xref: /linux/net/ipv6/syncookies.c (revision 07fdad3a93756b872da7b53647715c48d0f4a2d0)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  *  IPv6 Syncookies implementation for the Linux kernel
4  *
5  *  Authors:
6  *  Glenn Griffin	<ggriffin.kernel@gmail.com>
7  *
8  *  Based on IPv4 implementation by Andi Kleen
9  *  linux/net/ipv4/syncookies.c
10  */
11 
12 #include <linux/tcp.h>
13 #include <linux/random.h>
14 #include <linux/siphash.h>
15 #include <linux/kernel.h>
16 #include <net/secure_seq.h>
17 #include <net/ipv6.h>
18 #include <net/tcp.h>
19 #include <net/tcp_ecn.h>
20 
21 #define COOKIEBITS 24	/* Upper bits store count */
22 #define COOKIEMASK (((__u32)1 << COOKIEBITS) - 1)
23 
24 static siphash_aligned_key_t syncookie6_secret[2];
25 
26 /* RFC 2460, Section 8.3:
27  * [ipv6 tcp] MSS must be computed as the maximum packet size minus 60 [..]
28  *
29  * Due to IPV6_MIN_MTU=1280 the lowest possible MSS is 1220, which allows
30  * using higher values than ipv4 tcp syncookies.
31  * The other values are chosen based on ethernet (1500 and 9k MTU), plus
32  * one that accounts for common encap (PPPoe) overhead. Table must be sorted.
33  */
34 static __u16 const msstab[] = {
35 	1280 - 60, /* IPV6_MIN_MTU - 60 */
36 	1480 - 60,
37 	1500 - 60,
38 	9000 - 60,
39 };
40 
41 static u32 cookie_hash(const struct in6_addr *saddr,
42 		       const struct in6_addr *daddr,
43 		       __be16 sport, __be16 dport, u32 count, int c)
44 {
45 	const struct {
46 		struct in6_addr saddr;
47 		struct in6_addr daddr;
48 		u32 count;
49 		__be16 sport;
50 		__be16 dport;
51 	} __aligned(SIPHASH_ALIGNMENT) combined = {
52 		.saddr = *saddr,
53 		.daddr = *daddr,
54 		.count = count,
55 		.sport = sport,
56 		.dport = dport
57 	};
58 
59 	net_get_random_once(syncookie6_secret, sizeof(syncookie6_secret));
60 	return siphash(&combined, offsetofend(typeof(combined), dport),
61 		       &syncookie6_secret[c]);
62 }
63 
64 static __u32 secure_tcp_syn_cookie(const struct in6_addr *saddr,
65 				   const struct in6_addr *daddr,
66 				   __be16 sport, __be16 dport, __u32 sseq,
67 				   __u32 data)
68 {
69 	u32 count = tcp_cookie_time();
70 	return (cookie_hash(saddr, daddr, sport, dport, 0, 0) +
71 		sseq + (count << COOKIEBITS) +
72 		((cookie_hash(saddr, daddr, sport, dport, count, 1) + data)
73 		& COOKIEMASK));
74 }
75 
76 static __u32 check_tcp_syn_cookie(__u32 cookie, const struct in6_addr *saddr,
77 				  const struct in6_addr *daddr, __be16 sport,
78 				  __be16 dport, __u32 sseq)
79 {
80 	__u32 diff, count = tcp_cookie_time();
81 
82 	cookie -= cookie_hash(saddr, daddr, sport, dport, 0, 0) + sseq;
83 
84 	diff = (count - (cookie >> COOKIEBITS)) & ((__u32) -1 >> COOKIEBITS);
85 	if (diff >= MAX_SYNCOOKIE_AGE)
86 		return (__u32)-1;
87 
88 	return (cookie -
89 		cookie_hash(saddr, daddr, sport, dport, count - diff, 1))
90 		& COOKIEMASK;
91 }
92 
93 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
94 			      const struct tcphdr *th, __u16 *mssp)
95 {
96 	int mssind;
97 	const __u16 mss = *mssp;
98 
99 	for (mssind = ARRAY_SIZE(msstab) - 1; mssind ; mssind--)
100 		if (mss >= msstab[mssind])
101 			break;
102 
103 	*mssp = msstab[mssind];
104 
105 	return secure_tcp_syn_cookie(&iph->saddr, &iph->daddr, th->source,
106 				     th->dest, ntohl(th->seq), mssind);
107 }
108 EXPORT_SYMBOL_GPL(__cookie_v6_init_sequence);
109 
110 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mssp)
111 {
112 	const struct ipv6hdr *iph = ipv6_hdr(skb);
113 	const struct tcphdr *th = tcp_hdr(skb);
114 
115 	return __cookie_v6_init_sequence(iph, th, mssp);
116 }
117 
118 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th)
119 {
120 	__u32 cookie = ntohl(th->ack_seq) - 1;
121 	__u32 seq = ntohl(th->seq) - 1;
122 	__u32 mssind;
123 
124 	mssind = check_tcp_syn_cookie(cookie, &iph->saddr, &iph->daddr,
125 				      th->source, th->dest, seq);
126 
127 	return mssind < ARRAY_SIZE(msstab) ? msstab[mssind] : 0;
128 }
129 EXPORT_SYMBOL_GPL(__cookie_v6_check);
130 
131 static struct request_sock *cookie_tcp_check(struct net *net, struct sock *sk,
132 					     struct sk_buff *skb)
133 {
134 	struct tcp_options_received tcp_opt;
135 	u32 tsoff = 0;
136 	int mss;
137 
138 	if (tcp_synq_no_recent_overflow(sk))
139 		goto out;
140 
141 	mss = __cookie_v6_check(ipv6_hdr(skb), tcp_hdr(skb));
142 	if (!mss) {
143 		__NET_INC_STATS(net, LINUX_MIB_SYNCOOKIESFAILED);
144 		goto out;
145 	}
146 
147 	__NET_INC_STATS(net, LINUX_MIB_SYNCOOKIESRECV);
148 
149 	/* check for timestamp cookie support */
150 	memset(&tcp_opt, 0, sizeof(tcp_opt));
151 	tcp_parse_options(net, skb, &tcp_opt, 0, NULL);
152 
153 	if (tcp_opt.saw_tstamp && tcp_opt.rcv_tsecr) {
154 		tsoff = secure_tcpv6_ts_off(net,
155 					    ipv6_hdr(skb)->daddr.s6_addr32,
156 					    ipv6_hdr(skb)->saddr.s6_addr32);
157 		tcp_opt.rcv_tsecr -= tsoff;
158 	}
159 
160 	if (!cookie_timestamp_decode(net, &tcp_opt))
161 		goto out;
162 
163 	return cookie_tcp_reqsk_alloc(&tcp6_request_sock_ops, sk, skb,
164 				      &tcp_opt, mss, tsoff);
165 out:
166 	return ERR_PTR(-EINVAL);
167 }
168 
169 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb)
170 {
171 	const struct tcphdr *th = tcp_hdr(skb);
172 	struct ipv6_pinfo *np = inet6_sk(sk);
173 	struct tcp_sock *tp = tcp_sk(sk);
174 	struct inet_request_sock *ireq;
175 	struct net *net = sock_net(sk);
176 	struct request_sock *req;
177 	struct dst_entry *dst;
178 	struct sock *ret = sk;
179 	__u8 rcv_wscale;
180 	int full_space;
181 	SKB_DR(reason);
182 
183 	if (!READ_ONCE(net->ipv4.sysctl_tcp_syncookies) ||
184 	    !th->ack || th->rst)
185 		goto out;
186 
187 	if (cookie_bpf_ok(skb)) {
188 		req = cookie_bpf_check(sk, skb);
189 	} else {
190 		req = cookie_tcp_check(net, sk, skb);
191 		if (IS_ERR(req))
192 			goto out;
193 	}
194 	if (!req) {
195 		SKB_DR_SET(reason, NO_SOCKET);
196 		goto out_drop;
197 	}
198 
199 	ireq = inet_rsk(req);
200 
201 	ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr;
202 	ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr;
203 
204 	if (security_inet_conn_request(sk, skb, req)) {
205 		SKB_DR_SET(reason, SECURITY_HOOK);
206 		goto out_free;
207 	}
208 
209 	if (ipv6_opt_accepted(sk, skb, &TCP_SKB_CB(skb)->header.h6) ||
210 	    np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo ||
211 	    np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim) {
212 		refcount_inc(&skb->users);
213 		ireq->pktopts = skb;
214 	}
215 
216 	/* So that link locals have meaning */
217 	if (!sk->sk_bound_dev_if &&
218 	    ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL)
219 		ireq->ir_iif = tcp_v6_iif(skb);
220 
221 	tcp_ao_syncookie(sk, skb, req, AF_INET6);
222 
223 	/*
224 	 * We need to lookup the dst_entry to get the correct window size.
225 	 * This is taken from tcp_v6_syn_recv_sock.  Somebody please enlighten
226 	 * me if there is a preferred way.
227 	 */
228 	{
229 		struct in6_addr *final_p, final;
230 		struct flowi6 fl6;
231 		memset(&fl6, 0, sizeof(fl6));
232 		fl6.flowi6_proto = IPPROTO_TCP;
233 		fl6.daddr = ireq->ir_v6_rmt_addr;
234 		final_p = fl6_update_dst(&fl6, rcu_dereference(np->opt), &final);
235 		fl6.saddr = ireq->ir_v6_loc_addr;
236 		fl6.flowi6_oif = ireq->ir_iif;
237 		fl6.flowi6_mark = ireq->ir_mark;
238 		fl6.fl6_dport = ireq->ir_rmt_port;
239 		fl6.fl6_sport = inet_sk(sk)->inet_sport;
240 		fl6.flowi6_uid = sk_uid(sk);
241 		security_req_classify_flow(req, flowi6_to_flowi_common(&fl6));
242 
243 		dst = ip6_dst_lookup_flow(net, sk, &fl6, final_p);
244 		if (IS_ERR(dst)) {
245 			SKB_DR_SET(reason, IP_OUTNOROUTES);
246 			goto out_free;
247 		}
248 	}
249 
250 	req->rsk_window_clamp = READ_ONCE(tp->window_clamp) ? :dst_metric(dst, RTAX_WINDOW);
251 	/* limit the window selection if the user enforce a smaller rx buffer */
252 	full_space = tcp_full_space(sk);
253 	if (sk->sk_userlocks & SOCK_RCVBUF_LOCK &&
254 	    (req->rsk_window_clamp > full_space || req->rsk_window_clamp == 0))
255 		req->rsk_window_clamp = full_space;
256 
257 	tcp_select_initial_window(sk, full_space, req->mss,
258 				  &req->rsk_rcv_wnd, &req->rsk_window_clamp,
259 				  ireq->wscale_ok, &rcv_wscale,
260 				  dst_metric(dst, RTAX_INITRWND));
261 
262 	/* req->syncookie is set true only if ACK is validated
263 	 * by BPF kfunc, then, rcv_wscale is already configured.
264 	 */
265 	if (!req->syncookie)
266 		ireq->rcv_wscale = rcv_wscale;
267 	ireq->ecn_ok &= cookie_ecn_ok(net, dst);
268 	tcp_rsk(req)->accecn_ok = ireq->ecn_ok && cookie_accecn_ok(th);
269 
270 	ret = tcp_get_cookie_sock(sk, skb, req, dst);
271 	if (!ret) {
272 		SKB_DR_SET(reason, NO_SOCKET);
273 		goto out_drop;
274 	}
275 out:
276 	return ret;
277 out_free:
278 	reqsk_free(req);
279 out_drop:
280 	sk_skb_reason_drop(sk, skb, reason);
281 	return NULL;
282 }
283