Commit | Line | Data |
---|---|---|
2874c5fd | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
0e87506f ACM |
2 | /* |
3 | * NET Generic infrastructure for Network protocols. | |
4 | * | |
5 | * Authors: Arnaldo Carvalho de Melo <acme@conectiva.com.br> | |
6 | * | |
7 | * From code originally in include/net/tcp.h | |
0e87506f ACM |
8 | */ |
9 | ||
10 | #include <linux/module.h> | |
11 | #include <linux/random.h> | |
12 | #include <linux/slab.h> | |
13 | #include <linux/string.h> | |
8336886f | 14 | #include <linux/tcp.h> |
72a3effa | 15 | #include <linux/vmalloc.h> |
0e87506f ACM |
16 | |
17 | #include <net/request_sock.h> | |
18 | ||
e52c1f17 DM |
19 | /* |
20 | * Maximum number of SYN_RECV sockets in queue per LISTEN socket. | |
21 | * One SYN_RECV socket costs about 80bytes on a 32bit machine. | |
22 | * It would be better to replace it with a global counter for all sockets | |
23 | * but then some measure against one socket starving all other sockets | |
24 | * would be needed. | |
25 | * | |
99b53bdd | 26 | * The minimum value of it is 128. Experiments with real servers show that |
e52c1f17 | 27 | * it is absolutely not enough even at 100conn/sec. 256 cures most |
99b53bdd PP |
28 | * of problems. |
29 | * This value is adjusted to 128 for low memory machines, | |
30 | * and it will increase in proportion to the memory of machine. | |
72a3effa | 31 | * Note : Dont forget somaxconn that may limit backlog too. |
e52c1f17 | 32 | */ |
e52c1f17 | 33 | |
ef547f2a | 34 | void reqsk_queue_alloc(struct request_sock_queue *queue) |
0e87506f | 35 | { |
0536fcc0 ED |
36 | queue->fastopenq.rskq_rst_head = NULL; |
37 | queue->fastopenq.rskq_rst_tail = NULL; | |
38 | queue->fastopenq.qlen = 0; | |
0536fcc0 | 39 | |
3eb4801d | 40 | queue->rskq_accept_head = NULL; |
83e3609e ACM |
41 | } |
42 | ||
8336886f JC |
43 | /* |
44 | * This function is called to set a Fast Open socket's "fastopen_rsk" field | |
45 | * to NULL when a TFO socket no longer needs to access the request_sock. | |
46 | * This happens only after 3WHS has been either completed or aborted (e.g., | |
47 | * RST is received). | |
48 | * | |
49 | * Before TFO, a child socket is created only after 3WHS is completed, | |
50 | * hence it never needs to access the request_sock. things get a lot more | |
51 | * complex with TFO. A child socket, accepted or not, has to access its | |
52 | * request_sock for 3WHS processing, e.g., to retransmit SYN-ACK pkts, | |
53 | * until 3WHS is either completed or aborted. Afterwards the req will stay | |
54 | * until either the child socket is accepted, or in the rare case when the | |
55 | * listener is closed before the child is accepted. | |
56 | * | |
57 | * In short, a request socket is only freed after BOTH 3WHS has completed | |
58 | * (or aborted) and the child socket has been accepted (or listener closed). | |
59 | * When a child socket is accepted, its corresponding req->sk is set to | |
60 | * NULL since it's no longer needed. More importantly, "req->sk == NULL" | |
61 | * will be used by the code below to determine if a child socket has been | |
62 | * accepted or not, and the check is protected by the fastopenq->lock | |
63 | * described below. | |
64 | * | |
65 | * Note that fastopen_rsk is only accessed from the child socket's context | |
66 | * with its socket lock held. But a request_sock (req) can be accessed by | |
67 | * both its child socket through fastopen_rsk, and a listener socket through | |
68 | * icsk_accept_queue.rskq_accept_head. To protect the access a simple spin | |
69 | * lock per listener "icsk->icsk_accept_queue.fastopenq->lock" is created. | |
70 | * only in the rare case when both the listener and the child locks are held, | |
71 | * e.g., in inet_csk_listen_stop() do we not need to acquire the lock. | |
72 | * The lock also protects other fields such as fastopenq->qlen, which is | |
73 | * decremented by this function when fastopen_rsk is no longer needed. | |
74 | * | |
75 | * Note that another solution was to simply use the existing socket lock | |
76 | * from the listener. But first socket lock is difficult to use. It is not | |
77 | * a simple spin lock - one must consider sock_owned_by_user() and arrange | |
78 | * to use sk_add_backlog() stuff. But what really makes it infeasible is the | |
79 | * locking hierarchy violation. E.g., inet_csk_listen_stop() may try to | |
80 | * acquire a child's lock while holding listener's socket lock. A corner | |
81 | * case might also exist in tcp_v4_hnd_req() that will trigger this locking | |
82 | * order. | |
83 | * | |
9439ce00 ED |
84 | * This function also sets "treq->tfo_listener" to false. |
85 | * treq->tfo_listener is used by the listener so it is protected by the | |
8336886f JC |
86 | * fastopenq->lock in this function. |
87 | */ | |
88 | void reqsk_fastopen_remove(struct sock *sk, struct request_sock *req, | |
89 | bool reset) | |
90 | { | |
9439ce00 ED |
91 | struct sock *lsk = req->rsk_listener; |
92 | struct fastopen_queue *fastopenq; | |
93 | ||
0536fcc0 | 94 | fastopenq = &inet_csk(lsk)->icsk_accept_queue.fastopenq; |
8336886f | 95 | |
d983ea6f | 96 | RCU_INIT_POINTER(tcp_sk(sk)->fastopen_rsk, NULL); |
8336886f JC |
97 | spin_lock_bh(&fastopenq->lock); |
98 | fastopenq->qlen--; | |
9439ce00 | 99 | tcp_rsk(req)->tfo_listener = false; |
8336886f JC |
100 | if (req->sk) /* the child socket hasn't been accepted yet */ |
101 | goto out; | |
102 | ||
103 | if (!reset || lsk->sk_state != TCP_LISTEN) { | |
104 | /* If the listener has been closed don't bother with the | |
105 | * special RST handling below. | |
106 | */ | |
107 | spin_unlock_bh(&fastopenq->lock); | |
13854e5a | 108 | reqsk_put(req); |
8336886f JC |
109 | return; |
110 | } | |
111 | /* Wait for 60secs before removing a req that has triggered RST. | |
112 | * This is a simple defense against TFO spoofing attack - by | |
113 | * counting the req against fastopen.max_qlen, and disabling | |
114 | * TFO when the qlen exceeds max_qlen. | |
115 | * | |
116 | * For more details see CoNext'11 "TCP Fast Open" paper. | |
117 | */ | |
fa76ce73 | 118 | req->rsk_timer.expires = jiffies + 60*HZ; |
8336886f JC |
119 | if (fastopenq->rskq_rst_head == NULL) |
120 | fastopenq->rskq_rst_head = req; | |
121 | else | |
122 | fastopenq->rskq_rst_tail->dl_next = req; | |
123 | ||
124 | req->dl_next = NULL; | |
125 | fastopenq->rskq_rst_tail = req; | |
126 | fastopenq->qlen++; | |
127 | out: | |
128 | spin_unlock_bh(&fastopenq->lock); | |
8336886f | 129 | } |