Re: [RFC PATCH] packet: Add fanout support.

[Victor Julien <victor@inliniac.net>]

On 06/21/2011 11:53 AM, David Miller wrote:
> 
> This adds demuxing support for AF_PACKET sockets.  It's just to give
> people an idea, I've only build tested this patch.
> 
> Basically it allows to spread the AF_PACKET processing load amongst
> several AF_PACKET sockets.  The distribution can either be based upon
> hashing (PACKET_FANOUT_HASH) or round-robin based load-balancing
> (PACKET_FANOUT_LB).
> 
> The hash based fanout takes advantage of the precomputed skb->rxhash
> and only costs ~20 cpu cycles.
> 
> A restriction is that you must bind the AF_PACKET socket fully before
> you add it to a fanout.
> 
> The encoding of the PACKET_FANOUT socket option argument is:
> 
> 	(PACKET_FANOUT_{HASH,LB} << 16) | (ID & 0xffff)
> 
> All sockets adding themselves to the same fanout ID must all use
> the same PACKET_FANOUT_* type and also must be bound to the same
> device/protocol.
> 
> The implementation is agnostic to the type of AF_PACKET sockets in
> use.  You can use mmap based, and non-mmap based, AF_PACKET sockets.
> It simply doesn't care.

Thanks David! Looks interesting. I'm not familiar with the kernel
internals, so just a quick question. The hash based on skb->rxhash, does
that result in a "flow" based distribution over the listeners? So all
packets sharing a tuple being sent to the same socket?

Cheers,
Victor

> Signed-off-by: David S. Miller <davem@davemloft.net>
> 
> diff --git a/include/linux/if_packet.h b/include/linux/if_packet.h
> index 7b31863..1efa1cb 100644
> --- a/include/linux/if_packet.h
> +++ b/include/linux/if_packet.h
> @@ -49,6 +49,10 @@ struct sockaddr_ll {
>  #define PACKET_VNET_HDR			15
>  #define PACKET_TX_TIMESTAMP		16
>  #define PACKET_TIMESTAMP		17
> +#define PACKET_FANOUT			18
> +
> +#define PACKET_FANOUT_HASH		0
> +#define PACKET_FANOUT_LB		1
>  
>  struct tpacket_stats {
>  	unsigned int	tp_packets;
> diff --git a/net/packet/af_packet.c b/net/packet/af_packet.c
> index 461b16f..e6af2eb 100644
> --- a/net/packet/af_packet.c
> +++ b/net/packet/af_packet.c
> @@ -187,9 +187,11 @@ static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
>  
>  static void packet_flush_mclist(struct sock *sk);
>  
> +struct packet_fanout;
>  struct packet_sock {
>  	/* struct sock has to be the first member of packet_sock */
>  	struct sock		sk;
> +	struct packet_fanout	*fanout;
>  	struct tpacket_stats	stats;
>  	struct packet_ring_buffer	rx_ring;
>  	struct packet_ring_buffer	tx_ring;
> @@ -212,6 +214,22 @@ struct packet_sock {
>  	struct packet_type	prot_hook ____cacheline_aligned_in_smp;
>  };
>  
> +#define PACKET_FANOUT_MAX	2048
> +
> +struct packet_fanout {
> +#ifdef CONFIG_NET_NS
> +	struct net		*net;
> +#endif
> +	int			num_members;
> +	u16			id;
> +	u8			type;
> +	u8			pad;
> +	atomic_t		rr_cur;
> +	struct list_head	list;
> +	struct sock		*arr[PACKET_FANOUT_MAX];
> +	struct packet_type	prot_hook ____cacheline_aligned_in_smp;
> +};
> +
>  struct packet_skb_cb {
>  	unsigned int origlen;
>  	union {
> @@ -344,6 +362,164 @@ static void packet_sock_destruct(struct sock *sk)
>  	sk_refcnt_debug_dec(sk);
>  }
>  
> +static int fanout_rr_next(struct packet_fanout *f)
> +{
> +	int x = atomic_read(&f->rr_cur) + 1;
> +
> +	if (x >= f->num_members)
> +		x = 0;
> +
> +	return x;
> +}
> +
> +static struct sock *fanout_demux_hash(struct packet_fanout *f, struct sk_buff *skb)
> +{
> +	u32 idx = ((u64)skb->rxhash * f->num_members) >> 32;
> +
> +	return f->arr[idx];
> +}
> +
> +static struct sock *fanout_demux_lb(struct packet_fanout *f, struct sk_buff *skb)
> +{
> +	int cur, old;
> +
> +	cur = atomic_read(&f->rr_cur);
> +	while ((old = atomic_cmpxchg(&f->rr_cur, cur,
> +				     fanout_rr_next(f))) != cur)
> +		cur = old;
> +	return f->arr[cur];
> +}
> +
> +static int packet_rcv_fanout_hash(struct sk_buff *skb, struct net_device *dev,
> +				  struct packet_type *pt, struct net_device *orig_dev)
> +{
> +	struct packet_fanout *f = pt->af_packet_priv;
> +	struct packet_sock *po;
> +	struct sock *sk;
> +
> +	if (!net_eq(dev_net(dev), read_pnet(&f->net))) {
> +		kfree_skb(skb);
> +		return 0;
> +	}
> +
> +	sk = fanout_demux_hash(f, skb);
> +	po = pkt_sk(sk);
> +
> +	return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
> +}
> +
> +static int packet_rcv_fanout_lb(struct sk_buff *skb, struct net_device *dev,
> +				struct packet_type *pt, struct net_device *orig_dev)
> +{
> +	struct packet_fanout *f = pt->af_packet_priv;
> +	struct packet_sock *po;
> +	struct sock *sk;
> +
> +	if (!net_eq(dev_net(dev), read_pnet(&f->net))) {
> +		kfree_skb(skb);
> +		return 0;
> +	}
> +
> +	sk = fanout_demux_lb(f, skb);
> +	po = pkt_sk(sk);
> +
> +	return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
> +}
> +
> +static DEFINE_MUTEX(fanout_mutex);
> +static LIST_HEAD(fanout_list);
> +
> +static int fanout_add(struct sock *sk, u16 id, u8 type)
> +{
> +	struct packet_sock *po = pkt_sk(sk);
> +	struct packet_fanout *f, *match;
> +	int err;
> +
> +	switch (type) {
> +	case PACKET_FANOUT_HASH:
> +	case PACKET_FANOUT_LB:
> +		break;
> +	default:
> +		return -EINVAL;
> +	}
> +
> +	if (!po->running)
> +		return -EINVAL;
> +
> +	mutex_lock(&fanout_mutex);
> +	match = NULL;
> +	list_for_each_entry(f, &fanout_list, list) {
> +		if (f->id == id) {
> +			match = f;
> +			break;
> +		}
> +	}
> +	if (!match) {
> +		match = kzalloc(sizeof(*match), GFP_KERNEL);
> +		if (match) {
> +			write_pnet(&match->net, sock_net(sk));
> +			match->id = id;
> +			match->type = type;
> +			atomic_set(&match->rr_cur, 0);
> +			INIT_LIST_HEAD(&match->list);
> +			match->prot_hook.type = po->prot_hook.type;
> +			match->prot_hook.dev = po->prot_hook.dev;
> +			switch (type) {
> +			case PACKET_FANOUT_HASH:
> +				match->prot_hook.func = packet_rcv_fanout_hash;
> +				break;
> +			case PACKET_FANOUT_LB:
> +				match->prot_hook.func = packet_rcv_fanout_lb;
> +				break;
> +			}
> +			match->prot_hook.af_packet_priv = match;
> +			dev_add_pack(&match->prot_hook);
> +		}
> +	}
> +	err = -ENOMEM;
> +	if (match) {
> +		err = -EINVAL;
> +		if (match->type == type) {
> +			err = -ENOSPC;
> +			if (match->num_members < PACKET_FANOUT_MAX) {
> +				__dev_remove_pack(&po->prot_hook);
> +				po->fanout = match;
> +				match->arr[match->num_members] = sk;
> +				smp_wmb();
> +				match->num_members++;
> +				err = 0;
> +			}
> +		}
> +	}
> +	mutex_unlock(&fanout_mutex);
> +	return err;
> +}
> +
> +static void fanout_del(struct sock *sk)
> +{
> +	struct packet_sock *po = pkt_sk(sk);
> +	struct packet_fanout *f;
> +	int i;
> +
> +	f = po->fanout;
> +	po->fanout = NULL;
> +
> +	mutex_lock(&fanout_mutex);
> +	for (i = 0; i < f->num_members; i++) {
> +		if (f->arr[i] == sk)
> +			break;
> +	}
> +	BUG_ON(i >= f->num_members);
> +	f->arr[i] = f->arr[f->num_members - 1];
> +	f->num_members--;
> +
> +	if (!f->num_members) {
> +		list_del(&f->list);
> +		dev_remove_pack(&f->prot_hook);
> +		kfree(f);
> +	}
> +	mutex_unlock(&fanout_mutex);
> +}
>  
>  static const struct proto_ops packet_ops;
>  
> @@ -1343,7 +1519,10 @@ static int packet_release(struct socket *sock)
>  		 */
>  		po->running = 0;
>  		po->num = 0;
> -		__dev_remove_pack(&po->prot_hook);
> +		if (po->fanout)
> +			fanout_del(sk);
> +		else
> +			__dev_remove_pack(&po->prot_hook);
>  		__sock_put(sk);
>  	}
>  	if (po->prot_hook.dev) {
> @@ -1396,9 +1575,11 @@ static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protoc
>  		__sock_put(sk);
>  		po->running = 0;
>  		po->num = 0;
> -		spin_unlock(&po->bind_lock);
> -		dev_remove_pack(&po->prot_hook);
> -		spin_lock(&po->bind_lock);
> +		if (!po->fanout) {
> +			spin_unlock(&po->bind_lock);
> +			dev_remove_pack(&po->prot_hook);
> +			spin_lock(&po->bind_lock);
> +		}
>  	}
>  
>  	po->num = protocol;
> @@ -1413,7 +1594,8 @@ static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protoc
>  		goto out_unlock;
>  
>  	if (!dev || (dev->flags & IFF_UP)) {
> -		dev_add_pack(&po->prot_hook);
> +		if (!po->fanout)
> +			dev_add_pack(&po->prot_hook);
>  		sock_hold(sk);
>  		po->running = 1;
>  	} else {
> @@ -1542,7 +1724,8 @@ static int packet_create(struct net *net, struct socket *sock, int protocol,
>  
>  	if (proto) {
>  		po->prot_hook.type = proto;
> -		dev_add_pack(&po->prot_hook);
> +		if (!po->fanout)
> +			dev_add_pack(&po->prot_hook);
>  		sock_hold(sk);
>  		po->running = 1;
>  	}
> @@ -2109,6 +2292,17 @@ packet_setsockopt(struct socket *sock, int level, int optname, char __user *optv
>  		po->tp_tstamp = val;
>  		return 0;
>  	}
> +	case PACKET_FANOUT:
> +	{
> +		int val;
> +
> +		if (optlen != sizeof(val))
> +			return -EINVAL;
> +		if (copy_from_user(&val, optval, sizeof(val)))
> +			return -EFAULT;
> +
> +		return fanout_add(sk, val & 0xffff, val >> 16);
> +	}
>  	default:
>  		return -ENOPROTOOPT;
>  	}
> @@ -2207,6 +2401,15 @@ static int packet_getsockopt(struct socket *sock, int level, int optname,
>  		val = po->tp_tstamp;
>  		data = &val;
>  		break;
> +	case PACKET_FANOUT:
> +		if (len > sizeof(int))
> +			len = sizeof(int);
> +		val = (po->fanout ?
> +		       ((u32)po->fanout->id |
> +			((u32)po->fanout->type << 16)) :
> +		       0);
> +		data = &val;
> +		break;
>  	default:
>  		return -ENOPROTOOPT;
>  	}
> @@ -2260,7 +2463,8 @@ static int packet_notifier(struct notifier_block *this, unsigned long msg, void
>  			if (dev->ifindex == po->ifindex) {
>  				spin_lock(&po->bind_lock);
>  				if (po->num && !po->running) {
> -					dev_add_pack(&po->prot_hook);
> +					if (!po->fanout)
> +						dev_add_pack(&po->prot_hook);
>  					sock_hold(sk);
>  					po->running = 1;
>  				}
> @@ -2530,7 +2734,8 @@ static int packet_set_ring(struct sock *sk, struct tpacket_req *req,
>  	was_running = po->running;
>  	num = po->num;
>  	if (was_running) {
> -		__dev_remove_pack(&po->prot_hook);
> +		if (!po->fanout)
> +			__dev_remove_pack(&po->prot_hook);
>  		po->num = 0;
>  		po->running = 0;
>  		__sock_put(sk);
> @@ -2568,7 +2773,8 @@ static int packet_set_ring(struct sock *sk, struct tpacket_req *req,
>  		sock_hold(sk);
>  		po->running = 1;
>  		po->num = num;
> -		dev_add_pack(&po->prot_hook);
> +		if (!po->fanout)
> +			dev_add_pack(&po->prot_hook);
>  	}
>  	spin_unlock(&po->bind_lock);
>  
> 


-- 
---------------------------------------------
Victor Julien
http://www.inliniac.net/
PGP: http://www.inliniac.net/victorjulien.asc
---------------------------------------------