dst_output发包

不管是收到报文转发还是本机发送报文，最后都会调用dst_output

/* Output packet to network from transport.  */
static inline int dst_output(struct net *net, struct sock *sk, struct sk_buff *skb)
{
   /*
     * 如果是单播数据包，设置的是ip_output(),
     * 如果是组播数据包，设置的是ip_mc_output().dev_queue_xmit
     */
    return skb_dst(skb)->output(net, sk, skb);
}

 单播：

?

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22

/*  * 对于单播数据包，目的路由缓存项中的输出接口是ip_output().  */ int ip_output(struct net *net, struct sock *sk, struct sk_buff *skb) {     struct net_device *dev = skb_dst(skb)->dev;       IP_UPD_PO_STATS(net, IPSTATS_MIB_OUT, skb->len);     /*      * 设置数据包的输出网络设备和数据包网络      * 层协议类型。      */     skb->dev = dev;     skb->protocol = htons(ETH_P_IP); /*      * 经netfilter处理后，调用ip_finish_output()继续IP数据包的输出      */     return NF_HOOK_COND(NFPROTO_IPV4, NF_INET_POST_ROUTING,                 net, sk, skb, NULL, dev,                 ip_finish_output,                 !(IPCB(skb)->flags & IPSKB_REROUTED)); }

 

?

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28

static int ip_finish_output(struct net *net, struct sock *sk, struct sk_buff *skb) {     unsigned int mtu;   #if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)     /* Policy lookup after SNAT yielded a new policy */     if (skb_dst(skb)->xfrm) {         IPCB(skb)->flags |= IPSKB_REROUTED;         return dst_output(net, sk, skb);     } #endif /* //如果不支持TSO或者GSO，tcp发送的时候是按照mss来组织skb的， 所以skb->len会等于mtu  所以TCP叫分段，和IP分片不一样，只有UDP才有IP分片 //SKB不是gso类型，并且skb->len大于mtu则需要分片  对方接受后的分片重组在netfilter中的ipv4_conntrack_defrag */     mtu = ip_skb_dst_mtu(sk, skb);     if (skb_is_gso(skb))         return ip_finish_output_gso(net, sk, skb, mtu);  /* 如果数据包长度大于MTU，则调用ip_fragment()      * 对IP数据包进行分片。      */     if (skb->len > mtu || (IPCB(skb)->flags & IPSKB_FRAG_PMTU))         return ip_fragment(net, sk, skb, mtu, ip_finish_output2);       return ip_finish_output2(net, sk, skb); }

 

?

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52

/* ip send the packet  by ip_finish_output2*/ static int ip_finish_output2(struct net *net, struct sock *sk, struct sk_buff *skb) {     struct dst_entry *dst = skb_dst(skb);     struct rtable *rt = (struct rtable *)dst;     struct net_device *dev = dst->dev;     unsigned int hh_len = LL_RESERVED_SPACE(dev);     struct neighbour *neigh;     u32 nexthop;       if (rt->rt_type == RTN_MULTICAST) {         IP_UPD_PO_STATS(net, IPSTATS_MIB_OUTMCAST, skb->len);     } else if (rt->rt_type == RTN_BROADCAST)         IP_UPD_PO_STATS(net, IPSTATS_MIB_OUTBCAST, skb->len);       /* Be paranoid, rather than too clever. */      /* skb头部空间不能存储链路头 */     if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {         struct sk_buff *skb2;          /* 重新分配skb */         skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));         if (!skb2) {             kfree_skb(skb);             return -ENOMEM;         }         if (skb->sk)/* 关联控制块 */             skb_set_owner_w(skb2, skb->sk);         consume_skb(skb); /* 释放skb */         skb = skb2; /* 指向新的skb */     }       rcu_read_lock_bh();     /* 获取下一跳 */     nexthop = (__force u32) rt_nexthop(rt, ip_hdr(skb)->daddr);// get the dst ip address (u32)     neigh = __ipv4_neigh_lookup_noref(dev, nexthop);     //根据目的IP查找邻居项是否存在      //如果没有则创建邻居项，然后通过dst_neigh_output  发包     if (unlikely(!neigh))         neigh = __neigh_create(&arp_tbl, &nexthop, dev, false);     if (!IS_ERR(neigh)) {/* 成功 */         int res = dst_neigh_output(dst, neigh, skb); /* 通过邻居子系统输出 */           rcu_read_unlock_bh();         return res;     }     rcu_read_unlock_bh();       net_dbg_ratelimited("%s: No header cache and no neighbour!\n",                 __func__);     kfree_skb(skb);     return -EINVAL; }

 

最后通过令邻居子系统，调用dev_queue_xmit 将数据报文发送给链路层驱动