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/siphash.h>
11 #include <linux/kernel.h>
12 #include <linux/export.h>
13 #include <net/secure_seq.h>
15 #include <net/route.h>
17 static siphash_aligned_key_t syncookie_secret[2];
19 #define COOKIEBITS 24 /* Upper bits store count */
20 #define COOKIEMASK (((__u32)1 << COOKIEBITS) - 1)
22 /* TCP Timestamp: 6 lowest bits of timestamp sent in the cookie SYN-ACK
26 * | 31 ... 6 | 5 | 4 | 3 2 1 0 |
27 * | Timestamp | ECN | SACK | WScale |
29 * When we receive a valid cookie-ACK, we look at the echoed tsval (if
30 * any) to figure out which TCP options we should use for the rebuilt
33 * A WScale setting of '0xf' (which is an invalid scaling value)
34 * means that original syn did not include the TCP window scaling option.
36 #define TS_OPT_WSCALE_MASK 0xf
37 #define TS_OPT_SACK BIT(4)
38 #define TS_OPT_ECN BIT(5)
39 /* There is no TS_OPT_TIMESTAMP:
40 * if ACK contains timestamp option, we already know it was
41 * requested/supported by the syn/synack exchange.
44 #define TSMASK (((__u32)1 << TSBITS) - 1)
46 static u32 cookie_hash(__be32 saddr, __be32 daddr, __be16 sport, __be16 dport,
49 net_get_random_once(syncookie_secret, sizeof(syncookie_secret));
50 return siphash_4u32((__force u32)saddr, (__force u32)daddr,
51 (__force u32)sport << 16 | (__force u32)dport,
52 count, &syncookie_secret[c]);
57 * when syncookies are in effect and tcp timestamps are enabled we encode
58 * tcp options in the lower bits of the timestamp value that will be
59 * sent in the syn-ack.
60 * Since subsequent timestamps use the normal tcp_time_stamp value, we
61 * must make sure that the resulting initial timestamp is <= tcp_time_stamp.
63 u64 cookie_init_timestamp(struct request_sock *req, u64 now)
65 struct inet_request_sock *ireq;
66 u32 ts, ts_now = tcp_ns_to_ts(now);
71 options = ireq->wscale_ok ? ireq->snd_wscale : TS_OPT_WSCALE_MASK;
73 options |= TS_OPT_SACK;
75 options |= TS_OPT_ECN;
77 ts = ts_now & ~TSMASK;
85 return (u64)ts * (NSEC_PER_SEC / TCP_TS_HZ);
89 static __u32 secure_tcp_syn_cookie(__be32 saddr, __be32 daddr, __be16 sport,
90 __be16 dport, __u32 sseq, __u32 data)
93 * Compute the secure sequence number.
94 * The output should be:
95 * HASH(sec1,saddr,sport,daddr,dport,sec1) + sseq + (count * 2^24)
96 * + (HASH(sec2,saddr,sport,daddr,dport,count,sec2) % 2^24).
97 * Where sseq is their sequence number and count increases every
99 * As an extra hack, we add a small "data" value that encodes the
100 * MSS into the second hash value.
102 u32 count = tcp_cookie_time();
103 return (cookie_hash(saddr, daddr, sport, dport, 0, 0) +
104 sseq + (count << COOKIEBITS) +
105 ((cookie_hash(saddr, daddr, sport, dport, count, 1) + data)
110 * This retrieves the small "data" value from the syncookie.
111 * If the syncookie is bad, the data returned will be out of
112 * range. This must be checked by the caller.
114 * The count value used to generate the cookie must be less than
115 * MAX_SYNCOOKIE_AGE minutes in the past.
116 * The return value (__u32)-1 if this test fails.
118 static __u32 check_tcp_syn_cookie(__u32 cookie, __be32 saddr, __be32 daddr,
119 __be16 sport, __be16 dport, __u32 sseq)
121 u32 diff, count = tcp_cookie_time();
123 /* Strip away the layers from the cookie */
124 cookie -= cookie_hash(saddr, daddr, sport, dport, 0, 0) + sseq;
126 /* Cookie is now reduced to (count * 2^24) ^ (hash % 2^24) */
127 diff = (count - (cookie >> COOKIEBITS)) & ((__u32) -1 >> COOKIEBITS);
128 if (diff >= MAX_SYNCOOKIE_AGE)
132 cookie_hash(saddr, daddr, sport, dport, count - diff, 1))
133 & COOKIEMASK; /* Leaving the data behind */
137 * MSS Values are chosen based on the 2011 paper
138 * 'An Analysis of TCP Maximum Segement Sizes' by S. Alcock and R. Nelson.
140 * .. lower than 536 are rare (< 0.2%)
141 * .. between 537 and 1299 account for less than < 1.5% of observed values
142 * .. in the 1300-1349 range account for about 15 to 20% of observed mss values
143 * .. exceeding 1460 are very rare (< 0.04%)
145 * 1460 is the single most frequently announced mss value (30 to 46% depending
146 * on monitor location). Table must be sorted.
148 static __u16 const msstab[] = {
151 1440, /* 1440, 1452: PPPoE */
156 * Generate a syncookie. mssp points to the mss, which is returned
157 * rounded down to the value encoded in the cookie.
159 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
163 const __u16 mss = *mssp;
165 for (mssind = ARRAY_SIZE(msstab) - 1; mssind ; mssind--)
166 if (mss >= msstab[mssind])
168 *mssp = msstab[mssind];
170 return secure_tcp_syn_cookie(iph->saddr, iph->daddr,
171 th->source, th->dest, ntohl(th->seq),
174 EXPORT_SYMBOL_GPL(__cookie_v4_init_sequence);
176 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mssp)
178 const struct iphdr *iph = ip_hdr(skb);
179 const struct tcphdr *th = tcp_hdr(skb);
181 return __cookie_v4_init_sequence(iph, th, mssp);
185 * Check if a ack sequence number is a valid syncookie.
186 * Return the decoded mss if it is, or 0 if not.
188 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
191 __u32 seq = ntohl(th->seq) - 1;
192 __u32 mssind = check_tcp_syn_cookie(cookie, iph->saddr, iph->daddr,
193 th->source, th->dest, seq);
195 return mssind < ARRAY_SIZE(msstab) ? msstab[mssind] : 0;
197 EXPORT_SYMBOL_GPL(__cookie_v4_check);
199 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
200 struct request_sock *req,
201 struct dst_entry *dst, u32 tsoff)
203 struct inet_connection_sock *icsk = inet_csk(sk);
207 child = icsk->icsk_af_ops->syn_recv_sock(sk, skb, req, dst,
210 refcount_set(&req->rsk_refcnt, 1);
211 tcp_sk(child)->tsoffset = tsoff;
212 sock_rps_save_rxhash(child, skb);
214 if (rsk_drop_req(req)) {
219 if (inet_csk_reqsk_queue_add(sk, req, child))
222 bh_unlock_sock(child);
229 EXPORT_SYMBOL(tcp_get_cookie_sock);
232 * when syncookies are in effect and tcp timestamps are enabled we stored
233 * additional tcp options in the timestamp.
234 * This extracts these options from the timestamp echo.
236 * return false if we decode a tcp option that is disabled
239 bool cookie_timestamp_decode(const struct net *net,
240 struct tcp_options_received *tcp_opt)
242 /* echoed timestamp, lowest bits contain options */
243 u32 options = tcp_opt->rcv_tsecr;
245 if (!tcp_opt->saw_tstamp) {
246 tcp_clear_options(tcp_opt);
250 if (!READ_ONCE(net->ipv4.sysctl_tcp_timestamps))
253 tcp_opt->sack_ok = (options & TS_OPT_SACK) ? TCP_SACK_SEEN : 0;
255 if (tcp_opt->sack_ok && !READ_ONCE(net->ipv4.sysctl_tcp_sack))
258 if ((options & TS_OPT_WSCALE_MASK) == TS_OPT_WSCALE_MASK)
259 return true; /* no window scaling */
261 tcp_opt->wscale_ok = 1;
262 tcp_opt->snd_wscale = options & TS_OPT_WSCALE_MASK;
264 return READ_ONCE(net->ipv4.sysctl_tcp_window_scaling) != 0;
266 EXPORT_SYMBOL(cookie_timestamp_decode);
268 bool cookie_ecn_ok(const struct tcp_options_received *tcp_opt,
269 const struct net *net, const struct dst_entry *dst)
271 bool ecn_ok = tcp_opt->rcv_tsecr & TS_OPT_ECN;
276 if (READ_ONCE(net->ipv4.sysctl_tcp_ecn))
279 return dst_feature(dst, RTAX_FEATURE_ECN);
281 EXPORT_SYMBOL(cookie_ecn_ok);
283 struct request_sock *cookie_tcp_reqsk_alloc(const struct request_sock_ops *ops,
284 const struct tcp_request_sock_ops *af_ops,
288 struct tcp_request_sock *treq;
289 struct request_sock *req;
293 ops = &mptcp_subflow_request_sock_ops;
296 req = inet_reqsk_alloc(ops, sk, false);
302 /* treq->af_specific might be used to perform TCP_MD5 lookup */
303 treq->af_specific = af_ops;
305 treq->syn_tos = TCP_SKB_CB(skb)->ip_dsfield;
306 #if IS_ENABLED(CONFIG_MPTCP)
307 treq->is_mptcp = sk_is_mptcp(sk);
308 if (treq->is_mptcp) {
309 int err = mptcp_subflow_init_cookie_req(req, sk, skb);
320 EXPORT_SYMBOL_GPL(cookie_tcp_reqsk_alloc);
322 /* On input, sk is a listener.
323 * Output is listener if incoming packet would not create a child
324 * NULL if memory could not be allocated.
326 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb)
328 struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
329 struct tcp_options_received tcp_opt;
330 struct inet_request_sock *ireq;
331 struct tcp_request_sock *treq;
332 struct tcp_sock *tp = tcp_sk(sk);
333 const struct tcphdr *th = tcp_hdr(skb);
334 __u32 cookie = ntohl(th->ack_seq) - 1;
335 struct sock *ret = sk;
336 struct request_sock *req;
343 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies) ||
347 if (tcp_synq_no_recent_overflow(sk))
350 mss = __cookie_v4_check(ip_hdr(skb), th, cookie);
352 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESFAILED);
356 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESRECV);
358 /* check for timestamp cookie support */
359 memset(&tcp_opt, 0, sizeof(tcp_opt));
360 tcp_parse_options(sock_net(sk), skb, &tcp_opt, 0, NULL);
362 if (tcp_opt.saw_tstamp && tcp_opt.rcv_tsecr) {
363 tsoff = secure_tcp_ts_off(sock_net(sk),
366 tcp_opt.rcv_tsecr -= tsoff;
369 if (!cookie_timestamp_decode(sock_net(sk), &tcp_opt))
373 req = cookie_tcp_reqsk_alloc(&tcp_request_sock_ops,
374 &tcp_request_sock_ipv4_ops, sk, skb);
378 ireq = inet_rsk(req);
380 treq->rcv_isn = ntohl(th->seq) - 1;
381 treq->snt_isn = cookie;
383 treq->txhash = net_tx_rndhash();
385 ireq->ir_num = ntohs(th->dest);
386 ireq->ir_rmt_port = th->source;
387 sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr);
388 sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr);
389 ireq->ir_mark = inet_request_mark(sk, skb);
390 ireq->snd_wscale = tcp_opt.snd_wscale;
391 ireq->sack_ok = tcp_opt.sack_ok;
392 ireq->wscale_ok = tcp_opt.wscale_ok;
393 ireq->tstamp_ok = tcp_opt.saw_tstamp;
394 req->ts_recent = tcp_opt.saw_tstamp ? tcp_opt.rcv_tsval : 0;
395 treq->snt_synack = 0;
396 treq->tfo_listener = false;
398 if (IS_ENABLED(CONFIG_SMC))
401 ireq->ir_iif = inet_request_bound_dev_if(sk, skb);
403 /* We throwed the options of the initial SYN away, so we hope
404 * the ACK carries the same options again (see RFC1122 4.2.3.8)
406 RCU_INIT_POINTER(ireq->ireq_opt, tcp_v4_save_options(sock_net(sk), skb));
408 if (security_inet_conn_request(sk, skb, req)) {
413 req->num_retrans = 0;
416 * We need to lookup the route here to get at the correct
417 * window size. We should better make sure that the window size
418 * hasn't changed since we received the original syn, but I see
419 * no easy way to do this.
421 flowi4_init_output(&fl4, ireq->ir_iif, ireq->ir_mark,
422 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE, IPPROTO_TCP,
423 inet_sk_flowi_flags(sk),
424 opt->srr ? opt->faddr : ireq->ir_rmt_addr,
425 ireq->ir_loc_addr, th->source, th->dest, sk->sk_uid);
426 security_req_classify_flow(req, flowi4_to_flowi_common(&fl4));
427 rt = ip_route_output_key(sock_net(sk), &fl4);
433 /* Try to redo what tcp_v4_send_synack did. */
434 req->rsk_window_clamp = tp->window_clamp ? :dst_metric(&rt->dst, RTAX_WINDOW);
435 /* limit the window selection if the user enforce a smaller rx buffer */
436 full_space = tcp_full_space(sk);
437 if (sk->sk_userlocks & SOCK_RCVBUF_LOCK &&
438 (req->rsk_window_clamp > full_space || req->rsk_window_clamp == 0))
439 req->rsk_window_clamp = full_space;
441 tcp_select_initial_window(sk, full_space, req->mss,
442 &req->rsk_rcv_wnd, &req->rsk_window_clamp,
443 ireq->wscale_ok, &rcv_wscale,
444 dst_metric(&rt->dst, RTAX_INITRWND));
446 ireq->rcv_wscale = rcv_wscale;
447 ireq->ecn_ok = cookie_ecn_ok(&tcp_opt, sock_net(sk), &rt->dst);
449 ret = tcp_get_cookie_sock(sk, skb, req, &rt->dst, tsoff);
450 /* ip_queue_xmit() depends on our flow being setup
451 * Normal sockets get it right from inet_csk_route_child_sock()
454 inet_sk(ret)->cork.fl.u.ip4 = fl4;