xref: /openbmc/linux/net/rxrpc/peer_event.c (revision e285d5bf)
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 						     struct sockaddr_rxrpc *srx)
33 {
34 	struct sock_exterr_skb *serr = SKB_EXT_ERR(skb);
35 
36 	_enter("");
37 
38 	memset(srx, 0, sizeof(*srx));
39 	srx->transport_type = local->srx.transport_type;
40 	srx->transport_len = local->srx.transport_len;
41 	srx->transport.family = local->srx.transport.family;
42 
43 	/* Can we see an ICMP4 packet on an ICMP6 listening socket?  and vice
44 	 * versa?
45 	 */
46 	switch (srx->transport.family) {
47 	case AF_INET:
48 		srx->transport.sin.sin_port = serr->port;
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 		switch (serr->ee.ee_origin) {
73 		case SO_EE_ORIGIN_ICMP6:
74 			_net("Rx ICMP6");
75 			memcpy(&srx->transport.sin6.sin6_addr,
76 			       skb_network_header(skb) + serr->addr_offset,
77 			       sizeof(struct in6_addr));
78 			break;
79 		case SO_EE_ORIGIN_ICMP:
80 			_net("Rx ICMP on v6 sock");
81 			srx->transport.sin6.sin6_addr.s6_addr32[0] = 0;
82 			srx->transport.sin6.sin6_addr.s6_addr32[1] = 0;
83 			srx->transport.sin6.sin6_addr.s6_addr32[2] = htonl(0xffff);
84 			memcpy(srx->transport.sin6.sin6_addr.s6_addr + 12,
85 			       skb_network_header(skb) + serr->addr_offset,
86 			       sizeof(struct in_addr));
87 			break;
88 		default:
89 			memcpy(&srx->transport.sin6.sin6_addr,
90 			       &ipv6_hdr(skb)->saddr,
91 			       sizeof(struct in6_addr));
92 			break;
93 		}
94 		break;
95 #endif
96 
97 	default:
98 		BUG();
99 	}
100 
101 	return rxrpc_lookup_peer_rcu(local, srx);
102 }
103 
104 /*
105  * Handle an MTU/fragmentation problem.
106  */
107 static void rxrpc_adjust_mtu(struct rxrpc_peer *peer, struct sock_exterr_skb *serr)
108 {
109 	u32 mtu = serr->ee.ee_info;
110 
111 	_net("Rx ICMP Fragmentation Needed (%d)", mtu);
112 
113 	/* wind down the local interface MTU */
114 	if (mtu > 0 && peer->if_mtu == 65535 && mtu < peer->if_mtu) {
115 		peer->if_mtu = mtu;
116 		_net("I/F MTU %u", mtu);
117 	}
118 
119 	if (mtu == 0) {
120 		/* they didn't give us a size, estimate one */
121 		mtu = peer->if_mtu;
122 		if (mtu > 1500) {
123 			mtu >>= 1;
124 			if (mtu < 1500)
125 				mtu = 1500;
126 		} else {
127 			mtu -= 100;
128 			if (mtu < peer->hdrsize)
129 				mtu = peer->hdrsize + 4;
130 		}
131 	}
132 
133 	if (mtu < peer->mtu) {
134 		spin_lock_bh(&peer->lock);
135 		peer->mtu = mtu;
136 		peer->maxdata = peer->mtu - peer->hdrsize;
137 		spin_unlock_bh(&peer->lock);
138 		_net("Net MTU %u (maxdata %u)",
139 		     peer->mtu, peer->maxdata);
140 	}
141 }
142 
143 /*
144  * Handle an error received on the local endpoint.
145  */
146 void rxrpc_error_report(struct sock *sk)
147 {
148 	struct sock_exterr_skb *serr;
149 	struct sockaddr_rxrpc srx;
150 	struct rxrpc_local *local = sk->sk_user_data;
151 	struct rxrpc_peer *peer;
152 	struct sk_buff *skb;
153 
154 	_enter("%p{%d}", sk, local->debug_id);
155 
156 	skb = sock_dequeue_err_skb(sk);
157 	if (!skb) {
158 		_leave("UDP socket errqueue empty");
159 		return;
160 	}
161 	rxrpc_new_skb(skb, rxrpc_skb_rx_received);
162 	serr = SKB_EXT_ERR(skb);
163 	if (!skb->len && serr->ee.ee_origin == SO_EE_ORIGIN_TIMESTAMPING) {
164 		_leave("UDP empty message");
165 		rxrpc_free_skb(skb, rxrpc_skb_rx_freed);
166 		return;
167 	}
168 
169 	rcu_read_lock();
170 	peer = rxrpc_lookup_peer_icmp_rcu(local, skb, &srx);
171 	if (peer && !rxrpc_get_peer_maybe(peer))
172 		peer = NULL;
173 	if (!peer) {
174 		rcu_read_unlock();
175 		rxrpc_free_skb(skb, rxrpc_skb_rx_freed);
176 		_leave(" [no peer]");
177 		return;
178 	}
179 
180 	trace_rxrpc_rx_icmp(peer, &serr->ee, &srx);
181 
182 	if ((serr->ee.ee_origin == SO_EE_ORIGIN_ICMP &&
183 	     serr->ee.ee_type == ICMP_DEST_UNREACH &&
184 	     serr->ee.ee_code == ICMP_FRAG_NEEDED)) {
185 		rxrpc_adjust_mtu(peer, serr);
186 		rcu_read_unlock();
187 		rxrpc_free_skb(skb, rxrpc_skb_rx_freed);
188 		rxrpc_put_peer(peer);
189 		_leave(" [MTU update]");
190 		return;
191 	}
192 
193 	rxrpc_store_error(peer, serr);
194 	rcu_read_unlock();
195 	rxrpc_free_skb(skb, rxrpc_skb_rx_freed);
196 
197 	/* The ref we obtained is passed off to the work item */
198 	__rxrpc_queue_peer_error(peer);
199 	_leave("");
200 }
201 
202 /*
203  * Map an error report to error codes on the peer record.
204  */
205 static void rxrpc_store_error(struct rxrpc_peer *peer,
206 			      struct sock_exterr_skb *serr)
207 {
208 	struct sock_extended_err *ee;
209 	int err;
210 
211 	_enter("");
212 
213 	ee = &serr->ee;
214 
215 	err = ee->ee_errno;
216 
217 	switch (ee->ee_origin) {
218 	case SO_EE_ORIGIN_ICMP:
219 		switch (ee->ee_type) {
220 		case ICMP_DEST_UNREACH:
221 			switch (ee->ee_code) {
222 			case ICMP_NET_UNREACH:
223 				_net("Rx Received ICMP Network Unreachable");
224 				break;
225 			case ICMP_HOST_UNREACH:
226 				_net("Rx Received ICMP Host Unreachable");
227 				break;
228 			case ICMP_PORT_UNREACH:
229 				_net("Rx Received ICMP Port Unreachable");
230 				break;
231 			case ICMP_NET_UNKNOWN:
232 				_net("Rx Received ICMP Unknown Network");
233 				break;
234 			case ICMP_HOST_UNKNOWN:
235 				_net("Rx Received ICMP Unknown Host");
236 				break;
237 			default:
238 				_net("Rx Received ICMP DestUnreach code=%u",
239 				     ee->ee_code);
240 				break;
241 			}
242 			break;
243 
244 		case ICMP_TIME_EXCEEDED:
245 			_net("Rx Received ICMP TTL Exceeded");
246 			break;
247 
248 		default:
249 			_proto("Rx Received ICMP error { type=%u code=%u }",
250 			       ee->ee_type, ee->ee_code);
251 			break;
252 		}
253 		break;
254 
255 	case SO_EE_ORIGIN_NONE:
256 	case SO_EE_ORIGIN_LOCAL:
257 		_proto("Rx Received local error { error=%d }", err);
258 		err += RXRPC_LOCAL_ERROR_OFFSET;
259 		break;
260 
261 	case SO_EE_ORIGIN_ICMP6:
262 	default:
263 		_proto("Rx Received error report { orig=%u }", ee->ee_origin);
264 		break;
265 	}
266 
267 	peer->error_report = err;
268 }
269 
270 /*
271  * Distribute an error that occurred on a peer
272  */
273 void rxrpc_peer_error_distributor(struct work_struct *work)
274 {
275 	struct rxrpc_peer *peer =
276 		container_of(work, struct rxrpc_peer, error_distributor);
277 	struct rxrpc_call *call;
278 	enum rxrpc_call_completion compl;
279 	int error;
280 
281 	_enter("");
282 
283 	error = READ_ONCE(peer->error_report);
284 	if (error < RXRPC_LOCAL_ERROR_OFFSET) {
285 		compl = RXRPC_CALL_NETWORK_ERROR;
286 	} else {
287 		compl = RXRPC_CALL_LOCAL_ERROR;
288 		error -= RXRPC_LOCAL_ERROR_OFFSET;
289 	}
290 
291 	_debug("ISSUE ERROR %s %d", rxrpc_call_completions[compl], error);
292 
293 	spin_lock_bh(&peer->lock);
294 
295 	while (!hlist_empty(&peer->error_targets)) {
296 		call = hlist_entry(peer->error_targets.first,
297 				   struct rxrpc_call, error_link);
298 		hlist_del_init(&call->error_link);
299 		rxrpc_see_call(call);
300 
301 		if (rxrpc_set_call_completion(call, compl, 0, -error))
302 			rxrpc_notify_socket(call);
303 	}
304 
305 	spin_unlock_bh(&peer->lock);
306 
307 	rxrpc_put_peer(peer);
308 	_leave("");
309 }
310 
311 /*
312  * Add RTT information to cache.  This is called in softirq mode and has
313  * exclusive access to the peer RTT data.
314  */
315 void rxrpc_peer_add_rtt(struct rxrpc_call *call, enum rxrpc_rtt_rx_trace why,
316 			rxrpc_serial_t send_serial, rxrpc_serial_t resp_serial,
317 			ktime_t send_time, ktime_t resp_time)
318 {
319 	struct rxrpc_peer *peer = call->peer;
320 	s64 rtt;
321 	u64 sum = peer->rtt_sum, avg;
322 	u8 cursor = peer->rtt_cursor, usage = peer->rtt_usage;
323 
324 	rtt = ktime_to_ns(ktime_sub(resp_time, send_time));
325 	if (rtt < 0)
326 		return;
327 
328 	/* Replace the oldest datum in the RTT buffer */
329 	sum -= peer->rtt_cache[cursor];
330 	sum += rtt;
331 	peer->rtt_cache[cursor] = rtt;
332 	peer->rtt_cursor = (cursor + 1) & (RXRPC_RTT_CACHE_SIZE - 1);
333 	peer->rtt_sum = sum;
334 	if (usage < RXRPC_RTT_CACHE_SIZE) {
335 		usage++;
336 		peer->rtt_usage = usage;
337 	}
338 
339 	/* Now recalculate the average */
340 	if (usage == RXRPC_RTT_CACHE_SIZE) {
341 		avg = sum / RXRPC_RTT_CACHE_SIZE;
342 	} else {
343 		avg = sum;
344 		do_div(avg, usage);
345 	}
346 
347 	peer->rtt = avg;
348 	trace_rxrpc_rtt_rx(call, why, send_serial, resp_serial, rtt,
349 			   usage, avg);
350 }
351 
352 /*
353  * Perform keep-alive pings.
354  */
355 static void rxrpc_peer_keepalive_dispatch(struct rxrpc_net *rxnet,
356 					  struct list_head *collector,
357 					  time64_t base,
358 					  u8 cursor)
359 {
360 	struct rxrpc_peer *peer;
361 	const u8 mask = ARRAY_SIZE(rxnet->peer_keepalive) - 1;
362 	time64_t keepalive_at;
363 	int slot;
364 
365 	spin_lock_bh(&rxnet->peer_hash_lock);
366 
367 	while (!list_empty(collector)) {
368 		peer = list_entry(collector->next,
369 				  struct rxrpc_peer, keepalive_link);
370 
371 		list_del_init(&peer->keepalive_link);
372 		if (!rxrpc_get_peer_maybe(peer))
373 			continue;
374 
375 		spin_unlock_bh(&rxnet->peer_hash_lock);
376 
377 		keepalive_at = peer->last_tx_at + RXRPC_KEEPALIVE_TIME;
378 		slot = keepalive_at - base;
379 		_debug("%02x peer %u t=%d {%pISp}",
380 		       cursor, peer->debug_id, slot, &peer->srx.transport);
381 
382 		if (keepalive_at <= base ||
383 		    keepalive_at > base + RXRPC_KEEPALIVE_TIME) {
384 			rxrpc_send_keepalive(peer);
385 			slot = RXRPC_KEEPALIVE_TIME;
386 		}
387 
388 		/* A transmission to this peer occurred since last we examined
389 		 * it so put it into the appropriate future bucket.
390 		 */
391 		slot += cursor;
392 		slot &= mask;
393 		spin_lock_bh(&rxnet->peer_hash_lock);
394 		list_add_tail(&peer->keepalive_link,
395 			      &rxnet->peer_keepalive[slot & mask]);
396 		rxrpc_put_peer(peer);
397 	}
398 
399 	spin_unlock_bh(&rxnet->peer_hash_lock);
400 }
401 
402 /*
403  * Perform keep-alive pings with VERSION packets to keep any NAT alive.
404  */
405 void rxrpc_peer_keepalive_worker(struct work_struct *work)
406 {
407 	struct rxrpc_net *rxnet =
408 		container_of(work, struct rxrpc_net, peer_keepalive_work);
409 	const u8 mask = ARRAY_SIZE(rxnet->peer_keepalive) - 1;
410 	time64_t base, now, delay;
411 	u8 cursor, stop;
412 	LIST_HEAD(collector);
413 
414 	now = ktime_get_seconds();
415 	base = rxnet->peer_keepalive_base;
416 	cursor = rxnet->peer_keepalive_cursor;
417 	_enter("%lld,%u", base - now, cursor);
418 
419 	if (!rxnet->live)
420 		return;
421 
422 	/* Remove to a temporary list all the peers that are currently lodged
423 	 * in expired buckets plus all new peers.
424 	 *
425 	 * Everything in the bucket at the cursor is processed this
426 	 * second; the bucket at cursor + 1 goes at now + 1s and so
427 	 * on...
428 	 */
429 	spin_lock_bh(&rxnet->peer_hash_lock);
430 	list_splice_init(&rxnet->peer_keepalive_new, &collector);
431 
432 	stop = cursor + ARRAY_SIZE(rxnet->peer_keepalive);
433 	while (base <= now && (s8)(cursor - stop) < 0) {
434 		list_splice_tail_init(&rxnet->peer_keepalive[cursor & mask],
435 				      &collector);
436 		base++;
437 		cursor++;
438 	}
439 
440 	base = now;
441 	spin_unlock_bh(&rxnet->peer_hash_lock);
442 
443 	rxnet->peer_keepalive_base = base;
444 	rxnet->peer_keepalive_cursor = cursor;
445 	rxrpc_peer_keepalive_dispatch(rxnet, &collector, base, cursor);
446 	ASSERT(list_empty(&collector));
447 
448 	/* Schedule the timer for the next occupied timeslot. */
449 	cursor = rxnet->peer_keepalive_cursor;
450 	stop = cursor + RXRPC_KEEPALIVE_TIME - 1;
451 	for (; (s8)(cursor - stop) < 0; cursor++) {
452 		if (!list_empty(&rxnet->peer_keepalive[cursor & mask]))
453 			break;
454 		base++;
455 	}
456 
457 	now = ktime_get_seconds();
458 	delay = base - now;
459 	if (delay < 1)
460 		delay = 1;
461 	delay *= HZ;
462 	if (rxnet->live)
463 		timer_reduce(&rxnet->peer_keepalive_timer, jiffies + delay);
464 
465 	_leave("");
466 }
467