1/*
2 * Syncookies implementation for the Linux kernel
3 *
4 * Copyright (C) 1997 Andi Kleen
5 * Based on ideas by D.J.Bernstein and Eric Schenk.
6 *
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.
11 */
12
13#include <linux/tcp.h>
14#include <linux/slab.h>
15#include <linux/random.h>
16#include <linux/siphash.h>
17#include <linux/kernel.h>
18#include <linux/export.h>
19#include <net/secure_seq.h>
20#include <net/tcp.h>
21#include <net/route.h>
22
23static siphash_key_t syncookie_secret[2] __read_mostly;
24
25#define COOKIEBITS 24 /* Upper bits store count */
26#define COOKIEMASK (((__u32)1 << COOKIEBITS) - 1)
27
28/* TCP Timestamp: 6 lowest bits of timestamp sent in the cookie SYN-ACK
29 * stores TCP options:
30 *
31 * MSB LSB
32 * | 31 ... 6 | 5 | 4 | 3 2 1 0 |
33 * | Timestamp | ECN | SACK | WScale |
34 *
35 * When we receive a valid cookie-ACK, we look at the echoed tsval (if
36 * any) to figure out which TCP options we should use for the rebuilt
37 * connection.
38 *
39 * A WScale setting of '0xf' (which is an invalid scaling value)
40 * means that original syn did not include the TCP window scaling option.
41 */
42#define TS_OPT_WSCALE_MASK 0xf
43#define TS_OPT_SACK BIT(4)
44#define TS_OPT_ECN BIT(5)
45/* There is no TS_OPT_TIMESTAMP:
46 * if ACK contains timestamp option, we already know it was
47 * requested/supported by the syn/synack exchange.
48 */
49#define TSBITS 6
50#define TSMASK (((__u32)1 << TSBITS) - 1)
51
52static u32 cookie_hash(__be32 saddr, __be32 daddr, __be16 sport, __be16 dport,
53 u32 count, int c)
54{
55 net_get_random_once(syncookie_secret, sizeof(syncookie_secret));
56 return siphash_4u32((__force u32)saddr, (__force u32)daddr,
57 (__force u32)sport << 16 | (__force u32)dport,
58 count, &syncookie_secret[c]);
59}
60
61
62/*
63 * when syncookies are in effect and tcp timestamps are enabled we encode
64 * tcp options in the lower bits of the timestamp value that will be
65 * sent in the syn-ack.
66 * Since subsequent timestamps use the normal tcp_time_stamp value, we
67 * must make sure that the resulting initial timestamp is <= tcp_time_stamp.
68 */
69u64 cookie_init_timestamp(struct request_sock *req)
70{
71 struct inet_request_sock *ireq;
72 u32 ts, ts_now = tcp_time_stamp_raw();
73 u32 options = 0;
74
75 ireq = inet_rsk(req);
76
77 options = ireq->wscale_ok ? ireq->snd_wscale : TS_OPT_WSCALE_MASK;
78 if (ireq->sack_ok)
79 options |= TS_OPT_SACK;
80 if (ireq->ecn_ok)
81 options |= TS_OPT_ECN;
82
83 ts = ts_now & ~TSMASK;
84 ts |= options;
85 if (ts > ts_now) {
86 ts >>= TSBITS;
87 ts--;
88 ts <<= TSBITS;
89 ts |= options;
90 }
91 return (u64)ts * (NSEC_PER_SEC / TCP_TS_HZ);
92}
93
94
95static __u32 secure_tcp_syn_cookie(__be32 saddr, __be32 daddr, __be16 sport,
96 __be16 dport, __u32 sseq, __u32 data)
97{
98 /*
99 * Compute the secure sequence number.
100 * The output should be:
101 * HASH(sec1,saddr,sport,daddr,dport,sec1) + sseq + (count * 2^24)
102 * + (HASH(sec2,saddr,sport,daddr,dport,count,sec2) % 2^24).
103 * Where sseq is their sequence number and count increases every
104 * minute by 1.
105 * As an extra hack, we add a small "data" value that encodes the
106 * MSS into the second hash value.
107 */
108 u32 count = tcp_cookie_time();
109 return (cookie_hash(saddr, daddr, sport, dport, 0, 0) +
110 sseq + (count << COOKIEBITS) +
111 ((cookie_hash(saddr, daddr, sport, dport, count, 1) + data)
112 & COOKIEMASK));
113}
114
115/*
116 * This retrieves the small "data" value from the syncookie.
117 * If the syncookie is bad, the data returned will be out of
118 * range. This must be checked by the caller.
119 *
120 * The count value used to generate the cookie must be less than
121 * MAX_SYNCOOKIE_AGE minutes in the past.
122 * The return value (__u32)-1 if this test fails.
123 */
124static __u32 check_tcp_syn_cookie(__u32 cookie, __be32 saddr, __be32 daddr,
125 __be16 sport, __be16 dport, __u32 sseq)
126{
127 u32 diff, count = tcp_cookie_time();
128
129 /* Strip away the layers from the cookie */
130 cookie -= cookie_hash(saddr, daddr, sport, dport, 0, 0) + sseq;
131
132 /* Cookie is now reduced to (count * 2^24) ^ (hash % 2^24) */
133 diff = (count - (cookie >> COOKIEBITS)) & ((__u32) -1 >> COOKIEBITS);
134 if (diff >= MAX_SYNCOOKIE_AGE)
135 return (__u32)-1;
136
137 return (cookie -
138 cookie_hash(saddr, daddr, sport, dport, count - diff, 1))
139 & COOKIEMASK; /* Leaving the data behind */
140}
141
142/*
143 * MSS Values are chosen based on the 2011 paper
144 * 'An Analysis of TCP Maximum Segement Sizes' by S. Alcock and R. Nelson.
145 * Values ..
146 * .. lower than 536 are rare (< 0.2%)
147 * .. between 537 and 1299 account for less than < 1.5% of observed values
148 * .. in the 1300-1349 range account for about 15 to 20% of observed mss values
149 * .. exceeding 1460 are very rare (< 0.04%)
150 *
151 * 1460 is the single most frequently announced mss value (30 to 46% depending
152 * on monitor location). Table must be sorted.
153 */
154static __u16 const msstab[] = {
155 536,
156 1300,
157 1440, /* 1440, 1452: PPPoE */
158 1460,
159};
160
161/*
162 * Generate a syncookie. mssp points to the mss, which is returned
163 * rounded down to the value encoded in the cookie.
164 */
165u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
166 u16 *mssp)
167{
168 int mssind;
169 const __u16 mss = *mssp;
170
171 for (mssind = ARRAY_SIZE(msstab) - 1; mssind ; mssind--)
172 if (mss >= msstab[mssind])
173 break;
174 *mssp = msstab[mssind];
175
176 return secure_tcp_syn_cookie(iph->saddr, iph->daddr,
177 th->source, th->dest, ntohl(th->seq),
178 mssind);
179}
180EXPORT_SYMBOL_GPL(__cookie_v4_init_sequence);
181
182__u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mssp)
183{
184 const struct iphdr *iph = ip_hdr(skb);
185 const struct tcphdr *th = tcp_hdr(skb);
186
187 return __cookie_v4_init_sequence(iph, th, mssp);
188}
189
190/*
191 * Check if a ack sequence number is a valid syncookie.
192 * Return the decoded mss if it is, or 0 if not.
193 */
194int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
195 u32 cookie)
196{
197 __u32 seq = ntohl(th->seq) - 1;
198 __u32 mssind = check_tcp_syn_cookie(cookie, iph->saddr, iph->daddr,
199 th->source, th->dest, seq);
200
201 return mssind < ARRAY_SIZE(msstab) ? msstab[mssind] : 0;
202}
203EXPORT_SYMBOL_GPL(__cookie_v4_check);
204
205struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
206 struct request_sock *req,
207 struct dst_entry *dst, u32 tsoff)
208{
209 struct inet_connection_sock *icsk = inet_csk(sk);
210 struct sock *child;
211 bool own_req;
212
213 child = icsk->icsk_af_ops->syn_recv_sock(sk, skb, req, dst,
214 NULL, &own_req);
215 if (child) {
216 refcount_set(&req->rsk_refcnt, 1);
217 tcp_sk(child)->tsoffset = tsoff;
218 sock_rps_save_rxhash(child, skb);
219 if (!inet_csk_reqsk_queue_add(sk, req, child)) {
220 bh_unlock_sock(child);
221 sock_put(child);
222 child = NULL;
223 reqsk_put(req);
224 }
225 } else {
226 reqsk_free(req);
227 }
228 return child;
229}
230EXPORT_SYMBOL(tcp_get_cookie_sock);
231
232/*
233 * when syncookies are in effect and tcp timestamps are enabled we stored
234 * additional tcp options in the timestamp.
235 * This extracts these options from the timestamp echo.
236 *
237 * return false if we decode a tcp option that is disabled
238 * on the host.
239 */
240bool cookie_timestamp_decode(const struct net *net,
241 struct tcp_options_received *tcp_opt)
242{
243 /* echoed timestamp, lowest bits contain options */
244 u32 options = tcp_opt->rcv_tsecr;
245
246 if (!tcp_opt->saw_tstamp) {
247 tcp_clear_options(tcp_opt);
248 return true;
249 }
250
251 if (!net->ipv4.sysctl_tcp_timestamps)
252 return false;
253
254 tcp_opt->sack_ok = (options & TS_OPT_SACK) ? TCP_SACK_SEEN : 0;
255
256 if (tcp_opt->sack_ok && !net->ipv4.sysctl_tcp_sack)
257 return false;
258
259 if ((options & TS_OPT_WSCALE_MASK) == TS_OPT_WSCALE_MASK)
260 return true; /* no window scaling */
261
262 tcp_opt->wscale_ok = 1;
263 tcp_opt->snd_wscale = options & TS_OPT_WSCALE_MASK;
264
265 return net->ipv4.sysctl_tcp_window_scaling != 0;
266}
267EXPORT_SYMBOL(cookie_timestamp_decode);
268
269bool cookie_ecn_ok(const struct tcp_options_received *tcp_opt,
270 const struct net *net, const struct dst_entry *dst)
271{
272 bool ecn_ok = tcp_opt->rcv_tsecr & TS_OPT_ECN;
273
274 if (!ecn_ok)
275 return false;
276
277 if (net->ipv4.sysctl_tcp_ecn)
278 return true;
279
280 return dst_feature(dst, RTAX_FEATURE_ECN);
281}
282EXPORT_SYMBOL(cookie_ecn_ok);
283
284/* On input, sk is a listener.
285 * Output is listener if incoming packet would not create a child
286 * NULL if memory could not be allocated.
287 */
288struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb)
289{
290 struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
291 struct tcp_options_received tcp_opt;
292 struct inet_request_sock *ireq;
293 struct tcp_request_sock *treq;
294 struct tcp_sock *tp = tcp_sk(sk);
295 const struct tcphdr *th = tcp_hdr(skb);
296 __u32 cookie = ntohl(th->ack_seq) - 1;
297 struct sock *ret = sk;
298 struct request_sock *req;
299 int mss;
300 struct rtable *rt;
301 __u8 rcv_wscale;
302 struct flowi4 fl4;
303 u32 tsoff = 0;
304
305 if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies || !th->ack || th->rst)
306 goto out;
307
308 if (tcp_synq_no_recent_overflow(sk))
309 goto out;
310
311 mss = __cookie_v4_check(ip_hdr(skb), th, cookie);
312 if (mss == 0) {
313 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESFAILED);
314 goto out;
315 }
316
317 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESRECV);
318
319 /* check for timestamp cookie support */
320 memset(&tcp_opt, 0, sizeof(tcp_opt));
321 tcp_parse_options(sock_net(sk), skb, &tcp_opt, 0, NULL);
322
323 if (tcp_opt.saw_tstamp && tcp_opt.rcv_tsecr) {
324 tsoff = secure_tcp_ts_off(sock_net(sk),
325 ip_hdr(skb)->daddr,
326 ip_hdr(skb)->saddr);
327 tcp_opt.rcv_tsecr -= tsoff;
328 }
329
330 if (!cookie_timestamp_decode(sock_net(sk), &tcp_opt))
331 goto out;
332
333 ret = NULL;
334 req = inet_reqsk_alloc(&tcp_request_sock_ops, sk, false); /* for safety */
335 if (!req)
336 goto out;
337
338 ireq = inet_rsk(req);
339 treq = tcp_rsk(req);
340 treq->rcv_isn = ntohl(th->seq) - 1;
341 treq->snt_isn = cookie;
342 treq->ts_off = 0;
343 treq->txhash = net_tx_rndhash();
344 req->mss = mss;
345 ireq->ir_num = ntohs(th->dest);
346 ireq->ir_rmt_port = th->source;
347 sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr);
348 sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr);
349 ireq->ir_mark = inet_request_mark(sk, skb);
350 ireq->snd_wscale = tcp_opt.snd_wscale;
351 ireq->sack_ok = tcp_opt.sack_ok;
352 ireq->wscale_ok = tcp_opt.wscale_ok;
353 ireq->tstamp_ok = tcp_opt.saw_tstamp;
354 req->ts_recent = tcp_opt.saw_tstamp ? tcp_opt.rcv_tsval : 0;
355 treq->snt_synack = 0;
356 treq->tfo_listener = false;
357 if (IS_ENABLED(CONFIG_SMC))
358 ireq->smc_ok = 0;
359
360 ireq->ir_iif = inet_request_bound_dev_if(sk, skb);
361
362 /* We throwed the options of the initial SYN away, so we hope
363 * the ACK carries the same options again (see RFC1122 4.2.3.8)
364 */
365 RCU_INIT_POINTER(ireq->ireq_opt, tcp_v4_save_options(sock_net(sk), skb));
366
367 if (security_inet_conn_request(sk, skb, req)) {
368 reqsk_free(req);
369 goto out;
370 }
371
372 req->num_retrans = 0;
373
374 /*
375 * We need to lookup the route here to get at the correct
376 * window size. We should better make sure that the window size
377 * hasn't changed since we received the original syn, but I see
378 * no easy way to do this.
379 */
380 flowi4_init_output(&fl4, ireq->ir_iif, ireq->ir_mark,
381 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE, IPPROTO_TCP,
382 inet_sk_flowi_flags(sk),
383 opt->srr ? opt->faddr : ireq->ir_rmt_addr,
384 ireq->ir_loc_addr, th->source, th->dest, sk->sk_uid);
385 security_req_classify_flow(req, flowi4_to_flowi(&fl4));
386 rt = ip_route_output_key(sock_net(sk), &fl4);
387 if (IS_ERR(rt)) {
388 reqsk_free(req);
389 goto out;
390 }
391
392 /* Try to redo what tcp_v4_send_synack did. */
393 req->rsk_window_clamp = tp->window_clamp ? :dst_metric(&rt->dst, RTAX_WINDOW);
394
395 tcp_select_initial_window(sk, tcp_full_space(sk), req->mss,
396 &req->rsk_rcv_wnd, &req->rsk_window_clamp,
397 ireq->wscale_ok, &rcv_wscale,
398 dst_metric(&rt->dst, RTAX_INITRWND));
399
400 ireq->rcv_wscale = rcv_wscale;
401 ireq->ecn_ok = cookie_ecn_ok(&tcp_opt, sock_net(sk), &rt->dst);
402
403 ret = tcp_get_cookie_sock(sk, skb, req, &rt->dst, tsoff);
404 /* ip_queue_xmit() depends on our flow being setup
405 * Normal sockets get it right from inet_csk_route_child_sock()
406 */
407 if (ret)
408 inet_sk(ret)->cork.fl.u.ip4 = fl4;
409out: return ret;
410}
411