xref: /linux/fs/ext4/dir.c (revision c0e297dc61f8d4453e07afbea1fa8d0e67cd4a34)
1 /*
2  *  linux/fs/ext4/dir.c
3  *
4  * Copyright (C) 1992, 1993, 1994, 1995
5  * Remy Card (card@masi.ibp.fr)
6  * Laboratoire MASI - Institut Blaise Pascal
7  * Universite Pierre et Marie Curie (Paris VI)
8  *
9  *  from
10  *
11  *  linux/fs/minix/dir.c
12  *
13  *  Copyright (C) 1991, 1992  Linus Torvalds
14  *
15  *  ext4 directory handling functions
16  *
17  *  Big-endian to little-endian byte-swapping/bitmaps by
18  *        David S. Miller (davem@caip.rutgers.edu), 1995
19  *
20  * Hash Tree Directory indexing (c) 2001  Daniel Phillips
21  *
22  */
23 
24 #include <linux/fs.h>
25 #include <linux/buffer_head.h>
26 #include <linux/slab.h>
27 #include "ext4.h"
28 #include "xattr.h"
29 
30 static int ext4_dx_readdir(struct file *, struct dir_context *);
31 
32 /**
33  * Check if the given dir-inode refers to an htree-indexed directory
34  * (or a directory which could potentially get converted to use htree
35  * indexing).
36  *
37  * Return 1 if it is a dx dir, 0 if not
38  */
39 static int is_dx_dir(struct inode *inode)
40 {
41 	struct super_block *sb = inode->i_sb;
42 
43 	if (EXT4_HAS_COMPAT_FEATURE(inode->i_sb,
44 		     EXT4_FEATURE_COMPAT_DIR_INDEX) &&
45 	    ((ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) ||
46 	     ((inode->i_size >> sb->s_blocksize_bits) == 1) ||
47 	     ext4_has_inline_data(inode)))
48 		return 1;
49 
50 	return 0;
51 }
52 
53 /*
54  * Return 0 if the directory entry is OK, and 1 if there is a problem
55  *
56  * Note: this is the opposite of what ext2 and ext3 historically returned...
57  *
58  * bh passed here can be an inode block or a dir data block, depending
59  * on the inode inline data flag.
60  */
61 int __ext4_check_dir_entry(const char *function, unsigned int line,
62 			   struct inode *dir, struct file *filp,
63 			   struct ext4_dir_entry_2 *de,
64 			   struct buffer_head *bh, char *buf, int size,
65 			   unsigned int offset)
66 {
67 	const char *error_msg = NULL;
68 	const int rlen = ext4_rec_len_from_disk(de->rec_len,
69 						dir->i_sb->s_blocksize);
70 
71 	if (unlikely(rlen < EXT4_DIR_REC_LEN(1)))
72 		error_msg = "rec_len is smaller than minimal";
73 	else if (unlikely(rlen % 4 != 0))
74 		error_msg = "rec_len % 4 != 0";
75 	else if (unlikely(rlen < EXT4_DIR_REC_LEN(de->name_len)))
76 		error_msg = "rec_len is too small for name_len";
77 	else if (unlikely(((char *) de - buf) + rlen > size))
78 		error_msg = "directory entry across range";
79 	else if (unlikely(le32_to_cpu(de->inode) >
80 			le32_to_cpu(EXT4_SB(dir->i_sb)->s_es->s_inodes_count)))
81 		error_msg = "inode out of bounds";
82 	else
83 		return 0;
84 
85 	if (filp)
86 		ext4_error_file(filp, function, line, bh->b_blocknr,
87 				"bad entry in directory: %s - offset=%u(%u), "
88 				"inode=%u, rec_len=%d, name_len=%d",
89 				error_msg, (unsigned) (offset % size),
90 				offset, le32_to_cpu(de->inode),
91 				rlen, de->name_len);
92 	else
93 		ext4_error_inode(dir, function, line, bh->b_blocknr,
94 				"bad entry in directory: %s - offset=%u(%u), "
95 				"inode=%u, rec_len=%d, name_len=%d",
96 				error_msg, (unsigned) (offset % size),
97 				offset, le32_to_cpu(de->inode),
98 				rlen, de->name_len);
99 
100 	return 1;
101 }
102 
103 static int ext4_readdir(struct file *file, struct dir_context *ctx)
104 {
105 	unsigned int offset;
106 	int i;
107 	struct ext4_dir_entry_2 *de;
108 	int err;
109 	struct inode *inode = file_inode(file);
110 	struct super_block *sb = inode->i_sb;
111 	struct buffer_head *bh = NULL;
112 	int dir_has_error = 0;
113 	struct ext4_str fname_crypto_str = {.name = NULL, .len = 0};
114 
115 	if (is_dx_dir(inode)) {
116 		err = ext4_dx_readdir(file, ctx);
117 		if (err != ERR_BAD_DX_DIR) {
118 			return err;
119 		}
120 		/*
121 		 * We don't set the inode dirty flag since it's not
122 		 * critical that it get flushed back to the disk.
123 		 */
124 		ext4_clear_inode_flag(file_inode(file),
125 				      EXT4_INODE_INDEX);
126 	}
127 
128 	if (ext4_has_inline_data(inode)) {
129 		int has_inline_data = 1;
130 		err = ext4_read_inline_dir(file, ctx,
131 					   &has_inline_data);
132 		if (has_inline_data)
133 			return err;
134 	}
135 
136 	if (ext4_encrypted_inode(inode)) {
137 		err = ext4_fname_crypto_alloc_buffer(inode, EXT4_NAME_LEN,
138 						     &fname_crypto_str);
139 		if (err < 0)
140 			return err;
141 	}
142 
143 	offset = ctx->pos & (sb->s_blocksize - 1);
144 
145 	while (ctx->pos < inode->i_size) {
146 		struct ext4_map_blocks map;
147 
148 		map.m_lblk = ctx->pos >> EXT4_BLOCK_SIZE_BITS(sb);
149 		map.m_len = 1;
150 		err = ext4_map_blocks(NULL, inode, &map, 0);
151 		if (err > 0) {
152 			pgoff_t index = map.m_pblk >>
153 					(PAGE_CACHE_SHIFT - inode->i_blkbits);
154 			if (!ra_has_index(&file->f_ra, index))
155 				page_cache_sync_readahead(
156 					sb->s_bdev->bd_inode->i_mapping,
157 					&file->f_ra, file,
158 					index, 1);
159 			file->f_ra.prev_pos = (loff_t)index << PAGE_CACHE_SHIFT;
160 			bh = ext4_bread(NULL, inode, map.m_lblk, 0);
161 			if (IS_ERR(bh))
162 				return PTR_ERR(bh);
163 		}
164 
165 		if (!bh) {
166 			if (!dir_has_error) {
167 				EXT4_ERROR_FILE(file, 0,
168 						"directory contains a "
169 						"hole at offset %llu",
170 					   (unsigned long long) ctx->pos);
171 				dir_has_error = 1;
172 			}
173 			/* corrupt size?  Maybe no more blocks to read */
174 			if (ctx->pos > inode->i_blocks << 9)
175 				break;
176 			ctx->pos += sb->s_blocksize - offset;
177 			continue;
178 		}
179 
180 		/* Check the checksum */
181 		if (!buffer_verified(bh) &&
182 		    !ext4_dirent_csum_verify(inode,
183 				(struct ext4_dir_entry *)bh->b_data)) {
184 			EXT4_ERROR_FILE(file, 0, "directory fails checksum "
185 					"at offset %llu",
186 					(unsigned long long)ctx->pos);
187 			ctx->pos += sb->s_blocksize - offset;
188 			brelse(bh);
189 			bh = NULL;
190 			continue;
191 		}
192 		set_buffer_verified(bh);
193 
194 		/* If the dir block has changed since the last call to
195 		 * readdir(2), then we might be pointing to an invalid
196 		 * dirent right now.  Scan from the start of the block
197 		 * to make sure. */
198 		if (file->f_version != inode->i_version) {
199 			for (i = 0; i < sb->s_blocksize && i < offset; ) {
200 				de = (struct ext4_dir_entry_2 *)
201 					(bh->b_data + i);
202 				/* It's too expensive to do a full
203 				 * dirent test each time round this
204 				 * loop, but we do have to test at
205 				 * least that it is non-zero.  A
206 				 * failure will be detected in the
207 				 * dirent test below. */
208 				if (ext4_rec_len_from_disk(de->rec_len,
209 					sb->s_blocksize) < EXT4_DIR_REC_LEN(1))
210 					break;
211 				i += ext4_rec_len_from_disk(de->rec_len,
212 							    sb->s_blocksize);
213 			}
214 			offset = i;
215 			ctx->pos = (ctx->pos & ~(sb->s_blocksize - 1))
216 				| offset;
217 			file->f_version = inode->i_version;
218 		}
219 
220 		while (ctx->pos < inode->i_size
221 		       && offset < sb->s_blocksize) {
222 			de = (struct ext4_dir_entry_2 *) (bh->b_data + offset);
223 			if (ext4_check_dir_entry(inode, file, de, bh,
224 						 bh->b_data, bh->b_size,
225 						 offset)) {
226 				/*
227 				 * On error, skip to the next block
228 				 */
229 				ctx->pos = (ctx->pos |
230 						(sb->s_blocksize - 1)) + 1;
231 				break;
232 			}
233 			offset += ext4_rec_len_from_disk(de->rec_len,
234 					sb->s_blocksize);
235 			if (le32_to_cpu(de->inode)) {
236 				if (!ext4_encrypted_inode(inode)) {
237 					if (!dir_emit(ctx, de->name,
238 					    de->name_len,
239 					    le32_to_cpu(de->inode),
240 					    get_dtype(sb, de->file_type)))
241 						goto done;
242 				} else {
243 					int save_len = fname_crypto_str.len;
244 
245 					/* Directory is encrypted */
246 					err = ext4_fname_disk_to_usr(inode,
247 						NULL, de, &fname_crypto_str);
248 					fname_crypto_str.len = save_len;
249 					if (err < 0)
250 						goto errout;
251 					if (!dir_emit(ctx,
252 					    fname_crypto_str.name, err,
253 					    le32_to_cpu(de->inode),
254 					    get_dtype(sb, de->file_type)))
255 						goto done;
256 				}
257 			}
258 			ctx->pos += ext4_rec_len_from_disk(de->rec_len,
259 						sb->s_blocksize);
260 		}
261 		if ((ctx->pos < inode->i_size) && !dir_relax(inode))
262 			goto done;
263 		brelse(bh);
264 		bh = NULL;
265 		offset = 0;
266 	}
267 done:
268 	err = 0;
269 errout:
270 #ifdef CONFIG_EXT4_FS_ENCRYPTION
271 	ext4_fname_crypto_free_buffer(&fname_crypto_str);
272 #endif
273 	brelse(bh);
274 	return err;
275 }
276 
277 static inline int is_32bit_api(void)
278 {
279 #ifdef CONFIG_COMPAT
280 	return is_compat_task();
281 #else
282 	return (BITS_PER_LONG == 32);
283 #endif
284 }
285 
286 /*
287  * These functions convert from the major/minor hash to an f_pos
288  * value for dx directories
289  *
290  * Upper layer (for example NFS) should specify FMODE_32BITHASH or
291  * FMODE_64BITHASH explicitly. On the other hand, we allow ext4 to be mounted
292  * directly on both 32-bit and 64-bit nodes, under such case, neither
293  * FMODE_32BITHASH nor FMODE_64BITHASH is specified.
294  */
295 static inline loff_t hash2pos(struct file *filp, __u32 major, __u32 minor)
296 {
297 	if ((filp->f_mode & FMODE_32BITHASH) ||
298 	    (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
299 		return major >> 1;
300 	else
301 		return ((__u64)(major >> 1) << 32) | (__u64)minor;
302 }
303 
304 static inline __u32 pos2maj_hash(struct file *filp, loff_t pos)
305 {
306 	if ((filp->f_mode & FMODE_32BITHASH) ||
307 	    (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
308 		return (pos << 1) & 0xffffffff;
309 	else
310 		return ((pos >> 32) << 1) & 0xffffffff;
311 }
312 
313 static inline __u32 pos2min_hash(struct file *filp, loff_t pos)
314 {
315 	if ((filp->f_mode & FMODE_32BITHASH) ||
316 	    (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
317 		return 0;
318 	else
319 		return pos & 0xffffffff;
320 }
321 
322 /*
323  * Return 32- or 64-bit end-of-file for dx directories
324  */
325 static inline loff_t ext4_get_htree_eof(struct file *filp)
326 {
327 	if ((filp->f_mode & FMODE_32BITHASH) ||
328 	    (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
329 		return EXT4_HTREE_EOF_32BIT;
330 	else
331 		return EXT4_HTREE_EOF_64BIT;
332 }
333 
334 
335 /*
336  * ext4_dir_llseek() calls generic_file_llseek_size to handle htree
337  * directories, where the "offset" is in terms of the filename hash
338  * value instead of the byte offset.
339  *
340  * Because we may return a 64-bit hash that is well beyond offset limits,
341  * we need to pass the max hash as the maximum allowable offset in
342  * the htree directory case.
343  *
344  * For non-htree, ext4_llseek already chooses the proper max offset.
345  */
346 static loff_t ext4_dir_llseek(struct file *file, loff_t offset, int whence)
347 {
348 	struct inode *inode = file->f_mapping->host;
349 	int dx_dir = is_dx_dir(inode);
350 	loff_t htree_max = ext4_get_htree_eof(file);
351 
352 	if (likely(dx_dir))
353 		return generic_file_llseek_size(file, offset, whence,
354 						    htree_max, htree_max);
355 	else
356 		return ext4_llseek(file, offset, whence);
357 }
358 
359 /*
360  * This structure holds the nodes of the red-black tree used to store
361  * the directory entry in hash order.
362  */
363 struct fname {
364 	__u32		hash;
365 	__u32		minor_hash;
366 	struct rb_node	rb_hash;
367 	struct fname	*next;
368 	__u32		inode;
369 	__u8		name_len;
370 	__u8		file_type;
371 	char		name[0];
372 };
373 
374 /*
375  * This functoin implements a non-recursive way of freeing all of the
376  * nodes in the red-black tree.
377  */
378 static void free_rb_tree_fname(struct rb_root *root)
379 {
380 	struct fname *fname, *next;
381 
382 	rbtree_postorder_for_each_entry_safe(fname, next, root, rb_hash)
383 		while (fname) {
384 			struct fname *old = fname;
385 			fname = fname->next;
386 			kfree(old);
387 		}
388 
389 	*root = RB_ROOT;
390 }
391 
392 
393 static struct dir_private_info *ext4_htree_create_dir_info(struct file *filp,
394 							   loff_t pos)
395 {
396 	struct dir_private_info *p;
397 
398 	p = kzalloc(sizeof(struct dir_private_info), GFP_KERNEL);
399 	if (!p)
400 		return NULL;
401 	p->curr_hash = pos2maj_hash(filp, pos);
402 	p->curr_minor_hash = pos2min_hash(filp, pos);
403 	return p;
404 }
405 
406 void ext4_htree_free_dir_info(struct dir_private_info *p)
407 {
408 	free_rb_tree_fname(&p->root);
409 	kfree(p);
410 }
411 
412 /*
413  * Given a directory entry, enter it into the fname rb tree.
414  *
415  * When filename encryption is enabled, the dirent will hold the
416  * encrypted filename, while the htree will hold decrypted filename.
417  * The decrypted filename is passed in via ent_name.  parameter.
418  */
419 int ext4_htree_store_dirent(struct file *dir_file, __u32 hash,
420 			     __u32 minor_hash,
421 			    struct ext4_dir_entry_2 *dirent,
422 			    struct ext4_str *ent_name)
423 {
424 	struct rb_node **p, *parent = NULL;
425 	struct fname *fname, *new_fn;
426 	struct dir_private_info *info;
427 	int len;
428 
429 	info = dir_file->private_data;
430 	p = &info->root.rb_node;
431 
432 	/* Create and allocate the fname structure */
433 	len = sizeof(struct fname) + ent_name->len + 1;
434 	new_fn = kzalloc(len, GFP_KERNEL);
435 	if (!new_fn)
436 		return -ENOMEM;
437 	new_fn->hash = hash;
438 	new_fn->minor_hash = minor_hash;
439 	new_fn->inode = le32_to_cpu(dirent->inode);
440 	new_fn->name_len = ent_name->len;
441 	new_fn->file_type = dirent->file_type;
442 	memcpy(new_fn->name, ent_name->name, ent_name->len);
443 	new_fn->name[ent_name->len] = 0;
444 
445 	while (*p) {
446 		parent = *p;
447 		fname = rb_entry(parent, struct fname, rb_hash);
448 
449 		/*
450 		 * If the hash and minor hash match up, then we put
451 		 * them on a linked list.  This rarely happens...
452 		 */
453 		if ((new_fn->hash == fname->hash) &&
454 		    (new_fn->minor_hash == fname->minor_hash)) {
455 			new_fn->next = fname->next;
456 			fname->next = new_fn;
457 			return 0;
458 		}
459 
460 		if (new_fn->hash < fname->hash)
461 			p = &(*p)->rb_left;
462 		else if (new_fn->hash > fname->hash)
463 			p = &(*p)->rb_right;
464 		else if (new_fn->minor_hash < fname->minor_hash)
465 			p = &(*p)->rb_left;
466 		else /* if (new_fn->minor_hash > fname->minor_hash) */
467 			p = &(*p)->rb_right;
468 	}
469 
470 	rb_link_node(&new_fn->rb_hash, parent, p);
471 	rb_insert_color(&new_fn->rb_hash, &info->root);
472 	return 0;
473 }
474 
475 
476 
477 /*
478  * This is a helper function for ext4_dx_readdir.  It calls filldir
479  * for all entres on the fname linked list.  (Normally there is only
480  * one entry on the linked list, unless there are 62 bit hash collisions.)
481  */
482 static int call_filldir(struct file *file, struct dir_context *ctx,
483 			struct fname *fname)
484 {
485 	struct dir_private_info *info = file->private_data;
486 	struct inode *inode = file_inode(file);
487 	struct super_block *sb = inode->i_sb;
488 
489 	if (!fname) {
490 		ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: comm %s: "
491 			 "called with null fname?!?", __func__, __LINE__,
492 			 inode->i_ino, current->comm);
493 		return 0;
494 	}
495 	ctx->pos = hash2pos(file, fname->hash, fname->minor_hash);
496 	while (fname) {
497 		if (!dir_emit(ctx, fname->name,
498 				fname->name_len,
499 				fname->inode,
500 				get_dtype(sb, fname->file_type))) {
501 			info->extra_fname = fname;
502 			return 1;
503 		}
504 		fname = fname->next;
505 	}
506 	return 0;
507 }
508 
509 static int ext4_dx_readdir(struct file *file, struct dir_context *ctx)
510 {
511 	struct dir_private_info *info = file->private_data;
512 	struct inode *inode = file_inode(file);
513 	struct fname *fname;
514 	int	ret;
515 
516 	if (!info) {
517 		info = ext4_htree_create_dir_info(file, ctx->pos);
518 		if (!info)
519 			return -ENOMEM;
520 		file->private_data = info;
521 	}
522 
523 	if (ctx->pos == ext4_get_htree_eof(file))
524 		return 0;	/* EOF */
525 
526 	/* Some one has messed with f_pos; reset the world */
527 	if (info->last_pos != ctx->pos) {
528 		free_rb_tree_fname(&info->root);
529 		info->curr_node = NULL;
530 		info->extra_fname = NULL;
531 		info->curr_hash = pos2maj_hash(file, ctx->pos);
532 		info->curr_minor_hash = pos2min_hash(file, ctx->pos);
533 	}
534 
535 	/*
536 	 * If there are any leftover names on the hash collision
537 	 * chain, return them first.
538 	 */
539 	if (info->extra_fname) {
540 		if (call_filldir(file, ctx, info->extra_fname))
541 			goto finished;
542 		info->extra_fname = NULL;
543 		goto next_node;
544 	} else if (!info->curr_node)
545 		info->curr_node = rb_first(&info->root);
546 
547 	while (1) {
548 		/*
549 		 * Fill the rbtree if we have no more entries,
550 		 * or the inode has changed since we last read in the
551 		 * cached entries.
552 		 */
553 		if ((!info->curr_node) ||
554 		    (file->f_version != inode->i_version)) {
555 			info->curr_node = NULL;
556 			free_rb_tree_fname(&info->root);
557 			file->f_version = inode->i_version;
558 			ret = ext4_htree_fill_tree(file, info->curr_hash,
559 						   info->curr_minor_hash,
560 						   &info->next_hash);
561 			if (ret < 0)
562 				return ret;
563 			if (ret == 0) {
564 				ctx->pos = ext4_get_htree_eof(file);
565 				break;
566 			}
567 			info->curr_node = rb_first(&info->root);
568 		}
569 
570 		fname = rb_entry(info->curr_node, struct fname, rb_hash);
571 		info->curr_hash = fname->hash;
572 		info->curr_minor_hash = fname->minor_hash;
573 		if (call_filldir(file, ctx, fname))
574 			break;
575 	next_node:
576 		info->curr_node = rb_next(info->curr_node);
577 		if (info->curr_node) {
578 			fname = rb_entry(info->curr_node, struct fname,
579 					 rb_hash);
580 			info->curr_hash = fname->hash;
581 			info->curr_minor_hash = fname->minor_hash;
582 		} else {
583 			if (info->next_hash == ~0) {
584 				ctx->pos = ext4_get_htree_eof(file);
585 				break;
586 			}
587 			info->curr_hash = info->next_hash;
588 			info->curr_minor_hash = 0;
589 		}
590 	}
591 finished:
592 	info->last_pos = ctx->pos;
593 	return 0;
594 }
595 
596 static int ext4_dir_open(struct inode * inode, struct file * filp)
597 {
598 	if (ext4_encrypted_inode(inode))
599 		return ext4_get_encryption_info(inode) ? -EACCES : 0;
600 	return 0;
601 }
602 
603 static int ext4_release_dir(struct inode *inode, struct file *filp)
604 {
605 	if (filp->private_data)
606 		ext4_htree_free_dir_info(filp->private_data);
607 
608 	return 0;
609 }
610 
611 int ext4_check_all_de(struct inode *dir, struct buffer_head *bh, void *buf,
612 		      int buf_size)
613 {
614 	struct ext4_dir_entry_2 *de;
615 	int nlen, rlen;
616 	unsigned int offset = 0;
617 	char *top;
618 
619 	de = (struct ext4_dir_entry_2 *)buf;
620 	top = buf + buf_size;
621 	while ((char *) de < top) {
622 		if (ext4_check_dir_entry(dir, NULL, de, bh,
623 					 buf, buf_size, offset))
624 			return -EIO;
625 		nlen = EXT4_DIR_REC_LEN(de->name_len);
626 		rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
627 		de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
628 		offset += rlen;
629 	}
630 	if ((char *) de > top)
631 		return -EIO;
632 
633 	return 0;
634 }
635 
636 const struct file_operations ext4_dir_operations = {
637 	.llseek		= ext4_dir_llseek,
638 	.read		= generic_read_dir,
639 	.iterate	= ext4_readdir,
640 	.unlocked_ioctl = ext4_ioctl,
641 #ifdef CONFIG_COMPAT
642 	.compat_ioctl	= ext4_compat_ioctl,
643 #endif
644 	.fsync		= ext4_sync_file,
645 	.open		= ext4_dir_open,
646 	.release	= ext4_release_dir,
647 };
648