1 /* Local endpoint object management
3 * Copyright (C) 2016 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
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.
12 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
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>
20 #include <linux/hashtable.h>
23 #include <net/af_rxrpc.h>
24 #include "ar-internal.h"
26 static void rxrpc_local_processor(struct work_struct *);
27 static void rxrpc_local_rcu(struct rcu_head *);
30 * Compare a local to an address. Return -ve, 0 or +ve to indicate less than,
31 * same or greater than.
33 * We explicitly don't compare the RxRPC service ID as we want to reject
34 * conflicting uses by differing services. Further, we don't want to share
35 * addresses with different options (IPv6), so we don't compare those bits
38 static long rxrpc_local_cmp_key(const struct rxrpc_local *local,
39 const struct sockaddr_rxrpc *srx)
43 diff = ((local->srx.transport_type - srx->transport_type) ?:
44 (local->srx.transport_len - srx->transport_len) ?:
45 (local->srx.transport.family - srx->transport.family));
49 switch (srx->transport.family) {
51 /* If the choice of UDP port is left up to the transport, then
52 * the endpoint record doesn't match.
54 return ((u16 __force)local->srx.transport.sin.sin_port -
55 (u16 __force)srx->transport.sin.sin_port) ?:
56 memcmp(&local->srx.transport.sin.sin_addr,
57 &srx->transport.sin.sin_addr,
58 sizeof(struct in_addr));
59 #ifdef CONFIG_AF_RXRPC_IPV6
61 /* If the choice of UDP6 port is left up to the transport, then
62 * the endpoint record doesn't match.
64 return ((u16 __force)local->srx.transport.sin6.sin6_port -
65 (u16 __force)srx->transport.sin6.sin6_port) ?:
66 memcmp(&local->srx.transport.sin6.sin6_addr,
67 &srx->transport.sin6.sin6_addr,
68 sizeof(struct in6_addr));
76 * Allocate a new local endpoint.
78 static struct rxrpc_local *rxrpc_alloc_local(struct rxrpc_net *rxnet,
79 const struct sockaddr_rxrpc *srx)
81 struct rxrpc_local *local;
83 local = kzalloc(sizeof(struct rxrpc_local), GFP_KERNEL);
85 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 local->srx.srx_service = 0;
99 trace_rxrpc_local(local, rxrpc_local_new, 1, NULL);
102 _leave(" = %p", local);
107 * create the local socket
108 * - must be called with rxrpc_local_mutex locked
110 static int rxrpc_open_socket(struct rxrpc_local *local, struct net *net)
116 local, local->srx.transport_type, local->srx.transport.family);
118 /* create a socket to represent the local endpoint */
119 ret = sock_create_kern(net, local->srx.transport.family,
120 local->srx.transport_type, 0, &local->socket);
122 _leave(" = %d [socket]", ret);
126 /* set the socket up */
127 usk = local->socket->sk;
128 inet_sk(usk)->mc_loop = 0;
130 /* Enable CHECKSUM_UNNECESSARY to CHECKSUM_COMPLETE conversion */
131 inet_inc_convert_csum(usk);
133 rcu_assign_sk_user_data(usk, local);
135 udp_sk(usk)->encap_type = UDP_ENCAP_RXRPC;
136 udp_sk(usk)->encap_rcv = rxrpc_input_packet;
137 udp_sk(usk)->encap_destroy = NULL;
138 udp_sk(usk)->gro_receive = NULL;
139 udp_sk(usk)->gro_complete = NULL;
142 #if IS_ENABLED(CONFIG_AF_RXRPC_IPV6)
143 if (local->srx.transport.family == AF_INET6)
144 udpv6_encap_enable();
146 usk->sk_error_report = rxrpc_error_report;
148 /* if a local address was supplied then bind it */
149 if (local->srx.transport_len > sizeof(sa_family_t)) {
151 ret = kernel_bind(local->socket,
152 (struct sockaddr *)&local->srx.transport,
153 local->srx.transport_len);
155 _debug("bind failed %d", ret);
160 switch (local->srx.transport.family) {
162 /* we want to receive ICMPv6 errors */
164 ret = kernel_setsockopt(local->socket, SOL_IPV6, IPV6_RECVERR,
165 (char *) &opt, sizeof(opt));
167 _debug("setsockopt failed");
171 /* we want to set the don't fragment bit */
172 opt = IPV6_PMTUDISC_DO;
173 ret = kernel_setsockopt(local->socket, SOL_IPV6, IPV6_MTU_DISCOVER,
174 (char *) &opt, sizeof(opt));
176 _debug("setsockopt failed");
180 /* Fall through and set IPv4 options too otherwise we don't get
181 * errors from IPv4 packets sent through the IPv6 socket.
185 /* we want to receive ICMP errors */
187 ret = kernel_setsockopt(local->socket, SOL_IP, IP_RECVERR,
188 (char *) &opt, sizeof(opt));
190 _debug("setsockopt failed");
194 /* we want to set the don't fragment bit */
195 opt = IP_PMTUDISC_DO;
196 ret = kernel_setsockopt(local->socket, SOL_IP, IP_MTU_DISCOVER,
197 (char *) &opt, sizeof(opt));
199 _debug("setsockopt failed");
203 /* We want receive timestamps. */
205 ret = kernel_setsockopt(local->socket, SOL_SOCKET, SO_TIMESTAMPNS_OLD,
206 (char *)&opt, sizeof(opt));
208 _debug("setsockopt failed");
221 kernel_sock_shutdown(local->socket, SHUT_RDWR);
222 local->socket->sk->sk_user_data = NULL;
223 sock_release(local->socket);
224 local->socket = NULL;
226 _leave(" = %d", ret);
231 * Look up or create a new local endpoint using the specified local address.
233 struct rxrpc_local *rxrpc_lookup_local(struct net *net,
234 const struct sockaddr_rxrpc *srx)
236 struct rxrpc_local *local;
237 struct rxrpc_net *rxnet = rxrpc_net(net);
238 struct list_head *cursor;
243 _enter("{%d,%d,%pISp}",
244 srx->transport_type, srx->transport.family, &srx->transport);
246 mutex_lock(&rxnet->local_mutex);
248 for (cursor = rxnet->local_endpoints.next;
249 cursor != &rxnet->local_endpoints;
250 cursor = cursor->next) {
251 local = list_entry(cursor, struct rxrpc_local, link);
253 diff = rxrpc_local_cmp_key(local, srx);
259 /* Services aren't allowed to share transport sockets, so
260 * reject that here. It is possible that the object is dying -
261 * but it may also still have the local transport address that
264 if (srx->srx_service) {
269 /* Found a match. We replace a dying object. Attempting to
270 * bind the transport socket may still fail if we're attempting
271 * to use a local address that the dying object is still using.
273 if (!rxrpc_get_local_maybe(local)) {
274 cursor = cursor->next;
275 list_del_init(&local->link);
283 local = rxrpc_alloc_local(rxnet, srx);
287 ret = rxrpc_open_socket(local, net);
291 list_add_tail(&local->link, cursor);
295 mutex_unlock(&rxnet->local_mutex);
297 _net("LOCAL %s %d {%pISp}",
298 age, local->debug_id, &local->srx.transport);
300 _leave(" = %p", local);
306 mutex_unlock(&rxnet->local_mutex);
308 _leave(" = %d", ret);
312 mutex_unlock(&rxnet->local_mutex);
313 _leave(" = -EADDRINUSE");
314 return ERR_PTR(-EADDRINUSE);
318 * Get a ref on a local endpoint.
320 struct rxrpc_local *rxrpc_get_local(struct rxrpc_local *local)
322 const void *here = __builtin_return_address(0);
325 n = atomic_inc_return(&local->usage);
326 trace_rxrpc_local(local, rxrpc_local_got, n, here);
331 * Get a ref on a local endpoint unless its usage has already reached 0.
333 struct rxrpc_local *rxrpc_get_local_maybe(struct rxrpc_local *local)
335 const void *here = __builtin_return_address(0);
338 int n = atomic_fetch_add_unless(&local->usage, 1, 0);
340 trace_rxrpc_local(local, rxrpc_local_got, n + 1, here);
348 * Queue a local endpoint.
350 void rxrpc_queue_local(struct rxrpc_local *local)
352 const void *here = __builtin_return_address(0);
354 if (rxrpc_queue_work(&local->processor))
355 trace_rxrpc_local(local, rxrpc_local_queued,
356 atomic_read(&local->usage), here);
360 * A local endpoint reached its end of life.
362 static void __rxrpc_put_local(struct rxrpc_local *local)
364 _enter("%d", local->debug_id);
365 rxrpc_queue_work(&local->processor);
369 * Drop a ref on a local endpoint.
371 void rxrpc_put_local(struct rxrpc_local *local)
373 const void *here = __builtin_return_address(0);
377 n = atomic_dec_return(&local->usage);
378 trace_rxrpc_local(local, rxrpc_local_put, n, here);
381 __rxrpc_put_local(local);
386 * Destroy a local endpoint's socket and then hand the record to RCU to dispose
389 * Closing the socket cannot be done from bottom half context or RCU callback
390 * context because it might sleep.
392 static void rxrpc_local_destroyer(struct rxrpc_local *local)
394 struct socket *socket = local->socket;
395 struct rxrpc_net *rxnet = local->rxnet;
397 _enter("%d", local->debug_id);
399 /* We can get a race between an incoming call packet queueing the
400 * processor again and the work processor starting the destruction
401 * process which will shut down the UDP socket.
404 _leave(" [already dead]");
409 mutex_lock(&rxnet->local_mutex);
410 list_del_init(&local->link);
411 mutex_unlock(&rxnet->local_mutex);
413 ASSERT(RB_EMPTY_ROOT(&local->client_conns));
414 ASSERT(!local->service);
417 local->socket = NULL;
418 kernel_sock_shutdown(socket, SHUT_RDWR);
419 socket->sk->sk_user_data = NULL;
420 sock_release(socket);
423 /* At this point, there should be no more packets coming in to the
426 rxrpc_purge_queue(&local->reject_queue);
427 rxrpc_purge_queue(&local->event_queue);
429 _debug("rcu local %d", local->debug_id);
430 call_rcu(&local->rcu, rxrpc_local_rcu);
434 * Process events on an endpoint
436 static void rxrpc_local_processor(struct work_struct *work)
438 struct rxrpc_local *local =
439 container_of(work, struct rxrpc_local, processor);
442 trace_rxrpc_local(local, rxrpc_local_processing,
443 atomic_read(&local->usage), NULL);
447 if (atomic_read(&local->usage) == 0)
448 return rxrpc_local_destroyer(local);
450 if (!skb_queue_empty(&local->reject_queue)) {
451 rxrpc_reject_packets(local);
455 if (!skb_queue_empty(&local->event_queue)) {
456 rxrpc_process_local_events(local);
463 * Destroy a local endpoint after the RCU grace period expires.
465 static void rxrpc_local_rcu(struct rcu_head *rcu)
467 struct rxrpc_local *local = container_of(rcu, struct rxrpc_local, rcu);
469 _enter("%d", local->debug_id);
471 ASSERT(!work_pending(&local->processor));
473 _net("DESTROY LOCAL %d", local->debug_id);
479 * Verify the local endpoint list is empty by this point.
481 void rxrpc_destroy_all_locals(struct rxrpc_net *rxnet)
483 struct rxrpc_local *local;
487 flush_workqueue(rxrpc_workqueue);
489 if (!list_empty(&rxnet->local_endpoints)) {
490 mutex_lock(&rxnet->local_mutex);
491 list_for_each_entry(local, &rxnet->local_endpoints, link) {
492 pr_err("AF_RXRPC: Leaked local %p {%d}\n",
493 local, atomic_read(&local->usage));
495 mutex_unlock(&rxnet->local_mutex);