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 /* 86 * Allocate a new local endpoint. 87 */ 88 static struct rxrpc_local *rxrpc_alloc_local(struct net *net, 89 const struct sockaddr_rxrpc *srx) 90 { 91 struct rxrpc_local *local; 92 93 local = kzalloc(sizeof(struct rxrpc_local), GFP_KERNEL); 94 if (local) { 95 refcount_set(&local->ref, 1); 96 atomic_set(&local->active_users, 1); 97 local->net = net; 98 local->rxnet = rxrpc_net(net); 99 INIT_HLIST_NODE(&local->link); 100 init_rwsem(&local->defrag_sem); 101 init_completion(&local->io_thread_ready); 102 skb_queue_head_init(&local->rx_queue); 103 INIT_LIST_HEAD(&local->conn_attend_q); 104 INIT_LIST_HEAD(&local->call_attend_q); 105 local->client_bundles = RB_ROOT; 106 spin_lock_init(&local->client_bundles_lock); 107 spin_lock_init(&local->lock); 108 rwlock_init(&local->services_lock); 109 local->debug_id = atomic_inc_return(&rxrpc_debug_id); 110 memcpy(&local->srx, srx, sizeof(*srx)); 111 local->srx.srx_service = 0; 112 trace_rxrpc_local(local->debug_id, rxrpc_local_new, 1, 1); 113 } 114 115 _leave(" = %p", local); 116 return local; 117 } 118 119 /* 120 * create the local socket 121 * - must be called with rxrpc_local_mutex locked 122 */ 123 static int rxrpc_open_socket(struct rxrpc_local *local, struct net *net) 124 { 125 struct udp_tunnel_sock_cfg tuncfg = {NULL}; 126 struct sockaddr_rxrpc *srx = &local->srx; 127 struct udp_port_cfg udp_conf = {0}; 128 struct task_struct *io_thread; 129 struct sock *usk; 130 int ret; 131 132 _enter("%p{%d,%d}", 133 local, srx->transport_type, srx->transport.family); 134 135 udp_conf.family = srx->transport.family; 136 udp_conf.use_udp_checksums = true; 137 if (udp_conf.family == AF_INET) { 138 udp_conf.local_ip = srx->transport.sin.sin_addr; 139 udp_conf.local_udp_port = srx->transport.sin.sin_port; 140 #if IS_ENABLED(CONFIG_AF_RXRPC_IPV6) 141 } else { 142 udp_conf.local_ip6 = srx->transport.sin6.sin6_addr; 143 udp_conf.local_udp_port = srx->transport.sin6.sin6_port; 144 udp_conf.use_udp6_tx_checksums = true; 145 udp_conf.use_udp6_rx_checksums = true; 146 #endif 147 } 148 ret = udp_sock_create(net, &udp_conf, &local->socket); 149 if (ret < 0) { 150 _leave(" = %d [socket]", ret); 151 return ret; 152 } 153 154 tuncfg.encap_type = UDP_ENCAP_RXRPC; 155 tuncfg.encap_rcv = rxrpc_encap_rcv; 156 tuncfg.encap_err_rcv = rxrpc_encap_err_rcv; 157 tuncfg.sk_user_data = local; 158 setup_udp_tunnel_sock(net, local->socket, &tuncfg); 159 160 /* set the socket up */ 161 usk = local->socket->sk; 162 usk->sk_error_report = rxrpc_error_report; 163 164 switch (srx->transport.family) { 165 case AF_INET6: 166 /* we want to receive ICMPv6 errors */ 167 ip6_sock_set_recverr(usk); 168 169 /* Fall through and set IPv4 options too otherwise we don't get 170 * errors from IPv4 packets sent through the IPv6 socket. 171 */ 172 fallthrough; 173 case AF_INET: 174 /* we want to receive ICMP errors */ 175 ip_sock_set_recverr(usk); 176 177 /* we want to set the don't fragment bit */ 178 ip_sock_set_mtu_discover(usk, IP_PMTUDISC_DO); 179 180 /* We want receive timestamps. */ 181 sock_enable_timestamps(usk); 182 break; 183 184 default: 185 BUG(); 186 } 187 188 io_thread = kthread_run(rxrpc_io_thread, local, 189 "krxrpcio/%u", ntohs(udp_conf.local_udp_port)); 190 if (IS_ERR(io_thread)) { 191 ret = PTR_ERR(io_thread); 192 goto error_sock; 193 } 194 195 wait_for_completion(&local->io_thread_ready); 196 local->io_thread = io_thread; 197 _leave(" = 0"); 198 return 0; 199 200 error_sock: 201 kernel_sock_shutdown(local->socket, SHUT_RDWR); 202 local->socket->sk->sk_user_data = NULL; 203 sock_release(local->socket); 204 local->socket = NULL; 205 return ret; 206 } 207 208 /* 209 * Look up or create a new local endpoint using the specified local address. 210 */ 211 struct rxrpc_local *rxrpc_lookup_local(struct net *net, 212 const struct sockaddr_rxrpc *srx) 213 { 214 struct rxrpc_local *local; 215 struct rxrpc_net *rxnet = rxrpc_net(net); 216 struct hlist_node *cursor; 217 long diff; 218 int ret; 219 220 _enter("{%d,%d,%pISp}", 221 srx->transport_type, srx->transport.family, &srx->transport); 222 223 mutex_lock(&rxnet->local_mutex); 224 225 hlist_for_each(cursor, &rxnet->local_endpoints) { 226 local = hlist_entry(cursor, struct rxrpc_local, link); 227 228 diff = rxrpc_local_cmp_key(local, srx); 229 if (diff != 0) 230 continue; 231 232 /* Services aren't allowed to share transport sockets, so 233 * reject that here. It is possible that the object is dying - 234 * but it may also still have the local transport address that 235 * we want bound. 236 */ 237 if (srx->srx_service) { 238 local = NULL; 239 goto addr_in_use; 240 } 241 242 /* Found a match. We want to replace a dying object. 243 * Attempting to bind the transport socket may still fail if 244 * we're attempting to use a local address that the dying 245 * object is still using. 246 */ 247 if (!rxrpc_use_local(local, rxrpc_local_use_lookup)) 248 break; 249 250 goto found; 251 } 252 253 local = rxrpc_alloc_local(net, srx); 254 if (!local) 255 goto nomem; 256 257 ret = rxrpc_open_socket(local, net); 258 if (ret < 0) 259 goto sock_error; 260 261 if (cursor) { 262 hlist_replace_rcu(cursor, &local->link); 263 cursor->pprev = NULL; 264 } else { 265 hlist_add_head_rcu(&local->link, &rxnet->local_endpoints); 266 } 267 268 found: 269 mutex_unlock(&rxnet->local_mutex); 270 _leave(" = %p", local); 271 return local; 272 273 nomem: 274 ret = -ENOMEM; 275 sock_error: 276 mutex_unlock(&rxnet->local_mutex); 277 if (local) 278 call_rcu(&local->rcu, rxrpc_local_rcu); 279 _leave(" = %d", ret); 280 return ERR_PTR(ret); 281 282 addr_in_use: 283 mutex_unlock(&rxnet->local_mutex); 284 _leave(" = -EADDRINUSE"); 285 return ERR_PTR(-EADDRINUSE); 286 } 287 288 /* 289 * Get a ref on a local endpoint. 290 */ 291 struct rxrpc_local *rxrpc_get_local(struct rxrpc_local *local, 292 enum rxrpc_local_trace why) 293 { 294 int r, u; 295 296 u = atomic_read(&local->active_users); 297 __refcount_inc(&local->ref, &r); 298 trace_rxrpc_local(local->debug_id, why, r + 1, u); 299 return local; 300 } 301 302 /* 303 * Get a ref on a local endpoint unless its usage has already reached 0. 304 */ 305 struct rxrpc_local *rxrpc_get_local_maybe(struct rxrpc_local *local, 306 enum rxrpc_local_trace why) 307 { 308 int r, u; 309 310 if (local && __refcount_inc_not_zero(&local->ref, &r)) { 311 u = atomic_read(&local->active_users); 312 trace_rxrpc_local(local->debug_id, why, r + 1, u); 313 return local; 314 } 315 316 return NULL; 317 } 318 319 /* 320 * Drop a ref on a local endpoint. 321 */ 322 void rxrpc_put_local(struct rxrpc_local *local, enum rxrpc_local_trace why) 323 { 324 unsigned int debug_id; 325 bool dead; 326 int r, u; 327 328 if (local) { 329 debug_id = local->debug_id; 330 331 u = atomic_read(&local->active_users); 332 dead = __refcount_dec_and_test(&local->ref, &r); 333 trace_rxrpc_local(debug_id, why, r, u); 334 335 if (dead) 336 call_rcu(&local->rcu, rxrpc_local_rcu); 337 } 338 } 339 340 /* 341 * Start using a local endpoint. 342 */ 343 struct rxrpc_local *rxrpc_use_local(struct rxrpc_local *local, 344 enum rxrpc_local_trace why) 345 { 346 local = rxrpc_get_local_maybe(local, rxrpc_local_get_for_use); 347 if (!local) 348 return NULL; 349 350 if (!__rxrpc_use_local(local, why)) { 351 rxrpc_put_local(local, rxrpc_local_put_for_use); 352 return NULL; 353 } 354 355 return local; 356 } 357 358 /* 359 * Cease using a local endpoint. Once the number of active users reaches 0, we 360 * start the closure of the transport in the I/O thread.. 361 */ 362 void rxrpc_unuse_local(struct rxrpc_local *local, enum rxrpc_local_trace why) 363 { 364 unsigned int debug_id; 365 int r, u; 366 367 if (local) { 368 debug_id = local->debug_id; 369 r = refcount_read(&local->ref); 370 u = atomic_dec_return(&local->active_users); 371 trace_rxrpc_local(debug_id, why, r, u); 372 if (u == 0) 373 kthread_stop(local->io_thread); 374 } 375 } 376 377 /* 378 * Destroy a local endpoint's socket and then hand the record to RCU to dispose 379 * of. 380 * 381 * Closing the socket cannot be done from bottom half context or RCU callback 382 * context because it might sleep. 383 */ 384 void rxrpc_destroy_local(struct rxrpc_local *local) 385 { 386 struct socket *socket = local->socket; 387 struct rxrpc_net *rxnet = local->rxnet; 388 389 _enter("%d", local->debug_id); 390 391 local->dead = true; 392 393 mutex_lock(&rxnet->local_mutex); 394 hlist_del_init_rcu(&local->link); 395 mutex_unlock(&rxnet->local_mutex); 396 397 rxrpc_clean_up_local_conns(local); 398 rxrpc_service_connection_reaper(&rxnet->service_conn_reaper); 399 ASSERT(!local->service); 400 401 if (socket) { 402 local->socket = NULL; 403 kernel_sock_shutdown(socket, SHUT_RDWR); 404 socket->sk->sk_user_data = NULL; 405 sock_release(socket); 406 } 407 408 /* At this point, there should be no more packets coming in to the 409 * local endpoint. 410 */ 411 rxrpc_purge_queue(&local->rx_queue); 412 } 413 414 /* 415 * Destroy a local endpoint after the RCU grace period expires. 416 */ 417 static void rxrpc_local_rcu(struct rcu_head *rcu) 418 { 419 struct rxrpc_local *local = container_of(rcu, struct rxrpc_local, rcu); 420 421 rxrpc_see_local(local, rxrpc_local_free); 422 kfree(local); 423 } 424 425 /* 426 * Verify the local endpoint list is empty by this point. 427 */ 428 void rxrpc_destroy_all_locals(struct rxrpc_net *rxnet) 429 { 430 struct rxrpc_local *local; 431 432 _enter(""); 433 434 flush_workqueue(rxrpc_workqueue); 435 436 if (!hlist_empty(&rxnet->local_endpoints)) { 437 mutex_lock(&rxnet->local_mutex); 438 hlist_for_each_entry(local, &rxnet->local_endpoints, link) { 439 pr_err("AF_RXRPC: Leaked local %p {%d}\n", 440 local, refcount_read(&local->ref)); 441 } 442 mutex_unlock(&rxnet->local_mutex); 443 BUG(); 444 } 445 } 446