Better block record keeping, real mkfs
[btrfs-progs-unstable.git] / disk-io.c
blob0322c55162cba155c6b40cdb123ea20dba595d63
1 #define _XOPEN_SOURCE 500
2 #include <stdio.h>
3 #include <stdlib.h>
4 #include <sys/types.h>
5 #include <sys/stat.h>
6 #include <fcntl.h>
7 #include <unistd.h>
8 #include "kerncompat.h"
9 #include "radix-tree.h"
10 #include "ctree.h"
11 #include "disk-io.h"
12 #include "transaction.h"
14 static int allocated_blocks = 0;
15 int cache_max = 10000;
17 static int check_tree_block(struct btrfs_root *root, struct btrfs_buffer *buf)
19 if (buf->blocknr != btrfs_header_blocknr(&buf->node.header))
20 BUG();
21 if (root->node && btrfs_header_parentid(&buf->node.header) !=
22 btrfs_header_parentid(&root->node->node.header))
23 BUG();
24 return 0;
27 static int free_some_buffers(struct btrfs_root *root)
29 struct list_head *node, *next;
30 struct btrfs_buffer *b;
31 if (root->fs_info->cache_size < cache_max)
32 return 0;
33 list_for_each_safe(node, next, &root->fs_info->cache) {
34 b = list_entry(node, struct btrfs_buffer, cache);
35 if (b->count == 1) {
36 BUG_ON(!list_empty(&b->dirty));
37 list_del_init(&b->cache);
38 btrfs_block_release(root, b);
39 if (root->fs_info->cache_size < cache_max)
40 break;
43 return 0;
46 struct btrfs_buffer *alloc_tree_block(struct btrfs_root *root, u64 blocknr)
48 struct btrfs_buffer *buf;
49 int ret;
51 buf = malloc(sizeof(struct btrfs_buffer) + root->blocksize);
52 if (!buf)
53 return buf;
54 allocated_blocks++;
55 buf->blocknr = blocknr;
56 buf->count = 2;
57 INIT_LIST_HEAD(&buf->dirty);
58 free_some_buffers(root);
59 radix_tree_preload(GFP_KERNEL);
60 ret = radix_tree_insert(&root->fs_info->cache_radix, blocknr, buf);
61 radix_tree_preload_end();
62 list_add_tail(&buf->cache, &root->fs_info->cache);
63 root->fs_info->cache_size++;
64 if (ret) {
65 free(buf);
66 return NULL;
68 return buf;
71 struct btrfs_buffer *find_tree_block(struct btrfs_root *root, u64 blocknr)
73 struct btrfs_buffer *buf;
74 buf = radix_tree_lookup(&root->fs_info->cache_radix, blocknr);
75 if (buf) {
76 buf->count++;
77 } else {
78 buf = alloc_tree_block(root, blocknr);
79 if (!buf) {
80 BUG();
81 return NULL;
84 return buf;
87 struct btrfs_buffer *read_tree_block(struct btrfs_root *root, u64 blocknr)
89 loff_t offset = blocknr * root->blocksize;
90 struct btrfs_buffer *buf;
91 int ret;
93 buf = radix_tree_lookup(&root->fs_info->cache_radix, blocknr);
94 if (buf) {
95 buf->count++;
96 } else {
97 buf = alloc_tree_block(root, blocknr);
98 if (!buf)
99 return NULL;
100 ret = pread(root->fs_info->fp, &buf->node, root->blocksize,
101 offset);
102 if (ret != root->blocksize) {
103 free(buf);
104 return NULL;
107 if (check_tree_block(root, buf))
108 BUG();
109 return buf;
112 int dirty_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
113 struct btrfs_buffer *buf)
115 if (!list_empty(&buf->dirty))
116 return 0;
117 list_add_tail(&buf->dirty, &root->fs_info->trans);
118 buf->count++;
119 return 0;
122 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
123 struct btrfs_buffer *buf)
125 if (!list_empty(&buf->dirty)) {
126 list_del_init(&buf->dirty);
127 btrfs_block_release(root, buf);
129 return 0;
132 int write_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
133 struct btrfs_buffer *buf)
135 u64 blocknr = buf->blocknr;
136 loff_t offset = blocknr * root->blocksize;
137 int ret;
139 if (buf->blocknr != btrfs_header_blocknr(&buf->node.header))
140 BUG();
141 ret = pwrite(root->fs_info->fp, &buf->node, root->blocksize, offset);
142 if (ret != root->blocksize)
143 return ret;
144 return 0;
147 static int __commit_transaction(struct btrfs_trans_handle *trans, struct
148 btrfs_root *root)
150 struct btrfs_buffer *b;
151 int ret = 0;
152 int wret;
153 while(!list_empty(&root->fs_info->trans)) {
154 b = list_entry(root->fs_info->trans.next, struct btrfs_buffer,
155 dirty);
156 list_del_init(&b->dirty);
157 wret = write_tree_block(trans, root, b);
158 if (wret)
159 ret = wret;
160 btrfs_block_release(root, b);
162 return ret;
165 static int commit_tree_roots(struct btrfs_trans_handle *trans,
166 struct btrfs_fs_info *fs_info)
168 int ret;
169 u64 old_extent_block;
170 struct btrfs_root *tree_root = fs_info->tree_root;
171 struct btrfs_root *extent_root = fs_info->extent_root;
172 struct btrfs_root *inode_root = fs_info->inode_root;
174 btrfs_set_root_blocknr(&inode_root->root_item,
175 inode_root->node->blocknr);
176 ret = btrfs_update_root(trans, tree_root,
177 &inode_root->root_key,
178 &inode_root->root_item);
179 BUG_ON(ret);
180 while(1) {
181 old_extent_block = btrfs_root_blocknr(&extent_root->root_item);
182 if (old_extent_block == extent_root->node->blocknr)
183 break;
184 btrfs_set_root_blocknr(&extent_root->root_item,
185 extent_root->node->blocknr);
186 ret = btrfs_update_root(trans, tree_root,
187 &extent_root->root_key,
188 &extent_root->root_item);
189 BUG_ON(ret);
191 return 0;
194 int btrfs_commit_transaction(struct btrfs_trans_handle *trans, struct
195 btrfs_root *root, struct btrfs_super_block *s)
197 int ret = 0;
198 struct btrfs_buffer *snap = root->commit_root;
199 struct btrfs_key snap_key;
201 if (root->commit_root == root->node)
202 return 0;
204 memcpy(&snap_key, &root->root_key, sizeof(snap_key));
205 root->root_key.offset++;
207 btrfs_set_root_blocknr(&root->root_item, root->node->blocknr);
208 ret = btrfs_insert_root(trans, root->fs_info->tree_root,
209 &root->root_key, &root->root_item);
210 BUG_ON(ret);
212 ret = commit_tree_roots(trans, root->fs_info);
213 BUG_ON(ret);
215 ret = __commit_transaction(trans, root);
216 BUG_ON(ret);
218 write_ctree_super(trans, root, s);
219 btrfs_finish_extent_commit(trans, root->fs_info->extent_root);
220 btrfs_finish_extent_commit(trans, root->fs_info->tree_root);
222 root->commit_root = root->node;
223 root->node->count++;
224 ret = btrfs_drop_snapshot(trans, root, snap);
225 BUG_ON(ret);
227 ret = btrfs_del_root(trans, root->fs_info->tree_root, &snap_key);
228 BUG_ON(ret);
229 root->fs_info->generation = root->root_key.offset + 1;
231 return ret;
234 static int __setup_root(struct btrfs_super_block *super,
235 struct btrfs_root *root,
236 struct btrfs_fs_info *fs_info,
237 u64 objectid, int fp)
239 root->node = NULL;
240 root->commit_root = NULL;
241 root->blocksize = btrfs_super_blocksize(super);
242 root->ref_cows = 0;
243 root->fs_info = fs_info;
244 memset(&root->root_key, 0, sizeof(root->root_key));
245 memset(&root->root_item, 0, sizeof(root->root_item));
246 return 0;
249 static int find_and_setup_root(struct btrfs_super_block *super,
250 struct btrfs_root *tree_root,
251 struct btrfs_fs_info *fs_info,
252 u64 objectid,
253 struct btrfs_root *root, int fp)
255 int ret;
257 __setup_root(super, root, fs_info, objectid, fp);
258 ret = btrfs_find_last_root(tree_root, objectid,
259 &root->root_item, &root->root_key);
260 BUG_ON(ret);
262 root->node = read_tree_block(root,
263 btrfs_root_blocknr(&root->root_item));
264 BUG_ON(!root->node);
265 return 0;
268 struct btrfs_root *open_ctree(char *filename, struct btrfs_super_block *super)
270 struct btrfs_root *root = malloc(sizeof(struct btrfs_root));
271 struct btrfs_root *extent_root = malloc(sizeof(struct btrfs_root));
272 struct btrfs_root *tree_root = malloc(sizeof(struct btrfs_root));
273 struct btrfs_root *inode_root = malloc(sizeof(struct btrfs_root));
274 struct btrfs_fs_info *fs_info = malloc(sizeof(*fs_info));
275 int fp;
276 int ret;
278 fp = open(filename, O_CREAT | O_RDWR, 0600);
279 if (fp < 0) {
280 free(root);
281 return NULL;
283 INIT_RADIX_TREE(&fs_info->cache_radix, GFP_KERNEL);
284 INIT_RADIX_TREE(&fs_info->pinned_radix, GFP_KERNEL);
285 INIT_LIST_HEAD(&fs_info->trans);
286 INIT_LIST_HEAD(&fs_info->cache);
287 fs_info->cache_size = 0;
288 fs_info->fp = fp;
289 fs_info->running_transaction = NULL;
290 fs_info->fs_root = root;
291 fs_info->tree_root = tree_root;
292 fs_info->extent_root = extent_root;
293 fs_info->inode_root = inode_root;
294 fs_info->last_inode_alloc = 0;
295 fs_info->last_inode_alloc_dirid = 0;
296 fs_info->disk_super = super;
297 memset(&fs_info->current_insert, 0, sizeof(fs_info->current_insert));
298 memset(&fs_info->last_insert, 0, sizeof(fs_info->last_insert));
300 ret = pread(fp, super, sizeof(struct btrfs_super_block),
301 BTRFS_SUPER_INFO_OFFSET);
302 if (ret == 0 || btrfs_super_root(super) == 0) {
303 BUG();
304 return NULL;
306 BUG_ON(ret < 0);
308 __setup_root(super, tree_root, fs_info, BTRFS_ROOT_TREE_OBJECTID, fp);
309 tree_root->node = read_tree_block(tree_root, btrfs_super_root(super));
310 BUG_ON(!tree_root->node);
312 ret = find_and_setup_root(super, tree_root, fs_info,
313 BTRFS_EXTENT_TREE_OBJECTID, extent_root, fp);
314 BUG_ON(ret);
316 ret = find_and_setup_root(super, tree_root, fs_info,
317 BTRFS_INODE_MAP_OBJECTID, inode_root, fp);
318 BUG_ON(ret);
320 ret = find_and_setup_root(super, tree_root, fs_info,
321 BTRFS_FS_TREE_OBJECTID, root, fp);
322 BUG_ON(ret);
324 root->commit_root = root->node;
325 root->node->count++;
326 root->ref_cows = 1;
327 root->fs_info->generation = root->root_key.offset + 1;
328 return root;
331 int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
332 *root, struct btrfs_super_block *s)
334 int ret;
335 btrfs_set_super_root(s, root->fs_info->tree_root->node->blocknr);
336 ret = pwrite(root->fs_info->fp, s, sizeof(*s),
337 BTRFS_SUPER_INFO_OFFSET);
338 if (ret != sizeof(*s)) {
339 fprintf(stderr, "failed to write new super block err %d\n", ret);
340 return ret;
342 return 0;
345 static int drop_cache(struct btrfs_root *root)
347 while(!list_empty(&root->fs_info->cache)) {
348 struct btrfs_buffer *b = list_entry(root->fs_info->cache.next,
349 struct btrfs_buffer,
350 cache);
351 list_del_init(&b->cache);
352 btrfs_block_release(root, b);
354 return 0;
356 int close_ctree(struct btrfs_root *root, struct btrfs_super_block *s)
358 int ret;
359 struct btrfs_trans_handle *trans;
361 trans = root->fs_info->running_transaction;
362 btrfs_commit_transaction(trans, root, s);
363 ret = commit_tree_roots(trans, root->fs_info);
364 BUG_ON(ret);
365 ret = __commit_transaction(trans, root);
366 BUG_ON(ret);
367 write_ctree_super(trans, root, s);
368 drop_cache(root);
369 BUG_ON(!list_empty(&root->fs_info->trans));
371 close(root->fs_info->fp);
372 if (root->node)
373 btrfs_block_release(root, root->node);
374 if (root->fs_info->extent_root->node)
375 btrfs_block_release(root->fs_info->extent_root,
376 root->fs_info->extent_root->node);
377 if (root->fs_info->inode_root->node)
378 btrfs_block_release(root->fs_info->inode_root,
379 root->fs_info->inode_root->node);
380 if (root->fs_info->tree_root->node)
381 btrfs_block_release(root->fs_info->tree_root,
382 root->fs_info->tree_root->node);
383 btrfs_block_release(root, root->commit_root);
384 free(root);
385 printf("on close %d blocks are allocated\n", allocated_blocks);
386 return 0;
389 void btrfs_block_release(struct btrfs_root *root, struct btrfs_buffer *buf)
391 buf->count--;
392 if (buf->count < 0)
393 BUG();
394 if (buf->count == 0) {
395 BUG_ON(!list_empty(&buf->cache));
396 BUG_ON(!list_empty(&buf->dirty));
397 if (!radix_tree_lookup(&root->fs_info->cache_radix,
398 buf->blocknr))
399 BUG();
400 radix_tree_delete(&root->fs_info->cache_radix, buf->blocknr);
401 memset(buf, 0, sizeof(*buf));
402 free(buf);
403 BUG_ON(allocated_blocks == 0);
404 allocated_blocks--;
405 BUG_ON(root->fs_info->cache_size == 0);
406 root->fs_info->cache_size--;