blob: 3a622e7abc02d7ce061c1a1c3eb0548cc6969e4a [file] [log] [blame]
Glenn Griffinc6aefaf2008-02-07 21:49:26 -08001/*
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
24extern int sysctl_tcp_syncookies;
Florian Westphal2051f112008-03-23 22:21:28 -070025extern __u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS];
Glenn Griffinc6aefaf2008-02-07 21:49:26 -080026
27#define COOKIEBITS 24 /* Upper bits store count */
28#define COOKIEMASK (((__u32)1 << COOKIEBITS) - 1)
29
30/*
31 * This table has to be sorted and terminated with (__u16)-1.
32 * XXX generate a better table.
33 * Unresolved Issues: HIPPI with a 64k MSS is not well supported.
34 *
35 * Taken directly from ipv4 implementation.
36 * Should this list be modified for ipv6 use or is it close enough?
37 * rfc 2460 8.3 suggests mss values 20 bytes less than ipv4 counterpart
38 */
39static __u16 const msstab[] = {
40 64 - 1,
41 256 - 1,
42 512 - 1,
43 536 - 1,
44 1024 - 1,
45 1440 - 1,
46 1460 - 1,
47 4312 - 1,
48 (__u16)-1
49};
50/* The number doesn't include the -1 terminator */
51#define NUM_MSS (ARRAY_SIZE(msstab) - 1)
52
53/*
54 * This (misnamed) value is the age of syncookie which is permitted.
55 * Its ideal value should be dependent on TCP_TIMEOUT_INIT and
56 * sysctl_tcp_retries1. It's a rather complicated formula (exponential
57 * backoff) to compute at runtime so it's currently hardcoded here.
58 */
59#define COUNTER_TRIES 4
60
61static inline struct sock *get_cookie_sock(struct sock *sk, struct sk_buff *skb,
62 struct request_sock *req,
63 struct dst_entry *dst)
64{
65 struct inet_connection_sock *icsk = inet_csk(sk);
66 struct sock *child;
67
68 child = icsk->icsk_af_ops->syn_recv_sock(sk, skb, req, dst);
69 if (child)
70 inet_csk_reqsk_queue_add(sk, req, child);
71 else
72 reqsk_free(req);
73
74 return child;
75}
76
77static DEFINE_PER_CPU(__u32, cookie_scratch)[16 + 5 + SHA_WORKSPACE_WORDS];
78
79static u32 cookie_hash(struct in6_addr *saddr, struct in6_addr *daddr,
80 __be16 sport, __be16 dport, u32 count, int c)
81{
82 __u32 *tmp = __get_cpu_var(cookie_scratch);
83
84 /*
85 * we have 320 bits of information to hash, copy in the remaining
86 * 192 bits required for sha_transform, from the syncookie_secret
87 * and overwrite the digest with the secret
88 */
89 memcpy(tmp + 10, syncookie_secret[c], 44);
90 memcpy(tmp, saddr, 16);
91 memcpy(tmp + 4, daddr, 16);
92 tmp[8] = ((__force u32)sport << 16) + (__force u32)dport;
93 tmp[9] = count;
94 sha_transform(tmp + 16, (__u8 *)tmp, tmp + 16 + 5);
95
96 return tmp[17];
97}
98
99static __u32 secure_tcp_syn_cookie(struct in6_addr *saddr, struct in6_addr *daddr,
100 __be16 sport, __be16 dport, __u32 sseq,
101 __u32 count, __u32 data)
102{
103 return (cookie_hash(saddr, daddr, sport, dport, 0, 0) +
104 sseq + (count << COOKIEBITS) +
105 ((cookie_hash(saddr, daddr, sport, dport, count, 1) + data)
106 & COOKIEMASK));
107}
108
109static __u32 check_tcp_syn_cookie(__u32 cookie, struct in6_addr *saddr,
110 struct in6_addr *daddr, __be16 sport,
111 __be16 dport, __u32 sseq, __u32 count,
112 __u32 maxdiff)
113{
114 __u32 diff;
115
116 cookie -= cookie_hash(saddr, daddr, sport, dport, 0, 0) + sseq;
117
118 diff = (count - (cookie >> COOKIEBITS)) & ((__u32) -1 >> COOKIEBITS);
119 if (diff >= maxdiff)
120 return (__u32)-1;
121
122 return (cookie -
123 cookie_hash(saddr, daddr, sport, dport, count - diff, 1))
124 & COOKIEMASK;
125}
126
127__u32 cookie_v6_init_sequence(struct sock *sk, struct sk_buff *skb, __u16 *mssp)
128{
129 struct ipv6hdr *iph = ipv6_hdr(skb);
130 const struct tcphdr *th = tcp_hdr(skb);
131 int mssind;
132 const __u16 mss = *mssp;
133
134 tcp_sk(sk)->last_synq_overflow = jiffies;
135
136 for (mssind = 0; mss > msstab[mssind + 1]; mssind++)
137 ;
138 *mssp = msstab[mssind] + 1;
139
140 NET_INC_STATS_BH(LINUX_MIB_SYNCOOKIESSENT);
141
142 return secure_tcp_syn_cookie(&iph->saddr, &iph->daddr, th->source,
143 th->dest, ntohl(th->seq),
144 jiffies / (HZ * 60), mssind);
145}
146
147static inline int cookie_check(struct sk_buff *skb, __u32 cookie)
148{
149 struct ipv6hdr *iph = ipv6_hdr(skb);
150 const struct tcphdr *th = tcp_hdr(skb);
151 __u32 seq = ntohl(th->seq) - 1;
152 __u32 mssind = check_tcp_syn_cookie(cookie, &iph->saddr, &iph->daddr,
153 th->source, th->dest, seq,
154 jiffies / (HZ * 60), COUNTER_TRIES);
155
156 return mssind < NUM_MSS ? msstab[mssind] + 1 : 0;
157}
158
159struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb)
160{
161 struct inet_request_sock *ireq;
162 struct inet6_request_sock *ireq6;
163 struct tcp_request_sock *treq;
164 struct ipv6_pinfo *np = inet6_sk(sk);
165 struct tcp_sock *tp = tcp_sk(sk);
166 const struct tcphdr *th = tcp_hdr(skb);
167 __u32 cookie = ntohl(th->ack_seq) - 1;
168 struct sock *ret = sk;
169 struct request_sock *req;
170 int mss;
171 struct dst_entry *dst;
172 __u8 rcv_wscale;
173
174 if (!sysctl_tcp_syncookies || !th->ack)
175 goto out;
176
177 if (time_after(jiffies, tp->last_synq_overflow + TCP_TIMEOUT_INIT) ||
178 (mss = cookie_check(skb, cookie)) == 0) {
179 NET_INC_STATS_BH(LINUX_MIB_SYNCOOKIESFAILED);
180 goto out;
181 }
182
183 NET_INC_STATS_BH(LINUX_MIB_SYNCOOKIESRECV);
184
185 ret = NULL;
186 req = inet6_reqsk_alloc(&tcp6_request_sock_ops);
187 if (!req)
188 goto out;
189
190 ireq = inet_rsk(req);
191 ireq6 = inet6_rsk(req);
192 treq = tcp_rsk(req);
193 ireq6->pktopts = NULL;
194
195 if (security_inet_conn_request(sk, skb, req)) {
196 reqsk_free(req);
197 goto out;
198 }
199
200 req->mss = mss;
201 ireq->rmt_port = th->source;
202 ipv6_addr_copy(&ireq6->rmt_addr, &ipv6_hdr(skb)->saddr);
203 ipv6_addr_copy(&ireq6->loc_addr, &ipv6_hdr(skb)->daddr);
204 if (ipv6_opt_accepted(sk, skb) ||
205 np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo ||
206 np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim) {
207 atomic_inc(&skb->users);
208 ireq6->pktopts = skb;
209 }
210
211 ireq6->iif = sk->sk_bound_dev_if;
212 /* So that link locals have meaning */
213 if (!sk->sk_bound_dev_if &&
214 ipv6_addr_type(&ireq6->rmt_addr) & IPV6_ADDR_LINKLOCAL)
215 ireq6->iif = inet6_iif(skb);
216
217 req->expires = 0UL;
218 req->retrans = 0;
219 ireq->snd_wscale = ireq->rcv_wscale = ireq->tstamp_ok = 0;
220 ireq->wscale_ok = ireq->sack_ok = 0;
221 treq->rcv_isn = ntohl(th->seq) - 1;
222 treq->snt_isn = cookie;
223
224 /*
225 * We need to lookup the dst_entry to get the correct window size.
226 * This is taken from tcp_v6_syn_recv_sock. Somebody please enlighten
227 * me if there is a preferred way.
228 */
229 {
230 struct in6_addr *final_p = NULL, final;
231 struct flowi fl;
232 memset(&fl, 0, sizeof(fl));
233 fl.proto = IPPROTO_TCP;
234 ipv6_addr_copy(&fl.fl6_dst, &ireq6->rmt_addr);
235 if (np->opt && np->opt->srcrt) {
236 struct rt0_hdr *rt0 = (struct rt0_hdr *) np->opt->srcrt;
237 ipv6_addr_copy(&final, &fl.fl6_dst);
238 ipv6_addr_copy(&fl.fl6_dst, rt0->addr);
239 final_p = &final;
240 }
241 ipv6_addr_copy(&fl.fl6_src, &ireq6->loc_addr);
242 fl.oif = sk->sk_bound_dev_if;
243 fl.fl_ip_dport = inet_rsk(req)->rmt_port;
244 fl.fl_ip_sport = inet_sk(sk)->sport;
245 security_req_classify_flow(req, &fl);
246 if (ip6_dst_lookup(sk, &dst, &fl)) {
247 reqsk_free(req);
248 goto out;
249 }
250 if (final_p)
251 ipv6_addr_copy(&fl.fl6_dst, final_p);
252 if ((xfrm_lookup(&dst, &fl, sk, 0)) < 0)
253 goto out;
254 }
255
256 req->window_clamp = dst_metric(dst, RTAX_WINDOW);
257 tcp_select_initial_window(tcp_full_space(sk), req->mss,
258 &req->rcv_wnd, &req->window_clamp,
259 0, &rcv_wscale);
260
261 ireq->rcv_wscale = rcv_wscale;
262
263 ret = get_cookie_sock(sk, skb, req, dst);
264
265out: return ret;
266}
267