1 // SPDX-License-Identifier: GPL-2.0
5 * Copyright (C) 1992, 1993, 1994, 1995
6 * Remy Card (card@masi.ibp.fr)
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
19 * David S. Miller (davem@caip.rutgers.edu), 1995
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(((char *) de - buf) + rlen >
85 size - EXT4_DIR_REC_LEN(1) &&
86 ((char *) de - buf) + rlen != size)) {
87 error_msg = "directory entry too close to block end";
89 else if (unlikely(le32_to_cpu(de->inode) >
90 le32_to_cpu(EXT4_SB(dir->i_sb)->s_es->s_inodes_count)))
91 error_msg = "inode out of bounds";
96 ext4_error_file(filp, 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);
102 ext4_error_inode(dir, function, line, bh->b_blocknr,
103 "bad entry in directory: %s - offset=%u, "
104 "inode=%u, rec_len=%d, name_len=%d, size=%d",
105 error_msg, offset, le32_to_cpu(de->inode),
106 rlen, de->name_len, size);
111 static int ext4_readdir(struct file *file, struct dir_context *ctx)
115 struct ext4_dir_entry_2 *de;
117 struct inode *inode = file_inode(file);
118 struct super_block *sb = inode->i_sb;
119 struct buffer_head *bh = NULL;
120 struct fscrypt_str fstr = FSTR_INIT(NULL, 0);
122 if (IS_ENCRYPTED(inode)) {
123 err = fscrypt_get_encryption_info(inode);
124 if (err && err != -ENOKEY)
128 if (is_dx_dir(inode)) {
129 err = ext4_dx_readdir(file, ctx);
130 if (err != ERR_BAD_DX_DIR) {
133 /* Can we just clear INDEX flag to ignore htree information? */
134 if (!ext4_has_metadata_csum(sb)) {
136 * We don't set the inode dirty flag since it's not
137 * critical that it gets flushed back to the disk.
139 ext4_clear_inode_flag(inode, EXT4_INODE_INDEX);
143 if (ext4_has_inline_data(inode)) {
144 int has_inline_data = 1;
145 err = ext4_read_inline_dir(file, ctx,
151 if (IS_ENCRYPTED(inode)) {
152 err = fscrypt_fname_alloc_buffer(inode, EXT4_NAME_LEN, &fstr);
157 while (ctx->pos < inode->i_size) {
158 struct ext4_map_blocks map;
160 if (fatal_signal_pending(current)) {
165 offset = ctx->pos & (sb->s_blocksize - 1);
166 map.m_lblk = ctx->pos >> EXT4_BLOCK_SIZE_BITS(sb);
168 err = ext4_map_blocks(NULL, inode, &map, 0);
170 /* m_len should never be zero but let's avoid
171 * an infinite loop if it somehow is */
174 ctx->pos += map.m_len * sb->s_blocksize;
178 pgoff_t index = map.m_pblk >>
179 (PAGE_SHIFT - inode->i_blkbits);
180 if (!ra_has_index(&file->f_ra, index))
181 page_cache_sync_readahead(
182 sb->s_bdev->bd_inode->i_mapping,
185 file->f_ra.prev_pos = (loff_t)index << PAGE_SHIFT;
186 bh = ext4_bread(NULL, inode, map.m_lblk, 0);
195 /* corrupt size? Maybe no more blocks to read */
196 if (ctx->pos > inode->i_blocks << 9)
198 ctx->pos += sb->s_blocksize - offset;
202 /* Check the checksum */
203 if (!buffer_verified(bh) &&
204 !ext4_dirblock_csum_verify(inode, bh)) {
205 EXT4_ERROR_FILE(file, 0, "directory fails checksum "
207 (unsigned long long)ctx->pos);
208 ctx->pos += sb->s_blocksize - offset;
213 set_buffer_verified(bh);
215 /* If the dir block has changed since the last call to
216 * readdir(2), then we might be pointing to an invalid
217 * dirent right now. Scan from the start of the block
219 if (!inode_eq_iversion(inode, file->f_version)) {
220 for (i = 0; i < sb->s_blocksize && i < offset; ) {
221 de = (struct ext4_dir_entry_2 *)
223 /* It's too expensive to do a full
224 * dirent test each time round this
225 * loop, but we do have to test at
226 * least that it is non-zero. A
227 * failure will be detected in the
228 * dirent test below. */
229 if (ext4_rec_len_from_disk(de->rec_len,
230 sb->s_blocksize) < EXT4_DIR_REC_LEN(1))
232 i += ext4_rec_len_from_disk(de->rec_len,
236 ctx->pos = (ctx->pos & ~(sb->s_blocksize - 1))
238 file->f_version = inode_query_iversion(inode);
241 while (ctx->pos < inode->i_size
242 && offset < sb->s_blocksize) {
243 de = (struct ext4_dir_entry_2 *) (bh->b_data + offset);
244 if (ext4_check_dir_entry(inode, file, de, bh,
245 bh->b_data, bh->b_size,
248 * On error, skip to the next block
250 ctx->pos = (ctx->pos |
251 (sb->s_blocksize - 1)) + 1;
254 offset += ext4_rec_len_from_disk(de->rec_len,
256 if (le32_to_cpu(de->inode)) {
257 if (!IS_ENCRYPTED(inode)) {
258 if (!dir_emit(ctx, de->name,
260 le32_to_cpu(de->inode),
261 get_dtype(sb, de->file_type)))
264 int save_len = fstr.len;
265 struct fscrypt_str de_name =
269 /* Directory is encrypted */
270 err = fscrypt_fname_disk_to_usr(inode,
271 0, 0, &de_name, &fstr);
277 de_name.name, de_name.len,
278 le32_to_cpu(de->inode),
279 get_dtype(sb, de->file_type)))
283 ctx->pos += ext4_rec_len_from_disk(de->rec_len,
286 if ((ctx->pos < inode->i_size) && !dir_relax_shared(inode))
295 fscrypt_fname_free_buffer(&fstr);
300 static inline int is_32bit_api(void)
303 return in_compat_syscall();
305 return (BITS_PER_LONG == 32);
310 * These functions convert from the major/minor hash to an f_pos
311 * value for dx directories
313 * Upper layer (for example NFS) should specify FMODE_32BITHASH or
314 * FMODE_64BITHASH explicitly. On the other hand, we allow ext4 to be mounted
315 * directly on both 32-bit and 64-bit nodes, under such case, neither
316 * FMODE_32BITHASH nor FMODE_64BITHASH is specified.
318 static inline loff_t hash2pos(struct file *filp, __u32 major, __u32 minor)
320 if ((filp->f_mode & FMODE_32BITHASH) ||
321 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
324 return ((__u64)(major >> 1) << 32) | (__u64)minor;
327 static inline __u32 pos2maj_hash(struct file *filp, loff_t pos)
329 if ((filp->f_mode & FMODE_32BITHASH) ||
330 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
331 return (pos << 1) & 0xffffffff;
333 return ((pos >> 32) << 1) & 0xffffffff;
336 static inline __u32 pos2min_hash(struct file *filp, loff_t pos)
338 if ((filp->f_mode & FMODE_32BITHASH) ||
339 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
342 return pos & 0xffffffff;
346 * Return 32- or 64-bit end-of-file for dx directories
348 static inline loff_t ext4_get_htree_eof(struct file *filp)
350 if ((filp->f_mode & FMODE_32BITHASH) ||
351 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
352 return EXT4_HTREE_EOF_32BIT;
354 return EXT4_HTREE_EOF_64BIT;
359 * ext4_dir_llseek() calls generic_file_llseek_size to handle htree
360 * directories, where the "offset" is in terms of the filename hash
361 * value instead of the byte offset.
363 * Because we may return a 64-bit hash that is well beyond offset limits,
364 * we need to pass the max hash as the maximum allowable offset in
365 * the htree directory case.
367 * For non-htree, ext4_llseek already chooses the proper max offset.
369 static loff_t ext4_dir_llseek(struct file *file, loff_t offset, int whence)
371 struct inode *inode = file->f_mapping->host;
372 int dx_dir = is_dx_dir(inode);
373 loff_t ret, htree_max = ext4_get_htree_eof(file);
376 ret = generic_file_llseek_size(file, offset, whence,
377 htree_max, htree_max);
379 ret = ext4_llseek(file, offset, whence);
380 file->f_version = inode_peek_iversion(inode) - 1;
385 * This structure holds the nodes of the red-black tree used to store
386 * the directory entry in hash order.
391 struct rb_node rb_hash;
400 * This functoin implements a non-recursive way of freeing all of the
401 * nodes in the red-black tree.
403 static void free_rb_tree_fname(struct rb_root *root)
405 struct fname *fname, *next;
407 rbtree_postorder_for_each_entry_safe(fname, next, root, rb_hash)
409 struct fname *old = fname;
418 static struct dir_private_info *ext4_htree_create_dir_info(struct file *filp,
421 struct dir_private_info *p;
423 p = kzalloc(sizeof(*p), GFP_KERNEL);
426 p->curr_hash = pos2maj_hash(filp, pos);
427 p->curr_minor_hash = pos2min_hash(filp, pos);
431 void ext4_htree_free_dir_info(struct dir_private_info *p)
433 free_rb_tree_fname(&p->root);
438 * Given a directory entry, enter it into the fname rb tree.
440 * When filename encryption is enabled, the dirent will hold the
441 * encrypted filename, while the htree will hold decrypted filename.
442 * The decrypted filename is passed in via ent_name. parameter.
444 int ext4_htree_store_dirent(struct file *dir_file, __u32 hash,
446 struct ext4_dir_entry_2 *dirent,
447 struct fscrypt_str *ent_name)
449 struct rb_node **p, *parent = NULL;
450 struct fname *fname, *new_fn;
451 struct dir_private_info *info;
454 info = dir_file->private_data;
455 p = &info->root.rb_node;
457 /* Create and allocate the fname structure */
458 len = sizeof(struct fname) + ent_name->len + 1;
459 new_fn = kzalloc(len, GFP_KERNEL);
463 new_fn->minor_hash = minor_hash;
464 new_fn->inode = le32_to_cpu(dirent->inode);
465 new_fn->name_len = ent_name->len;
466 new_fn->file_type = dirent->file_type;
467 memcpy(new_fn->name, ent_name->name, ent_name->len);
468 new_fn->name[ent_name->len] = 0;
472 fname = rb_entry(parent, struct fname, rb_hash);
475 * If the hash and minor hash match up, then we put
476 * them on a linked list. This rarely happens...
478 if ((new_fn->hash == fname->hash) &&
479 (new_fn->minor_hash == fname->minor_hash)) {
480 new_fn->next = fname->next;
481 fname->next = new_fn;
485 if (new_fn->hash < fname->hash)
487 else if (new_fn->hash > fname->hash)
489 else if (new_fn->minor_hash < fname->minor_hash)
491 else /* if (new_fn->minor_hash > fname->minor_hash) */
495 rb_link_node(&new_fn->rb_hash, parent, p);
496 rb_insert_color(&new_fn->rb_hash, &info->root);
503 * This is a helper function for ext4_dx_readdir. It calls filldir
504 * for all entres on the fname linked list. (Normally there is only
505 * one entry on the linked list, unless there are 62 bit hash collisions.)
507 static int call_filldir(struct file *file, struct dir_context *ctx,
510 struct dir_private_info *info = file->private_data;
511 struct inode *inode = file_inode(file);
512 struct super_block *sb = inode->i_sb;
515 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: comm %s: "
516 "called with null fname?!?", __func__, __LINE__,
517 inode->i_ino, current->comm);
520 ctx->pos = hash2pos(file, fname->hash, fname->minor_hash);
522 if (!dir_emit(ctx, fname->name,
525 get_dtype(sb, fname->file_type))) {
526 info->extra_fname = fname;
534 static int ext4_dx_readdir(struct file *file, struct dir_context *ctx)
536 struct dir_private_info *info = file->private_data;
537 struct inode *inode = file_inode(file);
542 info = ext4_htree_create_dir_info(file, ctx->pos);
545 file->private_data = info;
548 if (ctx->pos == ext4_get_htree_eof(file))
551 /* Some one has messed with f_pos; reset the world */
552 if (info->last_pos != ctx->pos) {
553 free_rb_tree_fname(&info->root);
554 info->curr_node = NULL;
555 info->extra_fname = NULL;
556 info->curr_hash = pos2maj_hash(file, ctx->pos);
557 info->curr_minor_hash = pos2min_hash(file, ctx->pos);
561 * If there are any leftover names on the hash collision
562 * chain, return them first.
564 if (info->extra_fname) {
565 if (call_filldir(file, ctx, info->extra_fname))
567 info->extra_fname = NULL;
569 } else if (!info->curr_node)
570 info->curr_node = rb_first(&info->root);
574 * Fill the rbtree if we have no more entries,
575 * or the inode has changed since we last read in the
578 if ((!info->curr_node) ||
579 !inode_eq_iversion(inode, file->f_version)) {
580 info->curr_node = NULL;
581 free_rb_tree_fname(&info->root);
582 file->f_version = inode_query_iversion(inode);
583 ret = ext4_htree_fill_tree(file, info->curr_hash,
584 info->curr_minor_hash,
589 ctx->pos = ext4_get_htree_eof(file);
592 info->curr_node = rb_first(&info->root);
595 fname = rb_entry(info->curr_node, struct fname, rb_hash);
596 info->curr_hash = fname->hash;
597 info->curr_minor_hash = fname->minor_hash;
598 if (call_filldir(file, ctx, fname))
601 info->curr_node = rb_next(info->curr_node);
602 if (info->curr_node) {
603 fname = rb_entry(info->curr_node, struct fname,
605 info->curr_hash = fname->hash;
606 info->curr_minor_hash = fname->minor_hash;
608 if (info->next_hash == ~0) {
609 ctx->pos = ext4_get_htree_eof(file);
612 info->curr_hash = info->next_hash;
613 info->curr_minor_hash = 0;
617 info->last_pos = ctx->pos;
618 return ret < 0 ? ret : 0;
621 static int ext4_dir_open(struct inode * inode, struct file * filp)
623 if (IS_ENCRYPTED(inode))
624 return fscrypt_get_encryption_info(inode) ? -EACCES : 0;
628 static int ext4_release_dir(struct inode *inode, struct file *filp)
630 if (filp->private_data)
631 ext4_htree_free_dir_info(filp->private_data);
636 int ext4_check_all_de(struct inode *dir, struct buffer_head *bh, void *buf,
639 struct ext4_dir_entry_2 *de;
641 unsigned int offset = 0;
644 de = (struct ext4_dir_entry_2 *)buf;
645 top = buf + buf_size;
646 while ((char *) de < top) {
647 if (ext4_check_dir_entry(dir, NULL, de, bh,
648 buf, buf_size, offset))
649 return -EFSCORRUPTED;
650 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
651 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
654 if ((char *) de > top)
655 return -EFSCORRUPTED;
660 const struct file_operations ext4_dir_operations = {
661 .llseek = ext4_dir_llseek,
662 .read = generic_read_dir,
663 .iterate_shared = ext4_readdir,
664 .unlocked_ioctl = ext4_ioctl,
666 .compat_ioctl = ext4_compat_ioctl,
668 .fsync = ext4_sync_file,
669 .open = ext4_dir_open,
670 .release = ext4_release_dir,
673 #ifdef CONFIG_UNICODE
674 static int ext4_d_compare(const struct dentry *dentry, unsigned int len,
675 const char *str, const struct qstr *name)
677 struct qstr qstr = {.name = str, .len = len };
678 const struct dentry *parent = READ_ONCE(dentry->d_parent);
679 const struct inode *inode = READ_ONCE(parent->d_inode);
680 char strbuf[DNAME_INLINE_LEN];
682 if (!inode || !IS_CASEFOLDED(inode) ||
683 !EXT4_SB(inode->i_sb)->s_encoding) {
684 if (len != name->len)
686 return memcmp(str, name->name, len);
690 * If the dentry name is stored in-line, then it may be concurrently
691 * modified by a rename. If this happens, the VFS will eventually retry
692 * the lookup, so it doesn't matter what ->d_compare() returns.
693 * However, it's unsafe to call utf8_strncasecmp() with an unstable
694 * string. Therefore, we have to copy the name into a temporary buffer.
696 if (len <= DNAME_INLINE_LEN - 1) {
697 memcpy(strbuf, str, len);
700 /* prevent compiler from optimizing out the temporary buffer */
704 return ext4_ci_compare(inode, name, &qstr, false);
707 static int ext4_d_hash(const struct dentry *dentry, struct qstr *str)
709 const struct ext4_sb_info *sbi = EXT4_SB(dentry->d_sb);
710 const struct unicode_map *um = sbi->s_encoding;
711 const struct inode *inode = READ_ONCE(dentry->d_inode);
715 if (!inode || !IS_CASEFOLDED(inode) || !um)
718 norm = kmalloc(PATH_MAX, GFP_ATOMIC);
722 len = utf8_casefold(um, str, norm, PATH_MAX);
724 if (ext4_has_strict_mode(sbi))
728 str->hash = full_name_hash(dentry, norm, len);
734 const struct dentry_operations ext4_dentry_ops = {
735 .d_hash = ext4_d_hash,
736 .d_compare = ext4_d_compare,