net/rxrpc/peer_event.c

   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
  27 /*
  28  * Find the peer associated with an ICMP packet.
  29  */
  30 static struct rxrpc_peer *rxrpc_lookup_peer_icmp_rcu(struct rxrpc_local *local,
  31                                                      const struct sk_buff *skb)
  32 {
  33         struct sock_exterr_skb *serr = SKB_EXT_ERR(skb);
  34         struct sockaddr_rxrpc srx;
  35
  36         _enter("");
  37
  38         memset(&srx, 0, sizeof(srx));
  39         srx.transport_type = local->srx.transport_type;
  40         srx.transport.family = local->srx.transport.family;
  41
  42         /* Can we see an ICMP4 packet on an ICMP6 listening socket?  and vice
  43          * versa?
  44          */
  45         switch (srx.transport.family) {
  46         case AF_INET:
  47                 srx.transport.sin.sin_port = serr->port;
  48                 srx.transport_len = sizeof(struct sockaddr_in);
  49                 switch (serr->ee.ee_origin) {
  50                 case SO_EE_ORIGIN_ICMP:
  51                         _net("Rx ICMP");
  52                         memcpy(&srx.transport.sin.sin_addr,
  53                                skb_network_header(skb) + serr->addr_offset,
  54                                sizeof(struct in_addr));
  55                         break;
  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));
  61                         break;
  62                 default:
  63                         memcpy(&srx.transport.sin.sin_addr, &ip_hdr(skb)->saddr,
  64                                sizeof(struct in_addr));
  65                         break;
  66                 }
  67                 break;
  68
  69 #ifdef CONFIG_AF_RXRPC_IPV6
  70         case AF_INET6:
  71                 srx.transport.sin6.sin6_port = serr->port;
  72                 srx.transport_len = sizeof(struct sockaddr_in6);
  73                 switch (serr->ee.ee_origin) {
  74                 case SO_EE_ORIGIN_ICMP6:
  75                         _net("Rx ICMP6");
  76                         memcpy(&srx.transport.sin6.sin6_addr,
  77                                skb_network_header(skb) + serr->addr_offset,
  78                                sizeof(struct in6_addr));
  79                         break;
  80                 case SO_EE_ORIGIN_ICMP:
  81                         _net("Rx ICMP on v6 sock");
  82                         memcpy(srx.transport.sin6.sin6_addr.s6_addr + 12,
  83                                skb_network_header(skb) + serr->addr_offset,
  84                                sizeof(struct in_addr));
  85                         break;
  86                 default:
  87                         memcpy(&srx.transport.sin6.sin6_addr,
  88                                &ipv6_hdr(skb)->saddr,
  89                                sizeof(struct in6_addr));
  90                         break;
  91                 }
  92                 break;
  93 #endif
  94
  95         default:
  96                 BUG();
  97         }
  98
  99         return rxrpc_lookup_peer_rcu(local, &srx);
 100 }
 101
 102 /*
 103  * Handle an MTU/fragmentation problem.
 104  */
 105 static void rxrpc_adjust_mtu(struct rxrpc_peer *peer, struct sock_exterr_skb *serr)
 106 {
 107         u32 mtu = serr->ee.ee_info;
 108
 109         _net("Rx ICMP Fragmentation Needed (%d)", mtu);
 110
 111         /* wind down the local interface MTU */
 112         if (mtu > 0 && peer->if_mtu == 65535 && mtu < peer->if_mtu) {
 113                 peer->if_mtu = mtu;
 114                 _net("I/F MTU %u", mtu);
 115         }
 116
 117         if (mtu == 0) {
 118                 /* they didn't give us a size, estimate one */
 119                 mtu = peer->if_mtu;
 120                 if (mtu > 1500) {
 121                         mtu >>= 1;
 122                         if (mtu < 1500)
 123                                 mtu = 1500;
 124                 } else {
 125                         mtu -= 100;
 126                         if (mtu < peer->hdrsize)
 127                                 mtu = peer->hdrsize + 4;
 128                 }
 129         }
 130
 131         if (mtu < peer->mtu) {
 132                 spin_lock_bh(&peer->lock);
 133                 peer->mtu = mtu;
 134                 peer->maxdata = peer->mtu - peer->hdrsize;
 135                 spin_unlock_bh(&peer->lock);
 136                 _net("Net MTU %u (maxdata %u)",
 137                      peer->mtu, peer->maxdata);
 138         }
 139 }
 140
 141 /*
 142  * Handle an error received on the local endpoint.
 143  */
 144 void rxrpc_error_report(struct sock *sk)
 145 {
 146         struct sock_exterr_skb *serr;
 147         struct rxrpc_local *local = sk->sk_user_data;
 148         struct rxrpc_peer *peer;
 149         struct sk_buff *skb;
 150
 151         if (unlikely(!local))
 152                 return;
 153
 154         _enter("%p{%d}", sk, local->debug_id);
 155
 156         skb = sock_dequeue_err_skb(sk);
 157         if (!skb) {
 158                 _leave("UDP socket errqueue empty");
 159                 return;
 160         }
 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);
 166                 return;
 167         }
 168
 169         rcu_read_lock();
 170         peer = rxrpc_lookup_peer_icmp_rcu(local, skb);
 171         if (peer && !rxrpc_get_peer_maybe(peer))
 172                 peer = NULL;
 173         if (!peer) {
 174                 rcu_read_unlock();
 175                 rxrpc_free_skb(skb, rxrpc_skb_rx_freed);
 176                 _leave(" [no peer]");
 177                 return;
 178         }
 179
 180         if ((serr->ee.ee_origin == SO_EE_ORIGIN_ICMP &&
 181              serr->ee.ee_type == ICMP_DEST_UNREACH &&
 182              serr->ee.ee_code == ICMP_FRAG_NEEDED)) {
 183                 rxrpc_adjust_mtu(peer, serr);
 184                 rcu_read_unlock();
 185                 rxrpc_free_skb(skb, rxrpc_skb_rx_freed);
 186                 rxrpc_put_peer(peer);
 187                 _leave(" [MTU update]");
 188                 return;
 189         }
 190
 191         rxrpc_store_error(peer, serr);
 192         rcu_read_unlock();
 193         rxrpc_free_skb(skb, rxrpc_skb_rx_freed);
 194
 195         /* The ref we obtained is passed off to the work item */
 196         rxrpc_queue_work(&peer->error_distributor);
 197         _leave("");
 198 }
 199
 200 /*
 201  * Map an error report to error codes on the peer record.
 202  */
 203 static void rxrpc_store_error(struct rxrpc_peer *peer,
 204                               struct sock_exterr_skb *serr)
 205 {
 206         struct sock_extended_err *ee;
 207         int err;
 208
 209         _enter("");
 210
 211         ee = &serr->ee;
 212
 213         _net("Rx Error o=%d t=%d c=%d e=%d",
 214              ee->ee_origin, ee->ee_type, ee->ee_code, ee->ee_errno);
 215
 216         err = ee->ee_errno;
 217
 218         switch (ee->ee_origin) {
 219         case SO_EE_ORIGIN_ICMP:
 220                 switch (ee->ee_type) {
 221                 case ICMP_DEST_UNREACH:
 222                         switch (ee->ee_code) {
 223                         case ICMP_NET_UNREACH:
 224                                 _net("Rx Received ICMP Network Unreachable");
 225                                 break;
 226                         case ICMP_HOST_UNREACH:
 227                                 _net("Rx Received ICMP Host Unreachable");
 228                                 break;
 229                         case ICMP_PORT_UNREACH:
 230                                 _net("Rx Received ICMP Port Unreachable");
 231                                 break;
 232                         case ICMP_NET_UNKNOWN:
 233                                 _net("Rx Received ICMP Unknown Network");
 234                                 break;
 235                         case ICMP_HOST_UNKNOWN:
 236                                 _net("Rx Received ICMP Unknown Host");
 237                                 break;
 238                         default:
 239                                 _net("Rx Received ICMP DestUnreach code=%u",
 240                                      ee->ee_code);
 241                                 break;
 242                         }
 243                         break;
 244
 245                 case ICMP_TIME_EXCEEDED:
 246                         _net("Rx Received ICMP TTL Exceeded");
 247                         break;
 248
 249                 default:
 250                         _proto("Rx Received ICMP error { type=%u code=%u }",
 251                                ee->ee_type, ee->ee_code);
 252                         break;
 253                 }
 254                 break;
 255
 256         case SO_EE_ORIGIN_NONE:
 257         case SO_EE_ORIGIN_LOCAL:
 258                 _proto("Rx Received local error { error=%d }", err);
 259                 err += RXRPC_LOCAL_ERROR_OFFSET;
 260                 break;
 261
 262         case SO_EE_ORIGIN_ICMP6:
 263         default:
 264                 _proto("Rx Received error report { orig=%u }", ee->ee_origin);
 265                 break;
 266         }
 267
 268         peer->error_report = err;
 269 }
 270
 271 /*
 272  * Distribute an error that occurred on a peer
 273  */
 274 void rxrpc_peer_error_distributor(struct work_struct *work)
 275 {
 276         struct rxrpc_peer *peer =
 277                 container_of(work, struct rxrpc_peer, error_distributor);
 278         struct rxrpc_call *call;
 279         enum rxrpc_call_completion compl;
 280         int error;
 281
 282         _enter("");
 283
 284         error = READ_ONCE(peer->error_report);
 285         if (error < RXRPC_LOCAL_ERROR_OFFSET) {
 286                 compl = RXRPC_CALL_NETWORK_ERROR;
 287         } else {
 288                 compl = RXRPC_CALL_LOCAL_ERROR;
 289                 error -= RXRPC_LOCAL_ERROR_OFFSET;
 290         }
 291
 292         _debug("ISSUE ERROR %s %d", rxrpc_call_completions[compl], error);
 293
 294         spin_lock_bh(&peer->lock);
 295
 296         while (!hlist_empty(&peer->error_targets)) {
 297                 call = hlist_entry(peer->error_targets.first,
 298                                    struct rxrpc_call, error_link);
 299                 hlist_del_init(&call->error_link);
 300                 rxrpc_see_call(call);
 301
 302                 if (rxrpc_set_call_completion(call, compl, 0, error))
 303                         rxrpc_notify_socket(call);
 304         }
 305
 306         spin_unlock_bh(&peer->lock);
 307
 308         rxrpc_put_peer(peer);
 309         _leave("");
 310 }
 311
 312 /*
 313  * Add RTT information to cache.  This is called in softirq mode and has
 314  * exclusive access to the peer RTT data.
 315  */
 316 void rxrpc_peer_add_rtt(struct rxrpc_call *call, enum rxrpc_rtt_rx_trace why,
 317                         rxrpc_serial_t send_serial, rxrpc_serial_t resp_serial,
 318                         ktime_t send_time, ktime_t resp_time)
 319 {
 320         struct rxrpc_peer *peer = call->peer;
 321         s64 rtt;
 322         u64 sum = peer->rtt_sum, avg;
 323         u8 cursor = peer->rtt_cursor, usage = peer->rtt_usage;
 324
 325         rtt = ktime_to_ns(ktime_sub(resp_time, send_time));
 326         if (rtt < 0)
 327                 return;
 328
 329         /* Replace the oldest datum in the RTT buffer */
 330         sum -= peer->rtt_cache[cursor];
 331         sum += rtt;
 332         peer->rtt_cache[cursor] = rtt;
 333         peer->rtt_cursor = (cursor + 1) & (RXRPC_RTT_CACHE_SIZE - 1);
 334         peer->rtt_sum = sum;
 335         if (usage < RXRPC_RTT_CACHE_SIZE) {
 336                 usage++;
 337                 peer->rtt_usage = usage;
 338         }
 339
 340         /* Now recalculate the average */
 341         if (usage == RXRPC_RTT_CACHE_SIZE) {
 342                 avg = sum / RXRPC_RTT_CACHE_SIZE;
 343         } else {
 344                 avg = sum;
 345                 do_div(avg, usage);
 346         }
 347
 348         peer->rtt = avg;
 349         trace_rxrpc_rtt_rx(call, why, send_serial, resp_serial, rtt,
 350                            usage, avg);
 351 }