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/config.h>
  59 #include <linux/errno.h>
  60 #include <linux/types.h>
  61 #include <linux/socket.h>
  62 #include <linux/in.h>
  63 #include <linux/kernel.h>
  64 #include <linux/sockios.h>
  65 #include <linux/net.h>
  66 #include <linux/netdevice.h>
  67 #include <linux/inet.h>
  68 #include <linux/route.h>
  69 #include <linux/in_route.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/neighbour.h>
  82 #include <net/dst.h>
  83 #include <net/flow.h>
  84 #include <net/dn.h>
  85 #include <net/dn_dev.h>
  86 #include <net/dn_nsp.h>
  87 #include <net/dn_route.h>
  88 #include <net/dn_neigh.h>
  89 #include <net/dn_fib.h>
  90
  91 struct dn_rt_hash_bucket
  92 {
  93         struct dn_route *chain;
  94         spinlock_t lock;
  95 } __attribute__((__aligned__(8)));
  96
  97 extern struct neigh_table dn_neigh_table;
  98
  99
 100 static unsigned char dn_hiord_addr[6] = {0xAA,0x00,0x04,0x00,0x00,0x00};
 101
 102 static const int dn_rt_min_delay = 2 * HZ;
 103 static const int dn_rt_max_delay = 10 * HZ;
 104 static const int dn_rt_mtu_expires = 10 * 60 * HZ;
 105
 106 static unsigned long dn_rt_deadline;
 107
 108 static int dn_dst_gc(void);
 109 static struct dst_entry *dn_dst_check(struct dst_entry *, __u32);
 110 static struct dst_entry *dn_dst_negative_advice(struct dst_entry *);
 111 static void dn_dst_link_failure(struct sk_buff *);
 112 static void dn_dst_update_pmtu(struct dst_entry *dst, u32 mtu);
 113 static int dn_route_input(struct sk_buff *);
 114 static void dn_run_flush(unsigned long dummy);
 115
 116 static struct dn_rt_hash_bucket *dn_rt_hash_table;
 117 static unsigned dn_rt_hash_mask;
 118
 119 static struct timer_list dn_route_timer;
 120 static DEFINE_TIMER(dn_rt_flush_timer, dn_run_flush, 0, 0);
 121 int decnet_dst_gc_interval = 2;
 122
 123 static struct dst_ops dn_dst_ops = {
 124         .family =               PF_DECnet,
 125         .protocol =             __constant_htons(ETH_P_DNA_RT),
 126         .gc_thresh =            128,
 127         .gc =                   dn_dst_gc,
 128         .check =                dn_dst_check,
 129         .negative_advice =      dn_dst_negative_advice,
 130         .link_failure =         dn_dst_link_failure,
 131         .update_pmtu =          dn_dst_update_pmtu,
 132         .entry_size =           sizeof(struct dn_route),
 133         .entries =              ATOMIC_INIT(0),
 134 };
 135
 136 static __inline__ unsigned dn_hash(__le16 src, __le16 dst)
 137 {
 138         __u16 tmp = (__u16 __force)(src ^ dst);
 139         tmp ^= (tmp >> 3);
 140         tmp ^= (tmp >> 5);
 141         tmp ^= (tmp >> 10);
 142         return dn_rt_hash_mask & (unsigned)tmp;
 143 }
 144
 145 static inline void dnrt_free(struct dn_route *rt)
 146 {
 147         call_rcu_bh(&rt->u.dst.rcu_head, dst_rcu_free);
 148 }
 149
 150 static inline void dnrt_drop(struct dn_route *rt)
 151 {
 152         dst_release(&rt->u.dst);
 153         call_rcu_bh(&rt->u.dst.rcu_head, dst_rcu_free);
 154 }
 155
 156 static void dn_dst_check_expire(unsigned long dummy)
 157 {
 158         int i;
 159         struct dn_route *rt, **rtp;
 160         unsigned long now = jiffies;
 161         unsigned long expire = 120 * HZ;
 162
 163         for(i = 0; i <= dn_rt_hash_mask; i++) {
 164                 rtp = &dn_rt_hash_table[i].chain;
 165
 166                 spin_lock(&dn_rt_hash_table[i].lock);
 167                 while((rt=*rtp) != NULL) {
 168                         if (atomic_read(&rt->u.dst.__refcnt) ||
 169                                         (now - rt->u.dst.lastuse) < expire) {
 170                                 rtp = &rt->u.rt_next;
 171                                 continue;
 172                         }
 173                         *rtp = rt->u.rt_next;
 174                         rt->u.rt_next = NULL;
 175                         dnrt_free(rt);
 176                 }
 177                 spin_unlock(&dn_rt_hash_table[i].lock);
 178
 179                 if ((jiffies - now) > 0)
 180                         break;
 181         }
 182
 183         mod_timer(&dn_route_timer, now + decnet_dst_gc_interval * HZ);
 184 }
 185
 186 static int dn_dst_gc(void)
 187 {
 188         struct dn_route *rt, **rtp;
 189         int i;
 190         unsigned long now = jiffies;
 191         unsigned long expire = 10 * HZ;
 192
 193         for(i = 0; i <= dn_rt_hash_mask; i++) {
 194
 195                 spin_lock_bh(&dn_rt_hash_table[i].lock);
 196                 rtp = &dn_rt_hash_table[i].chain;
 197
 198                 while((rt=*rtp) != NULL) {
 199                         if (atomic_read(&rt->u.dst.__refcnt) ||
 200                                         (now - rt->u.dst.lastuse) < expire) {
 201                                 rtp = &rt->u.rt_next;
 202                                 continue;
 203                         }
 204                         *rtp = rt->u.rt_next;
 205                         rt->u.rt_next = NULL;
 206                         dnrt_drop(rt);
 207                         break;
 208                 }
 209                 spin_unlock_bh(&dn_rt_hash_table[i].lock);
 210         }
 211
 212         return 0;
 213 }
 214
 215 /*
 216  * The decnet standards don't impose a particular minimum mtu, what they
 217  * do insist on is that the routing layer accepts a datagram of at least
 218  * 230 bytes long. Here we have to subtract the routing header length from
 219  * 230 to get the minimum acceptable mtu. If there is no neighbour, then we
 220  * assume the worst and use a long header size.
 221  *
 222  * We update both the mtu and the advertised mss (i.e. the segment size we
 223  * advertise to the other end).
 224  */
 225 static void dn_dst_update_pmtu(struct dst_entry *dst, u32 mtu)
 226 {
 227         u32 min_mtu = 230;
 228         struct dn_dev *dn = dst->neighbour ?
 229                             (struct dn_dev *)dst->neighbour->dev->dn_ptr : NULL;
 230
 231         if (dn && dn->use_long == 0)
 232                 min_mtu -= 6;
 233         else
 234                 min_mtu -= 21;
 235
 236         if (dst->metrics[RTAX_MTU-1] > mtu && mtu >= min_mtu) {
 237                 if (!(dst_metric_locked(dst, RTAX_MTU))) {
 238                         dst->metrics[RTAX_MTU-1] = mtu;
 239                         dst_set_expires(dst, dn_rt_mtu_expires);
 240                 }
 241                 if (!(dst_metric_locked(dst, RTAX_ADVMSS))) {
 242                         u32 mss = mtu - DN_MAX_NSP_DATA_HEADER;
 243                         if (dst->metrics[RTAX_ADVMSS-1] > mss)
 244                                 dst->metrics[RTAX_ADVMSS-1] = mss;
 245                 }
 246         }
 247 }
 248
 249 /*
 250  * When a route has been marked obsolete. (e.g. routing cache flush)
 251  */
 252 static struct dst_entry *dn_dst_check(struct dst_entry *dst, __u32 cookie)
 253 {
 254         return NULL;
 255 }
 256
 257 static struct dst_entry *dn_dst_negative_advice(struct dst_entry *dst)
 258 {
 259         dst_release(dst);
 260         return NULL;
 261 }
 262
 263 static void dn_dst_link_failure(struct sk_buff *skb)
 264 {
 265         return;
 266 }
 267
 268 static inline int compare_keys(struct flowi *fl1, struct flowi *fl2)
 269 {
 270         return memcmp(&fl1->nl_u.dn_u, &fl2->nl_u.dn_u, sizeof(fl1->nl_u.dn_u)) == 0 &&
 271                 fl1->oif == fl2->oif &&
 272                 fl1->iif == fl2->iif;
 273 }
 274
 275 static int dn_insert_route(struct dn_route *rt, unsigned hash, struct dn_route **rp)
 276 {
 277         struct dn_route *rth, **rthp;
 278         unsigned long now = jiffies;
 279
 280         rthp = &dn_rt_hash_table[hash].chain;
 281
 282         spin_lock_bh(&dn_rt_hash_table[hash].lock);
 283         while((rth = *rthp) != NULL) {
 284                 if (compare_keys(&rth->fl, &rt->fl)) {
 285                         /* Put it first */
 286                         *rthp = rth->u.rt_next;
 287                         rcu_assign_pointer(rth->u.rt_next,
 288                                            dn_rt_hash_table[hash].chain);
 289                         rcu_assign_pointer(dn_rt_hash_table[hash].chain, rth);
 290
 291                         rth->u.dst.__use++;
 292                         dst_hold(&rth->u.dst);
 293                         rth->u.dst.lastuse = now;
 294                         spin_unlock_bh(&dn_rt_hash_table[hash].lock);
 295
 296                         dnrt_drop(rt);
 297                         *rp = rth;
 298                         return 0;
 299                 }
 300                 rthp = &rth->u.rt_next;
 301         }
 302
 303         rcu_assign_pointer(rt->u.rt_next, dn_rt_hash_table[hash].chain);
 304         rcu_assign_pointer(dn_rt_hash_table[hash].chain, rt);
 305
 306         dst_hold(&rt->u.dst);
 307         rt->u.dst.__use++;
 308         rt->u.dst.lastuse = now;
 309         spin_unlock_bh(&dn_rt_hash_table[hash].lock);
 310         *rp = rt;
 311         return 0;
 312 }
 313
 314 void dn_run_flush(unsigned long dummy)
 315 {
 316         int i;
 317         struct dn_route *rt, *next;
 318
 319         for(i = 0; i < dn_rt_hash_mask; i++) {
 320                 spin_lock_bh(&dn_rt_hash_table[i].lock);
 321
 322                 if ((rt = xchg(&dn_rt_hash_table[i].chain, NULL)) == NULL)
 323                         goto nothing_to_declare;
 324
 325                 for(; rt; rt=next) {
 326                         next = rt->u.rt_next;
 327                         rt->u.rt_next = NULL;
 328                         dst_free((struct dst_entry *)rt);
 329                 }
 330
 331 nothing_to_declare:
 332                 spin_unlock_bh(&dn_rt_hash_table[i].lock);
 333         }
 334 }
 335
 336 static DEFINE_SPINLOCK(dn_rt_flush_lock);
 337
 338 void dn_rt_cache_flush(int delay)
 339 {
 340         unsigned long now = jiffies;
 341         int user_mode = !in_interrupt();
 342
 343         if (delay < 0)
 344                 delay = dn_rt_min_delay;
 345
 346         spin_lock_bh(&dn_rt_flush_lock);
 347
 348         if (del_timer(&dn_rt_flush_timer) && delay > 0 && dn_rt_deadline) {
 349                 long tmo = (long)(dn_rt_deadline - now);
 350
 351                 if (user_mode && tmo < dn_rt_max_delay - dn_rt_min_delay)
 352                         tmo = 0;
 353
 354                 if (delay > tmo)
 355                         delay = tmo;
 356         }
 357
 358         if (delay <= 0) {
 359                 spin_unlock_bh(&dn_rt_flush_lock);
 360                 dn_run_flush(0);
 361                 return;
 362         }
 363
 364         if (dn_rt_deadline == 0)
 365                 dn_rt_deadline = now + dn_rt_max_delay;
 366
 367         dn_rt_flush_timer.expires = now + delay;
 368         add_timer(&dn_rt_flush_timer);
 369         spin_unlock_bh(&dn_rt_flush_lock);
 370 }
 371
 372 /**
 373  * dn_return_short - Return a short packet to its sender
 374  * @skb: The packet to return
 375  *
 376  */
 377 static int dn_return_short(struct sk_buff *skb)
 378 {
 379         struct dn_skb_cb *cb;
 380         unsigned char *ptr;
 381         __le16 *src;
 382         __le16 *dst;
 383         __le16 tmp;
 384
 385         /* Add back headers */
 386         skb_push(skb, skb->data - skb->nh.raw);
 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 source and destination */
 403         tmp  = *src;
 404         *src = *dst;
 405         *dst = tmp;
 406
 407         skb->pkt_type = PACKET_OUTGOING;
 408         dn_rt_finish_output(skb, NULL, NULL);
 409         return NET_RX_SUCCESS;
 410 }
 411
 412 /**
 413  * dn_return_long - Return a long packet to its sender
 414  * @skb: The long format packet to return
 415  *
 416  */
 417 static int dn_return_long(struct sk_buff *skb)
 418 {
 419         struct dn_skb_cb *cb;
 420         unsigned char *ptr;
 421         unsigned char *src_addr, *dst_addr;
 422         unsigned char tmp[ETH_ALEN];
 423
 424         /* Add back all headers */
 425         skb_push(skb, skb->data - skb->nh.raw);
 426
 427         if ((skb = skb_unshare(skb, GFP_ATOMIC)) == NULL)
 428                 return NET_RX_DROP;
 429
 430         cb = DN_SKB_CB(skb);
 431         /* Ignore packet length and point to flags */
 432         ptr = skb->data + 2;
 433
 434         /* Skip padding */
 435         if (*ptr & DN_RT_F_PF) {
 436                 char padlen = (*ptr & ~DN_RT_F_PF);
 437                 ptr += padlen;
 438         }
 439
 440         *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
 441         ptr += 2;
 442         dst_addr = ptr;
 443         ptr += 8;
 444         src_addr = ptr;
 445         ptr += 6;
 446         *ptr = 0; /* Zero hop count */
 447
 448         /* Swap source and destination */
 449         memcpy(tmp, src_addr, ETH_ALEN);
 450         memcpy(src_addr, dst_addr, ETH_ALEN);
 451         memcpy(dst_addr, tmp, ETH_ALEN);
 452
 453         skb->pkt_type = PACKET_OUTGOING;
 454         dn_rt_finish_output(skb, dst_addr, src_addr);
 455         return NET_RX_SUCCESS;
 456 }
 457
 458 /**
 459  * dn_route_rx_packet - Try and find a route for an incoming packet
 460  * @skb: The packet to find a route for
 461  *
 462  * Returns: result of input function if route is found, error code otherwise
 463  */
 464 static int dn_route_rx_packet(struct sk_buff *skb)
 465 {
 466         struct dn_skb_cb *cb = DN_SKB_CB(skb);
 467         int err;
 468
 469         if ((err = dn_route_input(skb)) == 0)
 470                 return dst_input(skb);
 471
 472         if (decnet_debug_level & 4) {
 473                 char *devname = skb->dev ? skb->dev->name : "???";
 474                 struct dn_skb_cb *cb = DN_SKB_CB(skb);
 475                 printk(KERN_DEBUG
 476                         "DECnet: dn_route_rx_packet: rt_flags=0x%02x dev=%s len=%d src=0x%04hx dst=0x%04hx err=%d type=%d\n",
 477                         (int)cb->rt_flags, devname, skb->len,
 478                         dn_ntohs(cb->src), dn_ntohs(cb->dst),
 479                         err, skb->pkt_type);
 480         }
 481
 482         if ((skb->pkt_type == PACKET_HOST) && (cb->rt_flags & DN_RT_F_RQR)) {
 483                 switch(cb->rt_flags & DN_RT_PKT_MSK) {
 484                         case DN_RT_PKT_SHORT:
 485                                 return dn_return_short(skb);
 486                         case DN_RT_PKT_LONG:
 487                                 return dn_return_long(skb);
 488                 }
 489         }
 490
 491         kfree_skb(skb);
 492         return NET_RX_DROP;
 493 }
 494
 495 static int dn_route_rx_long(struct sk_buff *skb)
 496 {
 497         struct dn_skb_cb *cb = DN_SKB_CB(skb);
 498         unsigned char *ptr = skb->data;
 499
 500         if (!pskb_may_pull(skb, 21)) /* 20 for long header, 1 for shortest nsp */
 501                 goto drop_it;
 502
 503         skb_pull(skb, 20);
 504         skb->h.raw = skb->data;
 505
 506         /* Destination info */
 507         ptr += 2;
 508         cb->dst = dn_eth2dn(ptr);
 509         if (memcmp(ptr, dn_hiord_addr, 4) != 0)
 510                 goto drop_it;
 511         ptr += 6;
 512
 513
 514         /* Source info */
 515         ptr += 2;
 516         cb->src = dn_eth2dn(ptr);
 517         if (memcmp(ptr, dn_hiord_addr, 4) != 0)
 518                 goto drop_it;
 519         ptr += 6;
 520         /* Other junk */
 521         ptr++;
 522         cb->hops = *ptr++; /* Visit Count */
 523
 524         return NF_HOOK(PF_DECnet, NF_DN_PRE_ROUTING, skb, skb->dev, NULL, dn_route_rx_packet);
 525
 526 drop_it:
 527         kfree_skb(skb);
 528         return NET_RX_DROP;
 529 }
 530
 531
 532
 533 static int dn_route_rx_short(struct sk_buff *skb)
 534 {
 535         struct dn_skb_cb *cb = DN_SKB_CB(skb);
 536         unsigned char *ptr = skb->data;
 537
 538         if (!pskb_may_pull(skb, 6)) /* 5 for short header + 1 for shortest nsp */
 539                 goto drop_it;
 540
 541         skb_pull(skb, 5);
 542         skb->h.raw = skb->data;
 543
 544         cb->dst = *(__le16 *)ptr;
 545         ptr += 2;
 546         cb->src = *(__le16 *)ptr;
 547         ptr += 2;
 548         cb->hops = *ptr & 0x3f;
 549
 550         return NF_HOOK(PF_DECnet, NF_DN_PRE_ROUTING, skb, skb->dev, NULL, dn_route_rx_packet);
 551
 552 drop_it:
 553         kfree_skb(skb);
 554         return NET_RX_DROP;
 555 }
 556
 557 static int dn_route_discard(struct sk_buff *skb)
 558 {
 559         /*
 560          * I know we drop the packet here, but thats considered success in
 561          * this case
 562          */
 563         kfree_skb(skb);
 564         return NET_RX_SUCCESS;
 565 }
 566
 567 static int dn_route_ptp_hello(struct sk_buff *skb)
 568 {
 569         dn_dev_hello(skb);
 570         dn_neigh_pointopoint_hello(skb);
 571         return NET_RX_SUCCESS;
 572 }
 573
 574 int dn_route_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
 575 {
 576         struct dn_skb_cb *cb;
 577         unsigned char flags = 0;
 578         __u16 len = dn_ntohs(*(__le16 *)skb->data);
 579         struct dn_dev *dn = (struct dn_dev *)dev->dn_ptr;
 580         unsigned char padlen = 0;
 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->nh.raw = skb->data;
 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;
 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;
 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;
 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  * Drop packet. This is used for endnodes and for
 765  * when we should not be forwarding packets from
 766  * this dest.
 767  */
 768 static int dn_blackhole(struct sk_buff *skb)
 769 {
 770         kfree_skb(skb);
 771         return NET_RX_DROP;
 772 }
 773
 774 /*
 775  * Used to catch bugs. This should never normally get
 776  * called.
 777  */
 778 static int dn_rt_bug(struct sk_buff *skb)
 779 {
 780         if (net_ratelimit()) {
 781                 struct dn_skb_cb *cb = DN_SKB_CB(skb);
 782
 783                 printk(KERN_DEBUG "dn_rt_bug: skb from:%04x to:%04x\n",
 784                                 dn_ntohs(cb->src), dn_ntohs(cb->dst));
 785         }
 786
 787         kfree_skb(skb);
 788
 789         return NET_RX_BAD;
 790 }
 791
 792 static int dn_rt_set_next_hop(struct dn_route *rt, struct dn_fib_res *res)
 793 {
 794         struct dn_fib_info *fi = res->fi;
 795         struct net_device *dev = rt->u.dst.dev;
 796         struct neighbour *n;
 797         unsigned mss;
 798
 799         if (fi) {
 800                 if (DN_FIB_RES_GW(*res) &&
 801                     DN_FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
 802                         rt->rt_gateway = DN_FIB_RES_GW(*res);
 803                 memcpy(rt->u.dst.metrics, fi->fib_metrics,
 804                        sizeof(rt->u.dst.metrics));
 805         }
 806         rt->rt_type = res->type;
 807
 808         if (dev != NULL && rt->u.dst.neighbour == NULL) {
 809                 n = __neigh_lookup_errno(&dn_neigh_table, &rt->rt_gateway, dev);
 810                 if (IS_ERR(n))
 811                         return PTR_ERR(n);
 812                 rt->u.dst.neighbour = n;
 813         }
 814
 815         if (rt->u.dst.metrics[RTAX_MTU-1] == 0 ||
 816             rt->u.dst.metrics[RTAX_MTU-1] > rt->u.dst.dev->mtu)
 817                 rt->u.dst.metrics[RTAX_MTU-1] = rt->u.dst.dev->mtu;
 818         mss = dn_mss_from_pmtu(dev, dst_mtu(&rt->u.dst));
 819         if (rt->u.dst.metrics[RTAX_ADVMSS-1] == 0 ||
 820             rt->u.dst.metrics[RTAX_ADVMSS-1] > mss)
 821                 rt->u.dst.metrics[RTAX_ADVMSS-1] = mss;
 822         return 0;
 823 }
 824
 825 static inline int dn_match_addr(__le16 addr1, __le16 addr2)
 826 {
 827         __u16 tmp = dn_ntohs(addr1) ^ dn_ntohs(addr2);
 828         int match = 16;
 829         while(tmp) {
 830                 tmp >>= 1;
 831                 match--;
 832         }
 833         return match;
 834 }
 835
 836 static __le16 dnet_select_source(const struct net_device *dev, __le16 daddr, int scope)
 837 {
 838         __le16 saddr = 0;
 839         struct dn_dev *dn_db = dev->dn_ptr;
 840         struct dn_ifaddr *ifa;
 841         int best_match = 0;
 842         int ret;
 843
 844         read_lock(&dev_base_lock);
 845         for(ifa = dn_db->ifa_list; ifa; ifa = ifa->ifa_next) {
 846                 if (ifa->ifa_scope > scope)
 847                         continue;
 848                 if (!daddr) {
 849                         saddr = ifa->ifa_local;
 850                         break;
 851                 }
 852                 ret = dn_match_addr(daddr, ifa->ifa_local);
 853                 if (ret > best_match)
 854                         saddr = ifa->ifa_local;
 855                 if (best_match == 0)
 856                         saddr = ifa->ifa_local;
 857         }
 858         read_unlock(&dev_base_lock);
 859
 860         return saddr;
 861 }
 862
 863 static inline __le16 __dn_fib_res_prefsrc(struct dn_fib_res *res)
 864 {
 865         return dnet_select_source(DN_FIB_RES_DEV(*res), DN_FIB_RES_GW(*res), res->scope);
 866 }
 867
 868 static inline __le16 dn_fib_rules_map_destination(__le16 daddr, struct dn_fib_res *res)
 869 {
 870         __le16 mask = dnet_make_mask(res->prefixlen);
 871         return (daddr&~mask)|res->fi->fib_nh->nh_gw;
 872 }
 873
 874 static int dn_route_output_slow(struct dst_entry **pprt, const struct flowi *oldflp, int try_hard)
 875 {
 876         struct flowi fl = { .nl_u = { .dn_u =
 877                                       { .daddr = oldflp->fld_dst,
 878                                         .saddr = oldflp->fld_src,
 879                                         .scope = RT_SCOPE_UNIVERSE,
 880 #ifdef CONFIG_DECNET_ROUTE_FWMARK
 881                                         .fwmark = oldflp->fld_fwmark
 882 #endif
 883                                      } },
 884                             .iif = loopback_dev.ifindex,
 885                             .oif = oldflp->oif };
 886         struct dn_route *rt = NULL;
 887         struct net_device *dev_out = NULL;
 888         struct neighbour *neigh = NULL;
 889         unsigned hash;
 890         unsigned flags = 0;
 891         struct dn_fib_res res = { .fi = NULL, .type = RTN_UNICAST };
 892         int err;
 893         int free_res = 0;
 894         __le16 gateway = 0;
 895
 896         if (decnet_debug_level & 16)
 897                 printk(KERN_DEBUG
 898                        "dn_route_output_slow: dst=%04x src=%04x mark=%d"
 899                        " iif=%d oif=%d\n", dn_ntohs(oldflp->fld_dst),
 900                        dn_ntohs(oldflp->fld_src),
 901                        oldflp->fld_fwmark, loopback_dev.ifindex, oldflp->oif);
 902
 903         /* If we have an output interface, verify its a DECnet device */
 904         if (oldflp->oif) {
 905                 dev_out = dev_get_by_index(oldflp->oif);
 906                 err = -ENODEV;
 907                 if (dev_out && dev_out->dn_ptr == NULL) {
 908                         dev_put(dev_out);
 909                         dev_out = NULL;
 910                 }
 911                 if (dev_out == NULL)
 912                         goto out;
 913         }
 914
 915         /* If we have a source address, verify that its a local address */
 916         if (oldflp->fld_src) {
 917                 err = -EADDRNOTAVAIL;
 918
 919                 if (dev_out) {
 920                         if (dn_dev_islocal(dev_out, oldflp->fld_src))
 921                                 goto source_ok;
 922                         dev_put(dev_out);
 923                         goto out;
 924                 }
 925                 read_lock(&dev_base_lock);
 926                 for(dev_out = dev_base; dev_out; dev_out = dev_out->next) {
 927                         if (!dev_out->dn_ptr)
 928                                 continue;
 929                         if (dn_dev_islocal(dev_out, oldflp->fld_src))
 930                                 break;
 931                 }
 932                 read_unlock(&dev_base_lock);
 933                 if (dev_out == NULL)
 934                         goto out;
 935                 dev_hold(dev_out);
 936 source_ok:
 937                 ;
 938         }
 939
 940         /* No destination? Assume its local */
 941         if (!fl.fld_dst) {
 942                 fl.fld_dst = fl.fld_src;
 943
 944                 err = -EADDRNOTAVAIL;
 945                 if (dev_out)
 946                         dev_put(dev_out);
 947                 dev_out = &loopback_dev;
 948                 dev_hold(dev_out);
 949                 if (!fl.fld_dst) {
 950                         fl.fld_dst =
 951                         fl.fld_src = dnet_select_source(dev_out, 0,
 952                                                        RT_SCOPE_HOST);
 953                         if (!fl.fld_dst)
 954                                 goto out;
 955                 }
 956                 fl.oif = loopback_dev.ifindex;
 957                 res.type = RTN_LOCAL;
 958                 goto make_route;
 959         }
 960
 961         if (decnet_debug_level & 16)
 962                 printk(KERN_DEBUG
 963                        "dn_route_output_slow: initial checks complete."
 964                        " dst=%o4x src=%04x oif=%d try_hard=%d\n",
 965                        dn_ntohs(fl.fld_dst), dn_ntohs(fl.fld_src),
 966                        fl.oif, try_hard);
 967
 968         /*
 969          * N.B. If the kernel is compiled without router support then
 970          * dn_fib_lookup() will evaluate to non-zero so this if () block
 971          * will always be executed.
 972          */
 973         err = -ESRCH;
 974         if (try_hard || (err = dn_fib_lookup(&fl, &res)) != 0) {
 975                 struct dn_dev *dn_db;
 976                 if (err != -ESRCH)
 977                         goto out;
 978                 /*
 979                  * Here the fallback is basically the standard algorithm for
 980                  * routing in endnodes which is described in the DECnet routing
 981                  * docs
 982                  *
 983                  * If we are not trying hard, look in neighbour cache.
 984                  * The result is tested to ensure that if a specific output
 985                  * device/source address was requested, then we honour that
 986                  * here
 987                  */
 988                 if (!try_hard) {
 989                         neigh = neigh_lookup_nodev(&dn_neigh_table, &fl.fld_dst);
 990                         if (neigh) {
 991                                 if ((oldflp->oif &&
 992                                     (neigh->dev->ifindex != oldflp->oif)) ||
 993                                     (oldflp->fld_src &&
 994                                     (!dn_dev_islocal(neigh->dev,
 995                                                       oldflp->fld_src)))) {
 996                                         neigh_release(neigh);
 997                                         neigh = NULL;
 998                                 } else {
 999                                         if (dev_out)
1000                                                 dev_put(dev_out);
1001                                         if (dn_dev_islocal(neigh->dev, fl.fld_dst)) {
1002                                                 dev_out = &loopback_dev;
1003                                                 res.type = RTN_LOCAL;
1004                                         } else {
1005                                                 dev_out = neigh->dev;
1006                                         }
1007                                         dev_hold(dev_out);
1008                                         goto select_source;
1009                                 }
1010                         }
1011                 }
1012
1013                 /* Not there? Perhaps its a local address */
1014                 if (dev_out == NULL)
1015                         dev_out = dn_dev_get_default();
1016                 err = -ENODEV;
1017                 if (dev_out == NULL)
1018                         goto out;
1019                 dn_db = dev_out->dn_ptr;
1020                 /* Possible improvement - check all devices for local addr */
1021                 if (dn_dev_islocal(dev_out, fl.fld_dst)) {
1022                         dev_put(dev_out);
1023                         dev_out = &loopback_dev;
1024                         dev_hold(dev_out);
1025                         res.type = RTN_LOCAL;
1026                         goto select_source;
1027                 }
1028                 /* Not local either.... try sending it to the default router */
1029                 neigh = neigh_clone(dn_db->router);
1030                 BUG_ON(neigh && neigh->dev != dev_out);
1031
1032                 /* Ok then, we assume its directly connected and move on */
1033 select_source:
1034                 if (neigh)
1035                         gateway = ((struct dn_neigh *)neigh)->addr;
1036                 if (gateway == 0)
1037                         gateway = fl.fld_dst;
1038                 if (fl.fld_src == 0) {
1039                         fl.fld_src = dnet_select_source(dev_out, gateway,
1040                                                          res.type == RTN_LOCAL ?
1041                                                          RT_SCOPE_HOST :
1042                                                          RT_SCOPE_LINK);
1043                         if (fl.fld_src == 0 && res.type != RTN_LOCAL)
1044                                 goto e_addr;
1045                 }
1046                 fl.oif = dev_out->ifindex;
1047                 goto make_route;
1048         }
1049         free_res = 1;
1050
1051         if (res.type == RTN_NAT)
1052                 goto e_inval;
1053
1054         if (res.type == RTN_LOCAL) {
1055                 if (!fl.fld_src)
1056                         fl.fld_src = fl.fld_dst;
1057                 if (dev_out)
1058                         dev_put(dev_out);
1059                 dev_out = &loopback_dev;
1060                 dev_hold(dev_out);
1061                 fl.oif = dev_out->ifindex;
1062                 if (res.fi)
1063                         dn_fib_info_put(res.fi);
1064                 res.fi = NULL;
1065                 goto make_route;
1066         }
1067
1068         if (res.fi->fib_nhs > 1 && fl.oif == 0)
1069                 dn_fib_select_multipath(&fl, &res);
1070
1071         /*
1072          * We could add some logic to deal with default routes here and
1073          * get rid of some of the special casing above.
1074          */
1075
1076         if (!fl.fld_src)
1077                 fl.fld_src = DN_FIB_RES_PREFSRC(res);
1078
1079         if (dev_out)
1080                 dev_put(dev_out);
1081         dev_out = DN_FIB_RES_DEV(res);
1082         dev_hold(dev_out);
1083         fl.oif = dev_out->ifindex;
1084         gateway = DN_FIB_RES_GW(res);
1085
1086 make_route:
1087         if (dev_out->flags & IFF_LOOPBACK)
1088                 flags |= RTCF_LOCAL;
1089
1090         rt = dst_alloc(&dn_dst_ops);
1091         if (rt == NULL)
1092                 goto e_nobufs;
1093
1094         atomic_set(&rt->u.dst.__refcnt, 1);
1095         rt->u.dst.flags   = DST_HOST;
1096
1097         rt->fl.fld_src    = oldflp->fld_src;
1098         rt->fl.fld_dst    = oldflp->fld_dst;
1099         rt->fl.oif        = oldflp->oif;
1100         rt->fl.iif        = 0;
1101 #ifdef CONFIG_DECNET_ROUTE_FWMARK
1102         rt->fl.fld_fwmark = oldflp->fld_fwmark;
1103 #endif
1104
1105         rt->rt_saddr      = fl.fld_src;
1106         rt->rt_daddr      = fl.fld_dst;
1107         rt->rt_gateway    = gateway ? gateway : fl.fld_dst;
1108         rt->rt_local_src  = fl.fld_src;
1109
1110         rt->rt_dst_map    = fl.fld_dst;
1111         rt->rt_src_map    = fl.fld_src;
1112
1113         rt->u.dst.dev = dev_out;
1114         dev_hold(dev_out);
1115         rt->u.dst.neighbour = neigh;
1116         neigh = NULL;
1117
1118         rt->u.dst.lastuse = jiffies;
1119         rt->u.dst.output  = dn_output;
1120         rt->u.dst.input   = dn_rt_bug;
1121         rt->rt_flags      = flags;
1122         if (flags & RTCF_LOCAL)
1123                 rt->u.dst.input = dn_nsp_rx;
1124
1125         err = dn_rt_set_next_hop(rt, &res);
1126         if (err)
1127                 goto e_neighbour;
1128
1129         hash = dn_hash(rt->fl.fld_src, rt->fl.fld_dst);
1130         dn_insert_route(rt, hash, (struct dn_route **)pprt);
1131
1132 done:
1133         if (neigh)
1134                 neigh_release(neigh);
1135         if (free_res)
1136                 dn_fib_res_put(&res);
1137         if (dev_out)
1138                 dev_put(dev_out);
1139 out:
1140         return err;
1141
1142 e_addr:
1143         err = -EADDRNOTAVAIL;
1144         goto done;
1145 e_inval:
1146         err = -EINVAL;
1147         goto done;
1148 e_nobufs:
1149         err = -ENOBUFS;
1150         goto done;
1151 e_neighbour:
1152         dst_free(&rt->u.dst);
1153         goto e_nobufs;
1154 }
1155
1156
1157 /*
1158  * N.B. The flags may be moved into the flowi at some future stage.
1159  */
1160 static int __dn_route_output_key(struct dst_entry **pprt, const struct flowi *flp, int flags)
1161 {
1162         unsigned hash = dn_hash(flp->fld_src, flp->fld_dst);
1163         struct dn_route *rt = NULL;
1164
1165         if (!(flags & MSG_TRYHARD)) {
1166                 rcu_read_lock_bh();
1167                 for(rt = rcu_dereference(dn_rt_hash_table[hash].chain); rt;
1168                         rt = rcu_dereference(rt->u.rt_next)) {
1169                         if ((flp->fld_dst == rt->fl.fld_dst) &&
1170                             (flp->fld_src == rt->fl.fld_src) &&
1171 #ifdef CONFIG_DECNET_ROUTE_FWMARK
1172                             (flp->fld_fwmark == rt->fl.fld_fwmark) &&
1173 #endif
1174                             (rt->fl.iif == 0) &&
1175                             (rt->fl.oif == flp->oif)) {
1176                                 rt->u.dst.lastuse = jiffies;
1177                                 dst_hold(&rt->u.dst);
1178                                 rt->u.dst.__use++;
1179                                 rcu_read_unlock_bh();
1180                                 *pprt = &rt->u.dst;
1181                                 return 0;
1182                         }
1183                 }
1184                 rcu_read_unlock_bh();
1185         }
1186
1187         return dn_route_output_slow(pprt, flp, flags);
1188 }
1189
1190 static int dn_route_output_key(struct dst_entry **pprt, struct flowi *flp, int flags)
1191 {
1192         int err;
1193
1194         err = __dn_route_output_key(pprt, flp, flags);
1195         if (err == 0 && flp->proto) {
1196                 err = xfrm_lookup(pprt, flp, NULL, 0);
1197         }
1198         return err;
1199 }
1200
1201 int dn_route_output_sock(struct dst_entry **pprt, struct flowi *fl, struct sock *sk, int flags)
1202 {
1203         int err;
1204
1205         err = __dn_route_output_key(pprt, fl, flags & MSG_TRYHARD);
1206         if (err == 0 && fl->proto) {
1207                 err = xfrm_lookup(pprt, fl, sk, !(flags & MSG_DONTWAIT));
1208         }
1209         return err;
1210 }
1211
1212 static int dn_route_input_slow(struct sk_buff *skb)
1213 {
1214         struct dn_route *rt = NULL;
1215         struct dn_skb_cb *cb = DN_SKB_CB(skb);
1216         struct net_device *in_dev = skb->dev;
1217         struct net_device *out_dev = NULL;
1218         struct dn_dev *dn_db;
1219         struct neighbour *neigh = NULL;
1220         unsigned hash;
1221         int flags = 0;
1222         __le16 gateway = 0;
1223         __le16 local_src = 0;
1224         struct flowi fl = { .nl_u = { .dn_u =
1225                                      { .daddr = cb->dst,
1226                                        .saddr = cb->src,
1227                                        .scope = RT_SCOPE_UNIVERSE,
1228 #ifdef CONFIG_DECNET_ROUTE_FWMARK
1229                                        .fwmark = skb->nfmark
1230 #endif
1231                                     } },
1232                             .iif = skb->dev->ifindex };
1233         struct dn_fib_res res = { .fi = NULL, .type = RTN_UNREACHABLE };
1234         int err = -EINVAL;
1235         int free_res = 0;
1236
1237         dev_hold(in_dev);
1238
1239         if ((dn_db = in_dev->dn_ptr) == NULL)
1240                 goto out;
1241
1242         /* Zero source addresses are not allowed */
1243         if (fl.fld_src == 0)
1244                 goto out;
1245
1246         /*
1247          * In this case we've just received a packet from a source
1248          * outside ourselves pretending to come from us. We don't
1249          * allow it any further to prevent routing loops, spoofing and
1250          * other nasties. Loopback packets already have the dst attached
1251          * so this only affects packets which have originated elsewhere.
1252          */
1253         err  = -ENOTUNIQ;
1254         if (dn_dev_islocal(in_dev, cb->src))
1255                 goto out;
1256
1257         err = dn_fib_lookup(&fl, &res);
1258         if (err) {
1259                 if (err != -ESRCH)
1260                         goto out;
1261                 /*
1262                  * Is the destination us ?
1263                  */
1264                 if (!dn_dev_islocal(in_dev, cb->dst))
1265                         goto e_inval;
1266
1267                 res.type = RTN_LOCAL;
1268                 flags |= RTCF_DIRECTSRC;
1269         } else {
1270                 __le16 src_map = fl.fld_src;
1271                 free_res = 1;
1272
1273                 out_dev = DN_FIB_RES_DEV(res);
1274                 if (out_dev == NULL) {
1275                         if (net_ratelimit())
1276                                 printk(KERN_CRIT "Bug in dn_route_input_slow() "
1277                                                  "No output device\n");
1278                         goto e_inval;
1279                 }
1280                 dev_hold(out_dev);
1281
1282                 if (res.r)
1283                         src_map = dn_fib_rules_policy(fl.fld_src, &res, &flags);
1284
1285                 gateway = DN_FIB_RES_GW(res);
1286                 if (res.type == RTN_NAT) {
1287                         fl.fld_dst = dn_fib_rules_map_destination(fl.fld_dst, &res);
1288                         dn_fib_res_put(&res);
1289                         free_res = 0;
1290                         if (dn_fib_lookup(&fl, &res))
1291                                 goto e_inval;
1292                         free_res = 1;
1293                         if (res.type != RTN_UNICAST)
1294                                 goto e_inval;
1295                         flags |= RTCF_DNAT;
1296                         gateway = fl.fld_dst;
1297                 }
1298                 fl.fld_src = src_map;
1299         }
1300
1301         switch(res.type) {
1302         case RTN_UNICAST:
1303                 /*
1304                  * Forwarding check here, we only check for forwarding
1305                  * being turned off, if you want to only forward intra
1306                  * area, its up to you to set the routing tables up
1307                  * correctly.
1308                  */
1309                 if (dn_db->parms.forwarding == 0)
1310                         goto e_inval;
1311
1312                 if (res.fi->fib_nhs > 1 && fl.oif == 0)
1313                         dn_fib_select_multipath(&fl, &res);
1314
1315                 /*
1316                  * Check for out_dev == in_dev. We use the RTCF_DOREDIRECT
1317                  * flag as a hint to set the intra-ethernet bit when
1318                  * forwarding. If we've got NAT in operation, we don't do
1319                  * this optimisation.
1320                  */
1321                 if (out_dev == in_dev && !(flags & RTCF_NAT))
1322                         flags |= RTCF_DOREDIRECT;
1323
1324                 local_src = DN_FIB_RES_PREFSRC(res);
1325
1326         case RTN_BLACKHOLE:
1327         case RTN_UNREACHABLE:
1328                 break;
1329         case RTN_LOCAL:
1330                 flags |= RTCF_LOCAL;
1331                 fl.fld_src = cb->dst;
1332                 fl.fld_dst = cb->src;
1333
1334                 /* Routing tables gave us a gateway */
1335                 if (gateway)
1336                         goto make_route;
1337
1338                 /* Packet was intra-ethernet, so we know its on-link */
1339                 if (cb->rt_flags | DN_RT_F_IE) {
1340                         gateway = cb->src;
1341                         flags |= RTCF_DIRECTSRC;
1342                         goto make_route;
1343                 }
1344
1345                 /* Use the default router if there is one */
1346                 neigh = neigh_clone(dn_db->router);
1347                 if (neigh) {
1348                         gateway = ((struct dn_neigh *)neigh)->addr;
1349                         goto make_route;
1350                 }
1351
1352                 /* Close eyes and pray */
1353                 gateway = cb->src;
1354                 flags |= RTCF_DIRECTSRC;
1355                 goto make_route;
1356         default:
1357                 goto e_inval;
1358         }
1359
1360 make_route:
1361         rt = dst_alloc(&dn_dst_ops);
1362         if (rt == NULL)
1363                 goto e_nobufs;
1364
1365         rt->rt_saddr      = fl.fld_src;
1366         rt->rt_daddr      = fl.fld_dst;
1367         rt->rt_gateway    = fl.fld_dst;
1368         if (gateway)
1369                 rt->rt_gateway = gateway;
1370         rt->rt_local_src  = local_src ? local_src : rt->rt_saddr;
1371
1372         rt->rt_dst_map    = fl.fld_dst;
1373         rt->rt_src_map    = fl.fld_src;
1374
1375         rt->fl.fld_src    = cb->src;
1376         rt->fl.fld_dst    = cb->dst;
1377         rt->fl.oif        = 0;
1378         rt->fl.iif        = in_dev->ifindex;
1379         rt->fl.fld_fwmark = fl.fld_fwmark;
1380
1381         rt->u.dst.flags = DST_HOST;
1382         rt->u.dst.neighbour = neigh;
1383         rt->u.dst.dev = out_dev;
1384         rt->u.dst.lastuse = jiffies;
1385         rt->u.dst.output = dn_rt_bug;
1386         switch(res.type) {
1387                 case RTN_UNICAST:
1388                         rt->u.dst.input = dn_forward;
1389                         break;
1390                 case RTN_LOCAL:
1391                         rt->u.dst.output = dn_output;
1392                         rt->u.dst.input = dn_nsp_rx;
1393                         rt->u.dst.dev = in_dev;
1394                         flags |= RTCF_LOCAL;
1395                         break;
1396                 default:
1397                 case RTN_UNREACHABLE:
1398                 case RTN_BLACKHOLE:
1399                         rt->u.dst.input = dn_blackhole;
1400         }
1401         rt->rt_flags = flags;
1402         if (rt->u.dst.dev)
1403                 dev_hold(rt->u.dst.dev);
1404
1405         err = dn_rt_set_next_hop(rt, &res);
1406         if (err)
1407                 goto e_neighbour;
1408
1409         hash = dn_hash(rt->fl.fld_src, rt->fl.fld_dst);
1410         dn_insert_route(rt, hash, (struct dn_route **)&skb->dst);
1411
1412 done:
1413         if (neigh)
1414                 neigh_release(neigh);
1415         if (free_res)
1416                 dn_fib_res_put(&res);
1417         dev_put(in_dev);
1418         if (out_dev)
1419                 dev_put(out_dev);
1420 out:
1421         return err;
1422
1423 e_inval:
1424         err = -EINVAL;
1425         goto done;
1426
1427 e_nobufs:
1428         err = -ENOBUFS;
1429         goto done;
1430
1431 e_neighbour:
1432         dst_free(&rt->u.dst);
1433         goto done;
1434 }
1435
1436 int dn_route_input(struct sk_buff *skb)
1437 {
1438         struct dn_route *rt;
1439         struct dn_skb_cb *cb = DN_SKB_CB(skb);
1440         unsigned hash = dn_hash(cb->src, cb->dst);
1441
1442         if (skb->dst)
1443                 return 0;
1444
1445         rcu_read_lock();
1446         for(rt = rcu_dereference(dn_rt_hash_table[hash].chain); rt != NULL;
1447             rt = rcu_dereference(rt->u.rt_next)) {
1448                 if ((rt->fl.fld_src == cb->src) &&
1449                     (rt->fl.fld_dst == cb->dst) &&
1450                     (rt->fl.oif == 0) &&
1451 #ifdef CONFIG_DECNET_ROUTE_FWMARK
1452                     (rt->fl.fld_fwmark == skb->nfmark) &&
1453 #endif
1454                     (rt->fl.iif == cb->iif)) {
1455                         rt->u.dst.lastuse = jiffies;
1456                         dst_hold(&rt->u.dst);
1457                         rt->u.dst.__use++;
1458                         rcu_read_unlock();
1459                         skb->dst = (struct dst_entry *)rt;
1460                         return 0;
1461                 }
1462         }
1463         rcu_read_unlock();
1464
1465         return dn_route_input_slow(skb);
1466 }
1467
1468 static int dn_rt_fill_info(struct sk_buff *skb, u32 pid, u32 seq,
1469                            int event, int nowait, unsigned int flags)
1470 {
1471         struct dn_route *rt = (struct dn_route *)skb->dst;
1472         struct rtmsg *r;
1473         struct nlmsghdr *nlh;
1474         unsigned char *b = skb->tail;
1475         struct rta_cacheinfo ci;
1476
1477         nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*r), flags);
1478         r = NLMSG_DATA(nlh);
1479         r->rtm_family = AF_DECnet;
1480         r->rtm_dst_len = 16;
1481         r->rtm_src_len = 0;
1482         r->rtm_tos = 0;
1483         r->rtm_table = RT_TABLE_MAIN;
1484         r->rtm_type = rt->rt_type;
1485         r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
1486         r->rtm_scope = RT_SCOPE_UNIVERSE;
1487         r->rtm_protocol = RTPROT_UNSPEC;
1488         if (rt->rt_flags & RTCF_NOTIFY)
1489                 r->rtm_flags |= RTM_F_NOTIFY;
1490         RTA_PUT(skb, RTA_DST, 2, &rt->rt_daddr);
1491         if (rt->fl.fld_src) {
1492                 r->rtm_src_len = 16;
1493                 RTA_PUT(skb, RTA_SRC, 2, &rt->fl.fld_src);
1494         }
1495         if (rt->u.dst.dev)
1496                 RTA_PUT(skb, RTA_OIF, sizeof(int), &rt->u.dst.dev->ifindex);
1497         /*
1498          * Note to self - change this if input routes reverse direction when
1499          * they deal only with inputs and not with replies like they do
1500          * currently.
1501          */
1502         RTA_PUT(skb, RTA_PREFSRC, 2, &rt->rt_local_src);
1503         if (rt->rt_daddr != rt->rt_gateway)
1504                 RTA_PUT(skb, RTA_GATEWAY, 2, &rt->rt_gateway);
1505         if (rtnetlink_put_metrics(skb, rt->u.dst.metrics) < 0)
1506                 goto rtattr_failure;
1507         ci.rta_lastuse = jiffies_to_clock_t(jiffies - rt->u.dst.lastuse);
1508         ci.rta_used     = rt->u.dst.__use;
1509         ci.rta_clntref  = atomic_read(&rt->u.dst.__refcnt);
1510         if (rt->u.dst.expires)
1511                 ci.rta_expires = jiffies_to_clock_t(rt->u.dst.expires - jiffies);
1512         else
1513                 ci.rta_expires = 0;
1514         ci.rta_error    = rt->u.dst.error;
1515         ci.rta_id       = ci.rta_ts = ci.rta_tsage = 0;
1516         RTA_PUT(skb, RTA_CACHEINFO, sizeof(ci), &ci);
1517         if (rt->fl.iif)
1518                 RTA_PUT(skb, RTA_IIF, sizeof(int), &rt->fl.iif);
1519
1520         nlh->nlmsg_len = skb->tail - b;
1521         return skb->len;
1522
1523 nlmsg_failure:
1524 rtattr_failure:
1525         skb_trim(skb, b - skb->data);
1526         return -1;
1527 }
1528
1529 /*
1530  * This is called by both endnodes and routers now.
1531  */
1532 int dn_cache_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh, void *arg)
1533 {
1534         struct rtattr **rta = arg;
1535         struct rtmsg *rtm = NLMSG_DATA(nlh);
1536         struct dn_route *rt = NULL;
1537         struct dn_skb_cb *cb;
1538         int err;
1539         struct sk_buff *skb;
1540         struct flowi fl;
1541
1542         memset(&fl, 0, sizeof(fl));
1543         fl.proto = DNPROTO_NSP;
1544
1545         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1546         if (skb == NULL)
1547                 return -ENOBUFS;
1548         skb->mac.raw = skb->data;
1549         cb = DN_SKB_CB(skb);
1550
1551         if (rta[RTA_SRC-1])
1552                 memcpy(&fl.fld_src, RTA_DATA(rta[RTA_SRC-1]), 2);
1553         if (rta[RTA_DST-1])
1554                 memcpy(&fl.fld_dst, RTA_DATA(rta[RTA_DST-1]), 2);
1555         if (rta[RTA_IIF-1])
1556                 memcpy(&fl.iif, RTA_DATA(rta[RTA_IIF-1]), sizeof(int));
1557
1558         if (fl.iif) {
1559                 struct net_device *dev;
1560                 if ((dev = dev_get_by_index(fl.iif)) == NULL) {
1561                         kfree_skb(skb);
1562                         return -ENODEV;
1563                 }
1564                 if (!dev->dn_ptr) {
1565                         dev_put(dev);
1566                         kfree_skb(skb);
1567                         return -ENODEV;
1568                 }
1569                 skb->protocol = __constant_htons(ETH_P_DNA_RT);
1570                 skb->dev = dev;
1571                 cb->src = fl.fld_src;
1572                 cb->dst = fl.fld_dst;
1573                 local_bh_disable();
1574                 err = dn_route_input(skb);
1575                 local_bh_enable();
1576                 memset(cb, 0, sizeof(struct dn_skb_cb));
1577                 rt = (struct dn_route *)skb->dst;
1578                 if (!err && -rt->u.dst.error)
1579                         err = rt->u.dst.error;
1580         } else {
1581                 int oif = 0;
1582                 if (rta[RTA_OIF - 1])
1583                         memcpy(&oif, RTA_DATA(rta[RTA_OIF - 1]), sizeof(int));
1584                 fl.oif = oif;
1585                 err = dn_route_output_key((struct dst_entry **)&rt, &fl, 0);
1586         }
1587
1588         if (skb->dev)
1589                 dev_put(skb->dev);
1590         skb->dev = NULL;
1591         if (err)
1592                 goto out_free;
1593         skb->dst = &rt->u.dst;
1594         if (rtm->rtm_flags & RTM_F_NOTIFY)
1595                 rt->rt_flags |= RTCF_NOTIFY;
1596
1597         NETLINK_CB(skb).dst_pid = NETLINK_CB(in_skb).pid;
1598
1599         err = dn_rt_fill_info(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq, RTM_NEWROUTE, 0, 0);
1600
1601         if (err == 0)
1602                 goto out_free;
1603         if (err < 0) {
1604                 err = -EMSGSIZE;
1605                 goto out_free;
1606         }
1607
1608         err = netlink_unicast(rtnl, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1609
1610         return err;
1611
1612 out_free:
1613         kfree_skb(skb);
1614         return err;
1615 }
1616
1617 /*
1618  * For routers, this is called from dn_fib_dump, but for endnodes its
1619  * called directly from the rtnetlink dispatch table.
1620  */
1621 int dn_cache_dump(struct sk_buff *skb, struct netlink_callback *cb)
1622 {
1623         struct dn_route *rt;
1624         int h, s_h;
1625         int idx, s_idx;
1626
1627         if (NLMSG_PAYLOAD(cb->nlh, 0) < sizeof(struct rtmsg))
1628                 return -EINVAL;
1629         if (!(((struct rtmsg *)NLMSG_DATA(cb->nlh))->rtm_flags&RTM_F_CLONED))
1630                 return 0;
1631
1632         s_h = cb->args[0];
1633         s_idx = idx = cb->args[1];
1634         for(h = 0; h <= dn_rt_hash_mask; h++) {
1635                 if (h < s_h)
1636                         continue;
1637                 if (h > s_h)
1638                         s_idx = 0;
1639                 rcu_read_lock_bh();
1640                 for(rt = rcu_dereference(dn_rt_hash_table[h].chain), idx = 0;
1641                         rt;
1642                         rt = rcu_dereference(rt->u.rt_next), idx++) {
1643                         if (idx < s_idx)
1644                                 continue;
1645                         skb->dst = dst_clone(&rt->u.dst);
1646                         if (dn_rt_fill_info(skb, NETLINK_CB(cb->skb).pid,
1647                                         cb->nlh->nlmsg_seq, RTM_NEWROUTE,
1648                                         1, NLM_F_MULTI) <= 0) {
1649                                 dst_release(xchg(&skb->dst, NULL));
1650                                 rcu_read_unlock_bh();
1651                                 goto done;
1652                         }
1653                         dst_release(xchg(&skb->dst, NULL));
1654                 }
1655                 rcu_read_unlock_bh();
1656         }
1657
1658 done:
1659         cb->args[0] = h;
1660         cb->args[1] = idx;
1661         return skb->len;
1662 }
1663
1664 #ifdef CONFIG_PROC_FS
1665 struct dn_rt_cache_iter_state {
1666         int bucket;
1667 };
1668
1669 static struct dn_route *dn_rt_cache_get_first(struct seq_file *seq)
1670 {
1671         struct dn_route *rt = NULL;
1672         struct dn_rt_cache_iter_state *s = seq->private;
1673
1674         for(s->bucket = dn_rt_hash_mask; s->bucket >= 0; --s->bucket) {
1675                 rcu_read_lock_bh();
1676                 rt = dn_rt_hash_table[s->bucket].chain;
1677                 if (rt)
1678                         break;
1679                 rcu_read_unlock_bh();
1680         }
1681         return rt;
1682 }
1683
1684 static struct dn_route *dn_rt_cache_get_next(struct seq_file *seq, struct dn_route *rt)
1685 {
1686         struct dn_rt_cache_iter_state *s = rcu_dereference(seq->private);
1687
1688         rt = rt->u.rt_next;
1689         while(!rt) {
1690                 rcu_read_unlock_bh();
1691                 if (--s->bucket < 0)
1692                         break;
1693                 rcu_read_lock_bh();
1694                 rt = dn_rt_hash_table[s->bucket].chain;
1695         }
1696         return rt;
1697 }
1698
1699 static void *dn_rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
1700 {
1701         struct dn_route *rt = dn_rt_cache_get_first(seq);
1702
1703         if (rt) {
1704                 while(*pos && (rt = dn_rt_cache_get_next(seq, rt)))
1705                         --*pos;
1706         }
1707         return *pos ? NULL : rt;
1708 }
1709
1710 static void *dn_rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1711 {
1712         struct dn_route *rt = dn_rt_cache_get_next(seq, v);
1713         ++*pos;
1714         return rt;
1715 }
1716
1717 static void dn_rt_cache_seq_stop(struct seq_file *seq, void *v)
1718 {
1719         if (v)
1720                 rcu_read_unlock_bh();
1721 }
1722
1723 static int dn_rt_cache_seq_show(struct seq_file *seq, void *v)
1724 {
1725         struct dn_route *rt = v;
1726         char buf1[DN_ASCBUF_LEN], buf2[DN_ASCBUF_LEN];
1727
1728         seq_printf(seq, "%-8s %-7s %-7s %04d %04d %04d\n",
1729                         rt->u.dst.dev ? rt->u.dst.dev->name : "*",
1730                         dn_addr2asc(dn_ntohs(rt->rt_daddr), buf1),
1731                         dn_addr2asc(dn_ntohs(rt->rt_saddr), buf2),
1732                         atomic_read(&rt->u.dst.__refcnt),
1733                         rt->u.dst.__use,
1734                         (int) dst_metric(&rt->u.dst, RTAX_RTT));
1735         return 0;
1736 }
1737
1738 static struct seq_operations dn_rt_cache_seq_ops = {
1739         .start  = dn_rt_cache_seq_start,
1740         .next   = dn_rt_cache_seq_next,
1741         .stop   = dn_rt_cache_seq_stop,
1742         .show   = dn_rt_cache_seq_show,
1743 };
1744
1745 static int dn_rt_cache_seq_open(struct inode *inode, struct file *file)
1746 {
1747         struct seq_file *seq;
1748         int rc = -ENOMEM;
1749         struct dn_rt_cache_iter_state *s = kmalloc(sizeof(*s), GFP_KERNEL);
1750
1751         if (!s)
1752                 goto out;
1753         rc = seq_open(file, &dn_rt_cache_seq_ops);
1754         if (rc)
1755                 goto out_kfree;
1756         seq             = file->private_data;
1757         seq->private    = s;
1758         memset(s, 0, sizeof(*s));
1759 out:
1760         return rc;
1761 out_kfree:
1762         kfree(s);
1763         goto out;
1764 }
1765
1766 static struct file_operations dn_rt_cache_seq_fops = {
1767         .owner   = THIS_MODULE,
1768         .open    = dn_rt_cache_seq_open,
1769         .read    = seq_read,
1770         .llseek  = seq_lseek,
1771         .release = seq_release_private,
1772 };
1773
1774 #endif /* CONFIG_PROC_FS */
1775
1776 void __init dn_route_init(void)
1777 {
1778         int i, goal, order;
1779
1780         dn_dst_ops.kmem_cachep = kmem_cache_create("dn_dst_cache",
1781                                                    sizeof(struct dn_route),
1782                                                    0, SLAB_HWCACHE_ALIGN,
1783                                                    NULL, NULL);
1784
1785         if (!dn_dst_ops.kmem_cachep)
1786                 panic("DECnet: Failed to allocate dn_dst_cache\n");
1787
1788         init_timer(&dn_route_timer);
1789         dn_route_timer.function = dn_dst_check_expire;
1790         dn_route_timer.expires = jiffies + decnet_dst_gc_interval * HZ;
1791         add_timer(&dn_route_timer);
1792
1793         goal = num_physpages >> (26 - PAGE_SHIFT);
1794
1795         for(order = 0; (1UL << order) < goal; order++)
1796                 /* NOTHING */;
1797
1798         /*
1799          * Only want 1024 entries max, since the table is very, very unlikely
1800          * to be larger than that.
1801          */
1802         while(order && ((((1UL << order) * PAGE_SIZE) /
1803                                 sizeof(struct dn_rt_hash_bucket)) >= 2048))
1804                 order--;
1805
1806         do {
1807                 dn_rt_hash_mask = (1UL << order) * PAGE_SIZE /
1808                         sizeof(struct dn_rt_hash_bucket);
1809                 while(dn_rt_hash_mask & (dn_rt_hash_mask - 1))
1810                         dn_rt_hash_mask--;
1811                 dn_rt_hash_table = (struct dn_rt_hash_bucket *)
1812                         __get_free_pages(GFP_ATOMIC, order);
1813         } while (dn_rt_hash_table == NULL && --order > 0);
1814
1815         if (!dn_rt_hash_table)
1816                 panic("Failed to allocate DECnet route cache hash table\n");
1817
1818         printk(KERN_INFO
1819                 "DECnet: Routing cache hash table of %u buckets, %ldKbytes\n",
1820                 dn_rt_hash_mask,
1821                 (long)(dn_rt_hash_mask*sizeof(struct dn_rt_hash_bucket))/1024);
1822
1823         dn_rt_hash_mask--;
1824         for(i = 0; i <= dn_rt_hash_mask; i++) {
1825                 spin_lock_init(&dn_rt_hash_table[i].lock);
1826                 dn_rt_hash_table[i].chain = NULL;
1827         }
1828
1829         dn_dst_ops.gc_thresh = (dn_rt_hash_mask + 1);
1830
1831         proc_net_fops_create("decnet_cache", S_IRUGO, &dn_rt_cache_seq_fops);
1832 }
1833
1834 void __exit dn_route_cleanup(void)
1835 {
1836         del_timer(&dn_route_timer);
1837         dn_run_flush(0);
1838
1839         proc_net_remove("decnet_cache");
1840 }
1841