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 <asm/errno.h>
  81 #include <net/net_namespace.h>
  82 #include <net/netlink.h>
  83 #include <net/neighbour.h>
  84 #include <net/dst.h>
  85 #include <net/flow.h>
  86 #include <net/fib_rules.h>
  87 #include <net/dn.h>
  88 #include <net/dn_dev.h>
  89 #include <net/dn_nsp.h>
  90 #include <net/dn_route.h>
  91 #include <net/dn_neigh.h>
  92 #include <net/dn_fib.h>
  93
  94 struct dn_rt_hash_bucket
  95 {
  96         struct dn_route *chain;
  97         spinlock_t lock;
  98 };
  99
 100 extern struct neigh_table dn_neigh_table;
 101
 102
 103 static unsigned char dn_hiord_addr[6] = {0xAA,0x00,0x04,0x00,0x00,0x00};
 104
 105 static const int dn_rt_min_delay = 2 * HZ;
 106 static const int dn_rt_max_delay = 10 * HZ;
 107 static const int dn_rt_mtu_expires = 10 * 60 * HZ;
 108
 109 static unsigned long dn_rt_deadline;
 110
 111 static int dn_dst_gc(struct dst_ops *ops);
 112 static struct dst_entry *dn_dst_check(struct dst_entry *, __u32);
 113 static struct dst_entry *dn_dst_negative_advice(struct dst_entry *);
 114 static void dn_dst_link_failure(struct sk_buff *);
 115 static void dn_dst_update_pmtu(struct dst_entry *dst, u32 mtu);
 116 static int dn_route_input(struct sk_buff *);
 117 static void dn_run_flush(unsigned long dummy);
 118
 119 static struct dn_rt_hash_bucket *dn_rt_hash_table;
 120 static unsigned dn_rt_hash_mask;
 121
 122 static struct timer_list dn_route_timer;
 123 static DEFINE_TIMER(dn_rt_flush_timer, dn_run_flush, 0, 0);
 124 int decnet_dst_gc_interval = 2;
 125
 126 static struct dst_ops dn_dst_ops = {
 127         .family =               PF_DECnet,
 128         .protocol =             cpu_to_be16(ETH_P_DNA_RT),
 129         .gc_thresh =            128,
 130         .gc =                   dn_dst_gc,
 131         .check =                dn_dst_check,
 132         .negative_advice =      dn_dst_negative_advice,
 133         .link_failure =         dn_dst_link_failure,
 134         .update_pmtu =          dn_dst_update_pmtu,
 135         .entries =              ATOMIC_INIT(0),
 136 };
 137
 138 static __inline__ unsigned dn_hash(__le16 src, __le16 dst)
 139 {
 140         __u16 tmp = (__u16 __force)(src ^ dst);
 141         tmp ^= (tmp >> 3);
 142         tmp ^= (tmp >> 5);
 143         tmp ^= (tmp >> 10);
 144         return dn_rt_hash_mask & (unsigned)tmp;
 145 }
 146
 147 static inline void dnrt_free(struct dn_route *rt)
 148 {
 149         call_rcu_bh(&rt->u.dst.rcu_head, dst_rcu_free);
 150 }
 151
 152 static inline void dnrt_drop(struct dn_route *rt)
 153 {
 154         dst_release(&rt->u.dst);
 155         call_rcu_bh(&rt->u.dst.rcu_head, dst_rcu_free);
 156 }
 157
 158 static void dn_dst_check_expire(unsigned long dummy)
 159 {
 160         int i;
 161         struct dn_route *rt, **rtp;
 162         unsigned long now = jiffies;
 163         unsigned long expire = 120 * HZ;
 164
 165         for(i = 0; i <= dn_rt_hash_mask; i++) {
 166                 rtp = &dn_rt_hash_table[i].chain;
 167
 168                 spin_lock(&dn_rt_hash_table[i].lock);
 169                 while((rt=*rtp) != NULL) {
 170                         if (atomic_read(&rt->u.dst.__refcnt) ||
 171                                         (now - rt->u.dst.lastuse) < expire) {
 172                                 rtp = &rt->u.dst.dn_next;
 173                                 continue;
 174                         }
 175                         *rtp = rt->u.dst.dn_next;
 176                         rt->u.dst.dn_next = NULL;
 177                         dnrt_free(rt);
 178                 }
 179                 spin_unlock(&dn_rt_hash_table[i].lock);
 180
 181                 if ((jiffies - now) > 0)
 182                         break;
 183         }
 184
 185         mod_timer(&dn_route_timer, now + decnet_dst_gc_interval * HZ);
 186 }
 187
 188 static int dn_dst_gc(struct dst_ops *ops)
 189 {
 190         struct dn_route *rt, **rtp;
 191         int i;
 192         unsigned long now = jiffies;
 193         unsigned long expire = 10 * HZ;
 194
 195         for(i = 0; i <= dn_rt_hash_mask; i++) {
 196
 197                 spin_lock_bh(&dn_rt_hash_table[i].lock);
 198                 rtp = &dn_rt_hash_table[i].chain;
 199
 200                 while((rt=*rtp) != NULL) {
 201                         if (atomic_read(&rt->u.dst.__refcnt) ||
 202                                         (now - rt->u.dst.lastuse) < expire) {
 203                                 rtp = &rt->u.dst.dn_next;
 204                                 continue;
 205                         }
 206                         *rtp = rt->u.dst.dn_next;
 207                         rt->u.dst.dn_next = NULL;
 208                         dnrt_drop(rt);
 209                         break;
 210                 }
 211                 spin_unlock_bh(&dn_rt_hash_table[i].lock);
 212         }
 213
 214         return 0;
 215 }
 216
 217 /*
 218  * The decnet standards don't impose a particular minimum mtu, what they
 219  * do insist on is that the routing layer accepts a datagram of at least
 220  * 230 bytes long. Here we have to subtract the routing header length from
 221  * 230 to get the minimum acceptable mtu. If there is no neighbour, then we
 222  * assume the worst and use a long header size.
 223  *
 224  * We update both the mtu and the advertised mss (i.e. the segment size we
 225  * advertise to the other end).
 226  */
 227 static void dn_dst_update_pmtu(struct dst_entry *dst, u32 mtu)
 228 {
 229         u32 min_mtu = 230;
 230         struct dn_dev *dn = dst->neighbour ?
 231                             (struct dn_dev *)dst->neighbour->dev->dn_ptr : NULL;
 232
 233         if (dn && dn->use_long == 0)
 234                 min_mtu -= 6;
 235         else
 236                 min_mtu -= 21;
 237
 238         if (dst_metric(dst, RTAX_MTU) > mtu && mtu >= min_mtu) {
 239                 if (!(dst_metric_locked(dst, RTAX_MTU))) {
 240                         dst->metrics[RTAX_MTU-1] = mtu;
 241                         dst_set_expires(dst, dn_rt_mtu_expires);
 242                 }
 243                 if (!(dst_metric_locked(dst, RTAX_ADVMSS))) {
 244                         u32 mss = mtu - DN_MAX_NSP_DATA_HEADER;
 245                         if (dst_metric(dst, RTAX_ADVMSS) > mss)
 246                                 dst->metrics[RTAX_ADVMSS-1] = mss;
 247                 }
 248         }
 249 }
 250
 251 /*
 252  * When a route has been marked obsolete. (e.g. routing cache flush)
 253  */
 254 static struct dst_entry *dn_dst_check(struct dst_entry *dst, __u32 cookie)
 255 {
 256         return NULL;
 257 }
 258
 259 static struct dst_entry *dn_dst_negative_advice(struct dst_entry *dst)
 260 {
 261         dst_release(dst);
 262         return NULL;
 263 }
 264
 265 static void dn_dst_link_failure(struct sk_buff *skb)
 266 {
 267         return;
 268 }
 269
 270 static inline int compare_keys(struct flowi *fl1, struct flowi *fl2)
 271 {
 272         return ((fl1->nl_u.dn_u.daddr ^ fl2->nl_u.dn_u.daddr) |
 273                 (fl1->nl_u.dn_u.saddr ^ fl2->nl_u.dn_u.saddr) |
 274                 (fl1->mark ^ fl2->mark) |
 275                 (fl1->nl_u.dn_u.scope ^ fl2->nl_u.dn_u.scope) |
 276                 (fl1->oif ^ fl2->oif) |
 277                 (fl1->iif ^ fl2->iif)) == 0;
 278 }
 279
 280 static int dn_insert_route(struct dn_route *rt, unsigned hash, struct dn_route **rp)
 281 {
 282         struct dn_route *rth, **rthp;
 283         unsigned long now = jiffies;
 284
 285         rthp = &dn_rt_hash_table[hash].chain;
 286
 287         spin_lock_bh(&dn_rt_hash_table[hash].lock);
 288         while((rth = *rthp) != NULL) {
 289                 if (compare_keys(&rth->fl, &rt->fl)) {
 290                         /* Put it first */
 291                         *rthp = rth->u.dst.dn_next;
 292                         rcu_assign_pointer(rth->u.dst.dn_next,
 293                                            dn_rt_hash_table[hash].chain);
 294                         rcu_assign_pointer(dn_rt_hash_table[hash].chain, rth);
 295
 296                         dst_use(&rth->u.dst, now);
 297                         spin_unlock_bh(&dn_rt_hash_table[hash].lock);
 298
 299                         dnrt_drop(rt);
 300                         *rp = rth;
 301                         return 0;
 302                 }
 303                 rthp = &rth->u.dst.dn_next;
 304         }
 305
 306         rcu_assign_pointer(rt->u.dst.dn_next, dn_rt_hash_table[hash].chain);
 307         rcu_assign_pointer(dn_rt_hash_table[hash].chain, rt);
 308
 309         dst_use(&rt->u.dst, now);
 310         spin_unlock_bh(&dn_rt_hash_table[hash].lock);
 311         *rp = rt;
 312         return 0;
 313 }
 314
 315 static void dn_run_flush(unsigned long dummy)
 316 {
 317         int i;
 318         struct dn_route *rt, *next;
 319
 320         for(i = 0; i < dn_rt_hash_mask; i++) {
 321                 spin_lock_bh(&dn_rt_hash_table[i].lock);
 322
 323                 if ((rt = xchg(&dn_rt_hash_table[i].chain, NULL)) == NULL)
 324                         goto nothing_to_declare;
 325
 326                 for(; rt; rt=next) {
 327                         next = rt->u.dst.dn_next;
 328                         rt->u.dst.dn_next = NULL;
 329                         dst_free((struct dst_entry *)rt);
 330                 }
 331
 332 nothing_to_declare:
 333                 spin_unlock_bh(&dn_rt_hash_table[i].lock);
 334         }
 335 }
 336
 337 static DEFINE_SPINLOCK(dn_rt_flush_lock);
 338
 339 void dn_rt_cache_flush(int delay)
 340 {
 341         unsigned long now = jiffies;
 342         int user_mode = !in_interrupt();
 343
 344         if (delay < 0)
 345                 delay = dn_rt_min_delay;
 346
 347         spin_lock_bh(&dn_rt_flush_lock);
 348
 349         if (del_timer(&dn_rt_flush_timer) && delay > 0 && dn_rt_deadline) {
 350                 long tmo = (long)(dn_rt_deadline - now);
 351
 352                 if (user_mode && tmo < dn_rt_max_delay - dn_rt_min_delay)
 353                         tmo = 0;
 354
 355                 if (delay > tmo)
 356                         delay = tmo;
 357         }
 358
 359         if (delay <= 0) {
 360                 spin_unlock_bh(&dn_rt_flush_lock);
 361                 dn_run_flush(0);
 362                 return;
 363         }
 364
 365         if (dn_rt_deadline == 0)
 366                 dn_rt_deadline = now + dn_rt_max_delay;
 367
 368         dn_rt_flush_timer.expires = now + delay;
 369         add_timer(&dn_rt_flush_timer);
 370         spin_unlock_bh(&dn_rt_flush_lock);
 371 }
 372
 373 /**
 374  * dn_return_short - Return a short packet to its sender
 375  * @skb: The packet to return
 376  *
 377  */
 378 static int dn_return_short(struct sk_buff *skb)
 379 {
 380         struct dn_skb_cb *cb;
 381         unsigned char *ptr;
 382         __le16 *src;
 383         __le16 *dst;
 384
 385         /* Add back headers */
 386         skb_push(skb, skb->data - skb_network_header(skb));
 387
 388         if ((skb = skb_unshare(skb, GFP_ATOMIC)) == NULL)
 389                 return NET_RX_DROP;
 390
 391         cb = DN_SKB_CB(skb);
 392         /* Skip packet length and point to flags */
 393         ptr = skb->data + 2;
 394         *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
 395
 396         dst = (__le16 *)ptr;
 397         ptr += 2;
 398         src = (__le16 *)ptr;
 399         ptr += 2;
 400         *ptr = 0; /* Zero hop count */
 401
 402         swap(*src, *dst);
 403
 404         skb->pkt_type = PACKET_OUTGOING;
 405         dn_rt_finish_output(skb, NULL, NULL);
 406         return NET_RX_SUCCESS;
 407 }
 408
 409 /**
 410  * dn_return_long - Return a long packet to its sender
 411  * @skb: The long format packet to return
 412  *
 413  */
 414 static int dn_return_long(struct sk_buff *skb)
 415 {
 416         struct dn_skb_cb *cb;
 417         unsigned char *ptr;
 418         unsigned char *src_addr, *dst_addr;
 419         unsigned char tmp[ETH_ALEN];
 420
 421         /* Add back all headers */
 422         skb_push(skb, skb->data - skb_network_header(skb));
 423
 424         if ((skb = skb_unshare(skb, GFP_ATOMIC)) == NULL)
 425                 return NET_RX_DROP;
 426
 427         cb = DN_SKB_CB(skb);
 428         /* Ignore packet length and point to flags */
 429         ptr = skb->data + 2;
 430
 431         /* Skip padding */
 432         if (*ptr & DN_RT_F_PF) {
 433                 char padlen = (*ptr & ~DN_RT_F_PF);
 434                 ptr += padlen;
 435         }
 436
 437         *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
 438         ptr += 2;
 439         dst_addr = ptr;
 440         ptr += 8;
 441         src_addr = ptr;
 442         ptr += 6;
 443         *ptr = 0; /* Zero hop count */
 444
 445         /* Swap source and destination */
 446         memcpy(tmp, src_addr, ETH_ALEN);
 447         memcpy(src_addr, dst_addr, ETH_ALEN);
 448         memcpy(dst_addr, tmp, ETH_ALEN);
 449
 450         skb->pkt_type = PACKET_OUTGOING;
 451         dn_rt_finish_output(skb, dst_addr, src_addr);
 452         return NET_RX_SUCCESS;
 453 }
 454
 455 /**
 456  * dn_route_rx_packet - Try and find a route for an incoming packet
 457  * @skb: The packet to find a route for
 458  *
 459  * Returns: result of input function if route is found, error code otherwise
 460  */
 461 static int dn_route_rx_packet(struct sk_buff *skb)
 462 {
 463         struct dn_skb_cb *cb = DN_SKB_CB(skb);
 464         int err;
 465
 466         if ((err = dn_route_input(skb)) == 0)
 467                 return dst_input(skb);
 468
 469         if (decnet_debug_level & 4) {
 470                 char *devname = skb->dev ? skb->dev->name : "???";
 471                 struct dn_skb_cb *cb = DN_SKB_CB(skb);
 472                 printk(KERN_DEBUG
 473                         "DECnet: dn_route_rx_packet: rt_flags=0x%02x dev=%s len=%d src=0x%04hx dst=0x%04hx err=%d type=%d\n",
 474                         (int)cb->rt_flags, devname, skb->len,
 475                         le16_to_cpu(cb->src), le16_to_cpu(cb->dst),
 476                         err, skb->pkt_type);
 477         }
 478
 479         if ((skb->pkt_type == PACKET_HOST) && (cb->rt_flags & DN_RT_F_RQR)) {
 480                 switch(cb->rt_flags & DN_RT_PKT_MSK) {
 481                         case DN_RT_PKT_SHORT:
 482                                 return dn_return_short(skb);
 483                         case DN_RT_PKT_LONG:
 484                                 return dn_return_long(skb);
 485                 }
 486         }
 487
 488         kfree_skb(skb);
 489         return NET_RX_DROP;
 490 }
 491
 492 static int dn_route_rx_long(struct sk_buff *skb)
 493 {
 494         struct dn_skb_cb *cb = DN_SKB_CB(skb);
 495         unsigned char *ptr = skb->data;
 496
 497         if (!pskb_may_pull(skb, 21)) /* 20 for long header, 1 for shortest nsp */
 498                 goto drop_it;
 499
 500         skb_pull(skb, 20);
 501         skb_reset_transport_header(skb);
 502
 503         /* Destination info */
 504         ptr += 2;
 505         cb->dst = dn_eth2dn(ptr);
 506         if (memcmp(ptr, dn_hiord_addr, 4) != 0)
 507                 goto drop_it;
 508         ptr += 6;
 509
 510
 511         /* Source info */
 512         ptr += 2;
 513         cb->src = dn_eth2dn(ptr);
 514         if (memcmp(ptr, dn_hiord_addr, 4) != 0)
 515                 goto drop_it;
 516         ptr += 6;
 517         /* Other junk */
 518         ptr++;
 519         cb->hops = *ptr++; /* Visit Count */
 520
 521         return NF_HOOK(PF_DECnet, NF_DN_PRE_ROUTING, skb, skb->dev, NULL, dn_route_rx_packet);
 522
 523 drop_it:
 524         kfree_skb(skb);
 525         return NET_RX_DROP;
 526 }
 527
 528
 529
 530 static int dn_route_rx_short(struct sk_buff *skb)
 531 {
 532         struct dn_skb_cb *cb = DN_SKB_CB(skb);
 533         unsigned char *ptr = skb->data;
 534
 535         if (!pskb_may_pull(skb, 6)) /* 5 for short header + 1 for shortest nsp */
 536                 goto drop_it;
 537
 538         skb_pull(skb, 5);
 539         skb_reset_transport_header(skb);
 540
 541         cb->dst = *(__le16 *)ptr;
 542         ptr += 2;
 543         cb->src = *(__le16 *)ptr;
 544         ptr += 2;
 545         cb->hops = *ptr & 0x3f;
 546
 547         return NF_HOOK(PF_DECnet, NF_DN_PRE_ROUTING, skb, skb->dev, NULL, dn_route_rx_packet);
 548
 549 drop_it:
 550         kfree_skb(skb);
 551         return NET_RX_DROP;
 552 }
 553
 554 static int dn_route_discard(struct sk_buff *skb)
 555 {
 556         /*
 557          * I know we drop the packet here, but thats considered success in
 558          * this case
 559          */
 560         kfree_skb(skb);
 561         return NET_RX_SUCCESS;
 562 }
 563
 564 static int dn_route_ptp_hello(struct sk_buff *skb)
 565 {
 566         dn_dev_hello(skb);
 567         dn_neigh_pointopoint_hello(skb);
 568         return NET_RX_SUCCESS;
 569 }
 570
 571 int dn_route_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
 572 {
 573         struct dn_skb_cb *cb;
 574         unsigned char flags = 0;
 575         __u16 len = le16_to_cpu(*(__le16 *)skb->data);
 576         struct dn_dev *dn = (struct dn_dev *)dev->dn_ptr;
 577         unsigned char padlen = 0;
 578
 579         if (!net_eq(dev_net(dev), &init_net))
 580                 goto dump_it;
 581
 582         if (dn == NULL)
 583                 goto dump_it;
 584
 585         if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
 586                 goto out;
 587
 588         if (!pskb_may_pull(skb, 3))
 589                 goto dump_it;
 590
 591         skb_pull(skb, 2);
 592
 593         if (len > skb->len)
 594                 goto dump_it;
 595
 596         skb_trim(skb, len);
 597
 598         flags = *skb->data;
 599
 600         cb = DN_SKB_CB(skb);
 601         cb->stamp = jiffies;
 602         cb->iif = dev->ifindex;
 603
 604         /*
 605          * If we have padding, remove it.
 606          */
 607         if (flags & DN_RT_F_PF) {
 608                 padlen = flags & ~DN_RT_F_PF;
 609                 if (!pskb_may_pull(skb, padlen + 1))
 610                         goto dump_it;
 611                 skb_pull(skb, padlen);
 612                 flags = *skb->data;
 613         }
 614
 615         skb_reset_network_header(skb);
 616
 617         /*
 618          * Weed out future version DECnet
 619          */
 620         if (flags & DN_RT_F_VER)
 621                 goto dump_it;
 622
 623         cb->rt_flags = flags;
 624
 625         if (decnet_debug_level & 1)
 626                 printk(KERN_DEBUG
 627                         "dn_route_rcv: got 0x%02x from %s [%d %d %d]\n",
 628                         (int)flags, (dev) ? dev->name : "???", len, skb->len,
 629                         padlen);
 630
 631         if (flags & DN_RT_PKT_CNTL) {
 632                 if (unlikely(skb_linearize(skb)))
 633                         goto dump_it;
 634
 635                 switch(flags & DN_RT_CNTL_MSK) {
 636                         case DN_RT_PKT_INIT:
 637                                 dn_dev_init_pkt(skb);
 638                                 break;
 639                         case DN_RT_PKT_VERI:
 640                                 dn_dev_veri_pkt(skb);
 641                                 break;
 642                 }
 643
 644                 if (dn->parms.state != DN_DEV_S_RU)
 645                         goto dump_it;
 646
 647                 switch(flags & DN_RT_CNTL_MSK) {
 648                         case DN_RT_PKT_HELO:
 649                                 return NF_HOOK(PF_DECnet, NF_DN_HELLO, skb, skb->dev, NULL, dn_route_ptp_hello);
 650
 651                         case DN_RT_PKT_L1RT:
 652                         case DN_RT_PKT_L2RT:
 653                                 return NF_HOOK(PF_DECnet, NF_DN_ROUTE, skb, skb->dev, NULL, dn_route_discard);
 654                         case DN_RT_PKT_ERTH:
 655                                 return NF_HOOK(PF_DECnet, NF_DN_HELLO, skb, skb->dev, NULL, dn_neigh_router_hello);
 656
 657                         case DN_RT_PKT_EEDH:
 658                                 return NF_HOOK(PF_DECnet, NF_DN_HELLO, skb, skb->dev, NULL, dn_neigh_endnode_hello);
 659                 }
 660         } else {
 661                 if (dn->parms.state != DN_DEV_S_RU)
 662                         goto dump_it;
 663
 664                 skb_pull(skb, 1); /* Pull flags */
 665
 666                 switch(flags & DN_RT_PKT_MSK) {
 667                         case DN_RT_PKT_LONG:
 668                                 return dn_route_rx_long(skb);
 669                         case DN_RT_PKT_SHORT:
 670                                 return dn_route_rx_short(skb);
 671                 }
 672         }
 673
 674 dump_it:
 675         kfree_skb(skb);
 676 out:
 677         return NET_RX_DROP;
 678 }
 679
 680 static int dn_output(struct sk_buff *skb)
 681 {
 682         struct dst_entry *dst = skb_dst(skb);
 683         struct dn_route *rt = (struct dn_route *)dst;
 684         struct net_device *dev = dst->dev;
 685         struct dn_skb_cb *cb = DN_SKB_CB(skb);
 686         struct neighbour *neigh;
 687
 688         int err = -EINVAL;
 689
 690         if ((neigh = dst->neighbour) == NULL)
 691                 goto error;
 692
 693         skb->dev = dev;
 694
 695         cb->src = rt->rt_saddr;
 696         cb->dst = rt->rt_daddr;
 697
 698         /*
 699          * Always set the Intra-Ethernet bit on all outgoing packets
 700          * originated on this node. Only valid flag from upper layers
 701          * is return-to-sender-requested. Set hop count to 0 too.
 702          */
 703         cb->rt_flags &= ~DN_RT_F_RQR;
 704         cb->rt_flags |= DN_RT_F_IE;
 705         cb->hops = 0;
 706
 707         return NF_HOOK(PF_DECnet, NF_DN_LOCAL_OUT, skb, NULL, dev, neigh->output);
 708
 709 error:
 710         if (net_ratelimit())
 711                 printk(KERN_DEBUG "dn_output: This should not happen\n");
 712
 713         kfree_skb(skb);
 714
 715         return err;
 716 }
 717
 718 static int dn_forward(struct sk_buff *skb)
 719 {
 720         struct dn_skb_cb *cb = DN_SKB_CB(skb);
 721         struct dst_entry *dst = skb_dst(skb);
 722         struct dn_dev *dn_db = dst->dev->dn_ptr;
 723         struct dn_route *rt;
 724         struct neighbour *neigh = dst->neighbour;
 725         int header_len;
 726 #ifdef CONFIG_NETFILTER
 727         struct net_device *dev = skb->dev;
 728 #endif
 729
 730         if (skb->pkt_type != PACKET_HOST)
 731                 goto drop;
 732
 733         /* Ensure that we have enough space for headers */
 734         rt = (struct dn_route *)skb_dst(skb);
 735         header_len = dn_db->use_long ? 21 : 6;
 736         if (skb_cow(skb, LL_RESERVED_SPACE(rt->u.dst.dev)+header_len))
 737                 goto drop;
 738
 739         /*
 740          * Hop count exceeded.
 741          */
 742         if (++cb->hops > 30)
 743                 goto drop;
 744
 745         skb->dev = rt->u.dst.dev;
 746
 747         /*
 748          * If packet goes out same interface it came in on, then set
 749          * the Intra-Ethernet bit. This has no effect for short
 750          * packets, so we don't need to test for them here.
 751          */
 752         cb->rt_flags &= ~DN_RT_F_IE;
 753         if (rt->rt_flags & RTCF_DOREDIRECT)
 754                 cb->rt_flags |= DN_RT_F_IE;
 755
 756         return NF_HOOK(PF_DECnet, NF_DN_FORWARD, skb, dev, skb->dev, neigh->output);
 757
 758 drop:
 759         kfree_skb(skb);
 760         return NET_RX_DROP;
 761 }
 762
 763 /*
 764  * Used to catch bugs. This should never normally get
 765  * called.
 766  */
 767 static int dn_rt_bug(struct sk_buff *skb)
 768 {
 769         if (net_ratelimit()) {
 770                 struct dn_skb_cb *cb = DN_SKB_CB(skb);
 771
 772                 printk(KERN_DEBUG "dn_rt_bug: skb from:%04x to:%04x\n",
 773                                 le16_to_cpu(cb->src), le16_to_cpu(cb->dst));
 774         }
 775
 776         kfree_skb(skb);
 777
 778         return NET_RX_DROP;
 779 }
 780
 781 static int dn_rt_set_next_hop(struct dn_route *rt, struct dn_fib_res *res)
 782 {
 783         struct dn_fib_info *fi = res->fi;
 784         struct net_device *dev = rt->u.dst.dev;
 785         struct neighbour *n;
 786         unsigned mss;
 787
 788         if (fi) {
 789                 if (DN_FIB_RES_GW(*res) &&
 790                     DN_FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
 791                         rt->rt_gateway = DN_FIB_RES_GW(*res);
 792                 memcpy(rt->u.dst.metrics, fi->fib_metrics,
 793                        sizeof(rt->u.dst.metrics));
 794         }
 795         rt->rt_type = res->type;
 796
 797         if (dev != NULL && rt->u.dst.neighbour == NULL) {
 798                 n = __neigh_lookup_errno(&dn_neigh_table, &rt->rt_gateway, dev);
 799                 if (IS_ERR(n))
 800                         return PTR_ERR(n);
 801                 rt->u.dst.neighbour = n;
 802         }
 803
 804         if (dst_metric(&rt->u.dst, RTAX_MTU) == 0 ||
 805             dst_metric(&rt->u.dst, RTAX_MTU) > rt->u.dst.dev->mtu)
 806                 rt->u.dst.metrics[RTAX_MTU-1] = rt->u.dst.dev->mtu;
 807         mss = dn_mss_from_pmtu(dev, dst_mtu(&rt->u.dst));
 808         if (dst_metric(&rt->u.dst, RTAX_ADVMSS) == 0 ||
 809             dst_metric(&rt->u.dst, RTAX_ADVMSS) > mss)
 810                 rt->u.dst.metrics[RTAX_ADVMSS-1] = mss;
 811         return 0;
 812 }
 813
 814 static inline int dn_match_addr(__le16 addr1, __le16 addr2)
 815 {
 816         __u16 tmp = le16_to_cpu(addr1) ^ le16_to_cpu(addr2);
 817         int match = 16;
 818         while(tmp) {
 819                 tmp >>= 1;
 820                 match--;
 821         }
 822         return match;
 823 }
 824
 825 static __le16 dnet_select_source(const struct net_device *dev, __le16 daddr, int scope)
 826 {
 827         __le16 saddr = 0;
 828         struct dn_dev *dn_db = dev->dn_ptr;
 829         struct dn_ifaddr *ifa;
 830         int best_match = 0;
 831         int ret;
 832
 833         read_lock(&dev_base_lock);
 834         for(ifa = dn_db->ifa_list; ifa; ifa = ifa->ifa_next) {
 835                 if (ifa->ifa_scope > scope)
 836                         continue;
 837                 if (!daddr) {
 838                         saddr = ifa->ifa_local;
 839                         break;
 840                 }
 841                 ret = dn_match_addr(daddr, ifa->ifa_local);
 842                 if (ret > best_match)
 843                         saddr = ifa->ifa_local;
 844                 if (best_match == 0)
 845                         saddr = ifa->ifa_local;
 846         }
 847         read_unlock(&dev_base_lock);
 848
 849         return saddr;
 850 }
 851
 852 static inline __le16 __dn_fib_res_prefsrc(struct dn_fib_res *res)
 853 {
 854         return dnet_select_source(DN_FIB_RES_DEV(*res), DN_FIB_RES_GW(*res), res->scope);
 855 }
 856
 857 static inline __le16 dn_fib_rules_map_destination(__le16 daddr, struct dn_fib_res *res)
 858 {
 859         __le16 mask = dnet_make_mask(res->prefixlen);
 860         return (daddr&~mask)|res->fi->fib_nh->nh_gw;
 861 }
 862
 863 static int dn_route_output_slow(struct dst_entry **pprt, const struct flowi *oldflp, int try_hard)
 864 {
 865         struct flowi fl = { .nl_u = { .dn_u =
 866                                       { .daddr = oldflp->fld_dst,
 867                                         .saddr = oldflp->fld_src,
 868                                         .scope = RT_SCOPE_UNIVERSE,
 869                                      } },
 870                             .mark = oldflp->mark,
 871                             .iif = init_net.loopback_dev->ifindex,
 872                             .oif = oldflp->oif };
 873         struct dn_route *rt = NULL;
 874         struct net_device *dev_out = NULL, *dev;
 875         struct neighbour *neigh = NULL;
 876         unsigned hash;
 877         unsigned flags = 0;
 878         struct dn_fib_res res = { .fi = NULL, .type = RTN_UNICAST };
 879         int err;
 880         int free_res = 0;
 881         __le16 gateway = 0;
 882
 883         if (decnet_debug_level & 16)
 884                 printk(KERN_DEBUG
 885                        "dn_route_output_slow: dst=%04x src=%04x mark=%d"
 886                        " iif=%d oif=%d\n", le16_to_cpu(oldflp->fld_dst),
 887                        le16_to_cpu(oldflp->fld_src),
 888                        oldflp->mark, init_net.loopback_dev->ifindex, oldflp->oif);
 889
 890         /* If we have an output interface, verify its a DECnet device */
 891         if (oldflp->oif) {
 892                 dev_out = dev_get_by_index(&init_net, oldflp->oif);
 893                 err = -ENODEV;
 894                 if (dev_out && dev_out->dn_ptr == NULL) {
 895                         dev_put(dev_out);
 896                         dev_out = NULL;
 897                 }
 898                 if (dev_out == NULL)
 899                         goto out;
 900         }
 901
 902         /* If we have a source address, verify that its a local address */
 903         if (oldflp->fld_src) {
 904                 err = -EADDRNOTAVAIL;
 905
 906                 if (dev_out) {
 907                         if (dn_dev_islocal(dev_out, oldflp->fld_src))
 908                                 goto source_ok;
 909                         dev_put(dev_out);
 910                         goto out;
 911                 }
 912                 rcu_read_lock();
 913                 for_each_netdev_rcu(&init_net, dev) {
 914                         if (!dev->dn_ptr)
 915                                 continue;
 916                         if (!dn_dev_islocal(dev, oldflp->fld_src))
 917                                 continue;
 918                         if ((dev->flags & IFF_LOOPBACK) &&
 919                             oldflp->fld_dst &&
 920                             !dn_dev_islocal(dev, oldflp->fld_dst))
 921                                 continue;
 922
 923                         dev_out = dev;
 924                         break;
 925                 }
 926                 rcu_read_unlock();
 927                 if (dev_out == NULL)
 928                         goto out;
 929                 dev_hold(dev_out);
 930 source_ok:
 931                 ;
 932         }
 933
 934         /* No destination? Assume its local */
 935         if (!fl.fld_dst) {
 936                 fl.fld_dst = fl.fld_src;
 937
 938                 err = -EADDRNOTAVAIL;
 939                 if (dev_out)
 940                         dev_put(dev_out);
 941                 dev_out = init_net.loopback_dev;
 942                 dev_hold(dev_out);
 943                 if (!fl.fld_dst) {
 944                         fl.fld_dst =
 945                         fl.fld_src = dnet_select_source(dev_out, 0,
 946                                                        RT_SCOPE_HOST);
 947                         if (!fl.fld_dst)
 948                                 goto out;
 949                 }
 950                 fl.oif = init_net.loopback_dev->ifindex;
 951                 res.type = RTN_LOCAL;
 952                 goto make_route;
 953         }
 954
 955         if (decnet_debug_level & 16)
 956                 printk(KERN_DEBUG
 957                        "dn_route_output_slow: initial checks complete."
 958                        " dst=%o4x src=%04x oif=%d try_hard=%d\n",
 959                        le16_to_cpu(fl.fld_dst), le16_to_cpu(fl.fld_src),
 960                        fl.oif, try_hard);
 961
 962         /*
 963          * N.B. If the kernel is compiled without router support then
 964          * dn_fib_lookup() will evaluate to non-zero so this if () block
 965          * will always be executed.
 966          */
 967         err = -ESRCH;
 968         if (try_hard || (err = dn_fib_lookup(&fl, &res)) != 0) {
 969                 struct dn_dev *dn_db;
 970                 if (err != -ESRCH)
 971                         goto out;
 972                 /*
 973                  * Here the fallback is basically the standard algorithm for
 974                  * routing in endnodes which is described in the DECnet routing
 975                  * docs
 976                  *
 977                  * If we are not trying hard, look in neighbour cache.
 978                  * The result is tested to ensure that if a specific output
 979                  * device/source address was requested, then we honour that
 980                  * here
 981                  */
 982                 if (!try_hard) {
 983                         neigh = neigh_lookup_nodev(&dn_neigh_table, &init_net, &fl.fld_dst);
 984                         if (neigh) {
 985                                 if ((oldflp->oif &&
 986                                     (neigh->dev->ifindex != oldflp->oif)) ||
 987                                     (oldflp->fld_src &&
 988                                     (!dn_dev_islocal(neigh->dev,
 989                                                       oldflp->fld_src)))) {
 990                                         neigh_release(neigh);
 991                                         neigh = NULL;
 992                                 } else {
 993                                         if (dev_out)
 994                                                 dev_put(dev_out);
 995                                         if (dn_dev_islocal(neigh->dev, fl.fld_dst)) {
 996                                                 dev_out = init_net.loopback_dev;
 997                                                 res.type = RTN_LOCAL;
 998                                         } else {
 999                                                 dev_out = neigh->dev;
1000                                         }
1001                                         dev_hold(dev_out);
1002                                         goto select_source;
1003                                 }
1004                         }
1005                 }
1006
1007                 /* Not there? Perhaps its a local address */
1008                 if (dev_out == NULL)
1009                         dev_out = dn_dev_get_default();
1010                 err = -ENODEV;
1011                 if (dev_out == NULL)
1012                         goto out;
1013                 dn_db = dev_out->dn_ptr;
1014                 /* Possible improvement - check all devices for local addr */
1015                 if (dn_dev_islocal(dev_out, fl.fld_dst)) {
1016                         dev_put(dev_out);
1017                         dev_out = init_net.loopback_dev;
1018                         dev_hold(dev_out);
1019                         res.type = RTN_LOCAL;
1020                         goto select_source;
1021                 }
1022                 /* Not local either.... try sending it to the default router */
1023                 neigh = neigh_clone(dn_db->router);
1024                 BUG_ON(neigh && neigh->dev != dev_out);
1025
1026                 /* Ok then, we assume its directly connected and move on */
1027 select_source:
1028                 if (neigh)
1029                         gateway = ((struct dn_neigh *)neigh)->addr;
1030                 if (gateway == 0)
1031                         gateway = fl.fld_dst;
1032                 if (fl.fld_src == 0) {
1033                         fl.fld_src = dnet_select_source(dev_out, gateway,
1034                                                          res.type == RTN_LOCAL ?
1035                                                          RT_SCOPE_HOST :
1036                                                          RT_SCOPE_LINK);
1037                         if (fl.fld_src == 0 && res.type != RTN_LOCAL)
1038                                 goto e_addr;
1039                 }
1040                 fl.oif = dev_out->ifindex;
1041                 goto make_route;
1042         }
1043         free_res = 1;
1044
1045         if (res.type == RTN_NAT)
1046                 goto e_inval;
1047
1048         if (res.type == RTN_LOCAL) {
1049                 if (!fl.fld_src)
1050                         fl.fld_src = fl.fld_dst;
1051                 if (dev_out)
1052                         dev_put(dev_out);
1053                 dev_out = init_net.loopback_dev;
1054                 dev_hold(dev_out);
1055                 fl.oif = dev_out->ifindex;
1056                 if (res.fi)
1057                         dn_fib_info_put(res.fi);
1058                 res.fi = NULL;
1059                 goto make_route;
1060         }
1061
1062         if (res.fi->fib_nhs > 1 && fl.oif == 0)
1063                 dn_fib_select_multipath(&fl, &res);
1064
1065         /*
1066          * We could add some logic to deal with default routes here and
1067          * get rid of some of the special casing above.
1068          */
1069
1070         if (!fl.fld_src)
1071                 fl.fld_src = DN_FIB_RES_PREFSRC(res);
1072
1073         if (dev_out)
1074                 dev_put(dev_out);
1075         dev_out = DN_FIB_RES_DEV(res);
1076         dev_hold(dev_out);
1077         fl.oif = dev_out->ifindex;
1078         gateway = DN_FIB_RES_GW(res);
1079
1080 make_route:
1081         if (dev_out->flags & IFF_LOOPBACK)
1082                 flags |= RTCF_LOCAL;
1083
1084         rt = dst_alloc(&dn_dst_ops);
1085         if (rt == NULL)
1086                 goto e_nobufs;
1087
1088         atomic_set(&rt->u.dst.__refcnt, 1);
1089         rt->u.dst.flags   = DST_HOST;
1090
1091         rt->fl.fld_src    = oldflp->fld_src;
1092         rt->fl.fld_dst    = oldflp->fld_dst;
1093         rt->fl.oif        = oldflp->oif;
1094         rt->fl.iif        = 0;
1095         rt->fl.mark       = oldflp->mark;
1096
1097         rt->rt_saddr      = fl.fld_src;
1098         rt->rt_daddr      = fl.fld_dst;
1099         rt->rt_gateway    = gateway ? gateway : fl.fld_dst;
1100         rt->rt_local_src  = fl.fld_src;
1101
1102         rt->rt_dst_map    = fl.fld_dst;
1103         rt->rt_src_map    = fl.fld_src;
1104
1105         rt->u.dst.dev = dev_out;
1106         dev_hold(dev_out);
1107         rt->u.dst.neighbour = neigh;
1108         neigh = NULL;
1109
1110         rt->u.dst.lastuse = jiffies;
1111         rt->u.dst.output  = dn_output;
1112         rt->u.dst.input   = dn_rt_bug;
1113         rt->rt_flags      = flags;
1114         if (flags & RTCF_LOCAL)
1115                 rt->u.dst.input = dn_nsp_rx;
1116
1117         err = dn_rt_set_next_hop(rt, &res);
1118         if (err)
1119                 goto e_neighbour;
1120
1121         hash = dn_hash(rt->fl.fld_src, rt->fl.fld_dst);
1122         dn_insert_route(rt, hash, (struct dn_route **)pprt);
1123
1124 done:
1125         if (neigh)
1126                 neigh_release(neigh);
1127         if (free_res)
1128                 dn_fib_res_put(&res);
1129         if (dev_out)
1130                 dev_put(dev_out);
1131 out:
1132         return err;
1133
1134 e_addr:
1135         err = -EADDRNOTAVAIL;
1136         goto done;
1137 e_inval:
1138         err = -EINVAL;
1139         goto done;
1140 e_nobufs:
1141         err = -ENOBUFS;
1142         goto done;
1143 e_neighbour:
1144         dst_free(&rt->u.dst);
1145         goto e_nobufs;
1146 }
1147
1148
1149 /*
1150  * N.B. The flags may be moved into the flowi at some future stage.
1151  */
1152 static int __dn_route_output_key(struct dst_entry **pprt, const struct flowi *flp, int flags)
1153 {
1154         unsigned hash = dn_hash(flp->fld_src, flp->fld_dst);
1155         struct dn_route *rt = NULL;
1156
1157         if (!(flags & MSG_TRYHARD)) {
1158                 rcu_read_lock_bh();
1159                 for (rt = rcu_dereference_bh(dn_rt_hash_table[hash].chain); rt;
1160                         rt = rcu_dereference_bh(rt->u.dst.dn_next)) {
1161                         if ((flp->fld_dst == rt->fl.fld_dst) &&
1162                             (flp->fld_src == rt->fl.fld_src) &&
1163                             (flp->mark == rt->fl.mark) &&
1164                             (rt->fl.iif == 0) &&
1165                             (rt->fl.oif == flp->oif)) {
1166                                 dst_use(&rt->u.dst, jiffies);
1167                                 rcu_read_unlock_bh();
1168                                 *pprt = &rt->u.dst;
1169                                 return 0;
1170                         }
1171                 }
1172                 rcu_read_unlock_bh();
1173         }
1174
1175         return dn_route_output_slow(pprt, flp, flags);
1176 }
1177
1178 static int dn_route_output_key(struct dst_entry **pprt, struct flowi *flp, int flags)
1179 {
1180         int err;
1181
1182         err = __dn_route_output_key(pprt, flp, flags);
1183         if (err == 0 && flp->proto) {
1184                 err = xfrm_lookup(&init_net, pprt, flp, NULL, 0);
1185         }
1186         return err;
1187 }
1188
1189 int dn_route_output_sock(struct dst_entry **pprt, struct flowi *fl, struct sock *sk, int flags)
1190 {
1191         int err;
1192
1193         err = __dn_route_output_key(pprt, fl, flags & MSG_TRYHARD);
1194         if (err == 0 && fl->proto) {
1195                 err = xfrm_lookup(&init_net, pprt, fl, sk,
1196                                  (flags & MSG_DONTWAIT) ? 0 : XFRM_LOOKUP_WAIT);
1197         }
1198         return err;
1199 }
1200
1201 static int dn_route_input_slow(struct sk_buff *skb)
1202 {
1203         struct dn_route *rt = NULL;
1204         struct dn_skb_cb *cb = DN_SKB_CB(skb);
1205         struct net_device *in_dev = skb->dev;
1206         struct net_device *out_dev = NULL;
1207         struct dn_dev *dn_db;
1208         struct neighbour *neigh = NULL;
1209         unsigned hash;
1210         int flags = 0;
1211         __le16 gateway = 0;
1212         __le16 local_src = 0;
1213         struct flowi fl = { .nl_u = { .dn_u =
1214                                      { .daddr = cb->dst,
1215                                        .saddr = cb->src,
1216                                        .scope = RT_SCOPE_UNIVERSE,
1217                                     } },
1218                             .mark = skb->mark,
1219                             .iif = skb->dev->ifindex };
1220         struct dn_fib_res res = { .fi = NULL, .type = RTN_UNREACHABLE };
1221         int err = -EINVAL;
1222         int free_res = 0;
1223
1224         dev_hold(in_dev);
1225
1226         if ((dn_db = in_dev->dn_ptr) == NULL)
1227                 goto out;
1228
1229         /* Zero source addresses are not allowed */
1230         if (fl.fld_src == 0)
1231                 goto out;
1232
1233         /*
1234          * In this case we've just received a packet from a source
1235          * outside ourselves pretending to come from us. We don't
1236          * allow it any further to prevent routing loops, spoofing and
1237          * other nasties. Loopback packets already have the dst attached
1238          * so this only affects packets which have originated elsewhere.
1239          */
1240         err  = -ENOTUNIQ;
1241         if (dn_dev_islocal(in_dev, cb->src))
1242                 goto out;
1243
1244         err = dn_fib_lookup(&fl, &res);
1245         if (err) {
1246                 if (err != -ESRCH)
1247                         goto out;
1248                 /*
1249                  * Is the destination us ?
1250                  */
1251                 if (!dn_dev_islocal(in_dev, cb->dst))
1252                         goto e_inval;
1253
1254                 res.type = RTN_LOCAL;
1255         } else {
1256                 __le16 src_map = fl.fld_src;
1257                 free_res = 1;
1258
1259                 out_dev = DN_FIB_RES_DEV(res);
1260                 if (out_dev == NULL) {
1261                         if (net_ratelimit())
1262                                 printk(KERN_CRIT "Bug in dn_route_input_slow() "
1263                                                  "No output device\n");
1264                         goto e_inval;
1265                 }
1266                 dev_hold(out_dev);
1267
1268                 if (res.r)
1269                         src_map = fl.fld_src; /* no NAT support for now */
1270
1271                 gateway = DN_FIB_RES_GW(res);
1272                 if (res.type == RTN_NAT) {
1273                         fl.fld_dst = dn_fib_rules_map_destination(fl.fld_dst, &res);
1274                         dn_fib_res_put(&res);
1275                         free_res = 0;
1276                         if (dn_fib_lookup(&fl, &res))
1277                                 goto e_inval;
1278                         free_res = 1;
1279                         if (res.type != RTN_UNICAST)
1280                                 goto e_inval;
1281                         flags |= RTCF_DNAT;
1282                         gateway = fl.fld_dst;
1283                 }
1284                 fl.fld_src = src_map;
1285         }
1286
1287         switch(res.type) {
1288         case RTN_UNICAST:
1289                 /*
1290                  * Forwarding check here, we only check for forwarding
1291                  * being turned off, if you want to only forward intra
1292                  * area, its up to you to set the routing tables up
1293                  * correctly.
1294                  */
1295                 if (dn_db->parms.forwarding == 0)
1296                         goto e_inval;
1297
1298                 if (res.fi->fib_nhs > 1 && fl.oif == 0)
1299                         dn_fib_select_multipath(&fl, &res);
1300
1301                 /*
1302                  * Check for out_dev == in_dev. We use the RTCF_DOREDIRECT
1303                  * flag as a hint to set the intra-ethernet bit when
1304                  * forwarding. If we've got NAT in operation, we don't do
1305                  * this optimisation.
1306                  */
1307                 if (out_dev == in_dev && !(flags & RTCF_NAT))
1308                         flags |= RTCF_DOREDIRECT;
1309
1310                 local_src = DN_FIB_RES_PREFSRC(res);
1311
1312         case RTN_BLACKHOLE:
1313         case RTN_UNREACHABLE:
1314                 break;
1315         case RTN_LOCAL:
1316                 flags |= RTCF_LOCAL;
1317                 fl.fld_src = cb->dst;
1318                 fl.fld_dst = cb->src;
1319
1320                 /* Routing tables gave us a gateway */
1321                 if (gateway)
1322                         goto make_route;
1323
1324                 /* Packet was intra-ethernet, so we know its on-link */
1325                 if (cb->rt_flags & DN_RT_F_IE) {
1326                         gateway = cb->src;
1327                         flags |= RTCF_DIRECTSRC;
1328                         goto make_route;
1329                 }
1330
1331                 /* Use the default router if there is one */
1332                 neigh = neigh_clone(dn_db->router);
1333                 if (neigh) {
1334                         gateway = ((struct dn_neigh *)neigh)->addr;
1335                         goto make_route;
1336                 }
1337
1338                 /* Close eyes and pray */
1339                 gateway = cb->src;
1340                 flags |= RTCF_DIRECTSRC;
1341                 goto make_route;
1342         default:
1343                 goto e_inval;
1344         }
1345
1346 make_route:
1347         rt = dst_alloc(&dn_dst_ops);
1348         if (rt == NULL)
1349                 goto e_nobufs;
1350
1351         rt->rt_saddr      = fl.fld_src;
1352         rt->rt_daddr      = fl.fld_dst;
1353         rt->rt_gateway    = fl.fld_dst;
1354         if (gateway)
1355                 rt->rt_gateway = gateway;
1356         rt->rt_local_src  = local_src ? local_src : rt->rt_saddr;
1357
1358         rt->rt_dst_map    = fl.fld_dst;
1359         rt->rt_src_map    = fl.fld_src;
1360
1361         rt->fl.fld_src    = cb->src;
1362         rt->fl.fld_dst    = cb->dst;
1363         rt->fl.oif        = 0;
1364         rt->fl.iif        = in_dev->ifindex;
1365         rt->fl.mark       = fl.mark;
1366
1367         rt->u.dst.flags = DST_HOST;
1368         rt->u.dst.neighbour = neigh;
1369         rt->u.dst.dev = out_dev;
1370         rt->u.dst.lastuse = jiffies;
1371         rt->u.dst.output = dn_rt_bug;
1372         switch(res.type) {
1373                 case RTN_UNICAST:
1374                         rt->u.dst.input = dn_forward;
1375                         break;
1376                 case RTN_LOCAL:
1377                         rt->u.dst.output = dn_output;
1378                         rt->u.dst.input = dn_nsp_rx;
1379                         rt->u.dst.dev = in_dev;
1380                         flags |= RTCF_LOCAL;
1381                         break;
1382                 default:
1383                 case RTN_UNREACHABLE:
1384                 case RTN_BLACKHOLE:
1385                         rt->u.dst.input = dst_discard;
1386         }
1387         rt->rt_flags = flags;
1388         if (rt->u.dst.dev)
1389                 dev_hold(rt->u.dst.dev);
1390
1391         err = dn_rt_set_next_hop(rt, &res);
1392         if (err)
1393                 goto e_neighbour;
1394
1395         hash = dn_hash(rt->fl.fld_src, rt->fl.fld_dst);
1396         dn_insert_route(rt, hash, &rt);
1397         skb_dst_set(skb, &rt->u.dst);
1398
1399 done:
1400         if (neigh)
1401                 neigh_release(neigh);
1402         if (free_res)
1403                 dn_fib_res_put(&res);
1404         dev_put(in_dev);
1405         if (out_dev)
1406                 dev_put(out_dev);
1407 out:
1408         return err;
1409
1410 e_inval:
1411         err = -EINVAL;
1412         goto done;
1413
1414 e_nobufs:
1415         err = -ENOBUFS;
1416         goto done;
1417
1418 e_neighbour:
1419         dst_free(&rt->u.dst);
1420         goto done;
1421 }
1422
1423 static int dn_route_input(struct sk_buff *skb)
1424 {
1425         struct dn_route *rt;
1426         struct dn_skb_cb *cb = DN_SKB_CB(skb);
1427         unsigned hash = dn_hash(cb->src, cb->dst);
1428
1429         if (skb_dst(skb))
1430                 return 0;
1431
1432         rcu_read_lock();
1433         for(rt = rcu_dereference(dn_rt_hash_table[hash].chain); rt != NULL;
1434             rt = rcu_dereference(rt->u.dst.dn_next)) {
1435                 if ((rt->fl.fld_src == cb->src) &&
1436                     (rt->fl.fld_dst == cb->dst) &&
1437                     (rt->fl.oif == 0) &&
1438                     (rt->fl.mark == skb->mark) &&
1439                     (rt->fl.iif == cb->iif)) {
1440                         dst_use(&rt->u.dst, jiffies);
1441                         rcu_read_unlock();
1442                         skb_dst_set(skb, (struct dst_entry *)rt);
1443                         return 0;
1444                 }
1445         }
1446         rcu_read_unlock();
1447
1448         return dn_route_input_slow(skb);
1449 }
1450
1451 static int dn_rt_fill_info(struct sk_buff *skb, u32 pid, u32 seq,
1452                            int event, int nowait, unsigned int flags)
1453 {
1454         struct dn_route *rt = (struct dn_route *)skb_dst(skb);
1455         struct rtmsg *r;
1456         struct nlmsghdr *nlh;
1457         unsigned char *b = skb_tail_pointer(skb);
1458         long expires;
1459
1460         nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*r), flags);
1461         r = NLMSG_DATA(nlh);
1462         r->rtm_family = AF_DECnet;
1463         r->rtm_dst_len = 16;
1464         r->rtm_src_len = 0;
1465         r->rtm_tos = 0;
1466         r->rtm_table = RT_TABLE_MAIN;
1467         RTA_PUT_U32(skb, RTA_TABLE, RT_TABLE_MAIN);
1468         r->rtm_type = rt->rt_type;
1469         r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
1470         r->rtm_scope = RT_SCOPE_UNIVERSE;
1471         r->rtm_protocol = RTPROT_UNSPEC;
1472         if (rt->rt_flags & RTCF_NOTIFY)
1473                 r->rtm_flags |= RTM_F_NOTIFY;
1474         RTA_PUT(skb, RTA_DST, 2, &rt->rt_daddr);
1475         if (rt->fl.fld_src) {
1476                 r->rtm_src_len = 16;
1477                 RTA_PUT(skb, RTA_SRC, 2, &rt->fl.fld_src);
1478         }
1479         if (rt->u.dst.dev)
1480                 RTA_PUT(skb, RTA_OIF, sizeof(int), &rt->u.dst.dev->ifindex);
1481         /*
1482          * Note to self - change this if input routes reverse direction when
1483          * they deal only with inputs and not with replies like they do
1484          * currently.
1485          */
1486         RTA_PUT(skb, RTA_PREFSRC, 2, &rt->rt_local_src);
1487         if (rt->rt_daddr != rt->rt_gateway)
1488                 RTA_PUT(skb, RTA_GATEWAY, 2, &rt->rt_gateway);
1489         if (rtnetlink_put_metrics(skb, rt->u.dst.metrics) < 0)
1490                 goto rtattr_failure;
1491         expires = rt->u.dst.expires ? rt->u.dst.expires - jiffies : 0;
1492         if (rtnl_put_cacheinfo(skb, &rt->u.dst, 0, 0, 0, expires,
1493                                rt->u.dst.error) < 0)
1494                 goto rtattr_failure;
1495         if (rt->fl.iif)
1496                 RTA_PUT(skb, RTA_IIF, sizeof(int), &rt->fl.iif);
1497
1498         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1499         return skb->len;
1500
1501 nlmsg_failure:
1502 rtattr_failure:
1503         nlmsg_trim(skb, b);
1504         return -1;
1505 }
1506
1507 /*
1508  * This is called by both endnodes and routers now.
1509  */
1510 static int dn_cache_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh, void *arg)
1511 {
1512         struct net *net = sock_net(in_skb->sk);
1513         struct rtattr **rta = arg;
1514         struct rtmsg *rtm = NLMSG_DATA(nlh);
1515         struct dn_route *rt = NULL;
1516         struct dn_skb_cb *cb;
1517         int err;
1518         struct sk_buff *skb;
1519         struct flowi fl;
1520
1521         if (!net_eq(net, &init_net))
1522                 return -EINVAL;
1523
1524         memset(&fl, 0, sizeof(fl));
1525         fl.proto = DNPROTO_NSP;
1526
1527         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1528         if (skb == NULL)
1529                 return -ENOBUFS;
1530         skb_reset_mac_header(skb);
1531         cb = DN_SKB_CB(skb);
1532
1533         if (rta[RTA_SRC-1])
1534                 memcpy(&fl.fld_src, RTA_DATA(rta[RTA_SRC-1]), 2);
1535         if (rta[RTA_DST-1])
1536                 memcpy(&fl.fld_dst, RTA_DATA(rta[RTA_DST-1]), 2);
1537         if (rta[RTA_IIF-1])
1538                 memcpy(&fl.iif, RTA_DATA(rta[RTA_IIF-1]), sizeof(int));
1539
1540         if (fl.iif) {
1541                 struct net_device *dev;
1542                 if ((dev = dev_get_by_index(&init_net, fl.iif)) == NULL) {
1543                         kfree_skb(skb);
1544                         return -ENODEV;
1545                 }
1546                 if (!dev->dn_ptr) {
1547                         dev_put(dev);
1548                         kfree_skb(skb);
1549                         return -ENODEV;
1550                 }
1551                 skb->protocol = htons(ETH_P_DNA_RT);
1552                 skb->dev = dev;
1553                 cb->src = fl.fld_src;
1554                 cb->dst = fl.fld_dst;
1555                 local_bh_disable();
1556                 err = dn_route_input(skb);
1557                 local_bh_enable();
1558                 memset(cb, 0, sizeof(struct dn_skb_cb));
1559                 rt = (struct dn_route *)skb_dst(skb);
1560                 if (!err && -rt->u.dst.error)
1561                         err = rt->u.dst.error;
1562         } else {
1563                 int oif = 0;
1564                 if (rta[RTA_OIF - 1])
1565                         memcpy(&oif, RTA_DATA(rta[RTA_OIF - 1]), sizeof(int));
1566                 fl.oif = oif;
1567                 err = dn_route_output_key((struct dst_entry **)&rt, &fl, 0);
1568         }
1569
1570         if (skb->dev)
1571                 dev_put(skb->dev);
1572         skb->dev = NULL;
1573         if (err)
1574                 goto out_free;
1575         skb_dst_set(skb, &rt->u.dst);
1576         if (rtm->rtm_flags & RTM_F_NOTIFY)
1577                 rt->rt_flags |= RTCF_NOTIFY;
1578
1579         err = dn_rt_fill_info(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq, RTM_NEWROUTE, 0, 0);
1580
1581         if (err == 0)
1582                 goto out_free;
1583         if (err < 0) {
1584                 err = -EMSGSIZE;
1585                 goto out_free;
1586         }
1587
1588         return rtnl_unicast(skb, &init_net, NETLINK_CB(in_skb).pid);
1589
1590 out_free:
1591         kfree_skb(skb);
1592         return err;
1593 }
1594
1595 /*
1596  * For routers, this is called from dn_fib_dump, but for endnodes its
1597  * called directly from the rtnetlink dispatch table.
1598  */
1599 int dn_cache_dump(struct sk_buff *skb, struct netlink_callback *cb)
1600 {
1601         struct net *net = sock_net(skb->sk);
1602         struct dn_route *rt;
1603         int h, s_h;
1604         int idx, s_idx;
1605
1606         if (!net_eq(net, &init_net))
1607                 return 0;
1608
1609         if (NLMSG_PAYLOAD(cb->nlh, 0) < sizeof(struct rtmsg))
1610                 return -EINVAL;
1611         if (!(((struct rtmsg *)NLMSG_DATA(cb->nlh))->rtm_flags&RTM_F_CLONED))
1612                 return 0;
1613
1614         s_h = cb->args[0];
1615         s_idx = idx = cb->args[1];
1616         for(h = 0; h <= dn_rt_hash_mask; h++) {
1617                 if (h < s_h)
1618                         continue;
1619                 if (h > s_h)
1620                         s_idx = 0;
1621                 rcu_read_lock_bh();
1622                 for(rt = rcu_dereference_bh(dn_rt_hash_table[h].chain), idx = 0;
1623                         rt;
1624                         rt = rcu_dereference_bh(rt->u.dst.dn_next), idx++) {
1625                         if (idx < s_idx)
1626                                 continue;
1627                         skb_dst_set(skb, dst_clone(&rt->u.dst));
1628                         if (dn_rt_fill_info(skb, NETLINK_CB(cb->skb).pid,
1629                                         cb->nlh->nlmsg_seq, RTM_NEWROUTE,
1630                                         1, NLM_F_MULTI) <= 0) {
1631                                 skb_dst_drop(skb);
1632                                 rcu_read_unlock_bh();
1633                                 goto done;
1634                         }
1635                         skb_dst_drop(skb);
1636                 }
1637                 rcu_read_unlock_bh();
1638         }
1639
1640 done:
1641         cb->args[0] = h;
1642         cb->args[1] = idx;
1643         return skb->len;
1644 }
1645
1646 #ifdef CONFIG_PROC_FS
1647 struct dn_rt_cache_iter_state {
1648         int bucket;
1649 };
1650
1651 static struct dn_route *dn_rt_cache_get_first(struct seq_file *seq)
1652 {
1653         struct dn_route *rt = NULL;
1654         struct dn_rt_cache_iter_state *s = seq->private;
1655
1656         for(s->bucket = dn_rt_hash_mask; s->bucket >= 0; --s->bucket) {
1657                 rcu_read_lock_bh();
1658                 rt = rcu_dereference_bh(dn_rt_hash_table[s->bucket].chain);
1659                 if (rt)
1660                         break;
1661                 rcu_read_unlock_bh();
1662         }
1663         return rt;
1664 }
1665
1666 static struct dn_route *dn_rt_cache_get_next(struct seq_file *seq, struct dn_route *rt)
1667 {
1668         struct dn_rt_cache_iter_state *s = seq->private;
1669
1670         rt = rt->u.dst.dn_next;
1671         while(!rt) {
1672                 rcu_read_unlock_bh();
1673                 if (--s->bucket < 0)
1674                         break;
1675                 rcu_read_lock_bh();
1676                 rt = dn_rt_hash_table[s->bucket].chain;
1677         }
1678         return rcu_dereference_bh(rt);
1679 }
1680
1681 static void *dn_rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
1682 {
1683         struct dn_route *rt = dn_rt_cache_get_first(seq);
1684
1685         if (rt) {
1686                 while(*pos && (rt = dn_rt_cache_get_next(seq, rt)))
1687                         --*pos;
1688         }
1689         return *pos ? NULL : rt;
1690 }
1691
1692 static void *dn_rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1693 {
1694         struct dn_route *rt = dn_rt_cache_get_next(seq, v);
1695         ++*pos;
1696         return rt;
1697 }
1698
1699 static void dn_rt_cache_seq_stop(struct seq_file *seq, void *v)
1700 {
1701         if (v)
1702                 rcu_read_unlock_bh();
1703 }
1704
1705 static int dn_rt_cache_seq_show(struct seq_file *seq, void *v)
1706 {
1707         struct dn_route *rt = v;
1708         char buf1[DN_ASCBUF_LEN], buf2[DN_ASCBUF_LEN];
1709
1710         seq_printf(seq, "%-8s %-7s %-7s %04d %04d %04d\n",
1711                         rt->u.dst.dev ? rt->u.dst.dev->name : "*",
1712                         dn_addr2asc(le16_to_cpu(rt->rt_daddr), buf1),
1713                         dn_addr2asc(le16_to_cpu(rt->rt_saddr), buf2),
1714                         atomic_read(&rt->u.dst.__refcnt),
1715                         rt->u.dst.__use,
1716                         (int) dst_metric(&rt->u.dst, RTAX_RTT));
1717         return 0;
1718 }
1719
1720 static const struct seq_operations dn_rt_cache_seq_ops = {
1721         .start  = dn_rt_cache_seq_start,
1722         .next   = dn_rt_cache_seq_next,
1723         .stop   = dn_rt_cache_seq_stop,
1724         .show   = dn_rt_cache_seq_show,
1725 };
1726
1727 static int dn_rt_cache_seq_open(struct inode *inode, struct file *file)
1728 {
1729         return seq_open_private(file, &dn_rt_cache_seq_ops,
1730                         sizeof(struct dn_rt_cache_iter_state));
1731 }
1732
1733 static const struct file_operations dn_rt_cache_seq_fops = {
1734         .owner   = THIS_MODULE,
1735         .open    = dn_rt_cache_seq_open,
1736         .read    = seq_read,
1737         .llseek  = seq_lseek,
1738         .release = seq_release_private,
1739 };
1740
1741 #endif /* CONFIG_PROC_FS */
1742
1743 void __init dn_route_init(void)
1744 {
1745         int i, goal, order;
1746
1747         dn_dst_ops.kmem_cachep =
1748                 kmem_cache_create("dn_dst_cache", sizeof(struct dn_route), 0,
1749                                   SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1750         setup_timer(&dn_route_timer, dn_dst_check_expire, 0);
1751         dn_route_timer.expires = jiffies + decnet_dst_gc_interval * HZ;
1752         add_timer(&dn_route_timer);
1753
1754         goal = totalram_pages >> (26 - PAGE_SHIFT);
1755
1756         for(order = 0; (1UL << order) < goal; order++)
1757                 /* NOTHING */;
1758
1759         /*
1760          * Only want 1024 entries max, since the table is very, very unlikely
1761          * to be larger than that.
1762          */
1763         while(order && ((((1UL << order) * PAGE_SIZE) /
1764                                 sizeof(struct dn_rt_hash_bucket)) >= 2048))
1765                 order--;
1766
1767         do {
1768                 dn_rt_hash_mask = (1UL << order) * PAGE_SIZE /
1769                         sizeof(struct dn_rt_hash_bucket);
1770                 while(dn_rt_hash_mask & (dn_rt_hash_mask - 1))
1771                         dn_rt_hash_mask--;
1772                 dn_rt_hash_table = (struct dn_rt_hash_bucket *)
1773                         __get_free_pages(GFP_ATOMIC, order);
1774         } while (dn_rt_hash_table == NULL && --order > 0);
1775
1776         if (!dn_rt_hash_table)
1777                 panic("Failed to allocate DECnet route cache hash table\n");
1778
1779         printk(KERN_INFO
1780                 "DECnet: Routing cache hash table of %u buckets, %ldKbytes\n",
1781                 dn_rt_hash_mask,
1782                 (long)(dn_rt_hash_mask*sizeof(struct dn_rt_hash_bucket))/1024);
1783
1784         dn_rt_hash_mask--;
1785         for(i = 0; i <= dn_rt_hash_mask; i++) {
1786                 spin_lock_init(&dn_rt_hash_table[i].lock);
1787                 dn_rt_hash_table[i].chain = NULL;
1788         }
1789
1790         dn_dst_ops.gc_thresh = (dn_rt_hash_mask + 1);
1791
1792         proc_net_fops_create(&init_net, "decnet_cache", S_IRUGO, &dn_rt_cache_seq_fops);
1793
1794 #ifdef CONFIG_DECNET_ROUTER
1795         rtnl_register(PF_DECnet, RTM_GETROUTE, dn_cache_getroute, dn_fib_dump);
1796 #else
1797         rtnl_register(PF_DECnet, RTM_GETROUTE, dn_cache_getroute,
1798                       dn_cache_dump);
1799 #endif
1800 }
1801
1802 void __exit dn_route_cleanup(void)
1803 {
1804         del_timer(&dn_route_timer);
1805         dn_run_flush(0);
1806
1807         proc_net_remove(&init_net, "decnet_cache");
1808 }
1809