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 __u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS]; 29 EXPORT_SYMBOL(syncookie_secret); 30 31 static __init int init_syncookies(void) 32 { 33 get_random_bytes(syncookie_secret, sizeof(syncookie_secret)); 34 return 0; 35 } 36 __initcall(init_syncookies); 37 38 #define COOKIEBITS 24 /* Upper bits store count */ 39 #define COOKIEMASK (((__u32)1 << COOKIEBITS) - 1) 40 41 static DEFINE_PER_CPU(__u32 [16 + 5 + SHA_WORKSPACE_WORDS], 42 ipv4_cookie_scratch); 43 44 static u32 cookie_hash(__be32 saddr, __be32 daddr, __be16 sport, __be16 dport, 45 u32 count, int c) 46 { 47 __u32 *tmp = __get_cpu_var(ipv4_cookie_scratch); 48 49 memcpy(tmp + 4, syncookie_secret[c], sizeof(syncookie_secret[c])); 50 tmp[0] = (__force u32)saddr; 51 tmp[1] = (__force u32)daddr; 52 tmp[2] = ((__force u32)sport << 16) + (__force u32)dport; 53 tmp[3] = count; 54 sha_transform(tmp + 16, (__u8 *)tmp, tmp + 16 + 5); 55 56 return tmp[17]; 57 } 58 59 60 /* 61 * when syncookies are in effect and tcp timestamps are enabled we encode 62 * tcp options in the lower bits of the timestamp value that will be 63 * sent in the syn-ack. 64 * Since subsequent timestamps use the normal tcp_time_stamp value, we 65 * must make sure that the resulting initial timestamp is <= tcp_time_stamp. 66 */ 67 __u32 cookie_init_timestamp(struct request_sock *req) 68 { 69 struct inet_request_sock *ireq; 70 u32 ts, ts_now = tcp_time_stamp; 71 u32 options = 0; 72 73 ireq = inet_rsk(req); 74 75 options = ireq->wscale_ok ? ireq->snd_wscale : 0xf; 76 options |= ireq->sack_ok << 4; 77 options |= ireq->ecn_ok << 5; 78 79 ts = ts_now & ~TSMASK; 80 ts |= options; 81 if (ts > ts_now) { 82 ts >>= TSBITS; 83 ts--; 84 ts <<= TSBITS; 85 ts |= options; 86 } 87 return ts; 88 } 89 90 91 static __u32 secure_tcp_syn_cookie(__be32 saddr, __be32 daddr, __be16 sport, 92 __be16 dport, __u32 sseq, __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 u32 count = tcp_cookie_time(); 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 less than 117 * MAX_SYNCOOKIE_AGE minutes in the past. 118 * The return value (__u32)-1 if this test fails. 119 */ 120 static __u32 check_tcp_syn_cookie(__u32 cookie, __be32 saddr, __be32 daddr, 121 __be16 sport, __be16 dport, __u32 sseq) 122 { 123 u32 diff, count = tcp_cookie_time(); 124 125 /* Strip away the layers from the cookie */ 126 cookie -= cookie_hash(saddr, daddr, sport, dport, 0, 0) + sseq; 127 128 /* Cookie is now reduced to (count * 2^24) ^ (hash % 2^24) */ 129 diff = (count - (cookie >> COOKIEBITS)) & ((__u32) - 1 >> COOKIEBITS); 130 if (diff >= MAX_SYNCOOKIE_AGE) 131 return (__u32)-1; 132 133 return (cookie - 134 cookie_hash(saddr, daddr, sport, dport, count - diff, 1)) 135 & COOKIEMASK; /* Leaving the data behind */ 136 } 137 138 /* 139 * MSS Values are chosen based on the 2011 paper 140 * 'An Analysis of TCP Maximum Segement Sizes' by S. Alcock and R. Nelson. 141 * Values .. 142 * .. lower than 536 are rare (< 0.2%) 143 * .. between 537 and 1299 account for less than < 1.5% of observed values 144 * .. in the 1300-1349 range account for about 15 to 20% of observed mss values 145 * .. exceeding 1460 are very rare (< 0.04%) 146 * 147 * 1460 is the single most frequently announced mss value (30 to 46% depending 148 * on monitor location). Table must be sorted. 149 */ 150 static __u16 const msstab[] = { 151 536, 152 1300, 153 1440, /* 1440, 1452: PPPoE */ 154 1460, 155 }; 156 157 /* 158 * Generate a syncookie. mssp points to the mss, which is returned 159 * rounded down to the value encoded in the cookie. 160 */ 161 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 162 u16 *mssp) 163 { 164 int mssind; 165 const __u16 mss = *mssp; 166 167 for (mssind = ARRAY_SIZE(msstab) - 1; mssind ; mssind--) 168 if (mss >= msstab[mssind]) 169 break; 170 *mssp = msstab[mssind]; 171 172 return secure_tcp_syn_cookie(iph->saddr, iph->daddr, 173 th->source, th->dest, ntohl(th->seq), 174 mssind); 175 } 176 EXPORT_SYMBOL_GPL(__cookie_v4_init_sequence); 177 178 __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb, __u16 *mssp) 179 { 180 const struct iphdr *iph = ip_hdr(skb); 181 const struct tcphdr *th = tcp_hdr(skb); 182 183 tcp_synq_overflow(sk); 184 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT); 185 186 return __cookie_v4_init_sequence(iph, th, mssp); 187 } 188 189 /* 190 * Check if a ack sequence number is a valid syncookie. 191 * Return the decoded mss if it is, or 0 if not. 192 */ 193 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th, 194 u32 cookie) 195 { 196 __u32 seq = ntohl(th->seq) - 1; 197 __u32 mssind = check_tcp_syn_cookie(cookie, iph->saddr, iph->daddr, 198 th->source, th->dest, seq); 199 200 return mssind < ARRAY_SIZE(msstab) ? msstab[mssind] : 0; 201 } 202 EXPORT_SYMBOL_GPL(__cookie_v4_check); 203 204 static inline struct sock *get_cookie_sock(struct sock *sk, struct sk_buff *skb, 205 struct request_sock *req, 206 struct dst_entry *dst) 207 { 208 struct inet_connection_sock *icsk = inet_csk(sk); 209 struct sock *child; 210 211 child = icsk->icsk_af_ops->syn_recv_sock(sk, skb, req, dst); 212 if (child) 213 inet_csk_reqsk_queue_add(sk, req, child); 214 else 215 reqsk_free(req); 216 217 return child; 218 } 219 220 221 /* 222 * when syncookies are in effect and tcp timestamps are enabled we stored 223 * additional tcp options in the timestamp. 224 * This extracts these options from the timestamp echo. 225 * 226 * The lowest 4 bits store snd_wscale. 227 * next 2 bits indicate SACK and ECN support. 228 * 229 * return false if we decode an option that should not be. 230 */ 231 bool cookie_check_timestamp(struct tcp_options_received *tcp_opt, 232 struct net *net, bool *ecn_ok) 233 { 234 /* echoed timestamp, lowest bits contain options */ 235 u32 options = tcp_opt->rcv_tsecr & TSMASK; 236 237 if (!tcp_opt->saw_tstamp) { 238 tcp_clear_options(tcp_opt); 239 return true; 240 } 241 242 if (!sysctl_tcp_timestamps) 243 return false; 244 245 tcp_opt->sack_ok = (options & (1 << 4)) ? TCP_SACK_SEEN : 0; 246 *ecn_ok = (options >> 5) & 1; 247 if (*ecn_ok && !net->ipv4.sysctl_tcp_ecn) 248 return false; 249 250 if (tcp_opt->sack_ok && !sysctl_tcp_sack) 251 return false; 252 253 if ((options & 0xf) == 0xf) 254 return true; /* no window scaling */ 255 256 tcp_opt->wscale_ok = 1; 257 tcp_opt->snd_wscale = options & 0xf; 258 return sysctl_tcp_window_scaling != 0; 259 } 260 EXPORT_SYMBOL(cookie_check_timestamp); 261 262 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb, 263 struct ip_options *opt) 264 { 265 struct tcp_options_received tcp_opt; 266 struct inet_request_sock *ireq; 267 struct tcp_request_sock *treq; 268 struct tcp_sock *tp = tcp_sk(sk); 269 const struct tcphdr *th = tcp_hdr(skb); 270 __u32 cookie = ntohl(th->ack_seq) - 1; 271 struct sock *ret = sk; 272 struct request_sock *req; 273 int mss; 274 struct rtable *rt; 275 __u8 rcv_wscale; 276 bool ecn_ok = false; 277 struct flowi4 fl4; 278 279 if (!sysctl_tcp_syncookies || !th->ack || th->rst) 280 goto out; 281 282 if (tcp_synq_no_recent_overflow(sk) || 283 (mss = __cookie_v4_check(ip_hdr(skb), th, cookie)) == 0) { 284 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SYNCOOKIESFAILED); 285 goto out; 286 } 287 288 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SYNCOOKIESRECV); 289 290 /* check for timestamp cookie support */ 291 memset(&tcp_opt, 0, sizeof(tcp_opt)); 292 tcp_parse_options(skb, &tcp_opt, 0, NULL); 293 294 if (!cookie_check_timestamp(&tcp_opt, sock_net(sk), &ecn_ok)) 295 goto out; 296 297 ret = NULL; 298 req = inet_reqsk_alloc(&tcp_request_sock_ops); /* for safety */ 299 if (!req) 300 goto out; 301 302 ireq = inet_rsk(req); 303 treq = tcp_rsk(req); 304 treq->rcv_isn = ntohl(th->seq) - 1; 305 treq->snt_isn = cookie; 306 req->mss = mss; 307 ireq->ir_num = ntohs(th->dest); 308 ireq->ir_rmt_port = th->source; 309 ireq->ir_loc_addr = ip_hdr(skb)->daddr; 310 ireq->ir_rmt_addr = ip_hdr(skb)->saddr; 311 ireq->ecn_ok = ecn_ok; 312 ireq->snd_wscale = tcp_opt.snd_wscale; 313 ireq->sack_ok = tcp_opt.sack_ok; 314 ireq->wscale_ok = tcp_opt.wscale_ok; 315 ireq->tstamp_ok = tcp_opt.saw_tstamp; 316 req->ts_recent = tcp_opt.saw_tstamp ? tcp_opt.rcv_tsval : 0; 317 treq->snt_synack = tcp_opt.saw_tstamp ? tcp_opt.rcv_tsecr : 0; 318 treq->listener = NULL; 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_rcu) + opt->optlen; 325 326 ireq->opt = kmalloc(opt_size, GFP_ATOMIC); 327 if (ireq->opt != NULL && ip_options_echo(&ireq->opt->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->num_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 flowi4_init_output(&fl4, sk->sk_bound_dev_if, sk->sk_mark, 348 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE, IPPROTO_TCP, 349 inet_sk_flowi_flags(sk), 350 (opt && opt->srr) ? opt->faddr : ireq->ir_rmt_addr, 351 ireq->ir_loc_addr, th->source, th->dest); 352 security_req_classify_flow(req, flowi4_to_flowi(&fl4)); 353 rt = ip_route_output_key(sock_net(sk), &fl4); 354 if (IS_ERR(rt)) { 355 reqsk_free(req); 356 goto out; 357 } 358 359 /* Try to redo what tcp_v4_send_synack did. */ 360 req->window_clamp = tp->window_clamp ? :dst_metric(&rt->dst, RTAX_WINDOW); 361 362 tcp_select_initial_window(tcp_full_space(sk), req->mss, 363 &req->rcv_wnd, &req->window_clamp, 364 ireq->wscale_ok, &rcv_wscale, 365 dst_metric(&rt->dst, RTAX_INITRWND)); 366 367 ireq->rcv_wscale = rcv_wscale; 368 369 ret = get_cookie_sock(sk, skb, req, &rt->dst); 370 /* ip_queue_xmit() depends on our flow being setup 371 * Normal sockets get it right from inet_csk_route_child_sock() 372 */ 373 if (ret) 374 inet_sk(ret)->cork.fl.u.ip4 = fl4; 375 out: return ret; 376 } 377