xref: /openbmc/linux/fs/hfsplus/bnode.c (revision 2d99a7ec)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *  linux/fs/hfsplus/bnode.c
4  *
5  * Copyright (C) 2001
6  * Brad Boyer (flar@allandria.com)
7  * (C) 2003 Ardis Technologies <roman@ardistech.com>
8  *
9  * Handle basic btree node operations
10  */
11 
12 #include <linux/string.h>
13 #include <linux/slab.h>
14 #include <linux/pagemap.h>
15 #include <linux/fs.h>
16 #include <linux/swap.h>
17 
18 #include "hfsplus_fs.h"
19 #include "hfsplus_raw.h"
20 
21 /* Copy a specified range of bytes from the raw data of a node */
22 void hfs_bnode_read(struct hfs_bnode *node, void *buf, int off, int len)
23 {
24 	struct page **pagep;
25 	int l;
26 
27 	off += node->page_offset;
28 	pagep = node->page + (off >> PAGE_SHIFT);
29 	off &= ~PAGE_MASK;
30 
31 	l = min_t(int, len, PAGE_SIZE - off);
32 	memcpy_from_page(buf, *pagep, off, l);
33 
34 	while ((len -= l) != 0) {
35 		buf += l;
36 		l = min_t(int, len, PAGE_SIZE);
37 		memcpy_from_page(buf, *++pagep, 0, l);
38 	}
39 }
40 
41 u16 hfs_bnode_read_u16(struct hfs_bnode *node, int off)
42 {
43 	__be16 data;
44 	/* TODO: optimize later... */
45 	hfs_bnode_read(node, &data, off, 2);
46 	return be16_to_cpu(data);
47 }
48 
49 u8 hfs_bnode_read_u8(struct hfs_bnode *node, int off)
50 {
51 	u8 data;
52 	/* TODO: optimize later... */
53 	hfs_bnode_read(node, &data, off, 1);
54 	return data;
55 }
56 
57 void hfs_bnode_read_key(struct hfs_bnode *node, void *key, int off)
58 {
59 	struct hfs_btree *tree;
60 	int key_len;
61 
62 	tree = node->tree;
63 	if (node->type == HFS_NODE_LEAF ||
64 	    tree->attributes & HFS_TREE_VARIDXKEYS ||
65 	    node->tree->cnid == HFSPLUS_ATTR_CNID)
66 		key_len = hfs_bnode_read_u16(node, off) + 2;
67 	else
68 		key_len = tree->max_key_len + 2;
69 
70 	hfs_bnode_read(node, key, off, key_len);
71 }
72 
73 void hfs_bnode_write(struct hfs_bnode *node, void *buf, int off, int len)
74 {
75 	struct page **pagep;
76 	int l;
77 
78 	off += node->page_offset;
79 	pagep = node->page + (off >> PAGE_SHIFT);
80 	off &= ~PAGE_MASK;
81 
82 	l = min_t(int, len, PAGE_SIZE - off);
83 	memcpy_to_page(*pagep, off, buf, l);
84 	set_page_dirty(*pagep);
85 
86 	while ((len -= l) != 0) {
87 		buf += l;
88 		l = min_t(int, len, PAGE_SIZE);
89 		memcpy_to_page(*++pagep, 0, buf, l);
90 		set_page_dirty(*pagep);
91 	}
92 }
93 
94 void hfs_bnode_write_u16(struct hfs_bnode *node, int off, u16 data)
95 {
96 	__be16 v = cpu_to_be16(data);
97 	/* TODO: optimize later... */
98 	hfs_bnode_write(node, &v, off, 2);
99 }
100 
101 void hfs_bnode_clear(struct hfs_bnode *node, int off, int len)
102 {
103 	struct page **pagep;
104 	int l;
105 
106 	off += node->page_offset;
107 	pagep = node->page + (off >> PAGE_SHIFT);
108 	off &= ~PAGE_MASK;
109 
110 	l = min_t(int, len, PAGE_SIZE - off);
111 	memzero_page(*pagep, off, l);
112 	set_page_dirty(*pagep);
113 
114 	while ((len -= l) != 0) {
115 		l = min_t(int, len, PAGE_SIZE);
116 		memzero_page(*++pagep, 0, l);
117 		set_page_dirty(*pagep);
118 	}
119 }
120 
121 void hfs_bnode_copy(struct hfs_bnode *dst_node, int dst,
122 		    struct hfs_bnode *src_node, int src, int len)
123 {
124 	struct page **src_page, **dst_page;
125 	int l;
126 
127 	hfs_dbg(BNODE_MOD, "copybytes: %u,%u,%u\n", dst, src, len);
128 	if (!len)
129 		return;
130 	src += src_node->page_offset;
131 	dst += dst_node->page_offset;
132 	src_page = src_node->page + (src >> PAGE_SHIFT);
133 	src &= ~PAGE_MASK;
134 	dst_page = dst_node->page + (dst >> PAGE_SHIFT);
135 	dst &= ~PAGE_MASK;
136 
137 	if (src == dst) {
138 		l = min_t(int, len, PAGE_SIZE - src);
139 		memcpy_page(*dst_page, src, *src_page, src, l);
140 		set_page_dirty(*dst_page);
141 
142 		while ((len -= l) != 0) {
143 			l = min_t(int, len, PAGE_SIZE);
144 			memcpy_page(*++dst_page, 0, *++src_page, 0, l);
145 			set_page_dirty(*dst_page);
146 		}
147 	} else {
148 		void *src_ptr, *dst_ptr;
149 
150 		do {
151 			dst_ptr = kmap_local_page(*dst_page) + dst;
152 			src_ptr = kmap_local_page(*src_page) + src;
153 			if (PAGE_SIZE - src < PAGE_SIZE - dst) {
154 				l = PAGE_SIZE - src;
155 				src = 0;
156 				dst += l;
157 			} else {
158 				l = PAGE_SIZE - dst;
159 				src += l;
160 				dst = 0;
161 			}
162 			l = min(len, l);
163 			memcpy(dst_ptr, src_ptr, l);
164 			kunmap_local(src_ptr);
165 			set_page_dirty(*dst_page);
166 			kunmap_local(dst_ptr);
167 			if (!dst)
168 				dst_page++;
169 			else
170 				src_page++;
171 		} while ((len -= l));
172 	}
173 }
174 
175 void hfs_bnode_move(struct hfs_bnode *node, int dst, int src, int len)
176 {
177 	struct page **src_page, **dst_page;
178 	void *src_ptr, *dst_ptr;
179 	int l;
180 
181 	hfs_dbg(BNODE_MOD, "movebytes: %u,%u,%u\n", dst, src, len);
182 	if (!len)
183 		return;
184 	src += node->page_offset;
185 	dst += node->page_offset;
186 	if (dst > src) {
187 		src += len - 1;
188 		src_page = node->page + (src >> PAGE_SHIFT);
189 		src = (src & ~PAGE_MASK) + 1;
190 		dst += len - 1;
191 		dst_page = node->page + (dst >> PAGE_SHIFT);
192 		dst = (dst & ~PAGE_MASK) + 1;
193 
194 		if (src == dst) {
195 			while (src < len) {
196 				dst_ptr = kmap_local_page(*dst_page);
197 				src_ptr = kmap_local_page(*src_page);
198 				memmove(dst_ptr, src_ptr, src);
199 				kunmap_local(src_ptr);
200 				set_page_dirty(*dst_page);
201 				kunmap_local(dst_ptr);
202 				len -= src;
203 				src = PAGE_SIZE;
204 				src_page--;
205 				dst_page--;
206 			}
207 			src -= len;
208 			dst_ptr = kmap_local_page(*dst_page);
209 			src_ptr = kmap_local_page(*src_page);
210 			memmove(dst_ptr + src, src_ptr + src, len);
211 			kunmap_local(src_ptr);
212 			set_page_dirty(*dst_page);
213 			kunmap_local(dst_ptr);
214 		} else {
215 			do {
216 				dst_ptr = kmap_local_page(*dst_page) + dst;
217 				src_ptr = kmap_local_page(*src_page) + src;
218 				if (src < dst) {
219 					l = src;
220 					src = PAGE_SIZE;
221 					dst -= l;
222 				} else {
223 					l = dst;
224 					src -= l;
225 					dst = PAGE_SIZE;
226 				}
227 				l = min(len, l);
228 				memmove(dst_ptr - l, src_ptr - l, l);
229 				kunmap_local(src_ptr);
230 				set_page_dirty(*dst_page);
231 				kunmap_local(dst_ptr);
232 				if (dst == PAGE_SIZE)
233 					dst_page--;
234 				else
235 					src_page--;
236 			} while ((len -= l));
237 		}
238 	} else {
239 		src_page = node->page + (src >> PAGE_SHIFT);
240 		src &= ~PAGE_MASK;
241 		dst_page = node->page + (dst >> PAGE_SHIFT);
242 		dst &= ~PAGE_MASK;
243 
244 		if (src == dst) {
245 			l = min_t(int, len, PAGE_SIZE - src);
246 
247 			dst_ptr = kmap_local_page(*dst_page) + src;
248 			src_ptr = kmap_local_page(*src_page) + src;
249 			memmove(dst_ptr, src_ptr, l);
250 			kunmap_local(src_ptr);
251 			set_page_dirty(*dst_page);
252 			kunmap_local(dst_ptr);
253 
254 			while ((len -= l) != 0) {
255 				l = min_t(int, len, PAGE_SIZE);
256 				dst_ptr = kmap_local_page(*++dst_page);
257 				src_ptr = kmap_local_page(*++src_page);
258 				memmove(dst_ptr, src_ptr, l);
259 				kunmap_local(src_ptr);
260 				set_page_dirty(*dst_page);
261 				kunmap_local(dst_ptr);
262 			}
263 		} else {
264 			do {
265 				dst_ptr = kmap_local_page(*dst_page) + dst;
266 				src_ptr = kmap_local_page(*src_page) + src;
267 				if (PAGE_SIZE - src <
268 						PAGE_SIZE - dst) {
269 					l = PAGE_SIZE - src;
270 					src = 0;
271 					dst += l;
272 				} else {
273 					l = PAGE_SIZE - dst;
274 					src += l;
275 					dst = 0;
276 				}
277 				l = min(len, l);
278 				memmove(dst_ptr, src_ptr, l);
279 				kunmap_local(src_ptr);
280 				set_page_dirty(*dst_page);
281 				kunmap_local(dst_ptr);
282 				if (!dst)
283 					dst_page++;
284 				else
285 					src_page++;
286 			} while ((len -= l));
287 		}
288 	}
289 }
290 
291 void hfs_bnode_dump(struct hfs_bnode *node)
292 {
293 	struct hfs_bnode_desc desc;
294 	__be32 cnid;
295 	int i, off, key_off;
296 
297 	hfs_dbg(BNODE_MOD, "bnode: %d\n", node->this);
298 	hfs_bnode_read(node, &desc, 0, sizeof(desc));
299 	hfs_dbg(BNODE_MOD, "%d, %d, %d, %d, %d\n",
300 		be32_to_cpu(desc.next), be32_to_cpu(desc.prev),
301 		desc.type, desc.height, be16_to_cpu(desc.num_recs));
302 
303 	off = node->tree->node_size - 2;
304 	for (i = be16_to_cpu(desc.num_recs); i >= 0; off -= 2, i--) {
305 		key_off = hfs_bnode_read_u16(node, off);
306 		hfs_dbg(BNODE_MOD, " %d", key_off);
307 		if (i && node->type == HFS_NODE_INDEX) {
308 			int tmp;
309 
310 			if (node->tree->attributes & HFS_TREE_VARIDXKEYS ||
311 					node->tree->cnid == HFSPLUS_ATTR_CNID)
312 				tmp = hfs_bnode_read_u16(node, key_off) + 2;
313 			else
314 				tmp = node->tree->max_key_len + 2;
315 			hfs_dbg_cont(BNODE_MOD, " (%d", tmp);
316 			hfs_bnode_read(node, &cnid, key_off + tmp, 4);
317 			hfs_dbg_cont(BNODE_MOD, ",%d)", be32_to_cpu(cnid));
318 		} else if (i && node->type == HFS_NODE_LEAF) {
319 			int tmp;
320 
321 			tmp = hfs_bnode_read_u16(node, key_off);
322 			hfs_dbg_cont(BNODE_MOD, " (%d)", tmp);
323 		}
324 	}
325 	hfs_dbg_cont(BNODE_MOD, "\n");
326 }
327 
328 void hfs_bnode_unlink(struct hfs_bnode *node)
329 {
330 	struct hfs_btree *tree;
331 	struct hfs_bnode *tmp;
332 	__be32 cnid;
333 
334 	tree = node->tree;
335 	if (node->prev) {
336 		tmp = hfs_bnode_find(tree, node->prev);
337 		if (IS_ERR(tmp))
338 			return;
339 		tmp->next = node->next;
340 		cnid = cpu_to_be32(tmp->next);
341 		hfs_bnode_write(tmp, &cnid,
342 			offsetof(struct hfs_bnode_desc, next), 4);
343 		hfs_bnode_put(tmp);
344 	} else if (node->type == HFS_NODE_LEAF)
345 		tree->leaf_head = node->next;
346 
347 	if (node->next) {
348 		tmp = hfs_bnode_find(tree, node->next);
349 		if (IS_ERR(tmp))
350 			return;
351 		tmp->prev = node->prev;
352 		cnid = cpu_to_be32(tmp->prev);
353 		hfs_bnode_write(tmp, &cnid,
354 			offsetof(struct hfs_bnode_desc, prev), 4);
355 		hfs_bnode_put(tmp);
356 	} else if (node->type == HFS_NODE_LEAF)
357 		tree->leaf_tail = node->prev;
358 
359 	/* move down? */
360 	if (!node->prev && !node->next)
361 		hfs_dbg(BNODE_MOD, "hfs_btree_del_level\n");
362 	if (!node->parent) {
363 		tree->root = 0;
364 		tree->depth = 0;
365 	}
366 	set_bit(HFS_BNODE_DELETED, &node->flags);
367 }
368 
369 static inline int hfs_bnode_hash(u32 num)
370 {
371 	num = (num >> 16) + num;
372 	num += num >> 8;
373 	return num & (NODE_HASH_SIZE - 1);
374 }
375 
376 struct hfs_bnode *hfs_bnode_findhash(struct hfs_btree *tree, u32 cnid)
377 {
378 	struct hfs_bnode *node;
379 
380 	if (cnid >= tree->node_count) {
381 		pr_err("request for non-existent node %d in B*Tree\n",
382 		       cnid);
383 		return NULL;
384 	}
385 
386 	for (node = tree->node_hash[hfs_bnode_hash(cnid)];
387 			node; node = node->next_hash)
388 		if (node->this == cnid)
389 			return node;
390 	return NULL;
391 }
392 
393 static struct hfs_bnode *__hfs_bnode_create(struct hfs_btree *tree, u32 cnid)
394 {
395 	struct hfs_bnode *node, *node2;
396 	struct address_space *mapping;
397 	struct page *page;
398 	int size, block, i, hash;
399 	loff_t off;
400 
401 	if (cnid >= tree->node_count) {
402 		pr_err("request for non-existent node %d in B*Tree\n",
403 		       cnid);
404 		return NULL;
405 	}
406 
407 	size = sizeof(struct hfs_bnode) + tree->pages_per_bnode *
408 		sizeof(struct page *);
409 	node = kzalloc(size, GFP_KERNEL);
410 	if (!node)
411 		return NULL;
412 	node->tree = tree;
413 	node->this = cnid;
414 	set_bit(HFS_BNODE_NEW, &node->flags);
415 	atomic_set(&node->refcnt, 1);
416 	hfs_dbg(BNODE_REFS, "new_node(%d:%d): 1\n",
417 		node->tree->cnid, node->this);
418 	init_waitqueue_head(&node->lock_wq);
419 	spin_lock(&tree->hash_lock);
420 	node2 = hfs_bnode_findhash(tree, cnid);
421 	if (!node2) {
422 		hash = hfs_bnode_hash(cnid);
423 		node->next_hash = tree->node_hash[hash];
424 		tree->node_hash[hash] = node;
425 		tree->node_hash_cnt++;
426 	} else {
427 		spin_unlock(&tree->hash_lock);
428 		kfree(node);
429 		wait_event(node2->lock_wq,
430 			!test_bit(HFS_BNODE_NEW, &node2->flags));
431 		return node2;
432 	}
433 	spin_unlock(&tree->hash_lock);
434 
435 	mapping = tree->inode->i_mapping;
436 	off = (loff_t)cnid << tree->node_size_shift;
437 	block = off >> PAGE_SHIFT;
438 	node->page_offset = off & ~PAGE_MASK;
439 	for (i = 0; i < tree->pages_per_bnode; block++, i++) {
440 		page = read_mapping_page(mapping, block, NULL);
441 		if (IS_ERR(page))
442 			goto fail;
443 		node->page[i] = page;
444 	}
445 
446 	return node;
447 fail:
448 	set_bit(HFS_BNODE_ERROR, &node->flags);
449 	return node;
450 }
451 
452 void hfs_bnode_unhash(struct hfs_bnode *node)
453 {
454 	struct hfs_bnode **p;
455 
456 	hfs_dbg(BNODE_REFS, "remove_node(%d:%d): %d\n",
457 		node->tree->cnid, node->this, atomic_read(&node->refcnt));
458 	for (p = &node->tree->node_hash[hfs_bnode_hash(node->this)];
459 	     *p && *p != node; p = &(*p)->next_hash)
460 		;
461 	BUG_ON(!*p);
462 	*p = node->next_hash;
463 	node->tree->node_hash_cnt--;
464 }
465 
466 /* Load a particular node out of a tree */
467 struct hfs_bnode *hfs_bnode_find(struct hfs_btree *tree, u32 num)
468 {
469 	struct hfs_bnode *node;
470 	struct hfs_bnode_desc *desc;
471 	int i, rec_off, off, next_off;
472 	int entry_size, key_size;
473 
474 	spin_lock(&tree->hash_lock);
475 	node = hfs_bnode_findhash(tree, num);
476 	if (node) {
477 		hfs_bnode_get(node);
478 		spin_unlock(&tree->hash_lock);
479 		wait_event(node->lock_wq,
480 			!test_bit(HFS_BNODE_NEW, &node->flags));
481 		if (test_bit(HFS_BNODE_ERROR, &node->flags))
482 			goto node_error;
483 		return node;
484 	}
485 	spin_unlock(&tree->hash_lock);
486 	node = __hfs_bnode_create(tree, num);
487 	if (!node)
488 		return ERR_PTR(-ENOMEM);
489 	if (test_bit(HFS_BNODE_ERROR, &node->flags))
490 		goto node_error;
491 	if (!test_bit(HFS_BNODE_NEW, &node->flags))
492 		return node;
493 
494 	desc = (struct hfs_bnode_desc *)(kmap_local_page(node->page[0]) +
495 							 node->page_offset);
496 	node->prev = be32_to_cpu(desc->prev);
497 	node->next = be32_to_cpu(desc->next);
498 	node->num_recs = be16_to_cpu(desc->num_recs);
499 	node->type = desc->type;
500 	node->height = desc->height;
501 	kunmap_local(desc);
502 
503 	switch (node->type) {
504 	case HFS_NODE_HEADER:
505 	case HFS_NODE_MAP:
506 		if (node->height != 0)
507 			goto node_error;
508 		break;
509 	case HFS_NODE_LEAF:
510 		if (node->height != 1)
511 			goto node_error;
512 		break;
513 	case HFS_NODE_INDEX:
514 		if (node->height <= 1 || node->height > tree->depth)
515 			goto node_error;
516 		break;
517 	default:
518 		goto node_error;
519 	}
520 
521 	rec_off = tree->node_size - 2;
522 	off = hfs_bnode_read_u16(node, rec_off);
523 	if (off != sizeof(struct hfs_bnode_desc))
524 		goto node_error;
525 	for (i = 1; i <= node->num_recs; off = next_off, i++) {
526 		rec_off -= 2;
527 		next_off = hfs_bnode_read_u16(node, rec_off);
528 		if (next_off <= off ||
529 		    next_off > tree->node_size ||
530 		    next_off & 1)
531 			goto node_error;
532 		entry_size = next_off - off;
533 		if (node->type != HFS_NODE_INDEX &&
534 		    node->type != HFS_NODE_LEAF)
535 			continue;
536 		key_size = hfs_bnode_read_u16(node, off) + 2;
537 		if (key_size >= entry_size || key_size & 1)
538 			goto node_error;
539 	}
540 	clear_bit(HFS_BNODE_NEW, &node->flags);
541 	wake_up(&node->lock_wq);
542 	return node;
543 
544 node_error:
545 	set_bit(HFS_BNODE_ERROR, &node->flags);
546 	clear_bit(HFS_BNODE_NEW, &node->flags);
547 	wake_up(&node->lock_wq);
548 	hfs_bnode_put(node);
549 	return ERR_PTR(-EIO);
550 }
551 
552 void hfs_bnode_free(struct hfs_bnode *node)
553 {
554 	int i;
555 
556 	for (i = 0; i < node->tree->pages_per_bnode; i++)
557 		if (node->page[i])
558 			put_page(node->page[i]);
559 	kfree(node);
560 }
561 
562 struct hfs_bnode *hfs_bnode_create(struct hfs_btree *tree, u32 num)
563 {
564 	struct hfs_bnode *node;
565 	struct page **pagep;
566 	int i;
567 
568 	spin_lock(&tree->hash_lock);
569 	node = hfs_bnode_findhash(tree, num);
570 	spin_unlock(&tree->hash_lock);
571 	if (node) {
572 		pr_crit("new node %u already hashed?\n", num);
573 		WARN_ON(1);
574 		return node;
575 	}
576 	node = __hfs_bnode_create(tree, num);
577 	if (!node)
578 		return ERR_PTR(-ENOMEM);
579 	if (test_bit(HFS_BNODE_ERROR, &node->flags)) {
580 		hfs_bnode_put(node);
581 		return ERR_PTR(-EIO);
582 	}
583 
584 	pagep = node->page;
585 	memzero_page(*pagep, node->page_offset,
586 		     min_t(int, PAGE_SIZE, tree->node_size));
587 	set_page_dirty(*pagep);
588 	for (i = 1; i < tree->pages_per_bnode; i++) {
589 		memzero_page(*++pagep, 0, PAGE_SIZE);
590 		set_page_dirty(*pagep);
591 	}
592 	clear_bit(HFS_BNODE_NEW, &node->flags);
593 	wake_up(&node->lock_wq);
594 
595 	return node;
596 }
597 
598 void hfs_bnode_get(struct hfs_bnode *node)
599 {
600 	if (node) {
601 		atomic_inc(&node->refcnt);
602 		hfs_dbg(BNODE_REFS, "get_node(%d:%d): %d\n",
603 			node->tree->cnid, node->this,
604 			atomic_read(&node->refcnt));
605 	}
606 }
607 
608 /* Dispose of resources used by a node */
609 void hfs_bnode_put(struct hfs_bnode *node)
610 {
611 	if (node) {
612 		struct hfs_btree *tree = node->tree;
613 		int i;
614 
615 		hfs_dbg(BNODE_REFS, "put_node(%d:%d): %d\n",
616 			node->tree->cnid, node->this,
617 			atomic_read(&node->refcnt));
618 		BUG_ON(!atomic_read(&node->refcnt));
619 		if (!atomic_dec_and_lock(&node->refcnt, &tree->hash_lock))
620 			return;
621 		for (i = 0; i < tree->pages_per_bnode; i++) {
622 			if (!node->page[i])
623 				continue;
624 			mark_page_accessed(node->page[i]);
625 		}
626 
627 		if (test_bit(HFS_BNODE_DELETED, &node->flags)) {
628 			hfs_bnode_unhash(node);
629 			spin_unlock(&tree->hash_lock);
630 			if (hfs_bnode_need_zeroout(tree))
631 				hfs_bnode_clear(node, 0, tree->node_size);
632 			hfs_bmap_free(node);
633 			hfs_bnode_free(node);
634 			return;
635 		}
636 		spin_unlock(&tree->hash_lock);
637 	}
638 }
639 
640 /*
641  * Unused nodes have to be zeroed if this is the catalog tree and
642  * a corresponding flag in the volume header is set.
643  */
644 bool hfs_bnode_need_zeroout(struct hfs_btree *tree)
645 {
646 	struct super_block *sb = tree->inode->i_sb;
647 	struct hfsplus_sb_info *sbi = HFSPLUS_SB(sb);
648 	const u32 volume_attr = be32_to_cpu(sbi->s_vhdr->attributes);
649 
650 	return tree->cnid == HFSPLUS_CAT_CNID &&
651 		volume_attr & HFSPLUS_VOL_UNUSED_NODE_FIX;
652 }
653