| 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 | * Set or clear the Don't Fragment flag on a socket. |
| 41 | */ |
| 42 | void rxrpc_local_dont_fragment(const struct rxrpc_local *local, bool set) |
| 43 | { |
| 44 | if (set) |
| 45 | ip_sock_set_mtu_discover(local->socket->sk, IP_PMTUDISC_DO); |
| 46 | else |
| 47 | ip_sock_set_mtu_discover(local->socket->sk, IP_PMTUDISC_DONT); |
| 48 | } |
| 49 | |
| 50 | /* |
| 51 | * Compare a local to an address. Return -ve, 0 or +ve to indicate less than, |
| 52 | * same or greater than. |
| 53 | * |
| 54 | * We explicitly don't compare the RxRPC service ID as we want to reject |
| 55 | * conflicting uses by differing services. Further, we don't want to share |
| 56 | * addresses with different options (IPv6), so we don't compare those bits |
| 57 | * either. |
| 58 | */ |
| 59 | static long rxrpc_local_cmp_key(const struct rxrpc_local *local, |
| 60 | const struct sockaddr_rxrpc *srx) |
| 61 | { |
| 62 | long diff; |
| 63 | |
| 64 | diff = ((local->srx.transport_type - srx->transport_type) ?: |
| 65 | (local->srx.transport_len - srx->transport_len) ?: |
| 66 | (local->srx.transport.family - srx->transport.family)); |
| 67 | if (diff != 0) |
| 68 | return diff; |
| 69 | |
| 70 | switch (srx->transport.family) { |
| 71 | case AF_INET: |
| 72 | /* If the choice of UDP port is left up to the transport, then |
| 73 | * the endpoint record doesn't match. |
| 74 | */ |
| 75 | return ((u16 __force)local->srx.transport.sin.sin_port - |
| 76 | (u16 __force)srx->transport.sin.sin_port) ?: |
| 77 | memcmp(&local->srx.transport.sin.sin_addr, |
| 78 | &srx->transport.sin.sin_addr, |
| 79 | sizeof(struct in_addr)); |
| 80 | #ifdef CONFIG_AF_RXRPC_IPV6 |
| 81 | case AF_INET6: |
| 82 | /* If the choice of UDP6 port is left up to the transport, then |
| 83 | * the endpoint record doesn't match. |
| 84 | */ |
| 85 | return ((u16 __force)local->srx.transport.sin6.sin6_port - |
| 86 | (u16 __force)srx->transport.sin6.sin6_port) ?: |
| 87 | memcmp(&local->srx.transport.sin6.sin6_addr, |
| 88 | &srx->transport.sin6.sin6_addr, |
| 89 | sizeof(struct in6_addr)); |
| 90 | #endif |
| 91 | default: |
| 92 | BUG(); |
| 93 | } |
| 94 | } |
| 95 | |
| 96 | static void rxrpc_client_conn_reap_timeout(struct timer_list *timer) |
| 97 | { |
| 98 | struct rxrpc_local *local = |
| 99 | container_of(timer, struct rxrpc_local, client_conn_reap_timer); |
| 100 | |
| 101 | if (!local->kill_all_client_conns && |
| 102 | test_and_set_bit(RXRPC_CLIENT_CONN_REAP_TIMER, &local->client_conn_flags)) |
| 103 | rxrpc_wake_up_io_thread(local); |
| 104 | } |
| 105 | |
| 106 | /* |
| 107 | * Allocate a new local endpoint. |
| 108 | */ |
| 109 | static struct rxrpc_local *rxrpc_alloc_local(struct net *net, |
| 110 | const struct sockaddr_rxrpc *srx) |
| 111 | { |
| 112 | struct rxrpc_local *local; |
| 113 | u32 tmp; |
| 114 | |
| 115 | local = kzalloc(sizeof(struct rxrpc_local), GFP_KERNEL); |
| 116 | if (local) { |
| 117 | refcount_set(&local->ref, 1); |
| 118 | atomic_set(&local->active_users, 1); |
| 119 | local->net = net; |
| 120 | local->rxnet = rxrpc_net(net); |
| 121 | INIT_HLIST_NODE(&local->link); |
| 122 | init_completion(&local->io_thread_ready); |
| 123 | #ifdef CONFIG_AF_RXRPC_INJECT_RX_DELAY |
| 124 | skb_queue_head_init(&local->rx_delay_queue); |
| 125 | #endif |
| 126 | skb_queue_head_init(&local->rx_queue); |
| 127 | INIT_LIST_HEAD(&local->conn_attend_q); |
| 128 | INIT_LIST_HEAD(&local->call_attend_q); |
| 129 | |
| 130 | local->client_bundles = RB_ROOT; |
| 131 | spin_lock_init(&local->client_bundles_lock); |
| 132 | local->kill_all_client_conns = false; |
| 133 | INIT_LIST_HEAD(&local->idle_client_conns); |
| 134 | timer_setup(&local->client_conn_reap_timer, |
| 135 | rxrpc_client_conn_reap_timeout, 0); |
| 136 | |
| 137 | spin_lock_init(&local->lock); |
| 138 | rwlock_init(&local->services_lock); |
| 139 | local->debug_id = atomic_inc_return(&rxrpc_debug_id); |
| 140 | memcpy(&local->srx, srx, sizeof(*srx)); |
| 141 | local->srx.srx_service = 0; |
| 142 | idr_init(&local->conn_ids); |
| 143 | get_random_bytes(&tmp, sizeof(tmp)); |
| 144 | tmp &= 0x3fffffff; |
| 145 | if (tmp == 0) |
| 146 | tmp = 1; |
| 147 | idr_set_cursor(&local->conn_ids, tmp); |
| 148 | INIT_LIST_HEAD(&local->new_client_calls); |
| 149 | spin_lock_init(&local->client_call_lock); |
| 150 | |
| 151 | trace_rxrpc_local(local->debug_id, rxrpc_local_new, 1, 1); |
| 152 | } |
| 153 | |
| 154 | _leave(" = %p", local); |
| 155 | return local; |
| 156 | } |
| 157 | |
| 158 | /* |
| 159 | * create the local socket |
| 160 | * - must be called with rxrpc_local_mutex locked |
| 161 | */ |
| 162 | static int rxrpc_open_socket(struct rxrpc_local *local, struct net *net) |
| 163 | { |
| 164 | struct udp_tunnel_sock_cfg tuncfg = {NULL}; |
| 165 | struct sockaddr_rxrpc *srx = &local->srx; |
| 166 | struct udp_port_cfg udp_conf = {0}; |
| 167 | struct task_struct *io_thread; |
| 168 | struct sock *usk; |
| 169 | int ret; |
| 170 | |
| 171 | _enter("%p{%d,%d}", |
| 172 | local, srx->transport_type, srx->transport.family); |
| 173 | |
| 174 | udp_conf.family = srx->transport.family; |
| 175 | udp_conf.use_udp_checksums = true; |
| 176 | if (udp_conf.family == AF_INET) { |
| 177 | udp_conf.local_ip = srx->transport.sin.sin_addr; |
| 178 | udp_conf.local_udp_port = srx->transport.sin.sin_port; |
| 179 | #if IS_ENABLED(CONFIG_AF_RXRPC_IPV6) |
| 180 | } else { |
| 181 | udp_conf.local_ip6 = srx->transport.sin6.sin6_addr; |
| 182 | udp_conf.local_udp_port = srx->transport.sin6.sin6_port; |
| 183 | udp_conf.use_udp6_tx_checksums = true; |
| 184 | udp_conf.use_udp6_rx_checksums = true; |
| 185 | #endif |
| 186 | } |
| 187 | ret = udp_sock_create(net, &udp_conf, &local->socket); |
| 188 | if (ret < 0) { |
| 189 | _leave(" = %d [socket]", ret); |
| 190 | return ret; |
| 191 | } |
| 192 | |
| 193 | tuncfg.encap_type = UDP_ENCAP_RXRPC; |
| 194 | tuncfg.encap_rcv = rxrpc_encap_rcv; |
| 195 | tuncfg.encap_err_rcv = rxrpc_encap_err_rcv; |
| 196 | tuncfg.sk_user_data = local; |
| 197 | setup_udp_tunnel_sock(net, local->socket, &tuncfg); |
| 198 | |
| 199 | /* set the socket up */ |
| 200 | usk = local->socket->sk; |
| 201 | usk->sk_error_report = rxrpc_error_report; |
| 202 | |
| 203 | switch (srx->transport.family) { |
| 204 | case AF_INET6: |
| 205 | /* we want to receive ICMPv6 errors */ |
| 206 | ip6_sock_set_recverr(usk); |
| 207 | |
| 208 | /* Fall through and set IPv4 options too otherwise we don't get |
| 209 | * errors from IPv4 packets sent through the IPv6 socket. |
| 210 | */ |
| 211 | fallthrough; |
| 212 | case AF_INET: |
| 213 | /* we want to receive ICMP errors */ |
| 214 | ip_sock_set_recverr(usk); |
| 215 | |
| 216 | /* we want to set the don't fragment bit */ |
| 217 | rxrpc_local_dont_fragment(local, true); |
| 218 | break; |
| 219 | |
| 220 | default: |
| 221 | BUG(); |
| 222 | } |
| 223 | |
| 224 | io_thread = kthread_run(rxrpc_io_thread, local, |
| 225 | "krxrpcio/%u", ntohs(udp_conf.local_udp_port)); |
| 226 | if (IS_ERR(io_thread)) { |
| 227 | ret = PTR_ERR(io_thread); |
| 228 | goto error_sock; |
| 229 | } |
| 230 | |
| 231 | wait_for_completion(&local->io_thread_ready); |
| 232 | WRITE_ONCE(local->io_thread, io_thread); |
| 233 | _leave(" = 0"); |
| 234 | return 0; |
| 235 | |
| 236 | error_sock: |
| 237 | kernel_sock_shutdown(local->socket, SHUT_RDWR); |
| 238 | local->socket->sk->sk_user_data = NULL; |
| 239 | sock_release(local->socket); |
| 240 | local->socket = NULL; |
| 241 | return ret; |
| 242 | } |
| 243 | |
| 244 | /* |
| 245 | * Look up or create a new local endpoint using the specified local address. |
| 246 | */ |
| 247 | struct rxrpc_local *rxrpc_lookup_local(struct net *net, |
| 248 | const struct sockaddr_rxrpc *srx) |
| 249 | { |
| 250 | struct rxrpc_local *local; |
| 251 | struct rxrpc_net *rxnet = rxrpc_net(net); |
| 252 | struct hlist_node *cursor; |
| 253 | long diff; |
| 254 | int ret; |
| 255 | |
| 256 | _enter("{%d,%d,%pISp}", |
| 257 | srx->transport_type, srx->transport.family, &srx->transport); |
| 258 | |
| 259 | mutex_lock(&rxnet->local_mutex); |
| 260 | |
| 261 | hlist_for_each(cursor, &rxnet->local_endpoints) { |
| 262 | local = hlist_entry(cursor, struct rxrpc_local, link); |
| 263 | |
| 264 | diff = rxrpc_local_cmp_key(local, srx); |
| 265 | if (diff != 0) |
| 266 | continue; |
| 267 | |
| 268 | /* Services aren't allowed to share transport sockets, so |
| 269 | * reject that here. It is possible that the object is dying - |
| 270 | * but it may also still have the local transport address that |
| 271 | * we want bound. |
| 272 | */ |
| 273 | if (srx->srx_service) { |
| 274 | local = NULL; |
| 275 | goto addr_in_use; |
| 276 | } |
| 277 | |
| 278 | /* Found a match. We want to replace a dying object. |
| 279 | * Attempting to bind the transport socket may still fail if |
| 280 | * we're attempting to use a local address that the dying |
| 281 | * object is still using. |
| 282 | */ |
| 283 | if (!rxrpc_use_local(local, rxrpc_local_use_lookup)) |
| 284 | break; |
| 285 | |
| 286 | goto found; |
| 287 | } |
| 288 | |
| 289 | local = rxrpc_alloc_local(net, srx); |
| 290 | if (!local) |
| 291 | goto nomem; |
| 292 | |
| 293 | ret = rxrpc_open_socket(local, net); |
| 294 | if (ret < 0) |
| 295 | goto sock_error; |
| 296 | |
| 297 | if (cursor) { |
| 298 | hlist_replace_rcu(cursor, &local->link); |
| 299 | cursor->pprev = NULL; |
| 300 | } else { |
| 301 | hlist_add_head_rcu(&local->link, &rxnet->local_endpoints); |
| 302 | } |
| 303 | |
| 304 | found: |
| 305 | mutex_unlock(&rxnet->local_mutex); |
| 306 | _leave(" = %p", local); |
| 307 | return local; |
| 308 | |
| 309 | nomem: |
| 310 | ret = -ENOMEM; |
| 311 | sock_error: |
| 312 | mutex_unlock(&rxnet->local_mutex); |
| 313 | if (local) |
| 314 | call_rcu(&local->rcu, rxrpc_local_rcu); |
| 315 | _leave(" = %d", ret); |
| 316 | return ERR_PTR(ret); |
| 317 | |
| 318 | addr_in_use: |
| 319 | mutex_unlock(&rxnet->local_mutex); |
| 320 | _leave(" = -EADDRINUSE"); |
| 321 | return ERR_PTR(-EADDRINUSE); |
| 322 | } |
| 323 | |
| 324 | /* |
| 325 | * Get a ref on a local endpoint. |
| 326 | */ |
| 327 | struct rxrpc_local *rxrpc_get_local(struct rxrpc_local *local, |
| 328 | enum rxrpc_local_trace why) |
| 329 | { |
| 330 | int r, u; |
| 331 | |
| 332 | u = atomic_read(&local->active_users); |
| 333 | __refcount_inc(&local->ref, &r); |
| 334 | trace_rxrpc_local(local->debug_id, why, r + 1, u); |
| 335 | return local; |
| 336 | } |
| 337 | |
| 338 | /* |
| 339 | * Get a ref on a local endpoint unless its usage has already reached 0. |
| 340 | */ |
| 341 | struct rxrpc_local *rxrpc_get_local_maybe(struct rxrpc_local *local, |
| 342 | enum rxrpc_local_trace why) |
| 343 | { |
| 344 | int r, u; |
| 345 | |
| 346 | if (local && __refcount_inc_not_zero(&local->ref, &r)) { |
| 347 | u = atomic_read(&local->active_users); |
| 348 | trace_rxrpc_local(local->debug_id, why, r + 1, u); |
| 349 | return local; |
| 350 | } |
| 351 | |
| 352 | return NULL; |
| 353 | } |
| 354 | |
| 355 | /* |
| 356 | * Drop a ref on a local endpoint. |
| 357 | */ |
| 358 | void rxrpc_put_local(struct rxrpc_local *local, enum rxrpc_local_trace why) |
| 359 | { |
| 360 | unsigned int debug_id; |
| 361 | bool dead; |
| 362 | int r, u; |
| 363 | |
| 364 | if (local) { |
| 365 | debug_id = local->debug_id; |
| 366 | |
| 367 | u = atomic_read(&local->active_users); |
| 368 | dead = __refcount_dec_and_test(&local->ref, &r); |
| 369 | trace_rxrpc_local(debug_id, why, r, u); |
| 370 | |
| 371 | if (dead) |
| 372 | call_rcu(&local->rcu, rxrpc_local_rcu); |
| 373 | } |
| 374 | } |
| 375 | |
| 376 | /* |
| 377 | * Start using a local endpoint. |
| 378 | */ |
| 379 | struct rxrpc_local *rxrpc_use_local(struct rxrpc_local *local, |
| 380 | enum rxrpc_local_trace why) |
| 381 | { |
| 382 | local = rxrpc_get_local_maybe(local, rxrpc_local_get_for_use); |
| 383 | if (!local) |
| 384 | return NULL; |
| 385 | |
| 386 | if (!__rxrpc_use_local(local, why)) { |
| 387 | rxrpc_put_local(local, rxrpc_local_put_for_use); |
| 388 | return NULL; |
| 389 | } |
| 390 | |
| 391 | return local; |
| 392 | } |
| 393 | |
| 394 | /* |
| 395 | * Cease using a local endpoint. Once the number of active users reaches 0, we |
| 396 | * start the closure of the transport in the I/O thread.. |
| 397 | */ |
| 398 | void rxrpc_unuse_local(struct rxrpc_local *local, enum rxrpc_local_trace why) |
| 399 | { |
| 400 | unsigned int debug_id; |
| 401 | int r, u; |
| 402 | |
| 403 | if (local) { |
| 404 | debug_id = local->debug_id; |
| 405 | r = refcount_read(&local->ref); |
| 406 | u = atomic_dec_return(&local->active_users); |
| 407 | trace_rxrpc_local(debug_id, why, r, u); |
| 408 | if (u == 0) |
| 409 | kthread_stop(local->io_thread); |
| 410 | } |
| 411 | } |
| 412 | |
| 413 | /* |
| 414 | * Destroy a local endpoint's socket and then hand the record to RCU to dispose |
| 415 | * of. |
| 416 | * |
| 417 | * Closing the socket cannot be done from bottom half context or RCU callback |
| 418 | * context because it might sleep. |
| 419 | */ |
| 420 | void rxrpc_destroy_local(struct rxrpc_local *local) |
| 421 | { |
| 422 | struct socket *socket = local->socket; |
| 423 | struct rxrpc_net *rxnet = local->rxnet; |
| 424 | |
| 425 | _enter("%d", local->debug_id); |
| 426 | |
| 427 | local->dead = true; |
| 428 | |
| 429 | mutex_lock(&rxnet->local_mutex); |
| 430 | hlist_del_init_rcu(&local->link); |
| 431 | mutex_unlock(&rxnet->local_mutex); |
| 432 | |
| 433 | rxrpc_clean_up_local_conns(local); |
| 434 | rxrpc_service_connection_reaper(&rxnet->service_conn_reaper); |
| 435 | ASSERT(!local->service); |
| 436 | |
| 437 | if (socket) { |
| 438 | local->socket = NULL; |
| 439 | kernel_sock_shutdown(socket, SHUT_RDWR); |
| 440 | socket->sk->sk_user_data = NULL; |
| 441 | sock_release(socket); |
| 442 | } |
| 443 | |
| 444 | /* At this point, there should be no more packets coming in to the |
| 445 | * local endpoint. |
| 446 | */ |
| 447 | #ifdef CONFIG_AF_RXRPC_INJECT_RX_DELAY |
| 448 | rxrpc_purge_queue(&local->rx_delay_queue); |
| 449 | #endif |
| 450 | rxrpc_purge_queue(&local->rx_queue); |
| 451 | rxrpc_purge_client_connections(local); |
| 452 | page_frag_cache_drain(&local->tx_alloc); |
| 453 | } |
| 454 | |
| 455 | /* |
| 456 | * Destroy a local endpoint after the RCU grace period expires. |
| 457 | */ |
| 458 | static void rxrpc_local_rcu(struct rcu_head *rcu) |
| 459 | { |
| 460 | struct rxrpc_local *local = container_of(rcu, struct rxrpc_local, rcu); |
| 461 | |
| 462 | rxrpc_see_local(local, rxrpc_local_free); |
| 463 | kfree(local); |
| 464 | } |
| 465 | |
| 466 | /* |
| 467 | * Verify the local endpoint list is empty by this point. |
| 468 | */ |
| 469 | void rxrpc_destroy_all_locals(struct rxrpc_net *rxnet) |
| 470 | { |
| 471 | struct rxrpc_local *local; |
| 472 | |
| 473 | _enter(""); |
| 474 | |
| 475 | flush_workqueue(rxrpc_workqueue); |
| 476 | |
| 477 | if (!hlist_empty(&rxnet->local_endpoints)) { |
| 478 | mutex_lock(&rxnet->local_mutex); |
| 479 | hlist_for_each_entry(local, &rxnet->local_endpoints, link) { |
| 480 | pr_err("AF_RXRPC: Leaked local %p {%d}\n", |
| 481 | local, refcount_read(&local->ref)); |
| 482 | } |
| 483 | mutex_unlock(&rxnet->local_mutex); |
| 484 | BUG(); |
| 485 | } |
| 486 | } |