Wireshark分析报文

对TCP三次握手过程进行抓包分析,并通过Wireshark的Analyze分析出tcp握手过程,通过截图体现传输内容。

   1、捕获大量的由本地主机到远程服务器的TCP分组;

   2、浏览追踪信息

在显示筛选规则编辑框中输入“tcp”,可以看到在本地主机和服务器之间传输的一系列tcp和HTTP消息,你应该能看到包含SYN Segment的三次握手。通过Analyze的Follow TCP Stream分析出传输内容。写出其中某TCP数据包的源IP地址,目的IP地址,源端口,目的端口,窗口大小。

筛选:

 

wireshark抓取tcp三次握手 wireshark分析tcp三次握手_wireshark抓取tcp三次握手

 

 分析某一条数据包:

 

wireshark抓取tcp三次握手 wireshark分析tcp三次握手_TCP_02

 

 

wireshark抓取tcp三次握手 wireshark分析tcp三次握手_TCP_03

 

 那么:

源IP地址:172.16.72.229

目的IP地址:36.110.171.40

源端口:51859

目的端口:80

窗口大小:8192

TCP的三次握手
TCP的三次握手大家都不陌生,下面用wireshark这个工具抓包,进一步的说明三次握手的细节。

wireshark抓取tcp三次握手 wireshark分析tcp三次握手_IP_04

 

 1.由客户端发起tcp连接的请求,此时客户端发送一条报文,其中包含SYN标志位,将SYN设置为1; 以及seq位。设seq = x ; 该报文段成为SYN报文段

2.服务器收到这条报文后,返回给客户端一条报文,包含Ack位,SYN,以及seq位。 其中ack = x+1; SYN = 1; seq = y。该报文段称为SYNACK报文段

3.当客户端收到SYNACK报文段之后,客户端需要再给服务器发送另外一个报文段,进行确认。该报文段的SYN = 0, seq = x +1, ack = y+1;

TCP协议源代码跟踪分析

  1.TCP的三次握手从用户程序的角度看就是客户端connect和服务端accept建立起连接时背后的完成的工作。由上次的实验我们可以知道,在socket接口层这两个socket API函数分别对应着sys_connect和sys_accept4函数,课上老师说明, sys_connect和sys_accecpt是通过函数指针sock->opt->connect和sock->opt->accept调用了具体的函数来实现的,在即调用了tcp_v4_connect函数和inet_csk_accept函数,这两个函数进一步触及TCP数据收发过程tcp_transmit_skb和tcp_v4_rcv函数。

  在net/ipv4/tcp-ipv4.c文件下的结构体变量struct proto tcp_prot指定了TCP协议栈的访问接口函数:

1 struct proto tcp_prot = {
 2         .name            = "TCP",
 3         .owner            = THIS_MODULE,
 4         .close            = tcp_close,
 5         .pre_connect        = tcp_v4_pre_connect,
 6         .connect        = tcp_v4_connect,
 7         .disconnect        = tcp_disconnect,
 8         .accept            = inet_csk_accept,
 9         .ioctl            = tcp_ioctl,
10         .init            = tcp_v4_init_sock,
11         .destroy        = tcp_v4_destroy_sock,
12         .shutdown        = tcp_shutdown,
13         .setsockopt        = tcp_setsockopt,
14         .getsockopt        = tcp_getsockopt,
15         .keepalive        = tcp_set_keepalive,
16         .recvmsg        = tcp_recvmsg,
17         .sendmsg        = tcp_sendmsg,
18         .sendpage        = tcp_sendpage,
19         .backlog_rcv        = tcp_v4_do_rcv,
20         .release_cb        = tcp_release_cb,
21         .hash            = inet_hash,
22         .unhash            = inet_unhash,
23         .get_port        = inet_csk_get_port,
24         .enter_memory_pressure    = tcp_enter_memory_pressure,
25         .leave_memory_pressure    = tcp_leave_memory_pressure,
26         .stream_memory_free    = tcp_stream_memory_free,
27         .sockets_allocated    = &tcp_sockets_allocated,
28         .orphan_count        = &tcp_orphan_count,
29         .memory_allocated    = &tcp_memory_allocated,
30         .memory_pressure    = &tcp_memory_pressure,
31         .sysctl_mem        = sysctl_tcp_mem,
32         .sysctl_wmem_offset    = offsetof(struct net, ipv4.sysctl_tcp_wmem),
33         .sysctl_rmem_offset    = offsetof(struct net, ipv4.sysctl_tcp_rmem),
34         .max_header        = MAX_TCP_HEADER,
35         .obj_size        = sizeof(struct tcp_sock),
36         .slab_flags        = SLAB_TYPESAFE_BY_RCU,
37         .twsk_prot        = &tcp_timewait_sock_ops,
38         .rsk_prot        = &tcp_request_sock_ops,
39         .h.hashinfo        = &tcp_hashinfo,
40         .no_autobind        = true,
41     #ifdef CONFIG_COMPAT
42         .compat_setsockopt    = compat_tcp_setsockopt,
43         .compat_getsockopt    = compat_tcp_getsockopt,
44     #endif
45         .diag_destroy        = tcp_abort,
46     };

在这里,我们可以看到socket接口层里sock->opt->connect和sock->opt->accept实际调用的函数tcp_v4_connect和inet_csk_accept。

2.接下来通过MenuOS的内核调试环境设置断点跟踪tcp_v4_connect函数和inet_csk_accept函数来进一步验证三次握手的过程。

  在tcp_v4_connect处打个断点:

wireshark抓取tcp三次握手 wireshark分析tcp三次握手_wireshark抓取tcp三次握手_05

 

 可以发现tcp_v4_connect函数在net/ipv4/tcp_ipv4.c处定义,看下代码:

1 /* This will initiate an outgoing connection. */
  2     int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
  3     {
  4         struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
  5         struct inet_sock *inet = inet_sk(sk);
  6         struct tcp_sock *tp = tcp_sk(sk);
  7         __be16 orig_sport, orig_dport;
  8         __be32 daddr, nexthop;
  9         struct flowi4 *fl4;
 10         struct rtable *rt;
 11         int err;
 12         struct ip_options_rcu *inet_opt;
 13         struct inet_timewait_death_row *tcp_death_row = &sock_net(sk)->ipv4.tcp_death_row;
 14     
 15 
 16         if (addr_len < sizeof(struct sockaddr_in))
 17             return -EINVAL;
 18     
 19 
 20         if (usin->sin_family != AF_INET)
 21             return -EAFNOSUPPORT;
 22     
 23 
 24         nexthop = daddr = usin->sin_addr.s_addr;
 25         inet_opt = rcu_dereference_protected(inet->inet_opt,
 26                              lockdep_sock_is_held(sk));
 27         if (inet_opt && inet_opt->opt.srr) {
 28             if (!daddr)
 29                 return -EINVAL;
 30             nexthop = inet_opt->opt.faddr;
 31         }
 32     
 33 
 34         orig_sport = inet->inet_sport;
 35         orig_dport = usin->sin_port;
 36         fl4 = &inet->cork.fl.u.ip4;
 37         rt = ip_route_connect(fl4, nexthop, inet->inet_saddr,
 38                       RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
 39                       IPPROTO_TCP,
 40                       orig_sport, orig_dport, sk);
 41         if (IS_ERR(rt)) {
 42             err = PTR_ERR(rt);
 43             if (err == -ENETUNREACH)
 44                 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
 45             return err;
 46         }
 47     
 48 
 49         if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
 50             ip_rt_put(rt);
 51             return -ENETUNREACH;
 52         }
 53     
 54 
 55         if (!inet_opt || !inet_opt->opt.srr)
 56             daddr = fl4->daddr;
 57     
 58 
 59         if (!inet->inet_saddr)
 60             inet->inet_saddr = fl4->saddr;
 61         sk_rcv_saddr_set(sk, inet->inet_saddr);
 62     
 63 
 64         if (tp->rx_opt.ts_recent_stamp && inet->inet_daddr != daddr) {
 65             /* Reset inherited state */
 66             tp->rx_opt.ts_recent       = 0;
 67             tp->rx_opt.ts_recent_stamp = 0;
 68             if (likely(!tp->repair))
 69                 tp->write_seq       = 0;
 70         }
 71     
 72 
 73         inet->inet_dport = usin->sin_port;
 74         sk_daddr_set(sk, daddr);
 75     
 76 
 77         inet_csk(sk)->icsk_ext_hdr_len = 0;
 78         if (inet_opt)
 79             inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
 80     
 81 
 82         tp->rx_opt.mss_clamp = TCP_MSS_DEFAULT;
 83     
 84 
 85         /* Socket identity is still unknown (sport may be zero).
 86          * However we set state to SYN-SENT and not releasing socket
 87          * lock select source port, enter ourselves into the hash tables and
 88          * complete initialization after this.
 89          */
 90         tcp_set_state(sk, TCP_SYN_SENT);
 91         err = inet_hash_connect(tcp_death_row, sk);
 92         if (err)
 93             goto failure;
 94     
 95 
 96         sk_set_txhash(sk);
 97     
 98 
 99         rt = ip_route_newports(fl4, rt, orig_sport, orig_dport,
100                        inet->inet_sport, inet->inet_dport, sk);
101         if (IS_ERR(rt)) {
102             err = PTR_ERR(rt);
103             rt = NULL;
104             goto failure;
105         }
106         /* OK, now commit destination to socket.  */
107         sk->sk_gso_type = SKB_GSO_TCPV4;
108         sk_setup_caps(sk, &rt->dst);
109         rt = NULL;
110     
111 
112         if (likely(!tp->repair)) {
113             if (!tp->write_seq)
114                 tp->write_seq = secure_tcp_seq(inet->inet_saddr,
115                                    inet->inet_daddr,
116                                    inet->inet_sport,
117                                    usin->sin_port);
118             tp->tsoffset = secure_tcp_ts_off(sock_net(sk),
119                              inet->inet_saddr,
120                              inet->inet_daddr);
121         }
122     
123 
124         inet->inet_id = tp->write_seq ^ jiffies;
125     
126 
127         if (tcp_fastopen_defer_connect(sk, &err))
128             return err;
129         if (err)
130             goto failure;
131     
132 
133         err = tcp_connect(sk);
134     
135 
136         if (err)
137             goto failure;
138     
139 
140         return 0;
141     
142 
143     failure:
144         /*
145          * This unhashes the socket and releases the local port,
146          * if necessary.
147          */
148         tcp_set_state(sk, TCP_CLOSE);
149         ip_rt_put(rt);
150         sk->sk_route_caps = 0;
151         inet->inet_dport = 0;
152         return err;
153     }

分析代码,可以看出tcp_v4_connect函数的主要作用就是发起一个TCP连接,从这个函数中可以看到它调用了IP层提供的一些服务,比如ip_route_connect和ip_route_newports,同时在tcp_v4_connect函数中,调用了tcp_set_state函数,它设置了TCP_SYN_SENT,并进一步调用了tcp_connect(sk)来实际构造SYN并发送出去。

  tcp_connect函数具体负责构造一个携带SYN标志位的TCP头并发送出去,同时还设置了计时器超时重发。这个函数定义在net/ipv4/tcp_output.c文件中,看看代码:

1 /* Build a SYN and send it off. */
 2     int tcp_connect(struct sock *sk)
 3     {
 4         struct tcp_sock *tp = tcp_sk(sk);
 5         struct sk_buff *buff;
 6         int err;
 7     
 8 
 9         tcp_call_bpf(sk, BPF_SOCK_OPS_TCP_CONNECT_CB, 0, NULL);
10     
11 
12         if (inet_csk(sk)->icsk_af_ops->rebuild_header(sk))
13             return -EHOSTUNREACH; /* Routing failure or similar. */
14     
15 
16         tcp_connect_init(sk);
17     
18 
19         if (unlikely(tp->repair)) {
20             tcp_finish_connect(sk, NULL);
21             return 0;
22         }
23     
24 
25         buff = sk_stream_alloc_skb(sk, 0, sk->sk_allocation, true);
26         if (unlikely(!buff))
27             return -ENOBUFS;
28     
29 
30         tcp_init_nondata_skb(buff, tp->write_seq++, TCPHDR_SYN);
31         tcp_mstamp_refresh(tp);
32         tp->retrans_stamp = tcp_time_stamp(tp);
33         tcp_connect_queue_skb(sk, buff);
34         tcp_ecn_send_syn(sk, buff);
35         tcp_rbtree_insert(&sk->tcp_rtx_queue, buff);
36     
37 
38         /* Send off SYN; include data in Fast Open. */
39         err = tp->fastopen_req ? tcp_send_syn_data(sk, buff) :
40               tcp_transmit_skb(sk, buff, 1, sk->sk_allocation);
41         if (err == -ECONNREFUSED)
42             return err;
43     
44 
45         /* We change tp->snd_nxt after the tcp_transmit_skb() call
46          * in order to make this packet get counted in tcpOutSegs.
47          */
48         tp->snd_nxt = tp->write_seq;
49         tp->pushed_seq = tp->write_seq;
50         buff = tcp_send_head(sk);
51         if (unlikely(buff)) {
52             tp->snd_nxt    = TCP_SKB_CB(buff)->seq;
53             tp->pushed_seq    = TCP_SKB_CB(buff)->seq;
54         }
55         TCP_INC_STATS(sock_net(sk), TCP_MIB_ACTIVEOPENS);
56     
57 
58         /* Timer for repeating the SYN until an answer. */
59         inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
60                       inet_csk(sk)->icsk_rto, TCP_RTO_MAX);
61         return 0;
62     }
63     EXPORT_SYMBOL(tcp_connect);

其中tcp_transmit_skb函数将tcp数据发送出去。

  这边,客户端的一个tcp数据包发送出去了,服务端将做出什么反应呢,下面来看看服务端的inet_csk_accept函数,首先在inet_csk_accept处打上断点:

wireshark抓取tcp三次握手 wireshark分析tcp三次握手_三次握手_06

 

inet_csk_accept函数在net/ipv4/inet_connection_sock.c文件中:

1 /*
 2      * This will accept the next outstanding connection.
 3      */
 4     struct sock *inet_csk_accept(struct sock *sk, int flags, int *err, bool kern)
 5     {
 6         struct inet_connection_sock *icsk = inet_csk(sk);
 7         struct request_sock_queue *queue = &icsk->icsk_accept_queue;
 8         struct request_sock *req;
 9         struct sock *newsk;
10         int error;
11     
12 
13         lock_sock(sk);
14     
15 
16         /* We need to make sure that this socket is listening,
17          * and that it has something pending.
18          */
19         error = -EINVAL;
20         if (sk->sk_state != TCP_LISTEN)
21             goto out_err;
22     
23 
24         /* Find already established connection */
25         if (reqsk_queue_empty(queue)) {
26             long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
27     
28 
29             /* If this is a non blocking socket don't sleep */
30             error = -EAGAIN;
31             if (!timeo)
32                 goto out_err;
33     
34 
35             error = inet_csk_wait_for_connect(sk, timeo);
36             if (error)
37                 goto out_err;
38         }
39         req = reqsk_queue_remove(queue, sk);
40         newsk = req->sk;
41     
42 
43         if (sk->sk_protocol == IPPROTO_TCP &&
44             tcp_rsk(req)->tfo_listener) {
45             spin_lock_bh(&queue->fastopenq.lock);
46             if (tcp_rsk(req)->tfo_listener) {
47                 /* We are still waiting for the final ACK from 3WHS
48                  * so can't free req now. Instead, we set req->sk to
49                  * NULL to signify that the child socket is taken
50                  * so reqsk_fastopen_remove() will free the req
51                  * when 3WHS finishes (or is aborted).
52                  */
53                 req->sk = NULL;
54                 req = NULL;
55             }
56             spin_unlock_bh(&queue->fastopenq.lock);
57         }
58     out:
59         release_sock(sk);
60         if (req)
61             reqsk_put(req);
62         return newsk;
63     out_err:
64         newsk = NULL;
65         req = NULL;
66         *err = error;
67         goto out;
68     }
69     EXPORT_SYMBOL(inet_csk_accept);

服务端的inet_csk_accept函数会从请求队列中取出一个连接请求,如果队列为空则通过inet_csk_wait_for_connect阻塞住等待客户端的连接。

  inet_csk_wait_for_connect函数定义在net/ipv4/inet_connection_sock.c文件中:

1 static int inet_csk_wait_for_connect(struct sock *sk, long timeo)
 2 {
 3     struct inet_connection_sock *icsk = inet_csk(sk);
 4     DEFINE_WAIT(wait);
 5     int err;
 6     for (;;) {
 7         prepare_to_wait_exclusive(sk_sleep(sk), &wait,
 8                       TASK_INTERRUPTIBLE);
 9         release_sock(sk);
10         if (reqsk_queue_empty(&icsk->icsk_accept_queue))
11             timeo = schedule_timeout(timeo);
12         sched_annotate_sleep();
13         lock_sock(sk);
14         err = 0;
15         if (!reqsk_queue_empty(&icsk->icsk_accept_queue))
16             break;
17         err = -EINVAL;
18         if (sk->sk_state != TCP_LISTEN)
19             break;
20         err = sock_intr_errno(timeo);
21         if (signal_pending(current))
22             break;
23         err = -EAGAIN;
24         if (!timeo)
25             break;
26     }
27     finish_wait(sk_sleep(sk), &wait);
28     return err;
29 }

根据代码可以分析出整个三次握手的过程为:客户端通过tcp_v4_connect函数调用到tcp_connect函数,将请求发送数据包出去,服务器端则通过inet_csk_accept函数调用inet_csk_wait_for_connect函数中的for循环进入阻塞,直到监听到请求才跳出循环。connect启动到返回和accept返回之间就是所谓三次握手的时间。

3.三次握手中携带SYN/ACK的TCP头数据的发送和接收

  以上分析了用户程序调用socket接口、通过系统调用陷入内核进入内核态的socket接口层代码,然后根据创建socket时指定协议选择适当的函数指针进入协议处理代码中。那么网卡接收到数据后是如何通知上层协议来接收并处理数据的呢。其实在TCP/IP协议栈的初始化过程中,协议栈将handler赋值为tcp_v4_rcv的函数指针,也就是TCP协议中负责接收处理数据的入口,接收TCP连接请求及进行三次握手处理过程也都是从这里开始。

  内核在处理接收到的TCP报文时使用了4个队列容器,分别为receive、out_of_order、prequeue、backlog队列。当网卡接收到报文并判断为TCP协议后,将会调用到内核的tcp_v4_rcv方法。tcp_v4_rcv方法会把这个报文直接插入到receive队列中。

  在该函数定义在net/ipv4/tcp_ipv4.c文件中。

tcp_v4_rcv函数只要做以下几个工作:

  (1) 设置TCP_CB

  (2) 查找控制块

  (3)根据控制块状态做不同处理,包括TCP_TIME_WAIT状态处理,TCP_NEW_SYN_RECV状态处理,TCP_LISTEN状态处理

  (4) 接收TCP段

  以上完成了将接收数据放入accept队列中,之后服务端接收客户端发来的tcp报文,并发送回SYN+ACK。

当前客户端处于TCP_SYN_SENT状态,并调用tcp_rcv_synsent_state_process处理SYN_SENT状态下接收到的TCP段,发送ACK报文  

  到这里,三次握手期间tcp接收处理数据包的过程基本完成。