1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* Local endpoint object management 3 * 4 * Copyright (C) 2016 Red Hat, Inc. All Rights Reserved. 5 * Written by David Howells (dhowells@redhat.com) 6 */ 7 8 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 9 10 #include <linux/module.h> 11 #include <linux/net.h> 12 #include <linux/skbuff.h> 13 #include <linux/slab.h> 14 #include <linux/udp.h> 15 #include <linux/ip.h> 16 #include <linux/hashtable.h> 17 #include <net/sock.h> 18 #include <net/udp.h> 19 #include <net/udp_tunnel.h> 20 #include <net/af_rxrpc.h> 21 #include "ar-internal.h" 22 23 static void rxrpc_local_rcu(struct rcu_head *); 24 25 /* 26 * Handle an ICMP/ICMP6 error turning up at the tunnel. Push it through the 27 * usual mechanism so that it gets parsed and presented through the UDP 28 * socket's error_report(). 29 */ 30 static void rxrpc_encap_err_rcv(struct sock *sk, struct sk_buff *skb, int err, 31 __be16 port, u32 info, u8 *payload) 32 { 33 if (ip_hdr(skb)->version == IPVERSION) 34 return ip_icmp_error(sk, skb, err, port, info, payload); 35 if (IS_ENABLED(CONFIG_AF_RXRPC_IPV6)) 36 return ipv6_icmp_error(sk, skb, err, port, info, payload); 37 } 38 39 /* 40 * Compare a local to an address. Return -ve, 0 or +ve to indicate less than, 41 * same or greater than. 42 * 43 * We explicitly don't compare the RxRPC service ID as we want to reject 44 * conflicting uses by differing services. Further, we don't want to share 45 * addresses with different options (IPv6), so we don't compare those bits 46 * either. 47 */ 48 static long rxrpc_local_cmp_key(const struct rxrpc_local *local, 49 const struct sockaddr_rxrpc *srx) 50 { 51 long diff; 52 53 diff = ((local->srx.transport_type - srx->transport_type) ?: 54 (local->srx.transport_len - srx->transport_len) ?: 55 (local->srx.transport.family - srx->transport.family)); 56 if (diff != 0) 57 return diff; 58 59 switch (srx->transport.family) { 60 case AF_INET: 61 /* If the choice of UDP port is left up to the transport, then 62 * the endpoint record doesn't match. 63 */ 64 return ((u16 __force)local->srx.transport.sin.sin_port - 65 (u16 __force)srx->transport.sin.sin_port) ?: 66 memcmp(&local->srx.transport.sin.sin_addr, 67 &srx->transport.sin.sin_addr, 68 sizeof(struct in_addr)); 69 #ifdef CONFIG_AF_RXRPC_IPV6 70 case AF_INET6: 71 /* If the choice of UDP6 port is left up to the transport, then 72 * the endpoint record doesn't match. 73 */ 74 return ((u16 __force)local->srx.transport.sin6.sin6_port - 75 (u16 __force)srx->transport.sin6.sin6_port) ?: 76 memcmp(&local->srx.transport.sin6.sin6_addr, 77 &srx->transport.sin6.sin6_addr, 78 sizeof(struct in6_addr)); 79 #endif 80 default: 81 BUG(); 82 } 83 } 84 85 static void rxrpc_client_conn_reap_timeout(struct timer_list *timer) 86 { 87 struct rxrpc_local *local = 88 container_of(timer, struct rxrpc_local, client_conn_reap_timer); 89 90 if (local->kill_all_client_conns && 91 test_and_set_bit(RXRPC_CLIENT_CONN_REAP_TIMER, &local->client_conn_flags)) 92 rxrpc_wake_up_io_thread(local); 93 } 94 95 /* 96 * Allocate a new local endpoint. 97 */ 98 static struct rxrpc_local *rxrpc_alloc_local(struct net *net, 99 const struct sockaddr_rxrpc *srx) 100 { 101 struct rxrpc_local *local; 102 u32 tmp; 103 104 local = kzalloc(sizeof(struct rxrpc_local), GFP_KERNEL); 105 if (local) { 106 refcount_set(&local->ref, 1); 107 atomic_set(&local->active_users, 1); 108 local->net = net; 109 local->rxnet = rxrpc_net(net); 110 INIT_HLIST_NODE(&local->link); 111 init_rwsem(&local->defrag_sem); 112 init_completion(&local->io_thread_ready); 113 skb_queue_head_init(&local->rx_queue); 114 INIT_LIST_HEAD(&local->conn_attend_q); 115 INIT_LIST_HEAD(&local->call_attend_q); 116 117 local->client_bundles = RB_ROOT; 118 spin_lock_init(&local->client_bundles_lock); 119 local->kill_all_client_conns = false; 120 INIT_LIST_HEAD(&local->idle_client_conns); 121 timer_setup(&local->client_conn_reap_timer, 122 rxrpc_client_conn_reap_timeout, 0); 123 124 spin_lock_init(&local->lock); 125 rwlock_init(&local->services_lock); 126 local->debug_id = atomic_inc_return(&rxrpc_debug_id); 127 memcpy(&local->srx, srx, sizeof(*srx)); 128 local->srx.srx_service = 0; 129 idr_init(&local->conn_ids); 130 get_random_bytes(&tmp, sizeof(tmp)); 131 tmp &= 0x3fffffff; 132 if (tmp == 0) 133 tmp = 1; 134 idr_set_cursor(&local->conn_ids, tmp); 135 INIT_LIST_HEAD(&local->new_client_calls); 136 spin_lock_init(&local->client_call_lock); 137 138 trace_rxrpc_local(local->debug_id, rxrpc_local_new, 1, 1); 139 } 140 141 _leave(" = %p", local); 142 return local; 143 } 144 145 /* 146 * create the local socket 147 * - must be called with rxrpc_local_mutex locked 148 */ 149 static int rxrpc_open_socket(struct rxrpc_local *local, struct net *net) 150 { 151 struct udp_tunnel_sock_cfg tuncfg = {NULL}; 152 struct sockaddr_rxrpc *srx = &local->srx; 153 struct udp_port_cfg udp_conf = {0}; 154 struct task_struct *io_thread; 155 struct sock *usk; 156 int ret; 157 158 _enter("%p{%d,%d}", 159 local, srx->transport_type, srx->transport.family); 160 161 udp_conf.family = srx->transport.family; 162 udp_conf.use_udp_checksums = true; 163 if (udp_conf.family == AF_INET) { 164 udp_conf.local_ip = srx->transport.sin.sin_addr; 165 udp_conf.local_udp_port = srx->transport.sin.sin_port; 166 #if IS_ENABLED(CONFIG_AF_RXRPC_IPV6) 167 } else { 168 udp_conf.local_ip6 = srx->transport.sin6.sin6_addr; 169 udp_conf.local_udp_port = srx->transport.sin6.sin6_port; 170 udp_conf.use_udp6_tx_checksums = true; 171 udp_conf.use_udp6_rx_checksums = true; 172 #endif 173 } 174 ret = udp_sock_create(net, &udp_conf, &local->socket); 175 if (ret < 0) { 176 _leave(" = %d [socket]", ret); 177 return ret; 178 } 179 180 tuncfg.encap_type = UDP_ENCAP_RXRPC; 181 tuncfg.encap_rcv = rxrpc_encap_rcv; 182 tuncfg.encap_err_rcv = rxrpc_encap_err_rcv; 183 tuncfg.sk_user_data = local; 184 setup_udp_tunnel_sock(net, local->socket, &tuncfg); 185 186 /* set the socket up */ 187 usk = local->socket->sk; 188 usk->sk_error_report = rxrpc_error_report; 189 190 switch (srx->transport.family) { 191 case AF_INET6: 192 /* we want to receive ICMPv6 errors */ 193 ip6_sock_set_recverr(usk); 194 195 /* Fall through and set IPv4 options too otherwise we don't get 196 * errors from IPv4 packets sent through the IPv6 socket. 197 */ 198 fallthrough; 199 case AF_INET: 200 /* we want to receive ICMP errors */ 201 ip_sock_set_recverr(usk); 202 203 /* we want to set the don't fragment bit */ 204 ip_sock_set_mtu_discover(usk, IP_PMTUDISC_DO); 205 206 /* We want receive timestamps. */ 207 sock_enable_timestamps(usk); 208 break; 209 210 default: 211 BUG(); 212 } 213 214 io_thread = kthread_run(rxrpc_io_thread, local, 215 "krxrpcio/%u", ntohs(udp_conf.local_udp_port)); 216 if (IS_ERR(io_thread)) { 217 ret = PTR_ERR(io_thread); 218 goto error_sock; 219 } 220 221 wait_for_completion(&local->io_thread_ready); 222 local->io_thread = io_thread; 223 _leave(" = 0"); 224 return 0; 225 226 error_sock: 227 kernel_sock_shutdown(local->socket, SHUT_RDWR); 228 local->socket->sk->sk_user_data = NULL; 229 sock_release(local->socket); 230 local->socket = NULL; 231 return ret; 232 } 233 234 /* 235 * Look up or create a new local endpoint using the specified local address. 236 */ 237 struct rxrpc_local *rxrpc_lookup_local(struct net *net, 238 const struct sockaddr_rxrpc *srx) 239 { 240 struct rxrpc_local *local; 241 struct rxrpc_net *rxnet = rxrpc_net(net); 242 struct hlist_node *cursor; 243 long diff; 244 int ret; 245 246 _enter("{%d,%d,%pISp}", 247 srx->transport_type, srx->transport.family, &srx->transport); 248 249 mutex_lock(&rxnet->local_mutex); 250 251 hlist_for_each(cursor, &rxnet->local_endpoints) { 252 local = hlist_entry(cursor, struct rxrpc_local, link); 253 254 diff = rxrpc_local_cmp_key(local, srx); 255 if (diff != 0) 256 continue; 257 258 /* Services aren't allowed to share transport sockets, so 259 * reject that here. It is possible that the object is dying - 260 * but it may also still have the local transport address that 261 * we want bound. 262 */ 263 if (srx->srx_service) { 264 local = NULL; 265 goto addr_in_use; 266 } 267 268 /* Found a match. We want to replace a dying object. 269 * Attempting to bind the transport socket may still fail if 270 * we're attempting to use a local address that the dying 271 * object is still using. 272 */ 273 if (!rxrpc_use_local(local, rxrpc_local_use_lookup)) 274 break; 275 276 goto found; 277 } 278 279 local = rxrpc_alloc_local(net, srx); 280 if (!local) 281 goto nomem; 282 283 ret = rxrpc_open_socket(local, net); 284 if (ret < 0) 285 goto sock_error; 286 287 if (cursor) { 288 hlist_replace_rcu(cursor, &local->link); 289 cursor->pprev = NULL; 290 } else { 291 hlist_add_head_rcu(&local->link, &rxnet->local_endpoints); 292 } 293 294 found: 295 mutex_unlock(&rxnet->local_mutex); 296 _leave(" = %p", local); 297 return local; 298 299 nomem: 300 ret = -ENOMEM; 301 sock_error: 302 mutex_unlock(&rxnet->local_mutex); 303 if (local) 304 call_rcu(&local->rcu, rxrpc_local_rcu); 305 _leave(" = %d", ret); 306 return ERR_PTR(ret); 307 308 addr_in_use: 309 mutex_unlock(&rxnet->local_mutex); 310 _leave(" = -EADDRINUSE"); 311 return ERR_PTR(-EADDRINUSE); 312 } 313 314 /* 315 * Get a ref on a local endpoint. 316 */ 317 struct rxrpc_local *rxrpc_get_local(struct rxrpc_local *local, 318 enum rxrpc_local_trace why) 319 { 320 int r, u; 321 322 u = atomic_read(&local->active_users); 323 __refcount_inc(&local->ref, &r); 324 trace_rxrpc_local(local->debug_id, why, r + 1, u); 325 return local; 326 } 327 328 /* 329 * Get a ref on a local endpoint unless its usage has already reached 0. 330 */ 331 struct rxrpc_local *rxrpc_get_local_maybe(struct rxrpc_local *local, 332 enum rxrpc_local_trace why) 333 { 334 int r, u; 335 336 if (local && __refcount_inc_not_zero(&local->ref, &r)) { 337 u = atomic_read(&local->active_users); 338 trace_rxrpc_local(local->debug_id, why, r + 1, u); 339 return local; 340 } 341 342 return NULL; 343 } 344 345 /* 346 * Drop a ref on a local endpoint. 347 */ 348 void rxrpc_put_local(struct rxrpc_local *local, enum rxrpc_local_trace why) 349 { 350 unsigned int debug_id; 351 bool dead; 352 int r, u; 353 354 if (local) { 355 debug_id = local->debug_id; 356 357 u = atomic_read(&local->active_users); 358 dead = __refcount_dec_and_test(&local->ref, &r); 359 trace_rxrpc_local(debug_id, why, r, u); 360 361 if (dead) 362 call_rcu(&local->rcu, rxrpc_local_rcu); 363 } 364 } 365 366 /* 367 * Start using a local endpoint. 368 */ 369 struct rxrpc_local *rxrpc_use_local(struct rxrpc_local *local, 370 enum rxrpc_local_trace why) 371 { 372 local = rxrpc_get_local_maybe(local, rxrpc_local_get_for_use); 373 if (!local) 374 return NULL; 375 376 if (!__rxrpc_use_local(local, why)) { 377 rxrpc_put_local(local, rxrpc_local_put_for_use); 378 return NULL; 379 } 380 381 return local; 382 } 383 384 /* 385 * Cease using a local endpoint. Once the number of active users reaches 0, we 386 * start the closure of the transport in the I/O thread.. 387 */ 388 void rxrpc_unuse_local(struct rxrpc_local *local, enum rxrpc_local_trace why) 389 { 390 unsigned int debug_id; 391 int r, u; 392 393 if (local) { 394 debug_id = local->debug_id; 395 r = refcount_read(&local->ref); 396 u = atomic_dec_return(&local->active_users); 397 trace_rxrpc_local(debug_id, why, r, u); 398 if (u == 0) 399 kthread_stop(local->io_thread); 400 } 401 } 402 403 /* 404 * Destroy a local endpoint's socket and then hand the record to RCU to dispose 405 * of. 406 * 407 * Closing the socket cannot be done from bottom half context or RCU callback 408 * context because it might sleep. 409 */ 410 void rxrpc_destroy_local(struct rxrpc_local *local) 411 { 412 struct socket *socket = local->socket; 413 struct rxrpc_net *rxnet = local->rxnet; 414 415 _enter("%d", local->debug_id); 416 417 local->dead = true; 418 419 mutex_lock(&rxnet->local_mutex); 420 hlist_del_init_rcu(&local->link); 421 mutex_unlock(&rxnet->local_mutex); 422 423 rxrpc_clean_up_local_conns(local); 424 rxrpc_service_connection_reaper(&rxnet->service_conn_reaper); 425 ASSERT(!local->service); 426 427 if (socket) { 428 local->socket = NULL; 429 kernel_sock_shutdown(socket, SHUT_RDWR); 430 socket->sk->sk_user_data = NULL; 431 sock_release(socket); 432 } 433 434 /* At this point, there should be no more packets coming in to the 435 * local endpoint. 436 */ 437 rxrpc_purge_queue(&local->rx_queue); 438 rxrpc_purge_client_connections(local); 439 } 440 441 /* 442 * Destroy a local endpoint after the RCU grace period expires. 443 */ 444 static void rxrpc_local_rcu(struct rcu_head *rcu) 445 { 446 struct rxrpc_local *local = container_of(rcu, struct rxrpc_local, rcu); 447 448 rxrpc_see_local(local, rxrpc_local_free); 449 kfree(local); 450 } 451 452 /* 453 * Verify the local endpoint list is empty by this point. 454 */ 455 void rxrpc_destroy_all_locals(struct rxrpc_net *rxnet) 456 { 457 struct rxrpc_local *local; 458 459 _enter(""); 460 461 flush_workqueue(rxrpc_workqueue); 462 463 if (!hlist_empty(&rxnet->local_endpoints)) { 464 mutex_lock(&rxnet->local_mutex); 465 hlist_for_each_entry(local, &rxnet->local_endpoints, link) { 466 pr_err("AF_RXRPC: Leaked local %p {%d}\n", 467 local, refcount_read(&local->ref)); 468 } 469 mutex_unlock(&rxnet->local_mutex); 470 BUG(); 471 } 472 } 473