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(&combined, offsetofend(typeof(combined), dport), 60 &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, 0, 0) + 70 sseq + (count << COOKIEBITS) + 71 ((cookie_hash(saddr, daddr, sport, dport, count, 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, 0, 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 - diff, 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(&iph->saddr, &iph->daddr, th->source, 105 th->dest, ntohl(th->seq), 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, &iph->saddr, &iph->daddr, 122 th->source, th->dest, 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); 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; 143 144 if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies || !th->ack || th->rst) 145 goto out; 146 147 if (tcp_synq_no_recent_overflow(sk)) 148 goto out; 149 150 mss = __cookie_v6_check(ipv6_hdr(skb), th, cookie); 151 if (mss == 0) { 152 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESFAILED); 153 goto out; 154 } 155 156 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESRECV); 157 158 /* check for timestamp cookie support */ 159 memset(&tcp_opt, 0, sizeof(tcp_opt)); 160 tcp_parse_options(sock_net(sk), skb, &tcp_opt, 0, NULL); 161 162 if (tcp_opt.saw_tstamp && tcp_opt.rcv_tsecr) { 163 tsoff = secure_tcpv6_ts_off(sock_net(sk), 164 ipv6_hdr(skb)->daddr.s6_addr32, 165 ipv6_hdr(skb)->saddr.s6_addr32); 166 tcp_opt.rcv_tsecr -= tsoff; 167 } 168 169 if (!cookie_timestamp_decode(sock_net(sk), &tcp_opt)) 170 goto out; 171 172 ret = NULL; 173 req = cookie_tcp_reqsk_alloc(&tcp6_request_sock_ops, sk, skb); 174 if (!req) 175 goto out; 176 177 ireq = inet_rsk(req); 178 treq = tcp_rsk(req); 179 treq->tfo_listener = false; 180 181 if (security_inet_conn_request(sk, skb, req)) 182 goto out_free; 183 184 req->mss = mss; 185 ireq->ir_rmt_port = th->source; 186 ireq->ir_num = ntohs(th->dest); 187 ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr; 188 ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr; 189 if (ipv6_opt_accepted(sk, skb, &TCP_SKB_CB(skb)->header.h6) || 190 np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo || 191 np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim) { 192 refcount_inc(&skb->users); 193 ireq->pktopts = skb; 194 } 195 196 ireq->ir_iif = inet_request_bound_dev_if(sk, skb); 197 /* So that link locals have meaning */ 198 if (!sk->sk_bound_dev_if && 199 ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL) 200 ireq->ir_iif = tcp_v6_iif(skb); 201 202 ireq->ir_mark = inet_request_mark(sk, skb); 203 204 req->num_retrans = 0; 205 ireq->snd_wscale = tcp_opt.snd_wscale; 206 ireq->sack_ok = tcp_opt.sack_ok; 207 ireq->wscale_ok = tcp_opt.wscale_ok; 208 ireq->tstamp_ok = tcp_opt.saw_tstamp; 209 req->ts_recent = tcp_opt.saw_tstamp ? tcp_opt.rcv_tsval : 0; 210 treq->snt_synack = 0; 211 treq->rcv_isn = ntohl(th->seq) - 1; 212 treq->snt_isn = cookie; 213 treq->ts_off = 0; 214 treq->txhash = net_tx_rndhash(); 215 if (IS_ENABLED(CONFIG_SMC)) 216 ireq->smc_ok = 0; 217 218 /* 219 * We need to lookup the dst_entry to get the correct window size. 220 * This is taken from tcp_v6_syn_recv_sock. Somebody please enlighten 221 * me if there is a preferred way. 222 */ 223 { 224 struct in6_addr *final_p, final; 225 struct flowi6 fl6; 226 memset(&fl6, 0, sizeof(fl6)); 227 fl6.flowi6_proto = IPPROTO_TCP; 228 fl6.daddr = ireq->ir_v6_rmt_addr; 229 final_p = fl6_update_dst(&fl6, rcu_dereference(np->opt), &final); 230 fl6.saddr = ireq->ir_v6_loc_addr; 231 fl6.flowi6_oif = ireq->ir_iif; 232 fl6.flowi6_mark = ireq->ir_mark; 233 fl6.fl6_dport = ireq->ir_rmt_port; 234 fl6.fl6_sport = inet_sk(sk)->inet_sport; 235 fl6.flowi6_uid = sk->sk_uid; 236 security_req_classify_flow(req, flowi6_to_flowi_common(&fl6)); 237 238 dst = ip6_dst_lookup_flow(sock_net(sk), sk, &fl6, final_p); 239 if (IS_ERR(dst)) 240 goto out_free; 241 } 242 243 req->rsk_window_clamp = tp->window_clamp ? :dst_metric(dst, RTAX_WINDOW); 244 /* limit the window selection if the user enforce a smaller rx buffer */ 245 full_space = tcp_full_space(sk); 246 if (sk->sk_userlocks & SOCK_RCVBUF_LOCK && 247 (req->rsk_window_clamp > full_space || req->rsk_window_clamp == 0)) 248 req->rsk_window_clamp = full_space; 249 250 tcp_select_initial_window(sk, full_space, req->mss, 251 &req->rsk_rcv_wnd, &req->rsk_window_clamp, 252 ireq->wscale_ok, &rcv_wscale, 253 dst_metric(dst, RTAX_INITRWND)); 254 255 ireq->rcv_wscale = rcv_wscale; 256 ireq->ecn_ok = cookie_ecn_ok(&tcp_opt, sock_net(sk), dst); 257 258 ret = tcp_get_cookie_sock(sk, skb, req, dst, tsoff); 259 out: 260 return ret; 261 out_free: 262 reqsk_free(req); 263 return NULL; 264 } 265