xref: /linux/fs/hfsplus/btree.c (revision 4e0ae876f77bc01a7e77724dea57b4b82bd53244)
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 
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 */
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 */
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 
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 
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 */
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 
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 
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