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/buffer_head.h>
26 #include <linux/slab.h>
30 static int ext4_dx_readdir(struct file *, struct dir_context *);
33 * Check if the given dir-inode refers to an htree-indexed directory
34 * (or a directory which could potentially get converted to use htree
37 * Return 1 if it is a dx dir, 0 if not
39 static int is_dx_dir(struct inode *inode)
41 struct super_block *sb = inode->i_sb;
43 if (ext4_has_feature_dir_index(inode->i_sb) &&
44 ((ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) ||
45 ((inode->i_size >> sb->s_blocksize_bits) == 1) ||
46 ext4_has_inline_data(inode)))
53 * Return 0 if the directory entry is OK, and 1 if there is a problem
55 * Note: this is the opposite of what ext2 and ext3 historically returned...
57 * bh passed here can be an inode block or a dir data block, depending
58 * on the inode inline data flag.
60 int __ext4_check_dir_entry(const char *function, unsigned int line,
61 struct inode *dir, struct file *filp,
62 struct ext4_dir_entry_2 *de,
63 struct buffer_head *bh, char *buf, int size,
66 const char *error_msg = NULL;
67 const int rlen = ext4_rec_len_from_disk(de->rec_len,
68 dir->i_sb->s_blocksize);
70 if (unlikely(rlen < EXT4_DIR_REC_LEN(1)))
71 error_msg = "rec_len is smaller than minimal";
72 else if (unlikely(rlen % 4 != 0))
73 error_msg = "rec_len % 4 != 0";
74 else if (unlikely(rlen < EXT4_DIR_REC_LEN(de->name_len)))
75 error_msg = "rec_len is too small for name_len";
76 else if (unlikely(((char *) de - buf) + rlen > size))
77 error_msg = "directory entry overrun";
78 else if (unlikely(((char *) de - buf) + rlen >
79 size - EXT4_DIR_REC_LEN(1) &&
80 ((char *) de - buf) + rlen != size)) {
81 error_msg = "directory entry too close to block end";
83 else if (unlikely(le32_to_cpu(de->inode) >
84 le32_to_cpu(EXT4_SB(dir->i_sb)->s_es->s_inodes_count)))
85 error_msg = "inode out of bounds";
90 ext4_error_file(filp, function, line, bh->b_blocknr,
91 "bad entry in directory: %s - offset=%u, "
92 "inode=%u, rec_len=%d, name_len=%d, size=%d",
93 error_msg, offset, le32_to_cpu(de->inode),
94 rlen, de->name_len, size);
96 ext4_error_inode(dir, function, line, bh->b_blocknr,
97 "bad entry in directory: %s - offset=%u, "
98 "inode=%u, rec_len=%d, name_len=%d, size=%d",
99 error_msg, offset, le32_to_cpu(de->inode),
100 rlen, de->name_len, size);
105 static int ext4_readdir(struct file *file, struct dir_context *ctx)
109 struct ext4_dir_entry_2 *de;
111 struct inode *inode = file_inode(file);
112 struct super_block *sb = inode->i_sb;
113 struct buffer_head *bh = NULL;
114 struct fscrypt_str fstr = FSTR_INIT(NULL, 0);
116 if (ext4_encrypted_inode(inode)) {
117 err = fscrypt_get_encryption_info(inode);
118 if (err && err != -ENOKEY)
122 if (is_dx_dir(inode)) {
123 err = ext4_dx_readdir(file, ctx);
124 if (err != ERR_BAD_DX_DIR) {
127 /* Can we just clear INDEX flag to ignore htree information? */
128 if (!ext4_has_metadata_csum(sb)) {
130 * We don't set the inode dirty flag since it's not
131 * critical that it gets flushed back to the disk.
133 ext4_clear_inode_flag(inode, EXT4_INODE_INDEX);
137 if (ext4_has_inline_data(inode)) {
138 int has_inline_data = 1;
139 err = ext4_read_inline_dir(file, ctx,
145 if (ext4_encrypted_inode(inode)) {
146 err = fscrypt_fname_alloc_buffer(inode, EXT4_NAME_LEN, &fstr);
151 while (ctx->pos < inode->i_size) {
152 struct ext4_map_blocks map;
154 if (fatal_signal_pending(current)) {
159 offset = ctx->pos & (sb->s_blocksize - 1);
160 map.m_lblk = ctx->pos >> EXT4_BLOCK_SIZE_BITS(sb);
162 err = ext4_map_blocks(NULL, inode, &map, 0);
164 /* m_len should never be zero but let's avoid
165 * an infinite loop if it somehow is */
168 ctx->pos += map.m_len * sb->s_blocksize;
172 pgoff_t index = map.m_pblk >>
173 (PAGE_SHIFT - inode->i_blkbits);
174 if (!ra_has_index(&file->f_ra, index))
175 page_cache_sync_readahead(
176 sb->s_bdev->bd_inode->i_mapping,
179 file->f_ra.prev_pos = (loff_t)index << PAGE_SHIFT;
180 bh = ext4_bread(NULL, inode, map.m_lblk, 0);
189 /* corrupt size? Maybe no more blocks to read */
190 if (ctx->pos > inode->i_blocks << 9)
192 ctx->pos += sb->s_blocksize - offset;
196 /* Check the checksum */
197 if (!buffer_verified(bh) &&
198 !ext4_dirent_csum_verify(inode,
199 (struct ext4_dir_entry *)bh->b_data)) {
200 EXT4_ERROR_FILE(file, 0, "directory fails checksum "
202 (unsigned long long)ctx->pos);
203 ctx->pos += sb->s_blocksize - offset;
208 set_buffer_verified(bh);
210 /* If the dir block has changed since the last call to
211 * readdir(2), then we might be pointing to an invalid
212 * dirent right now. Scan from the start of the block
214 if (file->f_version != inode->i_version) {
215 for (i = 0; i < sb->s_blocksize && i < offset; ) {
216 de = (struct ext4_dir_entry_2 *)
218 /* It's too expensive to do a full
219 * dirent test each time round this
220 * loop, but we do have to test at
221 * least that it is non-zero. A
222 * failure will be detected in the
223 * dirent test below. */
224 if (ext4_rec_len_from_disk(de->rec_len,
225 sb->s_blocksize) < EXT4_DIR_REC_LEN(1))
227 i += ext4_rec_len_from_disk(de->rec_len,
231 ctx->pos = (ctx->pos & ~(sb->s_blocksize - 1))
233 file->f_version = inode->i_version;
236 while (ctx->pos < inode->i_size
237 && offset < sb->s_blocksize) {
238 de = (struct ext4_dir_entry_2 *) (bh->b_data + offset);
239 if (ext4_check_dir_entry(inode, file, de, bh,
240 bh->b_data, bh->b_size,
243 * On error, skip to the next block
245 ctx->pos = (ctx->pos |
246 (sb->s_blocksize - 1)) + 1;
249 offset += ext4_rec_len_from_disk(de->rec_len,
251 if (le32_to_cpu(de->inode)) {
252 if (!ext4_encrypted_inode(inode)) {
253 if (!dir_emit(ctx, de->name,
255 le32_to_cpu(de->inode),
256 get_dtype(sb, de->file_type)))
259 int save_len = fstr.len;
260 struct fscrypt_str de_name =
264 /* Directory is encrypted */
265 err = fscrypt_fname_disk_to_usr(inode,
266 0, 0, &de_name, &fstr);
272 de_name.name, de_name.len,
273 le32_to_cpu(de->inode),
274 get_dtype(sb, de->file_type)))
278 ctx->pos += ext4_rec_len_from_disk(de->rec_len,
281 if ((ctx->pos < inode->i_size) && !dir_relax_shared(inode))
290 #ifdef CONFIG_EXT4_FS_ENCRYPTION
291 fscrypt_fname_free_buffer(&fstr);
297 static inline int is_32bit_api(void)
300 return in_compat_syscall();
302 return (BITS_PER_LONG == 32);
307 * These functions convert from the major/minor hash to an f_pos
308 * value for dx directories
310 * Upper layer (for example NFS) should specify FMODE_32BITHASH or
311 * FMODE_64BITHASH explicitly. On the other hand, we allow ext4 to be mounted
312 * directly on both 32-bit and 64-bit nodes, under such case, neither
313 * FMODE_32BITHASH nor FMODE_64BITHASH is specified.
315 static inline loff_t hash2pos(struct file *filp, __u32 major, __u32 minor)
317 if ((filp->f_mode & FMODE_32BITHASH) ||
318 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
321 return ((__u64)(major >> 1) << 32) | (__u64)minor;
324 static inline __u32 pos2maj_hash(struct file *filp, loff_t pos)
326 if ((filp->f_mode & FMODE_32BITHASH) ||
327 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
328 return (pos << 1) & 0xffffffff;
330 return ((pos >> 32) << 1) & 0xffffffff;
333 static inline __u32 pos2min_hash(struct file *filp, loff_t pos)
335 if ((filp->f_mode & FMODE_32BITHASH) ||
336 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
339 return pos & 0xffffffff;
343 * Return 32- or 64-bit end-of-file for dx directories
345 static inline loff_t ext4_get_htree_eof(struct file *filp)
347 if ((filp->f_mode & FMODE_32BITHASH) ||
348 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
349 return EXT4_HTREE_EOF_32BIT;
351 return EXT4_HTREE_EOF_64BIT;
356 * ext4_dir_llseek() calls generic_file_llseek_size to handle htree
357 * directories, where the "offset" is in terms of the filename hash
358 * value instead of the byte offset.
360 * Because we may return a 64-bit hash that is well beyond offset limits,
361 * we need to pass the max hash as the maximum allowable offset in
362 * the htree directory case.
364 * For non-htree, ext4_llseek already chooses the proper max offset.
366 static loff_t ext4_dir_llseek(struct file *file, loff_t offset, int whence)
368 struct inode *inode = file->f_mapping->host;
369 int dx_dir = is_dx_dir(inode);
370 loff_t htree_max = ext4_get_htree_eof(file);
373 return generic_file_llseek_size(file, offset, whence,
374 htree_max, htree_max);
376 return ext4_llseek(file, offset, whence);
380 * This structure holds the nodes of the red-black tree used to store
381 * the directory entry in hash order.
386 struct rb_node rb_hash;
395 * This functoin implements a non-recursive way of freeing all of the
396 * nodes in the red-black tree.
398 static void free_rb_tree_fname(struct rb_root *root)
400 struct fname *fname, *next;
402 rbtree_postorder_for_each_entry_safe(fname, next, root, rb_hash)
404 struct fname *old = fname;
413 static struct dir_private_info *ext4_htree_create_dir_info(struct file *filp,
416 struct dir_private_info *p;
418 p = kzalloc(sizeof(struct dir_private_info), GFP_KERNEL);
421 p->curr_hash = pos2maj_hash(filp, pos);
422 p->curr_minor_hash = pos2min_hash(filp, pos);
426 void ext4_htree_free_dir_info(struct dir_private_info *p)
428 free_rb_tree_fname(&p->root);
433 * Given a directory entry, enter it into the fname rb tree.
435 * When filename encryption is enabled, the dirent will hold the
436 * encrypted filename, while the htree will hold decrypted filename.
437 * The decrypted filename is passed in via ent_name. parameter.
439 int ext4_htree_store_dirent(struct file *dir_file, __u32 hash,
441 struct ext4_dir_entry_2 *dirent,
442 struct fscrypt_str *ent_name)
444 struct rb_node **p, *parent = NULL;
445 struct fname *fname, *new_fn;
446 struct dir_private_info *info;
449 info = dir_file->private_data;
450 p = &info->root.rb_node;
452 /* Create and allocate the fname structure */
453 len = sizeof(struct fname) + ent_name->len + 1;
454 new_fn = kzalloc(len, GFP_KERNEL);
458 new_fn->minor_hash = minor_hash;
459 new_fn->inode = le32_to_cpu(dirent->inode);
460 new_fn->name_len = ent_name->len;
461 new_fn->file_type = dirent->file_type;
462 memcpy(new_fn->name, ent_name->name, ent_name->len);
463 new_fn->name[ent_name->len] = 0;
467 fname = rb_entry(parent, struct fname, rb_hash);
470 * If the hash and minor hash match up, then we put
471 * them on a linked list. This rarely happens...
473 if ((new_fn->hash == fname->hash) &&
474 (new_fn->minor_hash == fname->minor_hash)) {
475 new_fn->next = fname->next;
476 fname->next = new_fn;
480 if (new_fn->hash < fname->hash)
482 else if (new_fn->hash > fname->hash)
484 else if (new_fn->minor_hash < fname->minor_hash)
486 else /* if (new_fn->minor_hash > fname->minor_hash) */
490 rb_link_node(&new_fn->rb_hash, parent, p);
491 rb_insert_color(&new_fn->rb_hash, &info->root);
498 * This is a helper function for ext4_dx_readdir. It calls filldir
499 * for all entres on the fname linked list. (Normally there is only
500 * one entry on the linked list, unless there are 62 bit hash collisions.)
502 static int call_filldir(struct file *file, struct dir_context *ctx,
505 struct dir_private_info *info = file->private_data;
506 struct inode *inode = file_inode(file);
507 struct super_block *sb = inode->i_sb;
510 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: comm %s: "
511 "called with null fname?!?", __func__, __LINE__,
512 inode->i_ino, current->comm);
515 ctx->pos = hash2pos(file, fname->hash, fname->minor_hash);
517 if (!dir_emit(ctx, fname->name,
520 get_dtype(sb, fname->file_type))) {
521 info->extra_fname = fname;
529 static int ext4_dx_readdir(struct file *file, struct dir_context *ctx)
531 struct dir_private_info *info = file->private_data;
532 struct inode *inode = file_inode(file);
537 info = ext4_htree_create_dir_info(file, ctx->pos);
540 file->private_data = info;
543 if (ctx->pos == ext4_get_htree_eof(file))
546 /* Some one has messed with f_pos; reset the world */
547 if (info->last_pos != ctx->pos) {
548 free_rb_tree_fname(&info->root);
549 info->curr_node = NULL;
550 info->extra_fname = NULL;
551 info->curr_hash = pos2maj_hash(file, ctx->pos);
552 info->curr_minor_hash = pos2min_hash(file, ctx->pos);
556 * If there are any leftover names on the hash collision
557 * chain, return them first.
559 if (info->extra_fname) {
560 if (call_filldir(file, ctx, info->extra_fname))
562 info->extra_fname = NULL;
564 } else if (!info->curr_node)
565 info->curr_node = rb_first(&info->root);
569 * Fill the rbtree if we have no more entries,
570 * or the inode has changed since we last read in the
573 if ((!info->curr_node) ||
574 (file->f_version != inode->i_version)) {
575 info->curr_node = NULL;
576 free_rb_tree_fname(&info->root);
577 file->f_version = inode->i_version;
578 ret = ext4_htree_fill_tree(file, info->curr_hash,
579 info->curr_minor_hash,
584 ctx->pos = ext4_get_htree_eof(file);
587 info->curr_node = rb_first(&info->root);
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(file, ctx, fname))
596 info->curr_node = rb_next(info->curr_node);
597 if (info->curr_node) {
598 fname = rb_entry(info->curr_node, struct fname,
600 info->curr_hash = fname->hash;
601 info->curr_minor_hash = fname->minor_hash;
603 if (info->next_hash == ~0) {
604 ctx->pos = ext4_get_htree_eof(file);
607 info->curr_hash = info->next_hash;
608 info->curr_minor_hash = 0;
612 info->last_pos = ctx->pos;
613 return ret < 0 ? ret : 0;
616 static int ext4_dir_open(struct inode * inode, struct file * filp)
618 if (ext4_encrypted_inode(inode))
619 return fscrypt_get_encryption_info(inode) ? -EACCES : 0;
623 static int ext4_release_dir(struct inode *inode, struct file *filp)
625 if (filp->private_data)
626 ext4_htree_free_dir_info(filp->private_data);
631 int ext4_check_all_de(struct inode *dir, struct buffer_head *bh, void *buf,
634 struct ext4_dir_entry_2 *de;
636 unsigned int offset = 0;
639 de = (struct ext4_dir_entry_2 *)buf;
640 top = buf + buf_size;
641 while ((char *) de < top) {
642 if (ext4_check_dir_entry(dir, NULL, de, bh,
643 buf, buf_size, offset))
644 return -EFSCORRUPTED;
645 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
646 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
649 if ((char *) de > top)
650 return -EFSCORRUPTED;
655 const struct file_operations ext4_dir_operations = {
656 .llseek = ext4_dir_llseek,
657 .read = generic_read_dir,
658 .iterate_shared = ext4_readdir,
659 .unlocked_ioctl = ext4_ioctl,
661 .compat_ioctl = ext4_compat_ioctl,
663 .fsync = ext4_sync_file,
664 .open = ext4_dir_open,
665 .release = ext4_release_dir,