linux的网络设计与实现

linux的网络设计与实现
(2241 个字於此篇帖子)
(已阅读: 2514 次)

  Bricks :)   http://bricks.yeah.net/=====================================linux的网络设计与实现=====================主要参考:UNIX高级教程 系统技术内幕              (USA) Uresh Vahalia 著      聊鸿斌 曲广之 王元鹏 等译              清华大学出版社        //该书内容极其丰富, 强力推荐, 实是不可不读Network Buffers And Memory Management(in KHG)      by Alan Cox               他是linux 内核的主要维护者, 也写了大量的程序,                在源程序中他的名字到处可见.linux的源程序并没有想象中的难读.一般的说法是linux不如FreeBSD的网络功能稳定. 可惜, 我不懂FreeBSD.如果有人理解这一点, 请于我们联系, 谢谢!~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~章节:一. Alan Cox 文章的翻译和补充二. 报文是如何从网卡传递到相应协议的一. Alan Cox 文章的翻译和补充-----------------------------linux中的网络层设计是一种面向对象的设计结构.一些key:Device or Interface:        一个网络接口是可以接收或发送packet. 包括, 网卡和lo.Protocol:每一种协议描述了一种网络语言. 在linux中, 一个协议是一组代码向socket层提供服务.Socket:一个socket是一个网络连接, 向用户程序提供类似于文件I/O风格的服务. 在kernel中, 个socket是一对结构描述了高层的socket接口和低层的socket接口.sk_buff:所有网络层使用的buffer都是sk_buffs. 由于这种统一性, 存在一组例程为整个网络层服务. sk_buffs为网络层提供了通用的缓冲和流控机制.一个sk_buff带有一块内存的控制结构. 它有两组例程库. 第一组用于操纵由sk_buffs组成的双向链表, 第二组用于操纵其控制的内存. 这种双向链表按一般网络操作的方式进行优化, 通常是从表头取数据向表尾加数据.下面是一个例子:void append_frame(char *buf, int len){  struct sk_buff *skb = alloc_skb(len, GFP_ATOMIC);  if (skb == NULL)    my_dropped++;  else {    skb_put(skb, len);    memcpy(skb->data, data, len);    skb_append(&my_list, skb);  }}void process_queue(void){  struct sk_buff *skb;  while ((skb = skb_dequeue(&my_list)) != NULL) {    process_data(skb);    kfree_skb(skb, FREE_READ);  }}append_frame()有点象网卡驱动程序通过中断来取得一个报文.process_frame()有点象将报文传递给相应的协议.在 net/core/dev.c 中的 netif_rx() 和 net_bh() 中也有类似的程序, 由于他们需要处理将报文正确投递和流控所以代码较复杂.skb_put() 增长数据区的长度来为memcpy准备空间. 许多的网络操作需要加入一些桢头, 这可以使用skb_push来将数据区向后推, 为头留出空间.请参见下图:----------------------------------------| head | data |                     |----------------------------------------     skb_put-----------------------------------------       | head | data | put_data |             |       -----------------------------------------     skb_push------------------------------------------        | head | push_data | data | put_data |  |------------------------------------------还有一个函数skb_reserve()用于在添加数据前, 移动头指针来保留空间.参见下面的例程:    skb = alloc_skb(len+headspace, GFP_KERNEL);    skb_reserve(skb, headspace);    skb_put(skb,len);    memcpy_fromfs(skb->data, data, len);    pass_to_m_protocol(skb);看一看下面的函数:   o skb_dequeue()struct sk_buff *__skb_dequeue(struct sk_buff_head *list);用于从一个链表中取得数据, 第一个参数为一个链表. 可以使用skb_queue_head 和 skb_queue_tail 来向链表中添加数据.   o skb_queue_head()void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);用于将一个缓冲区放到链表的开头.   o skb_queue_tail()void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);用于将一个缓冲区放到链表的结尾.   o skb_unlink()void skb_unlink(struct sk_buff *skb);从链表中删除一个缓冲区无论它在那个链表中. 该缓冲区并不被释放掉.   o skb_insert skb_appendvoid skb_insert(struct sk_buff *old, struct sk_buff *newsk);void skb_append(struct sk_buff *old, struct sk_buff *newsk);由于一些网络协议需要对它们的数据进行排序, 这两个函数是将一个新包加在某个包的前面或后面.   o alloc_skb()#define alloc_skb(size, priority)(struct sk_buff *) kmalloc(size,priority)用于创建一个新的sk_buff. 通常作如下的操作 : skb->free=1; 表明它是自由的.   o kfree_skb()void kfree_skb(struct sk_buff *skb);用于释放一个sk_buff.   o skb_clone()struct sk_buff *skb_clone(struct sk_buff *skb, int gfp_mask);作一份sk_buff的拷贝, 但并不拷贝数据区.   o skb_copy()int skb_copy_datagram(struct sk_buff *skb, int offset,      char *to, int size);拷贝一个sk_buff的内容.Higher Level Support Routines-----------------------------其中的sock_queue_rcv_skb() 用于处理接收到的数据并有流控处理. 它通常有下面的风格:    sk = my_find_socket(whatever);    if (sock_queue_rcv_skb(sk,skb) == -1) {        myproto_stats.dropped++;        kfree_skb(skb, FREE_READ);        return;    }该函数会阻止大量的数据进入socket. 当到达一定的限度时, 数据会被抛弃.在发送方, sock_alloc_send_skb() 用于处理一些标志, 如: 非阻塞等.    skb=sock_alloc_send_skb(sk,....)    if(skb == NULL)        return -err;    skb->sk = sk;    skb_reserve(skb, headroom);    skb_put(skb,len);    memcpy(skb->data, data, len);    protocol_do_something(skb);比较重要的一行是 skb->sk = sk;sock_alloc_send_skb() 改变了缓冲区指向的socket, 通过这一行我们告诉内核作kfree_skb()时, 要有sk作确认. 这样, 当一个设备发送完一个包并且释放了它, 用户可以继续发更多的包.Network Devices---------------所有的linux网络设备使用相同的函数接口(面向对象的设计).drivers/net/skeleton.c 包含了一个网络设备的轮廓.Basic Structure---------------      --------------------              -----------------------------      | dev_queue_xmit |         | netif_rx receivers frames |      |delivers packets |         | and queues them up.       |      --------------------         -----------------------------          |||||||                               ||||||||      ----------------------------------------------------------------      |            Methods and Variables (struct device)           |      ----------------------------------------------------------------|||||||              ||||||      ------------------   -------------------   ---------------------      |Initialisation |   | hard_start_time |   | mydev_interrupt   |      |Routine       |   | delivers frames |   | collects received |      ------------------   -------------------   | frames     |||||||| ---------------------|||||||              ||||||      ----------------------------------------------------------------      |                 Physical Device Media                   |      ----------------------------------------------------------------每个网络设备处理数据从协议层到物理介质的传输并从硬件上接收数据, 接收到的数据被放到网络层, 由netif_rx()完成, 改函数去掉桢头以被高层协议使用.每一个设备提供了以一套补充的方法来处理 停止, 开始, 控制和物理封装包.Naming------所有的linux网络设备有唯一的名字, 实际上, 网络设备并没有一个对应的文件, 虽然你可以自己建一个设备文件. 传统的, 名字标明的设备的类型而不是其制造商. 多个设备通过附加的数字来标明(数字从零开始).ethn以太设备trn令牌环slnSLIP设备pppnPPP设备plipnPLIP单元tunlnIPIP封装通道nrnNetROM 虚拟设备isdnnISDN设备dummyn空设备lo本地回绕设备Registering A Device--------------------每一个设备是通过添写一个 struct device 对象, 使用register_netdev(struct device *)来注册. 它将你的struct device连入内核的网络设备表. 你不可以释放该结构, 直到, 你使用 unregister_netdev(struct device *) 来释放该设备. 内核不支持多个设备使用同一个名字, 如果你的程序是一个模块, 你应该使用 struct device * dev_get(const char * name)来确定这个名字是否被使用了, 如果是, 你应该另选一个名字或是失败.一个典型的注册程序如下:int register_my_device(void){  int i = 0;  for(i = 0;i < 100;i++)  {    sprintf(mydevice.name, "mydev%d",i);    if(dev_get(mydevice.name) == NULL)    {      if(register_netdev(&mydevice) != 0)        return -EIO;      return 0;    }  }  printk("100 mydevs loaded. Unable to load more.");  return -ENFILE;}The Device Structure--------------------所有的一般性信息都放在 struct device 中, 为了创建一个设备你需要初始化它. 下面讨论它的数据.Naming------首先, name 包含了设备的名字, 这是一个字符串指针, 也可以是四个空格, 这样内核会自动分配一个ethn的名字, 但最好不要使用这一功能.Bus Interface Parameters------------------------这一部分用于维护网络设备在物理上的一些参数.Protocol Layer Variables------------------------关于协议层的一些数据. mtu等.二. 报文是如何从网卡传递到相应协议的------------------------------------* 我个人的一些心得(仅供参考)先来看一看网络部分是如何被初始化的. 下面是函数被调用的过程:start_kernel -> init -> do_basic_setup -> sock_init -> proto-init           inet_proto_init  -> ip_init -> dev_add_pack下面的结构定义了网络协议的初始化入口:struct net_proto{const char *name;/* Protocol name */void (*init_func)(struct net_proto *);/* Bootstrap */} protocols[];每一个协议提供了一个自己的init_func. 如IP提供了ip_init.dev_add_pack完成了实际的协议添加过程. 系统维护了两个协议表. 一个是单向链表,另一个是hash表(使用了桶形的hash表).每一个协议用一个struct packet_type来描述, 其中的func是其入口函数. 当系统从读到一个报文就会调用相应协议的func来完成实际的处理工作.struct packet_type{unsigned shorttype;/* This is really htons(ether_type).*/struct device*dev;/* NULL is wildcarded here*/int(*func) (struct sk_buff *, struct device *, struct packet_type *);void*data;/* Private to the packet type*/struct packet_type*next;};// 协议的hash表struct packet_type *ptype_base[16];/* Hashed types */// 协议的单向链表的头指针struct packet_type *ptype_all = NULL;/* Taps */dev_add_pack完成的工作其实很简单, 它将一个struct packet_type指针加入到相应的链表中. 请看源码:void dev_add_pack(struct packet_type *pt){int hash;//由pt->type来判断加到那一个链表中if(pt->type == htons(ETH_P_ALL)){netdev_nit++;// 添加到单向链表中pt->next = ptype_all;ptype_all = pt;}else{// 添加到hash表中hash = ntohs(pt->type)&15;~~~~~~~~~~~~~~~~~~~~~~~~~~//这是它的hash算法, 简单的只取低4位pt->next = ptype_base[hash];ptype_base[hash] = pt;}}ip_init使用dev_add_pack完成了实际的添加过程.下面是核心数据结构sk_buffstruct sk_buff {struct sk_buff* next;/* Next buffer in list*/struct sk_buff* prev;/* Previous buffer in list*/struct sk_buff_head * list;/* List we are on*/struct sock*sk;/* Socket we are owned by */struct timevalstamp;/* Time we arrived*/struct device*dev;/* Device we arrived on/are leaving by  *//* Transport layer header */union{struct tcphdr*th;struct udphdr*uh;struct icmphdr*icmph;struct igmphdr*igmph;struct iphdr*ipiph;struct spxhdr*spxh;unsigned char*raw;} h;/* Network layer header */union{struct iphdr*iph;struct ipv6hdr*ipv6h;struct arphdr*arph;struct ipxhdr*ipxh;unsigned char*raw;} nh;/* Link layer header */union{  struct ethhdr*ethernet;  unsigned char *raw;} mac;struct  dst_entry *dst;charcb[48];unsigned int len;/* Length of actual data    */unsigned intcsum;/* Checksum */volatile char used;/* Data moved to user and not MSG_PEEK */unsigned charis_clone,/* We are a clone   */cloned, /* head may be cloned(check refcnt to be sure)*/  pkt_type,/* Packet class    */  pkt_bridged,/* Tracker for bridging       */  ip_summed;/* Driver fed us an IP checksum      */__u32priority;/* Packet queueing priority      */atomic_tusers;/* User count - see datagram.c,tcp.c  */unsigned shortprotocol;/* Packet protocol from driver.       */unsigned shortsecurity;/* Security level of packet      */unsigned inttruesize;/* Buffer size        */unsigned char*head;/* Head of buffer        */unsigned char*data;/* Data head pointer       */unsigned char*tail;/* Tail pointer       */unsigned char *end;/* End pointer       */void (*destructor)(struct sk_buff *);/* Destruct function   */#ifdef CONFIG_IP_FIREWALL        __u32           fwmark;                 /* Label made by fwchains, used by pktsched  */#endif#if defined(CONFIG_SHAPER) || defined(CONFIG_SHAPER_MODULE)__u32shapelatency;/* Latency on frame */__u32shapeclock;/* Time it should go out */__u32shapelen;/* Frame length in clocks */__u32shapestamp;/* Stamp for shaper    */__u16shapepend;/* Pending */#endif#if defined(CONFIG_HIPPI)union{__u32ifield;} private;#endif};* arch/i386/kernel/irq.cvoid do_IRQ(unsigned int irq, struct pt_regs *regs){    int cpu = smp_processor_id();    kstat.irqs[cpu][irq]++;    irq_desc[irq].handler->handle(irq, regs);    /*     * This should be conditional: we should really get     * a return code from the irq handler to tell us     * whether the handler wants us to do software bottom     * half handling or not..     */    //每次处理中断, 都会进行底半处理. 现在还不符合底半的原意.    if (1) {        if (bh_active & bh_mask)            do_bottom_half();    }    __sti();/*VY*/}* net/core/dev.c/* *Device drivers call our routines to queue packets here. We empty the *queue in the bottom half handler. *///所有的报文先存放在这里, 随后由再net_bh发给相应的网络层模块, 如 ip_input.static struct sk_buff_head backlog;/* *Receive a packet from a device driver and queue it for the upper *(protocol) levels.  It always succeeds. *//* * 简单的将新到的包, 插到backlog中. * 网卡驱动程序会调用这个函数 */void netif_rx(struct sk_buff *skb){if(skb->stamp.tv_sec==0)get_fast_time(&skb->stamp); //取当前的时间/* The code is rearranged so that the path is the most   short when CPU is congested, but is still operating. */if (backlog.qlen <= netdev_max_backlog) {if (backlog.qlen) {if (netdev_dropping == 0) {skb_queue_tail(&backlog, skb);mark_bh(NET_BH);return;}atomic_inc(&netdev_rx_dropped);kfree_skb(skb);return;}netdev_dropping = 0;skb_queue_tail(&backlog, skb);mark_bh(NET_BH);return;}netdev_dropping = 1;atomic_inc(&netdev_rx_dropped);kfree_skb(skb);}/* *  网络的底半处理程序, 将报文由backlog中取出, 发给相应的网络层模块, *  如 ip_input. 有一点请大家注意一个报文可以被多个协议层所处理. 这 *  一点很重要. */void net_bh(void){struct packet_type *ptype;struct packet_type *pt_prev;unsigned short type;unsigned long start_time = jiffies;NET_PROFILE_ENTER(net_bh);/* *Can we send anything now? We want to clear the *decks for any more sends that get done as we *process the input. This also minimises the *latency on a transmit interrupt bh. */if (qdisc_head.forw != &qdisc_head)qdisc_run_queues();/* *Any data left to process. This may occur because a *mark_bh() is done after we empty the queue including *that from the device which does a mark_bh() just after *//* *While the queue is not empty.. * *Note that the queue never shrinks due to *an interrupt, so we can do this test without *disabling interrupts. */while (!skb_queue_empty(&backlog)){struct sk_buff * skb;/* Give chance to other bottom halves to run */if (jiffies - start_time > 1)goto net_bh_break;/* We have a packet. Therefore the queue has shrunk */skb = skb_dequeue(&backlog); //从backlog中取出一个包/*  *Bump the pointer to the next structure. * *On entry to the protocol layer. skb->data and *skb->nh.raw point to the MAC and encapsulated data *//* XXX until we figure out every place to modify.. */skb->h.raw = skb->nh.raw = skb->data;if (skb->mac.raw < skb->head || skb->mac.raw > skb->data) {printk(KERN_CRIT "%s: wrong mac.raw ptr, proto=%04x",                               skb->dev->name, skb->protocol);kfree_skb(skb);continue;}/* Fetch the packet protocol ID. */type = skb->protocol; //这里是指网络层协议/* *  We got a packet ID.  Now loop over the "known protocols" *  list. There are two lists. The ptype_all list of taps (normally empty) *  and the main protocol list which is hashed perfectly for normal protocols. */pt_prev = NULL;for (ptype = ptype_all; ptype!=NULL; ptype=ptype->next){if (!ptype->dev || ptype->dev == skb->dev) {if(pt_prev){struct sk_buff *skb2=skb_clone(skb, GFP_ATOMIC);if(skb2)pt_prev->func(skb2,skb->dev, pt_prev);}pt_prev=ptype;}}for (ptype = ptype_base[ntohs(type)&15]; ptype != NULL; ptype = ptype->next){if (ptype->type == type && (!ptype->dev || ptype->dev==skb->dev)){/* *We already have a match queued. Deliver *to it and then remember the new match */if(pt_prev){struct sk_buff *skb2;skb2=skb_clone(skb, GFP_ATOMIC);/* * Kick the protocol handler. This should be fast * and efficient code. */if(skb2)pt_prev->func(skb2, skb->dev, pt_prev);}/* Remember the current last to do */pt_prev=ptype;}} /* End of protocol list loop *//* Is there a last item to send to ? */if(pt_prev)pt_prev->func(skb, skb->dev, pt_prev);/* Has an unknown packet has been received ? */else {kfree_skb(skb);}  }/* End of queue loop */  /* We have emptied the queue *//* One last output flush. */if (qdisc_head.forw != &qdisc_head)qdisc_run_queues();netdev_dropping = 0;NET_PROFILE_LEAVE(net_bh);return;net_bh_break:mark_bh(NET_BH);NET_PROFILE_LEAVE(net_bh);return;}为了理清头绪, 我重新描述一下上面的过程:网卡驱动程序调用netif_rx将新收到的报文存在backlog队列中.在底半处理中, net_bh调用相应的协议模块来处理报文. 而目前linux的实现中, 每次中断都会调用底半处理.三. 报文是如何从协议传递到网卡的--------------------------------struct Qdisc{struct Qdisc_headh;int (*enqueue)(struct sk_buff *skb, struct Qdisc *dev);struct sk_buff *(*dequeue)(struct Qdisc *dev);unsignedflags;#define TCQ_F_BUILTIN1#define TCQ_F_THROTTLED2struct Qdisc_ops*ops;struct Qdisc*next;u32handle;atomic_trefcnt;struct sk_buff_headq;struct device *dev;struct tc_statsstats;unsigned longtx_timeo;unsigned longtx_last;int(*reshape_fail)(struct sk_buff *skb, struct Qdisc *q);/* This field is deprecated, but it is still used by CBQ * and it will live until better solution will be invented. */struct Qdisc*__parent;chardata[0];};int dev_queue_xmit(struct sk_buff *skb){struct device *dev = skb->dev;struct Qdisc  *q;#ifdef CONFIG_NET_PROFILEstart_bh_atomic();NET_PROFILE_ENTER(dev_queue_xmit);#endifstart_bh_atomic();q = dev->qdisc;if (q->enqueue) {q->enqueue(skb, q);qdisc_wakeup(dev);end_bh_atomic();#ifdef CONFIG_NET_PROFILE        NET_PROFILE_LEAVE(dev_queue_xmit);end_bh_atomic();#endifreturn 0;}/* The device has no queue. Common case for software devices:   loopback, all the sorts of tunnels...   Really, it is unlikely that bh protection is necessary here:   virtual devices do not generate EOI events.   However, it is possible, that they rely on bh protection   made by us here. */if (dev->flags&IFF_UP) {if (netdev_nit)dev_queue_xmit_nit(skb,dev);if (dev->hard_start_xmit(skb, dev) == 0) {end_bh_atomic();#ifdef CONFIG_NET_PROFILENET_PROFILE_LEAVE(dev_queue_xmit);end_bh_atomic();#endifreturn 0;}if (net_ratelimit())printk(KERN_DEBUG "Virtual device %s asks to queue packet!", dev->name);}end_bh_atomic();kfree_skb(skb);#ifdef CONFIG_NET_PROFILENET_PROFILE_LEAVE(dev_queue_xmit);end_bh_atomic();#endifreturn 0;}