xref: /linux/net/rxrpc/peer_event.c (revision 4f77c3462308c62ffe7129cc18b9ac937f44b5a5)
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 
21 static void rxrpc_store_error(struct rxrpc_peer *, struct sk_buff *);
22 static 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  */
28 static 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  */
103 static void rxrpc_adjust_mtu(struct rxrpc_peer *peer, unsigned int mtu)
104 {
105 	/* wind down the local interface MTU */
106 	if (mtu > 0 && peer->if_mtu == 65535 && mtu < peer->if_mtu)
107 		peer->if_mtu = mtu;
108 
109 	if (mtu == 0) {
110 		/* they didn't give us a size, estimate one */
111 		mtu = peer->if_mtu;
112 		if (mtu > 1500) {
113 			mtu >>= 1;
114 			if (mtu < 1500)
115 				mtu = 1500;
116 		} else {
117 			mtu -= 100;
118 			if (mtu < peer->hdrsize)
119 				mtu = peer->hdrsize + 4;
120 		}
121 	}
122 
123 	if (mtu < peer->mtu) {
124 		spin_lock(&peer->lock);
125 		peer->mtu = mtu;
126 		peer->maxdata = peer->mtu - peer->hdrsize;
127 		spin_unlock(&peer->lock);
128 	}
129 }
130 
131 /*
132  * Handle an error received on the local endpoint.
133  */
134 void rxrpc_input_error(struct rxrpc_local *local, struct sk_buff *skb)
135 {
136 	struct sock_exterr_skb *serr = SKB_EXT_ERR(skb);
137 	struct sockaddr_rxrpc srx;
138 	struct rxrpc_peer *peer = NULL;
139 
140 	_enter("L=%x", local->debug_id);
141 
142 	if (!skb->len && serr->ee.ee_origin == SO_EE_ORIGIN_TIMESTAMPING) {
143 		_leave("UDP empty message");
144 		return;
145 	}
146 
147 	rcu_read_lock();
148 	peer = rxrpc_lookup_peer_local_rcu(local, skb, &srx);
149 	if (peer && !rxrpc_get_peer_maybe(peer, rxrpc_peer_get_input_error))
150 		peer = NULL;
151 	rcu_read_unlock();
152 	if (!peer)
153 		return;
154 
155 	trace_rxrpc_rx_icmp(peer, &serr->ee, &srx);
156 
157 	if ((serr->ee.ee_origin == SO_EE_ORIGIN_ICMP &&
158 	     serr->ee.ee_type == ICMP_DEST_UNREACH &&
159 	     serr->ee.ee_code == ICMP_FRAG_NEEDED)) {
160 		rxrpc_adjust_mtu(peer, serr->ee.ee_info);
161 		goto out;
162 	}
163 
164 	rxrpc_store_error(peer, skb);
165 out:
166 	rxrpc_put_peer(peer, rxrpc_peer_put_input_error);
167 }
168 
169 /*
170  * Map an error report to error codes on the peer record.
171  */
172 static void rxrpc_store_error(struct rxrpc_peer *peer, struct sk_buff *skb)
173 {
174 	enum rxrpc_call_completion compl = RXRPC_CALL_NETWORK_ERROR;
175 	struct sock_exterr_skb *serr = SKB_EXT_ERR(skb);
176 	struct sock_extended_err *ee = &serr->ee;
177 	int err = ee->ee_errno;
178 
179 	_enter("");
180 
181 	switch (ee->ee_origin) {
182 	case SO_EE_ORIGIN_NONE:
183 	case SO_EE_ORIGIN_LOCAL:
184 		compl = RXRPC_CALL_LOCAL_ERROR;
185 		break;
186 
187 	case SO_EE_ORIGIN_ICMP6:
188 		if (err == EACCES)
189 			err = EHOSTUNREACH;
190 		fallthrough;
191 	case SO_EE_ORIGIN_ICMP:
192 	default:
193 		break;
194 	}
195 
196 	rxrpc_distribute_error(peer, skb, compl, err);
197 }
198 
199 /*
200  * Distribute an error that occurred on a peer.
201  */
202 static void rxrpc_distribute_error(struct rxrpc_peer *peer, struct sk_buff *skb,
203 				   enum rxrpc_call_completion compl, int err)
204 {
205 	struct rxrpc_call *call;
206 	HLIST_HEAD(error_targets);
207 
208 	spin_lock(&peer->lock);
209 	hlist_move_list(&peer->error_targets, &error_targets);
210 
211 	while (!hlist_empty(&error_targets)) {
212 		call = hlist_entry(error_targets.first,
213 				   struct rxrpc_call, error_link);
214 		hlist_del_init(&call->error_link);
215 		spin_unlock(&peer->lock);
216 
217 		rxrpc_see_call(call, rxrpc_call_see_distribute_error);
218 		rxrpc_set_call_completion(call, compl, 0, -err);
219 		rxrpc_input_call_event(call, skb);
220 
221 		spin_lock(&peer->lock);
222 	}
223 
224 	spin_unlock(&peer->lock);
225 }
226 
227 /*
228  * Perform keep-alive pings.
229  */
230 static void rxrpc_peer_keepalive_dispatch(struct rxrpc_net *rxnet,
231 					  struct list_head *collector,
232 					  time64_t base,
233 					  u8 cursor)
234 {
235 	struct rxrpc_peer *peer;
236 	const u8 mask = ARRAY_SIZE(rxnet->peer_keepalive) - 1;
237 	time64_t keepalive_at;
238 	bool use;
239 	int slot;
240 
241 	spin_lock(&rxnet->peer_hash_lock);
242 
243 	while (!list_empty(collector)) {
244 		peer = list_entry(collector->next,
245 				  struct rxrpc_peer, keepalive_link);
246 
247 		list_del_init(&peer->keepalive_link);
248 		if (!rxrpc_get_peer_maybe(peer, rxrpc_peer_get_keepalive))
249 			continue;
250 
251 		use = __rxrpc_use_local(peer->local, rxrpc_local_use_peer_keepalive);
252 		spin_unlock(&rxnet->peer_hash_lock);
253 
254 		if (use) {
255 			keepalive_at = peer->last_tx_at + RXRPC_KEEPALIVE_TIME;
256 			slot = keepalive_at - base;
257 			_debug("%02x peer %u t=%d {%pISp}",
258 			       cursor, peer->debug_id, slot, &peer->srx.transport);
259 
260 			if (keepalive_at <= base ||
261 			    keepalive_at > base + RXRPC_KEEPALIVE_TIME) {
262 				rxrpc_send_keepalive(peer);
263 				slot = RXRPC_KEEPALIVE_TIME;
264 			}
265 
266 			/* A transmission to this peer occurred since last we
267 			 * examined it so put it into the appropriate future
268 			 * bucket.
269 			 */
270 			slot += cursor;
271 			slot &= mask;
272 			spin_lock(&rxnet->peer_hash_lock);
273 			list_add_tail(&peer->keepalive_link,
274 				      &rxnet->peer_keepalive[slot & mask]);
275 			spin_unlock(&rxnet->peer_hash_lock);
276 			rxrpc_unuse_local(peer->local, rxrpc_local_unuse_peer_keepalive);
277 		}
278 		rxrpc_put_peer(peer, rxrpc_peer_put_keepalive);
279 		spin_lock(&rxnet->peer_hash_lock);
280 	}
281 
282 	spin_unlock(&rxnet->peer_hash_lock);
283 }
284 
285 /*
286  * Perform keep-alive pings with VERSION packets to keep any NAT alive.
287  */
288 void rxrpc_peer_keepalive_worker(struct work_struct *work)
289 {
290 	struct rxrpc_net *rxnet =
291 		container_of(work, struct rxrpc_net, peer_keepalive_work);
292 	const u8 mask = ARRAY_SIZE(rxnet->peer_keepalive) - 1;
293 	time64_t base, now, delay;
294 	u8 cursor, stop;
295 	LIST_HEAD(collector);
296 
297 	now = ktime_get_seconds();
298 	base = rxnet->peer_keepalive_base;
299 	cursor = rxnet->peer_keepalive_cursor;
300 	_enter("%lld,%u", base - now, cursor);
301 
302 	if (!rxnet->live)
303 		return;
304 
305 	/* Remove to a temporary list all the peers that are currently lodged
306 	 * in expired buckets plus all new peers.
307 	 *
308 	 * Everything in the bucket at the cursor is processed this
309 	 * second; the bucket at cursor + 1 goes at now + 1s and so
310 	 * on...
311 	 */
312 	spin_lock(&rxnet->peer_hash_lock);
313 	list_splice_init(&rxnet->peer_keepalive_new, &collector);
314 
315 	stop = cursor + ARRAY_SIZE(rxnet->peer_keepalive);
316 	while (base <= now && (s8)(cursor - stop) < 0) {
317 		list_splice_tail_init(&rxnet->peer_keepalive[cursor & mask],
318 				      &collector);
319 		base++;
320 		cursor++;
321 	}
322 
323 	base = now;
324 	spin_unlock(&rxnet->peer_hash_lock);
325 
326 	rxnet->peer_keepalive_base = base;
327 	rxnet->peer_keepalive_cursor = cursor;
328 	rxrpc_peer_keepalive_dispatch(rxnet, &collector, base, cursor);
329 	ASSERT(list_empty(&collector));
330 
331 	/* Schedule the timer for the next occupied timeslot. */
332 	cursor = rxnet->peer_keepalive_cursor;
333 	stop = cursor + RXRPC_KEEPALIVE_TIME - 1;
334 	for (; (s8)(cursor - stop) < 0; cursor++) {
335 		if (!list_empty(&rxnet->peer_keepalive[cursor & mask]))
336 			break;
337 		base++;
338 	}
339 
340 	now = ktime_get_seconds();
341 	delay = base - now;
342 	if (delay < 1)
343 		delay = 1;
344 	delay *= HZ;
345 	if (rxnet->live)
346 		timer_reduce(&rxnet->peer_keepalive_timer, jiffies + delay);
347 
348 	_leave("");
349 }
350