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