ip_forward参数对Linux内核转发影响分析

在进行Linux内核转发时，需要在proc文件系统的proc/sys目录设置转发的参数，可以使用下面的方法查看该参数的值 cat /proc/sys/net/ipv4/ip_forward，该参数的默认值为0，可以使用下面的方法进行修改该值，使能Linux内核的IP层的数据抓发，但是下面的方法在系统重启后不再生效。echo 1 > /proc/sys/net/ipv4/ip_forward         在Linux系统中也提供了一个系统的配置工具sysctl，使用它可以读取和配置Linux内核的一些参数。但是该方法和proc文件系统相关，使用该工具Linux内核需要支持proc文件系统。下面是使用sysctl配置内核的转发参数。# sysctl net.ipv4.ip_forwardnet.ipv4.ip_forward = 0/ # sysctl -w net.ipv4.ip_forward=1net.ipv4.ip_forward = 1/ # sysctl net.ipv4.ip_forwardnet.ipv4.ip_forward = 1/ #注意，参数 net.ipv4.ip_forward 实际是对应的 proc 目录/proc/sys/net/ipv4/ip_forward，选项 -w 表示配置该内核配置参数，没有选项表示读内核配置参数，不加任何选项信息，就表示读取操作。通过上面的方法我们可以设置和读取IP转发的参数。但是本文重点不是讲该参数如何配置，而是在配置完成后，在内核的转发过程中如何生效的，以及如何配置到内核中。既然，该参数是配置使能IP层的转发，那应该在Linux内核的转发部分对该参数进行了判断，该参数的判断实际上是在查找路由时进行判断的，下面这张图显示了其中的调用关系，在查路由的过程中，如果是转发的数据包调用下面的宏判断转发的参数是否开启。在函数ip_route_input_slow。if (!IN_DEV_FORWARD(in_dev))            gotoe_hostunreach;         看一下该宏是如何进行定义的，下面的宏定义在include/linux/inetdevice.h文件中。#defineIN_DEV_FORWARD(in_dev)                 IN_DEV_CONF_GET((in_dev), FORWARDING)在把IN_DEV_CONF_GET宏进一步展开了看：#define IN_DEV_CONF_GET(in_dev, attr) \         ipv4_devconf_get((in_dev),NET_IPV4_CONF_ ## attr)//这里的##表示连接两个字符串。         下面是ipv4_devconf_get函数的定义：staticinline int ipv4_devconf_get(struct in_device *in_dev, int index){         index--;//这里的index相当于NET_IPV4_CONF_FORWARDING         return in_dev->cnf.data[index];// init_net->ipv4.devconf_dfl.data[0]}（1）对于宏NET_IPV4_CONF_ FORWARDING，定义在include/linux/sysctl.h文件中，是一个枚举类型的。enum{         NET_IPV4_CONF_FORWARDING=1,         NET_IPV4_CONF_MC_FORWARDING=2,         NET_IPV4_CONF_PROXY_ARP=3,         NET_IPV4_CONF_ACCEPT_REDIRECTS=4,         NET_IPV4_CONF_SECURE_REDIRECTS=5,         NET_IPV4_CONF_SEND_REDIRECTS=6,         NET_IPV4_CONF_SHARED_MEDIA=7,         NET_IPV4_CONF_RP_FILTER=8,         NET_IPV4_CONF_ACCEPT_SOURCE_ROUTE=9,         NET_IPV4_CONF_BOOTP_RELAY=10,         NET_IPV4_CONF_LOG_MARTIANS=11,         NET_IPV4_CONF_TAG=12,         NET_IPV4_CONF_ARPFILTER=13,         NET_IPV4_CONF_MEDIUM_ID=14,         NET_IPV4_CONF_NOXFRM=15,         NET_IPV4_CONF_NOPOLICY=16,         NET_IPV4_CONF_FORCE_IGMP_VERSION=17,         NET_IPV4_CONF_ARP_ANNOUNCE=18,         NET_IPV4_CONF_ARP_IGNORE=19,         NET_IPV4_CONF_PROMOTE_SECONDARIES=20,         NET_IPV4_CONF_ARP_ACCEPT=21,         NET_IPV4_CONF_ARP_NOTIFY=22,         NET_IPV4_CONF_SRC_VMARK=24,         __NET_IPV4_CONF_MAX};（2）对于returnin_dev->cnf.data[index];返回的相当于in_dev->cnf.data[0],那下面我们看一下该初始值是如何产生的。         首先，in_dev是怎么获取到的，在ip_route_input_slow函数中通过structin_device *in_dev = in_dev_get(dev);函数获取，在in_dev_get函数中调用__in_dev_get_rcu，通过下面的赋值语句进行赋值struct in_device *in_dev = dev->ip_ptr;staticinline struct in_device *__in_dev_get_rcu(const struct net_device *dev){         struct in_device *in_dev =dev->ip_ptr;         if (in_dev)                   in_dev =rcu_dereference(in_dev);         return in_dev;} static__inline__ struct in_device *in_dev_get(conststruct net_device *dev){         struct in_device *in_dev;          rcu_read_lock();         in_dev = __in_dev_get_rcu(dev);         if (in_dev)                   atomic_inc(&in_dev->refcnt);         rcu_read_unlock();         return in_dev;}         dev->ip_ptr;又是什么时候赋值呢？答案是在net_device注册初始化函数inetdev_init中，static struct in_device*inetdev_init(struct net_device *dev){         structin_device *in_dev;          ASSERT_RTNL();          in_dev= kzalloc(sizeof(*in_dev), GFP_KERNEL);         if(!in_dev)                   gotoout;         memcpy(&in_dev->cnf,dev_net(dev)->ipv4.devconf_dflt,                            sizeof(in_dev->cnf));//这里对in_dev->cnt进行初始化操作，---（1）         in_dev->cnf.sysctl= NULL;         in_dev->dev= dev;         if((in_dev->arp_parms = neigh_parms_alloc(dev, &arp_tbl)) == NULL)                   gotoout_kfree;         if(IPV4_DEVCONF(in_dev->cnf, FORWARDING))                   dev_disable_lro(dev);         /*Reference in_dev->dev */         dev_hold(dev);         /*Account for reference dev->ip_ptr (below) */         in_dev_hold(in_dev);          devinet_sysctl_register(in_dev);         ip_mc_init_dev(in_dev);         if(dev->flags & IFF_UP)                   ip_mc_up(in_dev);          /*we can receive as soon as ip_ptr is set -- do this last */         rcu_assign_pointer(dev->ip_ptr,in_dev);//使用RCU保护锁机制对dev->ip_ptr进行赋值out:         returnin_dev;out_kfree:         kfree(in_dev);         in_dev= NULL;         gotoout;}（1）dev_net(dev)->ipv4.devconf_dfl 也就相当于init_net->ipv4.devconf_dfl，而devconf_dfl的初始化时在/net/ipv4/devinet.c文件中，devinet_init_net函数中，static struct ipv4_devconfipv4_devconf_dflt = {         .data = {                   [NET_IPV4_CONF_ACCEPT_REDIRECTS- 1] = 1,                   [NET_IPV4_CONF_SEND_REDIRECTS- 1] = 1,                   [NET_IPV4_CONF_SECURE_REDIRECTS- 1] = 1,                   [NET_IPV4_CONF_SHARED_MEDIA- 1] = 1,                   [NET_IPV4_CONF_ACCEPT_SOURCE_ROUTE- 1] = 1,         },};//这里并没有对FORWARDING进行赋值操作static __net_init intdevinet_init_net(struct net *net){         interr;         structipv4_devconf *all, *dflt;#ifdef CONFIG_SYSCTL         struct ctl_table *tbl = ctl_forward_entry;         structctl_table_header *forw_hdr;#endif          err= -ENOMEM;         all = &ipv4_devconf; //----------------------------进行初始化操作         dflt = &ipv4_devconf_dflt;          if(net != &init_net) {                   all= kmemdup(all, sizeof(ipv4_devconf), GFP_KERNEL);                   if(all == NULL)                            gotoerr_alloc_all;                    dflt= kmemdup(dflt, sizeof(ipv4_devconf_dflt), GFP_KERNEL);                   if(dflt == NULL)                            gotoerr_alloc_dflt; #ifdef CONFIG_SYSCTL                   tbl= kmemdup(tbl, sizeof(ctl_forward_entry), GFP_KERNEL);                   if(tbl == NULL)                            gotoerr_alloc_ctl;                    tbl[0].data= &all->data[NET_IPV4_CONF_FORWARDING - 1];                   tbl[0].extra1= all;                   tbl[0].extra2= net;#endif         } #ifdef CONFIG_SYSCTL         err= __devinet_sysctl_register(net, "all",                            NET_PROTO_CONF_ALL,all);         if(err                   gotoerr_reg_all;          err= __devinet_sysctl_register(net, "default",                            NET_PROTO_CONF_DEFAULT,dflt);         if(err                   gotoerr_reg_dflt;          err= -ENOMEM;         forw_hdr = register_net_sysctl_table(net, net_ipv4_path,tbl);         if(forw_hdr == NULL)                   gotoerr_reg_ctl;         net->ipv4.forw_hdr= forw_hdr;#endif          net->ipv4.devconf_all = all;//这里对net->ipv4_devconfi_all进行了初始化         net->ipv4.devconf_dflt = dflt;// //这里对net->devconf_dflt进行了初始化         return0;………………………….}上面的函数对net相关功能的初始化，在devinet.c文件中还有一个和ipv4_devconf_dflt类似的变量ipv4_devconf，但是IN_DEV_FORWARD(in_dev)宏读取的是结构体ipv4_devconf_dflt中变量的值，所以，如果要在Linux内核中修改转发的参数时，需要在ipv4_devconf_dflt中添加才能生效。static struct ipv4_devconf ipv4_devconf = {         .data = {                   [NET_IPV4_CONF_ACCEPT_REDIRECTS- 1] = 1,                   [NET_IPV4_CONF_SEND_REDIRECTS- 1] = 1,                   [NET_IPV4_CONF_SECURE_REDIRECTS- 1] = 1,                   [NET_IPV4_CONF_SHARED_MEDIA- 1] = 1,                   [NET_IPV4_CONF_FORCE_IGMP_VERSION-1]=2,         },};（3）下面看一下使用echo 1 > /proc/sys/net/ipv4/ip_forward配置语句如何是Linux内核IP转发生效的。      在上面的devinet_init_net()函数中，有下面的两段代码         struct ctl_table *tbl = ctl_forward_entry;         forw_hdr = register_net_sysctl_table(net, net_ipv4_path,tbl);其中ctl_forward_entry定义为下面的结构，static structctl_table ctl_forward_entry[] = {         {                   .ctl_name         = NET_IPV4_FORWARD,//一个ID，                   .procname        = "ip_forward",//字符串，包含在proc/sys下目录项，实际为proc/sys目录下的文件名                   .data                  = &ipv4_devconf.data[                                               NET_IPV4_CONF_FORWARDING- 1],//回调函数设置的值                   .maxlen             = sizeof(int),//设置值的最大长度                   .mode                = 0644,//文件的权限，也就是ip_forward文件的权限                   .proc_handler  = devinet_sysctl_forward,// 对/proc/sys下面的文件修改的时候调用该回调函数。                   .strategy = devinet_conf_sysctl,// 用sysctl读写系统参数时候调用该回调函数                   .extra1              = &ipv4_devconf,                   .extra2              = &init_net,         },         { },};         forw_hdr =register_net_sysctl_table(net, net_ipv4_path, tbl);用于动态注册系统控制功能，其中net_ipv4_path定义为下面的形式。也就是proc/sys/下的目录名，tbl就是上面的ctl_forward_entry[]结构体。static __net_initdata struct ctl_pathnet_ipv4_path[] = {         {.procname = "net", .ctl_name = CTL_NET, },         {.procname = "ipv4", .ctl_name = NET_IPV4, },         {},};         使用echo 1 >/proc/sys/net/ipv4/ip_forward调用devinet_sysctl_forward函数进行处理，下面是该函数的定义实现。其中参数write为1表示写配置，为0表示读取配置值，buffer是要写的值，lenp为buffer的大小，ppos为位置。这里的__user是告诉不应该解除该指针的引用，因为在当前地址空间中它是没有意义的，所以对于这种变量，在kernel中使用要用到copy_to_user和copy_from_userstaticint devinet_sysctl_forward(ctl_table *ctl, int write,                                       void __user *buffer,                                       size_t *lenp, loff_t *ppos){         int *valp = ctl->data;//获取&ipv4_devconf.data地址         int val = *valp;         loff_t pos = *ppos;         int ret = proc_dointvec(ctl, write,buffer, lenp, ppos);//该函数处理传进来的int型，proc_dostring处理传过来的字符串。/*ctl->data change  echo "0">/proc/sys/net/ipv4/ip_forward  write= 1 *valp = 0 val = 1 */         if (write && *valp != val) {                   struct net *net =ctl->extra2;                    if (valp !=&IPV4_DEVCONF_DFLT(net, FORWARDING)) {                            if (!rtnl_trylock()){                                     /*Restore the original values before restarting */                                     *valp =val;                                     *ppos =pos;                                     returnrestart_syscall();                            }                            if (valp ==&IPV4_DEVCONF_ALL(net, FORWARDING)) {                                     inet_forward_change(net);//调用该函数进行配置in_dev->cnf.data                            }else if (*valp) {                                     structipv4_devconf *cnf = ctl->extra1;                                     structin_device *idev =                                               container_of(cnf,struct in_device, cnf);                                     dev_disable_lro(idev->dev);                            }                            rtnl_unlock();                            rt_cache_flush(net,0);                   }         }          return ret;}下面是这个函数就是修改forward参数，static void inet_forward_change(struct net *net){         struct net_device*dev;         int on = IPV4_DEVCONF_ALL(net, FORWARDING);//获取配置的值          IPV4_DEVCONF_ALL(net,ACCEPT_REDIRECTS) = !on;         IPV4_DEVCONF_DFLT(net,FORWARDING) = on;//设置ipv4_devconf_dflt结构体，          read_lock(&dev_base_lock);         for_each_netdev(net,dev) {                   struct in_device*in_dev;                   if (on)                            dev_disable_lro(dev);                   rcu_read_lock();                   in_dev =__in_dev_get_rcu(dev);                   if (in_dev)                            IN_DEV_CONF_SET(in_dev,FORWARDING, on);//调用该宏设置in_dev->cnf.data                   rcu_read_unlock();         }         read_unlock(&dev_base_lock);} static inline void ipv4_devconf_set(struct in_device *in_dev, intindex,                                         int val){         index--;         set_bit(index,in_dev->cnf.state);         in_dev->cnf.data[index] = val;//设置in_dev的data，这里的Index为NET_IPV4_CONF_FORWARDING}其调用关系如下图：