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 *);
26 static void rxrpc_distribute_error(struct rxrpc_peer *, int,
27 enum rxrpc_call_completion);
30 * Find the peer associated with an ICMP packet.
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)
36 struct sock_exterr_skb *serr = SKB_EXT_ERR(skb);
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;
45 /* Can we see an ICMP4 packet on an ICMP6 listening socket? and vice
48 switch (srx->transport.family) {
50 srx->transport_len = sizeof(srx->transport.sin);
51 srx->transport.family = AF_INET;
52 srx->transport.sin.sin_port = serr->port;
53 switch (serr->ee.ee_origin) {
54 case SO_EE_ORIGIN_ICMP:
56 memcpy(&srx->transport.sin.sin_addr,
57 skb_network_header(skb) + serr->addr_offset,
58 sizeof(struct in_addr));
60 case SO_EE_ORIGIN_ICMP6:
61 _net("Rx ICMP6 on v4 sock");
62 memcpy(&srx->transport.sin.sin_addr,
63 skb_network_header(skb) + serr->addr_offset + 12,
64 sizeof(struct in_addr));
67 memcpy(&srx->transport.sin.sin_addr, &ip_hdr(skb)->saddr,
68 sizeof(struct in_addr));
73 #ifdef CONFIG_AF_RXRPC_IPV6
75 switch (serr->ee.ee_origin) {
76 case SO_EE_ORIGIN_ICMP6:
78 srx->transport.sin6.sin6_port = serr->port;
79 memcpy(&srx->transport.sin6.sin6_addr,
80 skb_network_header(skb) + serr->addr_offset,
81 sizeof(struct in6_addr));
83 case SO_EE_ORIGIN_ICMP:
84 _net("Rx ICMP on v6 sock");
85 srx->transport_len = sizeof(srx->transport.sin);
86 srx->transport.family = AF_INET;
87 srx->transport.sin.sin_port = serr->port;
88 memcpy(&srx->transport.sin.sin_addr,
89 skb_network_header(skb) + serr->addr_offset,
90 sizeof(struct in_addr));
93 memcpy(&srx->transport.sin6.sin6_addr,
94 &ipv6_hdr(skb)->saddr,
95 sizeof(struct in6_addr));
105 return rxrpc_lookup_peer_rcu(local, srx);
109 * Handle an MTU/fragmentation problem.
111 static void rxrpc_adjust_mtu(struct rxrpc_peer *peer, struct sock_exterr_skb *serr)
113 u32 mtu = serr->ee.ee_info;
115 _net("Rx ICMP Fragmentation Needed (%d)", mtu);
117 /* wind down the local interface MTU */
118 if (mtu > 0 && peer->if_mtu == 65535 && mtu < peer->if_mtu) {
120 _net("I/F MTU %u", mtu);
124 /* they didn't give us a size, estimate one */
132 if (mtu < peer->hdrsize)
133 mtu = peer->hdrsize + 4;
137 if (mtu < peer->mtu) {
138 spin_lock_bh(&peer->lock);
140 peer->maxdata = peer->mtu - peer->hdrsize;
141 spin_unlock_bh(&peer->lock);
142 _net("Net MTU %u (maxdata %u)",
143 peer->mtu, peer->maxdata);
148 * Handle an error received on the local endpoint.
150 void rxrpc_error_report(struct sock *sk)
152 struct sock_exterr_skb *serr;
153 struct sockaddr_rxrpc srx;
154 struct rxrpc_local *local = sk->sk_user_data;
155 struct rxrpc_peer *peer;
158 _enter("%p{%d}", sk, local->debug_id);
160 /* Clear the outstanding error value on the socket so that it doesn't
161 * cause kernel_sendmsg() to return it later.
165 skb = sock_dequeue_err_skb(sk);
167 _leave("UDP socket errqueue empty");
170 rxrpc_new_skb(skb, rxrpc_skb_rx_received);
171 serr = SKB_EXT_ERR(skb);
172 if (!skb->len && serr->ee.ee_origin == SO_EE_ORIGIN_TIMESTAMPING) {
173 _leave("UDP empty message");
174 rxrpc_free_skb(skb, rxrpc_skb_rx_freed);
179 peer = rxrpc_lookup_peer_icmp_rcu(local, skb, &srx);
180 if (peer && !rxrpc_get_peer_maybe(peer))
184 rxrpc_free_skb(skb, rxrpc_skb_rx_freed);
185 _leave(" [no peer]");
189 trace_rxrpc_rx_icmp(peer, &serr->ee, &srx);
191 if ((serr->ee.ee_origin == SO_EE_ORIGIN_ICMP &&
192 serr->ee.ee_type == ICMP_DEST_UNREACH &&
193 serr->ee.ee_code == ICMP_FRAG_NEEDED)) {
194 rxrpc_adjust_mtu(peer, serr);
196 rxrpc_free_skb(skb, rxrpc_skb_rx_freed);
197 rxrpc_put_peer(peer);
198 _leave(" [MTU update]");
202 rxrpc_store_error(peer, serr);
204 rxrpc_free_skb(skb, rxrpc_skb_rx_freed);
205 rxrpc_put_peer(peer);
211 * Map an error report to error codes on the peer record.
213 static void rxrpc_store_error(struct rxrpc_peer *peer,
214 struct sock_exterr_skb *serr)
216 enum rxrpc_call_completion compl = RXRPC_CALL_NETWORK_ERROR;
217 struct sock_extended_err *ee;
226 switch (ee->ee_origin) {
227 case SO_EE_ORIGIN_ICMP:
228 switch (ee->ee_type) {
229 case ICMP_DEST_UNREACH:
230 switch (ee->ee_code) {
231 case ICMP_NET_UNREACH:
232 _net("Rx Received ICMP Network Unreachable");
234 case ICMP_HOST_UNREACH:
235 _net("Rx Received ICMP Host Unreachable");
237 case ICMP_PORT_UNREACH:
238 _net("Rx Received ICMP Port Unreachable");
240 case ICMP_NET_UNKNOWN:
241 _net("Rx Received ICMP Unknown Network");
243 case ICMP_HOST_UNKNOWN:
244 _net("Rx Received ICMP Unknown Host");
247 _net("Rx Received ICMP DestUnreach code=%u",
253 case ICMP_TIME_EXCEEDED:
254 _net("Rx Received ICMP TTL Exceeded");
258 _proto("Rx Received ICMP error { type=%u code=%u }",
259 ee->ee_type, ee->ee_code);
264 case SO_EE_ORIGIN_NONE:
265 case SO_EE_ORIGIN_LOCAL:
266 _proto("Rx Received local error { error=%d }", err);
267 compl = RXRPC_CALL_LOCAL_ERROR;
270 case SO_EE_ORIGIN_ICMP6:
272 _proto("Rx Received error report { orig=%u }", ee->ee_origin);
276 rxrpc_distribute_error(peer, err, compl);
280 * Distribute an error that occurred on a peer.
282 static void rxrpc_distribute_error(struct rxrpc_peer *peer, int error,
283 enum rxrpc_call_completion compl)
285 struct rxrpc_call *call;
287 hlist_for_each_entry_rcu(call, &peer->error_targets, error_link) {
288 rxrpc_see_call(call);
289 if (call->state < RXRPC_CALL_COMPLETE &&
290 rxrpc_set_call_completion(call, compl, 0, -error))
291 rxrpc_notify_socket(call);
296 * Add RTT information to cache. This is called in softirq mode and has
297 * exclusive access to the peer RTT data.
299 void rxrpc_peer_add_rtt(struct rxrpc_call *call, enum rxrpc_rtt_rx_trace why,
300 rxrpc_serial_t send_serial, rxrpc_serial_t resp_serial,
301 ktime_t send_time, ktime_t resp_time)
303 struct rxrpc_peer *peer = call->peer;
305 u64 sum = peer->rtt_sum, avg;
306 u8 cursor = peer->rtt_cursor, usage = peer->rtt_usage;
308 rtt = ktime_to_ns(ktime_sub(resp_time, send_time));
312 spin_lock(&peer->rtt_input_lock);
314 /* Replace the oldest datum in the RTT buffer */
315 sum -= peer->rtt_cache[cursor];
317 peer->rtt_cache[cursor] = rtt;
318 peer->rtt_cursor = (cursor + 1) & (RXRPC_RTT_CACHE_SIZE - 1);
320 if (usage < RXRPC_RTT_CACHE_SIZE) {
322 peer->rtt_usage = usage;
325 spin_unlock(&peer->rtt_input_lock);
327 /* Now recalculate the average */
328 if (usage == RXRPC_RTT_CACHE_SIZE) {
329 avg = sum / RXRPC_RTT_CACHE_SIZE;
335 /* Don't need to update this under lock */
337 trace_rxrpc_rtt_rx(call, why, send_serial, resp_serial, rtt,
342 * Perform keep-alive pings.
344 static void rxrpc_peer_keepalive_dispatch(struct rxrpc_net *rxnet,
345 struct list_head *collector,
349 struct rxrpc_peer *peer;
350 const u8 mask = ARRAY_SIZE(rxnet->peer_keepalive) - 1;
351 time64_t keepalive_at;
354 spin_lock_bh(&rxnet->peer_hash_lock);
356 while (!list_empty(collector)) {
357 peer = list_entry(collector->next,
358 struct rxrpc_peer, keepalive_link);
360 list_del_init(&peer->keepalive_link);
361 if (!rxrpc_get_peer_maybe(peer))
364 spin_unlock_bh(&rxnet->peer_hash_lock);
366 keepalive_at = peer->last_tx_at + RXRPC_KEEPALIVE_TIME;
367 slot = keepalive_at - base;
368 _debug("%02x peer %u t=%d {%pISp}",
369 cursor, peer->debug_id, slot, &peer->srx.transport);
371 if (keepalive_at <= base ||
372 keepalive_at > base + RXRPC_KEEPALIVE_TIME) {
373 rxrpc_send_keepalive(peer);
374 slot = RXRPC_KEEPALIVE_TIME;
377 /* A transmission to this peer occurred since last we examined
378 * it so put it into the appropriate future bucket.
382 spin_lock_bh(&rxnet->peer_hash_lock);
383 list_add_tail(&peer->keepalive_link,
384 &rxnet->peer_keepalive[slot & mask]);
385 rxrpc_put_peer(peer);
388 spin_unlock_bh(&rxnet->peer_hash_lock);
392 * Perform keep-alive pings with VERSION packets to keep any NAT alive.
394 void rxrpc_peer_keepalive_worker(struct work_struct *work)
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;
401 LIST_HEAD(collector);
403 now = ktime_get_seconds();
404 base = rxnet->peer_keepalive_base;
405 cursor = rxnet->peer_keepalive_cursor;
406 _enter("%lld,%u", base - now, cursor);
411 /* Remove to a temporary list all the peers that are currently lodged
412 * in expired buckets plus all new peers.
414 * Everything in the bucket at the cursor is processed this
415 * second; the bucket at cursor + 1 goes at now + 1s and so
418 spin_lock_bh(&rxnet->peer_hash_lock);
419 list_splice_init(&rxnet->peer_keepalive_new, &collector);
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],
430 spin_unlock_bh(&rxnet->peer_hash_lock);
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));
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]))
446 now = ktime_get_seconds();
452 timer_reduce(&rxnet->peer_keepalive_timer, jiffies + delay);