1
2
3
4
5
6
7
8
9
10
11#include <linux/slab.h>
12#include <linux/pagemap.h>
13#include <linux/log2.h>
14
15#include "hfsplus_fs.h"
16#include "hfsplus_raw.h"
17
18
19
20
21
22#define CLUMP_ENTRIES 15
23
24static short clumptbl[CLUMP_ENTRIES * 3] = {
25
26
27
28
29 4, 4, 4,
30 6, 6, 4,
31 8, 8, 4,
32 11, 11, 5,
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61 64, 32, 5,
62 84, 49, 6,
63 111, 74, 7,
64 147, 111, 8,
65 194, 169, 9,
66 256, 256, 11,
67 294, 294, 14,
68 338, 338, 16,
69 388, 388, 20,
70 446, 446, 25,
71 512, 512, 32
72};
73
74u32 hfsplus_calc_btree_clump_size(u32 block_size, u32 node_size,
75 u64 sectors, int file_id)
76{
77 u32 mod = max(node_size, block_size);
78 u32 clump_size;
79 int column;
80 int i;
81
82
83 switch (file_id) {
84 case HFSPLUS_ATTR_CNID:
85 column = 0;
86 break;
87 case HFSPLUS_CAT_CNID:
88 column = 1;
89 break;
90 default:
91 column = 2;
92 break;
93 }
94
95
96
97
98
99 if (sectors < 0x200000) {
100 clump_size = sectors << 2;
101 if (clump_size < (8 * node_size))
102 clump_size = 8 * node_size;
103 } else {
104
105 for (i = 0, sectors = sectors >> 22;
106 sectors && (i < CLUMP_ENTRIES - 1);
107 ++i, sectors = sectors >> 1) {
108
109 }
110
111 clump_size = clumptbl[column + (i) * 3] * 1024 * 1024;
112 }
113
114
115
116
117
118 clump_size /= mod;
119 clump_size *= mod;
120
121
122
123
124
125 if (clump_size == 0)
126 clump_size = mod;
127
128 return clump_size;
129}
130
131
132struct hfs_btree *hfs_btree_open(struct super_block *sb, u32 id)
133{
134 struct hfs_btree *tree;
135 struct hfs_btree_header_rec *head;
136 struct address_space *mapping;
137 struct inode *inode;
138 struct page *page;
139 unsigned int size;
140
141 tree = kzalloc(sizeof(*tree), GFP_KERNEL);
142 if (!tree)
143 return NULL;
144
145 mutex_init(&tree->tree_lock);
146 spin_lock_init(&tree->hash_lock);
147 tree->sb = sb;
148 tree->cnid = id;
149 inode = hfsplus_iget(sb, id);
150 if (IS_ERR(inode))
151 goto free_tree;
152 tree->inode = inode;
153
154 if (!HFSPLUS_I(tree->inode)->first_blocks) {
155 pr_err("invalid btree extent records (0 size)\n");
156 goto free_inode;
157 }
158
159 mapping = tree->inode->i_mapping;
160 page = read_mapping_page(mapping, 0, NULL);
161 if (IS_ERR(page))
162 goto free_inode;
163
164
165 head = (struct hfs_btree_header_rec *)(kmap(page) +
166 sizeof(struct hfs_bnode_desc));
167 tree->root = be32_to_cpu(head->root);
168 tree->leaf_count = be32_to_cpu(head->leaf_count);
169 tree->leaf_head = be32_to_cpu(head->leaf_head);
170 tree->leaf_tail = be32_to_cpu(head->leaf_tail);
171 tree->node_count = be32_to_cpu(head->node_count);
172 tree->free_nodes = be32_to_cpu(head->free_nodes);
173 tree->attributes = be32_to_cpu(head->attributes);
174 tree->node_size = be16_to_cpu(head->node_size);
175 tree->max_key_len = be16_to_cpu(head->max_key_len);
176 tree->depth = be16_to_cpu(head->depth);
177
178
179 switch (id) {
180 case HFSPLUS_EXT_CNID:
181 if (tree->max_key_len != HFSPLUS_EXT_KEYLEN - sizeof(u16)) {
182 pr_err("invalid extent max_key_len %d\n",
183 tree->max_key_len);
184 goto fail_page;
185 }
186 if (tree->attributes & HFS_TREE_VARIDXKEYS) {
187 pr_err("invalid extent btree flag\n");
188 goto fail_page;
189 }
190
191 tree->keycmp = hfsplus_ext_cmp_key;
192 break;
193 case HFSPLUS_CAT_CNID:
194 if (tree->max_key_len != HFSPLUS_CAT_KEYLEN - sizeof(u16)) {
195 pr_err("invalid catalog max_key_len %d\n",
196 tree->max_key_len);
197 goto fail_page;
198 }
199 if (!(tree->attributes & HFS_TREE_VARIDXKEYS)) {
200 pr_err("invalid catalog btree flag\n");
201 goto fail_page;
202 }
203
204 if (test_bit(HFSPLUS_SB_HFSX, &HFSPLUS_SB(sb)->flags) &&
205 (head->key_type == HFSPLUS_KEY_BINARY))
206 tree->keycmp = hfsplus_cat_bin_cmp_key;
207 else {
208 tree->keycmp = hfsplus_cat_case_cmp_key;
209 set_bit(HFSPLUS_SB_CASEFOLD, &HFSPLUS_SB(sb)->flags);
210 }
211 break;
212 case HFSPLUS_ATTR_CNID:
213 if (tree->max_key_len != HFSPLUS_ATTR_KEYLEN - sizeof(u16)) {
214 pr_err("invalid attributes max_key_len %d\n",
215 tree->max_key_len);
216 goto fail_page;
217 }
218 tree->keycmp = hfsplus_attr_bin_cmp_key;
219 break;
220 default:
221 pr_err("unknown B*Tree requested\n");
222 goto fail_page;
223 }
224
225 if (!(tree->attributes & HFS_TREE_BIGKEYS)) {
226 pr_err("invalid btree flag\n");
227 goto fail_page;
228 }
229
230 size = tree->node_size;
231 if (!is_power_of_2(size))
232 goto fail_page;
233 if (!tree->node_count)
234 goto fail_page;
235
236 tree->node_size_shift = ffs(size) - 1;
237
238 tree->pages_per_bnode =
239 (tree->node_size + PAGE_SIZE - 1) >>
240 PAGE_SHIFT;
241
242 kunmap(page);
243 put_page(page);
244 return tree;
245
246 fail_page:
247 put_page(page);
248 free_inode:
249 tree->inode->i_mapping->a_ops = &hfsplus_aops;
250 iput(tree->inode);
251 free_tree:
252 kfree(tree);
253 return NULL;
254}
255
256
257void hfs_btree_close(struct hfs_btree *tree)
258{
259 struct hfs_bnode *node;
260 int i;
261
262 if (!tree)
263 return;
264
265 for (i = 0; i < NODE_HASH_SIZE; i++) {
266 while ((node = tree->node_hash[i])) {
267 tree->node_hash[i] = node->next_hash;
268 if (atomic_read(&node->refcnt))
269 pr_crit("node %d:%d "
270 "still has %d user(s)!\n",
271 node->tree->cnid, node->this,
272 atomic_read(&node->refcnt));
273 hfs_bnode_free(node);
274 tree->node_hash_cnt--;
275 }
276 }
277 iput(tree->inode);
278 kfree(tree);
279}
280
281int hfs_btree_write(struct hfs_btree *tree)
282{
283 struct hfs_btree_header_rec *head;
284 struct hfs_bnode *node;
285 struct page *page;
286
287 node = hfs_bnode_find(tree, 0);
288 if (IS_ERR(node))
289
290 return -EIO;
291
292 page = node->page[0];
293 head = (struct hfs_btree_header_rec *)(kmap(page) +
294 sizeof(struct hfs_bnode_desc));
295
296 head->root = cpu_to_be32(tree->root);
297 head->leaf_count = cpu_to_be32(tree->leaf_count);
298 head->leaf_head = cpu_to_be32(tree->leaf_head);
299 head->leaf_tail = cpu_to_be32(tree->leaf_tail);
300 head->node_count = cpu_to_be32(tree->node_count);
301 head->free_nodes = cpu_to_be32(tree->free_nodes);
302 head->attributes = cpu_to_be32(tree->attributes);
303 head->depth = cpu_to_be16(tree->depth);
304
305 kunmap(page);
306 set_page_dirty(page);
307 hfs_bnode_put(node);
308 return 0;
309}
310
311static struct hfs_bnode *hfs_bmap_new_bmap(struct hfs_bnode *prev, u32 idx)
312{
313 struct hfs_btree *tree = prev->tree;
314 struct hfs_bnode *node;
315 struct hfs_bnode_desc desc;
316 __be32 cnid;
317
318 node = hfs_bnode_create(tree, idx);
319 if (IS_ERR(node))
320 return node;
321
322 tree->free_nodes--;
323 prev->next = idx;
324 cnid = cpu_to_be32(idx);
325 hfs_bnode_write(prev, &cnid, offsetof(struct hfs_bnode_desc, next), 4);
326
327 node->type = HFS_NODE_MAP;
328 node->num_recs = 1;
329 hfs_bnode_clear(node, 0, tree->node_size);
330 desc.next = 0;
331 desc.prev = 0;
332 desc.type = HFS_NODE_MAP;
333 desc.height = 0;
334 desc.num_recs = cpu_to_be16(1);
335 desc.reserved = 0;
336 hfs_bnode_write(node, &desc, 0, sizeof(desc));
337 hfs_bnode_write_u16(node, 14, 0x8000);
338 hfs_bnode_write_u16(node, tree->node_size - 2, 14);
339 hfs_bnode_write_u16(node, tree->node_size - 4, tree->node_size - 6);
340
341 return node;
342}
343
344struct hfs_bnode *hfs_bmap_alloc(struct hfs_btree *tree)
345{
346 struct hfs_bnode *node, *next_node;
347 struct page **pagep;
348 u32 nidx, idx;
349 unsigned off;
350 u16 off16;
351 u16 len;
352 u8 *data, byte, m;
353 int i;
354
355 while (!tree->free_nodes) {
356 struct inode *inode = tree->inode;
357 struct hfsplus_inode_info *hip = HFSPLUS_I(inode);
358 u32 count;
359 int res;
360
361 res = hfsplus_file_extend(inode, hfs_bnode_need_zeroout(tree));
362 if (res)
363 return ERR_PTR(res);
364 hip->phys_size = inode->i_size =
365 (loff_t)hip->alloc_blocks <<
366 HFSPLUS_SB(tree->sb)->alloc_blksz_shift;
367 hip->fs_blocks =
368 hip->alloc_blocks << HFSPLUS_SB(tree->sb)->fs_shift;
369 inode_set_bytes(inode, inode->i_size);
370 count = inode->i_size >> tree->node_size_shift;
371 tree->free_nodes = count - tree->node_count;
372 tree->node_count = count;
373 }
374
375 nidx = 0;
376 node = hfs_bnode_find(tree, nidx);
377 if (IS_ERR(node))
378 return node;
379 len = hfs_brec_lenoff(node, 2, &off16);
380 off = off16;
381
382 off += node->page_offset;
383 pagep = node->page + (off >> PAGE_SHIFT);
384 data = kmap(*pagep);
385 off &= ~PAGE_MASK;
386 idx = 0;
387
388 for (;;) {
389 while (len) {
390 byte = data[off];
391 if (byte != 0xff) {
392 for (m = 0x80, i = 0; i < 8; m >>= 1, i++) {
393 if (!(byte & m)) {
394 idx += i;
395 data[off] |= m;
396 set_page_dirty(*pagep);
397 kunmap(*pagep);
398 tree->free_nodes--;
399 mark_inode_dirty(tree->inode);
400 hfs_bnode_put(node);
401 return hfs_bnode_create(tree,
402 idx);
403 }
404 }
405 }
406 if (++off >= PAGE_SIZE) {
407 kunmap(*pagep);
408 data = kmap(*++pagep);
409 off = 0;
410 }
411 idx += 8;
412 len--;
413 }
414 kunmap(*pagep);
415 nidx = node->next;
416 if (!nidx) {
417 hfs_dbg(BNODE_MOD, "create new bmap node\n");
418 next_node = hfs_bmap_new_bmap(node, idx);
419 } else
420 next_node = hfs_bnode_find(tree, nidx);
421 hfs_bnode_put(node);
422 if (IS_ERR(next_node))
423 return next_node;
424 node = next_node;
425
426 len = hfs_brec_lenoff(node, 0, &off16);
427 off = off16;
428 off += node->page_offset;
429 pagep = node->page + (off >> PAGE_SHIFT);
430 data = kmap(*pagep);
431 off &= ~PAGE_MASK;
432 }
433}
434
435void hfs_bmap_free(struct hfs_bnode *node)
436{
437 struct hfs_btree *tree;
438 struct page *page;
439 u16 off, len;
440 u32 nidx;
441 u8 *data, byte, m;
442
443 hfs_dbg(BNODE_MOD, "btree_free_node: %u\n", node->this);
444 BUG_ON(!node->this);
445 tree = node->tree;
446 nidx = node->this;
447 node = hfs_bnode_find(tree, 0);
448 if (IS_ERR(node))
449 return;
450 len = hfs_brec_lenoff(node, 2, &off);
451 while (nidx >= len * 8) {
452 u32 i;
453
454 nidx -= len * 8;
455 i = node->next;
456 hfs_bnode_put(node);
457 if (!i) {
458 ;
459 pr_crit("unable to free bnode %u. "
460 "bmap not found!\n",
461 node->this);
462 return;
463 }
464 node = hfs_bnode_find(tree, i);
465 if (IS_ERR(node))
466 return;
467 if (node->type != HFS_NODE_MAP) {
468 ;
469 pr_crit("invalid bmap found! "
470 "(%u,%d)\n",
471 node->this, node->type);
472 hfs_bnode_put(node);
473 return;
474 }
475 len = hfs_brec_lenoff(node, 0, &off);
476 }
477 off += node->page_offset + nidx / 8;
478 page = node->page[off >> PAGE_SHIFT];
479 data = kmap(page);
480 off &= ~PAGE_MASK;
481 m = 1 << (~nidx & 7);
482 byte = data[off];
483 if (!(byte & m)) {
484 pr_crit("trying to free free bnode "
485 "%u(%d)\n",
486 node->this, node->type);
487 kunmap(page);
488 hfs_bnode_put(node);
489 return;
490 }
491 data[off] = byte & ~m;
492 set_page_dirty(page);
493 kunmap(page);
494 hfs_bnode_put(node);
495 tree->free_nodes++;
496 mark_inode_dirty(tree->inode);
497}
498