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.
45 static u32 cookie_hash(__be32 saddr, __be32 daddr, __be16 sport, __be16 dport,
48 net_get_random_once(syncookie_secret, sizeof(syncookie_secret));
49 return siphash_4u32((__force u32)saddr, (__force u32)daddr,
50 (__force u32)sport << 16 | (__force u32)dport,
51 count, &syncookie_secret[c]);
54 /* Convert one nsec 64bit timestamp to ts (ms or usec resolution) */
55 static u64 tcp_ns_to_ts(bool usec_ts, u64 val)
58 return div_u64(val, NSEC_PER_USEC);
60 return div_u64(val, NSEC_PER_MSEC);
64 * when syncookies are in effect and tcp timestamps are enabled we encode
65 * tcp options in the lower bits of the timestamp value that will be
66 * sent in the syn-ack.
67 * Since subsequent timestamps use the normal tcp_time_stamp value, we
68 * must make sure that the resulting initial timestamp is <= tcp_time_stamp.
70 u64 cookie_init_timestamp(struct request_sock *req, u64 now)
72 const struct inet_request_sock *ireq = inet_rsk(req);
73 u64 ts, ts_now = tcp_ns_to_ts(false, now);
76 options = ireq->wscale_ok ? ireq->snd_wscale : TS_OPT_WSCALE_MASK;
78 options |= TS_OPT_SACK;
80 options |= TS_OPT_ECN;
82 ts = (ts_now >> TSBITS) << TSBITS;
85 ts -= (1UL << TSBITS);
87 if (tcp_rsk(req)->req_usec_ts)
88 return ts * NSEC_PER_USEC;
89 return ts * NSEC_PER_MSEC;
93 static __u32 secure_tcp_syn_cookie(__be32 saddr, __be32 daddr, __be16 sport,
94 __be16 dport, __u32 sseq, __u32 data)
97 * Compute the secure sequence number.
98 * The output should be:
99 * HASH(sec1,saddr,sport,daddr,dport,sec1) + sseq + (count * 2^24)
100 * + (HASH(sec2,saddr,sport,daddr,dport,count,sec2) % 2^24).
101 * Where sseq is their sequence number and count increases every
103 * As an extra hack, we add a small "data" value that encodes the
104 * MSS into the second hash value.
106 u32 count = tcp_cookie_time();
107 return (cookie_hash(saddr, daddr, sport, dport, 0, 0) +
108 sseq + (count << COOKIEBITS) +
109 ((cookie_hash(saddr, daddr, sport, dport, count, 1) + data)
114 * This retrieves the small "data" value from the syncookie.
115 * If the syncookie is bad, the data returned will be out of
116 * range. This must be checked by the caller.
118 * The count value used to generate the cookie must be less than
119 * MAX_SYNCOOKIE_AGE minutes in the past.
120 * The return value (__u32)-1 if this test fails.
122 static __u32 check_tcp_syn_cookie(__u32 cookie, __be32 saddr, __be32 daddr,
123 __be16 sport, __be16 dport, __u32 sseq)
125 u32 diff, count = tcp_cookie_time();
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);
132 if (diff >= MAX_SYNCOOKIE_AGE)
136 cookie_hash(saddr, daddr, sport, dport, count - diff, 1))
137 & COOKIEMASK; /* Leaving the data behind */
141 * MSS Values are chosen based on the 2011 paper
142 * 'An Analysis of TCP Maximum Segement Sizes' by S. Alcock and R. Nelson.
144 * .. lower than 536 are rare (< 0.2%)
145 * .. between 537 and 1299 account for less than < 1.5% of observed values
146 * .. in the 1300-1349 range account for about 15 to 20% of observed mss values
147 * .. exceeding 1460 are very rare (< 0.04%)
149 * 1460 is the single most frequently announced mss value (30 to 46% depending
150 * on monitor location). Table must be sorted.
152 static __u16 const msstab[] = {
155 1440, /* 1440, 1452: PPPoE */
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(const struct iphdr *iph, const struct tcphdr *th,
167 const __u16 mss = *mssp;
169 for (mssind = ARRAY_SIZE(msstab) - 1; mssind ; mssind--)
170 if (mss >= msstab[mssind])
172 *mssp = msstab[mssind];
174 return secure_tcp_syn_cookie(iph->saddr, iph->daddr,
175 th->source, th->dest, ntohl(th->seq),
178 EXPORT_SYMBOL_GPL(__cookie_v4_init_sequence);
180 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mssp)
182 const struct iphdr *iph = ip_hdr(skb);
183 const struct tcphdr *th = tcp_hdr(skb);
185 return __cookie_v4_init_sequence(iph, th, mssp);
189 * Check if a ack sequence number is a valid syncookie.
190 * Return the decoded mss if it is, or 0 if not.
192 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
195 __u32 seq = ntohl(th->seq) - 1;
196 __u32 mssind = check_tcp_syn_cookie(cookie, iph->saddr, iph->daddr,
197 th->source, th->dest, seq);
199 return mssind < ARRAY_SIZE(msstab) ? msstab[mssind] : 0;
201 EXPORT_SYMBOL_GPL(__cookie_v4_check);
203 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
204 struct request_sock *req,
205 struct dst_entry *dst, u32 tsoff)
207 struct inet_connection_sock *icsk = inet_csk(sk);
211 child = icsk->icsk_af_ops->syn_recv_sock(sk, skb, req, dst,
214 refcount_set(&req->rsk_refcnt, 1);
215 tcp_sk(child)->tsoffset = tsoff;
216 sock_rps_save_rxhash(child, skb);
218 if (rsk_drop_req(req)) {
223 if (inet_csk_reqsk_queue_add(sk, req, child))
226 bh_unlock_sock(child);
233 EXPORT_SYMBOL(tcp_get_cookie_sock);
236 * when syncookies are in effect and tcp timestamps are enabled we stored
237 * additional tcp options in the timestamp.
238 * This extracts these options from the timestamp echo.
240 * return false if we decode a tcp option that is disabled
243 bool cookie_timestamp_decode(const struct net *net,
244 struct tcp_options_received *tcp_opt)
246 /* echoed timestamp, lowest bits contain options */
247 u32 options = tcp_opt->rcv_tsecr;
249 if (!tcp_opt->saw_tstamp) {
250 tcp_clear_options(tcp_opt);
254 if (!READ_ONCE(net->ipv4.sysctl_tcp_timestamps))
257 tcp_opt->sack_ok = (options & TS_OPT_SACK) ? TCP_SACK_SEEN : 0;
259 if (tcp_opt->sack_ok && !READ_ONCE(net->ipv4.sysctl_tcp_sack))
262 if ((options & TS_OPT_WSCALE_MASK) == TS_OPT_WSCALE_MASK)
263 return true; /* no window scaling */
265 tcp_opt->wscale_ok = 1;
266 tcp_opt->snd_wscale = options & TS_OPT_WSCALE_MASK;
268 return READ_ONCE(net->ipv4.sysctl_tcp_window_scaling) != 0;
270 EXPORT_SYMBOL(cookie_timestamp_decode);
272 bool cookie_ecn_ok(const struct tcp_options_received *tcp_opt,
273 const struct net *net, const struct dst_entry *dst)
275 bool ecn_ok = tcp_opt->rcv_tsecr & TS_OPT_ECN;
280 if (READ_ONCE(net->ipv4.sysctl_tcp_ecn))
283 return dst_feature(dst, RTAX_FEATURE_ECN);
285 EXPORT_SYMBOL(cookie_ecn_ok);
287 struct request_sock *cookie_tcp_reqsk_alloc(const struct request_sock_ops *ops,
288 const struct tcp_request_sock_ops *af_ops,
292 struct tcp_request_sock *treq;
293 struct request_sock *req;
296 req = mptcp_subflow_reqsk_alloc(ops, sk, false);
298 req = inet_reqsk_alloc(ops, sk, false);
305 /* treq->af_specific might be used to perform TCP_MD5 lookup */
306 treq->af_specific = af_ops;
308 treq->syn_tos = TCP_SKB_CB(skb)->ip_dsfield;
309 treq->req_usec_ts = false;
311 #if IS_ENABLED(CONFIG_MPTCP)
312 treq->is_mptcp = sk_is_mptcp(sk);
313 if (treq->is_mptcp) {
314 int err = mptcp_subflow_init_cookie_req(req, sk, skb);
325 EXPORT_SYMBOL_GPL(cookie_tcp_reqsk_alloc);
327 /* On input, sk is a listener.
328 * Output is listener if incoming packet would not create a child
329 * NULL if memory could not be allocated.
331 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb)
333 struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
334 struct tcp_options_received tcp_opt;
335 struct inet_request_sock *ireq;
336 struct tcp_request_sock *treq;
337 struct tcp_sock *tp = tcp_sk(sk);
338 const struct tcphdr *th = tcp_hdr(skb);
339 __u32 cookie = ntohl(th->ack_seq) - 1;
340 struct sock *ret = sk;
341 struct request_sock *req;
349 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies) ||
353 if (tcp_synq_no_recent_overflow(sk))
356 mss = __cookie_v4_check(ip_hdr(skb), th, cookie);
358 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESFAILED);
362 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESRECV);
364 /* check for timestamp cookie support */
365 memset(&tcp_opt, 0, sizeof(tcp_opt));
366 tcp_parse_options(sock_net(sk), skb, &tcp_opt, 0, NULL);
368 if (tcp_opt.saw_tstamp && tcp_opt.rcv_tsecr) {
369 tsoff = secure_tcp_ts_off(sock_net(sk),
372 tcp_opt.rcv_tsecr -= tsoff;
375 if (!cookie_timestamp_decode(sock_net(sk), &tcp_opt))
379 req = cookie_tcp_reqsk_alloc(&tcp_request_sock_ops,
380 &tcp_request_sock_ipv4_ops, sk, skb);
384 ireq = inet_rsk(req);
386 treq->rcv_isn = ntohl(th->seq) - 1;
387 treq->snt_isn = cookie;
389 treq->txhash = net_tx_rndhash();
391 ireq->ir_num = ntohs(th->dest);
392 ireq->ir_rmt_port = th->source;
393 sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr);
394 sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr);
395 ireq->ir_mark = inet_request_mark(sk, skb);
396 ireq->snd_wscale = tcp_opt.snd_wscale;
397 ireq->sack_ok = tcp_opt.sack_ok;
398 ireq->wscale_ok = tcp_opt.wscale_ok;
399 ireq->tstamp_ok = tcp_opt.saw_tstamp;
400 req->ts_recent = tcp_opt.saw_tstamp ? tcp_opt.rcv_tsval : 0;
401 treq->snt_synack = 0;
402 treq->tfo_listener = false;
404 if (IS_ENABLED(CONFIG_SMC))
407 ireq->ir_iif = inet_request_bound_dev_if(sk, skb);
409 l3index = l3mdev_master_ifindex_by_index(sock_net(sk), ireq->ir_iif);
410 tcp_ao_syncookie(sk, skb, treq, AF_INET, l3index);
412 /* We throwed the options of the initial SYN away, so we hope
413 * the ACK carries the same options again (see RFC1122 4.2.3.8)
415 RCU_INIT_POINTER(ireq->ireq_opt, tcp_v4_save_options(sock_net(sk), skb));
417 if (security_inet_conn_request(sk, skb, req)) {
422 req->num_retrans = 0;
425 * We need to lookup the route here to get at the correct
426 * window size. We should better make sure that the window size
427 * hasn't changed since we received the original syn, but I see
428 * no easy way to do this.
430 flowi4_init_output(&fl4, ireq->ir_iif, ireq->ir_mark,
431 ip_sock_rt_tos(sk), ip_sock_rt_scope(sk),
432 IPPROTO_TCP, inet_sk_flowi_flags(sk),
433 opt->srr ? opt->faddr : ireq->ir_rmt_addr,
434 ireq->ir_loc_addr, th->source, th->dest, sk->sk_uid);
435 security_req_classify_flow(req, flowi4_to_flowi_common(&fl4));
436 rt = ip_route_output_key(sock_net(sk), &fl4);
442 /* Try to redo what tcp_v4_send_synack did. */
443 req->rsk_window_clamp = tp->window_clamp ? :dst_metric(&rt->dst, RTAX_WINDOW);
444 /* limit the window selection if the user enforce a smaller rx buffer */
445 full_space = tcp_full_space(sk);
446 if (sk->sk_userlocks & SOCK_RCVBUF_LOCK &&
447 (req->rsk_window_clamp > full_space || req->rsk_window_clamp == 0))
448 req->rsk_window_clamp = full_space;
450 tcp_select_initial_window(sk, full_space, req->mss,
451 &req->rsk_rcv_wnd, &req->rsk_window_clamp,
452 ireq->wscale_ok, &rcv_wscale,
453 dst_metric(&rt->dst, RTAX_INITRWND));
455 ireq->rcv_wscale = rcv_wscale;
456 ireq->ecn_ok = cookie_ecn_ok(&tcp_opt, sock_net(sk), &rt->dst);
458 ret = tcp_get_cookie_sock(sk, skb, req, &rt->dst, tsoff);
459 /* ip_queue_xmit() depends on our flow being setup
460 * Normal sockets get it right from inet_csk_route_child_sock()
463 inet_sk(ret)->cork.fl.u.ip4 = fl4;