Loading...
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/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>
16#include <net/tcp.h>
17#include <net/route.h>
18
19static siphash_key_t syncookie_secret[2] __read_mostly;
20
21#define COOKIEBITS 24 /* Upper bits store count */
22#define COOKIEMASK (((__u32)1 << COOKIEBITS) - 1)
23
24/* TCP Timestamp: 6 lowest bits of timestamp sent in the cookie SYN-ACK
25 * stores TCP options:
26 *
27 * MSB LSB
28 * | 31 ... 6 | 5 | 4 | 3 2 1 0 |
29 * | Timestamp | ECN | SACK | WScale |
30 *
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
33 * connection.
34 *
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.
37 */
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.
44 */
45#define TSBITS 6
46#define TSMASK (((__u32)1 << TSBITS) - 1)
47
48static u32 cookie_hash(__be32 saddr, __be32 daddr, __be16 sport, __be16 dport,
49 u32 count, int c)
50{
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]);
55}
56
57
58/*
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.
64 */
65u64 cookie_init_timestamp(struct request_sock *req, u64 now)
66{
67 struct inet_request_sock *ireq;
68 u32 ts, ts_now = tcp_ns_to_ts(now);
69 u32 options = 0;
70
71 ireq = inet_rsk(req);
72
73 options = ireq->wscale_ok ? ireq->snd_wscale : TS_OPT_WSCALE_MASK;
74 if (ireq->sack_ok)
75 options |= TS_OPT_SACK;
76 if (ireq->ecn_ok)
77 options |= TS_OPT_ECN;
78
79 ts = ts_now & ~TSMASK;
80 ts |= options;
81 if (ts > ts_now) {
82 ts >>= TSBITS;
83 ts--;
84 ts <<= TSBITS;
85 ts |= options;
86 }
87 return (u64)ts * (NSEC_PER_SEC / TCP_TS_HZ);
88}
89
90
91static __u32 secure_tcp_syn_cookie(__be32 saddr, __be32 daddr, __be16 sport,
92 __be16 dport, __u32 sseq, __u32 data)
93{
94 /*
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
100 * minute by 1.
101 * As an extra hack, we add a small "data" value that encodes the
102 * MSS into the second hash value.
103 */
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)
108 & COOKIEMASK));
109}
110
111/*
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.
115 *
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.
119 */
120static __u32 check_tcp_syn_cookie(__u32 cookie, __be32 saddr, __be32 daddr,
121 __be16 sport, __be16 dport, __u32 sseq)
122{
123 u32 diff, count = tcp_cookie_time();
124
125 /* Strip away the layers from the cookie */
126 cookie -= cookie_hash(saddr, daddr, sport, dport, 0, 0) + sseq;
127
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)
131 return (__u32)-1;
132
133 return (cookie -
134 cookie_hash(saddr, daddr, sport, dport, count - diff, 1))
135 & COOKIEMASK; /* Leaving the data behind */
136}
137
138/*
139 * MSS Values are chosen based on the 2011 paper
140 * 'An Analysis of TCP Maximum Segement Sizes' by S. Alcock and R. Nelson.
141 * Values ..
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%)
146 *
147 * 1460 is the single most frequently announced mss value (30 to 46% depending
148 * on monitor location). Table must be sorted.
149 */
150static __u16 const msstab[] = {
151 536,
152 1300,
153 1440, /* 1440, 1452: PPPoE */
154 1460,
155};
156
157/*
158 * Generate a syncookie. mssp points to the mss, which is returned
159 * rounded down to the value encoded in the cookie.
160 */
161u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
162 u16 *mssp)
163{
164 int mssind;
165 const __u16 mss = *mssp;
166
167 for (mssind = ARRAY_SIZE(msstab) - 1; mssind ; mssind--)
168 if (mss >= msstab[mssind])
169 break;
170 *mssp = msstab[mssind];
171
172 return secure_tcp_syn_cookie(iph->saddr, iph->daddr,
173 th->source, th->dest, ntohl(th->seq),
174 mssind);
175}
176EXPORT_SYMBOL_GPL(__cookie_v4_init_sequence);
177
178__u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mssp)
179{
180 const struct iphdr *iph = ip_hdr(skb);
181 const struct tcphdr *th = tcp_hdr(skb);
182
183 return __cookie_v4_init_sequence(iph, th, mssp);
184}
185
186/*
187 * Check if a ack sequence number is a valid syncookie.
188 * Return the decoded mss if it is, or 0 if not.
189 */
190int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
191 u32 cookie)
192{
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);
196
197 return mssind < ARRAY_SIZE(msstab) ? msstab[mssind] : 0;
198}
199EXPORT_SYMBOL_GPL(__cookie_v4_check);
200
201struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
202 struct request_sock *req,
203 struct dst_entry *dst, u32 tsoff)
204{
205 struct inet_connection_sock *icsk = inet_csk(sk);
206 struct sock *child;
207 bool own_req;
208
209 child = icsk->icsk_af_ops->syn_recv_sock(sk, skb, req, dst,
210 NULL, &own_req);
211 if (child) {
212 refcount_set(&req->rsk_refcnt, 1);
213 tcp_sk(child)->tsoffset = tsoff;
214 sock_rps_save_rxhash(child, skb);
215
216 if (rsk_drop_req(req)) {
217 reqsk_put(req);
218 return child;
219 }
220
221 if (inet_csk_reqsk_queue_add(sk, req, child))
222 return child;
223
224 bh_unlock_sock(child);
225 sock_put(child);
226 }
227 __reqsk_free(req);
228
229 return NULL;
230}
231EXPORT_SYMBOL(tcp_get_cookie_sock);
232
233/*
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.
237 *
238 * return false if we decode a tcp option that is disabled
239 * on the host.
240 */
241bool cookie_timestamp_decode(const struct net *net,
242 struct tcp_options_received *tcp_opt)
243{
244 /* echoed timestamp, lowest bits contain options */
245 u32 options = tcp_opt->rcv_tsecr;
246
247 if (!tcp_opt->saw_tstamp) {
248 tcp_clear_options(tcp_opt);
249 return true;
250 }
251
252 if (!net->ipv4.sysctl_tcp_timestamps)
253 return false;
254
255 tcp_opt->sack_ok = (options & TS_OPT_SACK) ? TCP_SACK_SEEN : 0;
256
257 if (tcp_opt->sack_ok && !net->ipv4.sysctl_tcp_sack)
258 return false;
259
260 if ((options & TS_OPT_WSCALE_MASK) == TS_OPT_WSCALE_MASK)
261 return true; /* no window scaling */
262
263 tcp_opt->wscale_ok = 1;
264 tcp_opt->snd_wscale = options & TS_OPT_WSCALE_MASK;
265
266 return net->ipv4.sysctl_tcp_window_scaling != 0;
267}
268EXPORT_SYMBOL(cookie_timestamp_decode);
269
270bool cookie_ecn_ok(const struct tcp_options_received *tcp_opt,
271 const struct net *net, const struct dst_entry *dst)
272{
273 bool ecn_ok = tcp_opt->rcv_tsecr & TS_OPT_ECN;
274
275 if (!ecn_ok)
276 return false;
277
278 if (net->ipv4.sysctl_tcp_ecn)
279 return true;
280
281 return dst_feature(dst, RTAX_FEATURE_ECN);
282}
283EXPORT_SYMBOL(cookie_ecn_ok);
284
285struct request_sock *cookie_tcp_reqsk_alloc(const struct request_sock_ops *ops,
286 struct sock *sk,
287 struct sk_buff *skb)
288{
289 struct request_sock *req;
290
291#ifdef CONFIG_MPTCP
292 struct tcp_request_sock *treq;
293
294 if (sk_is_mptcp(sk))
295 ops = &mptcp_subflow_request_sock_ops;
296#endif
297
298 req = inet_reqsk_alloc(ops, sk, false);
299 if (!req)
300 return NULL;
301
302#if IS_ENABLED(CONFIG_MPTCP)
303 treq = tcp_rsk(req);
304 treq->is_mptcp = sk_is_mptcp(sk);
305 if (treq->is_mptcp) {
306 int err = mptcp_subflow_init_cookie_req(req, sk, skb);
307
308 if (err) {
309 reqsk_free(req);
310 return NULL;
311 }
312 }
313#endif
314
315 return req;
316}
317EXPORT_SYMBOL_GPL(cookie_tcp_reqsk_alloc);
318
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.
322 */
323struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb)
324{
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;
334 int mss;
335 struct rtable *rt;
336 __u8 rcv_wscale;
337 struct flowi4 fl4;
338 u32 tsoff = 0;
339
340 if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies || !th->ack || th->rst)
341 goto out;
342
343 if (tcp_synq_no_recent_overflow(sk))
344 goto out;
345
346 mss = __cookie_v4_check(ip_hdr(skb), th, cookie);
347 if (mss == 0) {
348 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESFAILED);
349 goto out;
350 }
351
352 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESRECV);
353
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);
357
358 if (tcp_opt.saw_tstamp && tcp_opt.rcv_tsecr) {
359 tsoff = secure_tcp_ts_off(sock_net(sk),
360 ip_hdr(skb)->daddr,
361 ip_hdr(skb)->saddr);
362 tcp_opt.rcv_tsecr -= tsoff;
363 }
364
365 if (!cookie_timestamp_decode(sock_net(sk), &tcp_opt))
366 goto out;
367
368 ret = NULL;
369 req = cookie_tcp_reqsk_alloc(&tcp_request_sock_ops, sk, skb);
370 if (!req)
371 goto out;
372
373 ireq = inet_rsk(req);
374 treq = tcp_rsk(req);
375 treq->rcv_isn = ntohl(th->seq) - 1;
376 treq->snt_isn = cookie;
377 treq->ts_off = 0;
378 treq->txhash = net_tx_rndhash();
379 req->mss = mss;
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;
392
393 if (IS_ENABLED(CONFIG_SMC))
394 ireq->smc_ok = 0;
395
396 ireq->ir_iif = inet_request_bound_dev_if(sk, skb);
397
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)
400 */
401 RCU_INIT_POINTER(ireq->ireq_opt, tcp_v4_save_options(sock_net(sk), skb));
402
403 if (security_inet_conn_request(sk, skb, req)) {
404 reqsk_free(req);
405 goto out;
406 }
407
408 req->num_retrans = 0;
409
410 /*
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.
415 */
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(&fl4));
422 rt = ip_route_output_key(sock_net(sk), &fl4);
423 if (IS_ERR(rt)) {
424 reqsk_free(req);
425 goto out;
426 }
427
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
431 tcp_select_initial_window(sk, tcp_full_space(sk), req->mss,
432 &req->rsk_rcv_wnd, &req->rsk_window_clamp,
433 ireq->wscale_ok, &rcv_wscale,
434 dst_metric(&rt->dst, RTAX_INITRWND));
435
436 ireq->rcv_wscale = rcv_wscale;
437 ireq->ecn_ok = cookie_ecn_ok(&tcp_opt, sock_net(sk), &rt->dst);
438
439 ret = tcp_get_cookie_sock(sk, skb, req, &rt->dst, tsoff);
440 /* ip_queue_xmit() depends on our flow being setup
441 * Normal sockets get it right from inet_csk_route_child_sock()
442 */
443 if (ret)
444 inet_sk(ret)->cork.fl.u.ip4 = fl4;
445out: return ret;
446}
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/route.h>
16
17static siphash_aligned_key_t syncookie_secret[2];
18
19#define COOKIEBITS 24 /* Upper bits store count */
20#define COOKIEMASK (((__u32)1 << COOKIEBITS) - 1)
21
22/* TCP Timestamp: 6 lowest bits of timestamp sent in the cookie SYN-ACK
23 * stores TCP options:
24 *
25 * MSB LSB
26 * | 31 ... 6 | 5 | 4 | 3 2 1 0 |
27 * | Timestamp | ECN | SACK | WScale |
28 *
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
31 * connection.
32 *
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.
35 */
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.
42 */
43#define TSBITS 6
44#define TSMASK (((__u32)1 << TSBITS) - 1)
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((__force u32)saddr, (__force u32)daddr,
51 (__force u32)sport << 16 | (__force u32)dport,
52 count, &syncookie_secret[c]);
53}
54
55
56/*
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.
62 */
63u64 cookie_init_timestamp(struct request_sock *req, u64 now)
64{
65 struct inet_request_sock *ireq;
66 u32 ts, ts_now = tcp_ns_to_ts(now);
67 u32 options = 0;
68
69 ireq = inet_rsk(req);
70
71 options = ireq->wscale_ok ? ireq->snd_wscale : TS_OPT_WSCALE_MASK;
72 if (ireq->sack_ok)
73 options |= TS_OPT_SACK;
74 if (ireq->ecn_ok)
75 options |= TS_OPT_ECN;
76
77 ts = ts_now & ~TSMASK;
78 ts |= options;
79 if (ts > ts_now) {
80 ts >>= TSBITS;
81 ts--;
82 ts <<= TSBITS;
83 ts |= options;
84 }
85 return (u64)ts * (NSEC_PER_SEC / TCP_TS_HZ);
86}
87
88
89static __u32 secure_tcp_syn_cookie(__be32 saddr, __be32 daddr, __be16 sport,
90 __be16 dport, __u32 sseq, __u32 data)
91{
92 /*
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
98 * minute by 1.
99 * As an extra hack, we add a small "data" value that encodes the
100 * MSS into the second hash value.
101 */
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)
106 & COOKIEMASK));
107}
108
109/*
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.
113 *
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.
117 */
118static __u32 check_tcp_syn_cookie(__u32 cookie, __be32 saddr, __be32 daddr,
119 __be16 sport, __be16 dport, __u32 sseq)
120{
121 u32 diff, count = tcp_cookie_time();
122
123 /* Strip away the layers from the cookie */
124 cookie -= cookie_hash(saddr, daddr, sport, dport, 0, 0) + sseq;
125
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)
129 return (__u32)-1;
130
131 return (cookie -
132 cookie_hash(saddr, daddr, sport, dport, count - diff, 1))
133 & COOKIEMASK; /* Leaving the data behind */
134}
135
136/*
137 * MSS Values are chosen based on the 2011 paper
138 * 'An Analysis of TCP Maximum Segement Sizes' by S. Alcock and R. Nelson.
139 * Values ..
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%)
144 *
145 * 1460 is the single most frequently announced mss value (30 to 46% depending
146 * on monitor location). Table must be sorted.
147 */
148static __u16 const msstab[] = {
149 536,
150 1300,
151 1440, /* 1440, 1452: PPPoE */
152 1460,
153};
154
155/*
156 * Generate a syncookie. mssp points to the mss, which is returned
157 * rounded down to the value encoded in the cookie.
158 */
159u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
160 u16 *mssp)
161{
162 int mssind;
163 const __u16 mss = *mssp;
164
165 for (mssind = ARRAY_SIZE(msstab) - 1; mssind ; mssind--)
166 if (mss >= msstab[mssind])
167 break;
168 *mssp = msstab[mssind];
169
170 return secure_tcp_syn_cookie(iph->saddr, iph->daddr,
171 th->source, th->dest, ntohl(th->seq),
172 mssind);
173}
174EXPORT_SYMBOL_GPL(__cookie_v4_init_sequence);
175
176__u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mssp)
177{
178 const struct iphdr *iph = ip_hdr(skb);
179 const struct tcphdr *th = tcp_hdr(skb);
180
181 return __cookie_v4_init_sequence(iph, th, mssp);
182}
183
184/*
185 * Check if a ack sequence number is a valid syncookie.
186 * Return the decoded mss if it is, or 0 if not.
187 */
188int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
189 u32 cookie)
190{
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);
194
195 return mssind < ARRAY_SIZE(msstab) ? msstab[mssind] : 0;
196}
197EXPORT_SYMBOL_GPL(__cookie_v4_check);
198
199struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
200 struct request_sock *req,
201 struct dst_entry *dst, u32 tsoff)
202{
203 struct inet_connection_sock *icsk = inet_csk(sk);
204 struct sock *child;
205 bool own_req;
206
207 child = icsk->icsk_af_ops->syn_recv_sock(sk, skb, req, dst,
208 NULL, &own_req);
209 if (child) {
210 refcount_set(&req->rsk_refcnt, 1);
211 tcp_sk(child)->tsoffset = tsoff;
212 sock_rps_save_rxhash(child, skb);
213
214 if (rsk_drop_req(req)) {
215 reqsk_put(req);
216 return child;
217 }
218
219 if (inet_csk_reqsk_queue_add(sk, req, child))
220 return child;
221
222 bh_unlock_sock(child);
223 sock_put(child);
224 }
225 __reqsk_free(req);
226
227 return NULL;
228}
229EXPORT_SYMBOL(tcp_get_cookie_sock);
230
231/*
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.
235 *
236 * return false if we decode a tcp option that is disabled
237 * on the host.
238 */
239bool cookie_timestamp_decode(const struct net *net,
240 struct tcp_options_received *tcp_opt)
241{
242 /* echoed timestamp, lowest bits contain options */
243 u32 options = tcp_opt->rcv_tsecr;
244
245 if (!tcp_opt->saw_tstamp) {
246 tcp_clear_options(tcp_opt);
247 return true;
248 }
249
250 if (!READ_ONCE(net->ipv4.sysctl_tcp_timestamps))
251 return false;
252
253 tcp_opt->sack_ok = (options & TS_OPT_SACK) ? TCP_SACK_SEEN : 0;
254
255 if (tcp_opt->sack_ok && !READ_ONCE(net->ipv4.sysctl_tcp_sack))
256 return false;
257
258 if ((options & TS_OPT_WSCALE_MASK) == TS_OPT_WSCALE_MASK)
259 return true; /* no window scaling */
260
261 tcp_opt->wscale_ok = 1;
262 tcp_opt->snd_wscale = options & TS_OPT_WSCALE_MASK;
263
264 return READ_ONCE(net->ipv4.sysctl_tcp_window_scaling) != 0;
265}
266EXPORT_SYMBOL(cookie_timestamp_decode);
267
268bool cookie_ecn_ok(const struct tcp_options_received *tcp_opt,
269 const struct net *net, const struct dst_entry *dst)
270{
271 bool ecn_ok = tcp_opt->rcv_tsecr & TS_OPT_ECN;
272
273 if (!ecn_ok)
274 return false;
275
276 if (READ_ONCE(net->ipv4.sysctl_tcp_ecn))
277 return true;
278
279 return dst_feature(dst, RTAX_FEATURE_ECN);
280}
281EXPORT_SYMBOL(cookie_ecn_ok);
282
283struct request_sock *cookie_tcp_reqsk_alloc(const struct request_sock_ops *ops,
284 const struct tcp_request_sock_ops *af_ops,
285 struct sock *sk,
286 struct sk_buff *skb)
287{
288 struct tcp_request_sock *treq;
289 struct request_sock *req;
290
291 if (sk_is_mptcp(sk))
292 req = mptcp_subflow_reqsk_alloc(ops, sk, false);
293 else
294 req = inet_reqsk_alloc(ops, sk, false);
295
296 if (!req)
297 return NULL;
298
299 treq = tcp_rsk(req);
300
301 /* treq->af_specific might be used to perform TCP_MD5 lookup */
302 treq->af_specific = af_ops;
303
304 treq->syn_tos = TCP_SKB_CB(skb)->ip_dsfield;
305#if IS_ENABLED(CONFIG_MPTCP)
306 treq->is_mptcp = sk_is_mptcp(sk);
307 if (treq->is_mptcp) {
308 int err = mptcp_subflow_init_cookie_req(req, sk, skb);
309
310 if (err) {
311 reqsk_free(req);
312 return NULL;
313 }
314 }
315#endif
316
317 return req;
318}
319EXPORT_SYMBOL_GPL(cookie_tcp_reqsk_alloc);
320
321/* On input, sk is a listener.
322 * Output is listener if incoming packet would not create a child
323 * NULL if memory could not be allocated.
324 */
325struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb)
326{
327 struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
328 struct tcp_options_received tcp_opt;
329 struct inet_request_sock *ireq;
330 struct tcp_request_sock *treq;
331 struct tcp_sock *tp = tcp_sk(sk);
332 const struct tcphdr *th = tcp_hdr(skb);
333 __u32 cookie = ntohl(th->ack_seq) - 1;
334 struct sock *ret = sk;
335 struct request_sock *req;
336 int full_space, mss;
337 struct rtable *rt;
338 __u8 rcv_wscale;
339 struct flowi4 fl4;
340 u32 tsoff = 0;
341
342 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies) ||
343 !th->ack || th->rst)
344 goto out;
345
346 if (tcp_synq_no_recent_overflow(sk))
347 goto out;
348
349 mss = __cookie_v4_check(ip_hdr(skb), th, cookie);
350 if (mss == 0) {
351 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESFAILED);
352 goto out;
353 }
354
355 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESRECV);
356
357 /* check for timestamp cookie support */
358 memset(&tcp_opt, 0, sizeof(tcp_opt));
359 tcp_parse_options(sock_net(sk), skb, &tcp_opt, 0, NULL);
360
361 if (tcp_opt.saw_tstamp && tcp_opt.rcv_tsecr) {
362 tsoff = secure_tcp_ts_off(sock_net(sk),
363 ip_hdr(skb)->daddr,
364 ip_hdr(skb)->saddr);
365 tcp_opt.rcv_tsecr -= tsoff;
366 }
367
368 if (!cookie_timestamp_decode(sock_net(sk), &tcp_opt))
369 goto out;
370
371 ret = NULL;
372 req = cookie_tcp_reqsk_alloc(&tcp_request_sock_ops,
373 &tcp_request_sock_ipv4_ops, sk, skb);
374 if (!req)
375 goto out;
376
377 ireq = inet_rsk(req);
378 treq = tcp_rsk(req);
379 treq->rcv_isn = ntohl(th->seq) - 1;
380 treq->snt_isn = cookie;
381 treq->ts_off = 0;
382 treq->txhash = net_tx_rndhash();
383 req->mss = mss;
384 ireq->ir_num = ntohs(th->dest);
385 ireq->ir_rmt_port = th->source;
386 sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr);
387 sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr);
388 ireq->ir_mark = inet_request_mark(sk, skb);
389 ireq->snd_wscale = tcp_opt.snd_wscale;
390 ireq->sack_ok = tcp_opt.sack_ok;
391 ireq->wscale_ok = tcp_opt.wscale_ok;
392 ireq->tstamp_ok = tcp_opt.saw_tstamp;
393 req->ts_recent = tcp_opt.saw_tstamp ? tcp_opt.rcv_tsval : 0;
394 treq->snt_synack = 0;
395 treq->tfo_listener = false;
396
397 if (IS_ENABLED(CONFIG_SMC))
398 ireq->smc_ok = 0;
399
400 ireq->ir_iif = inet_request_bound_dev_if(sk, skb);
401
402 /* We throwed the options of the initial SYN away, so we hope
403 * the ACK carries the same options again (see RFC1122 4.2.3.8)
404 */
405 RCU_INIT_POINTER(ireq->ireq_opt, tcp_v4_save_options(sock_net(sk), skb));
406
407 if (security_inet_conn_request(sk, skb, req)) {
408 reqsk_free(req);
409 goto out;
410 }
411
412 req->num_retrans = 0;
413
414 /*
415 * We need to lookup the route here to get at the correct
416 * window size. We should better make sure that the window size
417 * hasn't changed since we received the original syn, but I see
418 * no easy way to do this.
419 */
420 flowi4_init_output(&fl4, ireq->ir_iif, ireq->ir_mark,
421 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE, IPPROTO_TCP,
422 inet_sk_flowi_flags(sk),
423 opt->srr ? opt->faddr : ireq->ir_rmt_addr,
424 ireq->ir_loc_addr, th->source, th->dest, sk->sk_uid);
425 security_req_classify_flow(req, flowi4_to_flowi_common(&fl4));
426 rt = ip_route_output_key(sock_net(sk), &fl4);
427 if (IS_ERR(rt)) {
428 reqsk_free(req);
429 goto out;
430 }
431
432 /* Try to redo what tcp_v4_send_synack did. */
433 req->rsk_window_clamp = tp->window_clamp ? :dst_metric(&rt->dst, RTAX_WINDOW);
434 /* limit the window selection if the user enforce a smaller rx buffer */
435 full_space = tcp_full_space(sk);
436 if (sk->sk_userlocks & SOCK_RCVBUF_LOCK &&
437 (req->rsk_window_clamp > full_space || req->rsk_window_clamp == 0))
438 req->rsk_window_clamp = full_space;
439
440 tcp_select_initial_window(sk, full_space, req->mss,
441 &req->rsk_rcv_wnd, &req->rsk_window_clamp,
442 ireq->wscale_ok, &rcv_wscale,
443 dst_metric(&rt->dst, RTAX_INITRWND));
444
445 ireq->rcv_wscale = rcv_wscale;
446 ireq->ecn_ok = cookie_ecn_ok(&tcp_opt, sock_net(sk), &rt->dst);
447
448 ret = tcp_get_cookie_sock(sk, skb, req, &rt->dst, tsoff);
449 /* ip_queue_xmit() depends on our flow being setup
450 * Normal sockets get it right from inet_csk_route_child_sock()
451 */
452 if (ret)
453 inet_sk(ret)->cork.fl.u.ip4 = fl4;
454out: return ret;
455}