[PATCH for 4.4.y-cip] cgroup-v1: Require capabilities to set release_agent

[PATCH for 4.4.y-cip] cgroup-v1: Require capabilities to set release_agent

[Masami Ichikawa]

From: "Eric W. Biederman" <ebiederm@xmission.com>

commit 24f6008564183aa120d07c03d9289519c2fe02af upstream.

The cgroup release_agent is called with call_usermodehelper.  The function
call_usermodehelper starts the release_agent with a full set fo capabilities.
Therefore require capabilities when setting the release_agaent.

Reported-by: Tabitha Sable <tabitha.c.sable@gmail.com>
Tested-by: Tabitha Sable <tabitha.c.sable@gmail.com>
Fixes: 81a6a5cdd2c5 ("Task Control Groups: automatic userspace notification of idle cgroups")
Cc: stable@vger.kernel.org # v2.6.24+
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
[mkoutny: Adjust for pre-fs_context, duplicate mount/remount check, drop log messages.]
Acked-by: Michal Koutný <mkoutny@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[masami: Backport patch from 4.9. Adjust to use current_user_ns() to get current user_ns.
Fix conflict in cgroup_release_agent_write().]
Reference: CVE-2022-0492
Signed-off-by: Masami Ichikawa(CIP) <masami.ichikawa@cybertrust.co.jp>
---
 kernel/cgroup.c | 25 +++++++++++++++++++++++++
 1 file changed, 25 insertions(+)

diff --git a/kernel/cgroup.c b/kernel/cgroup.c
index 1f5e7dcbfd40..af521a3da21c 100644
--- a/kernel/cgroup.c
+++ b/kernel/cgroup.c
@@ -1786,6 +1786,13 @@ static int cgroup_remount(struct kernfs_root *kf_root, int *flags, char *data)
 		pr_warn("option changes via remount are deprecated (pid=%d comm=%s)\n",
 			task_tgid_nr(current), current->comm);
 
+	/* See cgroup_mount release_agent handling */
+	if (opts.release_agent &&
+	    ((current_user_ns() != &init_user_ns) || !capable(CAP_SYS_ADMIN))) {
+		ret = -EINVAL;
+		goto out_unlock;
+	}
+
 	added_mask = opts.subsys_mask & ~root->subsys_mask;
 	removed_mask = root->subsys_mask & ~opts.subsys_mask;
 
@@ -2135,6 +2142,16 @@ static struct dentry *cgroup_mount(struct file_system_type *fs_type,
 		goto out_unlock;
 	}
 
+	/*
+	 * Release agent gets called with all capabilities,
+	 * require capabilities to set release agent.
+	 */
+	if (opts.release_agent &&
+	    ((current_user_ns() != &init_user_ns) || !capable(CAP_SYS_ADMIN))) {
+		ret = -EINVAL;
+		goto out_unlock;
+	}
+
 	root = kzalloc(sizeof(*root), GFP_KERNEL);
 	if (!root) {
 		ret = -ENOMEM;
@@ -2839,6 +2856,14 @@ static ssize_t cgroup_release_agent_write(struct kernfs_open_file *of,
 
 	BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX);
 
+	/*
+	 * Release agent gets called with all capabilities,
+	 * require capabilities to set release agent.
+	 */
+	if ((of->file->f_cred->user_ns != &init_user_ns) ||
+		!capable(CAP_SYS_ADMIN))
+			return -EPERM;
+
 	cgrp = cgroup_kn_lock_live(of->kn);
 	if (!cgrp)
 		return -ENODEV;
-- 
2.35.1

Re: [PATCH for 4.4.y-cip] cgroup-v1: Require capabilities to set release_agent

[Michal Koutný]

Thanks for sharing this Masami, I've been looking into pre-cgroup_ns
backport too.

On Wed, Feb 16, 2022 at 08:40:37AM +0900, Masami Ichikawa <masami256@gmail.com> wrote:
> [masami: Backport patch from 4.9. Adjust to use current_user_ns() to get current user_ns.
> Fix conflict in cgroup_release_agent_write().]

The condition to allow modifying release_agent is two-fold:
a) caller is capabable(CAP_SYS_ADMIN),
b) cgroup_ns is owned by init_user_ns.

In pre-cgroup_ns kernels, it is IMO safer to consider all (=the only)
cgroup_ns owned by init_user_ns.

So the (positive) condition translates into capable(CAP_SYS_ADMIN) only.

[
Additionally, there's invariant/implication
  capable(CAP_XXX) -> (current_user_ns() == &init_user_ns) ,
so the expression
  (current_user_ns() != &init_user_ns) || !capable(CAP_SYS_ADMIN)
simplifies to
  !capable(CAP_SYS_ADMIN) .
]


> @@ -2839,6 +2856,14 @@ static ssize_t cgroup_release_agent_write(struct kernfs_open_file *of,
>  
>  	BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX);
>  
> +	/*
> +	 * Release agent gets called with all capabilities,
> +	 * require capabilities to set release agent.
> +	 */
> +	if ((of->file->f_cred->user_ns != &init_user_ns) ||
> +		!capable(CAP_SYS_ADMIN))
> +			return -EPERM;
> +

Following the reasoning above, the check simplifies too but it should be
be against the opener, not the writer:
  file_ns_capable(of->file, &init_user_ns, CAP_SYS_ADMIN)

(It seems to be to be incorrect even in the original commit.
So I'll send a patch there to rectify that.)


Michal

Re: [PATCH for 4.4.y-cip] cgroup-v1: Require capabilities to set release_agent

[Masami Ichikawa]

On Wed, Feb 16, 2022 at 9:11 PM Michal Koutný <mkoutny@suse.com> wrote:
>
> Thanks for sharing this Masami, I've been looking into pre-cgroup_ns
> backport too.
>

Thank you for the review.

> On Wed, Feb 16, 2022 at 08:40:37AM +0900, Masami Ichikawa <masami256@gmail.com> wrote:
> > [masami: Backport patch from 4.9. Adjust to use current_user_ns() to get current user_ns.
> > Fix conflict in cgroup_release_agent_write().]
>
> The condition to allow modifying release_agent is two-fold:
> a) caller is capabable(CAP_SYS_ADMIN),
> b) cgroup_ns is owned by init_user_ns.
>
> In pre-cgroup_ns kernels, it is IMO safer to consider all (=the only)
> cgroup_ns owned by init_user_ns.
>
> So the (positive) condition translates into capable(CAP_SYS_ADMIN) only.
>

I see.  Thank you for the details.
If we treat cgroup_ns as only cgroup_ns that are owned by
init_user_ns,  we don't have to have an extra check.

> [
> Additionally, there's invariant/implication
>   capable(CAP_XXX) -> (current_user_ns() == &init_user_ns) ,
> so the expression
>   (current_user_ns() != &init_user_ns) || !capable(CAP_SYS_ADMIN)
> simplifies to
>   !capable(CAP_SYS_ADMIN) .
> ]
>
>
> > @@ -2839,6 +2856,14 @@ static ssize_t cgroup_release_agent_write(struct kernfs_open_file *of,
> >
> >       BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX);
> >
> > +     /*
> > +      * Release agent gets called with all capabilities,
> > +      * require capabilities to set release agent.
> > +      */
> > +     if ((of->file->f_cred->user_ns != &init_user_ns) ||
> > +             !capable(CAP_SYS_ADMIN))
> > +                     return -EPERM;
> > +
>
> Following the reasoning above, the check simplifies too but it should be
> be against the opener, not the writer:
>   file_ns_capable(of->file, &init_user_ns, CAP_SYS_ADMIN)
>

Thanks. I got it.

> (It seems to be to be incorrect even in the original commit.
> So I'll send a patch there to rectify that.)
>
>
> Michal
>
>

Regards,
-- 
/**
* Masami Ichikawa
* personal: masami256@gmail.com
* fedora project: masami@fedoraproject.org
*/