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 ([email protected]) 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 spin_lock_init(&peer->rtt_input_lock); 239 seqcount_init(&peer->mtu_lock); 240 peer->debug_id = atomic_inc_return(&rxrpc_debug_id); 241 242 rxrpc_peer_init_rtt(peer); 243 244 peer->cong_ssthresh = RXRPC_TX_MAX_WINDOW; 245 trace_rxrpc_peer(peer->debug_id, 1, why); 246 } 247 248 _leave(" = %p", peer); 249 return peer; 250 } 251 252 /* 253 * Initialise peer record. 254 */ 255 static void rxrpc_init_peer(struct rxrpc_local *local, struct rxrpc_peer *peer, 256 unsigned long hash_key) 257 { 258 peer->hash_key = hash_key; 259 260 261 switch (peer->srx.transport.family) { 262 case AF_INET: 263 peer->hdrsize = sizeof(struct iphdr); 264 break; 265 #ifdef CONFIG_AF_RXRPC_IPV6 266 case AF_INET6: 267 peer->hdrsize = sizeof(struct ipv6hdr); 268 break; 269 #endif 270 default: 271 BUG(); 272 } 273 274 switch (peer->srx.transport_type) { 275 case SOCK_DGRAM: 276 peer->hdrsize += sizeof(struct udphdr); 277 break; 278 default: 279 BUG(); 280 } 281 282 peer->hdrsize += sizeof(struct rxrpc_wire_header); 283 peer->max_data = peer->if_mtu - peer->hdrsize; 284 285 rxrpc_assess_MTU_size(local, peer); 286 287 peer->rtt_last_req = ktime_get_real(); 288 } 289 290 /* 291 * Set up a new peer. 292 */ 293 static struct rxrpc_peer *rxrpc_create_peer(struct rxrpc_local *local, 294 struct sockaddr_rxrpc *srx, 295 unsigned long hash_key, 296 gfp_t gfp) 297 { 298 struct rxrpc_peer *peer; 299 300 _enter(""); 301 302 peer = rxrpc_alloc_peer(local, gfp, rxrpc_peer_new_client); 303 if (peer) { 304 memcpy(&peer->srx, srx, sizeof(*srx)); 305 rxrpc_init_peer(local, peer, hash_key); 306 } 307 308 _leave(" = %p", peer); 309 return peer; 310 } 311 312 static void rxrpc_free_peer(struct rxrpc_peer *peer) 313 { 314 trace_rxrpc_peer(peer->debug_id, 0, rxrpc_peer_free); 315 rxrpc_put_local(peer->local, rxrpc_local_put_peer); 316 kfree_rcu(peer, rcu); 317 } 318 319 /* 320 * Set up a new incoming peer. There shouldn't be any other matching peers 321 * since we've already done a search in the list from the non-reentrant context 322 * (the data_ready handler) that is the only place we can add new peers. 323 */ 324 void rxrpc_new_incoming_peer(struct rxrpc_local *local, struct rxrpc_peer *peer) 325 { 326 struct rxrpc_net *rxnet = local->rxnet; 327 unsigned long hash_key; 328 329 hash_key = rxrpc_peer_hash_key(local, &peer->srx); 330 rxrpc_init_peer(local, peer, hash_key); 331 332 spin_lock(&rxnet->peer_hash_lock); 333 hash_add_rcu(rxnet->peer_hash, &peer->hash_link, hash_key); 334 list_add_tail(&peer->keepalive_link, &rxnet->peer_keepalive_new); 335 spin_unlock(&rxnet->peer_hash_lock); 336 } 337 338 /* 339 * obtain a remote transport endpoint for the specified address 340 */ 341 struct rxrpc_peer *rxrpc_lookup_peer(struct rxrpc_local *local, 342 struct sockaddr_rxrpc *srx, gfp_t gfp) 343 { 344 struct rxrpc_peer *peer, *candidate; 345 struct rxrpc_net *rxnet = local->rxnet; 346 unsigned long hash_key = rxrpc_peer_hash_key(local, srx); 347 348 _enter("{%pISp}", &srx->transport); 349 350 /* search the peer list first */ 351 rcu_read_lock(); 352 peer = __rxrpc_lookup_peer_rcu(local, srx, hash_key); 353 if (peer && !rxrpc_get_peer_maybe(peer, rxrpc_peer_get_lookup_client)) 354 peer = NULL; 355 rcu_read_unlock(); 356 357 if (!peer) { 358 /* The peer is not yet present in hash - create a candidate 359 * for a new record and then redo the search. 360 */ 361 candidate = rxrpc_create_peer(local, srx, hash_key, gfp); 362 if (!candidate) { 363 _leave(" = NULL [nomem]"); 364 return NULL; 365 } 366 367 spin_lock(&rxnet->peer_hash_lock); 368 369 /* Need to check that we aren't racing with someone else */ 370 peer = __rxrpc_lookup_peer_rcu(local, srx, hash_key); 371 if (peer && !rxrpc_get_peer_maybe(peer, rxrpc_peer_get_lookup_client)) 372 peer = NULL; 373 if (!peer) { 374 hash_add_rcu(rxnet->peer_hash, 375 &candidate->hash_link, hash_key); 376 list_add_tail(&candidate->keepalive_link, 377 &rxnet->peer_keepalive_new); 378 } 379 380 spin_unlock(&rxnet->peer_hash_lock); 381 382 if (peer) 383 rxrpc_free_peer(candidate); 384 else 385 peer = candidate; 386 } 387 388 _leave(" = %p {u=%d}", peer, refcount_read(&peer->ref)); 389 return peer; 390 } 391 392 /* 393 * Get a ref on a peer record. 394 */ 395 struct rxrpc_peer *rxrpc_get_peer(struct rxrpc_peer *peer, enum rxrpc_peer_trace why) 396 { 397 int r; 398 399 __refcount_inc(&peer->ref, &r); 400 trace_rxrpc_peer(peer->debug_id, r + 1, why); 401 return peer; 402 } 403 404 /* 405 * Get a ref on a peer record unless its usage has already reached 0. 406 */ 407 struct rxrpc_peer *rxrpc_get_peer_maybe(struct rxrpc_peer *peer, 408 enum rxrpc_peer_trace why) 409 { 410 int r; 411 412 if (peer) { 413 if (__refcount_inc_not_zero(&peer->ref, &r)) 414 trace_rxrpc_peer(peer->debug_id, r + 1, why); 415 else 416 peer = NULL; 417 } 418 return peer; 419 } 420 421 /* 422 * Discard a peer record. 423 */ 424 static void __rxrpc_put_peer(struct rxrpc_peer *peer) 425 { 426 struct rxrpc_net *rxnet = peer->local->rxnet; 427 428 ASSERT(hlist_empty(&peer->error_targets)); 429 430 spin_lock(&rxnet->peer_hash_lock); 431 hash_del_rcu(&peer->hash_link); 432 list_del_init(&peer->keepalive_link); 433 spin_unlock(&rxnet->peer_hash_lock); 434 435 rxrpc_free_peer(peer); 436 } 437 438 /* 439 * Drop a ref on a peer record. 440 */ 441 void rxrpc_put_peer(struct rxrpc_peer *peer, enum rxrpc_peer_trace why) 442 { 443 unsigned int debug_id; 444 bool dead; 445 int r; 446 447 if (peer) { 448 debug_id = peer->debug_id; 449 dead = __refcount_dec_and_test(&peer->ref, &r); 450 trace_rxrpc_peer(debug_id, r - 1, why); 451 if (dead) 452 __rxrpc_put_peer(peer); 453 } 454 } 455 456 /* 457 * Make sure all peer records have been discarded. 458 */ 459 void rxrpc_destroy_all_peers(struct rxrpc_net *rxnet) 460 { 461 struct rxrpc_peer *peer; 462 int i; 463 464 for (i = 0; i < HASH_SIZE(rxnet->peer_hash); i++) { 465 if (hlist_empty(&rxnet->peer_hash[i])) 466 continue; 467 468 hlist_for_each_entry(peer, &rxnet->peer_hash[i], hash_link) { 469 pr_err("Leaked peer %u {%u} %pISp\n", 470 peer->debug_id, 471 refcount_read(&peer->ref), 472 &peer->srx.transport); 473 } 474 } 475 } 476 477 /** 478 * rxrpc_kernel_get_call_peer - Get the peer address of a call 479 * @sock: The socket on which the call is in progress. 480 * @call: The call to query 481 * 482 * Get a record for the remote peer in a call. 483 */ 484 struct rxrpc_peer *rxrpc_kernel_get_call_peer(struct socket *sock, struct rxrpc_call *call) 485 { 486 return call->peer; 487 } 488 EXPORT_SYMBOL(rxrpc_kernel_get_call_peer); 489 490 /** 491 * rxrpc_kernel_get_srtt - Get a call's peer smoothed RTT 492 * @peer: The peer to query 493 * 494 * Get the call's peer smoothed RTT in uS or UINT_MAX if we have no samples. 495 */ 496 unsigned int rxrpc_kernel_get_srtt(const struct rxrpc_peer *peer) 497 { 498 return peer->rtt_count > 0 ? peer->srtt_us >> 3 : UINT_MAX; 499 } 500 EXPORT_SYMBOL(rxrpc_kernel_get_srtt); 501 502 /** 503 * rxrpc_kernel_remote_srx - Get the address of a peer 504 * @peer: The peer to query 505 * 506 * Get a pointer to the address from a peer record. The caller is responsible 507 * for making sure that the address is not deallocated. 508 */ 509 const struct sockaddr_rxrpc *rxrpc_kernel_remote_srx(const struct rxrpc_peer *peer) 510 { 511 return peer ? &peer->srx : &rxrpc_null_addr; 512 } 513 EXPORT_SYMBOL(rxrpc_kernel_remote_srx); 514 515 /** 516 * rxrpc_kernel_remote_addr - Get the peer transport address of a call 517 * @peer: The peer to query 518 * 519 * Get a pointer to the transport address from a peer record. The caller is 520 * responsible for making sure that the address is not deallocated. 521 */ 522 const struct sockaddr *rxrpc_kernel_remote_addr(const struct rxrpc_peer *peer) 523 { 524 return (const struct sockaddr *) 525 (peer ? &peer->srx.transport : &rxrpc_null_addr.transport); 526 } 527 EXPORT_SYMBOL(rxrpc_kernel_remote_addr); 528