cgroup: Add new capability to allow a process to migrate other tasks between cgroups
diff mbox

Message ID 1481949827-23613-1-git-send-email-john.stultz@linaro.org
State New
Headers show

Commit Message

John Stultz Dec. 17, 2016, 4:43 a.m. UTC
This patch adds CAP_GROUP and logic to allows a process to
migrate other tasks between cgroups.

In Android (where this feature originated), the ActivityManager
tracks various application states (TOP_APP, FOREGROUND,
BACKGROUND, SYSTEM, etc), and then as applications change
states, the SchedPolicy logic will migrate the application tasks
between different cgroups used to control the different
application states (for example, there is a background cpuset
cgroup which can limit background tasks to stay on one low-power
cpu, and the bg_non_interactive cpuctrl cgroup can then further
limit those background tasks to a small percentage of that one
cpu's cpu time).

However, for security reasons, Android doesn't want to make the
system_server (the process that runs the ActivityManager and
SchedPolicy logic), run as root. So in the Android common.git
kernel, they have some logic to allow cgroups to loosen their
permissions so CAP_SYS_NICE tasks can migrate other tasks between
cgroups.

I feel the approach taken there overloads CAP_SYS_NICE a bit much
for non-android environments. Efforts to re-use CAP_SYS_RESOURCE
for this purpose (which Android has since adopted) was also
stymied by concerns about risks from future cgroups that could be
considered "dangerous" by how they might change system semantics.

So to avoid overlapping usage, this patch adds a brand new
process capability flag (CAP_CGROUP), and uses it when checking
if a task can migrate other tasks between cgroups.

I've tested this with AOSP master (though its a bit hacked in as
I still need to properly get the selinux userspace bits aware of
the new capability bit) with selinux set to permissive and it
seems to be working well.

Thoughts and feedback would be appreciated!

(Note, I'm going on holiday break after today, so I may not
respond to feedback immediately, but I figured it would be
better to give folks the chance to review this rather then sit
it for two weeks. I'll resend after the new-year, addressing any
feedback I do get.)

Cc: Tejun Heo <tj@kernel.org>
Cc: Li Zefan <lizefan@huawei.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: cgroups@vger.kernel.org
Cc: Android Kernel Team <kernel-team@android.com>
Cc: Rom Lemarchand <romlem@android.com>
Cc: Colin Cross <ccross@android.com>
Cc: Dmitry Shmidt <dimitrysh@google.com>
Cc: Todd Kjos <tkjos@google.com>
Cc: Christian Poetzsch <christian.potzsch@imgtec.com>
Cc: Amit Pundir <amit.pundir@linaro.org>
Cc: Dmitry Torokhov <dmitry.torokhov@gmail.com>
Cc: Kees Cook <keescook@chromium.org>
Cc: Serge E. Hallyn <serge@hallyn.com>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: linux-api@vger.kernel.org
Cc: Paul Moore <paul@paul-moore.com>
Cc: Stephen Smalley <sds@tycho.nsa.gov>
Cc: Eric Paris <eparis@parisplace.org>
Cc: James Morris <james.l.morris@oracle.com>
Cc: Jeff Vander Stoep <jeffv@google.com>
Cc: linux-security-module@vger.kernel.org
Cc: selinux@tycho.nsa.gov
Signed-off-by: John Stultz <john.stultz@linaro.org>
---
v2: Renamed to just CAP_CGROUP_MIGRATE as reccomended by Tejun
v3: Switched to just using CAP_SYS_RESOURCE as suggested by Michael
v4: Send out properly folded down version of the patch. :P
v5: Switch back to CAP_CGROUP_MIGRATE due to concerns from Andy
v6: Rename to CAP_CGROUP, as it might be used for other purposes
    in the future. Also added selinux mappings for the new cap.
---
 include/uapi/linux/capability.h     | 5 ++++-
 kernel/cgroup.c                     | 3 ++-
 security/selinux/include/classmap.h | 4 ++--
 3 files changed, 8 insertions(+), 4 deletions(-)

Comments

Mickaël Salaün Dec. 17, 2016, 9:06 p.m. UTC | #1
Hi,

If I understand correctly, this patch is intended to add a delegation
feature to cgroup v1, which does not really make sense for the v2
because of the clean cgroup-v2 delegation design. However, this new
capability impact both versions.

As Michael said, capabilities are a limited numbers of silos and we
should try to use existing ones as much as possible but without falling
into the trap of using the same capability for everything (e.g.
CAP_SYS_ADMIN) [1]. The CAP_CGROUP looks a lot like another
CAP_SYS_ADMIN. It is not tied to any particular privilege.

Cgroups are not about resource limitation or any particular access
control, they are just about managing a set of processes, which may then
be subject to resource limitation or access control. This possible
limitations only depend on cgroup controllers or netfilter rules or BPF
programs.

To avoid the current capability issue [1], I think it should be a good
idea to reuse an existing capability (if one make sense) for each class
of constraints a controller/eBPF program/netfilter rule can enforce.
This may looks like CAP_NET_ADMIN (with network namespace handling) for
netfilter and eBPF *socket* programs but CAP_SYS_RESOURCE for the CPU,
memory and IO controllers.
As Tejun said, it will be more complicated to handle such a case, but I
don't see any other solution to keep a meaningful use of capabilities.

However, even if a cgroup does not directly involve a limitation, it may
be used to identify a group of processes for a security critical purpose
(e.g. kill a group of process). It can then make sense to have a
dedicated capability CAP_CGROUP to allow a process *without the right to
write in cgroup.procs* to be allowed to move a process out of its
current cgroup. This is similar to CAP_DAC_OVERRIDE but only for
cgroup/controllers files (but not necessarily sufficient to modify all
cgroups). This does not means that CAP_CGROUP should allow to move any
process from any cgroup. The cgroup_procs_write_permission() should
compose the checks for CAP_CGROUP and/or CAP_SYS_RESOURCE and/or
CAP_SYS_ADMIN depending on the current use of the cgroup (i.e. cgroup
controller, BPF program type, netfilter).

Regards,
 Mickaël


[1] https://forums.grsecurity.net/viewtopic.php?f=7&t=2522


On 17/12/2016 05:43, John Stultz wrote:
> This patch adds CAP_GROUP and logic to allows a process to
> migrate other tasks between cgroups.
> 
> In Android (where this feature originated), the ActivityManager
> tracks various application states (TOP_APP, FOREGROUND,
> BACKGROUND, SYSTEM, etc), and then as applications change
> states, the SchedPolicy logic will migrate the application tasks
> between different cgroups used to control the different
> application states (for example, there is a background cpuset
> cgroup which can limit background tasks to stay on one low-power
> cpu, and the bg_non_interactive cpuctrl cgroup can then further
> limit those background tasks to a small percentage of that one
> cpu's cpu time).
> 
> However, for security reasons, Android doesn't want to make the
> system_server (the process that runs the ActivityManager and
> SchedPolicy logic), run as root. So in the Android common.git
> kernel, they have some logic to allow cgroups to loosen their
> permissions so CAP_SYS_NICE tasks can migrate other tasks between
> cgroups.
> 
> I feel the approach taken there overloads CAP_SYS_NICE a bit much
> for non-android environments. Efforts to re-use CAP_SYS_RESOURCE
> for this purpose (which Android has since adopted) was also
> stymied by concerns about risks from future cgroups that could be
> considered "dangerous" by how they might change system semantics.
> 
> So to avoid overlapping usage, this patch adds a brand new
> process capability flag (CAP_CGROUP), and uses it when checking
> if a task can migrate other tasks between cgroups.
> 
> I've tested this with AOSP master (though its a bit hacked in as
> I still need to properly get the selinux userspace bits aware of
> the new capability bit) with selinux set to permissive and it
> seems to be working well.
> 
> Thoughts and feedback would be appreciated!
> 
> (Note, I'm going on holiday break after today, so I may not
> respond to feedback immediately, but I figured it would be
> better to give folks the chance to review this rather then sit
> it for two weeks. I'll resend after the new-year, addressing any
> feedback I do get.)
> 
> Cc: Tejun Heo <tj@kernel.org>
> Cc: Li Zefan <lizefan@huawei.com>
> Cc: Jonathan Corbet <corbet@lwn.net>
> Cc: cgroups@vger.kernel.org
> Cc: Android Kernel Team <kernel-team@android.com>
> Cc: Rom Lemarchand <romlem@android.com>
> Cc: Colin Cross <ccross@android.com>
> Cc: Dmitry Shmidt <dimitrysh@google.com>
> Cc: Todd Kjos <tkjos@google.com>
> Cc: Christian Poetzsch <christian.potzsch@imgtec.com>
> Cc: Amit Pundir <amit.pundir@linaro.org>
> Cc: Dmitry Torokhov <dmitry.torokhov@gmail.com>
> Cc: Kees Cook <keescook@chromium.org>
> Cc: Serge E. Hallyn <serge@hallyn.com>
> Cc: Andy Lutomirski <luto@amacapital.net>
> Cc: linux-api@vger.kernel.org
> Cc: Paul Moore <paul@paul-moore.com>
> Cc: Stephen Smalley <sds@tycho.nsa.gov>
> Cc: Eric Paris <eparis@parisplace.org>
> Cc: James Morris <james.l.morris@oracle.com>
> Cc: Jeff Vander Stoep <jeffv@google.com>
> Cc: linux-security-module@vger.kernel.org
> Cc: selinux@tycho.nsa.gov
> Signed-off-by: John Stultz <john.stultz@linaro.org>
> ---
> v2: Renamed to just CAP_CGROUP_MIGRATE as reccomended by Tejun
> v3: Switched to just using CAP_SYS_RESOURCE as suggested by Michael
> v4: Send out properly folded down version of the patch. :P
> v5: Switch back to CAP_CGROUP_MIGRATE due to concerns from Andy
> v6: Rename to CAP_CGROUP, as it might be used for other purposes
>     in the future. Also added selinux mappings for the new cap.
> ---
>  include/uapi/linux/capability.h     | 5 ++++-
>  kernel/cgroup.c                     | 3 ++-
>  security/selinux/include/classmap.h | 4 ++--
>  3 files changed, 8 insertions(+), 4 deletions(-)
> 
> diff --git a/include/uapi/linux/capability.h b/include/uapi/linux/capability.h
> index 49bc062..726f767 100644
> --- a/include/uapi/linux/capability.h
> +++ b/include/uapi/linux/capability.h
> @@ -349,8 +349,11 @@ struct vfs_cap_data {
>  
>  #define CAP_AUDIT_READ		37
>  
> +/* Allow migration of other tasks between cgroups */
>  
> -#define CAP_LAST_CAP         CAP_AUDIT_READ
> +#define CAP_CGROUP		38
> +
> +#define CAP_LAST_CAP         CAP_CGROUP
>  
>  #define cap_valid(x) ((x) >= 0 && (x) <= CAP_LAST_CAP)
>  
> diff --git a/kernel/cgroup.c b/kernel/cgroup.c
> index 2ee9ec3..8b42ae3 100644
> --- a/kernel/cgroup.c
> +++ b/kernel/cgroup.c
> @@ -2856,7 +2856,8 @@ static int cgroup_procs_write_permission(struct task_struct *task,
>  	 */
>  	if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) &&
>  	    !uid_eq(cred->euid, tcred->uid) &&
> -	    !uid_eq(cred->euid, tcred->suid))
> +	    !uid_eq(cred->euid, tcred->suid) &&
> +	    !ns_capable(tcred->user_ns, CAP_CGROUP))
>  		ret = -EACCES;
>  
>  	if (!ret && cgroup_on_dfl(dst_cgrp)) {
> diff --git a/security/selinux/include/classmap.h b/security/selinux/include/classmap.h
> index e2d4ad3a..ee8c1ed 100644
> --- a/security/selinux/include/classmap.h
> +++ b/security/selinux/include/classmap.h
> @@ -22,9 +22,9 @@
>  	    "audit_control", "setfcap"
>  
>  #define COMMON_CAP2_PERMS  "mac_override", "mac_admin", "syslog", \
> -		"wake_alarm", "block_suspend", "audit_read"
> +		"wake_alarm", "block_suspend", "audit_read", "cgroup"
>  
> -#if CAP_LAST_CAP > CAP_AUDIT_READ
> +#if CAP_LAST_CAP > CAP_CGROUP
>  #error New capability defined, please update COMMON_CAP2_PERMS.
>  #endif
>  
>
Tejun Heo Dec. 19, 2016, 1:11 p.m. UTC | #2
Hello,

On Sat, Dec 17, 2016 at 10:06:51PM +0100, Mickaël Salaün wrote:
> If I understand correctly, this patch is intended to add a delegation
> feature to cgroup v1, which does not really make sense for the v2

It's more about upstreaming a workaround for android somewhat like
including binder into kernel.  It isn't adding actual cgroup
delegation to v1.  It's just splitting a small piece of CAP_SYS_ADMIN
to accomodate what android has been doing.

> because of the clean cgroup-v2 delegation design. However, this new
> capability impact both versions.

In the same way but it's not about cgroup delegation.  It's just
allowing splitting up CAP_SYS_ADMIN so that "no extra restrictions on
cgroup" can be given away in a safer way.

> However, even if a cgroup does not directly involve a limitation, it may
> be used to identify a group of processes for a security critical purpose
> (e.g. kill a group of process). It can then make sense to have a
> dedicated capability CAP_CGROUP to allow a process *without the right to
> write in cgroup.procs* to be allowed to move a process out of its
> current cgroup. This is similar to CAP_DAC_OVERRIDE but only for
> cgroup/controllers files (but not necessarily sufficient to modify all
> cgroups). This does not means that CAP_CGROUP should allow to move any
> process from any cgroup. The cgroup_procs_write_permission() should
> compose the checks for CAP_CGROUP and/or CAP_SYS_RESOURCE and/or
> CAP_SYS_ADMIN depending on the current use of the cgroup (i.e. cgroup
> controller, BPF program type, netfilter).

There's no reason to invent a whole new set of security policies for
cgroup.  It already got one which follows the filesystem permissions
with some extra restrictions.  The CAP split is purely to accomodate
android and that's it.  If that isn't good enough a reason, then
android should just keep carrying the patches it needs.  This doesn't
justify bolting on another permission model on cgroup in any way.

Thanks.

Patch
diff mbox

diff --git a/include/uapi/linux/capability.h b/include/uapi/linux/capability.h
index 49bc062..726f767 100644
--- a/include/uapi/linux/capability.h
+++ b/include/uapi/linux/capability.h
@@ -349,8 +349,11 @@  struct vfs_cap_data {
 
 #define CAP_AUDIT_READ		37
 
+/* Allow migration of other tasks between cgroups */
 
-#define CAP_LAST_CAP         CAP_AUDIT_READ
+#define CAP_CGROUP		38
+
+#define CAP_LAST_CAP         CAP_CGROUP
 
 #define cap_valid(x) ((x) >= 0 && (x) <= CAP_LAST_CAP)
 
diff --git a/kernel/cgroup.c b/kernel/cgroup.c
index 2ee9ec3..8b42ae3 100644
--- a/kernel/cgroup.c
+++ b/kernel/cgroup.c
@@ -2856,7 +2856,8 @@  static int cgroup_procs_write_permission(struct task_struct *task,
 	 */
 	if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) &&
 	    !uid_eq(cred->euid, tcred->uid) &&
-	    !uid_eq(cred->euid, tcred->suid))
+	    !uid_eq(cred->euid, tcred->suid) &&
+	    !ns_capable(tcred->user_ns, CAP_CGROUP))
 		ret = -EACCES;
 
 	if (!ret && cgroup_on_dfl(dst_cgrp)) {
diff --git a/security/selinux/include/classmap.h b/security/selinux/include/classmap.h
index e2d4ad3a..ee8c1ed 100644
--- a/security/selinux/include/classmap.h
+++ b/security/selinux/include/classmap.h
@@ -22,9 +22,9 @@ 
 	    "audit_control", "setfcap"
 
 #define COMMON_CAP2_PERMS  "mac_override", "mac_admin", "syslog", \
-		"wake_alarm", "block_suspend", "audit_read"
+		"wake_alarm", "block_suspend", "audit_read", "cgroup"
 
-#if CAP_LAST_CAP > CAP_AUDIT_READ
+#if CAP_LAST_CAP > CAP_CGROUP
 #error New capability defined, please update COMMON_CAP2_PERMS.
 #endif