blob: 832867aef3dca61d8d2376e9ca7e14a24dbafd69 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/ext3/dir.c
3 *
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)
8 *
9 * from
10 *
11 * linux/fs/minix/dir.c
12 *
13 * Copyright (C) 1991, 1992 Linus Torvalds
14 *
15 * ext3 directory handling functions
16 *
17 * Big-endian to little-endian byte-swapping/bitmaps by
18 * David S. Miller (davem@caip.rutgers.edu), 1995
19 *
20 * Hash Tree Directory indexing (c) 2001 Daniel Phillips
21 *
22 */
23
24#include <linux/fs.h>
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>
31
32static unsigned char ext3_filetype_table[] = {
33 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
34};
35
36static int ext3_readdir(struct file *, void *, filldir_t);
37static int ext3_dx_readdir(struct file * filp,
38 void * dirent, filldir_t filldir);
39static int ext3_release_dir (struct inode * inode,
40 struct file * filp);
41
42struct 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,
50#endif
51};
52
53
54static unsigned char get_dtype(struct super_block *sb, int filetype)
55{
56 if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_FILETYPE) ||
57 (filetype >= EXT3_FT_MAX))
58 return DT_UNKNOWN;
59
60 return (ext3_filetype_table[filetype]);
61}
62
63
64int ext3_check_dir_entry (const char * function, struct inode * dir,
65 struct ext3_dir_entry_2 * de,
66 struct buffer_head * bh,
67 unsigned long offset)
68{
69 const char * error_msg = NULL;
70 const int rlen = le16_to_cpu(de->rec_len);
71
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";
83
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),
90 rlen, de->name_len);
91 return error_msg == NULL ? 1 : 0;
92}
93
94static int ext3_readdir(struct file * filp,
95 void * dirent, filldir_t filldir)
96{
97 int error = 0;
98 unsigned long offset, blk;
99 int i, num, stored;
100 struct buffer_head * bh, * tmp, * bha[16];
101 struct ext3_dir_entry_2 * de;
102 struct super_block * sb;
103 int err;
104 struct inode *inode = filp->f_dentry->d_inode;
105 int ret = 0;
106
107 sb = inode->i_sb;
108
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) {
116 ret = err;
117 goto out;
118 }
119 /*
120 * We don't set the inode dirty flag since it's not
121 * critical that it get flushed back to the disk.
122 */
123 EXT3_I(filp->f_dentry->d_inode)->i_flags &= ~EXT3_INDEX_FL;
124 }
125#endif
126 stored = 0;
127 bh = NULL;
128 offset = filp->f_pos & (sb->s_blocksize - 1);
129
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);
133 if (!bh) {
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 filp->f_pos += sb->s_blocksize - offset;
138 continue;
139 }
140
141 /*
142 * Do the readahead
143 */
144 if (!offset) {
145 for (i = 16 >> (EXT3_BLOCK_SIZE_BITS(sb) - 9), num = 0;
146 i > 0; i--) {
147 tmp = ext3_getblk (NULL, inode, ++blk, 0, &err);
148 if (tmp && !buffer_uptodate(tmp) &&
149 !buffer_locked(tmp))
150 bha[num++] = tmp;
151 else
152 brelse (tmp);
153 }
154 if (num) {
155 ll_rw_block (READA, num, bha);
156 for (i = 0; i < num; i++)
157 brelse (bha[i]);
158 }
159 }
160
161revalidate:
162 /* If the dir block has changed since the last call to
163 * readdir(2), then we might be pointing to an invalid
164 * dirent right now. Scan from the start of the block
165 * to make sure. */
166 if (filp->f_version != inode->i_version) {
167 for (i = 0; i < sb->s_blocksize && i < offset; ) {
168 de = (struct ext3_dir_entry_2 *)
169 (bh->b_data + i);
170 /* It's too expensive to do a full
171 * dirent test each time round this
172 * loop, but we do have to test at
173 * least that it is non-zero. A
174 * failure will be detected in the
175 * dirent test below. */
176 if (le16_to_cpu(de->rec_len) <
177 EXT3_DIR_REC_LEN(1))
178 break;
179 i += le16_to_cpu(de->rec_len);
180 }
181 offset = i;
182 filp->f_pos = (filp->f_pos & ~(sb->s_blocksize - 1))
183 | offset;
184 filp->f_version = inode->i_version;
185 }
186
187 while (!error && filp->f_pos < inode->i_size
188 && offset < sb->s_blocksize) {
189 de = (struct ext3_dir_entry_2 *) (bh->b_data + offset);
190 if (!ext3_check_dir_entry ("ext3_readdir", inode, de,
191 bh, offset)) {
192 /* On error, skip the f_pos to the
193 next block. */
194 filp->f_pos = (filp->f_pos |
195 (sb->s_blocksize - 1)) + 1;
196 brelse (bh);
197 ret = stored;
198 goto out;
199 }
200 offset += le16_to_cpu(de->rec_len);
201 if (le32_to_cpu(de->inode)) {
202 /* We might block in the next section
203 * if the data destination is
204 * currently swapped out. So, use a
205 * version stamp to detect whether or
206 * not the directory has been modified
207 * during the copy operation.
208 */
209 unsigned long version = filp->f_version;
210
211 error = filldir(dirent, de->name,
212 de->name_len,
213 filp->f_pos,
214 le32_to_cpu(de->inode),
215 get_dtype(sb, de->file_type));
216 if (error)
217 break;
218 if (version != filp->f_version)
219 goto revalidate;
220 stored ++;
221 }
222 filp->f_pos += le16_to_cpu(de->rec_len);
223 }
224 offset = 0;
225 brelse (bh);
226 }
227out:
228 return ret;
229}
230
231#ifdef CONFIG_EXT3_INDEX
232/*
233 * These functions convert from the major/minor hash to an f_pos
234 * value.
235 *
236 * Currently we only use major hash numer. This is unfortunate, but
237 * on 32-bit machines, the same VFS interface is used for lseek and
238 * llseek, so if we use the 64 bit offset, then the 32-bit versions of
239 * lseek/telldir/seekdir will blow out spectacularly, and from within
240 * the ext2 low-level routine, we don't know if we're being called by
241 * a 64-bit version of the system call or the 32-bit version of the
242 * system call. Worse yet, NFSv2 only allows for a 32-bit readdir
243 * cookie. Sigh.
244 */
245#define hash2pos(major, minor) (major >> 1)
246#define pos2maj_hash(pos) ((pos << 1) & 0xffffffff)
247#define pos2min_hash(pos) (0)
248
249/*
250 * This structure holds the nodes of the red-black tree used to store
251 * the directory entry in hash order.
252 */
253struct fname {
254 __u32 hash;
255 __u32 minor_hash;
256 struct rb_node rb_hash;
257 struct fname *next;
258 __u32 inode;
259 __u8 name_len;
260 __u8 file_type;
261 char name[0];
262};
263
264/*
265 * This functoin implements a non-recursive way of freeing all of the
266 * nodes in the red-black tree.
267 */
268static void free_rb_tree_fname(struct rb_root *root)
269{
270 struct rb_node *n = root->rb_node;
271 struct rb_node *parent;
272 struct fname *fname;
273
274 while (n) {
275 /* Do the node's children first */
276 if ((n)->rb_left) {
277 n = n->rb_left;
278 continue;
279 }
280 if (n->rb_right) {
281 n = n->rb_right;
282 continue;
283 }
284 /*
285 * The node has no children; free it, and then zero
286 * out parent's link to it. Finally go to the
287 * beginning of the loop and try to free the parent
288 * node.
289 */
290 parent = n->rb_parent;
291 fname = rb_entry(n, struct fname, rb_hash);
292 while (fname) {
293 struct fname * old = fname;
294 fname = fname->next;
295 kfree (old);
296 }
297 if (!parent)
298 root->rb_node = NULL;
299 else if (parent->rb_left == n)
300 parent->rb_left = NULL;
301 else if (parent->rb_right == n)
302 parent->rb_right = NULL;
303 n = parent;
304 }
305 root->rb_node = NULL;
306}
307
308
309static struct dir_private_info *create_dir_info(loff_t pos)
310{
311 struct dir_private_info *p;
312
313 p = kmalloc(sizeof(struct dir_private_info), GFP_KERNEL);
314 if (!p)
315 return NULL;
316 p->root.rb_node = NULL;
317 p->curr_node = NULL;
318 p->extra_fname = NULL;
319 p->last_pos = 0;
320 p->curr_hash = pos2maj_hash(pos);
321 p->curr_minor_hash = pos2min_hash(pos);
322 p->next_hash = 0;
323 return p;
324}
325
326void ext3_htree_free_dir_info(struct dir_private_info *p)
327{
328 free_rb_tree_fname(&p->root);
329 kfree(p);
330}
331
332/*
333 * Given a directory entry, enter it into the fname rb tree.
334 */
335int ext3_htree_store_dirent(struct file *dir_file, __u32 hash,
336 __u32 minor_hash,
337 struct ext3_dir_entry_2 *dirent)
338{
339 struct rb_node **p, *parent = NULL;
340 struct fname * fname, *new_fn;
341 struct dir_private_info *info;
342 int len;
343
344 info = (struct dir_private_info *) dir_file->private_data;
345 p = &info->root.rb_node;
346
347 /* Create and allocate the fname structure */
348 len = sizeof(struct fname) + dirent->name_len + 1;
349 new_fn = kmalloc(len, GFP_KERNEL);
350 if (!new_fn)
351 return -ENOMEM;
352 memset(new_fn, 0, len);
353 new_fn->hash = hash;
354 new_fn->minor_hash = minor_hash;
355 new_fn->inode = le32_to_cpu(dirent->inode);
356 new_fn->name_len = dirent->name_len;
357 new_fn->file_type = dirent->file_type;
358 memcpy(new_fn->name, dirent->name, dirent->name_len);
359 new_fn->name[dirent->name_len] = 0;
360
361 while (*p) {
362 parent = *p;
363 fname = rb_entry(parent, struct fname, rb_hash);
364
365 /*
366 * If the hash and minor hash match up, then we put
367 * them on a linked list. This rarely happens...
368 */
369 if ((new_fn->hash == fname->hash) &&
370 (new_fn->minor_hash == fname->minor_hash)) {
371 new_fn->next = fname->next;
372 fname->next = new_fn;
373 return 0;
374 }
375
376 if (new_fn->hash < fname->hash)
377 p = &(*p)->rb_left;
378 else if (new_fn->hash > fname->hash)
379 p = &(*p)->rb_right;
380 else if (new_fn->minor_hash < fname->minor_hash)
381 p = &(*p)->rb_left;
382 else /* if (new_fn->minor_hash > fname->minor_hash) */
383 p = &(*p)->rb_right;
384 }
385
386 rb_link_node(&new_fn->rb_hash, parent, p);
387 rb_insert_color(&new_fn->rb_hash, &info->root);
388 return 0;
389}
390
391
392
393/*
394 * This is a helper function for ext3_dx_readdir. It calls filldir
395 * for all entres on the fname linked list. (Normally there is only
396 * one entry on the linked list, unless there are 62 bit hash collisions.)
397 */
398static int call_filldir(struct file * filp, void * dirent,
399 filldir_t filldir, struct fname *fname)
400{
401 struct dir_private_info *info = filp->private_data;
402 loff_t curr_pos;
403 struct inode *inode = filp->f_dentry->d_inode;
404 struct super_block * sb;
405 int error;
406
407 sb = inode->i_sb;
408
409 if (!fname) {
410 printk("call_filldir: called with null fname?!?\n");
411 return 0;
412 }
413 curr_pos = hash2pos(fname->hash, fname->minor_hash);
414 while (fname) {
415 error = filldir(dirent, fname->name,
416 fname->name_len, curr_pos,
417 fname->inode,
418 get_dtype(sb, fname->file_type));
419 if (error) {
420 filp->f_pos = curr_pos;
421 info->extra_fname = fname->next;
422 return error;
423 }
424 fname = fname->next;
425 }
426 return 0;
427}
428
429static int ext3_dx_readdir(struct file * filp,
430 void * dirent, filldir_t filldir)
431{
432 struct dir_private_info *info = filp->private_data;
433 struct inode *inode = filp->f_dentry->d_inode;
434 struct fname *fname;
435 int ret;
436
437 if (!info) {
438 info = create_dir_info(filp->f_pos);
439 if (!info)
440 return -ENOMEM;
441 filp->private_data = info;
442 }
443
444 if (filp->f_pos == EXT3_HTREE_EOF)
445 return 0; /* EOF */
446
447 /* Some one has messed with f_pos; reset the world */
448 if (info->last_pos != filp->f_pos) {
449 free_rb_tree_fname(&info->root);
450 info->curr_node = NULL;
451 info->extra_fname = NULL;
452 info->curr_hash = pos2maj_hash(filp->f_pos);
453 info->curr_minor_hash = pos2min_hash(filp->f_pos);
454 }
455
456 /*
457 * If there are any leftover names on the hash collision
458 * chain, return them first.
459 */
460 if (info->extra_fname &&
461 call_filldir(filp, dirent, filldir, info->extra_fname))
462 goto finished;
463
464 if (!info->curr_node)
465 info->curr_node = rb_first(&info->root);
466
467 while (1) {
468 /*
469 * Fill the rbtree if we have no more entries,
470 * or the inode has changed since we last read in the
471 * cached entries.
472 */
473 if ((!info->curr_node) ||
474 (filp->f_version != inode->i_version)) {
475 info->curr_node = NULL;
476 free_rb_tree_fname(&info->root);
477 filp->f_version = inode->i_version;
478 ret = ext3_htree_fill_tree(filp, info->curr_hash,
479 info->curr_minor_hash,
480 &info->next_hash);
481 if (ret < 0)
482 return ret;
483 if (ret == 0) {
484 filp->f_pos = EXT3_HTREE_EOF;
485 break;
486 }
487 info->curr_node = rb_first(&info->root);
488 }
489
490 fname = rb_entry(info->curr_node, struct fname, rb_hash);
491 info->curr_hash = fname->hash;
492 info->curr_minor_hash = fname->minor_hash;
493 if (call_filldir(filp, dirent, filldir, fname))
494 break;
495
496 info->curr_node = rb_next(info->curr_node);
497 if (!info->curr_node) {
498 if (info->next_hash == ~0) {
499 filp->f_pos = EXT3_HTREE_EOF;
500 break;
501 }
502 info->curr_hash = info->next_hash;
503 info->curr_minor_hash = 0;
504 }
505 }
506finished:
507 info->last_pos = filp->f_pos;
508 return 0;
509}
510
511static int ext3_release_dir (struct inode * inode, struct file * filp)
512{
513 if (filp->private_data)
514 ext3_htree_free_dir_info(filp->private_data);
515
516 return 0;
517}
518
519#endif