773459164bb2cbd1b1641fd367b563d252ad3016
[linux-2.6.git] / fs / ext3 / dir.c
1 /*
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
32 static unsigned char ext3_filetype_table[] = {
33         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
34 };
35
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,
40                                 struct file * filp);
41
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,
50 #endif
51 };
52
53
54 static 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
64 int 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
94 static int ext3_readdir(struct file * filp,
95                          void * dirent, filldir_t filldir)
96 {
97         int error = 0;
98         unsigned long offset;
99         int i, stored;
100         struct ext3_dir_entry_2 *de;
101         struct super_block *sb;
102         int err;
103         struct inode *inode = filp->f_dentry->d_inode;
104         int ret = 0;
105
106         sb = inode->i_sb;
107
108 #ifdef CONFIG_EXT3_INDEX
109         if (EXT3_HAS_COMPAT_FEATURE(inode->i_sb,
110                                     EXT3_FEATURE_COMPAT_DIR_INDEX) &&
111             ((EXT3_I(inode)->i_flags & EXT3_INDEX_FL) ||
112              ((inode->i_size >> sb->s_blocksize_bits) == 1))) {
113                 err = ext3_dx_readdir(filp, dirent, filldir);
114                 if (err != ERR_BAD_DX_DIR) {
115                         ret = err;
116                         goto out;
117                 }
118                 /*
119                  * We don't set the inode dirty flag since it's not
120                  * critical that it get flushed back to the disk.
121                  */
122                 EXT3_I(filp->f_dentry->d_inode)->i_flags &= ~EXT3_INDEX_FL;
123         }
124 #endif
125         stored = 0;
126         offset = filp->f_pos & (sb->s_blocksize - 1);
127
128         while (!error && !stored && filp->f_pos < inode->i_size) {
129                 unsigned long blk = filp->f_pos >> EXT3_BLOCK_SIZE_BITS(sb);
130                 struct buffer_head map_bh;
131                 struct buffer_head *bh = NULL;
132
133                 map_bh.b_state = 0;
134                 err = ext3_get_block_handle(NULL, inode, blk, &map_bh, 0, 0);
135                 if (!err) {
136                         page_cache_readahead(sb->s_bdev->bd_inode->i_mapping,
137                                 &filp->f_ra,
138                                 filp,
139                                 map_bh.b_blocknr >>
140                                         (PAGE_CACHE_SHIFT - inode->i_blkbits),
141                                 1);
142                         bh = ext3_bread(NULL, inode, blk, 0, &err);
143                 }
144
145                 /*
146                  * We ignore I/O errors on directories so users have a chance
147                  * of recovering data when there's a bad sector
148                  */
149                 if (!bh) {
150                         ext3_error (sb, "ext3_readdir",
151                                 "directory #%lu contains a hole at offset %lu",
152                                 inode->i_ino, (unsigned long)filp->f_pos);
153                         filp->f_pos += sb->s_blocksize - offset;
154                         continue;
155                 }
156
157 revalidate:
158                 /* If the dir block has changed since the last call to
159                  * readdir(2), then we might be pointing to an invalid
160                  * dirent right now.  Scan from the start of the block
161                  * to make sure. */
162                 if (filp->f_version != inode->i_version) {
163                         for (i = 0; i < sb->s_blocksize && i < offset; ) {
164                                 de = (struct ext3_dir_entry_2 *) 
165                                         (bh->b_data + i);
166                                 /* It's too expensive to do a full
167                                  * dirent test each time round this
168                                  * loop, but we do have to test at
169                                  * least that it is non-zero.  A
170                                  * failure will be detected in the
171                                  * dirent test below. */
172                                 if (le16_to_cpu(de->rec_len) <
173                                                 EXT3_DIR_REC_LEN(1))
174                                         break;
175                                 i += le16_to_cpu(de->rec_len);
176                         }
177                         offset = i;
178                         filp->f_pos = (filp->f_pos & ~(sb->s_blocksize - 1))
179                                 | offset;
180                         filp->f_version = inode->i_version;
181                 }
182
183                 while (!error && filp->f_pos < inode->i_size 
184                        && offset < sb->s_blocksize) {
185                         de = (struct ext3_dir_entry_2 *) (bh->b_data + offset);
186                         if (!ext3_check_dir_entry ("ext3_readdir", inode, de,
187                                                    bh, offset)) {
188                                 /* On error, skip the f_pos to the
189                                    next block. */
190                                 filp->f_pos = (filp->f_pos |
191                                                 (sb->s_blocksize - 1)) + 1;
192                                 brelse (bh);
193                                 ret = stored;
194                                 goto out;
195                         }
196                         offset += le16_to_cpu(de->rec_len);
197                         if (le32_to_cpu(de->inode)) {
198                                 /* We might block in the next section
199                                  * if the data destination is
200                                  * currently swapped out.  So, use a
201                                  * version stamp to detect whether or
202                                  * not the directory has been modified
203                                  * during the copy operation.
204                                  */
205                                 unsigned long version = filp->f_version;
206
207                                 error = filldir(dirent, de->name,
208                                                 de->name_len,
209                                                 filp->f_pos,
210                                                 le32_to_cpu(de->inode),
211                                                 get_dtype(sb, de->file_type));
212                                 if (error)
213                                         break;
214                                 if (version != filp->f_version)
215                                         goto revalidate;
216                                 stored ++;
217                         }
218                         filp->f_pos += le16_to_cpu(de->rec_len);
219                 }
220                 offset = 0;
221                 brelse (bh);
222         }
223 out:
224         return ret;
225 }
226
227 #ifdef CONFIG_EXT3_INDEX
228 /*
229  * These functions convert from the major/minor hash to an f_pos
230  * value.
231  * 
232  * Currently we only use major hash numer.  This is unfortunate, but
233  * on 32-bit machines, the same VFS interface is used for lseek and
234  * llseek, so if we use the 64 bit offset, then the 32-bit versions of
235  * lseek/telldir/seekdir will blow out spectacularly, and from within
236  * the ext2 low-level routine, we don't know if we're being called by
237  * a 64-bit version of the system call or the 32-bit version of the
238  * system call.  Worse yet, NFSv2 only allows for a 32-bit readdir
239  * cookie.  Sigh.
240  */
241 #define hash2pos(major, minor)  (major >> 1)
242 #define pos2maj_hash(pos)       ((pos << 1) & 0xffffffff)
243 #define pos2min_hash(pos)       (0)
244
245 /*
246  * This structure holds the nodes of the red-black tree used to store
247  * the directory entry in hash order.
248  */
249 struct fname {
250         __u32           hash;
251         __u32           minor_hash;
252         struct rb_node  rb_hash; 
253         struct fname    *next;
254         __u32           inode;
255         __u8            name_len;
256         __u8            file_type;
257         char            name[0];
258 };
259
260 /*
261  * This functoin implements a non-recursive way of freeing all of the
262  * nodes in the red-black tree.
263  */
264 static void free_rb_tree_fname(struct rb_root *root)
265 {
266         struct rb_node  *n = root->rb_node;
267         struct rb_node  *parent;
268         struct fname    *fname;
269
270         while (n) {
271                 /* Do the node's children first */
272                 if ((n)->rb_left) {
273                         n = n->rb_left;
274                         continue;
275                 }
276                 if (n->rb_right) {
277                         n = n->rb_right;
278                         continue;
279                 }
280                 /*
281                  * The node has no children; free it, and then zero
282                  * out parent's link to it.  Finally go to the
283                  * beginning of the loop and try to free the parent
284                  * node.
285                  */
286                 parent = n->rb_parent;
287                 fname = rb_entry(n, struct fname, rb_hash);
288                 while (fname) {
289                         struct fname * old = fname;
290                         fname = fname->next;
291                         kfree (old);
292                 }
293                 if (!parent)
294                         root->rb_node = NULL;
295                 else if (parent->rb_left == n)
296                         parent->rb_left = NULL;
297                 else if (parent->rb_right == n)
298                         parent->rb_right = NULL;
299                 n = parent;
300         }
301         root->rb_node = NULL;
302 }
303
304
305 static struct dir_private_info *create_dir_info(loff_t pos)
306 {
307         struct dir_private_info *p;
308
309         p = kmalloc(sizeof(struct dir_private_info), GFP_KERNEL);
310         if (!p)
311                 return NULL;
312         p->root.rb_node = NULL;
313         p->curr_node = NULL;
314         p->extra_fname = NULL;
315         p->last_pos = 0;
316         p->curr_hash = pos2maj_hash(pos);
317         p->curr_minor_hash = pos2min_hash(pos);
318         p->next_hash = 0;
319         return p;
320 }
321
322 void ext3_htree_free_dir_info(struct dir_private_info *p)
323 {
324         free_rb_tree_fname(&p->root);
325         kfree(p);
326 }
327
328 /*
329  * Given a directory entry, enter it into the fname rb tree.
330  */
331 int ext3_htree_store_dirent(struct file *dir_file, __u32 hash,
332                              __u32 minor_hash,
333                              struct ext3_dir_entry_2 *dirent)
334 {
335         struct rb_node **p, *parent = NULL;
336         struct fname * fname, *new_fn;
337         struct dir_private_info *info;
338         int len;
339
340         info = (struct dir_private_info *) dir_file->private_data;
341         p = &info->root.rb_node;
342
343         /* Create and allocate the fname structure */
344         len = sizeof(struct fname) + dirent->name_len + 1;
345         new_fn = kmalloc(len, GFP_KERNEL);
346         if (!new_fn)
347                 return -ENOMEM;
348         memset(new_fn, 0, len);
349         new_fn->hash = hash;
350         new_fn->minor_hash = minor_hash;
351         new_fn->inode = le32_to_cpu(dirent->inode);
352         new_fn->name_len = dirent->name_len;
353         new_fn->file_type = dirent->file_type;
354         memcpy(new_fn->name, dirent->name, dirent->name_len);
355         new_fn->name[dirent->name_len] = 0;
356
357         while (*p) {
358                 parent = *p;
359                 fname = rb_entry(parent, struct fname, rb_hash);
360
361                 /*
362                  * If the hash and minor hash match up, then we put
363                  * them on a linked list.  This rarely happens...
364                  */
365                 if ((new_fn->hash == fname->hash) &&
366                     (new_fn->minor_hash == fname->minor_hash)) {
367                         new_fn->next = fname->next;
368                         fname->next = new_fn;
369                         return 0;
370                 }
371
372                 if (new_fn->hash < fname->hash)
373                         p = &(*p)->rb_left;
374                 else if (new_fn->hash > fname->hash)
375                         p = &(*p)->rb_right;
376                 else if (new_fn->minor_hash < fname->minor_hash)
377                         p = &(*p)->rb_left;
378                 else /* if (new_fn->minor_hash > fname->minor_hash) */
379                         p = &(*p)->rb_right;
380         }
381
382         rb_link_node(&new_fn->rb_hash, parent, p);
383         rb_insert_color(&new_fn->rb_hash, &info->root);
384         return 0;
385 }
386
387
388
389 /*
390  * This is a helper function for ext3_dx_readdir.  It calls filldir
391  * for all entres on the fname linked list.  (Normally there is only
392  * one entry on the linked list, unless there are 62 bit hash collisions.)
393  */
394 static int call_filldir(struct file * filp, void * dirent,
395                         filldir_t filldir, struct fname *fname)
396 {
397         struct dir_private_info *info = filp->private_data;
398         loff_t  curr_pos;
399         struct inode *inode = filp->f_dentry->d_inode;
400         struct super_block * sb;
401         int error;
402
403         sb = inode->i_sb;
404
405         if (!fname) {
406                 printk("call_filldir: called with null fname?!?\n");
407                 return 0;
408         }
409         curr_pos = hash2pos(fname->hash, fname->minor_hash);
410         while (fname) {
411                 error = filldir(dirent, fname->name,
412                                 fname->name_len, curr_pos, 
413                                 fname->inode,
414                                 get_dtype(sb, fname->file_type));
415                 if (error) {
416                         filp->f_pos = curr_pos;
417                         info->extra_fname = fname->next;
418                         return error;
419                 }
420                 fname = fname->next;
421         }
422         return 0;
423 }
424
425 static int ext3_dx_readdir(struct file * filp,
426                          void * dirent, filldir_t filldir)
427 {
428         struct dir_private_info *info = filp->private_data;
429         struct inode *inode = filp->f_dentry->d_inode;
430         struct fname *fname;
431         int     ret;
432
433         if (!info) {
434                 info = create_dir_info(filp->f_pos);
435                 if (!info)
436                         return -ENOMEM;
437                 filp->private_data = info;
438         }
439
440         if (filp->f_pos == EXT3_HTREE_EOF)
441                 return 0;       /* EOF */
442
443         /* Some one has messed with f_pos; reset the world */
444         if (info->last_pos != filp->f_pos) {
445                 free_rb_tree_fname(&info->root);
446                 info->curr_node = NULL;
447                 info->extra_fname = NULL;
448                 info->curr_hash = pos2maj_hash(filp->f_pos);
449                 info->curr_minor_hash = pos2min_hash(filp->f_pos);
450         }
451
452         /*
453          * If there are any leftover names on the hash collision
454          * chain, return them first.
455          */
456         if (info->extra_fname &&
457             call_filldir(filp, dirent, filldir, info->extra_fname))
458                 goto finished;
459
460         if (!info->curr_node)
461                 info->curr_node = rb_first(&info->root);
462
463         while (1) {
464                 /*
465                  * Fill the rbtree if we have no more entries,
466                  * or the inode has changed since we last read in the
467                  * cached entries. 
468                  */
469                 if ((!info->curr_node) ||
470                     (filp->f_version != inode->i_version)) {
471                         info->curr_node = NULL;
472                         free_rb_tree_fname(&info->root);
473                         filp->f_version = inode->i_version;
474                         ret = ext3_htree_fill_tree(filp, info->curr_hash,
475                                                    info->curr_minor_hash,
476                                                    &info->next_hash);
477                         if (ret < 0)
478                                 return ret;
479                         if (ret == 0) {
480                                 filp->f_pos = EXT3_HTREE_EOF;
481                                 break;
482                         }
483                         info->curr_node = rb_first(&info->root);
484                 }
485
486                 fname = rb_entry(info->curr_node, struct fname, rb_hash);
487                 info->curr_hash = fname->hash;
488                 info->curr_minor_hash = fname->minor_hash;
489                 if (call_filldir(filp, dirent, filldir, fname))
490                         break;
491
492                 info->curr_node = rb_next(info->curr_node);
493                 if (!info->curr_node) {
494                         if (info->next_hash == ~0) {
495                                 filp->f_pos = EXT3_HTREE_EOF;
496                                 break;
497                         }
498                         info->curr_hash = info->next_hash;
499                         info->curr_minor_hash = 0;
500                 }
501         }
502 finished:
503         info->last_pos = filp->f_pos;
504         return 0;
505 }
506
507 static int ext3_release_dir (struct inode * inode, struct file * filp)
508 {
509        if (filp->private_data)
510                 ext3_htree_free_dir_info(filp->private_data);
511
512         return 0;
513 }
514
515 #endif