Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * linux/fs/ext4/dir.c
4 *
5 * Copyright (C) 1992, 1993, 1994, 1995
6 * Remy Card (card@masi.ibp.fr)
7 * Laboratoire MASI - Institut Blaise Pascal
8 * Universite Pierre et Marie Curie (Paris VI)
9 *
10 * from
11 *
12 * linux/fs/minix/dir.c
13 *
14 * Copyright (C) 1991, 1992 Linus Torvalds
15 *
16 * ext4 directory handling functions
17 *
18 * Big-endian to little-endian byte-swapping/bitmaps by
19 * David S. Miller (davem@caip.rutgers.edu), 1995
20 *
21 * Hash Tree Directory indexing (c) 2001 Daniel Phillips
22 *
23 */
24
25#include <linux/fs.h>
26#include <linux/buffer_head.h>
27#include <linux/slab.h>
28#include <linux/iversion.h>
29#include <linux/unicode.h>
30#include "ext4.h"
31#include "xattr.h"
32
33static int ext4_dx_readdir(struct file *, struct dir_context *);
34
35/**
36 * is_dx_dir() - check if a directory is using htree indexing
37 * @inode: directory inode
38 *
39 * Check if the given dir-inode refers to an htree-indexed directory
40 * (or a directory which could potentially get converted to use htree
41 * indexing).
42 *
43 * Return 1 if it is a dx dir, 0 if not
44 */
45static int is_dx_dir(struct inode *inode)
46{
47 struct super_block *sb = inode->i_sb;
48
49 if (ext4_has_feature_dir_index(inode->i_sb) &&
50 ((ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) ||
51 ((inode->i_size >> sb->s_blocksize_bits) == 1) ||
52 ext4_has_inline_data(inode)))
53 return 1;
54
55 return 0;
56}
57
58/*
59 * Return 0 if the directory entry is OK, and 1 if there is a problem
60 *
61 * Note: this is the opposite of what ext2 and ext3 historically returned...
62 *
63 * bh passed here can be an inode block or a dir data block, depending
64 * on the inode inline data flag.
65 */
66int __ext4_check_dir_entry(const char *function, unsigned int line,
67 struct inode *dir, struct file *filp,
68 struct ext4_dir_entry_2 *de,
69 struct buffer_head *bh, char *buf, int size,
70 unsigned int offset)
71{
72 const char *error_msg = NULL;
73 const int rlen = ext4_rec_len_from_disk(de->rec_len,
74 dir->i_sb->s_blocksize);
75 const int next_offset = ((char *) de - buf) + rlen;
76
77 if (unlikely(rlen < EXT4_DIR_REC_LEN(1)))
78 error_msg = "rec_len is smaller than minimal";
79 else if (unlikely(rlen % 4 != 0))
80 error_msg = "rec_len % 4 != 0";
81 else if (unlikely(rlen < EXT4_DIR_REC_LEN(de->name_len)))
82 error_msg = "rec_len is too small for name_len";
83 else if (unlikely(next_offset > size))
84 error_msg = "directory entry overrun";
85 else if (unlikely(next_offset > size - EXT4_DIR_REC_LEN(1) &&
86 next_offset != size))
87 error_msg = "directory entry too close to block end";
88 else if (unlikely(le32_to_cpu(de->inode) >
89 le32_to_cpu(EXT4_SB(dir->i_sb)->s_es->s_inodes_count)))
90 error_msg = "inode out of bounds";
91 else
92 return 0;
93
94 if (filp)
95 ext4_error_file(filp, function, line, bh->b_blocknr,
96 "bad entry in directory: %s - offset=%u, "
97 "inode=%u, rec_len=%d, name_len=%d, size=%d",
98 error_msg, offset, le32_to_cpu(de->inode),
99 rlen, de->name_len, size);
100 else
101 ext4_error_inode(dir, function, line, bh->b_blocknr,
102 "bad entry in directory: %s - offset=%u, "
103 "inode=%u, rec_len=%d, name_len=%d, size=%d",
104 error_msg, offset, le32_to_cpu(de->inode),
105 rlen, de->name_len, size);
106
107 return 1;
108}
109
110static int ext4_readdir(struct file *file, struct dir_context *ctx)
111{
112 unsigned int offset;
113 int i;
114 struct ext4_dir_entry_2 *de;
115 int err;
116 struct inode *inode = file_inode(file);
117 struct super_block *sb = inode->i_sb;
118 struct buffer_head *bh = NULL;
119 struct fscrypt_str fstr = FSTR_INIT(NULL, 0);
120
121 if (IS_ENCRYPTED(inode)) {
122 err = fscrypt_get_encryption_info(inode);
123 if (err)
124 return err;
125 }
126
127 if (is_dx_dir(inode)) {
128 err = ext4_dx_readdir(file, ctx);
129 if (err != ERR_BAD_DX_DIR) {
130 return err;
131 }
132 /* Can we just clear INDEX flag to ignore htree information? */
133 if (!ext4_has_metadata_csum(sb)) {
134 /*
135 * We don't set the inode dirty flag since it's not
136 * critical that it gets flushed back to the disk.
137 */
138 ext4_clear_inode_flag(inode, EXT4_INODE_INDEX);
139 }
140 }
141
142 if (ext4_has_inline_data(inode)) {
143 int has_inline_data = 1;
144 err = ext4_read_inline_dir(file, ctx,
145 &has_inline_data);
146 if (has_inline_data)
147 return err;
148 }
149
150 if (IS_ENCRYPTED(inode)) {
151 err = fscrypt_fname_alloc_buffer(inode, EXT4_NAME_LEN, &fstr);
152 if (err < 0)
153 return err;
154 }
155
156 while (ctx->pos < inode->i_size) {
157 struct ext4_map_blocks map;
158
159 if (fatal_signal_pending(current)) {
160 err = -ERESTARTSYS;
161 goto errout;
162 }
163 cond_resched();
164 offset = ctx->pos & (sb->s_blocksize - 1);
165 map.m_lblk = ctx->pos >> EXT4_BLOCK_SIZE_BITS(sb);
166 map.m_len = 1;
167 err = ext4_map_blocks(NULL, inode, &map, 0);
168 if (err == 0) {
169 /* m_len should never be zero but let's avoid
170 * an infinite loop if it somehow is */
171 if (map.m_len == 0)
172 map.m_len = 1;
173 ctx->pos += map.m_len * sb->s_blocksize;
174 continue;
175 }
176 if (err > 0) {
177 pgoff_t index = map.m_pblk >>
178 (PAGE_SHIFT - inode->i_blkbits);
179 if (!ra_has_index(&file->f_ra, index))
180 page_cache_sync_readahead(
181 sb->s_bdev->bd_inode->i_mapping,
182 &file->f_ra, file,
183 index, 1);
184 file->f_ra.prev_pos = (loff_t)index << PAGE_SHIFT;
185 bh = ext4_bread(NULL, inode, map.m_lblk, 0);
186 if (IS_ERR(bh)) {
187 err = PTR_ERR(bh);
188 bh = NULL;
189 goto errout;
190 }
191 }
192
193 if (!bh) {
194 /* corrupt size? Maybe no more blocks to read */
195 if (ctx->pos > inode->i_blocks << 9)
196 break;
197 ctx->pos += sb->s_blocksize - offset;
198 continue;
199 }
200
201 /* Check the checksum */
202 if (!buffer_verified(bh) &&
203 !ext4_dirblock_csum_verify(inode, bh)) {
204 EXT4_ERROR_FILE(file, 0, "directory fails checksum "
205 "at offset %llu",
206 (unsigned long long)ctx->pos);
207 ctx->pos += sb->s_blocksize - offset;
208 brelse(bh);
209 bh = NULL;
210 continue;
211 }
212 set_buffer_verified(bh);
213
214 /* If the dir block has changed since the last call to
215 * readdir(2), then we might be pointing to an invalid
216 * dirent right now. Scan from the start of the block
217 * to make sure. */
218 if (!inode_eq_iversion(inode, file->f_version)) {
219 for (i = 0; i < sb->s_blocksize && i < offset; ) {
220 de = (struct ext4_dir_entry_2 *)
221 (bh->b_data + i);
222 /* It's too expensive to do a full
223 * dirent test each time round this
224 * loop, but we do have to test at
225 * least that it is non-zero. A
226 * failure will be detected in the
227 * dirent test below. */
228 if (ext4_rec_len_from_disk(de->rec_len,
229 sb->s_blocksize) < EXT4_DIR_REC_LEN(1))
230 break;
231 i += ext4_rec_len_from_disk(de->rec_len,
232 sb->s_blocksize);
233 }
234 offset = i;
235 ctx->pos = (ctx->pos & ~(sb->s_blocksize - 1))
236 | offset;
237 file->f_version = inode_query_iversion(inode);
238 }
239
240 while (ctx->pos < inode->i_size
241 && offset < sb->s_blocksize) {
242 de = (struct ext4_dir_entry_2 *) (bh->b_data + offset);
243 if (ext4_check_dir_entry(inode, file, de, bh,
244 bh->b_data, bh->b_size,
245 offset)) {
246 /*
247 * On error, skip to the next block
248 */
249 ctx->pos = (ctx->pos |
250 (sb->s_blocksize - 1)) + 1;
251 break;
252 }
253 offset += ext4_rec_len_from_disk(de->rec_len,
254 sb->s_blocksize);
255 if (le32_to_cpu(de->inode)) {
256 if (!IS_ENCRYPTED(inode)) {
257 if (!dir_emit(ctx, de->name,
258 de->name_len,
259 le32_to_cpu(de->inode),
260 get_dtype(sb, de->file_type)))
261 goto done;
262 } else {
263 int save_len = fstr.len;
264 struct fscrypt_str de_name =
265 FSTR_INIT(de->name,
266 de->name_len);
267
268 /* Directory is encrypted */
269 err = fscrypt_fname_disk_to_usr(inode,
270 0, 0, &de_name, &fstr);
271 de_name = fstr;
272 fstr.len = save_len;
273 if (err)
274 goto errout;
275 if (!dir_emit(ctx,
276 de_name.name, de_name.len,
277 le32_to_cpu(de->inode),
278 get_dtype(sb, de->file_type)))
279 goto done;
280 }
281 }
282 ctx->pos += ext4_rec_len_from_disk(de->rec_len,
283 sb->s_blocksize);
284 }
285 if ((ctx->pos < inode->i_size) && !dir_relax_shared(inode))
286 goto done;
287 brelse(bh);
288 bh = NULL;
289 offset = 0;
290 }
291done:
292 err = 0;
293errout:
294 fscrypt_fname_free_buffer(&fstr);
295 brelse(bh);
296 return err;
297}
298
299static inline int is_32bit_api(void)
300{
301#ifdef CONFIG_COMPAT
302 return in_compat_syscall();
303#else
304 return (BITS_PER_LONG == 32);
305#endif
306}
307
308/*
309 * These functions convert from the major/minor hash to an f_pos
310 * value for dx directories
311 *
312 * Upper layer (for example NFS) should specify FMODE_32BITHASH or
313 * FMODE_64BITHASH explicitly. On the other hand, we allow ext4 to be mounted
314 * directly on both 32-bit and 64-bit nodes, under such case, neither
315 * FMODE_32BITHASH nor FMODE_64BITHASH is specified.
316 */
317static inline loff_t hash2pos(struct file *filp, __u32 major, __u32 minor)
318{
319 if ((filp->f_mode & FMODE_32BITHASH) ||
320 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
321 return major >> 1;
322 else
323 return ((__u64)(major >> 1) << 32) | (__u64)minor;
324}
325
326static inline __u32 pos2maj_hash(struct file *filp, loff_t pos)
327{
328 if ((filp->f_mode & FMODE_32BITHASH) ||
329 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
330 return (pos << 1) & 0xffffffff;
331 else
332 return ((pos >> 32) << 1) & 0xffffffff;
333}
334
335static inline __u32 pos2min_hash(struct file *filp, loff_t pos)
336{
337 if ((filp->f_mode & FMODE_32BITHASH) ||
338 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
339 return 0;
340 else
341 return pos & 0xffffffff;
342}
343
344/*
345 * Return 32- or 64-bit end-of-file for dx directories
346 */
347static inline loff_t ext4_get_htree_eof(struct file *filp)
348{
349 if ((filp->f_mode & FMODE_32BITHASH) ||
350 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
351 return EXT4_HTREE_EOF_32BIT;
352 else
353 return EXT4_HTREE_EOF_64BIT;
354}
355
356
357/*
358 * ext4_dir_llseek() calls generic_file_llseek_size to handle htree
359 * directories, where the "offset" is in terms of the filename hash
360 * value instead of the byte offset.
361 *
362 * Because we may return a 64-bit hash that is well beyond offset limits,
363 * we need to pass the max hash as the maximum allowable offset in
364 * the htree directory case.
365 *
366 * For non-htree, ext4_llseek already chooses the proper max offset.
367 */
368static loff_t ext4_dir_llseek(struct file *file, loff_t offset, int whence)
369{
370 struct inode *inode = file->f_mapping->host;
371 int dx_dir = is_dx_dir(inode);
372 loff_t ret, htree_max = ext4_get_htree_eof(file);
373
374 if (likely(dx_dir))
375 ret = generic_file_llseek_size(file, offset, whence,
376 htree_max, htree_max);
377 else
378 ret = ext4_llseek(file, offset, whence);
379 file->f_version = inode_peek_iversion(inode) - 1;
380 return ret;
381}
382
383/*
384 * This structure holds the nodes of the red-black tree used to store
385 * the directory entry in hash order.
386 */
387struct fname {
388 __u32 hash;
389 __u32 minor_hash;
390 struct rb_node rb_hash;
391 struct fname *next;
392 __u32 inode;
393 __u8 name_len;
394 __u8 file_type;
395 char name[];
396};
397
398/*
399 * This functoin implements a non-recursive way of freeing all of the
400 * nodes in the red-black tree.
401 */
402static void free_rb_tree_fname(struct rb_root *root)
403{
404 struct fname *fname, *next;
405
406 rbtree_postorder_for_each_entry_safe(fname, next, root, rb_hash)
407 while (fname) {
408 struct fname *old = fname;
409 fname = fname->next;
410 kfree(old);
411 }
412
413 *root = RB_ROOT;
414}
415
416
417static struct dir_private_info *ext4_htree_create_dir_info(struct file *filp,
418 loff_t pos)
419{
420 struct dir_private_info *p;
421
422 p = kzalloc(sizeof(*p), GFP_KERNEL);
423 if (!p)
424 return NULL;
425 p->curr_hash = pos2maj_hash(filp, pos);
426 p->curr_minor_hash = pos2min_hash(filp, pos);
427 return p;
428}
429
430void ext4_htree_free_dir_info(struct dir_private_info *p)
431{
432 free_rb_tree_fname(&p->root);
433 kfree(p);
434}
435
436/*
437 * Given a directory entry, enter it into the fname rb tree.
438 *
439 * When filename encryption is enabled, the dirent will hold the
440 * encrypted filename, while the htree will hold decrypted filename.
441 * The decrypted filename is passed in via ent_name. parameter.
442 */
443int ext4_htree_store_dirent(struct file *dir_file, __u32 hash,
444 __u32 minor_hash,
445 struct ext4_dir_entry_2 *dirent,
446 struct fscrypt_str *ent_name)
447{
448 struct rb_node **p, *parent = NULL;
449 struct fname *fname, *new_fn;
450 struct dir_private_info *info;
451 int len;
452
453 info = dir_file->private_data;
454 p = &info->root.rb_node;
455
456 /* Create and allocate the fname structure */
457 len = sizeof(struct fname) + ent_name->len + 1;
458 new_fn = kzalloc(len, GFP_KERNEL);
459 if (!new_fn)
460 return -ENOMEM;
461 new_fn->hash = hash;
462 new_fn->minor_hash = minor_hash;
463 new_fn->inode = le32_to_cpu(dirent->inode);
464 new_fn->name_len = ent_name->len;
465 new_fn->file_type = dirent->file_type;
466 memcpy(new_fn->name, ent_name->name, ent_name->len);
467
468 while (*p) {
469 parent = *p;
470 fname = rb_entry(parent, struct fname, rb_hash);
471
472 /*
473 * If the hash and minor hash match up, then we put
474 * them on a linked list. This rarely happens...
475 */
476 if ((new_fn->hash == fname->hash) &&
477 (new_fn->minor_hash == fname->minor_hash)) {
478 new_fn->next = fname->next;
479 fname->next = new_fn;
480 return 0;
481 }
482
483 if (new_fn->hash < fname->hash)
484 p = &(*p)->rb_left;
485 else if (new_fn->hash > fname->hash)
486 p = &(*p)->rb_right;
487 else if (new_fn->minor_hash < fname->minor_hash)
488 p = &(*p)->rb_left;
489 else /* if (new_fn->minor_hash > fname->minor_hash) */
490 p = &(*p)->rb_right;
491 }
492
493 rb_link_node(&new_fn->rb_hash, parent, p);
494 rb_insert_color(&new_fn->rb_hash, &info->root);
495 return 0;
496}
497
498
499
500/*
501 * This is a helper function for ext4_dx_readdir. It calls filldir
502 * for all entres on the fname linked list. (Normally there is only
503 * one entry on the linked list, unless there are 62 bit hash collisions.)
504 */
505static int call_filldir(struct file *file, struct dir_context *ctx,
506 struct fname *fname)
507{
508 struct dir_private_info *info = file->private_data;
509 struct inode *inode = file_inode(file);
510 struct super_block *sb = inode->i_sb;
511
512 if (!fname) {
513 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: comm %s: "
514 "called with null fname?!?", __func__, __LINE__,
515 inode->i_ino, current->comm);
516 return 0;
517 }
518 ctx->pos = hash2pos(file, fname->hash, fname->minor_hash);
519 while (fname) {
520 if (!dir_emit(ctx, fname->name,
521 fname->name_len,
522 fname->inode,
523 get_dtype(sb, fname->file_type))) {
524 info->extra_fname = fname;
525 return 1;
526 }
527 fname = fname->next;
528 }
529 return 0;
530}
531
532static int ext4_dx_readdir(struct file *file, struct dir_context *ctx)
533{
534 struct dir_private_info *info = file->private_data;
535 struct inode *inode = file_inode(file);
536 struct fname *fname;
537 int ret;
538
539 if (!info) {
540 info = ext4_htree_create_dir_info(file, ctx->pos);
541 if (!info)
542 return -ENOMEM;
543 file->private_data = info;
544 }
545
546 if (ctx->pos == ext4_get_htree_eof(file))
547 return 0; /* EOF */
548
549 /* Some one has messed with f_pos; reset the world */
550 if (info->last_pos != ctx->pos) {
551 free_rb_tree_fname(&info->root);
552 info->curr_node = NULL;
553 info->extra_fname = NULL;
554 info->curr_hash = pos2maj_hash(file, ctx->pos);
555 info->curr_minor_hash = pos2min_hash(file, ctx->pos);
556 }
557
558 /*
559 * If there are any leftover names on the hash collision
560 * chain, return them first.
561 */
562 if (info->extra_fname) {
563 if (call_filldir(file, ctx, info->extra_fname))
564 goto finished;
565 info->extra_fname = NULL;
566 goto next_node;
567 } else if (!info->curr_node)
568 info->curr_node = rb_first(&info->root);
569
570 while (1) {
571 /*
572 * Fill the rbtree if we have no more entries,
573 * or the inode has changed since we last read in the
574 * cached entries.
575 */
576 if ((!info->curr_node) ||
577 !inode_eq_iversion(inode, file->f_version)) {
578 info->curr_node = NULL;
579 free_rb_tree_fname(&info->root);
580 file->f_version = inode_query_iversion(inode);
581 ret = ext4_htree_fill_tree(file, info->curr_hash,
582 info->curr_minor_hash,
583 &info->next_hash);
584 if (ret < 0)
585 return ret;
586 if (ret == 0) {
587 ctx->pos = ext4_get_htree_eof(file);
588 break;
589 }
590 info->curr_node = rb_first(&info->root);
591 }
592
593 fname = rb_entry(info->curr_node, struct fname, rb_hash);
594 info->curr_hash = fname->hash;
595 info->curr_minor_hash = fname->minor_hash;
596 if (call_filldir(file, ctx, fname))
597 break;
598 next_node:
599 info->curr_node = rb_next(info->curr_node);
600 if (info->curr_node) {
601 fname = rb_entry(info->curr_node, struct fname,
602 rb_hash);
603 info->curr_hash = fname->hash;
604 info->curr_minor_hash = fname->minor_hash;
605 } else {
606 if (info->next_hash == ~0) {
607 ctx->pos = ext4_get_htree_eof(file);
608 break;
609 }
610 info->curr_hash = info->next_hash;
611 info->curr_minor_hash = 0;
612 }
613 }
614finished:
615 info->last_pos = ctx->pos;
616 return 0;
617}
618
619static int ext4_dir_open(struct inode * inode, struct file * filp)
620{
621 if (IS_ENCRYPTED(inode))
622 return fscrypt_get_encryption_info(inode) ? -EACCES : 0;
623 return 0;
624}
625
626static int ext4_release_dir(struct inode *inode, struct file *filp)
627{
628 if (filp->private_data)
629 ext4_htree_free_dir_info(filp->private_data);
630
631 return 0;
632}
633
634int ext4_check_all_de(struct inode *dir, struct buffer_head *bh, void *buf,
635 int buf_size)
636{
637 struct ext4_dir_entry_2 *de;
638 int rlen;
639 unsigned int offset = 0;
640 char *top;
641
642 de = (struct ext4_dir_entry_2 *)buf;
643 top = buf + buf_size;
644 while ((char *) de < top) {
645 if (ext4_check_dir_entry(dir, NULL, de, bh,
646 buf, buf_size, offset))
647 return -EFSCORRUPTED;
648 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
649 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
650 offset += rlen;
651 }
652 if ((char *) de > top)
653 return -EFSCORRUPTED;
654
655 return 0;
656}
657
658const struct file_operations ext4_dir_operations = {
659 .llseek = ext4_dir_llseek,
660 .read = generic_read_dir,
661 .iterate_shared = ext4_readdir,
662 .unlocked_ioctl = ext4_ioctl,
663#ifdef CONFIG_COMPAT
664 .compat_ioctl = ext4_compat_ioctl,
665#endif
666 .fsync = ext4_sync_file,
667 .open = ext4_dir_open,
668 .release = ext4_release_dir,
669};
670
671#ifdef CONFIG_UNICODE
672static int ext4_d_compare(const struct dentry *dentry, unsigned int len,
673 const char *str, const struct qstr *name)
674{
675 struct qstr qstr = {.name = str, .len = len };
676 const struct dentry *parent = READ_ONCE(dentry->d_parent);
677 const struct inode *inode = READ_ONCE(parent->d_inode);
678 char strbuf[DNAME_INLINE_LEN];
679
680 if (!inode || !IS_CASEFOLDED(inode) ||
681 !EXT4_SB(inode->i_sb)->s_encoding) {
682 if (len != name->len)
683 return -1;
684 return memcmp(str, name->name, len);
685 }
686
687 /*
688 * If the dentry name is stored in-line, then it may be concurrently
689 * modified by a rename. If this happens, the VFS will eventually retry
690 * the lookup, so it doesn't matter what ->d_compare() returns.
691 * However, it's unsafe to call utf8_strncasecmp() with an unstable
692 * string. Therefore, we have to copy the name into a temporary buffer.
693 */
694 if (len <= DNAME_INLINE_LEN - 1) {
695 memcpy(strbuf, str, len);
696 strbuf[len] = 0;
697 qstr.name = strbuf;
698 /* prevent compiler from optimizing out the temporary buffer */
699 barrier();
700 }
701
702 return ext4_ci_compare(inode, name, &qstr, false);
703}
704
705static int ext4_d_hash(const struct dentry *dentry, struct qstr *str)
706{
707 const struct ext4_sb_info *sbi = EXT4_SB(dentry->d_sb);
708 const struct unicode_map *um = sbi->s_encoding;
709 const struct inode *inode = READ_ONCE(dentry->d_inode);
710 unsigned char *norm;
711 int len, ret = 0;
712
713 if (!inode || !IS_CASEFOLDED(inode) || !um)
714 return 0;
715
716 norm = kmalloc(PATH_MAX, GFP_ATOMIC);
717 if (!norm)
718 return -ENOMEM;
719
720 len = utf8_casefold(um, str, norm, PATH_MAX);
721 if (len < 0) {
722 if (ext4_has_strict_mode(sbi))
723 ret = -EINVAL;
724 goto out;
725 }
726 str->hash = full_name_hash(dentry, norm, len);
727out:
728 kfree(norm);
729 return ret;
730}
731
732const struct dentry_operations ext4_dentry_ops = {
733 .d_hash = ext4_d_hash,
734 .d_compare = ext4_d_compare,
735};
736#endif
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/jbd2.h>
26#include <linux/buffer_head.h>
27#include <linux/slab.h>
28#include <linux/rbtree.h>
29#include "ext4.h"
30
31static unsigned char ext4_filetype_table[] = {
32 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
33};
34
35static int ext4_dx_readdir(struct file *filp,
36 void *dirent, filldir_t filldir);
37
38static unsigned char get_dtype(struct super_block *sb, int filetype)
39{
40 if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE) ||
41 (filetype >= EXT4_FT_MAX))
42 return DT_UNKNOWN;
43
44 return (ext4_filetype_table[filetype]);
45}
46
47/**
48 * Check if the given dir-inode refers to an htree-indexed directory
49 * (or a directory which chould potentially get coverted to use htree
50 * indexing).
51 *
52 * Return 1 if it is a dx dir, 0 if not
53 */
54static int is_dx_dir(struct inode *inode)
55{
56 struct super_block *sb = inode->i_sb;
57
58 if (EXT4_HAS_COMPAT_FEATURE(inode->i_sb,
59 EXT4_FEATURE_COMPAT_DIR_INDEX) &&
60 ((ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) ||
61 ((inode->i_size >> sb->s_blocksize_bits) == 1)))
62 return 1;
63
64 return 0;
65}
66
67/*
68 * Return 0 if the directory entry is OK, and 1 if there is a problem
69 *
70 * Note: this is the opposite of what ext2 and ext3 historically returned...
71 */
72int __ext4_check_dir_entry(const char *function, unsigned int line,
73 struct inode *dir, struct file *filp,
74 struct ext4_dir_entry_2 *de,
75 struct buffer_head *bh,
76 unsigned int offset)
77{
78 const char *error_msg = NULL;
79 const int rlen = ext4_rec_len_from_disk(de->rec_len,
80 dir->i_sb->s_blocksize);
81
82 if (unlikely(rlen < EXT4_DIR_REC_LEN(1)))
83 error_msg = "rec_len is smaller than minimal";
84 else if (unlikely(rlen % 4 != 0))
85 error_msg = "rec_len % 4 != 0";
86 else if (unlikely(rlen < EXT4_DIR_REC_LEN(de->name_len)))
87 error_msg = "rec_len is too small for name_len";
88 else if (unlikely(((char *) de - bh->b_data) + rlen >
89 dir->i_sb->s_blocksize))
90 error_msg = "directory entry across blocks";
91 else if (unlikely(le32_to_cpu(de->inode) >
92 le32_to_cpu(EXT4_SB(dir->i_sb)->s_es->s_inodes_count)))
93 error_msg = "inode out of bounds";
94 else
95 return 0;
96
97 if (filp)
98 ext4_error_file(filp, function, line, bh->b_blocknr,
99 "bad entry in directory: %s - offset=%u(%u), "
100 "inode=%u, rec_len=%d, name_len=%d",
101 error_msg, (unsigned) (offset % bh->b_size),
102 offset, le32_to_cpu(de->inode),
103 rlen, de->name_len);
104 else
105 ext4_error_inode(dir, function, line, bh->b_blocknr,
106 "bad entry in directory: %s - offset=%u(%u), "
107 "inode=%u, rec_len=%d, name_len=%d",
108 error_msg, (unsigned) (offset % bh->b_size),
109 offset, le32_to_cpu(de->inode),
110 rlen, de->name_len);
111
112 return 1;
113}
114
115static int ext4_readdir(struct file *filp,
116 void *dirent, filldir_t filldir)
117{
118 int error = 0;
119 unsigned int offset;
120 int i, stored;
121 struct ext4_dir_entry_2 *de;
122 int err;
123 struct inode *inode = filp->f_path.dentry->d_inode;
124 struct super_block *sb = inode->i_sb;
125 int ret = 0;
126 int dir_has_error = 0;
127
128 if (is_dx_dir(inode)) {
129 err = ext4_dx_readdir(filp, dirent, filldir);
130 if (err != ERR_BAD_DX_DIR) {
131 ret = err;
132 goto out;
133 }
134 /*
135 * We don't set the inode dirty flag since it's not
136 * critical that it get flushed back to the disk.
137 */
138 ext4_clear_inode_flag(filp->f_path.dentry->d_inode,
139 EXT4_INODE_INDEX);
140 }
141 stored = 0;
142 offset = filp->f_pos & (sb->s_blocksize - 1);
143
144 while (!error && !stored && filp->f_pos < inode->i_size) {
145 struct ext4_map_blocks map;
146 struct buffer_head *bh = NULL;
147
148 map.m_lblk = filp->f_pos >> EXT4_BLOCK_SIZE_BITS(sb);
149 map.m_len = 1;
150 err = ext4_map_blocks(NULL, inode, &map, 0);
151 if (err > 0) {
152 pgoff_t index = map.m_pblk >>
153 (PAGE_CACHE_SHIFT - inode->i_blkbits);
154 if (!ra_has_index(&filp->f_ra, index))
155 page_cache_sync_readahead(
156 sb->s_bdev->bd_inode->i_mapping,
157 &filp->f_ra, filp,
158 index, 1);
159 filp->f_ra.prev_pos = (loff_t)index << PAGE_CACHE_SHIFT;
160 bh = ext4_bread(NULL, inode, map.m_lblk, 0, &err);
161 }
162
163 /*
164 * We ignore I/O errors on directories so users have a chance
165 * of recovering data when there's a bad sector
166 */
167 if (!bh) {
168 if (!dir_has_error) {
169 EXT4_ERROR_FILE(filp, 0,
170 "directory contains a "
171 "hole at offset %llu",
172 (unsigned long long) filp->f_pos);
173 dir_has_error = 1;
174 }
175 /* corrupt size? Maybe no more blocks to read */
176 if (filp->f_pos > inode->i_blocks << 9)
177 break;
178 filp->f_pos += sb->s_blocksize - offset;
179 continue;
180 }
181
182 /* Check the checksum */
183 if (!buffer_verified(bh) &&
184 !ext4_dirent_csum_verify(inode,
185 (struct ext4_dir_entry *)bh->b_data)) {
186 EXT4_ERROR_FILE(filp, 0, "directory fails checksum "
187 "at offset %llu",
188 (unsigned long long)filp->f_pos);
189 filp->f_pos += sb->s_blocksize - offset;
190 continue;
191 }
192 set_buffer_verified(bh);
193
194revalidate:
195 /* If the dir block has changed since the last call to
196 * readdir(2), then we might be pointing to an invalid
197 * dirent right now. Scan from the start of the block
198 * to make sure. */
199 if (filp->f_version != inode->i_version) {
200 for (i = 0; i < sb->s_blocksize && i < offset; ) {
201 de = (struct ext4_dir_entry_2 *)
202 (bh->b_data + i);
203 /* It's too expensive to do a full
204 * dirent test each time round this
205 * loop, but we do have to test at
206 * least that it is non-zero. A
207 * failure will be detected in the
208 * dirent test below. */
209 if (ext4_rec_len_from_disk(de->rec_len,
210 sb->s_blocksize) < EXT4_DIR_REC_LEN(1))
211 break;
212 i += ext4_rec_len_from_disk(de->rec_len,
213 sb->s_blocksize);
214 }
215 offset = i;
216 filp->f_pos = (filp->f_pos & ~(sb->s_blocksize - 1))
217 | offset;
218 filp->f_version = inode->i_version;
219 }
220
221 while (!error && filp->f_pos < inode->i_size
222 && offset < sb->s_blocksize) {
223 de = (struct ext4_dir_entry_2 *) (bh->b_data + offset);
224 if (ext4_check_dir_entry(inode, filp, de,
225 bh, offset)) {
226 /*
227 * On error, skip the f_pos to the next block
228 */
229 filp->f_pos = (filp->f_pos |
230 (sb->s_blocksize - 1)) + 1;
231 brelse(bh);
232 ret = stored;
233 goto out;
234 }
235 offset += ext4_rec_len_from_disk(de->rec_len,
236 sb->s_blocksize);
237 if (le32_to_cpu(de->inode)) {
238 /* We might block in the next section
239 * if the data destination is
240 * currently swapped out. So, use a
241 * version stamp to detect whether or
242 * not the directory has been modified
243 * during the copy operation.
244 */
245 u64 version = filp->f_version;
246
247 error = filldir(dirent, de->name,
248 de->name_len,
249 filp->f_pos,
250 le32_to_cpu(de->inode),
251 get_dtype(sb, de->file_type));
252 if (error)
253 break;
254 if (version != filp->f_version)
255 goto revalidate;
256 stored++;
257 }
258 filp->f_pos += ext4_rec_len_from_disk(de->rec_len,
259 sb->s_blocksize);
260 }
261 offset = 0;
262 brelse(bh);
263 }
264out:
265 return ret;
266}
267
268static inline int is_32bit_api(void)
269{
270#ifdef CONFIG_COMPAT
271 return is_compat_task();
272#else
273 return (BITS_PER_LONG == 32);
274#endif
275}
276
277/*
278 * These functions convert from the major/minor hash to an f_pos
279 * value for dx directories
280 *
281 * Upper layer (for example NFS) should specify FMODE_32BITHASH or
282 * FMODE_64BITHASH explicitly. On the other hand, we allow ext4 to be mounted
283 * directly on both 32-bit and 64-bit nodes, under such case, neither
284 * FMODE_32BITHASH nor FMODE_64BITHASH is specified.
285 */
286static inline loff_t hash2pos(struct file *filp, __u32 major, __u32 minor)
287{
288 if ((filp->f_mode & FMODE_32BITHASH) ||
289 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
290 return major >> 1;
291 else
292 return ((__u64)(major >> 1) << 32) | (__u64)minor;
293}
294
295static inline __u32 pos2maj_hash(struct file *filp, loff_t pos)
296{
297 if ((filp->f_mode & FMODE_32BITHASH) ||
298 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
299 return (pos << 1) & 0xffffffff;
300 else
301 return ((pos >> 32) << 1) & 0xffffffff;
302}
303
304static inline __u32 pos2min_hash(struct file *filp, loff_t pos)
305{
306 if ((filp->f_mode & FMODE_32BITHASH) ||
307 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
308 return 0;
309 else
310 return pos & 0xffffffff;
311}
312
313/*
314 * Return 32- or 64-bit end-of-file for dx directories
315 */
316static inline loff_t ext4_get_htree_eof(struct file *filp)
317{
318 if ((filp->f_mode & FMODE_32BITHASH) ||
319 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
320 return EXT4_HTREE_EOF_32BIT;
321 else
322 return EXT4_HTREE_EOF_64BIT;
323}
324
325
326/*
327 * ext4_dir_llseek() based on generic_file_llseek() to handle both
328 * non-htree and htree directories, where the "offset" is in terms
329 * of the filename hash value instead of the byte offset.
330 *
331 * NOTE: offsets obtained *before* ext4_set_inode_flag(dir, EXT4_INODE_INDEX)
332 * will be invalid once the directory was converted into a dx directory
333 */
334loff_t ext4_dir_llseek(struct file *file, loff_t offset, int origin)
335{
336 struct inode *inode = file->f_mapping->host;
337 loff_t ret = -EINVAL;
338 int dx_dir = is_dx_dir(inode);
339
340 mutex_lock(&inode->i_mutex);
341
342 /* NOTE: relative offsets with dx directories might not work
343 * as expected, as it is difficult to figure out the
344 * correct offset between dx hashes */
345
346 switch (origin) {
347 case SEEK_END:
348 if (unlikely(offset > 0))
349 goto out_err; /* not supported for directories */
350
351 /* so only negative offsets are left, does that have a
352 * meaning for directories at all? */
353 if (dx_dir)
354 offset += ext4_get_htree_eof(file);
355 else
356 offset += inode->i_size;
357 break;
358 case SEEK_CUR:
359 /*
360 * Here we special-case the lseek(fd, 0, SEEK_CUR)
361 * position-querying operation. Avoid rewriting the "same"
362 * f_pos value back to the file because a concurrent read(),
363 * write() or lseek() might have altered it
364 */
365 if (offset == 0) {
366 offset = file->f_pos;
367 goto out_ok;
368 }
369
370 offset += file->f_pos;
371 break;
372 }
373
374 if (unlikely(offset < 0))
375 goto out_err;
376
377 if (!dx_dir) {
378 if (offset > inode->i_sb->s_maxbytes)
379 goto out_err;
380 } else if (offset > ext4_get_htree_eof(file))
381 goto out_err;
382
383 /* Special lock needed here? */
384 if (offset != file->f_pos) {
385 file->f_pos = offset;
386 file->f_version = 0;
387 }
388
389out_ok:
390 ret = offset;
391out_err:
392 mutex_unlock(&inode->i_mutex);
393
394 return ret;
395}
396
397/*
398 * This structure holds the nodes of the red-black tree used to store
399 * the directory entry in hash order.
400 */
401struct fname {
402 __u32 hash;
403 __u32 minor_hash;
404 struct rb_node rb_hash;
405 struct fname *next;
406 __u32 inode;
407 __u8 name_len;
408 __u8 file_type;
409 char name[0];
410};
411
412/*
413 * This functoin implements a non-recursive way of freeing all of the
414 * nodes in the red-black tree.
415 */
416static void free_rb_tree_fname(struct rb_root *root)
417{
418 struct rb_node *n = root->rb_node;
419 struct rb_node *parent;
420 struct fname *fname;
421
422 while (n) {
423 /* Do the node's children first */
424 if (n->rb_left) {
425 n = n->rb_left;
426 continue;
427 }
428 if (n->rb_right) {
429 n = n->rb_right;
430 continue;
431 }
432 /*
433 * The node has no children; free it, and then zero
434 * out parent's link to it. Finally go to the
435 * beginning of the loop and try to free the parent
436 * node.
437 */
438 parent = rb_parent(n);
439 fname = rb_entry(n, struct fname, rb_hash);
440 while (fname) {
441 struct fname *old = fname;
442 fname = fname->next;
443 kfree(old);
444 }
445 if (!parent)
446 *root = RB_ROOT;
447 else if (parent->rb_left == n)
448 parent->rb_left = NULL;
449 else if (parent->rb_right == n)
450 parent->rb_right = NULL;
451 n = parent;
452 }
453}
454
455
456static struct dir_private_info *ext4_htree_create_dir_info(struct file *filp,
457 loff_t pos)
458{
459 struct dir_private_info *p;
460
461 p = kzalloc(sizeof(struct dir_private_info), GFP_KERNEL);
462 if (!p)
463 return NULL;
464 p->curr_hash = pos2maj_hash(filp, pos);
465 p->curr_minor_hash = pos2min_hash(filp, pos);
466 return p;
467}
468
469void ext4_htree_free_dir_info(struct dir_private_info *p)
470{
471 free_rb_tree_fname(&p->root);
472 kfree(p);
473}
474
475/*
476 * Given a directory entry, enter it into the fname rb tree.
477 */
478int ext4_htree_store_dirent(struct file *dir_file, __u32 hash,
479 __u32 minor_hash,
480 struct ext4_dir_entry_2 *dirent)
481{
482 struct rb_node **p, *parent = NULL;
483 struct fname *fname, *new_fn;
484 struct dir_private_info *info;
485 int len;
486
487 info = dir_file->private_data;
488 p = &info->root.rb_node;
489
490 /* Create and allocate the fname structure */
491 len = sizeof(struct fname) + dirent->name_len + 1;
492 new_fn = kzalloc(len, GFP_KERNEL);
493 if (!new_fn)
494 return -ENOMEM;
495 new_fn->hash = hash;
496 new_fn->minor_hash = minor_hash;
497 new_fn->inode = le32_to_cpu(dirent->inode);
498 new_fn->name_len = dirent->name_len;
499 new_fn->file_type = dirent->file_type;
500 memcpy(new_fn->name, dirent->name, dirent->name_len);
501 new_fn->name[dirent->name_len] = 0;
502
503 while (*p) {
504 parent = *p;
505 fname = rb_entry(parent, struct fname, rb_hash);
506
507 /*
508 * If the hash and minor hash match up, then we put
509 * them on a linked list. This rarely happens...
510 */
511 if ((new_fn->hash == fname->hash) &&
512 (new_fn->minor_hash == fname->minor_hash)) {
513 new_fn->next = fname->next;
514 fname->next = new_fn;
515 return 0;
516 }
517
518 if (new_fn->hash < fname->hash)
519 p = &(*p)->rb_left;
520 else if (new_fn->hash > fname->hash)
521 p = &(*p)->rb_right;
522 else if (new_fn->minor_hash < fname->minor_hash)
523 p = &(*p)->rb_left;
524 else /* if (new_fn->minor_hash > fname->minor_hash) */
525 p = &(*p)->rb_right;
526 }
527
528 rb_link_node(&new_fn->rb_hash, parent, p);
529 rb_insert_color(&new_fn->rb_hash, &info->root);
530 return 0;
531}
532
533
534
535/*
536 * This is a helper function for ext4_dx_readdir. It calls filldir
537 * for all entres on the fname linked list. (Normally there is only
538 * one entry on the linked list, unless there are 62 bit hash collisions.)
539 */
540static int call_filldir(struct file *filp, void *dirent,
541 filldir_t filldir, struct fname *fname)
542{
543 struct dir_private_info *info = filp->private_data;
544 loff_t curr_pos;
545 struct inode *inode = filp->f_path.dentry->d_inode;
546 struct super_block *sb;
547 int error;
548
549 sb = inode->i_sb;
550
551 if (!fname) {
552 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: comm %s: "
553 "called with null fname?!?", __func__, __LINE__,
554 inode->i_ino, current->comm);
555 return 0;
556 }
557 curr_pos = hash2pos(filp, fname->hash, fname->minor_hash);
558 while (fname) {
559 error = filldir(dirent, fname->name,
560 fname->name_len, curr_pos,
561 fname->inode,
562 get_dtype(sb, fname->file_type));
563 if (error) {
564 filp->f_pos = curr_pos;
565 info->extra_fname = fname;
566 return error;
567 }
568 fname = fname->next;
569 }
570 return 0;
571}
572
573static int ext4_dx_readdir(struct file *filp,
574 void *dirent, filldir_t filldir)
575{
576 struct dir_private_info *info = filp->private_data;
577 struct inode *inode = filp->f_path.dentry->d_inode;
578 struct fname *fname;
579 int ret;
580
581 if (!info) {
582 info = ext4_htree_create_dir_info(filp, filp->f_pos);
583 if (!info)
584 return -ENOMEM;
585 filp->private_data = info;
586 }
587
588 if (filp->f_pos == ext4_get_htree_eof(filp))
589 return 0; /* EOF */
590
591 /* Some one has messed with f_pos; reset the world */
592 if (info->last_pos != filp->f_pos) {
593 free_rb_tree_fname(&info->root);
594 info->curr_node = NULL;
595 info->extra_fname = NULL;
596 info->curr_hash = pos2maj_hash(filp, filp->f_pos);
597 info->curr_minor_hash = pos2min_hash(filp, filp->f_pos);
598 }
599
600 /*
601 * If there are any leftover names on the hash collision
602 * chain, return them first.
603 */
604 if (info->extra_fname) {
605 if (call_filldir(filp, dirent, filldir, info->extra_fname))
606 goto finished;
607 info->extra_fname = NULL;
608 goto next_node;
609 } else if (!info->curr_node)
610 info->curr_node = rb_first(&info->root);
611
612 while (1) {
613 /*
614 * Fill the rbtree if we have no more entries,
615 * or the inode has changed since we last read in the
616 * cached entries.
617 */
618 if ((!info->curr_node) ||
619 (filp->f_version != inode->i_version)) {
620 info->curr_node = NULL;
621 free_rb_tree_fname(&info->root);
622 filp->f_version = inode->i_version;
623 ret = ext4_htree_fill_tree(filp, info->curr_hash,
624 info->curr_minor_hash,
625 &info->next_hash);
626 if (ret < 0)
627 return ret;
628 if (ret == 0) {
629 filp->f_pos = ext4_get_htree_eof(filp);
630 break;
631 }
632 info->curr_node = rb_first(&info->root);
633 }
634
635 fname = rb_entry(info->curr_node, struct fname, rb_hash);
636 info->curr_hash = fname->hash;
637 info->curr_minor_hash = fname->minor_hash;
638 if (call_filldir(filp, dirent, filldir, fname))
639 break;
640 next_node:
641 info->curr_node = rb_next(info->curr_node);
642 if (info->curr_node) {
643 fname = rb_entry(info->curr_node, struct fname,
644 rb_hash);
645 info->curr_hash = fname->hash;
646 info->curr_minor_hash = fname->minor_hash;
647 } else {
648 if (info->next_hash == ~0) {
649 filp->f_pos = ext4_get_htree_eof(filp);
650 break;
651 }
652 info->curr_hash = info->next_hash;
653 info->curr_minor_hash = 0;
654 }
655 }
656finished:
657 info->last_pos = filp->f_pos;
658 return 0;
659}
660
661static int ext4_release_dir(struct inode *inode, struct file *filp)
662{
663 if (filp->private_data)
664 ext4_htree_free_dir_info(filp->private_data);
665
666 return 0;
667}
668
669const struct file_operations ext4_dir_operations = {
670 .llseek = ext4_dir_llseek,
671 .read = generic_read_dir,
672 .readdir = ext4_readdir,
673 .unlocked_ioctl = ext4_ioctl,
674#ifdef CONFIG_COMPAT
675 .compat_ioctl = ext4_compat_ioctl,
676#endif
677 .fsync = ext4_sync_file,
678 .release = ext4_release_dir,
679};