net: add Documentation/networking/scaling.txt
[linux-2.6-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
24extern int sysctl_tcp_syncookies;
2051f11f 25extern __u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS];
c6aefafb
GG
26
27#define COOKIEBITS 24 /* Upper bits store count */
28#define COOKIEMASK (((__u32)1 << COOKIEBITS) - 1)
29
5918e2fb 30/* Table must be sorted. */
c6aefafb 31static __u16 const msstab[] = {
5918e2fb
FW
32 64,
33 512,
34 536,
35 1280 - 60,
36 1480 - 60,
37 1500 - 60,
38 4460 - 60,
39 9000 - 60,
c6aefafb 40};
c6aefafb
GG
41
42/*
43 * This (misnamed) value is the age of syncookie which is permitted.
44 * Its ideal value should be dependent on TCP_TIMEOUT_INIT and
45 * sysctl_tcp_retries1. It's a rather complicated formula (exponential
46 * backoff) to compute at runtime so it's currently hardcoded here.
47 */
48#define COUNTER_TRIES 4
49
50static inline struct sock *get_cookie_sock(struct sock *sk, struct sk_buff *skb,
51 struct request_sock *req,
52 struct dst_entry *dst)
53{
54 struct inet_connection_sock *icsk = inet_csk(sk);
55 struct sock *child;
56
57 child = icsk->icsk_af_ops->syn_recv_sock(sk, skb, req, dst);
58 if (child)
59 inet_csk_reqsk_queue_add(sk, req, child);
60 else
61 reqsk_free(req);
62
63 return child;
64}
65
245b2e70
TH
66static DEFINE_PER_CPU(__u32 [16 + 5 + SHA_WORKSPACE_WORDS],
67 ipv6_cookie_scratch);
c6aefafb 68
b71d1d42 69static u32 cookie_hash(const struct in6_addr *saddr, const struct in6_addr *daddr,
c6aefafb
GG
70 __be16 sport, __be16 dport, u32 count, int c)
71{
245b2e70 72 __u32 *tmp = __get_cpu_var(ipv6_cookie_scratch);
c6aefafb
GG
73
74 /*
75 * we have 320 bits of information to hash, copy in the remaining
76 * 192 bits required for sha_transform, from the syncookie_secret
77 * and overwrite the digest with the secret
78 */
79 memcpy(tmp + 10, syncookie_secret[c], 44);
80 memcpy(tmp, saddr, 16);
81 memcpy(tmp + 4, daddr, 16);
82 tmp[8] = ((__force u32)sport << 16) + (__force u32)dport;
83 tmp[9] = count;
84 sha_transform(tmp + 16, (__u8 *)tmp, tmp + 16 + 5);
85
86 return tmp[17];
87}
88
b71d1d42
ED
89static __u32 secure_tcp_syn_cookie(const struct in6_addr *saddr,
90 const struct in6_addr *daddr,
c6aefafb
GG
91 __be16 sport, __be16 dport, __u32 sseq,
92 __u32 count, __u32 data)
93{
94 return (cookie_hash(saddr, daddr, sport, dport, 0, 0) +
95 sseq + (count << COOKIEBITS) +
96 ((cookie_hash(saddr, daddr, sport, dport, count, 1) + data)
97 & COOKIEMASK));
98}
99
b71d1d42
ED
100static __u32 check_tcp_syn_cookie(__u32 cookie, const struct in6_addr *saddr,
101 const struct in6_addr *daddr, __be16 sport,
c6aefafb
GG
102 __be16 dport, __u32 sseq, __u32 count,
103 __u32 maxdiff)
104{
105 __u32 diff;
106
107 cookie -= cookie_hash(saddr, daddr, sport, dport, 0, 0) + sseq;
108
109 diff = (count - (cookie >> COOKIEBITS)) & ((__u32) -1 >> COOKIEBITS);
110 if (diff >= maxdiff)
111 return (__u32)-1;
112
113 return (cookie -
114 cookie_hash(saddr, daddr, sport, dport, count - diff, 1))
115 & COOKIEMASK;
116}
117
118__u32 cookie_v6_init_sequence(struct sock *sk, struct sk_buff *skb, __u16 *mssp)
119{
b71d1d42 120 const struct ipv6hdr *iph = ipv6_hdr(skb);
c6aefafb
GG
121 const struct tcphdr *th = tcp_hdr(skb);
122 int mssind;
123 const __u16 mss = *mssp;
124
a0f82f64 125 tcp_synq_overflow(sk);
c6aefafb 126
5918e2fb
FW
127 for (mssind = ARRAY_SIZE(msstab) - 1; mssind ; mssind--)
128 if (mss >= msstab[mssind])
129 break;
130
131 *mssp = msstab[mssind];
c6aefafb 132
de0744af 133 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
c6aefafb
GG
134
135 return secure_tcp_syn_cookie(&iph->saddr, &iph->daddr, th->source,
136 th->dest, ntohl(th->seq),
137 jiffies / (HZ * 60), mssind);
138}
139
140static inline int cookie_check(struct sk_buff *skb, __u32 cookie)
141{
b71d1d42 142 const struct ipv6hdr *iph = ipv6_hdr(skb);
c6aefafb
GG
143 const struct tcphdr *th = tcp_hdr(skb);
144 __u32 seq = ntohl(th->seq) - 1;
145 __u32 mssind = check_tcp_syn_cookie(cookie, &iph->saddr, &iph->daddr,
146 th->source, th->dest, seq,
147 jiffies / (HZ * 60), COUNTER_TRIES);
148
5918e2fb 149 return mssind < ARRAY_SIZE(msstab) ? msstab[mssind] : 0;
c6aefafb
GG
150}
151
152struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb)
153{
4957faad
WAS
154 struct tcp_options_received tcp_opt;
155 u8 *hash_location;
c6aefafb
GG
156 struct inet_request_sock *ireq;
157 struct inet6_request_sock *ireq6;
158 struct tcp_request_sock *treq;
159 struct ipv6_pinfo *np = inet6_sk(sk);
160 struct tcp_sock *tp = tcp_sk(sk);
161 const struct tcphdr *th = tcp_hdr(skb);
162 __u32 cookie = ntohl(th->ack_seq) - 1;
163 struct sock *ret = sk;
164 struct request_sock *req;
165 int mss;
166 struct dst_entry *dst;
167 __u8 rcv_wscale;
172d69e6 168 bool ecn_ok;
c6aefafb 169
af9b4738 170 if (!sysctl_tcp_syncookies || !th->ack || th->rst)
c6aefafb
GG
171 goto out;
172
a0f82f64 173 if (tcp_synq_no_recent_overflow(sk) ||
c6aefafb 174 (mss = cookie_check(skb, cookie)) == 0) {
de0744af 175 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SYNCOOKIESFAILED);
c6aefafb
GG
176 goto out;
177 }
178
de0744af 179 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SYNCOOKIESRECV);
c6aefafb 180
bb5b7c11
DM
181 /* check for timestamp cookie support */
182 memset(&tcp_opt, 0, sizeof(tcp_opt));
183 tcp_parse_options(skb, &tcp_opt, &hash_location, 0);
184
172d69e6 185 if (!cookie_check_timestamp(&tcp_opt, &ecn_ok))
8c763681 186 goto out;
bb5b7c11 187
c6aefafb
GG
188 ret = NULL;
189 req = inet6_reqsk_alloc(&tcp6_request_sock_ops);
190 if (!req)
191 goto out;
192
193 ireq = inet_rsk(req);
194 ireq6 = inet6_rsk(req);
195 treq = tcp_rsk(req);
c6aefafb 196
1730554f
FW
197 if (security_inet_conn_request(sk, skb, req))
198 goto out_free;
c6aefafb
GG
199
200 req->mss = mss;
201 ireq->rmt_port = th->source;
fd507037 202 ireq->loc_port = th->dest;
c6aefafb
GG
203 ipv6_addr_copy(&ireq6->rmt_addr, &ipv6_hdr(skb)->saddr);
204 ipv6_addr_copy(&ireq6->loc_addr, &ipv6_hdr(skb)->daddr);
205 if (ipv6_opt_accepted(sk, skb) ||
206 np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo ||
207 np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim) {
208 atomic_inc(&skb->users);
209 ireq6->pktopts = skb;
210 }
211
212 ireq6->iif = sk->sk_bound_dev_if;
213 /* So that link locals have meaning */
214 if (!sk->sk_bound_dev_if &&
215 ipv6_addr_type(&ireq6->rmt_addr) & IPV6_ADDR_LINKLOCAL)
216 ireq6->iif = inet6_iif(skb);
217
218 req->expires = 0UL;
219 req->retrans = 0;
172d69e6 220 ireq->ecn_ok = ecn_ok;
bb5b7c11 221 ireq->snd_wscale = tcp_opt.snd_wscale;
bb5b7c11
DM
222 ireq->sack_ok = tcp_opt.sack_ok;
223 ireq->wscale_ok = tcp_opt.wscale_ok;
224 ireq->tstamp_ok = tcp_opt.saw_tstamp;
225 req->ts_recent = tcp_opt.saw_tstamp ? tcp_opt.rcv_tsval : 0;
9ad7c049 226 treq->snt_synack = tcp_opt.saw_tstamp ? tcp_opt.rcv_tsecr : 0;
c6aefafb
GG
227 treq->rcv_isn = ntohl(th->seq) - 1;
228 treq->snt_isn = cookie;
229
230 /*
231 * We need to lookup the dst_entry to get the correct window size.
232 * This is taken from tcp_v6_syn_recv_sock. Somebody please enlighten
233 * me if there is a preferred way.
234 */
235 {
20c59de2 236 struct in6_addr *final_p, final;
4c9483b2
DM
237 struct flowi6 fl6;
238 memset(&fl6, 0, sizeof(fl6));
239 fl6.flowi6_proto = IPPROTO_TCP;
240 ipv6_addr_copy(&fl6.daddr, &ireq6->rmt_addr);
241 final_p = fl6_update_dst(&fl6, np->opt, &final);
242 ipv6_addr_copy(&fl6.saddr, &ireq6->loc_addr);
243 fl6.flowi6_oif = sk->sk_bound_dev_if;
244 fl6.flowi6_mark = sk->sk_mark;
1958b856
DM
245 fl6.fl6_dport = inet_rsk(req)->rmt_port;
246 fl6.fl6_sport = inet_sk(sk)->inet_sport;
4c9483b2
DM
247 security_req_classify_flow(req, flowi6_to_flowi(&fl6));
248
249 dst = ip6_dst_lookup_flow(sk, &fl6, final_p, false);
68d0c6d3 250 if (IS_ERR(dst))
1730554f 251 goto out_free;
c6aefafb
GG
252 }
253
4dfc2817 254 req->window_clamp = tp->window_clamp ? :dst_metric(dst, RTAX_WINDOW);
c6aefafb
GG
255 tcp_select_initial_window(tcp_full_space(sk), req->mss,
256 &req->rcv_wnd, &req->window_clamp,
31d12926 257 ireq->wscale_ok, &rcv_wscale,
258 dst_metric(dst, RTAX_INITRWND));
c6aefafb
GG
259
260 ireq->rcv_wscale = rcv_wscale;
261
262 ret = get_cookie_sock(sk, skb, req, dst);
1730554f
FW
263out:
264 return ret;
265out_free:
266 reqsk_free(req);
267 return NULL;
c6aefafb
GG
268}
269