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.
13 #include <linux/tcp.h>
14 #include <linux/slab.h>
15 #include <linux/random.h>
16 #include <linux/cryptohash.h>
17 #include <linux/kernel.h>
18 #include <linux/export.h>
20 #include <net/route.h>
22 /* Timestamps: lowest bits store TCP options */
24 #define TSMASK (((__u32)1 << TSBITS) - 1)
26 extern int sysctl_tcp_syncookies;
28 static u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS] __read_mostly;
30 #define COOKIEBITS 24 /* Upper bits store count */
31 #define COOKIEMASK (((__u32)1 << COOKIEBITS) - 1)
33 static DEFINE_PER_CPU(__u32 [16 + 5 + SHA_WORKSPACE_WORDS],
36 static u32 cookie_hash(__be32 saddr, __be32 daddr, __be16 sport, __be16 dport,
41 net_get_random_once(syncookie_secret, sizeof(syncookie_secret));
43 tmp = this_cpu_ptr(ipv4_cookie_scratch);
44 memcpy(tmp + 4, syncookie_secret[c], sizeof(syncookie_secret[c]));
45 tmp[0] = (__force u32)saddr;
46 tmp[1] = (__force u32)daddr;
47 tmp[2] = ((__force u32)sport << 16) + (__force u32)dport;
49 sha_transform(tmp + 16, (__u8 *)tmp, tmp + 16 + 5);
56 * when syncookies are in effect and tcp timestamps are enabled we encode
57 * tcp options in the lower bits of the timestamp value that will be
58 * sent in the syn-ack.
59 * Since subsequent timestamps use the normal tcp_time_stamp value, we
60 * must make sure that the resulting initial timestamp is <= tcp_time_stamp.
62 __u32 cookie_init_timestamp(struct request_sock *req)
64 struct inet_request_sock *ireq;
65 u32 ts, ts_now = tcp_time_stamp;
70 options = ireq->wscale_ok ? ireq->snd_wscale : 0xf;
71 options |= ireq->sack_ok << 4;
72 options |= ireq->ecn_ok << 5;
74 ts = ts_now & ~TSMASK;
86 static __u32 secure_tcp_syn_cookie(__be32 saddr, __be32 daddr, __be16 sport,
87 __be16 dport, __u32 sseq, __u32 data)
90 * Compute the secure sequence number.
91 * The output should be:
92 * HASH(sec1,saddr,sport,daddr,dport,sec1) + sseq + (count * 2^24)
93 * + (HASH(sec2,saddr,sport,daddr,dport,count,sec2) % 2^24).
94 * Where sseq is their sequence number and count increases every
96 * As an extra hack, we add a small "data" value that encodes the
97 * MSS into the second hash value.
99 u32 count = tcp_cookie_time();
100 return (cookie_hash(saddr, daddr, sport, dport, 0, 0) +
101 sseq + (count << COOKIEBITS) +
102 ((cookie_hash(saddr, daddr, sport, dport, count, 1) + data)
107 * This retrieves the small "data" value from the syncookie.
108 * If the syncookie is bad, the data returned will be out of
109 * range. This must be checked by the caller.
111 * The count value used to generate the cookie must be less than
112 * MAX_SYNCOOKIE_AGE minutes in the past.
113 * The return value (__u32)-1 if this test fails.
115 static __u32 check_tcp_syn_cookie(__u32 cookie, __be32 saddr, __be32 daddr,
116 __be16 sport, __be16 dport, __u32 sseq)
118 u32 diff, count = tcp_cookie_time();
120 /* Strip away the layers from the cookie */
121 cookie -= cookie_hash(saddr, daddr, sport, dport, 0, 0) + sseq;
123 /* Cookie is now reduced to (count * 2^24) ^ (hash % 2^24) */
124 diff = (count - (cookie >> COOKIEBITS)) & ((__u32) -1 >> COOKIEBITS);
125 if (diff >= MAX_SYNCOOKIE_AGE)
129 cookie_hash(saddr, daddr, sport, dport, count - diff, 1))
130 & COOKIEMASK; /* Leaving the data behind */
134 * MSS Values are chosen based on the 2011 paper
135 * 'An Analysis of TCP Maximum Segement Sizes' by S. Alcock and R. Nelson.
137 * .. lower than 536 are rare (< 0.2%)
138 * .. between 537 and 1299 account for less than < 1.5% of observed values
139 * .. in the 1300-1349 range account for about 15 to 20% of observed mss values
140 * .. exceeding 1460 are very rare (< 0.04%)
142 * 1460 is the single most frequently announced mss value (30 to 46% depending
143 * on monitor location). Table must be sorted.
145 static __u16 const msstab[] = {
148 1440, /* 1440, 1452: PPPoE */
153 * Generate a syncookie. mssp points to the mss, which is returned
154 * rounded down to the value encoded in the cookie.
156 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
160 const __u16 mss = *mssp;
162 for (mssind = ARRAY_SIZE(msstab) - 1; mssind ; mssind--)
163 if (mss >= msstab[mssind])
165 *mssp = msstab[mssind];
167 return secure_tcp_syn_cookie(iph->saddr, iph->daddr,
168 th->source, th->dest, ntohl(th->seq),
171 EXPORT_SYMBOL_GPL(__cookie_v4_init_sequence);
173 __u32 cookie_v4_init_sequence(struct sock *sk, const struct sk_buff *skb,
176 const struct iphdr *iph = ip_hdr(skb);
177 const struct tcphdr *th = tcp_hdr(skb);
179 tcp_synq_overflow(sk);
180 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
182 return __cookie_v4_init_sequence(iph, th, mssp);
186 * Check if a ack sequence number is a valid syncookie.
187 * Return the decoded mss if it is, or 0 if not.
189 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
192 __u32 seq = ntohl(th->seq) - 1;
193 __u32 mssind = check_tcp_syn_cookie(cookie, iph->saddr, iph->daddr,
194 th->source, th->dest, seq);
196 return mssind < ARRAY_SIZE(msstab) ? msstab[mssind] : 0;
198 EXPORT_SYMBOL_GPL(__cookie_v4_check);
200 static inline struct sock *get_cookie_sock(struct sock *sk, struct sk_buff *skb,
201 struct request_sock *req,
202 struct dst_entry *dst)
204 struct inet_connection_sock *icsk = inet_csk(sk);
207 child = icsk->icsk_af_ops->syn_recv_sock(sk, skb, req, dst);
209 inet_csk_reqsk_queue_add(sk, req, child);
218 * when syncookies are in effect and tcp timestamps are enabled we stored
219 * additional tcp options in the timestamp.
220 * This extracts these options from the timestamp echo.
222 * The lowest 4 bits store snd_wscale.
223 * next 2 bits indicate SACK and ECN support.
225 * return false if we decode an option that should not be.
227 bool cookie_check_timestamp(struct tcp_options_received *tcp_opt,
228 struct net *net, bool *ecn_ok)
230 /* echoed timestamp, lowest bits contain options */
231 u32 options = tcp_opt->rcv_tsecr & TSMASK;
233 if (!tcp_opt->saw_tstamp) {
234 tcp_clear_options(tcp_opt);
238 if (!sysctl_tcp_timestamps)
241 tcp_opt->sack_ok = (options & (1 << 4)) ? TCP_SACK_SEEN : 0;
242 *ecn_ok = (options >> 5) & 1;
243 if (*ecn_ok && !net->ipv4.sysctl_tcp_ecn)
246 if (tcp_opt->sack_ok && !sysctl_tcp_sack)
249 if ((options & 0xf) == 0xf)
250 return true; /* no window scaling */
252 tcp_opt->wscale_ok = 1;
253 tcp_opt->snd_wscale = options & 0xf;
254 return sysctl_tcp_window_scaling != 0;
256 EXPORT_SYMBOL(cookie_check_timestamp);
258 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb)
260 struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
261 struct tcp_options_received tcp_opt;
262 struct inet_request_sock *ireq;
263 struct tcp_request_sock *treq;
264 struct tcp_sock *tp = tcp_sk(sk);
265 const struct tcphdr *th = tcp_hdr(skb);
266 __u32 cookie = ntohl(th->ack_seq) - 1;
267 struct sock *ret = sk;
268 struct request_sock *req;
275 if (!sysctl_tcp_syncookies || !th->ack || th->rst)
278 if (tcp_synq_no_recent_overflow(sk) ||
279 (mss = __cookie_v4_check(ip_hdr(skb), th, cookie)) == 0) {
280 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SYNCOOKIESFAILED);
284 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SYNCOOKIESRECV);
286 /* check for timestamp cookie support */
287 memset(&tcp_opt, 0, sizeof(tcp_opt));
288 tcp_parse_options(skb, &tcp_opt, 0, NULL);
290 if (!cookie_check_timestamp(&tcp_opt, sock_net(sk), &ecn_ok))
294 req = inet_reqsk_alloc(&tcp_request_sock_ops); /* for safety */
298 ireq = inet_rsk(req);
300 treq->rcv_isn = ntohl(th->seq) - 1;
301 treq->snt_isn = cookie;
303 ireq->ir_num = ntohs(th->dest);
304 ireq->ir_rmt_port = th->source;
305 ireq->ir_loc_addr = ip_hdr(skb)->daddr;
306 ireq->ir_rmt_addr = ip_hdr(skb)->saddr;
307 ireq->ir_mark = inet_request_mark(sk, skb);
308 ireq->ecn_ok = ecn_ok;
309 ireq->snd_wscale = tcp_opt.snd_wscale;
310 ireq->sack_ok = tcp_opt.sack_ok;
311 ireq->wscale_ok = tcp_opt.wscale_ok;
312 ireq->tstamp_ok = tcp_opt.saw_tstamp;
313 req->ts_recent = tcp_opt.saw_tstamp ? tcp_opt.rcv_tsval : 0;
314 treq->snt_synack = tcp_opt.saw_tstamp ? tcp_opt.rcv_tsecr : 0;
315 treq->listener = NULL;
317 /* We throwed the options of the initial SYN away, so we hope
318 * the ACK carries the same options again (see RFC1122 4.2.3.8)
320 ireq->opt = tcp_v4_save_options(skb);
322 if (security_inet_conn_request(sk, skb, req)) {
328 req->num_retrans = 0;
331 * We need to lookup the route here to get at the correct
332 * window size. We should better make sure that the window size
333 * hasn't changed since we received the original syn, but I see
334 * no easy way to do this.
336 flowi4_init_output(&fl4, sk->sk_bound_dev_if, ireq->ir_mark,
337 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE, IPPROTO_TCP,
338 inet_sk_flowi_flags(sk),
339 opt->srr ? opt->faddr : ireq->ir_rmt_addr,
340 ireq->ir_loc_addr, th->source, th->dest);
341 security_req_classify_flow(req, flowi4_to_flowi(&fl4));
342 rt = ip_route_output_key(sock_net(sk), &fl4);
348 /* Try to redo what tcp_v4_send_synack did. */
349 req->window_clamp = tp->window_clamp ? :dst_metric(&rt->dst, RTAX_WINDOW);
351 tcp_select_initial_window(tcp_full_space(sk), req->mss,
352 &req->rcv_wnd, &req->window_clamp,
353 ireq->wscale_ok, &rcv_wscale,
354 dst_metric(&rt->dst, RTAX_INITRWND));
356 ireq->rcv_wscale = rcv_wscale;
358 ret = get_cookie_sock(sk, skb, req, &rt->dst);
359 /* ip_queue_xmit() depends on our flow being setup
360 * Normal sockets get it right from inet_csk_route_child_sock()
363 inet_sk(ret)->cork.fl.u.ip4 = fl4;