inet6_hashtables.c 9.0 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322
  1. /*
  2. * INET An implementation of the TCP/IP protocol suite for the LINUX
  3. * operating system. INET is implemented using the BSD Socket
  4. * interface as the means of communication with the user level.
  5. *
  6. * Generic INET6 transport hashtables
  7. *
  8. * Authors: Lotsa people, from code originally in tcp, generalised here
  9. * by Arnaldo Carvalho de Melo <acme@mandriva.com>
  10. *
  11. * This program is free software; you can redistribute it and/or
  12. * modify it under the terms of the GNU General Public License
  13. * as published by the Free Software Foundation; either version
  14. * 2 of the License, or (at your option) any later version.
  15. */
  16. #include <linux/module.h>
  17. #include <linux/random.h>
  18. #include <net/addrconf.h>
  19. #include <net/inet_connection_sock.h>
  20. #include <net/inet_hashtables.h>
  21. #include <net/inet6_hashtables.h>
  22. #include <net/secure_seq.h>
  23. #include <net/ip.h>
  24. #include <net/sock_reuseport.h>
  25. u32 inet6_ehashfn(const struct net *net,
  26. const struct in6_addr *laddr, const u16 lport,
  27. const struct in6_addr *faddr, const __be16 fport)
  28. {
  29. static u32 inet6_ehash_secret __read_mostly;
  30. static u32 ipv6_hash_secret __read_mostly;
  31. u32 lhash, fhash;
  32. net_get_random_once(&inet6_ehash_secret, sizeof(inet6_ehash_secret));
  33. net_get_random_once(&ipv6_hash_secret, sizeof(ipv6_hash_secret));
  34. lhash = (__force u32)laddr->s6_addr32[3];
  35. fhash = __ipv6_addr_jhash(faddr, ipv6_hash_secret);
  36. return __inet6_ehashfn(lhash, lport, fhash, fport,
  37. inet6_ehash_secret + net_hash_mix(net));
  38. }
  39. /*
  40. * Sockets in TCP_CLOSE state are _always_ taken out of the hash, so
  41. * we need not check it for TCP lookups anymore, thanks Alexey. -DaveM
  42. *
  43. * The sockhash lock must be held as a reader here.
  44. */
  45. struct sock *__inet6_lookup_established(struct net *net,
  46. struct inet_hashinfo *hashinfo,
  47. const struct in6_addr *saddr,
  48. const __be16 sport,
  49. const struct in6_addr *daddr,
  50. const u16 hnum,
  51. const int dif)
  52. {
  53. struct sock *sk;
  54. const struct hlist_nulls_node *node;
  55. const __portpair ports = INET_COMBINED_PORTS(sport, hnum);
  56. /* Optimize here for direct hit, only listening connections can
  57. * have wildcards anyways.
  58. */
  59. unsigned int hash = inet6_ehashfn(net, daddr, hnum, saddr, sport);
  60. unsigned int slot = hash & hashinfo->ehash_mask;
  61. struct inet_ehash_bucket *head = &hashinfo->ehash[slot];
  62. begin:
  63. sk_nulls_for_each_rcu(sk, node, &head->chain) {
  64. if (sk->sk_hash != hash)
  65. continue;
  66. if (!INET6_MATCH(sk, net, saddr, daddr, ports, dif))
  67. continue;
  68. if (unlikely(!atomic_inc_not_zero(&sk->sk_refcnt)))
  69. goto out;
  70. if (unlikely(!INET6_MATCH(sk, net, saddr, daddr, ports, dif))) {
  71. sock_gen_put(sk);
  72. goto begin;
  73. }
  74. goto found;
  75. }
  76. if (get_nulls_value(node) != slot)
  77. goto begin;
  78. out:
  79. sk = NULL;
  80. found:
  81. return sk;
  82. }
  83. EXPORT_SYMBOL(__inet6_lookup_established);
  84. static inline int compute_score(struct sock *sk, struct net *net,
  85. const unsigned short hnum,
  86. const struct in6_addr *daddr,
  87. const int dif, bool exact_dif)
  88. {
  89. int score = -1;
  90. if (net_eq(sock_net(sk), net) && inet_sk(sk)->inet_num == hnum &&
  91. sk->sk_family == PF_INET6) {
  92. score = 1;
  93. if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr)) {
  94. if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr))
  95. return -1;
  96. score++;
  97. }
  98. if (sk->sk_bound_dev_if || exact_dif) {
  99. if (sk->sk_bound_dev_if != dif)
  100. return -1;
  101. score++;
  102. }
  103. if (sk->sk_incoming_cpu == raw_smp_processor_id())
  104. score++;
  105. }
  106. return score;
  107. }
  108. /* called with rcu_read_lock() */
  109. struct sock *inet6_lookup_listener(struct net *net,
  110. struct inet_hashinfo *hashinfo,
  111. struct sk_buff *skb, int doff,
  112. const struct in6_addr *saddr,
  113. const __be16 sport, const struct in6_addr *daddr,
  114. const unsigned short hnum, const int dif)
  115. {
  116. unsigned int hash = inet_lhashfn(net, hnum);
  117. struct inet_listen_hashbucket *ilb = &hashinfo->listening_hash[hash];
  118. int score, hiscore = 0, matches = 0, reuseport = 0;
  119. bool exact_dif = inet6_exact_dif_match(net, skb);
  120. struct sock *sk, *result = NULL;
  121. u32 phash = 0;
  122. sk_for_each(sk, &ilb->head) {
  123. score = compute_score(sk, net, hnum, daddr, dif, exact_dif);
  124. if (score > hiscore) {
  125. reuseport = sk->sk_reuseport;
  126. if (reuseport) {
  127. phash = inet6_ehashfn(net, daddr, hnum,
  128. saddr, sport);
  129. result = reuseport_select_sock(sk, phash,
  130. skb, doff);
  131. if (result)
  132. return result;
  133. matches = 1;
  134. }
  135. result = sk;
  136. hiscore = score;
  137. } else if (score == hiscore && reuseport) {
  138. matches++;
  139. if (reciprocal_scale(phash, matches) == 0)
  140. result = sk;
  141. phash = next_pseudo_random32(phash);
  142. }
  143. }
  144. return result;
  145. }
  146. EXPORT_SYMBOL_GPL(inet6_lookup_listener);
  147. struct sock *inet6_lookup(struct net *net, struct inet_hashinfo *hashinfo,
  148. struct sk_buff *skb, int doff,
  149. const struct in6_addr *saddr, const __be16 sport,
  150. const struct in6_addr *daddr, const __be16 dport,
  151. const int dif)
  152. {
  153. struct sock *sk;
  154. bool refcounted;
  155. sk = __inet6_lookup(net, hashinfo, skb, doff, saddr, sport, daddr,
  156. ntohs(dport), dif, &refcounted);
  157. if (sk && !refcounted && !atomic_inc_not_zero(&sk->sk_refcnt))
  158. sk = NULL;
  159. return sk;
  160. }
  161. EXPORT_SYMBOL_GPL(inet6_lookup);
  162. static int __inet6_check_established(struct inet_timewait_death_row *death_row,
  163. struct sock *sk, const __u16 lport,
  164. struct inet_timewait_sock **twp)
  165. {
  166. struct inet_hashinfo *hinfo = death_row->hashinfo;
  167. struct inet_sock *inet = inet_sk(sk);
  168. const struct in6_addr *daddr = &sk->sk_v6_rcv_saddr;
  169. const struct in6_addr *saddr = &sk->sk_v6_daddr;
  170. const int dif = sk->sk_bound_dev_if;
  171. const __portpair ports = INET_COMBINED_PORTS(inet->inet_dport, lport);
  172. struct net *net = sock_net(sk);
  173. const unsigned int hash = inet6_ehashfn(net, daddr, lport, saddr,
  174. inet->inet_dport);
  175. struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash);
  176. spinlock_t *lock = inet_ehash_lockp(hinfo, hash);
  177. struct sock *sk2;
  178. const struct hlist_nulls_node *node;
  179. struct inet_timewait_sock *tw = NULL;
  180. spin_lock(lock);
  181. sk_nulls_for_each(sk2, node, &head->chain) {
  182. if (sk2->sk_hash != hash)
  183. continue;
  184. if (likely(INET6_MATCH(sk2, net, saddr, daddr, ports, dif))) {
  185. if (sk2->sk_state == TCP_TIME_WAIT) {
  186. tw = inet_twsk(sk2);
  187. if (twsk_unique(sk, sk2, twp))
  188. break;
  189. }
  190. goto not_unique;
  191. }
  192. }
  193. /* Must record num and sport now. Otherwise we will see
  194. * in hash table socket with a funny identity.
  195. */
  196. inet->inet_num = lport;
  197. inet->inet_sport = htons(lport);
  198. sk->sk_hash = hash;
  199. WARN_ON(!sk_unhashed(sk));
  200. __sk_nulls_add_node_rcu(sk, &head->chain);
  201. if (tw) {
  202. sk_nulls_del_node_init_rcu((struct sock *)tw);
  203. __NET_INC_STATS(net, LINUX_MIB_TIMEWAITRECYCLED);
  204. }
  205. spin_unlock(lock);
  206. sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
  207. if (twp) {
  208. *twp = tw;
  209. } else if (tw) {
  210. /* Silly. Should hash-dance instead... */
  211. inet_twsk_deschedule_put(tw);
  212. }
  213. return 0;
  214. not_unique:
  215. spin_unlock(lock);
  216. return -EADDRNOTAVAIL;
  217. }
  218. static u32 inet6_sk_port_offset(const struct sock *sk)
  219. {
  220. const struct inet_sock *inet = inet_sk(sk);
  221. return secure_ipv6_port_ephemeral(sk->sk_v6_rcv_saddr.s6_addr32,
  222. sk->sk_v6_daddr.s6_addr32,
  223. inet->inet_dport);
  224. }
  225. int inet6_hash_connect(struct inet_timewait_death_row *death_row,
  226. struct sock *sk)
  227. {
  228. u32 port_offset = 0;
  229. if (!inet_sk(sk)->inet_num)
  230. port_offset = inet6_sk_port_offset(sk);
  231. return __inet_hash_connect(death_row, sk, port_offset,
  232. __inet6_check_established);
  233. }
  234. EXPORT_SYMBOL_GPL(inet6_hash_connect);
  235. int inet6_hash(struct sock *sk)
  236. {
  237. int err = 0;
  238. if (sk->sk_state != TCP_CLOSE) {
  239. local_bh_disable();
  240. err = __inet_hash(sk, NULL, ipv6_rcv_saddr_equal);
  241. local_bh_enable();
  242. }
  243. return err;
  244. }
  245. EXPORT_SYMBOL_GPL(inet6_hash);
  246. /* match_wildcard == true: IPV6_ADDR_ANY equals to any IPv6 addresses if IPv6
  247. * only, and any IPv4 addresses if not IPv6 only
  248. * match_wildcard == false: addresses must be exactly the same, i.e.
  249. * IPV6_ADDR_ANY only equals to IPV6_ADDR_ANY,
  250. * and 0.0.0.0 equals to 0.0.0.0 only
  251. */
  252. int ipv6_rcv_saddr_equal(const struct sock *sk, const struct sock *sk2,
  253. bool match_wildcard)
  254. {
  255. const struct in6_addr *sk2_rcv_saddr6 = inet6_rcv_saddr(sk2);
  256. int sk2_ipv6only = inet_v6_ipv6only(sk2);
  257. int addr_type = ipv6_addr_type(&sk->sk_v6_rcv_saddr);
  258. int addr_type2 = sk2_rcv_saddr6 ? ipv6_addr_type(sk2_rcv_saddr6) : IPV6_ADDR_MAPPED;
  259. /* if both are mapped, treat as IPv4 */
  260. if (addr_type == IPV6_ADDR_MAPPED && addr_type2 == IPV6_ADDR_MAPPED) {
  261. if (!sk2_ipv6only) {
  262. if (sk->sk_rcv_saddr == sk2->sk_rcv_saddr)
  263. return 1;
  264. if (!sk->sk_rcv_saddr || !sk2->sk_rcv_saddr)
  265. return match_wildcard;
  266. }
  267. return 0;
  268. }
  269. if (addr_type == IPV6_ADDR_ANY && addr_type2 == IPV6_ADDR_ANY)
  270. return 1;
  271. if (addr_type2 == IPV6_ADDR_ANY && match_wildcard &&
  272. !(sk2_ipv6only && addr_type == IPV6_ADDR_MAPPED))
  273. return 1;
  274. if (addr_type == IPV6_ADDR_ANY && match_wildcard &&
  275. !(ipv6_only_sock(sk) && addr_type2 == IPV6_ADDR_MAPPED))
  276. return 1;
  277. if (sk2_rcv_saddr6 &&
  278. ipv6_addr_equal(&sk->sk_v6_rcv_saddr, sk2_rcv_saddr6))
  279. return 1;
  280. return 0;
  281. }
  282. EXPORT_SYMBOL_GPL(ipv6_rcv_saddr_equal);