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 if (unlikely(!local))
157 return;
158
159 _enter("%p{%d}", sk, local->debug_id);
160
161 skb = sock_dequeue_err_skb(sk);
162 if (!skb) {
163 _leave("UDP socket errqueue empty");
164 return;
165 }
166 rxrpc_new_skb(skb, rxrpc_skb_rx_received);
167 serr = SKB_EXT_ERR(skb);
168 if (!skb->len && serr->ee.ee_origin == SO_EE_ORIGIN_TIMESTAMPING) {
169 _leave("UDP empty message");
170 rxrpc_free_skb(skb, rxrpc_skb_rx_freed);
171 return;
172 }
173
174 rcu_read_lock();
175 peer = rxrpc_lookup_peer_icmp_rcu(local, skb, &srx);
176 if (peer && !rxrpc_get_peer_maybe(peer))
177 peer = NULL;
178 if (!peer) {
179 rcu_read_unlock();
180 rxrpc_free_skb(skb, rxrpc_skb_rx_freed);
181 _leave(" [no peer]");
182 return;
183 }
184
185 trace_rxrpc_rx_icmp(peer, &serr->ee, &srx);
186
187 if ((serr->ee.ee_origin == SO_EE_ORIGIN_ICMP &&
188 serr->ee.ee_type == ICMP_DEST_UNREACH &&
189 serr->ee.ee_code == ICMP_FRAG_NEEDED)) {
190 rxrpc_adjust_mtu(peer, serr);
191 rcu_read_unlock();
192 rxrpc_free_skb(skb, rxrpc_skb_rx_freed);
193 rxrpc_put_peer(peer);
194 _leave(" [MTU update]");
195 return;
196 }
197
198 rxrpc_store_error(peer, serr);
199 rcu_read_unlock();
200 rxrpc_free_skb(skb, rxrpc_skb_rx_freed);
201 rxrpc_put_peer(peer);
202
203 _leave("");
204 }
205
206 /*
207 * Map an error report to error codes on the peer record.
208 */
rxrpc_store_error(struct rxrpc_peer * peer,struct sock_exterr_skb * serr)209 static void rxrpc_store_error(struct rxrpc_peer *peer,
210 struct sock_exterr_skb *serr)
211 {
212 enum rxrpc_call_completion compl = RXRPC_CALL_NETWORK_ERROR;
213 struct sock_extended_err *ee;
214 int err;
215
216 _enter("");
217
218 ee = &serr->ee;
219
220 err = ee->ee_errno;
221
222 switch (ee->ee_origin) {
223 case SO_EE_ORIGIN_ICMP:
224 switch (ee->ee_type) {
225 case ICMP_DEST_UNREACH:
226 switch (ee->ee_code) {
227 case ICMP_NET_UNREACH:
228 _net("Rx Received ICMP Network Unreachable");
229 break;
230 case ICMP_HOST_UNREACH:
231 _net("Rx Received ICMP Host Unreachable");
232 break;
233 case ICMP_PORT_UNREACH:
234 _net("Rx Received ICMP Port Unreachable");
235 break;
236 case ICMP_NET_UNKNOWN:
237 _net("Rx Received ICMP Unknown Network");
238 break;
239 case ICMP_HOST_UNKNOWN:
240 _net("Rx Received ICMP Unknown Host");
241 break;
242 default:
243 _net("Rx Received ICMP DestUnreach code=%u",
244 ee->ee_code);
245 break;
246 }
247 break;
248
249 case ICMP_TIME_EXCEEDED:
250 _net("Rx Received ICMP TTL Exceeded");
251 break;
252
253 default:
254 _proto("Rx Received ICMP error { type=%u code=%u }",
255 ee->ee_type, ee->ee_code);
256 break;
257 }
258 break;
259
260 case SO_EE_ORIGIN_NONE:
261 case SO_EE_ORIGIN_LOCAL:
262 _proto("Rx Received local error { error=%d }", err);
263 compl = RXRPC_CALL_LOCAL_ERROR;
264 break;
265
266 case SO_EE_ORIGIN_ICMP6:
267 default:
268 _proto("Rx Received error report { orig=%u }", ee->ee_origin);
269 break;
270 }
271
272 rxrpc_distribute_error(peer, err, compl);
273 }
274
275 /*
276 * Distribute an error that occurred on a peer.
277 */
rxrpc_distribute_error(struct rxrpc_peer * peer,int error,enum rxrpc_call_completion compl)278 static void rxrpc_distribute_error(struct rxrpc_peer *peer, int error,
279 enum rxrpc_call_completion compl)
280 {
281 struct rxrpc_call *call;
282
283 hlist_for_each_entry_rcu(call, &peer->error_targets, error_link) {
284 rxrpc_see_call(call);
285 if (call->state < RXRPC_CALL_COMPLETE &&
286 rxrpc_set_call_completion(call, compl, 0, -error))
287 rxrpc_notify_socket(call);
288 }
289 }
290
291 /*
292 * Add RTT information to cache. This is called in softirq mode and has
293 * exclusive access to the peer RTT data.
294 */
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)295 void rxrpc_peer_add_rtt(struct rxrpc_call *call, enum rxrpc_rtt_rx_trace why,
296 rxrpc_serial_t send_serial, rxrpc_serial_t resp_serial,
297 ktime_t send_time, ktime_t resp_time)
298 {
299 struct rxrpc_peer *peer = call->peer;
300 s64 rtt;
301 u64 sum = peer->rtt_sum, avg;
302 u8 cursor = peer->rtt_cursor, usage = peer->rtt_usage;
303
304 rtt = ktime_to_ns(ktime_sub(resp_time, send_time));
305 if (rtt < 0)
306 return;
307
308 spin_lock(&peer->rtt_input_lock);
309
310 /* Replace the oldest datum in the RTT buffer */
311 sum -= peer->rtt_cache[cursor];
312 sum += rtt;
313 peer->rtt_cache[cursor] = rtt;
314 peer->rtt_cursor = (cursor + 1) & (RXRPC_RTT_CACHE_SIZE - 1);
315 peer->rtt_sum = sum;
316 if (usage < RXRPC_RTT_CACHE_SIZE) {
317 usage++;
318 peer->rtt_usage = usage;
319 }
320
321 spin_unlock(&peer->rtt_input_lock);
322
323 /* Now recalculate the average */
324 if (usage == RXRPC_RTT_CACHE_SIZE) {
325 avg = sum / RXRPC_RTT_CACHE_SIZE;
326 } else {
327 avg = sum;
328 do_div(avg, usage);
329 }
330
331 /* Don't need to update this under lock */
332 peer->rtt = avg;
333 trace_rxrpc_rtt_rx(call, why, send_serial, resp_serial, rtt,
334 usage, avg);
335 }
336
337 /*
338 * Perform keep-alive pings.
339 */
rxrpc_peer_keepalive_dispatch(struct rxrpc_net * rxnet,struct list_head * collector,time64_t base,u8 cursor)340 static void rxrpc_peer_keepalive_dispatch(struct rxrpc_net *rxnet,
341 struct list_head *collector,
342 time64_t base,
343 u8 cursor)
344 {
345 struct rxrpc_peer *peer;
346 const u8 mask = ARRAY_SIZE(rxnet->peer_keepalive) - 1;
347 time64_t keepalive_at;
348 int slot;
349
350 spin_lock_bh(&rxnet->peer_hash_lock);
351
352 while (!list_empty(collector)) {
353 peer = list_entry(collector->next,
354 struct rxrpc_peer, keepalive_link);
355
356 list_del_init(&peer->keepalive_link);
357 if (!rxrpc_get_peer_maybe(peer))
358 continue;
359
360 if (__rxrpc_use_local(peer->local)) {
361 spin_unlock_bh(&rxnet->peer_hash_lock);
362
363 keepalive_at = peer->last_tx_at + RXRPC_KEEPALIVE_TIME;
364 slot = keepalive_at - base;
365 _debug("%02x peer %u t=%d {%pISp}",
366 cursor, peer->debug_id, slot, &peer->srx.transport);
367
368 if (keepalive_at <= base ||
369 keepalive_at > base + RXRPC_KEEPALIVE_TIME) {
370 rxrpc_send_keepalive(peer);
371 slot = RXRPC_KEEPALIVE_TIME;
372 }
373
374 /* A transmission to this peer occurred since last we
375 * examined it so put it into the appropriate future
376 * bucket.
377 */
378 slot += cursor;
379 slot &= mask;
380 spin_lock_bh(&rxnet->peer_hash_lock);
381 list_add_tail(&peer->keepalive_link,
382 &rxnet->peer_keepalive[slot & mask]);
383 rxrpc_unuse_local(peer->local);
384 }
385 rxrpc_put_peer_locked(peer);
386 }
387
388 spin_unlock_bh(&rxnet->peer_hash_lock);
389 }
390
391 /*
392 * Perform keep-alive pings with VERSION packets to keep any NAT alive.
393 */
rxrpc_peer_keepalive_worker(struct work_struct * work)394 void rxrpc_peer_keepalive_worker(struct work_struct *work)
395 {
396 struct rxrpc_net *rxnet =
397 container_of(work, struct rxrpc_net, peer_keepalive_work);
398 const u8 mask = ARRAY_SIZE(rxnet->peer_keepalive) - 1;
399 time64_t base, now, delay;
400 u8 cursor, stop;
401 LIST_HEAD(collector);
402
403 now = ktime_get_seconds();
404 base = rxnet->peer_keepalive_base;
405 cursor = rxnet->peer_keepalive_cursor;
406 _enter("%lld,%u", base - now, cursor);
407
408 if (!rxnet->live)
409 return;
410
411 /* Remove to a temporary list all the peers that are currently lodged
412 * in expired buckets plus all new peers.
413 *
414 * Everything in the bucket at the cursor is processed this
415 * second; the bucket at cursor + 1 goes at now + 1s and so
416 * on...
417 */
418 spin_lock_bh(&rxnet->peer_hash_lock);
419 list_splice_init(&rxnet->peer_keepalive_new, &collector);
420
421 stop = cursor + ARRAY_SIZE(rxnet->peer_keepalive);
422 while (base <= now && (s8)(cursor - stop) < 0) {
423 list_splice_tail_init(&rxnet->peer_keepalive[cursor & mask],
424 &collector);
425 base++;
426 cursor++;
427 }
428
429 base = now;
430 spin_unlock_bh(&rxnet->peer_hash_lock);
431
432 rxnet->peer_keepalive_base = base;
433 rxnet->peer_keepalive_cursor = cursor;
434 rxrpc_peer_keepalive_dispatch(rxnet, &collector, base, cursor);
435 ASSERT(list_empty(&collector));
436
437 /* Schedule the timer for the next occupied timeslot. */
438 cursor = rxnet->peer_keepalive_cursor;
439 stop = cursor + RXRPC_KEEPALIVE_TIME - 1;
440 for (; (s8)(cursor - stop) < 0; cursor++) {
441 if (!list_empty(&rxnet->peer_keepalive[cursor & mask]))
442 break;
443 base++;
444 }
445
446 now = ktime_get_seconds();
447 delay = base - now;
448 if (delay < 1)
449 delay = 1;
450 delay *= HZ;
451 if (rxnet->live)
452 timer_reduce(&rxnet->peer_keepalive_timer, jiffies + delay);
453
454 _leave("");
455 }
456