1// SPDX-License-Identifier: GPL-2.0
2/*
3 *
4 * Copyright (C) 2019-2021 Paragon Software GmbH, All rights reserved.
5 *
6 * Directory handling functions for NTFS-based filesystems.
7 *
8 */
9
10#include <linux/fs.h>
11#include <linux/nls.h>
12
13#include "debug.h"
14#include "ntfs.h"
15#include "ntfs_fs.h"
16
17/* Convert little endian UTF-16 to NLS string. */
18int ntfs_utf16_to_nls(struct ntfs_sb_info *sbi, const __le16 *name, u32 len,
19 u8 *buf, int buf_len)
20{
21 int ret, warn;
22 u8 *op;
23 struct nls_table *nls = sbi->options->nls;
24
25 static_assert(sizeof(wchar_t) == sizeof(__le16));
26
27 if (!nls) {
28 /* UTF-16 -> UTF-8 */
29 ret = utf16s_to_utf8s(pwcs: (wchar_t *)name, len, endian: UTF16_LITTLE_ENDIAN,
30 s: buf, maxlen: buf_len);
31 buf[ret] = '\0';
32 return ret;
33 }
34
35 op = buf;
36 warn = 0;
37
38 while (len--) {
39 u16 ec;
40 int charlen;
41 char dump[5];
42
43 if (buf_len < NLS_MAX_CHARSET_SIZE) {
44 ntfs_warn(sbi->sb,
45 "filename was truncated while converting.");
46 break;
47 }
48
49 ec = le16_to_cpu(*name++);
50 charlen = nls->uni2char(ec, op, buf_len);
51
52 if (charlen > 0) {
53 op += charlen;
54 buf_len -= charlen;
55 continue;
56 }
57
58 *op++ = '_';
59 buf_len -= 1;
60 if (warn)
61 continue;
62
63 warn = 1;
64 hex_byte_pack(buf: &dump[0], byte: ec >> 8);
65 hex_byte_pack(buf: &dump[2], byte: ec);
66 dump[4] = 0;
67
68 ntfs_err(sbi->sb, "failed to convert \"%s\" to %s", dump,
69 nls->charset);
70 }
71
72 *op = '\0';
73 return op - buf;
74}
75
76// clang-format off
77#define PLANE_SIZE 0x00010000
78
79#define SURROGATE_PAIR 0x0000d800
80#define SURROGATE_LOW 0x00000400
81#define SURROGATE_BITS 0x000003ff
82// clang-format on
83
84/*
85 * put_utf16 - Modified version of put_utf16 from fs/nls/nls_base.c
86 *
87 * Function is sparse warnings free.
88 */
89static inline void put_utf16(wchar_t *s, unsigned int c,
90 enum utf16_endian endian)
91{
92 static_assert(sizeof(wchar_t) == sizeof(__le16));
93 static_assert(sizeof(wchar_t) == sizeof(__be16));
94
95 switch (endian) {
96 default:
97 *s = (wchar_t)c;
98 break;
99 case UTF16_LITTLE_ENDIAN:
100 *(__le16 *)s = __cpu_to_le16(c);
101 break;
102 case UTF16_BIG_ENDIAN:
103 *(__be16 *)s = __cpu_to_be16(c);
104 break;
105 }
106}
107
108/*
109 * _utf8s_to_utf16s
110 *
111 * Modified version of 'utf8s_to_utf16s' allows to
112 * detect -ENAMETOOLONG without writing out of expected maximum.
113 */
114static int _utf8s_to_utf16s(const u8 *s, int inlen, enum utf16_endian endian,
115 wchar_t *pwcs, int maxout)
116{
117 u16 *op;
118 int size;
119 unicode_t u;
120
121 op = pwcs;
122 while (inlen > 0 && *s) {
123 if (*s & 0x80) {
124 size = utf8_to_utf32(s, len: inlen, pu: &u);
125 if (size < 0)
126 return -EINVAL;
127 s += size;
128 inlen -= size;
129
130 if (u >= PLANE_SIZE) {
131 if (maxout < 2)
132 return -ENAMETOOLONG;
133
134 u -= PLANE_SIZE;
135 put_utf16(s: op++,
136 SURROGATE_PAIR |
137 ((u >> 10) & SURROGATE_BITS),
138 endian);
139 put_utf16(s: op++,
140 SURROGATE_PAIR | SURROGATE_LOW |
141 (u & SURROGATE_BITS),
142 endian);
143 maxout -= 2;
144 } else {
145 if (maxout < 1)
146 return -ENAMETOOLONG;
147
148 put_utf16(s: op++, c: u, endian);
149 maxout--;
150 }
151 } else {
152 if (maxout < 1)
153 return -ENAMETOOLONG;
154
155 put_utf16(s: op++, c: *s++, endian);
156 inlen--;
157 maxout--;
158 }
159 }
160 return op - pwcs;
161}
162
163/*
164 * ntfs_nls_to_utf16 - Convert input string to UTF-16.
165 * @name: Input name.
166 * @name_len: Input name length.
167 * @uni: Destination memory.
168 * @max_ulen: Destination memory.
169 * @endian: Endian of target UTF-16 string.
170 *
171 * This function is called:
172 * - to create NTFS name
173 * - to create symlink
174 *
175 * Return: UTF-16 string length or error (if negative).
176 */
177int ntfs_nls_to_utf16(struct ntfs_sb_info *sbi, const u8 *name, u32 name_len,
178 struct cpu_str *uni, u32 max_ulen,
179 enum utf16_endian endian)
180{
181 int ret, slen;
182 const u8 *end;
183 struct nls_table *nls = sbi->options->nls;
184 u16 *uname = uni->name;
185
186 static_assert(sizeof(wchar_t) == sizeof(u16));
187
188 if (!nls) {
189 /* utf8 -> utf16 */
190 ret = _utf8s_to_utf16s(s: name, inlen: name_len, endian, pwcs: uname, maxout: max_ulen);
191 uni->len = ret;
192 return ret;
193 }
194
195 for (ret = 0, end = name + name_len; name < end; ret++, name += slen) {
196 if (ret >= max_ulen)
197 return -ENAMETOOLONG;
198
199 slen = nls->char2uni(name, end - name, uname + ret);
200 if (!slen)
201 return -EINVAL;
202 if (slen < 0)
203 return slen;
204 }
205
206#ifdef __BIG_ENDIAN
207 if (endian == UTF16_LITTLE_ENDIAN) {
208 int i = ret;
209
210 while (i--) {
211 __cpu_to_le16s(uname);
212 uname++;
213 }
214 }
215#else
216 if (endian == UTF16_BIG_ENDIAN) {
217 int i = ret;
218
219 while (i--) {
220 __cpu_to_be16s(uname);
221 uname++;
222 }
223 }
224#endif
225
226 uni->len = ret;
227 return ret;
228}
229
230/*
231 * dir_search_u - Helper function.
232 */
233struct inode *dir_search_u(struct inode *dir, const struct cpu_str *uni,
234 struct ntfs_fnd *fnd)
235{
236 int err = 0;
237 struct super_block *sb = dir->i_sb;
238 struct ntfs_sb_info *sbi = sb->s_fs_info;
239 struct ntfs_inode *ni = ntfs_i(inode: dir);
240 struct NTFS_DE *e;
241 int diff;
242 struct inode *inode = NULL;
243 struct ntfs_fnd *fnd_a = NULL;
244
245 if (!fnd) {
246 fnd_a = fnd_get();
247 if (!fnd_a) {
248 err = -ENOMEM;
249 goto out;
250 }
251 fnd = fnd_a;
252 }
253
254 err = indx_find(indx: &ni->dir, dir: ni, NULL, Key: uni, KeyLen: 0, param: sbi, diff: &diff, entry: &e, fnd);
255
256 if (err)
257 goto out;
258
259 if (diff) {
260 err = -ENOENT;
261 goto out;
262 }
263
264 inode = ntfs_iget5(sb, ref: &e->ref, name: uni);
265 if (!IS_ERR(ptr: inode) && is_bad_inode(inode)) {
266 iput(inode);
267 err = -EINVAL;
268 }
269out:
270 fnd_put(fnd: fnd_a);
271
272 return err == -ENOENT ? NULL : err ? ERR_PTR(error: err) : inode;
273}
274
275static inline int ntfs_filldir(struct ntfs_sb_info *sbi, struct ntfs_inode *ni,
276 const struct NTFS_DE *e, u8 *name,
277 struct dir_context *ctx)
278{
279 const struct ATTR_FILE_NAME *fname;
280 unsigned long ino;
281 int name_len;
282 u32 dt_type;
283
284 fname = Add2Ptr(e, sizeof(struct NTFS_DE));
285
286 if (fname->type == FILE_NAME_DOS)
287 return 0;
288
289 if (!mi_is_ref(mi: &ni->mi, ref: &fname->home))
290 return 0;
291
292 ino = ino_get(ref: &e->ref);
293
294 if (ino == MFT_REC_ROOT)
295 return 0;
296
297 /* Skip meta files. Unless option to show metafiles is set. */
298 if (!sbi->options->showmeta && ntfs_is_meta_file(sbi, rno: ino))
299 return 0;
300
301 if (sbi->options->nohidden && (fname->dup.fa & FILE_ATTRIBUTE_HIDDEN))
302 return 0;
303
304 name_len = ntfs_utf16_to_nls(sbi, name: fname->name, len: fname->name_len, buf: name,
305 PATH_MAX);
306 if (name_len <= 0) {
307 ntfs_warn(sbi->sb, "failed to convert name for inode %lx.",
308 ino);
309 return 0;
310 }
311
312 /*
313 * NTFS: symlinks are "dir + reparse" or "file + reparse"
314 * Unfortunately reparse attribute is used for many purposes (several dozens).
315 * It is not possible here to know is this name symlink or not.
316 * To get exactly the type of name we should to open inode (read mft).
317 * getattr for opened file (fstat) correctly returns symlink.
318 */
319 dt_type = (fname->dup.fa & FILE_ATTRIBUTE_DIRECTORY) ? DT_DIR : DT_REG;
320
321 /*
322 * It is not reliable to detect the type of name using duplicated information
323 * stored in parent directory.
324 * The only correct way to get the type of name - read MFT record and find ATTR_STD.
325 * The code below is not good idea.
326 * It does additional locks/reads just to get the type of name.
327 * Should we use additional mount option to enable branch below?
328 */
329 if ((fname->dup.fa & FILE_ATTRIBUTE_REPARSE_POINT) &&
330 ino != ni->mi.rno) {
331 struct inode *inode = ntfs_iget5(sb: sbi->sb, ref: &e->ref, NULL);
332 if (!IS_ERR_OR_NULL(ptr: inode)) {
333 dt_type = fs_umode_to_dtype(mode: inode->i_mode);
334 iput(inode);
335 }
336 }
337
338 return !dir_emit(ctx, name: (s8 *)name, namelen: name_len, ino, type: dt_type);
339}
340
341/*
342 * ntfs_read_hdr - Helper function for ntfs_readdir().
343 */
344static int ntfs_read_hdr(struct ntfs_sb_info *sbi, struct ntfs_inode *ni,
345 const struct INDEX_HDR *hdr, u64 vbo, u64 pos,
346 u8 *name, struct dir_context *ctx)
347{
348 int err;
349 const struct NTFS_DE *e;
350 u32 e_size;
351 u32 end = le32_to_cpu(hdr->used);
352 u32 off = le32_to_cpu(hdr->de_off);
353
354 for (;; off += e_size) {
355 if (off + sizeof(struct NTFS_DE) > end)
356 return -1;
357
358 e = Add2Ptr(hdr, off);
359 e_size = le16_to_cpu(e->size);
360 if (e_size < sizeof(struct NTFS_DE) || off + e_size > end)
361 return -1;
362
363 if (de_is_last(e))
364 return 0;
365
366 /* Skip already enumerated. */
367 if (vbo + off < pos)
368 continue;
369
370 if (le16_to_cpu(e->key_size) < SIZEOF_ATTRIBUTE_FILENAME)
371 return -1;
372
373 ctx->pos = vbo + off;
374
375 /* Submit the name to the filldir callback. */
376 err = ntfs_filldir(sbi, ni, e, name, ctx);
377 if (err)
378 return err;
379 }
380}
381
382/*
383 * ntfs_readdir - file_operations::iterate_shared
384 *
385 * Use non sorted enumeration.
386 * We have an example of broken volume where sorted enumeration
387 * counts each name twice.
388 */
389static int ntfs_readdir(struct file *file, struct dir_context *ctx)
390{
391 const struct INDEX_ROOT *root;
392 u64 vbo;
393 size_t bit;
394 loff_t eod;
395 int err = 0;
396 struct inode *dir = file_inode(f: file);
397 struct ntfs_inode *ni = ntfs_i(inode: dir);
398 struct super_block *sb = dir->i_sb;
399 struct ntfs_sb_info *sbi = sb->s_fs_info;
400 loff_t i_size = i_size_read(inode: dir);
401 u32 pos = ctx->pos;
402 u8 *name = NULL;
403 struct indx_node *node = NULL;
404 u8 index_bits = ni->dir.index_bits;
405
406 /* Name is a buffer of PATH_MAX length. */
407 static_assert(NTFS_NAME_LEN * 4 < PATH_MAX);
408
409 eod = i_size + sbi->record_size;
410
411 if (pos >= eod)
412 return 0;
413
414 if (!dir_emit_dots(file, ctx))
415 return 0;
416
417 /* Allocate PATH_MAX bytes. */
418 name = __getname();
419 if (!name)
420 return -ENOMEM;
421
422 if (!ni->mi_loaded && ni->attr_list.size) {
423 /*
424 * Directory inode is locked for read.
425 * Load all subrecords to avoid 'write' access to 'ni' during
426 * directory reading.
427 */
428 ni_lock(ni);
429 if (!ni->mi_loaded && ni->attr_list.size) {
430 err = ni_load_all_mi(ni);
431 if (!err)
432 ni->mi_loaded = true;
433 }
434 ni_unlock(ni);
435 if (err)
436 goto out;
437 }
438
439 root = indx_get_root(indx: &ni->dir, ni, NULL, NULL);
440 if (!root) {
441 err = -EINVAL;
442 goto out;
443 }
444
445 if (pos >= sbi->record_size) {
446 bit = (pos - sbi->record_size) >> index_bits;
447 } else {
448 err = ntfs_read_hdr(sbi, ni, hdr: &root->ihdr, vbo: 0, pos, name, ctx);
449 if (err)
450 goto out;
451 bit = 0;
452 }
453
454 if (!i_size) {
455 ctx->pos = eod;
456 goto out;
457 }
458
459 for (;;) {
460 vbo = (u64)bit << index_bits;
461 if (vbo >= i_size) {
462 ctx->pos = eod;
463 goto out;
464 }
465
466 err = indx_used_bit(indx: &ni->dir, ni, bit: &bit);
467 if (err)
468 goto out;
469
470 if (bit == MINUS_ONE_T) {
471 ctx->pos = eod;
472 goto out;
473 }
474
475 vbo = (u64)bit << index_bits;
476 if (vbo >= i_size) {
477 ntfs_inode_err(dir, "Looks like your dir is corrupt");
478 err = -EINVAL;
479 goto out;
480 }
481
482 err = indx_read(idx: &ni->dir, ni, vbn: bit << ni->dir.idx2vbn_bits,
483 node: &node);
484 if (err)
485 goto out;
486
487 err = ntfs_read_hdr(sbi, ni, hdr: &node->index->ihdr,
488 vbo: vbo + sbi->record_size, pos, name, ctx);
489 if (err)
490 goto out;
491
492 bit += 1;
493 }
494
495out:
496
497 __putname(name);
498 put_indx_node(in: node);
499
500 if (err == -ENOENT) {
501 err = 0;
502 ctx->pos = pos;
503 }
504
505 return err;
506}
507
508static int ntfs_dir_count(struct inode *dir, bool *is_empty, size_t *dirs,
509 size_t *files)
510{
511 int err = 0;
512 struct ntfs_inode *ni = ntfs_i(inode: dir);
513 struct NTFS_DE *e = NULL;
514 struct INDEX_ROOT *root;
515 struct INDEX_HDR *hdr;
516 const struct ATTR_FILE_NAME *fname;
517 u32 e_size, off, end;
518 size_t drs = 0, fles = 0, bit = 0;
519 struct indx_node *node = NULL;
520 size_t max_indx = i_size_read(inode: &ni->vfs_inode) >> ni->dir.index_bits;
521
522 if (is_empty)
523 *is_empty = true;
524
525 root = indx_get_root(indx: &ni->dir, ni, NULL, NULL);
526 if (!root)
527 return -EINVAL;
528
529 hdr = &root->ihdr;
530
531 for (;;) {
532 end = le32_to_cpu(hdr->used);
533 off = le32_to_cpu(hdr->de_off);
534
535 for (; off + sizeof(struct NTFS_DE) <= end; off += e_size) {
536 e = Add2Ptr(hdr, off);
537 e_size = le16_to_cpu(e->size);
538 if (e_size < sizeof(struct NTFS_DE) ||
539 off + e_size > end) {
540 /* Looks like corruption. */
541 break;
542 }
543
544 if (de_is_last(e))
545 break;
546
547 fname = de_get_fname(e);
548 if (!fname)
549 continue;
550
551 if (fname->type == FILE_NAME_DOS)
552 continue;
553
554 if (is_empty) {
555 *is_empty = false;
556 if (!dirs && !files)
557 goto out;
558 }
559
560 if (fname->dup.fa & FILE_ATTRIBUTE_DIRECTORY)
561 drs += 1;
562 else
563 fles += 1;
564 }
565
566 if (bit >= max_indx)
567 goto out;
568
569 err = indx_used_bit(indx: &ni->dir, ni, bit: &bit);
570 if (err)
571 goto out;
572
573 if (bit == MINUS_ONE_T)
574 goto out;
575
576 if (bit >= max_indx)
577 goto out;
578
579 err = indx_read(idx: &ni->dir, ni, vbn: bit << ni->dir.idx2vbn_bits,
580 node: &node);
581 if (err)
582 goto out;
583
584 hdr = &node->index->ihdr;
585 bit += 1;
586 }
587
588out:
589 put_indx_node(in: node);
590 if (dirs)
591 *dirs = drs;
592 if (files)
593 *files = fles;
594
595 return err;
596}
597
598bool dir_is_empty(struct inode *dir)
599{
600 bool is_empty = false;
601
602 ntfs_dir_count(dir, is_empty: &is_empty, NULL, NULL);
603
604 return is_empty;
605}
606
607// clang-format off
608const struct file_operations ntfs_dir_operations = {
609 .llseek = generic_file_llseek,
610 .read = generic_read_dir,
611 .iterate_shared = ntfs_readdir,
612 .fsync = generic_file_fsync,
613 .open = ntfs_file_open,
614 .unlocked_ioctl = ntfs_ioctl,
615#ifdef CONFIG_COMPAT
616 .compat_ioctl = ntfs_compat_ioctl,
617#endif
618};
619// clang-format on
620

source code of linux/fs/ntfs3/dir.c