Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Thu, 4 Aug 2011 03:15:51 +0200
From: Kay Sievers <kay.sievers@...y.org>
To: Linus Torvalds <torvalds@...ux-foundation.org>
Cc: Andrew Morton <akpm@...ux-foundation.org>, Marc Zyngier <marc.zyngier@....com>, 
	Vasiliy Kulikov <segoon@...nwall.com>, Manuel Lauss <manuel.lauss@...glemail.com>, 
	"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>, Richard Weinberger <richard@....at>, 
	Ingo Molnar <mingo@...e.hu>, 
	"kernel-hardening@...ts.openwall.com" <kernel-hardening@...ts.openwall.com>, 
	"Paul E. McKenney" <paul.mckenney@...aro.org>, Greg KH <greg@...ah.com>
Subject: Re: [PATCH] shm: fix a race between shm_exit() and shm_init()

On Thu, Aug 4, 2011 at 03:01, Linus Torvalds
<torvalds@...ux-foundation.org> wrote:
> On Wed, Aug 3, 2011 at 2:50 PM, Andrew Morton <akpm@...ux-foundation.org> wrote:
>>
>> Well, it's still a workaround.  We'll still have driver_init() trying
>> to run userspace helpers at an inappropriate time, and failing to do
>> so.  Either something will break or it should not be attempting
>> attempting to do this at all.
>
> Well, it has never worked before either, so I wouldn't call it a
> "workaround", more of a "insane subsystems do user-mode helpers for
> crazy small things, and at crazy times. Stop them from wasting our
> time and effort at those crazy times when we know they would fail
> anyway".
>
>> Perhaps Kay and Greg can suggest how we can fix all this up?
>
> I think just not calling user-mode when user-mode isn't ready is a real fix.

Yes, we should default to off, and probably only enable it just right
before we start /sbin/init from the mounted root or in initramfs.
There should be no useful use of /sbin/hotplug before we mounted root,
or unpacked initramfs.

> The fact that the device models send an absolutely *insane* amount of
> events for everything, and don't track "this is the bootup device
> scan" on their own is kind of sad, but it's how they roll. This fixes
> it at a core level, so that the device layer doesn't have to track the
> "am I booting, or is this a dynamic event?" at all.

It is insane, and the entire uevent hookup with /sbin/hotplug should
just be removed. We send from 1000 to many 1000s of events at bootup,
and forking a binary here makes not much sense these days.

All distros default to CONFIG_UEVENT_HELPER_PATH="" because nothing
else really works anymore, some boxes don't even boot up.
/sbin/hotplug worked fine when we all had one USB mouse connected and
needed to load like 5 drivers for the system, but not today.

It's a broken concept today, at least in the way we plug it into the
driver core makes no sense today, it is just a heritage from the old
days.

Kay

Powered by blists - more mailing lists

Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.