[SE/Linux] warning about debian hotplug package 20040329-9!

[SE/Linux] warning about debian hotplug package 20040329-9!

[Luke Kenneth Casson Leighton]

dear selinux and linux kernel,

i am after some assistance in clarifying how hotplug works, with
a view to solving an issue with SE/Linux where the default
SE/Linux policy is to deny write permission to /etc/hotplug
(with good reason) but the hotplug package is presently demanding
write permission.

a simple request for a change to writing to /var/state/hotplug
instead has thrown up a number of issues with kernel (2.6.6)
hotplugging and i would greatly appreciate some confirmation
and some assistance.


i raised a bug with the debian maintainer for hotplug [he
is known to be, how to describe in a few polite words, a
control freak.  coming from me, that's saying a lot].

i suggested that the location for files to be written to be moved
to /var/state/hotplug and he CLOSED the bug with a message saying
"you can't do that, /var might not be mounted at the time".

i reopened the bug, and then pointed out that the boottime order
was "Mounting local filesystems" (/etc/init.d/mountall.sh) followed
by "Starting Hotplug" (/etc/init.d/hotplug) and he responded with
"well the kernel can run hotplug at any time".

which seems to me to be a bit daft: if you haven't initialised
the hotplug system with /etc/init.d/hotplug, what's the point of
responding to kernel-triggered hotplug events?

sounds like an issue with the kernel, there, to me.

can anyone on the linux kernel mailing list who has knowledge of
the hotplug kernel stuff (and also who knows or is capable of knowing
how the debian hotplug system works) confirm whether hotplug
events will or will not be triggered by the kernel if you have NOT
run the /etc/init.d/hotplug script?


for example, i'm staring at line 176 of /etc/hotplug/usb.rc (v 1.22)
in version hotplug-20040329-8 and i note that usbdevfs is mounted
on /proc/bus/usb.

a comment at line 148 and 164 makes it clear (ish) that a distro
could mount usbdevfs, which could cause a "partial" initialisation
of the usb hotplugging subsystem, and hotplug events could have
been dropped as a result, or not properly started.

so, distros shouldn't _be_ doing a partial initialisation of the
usb subsystem, basically!

anyway.

from this cursory examination, i would conclude that the
argument used [to justify not moving state information to a
new directory /var/state/hotplug] by the debian maintainer of
hotplug that "the hotplug system may generate events at any
time" is bogus.

e.g. under such circumstances where a distribution has
initialised the usb hotplug system BEFORE /etc/init.d/hotplug start
gets a chance to run "/etc/hotplug/usb.rc start", hotplug
events are likely to get missed - not least because the
filesystem might not have been initialised / mounted (except
the root filesystem maybe) and consequently it's hazardous and
problematic, and the usb.rc script has to reinvent some of
the events it might have missed!!

l.

On Sun, Jul 11, 2004 at 12:40:43PM -0400, Jim McCullough wrote:
> Luke Kenneth Casson Leighton wrote:
> 
> >On Sun, Jul 11, 2004 at 08:50:08PM +1000, Russell Coker wrote:
> > 
> >
> >>On Sat, 10 Jul 2004 06:14, Luke Kenneth Casson Leighton <lkcl@lkcl.net> 
> >>wrote:
> >>   
> >>
> >>>warning warning warning, do not install hotplug version 20040329-9
> >>>from debian unstable because /etc/hotplug/net.rc attempts to write
> >>>to /etc/hotplug/net.enable.
> >>>
> >>>hotplug writing to /etc/ even /etc/hotplug is banned by the [present]
> >>>selinux policy.
> >>>
> >>>i'm raising a bugreport about this one if there isn't one already.
> >>>     
> >>>
> >>That package is broken in other ways.  

Re: [SE/Linux] warning about debian hotplug package 20040329-9!

[Chris Babcock]

So then what if (horror of horrors) somebody puts "/var" on a usb disk
device. (or some other type of device initialized by hotplug?)

> dear selinux and linux kernel,
>
> i am after some assistance in clarifying how hotplug works, with
> a view to solving an issue with SE/Linux where the default
> SE/Linux policy is to deny write permission to /etc/hotplug
> (with good reason) but the hotplug package is presently demanding
> write permission.
>
> a simple request for a change to writing to /var/state/hotplug
> instead has thrown up a number of issues with kernel (2.6.6)
> hotplugging and i would greatly appreciate some confirmation
> and some assistance.
>
>
> i raised a bug with the debian maintainer for hotplug [he
> is known to be, how to describe in a few polite words, a
> control freak.  coming from me, that's saying a lot].
>
> i suggested that the location for files to be written to be moved
> to /var/state/hotplug and he CLOSED the bug with a message saying
> "you can't do that, /var might not be mounted at the time".
>
> i reopened the bug, and then pointed out that the boottime order
> was "Mounting local filesystems" (/etc/init.d/mountall.sh) followed
> by "Starting Hotplug" (/etc/init.d/hotplug) and he responded with
> "well the kernel can run hotplug at any time".
>
> which seems to me to be a bit daft: if you haven't initialised
> the hotplug system with /etc/init.d/hotplug, what's the point of
> responding to kernel-triggered hotplug events?
>
> sounds like an issue with the kernel, there, to me.
>
> can anyone on the linux kernel mailing list who has knowledge of
> the hotplug kernel stuff (and also who knows or is capable of knowing
> how the debian hotplug system works) confirm whether hotplug
> events will or will not be triggered by the kernel if you have NOT
> run the /etc/init.d/hotplug script?
>
>
> for example, i'm staring at line 176 of /etc/hotplug/usb.rc (v 1.22)
> in version hotplug-20040329-8 and i note that usbdevfs is mounted
> on /proc/bus/usb.
>
> a comment at line 148 and 164 makes it clear (ish) that a distro
> could mount usbdevfs, which could cause a "partial" initialisation
> of the usb hotplugging subsystem, and hotplug events could have
> been dropped as a result, or not properly started.
>
> so, distros shouldn't _be_ doing a partial initialisation of the
> usb subsystem, basically!
>
> anyway.
>
> from this cursory examination, i would conclude that the
> argument used [to justify not moving state information to a
> new directory /var/state/hotplug] by the debian maintainer of
> hotplug that "the hotplug system may generate events at any
> time" is bogus.
>
> e.g. under such circumstances where a distribution has
> initialised the usb hotplug system BEFORE /etc/init.d/hotplug start
> gets a chance to run "/etc/hotplug/usb.rc start", hotplug
> events are likely to get missed - not least because the
> filesystem might not have been initialised / mounted (except
> the root filesystem maybe) and consequently it's hazardous and
> problematic, and the usb.rc script has to reinvent some of
> the events it might have missed!!
>
> l.
>
> On Sun, Jul 11, 2004 at 12:40:43PM -0400, Jim McCullough wrote:
>> Luke Kenneth Casson Leighton wrote:
>>
>> >On Sun, Jul 11, 2004 at 08:50:08PM +1000, Russell Coker wrote:
>> >
>> >
>> >>On Sat, 10 Jul 2004 06:14, Luke Kenneth Casson Leighton
>> <lkcl@lkcl.net>
>> >>wrote:
>> >>
>> >>
>> >>>warning warning warning, do not install hotplug version 20040329-9
>> >>>from debian unstable because /etc/hotplug/net.rc attempts to write
>> >>>to /etc/hotplug/net.enable.
>> >>>
>> >>>hotplug writing to /etc/ even /etc/hotplug is banned by the [present]
>> >>>selinux policy.
>> >>>
>> >>>i'm raising a bugreport about this one if there isn't one already.
>> >>>
>> >>>
>> >>That package is broken in other ways.
>
> --
> This message was distributed to subscribers of the selinux mailing list.
> If you no longer wish to subscribe, send mail to majordomo@tycho.nsa.gov
> with
> the words "unsubscribe selinux" without quotes as the message.
>

Re: [SE/Linux] warning about debian hotplug package 20040329-9!

[Luke Kenneth Casson Leighton]

On Mon, Jul 12, 2004 at 02:16:58PM -0700, Chris Babcock wrote:
> So then what if (horror of horrors) somebody puts "/var" on a usb disk
> device. (or some other type of device initialized by hotplug?)
 
 or an nfs-mounted partition.

 the suggestion in that case that i received by one of the hotplug
 developers / people-monitoring-debian-bugs-for-hotplug was that you
 should modify the /etc/hotplug scripts to use /devfs/shm/tmp instead,
 assuming that you have a debian initrd.

 now, on SE/Linux that isn't possible, and the reason isn't entirely
 clear, but i believe that the access permissions to the tmpfs
 created by the debian initrd are such that when the umount tmpfs
 occurs, it actually _does_ unmount it.

 on a standard debian/linux system (no selinux kernel) the initrd
 scripts attempt, amongst other things, to mount various filesystems
 and these are successful, but they are not _un_mounted properly later
 on.

 anyway, i digress: the idea i came up with was that the debian
 package be modified such that it's possible to specify the
 state directory, even if that's one of a list of possible
 locations e.g.  choose one: /etc/hotplug, /etc/hotplug/run,
 /var/run/hotplug, /devfs/shm/tmp, other.

 consequently, if this were to be implemented, at least people mad
 enough to use usb disks or nfs mounted stuff, they'd be able to
 at least get going without having to hack the source of hotplug.

 
> > dear selinux and linux kernel,
> >
> > i am after some assistance in clarifying how hotplug works, with
> > a view to solving an issue with SE/Linux where the default
> > SE/Linux policy is to deny write permission to /etc/hotplug
> > (with good reason) but the hotplug package is presently demanding
> > write permission.
> >
> > a simple request for a change to writing to /var/state/hotplug
> > instead has thrown up a number of issues with kernel (2.6.6)
> > hotplugging and i would greatly appreciate some confirmation
> > and some assistance.