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)))
58 static bool is_fake_dir_entry(struct ext4_dir_entry_2 *de)
60 /* Check if . or .. , or skip if namelen is 0 */
61 if ((de->name_len > 0) && (de->name_len <= 2) && (de->name[0] == '.') &&
62 (de->name[1] == '.' || de->name[1] == '\0'))
64 /* Check if this is a csum entry */
65 if (de->file_type == EXT4_FT_DIR_CSUM)
71 * Return 0 if the directory entry is OK, and 1 if there is a problem
73 * Note: this is the opposite of what ext2 and ext3 historically returned...
75 * bh passed here can be an inode block or a dir data block, depending
76 * on the inode inline data flag.
78 int __ext4_check_dir_entry(const char *function, unsigned int line,
79 struct inode *dir, struct file *filp,
80 struct ext4_dir_entry_2 *de,
81 struct buffer_head *bh, char *buf, int size,
84 const char *error_msg = NULL;
85 const int rlen = ext4_rec_len_from_disk(de->rec_len,
86 dir->i_sb->s_blocksize);
87 const int next_offset = ((char *) de - buf) + rlen;
88 bool fake = is_fake_dir_entry(de);
89 bool has_csum = ext4_has_metadata_csum(dir->i_sb);
91 if (unlikely(rlen < ext4_dir_rec_len(1, fake ? NULL : dir)))
92 error_msg = "rec_len is smaller than minimal";
93 else if (unlikely(rlen % 4 != 0))
94 error_msg = "rec_len % 4 != 0";
95 else if (unlikely(rlen < ext4_dir_rec_len(de->name_len,
97 error_msg = "rec_len is too small for name_len";
98 else if (unlikely(next_offset > size))
99 error_msg = "directory entry overrun";
100 else if (unlikely(next_offset > size - ext4_dir_rec_len(1,
101 has_csum ? NULL : dir) &&
102 next_offset != size))
103 error_msg = "directory entry too close to block end";
104 else if (unlikely(le32_to_cpu(de->inode) >
105 le32_to_cpu(EXT4_SB(dir->i_sb)->s_es->s_inodes_count)))
106 error_msg = "inode out of bounds";
111 ext4_error_file(filp, function, line, bh->b_blocknr,
112 "bad entry in directory: %s - offset=%u, "
113 "inode=%u, rec_len=%d, size=%d fake=%d",
114 error_msg, offset, le32_to_cpu(de->inode),
117 ext4_error_inode(dir, function, line, bh->b_blocknr,
118 "bad entry in directory: %s - offset=%u, "
119 "inode=%u, rec_len=%d, size=%d fake=%d",
120 error_msg, offset, le32_to_cpu(de->inode),
126 static int ext4_readdir(struct file *file, struct dir_context *ctx)
130 struct ext4_dir_entry_2 *de;
132 struct inode *inode = file_inode(file);
133 struct super_block *sb = inode->i_sb;
134 struct buffer_head *bh = NULL;
135 struct fscrypt_str fstr = FSTR_INIT(NULL, 0);
137 err = fscrypt_prepare_readdir(inode);
141 if (is_dx_dir(inode)) {
142 err = ext4_dx_readdir(file, ctx);
143 if (err != ERR_BAD_DX_DIR)
146 /* Can we just clear INDEX flag to ignore htree information? */
147 if (!ext4_has_metadata_csum(sb)) {
149 * We don't set the inode dirty flag since it's not
150 * critical that it gets flushed back to the disk.
152 ext4_clear_inode_flag(inode, EXT4_INODE_INDEX);
156 if (ext4_has_inline_data(inode)) {
157 int has_inline_data = 1;
158 err = ext4_read_inline_dir(file, ctx,
164 if (IS_ENCRYPTED(inode)) {
165 err = fscrypt_fname_alloc_buffer(EXT4_NAME_LEN, &fstr);
170 while (ctx->pos < inode->i_size) {
171 struct ext4_map_blocks map;
173 if (fatal_signal_pending(current)) {
178 offset = ctx->pos & (sb->s_blocksize - 1);
179 map.m_lblk = ctx->pos >> EXT4_BLOCK_SIZE_BITS(sb);
181 err = ext4_map_blocks(NULL, inode, &map, 0);
183 /* m_len should never be zero but let's avoid
184 * an infinite loop if it somehow is */
187 ctx->pos += map.m_len * sb->s_blocksize;
191 pgoff_t index = map.m_pblk >>
192 (PAGE_SHIFT - inode->i_blkbits);
193 if (!ra_has_index(&file->f_ra, index))
194 page_cache_sync_readahead(
195 sb->s_bdev->bd_inode->i_mapping,
198 file->f_ra.prev_pos = (loff_t)index << PAGE_SHIFT;
199 bh = ext4_bread(NULL, inode, map.m_lblk, 0);
208 /* corrupt size? Maybe no more blocks to read */
209 if (ctx->pos > inode->i_blocks << 9)
211 ctx->pos += sb->s_blocksize - offset;
215 /* Check the checksum */
216 if (!buffer_verified(bh) &&
217 !ext4_dirblock_csum_verify(inode, bh)) {
218 EXT4_ERROR_FILE(file, 0, "directory fails checksum "
220 (unsigned long long)ctx->pos);
221 ctx->pos += sb->s_blocksize - offset;
226 set_buffer_verified(bh);
228 /* If the dir block has changed since the last call to
229 * readdir(2), then we might be pointing to an invalid
230 * dirent right now. Scan from the start of the block
232 if (!inode_eq_iversion(inode, file->f_version)) {
233 for (i = 0; i < sb->s_blocksize && i < offset; ) {
234 de = (struct ext4_dir_entry_2 *)
236 /* It's too expensive to do a full
237 * dirent test each time round this
238 * loop, but we do have to test at
239 * least that it is non-zero. A
240 * failure will be detected in the
241 * dirent test below. */
242 if (ext4_rec_len_from_disk(de->rec_len,
243 sb->s_blocksize) < ext4_dir_rec_len(1,
246 i += ext4_rec_len_from_disk(de->rec_len,
250 ctx->pos = (ctx->pos & ~(sb->s_blocksize - 1))
252 file->f_version = inode_query_iversion(inode);
255 while (ctx->pos < inode->i_size
256 && offset < sb->s_blocksize) {
257 de = (struct ext4_dir_entry_2 *) (bh->b_data + offset);
258 if (ext4_check_dir_entry(inode, file, de, bh,
259 bh->b_data, bh->b_size,
262 * On error, skip to the next block
264 ctx->pos = (ctx->pos |
265 (sb->s_blocksize - 1)) + 1;
268 offset += ext4_rec_len_from_disk(de->rec_len,
270 if (le32_to_cpu(de->inode)) {
271 if (!IS_ENCRYPTED(inode)) {
272 if (!dir_emit(ctx, de->name,
274 le32_to_cpu(de->inode),
275 get_dtype(sb, de->file_type)))
278 int save_len = fstr.len;
279 struct fscrypt_str de_name =
283 /* Directory is encrypted */
284 err = fscrypt_fname_disk_to_usr(inode,
285 EXT4_DIRENT_HASH(de),
286 EXT4_DIRENT_MINOR_HASH(de),
293 de_name.name, de_name.len,
294 le32_to_cpu(de->inode),
295 get_dtype(sb, de->file_type)))
299 ctx->pos += ext4_rec_len_from_disk(de->rec_len,
302 if ((ctx->pos < inode->i_size) && !dir_relax_shared(inode))
310 fscrypt_fname_free_buffer(&fstr);
315 static inline int is_32bit_api(void)
318 return in_compat_syscall();
320 return (BITS_PER_LONG == 32);
325 * These functions convert from the major/minor hash to an f_pos
326 * value for dx directories
328 * Upper layer (for example NFS) should specify FMODE_32BITHASH or
329 * FMODE_64BITHASH explicitly. On the other hand, we allow ext4 to be mounted
330 * directly on both 32-bit and 64-bit nodes, under such case, neither
331 * FMODE_32BITHASH nor FMODE_64BITHASH is specified.
333 static inline loff_t hash2pos(struct file *filp, __u32 major, __u32 minor)
335 if ((filp->f_mode & FMODE_32BITHASH) ||
336 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
339 return ((__u64)(major >> 1) << 32) | (__u64)minor;
342 static inline __u32 pos2maj_hash(struct file *filp, loff_t pos)
344 if ((filp->f_mode & FMODE_32BITHASH) ||
345 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
346 return (pos << 1) & 0xffffffff;
348 return ((pos >> 32) << 1) & 0xffffffff;
351 static inline __u32 pos2min_hash(struct file *filp, loff_t pos)
353 if ((filp->f_mode & FMODE_32BITHASH) ||
354 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
357 return pos & 0xffffffff;
361 * Return 32- or 64-bit end-of-file for dx directories
363 static inline loff_t ext4_get_htree_eof(struct file *filp)
365 if ((filp->f_mode & FMODE_32BITHASH) ||
366 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
367 return EXT4_HTREE_EOF_32BIT;
369 return EXT4_HTREE_EOF_64BIT;
374 * ext4_dir_llseek() calls generic_file_llseek_size to handle htree
375 * directories, where the "offset" is in terms of the filename hash
376 * value instead of the byte offset.
378 * Because we may return a 64-bit hash that is well beyond offset limits,
379 * we need to pass the max hash as the maximum allowable offset in
380 * the htree directory case.
382 * For non-htree, ext4_llseek already chooses the proper max offset.
384 static loff_t ext4_dir_llseek(struct file *file, loff_t offset, int whence)
386 struct inode *inode = file->f_mapping->host;
387 int dx_dir = is_dx_dir(inode);
388 loff_t ret, htree_max = ext4_get_htree_eof(file);
391 ret = generic_file_llseek_size(file, offset, whence,
392 htree_max, htree_max);
394 ret = ext4_llseek(file, offset, whence);
395 file->f_version = inode_peek_iversion(inode) - 1;
400 * This structure holds the nodes of the red-black tree used to store
401 * the directory entry in hash order.
406 struct rb_node rb_hash;
415 * This function implements a non-recursive way of freeing all of the
416 * nodes in the red-black tree.
418 static void free_rb_tree_fname(struct rb_root *root)
420 struct fname *fname, *next;
422 rbtree_postorder_for_each_entry_safe(fname, next, root, rb_hash)
424 struct fname *old = fname;
433 static struct dir_private_info *ext4_htree_create_dir_info(struct file *filp,
436 struct dir_private_info *p;
438 p = kzalloc(sizeof(*p), GFP_KERNEL);
441 p->curr_hash = pos2maj_hash(filp, pos);
442 p->curr_minor_hash = pos2min_hash(filp, pos);
446 void ext4_htree_free_dir_info(struct dir_private_info *p)
448 free_rb_tree_fname(&p->root);
453 * Given a directory entry, enter it into the fname rb tree.
455 * When filename encryption is enabled, the dirent will hold the
456 * encrypted filename, while the htree will hold decrypted filename.
457 * The decrypted filename is passed in via ent_name. parameter.
459 int ext4_htree_store_dirent(struct file *dir_file, __u32 hash,
461 struct ext4_dir_entry_2 *dirent,
462 struct fscrypt_str *ent_name)
464 struct rb_node **p, *parent = NULL;
465 struct fname *fname, *new_fn;
466 struct dir_private_info *info;
469 info = dir_file->private_data;
470 p = &info->root.rb_node;
472 /* Create and allocate the fname structure */
473 len = sizeof(struct fname) + ent_name->len + 1;
474 new_fn = kzalloc(len, GFP_KERNEL);
478 new_fn->minor_hash = minor_hash;
479 new_fn->inode = le32_to_cpu(dirent->inode);
480 new_fn->name_len = ent_name->len;
481 new_fn->file_type = dirent->file_type;
482 memcpy(new_fn->name, ent_name->name, ent_name->len);
486 fname = rb_entry(parent, struct fname, rb_hash);
489 * If the hash and minor hash match up, then we put
490 * them on a linked list. This rarely happens...
492 if ((new_fn->hash == fname->hash) &&
493 (new_fn->minor_hash == fname->minor_hash)) {
494 new_fn->next = fname->next;
495 fname->next = new_fn;
499 if (new_fn->hash < fname->hash)
501 else if (new_fn->hash > fname->hash)
503 else if (new_fn->minor_hash < fname->minor_hash)
505 else /* if (new_fn->minor_hash > fname->minor_hash) */
509 rb_link_node(&new_fn->rb_hash, parent, p);
510 rb_insert_color(&new_fn->rb_hash, &info->root);
517 * This is a helper function for ext4_dx_readdir. It calls filldir
518 * for all entries on the fname linked list. (Normally there is only
519 * one entry on the linked list, unless there are 62 bit hash collisions.)
521 static int call_filldir(struct file *file, struct dir_context *ctx,
524 struct dir_private_info *info = file->private_data;
525 struct inode *inode = file_inode(file);
526 struct super_block *sb = inode->i_sb;
529 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: comm %s: "
530 "called with null fname?!?", __func__, __LINE__,
531 inode->i_ino, current->comm);
534 ctx->pos = hash2pos(file, fname->hash, fname->minor_hash);
536 if (!dir_emit(ctx, fname->name,
539 get_dtype(sb, fname->file_type))) {
540 info->extra_fname = fname;
548 static int ext4_dx_readdir(struct file *file, struct dir_context *ctx)
550 struct dir_private_info *info = file->private_data;
551 struct inode *inode = file_inode(file);
556 info = ext4_htree_create_dir_info(file, ctx->pos);
559 file->private_data = info;
562 if (ctx->pos == ext4_get_htree_eof(file))
565 /* Some one has messed with f_pos; reset the world */
566 if (info->last_pos != ctx->pos) {
567 free_rb_tree_fname(&info->root);
568 info->curr_node = NULL;
569 info->extra_fname = NULL;
570 info->curr_hash = pos2maj_hash(file, ctx->pos);
571 info->curr_minor_hash = pos2min_hash(file, ctx->pos);
575 * If there are any leftover names on the hash collision
576 * chain, return them first.
578 if (info->extra_fname) {
579 if (call_filldir(file, ctx, info->extra_fname))
581 info->extra_fname = NULL;
583 } else if (!info->curr_node)
584 info->curr_node = rb_first(&info->root);
588 * Fill the rbtree if we have no more entries,
589 * or the inode has changed since we last read in the
592 if ((!info->curr_node) ||
593 !inode_eq_iversion(inode, file->f_version)) {
594 info->curr_node = NULL;
595 free_rb_tree_fname(&info->root);
596 file->f_version = inode_query_iversion(inode);
597 ret = ext4_htree_fill_tree(file, info->curr_hash,
598 info->curr_minor_hash,
603 ctx->pos = ext4_get_htree_eof(file);
606 info->curr_node = rb_first(&info->root);
609 fname = rb_entry(info->curr_node, struct fname, rb_hash);
610 info->curr_hash = fname->hash;
611 info->curr_minor_hash = fname->minor_hash;
612 if (call_filldir(file, ctx, fname))
615 info->curr_node = rb_next(info->curr_node);
616 if (info->curr_node) {
617 fname = rb_entry(info->curr_node, struct fname,
619 info->curr_hash = fname->hash;
620 info->curr_minor_hash = fname->minor_hash;
622 if (info->next_hash == ~0) {
623 ctx->pos = ext4_get_htree_eof(file);
626 info->curr_hash = info->next_hash;
627 info->curr_minor_hash = 0;
631 info->last_pos = ctx->pos;
632 return ret < 0 ? ret : 0;
635 static int ext4_release_dir(struct inode *inode, struct file *filp)
637 if (filp->private_data)
638 ext4_htree_free_dir_info(filp->private_data);
643 int ext4_check_all_de(struct inode *dir, struct buffer_head *bh, void *buf,
646 struct ext4_dir_entry_2 *de;
648 unsigned int offset = 0;
652 top = buf + buf_size;
653 while ((char *) de < top) {
654 if (ext4_check_dir_entry(dir, NULL, de, bh,
655 buf, buf_size, offset))
656 return -EFSCORRUPTED;
657 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
658 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
661 if ((char *) de > top)
662 return -EFSCORRUPTED;
667 const struct file_operations ext4_dir_operations = {
668 .llseek = ext4_dir_llseek,
669 .read = generic_read_dir,
670 .iterate_shared = ext4_readdir,
671 .unlocked_ioctl = ext4_ioctl,
673 .compat_ioctl = ext4_compat_ioctl,
675 .fsync = ext4_sync_file,
676 .release = ext4_release_dir,