Re: [PATCH 3/4] net: use cgroup_attach method to migrate socket priotiy and classid

[Al Viro <viro@ZenIV.linux.org.uk>]

On Wed, Dec 21, 2011 at 09:39:49AM -0500, Neil Horman wrote:
> +static void cgrp_attach(struct cgroup_subsys *ss, struct cgroup *cgrp,
> +			struct cgroup *old_cgrp, struct task_struct *tsk)
> +{
> +	int fd, err;
> +	struct files_struct *files;
> +	struct file *file;
> +	struct socket *sock;
> +	struct cgroup_netprio_state *old, *new;
> +	pid_t task_pid;
> + 
> +	files = get_files_struct(tsk);
> +	task_pid = task_pid_nr(tsk);
> + 
> +	old = cgrp_netprio_state(old_cgrp);
> +	new = cgrp_netprio_state(cgrp);
> + 
> +	spin_lock(&files->file_lock);
> +	for (fd=0; fd < files_fdtable(files)->max_fds; fd++) {
> +		file = fcheck_files(files, fd);
> +		if (!file)
> +			continue;
> +		sock = sock_from_file(file, &err);
> +		if (!sock)
> +			continue;
> +		if (!sock->sk)
> +			continue;
> +		if (sock->sk->sk_cgrp_owner == task_pid)
> +			sock->sk->sk_cgrp_prioidx = new->prioidx;

This looks bogus *and* racy; what about e.g. files currently in SCM_RIGHTS
datagrams?