inet6_hashtables.c 8.1 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305
  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/inet_connection_sock.h>
  19. #include <net/inet_hashtables.h>
  20. #include <net/inet6_hashtables.h>
  21. #include <net/secure_seq.h>
  22. #include <net/ip.h>
  23. int __inet6_hash(struct sock *sk, struct inet_timewait_sock *tw)
  24. {
  25. struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
  26. int twrefcnt = 0;
  27. WARN_ON(!sk_unhashed(sk));
  28. if (sk->sk_state == TCP_LISTEN) {
  29. struct inet_listen_hashbucket *ilb;
  30. ilb = &hashinfo->listening_hash[inet_sk_listen_hashfn(sk)];
  31. spin_lock(&ilb->lock);
  32. __sk_nulls_add_node_rcu(sk, &ilb->head);
  33. spin_unlock(&ilb->lock);
  34. } else {
  35. unsigned int hash;
  36. struct hlist_nulls_head *list;
  37. spinlock_t *lock;
  38. sk->sk_hash = hash = inet6_sk_ehashfn(sk);
  39. list = &inet_ehash_bucket(hashinfo, hash)->chain;
  40. lock = inet_ehash_lockp(hashinfo, hash);
  41. spin_lock(lock);
  42. __sk_nulls_add_node_rcu(sk, list);
  43. if (tw) {
  44. WARN_ON(sk->sk_hash != tw->tw_hash);
  45. twrefcnt = inet_twsk_unhash(tw);
  46. }
  47. spin_unlock(lock);
  48. }
  49. sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
  50. return twrefcnt;
  51. }
  52. EXPORT_SYMBOL(__inet6_hash);
  53. /*
  54. * Sockets in TCP_CLOSE state are _always_ taken out of the hash, so
  55. * we need not check it for TCP lookups anymore, thanks Alexey. -DaveM
  56. *
  57. * The sockhash lock must be held as a reader here.
  58. */
  59. struct sock *__inet6_lookup_established(struct net *net,
  60. struct inet_hashinfo *hashinfo,
  61. const struct in6_addr *saddr,
  62. const __be16 sport,
  63. const struct in6_addr *daddr,
  64. const u16 hnum,
  65. const int dif)
  66. {
  67. struct sock *sk;
  68. const struct hlist_nulls_node *node;
  69. const __portpair ports = INET_COMBINED_PORTS(sport, hnum);
  70. /* Optimize here for direct hit, only listening connections can
  71. * have wildcards anyways.
  72. */
  73. unsigned int hash = inet6_ehashfn(net, daddr, hnum, saddr, sport);
  74. unsigned int slot = hash & hashinfo->ehash_mask;
  75. struct inet_ehash_bucket *head = &hashinfo->ehash[slot];
  76. rcu_read_lock();
  77. begin:
  78. sk_nulls_for_each_rcu(sk, node, &head->chain) {
  79. /* For IPV6 do the cheaper port and family tests first. */
  80. if (INET6_MATCH(sk, net, hash, saddr, daddr, ports, dif)) {
  81. if (unlikely(!atomic_inc_not_zero(&sk->sk_refcnt)))
  82. goto begintw;
  83. if (!INET6_MATCH(sk, net, hash, saddr, daddr, ports, dif)) {
  84. sock_put(sk);
  85. goto begin;
  86. }
  87. goto out;
  88. }
  89. }
  90. if (get_nulls_value(node) != slot)
  91. goto begin;
  92. begintw:
  93. /* Must check for a TIME_WAIT'er before going to listener hash. */
  94. sk_nulls_for_each_rcu(sk, node, &head->twchain) {
  95. if (INET6_TW_MATCH(sk, net, hash, saddr, daddr, ports, dif)) {
  96. if (unlikely(!atomic_inc_not_zero(&sk->sk_refcnt))) {
  97. sk = NULL;
  98. goto out;
  99. }
  100. if (!INET6_TW_MATCH(sk, net, hash, saddr, daddr, ports, dif)) {
  101. inet_twsk_put(inet_twsk(sk));
  102. goto begintw;
  103. }
  104. goto out;
  105. }
  106. }
  107. if (get_nulls_value(node) != slot)
  108. goto begintw;
  109. sk = NULL;
  110. out:
  111. rcu_read_unlock();
  112. return sk;
  113. }
  114. EXPORT_SYMBOL(__inet6_lookup_established);
  115. static inline int compute_score(struct sock *sk, struct net *net,
  116. const unsigned short hnum,
  117. const struct in6_addr *daddr,
  118. const int dif)
  119. {
  120. int score = -1;
  121. if (net_eq(sock_net(sk), net) && inet_sk(sk)->inet_num == hnum &&
  122. sk->sk_family == PF_INET6) {
  123. const struct ipv6_pinfo *np = inet6_sk(sk);
  124. score = 1;
  125. if (!ipv6_addr_any(&np->rcv_saddr)) {
  126. if (!ipv6_addr_equal(&np->rcv_saddr, daddr))
  127. return -1;
  128. score++;
  129. }
  130. if (sk->sk_bound_dev_if) {
  131. if (sk->sk_bound_dev_if != dif)
  132. return -1;
  133. score++;
  134. }
  135. }
  136. return score;
  137. }
  138. struct sock *inet6_lookup_listener(struct net *net,
  139. struct inet_hashinfo *hashinfo, const struct in6_addr *daddr,
  140. const unsigned short hnum, const int dif)
  141. {
  142. struct sock *sk;
  143. const struct hlist_nulls_node *node;
  144. struct sock *result;
  145. int score, hiscore;
  146. unsigned int hash = inet_lhashfn(net, hnum);
  147. struct inet_listen_hashbucket *ilb = &hashinfo->listening_hash[hash];
  148. rcu_read_lock();
  149. begin:
  150. result = NULL;
  151. hiscore = -1;
  152. sk_nulls_for_each(sk, node, &ilb->head) {
  153. score = compute_score(sk, net, hnum, daddr, dif);
  154. if (score > hiscore) {
  155. hiscore = score;
  156. result = sk;
  157. }
  158. }
  159. /*
  160. * if the nulls value we got at the end of this lookup is
  161. * not the expected one, we must restart lookup.
  162. * We probably met an item that was moved to another chain.
  163. */
  164. if (get_nulls_value(node) != hash + LISTENING_NULLS_BASE)
  165. goto begin;
  166. if (result) {
  167. if (unlikely(!atomic_inc_not_zero(&result->sk_refcnt)))
  168. result = NULL;
  169. else if (unlikely(compute_score(result, net, hnum, daddr,
  170. dif) < hiscore)) {
  171. sock_put(result);
  172. goto begin;
  173. }
  174. }
  175. rcu_read_unlock();
  176. return result;
  177. }
  178. EXPORT_SYMBOL_GPL(inet6_lookup_listener);
  179. struct sock *inet6_lookup(struct net *net, struct inet_hashinfo *hashinfo,
  180. const struct in6_addr *saddr, const __be16 sport,
  181. const struct in6_addr *daddr, const __be16 dport,
  182. const int dif)
  183. {
  184. struct sock *sk;
  185. local_bh_disable();
  186. sk = __inet6_lookup(net, hashinfo, saddr, sport, daddr, ntohs(dport), dif);
  187. local_bh_enable();
  188. return sk;
  189. }
  190. EXPORT_SYMBOL_GPL(inet6_lookup);
  191. static int __inet6_check_established(struct inet_timewait_death_row *death_row,
  192. struct sock *sk, const __u16 lport,
  193. struct inet_timewait_sock **twp)
  194. {
  195. struct inet_hashinfo *hinfo = death_row->hashinfo;
  196. struct inet_sock *inet = inet_sk(sk);
  197. const struct ipv6_pinfo *np = inet6_sk(sk);
  198. const struct in6_addr *daddr = &np->rcv_saddr;
  199. const struct in6_addr *saddr = &np->daddr;
  200. const int dif = sk->sk_bound_dev_if;
  201. const __portpair ports = INET_COMBINED_PORTS(inet->inet_dport, lport);
  202. struct net *net = sock_net(sk);
  203. const unsigned int hash = inet6_ehashfn(net, daddr, lport, saddr,
  204. inet->inet_dport);
  205. struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash);
  206. spinlock_t *lock = inet_ehash_lockp(hinfo, hash);
  207. struct sock *sk2;
  208. const struct hlist_nulls_node *node;
  209. struct inet_timewait_sock *tw;
  210. int twrefcnt = 0;
  211. spin_lock(lock);
  212. /* Check TIME-WAIT sockets first. */
  213. sk_nulls_for_each(sk2, node, &head->twchain) {
  214. tw = inet_twsk(sk2);
  215. if (INET6_TW_MATCH(sk2, net, hash, saddr, daddr, ports, dif)) {
  216. if (twsk_unique(sk, sk2, twp))
  217. goto unique;
  218. else
  219. goto not_unique;
  220. }
  221. }
  222. tw = NULL;
  223. /* And established part... */
  224. sk_nulls_for_each(sk2, node, &head->chain) {
  225. if (INET6_MATCH(sk2, net, hash, saddr, daddr, ports, dif))
  226. goto not_unique;
  227. }
  228. unique:
  229. /* Must record num and sport now. Otherwise we will see
  230. * in hash table socket with a funny identity. */
  231. inet->inet_num = lport;
  232. inet->inet_sport = htons(lport);
  233. sk->sk_hash = hash;
  234. WARN_ON(!sk_unhashed(sk));
  235. __sk_nulls_add_node_rcu(sk, &head->chain);
  236. if (tw) {
  237. twrefcnt = inet_twsk_unhash(tw);
  238. NET_INC_STATS_BH(net, LINUX_MIB_TIMEWAITRECYCLED);
  239. }
  240. spin_unlock(lock);
  241. if (twrefcnt)
  242. inet_twsk_put(tw);
  243. sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
  244. if (twp) {
  245. *twp = tw;
  246. } else if (tw) {
  247. /* Silly. Should hash-dance instead... */
  248. inet_twsk_deschedule(tw, death_row);
  249. inet_twsk_put(tw);
  250. }
  251. return 0;
  252. not_unique:
  253. spin_unlock(lock);
  254. return -EADDRNOTAVAIL;
  255. }
  256. static inline u32 inet6_sk_port_offset(const struct sock *sk)
  257. {
  258. const struct inet_sock *inet = inet_sk(sk);
  259. const struct ipv6_pinfo *np = inet6_sk(sk);
  260. return secure_ipv6_port_ephemeral(np->rcv_saddr.s6_addr32,
  261. np->daddr.s6_addr32,
  262. inet->inet_dport);
  263. }
  264. int inet6_hash_connect(struct inet_timewait_death_row *death_row,
  265. struct sock *sk)
  266. {
  267. return __inet_hash_connect(death_row, sk, inet6_sk_port_offset(sk),
  268. __inet6_check_established, __inet6_hash);
  269. }
  270. EXPORT_SYMBOL_GPL(inet6_hash_connect);