xref: /openbmc/linux/net/rxrpc/local_object.c (revision c4a7b9b5)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* Local endpoint object management
3  *
4  * Copyright (C) 2016 Red Hat, Inc. All Rights Reserved.
5  * Written by David Howells (dhowells@redhat.com)
6  */
7 
8 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
9 
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>
15 #include <linux/ip.h>
16 #include <linux/hashtable.h>
17 #include <net/sock.h>
18 #include <net/udp.h>
19 #include <net/udp_tunnel.h>
20 #include <net/af_rxrpc.h>
21 #include "ar-internal.h"
22 
23 static void rxrpc_local_processor(struct work_struct *);
24 static void rxrpc_local_rcu(struct rcu_head *);
25 
26 /*
27  * Compare a local to an address.  Return -ve, 0 or +ve to indicate less than,
28  * same or greater than.
29  *
30  * We explicitly don't compare the RxRPC service ID as we want to reject
31  * conflicting uses by differing services.  Further, we don't want to share
32  * addresses with different options (IPv6), so we don't compare those bits
33  * either.
34  */
35 static long rxrpc_local_cmp_key(const struct rxrpc_local *local,
36 				const struct sockaddr_rxrpc *srx)
37 {
38 	long diff;
39 
40 	diff = ((local->srx.transport_type - srx->transport_type) ?:
41 		(local->srx.transport_len - srx->transport_len) ?:
42 		(local->srx.transport.family - srx->transport.family));
43 	if (diff != 0)
44 		return diff;
45 
46 	switch (srx->transport.family) {
47 	case AF_INET:
48 		/* If the choice of UDP port is left up to the transport, then
49 		 * the endpoint record doesn't match.
50 		 */
51 		return ((u16 __force)local->srx.transport.sin.sin_port -
52 			(u16 __force)srx->transport.sin.sin_port) ?:
53 			memcmp(&local->srx.transport.sin.sin_addr,
54 			       &srx->transport.sin.sin_addr,
55 			       sizeof(struct in_addr));
56 #ifdef CONFIG_AF_RXRPC_IPV6
57 	case AF_INET6:
58 		/* If the choice of UDP6 port is left up to the transport, then
59 		 * the endpoint record doesn't match.
60 		 */
61 		return ((u16 __force)local->srx.transport.sin6.sin6_port -
62 			(u16 __force)srx->transport.sin6.sin6_port) ?:
63 			memcmp(&local->srx.transport.sin6.sin6_addr,
64 			       &srx->transport.sin6.sin6_addr,
65 			       sizeof(struct in6_addr));
66 #endif
67 	default:
68 		BUG();
69 	}
70 }
71 
72 /*
73  * Allocate a new local endpoint.
74  */
75 static struct rxrpc_local *rxrpc_alloc_local(struct rxrpc_net *rxnet,
76 					     const struct sockaddr_rxrpc *srx)
77 {
78 	struct rxrpc_local *local;
79 
80 	local = kzalloc(sizeof(struct rxrpc_local), GFP_KERNEL);
81 	if (local) {
82 		refcount_set(&local->ref, 1);
83 		atomic_set(&local->active_users, 1);
84 		local->rxnet = rxnet;
85 		INIT_HLIST_NODE(&local->link);
86 		INIT_WORK(&local->processor, rxrpc_local_processor);
87 		init_rwsem(&local->defrag_sem);
88 		skb_queue_head_init(&local->reject_queue);
89 		skb_queue_head_init(&local->event_queue);
90 		local->client_bundles = RB_ROOT;
91 		spin_lock_init(&local->client_bundles_lock);
92 		spin_lock_init(&local->lock);
93 		rwlock_init(&local->services_lock);
94 		local->debug_id = atomic_inc_return(&rxrpc_debug_id);
95 		memcpy(&local->srx, srx, sizeof(*srx));
96 		local->srx.srx_service = 0;
97 		trace_rxrpc_local(local->debug_id, rxrpc_local_new, 1, NULL);
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 udp_tunnel_sock_cfg tuncfg = {NULL};
111 	struct sockaddr_rxrpc *srx = &local->srx;
112 	struct udp_port_cfg udp_conf = {0};
113 	struct sock *usk;
114 	int ret;
115 
116 	_enter("%p{%d,%d}",
117 	       local, srx->transport_type, srx->transport.family);
118 
119 	udp_conf.family = srx->transport.family;
120 	udp_conf.use_udp_checksums = true;
121 	if (udp_conf.family == AF_INET) {
122 		udp_conf.local_ip = srx->transport.sin.sin_addr;
123 		udp_conf.local_udp_port = srx->transport.sin.sin_port;
124 #if IS_ENABLED(CONFIG_AF_RXRPC_IPV6)
125 	} else {
126 		udp_conf.local_ip6 = srx->transport.sin6.sin6_addr;
127 		udp_conf.local_udp_port = srx->transport.sin6.sin6_port;
128 		udp_conf.use_udp6_tx_checksums = true;
129 		udp_conf.use_udp6_rx_checksums = true;
130 #endif
131 	}
132 	ret = udp_sock_create(net, &udp_conf, &local->socket);
133 	if (ret < 0) {
134 		_leave(" = %d [socket]", ret);
135 		return ret;
136 	}
137 
138 	tuncfg.encap_type = UDP_ENCAP_RXRPC;
139 	tuncfg.encap_rcv = rxrpc_input_packet;
140 	tuncfg.encap_err_rcv = rxrpc_encap_err_rcv;
141 	tuncfg.sk_user_data = local;
142 	setup_udp_tunnel_sock(net, local->socket, &tuncfg);
143 
144 	/* set the socket up */
145 	usk = local->socket->sk;
146 	usk->sk_error_report = rxrpc_error_report;
147 
148 	switch (srx->transport.family) {
149 	case AF_INET6:
150 		/* we want to receive ICMPv6 errors */
151 		ip6_sock_set_recverr(usk);
152 
153 		/* Fall through and set IPv4 options too otherwise we don't get
154 		 * errors from IPv4 packets sent through the IPv6 socket.
155 		 */
156 		fallthrough;
157 	case AF_INET:
158 		/* we want to receive ICMP errors */
159 		ip_sock_set_recverr(usk);
160 
161 		/* we want to set the don't fragment bit */
162 		ip_sock_set_mtu_discover(usk, IP_PMTUDISC_DO);
163 
164 		/* We want receive timestamps. */
165 		sock_enable_timestamps(usk);
166 		break;
167 
168 	default:
169 		BUG();
170 	}
171 
172 	_leave(" = 0");
173 	return 0;
174 }
175 
176 /*
177  * Look up or create a new local endpoint using the specified local address.
178  */
179 struct rxrpc_local *rxrpc_lookup_local(struct net *net,
180 				       const struct sockaddr_rxrpc *srx)
181 {
182 	struct rxrpc_local *local;
183 	struct rxrpc_net *rxnet = rxrpc_net(net);
184 	struct hlist_node *cursor;
185 	const char *age;
186 	long diff;
187 	int ret;
188 
189 	_enter("{%d,%d,%pISp}",
190 	       srx->transport_type, srx->transport.family, &srx->transport);
191 
192 	mutex_lock(&rxnet->local_mutex);
193 
194 	hlist_for_each(cursor, &rxnet->local_endpoints) {
195 		local = hlist_entry(cursor, struct rxrpc_local, link);
196 
197 		diff = rxrpc_local_cmp_key(local, srx);
198 		if (diff != 0)
199 			continue;
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 want to replace a dying object.
212 		 * Attempting to bind the transport socket may still fail if
213 		 * we're attempting to use a local address that the dying
214 		 * object is still using.
215 		 */
216 		if (!rxrpc_use_local(local))
217 			break;
218 
219 		age = "old";
220 		goto found;
221 	}
222 
223 	local = rxrpc_alloc_local(rxnet, srx);
224 	if (!local)
225 		goto nomem;
226 
227 	ret = rxrpc_open_socket(local, net);
228 	if (ret < 0)
229 		goto sock_error;
230 
231 	if (cursor) {
232 		hlist_replace_rcu(cursor, &local->link);
233 		cursor->pprev = NULL;
234 	} else {
235 		hlist_add_head_rcu(&local->link, &rxnet->local_endpoints);
236 	}
237 	age = "new";
238 
239 found:
240 	mutex_unlock(&rxnet->local_mutex);
241 
242 	_net("LOCAL %s %d {%pISp}",
243 	     age, local->debug_id, &local->srx.transport);
244 
245 	_leave(" = %p", local);
246 	return local;
247 
248 nomem:
249 	ret = -ENOMEM;
250 sock_error:
251 	mutex_unlock(&rxnet->local_mutex);
252 	if (local)
253 		call_rcu(&local->rcu, rxrpc_local_rcu);
254 	_leave(" = %d", ret);
255 	return ERR_PTR(ret);
256 
257 addr_in_use:
258 	mutex_unlock(&rxnet->local_mutex);
259 	_leave(" = -EADDRINUSE");
260 	return ERR_PTR(-EADDRINUSE);
261 }
262 
263 /*
264  * Get a ref on a local endpoint.
265  */
266 struct rxrpc_local *rxrpc_get_local(struct rxrpc_local *local)
267 {
268 	const void *here = __builtin_return_address(0);
269 	int r;
270 
271 	__refcount_inc(&local->ref, &r);
272 	trace_rxrpc_local(local->debug_id, rxrpc_local_got, r + 1, here);
273 	return local;
274 }
275 
276 /*
277  * Get a ref on a local endpoint unless its usage has already reached 0.
278  */
279 struct rxrpc_local *rxrpc_get_local_maybe(struct rxrpc_local *local)
280 {
281 	const void *here = __builtin_return_address(0);
282 	int r;
283 
284 	if (local) {
285 		if (__refcount_inc_not_zero(&local->ref, &r))
286 			trace_rxrpc_local(local->debug_id, rxrpc_local_got,
287 					  r + 1, here);
288 		else
289 			local = NULL;
290 	}
291 	return local;
292 }
293 
294 /*
295  * Queue a local endpoint and pass the caller's reference to the work item.
296  */
297 void rxrpc_queue_local(struct rxrpc_local *local)
298 {
299 	const void *here = __builtin_return_address(0);
300 	unsigned int debug_id = local->debug_id;
301 	int r = refcount_read(&local->ref);
302 
303 	if (rxrpc_queue_work(&local->processor))
304 		trace_rxrpc_local(debug_id, rxrpc_local_queued, r + 1, here);
305 	else
306 		rxrpc_put_local(local);
307 }
308 
309 /*
310  * Drop a ref on a local endpoint.
311  */
312 void rxrpc_put_local(struct rxrpc_local *local)
313 {
314 	const void *here = __builtin_return_address(0);
315 	unsigned int debug_id;
316 	bool dead;
317 	int r;
318 
319 	if (local) {
320 		debug_id = local->debug_id;
321 
322 		dead = __refcount_dec_and_test(&local->ref, &r);
323 		trace_rxrpc_local(debug_id, rxrpc_local_put, r, here);
324 
325 		if (dead)
326 			call_rcu(&local->rcu, rxrpc_local_rcu);
327 	}
328 }
329 
330 /*
331  * Start using a local endpoint.
332  */
333 struct rxrpc_local *rxrpc_use_local(struct rxrpc_local *local)
334 {
335 	local = rxrpc_get_local_maybe(local);
336 	if (!local)
337 		return NULL;
338 
339 	if (!__rxrpc_use_local(local)) {
340 		rxrpc_put_local(local);
341 		return NULL;
342 	}
343 
344 	return local;
345 }
346 
347 /*
348  * Cease using a local endpoint.  Once the number of active users reaches 0, we
349  * start the closure of the transport in the work processor.
350  */
351 void rxrpc_unuse_local(struct rxrpc_local *local)
352 {
353 	if (local) {
354 		if (__rxrpc_unuse_local(local)) {
355 			rxrpc_get_local(local);
356 			rxrpc_queue_local(local);
357 		}
358 	}
359 }
360 
361 /*
362  * Destroy a local endpoint's socket and then hand the record to RCU to dispose
363  * of.
364  *
365  * Closing the socket cannot be done from bottom half context or RCU callback
366  * context because it might sleep.
367  */
368 static void rxrpc_local_destroyer(struct rxrpc_local *local)
369 {
370 	struct socket *socket = local->socket;
371 	struct rxrpc_net *rxnet = local->rxnet;
372 
373 	_enter("%d", local->debug_id);
374 
375 	local->dead = true;
376 
377 	mutex_lock(&rxnet->local_mutex);
378 	hlist_del_init_rcu(&local->link);
379 	mutex_unlock(&rxnet->local_mutex);
380 
381 	rxrpc_clean_up_local_conns(local);
382 	rxrpc_service_connection_reaper(&rxnet->service_conn_reaper);
383 	ASSERT(!local->service);
384 
385 	if (socket) {
386 		local->socket = NULL;
387 		kernel_sock_shutdown(socket, SHUT_RDWR);
388 		socket->sk->sk_user_data = NULL;
389 		sock_release(socket);
390 	}
391 
392 	/* At this point, there should be no more packets coming in to the
393 	 * local endpoint.
394 	 */
395 	rxrpc_purge_queue(&local->reject_queue);
396 	rxrpc_purge_queue(&local->event_queue);
397 }
398 
399 /*
400  * Process events on an endpoint.  The work item carries a ref which
401  * we must release.
402  */
403 static void rxrpc_local_processor(struct work_struct *work)
404 {
405 	struct rxrpc_local *local =
406 		container_of(work, struct rxrpc_local, processor);
407 	bool again;
408 
409 	if (local->dead)
410 		return;
411 
412 	trace_rxrpc_local(local->debug_id, rxrpc_local_processing,
413 			  refcount_read(&local->ref), NULL);
414 
415 	do {
416 		again = false;
417 		if (!__rxrpc_use_local(local)) {
418 			rxrpc_local_destroyer(local);
419 			break;
420 		}
421 
422 		if (!skb_queue_empty(&local->reject_queue)) {
423 			rxrpc_reject_packets(local);
424 			again = true;
425 		}
426 
427 		if (!skb_queue_empty(&local->event_queue)) {
428 			rxrpc_process_local_events(local);
429 			again = true;
430 		}
431 
432 		__rxrpc_unuse_local(local);
433 	} while (again);
434 
435 	rxrpc_put_local(local);
436 }
437 
438 /*
439  * Destroy a local endpoint after the RCU grace period expires.
440  */
441 static void rxrpc_local_rcu(struct rcu_head *rcu)
442 {
443 	struct rxrpc_local *local = container_of(rcu, struct rxrpc_local, rcu);
444 
445 	_enter("%d", local->debug_id);
446 
447 	ASSERT(!work_pending(&local->processor));
448 
449 	_net("DESTROY LOCAL %d", local->debug_id);
450 	kfree(local);
451 	_leave("");
452 }
453 
454 /*
455  * Verify the local endpoint list is empty by this point.
456  */
457 void rxrpc_destroy_all_locals(struct rxrpc_net *rxnet)
458 {
459 	struct rxrpc_local *local;
460 
461 	_enter("");
462 
463 	flush_workqueue(rxrpc_workqueue);
464 
465 	if (!hlist_empty(&rxnet->local_endpoints)) {
466 		mutex_lock(&rxnet->local_mutex);
467 		hlist_for_each_entry(local, &rxnet->local_endpoints, link) {
468 			pr_err("AF_RXRPC: Leaked local %p {%d}\n",
469 			       local, refcount_read(&local->ref));
470 		}
471 		mutex_unlock(&rxnet->local_mutex);
472 		BUG();
473 	}
474 }
475