1 /* RxRPC remote transport endpoint record management 2 * 3 * Copyright (C) 2007, 2016 Red Hat, Inc. All Rights Reserved. 4 * Written by David Howells (dhowells@redhat.com) 5 * 6 * This program is free software; you can redistribute it and/or 7 * modify it under the terms of the GNU General Public License 8 * as published by the Free Software Foundation; either version 9 * 2 of the License, or (at your option) any later version. 10 */ 11 12 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 13 14 #include <linux/module.h> 15 #include <linux/net.h> 16 #include <linux/skbuff.h> 17 #include <linux/udp.h> 18 #include <linux/in.h> 19 #include <linux/in6.h> 20 #include <linux/slab.h> 21 #include <linux/hashtable.h> 22 #include <net/sock.h> 23 #include <net/af_rxrpc.h> 24 #include <net/ip.h> 25 #include <net/route.h> 26 #include <net/ip6_route.h> 27 #include "ar-internal.h" 28 29 /* 30 * Hash a peer key. 31 */ 32 static unsigned long rxrpc_peer_hash_key(struct rxrpc_local *local, 33 const struct sockaddr_rxrpc *srx) 34 { 35 const u16 *p; 36 unsigned int i, size; 37 unsigned long hash_key; 38 39 _enter(""); 40 41 hash_key = (unsigned long)local / __alignof__(*local); 42 hash_key += srx->transport_type; 43 hash_key += srx->transport_len; 44 hash_key += srx->transport.family; 45 46 switch (srx->transport.family) { 47 case AF_INET: 48 hash_key += (u16 __force)srx->transport.sin.sin_port; 49 size = sizeof(srx->transport.sin.sin_addr); 50 p = (u16 *)&srx->transport.sin.sin_addr; 51 break; 52 #ifdef CONFIG_AF_RXRPC_IPV6 53 case AF_INET6: 54 hash_key += (u16 __force)srx->transport.sin.sin_port; 55 size = sizeof(srx->transport.sin6.sin6_addr); 56 p = (u16 *)&srx->transport.sin6.sin6_addr; 57 break; 58 #endif 59 default: 60 WARN(1, "AF_RXRPC: Unsupported transport address family\n"); 61 return 0; 62 } 63 64 /* Step through the peer address in 16-bit portions for speed */ 65 for (i = 0; i < size; i += sizeof(*p), p++) 66 hash_key += *p; 67 68 _leave(" 0x%lx", hash_key); 69 return hash_key; 70 } 71 72 /* 73 * Compare a peer to a key. Return -ve, 0 or +ve to indicate less than, same 74 * or greater than. 75 * 76 * Unfortunately, the primitives in linux/hashtable.h don't allow for sorted 77 * buckets and mid-bucket insertion, so we don't make full use of this 78 * information at this point. 79 */ 80 static long rxrpc_peer_cmp_key(const struct rxrpc_peer *peer, 81 struct rxrpc_local *local, 82 const struct sockaddr_rxrpc *srx, 83 unsigned long hash_key) 84 { 85 long diff; 86 87 diff = ((peer->hash_key - hash_key) ?: 88 ((unsigned long)peer->local - (unsigned long)local) ?: 89 (peer->srx.transport_type - srx->transport_type) ?: 90 (peer->srx.transport_len - srx->transport_len) ?: 91 (peer->srx.transport.family - srx->transport.family)); 92 if (diff != 0) 93 return diff; 94 95 switch (srx->transport.family) { 96 case AF_INET: 97 return ((u16 __force)peer->srx.transport.sin.sin_port - 98 (u16 __force)srx->transport.sin.sin_port) ?: 99 memcmp(&peer->srx.transport.sin.sin_addr, 100 &srx->transport.sin.sin_addr, 101 sizeof(struct in_addr)); 102 #ifdef CONFIG_AF_RXRPC_IPV6 103 case AF_INET6: 104 return ((u16 __force)peer->srx.transport.sin6.sin6_port - 105 (u16 __force)srx->transport.sin6.sin6_port) ?: 106 memcmp(&peer->srx.transport.sin6.sin6_addr, 107 &srx->transport.sin6.sin6_addr, 108 sizeof(struct in6_addr)); 109 #endif 110 default: 111 BUG(); 112 } 113 } 114 115 /* 116 * Look up a remote transport endpoint for the specified address using RCU. 117 */ 118 static struct rxrpc_peer *__rxrpc_lookup_peer_rcu( 119 struct rxrpc_local *local, 120 const struct sockaddr_rxrpc *srx, 121 unsigned long hash_key) 122 { 123 struct rxrpc_peer *peer; 124 struct rxrpc_net *rxnet = local->rxnet; 125 126 hash_for_each_possible_rcu(rxnet->peer_hash, peer, hash_link, hash_key) { 127 if (rxrpc_peer_cmp_key(peer, local, srx, hash_key) == 0 && 128 atomic_read(&peer->usage) > 0) 129 return peer; 130 } 131 132 return NULL; 133 } 134 135 /* 136 * Look up a remote transport endpoint for the specified address using RCU. 137 */ 138 struct rxrpc_peer *rxrpc_lookup_peer_rcu(struct rxrpc_local *local, 139 const struct sockaddr_rxrpc *srx) 140 { 141 struct rxrpc_peer *peer; 142 unsigned long hash_key = rxrpc_peer_hash_key(local, srx); 143 144 peer = __rxrpc_lookup_peer_rcu(local, srx, hash_key); 145 if (peer) { 146 _net("PEER %d {%pISp}", peer->debug_id, &peer->srx.transport); 147 _leave(" = %p {u=%d}", peer, atomic_read(&peer->usage)); 148 } 149 return peer; 150 } 151 152 /* 153 * assess the MTU size for the network interface through which this peer is 154 * reached 155 */ 156 static void rxrpc_assess_MTU_size(struct rxrpc_peer *peer) 157 { 158 struct dst_entry *dst; 159 struct rtable *rt; 160 struct flowi fl; 161 struct flowi4 *fl4 = &fl.u.ip4; 162 #ifdef CONFIG_AF_RXRPC_IPV6 163 struct flowi6 *fl6 = &fl.u.ip6; 164 #endif 165 166 peer->if_mtu = 1500; 167 168 memset(&fl, 0, sizeof(fl)); 169 switch (peer->srx.transport.family) { 170 case AF_INET: 171 rt = ip_route_output_ports( 172 &init_net, fl4, NULL, 173 peer->srx.transport.sin.sin_addr.s_addr, 0, 174 htons(7000), htons(7001), IPPROTO_UDP, 0, 0); 175 if (IS_ERR(rt)) { 176 _leave(" [route err %ld]", PTR_ERR(rt)); 177 return; 178 } 179 dst = &rt->dst; 180 break; 181 182 #ifdef CONFIG_AF_RXRPC_IPV6 183 case AF_INET6: 184 fl6->flowi6_iif = LOOPBACK_IFINDEX; 185 fl6->flowi6_scope = RT_SCOPE_UNIVERSE; 186 fl6->flowi6_proto = IPPROTO_UDP; 187 memcpy(&fl6->daddr, &peer->srx.transport.sin6.sin6_addr, 188 sizeof(struct in6_addr)); 189 fl6->fl6_dport = htons(7001); 190 fl6->fl6_sport = htons(7000); 191 dst = ip6_route_output(&init_net, NULL, fl6); 192 if (dst->error) { 193 _leave(" [route err %d]", dst->error); 194 return; 195 } 196 break; 197 #endif 198 199 default: 200 BUG(); 201 } 202 203 peer->if_mtu = dst_mtu(dst); 204 dst_release(dst); 205 206 _leave(" [if_mtu %u]", peer->if_mtu); 207 } 208 209 /* 210 * Allocate a peer. 211 */ 212 struct rxrpc_peer *rxrpc_alloc_peer(struct rxrpc_local *local, gfp_t gfp) 213 { 214 struct rxrpc_peer *peer; 215 216 _enter(""); 217 218 peer = kzalloc(sizeof(struct rxrpc_peer), gfp); 219 if (peer) { 220 atomic_set(&peer->usage, 1); 221 peer->local = local; 222 INIT_HLIST_HEAD(&peer->error_targets); 223 peer->service_conns = RB_ROOT; 224 seqlock_init(&peer->service_conn_lock); 225 spin_lock_init(&peer->lock); 226 peer->debug_id = atomic_inc_return(&rxrpc_debug_id); 227 228 if (RXRPC_TX_SMSS > 2190) 229 peer->cong_cwnd = 2; 230 else if (RXRPC_TX_SMSS > 1095) 231 peer->cong_cwnd = 3; 232 else 233 peer->cong_cwnd = 4; 234 } 235 236 _leave(" = %p", peer); 237 return peer; 238 } 239 240 /* 241 * Initialise peer record. 242 */ 243 static void rxrpc_init_peer(struct rxrpc_peer *peer, unsigned long hash_key) 244 { 245 peer->hash_key = hash_key; 246 rxrpc_assess_MTU_size(peer); 247 peer->mtu = peer->if_mtu; 248 peer->rtt_last_req = ktime_get_real(); 249 250 switch (peer->srx.transport.family) { 251 case AF_INET: 252 peer->hdrsize = sizeof(struct iphdr); 253 break; 254 #ifdef CONFIG_AF_RXRPC_IPV6 255 case AF_INET6: 256 peer->hdrsize = sizeof(struct ipv6hdr); 257 break; 258 #endif 259 default: 260 BUG(); 261 } 262 263 switch (peer->srx.transport_type) { 264 case SOCK_DGRAM: 265 peer->hdrsize += sizeof(struct udphdr); 266 break; 267 default: 268 BUG(); 269 } 270 271 peer->hdrsize += sizeof(struct rxrpc_wire_header); 272 peer->maxdata = peer->mtu - peer->hdrsize; 273 } 274 275 /* 276 * Set up a new peer. 277 */ 278 static struct rxrpc_peer *rxrpc_create_peer(struct rxrpc_local *local, 279 struct sockaddr_rxrpc *srx, 280 unsigned long hash_key, 281 gfp_t gfp) 282 { 283 struct rxrpc_peer *peer; 284 285 _enter(""); 286 287 peer = rxrpc_alloc_peer(local, gfp); 288 if (peer) { 289 memcpy(&peer->srx, srx, sizeof(*srx)); 290 rxrpc_init_peer(peer, hash_key); 291 } 292 293 _leave(" = %p", peer); 294 return peer; 295 } 296 297 /* 298 * Set up a new incoming peer. There shouldn't be any other matching peers 299 * since we've already done a search in the list from the non-reentrant context 300 * (the data_ready handler) that is the only place we can add new peers. 301 */ 302 void rxrpc_new_incoming_peer(struct rxrpc_local *local, struct rxrpc_peer *peer) 303 { 304 struct rxrpc_net *rxnet = local->rxnet; 305 unsigned long hash_key; 306 307 hash_key = rxrpc_peer_hash_key(local, &peer->srx); 308 peer->local = local; 309 rxrpc_init_peer(peer, hash_key); 310 311 spin_lock(&rxnet->peer_hash_lock); 312 hash_add_rcu(rxnet->peer_hash, &peer->hash_link, hash_key); 313 list_add_tail(&peer->keepalive_link, &rxnet->peer_keepalive_new); 314 spin_unlock(&rxnet->peer_hash_lock); 315 } 316 317 /* 318 * obtain a remote transport endpoint for the specified address 319 */ 320 struct rxrpc_peer *rxrpc_lookup_peer(struct rxrpc_local *local, 321 struct sockaddr_rxrpc *srx, gfp_t gfp) 322 { 323 struct rxrpc_peer *peer, *candidate; 324 struct rxrpc_net *rxnet = local->rxnet; 325 unsigned long hash_key = rxrpc_peer_hash_key(local, srx); 326 327 _enter("{%pISp}", &srx->transport); 328 329 /* search the peer list first */ 330 rcu_read_lock(); 331 peer = __rxrpc_lookup_peer_rcu(local, srx, hash_key); 332 if (peer && !rxrpc_get_peer_maybe(peer)) 333 peer = NULL; 334 rcu_read_unlock(); 335 336 if (!peer) { 337 /* The peer is not yet present in hash - create a candidate 338 * for a new record and then redo the search. 339 */ 340 candidate = rxrpc_create_peer(local, srx, hash_key, gfp); 341 if (!candidate) { 342 _leave(" = NULL [nomem]"); 343 return NULL; 344 } 345 346 spin_lock_bh(&rxnet->peer_hash_lock); 347 348 /* Need to check that we aren't racing with someone else */ 349 peer = __rxrpc_lookup_peer_rcu(local, srx, hash_key); 350 if (peer && !rxrpc_get_peer_maybe(peer)) 351 peer = NULL; 352 if (!peer) { 353 hash_add_rcu(rxnet->peer_hash, 354 &candidate->hash_link, hash_key); 355 list_add_tail(&candidate->keepalive_link, 356 &rxnet->peer_keepalive_new); 357 } 358 359 spin_unlock_bh(&rxnet->peer_hash_lock); 360 361 if (peer) 362 kfree(candidate); 363 else 364 peer = candidate; 365 } 366 367 _net("PEER %d {%pISp}", peer->debug_id, &peer->srx.transport); 368 369 _leave(" = %p {u=%d}", peer, atomic_read(&peer->usage)); 370 return peer; 371 } 372 373 /* 374 * Get a ref on a peer record. 375 */ 376 struct rxrpc_peer *rxrpc_get_peer(struct rxrpc_peer *peer) 377 { 378 const void *here = __builtin_return_address(0); 379 int n; 380 381 n = atomic_inc_return(&peer->usage); 382 trace_rxrpc_peer(peer, rxrpc_peer_got, n, here); 383 return peer; 384 } 385 386 /* 387 * Get a ref on a peer record unless its usage has already reached 0. 388 */ 389 struct rxrpc_peer *rxrpc_get_peer_maybe(struct rxrpc_peer *peer) 390 { 391 const void *here = __builtin_return_address(0); 392 393 if (peer) { 394 int n = atomic_fetch_add_unless(&peer->usage, 1, 0); 395 if (n > 0) 396 trace_rxrpc_peer(peer, rxrpc_peer_got, n + 1, here); 397 else 398 peer = NULL; 399 } 400 return peer; 401 } 402 403 /* 404 * Discard a peer record. 405 */ 406 static void __rxrpc_put_peer(struct rxrpc_peer *peer) 407 { 408 struct rxrpc_net *rxnet = peer->local->rxnet; 409 410 ASSERT(hlist_empty(&peer->error_targets)); 411 412 spin_lock_bh(&rxnet->peer_hash_lock); 413 hash_del_rcu(&peer->hash_link); 414 list_del_init(&peer->keepalive_link); 415 spin_unlock_bh(&rxnet->peer_hash_lock); 416 417 kfree_rcu(peer, rcu); 418 } 419 420 /* 421 * Drop a ref on a peer record. 422 */ 423 void rxrpc_put_peer(struct rxrpc_peer *peer) 424 { 425 const void *here = __builtin_return_address(0); 426 int n; 427 428 if (peer) { 429 n = atomic_dec_return(&peer->usage); 430 trace_rxrpc_peer(peer, rxrpc_peer_put, n, here); 431 if (n == 0) 432 __rxrpc_put_peer(peer); 433 } 434 } 435 436 /* 437 * Make sure all peer records have been discarded. 438 */ 439 void rxrpc_destroy_all_peers(struct rxrpc_net *rxnet) 440 { 441 struct rxrpc_peer *peer; 442 int i; 443 444 for (i = 0; i < HASH_SIZE(rxnet->peer_hash); i++) { 445 if (hlist_empty(&rxnet->peer_hash[i])) 446 continue; 447 448 hlist_for_each_entry(peer, &rxnet->peer_hash[i], hash_link) { 449 pr_err("Leaked peer %u {%u} %pISp\n", 450 peer->debug_id, 451 atomic_read(&peer->usage), 452 &peer->srx.transport); 453 } 454 } 455 } 456 457 /** 458 * rxrpc_kernel_get_peer - Get the peer address of a call 459 * @sock: The socket on which the call is in progress. 460 * @call: The call to query 461 * @_srx: Where to place the result 462 * 463 * Get the address of the remote peer in a call. 464 */ 465 void rxrpc_kernel_get_peer(struct socket *sock, struct rxrpc_call *call, 466 struct sockaddr_rxrpc *_srx) 467 { 468 *_srx = call->peer->srx; 469 } 470 EXPORT_SYMBOL(rxrpc_kernel_get_peer); 471 472 /** 473 * rxrpc_kernel_get_rtt - Get a call's peer RTT 474 * @sock: The socket on which the call is in progress. 475 * @call: The call to query 476 * 477 * Get the call's peer RTT. 478 */ 479 u64 rxrpc_kernel_get_rtt(struct socket *sock, struct rxrpc_call *call) 480 { 481 return call->peer->rtt; 482 } 483 EXPORT_SYMBOL(rxrpc_kernel_get_rtt); 484