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_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                 INIT_WORK(&peer->error_distributor,
 224                           &rxrpc_peer_error_distributor);
 225                 peer->service_conns = RB_ROOT;
 226                 seqlock_init(&peer->service_conn_lock);
 227                 spin_lock_init(&peer->lock);
 228                 peer->debug_id = atomic_inc_return(&rxrpc_debug_id);
 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         }
 237
 238         _leave(" = %p", peer);
 239         return peer;
 240 }
 241
 242 /*
 243  * Initialise peer record.
 244  */
 245 static void rxrpc_init_peer(struct rxrpc_peer *peer, unsigned long hash_key)
 246 {
 247         peer->hash_key = hash_key;
 248         rxrpc_assess_MTU_size(peer);
 249         peer->mtu = peer->if_mtu;
 250         peer->rtt_last_req = ktime_get_real();
 251
 252         switch (peer->srx.transport.family) {
 253         case AF_INET:
 254                 peer->hdrsize = sizeof(struct iphdr);
 255                 break;
 256 #ifdef CONFIG_AF_RXRPC_IPV6
 257         case AF_INET6:
 258                 peer->hdrsize = sizeof(struct ipv6hdr);
 259                 break;
 260 #endif
 261         default:
 262                 BUG();
 263         }
 264
 265         switch (peer->srx.transport_type) {
 266         case SOCK_DGRAM:
 267                 peer->hdrsize += sizeof(struct udphdr);
 268                 break;
 269         default:
 270                 BUG();
 271         }
 272
 273         peer->hdrsize += sizeof(struct rxrpc_wire_header);
 274         peer->maxdata = peer->mtu - peer->hdrsize;
 275 }
 276
 277 /*
 278  * Set up a new peer.
 279  */
 280 static struct rxrpc_peer *rxrpc_create_peer(struct rxrpc_local *local,
 281                                             struct sockaddr_rxrpc *srx,
 282                                             unsigned long hash_key,
 283                                             gfp_t gfp)
 284 {
 285         struct rxrpc_peer *peer;
 286
 287         _enter("");
 288
 289         peer = rxrpc_alloc_peer(local, gfp);
 290         if (peer) {
 291                 memcpy(&peer->srx, srx, sizeof(*srx));
 292                 rxrpc_init_peer(peer, hash_key);
 293         }
 294
 295         _leave(" = %p", peer);
 296         return peer;
 297 }
 298
 299 /*
 300  * Set up a new incoming peer.  There shouldn't be any other matching peers
 301  * since we've already done a search in the list from the non-reentrant context
 302  * (the data_ready handler) that is the only place we can add new peers.
 303  */
 304 void rxrpc_new_incoming_peer(struct rxrpc_local *local, struct rxrpc_peer *peer)
 305 {
 306         struct rxrpc_net *rxnet = local->rxnet;
 307         unsigned long hash_key;
 308
 309         hash_key = rxrpc_peer_hash_key(local, &peer->srx);
 310         peer->local = local;
 311         rxrpc_init_peer(peer, hash_key);
 312
 313         spin_lock(&rxnet->peer_hash_lock);
 314         hash_add_rcu(rxnet->peer_hash, &peer->hash_link, hash_key);
 315         list_add_tail(&peer->keepalive_link, &rxnet->peer_keepalive_new);
 316         spin_unlock(&rxnet->peer_hash_lock);
 317 }
 318
 319 /*
 320  * obtain a remote transport endpoint for the specified address
 321  */
 322 struct rxrpc_peer *rxrpc_lookup_peer(struct rxrpc_local *local,
 323                                      struct sockaddr_rxrpc *srx, gfp_t gfp)
 324 {
 325         struct rxrpc_peer *peer, *candidate;
 326         struct rxrpc_net *rxnet = local->rxnet;
 327         unsigned long hash_key = rxrpc_peer_hash_key(local, srx);
 328
 329         _enter("{%pISp}", &srx->transport);
 330
 331         /* search the peer list first */
 332         rcu_read_lock();
 333         peer = __rxrpc_lookup_peer_rcu(local, srx, hash_key);
 334         if (peer && !rxrpc_get_peer_maybe(peer))
 335                 peer = NULL;
 336         rcu_read_unlock();
 337
 338         if (!peer) {
 339                 /* The peer is not yet present in hash - create a candidate
 340                  * for a new record and then redo the search.
 341                  */
 342                 candidate = rxrpc_create_peer(local, srx, hash_key, gfp);
 343                 if (!candidate) {
 344                         _leave(" = NULL [nomem]");
 345                         return NULL;
 346                 }
 347
 348                 spin_lock_bh(&rxnet->peer_hash_lock);
 349
 350                 /* Need to check that we aren't racing with someone else */
 351                 peer = __rxrpc_lookup_peer_rcu(local, srx, hash_key);
 352                 if (peer && !rxrpc_get_peer_maybe(peer))
 353                         peer = NULL;
 354                 if (!peer) {
 355                         hash_add_rcu(rxnet->peer_hash,
 356                                      &candidate->hash_link, hash_key);
 357                         list_add_tail(&candidate->keepalive_link,
 358                                       &rxnet->peer_keepalive_new);
 359                 }
 360
 361                 spin_unlock_bh(&rxnet->peer_hash_lock);
 362
 363                 if (peer)
 364                         kfree(candidate);
 365                 else
 366                         peer = candidate;
 367         }
 368
 369         _net("PEER %d {%pISp}", peer->debug_id, &peer->srx.transport);
 370
 371         _leave(" = %p {u=%d}", peer, atomic_read(&peer->usage));
 372         return peer;
 373 }
 374
 375 /*
 376  * Get a ref on a peer record.
 377  */
 378 struct rxrpc_peer *rxrpc_get_peer(struct rxrpc_peer *peer)
 379 {
 380         const void *here = __builtin_return_address(0);
 381         int n;
 382
 383         n = atomic_inc_return(&peer->usage);
 384         trace_rxrpc_peer(peer, rxrpc_peer_got, n, here);
 385         return peer;
 386 }
 387
 388 /*
 389  * Get a ref on a peer record unless its usage has already reached 0.
 390  */
 391 struct rxrpc_peer *rxrpc_get_peer_maybe(struct rxrpc_peer *peer)
 392 {
 393         const void *here = __builtin_return_address(0);
 394
 395         if (peer) {
 396                 int n = atomic_fetch_add_unless(&peer->usage, 1, 0);
 397                 if (n > 0)
 398                         trace_rxrpc_peer(peer, rxrpc_peer_got, n + 1, here);
 399                 else
 400                         peer = NULL;
 401         }
 402         return peer;
 403 }
 404
 405 /*
 406  * Queue a peer record.  This passes the caller's ref to the workqueue.
 407  */
 408 void __rxrpc_queue_peer_error(struct rxrpc_peer *peer)
 409 {
 410         const void *here = __builtin_return_address(0);
 411         int n;
 412
 413         n = atomic_read(&peer->usage);
 414         if (rxrpc_queue_work(&peer->error_distributor))
 415                 trace_rxrpc_peer(peer, rxrpc_peer_queued_error, n, here);
 416         else
 417                 rxrpc_put_peer(peer);
 418 }
 419
 420 /*
 421  * Discard a peer record.
 422  */
 423 static void __rxrpc_put_peer(struct rxrpc_peer *peer)
 424 {
 425         struct rxrpc_net *rxnet = peer->local->rxnet;
 426
 427         ASSERT(hlist_empty(&peer->error_targets));
 428
 429         spin_lock_bh(&rxnet->peer_hash_lock);
 430         hash_del_rcu(&peer->hash_link);
 431         list_del_init(&peer->keepalive_link);
 432         spin_unlock_bh(&rxnet->peer_hash_lock);
 433
 434         kfree_rcu(peer, rcu);
 435 }
 436
 437 /*
 438  * Drop a ref on a peer record.
 439  */
 440 void rxrpc_put_peer(struct rxrpc_peer *peer)
 441 {
 442         const void *here = __builtin_return_address(0);
 443         int n;
 444
 445         if (peer) {
 446                 n = atomic_dec_return(&peer->usage);
 447                 trace_rxrpc_peer(peer, rxrpc_peer_put, n, here);
 448                 if (n == 0)
 449                         __rxrpc_put_peer(peer);
 450         }
 451 }
 452
 453 /*
 454  * Make sure all peer records have been discarded.
 455  */
 456 void rxrpc_destroy_all_peers(struct rxrpc_net *rxnet)
 457 {
 458         struct rxrpc_peer *peer;
 459         int i;
 460
 461         for (i = 0; i < HASH_SIZE(rxnet->peer_hash); i++) {
 462                 if (hlist_empty(&rxnet->peer_hash[i]))
 463                         continue;
 464
 465                 hlist_for_each_entry(peer, &rxnet->peer_hash[i], hash_link) {
 466                         pr_err("Leaked peer %u {%u} %pISp\n",
 467                                peer->debug_id,
 468                                atomic_read(&peer->usage),
 469                                &peer->srx.transport);
 470                 }
 471         }
 472 }
 473
 474 /**
 475  * rxrpc_kernel_get_peer - Get the peer address of a call
 476  * @sock: The socket on which the call is in progress.
 477  * @call: The call to query
 478  * @_srx: Where to place the result
 479  *
 480  * Get the address of the remote peer in a call.
 481  */
 482 void rxrpc_kernel_get_peer(struct socket *sock, struct rxrpc_call *call,
 483                            struct sockaddr_rxrpc *_srx)
 484 {
 485         *_srx = call->peer->srx;
 486 }
 487 EXPORT_SYMBOL(rxrpc_kernel_get_peer);
 488
 489 /**
 490  * rxrpc_kernel_get_rtt - Get a call's peer RTT
 491  * @sock: The socket on which the call is in progress.
 492  * @call: The call to query
 493  *
 494  * Get the call's peer RTT.
 495  */
 496 u64 rxrpc_kernel_get_rtt(struct socket *sock, struct rxrpc_call *call)
 497 {
 498         return call->peer->rtt;
 499 }
 500 EXPORT_SYMBOL(rxrpc_kernel_get_rtt);