Re: [PATCH 3/4] Add a back-pointer from struct cg_cgroup_link to struct cgroup

[Li Zefan <lizf@cn.fujitsu.com>]

Paul Menage wrote:
> Add a back-pointer from struct cg_cgroup_link to struct cgroup
> 
> Currently the cgroups code makes the assumption that the subsystem
> pointers in a struct css_set uniquely identify the hierarchy->cgroup
> mappings associated with the css_set; and there's no way to directly
> identify the associated set of cgroups other than by indirecting
> through the appropriate subsystem state pointers.
> 
> This patch removes the need for that assumption by adding a
> back-pointer from struct cg_cgroup_link object to its associated
> cgroup; this allows the set of cgroups to be determined by traversing
> the cg_links list in the struct css_set.
> 
> Signed-off-by: Paul Menage <menage@google.com>
> 

Reviewed-by: Li Zefan <lizf@cn.fujitsu.com>

...

> +#define MAX_TASKS_SHOWN_PER_CSS 25
> +static int cgroup_css_links_read(struct cgroup *cont,
> +				 struct cftype *cft,
> +				 struct seq_file *seq)
> +{
> +	struct cg_cgroup_link *link;
> +

I think we need to cgroup_enable_task_cg_lists():

	if (!use_task_css_set_links)
		cgroup_enable_task_cg_lists();

Otherwise we'll see no tasks when reading this debug file
if use_task_css_set_links == 0.

> +	read_lock(&css_set_lock);
> +	list_for_each_entry(link, &cont->css_sets, cgrp_link_list) {
> +		struct css_set *cg = link->cg;
> +		struct task_struct *task;
> +		int count = 0;
> +		seq_printf(seq, "css_set %p\n", cg);
> +		list_for_each_entry(task, &cg->tasks, cg_list) {
> +			if (count++ > MAX_TASKS_SHOWN_PER_CSS) {
> +				seq_puts(seq, "  ...\n");
> +				break;
> +			} else {
> +				seq_printf(seq, "  task %d\n",
> +					   task_pid_vnr(task));
> +			}
> +		}
> +	}
> +	read_unlock(&css_set_lock);
> +	return 0;
> +}