xref: /openbmc/linux/net/ipv4/syncookies.c (revision 930beb5a)
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 static u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS];
29 
30 #define COOKIEBITS 24	/* Upper bits store count */
31 #define COOKIEMASK (((__u32)1 << COOKIEBITS) - 1)
32 
33 static DEFINE_PER_CPU(__u32 [16 + 5 + SHA_WORKSPACE_WORDS],
34 		      ipv4_cookie_scratch);
35 
36 static u32 cookie_hash(__be32 saddr, __be32 daddr, __be16 sport, __be16 dport,
37 		       u32 count, int c)
38 {
39 	__u32 *tmp;
40 
41 	net_get_random_once(syncookie_secret, sizeof(syncookie_secret));
42 
43 	tmp  = __get_cpu_var(ipv4_cookie_scratch);
44 	memcpy(tmp + 4, syncookie_secret[c], sizeof(syncookie_secret[c]));
45 	tmp[0] = (__force u32)saddr;
46 	tmp[1] = (__force u32)daddr;
47 	tmp[2] = ((__force u32)sport << 16) + (__force u32)dport;
48 	tmp[3] = count;
49 	sha_transform(tmp + 16, (__u8 *)tmp, tmp + 16 + 5);
50 
51 	return tmp[17];
52 }
53 
54 
55 /*
56  * when syncookies are in effect and tcp timestamps are enabled we encode
57  * tcp options in the lower bits of the timestamp value that will be
58  * sent in the syn-ack.
59  * Since subsequent timestamps use the normal tcp_time_stamp value, we
60  * must make sure that the resulting initial timestamp is <= tcp_time_stamp.
61  */
62 __u32 cookie_init_timestamp(struct request_sock *req)
63 {
64 	struct inet_request_sock *ireq;
65 	u32 ts, ts_now = tcp_time_stamp;
66 	u32 options = 0;
67 
68 	ireq = inet_rsk(req);
69 
70 	options = ireq->wscale_ok ? ireq->snd_wscale : 0xf;
71 	options |= ireq->sack_ok << 4;
72 	options |= ireq->ecn_ok << 5;
73 
74 	ts = ts_now & ~TSMASK;
75 	ts |= options;
76 	if (ts > ts_now) {
77 		ts >>= TSBITS;
78 		ts--;
79 		ts <<= TSBITS;
80 		ts |= options;
81 	}
82 	return ts;
83 }
84 
85 
86 static __u32 secure_tcp_syn_cookie(__be32 saddr, __be32 daddr, __be16 sport,
87 				   __be16 dport, __u32 sseq, __u32 data)
88 {
89 	/*
90 	 * Compute the secure sequence number.
91 	 * The output should be:
92 	 *   HASH(sec1,saddr,sport,daddr,dport,sec1) + sseq + (count * 2^24)
93 	 *      + (HASH(sec2,saddr,sport,daddr,dport,count,sec2) % 2^24).
94 	 * Where sseq is their sequence number and count increases every
95 	 * minute by 1.
96 	 * As an extra hack, we add a small "data" value that encodes the
97 	 * MSS into the second hash value.
98 	 */
99 	u32 count = tcp_cookie_time();
100 	return (cookie_hash(saddr, daddr, sport, dport, 0, 0) +
101 		sseq + (count << COOKIEBITS) +
102 		((cookie_hash(saddr, daddr, sport, dport, count, 1) + data)
103 		 & COOKIEMASK));
104 }
105 
106 /*
107  * This retrieves the small "data" value from the syncookie.
108  * If the syncookie is bad, the data returned will be out of
109  * range.  This must be checked by the caller.
110  *
111  * The count value used to generate the cookie must be less than
112  * MAX_SYNCOOKIE_AGE minutes in the past.
113  * The return value (__u32)-1 if this test fails.
114  */
115 static __u32 check_tcp_syn_cookie(__u32 cookie, __be32 saddr, __be32 daddr,
116 				  __be16 sport, __be16 dport, __u32 sseq)
117 {
118 	u32 diff, count = tcp_cookie_time();
119 
120 	/* Strip away the layers from the cookie */
121 	cookie -= cookie_hash(saddr, daddr, sport, dport, 0, 0) + sseq;
122 
123 	/* Cookie is now reduced to (count * 2^24) ^ (hash % 2^24) */
124 	diff = (count - (cookie >> COOKIEBITS)) & ((__u32) - 1 >> COOKIEBITS);
125 	if (diff >= MAX_SYNCOOKIE_AGE)
126 		return (__u32)-1;
127 
128 	return (cookie -
129 		cookie_hash(saddr, daddr, sport, dport, count - diff, 1))
130 		& COOKIEMASK;	/* Leaving the data behind */
131 }
132 
133 /*
134  * MSS Values are chosen based on the 2011 paper
135  * 'An Analysis of TCP Maximum Segement Sizes' by S. Alcock and R. Nelson.
136  * Values ..
137  *  .. lower than 536 are rare (< 0.2%)
138  *  .. between 537 and 1299 account for less than < 1.5% of observed values
139  *  .. in the 1300-1349 range account for about 15 to 20% of observed mss values
140  *  .. exceeding 1460 are very rare (< 0.04%)
141  *
142  *  1460 is the single most frequently announced mss value (30 to 46% depending
143  *  on monitor location).  Table must be sorted.
144  */
145 static __u16 const msstab[] = {
146 	536,
147 	1300,
148 	1440,	/* 1440, 1452: PPPoE */
149 	1460,
150 };
151 
152 /*
153  * Generate a syncookie.  mssp points to the mss, which is returned
154  * rounded down to the value encoded in the cookie.
155  */
156 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
157 			      u16 *mssp)
158 {
159 	int mssind;
160 	const __u16 mss = *mssp;
161 
162 	for (mssind = ARRAY_SIZE(msstab) - 1; mssind ; mssind--)
163 		if (mss >= msstab[mssind])
164 			break;
165 	*mssp = msstab[mssind];
166 
167 	return secure_tcp_syn_cookie(iph->saddr, iph->daddr,
168 				     th->source, th->dest, ntohl(th->seq),
169 				     mssind);
170 }
171 EXPORT_SYMBOL_GPL(__cookie_v4_init_sequence);
172 
173 __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb, __u16 *mssp)
174 {
175 	const struct iphdr *iph = ip_hdr(skb);
176 	const struct tcphdr *th = tcp_hdr(skb);
177 
178 	tcp_synq_overflow(sk);
179 	NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
180 
181 	return __cookie_v4_init_sequence(iph, th, mssp);
182 }
183 
184 /*
185  * Check if a ack sequence number is a valid syncookie.
186  * Return the decoded mss if it is, or 0 if not.
187  */
188 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
189 		      u32 cookie)
190 {
191 	__u32 seq = ntohl(th->seq) - 1;
192 	__u32 mssind = check_tcp_syn_cookie(cookie, iph->saddr, iph->daddr,
193 					    th->source, th->dest, seq);
194 
195 	return mssind < ARRAY_SIZE(msstab) ? msstab[mssind] : 0;
196 }
197 EXPORT_SYMBOL_GPL(__cookie_v4_check);
198 
199 static inline struct sock *get_cookie_sock(struct sock *sk, struct sk_buff *skb,
200 					   struct request_sock *req,
201 					   struct dst_entry *dst)
202 {
203 	struct inet_connection_sock *icsk = inet_csk(sk);
204 	struct sock *child;
205 
206 	child = icsk->icsk_af_ops->syn_recv_sock(sk, skb, req, dst);
207 	if (child)
208 		inet_csk_reqsk_queue_add(sk, req, child);
209 	else
210 		reqsk_free(req);
211 
212 	return child;
213 }
214 
215 
216 /*
217  * when syncookies are in effect and tcp timestamps are enabled we stored
218  * additional tcp options in the timestamp.
219  * This extracts these options from the timestamp echo.
220  *
221  * The lowest 4 bits store snd_wscale.
222  * next 2 bits indicate SACK and ECN support.
223  *
224  * return false if we decode an option that should not be.
225  */
226 bool cookie_check_timestamp(struct tcp_options_received *tcp_opt,
227 			struct net *net, bool *ecn_ok)
228 {
229 	/* echoed timestamp, lowest bits contain options */
230 	u32 options = tcp_opt->rcv_tsecr & TSMASK;
231 
232 	if (!tcp_opt->saw_tstamp)  {
233 		tcp_clear_options(tcp_opt);
234 		return true;
235 	}
236 
237 	if (!sysctl_tcp_timestamps)
238 		return false;
239 
240 	tcp_opt->sack_ok = (options & (1 << 4)) ? TCP_SACK_SEEN : 0;
241 	*ecn_ok = (options >> 5) & 1;
242 	if (*ecn_ok && !net->ipv4.sysctl_tcp_ecn)
243 		return false;
244 
245 	if (tcp_opt->sack_ok && !sysctl_tcp_sack)
246 		return false;
247 
248 	if ((options & 0xf) == 0xf)
249 		return true; /* no window scaling */
250 
251 	tcp_opt->wscale_ok = 1;
252 	tcp_opt->snd_wscale = options & 0xf;
253 	return sysctl_tcp_window_scaling != 0;
254 }
255 EXPORT_SYMBOL(cookie_check_timestamp);
256 
257 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
258 			     struct ip_options *opt)
259 {
260 	struct tcp_options_received tcp_opt;
261 	struct inet_request_sock *ireq;
262 	struct tcp_request_sock *treq;
263 	struct tcp_sock *tp = tcp_sk(sk);
264 	const struct tcphdr *th = tcp_hdr(skb);
265 	__u32 cookie = ntohl(th->ack_seq) - 1;
266 	struct sock *ret = sk;
267 	struct request_sock *req;
268 	int mss;
269 	struct rtable *rt;
270 	__u8 rcv_wscale;
271 	bool ecn_ok = false;
272 	struct flowi4 fl4;
273 
274 	if (!sysctl_tcp_syncookies || !th->ack || th->rst)
275 		goto out;
276 
277 	if (tcp_synq_no_recent_overflow(sk) ||
278 	    (mss = __cookie_v4_check(ip_hdr(skb), th, cookie)) == 0) {
279 		NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SYNCOOKIESFAILED);
280 		goto out;
281 	}
282 
283 	NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SYNCOOKIESRECV);
284 
285 	/* check for timestamp cookie support */
286 	memset(&tcp_opt, 0, sizeof(tcp_opt));
287 	tcp_parse_options(skb, &tcp_opt, 0, NULL);
288 
289 	if (!cookie_check_timestamp(&tcp_opt, sock_net(sk), &ecn_ok))
290 		goto out;
291 
292 	ret = NULL;
293 	req = inet_reqsk_alloc(&tcp_request_sock_ops); /* for safety */
294 	if (!req)
295 		goto out;
296 
297 	ireq = inet_rsk(req);
298 	treq = tcp_rsk(req);
299 	treq->rcv_isn		= ntohl(th->seq) - 1;
300 	treq->snt_isn		= cookie;
301 	req->mss		= mss;
302 	ireq->ir_num		= ntohs(th->dest);
303 	ireq->ir_rmt_port	= th->source;
304 	ireq->ir_loc_addr	= ip_hdr(skb)->daddr;
305 	ireq->ir_rmt_addr	= ip_hdr(skb)->saddr;
306 	ireq->ecn_ok		= ecn_ok;
307 	ireq->snd_wscale	= tcp_opt.snd_wscale;
308 	ireq->sack_ok		= tcp_opt.sack_ok;
309 	ireq->wscale_ok		= tcp_opt.wscale_ok;
310 	ireq->tstamp_ok		= tcp_opt.saw_tstamp;
311 	req->ts_recent		= tcp_opt.saw_tstamp ? tcp_opt.rcv_tsval : 0;
312 	treq->snt_synack	= tcp_opt.saw_tstamp ? tcp_opt.rcv_tsecr : 0;
313 	treq->listener		= NULL;
314 
315 	/* We throwed the options of the initial SYN away, so we hope
316 	 * the ACK carries the same options again (see RFC1122 4.2.3.8)
317 	 */
318 	if (opt && opt->optlen) {
319 		int opt_size = sizeof(struct ip_options_rcu) + opt->optlen;
320 
321 		ireq->opt = kmalloc(opt_size, GFP_ATOMIC);
322 		if (ireq->opt != NULL && ip_options_echo(&ireq->opt->opt, skb)) {
323 			kfree(ireq->opt);
324 			ireq->opt = NULL;
325 		}
326 	}
327 
328 	if (security_inet_conn_request(sk, skb, req)) {
329 		reqsk_free(req);
330 		goto out;
331 	}
332 
333 	req->expires	= 0UL;
334 	req->num_retrans = 0;
335 
336 	/*
337 	 * We need to lookup the route here to get at the correct
338 	 * window size. We should better make sure that the window size
339 	 * hasn't changed since we received the original syn, but I see
340 	 * no easy way to do this.
341 	 */
342 	flowi4_init_output(&fl4, sk->sk_bound_dev_if, sk->sk_mark,
343 			   RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE, IPPROTO_TCP,
344 			   inet_sk_flowi_flags(sk),
345 			   (opt && opt->srr) ? opt->faddr : ireq->ir_rmt_addr,
346 			   ireq->ir_loc_addr, th->source, th->dest);
347 	security_req_classify_flow(req, flowi4_to_flowi(&fl4));
348 	rt = ip_route_output_key(sock_net(sk), &fl4);
349 	if (IS_ERR(rt)) {
350 		reqsk_free(req);
351 		goto out;
352 	}
353 
354 	/* Try to redo what tcp_v4_send_synack did. */
355 	req->window_clamp = tp->window_clamp ? :dst_metric(&rt->dst, RTAX_WINDOW);
356 
357 	tcp_select_initial_window(tcp_full_space(sk), req->mss,
358 				  &req->rcv_wnd, &req->window_clamp,
359 				  ireq->wscale_ok, &rcv_wscale,
360 				  dst_metric(&rt->dst, RTAX_INITRWND));
361 
362 	ireq->rcv_wscale  = rcv_wscale;
363 
364 	ret = get_cookie_sock(sk, skb, req, &rt->dst);
365 	/* ip_queue_xmit() depends on our flow being setup
366 	 * Normal sockets get it right from inet_csk_route_child_sock()
367 	 */
368 	if (ret)
369 		inet_sk(ret)->cork.fl.u.ip4 = fl4;
370 out:	return ret;
371 }
372