Re: SOCKS support?

Thanks for the rather technical details. So far I have not been running
into any (visible) problems at all, the only thing needed was linking with
bind-8.1.2 libbind.a instead of the AIX libbind.

Still, a "clean" implementation which addresses the issues you mentioned
would be very nice.

Cheers,

-Michael

>
> I belive you still need to patch squid to make it work correctly. It should
> work even without patching but there are
> some (minor) problems:
>
> 1. Socksified functions do change errno even on success while original
> functions do not touch errno
> if there were no error. Look at connect() in src/comm.c. Does it still
> relay on unchanged errno as in PATCH2 version of squid?
> If it does then you will get false errors on dnsserver (and ldap) startup.
>
> 2. NEC's libsock doesn't socksify getsockopt(). Does DEVEL3 still uses
> getsockopt() to check connection status? If it does you'll get wrong
> error codes (a la invalid file descriptor instead of connection refused)
> after socksification.
>
> Oleg
>
> >
> > > > environment is AIX432, IBM xlc compiler, squid-2.2.DEVEL3).
> > > ^^^^^^^^^^^^^^^^
> > > This is not supported; see the main web site.
> >
> > Do you want to tell me the supported version has SOCKS ;-)?
> >
> > >
> > > Socksifying Squid is going to force your proxy to have to support
> > > huge numbers of open file descriptors - it may well not do them
> > > anything like as well as Squid and may well run a process per
> > > connection, which is exactly what squid is trying to avoid, although
> > > it is coming back from that extreme in the threaded versions.
> >
> > Hmmm. I relay everything going toward the Internet through a proxy in
> > front of the firewall (single parent mode). I was under the impression
> > that not one socksd is run per every GET. I may have been wrong, but it
> > works nicely, admittadly in a small-medium environment.
> >
> > Cheers,
> >
> > -Michael
>
>
Received on Thu Apr 08 1999 - 05:03:59 MDT