1 /* Peer event handling, typically ICMP messages. 2 * 3 * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved. 4 * Written by David Howells (dhowells@redhat.com) 5 * 6 * This program is free software; you can redistribute it and/or 7 * modify it under the terms of the GNU General Public License 8 * as published by the Free Software Foundation; either version 9 * 2 of the License, or (at your option) any later version. 10 */ 11 12 #include <linux/module.h> 13 #include <linux/net.h> 14 #include <linux/skbuff.h> 15 #include <linux/errqueue.h> 16 #include <linux/udp.h> 17 #include <linux/in.h> 18 #include <linux/in6.h> 19 #include <linux/icmp.h> 20 #include <net/sock.h> 21 #include <net/af_rxrpc.h> 22 #include <net/ip.h> 23 #include "ar-internal.h" 24 25 static void rxrpc_store_error(struct rxrpc_peer *, struct sock_exterr_skb *); 26 static void rxrpc_distribute_error(struct rxrpc_peer *, int, 27 enum rxrpc_call_completion); 28 29 /* 30 * Find the peer associated with an ICMP packet. 31 */ 32 static struct rxrpc_peer *rxrpc_lookup_peer_icmp_rcu(struct rxrpc_local *local, 33 const struct sk_buff *skb, 34 struct sockaddr_rxrpc *srx) 35 { 36 struct sock_exterr_skb *serr = SKB_EXT_ERR(skb); 37 38 _enter(""); 39 40 memset(srx, 0, sizeof(*srx)); 41 srx->transport_type = local->srx.transport_type; 42 srx->transport_len = local->srx.transport_len; 43 srx->transport.family = local->srx.transport.family; 44 45 /* Can we see an ICMP4 packet on an ICMP6 listening socket? and vice 46 * versa? 47 */ 48 switch (srx->transport.family) { 49 case AF_INET: 50 srx->transport_len = sizeof(srx->transport.sin); 51 srx->transport.family = AF_INET; 52 srx->transport.sin.sin_port = serr->port; 53 switch (serr->ee.ee_origin) { 54 case SO_EE_ORIGIN_ICMP: 55 _net("Rx ICMP"); 56 memcpy(&srx->transport.sin.sin_addr, 57 skb_network_header(skb) + serr->addr_offset, 58 sizeof(struct in_addr)); 59 break; 60 case SO_EE_ORIGIN_ICMP6: 61 _net("Rx ICMP6 on v4 sock"); 62 memcpy(&srx->transport.sin.sin_addr, 63 skb_network_header(skb) + serr->addr_offset + 12, 64 sizeof(struct in_addr)); 65 break; 66 default: 67 memcpy(&srx->transport.sin.sin_addr, &ip_hdr(skb)->saddr, 68 sizeof(struct in_addr)); 69 break; 70 } 71 break; 72 73 #ifdef CONFIG_AF_RXRPC_IPV6 74 case AF_INET6: 75 switch (serr->ee.ee_origin) { 76 case SO_EE_ORIGIN_ICMP6: 77 _net("Rx ICMP6"); 78 srx->transport.sin6.sin6_port = serr->port; 79 memcpy(&srx->transport.sin6.sin6_addr, 80 skb_network_header(skb) + serr->addr_offset, 81 sizeof(struct in6_addr)); 82 break; 83 case SO_EE_ORIGIN_ICMP: 84 _net("Rx ICMP on v6 sock"); 85 srx->transport_len = sizeof(srx->transport.sin); 86 srx->transport.family = AF_INET; 87 srx->transport.sin.sin_port = serr->port; 88 memcpy(&srx->transport.sin.sin_addr, 89 skb_network_header(skb) + serr->addr_offset, 90 sizeof(struct in_addr)); 91 break; 92 default: 93 memcpy(&srx->transport.sin6.sin6_addr, 94 &ipv6_hdr(skb)->saddr, 95 sizeof(struct in6_addr)); 96 break; 97 } 98 break; 99 #endif 100 101 default: 102 BUG(); 103 } 104 105 return rxrpc_lookup_peer_rcu(local, srx); 106 } 107 108 /* 109 * Handle an MTU/fragmentation problem. 110 */ 111 static void rxrpc_adjust_mtu(struct rxrpc_peer *peer, struct sock_exterr_skb *serr) 112 { 113 u32 mtu = serr->ee.ee_info; 114 115 _net("Rx ICMP Fragmentation Needed (%d)", mtu); 116 117 /* wind down the local interface MTU */ 118 if (mtu > 0 && peer->if_mtu == 65535 && mtu < peer->if_mtu) { 119 peer->if_mtu = mtu; 120 _net("I/F MTU %u", mtu); 121 } 122 123 if (mtu == 0) { 124 /* they didn't give us a size, estimate one */ 125 mtu = peer->if_mtu; 126 if (mtu > 1500) { 127 mtu >>= 1; 128 if (mtu < 1500) 129 mtu = 1500; 130 } else { 131 mtu -= 100; 132 if (mtu < peer->hdrsize) 133 mtu = peer->hdrsize + 4; 134 } 135 } 136 137 if (mtu < peer->mtu) { 138 spin_lock_bh(&peer->lock); 139 peer->mtu = mtu; 140 peer->maxdata = peer->mtu - peer->hdrsize; 141 spin_unlock_bh(&peer->lock); 142 _net("Net MTU %u (maxdata %u)", 143 peer->mtu, peer->maxdata); 144 } 145 } 146 147 /* 148 * Handle an error received on the local endpoint. 149 */ 150 void rxrpc_error_report(struct sock *sk) 151 { 152 struct sock_exterr_skb *serr; 153 struct sockaddr_rxrpc srx; 154 struct rxrpc_local *local = sk->sk_user_data; 155 struct rxrpc_peer *peer; 156 struct sk_buff *skb; 157 158 _enter("%p{%d}", sk, local->debug_id); 159 160 skb = sock_dequeue_err_skb(sk); 161 if (!skb) { 162 _leave("UDP socket errqueue empty"); 163 return; 164 } 165 rxrpc_new_skb(skb, rxrpc_skb_rx_received); 166 serr = SKB_EXT_ERR(skb); 167 if (!skb->len && serr->ee.ee_origin == SO_EE_ORIGIN_TIMESTAMPING) { 168 _leave("UDP empty message"); 169 rxrpc_free_skb(skb, rxrpc_skb_rx_freed); 170 return; 171 } 172 173 rcu_read_lock(); 174 peer = rxrpc_lookup_peer_icmp_rcu(local, skb, &srx); 175 if (peer && !rxrpc_get_peer_maybe(peer)) 176 peer = NULL; 177 if (!peer) { 178 rcu_read_unlock(); 179 rxrpc_free_skb(skb, rxrpc_skb_rx_freed); 180 _leave(" [no peer]"); 181 return; 182 } 183 184 trace_rxrpc_rx_icmp(peer, &serr->ee, &srx); 185 186 if ((serr->ee.ee_origin == SO_EE_ORIGIN_ICMP && 187 serr->ee.ee_type == ICMP_DEST_UNREACH && 188 serr->ee.ee_code == ICMP_FRAG_NEEDED)) { 189 rxrpc_adjust_mtu(peer, serr); 190 rcu_read_unlock(); 191 rxrpc_free_skb(skb, rxrpc_skb_rx_freed); 192 rxrpc_put_peer(peer); 193 _leave(" [MTU update]"); 194 return; 195 } 196 197 rxrpc_store_error(peer, serr); 198 rcu_read_unlock(); 199 rxrpc_free_skb(skb, rxrpc_skb_rx_freed); 200 201 _leave(""); 202 } 203 204 /* 205 * Map an error report to error codes on the peer record. 206 */ 207 static void rxrpc_store_error(struct rxrpc_peer *peer, 208 struct sock_exterr_skb *serr) 209 { 210 enum rxrpc_call_completion compl = RXRPC_CALL_NETWORK_ERROR; 211 struct sock_extended_err *ee; 212 int err; 213 214 _enter(""); 215 216 ee = &serr->ee; 217 218 err = ee->ee_errno; 219 220 switch (ee->ee_origin) { 221 case SO_EE_ORIGIN_ICMP: 222 switch (ee->ee_type) { 223 case ICMP_DEST_UNREACH: 224 switch (ee->ee_code) { 225 case ICMP_NET_UNREACH: 226 _net("Rx Received ICMP Network Unreachable"); 227 break; 228 case ICMP_HOST_UNREACH: 229 _net("Rx Received ICMP Host Unreachable"); 230 break; 231 case ICMP_PORT_UNREACH: 232 _net("Rx Received ICMP Port Unreachable"); 233 break; 234 case ICMP_NET_UNKNOWN: 235 _net("Rx Received ICMP Unknown Network"); 236 break; 237 case ICMP_HOST_UNKNOWN: 238 _net("Rx Received ICMP Unknown Host"); 239 break; 240 default: 241 _net("Rx Received ICMP DestUnreach code=%u", 242 ee->ee_code); 243 break; 244 } 245 break; 246 247 case ICMP_TIME_EXCEEDED: 248 _net("Rx Received ICMP TTL Exceeded"); 249 break; 250 251 default: 252 _proto("Rx Received ICMP error { type=%u code=%u }", 253 ee->ee_type, ee->ee_code); 254 break; 255 } 256 break; 257 258 case SO_EE_ORIGIN_NONE: 259 case SO_EE_ORIGIN_LOCAL: 260 _proto("Rx Received local error { error=%d }", err); 261 compl = RXRPC_CALL_LOCAL_ERROR; 262 break; 263 264 case SO_EE_ORIGIN_ICMP6: 265 default: 266 _proto("Rx Received error report { orig=%u }", ee->ee_origin); 267 break; 268 } 269 270 rxrpc_distribute_error(peer, err, compl); 271 } 272 273 /* 274 * Distribute an error that occurred on a peer. 275 */ 276 static void rxrpc_distribute_error(struct rxrpc_peer *peer, int error, 277 enum rxrpc_call_completion compl) 278 { 279 struct rxrpc_call *call; 280 281 hlist_for_each_entry_rcu(call, &peer->error_targets, error_link) { 282 rxrpc_see_call(call); 283 if (call->state < RXRPC_CALL_COMPLETE && 284 rxrpc_set_call_completion(call, compl, 0, -error)) 285 rxrpc_notify_socket(call); 286 } 287 } 288 289 /* 290 * Add RTT information to cache. This is called in softirq mode and has 291 * exclusive access to the peer RTT data. 292 */ 293 void rxrpc_peer_add_rtt(struct rxrpc_call *call, enum rxrpc_rtt_rx_trace why, 294 rxrpc_serial_t send_serial, rxrpc_serial_t resp_serial, 295 ktime_t send_time, ktime_t resp_time) 296 { 297 struct rxrpc_peer *peer = call->peer; 298 s64 rtt; 299 u64 sum = peer->rtt_sum, avg; 300 u8 cursor = peer->rtt_cursor, usage = peer->rtt_usage; 301 302 rtt = ktime_to_ns(ktime_sub(resp_time, send_time)); 303 if (rtt < 0) 304 return; 305 306 /* Replace the oldest datum in the RTT buffer */ 307 sum -= peer->rtt_cache[cursor]; 308 sum += rtt; 309 peer->rtt_cache[cursor] = rtt; 310 peer->rtt_cursor = (cursor + 1) & (RXRPC_RTT_CACHE_SIZE - 1); 311 peer->rtt_sum = sum; 312 if (usage < RXRPC_RTT_CACHE_SIZE) { 313 usage++; 314 peer->rtt_usage = usage; 315 } 316 317 /* Now recalculate the average */ 318 if (usage == RXRPC_RTT_CACHE_SIZE) { 319 avg = sum / RXRPC_RTT_CACHE_SIZE; 320 } else { 321 avg = sum; 322 do_div(avg, usage); 323 } 324 325 peer->rtt = avg; 326 trace_rxrpc_rtt_rx(call, why, send_serial, resp_serial, rtt, 327 usage, avg); 328 } 329 330 /* 331 * Perform keep-alive pings. 332 */ 333 static void rxrpc_peer_keepalive_dispatch(struct rxrpc_net *rxnet, 334 struct list_head *collector, 335 time64_t base, 336 u8 cursor) 337 { 338 struct rxrpc_peer *peer; 339 const u8 mask = ARRAY_SIZE(rxnet->peer_keepalive) - 1; 340 time64_t keepalive_at; 341 int slot; 342 343 spin_lock_bh(&rxnet->peer_hash_lock); 344 345 while (!list_empty(collector)) { 346 peer = list_entry(collector->next, 347 struct rxrpc_peer, keepalive_link); 348 349 list_del_init(&peer->keepalive_link); 350 if (!rxrpc_get_peer_maybe(peer)) 351 continue; 352 353 spin_unlock_bh(&rxnet->peer_hash_lock); 354 355 keepalive_at = peer->last_tx_at + RXRPC_KEEPALIVE_TIME; 356 slot = keepalive_at - base; 357 _debug("%02x peer %u t=%d {%pISp}", 358 cursor, peer->debug_id, slot, &peer->srx.transport); 359 360 if (keepalive_at <= base || 361 keepalive_at > base + RXRPC_KEEPALIVE_TIME) { 362 rxrpc_send_keepalive(peer); 363 slot = RXRPC_KEEPALIVE_TIME; 364 } 365 366 /* A transmission to this peer occurred since last we examined 367 * it so put it into the appropriate future bucket. 368 */ 369 slot += cursor; 370 slot &= mask; 371 spin_lock_bh(&rxnet->peer_hash_lock); 372 list_add_tail(&peer->keepalive_link, 373 &rxnet->peer_keepalive[slot & mask]); 374 rxrpc_put_peer(peer); 375 } 376 377 spin_unlock_bh(&rxnet->peer_hash_lock); 378 } 379 380 /* 381 * Perform keep-alive pings with VERSION packets to keep any NAT alive. 382 */ 383 void rxrpc_peer_keepalive_worker(struct work_struct *work) 384 { 385 struct rxrpc_net *rxnet = 386 container_of(work, struct rxrpc_net, peer_keepalive_work); 387 const u8 mask = ARRAY_SIZE(rxnet->peer_keepalive) - 1; 388 time64_t base, now, delay; 389 u8 cursor, stop; 390 LIST_HEAD(collector); 391 392 now = ktime_get_seconds(); 393 base = rxnet->peer_keepalive_base; 394 cursor = rxnet->peer_keepalive_cursor; 395 _enter("%lld,%u", base - now, cursor); 396 397 if (!rxnet->live) 398 return; 399 400 /* Remove to a temporary list all the peers that are currently lodged 401 * in expired buckets plus all new peers. 402 * 403 * Everything in the bucket at the cursor is processed this 404 * second; the bucket at cursor + 1 goes at now + 1s and so 405 * on... 406 */ 407 spin_lock_bh(&rxnet->peer_hash_lock); 408 list_splice_init(&rxnet->peer_keepalive_new, &collector); 409 410 stop = cursor + ARRAY_SIZE(rxnet->peer_keepalive); 411 while (base <= now && (s8)(cursor - stop) < 0) { 412 list_splice_tail_init(&rxnet->peer_keepalive[cursor & mask], 413 &collector); 414 base++; 415 cursor++; 416 } 417 418 base = now; 419 spin_unlock_bh(&rxnet->peer_hash_lock); 420 421 rxnet->peer_keepalive_base = base; 422 rxnet->peer_keepalive_cursor = cursor; 423 rxrpc_peer_keepalive_dispatch(rxnet, &collector, base, cursor); 424 ASSERT(list_empty(&collector)); 425 426 /* Schedule the timer for the next occupied timeslot. */ 427 cursor = rxnet->peer_keepalive_cursor; 428 stop = cursor + RXRPC_KEEPALIVE_TIME - 1; 429 for (; (s8)(cursor - stop) < 0; cursor++) { 430 if (!list_empty(&rxnet->peer_keepalive[cursor & mask])) 431 break; 432 base++; 433 } 434 435 now = ktime_get_seconds(); 436 delay = base - now; 437 if (delay < 1) 438 delay = 1; 439 delay *= HZ; 440 if (rxnet->live) 441 timer_reduce(&rxnet->peer_keepalive_timer, jiffies + delay); 442 443 _leave(""); 444 } 445