peer_event.c

  1// SPDX-License-Identifier: GPL-2.0-or-later
  2/* Peer event handling, typically ICMP messages.
  3 *
  4 * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
  5 * Written by David Howells (dhowells@redhat.com)
  6 */
  7
  8#include <linux/module.h>
  9#include <linux/net.h>
 10#include <linux/skbuff.h>
 11#include <linux/errqueue.h>
 12#include <linux/udp.h>
 13#include <linux/in.h>
 14#include <linux/in6.h>
 15#include <linux/icmp.h>
 16#include <net/sock.h>
 17#include <net/af_rxrpc.h>
 18#include <net/ip.h>
 19#include "ar-internal.h"
 20
 21static void rxrpc_store_error(struct rxrpc_peer *, struct sk_buff *);
 22static void rxrpc_distribute_error(struct rxrpc_peer *, struct sk_buff *,
 23				   enum rxrpc_call_completion, int);
 24
 25/*
 26 * Find the peer associated with a local error.
 27 */
 28static struct rxrpc_peer *rxrpc_lookup_peer_local_rcu(struct rxrpc_local *local,
 29						      const struct sk_buff *skb,
 30						      struct sockaddr_rxrpc *srx)
 31{
 32	struct sock_exterr_skb *serr = SKB_EXT_ERR(skb);
 33
 34	_enter("");
 35
 36	memset(srx, 0, sizeof(*srx));
 37	srx->transport_type = local->srx.transport_type;
 38	srx->transport_len = local->srx.transport_len;
 39	srx->transport.family = local->srx.transport.family;
 40
 41	/* Can we see an ICMP4 packet on an ICMP6 listening socket?  and vice
 42	 * versa?
 43	 */
 44	switch (srx->transport.family) {
 45	case AF_INET:
 46		srx->transport_len = sizeof(srx->transport.sin);
 47		srx->transport.family = AF_INET;
 48		srx->transport.sin.sin_port = serr->port;
 49		switch (serr->ee.ee_origin) {
 50		case SO_EE_ORIGIN_ICMP:
 51			memcpy(&srx->transport.sin.sin_addr,
 52			       skb_network_header(skb) + serr->addr_offset,
 53			       sizeof(struct in_addr));
 54			break;
 55		case SO_EE_ORIGIN_ICMP6:
 56			memcpy(&srx->transport.sin.sin_addr,
 57			       skb_network_header(skb) + serr->addr_offset + 12,
 58			       sizeof(struct in_addr));
 59			break;
 60		default:
 61			memcpy(&srx->transport.sin.sin_addr, &ip_hdr(skb)->saddr,
 62			       sizeof(struct in_addr));
 63			break;
 64		}
 65		break;
 66
 67#ifdef CONFIG_AF_RXRPC_IPV6
 68	case AF_INET6:
 69		switch (serr->ee.ee_origin) {
 70		case SO_EE_ORIGIN_ICMP6:
 71			srx->transport.sin6.sin6_port = serr->port;
 72			memcpy(&srx->transport.sin6.sin6_addr,
 73			       skb_network_header(skb) + serr->addr_offset,
 74			       sizeof(struct in6_addr));
 75			break;
 76		case SO_EE_ORIGIN_ICMP:
 77			srx->transport_len = sizeof(srx->transport.sin);
 78			srx->transport.family = AF_INET;
 79			srx->transport.sin.sin_port = serr->port;
 80			memcpy(&srx->transport.sin.sin_addr,
 81			       skb_network_header(skb) + serr->addr_offset,
 82			       sizeof(struct in_addr));
 83			break;
 84		default:
 85			memcpy(&srx->transport.sin6.sin6_addr,
 86			       &ipv6_hdr(skb)->saddr,
 87			       sizeof(struct in6_addr));
 88			break;
 89		}
 90		break;
 91#endif
 92
 93	default:
 94		BUG();
 95	}
 96
 97	return rxrpc_lookup_peer_rcu(local, srx);
 98}
 99
100/*
101 * Handle an MTU/fragmentation problem.
102 */
103static void rxrpc_adjust_mtu(struct rxrpc_peer *peer, unsigned int mtu)
104{
105	unsigned int max_data;
106
107	/* wind down the local interface MTU */
108	if (mtu > 0 && peer->if_mtu == 65535 && mtu < peer->if_mtu)
109		peer->if_mtu = mtu;
110
111	if (mtu == 0) {
112		/* they didn't give us a size, estimate one */
113		mtu = peer->if_mtu;
114		if (mtu > 1500) {
115			mtu >>= 1;
116			if (mtu < 1500)
117				mtu = 1500;
118		} else {
119			mtu -= 100;
120			if (mtu < peer->hdrsize)
121				mtu = peer->hdrsize + 4;
122		}
123	}
124
125	max_data = max_t(int, mtu - peer->hdrsize, 500);
126	if (max_data < peer->max_data) {
127		if (peer->pmtud_good > max_data)
128			peer->pmtud_good = max_data;
129		if (peer->pmtud_bad > max_data + 1)
130			peer->pmtud_bad = max_data + 1;
131
132		trace_rxrpc_pmtud_reduce(peer, 0, max_data, rxrpc_pmtud_reduce_icmp);
133		peer->max_data = max_data;
134	}
135}
136
137/*
138 * Handle an error received on the local endpoint.
139 */
140void rxrpc_input_error(struct rxrpc_local *local, struct sk_buff *skb)
141{
142	struct sock_exterr_skb *serr = SKB_EXT_ERR(skb);
143	struct sockaddr_rxrpc srx;
144	struct rxrpc_peer *peer = NULL;
145
146	_enter("L=%x", local->debug_id);
147
148	if (!skb->len && serr->ee.ee_origin == SO_EE_ORIGIN_TIMESTAMPING) {
149		_leave("UDP empty message");
150		return;
151	}
152
153	rcu_read_lock();
154	peer = rxrpc_lookup_peer_local_rcu(local, skb, &srx);
155	if (peer && !rxrpc_get_peer_maybe(peer, rxrpc_peer_get_input_error))
156		peer = NULL;
157	rcu_read_unlock();
158	if (!peer)
159		return;
160
161	trace_rxrpc_rx_icmp(peer, &serr->ee, &srx);
162
163	if ((serr->ee.ee_origin == SO_EE_ORIGIN_ICMP &&
164	     serr->ee.ee_type == ICMP_DEST_UNREACH &&
165	     serr->ee.ee_code == ICMP_FRAG_NEEDED)) {
166		rxrpc_adjust_mtu(peer, serr->ee.ee_info);
167		goto out;
168	}
169
170	if ((serr->ee.ee_origin == SO_EE_ORIGIN_ICMP6 &&
171	     serr->ee.ee_type == ICMPV6_PKT_TOOBIG &&
172	     serr->ee.ee_code == 0)) {
173		rxrpc_adjust_mtu(peer, serr->ee.ee_info);
174		goto out;
175	}
176
177	rxrpc_store_error(peer, skb);
178out:
179	rxrpc_put_peer(peer, rxrpc_peer_put_input_error);
180}
181
182/*
183 * Map an error report to error codes on the peer record.
184 */
185static void rxrpc_store_error(struct rxrpc_peer *peer, struct sk_buff *skb)
186{
187	enum rxrpc_call_completion compl = RXRPC_CALL_NETWORK_ERROR;
188	struct sock_exterr_skb *serr = SKB_EXT_ERR(skb);
189	struct sock_extended_err *ee = &serr->ee;
190	int err = ee->ee_errno;
191
192	_enter("");
193
194	switch (ee->ee_origin) {
195	case SO_EE_ORIGIN_NONE:
196	case SO_EE_ORIGIN_LOCAL:
197		compl = RXRPC_CALL_LOCAL_ERROR;
198		break;
199
200	case SO_EE_ORIGIN_ICMP6:
201		if (err == EACCES)
202			err = EHOSTUNREACH;
203		fallthrough;
204	case SO_EE_ORIGIN_ICMP:
205	default:
206		break;
207	}
208
209	rxrpc_distribute_error(peer, skb, compl, err);
210}
211
212/*
213 * Distribute an error that occurred on a peer.
214 */
215static void rxrpc_distribute_error(struct rxrpc_peer *peer, struct sk_buff *skb,
216				   enum rxrpc_call_completion compl, int err)
217{
218	struct rxrpc_call *call;
219	HLIST_HEAD(error_targets);
220
221	spin_lock_irq(&peer->lock);
222	hlist_move_list(&peer->error_targets, &error_targets);
223
224	while (!hlist_empty(&error_targets)) {
225		call = hlist_entry(error_targets.first,
226				   struct rxrpc_call, error_link);
227		hlist_del_init(&call->error_link);
228		spin_unlock_irq(&peer->lock);
229
230		rxrpc_see_call(call, rxrpc_call_see_distribute_error);
231		rxrpc_set_call_completion(call, compl, 0, -err);
232		rxrpc_input_call_event(call);
233
234		spin_lock_irq(&peer->lock);
235	}
236
237	spin_unlock_irq(&peer->lock);
238}
239
240/*
241 * Reconstruct the last transmission time.  The difference calculated should be
242 * valid provided no more than ~68 years elapsed since the last transmission.
243 */
244static time64_t rxrpc_peer_get_tx_mark(const struct rxrpc_peer *peer, time64_t base)
245{
246	s32 last_tx_at = READ_ONCE(peer->last_tx_at);
247	s32 base_lsw = base;
248	s32 diff = last_tx_at - base_lsw;
249
250	diff = clamp(diff, -RXRPC_KEEPALIVE_TIME, RXRPC_KEEPALIVE_TIME);
251
252	return diff + base;
253}
254
255/*
256 * Perform keep-alive pings.
257 */
258static void rxrpc_peer_keepalive_dispatch(struct rxrpc_net *rxnet,
259					  struct list_head *collector,
260					  time64_t base,
261					  u8 cursor)
262{
263	struct rxrpc_peer *peer;
264	const u8 mask = ARRAY_SIZE(rxnet->peer_keepalive) - 1;
265	time64_t keepalive_at;
266	bool use;
267	int slot;
268
269	spin_lock_bh(&rxnet->peer_hash_lock);
270
271	while (!list_empty(collector)) {
272		peer = list_entry(collector->next,
273				  struct rxrpc_peer, keepalive_link);
274
275		list_del_init(&peer->keepalive_link);
276		if (!rxrpc_get_peer_maybe(peer, rxrpc_peer_get_keepalive))
277			continue;
278
279		use = __rxrpc_use_local(peer->local, rxrpc_local_use_peer_keepalive);
280		spin_unlock_bh(&rxnet->peer_hash_lock);
281
282		if (use) {
283			keepalive_at = rxrpc_peer_get_tx_mark(peer, base) + RXRPC_KEEPALIVE_TIME;
284			slot = keepalive_at - base;
285			_debug("%02x peer %u t=%d {%pISp}",
286			       cursor, peer->debug_id, slot, &peer->srx.transport);
287
288			if (keepalive_at <= base ||
289			    keepalive_at > base + RXRPC_KEEPALIVE_TIME) {
290				rxrpc_send_keepalive(peer);
291				slot = RXRPC_KEEPALIVE_TIME;
292			}
293
294			/* A transmission to this peer occurred since last we
295			 * examined it so put it into the appropriate future
296			 * bucket.
297			 */
298			slot += cursor;
299			slot &= mask;
300			spin_lock_bh(&rxnet->peer_hash_lock);
301			list_add_tail(&peer->keepalive_link,
302				      &rxnet->peer_keepalive[slot & mask]);
303			spin_unlock_bh(&rxnet->peer_hash_lock);
304			rxrpc_unuse_local(peer->local, rxrpc_local_unuse_peer_keepalive);
305		}
306		rxrpc_put_peer(peer, rxrpc_peer_put_keepalive);
307		spin_lock_bh(&rxnet->peer_hash_lock);
308	}
309
310	spin_unlock_bh(&rxnet->peer_hash_lock);
311}
312
313/*
314 * Perform keep-alive pings with VERSION packets to keep any NAT alive.
315 */
316void rxrpc_peer_keepalive_worker(struct work_struct *work)
317{
318	struct rxrpc_net *rxnet =
319		container_of(work, struct rxrpc_net, peer_keepalive_work);
320	const u8 mask = ARRAY_SIZE(rxnet->peer_keepalive) - 1;
321	time64_t base, now, delay;
322	u8 cursor, stop;
323	LIST_HEAD(collector);
324
325	now = ktime_get_seconds();
326	base = rxnet->peer_keepalive_base;
327	cursor = rxnet->peer_keepalive_cursor;
328	_enter("%lld,%u", base - now, cursor);
329
330	if (!rxnet->live)
331		return;
332
333	/* Remove to a temporary list all the peers that are currently lodged
334	 * in expired buckets plus all new peers.
335	 *
336	 * Everything in the bucket at the cursor is processed this
337	 * second; the bucket at cursor + 1 goes at now + 1s and so
338	 * on...
339	 */
340	spin_lock_bh(&rxnet->peer_hash_lock);
341	list_splice_init(&rxnet->peer_keepalive_new, &collector);
342
343	stop = cursor + ARRAY_SIZE(rxnet->peer_keepalive);
344	while (base <= now && (s8)(cursor - stop) < 0) {
345		list_splice_tail_init(&rxnet->peer_keepalive[cursor & mask],
346				      &collector);
347		base++;
348		cursor++;
349	}
350
351	base = now;
352	spin_unlock_bh(&rxnet->peer_hash_lock);
353
354	rxnet->peer_keepalive_base = base;
355	rxnet->peer_keepalive_cursor = cursor;
356	rxrpc_peer_keepalive_dispatch(rxnet, &collector, base, cursor);
357	ASSERT(list_empty(&collector));
358
359	/* Schedule the timer for the next occupied timeslot. */
360	cursor = rxnet->peer_keepalive_cursor;
361	stop = cursor + RXRPC_KEEPALIVE_TIME - 1;
362	for (; (s8)(cursor - stop) < 0; cursor++) {
363		if (!list_empty(&rxnet->peer_keepalive[cursor & mask]))
364			break;
365		base++;
366	}
367
368	now = ktime_get_seconds();
369	delay = base - now;
370	if (delay < 1)
371		delay = 1;
372	delay *= HZ;
373	if (rxnet->live)
374		timer_reduce(&rxnet->peer_keepalive_timer, jiffies + delay);
375
376	_leave("");
377}
378
379/*
380 * Do path MTU probing.
381 */
382void rxrpc_input_probe_for_pmtud(struct rxrpc_connection *conn, rxrpc_serial_t acked_serial,
383				 bool sendmsg_fail)
384{
385	struct rxrpc_peer *peer = conn->peer;
386	unsigned int max_data = peer->max_data;
387	int good, trial, bad, jumbo;
388
389	good  = peer->pmtud_good;
390	trial = peer->pmtud_trial;
391	bad   = peer->pmtud_bad;
392	if (good >= bad - 1) {
393		conn->pmtud_probe = 0;
394		peer->pmtud_lost = false;
395		return;
396	}
397
398	if (!peer->pmtud_probing)
399		goto send_probe;
400
401	if (sendmsg_fail || after(acked_serial, conn->pmtud_probe)) {
402		/* Retry a lost probe. */
403		if (!peer->pmtud_lost) {
404			trace_rxrpc_pmtud_lost(conn, acked_serial);
405			conn->pmtud_probe = 0;
406			peer->pmtud_lost = true;
407			goto send_probe;
408		}
409
410		/* The probed size didn't seem to get through. */
411		bad = trial;
412		peer->pmtud_bad = bad;
413		if (bad <= max_data)
414			max_data = bad - 1;
415	} else {
416		/* It did get through. */
417		good = trial;
418		peer->pmtud_good = good;
419		if (good > max_data)
420			max_data = good;
421	}
422
423	max_data = umin(max_data, peer->ackr_max_data);
424	if (max_data != peer->max_data)
425		peer->max_data = max_data;
426
427	jumbo = max_data + sizeof(struct rxrpc_jumbo_header);
428	jumbo /= RXRPC_JUMBO_SUBPKTLEN;
429	peer->pmtud_jumbo = jumbo;
430
431	trace_rxrpc_pmtud_rx(conn, acked_serial);
432	conn->pmtud_probe = 0;
433	peer->pmtud_lost = false;
434
435	if (good < RXRPC_JUMBO(2) && bad > RXRPC_JUMBO(2))
436		trial = RXRPC_JUMBO(2);
437	else if (good < RXRPC_JUMBO(4) && bad > RXRPC_JUMBO(4))
438		trial = RXRPC_JUMBO(4);
439	else if (good < RXRPC_JUMBO(3) && bad > RXRPC_JUMBO(3))
440		trial = RXRPC_JUMBO(3);
441	else if (good < RXRPC_JUMBO(6) && bad > RXRPC_JUMBO(6))
442		trial = RXRPC_JUMBO(6);
443	else if (good < RXRPC_JUMBO(5) && bad > RXRPC_JUMBO(5))
444		trial = RXRPC_JUMBO(5);
445	else if (good < RXRPC_JUMBO(8) && bad > RXRPC_JUMBO(8))
446		trial = RXRPC_JUMBO(8);
447	else if (good < RXRPC_JUMBO(7) && bad > RXRPC_JUMBO(7))
448		trial = RXRPC_JUMBO(7);
449	else
450		trial = (good + bad) / 2;
451	peer->pmtud_trial = trial;
452
453	if (good >= bad)
454		return;
455
456send_probe:
457	peer->pmtud_pending = true;
458}