diff options
| -rw-r--r-- | kernel/cgroup.c | 11 |
1 files changed, 7 insertions, 4 deletions
diff --git a/kernel/cgroup.c b/kernel/cgroup.c index 5eb20cd..f873c46 100644 --- a/kernel/cgroup.c +++ b/kernel/cgroup.c @@ -1694,6 +1694,7 @@ static struct dentry *cgroup_mount(struct file_system_type *fs_type, struct dentry *dentry; int ret; int i; + bool new_sb; /* * The first time anyone tries to mount a cgroup, enable the list @@ -1784,7 +1785,7 @@ static struct dentry *cgroup_mount(struct file_system_type *fs_type, * path is super cold. Let's just sleep a bit and retry. */ pinned_sb = kernfs_pin_sb(root->kf_root, NULL); - if (IS_ERR_OR_NULL(pinned_sb) || + if (IS_ERR(pinned_sb) || !percpu_ref_tryget_live(&root->cgrp.self.refcnt)) { mutex_unlock(&cgroup_mutex); if (!IS_ERR_OR_NULL(pinned_sb)) @@ -1830,16 +1831,18 @@ out_free: return ERR_PTR(ret); dentry = kernfs_mount(fs_type, flags, root->kf_root, - CGROUP_SUPER_MAGIC, NULL); - if (IS_ERR(dentry) || pinned_sb) + CGROUP_SUPER_MAGIC, &new_sb); + if (IS_ERR(dentry) || !new_sb) cgroup_put(&root->cgrp); /* * If @pinned_sb, we're reusing an existing root and holding an * extra ref on its sb. Mount is complete. Put the extra ref. */ - if (pinned_sb) + if (pinned_sb) { + WARN_ON(new_sb); deactivate_super(pinned_sb); + } return dentry; } |
