linux-kernel - Re: [RFC 0/5] forced comounts for cgroups.

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite for Android: free password hash cracker in your pocket

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20120905082947.GD3195@dhcp-172-17-108-109.mtv.corp.google.com>
Date:	Wed, 5 Sep 2012 01:29:47 -0700
From:	Tejun Heo <tj@...nel.org>
To:	Glauber Costa <glommer@...allels.com>
Cc:	linux-kernel@...r.kernel.org, cgroups@...r.kernel.org,
	linux-mm@...ck.org, davej@...hat.com, ben@...adent.org.uk,
	a.p.zijlstra@...llo.nl, pjt@...gle.com, lennart@...ttering.net,
	kay.sievers@...y.org
Subject: Re: [RFC 0/5] forced comounts for cgroups.

Hello, Glauber.

On Wed, Sep 05, 2012 at 12:17:11PM +0400, Glauber Costa wrote:
> > Distros can just co-mount them during boot.  What's the point of the
> > config options?
> 
> Pretty simple. The kernel can't assume the distro did. And then we still
> need to pay a stupid big price in the scheduler.
> 
> After this patchset, We can assume this. And cpuusage can totally be
> derived from the cpu cgroup. Because much more than "they can comount",
> we can assume they did.

As long as cpuacct and cpu are separate, I think it makes sense to
assume that they at least could be at different granularity.  As for
optimization for co-mounted case, if that is *really* necessary,
couldn't it be done dynamically?  It's not like CONFIG_XXX blocks are
pretty things and they're worse for runtime code path coverage.

> > Differing hierarchies in memcg and blkcg currently is the most
> > prominent case where the intersection in writeback is problematic and
> > your proposed solution doesn't help one way or the other.  What's the
> > point?
> 
> The point is that I am focusing at one problem at a time. But FWIW, I
> don't see why memcg/blkcg can't use a step just like this one in a
> separate pass.
> 
> If the goal is comounting them eventually, at some point when the issues
> are sorted out, just do it. Get a switch like this one, and then you
> will start being able to assume a lot of things in the code. Miracles
> can happen.

The problem is that I really don't see how this leads to where we
eventually wanna be.  Orthogonal hierarchies are bad because,

* It complicates the code.  This doesn't really help there much.

* Intersections between controllers are cumbersome to handle.  Again,
  this doesn't help much.

And this restricts the only valid use case for multiple hierarchies
which is applying differing level of granularity depending on
controllers.  So, I don't know.  Doesn't seem like a good idea to me.

Thanks.

-- 
tejun
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/