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 $
15 #include <linux/tcp.h>
16 #include <linux/slab.h>
17 #include <linux/random.h>
18 #include <linux/cryptohash.h>
19 #include <linux/kernel.h>
22 /* Timestamps: lowest 9 bits store TCP options */
24 #define TSMASK (((__u32)1 << TSBITS) - 1)
26 extern int sysctl_tcp_syncookies;
28 __u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS];
29 EXPORT_SYMBOL(syncookie_secret);
31 static __init int init_syncookies(void)
33 get_random_bytes(syncookie_secret, sizeof(syncookie_secret));
36 __initcall(init_syncookies);
38 #define COOKIEBITS 24 /* Upper bits store count */
39 #define COOKIEMASK (((__u32)1 << COOKIEBITS) - 1)
41 static DEFINE_PER_CPU(__u32, cookie_scratch)[16 + 5 + SHA_WORKSPACE_WORDS];
43 static u32 cookie_hash(__be32 saddr, __be32 daddr, __be16 sport, __be16 dport,
46 __u32 *tmp = __get_cpu_var(cookie_scratch);
48 memcpy(tmp + 4, syncookie_secret[c], sizeof(syncookie_secret[c]));
49 tmp[0] = (__force u32)saddr;
50 tmp[1] = (__force u32)daddr;
51 tmp[2] = ((__force u32)sport << 16) + (__force u32)dport;
53 sha_transform(tmp + 16, (__u8 *)tmp, tmp + 16 + 5);
60 * when syncookies are in effect and tcp timestamps are enabled we encode
61 * tcp options in the lowest 9 bits of the timestamp value that will be
62 * sent in the syn-ack.
63 * Since subsequent timestamps use the normal tcp_time_stamp value, we
64 * must make sure that the resulting initial timestamp is <= tcp_time_stamp.
66 __u32 cookie_init_timestamp(struct request_sock *req)
68 struct inet_request_sock *ireq;
69 u32 ts, ts_now = tcp_time_stamp;
73 if (ireq->wscale_ok) {
74 options = ireq->snd_wscale;
75 options |= ireq->rcv_wscale << 4;
77 options |= ireq->sack_ok << 8;
79 ts = ts_now & ~TSMASK;
91 static __u32 secure_tcp_syn_cookie(__be32 saddr, __be32 daddr, __be16 sport,
92 __be16 dport, __u32 sseq, __u32 count,
96 * Compute the secure sequence number.
97 * The output should be:
98 * HASH(sec1,saddr,sport,daddr,dport,sec1) + sseq + (count * 2^24)
99 * + (HASH(sec2,saddr,sport,daddr,dport,count,sec2) % 2^24).
100 * Where sseq is their sequence number and count increases every
102 * As an extra hack, we add a small "data" value that encodes the
103 * MSS into the second hash value.
106 return (cookie_hash(saddr, daddr, sport, dport, 0, 0) +
107 sseq + (count << COOKIEBITS) +
108 ((cookie_hash(saddr, daddr, sport, dport, count, 1) + data)
113 * This retrieves the small "data" value from the syncookie.
114 * If the syncookie is bad, the data returned will be out of
115 * range. This must be checked by the caller.
117 * The count value used to generate the cookie must be within
118 * "maxdiff" if the current (passed-in) "count". The return value
119 * is (__u32)-1 if this test fails.
121 static __u32 check_tcp_syn_cookie(__u32 cookie, __be32 saddr, __be32 daddr,
122 __be16 sport, __be16 dport, __u32 sseq,
123 __u32 count, __u32 maxdiff)
127 /* Strip away the layers from the cookie */
128 cookie -= cookie_hash(saddr, daddr, sport, dport, 0, 0) + sseq;
130 /* Cookie is now reduced to (count * 2^24) ^ (hash % 2^24) */
131 diff = (count - (cookie >> COOKIEBITS)) & ((__u32) - 1 >> COOKIEBITS);
136 cookie_hash(saddr, daddr, sport, dport, count - diff, 1))
137 & COOKIEMASK; /* Leaving the data behind */
141 * This table has to be sorted and terminated with (__u16)-1.
142 * XXX generate a better table.
143 * Unresolved Issues: HIPPI with a 64k MSS is not well supported.
145 static __u16 const msstab[] = {
156 /* The number doesn't include the -1 terminator */
157 #define NUM_MSS (ARRAY_SIZE(msstab) - 1)
160 * Generate a syncookie. mssp points to the mss, which is returned
161 * rounded down to the value encoded in the cookie.
163 __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb, __u16 *mssp)
165 struct tcp_sock *tp = tcp_sk(sk);
166 const struct iphdr *iph = ip_hdr(skb);
167 const struct tcphdr *th = tcp_hdr(skb);
169 const __u16 mss = *mssp;
171 tp->last_synq_overflow = jiffies;
173 /* XXX sort msstab[] by probability? Binary search? */
174 for (mssind = 0; mss > msstab[mssind + 1]; mssind++)
176 *mssp = msstab[mssind] + 1;
178 NET_INC_STATS_BH(LINUX_MIB_SYNCOOKIESSENT);
180 return secure_tcp_syn_cookie(iph->saddr, iph->daddr,
181 th->source, th->dest, ntohl(th->seq),
182 jiffies / (HZ * 60), mssind);
186 * This (misnamed) value is the age of syncookie which is permitted.
187 * Its ideal value should be dependent on TCP_TIMEOUT_INIT and
188 * sysctl_tcp_retries1. It's a rather complicated formula (exponential
189 * backoff) to compute at runtime so it's currently hardcoded here.
191 #define COUNTER_TRIES 4
193 * Check if a ack sequence number is a valid syncookie.
194 * Return the decoded mss if it is, or 0 if not.
196 static inline int cookie_check(struct sk_buff *skb, __u32 cookie)
198 const struct iphdr *iph = ip_hdr(skb);
199 const struct tcphdr *th = tcp_hdr(skb);
200 __u32 seq = ntohl(th->seq) - 1;
201 __u32 mssind = check_tcp_syn_cookie(cookie, iph->saddr, iph->daddr,
202 th->source, th->dest, seq,
206 return mssind < NUM_MSS ? msstab[mssind] + 1 : 0;
209 static inline struct sock *get_cookie_sock(struct sock *sk, struct sk_buff *skb,
210 struct request_sock *req,
211 struct dst_entry *dst)
213 struct inet_connection_sock *icsk = inet_csk(sk);
216 child = icsk->icsk_af_ops->syn_recv_sock(sk, skb, req, dst);
218 inet_csk_reqsk_queue_add(sk, req, child);
227 * when syncookies are in effect and tcp timestamps are enabled we stored
228 * additional tcp options in the timestamp.
229 * This extracts these options from the timestamp echo.
231 * The lowest 4 bits are for snd_wscale
232 * The next 4 lsb are for rcv_wscale
233 * The next lsb is for sack_ok
235 void cookie_check_timestamp(struct tcp_options_received *tcp_opt)
237 /* echoed timestamp, 9 lowest bits contain options */
238 u32 options = tcp_opt->rcv_tsecr & TSMASK;
240 tcp_opt->snd_wscale = options & 0xf;
242 tcp_opt->rcv_wscale = options & 0xf;
244 tcp_opt->sack_ok = (options >> 4) & 0x1;
246 if (tcp_opt->sack_ok)
247 tcp_sack_reset(tcp_opt);
249 if (tcp_opt->snd_wscale || tcp_opt->rcv_wscale)
250 tcp_opt->wscale_ok = 1;
252 EXPORT_SYMBOL(cookie_check_timestamp);
254 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
255 struct ip_options *opt)
257 struct inet_request_sock *ireq;
258 struct tcp_request_sock *treq;
259 struct tcp_sock *tp = tcp_sk(sk);
260 const struct tcphdr *th = tcp_hdr(skb);
261 __u32 cookie = ntohl(th->ack_seq) - 1;
262 struct sock *ret = sk;
263 struct request_sock *req;
267 struct tcp_options_received tcp_opt;
269 if (!sysctl_tcp_syncookies || !th->ack)
272 if (time_after(jiffies, tp->last_synq_overflow + TCP_TIMEOUT_INIT) ||
273 (mss = cookie_check(skb, cookie)) == 0) {
274 NET_INC_STATS_BH(LINUX_MIB_SYNCOOKIESFAILED);
278 NET_INC_STATS_BH(LINUX_MIB_SYNCOOKIESRECV);
280 /* check for timestamp cookie support */
281 memset(&tcp_opt, 0, sizeof(tcp_opt));
282 tcp_parse_options(skb, &tcp_opt, 0);
284 if (tcp_opt.saw_tstamp)
285 cookie_check_timestamp(&tcp_opt);
288 req = inet_reqsk_alloc(&tcp_request_sock_ops); /* for safety */
292 if (security_inet_conn_request(sk, skb, req)) {
296 ireq = inet_rsk(req);
298 treq->rcv_isn = ntohl(th->seq) - 1;
299 treq->snt_isn = cookie;
301 ireq->rmt_port = th->source;
302 ireq->loc_addr = ip_hdr(skb)->daddr;
303 ireq->rmt_addr = ip_hdr(skb)->saddr;
304 ireq->snd_wscale = tcp_opt.snd_wscale;
305 ireq->rcv_wscale = tcp_opt.rcv_wscale;
306 ireq->sack_ok = tcp_opt.sack_ok;
307 ireq->wscale_ok = tcp_opt.wscale_ok;
308 ireq->tstamp_ok = tcp_opt.saw_tstamp;
309 req->ts_recent = tcp_opt.saw_tstamp ? tcp_opt.rcv_tsval : 0;
311 /* We throwed the options of the initial SYN away, so we hope
312 * the ACK carries the same options again (see RFC1122 4.2.3.8)
314 if (opt && opt->optlen) {
315 int opt_size = sizeof(struct ip_options) + opt->optlen;
317 ireq->opt = kmalloc(opt_size, GFP_ATOMIC);
318 if (ireq->opt != NULL && ip_options_echo(ireq->opt, skb)) {
328 * We need to lookup the route here to get at the correct
329 * window size. We should better make sure that the window size
330 * hasn't changed since we received the original syn, but I see
331 * no easy way to do this.
334 struct flowi fl = { .nl_u = { .ip4_u =
335 { .daddr = ((opt && opt->srr) ?
338 .saddr = ireq->loc_addr,
339 .tos = RT_CONN_FLAGS(sk) } },
340 .proto = IPPROTO_TCP,
343 .dport = th->source } } };
344 security_req_classify_flow(req, &fl);
345 if (ip_route_output_key(&init_net, &rt, &fl)) {
351 /* Try to redo what tcp_v4_send_synack did. */
352 req->window_clamp = tp->window_clamp ? :dst_metric(&rt->u.dst, RTAX_WINDOW);
354 tcp_select_initial_window(tcp_full_space(sk), req->mss,
355 &req->rcv_wnd, &req->window_clamp,
356 ireq->wscale_ok, &rcv_wscale);
358 ireq->rcv_wscale = rcv_wscale;
360 ret = get_cookie_sock(sk, skb, req, &rt->u.dst);