Fix strtod integer/buffer overflow (bug 14459).

ChangeLog

@@ -1,3 +1,20 @@
+2012-08-27  Joseph Myers  <joseph@codesourcery.com>
+
+	[BZ #14459]
+	* stdlib/strtod_l.c: Include <stdint.h>.
+	(NDEBUG): Do not define.
+	(round_and_return): Change EXPONENT parameter to type intmax_t.
+	Rearrange calculations to avoid internal overflow possibilities.
+	(str_to_mpn): Change EXPONENT parameter to type intmax_t *.
+	Rearrange calculations to avoid internal overflow possibilities.
+	Assert that number fits inside MPNSIZE limbs.
+	(____STRTOF_INTERNAL): Change EXPONENT variable to type intmax_t.
+	Change DIG_NO, INT_NO and LEAD_ZERO to type size_t.  Rearrange
+	calculations and add assertions to avoid internal overflow
+	possibilities.  Add casts to avoid signed/unsigned operations.
+	* stdlib/tst-strtod-overflow.c: New file.
+	* stdlib/Makefile (tests): Add tst-strtod-overflow.
+
 2012-08-25  Marek Polacek  <polacek@redhat.com>
 
 	* time/time.h: Fix some typos in comments.

NEWS

@@ -11,7 +11,7 @@ Version 2.17
 
   6778, 6808, 9685, 11607, 13717, 13696, 13939, 14042, 14090, 14166, 14150,
   14151, 14154, 14157, 14166, 14173, 14195, 14252, 14283, 14298, 14303,
-  14307, 14328, 14331, 14336, 14337, 14347, 14349, 14505
+  14307, 14328, 14331, 14336, 14337, 14347, 14349, 14459, 14505
 
 * Support for STT_GNU_IFUNC symbols added for s390 and s390x.
   Optimized versions of memcpy, memset, and memcmp added for System z10 and

stdlib/Makefile

@@ -69,7 +69,7 @@ tests		:= tst-strtol tst-strtod testmb testrand testsort testdiv   \
 		   tst-makecontext tst-strtod4 tst-strtod5 tst-qsort2	    \
 		   tst-makecontext2 tst-strtod6 tst-unsetenv1		    \
 		   tst-makecontext3 bug-getcontext bug-fmtmsg1		    \
-		   tst-secure-getenv
+		   tst-secure-getenv tst-strtod-overflow
 tests-static	:= tst-secure-getenv
 
 include ../Makeconfig

stdlib/strtod_l.c

@@ -60,6 +60,7 @@ extern unsigned long long int ____strtoull_l_internal (const char *, char **,
 #include <math.h>
 #include <stdlib.h>
 #include <string.h>
+#include <stdint.h>
 
 /* The gmp headers need some configuration frobs.  */
 #define HAVE_ALLOCA 1
@@ -72,7 +73,6 @@ extern unsigned long long int ____strtoull_l_internal (const char *, char **,
 #include "longlong.h"
 #include "fpioconst.h"
 
-#define NDEBUG 1
 #include <assert.h>
 
 
@@ -174,19 +174,19 @@ extern const mp_limb_t _tens_in_limb[MAX_DIG_PER_LIMB + 1];
 /* Return a floating point number of the needed type according to the given
    multi-precision number after possible rounding.  */
 static FLOAT
-round_and_return (mp_limb_t *retval, int exponent, int negative,
+round_and_return (mp_limb_t *retval, intmax_t exponent, int negative,
 		  mp_limb_t round_limb, mp_size_t round_bit, int more_bits)
 {
   if (exponent < MIN_EXP - 1)
     {
-      mp_size_t shift = MIN_EXP - 1 - exponent;
+      if (exponent < MIN_EXP - 1 - MANT_DIG)
-
-      if (shift > MANT_DIG)
 	{
 	  __set_errno (ERANGE);
 	  return 0.0;
 	}
 
+      mp_size_t shift = MIN_EXP - 1 - exponent;
+
       more_bits |= (round_limb & ((((mp_limb_t) 1) << round_bit) - 1)) != 0;
       if (shift == MANT_DIG)
 	/* This is a special case to handle the very seldom case where
@@ -233,6 +233,9 @@ round_and_return (mp_limb_t *retval, int exponent, int negative,
       __set_errno (ERANGE);
     }
 
+  if (exponent > MAX_EXP)
+    goto overflow;
+
   if ((round_limb & (((mp_limb_t) 1) << round_bit)) != 0
       && (more_bits || (retval[0] & 1) != 0
 	  || (round_limb & ((((mp_limb_t) 1) << round_bit) - 1)) != 0))
@@ -258,6 +261,7 @@ round_and_return (mp_limb_t *retval, int exponent, int negative,
     }
 
   if (exponent > MAX_EXP)
+  overflow:
     return negative ? -FLOAT_HUGE_VAL : FLOAT_HUGE_VAL;
 
   return MPN2FLOAT (retval, exponent, negative);
@@ -271,7 +275,7 @@ round_and_return (mp_limb_t *retval, int exponent, int negative,
    factor for the resulting number (see code) multiply by it.  */
 static const STRING_TYPE *
 str_to_mpn (const STRING_TYPE *str, int digcnt, mp_limb_t *n, mp_size_t *nsize,
-	    int *exponent
+	    intmax_t *exponent
 #ifndef USE_WIDE_CHAR
 	    , const char *decimal, size_t decimal_len, const char *thousands
 #endif
@@ -301,6 +305,7 @@ str_to_mpn (const STRING_TYPE *str, int digcnt, mp_limb_t *n, mp_size_t *nsize,
 	      cy += __mpn_add_1 (n, n, *nsize, low);
 	      if (cy != 0)
 		{
+		  assert (*nsize < MPNSIZE);
 		  n[*nsize] = cy;
 		  ++(*nsize);
 		}
@@ -335,7 +340,7 @@ str_to_mpn (const STRING_TYPE *str, int digcnt, mp_limb_t *n, mp_size_t *nsize,
     }
   while (--digcnt > 0);
 
-  if (*exponent > 0 && cnt + *exponent <= MAX_DIG_PER_LIMB)
+  if (*exponent > 0 && *exponent <= MAX_DIG_PER_LIMB - cnt)
     {
       low *= _tens_in_limb[*exponent];
       start = _tens_in_limb[cnt + *exponent];
@@ -355,8 +360,11 @@ str_to_mpn (const STRING_TYPE *str, int digcnt, mp_limb_t *n, mp_size_t *nsize,
       cy = __mpn_mul_1 (n, n, *nsize, start);
       cy += __mpn_add_1 (n, n, *nsize, low);
       if (cy != 0)
+	{
+	  assert (*nsize < MPNSIZE);
 	  n[(*nsize)++] = cy;
 	}
+    }
 
   return str;
 }
@@ -413,7 +421,7 @@ ____STRTOF_INTERNAL (nptr, endptr, group, loc)
 {
   int negative;			/* The sign of the number.  */
   MPN_VAR (num);		/* MP representation of the number.  */
-  int exponent;			/* Exponent of the number.  */
+  intmax_t exponent;		/* Exponent of the number.  */
 
   /* Numbers starting `0X' or `0x' have to be processed with base 16.  */
   int base = 10;
@@ -435,7 +443,7 @@ ____STRTOF_INTERNAL (nptr, endptr, group, loc)
   /* Points at the character following the integer and fractional digits.  */
   const STRING_TYPE *expp;
   /* Total number of digit and number of digits in integer part.  */
-  int dig_no, int_no, lead_zero;
+  size_t dig_no, int_no, lead_zero;
   /* Contains the last character read.  */
   CHAR_TYPE c;
 
@@ -767,7 +775,7 @@ ____STRTOF_INTERNAL (nptr, endptr, group, loc)
      are all or any is really a fractional digit will be decided
      later.  */
   int_no = dig_no;
-  lead_zero = int_no == 0 ? -1 : 0;
+  lead_zero = int_no == 0 ? (size_t) -1 : 0;
 
   /* Read the fractional digits.  A special case are the 'american
      style' numbers like `16.' i.e. with decimal point but without
@@ -789,12 +797,13 @@ ____STRTOF_INTERNAL (nptr, endptr, group, loc)
 	     (base == 16 && ({ CHAR_TYPE lo = TOLOWER (c);
 			       lo >= L_('a') && lo <= L_('f'); })))
 	{
-	  if (c != L_('0') && lead_zero == -1)
+	  if (c != L_('0') && lead_zero == (size_t) -1)
 	    lead_zero = dig_no - int_no;
 	  ++dig_no;
 	  c = *++cp;
 	}
     }
+  assert (dig_no <= (uintmax_t) INTMAX_MAX);
 
   /* Remember start of exponent (if any).  */
   expp = cp;
@@ -817,24 +826,80 @@ ____STRTOF_INTERNAL (nptr, endptr, group, loc)
 
       if (c >= L_('0') && c <= L_('9'))
 	{
-	  int exp_limit;
+	  intmax_t exp_limit;
 
 	  /* Get the exponent limit. */
 	  if (base == 16)
-	    exp_limit = (exp_negative ?
+	    {
-			 -MIN_EXP + MANT_DIG + 4 * int_no :
+	      if (exp_negative)
-			 MAX_EXP - 4 * int_no + 4 * lead_zero + 3);
+		{
+		  assert (int_no <= (uintmax_t) (INTMAX_MAX
+						 + MIN_EXP - MANT_DIG) / 4);
+		  exp_limit = -MIN_EXP + MANT_DIG + 4 * (intmax_t) int_no;
+		}
 	      else
-	    exp_limit = (exp_negative ?
+		{
-			 -MIN_10_EXP + MANT_DIG + int_no :
+		  if (int_no)
-			 MAX_10_EXP - int_no + lead_zero + 1);
+		    {
+		      assert (lead_zero == 0
+			      && int_no <= (uintmax_t) INTMAX_MAX / 4);
+		      exp_limit = MAX_EXP - 4 * (intmax_t) int_no + 3;
+		    }
+		  else if (lead_zero == (size_t) -1)
+		    {
+		      /* The number is zero and this limit is
+			 arbitrary.  */
+		      exp_limit = MAX_EXP + 3;
+		    }
+		  else
+		    {
+		      assert (lead_zero
+			      <= (uintmax_t) (INTMAX_MAX - MAX_EXP - 3) / 4);
+		      exp_limit = (MAX_EXP
+				   + 4 * (intmax_t) lead_zero
+				   + 3);
+		    }
+		}
+	    }
+	  else
+	    {
+	      if (exp_negative)
+		{
+		  assert (int_no
+			  <= (uintmax_t) (INTMAX_MAX + MIN_10_EXP - MANT_DIG));
+		  exp_limit = -MIN_10_EXP + MANT_DIG + (intmax_t) int_no;
+		}
+	      else
+		{
+		  if (int_no)
+		    {
+		      assert (lead_zero == 0
+			      && int_no <= (uintmax_t) INTMAX_MAX);
+		      exp_limit = MAX_10_EXP - (intmax_t) int_no + 1;
+		    }
+		  else if (lead_zero == (size_t) -1)
+		    {
+		      /* The number is zero and this limit is
+			 arbitrary.  */
+		      exp_limit = MAX_10_EXP + 1;
+		    }
+		  else
+		    {
+		      assert (lead_zero
+			      <= (uintmax_t) (INTMAX_MAX - MAX_10_EXP - 1));
+		      exp_limit = MAX_10_EXP + (intmax_t) lead_zero + 1;
+		    }
+		}
+	    }
+
+	  if (exp_limit < 0)
+	    exp_limit = 0;
 
 	  do
 	    {
-	      exponent *= 10;
+	      if (__builtin_expect ((exponent > exp_limit / 10
-	      exponent += c - L_('0');
+				     || (exponent == exp_limit / 10
-
+					 && c - L_('0') > exp_limit % 10)), 0))
-	      if (__builtin_expect (exponent > exp_limit, 0))
 		/* The exponent is too large/small to represent a valid
 		   number.  */
 		{
@@ -843,7 +908,7 @@ ____STRTOF_INTERNAL (nptr, endptr, group, loc)
 		  /* We have to take care for special situation: a joker
 		     might have written "0.0e100000" which is in fact
 		     zero.  */
-		  if (lead_zero == -1)
+		  if (lead_zero == (size_t) -1)
 		    result = negative ? -0.0 : 0.0;
 		  else
 		    {
@@ -862,6 +927,9 @@ ____STRTOF_INTERNAL (nptr, endptr, group, loc)
 		  /* NOTREACHED */
 		}
 
+	      exponent *= 10;
+	      exponent += c - L_('0');
+
 	      c = *++cp;
 	    }
 	  while (c >= L_('0') && c <= L_('9'));
@@ -930,7 +998,14 @@ ____STRTOF_INTERNAL (nptr, endptr, group, loc)
 	}
 #endif
       startp += lead_zero + decimal_len;
-      exponent -= base == 16 ? 4 * lead_zero : lead_zero;
+      assert (lead_zero <= (base == 16
+			    ? (uintmax_t) INTMAX_MAX / 4
+			    : (uintmax_t) INTMAX_MAX));
+      assert (lead_zero <= (base == 16
+			    ? ((uintmax_t) exponent
+			       - (uintmax_t) INTMAX_MIN) / 4
+			    : ((uintmax_t) exponent - (uintmax_t) INTMAX_MIN)));
+      exponent -= base == 16 ? 4 * (intmax_t) lead_zero : (intmax_t) lead_zero;
       dig_no -= lead_zero;
     }
 
@@ -972,7 +1047,10 @@ ____STRTOF_INTERNAL (nptr, endptr, group, loc)
 	}
 
       /* Adjust the exponent for the bits we are shifting in.  */
-      exponent += bits - 1 + (int_no - 1) * 4;
+      assert (int_no <= (uintmax_t) (exponent < 0
+				     ? (INTMAX_MAX - bits + 1) / 4
+				     : (INTMAX_MAX - exponent - bits + 1) / 4));
+      exponent += bits - 1 + ((intmax_t) int_no - 1) * 4;
 
       while (--dig_no > 0 && idx >= 0)
 	{
@@ -1024,13 +1102,15 @@ ____STRTOF_INTERNAL (nptr, endptr, group, loc)
      really integer digits or belong to the fractional part; i.e. we normalize
      123e-2 to 1.23.  */
   {
-    register int incr = (exponent < 0 ? MAX (-int_no, exponent)
+    register intmax_t incr = (exponent < 0
-			 : MIN (dig_no - int_no, exponent));
+			      ? MAX (-(intmax_t) int_no, exponent)
+			      : MIN ((intmax_t) dig_no - (intmax_t) int_no,
+				     exponent));
     int_no += incr;
     exponent -= incr;
   }
 
-  if (__builtin_expect (int_no + exponent > MAX_10_EXP + 1, 0))
+  if (__builtin_expect (exponent > MAX_10_EXP + 1 - (intmax_t) int_no, 0))
     {
       __set_errno (ERANGE);
       return negative ? -FLOAT_HUGE_VAL : FLOAT_HUGE_VAL;
@@ -1215,7 +1295,7 @@ ____STRTOF_INTERNAL (nptr, endptr, group, loc)
        digits we should have enough bits for the result.  The remaining
        decimal digits give us the information that more bits are following.
        This can be used while rounding.  (Two added as a safety margin.)  */
-    if (dig_no - int_no > (MANT_DIG - bits + 2) / 3 + 2)
+    if ((intmax_t) dig_no > (intmax_t) int_no + (MANT_DIG - bits + 2) / 3 + 2)
       {
 	dig_no = int_no + (MANT_DIG - bits + 2) / 3 + 2;
 	more_bits = 1;
@@ -1223,7 +1303,7 @@ ____STRTOF_INTERNAL (nptr, endptr, group, loc)
     else
       more_bits = 0;
 
-    neg_exp = dig_no - int_no - exponent;
+    neg_exp = (intmax_t) dig_no - (intmax_t) int_no - exponent;
 
     /* Construct the denominator.  */
     densize = 0;

stdlib/tst-strtod-overflow.c

@@ -0,0 +1,48 @@
+/* Test for integer/buffer overflow in strtod.
+   Copyright (C) 2012 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
+   <http://www.gnu.org/licenses/>.  */
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+
+#define EXPONENT "e-2147483649"
+#define SIZE 214748364
+
+static int
+do_test (void)
+{
+  char *p = malloc (1 + SIZE + sizeof (EXPONENT));
+  if (p == NULL)
+    {
+      puts ("malloc failed, cannot test for overflow");
+      return 0;
+    }
+  p[0] = '1';
+  memset (p + 1, '0', SIZE);
+  memcpy (p + 1 + SIZE, EXPONENT, sizeof (EXPONENT));
+  double d = strtod (p, NULL);
+  if (d != 0)
+    {
+      printf ("strtod returned wrong value: %a\n", d);
+      return 1;
+    }
+  return 0;
+}
+
+#define TEST_FUNCTION do_test ()
+#include "../test-skeleton.c"