]> git.kernelconcepts.de Git - karo-tx-linux.git/blob - fs/ext4/dir.c
Merge remote-tracking branch 'ext4/dev'
[karo-tx-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 ext4_str fname_crypto_str = {.name = NULL, .len = 0};
113
114         if (is_dx_dir(inode)) {
115                 err = ext4_dx_readdir(file, ctx);
116                 if (err != ERR_BAD_DX_DIR) {
117                         return err;
118                 }
119                 /*
120                  * We don't set the inode dirty flag since it's not
121                  * critical that it get flushed back to the disk.
122                  */
123                 ext4_clear_inode_flag(file_inode(file),
124                                       EXT4_INODE_INDEX);
125         }
126
127         if (ext4_has_inline_data(inode)) {
128                 int has_inline_data = 1;
129                 err = ext4_read_inline_dir(file, ctx,
130                                            &has_inline_data);
131                 if (has_inline_data)
132                         return err;
133         }
134
135         if (ext4_encrypted_inode(inode)) {
136                 err = ext4_fname_crypto_alloc_buffer(inode, EXT4_NAME_LEN,
137                                                      &fname_crypto_str);
138                 if (err < 0)
139                         return err;
140         }
141
142         offset = ctx->pos & (sb->s_blocksize - 1);
143
144         while (ctx->pos < inode->i_size) {
145                 struct ext4_map_blocks map;
146
147                 map.m_lblk = ctx->pos >> EXT4_BLOCK_SIZE_BITS(sb);
148                 map.m_len = 1;
149                 err = ext4_map_blocks(NULL, inode, &map, 0);
150                 if (err > 0) {
151                         pgoff_t index = map.m_pblk >>
152                                         (PAGE_CACHE_SHIFT - inode->i_blkbits);
153                         if (!ra_has_index(&file->f_ra, index))
154                                 page_cache_sync_readahead(
155                                         sb->s_bdev->bd_inode->i_mapping,
156                                         &file->f_ra, file,
157                                         index, 1);
158                         file->f_ra.prev_pos = (loff_t)index << PAGE_CACHE_SHIFT;
159                         bh = ext4_bread(NULL, inode, map.m_lblk, 0);
160                         if (IS_ERR(bh))
161                                 return PTR_ERR(bh);
162                 }
163
164                 if (!bh) {
165                         if (!dir_has_error) {
166                                 EXT4_ERROR_FILE(file, 0,
167                                                 "directory contains a "
168                                                 "hole at offset %llu",
169                                            (unsigned long long) ctx->pos);
170                                 dir_has_error = 1;
171                         }
172                         /* corrupt size?  Maybe no more blocks to read */
173                         if (ctx->pos > inode->i_blocks << 9)
174                                 break;
175                         ctx->pos += sb->s_blocksize - offset;
176                         continue;
177                 }
178
179                 /* Check the checksum */
180                 if (!buffer_verified(bh) &&
181                     !ext4_dirent_csum_verify(inode,
182                                 (struct ext4_dir_entry *)bh->b_data)) {
183                         EXT4_ERROR_FILE(file, 0, "directory fails checksum "
184                                         "at offset %llu",
185                                         (unsigned long long)ctx->pos);
186                         ctx->pos += sb->s_blocksize - offset;
187                         brelse(bh);
188                         bh = NULL;
189                         continue;
190                 }
191                 set_buffer_verified(bh);
192
193                 /* If the dir block has changed since the last call to
194                  * readdir(2), then we might be pointing to an invalid
195                  * dirent right now.  Scan from the start of the block
196                  * to make sure. */
197                 if (file->f_version != inode->i_version) {
198                         for (i = 0; i < sb->s_blocksize && i < offset; ) {
199                                 de = (struct ext4_dir_entry_2 *)
200                                         (bh->b_data + i);
201                                 /* It's too expensive to do a full
202                                  * dirent test each time round this
203                                  * loop, but we do have to test at
204                                  * least that it is non-zero.  A
205                                  * failure will be detected in the
206                                  * dirent test below. */
207                                 if (ext4_rec_len_from_disk(de->rec_len,
208                                         sb->s_blocksize) < EXT4_DIR_REC_LEN(1))
209                                         break;
210                                 i += ext4_rec_len_from_disk(de->rec_len,
211                                                             sb->s_blocksize);
212                         }
213                         offset = i;
214                         ctx->pos = (ctx->pos & ~(sb->s_blocksize - 1))
215                                 | offset;
216                         file->f_version = inode->i_version;
217                 }
218
219                 while (ctx->pos < inode->i_size
220                        && offset < sb->s_blocksize) {
221                         de = (struct ext4_dir_entry_2 *) (bh->b_data + offset);
222                         if (ext4_check_dir_entry(inode, file, de, bh,
223                                                  bh->b_data, bh->b_size,
224                                                  offset)) {
225                                 /*
226                                  * On error, skip to the next block
227                                  */
228                                 ctx->pos = (ctx->pos |
229                                                 (sb->s_blocksize - 1)) + 1;
230                                 break;
231                         }
232                         offset += ext4_rec_len_from_disk(de->rec_len,
233                                         sb->s_blocksize);
234                         if (le32_to_cpu(de->inode)) {
235                                 if (!ext4_encrypted_inode(inode)) {
236                                         if (!dir_emit(ctx, de->name,
237                                             de->name_len,
238                                             le32_to_cpu(de->inode),
239                                             get_dtype(sb, de->file_type)))
240                                                 goto done;
241                                 } else {
242                                         int save_len = fname_crypto_str.len;
243
244                                         /* Directory is encrypted */
245                                         err = ext4_fname_disk_to_usr(inode,
246                                                 NULL, de, &fname_crypto_str);
247                                         fname_crypto_str.len = save_len;
248                                         if (err < 0)
249                                                 goto errout;
250                                         if (!dir_emit(ctx,
251                                             fname_crypto_str.name, err,
252                                             le32_to_cpu(de->inode),
253                                             get_dtype(sb, de->file_type)))
254                                                 goto done;
255                                 }
256                         }
257                         ctx->pos += ext4_rec_len_from_disk(de->rec_len,
258                                                 sb->s_blocksize);
259                 }
260                 if ((ctx->pos < inode->i_size) && !dir_relax(inode))
261                         goto done;
262                 brelse(bh);
263                 bh = NULL;
264                 offset = 0;
265         }
266 done:
267         err = 0;
268 errout:
269 #ifdef CONFIG_EXT4_FS_ENCRYPTION
270         ext4_fname_crypto_free_buffer(&fname_crypto_str);
271 #endif
272         brelse(bh);
273         return err;
274 }
275
276 static inline int is_32bit_api(void)
277 {
278 #ifdef CONFIG_COMPAT
279         return is_compat_task();
280 #else
281         return (BITS_PER_LONG == 32);
282 #endif
283 }
284
285 /*
286  * These functions convert from the major/minor hash to an f_pos
287  * value for dx directories
288  *
289  * Upper layer (for example NFS) should specify FMODE_32BITHASH or
290  * FMODE_64BITHASH explicitly. On the other hand, we allow ext4 to be mounted
291  * directly on both 32-bit and 64-bit nodes, under such case, neither
292  * FMODE_32BITHASH nor FMODE_64BITHASH is specified.
293  */
294 static inline loff_t hash2pos(struct file *filp, __u32 major, __u32 minor)
295 {
296         if ((filp->f_mode & FMODE_32BITHASH) ||
297             (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
298                 return major >> 1;
299         else
300                 return ((__u64)(major >> 1) << 32) | (__u64)minor;
301 }
302
303 static inline __u32 pos2maj_hash(struct file *filp, loff_t pos)
304 {
305         if ((filp->f_mode & FMODE_32BITHASH) ||
306             (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
307                 return (pos << 1) & 0xffffffff;
308         else
309                 return ((pos >> 32) << 1) & 0xffffffff;
310 }
311
312 static inline __u32 pos2min_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 0;
317         else
318                 return pos & 0xffffffff;
319 }
320
321 /*
322  * Return 32- or 64-bit end-of-file for dx directories
323  */
324 static inline loff_t ext4_get_htree_eof(struct file *filp)
325 {
326         if ((filp->f_mode & FMODE_32BITHASH) ||
327             (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
328                 return EXT4_HTREE_EOF_32BIT;
329         else
330                 return EXT4_HTREE_EOF_64BIT;
331 }
332
333
334 /*
335  * ext4_dir_llseek() calls generic_file_llseek_size to handle htree
336  * directories, where the "offset" is in terms of the filename hash
337  * value instead of the byte offset.
338  *
339  * Because we may return a 64-bit hash that is well beyond offset limits,
340  * we need to pass the max hash as the maximum allowable offset in
341  * the htree directory case.
342  *
343  * For non-htree, ext4_llseek already chooses the proper max offset.
344  */
345 static loff_t ext4_dir_llseek(struct file *file, loff_t offset, int whence)
346 {
347         struct inode *inode = file->f_mapping->host;
348         int dx_dir = is_dx_dir(inode);
349         loff_t htree_max = ext4_get_htree_eof(file);
350
351         if (likely(dx_dir))
352                 return generic_file_llseek_size(file, offset, whence,
353                                                     htree_max, htree_max);
354         else
355                 return ext4_llseek(file, offset, whence);
356 }
357
358 /*
359  * This structure holds the nodes of the red-black tree used to store
360  * the directory entry in hash order.
361  */
362 struct fname {
363         __u32           hash;
364         __u32           minor_hash;
365         struct rb_node  rb_hash;
366         struct fname    *next;
367         __u32           inode;
368         __u8            name_len;
369         __u8            file_type;
370         char            name[0];
371 };
372
373 /*
374  * This functoin implements a non-recursive way of freeing all of the
375  * nodes in the red-black tree.
376  */
377 static void free_rb_tree_fname(struct rb_root *root)
378 {
379         struct fname *fname, *next;
380
381         rbtree_postorder_for_each_entry_safe(fname, next, root, rb_hash)
382                 while (fname) {
383                         struct fname *old = fname;
384                         fname = fname->next;
385                         kfree(old);
386                 }
387
388         *root = RB_ROOT;
389 }
390
391
392 static struct dir_private_info *ext4_htree_create_dir_info(struct file *filp,
393                                                            loff_t pos)
394 {
395         struct dir_private_info *p;
396
397         p = kzalloc(sizeof(struct dir_private_info), GFP_KERNEL);
398         if (!p)
399                 return NULL;
400         p->curr_hash = pos2maj_hash(filp, pos);
401         p->curr_minor_hash = pos2min_hash(filp, pos);
402         return p;
403 }
404
405 void ext4_htree_free_dir_info(struct dir_private_info *p)
406 {
407         free_rb_tree_fname(&p->root);
408         kfree(p);
409 }
410
411 /*
412  * Given a directory entry, enter it into the fname rb tree.
413  *
414  * When filename encryption is enabled, the dirent will hold the
415  * encrypted filename, while the htree will hold decrypted filename.
416  * The decrypted filename is passed in via ent_name.  parameter.
417  */
418 int ext4_htree_store_dirent(struct file *dir_file, __u32 hash,
419                              __u32 minor_hash,
420                             struct ext4_dir_entry_2 *dirent,
421                             struct ext4_str *ent_name)
422 {
423         struct rb_node **p, *parent = NULL;
424         struct fname *fname, *new_fn;
425         struct dir_private_info *info;
426         int len;
427
428         info = dir_file->private_data;
429         p = &info->root.rb_node;
430
431         /* Create and allocate the fname structure */
432         len = sizeof(struct fname) + ent_name->len + 1;
433         new_fn = kzalloc(len, GFP_KERNEL);
434         if (!new_fn)
435                 return -ENOMEM;
436         new_fn->hash = hash;
437         new_fn->minor_hash = minor_hash;
438         new_fn->inode = le32_to_cpu(dirent->inode);
439         new_fn->name_len = ent_name->len;
440         new_fn->file_type = dirent->file_type;
441         memcpy(new_fn->name, ent_name->name, ent_name->len);
442         new_fn->name[ent_name->len] = 0;
443
444         while (*p) {
445                 parent = *p;
446                 fname = rb_entry(parent, struct fname, rb_hash);
447
448                 /*
449                  * If the hash and minor hash match up, then we put
450                  * them on a linked list.  This rarely happens...
451                  */
452                 if ((new_fn->hash == fname->hash) &&
453                     (new_fn->minor_hash == fname->minor_hash)) {
454                         new_fn->next = fname->next;
455                         fname->next = new_fn;
456                         return 0;
457                 }
458
459                 if (new_fn->hash < fname->hash)
460                         p = &(*p)->rb_left;
461                 else if (new_fn->hash > fname->hash)
462                         p = &(*p)->rb_right;
463                 else if (new_fn->minor_hash < fname->minor_hash)
464                         p = &(*p)->rb_left;
465                 else /* if (new_fn->minor_hash > fname->minor_hash) */
466                         p = &(*p)->rb_right;
467         }
468
469         rb_link_node(&new_fn->rb_hash, parent, p);
470         rb_insert_color(&new_fn->rb_hash, &info->root);
471         return 0;
472 }
473
474
475
476 /*
477  * This is a helper function for ext4_dx_readdir.  It calls filldir
478  * for all entres on the fname linked list.  (Normally there is only
479  * one entry on the linked list, unless there are 62 bit hash collisions.)
480  */
481 static int call_filldir(struct file *file, struct dir_context *ctx,
482                         struct fname *fname)
483 {
484         struct dir_private_info *info = file->private_data;
485         struct inode *inode = file_inode(file);
486         struct super_block *sb = inode->i_sb;
487
488         if (!fname) {
489                 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: comm %s: "
490                          "called with null fname?!?", __func__, __LINE__,
491                          inode->i_ino, current->comm);
492                 return 0;
493         }
494         ctx->pos = hash2pos(file, fname->hash, fname->minor_hash);
495         while (fname) {
496                 if (!dir_emit(ctx, fname->name,
497                                 fname->name_len,
498                                 fname->inode,
499                                 get_dtype(sb, fname->file_type))) {
500                         info->extra_fname = fname;
501                         return 1;
502                 }
503                 fname = fname->next;
504         }
505         return 0;
506 }
507
508 static int ext4_dx_readdir(struct file *file, struct dir_context *ctx)
509 {
510         struct dir_private_info *info = file->private_data;
511         struct inode *inode = file_inode(file);
512         struct fname *fname;
513         int     ret;
514
515         if (!info) {
516                 info = ext4_htree_create_dir_info(file, ctx->pos);
517                 if (!info)
518                         return -ENOMEM;
519                 file->private_data = info;
520         }
521
522         if (ctx->pos == ext4_get_htree_eof(file))
523                 return 0;       /* EOF */
524
525         /* Some one has messed with f_pos; reset the world */
526         if (info->last_pos != ctx->pos) {
527                 free_rb_tree_fname(&info->root);
528                 info->curr_node = NULL;
529                 info->extra_fname = NULL;
530                 info->curr_hash = pos2maj_hash(file, ctx->pos);
531                 info->curr_minor_hash = pos2min_hash(file, ctx->pos);
532         }
533
534         /*
535          * If there are any leftover names on the hash collision
536          * chain, return them first.
537          */
538         if (info->extra_fname) {
539                 if (call_filldir(file, ctx, info->extra_fname))
540                         goto finished;
541                 info->extra_fname = NULL;
542                 goto next_node;
543         } else if (!info->curr_node)
544                 info->curr_node = rb_first(&info->root);
545
546         while (1) {
547                 /*
548                  * Fill the rbtree if we have no more entries,
549                  * or the inode has changed since we last read in the
550                  * cached entries.
551                  */
552                 if ((!info->curr_node) ||
553                     (file->f_version != inode->i_version)) {
554                         info->curr_node = NULL;
555                         free_rb_tree_fname(&info->root);
556                         file->f_version = inode->i_version;
557                         ret = ext4_htree_fill_tree(file, info->curr_hash,
558                                                    info->curr_minor_hash,
559                                                    &info->next_hash);
560                         if (ret < 0)
561                                 return ret;
562                         if (ret == 0) {
563                                 ctx->pos = ext4_get_htree_eof(file);
564                                 break;
565                         }
566                         info->curr_node = rb_first(&info->root);
567                 }
568
569                 fname = rb_entry(info->curr_node, struct fname, rb_hash);
570                 info->curr_hash = fname->hash;
571                 info->curr_minor_hash = fname->minor_hash;
572                 if (call_filldir(file, ctx, fname))
573                         break;
574         next_node:
575                 info->curr_node = rb_next(info->curr_node);
576                 if (info->curr_node) {
577                         fname = rb_entry(info->curr_node, struct fname,
578                                          rb_hash);
579                         info->curr_hash = fname->hash;
580                         info->curr_minor_hash = fname->minor_hash;
581                 } else {
582                         if (info->next_hash == ~0) {
583                                 ctx->pos = ext4_get_htree_eof(file);
584                                 break;
585                         }
586                         info->curr_hash = info->next_hash;
587                         info->curr_minor_hash = 0;
588                 }
589         }
590 finished:
591         info->last_pos = ctx->pos;
592         return 0;
593 }
594
595 static int ext4_dir_open(struct inode * inode, struct file * filp)
596 {
597         if (ext4_encrypted_inode(inode))
598                 return ext4_get_encryption_info(inode) ? -EACCES : 0;
599         return 0;
600 }
601
602 static int ext4_release_dir(struct inode *inode, struct file *filp)
603 {
604         if (filp->private_data)
605                 ext4_htree_free_dir_info(filp->private_data);
606
607         return 0;
608 }
609
610 int ext4_check_all_de(struct inode *dir, struct buffer_head *bh, void *buf,
611                       int buf_size)
612 {
613         struct ext4_dir_entry_2 *de;
614         int nlen, rlen;
615         unsigned int offset = 0;
616         char *top;
617
618         de = (struct ext4_dir_entry_2 *)buf;
619         top = buf + buf_size;
620         while ((char *) de < top) {
621                 if (ext4_check_dir_entry(dir, NULL, de, bh,
622                                          buf, buf_size, offset))
623                         return -EFSCORRUPTED;
624                 nlen = EXT4_DIR_REC_LEN(de->name_len);
625                 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
626                 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
627                 offset += rlen;
628         }
629         if ((char *) de > top)
630                 return -EFSCORRUPTED;
631
632         return 0;
633 }
634
635 const struct file_operations ext4_dir_operations = {
636         .llseek         = ext4_dir_llseek,
637         .read           = generic_read_dir,
638         .iterate        = ext4_readdir,
639         .unlocked_ioctl = ext4_ioctl,
640 #ifdef CONFIG_COMPAT
641         .compat_ioctl   = ext4_compat_ioctl,
642 #endif
643         .fsync          = ext4_sync_file,
644         .open           = ext4_dir_open,
645         .release        = ext4_release_dir,
646 };