net/rxrpc/peer_object.c

   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_sock *rx,
 157                                   struct rxrpc_peer *peer)
 158 {
 159         struct net *net = sock_net(&rx->sk);
 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                         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(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                 peer->service_conns = RB_ROOT;
 226                 seqlock_init(&peer->service_conn_lock);
 227                 spin_lock_init(&peer->lock);
 228                 spin_lock_init(&peer->rtt_input_lock);
 229                 peer->debug_id = atomic_inc_return(&rxrpc_debug_id);
 230
 231                 if (RXRPC_TX_SMSS > 2190)
 232                         peer->cong_cwnd = 2;
 233                 else if (RXRPC_TX_SMSS > 1095)
 234                         peer->cong_cwnd = 3;
 235                 else
 236                         peer->cong_cwnd = 4;
 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         peer->local = local;
 315         rxrpc_init_peer(rx, peer, hash_key);
 316
 317         spin_lock(&rxnet->peer_hash_lock);
 318         hash_add_rcu(rxnet->peer_hash, &peer->hash_link, hash_key);
 319         list_add_tail(&peer->keepalive_link, &rxnet->peer_keepalive_new);
 320         spin_unlock(&rxnet->peer_hash_lock);
 321 }
 322
 323 /*
 324  * obtain a remote transport endpoint for the specified address
 325  */
 326 struct rxrpc_peer *rxrpc_lookup_peer(struct rxrpc_sock *rx,
 327                                      struct rxrpc_local *local,
 328                                      struct sockaddr_rxrpc *srx, gfp_t gfp)
 329 {
 330         struct rxrpc_peer *peer, *candidate;
 331         struct rxrpc_net *rxnet = local->rxnet;
 332         unsigned long hash_key = rxrpc_peer_hash_key(local, srx);
 333
 334         _enter("{%pISp}", &srx->transport);
 335
 336         /* search the peer list first */
 337         rcu_read_lock();
 338         peer = __rxrpc_lookup_peer_rcu(local, srx, hash_key);
 339         if (peer && !rxrpc_get_peer_maybe(peer))
 340                 peer = NULL;
 341         rcu_read_unlock();
 342
 343         if (!peer) {
 344                 /* The peer is not yet present in hash - create a candidate
 345                  * for a new record and then redo the search.
 346                  */
 347                 candidate = rxrpc_create_peer(rx, local, srx, hash_key, gfp);
 348                 if (!candidate) {
 349                         _leave(" = NULL [nomem]");
 350                         return NULL;
 351                 }
 352
 353                 spin_lock_bh(&rxnet->peer_hash_lock);
 354
 355                 /* Need to check that we aren't racing with someone else */
 356                 peer = __rxrpc_lookup_peer_rcu(local, srx, hash_key);
 357                 if (peer && !rxrpc_get_peer_maybe(peer))
 358                         peer = NULL;
 359                 if (!peer) {
 360                         hash_add_rcu(rxnet->peer_hash,
 361                                      &candidate->hash_link, hash_key);
 362                         list_add_tail(&candidate->keepalive_link,
 363                                       &rxnet->peer_keepalive_new);
 364                 }
 365
 366                 spin_unlock_bh(&rxnet->peer_hash_lock);
 367
 368                 if (peer)
 369                         kfree(candidate);
 370                 else
 371                         peer = candidate;
 372         }
 373
 374         _net("PEER %d {%pISp}", peer->debug_id, &peer->srx.transport);
 375
 376         _leave(" = %p {u=%d}", peer, atomic_read(&peer->usage));
 377         return peer;
 378 }
 379
 380 /*
 381  * Get a ref on a peer record.
 382  */
 383 struct rxrpc_peer *rxrpc_get_peer(struct rxrpc_peer *peer)
 384 {
 385         const void *here = __builtin_return_address(0);
 386         int n;
 387
 388         n = atomic_inc_return(&peer->usage);
 389         trace_rxrpc_peer(peer, rxrpc_peer_got, n, here);
 390         return peer;
 391 }
 392
 393 /*
 394  * Get a ref on a peer record unless its usage has already reached 0.
 395  */
 396 struct rxrpc_peer *rxrpc_get_peer_maybe(struct rxrpc_peer *peer)
 397 {
 398         const void *here = __builtin_return_address(0);
 399
 400         if (peer) {
 401                 int n = atomic_fetch_add_unless(&peer->usage, 1, 0);
 402                 if (n > 0)
 403                         trace_rxrpc_peer(peer, rxrpc_peer_got, n + 1, here);
 404                 else
 405                         peer = NULL;
 406         }
 407         return peer;
 408 }
 409
 410 /*
 411  * Discard a peer record.
 412  */
 413 static void __rxrpc_put_peer(struct rxrpc_peer *peer)
 414 {
 415         struct rxrpc_net *rxnet = peer->local->rxnet;
 416
 417         ASSERT(hlist_empty(&peer->error_targets));
 418
 419         spin_lock_bh(&rxnet->peer_hash_lock);
 420         hash_del_rcu(&peer->hash_link);
 421         list_del_init(&peer->keepalive_link);
 422         spin_unlock_bh(&rxnet->peer_hash_lock);
 423
 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         int n;
 434
 435         if (peer) {
 436                 n = atomic_dec_return(&peer->usage);
 437                 trace_rxrpc_peer(peer, rxrpc_peer_put, n, here);
 438                 if (n == 0)
 439                         __rxrpc_put_peer(peer);
 440         }
 441 }
 442
 443 /*
 444  * Make sure all peer records have been discarded.
 445  */
 446 void rxrpc_destroy_all_peers(struct rxrpc_net *rxnet)
 447 {
 448         struct rxrpc_peer *peer;
 449         int i;
 450
 451         for (i = 0; i < HASH_SIZE(rxnet->peer_hash); i++) {
 452                 if (hlist_empty(&rxnet->peer_hash[i]))
 453                         continue;
 454
 455                 hlist_for_each_entry(peer, &rxnet->peer_hash[i], hash_link) {
 456                         pr_err("Leaked peer %u {%u} %pISp\n",
 457                                peer->debug_id,
 458                                atomic_read(&peer->usage),
 459                                &peer->srx.transport);
 460                 }
 461         }
 462 }
 463
 464 /**
 465  * rxrpc_kernel_get_peer - Get the peer address of a call
 466  * @sock: The socket on which the call is in progress.
 467  * @call: The call to query
 468  * @_srx: Where to place the result
 469  *
 470  * Get the address of the remote peer in a call.
 471  */
 472 void rxrpc_kernel_get_peer(struct socket *sock, struct rxrpc_call *call,
 473                            struct sockaddr_rxrpc *_srx)
 474 {
 475         *_srx = call->peer->srx;
 476 }
 477 EXPORT_SYMBOL(rxrpc_kernel_get_peer);
 478
 479 /**
 480  * rxrpc_kernel_get_rtt - Get a call's peer RTT
 481  * @sock: The socket on which the call is in progress.
 482  * @call: The call to query
 483  *
 484  * Get the call's peer RTT.
 485  */
 486 u64 rxrpc_kernel_get_rtt(struct socket *sock, struct rxrpc_call *call)
 487 {
 488         return call->peer->rtt;
 489 }
 490 EXPORT_SYMBOL(rxrpc_kernel_get_rtt);