Merge branch 'for-linus' of git://one.firstfloor.org/home/andi/git/linux-2.6
[linux-2.6] / 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 const 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 #ifdef CONFIG_COMPAT
48         .compat_ioctl   = ext3_compat_ioctl,
49 #endif
50         .fsync          = ext3_sync_file,       /* BKL held */
51 #ifdef CONFIG_EXT3_INDEX
52         .release        = ext3_release_dir,
53 #endif
54 };
55
56
57 static unsigned char get_dtype(struct super_block *sb, int filetype)
58 {
59         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_FILETYPE) ||
60             (filetype >= EXT3_FT_MAX))
61                 return DT_UNKNOWN;
62
63         return (ext3_filetype_table[filetype]);
64 }
65
66
67 int ext3_check_dir_entry (const char * function, struct inode * dir,
68                           struct ext3_dir_entry_2 * de,
69                           struct buffer_head * bh,
70                           unsigned long offset)
71 {
72         const char * error_msg = NULL;
73         const int rlen = le16_to_cpu(de->rec_len);
74
75         if (rlen < EXT3_DIR_REC_LEN(1))
76                 error_msg = "rec_len is smaller than minimal";
77         else if (rlen % 4 != 0)
78                 error_msg = "rec_len % 4 != 0";
79         else if (rlen < EXT3_DIR_REC_LEN(de->name_len))
80                 error_msg = "rec_len is too small for name_len";
81         else if (((char *) de - bh->b_data) + rlen > dir->i_sb->s_blocksize)
82                 error_msg = "directory entry across blocks";
83         else if (le32_to_cpu(de->inode) >
84                         le32_to_cpu(EXT3_SB(dir->i_sb)->s_es->s_inodes_count))
85                 error_msg = "inode out of bounds";
86
87         if (error_msg != NULL)
88                 ext3_error (dir->i_sb, function,
89                         "bad entry in directory #%lu: %s - "
90                         "offset=%lu, inode=%lu, rec_len=%d, name_len=%d",
91                         dir->i_ino, error_msg, offset,
92                         (unsigned long) le32_to_cpu(de->inode),
93                         rlen, de->name_len);
94         return error_msg == NULL ? 1 : 0;
95 }
96
97 static int ext3_readdir(struct file * filp,
98                          void * dirent, filldir_t filldir)
99 {
100         int error = 0;
101         unsigned long offset;
102         int i, stored;
103         struct ext3_dir_entry_2 *de;
104         struct super_block *sb;
105         int err;
106         struct inode *inode = filp->f_dentry->d_inode;
107         int ret = 0;
108
109         sb = inode->i_sb;
110
111 #ifdef CONFIG_EXT3_INDEX
112         if (EXT3_HAS_COMPAT_FEATURE(inode->i_sb,
113                                     EXT3_FEATURE_COMPAT_DIR_INDEX) &&
114             ((EXT3_I(inode)->i_flags & EXT3_INDEX_FL) ||
115              ((inode->i_size >> sb->s_blocksize_bits) == 1))) {
116                 err = ext3_dx_readdir(filp, dirent, filldir);
117                 if (err != ERR_BAD_DX_DIR) {
118                         ret = err;
119                         goto out;
120                 }
121                 /*
122                  * We don't set the inode dirty flag since it's not
123                  * critical that it get flushed back to the disk.
124                  */
125                 EXT3_I(filp->f_dentry->d_inode)->i_flags &= ~EXT3_INDEX_FL;
126         }
127 #endif
128         stored = 0;
129         offset = filp->f_pos & (sb->s_blocksize - 1);
130
131         while (!error && !stored && filp->f_pos < inode->i_size) {
132                 unsigned long blk = filp->f_pos >> EXT3_BLOCK_SIZE_BITS(sb);
133                 struct buffer_head map_bh;
134                 struct buffer_head *bh = NULL;
135
136                 map_bh.b_state = 0;
137                 err = ext3_get_blocks_handle(NULL, inode, blk, 1,
138                                                 &map_bh, 0, 0);
139                 if (err > 0) {
140                         page_cache_readahead(sb->s_bdev->bd_inode->i_mapping,
141                                 &filp->f_ra,
142                                 filp,
143                                 map_bh.b_blocknr >>
144                                         (PAGE_CACHE_SHIFT - inode->i_blkbits),
145                                 1);
146                         bh = ext3_bread(NULL, inode, blk, 0, &err);
147                 }
148
149                 /*
150                  * We ignore I/O errors on directories so users have a chance
151                  * of recovering data when there's a bad sector
152                  */
153                 if (!bh) {
154                         ext3_error (sb, "ext3_readdir",
155                                 "directory #%lu contains a hole at offset %lu",
156                                 inode->i_ino, (unsigned long)filp->f_pos);
157                         /* corrupt size?  Maybe no more blocks to read */
158                         if (filp->f_pos > inode->i_blocks << 9)
159                                 break;
160                         filp->f_pos += sb->s_blocksize - offset;
161                         continue;
162                 }
163
164 revalidate:
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
168                  * to make sure. */
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 *)
172                                         (bh->b_data + i);
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) <
180                                                 EXT3_DIR_REC_LEN(1))
181                                         break;
182                                 i += le16_to_cpu(de->rec_len);
183                         }
184                         offset = i;
185                         filp->f_pos = (filp->f_pos & ~(sb->s_blocksize - 1))
186                                 | offset;
187                         filp->f_version = inode->i_version;
188                 }
189
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,
194                                                    bh, offset)) {
195                                 /* On error, skip the f_pos to the
196                                    next block. */
197                                 filp->f_pos = (filp->f_pos |
198                                                 (sb->s_blocksize - 1)) + 1;
199                                 brelse (bh);
200                                 ret = stored;
201                                 goto out;
202                         }
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.
211                                  */
212                                 unsigned long version = filp->f_version;
213
214                                 error = filldir(dirent, de->name,
215                                                 de->name_len,
216                                                 filp->f_pos,
217                                                 le32_to_cpu(de->inode),
218                                                 get_dtype(sb, de->file_type));
219                                 if (error)
220                                         break;
221                                 if (version != filp->f_version)
222                                         goto revalidate;
223                                 stored ++;
224                         }
225                         filp->f_pos += le16_to_cpu(de->rec_len);
226                 }
227                 offset = 0;
228                 brelse (bh);
229         }
230 out:
231         return ret;
232 }
233
234 #ifdef CONFIG_EXT3_INDEX
235 /*
236  * These functions convert from the major/minor hash to an f_pos
237  * value.
238  *
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
246  * cookie.  Sigh.
247  */
248 #define hash2pos(major, minor)  (major >> 1)
249 #define pos2maj_hash(pos)       ((pos << 1) & 0xffffffff)
250 #define pos2min_hash(pos)       (0)
251
252 /*
253  * This structure holds the nodes of the red-black tree used to store
254  * the directory entry in hash order.
255  */
256 struct fname {
257         __u32           hash;
258         __u32           minor_hash;
259         struct rb_node  rb_hash;
260         struct fname    *next;
261         __u32           inode;
262         __u8            name_len;
263         __u8            file_type;
264         char            name[0];
265 };
266
267 /*
268  * This functoin implements a non-recursive way of freeing all of the
269  * nodes in the red-black tree.
270  */
271 static void free_rb_tree_fname(struct rb_root *root)
272 {
273         struct rb_node  *n = root->rb_node;
274         struct rb_node  *parent;
275         struct fname    *fname;
276
277         while (n) {
278                 /* Do the node's children first */
279                 if ((n)->rb_left) {
280                         n = n->rb_left;
281                         continue;
282                 }
283                 if (n->rb_right) {
284                         n = n->rb_right;
285                         continue;
286                 }
287                 /*
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
291                  * node.
292                  */
293                 parent = rb_parent(n);
294                 fname = rb_entry(n, struct fname, rb_hash);
295                 while (fname) {
296                         struct fname * old = fname;
297                         fname = fname->next;
298                         kfree (old);
299                 }
300                 if (!parent)
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;
306                 n = parent;
307         }
308         root->rb_node = NULL;
309 }
310
311
312 static struct dir_private_info *create_dir_info(loff_t pos)
313 {
314         struct dir_private_info *p;
315
316         p = kmalloc(sizeof(struct dir_private_info), GFP_KERNEL);
317         if (!p)
318                 return NULL;
319         p->root.rb_node = NULL;
320         p->curr_node = NULL;
321         p->extra_fname = NULL;
322         p->last_pos = 0;
323         p->curr_hash = pos2maj_hash(pos);
324         p->curr_minor_hash = pos2min_hash(pos);
325         p->next_hash = 0;
326         return p;
327 }
328
329 void ext3_htree_free_dir_info(struct dir_private_info *p)
330 {
331         free_rb_tree_fname(&p->root);
332         kfree(p);
333 }
334
335 /*
336  * Given a directory entry, enter it into the fname rb tree.
337  */
338 int ext3_htree_store_dirent(struct file *dir_file, __u32 hash,
339                              __u32 minor_hash,
340                              struct ext3_dir_entry_2 *dirent)
341 {
342         struct rb_node **p, *parent = NULL;
343         struct fname * fname, *new_fn;
344         struct dir_private_info *info;
345         int len;
346
347         info = (struct dir_private_info *) dir_file->private_data;
348         p = &info->root.rb_node;
349
350         /* Create and allocate the fname structure */
351         len = sizeof(struct fname) + dirent->name_len + 1;
352         new_fn = kzalloc(len, GFP_KERNEL);
353         if (!new_fn)
354                 return -ENOMEM;
355         new_fn->hash = hash;
356         new_fn->minor_hash = minor_hash;
357         new_fn->inode = le32_to_cpu(dirent->inode);
358         new_fn->name_len = dirent->name_len;
359         new_fn->file_type = dirent->file_type;
360         memcpy(new_fn->name, dirent->name, dirent->name_len);
361         new_fn->name[dirent->name_len] = 0;
362
363         while (*p) {
364                 parent = *p;
365                 fname = rb_entry(parent, struct fname, rb_hash);
366
367                 /*
368                  * If the hash and minor hash match up, then we put
369                  * them on a linked list.  This rarely happens...
370                  */
371                 if ((new_fn->hash == fname->hash) &&
372                     (new_fn->minor_hash == fname->minor_hash)) {
373                         new_fn->next = fname->next;
374                         fname->next = new_fn;
375                         return 0;
376                 }
377
378                 if (new_fn->hash < fname->hash)
379                         p = &(*p)->rb_left;
380                 else if (new_fn->hash > fname->hash)
381                         p = &(*p)->rb_right;
382                 else if (new_fn->minor_hash < fname->minor_hash)
383                         p = &(*p)->rb_left;
384                 else /* if (new_fn->minor_hash > fname->minor_hash) */
385                         p = &(*p)->rb_right;
386         }
387
388         rb_link_node(&new_fn->rb_hash, parent, p);
389         rb_insert_color(&new_fn->rb_hash, &info->root);
390         return 0;
391 }
392
393
394
395 /*
396  * This is a helper function for ext3_dx_readdir.  It calls filldir
397  * for all entres on the fname linked list.  (Normally there is only
398  * one entry on the linked list, unless there are 62 bit hash collisions.)
399  */
400 static int call_filldir(struct file * filp, void * dirent,
401                         filldir_t filldir, struct fname *fname)
402 {
403         struct dir_private_info *info = filp->private_data;
404         loff_t  curr_pos;
405         struct inode *inode = filp->f_dentry->d_inode;
406         struct super_block * sb;
407         int error;
408
409         sb = inode->i_sb;
410
411         if (!fname) {
412                 printk("call_filldir: called with null fname?!?\n");
413                 return 0;
414         }
415         curr_pos = hash2pos(fname->hash, fname->minor_hash);
416         while (fname) {
417                 error = filldir(dirent, fname->name,
418                                 fname->name_len, curr_pos,
419                                 fname->inode,
420                                 get_dtype(sb, fname->file_type));
421                 if (error) {
422                         filp->f_pos = curr_pos;
423                         info->extra_fname = fname->next;
424                         return error;
425                 }
426                 fname = fname->next;
427         }
428         return 0;
429 }
430
431 static int ext3_dx_readdir(struct file * filp,
432                          void * dirent, filldir_t filldir)
433 {
434         struct dir_private_info *info = filp->private_data;
435         struct inode *inode = filp->f_dentry->d_inode;
436         struct fname *fname;
437         int     ret;
438
439         if (!info) {
440                 info = create_dir_info(filp->f_pos);
441                 if (!info)
442                         return -ENOMEM;
443                 filp->private_data = info;
444         }
445
446         if (filp->f_pos == EXT3_HTREE_EOF)
447                 return 0;       /* EOF */
448
449         /* Some one has messed with f_pos; reset the world */
450         if (info->last_pos != filp->f_pos) {
451                 free_rb_tree_fname(&info->root);
452                 info->curr_node = NULL;
453                 info->extra_fname = NULL;
454                 info->curr_hash = pos2maj_hash(filp->f_pos);
455                 info->curr_minor_hash = pos2min_hash(filp->f_pos);
456         }
457
458         /*
459          * If there are any leftover names on the hash collision
460          * chain, return them first.
461          */
462         if (info->extra_fname &&
463             call_filldir(filp, dirent, filldir, info->extra_fname))
464                 goto finished;
465
466         if (!info->curr_node)
467                 info->curr_node = rb_first(&info->root);
468
469         while (1) {
470                 /*
471                  * Fill the rbtree if we have no more entries,
472                  * or the inode has changed since we last read in the
473                  * cached entries.
474                  */
475                 if ((!info->curr_node) ||
476                     (filp->f_version != inode->i_version)) {
477                         info->curr_node = NULL;
478                         free_rb_tree_fname(&info->root);
479                         filp->f_version = inode->i_version;
480                         ret = ext3_htree_fill_tree(filp, info->curr_hash,
481                                                    info->curr_minor_hash,
482                                                    &info->next_hash);
483                         if (ret < 0)
484                                 return ret;
485                         if (ret == 0) {
486                                 filp->f_pos = EXT3_HTREE_EOF;
487                                 break;
488                         }
489                         info->curr_node = rb_first(&info->root);
490                 }
491
492                 fname = rb_entry(info->curr_node, struct fname, rb_hash);
493                 info->curr_hash = fname->hash;
494                 info->curr_minor_hash = fname->minor_hash;
495                 if (call_filldir(filp, dirent, filldir, fname))
496                         break;
497
498                 info->curr_node = rb_next(info->curr_node);
499                 if (!info->curr_node) {
500                         if (info->next_hash == ~0) {
501                                 filp->f_pos = EXT3_HTREE_EOF;
502                                 break;
503                         }
504                         info->curr_hash = info->next_hash;
505                         info->curr_minor_hash = 0;
506                 }
507         }
508 finished:
509         info->last_pos = filp->f_pos;
510         return 0;
511 }
512
513 static int ext3_release_dir (struct inode * inode, struct file * filp)
514 {
515        if (filp->private_data)
516                 ext3_htree_free_dir_info(filp->private_data);
517
518         return 0;
519 }
520
521 #endif