Linux内核收包精髓

本文原创为freas_1990，转载请表明出处：http://blog.csdn.net/freas_1990/article/details/42033025

网卡和磁盘是现代服务器里面性能要求最为苛刻的2个外设，我们来看一下Linux内核是如何处理高性能的网卡的。

int netif_rx(struct sk_buff *skb)
{
	int this_cpu = smp_processor_id();
	struct softnet_data *queue;
	unsigned long flags;


	if (skb->stamp.tv_sec == 0)
		do_gettimeofday(&skb->stamp);


	/* The code is rearranged so that the path is the most
	   short when CPU is congested, but is still operating.
	 */
	queue = &softnet_data[this_cpu];


	local_irq_save(flags);


	netdev_rx_stat[this_cpu].total++;
	if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
		if (queue->input_pkt_queue.qlen) {
			if (queue->throttle)
				goto drop;


enqueue:
			dev_hold(skb->dev);
			__skb_queue_tail(&queue->input_pkt_queue,skb);
			/* Runs from irqs or BH's, no need to wake BH */
			cpu_raise_softirq(this_cpu, NET_RX_SOFTIRQ);
			local_irq_restore(flags);
#ifndef OFFLINE_SAMPLE
			get_sample_stats(this_cpu);
#endif
			return softnet_data[this_cpu].cng_level;
		}


		if (queue->throttle) {
			queue->throttle = 0;
#ifdef CONFIG_NET_HW_FLOWCONTROL
			if (atomic_dec_and_test(&netdev_dropping))
				netdev_wakeup();
#endif
		}
		goto enqueue;
	}


	if (queue->throttle == 0) {
		queue->throttle = 1;
		netdev_rx_stat[this_cpu].throttled++;
#ifdef CONFIG_NET_HW_FLOWCONTROL
		atomic_inc(&netdev_dropping);
#endif
	}


drop:
	netdev_rx_stat[this_cpu].dropped++;
	local_irq_restore(flags);


	kfree_skb(skb);
	return NET_RX_DROP;
}


/* Deliver skb to an old protocol, which is not threaded well
   or which do not understand shared skbs.
 */
static int deliver_to_old_ones(struct packet_type *pt, struct sk_buff *skb, int last)
{
	static spinlock_t net_bh_lock = SPIN_LOCK_UNLOCKED;
	int ret = NET_RX_DROP;




	if (!last) {
		skb = skb_clone(skb, GFP_ATOMIC);
		if (skb == NULL)
			return ret;
	}
	if (skb_is_nonlinear(skb) && skb_linearize(skb, GFP_ATOMIC) != 0) {
		kfree_skb(skb);
		return ret;
	}


	/* The assumption (correct one) is that old protocols
	   did not depened on BHs different of NET_BH and TIMER_BH.
	 */


	/* Emulate NET_BH with special spinlock */
	spin_lock(&net_bh_lock);


	/* Disable timers and wait for all timers completion */
	tasklet_disable(bh_task_vec+TIMER_BH);


	ret = pt->func(skb, skb->dev, pt);


	tasklet_hi_enable(bh_task_vec+TIMER_BH);
	spin_unlock(&net_bh_lock);
	return ret;
}

static void net_rx_action(struct softirq_action *h)
{
	int this_cpu = smp_processor_id();
	struct softnet_data *queue = &softnet_data[this_cpu];
	unsigned long start_time = jiffies;
	int bugdet = netdev_max_backlog;

	br_read_lock(BR_NETPROTO_LOCK);

	for (;;) {
		struct sk_buff *skb;
		struct net_device *rx_dev;

		local_irq_disable();
		skb = __skb_dequeue(&queue->input_pkt_queue);
		local_irq_enable();

		if (skb == NULL)
			break;

		skb_bond(skb);

		rx_dev = skb->dev;

#ifdef CONFIG_NET_FASTROUTE
		if (skb->pkt_type == PACKET_FASTROUTE) {
			netdev_rx_stat[this_cpu].fastroute_deferred_out++;
			dev_queue_xmit(skb);
			dev_put(rx_dev);
			continue;
		}
#endif
		skb->h.raw = skb->nh.raw = skb->data;
		{
			struct packet_type *ptype, *pt_prev;
			unsigned short type = skb->protocol;

			pt_prev = NULL;
			for (ptype = ptype_all; ptype; ptype = ptype->next) {
				if (!ptype->dev || ptype->dev == skb->dev) {
					if (pt_prev) {
						if (!pt_prev->data) {
							deliver_to_old_ones(pt_prev, skb, 0);
						} else {
							atomic_inc(&skb->users);
							pt_prev->func(skb,
								      skb->dev,
								      pt_prev);
						}
					}
					pt_prev = ptype;
				}
			}

#ifdef CONFIG_NET_DIVERT
			if (skb->dev->divert && skb->dev->divert->divert)
				handle_diverter(skb);
#endif /* CONFIG_NET_DIVERT */

			
#if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
			if (skb->dev->br_port != NULL &&
			    br_handle_frame_hook != NULL) {
				handle_bridge(skb, pt_prev);
				dev_put(rx_dev);
				continue;
			}
#endif

			for (ptype=ptype_base[ntohs(type)&15];ptype;ptype=ptype->next) {
				if (ptype->type == type &&
				    (!ptype->dev || ptype->dev == skb->dev)) {
					if (pt_prev) {
						if (!pt_prev->data)
							deliver_to_old_ones(pt_prev, skb, 0);
						else {
							atomic_inc(&skb->users);
							pt_prev->func(skb,
								      skb->dev,
								      pt_prev);
						}
					}
					pt_prev = ptype;
				}
			}

			if (pt_prev) {
				if (!pt_prev->data)
					deliver_to_old_ones(pt_prev, skb, 1);
				else
					pt_prev->func(skb, skb->dev, pt_prev);
			} else
				kfree_skb(skb);
		}

		dev_put(rx_dev);

		if (bugdet-- < 0 || jiffies - start_time > 1)
			goto softnet_break;

#ifdef CONFIG_NET_HW_FLOWCONTROL
	if (queue->throttle && queue->input_pkt_queue.qlen < no_cong_thresh ) {
		if (atomic_dec_and_test(&netdev_dropping)) {
			queue->throttle = 0;
			netdev_wakeup();
			goto softnet_break;
		}
	}
#endif

	}
	br_read_unlock(BR_NETPROTO_LOCK);

	local_irq_disable();
	if (queue->throttle) {
		queue->throttle = 0;
#ifdef CONFIG_NET_HW_FLOWCONTROL
		if (atomic_dec_and_test(&netdev_dropping))
			netdev_wakeup();
#endif
	}
	local_irq_enable();

	NET_PROFILE_LEAVE(softnet_process);
	return;

softnet_break:
	br_read_unlock(BR_NETPROTO_LOCK);

	local_irq_disable();
	netdev_rx_stat[this_cpu].time_squeeze++;
	/* This already runs in BH context, no need to wake up BH's */
	cpu_raise_softirq(this_cpu, NET_RX_SOFTIRQ);
	local_irq_enable();

	NET_PROFILE_LEAVE(softnet_process);
	return;
}

麻痹的，Linux内核在网卡异步处理机制上比传统的BSD激进多了，性能优越得多，不过稳定性也随之下降！