Re: tux on 2.4.27 kernel and referrer checking

[Marek Habersack <grendel@caudium.net>]

On Wed, Oct 27, 2004 at 05:14:38PM -0400, Dean Lim scribbled:
> I just checked out the mailing list at kernel.org and alot of people
> are complaining about the new 2.6.9 kernel and all its bugs regarding
> production use. I would really like to use the 2.6 kernel series so I
> am wondering why you say its stable for production use.
It most definitely isn't. The fact that RedHat or SuSe ship with the kernels
doesn't mean the kernels are stable. The reality is that 2.6 has some
security issues that are pending fixes (see Alan's -ac kernels, or look up
the lkml archives for the "lost" threads thread where if you kill the main
thread of the process the remaining threads will remain hidden from ps(1)
and friends, if using NPTL). Also the inflow of changes in the Andrew
Morton's -mm series, which then are promoted to the mainline kernel, means
that there are still a lot of things happening below the surface of the
userland, which is not a good sign for stability and production use. As for
the opinion that the vendors ship stable kernels while the mainline kernel
is a "development" one, I can only say - bullshit. There has been a talk
about doing that _in the future_, but frankly I hope (and think) it won't be
so, as that would probably hurt the community as a whole. But that's OT, if
you want to discuss that with me, please followup privately.
As for 2.6... I would _love_ to use it in production for several reasons
(strict overcommit control, better SATA support, faster threads, better 
block device support, the BSD secure levels etc.) but as it stands now, it's
probably still test quality and will probably remain so until 2.6.10/11 (as
it was practically with every previous kernel series). So, my very humble
opinion is that you should let 2.6 stabilize for real for a while more and
only then hop in to use it.

If anybody wants to flame me for the opinions stated above, do it off the list,
please.

regards,

marek

Attachment: signature.asc
Description: Digital signature