xref: /openbmc/linux/net/rxrpc/local_object.c (revision e53134fe)
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 /*
29  * Compare a local to an address.  Return -ve, 0 or +ve to indicate less than,
30  * same or greater than.
31  *
32  * We explicitly don't compare the RxRPC service ID as we want to reject
33  * conflicting uses by differing services.  Further, we don't want to share
34  * addresses with different options (IPv6), so we don't compare those bits
35  * either.
36  */
37 static long rxrpc_local_cmp_key(const struct rxrpc_local *local,
38 				const struct sockaddr_rxrpc *srx)
39 {
40 	long diff;
41 
42 	diff = ((local->srx.transport_type - srx->transport_type) ?:
43 		(local->srx.transport_len - srx->transport_len) ?:
44 		(local->srx.transport.family - srx->transport.family));
45 	if (diff != 0)
46 		return diff;
47 
48 	switch (srx->transport.family) {
49 	case AF_INET:
50 		/* If the choice of UDP port is left up to the transport, then
51 		 * the endpoint record doesn't match.
52 		 */
53 		return ((u16 __force)local->srx.transport.sin.sin_port -
54 			(u16 __force)srx->transport.sin.sin_port) ?:
55 			memcmp(&local->srx.transport.sin.sin_addr,
56 			       &srx->transport.sin.sin_addr,
57 			       sizeof(struct in_addr));
58 #ifdef CONFIG_AF_RXRPC_IPV6
59 	case AF_INET6:
60 		/* If the choice of UDP6 port is left up to the transport, then
61 		 * the endpoint record doesn't match.
62 		 */
63 		return ((u16 __force)local->srx.transport.sin6.sin6_port -
64 			(u16 __force)srx->transport.sin6.sin6_port) ?:
65 			memcmp(&local->srx.transport.sin6.sin6_addr,
66 			       &srx->transport.sin6.sin6_addr,
67 			       sizeof(struct in6_addr));
68 #endif
69 	default:
70 		BUG();
71 	}
72 }
73 
74 /*
75  * Allocate a new local endpoint.
76  */
77 static struct rxrpc_local *rxrpc_alloc_local(struct rxrpc_net *rxnet,
78 					     const struct sockaddr_rxrpc *srx)
79 {
80 	struct rxrpc_local *local;
81 
82 	local = kzalloc(sizeof(struct rxrpc_local), GFP_KERNEL);
83 	if (local) {
84 		atomic_set(&local->usage, 1);
85 		local->rxnet = rxnet;
86 		INIT_LIST_HEAD(&local->link);
87 		INIT_WORK(&local->processor, rxrpc_local_processor);
88 		init_rwsem(&local->defrag_sem);
89 		skb_queue_head_init(&local->reject_queue);
90 		skb_queue_head_init(&local->event_queue);
91 		local->client_conns = RB_ROOT;
92 		spin_lock_init(&local->client_conns_lock);
93 		spin_lock_init(&local->lock);
94 		rwlock_init(&local->services_lock);
95 		local->debug_id = atomic_inc_return(&rxrpc_debug_id);
96 		memcpy(&local->srx, srx, sizeof(*srx));
97 		local->srx.srx_service = 0;
98 		trace_rxrpc_local(local, rxrpc_local_new, 1, NULL);
99 	}
100 
101 	_leave(" = %p", local);
102 	return local;
103 }
104 
105 /*
106  * create the local socket
107  * - must be called with rxrpc_local_mutex locked
108  */
109 static int rxrpc_open_socket(struct rxrpc_local *local, struct net *net)
110 {
111 	struct sock *sock;
112 	int ret, opt;
113 
114 	_enter("%p{%d,%d}",
115 	       local, local->srx.transport_type, local->srx.transport.family);
116 
117 	/* create a socket to represent the local endpoint */
118 	ret = sock_create_kern(net, local->srx.transport.family,
119 			       local->srx.transport_type, 0, &local->socket);
120 	if (ret < 0) {
121 		_leave(" = %d [socket]", ret);
122 		return ret;
123 	}
124 
125 	/* if a local address was supplied then bind it */
126 	if (local->srx.transport_len > sizeof(sa_family_t)) {
127 		_debug("bind");
128 		ret = kernel_bind(local->socket,
129 				  (struct sockaddr *)&local->srx.transport,
130 				  local->srx.transport_len);
131 		if (ret < 0) {
132 			_debug("bind failed %d", ret);
133 			goto error;
134 		}
135 	}
136 
137 	switch (local->srx.transport.family) {
138 	case AF_INET:
139 		/* we want to receive ICMP errors */
140 		opt = 1;
141 		ret = kernel_setsockopt(local->socket, SOL_IP, IP_RECVERR,
142 					(char *) &opt, sizeof(opt));
143 		if (ret < 0) {
144 			_debug("setsockopt failed");
145 			goto error;
146 		}
147 
148 		/* we want to set the don't fragment bit */
149 		opt = IP_PMTUDISC_DO;
150 		ret = kernel_setsockopt(local->socket, SOL_IP, IP_MTU_DISCOVER,
151 					(char *) &opt, sizeof(opt));
152 		if (ret < 0) {
153 			_debug("setsockopt failed");
154 			goto error;
155 		}
156 		break;
157 
158 	case AF_INET6:
159 		/* we want to receive ICMP errors */
160 		opt = 1;
161 		ret = kernel_setsockopt(local->socket, SOL_IPV6, IPV6_RECVERR,
162 					(char *) &opt, sizeof(opt));
163 		if (ret < 0) {
164 			_debug("setsockopt failed");
165 			goto error;
166 		}
167 
168 		/* we want to set the don't fragment bit */
169 		opt = IPV6_PMTUDISC_DO;
170 		ret = kernel_setsockopt(local->socket, SOL_IPV6, IPV6_MTU_DISCOVER,
171 					(char *) &opt, sizeof(opt));
172 		if (ret < 0) {
173 			_debug("setsockopt failed");
174 			goto error;
175 		}
176 		break;
177 
178 	default:
179 		BUG();
180 	}
181 
182 	/* set the socket up */
183 	sock = local->socket->sk;
184 	sock->sk_user_data	= local;
185 	sock->sk_data_ready	= rxrpc_data_ready;
186 	sock->sk_error_report	= rxrpc_error_report;
187 	_leave(" = 0");
188 	return 0;
189 
190 error:
191 	kernel_sock_shutdown(local->socket, SHUT_RDWR);
192 	local->socket->sk->sk_user_data = NULL;
193 	sock_release(local->socket);
194 	local->socket = NULL;
195 
196 	_leave(" = %d", ret);
197 	return ret;
198 }
199 
200 /*
201  * Look up or create a new local endpoint using the specified local address.
202  */
203 struct rxrpc_local *rxrpc_lookup_local(struct net *net,
204 				       const struct sockaddr_rxrpc *srx)
205 {
206 	struct rxrpc_local *local;
207 	struct rxrpc_net *rxnet = rxrpc_net(net);
208 	struct list_head *cursor;
209 	const char *age;
210 	long diff;
211 	int ret;
212 
213 	_enter("{%d,%d,%pISp}",
214 	       srx->transport_type, srx->transport.family, &srx->transport);
215 
216 	mutex_lock(&rxnet->local_mutex);
217 
218 	for (cursor = rxnet->local_endpoints.next;
219 	     cursor != &rxnet->local_endpoints;
220 	     cursor = cursor->next) {
221 		local = list_entry(cursor, struct rxrpc_local, link);
222 
223 		diff = rxrpc_local_cmp_key(local, srx);
224 		if (diff < 0)
225 			continue;
226 		if (diff > 0)
227 			break;
228 
229 		/* Services aren't allowed to share transport sockets, so
230 		 * reject that here.  It is possible that the object is dying -
231 		 * but it may also still have the local transport address that
232 		 * we want bound.
233 		 */
234 		if (srx->srx_service) {
235 			local = NULL;
236 			goto addr_in_use;
237 		}
238 
239 		/* Found a match.  We replace a dying object.  Attempting to
240 		 * bind the transport socket may still fail if we're attempting
241 		 * to use a local address that the dying object is still using.
242 		 */
243 		if (!rxrpc_get_local_maybe(local)) {
244 			cursor = cursor->next;
245 			list_del_init(&local->link);
246 			break;
247 		}
248 
249 		age = "old";
250 		goto found;
251 	}
252 
253 	local = rxrpc_alloc_local(rxnet, srx);
254 	if (!local)
255 		goto nomem;
256 
257 	ret = rxrpc_open_socket(local, net);
258 	if (ret < 0)
259 		goto sock_error;
260 
261 	list_add_tail(&local->link, cursor);
262 	age = "new";
263 
264 found:
265 	mutex_unlock(&rxnet->local_mutex);
266 
267 	_net("LOCAL %s %d {%pISp}",
268 	     age, local->debug_id, &local->srx.transport);
269 
270 	_leave(" = %p", local);
271 	return local;
272 
273 nomem:
274 	ret = -ENOMEM;
275 sock_error:
276 	mutex_unlock(&rxnet->local_mutex);
277 	kfree(local);
278 	_leave(" = %d", ret);
279 	return ERR_PTR(ret);
280 
281 addr_in_use:
282 	mutex_unlock(&rxnet->local_mutex);
283 	_leave(" = -EADDRINUSE");
284 	return ERR_PTR(-EADDRINUSE);
285 }
286 
287 /*
288  * Get a ref on a local endpoint.
289  */
290 struct rxrpc_local *rxrpc_get_local(struct rxrpc_local *local)
291 {
292 	const void *here = __builtin_return_address(0);
293 	int n;
294 
295 	n = atomic_inc_return(&local->usage);
296 	trace_rxrpc_local(local, rxrpc_local_got, n, here);
297 	return local;
298 }
299 
300 /*
301  * Get a ref on a local endpoint unless its usage has already reached 0.
302  */
303 struct rxrpc_local *rxrpc_get_local_maybe(struct rxrpc_local *local)
304 {
305 	const void *here = __builtin_return_address(0);
306 
307 	if (local) {
308 		int n = atomic_fetch_add_unless(&local->usage, 1, 0);
309 		if (n > 0)
310 			trace_rxrpc_local(local, rxrpc_local_got, n + 1, here);
311 		else
312 			local = NULL;
313 	}
314 	return local;
315 }
316 
317 /*
318  * Queue a local endpoint.
319  */
320 void rxrpc_queue_local(struct rxrpc_local *local)
321 {
322 	const void *here = __builtin_return_address(0);
323 
324 	if (rxrpc_queue_work(&local->processor))
325 		trace_rxrpc_local(local, rxrpc_local_queued,
326 				  atomic_read(&local->usage), here);
327 }
328 
329 /*
330  * A local endpoint reached its end of life.
331  */
332 static void __rxrpc_put_local(struct rxrpc_local *local)
333 {
334 	_enter("%d", local->debug_id);
335 	rxrpc_queue_work(&local->processor);
336 }
337 
338 /*
339  * Drop a ref on a local endpoint.
340  */
341 void rxrpc_put_local(struct rxrpc_local *local)
342 {
343 	const void *here = __builtin_return_address(0);
344 	int n;
345 
346 	if (local) {
347 		n = atomic_dec_return(&local->usage);
348 		trace_rxrpc_local(local, rxrpc_local_put, n, here);
349 
350 		if (n == 0)
351 			__rxrpc_put_local(local);
352 	}
353 }
354 
355 /*
356  * Destroy a local endpoint's socket and then hand the record to RCU to dispose
357  * of.
358  *
359  * Closing the socket cannot be done from bottom half context or RCU callback
360  * context because it might sleep.
361  */
362 static void rxrpc_local_destroyer(struct rxrpc_local *local)
363 {
364 	struct socket *socket = local->socket;
365 	struct rxrpc_net *rxnet = local->rxnet;
366 
367 	_enter("%d", local->debug_id);
368 
369 	/* We can get a race between an incoming call packet queueing the
370 	 * processor again and the work processor starting the destruction
371 	 * process which will shut down the UDP socket.
372 	 */
373 	if (local->dead) {
374 		_leave(" [already dead]");
375 		return;
376 	}
377 	local->dead = true;
378 
379 	mutex_lock(&rxnet->local_mutex);
380 	list_del_init(&local->link);
381 	mutex_unlock(&rxnet->local_mutex);
382 
383 	ASSERT(RB_EMPTY_ROOT(&local->client_conns));
384 	ASSERT(!local->service);
385 
386 	if (socket) {
387 		local->socket = NULL;
388 		kernel_sock_shutdown(socket, SHUT_RDWR);
389 		socket->sk->sk_user_data = NULL;
390 		sock_release(socket);
391 	}
392 
393 	/* At this point, there should be no more packets coming in to the
394 	 * local endpoint.
395 	 */
396 	rxrpc_purge_queue(&local->reject_queue);
397 	rxrpc_purge_queue(&local->event_queue);
398 
399 	_debug("rcu local %d", local->debug_id);
400 	call_rcu(&local->rcu, rxrpc_local_rcu);
401 }
402 
403 /*
404  * Process events on an endpoint
405  */
406 static void rxrpc_local_processor(struct work_struct *work)
407 {
408 	struct rxrpc_local *local =
409 		container_of(work, struct rxrpc_local, processor);
410 	bool again;
411 
412 	trace_rxrpc_local(local, rxrpc_local_processing,
413 			  atomic_read(&local->usage), NULL);
414 
415 	do {
416 		again = false;
417 		if (atomic_read(&local->usage) == 0)
418 			return rxrpc_local_destroyer(local);
419 
420 		if (!skb_queue_empty(&local->reject_queue)) {
421 			rxrpc_reject_packets(local);
422 			again = true;
423 		}
424 
425 		if (!skb_queue_empty(&local->event_queue)) {
426 			rxrpc_process_local_events(local);
427 			again = true;
428 		}
429 	} while (again);
430 }
431 
432 /*
433  * Destroy a local endpoint after the RCU grace period expires.
434  */
435 static void rxrpc_local_rcu(struct rcu_head *rcu)
436 {
437 	struct rxrpc_local *local = container_of(rcu, struct rxrpc_local, rcu);
438 
439 	_enter("%d", local->debug_id);
440 
441 	ASSERT(!work_pending(&local->processor));
442 
443 	_net("DESTROY LOCAL %d", local->debug_id);
444 	kfree(local);
445 	_leave("");
446 }
447 
448 /*
449  * Verify the local endpoint list is empty by this point.
450  */
451 void rxrpc_destroy_all_locals(struct rxrpc_net *rxnet)
452 {
453 	struct rxrpc_local *local;
454 
455 	_enter("");
456 
457 	flush_workqueue(rxrpc_workqueue);
458 
459 	if (!list_empty(&rxnet->local_endpoints)) {
460 		mutex_lock(&rxnet->local_mutex);
461 		list_for_each_entry(local, &rxnet->local_endpoints, link) {
462 			pr_err("AF_RXRPC: Leaked local %p {%d}\n",
463 			       local, atomic_read(&local->usage));
464 		}
465 		mutex_unlock(&rxnet->local_mutex);
466 		BUG();
467 	}
468 }
469