Re: [RFC][PATCH 00/20] Mount writer count and read-only bind mounts (v2)

[Herbert Poetzl]

On Sat, Jun 17, 2006 at 02:10:17AM +0200, Grzegorz Kulewski wrote:
> On Fri, 16 Jun 2006, Dave Hansen wrote:
> >On Sat, 2006-06-17 at 01:29 +0200, Grzegorz Kulewski wrote:
> >>Isn't this some kind of security risk (at least in my planned use)? I mean
> >>- for a small fraction of second somebody seeing /dest can write
> >>/source... No?
> >
> >I assume you're talking about this kind of situation:
> >
> >mount --bind /local/writable/dir /chroot/untrusted/area/
> >mount --o remount,ro /chroot/untrusted/area/
> 
> Well, actually about some kind of VPS: openvz or something like that.
> But yes, this is the same kind of scenario.

yes, Linux-VServer provides this kind of ro --bind mounts
without the race, as the the flags are passed on the actual
mount

> >This has no r/w window in the chroot area:
> >
> >mount --bind /local/writable/dir /tmp/area/
> >mount --o remount,ro /tmp/area/
> >mount --bind /tmp/area/ /chroot/untrusted/area/
> >umount /tmp/area/
> 
> Well, it looks a little scarry and complicated at first. And probably
> requires you to know that semantic of --bind lets you do the last
> unmount. But if you are saying that this makes kernel smaller, faster
> and less buggy then you are probably very right.

well, it makes the kernel more consistant in it's behaviour,
because especially for --rbind mounts, the logic what is
changed where and when is not as well defined as one would
wish ...

btw, you could get the same result by simply doing:

mount --bind /local/writable/dir /tmp/area/
mount --o remount,ro /tmp/area/
mount --move /tmp/area/ /chroot/untrusted/area/

without the duplicate mount and the unmount

HTH,
Herbert

> Thank you for your explanation,
> 
> Grzegorz Kulewski
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/