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 <net/tcp.h>
19#include <net/route.h>
20
21/* Timestamps: lowest bits store TCP options */
22#define TSBITS 6
23#define TSMASK (((__u32)1 << TSBITS) - 1)
24
25extern int sysctl_tcp_syncookies;
26
27__u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS];
28EXPORT_SYMBOL(syncookie_secret);
29
30static __init int init_syncookies(void)
31{
32	get_random_bytes(syncookie_secret, sizeof(syncookie_secret));
33	return 0;
34}
35__initcall(init_syncookies);
36
37#define COOKIEBITS 24	/* Upper bits store count */
38#define COOKIEMASK (((__u32)1 << COOKIEBITS) - 1)
39
40static DEFINE_PER_CPU(__u32 [16 + 5 + SHA_WORKSPACE_WORDS],
41		      ipv4_cookie_scratch);
42
43static u32 cookie_hash(__be32 saddr, __be32 daddr, __be16 sport, __be16 dport,
44		       u32 count, int c)
45{
46	__u32 *tmp = __get_cpu_var(ipv4_cookie_scratch);
47
48	memcpy(tmp + 4, syncookie_secret[c], sizeof(syncookie_secret[c]));
49	tmp[0] = (__force u32)saddr;
50	tmp[1] = (__force u32)daddr;
51	tmp[2] = ((__force u32)sport << 16) + (__force u32)dport;
52	tmp[3] = count;
53	sha_transform(tmp + 16, (__u8 *)tmp, tmp + 16 + 5);
54
55	return tmp[17];
56}
57
58
59/*
60 * when syncookies are in effect and tcp timestamps are enabled we encode
61 * tcp options in the lower bits of the timestamp value that will be
62 * sent in the syn-ack.
63 * Since subsequent timestamps use the normal tcp_time_stamp value, we
64 * must make sure that the resulting initial timestamp is <= tcp_time_stamp.
65 */
66__u32 cookie_init_timestamp(struct request_sock *req)
67{
68	struct inet_request_sock *ireq;
69	u32 ts, ts_now = tcp_time_stamp;
70	u32 options = 0;
71
72	ireq = inet_rsk(req);
73
74	options = ireq->wscale_ok ? ireq->snd_wscale : 0xf;
75	options |= ireq->sack_ok << 4;
76	options |= ireq->ecn_ok << 5;
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
90static __u32 secure_tcp_syn_cookie(__be32 saddr, __be32 daddr, __be16 sport,
91				   __be16 dport, __u32 sseq, __u32 count,
92				   __u32 data)
93{
94	/*
95	 * Compute the secure sequence number.
96	 * The output should be:
97	 *   HASH(sec1,saddr,sport,daddr,dport,sec1) + sseq + (count * 2^24)
98	 *      + (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 */
120static __u32 check_tcp_syn_cookie(__u32 cookie, __be32 saddr, __be32 daddr,
121				  __be16 sport, __be16 dport, __u32 sseq,
122				  __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
139/*
140 * MSS Values are taken from the 2009 paper
141 * 'Measuring TCP Maximum Segment Size' by S. Alcock and R. Nelson:
142 *  - values 1440 to 1460 accounted for 80% of observed mss values
143 *  - values outside the 536-1460 range are rare (<0.2%).
144 *
145 * Table must be sorted.
146 */
147static __u16 const msstab[] = {
148	64,
149	512,
150	536,
151	1024,
152	1440,
153	1460,
154	4312,
155	8960,
156};
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	const struct iphdr *iph = ip_hdr(skb);
165	const struct tcphdr *th = tcp_hdr(skb);
166	int mssind;
167	const __u16 mss = *mssp;
168
169	tcp_synq_overflow(sk);
170
171	for (mssind = ARRAY_SIZE(msstab) - 1; mssind ; mssind--)
172		if (mss >= msstab[mssind])
173			break;
174	*mssp = msstab[mssind];
175
176	NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
177
178	return secure_tcp_syn_cookie(iph->saddr, iph->daddr,
179				     th->source, th->dest, ntohl(th->seq),
180				     jiffies / (HZ * 60), mssind);
181}
182
183/*
184 * This (misnamed) value is the age of syncookie which is permitted.
185 * Its ideal value should be dependent on TCP_TIMEOUT_INIT and
186 * sysctl_tcp_retries1. It's a rather complicated formula (exponential
187 * backoff) to compute at runtime so it's currently hardcoded here.
188 */
189#define COUNTER_TRIES 4
190/*
191 * Check if a ack sequence number is a valid syncookie.
192 * Return the decoded mss if it is, or 0 if not.
193 */
194static inline int cookie_check(struct sk_buff *skb, __u32 cookie)
195{
196	const struct iphdr *iph = ip_hdr(skb);
197	const struct tcphdr *th = tcp_hdr(skb);
198	__u32 seq = ntohl(th->seq) - 1;
199	__u32 mssind = check_tcp_syn_cookie(cookie, iph->saddr, iph->daddr,
200					    th->source, th->dest, seq,
201					    jiffies / (HZ * 60),
202					    COUNTER_TRIES);
203
204	return mssind < ARRAY_SIZE(msstab) ? msstab[mssind] : 0;
205}
206
207static inline struct sock *get_cookie_sock(struct sock *sk, struct sk_buff *skb,
208					   struct request_sock *req,
209					   struct dst_entry *dst)
210{
211	struct inet_connection_sock *icsk = inet_csk(sk);
212	struct sock *child;
213
214	child = icsk->icsk_af_ops->syn_recv_sock(sk, skb, req, dst);
215	if (child)
216		inet_csk_reqsk_queue_add(sk, req, child);
217	else
218		reqsk_free(req);
219
220	return child;
221}
222
223
224/*
225 * when syncookies are in effect and tcp timestamps are enabled we stored
226 * additional tcp options in the timestamp.
227 * This extracts these options from the timestamp echo.
228 *
229 * The lowest 4 bits store snd_wscale.
230 * next 2 bits indicate SACK and ECN support.
231 *
232 * return false if we decode an option that should not be.
233 */
234bool cookie_check_timestamp(struct tcp_options_received *tcp_opt, bool *ecn_ok)
235{
236	/* echoed timestamp, lowest bits contain options */
237	u32 options = tcp_opt->rcv_tsecr & TSMASK;
238
239	if (!tcp_opt->saw_tstamp)  {
240		tcp_clear_options(tcp_opt);
241		return true;
242	}
243
244	if (!sysctl_tcp_timestamps)
245		return false;
246
247	tcp_opt->sack_ok = (options >> 4) & 0x1;
248	*ecn_ok = (options >> 5) & 1;
249	if (*ecn_ok && !sysctl_tcp_ecn)
250		return false;
251
252	if (tcp_opt->sack_ok && !sysctl_tcp_sack)
253		return false;
254
255	if ((options & 0xf) == 0xf)
256		return true; /* no window scaling */
257
258	tcp_opt->wscale_ok = 1;
259	tcp_opt->snd_wscale = options & 0xf;
260	return sysctl_tcp_window_scaling != 0;
261}
262EXPORT_SYMBOL(cookie_check_timestamp);
263
264struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
265			     struct ip_options *opt)
266{
267	struct tcp_options_received tcp_opt;
268	u8 *hash_location;
269	struct inet_request_sock *ireq;
270	struct tcp_request_sock *treq;
271	struct tcp_sock *tp = tcp_sk(sk);
272	const struct tcphdr *th = tcp_hdr(skb);
273	__u32 cookie = ntohl(th->ack_seq) - 1;
274	struct sock *ret = sk;
275	struct request_sock *req;
276	int mss;
277	struct rtable *rt;
278	__u8 rcv_wscale;
279	bool ecn_ok;
280
281	if (!sysctl_tcp_syncookies || !th->ack || th->rst)
282		goto out;
283
284	if (tcp_synq_no_recent_overflow(sk) ||
285	    (mss = cookie_check(skb, cookie)) == 0) {
286		NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SYNCOOKIESFAILED);
287		goto out;
288	}
289
290	NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SYNCOOKIESRECV);
291
292	/* check for timestamp cookie support */
293	memset(&tcp_opt, 0, sizeof(tcp_opt));
294	tcp_parse_options(skb, &tcp_opt, &hash_location, 0);
295
296	if (!cookie_check_timestamp(&tcp_opt, &ecn_ok))
297		goto out;
298
299	ret = NULL;
300	req = inet_reqsk_alloc(&tcp_request_sock_ops); /* for safety */
301	if (!req)
302		goto out;
303
304	ireq = inet_rsk(req);
305	treq = tcp_rsk(req);
306	treq->rcv_isn		= ntohl(th->seq) - 1;
307	treq->snt_isn		= cookie;
308	req->mss		= mss;
309	ireq->loc_port		= th->dest;
310	ireq->rmt_port		= th->source;
311	ireq->loc_addr		= ip_hdr(skb)->daddr;
312	ireq->rmt_addr		= ip_hdr(skb)->saddr;
313	ireq->ecn_ok		= ecn_ok;
314	ireq->snd_wscale	= tcp_opt.snd_wscale;
315	ireq->sack_ok		= tcp_opt.sack_ok;
316	ireq->wscale_ok		= tcp_opt.wscale_ok;
317	ireq->tstamp_ok		= tcp_opt.saw_tstamp;
318	req->ts_recent		= tcp_opt.saw_tstamp ? tcp_opt.rcv_tsval : 0;
319
320	/* We throwed the options of the initial SYN away, so we hope
321	 * the ACK carries the same options again (see RFC1122 4.2.3.8)
322	 */
323	if (opt && opt->optlen) {
324		int opt_size = sizeof(struct ip_options) + opt->optlen;
325
326		ireq->opt = kmalloc(opt_size, GFP_ATOMIC);
327		if (ireq->opt != NULL && ip_options_echo(ireq->opt, skb)) {
328			kfree(ireq->opt);
329			ireq->opt = NULL;
330		}
331	}
332
333	if (security_inet_conn_request(sk, skb, req)) {
334		reqsk_free(req);
335		goto out;
336	}
337
338	req->expires	= 0UL;
339	req->retrans	= 0;
340
341	/*
342	 * We need to lookup the route here to get at the correct
343	 * window size. We should better make sure that the window size
344	 * hasn't changed since we received the original syn, but I see
345	 * no easy way to do this.
346	 */
347	{
348		struct flowi fl = { .mark = sk->sk_mark,
349				    .nl_u = { .ip4_u =
350					      { .daddr = ((opt && opt->srr) ?
351							  opt->faddr :
352							  ireq->rmt_addr),
353						.saddr = ireq->loc_addr,
354						.tos = RT_CONN_FLAGS(sk) } },
355				    .proto = IPPROTO_TCP,
356				    .flags = inet_sk_flowi_flags(sk),
357				    .uli_u = { .ports =
358					       { .sport = th->dest,
359						 .dport = th->source } } };
360		security_req_classify_flow(req, &fl);
361		if (ip_route_output_key(sock_net(sk), &rt, &fl)) {
362			reqsk_free(req);
363			goto out;
364		}
365	}
366
367	/* Try to redo what tcp_v4_send_synack did. */
368	req->window_clamp = tp->window_clamp ? :dst_metric(&rt->dst, RTAX_WINDOW);
369
370	tcp_select_initial_window(tcp_full_space(sk), req->mss,
371				  &req->rcv_wnd, &req->window_clamp,
372				  ireq->wscale_ok, &rcv_wscale,
373				  dst_metric(&rt->dst, RTAX_INITRWND));
374
375	ireq->rcv_wscale  = rcv_wscale;
376
377	ret = get_cookie_sock(sk, skb, req, &rt->dst);
378out:	return ret;
379}
380