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 if (atomic_read(&peer->usage) == 0) 129 return NULL; 130 return peer; 131 } 132 } 133 134 return NULL; 135 } 136 137 /* 138 * Look up a remote transport endpoint for the specified address using RCU. 139 */ 140 struct rxrpc_peer *rxrpc_lookup_peer_rcu(struct rxrpc_local *local, 141 const struct sockaddr_rxrpc *srx) 142 { 143 struct rxrpc_peer *peer; 144 unsigned long hash_key = rxrpc_peer_hash_key(local, srx); 145 146 peer = __rxrpc_lookup_peer_rcu(local, srx, hash_key); 147 if (peer) { 148 _net("PEER %d {%pISp}", peer->debug_id, &peer->srx.transport); 149 _leave(" = %p {u=%d}", peer, atomic_read(&peer->usage)); 150 } 151 return peer; 152 } 153 154 /* 155 * assess the MTU size for the network interface through which this peer is 156 * reached 157 */ 158 static void rxrpc_assess_MTU_size(struct rxrpc_peer *peer) 159 { 160 struct dst_entry *dst; 161 struct rtable *rt; 162 struct flowi fl; 163 struct flowi4 *fl4 = &fl.u.ip4; 164 #ifdef CONFIG_AF_RXRPC_IPV6 165 struct flowi6 *fl6 = &fl.u.ip6; 166 #endif 167 168 peer->if_mtu = 1500; 169 170 memset(&fl, 0, sizeof(fl)); 171 switch (peer->srx.transport.family) { 172 case AF_INET: 173 rt = ip_route_output_ports( 174 &init_net, fl4, NULL, 175 peer->srx.transport.sin.sin_addr.s_addr, 0, 176 htons(7000), htons(7001), IPPROTO_UDP, 0, 0); 177 if (IS_ERR(rt)) { 178 _leave(" [route err %ld]", PTR_ERR(rt)); 179 return; 180 } 181 dst = &rt->dst; 182 break; 183 184 #ifdef CONFIG_AF_RXRPC_IPV6 185 case AF_INET6: 186 fl6->flowi6_iif = LOOPBACK_IFINDEX; 187 fl6->flowi6_scope = RT_SCOPE_UNIVERSE; 188 fl6->flowi6_proto = IPPROTO_UDP; 189 memcpy(&fl6->daddr, &peer->srx.transport.sin6.sin6_addr, 190 sizeof(struct in6_addr)); 191 fl6->fl6_dport = htons(7001); 192 fl6->fl6_sport = htons(7000); 193 dst = ip6_route_output(&init_net, NULL, fl6); 194 if (dst->error) { 195 _leave(" [route err %d]", dst->error); 196 return; 197 } 198 break; 199 #endif 200 201 default: 202 BUG(); 203 } 204 205 peer->if_mtu = dst_mtu(dst); 206 dst_release(dst); 207 208 _leave(" [if_mtu %u]", peer->if_mtu); 209 } 210 211 /* 212 * Allocate a peer. 213 */ 214 struct rxrpc_peer *rxrpc_alloc_peer(struct rxrpc_local *local, gfp_t gfp) 215 { 216 struct rxrpc_peer *peer; 217 218 _enter(""); 219 220 peer = kzalloc(sizeof(struct rxrpc_peer), gfp); 221 if (peer) { 222 atomic_set(&peer->usage, 1); 223 peer->local = local; 224 INIT_HLIST_HEAD(&peer->error_targets); 225 INIT_WORK(&peer->error_distributor, 226 &rxrpc_peer_error_distributor); 227 peer->service_conns = RB_ROOT; 228 seqlock_init(&peer->service_conn_lock); 229 spin_lock_init(&peer->lock); 230 peer->debug_id = atomic_inc_return(&rxrpc_debug_id); 231 232 if (RXRPC_TX_SMSS > 2190) 233 peer->cong_cwnd = 2; 234 else if (RXRPC_TX_SMSS > 1095) 235 peer->cong_cwnd = 3; 236 else 237 peer->cong_cwnd = 4; 238 } 239 240 _leave(" = %p", peer); 241 return peer; 242 } 243 244 /* 245 * Initialise peer record. 246 */ 247 static void rxrpc_init_peer(struct rxrpc_peer *peer, unsigned long hash_key) 248 { 249 peer->hash_key = hash_key; 250 rxrpc_assess_MTU_size(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_local *local, 283 struct sockaddr_rxrpc *srx, 284 unsigned long hash_key, 285 gfp_t gfp) 286 { 287 struct rxrpc_peer *peer; 288 289 _enter(""); 290 291 peer = rxrpc_alloc_peer(local, gfp); 292 if (peer) { 293 memcpy(&peer->srx, srx, sizeof(*srx)); 294 rxrpc_init_peer(peer, hash_key); 295 } 296 297 _leave(" = %p", peer); 298 return peer; 299 } 300 301 /* 302 * Set up a new incoming peer. The address is prestored in the preallocated 303 * peer. 304 */ 305 struct rxrpc_peer *rxrpc_lookup_incoming_peer(struct rxrpc_local *local, 306 struct rxrpc_peer *prealloc) 307 { 308 struct rxrpc_peer *peer; 309 struct rxrpc_net *rxnet = local->rxnet; 310 unsigned long hash_key; 311 312 hash_key = rxrpc_peer_hash_key(local, &prealloc->srx); 313 prealloc->local = local; 314 rxrpc_init_peer(prealloc, hash_key); 315 316 spin_lock(&rxnet->peer_hash_lock); 317 318 /* Need to check that we aren't racing with someone else */ 319 peer = __rxrpc_lookup_peer_rcu(local, &prealloc->srx, hash_key); 320 if (peer && !rxrpc_get_peer_maybe(peer)) 321 peer = NULL; 322 if (!peer) { 323 peer = prealloc; 324 hash_add_rcu(rxnet->peer_hash, &peer->hash_link, hash_key); 325 list_add_tail(&peer->keepalive_link, &rxnet->peer_keepalive_new); 326 } 327 328 spin_unlock(&rxnet->peer_hash_lock); 329 return peer; 330 } 331 332 /* 333 * obtain a remote transport endpoint for the specified address 334 */ 335 struct rxrpc_peer *rxrpc_lookup_peer(struct rxrpc_local *local, 336 struct sockaddr_rxrpc *srx, gfp_t gfp) 337 { 338 struct rxrpc_peer *peer, *candidate; 339 struct rxrpc_net *rxnet = local->rxnet; 340 unsigned long hash_key = rxrpc_peer_hash_key(local, srx); 341 342 _enter("{%pISp}", &srx->transport); 343 344 /* search the peer list first */ 345 rcu_read_lock(); 346 peer = __rxrpc_lookup_peer_rcu(local, srx, hash_key); 347 if (peer && !rxrpc_get_peer_maybe(peer)) 348 peer = NULL; 349 rcu_read_unlock(); 350 351 if (!peer) { 352 /* The peer is not yet present in hash - create a candidate 353 * for a new record and then redo the search. 354 */ 355 candidate = rxrpc_create_peer(local, srx, hash_key, gfp); 356 if (!candidate) { 357 _leave(" = NULL [nomem]"); 358 return NULL; 359 } 360 361 spin_lock_bh(&rxnet->peer_hash_lock); 362 363 /* Need to check that we aren't racing with someone else */ 364 peer = __rxrpc_lookup_peer_rcu(local, srx, hash_key); 365 if (peer && !rxrpc_get_peer_maybe(peer)) 366 peer = NULL; 367 if (!peer) { 368 hash_add_rcu(rxnet->peer_hash, 369 &candidate->hash_link, hash_key); 370 list_add_tail(&candidate->keepalive_link, 371 &rxnet->peer_keepalive_new); 372 } 373 374 spin_unlock_bh(&rxnet->peer_hash_lock); 375 376 if (peer) 377 kfree(candidate); 378 else 379 peer = candidate; 380 } 381 382 _net("PEER %d {%pISp}", peer->debug_id, &peer->srx.transport); 383 384 _leave(" = %p {u=%d}", peer, atomic_read(&peer->usage)); 385 return peer; 386 } 387 388 /* 389 * Get a ref on a peer record. 390 */ 391 struct rxrpc_peer *rxrpc_get_peer(struct rxrpc_peer *peer) 392 { 393 const void *here = __builtin_return_address(0); 394 int n; 395 396 n = atomic_inc_return(&peer->usage); 397 trace_rxrpc_peer(peer, rxrpc_peer_got, n, here); 398 return peer; 399 } 400 401 /* 402 * Get a ref on a peer record unless its usage has already reached 0. 403 */ 404 struct rxrpc_peer *rxrpc_get_peer_maybe(struct rxrpc_peer *peer) 405 { 406 const void *here = __builtin_return_address(0); 407 408 if (peer) { 409 int n = atomic_fetch_add_unless(&peer->usage, 1, 0); 410 if (n > 0) 411 trace_rxrpc_peer(peer, rxrpc_peer_got, n + 1, here); 412 else 413 peer = NULL; 414 } 415 return peer; 416 } 417 418 /* 419 * Queue a peer record. This passes the caller's ref to the workqueue. 420 */ 421 void __rxrpc_queue_peer_error(struct rxrpc_peer *peer) 422 { 423 const void *here = __builtin_return_address(0); 424 int n; 425 426 n = atomic_read(&peer->usage); 427 if (rxrpc_queue_work(&peer->error_distributor)) 428 trace_rxrpc_peer(peer, rxrpc_peer_queued_error, n, here); 429 else 430 rxrpc_put_peer(peer); 431 } 432 433 /* 434 * Discard a peer record. 435 */ 436 static void __rxrpc_put_peer(struct rxrpc_peer *peer) 437 { 438 struct rxrpc_net *rxnet = peer->local->rxnet; 439 440 ASSERT(hlist_empty(&peer->error_targets)); 441 442 spin_lock_bh(&rxnet->peer_hash_lock); 443 hash_del_rcu(&peer->hash_link); 444 list_del_init(&peer->keepalive_link); 445 spin_unlock_bh(&rxnet->peer_hash_lock); 446 447 kfree_rcu(peer, rcu); 448 } 449 450 /* 451 * Drop a ref on a peer record. 452 */ 453 void rxrpc_put_peer(struct rxrpc_peer *peer) 454 { 455 const void *here = __builtin_return_address(0); 456 int n; 457 458 if (peer) { 459 n = atomic_dec_return(&peer->usage); 460 trace_rxrpc_peer(peer, rxrpc_peer_put, n, here); 461 if (n == 0) 462 __rxrpc_put_peer(peer); 463 } 464 } 465 466 /* 467 * Make sure all peer records have been discarded. 468 */ 469 void rxrpc_destroy_all_peers(struct rxrpc_net *rxnet) 470 { 471 struct rxrpc_peer *peer; 472 int i; 473 474 for (i = 0; i < HASH_SIZE(rxnet->peer_hash); i++) { 475 if (hlist_empty(&rxnet->peer_hash[i])) 476 continue; 477 478 hlist_for_each_entry(peer, &rxnet->peer_hash[i], hash_link) { 479 pr_err("Leaked peer %u {%u} %pISp\n", 480 peer->debug_id, 481 atomic_read(&peer->usage), 482 &peer->srx.transport); 483 } 484 } 485 } 486 487 /** 488 * rxrpc_kernel_get_peer - Get the peer address of a call 489 * @sock: The socket on which the call is in progress. 490 * @call: The call to query 491 * @_srx: Where to place the result 492 * 493 * Get the address of the remote peer in a call. 494 */ 495 void rxrpc_kernel_get_peer(struct socket *sock, struct rxrpc_call *call, 496 struct sockaddr_rxrpc *_srx) 497 { 498 *_srx = call->peer->srx; 499 } 500 EXPORT_SYMBOL(rxrpc_kernel_get_peer); 501 502 /** 503 * rxrpc_kernel_get_rtt - Get a call's peer RTT 504 * @sock: The socket on which the call is in progress. 505 * @call: The call to query 506 * 507 * Get the call's peer RTT. 508 */ 509 u64 rxrpc_kernel_get_rtt(struct socket *sock, struct rxrpc_call *call) 510 { 511 return call->peer->rtt; 512 } 513 EXPORT_SYMBOL(rxrpc_kernel_get_rtt); 514