net/rxrpc/local_object.c

   1 /* Local endpoint object management
   2  *
   3  * Copyright (C) 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 Licence
   8  * as published by the Free Software Foundation; either version
   9  * 2 of the Licence, 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/slab.h>
  18 #include <linux/udp.h>
  19 #include <linux/ip.h>
  20 #include <linux/hashtable.h>
  21 #include <net/sock.h>
  22 #include <net/af_rxrpc.h>
  23 #include "ar-internal.h"
  24
  25 static void rxrpc_local_processor(struct work_struct *);
  26 static void rxrpc_local_rcu(struct rcu_head *);
  27
  28 /*
  29  * Compare a local to an address.  Return -ve, 0 or +ve to indicate less than,
  30  * same or greater than.
  31  *
  32  * We explicitly don't compare the RxRPC service ID as we want to reject
  33  * conflicting uses by differing services.  Further, we don't want to share
  34  * addresses with different options (IPv6), so we don't compare those bits
  35  * either.
  36  */
  37 static long rxrpc_local_cmp_key(const struct rxrpc_local *local,
  38                                 const struct sockaddr_rxrpc *srx)
  39 {
  40         long diff;
  41
  42         diff = ((local->srx.transport_type - srx->transport_type) ?:
  43                 (local->srx.transport_len - srx->transport_len) ?:
  44                 (local->srx.transport.family - srx->transport.family));
  45         if (diff != 0)
  46                 return diff;
  47
  48         switch (srx->transport.family) {
  49         case AF_INET:
  50                 /* If the choice of UDP port is left up to the transport, then
  51                  * the endpoint record doesn't match.
  52                  */
  53                 return ((u16 __force)local->srx.transport.sin.sin_port -
  54                         (u16 __force)srx->transport.sin.sin_port) ?:
  55                         memcmp(&local->srx.transport.sin.sin_addr,
  56                                &srx->transport.sin.sin_addr,
  57                                sizeof(struct in_addr));
  58 #ifdef CONFIG_AF_RXRPC_IPV6
  59         case AF_INET6:
  60                 /* If the choice of UDP6 port is left up to the transport, then
  61                  * the endpoint record doesn't match.
  62                  */
  63                 return ((u16 __force)local->srx.transport.sin6.sin6_port -
  64                         (u16 __force)srx->transport.sin6.sin6_port) ?:
  65                         memcmp(&local->srx.transport.sin6.sin6_addr,
  66                                &srx->transport.sin6.sin6_addr,
  67                                sizeof(struct in6_addr));
  68 #endif
  69         default:
  70                 BUG();
  71         }
  72 }
  73
  74 /*
  75  * Allocate a new local endpoint.
  76  */
  77 static struct rxrpc_local *rxrpc_alloc_local(struct rxrpc_net *rxnet,
  78                                              const struct sockaddr_rxrpc *srx)
  79 {
  80         struct rxrpc_local *local;
  81
  82         local = kzalloc(sizeof(struct rxrpc_local), GFP_KERNEL);
  83         if (local) {
  84                 atomic_set(&local->usage, 1);
  85                 local->rxnet = rxnet;
  86                 INIT_LIST_HEAD(&local->link);
  87                 INIT_WORK(&local->processor, rxrpc_local_processor);
  88                 init_rwsem(&local->defrag_sem);
  89                 skb_queue_head_init(&local->reject_queue);
  90                 skb_queue_head_init(&local->event_queue);
  91                 local->client_conns = RB_ROOT;
  92                 spin_lock_init(&local->client_conns_lock);
  93                 spin_lock_init(&local->lock);
  94                 rwlock_init(&local->services_lock);
  95                 local->debug_id = atomic_inc_return(&rxrpc_debug_id);
  96                 memcpy(&local->srx, srx, sizeof(*srx));
  97                 local->srx.srx_service = 0;
  98         }
  99
 100         _leave(" = %p", local);
 101         return local;
 102 }
 103
 104 /*
 105  * create the local socket
 106  * - must be called with rxrpc_local_mutex locked
 107  */
 108 static int rxrpc_open_socket(struct rxrpc_local *local, struct net *net)
 109 {
 110         struct sock *sock;
 111         int ret, opt;
 112
 113         _enter("%p{%d,%d}",
 114                local, local->srx.transport_type, local->srx.transport.family);
 115
 116         /* create a socket to represent the local endpoint */
 117         ret = sock_create_kern(net, local->srx.transport.family,
 118                                local->srx.transport_type, 0, &local->socket);
 119         if (ret < 0) {
 120                 _leave(" = %d [socket]", ret);
 121                 return ret;
 122         }
 123
 124         /* if a local address was supplied then bind it */
 125         if (local->srx.transport_len > sizeof(sa_family_t)) {
 126                 _debug("bind");
 127                 ret = kernel_bind(local->socket,
 128                                   (struct sockaddr *)&local->srx.transport,
 129                                   local->srx.transport_len);
 130                 if (ret < 0) {
 131                         _debug("bind failed %d", ret);
 132                         goto error;
 133                 }
 134         }
 135
 136         switch (local->srx.transport.family) {
 137         case AF_INET6:
 138                 /* we want to receive ICMPv6 errors */
 139                 opt = 1;
 140                 ret = kernel_setsockopt(local->socket, SOL_IPV6, IPV6_RECVERR,
 141                                         (char *) &opt, sizeof(opt));
 142                 if (ret < 0) {
 143                         _debug("setsockopt failed");
 144                         goto error;
 145                 }
 146
 147                 /* we want to set the don't fragment bit */
 148                 opt = IPV6_PMTUDISC_DO;
 149                 ret = kernel_setsockopt(local->socket, SOL_IPV6, IPV6_MTU_DISCOVER,
 150                                         (char *) &opt, sizeof(opt));
 151                 if (ret < 0) {
 152                         _debug("setsockopt failed");
 153                         goto error;
 154                 }
 155
 156                 /* Fall through and set IPv4 options too otherwise we don't get
 157                  * errors from IPv4 packets sent through the IPv6 socket.
 158                  */
 159
 160         case AF_INET:
 161                 /* we want to receive ICMP errors */
 162                 opt = 1;
 163                 ret = kernel_setsockopt(local->socket, SOL_IP, IP_RECVERR,
 164                                         (char *) &opt, sizeof(opt));
 165                 if (ret < 0) {
 166                         _debug("setsockopt failed");
 167                         goto error;
 168                 }
 169
 170                 /* we want to set the don't fragment bit */
 171                 opt = IP_PMTUDISC_DO;
 172                 ret = kernel_setsockopt(local->socket, SOL_IP, IP_MTU_DISCOVER,
 173                                         (char *) &opt, sizeof(opt));
 174                 if (ret < 0) {
 175                         _debug("setsockopt failed");
 176                         goto error;
 177                 }
 178                 break;
 179
 180         default:
 181                 BUG();
 182         }
 183
 184         /* set the socket up */
 185         sock = local->socket->sk;
 186         sock->sk_user_data      = local;
 187         sock->sk_data_ready     = rxrpc_data_ready;
 188         sock->sk_error_report   = rxrpc_error_report;
 189         _leave(" = 0");
 190         return 0;
 191
 192 error:
 193         kernel_sock_shutdown(local->socket, SHUT_RDWR);
 194         local->socket->sk->sk_user_data = NULL;
 195         sock_release(local->socket);
 196         local->socket = NULL;
 197
 198         _leave(" = %d", ret);
 199         return ret;
 200 }
 201
 202 /*
 203  * Look up or create a new local endpoint using the specified local address.
 204  */
 205 struct rxrpc_local *rxrpc_lookup_local(struct net *net,
 206                                        const struct sockaddr_rxrpc *srx)
 207 {
 208         struct rxrpc_local *local;
 209         struct rxrpc_net *rxnet = rxrpc_net(net);
 210         struct list_head *cursor;
 211         const char *age;
 212         long diff;
 213         int ret;
 214
 215         _enter("{%d,%d,%pISp}",
 216                srx->transport_type, srx->transport.family, &srx->transport);
 217
 218         mutex_lock(&rxnet->local_mutex);
 219
 220         for (cursor = rxnet->local_endpoints.next;
 221              cursor != &rxnet->local_endpoints;
 222              cursor = cursor->next) {
 223                 local = list_entry(cursor, struct rxrpc_local, link);
 224
 225                 diff = rxrpc_local_cmp_key(local, srx);
 226                 if (diff < 0)
 227                         continue;
 228                 if (diff > 0)
 229                         break;
 230
 231                 /* Services aren't allowed to share transport sockets, so
 232                  * reject that here.  It is possible that the object is dying -
 233                  * but it may also still have the local transport address that
 234                  * we want bound.
 235                  */
 236                 if (srx->srx_service) {
 237                         local = NULL;
 238                         goto addr_in_use;
 239                 }
 240
 241                 /* Found a match.  We replace a dying object.  Attempting to
 242                  * bind the transport socket may still fail if we're attempting
 243                  * to use a local address that the dying object is still using.
 244                  */
 245                 if (!rxrpc_get_local_maybe(local)) {
 246                         cursor = cursor->next;
 247                         list_del_init(&local->link);
 248                         break;
 249                 }
 250
 251                 age = "old";
 252                 goto found;
 253         }
 254
 255         local = rxrpc_alloc_local(rxnet, srx);
 256         if (!local)
 257                 goto nomem;
 258
 259         ret = rxrpc_open_socket(local, net);
 260         if (ret < 0)
 261                 goto sock_error;
 262
 263         list_add_tail(&local->link, cursor);
 264         age = "new";
 265
 266 found:
 267         mutex_unlock(&rxnet->local_mutex);
 268
 269         _net("LOCAL %s %d {%pISp}",
 270              age, local->debug_id, &local->srx.transport);
 271
 272         _leave(" = %p", local);
 273         return local;
 274
 275 nomem:
 276         ret = -ENOMEM;
 277 sock_error:
 278         mutex_unlock(&rxnet->local_mutex);
 279         kfree(local);
 280         _leave(" = %d", ret);
 281         return ERR_PTR(ret);
 282
 283 addr_in_use:
 284         mutex_unlock(&rxnet->local_mutex);
 285         _leave(" = -EADDRINUSE");
 286         return ERR_PTR(-EADDRINUSE);
 287 }
 288
 289 /*
 290  * A local endpoint reached its end of life.
 291  */
 292 void __rxrpc_put_local(struct rxrpc_local *local)
 293 {
 294         _enter("%d", local->debug_id);
 295         rxrpc_queue_work(&local->processor);
 296 }
 297
 298 /*
 299  * Destroy a local endpoint's socket and then hand the record to RCU to dispose
 300  * of.
 301  *
 302  * Closing the socket cannot be done from bottom half context or RCU callback
 303  * context because it might sleep.
 304  */
 305 static void rxrpc_local_destroyer(struct rxrpc_local *local)
 306 {
 307         struct socket *socket = local->socket;
 308         struct rxrpc_net *rxnet = local->rxnet;
 309
 310         _enter("%d", local->debug_id);
 311
 312         /* We can get a race between an incoming call packet queueing the
 313          * processor again and the work processor starting the destruction
 314          * process which will shut down the UDP socket.
 315          */
 316         if (local->dead) {
 317                 _leave(" [already dead]");
 318                 return;
 319         }
 320         local->dead = true;
 321
 322         mutex_lock(&rxnet->local_mutex);
 323         list_del_init(&local->link);
 324         mutex_unlock(&rxnet->local_mutex);
 325
 326         ASSERT(RB_EMPTY_ROOT(&local->client_conns));
 327         ASSERT(!local->service);
 328
 329         if (socket) {
 330                 local->socket = NULL;
 331                 kernel_sock_shutdown(socket, SHUT_RDWR);
 332                 socket->sk->sk_user_data = NULL;
 333                 sock_release(socket);
 334         }
 335
 336         /* At this point, there should be no more packets coming in to the
 337          * local endpoint.
 338          */
 339         rxrpc_purge_queue(&local->reject_queue);
 340         rxrpc_purge_queue(&local->event_queue);
 341
 342         _debug("rcu local %d", local->debug_id);
 343         call_rcu(&local->rcu, rxrpc_local_rcu);
 344 }
 345
 346 /*
 347  * Process events on an endpoint
 348  */
 349 static void rxrpc_local_processor(struct work_struct *work)
 350 {
 351         struct rxrpc_local *local =
 352                 container_of(work, struct rxrpc_local, processor);
 353         bool again;
 354
 355         _enter("%d", local->debug_id);
 356
 357         do {
 358                 again = false;
 359                 if (atomic_read(&local->usage) == 0)
 360                         return rxrpc_local_destroyer(local);
 361
 362                 if (!skb_queue_empty(&local->reject_queue)) {
 363                         rxrpc_reject_packets(local);
 364                         again = true;
 365                 }
 366
 367                 if (!skb_queue_empty(&local->event_queue)) {
 368                         rxrpc_process_local_events(local);
 369                         again = true;
 370                 }
 371         } while (again);
 372 }
 373
 374 /*
 375  * Destroy a local endpoint after the RCU grace period expires.
 376  */
 377 static void rxrpc_local_rcu(struct rcu_head *rcu)
 378 {
 379         struct rxrpc_local *local = container_of(rcu, struct rxrpc_local, rcu);
 380
 381         _enter("%d", local->debug_id);
 382
 383         ASSERT(!work_pending(&local->processor));
 384
 385         _net("DESTROY LOCAL %d", local->debug_id);
 386         kfree(local);
 387         _leave("");
 388 }
 389
 390 /*
 391  * Verify the local endpoint list is empty by this point.
 392  */
 393 void rxrpc_destroy_all_locals(struct rxrpc_net *rxnet)
 394 {
 395         struct rxrpc_local *local;
 396
 397         _enter("");
 398
 399         flush_workqueue(rxrpc_workqueue);
 400
 401         if (!list_empty(&rxnet->local_endpoints)) {
 402                 mutex_lock(&rxnet->local_mutex);
 403                 list_for_each_entry(local, &rxnet->local_endpoints, link) {
 404                         pr_err("AF_RXRPC: Leaked local %p {%d}\n",
 405                                local, atomic_read(&local->usage));
 406                 }
 407                 mutex_unlock(&rxnet->local_mutex);
 408                 BUG();
 409         }
 410 }