net: Use fwmark reflection in PMTU discovery.
[linux-block.git] / net / ipv6 / syncookies.c
CommitLineData
c6aefafb
GG
1/*
2 * IPv6 Syncookies implementation for the Linux kernel
3 *
4 * Authors:
5 * Glenn Griffin <ggriffin.kernel@gmail.com>
6 *
7 * Based on IPv4 implementation by Andi Kleen
8 * linux/net/ipv4/syncookies.c
9 *
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public License
12 * as published by the Free Software Foundation; either version
13 * 2 of the License, or (at your option) any later version.
14 *
15 */
16
17#include <linux/tcp.h>
18#include <linux/random.h>
19#include <linux/cryptohash.h>
20#include <linux/kernel.h>
21#include <net/ipv6.h>
22#include <net/tcp.h>
23
c6aefafb
GG
24#define COOKIEBITS 24 /* Upper bits store count */
25#define COOKIEMASK (((__u32)1 << COOKIEBITS) - 1)
26
b23a002f
HFS
27static u32 syncookie6_secret[2][16-4+SHA_DIGEST_WORDS];
28
08629354
FW
29/* RFC 2460, Section 8.3:
30 * [ipv6 tcp] MSS must be computed as the maximum packet size minus 60 [..]
31 *
32 * Due to IPV6_MIN_MTU=1280 the lowest possible MSS is 1220, which allows
33 * using higher values than ipv4 tcp syncookies.
34 * The other values are chosen based on ethernet (1500 and 9k MTU), plus
35 * one that accounts for common encap (PPPoe) overhead. Table must be sorted.
36 */
c6aefafb 37static __u16 const msstab[] = {
08629354 38 1280 - 60, /* IPV6_MIN_MTU - 60 */
5918e2fb
FW
39 1480 - 60,
40 1500 - 60,
5918e2fb 41 9000 - 60,
c6aefafb 42};
c6aefafb 43
c6aefafb
GG
44static inline struct sock *get_cookie_sock(struct sock *sk, struct sk_buff *skb,
45 struct request_sock *req,
46 struct dst_entry *dst)
47{
48 struct inet_connection_sock *icsk = inet_csk(sk);
49 struct sock *child;
50
51 child = icsk->icsk_af_ops->syn_recv_sock(sk, skb, req, dst);
52 if (child)
53 inet_csk_reqsk_queue_add(sk, req, child);
54 else
55 reqsk_free(req);
56
57 return child;
58}
59
245b2e70
TH
60static DEFINE_PER_CPU(__u32 [16 + 5 + SHA_WORKSPACE_WORDS],
61 ipv6_cookie_scratch);
c6aefafb 62
b71d1d42 63static u32 cookie_hash(const struct in6_addr *saddr, const struct in6_addr *daddr,
c6aefafb
GG
64 __be16 sport, __be16 dport, u32 count, int c)
65{
b23a002f
HFS
66 __u32 *tmp;
67
68 net_get_random_once(syncookie6_secret, sizeof(syncookie6_secret));
69
70 tmp = __get_cpu_var(ipv6_cookie_scratch);
c6aefafb
GG
71
72 /*
73 * we have 320 bits of information to hash, copy in the remaining
b23a002f 74 * 192 bits required for sha_transform, from the syncookie6_secret
c6aefafb
GG
75 * and overwrite the digest with the secret
76 */
b23a002f 77 memcpy(tmp + 10, syncookie6_secret[c], 44);
c6aefafb
GG
78 memcpy(tmp, saddr, 16);
79 memcpy(tmp + 4, daddr, 16);
80 tmp[8] = ((__force u32)sport << 16) + (__force u32)dport;
81 tmp[9] = count;
82 sha_transform(tmp + 16, (__u8 *)tmp, tmp + 16 + 5);
83
84 return tmp[17];
85}
86
b71d1d42
ED
87static __u32 secure_tcp_syn_cookie(const struct in6_addr *saddr,
88 const struct in6_addr *daddr,
c6aefafb 89 __be16 sport, __be16 dport, __u32 sseq,
8c27bd75 90 __u32 data)
c6aefafb 91{
8c27bd75 92 u32 count = tcp_cookie_time();
c6aefafb
GG
93 return (cookie_hash(saddr, daddr, sport, dport, 0, 0) +
94 sseq + (count << COOKIEBITS) +
95 ((cookie_hash(saddr, daddr, sport, dport, count, 1) + data)
96 & COOKIEMASK));
97}
98
b71d1d42
ED
99static __u32 check_tcp_syn_cookie(__u32 cookie, const struct in6_addr *saddr,
100 const struct in6_addr *daddr, __be16 sport,
8c27bd75 101 __be16 dport, __u32 sseq)
c6aefafb 102{
8c27bd75 103 __u32 diff, count = tcp_cookie_time();
c6aefafb
GG
104
105 cookie -= cookie_hash(saddr, daddr, sport, dport, 0, 0) + sseq;
106
107 diff = (count - (cookie >> COOKIEBITS)) & ((__u32) -1 >> COOKIEBITS);
8c27bd75 108 if (diff >= MAX_SYNCOOKIE_AGE)
c6aefafb
GG
109 return (__u32)-1;
110
111 return (cookie -
112 cookie_hash(saddr, daddr, sport, dport, count - diff, 1))
113 & COOKIEMASK;
114}
115
81eb6a14
PM
116u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
117 const struct tcphdr *th, __u16 *mssp)
c6aefafb 118{
c6aefafb
GG
119 int mssind;
120 const __u16 mss = *mssp;
121
5918e2fb
FW
122 for (mssind = ARRAY_SIZE(msstab) - 1; mssind ; mssind--)
123 if (mss >= msstab[mssind])
124 break;
125
126 *mssp = msstab[mssind];
c6aefafb 127
c6aefafb 128 return secure_tcp_syn_cookie(&iph->saddr, &iph->daddr, th->source,
8c27bd75 129 th->dest, ntohl(th->seq), mssind);
c6aefafb 130}
81eb6a14 131EXPORT_SYMBOL_GPL(__cookie_v6_init_sequence);
c6aefafb 132
81eb6a14 133__u32 cookie_v6_init_sequence(struct sock *sk, const struct sk_buff *skb, __u16 *mssp)
c6aefafb 134{
b71d1d42 135 const struct ipv6hdr *iph = ipv6_hdr(skb);
c6aefafb 136 const struct tcphdr *th = tcp_hdr(skb);
81eb6a14
PM
137
138 tcp_synq_overflow(sk);
139 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
140
141 return __cookie_v6_init_sequence(iph, th, mssp);
142}
143
144int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
145 __u32 cookie)
146{
c6aefafb
GG
147 __u32 seq = ntohl(th->seq) - 1;
148 __u32 mssind = check_tcp_syn_cookie(cookie, &iph->saddr, &iph->daddr,
8c27bd75 149 th->source, th->dest, seq);
c6aefafb 150
5918e2fb 151 return mssind < ARRAY_SIZE(msstab) ? msstab[mssind] : 0;
c6aefafb 152}
81eb6a14 153EXPORT_SYMBOL_GPL(__cookie_v6_check);
c6aefafb
GG
154
155struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb)
156{
4957faad 157 struct tcp_options_received tcp_opt;
c6aefafb 158 struct inet_request_sock *ireq;
c6aefafb
GG
159 struct tcp_request_sock *treq;
160 struct ipv6_pinfo *np = inet6_sk(sk);
161 struct tcp_sock *tp = tcp_sk(sk);
162 const struct tcphdr *th = tcp_hdr(skb);
163 __u32 cookie = ntohl(th->ack_seq) - 1;
164 struct sock *ret = sk;
165 struct request_sock *req;
166 int mss;
167 struct dst_entry *dst;
168 __u8 rcv_wscale;
f0e3d068 169 bool ecn_ok = false;
c6aefafb 170
af9b4738 171 if (!sysctl_tcp_syncookies || !th->ack || th->rst)
c6aefafb
GG
172 goto out;
173
a0f82f64 174 if (tcp_synq_no_recent_overflow(sk) ||
81eb6a14 175 (mss = __cookie_v6_check(ipv6_hdr(skb), th, cookie)) == 0) {
de0744af 176 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SYNCOOKIESFAILED);
c6aefafb
GG
177 goto out;
178 }
179
de0744af 180 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SYNCOOKIESRECV);
c6aefafb 181
bb5b7c11
DM
182 /* check for timestamp cookie support */
183 memset(&tcp_opt, 0, sizeof(tcp_opt));
1a2c6181 184 tcp_parse_options(skb, &tcp_opt, 0, NULL);
bb5b7c11 185
5d134f1c 186 if (!cookie_check_timestamp(&tcp_opt, sock_net(sk), &ecn_ok))
8c763681 187 goto out;
bb5b7c11 188
c6aefafb
GG
189 ret = NULL;
190 req = inet6_reqsk_alloc(&tcp6_request_sock_ops);
191 if (!req)
192 goto out;
193
194 ireq = inet_rsk(req);
c6aefafb 195 treq = tcp_rsk(req);
8336886f 196 treq->listener = NULL;
c6aefafb 197
1730554f
FW
198 if (security_inet_conn_request(sk, skb, req))
199 goto out_free;
c6aefafb
GG
200
201 req->mss = mss;
634fb979 202 ireq->ir_rmt_port = th->source;
b44084c2 203 ireq->ir_num = ntohs(th->dest);
634fb979
ED
204 ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr;
205 ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr;
c6aefafb
GG
206 if (ipv6_opt_accepted(sk, skb) ||
207 np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo ||
208 np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim) {
209 atomic_inc(&skb->users);
634fb979 210 ireq->pktopts = skb;
c6aefafb
GG
211 }
212
634fb979 213 ireq->ir_iif = sk->sk_bound_dev_if;
c6aefafb
GG
214 /* So that link locals have meaning */
215 if (!sk->sk_bound_dev_if &&
634fb979
ED
216 ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL)
217 ireq->ir_iif = inet6_iif(skb);
c6aefafb
GG
218
219 req->expires = 0UL;
e6c022a4 220 req->num_retrans = 0;
172d69e6 221 ireq->ecn_ok = ecn_ok;
bb5b7c11 222 ireq->snd_wscale = tcp_opt.snd_wscale;
bb5b7c11
DM
223 ireq->sack_ok = tcp_opt.sack_ok;
224 ireq->wscale_ok = tcp_opt.wscale_ok;
225 ireq->tstamp_ok = tcp_opt.saw_tstamp;
226 req->ts_recent = tcp_opt.saw_tstamp ? tcp_opt.rcv_tsval : 0;
9ad7c049 227 treq->snt_synack = tcp_opt.saw_tstamp ? tcp_opt.rcv_tsecr : 0;
c6aefafb
GG
228 treq->rcv_isn = ntohl(th->seq) - 1;
229 treq->snt_isn = cookie;
230
231 /*
232 * We need to lookup the dst_entry to get the correct window size.
233 * This is taken from tcp_v6_syn_recv_sock. Somebody please enlighten
234 * me if there is a preferred way.
235 */
236 {
20c59de2 237 struct in6_addr *final_p, final;
4c9483b2
DM
238 struct flowi6 fl6;
239 memset(&fl6, 0, sizeof(fl6));
240 fl6.flowi6_proto = IPPROTO_TCP;
634fb979 241 fl6.daddr = ireq->ir_v6_rmt_addr;
4c9483b2 242 final_p = fl6_update_dst(&fl6, np->opt, &final);
634fb979 243 fl6.saddr = ireq->ir_v6_loc_addr;
4c9483b2
DM
244 fl6.flowi6_oif = sk->sk_bound_dev_if;
245 fl6.flowi6_mark = sk->sk_mark;
634fb979 246 fl6.fl6_dport = ireq->ir_rmt_port;
1958b856 247 fl6.fl6_sport = inet_sk(sk)->inet_sport;
4c9483b2
DM
248 security_req_classify_flow(req, flowi6_to_flowi(&fl6));
249
0e0d44ab 250 dst = ip6_dst_lookup_flow(sk, &fl6, final_p);
68d0c6d3 251 if (IS_ERR(dst))
1730554f 252 goto out_free;
c6aefafb
GG
253 }
254
4dfc2817 255 req->window_clamp = tp->window_clamp ? :dst_metric(dst, RTAX_WINDOW);
c6aefafb
GG
256 tcp_select_initial_window(tcp_full_space(sk), req->mss,
257 &req->rcv_wnd, &req->window_clamp,
31d12926 258 ireq->wscale_ok, &rcv_wscale,
259 dst_metric(dst, RTAX_INITRWND));
c6aefafb
GG
260
261 ireq->rcv_wscale = rcv_wscale;
262
263 ret = get_cookie_sock(sk, skb, req, dst);
1730554f
FW
264out:
265 return ret;
266out_free:
267 reqsk_free(req);
268 return NULL;
c6aefafb
GG
269}
270