btrfs_fs_devices *seed_devices;
147 struct list_head *cur;
148 struct btrfs_device *device;
150 list_for_each(cur, &fs_devices->devices) {
151 device = list_entry(cur, struct btrfs_device, dev_list);
154 device->writeable = 0;
157 seed_devices = fs_devices->seed;
158 fs_devices->seed = NULL;
160 fs_devices = seed_devices;
167 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, int flags)
170 struct list_head *head = &fs_devices->devices;
171 struct list_head *cur;
172 struct btrfs_device *device;
175 list_for_each(cur, head) {
176 device = list_entry(cur, struct btrfs_device, dev_list);
178 fd = open(device->name, flags);
184 if (device->devid == fs_devices->latest_devid)
185 fs_devices->latest_bdev = fd;
186 if (device->devid == fs_devices->lowest_devid)
187 fs_devices->lowest_bdev = fd;
190 device->writeable = 1;
194 btrfs_close_devices(fs_devices);
198 int btrfs_scan_one_device(int fd, const char *path,
199 struct btrfs_fs_devices **fs_devices_ret,
200 u64 *total_devs, u64 super_offset)
202 struct btrfs_super_block *disk_super;
213 disk_super = (struct btrfs_super_block *)buf;
214 ret = btrfs_read_dev_super(fd, disk_super, super_offset);
219 devid = le64_to_cpu(disk_super->dev_item.devid);
220 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_METADUMP)
223 *total_devs = btrfs_super_num_devices(disk_super);
224 uuid_unparse(disk_super->fsid, uuidbuf);
226 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
235 * this uses a pretty simple search, the expectation is that it is
236 * called very infrequently and that a given device has a small number
239 static int find_free_dev_extent(struct btrfs_trans_handle *trans,
240 struct btrfs_device *device,
241 struct btrfs_path *path,
242 u64 num_bytes, u64 *start)
244 struct btrfs_key key;
245 struct btrfs_root *root = device->dev_root;
246 struct btrfs_dev_extent *dev_extent = NULL;
249 u64 search_start = 0;
250 u64 search_end = device->total_bytes;
254 struct extent_buffer *l;
259 /* FIXME use last free of some kind */
261 /* we don't want to overwrite the superblock on the drive,
262 * so we make sure to start at an offset of at least 1MB
264 search_start = max((u64)1024 * 1024, search_start);
266 if (root->fs_info->alloc_start + num_bytes <= device->total_bytes)
267 search_start = max(root->fs_info->alloc_start, search_start);
269 key.objectid = device->devid;
270 key.offset = search_start;
271 key.type = BTRFS_DEV_EXTENT_KEY;
272 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
275 ret = btrfs_previous_item(root, path, 0, key.type);
279 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
282 slot = path->slots[0];
283 if (slot >= btrfs_header_nritems(l)) {
284 ret = btrfs_next_leaf(root, path);
291 if (search_start >= search_end) {
295 *start = search_start;
299 *start = last_byte > search_start ?
300 last_byte : search_start;
301 if (search_end <= *start) {
307 btrfs_item_key_to_cpu(l, &key, slot);
309 if (key.objectid < device->devid)
312 if (key.objectid > device->devid)
315 if (key.offset >= search_start && key.offset > last_byte &&
317 if (last_byte < search_start)
318 last_byte = search_start;
319 hole_size = key.offset - last_byte;
320 if (key.offset > last_byte &&
321 hole_size >= num_bytes) {
326 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) {
331 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
332 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
338 /* we have to make sure we didn't find an extent that has already
339 * been allocated by the map tree or the original allocation
341 btrfs_release_path(root, path);
342 BUG_ON(*start < search_start);
344 if (*start + num_bytes > search_end) {
348 /* check for pending inserts here */
352 btrfs_release_path(root, path);
356 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
357 struct btrfs_device *device,
358 u64 chunk_tree, u64 chunk_objectid,
360 u64 num_bytes, u64 *start)
363 struct btrfs_path *path;
364 struct btrfs_root *root = device->dev_root;
365 struct btrfs_dev_extent *extent;
366 struct extent_buffer *leaf;
367 struct btrfs_key key;
369 path = btrfs_alloc_path();
373 ret = find_free_dev_extent(trans, device, path, num_bytes, start);
378 key.objectid = device->devid;
380 key.type = BTRFS_DEV_EXTENT_KEY;
381 ret = btrfs_insert_empty_item(trans, root, path, &key,
385 leaf = path->nodes[0];
386 extent = btrfs_item_ptr(leaf, path->slots[0],
387 struct btrfs_dev_extent);
388 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
389 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
390 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
392 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
393 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
396 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
397 btrfs_mark_buffer_dirty(leaf);
399 btrfs_free_path(path);
403 static int find_next_chunk(struct btrfs_root *root, u64 objectid, u64 *offset)
405 struct btrfs_path *path;
407 struct btrfs_key key;
408 struct btrfs_chunk *chunk;
409 struct btrfs_key found_key;
411 path = btrfs_alloc_path();
414 key.objectid = objectid;
415 key.offset = (u64)-1;
416 key.type = BTRFS_CHUNK_ITEM_KEY;
418 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
424 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
428 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
430 if (found_key.objectid != objectid)
433 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
435 *offset = found_key.offset +
436 btrfs_chunk_length(path->nodes[0], chunk);
441 btrfs_free_path(path);
445 static int find_next_devid(struct btrfs_root *root, struct btrfs_path *path,
449 struct btrfs_key key;
450 struct btrfs_key found_key;
452 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
453 key.type = BTRFS_DEV_ITEM_KEY;
454 key.offset = (u64)-1;
456 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
462 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
467 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
469 *objectid = found_key.offset + 1;
473 btrfs_release_path(root, path);
478 * the device information is stored in the chunk root
479 * the btrfs_device struct should be fully filled in
481 int btrfs_add_device(struct btrfs_trans_handle *trans,
482 struct btrfs_root *root,
483 struct btrfs_device *device)
486 struct btrfs_path *path;
487 struct btrfs_dev_item *dev_item;
488 struct extent_buffer *leaf;
489 struct btrfs_key key;
493 root = root->fs_info->chunk_root;
495 path = btrfs_alloc_path();
499 ret = find_next_devid(root, path, &free_devid);
503 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
504 key.type = BTRFS_DEV_ITEM_KEY;
505 key.offset = free_devid;
507 ret = btrfs_insert_empty_item(trans, root, path, &key,
512 leaf = path->nodes[0];
513 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
515 device->devid = free_devid;
516 btrfs_set_device_id(leaf, dev_item, device->devid);
517 btrfs_set_device_generation(leaf, dev_item, 0);
518 btrfs_set_device_type(leaf, dev_item, device->type);
519 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
520 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
521 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
522 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
523 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
524 btrfs_set_device_group(leaf, dev_item, 0);
525 btrfs_set_device_seek_speed(leaf, dev_item, 0);
526 btrfs_set_device_bandwidth(leaf, dev_item, 0);
527 btrfs_set_device_start_offset(leaf, dev_item, 0);
529 ptr = (unsigned long)btrfs_device_uuid(dev_item);
530 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
531 ptr = (unsigned long)btrfs_device_fsid(dev_item);
532 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
533 btrfs_mark_buffer_dirty(leaf);
537 btrfs_free_path(path);
541 int btrfs_update_device(struct btrfs_trans_handle *trans,
542 struct btrfs_device *device)
545 struct btrfs_path *path;
546 struct btrfs_root *root;
547 struct btrfs_dev_item *dev_item;
548 struct extent_buffer *leaf;
549 struct btrfs_key key;
551 root = device->dev_root->fs_info->chunk_root;
553 path = btrfs_alloc_path();
557 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
558 key.type = BTRFS_DEV_ITEM_KEY;
559 key.offset = device->devid;
561 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
570 leaf = path->nodes[0];
571 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
573 btrfs_set_device_id(leaf, dev_item, device->devid);
574 btrfs_set_device_type(leaf, dev_item, device->type);
575 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
576 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
577 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
578 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
579 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
580 btrfs_mark_buffer_dirty(leaf);
583 btrfs_free_path(path);
587 int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
588 struct btrfs_root *root,
589 struct btrfs_key *key,
590 struct btrfs_chunk *chunk, int item_size)
592 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
593 struct btrfs_disk_key disk_key;
597 array_size = btrfs_super_sys_array_size(super_copy);
598 if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
601 ptr = super_copy->sys_chunk_array + array_size;
602 btrfs_cpu_key_to_disk(&disk_key, key);
603 memcpy(ptr, &disk_key, sizeof(disk_key));
604 ptr += sizeof(disk_key);
605 memcpy(ptr, chunk, item_size);
606 item_size += sizeof(disk_key);
607 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
611 static u64 div_factor(u64 num, int factor)
619 static u64 chunk_bytes_by_type(u64 type, u64 calc_size, int num_stripes,
622 if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
624 else if (type & BTRFS_BLOCK_GROUP_RAID10)
625 return calc_size * (num_stripes / sub_stripes);
627 return calc_size * num_stripes;
631 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
632 struct btrfs_root *extent_root, u64 *start,
633 u64 *num_bytes, u64 type)
636 struct btrfs_fs_info *info = extent_root->fs_info;
637 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
638 struct btrfs_stripe *stripes;
639 struct btrfs_device *device = NULL;
640 struct btrfs_chunk *chunk;
641 struct list_head private_devs;
642 struct list_head *dev_list = &extent_root->fs_info->fs_devices->devices;
643 struct list_head *cur;
644 struct map_lookup *map;
645 int min_stripe_size = 1 * 1024 * 1024;
647 u64 calc_size = 8 * 1024 * 1024;
649 u64 max_chunk_size = 4 * calc_size;
659 int stripe_len = 64 * 1024;
660 struct btrfs_key key;
662 if (list_empty(dev_list)) {
666 if (type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
667 BTRFS_BLOCK_GROUP_RAID10 |
668 BTRFS_BLOCK_GROUP_DUP)) {
669 if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
670 calc_size = 8 * 1024 * 1024;
671 max_chunk_size = calc_size * 2;
672 min_stripe_size = 1 * 1024 * 1024;
673 } else if (type & BTRFS_BLOCK_GROUP_DATA) {
674 calc_size = 1024 * 1024 * 1024;
675 max_chunk_size = 10 * calc_size;
676 min_stripe_size = 64 * 1024 * 1024;
677 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
678 calc_size = 1024 * 1024 * 1024;
679 max_chunk_size = 4 * calc_size;
680 min_stripe_size = 32 * 1024 * 1024;
683 if (type & BTRFS_BLOCK_GROUP_RAID1) {
684 num_stripes = min_t(u64, 2,
685 btrfs_super_num_devices(&info->super_copy));
690 if (type & BTRFS_BLOCK_GROUP_DUP) {
694 if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
695 num_stripes = btrfs_super_num_devices(&info->super_copy);
698 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
699 num_stripes = btrfs_super_num_devices(&info->super_copy);
702 num_stripes &= ~(u32)1;
707 /* we don't want a chunk larger than 10% of the FS */
708 percent_max = div_factor(btrfs_super_total_bytes(&info->super_copy), 1);
709 max_chunk_size = min(percent_max, max_chunk_size);
712 if (chunk_bytes_by_type(type, calc_size, num_stripes, sub_stripes) >
714 calc_size = max_chunk_size;
715 calc_size /= num_stripes;
716 calc_size /= stripe_len;
717 calc_size *= stripe_len;
719 /* we don't want tiny stripes */
720 calc_size = max_t(u64, calc_size, min_stripe_size);
722 calc_size /= stripe_len;
723 calc_size *= stripe_len;
724 INIT_LIST_HEAD(&private_devs);
725 cur = dev_list->next;
728 if (type & BTRFS_BLOCK_GROUP_DUP)
729 min_free = calc_size * 2;
731 min_free = calc_size;
733 /* build a private list of devices we will allocate from */
734 while(index < num_stripes) {
735 device = list_entry(cur, struct btrfs_device, dev_list);
736 avail = device->total_bytes - device->bytes_used;
738 if (avail >= min_free) {
739 list_move_tail(&device->dev_list, &private_devs);
741 if (type & BTRFS_BLOCK_GROUP_DUP)
743 } else if (avail > max_avail)
748 if (index < num_stripes) {
749 list_splice(&private_devs, dev_list);
750 if (index >= min_stripes) {
752 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
753 num_stripes /= sub_stripes;
754 num_stripes *= sub_stripes;
759 if (!looped && max_avail > 0) {
761 calc_size = max_avail;
766 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
767 key.type = BTRFS_CHUNK_ITEM_KEY;
768 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
773 chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
777 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
783 stripes = &chunk->stripe;
784 *num_bytes = chunk_bytes_by_type(type, calc_size,
785 num_stripes, sub_stripes);
787 while(index < num_stripes) {
788 struct btrfs_stripe *stripe;
789 BUG_ON(list_empty(&private_devs));
790 cur = private_devs.next;
791 device = list_entry(cur, struct btrfs_device, dev_list);
793 /* loop over this device again if we're doing a dup group */
794 if (!(type & BTRFS_BLOCK_GROUP_DUP) ||
795 (index == num_stripes - 1))
796 list_move_tail(&device->dev_list, dev_list);
798 ret = btrfs_alloc_dev_extent(trans, device,
799 info->chunk_root->root_key.objectid,
800 BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
801 calc_size, &dev_offset);
804 device->bytes_used += calc_size;
805 ret = btrfs_update_device(trans, device);
808 map->stripes[index].dev = device;
809 map->stripes[index].physical = dev_offset;
810 stripe = stripes + index;
811 btrfs_set_stack_stripe_devid(stripe, device->devid);
812 btrfs_set_stack_stripe_offset(stripe, dev_offset);
813 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
814 physical = dev_offset;
817 BUG_ON(!list_empty(&private_devs));
819 /* key was set above */
820 btrfs_set_stack_chunk_length(chunk, *num_bytes);
821 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
822 btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
823 btrfs_set_stack_chunk_type(chunk, type);
824 btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
825 btrfs_set_stack_chunk_io_align(chunk, stripe_len);
826 btrfs_set_stack_chunk_io_width(chunk, stripe_len);
827 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
828 btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
829 map->sector_size = extent_root->sectorsize;
830 map->stripe_len = stripe_len;
831 map->io_align = stripe_len;
832 map->io_width = stripe_len;
834 map->num_stripes = num_stripes;
835 map->sub_stripes = sub_stripes;
837 ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
838 btrfs_chunk_item_size(num_stripes));
840 *start = key.offset;;
842 map->ce.start = key.offset;
843 map->ce.size = *num_bytes;
845 ret = insert_existing_cache_extent(
846 &extent_root->fs_info->mapping_tree.cache_tree,
850 if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
851 ret = btrfs_add_system_chunk(trans, chunk_root, &key,
852 chunk, btrfs_chunk_item_size(num_stripes));
860 int btrfs_alloc_data_chunk(struct btrfs_trans_handle *trans,
861 struct btrfs_root *extent_root, u64 *start,
862 u64 num_bytes, u64 type)
865 struct btrfs_fs_info *info = extent_root->fs_info;
866 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
867 struct btrfs_stripe *stripes;
868 struct btrfs_device *device = NULL;
869 struct btrfs_chunk *chunk;
870 struct list_head *dev_list = &extent_root->fs_info->fs_devices->devices;
871 struct list_head *cur;
872 struct map_lookup *map;
874 u64 calc_size = 8 * 1024 * 1024;
879 int stripe_len = 64 * 1024;
880 struct btrfs_key key;
882 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
883 key.type = BTRFS_CHUNK_ITEM_KEY;
884 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
889 chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
893 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
899 stripes = &chunk->stripe;
900 calc_size = num_bytes;
903 cur = dev_list->next;
904 device = list_entry(cur, struct btrfs_device, dev_list);
906 while (index < num_stripes) {
907 struct btrfs_stripe *stripe;
909 ret = btrfs_alloc_dev_extent(trans, device,
910 info->chunk_root->root_key.objectid,
911 BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
912 calc_size, &dev_offset);
915 device->bytes_used += calc_size;
916 ret = btrfs_update_device(trans, device);
919 map->stripes[index].dev = device;
920 map->stripes[index].physical = dev_offset;
921 stripe = stripes + index;
922 btrfs_set_stack_stripe_devid(stripe, device->devid);
923 btrfs_set_stack_stripe_offset(stripe, dev_offset);
924 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
925 physical = dev_offset;
929 /* key was set above */
930 btrfs_set_stack_chunk_length(chunk, num_bytes);
931 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
932 btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
933 btrfs_set_stack_chunk_type(chunk, type);
934 btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
935 btrfs_set_stack_chunk_io_align(chunk, stripe_len);
936 btrfs_set_stack_chunk_io_width(chunk, stripe_len);
937 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
938 btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
939 map->sector_size = extent_root->sectorsize;
940 map->stripe_len = stripe_len;
941 map->io_align = stripe_len;
942 map->io_width = stripe_len;
944 map->num_stripes = num_stripes;
945 map->sub_stripes = sub_stripes;
947 ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
948 btrfs_chunk_item_size(num_stripes));
952 map->ce.start = key.offset;
953 map->ce.size = num_bytes;
955 ret = insert_existing_cache_extent(
956 &extent_root->fs_info->mapping_tree.cache_tree,
964 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
966 cache_tree_init(&tree->cache_tree);
969 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
971 struct cache_extent *ce;
972 struct map_lookup *map;
976 ce = find_first_cache_extent(&map_tree->cache_tree, logical);
978 BUG_ON(ce->start > logical || ce->start + ce->size < logical);
979 map = container_of(ce, struct map_lookup, ce);
981 offset = logical - ce->start;
982 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
983 ret = map->num_stripes;
984 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
985 ret = map->sub_stripes;
991 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
992 u64 chunk_start, u64 physical, u64 devid,
993 u64 **logical, int *naddrs, int *stripe_len)
995 struct cache_extent *ce;
996 struct map_lookup *map;
1003 ce = find_first_cache_extent(&map_tree->cache_tree, chunk_start);
1005 map = container_of(ce, struct map_lookup, ce);
1008 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1009 length = ce->size / (map->num_stripes / map->sub_stripes);
1010 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
1011 length = ce->size / map->num_stripes;
1013 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
1015 for (i = 0; i < map->num_stripes; i++) {
1016 if (devid && map->stripes[i].dev->devid != devid)
1018 if (map->stripes[i].physical > physical ||
1019 map->stripes[i].physical + length <= physical)
1022 stripe_nr = (physical - map->stripes[i].physical) /
1025 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1026 stripe_nr = (stripe_nr * map->num_stripes + i) /
1028 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
1029 stripe_nr = stripe_nr * map->num_stripes + i;
1031 bytenr = ce->start + stripe_nr * map->stripe_len;
1032 for (j = 0; j < nr; j++) {
1033 if (buf[j] == bytenr)
1042 *stripe_len = map->stripe_len;
1047 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
1048 u64 logical, u64 *length,
1049 struct btrfs_multi_bio **multi_ret, int mirror_num)
1051 struct cache_extent *ce;
1052 struct map_lookup *map;
1056 int stripes_allocated = 8;
1057 int stripes_required = 1;
1060 struct btrfs_multi_bio *multi = NULL;
1062 if (multi_ret && rw == READ) {
1063 stripes_allocated = 1;
1067 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
1073 ce = find_first_cache_extent(&map_tree->cache_tree, logical);
1075 BUG_ON(ce->start > logical || ce->start + ce->size < logical);
1076 map = container_of(ce, struct map_lookup, ce);
1077 offset = logical - ce->start;
1080 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
1081 BTRFS_BLOCK_GROUP_DUP)) {
1082 stripes_required = map->num_stripes;
1083 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1084 stripes_required = map->sub_stripes;
1087 /* if our multi bio struct is too small, back off and try again */
1088 if (multi_ret && rw == WRITE &&
1089 stripes_allocated < stripes_required) {
1090 stripes_allocated = map->num_stripes;
1096 * stripe_nr counts the total number of stripes we have to stride
1097 * to get to this block
1099 stripe_nr = stripe_nr / map->stripe_len;
1101 stripe_offset = stripe_nr * map->stripe_len;
1102 BUG_ON(offset < stripe_offset);
1104 /* stripe_offset is the offset of this block in its stripe*/
1105 stripe_offset = offset - stripe_offset;
1107 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
1108 BTRFS_BLOCK_GROUP_RAID10 |
1109 BTRFS_BLOCK_GROUP_DUP)) {
1110 /* we limit the length of each bio to what fits in a stripe */
1111 *length = min_t(u64, ce->size - offset,
1112 map->stripe_len - stripe_offset);
1114 *length = ce->size - offset;
1120 multi->num_stripes = 1;
1122 if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
1124 multi->num_stripes = map->num_stripes;
1125 else if (mirror_num)
1126 stripe_index = mirror_num - 1;
1128 stripe_index = stripe_nr % map->num_stripes;
1129 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1130 int factor = map->num_stripes / map->sub_stripes;
1132 stripe_index = stripe_nr % factor;
1133 stripe_index *= map->sub_stripes;
1136 multi->num_stripes = map->sub_stripes;
1137 else if (mirror_num)
1138 stripe_index += mirror_num - 1;
1140 stripe_index = stripe_nr % map->sub_stripes;
1142 stripe_nr = stripe_nr / factor;
1143 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
1145 multi->num_stripes = map->num_stripes;
1146 else if (mirror_num)
1147 stripe_index = mirror_num - 1;
1150 * after this do_div call, stripe_nr is the number of stripes
1151 * on this device we have to walk to find the data, and
1152 * stripe_index is the number of our device in the stripe array
1154 stripe_index = stripe_nr % map->num_stripes;
1155 stripe_nr = stripe_nr / map->num_stripes;
1157 BUG_ON(stripe_index >= map->num_stripes);
1159 for (i = 0; i < multi->num_stripes; i++) {
1160 multi->stripes[i].physical =
1161 map->stripes[stripe_index].physical + stripe_offset +
1162 stripe_nr * map->stripe_len;
1163 multi->stripes[i].dev = map->stripes[stripe_index].dev;
1171 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
1174 struct btrfs_device *device;
1175 struct btrfs_fs_devices *cur_devices;
1177 cur_devices = root->fs_info->fs_devices;
1178 while (cur_devices) {
1180 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
1181 device = __find_device(&cur_devices->devices,
1186 cur_devices = cur_devices->seed;
1191 int btrfs_bootstrap_super_map(struct btrfs_mapping_tree *map_tree,
1192 struct btrfs_fs_devices *fs_devices)
1194 struct map_lookup *map;
1195 u64 logical = BTRFS_SUPER_INFO_OFFSET;
1196 u64 length = BTRFS_SUPER_INFO_SIZE;
1197 int num_stripes = 0;
1198 int sub_stripes = 0;
1201 struct list_head *cur;
1203 list_for_each(cur, &fs_devices->devices) {
1206 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
1210 map->ce.start = logical;
1211 map->ce.size = length;
1212 map->num_stripes = num_stripes;
1213 map->sub_stripes = sub_stripes;
1214 map->io_width = length;
1215 map->io_align = length;
1216 map->sector_size = length;
1217 map->stripe_len = length;
1218 map->type = BTRFS_BLOCK_GROUP_RAID1;
1221 list_for_each(cur, &fs_devices->devices) {
1222 struct btrfs_device *device = list_entry(cur,
1223 struct btrfs_device,
1225 map->stripes[i].physical = logical;
1226 map->stripes[i].dev = device;
1229 ret = insert_existing_cache_extent(&map_tree->cache_tree, &map->ce);
1230 if (ret == -EEXIST) {
1231 struct cache_extent *old;
1232 struct map_lookup *old_map;
1233 old = find_cache_extent(&map_tree->cache_tree, logical, length);
1234 old_map = container_of(old, struct map_lookup, ce);
1235 remove_cache_extent(&map_tree->cache_tree, old);
1237 ret = insert_existing_cache_extent(&map_tree->cache_tree,
1244 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
1246 struct cache_extent *ce;
1247 struct map_lookup *map;
1248 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
1252 ce = find_first_cache_extent(&map_tree->cache_tree, chunk_offset);
1255 map = container_of(ce, struct map_lookup, ce);
1256 for (i = 0; i < map->num_stripes; i++) {
1257 if (!map->stripes[i].dev->writeable) {
1266 static struct btrfs_device *fill_missing_device(u64 devid)
1268 struct btrfs_device *device;
1270 device = kzalloc(sizeof(*device), GFP_NOFS);
1271 device->devid = devid;
1276 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
1277 struct extent_buffer *leaf,
1278 struct btrfs_chunk *chunk)
1280 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
1281 struct map_lookup *map;
1282 struct cache_extent *ce;
1286 u8 uuid[BTRFS_UUID_SIZE];
1291 logical = key->offset;
1292 length = btrfs_chunk_length(leaf, chunk);
1294 ce = find_first_cache_extent(&map_tree->cache_tree, logical);
1296 /* already mapped? */
1297 if (ce && ce->start <= logical && ce->start + ce->size > logical) {
1301 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
1302 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
1306 map->ce.start = logical;
1307 map->ce.size = length;
1308 map->num_stripes = num_stripes;
1309 map->io_width = btrfs_chunk_io_width(leaf, chunk);
1310 map->io_align = btrfs_chunk_io_align(leaf, chunk);
1311 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
1312 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
1313 map->type = btrfs_chunk_type(leaf, chunk);
1314 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
1316 for (i = 0; i < num_stripes; i++) {
1317 map->stripes[i].physical =
1318 btrfs_stripe_offset_nr(leaf, chunk, i);
1319 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
1320 read_extent_buffer(leaf, uuid, (unsigned long)
1321 btrfs_stripe_dev_uuid_nr(chunk, i),
1323 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
1325 if (!map->stripes[i].dev) {
1326 map->stripes[i].dev = fill_missing_device(devid);
1327 printf("warning, device %llu is missing\n",
1328 (unsigned long long)devid);
1332 ret = insert_existing_cache_extent(&map_tree->cache_tree, &map->ce);
1338 static int fill_device_from_item(struct extent_buffer *leaf,
1339 struct btrfs_dev_item *dev_item,
1340 struct btrfs_device *device)
1344 device->devid = btrfs_device_id(leaf, dev_item);
1345 device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
1346 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
1347 device->type = btrfs_device_type(leaf, dev_item);
1348 device->io_align = btrfs_device_io_align(leaf, dev_item);
1349 device->io_width = btrfs_device_io_width(leaf, dev_item);
1350 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
1352 ptr = (unsigned long)btrfs_device_uuid(dev_item);
1353 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1358 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
1360 struct btrfs_fs_devices *fs_devices;
1363 fs_devices = root->fs_info->fs_devices->seed;
1364 while (fs_devices) {
1365 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
1369 fs_devices = fs_devices->seed;
1372 fs_devices = find_fsid(fsid);
1378 ret = btrfs_open_devices(fs_devices, O_RDONLY);
1382 fs_devices->seed = root->fs_info->fs_devices->seed;
1383 root->fs_info->fs_devices->seed = fs_devices;
1388 static int read_one_dev(struct btrfs_root *root,
1389 struct extent_buffer *leaf,
1390 struct btrfs_dev_item *dev_item)
1392 struct btrfs_device *device;
1395 u8 fs_uuid[BTRFS_UUID_SIZE];
1396 u8 dev_uuid[BTRFS_UUID_SIZE];
1398 devid = btrfs_device_id(leaf, dev_item);
1399 read_extent_buffer(leaf, dev_uuid,
1400 (unsigned long)btrfs_device_uuid(dev_item),
1402 read_extent_buffer(leaf, fs_uuid,
1403 (unsigned long)btrfs_device_fsid(dev_item),
1406 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
1407 ret = open_seed_devices(root, fs_uuid);
1412 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1414 printk("warning devid %llu not found already\n",
1415 (unsigned long long)devid);
1416 device = kmalloc(sizeof(*device), GFP_NOFS);
1419 device->total_ios = 0;
1420 list_add(&device->dev_list,
1421 &root->fs_info->fs_devices->devices);
1424 fill_device_from_item(leaf, dev_item, device);
1425 device->dev_root = root->fs_info->dev_root;
1429 int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
1431 struct btrfs_dev_item *dev_item;
1433 dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
1435 return read_one_dev(root, buf, dev_item);
1438 int btrfs_read_sys_array(struct btrfs_root *root)
1440 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1441 struct extent_buffer *sb;
1442 struct btrfs_disk_key *disk_key;
1443 struct btrfs_chunk *chunk;
1444 struct btrfs_key key;
1449 unsigned long sb_ptr;
1453 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
1454 BTRFS_SUPER_INFO_SIZE);
1457 btrfs_set_buffer_uptodate(sb);
1458 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
1459 array_size = btrfs_super_sys_array_size(super_copy);
1462 * we do this loop twice, once for the device items and
1463 * once for all of the chunks. This way there are device
1464 * structs filled in for every chunk
1466 ptr = super_copy->sys_chunk_array;
1467 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
1470 while (cur < array_size) {
1471 disk_key = (struct btrfs_disk_key *)ptr;
1472 btrfs_disk_key_to_cpu(&key, disk_key);
1474 len = sizeof(*disk_key);
1479 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1480 chunk = (struct btrfs_chunk *)sb_ptr;
1481 ret = read_one_chunk(root, &key, sb, chunk);
1483 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
1484 len = btrfs_chunk_item_size(num_stripes);
1492 free_extent_buffer(sb);
1496 int btrfs_read_chunk_tree(struct btrfs_root *root)
1498 struct btrfs_path *path;
1499 struct extent_buffer *leaf;
1500 struct btrfs_key key;
1501 struct btrfs_key found_key;
1505 root = root->fs_info->chunk_root;
1507 path = btrfs_alloc_path();
1511 /* first we search for all of the device items, and then we
1512 * read in all of the chunk items. This way we can create chunk
1513 * mappings that reference all of the devices that are afound
1515 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1519 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1521 leaf = path->nodes[0];
1522 slot = path->slots[0];
1523 if (slot >= btrfs_header_nritems(leaf)) {
1524 ret = btrfs_next_leaf(root, path);
1531 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1532 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
1533 if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
1535 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
1536 struct btrfs_dev_item *dev_item;
1537 dev_item = btrfs_item_ptr(leaf, slot,
1538 struct btrfs_dev_item);
1539 ret = read_one_dev(root, leaf, dev_item);
1542 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
1543 struct btrfs_chunk *chunk;
1544 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
1545 ret = read_one_chunk(root, &found_key, leaf, chunk);
1550 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
1552 btrfs_release_path(root, path);
1556 btrfs_free_path(path);
1562 struct list_head *btrfs_scanned_uuids(void)