Fix modification of string literal by swprintf

Fix modification of string literal by swprintf

[Thomas Preudhomme]

[-- Attachment #1: Type: text/plain, Size: 622 bytes --]

Don't over-read memory returned by _DTOA_R, and never write to it
since the result might be a string literal.

For example, when doing:
   swprintf(tt, 20, L"%.*f", 6, 0.0);

we will get back "0".

Instead, write the result returned by _DTOA_R to the output buffer.
After this, write the 0 chars directly to the the output buffer
(if there are any). This also has the (marginal) advantage that
we read/write less memory overall.

The patch, contributed by Silviu Baranga was tested against libcxx testsuite and 
showed no regression. Please find the patch in git format-patch format in 
attachment.

Best regards,

Thomas

[-- Attachment #2: 0001-Don-t-overread-or-write-memory-returned-by-_DTOA_R.patch --]
[-- Type: text/x-patch, Size: 2767 bytes --]

From 7a31cfb01a0b089daf2bed93b742b6edbf4cba0c Mon Sep 17 00:00:00 2001
From: Silviu Baranga <silviu.baranga@arm.cm>
Date: Mon, 5 Jun 2017 09:54:42 +0100
Subject: [PATCH] Don't overread or write memory returned by _DTOA_R

Don't over-read memory returned by _DTOA_R, and never write to it
since the result might be a string literal.

For example, when doing:
  swprintf(tt, 20, L"%.*f", 6, 0.0);

we will get back "0".

Instead, write the result returned by _DTOA_R to the output buffer.
After this, write the 0 chars directly to the the output buffer
(if there are any). This also has the (marginal) advantage that
we read/write less memory overall.
---
 newlib/libc/locale/setlocale.h |  2 +-
 newlib/libc/stdio/vfwprintf.c  | 24 ++++++++++++------------
 2 files changed, 13 insertions(+), 13 deletions(-)

diff --git a/newlib/libc/locale/setlocale.h b/newlib/libc/locale/setlocale.h
index 85a38d5..1440d0e 100644
--- a/newlib/libc/locale/setlocale.h
+++ b/newlib/libc/locale/setlocale.h
@@ -227,7 +227,7 @@ __get_locale_r (struct _reent *r)
 _ELIDABLE_INLINE struct __locale_t *
 __get_current_locale (void)
 {
-  return _REENT->_locale ?: __get_global_locale ();
+  return _REENT->_locale ?: __get_global_locale (); // version for !_MB_CAPABLE?
 }
 
 /* Only access fixed "C" locale using this function.  Fake for !_MB_CAPABLE
diff --git a/newlib/libc/stdio/vfwprintf.c b/newlib/libc/stdio/vfwprintf.c
index f0179a0..1bec9b2 100644
--- a/newlib/libc/stdio/vfwprintf.c
+++ b/newlib/libc/stdio/vfwprintf.c
@@ -1627,13 +1627,20 @@ wcvt(struct _reent *data, _PRINTF_FLOAT_TYPE value, int ndigits, int flags,
 
 	{
 	  char *digits, *bp, *rve;
-#ifndef _MB_CAPABLE
 	  int i;
-#endif
 
 	  digits = _DTOA_R (data, value, mode, ndigits, decpt, &dsgn, &rve);
 
+#ifdef _MB_CAPABLE
+	  _mbsnrtowcs_r (data, buf, (const char **) &digits, rve - digits,
+			 len, NULL);
+#else
+	  for (i = 0; i < rve - digits && i < len; ++i)
+	    buf[i] = (wchar_t) digits[i];
+#endif
+
 	  if ((ch != L'g' && ch != L'G') || flags & ALT) {	/* Print trailing zeros */
+		char *padding = rve;
 		bp = digits + ndigits;
 		if (ch == L'f' || ch == L'F') {
 			if (*digits == L'0' && value)
@@ -1642,18 +1649,11 @@ wcvt(struct _reent *data, _PRINTF_FLOAT_TYPE value, int ndigits, int flags,
 		}
 		if (value == 0)	/* kludge for __dtoa irregularity */
 			rve = bp;
-		while (rve < bp)
-			*rve++ = '0';
-	  }
 
+		for (i = padding - digits; i < rve - digits && i < len; ++i)
+			buf[i] = L'0';
+	  }
 	  *length = rve - digits; /* full length of the string */
-#ifdef _MB_CAPABLE
-	  _mbsnrtowcs_r (data, buf, (const char **) &digits, *length,
-			 len, NULL);
-#else
-	  for (i = 0; i < *length && i < len; ++i)
-	    buf[i] = (wchar_t) digits[i];
-#endif
 	  return buf;
 	}
 }
-- 
1.9.1

Re: Fix modification of string literal by swprintf

[Corinna Vinschen]

[-- Attachment #1: Type: text/plain, Size: 4037 bytes --]

On Jun  5 09:59, Thomas Preudhomme wrote:
> Don't over-read memory returned by _DTOA_R, and never write to it
> since the result might be a string literal.
> 
> For example, when doing:
>   swprintf(tt, 20, L"%.*f", 6, 0.0);
> 
> we will get back "0".
> 
> Instead, write the result returned by _DTOA_R to the output buffer.
> After this, write the 0 chars directly to the the output buffer
> (if there are any). This also has the (marginal) advantage that
> we read/write less memory overall.
> 
> The patch, contributed by Silviu Baranga was tested against libcxx testsuite
> and showed no regression. Please find the patch in git format-patch format
> in attachment.
> 
> Best regards,
> 
> Thomas

> >From 7a31cfb01a0b089daf2bed93b742b6edbf4cba0c Mon Sep 17 00:00:00 2001
> From: Silviu Baranga <silviu.baranga@arm.cm>
> Date: Mon, 5 Jun 2017 09:54:42 +0100
> Subject: [PATCH] Don't overread or write memory returned by _DTOA_R
> 
> Don't over-read memory returned by _DTOA_R, and never write to it
> since the result might be a string literal.
> 
> For example, when doing:
>   swprintf(tt, 20, L"%.*f", 6, 0.0);
> 
> we will get back "0".
> 
> Instead, write the result returned by _DTOA_R to the output buffer.
> After this, write the 0 chars directly to the the output buffer
> (if there are any). This also has the (marginal) advantage that
> we read/write less memory overall.
> ---
>  newlib/libc/locale/setlocale.h |  2 +-
>  newlib/libc/stdio/vfwprintf.c  | 24 ++++++++++++------------
>  2 files changed, 13 insertions(+), 13 deletions(-)
> 
> diff --git a/newlib/libc/locale/setlocale.h b/newlib/libc/locale/setlocale.h
> index 85a38d5..1440d0e 100644
> --- a/newlib/libc/locale/setlocale.h
> +++ b/newlib/libc/locale/setlocale.h
> @@ -227,7 +227,7 @@ __get_locale_r (struct _reent *r)
>  _ELIDABLE_INLINE struct __locale_t *
>  __get_current_locale (void)
>  {
> -  return _REENT->_locale ?: __get_global_locale ();
> +  return _REENT->_locale ?: __get_global_locale (); // version for !_MB_CAPABLE?
>  }

This doesn't belong here.  Also, the code is already fine for !_MB_CAPABLE
as well.

>  /* Only access fixed "C" locale using this function.  Fake for !_MB_CAPABLE
> diff --git a/newlib/libc/stdio/vfwprintf.c b/newlib/libc/stdio/vfwprintf.c
> index f0179a0..1bec9b2 100644
> --- a/newlib/libc/stdio/vfwprintf.c
> +++ b/newlib/libc/stdio/vfwprintf.c
> @@ -1627,13 +1627,20 @@ wcvt(struct _reent *data, _PRINTF_FLOAT_TYPE value, int ndigits, int flags,
>  
>  	{
>  	  char *digits, *bp, *rve;
> -#ifndef _MB_CAPABLE
>  	  int i;
> -#endif
>  
>  	  digits = _DTOA_R (data, value, mode, ndigits, decpt, &dsgn, &rve);
>  
> +#ifdef _MB_CAPABLE
> +	  _mbsnrtowcs_r (data, buf, (const char **) &digits, rve - digits,
> +			 len, NULL);
> +#else
> +	  for (i = 0; i < rve - digits && i < len; ++i)
> +	    buf[i] = (wchar_t) digits[i];
> +#endif
> +
>  	  if ((ch != L'g' && ch != L'G') || flags & ALT) {	/* Print trailing zeros */
> +		char *padding = rve;
>  		bp = digits + ndigits;
>  		if (ch == L'f' || ch == L'F') {
>  			if (*digits == L'0' && value)
> @@ -1642,18 +1649,11 @@ wcvt(struct _reent *data, _PRINTF_FLOAT_TYPE value, int ndigits, int flags,
>  		}
>  		if (value == 0)	/* kludge for __dtoa irregularity */
>  			rve = bp;
> -		while (rve < bp)
> -			*rve++ = '0';
> -	  }
>  
> +		for (i = padding - digits; i < rve - digits && i < len; ++i)
> +			buf[i] = L'0';

Appending zeros here without incrementing rve...

> +	  }
>  	  *length = rve - digits; /* full length of the string */

...leads to incorrect setting of *length here.  Or am I missing
something?

> -#ifdef _MB_CAPABLE
> -	  _mbsnrtowcs_r (data, buf, (const char **) &digits, *length,
> -			 len, NULL);
> -#else
> -	  for (i = 0; i < *length && i < len; ++i)
> -	    buf[i] = (wchar_t) digits[i];
> -#endif
>  	  return buf;
>  	}
>  }
> -- 
> 1.9.1
> 

Corinna

-- 
Corinna Vinschen
Cygwin Maintainer
Red Hat

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

Re: Fix modification of string literal by swprintf

[Thomas Preudhomme]

Hi Corinna,

On 07/06/17 11:01, Corinna Vinschen wrote:
> On Jun  5 09:59, Thomas Preudhomme wrote:
>> Don't over-read memory returned by _DTOA_R, and never write to it
>> since the result might be a string literal.
>>
>> For example, when doing:
>>   swprintf(tt, 20, L"%.*f", 6, 0.0);
>>
>> we will get back "0".
>>
>> Instead, write the result returned by _DTOA_R to the output buffer.
>> After this, write the 0 chars directly to the the output buffer
>> (if there are any). This also has the (marginal) advantage that
>> we read/write less memory overall.
>>
>> The patch, contributed by Silviu Baranga was tested against libcxx testsuite
>> and showed no regression. Please find the patch in git format-patch format
>> in attachment.
>>
>> Best regards,
>>
>> Thomas
>
>> >From 7a31cfb01a0b089daf2bed93b742b6edbf4cba0c Mon Sep 17 00:00:00 2001
>> From: Silviu Baranga <silviu.baranga@arm.cm>
>> Date: Mon, 5 Jun 2017 09:54:42 +0100
>> Subject: [PATCH] Don't overread or write memory returned by _DTOA_R
>>
>> Don't over-read memory returned by _DTOA_R, and never write to it
>> since the result might be a string literal.
>>
>> For example, when doing:
>>   swprintf(tt, 20, L"%.*f", 6, 0.0);
>>
>> we will get back "0".
>>
>> Instead, write the result returned by _DTOA_R to the output buffer.
>> After this, write the 0 chars directly to the the output buffer
>> (if there are any). This also has the (marginal) advantage that
>> we read/write less memory overall.
>> ---
>>  newlib/libc/locale/setlocale.h |  2 +-
>>  newlib/libc/stdio/vfwprintf.c  | 24 ++++++++++++------------
>>  2 files changed, 13 insertions(+), 13 deletions(-)
>>
>> diff --git a/newlib/libc/locale/setlocale.h b/newlib/libc/locale/setlocale.h
>> index 85a38d5..1440d0e 100644
>> --- a/newlib/libc/locale/setlocale.h
>> +++ b/newlib/libc/locale/setlocale.h
>> @@ -227,7 +227,7 @@ __get_locale_r (struct _reent *r)
>>  _ELIDABLE_INLINE struct __locale_t *
>>  __get_current_locale (void)
>>  {
>> -  return _REENT->_locale ?: __get_global_locale ();
>> +  return _REENT->_locale ?: __get_global_locale (); // version for !_MB_CAPABLE?
>>  }
>
> This doesn't belong here.  Also, the code is already fine for !_MB_CAPABLE
> as well.

My bad, this was a note I made to myself when reading on MB_CAPABLE some time 
ago. This does not belong to the patch and is indeed wrong.

>
>>  /* Only access fixed "C" locale using this function.  Fake for !_MB_CAPABLE
>> diff --git a/newlib/libc/stdio/vfwprintf.c b/newlib/libc/stdio/vfwprintf.c
>> index f0179a0..1bec9b2 100644
>> --- a/newlib/libc/stdio/vfwprintf.c
>> +++ b/newlib/libc/stdio/vfwprintf.c
>> @@ -1627,13 +1627,20 @@ wcvt(struct _reent *data, _PRINTF_FLOAT_TYPE value, int ndigits, int flags,
>>
>>  	{
>>  	  char *digits, *bp, *rve;
>> -#ifndef _MB_CAPABLE
>>  	  int i;
>> -#endif
>>
>>  	  digits = _DTOA_R (data, value, mode, ndigits, decpt, &dsgn, &rve);
>>
>> +#ifdef _MB_CAPABLE
>> +	  _mbsnrtowcs_r (data, buf, (const char **) &digits, rve - digits,
>> +			 len, NULL);
>> +#else
>> +	  for (i = 0; i < rve - digits && i < len; ++i)
>> +	    buf[i] = (wchar_t) digits[i];
>> +#endif
>> +
>>  	  if ((ch != L'g' && ch != L'G') || flags & ALT) {	/* Print trailing zeros */
>> +		char *padding = rve;
>>  		bp = digits + ndigits;
>>  		if (ch == L'f' || ch == L'F') {
>>  			if (*digits == L'0' && value)
>> @@ -1642,18 +1649,11 @@ wcvt(struct _reent *data, _PRINTF_FLOAT_TYPE value, int ndigits, int flags,
>>  		}
>>  		if (value == 0)	/* kludge for __dtoa irregularity */
>>  			rve = bp;
>> -		while (rve < bp)
>> -			*rve++ = '0';
>> -	  }
>>
>> +		for (i = padding - digits; i < rve - digits && i < len; ++i)
>> +			buf[i] = L'0';
>
> Appending zeros here without incrementing rve...
>
>> +	  }
>>  	  *length = rve - digits; /* full length of the string */
>
> ...leads to incorrect setting of *length here.  Or am I missing
> something?

I'll transmit to Silviu for him to answer.

Thanks for the comments.

Best regards,

Thomas

Re: Fix modification of string literal by swprintf

[Thomas Preudhomme]

[-- Attachment #1: Type: text/plain, Size: 4310 bytes --]

Hi Corinna,

On 07/06/17 11:41, Thomas Preudhomme wrote:
> Hi Corinna,
>
> On 07/06/17 11:01, Corinna Vinschen wrote:
>> On Jun  5 09:59, Thomas Preudhomme wrote:
>>> Don't over-read memory returned by _DTOA_R, and never write to it
>>> since the result might be a string literal.
>>>
>>> For example, when doing:
>>>   swprintf(tt, 20, L"%.*f", 6, 0.0);
>>>
>>> we will get back "0".
>>>
>>> Instead, write the result returned by _DTOA_R to the output buffer.
>>> After this, write the 0 chars directly to the the output buffer
>>> (if there are any). This also has the (marginal) advantage that
>>> we read/write less memory overall.
>>>
>>> The patch, contributed by Silviu Baranga was tested against libcxx testsuite
>>> and showed no regression. Please find the patch in git format-patch format
>>> in attachment.
>>>
>>> Best regards,
>>>
>>> Thomas
>>
>>> >From 7a31cfb01a0b089daf2bed93b742b6edbf4cba0c Mon Sep 17 00:00:00 2001
>>> From: Silviu Baranga <silviu.baranga@arm.cm>
>>> Date: Mon, 5 Jun 2017 09:54:42 +0100
>>> Subject: [PATCH] Don't overread or write memory returned by _DTOA_R
>>>
>>> Don't over-read memory returned by _DTOA_R, and never write to it
>>> since the result might be a string literal.
>>>
>>> For example, when doing:
>>>   swprintf(tt, 20, L"%.*f", 6, 0.0);
>>>
>>> we will get back "0".
>>>
>>> Instead, write the result returned by _DTOA_R to the output buffer.
>>> After this, write the 0 chars directly to the the output buffer
>>> (if there are any). This also has the (marginal) advantage that
>>> we read/write less memory overall.
>>> ---
>>>  newlib/libc/locale/setlocale.h |  2 +-
>>>  newlib/libc/stdio/vfwprintf.c  | 24 ++++++++++++------------
>>>  2 files changed, 13 insertions(+), 13 deletions(-)
>>>
>>> diff --git a/newlib/libc/locale/setlocale.h b/newlib/libc/locale/setlocale.h
>>> index 85a38d5..1440d0e 100644
>>> --- a/newlib/libc/locale/setlocale.h
>>> +++ b/newlib/libc/locale/setlocale.h
>>> @@ -227,7 +227,7 @@ __get_locale_r (struct _reent *r)
>>>  _ELIDABLE_INLINE struct __locale_t *
>>>  __get_current_locale (void)
>>>  {
>>> -  return _REENT->_locale ?: __get_global_locale ();
>>> +  return _REENT->_locale ?: __get_global_locale (); // version for
>>> !_MB_CAPABLE?
>>>  }
>>
>> This doesn't belong here.  Also, the code is already fine for !_MB_CAPABLE
>> as well.
>
> My bad, this was a note I made to myself when reading on MB_CAPABLE some time
> ago. This does not belong to the patch and is indeed wrong.
>
>>
>>>  /* Only access fixed "C" locale using this function.  Fake for !_MB_CAPABLE
>>> diff --git a/newlib/libc/stdio/vfwprintf.c b/newlib/libc/stdio/vfwprintf.c
>>> index f0179a0..1bec9b2 100644
>>> --- a/newlib/libc/stdio/vfwprintf.c
>>> +++ b/newlib/libc/stdio/vfwprintf.c
>>> @@ -1627,13 +1627,20 @@ wcvt(struct _reent *data, _PRINTF_FLOAT_TYPE value,
>>> int ndigits, int flags,
>>>
>>>      {
>>>        char *digits, *bp, *rve;
>>> -#ifndef _MB_CAPABLE
>>>        int i;
>>> -#endif
>>>
>>>        digits = _DTOA_R (data, value, mode, ndigits, decpt, &dsgn, &rve);
>>>
>>> +#ifdef _MB_CAPABLE
>>> +      _mbsnrtowcs_r (data, buf, (const char **) &digits, rve - digits,
>>> +             len, NULL);
>>> +#else
>>> +      for (i = 0; i < rve - digits && i < len; ++i)
>>> +        buf[i] = (wchar_t) digits[i];
>>> +#endif
>>> +
>>>        if ((ch != L'g' && ch != L'G') || flags & ALT) {    /* Print trailing
>>> zeros */
>>> +        char *padding = rve;
>>>          bp = digits + ndigits;
>>>          if (ch == L'f' || ch == L'F') {
>>>              if (*digits == L'0' && value)
>>> @@ -1642,18 +1649,11 @@ wcvt(struct _reent *data, _PRINTF_FLOAT_TYPE value,
>>> int ndigits, int flags,
>>>          }
>>>          if (value == 0)    /* kludge for __dtoa irregularity */
>>>              rve = bp;
>>> -        while (rve < bp)
>>> -            *rve++ = '0';
>>> -      }
>>>
>>> +        for (i = padding - digits; i < rve - digits && i < len; ++i)
>>> +            buf[i] = L'0';
>>
>> Appending zeros here without incrementing rve...
>>
>>> +      }
>>>        *length = rve - digits; /* full length of the string */
>>
>> ...leads to incorrect setting of *length here.  Or am I missing
>> something?
>
> I'll transmit to Silviu for him to answer.

Please find his updated patch attached.

Best regards,

Thomas

[-- Attachment #2: 0001-Don-t-overread-or-write-memory-returned-by-_DTOA_R.patch --]
[-- Type: text/x-patch, Size: 2173 bytes --]

From ee96a273969b0bccd308b81bcdf37a6fdefa06f6 Mon Sep 17 00:00:00 2001
From: Silviu Baranga <silviu.baranga@arm.cm>
Date: Mon, 5 Jun 2017 09:54:42 +0100
Subject: [PATCH] Don't overread or write memory returned by _DTOA_R

Don't over-read memory returned by _DTOA_R, and never write to it
since the result might be a string literal.

For example, when doing:
  swprintf(tt, 20, L"%.*f", 6, 0.0);

we will get back "0".

Instead, write the result returned by _DTOA_R to the output buffer.
After this, write the 0 chars directly to the the output buffer
(if there are any). This also has the (marginal) advantage that
we read/write less memory overall.
---
 newlib/libc/stdio/vfwprintf.c | 25 +++++++++++++------------
 1 file changed, 13 insertions(+), 12 deletions(-)

diff --git a/newlib/libc/stdio/vfwprintf.c b/newlib/libc/stdio/vfwprintf.c
index f0179a0..eb5a2dc 100644
--- a/newlib/libc/stdio/vfwprintf.c
+++ b/newlib/libc/stdio/vfwprintf.c
@@ -1627,12 +1627,18 @@ wcvt(struct _reent *data, _PRINTF_FLOAT_TYPE value, int ndigits, int flags,
 
 	{
 	  char *digits, *bp, *rve;
-#ifndef _MB_CAPABLE
 	  int i;
-#endif
 
 	  digits = _DTOA_R (data, value, mode, ndigits, decpt, &dsgn, &rve);
 
+#ifdef _MB_CAPABLE
+	  _mbsnrtowcs_r (data, buf, (const char **) &digits, rve - digits,
+			 len, NULL);
+#else
+	  for (i = 0; i < rve - digits && i < len; ++i)
+	    buf[i] = (wchar_t) digits[i];
+#endif
+
 	  if ((ch != L'g' && ch != L'G') || flags & ALT) {	/* Print trailing zeros */
 		bp = digits + ndigits;
 		if (ch == L'f' || ch == L'F') {
@@ -1642,18 +1648,13 @@ wcvt(struct _reent *data, _PRINTF_FLOAT_TYPE value, int ndigits, int flags,
 		}
 		if (value == 0)	/* kludge for __dtoa irregularity */
 			rve = bp;
-		while (rve < bp)
-			*rve++ = '0';
-	  }
 
+		for (i = rve - digits; i < bp - digits && i < len; ++i)
+			buf[i] = L'0';
+
+		rve = rve > bp ? rve : bp;
+	  }
 	  *length = rve - digits; /* full length of the string */
-#ifdef _MB_CAPABLE
-	  _mbsnrtowcs_r (data, buf, (const char **) &digits, *length,
-			 len, NULL);
-#else
-	  for (i = 0; i < *length && i < len; ++i)
-	    buf[i] = (wchar_t) digits[i];
-#endif
 	  return buf;
 	}
 }
-- 
1.9.1

Re: Fix modification of string literal by swprintf

[Corinna Vinschen]

[-- Attachment #1: Type: text/plain, Size: 911 bytes --]

On Jun  8 13:15, Thomas Preudhomme wrote:
> Please find his updated patch attached.
> 
> Best regards,
> 
> Thomas

> >From ee96a273969b0bccd308b81bcdf37a6fdefa06f6 Mon Sep 17 00:00:00 2001
> From: Silviu Baranga <silviu.baranga@arm.cm>
> Date: Mon, 5 Jun 2017 09:54:42 +0100
> Subject: [PATCH] Don't overread or write memory returned by _DTOA_R
> 
> Don't over-read memory returned by _DTOA_R, and never write to it
> since the result might be a string literal.
> 
> For example, when doing:
>   swprintf(tt, 20, L"%.*f", 6, 0.0);
> 
> we will get back "0".
> 
> Instead, write the result returned by _DTOA_R to the output buffer.
> After this, write the 0 chars directly to the the output buffer
> (if there are any). This also has the (marginal) advantage that
> we read/write less memory overall.

Pushed.


Thanks,
Corinna

-- 
Corinna Vinschen
Cygwin Maintainer
Red Hat

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

Re: Re: Fix modification of string literal by swprintf

[Renlin Li]

Hi there,

It seems this patch cause few regressions on arm/aarch64 baremetal targets.

libstdc++-v3:
FAIL: 21_strings/basic_string/numeric_conversions/wchar_t/dr1261.cc execution test
FAIL: 21_strings/basic_string/numeric_conversions/wchar_t/stof.cc execution test
FAIL: 21_strings/basic_string/numeric_conversions/wchar_t/to_wstring.cc execution test
FAIL: 27_io/basic_ostream/inserters_arithmetic/wchar_t/4402.cc execution test

newlib:
FAIL: newlib.stdio/swprintf.c execution

Regards,
Renlin

On 09/06/17 14:32, Corinna Vinschen wrote:
> On Jun  8 13:15, Thomas Preudhomme wrote:
>> Please find his updated patch attached.
>>
>> Best regards,
>>
>> Thomas
>
>> >From ee96a273969b0bccd308b81bcdf37a6fdefa06f6 Mon Sep 17 00:00:00 2001
>> From: Silviu Baranga <silviu.baranga@arm.cm>
>> Date: Mon, 5 Jun 2017 09:54:42 +0100
>> Subject: [PATCH] Don't overread or write memory returned by _DTOA_R
>>
>> Don't over-read memory returned by _DTOA_R, and never write to it
>> since the result might be a string literal.
>>
>> For example, when doing:
>>    swprintf(tt, 20, L"%.*f", 6, 0.0);
>>
>> we will get back "0".
>>
>> Instead, write the result returned by _DTOA_R to the output buffer.
>> After this, write the 0 chars directly to the the output buffer
>> (if there are any). This also has the (marginal) advantage that
>> we read/write less memory overall.
>
> Pushed.
>
>
> Thanks,
> Corinna
>

Re: Fix modification of string literal by swprintf

[Thomas Preudhomme]

Hi Renlin,

Here's Silviu's answer:

"This happens when  --enable-newlib-mb is used (which wasn't in my test 
configuration).

Unfortunately this complicates thing a lot and I have no easy fix for it.
The best action at the moment would be a revert."

Corinna, could you revert the patch for now?

Sorry for the trouble. Best regards,

Thomas

On 14/06/17 12:05, Renlin Li wrote:
> Hi there,
>
> It seems this patch cause few regressions on arm/aarch64 baremetal targets.
>
> libstdc++-v3:
> FAIL: 21_strings/basic_string/numeric_conversions/wchar_t/dr1261.cc execution test
> FAIL: 21_strings/basic_string/numeric_conversions/wchar_t/stof.cc execution test
> FAIL: 21_strings/basic_string/numeric_conversions/wchar_t/to_wstring.cc
> execution test
> FAIL: 27_io/basic_ostream/inserters_arithmetic/wchar_t/4402.cc execution test
>
> newlib:
> FAIL: newlib.stdio/swprintf.c execution
>
> Regards,
> Renlin
>
> On 09/06/17 14:32, Corinna Vinschen wrote:
>> On Jun  8 13:15, Thomas Preudhomme wrote:
>>> Please find his updated patch attached.
>>>
>>> Best regards,
>>>
>>> Thomas
>>
>>> >From ee96a273969b0bccd308b81bcdf37a6fdefa06f6 Mon Sep 17 00:00:00 2001
>>> From: Silviu Baranga <silviu.baranga@arm.cm>
>>> Date: Mon, 5 Jun 2017 09:54:42 +0100
>>> Subject: [PATCH] Don't overread or write memory returned by _DTOA_R
>>>
>>> Don't over-read memory returned by _DTOA_R, and never write to it
>>> since the result might be a string literal.
>>>
>>> For example, when doing:
>>>    swprintf(tt, 20, L"%.*f", 6, 0.0);
>>>
>>> we will get back "0".
>>>
>>> Instead, write the result returned by _DTOA_R to the output buffer.
>>> After this, write the 0 chars directly to the the output buffer
>>> (if there are any). This also has the (marginal) advantage that
>>> we read/write less memory overall.
>>
>> Pushed.
>>
>>
>> Thanks,
>> Corinna
>>

Re: Fix modification of string literal by swprintf

[Renlin Li]

Yes, in my configuration, we enable the newlib-mb feature. And the executing path to 
handle the trailing '0' is not consistent with how the digits is handled.

Thanks,
Renlin


On 15/06/17 09:00, Thomas Preudhomme wrote:
> Hi Renlin,
>
> Here's Silviu's answer:
>
> "This happens when  --enable-newlib-mb is used (which wasn't in my test configuration).
>
> Unfortunately this complicates thing a lot and I have no easy fix for it.
> The best action at the moment would be a revert."
>
> Corinna, could you revert the patch for now?
>
> Sorry for the trouble. Best regards,
>
> Thomas
>
> On 14/06/17 12:05, Renlin Li wrote:
>> Hi there,
>>
>> It seems this patch cause few regressions on arm/aarch64 baremetal targets.
>>
>> libstdc++-v3:
>> FAIL: 21_strings/basic_string/numeric_conversions/wchar_t/dr1261.cc execution test
>> FAIL: 21_strings/basic_string/numeric_conversions/wchar_t/stof.cc execution test
>> FAIL: 21_strings/basic_string/numeric_conversions/wchar_t/to_wstring.cc
>> execution test
>> FAIL: 27_io/basic_ostream/inserters_arithmetic/wchar_t/4402.cc execution test
>>
>> newlib:
>> FAIL: newlib.stdio/swprintf.c execution
>>
>> Regards,
>> Renlin
>>
>> On 09/06/17 14:32, Corinna Vinschen wrote:
>>> On Jun  8 13:15, Thomas Preudhomme wrote:
>>>> Please find his updated patch attached.
>>>>
>>>> Best regards,
>>>>
>>>> Thomas
>>>
>>>> >From ee96a273969b0bccd308b81bcdf37a6fdefa06f6 Mon Sep 17 00:00:00 2001
>>>> From: Silviu Baranga <silviu.baranga@arm.cm>
>>>> Date: Mon, 5 Jun 2017 09:54:42 +0100
>>>> Subject: [PATCH] Don't overread or write memory returned by _DTOA_R
>>>>
>>>> Don't over-read memory returned by _DTOA_R, and never write to it
>>>> since the result might be a string literal.
>>>>
>>>> For example, when doing:
>>>>    swprintf(tt, 20, L"%.*f", 6, 0.0);
>>>>
>>>> we will get back "0".
>>>>
>>>> Instead, write the result returned by _DTOA_R to the output buffer.
>>>> After this, write the 0 chars directly to the the output buffer
>>>> (if there are any). This also has the (marginal) advantage that
>>>> we read/write less memory overall.
>>>
>>> Pushed.
>>>
>>>
>>> Thanks,
>>> Corinna
>>>

Re: Fix modification of string literal by swprintf

[Corinna Vinschen]

[-- Attachment #1: Type: text/plain, Size: 445 bytes --]

On Jun 15 09:00, Thomas Preudhomme wrote:
> Hi Renlin,
> 
> Here's Silviu's answer:
> 
> "This happens when  --enable-newlib-mb is used (which wasn't in my test
> configuration).
> 
> Unfortunately this complicates thing a lot and I have no easy fix for it.
> The best action at the moment would be a revert."
> 
> Corinna, could you revert the patch for now?

Done.

Corinna

-- 
Corinna Vinschen
Cygwin Maintainer
Red Hat

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 819 bytes --]