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