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>
19 #include <net/route.h>
21 /* Timestamps: lowest bits store TCP options */
23 #define TSMASK (((__u32)1 << TSBITS) - 1)
25 extern int sysctl_tcp_syncookies;
27 __u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS];
28 EXPORT_SYMBOL(syncookie_secret);
30 static __init int init_syncookies(void)
32 get_random_bytes(syncookie_secret, sizeof(syncookie_secret));
35 __initcall(init_syncookies);
37 #define COOKIEBITS 24 /* Upper bits store count */
38 #define COOKIEMASK (((__u32)1 << COOKIEBITS) - 1)
40 static DEFINE_PER_CPU(__u32 [16 + 5 + SHA_WORKSPACE_WORDS],
43 static u32 cookie_hash(__be32 saddr, __be32 daddr, __be16 sport, __be16 dport,
46 __u32 *tmp = __get_cpu_var(ipv4_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 lower 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;
74 options = ireq->wscale_ok ? ireq->snd_wscale : 0xf;
75 options |= ireq->sack_ok << 4;
76 options |= ireq->ecn_ok << 5;
78 ts = ts_now & ~TSMASK;
90 static __u32 secure_tcp_syn_cookie(__be32 saddr, __be32 daddr, __be16 sport,
91 __be16 dport, __u32 sseq, __u32 count,
95 * Compute the secure sequence number.
96 * The output should be:
97 * HASH(sec1,saddr,sport,daddr,dport,sec1) + sseq + (count * 2^24)
98 * + (HASH(sec2,saddr,sport,daddr,dport,count,sec2) % 2^24).
99 * Where sseq is their sequence number and count increases every
101 * As an extra hack, we add a small "data" value that encodes the
102 * MSS into the second hash value.
105 return (cookie_hash(saddr, daddr, sport, dport, 0, 0) +
106 sseq + (count << COOKIEBITS) +
107 ((cookie_hash(saddr, daddr, sport, dport, count, 1) + data)
112 * This retrieves the small "data" value from the syncookie.
113 * If the syncookie is bad, the data returned will be out of
114 * range. This must be checked by the caller.
116 * The count value used to generate the cookie must be within
117 * "maxdiff" if the current (passed-in) "count". The return value
118 * is (__u32)-1 if this test fails.
120 static __u32 check_tcp_syn_cookie(__u32 cookie, __be32 saddr, __be32 daddr,
121 __be16 sport, __be16 dport, __u32 sseq,
122 __u32 count, __u32 maxdiff)
126 /* Strip away the layers from the cookie */
127 cookie -= cookie_hash(saddr, daddr, sport, dport, 0, 0) + sseq;
129 /* Cookie is now reduced to (count * 2^24) ^ (hash % 2^24) */
130 diff = (count - (cookie >> COOKIEBITS)) & ((__u32) - 1 >> COOKIEBITS);
135 cookie_hash(saddr, daddr, sport, dport, count - diff, 1))
136 & COOKIEMASK; /* Leaving the data behind */
140 * MSS Values are taken from the 2009 paper
141 * 'Measuring TCP Maximum Segment Size' by S. Alcock and R. Nelson:
142 * - values 1440 to 1460 accounted for 80% of observed mss values
143 * - values outside the 536-1460 range are rare (<0.2%).
145 * Table must be sorted.
147 static __u16 const msstab[] = {
159 * Generate a syncookie. mssp points to the mss, which is returned
160 * rounded down to the value encoded in the cookie.
162 __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb, __u16 *mssp)
164 const struct iphdr *iph = ip_hdr(skb);
165 const struct tcphdr *th = tcp_hdr(skb);
167 const __u16 mss = *mssp;
169 tcp_synq_overflow(sk);
171 for (mssind = ARRAY_SIZE(msstab) - 1; mssind ; mssind--)
172 if (mss >= msstab[mssind])
174 *mssp = msstab[mssind];
176 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
178 return secure_tcp_syn_cookie(iph->saddr, iph->daddr,
179 th->source, th->dest, ntohl(th->seq),
180 jiffies / (HZ * 60), mssind);
184 * This (misnamed) value is the age of syncookie which is permitted.
185 * Its ideal value should be dependent on TCP_TIMEOUT_INIT and
186 * sysctl_tcp_retries1. It's a rather complicated formula (exponential
187 * backoff) to compute at runtime so it's currently hardcoded here.
189 #define COUNTER_TRIES 4
191 * Check if a ack sequence number is a valid syncookie.
192 * Return the decoded mss if it is, or 0 if not.
194 static inline int cookie_check(struct sk_buff *skb, __u32 cookie)
196 const struct iphdr *iph = ip_hdr(skb);
197 const struct tcphdr *th = tcp_hdr(skb);
198 __u32 seq = ntohl(th->seq) - 1;
199 __u32 mssind = check_tcp_syn_cookie(cookie, iph->saddr, iph->daddr,
200 th->source, th->dest, seq,
204 return mssind < ARRAY_SIZE(msstab) ? msstab[mssind] : 0;
207 static inline struct sock *get_cookie_sock(struct sock *sk, struct sk_buff *skb,
208 struct request_sock *req,
209 struct dst_entry *dst)
211 struct inet_connection_sock *icsk = inet_csk(sk);
214 child = icsk->icsk_af_ops->syn_recv_sock(sk, skb, req, dst);
216 inet_csk_reqsk_queue_add(sk, req, child);
225 * when syncookies are in effect and tcp timestamps are enabled we stored
226 * additional tcp options in the timestamp.
227 * This extracts these options from the timestamp echo.
229 * The lowest 4 bits store snd_wscale.
230 * next 2 bits indicate SACK and ECN support.
232 * return false if we decode an option that should not be.
234 bool cookie_check_timestamp(struct tcp_options_received *tcp_opt, bool *ecn_ok)
236 /* echoed timestamp, lowest bits contain options */
237 u32 options = tcp_opt->rcv_tsecr & TSMASK;
239 if (!tcp_opt->saw_tstamp) {
240 tcp_clear_options(tcp_opt);
244 if (!sysctl_tcp_timestamps)
247 tcp_opt->sack_ok = (options >> 4) & 0x1;
248 *ecn_ok = (options >> 5) & 1;
249 if (*ecn_ok && !sysctl_tcp_ecn)
252 if (tcp_opt->sack_ok && !sysctl_tcp_sack)
255 if ((options & 0xf) == 0xf)
256 return true; /* no window scaling */
258 tcp_opt->wscale_ok = 1;
259 tcp_opt->snd_wscale = options & 0xf;
260 return sysctl_tcp_window_scaling != 0;
262 EXPORT_SYMBOL(cookie_check_timestamp);
264 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
265 struct ip_options *opt)
267 struct tcp_options_received tcp_opt;
269 struct inet_request_sock *ireq;
270 struct tcp_request_sock *treq;
271 struct tcp_sock *tp = tcp_sk(sk);
272 const struct tcphdr *th = tcp_hdr(skb);
273 __u32 cookie = ntohl(th->ack_seq) - 1;
274 struct sock *ret = sk;
275 struct request_sock *req;
281 if (!sysctl_tcp_syncookies || !th->ack || th->rst)
284 if (tcp_synq_no_recent_overflow(sk) ||
285 (mss = cookie_check(skb, cookie)) == 0) {
286 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SYNCOOKIESFAILED);
290 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SYNCOOKIESRECV);
292 /* check for timestamp cookie support */
293 memset(&tcp_opt, 0, sizeof(tcp_opt));
294 tcp_parse_options(skb, &tcp_opt, &hash_location, 0);
296 if (!cookie_check_timestamp(&tcp_opt, &ecn_ok))
300 req = inet_reqsk_alloc(&tcp_request_sock_ops); /* for safety */
304 ireq = inet_rsk(req);
306 treq->rcv_isn = ntohl(th->seq) - 1;
307 treq->snt_isn = cookie;
309 ireq->loc_port = th->dest;
310 ireq->rmt_port = th->source;
311 ireq->loc_addr = ip_hdr(skb)->daddr;
312 ireq->rmt_addr = ip_hdr(skb)->saddr;
313 ireq->ecn_ok = ecn_ok;
314 ireq->snd_wscale = tcp_opt.snd_wscale;
315 ireq->sack_ok = tcp_opt.sack_ok;
316 ireq->wscale_ok = tcp_opt.wscale_ok;
317 ireq->tstamp_ok = tcp_opt.saw_tstamp;
318 req->ts_recent = tcp_opt.saw_tstamp ? tcp_opt.rcv_tsval : 0;
320 /* We throwed the options of the initial SYN away, so we hope
321 * the ACK carries the same options again (see RFC1122 4.2.3.8)
323 if (opt && opt->optlen) {
324 int opt_size = sizeof(struct ip_options_rcu) + opt->optlen;
326 ireq->opt = kmalloc(opt_size, GFP_ATOMIC);
327 if (ireq->opt != NULL && ip_options_echo(&ireq->opt->opt, skb)) {
333 if (security_inet_conn_request(sk, skb, req)) {
342 * We need to lookup the route here to get at the correct
343 * window size. We should better make sure that the window size
344 * hasn't changed since we received the original syn, but I see
345 * no easy way to do this.
350 flowi4_init_output(&fl4, 0, sk->sk_mark, RT_CONN_FLAGS(sk),
351 RT_SCOPE_UNIVERSE, IPPROTO_TCP,
352 inet_sk_flowi_flags(sk),
353 (opt && opt->srr) ? opt->faddr : ireq->rmt_addr,
354 ireq->loc_addr, th->source, th->dest);
355 security_req_classify_flow(req, flowi4_to_flowi(&fl4));
356 rt = ip_route_output_key(sock_net(sk), &fl4);
363 /* Try to redo what tcp_v4_send_synack did. */
364 req->window_clamp = tp->window_clamp ? :dst_metric(&rt->dst, RTAX_WINDOW);
366 tcp_select_initial_window(tcp_full_space(sk), req->mss,
367 &req->rcv_wnd, &req->window_clamp,
368 ireq->wscale_ok, &rcv_wscale,
369 dst_metric(&rt->dst, RTAX_INITRWND));
371 ireq->rcv_wscale = rcv_wscale;
373 ret = get_cookie_sock(sk, skb, req, &rt->dst);