]>
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> | |
16 | #include <linux/cryptohash.h> | |
17 | #include <linux/kernel.h> | |
bc3b2d7f | 18 | #include <linux/export.h> |
1da177e4 | 19 | #include <net/tcp.h> |
86b08d86 | 20 | #include <net/route.h> |
1da177e4 LT |
21 | |
22 | extern int sysctl_tcp_syncookies; | |
23 | ||
253ff516 | 24 | static u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS] __read_mostly; |
1da177e4 LT |
25 | |
26 | #define COOKIEBITS 24 /* Upper bits store count */ | |
27 | #define COOKIEMASK (((__u32)1 << COOKIEBITS) - 1) | |
28 | ||
274e2da0 FW |
29 | /* TCP Timestamp: 6 lowest bits of timestamp sent in the cookie SYN-ACK |
30 | * stores TCP options: | |
31 | * | |
32 | * MSB LSB | |
33 | * | 31 ... 6 | 5 | 4 | 3 2 1 0 | | |
34 | * | Timestamp | ECN | SACK | WScale | | |
35 | * | |
36 | * When we receive a valid cookie-ACK, we look at the echoed tsval (if | |
37 | * any) to figure out which TCP options we should use for the rebuilt | |
38 | * connection. | |
39 | * | |
40 | * A WScale setting of '0xf' (which is an invalid scaling value) | |
41 | * means that original syn did not include the TCP window scaling option. | |
42 | */ | |
43 | #define TS_OPT_WSCALE_MASK 0xf | |
44 | #define TS_OPT_SACK BIT(4) | |
45 | #define TS_OPT_ECN BIT(5) | |
46 | /* There is no TS_OPT_TIMESTAMP: | |
47 | * if ACK contains timestamp option, we already know it was | |
48 | * requested/supported by the syn/synack exchange. | |
49 | */ | |
50 | #define TSBITS 6 | |
51 | #define TSMASK (((__u32)1 << TSBITS) - 1) | |
52 | ||
245b2e70 TH |
53 | static DEFINE_PER_CPU(__u32 [16 + 5 + SHA_WORKSPACE_WORDS], |
54 | ipv4_cookie_scratch); | |
11baab7a | 55 | |
714e85be | 56 | static u32 cookie_hash(__be32 saddr, __be32 daddr, __be16 sport, __be16 dport, |
1da177e4 LT |
57 | u32 count, int c) |
58 | { | |
b23a002f HFS |
59 | __u32 *tmp; |
60 | ||
61 | net_get_random_once(syncookie_secret, sizeof(syncookie_secret)); | |
1da177e4 | 62 | |
903ceff7 | 63 | tmp = this_cpu_ptr(ipv4_cookie_scratch); |
2051f11f | 64 | memcpy(tmp + 4, syncookie_secret[c], sizeof(syncookie_secret[c])); |
714e85be AV |
65 | tmp[0] = (__force u32)saddr; |
66 | tmp[1] = (__force u32)daddr; | |
67 | tmp[2] = ((__force u32)sport << 16) + (__force u32)dport; | |
1da177e4 LT |
68 | tmp[3] = count; |
69 | sha_transform(tmp + 16, (__u8 *)tmp, tmp + 16 + 5); | |
70 | ||
71 | return tmp[17]; | |
72 | } | |
73 | ||
4dfc2817 FW |
74 | |
75 | /* | |
76 | * when syncookies are in effect and tcp timestamps are enabled we encode | |
734f614b | 77 | * tcp options in the lower bits of the timestamp value that will be |
4dfc2817 FW |
78 | * sent in the syn-ack. |
79 | * Since subsequent timestamps use the normal tcp_time_stamp value, we | |
80 | * must make sure that the resulting initial timestamp is <= tcp_time_stamp. | |
81 | */ | |
82 | __u32 cookie_init_timestamp(struct request_sock *req) | |
83 | { | |
84 | struct inet_request_sock *ireq; | |
85 | u32 ts, ts_now = tcp_time_stamp; | |
86 | u32 options = 0; | |
87 | ||
88 | ireq = inet_rsk(req); | |
734f614b | 89 | |
274e2da0 FW |
90 | options = ireq->wscale_ok ? ireq->snd_wscale : TS_OPT_WSCALE_MASK; |
91 | if (ireq->sack_ok) | |
92 | options |= TS_OPT_SACK; | |
93 | if (ireq->ecn_ok) | |
94 | options |= TS_OPT_ECN; | |
4dfc2817 FW |
95 | |
96 | ts = ts_now & ~TSMASK; | |
97 | ts |= options; | |
98 | if (ts > ts_now) { | |
99 | ts >>= TSBITS; | |
100 | ts--; | |
101 | ts <<= TSBITS; | |
102 | ts |= options; | |
103 | } | |
104 | return ts; | |
105 | } | |
106 | ||
107 | ||
714e85be | 108 | static __u32 secure_tcp_syn_cookie(__be32 saddr, __be32 daddr, __be16 sport, |
8c27bd75 | 109 | __be16 dport, __u32 sseq, __u32 data) |
1da177e4 LT |
110 | { |
111 | /* | |
112 | * Compute the secure sequence number. | |
113 | * The output should be: | |
e905a9ed | 114 | * HASH(sec1,saddr,sport,daddr,dport,sec1) + sseq + (count * 2^24) |
1da177e4 LT |
115 | * + (HASH(sec2,saddr,sport,daddr,dport,count,sec2) % 2^24). |
116 | * Where sseq is their sequence number and count increases every | |
117 | * minute by 1. | |
118 | * As an extra hack, we add a small "data" value that encodes the | |
119 | * MSS into the second hash value. | |
120 | */ | |
8c27bd75 | 121 | u32 count = tcp_cookie_time(); |
1da177e4 LT |
122 | return (cookie_hash(saddr, daddr, sport, dport, 0, 0) + |
123 | sseq + (count << COOKIEBITS) + | |
124 | ((cookie_hash(saddr, daddr, sport, dport, count, 1) + data) | |
125 | & COOKIEMASK)); | |
126 | } | |
127 | ||
128 | /* | |
129 | * This retrieves the small "data" value from the syncookie. | |
130 | * If the syncookie is bad, the data returned will be out of | |
131 | * range. This must be checked by the caller. | |
132 | * | |
8c27bd75 FW |
133 | * The count value used to generate the cookie must be less than |
134 | * MAX_SYNCOOKIE_AGE minutes in the past. | |
135 | * The return value (__u32)-1 if this test fails. | |
1da177e4 | 136 | */ |
714e85be | 137 | static __u32 check_tcp_syn_cookie(__u32 cookie, __be32 saddr, __be32 daddr, |
8c27bd75 | 138 | __be16 sport, __be16 dport, __u32 sseq) |
1da177e4 | 139 | { |
8c27bd75 | 140 | u32 diff, count = tcp_cookie_time(); |
1da177e4 LT |
141 | |
142 | /* Strip away the layers from the cookie */ | |
143 | cookie -= cookie_hash(saddr, daddr, sport, dport, 0, 0) + sseq; | |
144 | ||
145 | /* Cookie is now reduced to (count * 2^24) ^ (hash % 2^24) */ | |
0c9a67d2 | 146 | diff = (count - (cookie >> COOKIEBITS)) & ((__u32) -1 >> COOKIEBITS); |
8c27bd75 | 147 | if (diff >= MAX_SYNCOOKIE_AGE) |
1da177e4 LT |
148 | return (__u32)-1; |
149 | ||
150 | return (cookie - | |
151 | cookie_hash(saddr, daddr, sport, dport, count - diff, 1)) | |
152 | & COOKIEMASK; /* Leaving the data behind */ | |
153 | } | |
154 | ||
e905a9ed | 155 | /* |
08629354 FW |
156 | * MSS Values are chosen based on the 2011 paper |
157 | * 'An Analysis of TCP Maximum Segement Sizes' by S. Alcock and R. Nelson. | |
158 | * Values .. | |
159 | * .. lower than 536 are rare (< 0.2%) | |
160 | * .. between 537 and 1299 account for less than < 1.5% of observed values | |
161 | * .. in the 1300-1349 range account for about 15 to 20% of observed mss values | |
162 | * .. exceeding 1460 are very rare (< 0.04%) | |
5918e2fb | 163 | * |
08629354 FW |
164 | * 1460 is the single most frequently announced mss value (30 to 46% depending |
165 | * on monitor location). Table must be sorted. | |
1da177e4 LT |
166 | */ |
167 | static __u16 const msstab[] = { | |
5918e2fb | 168 | 536, |
08629354 FW |
169 | 1300, |
170 | 1440, /* 1440, 1452: PPPoE */ | |
5918e2fb | 171 | 1460, |
1da177e4 | 172 | }; |
1da177e4 LT |
173 | |
174 | /* | |
175 | * Generate a syncookie. mssp points to the mss, which is returned | |
176 | * rounded down to the value encoded in the cookie. | |
177 | */ | |
0198230b PM |
178 | u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, |
179 | u16 *mssp) | |
1da177e4 | 180 | { |
1da177e4 LT |
181 | int mssind; |
182 | const __u16 mss = *mssp; | |
183 | ||
5918e2fb FW |
184 | for (mssind = ARRAY_SIZE(msstab) - 1; mssind ; mssind--) |
185 | if (mss >= msstab[mssind]) | |
186 | break; | |
187 | *mssp = msstab[mssind]; | |
1da177e4 | 188 | |
aa8223c7 ACM |
189 | return secure_tcp_syn_cookie(iph->saddr, iph->daddr, |
190 | th->source, th->dest, ntohl(th->seq), | |
8c27bd75 | 191 | mssind); |
1da177e4 | 192 | } |
0198230b PM |
193 | EXPORT_SYMBOL_GPL(__cookie_v4_init_sequence); |
194 | ||
3f684b4b | 195 | __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mssp) |
0198230b PM |
196 | { |
197 | const struct iphdr *iph = ip_hdr(skb); | |
198 | const struct tcphdr *th = tcp_hdr(skb); | |
199 | ||
0198230b PM |
200 | return __cookie_v4_init_sequence(iph, th, mssp); |
201 | } | |
1da177e4 | 202 | |
e905a9ed YH |
203 | /* |
204 | * Check if a ack sequence number is a valid syncookie. | |
1da177e4 LT |
205 | * Return the decoded mss if it is, or 0 if not. |
206 | */ | |
0198230b PM |
207 | int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th, |
208 | u32 cookie) | |
1da177e4 | 209 | { |
aa8223c7 ACM |
210 | __u32 seq = ntohl(th->seq) - 1; |
211 | __u32 mssind = check_tcp_syn_cookie(cookie, iph->saddr, iph->daddr, | |
8c27bd75 | 212 | th->source, th->dest, seq); |
1da177e4 | 213 | |
5918e2fb | 214 | return mssind < ARRAY_SIZE(msstab) ? msstab[mssind] : 0; |
1da177e4 | 215 | } |
0198230b | 216 | EXPORT_SYMBOL_GPL(__cookie_v4_check); |
1da177e4 | 217 | |
b80c0e78 ED |
218 | struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb, |
219 | struct request_sock *req, | |
220 | struct dst_entry *dst) | |
1da177e4 | 221 | { |
8292a17a | 222 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 LT |
223 | struct sock *child; |
224 | ||
8292a17a | 225 | child = icsk->icsk_af_ops->syn_recv_sock(sk, skb, req, dst); |
0470c8ca ED |
226 | if (child) { |
227 | atomic_set(&req->rsk_refcnt, 1); | |
6bcfd7f8 | 228 | sock_rps_save_rxhash(child, skb); |
463c84b9 | 229 | inet_csk_reqsk_queue_add(sk, req, child); |
0470c8ca | 230 | } else { |
60236fdd | 231 | reqsk_free(req); |
0470c8ca | 232 | } |
1da177e4 LT |
233 | return child; |
234 | } | |
b80c0e78 | 235 | EXPORT_SYMBOL(tcp_get_cookie_sock); |
4dfc2817 FW |
236 | |
237 | /* | |
238 | * when syncookies are in effect and tcp timestamps are enabled we stored | |
239 | * additional tcp options in the timestamp. | |
240 | * This extracts these options from the timestamp echo. | |
241 | * | |
f1673381 FW |
242 | * return false if we decode a tcp option that is disabled |
243 | * on the host. | |
4dfc2817 | 244 | */ |
f1673381 | 245 | bool cookie_timestamp_decode(struct tcp_options_received *tcp_opt) |
4dfc2817 | 246 | { |
734f614b | 247 | /* echoed timestamp, lowest bits contain options */ |
274e2da0 | 248 | u32 options = tcp_opt->rcv_tsecr; |
4dfc2817 | 249 | |
8c763681 FW |
250 | if (!tcp_opt->saw_tstamp) { |
251 | tcp_clear_options(tcp_opt); | |
252 | return true; | |
253 | } | |
254 | ||
255 | if (!sysctl_tcp_timestamps) | |
256 | return false; | |
257 | ||
274e2da0 | 258 | tcp_opt->sack_ok = (options & TS_OPT_SACK) ? TCP_SACK_SEEN : 0; |
4dfc2817 | 259 | |
8c763681 FW |
260 | if (tcp_opt->sack_ok && !sysctl_tcp_sack) |
261 | return false; | |
4dfc2817 | 262 | |
274e2da0 | 263 | if ((options & TS_OPT_WSCALE_MASK) == TS_OPT_WSCALE_MASK) |
734f614b FW |
264 | return true; /* no window scaling */ |
265 | ||
266 | tcp_opt->wscale_ok = 1; | |
274e2da0 FW |
267 | tcp_opt->snd_wscale = options & TS_OPT_WSCALE_MASK; |
268 | ||
734f614b | 269 | return sysctl_tcp_window_scaling != 0; |
4dfc2817 | 270 | } |
f1673381 FW |
271 | EXPORT_SYMBOL(cookie_timestamp_decode); |
272 | ||
273 | bool cookie_ecn_ok(const struct tcp_options_received *tcp_opt, | |
f7b3bec6 | 274 | const struct net *net, const struct dst_entry *dst) |
f1673381 FW |
275 | { |
276 | bool ecn_ok = tcp_opt->rcv_tsecr & TS_OPT_ECN; | |
277 | ||
278 | if (!ecn_ok) | |
279 | return false; | |
280 | ||
281 | if (net->ipv4.sysctl_tcp_ecn) | |
282 | return true; | |
283 | ||
f7b3bec6 | 284 | return dst_feature(dst, RTAX_FEATURE_ECN); |
f1673381 FW |
285 | } |
286 | EXPORT_SYMBOL(cookie_ecn_ok); | |
4dfc2817 | 287 | |
079096f1 ED |
288 | /* On input, sk is a listener. |
289 | * Output is listener if incoming packet would not create a child | |
290 | * NULL if memory could not be allocated. | |
291 | */ | |
461b74c3 | 292 | struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb) |
1da177e4 | 293 | { |
461b74c3 | 294 | struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; |
4957faad | 295 | struct tcp_options_received tcp_opt; |
2e6599cb ACM |
296 | struct inet_request_sock *ireq; |
297 | struct tcp_request_sock *treq; | |
1da177e4 | 298 | struct tcp_sock *tp = tcp_sk(sk); |
aa8223c7 ACM |
299 | const struct tcphdr *th = tcp_hdr(skb); |
300 | __u32 cookie = ntohl(th->ack_seq) - 1; | |
1da177e4 | 301 | struct sock *ret = sk; |
e905a9ed YH |
302 | struct request_sock *req; |
303 | int mss; | |
304 | struct rtable *rt; | |
1da177e4 | 305 | __u8 rcv_wscale; |
dfd25fff | 306 | struct flowi4 fl4; |
1da177e4 | 307 | |
af9b4738 | 308 | if (!sysctl_tcp_syncookies || !th->ack || th->rst) |
1da177e4 LT |
309 | goto out; |
310 | ||
646697b9 FW |
311 | if (tcp_synq_no_recent_overflow(sk)) |
312 | goto out; | |
313 | ||
314 | mss = __cookie_v4_check(ip_hdr(skb), th, cookie); | |
315 | if (mss == 0) { | |
de0744af | 316 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SYNCOOKIESFAILED); |
1da177e4 LT |
317 | goto out; |
318 | } | |
319 | ||
de0744af | 320 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SYNCOOKIESRECV); |
1da177e4 | 321 | |
bb5b7c11 DM |
322 | /* check for timestamp cookie support */ |
323 | memset(&tcp_opt, 0, sizeof(tcp_opt)); | |
1a2c6181 | 324 | tcp_parse_options(skb, &tcp_opt, 0, NULL); |
bb5b7c11 | 325 | |
f1673381 | 326 | if (!cookie_timestamp_decode(&tcp_opt)) |
8c763681 | 327 | goto out; |
bb5b7c11 | 328 | |
1da177e4 | 329 | ret = NULL; |
a1a5344d | 330 | req = inet_reqsk_alloc(&tcp_request_sock_ops, sk, false); /* for safety */ |
1da177e4 LT |
331 | if (!req) |
332 | goto out; | |
333 | ||
2e6599cb ACM |
334 | ireq = inet_rsk(req); |
335 | treq = tcp_rsk(req); | |
aa8223c7 | 336 | treq->rcv_isn = ntohl(th->seq) - 1; |
e905a9ed | 337 | treq->snt_isn = cookie; |
1da177e4 | 338 | req->mss = mss; |
b44084c2 ED |
339 | ireq->ir_num = ntohs(th->dest); |
340 | ireq->ir_rmt_port = th->source; | |
08d2cc3b ED |
341 | sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr); |
342 | sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr); | |
84f39b08 | 343 | ireq->ir_mark = inet_request_mark(sk, skb); |
bb5b7c11 | 344 | ireq->snd_wscale = tcp_opt.snd_wscale; |
bb5b7c11 DM |
345 | ireq->sack_ok = tcp_opt.sack_ok; |
346 | ireq->wscale_ok = tcp_opt.wscale_ok; | |
347 | ireq->tstamp_ok = tcp_opt.saw_tstamp; | |
348 | req->ts_recent = tcp_opt.saw_tstamp ? tcp_opt.rcv_tsval : 0; | |
0f1c28ae | 349 | treq->snt_synack.v64 = 0; |
9439ce00 | 350 | treq->tfo_listener = false; |
1da177e4 | 351 | |
16f86165 ED |
352 | ireq->ir_iif = sk->sk_bound_dev_if; |
353 | ||
1da177e4 LT |
354 | /* We throwed the options of the initial SYN away, so we hope |
355 | * the ACK carries the same options again (see RFC1122 4.2.3.8) | |
356 | */ | |
e25f866f | 357 | ireq->opt = tcp_v4_save_options(skb); |
1da177e4 | 358 | |
284904aa | 359 | if (security_inet_conn_request(sk, skb, req)) { |
0470c8ca | 360 | reqsk_free(req); |
284904aa PM |
361 | goto out; |
362 | } | |
363 | ||
e6c022a4 | 364 | req->num_retrans = 0; |
e905a9ed | 365 | |
1da177e4 LT |
366 | /* |
367 | * We need to lookup the route here to get at the correct | |
368 | * window size. We should better make sure that the window size | |
369 | * hasn't changed since we received the original syn, but I see | |
e905a9ed | 370 | * no easy way to do this. |
1da177e4 | 371 | */ |
84f39b08 | 372 | flowi4_init_output(&fl4, sk->sk_bound_dev_if, ireq->ir_mark, |
d66954a0 | 373 | RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE, IPPROTO_TCP, |
dfd25fff | 374 | inet_sk_flowi_flags(sk), |
461b74c3 | 375 | opt->srr ? opt->faddr : ireq->ir_rmt_addr, |
634fb979 | 376 | ireq->ir_loc_addr, th->source, th->dest); |
dfd25fff ED |
377 | security_req_classify_flow(req, flowi4_to_flowi(&fl4)); |
378 | rt = ip_route_output_key(sock_net(sk), &fl4); | |
379 | if (IS_ERR(rt)) { | |
0470c8ca | 380 | reqsk_free(req); |
dfd25fff | 381 | goto out; |
1da177e4 LT |
382 | } |
383 | ||
384 | /* Try to redo what tcp_v4_send_synack did. */ | |
d8d1f30b | 385 | req->window_clamp = tp->window_clamp ? :dst_metric(&rt->dst, RTAX_WINDOW); |
4dfc2817 | 386 | |
1da177e4 | 387 | tcp_select_initial_window(tcp_full_space(sk), req->mss, |
e905a9ed | 388 | &req->rcv_wnd, &req->window_clamp, |
31d12926 | 389 | ireq->wscale_ok, &rcv_wscale, |
d8d1f30b | 390 | dst_metric(&rt->dst, RTAX_INITRWND)); |
4dfc2817 | 391 | |
e905a9ed | 392 | ireq->rcv_wscale = rcv_wscale; |
f7b3bec6 | 393 | ireq->ecn_ok = cookie_ecn_ok(&tcp_opt, sock_net(sk), &rt->dst); |
1da177e4 | 394 | |
b80c0e78 | 395 | ret = tcp_get_cookie_sock(sk, skb, req, &rt->dst); |
dfd25fff ED |
396 | /* ip_queue_xmit() depends on our flow being setup |
397 | * Normal sockets get it right from inet_csk_route_child_sock() | |
398 | */ | |
399 | if (ret) | |
400 | inet_sk(ret)->cork.fl.u.ip4 = fl4; | |
1da177e4 LT |
401 | out: return ret; |
402 | } |