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 u32 cookie_hash(u32 saddr, u32 daddr, u32 sport, u32 dport,
41 __u32 tmp[16 + 5 + SHA_WORKSPACE_WORDS];
43 memcpy(tmp + 3, syncookie_secret[c], sizeof(syncookie_secret[c]));
46 tmp[2] = (sport << 16) + dport;
48 sha_transform(tmp + 16, (__u8 *)tmp, tmp + 16 + 5);
53 static __u32 secure_tcp_syn_cookie(__u32 saddr, __u32 daddr, __u16 sport,
54 __u16 dport, __u32 sseq, __u32 count,
58 * Compute the secure sequence number.
59 * The output should be:
60 * HASH(sec1,saddr,sport,daddr,dport,sec1) + sseq + (count * 2^24)
61 * + (HASH(sec2,saddr,sport,daddr,dport,count,sec2) % 2^24).
62 * Where sseq is their sequence number and count increases every
64 * As an extra hack, we add a small "data" value that encodes the
65 * MSS into the second hash value.
68 return (cookie_hash(saddr, daddr, sport, dport, 0, 0) +
69 sseq + (count << COOKIEBITS) +
70 ((cookie_hash(saddr, daddr, sport, dport, count, 1) + data)
75 * This retrieves the small "data" value from the syncookie.
76 * If the syncookie is bad, the data returned will be out of
77 * range. This must be checked by the caller.
79 * The count value used to generate the cookie must be within
80 * "maxdiff" if the current (passed-in) "count". The return value
81 * is (__u32)-1 if this test fails.
83 static __u32 check_tcp_syn_cookie(__u32 cookie, __u32 saddr, __u32 daddr,
84 __u16 sport, __u16 dport, __u32 sseq,
85 __u32 count, __u32 maxdiff)
89 /* Strip away the layers from the cookie */
90 cookie -= cookie_hash(saddr, daddr, sport, dport, 0, 0) + sseq;
92 /* Cookie is now reduced to (count * 2^24) ^ (hash % 2^24) */
93 diff = (count - (cookie >> COOKIEBITS)) & ((__u32) - 1 >> COOKIEBITS);
98 cookie_hash(saddr, daddr, sport, dport, count - diff, 1))
99 & COOKIEMASK; /* Leaving the data behind */
103 * This table has to be sorted and terminated with (__u16)-1.
104 * XXX generate a better table.
105 * Unresolved Issues: HIPPI with a 64k MSS is not well supported.
107 static __u16 const msstab[] = {
118 /* The number doesn't include the -1 terminator */
119 #define NUM_MSS (ARRAY_SIZE(msstab) - 1)
122 * Generate a syncookie. mssp points to the mss, which is returned
123 * rounded down to the value encoded in the cookie.
125 __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb, __u16 *mssp)
127 struct tcp_sock *tp = tcp_sk(sk);
129 const __u16 mss = *mssp;
132 tp->last_synq_overflow = jiffies;
134 /* XXX sort msstab[] by probability? Binary search? */
135 for (mssind = 0; mss > msstab[mssind + 1]; mssind++)
137 *mssp = msstab[mssind] + 1;
139 NET_INC_STATS_BH(LINUX_MIB_SYNCOOKIESSENT);
141 return secure_tcp_syn_cookie(skb->nh.iph->saddr, skb->nh.iph->daddr,
142 skb->h.th->source, skb->h.th->dest,
143 ntohl(skb->h.th->seq),
144 jiffies / (HZ * 60), mssind);
148 * This (misnamed) value is the age of syncookie which is permitted.
149 * Its ideal value should be dependent on TCP_TIMEOUT_INIT and
150 * sysctl_tcp_retries1. It's a rather complicated formula (exponential
151 * backoff) to compute at runtime so it's currently hardcoded here.
153 #define COUNTER_TRIES 4
155 * Check if a ack sequence number is a valid syncookie.
156 * Return the decoded mss if it is, or 0 if not.
158 static inline int cookie_check(struct sk_buff *skb, __u32 cookie)
163 seq = ntohl(skb->h.th->seq)-1;
164 mssind = check_tcp_syn_cookie(cookie,
165 skb->nh.iph->saddr, skb->nh.iph->daddr,
166 skb->h.th->source, skb->h.th->dest,
167 seq, jiffies / (HZ * 60), COUNTER_TRIES);
169 return mssind < NUM_MSS ? msstab[mssind] + 1 : 0;
172 static inline struct sock *get_cookie_sock(struct sock *sk, struct sk_buff *skb,
173 struct request_sock *req,
174 struct dst_entry *dst)
176 struct inet_connection_sock *icsk = inet_csk(sk);
179 child = icsk->icsk_af_ops->syn_recv_sock(sk, skb, req, dst);
181 inet_csk_reqsk_queue_add(sk, req, child);
188 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
189 struct ip_options *opt)
191 struct inet_request_sock *ireq;
192 struct tcp_request_sock *treq;
193 struct tcp_sock *tp = tcp_sk(sk);
194 __u32 cookie = ntohl(skb->h.th->ack_seq) - 1;
195 struct sock *ret = sk;
196 struct request_sock *req;
201 if (!sysctl_tcp_syncookies || !skb->h.th->ack)
204 if (time_after(jiffies, tp->last_synq_overflow + TCP_TIMEOUT_INIT) ||
205 (mss = cookie_check(skb, cookie)) == 0) {
206 NET_INC_STATS_BH(LINUX_MIB_SYNCOOKIESFAILED);
210 NET_INC_STATS_BH(LINUX_MIB_SYNCOOKIESRECV);
213 req = reqsk_alloc(&tcp_request_sock_ops); /* for safety */
217 if (security_inet_conn_request(sk, skb, req)) {
221 ireq = inet_rsk(req);
223 treq->rcv_isn = htonl(skb->h.th->seq) - 1;
224 treq->snt_isn = cookie;
226 ireq->rmt_port = skb->h.th->source;
227 ireq->loc_addr = skb->nh.iph->daddr;
228 ireq->rmt_addr = skb->nh.iph->saddr;
231 /* We throwed the options of the initial SYN away, so we hope
232 * the ACK carries the same options again (see RFC1122 4.2.3.8)
234 if (opt && opt->optlen) {
235 int opt_size = sizeof(struct ip_options) + opt->optlen;
237 ireq->opt = kmalloc(opt_size, GFP_ATOMIC);
238 if (ireq->opt != NULL && ip_options_echo(ireq->opt, skb)) {
244 ireq->snd_wscale = ireq->rcv_wscale = ireq->tstamp_ok = 0;
245 ireq->wscale_ok = ireq->sack_ok = 0;
250 * We need to lookup the route here to get at the correct
251 * window size. We should better make sure that the window size
252 * hasn't changed since we received the original syn, but I see
253 * no easy way to do this.
256 struct flowi fl = { .nl_u = { .ip4_u =
257 { .daddr = ((opt && opt->srr) ?
260 .saddr = ireq->loc_addr,
261 .tos = RT_CONN_FLAGS(sk) } },
262 .proto = IPPROTO_TCP,
264 { .sport = skb->h.th->dest,
265 .dport = skb->h.th->source } } };
266 security_req_classify_flow(req, &fl);
267 if (ip_route_output_key(&rt, &fl)) {
273 /* Try to redo what tcp_v4_send_synack did. */
274 req->window_clamp = dst_metric(&rt->u.dst, RTAX_WINDOW);
275 tcp_select_initial_window(tcp_full_space(sk), req->mss,
276 &req->rcv_wnd, &req->window_clamp,
278 /* BTW win scale with syncookies is 0 by definition */
279 ireq->rcv_wscale = rcv_wscale;
281 ret = get_cookie_sock(sk, skb, req, &rt->u.dst);