xref: /linux/fs/hfsplus/btree.c (revision a2cce7a9f1b8cc3d4edce106fb971529f1d4d9ce)
1 /*
2  *  linux/fs/hfsplus/btree.c
3  *
4  * Copyright (C) 2001
5  * Brad Boyer (flar@allandria.com)
6  * (C) 2003 Ardis Technologies <roman@ardistech.com>
7  *
8  * Handle opening/closing btree
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  * Initial source code of clump size calculation is gotten
20  * from http://opensource.apple.com/tarballs/diskdev_cmds/
21  */
22 #define CLUMP_ENTRIES	15
23 
24 static short clumptbl[CLUMP_ENTRIES * 3] = {
25 /*
26  *	    Volume	Attributes	 Catalog	 Extents
27  *	     Size	Clump (MB)	Clump (MB)	Clump (MB)
28  */
29 	/*   1GB */	  4,		  4,		 4,
30 	/*   2GB */	  6,		  6,		 4,
31 	/*   4GB */	  8,		  8,		 4,
32 	/*   8GB */	 11,		 11,		 5,
33 	/*
34 	 * For volumes 16GB and larger, we want to make sure that a full OS
35 	 * install won't require fragmentation of the Catalog or Attributes
36 	 * B-trees.  We do this by making the clump sizes sufficiently large,
37 	 * and by leaving a gap after the B-trees for them to grow into.
38 	 *
39 	 * For SnowLeopard 10A298, a FullNetInstall with all packages selected
40 	 * results in:
41 	 * Catalog B-tree Header
42 	 *	nodeSize:          8192
43 	 *	totalNodes:       31616
44 	 *	freeNodes:         1978
45 	 * (used = 231.55 MB)
46 	 * Attributes B-tree Header
47 	 *	nodeSize:          8192
48 	 *	totalNodes:       63232
49 	 *	freeNodes:          958
50 	 * (used = 486.52 MB)
51 	 *
52 	 * We also want Time Machine backup volumes to have a sufficiently
53 	 * large clump size to reduce fragmentation.
54 	 *
55 	 * The series of numbers for Catalog and Attribute form a geometric
56 	 * series. For Catalog (16GB to 512GB), each term is 8**(1/5) times
57 	 * the previous term.  For Attributes (16GB to 512GB), each term is
58 	 * 4**(1/5) times the previous term.  For 1TB to 16TB, each term is
59 	 * 2**(1/5) times the previous term.
60 	 */
61 	/*  16GB */	 64,		 32,		 5,
62 	/*  32GB */	 84,		 49,		 6,
63 	/*  64GB */	111,		 74,		 7,
64 	/* 128GB */	147,		111,		 8,
65 	/* 256GB */	194,		169,		 9,
66 	/* 512GB */	256,		256,		11,
67 	/*   1TB */	294,		294,		14,
68 	/*   2TB */	338,		338,		16,
69 	/*   4TB */	388,		388,		20,
70 	/*   8TB */	446,		446,		25,
71 	/*  16TB */	512,		512,		32
72 };
73 
74 u32 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 	/* Figure out which column of the above table to use for this file. */
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 	 * The default clump size is 0.8% of the volume size. And
97 	 * it must also be a multiple of the node and block size.
98 	 */
99 	if (sectors < 0x200000) {
100 		clump_size = sectors << 2;	/*  0.8 %  */
101 		if (clump_size < (8 * node_size))
102 			clump_size = 8 * node_size;
103 	} else {
104 		/* turn exponent into table index... */
105 		for (i = 0, sectors = sectors >> 22;
106 		     sectors && (i < CLUMP_ENTRIES - 1);
107 		     ++i, sectors = sectors >> 1) {
108 			/* empty body */
109 		}
110 
111 		clump_size = clumptbl[column + (i) * 3] * 1024 * 1024;
112 	}
113 
114 	/*
115 	 * Round the clump size to a multiple of node and block size.
116 	 * NOTE: This rounds down.
117 	 */
118 	clump_size /= mod;
119 	clump_size *= mod;
120 
121 	/*
122 	 * Rounding down could have rounded down to 0 if the block size was
123 	 * greater than the clump size.  If so, just use one block or node.
124 	 */
125 	if (clump_size == 0)
126 		clump_size = mod;
127 
128 	return clump_size;
129 }
130 
131 /* Get a reference to a B*Tree and do some initial checks */
132 struct 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 	/* Load the header */
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 	/* Verify the tree and set the correct compare function */
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_CACHE_SIZE - 1) >>
240 		PAGE_CACHE_SHIFT;
241 
242 	kunmap(page);
243 	page_cache_release(page);
244 	return tree;
245 
246  fail_page:
247 	page_cache_release(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 /* Release resources used by a btree */
257 void 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 
281 int 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 		/* panic? */
290 		return -EIO;
291 	/* Load the header */
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 
311 static 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 
344 struct 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_CACHE_SHIFT);
384 	data = kmap(*pagep);
385 	off &= ~PAGE_CACHE_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_CACHE_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_CACHE_SHIFT);
430 		data = kmap(*pagep);
431 		off &= ~PAGE_CACHE_MASK;
432 	}
433 }
434 
435 void 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 			/* panic */;
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 			/* panic */;
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_CACHE_SHIFT];
479 	data = kmap(page);
480 	off &= ~PAGE_CACHE_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