Merge tag 'configfs-for-4.8' of git://git.infradead.org/users/hch/configfs
[deliverable/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 fstr.len = save_len;
264 if (err < 0)
265 goto errout;
266 if (!dir_emit(ctx,
267 fstr.name, err,
268 le32_to_cpu(de->inode),
269 get_dtype(sb, de->file_type)))
270 goto done;
271 }
272 }
273 ctx->pos += ext4_rec_len_from_disk(de->rec_len,
274 sb->s_blocksize);
275 }
276 if ((ctx->pos < inode->i_size) && !dir_relax_shared(inode))
277 goto done;
278 brelse(bh);
279 bh = NULL;
280 offset = 0;
281 }
282 done:
283 err = 0;
284 errout:
285 #ifdef CONFIG_EXT4_FS_ENCRYPTION
286 fscrypt_fname_free_buffer(&fstr);
287 #endif
288 brelse(bh);
289 return err;
290 }
291
292 static inline int is_32bit_api(void)
293 {
294 #ifdef CONFIG_COMPAT
295 return in_compat_syscall();
296 #else
297 return (BITS_PER_LONG == 32);
298 #endif
299 }
300
301 /*
302 * These functions convert from the major/minor hash to an f_pos
303 * value for dx directories
304 *
305 * Upper layer (for example NFS) should specify FMODE_32BITHASH or
306 * FMODE_64BITHASH explicitly. On the other hand, we allow ext4 to be mounted
307 * directly on both 32-bit and 64-bit nodes, under such case, neither
308 * FMODE_32BITHASH nor FMODE_64BITHASH is specified.
309 */
310 static inline loff_t hash2pos(struct file *filp, __u32 major, __u32 minor)
311 {
312 if ((filp->f_mode & FMODE_32BITHASH) ||
313 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
314 return major >> 1;
315 else
316 return ((__u64)(major >> 1) << 32) | (__u64)minor;
317 }
318
319 static inline __u32 pos2maj_hash(struct file *filp, loff_t pos)
320 {
321 if ((filp->f_mode & FMODE_32BITHASH) ||
322 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
323 return (pos << 1) & 0xffffffff;
324 else
325 return ((pos >> 32) << 1) & 0xffffffff;
326 }
327
328 static inline __u32 pos2min_hash(struct file *filp, loff_t pos)
329 {
330 if ((filp->f_mode & FMODE_32BITHASH) ||
331 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
332 return 0;
333 else
334 return pos & 0xffffffff;
335 }
336
337 /*
338 * Return 32- or 64-bit end-of-file for dx directories
339 */
340 static inline loff_t ext4_get_htree_eof(struct file *filp)
341 {
342 if ((filp->f_mode & FMODE_32BITHASH) ||
343 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
344 return EXT4_HTREE_EOF_32BIT;
345 else
346 return EXT4_HTREE_EOF_64BIT;
347 }
348
349
350 /*
351 * ext4_dir_llseek() calls generic_file_llseek_size to handle htree
352 * directories, where the "offset" is in terms of the filename hash
353 * value instead of the byte offset.
354 *
355 * Because we may return a 64-bit hash that is well beyond offset limits,
356 * we need to pass the max hash as the maximum allowable offset in
357 * the htree directory case.
358 *
359 * For non-htree, ext4_llseek already chooses the proper max offset.
360 */
361 static loff_t ext4_dir_llseek(struct file *file, loff_t offset, int whence)
362 {
363 struct inode *inode = file->f_mapping->host;
364 int dx_dir = is_dx_dir(inode);
365 loff_t htree_max = ext4_get_htree_eof(file);
366
367 if (likely(dx_dir))
368 return generic_file_llseek_size(file, offset, whence,
369 htree_max, htree_max);
370 else
371 return ext4_llseek(file, offset, whence);
372 }
373
374 /*
375 * This structure holds the nodes of the red-black tree used to store
376 * the directory entry in hash order.
377 */
378 struct fname {
379 __u32 hash;
380 __u32 minor_hash;
381 struct rb_node rb_hash;
382 struct fname *next;
383 __u32 inode;
384 __u8 name_len;
385 __u8 file_type;
386 char name[0];
387 };
388
389 /*
390 * This functoin implements a non-recursive way of freeing all of the
391 * nodes in the red-black tree.
392 */
393 static void free_rb_tree_fname(struct rb_root *root)
394 {
395 struct fname *fname, *next;
396
397 rbtree_postorder_for_each_entry_safe(fname, next, root, rb_hash)
398 while (fname) {
399 struct fname *old = fname;
400 fname = fname->next;
401 kfree(old);
402 }
403
404 *root = RB_ROOT;
405 }
406
407
408 static struct dir_private_info *ext4_htree_create_dir_info(struct file *filp,
409 loff_t pos)
410 {
411 struct dir_private_info *p;
412
413 p = kzalloc(sizeof(struct dir_private_info), GFP_KERNEL);
414 if (!p)
415 return NULL;
416 p->curr_hash = pos2maj_hash(filp, pos);
417 p->curr_minor_hash = pos2min_hash(filp, pos);
418 return p;
419 }
420
421 void ext4_htree_free_dir_info(struct dir_private_info *p)
422 {
423 free_rb_tree_fname(&p->root);
424 kfree(p);
425 }
426
427 /*
428 * Given a directory entry, enter it into the fname rb tree.
429 *
430 * When filename encryption is enabled, the dirent will hold the
431 * encrypted filename, while the htree will hold decrypted filename.
432 * The decrypted filename is passed in via ent_name. parameter.
433 */
434 int ext4_htree_store_dirent(struct file *dir_file, __u32 hash,
435 __u32 minor_hash,
436 struct ext4_dir_entry_2 *dirent,
437 struct fscrypt_str *ent_name)
438 {
439 struct rb_node **p, *parent = NULL;
440 struct fname *fname, *new_fn;
441 struct dir_private_info *info;
442 int len;
443
444 info = dir_file->private_data;
445 p = &info->root.rb_node;
446
447 /* Create and allocate the fname structure */
448 len = sizeof(struct fname) + ent_name->len + 1;
449 new_fn = kzalloc(len, GFP_KERNEL);
450 if (!new_fn)
451 return -ENOMEM;
452 new_fn->hash = hash;
453 new_fn->minor_hash = minor_hash;
454 new_fn->inode = le32_to_cpu(dirent->inode);
455 new_fn->name_len = ent_name->len;
456 new_fn->file_type = dirent->file_type;
457 memcpy(new_fn->name, ent_name->name, ent_name->len);
458 new_fn->name[ent_name->len] = 0;
459
460 while (*p) {
461 parent = *p;
462 fname = rb_entry(parent, struct fname, rb_hash);
463
464 /*
465 * If the hash and minor hash match up, then we put
466 * them on a linked list. This rarely happens...
467 */
468 if ((new_fn->hash == fname->hash) &&
469 (new_fn->minor_hash == fname->minor_hash)) {
470 new_fn->next = fname->next;
471 fname->next = new_fn;
472 return 0;
473 }
474
475 if (new_fn->hash < fname->hash)
476 p = &(*p)->rb_left;
477 else if (new_fn->hash > fname->hash)
478 p = &(*p)->rb_right;
479 else if (new_fn->minor_hash < fname->minor_hash)
480 p = &(*p)->rb_left;
481 else /* if (new_fn->minor_hash > fname->minor_hash) */
482 p = &(*p)->rb_right;
483 }
484
485 rb_link_node(&new_fn->rb_hash, parent, p);
486 rb_insert_color(&new_fn->rb_hash, &info->root);
487 return 0;
488 }
489
490
491
492 /*
493 * This is a helper function for ext4_dx_readdir. It calls filldir
494 * for all entres on the fname linked list. (Normally there is only
495 * one entry on the linked list, unless there are 62 bit hash collisions.)
496 */
497 static int call_filldir(struct file *file, struct dir_context *ctx,
498 struct fname *fname)
499 {
500 struct dir_private_info *info = file->private_data;
501 struct inode *inode = file_inode(file);
502 struct super_block *sb = inode->i_sb;
503
504 if (!fname) {
505 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: comm %s: "
506 "called with null fname?!?", __func__, __LINE__,
507 inode->i_ino, current->comm);
508 return 0;
509 }
510 ctx->pos = hash2pos(file, fname->hash, fname->minor_hash);
511 while (fname) {
512 if (!dir_emit(ctx, fname->name,
513 fname->name_len,
514 fname->inode,
515 get_dtype(sb, fname->file_type))) {
516 info->extra_fname = fname;
517 return 1;
518 }
519 fname = fname->next;
520 }
521 return 0;
522 }
523
524 static int ext4_dx_readdir(struct file *file, struct dir_context *ctx)
525 {
526 struct dir_private_info *info = file->private_data;
527 struct inode *inode = file_inode(file);
528 struct fname *fname;
529 int ret;
530
531 if (!info) {
532 info = ext4_htree_create_dir_info(file, ctx->pos);
533 if (!info)
534 return -ENOMEM;
535 file->private_data = info;
536 }
537
538 if (ctx->pos == ext4_get_htree_eof(file))
539 return 0; /* EOF */
540
541 /* Some one has messed with f_pos; reset the world */
542 if (info->last_pos != ctx->pos) {
543 free_rb_tree_fname(&info->root);
544 info->curr_node = NULL;
545 info->extra_fname = NULL;
546 info->curr_hash = pos2maj_hash(file, ctx->pos);
547 info->curr_minor_hash = pos2min_hash(file, ctx->pos);
548 }
549
550 /*
551 * If there are any leftover names on the hash collision
552 * chain, return them first.
553 */
554 if (info->extra_fname) {
555 if (call_filldir(file, ctx, info->extra_fname))
556 goto finished;
557 info->extra_fname = NULL;
558 goto next_node;
559 } else if (!info->curr_node)
560 info->curr_node = rb_first(&info->root);
561
562 while (1) {
563 /*
564 * Fill the rbtree if we have no more entries,
565 * or the inode has changed since we last read in the
566 * cached entries.
567 */
568 if ((!info->curr_node) ||
569 (file->f_version != inode->i_version)) {
570 info->curr_node = NULL;
571 free_rb_tree_fname(&info->root);
572 file->f_version = inode->i_version;
573 ret = ext4_htree_fill_tree(file, info->curr_hash,
574 info->curr_minor_hash,
575 &info->next_hash);
576 if (ret < 0)
577 return ret;
578 if (ret == 0) {
579 ctx->pos = ext4_get_htree_eof(file);
580 break;
581 }
582 info->curr_node = rb_first(&info->root);
583 }
584
585 fname = rb_entry(info->curr_node, struct fname, rb_hash);
586 info->curr_hash = fname->hash;
587 info->curr_minor_hash = fname->minor_hash;
588 if (call_filldir(file, ctx, fname))
589 break;
590 next_node:
591 info->curr_node = rb_next(info->curr_node);
592 if (info->curr_node) {
593 fname = rb_entry(info->curr_node, struct fname,
594 rb_hash);
595 info->curr_hash = fname->hash;
596 info->curr_minor_hash = fname->minor_hash;
597 } else {
598 if (info->next_hash == ~0) {
599 ctx->pos = ext4_get_htree_eof(file);
600 break;
601 }
602 info->curr_hash = info->next_hash;
603 info->curr_minor_hash = 0;
604 }
605 }
606 finished:
607 info->last_pos = ctx->pos;
608 return 0;
609 }
610
611 static int ext4_dir_open(struct inode * inode, struct file * filp)
612 {
613 if (ext4_encrypted_inode(inode))
614 return fscrypt_get_encryption_info(inode) ? -EACCES : 0;
615 return 0;
616 }
617
618 static int ext4_release_dir(struct inode *inode, struct file *filp)
619 {
620 if (filp->private_data)
621 ext4_htree_free_dir_info(filp->private_data);
622
623 return 0;
624 }
625
626 int ext4_check_all_de(struct inode *dir, struct buffer_head *bh, void *buf,
627 int buf_size)
628 {
629 struct ext4_dir_entry_2 *de;
630 int nlen, rlen;
631 unsigned int offset = 0;
632 char *top;
633
634 de = (struct ext4_dir_entry_2 *)buf;
635 top = buf + buf_size;
636 while ((char *) de < top) {
637 if (ext4_check_dir_entry(dir, NULL, de, bh,
638 buf, buf_size, offset))
639 return -EFSCORRUPTED;
640 nlen = EXT4_DIR_REC_LEN(de->name_len);
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 };
This page took 0.046703 seconds and 5 git commands to generate.