peer_event.c 10 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417
  1. // SPDX-License-Identifier: GPL-2.0-or-later
  2. /* Peer event handling, typically ICMP messages.
  3. *
  4. * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
  5. * Written by David Howells (dhowells@redhat.com)
  6. */
  7. #include <linux/module.h>
  8. #include <linux/net.h>
  9. #include <linux/skbuff.h>
  10. #include <linux/errqueue.h>
  11. #include <linux/udp.h>
  12. #include <linux/in.h>
  13. #include <linux/in6.h>
  14. #include <linux/icmp.h>
  15. #include <net/sock.h>
  16. #include <net/af_rxrpc.h>
  17. #include <net/ip.h>
  18. #include "ar-internal.h"
  19. static void rxrpc_store_error(struct rxrpc_peer *, struct sock_exterr_skb *);
  20. static void rxrpc_distribute_error(struct rxrpc_peer *, int,
  21. enum rxrpc_call_completion);
  22. /*
  23. * Find the peer associated with an ICMP packet.
  24. */
  25. static struct rxrpc_peer *rxrpc_lookup_peer_icmp_rcu(struct rxrpc_local *local,
  26. const struct sk_buff *skb,
  27. struct sockaddr_rxrpc *srx)
  28. {
  29. struct sock_exterr_skb *serr = SKB_EXT_ERR(skb);
  30. _enter("");
  31. memset(srx, 0, sizeof(*srx));
  32. srx->transport_type = local->srx.transport_type;
  33. srx->transport_len = local->srx.transport_len;
  34. srx->transport.family = local->srx.transport.family;
  35. /* Can we see an ICMP4 packet on an ICMP6 listening socket? and vice
  36. * versa?
  37. */
  38. switch (srx->transport.family) {
  39. case AF_INET:
  40. srx->transport_len = sizeof(srx->transport.sin);
  41. srx->transport.family = AF_INET;
  42. srx->transport.sin.sin_port = serr->port;
  43. switch (serr->ee.ee_origin) {
  44. case SO_EE_ORIGIN_ICMP:
  45. _net("Rx ICMP");
  46. memcpy(&srx->transport.sin.sin_addr,
  47. skb_network_header(skb) + serr->addr_offset,
  48. sizeof(struct in_addr));
  49. break;
  50. case SO_EE_ORIGIN_ICMP6:
  51. _net("Rx ICMP6 on v4 sock");
  52. memcpy(&srx->transport.sin.sin_addr,
  53. skb_network_header(skb) + serr->addr_offset + 12,
  54. sizeof(struct in_addr));
  55. break;
  56. default:
  57. memcpy(&srx->transport.sin.sin_addr, &ip_hdr(skb)->saddr,
  58. sizeof(struct in_addr));
  59. break;
  60. }
  61. break;
  62. #ifdef CONFIG_AF_RXRPC_IPV6
  63. case AF_INET6:
  64. switch (serr->ee.ee_origin) {
  65. case SO_EE_ORIGIN_ICMP6:
  66. _net("Rx ICMP6");
  67. srx->transport.sin6.sin6_port = serr->port;
  68. memcpy(&srx->transport.sin6.sin6_addr,
  69. skb_network_header(skb) + serr->addr_offset,
  70. sizeof(struct in6_addr));
  71. break;
  72. case SO_EE_ORIGIN_ICMP:
  73. _net("Rx ICMP on v6 sock");
  74. srx->transport_len = sizeof(srx->transport.sin);
  75. srx->transport.family = AF_INET;
  76. srx->transport.sin.sin_port = serr->port;
  77. memcpy(&srx->transport.sin.sin_addr,
  78. skb_network_header(skb) + serr->addr_offset,
  79. sizeof(struct in_addr));
  80. break;
  81. default:
  82. memcpy(&srx->transport.sin6.sin6_addr,
  83. &ipv6_hdr(skb)->saddr,
  84. sizeof(struct in6_addr));
  85. break;
  86. }
  87. break;
  88. #endif
  89. default:
  90. BUG();
  91. }
  92. return rxrpc_lookup_peer_rcu(local, srx);
  93. }
  94. /*
  95. * Handle an MTU/fragmentation problem.
  96. */
  97. static void rxrpc_adjust_mtu(struct rxrpc_peer *peer, struct sock_exterr_skb *serr)
  98. {
  99. u32 mtu = serr->ee.ee_info;
  100. _net("Rx ICMP Fragmentation Needed (%d)", mtu);
  101. /* wind down the local interface MTU */
  102. if (mtu > 0 && peer->if_mtu == 65535 && mtu < peer->if_mtu) {
  103. peer->if_mtu = mtu;
  104. _net("I/F MTU %u", mtu);
  105. }
  106. if (mtu == 0) {
  107. /* they didn't give us a size, estimate one */
  108. mtu = peer->if_mtu;
  109. if (mtu > 1500) {
  110. mtu >>= 1;
  111. if (mtu < 1500)
  112. mtu = 1500;
  113. } else {
  114. mtu -= 100;
  115. if (mtu < peer->hdrsize)
  116. mtu = peer->hdrsize + 4;
  117. }
  118. }
  119. if (mtu < peer->mtu) {
  120. spin_lock_bh(&peer->lock);
  121. peer->mtu = mtu;
  122. peer->maxdata = peer->mtu - peer->hdrsize;
  123. spin_unlock_bh(&peer->lock);
  124. _net("Net MTU %u (maxdata %u)",
  125. peer->mtu, peer->maxdata);
  126. }
  127. }
  128. /*
  129. * Handle an error received on the local endpoint.
  130. */
  131. void rxrpc_error_report(struct sock *sk)
  132. {
  133. struct sock_exterr_skb *serr;
  134. struct sockaddr_rxrpc srx;
  135. struct rxrpc_local *local;
  136. struct rxrpc_peer *peer;
  137. struct sk_buff *skb;
  138. rcu_read_lock();
  139. local = rcu_dereference_sk_user_data(sk);
  140. if (unlikely(!local)) {
  141. rcu_read_unlock();
  142. return;
  143. }
  144. _enter("%p{%d}", sk, local->debug_id);
  145. /* Clear the outstanding error value on the socket so that it doesn't
  146. * cause kernel_sendmsg() to return it later.
  147. */
  148. sock_error(sk);
  149. skb = sock_dequeue_err_skb(sk);
  150. if (!skb) {
  151. rcu_read_unlock();
  152. _leave("UDP socket errqueue empty");
  153. return;
  154. }
  155. rxrpc_new_skb(skb, rxrpc_skb_received);
  156. serr = SKB_EXT_ERR(skb);
  157. if (!skb->len && serr->ee.ee_origin == SO_EE_ORIGIN_TIMESTAMPING) {
  158. _leave("UDP empty message");
  159. rcu_read_unlock();
  160. rxrpc_free_skb(skb, rxrpc_skb_freed);
  161. return;
  162. }
  163. peer = rxrpc_lookup_peer_icmp_rcu(local, skb, &srx);
  164. if (peer && !rxrpc_get_peer_maybe(peer))
  165. peer = NULL;
  166. if (!peer) {
  167. rcu_read_unlock();
  168. rxrpc_free_skb(skb, rxrpc_skb_freed);
  169. _leave(" [no peer]");
  170. return;
  171. }
  172. trace_rxrpc_rx_icmp(peer, &serr->ee, &srx);
  173. if ((serr->ee.ee_origin == SO_EE_ORIGIN_ICMP &&
  174. serr->ee.ee_type == ICMP_DEST_UNREACH &&
  175. serr->ee.ee_code == ICMP_FRAG_NEEDED)) {
  176. rxrpc_adjust_mtu(peer, serr);
  177. rcu_read_unlock();
  178. rxrpc_free_skb(skb, rxrpc_skb_freed);
  179. rxrpc_put_peer(peer);
  180. _leave(" [MTU update]");
  181. return;
  182. }
  183. rxrpc_store_error(peer, serr);
  184. rcu_read_unlock();
  185. rxrpc_free_skb(skb, rxrpc_skb_freed);
  186. rxrpc_put_peer(peer);
  187. _leave("");
  188. }
  189. /*
  190. * Map an error report to error codes on the peer record.
  191. */
  192. static void rxrpc_store_error(struct rxrpc_peer *peer,
  193. struct sock_exterr_skb *serr)
  194. {
  195. enum rxrpc_call_completion compl = RXRPC_CALL_NETWORK_ERROR;
  196. struct sock_extended_err *ee;
  197. int err;
  198. _enter("");
  199. ee = &serr->ee;
  200. err = ee->ee_errno;
  201. switch (ee->ee_origin) {
  202. case SO_EE_ORIGIN_ICMP:
  203. switch (ee->ee_type) {
  204. case ICMP_DEST_UNREACH:
  205. switch (ee->ee_code) {
  206. case ICMP_NET_UNREACH:
  207. _net("Rx Received ICMP Network Unreachable");
  208. break;
  209. case ICMP_HOST_UNREACH:
  210. _net("Rx Received ICMP Host Unreachable");
  211. break;
  212. case ICMP_PORT_UNREACH:
  213. _net("Rx Received ICMP Port Unreachable");
  214. break;
  215. case ICMP_NET_UNKNOWN:
  216. _net("Rx Received ICMP Unknown Network");
  217. break;
  218. case ICMP_HOST_UNKNOWN:
  219. _net("Rx Received ICMP Unknown Host");
  220. break;
  221. default:
  222. _net("Rx Received ICMP DestUnreach code=%u",
  223. ee->ee_code);
  224. break;
  225. }
  226. break;
  227. case ICMP_TIME_EXCEEDED:
  228. _net("Rx Received ICMP TTL Exceeded");
  229. break;
  230. default:
  231. _proto("Rx Received ICMP error { type=%u code=%u }",
  232. ee->ee_type, ee->ee_code);
  233. break;
  234. }
  235. break;
  236. case SO_EE_ORIGIN_NONE:
  237. case SO_EE_ORIGIN_LOCAL:
  238. _proto("Rx Received local error { error=%d }", err);
  239. compl = RXRPC_CALL_LOCAL_ERROR;
  240. break;
  241. case SO_EE_ORIGIN_ICMP6:
  242. if (err == EACCES)
  243. err = EHOSTUNREACH;
  244. fallthrough;
  245. default:
  246. _proto("Rx Received error report { orig=%u }", ee->ee_origin);
  247. break;
  248. }
  249. rxrpc_distribute_error(peer, err, compl);
  250. }
  251. /*
  252. * Distribute an error that occurred on a peer.
  253. */
  254. static void rxrpc_distribute_error(struct rxrpc_peer *peer, int error,
  255. enum rxrpc_call_completion compl)
  256. {
  257. struct rxrpc_call *call;
  258. hlist_for_each_entry_rcu(call, &peer->error_targets, error_link) {
  259. rxrpc_see_call(call);
  260. rxrpc_set_call_completion(call, compl, 0, -error);
  261. }
  262. }
  263. /*
  264. * Perform keep-alive pings.
  265. */
  266. static void rxrpc_peer_keepalive_dispatch(struct rxrpc_net *rxnet,
  267. struct list_head *collector,
  268. time64_t base,
  269. u8 cursor)
  270. {
  271. struct rxrpc_peer *peer;
  272. const u8 mask = ARRAY_SIZE(rxnet->peer_keepalive) - 1;
  273. time64_t keepalive_at;
  274. int slot;
  275. spin_lock_bh(&rxnet->peer_hash_lock);
  276. while (!list_empty(collector)) {
  277. peer = list_entry(collector->next,
  278. struct rxrpc_peer, keepalive_link);
  279. list_del_init(&peer->keepalive_link);
  280. if (!rxrpc_get_peer_maybe(peer))
  281. continue;
  282. if (__rxrpc_use_local(peer->local)) {
  283. spin_unlock_bh(&rxnet->peer_hash_lock);
  284. keepalive_at = peer->last_tx_at + RXRPC_KEEPALIVE_TIME;
  285. slot = keepalive_at - base;
  286. _debug("%02x peer %u t=%d {%pISp}",
  287. cursor, peer->debug_id, slot, &peer->srx.transport);
  288. if (keepalive_at <= base ||
  289. keepalive_at > base + RXRPC_KEEPALIVE_TIME) {
  290. rxrpc_send_keepalive(peer);
  291. slot = RXRPC_KEEPALIVE_TIME;
  292. }
  293. /* A transmission to this peer occurred since last we
  294. * examined it so put it into the appropriate future
  295. * bucket.
  296. */
  297. slot += cursor;
  298. slot &= mask;
  299. spin_lock_bh(&rxnet->peer_hash_lock);
  300. list_add_tail(&peer->keepalive_link,
  301. &rxnet->peer_keepalive[slot & mask]);
  302. rxrpc_unuse_local(peer->local);
  303. }
  304. rxrpc_put_peer_locked(peer);
  305. }
  306. spin_unlock_bh(&rxnet->peer_hash_lock);
  307. }
  308. /*
  309. * Perform keep-alive pings with VERSION packets to keep any NAT alive.
  310. */
  311. void rxrpc_peer_keepalive_worker(struct work_struct *work)
  312. {
  313. struct rxrpc_net *rxnet =
  314. container_of(work, struct rxrpc_net, peer_keepalive_work);
  315. const u8 mask = ARRAY_SIZE(rxnet->peer_keepalive) - 1;
  316. time64_t base, now, delay;
  317. u8 cursor, stop;
  318. LIST_HEAD(collector);
  319. now = ktime_get_seconds();
  320. base = rxnet->peer_keepalive_base;
  321. cursor = rxnet->peer_keepalive_cursor;
  322. _enter("%lld,%u", base - now, cursor);
  323. if (!rxnet->live)
  324. return;
  325. /* Remove to a temporary list all the peers that are currently lodged
  326. * in expired buckets plus all new peers.
  327. *
  328. * Everything in the bucket at the cursor is processed this
  329. * second; the bucket at cursor + 1 goes at now + 1s and so
  330. * on...
  331. */
  332. spin_lock_bh(&rxnet->peer_hash_lock);
  333. list_splice_init(&rxnet->peer_keepalive_new, &collector);
  334. stop = cursor + ARRAY_SIZE(rxnet->peer_keepalive);
  335. while (base <= now && (s8)(cursor - stop) < 0) {
  336. list_splice_tail_init(&rxnet->peer_keepalive[cursor & mask],
  337. &collector);
  338. base++;
  339. cursor++;
  340. }
  341. base = now;
  342. spin_unlock_bh(&rxnet->peer_hash_lock);
  343. rxnet->peer_keepalive_base = base;
  344. rxnet->peer_keepalive_cursor = cursor;
  345. rxrpc_peer_keepalive_dispatch(rxnet, &collector, base, cursor);
  346. ASSERT(list_empty(&collector));
  347. /* Schedule the timer for the next occupied timeslot. */
  348. cursor = rxnet->peer_keepalive_cursor;
  349. stop = cursor + RXRPC_KEEPALIVE_TIME - 1;
  350. for (; (s8)(cursor - stop) < 0; cursor++) {
  351. if (!list_empty(&rxnet->peer_keepalive[cursor & mask]))
  352. break;
  353. base++;
  354. }
  355. now = ktime_get_seconds();
  356. delay = base - now;
  357. if (delay < 1)
  358. delay = 1;
  359. delay *= HZ;
  360. if (rxnet->live)
  361. timer_reduce(&rxnet->peer_keepalive_timer, jiffies + delay);
  362. _leave("");
  363. }