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 22extern int sysctl_tcp_syncookies; 23 24static u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS] __read_mostly; 25 26#define COOKIEBITS 24 /* Upper bits store count */ 27#define COOKIEMASK (((__u32)1 << COOKIEBITS) - 1) 28 29/* TCP Timestamp: 6 lowest bits of timestamp sent in the cookie SYN-ACK 30 * stores TCP options: 31 * 32 * MSB LSB 33 * | 31 ... 6 | 5 | 4 | 3 2 1 0 | 34 * | Timestamp | ECN | SACK | WScale | 35 * 36 * When we receive a valid cookie-ACK, we look at the echoed tsval (if 37 * any) to figure out which TCP options we should use for the rebuilt 38 * connection. 39 * 40 * A WScale setting of '0xf' (which is an invalid scaling value) 41 * means that original syn did not include the TCP window scaling option. 42 */ 43#define TS_OPT_WSCALE_MASK 0xf 44#define TS_OPT_SACK BIT(4) 45#define TS_OPT_ECN BIT(5) 46/* There is no TS_OPT_TIMESTAMP: 47 * if ACK contains timestamp option, we already know it was 48 * requested/supported by the syn/synack exchange. 49 */ 50#define TSBITS 6 51#define TSMASK (((__u32)1 << TSBITS) - 1) 52 53static DEFINE_PER_CPU(__u32 [16 + 5 + SHA_WORKSPACE_WORDS], 54 ipv4_cookie_scratch); 55 56static u32 cookie_hash(__be32 saddr, __be32 daddr, __be16 sport, __be16 dport, 57 u32 count, int c) 58{ 59 __u32 *tmp; 60 61 net_get_random_once(syncookie_secret, sizeof(syncookie_secret)); 62 63 tmp = this_cpu_ptr(ipv4_cookie_scratch); 64 memcpy(tmp + 4, syncookie_secret[c], sizeof(syncookie_secret[c])); 65 tmp[0] = (__force u32)saddr; 66 tmp[1] = (__force u32)daddr; 67 tmp[2] = ((__force u32)sport << 16) + (__force u32)dport; 68 tmp[3] = count; 69 sha_transform(tmp + 16, (__u8 *)tmp, tmp + 16 + 5); 70 71 return tmp[17]; 72} 73 74 75/* 76 * when syncookies are in effect and tcp timestamps are enabled we encode 77 * tcp options in the lower bits of the timestamp value that will be 78 * sent in the syn-ack. 79 * Since subsequent timestamps use the normal tcp_time_stamp value, we 80 * must make sure that the resulting initial timestamp is <= tcp_time_stamp. 81 */ 82__u32 cookie_init_timestamp(struct request_sock *req) 83{ 84 struct inet_request_sock *ireq; 85 u32 ts, ts_now = tcp_time_stamp; 86 u32 options = 0; 87 88 ireq = inet_rsk(req); 89 90 options = ireq->wscale_ok ? ireq->snd_wscale : TS_OPT_WSCALE_MASK; 91 if (ireq->sack_ok) 92 options |= TS_OPT_SACK; 93 if (ireq->ecn_ok) 94 options |= TS_OPT_ECN; 95 96 ts = ts_now & ~TSMASK; 97 ts |= options; 98 if (ts > ts_now) { 99 ts >>= TSBITS; 100 ts--; 101 ts <<= TSBITS; 102 ts |= options; 103 } 104 return ts; 105} 106 107 108static __u32 secure_tcp_syn_cookie(__be32 saddr, __be32 daddr, __be16 sport, 109 __be16 dport, __u32 sseq, __u32 data) 110{ 111 /* 112 * Compute the secure sequence number. 113 * The output should be: 114 * HASH(sec1,saddr,sport,daddr,dport,sec1) + sseq + (count * 2^24) 115 * + (HASH(sec2,saddr,sport,daddr,dport,count,sec2) % 2^24). 116 * Where sseq is their sequence number and count increases every 117 * minute by 1. 118 * As an extra hack, we add a small "data" value that encodes the 119 * MSS into the second hash value. 120 */ 121 u32 count = tcp_cookie_time(); 122 return (cookie_hash(saddr, daddr, sport, dport, 0, 0) + 123 sseq + (count << COOKIEBITS) + 124 ((cookie_hash(saddr, daddr, sport, dport, count, 1) + data) 125 & COOKIEMASK)); 126} 127 128/* 129 * This retrieves the small "data" value from the syncookie. 130 * If the syncookie is bad, the data returned will be out of 131 * range. This must be checked by the caller. 132 * 133 * The count value used to generate the cookie must be less than 134 * MAX_SYNCOOKIE_AGE minutes in the past. 135 * The return value (__u32)-1 if this test fails. 136 */ 137static __u32 check_tcp_syn_cookie(__u32 cookie, __be32 saddr, __be32 daddr, 138 __be16 sport, __be16 dport, __u32 sseq) 139{ 140 u32 diff, count = tcp_cookie_time(); 141 142 /* Strip away the layers from the cookie */ 143 cookie -= cookie_hash(saddr, daddr, sport, dport, 0, 0) + sseq; 144 145 /* Cookie is now reduced to (count * 2^24) ^ (hash % 2^24) */ 146 diff = (count - (cookie >> COOKIEBITS)) & ((__u32) -1 >> COOKIEBITS); 147 if (diff >= MAX_SYNCOOKIE_AGE) 148 return (__u32)-1; 149 150 return (cookie - 151 cookie_hash(saddr, daddr, sport, dport, count - diff, 1)) 152 & COOKIEMASK; /* Leaving the data behind */ 153} 154 155/* 156 * MSS Values are chosen based on the 2011 paper 157 * 'An Analysis of TCP Maximum Segement Sizes' by S. Alcock and R. Nelson. 158 * Values .. 159 * .. lower than 536 are rare (< 0.2%) 160 * .. between 537 and 1299 account for less than < 1.5% of observed values 161 * .. in the 1300-1349 range account for about 15 to 20% of observed mss values 162 * .. exceeding 1460 are very rare (< 0.04%) 163 * 164 * 1460 is the single most frequently announced mss value (30 to 46% depending 165 * on monitor location). Table must be sorted. 166 */ 167static __u16 const msstab[] = { 168 536, 169 1300, 170 1440, /* 1440, 1452: PPPoE */ 171 1460, 172}; 173 174/* 175 * Generate a syncookie. mssp points to the mss, which is returned 176 * rounded down to the value encoded in the cookie. 177 */ 178u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, 179 u16 *mssp) 180{ 181 int mssind; 182 const __u16 mss = *mssp; 183 184 for (mssind = ARRAY_SIZE(msstab) - 1; mssind ; mssind--) 185 if (mss >= msstab[mssind]) 186 break; 187 *mssp = msstab[mssind]; 188 189 return secure_tcp_syn_cookie(iph->saddr, iph->daddr, 190 th->source, th->dest, ntohl(th->seq), 191 mssind); 192} 193EXPORT_SYMBOL_GPL(__cookie_v4_init_sequence); 194 195__u32 cookie_v4_init_sequence(struct sock *sk, const struct sk_buff *skb, 196 __u16 *mssp) 197{ 198 const struct iphdr *iph = ip_hdr(skb); 199 const struct tcphdr *th = tcp_hdr(skb); 200 201 tcp_synq_overflow(sk); 202 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT); 203 204 return __cookie_v4_init_sequence(iph, th, mssp); 205} 206 207/* 208 * Check if a ack sequence number is a valid syncookie. 209 * Return the decoded mss if it is, or 0 if not. 210 */ 211int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th, 212 u32 cookie) 213{ 214 __u32 seq = ntohl(th->seq) - 1; 215 __u32 mssind = check_tcp_syn_cookie(cookie, iph->saddr, iph->daddr, 216 th->source, th->dest, seq); 217 218 return mssind < ARRAY_SIZE(msstab) ? msstab[mssind] : 0; 219} 220EXPORT_SYMBOL_GPL(__cookie_v4_check); 221 222static struct sock *get_cookie_sock(struct sock *sk, struct sk_buff *skb, 223 struct request_sock *req, 224 struct dst_entry *dst) 225{ 226 struct inet_connection_sock *icsk = inet_csk(sk); 227 struct sock *child; 228 229 child = icsk->icsk_af_ops->syn_recv_sock(sk, skb, req, dst); 230 if (child) { 231 atomic_set(&req->rsk_refcnt, 1); 232 inet_csk_reqsk_queue_add(sk, req, child); 233 } else { 234 reqsk_free(req); 235 } 236 return child; 237} 238 239 240/* 241 * when syncookies are in effect and tcp timestamps are enabled we stored 242 * additional tcp options in the timestamp. 243 * This extracts these options from the timestamp echo. 244 * 245 * return false if we decode a tcp option that is disabled 246 * on the host. 247 */ 248bool cookie_timestamp_decode(struct tcp_options_received *tcp_opt) 249{ 250 /* echoed timestamp, lowest bits contain options */ 251 u32 options = tcp_opt->rcv_tsecr; 252 253 if (!tcp_opt->saw_tstamp) { 254 tcp_clear_options(tcp_opt); 255 return true; 256 } 257 258 if (!sysctl_tcp_timestamps) 259 return false; 260 261 tcp_opt->sack_ok = (options & TS_OPT_SACK) ? TCP_SACK_SEEN : 0; 262 263 if (tcp_opt->sack_ok && !sysctl_tcp_sack) 264 return false; 265 266 if ((options & TS_OPT_WSCALE_MASK) == TS_OPT_WSCALE_MASK) 267 return true; /* no window scaling */ 268 269 tcp_opt->wscale_ok = 1; 270 tcp_opt->snd_wscale = options & TS_OPT_WSCALE_MASK; 271 272 return sysctl_tcp_window_scaling != 0; 273} 274EXPORT_SYMBOL(cookie_timestamp_decode); 275 276bool cookie_ecn_ok(const struct tcp_options_received *tcp_opt, 277 const struct net *net, const struct dst_entry *dst) 278{ 279 bool ecn_ok = tcp_opt->rcv_tsecr & TS_OPT_ECN; 280 281 if (!ecn_ok) 282 return false; 283 284 if (net->ipv4.sysctl_tcp_ecn) 285 return true; 286 287 return dst_feature(dst, RTAX_FEATURE_ECN); 288} 289EXPORT_SYMBOL(cookie_ecn_ok); 290 291struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb) 292{ 293 struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; 294 struct tcp_options_received tcp_opt; 295 struct inet_request_sock *ireq; 296 struct tcp_request_sock *treq; 297 struct tcp_sock *tp = tcp_sk(sk); 298 const struct tcphdr *th = tcp_hdr(skb); 299 __u32 cookie = ntohl(th->ack_seq) - 1; 300 struct sock *ret = sk; 301 struct request_sock *req; 302 int mss; 303 struct rtable *rt; 304 __u8 rcv_wscale; 305 struct flowi4 fl4; 306 307 if (!sysctl_tcp_syncookies || !th->ack || th->rst) 308 goto out; 309 310 if (tcp_synq_no_recent_overflow(sk)) 311 goto out; 312 313 mss = __cookie_v4_check(ip_hdr(skb), th, cookie); 314 if (mss == 0) { 315 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SYNCOOKIESFAILED); 316 goto out; 317 } 318 319 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SYNCOOKIESRECV); 320 321 /* check for timestamp cookie support */ 322 memset(&tcp_opt, 0, sizeof(tcp_opt)); 323 tcp_parse_options(skb, &tcp_opt, 0, NULL); 324 325 if (!cookie_timestamp_decode(&tcp_opt)) 326 goto out; 327 328 ret = NULL; 329 req = inet_reqsk_alloc(&tcp_request_sock_ops, sk); /* for safety */ 330 if (!req) 331 goto out; 332 333 ireq = inet_rsk(req); 334 treq = tcp_rsk(req); 335 treq->rcv_isn = ntohl(th->seq) - 1; 336 treq->snt_isn = cookie; 337 req->mss = mss; 338 ireq->ir_num = ntohs(th->dest); 339 ireq->ir_rmt_port = th->source; 340 sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr); 341 sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr); 342 ireq->ir_mark = inet_request_mark(sk, skb); 343 ireq->snd_wscale = tcp_opt.snd_wscale; 344 ireq->sack_ok = tcp_opt.sack_ok; 345 ireq->wscale_ok = tcp_opt.wscale_ok; 346 ireq->tstamp_ok = tcp_opt.saw_tstamp; 347 req->ts_recent = tcp_opt.saw_tstamp ? tcp_opt.rcv_tsval : 0; 348 treq->snt_synack = tcp_opt.saw_tstamp ? tcp_opt.rcv_tsecr : 0; 349 treq->tfo_listener = false; 350 351 ireq->ir_iif = sk->sk_bound_dev_if; 352 353 /* We throwed the options of the initial SYN away, so we hope 354 * the ACK carries the same options again (see RFC1122 4.2.3.8) 355 */ 356 ireq->opt = tcp_v4_save_options(skb); 357 358 if (security_inet_conn_request(sk, skb, req)) { 359 reqsk_free(req); 360 goto out; 361 } 362 363 req->num_retrans = 0; 364 365 /* 366 * We need to lookup the route here to get at the correct 367 * window size. We should better make sure that the window size 368 * hasn't changed since we received the original syn, but I see 369 * no easy way to do this. 370 */ 371 flowi4_init_output(&fl4, sk->sk_bound_dev_if, ireq->ir_mark, 372 RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE, IPPROTO_TCP, 373 inet_sk_flowi_flags(sk), 374 opt->srr ? opt->faddr : ireq->ir_rmt_addr, 375 ireq->ir_loc_addr, th->source, th->dest); 376 security_req_classify_flow(req, flowi4_to_flowi(&fl4)); 377 rt = ip_route_output_key(sock_net(sk), &fl4); 378 if (IS_ERR(rt)) { 379 reqsk_free(req); 380 goto out; 381 } 382 383 /* Try to redo what tcp_v4_send_synack did. */ 384 req->window_clamp = tp->window_clamp ? :dst_metric(&rt->dst, RTAX_WINDOW); 385 386 tcp_select_initial_window(tcp_full_space(sk), req->mss, 387 &req->rcv_wnd, &req->window_clamp, 388 ireq->wscale_ok, &rcv_wscale, 389 dst_metric(&rt->dst, RTAX_INITRWND)); 390 391 ireq->rcv_wscale = rcv_wscale; 392 ireq->ecn_ok = cookie_ecn_ok(&tcp_opt, sock_net(sk), &rt->dst); 393 394 ret = get_cookie_sock(sk, skb, req, &rt->dst); 395 /* ip_queue_xmit() depends on our flow being setup 396 * Normal sockets get it right from inet_csk_route_child_sock() 397 */ 398 if (ret) 399 inet_sk(ret)->cork.fl.u.ip4 = fl4; 400out: return ret; 401} 402