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_COMPAT_FEATURE(inode->i_sb,
44 EXT4_FEATURE_COMPAT_DIR_INDEX) &&
45 ((ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) ||
46 ((inode->i_size >> sb->s_blocksize_bits) == 1) ||
47 ext4_has_inline_data(inode)))
54 * Return 0 if the directory entry is OK, and 1 if there is a problem
56 * Note: this is the opposite of what ext2 and ext3 historically returned...
58 * bh passed here can be an inode block or a dir data block, depending
59 * on the inode inline data flag.
61 int __ext4_check_dir_entry(const char *function, unsigned int line,
62 struct inode *dir, struct file *filp,
63 struct ext4_dir_entry_2 *de,
64 struct buffer_head *bh, char *buf, int size,
67 const char *error_msg = NULL;
68 const int rlen = ext4_rec_len_from_disk(de->rec_len,
69 dir->i_sb->s_blocksize);
71 if (unlikely(rlen < EXT4_DIR_REC_LEN(1)))
72 error_msg = "rec_len is smaller than minimal";
73 else if (unlikely(rlen % 4 != 0))
74 error_msg = "rec_len % 4 != 0";
75 else if (unlikely(rlen < EXT4_DIR_REC_LEN(de->name_len)))
76 error_msg = "rec_len is too small for name_len";
77 else if (unlikely(((char *) de - buf) + rlen > size))
78 error_msg = "directory entry across range";
79 else if (unlikely(le32_to_cpu(de->inode) >
80 le32_to_cpu(EXT4_SB(dir->i_sb)->s_es->s_inodes_count)))
81 error_msg = "inode out of bounds";
86 ext4_error_file(filp, function, line, bh->b_blocknr,
87 "bad entry in directory: %s - offset=%u(%u), "
88 "inode=%u, rec_len=%d, name_len=%d",
89 error_msg, (unsigned) (offset % size),
90 offset, le32_to_cpu(de->inode),
93 ext4_error_inode(dir, function, line, bh->b_blocknr,
94 "bad entry in directory: %s - offset=%u(%u), "
95 "inode=%u, rec_len=%d, name_len=%d",
96 error_msg, (unsigned) (offset % size),
97 offset, le32_to_cpu(de->inode),
103 static int ext4_readdir(struct file *file, struct dir_context *ctx)
107 struct ext4_dir_entry_2 *de;
109 struct inode *inode = file_inode(file);
110 struct super_block *sb = inode->i_sb;
111 struct buffer_head *bh = NULL;
112 int dir_has_error = 0;
113 struct ext4_fname_crypto_ctx *enc_ctx = NULL;
114 struct ext4_str fname_crypto_str = {.name = NULL, .len = 0};
116 if (is_dx_dir(inode)) {
117 err = ext4_dx_readdir(file, ctx);
118 if (err != ERR_BAD_DX_DIR) {
122 * We don't set the inode dirty flag since it's not
123 * critical that it get flushed back to the disk.
125 ext4_clear_inode_flag(file_inode(file),
129 if (ext4_has_inline_data(inode)) {
130 int has_inline_data = 1;
131 err = ext4_read_inline_dir(file, ctx,
137 enc_ctx = ext4_get_fname_crypto_ctx(inode, EXT4_NAME_LEN);
139 return PTR_ERR(enc_ctx);
141 err = ext4_fname_crypto_alloc_buffer(enc_ctx, EXT4_NAME_LEN,
144 ext4_put_fname_crypto_ctx(&enc_ctx);
149 offset = ctx->pos & (sb->s_blocksize - 1);
151 while (ctx->pos < inode->i_size) {
152 struct ext4_map_blocks map;
154 map.m_lblk = ctx->pos >> EXT4_BLOCK_SIZE_BITS(sb);
156 err = ext4_map_blocks(NULL, inode, &map, 0);
158 pgoff_t index = map.m_pblk >>
159 (PAGE_CACHE_SHIFT - inode->i_blkbits);
160 if (!ra_has_index(&file->f_ra, index))
161 page_cache_sync_readahead(
162 sb->s_bdev->bd_inode->i_mapping,
165 file->f_ra.prev_pos = (loff_t)index << PAGE_CACHE_SHIFT;
166 bh = ext4_bread(NULL, inode, map.m_lblk, 0);
172 if (!dir_has_error) {
173 EXT4_ERROR_FILE(file, 0,
174 "directory contains a "
175 "hole at offset %llu",
176 (unsigned long long) ctx->pos);
179 /* corrupt size? Maybe no more blocks to read */
180 if (ctx->pos > inode->i_blocks << 9)
182 ctx->pos += sb->s_blocksize - offset;
186 /* Check the checksum */
187 if (!buffer_verified(bh) &&
188 !ext4_dirent_csum_verify(inode,
189 (struct ext4_dir_entry *)bh->b_data)) {
190 EXT4_ERROR_FILE(file, 0, "directory fails checksum "
192 (unsigned long long)ctx->pos);
193 ctx->pos += sb->s_blocksize - offset;
198 set_buffer_verified(bh);
200 /* If the dir block has changed since the last call to
201 * readdir(2), then we might be pointing to an invalid
202 * dirent right now. Scan from the start of the block
204 if (file->f_version != inode->i_version) {
205 for (i = 0; i < sb->s_blocksize && i < offset; ) {
206 de = (struct ext4_dir_entry_2 *)
208 /* It's too expensive to do a full
209 * dirent test each time round this
210 * loop, but we do have to test at
211 * least that it is non-zero. A
212 * failure will be detected in the
213 * dirent test below. */
214 if (ext4_rec_len_from_disk(de->rec_len,
215 sb->s_blocksize) < EXT4_DIR_REC_LEN(1))
217 i += ext4_rec_len_from_disk(de->rec_len,
221 ctx->pos = (ctx->pos & ~(sb->s_blocksize - 1))
223 file->f_version = inode->i_version;
226 while (ctx->pos < inode->i_size
227 && offset < sb->s_blocksize) {
228 de = (struct ext4_dir_entry_2 *) (bh->b_data + offset);
229 if (ext4_check_dir_entry(inode, file, de, bh,
230 bh->b_data, bh->b_size,
233 * On error, skip to the next block
235 ctx->pos = (ctx->pos |
236 (sb->s_blocksize - 1)) + 1;
239 offset += ext4_rec_len_from_disk(de->rec_len,
241 if (le32_to_cpu(de->inode)) {
242 if (enc_ctx == NULL) {
243 /* Directory is not encrypted */
244 if (!dir_emit(ctx, de->name,
246 le32_to_cpu(de->inode),
247 get_dtype(sb, de->file_type)))
250 /* Directory is encrypted */
251 err = ext4_fname_disk_to_usr(enc_ctx,
252 NULL, de, &fname_crypto_str);
256 fname_crypto_str.name, err,
257 le32_to_cpu(de->inode),
258 get_dtype(sb, de->file_type)))
262 ctx->pos += ext4_rec_len_from_disk(de->rec_len,
265 if ((ctx->pos < inode->i_size) && !dir_relax(inode))
274 #ifdef CONFIG_EXT4_FS_ENCRYPTION
275 ext4_put_fname_crypto_ctx(&enc_ctx);
276 ext4_fname_crypto_free_buffer(&fname_crypto_str);
282 static inline int is_32bit_api(void)
285 return is_compat_task();
287 return (BITS_PER_LONG == 32);
292 * These functions convert from the major/minor hash to an f_pos
293 * value for dx directories
295 * Upper layer (for example NFS) should specify FMODE_32BITHASH or
296 * FMODE_64BITHASH explicitly. On the other hand, we allow ext4 to be mounted
297 * directly on both 32-bit and 64-bit nodes, under such case, neither
298 * FMODE_32BITHASH nor FMODE_64BITHASH is specified.
300 static inline loff_t hash2pos(struct file *filp, __u32 major, __u32 minor)
302 if ((filp->f_mode & FMODE_32BITHASH) ||
303 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
306 return ((__u64)(major >> 1) << 32) | (__u64)minor;
309 static inline __u32 pos2maj_hash(struct file *filp, loff_t pos)
311 if ((filp->f_mode & FMODE_32BITHASH) ||
312 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
313 return (pos << 1) & 0xffffffff;
315 return ((pos >> 32) << 1) & 0xffffffff;
318 static inline __u32 pos2min_hash(struct file *filp, loff_t pos)
320 if ((filp->f_mode & FMODE_32BITHASH) ||
321 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
324 return pos & 0xffffffff;
328 * Return 32- or 64-bit end-of-file for dx directories
330 static inline loff_t ext4_get_htree_eof(struct file *filp)
332 if ((filp->f_mode & FMODE_32BITHASH) ||
333 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
334 return EXT4_HTREE_EOF_32BIT;
336 return EXT4_HTREE_EOF_64BIT;
341 * ext4_dir_llseek() calls generic_file_llseek_size to handle htree
342 * directories, where the "offset" is in terms of the filename hash
343 * value instead of the byte offset.
345 * Because we may return a 64-bit hash that is well beyond offset limits,
346 * we need to pass the max hash as the maximum allowable offset in
347 * the htree directory case.
349 * For non-htree, ext4_llseek already chooses the proper max offset.
351 static loff_t ext4_dir_llseek(struct file *file, loff_t offset, int whence)
353 struct inode *inode = file->f_mapping->host;
354 int dx_dir = is_dx_dir(inode);
355 loff_t htree_max = ext4_get_htree_eof(file);
358 return generic_file_llseek_size(file, offset, whence,
359 htree_max, htree_max);
361 return ext4_llseek(file, offset, whence);
365 * This structure holds the nodes of the red-black tree used to store
366 * the directory entry in hash order.
371 struct rb_node rb_hash;
380 * This functoin implements a non-recursive way of freeing all of the
381 * nodes in the red-black tree.
383 static void free_rb_tree_fname(struct rb_root *root)
385 struct fname *fname, *next;
387 rbtree_postorder_for_each_entry_safe(fname, next, root, rb_hash)
389 struct fname *old = fname;
398 static struct dir_private_info *ext4_htree_create_dir_info(struct file *filp,
401 struct dir_private_info *p;
403 p = kzalloc(sizeof(struct dir_private_info), GFP_KERNEL);
406 p->curr_hash = pos2maj_hash(filp, pos);
407 p->curr_minor_hash = pos2min_hash(filp, pos);
411 void ext4_htree_free_dir_info(struct dir_private_info *p)
413 free_rb_tree_fname(&p->root);
418 * Given a directory entry, enter it into the fname rb tree.
420 * When filename encryption is enabled, the dirent will hold the
421 * encrypted filename, while the htree will hold decrypted filename.
422 * The decrypted filename is passed in via ent_name. parameter.
424 int ext4_htree_store_dirent(struct file *dir_file, __u32 hash,
426 struct ext4_dir_entry_2 *dirent,
427 struct ext4_str *ent_name)
429 struct rb_node **p, *parent = NULL;
430 struct fname *fname, *new_fn;
431 struct dir_private_info *info;
434 info = dir_file->private_data;
435 p = &info->root.rb_node;
437 /* Create and allocate the fname structure */
438 len = sizeof(struct fname) + ent_name->len + 1;
439 new_fn = kzalloc(len, GFP_KERNEL);
443 new_fn->minor_hash = minor_hash;
444 new_fn->inode = le32_to_cpu(dirent->inode);
445 new_fn->name_len = ent_name->len;
446 new_fn->file_type = dirent->file_type;
447 memcpy(new_fn->name, ent_name->name, ent_name->len);
448 new_fn->name[ent_name->len] = 0;
452 fname = rb_entry(parent, struct fname, rb_hash);
455 * If the hash and minor hash match up, then we put
456 * them on a linked list. This rarely happens...
458 if ((new_fn->hash == fname->hash) &&
459 (new_fn->minor_hash == fname->minor_hash)) {
460 new_fn->next = fname->next;
461 fname->next = new_fn;
465 if (new_fn->hash < fname->hash)
467 else if (new_fn->hash > fname->hash)
469 else if (new_fn->minor_hash < fname->minor_hash)
471 else /* if (new_fn->minor_hash > fname->minor_hash) */
475 rb_link_node(&new_fn->rb_hash, parent, p);
476 rb_insert_color(&new_fn->rb_hash, &info->root);
483 * This is a helper function for ext4_dx_readdir. It calls filldir
484 * for all entres on the fname linked list. (Normally there is only
485 * one entry on the linked list, unless there are 62 bit hash collisions.)
487 static int call_filldir(struct file *file, struct dir_context *ctx,
490 struct dir_private_info *info = file->private_data;
491 struct inode *inode = file_inode(file);
492 struct super_block *sb = inode->i_sb;
495 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: comm %s: "
496 "called with null fname?!?", __func__, __LINE__,
497 inode->i_ino, current->comm);
500 ctx->pos = hash2pos(file, fname->hash, fname->minor_hash);
502 if (!dir_emit(ctx, fname->name,
505 get_dtype(sb, fname->file_type))) {
506 info->extra_fname = fname;
514 static int ext4_dx_readdir(struct file *file, struct dir_context *ctx)
516 struct dir_private_info *info = file->private_data;
517 struct inode *inode = file_inode(file);
522 info = ext4_htree_create_dir_info(file, ctx->pos);
525 file->private_data = info;
528 if (ctx->pos == ext4_get_htree_eof(file))
531 /* Some one has messed with f_pos; reset the world */
532 if (info->last_pos != ctx->pos) {
533 free_rb_tree_fname(&info->root);
534 info->curr_node = NULL;
535 info->extra_fname = NULL;
536 info->curr_hash = pos2maj_hash(file, ctx->pos);
537 info->curr_minor_hash = pos2min_hash(file, ctx->pos);
541 * If there are any leftover names on the hash collision
542 * chain, return them first.
544 if (info->extra_fname) {
545 if (call_filldir(file, ctx, info->extra_fname))
547 info->extra_fname = NULL;
549 } else if (!info->curr_node)
550 info->curr_node = rb_first(&info->root);
554 * Fill the rbtree if we have no more entries,
555 * or the inode has changed since we last read in the
558 if ((!info->curr_node) ||
559 (file->f_version != inode->i_version)) {
560 info->curr_node = NULL;
561 free_rb_tree_fname(&info->root);
562 file->f_version = inode->i_version;
563 ret = ext4_htree_fill_tree(file, info->curr_hash,
564 info->curr_minor_hash,
569 ctx->pos = ext4_get_htree_eof(file);
572 info->curr_node = rb_first(&info->root);
575 fname = rb_entry(info->curr_node, struct fname, rb_hash);
576 info->curr_hash = fname->hash;
577 info->curr_minor_hash = fname->minor_hash;
578 if (call_filldir(file, ctx, fname))
581 info->curr_node = rb_next(info->curr_node);
582 if (info->curr_node) {
583 fname = rb_entry(info->curr_node, struct fname,
585 info->curr_hash = fname->hash;
586 info->curr_minor_hash = fname->minor_hash;
588 if (info->next_hash == ~0) {
589 ctx->pos = ext4_get_htree_eof(file);
592 info->curr_hash = info->next_hash;
593 info->curr_minor_hash = 0;
597 info->last_pos = ctx->pos;
601 static int ext4_release_dir(struct inode *inode, struct file *filp)
603 if (filp->private_data)
604 ext4_htree_free_dir_info(filp->private_data);
609 int ext4_check_all_de(struct inode *dir, struct buffer_head *bh, void *buf,
612 struct ext4_dir_entry_2 *de;
614 unsigned int offset = 0;
617 de = (struct ext4_dir_entry_2 *)buf;
618 top = buf + buf_size;
619 while ((char *) de < top) {
620 if (ext4_check_dir_entry(dir, NULL, de, bh,
621 buf, buf_size, offset))
623 nlen = EXT4_DIR_REC_LEN(de->name_len);
624 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
625 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
628 if ((char *) de > top)
634 const struct file_operations ext4_dir_operations = {
635 .llseek = ext4_dir_llseek,
636 .read = generic_read_dir,
637 .iterate = ext4_readdir,
638 .unlocked_ioctl = ext4_ioctl,
640 .compat_ioctl = ext4_compat_ioctl,
642 .fsync = ext4_sync_file,
643 .release = ext4_release_dir,