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 static DEFINE_MUTEX(rxrpc_local_mutex);
  29 static LIST_HEAD(rxrpc_local_endpoints);
  30
  31 /*
  32  * Compare a local to an address.  Return -ve, 0 or +ve to indicate less than,
  33  * same or greater than.
  34  *
  35  * We explicitly don't compare the RxRPC service ID as we want to reject
  36  * conflicting uses by differing services.  Further, we don't want to share
  37  * addresses with different options (IPv6), so we don't compare those bits
  38  * either.
  39  */
  40 static long rxrpc_local_cmp_key(const struct rxrpc_local *local,
  41                                 const struct sockaddr_rxrpc *srx)
  42 {
  43         long diff;
  44
  45         diff = ((local->srx.transport_type - srx->transport_type) ?:
  46                 (local->srx.transport_len - srx->transport_len) ?:
  47                 (local->srx.transport.family - srx->transport.family));
  48         if (diff != 0)
  49                 return diff;
  50
  51         switch (srx->transport.family) {
  52         case AF_INET:
  53                 /* If the choice of UDP port is left up to the transport, then
  54                  * the endpoint record doesn't match.
  55                  */
  56                 return ((u16 __force)local->srx.transport.sin.sin_port -
  57                         (u16 __force)srx->transport.sin.sin_port) ?:
  58                         memcmp(&local->srx.transport.sin.sin_addr,
  59                                &srx->transport.sin.sin_addr,
  60                                sizeof(struct in_addr));
  61 #ifdef CONFIG_AF_RXRPC_IPV6
  62         case AF_INET6:
  63                 /* If the choice of UDP6 port is left up to the transport, then
  64                  * the endpoint record doesn't match.
  65                  */
  66                 return ((u16 __force)local->srx.transport.sin6.sin6_port -
  67                         (u16 __force)srx->transport.sin6.sin6_port) ?:
  68                         memcmp(&local->srx.transport.sin6.sin6_addr,
  69                                &srx->transport.sin6.sin6_addr,
  70                                sizeof(struct in6_addr));
  71 #endif
  72         default:
  73                 BUG();
  74         }
  75 }
  76
  77 /*
  78  * Allocate a new local endpoint.
  79  */
  80 static struct rxrpc_local *rxrpc_alloc_local(const struct sockaddr_rxrpc *srx)
  81 {
  82         struct rxrpc_local *local;
  83
  84         local = kzalloc(sizeof(struct rxrpc_local), GFP_KERNEL);
  85         if (local) {
  86                 atomic_set(&local->usage, 1);
  87                 INIT_LIST_HEAD(&local->link);
  88                 INIT_WORK(&local->processor, rxrpc_local_processor);
  89                 init_rwsem(&local->defrag_sem);
  90                 skb_queue_head_init(&local->reject_queue);
  91                 skb_queue_head_init(&local->event_queue);
  92                 local->client_conns = RB_ROOT;
  93                 spin_lock_init(&local->client_conns_lock);
  94                 spin_lock_init(&local->lock);
  95                 rwlock_init(&local->services_lock);
  96                 local->debug_id = atomic_inc_return(&rxrpc_debug_id);
  97                 memcpy(&local->srx, srx, sizeof(*srx));
  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)
 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(&init_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(const struct sockaddr_rxrpc *srx)
 176 {
 177         struct rxrpc_local *local;
 178         struct list_head *cursor;
 179         const char *age;
 180         long diff;
 181         int ret;
 182
 183         _enter("{%d,%d,%pISp}",
 184                srx->transport_type, srx->transport.family, &srx->transport);
 185
 186         mutex_lock(&rxrpc_local_mutex);
 187
 188         for (cursor = rxrpc_local_endpoints.next;
 189              cursor != &rxrpc_local_endpoints;
 190              cursor = cursor->next) {
 191                 local = list_entry(cursor, struct rxrpc_local, link);
 192
 193                 diff = rxrpc_local_cmp_key(local, srx);
 194                 if (diff < 0)
 195                         continue;
 196                 if (diff > 0)
 197                         break;
 198
 199                 /* Services aren't allowed to share transport sockets, so
 200                  * reject that here.  It is possible that the object is dying -
 201                  * but it may also still have the local transport address that
 202                  * we want bound.
 203                  */
 204                 if (srx->srx_service) {
 205                         local = NULL;
 206                         goto addr_in_use;
 207                 }
 208
 209                 /* Found a match.  We replace a dying object.  Attempting to
 210                  * bind the transport socket may still fail if we're attempting
 211                  * to use a local address that the dying object is still using.
 212                  */
 213                 if (!rxrpc_get_local_maybe(local)) {
 214                         cursor = cursor->next;
 215                         list_del_init(&local->link);
 216                         break;
 217                 }
 218
 219                 age = "old";
 220                 goto found;
 221         }
 222
 223         local = rxrpc_alloc_local(srx);
 224         if (!local)
 225                 goto nomem;
 226
 227         ret = rxrpc_open_socket(local);
 228         if (ret < 0)
 229                 goto sock_error;
 230
 231         list_add_tail(&local->link, cursor);
 232         age = "new";
 233
 234 found:
 235         mutex_unlock(&rxrpc_local_mutex);
 236
 237         _net("LOCAL %s %d {%pISp}",
 238              age, local->debug_id, &local->srx.transport);
 239
 240         _leave(" = %p", local);
 241         return local;
 242
 243 nomem:
 244         ret = -ENOMEM;
 245 sock_error:
 246         mutex_unlock(&rxrpc_local_mutex);
 247         kfree(local);
 248         _leave(" = %d", ret);
 249         return ERR_PTR(ret);
 250
 251 addr_in_use:
 252         mutex_unlock(&rxrpc_local_mutex);
 253         _leave(" = -EADDRINUSE");
 254         return ERR_PTR(-EADDRINUSE);
 255 }
 256
 257 /*
 258  * A local endpoint reached its end of life.
 259  */
 260 void __rxrpc_put_local(struct rxrpc_local *local)
 261 {
 262         _enter("%d", local->debug_id);
 263         rxrpc_queue_work(&local->processor);
 264 }
 265
 266 /*
 267  * Destroy a local endpoint's socket and then hand the record to RCU to dispose
 268  * of.
 269  *
 270  * Closing the socket cannot be done from bottom half context or RCU callback
 271  * context because it might sleep.
 272  */
 273 static void rxrpc_local_destroyer(struct rxrpc_local *local)
 274 {
 275         struct socket *socket = local->socket;
 276
 277         _enter("%d", local->debug_id);
 278
 279         /* We can get a race between an incoming call packet queueing the
 280          * processor again and the work processor starting the destruction
 281          * process which will shut down the UDP socket.
 282          */
 283         if (local->dead) {
 284                 _leave(" [already dead]");
 285                 return;
 286         }
 287         local->dead = true;
 288
 289         mutex_lock(&rxrpc_local_mutex);
 290         list_del_init(&local->link);
 291         mutex_unlock(&rxrpc_local_mutex);
 292
 293         ASSERT(RB_EMPTY_ROOT(&local->client_conns));
 294         ASSERT(!local->service);
 295
 296         if (socket) {
 297                 local->socket = NULL;
 298                 kernel_sock_shutdown(socket, SHUT_RDWR);
 299                 socket->sk->sk_user_data = NULL;
 300                 sock_release(socket);
 301         }
 302
 303         /* At this point, there should be no more packets coming in to the
 304          * local endpoint.
 305          */
 306         rxrpc_purge_queue(&local->reject_queue);
 307         rxrpc_purge_queue(&local->event_queue);
 308
 309         _debug("rcu local %d", local->debug_id);
 310         call_rcu(&local->rcu, rxrpc_local_rcu);
 311 }
 312
 313 /*
 314  * Process events on an endpoint
 315  */
 316 static void rxrpc_local_processor(struct work_struct *work)
 317 {
 318         struct rxrpc_local *local =
 319                 container_of(work, struct rxrpc_local, processor);
 320         bool again;
 321
 322         _enter("%d", local->debug_id);
 323
 324         do {
 325                 again = false;
 326                 if (atomic_read(&local->usage) == 0)
 327                         return rxrpc_local_destroyer(local);
 328
 329                 if (!skb_queue_empty(&local->reject_queue)) {
 330                         rxrpc_reject_packets(local);
 331                         again = true;
 332                 }
 333
 334                 if (!skb_queue_empty(&local->event_queue)) {
 335                         rxrpc_process_local_events(local);
 336                         again = true;
 337                 }
 338         } while (again);
 339 }
 340
 341 /*
 342  * Destroy a local endpoint after the RCU grace period expires.
 343  */
 344 static void rxrpc_local_rcu(struct rcu_head *rcu)
 345 {
 346         struct rxrpc_local *local = container_of(rcu, struct rxrpc_local, rcu);
 347
 348         _enter("%d", local->debug_id);
 349
 350         ASSERT(!work_pending(&local->processor));
 351
 352         _net("DESTROY LOCAL %d", local->debug_id);
 353         kfree(local);
 354         _leave("");
 355 }
 356
 357 /*
 358  * Verify the local endpoint list is empty by this point.
 359  */
 360 void __exit rxrpc_destroy_all_locals(void)
 361 {
 362         struct rxrpc_local *local;
 363
 364         _enter("");
 365
 366         flush_workqueue(rxrpc_workqueue);
 367
 368         if (!list_empty(&rxrpc_local_endpoints)) {
 369                 mutex_lock(&rxrpc_local_mutex);
 370                 list_for_each_entry(local, &rxrpc_local_endpoints, link) {
 371                         pr_err("AF_RXRPC: Leaked local %p {%d}\n",
 372                                local, atomic_read(&local->usage));
 373                 }
 374                 mutex_unlock(&rxrpc_local_mutex);
 375                 BUG();
 376         }
 377
 378         rcu_barrier();
 379 }