xref: /openbmc/linux/fs/ext4/dir.c (revision 4161b450)
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 *, struct dir_context *);
33 
34 /**
35  * Check if the given dir-inode refers to an htree-indexed directory
36  * (or a directory which could potentially get converted to use htree
37  * indexing).
38  *
39  * Return 1 if it is a dx dir, 0 if not
40  */
41 static int is_dx_dir(struct inode *inode)
42 {
43 	struct super_block *sb = inode->i_sb;
44 
45 	if (EXT4_HAS_COMPAT_FEATURE(inode->i_sb,
46 		     EXT4_FEATURE_COMPAT_DIR_INDEX) &&
47 	    ((ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) ||
48 	     ((inode->i_size >> sb->s_blocksize_bits) == 1) ||
49 	     ext4_has_inline_data(inode)))
50 		return 1;
51 
52 	return 0;
53 }
54 
55 /*
56  * Return 0 if the directory entry is OK, and 1 if there is a problem
57  *
58  * Note: this is the opposite of what ext2 and ext3 historically returned...
59  *
60  * bh passed here can be an inode block or a dir data block, depending
61  * on the inode inline data flag.
62  */
63 int __ext4_check_dir_entry(const char *function, unsigned int line,
64 			   struct inode *dir, struct file *filp,
65 			   struct ext4_dir_entry_2 *de,
66 			   struct buffer_head *bh, char *buf, int size,
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 (unlikely(rlen < EXT4_DIR_REC_LEN(1)))
74 		error_msg = "rec_len is smaller than minimal";
75 	else if (unlikely(rlen % 4 != 0))
76 		error_msg = "rec_len % 4 != 0";
77 	else if (unlikely(rlen < EXT4_DIR_REC_LEN(de->name_len)))
78 		error_msg = "rec_len is too small for name_len";
79 	else if (unlikely(((char *) de - buf) + rlen > size))
80 		error_msg = "directory entry across range";
81 	else if (unlikely(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 	else
85 		return 0;
86 
87 	if (filp)
88 		ext4_error_file(filp, function, line, bh->b_blocknr,
89 				"bad entry in directory: %s - offset=%u(%u), "
90 				"inode=%u, rec_len=%d, name_len=%d",
91 				error_msg, (unsigned) (offset % size),
92 				offset, le32_to_cpu(de->inode),
93 				rlen, de->name_len);
94 	else
95 		ext4_error_inode(dir, function, line, bh->b_blocknr,
96 				"bad entry in directory: %s - offset=%u(%u), "
97 				"inode=%u, rec_len=%d, name_len=%d",
98 				error_msg, (unsigned) (offset % size),
99 				offset, le32_to_cpu(de->inode),
100 				rlen, de->name_len);
101 
102 	return 1;
103 }
104 
105 static int ext4_readdir(struct file *file, struct dir_context *ctx)
106 {
107 	unsigned int offset;
108 	int i;
109 	struct ext4_dir_entry_2 *de;
110 	int err;
111 	struct inode *inode = file_inode(file);
112 	struct super_block *sb = inode->i_sb;
113 	int dir_has_error = 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 		int ret = ext4_read_inline_dir(file, ctx,
131 					   &has_inline_data);
132 		if (has_inline_data)
133 			return ret;
134 	}
135 
136 	offset = ctx->pos & (sb->s_blocksize - 1);
137 
138 	while (ctx->pos < inode->i_size) {
139 		struct ext4_map_blocks map;
140 		struct buffer_head *bh = NULL;
141 
142 		map.m_lblk = ctx->pos >> EXT4_BLOCK_SIZE_BITS(sb);
143 		map.m_len = 1;
144 		err = ext4_map_blocks(NULL, inode, &map, 0);
145 		if (err > 0) {
146 			pgoff_t index = map.m_pblk >>
147 					(PAGE_CACHE_SHIFT - inode->i_blkbits);
148 			if (!ra_has_index(&file->f_ra, index))
149 				page_cache_sync_readahead(
150 					sb->s_bdev->bd_inode->i_mapping,
151 					&file->f_ra, file,
152 					index, 1);
153 			file->f_ra.prev_pos = (loff_t)index << PAGE_CACHE_SHIFT;
154 			bh = ext4_bread(NULL, inode, map.m_lblk, 0);
155 			if (IS_ERR(bh))
156 				return PTR_ERR(bh);
157 		}
158 
159 		if (!bh) {
160 			if (!dir_has_error) {
161 				EXT4_ERROR_FILE(file, 0,
162 						"directory contains a "
163 						"hole at offset %llu",
164 					   (unsigned long long) ctx->pos);
165 				dir_has_error = 1;
166 			}
167 			/* corrupt size?  Maybe no more blocks to read */
168 			if (ctx->pos > inode->i_blocks << 9)
169 				break;
170 			ctx->pos += sb->s_blocksize - offset;
171 			continue;
172 		}
173 
174 		/* Check the checksum */
175 		if (!buffer_verified(bh) &&
176 		    !ext4_dirent_csum_verify(inode,
177 				(struct ext4_dir_entry *)bh->b_data)) {
178 			EXT4_ERROR_FILE(file, 0, "directory fails checksum "
179 					"at offset %llu",
180 					(unsigned long long)ctx->pos);
181 			ctx->pos += sb->s_blocksize - offset;
182 			brelse(bh);
183 			continue;
184 		}
185 		set_buffer_verified(bh);
186 
187 		/* If the dir block has changed since the last call to
188 		 * readdir(2), then we might be pointing to an invalid
189 		 * dirent right now.  Scan from the start of the block
190 		 * to make sure. */
191 		if (file->f_version != inode->i_version) {
192 			for (i = 0; i < sb->s_blocksize && i < offset; ) {
193 				de = (struct ext4_dir_entry_2 *)
194 					(bh->b_data + i);
195 				/* It's too expensive to do a full
196 				 * dirent test each time round this
197 				 * loop, but we do have to test at
198 				 * least that it is non-zero.  A
199 				 * failure will be detected in the
200 				 * dirent test below. */
201 				if (ext4_rec_len_from_disk(de->rec_len,
202 					sb->s_blocksize) < EXT4_DIR_REC_LEN(1))
203 					break;
204 				i += ext4_rec_len_from_disk(de->rec_len,
205 							    sb->s_blocksize);
206 			}
207 			offset = i;
208 			ctx->pos = (ctx->pos & ~(sb->s_blocksize - 1))
209 				| offset;
210 			file->f_version = inode->i_version;
211 		}
212 
213 		while (ctx->pos < inode->i_size
214 		       && offset < sb->s_blocksize) {
215 			de = (struct ext4_dir_entry_2 *) (bh->b_data + offset);
216 			if (ext4_check_dir_entry(inode, file, de, bh,
217 						 bh->b_data, bh->b_size,
218 						 offset)) {
219 				/*
220 				 * On error, skip to the next block
221 				 */
222 				ctx->pos = (ctx->pos |
223 						(sb->s_blocksize - 1)) + 1;
224 				break;
225 			}
226 			offset += ext4_rec_len_from_disk(de->rec_len,
227 					sb->s_blocksize);
228 			if (le32_to_cpu(de->inode)) {
229 				if (!dir_emit(ctx, de->name,
230 						de->name_len,
231 						le32_to_cpu(de->inode),
232 						get_dtype(sb, de->file_type))) {
233 					brelse(bh);
234 					return 0;
235 				}
236 			}
237 			ctx->pos += ext4_rec_len_from_disk(de->rec_len,
238 						sb->s_blocksize);
239 		}
240 		offset = 0;
241 		brelse(bh);
242 		if (ctx->pos < inode->i_size) {
243 			if (!dir_relax(inode))
244 				return 0;
245 		}
246 	}
247 	return 0;
248 }
249 
250 static inline int is_32bit_api(void)
251 {
252 #ifdef CONFIG_COMPAT
253 	return is_compat_task();
254 #else
255 	return (BITS_PER_LONG == 32);
256 #endif
257 }
258 
259 /*
260  * These functions convert from the major/minor hash to an f_pos
261  * value for dx directories
262  *
263  * Upper layer (for example NFS) should specify FMODE_32BITHASH or
264  * FMODE_64BITHASH explicitly. On the other hand, we allow ext4 to be mounted
265  * directly on both 32-bit and 64-bit nodes, under such case, neither
266  * FMODE_32BITHASH nor FMODE_64BITHASH is specified.
267  */
268 static inline loff_t hash2pos(struct file *filp, __u32 major, __u32 minor)
269 {
270 	if ((filp->f_mode & FMODE_32BITHASH) ||
271 	    (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
272 		return major >> 1;
273 	else
274 		return ((__u64)(major >> 1) << 32) | (__u64)minor;
275 }
276 
277 static inline __u32 pos2maj_hash(struct file *filp, loff_t pos)
278 {
279 	if ((filp->f_mode & FMODE_32BITHASH) ||
280 	    (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
281 		return (pos << 1) & 0xffffffff;
282 	else
283 		return ((pos >> 32) << 1) & 0xffffffff;
284 }
285 
286 static inline __u32 pos2min_hash(struct file *filp, loff_t pos)
287 {
288 	if ((filp->f_mode & FMODE_32BITHASH) ||
289 	    (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
290 		return 0;
291 	else
292 		return pos & 0xffffffff;
293 }
294 
295 /*
296  * Return 32- or 64-bit end-of-file for dx directories
297  */
298 static inline loff_t ext4_get_htree_eof(struct file *filp)
299 {
300 	if ((filp->f_mode & FMODE_32BITHASH) ||
301 	    (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
302 		return EXT4_HTREE_EOF_32BIT;
303 	else
304 		return EXT4_HTREE_EOF_64BIT;
305 }
306 
307 
308 /*
309  * ext4_dir_llseek() calls generic_file_llseek_size to handle htree
310  * directories, where the "offset" is in terms of the filename hash
311  * value instead of the byte offset.
312  *
313  * Because we may return a 64-bit hash that is well beyond offset limits,
314  * we need to pass the max hash as the maximum allowable offset in
315  * the htree directory case.
316  *
317  * For non-htree, ext4_llseek already chooses the proper max offset.
318  */
319 static loff_t ext4_dir_llseek(struct file *file, loff_t offset, int whence)
320 {
321 	struct inode *inode = file->f_mapping->host;
322 	int dx_dir = is_dx_dir(inode);
323 	loff_t htree_max = ext4_get_htree_eof(file);
324 
325 	if (likely(dx_dir))
326 		return generic_file_llseek_size(file, offset, whence,
327 						    htree_max, htree_max);
328 	else
329 		return ext4_llseek(file, offset, whence);
330 }
331 
332 /*
333  * This structure holds the nodes of the red-black tree used to store
334  * the directory entry in hash order.
335  */
336 struct fname {
337 	__u32		hash;
338 	__u32		minor_hash;
339 	struct rb_node	rb_hash;
340 	struct fname	*next;
341 	__u32		inode;
342 	__u8		name_len;
343 	__u8		file_type;
344 	char		name[0];
345 };
346 
347 /*
348  * This functoin implements a non-recursive way of freeing all of the
349  * nodes in the red-black tree.
350  */
351 static void free_rb_tree_fname(struct rb_root *root)
352 {
353 	struct fname *fname, *next;
354 
355 	rbtree_postorder_for_each_entry_safe(fname, next, root, rb_hash)
356 		while (fname) {
357 			struct fname *old = fname;
358 			fname = fname->next;
359 			kfree(old);
360 		}
361 
362 	*root = RB_ROOT;
363 }
364 
365 
366 static struct dir_private_info *ext4_htree_create_dir_info(struct file *filp,
367 							   loff_t pos)
368 {
369 	struct dir_private_info *p;
370 
371 	p = kzalloc(sizeof(struct dir_private_info), GFP_KERNEL);
372 	if (!p)
373 		return NULL;
374 	p->curr_hash = pos2maj_hash(filp, pos);
375 	p->curr_minor_hash = pos2min_hash(filp, pos);
376 	return p;
377 }
378 
379 void ext4_htree_free_dir_info(struct dir_private_info *p)
380 {
381 	free_rb_tree_fname(&p->root);
382 	kfree(p);
383 }
384 
385 /*
386  * Given a directory entry, enter it into the fname rb tree.
387  */
388 int ext4_htree_store_dirent(struct file *dir_file, __u32 hash,
389 			     __u32 minor_hash,
390 			     struct ext4_dir_entry_2 *dirent)
391 {
392 	struct rb_node **p, *parent = NULL;
393 	struct fname *fname, *new_fn;
394 	struct dir_private_info *info;
395 	int len;
396 
397 	info = dir_file->private_data;
398 	p = &info->root.rb_node;
399 
400 	/* Create and allocate the fname structure */
401 	len = sizeof(struct fname) + dirent->name_len + 1;
402 	new_fn = kzalloc(len, GFP_KERNEL);
403 	if (!new_fn)
404 		return -ENOMEM;
405 	new_fn->hash = hash;
406 	new_fn->minor_hash = minor_hash;
407 	new_fn->inode = le32_to_cpu(dirent->inode);
408 	new_fn->name_len = dirent->name_len;
409 	new_fn->file_type = dirent->file_type;
410 	memcpy(new_fn->name, dirent->name, dirent->name_len);
411 	new_fn->name[dirent->name_len] = 0;
412 
413 	while (*p) {
414 		parent = *p;
415 		fname = rb_entry(parent, struct fname, rb_hash);
416 
417 		/*
418 		 * If the hash and minor hash match up, then we put
419 		 * them on a linked list.  This rarely happens...
420 		 */
421 		if ((new_fn->hash == fname->hash) &&
422 		    (new_fn->minor_hash == fname->minor_hash)) {
423 			new_fn->next = fname->next;
424 			fname->next = new_fn;
425 			return 0;
426 		}
427 
428 		if (new_fn->hash < fname->hash)
429 			p = &(*p)->rb_left;
430 		else if (new_fn->hash > fname->hash)
431 			p = &(*p)->rb_right;
432 		else if (new_fn->minor_hash < fname->minor_hash)
433 			p = &(*p)->rb_left;
434 		else /* if (new_fn->minor_hash > fname->minor_hash) */
435 			p = &(*p)->rb_right;
436 	}
437 
438 	rb_link_node(&new_fn->rb_hash, parent, p);
439 	rb_insert_color(&new_fn->rb_hash, &info->root);
440 	return 0;
441 }
442 
443 
444 
445 /*
446  * This is a helper function for ext4_dx_readdir.  It calls filldir
447  * for all entres on the fname linked list.  (Normally there is only
448  * one entry on the linked list, unless there are 62 bit hash collisions.)
449  */
450 static int call_filldir(struct file *file, struct dir_context *ctx,
451 			struct fname *fname)
452 {
453 	struct dir_private_info *info = file->private_data;
454 	struct inode *inode = file_inode(file);
455 	struct super_block *sb = inode->i_sb;
456 
457 	if (!fname) {
458 		ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: comm %s: "
459 			 "called with null fname?!?", __func__, __LINE__,
460 			 inode->i_ino, current->comm);
461 		return 0;
462 	}
463 	ctx->pos = hash2pos(file, fname->hash, fname->minor_hash);
464 	while (fname) {
465 		if (!dir_emit(ctx, fname->name,
466 				fname->name_len,
467 				fname->inode,
468 				get_dtype(sb, fname->file_type))) {
469 			info->extra_fname = fname;
470 			return 1;
471 		}
472 		fname = fname->next;
473 	}
474 	return 0;
475 }
476 
477 static int ext4_dx_readdir(struct file *file, struct dir_context *ctx)
478 {
479 	struct dir_private_info *info = file->private_data;
480 	struct inode *inode = file_inode(file);
481 	struct fname *fname;
482 	int	ret;
483 
484 	if (!info) {
485 		info = ext4_htree_create_dir_info(file, ctx->pos);
486 		if (!info)
487 			return -ENOMEM;
488 		file->private_data = info;
489 	}
490 
491 	if (ctx->pos == ext4_get_htree_eof(file))
492 		return 0;	/* EOF */
493 
494 	/* Some one has messed with f_pos; reset the world */
495 	if (info->last_pos != ctx->pos) {
496 		free_rb_tree_fname(&info->root);
497 		info->curr_node = NULL;
498 		info->extra_fname = NULL;
499 		info->curr_hash = pos2maj_hash(file, ctx->pos);
500 		info->curr_minor_hash = pos2min_hash(file, ctx->pos);
501 	}
502 
503 	/*
504 	 * If there are any leftover names on the hash collision
505 	 * chain, return them first.
506 	 */
507 	if (info->extra_fname) {
508 		if (call_filldir(file, ctx, info->extra_fname))
509 			goto finished;
510 		info->extra_fname = NULL;
511 		goto next_node;
512 	} else if (!info->curr_node)
513 		info->curr_node = rb_first(&info->root);
514 
515 	while (1) {
516 		/*
517 		 * Fill the rbtree if we have no more entries,
518 		 * or the inode has changed since we last read in the
519 		 * cached entries.
520 		 */
521 		if ((!info->curr_node) ||
522 		    (file->f_version != inode->i_version)) {
523 			info->curr_node = NULL;
524 			free_rb_tree_fname(&info->root);
525 			file->f_version = inode->i_version;
526 			ret = ext4_htree_fill_tree(file, info->curr_hash,
527 						   info->curr_minor_hash,
528 						   &info->next_hash);
529 			if (ret < 0)
530 				return ret;
531 			if (ret == 0) {
532 				ctx->pos = ext4_get_htree_eof(file);
533 				break;
534 			}
535 			info->curr_node = rb_first(&info->root);
536 		}
537 
538 		fname = rb_entry(info->curr_node, struct fname, rb_hash);
539 		info->curr_hash = fname->hash;
540 		info->curr_minor_hash = fname->minor_hash;
541 		if (call_filldir(file, ctx, fname))
542 			break;
543 	next_node:
544 		info->curr_node = rb_next(info->curr_node);
545 		if (info->curr_node) {
546 			fname = rb_entry(info->curr_node, struct fname,
547 					 rb_hash);
548 			info->curr_hash = fname->hash;
549 			info->curr_minor_hash = fname->minor_hash;
550 		} else {
551 			if (info->next_hash == ~0) {
552 				ctx->pos = ext4_get_htree_eof(file);
553 				break;
554 			}
555 			info->curr_hash = info->next_hash;
556 			info->curr_minor_hash = 0;
557 		}
558 	}
559 finished:
560 	info->last_pos = ctx->pos;
561 	return 0;
562 }
563 
564 static int ext4_release_dir(struct inode *inode, struct file *filp)
565 {
566 	if (filp->private_data)
567 		ext4_htree_free_dir_info(filp->private_data);
568 
569 	return 0;
570 }
571 
572 int ext4_check_all_de(struct inode *dir, struct buffer_head *bh, void *buf,
573 		      int buf_size)
574 {
575 	struct ext4_dir_entry_2 *de;
576 	int nlen, rlen;
577 	unsigned int offset = 0;
578 	char *top;
579 
580 	de = (struct ext4_dir_entry_2 *)buf;
581 	top = buf + buf_size;
582 	while ((char *) de < top) {
583 		if (ext4_check_dir_entry(dir, NULL, de, bh,
584 					 buf, buf_size, offset))
585 			return -EIO;
586 		nlen = EXT4_DIR_REC_LEN(de->name_len);
587 		rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
588 		de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
589 		offset += rlen;
590 	}
591 	if ((char *) de > top)
592 		return -EIO;
593 
594 	return 0;
595 }
596 
597 const struct file_operations ext4_dir_operations = {
598 	.llseek		= ext4_dir_llseek,
599 	.read		= generic_read_dir,
600 	.iterate	= ext4_readdir,
601 	.unlocked_ioctl = ext4_ioctl,
602 #ifdef CONFIG_COMPAT
603 	.compat_ioctl	= ext4_compat_ioctl,
604 #endif
605 	.fsync		= ext4_sync_file,
606 	.release	= ext4_release_dir,
607 };
608