Fix nan functions handling of payload strings (bug 16961, bug 16962).

The nan, nanf and nanl functions handle payload strings by doing e.g.:

  if (tagp[0] != '\0')
    {
      char buf[6 + strlen (tagp)];
      sprintf (buf, "NAN(%s)", tagp);
      return strtod (buf, NULL);
    }

This is an unbounded stack allocation based on the length of the
argument.  Furthermore, if the argument starts with an n-char-sequence
followed by ')', that n-char-sequence is wrongly treated as
significant for determining the payload of the resulting NaN, when ISO
C says the call should be equivalent to strtod ("NAN", NULL), without
being affected by that initial n-char-sequence.  This patch fixes both
those problems by using the __strtod_nan etc. functions recently
factored out of strtod etc. for that purpose, with those functions
being exported from libc at version GLIBC_PRIVATE.

Tested for x86_64, x86, mips64 and powerpc.

	[BZ #16961]
	[BZ #16962]
	* math/s_nan.c (__nan): Use __strtod_nan instead of constructing a
	string on the stack for strtod.
	* math/s_nanf.c (__nanf): Use __strtof_nan instead of constructing
	a string on the stack for strtof.
	* math/s_nanl.c (__nanl): Use __strtold_nan instead of
	constructing a string on the stack for strtold.
	* stdlib/Versions (libc): Add __strtof_nan, __strtod_nan and
	__strtold_nan to GLIBC_PRIVATE.
	* math/test-nan-overflow.c: New file.
	* math/test-nan-payload.c: Likewise.
	* math/Makefile (tests): Add test-nan-overflow and
	test-nan-payload.

1 files changed, 1 insertions, 8 deletions

diff --git a/math/s_nanl.c b/math/s_nanl.c
index 73387a9116..4e1fc806fd 100644
--- a/math/s_nanl.c
+++ b/math/s_nanl.c
@@ -28,13 +28,6 @@
 long double
 __nanl (const char *tagp)
 {
-  if (tagp[0] != '\0')
-    {
-      char buf[6 + strlen (tagp)];
-      sprintf (buf, "NAN(%s)", tagp);
-      return strtold (buf, NULL);
-    }
-
-  return NAN;
+  return __strtold_nan (tagp, NULL, 0);
 }
 weak_alias (__nanl, nanl)