Date: Tue, 15 Oct 1996 14:00:16 -0700 (MST) From: "Jeffrey D. Wheelhouse" <jdw@wwwi.com> To: freebsd-hackers@freebsd.org Subject: Re: IP bugs in FreeBSD 2.1.5 Message-ID: <199610152100.OAA03005@wwwi.com>
next in thread | raw e-mail | index | archive | help
<moved to FreeBSD-hackers, not "technically" a bug>
Let me begin by saying that is absolutely not my intention to
flame Julian or anyone else... Julian has contributed infinitely
more to the FreeBSD project than I have.. However I've noticed
an attitude recently which disturbs me.
I'm one of the most vehement toers-of-the-line when it comes to
current vs. stable. In fact, despite the advances in -current
that would make my life a little easier (and possibly faster),
every single FreeBSD machine I have deployed, for myself and my
customers, is running stable/2.1.5. All over the web site and
in the docs it's billed as the "stable" version vs. current as
an "experimental" version... and it is just that... very stable.
Haven't had a single FreeBSD machine panic in ages... some under
some very heavy loads.
Meanwhile, I sat back and watched the debate about -current and
whether it should be always buildable, etc, for the sake of the
people who complained when it wasn't. I resisted the urge to
say something about the silliness of expecting the experimental
version to work 100% of the time on the old "If you don't have
anything nice to say..." rule.
Now, however, I'm a little concerned. Work on -stable has
effectively stopped with 2.1.5 (reasons why no mystery to list
readers). However, Brian's message is not the first time that
someone has said "here is a problem in 2.1.5" and gotten a
reponse of "that is fixed in 2.2" or "the behavior is different
under 2.2" or even the dreaded "have you tried it on -current/2.2?"
You [all] told me (on the web pages) not to run 2.2!
I agreed 100% with the reasons listed why a person should
choose -stable over -current for their production servers, and I did.
Looking at the flurry of messages on the -current list about "oops,
now it panics at [xxx] after [xxx] minutes" I think this was the
right choice.
I'm not crying for tech support of any kind, and I'm certainly not
insisting that "somebody stop what you're doing and come fix my bugs!"
I run stable, it is stable and I make no complaints. I have an HP
CD-R in an NT machine because it won't work in 2.1.5 ("you should upgrade
to current!") and the problem with pinging multiple IP address Brian
mentioned is most annoying, but certainly of a minor impact.
It's just that when someone says "I have a problem," and the answer is
"We fixed that in the version you shouldn't use." That is like saying
"Have this yummy cupcake," and then putting the cupcake behind 2 inches
of plexiglass with "WARNING: hard hat area; experts only" painted on it.
Ok, maybe it is not quite like that, but you see what I mean. ;-)
I guess -current is getting awfully far ahead of -stable. Which is
natural, since it's moving and -stable isn't. Maybe I'm the only one
that's starting to feel left behind, like -stable's been unhooked
from the FreeBSD train (partially because it was slowing the train down).
Later,
Jeff
>Date: Tue, 15 Oct 1996 12:54:36 -0700
>From: Julian Elischer <julian@whistle.com>
>Organization: Whistle Communications
>To: Brian Candler <B.Candler@pobox.com>
>CC: bugs@FreeBSD.org, nato-ws@ripe.net, t12@psg.com
>Subject: Re: IP bugs in FreeBSD 2.1.5
>References: <199610151340.OAA00334@gazebo.candler.demon.co.uk>
>Sender: owner-bugs@FreeBSD.org
>X-Loop: FreeBSD.org
>
>Brian Candler wrote:
>>
>> Dear FreeBSD team,
>>
>> I have just returned from a NATO-sponsored Advanced Networking Course held
>> in St Petersburg, Russia, where I was the chief instructor.
>>
>> For our PC-based practical sessions we used FreeBSD 2.1.5.
>>
>NEAT!
>
>I hope each participant got a FreeBSD cdrom as part of the
>course materials! :)
>
>> We uncovered a
>> number of bugs in the TCP/IP code, which I would like to report in the hope
>> they can be fixed in some future release.
>
>certainly..
>Have you looked at any of these problems with 2.2?
>
>>
>> 1. On several occasions we found that although a default route appeared to
>> be in the kernel forwarding table (as shown by netstat -nr), it did
>> not work. However simply by deleting and reinserting the exact same
>> default route, it then worked fine. I'm afraid I can't give you
>> a set of circumstances which can cause this problem to be
>> reproduced.
>>
>possibly another route preceeded it in some way
>I can imagine such a case but can't quite put my finger on a
>mechanism.
>
>> 2. When you set up a ppp link, you cannot ping your own IP number
>> locally. (This is using kernel-based ppp, with pppd to set up the
>> link)
>
>Under -current pinging the local end address certainly works..
>even if the link is not up. of course pinging the remote end fails :)
>
>
>
>> 3. When you use ipalias to set up multiple IP addresses for an
>> ethernet card, you cannot ping those additional IP numbers locally.
>
>hmmmmm interesting... they are logically identical to the original.
>address, it certainly works for me under -current.
>
>
>>
>> 4. If you 'ifconfig down' an interface, then set up a default route
>> via another interface, you still cannot ping the range of IP numbers
>> which the original interface covered - presumably the kernel still
>> tries to send them via the (downed) interface.
>
>I think I've seen related things
>sometimes the kernel keeps hold of rtentry structs with outdated
>information, even though they are not in teh table any more..
>this is because the protocol PCBs 'cache' them and
>there is no way to flush cached rtentry's.
>It's on my "look at this some more" list
>
>>
>> 5. On one occasion the kernel forwarding table had a bad entry (I
>> think a "link #1" type entry) which could not be removed, apart from
>> by rebooting the machine.
>>
>
>sorry, no idea
>
>
>
>> I am unlikely to have time to look at the source code myself I'm
>> afraid, but I hope you don't mind me submitting the notes above. Apart
>> from these points, and "vi" dumping core a few times, FreeBSD was
>> stable for the whole week we were using it.
>>
>without such feedback we'd have a much harder time fixing the system!
>
>> Thanks in advance for your attention,
>>
>> Brian Candler <B.Candler@pobox.com>
>
>
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199610152100.OAA03005>