xref: /openbmc/linux/net/rxrpc/local_object.c (revision bc5aa3a0)
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 static DEFINE_MUTEX(rxrpc_local_mutex);
29 static LIST_HEAD(rxrpc_local_endpoints);
30 
31 /*
32  * Compare a local to an address.  Return -ve, 0 or +ve to indicate less than,
33  * same or greater than.
34  *
35  * We explicitly don't compare the RxRPC service ID as we want to reject
36  * conflicting uses by differing services.  Further, we don't want to share
37  * addresses with different options (IPv6), so we don't compare those bits
38  * either.
39  */
40 static long rxrpc_local_cmp_key(const struct rxrpc_local *local,
41 				const struct sockaddr_rxrpc *srx)
42 {
43 	long diff;
44 
45 	diff = ((local->srx.transport_type - srx->transport_type) ?:
46 		(local->srx.transport_len - srx->transport_len) ?:
47 		(local->srx.transport.family - srx->transport.family));
48 	if (diff != 0)
49 		return diff;
50 
51 	switch (srx->transport.family) {
52 	case AF_INET:
53 		/* If the choice of UDP port is left up to the transport, then
54 		 * the endpoint record doesn't match.
55 		 */
56 		return ((u16 __force)local->srx.transport.sin.sin_port -
57 			(u16 __force)srx->transport.sin.sin_port) ?:
58 			memcmp(&local->srx.transport.sin.sin_addr,
59 			       &srx->transport.sin.sin_addr,
60 			       sizeof(struct in_addr));
61 	default:
62 		BUG();
63 	}
64 }
65 
66 /*
67  * Allocate a new local endpoint.
68  */
69 static struct rxrpc_local *rxrpc_alloc_local(const struct sockaddr_rxrpc *srx)
70 {
71 	struct rxrpc_local *local;
72 
73 	local = kzalloc(sizeof(struct rxrpc_local), GFP_KERNEL);
74 	if (local) {
75 		atomic_set(&local->usage, 1);
76 		INIT_LIST_HEAD(&local->link);
77 		INIT_WORK(&local->processor, rxrpc_local_processor);
78 		INIT_LIST_HEAD(&local->services);
79 		init_rwsem(&local->defrag_sem);
80 		skb_queue_head_init(&local->accept_queue);
81 		skb_queue_head_init(&local->reject_queue);
82 		skb_queue_head_init(&local->event_queue);
83 		local->client_conns = RB_ROOT;
84 		spin_lock_init(&local->client_conns_lock);
85 		spin_lock_init(&local->lock);
86 		rwlock_init(&local->services_lock);
87 		local->debug_id = atomic_inc_return(&rxrpc_debug_id);
88 		memcpy(&local->srx, srx, sizeof(*srx));
89 	}
90 
91 	_leave(" = %p", local);
92 	return local;
93 }
94 
95 /*
96  * create the local socket
97  * - must be called with rxrpc_local_mutex locked
98  */
99 static int rxrpc_open_socket(struct rxrpc_local *local)
100 {
101 	struct sock *sock;
102 	int ret, opt;
103 
104 	_enter("%p{%d}", local, local->srx.transport_type);
105 
106 	/* create a socket to represent the local endpoint */
107 	ret = sock_create_kern(&init_net, PF_INET, local->srx.transport_type,
108 			       IPPROTO_UDP, &local->socket);
109 	if (ret < 0) {
110 		_leave(" = %d [socket]", ret);
111 		return ret;
112 	}
113 
114 	/* if a local address was supplied then bind it */
115 	if (local->srx.transport_len > sizeof(sa_family_t)) {
116 		_debug("bind");
117 		ret = kernel_bind(local->socket,
118 				  (struct sockaddr *)&local->srx.transport,
119 				  local->srx.transport_len);
120 		if (ret < 0) {
121 			_debug("bind failed %d", ret);
122 			goto error;
123 		}
124 	}
125 
126 	/* we want to receive ICMP errors */
127 	opt = 1;
128 	ret = kernel_setsockopt(local->socket, SOL_IP, IP_RECVERR,
129 				(char *) &opt, sizeof(opt));
130 	if (ret < 0) {
131 		_debug("setsockopt failed");
132 		goto error;
133 	}
134 
135 	/* we want to set the don't fragment bit */
136 	opt = IP_PMTUDISC_DO;
137 	ret = kernel_setsockopt(local->socket, SOL_IP, IP_MTU_DISCOVER,
138 				(char *) &opt, sizeof(opt));
139 	if (ret < 0) {
140 		_debug("setsockopt failed");
141 		goto error;
142 	}
143 
144 	/* set the socket up */
145 	sock = local->socket->sk;
146 	sock->sk_user_data	= local;
147 	sock->sk_data_ready	= rxrpc_data_ready;
148 	sock->sk_error_report	= rxrpc_error_report;
149 	_leave(" = 0");
150 	return 0;
151 
152 error:
153 	kernel_sock_shutdown(local->socket, SHUT_RDWR);
154 	local->socket->sk->sk_user_data = NULL;
155 	sock_release(local->socket);
156 	local->socket = NULL;
157 
158 	_leave(" = %d", ret);
159 	return ret;
160 }
161 
162 /*
163  * Look up or create a new local endpoint using the specified local address.
164  */
165 struct rxrpc_local *rxrpc_lookup_local(const struct sockaddr_rxrpc *srx)
166 {
167 	struct rxrpc_local *local;
168 	struct list_head *cursor;
169 	const char *age;
170 	long diff;
171 	int ret;
172 
173 	if (srx->transport.family == AF_INET) {
174 		_enter("{%d,%u,%pI4+%hu}",
175 		       srx->transport_type,
176 		       srx->transport.family,
177 		       &srx->transport.sin.sin_addr,
178 		       ntohs(srx->transport.sin.sin_port));
179 	} else {
180 		_enter("{%d,%u}",
181 		       srx->transport_type,
182 		       srx->transport.family);
183 		return ERR_PTR(-EAFNOSUPPORT);
184 	}
185 
186 	mutex_lock(&rxrpc_local_mutex);
187 
188 	for (cursor = rxrpc_local_endpoints.next;
189 	     cursor != &rxrpc_local_endpoints;
190 	     cursor = cursor->next) {
191 		local = list_entry(cursor, struct rxrpc_local, link);
192 
193 		diff = rxrpc_local_cmp_key(local, srx);
194 		if (diff < 0)
195 			continue;
196 		if (diff > 0)
197 			break;
198 
199 		/* Services aren't allowed to share transport sockets, so
200 		 * reject that here.  It is possible that the object is dying -
201 		 * but it may also still have the local transport address that
202 		 * we want bound.
203 		 */
204 		if (srx->srx_service) {
205 			local = NULL;
206 			goto addr_in_use;
207 		}
208 
209 		/* Found a match.  We replace a dying object.  Attempting to
210 		 * bind the transport socket may still fail if we're attempting
211 		 * to use a local address that the dying object is still using.
212 		 */
213 		if (!rxrpc_get_local_maybe(local)) {
214 			cursor = cursor->next;
215 			list_del_init(&local->link);
216 			break;
217 		}
218 
219 		age = "old";
220 		goto found;
221 	}
222 
223 	local = rxrpc_alloc_local(srx);
224 	if (!local)
225 		goto nomem;
226 
227 	ret = rxrpc_open_socket(local);
228 	if (ret < 0)
229 		goto sock_error;
230 
231 	list_add_tail(&local->link, cursor);
232 	age = "new";
233 
234 found:
235 	mutex_unlock(&rxrpc_local_mutex);
236 
237 	_net("LOCAL %s %d {%d,%u,%pI4+%hu}",
238 	     age,
239 	     local->debug_id,
240 	     local->srx.transport_type,
241 	     local->srx.transport.family,
242 	     &local->srx.transport.sin.sin_addr,
243 	     ntohs(local->srx.transport.sin.sin_port));
244 
245 	_leave(" = %p", local);
246 	return local;
247 
248 nomem:
249 	ret = -ENOMEM;
250 sock_error:
251 	mutex_unlock(&rxrpc_local_mutex);
252 	kfree(local);
253 	_leave(" = %d", ret);
254 	return ERR_PTR(ret);
255 
256 addr_in_use:
257 	mutex_unlock(&rxrpc_local_mutex);
258 	_leave(" = -EADDRINUSE");
259 	return ERR_PTR(-EADDRINUSE);
260 }
261 
262 /*
263  * A local endpoint reached its end of life.
264  */
265 void __rxrpc_put_local(struct rxrpc_local *local)
266 {
267 	_enter("%d", local->debug_id);
268 	rxrpc_queue_work(&local->processor);
269 }
270 
271 /*
272  * Destroy a local endpoint's socket and then hand the record to RCU to dispose
273  * of.
274  *
275  * Closing the socket cannot be done from bottom half context or RCU callback
276  * context because it might sleep.
277  */
278 static void rxrpc_local_destroyer(struct rxrpc_local *local)
279 {
280 	struct socket *socket = local->socket;
281 
282 	_enter("%d", local->debug_id);
283 
284 	/* We can get a race between an incoming call packet queueing the
285 	 * processor again and the work processor starting the destruction
286 	 * process which will shut down the UDP socket.
287 	 */
288 	if (local->dead) {
289 		_leave(" [already dead]");
290 		return;
291 	}
292 	local->dead = true;
293 
294 	mutex_lock(&rxrpc_local_mutex);
295 	list_del_init(&local->link);
296 	mutex_unlock(&rxrpc_local_mutex);
297 
298 	ASSERT(RB_EMPTY_ROOT(&local->client_conns));
299 	ASSERT(list_empty(&local->services));
300 
301 	if (socket) {
302 		local->socket = NULL;
303 		kernel_sock_shutdown(socket, SHUT_RDWR);
304 		socket->sk->sk_user_data = NULL;
305 		sock_release(socket);
306 	}
307 
308 	/* At this point, there should be no more packets coming in to the
309 	 * local endpoint.
310 	 */
311 	rxrpc_purge_queue(&local->accept_queue);
312 	rxrpc_purge_queue(&local->reject_queue);
313 	rxrpc_purge_queue(&local->event_queue);
314 
315 	_debug("rcu local %d", local->debug_id);
316 	call_rcu(&local->rcu, rxrpc_local_rcu);
317 }
318 
319 /*
320  * Process events on an endpoint
321  */
322 static void rxrpc_local_processor(struct work_struct *work)
323 {
324 	struct rxrpc_local *local =
325 		container_of(work, struct rxrpc_local, processor);
326 	bool again;
327 
328 	_enter("%d", local->debug_id);
329 
330 	do {
331 		again = false;
332 		if (atomic_read(&local->usage) == 0)
333 			return rxrpc_local_destroyer(local);
334 
335 		if (!skb_queue_empty(&local->accept_queue)) {
336 			rxrpc_accept_incoming_calls(local);
337 			again = true;
338 		}
339 
340 		if (!skb_queue_empty(&local->reject_queue)) {
341 			rxrpc_reject_packets(local);
342 			again = true;
343 		}
344 
345 		if (!skb_queue_empty(&local->event_queue)) {
346 			rxrpc_process_local_events(local);
347 			again = true;
348 		}
349 	} while (again);
350 }
351 
352 /*
353  * Destroy a local endpoint after the RCU grace period expires.
354  */
355 static void rxrpc_local_rcu(struct rcu_head *rcu)
356 {
357 	struct rxrpc_local *local = container_of(rcu, struct rxrpc_local, rcu);
358 
359 	_enter("%d", local->debug_id);
360 
361 	ASSERT(!work_pending(&local->processor));
362 
363 	_net("DESTROY LOCAL %d", local->debug_id);
364 	kfree(local);
365 	_leave("");
366 }
367 
368 /*
369  * Verify the local endpoint list is empty by this point.
370  */
371 void __exit rxrpc_destroy_all_locals(void)
372 {
373 	struct rxrpc_local *local;
374 
375 	_enter("");
376 
377 	flush_workqueue(rxrpc_workqueue);
378 
379 	if (!list_empty(&rxrpc_local_endpoints)) {
380 		mutex_lock(&rxrpc_local_mutex);
381 		list_for_each_entry(local, &rxrpc_local_endpoints, link) {
382 			pr_err("AF_RXRPC: Leaked local %p {%d}\n",
383 			       local, atomic_read(&local->usage));
384 		}
385 		mutex_unlock(&rxrpc_local_mutex);
386 		BUG();
387 	}
388 
389 	rcu_barrier();
390 }
391