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         default:
  62                 BUG();
  63         }
  64 }
  65
  66 /*
  67  * Allocate a new local endpoint.
  68  */
  69 static struct rxrpc_local *rxrpc_alloc_local(const struct sockaddr_rxrpc *srx)
  70 {
  71         struct rxrpc_local *local;
  72
  73         local = kzalloc(sizeof(struct rxrpc_local), GFP_KERNEL);
  74         if (local) {
  75                 atomic_set(&local->usage, 1);
  76                 INIT_LIST_HEAD(&local->link);
  77                 INIT_WORK(&local->processor, rxrpc_local_processor);
  78                 INIT_HLIST_HEAD(&local->services);
  79                 init_rwsem(&local->defrag_sem);
  80                 skb_queue_head_init(&local->reject_queue);
  81                 skb_queue_head_init(&local->event_queue);
  82                 local->client_conns = RB_ROOT;
  83                 spin_lock_init(&local->client_conns_lock);
  84                 spin_lock_init(&local->lock);
  85                 rwlock_init(&local->services_lock);
  86                 local->debug_id = atomic_inc_return(&rxrpc_debug_id);
  87                 memcpy(&local->srx, srx, sizeof(*srx));
  88         }
  89
  90         _leave(" = %p", local);
  91         return local;
  92 }
  93
  94 /*
  95  * create the local socket
  96  * - must be called with rxrpc_local_mutex locked
  97  */
  98 static int rxrpc_open_socket(struct rxrpc_local *local)
  99 {
 100         struct sock *sock;
 101         int ret, opt;
 102
 103         _enter("%p{%d}", local, local->srx.transport_type);
 104
 105         /* create a socket to represent the local endpoint */
 106         ret = sock_create_kern(&init_net, PF_INET, local->srx.transport_type,
 107                                IPPROTO_UDP, &local->socket);
 108         if (ret < 0) {
 109                 _leave(" = %d [socket]", ret);
 110                 return ret;
 111         }
 112
 113         /* if a local address was supplied then bind it */
 114         if (local->srx.transport_len > sizeof(sa_family_t)) {
 115                 _debug("bind");
 116                 ret = kernel_bind(local->socket,
 117                                   (struct sockaddr *)&local->srx.transport,
 118                                   local->srx.transport_len);
 119                 if (ret < 0) {
 120                         _debug("bind failed %d", ret);
 121                         goto error;
 122                 }
 123         }
 124
 125         /* we want to receive ICMP errors */
 126         opt = 1;
 127         ret = kernel_setsockopt(local->socket, SOL_IP, IP_RECVERR,
 128                                 (char *) &opt, sizeof(opt));
 129         if (ret < 0) {
 130                 _debug("setsockopt failed");
 131                 goto error;
 132         }
 133
 134         /* we want to set the don't fragment bit */
 135         opt = IP_PMTUDISC_DO;
 136         ret = kernel_setsockopt(local->socket, SOL_IP, IP_MTU_DISCOVER,
 137                                 (char *) &opt, sizeof(opt));
 138         if (ret < 0) {
 139                 _debug("setsockopt failed");
 140                 goto error;
 141         }
 142
 143         /* set the socket up */
 144         sock = local->socket->sk;
 145         sock->sk_user_data      = local;
 146         sock->sk_data_ready     = rxrpc_data_ready;
 147         sock->sk_error_report   = rxrpc_error_report;
 148         _leave(" = 0");
 149         return 0;
 150
 151 error:
 152         kernel_sock_shutdown(local->socket, SHUT_RDWR);
 153         local->socket->sk->sk_user_data = NULL;
 154         sock_release(local->socket);
 155         local->socket = NULL;
 156
 157         _leave(" = %d", ret);
 158         return ret;
 159 }
 160
 161 /*
 162  * Look up or create a new local endpoint using the specified local address.
 163  */
 164 struct rxrpc_local *rxrpc_lookup_local(const struct sockaddr_rxrpc *srx)
 165 {
 166         struct rxrpc_local *local;
 167         struct list_head *cursor;
 168         const char *age;
 169         long diff;
 170         int ret;
 171
 172         if (srx->transport.family == AF_INET) {
 173                 _enter("{%d,%u,%pI4+%hu}",
 174                        srx->transport_type,
 175                        srx->transport.family,
 176                        &srx->transport.sin.sin_addr,
 177                        ntohs(srx->transport.sin.sin_port));
 178         } else {
 179                 _enter("{%d,%u}",
 180                        srx->transport_type,
 181                        srx->transport.family);
 182                 return ERR_PTR(-EAFNOSUPPORT);
 183         }
 184
 185         mutex_lock(&rxrpc_local_mutex);
 186
 187         for (cursor = rxrpc_local_endpoints.next;
 188              cursor != &rxrpc_local_endpoints;
 189              cursor = cursor->next) {
 190                 local = list_entry(cursor, struct rxrpc_local, link);
 191
 192                 diff = rxrpc_local_cmp_key(local, srx);
 193                 if (diff < 0)
 194                         continue;
 195                 if (diff > 0)
 196                         break;
 197
 198                 /* Services aren't allowed to share transport sockets, so
 199                  * reject that here.  It is possible that the object is dying -
 200                  * but it may also still have the local transport address that
 201                  * we want bound.
 202                  */
 203                 if (srx->srx_service) {
 204                         local = NULL;
 205                         goto addr_in_use;
 206                 }
 207
 208                 /* Found a match.  We replace a dying object.  Attempting to
 209                  * bind the transport socket may still fail if we're attempting
 210                  * to use a local address that the dying object is still using.
 211                  */
 212                 if (!rxrpc_get_local_maybe(local)) {
 213                         cursor = cursor->next;
 214                         list_del_init(&local->link);
 215                         break;
 216                 }
 217
 218                 age = "old";
 219                 goto found;
 220         }
 221
 222         local = rxrpc_alloc_local(srx);
 223         if (!local)
 224                 goto nomem;
 225
 226         ret = rxrpc_open_socket(local);
 227         if (ret < 0)
 228                 goto sock_error;
 229
 230         list_add_tail(&local->link, cursor);
 231         age = "new";
 232
 233 found:
 234         mutex_unlock(&rxrpc_local_mutex);
 235
 236         _net("LOCAL %s %d {%d,%u,%pI4+%hu}",
 237              age,
 238              local->debug_id,
 239              local->srx.transport_type,
 240              local->srx.transport.family,
 241              &local->srx.transport.sin.sin_addr,
 242              ntohs(local->srx.transport.sin.sin_port));
 243
 244         _leave(" = %p", local);
 245         return local;
 246
 247 nomem:
 248         ret = -ENOMEM;
 249 sock_error:
 250         mutex_unlock(&rxrpc_local_mutex);
 251         kfree(local);
 252         _leave(" = %d", ret);
 253         return ERR_PTR(ret);
 254
 255 addr_in_use:
 256         mutex_unlock(&rxrpc_local_mutex);
 257         _leave(" = -EADDRINUSE");
 258         return ERR_PTR(-EADDRINUSE);
 259 }
 260
 261 /*
 262  * A local endpoint reached its end of life.
 263  */
 264 void __rxrpc_put_local(struct rxrpc_local *local)
 265 {
 266         _enter("%d", local->debug_id);
 267         rxrpc_queue_work(&local->processor);
 268 }
 269
 270 /*
 271  * Destroy a local endpoint's socket and then hand the record to RCU to dispose
 272  * of.
 273  *
 274  * Closing the socket cannot be done from bottom half context or RCU callback
 275  * context because it might sleep.
 276  */
 277 static void rxrpc_local_destroyer(struct rxrpc_local *local)
 278 {
 279         struct socket *socket = local->socket;
 280
 281         _enter("%d", local->debug_id);
 282
 283         /* We can get a race between an incoming call packet queueing the
 284          * processor again and the work processor starting the destruction
 285          * process which will shut down the UDP socket.
 286          */
 287         if (local->dead) {
 288                 _leave(" [already dead]");
 289                 return;
 290         }
 291         local->dead = true;
 292
 293         mutex_lock(&rxrpc_local_mutex);
 294         list_del_init(&local->link);
 295         mutex_unlock(&rxrpc_local_mutex);
 296
 297         ASSERT(RB_EMPTY_ROOT(&local->client_conns));
 298         ASSERT(hlist_empty(&local->services));
 299
 300         if (socket) {
 301                 local->socket = NULL;
 302                 kernel_sock_shutdown(socket, SHUT_RDWR);
 303                 socket->sk->sk_user_data = NULL;
 304                 sock_release(socket);
 305         }
 306
 307         /* At this point, there should be no more packets coming in to the
 308          * local endpoint.
 309          */
 310         rxrpc_purge_queue(&local->reject_queue);
 311         rxrpc_purge_queue(&local->event_queue);
 312
 313         _debug("rcu local %d", local->debug_id);
 314         call_rcu(&local->rcu, rxrpc_local_rcu);
 315 }
 316
 317 /*
 318  * Process events on an endpoint
 319  */
 320 static void rxrpc_local_processor(struct work_struct *work)
 321 {
 322         struct rxrpc_local *local =
 323                 container_of(work, struct rxrpc_local, processor);
 324         bool again;
 325
 326         _enter("%d", local->debug_id);
 327
 328         do {
 329                 again = false;
 330                 if (atomic_read(&local->usage) == 0)
 331                         return rxrpc_local_destroyer(local);
 332
 333                 if (!skb_queue_empty(&local->reject_queue)) {
 334                         rxrpc_reject_packets(local);
 335                         again = true;
 336                 }
 337
 338                 if (!skb_queue_empty(&local->event_queue)) {
 339                         rxrpc_process_local_events(local);
 340                         again = true;
 341                 }
 342         } while (again);
 343 }
 344
 345 /*
 346  * Destroy a local endpoint after the RCU grace period expires.
 347  */
 348 static void rxrpc_local_rcu(struct rcu_head *rcu)
 349 {
 350         struct rxrpc_local *local = container_of(rcu, struct rxrpc_local, rcu);
 351
 352         _enter("%d", local->debug_id);
 353
 354         ASSERT(!work_pending(&local->processor));
 355
 356         _net("DESTROY LOCAL %d", local->debug_id);
 357         kfree(local);
 358         _leave("");
 359 }
 360
 361 /*
 362  * Verify the local endpoint list is empty by this point.
 363  */
 364 void __exit rxrpc_destroy_all_locals(void)
 365 {
 366         struct rxrpc_local *local;
 367
 368         _enter("");
 369
 370         flush_workqueue(rxrpc_workqueue);
 371
 372         if (!list_empty(&rxrpc_local_endpoints)) {
 373                 mutex_lock(&rxrpc_local_mutex);
 374                 list_for_each_entry(local, &rxrpc_local_endpoints, link) {
 375                         pr_err("AF_RXRPC: Leaked local %p {%d}\n",
 376                                local, atomic_read(&local->usage));
 377                 }
 378                 mutex_unlock(&rxrpc_local_mutex);
 379                 BUG();
 380         }
 381
 382         rcu_barrier();
 383 }