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(NFPROTO_DECNET, NF_DN_PRE_ROUTING, skb, skb->dev, NULL,
 522                        dn_route_rx_packet);
 523
 524 drop_it:
 525         kfree_skb(skb);
 526         return NET_RX_DROP;
 527 }
 528
 529
 530
 531 static int dn_route_rx_short(struct sk_buff *skb)
 532 {
 533         struct dn_skb_cb *cb = DN_SKB_CB(skb);
 534         unsigned char *ptr = skb->data;
 535
 536         if (!pskb_may_pull(skb, 6)) /* 5 for short header + 1 for shortest nsp */
 537                 goto drop_it;
 538
 539         skb_pull(skb, 5);
 540         skb_reset_transport_header(skb);
 541
 542         cb->dst = *(__le16 *)ptr;
 543         ptr += 2;
 544         cb->src = *(__le16 *)ptr;
 545         ptr += 2;
 546         cb->hops = *ptr & 0x3f;
 547
 548         return NF_HOOK(NFPROTO_DECNET, NF_DN_PRE_ROUTING, skb, skb->dev, NULL,
 549                        dn_route_rx_packet);
 550
 551 drop_it:
 552         kfree_skb(skb);
 553         return NET_RX_DROP;
 554 }
 555
 556 static int dn_route_discard(struct sk_buff *skb)
 557 {
 558         /*
 559          * I know we drop the packet here, but thats considered success in
 560          * this case
 561          */
 562         kfree_skb(skb);
 563         return NET_RX_SUCCESS;
 564 }
 565
 566 static int dn_route_ptp_hello(struct sk_buff *skb)
 567 {
 568         dn_dev_hello(skb);
 569         dn_neigh_pointopoint_hello(skb);
 570         return NET_RX_SUCCESS;
 571 }
 572
 573 int dn_route_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
 574 {
 575         struct dn_skb_cb *cb;
 576         unsigned char flags = 0;
 577         __u16 len = le16_to_cpu(*(__le16 *)skb->data);
 578         struct dn_dev *dn = (struct dn_dev *)dev->dn_ptr;
 579         unsigned char padlen = 0;
 580
 581         if (!net_eq(dev_net(dev), &init_net))
 582                 goto dump_it;
 583
 584         if (dn == NULL)
 585                 goto dump_it;
 586
 587         if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
 588                 goto out;
 589
 590         if (!pskb_may_pull(skb, 3))
 591                 goto dump_it;
 592
 593         skb_pull(skb, 2);
 594
 595         if (len > skb->len)
 596                 goto dump_it;
 597
 598         skb_trim(skb, len);
 599
 600         flags = *skb->data;
 601
 602         cb = DN_SKB_CB(skb);
 603         cb->stamp = jiffies;
 604         cb->iif = dev->ifindex;
 605
 606         /*
 607          * If we have padding, remove it.
 608          */
 609         if (flags & DN_RT_F_PF) {
 610                 padlen = flags & ~DN_RT_F_PF;
 611                 if (!pskb_may_pull(skb, padlen + 1))
 612                         goto dump_it;
 613                 skb_pull(skb, padlen);
 614                 flags = *skb->data;
 615         }
 616
 617         skb_reset_network_header(skb);
 618
 619         /*
 620          * Weed out future version DECnet
 621          */
 622         if (flags & DN_RT_F_VER)
 623                 goto dump_it;
 624
 625         cb->rt_flags = flags;
 626
 627         if (decnet_debug_level & 1)
 628                 printk(KERN_DEBUG
 629                         "dn_route_rcv: got 0x%02x from %s [%d %d %d]\n",
 630                         (int)flags, (dev) ? dev->name : "???", len, skb->len,
 631                         padlen);
 632
 633         if (flags & DN_RT_PKT_CNTL) {
 634                 if (unlikely(skb_linearize(skb)))
 635                         goto dump_it;
 636
 637                 switch(flags & DN_RT_CNTL_MSK) {
 638                         case DN_RT_PKT_INIT:
 639                                 dn_dev_init_pkt(skb);
 640                                 break;
 641                         case DN_RT_PKT_VERI:
 642                                 dn_dev_veri_pkt(skb);
 643                                 break;
 644                 }
 645
 646                 if (dn->parms.state != DN_DEV_S_RU)
 647                         goto dump_it;
 648
 649                 switch(flags & DN_RT_CNTL_MSK) {
 650                         case DN_RT_PKT_HELO:
 651                                 return NF_HOOK(NFPROTO_DECNET, NF_DN_HELLO,
 652                                                skb, skb->dev, NULL,
 653                                                dn_route_ptp_hello);
 654
 655                         case DN_RT_PKT_L1RT:
 656                         case DN_RT_PKT_L2RT:
 657                                 return NF_HOOK(NFPROTO_DECNET, NF_DN_ROUTE,
 658                                                skb, skb->dev, NULL,
 659                                                dn_route_discard);
 660                         case DN_RT_PKT_ERTH:
 661                                 return NF_HOOK(NFPROTO_DECNET, NF_DN_HELLO,
 662                                                skb, skb->dev, NULL,
 663                                                dn_neigh_router_hello);
 664
 665                         case DN_RT_PKT_EEDH:
 666                                 return NF_HOOK(NFPROTO_DECNET, NF_DN_HELLO,
 667                                                skb, skb->dev, NULL,
 668                                                dn_neigh_endnode_hello);
 669                 }
 670         } else {
 671                 if (dn->parms.state != DN_DEV_S_RU)
 672                         goto dump_it;
 673
 674                 skb_pull(skb, 1); /* Pull flags */
 675
 676                 switch(flags & DN_RT_PKT_MSK) {
 677                         case DN_RT_PKT_LONG:
 678                                 return dn_route_rx_long(skb);
 679                         case DN_RT_PKT_SHORT:
 680                                 return dn_route_rx_short(skb);
 681                 }
 682         }
 683
 684 dump_it:
 685         kfree_skb(skb);
 686 out:
 687         return NET_RX_DROP;
 688 }
 689
 690 static int dn_output(struct sk_buff *skb)
 691 {
 692         struct dst_entry *dst = skb_dst(skb);
 693         struct dn_route *rt = (struct dn_route *)dst;
 694         struct net_device *dev = dst->dev;
 695         struct dn_skb_cb *cb = DN_SKB_CB(skb);
 696         struct neighbour *neigh;
 697
 698         int err = -EINVAL;
 699
 700         if ((neigh = dst->neighbour) == NULL)
 701                 goto error;
 702
 703         skb->dev = dev;
 704
 705         cb->src = rt->rt_saddr;
 706         cb->dst = rt->rt_daddr;
 707
 708         /*
 709          * Always set the Intra-Ethernet bit on all outgoing packets
 710          * originated on this node. Only valid flag from upper layers
 711          * is return-to-sender-requested. Set hop count to 0 too.
 712          */
 713         cb->rt_flags &= ~DN_RT_F_RQR;
 714         cb->rt_flags |= DN_RT_F_IE;
 715         cb->hops = 0;
 716
 717         return NF_HOOK(NFPROTO_DECNET, NF_DN_LOCAL_OUT, skb, NULL, dev,
 718                        neigh->output);
 719
 720 error:
 721         if (net_ratelimit())
 722                 printk(KERN_DEBUG "dn_output: This should not happen\n");
 723
 724         kfree_skb(skb);
 725
 726         return err;
 727 }
 728
 729 static int dn_forward(struct sk_buff *skb)
 730 {
 731         struct dn_skb_cb *cb = DN_SKB_CB(skb);
 732         struct dst_entry *dst = skb_dst(skb);
 733         struct dn_dev *dn_db = dst->dev->dn_ptr;
 734         struct dn_route *rt;
 735         struct neighbour *neigh = dst->neighbour;
 736         int header_len;
 737 #ifdef CONFIG_NETFILTER
 738         struct net_device *dev = skb->dev;
 739 #endif
 740
 741         if (skb->pkt_type != PACKET_HOST)
 742                 goto drop;
 743
 744         /* Ensure that we have enough space for headers */
 745         rt = (struct dn_route *)skb_dst(skb);
 746         header_len = dn_db->use_long ? 21 : 6;
 747         if (skb_cow(skb, LL_RESERVED_SPACE(rt->u.dst.dev)+header_len))
 748                 goto drop;
 749
 750         /*
 751          * Hop count exceeded.
 752          */
 753         if (++cb->hops > 30)
 754                 goto drop;
 755
 756         skb->dev = rt->u.dst.dev;
 757
 758         /*
 759          * If packet goes out same interface it came in on, then set
 760          * the Intra-Ethernet bit. This has no effect for short
 761          * packets, so we don't need to test for them here.
 762          */
 763         cb->rt_flags &= ~DN_RT_F_IE;
 764         if (rt->rt_flags & RTCF_DOREDIRECT)
 765                 cb->rt_flags |= DN_RT_F_IE;
 766
 767         return NF_HOOK(NFPROTO_DECNET, NF_DN_FORWARD, skb, dev, skb->dev,
 768                        neigh->output);
 769
 770 drop:
 771         kfree_skb(skb);
 772         return NET_RX_DROP;
 773 }
 774
 775 /*
 776  * Used to catch bugs. This should never normally get
 777  * called.
 778  */
 779 static int dn_rt_bug(struct sk_buff *skb)
 780 {
 781         if (net_ratelimit()) {
 782                 struct dn_skb_cb *cb = DN_SKB_CB(skb);
 783
 784                 printk(KERN_DEBUG "dn_rt_bug: skb from:%04x to:%04x\n",
 785                                 le16_to_cpu(cb->src), le16_to_cpu(cb->dst));
 786         }
 787
 788         kfree_skb(skb);
 789
 790         return NET_RX_DROP;
 791 }
 792
 793 static int dn_rt_set_next_hop(struct dn_route *rt, struct dn_fib_res *res)
 794 {
 795         struct dn_fib_info *fi = res->fi;
 796         struct net_device *dev = rt->u.dst.dev;
 797         struct neighbour *n;
 798         unsigned mss;
 799
 800         if (fi) {
 801                 if (DN_FIB_RES_GW(*res) &&
 802                     DN_FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
 803                         rt->rt_gateway = DN_FIB_RES_GW(*res);
 804                 memcpy(rt->u.dst.metrics, fi->fib_metrics,
 805                        sizeof(rt->u.dst.metrics));
 806         }
 807         rt->rt_type = res->type;
 808
 809         if (dev != NULL && rt->u.dst.neighbour == NULL) {
 810                 n = __neigh_lookup_errno(&dn_neigh_table, &rt->rt_gateway, dev);
 811                 if (IS_ERR(n))
 812                         return PTR_ERR(n);
 813                 rt->u.dst.neighbour = n;
 814         }
 815
 816         if (dst_metric(&rt->u.dst, RTAX_MTU) == 0 ||
 817             dst_metric(&rt->u.dst, RTAX_MTU) > rt->u.dst.dev->mtu)
 818                 rt->u.dst.metrics[RTAX_MTU-1] = rt->u.dst.dev->mtu;
 819         mss = dn_mss_from_pmtu(dev, dst_mtu(&rt->u.dst));
 820         if (dst_metric(&rt->u.dst, RTAX_ADVMSS) == 0 ||
 821             dst_metric(&rt->u.dst, RTAX_ADVMSS) > mss)
 822                 rt->u.dst.metrics[RTAX_ADVMSS-1] = mss;
 823         return 0;
 824 }
 825
 826 static inline int dn_match_addr(__le16 addr1, __le16 addr2)
 827 {
 828         __u16 tmp = le16_to_cpu(addr1) ^ le16_to_cpu(addr2);
 829         int match = 16;
 830         while(tmp) {
 831                 tmp >>= 1;
 832                 match--;
 833         }
 834         return match;
 835 }
 836
 837 static __le16 dnet_select_source(const struct net_device *dev, __le16 daddr, int scope)
 838 {
 839         __le16 saddr = 0;
 840         struct dn_dev *dn_db = dev->dn_ptr;
 841         struct dn_ifaddr *ifa;
 842         int best_match = 0;
 843         int ret;
 844
 845         read_lock(&dev_base_lock);
 846         for(ifa = dn_db->ifa_list; ifa; ifa = ifa->ifa_next) {
 847                 if (ifa->ifa_scope > scope)
 848                         continue;
 849                 if (!daddr) {
 850                         saddr = ifa->ifa_local;
 851                         break;
 852                 }
 853                 ret = dn_match_addr(daddr, ifa->ifa_local);
 854                 if (ret > best_match)
 855                         saddr = ifa->ifa_local;
 856                 if (best_match == 0)
 857                         saddr = ifa->ifa_local;
 858         }
 859         read_unlock(&dev_base_lock);
 860
 861         return saddr;
 862 }
 863
 864 static inline __le16 __dn_fib_res_prefsrc(struct dn_fib_res *res)
 865 {
 866         return dnet_select_source(DN_FIB_RES_DEV(*res), DN_FIB_RES_GW(*res), res->scope);
 867 }
 868
 869 static inline __le16 dn_fib_rules_map_destination(__le16 daddr, struct dn_fib_res *res)
 870 {
 871         __le16 mask = dnet_make_mask(res->prefixlen);
 872         return (daddr&~mask)|res->fi->fib_nh->nh_gw;
 873 }
 874
 875 static int dn_route_output_slow(struct dst_entry **pprt, const struct flowi *oldflp, int try_hard)
 876 {
 877         struct flowi fl = { .nl_u = { .dn_u =
 878                                       { .daddr = oldflp->fld_dst,
 879                                         .saddr = oldflp->fld_src,
 880                                         .scope = RT_SCOPE_UNIVERSE,
 881                                      } },
 882                             .mark = oldflp->mark,
 883                             .iif = init_net.loopback_dev->ifindex,
 884                             .oif = oldflp->oif };
 885         struct dn_route *rt = NULL;
 886         struct net_device *dev_out = NULL, *dev;
 887         struct neighbour *neigh = NULL;
 888         unsigned hash;
 889         unsigned flags = 0;
 890         struct dn_fib_res res = { .fi = NULL, .type = RTN_UNICAST };
 891         int err;
 892         int free_res = 0;
 893         __le16 gateway = 0;
 894
 895         if (decnet_debug_level & 16)
 896                 printk(KERN_DEBUG
 897                        "dn_route_output_slow: dst=%04x src=%04x mark=%d"
 898                        " iif=%d oif=%d\n", le16_to_cpu(oldflp->fld_dst),
 899                        le16_to_cpu(oldflp->fld_src),
 900                        oldflp->mark, init_net.loopback_dev->ifindex, oldflp->oif);
 901
 902         /* If we have an output interface, verify its a DECnet device */
 903         if (oldflp->oif) {
 904                 dev_out = dev_get_by_index(&init_net, oldflp->oif);
 905                 err = -ENODEV;
 906                 if (dev_out && dev_out->dn_ptr == NULL) {
 907                         dev_put(dev_out);
 908                         dev_out = NULL;
 909                 }
 910                 if (dev_out == NULL)
 911                         goto out;
 912         }
 913
 914         /* If we have a source address, verify that its a local address */
 915         if (oldflp->fld_src) {
 916                 err = -EADDRNOTAVAIL;
 917
 918                 if (dev_out) {
 919                         if (dn_dev_islocal(dev_out, oldflp->fld_src))
 920                                 goto source_ok;
 921                         dev_put(dev_out);
 922                         goto out;
 923                 }
 924                 rcu_read_lock();
 925                 for_each_netdev_rcu(&init_net, dev) {
 926                         if (!dev->dn_ptr)
 927                                 continue;
 928                         if (!dn_dev_islocal(dev, oldflp->fld_src))
 929                                 continue;
 930                         if ((dev->flags & IFF_LOOPBACK) &&
 931                             oldflp->fld_dst &&
 932                             !dn_dev_islocal(dev, oldflp->fld_dst))
 933                                 continue;
 934
 935                         dev_out = dev;
 936                         break;
 937                 }
 938                 rcu_read_unlock();
 939                 if (dev_out == NULL)
 940                         goto out;
 941                 dev_hold(dev_out);
 942 source_ok:
 943                 ;
 944         }
 945
 946         /* No destination? Assume its local */
 947         if (!fl.fld_dst) {
 948                 fl.fld_dst = fl.fld_src;
 949
 950                 err = -EADDRNOTAVAIL;
 951                 if (dev_out)
 952                         dev_put(dev_out);
 953                 dev_out = init_net.loopback_dev;
 954                 dev_hold(dev_out);
 955                 if (!fl.fld_dst) {
 956                         fl.fld_dst =
 957                         fl.fld_src = dnet_select_source(dev_out, 0,
 958                                                        RT_SCOPE_HOST);
 959                         if (!fl.fld_dst)
 960                                 goto out;
 961                 }
 962                 fl.oif = init_net.loopback_dev->ifindex;
 963                 res.type = RTN_LOCAL;
 964                 goto make_route;
 965         }
 966
 967         if (decnet_debug_level & 16)
 968                 printk(KERN_DEBUG
 969                        "dn_route_output_slow: initial checks complete."
 970                        " dst=%o4x src=%04x oif=%d try_hard=%d\n",
 971                        le16_to_cpu(fl.fld_dst), le16_to_cpu(fl.fld_src),
 972                        fl.oif, try_hard);
 973
 974         /*
 975          * N.B. If the kernel is compiled without router support then
 976          * dn_fib_lookup() will evaluate to non-zero so this if () block
 977          * will always be executed.
 978          */
 979         err = -ESRCH;
 980         if (try_hard || (err = dn_fib_lookup(&fl, &res)) != 0) {
 981                 struct dn_dev *dn_db;
 982                 if (err != -ESRCH)
 983                         goto out;
 984                 /*
 985                  * Here the fallback is basically the standard algorithm for
 986                  * routing in endnodes which is described in the DECnet routing
 987                  * docs
 988                  *
 989                  * If we are not trying hard, look in neighbour cache.
 990                  * The result is tested to ensure that if a specific output
 991                  * device/source address was requested, then we honour that
 992                  * here
 993                  */
 994                 if (!try_hard) {
 995                         neigh = neigh_lookup_nodev(&dn_neigh_table, &init_net, &fl.fld_dst);
 996                         if (neigh) {
 997                                 if ((oldflp->oif &&
 998                                     (neigh->dev->ifindex != oldflp->oif)) ||
 999                                     (oldflp->fld_src &&
1000                                     (!dn_dev_islocal(neigh->dev,
1001                                                       oldflp->fld_src)))) {
1002                                         neigh_release(neigh);
1003                                         neigh = NULL;
1004                                 } else {
1005                                         if (dev_out)
1006                                                 dev_put(dev_out);
1007                                         if (dn_dev_islocal(neigh->dev, fl.fld_dst)) {
1008                                                 dev_out = init_net.loopback_dev;
1009                                                 res.type = RTN_LOCAL;
1010                                         } else {
1011                                                 dev_out = neigh->dev;
1012                                         }
1013                                         dev_hold(dev_out);
1014                                         goto select_source;
1015                                 }
1016                         }
1017                 }
1018
1019                 /* Not there? Perhaps its a local address */
1020                 if (dev_out == NULL)
1021                         dev_out = dn_dev_get_default();
1022                 err = -ENODEV;
1023                 if (dev_out == NULL)
1024                         goto out;
1025                 dn_db = dev_out->dn_ptr;
1026                 /* Possible improvement - check all devices for local addr */
1027                 if (dn_dev_islocal(dev_out, fl.fld_dst)) {
1028                         dev_put(dev_out);
1029                         dev_out = init_net.loopback_dev;
1030                         dev_hold(dev_out);
1031                         res.type = RTN_LOCAL;
1032                         goto select_source;
1033                 }
1034                 /* Not local either.... try sending it to the default router */
1035                 neigh = neigh_clone(dn_db->router);
1036                 BUG_ON(neigh && neigh->dev != dev_out);
1037
1038                 /* Ok then, we assume its directly connected and move on */
1039 select_source:
1040                 if (neigh)
1041                         gateway = ((struct dn_neigh *)neigh)->addr;
1042                 if (gateway == 0)
1043                         gateway = fl.fld_dst;
1044                 if (fl.fld_src == 0) {
1045                         fl.fld_src = dnet_select_source(dev_out, gateway,
1046                                                          res.type == RTN_LOCAL ?
1047                                                          RT_SCOPE_HOST :
1048                                                          RT_SCOPE_LINK);
1049                         if (fl.fld_src == 0 && res.type != RTN_LOCAL)
1050                                 goto e_addr;
1051                 }
1052                 fl.oif = dev_out->ifindex;
1053                 goto make_route;
1054         }
1055         free_res = 1;
1056
1057         if (res.type == RTN_NAT)
1058                 goto e_inval;
1059
1060         if (res.type == RTN_LOCAL) {
1061                 if (!fl.fld_src)
1062                         fl.fld_src = fl.fld_dst;
1063                 if (dev_out)
1064                         dev_put(dev_out);
1065                 dev_out = init_net.loopback_dev;
1066                 dev_hold(dev_out);
1067                 fl.oif = dev_out->ifindex;
1068                 if (res.fi)
1069                         dn_fib_info_put(res.fi);
1070                 res.fi = NULL;
1071                 goto make_route;
1072         }
1073
1074         if (res.fi->fib_nhs > 1 && fl.oif == 0)
1075                 dn_fib_select_multipath(&fl, &res);
1076
1077         /*
1078          * We could add some logic to deal with default routes here and
1079          * get rid of some of the special casing above.
1080          */
1081
1082         if (!fl.fld_src)
1083                 fl.fld_src = DN_FIB_RES_PREFSRC(res);
1084
1085         if (dev_out)
1086                 dev_put(dev_out);
1087         dev_out = DN_FIB_RES_DEV(res);
1088         dev_hold(dev_out);
1089         fl.oif = dev_out->ifindex;
1090         gateway = DN_FIB_RES_GW(res);
1091
1092 make_route:
1093         if (dev_out->flags & IFF_LOOPBACK)
1094                 flags |= RTCF_LOCAL;
1095
1096         rt = dst_alloc(&dn_dst_ops);
1097         if (rt == NULL)
1098                 goto e_nobufs;
1099
1100         atomic_set(&rt->u.dst.__refcnt, 1);
1101         rt->u.dst.flags   = DST_HOST;
1102
1103         rt->fl.fld_src    = oldflp->fld_src;
1104         rt->fl.fld_dst    = oldflp->fld_dst;
1105         rt->fl.oif        = oldflp->oif;
1106         rt->fl.iif        = 0;
1107         rt->fl.mark       = oldflp->mark;
1108
1109         rt->rt_saddr      = fl.fld_src;
1110         rt->rt_daddr      = fl.fld_dst;
1111         rt->rt_gateway    = gateway ? gateway : fl.fld_dst;
1112         rt->rt_local_src  = fl.fld_src;
1113
1114         rt->rt_dst_map    = fl.fld_dst;
1115         rt->rt_src_map    = fl.fld_src;
1116
1117         rt->u.dst.dev = dev_out;
1118         dev_hold(dev_out);
1119         rt->u.dst.neighbour = neigh;
1120         neigh = NULL;
1121
1122         rt->u.dst.lastuse = jiffies;
1123         rt->u.dst.output  = dn_output;
1124         rt->u.dst.input   = dn_rt_bug;
1125         rt->rt_flags      = flags;
1126         if (flags & RTCF_LOCAL)
1127                 rt->u.dst.input = dn_nsp_rx;
1128
1129         err = dn_rt_set_next_hop(rt, &res);
1130         if (err)
1131                 goto e_neighbour;
1132
1133         hash = dn_hash(rt->fl.fld_src, rt->fl.fld_dst);
1134         dn_insert_route(rt, hash, (struct dn_route **)pprt);
1135
1136 done:
1137         if (neigh)
1138                 neigh_release(neigh);
1139         if (free_res)
1140                 dn_fib_res_put(&res);
1141         if (dev_out)
1142                 dev_put(dev_out);
1143 out:
1144         return err;
1145
1146 e_addr:
1147         err = -EADDRNOTAVAIL;
1148         goto done;
1149 e_inval:
1150         err = -EINVAL;
1151         goto done;
1152 e_nobufs:
1153         err = -ENOBUFS;
1154         goto done;
1155 e_neighbour:
1156         dst_free(&rt->u.dst);
1157         goto e_nobufs;
1158 }
1159
1160
1161 /*
1162  * N.B. The flags may be moved into the flowi at some future stage.
1163  */
1164 static int __dn_route_output_key(struct dst_entry **pprt, const struct flowi *flp, int flags)
1165 {
1166         unsigned hash = dn_hash(flp->fld_src, flp->fld_dst);
1167         struct dn_route *rt = NULL;
1168
1169         if (!(flags & MSG_TRYHARD)) {
1170                 rcu_read_lock_bh();
1171                 for (rt = rcu_dereference_bh(dn_rt_hash_table[hash].chain); rt;
1172                         rt = rcu_dereference_bh(rt->u.dst.dn_next)) {
1173                         if ((flp->fld_dst == rt->fl.fld_dst) &&
1174                             (flp->fld_src == rt->fl.fld_src) &&
1175                             (flp->mark == rt->fl.mark) &&
1176                             (rt->fl.iif == 0) &&
1177                             (rt->fl.oif == flp->oif)) {
1178                                 dst_use(&rt->u.dst, jiffies);
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(&init_net, 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(&init_net, pprt, fl, sk,
1208                                  (flags & MSG_DONTWAIT) ? 0 : XFRM_LOOKUP_WAIT);
1209         }
1210         return err;
1211 }
1212
1213 static int dn_route_input_slow(struct sk_buff *skb)
1214 {
1215         struct dn_route *rt = NULL;
1216         struct dn_skb_cb *cb = DN_SKB_CB(skb);
1217         struct net_device *in_dev = skb->dev;
1218         struct net_device *out_dev = NULL;
1219         struct dn_dev *dn_db;
1220         struct neighbour *neigh = NULL;
1221         unsigned hash;
1222         int flags = 0;
1223         __le16 gateway = 0;
1224         __le16 local_src = 0;
1225         struct flowi fl = { .nl_u = { .dn_u =
1226                                      { .daddr = cb->dst,
1227                                        .saddr = cb->src,
1228                                        .scope = RT_SCOPE_UNIVERSE,
1229                                     } },
1230                             .mark = skb->mark,
1231                             .iif = skb->dev->ifindex };
1232         struct dn_fib_res res = { .fi = NULL, .type = RTN_UNREACHABLE };
1233         int err = -EINVAL;
1234         int free_res = 0;
1235
1236         dev_hold(in_dev);
1237
1238         if ((dn_db = in_dev->dn_ptr) == NULL)
1239                 goto out;
1240
1241         /* Zero source addresses are not allowed */
1242         if (fl.fld_src == 0)
1243                 goto out;
1244
1245         /*
1246          * In this case we've just received a packet from a source
1247          * outside ourselves pretending to come from us. We don't
1248          * allow it any further to prevent routing loops, spoofing and
1249          * other nasties. Loopback packets already have the dst attached
1250          * so this only affects packets which have originated elsewhere.
1251          */
1252         err  = -ENOTUNIQ;
1253         if (dn_dev_islocal(in_dev, cb->src))
1254                 goto out;
1255
1256         err = dn_fib_lookup(&fl, &res);
1257         if (err) {
1258                 if (err != -ESRCH)
1259                         goto out;
1260                 /*
1261                  * Is the destination us ?
1262                  */
1263                 if (!dn_dev_islocal(in_dev, cb->dst))
1264                         goto e_inval;
1265
1266                 res.type = RTN_LOCAL;
1267         } else {
1268                 __le16 src_map = fl.fld_src;
1269                 free_res = 1;
1270
1271                 out_dev = DN_FIB_RES_DEV(res);
1272                 if (out_dev == NULL) {
1273                         if (net_ratelimit())
1274                                 printk(KERN_CRIT "Bug in dn_route_input_slow() "
1275                                                  "No output device\n");
1276                         goto e_inval;
1277                 }
1278                 dev_hold(out_dev);
1279
1280                 if (res.r)
1281                         src_map = fl.fld_src; /* no NAT support for now */
1282
1283                 gateway = DN_FIB_RES_GW(res);
1284                 if (res.type == RTN_NAT) {
1285                         fl.fld_dst = dn_fib_rules_map_destination(fl.fld_dst, &res);
1286                         dn_fib_res_put(&res);
1287                         free_res = 0;
1288                         if (dn_fib_lookup(&fl, &res))
1289                                 goto e_inval;
1290                         free_res = 1;
1291                         if (res.type != RTN_UNICAST)
1292                                 goto e_inval;
1293                         flags |= RTCF_DNAT;
1294                         gateway = fl.fld_dst;
1295                 }
1296                 fl.fld_src = src_map;
1297         }
1298
1299         switch(res.type) {
1300         case RTN_UNICAST:
1301                 /*
1302                  * Forwarding check here, we only check for forwarding
1303                  * being turned off, if you want to only forward intra
1304                  * area, its up to you to set the routing tables up
1305                  * correctly.
1306                  */
1307                 if (dn_db->parms.forwarding == 0)
1308                         goto e_inval;
1309
1310                 if (res.fi->fib_nhs > 1 && fl.oif == 0)
1311                         dn_fib_select_multipath(&fl, &res);
1312
1313                 /*
1314                  * Check for out_dev == in_dev. We use the RTCF_DOREDIRECT
1315                  * flag as a hint to set the intra-ethernet bit when
1316                  * forwarding. If we've got NAT in operation, we don't do
1317                  * this optimisation.
1318                  */
1319                 if (out_dev == in_dev && !(flags & RTCF_NAT))
1320                         flags |= RTCF_DOREDIRECT;
1321
1322                 local_src = DN_FIB_RES_PREFSRC(res);
1323
1324         case RTN_BLACKHOLE:
1325         case RTN_UNREACHABLE:
1326                 break;
1327         case RTN_LOCAL:
1328                 flags |= RTCF_LOCAL;
1329                 fl.fld_src = cb->dst;
1330                 fl.fld_dst = cb->src;
1331
1332                 /* Routing tables gave us a gateway */
1333                 if (gateway)
1334                         goto make_route;
1335
1336                 /* Packet was intra-ethernet, so we know its on-link */
1337                 if (cb->rt_flags & DN_RT_F_IE) {
1338                         gateway = cb->src;
1339                         flags |= RTCF_DIRECTSRC;
1340                         goto make_route;
1341                 }
1342
1343                 /* Use the default router if there is one */
1344                 neigh = neigh_clone(dn_db->router);
1345                 if (neigh) {
1346                         gateway = ((struct dn_neigh *)neigh)->addr;
1347                         goto make_route;
1348                 }
1349
1350                 /* Close eyes and pray */
1351                 gateway = cb->src;
1352                 flags |= RTCF_DIRECTSRC;
1353                 goto make_route;
1354         default:
1355                 goto e_inval;
1356         }
1357
1358 make_route:
1359         rt = dst_alloc(&dn_dst_ops);
1360         if (rt == NULL)
1361                 goto e_nobufs;
1362
1363         rt->rt_saddr      = fl.fld_src;
1364         rt->rt_daddr      = fl.fld_dst;
1365         rt->rt_gateway    = fl.fld_dst;
1366         if (gateway)
1367                 rt->rt_gateway = gateway;
1368         rt->rt_local_src  = local_src ? local_src : rt->rt_saddr;
1369
1370         rt->rt_dst_map    = fl.fld_dst;
1371         rt->rt_src_map    = fl.fld_src;
1372
1373         rt->fl.fld_src    = cb->src;
1374         rt->fl.fld_dst    = cb->dst;
1375         rt->fl.oif        = 0;
1376         rt->fl.iif        = in_dev->ifindex;
1377         rt->fl.mark       = fl.mark;
1378
1379         rt->u.dst.flags = DST_HOST;
1380         rt->u.dst.neighbour = neigh;
1381         rt->u.dst.dev = out_dev;
1382         rt->u.dst.lastuse = jiffies;
1383         rt->u.dst.output = dn_rt_bug;
1384         switch(res.type) {
1385                 case RTN_UNICAST:
1386                         rt->u.dst.input = dn_forward;
1387                         break;
1388                 case RTN_LOCAL:
1389                         rt->u.dst.output = dn_output;
1390                         rt->u.dst.input = dn_nsp_rx;
1391                         rt->u.dst.dev = in_dev;
1392                         flags |= RTCF_LOCAL;
1393                         break;
1394                 default:
1395                 case RTN_UNREACHABLE:
1396                 case RTN_BLACKHOLE:
1397                         rt->u.dst.input = dst_discard;
1398         }
1399         rt->rt_flags = flags;
1400         if (rt->u.dst.dev)
1401                 dev_hold(rt->u.dst.dev);
1402
1403         err = dn_rt_set_next_hop(rt, &res);
1404         if (err)
1405                 goto e_neighbour;
1406
1407         hash = dn_hash(rt->fl.fld_src, rt->fl.fld_dst);
1408         dn_insert_route(rt, hash, &rt);
1409         skb_dst_set(skb, &rt->u.dst);
1410
1411 done:
1412         if (neigh)
1413                 neigh_release(neigh);
1414         if (free_res)
1415                 dn_fib_res_put(&res);
1416         dev_put(in_dev);
1417         if (out_dev)
1418                 dev_put(out_dev);
1419 out:
1420         return err;
1421
1422 e_inval:
1423         err = -EINVAL;
1424         goto done;
1425
1426 e_nobufs:
1427         err = -ENOBUFS;
1428         goto done;
1429
1430 e_neighbour:
1431         dst_free(&rt->u.dst);
1432         goto done;
1433 }
1434
1435 static int dn_route_input(struct sk_buff *skb)
1436 {
1437         struct dn_route *rt;
1438         struct dn_skb_cb *cb = DN_SKB_CB(skb);
1439         unsigned hash = dn_hash(cb->src, cb->dst);
1440
1441         if (skb_dst(skb))
1442                 return 0;
1443
1444         rcu_read_lock();
1445         for(rt = rcu_dereference(dn_rt_hash_table[hash].chain); rt != NULL;
1446             rt = rcu_dereference(rt->u.dst.dn_next)) {
1447                 if ((rt->fl.fld_src == cb->src) &&
1448                     (rt->fl.fld_dst == cb->dst) &&
1449                     (rt->fl.oif == 0) &&
1450                     (rt->fl.mark == skb->mark) &&
1451                     (rt->fl.iif == cb->iif)) {
1452                         dst_use(&rt->u.dst, jiffies);
1453                         rcu_read_unlock();
1454                         skb_dst_set(skb, (struct dst_entry *)rt);
1455                         return 0;
1456                 }
1457         }
1458         rcu_read_unlock();
1459
1460         return dn_route_input_slow(skb);
1461 }
1462
1463 static int dn_rt_fill_info(struct sk_buff *skb, u32 pid, u32 seq,
1464                            int event, int nowait, unsigned int flags)
1465 {
1466         struct dn_route *rt = (struct dn_route *)skb_dst(skb);
1467         struct rtmsg *r;
1468         struct nlmsghdr *nlh;
1469         unsigned char *b = skb_tail_pointer(skb);
1470         long expires;
1471
1472         nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*r), flags);
1473         r = NLMSG_DATA(nlh);
1474         r->rtm_family = AF_DECnet;
1475         r->rtm_dst_len = 16;
1476         r->rtm_src_len = 0;
1477         r->rtm_tos = 0;
1478         r->rtm_table = RT_TABLE_MAIN;
1479         RTA_PUT_U32(skb, RTA_TABLE, RT_TABLE_MAIN);
1480         r->rtm_type = rt->rt_type;
1481         r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
1482         r->rtm_scope = RT_SCOPE_UNIVERSE;
1483         r->rtm_protocol = RTPROT_UNSPEC;
1484         if (rt->rt_flags & RTCF_NOTIFY)
1485                 r->rtm_flags |= RTM_F_NOTIFY;
1486         RTA_PUT(skb, RTA_DST, 2, &rt->rt_daddr);
1487         if (rt->fl.fld_src) {
1488                 r->rtm_src_len = 16;
1489                 RTA_PUT(skb, RTA_SRC, 2, &rt->fl.fld_src);
1490         }
1491         if (rt->u.dst.dev)
1492                 RTA_PUT(skb, RTA_OIF, sizeof(int), &rt->u.dst.dev->ifindex);
1493         /*
1494          * Note to self - change this if input routes reverse direction when
1495          * they deal only with inputs and not with replies like they do
1496          * currently.
1497          */
1498         RTA_PUT(skb, RTA_PREFSRC, 2, &rt->rt_local_src);
1499         if (rt->rt_daddr != rt->rt_gateway)
1500                 RTA_PUT(skb, RTA_GATEWAY, 2, &rt->rt_gateway);
1501         if (rtnetlink_put_metrics(skb, rt->u.dst.metrics) < 0)
1502                 goto rtattr_failure;
1503         expires = rt->u.dst.expires ? rt->u.dst.expires - jiffies : 0;
1504         if (rtnl_put_cacheinfo(skb, &rt->u.dst, 0, 0, 0, expires,
1505                                rt->u.dst.error) < 0)
1506                 goto rtattr_failure;
1507         if (rt->fl.iif)
1508                 RTA_PUT(skb, RTA_IIF, sizeof(int), &rt->fl.iif);
1509
1510         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1511         return skb->len;
1512
1513 nlmsg_failure:
1514 rtattr_failure:
1515         nlmsg_trim(skb, b);
1516         return -1;
1517 }
1518
1519 /*
1520  * This is called by both endnodes and routers now.
1521  */
1522 static int dn_cache_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh, void *arg)
1523 {
1524         struct net *net = sock_net(in_skb->sk);
1525         struct rtattr **rta = arg;
1526         struct rtmsg *rtm = NLMSG_DATA(nlh);
1527         struct dn_route *rt = NULL;
1528         struct dn_skb_cb *cb;
1529         int err;
1530         struct sk_buff *skb;
1531         struct flowi fl;
1532
1533         if (!net_eq(net, &init_net))
1534                 return -EINVAL;
1535
1536         memset(&fl, 0, sizeof(fl));
1537         fl.proto = DNPROTO_NSP;
1538
1539         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1540         if (skb == NULL)
1541                 return -ENOBUFS;
1542         skb_reset_mac_header(skb);
1543         cb = DN_SKB_CB(skb);
1544
1545         if (rta[RTA_SRC-1])
1546                 memcpy(&fl.fld_src, RTA_DATA(rta[RTA_SRC-1]), 2);
1547         if (rta[RTA_DST-1])
1548                 memcpy(&fl.fld_dst, RTA_DATA(rta[RTA_DST-1]), 2);
1549         if (rta[RTA_IIF-1])
1550                 memcpy(&fl.iif, RTA_DATA(rta[RTA_IIF-1]), sizeof(int));
1551
1552         if (fl.iif) {
1553                 struct net_device *dev;
1554                 if ((dev = dev_get_by_index(&init_net, fl.iif)) == NULL) {
1555                         kfree_skb(skb);
1556                         return -ENODEV;
1557                 }
1558                 if (!dev->dn_ptr) {
1559                         dev_put(dev);
1560                         kfree_skb(skb);
1561                         return -ENODEV;
1562                 }
1563                 skb->protocol = htons(ETH_P_DNA_RT);
1564                 skb->dev = dev;
1565                 cb->src = fl.fld_src;
1566                 cb->dst = fl.fld_dst;
1567                 local_bh_disable();
1568                 err = dn_route_input(skb);
1569                 local_bh_enable();
1570                 memset(cb, 0, sizeof(struct dn_skb_cb));
1571                 rt = (struct dn_route *)skb_dst(skb);
1572                 if (!err && -rt->u.dst.error)
1573                         err = rt->u.dst.error;
1574         } else {
1575                 int oif = 0;
1576                 if (rta[RTA_OIF - 1])
1577                         memcpy(&oif, RTA_DATA(rta[RTA_OIF - 1]), sizeof(int));
1578                 fl.oif = oif;
1579                 err = dn_route_output_key((struct dst_entry **)&rt, &fl, 0);
1580         }
1581
1582         if (skb->dev)
1583                 dev_put(skb->dev);
1584         skb->dev = NULL;
1585         if (err)
1586                 goto out_free;
1587         skb_dst_set(skb, &rt->u.dst);
1588         if (rtm->rtm_flags & RTM_F_NOTIFY)
1589                 rt->rt_flags |= RTCF_NOTIFY;
1590
1591         err = dn_rt_fill_info(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq, RTM_NEWROUTE, 0, 0);
1592
1593         if (err == 0)
1594                 goto out_free;
1595         if (err < 0) {
1596                 err = -EMSGSIZE;
1597                 goto out_free;
1598         }
1599
1600         return rtnl_unicast(skb, &init_net, NETLINK_CB(in_skb).pid);
1601
1602 out_free:
1603         kfree_skb(skb);
1604         return err;
1605 }
1606
1607 /*
1608  * For routers, this is called from dn_fib_dump, but for endnodes its
1609  * called directly from the rtnetlink dispatch table.
1610  */
1611 int dn_cache_dump(struct sk_buff *skb, struct netlink_callback *cb)
1612 {
1613         struct net *net = sock_net(skb->sk);
1614         struct dn_route *rt;
1615         int h, s_h;
1616         int idx, s_idx;
1617
1618         if (!net_eq(net, &init_net))
1619                 return 0;
1620
1621         if (NLMSG_PAYLOAD(cb->nlh, 0) < sizeof(struct rtmsg))
1622                 return -EINVAL;
1623         if (!(((struct rtmsg *)NLMSG_DATA(cb->nlh))->rtm_flags&RTM_F_CLONED))
1624                 return 0;
1625
1626         s_h = cb->args[0];
1627         s_idx = idx = cb->args[1];
1628         for(h = 0; h <= dn_rt_hash_mask; h++) {
1629                 if (h < s_h)
1630                         continue;
1631                 if (h > s_h)
1632                         s_idx = 0;
1633                 rcu_read_lock_bh();
1634                 for(rt = rcu_dereference_bh(dn_rt_hash_table[h].chain), idx = 0;
1635                         rt;
1636                         rt = rcu_dereference_bh(rt->u.dst.dn_next), idx++) {
1637                         if (idx < s_idx)
1638                                 continue;
1639                         skb_dst_set(skb, dst_clone(&rt->u.dst));
1640                         if (dn_rt_fill_info(skb, NETLINK_CB(cb->skb).pid,
1641                                         cb->nlh->nlmsg_seq, RTM_NEWROUTE,
1642                                         1, NLM_F_MULTI) <= 0) {
1643                                 skb_dst_drop(skb);
1644                                 rcu_read_unlock_bh();
1645                                 goto done;
1646                         }
1647                         skb_dst_drop(skb);
1648                 }
1649                 rcu_read_unlock_bh();
1650         }
1651
1652 done:
1653         cb->args[0] = h;
1654         cb->args[1] = idx;
1655         return skb->len;
1656 }
1657
1658 #ifdef CONFIG_PROC_FS
1659 struct dn_rt_cache_iter_state {
1660         int bucket;
1661 };
1662
1663 static struct dn_route *dn_rt_cache_get_first(struct seq_file *seq)
1664 {
1665         struct dn_route *rt = NULL;
1666         struct dn_rt_cache_iter_state *s = seq->private;
1667
1668         for(s->bucket = dn_rt_hash_mask; s->bucket >= 0; --s->bucket) {
1669                 rcu_read_lock_bh();
1670                 rt = rcu_dereference_bh(dn_rt_hash_table[s->bucket].chain);
1671                 if (rt)
1672                         break;
1673                 rcu_read_unlock_bh();
1674         }
1675         return rt;
1676 }
1677
1678 static struct dn_route *dn_rt_cache_get_next(struct seq_file *seq, struct dn_route *rt)
1679 {
1680         struct dn_rt_cache_iter_state *s = seq->private;
1681
1682         rt = rt->u.dst.dn_next;
1683         while(!rt) {
1684                 rcu_read_unlock_bh();
1685                 if (--s->bucket < 0)
1686                         break;
1687                 rcu_read_lock_bh();
1688                 rt = dn_rt_hash_table[s->bucket].chain;
1689         }
1690         return rcu_dereference_bh(rt);
1691 }
1692
1693 static void *dn_rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
1694 {
1695         struct dn_route *rt = dn_rt_cache_get_first(seq);
1696
1697         if (rt) {
1698                 while(*pos && (rt = dn_rt_cache_get_next(seq, rt)))
1699                         --*pos;
1700         }
1701         return *pos ? NULL : rt;
1702 }
1703
1704 static void *dn_rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1705 {
1706         struct dn_route *rt = dn_rt_cache_get_next(seq, v);
1707         ++*pos;
1708         return rt;
1709 }
1710
1711 static void dn_rt_cache_seq_stop(struct seq_file *seq, void *v)
1712 {
1713         if (v)
1714                 rcu_read_unlock_bh();
1715 }
1716
1717 static int dn_rt_cache_seq_show(struct seq_file *seq, void *v)
1718 {
1719         struct dn_route *rt = v;
1720         char buf1[DN_ASCBUF_LEN], buf2[DN_ASCBUF_LEN];
1721
1722         seq_printf(seq, "%-8s %-7s %-7s %04d %04d %04d\n",
1723                         rt->u.dst.dev ? rt->u.dst.dev->name : "*",
1724                         dn_addr2asc(le16_to_cpu(rt->rt_daddr), buf1),
1725                         dn_addr2asc(le16_to_cpu(rt->rt_saddr), buf2),
1726                         atomic_read(&rt->u.dst.__refcnt),
1727                         rt->u.dst.__use,
1728                         (int) dst_metric(&rt->u.dst, RTAX_RTT));
1729         return 0;
1730 }
1731
1732 static const struct seq_operations dn_rt_cache_seq_ops = {
1733         .start  = dn_rt_cache_seq_start,
1734         .next   = dn_rt_cache_seq_next,
1735         .stop   = dn_rt_cache_seq_stop,
1736         .show   = dn_rt_cache_seq_show,
1737 };
1738
1739 static int dn_rt_cache_seq_open(struct inode *inode, struct file *file)
1740 {
1741         return seq_open_private(file, &dn_rt_cache_seq_ops,
1742                         sizeof(struct dn_rt_cache_iter_state));
1743 }
1744
1745 static const struct file_operations dn_rt_cache_seq_fops = {
1746         .owner   = THIS_MODULE,
1747         .open    = dn_rt_cache_seq_open,
1748         .read    = seq_read,
1749         .llseek  = seq_lseek,
1750         .release = seq_release_private,
1751 };
1752
1753 #endif /* CONFIG_PROC_FS */
1754
1755 void __init dn_route_init(void)
1756 {
1757         int i, goal, order;
1758
1759         dn_dst_ops.kmem_cachep =
1760                 kmem_cache_create("dn_dst_cache", sizeof(struct dn_route), 0,
1761                                   SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1762         setup_timer(&dn_route_timer, dn_dst_check_expire, 0);
1763         dn_route_timer.expires = jiffies + decnet_dst_gc_interval * HZ;
1764         add_timer(&dn_route_timer);
1765
1766         goal = totalram_pages >> (26 - PAGE_SHIFT);
1767
1768         for(order = 0; (1UL << order) < goal; order++)
1769                 /* NOTHING */;
1770
1771         /*
1772          * Only want 1024 entries max, since the table is very, very unlikely
1773          * to be larger than that.
1774          */
1775         while(order && ((((1UL << order) * PAGE_SIZE) /
1776                                 sizeof(struct dn_rt_hash_bucket)) >= 2048))
1777                 order--;
1778
1779         do {
1780                 dn_rt_hash_mask = (1UL << order) * PAGE_SIZE /
1781                         sizeof(struct dn_rt_hash_bucket);
1782                 while(dn_rt_hash_mask & (dn_rt_hash_mask - 1))
1783                         dn_rt_hash_mask--;
1784                 dn_rt_hash_table = (struct dn_rt_hash_bucket *)
1785                         __get_free_pages(GFP_ATOMIC, order);
1786         } while (dn_rt_hash_table == NULL && --order > 0);
1787
1788         if (!dn_rt_hash_table)
1789                 panic("Failed to allocate DECnet route cache hash table\n");
1790
1791         printk(KERN_INFO
1792                 "DECnet: Routing cache hash table of %u buckets, %ldKbytes\n",
1793                 dn_rt_hash_mask,
1794                 (long)(dn_rt_hash_mask*sizeof(struct dn_rt_hash_bucket))/1024);
1795
1796         dn_rt_hash_mask--;
1797         for(i = 0; i <= dn_rt_hash_mask; i++) {
1798                 spin_lock_init(&dn_rt_hash_table[i].lock);
1799                 dn_rt_hash_table[i].chain = NULL;
1800         }
1801
1802         dn_dst_ops.gc_thresh = (dn_rt_hash_mask + 1);
1803
1804         proc_net_fops_create(&init_net, "decnet_cache", S_IRUGO, &dn_rt_cache_seq_fops);
1805
1806 #ifdef CONFIG_DECNET_ROUTER
1807         rtnl_register(PF_DECnet, RTM_GETROUTE, dn_cache_getroute, dn_fib_dump);
1808 #else
1809         rtnl_register(PF_DECnet, RTM_GETROUTE, dn_cache_getroute,
1810                       dn_cache_dump);
1811 #endif
1812 }
1813
1814 void __exit dn_route_cleanup(void)
1815 {
1816         del_timer(&dn_route_timer);
1817         dn_run_flush(0);
1818
1819         proc_net_remove(&init_net, "decnet_cache");
1820 }
1821