1 // SPDX-License-Identifier: GPL-2.0
5 * Copyright (C) 1992, 1993, 1994, 1995
7 * Laboratoire MASI - Institut Blaise Pascal
8 * Universite Pierre et Marie Curie (Paris VI)
12 * linux/fs/minix/dir.c
14 * Copyright (C) 1991, 1992 Linus Torvalds
16 * ext4 directory handling functions
18 * Big-endian to little-endian byte-swapping/bitmaps by
21 * Hash Tree Directory indexing (c) 2001 Daniel Phillips
26 #include <linux/buffer_head.h>
27 #include <linux/slab.h>
28 #include <linux/iversion.h>
29 #include <linux/unicode.h>
33 static int ext4_dx_readdir(struct file *, struct dir_context *);
36 * is_dx_dir() - check if a directory is using htree indexing
37 * @inode: directory inode
39 * Check if the given dir-inode refers to an htree-indexed directory
40 * (or a directory which could potentially get converted to use htree
43 * Return 1 if it is a dx dir, 0 if not
45 static int is_dx_dir(struct inode *inode)
47 struct super_block *sb = inode->i_sb;
49 if (ext4_has_feature_dir_index(inode->i_sb) &&
50 ((ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) ||
51 ((inode->i_size >> sb->s_blocksize_bits) == 1) ||
52 ext4_has_inline_data(inode)))
59 * Return 0 if the directory entry is OK, and 1 if there is a problem
61 * Note: this is the opposite of what ext2 and ext3 historically returned...
63 * bh passed here can be an inode block or a dir data block, depending
64 * on the inode inline data flag.
66 int __ext4_check_dir_entry(const char *function, unsigned int line,
67 struct inode *dir, struct file *filp,
68 struct ext4_dir_entry_2 *de,
69 struct buffer_head *bh, char *buf, int size,
72 const char *error_msg = NULL;
73 const int rlen = ext4_rec_len_from_disk(de->rec_len,
74 dir->i_sb->s_blocksize);
76 if (unlikely(rlen < EXT4_DIR_REC_LEN(1)))
77 error_msg = "rec_len is smaller than minimal";
78 else if (unlikely(rlen % 4 != 0))
79 error_msg = "rec_len % 4 != 0";
80 else if (unlikely(rlen < EXT4_DIR_REC_LEN(de->name_len)))
81 error_msg = "rec_len is too small for name_len";
82 else if (unlikely(((char *) de - buf) + rlen > size))
83 error_msg = "directory entry overrun";
84 else if (unlikely(le32_to_cpu(de->inode) >
85 le32_to_cpu(EXT4_SB(dir->i_sb)->s_es->s_inodes_count)))
86 error_msg = "inode out of bounds";
91 ext4_error_file(filp, function, line, bh->b_blocknr,
92 "bad entry in directory: %s - offset=%u, "
93 "inode=%u, rec_len=%d, name_len=%d, size=%d",
94 error_msg, offset, le32_to_cpu(de->inode),
95 rlen, de->name_len, size);
97 ext4_error_inode(dir, function, line, bh->b_blocknr,
98 "bad entry in directory: %s - offset=%u, "
99 "inode=%u, rec_len=%d, name_len=%d, size=%d",
100 error_msg, offset, le32_to_cpu(de->inode),
101 rlen, de->name_len, size);
106 static int ext4_readdir(struct file *file, struct dir_context *ctx)
110 struct ext4_dir_entry_2 *de;
112 struct inode *inode = file_inode(file);
113 struct super_block *sb = inode->i_sb;
114 struct buffer_head *bh = NULL;
115 struct fscrypt_str fstr = FSTR_INIT(NULL, 0);
117 if (IS_ENCRYPTED(inode)) {
118 err = fscrypt_get_encryption_info(inode);
119 if (err && err != -ENOKEY)
123 if (is_dx_dir(inode)) {
124 err = ext4_dx_readdir(file, ctx);
125 if (err != ERR_BAD_DX_DIR) {
129 * We don't set the inode dirty flag since it's not
130 * critical that it get flushed back to the disk.
132 ext4_clear_inode_flag(file_inode(file),
136 if (ext4_has_inline_data(inode)) {
137 int has_inline_data = 1;
138 err = ext4_read_inline_dir(file, ctx,
144 if (IS_ENCRYPTED(inode)) {
145 err = fscrypt_fname_alloc_buffer(inode, EXT4_NAME_LEN, &fstr);
150 while (ctx->pos < inode->i_size) {
151 struct ext4_map_blocks map;
153 if (fatal_signal_pending(current)) {
158 offset = ctx->pos & (sb->s_blocksize - 1);
159 map.m_lblk = ctx->pos >> EXT4_BLOCK_SIZE_BITS(sb);
161 err = ext4_map_blocks(NULL, inode, &map, 0);
163 /* m_len should never be zero but let's avoid
164 * an infinite loop if it somehow is */
167 ctx->pos += map.m_len * sb->s_blocksize;
171 pgoff_t index = map.m_pblk >>
172 (PAGE_SHIFT - inode->i_blkbits);
173 if (!ra_has_index(&file->f_ra, index))
174 page_cache_sync_readahead(
175 sb->s_bdev->bd_inode->i_mapping,
178 file->f_ra.prev_pos = (loff_t)index << PAGE_SHIFT;
179 bh = ext4_bread(NULL, inode, map.m_lblk, 0);
188 /* corrupt size? Maybe no more blocks to read */
189 if (ctx->pos > inode->i_blocks << 9)
191 ctx->pos += sb->s_blocksize - offset;
195 /* Check the checksum */
196 if (!buffer_verified(bh) &&
197 !ext4_dirblock_csum_verify(inode, bh)) {
198 EXT4_ERROR_FILE(file, 0, "directory fails checksum "
200 (unsigned long long)ctx->pos);
201 ctx->pos += sb->s_blocksize - offset;
206 set_buffer_verified(bh);
208 /* If the dir block has changed since the last call to
209 * readdir(2), then we might be pointing to an invalid
210 * dirent right now. Scan from the start of the block
212 if (!inode_eq_iversion(inode, file->f_version)) {
213 for (i = 0; i < sb->s_blocksize && i < offset; ) {
214 de = (struct ext4_dir_entry_2 *)
216 /* It's too expensive to do a full
217 * dirent test each time round this
218 * loop, but we do have to test at
219 * least that it is non-zero. A
220 * failure will be detected in the
221 * dirent test below. */
222 if (ext4_rec_len_from_disk(de->rec_len,
223 sb->s_blocksize) < EXT4_DIR_REC_LEN(1))
225 i += ext4_rec_len_from_disk(de->rec_len,
229 ctx->pos = (ctx->pos & ~(sb->s_blocksize - 1))
231 file->f_version = inode_query_iversion(inode);
234 while (ctx->pos < inode->i_size
235 && offset < sb->s_blocksize) {
236 de = (struct ext4_dir_entry_2 *) (bh->b_data + offset);
237 if (ext4_check_dir_entry(inode, file, de, bh,
238 bh->b_data, bh->b_size,
241 * On error, skip to the next block
243 ctx->pos = (ctx->pos |
244 (sb->s_blocksize - 1)) + 1;
247 offset += ext4_rec_len_from_disk(de->rec_len,
249 if (le32_to_cpu(de->inode)) {
250 if (!IS_ENCRYPTED(inode)) {
251 if (!dir_emit(ctx, de->name,
253 le32_to_cpu(de->inode),
254 get_dtype(sb, de->file_type)))
257 int save_len = fstr.len;
258 struct fscrypt_str de_name =
262 /* Directory is encrypted */
263 err = fscrypt_fname_disk_to_usr(inode,
264 0, 0, &de_name, &fstr);
270 de_name.name, de_name.len,
271 le32_to_cpu(de->inode),
272 get_dtype(sb, de->file_type)))
276 ctx->pos += ext4_rec_len_from_disk(de->rec_len,
279 if ((ctx->pos < inode->i_size) && !dir_relax_shared(inode))
288 fscrypt_fname_free_buffer(&fstr);
293 static inline int is_32bit_api(void)
296 return in_compat_syscall();
298 return (BITS_PER_LONG == 32);
303 * These functions convert from the major/minor hash to an f_pos
304 * value for dx directories
306 * Upper layer (for example NFS) should specify FMODE_32BITHASH or
307 * FMODE_64BITHASH explicitly. On the other hand, we allow ext4 to be mounted
308 * directly on both 32-bit and 64-bit nodes, under such case, neither
309 * FMODE_32BITHASH nor FMODE_64BITHASH is specified.
311 static inline loff_t hash2pos(struct file *filp, __u32 major, __u32 minor)
313 if ((filp->f_mode & FMODE_32BITHASH) ||
314 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
317 return ((__u64)(major >> 1) << 32) | (__u64)minor;
320 static inline __u32 pos2maj_hash(struct file *filp, loff_t pos)
322 if ((filp->f_mode & FMODE_32BITHASH) ||
323 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
324 return (pos << 1) & 0xffffffff;
326 return ((pos >> 32) << 1) & 0xffffffff;
329 static inline __u32 pos2min_hash(struct file *filp, loff_t pos)
331 if ((filp->f_mode & FMODE_32BITHASH) ||
332 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
335 return pos & 0xffffffff;
339 * Return 32- or 64-bit end-of-file for dx directories
341 static inline loff_t ext4_get_htree_eof(struct file *filp)
343 if ((filp->f_mode & FMODE_32BITHASH) ||
344 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
345 return EXT4_HTREE_EOF_32BIT;
347 return EXT4_HTREE_EOF_64BIT;
352 * ext4_dir_llseek() calls generic_file_llseek_size to handle htree
353 * directories, where the "offset" is in terms of the filename hash
354 * value instead of the byte offset.
356 * Because we may return a 64-bit hash that is well beyond offset limits,
357 * we need to pass the max hash as the maximum allowable offset in
358 * the htree directory case.
360 * For non-htree, ext4_llseek already chooses the proper max offset.
362 static loff_t ext4_dir_llseek(struct file *file, loff_t offset, int whence)
364 struct inode *inode = file->f_mapping->host;
365 int dx_dir = is_dx_dir(inode);
366 loff_t ret, htree_max = ext4_get_htree_eof(file);
369 ret = generic_file_llseek_size(file, offset, whence,
370 htree_max, htree_max);
372 ret = ext4_llseek(file, offset, whence);
373 file->f_version = inode_peek_iversion(inode) - 1;
378 * This structure holds the nodes of the red-black tree used to store
379 * the directory entry in hash order.
384 struct rb_node rb_hash;
393 * This functoin implements a non-recursive way of freeing all of the
394 * nodes in the red-black tree.
396 static void free_rb_tree_fname(struct rb_root *root)
398 struct fname *fname, *next;
400 rbtree_postorder_for_each_entry_safe(fname, next, root, rb_hash)
402 struct fname *old = fname;
411 static struct dir_private_info *ext4_htree_create_dir_info(struct file *filp,
414 struct dir_private_info *p;
416 p = kzalloc(sizeof(*p), GFP_KERNEL);
419 p->curr_hash = pos2maj_hash(filp, pos);
420 p->curr_minor_hash = pos2min_hash(filp, pos);
424 void ext4_htree_free_dir_info(struct dir_private_info *p)
426 free_rb_tree_fname(&p->root);
431 * Given a directory entry, enter it into the fname rb tree.
433 * When filename encryption is enabled, the dirent will hold the
434 * encrypted filename, while the htree will hold decrypted filename.
435 * The decrypted filename is passed in via ent_name. parameter.
437 int ext4_htree_store_dirent(struct file *dir_file, __u32 hash,
439 struct ext4_dir_entry_2 *dirent,
440 struct fscrypt_str *ent_name)
442 struct rb_node **p, *parent = NULL;
443 struct fname *fname, *new_fn;
444 struct dir_private_info *info;
447 info = dir_file->private_data;
448 p = &info->root.rb_node;
450 /* Create and allocate the fname structure */
451 len = sizeof(struct fname) + ent_name->len + 1;
452 new_fn = kzalloc(len, GFP_KERNEL);
456 new_fn->minor_hash = minor_hash;
457 new_fn->inode = le32_to_cpu(dirent->inode);
458 new_fn->name_len = ent_name->len;
459 new_fn->file_type = dirent->file_type;
460 memcpy(new_fn->name, ent_name->name, ent_name->len);
461 new_fn->name[ent_name->len] = 0;
465 fname = rb_entry(parent, struct fname, rb_hash);
468 * If the hash and minor hash match up, then we put
469 * them on a linked list. This rarely happens...
471 if ((new_fn->hash == fname->hash) &&
472 (new_fn->minor_hash == fname->minor_hash)) {
473 new_fn->next = fname->next;
474 fname->next = new_fn;
478 if (new_fn->hash < fname->hash)
480 else if (new_fn->hash > fname->hash)
482 else if (new_fn->minor_hash < fname->minor_hash)
484 else /* if (new_fn->minor_hash > fname->minor_hash) */
488 rb_link_node(&new_fn->rb_hash, parent, p);
489 rb_insert_color(&new_fn->rb_hash, &info->root);
496 * This is a helper function for ext4_dx_readdir. It calls filldir
497 * for all entres on the fname linked list. (Normally there is only
498 * one entry on the linked list, unless there are 62 bit hash collisions.)
500 static int call_filldir(struct file *file, struct dir_context *ctx,
503 struct dir_private_info *info = file->private_data;
504 struct inode *inode = file_inode(file);
505 struct super_block *sb = inode->i_sb;
508 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: comm %s: "
509 "called with null fname?!?", __func__, __LINE__,
510 inode->i_ino, current->comm);
513 ctx->pos = hash2pos(file, fname->hash, fname->minor_hash);
515 if (!dir_emit(ctx, fname->name,
518 get_dtype(sb, fname->file_type))) {
519 info->extra_fname = fname;
527 static int ext4_dx_readdir(struct file *file, struct dir_context *ctx)
529 struct dir_private_info *info = file->private_data;
530 struct inode *inode = file_inode(file);
535 info = ext4_htree_create_dir_info(file, ctx->pos);
538 file->private_data = info;
541 if (ctx->pos == ext4_get_htree_eof(file))
544 /* Some one has messed with f_pos; reset the world */
545 if (info->last_pos != ctx->pos) {
546 free_rb_tree_fname(&info->root);
547 info->curr_node = NULL;
548 info->extra_fname = NULL;
549 info->curr_hash = pos2maj_hash(file, ctx->pos);
550 info->curr_minor_hash = pos2min_hash(file, ctx->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(file, ctx, 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 !inode_eq_iversion(inode, file->f_version)) {
573 info->curr_node = NULL;
574 free_rb_tree_fname(&info->root);
575 file->f_version = inode_query_iversion(inode);
576 ret = ext4_htree_fill_tree(file, info->curr_hash,
577 info->curr_minor_hash,
582 ctx->pos = ext4_get_htree_eof(file);
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(file, ctx, 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 ctx->pos = ext4_get_htree_eof(file);
605 info->curr_hash = info->next_hash;
606 info->curr_minor_hash = 0;
610 info->last_pos = ctx->pos;
614 static int ext4_dir_open(struct inode * inode, struct file * filp)
616 if (IS_ENCRYPTED(inode))
617 return fscrypt_get_encryption_info(inode) ? -EACCES : 0;
621 static int ext4_release_dir(struct inode *inode, struct file *filp)
623 if (filp->private_data)
624 ext4_htree_free_dir_info(filp->private_data);
629 int ext4_check_all_de(struct inode *dir, struct buffer_head *bh, void *buf,
632 struct ext4_dir_entry_2 *de;
634 unsigned int offset = 0;
637 de = (struct ext4_dir_entry_2 *)buf;
638 top = buf + buf_size;
639 while ((char *) de < top) {
640 if (ext4_check_dir_entry(dir, NULL, de, bh,
641 buf, buf_size, offset))
642 return -EFSCORRUPTED;
643 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
644 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
647 if ((char *) de > top)
648 return -EFSCORRUPTED;
653 const struct file_operations ext4_dir_operations = {
654 .llseek = ext4_dir_llseek,
655 .read = generic_read_dir,
656 .iterate_shared = ext4_readdir,
657 .unlocked_ioctl = ext4_ioctl,
659 .compat_ioctl = ext4_compat_ioctl,
661 .fsync = ext4_sync_file,
662 .open = ext4_dir_open,
663 .release = ext4_release_dir,
666 #ifdef CONFIG_UNICODE
667 static int ext4_d_compare(const struct dentry *dentry, unsigned int len,
668 const char *str, const struct qstr *name)
670 struct qstr qstr = {.name = str, .len = len };
671 struct inode *inode = dentry->d_parent->d_inode;
673 if (!IS_CASEFOLDED(inode) || !EXT4_SB(inode->i_sb)->s_encoding) {
674 if (len != name->len)
676 return memcmp(str, name->name, len);
679 return ext4_ci_compare(inode, name, &qstr, false);
682 static int ext4_d_hash(const struct dentry *dentry, struct qstr *str)
684 const struct ext4_sb_info *sbi = EXT4_SB(dentry->d_sb);
685 const struct unicode_map *um = sbi->s_encoding;
689 if (!IS_CASEFOLDED(dentry->d_inode) || !um)
692 norm = kmalloc(PATH_MAX, GFP_ATOMIC);
696 len = utf8_casefold(um, str, norm, PATH_MAX);
698 if (ext4_has_strict_mode(sbi))
702 str->hash = full_name_hash(dentry, norm, len);
708 const struct dentry_operations ext4_dentry_ops = {
709 .d_hash = ext4_d_hash,
710 .d_compare = ext4_d_compare,