On Wed, Jan 08, 2020 at 07:10:49PM +0100, Christian Brauner wrote:

On Wed, Jan 08, 2020 at 05:01:02PM +0100, Michal Koutný wrote:

quoted

On Mon, Dec 23, 2019 at 07:15:03AM +0100, Christian Brauner [off-list ref] wrote:

quoted

This adds support for creating a process in a different cgroup than its
parent.

Binding fork and migration together looks useful.

quoted

--- a/kernel/cgroup/cgroup.c
+++ b/kernel/cgroup/cgroup.c

@@ -5882,21 +5882,176 @@ void cgroup_fork(struct task_struct *child)
 	INIT_LIST_HEAD(&child->cg_list);

Just a nitpick, I noticed the comment for cgroup_fork should be updated
too (generic migration happens in cgroup_post_fork).

Thanks.

quoted

quoted

--- a/kernel/fork.c
+++ b/kernel/fork.c

[...]

@@ -2279,8 +2278,7 @@ static __latent_entropy struct task_struct *copy_process(
 	write_unlock_irq(&tasklist_lock);
 
 	proc_fork_connector(p);
-	cgroup_post_fork(p);
-	cgroup_threadgroup_change_end(current);
+	cgroup_post_fork(current, p, args);

I can see that when CLONE_INTO_CGROUP | CLONE_NEWCGROUP is passed, then
the child's cgroup NS will be rooted at parent's css set
(copy_namespaces precedes cgroup_post_fork).

Wouldn't it make better sense if this flags combination resulted in
child's NS rooted in its css set?

I need to take a closer look but it sounds like we should move the
copying of the cgroup namespace to a later point; but again I need to
look into this.

Ok, this is way simpler I think, we just set the root_cset of the new
cgroup namespace to the child's cset in cgroup_post_fork() if
CLONE_NEWCGROUP is requested.

Christian