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         /* we want to receive ICMP errors */
 137         opt = 1;
 138         ret = kernel_setsockopt(local->socket, SOL_IP, IP_RECVERR,
 139                                 (char *) &opt, sizeof(opt));
 140         if (ret < 0) {
 141                 _debug("setsockopt failed");
 142                 goto error;
 143         }
 144
 145         /* we want to set the don't fragment bit */
 146         opt = IP_PMTUDISC_DO;
 147         ret = kernel_setsockopt(local->socket, SOL_IP, IP_MTU_DISCOVER,
 148                                 (char *) &opt, sizeof(opt));
 149         if (ret < 0) {
 150                 _debug("setsockopt failed");
 151                 goto error;
 152         }
 153
 154         /* set the socket up */
 155         sock = local->socket->sk;
 156         sock->sk_user_data      = local;
 157         sock->sk_data_ready     = rxrpc_data_ready;
 158         sock->sk_error_report   = rxrpc_error_report;
 159         _leave(" = 0");
 160         return 0;
 161
 162 error:
 163         kernel_sock_shutdown(local->socket, SHUT_RDWR);
 164         local->socket->sk->sk_user_data = NULL;
 165         sock_release(local->socket);
 166         local->socket = NULL;
 167
 168         _leave(" = %d", ret);
 169         return ret;
 170 }
 171
 172 /*
 173  * Look up or create a new local endpoint using the specified local address.
 174  */
 175 struct rxrpc_local *rxrpc_lookup_local(struct net *net,
 176                                        const struct sockaddr_rxrpc *srx)
 177 {
 178         struct rxrpc_local *local;
 179         struct rxrpc_net *rxnet = rxrpc_net(net);
 180         struct list_head *cursor;
 181         const char *age;
 182         long diff;
 183         int ret;
 184
 185         _enter("{%d,%d,%pISp}",
 186                srx->transport_type, srx->transport.family, &srx->transport);
 187
 188         mutex_lock(&rxnet->local_mutex);
 189
 190         for (cursor = rxnet->local_endpoints.next;
 191              cursor != &rxnet->local_endpoints;
 192              cursor = cursor->next) {
 193                 local = list_entry(cursor, struct rxrpc_local, link);
 194
 195                 diff = rxrpc_local_cmp_key(local, srx);
 196                 if (diff < 0)
 197                         continue;
 198                 if (diff > 0)
 199                         break;
 200
 201                 /* Services aren't allowed to share transport sockets, so
 202                  * reject that here.  It is possible that the object is dying -
 203                  * but it may also still have the local transport address that
 204                  * we want bound.
 205                  */
 206                 if (srx->srx_service) {
 207                         local = NULL;
 208                         goto addr_in_use;
 209                 }
 210
 211                 /* Found a match.  We replace a dying object.  Attempting to
 212                  * bind the transport socket may still fail if we're attempting
 213                  * to use a local address that the dying object is still using.
 214                  */
 215                 if (!rxrpc_get_local_maybe(local)) {
 216                         cursor = cursor->next;
 217                         list_del_init(&local->link);
 218                         break;
 219                 }
 220
 221                 age = "old";
 222                 goto found;
 223         }
 224
 225         local = rxrpc_alloc_local(rxnet, srx);
 226         if (!local)
 227                 goto nomem;
 228
 229         ret = rxrpc_open_socket(local, net);
 230         if (ret < 0)
 231                 goto sock_error;
 232
 233         list_add_tail(&local->link, cursor);
 234         age = "new";
 235
 236 found:
 237         mutex_unlock(&rxnet->local_mutex);
 238
 239         _net("LOCAL %s %d {%pISp}",
 240              age, local->debug_id, &local->srx.transport);
 241
 242         _leave(" = %p", local);
 243         return local;
 244
 245 nomem:
 246         ret = -ENOMEM;
 247 sock_error:
 248         mutex_unlock(&rxnet->local_mutex);
 249         kfree(local);
 250         _leave(" = %d", ret);
 251         return ERR_PTR(ret);
 252
 253 addr_in_use:
 254         mutex_unlock(&rxnet->local_mutex);
 255         _leave(" = -EADDRINUSE");
 256         return ERR_PTR(-EADDRINUSE);
 257 }
 258
 259 /*
 260  * A local endpoint reached its end of life.
 261  */
 262 void __rxrpc_put_local(struct rxrpc_local *local)
 263 {
 264         _enter("%d", local->debug_id);
 265         rxrpc_queue_work(&local->processor);
 266 }
 267
 268 /*
 269  * Destroy a local endpoint's socket and then hand the record to RCU to dispose
 270  * of.
 271  *
 272  * Closing the socket cannot be done from bottom half context or RCU callback
 273  * context because it might sleep.
 274  */
 275 static void rxrpc_local_destroyer(struct rxrpc_local *local)
 276 {
 277         struct socket *socket = local->socket;
 278         struct rxrpc_net *rxnet = local->rxnet;
 279
 280         _enter("%d", local->debug_id);
 281
 282         /* We can get a race between an incoming call packet queueing the
 283          * processor again and the work processor starting the destruction
 284          * process which will shut down the UDP socket.
 285          */
 286         if (local->dead) {
 287                 _leave(" [already dead]");
 288                 return;
 289         }
 290         local->dead = true;
 291
 292         mutex_lock(&rxnet->local_mutex);
 293         list_del_init(&local->link);
 294         mutex_unlock(&rxnet->local_mutex);
 295
 296         ASSERT(RB_EMPTY_ROOT(&local->client_conns));
 297         ASSERT(!local->service);
 298
 299         if (socket) {
 300                 local->socket = NULL;
 301                 kernel_sock_shutdown(socket, SHUT_RDWR);
 302                 socket->sk->sk_user_data = NULL;
 303                 sock_release(socket);
 304         }
 305
 306         /* At this point, there should be no more packets coming in to the
 307          * local endpoint.
 308          */
 309         rxrpc_purge_queue(&local->reject_queue);
 310         rxrpc_purge_queue(&local->event_queue);
 311
 312         _debug("rcu local %d", local->debug_id);
 313         call_rcu(&local->rcu, rxrpc_local_rcu);
 314 }
 315
 316 /*
 317  * Process events on an endpoint
 318  */
 319 static void rxrpc_local_processor(struct work_struct *work)
 320 {
 321         struct rxrpc_local *local =
 322                 container_of(work, struct rxrpc_local, processor);
 323         bool again;
 324
 325         _enter("%d", local->debug_id);
 326
 327         do {
 328                 again = false;
 329                 if (atomic_read(&local->usage) == 0)
 330                         return rxrpc_local_destroyer(local);
 331
 332                 if (!skb_queue_empty(&local->reject_queue)) {
 333                         rxrpc_reject_packets(local);
 334                         again = true;
 335                 }
 336
 337                 if (!skb_queue_empty(&local->event_queue)) {
 338                         rxrpc_process_local_events(local);
 339                         again = true;
 340                 }
 341         } while (again);
 342 }
 343
 344 /*
 345  * Destroy a local endpoint after the RCU grace period expires.
 346  */
 347 static void rxrpc_local_rcu(struct rcu_head *rcu)
 348 {
 349         struct rxrpc_local *local = container_of(rcu, struct rxrpc_local, rcu);
 350
 351         _enter("%d", local->debug_id);
 352
 353         ASSERT(!work_pending(&local->processor));
 354
 355         _net("DESTROY LOCAL %d", local->debug_id);
 356         kfree(local);
 357         _leave("");
 358 }
 359
 360 /*
 361  * Verify the local endpoint list is empty by this point.
 362  */
 363 void rxrpc_destroy_all_locals(struct rxrpc_net *rxnet)
 364 {
 365         struct rxrpc_local *local;
 366
 367         _enter("");
 368
 369         flush_workqueue(rxrpc_workqueue);
 370
 371         if (!list_empty(&rxnet->local_endpoints)) {
 372                 mutex_lock(&rxnet->local_mutex);
 373                 list_for_each_entry(local, &rxnet->local_endpoints, link) {
 374                         pr_err("AF_RXRPC: Leaked local %p {%d}\n",
 375                                local, atomic_read(&local->usage));
 376                 }
 377                 mutex_unlock(&rxnet->local_mutex);
 378                 BUG();
 379         }
 380 }