xref: /openbmc/linux/fs/ext4/dir.c (revision 37185b33)
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 
31 static unsigned char ext4_filetype_table[] = {
32 	DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
33 };
34 
35 static int ext4_dx_readdir(struct file *filp,
36 			   void *dirent, filldir_t filldir);
37 
38 static unsigned char get_dtype(struct super_block *sb, int filetype)
39 {
40 	if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE) ||
41 	    (filetype >= EXT4_FT_MAX))
42 		return DT_UNKNOWN;
43 
44 	return (ext4_filetype_table[filetype]);
45 }
46 
47 /**
48  * Check if the given dir-inode refers to an htree-indexed directory
49  * (or a directory which chould potentially get coverted to use htree
50  * indexing).
51  *
52  * Return 1 if it is a dx dir, 0 if not
53  */
54 static int is_dx_dir(struct inode *inode)
55 {
56 	struct super_block *sb = inode->i_sb;
57 
58 	if (EXT4_HAS_COMPAT_FEATURE(inode->i_sb,
59 		     EXT4_FEATURE_COMPAT_DIR_INDEX) &&
60 	    ((ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) ||
61 	     ((inode->i_size >> sb->s_blocksize_bits) == 1)))
62 		return 1;
63 
64 	return 0;
65 }
66 
67 /*
68  * Return 0 if the directory entry is OK, and 1 if there is a problem
69  *
70  * Note: this is the opposite of what ext2 and ext3 historically returned...
71  */
72 int __ext4_check_dir_entry(const char *function, unsigned int line,
73 			   struct inode *dir, struct file *filp,
74 			   struct ext4_dir_entry_2 *de,
75 			   struct buffer_head *bh,
76 			   unsigned int offset)
77 {
78 	const char *error_msg = NULL;
79 	const int rlen = ext4_rec_len_from_disk(de->rec_len,
80 						dir->i_sb->s_blocksize);
81 
82 	if (unlikely(rlen < EXT4_DIR_REC_LEN(1)))
83 		error_msg = "rec_len is smaller than minimal";
84 	else if (unlikely(rlen % 4 != 0))
85 		error_msg = "rec_len % 4 != 0";
86 	else if (unlikely(rlen < EXT4_DIR_REC_LEN(de->name_len)))
87 		error_msg = "rec_len is too small for name_len";
88 	else if (unlikely(((char *) de - bh->b_data) + rlen >
89 			  dir->i_sb->s_blocksize))
90 		error_msg = "directory entry across blocks";
91 	else if (unlikely(le32_to_cpu(de->inode) >
92 			le32_to_cpu(EXT4_SB(dir->i_sb)->s_es->s_inodes_count)))
93 		error_msg = "inode out of bounds";
94 	else
95 		return 0;
96 
97 	if (filp)
98 		ext4_error_file(filp, function, line, bh->b_blocknr,
99 				"bad entry in directory: %s - offset=%u(%u), "
100 				"inode=%u, rec_len=%d, name_len=%d",
101 				error_msg, (unsigned) (offset % bh->b_size),
102 				offset, le32_to_cpu(de->inode),
103 				rlen, de->name_len);
104 	else
105 		ext4_error_inode(dir, function, line, bh->b_blocknr,
106 				"bad entry in directory: %s - offset=%u(%u), "
107 				"inode=%u, rec_len=%d, name_len=%d",
108 				error_msg, (unsigned) (offset % bh->b_size),
109 				offset, le32_to_cpu(de->inode),
110 				rlen, de->name_len);
111 
112 	return 1;
113 }
114 
115 static int ext4_readdir(struct file *filp,
116 			 void *dirent, filldir_t filldir)
117 {
118 	int error = 0;
119 	unsigned int offset;
120 	int i, stored;
121 	struct ext4_dir_entry_2 *de;
122 	int err;
123 	struct inode *inode = filp->f_path.dentry->d_inode;
124 	struct super_block *sb = inode->i_sb;
125 	int ret = 0;
126 	int dir_has_error = 0;
127 
128 	if (is_dx_dir(inode)) {
129 		err = ext4_dx_readdir(filp, dirent, filldir);
130 		if (err != ERR_BAD_DX_DIR) {
131 			ret = err;
132 			goto out;
133 		}
134 		/*
135 		 * We don't set the inode dirty flag since it's not
136 		 * critical that it get flushed back to the disk.
137 		 */
138 		ext4_clear_inode_flag(filp->f_path.dentry->d_inode,
139 				      EXT4_INODE_INDEX);
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 			continue;
191 		}
192 		set_buffer_verified(bh);
193 
194 revalidate:
195 		/* If the dir block has changed since the last call to
196 		 * readdir(2), then we might be pointing to an invalid
197 		 * dirent right now.  Scan from the start of the block
198 		 * to make sure. */
199 		if (filp->f_version != inode->i_version) {
200 			for (i = 0; i < sb->s_blocksize && i < offset; ) {
201 				de = (struct ext4_dir_entry_2 *)
202 					(bh->b_data + i);
203 				/* It's too expensive to do a full
204 				 * dirent test each time round this
205 				 * loop, but we do have to test at
206 				 * least that it is non-zero.  A
207 				 * failure will be detected in the
208 				 * dirent test below. */
209 				if (ext4_rec_len_from_disk(de->rec_len,
210 					sb->s_blocksize) < EXT4_DIR_REC_LEN(1))
211 					break;
212 				i += ext4_rec_len_from_disk(de->rec_len,
213 							    sb->s_blocksize);
214 			}
215 			offset = i;
216 			filp->f_pos = (filp->f_pos & ~(sb->s_blocksize - 1))
217 				| offset;
218 			filp->f_version = inode->i_version;
219 		}
220 
221 		while (!error && filp->f_pos < inode->i_size
222 		       && offset < sb->s_blocksize) {
223 			de = (struct ext4_dir_entry_2 *) (bh->b_data + offset);
224 			if (ext4_check_dir_entry(inode, filp, de,
225 						 bh, offset)) {
226 				/*
227 				 * On error, skip the f_pos to the next block
228 				 */
229 				filp->f_pos = (filp->f_pos |
230 						(sb->s_blocksize - 1)) + 1;
231 				brelse(bh);
232 				ret = stored;
233 				goto out;
234 			}
235 			offset += ext4_rec_len_from_disk(de->rec_len,
236 					sb->s_blocksize);
237 			if (le32_to_cpu(de->inode)) {
238 				/* We might block in the next section
239 				 * if the data destination is
240 				 * currently swapped out.  So, use a
241 				 * version stamp to detect whether or
242 				 * not the directory has been modified
243 				 * during the copy operation.
244 				 */
245 				u64 version = filp->f_version;
246 
247 				error = filldir(dirent, de->name,
248 						de->name_len,
249 						filp->f_pos,
250 						le32_to_cpu(de->inode),
251 						get_dtype(sb, de->file_type));
252 				if (error)
253 					break;
254 				if (version != filp->f_version)
255 					goto revalidate;
256 				stored++;
257 			}
258 			filp->f_pos += ext4_rec_len_from_disk(de->rec_len,
259 						sb->s_blocksize);
260 		}
261 		offset = 0;
262 		brelse(bh);
263 	}
264 out:
265 	return ret;
266 }
267 
268 static inline int is_32bit_api(void)
269 {
270 #ifdef CONFIG_COMPAT
271 	return is_compat_task();
272 #else
273 	return (BITS_PER_LONG == 32);
274 #endif
275 }
276 
277 /*
278  * These functions convert from the major/minor hash to an f_pos
279  * value for dx directories
280  *
281  * Upper layer (for example NFS) should specify FMODE_32BITHASH or
282  * FMODE_64BITHASH explicitly. On the other hand, we allow ext4 to be mounted
283  * directly on both 32-bit and 64-bit nodes, under such case, neither
284  * FMODE_32BITHASH nor FMODE_64BITHASH is specified.
285  */
286 static inline loff_t hash2pos(struct file *filp, __u32 major, __u32 minor)
287 {
288 	if ((filp->f_mode & FMODE_32BITHASH) ||
289 	    (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
290 		return major >> 1;
291 	else
292 		return ((__u64)(major >> 1) << 32) | (__u64)minor;
293 }
294 
295 static inline __u32 pos2maj_hash(struct file *filp, loff_t pos)
296 {
297 	if ((filp->f_mode & FMODE_32BITHASH) ||
298 	    (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
299 		return (pos << 1) & 0xffffffff;
300 	else
301 		return ((pos >> 32) << 1) & 0xffffffff;
302 }
303 
304 static inline __u32 pos2min_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 0;
309 	else
310 		return pos & 0xffffffff;
311 }
312 
313 /*
314  * Return 32- or 64-bit end-of-file for dx directories
315  */
316 static inline loff_t ext4_get_htree_eof(struct file *filp)
317 {
318 	if ((filp->f_mode & FMODE_32BITHASH) ||
319 	    (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
320 		return EXT4_HTREE_EOF_32BIT;
321 	else
322 		return EXT4_HTREE_EOF_64BIT;
323 }
324 
325 
326 /*
327  * ext4_dir_llseek() calls generic_file_llseek_size to handle htree
328  * directories, where the "offset" is in terms of the filename hash
329  * value instead of the byte offset.
330  *
331  * Because we may return a 64-bit hash that is well beyond offset limits,
332  * we need to pass the max hash as the maximum allowable offset in
333  * the htree directory case.
334  *
335  * For non-htree, ext4_llseek already chooses the proper max offset.
336  */
337 loff_t ext4_dir_llseek(struct file *file, loff_t offset, int origin)
338 {
339 	struct inode *inode = file->f_mapping->host;
340 	int dx_dir = is_dx_dir(inode);
341 	loff_t htree_max = ext4_get_htree_eof(file);
342 
343 	if (likely(dx_dir))
344 		return generic_file_llseek_size(file, offset, origin,
345 						    htree_max, htree_max);
346 	else
347 		return ext4_llseek(file, offset, origin);
348 }
349 
350 /*
351  * This structure holds the nodes of the red-black tree used to store
352  * the directory entry in hash order.
353  */
354 struct fname {
355 	__u32		hash;
356 	__u32		minor_hash;
357 	struct rb_node	rb_hash;
358 	struct fname	*next;
359 	__u32		inode;
360 	__u8		name_len;
361 	__u8		file_type;
362 	char		name[0];
363 };
364 
365 /*
366  * This functoin implements a non-recursive way of freeing all of the
367  * nodes in the red-black tree.
368  */
369 static void free_rb_tree_fname(struct rb_root *root)
370 {
371 	struct rb_node	*n = root->rb_node;
372 	struct rb_node	*parent;
373 	struct fname	*fname;
374 
375 	while (n) {
376 		/* Do the node's children first */
377 		if (n->rb_left) {
378 			n = n->rb_left;
379 			continue;
380 		}
381 		if (n->rb_right) {
382 			n = n->rb_right;
383 			continue;
384 		}
385 		/*
386 		 * The node has no children; free it, and then zero
387 		 * out parent's link to it.  Finally go to the
388 		 * beginning of the loop and try to free the parent
389 		 * node.
390 		 */
391 		parent = rb_parent(n);
392 		fname = rb_entry(n, struct fname, rb_hash);
393 		while (fname) {
394 			struct fname *old = fname;
395 			fname = fname->next;
396 			kfree(old);
397 		}
398 		if (!parent)
399 			*root = RB_ROOT;
400 		else if (parent->rb_left == n)
401 			parent->rb_left = NULL;
402 		else if (parent->rb_right == n)
403 			parent->rb_right = NULL;
404 		n = parent;
405 	}
406 }
407 
408 
409 static struct dir_private_info *ext4_htree_create_dir_info(struct file *filp,
410 							   loff_t pos)
411 {
412 	struct dir_private_info *p;
413 
414 	p = kzalloc(sizeof(struct dir_private_info), GFP_KERNEL);
415 	if (!p)
416 		return NULL;
417 	p->curr_hash = pos2maj_hash(filp, pos);
418 	p->curr_minor_hash = pos2min_hash(filp, pos);
419 	return p;
420 }
421 
422 void ext4_htree_free_dir_info(struct dir_private_info *p)
423 {
424 	free_rb_tree_fname(&p->root);
425 	kfree(p);
426 }
427 
428 /*
429  * Given a directory entry, enter it into the fname rb tree.
430  */
431 int ext4_htree_store_dirent(struct file *dir_file, __u32 hash,
432 			     __u32 minor_hash,
433 			     struct ext4_dir_entry_2 *dirent)
434 {
435 	struct rb_node **p, *parent = NULL;
436 	struct fname *fname, *new_fn;
437 	struct dir_private_info *info;
438 	int len;
439 
440 	info = dir_file->private_data;
441 	p = &info->root.rb_node;
442 
443 	/* Create and allocate the fname structure */
444 	len = sizeof(struct fname) + dirent->name_len + 1;
445 	new_fn = kzalloc(len, GFP_KERNEL);
446 	if (!new_fn)
447 		return -ENOMEM;
448 	new_fn->hash = hash;
449 	new_fn->minor_hash = minor_hash;
450 	new_fn->inode = le32_to_cpu(dirent->inode);
451 	new_fn->name_len = dirent->name_len;
452 	new_fn->file_type = dirent->file_type;
453 	memcpy(new_fn->name, dirent->name, dirent->name_len);
454 	new_fn->name[dirent->name_len] = 0;
455 
456 	while (*p) {
457 		parent = *p;
458 		fname = rb_entry(parent, struct fname, rb_hash);
459 
460 		/*
461 		 * If the hash and minor hash match up, then we put
462 		 * them on a linked list.  This rarely happens...
463 		 */
464 		if ((new_fn->hash == fname->hash) &&
465 		    (new_fn->minor_hash == fname->minor_hash)) {
466 			new_fn->next = fname->next;
467 			fname->next = new_fn;
468 			return 0;
469 		}
470 
471 		if (new_fn->hash < fname->hash)
472 			p = &(*p)->rb_left;
473 		else if (new_fn->hash > fname->hash)
474 			p = &(*p)->rb_right;
475 		else if (new_fn->minor_hash < fname->minor_hash)
476 			p = &(*p)->rb_left;
477 		else /* if (new_fn->minor_hash > fname->minor_hash) */
478 			p = &(*p)->rb_right;
479 	}
480 
481 	rb_link_node(&new_fn->rb_hash, parent, p);
482 	rb_insert_color(&new_fn->rb_hash, &info->root);
483 	return 0;
484 }
485 
486 
487 
488 /*
489  * This is a helper function for ext4_dx_readdir.  It calls filldir
490  * for all entres on the fname linked list.  (Normally there is only
491  * one entry on the linked list, unless there are 62 bit hash collisions.)
492  */
493 static int call_filldir(struct file *filp, void *dirent,
494 			filldir_t filldir, struct fname *fname)
495 {
496 	struct dir_private_info *info = filp->private_data;
497 	loff_t	curr_pos;
498 	struct inode *inode = filp->f_path.dentry->d_inode;
499 	struct super_block *sb;
500 	int error;
501 
502 	sb = inode->i_sb;
503 
504 	if (!fname) {
505 		ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: comm %s: "
506 			 "called with null fname?!?", __func__, __LINE__,
507 			 inode->i_ino, current->comm);
508 		return 0;
509 	}
510 	curr_pos = hash2pos(filp, fname->hash, fname->minor_hash);
511 	while (fname) {
512 		error = filldir(dirent, fname->name,
513 				fname->name_len, curr_pos,
514 				fname->inode,
515 				get_dtype(sb, fname->file_type));
516 		if (error) {
517 			filp->f_pos = curr_pos;
518 			info->extra_fname = fname;
519 			return error;
520 		}
521 		fname = fname->next;
522 	}
523 	return 0;
524 }
525 
526 static int ext4_dx_readdir(struct file *filp,
527 			 void *dirent, filldir_t filldir)
528 {
529 	struct dir_private_info *info = filp->private_data;
530 	struct inode *inode = filp->f_path.dentry->d_inode;
531 	struct fname *fname;
532 	int	ret;
533 
534 	if (!info) {
535 		info = ext4_htree_create_dir_info(filp, filp->f_pos);
536 		if (!info)
537 			return -ENOMEM;
538 		filp->private_data = info;
539 	}
540 
541 	if (filp->f_pos == ext4_get_htree_eof(filp))
542 		return 0;	/* EOF */
543 
544 	/* Some one has messed with f_pos; reset the world */
545 	if (info->last_pos != filp->f_pos) {
546 		free_rb_tree_fname(&info->root);
547 		info->curr_node = NULL;
548 		info->extra_fname = NULL;
549 		info->curr_hash = pos2maj_hash(filp, filp->f_pos);
550 		info->curr_minor_hash = pos2min_hash(filp, filp->f_pos);
551 	}
552 
553 	/*
554 	 * If there are any leftover names on the hash collision
555 	 * chain, return them first.
556 	 */
557 	if (info->extra_fname) {
558 		if (call_filldir(filp, dirent, filldir, info->extra_fname))
559 			goto finished;
560 		info->extra_fname = NULL;
561 		goto next_node;
562 	} else if (!info->curr_node)
563 		info->curr_node = rb_first(&info->root);
564 
565 	while (1) {
566 		/*
567 		 * Fill the rbtree if we have no more entries,
568 		 * or the inode has changed since we last read in the
569 		 * cached entries.
570 		 */
571 		if ((!info->curr_node) ||
572 		    (filp->f_version != inode->i_version)) {
573 			info->curr_node = NULL;
574 			free_rb_tree_fname(&info->root);
575 			filp->f_version = inode->i_version;
576 			ret = ext4_htree_fill_tree(filp, info->curr_hash,
577 						   info->curr_minor_hash,
578 						   &info->next_hash);
579 			if (ret < 0)
580 				return ret;
581 			if (ret == 0) {
582 				filp->f_pos = ext4_get_htree_eof(filp);
583 				break;
584 			}
585 			info->curr_node = rb_first(&info->root);
586 		}
587 
588 		fname = rb_entry(info->curr_node, struct fname, rb_hash);
589 		info->curr_hash = fname->hash;
590 		info->curr_minor_hash = fname->minor_hash;
591 		if (call_filldir(filp, dirent, filldir, fname))
592 			break;
593 	next_node:
594 		info->curr_node = rb_next(info->curr_node);
595 		if (info->curr_node) {
596 			fname = rb_entry(info->curr_node, struct fname,
597 					 rb_hash);
598 			info->curr_hash = fname->hash;
599 			info->curr_minor_hash = fname->minor_hash;
600 		} else {
601 			if (info->next_hash == ~0) {
602 				filp->f_pos = ext4_get_htree_eof(filp);
603 				break;
604 			}
605 			info->curr_hash = info->next_hash;
606 			info->curr_minor_hash = 0;
607 		}
608 	}
609 finished:
610 	info->last_pos = filp->f_pos;
611 	return 0;
612 }
613 
614 static int ext4_release_dir(struct inode *inode, struct file *filp)
615 {
616 	if (filp->private_data)
617 		ext4_htree_free_dir_info(filp->private_data);
618 
619 	return 0;
620 }
621 
622 const struct file_operations ext4_dir_operations = {
623 	.llseek		= ext4_dir_llseek,
624 	.read		= generic_read_dir,
625 	.readdir	= ext4_readdir,
626 	.unlocked_ioctl = ext4_ioctl,
627 #ifdef CONFIG_COMPAT
628 	.compat_ioctl	= ext4_compat_ioctl,
629 #endif
630 	.fsync		= ext4_sync_file,
631 	.release	= ext4_release_dir,
632 };
633