深入理解TCP协议：三次握手详解

72  tcp_set_state(sk, TCP_SYN_SENT)

109 err = tcp_connect(sk);

 1 void tcp_set_state(struct sock *sk, int state)
 2 {
 3     int oldstate = sk->sk_state;
 4 
 5     /* We defined a new enum for TCP states that are exported in BPF
 6      * so as not force the internal TCP states to be frozen. The
 7      * following checks will detect if an internal state value ever
 8      * differs from the BPF value. If this ever happens, then we will
 9      * need to remap the internal value to the BPF value before calling
10      * tcp_call_bpf_2arg.
11      */
12     BUILD_BUG_ON((int)BPF_TCP_ESTABLISHED != (int)TCP_ESTABLISHED);
13     BUILD_BUG_ON((int)BPF_TCP_SYN_SENT != (int)TCP_SYN_SENT);
14     BUILD_BUG_ON((int)BPF_TCP_SYN_RECV != (int)TCP_SYN_RECV);
15     BUILD_BUG_ON((int)BPF_TCP_FIN_WAIT1 != (int)TCP_FIN_WAIT1);
16     BUILD_BUG_ON((int)BPF_TCP_FIN_WAIT2 != (int)TCP_FIN_WAIT2);
17     BUILD_BUG_ON((int)BPF_TCP_TIME_WAIT != (int)TCP_TIME_WAIT);
18     BUILD_BUG_ON((int)BPF_TCP_CLOSE != (int)TCP_CLOSE);
19     BUILD_BUG_ON((int)BPF_TCP_CLOSE_WAIT != (int)TCP_CLOSE_WAIT);
20     BUILD_BUG_ON((int)BPF_TCP_LAST_ACK != (int)TCP_LAST_ACK);
21     BUILD_BUG_ON((int)BPF_TCP_LISTEN != (int)TCP_LISTEN);
22     BUILD_BUG_ON((int)BPF_TCP_CLOSING != (int)TCP_CLOSING);
23     BUILD_BUG_ON((int)BPF_TCP_NEW_SYN_RECV != (int)TCP_NEW_SYN_RECV);
24     BUILD_BUG_ON((int)BPF_TCP_MAX_STATES != (int)TCP_MAX_STATES);
25 
26     if (BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk), BPF_SOCK_OPS_STATE_CB_FLAG))
27         tcp_call_bpf_2arg(sk, BPF_SOCK_OPS_STATE_CB, oldstate, state);
28 
29     switch (state) {
30     case TCP_ESTABLISHED:
31         if (oldstate != TCP_ESTABLISHED)
32             TCP_INC_STATS(sock_net(sk), TCP_MIB_CURRESTAB);
33         break;
34 
35     case TCP_CLOSE:
36         if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED)
37             TCP_INC_STATS(sock_net(sk), TCP_MIB_ESTABRESETS);
38 
39         sk->sk_prot->unhash(sk);
40         if (inet_csk(sk)->icsk_bind_hash &&
41             !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
42             inet_put_port(sk);
43         /* fall through */
44     default:
45         if (oldstate == TCP_ESTABLISHED)
46             TCP_DEC_STATS(sock_net(sk), TCP_MIB_CURRESTAB);
47     }
48 
49     /* Change state AFTER socket is unhashed to avoid closed
50      * socket sitting in hash tables.
51      */
52     inet_sk_state_store(sk, state);
53 }

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

 1 static void tcp_init_nondata_skb(struct sk_buff *skb, u32 seq, u8 flags)
 2 {
 3     skb->ip_summed = CHECKSUM_PARTIAL;
 4 
 5     TCP_SKB_CB(skb)->tcp_flags = flags;
 6     TCP_SKB_CB(skb)->sacked = 0;
 7 
 8     tcp_skb_pcount_set(skb, 1);
 9 
10     TCP_SKB_CB(skb)->seq = seq;
11     if (flags & (TCPHDR_SYN | TCPHDR_FIN))
12         seq++;
13     TCP_SKB_CB(skb)->end_seq = seq;
14 }

 1 int __sys_accept4(int fd, struct sockaddr __user *upeer_sockaddr,
 2           int __user *upeer_addrlen, int flags)
 3 {
 4     struct socket *sock, *newsock;
 5     struct file *newfile;
 6     int err, len, newfd, fput_needed;
 7     struct sockaddr_storage address;
 8 
 9     if (flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
10         return -EINVAL;
11 
12     if (SOCK_NONBLOCK != O_NONBLOCK && (flags & SOCK_NONBLOCK))
13         flags = (flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
14 
15     sock = sockfd_lookup_light(fd, &err, &fput_needed);
16     if (!sock)
17         goto out;
18 
19     err = -ENFILE;
20     newsock = sock_alloc();
21     if (!newsock)
22         goto out_put;
23 
24     newsock->type = sock->type;
25     newsock->ops = sock->ops;
26 
27     /*
28      * We don‘t need try_module_get here, as the listening socket (sock)
29      * has the protocol module (sock->ops->owner) held.
30      */
31     __module_get(newsock->ops->owner);
32 
33     newfd = get_unused_fd_flags(flags);
34     if (unlikely(newfd < 0)) {
35         err = newfd;
36         sock_release(newsock);
37         goto out_put;
38     }
39     newfile = sock_alloc_file(newsock, flags, sock->sk->sk_prot_creator->name);
40     if (IS_ERR(newfile)) {
41         err = PTR_ERR(newfile);
42         put_unused_fd(newfd);
43         goto out_put;
44     }
45 
46     err = security_socket_accept(sock, newsock);
47     if (err)
48         goto out_fd;
49 
50     err = sock->ops->accept(sock, newsock, sock->file->f_flags, false);
51     if (err < 0)
52         goto out_fd;
53 
54     if (upeer_sockaddr) {
55         len = newsock->ops->getname(newsock,
56                     (struct sockaddr *)&address, 2);
57         if (len < 0) {
58             err = -ECONNABORTED;
59             goto out_fd;
60         }
61         err = move_addr_to_user(&address,
62                     len, upeer_sockaddr, upeer_addrlen);
63         if (err < 0)
64             goto out_fd;
65     }
66 
67     /* File flags are not inherited via accept() unlike another OSes. */
68 
69     fd_install(newfd, newfile);
70     err = newfd;
71 
72 out_put:
73     fput_light(sock->file, fput_needed);
74 out:
75     return err;
76 out_fd:
77     fput(newfile);
78     put_unused_fd(newfd);
79     goto out_put;
80 }

err = sock->ops->accept(sock, newsock, sock->file->f_flags, false);

 1 struct sock *inet_csk_accept(struct sock *sk, int flags, int *err, bool kern)
 2 {
 3     struct inet_connection_sock *icsk = inet_csk(sk);
 4     struct request_sock_queue *queue = &icsk->icsk_accept_queue;
 5     struct request_sock *req;
 6     struct sock *newsk;
 7     int error;
 8 
 9     lock_sock(sk);
10 
11     /* We need to make sure that this socket is listening,
12      * and that it has something pending.
13      */
14     error = -EINVAL;
15     if (sk->sk_state != TCP_LISTEN)
16         goto out_err;
17 
18     /* Find already established connection */
19     if (reqsk_queue_empty(queue)) {
20         long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
21 
22         /* If this is a non blocking socket don‘t sleep */
23         error = -EAGAIN;
24         if (!timeo)
25             goto out_err;
26 
27         error = inet_csk_wait_for_connect(sk, timeo);
28         if (error)
29             goto out_err;
30     }
31     req = reqsk_queue_remove(queue, sk);
32     newsk = req->sk;
33 
34     if (sk->sk_protocol == IPPROTO_TCP &&
35         tcp_rsk(req)->tfo_listener) {
36         spin_lock_bh(&queue->fastopenq.lock);
37         if (tcp_rsk(req)->tfo_listener) {
38             /* We are still waiting for the final ACK from 3WHS
39              * so can‘t free req now. Instead, we set req->sk to
40              * NULL to signify that the child socket is taken
41              * so reqsk_fastopen_remove() will free the req
42              * when 3WHS finishes (or is aborted).
43              */
44             req->sk = NULL;
45             req = NULL;
46         }
47         spin_unlock_bh(&queue->fastopenq.lock);
48     }
49 out:
50     release_sock(sk);
51     if (req)
52         reqsk_put(req);
53     return newsk;
54 out_err:
55     newsk = NULL;
56     req = NULL;
57     *err = error;
58     goto out;
59 }