Chinaunix首页 | 论坛 | 博客
  • 博客访问: 4516189
  • 博文数量: 252
  • 博客积分: 5347
  • 博客等级: 大校
  • 技术积分: 13838
  • 用 户 组: 普通用户
  • 注册时间: 2009-09-30 10:13
文章分类
文章存档

2022年(12)

2017年(11)

2016年(7)

2015年(14)

2014年(20)

2012年(9)

2011年(20)

2010年(153)

2009年(6)

分类: LINUX

2014-08-15 19:25:32

   转载请注明出处:http://blog.chinaunix.net/uid-20788636-id-4420258.html

2.3  inet_csk_get_port函数

inet_connection_sock.c文件中的inet_csk_get_port函数分析。

int inet_csk_get_port(struct sock *sk, unsigned short snum)

{

         struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;/*TCP散列表管理结构实例tcp_hashinfo,在tcp.c文件中tcp_init函数中进行了初始化工作,在tcp_ipv4.c文件中,struct proto tcp_prot结构体对其进行赋值 .h.hashinfo                 = &tcp_hashinfo,*/

         struct inet_bind_hashbucket *head;

         struct inet_bind_bucket *tb;

         int ret, attempts = 5;

         struct net *net = sock_net(sk);

         int smallest_size = -1, smallest_rover;

         kuid_t uid = sock_i_uid(sk);//运行经常的用户ID

 

         local_bh_disable();

         if (!snum) {//如果用户绑定的端口为0,就选择一个可用的本地端口

                   int remaining, rover, low, high;

 

again:

                   inet_get_local_port_range(net, &low, &high); //获取到本地可以使用的端口范围--1

                   remaining = (high - low) + 1; //最大重新分配的次数

                   smallest_rover = rover = prandom_u32() % remaining + low; //随机生成的端口号,赋值给rover.

 

                   smallest_size = -1;

//下面的while循环代码是根据获取到的空闲的端口号和bhash_sizebhash上取得HASH值对应的链表,然后遍历链表,对比链表中是否有获取到的空闲端口号,如果有该端口号,说明获取的该端口号已经被占用,如果已经被占用则将获取的端口号加一,如果大于最大值,则从最小值开始重新遍历端口列表,直到尝试成功的次数为remaining.

                   do {

                            if (inet_is_reserved_local_port(rover))//如果是保留端口直接寻找下一个接口

                                     goto next_nolock;

                            head = &hashinfo->bhash[inet_bhashfn(net, rover,

                                               hashinfo->bhash_size)];

                            spin_lock(&head->lock);

                            inet_bind_bucket_for_each(tb, &head->chain)

                                     if (net_eq(ib_net(tb), net) && tb->port == rover) {

/*下面的这段代码的判断就是判断端口是否可以被复用,如果可以被复用即使在绑定表中,也优先使用可以复用的端口*/

                                              if (((tb->fastreuse > 0 &&

                                                     sk->sk_reuse &&

                                                     sk->sk_state != TCP_LISTEN) ||

                                                    (tb->fastreuseport > 0 &&

                                                     sk->sk_reuseport &&

                                                     uid_eq(tb->fastuid, uid))) &&

                                                   (tb->num_owners < smallest_size || smallest_size == -1)) {

                                                        smallest_size = tb->num_owners;//记下端口使用者的个数

                                                        smallest_rover = rover;

/*如果绑定端口的个数大于端口的可用个数,就会判断是否有绑定冲突*/

                                                        if (atomic_read(&hashinfo->bsockets) > (high - low) + 1 &&

                                                            !inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb, false)) {

                                                                 snum = smallest_rover;//如果没有绑定冲突使用该端口 调用inet_csk_bind_conflict

                                                                 goto tb_found; //跳转到找到该端口处理

                                                        }

                                               }

//检查端口绑定是否有冲突,如果没有冲突就使用该端口inet_csk_bind_conflict

                                               if (!inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb, false)) {

                                                        snum = rover;

                                                        goto tb_found; //跳转到找到该端口处理

                                               }

                                               goto next;//此端口在绑定表中,但是不能复用,寻找下一个

                                     }

                            break;//如果不在绑定表中,则该端口可以使用,直接跳出循环

                   next:

                            spin_unlock(&head->lock);

                   next_nolock:

                            if (++rover > high)//如果找到的端口号,大于端口的上限值,则把最小端口赋值给rover

                                     rover = low;

                   } while (--remaining > 0);

 

                   /* Exhausted local port range during search?  It is not

                    * possible for us to be holding one of the bind hash

                    * locks if this test triggers, because if 'remaining'

                    * drops to zero, we broke out of the do/while loop at

                    * the top level, not from the 'break;' statement.

                    */

                   ret = 1;

                   if (remaining <= 0) {//这里主要是在没有查到的情况下,再给次最后一次机会

                            if (smallest_size != -1) {

                                     snum = smallest_rover;

                                     goto have_snum;

                            }

                            goto fail;

                   }

                   /* OK, here is the one we will use.  HEAD is

                    * non-NULL and we hold it's mutex.

                    */

                   snum = rover;//找到绑定的端口号

         } else {//如果指定端口号,则在相应的绑定链表中进行查询。

have_snum:

                   head = &hashinfo->bhash[inet_bhashfn(net, snum,

                                     hashinfo->bhash_size)];

                   spin_lock(&head->lock);

                   inet_bind_bucket_for_each(tb, &head->chain)

                            if (net_eq(ib_net(tb), net) && tb->port == snum)

                                     goto tb_found;//在绑定表中查找,表示该端口已经绑定

         }

         tb = NULL;//如果指定的端口在绑定表中没有发现,直接创建

         goto tb_not_found;

tb_found:

         if (!hlist_empty(&tb->owners)) {//该端口绑定Socket

                   if (sk->sk_reuse == SK_FORCE_REUSE)

                            goto success;//如果该Socket设置了SK_FORCE_REUSE,表示可以强制复用

 

                   if (((tb->fastreuse > 0 &&

                         sk->sk_reuse && sk->sk_state != TCP_LISTEN) ||

                        (tb->fastreuseport > 0 &&

                         sk->sk_reuseport && uid_eq(tb->fastuid, uid))) &&

                       smallest_size == -1) {/*判断端口是否可以复用,其中fastreuseportGoogle添加的一个SOCKET bind选项信息,tb 配置启用了 reuseport,并且当前 socket 也设置 reuseport,且 tb 和当前 socket UID 一样,可以认为当前 socket 也可以放到 bind hash 中,随后会调用 inet_bind_hash 将当前 sock 也加入到 tb->owners 链表中*/

                            goto success;

                   } else {//如果是指定端口的话,else应该不会执行

                            ret = 1;

                            if (inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb, true)) {//如果绑定冲突,进行5次尝试查找端口号 attempts = 5;由于在查找时进行了类似的判断,该判断条件基本不会成立,直接执行tb_not_found点,这时tb不为空

                                     if (((sk->sk_reuse && sk->sk_state != TCP_LISTEN) ||

                                          (tb->fastreuseport > 0 &&

                                           sk->sk_reuseport && uid_eq(tb->fastuid, uid))) &&

                                         smallest_size != -1 && --attempts >= 0) {

                                               spin_unlock(&head->lock);

                                               goto again;

                                     }

 

                                     goto fail_unlock;

                            }

                   }

         }

tb_not_found:

         ret = 1;//如果在绑定表中没有发现,则创建

         if (!tb && (tb = inet_bind_bucket_create(hashinfo->bind_bucket_cachep,

                                               net, head, snum)) == NULL)

                   goto fail_unlock;

         if (hlist_empty(&tb->owners)) {//如果没有绑定Socket

                   if (sk->sk_reuse && sk->sk_state != TCP_LISTEN)

                            tb->fastreuse = 1;

                   else

                            tb->fastreuse = 0;

                   if (sk->sk_reuseport) {//如果Socket设定了SO_REUSEPORT选项,就对fastreuseport进行赋值1

                            tb->fastreuseport = 1;

                            tb->fastuid = uid;//创建当前fd的的UID

                   } else

                            tb->fastreuseport = 0;

         } else {//如果绑定了Socket

                   if (tb->fastreuse &&

                       (!sk->sk_reuse || sk->sk_state == TCP_LISTEN))

                            tb->fastreuse = 0;

                   if (tb->fastreuseport &&

                       (!sk->sk_reuseport || !uid_eq(tb->fastuid, uid)))

                            tb->fastreuseport = 0;

         }

success://如果成功找到一个可用的端口。添加到绑定表中

         if (!inet_csk(sk)->icsk_bind_hash)

                   inet_bind_hash(sk, tb, snum);//把当前的sock插入到owers中,

         WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);

         ret = 0;

 

fail_unlock:

         spin_unlock(&head->lock);

fail:

         local_bh_enable();

         return ret;

}

2.3.1  inet_get_local_port_range()

inet_get_local_port_range()获取本地可用端口的范围,从下面的定义可以知道端口的范围为32768—61000。如果用户空间绑定的本地端口为0的话,会自动为套接口分配一个可用的端口。

/*

 * This struct holds the first and last local port number.

 */

struct local_ports sysctl_local_ports __read_mostly = {

         .lock = SEQLOCK_UNLOCKED,

         .range = { 32768, 61000 },

};

 

void inet_get_local_port_range(int *low, int *high)

{

         unsigned seq;

         do {

                   seq = read_seqbegin(&sysctl_local_ports.lock);

 

                   *low = sysctl_local_ports.range[0];

                   *high = sysctl_local_ports.range[1];

         } while (read_seqretry(&sysctl_local_ports.lock, seq));

}

2.3.2 本地端口可以被复用的条件

本地端口可以被复用的几个条件如下:

          = 1 \* alphabetic a) 如果Socket绑定在不同的接口上,可以共享同一个本地端口。

          = 2 \* alphabetic b)如果sockets设置了sk->sk_reuse,并且这些Sockets的状态都不是TCP_LISTEN,端口可以被复用

 = 3 \* alphabetic c 如果Scokets绑定了特定的inet_sk(sk)->rcv_saddr 本地地址,并且这样地址不相同,端口可以被复用。

如果不满足上面三个条件之一,则端口不能够被复用。

2.3.3  inet_bind_bucket 结构体

struct inet_bind_bucket {

#ifdef CONFIG_NET_NS

         struct net                   *ib_net;

#endif

         unsigned short                   port;//端口号

         signed char                fastreuse;//地址复用SO_REUSEADDR

         signed char                fastreuseport;//端口号复用

         kuid_t                          fastuid;//进程的用户ID

         int                       num_owners;//端口使用者的个数

         struct hlist_node     node;//指向下一个端口的inet_bind_bucket

         struct hlist_head     owners;//使用这个端口的Scoket链表

};

2.3.4  inet_csk_bind_conflict函数

inet_csk_bind_conflict检查端口是否冲突,返回0表示可以绑定,不冲突,返回1表示无法绑定该端口号

int inet_csk_bind_conflict(const struct sock *sk,

                               const struct inet_bind_bucket *tb, bool relax)

{

         struct sock *sk2;

         int reuse = sk->sk_reuse;// SO_REUSEADDR

         int reuseport = sk->sk_reuseport;

         kuid_t uid = sock_i_uid((struct sock *)sk);

 

         /*

          * Unlike other sk lookup places we do not check

          * for sk_net here, since _all_ the socks listed

          * in tb->owners list belong to the same net - the

          * one this bucket belongs to.

          */

/* tb->owners链表中循环检查绑定该端口的Socket,确定该端口是否冲突*/

         sk_for_each_bound(sk2, &tb->owners) {

/*这里的判断看是否冲突:第一If是判断:不是同一个socket,并且没有绑定设备,或者绑定的设备为相同*/

                   if (sk != sk2 &&

                       !inet_v6_ipv6only(sk2) &&

                       (!sk->sk_bound_dev_if ||

                        !sk2->sk_bound_dev_if ||

                        sk->sk_bound_dev_if == sk2->sk_bound_dev_if)) {

/*满足下面的条件之一a为真:1)绑定的socket不能复用;2)查找到的socket不能复用;3)查找到的socket处于监听状态;

满足下面的条件之一b为真:1)要绑定的Socket端口不能复用;2)查找到的Socket的端口不允许复用 3)在链表中查找到的socket的状态为TCP_TIME_WAIT,并且两个socket的用户ID不相等

如果ab都为真,再判断绑定的IP地址是否相同*/

                            if ((!reuse || !sk2->sk_reuse ||

                                sk2->sk_state == TCP_LISTEN) &&

                                (!reuseport || !sk2->sk_reuseport ||

                                (sk2->sk_state != TCP_TIME_WAIT &&

                                 !uid_eq(uid, sock_i_uid(sk2))))) {

      /*绑定的相同的IP*/

                                     if (!sk2->sk_rcv_saddr || !sk->sk_rcv_saddr ||

                                         sk2->sk_rcv_saddr == sk->sk_rcv_saddr)

                                               break;//检查到冲突

                            }

/*如果relexFalse,就不需要判断端口号是否可以复用,只判断地址是否可以复用*/

                            if (!relax && reuse && sk2->sk_reuse &&

                                sk2->sk_state != TCP_LISTEN) {

 

                                     if (!sk2->sk_rcv_saddr || !sk->sk_rcv_saddr ||

                                         sk2->sk_rcv_saddr == sk->sk_rcv_saddr)

                                               break;

                            }

                   }

         }

         return sk2 != NULL;//sk2不等于空,说明有冲突

}

2.3.5 inet_bind_bucket_create函数

inet_bind_bucket_create函数分配一个inet_bind_bucket结构体实例并进行初始化操作,然后绑定到已绑定端口的散列表中

struct inet_bind_bucket *inet_bind_bucket_create(struct kmem_cache *cachep,

                                                         struct net *net,

                                                         struct inet_bind_hashbucket *head,

                                                         const unsigned short snum)

{

         struct inet_bind_bucket *tb = kmem_cache_alloc(cachep, GFP_ATOMIC);

 

         if (tb != NULL) {

                   write_pnet(&tb->ib_net, hold_net(net));

                   tb->port      = snum;

                   tb->fastreuse = 0;

                   tb->fastreuseport = 0;

                   tb->num_owners = 0;//这些初始化,会根据配置的socket参数进行修改

                   INIT_HLIST_HEAD(&tb->owners);

                   hlist_add_head(&tb->node, &head->chain);

         }

         return tb;

}

2.3.6  inet_bind_hash 函数

inet_bind_hash 函数更新变量

void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb,

                       const unsigned short snum)

{

         struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;//TCP散列表管理结构实例TCP_hashinfo

 

         atomic_inc(&hashinfo->bsockets);//绑定次数加1

 

         inet_sk(sk)->inet_num = snum;//端口号赋值

         sk_add_bind_node(sk, &tb->owners);//Socket加入到tb->ownershash表中

         tb->num_owners++;//端口的绑定次数增加

         inet_csk(sk)->icsk_bind_hash = tb;

}

2.3.7 数据结构之间的关系

各种数据结构之间的关系

 

2.3.8  inet_hashinfo 结构体

struct inet_hashinfo {

         /* This is for sockets with full identity only.  Sockets here will

          * always be without wildcards and will have the following invariant:

          *

          *          TCP_ESTABLISHED <= sk->sk_state < TCP_CLOSE

          *

          */

         struct inet_ehash_bucket        *ehash;

         spinlock_t                            *ehash_locks;

         unsigned int                        ehash_mask;

         unsigned int                        ehash_locks_mask;

 

         /* Ok, let's try this, I give up, we do need a local binding

          * TCP hash as well as the others for fast bind/connect.

          */

         struct inet_bind_hashbucket  *bhash;

 

         unsigned int                        bhash_size;

         /* 4 bytes hole on 64 bit */

 

         struct kmem_cache                   *bind_bucket_cachep;

 

         /* All the above members are written once at bootup and

          * never written again _or_ are predominantly read-access.

          *

          * Now align to a new cache line as all the following members

          * might be often dirty.

          */

         /* All sockets in TCP_LISTEN state will be in here.  This is the only

          * table where wildcard'd TCP sockets can exist.  Hash function here

          * is just local port number.

          */

         struct inet_listen_hashbucket         listening_hash[INET_LHTABLE_SIZE]

                                               ____cacheline_aligned_in_smp;

 

         atomic_t                    bsockets;

};

tcp表分成了三张表ehash, bhash, listening_hash,其中ehash, listening_hash对应于socket处在TCPESTABLISHED, LISTEN状态,bhash对应于socket已绑定了本地地址。

2.4 流程和总结

1bind主要的主要是选择一个可用的端口号,如果用户没有指定端口号,则会按照一定的规则进行选择一个可用的端口号。

 

附录:

    对于Google REUSEPORT 新特性,支持多个进程或者线程绑定到相同的 IP 和端口,以提高 server 的性能。

    该特性实现了 IPv4/IPv6 TCP/UDP 协议的支持, 已经集成到 kernel 3.9 中。

核心的实现主要有三点:

1)扩展 socket option,增加 SO_REUSEPORT 选项,用来设置 reuseport

2)修改 bind 系统调用实现,以便支持可以绑定到相同的 IP 和端口

3)修改处理新建连接的实现,查找 listener 的时候,能够支持在监听相同 IP 和端口的多个 sock 之间均衡选择。

请参考:http://blog.chinaunix.net/uid-10167808-id-3807060.html

参考资料:

http://blog.csdn.net/zhangskd/article/details/13631715

http://tsecer.blog.163.com/blog/static/1501817201281211321031

阅读(14429) | 评论(0) | 转发(4) |
给主人留下些什么吧!~~