ioctl() on socket fd's take 3 seconds on 1.7.7

[Corinna Vinschen]

On Nov 20 18:25, Jason Curl wrote:
> On 15/11/2010 17:22, Corinna Vinschen wrote:
> >On Nov  9 09:20, Jason Curl wrote:
> >>   166   65418 [main] ipcheck 5580 ioctl: fd 3, cmd 80087364
> >>--- Process 5580, exception C0000005 at 610C8C86
> >
> >Crash in Cygwin, but the address doesn't help much, unfortunately.
> >
> >>Interestingly enough, the program works. That is, it finds all
> >>interfaces and returns correct values (except interface names are
> >>UUIDs instead of something more friendly like "eth0" that existed in
> >>1.5.26).
> >>
> >>It will be a little later when I'm able to generate a working test case.
> >
> >T'would be nice.  Please make it short and simple and, if possible, in
> >plain C.
> 
> I've generated a relatively simple test case that I can compile
> using cygwin1-20101102.dll.

Thank you.  I can't reproduce any slowness, each ioctl takes only a
couple of milliseconds.  At least I could reproduce the exception when
stracing the testcase and I've applied a matching patch to Cygwin.

Can you retry with the next developer snapshot?  Are the ioctl calls
still slow?  If so, I'm wondering if the GetAdaptersAddresses call is
rather slow if a lot of interfaces exist.  Every single ioctl in your
application calls GetAdaptersAddresses twice, like this:

bool
get_adapters_addresses (PIP_ADAPTER_ADDRESSES *pa_ret, ULONG family)
{
  DWORD ret, size = 0;
  PIP_ADAPTER_ADDRESSES pa0 = NULL;

  if (!pa_ret)
    return ERROR_BUFFER_OVERFLOW
           == GetAdaptersAddresses (family, GAA_FLAG_INCLUDE_PREFIX
                                            | GAA_FLAG_INCLUDE_ALL_INTERFACES,
                                    NULL, NULL, &size);
  do
    {
      ret = GetAdaptersAddresses (family, GAA_FLAG_INCLUDE_PREFIX
                                          | GAA_FLAG_INCLUDE_ALL_INTERFACES,
                                  NULL, pa0, &size);
      if (ret == ERROR_BUFFER_OVERFLOW
          && !(pa0 = (PIP_ADAPTER_ADDRESSES) realloc (pa0, size)))
        break;
    }
  while (ret == ERROR_BUFFER_OVERFLOW);
  if (ret != ERROR_SUCCESS)
    {
      if (pa0)
        free (pa0);
      *pa_ret = NULL;
      return false;
    }
  *pa_ret = pa0;
  return true;
}

Can you test how long this call takes in your scenario, and if it
might be the culprit?


Corinna

-- 
Corinna Vinschen                  Please, send mails regarding Cygwin to
Cygwin Project Co-Leader          cygwin AT cygwin DOT com
Red Hat

--
Problem reports:       http://cygwin.com/problems.html
FAQ:                   http://cygwin.com/faq/
Documentation:         http://cygwin.com/docs.html
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple