linux网络协议栈--路由流程分析

转：http://blog.csdn.net/hsly_support/article/details/8797976

来吧，路由

路由是网络的核心，是linux网络协议栈的核心，我们找个入口进去看看

---

还记得在笔记5-IP层的处理1中ip_rcv_finish走到过一个岔口

->ip_rcv_finish()

     ->ip_route_input()  查找路由信息

     ->if (iph->ihl > 5 && ip_rcv_options(skb)) 如果IP头部大于20字节，则表示IP头部包含IP选项，需要进行选项处理.
          goto drop;

     ->dst_input(skb);      dst_input实际上会调用skb->dst->input(skb).input函数会根据路由信息设置为合适的函数指针，

                                      如果是则递交到本地的则为ip_local_deliver，若是转发则为ip_forward

两条路径：

1) ip_local_deliver

2) ip_forward

是什么导致路径不同呢，我们看一看ip_route_input() 干了啥

---

int ip_route_input(struct sk_buff *skb, __be32 daddr, __be32 saddr,
             u8 tos, struct net_device *dev)

     ->net = dev_net(dev);

     ->hash = rt_hash(daddr, saddr, iif, rt_genid(net));        计算hash值，注意hash因子    

     ->                                                                             既然hash值算出来了，我们就去找吧

     rcu_read_lock();

for (rth = rcu_dereference(rt_hash_table[hash].chain); rth;      
          rth = rcu_dereference(rth->u.dst.rt_next)) {
          if (((rth->fl.fl4_dst ^ daddr) |                    异或 相同为0            
          
               (rth->fl.fl4_src ^ saddr) |
               (rth->fl.iif ^ iif) |
               rth->fl.oif |
               (rth->fl.fl4_tos ^ tos)) == 0 &&
              rth->fl.mark == skb->mark &&
              net_eq(dev_net(rth->u.dst.dev), net) &&           判断路由和报文的struct net指针地址是否相同
              !rt_is_expired(rth)) {                                        路由项是否过期

                           找到了
               dst_use(&rth->u.dst, jiffies);                            表示路由的使用时间
               RT_CACHE_STAT_INC(in_hit);
               rcu_read_unlock();
               skb_dst_set(skb, &rth->u.dst);                         设置到skb中去
               return 0;
          }
          RT_CACHE_STAT_INC(in_hlist_search);
     }
     rcu_read_unlock();

---

static struct rt_hash_bucket      *rt_hash_table __read_mostly;

struct rt_hash_bucket {
     struct rtable     *chain;
};

struct rtable
{
     union
     {
          struct dst_entry     dst;
     } u;
     /* Cache lookup keys */
     struct flowi          fl;                     存放的是查找该路由节点的哈希值，该哈希值用源IP,目的地址，TOS一起确定
     struct in_device     *idev;
    
     int               rt_genid;
     unsigned          rt_flags;               一些结构性的标志，例如，RTF_UP表示这条路由可用
     __u16               rt_type;               表明了目标地址的类型，例如RTN_LOCAL,RTN_MULTICAST

__be32               rt_dst;     /* Path destination     */              用来存放目标的IP地址
     __be32               rt_src;     /* Path source          */               路由路径的起点ip地址
     int               rt_iif;

/* Info on neighbour */
     __be32               rt_gateway;                     网关信息

/* Miscellaneous cached information */
     __be32               rt_spec_dst; /* RFC1122 specific destination */
     struct inet_peer     *peer; /* long-living peer info */
};

---

当然，在rt_hash_table中查不到  则处理多播ipv4_is_multicast(daddr)   ->ip_route_input_mc

再进ip_route_input_slow(skb, daddr, saddr, tos, dev);  rt_hash_table为路由高速缓存

我们主要分析ip_route_input_slow()

static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
                      u8 tos, struct net_device *dev)
{
     struct fib_result res;
     struct in_device *in_dev = in_dev_get(dev);
     struct flowi fl = { .nl_u = { .ip4_u =
                          { .daddr = daddr,
                         .saddr = saddr,
                         .tos = tos,
                         .scope = RT_SCOPE_UNIVERSE,
                          } },
                   .mark = skb->mark,
                   .iif = dev->ifindex };                       初始化路由键值
     unsigned     flags = 0;
     u32          itag = 0;
     struct rtable * rth;
     unsigned     hash;
     __be32          spec_dst;
     int          err = -EINVAL;
     int          free_res = 0;
     struct net    * net = dev_net(dev);

/* IP on this device is disabled. */

if (!in_dev)
          goto out;

/* Check for the most weird martians, which can be not detected
        by fib_lookup.
     */

if (ipv4_is_multicast(saddr) || ipv4_is_lbcast(saddr) ||         是否是多播、组播、环回地址等
         ipv4_is_loopback(saddr))
          goto martian_source;

if (daddr == htonl(0xFFFFFFFF) || (saddr == 0 && daddr == 0))
          goto brd_input;

/* Accept zero addresses only to limited broadcast;
     * I even do not know to fix it or not. Waiting for complains :-)
     */
     if (ipv4_is_zeronet(saddr))               源地址是否是零网地址类型
          goto martian_source;

if (ipv4_is_lbcast(daddr) || ipv4_is_zeronet(daddr) ||
         ipv4_is_loopback(daddr))
          goto martian_destination;

/*
     *     Now we are ready to route packet.
     */
     if ((err = fib_lookup(net, &fl, &res)) != 0) {                 通过路由函数查找目标地址   结果记录在res中
          if (!IN_DEV_FORWARD(in_dev))                         如果设备不支持转发
               goto e_hostunreach;
          goto no_route;
     }
     free_res = 1;                                          默认为释放查找结果

RT_CACHE_STAT_INC(in_slow_tot);

if (res.type == RTN_BROADCAST)            路由类型为广播   
          goto brd_input;

if (res.type == RTN_LOCAL) {                  路由类型为本地类型
          int result;
          result = fib_validate_source(saddr, daddr, tos,                         检查源地址
                              net->loopback_dev->ifindex,
                              dev, &spec_dst, &itag, skb->mark);
          if (result < 0)
               goto martian_source;                     源地址错误
          if (result)
               flags |= RTCF_DIRECTSRC;         
          spec_dst = daddr;                              记录目标地址
          goto local_input;                                本地输入，跳转
     }

if (!IN_DEV_FORWARD(in_dev))                       如果设备不支持转发
          goto e_hostunreach;
     if (res.type != RTN_UNICAST)                         如果目标地址不是单播类型
          goto martian_destination;                          目标地址错误，跳转

err = ip_mkroute_input(skb, &res, &fl, in_dev, daddr, saddr, tos);         创建用于转发的路由表   
done:
     in_dev_put(in_dev);                
     if (free_res)                                             如果需要释放
          fib_res_put(&res);
out:     return err;

brd_input:                            广播输入
     if (skb->protocol != htons(ETH_P_IP))            不是ip协议
          goto e_inval;

if (ipv4_is_zeronet(saddr))
          spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
     else {
          err = fib_validate_source(saddr, 0, tos, 0, dev, &spec_dst,          检查源地址的有效性
                           &itag, skb->mark);
          if (err < 0)
               goto martian_source;
          if (err)
               flags |= RTCF_DIRECTSRC;        
     }
     flags |= RTCF_BROADCAST;             增加广播标志

     res.type = RTN_BROADCAST;               设置地址类型为广播类型
     RT_CACHE_STAT_INC(in_brd);

local_input:                                            本地输入
     rth = dst_alloc(&ipv4_dst_ops);           创建路由表
     if (!rth)
          goto e_nobufs;

rth->u.dst.output= ip_rt_bug;                       设置输出方向的函数
     rth->rt_genid = rt_genid(net);                         产生随机值

atomic_set(&rth->u.dst.__refcnt, 1);              
     rth->u.dst.flags= DST_HOST;                            设置路由标志
     if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
          rth->u.dst.flags |= DST_NOPOLICY;
     rth->fl.fl4_dst     = daddr;                      记录目标地址
     rth->rt_dst     = daddr;                     
     rth->fl.fl4_tos     = tos;                       记录TOS
     rth->fl.mark    = skb->mark;                  记录掩码
     rth->fl.fl4_src     = saddr;                    记录源地址
     rth->rt_src     = saddr;
#ifdef CONFIG_NET_CLS_ROUTE
     rth->u.dst.tclassid = itag;
#endif
     rth->rt_iif     =
     rth->fl.iif     = dev->ifindex;                          记录网络设备id
     rth->u.dst.dev     = net->loopback_dev;         记录环回设备
     dev_hold(rth->u.dst.dev);
     rth->idev     = in_dev_get(rth->u.dst.dev);
     rth->rt_gateway     = daddr;                          记录网关地址
     rth->rt_spec_dst= spec_dst;                          记录指定目标地址
     rth->u.dst.input= ip_local_deliver;                   设置输入函数         ip_local_deliver在这里设置，(文章开头的疑问？）
     rth->rt_flags      = flags|RTCF_LOCAL;           增加本地路由地址
     if (res.type == RTN_UNREACHABLE) {             如果目标地址不可达
          rth->u.dst.input= ip_error;
          rth->u.dst.error= -err;
          rth->rt_flags      &= ~RTCF_LOCAL;
     }
     rth->rt_type     = res.type;                           设置地址类型
     hash = rt_hash(daddr, saddr, fl.iif, rt_genid(net));          计算hash值
     err = rt_intern_hash(hash, rth, NULL, skb);        将路由表插入hash队列并记录到数据包中
     goto done;

no_route:
     RT_CACHE_STAT_INC(in_no_route);
     spec_dst = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE);   确定指定目标
     res.type = RTN_UNREACHABLE;                          设置不可达
     if (err == -ESRCH)
          err = -ENETUNREACH;
     goto local_input;

/*
     *     Do not cache martian addresses: they should be logged (RFC1812)
     */
martian_destination:                                    目标地址错
     RT_CACHE_STAT_INC(in_martian_dst);
#ifdef CONFIG_IP_ROUTE_VERBOSE
     if (IN_DEV_LOG_MARTIANS(in_dev) && net_ratelimit())
          printk(KERN_WARNING "martian destination %pI4 from %pI4, dev %s\n",
               &daddr, &saddr, dev->name);
#endif

e_hostunreach:                                           主机不可达错误
     err = -EHOSTUNREACH;
     goto done;

e_inval:                                                       无法识别
     err = -EINVAL;
     goto done;

e_nobufs:                                                     空间不足
     err = -ENOBUFS;
     goto done;

martian_source:                                           源地址错误
     ip_handle_martian_source(dev, in_dev, skb, daddr, saddr);
     goto e_inval;
}

---

在上面函数的分析中可以看到，对于目标地址是转发情况的，调用ip_mkroute_input()函数创建转发路由表

对于广播或者本地类型，直接分配路由表并初始化，并直接指定下一步处理函数为ip_local_deliver()

---

static int ip_mkroute_input(struct sk_buff *skb,
                   struct fib_result *res,
                   const struct flowi *fl,
                   struct in_device *in_dev,
                   __be32 daddr, __be32 saddr, u32 tos)
{
     struct rtable* rth = NULL;              
     int err;
     unsigned hash;

#ifdef CONFIG_IP_ROUTE_MULTIPATH                   多路径选择
     if (res->fi && res->fi->fib_nhs > 1 && fl->oif == 0)
          fib_select_multipath(fl, res); 
#endif

/* create a routing cache entry */
     err = __mkroute_input(skb, res, in_dev, daddr, saddr, tos, &rth);        创建路由表
     if (err)
          return err;

/* put it into the cache */
     hash = rt_hash(daddr, saddr, fl->iif,
                 rt_genid(dev_net(rth->u.dst.dev)));    计算hash值
     return rt_intern_hash(hash, rth, NULL, skb);    将路由表插入路由高速缓存队列中，并记录到数据包中
}

---

static int __mkroute_input(struct sk_buff *skb,
                  struct fib_result *res,
                  struct in_device *in_dev,
                  __be32 daddr, __be32 saddr, u32 tos,
                  struct rtable **result)
{

struct rtable *rth;
     int err;
     struct in_device *out_dev;
     unsigned flags = 0;
     __be32 spec_dst;
     u32 itag;

/* get a working reference to the output device */
     out_dev = in_dev_get(FIB_RES_DEV(*res));               取出输出设备的配置结构
     if (out_dev == NULL) {             
          if (net_ratelimit())
               printk(KERN_CRIT "Bug in ip_route_input" \
                      "_slow(). Please, report\n");
          return -EINVAL;
     }

err = fib_validate_source(saddr, daddr, tos, FIB_RES_OIF(*res),
                      in_dev->dev, &spec_dst, &itag, skb->mark);        检查源地址
     if (err < 0) {
          ip_handle_martian_source(in_dev->dev, in_dev, skb, daddr,
                         saddr);

err = -EINVAL;
          goto cleanup;
     }

if (err)
          flags |= RTCF_DIRECTSRC;

if (out_dev == in_dev && err &&
         (IN_DEV_SHARED_MEDIA(out_dev) ||
          inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
          flags |= RTCF_DOREDIRECT;

if (skb->protocol != htons(ETH_P_IP)) {               如果不是ip
          /* Not IP (i.e. ARP). Do not create route, if it is
          * invalid for proxy arp. DNAT routes are always valid.
          */
          if (out_dev == in_dev) {
               err = -EINVAL;
               goto cleanup;
          }
     }

rth = dst_alloc(&ipv4_dst_ops);                申请路由项
     if (!rth) {
          err = -ENOBUFS;
          goto cleanup;
     }

atomic_set(&rth->u.dst.__refcnt, 1);
     rth->u.dst.flags= DST_HOST;
     if (IN_DEV_CONF_GET(in_dev, NOPOLICY))
          rth->u.dst.flags |= DST_NOPOLICY;
     if (IN_DEV_CONF_GET(out_dev, NOXFRM))
          rth->u.dst.flags |= DST_NOXFRM;
     rth->fl.fl4_dst     = daddr;
     rth->rt_dst     = daddr;
     rth->fl.fl4_tos     = tos;
     rth->fl.mark    = skb->mark;
     rth->fl.fl4_src     = saddr;
     rth->rt_src     = saddr;
     rth->rt_gateway     = daddr;
     rth->rt_iif      =
          rth->fl.iif     = in_dev->dev->ifindex;
     rth->u.dst.dev     = (out_dev)->dev;
     dev_hold(rth->u.dst.dev);
     rth->idev     = in_dev_get(rth->u.dst.dev);
     rth->fl.oif      = 0;
     rth->rt_spec_dst= spec_dst;

rth->u.dst.input = ip_forward;                     设置输入函数         ip_forward在这里设置，(文章开头的疑问？）

     rth->u.dst.output = ip_output;                      设置输出函数
     rth->rt_genid = rt_genid(dev_net(rth->u.dst.dev));

rt_set_nexthop(rth, res, itag);

rth->rt_flags = flags;

*result = rth;
     err = 0;
cleanup:
     /* release the working reference to the output device */
     in_dev_put(out_dev);
     return err;
}

---

经过路由查找分析，我们看到ip层在选择转发或者本地上送的选择是从路由信息里来的，而查找路由信息的过程则是先从

rt_hash_table中查找，如果其中查不到则通过fib_lookup进行
查找，形成路由信息，根据目标地址类型是本地还是转发选择了不同的输入函数，这样ip层的后续投递就会有了两种选择路径， 当然
把查到的路由信息插入到rt_hash_table，更新skb的dst。