Merge tag 'ext4_for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tytso...
[sfrench/cifs-2.6.git] / fs / ext4 / dir.c
1 /*
2  *  linux/fs/ext4/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  *  ext4 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/buffer_head.h>
26 #include <linux/slab.h>
27 #include "ext4.h"
28 #include "xattr.h"
29
30 static int ext4_dx_readdir(struct file *, struct dir_context *);
31
32 /**
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
35  * indexing).
36  *
37  * Return 1 if it is a dx dir, 0 if not
38  */
39 static int is_dx_dir(struct inode *inode)
40 {
41         struct super_block *sb = inode->i_sb;
42
43         if (ext4_has_feature_dir_index(inode->i_sb) &&
44             ((ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) ||
45              ((inode->i_size >> sb->s_blocksize_bits) == 1) ||
46              ext4_has_inline_data(inode)))
47                 return 1;
48
49         return 0;
50 }
51
52 /*
53  * Return 0 if the directory entry is OK, and 1 if there is a problem
54  *
55  * Note: this is the opposite of what ext2 and ext3 historically returned...
56  *
57  * bh passed here can be an inode block or a dir data block, depending
58  * on the inode inline data flag.
59  */
60 int __ext4_check_dir_entry(const char *function, unsigned int line,
61                            struct inode *dir, struct file *filp,
62                            struct ext4_dir_entry_2 *de,
63                            struct buffer_head *bh, char *buf, int size,
64                            unsigned int offset)
65 {
66         const char *error_msg = NULL;
67         const int rlen = ext4_rec_len_from_disk(de->rec_len,
68                                                 dir->i_sb->s_blocksize);
69
70         if (unlikely(rlen < EXT4_DIR_REC_LEN(1)))
71                 error_msg = "rec_len is smaller than minimal";
72         else if (unlikely(rlen % 4 != 0))
73                 error_msg = "rec_len % 4 != 0";
74         else if (unlikely(rlen < EXT4_DIR_REC_LEN(de->name_len)))
75                 error_msg = "rec_len is too small for name_len";
76         else if (unlikely(((char *) de - buf) + rlen > size))
77                 error_msg = "directory entry across range";
78         else if (unlikely(le32_to_cpu(de->inode) >
79                         le32_to_cpu(EXT4_SB(dir->i_sb)->s_es->s_inodes_count)))
80                 error_msg = "inode out of bounds";
81         else
82                 return 0;
83
84         if (filp)
85                 ext4_error_file(filp, function, line, bh->b_blocknr,
86                                 "bad entry in directory: %s - offset=%u(%u), "
87                                 "inode=%u, rec_len=%d, name_len=%d",
88                                 error_msg, (unsigned) (offset % size),
89                                 offset, le32_to_cpu(de->inode),
90                                 rlen, de->name_len);
91         else
92                 ext4_error_inode(dir, function, line, bh->b_blocknr,
93                                 "bad entry in directory: %s - offset=%u(%u), "
94                                 "inode=%u, rec_len=%d, name_len=%d",
95                                 error_msg, (unsigned) (offset % size),
96                                 offset, le32_to_cpu(de->inode),
97                                 rlen, de->name_len);
98
99         return 1;
100 }
101
102 static int ext4_readdir(struct file *file, struct dir_context *ctx)
103 {
104         unsigned int offset;
105         int i;
106         struct ext4_dir_entry_2 *de;
107         int err;
108         struct inode *inode = file_inode(file);
109         struct super_block *sb = inode->i_sb;
110         struct buffer_head *bh = NULL;
111         int dir_has_error = 0;
112         struct ext4_str fname_crypto_str = {.name = NULL, .len = 0};
113
114         if (ext4_encrypted_inode(inode)) {
115                 err = ext4_get_encryption_info(inode);
116                 if (err && err != -ENOKEY)
117                         return err;
118         }
119
120         if (is_dx_dir(inode)) {
121                 err = ext4_dx_readdir(file, ctx);
122                 if (err != ERR_BAD_DX_DIR) {
123                         return err;
124                 }
125                 /*
126                  * We don't set the inode dirty flag since it's not
127                  * critical that it get flushed back to the disk.
128                  */
129                 ext4_clear_inode_flag(file_inode(file),
130                                       EXT4_INODE_INDEX);
131         }
132
133         if (ext4_has_inline_data(inode)) {
134                 int has_inline_data = 1;
135                 err = ext4_read_inline_dir(file, ctx,
136                                            &has_inline_data);
137                 if (has_inline_data)
138                         return err;
139         }
140
141         if (ext4_encrypted_inode(inode)) {
142                 err = ext4_fname_crypto_alloc_buffer(inode, EXT4_NAME_LEN,
143                                                      &fname_crypto_str);
144                 if (err < 0)
145                         return err;
146         }
147
148         offset = ctx->pos & (sb->s_blocksize - 1);
149
150         while (ctx->pos < inode->i_size) {
151                 struct ext4_map_blocks map;
152
153                 if (fatal_signal_pending(current)) {
154                         err = -ERESTARTSYS;
155                         goto errout;
156                 }
157                 cond_resched();
158                 map.m_lblk = ctx->pos >> EXT4_BLOCK_SIZE_BITS(sb);
159                 map.m_len = 1;
160                 err = ext4_map_blocks(NULL, inode, &map, 0);
161                 if (err > 0) {
162                         pgoff_t index = map.m_pblk >>
163                                         (PAGE_SHIFT - inode->i_blkbits);
164                         if (!ra_has_index(&file->f_ra, index))
165                                 page_cache_sync_readahead(
166                                         sb->s_bdev->bd_inode->i_mapping,
167                                         &file->f_ra, file,
168                                         index, 1);
169                         file->f_ra.prev_pos = (loff_t)index << PAGE_SHIFT;
170                         bh = ext4_bread(NULL, inode, map.m_lblk, 0);
171                         if (IS_ERR(bh)) {
172                                 err = PTR_ERR(bh);
173                                 bh = NULL;
174                                 goto errout;
175                         }
176                 }
177
178                 if (!bh) {
179                         if (!dir_has_error) {
180                                 EXT4_ERROR_FILE(file, 0,
181                                                 "directory contains a "
182                                                 "hole at offset %llu",
183                                            (unsigned long long) ctx->pos);
184                                 dir_has_error = 1;
185                         }
186                         /* corrupt size?  Maybe no more blocks to read */
187                         if (ctx->pos > inode->i_blocks << 9)
188                                 break;
189                         ctx->pos += sb->s_blocksize - offset;
190                         continue;
191                 }
192
193                 /* Check the checksum */
194                 if (!buffer_verified(bh) &&
195                     !ext4_dirent_csum_verify(inode,
196                                 (struct ext4_dir_entry *)bh->b_data)) {
197                         EXT4_ERROR_FILE(file, 0, "directory fails checksum "
198                                         "at offset %llu",
199                                         (unsigned long long)ctx->pos);
200                         ctx->pos += sb->s_blocksize - offset;
201                         brelse(bh);
202                         bh = NULL;
203                         continue;
204                 }
205                 set_buffer_verified(bh);
206
207                 /* If the dir block has changed since the last call to
208                  * readdir(2), then we might be pointing to an invalid
209                  * dirent right now.  Scan from the start of the block
210                  * to make sure. */
211                 if (file->f_version != inode->i_version) {
212                         for (i = 0; i < sb->s_blocksize && i < offset; ) {
213                                 de = (struct ext4_dir_entry_2 *)
214                                         (bh->b_data + i);
215                                 /* It's too expensive to do a full
216                                  * dirent test each time round this
217                                  * loop, but we do have to test at
218                                  * least that it is non-zero.  A
219                                  * failure will be detected in the
220                                  * dirent test below. */
221                                 if (ext4_rec_len_from_disk(de->rec_len,
222                                         sb->s_blocksize) < EXT4_DIR_REC_LEN(1))
223                                         break;
224                                 i += ext4_rec_len_from_disk(de->rec_len,
225                                                             sb->s_blocksize);
226                         }
227                         offset = i;
228                         ctx->pos = (ctx->pos & ~(sb->s_blocksize - 1))
229                                 | offset;
230                         file->f_version = inode->i_version;
231                 }
232
233                 while (ctx->pos < inode->i_size
234                        && offset < sb->s_blocksize) {
235                         de = (struct ext4_dir_entry_2 *) (bh->b_data + offset);
236                         if (ext4_check_dir_entry(inode, file, de, bh,
237                                                  bh->b_data, bh->b_size,
238                                                  offset)) {
239                                 /*
240                                  * On error, skip to the next block
241                                  */
242                                 ctx->pos = (ctx->pos |
243                                                 (sb->s_blocksize - 1)) + 1;
244                                 break;
245                         }
246                         offset += ext4_rec_len_from_disk(de->rec_len,
247                                         sb->s_blocksize);
248                         if (le32_to_cpu(de->inode)) {
249                                 if (!ext4_encrypted_inode(inode)) {
250                                         if (!dir_emit(ctx, de->name,
251                                             de->name_len,
252                                             le32_to_cpu(de->inode),
253                                             get_dtype(sb, de->file_type)))
254                                                 goto done;
255                                 } else {
256                                         int save_len = fname_crypto_str.len;
257
258                                         /* Directory is encrypted */
259                                         err = ext4_fname_disk_to_usr(inode,
260                                                 NULL, de, &fname_crypto_str);
261                                         fname_crypto_str.len = save_len;
262                                         if (err < 0)
263                                                 goto errout;
264                                         if (!dir_emit(ctx,
265                                             fname_crypto_str.name, err,
266                                             le32_to_cpu(de->inode),
267                                             get_dtype(sb, de->file_type)))
268                                                 goto done;
269                                 }
270                         }
271                         ctx->pos += ext4_rec_len_from_disk(de->rec_len,
272                                                 sb->s_blocksize);
273                 }
274                 if ((ctx->pos < inode->i_size) && !dir_relax_shared(inode))
275                         goto done;
276                 brelse(bh);
277                 bh = NULL;
278                 offset = 0;
279         }
280 done:
281         err = 0;
282 errout:
283 #ifdef CONFIG_EXT4_FS_ENCRYPTION
284         ext4_fname_crypto_free_buffer(&fname_crypto_str);
285 #endif
286         brelse(bh);
287         return err;
288 }
289
290 static inline int is_32bit_api(void)
291 {
292 #ifdef CONFIG_COMPAT
293         return in_compat_syscall();
294 #else
295         return (BITS_PER_LONG == 32);
296 #endif
297 }
298
299 /*
300  * These functions convert from the major/minor hash to an f_pos
301  * value for dx directories
302  *
303  * Upper layer (for example NFS) should specify FMODE_32BITHASH or
304  * FMODE_64BITHASH explicitly. On the other hand, we allow ext4 to be mounted
305  * directly on both 32-bit and 64-bit nodes, under such case, neither
306  * FMODE_32BITHASH nor FMODE_64BITHASH is specified.
307  */
308 static inline loff_t hash2pos(struct file *filp, __u32 major, __u32 minor)
309 {
310         if ((filp->f_mode & FMODE_32BITHASH) ||
311             (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
312                 return major >> 1;
313         else
314                 return ((__u64)(major >> 1) << 32) | (__u64)minor;
315 }
316
317 static inline __u32 pos2maj_hash(struct file *filp, loff_t pos)
318 {
319         if ((filp->f_mode & FMODE_32BITHASH) ||
320             (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
321                 return (pos << 1) & 0xffffffff;
322         else
323                 return ((pos >> 32) << 1) & 0xffffffff;
324 }
325
326 static inline __u32 pos2min_hash(struct file *filp, loff_t pos)
327 {
328         if ((filp->f_mode & FMODE_32BITHASH) ||
329             (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
330                 return 0;
331         else
332                 return pos & 0xffffffff;
333 }
334
335 /*
336  * Return 32- or 64-bit end-of-file for dx directories
337  */
338 static inline loff_t ext4_get_htree_eof(struct file *filp)
339 {
340         if ((filp->f_mode & FMODE_32BITHASH) ||
341             (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
342                 return EXT4_HTREE_EOF_32BIT;
343         else
344                 return EXT4_HTREE_EOF_64BIT;
345 }
346
347
348 /*
349  * ext4_dir_llseek() calls generic_file_llseek_size to handle htree
350  * directories, where the "offset" is in terms of the filename hash
351  * value instead of the byte offset.
352  *
353  * Because we may return a 64-bit hash that is well beyond offset limits,
354  * we need to pass the max hash as the maximum allowable offset in
355  * the htree directory case.
356  *
357  * For non-htree, ext4_llseek already chooses the proper max offset.
358  */
359 static loff_t ext4_dir_llseek(struct file *file, loff_t offset, int whence)
360 {
361         struct inode *inode = file->f_mapping->host;
362         int dx_dir = is_dx_dir(inode);
363         loff_t htree_max = ext4_get_htree_eof(file);
364
365         if (likely(dx_dir))
366                 return generic_file_llseek_size(file, offset, whence,
367                                                     htree_max, htree_max);
368         else
369                 return ext4_llseek(file, offset, whence);
370 }
371
372 /*
373  * This structure holds the nodes of the red-black tree used to store
374  * the directory entry in hash order.
375  */
376 struct fname {
377         __u32           hash;
378         __u32           minor_hash;
379         struct rb_node  rb_hash;
380         struct fname    *next;
381         __u32           inode;
382         __u8            name_len;
383         __u8            file_type;
384         char            name[0];
385 };
386
387 /*
388  * This functoin implements a non-recursive way of freeing all of the
389  * nodes in the red-black tree.
390  */
391 static void free_rb_tree_fname(struct rb_root *root)
392 {
393         struct fname *fname, *next;
394
395         rbtree_postorder_for_each_entry_safe(fname, next, root, rb_hash)
396                 while (fname) {
397                         struct fname *old = fname;
398                         fname = fname->next;
399                         kfree(old);
400                 }
401
402         *root = RB_ROOT;
403 }
404
405
406 static struct dir_private_info *ext4_htree_create_dir_info(struct file *filp,
407                                                            loff_t pos)
408 {
409         struct dir_private_info *p;
410
411         p = kzalloc(sizeof(struct dir_private_info), GFP_KERNEL);
412         if (!p)
413                 return NULL;
414         p->curr_hash = pos2maj_hash(filp, pos);
415         p->curr_minor_hash = pos2min_hash(filp, pos);
416         return p;
417 }
418
419 void ext4_htree_free_dir_info(struct dir_private_info *p)
420 {
421         free_rb_tree_fname(&p->root);
422         kfree(p);
423 }
424
425 /*
426  * Given a directory entry, enter it into the fname rb tree.
427  *
428  * When filename encryption is enabled, the dirent will hold the
429  * encrypted filename, while the htree will hold decrypted filename.
430  * The decrypted filename is passed in via ent_name.  parameter.
431  */
432 int ext4_htree_store_dirent(struct file *dir_file, __u32 hash,
433                              __u32 minor_hash,
434                             struct ext4_dir_entry_2 *dirent,
435                             struct ext4_str *ent_name)
436 {
437         struct rb_node **p, *parent = NULL;
438         struct fname *fname, *new_fn;
439         struct dir_private_info *info;
440         int len;
441
442         info = dir_file->private_data;
443         p = &info->root.rb_node;
444
445         /* Create and allocate the fname structure */
446         len = sizeof(struct fname) + ent_name->len + 1;
447         new_fn = kzalloc(len, GFP_KERNEL);
448         if (!new_fn)
449                 return -ENOMEM;
450         new_fn->hash = hash;
451         new_fn->minor_hash = minor_hash;
452         new_fn->inode = le32_to_cpu(dirent->inode);
453         new_fn->name_len = ent_name->len;
454         new_fn->file_type = dirent->file_type;
455         memcpy(new_fn->name, ent_name->name, ent_name->len);
456         new_fn->name[ent_name->len] = 0;
457
458         while (*p) {
459                 parent = *p;
460                 fname = rb_entry(parent, struct fname, rb_hash);
461
462                 /*
463                  * If the hash and minor hash match up, then we put
464                  * them on a linked list.  This rarely happens...
465                  */
466                 if ((new_fn->hash == fname->hash) &&
467                     (new_fn->minor_hash == fname->minor_hash)) {
468                         new_fn->next = fname->next;
469                         fname->next = new_fn;
470                         return 0;
471                 }
472
473                 if (new_fn->hash < fname->hash)
474                         p = &(*p)->rb_left;
475                 else if (new_fn->hash > fname->hash)
476                         p = &(*p)->rb_right;
477                 else if (new_fn->minor_hash < fname->minor_hash)
478                         p = &(*p)->rb_left;
479                 else /* if (new_fn->minor_hash > fname->minor_hash) */
480                         p = &(*p)->rb_right;
481         }
482
483         rb_link_node(&new_fn->rb_hash, parent, p);
484         rb_insert_color(&new_fn->rb_hash, &info->root);
485         return 0;
486 }
487
488
489
490 /*
491  * This is a helper function for ext4_dx_readdir.  It calls filldir
492  * for all entres on the fname linked list.  (Normally there is only
493  * one entry on the linked list, unless there are 62 bit hash collisions.)
494  */
495 static int call_filldir(struct file *file, struct dir_context *ctx,
496                         struct fname *fname)
497 {
498         struct dir_private_info *info = file->private_data;
499         struct inode *inode = file_inode(file);
500         struct super_block *sb = inode->i_sb;
501
502         if (!fname) {
503                 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: comm %s: "
504                          "called with null fname?!?", __func__, __LINE__,
505                          inode->i_ino, current->comm);
506                 return 0;
507         }
508         ctx->pos = hash2pos(file, fname->hash, fname->minor_hash);
509         while (fname) {
510                 if (!dir_emit(ctx, fname->name,
511                                 fname->name_len,
512                                 fname->inode,
513                                 get_dtype(sb, fname->file_type))) {
514                         info->extra_fname = fname;
515                         return 1;
516                 }
517                 fname = fname->next;
518         }
519         return 0;
520 }
521
522 static int ext4_dx_readdir(struct file *file, struct dir_context *ctx)
523 {
524         struct dir_private_info *info = file->private_data;
525         struct inode *inode = file_inode(file);
526         struct fname *fname;
527         int     ret;
528
529         if (!info) {
530                 info = ext4_htree_create_dir_info(file, ctx->pos);
531                 if (!info)
532                         return -ENOMEM;
533                 file->private_data = info;
534         }
535
536         if (ctx->pos == ext4_get_htree_eof(file))
537                 return 0;       /* EOF */
538
539         /* Some one has messed with f_pos; reset the world */
540         if (info->last_pos != ctx->pos) {
541                 free_rb_tree_fname(&info->root);
542                 info->curr_node = NULL;
543                 info->extra_fname = NULL;
544                 info->curr_hash = pos2maj_hash(file, ctx->pos);
545                 info->curr_minor_hash = pos2min_hash(file, ctx->pos);
546         }
547
548         /*
549          * If there are any leftover names on the hash collision
550          * chain, return them first.
551          */
552         if (info->extra_fname) {
553                 if (call_filldir(file, ctx, info->extra_fname))
554                         goto finished;
555                 info->extra_fname = NULL;
556                 goto next_node;
557         } else if (!info->curr_node)
558                 info->curr_node = rb_first(&info->root);
559
560         while (1) {
561                 /*
562                  * Fill the rbtree if we have no more entries,
563                  * or the inode has changed since we last read in the
564                  * cached entries.
565                  */
566                 if ((!info->curr_node) ||
567                     (file->f_version != inode->i_version)) {
568                         info->curr_node = NULL;
569                         free_rb_tree_fname(&info->root);
570                         file->f_version = inode->i_version;
571                         ret = ext4_htree_fill_tree(file, info->curr_hash,
572                                                    info->curr_minor_hash,
573                                                    &info->next_hash);
574                         if (ret < 0)
575                                 return ret;
576                         if (ret == 0) {
577                                 ctx->pos = ext4_get_htree_eof(file);
578                                 break;
579                         }
580                         info->curr_node = rb_first(&info->root);
581                 }
582
583                 fname = rb_entry(info->curr_node, struct fname, rb_hash);
584                 info->curr_hash = fname->hash;
585                 info->curr_minor_hash = fname->minor_hash;
586                 if (call_filldir(file, ctx, fname))
587                         break;
588         next_node:
589                 info->curr_node = rb_next(info->curr_node);
590                 if (info->curr_node) {
591                         fname = rb_entry(info->curr_node, struct fname,
592                                          rb_hash);
593                         info->curr_hash = fname->hash;
594                         info->curr_minor_hash = fname->minor_hash;
595                 } else {
596                         if (info->next_hash == ~0) {
597                                 ctx->pos = ext4_get_htree_eof(file);
598                                 break;
599                         }
600                         info->curr_hash = info->next_hash;
601                         info->curr_minor_hash = 0;
602                 }
603         }
604 finished:
605         info->last_pos = ctx->pos;
606         return 0;
607 }
608
609 static int ext4_dir_open(struct inode * inode, struct file * filp)
610 {
611         if (ext4_encrypted_inode(inode))
612                 return ext4_get_encryption_info(inode) ? -EACCES : 0;
613         return 0;
614 }
615
616 static int ext4_release_dir(struct inode *inode, struct file *filp)
617 {
618         if (filp->private_data)
619                 ext4_htree_free_dir_info(filp->private_data);
620
621         return 0;
622 }
623
624 int ext4_check_all_de(struct inode *dir, struct buffer_head *bh, void *buf,
625                       int buf_size)
626 {
627         struct ext4_dir_entry_2 *de;
628         int nlen, rlen;
629         unsigned int offset = 0;
630         char *top;
631
632         de = (struct ext4_dir_entry_2 *)buf;
633         top = buf + buf_size;
634         while ((char *) de < top) {
635                 if (ext4_check_dir_entry(dir, NULL, de, bh,
636                                          buf, buf_size, offset))
637                         return -EFSCORRUPTED;
638                 nlen = EXT4_DIR_REC_LEN(de->name_len);
639                 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
640                 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
641                 offset += rlen;
642         }
643         if ((char *) de > top)
644                 return -EFSCORRUPTED;
645
646         return 0;
647 }
648
649 const struct file_operations ext4_dir_operations = {
650         .llseek         = ext4_dir_llseek,
651         .read           = generic_read_dir,
652         .iterate_shared = ext4_readdir,
653         .unlocked_ioctl = ext4_ioctl,
654 #ifdef CONFIG_COMPAT
655         .compat_ioctl   = ext4_compat_ioctl,
656 #endif
657         .fsync          = ext4_sync_file,
658         .open           = ext4_dir_open,
659         .release        = ext4_release_dir,
660 };