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