Fix buffer overruns with long language strings
Resolves: RHEL-169755
This commit is contained in:
parent
2b9f6dd2df
commit
5edc9f4666
@ -0,0 +1,97 @@
|
||||
From c0e4c08533e3e899d7e1cc0d9885e83ea1e1bdfb Mon Sep 17 00:00:00 2001
|
||||
From: Dave Cantrell <dcantrell@redhat.com>
|
||||
Date: Mon, 20 Apr 2026 15:05:04 -0400
|
||||
Subject: [PATCH] Prevent buffer overruns in findPreambleTag() for language
|
||||
string
|
||||
MIME-Version: 1.0
|
||||
Content-Type: text/plain; charset=UTF-8
|
||||
Content-Transfer-Encoding: 8bit
|
||||
|
||||
This is technically possible and there is a reproducer for it, but I
|
||||
would not consider this a critical problem. If you have a really long
|
||||
language identifier string in a preamble tag and it's larger than
|
||||
BUFSIZ on the platform, you get a SIGSEGV. Not a surprise.
|
||||
|
||||
In spec files you can have stuff like this in the preamble:
|
||||
|
||||
Summary: Here is a short summary
|
||||
Summary(de): Hier ist eine kurze Zusammenfassung
|
||||
Summary(eo): Jen mallonga resumo
|
||||
Summary(Elvish): Sí na- a estent summarui
|
||||
Summary(ga_ie): Is coimriú ghearr é seo
|
||||
Summary(Klingon): naDev 'oH ngaj summary
|
||||
|
||||
And findPreambleTag() is eventually called to pick up those language
|
||||
identifiers in parens. Usually the identifiers are two characters,
|
||||
but sometimes they are three or four. The parser in the library scans
|
||||
a character at a time until it hits the closing paren and just stuffs
|
||||
it all in the 'lang' buffer. The problem is that buffer is BUFSIZ and
|
||||
there is no bounds checking to see if the string in the spec file in
|
||||
parens is larger than what BUFSIZ can hold. So it is technically
|
||||
possible to provide a spec file with a completely useless huge string
|
||||
as a language identifier that then crashes the spec file parser.
|
||||
|
||||
This patch adds some bounds checking to that reading loop to prevent
|
||||
this incredibly rare yet technically possible issue. I don't bother
|
||||
growing the buffer if we're still reading characters because honestly
|
||||
if we have more than BUFSIZ in parens, we've got a garbage spec file.
|
||||
|
||||
The patch does ensure the unused space in BUFSIZ is NULL and that the
|
||||
lang buffer is NULL terminated so it's moderately useful in later
|
||||
parts of the code.
|
||||
|
||||
Signed-off-by: Dave Cantrell <dcantrell@redhat.com>
|
||||
(backported from commit b6f2cc52fcc7eedf0095c2bf0495e4b6aafd87cb)
|
||||
---
|
||||
build/parsePreamble.c | 12 ++++++++----
|
||||
1 file changed, 8 insertions(+), 4 deletions(-)
|
||||
|
||||
diff --git a/build/parsePreamble.c b/build/parsePreamble.c
|
||||
index 3693746f8..e14664939 100644
|
||||
--- a/build/parsePreamble.c
|
||||
+++ b/build/parsePreamble.c
|
||||
@@ -1064,10 +1064,11 @@ static struct PreambleRec_s const preambleList[] = {
|
||||
/**
|
||||
*/
|
||||
static int findPreambleTag(rpmSpec spec,rpmTagVal * tag,
|
||||
- const char ** macro, char * lang)
|
||||
+ const char ** macro, char * lang, size_t langsize)
|
||||
{
|
||||
PreambleRec p;
|
||||
char *s;
|
||||
+ size_t l = 0;
|
||||
|
||||
for (p = preambleList; p->token != NULL; p++) {
|
||||
if (!(p->token && !rstrncasecmp(spec->line, p->token, p->len)))
|
||||
@@ -1097,14 +1098,17 @@ static int findPreambleTag(rpmSpec spec,rpmTagVal * tag,
|
||||
case 2:
|
||||
if (*s == ':') {
|
||||
/* Type 1 is multilang, 2 is qualifiers with no defaults */
|
||||
- strcpy(lang, (p->type == 1) ? RPMBUILD_DEFAULT_LANG : "");
|
||||
+ rstrlcpy(lang, (p->type == 1) ? RPMBUILD_DEFAULT_LANG : "", langsize);
|
||||
+ l = strlen(lang);
|
||||
break;
|
||||
}
|
||||
if (*s != '(') return 1;
|
||||
s++;
|
||||
SKIPSPACE(s);
|
||||
- while (!risspace(*s) && *s != ')')
|
||||
+ while (!risspace(*s) && *s != ')' && l < (langsize - 1)) {
|
||||
*lang++ = *s++;
|
||||
+ l++;
|
||||
+ }
|
||||
*lang = '\0';
|
||||
SKIPSPACE(s);
|
||||
if (*s != ')') return 1;
|
||||
@@ -1173,7 +1177,7 @@ int parsePreamble(rpmSpec spec, int initialPackage)
|
||||
linep = spec->line;
|
||||
SKIPSPACE(linep);
|
||||
if (*linep != '\0') {
|
||||
- if (findPreambleTag(spec, &tag, ¯o, lang)) {
|
||||
+ if (findPreambleTag(spec, &tag, ¯o, lang, sizeof(lang))) {
|
||||
if (spec->lineNum == 1 &&
|
||||
(unsigned char)(spec->line[0]) == 0xed &&
|
||||
(unsigned char)(spec->line[1]) == 0xab &&
|
||||
--
|
||||
2.54.0
|
||||
|
||||
1
rpm.spec
1
rpm.spec
@ -673,6 +673,7 @@ fi
|
||||
- Add support for %%autosetup -C (RHEL-141269)
|
||||
- Add support for database parking (RHEL-126405)
|
||||
- Make syslog plugin actually useful and also required (RHEL-155272)
|
||||
- Fix buffer overruns with long language strings (RHEL-169755)
|
||||
|
||||
* Thu Feb 05 2026 Michal Domonkos <mdomonko@redhat.com> - 4.19.1.1-23
|
||||
- Fix key import API to return NOTTRUSTED for disabled algorithms (RHEL-112394)
|
||||
|
||||
Loading…
Reference in New Issue
Block a user