blob: d346c22aa6ae170a45d540f27cc04c3028f16c6e [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * Syncookies implementation for the Linux kernel
3 *
4 * Copyright (C) 1997 Andi Kleen
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +09005 * Based on ideas by D.J.Bernstein and Eric Schenk.
Linus Torvalds1da177e2005-04-16 15:20:36 -07006 *
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.
Linus Torvalds1da177e2005-04-16 15:20:36 -070011 */
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>
18#include <net/tcp.h>
KOVACS Krisztian86b08d82008-10-01 07:44:42 -070019#include <net/route.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020
Florian Westphal4dfc2812008-04-10 03:12:40 -070021/* Timestamps: lowest 9 bits store TCP options */
22#define TSBITS 9
23#define TSMASK (((__u32)1 << TSBITS) - 1)
24
Linus Torvalds1da177e2005-04-16 15:20:36 -070025extern int sysctl_tcp_syncookies;
26
Florian Westphal2051f112008-03-23 22:21:28 -070027__u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS];
Glenn Griffinc6aefaf2008-02-07 21:49:26 -080028EXPORT_SYMBOL(syncookie_secret);
Linus Torvalds1da177e2005-04-16 15:20:36 -070029
30static __init int init_syncookies(void)
31{
32 get_random_bytes(syncookie_secret, sizeof(syncookie_secret));
33 return 0;
34}
Glenn Griffinc6aefaf2008-02-07 21:49:26 -080035__initcall(init_syncookies);
Linus Torvalds1da177e2005-04-16 15:20:36 -070036
37#define COOKIEBITS 24 /* Upper bits store count */
38#define COOKIEMASK (((__u32)1 << COOKIEBITS) - 1)
39
Eric Dumazet11baab72008-02-07 10:40:19 +010040static DEFINE_PER_CPU(__u32, cookie_scratch)[16 + 5 + SHA_WORKSPACE_WORDS];
41
Al Viro714e85b2006-11-14 20:51:49 -080042static u32 cookie_hash(__be32 saddr, __be32 daddr, __be16 sport, __be16 dport,
Linus Torvalds1da177e2005-04-16 15:20:36 -070043 u32 count, int c)
44{
Eric Dumazet11baab72008-02-07 10:40:19 +010045 __u32 *tmp = __get_cpu_var(cookie_scratch);
Linus Torvalds1da177e2005-04-16 15:20:36 -070046
Florian Westphal2051f112008-03-23 22:21:28 -070047 memcpy(tmp + 4, syncookie_secret[c], sizeof(syncookie_secret[c]));
Al Viro714e85b2006-11-14 20:51:49 -080048 tmp[0] = (__force u32)saddr;
49 tmp[1] = (__force u32)daddr;
50 tmp[2] = ((__force u32)sport << 16) + (__force u32)dport;
Linus Torvalds1da177e2005-04-16 15:20:36 -070051 tmp[3] = count;
52 sha_transform(tmp + 16, (__u8 *)tmp, tmp + 16 + 5);
53
54 return tmp[17];
55}
56
Florian Westphal4dfc2812008-04-10 03:12:40 -070057
58/*
59 * when syncookies are in effect and tcp timestamps are enabled we encode
60 * tcp options in the lowest 9 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 */
65__u32 cookie_init_timestamp(struct request_sock *req)
66{
67 struct inet_request_sock *ireq;
68 u32 ts, ts_now = tcp_time_stamp;
69 u32 options = 0;
70
71 ireq = inet_rsk(req);
72 if (ireq->wscale_ok) {
73 options = ireq->snd_wscale;
74 options |= ireq->rcv_wscale << 4;
75 }
76 options |= ireq->sack_ok << 8;
77
78 ts = ts_now & ~TSMASK;
79 ts |= options;
80 if (ts > ts_now) {
81 ts >>= TSBITS;
82 ts--;
83 ts <<= TSBITS;
84 ts |= options;
85 }
86 return ts;
87}
88
89
Al Viro714e85b2006-11-14 20:51:49 -080090static __u32 secure_tcp_syn_cookie(__be32 saddr, __be32 daddr, __be16 sport,
91 __be16 dport, __u32 sseq, __u32 count,
Linus Torvalds1da177e2005-04-16 15:20:36 -070092 __u32 data)
93{
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 */
104
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 within
117 * "maxdiff" if the current (passed-in) "count". The return value
118 * is (__u32)-1 if this test fails.
119 */
Al Viro714e85b2006-11-14 20:51:49 -0800120static __u32 check_tcp_syn_cookie(__u32 cookie, __be32 saddr, __be32 daddr,
121 __be16 sport, __be16 dport, __u32 sseq,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122 __u32 count, __u32 maxdiff)
123{
124 __u32 diff;
125
126 /* Strip away the layers from the cookie */
127 cookie -= cookie_hash(saddr, daddr, sport, dport, 0, 0) + sseq;
128
129 /* Cookie is now reduced to (count * 2^24) ^ (hash % 2^24) */
130 diff = (count - (cookie >> COOKIEBITS)) & ((__u32) - 1 >> COOKIEBITS);
131 if (diff >= maxdiff)
132 return (__u32)-1;
133
134 return (cookie -
135 cookie_hash(saddr, daddr, sport, dport, count - diff, 1))
136 & COOKIEMASK; /* Leaving the data behind */
137}
138
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900139/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140 * This table has to be sorted and terminated with (__u16)-1.
141 * XXX generate a better table.
142 * Unresolved Issues: HIPPI with a 64k MSS is not well supported.
143 */
144static __u16 const msstab[] = {
145 64 - 1,
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900146 256 - 1,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147 512 - 1,
148 536 - 1,
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900149 1024 - 1,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150 1440 - 1,
151 1460 - 1,
152 4312 - 1,
153 (__u16)-1
154};
155/* The number doesn't include the -1 terminator */
156#define NUM_MSS (ARRAY_SIZE(msstab) - 1)
157
158/*
159 * Generate a syncookie. mssp points to the mss, which is returned
160 * rounded down to the value encoded in the cookie.
161 */
162__u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb, __u16 *mssp)
163{
164 struct tcp_sock *tp = tcp_sk(sk);
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -0700165 const struct iphdr *iph = ip_hdr(skb);
166 const struct tcphdr *th = tcp_hdr(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167 int mssind;
168 const __u16 mss = *mssp;
169
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170 tp->last_synq_overflow = jiffies;
171
172 /* XXX sort msstab[] by probability? Binary search? */
173 for (mssind = 0; mss > msstab[mssind + 1]; mssind++)
174 ;
175 *mssp = msstab[mssind] + 1;
176
Pavel Emelyanovde0744a2008-07-16 20:31:16 -0700177 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -0700179 return secure_tcp_syn_cookie(iph->saddr, iph->daddr,
180 th->source, th->dest, ntohl(th->seq),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181 jiffies / (HZ * 60), mssind);
182}
183
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900184/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185 * This (misnamed) value is the age of syncookie which is permitted.
186 * Its ideal value should be dependent on TCP_TIMEOUT_INIT and
187 * sysctl_tcp_retries1. It's a rather complicated formula (exponential
188 * backoff) to compute at runtime so it's currently hardcoded here.
189 */
190#define COUNTER_TRIES 4
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900191/*
192 * Check if a ack sequence number is a valid syncookie.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193 * Return the decoded mss if it is, or 0 if not.
194 */
195static inline int cookie_check(struct sk_buff *skb, __u32 cookie)
196{
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -0700197 const struct iphdr *iph = ip_hdr(skb);
198 const struct tcphdr *th = tcp_hdr(skb);
199 __u32 seq = ntohl(th->seq) - 1;
200 __u32 mssind = check_tcp_syn_cookie(cookie, iph->saddr, iph->daddr,
201 th->source, th->dest, seq,
202 jiffies / (HZ * 60),
203 COUNTER_TRIES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204
205 return mssind < NUM_MSS ? msstab[mssind] + 1 : 0;
206}
207
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208static inline struct sock *get_cookie_sock(struct sock *sk, struct sk_buff *skb,
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -0700209 struct request_sock *req,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210 struct dst_entry *dst)
211{
Arnaldo Carvalho de Melo8292a172005-12-13 23:15:52 -0800212 struct inet_connection_sock *icsk = inet_csk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213 struct sock *child;
214
Arnaldo Carvalho de Melo8292a172005-12-13 23:15:52 -0800215 child = icsk->icsk_af_ops->syn_recv_sock(sk, skb, req, dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216 if (child)
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700217 inet_csk_reqsk_queue_add(sk, req, child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218 else
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -0700219 reqsk_free(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220
221 return child;
222}
223
Florian Westphal4dfc2812008-04-10 03:12:40 -0700224
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 *
230 * The lowest 4 bits are for snd_wscale
231 * The next 4 lsb are for rcv_wscale
232 * The next lsb is for sack_ok
233 */
234void cookie_check_timestamp(struct tcp_options_received *tcp_opt)
235{
236 /* echoed timestamp, 9 lowest bits contain options */
237 u32 options = tcp_opt->rcv_tsecr & TSMASK;
238
239 tcp_opt->snd_wscale = options & 0xf;
240 options >>= 4;
241 tcp_opt->rcv_wscale = options & 0xf;
242
243 tcp_opt->sack_ok = (options >> 4) & 0x1;
244
245 if (tcp_opt->sack_ok)
246 tcp_sack_reset(tcp_opt);
247
248 if (tcp_opt->snd_wscale || tcp_opt->rcv_wscale)
249 tcp_opt->wscale_ok = 1;
250}
251EXPORT_SYMBOL(cookie_check_timestamp);
252
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
254 struct ip_options *opt)
255{
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -0700256 struct inet_request_sock *ireq;
257 struct tcp_request_sock *treq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 struct tcp_sock *tp = tcp_sk(sk);
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -0700259 const struct tcphdr *th = tcp_hdr(skb);
260 __u32 cookie = ntohl(th->ack_seq) - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 struct sock *ret = sk;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900262 struct request_sock *req;
263 int mss;
264 struct rtable *rt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265 __u8 rcv_wscale;
Florian Westphal4dfc2812008-04-10 03:12:40 -0700266 struct tcp_options_received tcp_opt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -0700268 if (!sysctl_tcp_syncookies || !th->ack)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269 goto out;
270
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900271 if (time_after(jiffies, tp->last_synq_overflow + TCP_TIMEOUT_INIT) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272 (mss = cookie_check(skb, cookie)) == 0) {
Pavel Emelyanovde0744a2008-07-16 20:31:16 -0700273 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SYNCOOKIESFAILED);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274 goto out;
275 }
276
Pavel Emelyanovde0744a2008-07-16 20:31:16 -0700277 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SYNCOOKIESRECV);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278
Florian Westphal4dfc2812008-04-10 03:12:40 -0700279 /* check for timestamp cookie support */
280 memset(&tcp_opt, 0, sizeof(tcp_opt));
281 tcp_parse_options(skb, &tcp_opt, 0);
282
283 if (tcp_opt.saw_tstamp)
284 cookie_check_timestamp(&tcp_opt);
285
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286 ret = NULL;
Arnaldo Carvalho de Meloce4a7d02008-06-10 12:39:35 -0700287 req = inet_reqsk_alloc(&tcp_request_sock_ops); /* for safety */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288 if (!req)
289 goto out;
290
Venkat Yekkirala4237c752006-07-24 23:32:50 -0700291 if (security_inet_conn_request(sk, skb, req)) {
292 reqsk_free(req);
293 goto out;
294 }
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -0700295 ireq = inet_rsk(req);
296 treq = tcp_rsk(req);
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -0700297 treq->rcv_isn = ntohl(th->seq) - 1;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900298 treq->snt_isn = cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299 req->mss = mss;
KOVACS Krisztiana3116ac2008-10-01 07:46:49 -0700300 ireq->loc_port = th->dest;
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -0700301 ireq->rmt_port = th->source;
Arnaldo Carvalho de Meloeddc9ec2007-04-20 22:47:35 -0700302 ireq->loc_addr = ip_hdr(skb)->daddr;
303 ireq->rmt_addr = ip_hdr(skb)->saddr;
Florian Westphal16df8452008-07-26 02:21:54 -0700304 ireq->ecn_ok = 0;
Florian Westphal4dfc2812008-04-10 03:12:40 -0700305 ireq->snd_wscale = tcp_opt.snd_wscale;
306 ireq->rcv_wscale = tcp_opt.rcv_wscale;
307 ireq->sack_ok = tcp_opt.sack_ok;
308 ireq->wscale_ok = tcp_opt.wscale_ok;
309 ireq->tstamp_ok = tcp_opt.saw_tstamp;
310 req->ts_recent = tcp_opt.saw_tstamp ? tcp_opt.rcv_tsval : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311
312 /* We throwed the options of the initial SYN away, so we hope
313 * the ACK carries the same options again (see RFC1122 4.2.3.8)
314 */
315 if (opt && opt->optlen) {
316 int opt_size = sizeof(struct ip_options) + opt->optlen;
317
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -0700318 ireq->opt = kmalloc(opt_size, GFP_ATOMIC);
319 if (ireq->opt != NULL && ip_options_echo(ireq->opt, skb)) {
320 kfree(ireq->opt);
321 ireq->opt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322 }
323 }
324
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900325 req->expires = 0UL;
326 req->retrans = 0;
327
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328 /*
329 * We need to lookup the route here to get at the correct
330 * window size. We should better make sure that the window size
331 * hasn't changed since we received the original syn, but I see
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900332 * no easy way to do this.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333 */
334 {
335 struct flowi fl = { .nl_u = { .ip4_u =
336 { .daddr = ((opt && opt->srr) ?
337 opt->faddr :
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -0700338 ireq->rmt_addr),
339 .saddr = ireq->loc_addr,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340 .tos = RT_CONN_FLAGS(sk) } },
341 .proto = IPPROTO_TCP,
KOVACS Krisztian86b08d82008-10-01 07:44:42 -0700342 .flags = inet_sk_flowi_flags(sk),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343 .uli_u = { .ports =
Arnaldo Carvalho de Meloaa8223c2007-04-10 21:04:22 -0700344 { .sport = th->dest,
345 .dport = th->source } } };
Venkat Yekkirala4237c752006-07-24 23:32:50 -0700346 security_req_classify_flow(req, &fl);
Denis V. Lunevf2063512008-01-22 22:07:34 -0800347 if (ip_route_output_key(&init_net, &rt, &fl)) {
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -0700348 reqsk_free(req);
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900349 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350 }
351 }
352
353 /* Try to redo what tcp_v4_send_synack did. */
Florian Westphal4dfc2812008-04-10 03:12:40 -0700354 req->window_clamp = tp->window_clamp ? :dst_metric(&rt->u.dst, RTAX_WINDOW);
355
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356 tcp_select_initial_window(tcp_full_space(sk), req->mss,
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900357 &req->rcv_wnd, &req->window_clamp,
Florian Westphal4dfc2812008-04-10 03:12:40 -0700358 ireq->wscale_ok, &rcv_wscale);
359
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900360 ireq->rcv_wscale = rcv_wscale;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361
362 ret = get_cookie_sock(sk, skb, req, &rt->u.dst);
363out: return ret;
364}