blob: f0794f0232bae749244fff35d8b96b1f561a5e87 [file] [log] [blame]
Thomas Gleixner2874c5f2019-05-27 08:55:01 +02001// SPDX-License-Identifier: GPL-2.0-or-later
Linus Torvalds1da177e2005-04-16 15:20:36 -07002/*
3 * Syncookies implementation for the Linux kernel
4 *
5 * Copyright (C) 1997 Andi Kleen
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09006 * Based on ideas by D.J.Bernstein and Eric Schenk.
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 */
8
9#include <linux/tcp.h>
10#include <linux/slab.h>
11#include <linux/random.h>
Jason A. Donenfeldfe62d052017-01-08 13:54:03 +010012#include <linux/siphash.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070013#include <linux/kernel.h>
Paul Gortmakerbc3b2d72011-07-15 11:47:34 -040014#include <linux/export.h>
Eric Dumazet84b114b2017-05-05 06:56:54 -070015#include <net/secure_seq.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070016#include <net/tcp.h>
KOVACS Krisztian86b08d82008-10-01 07:44:42 -070017#include <net/route.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018
Jason A. Donenfeldfe62d052017-01-08 13:54:03 +010019static siphash_key_t syncookie_secret[2] __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070020
21#define COOKIEBITS 24 /* Upper bits store count */
22#define COOKIEMASK (((__u32)1 << COOKIEBITS) - 1)
23
Florian Westphal274e2da2014-11-03 17:35:01 +010024/* 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
Al Viro714e85b2006-11-14 20:51:49 -080048static u32 cookie_hash(__be32 saddr, __be32 daddr, __be16 sport, __be16 dport,
Linus Torvalds1da177e2005-04-16 15:20:36 -070049 u32 count, int c)
50{
Hannes Frederic Sowab23a0022013-10-19 21:48:56 +020051 net_get_random_once(syncookie_secret, sizeof(syncookie_secret));
Jason A. Donenfeldfe62d052017-01-08 13:54:03 +010052 return siphash_4u32((__force u32)saddr, (__force u32)daddr,
53 (__force u32)sport << 16 | (__force u32)dport,
54 count, &syncookie_secret[c]);
Linus Torvalds1da177e2005-04-16 15:20:36 -070055}
56
Florian Westphal4dfc2812008-04-10 03:12:40 -070057
58/*
59 * when syncookies are in effect and tcp timestamps are enabled we encode
Florian Westphal734f6142010-06-21 11:48:44 +000060 * tcp options in the lower bits of the timestamp value that will be
Florian Westphal4dfc2812008-04-10 03:12:40 -070061 * 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 */
Eric Dumazet200ecef2019-11-07 11:51:18 -080065u64 cookie_init_timestamp(struct request_sock *req, u64 now)
Florian Westphal4dfc2812008-04-10 03:12:40 -070066{
67 struct inet_request_sock *ireq;
Eric Dumazet200ecef2019-11-07 11:51:18 -080068 u32 ts, ts_now = tcp_ns_to_ts(now);
Florian Westphal4dfc2812008-04-10 03:12:40 -070069 u32 options = 0;
70
71 ireq = inet_rsk(req);
Florian Westphal734f6142010-06-21 11:48:44 +000072
Florian Westphal274e2da2014-11-03 17:35:01 +010073 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;
Florian Westphal4dfc2812008-04-10 03:12:40 -070078
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 }
Eric Dumazetd3edd062018-09-21 08:51:50 -070087 return (u64)ts * (NSEC_PER_SEC / TCP_TS_HZ);
Florian Westphal4dfc2812008-04-10 03:12:40 -070088}
89
90
Al Viro714e85b2006-11-14 20:51:49 -080091static __u32 secure_tcp_syn_cookie(__be32 saddr, __be32 daddr, __be16 sport,
Florian Westphal8c27bd72013-09-20 22:32:55 +020092 __be16 dport, __u32 sseq, __u32 data)
Linus Torvalds1da177e2005-04-16 15:20:36 -070093{
94 /*
95 * Compute the secure sequence number.
96 * The output should be:
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +090097 * HASH(sec1,saddr,sport,daddr,dport,sec1) + sseq + (count * 2^24)
Linus Torvalds1da177e2005-04-16 15:20:36 -070098 * + (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 */
Florian Westphal8c27bd72013-09-20 22:32:55 +0200104 u32 count = tcp_cookie_time();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105 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 *
Florian Westphal8c27bd72013-09-20 22:32:55 +0200116 * 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.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119 */
Al Viro714e85b2006-11-14 20:51:49 -0800120static __u32 check_tcp_syn_cookie(__u32 cookie, __be32 saddr, __be32 daddr,
Florian Westphal8c27bd72013-09-20 22:32:55 +0200121 __be16 sport, __be16 dport, __u32 sseq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122{
Florian Westphal8c27bd72013-09-20 22:32:55 +0200123 u32 diff, count = tcp_cookie_time();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124
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) */
Weilong Chen0c9a67d2013-12-23 14:37:27 +0800129 diff = (count - (cookie >> COOKIEBITS)) & ((__u32) -1 >> COOKIEBITS);
Florian Westphal8c27bd72013-09-20 22:32:55 +0200130 if (diff >= MAX_SYNCOOKIE_AGE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131 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
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900138/*
Florian Westphal08629352013-09-20 22:32:56 +0200139 * 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%)
Florian Westphal5918e2f2010-06-03 00:43:57 +0000146 *
Florian Westphal08629352013-09-20 22:32:56 +0200147 * 1460 is the single most frequently announced mss value (30 to 46% depending
148 * on monitor location). Table must be sorted.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149 */
150static __u16 const msstab[] = {
Florian Westphal5918e2f2010-06-03 00:43:57 +0000151 536,
Florian Westphal08629352013-09-20 22:32:56 +0200152 1300,
153 1440, /* 1440, 1452: PPPoE */
Florian Westphal5918e2f2010-06-03 00:43:57 +0000154 1460,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156
157/*
158 * Generate a syncookie. mssp points to the mss, which is returned
159 * rounded down to the value encoded in the cookie.
160 */
Patrick McHardy01982302013-08-27 08:50:13 +0200161u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
162 u16 *mssp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164 int mssind;
165 const __u16 mss = *mssp;
166
Florian Westphal5918e2f2010-06-03 00:43:57 +0000167 for (mssind = ARRAY_SIZE(msstab) - 1; mssind ; mssind--)
168 if (mss >= msstab[mssind])
169 break;
170 *mssp = msstab[mssind];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -0700172 return secure_tcp_syn_cookie(iph->saddr, iph->daddr,
173 th->source, th->dest, ntohl(th->seq),
Florian Westphal8c27bd72013-09-20 22:32:55 +0200174 mssind);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175}
Patrick McHardy01982302013-08-27 08:50:13 +0200176EXPORT_SYMBOL_GPL(__cookie_v4_init_sequence);
177
Eric Dumazet3f684b42015-09-29 07:42:49 -0700178__u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mssp)
Patrick McHardy01982302013-08-27 08:50:13 +0200179{
180 const struct iphdr *iph = ip_hdr(skb);
181 const struct tcphdr *th = tcp_hdr(skb);
182
Patrick McHardy01982302013-08-27 08:50:13 +0200183 return __cookie_v4_init_sequence(iph, th, mssp);
184}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900186/*
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900187 * Check if a ack sequence number is a valid syncookie.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188 * Return the decoded mss if it is, or 0 if not.
189 */
Patrick McHardy01982302013-08-27 08:50:13 +0200190int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
191 u32 cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192{
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -0700193 __u32 seq = ntohl(th->seq) - 1;
194 __u32 mssind = check_tcp_syn_cookie(cookie, iph->saddr, iph->daddr,
Florian Westphal8c27bd72013-09-20 22:32:55 +0200195 th->source, th->dest, seq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196
Florian Westphal5918e2f2010-06-03 00:43:57 +0000197 return mssind < ARRAY_SIZE(msstab) ? msstab[mssind] : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198}
Patrick McHardy01982302013-08-27 08:50:13 +0200199EXPORT_SYMBOL_GPL(__cookie_v4_check);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200
Eric Dumazetb80c0e72015-06-04 18:30:43 -0700201struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
202 struct request_sock *req,
Eric Dumazet84b114b2017-05-05 06:56:54 -0700203 struct dst_entry *dst, u32 tsoff)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204{
Arnaldo Carvalho de Melo8292a172005-12-13 23:15:52 -0800205 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206 struct sock *child;
Eric Dumazet5e0724d2015-10-22 08:20:46 -0700207 bool own_req;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208
Eric Dumazet5e0724d2015-10-22 08:20:46 -0700209 child = icsk->icsk_af_ops->syn_recv_sock(sk, skb, req, dst,
210 NULL, &own_req);
Eric Dumazet0470c8c2015-03-17 18:32:31 -0700211 if (child) {
Reshetova, Elena41c6d652017-06-30 13:08:01 +0300212 refcount_set(&req->rsk_refcnt, 1);
Eric Dumazet84b114b2017-05-05 06:56:54 -0700213 tcp_sk(child)->tsoffset = tsoff;
Eric Dumazet6bcfd7f2015-10-08 11:16:48 -0700214 sock_rps_save_rxhash(child, skb);
Florian Westphal9466a1c2020-07-30 21:25:56 +0200215
Eric Dumazet0e8642c2020-07-31 19:09:29 -0700216 if (rsk_drop_req(req)) {
Florian Westphal9466a1c2020-07-30 21:25:56 +0200217 refcount_set(&req->rsk_refcnt, 2);
218 return child;
219 }
220
Guillaume Nault9403cf22019-03-19 16:05:44 +0100221 if (inet_csk_reqsk_queue_add(sk, req, child))
222 return child;
223
224 bh_unlock_sock(child);
225 sock_put(child);
Eric Dumazet0470c8c2015-03-17 18:32:31 -0700226 }
Guillaume Nault9403cf22019-03-19 16:05:44 +0100227 __reqsk_free(req);
228
229 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230}
Eric Dumazetb80c0e72015-06-04 18:30:43 -0700231EXPORT_SYMBOL(tcp_get_cookie_sock);
Florian Westphal4dfc2812008-04-10 03:12:40 -0700232
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 *
Florian Westphalf1673382014-11-03 17:35:02 +0100238 * return false if we decode a tcp option that is disabled
239 * on the host.
Florian Westphal4dfc2812008-04-10 03:12:40 -0700240 */
Eric Dumazetf9301032017-06-07 10:34:37 -0700241bool cookie_timestamp_decode(const struct net *net,
242 struct tcp_options_received *tcp_opt)
Florian Westphal4dfc2812008-04-10 03:12:40 -0700243{
Florian Westphal734f6142010-06-21 11:48:44 +0000244 /* echoed timestamp, lowest bits contain options */
Florian Westphal274e2da2014-11-03 17:35:01 +0100245 u32 options = tcp_opt->rcv_tsecr;
Florian Westphal4dfc2812008-04-10 03:12:40 -0700246
Florian Westphal8c763682010-06-16 14:42:15 -0700247 if (!tcp_opt->saw_tstamp) {
248 tcp_clear_options(tcp_opt);
249 return true;
250 }
251
Eric Dumazet5d2ed052017-06-07 10:34:39 -0700252 if (!net->ipv4.sysctl_tcp_timestamps)
Florian Westphal8c763682010-06-16 14:42:15 -0700253 return false;
254
Florian Westphal274e2da2014-11-03 17:35:01 +0100255 tcp_opt->sack_ok = (options & TS_OPT_SACK) ? TCP_SACK_SEEN : 0;
Florian Westphal4dfc2812008-04-10 03:12:40 -0700256
Eric Dumazetf9301032017-06-07 10:34:37 -0700257 if (tcp_opt->sack_ok && !net->ipv4.sysctl_tcp_sack)
Florian Westphal8c763682010-06-16 14:42:15 -0700258 return false;
Florian Westphal4dfc2812008-04-10 03:12:40 -0700259
Florian Westphal274e2da2014-11-03 17:35:01 +0100260 if ((options & TS_OPT_WSCALE_MASK) == TS_OPT_WSCALE_MASK)
Florian Westphal734f6142010-06-21 11:48:44 +0000261 return true; /* no window scaling */
262
263 tcp_opt->wscale_ok = 1;
Florian Westphal274e2da2014-11-03 17:35:01 +0100264 tcp_opt->snd_wscale = options & TS_OPT_WSCALE_MASK;
265
Eric Dumazet9bb37ef2017-06-07 10:34:38 -0700266 return net->ipv4.sysctl_tcp_window_scaling != 0;
Florian Westphal4dfc2812008-04-10 03:12:40 -0700267}
Florian Westphalf1673382014-11-03 17:35:02 +0100268EXPORT_SYMBOL(cookie_timestamp_decode);
269
270bool cookie_ecn_ok(const struct tcp_options_received *tcp_opt,
Florian Westphalf7b3bec2014-11-03 17:35:03 +0100271 const struct net *net, const struct dst_entry *dst)
Florian Westphalf1673382014-11-03 17:35:02 +0100272{
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
Florian Westphalf7b3bec2014-11-03 17:35:03 +0100281 return dst_feature(dst, RTAX_FEATURE_ECN);
Florian Westphalf1673382014-11-03 17:35:02 +0100282}
283EXPORT_SYMBOL(cookie_ecn_ok);
Florian Westphal4dfc2812008-04-10 03:12:40 -0700284
Florian Westphal6fc8c822020-07-30 21:25:55 +0200285struct request_sock *cookie_tcp_reqsk_alloc(const struct request_sock_ops *ops,
286 struct sock *sk,
287 struct sk_buff *skb)
288{
Florian Westphal6fc8c822020-07-30 21:25:55 +0200289 struct request_sock *req;
290
291#ifdef CONFIG_MPTCP
Eric Dumazet0e8642c2020-07-31 19:09:29 -0700292 struct tcp_request_sock *treq;
293
Florian Westphal6fc8c822020-07-30 21:25:55 +0200294 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
Eric Dumazet079096f2015-10-02 11:43:32 -0700319/* 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 */
Cong Wang461b74c2014-10-15 14:33:22 -0700323struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324{
Cong Wang461b74c2014-10-15 14:33:22 -0700325 struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
William Allen Simpson4957faade2009-12-02 18:25:27 +0000326 struct tcp_options_received tcp_opt;
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -0700327 struct inet_request_sock *ireq;
328 struct tcp_request_sock *treq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329 struct tcp_sock *tp = tcp_sk(sk);
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -0700330 const struct tcphdr *th = tcp_hdr(skb);
331 __u32 cookie = ntohl(th->ack_seq) - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 struct sock *ret = sk;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900333 struct request_sock *req;
334 int mss;
335 struct rtable *rt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336 __u8 rcv_wscale;
Eric Dumazetdfd25ff2012-03-10 09:20:21 +0000337 struct flowi4 fl4;
Eric Dumazet84b114b2017-05-05 06:56:54 -0700338 u32 tsoff = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339
Nikolay Borisov12ed8242016-02-03 09:46:51 +0200340 if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies || !th->ack || th->rst)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341 goto out;
342
Florian Westphal646697b2014-10-30 02:55:38 +0100343 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) {
Eric Dumazet02a1d6e2016-04-27 16:44:39 -0700348 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESFAILED);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349 goto out;
350 }
351
Eric Dumazet02a1d6e2016-04-27 16:44:39 -0700352 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESRECV);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353
David S. Millerbb5b7c12009-12-15 20:56:42 -0800354 /* check for timestamp cookie support */
355 memset(&tcp_opt, 0, sizeof(tcp_opt));
Eric Dumazeteed29f12017-06-07 10:34:36 -0700356 tcp_parse_options(sock_net(sk), skb, &tcp_opt, 0, NULL);
David S. Millerbb5b7c12009-12-15 20:56:42 -0800357
Eric Dumazet84b114b2017-05-05 06:56:54 -0700358 if (tcp_opt.saw_tstamp && tcp_opt.rcv_tsecr) {
Eric Dumazet5d2ed052017-06-07 10:34:39 -0700359 tsoff = secure_tcp_ts_off(sock_net(sk),
360 ip_hdr(skb)->daddr,
361 ip_hdr(skb)->saddr);
Eric Dumazet84b114b2017-05-05 06:56:54 -0700362 tcp_opt.rcv_tsecr -= tsoff;
363 }
364
Eric Dumazetf9301032017-06-07 10:34:37 -0700365 if (!cookie_timestamp_decode(sock_net(sk), &tcp_opt))
Florian Westphal8c763682010-06-16 14:42:15 -0700366 goto out;
David S. Millerbb5b7c12009-12-15 20:56:42 -0800367
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368 ret = NULL;
Florian Westphal6fc8c822020-07-30 21:25:55 +0200369 req = cookie_tcp_reqsk_alloc(&tcp_request_sock_ops, sk, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370 if (!req)
371 goto out;
372
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -0700373 ireq = inet_rsk(req);
374 treq = tcp_rsk(req);
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -0700375 treq->rcv_isn = ntohl(th->seq) - 1;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900376 treq->snt_isn = cookie;
Florian Westphal95a22ca2016-12-01 11:32:06 +0100377 treq->ts_off = 0;
Alexander Potapenko18bcf292017-07-17 12:35:58 +0200378 treq->txhash = net_tx_rndhash();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379 req->mss = mss;
Eric Dumazetb44084c2013-10-10 00:04:37 -0700380 ireq->ir_num = ntohs(th->dest);
381 ireq->ir_rmt_port = th->source;
Eric Dumazet08d2cc3b2015-03-18 14:05:38 -0700382 sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr);
383 sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr);
Lorenzo Colitti84f39b02014-05-13 10:17:35 -0700384 ireq->ir_mark = inet_request_mark(sk, skb);
David S. Millerbb5b7c12009-12-15 20:56:42 -0800385 ireq->snd_wscale = tcp_opt.snd_wscale;
David S. Millerbb5b7c12009-12-15 20:56:42 -0800386 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;
Eric Dumazet9a568de2017-05-16 14:00:14 -0700390 treq->snt_synack = 0;
Eric Dumazet9439ce02015-03-17 18:32:29 -0700391 treq->tfo_listener = false;
Florian Westphalae2dd712020-01-29 15:54:46 +0100392
Hans Wippelbc58a1b2018-03-23 11:05:45 +0100393 if (IS_ENABLED(CONFIG_SMC))
394 ireq->smc_ok = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395
David Ahern6dd9a142015-12-16 13:20:44 -0800396 ireq->ir_iif = inet_request_bound_dev_if(sk, skb);
Eric Dumazet16f86162015-03-13 15:51:10 -0700397
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398 /* 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 */
Eric Dumazetc92e8c02017-10-20 09:04:13 -0700401 RCU_INIT_POINTER(ireq->ireq_opt, tcp_v4_save_options(sock_net(sk), skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402
Paul Moore284904a2009-03-27 17:10:28 -0400403 if (security_inet_conn_request(sk, skb, req)) {
Eric Dumazet0470c8c2015-03-17 18:32:31 -0700404 reqsk_free(req);
Paul Moore284904a2009-03-27 17:10:28 -0400405 goto out;
406 }
407
Eric Dumazete6c022a4f2012-10-27 23:16:46 +0000408 req->num_retrans = 0;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900409
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410 /*
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
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900414 * no easy way to do this.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 */
David Ahern6dd9a142015-12-16 13:20:44 -0800416 flowi4_init_output(&fl4, ireq->ir_iif, ireq->ir_mark,
Dmitry Popovd66954a2013-04-11 08:55:07 +0000417 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE, IPPROTO_TCP,
Eric Dumazetdfd25ff2012-03-10 09:20:21 +0000418 inet_sk_flowi_flags(sk),
Cong Wang461b74c2014-10-15 14:33:22 -0700419 opt->srr ? opt->faddr : ireq->ir_rmt_addr,
Lorenzo Colittie2d118a2016-11-04 02:23:43 +0900420 ireq->ir_loc_addr, th->source, th->dest, sk->sk_uid);
Eric Dumazetdfd25ff2012-03-10 09:20:21 +0000421 security_req_classify_flow(req, flowi4_to_flowi(&fl4));
422 rt = ip_route_output_key(sock_net(sk), &fl4);
423 if (IS_ERR(rt)) {
Eric Dumazet0470c8c2015-03-17 18:32:31 -0700424 reqsk_free(req);
Eric Dumazetdfd25ff2012-03-10 09:20:21 +0000425 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426 }
427
428 /* Try to redo what tcp_v4_send_synack did. */
Eric Dumazeted53d0a2015-10-08 19:33:23 -0700429 req->rsk_window_clamp = tp->window_clamp ? :dst_metric(&rt->dst, RTAX_WINDOW);
Florian Westphal4dfc2812008-04-10 03:12:40 -0700430
Eric Dumazetceef9ab2017-10-27 07:47:24 -0700431 tcp_select_initial_window(sk, tcp_full_space(sk), req->mss,
Eric Dumazeted53d0a2015-10-08 19:33:23 -0700432 &req->rsk_rcv_wnd, &req->rsk_window_clamp,
laurent chavey31d12922009-12-15 11:15:28 +0000433 ireq->wscale_ok, &rcv_wscale,
Changli Gaod8d1f302010-06-10 23:31:35 -0700434 dst_metric(&rt->dst, RTAX_INITRWND));
Florian Westphal4dfc2812008-04-10 03:12:40 -0700435
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900436 ireq->rcv_wscale = rcv_wscale;
Florian Westphalf7b3bec2014-11-03 17:35:03 +0100437 ireq->ecn_ok = cookie_ecn_ok(&tcp_opt, sock_net(sk), &rt->dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438
Eric Dumazet84b114b2017-05-05 06:56:54 -0700439 ret = tcp_get_cookie_sock(sk, skb, req, &rt->dst, tsoff);
Eric Dumazetdfd25ff2012-03-10 09:20:21 +0000440 /* 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;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445out: return ret;
446}