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 "ext4.h"
30#include "xattr.h"
31
32static int ext4_dx_readdir(struct file *, struct dir_context *);
33
34/**
35 * Check if the given dir-inode refers to an htree-indexed directory
36 * (or a directory which could potentially get converted to use htree
37 * indexing).
38 *
39 * Return 1 if it is a dx dir, 0 if not
40 */
41static int is_dx_dir(struct inode *inode)
42{
43 struct super_block *sb = inode->i_sb;
44
45 if (ext4_has_feature_dir_index(inode->i_sb) &&
46 ((ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) ||
47 ((inode->i_size >> sb->s_blocksize_bits) == 1) ||
48 ext4_has_inline_data(inode)))
49 return 1;
50
51 return 0;
52}
53
54/*
55 * Return 0 if the directory entry is OK, and 1 if there is a problem
56 *
57 * Note: this is the opposite of what ext2 and ext3 historically returned...
58 *
59 * bh passed here can be an inode block or a dir data block, depending
60 * on the inode inline data flag.
61 */
62int __ext4_check_dir_entry(const char *function, unsigned int line,
63 struct inode *dir, struct file *filp,
64 struct ext4_dir_entry_2 *de,
65 struct buffer_head *bh, char *buf, int size,
66 unsigned int offset)
67{
68 const char *error_msg = NULL;
69 const int rlen = ext4_rec_len_from_disk(de->rec_len,
70 dir->i_sb->s_blocksize);
71
72 if (unlikely(rlen < EXT4_DIR_REC_LEN(1)))
73 error_msg = "rec_len is smaller than minimal";
74 else if (unlikely(rlen % 4 != 0))
75 error_msg = "rec_len % 4 != 0";
76 else if (unlikely(rlen < EXT4_DIR_REC_LEN(de->name_len)))
77 error_msg = "rec_len is too small for name_len";
78 else if (unlikely(((char *) de - buf) + rlen > size))
79 error_msg = "directory entry across range";
80 else if (unlikely(le32_to_cpu(de->inode) >
81 le32_to_cpu(EXT4_SB(dir->i_sb)->s_es->s_inodes_count)))
82 error_msg = "inode out of bounds";
83 else
84 return 0;
85
86 if (filp)
87 ext4_error_file(filp, function, line, bh->b_blocknr,
88 "bad entry in directory: %s - offset=%u(%u), "
89 "inode=%u, rec_len=%d, name_len=%d",
90 error_msg, (unsigned) (offset % size),
91 offset, le32_to_cpu(de->inode),
92 rlen, de->name_len);
93 else
94 ext4_error_inode(dir, function, line, bh->b_blocknr,
95 "bad entry in directory: %s - offset=%u(%u), "
96 "inode=%u, rec_len=%d, name_len=%d",
97 error_msg, (unsigned) (offset % size),
98 offset, le32_to_cpu(de->inode),
99 rlen, de->name_len);
100
101 return 1;
102}
103
104static int ext4_readdir(struct file *file, struct dir_context *ctx)
105{
106 unsigned int offset;
107 int i;
108 struct ext4_dir_entry_2 *de;
109 int err;
110 struct inode *inode = file_inode(file);
111 struct super_block *sb = inode->i_sb;
112 struct buffer_head *bh = NULL;
113 int dir_has_error = 0;
114 struct fscrypt_str fstr = FSTR_INIT(NULL, 0);
115
116 if (ext4_encrypted_inode(inode)) {
117 err = fscrypt_get_encryption_info(inode);
118 if (err && err != -ENOKEY)
119 return err;
120 }
121
122 if (is_dx_dir(inode)) {
123 err = ext4_dx_readdir(file, ctx);
124 if (err != ERR_BAD_DX_DIR) {
125 return err;
126 }
127 /*
128 * We don't set the inode dirty flag since it's not
129 * critical that it get flushed back to the disk.
130 */
131 ext4_clear_inode_flag(file_inode(file),
132 EXT4_INODE_INDEX);
133 }
134
135 if (ext4_has_inline_data(inode)) {
136 int has_inline_data = 1;
137 err = ext4_read_inline_dir(file, ctx,
138 &has_inline_data);
139 if (has_inline_data)
140 return err;
141 }
142
143 if (ext4_encrypted_inode(inode)) {
144 err = fscrypt_fname_alloc_buffer(inode, EXT4_NAME_LEN, &fstr);
145 if (err < 0)
146 return err;
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 if (fatal_signal_pending(current)) {
155 err = -ERESTARTSYS;
156 goto errout;
157 }
158 cond_resched();
159 map.m_lblk = ctx->pos >> EXT4_BLOCK_SIZE_BITS(sb);
160 map.m_len = 1;
161 err = ext4_map_blocks(NULL, inode, &map, 0);
162 if (err > 0) {
163 pgoff_t index = map.m_pblk >>
164 (PAGE_SHIFT - inode->i_blkbits);
165 if (!ra_has_index(&file->f_ra, index))
166 page_cache_sync_readahead(
167 sb->s_bdev->bd_inode->i_mapping,
168 &file->f_ra, file,
169 index, 1);
170 file->f_ra.prev_pos = (loff_t)index << PAGE_SHIFT;
171 bh = ext4_bread(NULL, inode, map.m_lblk, 0);
172 if (IS_ERR(bh)) {
173 err = PTR_ERR(bh);
174 bh = NULL;
175 goto errout;
176 }
177 }
178
179 if (!bh) {
180 if (!dir_has_error) {
181 EXT4_ERROR_FILE(file, 0,
182 "directory contains a "
183 "hole at offset %llu",
184 (unsigned long long) ctx->pos);
185 dir_has_error = 1;
186 }
187 /* corrupt size? Maybe no more blocks to read */
188 if (ctx->pos > inode->i_blocks << 9)
189 break;
190 ctx->pos += sb->s_blocksize - offset;
191 continue;
192 }
193
194 /* Check the checksum */
195 if (!buffer_verified(bh) &&
196 !ext4_dirent_csum_verify(inode,
197 (struct ext4_dir_entry *)bh->b_data)) {
198 EXT4_ERROR_FILE(file, 0, "directory fails checksum "
199 "at offset %llu",
200 (unsigned long long)ctx->pos);
201 ctx->pos += sb->s_blocksize - offset;
202 brelse(bh);
203 bh = NULL;
204 continue;
205 }
206 set_buffer_verified(bh);
207
208 /* If the dir block has changed since the last call to
209 * readdir(2), then we might be pointing to an invalid
210 * dirent right now. Scan from the start of the block
211 * to make sure. */
212 if (!inode_eq_iversion(inode, file->f_version)) {
213 for (i = 0; i < sb->s_blocksize && i < offset; ) {
214 de = (struct ext4_dir_entry_2 *)
215 (bh->b_data + i);
216 /* It's too expensive to do a full
217 * dirent test each time round this
218 * loop, but we do have to test at
219 * least that it is non-zero. A
220 * failure will be detected in the
221 * dirent test below. */
222 if (ext4_rec_len_from_disk(de->rec_len,
223 sb->s_blocksize) < EXT4_DIR_REC_LEN(1))
224 break;
225 i += ext4_rec_len_from_disk(de->rec_len,
226 sb->s_blocksize);
227 }
228 offset = i;
229 ctx->pos = (ctx->pos & ~(sb->s_blocksize - 1))
230 | offset;
231 file->f_version = inode_query_iversion(inode);
232 }
233
234 while (ctx->pos < inode->i_size
235 && offset < sb->s_blocksize) {
236 de = (struct ext4_dir_entry_2 *) (bh->b_data + offset);
237 if (ext4_check_dir_entry(inode, file, de, bh,
238 bh->b_data, bh->b_size,
239 offset)) {
240 /*
241 * On error, skip to the next block
242 */
243 ctx->pos = (ctx->pos |
244 (sb->s_blocksize - 1)) + 1;
245 break;
246 }
247 offset += ext4_rec_len_from_disk(de->rec_len,
248 sb->s_blocksize);
249 if (le32_to_cpu(de->inode)) {
250 if (!ext4_encrypted_inode(inode)) {
251 if (!dir_emit(ctx, de->name,
252 de->name_len,
253 le32_to_cpu(de->inode),
254 get_dtype(sb, de->file_type)))
255 goto done;
256 } else {
257 int save_len = fstr.len;
258 struct fscrypt_str de_name =
259 FSTR_INIT(de->name,
260 de->name_len);
261
262 /* Directory is encrypted */
263 err = fscrypt_fname_disk_to_usr(inode,
264 0, 0, &de_name, &fstr);
265 de_name = fstr;
266 fstr.len = save_len;
267 if (err)
268 goto errout;
269 if (!dir_emit(ctx,
270 de_name.name, de_name.len,
271 le32_to_cpu(de->inode),
272 get_dtype(sb, de->file_type)))
273 goto done;
274 }
275 }
276 ctx->pos += ext4_rec_len_from_disk(de->rec_len,
277 sb->s_blocksize);
278 }
279 if ((ctx->pos < inode->i_size) && !dir_relax_shared(inode))
280 goto done;
281 brelse(bh);
282 bh = NULL;
283 offset = 0;
284 }
285done:
286 err = 0;
287errout:
288#ifdef CONFIG_EXT4_FS_ENCRYPTION
289 fscrypt_fname_free_buffer(&fstr);
290#endif
291 brelse(bh);
292 return err;
293}
294
295static inline int is_32bit_api(void)
296{
297#ifdef CONFIG_COMPAT
298 return in_compat_syscall();
299#else
300 return (BITS_PER_LONG == 32);
301#endif
302}
303
304/*
305 * These functions convert from the major/minor hash to an f_pos
306 * value for dx directories
307 *
308 * Upper layer (for example NFS) should specify FMODE_32BITHASH or
309 * FMODE_64BITHASH explicitly. On the other hand, we allow ext4 to be mounted
310 * directly on both 32-bit and 64-bit nodes, under such case, neither
311 * FMODE_32BITHASH nor FMODE_64BITHASH is specified.
312 */
313static inline loff_t hash2pos(struct file *filp, __u32 major, __u32 minor)
314{
315 if ((filp->f_mode & FMODE_32BITHASH) ||
316 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
317 return major >> 1;
318 else
319 return ((__u64)(major >> 1) << 32) | (__u64)minor;
320}
321
322static inline __u32 pos2maj_hash(struct file *filp, loff_t pos)
323{
324 if ((filp->f_mode & FMODE_32BITHASH) ||
325 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
326 return (pos << 1) & 0xffffffff;
327 else
328 return ((pos >> 32) << 1) & 0xffffffff;
329}
330
331static inline __u32 pos2min_hash(struct file *filp, loff_t pos)
332{
333 if ((filp->f_mode & FMODE_32BITHASH) ||
334 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
335 return 0;
336 else
337 return pos & 0xffffffff;
338}
339
340/*
341 * Return 32- or 64-bit end-of-file for dx directories
342 */
343static inline loff_t ext4_get_htree_eof(struct file *filp)
344{
345 if ((filp->f_mode & FMODE_32BITHASH) ||
346 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
347 return EXT4_HTREE_EOF_32BIT;
348 else
349 return EXT4_HTREE_EOF_64BIT;
350}
351
352
353/*
354 * ext4_dir_llseek() calls generic_file_llseek_size to handle htree
355 * directories, where the "offset" is in terms of the filename hash
356 * value instead of the byte offset.
357 *
358 * Because we may return a 64-bit hash that is well beyond offset limits,
359 * we need to pass the max hash as the maximum allowable offset in
360 * the htree directory case.
361 *
362 * For non-htree, ext4_llseek already chooses the proper max offset.
363 */
364static loff_t ext4_dir_llseek(struct file *file, loff_t offset, int whence)
365{
366 struct inode *inode = file->f_mapping->host;
367 int dx_dir = is_dx_dir(inode);
368 loff_t ret, htree_max = ext4_get_htree_eof(file);
369
370 if (likely(dx_dir))
371 ret = generic_file_llseek_size(file, offset, whence,
372 htree_max, htree_max);
373 else
374 ret = ext4_llseek(file, offset, whence);
375 file->f_version = inode_peek_iversion(inode) - 1;
376 return ret;
377}
378
379/*
380 * This structure holds the nodes of the red-black tree used to store
381 * the directory entry in hash order.
382 */
383struct fname {
384 __u32 hash;
385 __u32 minor_hash;
386 struct rb_node rb_hash;
387 struct fname *next;
388 __u32 inode;
389 __u8 name_len;
390 __u8 file_type;
391 char name[0];
392};
393
394/*
395 * This functoin implements a non-recursive way of freeing all of the
396 * nodes in the red-black tree.
397 */
398static void free_rb_tree_fname(struct rb_root *root)
399{
400 struct fname *fname, *next;
401
402 rbtree_postorder_for_each_entry_safe(fname, next, root, rb_hash)
403 while (fname) {
404 struct fname *old = fname;
405 fname = fname->next;
406 kfree(old);
407 }
408
409 *root = RB_ROOT;
410}
411
412
413static struct dir_private_info *ext4_htree_create_dir_info(struct file *filp,
414 loff_t pos)
415{
416 struct dir_private_info *p;
417
418 p = kzalloc(sizeof(*p), GFP_KERNEL);
419 if (!p)
420 return NULL;
421 p->curr_hash = pos2maj_hash(filp, pos);
422 p->curr_minor_hash = pos2min_hash(filp, pos);
423 return p;
424}
425
426void ext4_htree_free_dir_info(struct dir_private_info *p)
427{
428 free_rb_tree_fname(&p->root);
429 kfree(p);
430}
431
432/*
433 * Given a directory entry, enter it into the fname rb tree.
434 *
435 * When filename encryption is enabled, the dirent will hold the
436 * encrypted filename, while the htree will hold decrypted filename.
437 * The decrypted filename is passed in via ent_name. parameter.
438 */
439int ext4_htree_store_dirent(struct file *dir_file, __u32 hash,
440 __u32 minor_hash,
441 struct ext4_dir_entry_2 *dirent,
442 struct fscrypt_str *ent_name)
443{
444 struct rb_node **p, *parent = NULL;
445 struct fname *fname, *new_fn;
446 struct dir_private_info *info;
447 int len;
448
449 info = dir_file->private_data;
450 p = &info->root.rb_node;
451
452 /* Create and allocate the fname structure */
453 len = sizeof(struct fname) + ent_name->len + 1;
454 new_fn = kzalloc(len, GFP_KERNEL);
455 if (!new_fn)
456 return -ENOMEM;
457 new_fn->hash = hash;
458 new_fn->minor_hash = minor_hash;
459 new_fn->inode = le32_to_cpu(dirent->inode);
460 new_fn->name_len = ent_name->len;
461 new_fn->file_type = dirent->file_type;
462 memcpy(new_fn->name, ent_name->name, ent_name->len);
463 new_fn->name[ent_name->len] = 0;
464
465 while (*p) {
466 parent = *p;
467 fname = rb_entry(parent, struct fname, rb_hash);
468
469 /*
470 * If the hash and minor hash match up, then we put
471 * them on a linked list. This rarely happens...
472 */
473 if ((new_fn->hash == fname->hash) &&
474 (new_fn->minor_hash == fname->minor_hash)) {
475 new_fn->next = fname->next;
476 fname->next = new_fn;
477 return 0;
478 }
479
480 if (new_fn->hash < fname->hash)
481 p = &(*p)->rb_left;
482 else if (new_fn->hash > fname->hash)
483 p = &(*p)->rb_right;
484 else if (new_fn->minor_hash < fname->minor_hash)
485 p = &(*p)->rb_left;
486 else /* if (new_fn->minor_hash > fname->minor_hash) */
487 p = &(*p)->rb_right;
488 }
489
490 rb_link_node(&new_fn->rb_hash, parent, p);
491 rb_insert_color(&new_fn->rb_hash, &info->root);
492 return 0;
493}
494
495
496
497/*
498 * This is a helper function for ext4_dx_readdir. It calls filldir
499 * for all entres on the fname linked list. (Normally there is only
500 * one entry on the linked list, unless there are 62 bit hash collisions.)
501 */
502static int call_filldir(struct file *file, struct dir_context *ctx,
503 struct fname *fname)
504{
505 struct dir_private_info *info = file->private_data;
506 struct inode *inode = file_inode(file);
507 struct super_block *sb = inode->i_sb;
508
509 if (!fname) {
510 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: comm %s: "
511 "called with null fname?!?", __func__, __LINE__,
512 inode->i_ino, current->comm);
513 return 0;
514 }
515 ctx->pos = hash2pos(file, fname->hash, fname->minor_hash);
516 while (fname) {
517 if (!dir_emit(ctx, fname->name,
518 fname->name_len,
519 fname->inode,
520 get_dtype(sb, fname->file_type))) {
521 info->extra_fname = fname;
522 return 1;
523 }
524 fname = fname->next;
525 }
526 return 0;
527}
528
529static int ext4_dx_readdir(struct file *file, struct dir_context *ctx)
530{
531 struct dir_private_info *info = file->private_data;
532 struct inode *inode = file_inode(file);
533 struct fname *fname;
534 int ret;
535
536 if (!info) {
537 info = ext4_htree_create_dir_info(file, ctx->pos);
538 if (!info)
539 return -ENOMEM;
540 file->private_data = info;
541 }
542
543 if (ctx->pos == ext4_get_htree_eof(file))
544 return 0; /* EOF */
545
546 /* Some one has messed with f_pos; reset the world */
547 if (info->last_pos != ctx->pos) {
548 free_rb_tree_fname(&info->root);
549 info->curr_node = NULL;
550 info->extra_fname = NULL;
551 info->curr_hash = pos2maj_hash(file, ctx->pos);
552 info->curr_minor_hash = pos2min_hash(file, ctx->pos);
553 }
554
555 /*
556 * If there are any leftover names on the hash collision
557 * chain, return them first.
558 */
559 if (info->extra_fname) {
560 if (call_filldir(file, ctx, info->extra_fname))
561 goto finished;
562 info->extra_fname = NULL;
563 goto next_node;
564 } else if (!info->curr_node)
565 info->curr_node = rb_first(&info->root);
566
567 while (1) {
568 /*
569 * Fill the rbtree if we have no more entries,
570 * or the inode has changed since we last read in the
571 * cached entries.
572 */
573 if ((!info->curr_node) ||
574 !inode_eq_iversion(inode, file->f_version)) {
575 info->curr_node = NULL;
576 free_rb_tree_fname(&info->root);
577 file->f_version = inode_query_iversion(inode);
578 ret = ext4_htree_fill_tree(file, info->curr_hash,
579 info->curr_minor_hash,
580 &info->next_hash);
581 if (ret < 0)
582 return ret;
583 if (ret == 0) {
584 ctx->pos = ext4_get_htree_eof(file);
585 break;
586 }
587 info->curr_node = rb_first(&info->root);
588 }
589
590 fname = rb_entry(info->curr_node, struct fname, rb_hash);
591 info->curr_hash = fname->hash;
592 info->curr_minor_hash = fname->minor_hash;
593 if (call_filldir(file, ctx, fname))
594 break;
595 next_node:
596 info->curr_node = rb_next(info->curr_node);
597 if (info->curr_node) {
598 fname = rb_entry(info->curr_node, struct fname,
599 rb_hash);
600 info->curr_hash = fname->hash;
601 info->curr_minor_hash = fname->minor_hash;
602 } else {
603 if (info->next_hash == ~0) {
604 ctx->pos = ext4_get_htree_eof(file);
605 break;
606 }
607 info->curr_hash = info->next_hash;
608 info->curr_minor_hash = 0;
609 }
610 }
611finished:
612 info->last_pos = ctx->pos;
613 return 0;
614}
615
616static int ext4_dir_open(struct inode * inode, struct file * filp)
617{
618 if (ext4_encrypted_inode(inode))
619 return fscrypt_get_encryption_info(inode) ? -EACCES : 0;
620 return 0;
621}
622
623static int ext4_release_dir(struct inode *inode, struct file *filp)
624{
625 if (filp->private_data)
626 ext4_htree_free_dir_info(filp->private_data);
627
628 return 0;
629}
630
631int ext4_check_all_de(struct inode *dir, struct buffer_head *bh, void *buf,
632 int buf_size)
633{
634 struct ext4_dir_entry_2 *de;
635 int rlen;
636 unsigned int offset = 0;
637 char *top;
638
639 de = (struct ext4_dir_entry_2 *)buf;
640 top = buf + buf_size;
641 while ((char *) de < top) {
642 if (ext4_check_dir_entry(dir, NULL, de, bh,
643 buf, buf_size, offset))
644 return -EFSCORRUPTED;
645 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
646 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
647 offset += rlen;
648 }
649 if ((char *) de > top)
650 return -EFSCORRUPTED;
651
652 return 0;
653}
654
655const struct file_operations ext4_dir_operations = {
656 .llseek = ext4_dir_llseek,
657 .read = generic_read_dir,
658 .iterate_shared = ext4_readdir,
659 .unlocked_ioctl = ext4_ioctl,
660#ifdef CONFIG_COMPAT
661 .compat_ioctl = ext4_compat_ioctl,
662#endif
663 .fsync = ext4_sync_file,
664 .open = ext4_dir_open,
665 .release = ext4_release_dir,
666};
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