Merge branch 'akpm' (patches from Andrew)
[cascardo/linux.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 fscrypt_str fstr = FSTR_INIT(NULL, 0);
113
114         if (ext4_encrypted_inode(inode)) {
115                 err = fscrypt_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 = fscrypt_fname_alloc_buffer(inode, EXT4_NAME_LEN, &fstr);
143                 if (err < 0)
144                         return err;
145         }
146
147         offset = ctx->pos & (sb->s_blocksize - 1);
148
149         while (ctx->pos < inode->i_size) {
150                 struct ext4_map_blocks map;
151
152                 if (fatal_signal_pending(current)) {
153                         err = -ERESTARTSYS;
154                         goto errout;
155                 }
156                 cond_resched();
157                 map.m_lblk = ctx->pos >> EXT4_BLOCK_SIZE_BITS(sb);
158                 map.m_len = 1;
159                 err = ext4_map_blocks(NULL, inode, &map, 0);
160                 if (err > 0) {
161                         pgoff_t index = map.m_pblk >>
162                                         (PAGE_SHIFT - inode->i_blkbits);
163                         if (!ra_has_index(&file->f_ra, index))
164                                 page_cache_sync_readahead(
165                                         sb->s_bdev->bd_inode->i_mapping,
166                                         &file->f_ra, file,
167                                         index, 1);
168                         file->f_ra.prev_pos = (loff_t)index << PAGE_SHIFT;
169                         bh = ext4_bread(NULL, inode, map.m_lblk, 0);
170                         if (IS_ERR(bh)) {
171                                 err = PTR_ERR(bh);
172                                 bh = NULL;
173                                 goto errout;
174                         }
175                 }
176
177                 if (!bh) {
178                         if (!dir_has_error) {
179                                 EXT4_ERROR_FILE(file, 0,
180                                                 "directory contains a "
181                                                 "hole at offset %llu",
182                                            (unsigned long long) ctx->pos);
183                                 dir_has_error = 1;
184                         }
185                         /* corrupt size?  Maybe no more blocks to read */
186                         if (ctx->pos > inode->i_blocks << 9)
187                                 break;
188                         ctx->pos += sb->s_blocksize - offset;
189                         continue;
190                 }
191
192                 /* Check the checksum */
193                 if (!buffer_verified(bh) &&
194                     !ext4_dirent_csum_verify(inode,
195                                 (struct ext4_dir_entry *)bh->b_data)) {
196                         EXT4_ERROR_FILE(file, 0, "directory fails checksum "
197                                         "at offset %llu",
198                                         (unsigned long long)ctx->pos);
199                         ctx->pos += sb->s_blocksize - offset;
200                         brelse(bh);
201                         bh = NULL;
202                         continue;
203                 }
204                 set_buffer_verified(bh);
205
206                 /* If the dir block has changed since the last call to
207                  * readdir(2), then we might be pointing to an invalid
208                  * dirent right now.  Scan from the start of the block
209                  * to make sure. */
210                 if (file->f_version != inode->i_version) {
211                         for (i = 0; i < sb->s_blocksize && i < offset; ) {
212                                 de = (struct ext4_dir_entry_2 *)
213                                         (bh->b_data + i);
214                                 /* It's too expensive to do a full
215                                  * dirent test each time round this
216                                  * loop, but we do have to test at
217                                  * least that it is non-zero.  A
218                                  * failure will be detected in the
219                                  * dirent test below. */
220                                 if (ext4_rec_len_from_disk(de->rec_len,
221                                         sb->s_blocksize) < EXT4_DIR_REC_LEN(1))
222                                         break;
223                                 i += ext4_rec_len_from_disk(de->rec_len,
224                                                             sb->s_blocksize);
225                         }
226                         offset = i;
227                         ctx->pos = (ctx->pos & ~(sb->s_blocksize - 1))
228                                 | offset;
229                         file->f_version = inode->i_version;
230                 }
231
232                 while (ctx->pos < inode->i_size
233                        && offset < sb->s_blocksize) {
234                         de = (struct ext4_dir_entry_2 *) (bh->b_data + offset);
235                         if (ext4_check_dir_entry(inode, file, de, bh,
236                                                  bh->b_data, bh->b_size,
237                                                  offset)) {
238                                 /*
239                                  * On error, skip to the next block
240                                  */
241                                 ctx->pos = (ctx->pos |
242                                                 (sb->s_blocksize - 1)) + 1;
243                                 break;
244                         }
245                         offset += ext4_rec_len_from_disk(de->rec_len,
246                                         sb->s_blocksize);
247                         if (le32_to_cpu(de->inode)) {
248                                 if (!ext4_encrypted_inode(inode)) {
249                                         if (!dir_emit(ctx, de->name,
250                                             de->name_len,
251                                             le32_to_cpu(de->inode),
252                                             get_dtype(sb, de->file_type)))
253                                                 goto done;
254                                 } else {
255                                         int save_len = fstr.len;
256                                         struct fscrypt_str de_name =
257                                                         FSTR_INIT(de->name,
258                                                                 de->name_len);
259
260                                         /* Directory is encrypted */
261                                         err = fscrypt_fname_disk_to_usr(inode,
262                                                 0, 0, &de_name, &fstr);
263                                         de_name = fstr;
264                                         fstr.len = save_len;
265                                         if (err)
266                                                 goto errout;
267                                         if (!dir_emit(ctx,
268                                             de_name.name, de_name.len,
269                                             le32_to_cpu(de->inode),
270                                             get_dtype(sb, de->file_type)))
271                                                 goto done;
272                                 }
273                         }
274                         ctx->pos += ext4_rec_len_from_disk(de->rec_len,
275                                                 sb->s_blocksize);
276                 }
277                 if ((ctx->pos < inode->i_size) && !dir_relax_shared(inode))
278                         goto done;
279                 brelse(bh);
280                 bh = NULL;
281                 offset = 0;
282         }
283 done:
284         err = 0;
285 errout:
286 #ifdef CONFIG_EXT4_FS_ENCRYPTION
287         fscrypt_fname_free_buffer(&fstr);
288 #endif
289         brelse(bh);
290         return err;
291 }
292
293 static inline int is_32bit_api(void)
294 {
295 #ifdef CONFIG_COMPAT
296         return in_compat_syscall();
297 #else
298         return (BITS_PER_LONG == 32);
299 #endif
300 }
301
302 /*
303  * These functions convert from the major/minor hash to an f_pos
304  * value for dx directories
305  *
306  * Upper layer (for example NFS) should specify FMODE_32BITHASH or
307  * FMODE_64BITHASH explicitly. On the other hand, we allow ext4 to be mounted
308  * directly on both 32-bit and 64-bit nodes, under such case, neither
309  * FMODE_32BITHASH nor FMODE_64BITHASH is specified.
310  */
311 static inline loff_t hash2pos(struct file *filp, __u32 major, __u32 minor)
312 {
313         if ((filp->f_mode & FMODE_32BITHASH) ||
314             (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
315                 return major >> 1;
316         else
317                 return ((__u64)(major >> 1) << 32) | (__u64)minor;
318 }
319
320 static inline __u32 pos2maj_hash(struct file *filp, loff_t pos)
321 {
322         if ((filp->f_mode & FMODE_32BITHASH) ||
323             (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
324                 return (pos << 1) & 0xffffffff;
325         else
326                 return ((pos >> 32) << 1) & 0xffffffff;
327 }
328
329 static inline __u32 pos2min_hash(struct file *filp, loff_t pos)
330 {
331         if ((filp->f_mode & FMODE_32BITHASH) ||
332             (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
333                 return 0;
334         else
335                 return pos & 0xffffffff;
336 }
337
338 /*
339  * Return 32- or 64-bit end-of-file for dx directories
340  */
341 static inline loff_t ext4_get_htree_eof(struct file *filp)
342 {
343         if ((filp->f_mode & FMODE_32BITHASH) ||
344             (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
345                 return EXT4_HTREE_EOF_32BIT;
346         else
347                 return EXT4_HTREE_EOF_64BIT;
348 }
349
350
351 /*
352  * ext4_dir_llseek() calls generic_file_llseek_size to handle htree
353  * directories, where the "offset" is in terms of the filename hash
354  * value instead of the byte offset.
355  *
356  * Because we may return a 64-bit hash that is well beyond offset limits,
357  * we need to pass the max hash as the maximum allowable offset in
358  * the htree directory case.
359  *
360  * For non-htree, ext4_llseek already chooses the proper max offset.
361  */
362 static loff_t ext4_dir_llseek(struct file *file, loff_t offset, int whence)
363 {
364         struct inode *inode = file->f_mapping->host;
365         int dx_dir = is_dx_dir(inode);
366         loff_t htree_max = ext4_get_htree_eof(file);
367
368         if (likely(dx_dir))
369                 return generic_file_llseek_size(file, offset, whence,
370                                                     htree_max, htree_max);
371         else
372                 return ext4_llseek(file, offset, whence);
373 }
374
375 /*
376  * This structure holds the nodes of the red-black tree used to store
377  * the directory entry in hash order.
378  */
379 struct fname {
380         __u32           hash;
381         __u32           minor_hash;
382         struct rb_node  rb_hash;
383         struct fname    *next;
384         __u32           inode;
385         __u8            name_len;
386         __u8            file_type;
387         char            name[0];
388 };
389
390 /*
391  * This functoin implements a non-recursive way of freeing all of the
392  * nodes in the red-black tree.
393  */
394 static void free_rb_tree_fname(struct rb_root *root)
395 {
396         struct fname *fname, *next;
397
398         rbtree_postorder_for_each_entry_safe(fname, next, root, rb_hash)
399                 while (fname) {
400                         struct fname *old = fname;
401                         fname = fname->next;
402                         kfree(old);
403                 }
404
405         *root = RB_ROOT;
406 }
407
408
409 static struct dir_private_info *ext4_htree_create_dir_info(struct file *filp,
410                                                            loff_t pos)
411 {
412         struct dir_private_info *p;
413
414         p = kzalloc(sizeof(struct dir_private_info), GFP_KERNEL);
415         if (!p)
416                 return NULL;
417         p->curr_hash = pos2maj_hash(filp, pos);
418         p->curr_minor_hash = pos2min_hash(filp, pos);
419         return p;
420 }
421
422 void ext4_htree_free_dir_info(struct dir_private_info *p)
423 {
424         free_rb_tree_fname(&p->root);
425         kfree(p);
426 }
427
428 /*
429  * Given a directory entry, enter it into the fname rb tree.
430  *
431  * When filename encryption is enabled, the dirent will hold the
432  * encrypted filename, while the htree will hold decrypted filename.
433  * The decrypted filename is passed in via ent_name.  parameter.
434  */
435 int ext4_htree_store_dirent(struct file *dir_file, __u32 hash,
436                              __u32 minor_hash,
437                             struct ext4_dir_entry_2 *dirent,
438                             struct fscrypt_str *ent_name)
439 {
440         struct rb_node **p, *parent = NULL;
441         struct fname *fname, *new_fn;
442         struct dir_private_info *info;
443         int len;
444
445         info = dir_file->private_data;
446         p = &info->root.rb_node;
447
448         /* Create and allocate the fname structure */
449         len = sizeof(struct fname) + ent_name->len + 1;
450         new_fn = kzalloc(len, GFP_KERNEL);
451         if (!new_fn)
452                 return -ENOMEM;
453         new_fn->hash = hash;
454         new_fn->minor_hash = minor_hash;
455         new_fn->inode = le32_to_cpu(dirent->inode);
456         new_fn->name_len = ent_name->len;
457         new_fn->file_type = dirent->file_type;
458         memcpy(new_fn->name, ent_name->name, ent_name->len);
459         new_fn->name[ent_name->len] = 0;
460
461         while (*p) {
462                 parent = *p;
463                 fname = rb_entry(parent, struct fname, rb_hash);
464
465                 /*
466                  * If the hash and minor hash match up, then we put
467                  * them on a linked list.  This rarely happens...
468                  */
469                 if ((new_fn->hash == fname->hash) &&
470                     (new_fn->minor_hash == fname->minor_hash)) {
471                         new_fn->next = fname->next;
472                         fname->next = new_fn;
473                         return 0;
474                 }
475
476                 if (new_fn->hash < fname->hash)
477                         p = &(*p)->rb_left;
478                 else if (new_fn->hash > fname->hash)
479                         p = &(*p)->rb_right;
480                 else if (new_fn->minor_hash < fname->minor_hash)
481                         p = &(*p)->rb_left;
482                 else /* if (new_fn->minor_hash > fname->minor_hash) */
483                         p = &(*p)->rb_right;
484         }
485
486         rb_link_node(&new_fn->rb_hash, parent, p);
487         rb_insert_color(&new_fn->rb_hash, &info->root);
488         return 0;
489 }
490
491
492
493 /*
494  * This is a helper function for ext4_dx_readdir.  It calls filldir
495  * for all entres on the fname linked list.  (Normally there is only
496  * one entry on the linked list, unless there are 62 bit hash collisions.)
497  */
498 static int call_filldir(struct file *file, struct dir_context *ctx,
499                         struct fname *fname)
500 {
501         struct dir_private_info *info = file->private_data;
502         struct inode *inode = file_inode(file);
503         struct super_block *sb = inode->i_sb;
504
505         if (!fname) {
506                 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: comm %s: "
507                          "called with null fname?!?", __func__, __LINE__,
508                          inode->i_ino, current->comm);
509                 return 0;
510         }
511         ctx->pos = hash2pos(file, fname->hash, fname->minor_hash);
512         while (fname) {
513                 if (!dir_emit(ctx, fname->name,
514                                 fname->name_len,
515                                 fname->inode,
516                                 get_dtype(sb, fname->file_type))) {
517                         info->extra_fname = fname;
518                         return 1;
519                 }
520                 fname = fname->next;
521         }
522         return 0;
523 }
524
525 static int ext4_dx_readdir(struct file *file, struct dir_context *ctx)
526 {
527         struct dir_private_info *info = file->private_data;
528         struct inode *inode = file_inode(file);
529         struct fname *fname;
530         int     ret;
531
532         if (!info) {
533                 info = ext4_htree_create_dir_info(file, ctx->pos);
534                 if (!info)
535                         return -ENOMEM;
536                 file->private_data = info;
537         }
538
539         if (ctx->pos == ext4_get_htree_eof(file))
540                 return 0;       /* EOF */
541
542         /* Some one has messed with f_pos; reset the world */
543         if (info->last_pos != ctx->pos) {
544                 free_rb_tree_fname(&info->root);
545                 info->curr_node = NULL;
546                 info->extra_fname = NULL;
547                 info->curr_hash = pos2maj_hash(file, ctx->pos);
548                 info->curr_minor_hash = pos2min_hash(file, ctx->pos);
549         }
550
551         /*
552          * If there are any leftover names on the hash collision
553          * chain, return them first.
554          */
555         if (info->extra_fname) {
556                 if (call_filldir(file, ctx, info->extra_fname))
557                         goto finished;
558                 info->extra_fname = NULL;
559                 goto next_node;
560         } else if (!info->curr_node)
561                 info->curr_node = rb_first(&info->root);
562
563         while (1) {
564                 /*
565                  * Fill the rbtree if we have no more entries,
566                  * or the inode has changed since we last read in the
567                  * cached entries.
568                  */
569                 if ((!info->curr_node) ||
570                     (file->f_version != inode->i_version)) {
571                         info->curr_node = NULL;
572                         free_rb_tree_fname(&info->root);
573                         file->f_version = inode->i_version;
574                         ret = ext4_htree_fill_tree(file, info->curr_hash,
575                                                    info->curr_minor_hash,
576                                                    &info->next_hash);
577                         if (ret < 0)
578                                 return ret;
579                         if (ret == 0) {
580                                 ctx->pos = ext4_get_htree_eof(file);
581                                 break;
582                         }
583                         info->curr_node = rb_first(&info->root);
584                 }
585
586                 fname = rb_entry(info->curr_node, struct fname, rb_hash);
587                 info->curr_hash = fname->hash;
588                 info->curr_minor_hash = fname->minor_hash;
589                 if (call_filldir(file, ctx, fname))
590                         break;
591         next_node:
592                 info->curr_node = rb_next(info->curr_node);
593                 if (info->curr_node) {
594                         fname = rb_entry(info->curr_node, struct fname,
595                                          rb_hash);
596                         info->curr_hash = fname->hash;
597                         info->curr_minor_hash = fname->minor_hash;
598                 } else {
599                         if (info->next_hash == ~0) {
600                                 ctx->pos = ext4_get_htree_eof(file);
601                                 break;
602                         }
603                         info->curr_hash = info->next_hash;
604                         info->curr_minor_hash = 0;
605                 }
606         }
607 finished:
608         info->last_pos = ctx->pos;
609         return 0;
610 }
611
612 static int ext4_dir_open(struct inode * inode, struct file * filp)
613 {
614         if (ext4_encrypted_inode(inode))
615                 return fscrypt_get_encryption_info(inode) ? -EACCES : 0;
616         return 0;
617 }
618
619 static int ext4_release_dir(struct inode *inode, struct file *filp)
620 {
621         if (filp->private_data)
622                 ext4_htree_free_dir_info(filp->private_data);
623
624         return 0;
625 }
626
627 int ext4_check_all_de(struct inode *dir, struct buffer_head *bh, void *buf,
628                       int buf_size)
629 {
630         struct ext4_dir_entry_2 *de;
631         int rlen;
632         unsigned int offset = 0;
633         char *top;
634
635         de = (struct ext4_dir_entry_2 *)buf;
636         top = buf + buf_size;
637         while ((char *) de < top) {
638                 if (ext4_check_dir_entry(dir, NULL, de, bh,
639                                          buf, buf_size, offset))
640                         return -EFSCORRUPTED;
641                 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
642                 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
643                 offset += rlen;
644         }
645         if ((char *) de > top)
646                 return -EFSCORRUPTED;
647
648         return 0;
649 }
650
651 const struct file_operations ext4_dir_operations = {
652         .llseek         = ext4_dir_llseek,
653         .read           = generic_read_dir,
654         .iterate_shared = ext4_readdir,
655         .unlocked_ioctl = ext4_ioctl,
656 #ifdef CONFIG_COMPAT
657         .compat_ioctl   = ext4_compat_ioctl,
658 #endif
659         .fsync          = ext4_sync_file,
660         .open           = ext4_dir_open,
661         .release        = ext4_release_dir,
662 };