linux-kernel - Re: Containers: css

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: Windows password security audit tool. GUI, reports in PDF.

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

Date:	Mon, 16 Jul 2007 12:03:18 -0700
From:	"Paul (宝瑠) Menage" 
	<menage@...gle.com>
To:	balbir@...ux.vnet.ibm.com
Cc:	"Andrew Morton" <akpm@...ux-foundation.org>,
	"Srivatsa Vaddagiri" <vatsa@...ibm.com>,
	"Pavel Emelianov" <xemul@...ru>,
	"Vaidyanathan Srinivasan" <svaidy@...ux.vnet.ibm.com>,
	"Serge E. Hallyn" <serue@...ibm.com>,
	"linux kernel mailing list" <linux-kernel@...r.kernel.org>,
	"Linux Containers" <containers@...ts.osdl.org>,
	"Paul Jackson" <pj@....com>
Subject: Re: Containers: css_put() dilemma

On 7/16/07, Balbir Singh <balbir@...ux.vnet.ibm.com> wrote:
> Hi, Paul,
>
> I've run into a strange problem with css_put(). After the changes for notify_on_release(), the css_put() routine can now block and it blocks on
> the container_mutex. This implies that css_put() cannot be called if
>
> 1. We cannot block
> 2. We already hold the container_mutex
>
> The problem I have is that of preventing the destruction of my container
> (when the user does rmdir). If the user migrates away all tasks and does
> an rmdir, the only way to prevent the container from going away is through
> css_get() references. In my case, some pages have been allocated from the
> container and hence I do not want it to go away, until all the pages
> charged to it are freed. When I use css_get/put() to prevent destruction
> I am blocked by the limitations of css_put() listed above.
>
> Do you have any recommendations for a cleaner solution? I suspect we'll
> need can_destroy() callbacks (similar to can_attach()).

I think moving the release_list synchronization inside a separate
spinlock, and thus not requiring container_mutex to be held for
check_for_release(), is the simplest solution. I'll do that. I'm
hoping to get a new set of patches to Andrew today or tomorrow.

Adding a can_destroy() callback is possible, but since I envisage that
most subsystems that would want to implement it would basically be
doing reference counting anyway, it seems worth having a generic
reference counting mechanism in the framework. In particular, since
once the container does become releasable due to all the
subsystem-specific refcounts being released, we want to be able to
invoke the release agent, we'll end up with the same synchronization
problems that we have now if we just pushed everything into a
can_destroy() method. (Unless the framework polled all can_destroy()
methods for potentially-removable containers, which seems a bit
nasty).

We can add can_destroy() if we encounter a situation that can't be
handled by generic reference counting.

Paul
-
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/