net/decnet/dn_route.c

   1 /*
   2  * DECnet       An implementation of the DECnet protocol suite for the LINUX
   3  *              operating system.  DECnet is implemented using the  BSD Socket
   4  *              interface as the means of communication with the user level.
   5  *
   6  *              DECnet Routing Functions (Endnode and Router)
   7  *
   8  * Authors:     Steve Whitehouse <SteveW@ACM.org>
   9  *              Eduardo Marcelo Serrat <emserrat@geocities.com>
  10  *
  11  * Changes:
  12  *              Steve Whitehouse : Fixes to allow "intra-ethernet" and
  13  *                                 "return-to-sender" bits on outgoing
  14  *                                 packets.
  15  *              Steve Whitehouse : Timeouts for cached routes.
  16  *              Steve Whitehouse : Use dst cache for input routes too.
  17  *              Steve Whitehouse : Fixed error values in dn_send_skb.
  18  *              Steve Whitehouse : Rework routing functions to better fit
  19  *                                 DECnet routing design
  20  *              Alexey Kuznetsov : New SMP locking
  21  *              Steve Whitehouse : More SMP locking changes & dn_cache_dump()
  22  *              Steve Whitehouse : Prerouting NF hook, now really is prerouting.
  23  *                                 Fixed possible skb leak in rtnetlink funcs.
  24  *              Steve Whitehouse : Dave Miller's dynamic hash table sizing and
  25  *                                 Alexey Kuznetsov's finer grained locking
  26  *                                 from ipv4/route.c.
  27  *              Steve Whitehouse : Routing is now starting to look like a
  28  *                                 sensible set of code now, mainly due to
  29  *                                 my copying the IPv4 routing code. The
  30  *                                 hooks here are modified and will continue
  31  *                                 to evolve for a while.
  32  *              Steve Whitehouse : Real SMP at last :-) Also new netfilter
  33  *                                 stuff. Look out raw sockets your days
  34  *                                 are numbered!
  35  *              Steve Whitehouse : Added return-to-sender functions. Added
  36  *                                 backlog congestion level return codes.
  37  *              Steve Whitehouse : Fixed bug where routes were set up with
  38  *                                 no ref count on net devices.
  39  *              Steve Whitehouse : RCU for the route cache
  40  *              Steve Whitehouse : Preparations for the flow cache
  41  *              Steve Whitehouse : Prepare for nonlinear skbs
  42  */
  43
  44 /******************************************************************************
  45     (c) 1995-1998 E.M. Serrat           emserrat@geocities.com
  46
  47     This program is free software; you can redistribute it and/or modify
  48     it under the terms of the GNU General Public License as published by
  49     the Free Software Foundation; either version 2 of the License, or
  50     any later version.
  51
  52     This program is distributed in the hope that it will be useful,
  53     but WITHOUT ANY WARRANTY; without even the implied warranty of
  54     MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  55     GNU General Public License for more details.
  56 *******************************************************************************/
  57
  58 #include <linux/errno.h>
  59 #include <linux/types.h>
  60 #include <linux/socket.h>
  61 #include <linux/in.h>
  62 #include <linux/kernel.h>
  63 #include <linux/sockios.h>
  64 #include <linux/net.h>
  65 #include <linux/netdevice.h>
  66 #include <linux/inet.h>
  67 #include <linux/route.h>
  68 #include <linux/in_route.h>
  69 #include <linux/slab.h>
  70 #include <net/sock.h>
  71 #include <linux/mm.h>
  72 #include <linux/proc_fs.h>
  73 #include <linux/seq_file.h>
  74 #include <linux/init.h>
  75 #include <linux/rtnetlink.h>
  76 #include <linux/string.h>
  77 #include <linux/netfilter_decnet.h>
  78 #include <linux/rcupdate.h>
  79 #include <linux/times.h>
  80 #include <linux/export.h>
  81 #include <asm/errno.h>
  82 #include <net/net_namespace.h>
  83 #include <net/netlink.h>
  84 #include <net/neighbour.h>
  85 #include <net/dst.h>
  86 #include <net/flow.h>
  87 #include <net/fib_rules.h>
  88 #include <net/dn.h>
  89 #include <net/dn_dev.h>
  90 #include <net/dn_nsp.h>
  91 #include <net/dn_route.h>
  92 #include <net/dn_neigh.h>
  93 #include <net/dn_fib.h>
  94
  95 struct dn_rt_hash_bucket
  96 {
  97         struct dn_route __rcu *chain;
  98         spinlock_t lock;
  99 };
 100
 101 extern struct neigh_table dn_neigh_table;
 102
 103
 104 static unsigned char dn_hiord_addr[6] = {0xAA,0x00,0x04,0x00,0x00,0x00};
 105
 106 static const int dn_rt_min_delay = 2 * HZ;
 107 static const int dn_rt_max_delay = 10 * HZ;
 108 static const int dn_rt_mtu_expires = 10 * 60 * HZ;
 109
 110 static unsigned long dn_rt_deadline;
 111
 112 static int dn_dst_gc(struct dst_ops *ops);
 113 static struct dst_entry *dn_dst_check(struct dst_entry *, __u32);
 114 static unsigned int dn_dst_default_advmss(const struct dst_entry *dst);
 115 static unsigned int dn_dst_mtu(const struct dst_entry *dst);
 116 static void dn_dst_destroy(struct dst_entry *);
 117 static void dn_dst_ifdown(struct dst_entry *, struct net_device *dev, int how);
 118 static struct dst_entry *dn_dst_negative_advice(struct dst_entry *);
 119 static void dn_dst_link_failure(struct sk_buff *);
 120 static void dn_dst_update_pmtu(struct dst_entry *dst, struct sock *sk,
 121                                struct sk_buff *skb , u32 mtu,
 122                                bool confirm_neigh);
 123 static void dn_dst_redirect(struct dst_entry *dst, struct sock *sk,
 124                             struct sk_buff *skb);
 125 static struct neighbour *dn_dst_neigh_lookup(const struct dst_entry *dst,
 126                                              struct sk_buff *skb,
 127                                              const void *daddr);
 128 static int dn_route_input(struct sk_buff *);
 129 static void dn_run_flush(unsigned long dummy);
 130
 131 static struct dn_rt_hash_bucket *dn_rt_hash_table;
 132 static unsigned int dn_rt_hash_mask;
 133
 134 static struct timer_list dn_route_timer;
 135 static DEFINE_TIMER(dn_rt_flush_timer, dn_run_flush, 0, 0);
 136 int decnet_dst_gc_interval = 2;
 137
 138 static struct dst_ops dn_dst_ops = {
 139         .family =               PF_DECnet,
 140         .gc_thresh =            128,
 141         .gc =                   dn_dst_gc,
 142         .check =                dn_dst_check,
 143         .default_advmss =       dn_dst_default_advmss,
 144         .mtu =                  dn_dst_mtu,
 145         .cow_metrics =          dst_cow_metrics_generic,
 146         .destroy =              dn_dst_destroy,
 147         .ifdown =               dn_dst_ifdown,
 148         .negative_advice =      dn_dst_negative_advice,
 149         .link_failure =         dn_dst_link_failure,
 150         .update_pmtu =          dn_dst_update_pmtu,
 151         .redirect =             dn_dst_redirect,
 152         .neigh_lookup =         dn_dst_neigh_lookup,
 153 };
 154
 155 static void dn_dst_destroy(struct dst_entry *dst)
 156 {
 157         struct dn_route *rt = (struct dn_route *) dst;
 158
 159         if (rt->n)
 160                 neigh_release(rt->n);
 161         dst_destroy_metrics_generic(dst);
 162 }
 163
 164 static void dn_dst_ifdown(struct dst_entry *dst, struct net_device *dev, int how)
 165 {
 166         if (how) {
 167                 struct dn_route *rt = (struct dn_route *) dst;
 168                 struct neighbour *n = rt->n;
 169
 170                 if (n && n->dev == dev) {
 171                         n->dev = dev_net(dev)->loopback_dev;
 172                         dev_hold(n->dev);
 173                         dev_put(dev);
 174                 }
 175         }
 176 }
 177
 178 static __inline__ unsigned int dn_hash(__le16 src, __le16 dst)
 179 {
 180         __u16 tmp = (__u16 __force)(src ^ dst);
 181         tmp ^= (tmp >> 3);
 182         tmp ^= (tmp >> 5);
 183         tmp ^= (tmp >> 10);
 184         return dn_rt_hash_mask & (unsigned int)tmp;
 185 }
 186
 187 static void dn_dst_check_expire(unsigned long dummy)
 188 {
 189         int i;
 190         struct dn_route *rt;
 191         struct dn_route __rcu **rtp;
 192         unsigned long now = jiffies;
 193         unsigned long expire = 120 * HZ;
 194
 195         for (i = 0; i <= dn_rt_hash_mask; i++) {
 196                 rtp = &dn_rt_hash_table[i].chain;
 197
 198                 spin_lock(&dn_rt_hash_table[i].lock);
 199                 while ((rt = rcu_dereference_protected(*rtp,
 200                                                 lockdep_is_held(&dn_rt_hash_table[i].lock))) != NULL) {
 201                         if (atomic_read(&rt->dst.__refcnt) > 1 ||
 202                             (now - rt->dst.lastuse) < expire) {
 203                                 rtp = &rt->dst.dn_next;
 204                                 continue;
 205                         }
 206                         *rtp = rt->dst.dn_next;
 207                         rt->dst.dn_next = NULL;
 208                         dst_dev_put(&rt->dst);
 209                         dst_release(&rt->dst);
 210                 }
 211                 spin_unlock(&dn_rt_hash_table[i].lock);
 212
 213                 if ((jiffies - now) > 0)
 214                         break;
 215         }
 216
 217         mod_timer(&dn_route_timer, now + decnet_dst_gc_interval * HZ);
 218 }
 219
 220 static int dn_dst_gc(struct dst_ops *ops)
 221 {
 222         struct dn_route *rt;
 223         struct dn_route __rcu **rtp;
 224         int i;
 225         unsigned long now = jiffies;
 226         unsigned long expire = 10 * HZ;
 227
 228         for (i = 0; i <= dn_rt_hash_mask; i++) {
 229
 230                 spin_lock_bh(&dn_rt_hash_table[i].lock);
 231                 rtp = &dn_rt_hash_table[i].chain;
 232
 233                 while ((rt = rcu_dereference_protected(*rtp,
 234                                                 lockdep_is_held(&dn_rt_hash_table[i].lock))) != NULL) {
 235                         if (atomic_read(&rt->dst.__refcnt) > 1 ||
 236                             (now - rt->dst.lastuse) < expire) {
 237                                 rtp = &rt->dst.dn_next;
 238                                 continue;
 239                         }
 240                         *rtp = rt->dst.dn_next;
 241                         rt->dst.dn_next = NULL;
 242                         dst_dev_put(&rt->dst);
 243                         dst_release(&rt->dst);
 244                         break;
 245                 }
 246                 spin_unlock_bh(&dn_rt_hash_table[i].lock);
 247         }
 248
 249         return 0;
 250 }
 251
 252 /*
 253  * The decnet standards don't impose a particular minimum mtu, what they
 254  * do insist on is that the routing layer accepts a datagram of at least
 255  * 230 bytes long. Here we have to subtract the routing header length from
 256  * 230 to get the minimum acceptable mtu. If there is no neighbour, then we
 257  * assume the worst and use a long header size.
 258  *
 259  * We update both the mtu and the advertised mss (i.e. the segment size we
 260  * advertise to the other end).
 261  */
 262 static void dn_dst_update_pmtu(struct dst_entry *dst, struct sock *sk,
 263                                struct sk_buff *skb, u32 mtu,
 264                                bool confirm_neigh)
 265 {
 266         struct dn_route *rt = (struct dn_route *) dst;
 267         struct neighbour *n = rt->n;
 268         u32 min_mtu = 230;
 269         struct dn_dev *dn;
 270
 271         dn = n ? rcu_dereference_raw(n->dev->dn_ptr) : NULL;
 272
 273         if (dn && dn->use_long == 0)
 274                 min_mtu -= 6;
 275         else
 276                 min_mtu -= 21;
 277
 278         if (dst_metric(dst, RTAX_MTU) > mtu && mtu >= min_mtu) {
 279                 if (!(dst_metric_locked(dst, RTAX_MTU))) {
 280                         dst_metric_set(dst, RTAX_MTU, mtu);
 281                         dst_set_expires(dst, dn_rt_mtu_expires);
 282                 }
 283                 if (!(dst_metric_locked(dst, RTAX_ADVMSS))) {
 284                         u32 mss = mtu - DN_MAX_NSP_DATA_HEADER;
 285                         u32 existing_mss = dst_metric_raw(dst, RTAX_ADVMSS);
 286                         if (!existing_mss || existing_mss > mss)
 287                                 dst_metric_set(dst, RTAX_ADVMSS, mss);
 288                 }
 289         }
 290 }
 291
 292 static void dn_dst_redirect(struct dst_entry *dst, struct sock *sk,
 293                             struct sk_buff *skb)
 294 {
 295 }
 296
 297 /*
 298  * When a route has been marked obsolete. (e.g. routing cache flush)
 299  */
 300 static struct dst_entry *dn_dst_check(struct dst_entry *dst, __u32 cookie)
 301 {
 302         return NULL;
 303 }
 304
 305 static struct dst_entry *dn_dst_negative_advice(struct dst_entry *dst)
 306 {
 307         dst_release(dst);
 308         return NULL;
 309 }
 310
 311 static void dn_dst_link_failure(struct sk_buff *skb)
 312 {
 313 }
 314
 315 static inline int compare_keys(struct flowidn *fl1, struct flowidn *fl2)
 316 {
 317         return ((fl1->daddr ^ fl2->daddr) |
 318                 (fl1->saddr ^ fl2->saddr) |
 319                 (fl1->flowidn_mark ^ fl2->flowidn_mark) |
 320                 (fl1->flowidn_scope ^ fl2->flowidn_scope) |
 321                 (fl1->flowidn_oif ^ fl2->flowidn_oif) |
 322                 (fl1->flowidn_iif ^ fl2->flowidn_iif)) == 0;
 323 }
 324
 325 static int dn_insert_route(struct dn_route *rt, unsigned int hash, struct dn_route **rp)
 326 {
 327         struct dn_route *rth;
 328         struct dn_route __rcu **rthp;
 329         unsigned long now = jiffies;
 330
 331         rthp = &dn_rt_hash_table[hash].chain;
 332
 333         spin_lock_bh(&dn_rt_hash_table[hash].lock);
 334         while ((rth = rcu_dereference_protected(*rthp,
 335                                                 lockdep_is_held(&dn_rt_hash_table[hash].lock))) != NULL) {
 336                 if (compare_keys(&rth->fld, &rt->fld)) {
 337                         /* Put it first */
 338                         *rthp = rth->dst.dn_next;
 339                         rcu_assign_pointer(rth->dst.dn_next,
 340                                            dn_rt_hash_table[hash].chain);
 341                         rcu_assign_pointer(dn_rt_hash_table[hash].chain, rth);
 342
 343                         dst_use(&rth->dst, now);
 344                         spin_unlock_bh(&dn_rt_hash_table[hash].lock);
 345
 346                         dst_release_immediate(&rt->dst);
 347                         *rp = rth;
 348                         return 0;
 349                 }
 350                 rthp = &rth->dst.dn_next;
 351         }
 352
 353         rcu_assign_pointer(rt->dst.dn_next, dn_rt_hash_table[hash].chain);
 354         rcu_assign_pointer(dn_rt_hash_table[hash].chain, rt);
 355
 356         dst_use(&rt->dst, now);
 357         spin_unlock_bh(&dn_rt_hash_table[hash].lock);
 358         *rp = rt;
 359         return 0;
 360 }
 361
 362 static void dn_run_flush(unsigned long dummy)
 363 {
 364         int i;
 365         struct dn_route *rt, *next;
 366
 367         for (i = 0; i < dn_rt_hash_mask; i++) {
 368                 spin_lock_bh(&dn_rt_hash_table[i].lock);
 369
 370                 if ((rt = xchg((struct dn_route **)&dn_rt_hash_table[i].chain, NULL)) == NULL)
 371                         goto nothing_to_declare;
 372
 373                 for(; rt; rt = next) {
 374                         next = rcu_dereference_raw(rt->dst.dn_next);
 375                         RCU_INIT_POINTER(rt->dst.dn_next, NULL);
 376                         dst_dev_put(&rt->dst);
 377                         dst_release(&rt->dst);
 378                 }
 379
 380 nothing_to_declare:
 381                 spin_unlock_bh(&dn_rt_hash_table[i].lock);
 382         }
 383 }
 384
 385 static DEFINE_SPINLOCK(dn_rt_flush_lock);
 386
 387 void dn_rt_cache_flush(int delay)
 388 {
 389         unsigned long now = jiffies;
 390         int user_mode = !in_interrupt();
 391
 392         if (delay < 0)
 393                 delay = dn_rt_min_delay;
 394
 395         spin_lock_bh(&dn_rt_flush_lock);
 396
 397         if (del_timer(&dn_rt_flush_timer) && delay > 0 && dn_rt_deadline) {
 398                 long tmo = (long)(dn_rt_deadline - now);
 399
 400                 if (user_mode && tmo < dn_rt_max_delay - dn_rt_min_delay)
 401                         tmo = 0;
 402
 403                 if (delay > tmo)
 404                         delay = tmo;
 405         }
 406
 407         if (delay <= 0) {
 408                 spin_unlock_bh(&dn_rt_flush_lock);
 409                 dn_run_flush(0);
 410                 return;
 411         }
 412
 413         if (dn_rt_deadline == 0)
 414                 dn_rt_deadline = now + dn_rt_max_delay;
 415
 416         dn_rt_flush_timer.expires = now + delay;
 417         add_timer(&dn_rt_flush_timer);
 418         spin_unlock_bh(&dn_rt_flush_lock);
 419 }
 420
 421 /**
 422  * dn_return_short - Return a short packet to its sender
 423  * @skb: The packet to return
 424  *
 425  */
 426 static int dn_return_short(struct sk_buff *skb)
 427 {
 428         struct dn_skb_cb *cb;
 429         unsigned char *ptr;
 430         __le16 *src;
 431         __le16 *dst;
 432
 433         /* Add back headers */
 434         skb_push(skb, skb->data - skb_network_header(skb));
 435
 436         if ((skb = skb_unshare(skb, GFP_ATOMIC)) == NULL)
 437                 return NET_RX_DROP;
 438
 439         cb = DN_SKB_CB(skb);
 440         /* Skip packet length and point to flags */
 441         ptr = skb->data + 2;
 442         *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
 443
 444         dst = (__le16 *)ptr;
 445         ptr += 2;
 446         src = (__le16 *)ptr;
 447         ptr += 2;
 448         *ptr = 0; /* Zero hop count */
 449
 450         swap(*src, *dst);
 451
 452         skb->pkt_type = PACKET_OUTGOING;
 453         dn_rt_finish_output(skb, NULL, NULL);
 454         return NET_RX_SUCCESS;
 455 }
 456
 457 /**
 458  * dn_return_long - Return a long packet to its sender
 459  * @skb: The long format packet to return
 460  *
 461  */
 462 static int dn_return_long(struct sk_buff *skb)
 463 {
 464         struct dn_skb_cb *cb;
 465         unsigned char *ptr;
 466         unsigned char *src_addr, *dst_addr;
 467         unsigned char tmp[ETH_ALEN];
 468
 469         /* Add back all headers */
 470         skb_push(skb, skb->data - skb_network_header(skb));
 471
 472         if ((skb = skb_unshare(skb, GFP_ATOMIC)) == NULL)
 473                 return NET_RX_DROP;
 474
 475         cb = DN_SKB_CB(skb);
 476         /* Ignore packet length and point to flags */
 477         ptr = skb->data + 2;
 478
 479         /* Skip padding */
 480         if (*ptr & DN_RT_F_PF) {
 481                 char padlen = (*ptr & ~DN_RT_F_PF);
 482                 ptr += padlen;
 483         }
 484
 485         *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
 486         ptr += 2;
 487         dst_addr = ptr;
 488         ptr += 8;
 489         src_addr = ptr;
 490         ptr += 6;
 491         *ptr = 0; /* Zero hop count */
 492
 493         /* Swap source and destination */
 494         memcpy(tmp, src_addr, ETH_ALEN);
 495         memcpy(src_addr, dst_addr, ETH_ALEN);
 496         memcpy(dst_addr, tmp, ETH_ALEN);
 497
 498         skb->pkt_type = PACKET_OUTGOING;
 499         dn_rt_finish_output(skb, dst_addr, src_addr);
 500         return NET_RX_SUCCESS;
 501 }
 502
 503 /**
 504  * dn_route_rx_packet - Try and find a route for an incoming packet
 505  * @skb: The packet to find a route for
 506  *
 507  * Returns: result of input function if route is found, error code otherwise
 508  */
 509 static int dn_route_rx_packet(struct net *net, struct sock *sk, struct sk_buff *skb)
 510 {
 511         struct dn_skb_cb *cb;
 512         int err;
 513
 514         if ((err = dn_route_input(skb)) == 0)
 515                 return dst_input(skb);
 516
 517         cb = DN_SKB_CB(skb);
 518         if (decnet_debug_level & 4) {
 519                 char *devname = skb->dev ? skb->dev->name : "???";
 520
 521                 printk(KERN_DEBUG
 522                         "DECnet: dn_route_rx_packet: rt_flags=0x%02x dev=%s len=%d src=0x%04hx dst=0x%04hx err=%d type=%d\n",
 523                         (int)cb->rt_flags, devname, skb->len,
 524                         le16_to_cpu(cb->src), le16_to_cpu(cb->dst),
 525                         err, skb->pkt_type);
 526         }
 527
 528         if ((skb->pkt_type == PACKET_HOST) && (cb->rt_flags & DN_RT_F_RQR)) {
 529                 switch (cb->rt_flags & DN_RT_PKT_MSK) {
 530                 case DN_RT_PKT_SHORT:
 531                         return dn_return_short(skb);
 532                 case DN_RT_PKT_LONG:
 533                         return dn_return_long(skb);
 534                 }
 535         }
 536
 537         kfree_skb(skb);
 538         return NET_RX_DROP;
 539 }
 540
 541 static int dn_route_rx_long(struct sk_buff *skb)
 542 {
 543         struct dn_skb_cb *cb = DN_SKB_CB(skb);
 544         unsigned char *ptr = skb->data;
 545
 546         if (!pskb_may_pull(skb, 21)) /* 20 for long header, 1 for shortest nsp */
 547                 goto drop_it;
 548
 549         skb_pull(skb, 20);
 550         skb_reset_transport_header(skb);
 551
 552         /* Destination info */
 553         ptr += 2;
 554         cb->dst = dn_eth2dn(ptr);
 555         if (memcmp(ptr, dn_hiord_addr, 4) != 0)
 556                 goto drop_it;
 557         ptr += 6;
 558
 559
 560         /* Source info */
 561         ptr += 2;
 562         cb->src = dn_eth2dn(ptr);
 563         if (memcmp(ptr, dn_hiord_addr, 4) != 0)
 564                 goto drop_it;
 565         ptr += 6;
 566         /* Other junk */
 567         ptr++;
 568         cb->hops = *ptr++; /* Visit Count */
 569
 570         return NF_HOOK(NFPROTO_DECNET, NF_DN_PRE_ROUTING,
 571                        &init_net, NULL, skb, skb->dev, NULL,
 572                        dn_route_rx_packet);
 573
 574 drop_it:
 575         kfree_skb(skb);
 576         return NET_RX_DROP;
 577 }
 578
 579
 580
 581 static int dn_route_rx_short(struct sk_buff *skb)
 582 {
 583         struct dn_skb_cb *cb = DN_SKB_CB(skb);
 584         unsigned char *ptr = skb->data;
 585
 586         if (!pskb_may_pull(skb, 6)) /* 5 for short header + 1 for shortest nsp */
 587                 goto drop_it;
 588
 589         skb_pull(skb, 5);
 590         skb_reset_transport_header(skb);
 591
 592         cb->dst = *(__le16 *)ptr;
 593         ptr += 2;
 594         cb->src = *(__le16 *)ptr;
 595         ptr += 2;
 596         cb->hops = *ptr & 0x3f;
 597
 598         return NF_HOOK(NFPROTO_DECNET, NF_DN_PRE_ROUTING,
 599                        &init_net, NULL, skb, skb->dev, NULL,
 600                        dn_route_rx_packet);
 601
 602 drop_it:
 603         kfree_skb(skb);
 604         return NET_RX_DROP;
 605 }
 606
 607 static int dn_route_discard(struct net *net, struct sock *sk, struct sk_buff *skb)
 608 {
 609         /*
 610          * I know we drop the packet here, but thats considered success in
 611          * this case
 612          */
 613         kfree_skb(skb);
 614         return NET_RX_SUCCESS;
 615 }
 616
 617 static int dn_route_ptp_hello(struct net *net, struct sock *sk, struct sk_buff *skb)
 618 {
 619         dn_dev_hello(skb);
 620         dn_neigh_pointopoint_hello(skb);
 621         return NET_RX_SUCCESS;
 622 }
 623
 624 int dn_route_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
 625 {
 626         struct dn_skb_cb *cb;
 627         unsigned char flags = 0;
 628         __u16 len = le16_to_cpu(*(__le16 *)skb->data);
 629         struct dn_dev *dn = rcu_dereference(dev->dn_ptr);
 630         unsigned char padlen = 0;
 631
 632         if (!net_eq(dev_net(dev), &init_net))
 633                 goto dump_it;
 634
 635         if (dn == NULL)
 636                 goto dump_it;
 637
 638         if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
 639                 goto out;
 640
 641         if (!pskb_may_pull(skb, 3))
 642                 goto dump_it;
 643
 644         skb_pull(skb, 2);
 645
 646         if (len > skb->len)
 647                 goto dump_it;
 648
 649         skb_trim(skb, len);
 650
 651         flags = *skb->data;
 652
 653         cb = DN_SKB_CB(skb);
 654         cb->stamp = jiffies;
 655         cb->iif = dev->ifindex;
 656
 657         /*
 658          * If we have padding, remove it.
 659          */
 660         if (flags & DN_RT_F_PF) {
 661                 padlen = flags & ~DN_RT_F_PF;
 662                 if (!pskb_may_pull(skb, padlen + 1))
 663                         goto dump_it;
 664                 skb_pull(skb, padlen);
 665                 flags = *skb->data;
 666         }
 667
 668         skb_reset_network_header(skb);
 669
 670         /*
 671          * Weed out future version DECnet
 672          */
 673         if (flags & DN_RT_F_VER)
 674                 goto dump_it;
 675
 676         cb->rt_flags = flags;
 677
 678         if (decnet_debug_level & 1)
 679                 printk(KERN_DEBUG
 680                         "dn_route_rcv: got 0x%02x from %s [%d %d %d]\n",
 681                         (int)flags, (dev) ? dev->name : "???", len, skb->len,
 682                         padlen);
 683
 684         if (flags & DN_RT_PKT_CNTL) {
 685                 if (unlikely(skb_linearize(skb)))
 686                         goto dump_it;
 687
 688                 switch (flags & DN_RT_CNTL_MSK) {
 689                 case DN_RT_PKT_INIT:
 690                         dn_dev_init_pkt(skb);
 691                         break;
 692                 case DN_RT_PKT_VERI:
 693                         dn_dev_veri_pkt(skb);
 694                         break;
 695                 }
 696
 697                 if (dn->parms.state != DN_DEV_S_RU)
 698                         goto dump_it;
 699
 700                 switch (flags & DN_RT_CNTL_MSK) {
 701                 case DN_RT_PKT_HELO:
 702                         return NF_HOOK(NFPROTO_DECNET, NF_DN_HELLO,
 703                                        &init_net, NULL, skb, skb->dev, NULL,
 704                                        dn_route_ptp_hello);
 705
 706                 case DN_RT_PKT_L1RT:
 707                 case DN_RT_PKT_L2RT:
 708                         return NF_HOOK(NFPROTO_DECNET, NF_DN_ROUTE,
 709                                        &init_net, NULL, skb, skb->dev, NULL,
 710                                        dn_route_discard);
 711                 case DN_RT_PKT_ERTH:
 712                         return NF_HOOK(NFPROTO_DECNET, NF_DN_HELLO,
 713                                        &init_net, NULL, skb, skb->dev, NULL,
 714                                        dn_neigh_router_hello);
 715
 716                 case DN_RT_PKT_EEDH:
 717                         return NF_HOOK(NFPROTO_DECNET, NF_DN_HELLO,
 718                                        &init_net, NULL, skb, skb->dev, NULL,
 719                                        dn_neigh_endnode_hello);
 720                 }
 721         } else {
 722                 if (dn->parms.state != DN_DEV_S_RU)
 723                         goto dump_it;
 724
 725                 skb_pull(skb, 1); /* Pull flags */
 726
 727                 switch (flags & DN_RT_PKT_MSK) {
 728                 case DN_RT_PKT_LONG:
 729                         return dn_route_rx_long(skb);
 730                 case DN_RT_PKT_SHORT:
 731                         return dn_route_rx_short(skb);
 732                 }
 733         }
 734
 735 dump_it:
 736         kfree_skb(skb);
 737 out:
 738         return NET_RX_DROP;
 739 }
 740
 741 static int dn_output(struct net *net, struct sock *sk, struct sk_buff *skb)
 742 {
 743         struct dst_entry *dst = skb_dst(skb);
 744         struct dn_route *rt = (struct dn_route *)dst;
 745         struct net_device *dev = dst->dev;
 746         struct dn_skb_cb *cb = DN_SKB_CB(skb);
 747
 748         int err = -EINVAL;
 749
 750         if (rt->n == NULL)
 751                 goto error;
 752
 753         skb->dev = dev;
 754
 755         cb->src = rt->rt_saddr;
 756         cb->dst = rt->rt_daddr;
 757
 758         /*
 759          * Always set the Intra-Ethernet bit on all outgoing packets
 760          * originated on this node. Only valid flag from upper layers
 761          * is return-to-sender-requested. Set hop count to 0 too.
 762          */
 763         cb->rt_flags &= ~DN_RT_F_RQR;
 764         cb->rt_flags |= DN_RT_F_IE;
 765         cb->hops = 0;
 766
 767         return NF_HOOK(NFPROTO_DECNET, NF_DN_LOCAL_OUT,
 768                        &init_net, sk, skb, NULL, dev,
 769                        dn_to_neigh_output);
 770
 771 error:
 772         net_dbg_ratelimited("dn_output: This should not happen\n");
 773
 774         kfree_skb(skb);
 775
 776         return err;
 777 }
 778
 779 static int dn_forward(struct sk_buff *skb)
 780 {
 781         struct dn_skb_cb *cb = DN_SKB_CB(skb);
 782         struct dst_entry *dst = skb_dst(skb);
 783         struct dn_dev *dn_db = rcu_dereference(dst->dev->dn_ptr);
 784         struct dn_route *rt;
 785         int header_len;
 786         struct net_device *dev = skb->dev;
 787
 788         if (skb->pkt_type != PACKET_HOST)
 789                 goto drop;
 790
 791         /* Ensure that we have enough space for headers */
 792         rt = (struct dn_route *)skb_dst(skb);
 793         header_len = dn_db->use_long ? 21 : 6;
 794         if (skb_cow(skb, LL_RESERVED_SPACE(rt->dst.dev)+header_len))
 795                 goto drop;
 796
 797         /*
 798          * Hop count exceeded.
 799          */
 800         if (++cb->hops > 30)
 801                 goto drop;
 802
 803         skb->dev = rt->dst.dev;
 804
 805         /*
 806          * If packet goes out same interface it came in on, then set
 807          * the Intra-Ethernet bit. This has no effect for short
 808          * packets, so we don't need to test for them here.
 809          */
 810         cb->rt_flags &= ~DN_RT_F_IE;
 811         if (rt->rt_flags & RTCF_DOREDIRECT)
 812                 cb->rt_flags |= DN_RT_F_IE;
 813
 814         return NF_HOOK(NFPROTO_DECNET, NF_DN_FORWARD,
 815                        &init_net, NULL, skb, dev, skb->dev,
 816                        dn_to_neigh_output);
 817
 818 drop:
 819         kfree_skb(skb);
 820         return NET_RX_DROP;
 821 }
 822
 823 /*
 824  * Used to catch bugs. This should never normally get
 825  * called.
 826  */
 827 static int dn_rt_bug_out(struct net *net, struct sock *sk, struct sk_buff *skb)
 828 {
 829         struct dn_skb_cb *cb = DN_SKB_CB(skb);
 830
 831         net_dbg_ratelimited("dn_rt_bug: skb from:%04x to:%04x\n",
 832                             le16_to_cpu(cb->src), le16_to_cpu(cb->dst));
 833
 834         kfree_skb(skb);
 835
 836         return NET_RX_DROP;
 837 }
 838
 839 static int dn_rt_bug(struct sk_buff *skb)
 840 {
 841         struct dn_skb_cb *cb = DN_SKB_CB(skb);
 842
 843         net_dbg_ratelimited("dn_rt_bug: skb from:%04x to:%04x\n",
 844                             le16_to_cpu(cb->src), le16_to_cpu(cb->dst));
 845
 846         kfree_skb(skb);
 847
 848         return NET_RX_DROP;
 849 }
 850
 851 static unsigned int dn_dst_default_advmss(const struct dst_entry *dst)
 852 {
 853         return dn_mss_from_pmtu(dst->dev, dst_mtu(dst));
 854 }
 855
 856 static unsigned int dn_dst_mtu(const struct dst_entry *dst)
 857 {
 858         unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
 859
 860         return mtu ? : dst->dev->mtu;
 861 }
 862
 863 static struct neighbour *dn_dst_neigh_lookup(const struct dst_entry *dst,
 864                                              struct sk_buff *skb,
 865                                              const void *daddr)
 866 {
 867         return __neigh_lookup_errno(&dn_neigh_table, daddr, dst->dev);
 868 }
 869
 870 static int dn_rt_set_next_hop(struct dn_route *rt, struct dn_fib_res *res)
 871 {
 872         struct dn_fib_info *fi = res->fi;
 873         struct net_device *dev = rt->dst.dev;
 874         unsigned int mss_metric;
 875         struct neighbour *n;
 876
 877         if (fi) {
 878                 if (DN_FIB_RES_GW(*res) &&
 879                     DN_FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
 880                         rt->rt_gateway = DN_FIB_RES_GW(*res);
 881                 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
 882         }
 883         rt->rt_type = res->type;
 884
 885         if (dev != NULL && rt->n == NULL) {
 886                 n = __neigh_lookup_errno(&dn_neigh_table, &rt->rt_gateway, dev);
 887                 if (IS_ERR(n))
 888                         return PTR_ERR(n);
 889                 rt->n = n;
 890         }
 891
 892         if (dst_metric(&rt->dst, RTAX_MTU) > rt->dst.dev->mtu)
 893                 dst_metric_set(&rt->dst, RTAX_MTU, rt->dst.dev->mtu);
 894         mss_metric = dst_metric_raw(&rt->dst, RTAX_ADVMSS);
 895         if (mss_metric) {
 896                 unsigned int mss = dn_mss_from_pmtu(dev, dst_mtu(&rt->dst));
 897                 if (mss_metric > mss)
 898                         dst_metric_set(&rt->dst, RTAX_ADVMSS, mss);
 899         }
 900         return 0;
 901 }
 902
 903 static inline int dn_match_addr(__le16 addr1, __le16 addr2)
 904 {
 905         __u16 tmp = le16_to_cpu(addr1) ^ le16_to_cpu(addr2);
 906         int match = 16;
 907         while(tmp) {
 908                 tmp >>= 1;
 909                 match--;
 910         }
 911         return match;
 912 }
 913
 914 static __le16 dnet_select_source(const struct net_device *dev, __le16 daddr, int scope)
 915 {
 916         __le16 saddr = 0;
 917         struct dn_dev *dn_db;
 918         struct dn_ifaddr *ifa;
 919         int best_match = 0;
 920         int ret;
 921
 922         rcu_read_lock();
 923         dn_db = rcu_dereference(dev->dn_ptr);
 924         for (ifa = rcu_dereference(dn_db->ifa_list);
 925              ifa != NULL;
 926              ifa = rcu_dereference(ifa->ifa_next)) {
 927                 if (ifa->ifa_scope > scope)
 928                         continue;
 929                 if (!daddr) {
 930                         saddr = ifa->ifa_local;
 931                         break;
 932                 }
 933                 ret = dn_match_addr(daddr, ifa->ifa_local);
 934                 if (ret > best_match)
 935                         saddr = ifa->ifa_local;
 936                 if (best_match == 0)
 937                         saddr = ifa->ifa_local;
 938         }
 939         rcu_read_unlock();
 940
 941         return saddr;
 942 }
 943
 944 static inline __le16 __dn_fib_res_prefsrc(struct dn_fib_res *res)
 945 {
 946         return dnet_select_source(DN_FIB_RES_DEV(*res), DN_FIB_RES_GW(*res), res->scope);
 947 }
 948
 949 static inline __le16 dn_fib_rules_map_destination(__le16 daddr, struct dn_fib_res *res)
 950 {
 951         __le16 mask = dnet_make_mask(res->prefixlen);
 952         return (daddr&~mask)|res->fi->fib_nh->nh_gw;
 953 }
 954
 955 static int dn_route_output_slow(struct dst_entry **pprt, const struct flowidn *oldflp, int try_hard)
 956 {
 957         struct flowidn fld = {
 958                 .daddr = oldflp->daddr,
 959                 .saddr = oldflp->saddr,
 960                 .flowidn_scope = RT_SCOPE_UNIVERSE,
 961                 .flowidn_mark = oldflp->flowidn_mark,
 962                 .flowidn_iif = LOOPBACK_IFINDEX,
 963                 .flowidn_oif = oldflp->flowidn_oif,
 964         };
 965         struct dn_route *rt = NULL;
 966         struct net_device *dev_out = NULL, *dev;
 967         struct neighbour *neigh = NULL;
 968         unsigned int hash;
 969         unsigned int flags = 0;
 970         struct dn_fib_res res = { .fi = NULL, .type = RTN_UNICAST };
 971         int err;
 972         int free_res = 0;
 973         __le16 gateway = 0;
 974
 975         if (decnet_debug_level & 16)
 976                 printk(KERN_DEBUG
 977                        "dn_route_output_slow: dst=%04x src=%04x mark=%d"
 978                        " iif=%d oif=%d\n", le16_to_cpu(oldflp->daddr),
 979                        le16_to_cpu(oldflp->saddr),
 980                        oldflp->flowidn_mark, LOOPBACK_IFINDEX,
 981                        oldflp->flowidn_oif);
 982
 983         /* If we have an output interface, verify its a DECnet device */
 984         if (oldflp->flowidn_oif) {
 985                 dev_out = dev_get_by_index(&init_net, oldflp->flowidn_oif);
 986                 err = -ENODEV;
 987                 if (dev_out && dev_out->dn_ptr == NULL) {
 988                         dev_put(dev_out);
 989                         dev_out = NULL;
 990                 }
 991                 if (dev_out == NULL)
 992                         goto out;
 993         }
 994
 995         /* If we have a source address, verify that its a local address */
 996         if (oldflp->saddr) {
 997                 err = -EADDRNOTAVAIL;
 998
 999                 if (dev_out) {
1000                         if (dn_dev_islocal(dev_out, oldflp->saddr))
1001                                 goto source_ok;
1002                         dev_put(dev_out);
1003                         goto out;
1004                 }
1005                 rcu_read_lock();
1006                 for_each_netdev_rcu(&init_net, dev) {
1007                         if (!dev->dn_ptr)
1008                                 continue;
1009                         if (!dn_dev_islocal(dev, oldflp->saddr))
1010                                 continue;
1011                         if ((dev->flags & IFF_LOOPBACK) &&
1012                             oldflp->daddr &&
1013                             !dn_dev_islocal(dev, oldflp->daddr))
1014                                 continue;
1015
1016                         dev_out = dev;
1017                         break;
1018                 }
1019                 rcu_read_unlock();
1020                 if (dev_out == NULL)
1021                         goto out;
1022                 dev_hold(dev_out);
1023 source_ok:
1024                 ;
1025         }
1026
1027         /* No destination? Assume its local */
1028         if (!fld.daddr) {
1029                 fld.daddr = fld.saddr;
1030
1031                 if (dev_out)
1032                         dev_put(dev_out);
1033                 err = -EINVAL;
1034                 dev_out = init_net.loopback_dev;
1035                 if (!dev_out->dn_ptr)
1036                         goto out;
1037                 err = -EADDRNOTAVAIL;
1038                 dev_hold(dev_out);
1039                 if (!fld.daddr) {
1040                         fld.daddr =
1041                         fld.saddr = dnet_select_source(dev_out, 0,
1042                                                        RT_SCOPE_HOST);
1043                         if (!fld.daddr)
1044                                 goto out;
1045                 }
1046                 fld.flowidn_oif = LOOPBACK_IFINDEX;
1047                 res.type = RTN_LOCAL;
1048                 goto make_route;
1049         }
1050
1051         if (decnet_debug_level & 16)
1052                 printk(KERN_DEBUG
1053                        "dn_route_output_slow: initial checks complete."
1054                        " dst=%04x src=%04x oif=%d try_hard=%d\n",
1055                        le16_to_cpu(fld.daddr), le16_to_cpu(fld.saddr),
1056                        fld.flowidn_oif, try_hard);
1057
1058         /*
1059          * N.B. If the kernel is compiled without router support then
1060          * dn_fib_lookup() will evaluate to non-zero so this if () block
1061          * will always be executed.
1062          */
1063         err = -ESRCH;
1064         if (try_hard || (err = dn_fib_lookup(&fld, &res)) != 0) {
1065                 struct dn_dev *dn_db;
1066                 if (err != -ESRCH)
1067                         goto out;
1068                 /*
1069                  * Here the fallback is basically the standard algorithm for
1070                  * routing in endnodes which is described in the DECnet routing
1071                  * docs
1072                  *
1073                  * If we are not trying hard, look in neighbour cache.
1074                  * The result is tested to ensure that if a specific output
1075                  * device/source address was requested, then we honour that
1076                  * here
1077                  */
1078                 if (!try_hard) {
1079                         neigh = neigh_lookup_nodev(&dn_neigh_table, &init_net, &fld.daddr);
1080                         if (neigh) {
1081                                 if ((oldflp->flowidn_oif &&
1082                                     (neigh->dev->ifindex != oldflp->flowidn_oif)) ||
1083                                     (oldflp->saddr &&
1084                                     (!dn_dev_islocal(neigh->dev,
1085                                                      oldflp->saddr)))) {
1086                                         neigh_release(neigh);
1087                                         neigh = NULL;
1088                                 } else {
1089                                         if (dev_out)
1090                                                 dev_put(dev_out);
1091                                         if (dn_dev_islocal(neigh->dev, fld.daddr)) {
1092                                                 dev_out = init_net.loopback_dev;
1093                                                 res.type = RTN_LOCAL;
1094                                         } else {
1095                                                 dev_out = neigh->dev;
1096                                         }
1097                                         dev_hold(dev_out);
1098                                         goto select_source;
1099                                 }
1100                         }
1101                 }
1102
1103                 /* Not there? Perhaps its a local address */
1104                 if (dev_out == NULL)
1105                         dev_out = dn_dev_get_default();
1106                 err = -ENODEV;
1107                 if (dev_out == NULL)
1108                         goto out;
1109                 dn_db = rcu_dereference_raw(dev_out->dn_ptr);
1110                 if (!dn_db)
1111                         goto e_inval;
1112                 /* Possible improvement - check all devices for local addr */
1113                 if (dn_dev_islocal(dev_out, fld.daddr)) {
1114                         dev_put(dev_out);
1115                         dev_out = init_net.loopback_dev;
1116                         dev_hold(dev_out);
1117                         res.type = RTN_LOCAL;
1118                         goto select_source;
1119                 }
1120                 /* Not local either.... try sending it to the default router */
1121                 neigh = neigh_clone(dn_db->router);
1122                 BUG_ON(neigh && neigh->dev != dev_out);
1123
1124                 /* Ok then, we assume its directly connected and move on */
1125 select_source:
1126                 if (neigh)
1127                         gateway = ((struct dn_neigh *)neigh)->addr;
1128                 if (gateway == 0)
1129                         gateway = fld.daddr;
1130                 if (fld.saddr == 0) {
1131                         fld.saddr = dnet_select_source(dev_out, gateway,
1132                                                        res.type == RTN_LOCAL ?
1133                                                        RT_SCOPE_HOST :
1134                                                        RT_SCOPE_LINK);
1135                         if (fld.saddr == 0 && res.type != RTN_LOCAL)
1136                                 goto e_addr;
1137                 }
1138                 fld.flowidn_oif = dev_out->ifindex;
1139                 goto make_route;
1140         }
1141         free_res = 1;
1142
1143         if (res.type == RTN_NAT)
1144                 goto e_inval;
1145
1146         if (res.type == RTN_LOCAL) {
1147                 if (!fld.saddr)
1148                         fld.saddr = fld.daddr;
1149                 if (dev_out)
1150                         dev_put(dev_out);
1151                 dev_out = init_net.loopback_dev;
1152                 dev_hold(dev_out);
1153                 if (!dev_out->dn_ptr)
1154                         goto e_inval;
1155                 fld.flowidn_oif = dev_out->ifindex;
1156                 if (res.fi)
1157                         dn_fib_info_put(res.fi);
1158                 res.fi = NULL;
1159                 goto make_route;
1160         }
1161
1162         if (res.fi->fib_nhs > 1 && fld.flowidn_oif == 0)
1163                 dn_fib_select_multipath(&fld, &res);
1164
1165         /*
1166          * We could add some logic to deal with default routes here and
1167          * get rid of some of the special casing above.
1168          */
1169
1170         if (!fld.saddr)
1171                 fld.saddr = DN_FIB_RES_PREFSRC(res);
1172
1173         if (dev_out)
1174                 dev_put(dev_out);
1175         dev_out = DN_FIB_RES_DEV(res);
1176         dev_hold(dev_out);
1177         fld.flowidn_oif = dev_out->ifindex;
1178         gateway = DN_FIB_RES_GW(res);
1179
1180 make_route:
1181         if (dev_out->flags & IFF_LOOPBACK)
1182                 flags |= RTCF_LOCAL;
1183
1184         rt = dst_alloc(&dn_dst_ops, dev_out, 0, DST_OBSOLETE_NONE, DST_HOST);
1185         if (rt == NULL)
1186                 goto e_nobufs;
1187
1188         memset(&rt->fld, 0, sizeof(rt->fld));
1189         rt->fld.saddr        = oldflp->saddr;
1190         rt->fld.daddr        = oldflp->daddr;
1191         rt->fld.flowidn_oif  = oldflp->flowidn_oif;
1192         rt->fld.flowidn_iif  = 0;
1193         rt->fld.flowidn_mark = oldflp->flowidn_mark;
1194
1195         rt->rt_saddr      = fld.saddr;
1196         rt->rt_daddr      = fld.daddr;
1197         rt->rt_gateway    = gateway ? gateway : fld.daddr;
1198         rt->rt_local_src  = fld.saddr;
1199
1200         rt->rt_dst_map    = fld.daddr;
1201         rt->rt_src_map    = fld.saddr;
1202
1203         rt->n = neigh;
1204         neigh = NULL;
1205
1206         rt->dst.lastuse = jiffies;
1207         rt->dst.output  = dn_output;
1208         rt->dst.input   = dn_rt_bug;
1209         rt->rt_flags      = flags;
1210         if (flags & RTCF_LOCAL)
1211                 rt->dst.input = dn_nsp_rx;
1212
1213         err = dn_rt_set_next_hop(rt, &res);
1214         if (err)
1215                 goto e_neighbour;
1216
1217         hash = dn_hash(rt->fld.saddr, rt->fld.daddr);
1218         /* dn_insert_route() increments dst->__refcnt */
1219         dn_insert_route(rt, hash, (struct dn_route **)pprt);
1220
1221 done:
1222         if (neigh)
1223                 neigh_release(neigh);
1224         if (free_res)
1225                 dn_fib_res_put(&res);
1226         if (dev_out)
1227                 dev_put(dev_out);
1228 out:
1229         return err;
1230
1231 e_addr:
1232         err = -EADDRNOTAVAIL;
1233         goto done;
1234 e_inval:
1235         err = -EINVAL;
1236         goto done;
1237 e_nobufs:
1238         err = -ENOBUFS;
1239         goto done;
1240 e_neighbour:
1241         dst_release_immediate(&rt->dst);
1242         goto e_nobufs;
1243 }
1244
1245
1246 /*
1247  * N.B. The flags may be moved into the flowi at some future stage.
1248  */
1249 static int __dn_route_output_key(struct dst_entry **pprt, const struct flowidn *flp, int flags)
1250 {
1251         unsigned int hash = dn_hash(flp->saddr, flp->daddr);
1252         struct dn_route *rt = NULL;
1253
1254         if (!(flags & MSG_TRYHARD)) {
1255                 rcu_read_lock_bh();
1256                 for (rt = rcu_dereference_bh(dn_rt_hash_table[hash].chain); rt;
1257                         rt = rcu_dereference_bh(rt->dst.dn_next)) {
1258                         if ((flp->daddr == rt->fld.daddr) &&
1259                             (flp->saddr == rt->fld.saddr) &&
1260                             (flp->flowidn_mark == rt->fld.flowidn_mark) &&
1261                             dn_is_output_route(rt) &&
1262                             (rt->fld.flowidn_oif == flp->flowidn_oif)) {
1263                                 dst_use(&rt->dst, jiffies);
1264                                 rcu_read_unlock_bh();
1265                                 *pprt = &rt->dst;
1266                                 return 0;
1267                         }
1268                 }
1269                 rcu_read_unlock_bh();
1270         }
1271
1272         return dn_route_output_slow(pprt, flp, flags);
1273 }
1274
1275 static int dn_route_output_key(struct dst_entry **pprt, struct flowidn *flp, int flags)
1276 {
1277         int err;
1278
1279         err = __dn_route_output_key(pprt, flp, flags);
1280         if (err == 0 && flp->flowidn_proto) {
1281                 *pprt = xfrm_lookup(&init_net, *pprt,
1282                                     flowidn_to_flowi(flp), NULL, 0);
1283                 if (IS_ERR(*pprt)) {
1284                         err = PTR_ERR(*pprt);
1285                         *pprt = NULL;
1286                 }
1287         }
1288         return err;
1289 }
1290
1291 int dn_route_output_sock(struct dst_entry __rcu **pprt, struct flowidn *fl, struct sock *sk, int flags)
1292 {
1293         int err;
1294
1295         err = __dn_route_output_key(pprt, fl, flags & MSG_TRYHARD);
1296         if (err == 0 && fl->flowidn_proto) {
1297                 *pprt = xfrm_lookup(&init_net, *pprt,
1298                                     flowidn_to_flowi(fl), sk, 0);
1299                 if (IS_ERR(*pprt)) {
1300                         err = PTR_ERR(*pprt);
1301                         *pprt = NULL;
1302                 }
1303         }
1304         return err;
1305 }
1306
1307 static int dn_route_input_slow(struct sk_buff *skb)
1308 {
1309         struct dn_route *rt = NULL;
1310         struct dn_skb_cb *cb = DN_SKB_CB(skb);
1311         struct net_device *in_dev = skb->dev;
1312         struct net_device *out_dev = NULL;
1313         struct dn_dev *dn_db;
1314         struct neighbour *neigh = NULL;
1315         unsigned int hash;
1316         int flags = 0;
1317         __le16 gateway = 0;
1318         __le16 local_src = 0;
1319         struct flowidn fld = {
1320                 .daddr = cb->dst,
1321                 .saddr = cb->src,
1322                 .flowidn_scope = RT_SCOPE_UNIVERSE,
1323                 .flowidn_mark = skb->mark,
1324                 .flowidn_iif = skb->dev->ifindex,
1325         };
1326         struct dn_fib_res res = { .fi = NULL, .type = RTN_UNREACHABLE };
1327         int err = -EINVAL;
1328         int free_res = 0;
1329
1330         dev_hold(in_dev);
1331
1332         if ((dn_db = rcu_dereference(in_dev->dn_ptr)) == NULL)
1333                 goto out;
1334
1335         /* Zero source addresses are not allowed */
1336         if (fld.saddr == 0)
1337                 goto out;
1338
1339         /*
1340          * In this case we've just received a packet from a source
1341          * outside ourselves pretending to come from us. We don't
1342          * allow it any further to prevent routing loops, spoofing and
1343          * other nasties. Loopback packets already have the dst attached
1344          * so this only affects packets which have originated elsewhere.
1345          */
1346         err  = -ENOTUNIQ;
1347         if (dn_dev_islocal(in_dev, cb->src))
1348                 goto out;
1349
1350         err = dn_fib_lookup(&fld, &res);
1351         if (err) {
1352                 if (err != -ESRCH)
1353                         goto out;
1354                 /*
1355                  * Is the destination us ?
1356                  */
1357                 if (!dn_dev_islocal(in_dev, cb->dst))
1358                         goto e_inval;
1359
1360                 res.type = RTN_LOCAL;
1361         } else {
1362                 __le16 src_map = fld.saddr;
1363                 free_res = 1;
1364
1365                 out_dev = DN_FIB_RES_DEV(res);
1366                 if (out_dev == NULL) {
1367                         net_crit_ratelimited("Bug in dn_route_input_slow() No output device\n");
1368                         goto e_inval;
1369                 }
1370                 dev_hold(out_dev);
1371
1372                 if (res.r)
1373                         src_map = fld.saddr; /* no NAT support for now */
1374
1375                 gateway = DN_FIB_RES_GW(res);
1376                 if (res.type == RTN_NAT) {
1377                         fld.daddr = dn_fib_rules_map_destination(fld.daddr, &res);
1378                         dn_fib_res_put(&res);
1379                         free_res = 0;
1380                         if (dn_fib_lookup(&fld, &res))
1381                                 goto e_inval;
1382                         free_res = 1;
1383                         if (res.type != RTN_UNICAST)
1384                                 goto e_inval;
1385                         flags |= RTCF_DNAT;
1386                         gateway = fld.daddr;
1387                 }
1388                 fld.saddr = src_map;
1389         }
1390
1391         switch(res.type) {
1392         case RTN_UNICAST:
1393                 /*
1394                  * Forwarding check here, we only check for forwarding
1395                  * being turned off, if you want to only forward intra
1396                  * area, its up to you to set the routing tables up
1397                  * correctly.
1398                  */
1399                 if (dn_db->parms.forwarding == 0)
1400                         goto e_inval;
1401
1402                 if (res.fi->fib_nhs > 1 && fld.flowidn_oif == 0)
1403                         dn_fib_select_multipath(&fld, &res);
1404
1405                 /*
1406                  * Check for out_dev == in_dev. We use the RTCF_DOREDIRECT
1407                  * flag as a hint to set the intra-ethernet bit when
1408                  * forwarding. If we've got NAT in operation, we don't do
1409                  * this optimisation.
1410                  */
1411                 if (out_dev == in_dev && !(flags & RTCF_NAT))
1412                         flags |= RTCF_DOREDIRECT;
1413
1414                 local_src = DN_FIB_RES_PREFSRC(res);
1415
1416         case RTN_BLACKHOLE:
1417         case RTN_UNREACHABLE:
1418                 break;
1419         case RTN_LOCAL:
1420                 flags |= RTCF_LOCAL;
1421                 fld.saddr = cb->dst;
1422                 fld.daddr = cb->src;
1423
1424                 /* Routing tables gave us a gateway */
1425                 if (gateway)
1426                         goto make_route;
1427
1428                 /* Packet was intra-ethernet, so we know its on-link */
1429                 if (cb->rt_flags & DN_RT_F_IE) {
1430                         gateway = cb->src;
1431                         goto make_route;
1432                 }
1433
1434                 /* Use the default router if there is one */
1435                 neigh = neigh_clone(dn_db->router);
1436                 if (neigh) {
1437                         gateway = ((struct dn_neigh *)neigh)->addr;
1438                         goto make_route;
1439                 }
1440
1441                 /* Close eyes and pray */
1442                 gateway = cb->src;
1443                 goto make_route;
1444         default:
1445                 goto e_inval;
1446         }
1447
1448 make_route:
1449         rt = dst_alloc(&dn_dst_ops, out_dev, 1, DST_OBSOLETE_NONE, DST_HOST);
1450         if (rt == NULL)
1451                 goto e_nobufs;
1452
1453         memset(&rt->fld, 0, sizeof(rt->fld));
1454         rt->rt_saddr      = fld.saddr;
1455         rt->rt_daddr      = fld.daddr;
1456         rt->rt_gateway    = fld.daddr;
1457         if (gateway)
1458                 rt->rt_gateway = gateway;
1459         rt->rt_local_src  = local_src ? local_src : rt->rt_saddr;
1460
1461         rt->rt_dst_map    = fld.daddr;
1462         rt->rt_src_map    = fld.saddr;
1463
1464         rt->fld.saddr        = cb->src;
1465         rt->fld.daddr        = cb->dst;
1466         rt->fld.flowidn_oif  = 0;
1467         rt->fld.flowidn_iif  = in_dev->ifindex;
1468         rt->fld.flowidn_mark = fld.flowidn_mark;
1469
1470         rt->n = neigh;
1471         rt->dst.lastuse = jiffies;
1472         rt->dst.output = dn_rt_bug_out;
1473         switch (res.type) {
1474         case RTN_UNICAST:
1475                 rt->dst.input = dn_forward;
1476                 break;
1477         case RTN_LOCAL:
1478                 rt->dst.output = dn_output;
1479                 rt->dst.input = dn_nsp_rx;
1480                 rt->dst.dev = in_dev;
1481                 flags |= RTCF_LOCAL;
1482                 break;
1483         default:
1484         case RTN_UNREACHABLE:
1485         case RTN_BLACKHOLE:
1486                 rt->dst.input = dst_discard;
1487         }
1488         rt->rt_flags = flags;
1489
1490         err = dn_rt_set_next_hop(rt, &res);
1491         if (err)
1492                 goto e_neighbour;
1493
1494         hash = dn_hash(rt->fld.saddr, rt->fld.daddr);
1495         /* dn_insert_route() increments dst->__refcnt */
1496         dn_insert_route(rt, hash, &rt);
1497         skb_dst_set(skb, &rt->dst);
1498
1499 done:
1500         if (neigh)
1501                 neigh_release(neigh);
1502         if (free_res)
1503                 dn_fib_res_put(&res);
1504         dev_put(in_dev);
1505         if (out_dev)
1506                 dev_put(out_dev);
1507 out:
1508         return err;
1509
1510 e_inval:
1511         err = -EINVAL;
1512         goto done;
1513
1514 e_nobufs:
1515         err = -ENOBUFS;
1516         goto done;
1517
1518 e_neighbour:
1519         dst_release_immediate(&rt->dst);
1520         goto done;
1521 }
1522
1523 static int dn_route_input(struct sk_buff *skb)
1524 {
1525         struct dn_route *rt;
1526         struct dn_skb_cb *cb = DN_SKB_CB(skb);
1527         unsigned int hash = dn_hash(cb->src, cb->dst);
1528
1529         if (skb_dst(skb))
1530                 return 0;
1531
1532         rcu_read_lock();
1533         for(rt = rcu_dereference(dn_rt_hash_table[hash].chain); rt != NULL;
1534             rt = rcu_dereference(rt->dst.dn_next)) {
1535                 if ((rt->fld.saddr == cb->src) &&
1536                     (rt->fld.daddr == cb->dst) &&
1537                     (rt->fld.flowidn_oif == 0) &&
1538                     (rt->fld.flowidn_mark == skb->mark) &&
1539                     (rt->fld.flowidn_iif == cb->iif)) {
1540                         dst_use(&rt->dst, jiffies);
1541                         rcu_read_unlock();
1542                         skb_dst_set(skb, (struct dst_entry *)rt);
1543                         return 0;
1544                 }
1545         }
1546         rcu_read_unlock();
1547
1548         return dn_route_input_slow(skb);
1549 }
1550
1551 static int dn_rt_fill_info(struct sk_buff *skb, u32 portid, u32 seq,
1552                            int event, int nowait, unsigned int flags)
1553 {
1554         struct dn_route *rt = (struct dn_route *)skb_dst(skb);
1555         struct rtmsg *r;
1556         struct nlmsghdr *nlh;
1557         long expires;
1558
1559         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*r), flags);
1560         if (!nlh)
1561                 return -EMSGSIZE;
1562
1563         r = nlmsg_data(nlh);
1564         r->rtm_family = AF_DECnet;
1565         r->rtm_dst_len = 16;
1566         r->rtm_src_len = 0;
1567         r->rtm_tos = 0;
1568         r->rtm_table = RT_TABLE_MAIN;
1569         r->rtm_type = rt->rt_type;
1570         r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
1571         r->rtm_scope = RT_SCOPE_UNIVERSE;
1572         r->rtm_protocol = RTPROT_UNSPEC;
1573
1574         if (rt->rt_flags & RTCF_NOTIFY)
1575                 r->rtm_flags |= RTM_F_NOTIFY;
1576
1577         if (nla_put_u32(skb, RTA_TABLE, RT_TABLE_MAIN) < 0 ||
1578             nla_put_le16(skb, RTA_DST, rt->rt_daddr) < 0)
1579                 goto errout;
1580
1581         if (rt->fld.saddr) {
1582                 r->rtm_src_len = 16;
1583                 if (nla_put_le16(skb, RTA_SRC, rt->fld.saddr) < 0)
1584                         goto errout;
1585         }
1586         if (rt->dst.dev &&
1587             nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex) < 0)
1588                 goto errout;
1589
1590         /*
1591          * Note to self - change this if input routes reverse direction when
1592          * they deal only with inputs and not with replies like they do
1593          * currently.
1594          */
1595         if (nla_put_le16(skb, RTA_PREFSRC, rt->rt_local_src) < 0)
1596                 goto errout;
1597
1598         if (rt->rt_daddr != rt->rt_gateway &&
1599             nla_put_le16(skb, RTA_GATEWAY, rt->rt_gateway) < 0)
1600                 goto errout;
1601
1602         if (rtnetlink_put_metrics(skb, dst_metrics_ptr(&rt->dst)) < 0)
1603                 goto errout;
1604
1605         expires = rt->dst.expires ? rt->dst.expires - jiffies : 0;
1606         if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires,
1607                                rt->dst.error) < 0)
1608                 goto errout;
1609
1610         if (dn_is_input_route(rt) &&
1611             nla_put_u32(skb, RTA_IIF, rt->fld.flowidn_iif) < 0)
1612                 goto errout;
1613
1614         nlmsg_end(skb, nlh);
1615         return 0;
1616
1617 errout:
1618         nlmsg_cancel(skb, nlh);
1619         return -EMSGSIZE;
1620 }
1621
1622 const struct nla_policy rtm_dn_policy[RTA_MAX + 1] = {
1623         [RTA_DST]               = { .type = NLA_U16 },
1624         [RTA_SRC]               = { .type = NLA_U16 },
1625         [RTA_IIF]               = { .type = NLA_U32 },
1626         [RTA_OIF]               = { .type = NLA_U32 },
1627         [RTA_GATEWAY]           = { .type = NLA_U16 },
1628         [RTA_PRIORITY]          = { .type = NLA_U32 },
1629         [RTA_PREFSRC]           = { .type = NLA_U16 },
1630         [RTA_METRICS]           = { .type = NLA_NESTED },
1631         [RTA_MULTIPATH]         = { .type = NLA_NESTED },
1632         [RTA_TABLE]             = { .type = NLA_U32 },
1633         [RTA_MARK]              = { .type = NLA_U32 },
1634 };
1635
1636 /*
1637  * This is called by both endnodes and routers now.
1638  */
1639 static int dn_cache_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
1640                              struct netlink_ext_ack *extack)
1641 {
1642         struct net *net = sock_net(in_skb->sk);
1643         struct rtmsg *rtm = nlmsg_data(nlh);
1644         struct dn_route *rt = NULL;
1645         struct dn_skb_cb *cb;
1646         int err;
1647         struct sk_buff *skb;
1648         struct flowidn fld;
1649         struct nlattr *tb[RTA_MAX+1];
1650
1651         if (!net_eq(net, &init_net))
1652                 return -EINVAL;
1653
1654         err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_dn_policy,
1655                           extack);
1656         if (err < 0)
1657                 return err;
1658
1659         memset(&fld, 0, sizeof(fld));
1660         fld.flowidn_proto = DNPROTO_NSP;
1661
1662         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1663         if (skb == NULL)
1664                 return -ENOBUFS;
1665         skb_reset_mac_header(skb);
1666         cb = DN_SKB_CB(skb);
1667
1668         if (tb[RTA_SRC])
1669                 fld.saddr = nla_get_le16(tb[RTA_SRC]);
1670
1671         if (tb[RTA_DST])
1672                 fld.daddr = nla_get_le16(tb[RTA_DST]);
1673
1674         if (tb[RTA_IIF])
1675                 fld.flowidn_iif = nla_get_u32(tb[RTA_IIF]);
1676
1677         if (fld.flowidn_iif) {
1678                 struct net_device *dev;
1679                 dev = __dev_get_by_index(&init_net, fld.flowidn_iif);
1680                 if (!dev || !dev->dn_ptr) {
1681                         kfree_skb(skb);
1682                         return -ENODEV;
1683                 }
1684                 skb->protocol = htons(ETH_P_DNA_RT);
1685                 skb->dev = dev;
1686                 cb->src = fld.saddr;
1687                 cb->dst = fld.daddr;
1688                 local_bh_disable();
1689                 err = dn_route_input(skb);
1690                 local_bh_enable();
1691                 memset(cb, 0, sizeof(struct dn_skb_cb));
1692                 rt = (struct dn_route *)skb_dst(skb);
1693                 if (!err && -rt->dst.error)
1694                         err = rt->dst.error;
1695         } else {
1696                 if (tb[RTA_OIF])
1697                         fld.flowidn_oif = nla_get_u32(tb[RTA_OIF]);
1698
1699                 err = dn_route_output_key((struct dst_entry **)&rt, &fld, 0);
1700         }
1701
1702         skb->dev = NULL;
1703         if (err)
1704                 goto out_free;
1705         skb_dst_set(skb, &rt->dst);
1706         if (rtm->rtm_flags & RTM_F_NOTIFY)
1707                 rt->rt_flags |= RTCF_NOTIFY;
1708
1709         err = dn_rt_fill_info(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq, RTM_NEWROUTE, 0, 0);
1710         if (err < 0) {
1711                 err = -EMSGSIZE;
1712                 goto out_free;
1713         }
1714
1715         return rtnl_unicast(skb, &init_net, NETLINK_CB(in_skb).portid);
1716
1717 out_free:
1718         kfree_skb(skb);
1719         return err;
1720 }
1721
1722 /*
1723  * For routers, this is called from dn_fib_dump, but for endnodes its
1724  * called directly from the rtnetlink dispatch table.
1725  */
1726 int dn_cache_dump(struct sk_buff *skb, struct netlink_callback *cb)
1727 {
1728         struct net *net = sock_net(skb->sk);
1729         struct dn_route *rt;
1730         int h, s_h;
1731         int idx, s_idx;
1732         struct rtmsg *rtm;
1733
1734         if (!net_eq(net, &init_net))
1735                 return 0;
1736
1737         if (nlmsg_len(cb->nlh) < sizeof(struct rtmsg))
1738                 return -EINVAL;
1739
1740         rtm = nlmsg_data(cb->nlh);
1741         if (!(rtm->rtm_flags & RTM_F_CLONED))
1742                 return 0;
1743
1744         s_h = cb->args[0];
1745         s_idx = idx = cb->args[1];
1746         for(h = 0; h <= dn_rt_hash_mask; h++) {
1747                 if (h < s_h)
1748                         continue;
1749                 if (h > s_h)
1750                         s_idx = 0;
1751                 rcu_read_lock_bh();
1752                 for(rt = rcu_dereference_bh(dn_rt_hash_table[h].chain), idx = 0;
1753                         rt;
1754                         rt = rcu_dereference_bh(rt->dst.dn_next), idx++) {
1755                         if (idx < s_idx)
1756                                 continue;
1757                         skb_dst_set(skb, dst_clone(&rt->dst));
1758                         if (dn_rt_fill_info(skb, NETLINK_CB(cb->skb).portid,
1759                                         cb->nlh->nlmsg_seq, RTM_NEWROUTE,
1760                                         1, NLM_F_MULTI) < 0) {
1761                                 skb_dst_drop(skb);
1762                                 rcu_read_unlock_bh();
1763                                 goto done;
1764                         }
1765                         skb_dst_drop(skb);
1766                 }
1767                 rcu_read_unlock_bh();
1768         }
1769
1770 done:
1771         cb->args[0] = h;
1772         cb->args[1] = idx;
1773         return skb->len;
1774 }
1775
1776 #ifdef CONFIG_PROC_FS
1777 struct dn_rt_cache_iter_state {
1778         int bucket;
1779 };
1780
1781 static struct dn_route *dn_rt_cache_get_first(struct seq_file *seq)
1782 {
1783         struct dn_route *rt = NULL;
1784         struct dn_rt_cache_iter_state *s = seq->private;
1785
1786         for(s->bucket = dn_rt_hash_mask; s->bucket >= 0; --s->bucket) {
1787                 rcu_read_lock_bh();
1788                 rt = rcu_dereference_bh(dn_rt_hash_table[s->bucket].chain);
1789                 if (rt)
1790                         break;
1791                 rcu_read_unlock_bh();
1792         }
1793         return rt;
1794 }
1795
1796 static struct dn_route *dn_rt_cache_get_next(struct seq_file *seq, struct dn_route *rt)
1797 {
1798         struct dn_rt_cache_iter_state *s = seq->private;
1799
1800         rt = rcu_dereference_bh(rt->dst.dn_next);
1801         while (!rt) {
1802                 rcu_read_unlock_bh();
1803                 if (--s->bucket < 0)
1804                         break;
1805                 rcu_read_lock_bh();
1806                 rt = rcu_dereference_bh(dn_rt_hash_table[s->bucket].chain);
1807         }
1808         return rt;
1809 }
1810
1811 static void *dn_rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
1812 {
1813         struct dn_route *rt = dn_rt_cache_get_first(seq);
1814
1815         if (rt) {
1816                 while(*pos && (rt = dn_rt_cache_get_next(seq, rt)))
1817                         --*pos;
1818         }
1819         return *pos ? NULL : rt;
1820 }
1821
1822 static void *dn_rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1823 {
1824         struct dn_route *rt = dn_rt_cache_get_next(seq, v);
1825         ++*pos;
1826         return rt;
1827 }
1828
1829 static void dn_rt_cache_seq_stop(struct seq_file *seq, void *v)
1830 {
1831         if (v)
1832                 rcu_read_unlock_bh();
1833 }
1834
1835 static int dn_rt_cache_seq_show(struct seq_file *seq, void *v)
1836 {
1837         struct dn_route *rt = v;
1838         char buf1[DN_ASCBUF_LEN], buf2[DN_ASCBUF_LEN];
1839
1840         seq_printf(seq, "%-8s %-7s %-7s %04d %04d %04d\n",
1841                    rt->dst.dev ? rt->dst.dev->name : "*",
1842                    dn_addr2asc(le16_to_cpu(rt->rt_daddr), buf1),
1843                    dn_addr2asc(le16_to_cpu(rt->rt_saddr), buf2),
1844                    atomic_read(&rt->dst.__refcnt),
1845                    rt->dst.__use, 0);
1846         return 0;
1847 }
1848
1849 static const struct seq_operations dn_rt_cache_seq_ops = {
1850         .start  = dn_rt_cache_seq_start,
1851         .next   = dn_rt_cache_seq_next,
1852         .stop   = dn_rt_cache_seq_stop,
1853         .show   = dn_rt_cache_seq_show,
1854 };
1855
1856 static int dn_rt_cache_seq_open(struct inode *inode, struct file *file)
1857 {
1858         return seq_open_private(file, &dn_rt_cache_seq_ops,
1859                         sizeof(struct dn_rt_cache_iter_state));
1860 }
1861
1862 static const struct file_operations dn_rt_cache_seq_fops = {
1863         .owner   = THIS_MODULE,
1864         .open    = dn_rt_cache_seq_open,
1865         .read    = seq_read,
1866         .llseek  = seq_lseek,
1867         .release = seq_release_private,
1868 };
1869
1870 #endif /* CONFIG_PROC_FS */
1871
1872 void __init dn_route_init(void)
1873 {
1874         int i, goal, order;
1875
1876         dn_dst_ops.kmem_cachep =
1877                 kmem_cache_create("dn_dst_cache", sizeof(struct dn_route), 0,
1878                                   SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1879         dst_entries_init(&dn_dst_ops);
1880         setup_timer(&dn_route_timer, dn_dst_check_expire, 0);
1881         dn_route_timer.expires = jiffies + decnet_dst_gc_interval * HZ;
1882         add_timer(&dn_route_timer);
1883
1884         goal = totalram_pages >> (26 - PAGE_SHIFT);
1885
1886         for(order = 0; (1UL << order) < goal; order++)
1887                 /* NOTHING */;
1888
1889         /*
1890          * Only want 1024 entries max, since the table is very, very unlikely
1891          * to be larger than that.
1892          */
1893         while(order && ((((1UL << order) * PAGE_SIZE) /
1894                                 sizeof(struct dn_rt_hash_bucket)) >= 2048))
1895                 order--;
1896
1897         do {
1898                 dn_rt_hash_mask = (1UL << order) * PAGE_SIZE /
1899                         sizeof(struct dn_rt_hash_bucket);
1900                 while(dn_rt_hash_mask & (dn_rt_hash_mask - 1))
1901                         dn_rt_hash_mask--;
1902                 dn_rt_hash_table = (struct dn_rt_hash_bucket *)
1903                         __get_free_pages(GFP_ATOMIC, order);
1904         } while (dn_rt_hash_table == NULL && --order > 0);
1905
1906         if (!dn_rt_hash_table)
1907                 panic("Failed to allocate DECnet route cache hash table\n");
1908
1909         printk(KERN_INFO
1910                 "DECnet: Routing cache hash table of %u buckets, %ldKbytes\n",
1911                 dn_rt_hash_mask,
1912                 (long)(dn_rt_hash_mask*sizeof(struct dn_rt_hash_bucket))/1024);
1913
1914         dn_rt_hash_mask--;
1915         for(i = 0; i <= dn_rt_hash_mask; i++) {
1916                 spin_lock_init(&dn_rt_hash_table[i].lock);
1917                 dn_rt_hash_table[i].chain = NULL;
1918         }
1919
1920         dn_dst_ops.gc_thresh = (dn_rt_hash_mask + 1);
1921
1922         proc_create("decnet_cache", S_IRUGO, init_net.proc_net,
1923                     &dn_rt_cache_seq_fops);
1924
1925 #ifdef CONFIG_DECNET_ROUTER
1926         rtnl_register(PF_DECnet, RTM_GETROUTE, dn_cache_getroute,
1927                       dn_fib_dump, 0);
1928 #else
1929         rtnl_register(PF_DECnet, RTM_GETROUTE, dn_cache_getroute,
1930                       dn_cache_dump, 0);
1931 #endif
1932 }
1933
1934 void __exit dn_route_cleanup(void)
1935 {
1936         del_timer(&dn_route_timer);
1937         dn_run_flush(0);
1938
1939         remove_proc_entry("decnet_cache", init_net.proc_net);
1940         dst_entries_destroy(&dn_dst_ops);
1941 }
1942