blob: 5d26056b6d8f01f4db815423d2d78f6f7d25d55f [file] [log] [blame]
Arnaldo Carvalho de Melo0e875062005-06-18 22:47:59 -07001/*
2 * NET Generic infrastructure for Network protocols.
3 *
4 * Authors: Arnaldo Carvalho de Melo <acme@conectiva.com.br>
5 *
6 * From code originally in include/net/tcp.h
7 *
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License
10 * as published by the Free Software Foundation; either version
11 * 2 of the License, or (at your option) any later version.
12 */
13
14#include <linux/module.h>
15#include <linux/random.h>
16#include <linux/slab.h>
17#include <linux/string.h>
Jerry Chu83368862012-08-31 12:29:12 +000018#include <linux/tcp.h>
Eric Dumazet72a3eff2006-11-16 02:30:37 -080019#include <linux/vmalloc.h>
Arnaldo Carvalho de Melo0e875062005-06-18 22:47:59 -070020
21#include <net/request_sock.h>
22
David S. Millere52c1f12005-06-18 22:49:40 -070023/*
24 * Maximum number of SYN_RECV sockets in queue per LISTEN socket.
25 * One SYN_RECV socket costs about 80bytes on a 32bit machine.
26 * It would be better to replace it with a global counter for all sockets
27 * but then some measure against one socket starving all other sockets
28 * would be needed.
29 *
Peter Pan(潘卫平)99b53bd2011-12-05 21:39:41 +000030 * The minimum value of it is 128. Experiments with real servers show that
David S. Millere52c1f12005-06-18 22:49:40 -070031 * it is absolutely not enough even at 100conn/sec. 256 cures most
Peter Pan(潘卫平)99b53bd2011-12-05 21:39:41 +000032 * of problems.
33 * This value is adjusted to 128 for low memory machines,
34 * and it will increase in proportion to the memory of machine.
Eric Dumazet72a3eff2006-11-16 02:30:37 -080035 * Note : Dont forget somaxconn that may limit backlog too.
David S. Millere52c1f12005-06-18 22:49:40 -070036 */
37int sysctl_max_syn_backlog = 256;
David S. Miller493f3772010-12-02 12:14:29 -080038EXPORT_SYMBOL(sysctl_max_syn_backlog);
David S. Millere52c1f12005-06-18 22:49:40 -070039
Eric Dumazetef547f22015-10-02 11:43:37 -070040void reqsk_queue_alloc(struct request_sock_queue *queue)
Arnaldo Carvalho de Melo0e875062005-06-18 22:47:59 -070041{
Eric Dumazetfff1f302015-10-02 11:43:23 -070042 spin_lock_init(&queue->rskq_lock);
Eric Dumazet0536fcc2015-09-29 07:42:52 -070043
44 spin_lock_init(&queue->fastopenq.lock);
45 queue->fastopenq.rskq_rst_head = NULL;
46 queue->fastopenq.rskq_rst_tail = NULL;
47 queue->fastopenq.qlen = 0;
Eric Dumazet0536fcc2015-09-29 07:42:52 -070048
Norbert Kiesel3eb48012006-03-26 17:39:55 -080049 queue->rskq_accept_head = NULL;
Arnaldo Carvalho de Melo83e36092005-08-09 19:33:31 -070050}
51
Jerry Chu83368862012-08-31 12:29:12 +000052/*
53 * This function is called to set a Fast Open socket's "fastopen_rsk" field
54 * to NULL when a TFO socket no longer needs to access the request_sock.
55 * This happens only after 3WHS has been either completed or aborted (e.g.,
56 * RST is received).
57 *
58 * Before TFO, a child socket is created only after 3WHS is completed,
59 * hence it never needs to access the request_sock. things get a lot more
60 * complex with TFO. A child socket, accepted or not, has to access its
61 * request_sock for 3WHS processing, e.g., to retransmit SYN-ACK pkts,
62 * until 3WHS is either completed or aborted. Afterwards the req will stay
63 * until either the child socket is accepted, or in the rare case when the
64 * listener is closed before the child is accepted.
65 *
66 * In short, a request socket is only freed after BOTH 3WHS has completed
67 * (or aborted) and the child socket has been accepted (or listener closed).
68 * When a child socket is accepted, its corresponding req->sk is set to
69 * NULL since it's no longer needed. More importantly, "req->sk == NULL"
70 * will be used by the code below to determine if a child socket has been
71 * accepted or not, and the check is protected by the fastopenq->lock
72 * described below.
73 *
74 * Note that fastopen_rsk is only accessed from the child socket's context
75 * with its socket lock held. But a request_sock (req) can be accessed by
76 * both its child socket through fastopen_rsk, and a listener socket through
77 * icsk_accept_queue.rskq_accept_head. To protect the access a simple spin
78 * lock per listener "icsk->icsk_accept_queue.fastopenq->lock" is created.
79 * only in the rare case when both the listener and the child locks are held,
80 * e.g., in inet_csk_listen_stop() do we not need to acquire the lock.
81 * The lock also protects other fields such as fastopenq->qlen, which is
82 * decremented by this function when fastopen_rsk is no longer needed.
83 *
84 * Note that another solution was to simply use the existing socket lock
85 * from the listener. But first socket lock is difficult to use. It is not
86 * a simple spin lock - one must consider sock_owned_by_user() and arrange
87 * to use sk_add_backlog() stuff. But what really makes it infeasible is the
88 * locking hierarchy violation. E.g., inet_csk_listen_stop() may try to
89 * acquire a child's lock while holding listener's socket lock. A corner
90 * case might also exist in tcp_v4_hnd_req() that will trigger this locking
91 * order.
92 *
Eric Dumazet9439ce02015-03-17 18:32:29 -070093 * This function also sets "treq->tfo_listener" to false.
94 * treq->tfo_listener is used by the listener so it is protected by the
Jerry Chu83368862012-08-31 12:29:12 +000095 * fastopenq->lock in this function.
96 */
97void reqsk_fastopen_remove(struct sock *sk, struct request_sock *req,
98 bool reset)
99{
Eric Dumazet9439ce02015-03-17 18:32:29 -0700100 struct sock *lsk = req->rsk_listener;
101 struct fastopen_queue *fastopenq;
102
Eric Dumazet0536fcc2015-09-29 07:42:52 -0700103 fastopenq = &inet_csk(lsk)->icsk_accept_queue.fastopenq;
Jerry Chu83368862012-08-31 12:29:12 +0000104
Jerry Chu83368862012-08-31 12:29:12 +0000105 tcp_sk(sk)->fastopen_rsk = NULL;
106 spin_lock_bh(&fastopenq->lock);
107 fastopenq->qlen--;
Eric Dumazet9439ce02015-03-17 18:32:29 -0700108 tcp_rsk(req)->tfo_listener = false;
Jerry Chu83368862012-08-31 12:29:12 +0000109 if (req->sk) /* the child socket hasn't been accepted yet */
110 goto out;
111
112 if (!reset || lsk->sk_state != TCP_LISTEN) {
113 /* If the listener has been closed don't bother with the
114 * special RST handling below.
115 */
116 spin_unlock_bh(&fastopenq->lock);
Eric Dumazet13854e52015-03-15 21:12:16 -0700117 reqsk_put(req);
Jerry Chu83368862012-08-31 12:29:12 +0000118 return;
119 }
120 /* Wait for 60secs before removing a req that has triggered RST.
121 * This is a simple defense against TFO spoofing attack - by
122 * counting the req against fastopen.max_qlen, and disabling
123 * TFO when the qlen exceeds max_qlen.
124 *
125 * For more details see CoNext'11 "TCP Fast Open" paper.
126 */
Eric Dumazetfa76ce732015-03-19 19:04:20 -0700127 req->rsk_timer.expires = jiffies + 60*HZ;
Jerry Chu83368862012-08-31 12:29:12 +0000128 if (fastopenq->rskq_rst_head == NULL)
129 fastopenq->rskq_rst_head = req;
130 else
131 fastopenq->rskq_rst_tail->dl_next = req;
132
133 req->dl_next = NULL;
134 fastopenq->rskq_rst_tail = req;
135 fastopenq->qlen++;
136out:
137 spin_unlock_bh(&fastopenq->lock);
Jerry Chu83368862012-08-31 12:29:12 +0000138}