Skip to contentSkip to conversation

vsprintf: ignore arguments to %n

PATCH In review Kees Cook 3 participants 6 messages Jan 28, 2014

Subsystems (3)
lore
Share & tools
lore

Kees Cook

author
#0 lore raw

PATCH 1/1 vsprintf: ignore arguments to %n

If arguments are consumed without output when encountering %n, it could be used to benefit or improve information leak attacks that were exposed via a limited size buffer. Since %n is not used by the kernel, there is no reason to make an info leak attack any easier.

---
 lib/vsprintf.c |   10 +++-------
 1 file changed, 3 insertions(+), 7 deletions(-)
lib/vsprintf.c +3 -7
index 185b6d300ebc..9d5c48b705f9 100644
@@ -1735,14 +1735,10 @@ int vsnprintf(char *buf, size_t size, const char *fmt, va_list args)
 		case FORMAT_TYPE_NRCHARS: {
 			/*
 			 * Since %n poses a greater security risk than
-			 * utility, ignore %n and skip its argument.
+			 * utility, it should not be implemented. Instead,
+			 * when encountering %n, ignore the arguments.
 			 */
-			void *skip_arg;
-
-			WARN_ONCE(1, "Please remove ignored %%n in '%s'\n",
-					old_fmt);
-
-			skip_arg = va_arg(args, void *);
+			WARN_ONCE(1, "Ignored %%n in '%s'\n", old_fmt);
 			break;
 		}
 
Signature · 6 lines
1.7.9.5


-- 
Kees Cook
Chrome OS Security
maintainer
#1 lore raw

lib/vsprintf.c

Quoted 1 line · Kees Cook
lib/vsprintf.c +0 -0

[]

Quoted 7 lines
lib/vsprintf.c +2 -1
@@ -1735,14 +1735,10 @@ int vsnprintf(char *buf, size_t size, const char *fmt, va_list args)
 		case FORMAT_TYPE_NRCHARS: {
 			/*
 			 * Since %n poses a greater security risk than
-			 * utility, ignore %n and skip its argument.
+			 * utility, it should not be implemented. Instead,
+			 * when encountering %n, ignore the arguments.

lib/vsprintf.c int vsnprintf(char *buf, size_t size, const char *fmt, va_list args) · line 1743

Quoted 10 lines
lib/vsprintf.c +1 -6
 			 */
-			void *skip_arg;
-
-			WARN_ONCE(1, "Please remove ignored %%n in '%s'\n",
-					old_fmt);
-
-			skip_arg = va_arg(args, void *);
+			WARN_ONCE(1, "Ignored %%n in '%s'\n", old_fmt);
 			break;
 		}

I don't think this is better either.

Why not tell people to fix the code?
checkpatch isn't run on quite a lot of patches.

anyway, here's a possible checkpatch test too.

---
 scripts/checkpatch.pl | 9 +++++++++
 1 file changed, 9 insertions(+)
scripts/checkpatch.pl +9 -0
index 1dbd6d1..95c3264 100755
@@ -4069,6 +4069,15 @@ sub process {
 			}
 		}
 
+# check for unsupported uses of %n in what appear to be formats
+		if ($sline =~ /\b$logFunctions\s*\(.*"/) {
+			my $fmt = get_quoted_string($line, $rawline);
+			if ($fmt ne "" && $fmt =~ /[^\%]\%n/) {
+				WARN("PRINTF_CONTROL_N",
+				     "printf style use of \%n is not supported\n" . $herecurr);
+			}
+		}
+
 # Check for misused memsets
 		if ($^V && $^V ge 5.10.0 &&
 		    defined $stat &&
#2 lore raw
In reply to Kees Cook
Quoted 4 lines · Kees Cook

If arguments are consumed without output when encountering %n, it could be used to benefit or improve information leak attacks that were exposed via a limited size buffer. Since %n is not used by the kernel, there is no reason to make an info leak attack any easier.

I was thinking more like the following. Print the warning if %n is detected in format_decode(), but otherwise just remove the handling of %n outright and treat it like any other invalid format specifier. Something like this completely untested patch. Thoughts?

~Ryan

---

lib/vsprintf.c +9 -36
index 10909c5..4e24009 100644
@@ -364,7 +364,6 @@ enum format_type {
 	FORMAT_TYPE_SHORT,
 	FORMAT_TYPE_UINT,
 	FORMAT_TYPE_INT,
-	FORMAT_TYPE_NRCHARS,
 	FORMAT_TYPE_SIZE_T,
 	FORMAT_TYPE_PTRDIFF
 };
@@ -1512,10 +1511,6 @@ qualifier:
 		return fmt - start;
 		/* skip alnum */
 
-	case 'n':
-		spec->type = FORMAT_TYPE_NRCHARS;
-		return ++fmt - start;
-
 	case '%':
 		spec->type = FORMAT_TYPE_PERCENT_CHAR;
 		return ++fmt - start;
@@ -1538,6 +1533,15 @@ qualifier:
 	case 'u':
 		break;
 
+	case 'n':
+		/*
+		 * Since %n poses a greater security risk than utility, treat
+		 * it as an invalid format specifier. Warn about it use, so
+		 * new instances don't get added.
+		 */
+		WARN_ONCE(1, "Please remove ignored %%n in '%s'\n", fmt);
+		/* Fall-through */
+
 	default:
 		spec->type = FORMAT_TYPE_INVALID;
 		return fmt - start;
@@ -1711,20 +1715,6 @@ int vsnprintf(char *buf, size_t size, const char *fmt, va_list args)
 			++str;
 			break;
 
-		case FORMAT_TYPE_NRCHARS: {
-			/*
-			 * Since %n poses a greater security risk than
-			 * utility, ignore %n and skip its argument.
-			 */
-			void *skip_arg;
-
-			WARN_ONCE(1, "Please remove ignored %%n in '%s'\n",
-					old_fmt);
-
-			skip_arg = va_arg(args, void *);
-			break;
-		}
-
 		default:
 			switch (spec.type) {
 			case FORMAT_TYPE_LONG_LONG:
@@ -1999,19 +1989,6 @@ do {									\
 				fmt++;
 			break;
 
-		case FORMAT_TYPE_NRCHARS: {
-			/* skip %n 's argument */
-			u8 qualifier = spec.qualifier;
-			void *skip_arg;
-			if (qualifier == 'l')
-				skip_arg = va_arg(args, long *);
-			else if (_tolower(qualifier) == 'z')
-				skip_arg = va_arg(args, size_t *);
-			else
-				skip_arg = va_arg(args, int *);
-			break;
-		}
-
 		default:
 			switch (spec.type) {
 
@@ -2170,10 +2147,6 @@ int bstr_printf(char *buf, size_t size, const char *fmt, const u32 *bin_buf)
 			++str;
 			break;
 
-		case FORMAT_TYPE_NRCHARS:
-			/* skip */
-			break;
-
 		default: {
 			unsigned long long num;
 
author Acked-by
#3 lore raw
Quoted 10 lines · Ryan Mallon
Quoted 4 lines · Kees Cook

If arguments are consumed without output when encountering %n, it could be used to benefit or improve information leak attacks that were exposed via a limited size buffer. Since %n is not used by the kernel, there is no reason to make an info leak attack any easier.

I was thinking more like the following. Print the warning if %n is detected in format_decode(), but otherwise just remove the handling of %n outright and treat it like any other invalid format specifier. Something like this completely untested patch. Thoughts?

I'd be totally fine with it. Minor typo in the comment before the WARN_ONCE (should be "its" instead of "it"), but otherwise looks good. Consider it:

It builds and boots fine for me, FWIW.

-Kees

Signature · 2 lines
Kees Cook
Chrome OS Security
#4 lore raw
Quoted 16 lines · Kees Cook
Quoted 10 lines · Ryan Mallon
Quoted 4 lines · Kees Cook

If arguments are consumed without output when encountering %n, it could be used to benefit or improve information leak attacks that were exposed via a limited size buffer. Since %n is not used by the kernel, there is no reason to make an info leak attack any easier.

I was thinking more like the following. Print the warning if %n is detected in format_decode(), but otherwise just remove the handling of %n outright and treat it like any other invalid format specifier. Something like this completely untested patch. Thoughts?

I'd be totally fine with it. Minor typo in the comment before the WARN_ONCE (should be "its" instead of "it"), but otherwise looks good. Consider it:

Quoted 6 lines

It builds and boots fine for me, FWIW.

-Kees

It looks like your second version already got added to Andrew's mm tree. I'm happy to repost mine with a fixed typo and proper signed-off by if you'd rather use that version.

~Ryan

author
#5 lore raw
Quoted 22 lines · Ryan Mallon
Quoted 20 lines · Kees Cook
Quoted 10 lines · Ryan Mallon
Quoted 4 lines · Kees Cook

If arguments are consumed without output when encountering %n, it could be used to benefit or improve information leak attacks that were exposed via a limited size buffer. Since %n is not used by the kernel, there is no reason to make an info leak attack any easier.

I was thinking more like the following. Print the warning if %n is detected in format_decode(), but otherwise just remove the handling of %n outright and treat it like any other invalid format specifier. Something like this completely untested patch. Thoughts?

I'd be totally fine with it. Minor typo in the comment before the WARN_ONCE (should be "its" instead of "it"), but otherwise looks good. Consider it:

It builds and boots fine for me, FWIW.

Quoted 7 lines
Quoted 2 lines

-Kees

It looks like your second version already got added to Andrew's mm tree. I'm happy to repost mine with a fixed typo and proper signed-off by if you'd rather use that version.

I think yours is much cleaner: it entirely removes the %n processing logic.

-Kees

Signature · 2 lines
Kees Cook
Chrome OS Security