The original model of device_cgroup is having a whitelist where all the
allowed devices are listed. The problem with this approach is that is
impossible to have the case of allowing everything but few devices.
The reason for that lies in the way the whitelist is handled internally:
since there's only a whitelist, the "all devices" entry would have to be
removed and replaced by the entire list of possible devices but the ones
that are being denied. Since dev_t is 32 bits long, representing the allowed
devices as a bitfield is not memory efficient.
This patch replaces the "whitelist" by a "exceptions" list and the default
policy is kept as "deny_all" variable in dev_cgroup structure.
The current interface determines that whenever "a" is written to devices.allow
or devices.deny, the entry masking all devices will be added or removed,
respectively. This behavior is kept and it's what will determine the default
policy:
# cat devices.list
a *:* rwm
# echo a >devices.deny
# cat devices.list
# echo a >devices.allow
# cat devices.list
a *:* rwm
The interface is also preserved. For example, if one wants to block only access
to /dev/null:
# ls -l /dev/null
crw-rw-rw- 1 root root 1, 3 Jul 24 16:17 /dev/null
# echo a >devices.allow
# echo "c 1:3 rwm" >devices.deny
# cat /dev/null
cat: /dev/null: Operation not permitted
# echo >/dev/null
bash: /dev/null: Operation not permitted
mknod /tmp/null c 1 3
mknod: ‘/tmp/null’: Operation not permitted
# echo "c 1:3 r" >devices.allow
# cat /dev/null
# echo >/dev/null
bash: /dev/null: Operation not permitted
mknod /tmp/null c 1 3
mknod: ‘/tmp/null’: Operation not permitted
# echo "c 1:3 rw" >devices.allow
# echo >/dev/null
# cat /dev/null
# mknod /tmp/null c 1 3
mknod: ‘/tmp/null’: Operation not permitted
# echo "c 1:3 rwm" >devices.allow
# echo >/dev/null
# cat /dev/null
# mknod /tmp/null c 1 3
#
device_cgroup.c | 371 ++++++++++++++++++++++++++++++++------------------------
1 file changed, 212 insertions(+), 159 deletions(-)
On Thu, 09 Aug 2012 15:04:14 -0400
[email protected] wrote:
> The original model of device_cgroup is having a whitelist where all the
> allowed devices are listed. The problem with this approach is that is
> impossible to have the case of allowing everything but few devices.
>
> The reason for that lies in the way the whitelist is handled internally:
> since there's only a whitelist, the "all devices" entry would have to be
> removed and replaced by the entire list of possible devices but the ones
> that are being denied. Since dev_t is 32 bits long, representing the allowed
> devices as a bitfield is not memory efficient.
>
> This patch replaces the "whitelist" by a "exceptions" list and the default
> policy is kept as "deny_all" variable in dev_cgroup structure.
The patches look reasonable to me. I suggest you resend them after
kernel summit and cc a few additional people: Serge Hallyn, Pavel
Emelyanov and James Morris.
On Thu, Aug 16, 2012 at 01:08:44PM -0700, Andrew Morton wrote:
> On Thu, 09 Aug 2012 15:04:14 -0400
> [email protected] wrote:
>
> > The original model of device_cgroup is having a whitelist where all the
> > allowed devices are listed. The problem with this approach is that is
> > impossible to have the case of allowing everything but few devices.
> >
> > The reason for that lies in the way the whitelist is handled internally:
> > since there's only a whitelist, the "all devices" entry would have to be
> > removed and replaced by the entire list of possible devices but the ones
> > that are being denied. Since dev_t is 32 bits long, representing the allowed
> > devices as a bitfield is not memory efficient.
> >
> > This patch replaces the "whitelist" by a "exceptions" list and the default
> > policy is kept as "deny_all" variable in dev_cgroup structure.
>
> The patches look reasonable to me. I suggest you resend them after
> kernel summit and cc a few additional people: Serge Hallyn, Pavel
> Emelyanov and James Morris.
thanks, will do. will include the checkpatch fixes as well.
--
Aristeu