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 /* 26 * Hash a peer key. 27 */ 28 static unsigned long rxrpc_peer_hash_key(struct rxrpc_local *local, 29 const struct sockaddr_rxrpc *srx) 30 { 31 const u16 *p; 32 unsigned int i, size; 33 unsigned long hash_key; 34 35 _enter(""); 36 37 hash_key = (unsigned long)local / __alignof__(*local); 38 hash_key += srx->transport_type; 39 hash_key += srx->transport_len; 40 hash_key += srx->transport.family; 41 42 switch (srx->transport.family) { 43 case AF_INET: 44 hash_key += (u16 __force)srx->transport.sin.sin_port; 45 size = sizeof(srx->transport.sin.sin_addr); 46 p = (u16 *)&srx->transport.sin.sin_addr; 47 break; 48 #ifdef CONFIG_AF_RXRPC_IPV6 49 case AF_INET6: 50 hash_key += (u16 __force)srx->transport.sin.sin_port; 51 size = sizeof(srx->transport.sin6.sin6_addr); 52 p = (u16 *)&srx->transport.sin6.sin6_addr; 53 break; 54 #endif 55 default: 56 WARN(1, "AF_RXRPC: Unsupported transport address family\n"); 57 return 0; 58 } 59 60 /* Step through the peer address in 16-bit portions for speed */ 61 for (i = 0; i < size; i += sizeof(*p), p++) 62 hash_key += *p; 63 64 _leave(" 0x%lx", hash_key); 65 return hash_key; 66 } 67 68 /* 69 * Compare a peer to a key. Return -ve, 0 or +ve to indicate less than, same 70 * or greater than. 71 * 72 * Unfortunately, the primitives in linux/hashtable.h don't allow for sorted 73 * buckets and mid-bucket insertion, so we don't make full use of this 74 * information at this point. 75 */ 76 static long rxrpc_peer_cmp_key(const struct rxrpc_peer *peer, 77 struct rxrpc_local *local, 78 const struct sockaddr_rxrpc *srx, 79 unsigned long hash_key) 80 { 81 long diff; 82 83 diff = ((peer->hash_key - hash_key) ?: 84 ((unsigned long)peer->local - (unsigned long)local) ?: 85 (peer->srx.transport_type - srx->transport_type) ?: 86 (peer->srx.transport_len - srx->transport_len) ?: 87 (peer->srx.transport.family - srx->transport.family)); 88 if (diff != 0) 89 return diff; 90 91 switch (srx->transport.family) { 92 case AF_INET: 93 return ((u16 __force)peer->srx.transport.sin.sin_port - 94 (u16 __force)srx->transport.sin.sin_port) ?: 95 memcmp(&peer->srx.transport.sin.sin_addr, 96 &srx->transport.sin.sin_addr, 97 sizeof(struct in_addr)); 98 #ifdef CONFIG_AF_RXRPC_IPV6 99 case AF_INET6: 100 return ((u16 __force)peer->srx.transport.sin6.sin6_port - 101 (u16 __force)srx->transport.sin6.sin6_port) ?: 102 memcmp(&peer->srx.transport.sin6.sin6_addr, 103 &srx->transport.sin6.sin6_addr, 104 sizeof(struct in6_addr)); 105 #endif 106 default: 107 BUG(); 108 } 109 } 110 111 /* 112 * Look up a remote transport endpoint for the specified address using RCU. 113 */ 114 static struct rxrpc_peer *__rxrpc_lookup_peer_rcu( 115 struct rxrpc_local *local, 116 const struct sockaddr_rxrpc *srx, 117 unsigned long hash_key) 118 { 119 struct rxrpc_peer *peer; 120 struct rxrpc_net *rxnet = local->rxnet; 121 122 hash_for_each_possible_rcu(rxnet->peer_hash, peer, hash_link, hash_key) { 123 if (rxrpc_peer_cmp_key(peer, local, srx, hash_key) == 0 && 124 atomic_read(&peer->usage) > 0) 125 return peer; 126 } 127 128 return NULL; 129 } 130 131 /* 132 * Look up a remote transport endpoint for the specified address using RCU. 133 */ 134 struct rxrpc_peer *rxrpc_lookup_peer_rcu(struct rxrpc_local *local, 135 const struct sockaddr_rxrpc *srx) 136 { 137 struct rxrpc_peer *peer; 138 unsigned long hash_key = rxrpc_peer_hash_key(local, srx); 139 140 peer = __rxrpc_lookup_peer_rcu(local, srx, hash_key); 141 if (peer) { 142 _net("PEER %d {%pISp}", peer->debug_id, &peer->srx.transport); 143 _leave(" = %p {u=%d}", peer, atomic_read(&peer->usage)); 144 } 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_sock *rx, 153 struct rxrpc_peer *peer) 154 { 155 struct net *net = sock_net(&rx->sk); 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; 165 166 memset(&fl, 0, sizeof(fl)); 167 switch (peer->srx.transport.family) { 168 case AF_INET: 169 rt = ip_route_output_ports( 170 net, fl4, NULL, 171 peer->srx.transport.sin.sin_addr.s_addr, 0, 172 htons(7000), htons(7001), IPPROTO_UDP, 0, 0); 173 if (IS_ERR(rt)) { 174 _leave(" [route err %ld]", PTR_ERR(rt)); 175 return; 176 } 177 dst = &rt->dst; 178 break; 179 180 #ifdef CONFIG_AF_RXRPC_IPV6 181 case AF_INET6: 182 fl6->flowi6_iif = LOOPBACK_IFINDEX; 183 fl6->flowi6_scope = RT_SCOPE_UNIVERSE; 184 fl6->flowi6_proto = IPPROTO_UDP; 185 memcpy(&fl6->daddr, &peer->srx.transport.sin6.sin6_addr, 186 sizeof(struct in6_addr)); 187 fl6->fl6_dport = htons(7001); 188 fl6->fl6_sport = htons(7000); 189 dst = ip6_route_output(net, NULL, fl6); 190 if (dst->error) { 191 _leave(" [route err %d]", dst->error); 192 return; 193 } 194 break; 195 #endif 196 197 default: 198 BUG(); 199 } 200 201 peer->if_mtu = dst_mtu(dst); 202 dst_release(dst); 203 204 _leave(" [if_mtu %u]", peer->if_mtu); 205 } 206 207 /* 208 * Allocate a peer. 209 */ 210 struct rxrpc_peer *rxrpc_alloc_peer(struct rxrpc_local *local, gfp_t gfp) 211 { 212 const void *here = __builtin_return_address(0); 213 struct rxrpc_peer *peer; 214 215 _enter(""); 216 217 peer = kzalloc(sizeof(struct rxrpc_peer), gfp); 218 if (peer) { 219 atomic_set(&peer->usage, 1); 220 peer->local = rxrpc_get_local(local); 221 INIT_HLIST_HEAD(&peer->error_targets); 222 peer->service_conns = RB_ROOT; 223 seqlock_init(&peer->service_conn_lock); 224 spin_lock_init(&peer->lock); 225 spin_lock_init(&peer->rtt_input_lock); 226 peer->debug_id = atomic_inc_return(&rxrpc_debug_id); 227 228 rxrpc_peer_init_rtt(peer); 229 230 if (RXRPC_TX_SMSS > 2190) 231 peer->cong_cwnd = 2; 232 else if (RXRPC_TX_SMSS > 1095) 233 peer->cong_cwnd = 3; 234 else 235 peer->cong_cwnd = 4; 236 trace_rxrpc_peer(peer->debug_id, rxrpc_peer_new, 1, here); 237 } 238 239 _leave(" = %p", peer); 240 return peer; 241 } 242 243 /* 244 * Initialise peer record. 245 */ 246 static void rxrpc_init_peer(struct rxrpc_sock *rx, struct rxrpc_peer *peer, 247 unsigned long hash_key) 248 { 249 peer->hash_key = hash_key; 250 rxrpc_assess_MTU_size(rx, peer); 251 peer->mtu = peer->if_mtu; 252 peer->rtt_last_req = ktime_get_real(); 253 254 switch (peer->srx.transport.family) { 255 case AF_INET: 256 peer->hdrsize = sizeof(struct iphdr); 257 break; 258 #ifdef CONFIG_AF_RXRPC_IPV6 259 case AF_INET6: 260 peer->hdrsize = sizeof(struct ipv6hdr); 261 break; 262 #endif 263 default: 264 BUG(); 265 } 266 267 switch (peer->srx.transport_type) { 268 case SOCK_DGRAM: 269 peer->hdrsize += sizeof(struct udphdr); 270 break; 271 default: 272 BUG(); 273 } 274 275 peer->hdrsize += sizeof(struct rxrpc_wire_header); 276 peer->maxdata = peer->mtu - peer->hdrsize; 277 } 278 279 /* 280 * Set up a new peer. 281 */ 282 static struct rxrpc_peer *rxrpc_create_peer(struct rxrpc_sock *rx, 283 struct rxrpc_local *local, 284 struct sockaddr_rxrpc *srx, 285 unsigned long hash_key, 286 gfp_t gfp) 287 { 288 struct rxrpc_peer *peer; 289 290 _enter(""); 291 292 peer = rxrpc_alloc_peer(local, gfp); 293 if (peer) { 294 memcpy(&peer->srx, srx, sizeof(*srx)); 295 rxrpc_init_peer(rx, peer, hash_key); 296 } 297 298 _leave(" = %p", peer); 299 return peer; 300 } 301 302 /* 303 * Set up a new incoming peer. There shouldn't be any other matching peers 304 * since we've already done a search in the list from the non-reentrant context 305 * (the data_ready handler) that is the only place we can add new peers. 306 */ 307 void rxrpc_new_incoming_peer(struct rxrpc_sock *rx, struct rxrpc_local *local, 308 struct rxrpc_peer *peer) 309 { 310 struct rxrpc_net *rxnet = local->rxnet; 311 unsigned long hash_key; 312 313 hash_key = rxrpc_peer_hash_key(local, &peer->srx); 314 rxrpc_init_peer(rx, peer, hash_key); 315 316 spin_lock(&rxnet->peer_hash_lock); 317 hash_add_rcu(rxnet->peer_hash, &peer->hash_link, hash_key); 318 list_add_tail(&peer->keepalive_link, &rxnet->peer_keepalive_new); 319 spin_unlock(&rxnet->peer_hash_lock); 320 } 321 322 /* 323 * obtain a remote transport endpoint for the specified address 324 */ 325 struct rxrpc_peer *rxrpc_lookup_peer(struct rxrpc_sock *rx, 326 struct rxrpc_local *local, 327 struct sockaddr_rxrpc *srx, gfp_t gfp) 328 { 329 struct rxrpc_peer *peer, *candidate; 330 struct rxrpc_net *rxnet = local->rxnet; 331 unsigned long hash_key = rxrpc_peer_hash_key(local, srx); 332 333 _enter("{%pISp}", &srx->transport); 334 335 /* search the peer list first */ 336 rcu_read_lock(); 337 peer = __rxrpc_lookup_peer_rcu(local, srx, hash_key); 338 if (peer && !rxrpc_get_peer_maybe(peer)) 339 peer = NULL; 340 rcu_read_unlock(); 341 342 if (!peer) { 343 /* The peer is not yet present in hash - create a candidate 344 * for a new record and then redo the search. 345 */ 346 candidate = rxrpc_create_peer(rx, local, srx, hash_key, gfp); 347 if (!candidate) { 348 _leave(" = NULL [nomem]"); 349 return NULL; 350 } 351 352 spin_lock_bh(&rxnet->peer_hash_lock); 353 354 /* Need to check that we aren't racing with someone else */ 355 peer = __rxrpc_lookup_peer_rcu(local, srx, hash_key); 356 if (peer && !rxrpc_get_peer_maybe(peer)) 357 peer = NULL; 358 if (!peer) { 359 hash_add_rcu(rxnet->peer_hash, 360 &candidate->hash_link, hash_key); 361 list_add_tail(&candidate->keepalive_link, 362 &rxnet->peer_keepalive_new); 363 } 364 365 spin_unlock_bh(&rxnet->peer_hash_lock); 366 367 if (peer) 368 kfree(candidate); 369 else 370 peer = candidate; 371 } 372 373 _net("PEER %d {%pISp}", peer->debug_id, &peer->srx.transport); 374 375 _leave(" = %p {u=%d}", peer, atomic_read(&peer->usage)); 376 return peer; 377 } 378 379 /* 380 * Get a ref on a peer record. 381 */ 382 struct rxrpc_peer *rxrpc_get_peer(struct rxrpc_peer *peer) 383 { 384 const void *here = __builtin_return_address(0); 385 int n; 386 387 n = atomic_inc_return(&peer->usage); 388 trace_rxrpc_peer(peer->debug_id, rxrpc_peer_got, n, here); 389 return peer; 390 } 391 392 /* 393 * Get a ref on a peer record unless its usage has already reached 0. 394 */ 395 struct rxrpc_peer *rxrpc_get_peer_maybe(struct rxrpc_peer *peer) 396 { 397 const void *here = __builtin_return_address(0); 398 399 if (peer) { 400 int n = atomic_fetch_add_unless(&peer->usage, 1, 0); 401 if (n > 0) 402 trace_rxrpc_peer(peer->debug_id, rxrpc_peer_got, n + 1, here); 403 else 404 peer = NULL; 405 } 406 return peer; 407 } 408 409 /* 410 * Discard a peer record. 411 */ 412 static void __rxrpc_put_peer(struct rxrpc_peer *peer) 413 { 414 struct rxrpc_net *rxnet = peer->local->rxnet; 415 416 ASSERT(hlist_empty(&peer->error_targets)); 417 418 spin_lock_bh(&rxnet->peer_hash_lock); 419 hash_del_rcu(&peer->hash_link); 420 list_del_init(&peer->keepalive_link); 421 spin_unlock_bh(&rxnet->peer_hash_lock); 422 423 rxrpc_put_local(peer->local); 424 kfree_rcu(peer, rcu); 425 } 426 427 /* 428 * Drop a ref on a peer record. 429 */ 430 void rxrpc_put_peer(struct rxrpc_peer *peer) 431 { 432 const void *here = __builtin_return_address(0); 433 unsigned int debug_id; 434 int n; 435 436 if (peer) { 437 debug_id = peer->debug_id; 438 n = atomic_dec_return(&peer->usage); 439 trace_rxrpc_peer(debug_id, rxrpc_peer_put, n, here); 440 if (n == 0) 441 __rxrpc_put_peer(peer); 442 } 443 } 444 445 /* 446 * Drop a ref on a peer record where the caller already holds the 447 * peer_hash_lock. 448 */ 449 void rxrpc_put_peer_locked(struct rxrpc_peer *peer) 450 { 451 const void *here = __builtin_return_address(0); 452 unsigned int debug_id = peer->debug_id; 453 int n; 454 455 n = atomic_dec_return(&peer->usage); 456 trace_rxrpc_peer(debug_id, rxrpc_peer_put, n, here); 457 if (n == 0) { 458 hash_del_rcu(&peer->hash_link); 459 list_del_init(&peer->keepalive_link); 460 rxrpc_put_local(peer->local); 461 kfree_rcu(peer, rcu); 462 } 463 } 464 465 /* 466 * Make sure all peer records have been discarded. 467 */ 468 void rxrpc_destroy_all_peers(struct rxrpc_net *rxnet) 469 { 470 struct rxrpc_peer *peer; 471 int i; 472 473 for (i = 0; i < HASH_SIZE(rxnet->peer_hash); i++) { 474 if (hlist_empty(&rxnet->peer_hash[i])) 475 continue; 476 477 hlist_for_each_entry(peer, &rxnet->peer_hash[i], hash_link) { 478 pr_err("Leaked peer %u {%u} %pISp\n", 479 peer->debug_id, 480 atomic_read(&peer->usage), 481 &peer->srx.transport); 482 } 483 } 484 } 485 486 /** 487 * rxrpc_kernel_get_peer - Get the peer address of a call 488 * @sock: The socket on which the call is in progress. 489 * @call: The call to query 490 * @_srx: Where to place the result 491 * 492 * Get the address of the remote peer in a call. 493 */ 494 void rxrpc_kernel_get_peer(struct socket *sock, struct rxrpc_call *call, 495 struct sockaddr_rxrpc *_srx) 496 { 497 *_srx = call->peer->srx; 498 } 499 EXPORT_SYMBOL(rxrpc_kernel_get_peer); 500 501 /** 502 * rxrpc_kernel_get_srtt - Get a call's peer smoothed RTT 503 * @sock: The socket on which the call is in progress. 504 * @call: The call to query 505 * @_srtt: Where to store the SRTT value. 506 * 507 * Get the call's peer smoothed RTT in uS. 508 */ 509 bool rxrpc_kernel_get_srtt(struct socket *sock, struct rxrpc_call *call, 510 u32 *_srtt) 511 { 512 struct rxrpc_peer *peer = call->peer; 513 514 if (peer->rtt_count == 0) { 515 *_srtt = 1000000; /* 1S */ 516 return false; 517 } 518 519 *_srtt = call->peer->srtt_us >> 3; 520 return true; 521 } 522 EXPORT_SYMBOL(rxrpc_kernel_get_srtt); 523