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 cache_extent ce;
35 struct btrfs_device *dev;
38 static LIST_HEAD(fs_uuids);
40 static struct btrfs_device *__find_device(struct list_head *head, u64 devid)
42 struct btrfs_device *dev;
43 struct list_head *cur;
45 list_for_each(cur, head) {
46 dev = list_entry(cur, struct btrfs_device, dev_list);
47 if (dev->devid == devid)
53 static struct btrfs_fs_devices *find_fsid(u8 *fsid)
55 struct list_head *cur;
56 struct btrfs_fs_devices *fs_devices;
58 list_for_each(cur, &fs_uuids) {
59 fs_devices = list_entry(cur, struct btrfs_fs_devices, list);
60 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
66 static int device_list_add(const char *path,
67 struct btrfs_super_block *disk_super,
68 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
70 struct btrfs_device *device;
71 struct btrfs_fs_devices *fs_devices;
72 u64 found_transid = btrfs_super_generation(disk_super);
74 fs_devices = find_fsid(disk_super->fsid);
76 fs_devices = kmalloc(sizeof(*fs_devices), GFP_NOFS);
79 INIT_LIST_HEAD(&fs_devices->devices);
80 list_add(&fs_devices->list, &fs_uuids);
81 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
82 fs_devices->latest_devid = devid;
83 fs_devices->latest_trans = found_transid;
84 fs_devices->lowest_devid = (u64)-1;
87 device = __find_device(&fs_devices->devices, devid);
90 device = kzalloc(sizeof(*device), GFP_NOFS);
92 /* we can safely leave the fs_devices entry around */
95 device->devid = devid;
96 device->name = kstrdup(path, GFP_NOFS);
101 list_add(&device->dev_list, &fs_devices->devices);
104 if (found_transid > fs_devices->latest_trans) {
105 fs_devices->latest_devid = devid;
106 fs_devices->latest_trans = found_transid;
108 if (fs_devices->lowest_devid > devid) {
109 fs_devices->lowest_devid = devid;
110 printk("lowest devid now %Lu\n", devid);
112 *fs_devices_ret = fs_devices;
116 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
118 struct list_head *head = &fs_devices->devices;
119 struct list_head *cur;
120 struct btrfs_device *device;
122 list_for_each(cur, head) {
123 device = list_entry(cur, struct btrfs_device, dev_list);
129 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, int flags)
132 struct list_head *head = &fs_devices->devices;
133 struct list_head *cur;
134 struct btrfs_device *device;
137 list_for_each(cur, head) {
138 device = list_entry(cur, struct btrfs_device, dev_list);
139 fd = open(device->name, flags);
140 printk("opening %s devid %Lu fd %d\n", device->name, device->devid, fd);
145 if (device->devid == fs_devices->latest_devid)
146 fs_devices->latest_bdev = fd;
147 if (device->devid == fs_devices->lowest_devid)
148 fs_devices->lowest_bdev = fd;
153 btrfs_close_devices(fs_devices);
157 int btrfs_scan_one_device(int fd, const char *path,
158 struct btrfs_fs_devices **fs_devices_ret,
159 u64 *total_devs, u64 super_offset)
161 struct btrfs_super_block *disk_super;
171 ret = pread(fd, buf, 4096, super_offset);
176 disk_super = (struct btrfs_super_block *)buf;
177 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
178 sizeof(disk_super->magic))) {
182 devid = le64_to_cpu(disk_super->dev_item.devid);
183 *total_devs = btrfs_super_num_devices(disk_super);
184 printk("found device %Lu on %s\n", devid, path);
185 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
194 * this uses a pretty simple search, the expectation is that it is
195 * called very infrequently and that a given device has a small number
198 static int find_free_dev_extent(struct btrfs_trans_handle *trans,
199 struct btrfs_device *device,
200 struct btrfs_path *path,
201 u64 num_bytes, u64 *start)
203 struct btrfs_key key;
204 struct btrfs_root *root = device->dev_root;
205 struct btrfs_dev_extent *dev_extent = NULL;
208 u64 search_start = 0;
209 u64 search_end = device->total_bytes;
213 struct extent_buffer *l;
218 /* FIXME use last free of some kind */
220 key.objectid = device->devid;
221 key.offset = search_start;
222 key.type = BTRFS_DEV_EXTENT_KEY;
223 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
226 ret = btrfs_previous_item(root, path, 0, key.type);
230 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
233 slot = path->slots[0];
234 if (slot >= btrfs_header_nritems(l)) {
235 ret = btrfs_next_leaf(root, path);
242 if (search_start >= search_end) {
246 *start = search_start;
250 *start = last_byte > search_start ?
251 last_byte : search_start;
252 if (search_end <= *start) {
258 btrfs_item_key_to_cpu(l, &key, slot);
260 if (key.objectid < device->devid)
263 if (key.objectid > device->devid)
266 if (key.offset >= search_start && key.offset > last_byte &&
268 if (last_byte < search_start)
269 last_byte = search_start;
270 hole_size = key.offset - last_byte;
271 if (key.offset > last_byte &&
272 hole_size >= num_bytes) {
277 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) {
282 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
283 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
289 /* we have to make sure we didn't find an extent that has already
290 * been allocated by the map tree or the original allocation
292 btrfs_release_path(root, path);
293 BUG_ON(*start < search_start);
295 if (*start + num_bytes > search_end) {
299 /* check for pending inserts here */
303 btrfs_release_path(root, path);
307 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
308 struct btrfs_device *device,
309 u64 owner, u64 num_bytes, u64 *start)
312 struct btrfs_path *path;
313 struct btrfs_root *root = device->dev_root;
314 struct btrfs_dev_extent *extent;
315 struct extent_buffer *leaf;
316 struct btrfs_key key;
318 path = btrfs_alloc_path();
322 ret = find_free_dev_extent(trans, device, path, num_bytes, start);
327 key.objectid = device->devid;
329 key.type = BTRFS_DEV_EXTENT_KEY;
330 ret = btrfs_insert_empty_item(trans, root, path, &key,
334 leaf = path->nodes[0];
335 extent = btrfs_item_ptr(leaf, path->slots[0],
336 struct btrfs_dev_extent);
337 btrfs_set_dev_extent_owner(leaf, extent, owner);
338 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
339 btrfs_mark_buffer_dirty(leaf);
341 btrfs_free_path(path);
345 static int find_next_chunk(struct btrfs_root *root, u64 *objectid)
347 struct btrfs_path *path;
349 struct btrfs_key key;
350 struct btrfs_key found_key;
352 path = btrfs_alloc_path();
355 key.objectid = (u64)-1;
356 key.offset = (u64)-1;
357 key.type = BTRFS_CHUNK_ITEM_KEY;
359 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
365 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
369 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
371 *objectid = found_key.objectid + found_key.offset;
375 btrfs_free_path(path);
379 static int find_next_devid(struct btrfs_root *root, struct btrfs_path *path,
383 struct btrfs_key key;
384 struct btrfs_key found_key;
386 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
387 key.type = BTRFS_DEV_ITEM_KEY;
388 key.offset = (u64)-1;
390 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
396 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
401 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
403 *objectid = found_key.offset + 1;
407 btrfs_release_path(root, path);
412 * the device information is stored in the chunk root
413 * the btrfs_device struct should be fully filled in
415 int btrfs_add_device(struct btrfs_trans_handle *trans,
416 struct btrfs_root *root,
417 struct btrfs_device *device)
420 struct btrfs_path *path;
421 struct btrfs_dev_item *dev_item;
422 struct extent_buffer *leaf;
423 struct btrfs_key key;
427 root = root->fs_info->chunk_root;
429 path = btrfs_alloc_path();
433 ret = find_next_devid(root, path, &free_devid);
437 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
438 key.type = BTRFS_DEV_ITEM_KEY;
439 key.offset = free_devid;
441 ret = btrfs_insert_empty_item(trans, root, path, &key,
446 leaf = path->nodes[0];
447 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
449 device->devid = free_devid;
450 btrfs_set_device_id(leaf, dev_item, device->devid);
451 btrfs_set_device_type(leaf, dev_item, device->type);
452 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
453 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
454 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
455 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
456 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
458 ptr = (unsigned long)btrfs_device_uuid(dev_item);
459 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_DEV_UUID_SIZE);
460 btrfs_mark_buffer_dirty(leaf);
464 btrfs_free_path(path);
468 int btrfs_update_device(struct btrfs_trans_handle *trans,
469 struct btrfs_device *device)
472 struct btrfs_path *path;
473 struct btrfs_root *root;
474 struct btrfs_dev_item *dev_item;
475 struct extent_buffer *leaf;
476 struct btrfs_key key;
478 root = device->dev_root->fs_info->chunk_root;
480 path = btrfs_alloc_path();
484 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
485 key.type = BTRFS_DEV_ITEM_KEY;
486 key.offset = device->devid;
488 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
497 leaf = path->nodes[0];
498 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
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_mark_buffer_dirty(leaf);
510 btrfs_free_path(path);
514 int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
515 struct btrfs_root *root,
516 struct btrfs_key *key,
517 struct btrfs_chunk *chunk, int item_size)
519 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
520 struct btrfs_disk_key disk_key;
524 array_size = btrfs_super_sys_array_size(super_copy);
525 if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
528 ptr = super_copy->sys_chunk_array + array_size;
529 btrfs_cpu_key_to_disk(&disk_key, key);
530 memcpy(ptr, &disk_key, sizeof(disk_key));
531 ptr += sizeof(disk_key);
532 memcpy(ptr, chunk, item_size);
533 item_size += sizeof(disk_key);
534 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
538 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
539 struct btrfs_root *extent_root, u64 *start,
540 u64 *num_bytes, u64 type)
543 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
544 struct btrfs_stripe *stripes;
545 struct btrfs_device *device = NULL;
546 struct btrfs_chunk *chunk;
547 struct list_head private_devs;
548 struct list_head *dev_list = &extent_root->fs_info->fs_devices->devices;
549 struct list_head *cur;
550 struct map_lookup *map;
552 u64 calc_size = 128 * 1024 * 1024;
559 struct btrfs_key key;
561 if (list_empty(dev_list))
564 INIT_LIST_HEAD(&private_devs);
565 cur = dev_list->next;
567 /* build a private list of devices we will allocate from */
568 while(index < num_stripes) {
569 device = list_entry(cur, struct btrfs_device, dev_list);
570 avail = device->total_bytes - device->bytes_used;
572 if (avail > max_avail)
574 if (avail >= calc_size) {
575 list_move_tail(&device->dev_list, &private_devs);
581 if (index < num_stripes) {
582 list_splice(&private_devs, dev_list);
583 if (!looped && max_avail > 0) {
585 calc_size = max_avail;
591 ret = find_next_chunk(chunk_root, &key.objectid);
595 chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
599 stripes = &chunk->stripe;
601 *num_bytes = calc_size;
603 while(index < num_stripes) {
604 BUG_ON(list_empty(&private_devs));
605 cur = private_devs.next;
606 device = list_entry(cur, struct btrfs_device, dev_list);
607 list_move_tail(&device->dev_list, dev_list);
609 ret = btrfs_alloc_dev_extent(trans, device,
611 calc_size, &dev_offset);
613 printk("alloc chunk size %Lu from dev %Lu\n", calc_size, device->devid);
614 device->bytes_used += calc_size;
615 ret = btrfs_update_device(trans, device);
618 btrfs_set_stack_stripe_devid(stripes + index, device->devid);
619 btrfs_set_stack_stripe_offset(stripes + index, dev_offset);
620 physical = dev_offset;
623 BUG_ON(!list_empty(&private_devs));
625 /* key.objectid was set above */
626 key.offset = *num_bytes;
627 key.type = BTRFS_CHUNK_ITEM_KEY;
628 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
629 btrfs_set_stack_chunk_stripe_len(chunk, 64 * 1024);
630 btrfs_set_stack_chunk_type(chunk, type);
631 btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
632 btrfs_set_stack_chunk_io_align(chunk, extent_root->sectorsize);
633 btrfs_set_stack_chunk_io_width(chunk, extent_root->sectorsize);
634 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
636 ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
637 btrfs_chunk_item_size(num_stripes));
639 *start = key.objectid;
641 map = kmalloc(sizeof(*map), GFP_NOFS);
645 map->ce.start = key.objectid;
646 map->ce.size = key.offset;
648 map->physical = physical;
655 ret = insert_existing_cache_extent(
656 &extent_root->fs_info->mapping_tree.cache_tree,
664 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
666 cache_tree_init(&tree->cache_tree);
669 int btrfs_map_block(struct btrfs_mapping_tree *map_tree,
670 u64 logical, u64 *phys, u64 *length,
671 struct btrfs_device **dev)
673 struct cache_extent *ce;
674 struct map_lookup *map;
677 ce = find_first_cache_extent(&map_tree->cache_tree, logical);
679 BUG_ON(ce->start > logical || ce->start + ce->size < logical);
680 map = container_of(ce, struct map_lookup, ce);
681 offset = logical - ce->start;
682 *phys = map->physical + offset;
683 *length = ce->size - offset;
688 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid)
690 struct list_head *head = &root->fs_info->fs_devices->devices;
692 return __find_device(head, devid);
695 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
696 struct extent_buffer *leaf,
697 struct btrfs_chunk *chunk)
699 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
700 struct map_lookup *map;
701 struct cache_extent *ce;
707 logical = key->objectid;
708 length = key->offset;
709 ce = find_first_cache_extent(&map_tree->cache_tree, logical);
711 /* already mapped? */
712 if (ce && ce->start <= logical && ce->start + ce->size > logical) {
716 map = kmalloc(sizeof(*map), GFP_NOFS);
720 map->ce.start = logical;
721 map->ce.size = length;
723 map->physical = btrfs_stripe_offset_nr(leaf, chunk, 0);
725 devid = btrfs_stripe_devid_nr(leaf, chunk, 0);
726 map->dev = btrfs_find_device(root, devid);
732 ret = insert_existing_cache_extent(&map_tree->cache_tree, &map->ce);
738 static int fill_device_from_item(struct extent_buffer *leaf,
739 struct btrfs_dev_item *dev_item,
740 struct btrfs_device *device)
744 device->devid = btrfs_device_id(leaf, dev_item);
745 device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
746 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
747 device->type = btrfs_device_type(leaf, dev_item);
748 device->io_align = btrfs_device_io_align(leaf, dev_item);
749 device->io_width = btrfs_device_io_width(leaf, dev_item);
750 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
752 ptr = (unsigned long)btrfs_device_uuid(dev_item);
753 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_DEV_UUID_SIZE);
758 static int read_one_dev(struct btrfs_root *root,
759 struct extent_buffer *leaf,
760 struct btrfs_dev_item *dev_item)
762 struct btrfs_device *device;
766 devid = btrfs_device_id(leaf, dev_item);
767 device = btrfs_find_device(root, devid);
769 printk("warning devid %Lu not found already\n", devid);
770 device = kmalloc(sizeof(*device), GFP_NOFS);
773 list_add(&device->dev_list,
774 &root->fs_info->fs_devices->devices);
777 fill_device_from_item(leaf, dev_item, device);
778 device->dev_root = root->fs_info->dev_root;
782 int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
784 struct btrfs_dev_item *dev_item;
786 dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
788 return read_one_dev(root, buf, dev_item);
791 int btrfs_read_sys_array(struct btrfs_root *root)
793 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
794 struct extent_buffer *sb = root->fs_info->sb_buffer;
795 struct btrfs_disk_key *disk_key;
796 struct btrfs_chunk *chunk;
797 struct btrfs_key key;
802 unsigned long sb_ptr;
806 array_size = btrfs_super_sys_array_size(super_copy);
809 * we do this loop twice, once for the device items and
810 * once for all of the chunks. This way there are device
811 * structs filled in for every chunk
813 ptr = super_copy->sys_chunk_array;
814 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
817 while (cur < array_size) {
818 disk_key = (struct btrfs_disk_key *)ptr;
819 btrfs_disk_key_to_cpu(&key, disk_key);
821 len = sizeof(*disk_key);
826 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
827 chunk = (struct btrfs_chunk *)sb_ptr;
828 ret = read_one_chunk(root, &key, sb, chunk);
830 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
831 len = btrfs_chunk_item_size(num_stripes);
842 int btrfs_read_chunk_tree(struct btrfs_root *root)
844 struct btrfs_path *path;
845 struct extent_buffer *leaf;
846 struct btrfs_key key;
847 struct btrfs_key found_key;
851 root = root->fs_info->chunk_root;
853 path = btrfs_alloc_path();
857 /* first we search for all of the device items, and then we
858 * read in all of the chunk items. This way we can create chunk
859 * mappings that reference all of the devices that are afound
861 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
865 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
867 leaf = path->nodes[0];
868 slot = path->slots[0];
869 if (slot >= btrfs_header_nritems(leaf)) {
870 ret = btrfs_next_leaf(root, path);
877 btrfs_item_key_to_cpu(leaf, &found_key, slot);
878 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
879 if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
881 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
882 struct btrfs_dev_item *dev_item;
883 dev_item = btrfs_item_ptr(leaf, slot,
884 struct btrfs_dev_item);
885 ret = read_one_dev(root, leaf, dev_item);
888 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
889 struct btrfs_chunk *chunk;
890 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
891 ret = read_one_chunk(root, &found_key, leaf, chunk);
895 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
897 btrfs_release_path(root, path);
901 btrfs_free_path(path);