xref: /linux/fs/ext4/dir.c (revision 6eb2fb3170549737207974c2c6ad34bcc2f3025e)
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/jbd2.h>
26 #include <linux/buffer_head.h>
27 #include <linux/slab.h>
28 #include <linux/rbtree.h>
29 #include "ext4.h"
30 #include "xattr.h"
31 
32 static int ext4_dx_readdir(struct file *filp,
33 			   void *dirent, filldir_t filldir);
34 
35 /**
36  * Check if the given dir-inode refers to an htree-indexed directory
37  * (or a directory which chould potentially get coverted to use htree
38  * indexing).
39  *
40  * Return 1 if it is a dx dir, 0 if not
41  */
42 static int is_dx_dir(struct inode *inode)
43 {
44 	struct super_block *sb = inode->i_sb;
45 
46 	if (EXT4_HAS_COMPAT_FEATURE(inode->i_sb,
47 		     EXT4_FEATURE_COMPAT_DIR_INDEX) &&
48 	    ((ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) ||
49 	     ((inode->i_size >> sb->s_blocksize_bits) == 1) ||
50 	     ext4_has_inline_data(inode)))
51 		return 1;
52 
53 	return 0;
54 }
55 
56 /*
57  * Return 0 if the directory entry is OK, and 1 if there is a problem
58  *
59  * Note: this is the opposite of what ext2 and ext3 historically returned...
60  *
61  * bh passed here can be an inode block or a dir data block, depending
62  * on the inode inline data flag.
63  */
64 int __ext4_check_dir_entry(const char *function, unsigned int line,
65 			   struct inode *dir, struct file *filp,
66 			   struct ext4_dir_entry_2 *de,
67 			   struct buffer_head *bh, char *buf, int size,
68 			   unsigned int offset)
69 {
70 	const char *error_msg = NULL;
71 	const int rlen = ext4_rec_len_from_disk(de->rec_len,
72 						dir->i_sb->s_blocksize);
73 
74 	if (unlikely(rlen < EXT4_DIR_REC_LEN(1)))
75 		error_msg = "rec_len is smaller than minimal";
76 	else if (unlikely(rlen % 4 != 0))
77 		error_msg = "rec_len % 4 != 0";
78 	else if (unlikely(rlen < EXT4_DIR_REC_LEN(de->name_len)))
79 		error_msg = "rec_len is too small for name_len";
80 	else if (unlikely(((char *) de - buf) + rlen > size))
81 		error_msg = "directory entry across range";
82 	else if (unlikely(le32_to_cpu(de->inode) >
83 			le32_to_cpu(EXT4_SB(dir->i_sb)->s_es->s_inodes_count)))
84 		error_msg = "inode out of bounds";
85 	else
86 		return 0;
87 
88 	if (filp)
89 		ext4_error_file(filp, function, line, bh->b_blocknr,
90 				"bad entry in directory: %s - offset=%u(%u), "
91 				"inode=%u, rec_len=%d, name_len=%d",
92 				error_msg, (unsigned) (offset % size),
93 				offset, le32_to_cpu(de->inode),
94 				rlen, de->name_len);
95 	else
96 		ext4_error_inode(dir, function, line, bh->b_blocknr,
97 				"bad entry in directory: %s - offset=%u(%u), "
98 				"inode=%u, rec_len=%d, name_len=%d",
99 				error_msg, (unsigned) (offset % size),
100 				offset, le32_to_cpu(de->inode),
101 				rlen, de->name_len);
102 
103 	return 1;
104 }
105 
106 static int ext4_readdir(struct file *filp,
107 			 void *dirent, filldir_t filldir)
108 {
109 	int error = 0;
110 	unsigned int offset;
111 	int i, stored;
112 	struct ext4_dir_entry_2 *de;
113 	int err;
114 	struct inode *inode = file_inode(filp);
115 	struct super_block *sb = inode->i_sb;
116 	int ret = 0;
117 	int dir_has_error = 0;
118 
119 	if (is_dx_dir(inode)) {
120 		err = ext4_dx_readdir(filp, dirent, filldir);
121 		if (err != ERR_BAD_DX_DIR) {
122 			ret = err;
123 			goto out;
124 		}
125 		/*
126 		 * We don't set the inode dirty flag since it's not
127 		 * critical that it get flushed back to the disk.
128 		 */
129 		ext4_clear_inode_flag(file_inode(filp),
130 				      EXT4_INODE_INDEX);
131 	}
132 
133 	if (ext4_has_inline_data(inode)) {
134 		int has_inline_data = 1;
135 		ret = ext4_read_inline_dir(filp, dirent, filldir,
136 					   &has_inline_data);
137 		if (has_inline_data)
138 			return ret;
139 	}
140 
141 	stored = 0;
142 	offset = filp->f_pos & (sb->s_blocksize - 1);
143 
144 	while (!error && !stored && filp->f_pos < inode->i_size) {
145 		struct ext4_map_blocks map;
146 		struct buffer_head *bh = NULL;
147 
148 		map.m_lblk = filp->f_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(&filp->f_ra, index))
155 				page_cache_sync_readahead(
156 					sb->s_bdev->bd_inode->i_mapping,
157 					&filp->f_ra, filp,
158 					index, 1);
159 			filp->f_ra.prev_pos = (loff_t)index << PAGE_CACHE_SHIFT;
160 			bh = ext4_bread(NULL, inode, map.m_lblk, 0, &err);
161 		}
162 
163 		/*
164 		 * We ignore I/O errors on directories so users have a chance
165 		 * of recovering data when there's a bad sector
166 		 */
167 		if (!bh) {
168 			if (!dir_has_error) {
169 				EXT4_ERROR_FILE(filp, 0,
170 						"directory contains a "
171 						"hole at offset %llu",
172 					   (unsigned long long) filp->f_pos);
173 				dir_has_error = 1;
174 			}
175 			/* corrupt size?  Maybe no more blocks to read */
176 			if (filp->f_pos > inode->i_blocks << 9)
177 				break;
178 			filp->f_pos += sb->s_blocksize - offset;
179 			continue;
180 		}
181 
182 		/* Check the checksum */
183 		if (!buffer_verified(bh) &&
184 		    !ext4_dirent_csum_verify(inode,
185 				(struct ext4_dir_entry *)bh->b_data)) {
186 			EXT4_ERROR_FILE(filp, 0, "directory fails checksum "
187 					"at offset %llu",
188 					(unsigned long long)filp->f_pos);
189 			filp->f_pos += sb->s_blocksize - offset;
190 			brelse(bh);
191 			continue;
192 		}
193 		set_buffer_verified(bh);
194 
195 revalidate:
196 		/* If the dir block has changed since the last call to
197 		 * readdir(2), then we might be pointing to an invalid
198 		 * dirent right now.  Scan from the start of the block
199 		 * to make sure. */
200 		if (filp->f_version != inode->i_version) {
201 			for (i = 0; i < sb->s_blocksize && i < offset; ) {
202 				de = (struct ext4_dir_entry_2 *)
203 					(bh->b_data + i);
204 				/* It's too expensive to do a full
205 				 * dirent test each time round this
206 				 * loop, but we do have to test at
207 				 * least that it is non-zero.  A
208 				 * failure will be detected in the
209 				 * dirent test below. */
210 				if (ext4_rec_len_from_disk(de->rec_len,
211 					sb->s_blocksize) < EXT4_DIR_REC_LEN(1))
212 					break;
213 				i += ext4_rec_len_from_disk(de->rec_len,
214 							    sb->s_blocksize);
215 			}
216 			offset = i;
217 			filp->f_pos = (filp->f_pos & ~(sb->s_blocksize - 1))
218 				| offset;
219 			filp->f_version = inode->i_version;
220 		}
221 
222 		while (!error && filp->f_pos < inode->i_size
223 		       && offset < sb->s_blocksize) {
224 			de = (struct ext4_dir_entry_2 *) (bh->b_data + offset);
225 			if (ext4_check_dir_entry(inode, filp, de, bh,
226 						 bh->b_data, bh->b_size,
227 						 offset)) {
228 				/*
229 				 * On error, skip the f_pos to the next block
230 				 */
231 				filp->f_pos = (filp->f_pos |
232 						(sb->s_blocksize - 1)) + 1;
233 				brelse(bh);
234 				ret = stored;
235 				goto out;
236 			}
237 			offset += ext4_rec_len_from_disk(de->rec_len,
238 					sb->s_blocksize);
239 			if (le32_to_cpu(de->inode)) {
240 				/* We might block in the next section
241 				 * if the data destination is
242 				 * currently swapped out.  So, use a
243 				 * version stamp to detect whether or
244 				 * not the directory has been modified
245 				 * during the copy operation.
246 				 */
247 				u64 version = filp->f_version;
248 
249 				error = filldir(dirent, de->name,
250 						de->name_len,
251 						filp->f_pos,
252 						le32_to_cpu(de->inode),
253 						get_dtype(sb, de->file_type));
254 				if (error)
255 					break;
256 				if (version != filp->f_version)
257 					goto revalidate;
258 				stored++;
259 			}
260 			filp->f_pos += ext4_rec_len_from_disk(de->rec_len,
261 						sb->s_blocksize);
262 		}
263 		offset = 0;
264 		brelse(bh);
265 	}
266 out:
267 	return ret;
268 }
269 
270 static inline int is_32bit_api(void)
271 {
272 #ifdef CONFIG_COMPAT
273 	return is_compat_task();
274 #else
275 	return (BITS_PER_LONG == 32);
276 #endif
277 }
278 
279 /*
280  * These functions convert from the major/minor hash to an f_pos
281  * value for dx directories
282  *
283  * Upper layer (for example NFS) should specify FMODE_32BITHASH or
284  * FMODE_64BITHASH explicitly. On the other hand, we allow ext4 to be mounted
285  * directly on both 32-bit and 64-bit nodes, under such case, neither
286  * FMODE_32BITHASH nor FMODE_64BITHASH is specified.
287  */
288 static inline loff_t hash2pos(struct file *filp, __u32 major, __u32 minor)
289 {
290 	if ((filp->f_mode & FMODE_32BITHASH) ||
291 	    (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
292 		return major >> 1;
293 	else
294 		return ((__u64)(major >> 1) << 32) | (__u64)minor;
295 }
296 
297 static inline __u32 pos2maj_hash(struct file *filp, loff_t pos)
298 {
299 	if ((filp->f_mode & FMODE_32BITHASH) ||
300 	    (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
301 		return (pos << 1) & 0xffffffff;
302 	else
303 		return ((pos >> 32) << 1) & 0xffffffff;
304 }
305 
306 static inline __u32 pos2min_hash(struct file *filp, loff_t pos)
307 {
308 	if ((filp->f_mode & FMODE_32BITHASH) ||
309 	    (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
310 		return 0;
311 	else
312 		return pos & 0xffffffff;
313 }
314 
315 /*
316  * Return 32- or 64-bit end-of-file for dx directories
317  */
318 static inline loff_t ext4_get_htree_eof(struct file *filp)
319 {
320 	if ((filp->f_mode & FMODE_32BITHASH) ||
321 	    (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
322 		return EXT4_HTREE_EOF_32BIT;
323 	else
324 		return EXT4_HTREE_EOF_64BIT;
325 }
326 
327 
328 /*
329  * ext4_dir_llseek() calls generic_file_llseek_size to handle htree
330  * directories, where the "offset" is in terms of the filename hash
331  * value instead of the byte offset.
332  *
333  * Because we may return a 64-bit hash that is well beyond offset limits,
334  * we need to pass the max hash as the maximum allowable offset in
335  * the htree directory case.
336  *
337  * For non-htree, ext4_llseek already chooses the proper max offset.
338  */
339 static loff_t ext4_dir_llseek(struct file *file, loff_t offset, int whence)
340 {
341 	struct inode *inode = file->f_mapping->host;
342 	int dx_dir = is_dx_dir(inode);
343 	loff_t htree_max = ext4_get_htree_eof(file);
344 
345 	if (likely(dx_dir))
346 		return generic_file_llseek_size(file, offset, whence,
347 						    htree_max, htree_max);
348 	else
349 		return ext4_llseek(file, offset, whence);
350 }
351 
352 /*
353  * This structure holds the nodes of the red-black tree used to store
354  * the directory entry in hash order.
355  */
356 struct fname {
357 	__u32		hash;
358 	__u32		minor_hash;
359 	struct rb_node	rb_hash;
360 	struct fname	*next;
361 	__u32		inode;
362 	__u8		name_len;
363 	__u8		file_type;
364 	char		name[0];
365 };
366 
367 /*
368  * This functoin implements a non-recursive way of freeing all of the
369  * nodes in the red-black tree.
370  */
371 static void free_rb_tree_fname(struct rb_root *root)
372 {
373 	struct rb_node	*n = root->rb_node;
374 	struct rb_node	*parent;
375 	struct fname	*fname;
376 
377 	while (n) {
378 		/* Do the node's children first */
379 		if (n->rb_left) {
380 			n = n->rb_left;
381 			continue;
382 		}
383 		if (n->rb_right) {
384 			n = n->rb_right;
385 			continue;
386 		}
387 		/*
388 		 * The node has no children; free it, and then zero
389 		 * out parent's link to it.  Finally go to the
390 		 * beginning of the loop and try to free the parent
391 		 * node.
392 		 */
393 		parent = rb_parent(n);
394 		fname = rb_entry(n, struct fname, rb_hash);
395 		while (fname) {
396 			struct fname *old = fname;
397 			fname = fname->next;
398 			kfree(old);
399 		}
400 		if (!parent)
401 			*root = RB_ROOT;
402 		else if (parent->rb_left == n)
403 			parent->rb_left = NULL;
404 		else if (parent->rb_right == n)
405 			parent->rb_right = NULL;
406 		n = parent;
407 	}
408 }
409 
410 
411 static struct dir_private_info *ext4_htree_create_dir_info(struct file *filp,
412 							   loff_t pos)
413 {
414 	struct dir_private_info *p;
415 
416 	p = kzalloc(sizeof(struct dir_private_info), GFP_KERNEL);
417 	if (!p)
418 		return NULL;
419 	p->curr_hash = pos2maj_hash(filp, pos);
420 	p->curr_minor_hash = pos2min_hash(filp, pos);
421 	return p;
422 }
423 
424 void ext4_htree_free_dir_info(struct dir_private_info *p)
425 {
426 	free_rb_tree_fname(&p->root);
427 	kfree(p);
428 }
429 
430 /*
431  * Given a directory entry, enter it into the fname rb tree.
432  */
433 int ext4_htree_store_dirent(struct file *dir_file, __u32 hash,
434 			     __u32 minor_hash,
435 			     struct ext4_dir_entry_2 *dirent)
436 {
437 	struct rb_node **p, *parent = NULL;
438 	struct fname *fname, *new_fn;
439 	struct dir_private_info *info;
440 	int len;
441 
442 	info = dir_file->private_data;
443 	p = &info->root.rb_node;
444 
445 	/* Create and allocate the fname structure */
446 	len = sizeof(struct fname) + dirent->name_len + 1;
447 	new_fn = kzalloc(len, GFP_KERNEL);
448 	if (!new_fn)
449 		return -ENOMEM;
450 	new_fn->hash = hash;
451 	new_fn->minor_hash = minor_hash;
452 	new_fn->inode = le32_to_cpu(dirent->inode);
453 	new_fn->name_len = dirent->name_len;
454 	new_fn->file_type = dirent->file_type;
455 	memcpy(new_fn->name, dirent->name, dirent->name_len);
456 	new_fn->name[dirent->name_len] = 0;
457 
458 	while (*p) {
459 		parent = *p;
460 		fname = rb_entry(parent, struct fname, rb_hash);
461 
462 		/*
463 		 * If the hash and minor hash match up, then we put
464 		 * them on a linked list.  This rarely happens...
465 		 */
466 		if ((new_fn->hash == fname->hash) &&
467 		    (new_fn->minor_hash == fname->minor_hash)) {
468 			new_fn->next = fname->next;
469 			fname->next = new_fn;
470 			return 0;
471 		}
472 
473 		if (new_fn->hash < fname->hash)
474 			p = &(*p)->rb_left;
475 		else if (new_fn->hash > fname->hash)
476 			p = &(*p)->rb_right;
477 		else if (new_fn->minor_hash < fname->minor_hash)
478 			p = &(*p)->rb_left;
479 		else /* if (new_fn->minor_hash > fname->minor_hash) */
480 			p = &(*p)->rb_right;
481 	}
482 
483 	rb_link_node(&new_fn->rb_hash, parent, p);
484 	rb_insert_color(&new_fn->rb_hash, &info->root);
485 	return 0;
486 }
487 
488 
489 
490 /*
491  * This is a helper function for ext4_dx_readdir.  It calls filldir
492  * for all entres on the fname linked list.  (Normally there is only
493  * one entry on the linked list, unless there are 62 bit hash collisions.)
494  */
495 static int call_filldir(struct file *filp, void *dirent,
496 			filldir_t filldir, struct fname *fname)
497 {
498 	struct dir_private_info *info = filp->private_data;
499 	loff_t	curr_pos;
500 	struct inode *inode = file_inode(filp);
501 	struct super_block *sb;
502 	int error;
503 
504 	sb = inode->i_sb;
505 
506 	if (!fname) {
507 		ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: comm %s: "
508 			 "called with null fname?!?", __func__, __LINE__,
509 			 inode->i_ino, current->comm);
510 		return 0;
511 	}
512 	curr_pos = hash2pos(filp, fname->hash, fname->minor_hash);
513 	while (fname) {
514 		error = filldir(dirent, fname->name,
515 				fname->name_len, curr_pos,
516 				fname->inode,
517 				get_dtype(sb, fname->file_type));
518 		if (error) {
519 			filp->f_pos = curr_pos;
520 			info->extra_fname = fname;
521 			return error;
522 		}
523 		fname = fname->next;
524 	}
525 	return 0;
526 }
527 
528 static int ext4_dx_readdir(struct file *filp,
529 			 void *dirent, filldir_t filldir)
530 {
531 	struct dir_private_info *info = filp->private_data;
532 	struct inode *inode = file_inode(filp);
533 	struct fname *fname;
534 	int	ret;
535 
536 	if (!info) {
537 		info = ext4_htree_create_dir_info(filp, filp->f_pos);
538 		if (!info)
539 			return -ENOMEM;
540 		filp->private_data = info;
541 	}
542 
543 	if (filp->f_pos == ext4_get_htree_eof(filp))
544 		return 0;	/* EOF */
545 
546 	/* Some one has messed with f_pos; reset the world */
547 	if (info->last_pos != filp->f_pos) {
548 		free_rb_tree_fname(&info->root);
549 		info->curr_node = NULL;
550 		info->extra_fname = NULL;
551 		info->curr_hash = pos2maj_hash(filp, filp->f_pos);
552 		info->curr_minor_hash = pos2min_hash(filp, filp->f_pos);
553 	}
554 
555 	/*
556 	 * If there are any leftover names on the hash collision
557 	 * chain, return them first.
558 	 */
559 	if (info->extra_fname) {
560 		if (call_filldir(filp, dirent, filldir, info->extra_fname))
561 			goto finished;
562 		info->extra_fname = NULL;
563 		goto next_node;
564 	} else if (!info->curr_node)
565 		info->curr_node = rb_first(&info->root);
566 
567 	while (1) {
568 		/*
569 		 * Fill the rbtree if we have no more entries,
570 		 * or the inode has changed since we last read in the
571 		 * cached entries.
572 		 */
573 		if ((!info->curr_node) ||
574 		    (filp->f_version != inode->i_version)) {
575 			info->curr_node = NULL;
576 			free_rb_tree_fname(&info->root);
577 			filp->f_version = inode->i_version;
578 			ret = ext4_htree_fill_tree(filp, info->curr_hash,
579 						   info->curr_minor_hash,
580 						   &info->next_hash);
581 			if (ret < 0)
582 				return ret;
583 			if (ret == 0) {
584 				filp->f_pos = ext4_get_htree_eof(filp);
585 				break;
586 			}
587 			info->curr_node = rb_first(&info->root);
588 		}
589 
590 		fname = rb_entry(info->curr_node, struct fname, rb_hash);
591 		info->curr_hash = fname->hash;
592 		info->curr_minor_hash = fname->minor_hash;
593 		if (call_filldir(filp, dirent, filldir, fname))
594 			break;
595 	next_node:
596 		info->curr_node = rb_next(info->curr_node);
597 		if (info->curr_node) {
598 			fname = rb_entry(info->curr_node, struct fname,
599 					 rb_hash);
600 			info->curr_hash = fname->hash;
601 			info->curr_minor_hash = fname->minor_hash;
602 		} else {
603 			if (info->next_hash == ~0) {
604 				filp->f_pos = ext4_get_htree_eof(filp);
605 				break;
606 			}
607 			info->curr_hash = info->next_hash;
608 			info->curr_minor_hash = 0;
609 		}
610 	}
611 finished:
612 	info->last_pos = filp->f_pos;
613 	return 0;
614 }
615 
616 static int ext4_release_dir(struct inode *inode, struct file *filp)
617 {
618 	if (filp->private_data)
619 		ext4_htree_free_dir_info(filp->private_data);
620 
621 	return 0;
622 }
623 
624 const struct file_operations ext4_dir_operations = {
625 	.llseek		= ext4_dir_llseek,
626 	.read		= generic_read_dir,
627 	.readdir	= ext4_readdir,
628 	.unlocked_ioctl = ext4_ioctl,
629 #ifdef CONFIG_COMPAT
630 	.compat_ioctl	= ext4_compat_ioctl,
631 #endif
632 	.fsync		= ext4_sync_file,
633 	.release	= ext4_release_dir,
634 };
635