cgroup: Honor caller's cgroup NS when resolving path
authorMichal Koutný <mkoutny@suse.com>
Fri, 26 Aug 2022 16:52:35 +0000 (18:52 +0200)
committerTejun Heo <tj@kernel.org>
Fri, 26 Aug 2022 20:56:26 +0000 (10:56 -1000)
cgroup_get_from_path() is not widely used function. Its callers presume
the path is resolved under cgroup namespace. (There is one caller
currently and resolving in init NS won't make harm (netfilter). However,
future users may be subject to different effects when resolving
globally.)
Since, there's currently no use for the global resolution, modify the
existing function to take cgroup NS into account.

Fixes: a79a908fd2b0 ("cgroup: introduce cgroup namespaces")
Signed-off-by: Michal Koutný <mkoutny@suse.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
kernel/cgroup/cgroup.c

index 627ff0f..1c7ab41 100644 (file)
@@ -6602,8 +6602,12 @@ struct cgroup *cgroup_get_from_path(const char *path)
 {
        struct kernfs_node *kn;
        struct cgroup *cgrp = ERR_PTR(-ENOENT);
+       struct cgroup *root_cgrp;
 
-       kn = kernfs_walk_and_get(cgrp_dfl_root.cgrp.kn, path);
+       spin_lock_irq(&css_set_lock);
+       root_cgrp = current_cgns_cgroup_from_root(&cgrp_dfl_root);
+       kn = kernfs_walk_and_get(root_cgrp->kn, path);
+       spin_unlock_irq(&css_set_lock);
        if (!kn)
                goto out;