Avoid ldbl-96 stack corruption from range reduction of pseudo-zero (bug 25487) [committed]
Commit Message
Bug 25487 reports stack corruption in ldbl-96 sinl on a pseudo-zero
argument (an representation where all the significand bits, including
the explicit high bit, are zero, but the exponent is not zero, which
is not a valid representation for the long double type).
Although this is not a valid long double representation, existing
practice in this area (see bug 4586, originally marked invalid but
subsequently fixed) is that we still seek to avoid invalid memory
accesses as a result, in case of programs that treat arbitrary binary
data as long double representations, although the invalid
representations of the ldbl-96 format do not need to be consistently
handled the same as any particular valid representation.
This patch makes the range reduction detect pseudo-zero and unnormal
representations that would otherwise go to __kernel_rem_pio2, and
returns a NaN for them instead of continuing with the range reduction
process. (Pseudo-zero and unnormal representations whose unbiased
exponent is less than -1 have already been safely returned from the
function before this point without going through the rest of range
reduction.) Pseudo-zero representations would previously result in
the value passed to __kernel_rem_pio2 being all-zero, which is
definitely unsafe; unnormal representations would previously result in
a value passed whose high bit is zero, which might well be unsafe
since that is not a form of input expected by __kernel_rem_pio2.
Tested for x86_64.
---
Committed.
Comments
* Joseph Myers:
> Bug 25487 reports stack corruption in ldbl-96 sinl on a pseudo-zero
> argument (an representation where all the significand bits, including
> diff --git a/sysdeps/ieee754/ldbl-96/Makefile b/sysdeps/ieee754/ldbl-96/Makefile
> index 995e90d6da..318628aed6 100644
> --- a/sysdeps/ieee754/ldbl-96/Makefile
> +++ b/sysdeps/ieee754/ldbl-96/Makefile
> @@ -17,5 +17,6 @@
> # <https://www.gnu.org/licenses/>.
>
> ifeq ($(subdir),math)
> -tests += test-canonical-ldbl-96 test-totalorderl-ldbl-96
> +tests += test-canonical-ldbl-96 test-totalorderl-ldbl-96 test-sinl-pseudo
> +CFLAGS-test-sinl-pseudo.c += -fstack-protector-all
> endif
> diff --git a/sysdeps/ieee754/ldbl-96/e_rem_pio2l.c b/sysdeps/ieee754/ldbl-96/e_rem_pio2l.c
Thanks for fixing this. Unfortunately, the test doesn't build on ia64
because GCC does not support -fstack-protector-all there:
cc1: error: ‘-fstack-protector’ not supported for this target [-Werror]
cc1: all warnings being treated as errors
make[3]: *** [/mnt/scratch1/fw/bmg/default/build/glibcs/ia64-linux-gnu/glibc/sysd-rules:639: /mnt/scratch1/fw/bmg/default/build/glibcs/ia64-linux-gnu/glibc/math/test-sinl-pseudo.o] Error 1
On Thu, 13 Feb 2020, Florian Weimer wrote:
> Thanks for fixing this. Unfortunately, the test doesn't build on ia64
> because GCC does not support -fstack-protector-all there:
>
> cc1: error: ‘-fstack-protector’ not supported for this target [-Werror]
> cc1: all warnings being treated as errors
> make[3]: *** [/mnt/scratch1/fw/bmg/default/build/glibcs/ia64-linux-gnu/glibc/sysd-rules:639: /mnt/scratch1/fw/bmg/default/build/glibcs/ia64-linux-gnu/glibc/math/test-sinl-pseudo.o] Error 1
It looks like ifeq ($(have-ssp),yes) will be needed around that CFLAGS
setting. (I took the unconditional CFLAGS setting from where such a
setting is used in debug/Makefile, but I see that in fact the addition of
the test there is conditional. In this case, the test is appropriate
unconditionally, it's just the use of -fstack-protector-all that needs to
be conditional.)
@@ -17,5 +17,6 @@
# <https://www.gnu.org/licenses/>.
ifeq ($(subdir),math)
-tests += test-canonical-ldbl-96 test-totalorderl-ldbl-96
+tests += test-canonical-ldbl-96 test-totalorderl-ldbl-96 test-sinl-pseudo
+CFLAGS-test-sinl-pseudo.c += -fstack-protector-all
endif
@@ -210,6 +210,18 @@ __ieee754_rem_pio2l (long double x, long double *y)
return 0;
}
+ if ((i0 & 0x80000000) == 0)
+ {
+ /* Pseudo-zero and unnormal representations are not valid
+ representations of long double. We need to avoid stack
+ corruption in __kernel_rem_pio2, which expects input in a
+ particular normal form, but those representations do not need
+ to be consistently handled like any particular floating-point
+ value. */
+ y[1] = y[0] = __builtin_nanl ("");
+ return 0;
+ }
+
/* Split the 64 bits of the mantissa into three 24-bit integers
stored in a double array. */
exp = j0 - 23;
new file mode 100644
@@ -0,0 +1,41 @@
+/* Test sinl for pseudo-zeros and unnormals for ldbl-96 (bug 25487).
+ Copyright (C) 2020 Free Software Foundation, Inc.
+ This file is part of the GNU C Library.
+
+ The GNU C Library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 2.1 of the License, or (at your option) any later version.
+
+ The GNU C Library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with the GNU C Library; if not, see
+ <https://www.gnu.org/licenses/>. */
+
+#include <math.h>
+#include <math_ldbl.h>
+#include <stdint.h>
+
+static int
+do_test (void)
+{
+ for (int i = 0; i < 64; i++)
+ {
+ uint64_t sig = i == 63 ? 0 : 1ULL << i;
+ long double ld;
+ SET_LDOUBLE_WORDS (ld, 0x4141,
+ sig >> 32, sig & 0xffffffffULL);
+ /* The requirement is that no stack overflow occurs when the
+ pseudo-zero or unnormal goes through range reduction. */
+ volatile long double ldr;
+ ldr = sinl (ld);
+ (void) ldr;
+ }
+ return 0;
+}
+
+#include <support/test-driver.c>