1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Syncookies implementation for the Linux kernel
5 * Copyright (C) 1997 Andi Kleen
6 * Based on ideas by D.J.Bernstein and Eric Schenk.
10 #include <linux/slab.h>
11 #include <linux/random.h>
12 #include <linux/siphash.h>
13 #include <linux/kernel.h>
14 #include <linux/export.h>
15 #include <net/secure_seq.h>
17 #include <net/route.h>
19 static siphash_key_t syncookie_secret[2] __read_mostly;
21 #define COOKIEBITS 24 /* Upper bits store count */
22 #define COOKIEMASK (((__u32)1 << COOKIEBITS) - 1)
24 /* TCP Timestamp: 6 lowest bits of timestamp sent in the cookie SYN-ACK
28 * | 31 ... 6 | 5 | 4 | 3 2 1 0 |
29 * | Timestamp | ECN | SACK | WScale |
31 * When we receive a valid cookie-ACK, we look at the echoed tsval (if
32 * any) to figure out which TCP options we should use for the rebuilt
35 * A WScale setting of '0xf' (which is an invalid scaling value)
36 * means that original syn did not include the TCP window scaling option.
38 #define TS_OPT_WSCALE_MASK 0xf
39 #define TS_OPT_SACK BIT(4)
40 #define TS_OPT_ECN BIT(5)
41 /* There is no TS_OPT_TIMESTAMP:
42 * if ACK contains timestamp option, we already know it was
43 * requested/supported by the syn/synack exchange.
46 #define TSMASK (((__u32)1 << TSBITS) - 1)
48 static u32 cookie_hash(__be32 saddr, __be32 daddr, __be16 sport, __be16 dport,
51 net_get_random_once(syncookie_secret, sizeof(syncookie_secret));
52 return siphash_4u32((__force u32)saddr, (__force u32)daddr,
53 (__force u32)sport << 16 | (__force u32)dport,
54 count, &syncookie_secret[c]);
59 * when syncookies are in effect and tcp timestamps are enabled we encode
60 * tcp options in the lower bits of the timestamp value that will be
61 * sent in the syn-ack.
62 * Since subsequent timestamps use the normal tcp_time_stamp value, we
63 * must make sure that the resulting initial timestamp is <= tcp_time_stamp.
65 u64 cookie_init_timestamp(struct request_sock *req, u64 now)
67 struct inet_request_sock *ireq;
68 u32 ts, ts_now = tcp_ns_to_ts(now);
73 options = ireq->wscale_ok ? ireq->snd_wscale : TS_OPT_WSCALE_MASK;
75 options |= TS_OPT_SACK;
77 options |= TS_OPT_ECN;
79 ts = ts_now & ~TSMASK;
87 return (u64)ts * (NSEC_PER_SEC / TCP_TS_HZ);
91 static __u32 secure_tcp_syn_cookie(__be32 saddr, __be32 daddr, __be16 sport,
92 __be16 dport, __u32 sseq, __u32 data)
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.
104 u32 count = tcp_cookie_time();
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 less than
117 * MAX_SYNCOOKIE_AGE minutes in the past.
118 * The return value (__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)
123 u32 diff, count = tcp_cookie_time();
125 /* Strip away the layers from the cookie */
126 cookie -= cookie_hash(saddr, daddr, sport, dport, 0, 0) + sseq;
128 /* Cookie is now reduced to (count * 2^24) ^ (hash % 2^24) */
129 diff = (count - (cookie >> COOKIEBITS)) & ((__u32) -1 >> COOKIEBITS);
130 if (diff >= MAX_SYNCOOKIE_AGE)
134 cookie_hash(saddr, daddr, sport, dport, count - diff, 1))
135 & COOKIEMASK; /* Leaving the data behind */
139 * MSS Values are chosen based on the 2011 paper
140 * 'An Analysis of TCP Maximum Segement Sizes' by S. Alcock and R. Nelson.
142 * .. lower than 536 are rare (< 0.2%)
143 * .. between 537 and 1299 account for less than < 1.5% of observed values
144 * .. in the 1300-1349 range account for about 15 to 20% of observed mss values
145 * .. exceeding 1460 are very rare (< 0.04%)
147 * 1460 is the single most frequently announced mss value (30 to 46% depending
148 * on monitor location). Table must be sorted.
150 static __u16 const msstab[] = {
153 1440, /* 1440, 1452: PPPoE */
158 * Generate a syncookie. mssp points to the mss, which is returned
159 * rounded down to the value encoded in the cookie.
161 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
165 const __u16 mss = *mssp;
167 for (mssind = ARRAY_SIZE(msstab) - 1; mssind ; mssind--)
168 if (mss >= msstab[mssind])
170 *mssp = msstab[mssind];
172 return secure_tcp_syn_cookie(iph->saddr, iph->daddr,
173 th->source, th->dest, ntohl(th->seq),
176 EXPORT_SYMBOL_GPL(__cookie_v4_init_sequence);
178 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mssp)
180 const struct iphdr *iph = ip_hdr(skb);
181 const struct tcphdr *th = tcp_hdr(skb);
183 return __cookie_v4_init_sequence(iph, th, mssp);
187 * Check if a ack sequence number is a valid syncookie.
188 * Return the decoded mss if it is, or 0 if not.
190 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
193 __u32 seq = ntohl(th->seq) - 1;
194 __u32 mssind = check_tcp_syn_cookie(cookie, iph->saddr, iph->daddr,
195 th->source, th->dest, seq);
197 return mssind < ARRAY_SIZE(msstab) ? msstab[mssind] : 0;
199 EXPORT_SYMBOL_GPL(__cookie_v4_check);
201 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
202 struct request_sock *req,
203 struct dst_entry *dst, u32 tsoff)
205 struct inet_connection_sock *icsk = inet_csk(sk);
209 child = icsk->icsk_af_ops->syn_recv_sock(sk, skb, req, dst,
212 refcount_set(&req->rsk_refcnt, 1);
213 tcp_sk(child)->tsoffset = tsoff;
214 sock_rps_save_rxhash(child, skb);
216 if (rsk_drop_req(req)) {
221 if (inet_csk_reqsk_queue_add(sk, req, child))
224 bh_unlock_sock(child);
231 EXPORT_SYMBOL(tcp_get_cookie_sock);
234 * when syncookies are in effect and tcp timestamps are enabled we stored
235 * additional tcp options in the timestamp.
236 * This extracts these options from the timestamp echo.
238 * return false if we decode a tcp option that is disabled
241 bool cookie_timestamp_decode(const struct net *net,
242 struct tcp_options_received *tcp_opt)
244 /* echoed timestamp, lowest bits contain options */
245 u32 options = tcp_opt->rcv_tsecr;
247 if (!tcp_opt->saw_tstamp) {
248 tcp_clear_options(tcp_opt);
252 if (!net->ipv4.sysctl_tcp_timestamps)
255 tcp_opt->sack_ok = (options & TS_OPT_SACK) ? TCP_SACK_SEEN : 0;
257 if (tcp_opt->sack_ok && !net->ipv4.sysctl_tcp_sack)
260 if ((options & TS_OPT_WSCALE_MASK) == TS_OPT_WSCALE_MASK)
261 return true; /* no window scaling */
263 tcp_opt->wscale_ok = 1;
264 tcp_opt->snd_wscale = options & TS_OPT_WSCALE_MASK;
266 return net->ipv4.sysctl_tcp_window_scaling != 0;
268 EXPORT_SYMBOL(cookie_timestamp_decode);
270 bool cookie_ecn_ok(const struct tcp_options_received *tcp_opt,
271 const struct net *net, const struct dst_entry *dst)
273 bool ecn_ok = tcp_opt->rcv_tsecr & TS_OPT_ECN;
278 if (net->ipv4.sysctl_tcp_ecn)
281 return dst_feature(dst, RTAX_FEATURE_ECN);
283 EXPORT_SYMBOL(cookie_ecn_ok);
285 struct request_sock *cookie_tcp_reqsk_alloc(const struct request_sock_ops *ops,
289 struct tcp_request_sock *treq;
290 struct request_sock *req;
294 ops = &mptcp_subflow_request_sock_ops;
297 req = inet_reqsk_alloc(ops, sk, false);
302 treq->syn_tos = TCP_SKB_CB(skb)->ip_dsfield;
303 #if IS_ENABLED(CONFIG_MPTCP)
304 treq->is_mptcp = sk_is_mptcp(sk);
305 if (treq->is_mptcp) {
306 int err = mptcp_subflow_init_cookie_req(req, sk, skb);
317 EXPORT_SYMBOL_GPL(cookie_tcp_reqsk_alloc);
319 /* On input, sk is a listener.
320 * Output is listener if incoming packet would not create a child
321 * NULL if memory could not be allocated.
323 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb)
325 struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
326 struct tcp_options_received tcp_opt;
327 struct inet_request_sock *ireq;
328 struct tcp_request_sock *treq;
329 struct tcp_sock *tp = tcp_sk(sk);
330 const struct tcphdr *th = tcp_hdr(skb);
331 __u32 cookie = ntohl(th->ack_seq) - 1;
332 struct sock *ret = sk;
333 struct request_sock *req;
340 if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies || !th->ack || th->rst)
343 if (tcp_synq_no_recent_overflow(sk))
346 mss = __cookie_v4_check(ip_hdr(skb), th, cookie);
348 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESFAILED);
352 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESRECV);
354 /* check for timestamp cookie support */
355 memset(&tcp_opt, 0, sizeof(tcp_opt));
356 tcp_parse_options(sock_net(sk), skb, &tcp_opt, 0, NULL);
358 if (tcp_opt.saw_tstamp && tcp_opt.rcv_tsecr) {
359 tsoff = secure_tcp_ts_off(sock_net(sk),
362 tcp_opt.rcv_tsecr -= tsoff;
365 if (!cookie_timestamp_decode(sock_net(sk), &tcp_opt))
369 req = cookie_tcp_reqsk_alloc(&tcp_request_sock_ops, sk, skb);
373 ireq = inet_rsk(req);
375 treq->rcv_isn = ntohl(th->seq) - 1;
376 treq->snt_isn = cookie;
378 treq->txhash = net_tx_rndhash();
380 ireq->ir_num = ntohs(th->dest);
381 ireq->ir_rmt_port = th->source;
382 sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr);
383 sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr);
384 ireq->ir_mark = inet_request_mark(sk, skb);
385 ireq->snd_wscale = tcp_opt.snd_wscale;
386 ireq->sack_ok = tcp_opt.sack_ok;
387 ireq->wscale_ok = tcp_opt.wscale_ok;
388 ireq->tstamp_ok = tcp_opt.saw_tstamp;
389 req->ts_recent = tcp_opt.saw_tstamp ? tcp_opt.rcv_tsval : 0;
390 treq->snt_synack = 0;
391 treq->tfo_listener = false;
393 if (IS_ENABLED(CONFIG_SMC))
396 ireq->ir_iif = inet_request_bound_dev_if(sk, skb);
398 /* We throwed the options of the initial SYN away, so we hope
399 * the ACK carries the same options again (see RFC1122 4.2.3.8)
401 RCU_INIT_POINTER(ireq->ireq_opt, tcp_v4_save_options(sock_net(sk), skb));
403 if (security_inet_conn_request(sk, skb, req)) {
408 req->num_retrans = 0;
411 * We need to lookup the route here to get at the correct
412 * window size. We should better make sure that the window size
413 * hasn't changed since we received the original syn, but I see
414 * no easy way to do this.
416 flowi4_init_output(&fl4, ireq->ir_iif, ireq->ir_mark,
417 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE, IPPROTO_TCP,
418 inet_sk_flowi_flags(sk),
419 opt->srr ? opt->faddr : ireq->ir_rmt_addr,
420 ireq->ir_loc_addr, th->source, th->dest, sk->sk_uid);
421 security_req_classify_flow(req, flowi4_to_flowi_common(&fl4));
422 rt = ip_route_output_key(sock_net(sk), &fl4);
428 /* Try to redo what tcp_v4_send_synack did. */
429 req->rsk_window_clamp = tp->window_clamp ? :dst_metric(&rt->dst, RTAX_WINDOW);
430 /* limit the window selection if the user enforce a smaller rx buffer */
431 full_space = tcp_full_space(sk);
432 if (sk->sk_userlocks & SOCK_RCVBUF_LOCK &&
433 (req->rsk_window_clamp > full_space || req->rsk_window_clamp == 0))
434 req->rsk_window_clamp = full_space;
436 tcp_select_initial_window(sk, full_space, req->mss,
437 &req->rsk_rcv_wnd, &req->rsk_window_clamp,
438 ireq->wscale_ok, &rcv_wscale,
439 dst_metric(&rt->dst, RTAX_INITRWND));
441 ireq->rcv_wscale = rcv_wscale;
442 ireq->ecn_ok = cookie_ecn_ok(&tcp_opt, sock_net(sk), &rt->dst);
444 ret = tcp_get_cookie_sock(sk, skb, req, &rt->dst, tsoff);
445 /* ip_queue_xmit() depends on our flow being setup
446 * Normal sockets get it right from inet_csk_route_child_sock()
449 inet_sk(ret)->cork.fl.u.ip4 = fl4;