本文分析基于内核Linux Kernel 1.2.13
作者:闫明 https://blog.csdn.net/geekcome/article/details/7532512
更多请看专栏,地址http://blog.csdn.net/column/details/linux-kernel-net.html
注:标题中的”(上)“,”(下)“表示分析过程基于数据包的传递方向:”(上)“表示分析是从底层向上分析、”(下)“表示分析是从上向下分析。
这里看看数据包从IP层是如何交给传输层来处理的,为了方便,这里传输层以UDP协议为例来分析。
从ip_rcv()函数中可以看到
/*
* Pass on the datagram to each protocol that wants it,
* based on the datagram protocol. We should really
* check the protocol handler's return values here...
*/
ipprot->handler(skb2, dev, opts_p ? &opt : 0, iph->daddr,
(ntohs(iph->tot_len) - (iph->ihl * 4)),
iph->saddr, 0, ipprot);
这里调用指定协议的handler函数,如果是UDP协议,该函数的定义 udp_protocol如下
static struct inet_protocol udp_protocol = {
udp_rcv, /* UDP handler */
NULL, /* Will be UDP fraglist handler */
udp_err, /* UDP error control */
&tcp_protocol, /* next */
IPPROTO_UDP, /* protocol ID */
0, /* copy */
NULL, /* data */
"UDP" /* name */
};
先看UDP协议数据报的报头定义如下:比较简单
struct udphdr {
unsigned short source;//源端口
unsigned short dest;//目的端口
unsigned short len;//数据包长度
unsigned short check;//检验和
};
下面就分析下udp_rcv()函数,流程图:
/*
* All we need to do is get the socket, and then do a checksum.
*/
int udp_rcv(struct sk_buff *skb, struct device *dev, struct options *opt,
unsigned long daddr, unsigned short len,
unsigned long saddr, int redo, struct inet_protocol *protocol)
{
struct sock *sk;
struct udphdr *uh;
unsigned short ulen;
int addr_type = IS_MYADDR;
if(!dev || dev->pa_addr!=daddr)//检查这个数据包是不是发送给本地的数据包
addr_type=ip_chk_addr(daddr);//该函数定义在devinet.c中,用于检查ip地址是否是本地或多播、广播地址
/*
* Get the header.
*/
uh = (struct udphdr *) skb->h.uh;//获得UDP数据报的报头
ip_statistics.IpInDelivers++;
/*
* Validate the packet and the UDP length.
*/
ulen = ntohs(uh->len);
//参数len表示ip负载长度(IP数据报的数据部分长度)= UDP数据包头+UDP数据包的数据部分+填充部分长度
//ulen表示的是UDP数据报首部和负载部分的长度,所以正常情况下len>=ulen
if (ulen > len || len < sizeof(*uh) || ulen < sizeof(*uh))
{
printk("UDP: short packet: %d/%d\n", ulen, len);
udp_statistics.UdpInErrors++;
kfree_skb(skb, FREE_WRITE);
return(0);
}
if (uh->check && udp_check(uh, len, saddr, daddr)) //进行UDP数据包校验
{
/* <mea@utu.fi> wants to know, who sent it, to
go and stomp on the garbage sender... */
printk("UDP: bad checksum. From %08lX:%d to %08lX:%d ulen %d\n",
ntohl(saddr),ntohs(uh->source),
ntohl(daddr),ntohs(uh->dest),
ulen);
udp_statistics.UdpInErrors++;
kfree_skb(skb, FREE_WRITE);
return(0);
}
len=ulen;//对len赋值为实际的UDP数据报长度
#ifdef CONFIG_IP_MULTICAST//对多播情况进行处理
if (addr_type!=IS_MYADDR)
{
/*
* Multicasts and broadcasts go to each listener.
*/
struct sock *sknext=NULL;//next指针
/*get_sock_mcast 获取在对应端口的多播套接字队列
*下面函数的参数依次表示:sock结构指针,本地端口,远端地址,远端端口,本地地址
*/
sk=get_sock_mcast(udp_prot.sock_array[ntohs(uh->dest)&(SOCK_ARRAY_SIZE-1)], uh->dest,
saddr, uh->source, daddr);
if(sk)
{
do
{
struct sk_buff *skb1;
sknext=get_sock_mcast(sk->next, uh->dest, saddr, uh->source, daddr);//下一个满足条件的套接字
if(sknext)
skb1=skb_clone(skb,GFP_ATOMIC);
else
skb1=skb;
if(skb1)
udp_deliver(sk, uh, skb1, dev,saddr,daddr,len);//对满足条件的套接字调用发送函数发送
sk=sknext;
}
while(sknext!=NULL);
}
else
kfree_skb(skb, FREE_READ);
return 0;
}
#endif
sk = get_sock(&udp_prot, uh->dest, saddr, uh->source, daddr);
if (sk == NULL) //没有找到本地对应的套接字,则进行出错处理
{
udp_statistics.UdpNoPorts++;
if (addr_type == IS_MYADDR)
{
icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0, dev);//回复ICMP出错报文,目的主机不可达
}
/*
* Hmm. We got an UDP broadcast to a port to which we
* don't wanna listen. Ignore it.
*/
skb->sk = NULL;
kfree_skb(skb, FREE_WRITE);
return(0);
}
return udp_deliver(sk,uh,skb,dev, saddr, daddr, len);//调用函数发送套接字
}
上面函数中调用了get_sock_mcast()函数,下面具体分析一下该函数的功能,该函数定义的位置在文件af_inet.c中
/*
* Deliver a datagram to broadcast/multicast sockets.
*/
struct sock *get_sock_mcast(struct sock *sk, //套接字指针
unsigned short num,//本地端口
unsigned long raddr,//远端地址
unsigned short rnum,//远端端口
unsigned long laddr)//本地地址
{
struct sock *s;
unsigned short hnum;
hnum = ntohs(num);
/*
* SOCK_ARRAY_SIZE must be a power of two. This will work better
* than a prime unless 3 or more sockets end up using the same
* array entry. This should not be a problem because most
* well known sockets don't overlap that much, and for
* the other ones, we can just be careful about picking our
* socket number when we choose an arbitrary one.
*/
s=sk;
for(; s != NULL; s = s->next)
{
if (s->num != hnum) //本地端口不符合,跳过
continue;
if(s->dead && (s->state == TCP_CLOSE))//dead=1表示该sock结构已经处于释放状态
continue;
if(s->daddr && s->daddr!=raddr)//sock的远端地址不等于条件中的远端地址
continue;
if (s->dummy_th.dest != rnum && s->dummy_th.dest != 0)
continue;
if(s->saddr && s->saddr!=laddr)//sock的本地地址不等于条件的本地地址
continue;
return(s);
}
return(NULL);
}
下面是udp_rcv调用的udp_deliver()函数
static int udp_deliver(struct sock *sk,//sock结构指针
struct udphdr *uh,//UDP头指针
struct sk_buff *skb,//sk_buff
struct device *dev,//接收的网络设备
long saddr,//本地地址
long daddr,//远端地址
int len)//数据包的长度
{
//对skb结构相应字段赋值
skb->sk = sk;
skb->dev = dev;
//skb->len = len;
/*
* These are supposed to be switched.
*/
skb->daddr = saddr;//设置目的地址为本地地址
skb->saddr = daddr;//设置源地址为远端地址
/*
* Charge it to the socket, dropping if the queue is full.
*/
skb->len = len - sizeof(*uh);
if (sock_queue_rcv_skb(sk,skb)<0) //调用sock_queu_rcv_skb()函数,将skb挂到sk接构中的接收队列中
{
udp_statistics.UdpInErrors++;
ip_statistics.IpInDiscards++;
ip_statistics.IpInDelivers--;
skb->sk = NULL;
kfree_skb(skb, FREE_WRITE);
release_sock(sk);
return(0);
}
udp_statistics.UdpInDatagrams++;
release_sock(sk);
return(0);
}
sock_queu_rcv_skb()函数的实现如下:
/*
* Queue a received datagram if it will fit. Stream and sequenced protocols
* can't normally use this as they need to fit buffers in and play with them.
*/
int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
{
unsigned long flags;
if(sk->rmem_alloc + skb->mem_len >= sk->rcvbuf)
return -ENOMEM;
save_flags(flags);
cli();
sk->rmem_alloc+=skb->mem_len;
skb->sk=sk;
restore_flags(flags);
skb_queue_tail(&sk->receive_queue,skb);
if(!sk->dead)
sk->data_ready(sk,skb->len);
return 0;
}
这里就完成了数据包从网络层到传输层的传输。下面的博文将会分析数据包的从上到下的传输过程。