xref: /linux/net/rxrpc/peer_event.c (revision e5c86679d5e864947a52fb31e45a425dea3e7fa9)
1 /* Peer event handling, typically ICMP messages.
2  *
3  * Copyright (C) 2007 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 License
8  * as published by the Free Software Foundation; either version
9  * 2 of the License, or (at your option) any later version.
10  */
11 
12 #include <linux/module.h>
13 #include <linux/net.h>
14 #include <linux/skbuff.h>
15 #include <linux/errqueue.h>
16 #include <linux/udp.h>
17 #include <linux/in.h>
18 #include <linux/in6.h>
19 #include <linux/icmp.h>
20 #include <net/sock.h>
21 #include <net/af_rxrpc.h>
22 #include <net/ip.h>
23 #include "ar-internal.h"
24 
25 static void rxrpc_store_error(struct rxrpc_peer *, struct sock_exterr_skb *);
26 
27 /*
28  * Find the peer associated with an ICMP packet.
29  */
30 static struct rxrpc_peer *rxrpc_lookup_peer_icmp_rcu(struct rxrpc_local *local,
31 						     const struct sk_buff *skb)
32 {
33 	struct sock_exterr_skb *serr = SKB_EXT_ERR(skb);
34 	struct sockaddr_rxrpc srx;
35 
36 	_enter("");
37 
38 	memset(&srx, 0, sizeof(srx));
39 	srx.transport_type = local->srx.transport_type;
40 	srx.transport.family = local->srx.transport.family;
41 
42 	/* Can we see an ICMP4 packet on an ICMP6 listening socket?  and vice
43 	 * versa?
44 	 */
45 	switch (srx.transport.family) {
46 	case AF_INET:
47 		srx.transport.sin.sin_port = serr->port;
48 		srx.transport_len = sizeof(struct sockaddr_in);
49 		switch (serr->ee.ee_origin) {
50 		case SO_EE_ORIGIN_ICMP:
51 			_net("Rx ICMP");
52 			memcpy(&srx.transport.sin.sin_addr,
53 			       skb_network_header(skb) + serr->addr_offset,
54 			       sizeof(struct in_addr));
55 			break;
56 		case SO_EE_ORIGIN_ICMP6:
57 			_net("Rx ICMP6 on v4 sock");
58 			memcpy(&srx.transport.sin.sin_addr,
59 			       skb_network_header(skb) + serr->addr_offset + 12,
60 			       sizeof(struct in_addr));
61 			break;
62 		default:
63 			memcpy(&srx.transport.sin.sin_addr, &ip_hdr(skb)->saddr,
64 			       sizeof(struct in_addr));
65 			break;
66 		}
67 		break;
68 
69 #ifdef CONFIG_AF_RXRPC_IPV6
70 	case AF_INET6:
71 		srx.transport.sin6.sin6_port = serr->port;
72 		srx.transport_len = sizeof(struct sockaddr_in6);
73 		switch (serr->ee.ee_origin) {
74 		case SO_EE_ORIGIN_ICMP6:
75 			_net("Rx ICMP6");
76 			memcpy(&srx.transport.sin6.sin6_addr,
77 			       skb_network_header(skb) + serr->addr_offset,
78 			       sizeof(struct in6_addr));
79 			break;
80 		case SO_EE_ORIGIN_ICMP:
81 			_net("Rx ICMP on v6 sock");
82 			memcpy(srx.transport.sin6.sin6_addr.s6_addr + 12,
83 			       skb_network_header(skb) + serr->addr_offset,
84 			       sizeof(struct in_addr));
85 			break;
86 		default:
87 			memcpy(&srx.transport.sin6.sin6_addr,
88 			       &ipv6_hdr(skb)->saddr,
89 			       sizeof(struct in6_addr));
90 			break;
91 		}
92 		break;
93 #endif
94 
95 	default:
96 		BUG();
97 	}
98 
99 	return rxrpc_lookup_peer_rcu(local, &srx);
100 }
101 
102 /*
103  * Handle an MTU/fragmentation problem.
104  */
105 static void rxrpc_adjust_mtu(struct rxrpc_peer *peer, struct sock_exterr_skb *serr)
106 {
107 	u32 mtu = serr->ee.ee_info;
108 
109 	_net("Rx ICMP Fragmentation Needed (%d)", mtu);
110 
111 	/* wind down the local interface MTU */
112 	if (mtu > 0 && peer->if_mtu == 65535 && mtu < peer->if_mtu) {
113 		peer->if_mtu = mtu;
114 		_net("I/F MTU %u", mtu);
115 	}
116 
117 	if (mtu == 0) {
118 		/* they didn't give us a size, estimate one */
119 		mtu = peer->if_mtu;
120 		if (mtu > 1500) {
121 			mtu >>= 1;
122 			if (mtu < 1500)
123 				mtu = 1500;
124 		} else {
125 			mtu -= 100;
126 			if (mtu < peer->hdrsize)
127 				mtu = peer->hdrsize + 4;
128 		}
129 	}
130 
131 	if (mtu < peer->mtu) {
132 		spin_lock_bh(&peer->lock);
133 		peer->mtu = mtu;
134 		peer->maxdata = peer->mtu - peer->hdrsize;
135 		spin_unlock_bh(&peer->lock);
136 		_net("Net MTU %u (maxdata %u)",
137 		     peer->mtu, peer->maxdata);
138 	}
139 }
140 
141 /*
142  * Handle an error received on the local endpoint.
143  */
144 void rxrpc_error_report(struct sock *sk)
145 {
146 	struct sock_exterr_skb *serr;
147 	struct rxrpc_local *local = sk->sk_user_data;
148 	struct rxrpc_peer *peer;
149 	struct sk_buff *skb;
150 
151 	_enter("%p{%d}", sk, local->debug_id);
152 
153 	skb = sock_dequeue_err_skb(sk);
154 	if (!skb) {
155 		_leave("UDP socket errqueue empty");
156 		return;
157 	}
158 	rxrpc_new_skb(skb, rxrpc_skb_rx_received);
159 	serr = SKB_EXT_ERR(skb);
160 	if (!skb->len && serr->ee.ee_origin == SO_EE_ORIGIN_TIMESTAMPING) {
161 		_leave("UDP empty message");
162 		rxrpc_free_skb(skb, rxrpc_skb_rx_freed);
163 		return;
164 	}
165 
166 	rcu_read_lock();
167 	peer = rxrpc_lookup_peer_icmp_rcu(local, skb);
168 	if (peer && !rxrpc_get_peer_maybe(peer))
169 		peer = NULL;
170 	if (!peer) {
171 		rcu_read_unlock();
172 		rxrpc_free_skb(skb, rxrpc_skb_rx_freed);
173 		_leave(" [no peer]");
174 		return;
175 	}
176 
177 	if ((serr->ee.ee_origin == SO_EE_ORIGIN_ICMP &&
178 	     serr->ee.ee_type == ICMP_DEST_UNREACH &&
179 	     serr->ee.ee_code == ICMP_FRAG_NEEDED)) {
180 		rxrpc_adjust_mtu(peer, serr);
181 		rcu_read_unlock();
182 		rxrpc_free_skb(skb, rxrpc_skb_rx_freed);
183 		rxrpc_put_peer(peer);
184 		_leave(" [MTU update]");
185 		return;
186 	}
187 
188 	rxrpc_store_error(peer, serr);
189 	rcu_read_unlock();
190 	rxrpc_free_skb(skb, rxrpc_skb_rx_freed);
191 
192 	/* The ref we obtained is passed off to the work item */
193 	rxrpc_queue_work(&peer->error_distributor);
194 	_leave("");
195 }
196 
197 /*
198  * Map an error report to error codes on the peer record.
199  */
200 static void rxrpc_store_error(struct rxrpc_peer *peer,
201 			      struct sock_exterr_skb *serr)
202 {
203 	struct sock_extended_err *ee;
204 	int err;
205 
206 	_enter("");
207 
208 	ee = &serr->ee;
209 
210 	_net("Rx Error o=%d t=%d c=%d e=%d",
211 	     ee->ee_origin, ee->ee_type, ee->ee_code, ee->ee_errno);
212 
213 	err = ee->ee_errno;
214 
215 	switch (ee->ee_origin) {
216 	case SO_EE_ORIGIN_ICMP:
217 		switch (ee->ee_type) {
218 		case ICMP_DEST_UNREACH:
219 			switch (ee->ee_code) {
220 			case ICMP_NET_UNREACH:
221 				_net("Rx Received ICMP Network Unreachable");
222 				break;
223 			case ICMP_HOST_UNREACH:
224 				_net("Rx Received ICMP Host Unreachable");
225 				break;
226 			case ICMP_PORT_UNREACH:
227 				_net("Rx Received ICMP Port Unreachable");
228 				break;
229 			case ICMP_NET_UNKNOWN:
230 				_net("Rx Received ICMP Unknown Network");
231 				break;
232 			case ICMP_HOST_UNKNOWN:
233 				_net("Rx Received ICMP Unknown Host");
234 				break;
235 			default:
236 				_net("Rx Received ICMP DestUnreach code=%u",
237 				     ee->ee_code);
238 				break;
239 			}
240 			break;
241 
242 		case ICMP_TIME_EXCEEDED:
243 			_net("Rx Received ICMP TTL Exceeded");
244 			break;
245 
246 		default:
247 			_proto("Rx Received ICMP error { type=%u code=%u }",
248 			       ee->ee_type, ee->ee_code);
249 			break;
250 		}
251 		break;
252 
253 	case SO_EE_ORIGIN_NONE:
254 	case SO_EE_ORIGIN_LOCAL:
255 		_proto("Rx Received local error { error=%d }", err);
256 		err += RXRPC_LOCAL_ERROR_OFFSET;
257 		break;
258 
259 	case SO_EE_ORIGIN_ICMP6:
260 	default:
261 		_proto("Rx Received error report { orig=%u }", ee->ee_origin);
262 		break;
263 	}
264 
265 	peer->error_report = err;
266 }
267 
268 /*
269  * Distribute an error that occurred on a peer
270  */
271 void rxrpc_peer_error_distributor(struct work_struct *work)
272 {
273 	struct rxrpc_peer *peer =
274 		container_of(work, struct rxrpc_peer, error_distributor);
275 	struct rxrpc_call *call;
276 	enum rxrpc_call_completion compl;
277 	int error;
278 
279 	_enter("");
280 
281 	error = READ_ONCE(peer->error_report);
282 	if (error < RXRPC_LOCAL_ERROR_OFFSET) {
283 		compl = RXRPC_CALL_NETWORK_ERROR;
284 	} else {
285 		compl = RXRPC_CALL_LOCAL_ERROR;
286 		error -= RXRPC_LOCAL_ERROR_OFFSET;
287 	}
288 
289 	_debug("ISSUE ERROR %s %d", rxrpc_call_completions[compl], error);
290 
291 	spin_lock_bh(&peer->lock);
292 
293 	while (!hlist_empty(&peer->error_targets)) {
294 		call = hlist_entry(peer->error_targets.first,
295 				   struct rxrpc_call, error_link);
296 		hlist_del_init(&call->error_link);
297 		rxrpc_see_call(call);
298 
299 		if (rxrpc_set_call_completion(call, compl, 0, error))
300 			rxrpc_notify_socket(call);
301 	}
302 
303 	spin_unlock_bh(&peer->lock);
304 
305 	rxrpc_put_peer(peer);
306 	_leave("");
307 }
308 
309 /*
310  * Add RTT information to cache.  This is called in softirq mode and has
311  * exclusive access to the peer RTT data.
312  */
313 void rxrpc_peer_add_rtt(struct rxrpc_call *call, enum rxrpc_rtt_rx_trace why,
314 			rxrpc_serial_t send_serial, rxrpc_serial_t resp_serial,
315 			ktime_t send_time, ktime_t resp_time)
316 {
317 	struct rxrpc_peer *peer = call->peer;
318 	s64 rtt;
319 	u64 sum = peer->rtt_sum, avg;
320 	u8 cursor = peer->rtt_cursor, usage = peer->rtt_usage;
321 
322 	rtt = ktime_to_ns(ktime_sub(resp_time, send_time));
323 	if (rtt < 0)
324 		return;
325 
326 	/* Replace the oldest datum in the RTT buffer */
327 	sum -= peer->rtt_cache[cursor];
328 	sum += rtt;
329 	peer->rtt_cache[cursor] = rtt;
330 	peer->rtt_cursor = (cursor + 1) & (RXRPC_RTT_CACHE_SIZE - 1);
331 	peer->rtt_sum = sum;
332 	if (usage < RXRPC_RTT_CACHE_SIZE) {
333 		usage++;
334 		peer->rtt_usage = usage;
335 	}
336 
337 	/* Now recalculate the average */
338 	if (usage == RXRPC_RTT_CACHE_SIZE) {
339 		avg = sum / RXRPC_RTT_CACHE_SIZE;
340 	} else {
341 		avg = sum;
342 		do_div(avg, usage);
343 	}
344 
345 	peer->rtt = avg;
346 	trace_rxrpc_rtt_rx(call, why, send_serial, resp_serial, rtt,
347 			   usage, avg);
348 }
349