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 8 #include <linux/module.h> 9 #include <linux/net.h> 10 #include <linux/skbuff.h> 11 #include <linux/errqueue.h> 12 #include <linux/udp.h> 13 #include <linux/in.h> 14 #include <linux/in6.h> 15 #include <linux/icmp.h> 16 #include <net/sock.h> 17 #include <net/af_rxrpc.h> 18 #include <net/ip.h> 19 #include "ar-internal.h" 20 21 static void rxrpc_store_error(struct rxrpc_peer *, struct sock_exterr_skb *); 22 static void rxrpc_distribute_error(struct rxrpc_peer *, int, 23 enum rxrpc_call_completion); 24 25 /* 26 * Find the peer associated with a local error. 27 */ 28 static struct rxrpc_peer *rxrpc_lookup_peer_local_rcu(struct rxrpc_local *local, 29 const struct sk_buff *skb, 30 struct sockaddr_rxrpc *srx) 31 { 32 struct sock_exterr_skb *serr = SKB_EXT_ERR(skb); 33 34 _enter(""); 35 36 memset(srx, 0, sizeof(*srx)); 37 srx->transport_type = local->srx.transport_type; 38 srx->transport_len = local->srx.transport_len; 39 srx->transport.family = local->srx.transport.family; 40 41 /* Can we see an ICMP4 packet on an ICMP6 listening socket? and vice 42 * versa? 43 */ 44 switch (srx->transport.family) { 45 case AF_INET: 46 srx->transport_len = sizeof(srx->transport.sin); 47 srx->transport.family = 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 switch (serr->ee.ee_origin) { 72 case SO_EE_ORIGIN_ICMP6: 73 _net("Rx ICMP6"); 74 srx->transport.sin6.sin6_port = serr->port; 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_len = sizeof(srx->transport.sin); 82 srx->transport.family = AF_INET; 83 srx->transport.sin.sin_port = serr->port; 84 memcpy(&srx->transport.sin.sin_addr, 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, unsigned int mtu) 108 { 109 _net("Rx ICMP Fragmentation Needed (%d)", mtu); 110 111 /* wind down the local interface MTU */ 112 if (mtu > 0 && peer->if_mtu == 65535 && mtu < peer->if_mtu) { 113 peer->if_mtu = mtu; 114 _net("I/F MTU %u", mtu); 115 } 116 117 if (mtu == 0) { 118 /* they didn't give us a size, estimate one */ 119 mtu = peer->if_mtu; 120 if (mtu > 1500) { 121 mtu >>= 1; 122 if (mtu < 1500) 123 mtu = 1500; 124 } else { 125 mtu -= 100; 126 if (mtu < peer->hdrsize) 127 mtu = peer->hdrsize + 4; 128 } 129 } 130 131 if (mtu < peer->mtu) { 132 spin_lock_bh(&peer->lock); 133 peer->mtu = mtu; 134 peer->maxdata = peer->mtu - peer->hdrsize; 135 spin_unlock_bh(&peer->lock); 136 _net("Net MTU %u (maxdata %u)", 137 peer->mtu, peer->maxdata); 138 } 139 } 140 141 /* 142 * Handle an error received on the local endpoint. 143 */ 144 void rxrpc_error_report(struct sock *sk) 145 { 146 struct sock_exterr_skb *serr; 147 struct sockaddr_rxrpc srx; 148 struct rxrpc_local *local; 149 struct rxrpc_peer *peer = NULL; 150 struct sk_buff *skb; 151 152 rcu_read_lock(); 153 local = rcu_dereference_sk_user_data(sk); 154 if (unlikely(!local)) { 155 rcu_read_unlock(); 156 return; 157 } 158 _enter("%p{%d}", sk, local->debug_id); 159 160 /* Clear the outstanding error value on the socket so that it doesn't 161 * cause kernel_sendmsg() to return it later. 162 */ 163 sock_error(sk); 164 165 skb = sock_dequeue_err_skb(sk); 166 if (!skb) { 167 rcu_read_unlock(); 168 _leave("UDP socket errqueue empty"); 169 return; 170 } 171 rxrpc_new_skb(skb, rxrpc_skb_received); 172 serr = SKB_EXT_ERR(skb); 173 if (!skb->len && serr->ee.ee_origin == SO_EE_ORIGIN_TIMESTAMPING) { 174 _leave("UDP empty message"); 175 rcu_read_unlock(); 176 rxrpc_free_skb(skb, rxrpc_skb_freed); 177 return; 178 } 179 180 peer = rxrpc_lookup_peer_local_rcu(local, skb, &srx); 181 if (peer && !rxrpc_get_peer_maybe(peer)) 182 peer = NULL; 183 if (!peer) { 184 rcu_read_unlock(); 185 rxrpc_free_skb(skb, rxrpc_skb_freed); 186 _leave(" [no peer]"); 187 return; 188 } 189 190 trace_rxrpc_rx_icmp(peer, &serr->ee, &srx); 191 192 if ((serr->ee.ee_origin == SO_EE_ORIGIN_ICMP && 193 serr->ee.ee_type == ICMP_DEST_UNREACH && 194 serr->ee.ee_code == ICMP_FRAG_NEEDED)) { 195 rxrpc_adjust_mtu(peer, serr->ee.ee_info); 196 goto out; 197 } 198 199 rxrpc_store_error(peer, serr); 200 out: 201 rcu_read_unlock(); 202 rxrpc_free_skb(skb, rxrpc_skb_freed); 203 rxrpc_put_peer(peer); 204 205 _leave(""); 206 } 207 208 /* 209 * Map an error report to error codes on the peer record. 210 */ 211 static void rxrpc_store_error(struct rxrpc_peer *peer, 212 struct sock_exterr_skb *serr) 213 { 214 enum rxrpc_call_completion compl = RXRPC_CALL_NETWORK_ERROR; 215 struct sock_extended_err *ee; 216 int err; 217 218 _enter(""); 219 220 ee = &serr->ee; 221 222 err = ee->ee_errno; 223 224 switch (ee->ee_origin) { 225 case SO_EE_ORIGIN_ICMP: 226 switch (ee->ee_type) { 227 case ICMP_DEST_UNREACH: 228 switch (ee->ee_code) { 229 case ICMP_NET_UNREACH: 230 _net("Rx Received ICMP Network Unreachable"); 231 break; 232 case ICMP_HOST_UNREACH: 233 _net("Rx Received ICMP Host Unreachable"); 234 break; 235 case ICMP_PORT_UNREACH: 236 _net("Rx Received ICMP Port Unreachable"); 237 break; 238 case ICMP_NET_UNKNOWN: 239 _net("Rx Received ICMP Unknown Network"); 240 break; 241 case ICMP_HOST_UNKNOWN: 242 _net("Rx Received ICMP Unknown Host"); 243 break; 244 default: 245 _net("Rx Received ICMP DestUnreach code=%u", 246 ee->ee_code); 247 break; 248 } 249 break; 250 251 case ICMP_TIME_EXCEEDED: 252 _net("Rx Received ICMP TTL Exceeded"); 253 break; 254 255 default: 256 _proto("Rx Received ICMP error { type=%u code=%u }", 257 ee->ee_type, ee->ee_code); 258 break; 259 } 260 break; 261 262 case SO_EE_ORIGIN_NONE: 263 case SO_EE_ORIGIN_LOCAL: 264 _proto("Rx Received local error { error=%d }", err); 265 compl = RXRPC_CALL_LOCAL_ERROR; 266 break; 267 268 case SO_EE_ORIGIN_ICMP6: 269 if (err == EACCES) 270 err = EHOSTUNREACH; 271 fallthrough; 272 default: 273 _proto("Rx Received error report { orig=%u }", ee->ee_origin); 274 break; 275 } 276 277 rxrpc_distribute_error(peer, err, compl); 278 } 279 280 /* 281 * Distribute an error that occurred on a peer. 282 */ 283 static void rxrpc_distribute_error(struct rxrpc_peer *peer, int error, 284 enum rxrpc_call_completion compl) 285 { 286 struct rxrpc_call *call; 287 288 hlist_for_each_entry_rcu(call, &peer->error_targets, error_link) { 289 rxrpc_see_call(call); 290 rxrpc_set_call_completion(call, compl, 0, -error); 291 } 292 } 293 294 /* 295 * Perform keep-alive pings. 296 */ 297 static void rxrpc_peer_keepalive_dispatch(struct rxrpc_net *rxnet, 298 struct list_head *collector, 299 time64_t base, 300 u8 cursor) 301 { 302 struct rxrpc_peer *peer; 303 const u8 mask = ARRAY_SIZE(rxnet->peer_keepalive) - 1; 304 time64_t keepalive_at; 305 int slot; 306 307 spin_lock_bh(&rxnet->peer_hash_lock); 308 309 while (!list_empty(collector)) { 310 peer = list_entry(collector->next, 311 struct rxrpc_peer, keepalive_link); 312 313 list_del_init(&peer->keepalive_link); 314 if (!rxrpc_get_peer_maybe(peer)) 315 continue; 316 317 if (__rxrpc_use_local(peer->local)) { 318 spin_unlock_bh(&rxnet->peer_hash_lock); 319 320 keepalive_at = peer->last_tx_at + RXRPC_KEEPALIVE_TIME; 321 slot = keepalive_at - base; 322 _debug("%02x peer %u t=%d {%pISp}", 323 cursor, peer->debug_id, slot, &peer->srx.transport); 324 325 if (keepalive_at <= base || 326 keepalive_at > base + RXRPC_KEEPALIVE_TIME) { 327 rxrpc_send_keepalive(peer); 328 slot = RXRPC_KEEPALIVE_TIME; 329 } 330 331 /* A transmission to this peer occurred since last we 332 * examined it so put it into the appropriate future 333 * bucket. 334 */ 335 slot += cursor; 336 slot &= mask; 337 spin_lock_bh(&rxnet->peer_hash_lock); 338 list_add_tail(&peer->keepalive_link, 339 &rxnet->peer_keepalive[slot & mask]); 340 rxrpc_unuse_local(peer->local); 341 } 342 rxrpc_put_peer_locked(peer); 343 } 344 345 spin_unlock_bh(&rxnet->peer_hash_lock); 346 } 347 348 /* 349 * Perform keep-alive pings with VERSION packets to keep any NAT alive. 350 */ 351 void rxrpc_peer_keepalive_worker(struct work_struct *work) 352 { 353 struct rxrpc_net *rxnet = 354 container_of(work, struct rxrpc_net, peer_keepalive_work); 355 const u8 mask = ARRAY_SIZE(rxnet->peer_keepalive) - 1; 356 time64_t base, now, delay; 357 u8 cursor, stop; 358 LIST_HEAD(collector); 359 360 now = ktime_get_seconds(); 361 base = rxnet->peer_keepalive_base; 362 cursor = rxnet->peer_keepalive_cursor; 363 _enter("%lld,%u", base - now, cursor); 364 365 if (!rxnet->live) 366 return; 367 368 /* Remove to a temporary list all the peers that are currently lodged 369 * in expired buckets plus all new peers. 370 * 371 * Everything in the bucket at the cursor is processed this 372 * second; the bucket at cursor + 1 goes at now + 1s and so 373 * on... 374 */ 375 spin_lock_bh(&rxnet->peer_hash_lock); 376 list_splice_init(&rxnet->peer_keepalive_new, &collector); 377 378 stop = cursor + ARRAY_SIZE(rxnet->peer_keepalive); 379 while (base <= now && (s8)(cursor - stop) < 0) { 380 list_splice_tail_init(&rxnet->peer_keepalive[cursor & mask], 381 &collector); 382 base++; 383 cursor++; 384 } 385 386 base = now; 387 spin_unlock_bh(&rxnet->peer_hash_lock); 388 389 rxnet->peer_keepalive_base = base; 390 rxnet->peer_keepalive_cursor = cursor; 391 rxrpc_peer_keepalive_dispatch(rxnet, &collector, base, cursor); 392 ASSERT(list_empty(&collector)); 393 394 /* Schedule the timer for the next occupied timeslot. */ 395 cursor = rxnet->peer_keepalive_cursor; 396 stop = cursor + RXRPC_KEEPALIVE_TIME - 1; 397 for (; (s8)(cursor - stop) < 0; cursor++) { 398 if (!list_empty(&rxnet->peer_keepalive[cursor & mask])) 399 break; 400 base++; 401 } 402 403 now = ktime_get_seconds(); 404 delay = base - now; 405 if (delay < 1) 406 delay = 1; 407 delay *= HZ; 408 if (rxnet->live) 409 timer_reduce(&rxnet->peer_keepalive_timer, jiffies + delay); 410 411 _leave(""); 412 } 413