1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * Syncookies implementation for the Linux kernel
4 *
5 * Copyright (C) 1997 Andi Kleen
6 * Based on ideas by D.J.Bernstein and Eric Schenk.
7 */
8
9#include <linux/tcp.h>
10#include <linux/siphash.h>
11#include <linux/kernel.h>
12#include <linux/export.h>
13#include <net/secure_seq.h>
14#include <net/tcp.h>
15#include <net/tcp_ecn.h>
16#include <net/route.h>
17
18static siphash_aligned_key_t syncookie_secret[2];
19
20#define COOKIEBITS 24 /* Upper bits store count */
21#define COOKIEMASK (((__u32)1 << COOKIEBITS) - 1)
22
23/* TCP Timestamp: 6 lowest bits of timestamp sent in the cookie SYN-ACK
24 * stores TCP options:
25 *
26 * MSB LSB
27 * | 31 ... 6 | 5 | 4 | 3 2 1 0 |
28 * | Timestamp | ECN | SACK | WScale |
29 *
30 * When we receive a valid cookie-ACK, we look at the echoed tsval (if
31 * any) to figure out which TCP options we should use for the rebuilt
32 * connection.
33 *
34 * A WScale setting of '0xf' (which is an invalid scaling value)
35 * means that original syn did not include the TCP window scaling option.
36 */
37#define TS_OPT_WSCALE_MASK 0xf
38#define TS_OPT_SACK BIT(4)
39#define TS_OPT_ECN BIT(5)
40/* There is no TS_OPT_TIMESTAMP:
41 * if ACK contains timestamp option, we already know it was
42 * requested/supported by the syn/synack exchange.
43 */
44#define TSBITS 6
45
46static u32 cookie_hash(__be32 saddr, __be32 daddr, __be16 sport, __be16 dport,
47 u32 count, int c)
48{
49 net_get_random_once(syncookie_secret, sizeof(syncookie_secret));
50 return siphash_4u32(a: (__force u32)saddr, b: (__force u32)daddr,
51 c: (__force u32)sport << 16 | (__force u32)dport,
52 d: count, key: &syncookie_secret[c]);
53}
54
55/*
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.
61 */
62u64 cookie_init_timestamp(struct request_sock *req, u64 now)
63{
64 const struct inet_request_sock *ireq = inet_rsk(sk: req);
65 u64 ts, ts_now = tcp_ns_to_ts(usec_ts: false, val: now);
66 u32 options = 0;
67
68 options = ireq->wscale_ok ? ireq->snd_wscale : TS_OPT_WSCALE_MASK;
69 if (ireq->sack_ok)
70 options |= TS_OPT_SACK;
71 if (ireq->ecn_ok)
72 options |= TS_OPT_ECN;
73
74 ts = (ts_now >> TSBITS) << TSBITS;
75 ts |= options;
76 if (ts > ts_now)
77 ts -= (1UL << TSBITS);
78
79 if (tcp_rsk(req)->req_usec_ts)
80 return ts * NSEC_PER_USEC;
81 return ts * NSEC_PER_MSEC;
82}
83
84
85static __u32 secure_tcp_syn_cookie(__be32 saddr, __be32 daddr, __be16 sport,
86 __be16 dport, __u32 sseq, __u32 data)
87{
88 /*
89 * Compute the secure sequence number.
90 * The output should be:
91 * HASH(sec1,saddr,sport,daddr,dport,sec1) + sseq + (count * 2^24)
92 * + (HASH(sec2,saddr,sport,daddr,dport,count,sec2) % 2^24).
93 * Where sseq is their sequence number and count increases every
94 * minute by 1.
95 * As an extra hack, we add a small "data" value that encodes the
96 * MSS into the second hash value.
97 */
98 u32 count = tcp_cookie_time();
99 return (cookie_hash(saddr, daddr, sport, dport, count: 0, c: 0) +
100 sseq + (count << COOKIEBITS) +
101 ((cookie_hash(saddr, daddr, sport, dport, count, c: 1) + data)
102 & COOKIEMASK));
103}
104
105/*
106 * This retrieves the small "data" value from the syncookie.
107 * If the syncookie is bad, the data returned will be out of
108 * range. This must be checked by the caller.
109 *
110 * The count value used to generate the cookie must be less than
111 * MAX_SYNCOOKIE_AGE minutes in the past.
112 * The return value (__u32)-1 if this test fails.
113 */
114static __u32 check_tcp_syn_cookie(__u32 cookie, __be32 saddr, __be32 daddr,
115 __be16 sport, __be16 dport, __u32 sseq)
116{
117 u32 diff, count = tcp_cookie_time();
118
119 /* Strip away the layers from the cookie */
120 cookie -= cookie_hash(saddr, daddr, sport, dport, count: 0, c: 0) + sseq;
121
122 /* Cookie is now reduced to (count * 2^24) ^ (hash % 2^24) */
123 diff = (count - (cookie >> COOKIEBITS)) & ((__u32) -1 >> COOKIEBITS);
124 if (diff >= MAX_SYNCOOKIE_AGE)
125 return (__u32)-1;
126
127 return (cookie -
128 cookie_hash(saddr, daddr, sport, dport, count: count - diff, c: 1))
129 & COOKIEMASK; /* Leaving the data behind */
130}
131
132/*
133 * MSS Values are chosen based on the 2011 paper
134 * 'An Analysis of TCP Maximum Segement Sizes' by S. Alcock and R. Nelson.
135 * Values ..
136 * .. lower than 536 are rare (< 0.2%)
137 * .. between 537 and 1299 account for less than < 1.5% of observed values
138 * .. in the 1300-1349 range account for about 15 to 20% of observed mss values
139 * .. exceeding 1460 are very rare (< 0.04%)
140 *
141 * 1460 is the single most frequently announced mss value (30 to 46% depending
142 * on monitor location). Table must be sorted.
143 */
144static __u16 const msstab[] = {
145 536,
146 1300,
147 1440, /* 1440, 1452: PPPoE */
148 1460,
149};
150
151/*
152 * Generate a syncookie. mssp points to the mss, which is returned
153 * rounded down to the value encoded in the cookie.
154 */
155u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
156 u16 *mssp)
157{
158 int mssind;
159 const __u16 mss = *mssp;
160
161 for (mssind = ARRAY_SIZE(msstab) - 1; mssind ; mssind--)
162 if (mss >= msstab[mssind])
163 break;
164 *mssp = msstab[mssind];
165
166 return secure_tcp_syn_cookie(saddr: iph->saddr, daddr: iph->daddr,
167 sport: th->source, dport: th->dest, ntohl(th->seq),
168 data: mssind);
169}
170EXPORT_SYMBOL_GPL(__cookie_v4_init_sequence);
171
172__u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mssp)
173{
174 const struct iphdr *iph = ip_hdr(skb);
175 const struct tcphdr *th = tcp_hdr(skb);
176
177 return __cookie_v4_init_sequence(iph, th, mssp);
178}
179
180/*
181 * Check if a ack sequence number is a valid syncookie.
182 * Return the decoded mss if it is, or 0 if not.
183 */
184int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th)
185{
186 __u32 cookie = ntohl(th->ack_seq) - 1;
187 __u32 seq = ntohl(th->seq) - 1;
188 __u32 mssind;
189
190 mssind = check_tcp_syn_cookie(cookie, saddr: iph->saddr, daddr: iph->daddr,
191 sport: th->source, dport: th->dest, sseq: seq);
192
193 return mssind < ARRAY_SIZE(msstab) ? msstab[mssind] : 0;
194}
195EXPORT_SYMBOL_GPL(__cookie_v4_check);
196
197struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
198 struct request_sock *req,
199 struct dst_entry *dst)
200{
201 struct inet_connection_sock *icsk = inet_csk(sk);
202 struct sock *child;
203 bool own_req;
204
205 child = icsk->icsk_af_ops->syn_recv_sock(sk, skb, req, dst,
206 NULL, &own_req);
207 if (child) {
208 refcount_set(r: &req->rsk_refcnt, n: 1);
209 sock_rps_save_rxhash(sk: child, skb);
210
211 if (rsk_drop_req(req)) {
212 reqsk_put(req);
213 return child;
214 }
215
216 if (inet_csk_reqsk_queue_add(sk, req, child))
217 return child;
218
219 bh_unlock_sock(child);
220 sock_put(sk: child);
221 }
222 __reqsk_free(req);
223
224 return NULL;
225}
226EXPORT_IPV6_MOD(tcp_get_cookie_sock);
227
228/*
229 * when syncookies are in effect and tcp timestamps are enabled we stored
230 * additional tcp options in the timestamp.
231 * This extracts these options from the timestamp echo.
232 *
233 * return false if we decode a tcp option that is disabled
234 * on the host.
235 */
236bool cookie_timestamp_decode(const struct net *net,
237 struct tcp_options_received *tcp_opt)
238{
239 /* echoed timestamp, lowest bits contain options */
240 u32 options = tcp_opt->rcv_tsecr;
241
242 if (!tcp_opt->saw_tstamp) {
243 tcp_clear_options(rx_opt: tcp_opt);
244 return true;
245 }
246
247 if (!READ_ONCE(net->ipv4.sysctl_tcp_timestamps))
248 return false;
249
250 tcp_opt->sack_ok = (options & TS_OPT_SACK) ? TCP_SACK_SEEN : 0;
251
252 if (tcp_opt->sack_ok && !READ_ONCE(net->ipv4.sysctl_tcp_sack))
253 return false;
254
255 if ((options & TS_OPT_WSCALE_MASK) == TS_OPT_WSCALE_MASK)
256 return true; /* no window scaling */
257
258 tcp_opt->wscale_ok = 1;
259 tcp_opt->snd_wscale = options & TS_OPT_WSCALE_MASK;
260
261 return READ_ONCE(net->ipv4.sysctl_tcp_window_scaling) != 0;
262}
263EXPORT_IPV6_MOD(cookie_timestamp_decode);
264
265static int cookie_tcp_reqsk_init(struct sock *sk, struct sk_buff *skb,
266 struct request_sock *req)
267{
268 struct inet_request_sock *ireq = inet_rsk(sk: req);
269 struct tcp_request_sock *treq = tcp_rsk(req);
270 const struct tcphdr *th = tcp_hdr(skb);
271
272 req->num_retrans = 0;
273
274 ireq->ir_num = ntohs(th->dest);
275 ireq->ir_rmt_port = th->source;
276 ireq->ir_iif = inet_request_bound_dev_if(sk, skb);
277 ireq->ir_mark = inet_request_mark(sk, skb);
278
279 if (IS_ENABLED(CONFIG_SMC))
280 ireq->smc_ok = 0;
281
282 treq->snt_synack = 0;
283 treq->snt_tsval_first = 0;
284 treq->tfo_listener = false;
285 treq->txhash = net_tx_rndhash();
286 treq->rcv_isn = ntohl(th->seq) - 1;
287 treq->snt_isn = ntohl(th->ack_seq) - 1;
288 treq->syn_tos = TCP_SKB_CB(skb)->ip_dsfield;
289 treq->req_usec_ts = false;
290
291#if IS_ENABLED(CONFIG_MPTCP)
292 treq->is_mptcp = sk_is_mptcp(sk);
293 if (treq->is_mptcp)
294 return mptcp_subflow_init_cookie_req(req, sk, skb);
295#endif
296
297 return 0;
298}
299
300#if IS_ENABLED(CONFIG_BPF)
301struct request_sock *cookie_bpf_check(struct sock *sk, struct sk_buff *skb)
302{
303 struct request_sock *req = inet_reqsk(sk: skb->sk);
304
305 skb->sk = NULL;
306 skb->destructor = NULL;
307
308 if (cookie_tcp_reqsk_init(sk, skb, req)) {
309 reqsk_free(req);
310 req = NULL;
311 }
312
313 return req;
314}
315EXPORT_IPV6_MOD_GPL(cookie_bpf_check);
316#endif
317
318struct request_sock *cookie_tcp_reqsk_alloc(const struct request_sock_ops *ops,
319 struct sock *sk, struct sk_buff *skb,
320 struct tcp_options_received *tcp_opt,
321 int mss, u32 tsoff)
322{
323 struct inet_request_sock *ireq;
324 struct tcp_request_sock *treq;
325 struct request_sock *req;
326
327 if (sk_is_mptcp(sk))
328 req = mptcp_subflow_reqsk_alloc(ops, sk_listener: sk, attach_listener: false);
329 else
330 req = inet_reqsk_alloc(ops, sk_listener: sk, attach_listener: false);
331
332 if (!req)
333 return NULL;
334
335 if (cookie_tcp_reqsk_init(sk, skb, req)) {
336 reqsk_free(req);
337 return NULL;
338 }
339
340 ireq = inet_rsk(sk: req);
341 treq = tcp_rsk(req);
342
343 req->mss = mss;
344 req->ts_recent = tcp_opt->saw_tstamp ? tcp_opt->rcv_tsval : 0;
345
346 ireq->snd_wscale = tcp_opt->snd_wscale;
347 ireq->tstamp_ok = tcp_opt->saw_tstamp;
348 ireq->sack_ok = tcp_opt->sack_ok;
349 ireq->wscale_ok = tcp_opt->wscale_ok;
350 ireq->ecn_ok = !!(tcp_opt->rcv_tsecr & TS_OPT_ECN);
351
352 treq->ts_off = tsoff;
353
354 return req;
355}
356EXPORT_IPV6_MOD_GPL(cookie_tcp_reqsk_alloc);
357
358static struct request_sock *cookie_tcp_check(struct net *net, struct sock *sk,
359 struct sk_buff *skb)
360{
361 struct tcp_options_received tcp_opt;
362 u32 tsoff = 0;
363 int mss;
364
365 if (tcp_synq_no_recent_overflow(sk))
366 goto out;
367
368 mss = __cookie_v4_check(ip_hdr(skb), tcp_hdr(skb));
369 if (!mss) {
370 __NET_INC_STATS(net, LINUX_MIB_SYNCOOKIESFAILED);
371 goto out;
372 }
373
374 __NET_INC_STATS(net, LINUX_MIB_SYNCOOKIESRECV);
375
376 /* check for timestamp cookie support */
377 memset(s: &tcp_opt, c: 0, n: sizeof(tcp_opt));
378 tcp_parse_options(net, skb, opt_rx: &tcp_opt, estab: 0, NULL);
379
380 if (tcp_opt.saw_tstamp && tcp_opt.rcv_tsecr) {
381 tsoff = secure_tcp_ts_off(net,
382 saddr: ip_hdr(skb)->daddr,
383 daddr: ip_hdr(skb)->saddr);
384 tcp_opt.rcv_tsecr -= tsoff;
385 }
386
387 if (!cookie_timestamp_decode(net, tcp_opt: &tcp_opt))
388 goto out;
389
390 return cookie_tcp_reqsk_alloc(ops: &tcp_request_sock_ops, sk, skb,
391 tcp_opt: &tcp_opt, mss, tsoff);
392out:
393 return ERR_PTR(error: -EINVAL);
394}
395
396/* On input, sk is a listener.
397 * Output is listener if incoming packet would not create a child
398 * NULL if memory could not be allocated.
399 */
400struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb)
401{
402 struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
403 const struct tcphdr *th = tcp_hdr(skb);
404 struct tcp_sock *tp = tcp_sk(sk);
405 struct inet_request_sock *ireq;
406 struct net *net = sock_net(sk);
407 struct tcp_request_sock *treq;
408 struct request_sock *req;
409 struct sock *ret = sk;
410 struct flowi4 fl4;
411 struct rtable *rt;
412 __u8 rcv_wscale;
413 int full_space;
414 SKB_DR(reason);
415
416 if (!READ_ONCE(net->ipv4.sysctl_tcp_syncookies) ||
417 !th->ack || th->rst)
418 goto out;
419
420 if (cookie_bpf_ok(skb)) {
421 req = cookie_bpf_check(sk, skb);
422 } else {
423 req = cookie_tcp_check(net, sk, skb);
424 if (IS_ERR(ptr: req))
425 goto out;
426 }
427 if (!req) {
428 SKB_DR_SET(reason, NO_SOCKET);
429 goto out_drop;
430 }
431
432 ireq = inet_rsk(sk: req);
433 treq = tcp_rsk(req);
434
435 sk_rcv_saddr_set(sk: req_to_sk(req), addr: ip_hdr(skb)->daddr);
436 sk_daddr_set(sk: req_to_sk(req), addr: ip_hdr(skb)->saddr);
437
438 /* We throwed the options of the initial SYN away, so we hope
439 * the ACK carries the same options again (see RFC1122 4.2.3.8)
440 */
441 RCU_INIT_POINTER(ireq->ireq_opt, tcp_v4_save_options(net, skb));
442
443 if (security_inet_conn_request(sk, skb, req)) {
444 SKB_DR_SET(reason, SECURITY_HOOK);
445 goto out_free;
446 }
447
448 tcp_ao_syncookie(sk, skb, req, AF_INET);
449
450 /*
451 * We need to lookup the route here to get at the correct
452 * window size. We should better make sure that the window size
453 * hasn't changed since we received the original syn, but I see
454 * no easy way to do this.
455 */
456 flowi4_init_output(fl4: &fl4, oif: ireq->ir_iif, mark: ireq->ir_mark,
457 tos: ip_sock_rt_tos(sk), scope: ip_sock_rt_scope(sk),
458 IPPROTO_TCP, flags: inet_sk_flowi_flags(sk),
459 daddr: opt->srr ? opt->faddr : ireq->ir_rmt_addr,
460 saddr: ireq->ir_loc_addr, dport: th->source, sport: th->dest,
461 uid: sk_uid(sk));
462 security_req_classify_flow(req, flic: flowi4_to_flowi_common(fl4: &fl4));
463 rt = ip_route_output_key(net, flp: &fl4);
464 if (IS_ERR(ptr: rt)) {
465 SKB_DR_SET(reason, IP_OUTNOROUTES);
466 goto out_free;
467 }
468
469 /* Try to redo what tcp_v4_send_synack did. */
470 req->rsk_window_clamp = READ_ONCE(tp->window_clamp) ? :
471 dst_metric(dst: &rt->dst, RTAX_WINDOW);
472 /* limit the window selection if the user enforce a smaller rx buffer */
473 full_space = tcp_full_space(sk);
474 if (sk->sk_userlocks & SOCK_RCVBUF_LOCK &&
475 (req->rsk_window_clamp > full_space || req->rsk_window_clamp == 0))
476 req->rsk_window_clamp = full_space;
477
478 tcp_select_initial_window(sk, space: full_space, mss: req->mss,
479 rcv_wnd: &req->rsk_rcv_wnd, window_clamp: &req->rsk_window_clamp,
480 wscale_ok: ireq->wscale_ok, rcv_wscale: &rcv_wscale,
481 init_rcv_wnd: dst_metric(dst: &rt->dst, RTAX_INITRWND));
482
483 /* req->syncookie is set true only if ACK is validated
484 * by BPF kfunc, then, rcv_wscale is already configured.
485 */
486 if (!req->syncookie)
487 ireq->rcv_wscale = rcv_wscale;
488 ireq->ecn_ok &= cookie_ecn_ok(net, dst: &rt->dst);
489 treq->accecn_ok = ireq->ecn_ok && cookie_accecn_ok(th);
490
491 ret = tcp_get_cookie_sock(sk, skb, req, dst: &rt->dst);
492 /* ip_queue_xmit() depends on our flow being setup
493 * Normal sockets get it right from inet_csk_route_child_sock()
494 */
495 if (!ret) {
496 SKB_DR_SET(reason, NO_SOCKET);
497 goto out_drop;
498 }
499 inet_sk(ret)->cork.fl.u.ip4 = fl4;
500out:
501 return ret;
502out_free:
503 reqsk_free(req);
504out_drop:
505 sk_skb_reason_drop(sk, skb, reason);
506 return NULL;
507}
508