net/rxrpc/local_object.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /* Local endpoint object management
   3  *
   4  * Copyright (C) 2016 Red Hat, Inc. All Rights Reserved.
   5  * Written by David Howells (dhowells@redhat.com)
   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/slab.h>
  14 #include <linux/udp.h>
  15 #include <linux/ip.h>
  16 #include <linux/hashtable.h>
  17 #include <net/sock.h>
  18 #include <net/udp.h>
  19 #include <net/af_rxrpc.h>
  20 #include "ar-internal.h"
  21
  22 static void rxrpc_local_processor(struct work_struct *);
  23 static void rxrpc_local_rcu(struct rcu_head *);
  24
  25 /*
  26  * Compare a local to an address.  Return -ve, 0 or +ve to indicate less than,
  27  * same or greater than.
  28  *
  29  * We explicitly don't compare the RxRPC service ID as we want to reject
  30  * conflicting uses by differing services.  Further, we don't want to share
  31  * addresses with different options (IPv6), so we don't compare those bits
  32  * either.
  33  */
  34 static long rxrpc_local_cmp_key(const struct rxrpc_local *local,
  35                                 const struct sockaddr_rxrpc *srx)
  36 {
  37         long diff;
  38
  39         diff = ((local->srx.transport_type - srx->transport_type) ?:
  40                 (local->srx.transport_len - srx->transport_len) ?:
  41                 (local->srx.transport.family - srx->transport.family));
  42         if (diff != 0)
  43                 return diff;
  44
  45         switch (srx->transport.family) {
  46         case AF_INET:
  47                 /* If the choice of UDP port is left up to the transport, then
  48                  * the endpoint record doesn't match.
  49                  */
  50                 return ((u16 __force)local->srx.transport.sin.sin_port -
  51                         (u16 __force)srx->transport.sin.sin_port) ?:
  52                         memcmp(&local->srx.transport.sin.sin_addr,
  53                                &srx->transport.sin.sin_addr,
  54                                sizeof(struct in_addr));
  55 #ifdef CONFIG_AF_RXRPC_IPV6
  56         case AF_INET6:
  57                 /* If the choice of UDP6 port is left up to the transport, then
  58                  * the endpoint record doesn't match.
  59                  */
  60                 return ((u16 __force)local->srx.transport.sin6.sin6_port -
  61                         (u16 __force)srx->transport.sin6.sin6_port) ?:
  62                         memcmp(&local->srx.transport.sin6.sin6_addr,
  63                                &srx->transport.sin6.sin6_addr,
  64                                sizeof(struct in6_addr));
  65 #endif
  66         default:
  67                 BUG();
  68         }
  69 }
  70
  71 /*
  72  * Allocate a new local endpoint.
  73  */
  74 static struct rxrpc_local *rxrpc_alloc_local(struct rxrpc_net *rxnet,
  75                                              const struct sockaddr_rxrpc *srx)
  76 {
  77         struct rxrpc_local *local;
  78
  79         local = kzalloc(sizeof(struct rxrpc_local), GFP_KERNEL);
  80         if (local) {
  81                 atomic_set(&local->usage, 1);
  82                 atomic_set(&local->active_users, 1);
  83                 local->rxnet = rxnet;
  84                 INIT_LIST_HEAD(&local->link);
  85                 INIT_WORK(&local->processor, rxrpc_local_processor);
  86                 init_rwsem(&local->defrag_sem);
  87                 skb_queue_head_init(&local->reject_queue);
  88                 skb_queue_head_init(&local->event_queue);
  89                 local->client_bundles = RB_ROOT;
  90                 spin_lock_init(&local->client_bundles_lock);
  91                 spin_lock_init(&local->lock);
  92                 rwlock_init(&local->services_lock);
  93                 local->debug_id = atomic_inc_return(&rxrpc_debug_id);
  94                 memcpy(&local->srx, srx, sizeof(*srx));
  95                 local->srx.srx_service = 0;
  96                 trace_rxrpc_local(local->debug_id, rxrpc_local_new, 1, NULL);
  97         }
  98
  99         _leave(" = %p", local);
 100         return local;
 101 }
 102
 103 /*
 104  * create the local socket
 105  * - must be called with rxrpc_local_mutex locked
 106  */
 107 static int rxrpc_open_socket(struct rxrpc_local *local, struct net *net)
 108 {
 109         struct sock *usk;
 110         int ret;
 111
 112         _enter("%p{%d,%d}",
 113                local, local->srx.transport_type, local->srx.transport.family);
 114
 115         /* create a socket to represent the local endpoint */
 116         ret = sock_create_kern(net, local->srx.transport.family,
 117                                local->srx.transport_type, 0, &local->socket);
 118         if (ret < 0) {
 119                 _leave(" = %d [socket]", ret);
 120                 return ret;
 121         }
 122
 123         /* set the socket up */
 124         usk = local->socket->sk;
 125         inet_sk(usk)->mc_loop = 0;
 126
 127         /* Enable CHECKSUM_UNNECESSARY to CHECKSUM_COMPLETE conversion */
 128         inet_inc_convert_csum(usk);
 129
 130         rcu_assign_sk_user_data(usk, local);
 131
 132         udp_sk(usk)->encap_type = UDP_ENCAP_RXRPC;
 133         udp_sk(usk)->encap_rcv = rxrpc_input_packet;
 134         udp_sk(usk)->encap_destroy = NULL;
 135         udp_sk(usk)->gro_receive = NULL;
 136         udp_sk(usk)->gro_complete = NULL;
 137
 138         udp_encap_enable();
 139 #if IS_ENABLED(CONFIG_AF_RXRPC_IPV6)
 140         if (local->srx.transport.family == AF_INET6)
 141                 udpv6_encap_enable();
 142 #endif
 143         usk->sk_error_report = rxrpc_error_report;
 144
 145         /* if a local address was supplied then bind it */
 146         if (local->srx.transport_len > sizeof(sa_family_t)) {
 147                 _debug("bind");
 148                 ret = kernel_bind(local->socket,
 149                                   (struct sockaddr *)&local->srx.transport,
 150                                   local->srx.transport_len);
 151                 if (ret < 0) {
 152                         _debug("bind failed %d", ret);
 153                         goto error;
 154                 }
 155         }
 156
 157         switch (local->srx.transport.family) {
 158         case AF_INET6:
 159                 /* we want to receive ICMPv6 errors */
 160                 ip6_sock_set_recverr(local->socket->sk);
 161
 162                 /* Fall through and set IPv4 options too otherwise we don't get
 163                  * errors from IPv4 packets sent through the IPv6 socket.
 164                  */
 165                 fallthrough;
 166         case AF_INET:
 167                 /* we want to receive ICMP errors */
 168                 ip_sock_set_recverr(local->socket->sk);
 169
 170                 /* we want to set the don't fragment bit */
 171                 ip_sock_set_mtu_discover(local->socket->sk, IP_PMTUDISC_DO);
 172
 173                 /* We want receive timestamps. */
 174                 sock_enable_timestamps(local->socket->sk);
 175                 break;
 176
 177         default:
 178                 BUG();
 179         }
 180
 181         _leave(" = 0");
 182         return 0;
 183
 184 error:
 185         kernel_sock_shutdown(local->socket, SHUT_RDWR);
 186         local->socket->sk->sk_user_data = NULL;
 187         sock_release(local->socket);
 188         local->socket = NULL;
 189
 190         _leave(" = %d", ret);
 191         return ret;
 192 }
 193
 194 /*
 195  * Look up or create a new local endpoint using the specified local address.
 196  */
 197 struct rxrpc_local *rxrpc_lookup_local(struct net *net,
 198                                        const struct sockaddr_rxrpc *srx)
 199 {
 200         struct rxrpc_local *local;
 201         struct rxrpc_net *rxnet = rxrpc_net(net);
 202         struct list_head *cursor;
 203         const char *age;
 204         long diff;
 205         int ret;
 206
 207         _enter("{%d,%d,%pISp}",
 208                srx->transport_type, srx->transport.family, &srx->transport);
 209
 210         mutex_lock(&rxnet->local_mutex);
 211
 212         for (cursor = rxnet->local_endpoints.next;
 213              cursor != &rxnet->local_endpoints;
 214              cursor = cursor->next) {
 215                 local = list_entry(cursor, struct rxrpc_local, link);
 216
 217                 diff = rxrpc_local_cmp_key(local, srx);
 218                 if (diff < 0)
 219                         continue;
 220                 if (diff > 0)
 221                         break;
 222
 223                 /* Services aren't allowed to share transport sockets, so
 224                  * reject that here.  It is possible that the object is dying -
 225                  * but it may also still have the local transport address that
 226                  * we want bound.
 227                  */
 228                 if (srx->srx_service) {
 229                         local = NULL;
 230                         goto addr_in_use;
 231                 }
 232
 233                 /* Found a match.  We replace a dying object.  Attempting to
 234                  * bind the transport socket may still fail if we're attempting
 235                  * to use a local address that the dying object is still using.
 236                  */
 237                 if (!rxrpc_use_local(local))
 238                         break;
 239
 240                 age = "old";
 241                 goto found;
 242         }
 243
 244         local = rxrpc_alloc_local(rxnet, srx);
 245         if (!local)
 246                 goto nomem;
 247
 248         ret = rxrpc_open_socket(local, net);
 249         if (ret < 0)
 250                 goto sock_error;
 251
 252         if (cursor != &rxnet->local_endpoints)
 253                 list_replace_init(cursor, &local->link);
 254         else
 255                 list_add_tail(&local->link, cursor);
 256         age = "new";
 257
 258 found:
 259         mutex_unlock(&rxnet->local_mutex);
 260
 261         _net("LOCAL %s %d {%pISp}",
 262              age, local->debug_id, &local->srx.transport);
 263
 264         _leave(" = %p", local);
 265         return local;
 266
 267 nomem:
 268         ret = -ENOMEM;
 269 sock_error:
 270         mutex_unlock(&rxnet->local_mutex);
 271         if (local)
 272                 call_rcu(&local->rcu, rxrpc_local_rcu);
 273         _leave(" = %d", ret);
 274         return ERR_PTR(ret);
 275
 276 addr_in_use:
 277         mutex_unlock(&rxnet->local_mutex);
 278         _leave(" = -EADDRINUSE");
 279         return ERR_PTR(-EADDRINUSE);
 280 }
 281
 282 /*
 283  * Get a ref on a local endpoint.
 284  */
 285 struct rxrpc_local *rxrpc_get_local(struct rxrpc_local *local)
 286 {
 287         const void *here = __builtin_return_address(0);
 288         int n;
 289
 290         n = atomic_inc_return(&local->usage);
 291         trace_rxrpc_local(local->debug_id, rxrpc_local_got, n, here);
 292         return local;
 293 }
 294
 295 /*
 296  * Get a ref on a local endpoint unless its usage has already reached 0.
 297  */
 298 struct rxrpc_local *rxrpc_get_local_maybe(struct rxrpc_local *local)
 299 {
 300         const void *here = __builtin_return_address(0);
 301
 302         if (local) {
 303                 int n = atomic_fetch_add_unless(&local->usage, 1, 0);
 304                 if (n > 0)
 305                         trace_rxrpc_local(local->debug_id, rxrpc_local_got,
 306                                           n + 1, here);
 307                 else
 308                         local = NULL;
 309         }
 310         return local;
 311 }
 312
 313 /*
 314  * Queue a local endpoint and pass the caller's reference to the work item.
 315  */
 316 void rxrpc_queue_local(struct rxrpc_local *local)
 317 {
 318         const void *here = __builtin_return_address(0);
 319         unsigned int debug_id = local->debug_id;
 320         int n = atomic_read(&local->usage);
 321
 322         if (rxrpc_queue_work(&local->processor))
 323                 trace_rxrpc_local(debug_id, rxrpc_local_queued, n, here);
 324         else
 325                 rxrpc_put_local(local);
 326 }
 327
 328 /*
 329  * Drop a ref on a local endpoint.
 330  */
 331 void rxrpc_put_local(struct rxrpc_local *local)
 332 {
 333         const void *here = __builtin_return_address(0);
 334         unsigned int debug_id;
 335         int n;
 336
 337         if (local) {
 338                 debug_id = local->debug_id;
 339
 340                 n = atomic_dec_return(&local->usage);
 341                 trace_rxrpc_local(debug_id, rxrpc_local_put, n, here);
 342
 343                 if (n == 0)
 344                         call_rcu(&local->rcu, rxrpc_local_rcu);
 345         }
 346 }
 347
 348 /*
 349  * Start using a local endpoint.
 350  */
 351 struct rxrpc_local *rxrpc_use_local(struct rxrpc_local *local)
 352 {
 353         local = rxrpc_get_local_maybe(local);
 354         if (!local)
 355                 return NULL;
 356
 357         if (!__rxrpc_use_local(local)) {
 358                 rxrpc_put_local(local);
 359                 return NULL;
 360         }
 361
 362         return local;
 363 }
 364
 365 /*
 366  * Cease using a local endpoint.  Once the number of active users reaches 0, we
 367  * start the closure of the transport in the work processor.
 368  */
 369 void rxrpc_unuse_local(struct rxrpc_local *local)
 370 {
 371         if (local) {
 372                 if (__rxrpc_unuse_local(local)) {
 373                         rxrpc_get_local(local);
 374                         rxrpc_queue_local(local);
 375                 }
 376         }
 377 }
 378
 379 /*
 380  * Destroy a local endpoint's socket and then hand the record to RCU to dispose
 381  * of.
 382  *
 383  * Closing the socket cannot be done from bottom half context or RCU callback
 384  * context because it might sleep.
 385  */
 386 static void rxrpc_local_destroyer(struct rxrpc_local *local)
 387 {
 388         struct socket *socket = local->socket;
 389         struct rxrpc_net *rxnet = local->rxnet;
 390
 391         _enter("%d", local->debug_id);
 392
 393         local->dead = true;
 394
 395         mutex_lock(&rxnet->local_mutex);
 396         list_del_init(&local->link);
 397         mutex_unlock(&rxnet->local_mutex);
 398
 399         rxrpc_clean_up_local_conns(local);
 400         rxrpc_service_connection_reaper(&rxnet->service_conn_reaper);
 401         ASSERT(!local->service);
 402
 403         if (socket) {
 404                 local->socket = NULL;
 405                 kernel_sock_shutdown(socket, SHUT_RDWR);
 406                 socket->sk->sk_user_data = NULL;
 407                 sock_release(socket);
 408         }
 409
 410         /* At this point, there should be no more packets coming in to the
 411          * local endpoint.
 412          */
 413         rxrpc_purge_queue(&local->reject_queue);
 414         rxrpc_purge_queue(&local->event_queue);
 415 }
 416
 417 /*
 418  * Process events on an endpoint.  The work item carries a ref which
 419  * we must release.
 420  */
 421 static void rxrpc_local_processor(struct work_struct *work)
 422 {
 423         struct rxrpc_local *local =
 424                 container_of(work, struct rxrpc_local, processor);
 425         bool again;
 426
 427         trace_rxrpc_local(local->debug_id, rxrpc_local_processing,
 428                           atomic_read(&local->usage), NULL);
 429
 430         do {
 431                 again = false;
 432                 if (!__rxrpc_use_local(local)) {
 433                         rxrpc_local_destroyer(local);
 434                         break;
 435                 }
 436
 437                 if (!skb_queue_empty(&local->reject_queue)) {
 438                         rxrpc_reject_packets(local);
 439                         again = true;
 440                 }
 441
 442                 if (!skb_queue_empty(&local->event_queue)) {
 443                         rxrpc_process_local_events(local);
 444                         again = true;
 445                 }
 446
 447                 __rxrpc_unuse_local(local);
 448         } while (again);
 449
 450         rxrpc_put_local(local);
 451 }
 452
 453 /*
 454  * Destroy a local endpoint after the RCU grace period expires.
 455  */
 456 static void rxrpc_local_rcu(struct rcu_head *rcu)
 457 {
 458         struct rxrpc_local *local = container_of(rcu, struct rxrpc_local, rcu);
 459
 460         _enter("%d", local->debug_id);
 461
 462         ASSERT(!work_pending(&local->processor));
 463
 464         _net("DESTROY LOCAL %d", local->debug_id);
 465         kfree(local);
 466         _leave("");
 467 }
 468
 469 /*
 470  * Verify the local endpoint list is empty by this point.
 471  */
 472 void rxrpc_destroy_all_locals(struct rxrpc_net *rxnet)
 473 {
 474         struct rxrpc_local *local;
 475
 476         _enter("");
 477
 478         flush_workqueue(rxrpc_workqueue);
 479
 480         if (!list_empty(&rxnet->local_endpoints)) {
 481                 mutex_lock(&rxnet->local_mutex);
 482                 list_for_each_entry(local, &rxnet->local_endpoints, link) {
 483                         pr_err("AF_RXRPC: Leaked local %p {%d}\n",
 484                                local, atomic_read(&local->usage));
 485                 }
 486                 mutex_unlock(&rxnet->local_mutex);
 487                 BUG();
 488         }
 489 }