1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* RxRPC remote transport endpoint record management 3 * 4 * Copyright (C) 2007, 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/udp.h> 14 #include <linux/in.h> 15 #include <linux/in6.h> 16 #include <linux/slab.h> 17 #include <linux/hashtable.h> 18 #include <net/sock.h> 19 #include <net/af_rxrpc.h> 20 #include <net/ip.h> 21 #include <net/route.h> 22 #include <net/ip6_route.h> 23 #include "ar-internal.h" 24 25 static const struct sockaddr_rxrpc rxrpc_null_addr; 26 27 /* 28 * Hash a peer key. 29 */ 30 static unsigned long rxrpc_peer_hash_key(struct rxrpc_local *local, 31 const struct sockaddr_rxrpc *srx) 32 { 33 const u16 *p; 34 unsigned int i, size; 35 unsigned long hash_key; 36 37 _enter(""); 38 39 hash_key = (unsigned long)local / __alignof__(*local); 40 hash_key += srx->transport_type; 41 hash_key += srx->transport_len; 42 hash_key += srx->transport.family; 43 44 switch (srx->transport.family) { 45 case AF_INET: 46 hash_key += (u16 __force)srx->transport.sin.sin_port; 47 size = sizeof(srx->transport.sin.sin_addr); 48 p = (u16 *)&srx->transport.sin.sin_addr; 49 break; 50 #ifdef CONFIG_AF_RXRPC_IPV6 51 case AF_INET6: 52 hash_key += (u16 __force)srx->transport.sin.sin_port; 53 size = sizeof(srx->transport.sin6.sin6_addr); 54 p = (u16 *)&srx->transport.sin6.sin6_addr; 55 break; 56 #endif 57 default: 58 WARN(1, "AF_RXRPC: Unsupported transport address family\n"); 59 return 0; 60 } 61 62 /* Step through the peer address in 16-bit portions for speed */ 63 for (i = 0; i < size; i += sizeof(*p), p++) 64 hash_key += *p; 65 66 _leave(" 0x%lx", hash_key); 67 return hash_key; 68 } 69 70 /* 71 * Compare a peer to a key. Return -ve, 0 or +ve to indicate less than, same 72 * or greater than. 73 * 74 * Unfortunately, the primitives in linux/hashtable.h don't allow for sorted 75 * buckets and mid-bucket insertion, so we don't make full use of this 76 * information at this point. 77 */ 78 static long rxrpc_peer_cmp_key(const struct rxrpc_peer *peer, 79 struct rxrpc_local *local, 80 const struct sockaddr_rxrpc *srx, 81 unsigned long hash_key) 82 { 83 long diff; 84 85 diff = ((peer->hash_key - hash_key) ?: 86 ((unsigned long)peer->local - (unsigned long)local) ?: 87 (peer->srx.transport_type - srx->transport_type) ?: 88 (peer->srx.transport_len - srx->transport_len) ?: 89 (peer->srx.transport.family - srx->transport.family)); 90 if (diff != 0) 91 return diff; 92 93 switch (srx->transport.family) { 94 case AF_INET: 95 return ((u16 __force)peer->srx.transport.sin.sin_port - 96 (u16 __force)srx->transport.sin.sin_port) ?: 97 memcmp(&peer->srx.transport.sin.sin_addr, 98 &srx->transport.sin.sin_addr, 99 sizeof(struct in_addr)); 100 #ifdef CONFIG_AF_RXRPC_IPV6 101 case AF_INET6: 102 return ((u16 __force)peer->srx.transport.sin6.sin6_port - 103 (u16 __force)srx->transport.sin6.sin6_port) ?: 104 memcmp(&peer->srx.transport.sin6.sin6_addr, 105 &srx->transport.sin6.sin6_addr, 106 sizeof(struct in6_addr)); 107 #endif 108 default: 109 BUG(); 110 } 111 } 112 113 /* 114 * Look up a remote transport endpoint for the specified address using RCU. 115 */ 116 static struct rxrpc_peer *__rxrpc_lookup_peer_rcu( 117 struct rxrpc_local *local, 118 const struct sockaddr_rxrpc *srx, 119 unsigned long hash_key) 120 { 121 struct rxrpc_peer *peer; 122 struct rxrpc_net *rxnet = local->rxnet; 123 124 hash_for_each_possible_rcu(rxnet->peer_hash, peer, hash_link, hash_key) { 125 if (rxrpc_peer_cmp_key(peer, local, srx, hash_key) == 0 && 126 refcount_read(&peer->ref) > 0) 127 return peer; 128 } 129 130 return NULL; 131 } 132 133 /* 134 * Look up a remote transport endpoint for the specified address using RCU. 135 */ 136 struct rxrpc_peer *rxrpc_lookup_peer_rcu(struct rxrpc_local *local, 137 const struct sockaddr_rxrpc *srx) 138 { 139 struct rxrpc_peer *peer; 140 unsigned long hash_key = rxrpc_peer_hash_key(local, srx); 141 142 peer = __rxrpc_lookup_peer_rcu(local, srx, hash_key); 143 if (peer) 144 _leave(" = %p {u=%d}", peer, refcount_read(&peer->ref)); 145 return peer; 146 } 147 148 /* 149 * assess the MTU size for the network interface through which this peer is 150 * reached 151 */ 152 static void rxrpc_assess_MTU_size(struct rxrpc_local *local, 153 struct rxrpc_peer *peer) 154 { 155 struct net *net = local->net; 156 struct dst_entry *dst; 157 struct rtable *rt; 158 struct flowi fl; 159 struct flowi4 *fl4 = &fl.u.ip4; 160 #ifdef CONFIG_AF_RXRPC_IPV6 161 struct flowi6 *fl6 = &fl.u.ip6; 162 #endif 163 164 peer->if_mtu = 1500; 165 if (peer->max_data < peer->if_mtu - peer->hdrsize) { 166 trace_rxrpc_pmtud_reduce(peer, 0, peer->if_mtu - peer->hdrsize, 167 rxrpc_pmtud_reduce_route); 168 peer->max_data = peer->if_mtu - peer->hdrsize; 169 } 170 171 memset(&fl, 0, sizeof(fl)); 172 switch (peer->srx.transport.family) { 173 case AF_INET: 174 rt = ip_route_output_ports( 175 net, fl4, NULL, 176 peer->srx.transport.sin.sin_addr.s_addr, 0, 177 htons(7000), htons(7001), IPPROTO_UDP, 0, 0); 178 if (IS_ERR(rt)) { 179 _leave(" [route err %ld]", PTR_ERR(rt)); 180 return; 181 } 182 dst = &rt->dst; 183 break; 184 185 #ifdef CONFIG_AF_RXRPC_IPV6 186 case AF_INET6: 187 fl6->flowi6_iif = LOOPBACK_IFINDEX; 188 fl6->flowi6_scope = RT_SCOPE_UNIVERSE; 189 fl6->flowi6_proto = IPPROTO_UDP; 190 memcpy(&fl6->daddr, &peer->srx.transport.sin6.sin6_addr, 191 sizeof(struct in6_addr)); 192 fl6->fl6_dport = htons(7001); 193 fl6->fl6_sport = htons(7000); 194 dst = ip6_route_output(net, NULL, fl6); 195 if (dst->error) { 196 _leave(" [route err %d]", dst->error); 197 return; 198 } 199 break; 200 #endif 201 202 default: 203 BUG(); 204 } 205 206 peer->if_mtu = dst_mtu(dst); 207 peer->hdrsize += dst->header_len + dst->trailer_len; 208 peer->tx_seg_max = dst->dev->gso_max_segs; 209 dst_release(dst); 210 211 peer->max_data = umin(RXRPC_JUMBO(1), peer->if_mtu - peer->hdrsize); 212 peer->pmtud_good = 500; 213 peer->pmtud_bad = peer->if_mtu - peer->hdrsize + 1; 214 peer->pmtud_trial = umin(peer->max_data, peer->pmtud_bad - 1); 215 peer->pmtud_pending = true; 216 217 _leave(" [if_mtu %u]", peer->if_mtu); 218 } 219 220 /* 221 * Allocate a peer. 222 */ 223 struct rxrpc_peer *rxrpc_alloc_peer(struct rxrpc_local *local, gfp_t gfp, 224 enum rxrpc_peer_trace why) 225 { 226 struct rxrpc_peer *peer; 227 228 _enter(""); 229 230 peer = kzalloc(sizeof(struct rxrpc_peer), gfp); 231 if (peer) { 232 refcount_set(&peer->ref, 1); 233 peer->local = rxrpc_get_local(local, rxrpc_local_get_peer); 234 INIT_HLIST_HEAD(&peer->error_targets); 235 peer->service_conns = RB_ROOT; 236 seqlock_init(&peer->service_conn_lock); 237 spin_lock_init(&peer->lock); 238 peer->debug_id = atomic_inc_return(&rxrpc_debug_id); 239 peer->recent_srtt_us = UINT_MAX; 240 peer->cong_ssthresh = RXRPC_TX_MAX_WINDOW; 241 trace_rxrpc_peer(peer->debug_id, 1, why); 242 } 243 244 _leave(" = %p", peer); 245 return peer; 246 } 247 248 /* 249 * Initialise peer record. 250 */ 251 static void rxrpc_init_peer(struct rxrpc_local *local, struct rxrpc_peer *peer, 252 unsigned long hash_key) 253 { 254 peer->hash_key = hash_key; 255 256 257 switch (peer->srx.transport.family) { 258 case AF_INET: 259 peer->hdrsize = sizeof(struct iphdr); 260 break; 261 #ifdef CONFIG_AF_RXRPC_IPV6 262 case AF_INET6: 263 peer->hdrsize = sizeof(struct ipv6hdr); 264 break; 265 #endif 266 default: 267 BUG(); 268 } 269 270 switch (peer->srx.transport_type) { 271 case SOCK_DGRAM: 272 peer->hdrsize += sizeof(struct udphdr); 273 break; 274 default: 275 BUG(); 276 } 277 278 peer->hdrsize += sizeof(struct rxrpc_wire_header); 279 peer->max_data = peer->if_mtu - peer->hdrsize; 280 281 rxrpc_assess_MTU_size(local, peer); 282 } 283 284 /* 285 * Set up a new peer. 286 */ 287 static struct rxrpc_peer *rxrpc_create_peer(struct rxrpc_local *local, 288 struct sockaddr_rxrpc *srx, 289 unsigned long hash_key, 290 gfp_t gfp) 291 { 292 struct rxrpc_peer *peer; 293 294 _enter(""); 295 296 peer = rxrpc_alloc_peer(local, gfp, rxrpc_peer_new_client); 297 if (peer) { 298 memcpy(&peer->srx, srx, sizeof(*srx)); 299 rxrpc_init_peer(local, peer, hash_key); 300 } 301 302 _leave(" = %p", peer); 303 return peer; 304 } 305 306 static void rxrpc_free_peer(struct rxrpc_peer *peer) 307 { 308 trace_rxrpc_peer(peer->debug_id, 0, rxrpc_peer_free); 309 rxrpc_put_local(peer->local, rxrpc_local_put_peer); 310 kfree_rcu(peer, rcu); 311 } 312 313 /* 314 * Set up a new incoming peer. There shouldn't be any other matching peers 315 * since we've already done a search in the list from the non-reentrant context 316 * (the data_ready handler) that is the only place we can add new peers. 317 * Called with interrupts disabled. 318 */ 319 void rxrpc_new_incoming_peer(struct rxrpc_local *local, struct rxrpc_peer *peer) 320 { 321 struct rxrpc_net *rxnet = local->rxnet; 322 unsigned long hash_key; 323 324 hash_key = rxrpc_peer_hash_key(local, &peer->srx); 325 rxrpc_init_peer(local, peer, hash_key); 326 327 spin_lock(&rxnet->peer_hash_lock); 328 hash_add_rcu(rxnet->peer_hash, &peer->hash_link, hash_key); 329 list_add_tail(&peer->keepalive_link, &rxnet->peer_keepalive_new); 330 spin_unlock(&rxnet->peer_hash_lock); 331 } 332 333 /* 334 * obtain a remote transport endpoint for the specified address 335 */ 336 struct rxrpc_peer *rxrpc_lookup_peer(struct rxrpc_local *local, 337 struct sockaddr_rxrpc *srx, gfp_t gfp) 338 { 339 struct rxrpc_peer *peer, *candidate; 340 struct rxrpc_net *rxnet = local->rxnet; 341 unsigned long hash_key = rxrpc_peer_hash_key(local, srx); 342 343 _enter("{%pISp}", &srx->transport); 344 345 /* search the peer list first */ 346 rcu_read_lock(); 347 peer = __rxrpc_lookup_peer_rcu(local, srx, hash_key); 348 if (peer && !rxrpc_get_peer_maybe(peer, rxrpc_peer_get_lookup_client)) 349 peer = NULL; 350 rcu_read_unlock(); 351 352 if (!peer) { 353 /* The peer is not yet present in hash - create a candidate 354 * for a new record and then redo the search. 355 */ 356 candidate = rxrpc_create_peer(local, srx, hash_key, gfp); 357 if (!candidate) { 358 _leave(" = NULL [nomem]"); 359 return NULL; 360 } 361 362 spin_lock_bh(&rxnet->peer_hash_lock); 363 364 /* Need to check that we aren't racing with someone else */ 365 peer = __rxrpc_lookup_peer_rcu(local, srx, hash_key); 366 if (peer && !rxrpc_get_peer_maybe(peer, rxrpc_peer_get_lookup_client)) 367 peer = NULL; 368 if (!peer) { 369 hash_add_rcu(rxnet->peer_hash, 370 &candidate->hash_link, hash_key); 371 list_add_tail(&candidate->keepalive_link, 372 &rxnet->peer_keepalive_new); 373 } 374 375 spin_unlock_bh(&rxnet->peer_hash_lock); 376 377 if (peer) 378 rxrpc_free_peer(candidate); 379 else 380 peer = candidate; 381 } 382 383 _leave(" = %p {u=%d}", peer, refcount_read(&peer->ref)); 384 return peer; 385 } 386 387 /* 388 * Get a ref on a peer record. 389 */ 390 struct rxrpc_peer *rxrpc_get_peer(struct rxrpc_peer *peer, enum rxrpc_peer_trace why) 391 { 392 int r; 393 394 __refcount_inc(&peer->ref, &r); 395 trace_rxrpc_peer(peer->debug_id, r + 1, why); 396 return peer; 397 } 398 399 /* 400 * Get a ref on a peer record unless its usage has already reached 0. 401 */ 402 struct rxrpc_peer *rxrpc_get_peer_maybe(struct rxrpc_peer *peer, 403 enum rxrpc_peer_trace why) 404 { 405 int r; 406 407 if (peer) { 408 if (__refcount_inc_not_zero(&peer->ref, &r)) 409 trace_rxrpc_peer(peer->debug_id, r + 1, why); 410 else 411 peer = NULL; 412 } 413 return peer; 414 } 415 416 /* 417 * Discard a peer record. 418 */ 419 static void __rxrpc_put_peer(struct rxrpc_peer *peer) 420 { 421 struct rxrpc_net *rxnet = peer->local->rxnet; 422 423 ASSERT(hlist_empty(&peer->error_targets)); 424 425 spin_lock_bh(&rxnet->peer_hash_lock); 426 hash_del_rcu(&peer->hash_link); 427 list_del_init(&peer->keepalive_link); 428 spin_unlock_bh(&rxnet->peer_hash_lock); 429 430 rxrpc_free_peer(peer); 431 } 432 433 /* 434 * Drop a ref on a peer record. 435 */ 436 void rxrpc_put_peer(struct rxrpc_peer *peer, enum rxrpc_peer_trace why) 437 { 438 unsigned int debug_id; 439 bool dead; 440 int r; 441 442 if (peer) { 443 debug_id = peer->debug_id; 444 dead = __refcount_dec_and_test(&peer->ref, &r); 445 trace_rxrpc_peer(debug_id, r - 1, why); 446 if (dead) 447 __rxrpc_put_peer(peer); 448 } 449 } 450 451 /* 452 * Make sure all peer records have been discarded. 453 */ 454 void rxrpc_destroy_all_peers(struct rxrpc_net *rxnet) 455 { 456 struct rxrpc_peer *peer; 457 int i; 458 459 for (i = 0; i < HASH_SIZE(rxnet->peer_hash); i++) { 460 if (hlist_empty(&rxnet->peer_hash[i])) 461 continue; 462 463 hlist_for_each_entry(peer, &rxnet->peer_hash[i], hash_link) { 464 pr_err("Leaked peer %x {%u} %pISp\n", 465 peer->debug_id, 466 refcount_read(&peer->ref), 467 &peer->srx.transport); 468 } 469 } 470 } 471 472 /** 473 * rxrpc_kernel_get_call_peer - Get the peer address of a call 474 * @sock: The socket on which the call is in progress. 475 * @call: The call to query 476 * 477 * Get a record for the remote peer in a call. 478 */ 479 struct rxrpc_peer *rxrpc_kernel_get_call_peer(struct socket *sock, struct rxrpc_call *call) 480 { 481 return rxrpc_get_peer(call->peer, rxrpc_peer_get_application); 482 } 483 EXPORT_SYMBOL(rxrpc_kernel_get_call_peer); 484 485 /** 486 * rxrpc_kernel_get_srtt - Get a call's peer smoothed RTT 487 * @peer: The peer to query 488 * 489 * Get the call's peer smoothed RTT in uS or UINT_MAX if we have no samples. 490 */ 491 unsigned int rxrpc_kernel_get_srtt(const struct rxrpc_peer *peer) 492 { 493 return READ_ONCE(peer->recent_srtt_us); 494 } 495 EXPORT_SYMBOL(rxrpc_kernel_get_srtt); 496 497 /** 498 * rxrpc_kernel_remote_srx - Get the address of a peer 499 * @peer: The peer to query 500 * 501 * Get a pointer to the address from a peer record. The caller is responsible 502 * for making sure that the address is not deallocated. 503 */ 504 const struct sockaddr_rxrpc *rxrpc_kernel_remote_srx(const struct rxrpc_peer *peer) 505 { 506 return peer ? &peer->srx : &rxrpc_null_addr; 507 } 508 EXPORT_SYMBOL(rxrpc_kernel_remote_srx); 509 510 /** 511 * rxrpc_kernel_remote_addr - Get the peer transport address of a call 512 * @peer: The peer to query 513 * 514 * Get a pointer to the transport address from a peer record. The caller is 515 * responsible for making sure that the address is not deallocated. 516 */ 517 const struct sockaddr *rxrpc_kernel_remote_addr(const struct rxrpc_peer *peer) 518 { 519 return (const struct sockaddr *) 520 (peer ? &peer->srx.transport : &rxrpc_null_addr.transport); 521 } 522 EXPORT_SYMBOL(rxrpc_kernel_remote_addr); 523 524 /** 525 * rxrpc_kernel_set_peer_data - Set app-specific data on a peer. 526 * @peer: The peer to alter 527 * @app_data: The data to set 528 * 529 * Set the app-specific data on a peer. AF_RXRPC makes no effort to retain 530 * anything the data might refer to. The previous app_data is returned. 531 */ 532 unsigned long rxrpc_kernel_set_peer_data(struct rxrpc_peer *peer, unsigned long app_data) 533 { 534 return xchg(&peer->app_data, app_data); 535 } 536 EXPORT_SYMBOL(rxrpc_kernel_set_peer_data); 537 538 /** 539 * rxrpc_kernel_get_peer_data - Get app-specific data from a peer. 540 * @peer: The peer to query 541 * 542 * Retrieve the app-specific data from a peer. 543 */ 544 unsigned long rxrpc_kernel_get_peer_data(const struct rxrpc_peer *peer) 545 { 546 return peer->app_data; 547 } 548 EXPORT_SYMBOL(rxrpc_kernel_get_peer_data); 549