[v2] openvswitch: Add support to count upall packets

  Added the total number of upcalls and the total of upcall failures.
Due to ovs-userspace do not support NLA_NESTED, here still use the
"struct ovs_vport_upcall_stats"

Thank you
wangchuanlei

On 14 Sep 2022, at 14:14, wangchuanlei wrote:

> Add support to count upcall packets on every interface.
> I always encounter high cpu use of ovs-vswictchd, this help to check
> which interface send too many packets without open vlog/set switch,
> and have no influence on datapath.

Hi,

I did not do a full review, but I think we should not try to make the
 same mistake as before and embed a structure inside a netlink message.

You are adding “struct ovs_vport_upcall_stats” but in theory,
 you could have just added the new entry to “ovs_vport_stats”.
 But this is breaking userspace as it expects an exact structure size :(

So I think the right approach would be to have “
OVS_VPORT_ATTR_UPCALL_STATS” be an NLA_NESTED type, and have
individual stat attributes as NLA_U64 (or whatever type you need).

What is also confusing is that you use upcall_packets in
 ovs_vport_upcall_stats, which to me are the total of up calls,
but you called it n_missed in your stats. I think you should try to
 avoid missed in the upcall path, and just call it n_upcall_packets also.

In addition, I think you should keep two types of statics, and make them
 available, namely the total number of upcalls and the total of upcall
failures.

Cheers,

Eelco

Signed-off-by: wangchuanlei <wangchuanlei@inspur.com>
---
 include/uapi/linux/openvswitch.h |  6 ++++
 net/openvswitch/datapath.c       | 54 +++++++++++++++++++++++++++++++-
 net/openvswitch/datapath.h       | 12 +++++++
 net/openvswitch/vport.c          | 33 +++++++++++++++++++
 net/openvswitch/vport.h          |  4 +++
 5 files changed, 108 insertions(+), 1 deletion(-)
  

On 15 Nov 2022, at 8:28, wangchuanlei wrote:

> Added the total number of upcalls and the total of upcall failures.
> Due to ovs-userspace do not support NLA_NESTED, here still use the
> "struct ovs_vport_upcall_stats"

I guess that was the idea, we should start using the NLA_NESTED attribute rather than just embedding data structures.
The required backend support should be there, functions likes nl_msg_start_nested(), and NL_NESTED_FOR_EACH() exists.

> Thank you
> wangchuanlei
>
> On 14 Sep 2022, at 14:14, wangchuanlei wrote:
>
>> Add support to count upcall packets on every interface.
>> I always encounter high cpu use of ovs-vswictchd, this help to check
>> which interface send too many packets without open vlog/set switch,
>> and have no influence on datapath.
>
> Hi,
>
> I did not do a full review, but I think we should not try to make the
>  same mistake as before and embed a structure inside a netlink message.
>
> You are adding “struct ovs_vport_upcall_stats” but in theory,
>  you could have just added the new entry to “ovs_vport_stats”.
>  But this is breaking userspace as it expects an exact structure size :(
>
> So I think the right approach would be to have “
> OVS_VPORT_ATTR_UPCALL_STATS” be an NLA_NESTED type, and have
> individual stat attributes as NLA_U64 (or whatever type you need).
>
> What is also confusing is that you use upcall_packets in
>  ovs_vport_upcall_stats, which to me are the total of up calls,
> but you called it n_missed in your stats. I think you should try to
>  avoid missed in the upcall path, and just call it n_upcall_packets also.
>
> In addition, I think you should keep two types of statics, and make them
>  available, namely the total number of upcalls and the total of upcall
> failures.
>
> Cheers,
>
> Eelco
>
> Signed-off-by: wangchuanlei <wangchuanlei@inspur.com>
> ---
>  include/uapi/linux/openvswitch.h |  6 ++++
>  net/openvswitch/datapath.c       | 54 +++++++++++++++++++++++++++++++-
>  net/openvswitch/datapath.h       | 12 +++++++
>  net/openvswitch/vport.c          | 33 +++++++++++++++++++
>  net/openvswitch/vport.h          |  4 +++
>  5 files changed, 108 insertions(+), 1 deletion(-)
>
> diff --git a/include/uapi/linux/openvswitch.h b/include/uapi/linux/openvswitch.h
> index 94066f87e9ee..bb671d92b711 100644
> --- a/include/uapi/linux/openvswitch.h
> +++ b/include/uapi/linux/openvswitch.h
> @@ -126,6 +126,11 @@ struct ovs_vport_stats {
>  	__u64   tx_dropped;		/* no space available in linux  */
>  };
>
> +struct ovs_vport_upcall_stats {
> +	__u64   upcall_success;             /* total packets upcalls  succeed */
> +	__u64   upcall_fail;				/* total packets upcalls  failed */
> +};
> +
>  /* Allow last Netlink attribute to be unaligned */
>  #define OVS_DP_F_UNALIGNED	(1 << 0)
>
> @@ -277,6 +282,7 @@ enum ovs_vport_attr {
>  	OVS_VPORT_ATTR_PAD,
>  	OVS_VPORT_ATTR_IFINDEX,
>  	OVS_VPORT_ATTR_NETNSID,
> +	OVS_VPORT_ATTR_UPCALL_STATS, /* struct ovs_vport_upcall_stats */
>  	__OVS_VPORT_ATTR_MAX
>  };
>
> diff --git a/net/openvswitch/datapath.c b/net/openvswitch/datapath.c
> index c8a9075ddd0a..8b8ea95f94ae 100644
> --- a/net/openvswitch/datapath.c
> +++ b/net/openvswitch/datapath.c
> @@ -209,6 +209,28 @@ static struct vport *new_vport(const struct vport_parms *parms)
>  	return vport;
>  }
>
> +static void ovs_vport_upcalls(struct sk_buff *skb,
> +			      const struct dp_upcall_info *upcall_info,
> +			      bool upcall_success)
> +{
> +	if (upcall_info->cmd == OVS_PACKET_CMD_MISS ||
> +	    upcall_info->cmd == OVS_PACKET_CMD_ACTION) {
> +		const struct vport *p = OVS_CB(skb)->input_vport;
> +		struct vport_upcall_stats_percpu *vport_stats;
> +		u64 *stats_counter_upcall;
> +
> +		vport_stats = this_cpu_ptr(p->vport_upcall_stats_percpu);
> +		if (upcall_success)
> +			stats_counter_upcall = &vport_stats->n_upcall_success;
> +		else
> +			stats_counter_upcall = &vport_stats->n_upcall_fail;
> +
> +		u64_stats_update_begin(&vport_stats->syncp);
> +		(*stats_counter_upcall)++;
> +		u64_stats_update_end(&vport_stats->syncp);
> +	}
> +}
> +
>  void ovs_dp_detach_port(struct vport *p)
>  {
>  	ASSERT_OVSL();
> @@ -216,6 +238,9 @@ void ovs_dp_detach_port(struct vport *p)
>  	/* First drop references to device. */
>  	hlist_del_rcu(&p->dp_hash_node);
>
> +	/* Free percpu memory */
> +	free_percpu(p->vport_upcall_stats_percpu);
> +
>  	/* Then destroy it. */
>  	ovs_vport_del(p);
>  }
> @@ -305,8 +330,12 @@ int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
>  		err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
>  	else
>  		err = queue_gso_packets(dp, skb, key, upcall_info, cutlen);
> -	if (err)
> +	if (err) {
> +		ovs_vport_upcalls(skb, upcall_info, false);
>  		goto err;
> +	} else {
> +		ovs_vport_upcalls(skb, upcall_info, true);
> +	}
>
>  	return 0;
>
> @@ -1825,6 +1854,13 @@ static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
>  		goto err_destroy_portids;
>  	}
>
> +	vport->vport_upcall_stats_percpu =
> +				netdev_alloc_pcpu_stats(struct vport_upcall_stats_percpu);
> +	if (!vport->vport_upcall_stats_percpu) {
> +		err = -ENOMEM;
> +		goto err_destroy_portids;
> +	}
> +
>  	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
>  				   info->snd_seq, 0, OVS_DP_CMD_NEW);
>  	BUG_ON(err < 0);
> @@ -2068,6 +2104,7 @@ static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
>  {
>  	struct ovs_header *ovs_header;
>  	struct ovs_vport_stats vport_stats;
> +	struct ovs_vport_upcall_stats vport_upcall_stats;
>  	int err;
>
>  	ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
> @@ -2097,6 +2134,13 @@ static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
>  			  OVS_VPORT_ATTR_PAD))
>  		goto nla_put_failure;
>
> +	ovs_vport_get_upcall_stats(vport, &vport_upcall_stats);
> +	if (nla_put_64bit(skb, OVS_VPORT_ATTR_UPCALL_STATS,
> +			  sizeof(struct ovs_vport_upcall_stats),
> +			  &vport_upcall_stats,
> +			  OVS_VPORT_ATTR_PAD))
> +		goto nla_put_failure;
> +
>  	if (ovs_vport_get_upcall_portids(vport, skb))
>  		goto nla_put_failure;
>
> @@ -2278,6 +2322,14 @@ static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
>  		goto exit_unlock_free;
>  	}
>
> +	vport->vport_upcall_stats_percpu =
> +		netdev_alloc_pcpu_stats(struct vport_upcall_stats_percpu);
> +
> +	if (!vport->vport_upcall_stats_percpu) {
> +		err = -ENOMEM;
> +		goto exit_unlock_free;
> +	}
> +
>  	err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
>  				      info->snd_portid, info->snd_seq, 0,
>  				      OVS_VPORT_CMD_NEW, GFP_KERNEL);
> diff --git a/net/openvswitch/datapath.h b/net/openvswitch/datapath.h
> index 0cd29971a907..2f40db78d617 100644
> --- a/net/openvswitch/datapath.h
> +++ b/net/openvswitch/datapath.h
> @@ -50,6 +50,18 @@ struct dp_stats_percpu {
>  	struct u64_stats_sync syncp;
>  };
>
> +/**
> + * struct vport_upcall_stats_percpu - per-cpu packet upcall statistics for
> + * a given vport.
> + * @n_upcall_success: Number of packets that upcall to userspace succeed.
> + * @n_upcall_fail:    Number of packets that upcall to userspace failed.
> + */
> +struct vport_upcall_stats_percpu {
> +	u64 n_upcall_success;
> +	u64 n_upcall_fail;
> +	struct u64_stats_sync syncp;
> +};
> +
>  /**
>   * struct dp_nlsk_pids - array of netlink portids of for a datapath.
>   *                       This is used when OVS_DP_F_DISPATCH_UPCALL_PER_CPU
> diff --git a/net/openvswitch/vport.c b/net/openvswitch/vport.c
> index 82a74f998966..39b018da685e 100644
> --- a/net/openvswitch/vport.c
> +++ b/net/openvswitch/vport.c
> @@ -284,6 +284,39 @@ void ovs_vport_get_stats(struct vport *vport, struct ovs_vport_stats *stats)
>  	stats->tx_packets = dev_stats->tx_packets;
>  }
>
> +/**
> + *	ovs_vport_get_upcall_stats - retrieve upcall stats
> + *
> + * @vport: vport from which to retrieve the stats
> + * @ovs_vport_upcall_stats: location to store stats
> + *
> + * Retrieves upcall stats for the given device.
> + *
> + * Must be called with ovs_mutex or rcu_read_lock.
> + */
> +void ovs_vport_get_upcall_stats(struct vport *vport, struct ovs_vport_upcall_stats *stats)
> +{
> +	int i;
> +
> +	stats->upcall_success = 0;
> +	stats->upcall_fail = 0;
> +
> +	for_each_possible_cpu(i) {
> +		const struct vport_upcall_stats_percpu *percpu_upcall_stats;
> +		struct vport_upcall_stats_percpu local_stats;
> +		unsigned int start;
> +
> +		percpu_upcall_stats = per_cpu_ptr(vport->vport_upcall_stats_percpu, i);
> +		do {
> +			start = u64_stats_fetch_begin_irq(&percpu_upcall_stats->syncp);
> +			local_stats = *percpu_upcall_stats;
> +		} while (u64_stats_fetch_retry_irq(&percpu_upcall_stats->syncp, start));
> +
> +		stats->upcall_success += local_stats.n_upcall_success;
> +		stats->upcall_fail += local_stats.n_upcall_fail;
> +	}
> +}
> +
>  /**
>   *	ovs_vport_get_options - retrieve device options
>   *
> diff --git a/net/openvswitch/vport.h b/net/openvswitch/vport.h
> index 7d276f60c000..6defacd6d718 100644
> --- a/net/openvswitch/vport.h
> +++ b/net/openvswitch/vport.h
> @@ -32,6 +32,9 @@ struct vport *ovs_vport_locate(const struct net *net, const char *name);
>
>  void ovs_vport_get_stats(struct vport *, struct ovs_vport_stats *);
>
> +void ovs_vport_get_upcall_stats(struct vport *vport,
> +				struct ovs_vport_upcall_stats *stats);
> +
>  int ovs_vport_set_options(struct vport *, struct nlattr *options);
>  int ovs_vport_get_options(const struct vport *, struct sk_buff *);
>
> @@ -78,6 +81,7 @@ struct vport {
>  	struct hlist_node hash_node;
>  	struct hlist_node dp_hash_node;
>  	const struct vport_ops *ops;
> +	struct vport_upcall_stats_percpu __percpu *vport_upcall_stats_percpu;
>
>  	struct list_head detach_list;
>  	struct rcu_head rcu;
> -- 
> 2.27.0