xref: /freebsd/lib/libc/gen/glob-compat11.c (revision c19fb1f963e3dc88a82b20d1b17f94a4cd321e74)
1 /*
2  * Copyright (c) 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * This code is derived from software contributed to Berkeley by
6  * Guido van Rossum.
7  *
8  * Copyright (c) 2011 The FreeBSD Foundation
9  * All rights reserved.
10  * Portions of this software were developed by David Chisnall
11  * under sponsorship from the FreeBSD Foundation.
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
16  * 1. Redistributions of source code must retain the above copyright
17  *    notice, this list of conditions and the following disclaimer.
18  * 2. Redistributions in binary form must reproduce the above copyright
19  *    notice, this list of conditions and the following disclaimer in the
20  *    documentation and/or other materials provided with the distribution.
21  * 3. Neither the name of the University nor the names of its contributors
22  *    may be used to endorse or promote products derived from this software
23  *    without specific prior written permission.
24  *
25  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35  * SUCH DAMAGE.
36  * From: FreeBSD: head/lib/libc/gen/glob.c 317913 2017-05-07 19:52:56Z jilles
37  */
38 
39 #include <sys/param.h>
40 #define	_WANT_FREEBSD11_STAT
41 #include <sys/stat.h>
42 
43 #include <ctype.h>
44 #define	_WANT_FREEBSD11_DIRENT
45 #include <dirent.h>
46 #include <errno.h>
47 #include <glob.h>
48 #include <limits.h>
49 #include <pwd.h>
50 #include <stdint.h>
51 #include <stdio.h>
52 #include <stdlib.h>
53 #include <string.h>
54 #include <unistd.h>
55 #include <wchar.h>
56 
57 #include "collate.h"
58 #include "gen-compat.h"
59 #include "glob-compat11.h"
60 
61 /*
62  * glob(3) expansion limits. Stop the expansion if any of these limits
63  * is reached. This caps the runtime in the face of DoS attacks. See
64  * also CVE-2010-2632
65  */
66 #define	GLOB_LIMIT_BRACE	128	/* number of brace calls */
67 #define	GLOB_LIMIT_PATH		65536	/* number of path elements */
68 #define	GLOB_LIMIT_READDIR	16384	/* number of readdirs */
69 #define	GLOB_LIMIT_STAT		1024	/* number of stat system calls */
70 #define	GLOB_LIMIT_STRING	ARG_MAX	/* maximum total size for paths */
71 
72 struct glob_limit {
73 	size_t	l_brace_cnt;
74 	size_t	l_path_lim;
75 	size_t	l_readdir_cnt;
76 	size_t	l_stat_cnt;
77 	size_t	l_string_cnt;
78 };
79 
80 #define	DOT		L'.'
81 #define	EOS		L'\0'
82 #define	LBRACKET	L'['
83 #define	NOT		L'!'
84 #define	QUESTION	L'?'
85 #define	QUOTE		L'\\'
86 #define	RANGE		L'-'
87 #define	RBRACKET	L']'
88 #define	SEP		L'/'
89 #define	STAR		L'*'
90 #define	TILDE		L'~'
91 #define	LBRACE		L'{'
92 #define	RBRACE		L'}'
93 #define	COMMA		L','
94 
95 #define	M_QUOTE		0x8000000000ULL
96 #define	M_PROTECT	0x4000000000ULL
97 #define	M_MASK		0xffffffffffULL
98 #define	M_CHAR		0x00ffffffffULL
99 
100 typedef uint_fast64_t Char;
101 
102 #define	CHAR(c)		((Char)((c)&M_CHAR))
103 #define	META(c)		((Char)((c)|M_QUOTE))
104 #define	UNPROT(c)	((c) & ~M_PROTECT)
105 #define	M_ALL		META(L'*')
106 #define	M_END		META(L']')
107 #define	M_NOT		META(L'!')
108 #define	M_ONE		META(L'?')
109 #define	M_RNG		META(L'-')
110 #define	M_SET		META(L'[')
111 #define	ismeta(c)	(((c)&M_QUOTE) != 0)
112 #ifdef DEBUG
113 #define	isprot(c)	(((c)&M_PROTECT) != 0)
114 #endif
115 
116 static int	 compare(const void *, const void *);
117 static int	 g_Ctoc(const Char *, char *, size_t);
118 static int	 g_lstat(Char *, struct freebsd11_stat *, glob11_t *);
119 static DIR	*g_opendir(Char *, glob11_t *);
120 static const Char *g_strchr(const Char *, wchar_t);
121 #ifdef notdef
122 static Char	*g_strcat(Char *, const Char *);
123 #endif
124 static int	 g_stat(Char *, struct freebsd11_stat *, glob11_t *);
125 static int	 glob0(const Char *, glob11_t *, struct glob_limit *,
126     const char *);
127 static int	 glob1(Char *, glob11_t *, struct glob_limit *);
128 static int	 glob2(Char *, Char *, Char *, Char *, glob11_t *,
129     struct glob_limit *);
130 static int	 glob3(Char *, Char *, Char *, Char *, Char *, glob11_t *,
131     struct glob_limit *);
132 static int	 globextend(const Char *, glob11_t *, struct glob_limit *,
133     const char *);
134 static const Char *
135 		 globtilde(const Char *, Char *, size_t, glob11_t *);
136 static int	 globexp0(const Char *, glob11_t *, struct glob_limit *,
137     const char *);
138 static int	 globexp1(const Char *, glob11_t *, struct glob_limit *);
139 static int	 globexp2(const Char *, const Char *, glob11_t *,
140     struct glob_limit *);
141 static int	 globfinal(glob11_t *, struct glob_limit *, size_t,
142     const char *);
143 static int	 match(Char *, Char *, Char *);
144 static int	 err_nomatch(glob11_t *, struct glob_limit *, const char *);
145 static int	 err_aborted(glob11_t *, int, char *);
146 #ifdef DEBUG
147 static void	 qprintf(const char *, Char *);
148 #endif
149 
150 int
151 freebsd11_glob(const char * __restrict pattern, int flags,
152 	 int (*errfunc)(const char *, int), glob11_t * __restrict pglob)
153 {
154 	struct glob_limit limit = { 0, 0, 0, 0, 0 };
155 	const char *patnext;
156 	Char *bufnext, *bufend, patbuf[MAXPATHLEN], prot;
157 	mbstate_t mbs;
158 	wchar_t wc;
159 	size_t clen;
160 	int too_long;
161 
162 	patnext = pattern;
163 	if (!(flags & GLOB_APPEND)) {
164 		pglob->gl_pathc = 0;
165 		pglob->gl_pathv = NULL;
166 		if (!(flags & GLOB_DOOFFS))
167 			pglob->gl_offs = 0;
168 	}
169 	if (flags & GLOB_LIMIT) {
170 		limit.l_path_lim = pglob->gl_matchc;
171 		if (limit.l_path_lim == 0)
172 			limit.l_path_lim = GLOB_LIMIT_PATH;
173 	}
174 	pglob->gl_flags = flags & ~GLOB_MAGCHAR;
175 	pglob->gl_errfunc = errfunc;
176 	pglob->gl_matchc = 0;
177 
178 	bufnext = patbuf;
179 	bufend = bufnext + MAXPATHLEN - 1;
180 	too_long = 1;
181 	if (flags & GLOB_NOESCAPE) {
182 		memset(&mbs, 0, sizeof(mbs));
183 		while (bufnext <= bufend) {
184 			clen = mbrtowc(&wc, patnext, MB_LEN_MAX, &mbs);
185 			if (clen == (size_t)-1 || clen == (size_t)-2)
186 				return (err_nomatch(pglob, &limit, pattern));
187 			else if (clen == 0) {
188 				too_long = 0;
189 				break;
190 			}
191 			*bufnext++ = wc;
192 			patnext += clen;
193 		}
194 	} else {
195 		/* Protect the quoted characters. */
196 		memset(&mbs, 0, sizeof(mbs));
197 		while (bufnext <= bufend) {
198 			if (*patnext == '\\') {
199 				if (*++patnext == '\0') {
200 					*bufnext++ = QUOTE;
201 					continue;
202 				}
203 				prot = M_PROTECT;
204 			} else
205 				prot = 0;
206 			clen = mbrtowc(&wc, patnext, MB_LEN_MAX, &mbs);
207 			if (clen == (size_t)-1 || clen == (size_t)-2)
208 				return (err_nomatch(pglob, &limit, pattern));
209 			else if (clen == 0) {
210 				too_long = 0;
211 				break;
212 			}
213 			*bufnext++ = wc | prot;
214 			patnext += clen;
215 		}
216 	}
217 	if (too_long)
218 		return (err_nomatch(pglob, &limit, pattern));
219 	*bufnext = EOS;
220 
221 	if (flags & GLOB_BRACE)
222 	    return (globexp0(patbuf, pglob, &limit, pattern));
223 	else
224 	    return (glob0(patbuf, pglob, &limit, pattern));
225 }
226 
227 static int
228 globexp0(const Char *pattern, glob11_t *pglob, struct glob_limit *limit,
229     const char *origpat) {
230 	int rv;
231 	size_t oldpathc;
232 
233 	/* Protect a single {}, for find(1), like csh */
234 	if (pattern[0] == LBRACE && pattern[1] == RBRACE && pattern[2] == EOS) {
235 		if ((pglob->gl_flags & GLOB_LIMIT) &&
236 		    limit->l_brace_cnt++ >= GLOB_LIMIT_BRACE) {
237 			errno = E2BIG;
238 			return (GLOB_NOSPACE);
239 		}
240 		return (glob0(pattern, pglob, limit, origpat));
241 	}
242 
243 	oldpathc = pglob->gl_pathc;
244 
245 	if ((rv = globexp1(pattern, pglob, limit)) != 0)
246 		return rv;
247 
248 	return (globfinal(pglob, limit, oldpathc, origpat));
249 }
250 
251 /*
252  * Expand recursively a glob {} pattern. When there is no more expansion
253  * invoke the standard globbing routine to glob the rest of the magic
254  * characters
255  */
256 static int
257 globexp1(const Char *pattern, glob11_t *pglob, struct glob_limit *limit)
258 {
259 	const Char* ptr;
260 
261 	if ((ptr = g_strchr(pattern, LBRACE)) != NULL) {
262 		if ((pglob->gl_flags & GLOB_LIMIT) &&
263 		    limit->l_brace_cnt++ >= GLOB_LIMIT_BRACE) {
264 			errno = E2BIG;
265 			return (GLOB_NOSPACE);
266 		}
267 		return (globexp2(ptr, pattern, pglob, limit));
268 	}
269 
270 	return (glob0(pattern, pglob, limit, NULL));
271 }
272 
273 
274 /*
275  * Recursive brace globbing helper. Tries to expand a single brace.
276  * If it succeeds then it invokes globexp1 with the new pattern.
277  * If it fails then it tries to glob the rest of the pattern and returns.
278  */
279 static int
280 globexp2(const Char *ptr, const Char *pattern, glob11_t *pglob,
281     struct glob_limit *limit)
282 {
283 	int     i, rv;
284 	Char   *lm, *ls;
285 	const Char *pe, *pm, *pm1, *pl;
286 	Char    patbuf[MAXPATHLEN];
287 
288 	/* copy part up to the brace */
289 	for (lm = patbuf, pm = pattern; pm != ptr; *lm++ = *pm++)
290 		continue;
291 	*lm = EOS;
292 	ls = lm;
293 
294 	/* Find the balanced brace */
295 	for (i = 0, pe = ++ptr; *pe != EOS; pe++)
296 		if (*pe == LBRACKET) {
297 			/* Ignore everything between [] */
298 			for (pm = pe++; *pe != RBRACKET && *pe != EOS; pe++)
299 				continue;
300 			if (*pe == EOS) {
301 				/*
302 				 * We could not find a matching RBRACKET.
303 				 * Ignore and just look for RBRACE
304 				 */
305 				pe = pm;
306 			}
307 		}
308 		else if (*pe == LBRACE)
309 			i++;
310 		else if (*pe == RBRACE) {
311 			if (i == 0)
312 				break;
313 			i--;
314 		}
315 
316 	/* Non matching braces; just glob the pattern */
317 	if (i != 0 || *pe == EOS)
318 		return (glob0(pattern, pglob, limit, NULL));
319 
320 	for (i = 0, pl = pm = ptr; pm <= pe; pm++)
321 		switch (*pm) {
322 		case LBRACKET:
323 			/* Ignore everything between [] */
324 			for (pm1 = pm++; *pm != RBRACKET && *pm != EOS; pm++)
325 				continue;
326 			if (*pm == EOS) {
327 				/*
328 				 * We could not find a matching RBRACKET.
329 				 * Ignore and just look for RBRACE
330 				 */
331 				pm = pm1;
332 			}
333 			break;
334 
335 		case LBRACE:
336 			i++;
337 			break;
338 
339 		case RBRACE:
340 			if (i) {
341 			    i--;
342 			    break;
343 			}
344 			/* FALLTHROUGH */
345 		case COMMA:
346 			if (i && *pm == COMMA)
347 				break;
348 			else {
349 				/* Append the current string */
350 				for (lm = ls; (pl < pm); *lm++ = *pl++)
351 					continue;
352 				/*
353 				 * Append the rest of the pattern after the
354 				 * closing brace
355 				 */
356 				for (pl = pe + 1; (*lm++ = *pl++) != EOS;)
357 					continue;
358 
359 				/* Expand the current pattern */
360 #ifdef DEBUG
361 				qprintf("globexp2:", patbuf);
362 #endif
363 				rv = globexp1(patbuf, pglob, limit);
364 				if (rv)
365 					return (rv);
366 
367 				/* move after the comma, to the next string */
368 				pl = pm + 1;
369 			}
370 			break;
371 
372 		default:
373 			break;
374 		}
375 	return (0);
376 }
377 
378 
379 
380 /*
381  * expand tilde from the passwd file.
382  */
383 static const Char *
384 globtilde(const Char *pattern, Char *patbuf, size_t patbuf_len, glob11_t *pglob)
385 {
386 	struct passwd *pwd;
387 	char *h, *sc;
388 	const Char *p;
389 	Char *b, *eb;
390 	wchar_t wc;
391 	wchar_t wbuf[MAXPATHLEN];
392 	wchar_t *wbufend, *dc;
393 	size_t clen;
394 	mbstate_t mbs;
395 	int too_long;
396 
397 	if (*pattern != TILDE || !(pglob->gl_flags & GLOB_TILDE))
398 		return (pattern);
399 
400 	/*
401 	 * Copy up to the end of the string or /
402 	 */
403 	eb = &patbuf[patbuf_len - 1];
404 	for (p = pattern + 1, b = patbuf;
405 	    b < eb && *p != EOS && UNPROT(*p) != SEP; *b++ = *p++)
406 		continue;
407 
408 	if (*p != EOS && UNPROT(*p) != SEP)
409 		return (NULL);
410 
411 	*b = EOS;
412 	h = NULL;
413 
414 	if (patbuf[0] == EOS) {
415 		/*
416 		 * handle a plain ~ or ~/ by expanding $HOME first (iff
417 		 * we're not running setuid or setgid) and then trying
418 		 * the password file
419 		 */
420 		if ((h = secure_getenv("HOME")) == NULL) {
421 			if (((h = getlogin()) != NULL &&
422 			     (pwd = getpwnam(h)) != NULL) ||
423 			    (pwd = getpwuid(getuid())) != NULL)
424 				h = pwd->pw_dir;
425 			else
426 				return (pattern);
427 		}
428 	}
429 	else {
430 		/*
431 		 * Expand a ~user
432 		 */
433 		if (g_Ctoc(patbuf, (char *)wbuf, sizeof(wbuf)))
434 			return (NULL);
435 		if ((pwd = getpwnam((char *)wbuf)) == NULL)
436 			return (pattern);
437 		else
438 			h = pwd->pw_dir;
439 	}
440 
441 	/* Copy the home directory */
442 	dc = wbuf;
443 	sc = h;
444 	wbufend = wbuf + MAXPATHLEN - 1;
445 	too_long = 1;
446 	memset(&mbs, 0, sizeof(mbs));
447 	while (dc <= wbufend) {
448 		clen = mbrtowc(&wc, sc, MB_LEN_MAX, &mbs);
449 		if (clen == (size_t)-1 || clen == (size_t)-2) {
450 			/* XXX See initial comment #2. */
451 			wc = (unsigned char)*sc;
452 			clen = 1;
453 			memset(&mbs, 0, sizeof(mbs));
454 		}
455 		if ((*dc++ = wc) == EOS) {
456 			too_long = 0;
457 			break;
458 		}
459 		sc += clen;
460 	}
461 	if (too_long)
462 		return (NULL);
463 
464 	dc = wbuf;
465 	for (b = patbuf; b < eb && *dc != EOS; *b++ = *dc++ | M_PROTECT)
466 		continue;
467 	if (*dc != EOS)
468 		return (NULL);
469 
470 	/* Append the rest of the pattern */
471 	if (*p != EOS) {
472 		too_long = 1;
473 		while (b <= eb) {
474 			if ((*b++ = *p++) == EOS) {
475 				too_long = 0;
476 				break;
477 			}
478 		}
479 		if (too_long)
480 			return (NULL);
481 	} else
482 		*b = EOS;
483 
484 	return (patbuf);
485 }
486 
487 
488 /*
489  * The main glob() routine: compiles the pattern (optionally processing
490  * quotes), calls glob1() to do the real pattern matching, and finally
491  * sorts the list (unless unsorted operation is requested).  Returns 0
492  * if things went well, nonzero if errors occurred.
493  */
494 static int
495 glob0(const Char *pattern, glob11_t *pglob, struct glob_limit *limit,
496     const char *origpat) {
497 	const Char *qpatnext;
498 	int err;
499 	size_t oldpathc;
500 	Char *bufnext, c, patbuf[MAXPATHLEN];
501 
502 	qpatnext = globtilde(pattern, patbuf, MAXPATHLEN, pglob);
503 	if (qpatnext == NULL) {
504 		errno = E2BIG;
505 		return (GLOB_NOSPACE);
506 	}
507 	oldpathc = pglob->gl_pathc;
508 	bufnext = patbuf;
509 
510 	/* We don't need to check for buffer overflow any more. */
511 	while ((c = *qpatnext++) != EOS) {
512 		switch (c) {
513 		case LBRACKET:
514 			c = *qpatnext;
515 			if (c == NOT)
516 				++qpatnext;
517 			if (*qpatnext == EOS ||
518 			    g_strchr(qpatnext+1, RBRACKET) == NULL) {
519 				*bufnext++ = LBRACKET;
520 				if (c == NOT)
521 					--qpatnext;
522 				break;
523 			}
524 			*bufnext++ = M_SET;
525 			if (c == NOT)
526 				*bufnext++ = M_NOT;
527 			c = *qpatnext++;
528 			do {
529 				*bufnext++ = CHAR(c);
530 				if (*qpatnext == RANGE &&
531 				    (c = qpatnext[1]) != RBRACKET) {
532 					*bufnext++ = M_RNG;
533 					*bufnext++ = CHAR(c);
534 					qpatnext += 2;
535 				}
536 			} while ((c = *qpatnext++) != RBRACKET);
537 			pglob->gl_flags |= GLOB_MAGCHAR;
538 			*bufnext++ = M_END;
539 			break;
540 		case QUESTION:
541 			pglob->gl_flags |= GLOB_MAGCHAR;
542 			*bufnext++ = M_ONE;
543 			break;
544 		case STAR:
545 			pglob->gl_flags |= GLOB_MAGCHAR;
546 			/* collapse adjacent stars to one,
547 			 * to avoid exponential behavior
548 			 */
549 			if (bufnext == patbuf || bufnext[-1] != M_ALL)
550 			    *bufnext++ = M_ALL;
551 			break;
552 		default:
553 			*bufnext++ = CHAR(c);
554 			break;
555 		}
556 	}
557 	*bufnext = EOS;
558 #ifdef DEBUG
559 	qprintf("glob0:", patbuf);
560 #endif
561 
562 	if ((err = glob1(patbuf, pglob, limit)) != 0)
563 		return(err);
564 
565 	if (origpat != NULL)
566 		return (globfinal(pglob, limit, oldpathc, origpat));
567 
568 	return (0);
569 }
570 
571 static int
572 globfinal(glob11_t *pglob, struct glob_limit *limit, size_t oldpathc,
573     const char *origpat) {
574 	if (pglob->gl_pathc == oldpathc)
575 		return (err_nomatch(pglob, limit, origpat));
576 
577 	if (!(pglob->gl_flags & GLOB_NOSORT))
578 		qsort(pglob->gl_pathv + pglob->gl_offs + oldpathc,
579 		    pglob->gl_pathc - oldpathc, sizeof(char *), compare);
580 
581 	return (0);
582 }
583 
584 static int
585 compare(const void *p, const void *q)
586 {
587 	return (strcoll(*(char **)p, *(char **)q));
588 }
589 
590 static int
591 glob1(Char *pattern, glob11_t *pglob, struct glob_limit *limit)
592 {
593 	Char pathbuf[MAXPATHLEN];
594 
595 	/* A null pathname is invalid -- POSIX 1003.1 sect. 2.4. */
596 	if (*pattern == EOS)
597 		return (0);
598 	return (glob2(pathbuf, pathbuf, pathbuf + MAXPATHLEN - 1,
599 	    pattern, pglob, limit));
600 }
601 
602 /*
603  * The functions glob2 and glob3 are mutually recursive; there is one level
604  * of recursion for each segment in the pattern that contains one or more
605  * meta characters.
606  */
607 static int
608 glob2(Char *pathbuf, Char *pathend, Char *pathend_last, Char *pattern,
609       glob11_t *pglob, struct glob_limit *limit)
610 {
611 	struct freebsd11_stat sb;
612 	Char *p, *q;
613 	int anymeta;
614 
615 	/*
616 	 * Loop over pattern segments until end of pattern or until
617 	 * segment with meta character found.
618 	 */
619 	for (anymeta = 0;;) {
620 		if (*pattern == EOS) {		/* End of pattern? */
621 			*pathend = EOS;
622 			if (g_lstat(pathbuf, &sb, pglob))
623 				return (0);
624 
625 			if ((pglob->gl_flags & GLOB_LIMIT) &&
626 			    limit->l_stat_cnt++ >= GLOB_LIMIT_STAT) {
627 				errno = E2BIG;
628 				return (GLOB_NOSPACE);
629 			}
630 			if ((pglob->gl_flags & GLOB_MARK) &&
631 			    UNPROT(pathend[-1]) != SEP &&
632 			    (S_ISDIR(sb.st_mode) ||
633 			    (S_ISLNK(sb.st_mode) &&
634 			    g_stat(pathbuf, &sb, pglob) == 0 &&
635 			    S_ISDIR(sb.st_mode)))) {
636 				if (pathend + 1 > pathend_last) {
637 					errno = E2BIG;
638 					return (GLOB_NOSPACE);
639 				}
640 				*pathend++ = SEP;
641 				*pathend = EOS;
642 			}
643 			++pglob->gl_matchc;
644 			return (globextend(pathbuf, pglob, limit, NULL));
645 		}
646 
647 		/* Find end of next segment, copy tentatively to pathend. */
648 		q = pathend;
649 		p = pattern;
650 		while (*p != EOS && UNPROT(*p) != SEP) {
651 			if (ismeta(*p))
652 				anymeta = 1;
653 			if (q + 1 > pathend_last) {
654 				errno = E2BIG;
655 				return (GLOB_NOSPACE);
656 			}
657 			*q++ = *p++;
658 		}
659 
660 		if (!anymeta) {		/* No expansion, do next segment. */
661 			pathend = q;
662 			pattern = p;
663 			while (UNPROT(*pattern) == SEP) {
664 				if (pathend + 1 > pathend_last) {
665 					errno = E2BIG;
666 					return (GLOB_NOSPACE);
667 				}
668 				*pathend++ = *pattern++;
669 			}
670 		} else			/* Need expansion, recurse. */
671 			return (glob3(pathbuf, pathend, pathend_last, pattern,
672 			    p, pglob, limit));
673 	}
674 	/* NOTREACHED */
675 }
676 
677 static int
678 glob3(Char *pathbuf, Char *pathend, Char *pathend_last,
679       Char *pattern, Char *restpattern,
680       glob11_t *pglob, struct glob_limit *limit)
681 {
682 	struct freebsd11_dirent *dp;
683 	DIR *dirp;
684 	int err, too_long, saverrno, saverrno2;
685 	char buf[MAXPATHLEN + MB_LEN_MAX - 1];
686 
687 	struct freebsd11_dirent *(*readdirfunc)(DIR *);
688 
689 	if (pathend > pathend_last) {
690 		errno = E2BIG;
691 		return (GLOB_NOSPACE);
692 	}
693 	*pathend = EOS;
694 	if (pglob->gl_errfunc != NULL &&
695 	    g_Ctoc(pathbuf, buf, sizeof(buf))) {
696 		errno = E2BIG;
697 		return (GLOB_NOSPACE);
698 	}
699 
700 	saverrno = errno;
701 	errno = 0;
702 	if ((dirp = g_opendir(pathbuf, pglob)) == NULL) {
703 		if (errno == ENOENT || errno == ENOTDIR)
704 			return (0);
705 		err = err_aborted(pglob, errno, buf);
706 		if (errno == 0)
707 			errno = saverrno;
708 		return (err);
709 	}
710 
711 	err = 0;
712 
713 	/* pglob->gl_readdir takes a void *, fix this manually */
714 	if (pglob->gl_flags & GLOB_ALTDIRFUNC)
715 		readdirfunc =
716 		    (struct freebsd11_dirent *(*)(DIR *))pglob->gl_readdir;
717 	else
718 		readdirfunc = freebsd11_readdir;
719 
720 	errno = 0;
721 	/* Search directory for matching names. */
722 	while ((dp = (*readdirfunc)(dirp)) != NULL) {
723 		char *sc;
724 		Char *dc;
725 		wchar_t wc;
726 		size_t clen;
727 		mbstate_t mbs;
728 
729 		if ((pglob->gl_flags & GLOB_LIMIT) &&
730 		    limit->l_readdir_cnt++ >= GLOB_LIMIT_READDIR) {
731 			errno = E2BIG;
732 			err = GLOB_NOSPACE;
733 			break;
734 		}
735 
736 		/* Initial DOT must be matched literally. */
737 		if (dp->d_name[0] == '.' && UNPROT(*pattern) != DOT) {
738 			errno = 0;
739 			continue;
740 		}
741 		memset(&mbs, 0, sizeof(mbs));
742 		dc = pathend;
743 		sc = dp->d_name;
744 		too_long = 1;
745 		while (dc <= pathend_last) {
746 			clen = mbrtowc(&wc, sc, MB_LEN_MAX, &mbs);
747 			if (clen == (size_t)-1 || clen == (size_t)-2) {
748 				/* XXX See initial comment #2. */
749 				wc = (unsigned char)*sc;
750 				clen = 1;
751 				memset(&mbs, 0, sizeof(mbs));
752 			}
753 			if ((*dc++ = wc) == EOS) {
754 				too_long = 0;
755 				break;
756 			}
757 			sc += clen;
758 		}
759 		if (too_long && (err = err_aborted(pglob, ENAMETOOLONG,
760 		    buf))) {
761 			errno = ENAMETOOLONG;
762 			break;
763 		}
764 		if (too_long || !match(pathend, pattern, restpattern)) {
765 			*pathend = EOS;
766 			errno = 0;
767 			continue;
768 		}
769 		if (errno == 0)
770 			errno = saverrno;
771 		err = glob2(pathbuf, --dc, pathend_last, restpattern,
772 		    pglob, limit);
773 		if (err)
774 			break;
775 		errno = 0;
776 	}
777 
778 	saverrno2 = errno;
779 	if (pglob->gl_flags & GLOB_ALTDIRFUNC)
780 		(*pglob->gl_closedir)(dirp);
781 	else
782 		closedir(dirp);
783 	errno = saverrno2;
784 
785 	if (err)
786 		return (err);
787 
788 	if (dp == NULL && errno != 0 &&
789 	    (err = err_aborted(pglob, errno, buf)))
790 		return (err);
791 
792 	if (errno == 0)
793 		errno = saverrno;
794 	return (0);
795 }
796 
797 
798 /*
799  * Extend the gl_pathv member of a glob11_t structure to accommodate a new item,
800  * add the new item, and update gl_pathc.
801  *
802  * This assumes the BSD realloc, which only copies the block when its size
803  * crosses a power-of-two boundary; for v7 realloc, this would cause quadratic
804  * behavior.
805  *
806  * Return 0 if new item added, error code if memory couldn't be allocated.
807  *
808  * Invariant of the glob11_t structure:
809  *	Either gl_pathc is zero and gl_pathv is NULL; or gl_pathc > 0 and
810  *	gl_pathv points to (gl_offs + gl_pathc + 1) items.
811  */
812 static int
813 globextend(const Char *path, glob11_t *pglob, struct glob_limit *limit,
814     const char *origpat)
815 {
816 	char **pathv;
817 	size_t i, newn, len;
818 	char *copy;
819 	const Char *p;
820 
821 	if ((pglob->gl_flags & GLOB_LIMIT) &&
822 	    pglob->gl_matchc > limit->l_path_lim) {
823 		errno = E2BIG;
824 		return (GLOB_NOSPACE);
825 	}
826 
827 	newn = 2 + pglob->gl_pathc + pglob->gl_offs;
828 	/* reallocarray(NULL, newn, size) is equivalent to malloc(newn*size). */
829 	pathv = reallocarray(pglob->gl_pathv, newn, sizeof(*pathv));
830 	if (pathv == NULL)
831 		return (GLOB_NOSPACE);
832 
833 	if (pglob->gl_pathv == NULL && pglob->gl_offs > 0) {
834 		/* first time around -- clear initial gl_offs items */
835 		pathv += pglob->gl_offs;
836 		for (i = pglob->gl_offs + 1; --i > 0; )
837 			*--pathv = NULL;
838 	}
839 	pglob->gl_pathv = pathv;
840 
841 	if (origpat != NULL)
842 		copy = strdup(origpat);
843 	else {
844 		for (p = path; *p++ != EOS;)
845 			continue;
846 		len = MB_CUR_MAX * (size_t)(p - path); /* XXX overallocation */
847 		if ((copy = malloc(len)) != NULL) {
848 			if (g_Ctoc(path, copy, len)) {
849 				free(copy);
850 				errno = E2BIG;
851 				return (GLOB_NOSPACE);
852 			}
853 		}
854 	}
855 	if (copy != NULL) {
856 		limit->l_string_cnt += strlen(copy) + 1;
857 		if ((pglob->gl_flags & GLOB_LIMIT) &&
858 		    limit->l_string_cnt >= GLOB_LIMIT_STRING) {
859 			free(copy);
860 			errno = E2BIG;
861 			return (GLOB_NOSPACE);
862 		}
863 		pathv[pglob->gl_offs + pglob->gl_pathc++] = copy;
864 	}
865 	pathv[pglob->gl_offs + pglob->gl_pathc] = NULL;
866 	return (copy == NULL ? GLOB_NOSPACE : 0);
867 }
868 
869 /*
870  * pattern matching function for filenames.
871  */
872 static int
873 match(Char *name, Char *pat, Char *patend)
874 {
875 	int ok, negate_range;
876 	Char c, k, *nextp, *nextn;
877 	struct xlocale_collate *table =
878 		(struct xlocale_collate*)__get_locale()->components[XLC_COLLATE];
879 
880 	nextn = NULL;
881 	nextp = NULL;
882 
883 	while (1) {
884 		while (pat < patend) {
885 			c = *pat++;
886 			switch (c & M_MASK) {
887 			case M_ALL:
888 				if (pat == patend)
889 					return (1);
890 				if (*name == EOS)
891 					return (0);
892 				nextn = name + 1;
893 				nextp = pat - 1;
894 				break;
895 			case M_ONE:
896 				if (*name++ == EOS)
897 					goto fail;
898 				break;
899 			case M_SET:
900 				ok = 0;
901 				if ((k = *name++) == EOS)
902 					goto fail;
903 				negate_range = ((*pat & M_MASK) == M_NOT);
904 				if (negate_range != 0)
905 					++pat;
906 				while (((c = *pat++) & M_MASK) != M_END)
907 					if ((*pat & M_MASK) == M_RNG) {
908 						if (table->__collate_load_error ?
909 						    CHAR(c) <= CHAR(k) &&
910 						    CHAR(k) <= CHAR(pat[1]) :
911 						    __wcollate_range_cmp(CHAR(c),
912 						    CHAR(k)) <= 0 &&
913 						    __wcollate_range_cmp(CHAR(k),
914 						    CHAR(pat[1])) <= 0)
915 							ok = 1;
916 						pat += 2;
917 					} else if (c == k)
918 						ok = 1;
919 				if (ok == negate_range)
920 					goto fail;
921 				break;
922 			default:
923 				if (*name++ != c)
924 					goto fail;
925 				break;
926 			}
927 		}
928 		if (*name == EOS)
929 			return (1);
930 
931 	fail:
932 		if (nextn == NULL)
933 			break;
934 		pat = nextp;
935 		name = nextn;
936 	}
937 	return (0);
938 }
939 
940 /* Free allocated data belonging to a glob11_t structure. */
941 void
942 freebsd11_globfree(glob11_t *pglob)
943 {
944 	size_t i;
945 	char **pp;
946 
947 	if (pglob->gl_pathv != NULL) {
948 		pp = pglob->gl_pathv + pglob->gl_offs;
949 		for (i = pglob->gl_pathc; i--; ++pp)
950 			if (*pp)
951 				free(*pp);
952 		free(pglob->gl_pathv);
953 		pglob->gl_pathv = NULL;
954 	}
955 }
956 
957 static DIR *
958 g_opendir(Char *str, glob11_t *pglob)
959 {
960 	char buf[MAXPATHLEN + MB_LEN_MAX - 1];
961 
962 	if (*str == EOS)
963 		strcpy(buf, ".");
964 	else {
965 		if (g_Ctoc(str, buf, sizeof(buf))) {
966 			errno = ENAMETOOLONG;
967 			return (NULL);
968 		}
969 	}
970 
971 	if (pglob->gl_flags & GLOB_ALTDIRFUNC)
972 		return ((*pglob->gl_opendir)(buf));
973 
974 	return (opendir(buf));
975 }
976 
977 static int
978 g_lstat(Char *fn, struct freebsd11_stat *sb, glob11_t *pglob)
979 {
980 	char buf[MAXPATHLEN + MB_LEN_MAX - 1];
981 
982 	if (g_Ctoc(fn, buf, sizeof(buf))) {
983 		errno = ENAMETOOLONG;
984 		return (-1);
985 	}
986 	if (pglob->gl_flags & GLOB_ALTDIRFUNC)
987 		return((*pglob->gl_lstat)(buf, sb));
988 	return (freebsd11_lstat(buf, sb));
989 }
990 
991 static int
992 g_stat(Char *fn, struct freebsd11_stat *sb, glob11_t *pglob)
993 {
994 	char buf[MAXPATHLEN + MB_LEN_MAX - 1];
995 
996 	if (g_Ctoc(fn, buf, sizeof(buf))) {
997 		errno = ENAMETOOLONG;
998 		return (-1);
999 	}
1000 	if (pglob->gl_flags & GLOB_ALTDIRFUNC)
1001 		return ((*pglob->gl_stat)(buf, sb));
1002 	return (freebsd11_stat(buf, sb));
1003 }
1004 
1005 static const Char *
1006 g_strchr(const Char *str, wchar_t ch)
1007 {
1008 
1009 	do {
1010 		if (*str == ch)
1011 			return (str);
1012 	} while (*str++);
1013 	return (NULL);
1014 }
1015 
1016 static int
1017 g_Ctoc(const Char *str, char *buf, size_t len)
1018 {
1019 	mbstate_t mbs;
1020 	size_t clen;
1021 
1022 	memset(&mbs, 0, sizeof(mbs));
1023 	while (len >= MB_CUR_MAX) {
1024 		clen = wcrtomb(buf, CHAR(*str), &mbs);
1025 		if (clen == (size_t)-1) {
1026 			/* XXX See initial comment #2. */
1027 			*buf = (char)CHAR(*str);
1028 			clen = 1;
1029 			memset(&mbs, 0, sizeof(mbs));
1030 		}
1031 		if (CHAR(*str) == EOS)
1032 			return (0);
1033 		str++;
1034 		buf += clen;
1035 		len -= clen;
1036 	}
1037 	return (1);
1038 }
1039 
1040 static int
1041 err_nomatch(glob11_t *pglob, struct glob_limit *limit, const char *origpat) {
1042 	/*
1043 	 * If there was no match we are going to append the origpat
1044 	 * if GLOB_NOCHECK was specified or if GLOB_NOMAGIC was specified
1045 	 * and the origpat did not contain any magic characters
1046 	 * GLOB_NOMAGIC is there just for compatibility with csh.
1047 	 */
1048 	if ((pglob->gl_flags & GLOB_NOCHECK) ||
1049 	    ((pglob->gl_flags & GLOB_NOMAGIC) &&
1050 	    !(pglob->gl_flags & GLOB_MAGCHAR)))
1051 		return (globextend(NULL, pglob, limit, origpat));
1052 	return (GLOB_NOMATCH);
1053 }
1054 
1055 static int
1056 err_aborted(glob11_t *pglob, int err, char *buf) {
1057 	if ((pglob->gl_errfunc != NULL && pglob->gl_errfunc(buf, err)) ||
1058 	    (pglob->gl_flags & GLOB_ERR))
1059 		return (GLOB_ABORTED);
1060 	return (0);
1061 }
1062 
1063 #ifdef DEBUG
1064 static void
1065 qprintf(const char *str, Char *s)
1066 {
1067 	Char *p;
1068 
1069 	(void)printf("%s\n", str);
1070 	if (s != NULL) {
1071 		for (p = s; *p != EOS; p++)
1072 			(void)printf("%c", (char)CHAR(*p));
1073 		(void)printf("\n");
1074 		for (p = s; *p != EOS; p++)
1075 			(void)printf("%c", (isprot(*p) ? '\\' : ' '));
1076 		(void)printf("\n");
1077 		for (p = s; *p != EOS; p++)
1078 			(void)printf("%c", (ismeta(*p) ? '_' : ' '));
1079 		(void)printf("\n");
1080 	}
1081 }
1082 #endif
1083 
1084 __sym_compat(glob, freebsd11_glob, FBSD_1.0);
1085 __sym_compat(globfree, freebsd11_globfree, FBSD_1.0);
1086