NetBSD-Bugs archive

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Old Index]

lib/61021: libc/regex: out-of-bounds read in backref() ONWBND at slice boundaries



>Number:         61021
>Category:       lib
>Synopsis:       libc/regex: out-of-bounds read in backref() ONWBND at slice boundaries
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    lib-bug-people
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Mon Nov 16 23:30:00 +0000 2026
>Originator:     Nick Leppelmeier
>Release:        NetBSD-current (src/lib/libc/regex/engine.c rev 1.30)
>Organization:
>Environment:
NetBSD-current src/lib/libc/regex (engine.c 1.30, regcomp.c 1.50, regexec.c 1.26); reproduced standalone on Linux 6.6.114 x86_64 (GCC 15.2) and Android Bionic (aarch64/x86_64)
>Description:
In lib/libc/regex/engine.c, the ONWBND (\B) handling inside backref() reads
out of bounds when matching at the start or end of a subject slice:

	if (((sp == m->beginp) && !ISWORD(*sp)) ||
	    (sp == m->endp && !ISWORD(*(sp - 1))))
		{ /* yes, beginning/end of subject */ }
	else if (ISWORD(*(sp - 1)) == ISWORD(*sp))
		{ /* yes, beginning/end of subject */ }

1. When the subject is empty (sp == m->beginp && sp == m->endp), ISWORD(*sp)
   reads 1 byte past m->endp.
2. When sp == m->beginp and *sp is a word character (for example, matching
   "()\1(\B.|a)" against "a"), the first condition evaluates to false and
   execution falls through to ISWORD(*(sp - 1)), reading 1 byte before
   m->beginp.
3. Similarly, when sp == m->endp and *(sp - 1) is a word character (for
   example, matching "()\1(a\B|a)" against "a"), execution falls through to
   ISWORD(*sp), reading 1 byte past m->endp.

If the subject buffer sits directly adjacent to an unmapped or PROT_NONE
guard page (such as with REG_STARTEND), regexec() crashes with SIGSEGV.
>How-To-Repeat:
/* Compile and run on NetBSD: cc -Wall -O2 repro.c -o repro && ./repro */
#include <sys/mman.h>
#include <regex.h>
#include <stdio.h>
#include <string.h>
#include <unistd.h>

int
main(void)
{
	long pgsz = sysconf(_SC_PAGESIZE);
	char *buf = mmap(NULL, (size_t)pgsz * 3, PROT_NONE,
	    MAP_PRIVATE | MAP_ANON, -1, 0);
	if (buf == MAP_FAILED)
		return 1;

	char *page = buf + pgsz;
	if (mprotect(page, (size_t)pgsz, PROT_READ | PROT_WRITE) != 0)
		return 1;

	/* Place "a" right at the start of the mapped page after PROT_NONE */
	page[0] = 'a';
	page[1] = '\0';

	regex_t re;
	regmatch_t pm = { .rm_so = 0, .rm_eo = 1 };
	if (regcomp(&re, "()\\1(\\B.|a)", REG_EXTENDED | REG_GNU) != 0)
		return 1;

	/* Triggers read of *(m->beginp - 1) -> SIGSEGV */
	int rc = regexec(&re, page, 1, &pm, REG_STARTEND);
	printf("rc=%d\n", rc);
	regfree(&re);
	return 0;
}
>Fix:
Index: lib/libc/regex/engine.c
===================================================================
--- lib/libc/regex/engine.c
+++ lib/libc/regex/engine.c
@@ -732,11 +732,20 @@ backref(
 			return(NULL);
 		break;
 	case ONWBND:
-		if (((sp == m->beginp) && !ISWORD(*sp)) ||
-		    (sp == m->endp && !ISWORD(*(sp - 1))))
-			{ /* yes, beginning/end of subject */ }
-		else if (ISWORD(*(sp - 1)) == ISWORD(*sp))
-			{ /* yes, beginning/end of subject */ }
+		if (sp == m->beginp && sp == m->endp)
+			{ /* yes, empty subject has no word chars */ }
+		else if (sp == m->beginp) {
+			if (!ISWORD(*sp))
+				{ /* yes, beginning of subject */ }
+			else
+				return(NULL);
+		} else if (sp == m->endp) {
+			if (!ISWORD(*(sp - 1)))
+				{ /* yes, end of subject */ }
+			else
+				return(NULL);
+		} else if (ISWORD(*(sp - 1)) == ISWORD(*sp))
+			{ /* yes, neither or both are word chars */ }
 		else
 			return(NULL);
 		break;




Home | Main Index | Thread Index | Old Index