1 /* Peer event handling, typically ICMP messages.
3 * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
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.
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>
18 #include <linux/in6.h>
19 #include <linux/icmp.h>
21 #include <net/af_rxrpc.h>
23 #include "ar-internal.h"
25 static void rxrpc_store_error(struct rxrpc_peer *, struct sock_exterr_skb *);
28 * Find the peer associated with an ICMP packet.
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)
34 struct sock_exterr_skb *serr = SKB_EXT_ERR(skb);
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;
43 /* Can we see an ICMP4 packet on an ICMP6 listening socket? and vice
46 switch (srx->transport.family) {
48 srx->transport.sin.sin_port = serr->port;
49 switch (serr->ee.ee_origin) {
50 case SO_EE_ORIGIN_ICMP:
52 memcpy(&srx->transport.sin.sin_addr,
53 skb_network_header(skb) + serr->addr_offset,
54 sizeof(struct in_addr));
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));
63 memcpy(&srx->transport.sin.sin_addr, &ip_hdr(skb)->saddr,
64 sizeof(struct in_addr));
69 #ifdef CONFIG_AF_RXRPC_IPV6
71 srx->transport.sin6.sin6_port = serr->port;
72 switch (serr->ee.ee_origin) {
73 case SO_EE_ORIGIN_ICMP6:
75 memcpy(&srx->transport.sin6.sin6_addr,
76 skb_network_header(skb) + serr->addr_offset,
77 sizeof(struct in6_addr));
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));
89 memcpy(&srx->transport.sin6.sin6_addr,
90 &ipv6_hdr(skb)->saddr,
91 sizeof(struct in6_addr));
101 return rxrpc_lookup_peer_rcu(local, srx);
105 * Handle an MTU/fragmentation problem.
107 static void rxrpc_adjust_mtu(struct rxrpc_peer *peer, struct sock_exterr_skb *serr)
109 u32 mtu = serr->ee.ee_info;
111 _net("Rx ICMP Fragmentation Needed (%d)", mtu);
113 /* wind down the local interface MTU */
114 if (mtu > 0 && peer->if_mtu == 65535 && mtu < peer->if_mtu) {
116 _net("I/F MTU %u", mtu);
120 /* they didn't give us a size, estimate one */
128 if (mtu < peer->hdrsize)
129 mtu = peer->hdrsize + 4;
133 if (mtu < peer->mtu) {
134 spin_lock_bh(&peer->lock);
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);
144 * Handle an error received on the local endpoint.
146 void rxrpc_error_report(struct sock *sk)
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;
154 _enter("%p{%d}", sk, local->debug_id);
156 skb = sock_dequeue_err_skb(sk);
158 _leave("UDP socket errqueue empty");
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);
170 peer = rxrpc_lookup_peer_icmp_rcu(local, skb, &srx);
171 if (peer && !rxrpc_get_peer_maybe(peer))
175 rxrpc_free_skb(skb, rxrpc_skb_rx_freed);
176 _leave(" [no peer]");
180 trace_rxrpc_rx_icmp(peer, &serr->ee, &srx);
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);
187 rxrpc_free_skb(skb, rxrpc_skb_rx_freed);
188 rxrpc_put_peer(peer);
189 _leave(" [MTU update]");
193 rxrpc_store_error(peer, serr);
195 rxrpc_free_skb(skb, rxrpc_skb_rx_freed);
197 /* The ref we obtained is passed off to the work item */
198 __rxrpc_queue_peer_error(peer);
203 * Map an error report to error codes on the peer record.
205 static void rxrpc_store_error(struct rxrpc_peer *peer,
206 struct sock_exterr_skb *serr)
208 struct sock_extended_err *ee;
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");
225 case ICMP_HOST_UNREACH:
226 _net("Rx Received ICMP Host Unreachable");
228 case ICMP_PORT_UNREACH:
229 _net("Rx Received ICMP Port Unreachable");
231 case ICMP_NET_UNKNOWN:
232 _net("Rx Received ICMP Unknown Network");
234 case ICMP_HOST_UNKNOWN:
235 _net("Rx Received ICMP Unknown Host");
238 _net("Rx Received ICMP DestUnreach code=%u",
244 case ICMP_TIME_EXCEEDED:
245 _net("Rx Received ICMP TTL Exceeded");
249 _proto("Rx Received ICMP error { type=%u code=%u }",
250 ee->ee_type, ee->ee_code);
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;
261 case SO_EE_ORIGIN_ICMP6:
263 _proto("Rx Received error report { orig=%u }", ee->ee_origin);
267 peer->error_report = err;
271 * Distribute an error that occurred on a peer
273 void rxrpc_peer_error_distributor(struct work_struct *work)
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;
283 error = READ_ONCE(peer->error_report);
284 if (error < RXRPC_LOCAL_ERROR_OFFSET) {
285 compl = RXRPC_CALL_NETWORK_ERROR;
287 compl = RXRPC_CALL_LOCAL_ERROR;
288 error -= RXRPC_LOCAL_ERROR_OFFSET;
291 _debug("ISSUE ERROR %s %d", rxrpc_call_completions[compl], error);
293 spin_lock_bh(&peer->lock);
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);
301 if (rxrpc_set_call_completion(call, compl, 0, -error))
302 rxrpc_notify_socket(call);
305 spin_unlock_bh(&peer->lock);
307 rxrpc_put_peer(peer);
312 * Add RTT information to cache. This is called in softirq mode and has
313 * exclusive access to the peer RTT data.
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)
319 struct rxrpc_peer *peer = call->peer;
321 u64 sum = peer->rtt_sum, avg;
322 u8 cursor = peer->rtt_cursor, usage = peer->rtt_usage;
324 rtt = ktime_to_ns(ktime_sub(resp_time, send_time));
328 /* Replace the oldest datum in the RTT buffer */
329 sum -= peer->rtt_cache[cursor];
331 peer->rtt_cache[cursor] = rtt;
332 peer->rtt_cursor = (cursor + 1) & (RXRPC_RTT_CACHE_SIZE - 1);
334 if (usage < RXRPC_RTT_CACHE_SIZE) {
336 peer->rtt_usage = usage;
339 /* Now recalculate the average */
340 if (usage == RXRPC_RTT_CACHE_SIZE) {
341 avg = sum / RXRPC_RTT_CACHE_SIZE;
348 trace_rxrpc_rtt_rx(call, why, send_serial, resp_serial, rtt,
353 * Perform keep-alive pings.
355 static void rxrpc_peer_keepalive_dispatch(struct rxrpc_net *rxnet,
356 struct list_head *collector,
360 struct rxrpc_peer *peer;
361 const u8 mask = ARRAY_SIZE(rxnet->peer_keepalive) - 1;
362 time64_t keepalive_at;
365 spin_lock_bh(&rxnet->peer_hash_lock);
367 while (!list_empty(collector)) {
368 peer = list_entry(collector->next,
369 struct rxrpc_peer, keepalive_link);
371 list_del_init(&peer->keepalive_link);
372 if (!rxrpc_get_peer_maybe(peer))
375 spin_unlock_bh(&rxnet->peer_hash_lock);
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);
382 if (keepalive_at <= base ||
383 keepalive_at > base + RXRPC_KEEPALIVE_TIME) {
384 rxrpc_send_keepalive(peer);
385 slot = RXRPC_KEEPALIVE_TIME;
388 /* A transmission to this peer occurred since last we examined
389 * it so put it into the appropriate future bucket.
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);
399 spin_unlock_bh(&rxnet->peer_hash_lock);
403 * Perform keep-alive pings with VERSION packets to keep any NAT alive.
405 void rxrpc_peer_keepalive_worker(struct work_struct *work)
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;
412 LIST_HEAD(collector);
414 now = ktime_get_seconds();
415 base = rxnet->peer_keepalive_base;
416 cursor = rxnet->peer_keepalive_cursor;
417 _enter("%lld,%u", base - now, cursor);
422 /* Remove to a temporary list all the peers that are currently lodged
423 * in expired buckets plus all new peers.
425 * Everything in the bucket at the cursor is processed this
426 * second; the bucket at cursor + 1 goes at now + 1s and so
429 spin_lock_bh(&rxnet->peer_hash_lock);
430 list_splice_init(&rxnet->peer_keepalive_new, &collector);
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],
441 spin_unlock_bh(&rxnet->peer_hash_lock);
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));
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]))
457 now = ktime_get_seconds();
463 timer_reduce(&rxnet->peer_keepalive_timer, jiffies + delay);