syncookies.c 12 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410
  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. #include <linux/tcp.h>
  13. #include <linux/slab.h>
  14. #include <linux/random.h>
  15. #include <linux/cryptohash.h>
  16. #include <linux/kernel.h>
  17. #include <linux/export.h>
  18. #include <net/tcp.h>
  19. #include <net/route.h>
  20. extern int sysctl_tcp_syncookies;
  21. static u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS] __read_mostly;
  22. #define COOKIEBITS 24 /* Upper bits store count */
  23. #define COOKIEMASK (((__u32)1 << COOKIEBITS) - 1)
  24. /* TCP Timestamp: 6 lowest bits of timestamp sent in the cookie SYN-ACK
  25. * stores TCP options:
  26. *
  27. * MSB LSB
  28. * | 31 ... 6 | 5 | 4 | 3 2 1 0 |
  29. * | Timestamp | ECN | SACK | WScale |
  30. *
  31. * When we receive a valid cookie-ACK, we look at the echoed tsval (if
  32. * any) to figure out which TCP options we should use for the rebuilt
  33. * connection.
  34. *
  35. * A WScale setting of '0xf' (which is an invalid scaling value)
  36. * means that original syn did not include the TCP window scaling option.
  37. */
  38. #define TS_OPT_WSCALE_MASK 0xf
  39. #define TS_OPT_SACK BIT(4)
  40. #define TS_OPT_ECN BIT(5)
  41. /* There is no TS_OPT_TIMESTAMP:
  42. * if ACK contains timestamp option, we already know it was
  43. * requested/supported by the syn/synack exchange.
  44. */
  45. #define TSBITS 6
  46. #define TSMASK (((__u32)1 << TSBITS) - 1)
  47. static DEFINE_PER_CPU(__u32 [16 + 5 + SHA_WORKSPACE_WORDS],
  48. ipv4_cookie_scratch);
  49. static u32 cookie_hash(__be32 saddr, __be32 daddr, __be16 sport, __be16 dport,
  50. u32 count, int c)
  51. {
  52. __u32 *tmp;
  53. net_get_random_once(syncookie_secret, sizeof(syncookie_secret));
  54. tmp = this_cpu_ptr(ipv4_cookie_scratch);
  55. memcpy(tmp + 4, syncookie_secret[c], sizeof(syncookie_secret[c]));
  56. tmp[0] = (__force u32)saddr;
  57. tmp[1] = (__force u32)daddr;
  58. tmp[2] = ((__force u32)sport << 16) + (__force u32)dport;
  59. tmp[3] = count;
  60. sha_transform(tmp + 16, (__u8 *)tmp, tmp + 16 + 5);
  61. return tmp[17];
  62. }
  63. /*
  64. * when syncookies are in effect and tcp timestamps are enabled we encode
  65. * tcp options in the lower bits of the timestamp value that will be
  66. * sent in the syn-ack.
  67. * Since subsequent timestamps use the normal tcp_time_stamp value, we
  68. * must make sure that the resulting initial timestamp is <= tcp_time_stamp.
  69. */
  70. __u32 cookie_init_timestamp(struct request_sock *req)
  71. {
  72. struct inet_request_sock *ireq;
  73. u32 ts, ts_now = tcp_time_stamp;
  74. u32 options = 0;
  75. ireq = inet_rsk(req);
  76. options = ireq->wscale_ok ? ireq->snd_wscale : TS_OPT_WSCALE_MASK;
  77. if (ireq->sack_ok)
  78. options |= TS_OPT_SACK;
  79. if (ireq->ecn_ok)
  80. options |= TS_OPT_ECN;
  81. ts = ts_now & ~TSMASK;
  82. ts |= options;
  83. if (ts > ts_now) {
  84. ts >>= TSBITS;
  85. ts--;
  86. ts <<= TSBITS;
  87. ts |= options;
  88. }
  89. return ts;
  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. * This retrieves the small "data" value from the syncookie.
  112. * If the syncookie is bad, the data returned will be out of
  113. * range. This must be checked by the caller.
  114. *
  115. * The count value used to generate the cookie must be less than
  116. * MAX_SYNCOOKIE_AGE minutes in the past.
  117. * The return value (__u32)-1 if this test fails.
  118. */
  119. static __u32 check_tcp_syn_cookie(__u32 cookie, __be32 saddr, __be32 daddr,
  120. __be16 sport, __be16 dport, __u32 sseq)
  121. {
  122. u32 diff, count = tcp_cookie_time();
  123. /* Strip away the layers from the cookie */
  124. cookie -= cookie_hash(saddr, daddr, sport, dport, 0, 0) + sseq;
  125. /* Cookie is now reduced to (count * 2^24) ^ (hash % 2^24) */
  126. diff = (count - (cookie >> COOKIEBITS)) & ((__u32) -1 >> COOKIEBITS);
  127. if (diff >= MAX_SYNCOOKIE_AGE)
  128. return (__u32)-1;
  129. return (cookie -
  130. cookie_hash(saddr, daddr, sport, dport, count - diff, 1))
  131. & COOKIEMASK; /* Leaving the data behind */
  132. }
  133. /*
  134. * MSS Values are chosen based on the 2011 paper
  135. * 'An Analysis of TCP Maximum Segement Sizes' by S. Alcock and R. Nelson.
  136. * Values ..
  137. * .. lower than 536 are rare (< 0.2%)
  138. * .. between 537 and 1299 account for less than < 1.5% of observed values
  139. * .. in the 1300-1349 range account for about 15 to 20% of observed mss values
  140. * .. exceeding 1460 are very rare (< 0.04%)
  141. *
  142. * 1460 is the single most frequently announced mss value (30 to 46% depending
  143. * on monitor location). Table must be sorted.
  144. */
  145. static __u16 const msstab[] = {
  146. 536,
  147. 1300,
  148. 1440, /* 1440, 1452: PPPoE */
  149. 1460,
  150. };
  151. /*
  152. * Generate a syncookie. mssp points to the mss, which is returned
  153. * rounded down to the value encoded in the cookie.
  154. */
  155. u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
  156. u16 *mssp)
  157. {
  158. int mssind;
  159. const __u16 mss = *mssp;
  160. for (mssind = ARRAY_SIZE(msstab) - 1; mssind ; mssind--)
  161. if (mss >= msstab[mssind])
  162. break;
  163. *mssp = msstab[mssind];
  164. return secure_tcp_syn_cookie(iph->saddr, iph->daddr,
  165. th->source, th->dest, ntohl(th->seq),
  166. mssind);
  167. }
  168. EXPORT_SYMBOL_GPL(__cookie_v4_init_sequence);
  169. __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mssp)
  170. {
  171. const struct iphdr *iph = ip_hdr(skb);
  172. const struct tcphdr *th = tcp_hdr(skb);
  173. return __cookie_v4_init_sequence(iph, th, mssp);
  174. }
  175. /*
  176. * Check if a ack sequence number is a valid syncookie.
  177. * Return the decoded mss if it is, or 0 if not.
  178. */
  179. int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
  180. u32 cookie)
  181. {
  182. __u32 seq = ntohl(th->seq) - 1;
  183. __u32 mssind = check_tcp_syn_cookie(cookie, iph->saddr, iph->daddr,
  184. th->source, th->dest, seq);
  185. return mssind < ARRAY_SIZE(msstab) ? msstab[mssind] : 0;
  186. }
  187. EXPORT_SYMBOL_GPL(__cookie_v4_check);
  188. struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
  189. struct request_sock *req,
  190. struct dst_entry *dst)
  191. {
  192. struct inet_connection_sock *icsk = inet_csk(sk);
  193. struct sock *child;
  194. bool own_req;
  195. child = icsk->icsk_af_ops->syn_recv_sock(sk, skb, req, dst,
  196. NULL, &own_req);
  197. if (child) {
  198. atomic_set(&req->rsk_refcnt, 1);
  199. sock_rps_save_rxhash(child, skb);
  200. if (!inet_csk_reqsk_queue_add(sk, req, child)) {
  201. bh_unlock_sock(child);
  202. sock_put(child);
  203. child = NULL;
  204. reqsk_put(req);
  205. }
  206. } else {
  207. reqsk_free(req);
  208. }
  209. return child;
  210. }
  211. EXPORT_SYMBOL(tcp_get_cookie_sock);
  212. /*
  213. * when syncookies are in effect and tcp timestamps are enabled we stored
  214. * additional tcp options in the timestamp.
  215. * This extracts these options from the timestamp echo.
  216. *
  217. * return false if we decode a tcp option that is disabled
  218. * on the host.
  219. */
  220. bool cookie_timestamp_decode(struct tcp_options_received *tcp_opt)
  221. {
  222. /* echoed timestamp, lowest bits contain options */
  223. u32 options = tcp_opt->rcv_tsecr;
  224. if (!tcp_opt->saw_tstamp) {
  225. tcp_clear_options(tcp_opt);
  226. return true;
  227. }
  228. if (!sysctl_tcp_timestamps)
  229. return false;
  230. tcp_opt->sack_ok = (options & TS_OPT_SACK) ? TCP_SACK_SEEN : 0;
  231. if (tcp_opt->sack_ok && !sysctl_tcp_sack)
  232. return false;
  233. if ((options & TS_OPT_WSCALE_MASK) == TS_OPT_WSCALE_MASK)
  234. return true; /* no window scaling */
  235. tcp_opt->wscale_ok = 1;
  236. tcp_opt->snd_wscale = options & TS_OPT_WSCALE_MASK;
  237. return sysctl_tcp_window_scaling != 0;
  238. }
  239. EXPORT_SYMBOL(cookie_timestamp_decode);
  240. bool cookie_ecn_ok(const struct tcp_options_received *tcp_opt,
  241. const struct net *net, const struct dst_entry *dst)
  242. {
  243. bool ecn_ok = tcp_opt->rcv_tsecr & TS_OPT_ECN;
  244. if (!ecn_ok)
  245. return false;
  246. if (net->ipv4.sysctl_tcp_ecn)
  247. return true;
  248. return dst_feature(dst, RTAX_FEATURE_ECN);
  249. }
  250. EXPORT_SYMBOL(cookie_ecn_ok);
  251. /* On input, sk is a listener.
  252. * Output is listener if incoming packet would not create a child
  253. * NULL if memory could not be allocated.
  254. */
  255. struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb)
  256. {
  257. struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
  258. struct tcp_options_received tcp_opt;
  259. struct inet_request_sock *ireq;
  260. struct tcp_request_sock *treq;
  261. struct tcp_sock *tp = tcp_sk(sk);
  262. const struct tcphdr *th = tcp_hdr(skb);
  263. __u32 cookie = ntohl(th->ack_seq) - 1;
  264. struct sock *ret = sk;
  265. struct request_sock *req;
  266. int mss;
  267. struct rtable *rt;
  268. __u8 rcv_wscale;
  269. struct flowi4 fl4;
  270. if (!sysctl_tcp_syncookies || !th->ack || th->rst)
  271. goto out;
  272. if (tcp_synq_no_recent_overflow(sk))
  273. goto out;
  274. mss = __cookie_v4_check(ip_hdr(skb), th, cookie);
  275. if (mss == 0) {
  276. NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SYNCOOKIESFAILED);
  277. goto out;
  278. }
  279. NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SYNCOOKIESRECV);
  280. /* check for timestamp cookie support */
  281. memset(&tcp_opt, 0, sizeof(tcp_opt));
  282. tcp_parse_options(skb, &tcp_opt, 0, NULL);
  283. if (!cookie_timestamp_decode(&tcp_opt))
  284. goto out;
  285. ret = NULL;
  286. req = inet_reqsk_alloc(&tcp_request_sock_ops, sk, false); /* for safety */
  287. if (!req)
  288. goto out;
  289. ireq = inet_rsk(req);
  290. treq = tcp_rsk(req);
  291. treq->rcv_isn = ntohl(th->seq) - 1;
  292. treq->snt_isn = cookie;
  293. treq->txhash = net_tx_rndhash();
  294. req->mss = mss;
  295. ireq->ir_num = ntohs(th->dest);
  296. ireq->ir_rmt_port = th->source;
  297. sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr);
  298. sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr);
  299. ireq->ir_mark = inet_request_mark(sk, skb);
  300. ireq->snd_wscale = tcp_opt.snd_wscale;
  301. ireq->sack_ok = tcp_opt.sack_ok;
  302. ireq->wscale_ok = tcp_opt.wscale_ok;
  303. ireq->tstamp_ok = tcp_opt.saw_tstamp;
  304. req->ts_recent = tcp_opt.saw_tstamp ? tcp_opt.rcv_tsval : 0;
  305. treq->snt_synack.v64 = 0;
  306. treq->tfo_listener = false;
  307. ireq->ir_iif = sk->sk_bound_dev_if;
  308. /* We throwed the options of the initial SYN away, so we hope
  309. * the ACK carries the same options again (see RFC1122 4.2.3.8)
  310. */
  311. RCU_INIT_POINTER(ireq->ireq_opt, tcp_v4_save_options(skb));
  312. if (security_inet_conn_request(sk, skb, req)) {
  313. reqsk_free(req);
  314. goto out;
  315. }
  316. req->num_retrans = 0;
  317. /*
  318. * We need to lookup the route here to get at the correct
  319. * window size. We should better make sure that the window size
  320. * hasn't changed since we received the original syn, but I see
  321. * no easy way to do this.
  322. */
  323. flowi4_init_output(&fl4, sk->sk_bound_dev_if, ireq->ir_mark,
  324. RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE, IPPROTO_TCP,
  325. inet_sk_flowi_flags(sk),
  326. opt->srr ? opt->faddr : ireq->ir_rmt_addr,
  327. ireq->ir_loc_addr, th->source, th->dest);
  328. security_req_classify_flow(req, flowi4_to_flowi(&fl4));
  329. rt = ip_route_output_key(sock_net(sk), &fl4);
  330. if (IS_ERR(rt)) {
  331. reqsk_free(req);
  332. goto out;
  333. }
  334. /* Try to redo what tcp_v4_send_synack did. */
  335. req->rsk_window_clamp = tp->window_clamp ? :dst_metric(&rt->dst, RTAX_WINDOW);
  336. tcp_select_initial_window(tcp_full_space(sk), req->mss,
  337. &req->rsk_rcv_wnd, &req->rsk_window_clamp,
  338. ireq->wscale_ok, &rcv_wscale,
  339. dst_metric(&rt->dst, RTAX_INITRWND));
  340. ireq->rcv_wscale = rcv_wscale;
  341. ireq->ecn_ok = cookie_ecn_ok(&tcp_opt, sock_net(sk), &rt->dst);
  342. ret = tcp_get_cookie_sock(sk, skb, req, &rt->dst);
  343. /* ip_queue_xmit() depends on our flow being setup
  344. * Normal sockets get it right from inet_csk_route_child_sock()
  345. */
  346. if (ret)
  347. inet_sk(ret)->cork.fl.u.ip4 = fl4;
  348. out: return ret;
  349. }