net/rxrpc/peer_event.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /* Peer event handling, typically ICMP messages.
   3  *
   4  * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
   5  * Written by David Howells (dhowells@redhat.com)
   6  */
   7
   8 #include <linux/module.h>
   9 #include <linux/net.h>
  10 #include <linux/skbuff.h>
  11 #include <linux/errqueue.h>
  12 #include <linux/udp.h>
  13 #include <linux/in.h>
  14 #include <linux/in6.h>
  15 #include <linux/icmp.h>
  16 #include <net/sock.h>
  17 #include <net/af_rxrpc.h>
  18 #include <net/ip.h>
  19 #include "ar-internal.h"
  20
  21 static void rxrpc_store_error(struct rxrpc_peer *, struct sock_exterr_skb *);
  22 static void rxrpc_distribute_error(struct rxrpc_peer *, int,
  23                                    enum rxrpc_call_completion);
  24
  25 /*
  26  * Find the peer associated with a local error.
  27  */
  28 static struct rxrpc_peer *rxrpc_lookup_peer_local_rcu(struct rxrpc_local *local,
  29                                                       const struct sk_buff *skb,
  30                                                       struct sockaddr_rxrpc *srx)
  31 {
  32         struct sock_exterr_skb *serr = SKB_EXT_ERR(skb);
  33
  34         _enter("");
  35
  36         memset(srx, 0, sizeof(*srx));
  37         srx->transport_type = local->srx.transport_type;
  38         srx->transport_len = local->srx.transport_len;
  39         srx->transport.family = local->srx.transport.family;
  40
  41         /* Can we see an ICMP4 packet on an ICMP6 listening socket?  and vice
  42          * versa?
  43          */
  44         switch (srx->transport.family) {
  45         case AF_INET:
  46                 srx->transport_len = sizeof(srx->transport.sin);
  47                 srx->transport.family = AF_INET;
  48                 srx->transport.sin.sin_port = serr->port;
  49                 switch (serr->ee.ee_origin) {
  50                 case SO_EE_ORIGIN_ICMP:
  51                         memcpy(&srx->transport.sin.sin_addr,
  52                                skb_network_header(skb) + serr->addr_offset,
  53                                sizeof(struct in_addr));
  54                         break;
  55                 case SO_EE_ORIGIN_ICMP6:
  56                         memcpy(&srx->transport.sin.sin_addr,
  57                                skb_network_header(skb) + serr->addr_offset + 12,
  58                                sizeof(struct in_addr));
  59                         break;
  60                 default:
  61                         memcpy(&srx->transport.sin.sin_addr, &ip_hdr(skb)->saddr,
  62                                sizeof(struct in_addr));
  63                         break;
  64                 }
  65                 break;
  66
  67 #ifdef CONFIG_AF_RXRPC_IPV6
  68         case AF_INET6:
  69                 switch (serr->ee.ee_origin) {
  70                 case SO_EE_ORIGIN_ICMP6:
  71                         srx->transport.sin6.sin6_port = serr->port;
  72                         memcpy(&srx->transport.sin6.sin6_addr,
  73                                skb_network_header(skb) + serr->addr_offset,
  74                                sizeof(struct in6_addr));
  75                         break;
  76                 case SO_EE_ORIGIN_ICMP:
  77                         srx->transport_len = sizeof(srx->transport.sin);
  78                         srx->transport.family = AF_INET;
  79                         srx->transport.sin.sin_port = serr->port;
  80                         memcpy(&srx->transport.sin.sin_addr,
  81                                skb_network_header(skb) + serr->addr_offset,
  82                                sizeof(struct in_addr));
  83                         break;
  84                 default:
  85                         memcpy(&srx->transport.sin6.sin6_addr,
  86                                &ipv6_hdr(skb)->saddr,
  87                                sizeof(struct in6_addr));
  88                         break;
  89                 }
  90                 break;
  91 #endif
  92
  93         default:
  94                 BUG();
  95         }
  96
  97         return rxrpc_lookup_peer_rcu(local, srx);
  98 }
  99
 100 /*
 101  * Handle an MTU/fragmentation problem.
 102  */
 103 static void rxrpc_adjust_mtu(struct rxrpc_peer *peer, unsigned int mtu)
 104 {
 105         /* wind down the local interface MTU */
 106         if (mtu > 0 && peer->if_mtu == 65535 && mtu < peer->if_mtu)
 107                 peer->if_mtu = mtu;
 108
 109         if (mtu == 0) {
 110                 /* they didn't give us a size, estimate one */
 111                 mtu = peer->if_mtu;
 112                 if (mtu > 1500) {
 113                         mtu >>= 1;
 114                         if (mtu < 1500)
 115                                 mtu = 1500;
 116                 } else {
 117                         mtu -= 100;
 118                         if (mtu < peer->hdrsize)
 119                                 mtu = peer->hdrsize + 4;
 120                 }
 121         }
 122
 123         if (mtu < peer->mtu) {
 124                 spin_lock_bh(&peer->lock);
 125                 peer->mtu = mtu;
 126                 peer->maxdata = peer->mtu - peer->hdrsize;
 127                 spin_unlock_bh(&peer->lock);
 128         }
 129 }
 130
 131 /*
 132  * Handle an error received on the local endpoint.
 133  */
 134 void rxrpc_error_report(struct sock *sk)
 135 {
 136         struct sock_exterr_skb *serr;
 137         struct sockaddr_rxrpc srx;
 138         struct rxrpc_local *local;
 139         struct rxrpc_peer *peer = NULL;
 140         struct sk_buff *skb;
 141
 142         rcu_read_lock();
 143         local = rcu_dereference_sk_user_data(sk);
 144         if (unlikely(!local)) {
 145                 rcu_read_unlock();
 146                 return;
 147         }
 148         _enter("%p{%d}", sk, local->debug_id);
 149
 150         /* Clear the outstanding error value on the socket so that it doesn't
 151          * cause kernel_sendmsg() to return it later.
 152          */
 153         sock_error(sk);
 154
 155         skb = sock_dequeue_err_skb(sk);
 156         if (!skb) {
 157                 rcu_read_unlock();
 158                 _leave("UDP socket errqueue empty");
 159                 return;
 160         }
 161         rxrpc_new_skb(skb, rxrpc_skb_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                 rcu_read_unlock();
 166                 rxrpc_free_skb(skb, rxrpc_skb_freed);
 167                 return;
 168         }
 169
 170         peer = rxrpc_lookup_peer_local_rcu(local, skb, &srx);
 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_freed);
 176                 _leave(" [no peer]");
 177                 return;
 178         }
 179
 180         trace_rxrpc_rx_icmp(peer, &serr->ee, &srx);
 181
 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->ee.ee_info);
 186                 goto out;
 187         }
 188
 189         rxrpc_store_error(peer, serr);
 190 out:
 191         rcu_read_unlock();
 192         rxrpc_free_skb(skb, rxrpc_skb_freed);
 193         rxrpc_put_peer(peer);
 194
 195         _leave("");
 196 }
 197
 198 /*
 199  * Map an error report to error codes on the peer record.
 200  */
 201 static void rxrpc_store_error(struct rxrpc_peer *peer,
 202                               struct sock_exterr_skb *serr)
 203 {
 204         enum rxrpc_call_completion compl = RXRPC_CALL_NETWORK_ERROR;
 205         struct sock_extended_err *ee;
 206         int err;
 207
 208         _enter("");
 209
 210         ee = &serr->ee;
 211
 212         err = ee->ee_errno;
 213
 214         switch (ee->ee_origin) {
 215         case SO_EE_ORIGIN_NONE:
 216         case SO_EE_ORIGIN_LOCAL:
 217                 compl = RXRPC_CALL_LOCAL_ERROR;
 218                 break;
 219
 220         case SO_EE_ORIGIN_ICMP6:
 221                 if (err == EACCES)
 222                         err = EHOSTUNREACH;
 223                 fallthrough;
 224         case SO_EE_ORIGIN_ICMP:
 225         default:
 226                 break;
 227         }
 228
 229         rxrpc_distribute_error(peer, err, compl);
 230 }
 231
 232 /*
 233  * Distribute an error that occurred on a peer.
 234  */
 235 static void rxrpc_distribute_error(struct rxrpc_peer *peer, int error,
 236                                    enum rxrpc_call_completion compl)
 237 {
 238         struct rxrpc_call *call;
 239
 240         hlist_for_each_entry_rcu(call, &peer->error_targets, error_link) {
 241                 rxrpc_see_call(call);
 242                 rxrpc_set_call_completion(call, compl, 0, -error);
 243         }
 244 }
 245
 246 /*
 247  * Perform keep-alive pings.
 248  */
 249 static void rxrpc_peer_keepalive_dispatch(struct rxrpc_net *rxnet,
 250                                           struct list_head *collector,
 251                                           time64_t base,
 252                                           u8 cursor)
 253 {
 254         struct rxrpc_peer *peer;
 255         const u8 mask = ARRAY_SIZE(rxnet->peer_keepalive) - 1;
 256         time64_t keepalive_at;
 257         int slot;
 258
 259         spin_lock_bh(&rxnet->peer_hash_lock);
 260
 261         while (!list_empty(collector)) {
 262                 peer = list_entry(collector->next,
 263                                   struct rxrpc_peer, keepalive_link);
 264
 265                 list_del_init(&peer->keepalive_link);
 266                 if (!rxrpc_get_peer_maybe(peer))
 267                         continue;
 268
 269                 if (__rxrpc_use_local(peer->local, rxrpc_local_use_peer_keepalive)) {
 270                         spin_unlock_bh(&rxnet->peer_hash_lock);
 271
 272                         keepalive_at = peer->last_tx_at + RXRPC_KEEPALIVE_TIME;
 273                         slot = keepalive_at - base;
 274                         _debug("%02x peer %u t=%d {%pISp}",
 275                                cursor, peer->debug_id, slot, &peer->srx.transport);
 276
 277                         if (keepalive_at <= base ||
 278                             keepalive_at > base + RXRPC_KEEPALIVE_TIME) {
 279                                 rxrpc_send_keepalive(peer);
 280                                 slot = RXRPC_KEEPALIVE_TIME;
 281                         }
 282
 283                         /* A transmission to this peer occurred since last we
 284                          * examined it so put it into the appropriate future
 285                          * bucket.
 286                          */
 287                         slot += cursor;
 288                         slot &= mask;
 289                         spin_lock_bh(&rxnet->peer_hash_lock);
 290                         list_add_tail(&peer->keepalive_link,
 291                                       &rxnet->peer_keepalive[slot & mask]);
 292                         rxrpc_unuse_local(peer->local, rxrpc_local_unuse_peer_keepalive);
 293                 }
 294                 rxrpc_put_peer_locked(peer);
 295         }
 296
 297         spin_unlock_bh(&rxnet->peer_hash_lock);
 298 }
 299
 300 /*
 301  * Perform keep-alive pings with VERSION packets to keep any NAT alive.
 302  */
 303 void rxrpc_peer_keepalive_worker(struct work_struct *work)
 304 {
 305         struct rxrpc_net *rxnet =
 306                 container_of(work, struct rxrpc_net, peer_keepalive_work);
 307         const u8 mask = ARRAY_SIZE(rxnet->peer_keepalive) - 1;
 308         time64_t base, now, delay;
 309         u8 cursor, stop;
 310         LIST_HEAD(collector);
 311
 312         now = ktime_get_seconds();
 313         base = rxnet->peer_keepalive_base;
 314         cursor = rxnet->peer_keepalive_cursor;
 315         _enter("%lld,%u", base - now, cursor);
 316
 317         if (!rxnet->live)
 318                 return;
 319
 320         /* Remove to a temporary list all the peers that are currently lodged
 321          * in expired buckets plus all new peers.
 322          *
 323          * Everything in the bucket at the cursor is processed this
 324          * second; the bucket at cursor + 1 goes at now + 1s and so
 325          * on...
 326          */
 327         spin_lock_bh(&rxnet->peer_hash_lock);
 328         list_splice_init(&rxnet->peer_keepalive_new, &collector);
 329
 330         stop = cursor + ARRAY_SIZE(rxnet->peer_keepalive);
 331         while (base <= now && (s8)(cursor - stop) < 0) {
 332                 list_splice_tail_init(&rxnet->peer_keepalive[cursor & mask],
 333                                       &collector);
 334                 base++;
 335                 cursor++;
 336         }
 337
 338         base = now;
 339         spin_unlock_bh(&rxnet->peer_hash_lock);
 340
 341         rxnet->peer_keepalive_base = base;
 342         rxnet->peer_keepalive_cursor = cursor;
 343         rxrpc_peer_keepalive_dispatch(rxnet, &collector, base, cursor);
 344         ASSERT(list_empty(&collector));
 345
 346         /* Schedule the timer for the next occupied timeslot. */
 347         cursor = rxnet->peer_keepalive_cursor;
 348         stop = cursor + RXRPC_KEEPALIVE_TIME - 1;
 349         for (; (s8)(cursor - stop) < 0; cursor++) {
 350                 if (!list_empty(&rxnet->peer_keepalive[cursor & mask]))
 351                         break;
 352                 base++;
 353         }
 354
 355         now = ktime_get_seconds();
 356         delay = base - now;
 357         if (delay < 1)
 358                 delay = 1;
 359         delay *= HZ;
 360         if (rxnet->live)
 361                 timer_reduce(&rxnet->peer_keepalive_timer, jiffies + delay);
 362
 363         _leave("");
 364 }