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 * ext3 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/jbd.h>
26 #include <linux/ext3_fs.h>
27 #include <linux/buffer_head.h>
28 #include <linux/smp_lock.h>
29 #include <linux/slab.h>
30 #include <linux/rbtree.h>
32 static unsigned char ext3_filetype_table[] = {
33 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
36 static int ext3_readdir(struct file *, void *, filldir_t);
37 static int ext3_dx_readdir(struct file * filp,
38 void * dirent, filldir_t filldir);
39 static int ext3_release_dir (struct inode * inode,
42 struct file_operations ext3_dir_operations = {
43 .llseek = generic_file_llseek,
44 .read = generic_read_dir,
45 .readdir = ext3_readdir, /* we take BKL. needed?*/
46 .ioctl = ext3_ioctl, /* BKL held */
47 .fsync = ext3_sync_file, /* BKL held */
48 #ifdef CONFIG_EXT3_INDEX
49 .release = ext3_release_dir,
54 static unsigned char get_dtype(struct super_block *sb, int filetype)
56 if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_FILETYPE) ||
57 (filetype >= EXT3_FT_MAX))
60 return (ext3_filetype_table[filetype]);
64 int ext3_check_dir_entry (const char * function, struct inode * dir,
65 struct ext3_dir_entry_2 * de,
66 struct buffer_head * bh,
69 const char * error_msg = NULL;
70 const int rlen = le16_to_cpu(de->rec_len);
72 if (rlen < EXT3_DIR_REC_LEN(1))
73 error_msg = "rec_len is smaller than minimal";
74 else if (rlen % 4 != 0)
75 error_msg = "rec_len % 4 != 0";
76 else if (rlen < EXT3_DIR_REC_LEN(de->name_len))
77 error_msg = "rec_len is too small for name_len";
78 else if (((char *) de - bh->b_data) + rlen > dir->i_sb->s_blocksize)
79 error_msg = "directory entry across blocks";
80 else if (le32_to_cpu(de->inode) >
81 le32_to_cpu(EXT3_SB(dir->i_sb)->s_es->s_inodes_count))
82 error_msg = "inode out of bounds";
84 if (error_msg != NULL)
85 ext3_error (dir->i_sb, function,
86 "bad entry in directory #%lu: %s - "
87 "offset=%lu, inode=%lu, rec_len=%d, name_len=%d",
88 dir->i_ino, error_msg, offset,
89 (unsigned long) le32_to_cpu(de->inode),
91 return error_msg == NULL ? 1 : 0;
94 static int ext3_readdir(struct file * filp,
95 void * dirent, filldir_t filldir)
98 unsigned long offset, blk;
100 struct buffer_head * bh, * tmp, * bha[16];
101 struct ext3_dir_entry_2 * de;
102 struct super_block * sb;
104 struct inode *inode = filp->f_dentry->d_inode;
109 #ifdef CONFIG_EXT3_INDEX
110 if (EXT3_HAS_COMPAT_FEATURE(inode->i_sb,
111 EXT3_FEATURE_COMPAT_DIR_INDEX) &&
112 ((EXT3_I(inode)->i_flags & EXT3_INDEX_FL) ||
113 ((inode->i_size >> sb->s_blocksize_bits) == 1))) {
114 err = ext3_dx_readdir(filp, dirent, filldir);
115 if (err != ERR_BAD_DX_DIR) {
120 * We don't set the inode dirty flag since it's not
121 * critical that it get flushed back to the disk.
123 EXT3_I(filp->f_dentry->d_inode)->i_flags &= ~EXT3_INDEX_FL;
128 offset = filp->f_pos & (sb->s_blocksize - 1);
130 while (!error && !stored && filp->f_pos < inode->i_size) {
131 blk = (filp->f_pos) >> EXT3_BLOCK_SIZE_BITS(sb);
132 bh = ext3_bread(NULL, inode, blk, 0, &err);
134 ext3_error (sb, "ext3_readdir",
135 "directory #%lu contains a hole at offset %lu",
136 inode->i_ino, (unsigned long)filp->f_pos);
137 /* corrupt size? Maybe no more blocks to read */
138 if (filp->f_pos > inode->i_blocks << 9)
140 filp->f_pos += sb->s_blocksize - offset;
148 for (i = 16 >> (EXT3_BLOCK_SIZE_BITS(sb) - 9), num = 0;
150 tmp = ext3_getblk (NULL, inode, ++blk, 0, &err);
151 if (tmp && !buffer_uptodate(tmp) &&
158 ll_rw_block (READA, num, bha);
159 for (i = 0; i < num; i++)
165 /* If the dir block has changed since the last call to
166 * readdir(2), then we might be pointing to an invalid
167 * dirent right now. Scan from the start of the block
169 if (filp->f_version != inode->i_version) {
170 for (i = 0; i < sb->s_blocksize && i < offset; ) {
171 de = (struct ext3_dir_entry_2 *)
173 /* It's too expensive to do a full
174 * dirent test each time round this
175 * loop, but we do have to test at
176 * least that it is non-zero. A
177 * failure will be detected in the
178 * dirent test below. */
179 if (le16_to_cpu(de->rec_len) <
182 i += le16_to_cpu(de->rec_len);
185 filp->f_pos = (filp->f_pos & ~(sb->s_blocksize - 1))
187 filp->f_version = inode->i_version;
190 while (!error && filp->f_pos < inode->i_size
191 && offset < sb->s_blocksize) {
192 de = (struct ext3_dir_entry_2 *) (bh->b_data + offset);
193 if (!ext3_check_dir_entry ("ext3_readdir", inode, de,
195 /* On error, skip the f_pos to the
197 filp->f_pos = (filp->f_pos |
198 (sb->s_blocksize - 1)) + 1;
203 offset += le16_to_cpu(de->rec_len);
204 if (le32_to_cpu(de->inode)) {
205 /* We might block in the next section
206 * if the data destination is
207 * currently swapped out. So, use a
208 * version stamp to detect whether or
209 * not the directory has been modified
210 * during the copy operation.
212 unsigned long version = filp->f_version;
214 error = filldir(dirent, de->name,
217 le32_to_cpu(de->inode),
218 get_dtype(sb, de->file_type));
221 if (version != filp->f_version)
225 filp->f_pos += le16_to_cpu(de->rec_len);
234 #ifdef CONFIG_EXT3_INDEX
236 * These functions convert from the major/minor hash to an f_pos
239 * Currently we only use major hash numer. This is unfortunate, but
240 * on 32-bit machines, the same VFS interface is used for lseek and
241 * llseek, so if we use the 64 bit offset, then the 32-bit versions of
242 * lseek/telldir/seekdir will blow out spectacularly, and from within
243 * the ext2 low-level routine, we don't know if we're being called by
244 * a 64-bit version of the system call or the 32-bit version of the
245 * system call. Worse yet, NFSv2 only allows for a 32-bit readdir
248 #define hash2pos(major, minor) (major >> 1)
249 #define pos2maj_hash(pos) ((pos << 1) & 0xffffffff)
250 #define pos2min_hash(pos) (0)
253 * This structure holds the nodes of the red-black tree used to store
254 * the directory entry in hash order.
259 struct rb_node rb_hash;
268 * This functoin implements a non-recursive way of freeing all of the
269 * nodes in the red-black tree.
271 static void free_rb_tree_fname(struct rb_root *root)
273 struct rb_node *n = root->rb_node;
274 struct rb_node *parent;
278 /* Do the node's children first */
288 * The node has no children; free it, and then zero
289 * out parent's link to it. Finally go to the
290 * beginning of the loop and try to free the parent
293 parent = n->rb_parent;
294 fname = rb_entry(n, struct fname, rb_hash);
296 struct fname * old = fname;
301 root->rb_node = NULL;
302 else if (parent->rb_left == n)
303 parent->rb_left = NULL;
304 else if (parent->rb_right == n)
305 parent->rb_right = NULL;
308 root->rb_node = NULL;
312 static struct dir_private_info *create_dir_info(loff_t pos)
314 struct dir_private_info *p;
316 p = kmalloc(sizeof(struct dir_private_info), GFP_KERNEL);
319 p->root.rb_node = NULL;
321 p->extra_fname = NULL;
323 p->curr_hash = pos2maj_hash(pos);
324 p->curr_minor_hash = pos2min_hash(pos);
329 void ext3_htree_free_dir_info(struct dir_private_info *p)
331 free_rb_tree_fname(&p->root);
336 * Given a directory entry, enter it into the fname rb tree.
338 int ext3_htree_store_dirent(struct file *dir_file, __u32 hash,
340 struct ext3_dir_entry_2 *dirent)
342 struct rb_node **p, *parent = NULL;
343 struct fname * fname, *new_fn;
344 struct dir_private_info *info;
347 info = (struct dir_private_info *) dir_file->private_data;
348 p = &info->root.rb_node;
350 /* Create and allocate the fname structure */
351 len = sizeof(struct fname) + dirent->name_len + 1;
352 new_fn = kmalloc(len, GFP_KERNEL);
355 memset(new_fn, 0, len);
357 new_fn->minor_hash = minor_hash;
358 new_fn->inode = le32_to_cpu(dirent->inode);
359 new_fn->name_len = dirent->name_len;
360 new_fn->file_type = dirent->file_type;
361 memcpy(new_fn->name, dirent->name, dirent->name_len);
362 new_fn->name[dirent->name_len] = 0;
366 fname = rb_entry(parent, struct fname, rb_hash);
369 * If the hash and minor hash match up, then we put
370 * them on a linked list. This rarely happens...
372 if ((new_fn->hash == fname->hash) &&
373 (new_fn->minor_hash == fname->minor_hash)) {
374 new_fn->next = fname->next;
375 fname->next = new_fn;
379 if (new_fn->hash < fname->hash)
381 else if (new_fn->hash > fname->hash)
383 else if (new_fn->minor_hash < fname->minor_hash)
385 else /* if (new_fn->minor_hash > fname->minor_hash) */
389 rb_link_node(&new_fn->rb_hash, parent, p);
390 rb_insert_color(&new_fn->rb_hash, &info->root);
397 * This is a helper function for ext3_dx_readdir. It calls filldir
398 * for all entres on the fname linked list. (Normally there is only
399 * one entry on the linked list, unless there are 62 bit hash collisions.)
401 static int call_filldir(struct file * filp, void * dirent,
402 filldir_t filldir, struct fname *fname)
404 struct dir_private_info *info = filp->private_data;
406 struct inode *inode = filp->f_dentry->d_inode;
407 struct super_block * sb;
413 printk("call_filldir: called with null fname?!?\n");
416 curr_pos = hash2pos(fname->hash, fname->minor_hash);
418 error = filldir(dirent, fname->name,
419 fname->name_len, curr_pos,
421 get_dtype(sb, fname->file_type));
423 filp->f_pos = curr_pos;
424 info->extra_fname = fname->next;
432 static int ext3_dx_readdir(struct file * filp,
433 void * dirent, filldir_t filldir)
435 struct dir_private_info *info = filp->private_data;
436 struct inode *inode = filp->f_dentry->d_inode;
441 info = create_dir_info(filp->f_pos);
444 filp->private_data = info;
447 if (filp->f_pos == EXT3_HTREE_EOF)
450 /* Some one has messed with f_pos; reset the world */
451 if (info->last_pos != filp->f_pos) {
452 free_rb_tree_fname(&info->root);
453 info->curr_node = NULL;
454 info->extra_fname = NULL;
455 info->curr_hash = pos2maj_hash(filp->f_pos);
456 info->curr_minor_hash = pos2min_hash(filp->f_pos);
460 * If there are any leftover names on the hash collision
461 * chain, return them first.
463 if (info->extra_fname &&
464 call_filldir(filp, dirent, filldir, info->extra_fname))
467 if (!info->curr_node)
468 info->curr_node = rb_first(&info->root);
472 * Fill the rbtree if we have no more entries,
473 * or the inode has changed since we last read in the
476 if ((!info->curr_node) ||
477 (filp->f_version != inode->i_version)) {
478 info->curr_node = NULL;
479 free_rb_tree_fname(&info->root);
480 filp->f_version = inode->i_version;
481 ret = ext3_htree_fill_tree(filp, info->curr_hash,
482 info->curr_minor_hash,
487 filp->f_pos = EXT3_HTREE_EOF;
490 info->curr_node = rb_first(&info->root);
493 fname = rb_entry(info->curr_node, struct fname, rb_hash);
494 info->curr_hash = fname->hash;
495 info->curr_minor_hash = fname->minor_hash;
496 if (call_filldir(filp, dirent, filldir, fname))
499 info->curr_node = rb_next(info->curr_node);
500 if (!info->curr_node) {
501 if (info->next_hash == ~0) {
502 filp->f_pos = EXT3_HTREE_EOF;
505 info->curr_hash = info->next_hash;
506 info->curr_minor_hash = 0;
510 info->last_pos = filp->f_pos;
514 static int ext3_release_dir (struct inode * inode, struct file * filp)
516 if (filp->private_data)
517 ext3_htree_free_dir_info(filp->private_data);