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 27 /* 28 * Find the peer associated with an ICMP packet. 29 */ 30 static struct rxrpc_peer *rxrpc_lookup_peer_icmp_rcu(struct rxrpc_local *local, 31 const struct sk_buff *skb) 32 { 33 struct sock_exterr_skb *serr = SKB_EXT_ERR(skb); 34 struct sockaddr_rxrpc srx; 35 36 _enter(""); 37 38 memset(&srx, 0, sizeof(srx)); 39 srx.transport_type = local->srx.transport_type; 40 srx.transport_len = local->srx.transport_len; 41 srx.transport.family = local->srx.transport.family; 42 43 /* Can we see an ICMP4 packet on an ICMP6 listening socket? and vice 44 * versa? 45 */ 46 switch (srx.transport.family) { 47 case AF_INET: 48 srx.transport.sin.sin_port = serr->port; 49 switch (serr->ee.ee_origin) { 50 case SO_EE_ORIGIN_ICMP: 51 _net("Rx ICMP"); 52 memcpy(&srx.transport.sin.sin_addr, 53 skb_network_header(skb) + serr->addr_offset, 54 sizeof(struct in_addr)); 55 break; 56 case SO_EE_ORIGIN_ICMP6: 57 _net("Rx ICMP6 on v4 sock"); 58 memcpy(&srx.transport.sin.sin_addr, 59 skb_network_header(skb) + serr->addr_offset + 12, 60 sizeof(struct in_addr)); 61 break; 62 default: 63 memcpy(&srx.transport.sin.sin_addr, &ip_hdr(skb)->saddr, 64 sizeof(struct in_addr)); 65 break; 66 } 67 break; 68 69 #ifdef CONFIG_AF_RXRPC_IPV6 70 case AF_INET6: 71 srx.transport.sin6.sin6_port = serr->port; 72 switch (serr->ee.ee_origin) { 73 case SO_EE_ORIGIN_ICMP6: 74 _net("Rx ICMP6"); 75 memcpy(&srx.transport.sin6.sin6_addr, 76 skb_network_header(skb) + serr->addr_offset, 77 sizeof(struct in6_addr)); 78 break; 79 case SO_EE_ORIGIN_ICMP: 80 _net("Rx ICMP on v6 sock"); 81 srx.transport.sin6.sin6_addr.s6_addr32[0] = 0; 82 srx.transport.sin6.sin6_addr.s6_addr32[1] = 0; 83 srx.transport.sin6.sin6_addr.s6_addr32[2] = htonl(0xffff); 84 memcpy(srx.transport.sin6.sin6_addr.s6_addr + 12, 85 skb_network_header(skb) + serr->addr_offset, 86 sizeof(struct in_addr)); 87 break; 88 default: 89 memcpy(&srx.transport.sin6.sin6_addr, 90 &ipv6_hdr(skb)->saddr, 91 sizeof(struct in6_addr)); 92 break; 93 } 94 break; 95 #endif 96 97 default: 98 BUG(); 99 } 100 101 return rxrpc_lookup_peer_rcu(local, &srx); 102 } 103 104 /* 105 * Handle an MTU/fragmentation problem. 106 */ 107 static void rxrpc_adjust_mtu(struct rxrpc_peer *peer, struct sock_exterr_skb *serr) 108 { 109 u32 mtu = serr->ee.ee_info; 110 111 _net("Rx ICMP Fragmentation Needed (%d)", mtu); 112 113 /* wind down the local interface MTU */ 114 if (mtu > 0 && peer->if_mtu == 65535 && mtu < peer->if_mtu) { 115 peer->if_mtu = mtu; 116 _net("I/F MTU %u", mtu); 117 } 118 119 if (mtu == 0) { 120 /* they didn't give us a size, estimate one */ 121 mtu = peer->if_mtu; 122 if (mtu > 1500) { 123 mtu >>= 1; 124 if (mtu < 1500) 125 mtu = 1500; 126 } else { 127 mtu -= 100; 128 if (mtu < peer->hdrsize) 129 mtu = peer->hdrsize + 4; 130 } 131 } 132 133 if (mtu < peer->mtu) { 134 spin_lock_bh(&peer->lock); 135 peer->mtu = mtu; 136 peer->maxdata = peer->mtu - peer->hdrsize; 137 spin_unlock_bh(&peer->lock); 138 _net("Net MTU %u (maxdata %u)", 139 peer->mtu, peer->maxdata); 140 } 141 } 142 143 /* 144 * Handle an error received on the local endpoint. 145 */ 146 void rxrpc_error_report(struct sock *sk) 147 { 148 struct sock_exterr_skb *serr; 149 struct rxrpc_local *local = sk->sk_user_data; 150 struct rxrpc_peer *peer; 151 struct sk_buff *skb; 152 153 _enter("%p{%d}", sk, local->debug_id); 154 155 skb = sock_dequeue_err_skb(sk); 156 if (!skb) { 157 _leave("UDP socket errqueue empty"); 158 return; 159 } 160 rxrpc_new_skb(skb, rxrpc_skb_rx_received); 161 serr = SKB_EXT_ERR(skb); 162 if (!skb->len && serr->ee.ee_origin == SO_EE_ORIGIN_TIMESTAMPING) { 163 _leave("UDP empty message"); 164 rxrpc_free_skb(skb, rxrpc_skb_rx_freed); 165 return; 166 } 167 168 rcu_read_lock(); 169 peer = rxrpc_lookup_peer_icmp_rcu(local, skb); 170 if (peer && !rxrpc_get_peer_maybe(peer)) 171 peer = NULL; 172 if (!peer) { 173 rcu_read_unlock(); 174 rxrpc_free_skb(skb, rxrpc_skb_rx_freed); 175 _leave(" [no peer]"); 176 return; 177 } 178 179 if ((serr->ee.ee_origin == SO_EE_ORIGIN_ICMP && 180 serr->ee.ee_type == ICMP_DEST_UNREACH && 181 serr->ee.ee_code == ICMP_FRAG_NEEDED)) { 182 rxrpc_adjust_mtu(peer, serr); 183 rcu_read_unlock(); 184 rxrpc_free_skb(skb, rxrpc_skb_rx_freed); 185 rxrpc_put_peer(peer); 186 _leave(" [MTU update]"); 187 return; 188 } 189 190 rxrpc_store_error(peer, serr); 191 rcu_read_unlock(); 192 rxrpc_free_skb(skb, rxrpc_skb_rx_freed); 193 194 /* The ref we obtained is passed off to the work item */ 195 __rxrpc_queue_peer_error(peer); 196 _leave(""); 197 } 198 199 /* 200 * Map an error report to error codes on the peer record. 201 */ 202 static void rxrpc_store_error(struct rxrpc_peer *peer, 203 struct sock_exterr_skb *serr) 204 { 205 struct sock_extended_err *ee; 206 int err; 207 208 _enter(""); 209 210 ee = &serr->ee; 211 212 _net("Rx Error o=%d t=%d c=%d e=%d", 213 ee->ee_origin, ee->ee_type, ee->ee_code, ee->ee_errno); 214 215 err = ee->ee_errno; 216 217 switch (ee->ee_origin) { 218 case SO_EE_ORIGIN_ICMP: 219 switch (ee->ee_type) { 220 case ICMP_DEST_UNREACH: 221 switch (ee->ee_code) { 222 case ICMP_NET_UNREACH: 223 _net("Rx Received ICMP Network Unreachable"); 224 break; 225 case ICMP_HOST_UNREACH: 226 _net("Rx Received ICMP Host Unreachable"); 227 break; 228 case ICMP_PORT_UNREACH: 229 _net("Rx Received ICMP Port Unreachable"); 230 break; 231 case ICMP_NET_UNKNOWN: 232 _net("Rx Received ICMP Unknown Network"); 233 break; 234 case ICMP_HOST_UNKNOWN: 235 _net("Rx Received ICMP Unknown Host"); 236 break; 237 default: 238 _net("Rx Received ICMP DestUnreach code=%u", 239 ee->ee_code); 240 break; 241 } 242 break; 243 244 case ICMP_TIME_EXCEEDED: 245 _net("Rx Received ICMP TTL Exceeded"); 246 break; 247 248 default: 249 _proto("Rx Received ICMP error { type=%u code=%u }", 250 ee->ee_type, ee->ee_code); 251 break; 252 } 253 break; 254 255 case SO_EE_ORIGIN_NONE: 256 case SO_EE_ORIGIN_LOCAL: 257 _proto("Rx Received local error { error=%d }", err); 258 err += RXRPC_LOCAL_ERROR_OFFSET; 259 break; 260 261 case SO_EE_ORIGIN_ICMP6: 262 default: 263 _proto("Rx Received error report { orig=%u }", ee->ee_origin); 264 break; 265 } 266 267 peer->error_report = err; 268 } 269 270 /* 271 * Distribute an error that occurred on a peer 272 */ 273 void rxrpc_peer_error_distributor(struct work_struct *work) 274 { 275 struct rxrpc_peer *peer = 276 container_of(work, struct rxrpc_peer, error_distributor); 277 struct rxrpc_call *call; 278 enum rxrpc_call_completion compl; 279 int error; 280 281 _enter(""); 282 283 error = READ_ONCE(peer->error_report); 284 if (error < RXRPC_LOCAL_ERROR_OFFSET) { 285 compl = RXRPC_CALL_NETWORK_ERROR; 286 } else { 287 compl = RXRPC_CALL_LOCAL_ERROR; 288 error -= RXRPC_LOCAL_ERROR_OFFSET; 289 } 290 291 _debug("ISSUE ERROR %s %d", rxrpc_call_completions[compl], error); 292 293 spin_lock_bh(&peer->lock); 294 295 while (!hlist_empty(&peer->error_targets)) { 296 call = hlist_entry(peer->error_targets.first, 297 struct rxrpc_call, error_link); 298 hlist_del_init(&call->error_link); 299 rxrpc_see_call(call); 300 301 if (rxrpc_set_call_completion(call, compl, 0, -error)) 302 rxrpc_notify_socket(call); 303 } 304 305 spin_unlock_bh(&peer->lock); 306 307 rxrpc_put_peer(peer); 308 _leave(""); 309 } 310 311 /* 312 * Add RTT information to cache. This is called in softirq mode and has 313 * exclusive access to the peer RTT data. 314 */ 315 void rxrpc_peer_add_rtt(struct rxrpc_call *call, enum rxrpc_rtt_rx_trace why, 316 rxrpc_serial_t send_serial, rxrpc_serial_t resp_serial, 317 ktime_t send_time, ktime_t resp_time) 318 { 319 struct rxrpc_peer *peer = call->peer; 320 s64 rtt; 321 u64 sum = peer->rtt_sum, avg; 322 u8 cursor = peer->rtt_cursor, usage = peer->rtt_usage; 323 324 rtt = ktime_to_ns(ktime_sub(resp_time, send_time)); 325 if (rtt < 0) 326 return; 327 328 /* Replace the oldest datum in the RTT buffer */ 329 sum -= peer->rtt_cache[cursor]; 330 sum += rtt; 331 peer->rtt_cache[cursor] = rtt; 332 peer->rtt_cursor = (cursor + 1) & (RXRPC_RTT_CACHE_SIZE - 1); 333 peer->rtt_sum = sum; 334 if (usage < RXRPC_RTT_CACHE_SIZE) { 335 usage++; 336 peer->rtt_usage = usage; 337 } 338 339 /* Now recalculate the average */ 340 if (usage == RXRPC_RTT_CACHE_SIZE) { 341 avg = sum / RXRPC_RTT_CACHE_SIZE; 342 } else { 343 avg = sum; 344 do_div(avg, usage); 345 } 346 347 peer->rtt = avg; 348 trace_rxrpc_rtt_rx(call, why, send_serial, resp_serial, rtt, 349 usage, avg); 350 } 351 352 /* 353 * Perform keep-alive pings with VERSION packets to keep any NAT alive. 354 */ 355 void rxrpc_peer_keepalive_worker(struct work_struct *work) 356 { 357 struct rxrpc_net *rxnet = 358 container_of(work, struct rxrpc_net, peer_keepalive_work); 359 struct rxrpc_peer *peer; 360 unsigned long delay; 361 ktime_t base, now = ktime_get_real(); 362 s64 diff; 363 u8 cursor, slot; 364 365 base = rxnet->peer_keepalive_base; 366 cursor = rxnet->peer_keepalive_cursor; 367 368 _enter("%u,%lld", cursor, ktime_sub(now, base)); 369 370 next_bucket: 371 diff = ktime_to_ns(ktime_sub(now, base)); 372 if (diff < 0) 373 goto resched; 374 375 _debug("at %u", cursor); 376 spin_lock_bh(&rxnet->peer_hash_lock); 377 next_peer: 378 if (!rxnet->live) { 379 spin_unlock_bh(&rxnet->peer_hash_lock); 380 goto out; 381 } 382 383 /* Everything in the bucket at the cursor is processed this second; the 384 * bucket at cursor + 1 goes now + 1s and so on... 385 */ 386 if (hlist_empty(&rxnet->peer_keepalive[cursor])) { 387 if (hlist_empty(&rxnet->peer_keepalive_new)) { 388 spin_unlock_bh(&rxnet->peer_hash_lock); 389 goto emptied_bucket; 390 } 391 392 hlist_move_list(&rxnet->peer_keepalive_new, 393 &rxnet->peer_keepalive[cursor]); 394 } 395 396 peer = hlist_entry(rxnet->peer_keepalive[cursor].first, 397 struct rxrpc_peer, keepalive_link); 398 hlist_del_init(&peer->keepalive_link); 399 if (!rxrpc_get_peer_maybe(peer)) 400 goto next_peer; 401 402 spin_unlock_bh(&rxnet->peer_hash_lock); 403 404 _debug("peer %u {%pISp}", peer->debug_id, &peer->srx.transport); 405 406 recalc: 407 diff = ktime_divns(ktime_sub(peer->last_tx_at, base), NSEC_PER_SEC); 408 if (diff < -30 || diff > 30) 409 goto send; /* LSW of 64-bit time probably wrapped on 32-bit */ 410 diff += RXRPC_KEEPALIVE_TIME - 1; 411 if (diff < 0) 412 goto send; 413 414 slot = (diff > RXRPC_KEEPALIVE_TIME - 1) ? RXRPC_KEEPALIVE_TIME - 1 : diff; 415 if (slot == 0) 416 goto send; 417 418 /* A transmission to this peer occurred since last we examined it so 419 * put it into the appropriate future bucket. 420 */ 421 slot = (slot + cursor) % ARRAY_SIZE(rxnet->peer_keepalive); 422 spin_lock_bh(&rxnet->peer_hash_lock); 423 hlist_add_head(&peer->keepalive_link, &rxnet->peer_keepalive[slot]); 424 rxrpc_put_peer(peer); 425 goto next_peer; 426 427 send: 428 rxrpc_send_keepalive(peer); 429 now = ktime_get_real(); 430 goto recalc; 431 432 emptied_bucket: 433 cursor++; 434 if (cursor >= ARRAY_SIZE(rxnet->peer_keepalive)) 435 cursor = 0; 436 base = ktime_add_ns(base, NSEC_PER_SEC); 437 goto next_bucket; 438 439 resched: 440 rxnet->peer_keepalive_base = base; 441 rxnet->peer_keepalive_cursor = cursor; 442 delay = nsecs_to_jiffies(-diff) + 1; 443 timer_reduce(&rxnet->peer_keepalive_timer, jiffies + delay); 444 out: 445 _leave(""); 446 } 447