Question regarding gcc 12 Werror=stringop-overflow in iconv/loop.c

Question regarding gcc 12 Werror=stringop-overflow in iconv/loop.c

[Stefan Liebler]

Hi,

with GCC 12, when building with -O3 I'm getting this werror on s390x
when compiling sysdeps/s390/utf8-utf16-z9.c:
../iconv/loop.c: In function ‘__from_utf8_loop_etf3eh_single’:
../iconv/loop.c:445:22: error: writing 1 byte into a region of size 0
[-Werror=stringop-overflow=]
  445 |     bytebuf[inlen++] = *inptr++;
      |     ~~~~~~~~~~~~~~~~~^~~~~~~~~~
../iconv/loop.c:381:17: note: at offset 4 into destination object
‘bytebuf’ of size 4
  381 |   unsigned char bytebuf[MAX_NEEDED_INPUT];
      |                 ^~~~~~~
../iconv/loop.c:445:22: error: writing 1 byte into a region of size 0
[-Werror=stringop-overflow=]
  445 |     bytebuf[inlen++] = *inptr++;
      |     ~~~~~~~~~~~~~~~~~^~~~~~~~~~
../iconv/loop.c:381:17: note: at offset 5 into destination object
‘bytebuf’ of size 4
  381 |   unsigned char bytebuf[MAX_NEEDED_INPUT];
      |                 ^~~~~~~

In this case MAX_NEEDED_INPUT is 4.
I can silence the warning if I use a while-loop in loop.c:
  while (inlen < MAX_NEEDED_INPUT && inptr < inend)
    bytebuf[inlen++] = *inptr++;
instead of the do-while-loop.

But then I'm get an werror with GCC 11.2 in iconvdata/gbgbk.c.

Is there a guarantee that (inptr < inend) is always met?
Then extending the unreachable-case above the loop would silence the
werrors in both files:
if (inlen >= MAX_NEEDED_INPUT || inptr >= inend)
    /* Avoid a -Wstringop-overflow= warning when this loop is
       unrolled.  The compiler cannot otherwise see that this is
       unreachable because it depends on (state->__count & 7) not
       being too large after a previous conversion step.  */
    __builtin_unreachable ();


Otherwise an extra check above the loop like this, would also silence
the werrors in both files:
  if (inptr >= inend)
    return __GCONV_INCOMPLETE_INPUT;

Thanks in advance for any help and hints.
Stefan

Re: Question regarding gcc 12 Werror=stringop-overflow in iconv/loop.c

[Florian Weimer]

* Stefan Liebler via Libc-alpha:

> Is there a guarantee that (inptr < inend) is always met?

When calling iconv (the external API), inptr <= inend must hold (no
pointer wraparound).  I think there is a minimum-input-size check in
iconv/loop.c, so it follows that we have inptr < inend.  I think.

Thanks,
Florian