]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Syncookies implementation for the Linux kernel | |
3 | * | |
4 | * Copyright (C) 1997 Andi Kleen | |
e905a9ed | 5 | * Based on ideas by D.J.Bernstein and Eric Schenk. |
1da177e4 LT |
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. | |
1da177e4 LT |
11 | */ |
12 | ||
13 | #include <linux/tcp.h> | |
14 | #include <linux/slab.h> | |
15 | #include <linux/random.h> | |
fe62d05b | 16 | #include <linux/siphash.h> |
1da177e4 | 17 | #include <linux/kernel.h> |
bc3b2d7f | 18 | #include <linux/export.h> |
1da177e4 | 19 | #include <net/tcp.h> |
86b08d86 | 20 | #include <net/route.h> |
1da177e4 | 21 | |
fe62d05b | 22 | static siphash_key_t syncookie_secret[2] __read_mostly; |
1da177e4 LT |
23 | |
24 | #define COOKIEBITS 24 /* Upper bits store count */ | |
25 | #define COOKIEMASK (((__u32)1 << COOKIEBITS) - 1) | |
26 | ||
274e2da0 FW |
27 | /* TCP Timestamp: 6 lowest bits of timestamp sent in the cookie SYN-ACK |
28 | * stores TCP options: | |
29 | * | |
30 | * MSB LSB | |
31 | * | 31 ... 6 | 5 | 4 | 3 2 1 0 | | |
32 | * | Timestamp | ECN | SACK | WScale | | |
33 | * | |
34 | * When we receive a valid cookie-ACK, we look at the echoed tsval (if | |
35 | * any) to figure out which TCP options we should use for the rebuilt | |
36 | * connection. | |
37 | * | |
38 | * A WScale setting of '0xf' (which is an invalid scaling value) | |
39 | * means that original syn did not include the TCP window scaling option. | |
40 | */ | |
41 | #define TS_OPT_WSCALE_MASK 0xf | |
42 | #define TS_OPT_SACK BIT(4) | |
43 | #define TS_OPT_ECN BIT(5) | |
44 | /* There is no TS_OPT_TIMESTAMP: | |
45 | * if ACK contains timestamp option, we already know it was | |
46 | * requested/supported by the syn/synack exchange. | |
47 | */ | |
48 | #define TSBITS 6 | |
49 | #define TSMASK (((__u32)1 << TSBITS) - 1) | |
50 | ||
714e85be | 51 | static u32 cookie_hash(__be32 saddr, __be32 daddr, __be16 sport, __be16 dport, |
1da177e4 LT |
52 | u32 count, int c) |
53 | { | |
b23a002f | 54 | net_get_random_once(syncookie_secret, sizeof(syncookie_secret)); |
fe62d05b JD |
55 | return siphash_4u32((__force u32)saddr, (__force u32)daddr, |
56 | (__force u32)sport << 16 | (__force u32)dport, | |
57 | count, &syncookie_secret[c]); | |
1da177e4 LT |
58 | } |
59 | ||
4dfc2817 FW |
60 | |
61 | /* | |
62 | * when syncookies are in effect and tcp timestamps are enabled we encode | |
734f614b | 63 | * tcp options in the lower bits of the timestamp value that will be |
4dfc2817 FW |
64 | * sent in the syn-ack. |
65 | * Since subsequent timestamps use the normal tcp_time_stamp value, we | |
66 | * must make sure that the resulting initial timestamp is <= tcp_time_stamp. | |
67 | */ | |
68 | __u32 cookie_init_timestamp(struct request_sock *req) | |
69 | { | |
70 | struct inet_request_sock *ireq; | |
71 | u32 ts, ts_now = tcp_time_stamp; | |
72 | u32 options = 0; | |
73 | ||
74 | ireq = inet_rsk(req); | |
734f614b | 75 | |
274e2da0 FW |
76 | options = ireq->wscale_ok ? ireq->snd_wscale : TS_OPT_WSCALE_MASK; |
77 | if (ireq->sack_ok) | |
78 | options |= TS_OPT_SACK; | |
79 | if (ireq->ecn_ok) | |
80 | options |= TS_OPT_ECN; | |
4dfc2817 FW |
81 | |
82 | ts = ts_now & ~TSMASK; | |
83 | ts |= options; | |
84 | if (ts > ts_now) { | |
85 | ts >>= TSBITS; | |
86 | ts--; | |
87 | ts <<= TSBITS; | |
88 | ts |= options; | |
89 | } | |
90 | return ts; | |
91 | } | |
92 | ||
93 | ||
714e85be | 94 | static __u32 secure_tcp_syn_cookie(__be32 saddr, __be32 daddr, __be16 sport, |
8c27bd75 | 95 | __be16 dport, __u32 sseq, __u32 data) |
1da177e4 LT |
96 | { |
97 | /* | |
98 | * Compute the secure sequence number. | |
99 | * The output should be: | |
e905a9ed | 100 | * HASH(sec1,saddr,sport,daddr,dport,sec1) + sseq + (count * 2^24) |
1da177e4 LT |
101 | * + (HASH(sec2,saddr,sport,daddr,dport,count,sec2) % 2^24). |
102 | * Where sseq is their sequence number and count increases every | |
103 | * minute by 1. | |
104 | * As an extra hack, we add a small "data" value that encodes the | |
105 | * MSS into the second hash value. | |
106 | */ | |
8c27bd75 | 107 | u32 count = tcp_cookie_time(); |
1da177e4 LT |
108 | return (cookie_hash(saddr, daddr, sport, dport, 0, 0) + |
109 | sseq + (count << COOKIEBITS) + | |
110 | ((cookie_hash(saddr, daddr, sport, dport, count, 1) + data) | |
111 | & COOKIEMASK)); | |
112 | } | |
113 | ||
114 | /* | |
115 | * This retrieves the small "data" value from the syncookie. | |
116 | * If the syncookie is bad, the data returned will be out of | |
117 | * range. This must be checked by the caller. | |
118 | * | |
8c27bd75 FW |
119 | * The count value used to generate the cookie must be less than |
120 | * MAX_SYNCOOKIE_AGE minutes in the past. | |
121 | * The return value (__u32)-1 if this test fails. | |
1da177e4 | 122 | */ |
714e85be | 123 | static __u32 check_tcp_syn_cookie(__u32 cookie, __be32 saddr, __be32 daddr, |
8c27bd75 | 124 | __be16 sport, __be16 dport, __u32 sseq) |
1da177e4 | 125 | { |
8c27bd75 | 126 | u32 diff, count = tcp_cookie_time(); |
1da177e4 LT |
127 | |
128 | /* Strip away the layers from the cookie */ | |
129 | cookie -= cookie_hash(saddr, daddr, sport, dport, 0, 0) + sseq; | |
130 | ||
131 | /* Cookie is now reduced to (count * 2^24) ^ (hash % 2^24) */ | |
0c9a67d2 | 132 | diff = (count - (cookie >> COOKIEBITS)) & ((__u32) -1 >> COOKIEBITS); |
8c27bd75 | 133 | if (diff >= MAX_SYNCOOKIE_AGE) |
1da177e4 LT |
134 | return (__u32)-1; |
135 | ||
136 | return (cookie - | |
137 | cookie_hash(saddr, daddr, sport, dport, count - diff, 1)) | |
138 | & COOKIEMASK; /* Leaving the data behind */ | |
139 | } | |
140 | ||
e905a9ed | 141 | /* |
08629354 FW |
142 | * MSS Values are chosen based on the 2011 paper |
143 | * 'An Analysis of TCP Maximum Segement Sizes' by S. Alcock and R. Nelson. | |
144 | * Values .. | |
145 | * .. lower than 536 are rare (< 0.2%) | |
146 | * .. between 537 and 1299 account for less than < 1.5% of observed values | |
147 | * .. in the 1300-1349 range account for about 15 to 20% of observed mss values | |
148 | * .. exceeding 1460 are very rare (< 0.04%) | |
5918e2fb | 149 | * |
08629354 FW |
150 | * 1460 is the single most frequently announced mss value (30 to 46% depending |
151 | * on monitor location). Table must be sorted. | |
1da177e4 LT |
152 | */ |
153 | static __u16 const msstab[] = { | |
5918e2fb | 154 | 536, |
08629354 FW |
155 | 1300, |
156 | 1440, /* 1440, 1452: PPPoE */ | |
5918e2fb | 157 | 1460, |
1da177e4 | 158 | }; |
1da177e4 LT |
159 | |
160 | /* | |
161 | * Generate a syncookie. mssp points to the mss, which is returned | |
162 | * rounded down to the value encoded in the cookie. | |
163 | */ | |
0198230b PM |
164 | u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, |
165 | u16 *mssp) | |
1da177e4 | 166 | { |
1da177e4 LT |
167 | int mssind; |
168 | const __u16 mss = *mssp; | |
169 | ||
5918e2fb FW |
170 | for (mssind = ARRAY_SIZE(msstab) - 1; mssind ; mssind--) |
171 | if (mss >= msstab[mssind]) | |
172 | break; | |
173 | *mssp = msstab[mssind]; | |
1da177e4 | 174 | |
aa8223c7 ACM |
175 | return secure_tcp_syn_cookie(iph->saddr, iph->daddr, |
176 | th->source, th->dest, ntohl(th->seq), | |
8c27bd75 | 177 | mssind); |
1da177e4 | 178 | } |
0198230b PM |
179 | EXPORT_SYMBOL_GPL(__cookie_v4_init_sequence); |
180 | ||
3f684b4b | 181 | __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mssp) |
0198230b PM |
182 | { |
183 | const struct iphdr *iph = ip_hdr(skb); | |
184 | const struct tcphdr *th = tcp_hdr(skb); | |
185 | ||
0198230b PM |
186 | return __cookie_v4_init_sequence(iph, th, mssp); |
187 | } | |
1da177e4 | 188 | |
e905a9ed YH |
189 | /* |
190 | * Check if a ack sequence number is a valid syncookie. | |
1da177e4 LT |
191 | * Return the decoded mss if it is, or 0 if not. |
192 | */ | |
0198230b PM |
193 | int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th, |
194 | u32 cookie) | |
1da177e4 | 195 | { |
aa8223c7 ACM |
196 | __u32 seq = ntohl(th->seq) - 1; |
197 | __u32 mssind = check_tcp_syn_cookie(cookie, iph->saddr, iph->daddr, | |
8c27bd75 | 198 | th->source, th->dest, seq); |
1da177e4 | 199 | |
5918e2fb | 200 | return mssind < ARRAY_SIZE(msstab) ? msstab[mssind] : 0; |
1da177e4 | 201 | } |
0198230b | 202 | EXPORT_SYMBOL_GPL(__cookie_v4_check); |
1da177e4 | 203 | |
b80c0e78 ED |
204 | struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb, |
205 | struct request_sock *req, | |
206 | struct dst_entry *dst) | |
1da177e4 | 207 | { |
8292a17a | 208 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 | 209 | struct sock *child; |
5e0724d0 | 210 | bool own_req; |
1da177e4 | 211 | |
5e0724d0 ED |
212 | child = icsk->icsk_af_ops->syn_recv_sock(sk, skb, req, dst, |
213 | NULL, &own_req); | |
0470c8ca ED |
214 | if (child) { |
215 | atomic_set(&req->rsk_refcnt, 1); | |
6bcfd7f8 | 216 | sock_rps_save_rxhash(child, skb); |
463c84b9 | 217 | inet_csk_reqsk_queue_add(sk, req, child); |
0470c8ca | 218 | } else { |
60236fdd | 219 | reqsk_free(req); |
0470c8ca | 220 | } |
1da177e4 LT |
221 | return child; |
222 | } | |
b80c0e78 | 223 | EXPORT_SYMBOL(tcp_get_cookie_sock); |
4dfc2817 FW |
224 | |
225 | /* | |
226 | * when syncookies are in effect and tcp timestamps are enabled we stored | |
227 | * additional tcp options in the timestamp. | |
228 | * This extracts these options from the timestamp echo. | |
229 | * | |
f1673381 FW |
230 | * return false if we decode a tcp option that is disabled |
231 | * on the host. | |
4dfc2817 | 232 | */ |
f1673381 | 233 | bool cookie_timestamp_decode(struct tcp_options_received *tcp_opt) |
4dfc2817 | 234 | { |
734f614b | 235 | /* echoed timestamp, lowest bits contain options */ |
274e2da0 | 236 | u32 options = tcp_opt->rcv_tsecr; |
4dfc2817 | 237 | |
8c763681 FW |
238 | if (!tcp_opt->saw_tstamp) { |
239 | tcp_clear_options(tcp_opt); | |
240 | return true; | |
241 | } | |
242 | ||
243 | if (!sysctl_tcp_timestamps) | |
244 | return false; | |
245 | ||
274e2da0 | 246 | tcp_opt->sack_ok = (options & TS_OPT_SACK) ? TCP_SACK_SEEN : 0; |
4dfc2817 | 247 | |
8c763681 FW |
248 | if (tcp_opt->sack_ok && !sysctl_tcp_sack) |
249 | return false; | |
4dfc2817 | 250 | |
274e2da0 | 251 | if ((options & TS_OPT_WSCALE_MASK) == TS_OPT_WSCALE_MASK) |
734f614b FW |
252 | return true; /* no window scaling */ |
253 | ||
254 | tcp_opt->wscale_ok = 1; | |
274e2da0 FW |
255 | tcp_opt->snd_wscale = options & TS_OPT_WSCALE_MASK; |
256 | ||
734f614b | 257 | return sysctl_tcp_window_scaling != 0; |
4dfc2817 | 258 | } |
f1673381 FW |
259 | EXPORT_SYMBOL(cookie_timestamp_decode); |
260 | ||
261 | bool cookie_ecn_ok(const struct tcp_options_received *tcp_opt, | |
f7b3bec6 | 262 | const struct net *net, const struct dst_entry *dst) |
f1673381 FW |
263 | { |
264 | bool ecn_ok = tcp_opt->rcv_tsecr & TS_OPT_ECN; | |
265 | ||
266 | if (!ecn_ok) | |
267 | return false; | |
268 | ||
269 | if (net->ipv4.sysctl_tcp_ecn) | |
270 | return true; | |
271 | ||
f7b3bec6 | 272 | return dst_feature(dst, RTAX_FEATURE_ECN); |
f1673381 FW |
273 | } |
274 | EXPORT_SYMBOL(cookie_ecn_ok); | |
4dfc2817 | 275 | |
079096f1 ED |
276 | /* On input, sk is a listener. |
277 | * Output is listener if incoming packet would not create a child | |
278 | * NULL if memory could not be allocated. | |
279 | */ | |
461b74c3 | 280 | struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb) |
1da177e4 | 281 | { |
461b74c3 | 282 | struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; |
4957faad | 283 | struct tcp_options_received tcp_opt; |
2e6599cb ACM |
284 | struct inet_request_sock *ireq; |
285 | struct tcp_request_sock *treq; | |
1da177e4 | 286 | struct tcp_sock *tp = tcp_sk(sk); |
aa8223c7 ACM |
287 | const struct tcphdr *th = tcp_hdr(skb); |
288 | __u32 cookie = ntohl(th->ack_seq) - 1; | |
1da177e4 | 289 | struct sock *ret = sk; |
e905a9ed YH |
290 | struct request_sock *req; |
291 | int mss; | |
292 | struct rtable *rt; | |
1da177e4 | 293 | __u8 rcv_wscale; |
dfd25fff | 294 | struct flowi4 fl4; |
1da177e4 | 295 | |
12ed8244 | 296 | if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies || !th->ack || th->rst) |
1da177e4 LT |
297 | goto out; |
298 | ||
646697b9 FW |
299 | if (tcp_synq_no_recent_overflow(sk)) |
300 | goto out; | |
301 | ||
302 | mss = __cookie_v4_check(ip_hdr(skb), th, cookie); | |
303 | if (mss == 0) { | |
02a1d6e7 | 304 | __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESFAILED); |
1da177e4 LT |
305 | goto out; |
306 | } | |
307 | ||
02a1d6e7 | 308 | __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESRECV); |
1da177e4 | 309 | |
bb5b7c11 DM |
310 | /* check for timestamp cookie support */ |
311 | memset(&tcp_opt, 0, sizeof(tcp_opt)); | |
1a2c6181 | 312 | tcp_parse_options(skb, &tcp_opt, 0, NULL); |
bb5b7c11 | 313 | |
f1673381 | 314 | if (!cookie_timestamp_decode(&tcp_opt)) |
8c763681 | 315 | goto out; |
bb5b7c11 | 316 | |
1da177e4 | 317 | ret = NULL; |
a1a5344d | 318 | req = inet_reqsk_alloc(&tcp_request_sock_ops, sk, false); /* for safety */ |
1da177e4 LT |
319 | if (!req) |
320 | goto out; | |
321 | ||
2e6599cb ACM |
322 | ireq = inet_rsk(req); |
323 | treq = tcp_rsk(req); | |
aa8223c7 | 324 | treq->rcv_isn = ntohl(th->seq) - 1; |
e905a9ed | 325 | treq->snt_isn = cookie; |
95a22cae | 326 | treq->ts_off = 0; |
1da177e4 | 327 | req->mss = mss; |
b44084c2 ED |
328 | ireq->ir_num = ntohs(th->dest); |
329 | ireq->ir_rmt_port = th->source; | |
08d2cc3b ED |
330 | sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr); |
331 | sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr); | |
84f39b08 | 332 | ireq->ir_mark = inet_request_mark(sk, skb); |
bb5b7c11 | 333 | ireq->snd_wscale = tcp_opt.snd_wscale; |
bb5b7c11 DM |
334 | ireq->sack_ok = tcp_opt.sack_ok; |
335 | ireq->wscale_ok = tcp_opt.wscale_ok; | |
336 | ireq->tstamp_ok = tcp_opt.saw_tstamp; | |
337 | req->ts_recent = tcp_opt.saw_tstamp ? tcp_opt.rcv_tsval : 0; | |
0f1c28ae | 338 | treq->snt_synack.v64 = 0; |
9439ce00 | 339 | treq->tfo_listener = false; |
1da177e4 | 340 | |
6dd9a14e | 341 | ireq->ir_iif = inet_request_bound_dev_if(sk, skb); |
16f86165 | 342 | |
1da177e4 LT |
343 | /* We throwed the options of the initial SYN away, so we hope |
344 | * the ACK carries the same options again (see RFC1122 4.2.3.8) | |
345 | */ | |
e25f866f | 346 | ireq->opt = tcp_v4_save_options(skb); |
1da177e4 | 347 | |
284904aa | 348 | if (security_inet_conn_request(sk, skb, req)) { |
0470c8ca | 349 | reqsk_free(req); |
284904aa PM |
350 | goto out; |
351 | } | |
352 | ||
e6c022a4 | 353 | req->num_retrans = 0; |
e905a9ed | 354 | |
1da177e4 LT |
355 | /* |
356 | * We need to lookup the route here to get at the correct | |
357 | * window size. We should better make sure that the window size | |
358 | * hasn't changed since we received the original syn, but I see | |
e905a9ed | 359 | * no easy way to do this. |
1da177e4 | 360 | */ |
6dd9a14e | 361 | flowi4_init_output(&fl4, ireq->ir_iif, ireq->ir_mark, |
d66954a0 | 362 | RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE, IPPROTO_TCP, |
dfd25fff | 363 | inet_sk_flowi_flags(sk), |
461b74c3 | 364 | opt->srr ? opt->faddr : ireq->ir_rmt_addr, |
e2d118a1 | 365 | ireq->ir_loc_addr, th->source, th->dest, sk->sk_uid); |
dfd25fff ED |
366 | security_req_classify_flow(req, flowi4_to_flowi(&fl4)); |
367 | rt = ip_route_output_key(sock_net(sk), &fl4); | |
368 | if (IS_ERR(rt)) { | |
0470c8ca | 369 | reqsk_free(req); |
dfd25fff | 370 | goto out; |
1da177e4 LT |
371 | } |
372 | ||
373 | /* Try to redo what tcp_v4_send_synack did. */ | |
ed53d0ab | 374 | req->rsk_window_clamp = tp->window_clamp ? :dst_metric(&rt->dst, RTAX_WINDOW); |
4dfc2817 | 375 | |
1da177e4 | 376 | tcp_select_initial_window(tcp_full_space(sk), req->mss, |
ed53d0ab | 377 | &req->rsk_rcv_wnd, &req->rsk_window_clamp, |
31d12926 | 378 | ireq->wscale_ok, &rcv_wscale, |
d8d1f30b | 379 | dst_metric(&rt->dst, RTAX_INITRWND)); |
4dfc2817 | 380 | |
e905a9ed | 381 | ireq->rcv_wscale = rcv_wscale; |
f7b3bec6 | 382 | ireq->ecn_ok = cookie_ecn_ok(&tcp_opt, sock_net(sk), &rt->dst); |
1da177e4 | 383 | |
b80c0e78 | 384 | ret = tcp_get_cookie_sock(sk, skb, req, &rt->dst); |
dfd25fff ED |
385 | /* ip_queue_xmit() depends on our flow being setup |
386 | * Normal sockets get it right from inet_csk_route_child_sock() | |
387 | */ | |
388 | if (ret) | |
389 | inet_sk(ret)->cork.fl.u.ip4 = fl4; | |
1da177e4 LT |
390 | out: return ret; |
391 | } |