xref: /openbmc/linux/net/rxrpc/local_object.c (revision 7fc96d71)
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.sk_user_data = local;
141 	setup_udp_tunnel_sock(net, local->socket, &tuncfg);
142 
143 	/* set the socket up */
144 	usk = local->socket->sk;
145 	usk->sk_error_report = rxrpc_error_report;
146 
147 	switch (srx->transport.family) {
148 	case AF_INET6:
149 		/* we want to receive ICMPv6 errors */
150 		ip6_sock_set_recverr(usk);
151 
152 		/* Fall through and set IPv4 options too otherwise we don't get
153 		 * errors from IPv4 packets sent through the IPv6 socket.
154 		 */
155 		fallthrough;
156 	case AF_INET:
157 		/* we want to receive ICMP errors */
158 		ip_sock_set_recverr(usk);
159 
160 		/* we want to set the don't fragment bit */
161 		ip_sock_set_mtu_discover(usk, IP_PMTUDISC_DO);
162 
163 		/* We want receive timestamps. */
164 		sock_enable_timestamps(usk);
165 		break;
166 
167 	default:
168 		BUG();
169 	}
170 
171 	_leave(" = 0");
172 	return 0;
173 }
174 
175 /*
176  * Look up or create a new local endpoint using the specified local address.
177  */
178 struct rxrpc_local *rxrpc_lookup_local(struct net *net,
179 				       const struct sockaddr_rxrpc *srx)
180 {
181 	struct rxrpc_local *local;
182 	struct rxrpc_net *rxnet = rxrpc_net(net);
183 	struct hlist_node *cursor;
184 	const char *age;
185 	long diff;
186 	int ret;
187 
188 	_enter("{%d,%d,%pISp}",
189 	       srx->transport_type, srx->transport.family, &srx->transport);
190 
191 	mutex_lock(&rxnet->local_mutex);
192 
193 	hlist_for_each(cursor, &rxnet->local_endpoints) {
194 		local = hlist_entry(cursor, struct rxrpc_local, link);
195 
196 		diff = rxrpc_local_cmp_key(local, srx);
197 		if (diff != 0)
198 			continue;
199 
200 		/* Services aren't allowed to share transport sockets, so
201 		 * reject that here.  It is possible that the object is dying -
202 		 * but it may also still have the local transport address that
203 		 * we want bound.
204 		 */
205 		if (srx->srx_service) {
206 			local = NULL;
207 			goto addr_in_use;
208 		}
209 
210 		/* Found a match.  We want to replace a dying object.
211 		 * Attempting to bind the transport socket may still fail if
212 		 * we're attempting to use a local address that the dying
213 		 * object is still using.
214 		 */
215 		if (!rxrpc_use_local(local))
216 			break;
217 
218 		age = "old";
219 		goto found;
220 	}
221 
222 	local = rxrpc_alloc_local(rxnet, srx);
223 	if (!local)
224 		goto nomem;
225 
226 	ret = rxrpc_open_socket(local, net);
227 	if (ret < 0)
228 		goto sock_error;
229 
230 	if (cursor) {
231 		hlist_replace_rcu(cursor, &local->link);
232 		cursor->pprev = NULL;
233 	} else {
234 		hlist_add_head_rcu(&local->link, &rxnet->local_endpoints);
235 	}
236 	age = "new";
237 
238 found:
239 	mutex_unlock(&rxnet->local_mutex);
240 
241 	_net("LOCAL %s %d {%pISp}",
242 	     age, local->debug_id, &local->srx.transport);
243 
244 	_leave(" = %p", local);
245 	return local;
246 
247 nomem:
248 	ret = -ENOMEM;
249 sock_error:
250 	mutex_unlock(&rxnet->local_mutex);
251 	if (local)
252 		call_rcu(&local->rcu, rxrpc_local_rcu);
253 	_leave(" = %d", ret);
254 	return ERR_PTR(ret);
255 
256 addr_in_use:
257 	mutex_unlock(&rxnet->local_mutex);
258 	_leave(" = -EADDRINUSE");
259 	return ERR_PTR(-EADDRINUSE);
260 }
261 
262 /*
263  * Get a ref on a local endpoint.
264  */
265 struct rxrpc_local *rxrpc_get_local(struct rxrpc_local *local)
266 {
267 	const void *here = __builtin_return_address(0);
268 	int r;
269 
270 	__refcount_inc(&local->ref, &r);
271 	trace_rxrpc_local(local->debug_id, rxrpc_local_got, r + 1, here);
272 	return local;
273 }
274 
275 /*
276  * Get a ref on a local endpoint unless its usage has already reached 0.
277  */
278 struct rxrpc_local *rxrpc_get_local_maybe(struct rxrpc_local *local)
279 {
280 	const void *here = __builtin_return_address(0);
281 	int r;
282 
283 	if (local) {
284 		if (__refcount_inc_not_zero(&local->ref, &r))
285 			trace_rxrpc_local(local->debug_id, rxrpc_local_got,
286 					  r + 1, here);
287 		else
288 			local = NULL;
289 	}
290 	return local;
291 }
292 
293 /*
294  * Queue a local endpoint and pass the caller's reference to the work item.
295  */
296 void rxrpc_queue_local(struct rxrpc_local *local)
297 {
298 	const void *here = __builtin_return_address(0);
299 	unsigned int debug_id = local->debug_id;
300 	int r = refcount_read(&local->ref);
301 
302 	if (rxrpc_queue_work(&local->processor))
303 		trace_rxrpc_local(debug_id, rxrpc_local_queued, r + 1, here);
304 	else
305 		rxrpc_put_local(local);
306 }
307 
308 /*
309  * Drop a ref on a local endpoint.
310  */
311 void rxrpc_put_local(struct rxrpc_local *local)
312 {
313 	const void *here = __builtin_return_address(0);
314 	unsigned int debug_id;
315 	bool dead;
316 	int r;
317 
318 	if (local) {
319 		debug_id = local->debug_id;
320 
321 		dead = __refcount_dec_and_test(&local->ref, &r);
322 		trace_rxrpc_local(debug_id, rxrpc_local_put, r, here);
323 
324 		if (dead)
325 			call_rcu(&local->rcu, rxrpc_local_rcu);
326 	}
327 }
328 
329 /*
330  * Start using a local endpoint.
331  */
332 struct rxrpc_local *rxrpc_use_local(struct rxrpc_local *local)
333 {
334 	local = rxrpc_get_local_maybe(local);
335 	if (!local)
336 		return NULL;
337 
338 	if (!__rxrpc_use_local(local)) {
339 		rxrpc_put_local(local);
340 		return NULL;
341 	}
342 
343 	return local;
344 }
345 
346 /*
347  * Cease using a local endpoint.  Once the number of active users reaches 0, we
348  * start the closure of the transport in the work processor.
349  */
350 void rxrpc_unuse_local(struct rxrpc_local *local)
351 {
352 	if (local) {
353 		if (__rxrpc_unuse_local(local)) {
354 			rxrpc_get_local(local);
355 			rxrpc_queue_local(local);
356 		}
357 	}
358 }
359 
360 /*
361  * Destroy a local endpoint's socket and then hand the record to RCU to dispose
362  * of.
363  *
364  * Closing the socket cannot be done from bottom half context or RCU callback
365  * context because it might sleep.
366  */
367 static void rxrpc_local_destroyer(struct rxrpc_local *local)
368 {
369 	struct socket *socket = local->socket;
370 	struct rxrpc_net *rxnet = local->rxnet;
371 
372 	_enter("%d", local->debug_id);
373 
374 	local->dead = true;
375 
376 	mutex_lock(&rxnet->local_mutex);
377 	hlist_del_init_rcu(&local->link);
378 	mutex_unlock(&rxnet->local_mutex);
379 
380 	rxrpc_clean_up_local_conns(local);
381 	rxrpc_service_connection_reaper(&rxnet->service_conn_reaper);
382 	ASSERT(!local->service);
383 
384 	if (socket) {
385 		local->socket = NULL;
386 		kernel_sock_shutdown(socket, SHUT_RDWR);
387 		socket->sk->sk_user_data = NULL;
388 		sock_release(socket);
389 	}
390 
391 	/* At this point, there should be no more packets coming in to the
392 	 * local endpoint.
393 	 */
394 	rxrpc_purge_queue(&local->reject_queue);
395 	rxrpc_purge_queue(&local->event_queue);
396 }
397 
398 /*
399  * Process events on an endpoint.  The work item carries a ref which
400  * we must release.
401  */
402 static void rxrpc_local_processor(struct work_struct *work)
403 {
404 	struct rxrpc_local *local =
405 		container_of(work, struct rxrpc_local, processor);
406 	bool again;
407 
408 	trace_rxrpc_local(local->debug_id, rxrpc_local_processing,
409 			  refcount_read(&local->ref), NULL);
410 
411 	do {
412 		again = false;
413 		if (!__rxrpc_use_local(local)) {
414 			rxrpc_local_destroyer(local);
415 			break;
416 		}
417 
418 		if (!skb_queue_empty(&local->reject_queue)) {
419 			rxrpc_reject_packets(local);
420 			again = true;
421 		}
422 
423 		if (!skb_queue_empty(&local->event_queue)) {
424 			rxrpc_process_local_events(local);
425 			again = true;
426 		}
427 
428 		__rxrpc_unuse_local(local);
429 	} while (again);
430 
431 	rxrpc_put_local(local);
432 }
433 
434 /*
435  * Destroy a local endpoint after the RCU grace period expires.
436  */
437 static void rxrpc_local_rcu(struct rcu_head *rcu)
438 {
439 	struct rxrpc_local *local = container_of(rcu, struct rxrpc_local, rcu);
440 
441 	_enter("%d", local->debug_id);
442 
443 	ASSERT(!work_pending(&local->processor));
444 
445 	_net("DESTROY LOCAL %d", local->debug_id);
446 	kfree(local);
447 	_leave("");
448 }
449 
450 /*
451  * Verify the local endpoint list is empty by this point.
452  */
453 void rxrpc_destroy_all_locals(struct rxrpc_net *rxnet)
454 {
455 	struct rxrpc_local *local;
456 
457 	_enter("");
458 
459 	flush_workqueue(rxrpc_workqueue);
460 
461 	if (!hlist_empty(&rxnet->local_endpoints)) {
462 		mutex_lock(&rxnet->local_mutex);
463 		hlist_for_each_entry(local, &rxnet->local_endpoints, link) {
464 			pr_err("AF_RXRPC: Leaked local %p {%d}\n",
465 			       local, refcount_read(&local->ref));
466 		}
467 		mutex_unlock(&rxnet->local_mutex);
468 		BUG();
469 	}
470 }
471