Milter-greylist

--- In milter-greylist@yahoogroups.com, manu@n... wrote:
> Ranko Zivojnovic <ranko@s...> wrote:
> 
> > Looking further through the code, I saw that local_addr()
function can
> > return three values: 1, 0 and -1, but it is used in the code as
"true" -
> > "false" kind of call:
> >    if(local_addr(...)) {
> >      ...
> >    }
> > 
> > Then I went back couple of days in my logs and found the following
> > error:
> > local_addr: bind failed: Bad file number
> (snip)
> > I did fix it localy to return 0 rather than -1, however I thought
I
> > should let you know of the issue and you might want to reconsider
the
> > use of local_addr() in a different way.
> 
> I think your fix just ignore if the address is local or not, and
force
> it to be non local. 
> 
> Here socket() returned a positive file descriptor, but it's still a
bad
> file descriptor. Bad file number is EBADF, right? Anyone used to
Solaris
> can comment on this?

True. bind() should have set errno to EADDRNOTAVAIL, however (here at
least) to me it _always_ gives EBADF, on both systems, so what I've
changed was:
  if (errno != EADDRNOTAVAIL)

to 
  if (errno != EADDRNOTAVAIL && errno != EBADF)

... in order to bypass the problem.

I did however further tests with regards to the behavior on Solaris
and in all the tests I did, errno was always set to EADDRNOTAVAIL as
it should. I could not reproduce (yet) the case outside sync.c code
where bind() would give me EBADF on a valid socket.

Something is really strange here...

R.