2 * Syncookies implementation for the Linux kernel
4 * Copyright (C) 1997 Andi Kleen
5 * Based on ideas by D.J.Bernstein and Eric Schenk.
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public License
9 * as published by the Free Software Foundation; either version
10 * 2 of the License, or (at your option) any later version.
12 * $Id: syncookies.c,v 1.18 2002/02/01 22:01:04 davem Exp $
14 * Missing: IPv6 support.
17 #include <linux/tcp.h>
18 #include <linux/slab.h>
19 #include <linux/random.h>
20 #include <linux/cryptohash.h>
21 #include <linux/kernel.h>
24 extern int sysctl_tcp_syncookies;
26 static __u32 syncookie_secret[2][16-3+SHA_DIGEST_WORDS];
28 static __init int init_syncookies(void)
30 get_random_bytes(syncookie_secret, sizeof(syncookie_secret));
33 module_init(init_syncookies);
35 #define COOKIEBITS 24 /* Upper bits store count */
36 #define COOKIEMASK (((__u32)1 << COOKIEBITS) - 1)
38 static DEFINE_PER_CPU(__u32, cookie_scratch)[16 + 5 + SHA_WORKSPACE_WORDS];
40 static u32 cookie_hash(__be32 saddr, __be32 daddr, __be16 sport, __be16 dport,
43 __u32 *tmp = __get_cpu_var(cookie_scratch);
45 memcpy(tmp + 3, syncookie_secret[c], sizeof(syncookie_secret[c]));
46 tmp[0] = (__force u32)saddr;
47 tmp[1] = (__force u32)daddr;
48 tmp[2] = ((__force u32)sport << 16) + (__force u32)dport;
50 sha_transform(tmp + 16, (__u8 *)tmp, tmp + 16 + 5);
55 static __u32 secure_tcp_syn_cookie(__be32 saddr, __be32 daddr, __be16 sport,
56 __be16 dport, __u32 sseq, __u32 count,
60 * Compute the secure sequence number.
61 * The output should be:
62 * HASH(sec1,saddr,sport,daddr,dport,sec1) + sseq + (count * 2^24)
63 * + (HASH(sec2,saddr,sport,daddr,dport,count,sec2) % 2^24).
64 * Where sseq is their sequence number and count increases every
66 * As an extra hack, we add a small "data" value that encodes the
67 * MSS into the second hash value.
70 return (cookie_hash(saddr, daddr, sport, dport, 0, 0) +
71 sseq + (count << COOKIEBITS) +
72 ((cookie_hash(saddr, daddr, sport, dport, count, 1) + data)
77 * This retrieves the small "data" value from the syncookie.
78 * If the syncookie is bad, the data returned will be out of
79 * range. This must be checked by the caller.
81 * The count value used to generate the cookie must be within
82 * "maxdiff" if the current (passed-in) "count". The return value
83 * is (__u32)-1 if this test fails.
85 static __u32 check_tcp_syn_cookie(__u32 cookie, __be32 saddr, __be32 daddr,
86 __be16 sport, __be16 dport, __u32 sseq,
87 __u32 count, __u32 maxdiff)
91 /* Strip away the layers from the cookie */
92 cookie -= cookie_hash(saddr, daddr, sport, dport, 0, 0) + sseq;
94 /* Cookie is now reduced to (count * 2^24) ^ (hash % 2^24) */
95 diff = (count - (cookie >> COOKIEBITS)) & ((__u32) - 1 >> COOKIEBITS);
100 cookie_hash(saddr, daddr, sport, dport, count - diff, 1))
101 & COOKIEMASK; /* Leaving the data behind */
105 * This table has to be sorted and terminated with (__u16)-1.
106 * XXX generate a better table.
107 * Unresolved Issues: HIPPI with a 64k MSS is not well supported.
109 static __u16 const msstab[] = {
120 /* The number doesn't include the -1 terminator */
121 #define NUM_MSS (ARRAY_SIZE(msstab) - 1)
124 * Generate a syncookie. mssp points to the mss, which is returned
125 * rounded down to the value encoded in the cookie.
127 __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb, __u16 *mssp)
129 struct tcp_sock *tp = tcp_sk(sk);
130 const struct iphdr *iph = ip_hdr(skb);
131 const struct tcphdr *th = tcp_hdr(skb);
133 const __u16 mss = *mssp;
135 tp->last_synq_overflow = jiffies;
137 /* XXX sort msstab[] by probability? Binary search? */
138 for (mssind = 0; mss > msstab[mssind + 1]; mssind++)
140 *mssp = msstab[mssind] + 1;
142 NET_INC_STATS_BH(LINUX_MIB_SYNCOOKIESSENT);
144 return secure_tcp_syn_cookie(iph->saddr, iph->daddr,
145 th->source, th->dest, ntohl(th->seq),
146 jiffies / (HZ * 60), mssind);
150 * This (misnamed) value is the age of syncookie which is permitted.
151 * Its ideal value should be dependent on TCP_TIMEOUT_INIT and
152 * sysctl_tcp_retries1. It's a rather complicated formula (exponential
153 * backoff) to compute at runtime so it's currently hardcoded here.
155 #define COUNTER_TRIES 4
157 * Check if a ack sequence number is a valid syncookie.
158 * Return the decoded mss if it is, or 0 if not.
160 static inline int cookie_check(struct sk_buff *skb, __u32 cookie)
162 const struct iphdr *iph = ip_hdr(skb);
163 const struct tcphdr *th = tcp_hdr(skb);
164 __u32 seq = ntohl(th->seq) - 1;
165 __u32 mssind = check_tcp_syn_cookie(cookie, iph->saddr, iph->daddr,
166 th->source, th->dest, seq,
170 return mssind < NUM_MSS ? msstab[mssind] + 1 : 0;
173 static inline struct sock *get_cookie_sock(struct sock *sk, struct sk_buff *skb,
174 struct request_sock *req,
175 struct dst_entry *dst)
177 struct inet_connection_sock *icsk = inet_csk(sk);
180 child = icsk->icsk_af_ops->syn_recv_sock(sk, skb, req, dst);
182 inet_csk_reqsk_queue_add(sk, req, child);
189 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
190 struct ip_options *opt)
192 struct inet_request_sock *ireq;
193 struct tcp_request_sock *treq;
194 struct tcp_sock *tp = tcp_sk(sk);
195 const struct tcphdr *th = tcp_hdr(skb);
196 __u32 cookie = ntohl(th->ack_seq) - 1;
197 struct sock *ret = sk;
198 struct request_sock *req;
203 if (!sysctl_tcp_syncookies || !th->ack)
206 if (time_after(jiffies, tp->last_synq_overflow + TCP_TIMEOUT_INIT) ||
207 (mss = cookie_check(skb, cookie)) == 0) {
208 NET_INC_STATS_BH(LINUX_MIB_SYNCOOKIESFAILED);
212 NET_INC_STATS_BH(LINUX_MIB_SYNCOOKIESRECV);
215 req = reqsk_alloc(&tcp_request_sock_ops); /* for safety */
219 if (security_inet_conn_request(sk, skb, req)) {
223 ireq = inet_rsk(req);
225 treq->rcv_isn = ntohl(th->seq) - 1;
226 treq->snt_isn = cookie;
228 ireq->rmt_port = th->source;
229 ireq->loc_addr = ip_hdr(skb)->daddr;
230 ireq->rmt_addr = ip_hdr(skb)->saddr;
233 /* We throwed the options of the initial SYN away, so we hope
234 * the ACK carries the same options again (see RFC1122 4.2.3.8)
236 if (opt && opt->optlen) {
237 int opt_size = sizeof(struct ip_options) + opt->optlen;
239 ireq->opt = kmalloc(opt_size, GFP_ATOMIC);
240 if (ireq->opt != NULL && ip_options_echo(ireq->opt, skb)) {
246 ireq->snd_wscale = ireq->rcv_wscale = ireq->tstamp_ok = 0;
247 ireq->wscale_ok = ireq->sack_ok = 0;
252 * We need to lookup the route here to get at the correct
253 * window size. We should better make sure that the window size
254 * hasn't changed since we received the original syn, but I see
255 * no easy way to do this.
258 struct flowi fl = { .nl_u = { .ip4_u =
259 { .daddr = ((opt && opt->srr) ?
262 .saddr = ireq->loc_addr,
263 .tos = RT_CONN_FLAGS(sk) } },
264 .proto = IPPROTO_TCP,
267 .dport = th->source } } };
268 security_req_classify_flow(req, &fl);
269 if (ip_route_output_key(&init_net, &rt, &fl)) {
275 /* Try to redo what tcp_v4_send_synack did. */
276 req->window_clamp = dst_metric(&rt->u.dst, RTAX_WINDOW);
277 tcp_select_initial_window(tcp_full_space(sk), req->mss,
278 &req->rcv_wnd, &req->window_clamp,
280 /* BTW win scale with syncookies is 0 by definition */
281 ireq->rcv_wscale = rcv_wscale;
283 ret = get_cookie_sock(sk, skb, req, &rt->u.dst);