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)
11 * linux/fs/minix/dir.c
13 * Copyright (C) 1991, 1992 Linus Torvalds
15 * ext4 directory handling functions
17 * Big-endian to little-endian byte-swapping/bitmaps by
18 * David S. Miller (davem@caip.rutgers.edu), 1995
20 * Hash Tree Directory indexing (c) 2001 Daniel Phillips
25 #include <linux/jbd2.h>
26 #include <linux/buffer_head.h>
27 #include <linux/slab.h>
28 #include <linux/rbtree.h>
31 static unsigned char ext4_filetype_table[] = {
32 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
35 static int ext4_dx_readdir(struct file *filp,
36 void *dirent, filldir_t filldir);
38 static unsigned char get_dtype(struct super_block *sb, int filetype)
40 if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE) ||
41 (filetype >= EXT4_FT_MAX))
44 return (ext4_filetype_table[filetype]);
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
52 * Return 1 if it is a dx dir, 0 if not
54 static int is_dx_dir(struct inode *inode)
56 struct super_block *sb = inode->i_sb;
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)))
68 * Return 0 if the directory entry is OK, and 1 if there is a problem
70 * Note: this is the opposite of what ext2 and ext3 historically returned...
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, char *buf, int size,
78 const char *error_msg = NULL;
79 const int rlen = ext4_rec_len_from_disk(de->rec_len,
80 dir->i_sb->s_blocksize);
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 - buf) + rlen > size))
89 error_msg = "directory entry across range";
90 else if (unlikely(le32_to_cpu(de->inode) >
91 le32_to_cpu(EXT4_SB(dir->i_sb)->s_es->s_inodes_count)))
92 error_msg = "inode out of bounds";
97 ext4_error_file(filp, function, line, bh->b_blocknr,
98 "bad entry in directory: %s - offset=%u(%u), "
99 "inode=%u, rec_len=%d, name_len=%d",
100 error_msg, (unsigned) (offset % size),
101 offset, le32_to_cpu(de->inode),
104 ext4_error_inode(dir, function, line, bh->b_blocknr,
105 "bad entry in directory: %s - offset=%u(%u), "
106 "inode=%u, rec_len=%d, name_len=%d",
107 error_msg, (unsigned) (offset % size),
108 offset, le32_to_cpu(de->inode),
114 static int ext4_readdir(struct file *filp,
115 void *dirent, filldir_t filldir)
120 struct ext4_dir_entry_2 *de;
122 struct inode *inode = filp->f_path.dentry->d_inode;
123 struct super_block *sb = inode->i_sb;
125 int dir_has_error = 0;
127 if (is_dx_dir(inode)) {
128 err = ext4_dx_readdir(filp, dirent, filldir);
129 if (err != ERR_BAD_DX_DIR) {
134 * We don't set the inode dirty flag since it's not
135 * critical that it get flushed back to the disk.
137 ext4_clear_inode_flag(filp->f_path.dentry->d_inode,
141 offset = filp->f_pos & (sb->s_blocksize - 1);
143 while (!error && !stored && filp->f_pos < inode->i_size) {
144 struct ext4_map_blocks map;
145 struct buffer_head *bh = NULL;
147 map.m_lblk = filp->f_pos >> EXT4_BLOCK_SIZE_BITS(sb);
149 err = ext4_map_blocks(NULL, inode, &map, 0);
151 pgoff_t index = map.m_pblk >>
152 (PAGE_CACHE_SHIFT - inode->i_blkbits);
153 if (!ra_has_index(&filp->f_ra, index))
154 page_cache_sync_readahead(
155 sb->s_bdev->bd_inode->i_mapping,
158 filp->f_ra.prev_pos = (loff_t)index << PAGE_CACHE_SHIFT;
159 bh = ext4_bread(NULL, inode, map.m_lblk, 0, &err);
163 * We ignore I/O errors on directories so users have a chance
164 * of recovering data when there's a bad sector
167 if (!dir_has_error) {
168 EXT4_ERROR_FILE(filp, 0,
169 "directory contains a "
170 "hole at offset %llu",
171 (unsigned long long) filp->f_pos);
174 /* corrupt size? Maybe no more blocks to read */
175 if (filp->f_pos > inode->i_blocks << 9)
177 filp->f_pos += sb->s_blocksize - offset;
181 /* Check the checksum */
182 if (!buffer_verified(bh) &&
183 !ext4_dirent_csum_verify(inode,
184 (struct ext4_dir_entry *)bh->b_data)) {
185 EXT4_ERROR_FILE(filp, 0, "directory fails checksum "
187 (unsigned long long)filp->f_pos);
188 filp->f_pos += sb->s_blocksize - offset;
191 set_buffer_verified(bh);
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
198 if (filp->f_version != inode->i_version) {
199 for (i = 0; i < sb->s_blocksize && i < offset; ) {
200 de = (struct ext4_dir_entry_2 *)
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))
211 i += ext4_rec_len_from_disk(de->rec_len,
215 filp->f_pos = (filp->f_pos & ~(sb->s_blocksize - 1))
217 filp->f_version = inode->i_version;
220 while (!error && filp->f_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, filp, de, bh,
224 bh->b_data, bh->b_size,
227 * On error, skip the f_pos to the next block
229 filp->f_pos = (filp->f_pos |
230 (sb->s_blocksize - 1)) + 1;
235 offset += ext4_rec_len_from_disk(de->rec_len,
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.
245 u64 version = filp->f_version;
247 error = filldir(dirent, de->name,
250 le32_to_cpu(de->inode),
251 get_dtype(sb, de->file_type));
254 if (version != filp->f_version)
258 filp->f_pos += ext4_rec_len_from_disk(de->rec_len,
268 static inline int is_32bit_api(void)
271 return is_compat_task();
273 return (BITS_PER_LONG == 32);
278 * These functions convert from the major/minor hash to an f_pos
279 * value for dx directories
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.
286 static inline loff_t hash2pos(struct file *filp, __u32 major, __u32 minor)
288 if ((filp->f_mode & FMODE_32BITHASH) ||
289 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
292 return ((__u64)(major >> 1) << 32) | (__u64)minor;
295 static inline __u32 pos2maj_hash(struct file *filp, loff_t pos)
297 if ((filp->f_mode & FMODE_32BITHASH) ||
298 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
299 return (pos << 1) & 0xffffffff;
301 return ((pos >> 32) << 1) & 0xffffffff;
304 static inline __u32 pos2min_hash(struct file *filp, loff_t pos)
306 if ((filp->f_mode & FMODE_32BITHASH) ||
307 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
310 return pos & 0xffffffff;
314 * Return 32- or 64-bit end-of-file for dx directories
316 static inline loff_t ext4_get_htree_eof(struct file *filp)
318 if ((filp->f_mode & FMODE_32BITHASH) ||
319 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
320 return EXT4_HTREE_EOF_32BIT;
322 return EXT4_HTREE_EOF_64BIT;
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.
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.
335 * For non-htree, ext4_llseek already chooses the proper max offset.
337 loff_t ext4_dir_llseek(struct file *file, loff_t offset, int origin)
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);
344 return generic_file_llseek_size(file, offset, origin,
345 htree_max, htree_max);
347 return ext4_llseek(file, offset, origin);
351 * This structure holds the nodes of the red-black tree used to store
352 * the directory entry in hash order.
357 struct rb_node rb_hash;
366 * This functoin implements a non-recursive way of freeing all of the
367 * nodes in the red-black tree.
369 static void free_rb_tree_fname(struct rb_root *root)
371 struct rb_node *n = root->rb_node;
372 struct rb_node *parent;
376 /* Do the node's children first */
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
391 parent = rb_parent(n);
392 fname = rb_entry(n, struct fname, rb_hash);
394 struct fname *old = fname;
400 else if (parent->rb_left == n)
401 parent->rb_left = NULL;
402 else if (parent->rb_right == n)
403 parent->rb_right = NULL;
409 static struct dir_private_info *ext4_htree_create_dir_info(struct file *filp,
412 struct dir_private_info *p;
414 p = kzalloc(sizeof(struct dir_private_info), GFP_KERNEL);
417 p->curr_hash = pos2maj_hash(filp, pos);
418 p->curr_minor_hash = pos2min_hash(filp, pos);
422 void ext4_htree_free_dir_info(struct dir_private_info *p)
424 free_rb_tree_fname(&p->root);
429 * Given a directory entry, enter it into the fname rb tree.
431 int ext4_htree_store_dirent(struct file *dir_file, __u32 hash,
433 struct ext4_dir_entry_2 *dirent)
435 struct rb_node **p, *parent = NULL;
436 struct fname *fname, *new_fn;
437 struct dir_private_info *info;
440 info = dir_file->private_data;
441 p = &info->root.rb_node;
443 /* Create and allocate the fname structure */
444 len = sizeof(struct fname) + dirent->name_len + 1;
445 new_fn = kzalloc(len, GFP_KERNEL);
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;
458 fname = rb_entry(parent, struct fname, rb_hash);
461 * If the hash and minor hash match up, then we put
462 * them on a linked list. This rarely happens...
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;
471 if (new_fn->hash < fname->hash)
473 else if (new_fn->hash > fname->hash)
475 else if (new_fn->minor_hash < fname->minor_hash)
477 else /* if (new_fn->minor_hash > fname->minor_hash) */
481 rb_link_node(&new_fn->rb_hash, parent, p);
482 rb_insert_color(&new_fn->rb_hash, &info->root);
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.)
493 static int call_filldir(struct file *filp, void *dirent,
494 filldir_t filldir, struct fname *fname)
496 struct dir_private_info *info = filp->private_data;
498 struct inode *inode = filp->f_path.dentry->d_inode;
499 struct super_block *sb;
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);
510 curr_pos = hash2pos(filp, fname->hash, fname->minor_hash);
512 error = filldir(dirent, fname->name,
513 fname->name_len, curr_pos,
515 get_dtype(sb, fname->file_type));
517 filp->f_pos = curr_pos;
518 info->extra_fname = fname;
526 static int ext4_dx_readdir(struct file *filp,
527 void *dirent, filldir_t filldir)
529 struct dir_private_info *info = filp->private_data;
530 struct inode *inode = filp->f_path.dentry->d_inode;
535 info = ext4_htree_create_dir_info(filp, filp->f_pos);
538 filp->private_data = info;
541 if (filp->f_pos == ext4_get_htree_eof(filp))
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);
554 * If there are any leftover names on the hash collision
555 * chain, return them first.
557 if (info->extra_fname) {
558 if (call_filldir(filp, dirent, filldir, info->extra_fname))
560 info->extra_fname = NULL;
562 } else if (!info->curr_node)
563 info->curr_node = rb_first(&info->root);
567 * Fill the rbtree if we have no more entries,
568 * or the inode has changed since we last read in the
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,
582 filp->f_pos = ext4_get_htree_eof(filp);
585 info->curr_node = rb_first(&info->root);
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))
594 info->curr_node = rb_next(info->curr_node);
595 if (info->curr_node) {
596 fname = rb_entry(info->curr_node, struct fname,
598 info->curr_hash = fname->hash;
599 info->curr_minor_hash = fname->minor_hash;
601 if (info->next_hash == ~0) {
602 filp->f_pos = ext4_get_htree_eof(filp);
605 info->curr_hash = info->next_hash;
606 info->curr_minor_hash = 0;
610 info->last_pos = filp->f_pos;
614 static int ext4_release_dir(struct inode *inode, struct file *filp)
616 if (filp->private_data)
617 ext4_htree_free_dir_info(filp->private_data);
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,
628 .compat_ioctl = ext4_compat_ioctl,
630 .fsync = ext4_sync_file,
631 .release = ext4_release_dir,