GNU Linux-libre 4.4.289-gnu1
[releases.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 overrun";
78         else if (unlikely(((char *) de - buf) + rlen >
79                           size - EXT4_DIR_REC_LEN(1) &&
80                           ((char *) de - buf) + rlen != size)) {
81                 error_msg = "directory entry too close to block end";
82         }
83         else if (unlikely(le32_to_cpu(de->inode) >
84                         le32_to_cpu(EXT4_SB(dir->i_sb)->s_es->s_inodes_count)))
85                 error_msg = "inode out of bounds";
86         else
87                 return 0;
88
89         if (filp)
90                 ext4_error_file(filp, function, line, bh->b_blocknr,
91                                 "bad entry in directory: %s - offset=%u, "
92                                 "inode=%u, rec_len=%d, name_len=%d, size=%d",
93                                 error_msg, offset, le32_to_cpu(de->inode),
94                                 rlen, de->name_len, size);
95         else
96                 ext4_error_inode(dir, function, line, bh->b_blocknr,
97                                 "bad entry in directory: %s - offset=%u, "
98                                 "inode=%u, rec_len=%d, name_len=%d, size=%d",
99                                  error_msg, offset, le32_to_cpu(de->inode),
100                                  rlen, de->name_len, size);
101
102         return 1;
103 }
104
105 static int ext4_readdir(struct file *file, struct dir_context *ctx)
106 {
107         unsigned int offset;
108         int i;
109         struct ext4_dir_entry_2 *de;
110         int err;
111         struct inode *inode = file_inode(file);
112         struct super_block *sb = inode->i_sb;
113         struct buffer_head *bh = NULL;
114         int dir_has_error = 0;
115         struct ext4_str fname_crypto_str = {.name = NULL, .len = 0};
116
117         if (ext4_encrypted_inode(inode)) {
118                 err = ext4_get_encryption_info(inode);
119                 if (err && err != -ENOKEY)
120                         return err;
121         }
122
123         if (is_dx_dir(inode)) {
124                 err = ext4_dx_readdir(file, ctx);
125                 if (err != ERR_BAD_DX_DIR) {
126                         return err;
127                 }
128                 /* Can we just clear INDEX flag to ignore htree information? */
129                 if (!ext4_has_metadata_csum(sb)) {
130                         /*
131                          * We don't set the inode dirty flag since it's not
132                          * critical that it gets flushed back to the disk.
133                          */
134                         ext4_clear_inode_flag(inode, EXT4_INODE_INDEX);
135                 }
136         }
137
138         if (ext4_has_inline_data(inode)) {
139                 int has_inline_data = 1;
140                 err = ext4_read_inline_dir(file, ctx,
141                                            &has_inline_data);
142                 if (has_inline_data)
143                         return err;
144         }
145
146         if (ext4_encrypted_inode(inode)) {
147                 err = ext4_fname_crypto_alloc_buffer(inode, EXT4_NAME_LEN,
148                                                      &fname_crypto_str);
149                 if (err < 0)
150                         return err;
151         }
152
153         offset = ctx->pos & (sb->s_blocksize - 1);
154
155         while (ctx->pos < inode->i_size) {
156                 struct ext4_map_blocks map;
157
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_CACHE_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_CACHE_SHIFT;
170                         bh = ext4_bread(NULL, inode, map.m_lblk, 0);
171                         if (IS_ERR(bh))
172                                 return PTR_ERR(bh);
173                 }
174
175                 if (!bh) {
176                         if (!dir_has_error) {
177                                 EXT4_ERROR_FILE(file, 0,
178                                                 "directory contains a "
179                                                 "hole at offset %llu",
180                                            (unsigned long long) ctx->pos);
181                                 dir_has_error = 1;
182                         }
183                         /* corrupt size?  Maybe no more blocks to read */
184                         if (ctx->pos > inode->i_blocks << 9)
185                                 break;
186                         ctx->pos += sb->s_blocksize - offset;
187                         continue;
188                 }
189
190                 /* Check the checksum */
191                 if (!buffer_verified(bh) &&
192                     !ext4_dirent_csum_verify(inode,
193                                 (struct ext4_dir_entry *)bh->b_data)) {
194                         EXT4_ERROR_FILE(file, 0, "directory fails checksum "
195                                         "at offset %llu",
196                                         (unsigned long long)ctx->pos);
197                         ctx->pos += sb->s_blocksize - offset;
198                         brelse(bh);
199                         bh = NULL;
200                         continue;
201                 }
202                 set_buffer_verified(bh);
203
204                 /* If the dir block has changed since the last call to
205                  * readdir(2), then we might be pointing to an invalid
206                  * dirent right now.  Scan from the start of the block
207                  * to make sure. */
208                 if (file->f_version != inode->i_version) {
209                         for (i = 0; i < sb->s_blocksize && i < offset; ) {
210                                 de = (struct ext4_dir_entry_2 *)
211                                         (bh->b_data + i);
212                                 /* It's too expensive to do a full
213                                  * dirent test each time round this
214                                  * loop, but we do have to test at
215                                  * least that it is non-zero.  A
216                                  * failure will be detected in the
217                                  * dirent test below. */
218                                 if (ext4_rec_len_from_disk(de->rec_len,
219                                         sb->s_blocksize) < EXT4_DIR_REC_LEN(1))
220                                         break;
221                                 i += ext4_rec_len_from_disk(de->rec_len,
222                                                             sb->s_blocksize);
223                         }
224                         offset = i;
225                         ctx->pos = (ctx->pos & ~(sb->s_blocksize - 1))
226                                 | offset;
227                         file->f_version = inode->i_version;
228                 }
229
230                 while (ctx->pos < inode->i_size
231                        && offset < sb->s_blocksize) {
232                         de = (struct ext4_dir_entry_2 *) (bh->b_data + offset);
233                         if (ext4_check_dir_entry(inode, file, de, bh,
234                                                  bh->b_data, bh->b_size,
235                                                  offset)) {
236                                 /*
237                                  * On error, skip to the next block
238                                  */
239                                 ctx->pos = (ctx->pos |
240                                                 (sb->s_blocksize - 1)) + 1;
241                                 break;
242                         }
243                         offset += ext4_rec_len_from_disk(de->rec_len,
244                                         sb->s_blocksize);
245                         if (le32_to_cpu(de->inode)) {
246                                 if (!ext4_encrypted_inode(inode)) {
247                                         if (!dir_emit(ctx, de->name,
248                                             de->name_len,
249                                             le32_to_cpu(de->inode),
250                                             get_dtype(sb, de->file_type)))
251                                                 goto done;
252                                 } else {
253                                         int save_len = fname_crypto_str.len;
254
255                                         /* Directory is encrypted */
256                                         err = ext4_fname_disk_to_usr(inode,
257                                                 NULL, de, &fname_crypto_str);
258                                         fname_crypto_str.len = save_len;
259                                         if (err < 0)
260                                                 goto errout;
261                                         if (!dir_emit(ctx,
262                                             fname_crypto_str.name, err,
263                                             le32_to_cpu(de->inode),
264                                             get_dtype(sb, de->file_type)))
265                                                 goto done;
266                                 }
267                         }
268                         ctx->pos += ext4_rec_len_from_disk(de->rec_len,
269                                                 sb->s_blocksize);
270                 }
271                 if ((ctx->pos < inode->i_size) && !dir_relax(inode))
272                         goto done;
273                 brelse(bh);
274                 bh = NULL;
275                 offset = 0;
276         }
277 done:
278         err = 0;
279 errout:
280 #ifdef CONFIG_EXT4_FS_ENCRYPTION
281         ext4_fname_crypto_free_buffer(&fname_crypto_str);
282 #endif
283         brelse(bh);
284         return err;
285 }
286
287 static inline int is_32bit_api(void)
288 {
289 #ifdef CONFIG_COMPAT
290         return is_compat_task();
291 #else
292         return (BITS_PER_LONG == 32);
293 #endif
294 }
295
296 /*
297  * These functions convert from the major/minor hash to an f_pos
298  * value for dx directories
299  *
300  * Upper layer (for example NFS) should specify FMODE_32BITHASH or
301  * FMODE_64BITHASH explicitly. On the other hand, we allow ext4 to be mounted
302  * directly on both 32-bit and 64-bit nodes, under such case, neither
303  * FMODE_32BITHASH nor FMODE_64BITHASH is specified.
304  */
305 static inline loff_t hash2pos(struct file *filp, __u32 major, __u32 minor)
306 {
307         if ((filp->f_mode & FMODE_32BITHASH) ||
308             (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
309                 return major >> 1;
310         else
311                 return ((__u64)(major >> 1) << 32) | (__u64)minor;
312 }
313
314 static inline __u32 pos2maj_hash(struct file *filp, loff_t pos)
315 {
316         if ((filp->f_mode & FMODE_32BITHASH) ||
317             (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
318                 return (pos << 1) & 0xffffffff;
319         else
320                 return ((pos >> 32) << 1) & 0xffffffff;
321 }
322
323 static inline __u32 pos2min_hash(struct file *filp, loff_t pos)
324 {
325         if ((filp->f_mode & FMODE_32BITHASH) ||
326             (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
327                 return 0;
328         else
329                 return pos & 0xffffffff;
330 }
331
332 /*
333  * Return 32- or 64-bit end-of-file for dx directories
334  */
335 static inline loff_t ext4_get_htree_eof(struct file *filp)
336 {
337         if ((filp->f_mode & FMODE_32BITHASH) ||
338             (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
339                 return EXT4_HTREE_EOF_32BIT;
340         else
341                 return EXT4_HTREE_EOF_64BIT;
342 }
343
344
345 /*
346  * ext4_dir_llseek() calls generic_file_llseek_size to handle htree
347  * directories, where the "offset" is in terms of the filename hash
348  * value instead of the byte offset.
349  *
350  * Because we may return a 64-bit hash that is well beyond offset limits,
351  * we need to pass the max hash as the maximum allowable offset in
352  * the htree directory case.
353  *
354  * For non-htree, ext4_llseek already chooses the proper max offset.
355  */
356 static loff_t ext4_dir_llseek(struct file *file, loff_t offset, int whence)
357 {
358         struct inode *inode = file->f_mapping->host;
359         int dx_dir = is_dx_dir(inode);
360         loff_t htree_max = ext4_get_htree_eof(file);
361
362         if (likely(dx_dir))
363                 return generic_file_llseek_size(file, offset, whence,
364                                                     htree_max, htree_max);
365         else
366                 return ext4_llseek(file, offset, whence);
367 }
368
369 /*
370  * This structure holds the nodes of the red-black tree used to store
371  * the directory entry in hash order.
372  */
373 struct fname {
374         __u32           hash;
375         __u32           minor_hash;
376         struct rb_node  rb_hash;
377         struct fname    *next;
378         __u32           inode;
379         __u8            name_len;
380         __u8            file_type;
381         char            name[0];
382 };
383
384 /*
385  * This functoin implements a non-recursive way of freeing all of the
386  * nodes in the red-black tree.
387  */
388 static void free_rb_tree_fname(struct rb_root *root)
389 {
390         struct fname *fname, *next;
391
392         rbtree_postorder_for_each_entry_safe(fname, next, root, rb_hash)
393                 while (fname) {
394                         struct fname *old = fname;
395                         fname = fname->next;
396                         kfree(old);
397                 }
398
399         *root = RB_ROOT;
400 }
401
402
403 static struct dir_private_info *ext4_htree_create_dir_info(struct file *filp,
404                                                            loff_t pos)
405 {
406         struct dir_private_info *p;
407
408         p = kzalloc(sizeof(struct dir_private_info), GFP_KERNEL);
409         if (!p)
410                 return NULL;
411         p->curr_hash = pos2maj_hash(filp, pos);
412         p->curr_minor_hash = pos2min_hash(filp, pos);
413         return p;
414 }
415
416 void ext4_htree_free_dir_info(struct dir_private_info *p)
417 {
418         free_rb_tree_fname(&p->root);
419         kfree(p);
420 }
421
422 /*
423  * Given a directory entry, enter it into the fname rb tree.
424  *
425  * When filename encryption is enabled, the dirent will hold the
426  * encrypted filename, while the htree will hold decrypted filename.
427  * The decrypted filename is passed in via ent_name.  parameter.
428  */
429 int ext4_htree_store_dirent(struct file *dir_file, __u32 hash,
430                              __u32 minor_hash,
431                             struct ext4_dir_entry_2 *dirent,
432                             struct ext4_str *ent_name)
433 {
434         struct rb_node **p, *parent = NULL;
435         struct fname *fname, *new_fn;
436         struct dir_private_info *info;
437         int len;
438
439         info = dir_file->private_data;
440         p = &info->root.rb_node;
441
442         /* Create and allocate the fname structure */
443         len = sizeof(struct fname) + ent_name->len + 1;
444         new_fn = kzalloc(len, GFP_KERNEL);
445         if (!new_fn)
446                 return -ENOMEM;
447         new_fn->hash = hash;
448         new_fn->minor_hash = minor_hash;
449         new_fn->inode = le32_to_cpu(dirent->inode);
450         new_fn->name_len = ent_name->len;
451         new_fn->file_type = dirent->file_type;
452         memcpy(new_fn->name, ent_name->name, ent_name->len);
453         new_fn->name[ent_name->len] = 0;
454
455         while (*p) {
456                 parent = *p;
457                 fname = rb_entry(parent, struct fname, rb_hash);
458
459                 /*
460                  * If the hash and minor hash match up, then we put
461                  * them on a linked list.  This rarely happens...
462                  */
463                 if ((new_fn->hash == fname->hash) &&
464                     (new_fn->minor_hash == fname->minor_hash)) {
465                         new_fn->next = fname->next;
466                         fname->next = new_fn;
467                         return 0;
468                 }
469
470                 if (new_fn->hash < fname->hash)
471                         p = &(*p)->rb_left;
472                 else if (new_fn->hash > fname->hash)
473                         p = &(*p)->rb_right;
474                 else if (new_fn->minor_hash < fname->minor_hash)
475                         p = &(*p)->rb_left;
476                 else /* if (new_fn->minor_hash > fname->minor_hash) */
477                         p = &(*p)->rb_right;
478         }
479
480         rb_link_node(&new_fn->rb_hash, parent, p);
481         rb_insert_color(&new_fn->rb_hash, &info->root);
482         return 0;
483 }
484
485
486
487 /*
488  * This is a helper function for ext4_dx_readdir.  It calls filldir
489  * for all entres on the fname linked list.  (Normally there is only
490  * one entry on the linked list, unless there are 62 bit hash collisions.)
491  */
492 static int call_filldir(struct file *file, struct dir_context *ctx,
493                         struct fname *fname)
494 {
495         struct dir_private_info *info = file->private_data;
496         struct inode *inode = file_inode(file);
497         struct super_block *sb = inode->i_sb;
498
499         if (!fname) {
500                 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: comm %s: "
501                          "called with null fname?!?", __func__, __LINE__,
502                          inode->i_ino, current->comm);
503                 return 0;
504         }
505         ctx->pos = hash2pos(file, fname->hash, fname->minor_hash);
506         while (fname) {
507                 if (!dir_emit(ctx, fname->name,
508                                 fname->name_len,
509                                 fname->inode,
510                                 get_dtype(sb, fname->file_type))) {
511                         info->extra_fname = fname;
512                         return 1;
513                 }
514                 fname = fname->next;
515         }
516         return 0;
517 }
518
519 static int ext4_dx_readdir(struct file *file, struct dir_context *ctx)
520 {
521         struct dir_private_info *info = file->private_data;
522         struct inode *inode = file_inode(file);
523         struct fname *fname;
524         int ret = 0;
525
526         if (!info) {
527                 info = ext4_htree_create_dir_info(file, ctx->pos);
528                 if (!info)
529                         return -ENOMEM;
530                 file->private_data = info;
531         }
532
533         if (ctx->pos == ext4_get_htree_eof(file))
534                 return 0;       /* EOF */
535
536         /* Some one has messed with f_pos; reset the world */
537         if (info->last_pos != ctx->pos) {
538                 free_rb_tree_fname(&info->root);
539                 info->curr_node = NULL;
540                 info->extra_fname = NULL;
541                 info->curr_hash = pos2maj_hash(file, ctx->pos);
542                 info->curr_minor_hash = pos2min_hash(file, ctx->pos);
543         }
544
545         /*
546          * If there are any leftover names on the hash collision
547          * chain, return them first.
548          */
549         if (info->extra_fname) {
550                 if (call_filldir(file, ctx, info->extra_fname))
551                         goto finished;
552                 info->extra_fname = NULL;
553                 goto next_node;
554         } else if (!info->curr_node)
555                 info->curr_node = rb_first(&info->root);
556
557         while (1) {
558                 /*
559                  * Fill the rbtree if we have no more entries,
560                  * or the inode has changed since we last read in the
561                  * cached entries.
562                  */
563                 if ((!info->curr_node) ||
564                     (file->f_version != inode->i_version)) {
565                         info->curr_node = NULL;
566                         free_rb_tree_fname(&info->root);
567                         file->f_version = inode->i_version;
568                         ret = ext4_htree_fill_tree(file, info->curr_hash,
569                                                    info->curr_minor_hash,
570                                                    &info->next_hash);
571                         if (ret < 0)
572                                 goto finished;
573                         if (ret == 0) {
574                                 ctx->pos = ext4_get_htree_eof(file);
575                                 break;
576                         }
577                         info->curr_node = rb_first(&info->root);
578                 }
579
580                 fname = rb_entry(info->curr_node, struct fname, rb_hash);
581                 info->curr_hash = fname->hash;
582                 info->curr_minor_hash = fname->minor_hash;
583                 if (call_filldir(file, ctx, fname))
584                         break;
585         next_node:
586                 info->curr_node = rb_next(info->curr_node);
587                 if (info->curr_node) {
588                         fname = rb_entry(info->curr_node, struct fname,
589                                          rb_hash);
590                         info->curr_hash = fname->hash;
591                         info->curr_minor_hash = fname->minor_hash;
592                 } else {
593                         if (info->next_hash == ~0) {
594                                 ctx->pos = ext4_get_htree_eof(file);
595                                 break;
596                         }
597                         info->curr_hash = info->next_hash;
598                         info->curr_minor_hash = 0;
599                 }
600         }
601 finished:
602         info->last_pos = ctx->pos;
603         return ret < 0 ? ret : 0;
604 }
605
606 static int ext4_dir_open(struct inode * inode, struct file * filp)
607 {
608         if (ext4_encrypted_inode(inode))
609                 return ext4_get_encryption_info(inode) ? -EACCES : 0;
610         return 0;
611 }
612
613 static int ext4_release_dir(struct inode *inode, struct file *filp)
614 {
615         if (filp->private_data)
616                 ext4_htree_free_dir_info(filp->private_data);
617
618         return 0;
619 }
620
621 int ext4_check_all_de(struct inode *dir, struct buffer_head *bh, void *buf,
622                       int buf_size)
623 {
624         struct ext4_dir_entry_2 *de;
625         int nlen, rlen;
626         unsigned int offset = 0;
627         char *top;
628
629         de = (struct ext4_dir_entry_2 *)buf;
630         top = buf + buf_size;
631         while ((char *) de < top) {
632                 if (ext4_check_dir_entry(dir, NULL, de, bh,
633                                          buf, buf_size, offset))
634                         return -EFSCORRUPTED;
635                 nlen = EXT4_DIR_REC_LEN(de->name_len);
636                 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
637                 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
638                 offset += rlen;
639         }
640         if ((char *) de > top)
641                 return -EFSCORRUPTED;
642
643         return 0;
644 }
645
646 const struct file_operations ext4_dir_operations = {
647         .llseek         = ext4_dir_llseek,
648         .read           = generic_read_dir,
649         .iterate        = ext4_readdir,
650         .unlocked_ioctl = ext4_ioctl,
651 #ifdef CONFIG_COMPAT
652         .compat_ioctl   = ext4_compat_ioctl,
653 #endif
654         .fsync          = ext4_sync_file,
655         .open           = ext4_dir_open,
656         .release        = ext4_release_dir,
657 };