1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * linux/fs/hfsplus/btree.c
4 *
5 * Copyright (C) 2001
6 * Brad Boyer (flar@allandria.com)
7 * (C) 2003 Ardis Technologies <roman@ardistech.com>
8 *
9 * Handle opening/closing btree
10 */
11
12 #include <linux/slab.h>
13 #include <linux/pagemap.h>
14 #include <linux/log2.h>
15
16 #include "hfsplus_fs.h"
17 #include "hfsplus_raw.h"
18
19 /*
20 * Initial source code of clump size calculation is gotten
21 * from http://opensource.apple.com/tarballs/diskdev_cmds/
22 */
23 #define CLUMP_ENTRIES 15
24
25 static short clumptbl[CLUMP_ENTRIES * 3] = {
26 /*
27 * Volume Attributes Catalog Extents
28 * Size Clump (MB) Clump (MB) Clump (MB)
29 */
30 /* 1GB */ 4, 4, 4,
31 /* 2GB */ 6, 6, 4,
32 /* 4GB */ 8, 8, 4,
33 /* 8GB */ 11, 11, 5,
34 /*
35 * For volumes 16GB and larger, we want to make sure that a full OS
36 * install won't require fragmentation of the Catalog or Attributes
37 * B-trees. We do this by making the clump sizes sufficiently large,
38 * and by leaving a gap after the B-trees for them to grow into.
39 *
40 * For SnowLeopard 10A298, a FullNetInstall with all packages selected
41 * results in:
42 * Catalog B-tree Header
43 * nodeSize: 8192
44 * totalNodes: 31616
45 * freeNodes: 1978
46 * (used = 231.55 MB)
47 * Attributes B-tree Header
48 * nodeSize: 8192
49 * totalNodes: 63232
50 * freeNodes: 958
51 * (used = 486.52 MB)
52 *
53 * We also want Time Machine backup volumes to have a sufficiently
54 * large clump size to reduce fragmentation.
55 *
56 * The series of numbers for Catalog and Attribute form a geometric
57 * series. For Catalog (16GB to 512GB), each term is 8**(1/5) times
58 * the previous term. For Attributes (16GB to 512GB), each term is
59 * 4**(1/5) times the previous term. For 1TB to 16TB, each term is
60 * 2**(1/5) times the previous term.
61 */
62 /* 16GB */ 64, 32, 5,
63 /* 32GB */ 84, 49, 6,
64 /* 64GB */ 111, 74, 7,
65 /* 128GB */ 147, 111, 8,
66 /* 256GB */ 194, 169, 9,
67 /* 512GB */ 256, 256, 11,
68 /* 1TB */ 294, 294, 14,
69 /* 2TB */ 338, 338, 16,
70 /* 4TB */ 388, 388, 20,
71 /* 8TB */ 446, 446, 25,
72 /* 16TB */ 512, 512, 32
73 };
74
hfsplus_calc_btree_clump_size(u32 block_size,u32 node_size,u64 sectors,int file_id)75 u32 hfsplus_calc_btree_clump_size(u32 block_size, u32 node_size,
76 u64 sectors, int file_id)
77 {
78 u32 mod = max(node_size, block_size);
79 u32 clump_size;
80 int column;
81 int i;
82
83 /* Figure out which column of the above table to use for this file. */
84 switch (file_id) {
85 case HFSPLUS_ATTR_CNID:
86 column = 0;
87 break;
88 case HFSPLUS_CAT_CNID:
89 column = 1;
90 break;
91 default:
92 column = 2;
93 break;
94 }
95
96 /*
97 * The default clump size is 0.8% of the volume size. And
98 * it must also be a multiple of the node and block size.
99 */
100 if (sectors < 0x200000) {
101 clump_size = sectors << 2; /* 0.8 % */
102 if (clump_size < (8 * node_size))
103 clump_size = 8 * node_size;
104 } else {
105 /* turn exponent into table index... */
106 for (i = 0, sectors = sectors >> 22;
107 sectors && (i < CLUMP_ENTRIES - 1);
108 ++i, sectors = sectors >> 1) {
109 /* empty body */
110 }
111
112 clump_size = clumptbl[column + (i) * 3] * 1024 * 1024;
113 }
114
115 /*
116 * Round the clump size to a multiple of node and block size.
117 * NOTE: This rounds down.
118 */
119 clump_size /= mod;
120 clump_size *= mod;
121
122 /*
123 * Rounding down could have rounded down to 0 if the block size was
124 * greater than the clump size. If so, just use one block or node.
125 */
126 if (clump_size == 0)
127 clump_size = mod;
128
129 return clump_size;
130 }
131
132 /* Get a reference to a B*Tree and do some initial checks */
hfs_btree_open(struct super_block * sb,u32 id)133 struct hfs_btree *hfs_btree_open(struct super_block *sb, u32 id)
134 {
135 struct hfs_btree *tree;
136 struct hfs_btree_header_rec *head;
137 struct address_space *mapping;
138 struct inode *inode;
139 struct page *page;
140 unsigned int size;
141
142 tree = kzalloc(sizeof(*tree), GFP_KERNEL);
143 if (!tree)
144 return NULL;
145
146 mutex_init(&tree->tree_lock);
147 spin_lock_init(&tree->hash_lock);
148 tree->sb = sb;
149 tree->cnid = id;
150 inode = hfsplus_iget(sb, id);
151 if (IS_ERR(inode))
152 goto free_tree;
153 tree->inode = inode;
154
155 if (!HFSPLUS_I(tree->inode)->first_blocks) {
156 pr_err("invalid btree extent records (0 size)\n");
157 goto free_inode;
158 }
159
160 mapping = tree->inode->i_mapping;
161 page = read_mapping_page(mapping, 0, NULL);
162 if (IS_ERR(page))
163 goto free_inode;
164
165 /* Load the header */
166 head = (struct hfs_btree_header_rec *)(kmap(page) +
167 sizeof(struct hfs_bnode_desc));
168 tree->root = be32_to_cpu(head->root);
169 tree->leaf_count = be32_to_cpu(head->leaf_count);
170 tree->leaf_head = be32_to_cpu(head->leaf_head);
171 tree->leaf_tail = be32_to_cpu(head->leaf_tail);
172 tree->node_count = be32_to_cpu(head->node_count);
173 tree->free_nodes = be32_to_cpu(head->free_nodes);
174 tree->attributes = be32_to_cpu(head->attributes);
175 tree->node_size = be16_to_cpu(head->node_size);
176 tree->max_key_len = be16_to_cpu(head->max_key_len);
177 tree->depth = be16_to_cpu(head->depth);
178
179 /* Verify the tree and set the correct compare function */
180 switch (id) {
181 case HFSPLUS_EXT_CNID:
182 if (tree->max_key_len != HFSPLUS_EXT_KEYLEN - sizeof(u16)) {
183 pr_err("invalid extent max_key_len %d\n",
184 tree->max_key_len);
185 goto fail_page;
186 }
187 if (tree->attributes & HFS_TREE_VARIDXKEYS) {
188 pr_err("invalid extent btree flag\n");
189 goto fail_page;
190 }
191
192 tree->keycmp = hfsplus_ext_cmp_key;
193 break;
194 case HFSPLUS_CAT_CNID:
195 if (tree->max_key_len != HFSPLUS_CAT_KEYLEN - sizeof(u16)) {
196 pr_err("invalid catalog max_key_len %d\n",
197 tree->max_key_len);
198 goto fail_page;
199 }
200 if (!(tree->attributes & HFS_TREE_VARIDXKEYS)) {
201 pr_err("invalid catalog btree flag\n");
202 goto fail_page;
203 }
204
205 if (test_bit(HFSPLUS_SB_HFSX, &HFSPLUS_SB(sb)->flags) &&
206 (head->key_type == HFSPLUS_KEY_BINARY))
207 tree->keycmp = hfsplus_cat_bin_cmp_key;
208 else {
209 tree->keycmp = hfsplus_cat_case_cmp_key;
210 set_bit(HFSPLUS_SB_CASEFOLD, &HFSPLUS_SB(sb)->flags);
211 }
212 break;
213 case HFSPLUS_ATTR_CNID:
214 if (tree->max_key_len != HFSPLUS_ATTR_KEYLEN - sizeof(u16)) {
215 pr_err("invalid attributes max_key_len %d\n",
216 tree->max_key_len);
217 goto fail_page;
218 }
219 tree->keycmp = hfsplus_attr_bin_cmp_key;
220 break;
221 default:
222 pr_err("unknown B*Tree requested\n");
223 goto fail_page;
224 }
225
226 if (!(tree->attributes & HFS_TREE_BIGKEYS)) {
227 pr_err("invalid btree flag\n");
228 goto fail_page;
229 }
230
231 size = tree->node_size;
232 if (!is_power_of_2(size))
233 goto fail_page;
234 if (!tree->node_count)
235 goto fail_page;
236
237 tree->node_size_shift = ffs(size) - 1;
238
239 tree->pages_per_bnode =
240 (tree->node_size + PAGE_SIZE - 1) >>
241 PAGE_SHIFT;
242
243 kunmap(page);
244 put_page(page);
245 return tree;
246
247 fail_page:
248 put_page(page);
249 free_inode:
250 tree->inode->i_mapping->a_ops = &hfsplus_aops;
251 iput(tree->inode);
252 free_tree:
253 kfree(tree);
254 return NULL;
255 }
256
257 /* Release resources used by a btree */
hfs_btree_close(struct hfs_btree * tree)258 void hfs_btree_close(struct hfs_btree *tree)
259 {
260 struct hfs_bnode *node;
261 int i;
262
263 if (!tree)
264 return;
265
266 for (i = 0; i < NODE_HASH_SIZE; i++) {
267 while ((node = tree->node_hash[i])) {
268 tree->node_hash[i] = node->next_hash;
269 if (atomic_read(&node->refcnt))
270 pr_crit("node %d:%d "
271 "still has %d user(s)!\n",
272 node->tree->cnid, node->this,
273 atomic_read(&node->refcnt));
274 hfs_bnode_free(node);
275 tree->node_hash_cnt--;
276 }
277 }
278 iput(tree->inode);
279 kfree(tree);
280 }
281
hfs_btree_write(struct hfs_btree * tree)282 int hfs_btree_write(struct hfs_btree *tree)
283 {
284 struct hfs_btree_header_rec *head;
285 struct hfs_bnode *node;
286 struct page *page;
287
288 node = hfs_bnode_find(tree, 0);
289 if (IS_ERR(node))
290 /* panic? */
291 return -EIO;
292 /* Load the header */
293 page = node->page[0];
294 head = (struct hfs_btree_header_rec *)(kmap(page) +
295 sizeof(struct hfs_bnode_desc));
296
297 head->root = cpu_to_be32(tree->root);
298 head->leaf_count = cpu_to_be32(tree->leaf_count);
299 head->leaf_head = cpu_to_be32(tree->leaf_head);
300 head->leaf_tail = cpu_to_be32(tree->leaf_tail);
301 head->node_count = cpu_to_be32(tree->node_count);
302 head->free_nodes = cpu_to_be32(tree->free_nodes);
303 head->attributes = cpu_to_be32(tree->attributes);
304 head->depth = cpu_to_be16(tree->depth);
305
306 kunmap(page);
307 set_page_dirty(page);
308 hfs_bnode_put(node);
309 return 0;
310 }
311
hfs_bmap_new_bmap(struct hfs_bnode * prev,u32 idx)312 static struct hfs_bnode *hfs_bmap_new_bmap(struct hfs_bnode *prev, u32 idx)
313 {
314 struct hfs_btree *tree = prev->tree;
315 struct hfs_bnode *node;
316 struct hfs_bnode_desc desc;
317 __be32 cnid;
318
319 node = hfs_bnode_create(tree, idx);
320 if (IS_ERR(node))
321 return node;
322
323 tree->free_nodes--;
324 prev->next = idx;
325 cnid = cpu_to_be32(idx);
326 hfs_bnode_write(prev, &cnid, offsetof(struct hfs_bnode_desc, next), 4);
327
328 node->type = HFS_NODE_MAP;
329 node->num_recs = 1;
330 hfs_bnode_clear(node, 0, tree->node_size);
331 desc.next = 0;
332 desc.prev = 0;
333 desc.type = HFS_NODE_MAP;
334 desc.height = 0;
335 desc.num_recs = cpu_to_be16(1);
336 desc.reserved = 0;
337 hfs_bnode_write(node, &desc, 0, sizeof(desc));
338 hfs_bnode_write_u16(node, 14, 0x8000);
339 hfs_bnode_write_u16(node, tree->node_size - 2, 14);
340 hfs_bnode_write_u16(node, tree->node_size - 4, tree->node_size - 6);
341
342 return node;
343 }
344
345 /* Make sure @tree has enough space for the @rsvd_nodes */
hfs_bmap_reserve(struct hfs_btree * tree,int rsvd_nodes)346 int hfs_bmap_reserve(struct hfs_btree *tree, int rsvd_nodes)
347 {
348 struct inode *inode = tree->inode;
349 struct hfsplus_inode_info *hip = HFSPLUS_I(inode);
350 u32 count;
351 int res;
352
353 if (rsvd_nodes <= 0)
354 return 0;
355
356 while (tree->free_nodes < rsvd_nodes) {
357 res = hfsplus_file_extend(inode, hfs_bnode_need_zeroout(tree));
358 if (res)
359 return res;
360 hip->phys_size = inode->i_size =
361 (loff_t)hip->alloc_blocks <<
362 HFSPLUS_SB(tree->sb)->alloc_blksz_shift;
363 hip->fs_blocks =
364 hip->alloc_blocks << HFSPLUS_SB(tree->sb)->fs_shift;
365 inode_set_bytes(inode, inode->i_size);
366 count = inode->i_size >> tree->node_size_shift;
367 tree->free_nodes += count - tree->node_count;
368 tree->node_count = count;
369 }
370 return 0;
371 }
372
hfs_bmap_alloc(struct hfs_btree * tree)373 struct hfs_bnode *hfs_bmap_alloc(struct hfs_btree *tree)
374 {
375 struct hfs_bnode *node, *next_node;
376 struct page **pagep;
377 u32 nidx, idx;
378 unsigned off;
379 u16 off16;
380 u16 len;
381 u8 *data, byte, m;
382 int i, res;
383
384 res = hfs_bmap_reserve(tree, 1);
385 if (res)
386 return ERR_PTR(res);
387
388 nidx = 0;
389 node = hfs_bnode_find(tree, nidx);
390 if (IS_ERR(node))
391 return node;
392 len = hfs_brec_lenoff(node, 2, &off16);
393 off = off16;
394
395 off += node->page_offset;
396 pagep = node->page + (off >> PAGE_SHIFT);
397 data = kmap(*pagep);
398 off &= ~PAGE_MASK;
399 idx = 0;
400
401 for (;;) {
402 while (len) {
403 byte = data[off];
404 if (byte != 0xff) {
405 for (m = 0x80, i = 0; i < 8; m >>= 1, i++) {
406 if (!(byte & m)) {
407 idx += i;
408 data[off] |= m;
409 set_page_dirty(*pagep);
410 kunmap(*pagep);
411 tree->free_nodes--;
412 mark_inode_dirty(tree->inode);
413 hfs_bnode_put(node);
414 return hfs_bnode_create(tree,
415 idx);
416 }
417 }
418 }
419 if (++off >= PAGE_SIZE) {
420 kunmap(*pagep);
421 data = kmap(*++pagep);
422 off = 0;
423 }
424 idx += 8;
425 len--;
426 }
427 kunmap(*pagep);
428 nidx = node->next;
429 if (!nidx) {
430 hfs_dbg(BNODE_MOD, "create new bmap node\n");
431 next_node = hfs_bmap_new_bmap(node, idx);
432 } else
433 next_node = hfs_bnode_find(tree, nidx);
434 hfs_bnode_put(node);
435 if (IS_ERR(next_node))
436 return next_node;
437 node = next_node;
438
439 len = hfs_brec_lenoff(node, 0, &off16);
440 off = off16;
441 off += node->page_offset;
442 pagep = node->page + (off >> PAGE_SHIFT);
443 data = kmap(*pagep);
444 off &= ~PAGE_MASK;
445 }
446 }
447
hfs_bmap_free(struct hfs_bnode * node)448 void hfs_bmap_free(struct hfs_bnode *node)
449 {
450 struct hfs_btree *tree;
451 struct page *page;
452 u16 off, len;
453 u32 nidx;
454 u8 *data, byte, m;
455
456 hfs_dbg(BNODE_MOD, "btree_free_node: %u\n", node->this);
457 BUG_ON(!node->this);
458 tree = node->tree;
459 nidx = node->this;
460 node = hfs_bnode_find(tree, 0);
461 if (IS_ERR(node))
462 return;
463 len = hfs_brec_lenoff(node, 2, &off);
464 while (nidx >= len * 8) {
465 u32 i;
466
467 nidx -= len * 8;
468 i = node->next;
469 if (!i) {
470 /* panic */;
471 pr_crit("unable to free bnode %u. "
472 "bmap not found!\n",
473 node->this);
474 hfs_bnode_put(node);
475 return;
476 }
477 hfs_bnode_put(node);
478 node = hfs_bnode_find(tree, i);
479 if (IS_ERR(node))
480 return;
481 if (node->type != HFS_NODE_MAP) {
482 /* panic */;
483 pr_crit("invalid bmap found! "
484 "(%u,%d)\n",
485 node->this, node->type);
486 hfs_bnode_put(node);
487 return;
488 }
489 len = hfs_brec_lenoff(node, 0, &off);
490 }
491 off += node->page_offset + nidx / 8;
492 page = node->page[off >> PAGE_SHIFT];
493 data = kmap(page);
494 off &= ~PAGE_MASK;
495 m = 1 << (~nidx & 7);
496 byte = data[off];
497 if (!(byte & m)) {
498 pr_crit("trying to free free bnode "
499 "%u(%d)\n",
500 node->this, node->type);
501 kunmap(page);
502 hfs_bnode_put(node);
503 return;
504 }
505 data[off] = byte & ~m;
506 set_page_dirty(page);
507 kunmap(page);
508 hfs_bnode_put(node);
509 tree->free_nodes++;
510 mark_inode_dirty(tree->inode);
511 }
512