xref: /linux/fs/ext4/dir.c (revision 0dd9ac63ce26ec87b080ca9c3e6efed33c23ace6)
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_readdir(struct file *, void *, filldir_t);
36 static int ext4_dx_readdir(struct file *filp,
37 			   void *dirent, filldir_t filldir);
38 static int ext4_release_dir(struct inode *inode,
39 				struct file *filp);
40 
41 const struct file_operations ext4_dir_operations = {
42 	.llseek		= generic_file_llseek,
43 	.read		= generic_read_dir,
44 	.readdir	= ext4_readdir,		/* we take BKL. needed?*/
45 	.unlocked_ioctl = ext4_ioctl,
46 #ifdef CONFIG_COMPAT
47 	.compat_ioctl	= ext4_compat_ioctl,
48 #endif
49 	.fsync		= ext4_sync_file,
50 	.release	= ext4_release_dir,
51 };
52 
53 
54 static unsigned char get_dtype(struct super_block *sb, int filetype)
55 {
56 	if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE) ||
57 	    (filetype >= EXT4_FT_MAX))
58 		return DT_UNKNOWN;
59 
60 	return (ext4_filetype_table[filetype]);
61 }
62 
63 
64 int ext4_check_dir_entry(const char *function, struct inode *dir,
65 			 struct ext4_dir_entry_2 *de,
66 			 struct buffer_head *bh,
67 			 unsigned int offset)
68 {
69 	const char *error_msg = NULL;
70 	const int rlen = ext4_rec_len_from_disk(de->rec_len,
71 						dir->i_sb->s_blocksize);
72 
73 	if (rlen < EXT4_DIR_REC_LEN(1))
74 		error_msg = "rec_len is smaller than minimal";
75 	else if (rlen % 4 != 0)
76 		error_msg = "rec_len % 4 != 0";
77 	else if (rlen < EXT4_DIR_REC_LEN(de->name_len))
78 		error_msg = "rec_len is too small for name_len";
79 	else if (((char *) de - bh->b_data) + rlen > dir->i_sb->s_blocksize)
80 		error_msg = "directory entry across blocks";
81 	else if (le32_to_cpu(de->inode) >
82 			le32_to_cpu(EXT4_SB(dir->i_sb)->s_es->s_inodes_count))
83 		error_msg = "inode out of bounds";
84 
85 	if (error_msg != NULL)
86 		ext4_error_inode(function, dir,
87 			"bad entry in directory: %s - block=%llu"
88 			"offset=%u(%u), inode=%u, rec_len=%d, name_len=%d",
89 			error_msg, (unsigned long long) bh->b_blocknr,
90 			(unsigned) (offset%bh->b_size), offset,
91 			le32_to_cpu(de->inode),
92 			rlen, de->name_len);
93 	return error_msg == NULL ? 1 : 0;
94 }
95 
96 static int ext4_readdir(struct file *filp,
97 			 void *dirent, filldir_t filldir)
98 {
99 	int error = 0;
100 	unsigned int offset;
101 	int i, stored;
102 	struct ext4_dir_entry_2 *de;
103 	struct super_block *sb;
104 	int err;
105 	struct inode *inode = filp->f_path.dentry->d_inode;
106 	int ret = 0;
107 	int dir_has_error = 0;
108 
109 	sb = inode->i_sb;
110 
111 	if (EXT4_HAS_COMPAT_FEATURE(inode->i_sb,
112 				    EXT4_FEATURE_COMPAT_DIR_INDEX) &&
113 	    ((ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) ||
114 	     ((inode->i_size >> sb->s_blocksize_bits) == 1))) {
115 		err = ext4_dx_readdir(filp, dirent, filldir);
116 		if (err != ERR_BAD_DX_DIR) {
117 			ret = err;
118 			goto out;
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(filp->f_path.dentry->d_inode, EXT4_INODE_INDEX);
125 	}
126 	stored = 0;
127 	offset = filp->f_pos & (sb->s_blocksize - 1);
128 
129 	while (!error && !stored && filp->f_pos < inode->i_size) {
130 		struct ext4_map_blocks map;
131 		struct buffer_head *bh = NULL;
132 
133 		map.m_lblk = filp->f_pos >> EXT4_BLOCK_SIZE_BITS(sb);
134 		map.m_len = 1;
135 		err = ext4_map_blocks(NULL, inode, &map, 0);
136 		if (err > 0) {
137 			pgoff_t index = map.m_pblk >>
138 					(PAGE_CACHE_SHIFT - inode->i_blkbits);
139 			if (!ra_has_index(&filp->f_ra, index))
140 				page_cache_sync_readahead(
141 					sb->s_bdev->bd_inode->i_mapping,
142 					&filp->f_ra, filp,
143 					index, 1);
144 			filp->f_ra.prev_pos = (loff_t)index << PAGE_CACHE_SHIFT;
145 			bh = ext4_bread(NULL, inode, map.m_lblk, 0, &err);
146 		}
147 
148 		/*
149 		 * We ignore I/O errors on directories so users have a chance
150 		 * of recovering data when there's a bad sector
151 		 */
152 		if (!bh) {
153 			if (!dir_has_error) {
154 				EXT4_ERROR_INODE(inode, "directory "
155 					   "contains a hole at offset %Lu",
156 					   (unsigned long long) filp->f_pos);
157 				dir_has_error = 1;
158 			}
159 			/* corrupt size?  Maybe no more blocks to read */
160 			if (filp->f_pos > inode->i_blocks << 9)
161 				break;
162 			filp->f_pos += sb->s_blocksize - offset;
163 			continue;
164 		}
165 
166 revalidate:
167 		/* If the dir block has changed since the last call to
168 		 * readdir(2), then we might be pointing to an invalid
169 		 * dirent right now.  Scan from the start of the block
170 		 * to make sure. */
171 		if (filp->f_version != inode->i_version) {
172 			for (i = 0; i < sb->s_blocksize && i < offset; ) {
173 				de = (struct ext4_dir_entry_2 *)
174 					(bh->b_data + i);
175 				/* It's too expensive to do a full
176 				 * dirent test each time round this
177 				 * loop, but we do have to test at
178 				 * least that it is non-zero.  A
179 				 * failure will be detected in the
180 				 * dirent test below. */
181 				if (ext4_rec_len_from_disk(de->rec_len,
182 					sb->s_blocksize) < EXT4_DIR_REC_LEN(1))
183 					break;
184 				i += ext4_rec_len_from_disk(de->rec_len,
185 							    sb->s_blocksize);
186 			}
187 			offset = i;
188 			filp->f_pos = (filp->f_pos & ~(sb->s_blocksize - 1))
189 				| offset;
190 			filp->f_version = inode->i_version;
191 		}
192 
193 		while (!error && filp->f_pos < inode->i_size
194 		       && offset < sb->s_blocksize) {
195 			de = (struct ext4_dir_entry_2 *) (bh->b_data + offset);
196 			if (!ext4_check_dir_entry("ext4_readdir", inode, de,
197 						  bh, offset)) {
198 				/*
199 				 * On error, skip the f_pos to the next block
200 				 */
201 				filp->f_pos = (filp->f_pos |
202 						(sb->s_blocksize - 1)) + 1;
203 				brelse(bh);
204 				ret = stored;
205 				goto out;
206 			}
207 			offset += ext4_rec_len_from_disk(de->rec_len,
208 					sb->s_blocksize);
209 			if (le32_to_cpu(de->inode)) {
210 				/* We might block in the next section
211 				 * if the data destination is
212 				 * currently swapped out.  So, use a
213 				 * version stamp to detect whether or
214 				 * not the directory has been modified
215 				 * during the copy operation.
216 				 */
217 				u64 version = filp->f_version;
218 
219 				error = filldir(dirent, de->name,
220 						de->name_len,
221 						filp->f_pos,
222 						le32_to_cpu(de->inode),
223 						get_dtype(sb, de->file_type));
224 				if (error)
225 					break;
226 				if (version != filp->f_version)
227 					goto revalidate;
228 				stored++;
229 			}
230 			filp->f_pos += ext4_rec_len_from_disk(de->rec_len,
231 						sb->s_blocksize);
232 		}
233 		offset = 0;
234 		brelse(bh);
235 	}
236 out:
237 	return ret;
238 }
239 
240 /*
241  * These functions convert from the major/minor hash to an f_pos
242  * value.
243  *
244  * Currently we only use major hash numer.  This is unfortunate, but
245  * on 32-bit machines, the same VFS interface is used for lseek and
246  * llseek, so if we use the 64 bit offset, then the 32-bit versions of
247  * lseek/telldir/seekdir will blow out spectacularly, and from within
248  * the ext2 low-level routine, we don't know if we're being called by
249  * a 64-bit version of the system call or the 32-bit version of the
250  * system call.  Worse yet, NFSv2 only allows for a 32-bit readdir
251  * cookie.  Sigh.
252  */
253 #define hash2pos(major, minor)	(major >> 1)
254 #define pos2maj_hash(pos)	((pos << 1) & 0xffffffff)
255 #define pos2min_hash(pos)	(0)
256 
257 /*
258  * This structure holds the nodes of the red-black tree used to store
259  * the directory entry in hash order.
260  */
261 struct fname {
262 	__u32		hash;
263 	__u32		minor_hash;
264 	struct rb_node	rb_hash;
265 	struct fname	*next;
266 	__u32		inode;
267 	__u8		name_len;
268 	__u8		file_type;
269 	char		name[0];
270 };
271 
272 /*
273  * This functoin implements a non-recursive way of freeing all of the
274  * nodes in the red-black tree.
275  */
276 static void free_rb_tree_fname(struct rb_root *root)
277 {
278 	struct rb_node	*n = root->rb_node;
279 	struct rb_node	*parent;
280 	struct fname	*fname;
281 
282 	while (n) {
283 		/* Do the node's children first */
284 		if (n->rb_left) {
285 			n = n->rb_left;
286 			continue;
287 		}
288 		if (n->rb_right) {
289 			n = n->rb_right;
290 			continue;
291 		}
292 		/*
293 		 * The node has no children; free it, and then zero
294 		 * out parent's link to it.  Finally go to the
295 		 * beginning of the loop and try to free the parent
296 		 * node.
297 		 */
298 		parent = rb_parent(n);
299 		fname = rb_entry(n, struct fname, rb_hash);
300 		while (fname) {
301 			struct fname *old = fname;
302 			fname = fname->next;
303 			kfree(old);
304 		}
305 		if (!parent)
306 			*root = RB_ROOT;
307 		else if (parent->rb_left == n)
308 			parent->rb_left = NULL;
309 		else if (parent->rb_right == n)
310 			parent->rb_right = NULL;
311 		n = parent;
312 	}
313 }
314 
315 
316 static struct dir_private_info *ext4_htree_create_dir_info(loff_t pos)
317 {
318 	struct dir_private_info *p;
319 
320 	p = kzalloc(sizeof(struct dir_private_info), GFP_KERNEL);
321 	if (!p)
322 		return NULL;
323 	p->curr_hash = pos2maj_hash(pos);
324 	p->curr_minor_hash = pos2min_hash(pos);
325 	return p;
326 }
327 
328 void ext4_htree_free_dir_info(struct dir_private_info *p)
329 {
330 	free_rb_tree_fname(&p->root);
331 	kfree(p);
332 }
333 
334 /*
335  * Given a directory entry, enter it into the fname rb tree.
336  */
337 int ext4_htree_store_dirent(struct file *dir_file, __u32 hash,
338 			     __u32 minor_hash,
339 			     struct ext4_dir_entry_2 *dirent)
340 {
341 	struct rb_node **p, *parent = NULL;
342 	struct fname *fname, *new_fn;
343 	struct dir_private_info *info;
344 	int len;
345 
346 	info = (struct dir_private_info *) dir_file->private_data;
347 	p = &info->root.rb_node;
348 
349 	/* Create and allocate the fname structure */
350 	len = sizeof(struct fname) + dirent->name_len + 1;
351 	new_fn = kzalloc(len, GFP_KERNEL);
352 	if (!new_fn)
353 		return -ENOMEM;
354 	new_fn->hash = hash;
355 	new_fn->minor_hash = minor_hash;
356 	new_fn->inode = le32_to_cpu(dirent->inode);
357 	new_fn->name_len = dirent->name_len;
358 	new_fn->file_type = dirent->file_type;
359 	memcpy(new_fn->name, dirent->name, dirent->name_len);
360 	new_fn->name[dirent->name_len] = 0;
361 
362 	while (*p) {
363 		parent = *p;
364 		fname = rb_entry(parent, struct fname, rb_hash);
365 
366 		/*
367 		 * If the hash and minor hash match up, then we put
368 		 * them on a linked list.  This rarely happens...
369 		 */
370 		if ((new_fn->hash == fname->hash) &&
371 		    (new_fn->minor_hash == fname->minor_hash)) {
372 			new_fn->next = fname->next;
373 			fname->next = new_fn;
374 			return 0;
375 		}
376 
377 		if (new_fn->hash < fname->hash)
378 			p = &(*p)->rb_left;
379 		else if (new_fn->hash > fname->hash)
380 			p = &(*p)->rb_right;
381 		else if (new_fn->minor_hash < fname->minor_hash)
382 			p = &(*p)->rb_left;
383 		else /* if (new_fn->minor_hash > fname->minor_hash) */
384 			p = &(*p)->rb_right;
385 	}
386 
387 	rb_link_node(&new_fn->rb_hash, parent, p);
388 	rb_insert_color(&new_fn->rb_hash, &info->root);
389 	return 0;
390 }
391 
392 
393 
394 /*
395  * This is a helper function for ext4_dx_readdir.  It calls filldir
396  * for all entres on the fname linked list.  (Normally there is only
397  * one entry on the linked list, unless there are 62 bit hash collisions.)
398  */
399 static int call_filldir(struct file *filp, void *dirent,
400 			filldir_t filldir, struct fname *fname)
401 {
402 	struct dir_private_info *info = filp->private_data;
403 	loff_t	curr_pos;
404 	struct inode *inode = filp->f_path.dentry->d_inode;
405 	struct super_block *sb;
406 	int error;
407 
408 	sb = inode->i_sb;
409 
410 	if (!fname) {
411 		printk(KERN_ERR "EXT4-fs: call_filldir: called with "
412 		       "null fname?!?\n");
413 		return 0;
414 	}
415 	curr_pos = hash2pos(fname->hash, fname->minor_hash);
416 	while (fname) {
417 		error = filldir(dirent, fname->name,
418 				fname->name_len, curr_pos,
419 				fname->inode,
420 				get_dtype(sb, fname->file_type));
421 		if (error) {
422 			filp->f_pos = curr_pos;
423 			info->extra_fname = fname;
424 			return error;
425 		}
426 		fname = fname->next;
427 	}
428 	return 0;
429 }
430 
431 static int ext4_dx_readdir(struct file *filp,
432 			 void *dirent, filldir_t filldir)
433 {
434 	struct dir_private_info *info = filp->private_data;
435 	struct inode *inode = filp->f_path.dentry->d_inode;
436 	struct fname *fname;
437 	int	ret;
438 
439 	if (!info) {
440 		info = ext4_htree_create_dir_info(filp->f_pos);
441 		if (!info)
442 			return -ENOMEM;
443 		filp->private_data = info;
444 	}
445 
446 	if (filp->f_pos == EXT4_HTREE_EOF)
447 		return 0;	/* EOF */
448 
449 	/* Some one has messed with f_pos; reset the world */
450 	if (info->last_pos != filp->f_pos) {
451 		free_rb_tree_fname(&info->root);
452 		info->curr_node = NULL;
453 		info->extra_fname = NULL;
454 		info->curr_hash = pos2maj_hash(filp->f_pos);
455 		info->curr_minor_hash = pos2min_hash(filp->f_pos);
456 	}
457 
458 	/*
459 	 * If there are any leftover names on the hash collision
460 	 * chain, return them first.
461 	 */
462 	if (info->extra_fname) {
463 		if (call_filldir(filp, dirent, filldir, info->extra_fname))
464 			goto finished;
465 		info->extra_fname = NULL;
466 		goto next_node;
467 	} else if (!info->curr_node)
468 		info->curr_node = rb_first(&info->root);
469 
470 	while (1) {
471 		/*
472 		 * Fill the rbtree if we have no more entries,
473 		 * or the inode has changed since we last read in the
474 		 * cached entries.
475 		 */
476 		if ((!info->curr_node) ||
477 		    (filp->f_version != inode->i_version)) {
478 			info->curr_node = NULL;
479 			free_rb_tree_fname(&info->root);
480 			filp->f_version = inode->i_version;
481 			ret = ext4_htree_fill_tree(filp, info->curr_hash,
482 						   info->curr_minor_hash,
483 						   &info->next_hash);
484 			if (ret < 0)
485 				return ret;
486 			if (ret == 0) {
487 				filp->f_pos = EXT4_HTREE_EOF;
488 				break;
489 			}
490 			info->curr_node = rb_first(&info->root);
491 		}
492 
493 		fname = rb_entry(info->curr_node, struct fname, rb_hash);
494 		info->curr_hash = fname->hash;
495 		info->curr_minor_hash = fname->minor_hash;
496 		if (call_filldir(filp, dirent, filldir, fname))
497 			break;
498 	next_node:
499 		info->curr_node = rb_next(info->curr_node);
500 		if (info->curr_node) {
501 			fname = rb_entry(info->curr_node, struct fname,
502 					 rb_hash);
503 			info->curr_hash = fname->hash;
504 			info->curr_minor_hash = fname->minor_hash;
505 		} else {
506 			if (info->next_hash == ~0) {
507 				filp->f_pos = EXT4_HTREE_EOF;
508 				break;
509 			}
510 			info->curr_hash = info->next_hash;
511 			info->curr_minor_hash = 0;
512 		}
513 	}
514 finished:
515 	info->last_pos = filp->f_pos;
516 	return 0;
517 }
518 
519 static int ext4_release_dir(struct inode *inode, struct file *filp)
520 {
521 	if (filp->private_data)
522 		ext4_htree_free_dir_info(filp->private_data);
523 
524 	return 0;
525 }
526