Root/
1 | /* |
2 | * Syncookies implementation for the Linux kernel |
3 | * |
4 | * Copyright (C) 1997 Andi Kleen |
5 | * Based on ideas by D.J.Bernstein and Eric Schenk. |
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. |
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> |
18 | #include <linux/export.h> |
19 | #include <net/tcp.h> |
20 | #include <net/route.h> |
21 | |
22 | /* Timestamps: lowest bits store TCP options */ |
23 | #define TSBITS 6 |
24 | #define TSMASK (((__u32)1 << TSBITS) - 1) |
25 | |
26 | extern int sysctl_tcp_syncookies; |
27 | |
28 | __u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS]; |
29 | EXPORT_SYMBOL(syncookie_secret); |
30 | |
31 | static __init int init_syncookies(void) |
32 | { |
33 | get_random_bytes(syncookie_secret, sizeof(syncookie_secret)); |
34 | return 0; |
35 | } |
36 | __initcall(init_syncookies); |
37 | |
38 | #define COOKIEBITS 24 /* Upper bits store count */ |
39 | #define COOKIEMASK (((__u32)1 << COOKIEBITS) - 1) |
40 | |
41 | static DEFINE_PER_CPU(__u32 [16 + 5 + SHA_WORKSPACE_WORDS], |
42 | ipv4_cookie_scratch); |
43 | |
44 | static u32 cookie_hash(__be32 saddr, __be32 daddr, __be16 sport, __be16 dport, |
45 | u32 count, int c) |
46 | { |
47 | __u32 *tmp = __get_cpu_var(ipv4_cookie_scratch); |
48 | |
49 | memcpy(tmp + 4, syncookie_secret[c], sizeof(syncookie_secret[c])); |
50 | tmp[0] = (__force u32)saddr; |
51 | tmp[1] = (__force u32)daddr; |
52 | tmp[2] = ((__force u32)sport << 16) + (__force u32)dport; |
53 | tmp[3] = count; |
54 | sha_transform(tmp + 16, (__u8 *)tmp, tmp + 16 + 5); |
55 | |
56 | return tmp[17]; |
57 | } |
58 | |
59 | |
60 | /* |
61 | * when syncookies are in effect and tcp timestamps are enabled we encode |
62 | * tcp options in the lower bits of the timestamp value that will be |
63 | * sent in the syn-ack. |
64 | * Since subsequent timestamps use the normal tcp_time_stamp value, we |
65 | * must make sure that the resulting initial timestamp is <= tcp_time_stamp. |
66 | */ |
67 | __u32 cookie_init_timestamp(struct request_sock *req) |
68 | { |
69 | struct inet_request_sock *ireq; |
70 | u32 ts, ts_now = tcp_time_stamp; |
71 | u32 options = 0; |
72 | |
73 | ireq = inet_rsk(req); |
74 | |
75 | options = ireq->wscale_ok ? ireq->snd_wscale : 0xf; |
76 | options |= ireq->sack_ok << 4; |
77 | options |= ireq->ecn_ok << 5; |
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 ts; |
88 | } |
89 | |
90 | |
91 | static __u32 secure_tcp_syn_cookie(__be32 saddr, __be32 daddr, __be16 sport, |
92 | __be16 dport, __u32 sseq, __u32 count, |
93 | __u32 data) |
94 | { |
95 | /* |
96 | * Compute the secure sequence number. |
97 | * The output should be: |
98 | * HASH(sec1,saddr,sport,daddr,dport,sec1) + sseq + (count * 2^24) |
99 | * + (HASH(sec2,saddr,sport,daddr,dport,count,sec2) % 2^24). |
100 | * Where sseq is their sequence number and count increases every |
101 | * minute by 1. |
102 | * As an extra hack, we add a small "data" value that encodes the |
103 | * MSS into the second hash value. |
104 | */ |
105 | |
106 | return (cookie_hash(saddr, daddr, sport, dport, 0, 0) + |
107 | sseq + (count << COOKIEBITS) + |
108 | ((cookie_hash(saddr, daddr, sport, dport, count, 1) + data) |
109 | & COOKIEMASK)); |
110 | } |
111 | |
112 | /* |
113 | * This retrieves the small "data" value from the syncookie. |
114 | * If the syncookie is bad, the data returned will be out of |
115 | * range. This must be checked by the caller. |
116 | * |
117 | * The count value used to generate the cookie must be within |
118 | * "maxdiff" if the current (passed-in) "count". The return value |
119 | * is (__u32)-1 if this test fails. |
120 | */ |
121 | static __u32 check_tcp_syn_cookie(__u32 cookie, __be32 saddr, __be32 daddr, |
122 | __be16 sport, __be16 dport, __u32 sseq, |
123 | __u32 count, __u32 maxdiff) |
124 | { |
125 | __u32 diff; |
126 | |
127 | /* Strip away the layers from the cookie */ |
128 | cookie -= cookie_hash(saddr, daddr, sport, dport, 0, 0) + sseq; |
129 | |
130 | /* Cookie is now reduced to (count * 2^24) ^ (hash % 2^24) */ |
131 | diff = (count - (cookie >> COOKIEBITS)) & ((__u32) - 1 >> COOKIEBITS); |
132 | if (diff >= maxdiff) |
133 | return (__u32)-1; |
134 | |
135 | return (cookie - |
136 | cookie_hash(saddr, daddr, sport, dport, count - diff, 1)) |
137 | & COOKIEMASK; /* Leaving the data behind */ |
138 | } |
139 | |
140 | /* |
141 | * MSS Values are taken from the 2009 paper |
142 | * 'Measuring TCP Maximum Segment Size' by S. Alcock and R. Nelson: |
143 | * - values 1440 to 1460 accounted for 80% of observed mss values |
144 | * - values outside the 536-1460 range are rare (<0.2%). |
145 | * |
146 | * Table must be sorted. |
147 | */ |
148 | static __u16 const msstab[] = { |
149 | 64, |
150 | 512, |
151 | 536, |
152 | 1024, |
153 | 1440, |
154 | 1460, |
155 | 4312, |
156 | 8960, |
157 | }; |
158 | |
159 | /* |
160 | * Generate a syncookie. mssp points to the mss, which is returned |
161 | * rounded down to the value encoded in the cookie. |
162 | */ |
163 | __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb, __u16 *mssp) |
164 | { |
165 | const struct iphdr *iph = ip_hdr(skb); |
166 | const struct tcphdr *th = tcp_hdr(skb); |
167 | int mssind; |
168 | const __u16 mss = *mssp; |
169 | |
170 | tcp_synq_overflow(sk); |
171 | |
172 | for (mssind = ARRAY_SIZE(msstab) - 1; mssind ; mssind--) |
173 | if (mss >= msstab[mssind]) |
174 | break; |
175 | *mssp = msstab[mssind]; |
176 | |
177 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT); |
178 | |
179 | return secure_tcp_syn_cookie(iph->saddr, iph->daddr, |
180 | th->source, th->dest, ntohl(th->seq), |
181 | jiffies / (HZ * 60), mssind); |
182 | } |
183 | |
184 | /* |
185 | * 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 |
191 | /* |
192 | * Check if a ack sequence number is a valid syncookie. |
193 | * Return the decoded mss if it is, or 0 if not. |
194 | */ |
195 | static inline int cookie_check(struct sk_buff *skb, __u32 cookie) |
196 | { |
197 | 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); |
204 | |
205 | return mssind < ARRAY_SIZE(msstab) ? msstab[mssind] : 0; |
206 | } |
207 | |
208 | static inline struct sock *get_cookie_sock(struct sock *sk, struct sk_buff *skb, |
209 | struct request_sock *req, |
210 | struct dst_entry *dst) |
211 | { |
212 | struct inet_connection_sock *icsk = inet_csk(sk); |
213 | struct sock *child; |
214 | |
215 | child = icsk->icsk_af_ops->syn_recv_sock(sk, skb, req, dst); |
216 | if (child) |
217 | inet_csk_reqsk_queue_add(sk, req, child); |
218 | else |
219 | reqsk_free(req); |
220 | |
221 | return child; |
222 | } |
223 | |
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 | * |
230 | * The lowest 4 bits store snd_wscale. |
231 | * next 2 bits indicate SACK and ECN support. |
232 | * |
233 | * return false if we decode an option that should not be. |
234 | */ |
235 | bool cookie_check_timestamp(struct tcp_options_received *tcp_opt, |
236 | struct net *net, bool *ecn_ok) |
237 | { |
238 | /* echoed timestamp, lowest bits contain options */ |
239 | u32 options = tcp_opt->rcv_tsecr & TSMASK; |
240 | |
241 | if (!tcp_opt->saw_tstamp) { |
242 | tcp_clear_options(tcp_opt); |
243 | return true; |
244 | } |
245 | |
246 | if (!sysctl_tcp_timestamps) |
247 | return false; |
248 | |
249 | tcp_opt->sack_ok = (options & (1 << 4)) ? TCP_SACK_SEEN : 0; |
250 | *ecn_ok = (options >> 5) & 1; |
251 | if (*ecn_ok && !net->ipv4.sysctl_tcp_ecn) |
252 | return false; |
253 | |
254 | if (tcp_opt->sack_ok && !sysctl_tcp_sack) |
255 | return false; |
256 | |
257 | if ((options & 0xf) == 0xf) |
258 | return true; /* no window scaling */ |
259 | |
260 | tcp_opt->wscale_ok = 1; |
261 | tcp_opt->snd_wscale = options & 0xf; |
262 | return sysctl_tcp_window_scaling != 0; |
263 | } |
264 | EXPORT_SYMBOL(cookie_check_timestamp); |
265 | |
266 | struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb, |
267 | struct ip_options *opt) |
268 | { |
269 | struct tcp_options_received tcp_opt; |
270 | const u8 *hash_location; |
271 | struct inet_request_sock *ireq; |
272 | struct tcp_request_sock *treq; |
273 | struct tcp_sock *tp = tcp_sk(sk); |
274 | const struct tcphdr *th = tcp_hdr(skb); |
275 | __u32 cookie = ntohl(th->ack_seq) - 1; |
276 | struct sock *ret = sk; |
277 | struct request_sock *req; |
278 | int mss; |
279 | struct rtable *rt; |
280 | __u8 rcv_wscale; |
281 | bool ecn_ok = false; |
282 | struct flowi4 fl4; |
283 | |
284 | if (!sysctl_tcp_syncookies || !th->ack || th->rst) |
285 | goto out; |
286 | |
287 | if (tcp_synq_no_recent_overflow(sk) || |
288 | (mss = cookie_check(skb, cookie)) == 0) { |
289 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SYNCOOKIESFAILED); |
290 | goto out; |
291 | } |
292 | |
293 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SYNCOOKIESRECV); |
294 | |
295 | /* check for timestamp cookie support */ |
296 | memset(&tcp_opt, 0, sizeof(tcp_opt)); |
297 | tcp_parse_options(skb, &tcp_opt, &hash_location, 0, NULL); |
298 | |
299 | if (!cookie_check_timestamp(&tcp_opt, sock_net(sk), &ecn_ok)) |
300 | goto out; |
301 | |
302 | ret = NULL; |
303 | req = inet_reqsk_alloc(&tcp_request_sock_ops); /* for safety */ |
304 | if (!req) |
305 | goto out; |
306 | |
307 | ireq = inet_rsk(req); |
308 | treq = tcp_rsk(req); |
309 | treq->rcv_isn = ntohl(th->seq) - 1; |
310 | treq->snt_isn = cookie; |
311 | req->mss = mss; |
312 | ireq->loc_port = th->dest; |
313 | ireq->rmt_port = th->source; |
314 | ireq->loc_addr = ip_hdr(skb)->daddr; |
315 | ireq->rmt_addr = ip_hdr(skb)->saddr; |
316 | ireq->ecn_ok = ecn_ok; |
317 | ireq->snd_wscale = tcp_opt.snd_wscale; |
318 | ireq->sack_ok = tcp_opt.sack_ok; |
319 | ireq->wscale_ok = tcp_opt.wscale_ok; |
320 | ireq->tstamp_ok = tcp_opt.saw_tstamp; |
321 | req->ts_recent = tcp_opt.saw_tstamp ? tcp_opt.rcv_tsval : 0; |
322 | treq->snt_synack = tcp_opt.saw_tstamp ? tcp_opt.rcv_tsecr : 0; |
323 | treq->listener = NULL; |
324 | |
325 | /* We throwed the options of the initial SYN away, so we hope |
326 | * the ACK carries the same options again (see RFC1122 4.2.3.8) |
327 | */ |
328 | if (opt && opt->optlen) { |
329 | int opt_size = sizeof(struct ip_options_rcu) + opt->optlen; |
330 | |
331 | ireq->opt = kmalloc(opt_size, GFP_ATOMIC); |
332 | if (ireq->opt != NULL && ip_options_echo(&ireq->opt->opt, skb)) { |
333 | kfree(ireq->opt); |
334 | ireq->opt = NULL; |
335 | } |
336 | } |
337 | |
338 | if (security_inet_conn_request(sk, skb, req)) { |
339 | reqsk_free(req); |
340 | goto out; |
341 | } |
342 | |
343 | req->expires = 0UL; |
344 | req->num_retrans = 0; |
345 | |
346 | /* |
347 | * We need to lookup the route here to get at the correct |
348 | * window size. We should better make sure that the window size |
349 | * hasn't changed since we received the original syn, but I see |
350 | * no easy way to do this. |
351 | */ |
352 | flowi4_init_output(&fl4, sk->sk_bound_dev_if, sk->sk_mark, |
353 | RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE, IPPROTO_TCP, |
354 | inet_sk_flowi_flags(sk), |
355 | (opt && opt->srr) ? opt->faddr : ireq->rmt_addr, |
356 | ireq->loc_addr, th->source, th->dest); |
357 | security_req_classify_flow(req, flowi4_to_flowi(&fl4)); |
358 | rt = ip_route_output_key(sock_net(sk), &fl4); |
359 | if (IS_ERR(rt)) { |
360 | reqsk_free(req); |
361 | goto out; |
362 | } |
363 | |
364 | /* Try to redo what tcp_v4_send_synack did. */ |
365 | req->window_clamp = tp->window_clamp ? :dst_metric(&rt->dst, RTAX_WINDOW); |
366 | |
367 | tcp_select_initial_window(tcp_full_space(sk), req->mss, |
368 | &req->rcv_wnd, &req->window_clamp, |
369 | ireq->wscale_ok, &rcv_wscale, |
370 | dst_metric(&rt->dst, RTAX_INITRWND)); |
371 | |
372 | ireq->rcv_wscale = rcv_wscale; |
373 | |
374 | ret = get_cookie_sock(sk, skb, req, &rt->dst); |
375 | /* ip_queue_xmit() depends on our flow being setup |
376 | * Normal sockets get it right from inet_csk_route_child_sock() |
377 | */ |
378 | if (ret) |
379 | inet_sk(ret)->cork.fl.u.ip4 = fl4; |
380 | out: return ret; |
381 | } |
382 |
Branches:
ben-wpan
ben-wpan-stefan
javiroman/ks7010
jz-2.6.34
jz-2.6.34-rc5
jz-2.6.34-rc6
jz-2.6.34-rc7
jz-2.6.35
jz-2.6.36
jz-2.6.37
jz-2.6.38
jz-2.6.39
jz-3.0
jz-3.1
jz-3.11
jz-3.12
jz-3.13
jz-3.15
jz-3.16
jz-3.18-dt
jz-3.2
jz-3.3
jz-3.4
jz-3.5
jz-3.6
jz-3.6-rc2-pwm
jz-3.9
jz-3.9-clk
jz-3.9-rc8
jz47xx
jz47xx-2.6.38
master
Tags:
od-2011-09-04
od-2011-09-18
v2.6.34-rc5
v2.6.34-rc6
v2.6.34-rc7
v3.9