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 | |
20 | #define COOKIEBITS 24 /* Upper bits store count */ |
21 | #define COOKIEMASK (((__u32)1 << COOKIEBITS) - 1) |
22 | |
23 | static siphash_aligned_key_t syncookie6_secret[2]; |
24 | |
25 | /* RFC 2460, Section 8.3: |
26 | * [ipv6 tcp] MSS must be computed as the maximum packet size minus 60 [..] |
27 | * |
28 | * Due to IPV6_MIN_MTU=1280 the lowest possible MSS is 1220, which allows |
29 | * using higher values than ipv4 tcp syncookies. |
30 | * The other values are chosen based on ethernet (1500 and 9k MTU), plus |
31 | * one that accounts for common encap (PPPoe) overhead. Table must be sorted. |
32 | */ |
33 | static __u16 const msstab[] = { |
34 | 1280 - 60, /* IPV6_MIN_MTU - 60 */ |
35 | 1480 - 60, |
36 | 1500 - 60, |
37 | 9000 - 60, |
38 | }; |
39 | |
40 | static u32 cookie_hash(const struct in6_addr *saddr, |
41 | const struct in6_addr *daddr, |
42 | __be16 sport, __be16 dport, u32 count, int c) |
43 | { |
44 | const struct { |
45 | struct in6_addr saddr; |
46 | struct in6_addr daddr; |
47 | u32 count; |
48 | __be16 sport; |
49 | __be16 dport; |
50 | } __aligned(SIPHASH_ALIGNMENT) combined = { |
51 | .saddr = *saddr, |
52 | .daddr = *daddr, |
53 | .count = count, |
54 | .sport = sport, |
55 | .dport = dport |
56 | }; |
57 | |
58 | net_get_random_once(syncookie6_secret, sizeof(syncookie6_secret)); |
59 | return siphash(data: &combined, offsetofend(typeof(combined), dport), |
60 | key: &syncookie6_secret[c]); |
61 | } |
62 | |
63 | static __u32 secure_tcp_syn_cookie(const struct in6_addr *saddr, |
64 | const struct in6_addr *daddr, |
65 | __be16 sport, __be16 dport, __u32 sseq, |
66 | __u32 data) |
67 | { |
68 | u32 count = tcp_cookie_time(); |
69 | return (cookie_hash(saddr, daddr, sport, dport, count: 0, c: 0) + |
70 | sseq + (count << COOKIEBITS) + |
71 | ((cookie_hash(saddr, daddr, sport, dport, count, c: 1) + data) |
72 | & COOKIEMASK)); |
73 | } |
74 | |
75 | static __u32 check_tcp_syn_cookie(__u32 cookie, const struct in6_addr *saddr, |
76 | const struct in6_addr *daddr, __be16 sport, |
77 | __be16 dport, __u32 sseq) |
78 | { |
79 | __u32 diff, count = tcp_cookie_time(); |
80 | |
81 | cookie -= cookie_hash(saddr, daddr, sport, dport, count: 0, c: 0) + sseq; |
82 | |
83 | diff = (count - (cookie >> COOKIEBITS)) & ((__u32) -1 >> COOKIEBITS); |
84 | if (diff >= MAX_SYNCOOKIE_AGE) |
85 | return (__u32)-1; |
86 | |
87 | return (cookie - |
88 | cookie_hash(saddr, daddr, sport, dport, count: count - diff, c: 1)) |
89 | & COOKIEMASK; |
90 | } |
91 | |
92 | u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, |
93 | const struct tcphdr *th, __u16 *mssp) |
94 | { |
95 | int mssind; |
96 | const __u16 mss = *mssp; |
97 | |
98 | for (mssind = ARRAY_SIZE(msstab) - 1; mssind ; mssind--) |
99 | if (mss >= msstab[mssind]) |
100 | break; |
101 | |
102 | *mssp = msstab[mssind]; |
103 | |
104 | return secure_tcp_syn_cookie(saddr: &iph->saddr, daddr: &iph->daddr, sport: th->source, |
105 | dport: th->dest, ntohl(th->seq), data: mssind); |
106 | } |
107 | EXPORT_SYMBOL_GPL(__cookie_v6_init_sequence); |
108 | |
109 | __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mssp) |
110 | { |
111 | const struct ipv6hdr *iph = ipv6_hdr(skb); |
112 | const struct tcphdr *th = tcp_hdr(skb); |
113 | |
114 | return __cookie_v6_init_sequence(iph, th, mssp); |
115 | } |
116 | |
117 | int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th, |
118 | __u32 cookie) |
119 | { |
120 | __u32 seq = ntohl(th->seq) - 1; |
121 | __u32 mssind = check_tcp_syn_cookie(cookie, saddr: &iph->saddr, daddr: &iph->daddr, |
122 | sport: th->source, dport: th->dest, sseq: seq); |
123 | |
124 | return mssind < ARRAY_SIZE(msstab) ? msstab[mssind] : 0; |
125 | } |
126 | EXPORT_SYMBOL_GPL(__cookie_v6_check); |
127 | |
128 | struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb) |
129 | { |
130 | struct tcp_options_received tcp_opt; |
131 | struct inet_request_sock *ireq; |
132 | struct tcp_request_sock *treq; |
133 | struct ipv6_pinfo *np = inet6_sk(sk: sk); |
134 | struct tcp_sock *tp = tcp_sk(sk); |
135 | const struct tcphdr *th = tcp_hdr(skb); |
136 | __u32 cookie = ntohl(th->ack_seq) - 1; |
137 | struct sock *ret = sk; |
138 | struct request_sock *req; |
139 | int full_space, mss; |
140 | struct dst_entry *dst; |
141 | __u8 rcv_wscale; |
142 | u32 tsoff = 0; |
143 | int l3index; |
144 | |
145 | if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies) || |
146 | !th->ack || th->rst) |
147 | goto out; |
148 | |
149 | if (tcp_synq_no_recent_overflow(sk)) |
150 | goto out; |
151 | |
152 | mss = __cookie_v6_check(ipv6_hdr(skb), th, cookie); |
153 | if (mss == 0) { |
154 | __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESFAILED); |
155 | goto out; |
156 | } |
157 | |
158 | __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESRECV); |
159 | |
160 | /* check for timestamp cookie support */ |
161 | memset(&tcp_opt, 0, sizeof(tcp_opt)); |
162 | tcp_parse_options(net: sock_net(sk), skb, opt_rx: &tcp_opt, estab: 0, NULL); |
163 | |
164 | if (tcp_opt.saw_tstamp && tcp_opt.rcv_tsecr) { |
165 | tsoff = secure_tcpv6_ts_off(net: sock_net(sk), |
166 | saddr: ipv6_hdr(skb)->daddr.s6_addr32, |
167 | daddr: ipv6_hdr(skb)->saddr.s6_addr32); |
168 | tcp_opt.rcv_tsecr -= tsoff; |
169 | } |
170 | |
171 | if (!cookie_timestamp_decode(net: sock_net(sk), opt: &tcp_opt)) |
172 | goto out; |
173 | |
174 | ret = NULL; |
175 | req = cookie_tcp_reqsk_alloc(ops: &tcp6_request_sock_ops, |
176 | af_ops: &tcp_request_sock_ipv6_ops, sk, skb); |
177 | if (!req) |
178 | goto out; |
179 | |
180 | ireq = inet_rsk(sk: req); |
181 | treq = tcp_rsk(req); |
182 | treq->tfo_listener = false; |
183 | |
184 | if (security_inet_conn_request(sk, skb, req)) |
185 | goto out_free; |
186 | |
187 | req->mss = mss; |
188 | ireq->ir_rmt_port = th->source; |
189 | ireq->ir_num = ntohs(th->dest); |
190 | ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr; |
191 | ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr; |
192 | if (ipv6_opt_accepted(sk, skb, opt: &TCP_SKB_CB(skb)->header.h6) || |
193 | np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo || |
194 | np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim) { |
195 | refcount_inc(r: &skb->users); |
196 | ireq->pktopts = skb; |
197 | } |
198 | |
199 | ireq->ir_iif = inet_request_bound_dev_if(sk, skb); |
200 | /* So that link locals have meaning */ |
201 | if (!sk->sk_bound_dev_if && |
202 | ipv6_addr_type(addr: &ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL) |
203 | ireq->ir_iif = tcp_v6_iif(skb); |
204 | |
205 | ireq->ir_mark = inet_request_mark(sk, skb); |
206 | |
207 | req->num_retrans = 0; |
208 | ireq->snd_wscale = tcp_opt.snd_wscale; |
209 | ireq->sack_ok = tcp_opt.sack_ok; |
210 | ireq->wscale_ok = tcp_opt.wscale_ok; |
211 | ireq->tstamp_ok = tcp_opt.saw_tstamp; |
212 | req->ts_recent = tcp_opt.saw_tstamp ? tcp_opt.rcv_tsval : 0; |
213 | treq->snt_synack = 0; |
214 | treq->rcv_isn = ntohl(th->seq) - 1; |
215 | treq->snt_isn = cookie; |
216 | treq->ts_off = 0; |
217 | treq->txhash = net_tx_rndhash(); |
218 | |
219 | l3index = l3mdev_master_ifindex_by_index(net: sock_net(sk), ifindex: ireq->ir_iif); |
220 | tcp_ao_syncookie(sk, skb, treq, AF_INET6, l3index); |
221 | |
222 | if (IS_ENABLED(CONFIG_SMC)) |
223 | ireq->smc_ok = 0; |
224 | |
225 | /* |
226 | * We need to lookup the dst_entry to get the correct window size. |
227 | * This is taken from tcp_v6_syn_recv_sock. Somebody please enlighten |
228 | * me if there is a preferred way. |
229 | */ |
230 | { |
231 | struct in6_addr *final_p, final; |
232 | struct flowi6 fl6; |
233 | memset(&fl6, 0, sizeof(fl6)); |
234 | fl6.flowi6_proto = IPPROTO_TCP; |
235 | fl6.daddr = ireq->ir_v6_rmt_addr; |
236 | final_p = fl6_update_dst(fl6: &fl6, rcu_dereference(np->opt), orig: &final); |
237 | fl6.saddr = ireq->ir_v6_loc_addr; |
238 | fl6.flowi6_oif = ireq->ir_iif; |
239 | fl6.flowi6_mark = ireq->ir_mark; |
240 | fl6.fl6_dport = ireq->ir_rmt_port; |
241 | fl6.fl6_sport = inet_sk(sk)->inet_sport; |
242 | fl6.flowi6_uid = sk->sk_uid; |
243 | security_req_classify_flow(req, flic: flowi6_to_flowi_common(fl6: &fl6)); |
244 | |
245 | dst = ip6_dst_lookup_flow(net: sock_net(sk), sk, fl6: &fl6, final_dst: final_p); |
246 | if (IS_ERR(ptr: dst)) |
247 | goto out_free; |
248 | } |
249 | |
250 | req->rsk_window_clamp = 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, space: full_space, mss: req->mss, |
258 | rcv_wnd: &req->rsk_rcv_wnd, window_clamp: &req->rsk_window_clamp, |
259 | wscale_ok: ireq->wscale_ok, rcv_wscale: &rcv_wscale, |
260 | init_rcv_wnd: dst_metric(dst, RTAX_INITRWND)); |
261 | |
262 | ireq->rcv_wscale = rcv_wscale; |
263 | ireq->ecn_ok = cookie_ecn_ok(opt: &tcp_opt, net: sock_net(sk), dst); |
264 | |
265 | ret = tcp_get_cookie_sock(sk, skb, req, dst, tsoff); |
266 | out: |
267 | return ret; |
268 | out_free: |
269 | reqsk_free(req); |
270 | return NULL; |
271 | } |
272 | |