Re: [PATCH 1/3] perf: add PERF_RECORD_NAMESPACES to include namespaces related info

[Peter Zijlstra <peterz@infradead.org>]

On Thu, Nov 10, 2016 at 05:08:06PM +0530, Hari Bathini wrote:

> diff --git a/include/uapi/linux/perf_event.h b/include/uapi/linux/perf_event.h
> index c66a485..575aed6 100644
> --- a/include/uapi/linux/perf_event.h
> +++ b/include/uapi/linux/perf_event.h
> @@ -344,7 +344,8 @@ struct perf_event_attr {
>  				use_clockid    :  1, /* use @clockid for time fields */
>  				context_switch :  1, /* context switch data */
>  				write_backward :  1, /* Write ring buffer from end to beginning */
> -				__reserved_1   : 36;
> +				namespaces     :  1, /* include namespaces data */
> +				__reserved_1   : 35;
>  
>  	union {
>  		__u32		wakeup_events;	  /* wakeup every n events */
> @@ -862,6 +863,24 @@ enum perf_event_type {
>  	 */
>  	PERF_RECORD_SWITCH_CPU_WIDE		= 15,
>  
> +	/*
> +	 * struct {
> +	 *	struct perf_event_header	header;
> +	 *
> +	 *	u32				pid, tid;
> +	 *	u64				time;
> +	 *	u32				uts_ns_inum;
> +	 *	u32				ipc_ns_inum;
> +	 *	u32				mnt_ns_inum;
> +	 *	u32				pid_ns_inum;
> +	 *	u32				net_ns_inum;
> +	 *	u32				cgroup_ns_inum;
> +	 *	u32				user_ns_inum;
> +	 * 	struct sample_id		sample_id;
> +	 * };
> +	 */
> +	PERF_RECORD_NAMESPACES			= 16,

So this format is not extensible, that is, if someone adds yet another
namespace, we'll need to introduce PERF_RECORD_NAMESPACES2.

Is there a 'natural' and exposed namespace index that we can use to
change it like:

	u32	nr_nss;
	u32	namespace[nr_nss];

?

> +static void perf_event_namespaces_output(struct perf_event *event,
> +					 void *data)
> +{
> +	struct perf_namespaces_event *namespaces_event = data;
> +	struct perf_output_handle handle;
> +	struct perf_sample_data sample;
> +	int size = namespaces_event->event_id.header.size;
> +	struct nsproxy *nsproxy;
> +	int ret;
> +
> +	if (!perf_event_namespaces_match(event))
> +		return;
> +
> +	perf_event_header__init_id(&namespaces_event->event_id.header,
> +				   &sample, event);
> +	ret = perf_output_begin(&handle, event,
> +				namespaces_event->event_id.header.size);
> +
> +	if (ret)
> +		goto out;

If you were to introduce:

	struct ns_event_id *ei = &namespace_event->event_id;

> +
> +	namespaces_event->event_id.pid = perf_event_pid(event,
> +							namespaces_event->task);
> +	namespaces_event->event_id.tid = perf_event_tid(event,
> +							namespaces_event->task);
> +
> +	if (namespaces_event->task != current)
> +		task_lock(namespaces_event->task);
> +
> +	nsproxy = namespaces_event->task->nsproxy;
> +	if (nsproxy != NULL) {
> +		namespaces_event->event_id.uts_ns_inum =
> +					nsproxy->uts_ns->ns.inum;
> +#ifdef CONFIG_IPC_NS
> +		namespaces_event->event_id.ipc_ns_inum =
> +					nsproxy->ipc_ns->ns.inum;
> +#endif
> +		namespaces_event->event_id.mnt_ns_inum =
> +					nsproxy->mnt_ns->ns.inum;
> +		namespaces_event->event_id.pid_ns_inum =
> +					nsproxy->pid_ns_for_children->ns.inum;
> +#ifdef CONFIG_NET
> +		namespaces_event->event_id.net_ns_inum =
> +					nsproxy->net_ns->ns.inum;
> +#endif
> +#ifdef CONFIG_CGROUPS
> +		namespaces_event->event_id.cgroup_ns_inum =
> +					nsproxy->cgroup_ns->ns.inum;
> +#endif
> +	}
> +
> +	namespaces_event->event_id.user_ns_inum =
> +			__task_cred(namespaces_event->task)->user_ns->ns.inum;

You can do s/namespace_event->event_id./ei->/ which is tons shorter and
would result in less wrapping of lines and generally improve
readability.

> +
> +	if (namespaces_event->task != current)
> +		task_unlock(namespaces_event->task);
> +
> +	namespaces_event->event_id.time = perf_event_clock(event);
> +
> +	perf_output_put(&handle, namespaces_event->event_id);
> +
> +	perf_event__output_id_sample(event, &handle, &sample);
> +
> +	perf_output_end(&handle);
> +out:
> +	namespaces_event->event_id.header.size = size;
> +}
> +
> +void perf_event_namespaces(struct task_struct *task)
> +{
> +	struct perf_namespaces_event namespaces_event;
> +
> +	if (!atomic_read(&nr_namespaces_events))
> +		return;
> +
> +	namespaces_event = (struct perf_namespaces_event){
> +		.task	= task,
> +		.event_id  = {
> +			.header = {
> +				.type = PERF_RECORD_NAMESPACES,
> +				.misc = 0,
> +				.size = sizeof(namespaces_event.event_id),
> +			},
> +			/* .pid */
> +			/* .tid */
> +			/* .time */
> +			/* .uts_ns_inum */
> +			/* .ipc_ns_inum */
> +			/* .mnt_ns_inum */
> +			/* .pid_ns_inum */
> +			/* .net_ns_inum */
> +			/* .cgroup_ns_inum */
> +			/* .user_ns_inum */
> +		},
> +	};
> +
> +	perf_iterate_sb(perf_event_namespaces_output,
> +			&namespaces_event,
> +			NULL);
> +}

> diff --git a/kernel/fork.c b/kernel/fork.c
> index 997ac1d..3faca3d 100644
> --- a/kernel/fork.c
> +++ b/kernel/fork.c
> @@ -1818,6 +1818,7 @@ static __latent_entropy struct task_struct *copy_process(
>  	cgroup_post_fork(p);
>  	threadgroup_change_end(current);
>  	perf_event_fork(p);
> +	perf_event_namespaces(p);

I would much prefer calling perf_event_namespace() from
perf_event_fork() and reduce the external interface.