1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* Local endpoint object management
4 * Copyright (C) 2016 Red Hat, Inc. All Rights Reserved.
5 * Written by David Howells (dhowells@redhat.com)
8 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
10 #include <linux/module.h>
11 #include <linux/net.h>
12 #include <linux/skbuff.h>
13 #include <linux/slab.h>
14 #include <linux/udp.h>
16 #include <linux/hashtable.h>
19 #include <net/udp_tunnel.h>
20 #include <net/af_rxrpc.h>
21 #include "ar-internal.h"
23 static void rxrpc_local_rcu(struct rcu_head *);
26 * Handle an ICMP/ICMP6 error turning up at the tunnel. Push it through the
27 * usual mechanism so that it gets parsed and presented through the UDP
28 * socket's error_report().
30 static void rxrpc_encap_err_rcv(struct sock *sk, struct sk_buff *skb, int err,
31 __be16 port, u32 info, u8 *payload)
33 if (ip_hdr(skb)->version == IPVERSION)
34 return ip_icmp_error(sk, skb, err, port, info, payload);
35 if (IS_ENABLED(CONFIG_AF_RXRPC_IPV6))
36 return ipv6_icmp_error(sk, skb, err, port, info, payload);
40 * Compare a local to an address. Return -ve, 0 or +ve to indicate less than,
41 * same or greater than.
43 * We explicitly don't compare the RxRPC service ID as we want to reject
44 * conflicting uses by differing services. Further, we don't want to share
45 * addresses with different options (IPv6), so we don't compare those bits
48 static long rxrpc_local_cmp_key(const struct rxrpc_local *local,
49 const struct sockaddr_rxrpc *srx)
53 diff = ((local->srx.transport_type - srx->transport_type) ?:
54 (local->srx.transport_len - srx->transport_len) ?:
55 (local->srx.transport.family - srx->transport.family));
59 switch (srx->transport.family) {
61 /* If the choice of UDP port is left up to the transport, then
62 * the endpoint record doesn't match.
64 return ((u16 __force)local->srx.transport.sin.sin_port -
65 (u16 __force)srx->transport.sin.sin_port) ?:
66 memcmp(&local->srx.transport.sin.sin_addr,
67 &srx->transport.sin.sin_addr,
68 sizeof(struct in_addr));
69 #ifdef CONFIG_AF_RXRPC_IPV6
71 /* If the choice of UDP6 port is left up to the transport, then
72 * the endpoint record doesn't match.
74 return ((u16 __force)local->srx.transport.sin6.sin6_port -
75 (u16 __force)srx->transport.sin6.sin6_port) ?:
76 memcmp(&local->srx.transport.sin6.sin6_addr,
77 &srx->transport.sin6.sin6_addr,
78 sizeof(struct in6_addr));
86 * Allocate a new local endpoint.
88 static struct rxrpc_local *rxrpc_alloc_local(struct rxrpc_net *rxnet,
89 const struct sockaddr_rxrpc *srx)
91 struct rxrpc_local *local;
93 local = kzalloc(sizeof(struct rxrpc_local), GFP_KERNEL);
95 refcount_set(&local->ref, 1);
96 atomic_set(&local->active_users, 1);
98 INIT_HLIST_NODE(&local->link);
99 INIT_LIST_HEAD(&local->ack_tx_queue);
100 spin_lock_init(&local->ack_tx_lock);
101 init_rwsem(&local->defrag_sem);
102 skb_queue_head_init(&local->rx_queue);
103 INIT_LIST_HEAD(&local->call_attend_q);
104 local->client_bundles = RB_ROOT;
105 spin_lock_init(&local->client_bundles_lock);
106 spin_lock_init(&local->lock);
107 rwlock_init(&local->services_lock);
108 local->debug_id = atomic_inc_return(&rxrpc_debug_id);
109 memcpy(&local->srx, srx, sizeof(*srx));
110 local->srx.srx_service = 0;
111 trace_rxrpc_local(local->debug_id, rxrpc_local_new, 1, 1);
114 _leave(" = %p", local);
119 * create the local socket
120 * - must be called with rxrpc_local_mutex locked
122 static int rxrpc_open_socket(struct rxrpc_local *local, struct net *net)
124 struct udp_tunnel_sock_cfg tuncfg = {NULL};
125 struct sockaddr_rxrpc *srx = &local->srx;
126 struct udp_port_cfg udp_conf = {0};
127 struct task_struct *io_thread;
132 local, srx->transport_type, srx->transport.family);
134 udp_conf.family = srx->transport.family;
135 udp_conf.use_udp_checksums = true;
136 if (udp_conf.family == AF_INET) {
137 udp_conf.local_ip = srx->transport.sin.sin_addr;
138 udp_conf.local_udp_port = srx->transport.sin.sin_port;
139 #if IS_ENABLED(CONFIG_AF_RXRPC_IPV6)
141 udp_conf.local_ip6 = srx->transport.sin6.sin6_addr;
142 udp_conf.local_udp_port = srx->transport.sin6.sin6_port;
143 udp_conf.use_udp6_tx_checksums = true;
144 udp_conf.use_udp6_rx_checksums = true;
147 ret = udp_sock_create(net, &udp_conf, &local->socket);
149 _leave(" = %d [socket]", ret);
153 tuncfg.encap_type = UDP_ENCAP_RXRPC;
154 tuncfg.encap_rcv = rxrpc_encap_rcv;
155 tuncfg.encap_err_rcv = rxrpc_encap_err_rcv;
156 tuncfg.sk_user_data = local;
157 setup_udp_tunnel_sock(net, local->socket, &tuncfg);
159 /* set the socket up */
160 usk = local->socket->sk;
161 usk->sk_error_report = rxrpc_error_report;
163 switch (srx->transport.family) {
165 /* we want to receive ICMPv6 errors */
166 ip6_sock_set_recverr(usk);
168 /* Fall through and set IPv4 options too otherwise we don't get
169 * errors from IPv4 packets sent through the IPv6 socket.
173 /* we want to receive ICMP errors */
174 ip_sock_set_recverr(usk);
176 /* we want to set the don't fragment bit */
177 ip_sock_set_mtu_discover(usk, IP_PMTUDISC_DO);
179 /* We want receive timestamps. */
180 sock_enable_timestamps(usk);
187 io_thread = kthread_run(rxrpc_io_thread, local,
188 "krxrpcio/%u", ntohs(udp_conf.local_udp_port));
189 if (IS_ERR(io_thread)) {
190 ret = PTR_ERR(io_thread);
194 local->io_thread = io_thread;
199 kernel_sock_shutdown(local->socket, SHUT_RDWR);
200 local->socket->sk->sk_user_data = NULL;
201 sock_release(local->socket);
202 local->socket = NULL;
207 * Look up or create a new local endpoint using the specified local address.
209 struct rxrpc_local *rxrpc_lookup_local(struct net *net,
210 const struct sockaddr_rxrpc *srx)
212 struct rxrpc_local *local;
213 struct rxrpc_net *rxnet = rxrpc_net(net);
214 struct hlist_node *cursor;
218 _enter("{%d,%d,%pISp}",
219 srx->transport_type, srx->transport.family, &srx->transport);
221 mutex_lock(&rxnet->local_mutex);
223 hlist_for_each(cursor, &rxnet->local_endpoints) {
224 local = hlist_entry(cursor, struct rxrpc_local, link);
226 diff = rxrpc_local_cmp_key(local, srx);
230 /* Services aren't allowed to share transport sockets, so
231 * reject that here. It is possible that the object is dying -
232 * but it may also still have the local transport address that
235 if (srx->srx_service) {
240 /* Found a match. We want to replace a dying object.
241 * Attempting to bind the transport socket may still fail if
242 * we're attempting to use a local address that the dying
243 * object is still using.
245 if (!rxrpc_use_local(local, rxrpc_local_use_lookup))
251 local = rxrpc_alloc_local(rxnet, srx);
255 ret = rxrpc_open_socket(local, net);
260 hlist_replace_rcu(cursor, &local->link);
261 cursor->pprev = NULL;
263 hlist_add_head_rcu(&local->link, &rxnet->local_endpoints);
267 mutex_unlock(&rxnet->local_mutex);
268 _leave(" = %p", local);
274 mutex_unlock(&rxnet->local_mutex);
276 call_rcu(&local->rcu, rxrpc_local_rcu);
277 _leave(" = %d", ret);
281 mutex_unlock(&rxnet->local_mutex);
282 _leave(" = -EADDRINUSE");
283 return ERR_PTR(-EADDRINUSE);
287 * Get a ref on a local endpoint.
289 struct rxrpc_local *rxrpc_get_local(struct rxrpc_local *local,
290 enum rxrpc_local_trace why)
294 u = atomic_read(&local->active_users);
295 __refcount_inc(&local->ref, &r);
296 trace_rxrpc_local(local->debug_id, why, r + 1, u);
301 * Get a ref on a local endpoint unless its usage has already reached 0.
303 struct rxrpc_local *rxrpc_get_local_maybe(struct rxrpc_local *local,
304 enum rxrpc_local_trace why)
308 if (local && __refcount_inc_not_zero(&local->ref, &r)) {
309 u = atomic_read(&local->active_users);
310 trace_rxrpc_local(local->debug_id, why, r + 1, u);
318 * Drop a ref on a local endpoint.
320 void rxrpc_put_local(struct rxrpc_local *local, enum rxrpc_local_trace why)
322 unsigned int debug_id;
327 debug_id = local->debug_id;
329 u = atomic_read(&local->active_users);
330 dead = __refcount_dec_and_test(&local->ref, &r);
331 trace_rxrpc_local(debug_id, why, r, u);
334 call_rcu(&local->rcu, rxrpc_local_rcu);
339 * Start using a local endpoint.
341 struct rxrpc_local *rxrpc_use_local(struct rxrpc_local *local,
342 enum rxrpc_local_trace why)
344 local = rxrpc_get_local_maybe(local, rxrpc_local_get_for_use);
348 if (!__rxrpc_use_local(local, why)) {
349 rxrpc_put_local(local, rxrpc_local_put_for_use);
357 * Cease using a local endpoint. Once the number of active users reaches 0, we
358 * start the closure of the transport in the I/O thread..
360 void rxrpc_unuse_local(struct rxrpc_local *local, enum rxrpc_local_trace why)
362 unsigned int debug_id = local->debug_id;
366 r = refcount_read(&local->ref);
367 u = atomic_dec_return(&local->active_users);
368 trace_rxrpc_local(debug_id, why, r, u);
370 kthread_stop(local->io_thread);
375 * Destroy a local endpoint's socket and then hand the record to RCU to dispose
378 * Closing the socket cannot be done from bottom half context or RCU callback
379 * context because it might sleep.
381 void rxrpc_destroy_local(struct rxrpc_local *local)
383 struct socket *socket = local->socket;
384 struct rxrpc_net *rxnet = local->rxnet;
386 _enter("%d", local->debug_id);
390 mutex_lock(&rxnet->local_mutex);
391 hlist_del_init_rcu(&local->link);
392 mutex_unlock(&rxnet->local_mutex);
394 rxrpc_clean_up_local_conns(local);
395 rxrpc_service_connection_reaper(&rxnet->service_conn_reaper);
396 ASSERT(!local->service);
399 local->socket = NULL;
400 kernel_sock_shutdown(socket, SHUT_RDWR);
401 socket->sk->sk_user_data = NULL;
402 sock_release(socket);
405 /* At this point, there should be no more packets coming in to the
408 rxrpc_purge_queue(&local->rx_queue);
412 * Destroy a local endpoint after the RCU grace period expires.
414 static void rxrpc_local_rcu(struct rcu_head *rcu)
416 struct rxrpc_local *local = container_of(rcu, struct rxrpc_local, rcu);
418 rxrpc_see_local(local, rxrpc_local_free);
423 * Verify the local endpoint list is empty by this point.
425 void rxrpc_destroy_all_locals(struct rxrpc_net *rxnet)
427 struct rxrpc_local *local;
431 flush_workqueue(rxrpc_workqueue);
433 if (!hlist_empty(&rxnet->local_endpoints)) {
434 mutex_lock(&rxnet->local_mutex);
435 hlist_for_each_entry(local, &rxnet->local_endpoints, link) {
436 pr_err("AF_RXRPC: Leaked local %p {%d}\n",
437 local, refcount_read(&local->ref));
439 mutex_unlock(&rxnet->local_mutex);