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 static void rxrpc_distribute_error(struct rxrpc_peer *, int,
27 enum rxrpc_call_completion);
28
29 /*
30 * Find the peer associated with an ICMP packet.
31 */
rxrpc_lookup_peer_icmp_rcu(struct rxrpc_local * local,const struct sk_buff * skb,struct sockaddr_rxrpc * srx)32 static struct rxrpc_peer *rxrpc_lookup_peer_icmp_rcu(struct rxrpc_local *local,
33 const struct sk_buff *skb,
34 struct sockaddr_rxrpc *srx)
35 {
36 struct sock_exterr_skb *serr = SKB_EXT_ERR(skb);
37
38 _enter("");
39
40 memset(srx, 0, sizeof(*srx));
41 srx->transport_type = local->srx.transport_type;
42 srx->transport_len = local->srx.transport_len;
43 srx->transport.family = local->srx.transport.family;
44
45 /* Can we see an ICMP4 packet on an ICMP6 listening socket? and vice
46 * versa?
47 */
48 switch (srx->transport.family) {
49 case AF_INET:
50 srx->transport.sin.sin_port = serr->port;
51 switch (serr->ee.ee_origin) {
52 case SO_EE_ORIGIN_ICMP:
53 _net("Rx ICMP");
54 memcpy(&srx->transport.sin.sin_addr,
55 skb_network_header(skb) + serr->addr_offset,
56 sizeof(struct in_addr));
57 break;
58 case SO_EE_ORIGIN_ICMP6:
59 _net("Rx ICMP6 on v4 sock");
60 memcpy(&srx->transport.sin.sin_addr,
61 skb_network_header(skb) + serr->addr_offset + 12,
62 sizeof(struct in_addr));
63 break;
64 default:
65 memcpy(&srx->transport.sin.sin_addr, &ip_hdr(skb)->saddr,
66 sizeof(struct in_addr));
67 break;
68 }
69 break;
70
71 #ifdef CONFIG_AF_RXRPC_IPV6
72 case AF_INET6:
73 srx->transport.sin6.sin6_port = serr->port;
74 switch (serr->ee.ee_origin) {
75 case SO_EE_ORIGIN_ICMP6:
76 _net("Rx ICMP6");
77 memcpy(&srx->transport.sin6.sin6_addr,
78 skb_network_header(skb) + serr->addr_offset,
79 sizeof(struct in6_addr));
80 break;
81 case SO_EE_ORIGIN_ICMP:
82 _net("Rx ICMP on v6 sock");
83 srx->transport.sin6.sin6_addr.s6_addr32[0] = 0;
84 srx->transport.sin6.sin6_addr.s6_addr32[1] = 0;
85 srx->transport.sin6.sin6_addr.s6_addr32[2] = htonl(0xffff);
86 memcpy(srx->transport.sin6.sin6_addr.s6_addr + 12,
87 skb_network_header(skb) + serr->addr_offset,
88 sizeof(struct in_addr));
89 break;
90 default:
91 memcpy(&srx->transport.sin6.sin6_addr,
92 &ipv6_hdr(skb)->saddr,
93 sizeof(struct in6_addr));
94 break;
95 }
96 break;
97 #endif
98
99 default:
100 BUG();
101 }
102
103 return rxrpc_lookup_peer_rcu(local, srx);
104 }
105
106 /*
107 * Handle an MTU/fragmentation problem.
108 */
rxrpc_adjust_mtu(struct rxrpc_peer * peer,struct sock_exterr_skb * serr)109 static void rxrpc_adjust_mtu(struct rxrpc_peer *peer, struct sock_exterr_skb *serr)
110 {
111 u32 mtu = serr->ee.ee_info;
112
113 _net("Rx ICMP Fragmentation Needed (%d)", mtu);
114
115 /* wind down the local interface MTU */
116 if (mtu > 0 && peer->if_mtu == 65535 && mtu < peer->if_mtu) {
117 peer->if_mtu = mtu;
118 _net("I/F MTU %u", mtu);
119 }
120
121 if (mtu == 0) {
122 /* they didn't give us a size, estimate one */
123 mtu = peer->if_mtu;
124 if (mtu > 1500) {
125 mtu >>= 1;
126 if (mtu < 1500)
127 mtu = 1500;
128 } else {
129 mtu -= 100;
130 if (mtu < peer->hdrsize)
131 mtu = peer->hdrsize + 4;
132 }
133 }
134
135 if (mtu < peer->mtu) {
136 spin_lock_bh(&peer->lock);
137 peer->mtu = mtu;
138 peer->maxdata = peer->mtu - peer->hdrsize;
139 spin_unlock_bh(&peer->lock);
140 _net("Net MTU %u (maxdata %u)",
141 peer->mtu, peer->maxdata);
142 }
143 }
144
145 /*
146 * Handle an error received on the local endpoint.
147 */
rxrpc_error_report(struct sock * sk)148 void rxrpc_error_report(struct sock *sk)
149 {
150 struct sock_exterr_skb *serr;
151 struct sockaddr_rxrpc srx;
152 struct rxrpc_local *local = sk->sk_user_data;
153 struct rxrpc_peer *peer;
154 struct sk_buff *skb;
155
156 _enter("%p{%d}", sk, local->debug_id);
157
158 skb = sock_dequeue_err_skb(sk);
159 if (!skb) {
160 _leave("UDP socket errqueue empty");
161 return;
162 }
163 rxrpc_new_skb(skb, rxrpc_skb_rx_received);
164 serr = SKB_EXT_ERR(skb);
165 if (!skb->len && serr->ee.ee_origin == SO_EE_ORIGIN_TIMESTAMPING) {
166 _leave("UDP empty message");
167 rxrpc_free_skb(skb, rxrpc_skb_rx_freed);
168 return;
169 }
170
171 rcu_read_lock();
172 peer = rxrpc_lookup_peer_icmp_rcu(local, skb, &srx);
173 if (peer && !rxrpc_get_peer_maybe(peer))
174 peer = NULL;
175 if (!peer) {
176 rcu_read_unlock();
177 rxrpc_free_skb(skb, rxrpc_skb_rx_freed);
178 _leave(" [no peer]");
179 return;
180 }
181
182 trace_rxrpc_rx_icmp(peer, &serr->ee, &srx);
183
184 if ((serr->ee.ee_origin == SO_EE_ORIGIN_ICMP &&
185 serr->ee.ee_type == ICMP_DEST_UNREACH &&
186 serr->ee.ee_code == ICMP_FRAG_NEEDED)) {
187 rxrpc_adjust_mtu(peer, serr);
188 rcu_read_unlock();
189 rxrpc_free_skb(skb, rxrpc_skb_rx_freed);
190 rxrpc_put_peer(peer);
191 _leave(" [MTU update]");
192 return;
193 }
194
195 rxrpc_store_error(peer, serr);
196 rcu_read_unlock();
197 rxrpc_free_skb(skb, rxrpc_skb_rx_freed);
198 rxrpc_put_peer(peer);
199
200 _leave("");
201 }
202
203 /*
204 * Map an error report to error codes on the peer record.
205 */
rxrpc_store_error(struct rxrpc_peer * peer,struct sock_exterr_skb * serr)206 static void rxrpc_store_error(struct rxrpc_peer *peer,
207 struct sock_exterr_skb *serr)
208 {
209 enum rxrpc_call_completion compl = RXRPC_CALL_NETWORK_ERROR;
210 struct sock_extended_err *ee;
211 int err;
212
213 _enter("");
214
215 ee = &serr->ee;
216
217 err = ee->ee_errno;
218
219 switch (ee->ee_origin) {
220 case SO_EE_ORIGIN_ICMP:
221 switch (ee->ee_type) {
222 case ICMP_DEST_UNREACH:
223 switch (ee->ee_code) {
224 case ICMP_NET_UNREACH:
225 _net("Rx Received ICMP Network Unreachable");
226 break;
227 case ICMP_HOST_UNREACH:
228 _net("Rx Received ICMP Host Unreachable");
229 break;
230 case ICMP_PORT_UNREACH:
231 _net("Rx Received ICMP Port Unreachable");
232 break;
233 case ICMP_NET_UNKNOWN:
234 _net("Rx Received ICMP Unknown Network");
235 break;
236 case ICMP_HOST_UNKNOWN:
237 _net("Rx Received ICMP Unknown Host");
238 break;
239 default:
240 _net("Rx Received ICMP DestUnreach code=%u",
241 ee->ee_code);
242 break;
243 }
244 break;
245
246 case ICMP_TIME_EXCEEDED:
247 _net("Rx Received ICMP TTL Exceeded");
248 break;
249
250 default:
251 _proto("Rx Received ICMP error { type=%u code=%u }",
252 ee->ee_type, ee->ee_code);
253 break;
254 }
255 break;
256
257 case SO_EE_ORIGIN_NONE:
258 case SO_EE_ORIGIN_LOCAL:
259 _proto("Rx Received local error { error=%d }", err);
260 compl = RXRPC_CALL_LOCAL_ERROR;
261 break;
262
263 case SO_EE_ORIGIN_ICMP6:
264 default:
265 _proto("Rx Received error report { orig=%u }", ee->ee_origin);
266 break;
267 }
268
269 rxrpc_distribute_error(peer, err, compl);
270 }
271
272 /*
273 * Distribute an error that occurred on a peer.
274 */
rxrpc_distribute_error(struct rxrpc_peer * peer,int error,enum rxrpc_call_completion compl)275 static void rxrpc_distribute_error(struct rxrpc_peer *peer, int error,
276 enum rxrpc_call_completion compl)
277 {
278 struct rxrpc_call *call;
279
280 hlist_for_each_entry_rcu(call, &peer->error_targets, error_link) {
281 rxrpc_see_call(call);
282 if (call->state < RXRPC_CALL_COMPLETE &&
283 rxrpc_set_call_completion(call, compl, 0, -error))
284 rxrpc_notify_socket(call);
285 }
286 }
287
288 /*
289 * Add RTT information to cache. This is called in softirq mode and has
290 * exclusive access to the peer RTT data.
291 */
rxrpc_peer_add_rtt(struct rxrpc_call * call,enum rxrpc_rtt_rx_trace why,rxrpc_serial_t send_serial,rxrpc_serial_t resp_serial,ktime_t send_time,ktime_t resp_time)292 void rxrpc_peer_add_rtt(struct rxrpc_call *call, enum rxrpc_rtt_rx_trace why,
293 rxrpc_serial_t send_serial, rxrpc_serial_t resp_serial,
294 ktime_t send_time, ktime_t resp_time)
295 {
296 struct rxrpc_peer *peer = call->peer;
297 s64 rtt;
298 u64 sum = peer->rtt_sum, avg;
299 u8 cursor = peer->rtt_cursor, usage = peer->rtt_usage;
300
301 rtt = ktime_to_ns(ktime_sub(resp_time, send_time));
302 if (rtt < 0)
303 return;
304
305 spin_lock(&peer->rtt_input_lock);
306
307 /* Replace the oldest datum in the RTT buffer */
308 sum -= peer->rtt_cache[cursor];
309 sum += rtt;
310 peer->rtt_cache[cursor] = rtt;
311 peer->rtt_cursor = (cursor + 1) & (RXRPC_RTT_CACHE_SIZE - 1);
312 peer->rtt_sum = sum;
313 if (usage < RXRPC_RTT_CACHE_SIZE) {
314 usage++;
315 peer->rtt_usage = usage;
316 }
317
318 spin_unlock(&peer->rtt_input_lock);
319
320 /* Now recalculate the average */
321 if (usage == RXRPC_RTT_CACHE_SIZE) {
322 avg = sum / RXRPC_RTT_CACHE_SIZE;
323 } else {
324 avg = sum;
325 do_div(avg, usage);
326 }
327
328 /* Don't need to update this under lock */
329 peer->rtt = avg;
330 trace_rxrpc_rtt_rx(call, why, send_serial, resp_serial, rtt,
331 usage, avg);
332 }
333
334 /*
335 * Perform keep-alive pings.
336 */
rxrpc_peer_keepalive_dispatch(struct rxrpc_net * rxnet,struct list_head * collector,time64_t base,u8 cursor)337 static void rxrpc_peer_keepalive_dispatch(struct rxrpc_net *rxnet,
338 struct list_head *collector,
339 time64_t base,
340 u8 cursor)
341 {
342 struct rxrpc_peer *peer;
343 const u8 mask = ARRAY_SIZE(rxnet->peer_keepalive) - 1;
344 time64_t keepalive_at;
345 int slot;
346
347 spin_lock_bh(&rxnet->peer_hash_lock);
348
349 while (!list_empty(collector)) {
350 peer = list_entry(collector->next,
351 struct rxrpc_peer, keepalive_link);
352
353 list_del_init(&peer->keepalive_link);
354 if (!rxrpc_get_peer_maybe(peer))
355 continue;
356
357 spin_unlock_bh(&rxnet->peer_hash_lock);
358
359 keepalive_at = peer->last_tx_at + RXRPC_KEEPALIVE_TIME;
360 slot = keepalive_at - base;
361 _debug("%02x peer %u t=%d {%pISp}",
362 cursor, peer->debug_id, slot, &peer->srx.transport);
363
364 if (keepalive_at <= base ||
365 keepalive_at > base + RXRPC_KEEPALIVE_TIME) {
366 rxrpc_send_keepalive(peer);
367 slot = RXRPC_KEEPALIVE_TIME;
368 }
369
370 /* A transmission to this peer occurred since last we examined
371 * it so put it into the appropriate future bucket.
372 */
373 slot += cursor;
374 slot &= mask;
375 spin_lock_bh(&rxnet->peer_hash_lock);
376 list_add_tail(&peer->keepalive_link,
377 &rxnet->peer_keepalive[slot & mask]);
378 rxrpc_put_peer(peer);
379 }
380
381 spin_unlock_bh(&rxnet->peer_hash_lock);
382 }
383
384 /*
385 * Perform keep-alive pings with VERSION packets to keep any NAT alive.
386 */
rxrpc_peer_keepalive_worker(struct work_struct * work)387 void rxrpc_peer_keepalive_worker(struct work_struct *work)
388 {
389 struct rxrpc_net *rxnet =
390 container_of(work, struct rxrpc_net, peer_keepalive_work);
391 const u8 mask = ARRAY_SIZE(rxnet->peer_keepalive) - 1;
392 time64_t base, now, delay;
393 u8 cursor, stop;
394 LIST_HEAD(collector);
395
396 now = ktime_get_seconds();
397 base = rxnet->peer_keepalive_base;
398 cursor = rxnet->peer_keepalive_cursor;
399 _enter("%lld,%u", base - now, cursor);
400
401 if (!rxnet->live)
402 return;
403
404 /* Remove to a temporary list all the peers that are currently lodged
405 * in expired buckets plus all new peers.
406 *
407 * Everything in the bucket at the cursor is processed this
408 * second; the bucket at cursor + 1 goes at now + 1s and so
409 * on...
410 */
411 spin_lock_bh(&rxnet->peer_hash_lock);
412 list_splice_init(&rxnet->peer_keepalive_new, &collector);
413
414 stop = cursor + ARRAY_SIZE(rxnet->peer_keepalive);
415 while (base <= now && (s8)(cursor - stop) < 0) {
416 list_splice_tail_init(&rxnet->peer_keepalive[cursor & mask],
417 &collector);
418 base++;
419 cursor++;
420 }
421
422 base = now;
423 spin_unlock_bh(&rxnet->peer_hash_lock);
424
425 rxnet->peer_keepalive_base = base;
426 rxnet->peer_keepalive_cursor = cursor;
427 rxrpc_peer_keepalive_dispatch(rxnet, &collector, base, cursor);
428 ASSERT(list_empty(&collector));
429
430 /* Schedule the timer for the next occupied timeslot. */
431 cursor = rxnet->peer_keepalive_cursor;
432 stop = cursor + RXRPC_KEEPALIVE_TIME - 1;
433 for (; (s8)(cursor - stop) < 0; cursor++) {
434 if (!list_empty(&rxnet->peer_keepalive[cursor & mask]))
435 break;
436 base++;
437 }
438
439 now = ktime_get_seconds();
440 delay = base - now;
441 if (delay < 1)
442 delay = 1;
443 delay *= HZ;
444 if (rxnet->live)
445 timer_reduce(&rxnet->peer_keepalive_timer, jiffies + delay);
446
447 _leave("");
448 }
449
