Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / 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_COMPAT_FEATURE(inode->i_sb,
44                      EXT4_FEATURE_COMPAT_DIR_INDEX) &&
45             ((ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) ||
46              ((inode->i_size >> sb->s_blocksize_bits) == 1) ||
47              ext4_has_inline_data(inode)))
48                 return 1;
49
50         return 0;
51 }
52
53 /*
54  * Return 0 if the directory entry is OK, and 1 if there is a problem
55  *
56  * Note: this is the opposite of what ext2 and ext3 historically returned...
57  *
58  * bh passed here can be an inode block or a dir data block, depending
59  * on the inode inline data flag.
60  */
61 int __ext4_check_dir_entry(const char *function, unsigned int line,
62                            struct inode *dir, struct file *filp,
63                            struct ext4_dir_entry_2 *de,
64                            struct buffer_head *bh, char *buf, int size,
65                            unsigned int offset)
66 {
67         const char *error_msg = NULL;
68         const int rlen = ext4_rec_len_from_disk(de->rec_len,
69                                                 dir->i_sb->s_blocksize);
70
71         if (unlikely(rlen < EXT4_DIR_REC_LEN(1)))
72                 error_msg = "rec_len is smaller than minimal";
73         else if (unlikely(rlen % 4 != 0))
74                 error_msg = "rec_len % 4 != 0";
75         else if (unlikely(rlen < EXT4_DIR_REC_LEN(de->name_len)))
76                 error_msg = "rec_len is too small for name_len";
77         else if (unlikely(((char *) de - buf) + rlen > size))
78                 error_msg = "directory entry across range";
79         else if (unlikely(le32_to_cpu(de->inode) >
80                         le32_to_cpu(EXT4_SB(dir->i_sb)->s_es->s_inodes_count)))
81                 error_msg = "inode out of bounds";
82         else
83                 return 0;
84
85         if (filp)
86                 ext4_error_file(filp, function, line, bh->b_blocknr,
87                                 "bad entry in directory: %s - offset=%u(%u), "
88                                 "inode=%u, rec_len=%d, name_len=%d",
89                                 error_msg, (unsigned) (offset % size),
90                                 offset, le32_to_cpu(de->inode),
91                                 rlen, de->name_len);
92         else
93                 ext4_error_inode(dir, function, line, bh->b_blocknr,
94                                 "bad entry in directory: %s - offset=%u(%u), "
95                                 "inode=%u, rec_len=%d, name_len=%d",
96                                 error_msg, (unsigned) (offset % size),
97                                 offset, le32_to_cpu(de->inode),
98                                 rlen, de->name_len);
99
100         return 1;
101 }
102
103 static int ext4_readdir(struct file *file, struct dir_context *ctx)
104 {
105         unsigned int offset;
106         int i;
107         struct ext4_dir_entry_2 *de;
108         int err;
109         struct inode *inode = file_inode(file);
110         struct super_block *sb = inode->i_sb;
111         struct buffer_head *bh = NULL;
112         int dir_has_error = 0;
113         struct ext4_fname_crypto_ctx *enc_ctx = NULL;
114         struct ext4_str fname_crypto_str = {.name = NULL, .len = 0};
115
116         if (is_dx_dir(inode)) {
117                 err = ext4_dx_readdir(file, ctx);
118                 if (err != ERR_BAD_DX_DIR) {
119                         return err;
120                 }
121                 /*
122                  * We don't set the inode dirty flag since it's not
123                  * critical that it get flushed back to the disk.
124                  */
125                 ext4_clear_inode_flag(file_inode(file),
126                                       EXT4_INODE_INDEX);
127         }
128
129         if (ext4_has_inline_data(inode)) {
130                 int has_inline_data = 1;
131                 err = ext4_read_inline_dir(file, ctx,
132                                            &has_inline_data);
133                 if (has_inline_data)
134                         return err;
135         }
136
137         enc_ctx = ext4_get_fname_crypto_ctx(inode, EXT4_NAME_LEN);
138         if (IS_ERR(enc_ctx))
139                 return PTR_ERR(enc_ctx);
140         if (enc_ctx) {
141                 err = ext4_fname_crypto_alloc_buffer(enc_ctx, EXT4_NAME_LEN,
142                                                      &fname_crypto_str);
143                 if (err < 0) {
144                         ext4_put_fname_crypto_ctx(&enc_ctx);
145                         return err;
146                 }
147         }
148
149         offset = ctx->pos & (sb->s_blocksize - 1);
150
151         while (ctx->pos < inode->i_size) {
152                 struct ext4_map_blocks map;
153
154                 map.m_lblk = ctx->pos >> EXT4_BLOCK_SIZE_BITS(sb);
155                 map.m_len = 1;
156                 err = ext4_map_blocks(NULL, inode, &map, 0);
157                 if (err > 0) {
158                         pgoff_t index = map.m_pblk >>
159                                         (PAGE_CACHE_SHIFT - inode->i_blkbits);
160                         if (!ra_has_index(&file->f_ra, index))
161                                 page_cache_sync_readahead(
162                                         sb->s_bdev->bd_inode->i_mapping,
163                                         &file->f_ra, file,
164                                         index, 1);
165                         file->f_ra.prev_pos = (loff_t)index << PAGE_CACHE_SHIFT;
166                         bh = ext4_bread(NULL, inode, map.m_lblk, 0);
167                         if (IS_ERR(bh))
168                                 return PTR_ERR(bh);
169                 }
170
171                 if (!bh) {
172                         if (!dir_has_error) {
173                                 EXT4_ERROR_FILE(file, 0,
174                                                 "directory contains a "
175                                                 "hole at offset %llu",
176                                            (unsigned long long) ctx->pos);
177                                 dir_has_error = 1;
178                         }
179                         /* corrupt size?  Maybe no more blocks to read */
180                         if (ctx->pos > inode->i_blocks << 9)
181                                 break;
182                         ctx->pos += sb->s_blocksize - offset;
183                         continue;
184                 }
185
186                 /* Check the checksum */
187                 if (!buffer_verified(bh) &&
188                     !ext4_dirent_csum_verify(inode,
189                                 (struct ext4_dir_entry *)bh->b_data)) {
190                         EXT4_ERROR_FILE(file, 0, "directory fails checksum "
191                                         "at offset %llu",
192                                         (unsigned long long)ctx->pos);
193                         ctx->pos += sb->s_blocksize - offset;
194                         brelse(bh);
195                         bh = NULL;
196                         continue;
197                 }
198                 set_buffer_verified(bh);
199
200                 /* If the dir block has changed since the last call to
201                  * readdir(2), then we might be pointing to an invalid
202                  * dirent right now.  Scan from the start of the block
203                  * to make sure. */
204                 if (file->f_version != inode->i_version) {
205                         for (i = 0; i < sb->s_blocksize && i < offset; ) {
206                                 de = (struct ext4_dir_entry_2 *)
207                                         (bh->b_data + i);
208                                 /* It's too expensive to do a full
209                                  * dirent test each time round this
210                                  * loop, but we do have to test at
211                                  * least that it is non-zero.  A
212                                  * failure will be detected in the
213                                  * dirent test below. */
214                                 if (ext4_rec_len_from_disk(de->rec_len,
215                                         sb->s_blocksize) < EXT4_DIR_REC_LEN(1))
216                                         break;
217                                 i += ext4_rec_len_from_disk(de->rec_len,
218                                                             sb->s_blocksize);
219                         }
220                         offset = i;
221                         ctx->pos = (ctx->pos & ~(sb->s_blocksize - 1))
222                                 | offset;
223                         file->f_version = inode->i_version;
224                 }
225
226                 while (ctx->pos < inode->i_size
227                        && offset < sb->s_blocksize) {
228                         de = (struct ext4_dir_entry_2 *) (bh->b_data + offset);
229                         if (ext4_check_dir_entry(inode, file, de, bh,
230                                                  bh->b_data, bh->b_size,
231                                                  offset)) {
232                                 /*
233                                  * On error, skip to the next block
234                                  */
235                                 ctx->pos = (ctx->pos |
236                                                 (sb->s_blocksize - 1)) + 1;
237                                 break;
238                         }
239                         offset += ext4_rec_len_from_disk(de->rec_len,
240                                         sb->s_blocksize);
241                         if (le32_to_cpu(de->inode)) {
242                                 if (enc_ctx == NULL) {
243                                         /* Directory is not encrypted */
244                                         if (!dir_emit(ctx, de->name,
245                                             de->name_len,
246                                             le32_to_cpu(de->inode),
247                                             get_dtype(sb, de->file_type)))
248                                                 goto done;
249                                 } else {
250                                         /* Directory is encrypted */
251                                         err = ext4_fname_disk_to_usr(enc_ctx,
252                                                 NULL, de, &fname_crypto_str);
253                                         if (err < 0)
254                                                 goto errout;
255                                         if (!dir_emit(ctx,
256                                             fname_crypto_str.name, err,
257                                             le32_to_cpu(de->inode),
258                                             get_dtype(sb, de->file_type)))
259                                                 goto done;
260                                 }
261                         }
262                         ctx->pos += ext4_rec_len_from_disk(de->rec_len,
263                                                 sb->s_blocksize);
264                 }
265                 if ((ctx->pos < inode->i_size) && !dir_relax(inode))
266                         goto done;
267                 brelse(bh);
268                 bh = NULL;
269                 offset = 0;
270         }
271 done:
272         err = 0;
273 errout:
274 #ifdef CONFIG_EXT4_FS_ENCRYPTION
275         ext4_put_fname_crypto_ctx(&enc_ctx);
276         ext4_fname_crypto_free_buffer(&fname_crypto_str);
277 #endif
278         brelse(bh);
279         return err;
280 }
281
282 static inline int is_32bit_api(void)
283 {
284 #ifdef CONFIG_COMPAT
285         return is_compat_task();
286 #else
287         return (BITS_PER_LONG == 32);
288 #endif
289 }
290
291 /*
292  * These functions convert from the major/minor hash to an f_pos
293  * value for dx directories
294  *
295  * Upper layer (for example NFS) should specify FMODE_32BITHASH or
296  * FMODE_64BITHASH explicitly. On the other hand, we allow ext4 to be mounted
297  * directly on both 32-bit and 64-bit nodes, under such case, neither
298  * FMODE_32BITHASH nor FMODE_64BITHASH is specified.
299  */
300 static inline loff_t hash2pos(struct file *filp, __u32 major, __u32 minor)
301 {
302         if ((filp->f_mode & FMODE_32BITHASH) ||
303             (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
304                 return major >> 1;
305         else
306                 return ((__u64)(major >> 1) << 32) | (__u64)minor;
307 }
308
309 static inline __u32 pos2maj_hash(struct file *filp, loff_t pos)
310 {
311         if ((filp->f_mode & FMODE_32BITHASH) ||
312             (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
313                 return (pos << 1) & 0xffffffff;
314         else
315                 return ((pos >> 32) << 1) & 0xffffffff;
316 }
317
318 static inline __u32 pos2min_hash(struct file *filp, loff_t pos)
319 {
320         if ((filp->f_mode & FMODE_32BITHASH) ||
321             (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
322                 return 0;
323         else
324                 return pos & 0xffffffff;
325 }
326
327 /*
328  * Return 32- or 64-bit end-of-file for dx directories
329  */
330 static inline loff_t ext4_get_htree_eof(struct file *filp)
331 {
332         if ((filp->f_mode & FMODE_32BITHASH) ||
333             (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
334                 return EXT4_HTREE_EOF_32BIT;
335         else
336                 return EXT4_HTREE_EOF_64BIT;
337 }
338
339
340 /*
341  * ext4_dir_llseek() calls generic_file_llseek_size to handle htree
342  * directories, where the "offset" is in terms of the filename hash
343  * value instead of the byte offset.
344  *
345  * Because we may return a 64-bit hash that is well beyond offset limits,
346  * we need to pass the max hash as the maximum allowable offset in
347  * the htree directory case.
348  *
349  * For non-htree, ext4_llseek already chooses the proper max offset.
350  */
351 static loff_t ext4_dir_llseek(struct file *file, loff_t offset, int whence)
352 {
353         struct inode *inode = file->f_mapping->host;
354         int dx_dir = is_dx_dir(inode);
355         loff_t htree_max = ext4_get_htree_eof(file);
356
357         if (likely(dx_dir))
358                 return generic_file_llseek_size(file, offset, whence,
359                                                     htree_max, htree_max);
360         else
361                 return ext4_llseek(file, offset, whence);
362 }
363
364 /*
365  * This structure holds the nodes of the red-black tree used to store
366  * the directory entry in hash order.
367  */
368 struct fname {
369         __u32           hash;
370         __u32           minor_hash;
371         struct rb_node  rb_hash;
372         struct fname    *next;
373         __u32           inode;
374         __u8            name_len;
375         __u8            file_type;
376         char            name[0];
377 };
378
379 /*
380  * This functoin implements a non-recursive way of freeing all of the
381  * nodes in the red-black tree.
382  */
383 static void free_rb_tree_fname(struct rb_root *root)
384 {
385         struct fname *fname, *next;
386
387         rbtree_postorder_for_each_entry_safe(fname, next, root, rb_hash)
388                 while (fname) {
389                         struct fname *old = fname;
390                         fname = fname->next;
391                         kfree(old);
392                 }
393
394         *root = RB_ROOT;
395 }
396
397
398 static struct dir_private_info *ext4_htree_create_dir_info(struct file *filp,
399                                                            loff_t pos)
400 {
401         struct dir_private_info *p;
402
403         p = kzalloc(sizeof(struct dir_private_info), GFP_KERNEL);
404         if (!p)
405                 return NULL;
406         p->curr_hash = pos2maj_hash(filp, pos);
407         p->curr_minor_hash = pos2min_hash(filp, pos);
408         return p;
409 }
410
411 void ext4_htree_free_dir_info(struct dir_private_info *p)
412 {
413         free_rb_tree_fname(&p->root);
414         kfree(p);
415 }
416
417 /*
418  * Given a directory entry, enter it into the fname rb tree.
419  *
420  * When filename encryption is enabled, the dirent will hold the
421  * encrypted filename, while the htree will hold decrypted filename.
422  * The decrypted filename is passed in via ent_name.  parameter.
423  */
424 int ext4_htree_store_dirent(struct file *dir_file, __u32 hash,
425                              __u32 minor_hash,
426                             struct ext4_dir_entry_2 *dirent,
427                             struct ext4_str *ent_name)
428 {
429         struct rb_node **p, *parent = NULL;
430         struct fname *fname, *new_fn;
431         struct dir_private_info *info;
432         int len;
433
434         info = dir_file->private_data;
435         p = &info->root.rb_node;
436
437         /* Create and allocate the fname structure */
438         len = sizeof(struct fname) + ent_name->len + 1;
439         new_fn = kzalloc(len, GFP_KERNEL);
440         if (!new_fn)
441                 return -ENOMEM;
442         new_fn->hash = hash;
443         new_fn->minor_hash = minor_hash;
444         new_fn->inode = le32_to_cpu(dirent->inode);
445         new_fn->name_len = ent_name->len;
446         new_fn->file_type = dirent->file_type;
447         memcpy(new_fn->name, ent_name->name, ent_name->len);
448         new_fn->name[ent_name->len] = 0;
449
450         while (*p) {
451                 parent = *p;
452                 fname = rb_entry(parent, struct fname, rb_hash);
453
454                 /*
455                  * If the hash and minor hash match up, then we put
456                  * them on a linked list.  This rarely happens...
457                  */
458                 if ((new_fn->hash == fname->hash) &&
459                     (new_fn->minor_hash == fname->minor_hash)) {
460                         new_fn->next = fname->next;
461                         fname->next = new_fn;
462                         return 0;
463                 }
464
465                 if (new_fn->hash < fname->hash)
466                         p = &(*p)->rb_left;
467                 else if (new_fn->hash > fname->hash)
468                         p = &(*p)->rb_right;
469                 else if (new_fn->minor_hash < fname->minor_hash)
470                         p = &(*p)->rb_left;
471                 else /* if (new_fn->minor_hash > fname->minor_hash) */
472                         p = &(*p)->rb_right;
473         }
474
475         rb_link_node(&new_fn->rb_hash, parent, p);
476         rb_insert_color(&new_fn->rb_hash, &info->root);
477         return 0;
478 }
479
480
481
482 /*
483  * This is a helper function for ext4_dx_readdir.  It calls filldir
484  * for all entres on the fname linked list.  (Normally there is only
485  * one entry on the linked list, unless there are 62 bit hash collisions.)
486  */
487 static int call_filldir(struct file *file, struct dir_context *ctx,
488                         struct fname *fname)
489 {
490         struct dir_private_info *info = file->private_data;
491         struct inode *inode = file_inode(file);
492         struct super_block *sb = inode->i_sb;
493
494         if (!fname) {
495                 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: comm %s: "
496                          "called with null fname?!?", __func__, __LINE__,
497                          inode->i_ino, current->comm);
498                 return 0;
499         }
500         ctx->pos = hash2pos(file, fname->hash, fname->minor_hash);
501         while (fname) {
502                 if (!dir_emit(ctx, fname->name,
503                                 fname->name_len,
504                                 fname->inode,
505                                 get_dtype(sb, fname->file_type))) {
506                         info->extra_fname = fname;
507                         return 1;
508                 }
509                 fname = fname->next;
510         }
511         return 0;
512 }
513
514 static int ext4_dx_readdir(struct file *file, struct dir_context *ctx)
515 {
516         struct dir_private_info *info = file->private_data;
517         struct inode *inode = file_inode(file);
518         struct fname *fname;
519         int     ret;
520
521         if (!info) {
522                 info = ext4_htree_create_dir_info(file, ctx->pos);
523                 if (!info)
524                         return -ENOMEM;
525                 file->private_data = info;
526         }
527
528         if (ctx->pos == ext4_get_htree_eof(file))
529                 return 0;       /* EOF */
530
531         /* Some one has messed with f_pos; reset the world */
532         if (info->last_pos != ctx->pos) {
533                 free_rb_tree_fname(&info->root);
534                 info->curr_node = NULL;
535                 info->extra_fname = NULL;
536                 info->curr_hash = pos2maj_hash(file, ctx->pos);
537                 info->curr_minor_hash = pos2min_hash(file, ctx->pos);
538         }
539
540         /*
541          * If there are any leftover names on the hash collision
542          * chain, return them first.
543          */
544         if (info->extra_fname) {
545                 if (call_filldir(file, ctx, info->extra_fname))
546                         goto finished;
547                 info->extra_fname = NULL;
548                 goto next_node;
549         } else if (!info->curr_node)
550                 info->curr_node = rb_first(&info->root);
551
552         while (1) {
553                 /*
554                  * Fill the rbtree if we have no more entries,
555                  * or the inode has changed since we last read in the
556                  * cached entries.
557                  */
558                 if ((!info->curr_node) ||
559                     (file->f_version != inode->i_version)) {
560                         info->curr_node = NULL;
561                         free_rb_tree_fname(&info->root);
562                         file->f_version = inode->i_version;
563                         ret = ext4_htree_fill_tree(file, info->curr_hash,
564                                                    info->curr_minor_hash,
565                                                    &info->next_hash);
566                         if (ret < 0)
567                                 return ret;
568                         if (ret == 0) {
569                                 ctx->pos = ext4_get_htree_eof(file);
570                                 break;
571                         }
572                         info->curr_node = rb_first(&info->root);
573                 }
574
575                 fname = rb_entry(info->curr_node, struct fname, rb_hash);
576                 info->curr_hash = fname->hash;
577                 info->curr_minor_hash = fname->minor_hash;
578                 if (call_filldir(file, ctx, fname))
579                         break;
580         next_node:
581                 info->curr_node = rb_next(info->curr_node);
582                 if (info->curr_node) {
583                         fname = rb_entry(info->curr_node, struct fname,
584                                          rb_hash);
585                         info->curr_hash = fname->hash;
586                         info->curr_minor_hash = fname->minor_hash;
587                 } else {
588                         if (info->next_hash == ~0) {
589                                 ctx->pos = ext4_get_htree_eof(file);
590                                 break;
591                         }
592                         info->curr_hash = info->next_hash;
593                         info->curr_minor_hash = 0;
594                 }
595         }
596 finished:
597         info->last_pos = ctx->pos;
598         return 0;
599 }
600
601 static int ext4_release_dir(struct inode *inode, struct file *filp)
602 {
603         if (filp->private_data)
604                 ext4_htree_free_dir_info(filp->private_data);
605
606         return 0;
607 }
608
609 int ext4_check_all_de(struct inode *dir, struct buffer_head *bh, void *buf,
610                       int buf_size)
611 {
612         struct ext4_dir_entry_2 *de;
613         int nlen, rlen;
614         unsigned int offset = 0;
615         char *top;
616
617         de = (struct ext4_dir_entry_2 *)buf;
618         top = buf + buf_size;
619         while ((char *) de < top) {
620                 if (ext4_check_dir_entry(dir, NULL, de, bh,
621                                          buf, buf_size, offset))
622                         return -EIO;
623                 nlen = EXT4_DIR_REC_LEN(de->name_len);
624                 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
625                 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
626                 offset += rlen;
627         }
628         if ((char *) de > top)
629                 return -EIO;
630
631         return 0;
632 }
633
634 const struct file_operations ext4_dir_operations = {
635         .llseek         = ext4_dir_llseek,
636         .read           = generic_read_dir,
637         .iterate        = ext4_readdir,
638         .unlocked_ioctl = ext4_ioctl,
639 #ifdef CONFIG_COMPAT
640         .compat_ioctl   = ext4_compat_ioctl,
641 #endif
642         .fsync          = ext4_sync_file,
643         .release        = ext4_release_dir,
644 };