Re: [PATCH 2/3] string: provide strscpy() and strscpy_truncate()

[Chris Metcalf <cmetcalf@ezchip.com>]

On 05/07/2015 05:00 AM, Dan Carpenter wrote:
> On Wed, May 06, 2015 at 06:45:56PM +0200, Geert Uytterhoeven wrote:
>> On Wed, May 6, 2015 at 5:59 PM, Dan Carpenter <dan.carpenter@oracle.com> wrote:
>>> We actually do have a __must_check tag so it's easy enough to force
>>> people to check.  A different option is we could make it trigger a
>> People tend to ignore compiler warnings...
> We're doing a lot better these days with zero day build testing.  There
> is not even one ignored __must_check return in my allmodconfig.

If we keep the strscpy/strscpy_truncate distinction, I agree that having
__must_check on strscpy seems like a good idea.

>>> WARN_ONCE().
>>>
>>> #define strXcpy(dest, src, len) (({                             \
>>>          ssize_t __ret = strscpy_truncate(dest, src, len);       \
>>>          WARN_ONCE(__ret < 0, "strXcpy trancates\n");            \
>>>          __ret; }))
>> Which will probably trigger only in extreme cases in the wild, not during
>> development.
> It's less subtle than just putting an empty string there so we're more
> likely to get bug reports than with the original code.

The problem with WARN_ONCE() here is that we may be using strscpy()
to take user input of some kind.  If so, we don't want to warn if we
are truncating the string - we just want to return a suitable error up
the call stack.

-- 
Chris Metcalf, EZChip Semiconductor
http://www.ezchip.com