net/rxrpc/peer_object.c

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