本文分析基于Linux Kernel 1.2.13
原创作品，转载请标明http://blog.csdn.net/yming0221/article/details/7552455
更多请查看专栏，地址http://blog.csdn.net/column/details/linux-kernel-net.html
作者：闫明
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。
上篇博文分析传输层最终会调用函数ip_queue_xmit()函数，将发送数据的任务交给网络层，下面就分析了下该函数：
该函数的主要函数调用关系图如下：
/* * Queues a packet to be sent, and starts the transmitter
* if necessary. if free = 1 then we free the block after
* transmit, otherwise we don't. If free==2 we not only
* free the block but also don't assign a new ip seq number.
* This routine also needs to put in the total length,
* and compute the checksum
*/
void ip_queue_xmit(struct sock *sk, //发送数据的队列所对应的sock结构
struct device *dev,//发送该数据包的网卡设备
struct sk_buff *skb,//封装好的sk_buff结构，要发送的数据在该结构中
int free)//主要配合TCP协议使用，用于数据包的重发，UDP等协议调用是free=1
{
struct iphdr *iph;//IP数据报首部指针
unsigned char *ptr;
/* Sanity check */
if (dev == NULL)
{
printk("IP: ip_queue_xmit dev = NULL\n");
return;
}
IS_SKB(skb);
/*
*Do some book-keeping in the packet for later
*/
skb->dev = dev;//进一步完整sk_buff的相应字段
skb->when = jiffies;//用于TCP协议的超时重传
/*
*Find the IP header and set the length. This is bad
*but once we get the skb data handling code in the
*hardware will push its header sensibly and we will
*set skb->ip_hdr to avoid this mess and the fixed
*header length problem
*/
ptr = skb->data;//指针指向sk_buff中的数据部分
ptr += dev->hard_header_len;//hard_header_len为硬件首部长度，在net_init.c的函数eth_setup()函数中设置的，dev->hard_header_len = ETH_HLEN; 以太网首部长度为14
iph = (struct iphdr *)ptr;//prt已经指向IP数据包的首部
skb->ip_hdr = iph;
iph->tot_len = ntohs(skb->len-dev->hard_header_len);//计算IP数据报的总长度
#ifdef CONFIG_IP_FIREWALL
if(ip_fw_chk(iph, dev, ip_fw_blk_chain, ip_fw_blk_policy, 0) != 1)
/* just don't send this packet */
return;
#endif
/*
*No reassigning numbers to fragments...
*/
if(free!=2)
iph->id = htons(ip_id_count++);
else
free=1;
/* All buffers without an owner socket get freed */
if (sk == NULL)
free = 1;
skb->free = free;//设置skb的free值，free=1，发送后立即释放；free=2，不但释放缓存，而且不分配新的序列号
/*
*Do we need to fragment. Again this is inefficient.
*We need to somehow lock the original buffer and use
*bits of it.
*/
//数据帧中的数据部分必须小于等于MTU
if(skb->len > dev->mtu + dev->hard_header_len)//发送的数据长度大于数据帧的数据部分和帧首部之和，则需要分片
{
ip_fragment(sk,skb,dev,0);//对数据报分片后继续调用ip _queue_xmit()函数发送数据
IS_SKB(skb);
kfree_skb(skb,FREE_WRITE);
return;
}
/*
*Add an IP checksum
*/
ip_send_check(iph);//IP数据报首部检查
/*
*Print the frame when debugging
*/
/*
*More debugging. You cannot queue a packet already on a list
*Spot this and moan loudly.
*/
if (skb->next != NULL)//说明该数据包仍然存在于某个缓存队列
{
printk("ip_queue_xmit: next != NULL\n");
skb_unlink(skb);//将其从缓存链表中删除，否则可能导致内核错误
}
/*
*If a sender wishes the packet to remain unfreed
*we add it to his send queue. This arguably belongs
*in the TCP level since nobody else uses it. BUT
*remember IPng might change all the rules.
*/
if (!free)//free=0
{
unsigned long flags;
/* The socket now has more outstanding blocks */
sk->packets_out++;
/* Protect the list for a moment */
save_flags(flags);
cli();
if (skb->link3 != NULL)//link3指向数据报道呃重发队列
{
printk("ip.c: link3 != NULL\n");
skb->link3 = NULL;
}
//sk中send_tail和send_head是用户缓存的单向链表表尾和表头
if (sk->send_head == NULL)
{
sk->send_tail = skb;
sk->send_head = skb;
}
else
{
sk->send_tail->link3 = skb;//link3指针用于数据包的连接
sk->send_tail = skb;
}
/* skb->link3 is NULL */
/* Interrupt restore */
restore_flags(flags);
}
else
/* Remember who owns the buffer */
skb->sk = sk;
/*
*If the indicated interface is up and running, send the packet.
*/
ip_statistics.IpOutRequests++;
#ifdef CONFIG_IP_ACCT
ip_acct_cnt(iph,dev, ip_acct_chain);
#endif
#ifdef CONFIG_IP_MULTICAST//这部分是IP数据报的多播处理
/*
*Multicasts are looped back for other local users
*/
.......................................
#endif
if((dev->flags&IFF_BROADCAST) && iph->daddr==dev->pa_brdaddr && !(dev->flags&IFF_LOOPBACK))//广播数据包的处理
ip_loopback(dev,skb);
if (dev->flags & IFF_UP)//设备状态正常
{
/*
*If we have an owner use its priority setting,
*otherwise use NORMAL
*/
//调用设备接口层函数发送数据: dev_queue_xmit()函数
if (sk != NULL)
{
dev_queue_xmit(skb, dev, sk->priority);
}
else
{
dev_queue_xmit(skb, dev, SOPRI_NORMAL);
}
}
else//设备状态不正常
{
ip_statistics.IpOutDiscards++;
if (free)
kfree_skb(skb, FREE_WRITE);
}
}
这个函数中对长度过长的数据包进行了分片，ip_fragment()函数，该函数没有详细分析。
void ip_fragment(struct sock *sk, struct sk_buff *skb, struct device *dev, int is_frag){struct iphdr *iph;unsigned char *raw;unsigned char *ptr;struct sk_buff *skb2;int left, mtu, hlen, len;int offset;unsigned long flags;/* *Point into the IP datagram header. */raw = skb->data;iph = (struct iphdr *) (raw + dev->hard_header_len);skb->ip_hdr = iph;/* *Setup starting values. */hlen = (iph->ihl * sizeof(unsigned long));left = ntohs(iph->tot_len) - hlen;/* Space per frame */hlen += dev->hard_header_len;/* Total header size */mtu = (dev->mtu - hlen);/* Size of data space */ptr = (raw + hlen);/* Where to start from *//* *Check for any "DF" flag. [DF means do not fragment] */if (ntohs(iph->frag_off) & IP_DF){/* *Reply giving the MTU of the failed hop. */ip_statistics.IpFragFails++;icmp_send(skb,ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, dev->mtu, dev);return;}/* *The protocol doesn't seem to say what to do in the case that the *frame + options doesn't fit the mtu. As it used to fall down dead *in this case we were fortunate it didn't happen */if(mtu<8){/* It's wrong but it's better than nothing */icmp_send(skb,ICMP_DEST_UNREACH,ICMP_FRAG_NEEDED,dev->mtu, dev);ip_statistics.IpFragFails++;return;}/* *Fragment the datagram. *//* *The initial offset is 0 for a complete frame. When *fragmenting fragments it's wherever this one starts. */if (is_frag & 2)offset = (ntohs(iph->frag_off) & 0x1fff) << 3;elseoffset = 0;/* *Keep copying data until we run out. */while(left > 0){len = left;/* IF: it doesn't fit, use 'mtu' - the data space left */if (len > mtu)len = mtu;/* IF: we are not sending upto and including the packet end then align the next start on an eight byte boundary */if (len < left){len/=8;len*=8;}/* *Allocate buffer. */if ((skb2 = alloc_skb(len + hlen,GFP_ATOMIC)) == NULL){printk("IP: frag: no memory for new fragment!\n");ip_statistics.IpFragFails++;return;}/* *Set up data on packet */skb2->arp = skb->arp;if(skb->free==0)printk("IP fragmenter: BUG free!=1 in fragmenter\n");skb2->free = 1;skb2->len = len + hlen;skb2->h.raw=(char *) skb2->data;/* *Charge the memory for the fragment to any owner *it might possess */save_flags(flags);if (sk){cli();sk->wmem_alloc += skb2->mem_len;skb2->sk=sk;}restore_flags(flags);skb2->raddr = skb->raddr;/* For rebuild_header - must be here *//* *Copy the packet header into the new buffer. */memcpy(skb2->h.raw, raw, hlen);/* *Copy a block of the IP datagram. */memcpy(skb2->h.raw + hlen, ptr, len);left -= len;skb2->h.raw+=dev->hard_header_len;/* *Fill in the new header fields. */iph = (struct iphdr *)(skb2->h.raw/*+dev->hard_header_len*/);iph->frag_off = htons((offset >> 3));/* *Added AC : If we are fragmenting a fragment thats not the * last fragment then keep MF on each bit */if (left > 0 || (is_frag & 1))iph->frag_off |= htons(IP_MF);ptr += len;offset += len;/* *Put this fragment into the sending queue. */ip_statistics.IpFragCreates++;ip_queue_xmit(sk, dev, skb2, 2);//还是调用ip_queue_xmit()函数来发送分片后的数据}ip_statistics.IpFragOKs++;}
网络层的发送函数调用了设备接口层，相当于网络模型的链路层的发送函数dev_queue_xmit()
该函数的调用关系如下：
/* *Send (or queue for sending) a packet. * *IMPORTANT: When this is called to resend frames. The caller MUST *already have locked the sk_buff. Apart from that we do the *rest of the magic. */void dev_queue_xmit(struct sk_buff *skb, struct device *dev, int pri){unsigned long flags;int nitcount;struct packet_type *ptype;int where = 0;/* used to say if the packet should go*//* at the front or the back of the*//* queue - front is a retransmit try*//* where=0 表示是刚从上层传递的新数据包；where=1 表示从硬件队列中取出的数据包*/if (dev == NULL) {printk("dev.c: dev_queue_xmit: dev = NULL\n");return;}if(pri>=0 && !skb_device_locked(skb))//锁定该skb再进行操作，避免造成内核的不一致情况skb_device_lock(skb);/* Shove a lock on the frame */#ifdef CONFIG_SLAVE_BALANCINGsave_flags(flags);cli();if(dev->slave!=NULL && dev->slave->pkt_queue < dev->pkt_queue &&(dev->slave->flags & IFF_UP))dev=dev->slave;restore_flags(flags);#endif#ifdef CONFIG_SKB_CHECK IS_SKB(skb);#endif skb->dev = dev;/* *This just eliminates some race conditions, but not all... */if (skb->next != NULL) //这种条件似乎永远不能成立，因为发送数据包前，数据包已经从缓存队列摘下{//以防内核代码有BUG/* *Make sure we haven't missed an interrupt. */printk("dev_queue_xmit: worked around a missed interrupt\n");start_bh_atomic();dev->hard_start_xmit(NULL, dev);end_bh_atomic();return; }/* *Negative priority is used to flag a frame that is being pulled from the *queue front as a retransmit attempt. It therefore goes back on the queue *start on a failure. */ if (pri < 0) //优先级小于0表示是从硬件队列中取出的数据包 {pri = -pri-1;where = 1; }if (pri >= DEV_NUMBUFFS) {printk("bad priority in dev_queue_xmit.\n");pri = 1;}/* *If the address has not been resolved. Call the device header rebuilder. *This can cover all protocols and technically not just ARP either. */ if (!skb->arp && dev->rebuild_header(skb->data, dev, skb->raddr, skb)) {//用于ARP协议，并重建MAC帧首部return;}save_flags(flags);cli();if (!where) {//表示是新数据包，需要将其加入设备队列中#ifdef CONFIG_SLAVE_BALANCINGskb->in_dev_queue=1;//该数据包在设备队列#endifskb_queue_tail(dev->buffs + pri,skb);//将发送数据包加入硬件队列skb_device_unlock(skb);/* Buffer is on the device queue and can be freed safely */skb = skb_dequeue(dev->buffs + pri);//从硬件队列中取出一个数据包skb_device_lock(skb);/* New buffer needs locking down */#ifdef CONFIG_SLAVE_BALANCINGskb->in_dev_queue=0;#endif}restore_flags(flags);/* copy outgoing packets to any sniffer packet handlers */if(!where)//对于新的数据包，则遍历网络层协议队列，内核支持混杂模式{for (nitcount= dev_nit, ptype = ptype_base; nitcount > 0 && ptype != NULL; ptype = ptype->next) {/* Never send packets back to the socket * they originated from - MvS (miquels@drinkel.ow.org) */if (ptype->type == htons(ETH_P_ALL) && (ptype->dev == dev || !ptype->dev) && ((struct sock *)ptype->data != skb->sk)){struct sk_buff *skb2;if ((skb2 = skb_clone(skb, GFP_ATOMIC)) == NULL)break;/* *The protocol knows this has (for other paths) been taken off *and adds it back. */skb2->len-=skb->dev->hard_header_len;ptype->func(skb2, skb->dev, ptype);//IP层函数对应func为ip_rcv()，将发送的数据回送一份给对应的网络层协议nitcount--;//用于及时退出循环}}}start_bh_atomic();//开始原子操作if (dev->hard_start_xmit(skb, dev) == 0) {//调用硬件的发送函数发送数据end_bh_atomic();//结束原子操作/* *Packet is now solely the responsibility of the driver */return;//到这里说明数据包成功发送}//数据包没有成功发送，进行处理，将数据包从新加入硬件队列end_bh_atomic();/* *Transmission failed, put skb back into a list. Once on the list it's safe and *no longer device locked (it can be freed safely from the device queue) */cli();#ifdef CONFIG_SLAVE_BALANCINGskb->in_dev_queue=1;dev->pkt_queue++;#endifskb_device_unlock(skb);//对SKB解锁skb_queue_head(dev->buffs + pri,skb);//这次采用头插法插入硬件发送队列restore_flags(flags);}
具体的硬件发送函数dev->hard_start_xmit的实现将做下篇博文中分析。