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 __rcu *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 unsigned int dn_dst_default_advmss(const struct dst_entry *dst);
 114 static unsigned int dn_dst_default_mtu(const struct dst_entry *dst);
 115 static void dn_dst_destroy(struct dst_entry *);
 116 static struct dst_entry *dn_dst_negative_advice(struct dst_entry *);
 117 static void dn_dst_link_failure(struct sk_buff *);
 118 static void dn_dst_update_pmtu(struct dst_entry *dst, u32 mtu);
 119 static int dn_route_input(struct sk_buff *);
 120 static void dn_run_flush(unsigned long dummy);
 121
 122 static struct dn_rt_hash_bucket *dn_rt_hash_table;
 123 static unsigned dn_rt_hash_mask;
 124
 125 static struct timer_list dn_route_timer;
 126 static DEFINE_TIMER(dn_rt_flush_timer, dn_run_flush, 0, 0);
 127 int decnet_dst_gc_interval = 2;
 128
 129 static struct dst_ops dn_dst_ops = {
 130         .family =               PF_DECnet,
 131         .protocol =             cpu_to_be16(ETH_P_DNA_RT),
 132         .gc_thresh =            128,
 133         .gc =                   dn_dst_gc,
 134         .check =                dn_dst_check,
 135         .default_advmss =       dn_dst_default_advmss,
 136         .default_mtu =          dn_dst_default_mtu,
 137         .cow_metrics =          dst_cow_metrics_generic,
 138         .destroy =              dn_dst_destroy,
 139         .negative_advice =      dn_dst_negative_advice,
 140         .link_failure =         dn_dst_link_failure,
 141         .update_pmtu =          dn_dst_update_pmtu,
 142 };
 143
 144 static void dn_dst_destroy(struct dst_entry *dst)
 145 {
 146         dst_destroy_metrics_generic(dst);
 147 }
 148
 149 static __inline__ unsigned dn_hash(__le16 src, __le16 dst)
 150 {
 151         __u16 tmp = (__u16 __force)(src ^ dst);
 152         tmp ^= (tmp >> 3);
 153         tmp ^= (tmp >> 5);
 154         tmp ^= (tmp >> 10);
 155         return dn_rt_hash_mask & (unsigned)tmp;
 156 }
 157
 158 static inline void dnrt_free(struct dn_route *rt)
 159 {
 160         call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
 161 }
 162
 163 static inline void dnrt_drop(struct dn_route *rt)
 164 {
 165         dst_release(&rt->dst);
 166         call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
 167 }
 168
 169 static void dn_dst_check_expire(unsigned long dummy)
 170 {
 171         int i;
 172         struct dn_route *rt;
 173         struct dn_route __rcu **rtp;
 174         unsigned long now = jiffies;
 175         unsigned long expire = 120 * HZ;
 176
 177         for (i = 0; i <= dn_rt_hash_mask; i++) {
 178                 rtp = &dn_rt_hash_table[i].chain;
 179
 180                 spin_lock(&dn_rt_hash_table[i].lock);
 181                 while ((rt = rcu_dereference_protected(*rtp,
 182                                                 lockdep_is_held(&dn_rt_hash_table[i].lock))) != NULL) {
 183                         if (atomic_read(&rt->dst.__refcnt) ||
 184                                         (now - rt->dst.lastuse) < expire) {
 185                                 rtp = &rt->dst.dn_next;
 186                                 continue;
 187                         }
 188                         *rtp = rt->dst.dn_next;
 189                         rt->dst.dn_next = NULL;
 190                         dnrt_free(rt);
 191                 }
 192                 spin_unlock(&dn_rt_hash_table[i].lock);
 193
 194                 if ((jiffies - now) > 0)
 195                         break;
 196         }
 197
 198         mod_timer(&dn_route_timer, now + decnet_dst_gc_interval * HZ);
 199 }
 200
 201 static int dn_dst_gc(struct dst_ops *ops)
 202 {
 203         struct dn_route *rt;
 204         struct dn_route __rcu **rtp;
 205         int i;
 206         unsigned long now = jiffies;
 207         unsigned long expire = 10 * HZ;
 208
 209         for (i = 0; i <= dn_rt_hash_mask; i++) {
 210
 211                 spin_lock_bh(&dn_rt_hash_table[i].lock);
 212                 rtp = &dn_rt_hash_table[i].chain;
 213
 214                 while ((rt = rcu_dereference_protected(*rtp,
 215                                                 lockdep_is_held(&dn_rt_hash_table[i].lock))) != NULL) {
 216                         if (atomic_read(&rt->dst.__refcnt) ||
 217                                         (now - rt->dst.lastuse) < expire) {
 218                                 rtp = &rt->dst.dn_next;
 219                                 continue;
 220                         }
 221                         *rtp = rt->dst.dn_next;
 222                         rt->dst.dn_next = NULL;
 223                         dnrt_drop(rt);
 224                         break;
 225                 }
 226                 spin_unlock_bh(&dn_rt_hash_table[i].lock);
 227         }
 228
 229         return 0;
 230 }
 231
 232 /*
 233  * The decnet standards don't impose a particular minimum mtu, what they
 234  * do insist on is that the routing layer accepts a datagram of at least
 235  * 230 bytes long. Here we have to subtract the routing header length from
 236  * 230 to get the minimum acceptable mtu. If there is no neighbour, then we
 237  * assume the worst and use a long header size.
 238  *
 239  * We update both the mtu and the advertised mss (i.e. the segment size we
 240  * advertise to the other end).
 241  */
 242 static void dn_dst_update_pmtu(struct dst_entry *dst, u32 mtu)
 243 {
 244         u32 min_mtu = 230;
 245         struct dn_dev *dn = dst->neighbour ?
 246                             rcu_dereference_raw(dst->neighbour->dev->dn_ptr) : NULL;
 247
 248         if (dn && dn->use_long == 0)
 249                 min_mtu -= 6;
 250         else
 251                 min_mtu -= 21;
 252
 253         if (dst_metric(dst, RTAX_MTU) > mtu && mtu >= min_mtu) {
 254                 if (!(dst_metric_locked(dst, RTAX_MTU))) {
 255                         dst_metric_set(dst, RTAX_MTU, mtu);
 256                         dst_set_expires(dst, dn_rt_mtu_expires);
 257                 }
 258                 if (!(dst_metric_locked(dst, RTAX_ADVMSS))) {
 259                         u32 mss = mtu - DN_MAX_NSP_DATA_HEADER;
 260                         u32 existing_mss = dst_metric_raw(dst, RTAX_ADVMSS);
 261                         if (!existing_mss || existing_mss > mss)
 262                                 dst_metric_set(dst, RTAX_ADVMSS, mss);
 263                 }
 264         }
 265 }
 266
 267 /*
 268  * When a route has been marked obsolete. (e.g. routing cache flush)
 269  */
 270 static struct dst_entry *dn_dst_check(struct dst_entry *dst, __u32 cookie)
 271 {
 272         return NULL;
 273 }
 274
 275 static struct dst_entry *dn_dst_negative_advice(struct dst_entry *dst)
 276 {
 277         dst_release(dst);
 278         return NULL;
 279 }
 280
 281 static void dn_dst_link_failure(struct sk_buff *skb)
 282 {
 283 }
 284
 285 static inline int compare_keys(struct flowidn *fl1, struct flowidn *fl2)
 286 {
 287         return ((fl1->daddr ^ fl2->daddr) |
 288                 (fl1->saddr ^ fl2->saddr) |
 289                 (fl1->flowidn_mark ^ fl2->flowidn_mark) |
 290                 (fl1->flowidn_scope ^ fl2->flowidn_scope) |
 291                 (fl1->flowidn_oif ^ fl2->flowidn_oif) |
 292                 (fl1->flowidn_iif ^ fl2->flowidn_iif)) == 0;
 293 }
 294
 295 static int dn_insert_route(struct dn_route *rt, unsigned hash, struct dn_route **rp)
 296 {
 297         struct dn_route *rth;
 298         struct dn_route __rcu **rthp;
 299         unsigned long now = jiffies;
 300
 301         rthp = &dn_rt_hash_table[hash].chain;
 302
 303         spin_lock_bh(&dn_rt_hash_table[hash].lock);
 304         while ((rth = rcu_dereference_protected(*rthp,
 305                                                 lockdep_is_held(&dn_rt_hash_table[hash].lock))) != NULL) {
 306                 if (compare_keys(&rth->fld, &rt->fld)) {
 307                         /* Put it first */
 308                         *rthp = rth->dst.dn_next;
 309                         rcu_assign_pointer(rth->dst.dn_next,
 310                                            dn_rt_hash_table[hash].chain);
 311                         rcu_assign_pointer(dn_rt_hash_table[hash].chain, rth);
 312
 313                         dst_use(&rth->dst, now);
 314                         spin_unlock_bh(&dn_rt_hash_table[hash].lock);
 315
 316                         dnrt_drop(rt);
 317                         *rp = rth;
 318                         return 0;
 319                 }
 320                 rthp = &rth->dst.dn_next;
 321         }
 322
 323         rcu_assign_pointer(rt->dst.dn_next, dn_rt_hash_table[hash].chain);
 324         rcu_assign_pointer(dn_rt_hash_table[hash].chain, rt);
 325
 326         dst_use(&rt->dst, now);
 327         spin_unlock_bh(&dn_rt_hash_table[hash].lock);
 328         *rp = rt;
 329         return 0;
 330 }
 331
 332 static void dn_run_flush(unsigned long dummy)
 333 {
 334         int i;
 335         struct dn_route *rt, *next;
 336
 337         for (i = 0; i < dn_rt_hash_mask; i++) {
 338                 spin_lock_bh(&dn_rt_hash_table[i].lock);
 339
 340                 if ((rt = xchg((struct dn_route **)&dn_rt_hash_table[i].chain, NULL)) == NULL)
 341                         goto nothing_to_declare;
 342
 343                 for(; rt; rt = next) {
 344                         next = rcu_dereference_raw(rt->dst.dn_next);
 345                         RCU_INIT_POINTER(rt->dst.dn_next, NULL);
 346                         dst_free((struct dst_entry *)rt);
 347                 }
 348
 349 nothing_to_declare:
 350                 spin_unlock_bh(&dn_rt_hash_table[i].lock);
 351         }
 352 }
 353
 354 static DEFINE_SPINLOCK(dn_rt_flush_lock);
 355
 356 void dn_rt_cache_flush(int delay)
 357 {
 358         unsigned long now = jiffies;
 359         int user_mode = !in_interrupt();
 360
 361         if (delay < 0)
 362                 delay = dn_rt_min_delay;
 363
 364         spin_lock_bh(&dn_rt_flush_lock);
 365
 366         if (del_timer(&dn_rt_flush_timer) && delay > 0 && dn_rt_deadline) {
 367                 long tmo = (long)(dn_rt_deadline - now);
 368
 369                 if (user_mode && tmo < dn_rt_max_delay - dn_rt_min_delay)
 370                         tmo = 0;
 371
 372                 if (delay > tmo)
 373                         delay = tmo;
 374         }
 375
 376         if (delay <= 0) {
 377                 spin_unlock_bh(&dn_rt_flush_lock);
 378                 dn_run_flush(0);
 379                 return;
 380         }
 381
 382         if (dn_rt_deadline == 0)
 383                 dn_rt_deadline = now + dn_rt_max_delay;
 384
 385         dn_rt_flush_timer.expires = now + delay;
 386         add_timer(&dn_rt_flush_timer);
 387         spin_unlock_bh(&dn_rt_flush_lock);
 388 }
 389
 390 /**
 391  * dn_return_short - Return a short packet to its sender
 392  * @skb: The packet to return
 393  *
 394  */
 395 static int dn_return_short(struct sk_buff *skb)
 396 {
 397         struct dn_skb_cb *cb;
 398         unsigned char *ptr;
 399         __le16 *src;
 400         __le16 *dst;
 401
 402         /* Add back headers */
 403         skb_push(skb, skb->data - skb_network_header(skb));
 404
 405         if ((skb = skb_unshare(skb, GFP_ATOMIC)) == NULL)
 406                 return NET_RX_DROP;
 407
 408         cb = DN_SKB_CB(skb);
 409         /* Skip packet length and point to flags */
 410         ptr = skb->data + 2;
 411         *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
 412
 413         dst = (__le16 *)ptr;
 414         ptr += 2;
 415         src = (__le16 *)ptr;
 416         ptr += 2;
 417         *ptr = 0; /* Zero hop count */
 418
 419         swap(*src, *dst);
 420
 421         skb->pkt_type = PACKET_OUTGOING;
 422         dn_rt_finish_output(skb, NULL, NULL);
 423         return NET_RX_SUCCESS;
 424 }
 425
 426 /**
 427  * dn_return_long - Return a long packet to its sender
 428  * @skb: The long format packet to return
 429  *
 430  */
 431 static int dn_return_long(struct sk_buff *skb)
 432 {
 433         struct dn_skb_cb *cb;
 434         unsigned char *ptr;
 435         unsigned char *src_addr, *dst_addr;
 436         unsigned char tmp[ETH_ALEN];
 437
 438         /* Add back all headers */
 439         skb_push(skb, skb->data - skb_network_header(skb));
 440
 441         if ((skb = skb_unshare(skb, GFP_ATOMIC)) == NULL)
 442                 return NET_RX_DROP;
 443
 444         cb = DN_SKB_CB(skb);
 445         /* Ignore packet length and point to flags */
 446         ptr = skb->data + 2;
 447
 448         /* Skip padding */
 449         if (*ptr & DN_RT_F_PF) {
 450                 char padlen = (*ptr & ~DN_RT_F_PF);
 451                 ptr += padlen;
 452         }
 453
 454         *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
 455         ptr += 2;
 456         dst_addr = ptr;
 457         ptr += 8;
 458         src_addr = ptr;
 459         ptr += 6;
 460         *ptr = 0; /* Zero hop count */
 461
 462         /* Swap source and destination */
 463         memcpy(tmp, src_addr, ETH_ALEN);
 464         memcpy(src_addr, dst_addr, ETH_ALEN);
 465         memcpy(dst_addr, tmp, ETH_ALEN);
 466
 467         skb->pkt_type = PACKET_OUTGOING;
 468         dn_rt_finish_output(skb, dst_addr, src_addr);
 469         return NET_RX_SUCCESS;
 470 }
 471
 472 /**
 473  * dn_route_rx_packet - Try and find a route for an incoming packet
 474  * @skb: The packet to find a route for
 475  *
 476  * Returns: result of input function if route is found, error code otherwise
 477  */
 478 static int dn_route_rx_packet(struct sk_buff *skb)
 479 {
 480         struct dn_skb_cb *cb;
 481         int err;
 482
 483         if ((err = dn_route_input(skb)) == 0)
 484                 return dst_input(skb);
 485
 486         cb = DN_SKB_CB(skb);
 487         if (decnet_debug_level & 4) {
 488                 char *devname = skb->dev ? skb->dev->name : "???";
 489
 490                 printk(KERN_DEBUG
 491                         "DECnet: dn_route_rx_packet: rt_flags=0x%02x dev=%s len=%d src=0x%04hx dst=0x%04hx err=%d type=%d\n",
 492                         (int)cb->rt_flags, devname, skb->len,
 493                         le16_to_cpu(cb->src), le16_to_cpu(cb->dst),
 494                         err, skb->pkt_type);
 495         }
 496
 497         if ((skb->pkt_type == PACKET_HOST) && (cb->rt_flags & DN_RT_F_RQR)) {
 498                 switch(cb->rt_flags & DN_RT_PKT_MSK) {
 499                         case DN_RT_PKT_SHORT:
 500                                 return dn_return_short(skb);
 501                         case DN_RT_PKT_LONG:
 502                                 return dn_return_long(skb);
 503                 }
 504         }
 505
 506         kfree_skb(skb);
 507         return NET_RX_DROP;
 508 }
 509
 510 static int dn_route_rx_long(struct sk_buff *skb)
 511 {
 512         struct dn_skb_cb *cb = DN_SKB_CB(skb);
 513         unsigned char *ptr = skb->data;
 514
 515         if (!pskb_may_pull(skb, 21)) /* 20 for long header, 1 for shortest nsp */
 516                 goto drop_it;
 517
 518         skb_pull(skb, 20);
 519         skb_reset_transport_header(skb);
 520
 521         /* Destination info */
 522         ptr += 2;
 523         cb->dst = dn_eth2dn(ptr);
 524         if (memcmp(ptr, dn_hiord_addr, 4) != 0)
 525                 goto drop_it;
 526         ptr += 6;
 527
 528
 529         /* Source info */
 530         ptr += 2;
 531         cb->src = dn_eth2dn(ptr);
 532         if (memcmp(ptr, dn_hiord_addr, 4) != 0)
 533                 goto drop_it;
 534         ptr += 6;
 535         /* Other junk */
 536         ptr++;
 537         cb->hops = *ptr++; /* Visit Count */
 538
 539         return NF_HOOK(NFPROTO_DECNET, NF_DN_PRE_ROUTING, skb, skb->dev, NULL,
 540                        dn_route_rx_packet);
 541
 542 drop_it:
 543         kfree_skb(skb);
 544         return NET_RX_DROP;
 545 }
 546
 547
 548
 549 static int dn_route_rx_short(struct sk_buff *skb)
 550 {
 551         struct dn_skb_cb *cb = DN_SKB_CB(skb);
 552         unsigned char *ptr = skb->data;
 553
 554         if (!pskb_may_pull(skb, 6)) /* 5 for short header + 1 for shortest nsp */
 555                 goto drop_it;
 556
 557         skb_pull(skb, 5);
 558         skb_reset_transport_header(skb);
 559
 560         cb->dst = *(__le16 *)ptr;
 561         ptr += 2;
 562         cb->src = *(__le16 *)ptr;
 563         ptr += 2;
 564         cb->hops = *ptr & 0x3f;
 565
 566         return NF_HOOK(NFPROTO_DECNET, NF_DN_PRE_ROUTING, skb, skb->dev, NULL,
 567                        dn_route_rx_packet);
 568
 569 drop_it:
 570         kfree_skb(skb);
 571         return NET_RX_DROP;
 572 }
 573
 574 static int dn_route_discard(struct sk_buff *skb)
 575 {
 576         /*
 577          * I know we drop the packet here, but thats considered success in
 578          * this case
 579          */
 580         kfree_skb(skb);
 581         return NET_RX_SUCCESS;
 582 }
 583
 584 static int dn_route_ptp_hello(struct sk_buff *skb)
 585 {
 586         dn_dev_hello(skb);
 587         dn_neigh_pointopoint_hello(skb);
 588         return NET_RX_SUCCESS;
 589 }
 590
 591 int dn_route_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
 592 {
 593         struct dn_skb_cb *cb;
 594         unsigned char flags = 0;
 595         __u16 len = le16_to_cpu(*(__le16 *)skb->data);
 596         struct dn_dev *dn = rcu_dereference(dev->dn_ptr);
 597         unsigned char padlen = 0;
 598
 599         if (!net_eq(dev_net(dev), &init_net))
 600                 goto dump_it;
 601
 602         if (dn == NULL)
 603                 goto dump_it;
 604
 605         if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
 606                 goto out;
 607
 608         if (!pskb_may_pull(skb, 3))
 609                 goto dump_it;
 610
 611         skb_pull(skb, 2);
 612
 613         if (len > skb->len)
 614                 goto dump_it;
 615
 616         skb_trim(skb, len);
 617
 618         flags = *skb->data;
 619
 620         cb = DN_SKB_CB(skb);
 621         cb->stamp = jiffies;
 622         cb->iif = dev->ifindex;
 623
 624         /*
 625          * If we have padding, remove it.
 626          */
 627         if (flags & DN_RT_F_PF) {
 628                 padlen = flags & ~DN_RT_F_PF;
 629                 if (!pskb_may_pull(skb, padlen + 1))
 630                         goto dump_it;
 631                 skb_pull(skb, padlen);
 632                 flags = *skb->data;
 633         }
 634
 635         skb_reset_network_header(skb);
 636
 637         /*
 638          * Weed out future version DECnet
 639          */
 640         if (flags & DN_RT_F_VER)
 641                 goto dump_it;
 642
 643         cb->rt_flags = flags;
 644
 645         if (decnet_debug_level & 1)
 646                 printk(KERN_DEBUG
 647                         "dn_route_rcv: got 0x%02x from %s [%d %d %d]\n",
 648                         (int)flags, (dev) ? dev->name : "???", len, skb->len,
 649                         padlen);
 650
 651         if (flags & DN_RT_PKT_CNTL) {
 652                 if (unlikely(skb_linearize(skb)))
 653                         goto dump_it;
 654
 655                 switch(flags & DN_RT_CNTL_MSK) {
 656                         case DN_RT_PKT_INIT:
 657                                 dn_dev_init_pkt(skb);
 658                                 break;
 659                         case DN_RT_PKT_VERI:
 660                                 dn_dev_veri_pkt(skb);
 661                                 break;
 662                 }
 663
 664                 if (dn->parms.state != DN_DEV_S_RU)
 665                         goto dump_it;
 666
 667                 switch(flags & DN_RT_CNTL_MSK) {
 668                         case DN_RT_PKT_HELO:
 669                                 return NF_HOOK(NFPROTO_DECNET, NF_DN_HELLO,
 670                                                skb, skb->dev, NULL,
 671                                                dn_route_ptp_hello);
 672
 673                         case DN_RT_PKT_L1RT:
 674                         case DN_RT_PKT_L2RT:
 675                                 return NF_HOOK(NFPROTO_DECNET, NF_DN_ROUTE,
 676                                                skb, skb->dev, NULL,
 677                                                dn_route_discard);
 678                         case DN_RT_PKT_ERTH:
 679                                 return NF_HOOK(NFPROTO_DECNET, NF_DN_HELLO,
 680                                                skb, skb->dev, NULL,
 681                                                dn_neigh_router_hello);
 682
 683                         case DN_RT_PKT_EEDH:
 684                                 return NF_HOOK(NFPROTO_DECNET, NF_DN_HELLO,
 685                                                skb, skb->dev, NULL,
 686                                                dn_neigh_endnode_hello);
 687                 }
 688         } else {
 689                 if (dn->parms.state != DN_DEV_S_RU)
 690                         goto dump_it;
 691
 692                 skb_pull(skb, 1); /* Pull flags */
 693
 694                 switch(flags & DN_RT_PKT_MSK) {
 695                         case DN_RT_PKT_LONG:
 696                                 return dn_route_rx_long(skb);
 697                         case DN_RT_PKT_SHORT:
 698                                 return dn_route_rx_short(skb);
 699                 }
 700         }
 701
 702 dump_it:
 703         kfree_skb(skb);
 704 out:
 705         return NET_RX_DROP;
 706 }
 707
 708 static int dn_output(struct sk_buff *skb)
 709 {
 710         struct dst_entry *dst = skb_dst(skb);
 711         struct dn_route *rt = (struct dn_route *)dst;
 712         struct net_device *dev = dst->dev;
 713         struct dn_skb_cb *cb = DN_SKB_CB(skb);
 714         struct neighbour *neigh;
 715
 716         int err = -EINVAL;
 717
 718         if ((neigh = dst->neighbour) == NULL)
 719                 goto error;
 720
 721         skb->dev = dev;
 722
 723         cb->src = rt->rt_saddr;
 724         cb->dst = rt->rt_daddr;
 725
 726         /*
 727          * Always set the Intra-Ethernet bit on all outgoing packets
 728          * originated on this node. Only valid flag from upper layers
 729          * is return-to-sender-requested. Set hop count to 0 too.
 730          */
 731         cb->rt_flags &= ~DN_RT_F_RQR;
 732         cb->rt_flags |= DN_RT_F_IE;
 733         cb->hops = 0;
 734
 735         return NF_HOOK(NFPROTO_DECNET, NF_DN_LOCAL_OUT, skb, NULL, dev,
 736                        neigh->output);
 737
 738 error:
 739         if (net_ratelimit())
 740                 printk(KERN_DEBUG "dn_output: This should not happen\n");
 741
 742         kfree_skb(skb);
 743
 744         return err;
 745 }
 746
 747 static int dn_forward(struct sk_buff *skb)
 748 {
 749         struct dn_skb_cb *cb = DN_SKB_CB(skb);
 750         struct dst_entry *dst = skb_dst(skb);
 751         struct dn_dev *dn_db = rcu_dereference(dst->dev->dn_ptr);
 752         struct dn_route *rt;
 753         struct neighbour *neigh = dst->neighbour;
 754         int header_len;
 755 #ifdef CONFIG_NETFILTER
 756         struct net_device *dev = skb->dev;
 757 #endif
 758
 759         if (skb->pkt_type != PACKET_HOST)
 760                 goto drop;
 761
 762         /* Ensure that we have enough space for headers */
 763         rt = (struct dn_route *)skb_dst(skb);
 764         header_len = dn_db->use_long ? 21 : 6;
 765         if (skb_cow(skb, LL_RESERVED_SPACE(rt->dst.dev)+header_len))
 766                 goto drop;
 767
 768         /*
 769          * Hop count exceeded.
 770          */
 771         if (++cb->hops > 30)
 772                 goto drop;
 773
 774         skb->dev = rt->dst.dev;
 775
 776         /*
 777          * If packet goes out same interface it came in on, then set
 778          * the Intra-Ethernet bit. This has no effect for short
 779          * packets, so we don't need to test for them here.
 780          */
 781         cb->rt_flags &= ~DN_RT_F_IE;
 782         if (rt->rt_flags & RTCF_DOREDIRECT)
 783                 cb->rt_flags |= DN_RT_F_IE;
 784
 785         return NF_HOOK(NFPROTO_DECNET, NF_DN_FORWARD, skb, dev, skb->dev,
 786                        neigh->output);
 787
 788 drop:
 789         kfree_skb(skb);
 790         return NET_RX_DROP;
 791 }
 792
 793 /*
 794  * Used to catch bugs. This should never normally get
 795  * called.
 796  */
 797 static int dn_rt_bug(struct sk_buff *skb)
 798 {
 799         if (net_ratelimit()) {
 800                 struct dn_skb_cb *cb = DN_SKB_CB(skb);
 801
 802                 printk(KERN_DEBUG "dn_rt_bug: skb from:%04x to:%04x\n",
 803                                 le16_to_cpu(cb->src), le16_to_cpu(cb->dst));
 804         }
 805
 806         kfree_skb(skb);
 807
 808         return NET_RX_DROP;
 809 }
 810
 811 static unsigned int dn_dst_default_advmss(const struct dst_entry *dst)
 812 {
 813         return dn_mss_from_pmtu(dst->dev, dst_mtu(dst));
 814 }
 815
 816 static unsigned int dn_dst_default_mtu(const struct dst_entry *dst)
 817 {
 818         return dst->dev->mtu;
 819 }
 820
 821 static int dn_rt_set_next_hop(struct dn_route *rt, struct dn_fib_res *res)
 822 {
 823         struct dn_fib_info *fi = res->fi;
 824         struct net_device *dev = rt->dst.dev;
 825         unsigned int mss_metric;
 826         struct neighbour *n;
 827
 828         if (fi) {
 829                 if (DN_FIB_RES_GW(*res) &&
 830                     DN_FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
 831                         rt->rt_gateway = DN_FIB_RES_GW(*res);
 832                 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
 833         }
 834         rt->rt_type = res->type;
 835
 836         if (dev != NULL && rt->dst.neighbour == NULL) {
 837                 n = __neigh_lookup_errno(&dn_neigh_table, &rt->rt_gateway, dev);
 838                 if (IS_ERR(n))
 839                         return PTR_ERR(n);
 840                 rt->dst.neighbour = n;
 841         }
 842
 843         if (dst_metric(&rt->dst, RTAX_MTU) > rt->dst.dev->mtu)
 844                 dst_metric_set(&rt->dst, RTAX_MTU, rt->dst.dev->mtu);
 845         mss_metric = dst_metric_raw(&rt->dst, RTAX_ADVMSS);
 846         if (mss_metric) {
 847                 unsigned int mss = dn_mss_from_pmtu(dev, dst_mtu(&rt->dst));
 848                 if (mss_metric > mss)
 849                         dst_metric_set(&rt->dst, RTAX_ADVMSS, mss);
 850         }
 851         return 0;
 852 }
 853
 854 static inline int dn_match_addr(__le16 addr1, __le16 addr2)
 855 {
 856         __u16 tmp = le16_to_cpu(addr1) ^ le16_to_cpu(addr2);
 857         int match = 16;
 858         while(tmp) {
 859                 tmp >>= 1;
 860                 match--;
 861         }
 862         return match;
 863 }
 864
 865 static __le16 dnet_select_source(const struct net_device *dev, __le16 daddr, int scope)
 866 {
 867         __le16 saddr = 0;
 868         struct dn_dev *dn_db;
 869         struct dn_ifaddr *ifa;
 870         int best_match = 0;
 871         int ret;
 872
 873         rcu_read_lock();
 874         dn_db = rcu_dereference(dev->dn_ptr);
 875         for (ifa = rcu_dereference(dn_db->ifa_list);
 876              ifa != NULL;
 877              ifa = rcu_dereference(ifa->ifa_next)) {
 878                 if (ifa->ifa_scope > scope)
 879                         continue;
 880                 if (!daddr) {
 881                         saddr = ifa->ifa_local;
 882                         break;
 883                 }
 884                 ret = dn_match_addr(daddr, ifa->ifa_local);
 885                 if (ret > best_match)
 886                         saddr = ifa->ifa_local;
 887                 if (best_match == 0)
 888                         saddr = ifa->ifa_local;
 889         }
 890         rcu_read_unlock();
 891
 892         return saddr;
 893 }
 894
 895 static inline __le16 __dn_fib_res_prefsrc(struct dn_fib_res *res)
 896 {
 897         return dnet_select_source(DN_FIB_RES_DEV(*res), DN_FIB_RES_GW(*res), res->scope);
 898 }
 899
 900 static inline __le16 dn_fib_rules_map_destination(__le16 daddr, struct dn_fib_res *res)
 901 {
 902         __le16 mask = dnet_make_mask(res->prefixlen);
 903         return (daddr&~mask)|res->fi->fib_nh->nh_gw;
 904 }
 905
 906 static int dn_route_output_slow(struct dst_entry **pprt, const struct flowidn *oldflp, int try_hard)
 907 {
 908         struct flowidn fld = {
 909                 .daddr = oldflp->daddr,
 910                 .saddr = oldflp->saddr,
 911                 .flowidn_scope = RT_SCOPE_UNIVERSE,
 912                 .flowidn_mark = oldflp->flowidn_mark,
 913                 .flowidn_iif = init_net.loopback_dev->ifindex,
 914                 .flowidn_oif = oldflp->flowidn_oif,
 915         };
 916         struct dn_route *rt = NULL;
 917         struct net_device *dev_out = NULL, *dev;
 918         struct neighbour *neigh = NULL;
 919         unsigned hash;
 920         unsigned flags = 0;
 921         struct dn_fib_res res = { .fi = NULL, .type = RTN_UNICAST };
 922         int err;
 923         int free_res = 0;
 924         __le16 gateway = 0;
 925
 926         if (decnet_debug_level & 16)
 927                 printk(KERN_DEBUG
 928                        "dn_route_output_slow: dst=%04x src=%04x mark=%d"
 929                        " iif=%d oif=%d\n", le16_to_cpu(oldflp->daddr),
 930                        le16_to_cpu(oldflp->saddr),
 931                        oldflp->flowidn_mark, init_net.loopback_dev->ifindex,
 932                        oldflp->flowidn_oif);
 933
 934         /* If we have an output interface, verify its a DECnet device */
 935         if (oldflp->flowidn_oif) {
 936                 dev_out = dev_get_by_index(&init_net, oldflp->flowidn_oif);
 937                 err = -ENODEV;
 938                 if (dev_out && dev_out->dn_ptr == NULL) {
 939                         dev_put(dev_out);
 940                         dev_out = NULL;
 941                 }
 942                 if (dev_out == NULL)
 943                         goto out;
 944         }
 945
 946         /* If we have a source address, verify that its a local address */
 947         if (oldflp->saddr) {
 948                 err = -EADDRNOTAVAIL;
 949
 950                 if (dev_out) {
 951                         if (dn_dev_islocal(dev_out, oldflp->saddr))
 952                                 goto source_ok;
 953                         dev_put(dev_out);
 954                         goto out;
 955                 }
 956                 rcu_read_lock();
 957                 for_each_netdev_rcu(&init_net, dev) {
 958                         if (!dev->dn_ptr)
 959                                 continue;
 960                         if (!dn_dev_islocal(dev, oldflp->saddr))
 961                                 continue;
 962                         if ((dev->flags & IFF_LOOPBACK) &&
 963                             oldflp->daddr &&
 964                             !dn_dev_islocal(dev, oldflp->daddr))
 965                                 continue;
 966
 967                         dev_out = dev;
 968                         break;
 969                 }
 970                 rcu_read_unlock();
 971                 if (dev_out == NULL)
 972                         goto out;
 973                 dev_hold(dev_out);
 974 source_ok:
 975                 ;
 976         }
 977
 978         /* No destination? Assume its local */
 979         if (!fld.daddr) {
 980                 fld.daddr = fld.saddr;
 981
 982                 err = -EADDRNOTAVAIL;
 983                 if (dev_out)
 984                         dev_put(dev_out);
 985                 dev_out = init_net.loopback_dev;
 986                 dev_hold(dev_out);
 987                 if (!fld.daddr) {
 988                         fld.daddr =
 989                         fld.saddr = dnet_select_source(dev_out, 0,
 990                                                        RT_SCOPE_HOST);
 991                         if (!fld.daddr)
 992                                 goto out;
 993                 }
 994                 fld.flowidn_oif = init_net.loopback_dev->ifindex;
 995                 res.type = RTN_LOCAL;
 996                 goto make_route;
 997         }
 998
 999         if (decnet_debug_level & 16)
1000                 printk(KERN_DEBUG
1001                        "dn_route_output_slow: initial checks complete."
1002                        " dst=%o4x src=%04x oif=%d try_hard=%d\n",
1003                        le16_to_cpu(fld.daddr), le16_to_cpu(fld.saddr),
1004                        fld.flowidn_oif, try_hard);
1005
1006         /*
1007          * N.B. If the kernel is compiled without router support then
1008          * dn_fib_lookup() will evaluate to non-zero so this if () block
1009          * will always be executed.
1010          */
1011         err = -ESRCH;
1012         if (try_hard || (err = dn_fib_lookup(&fld, &res)) != 0) {
1013                 struct dn_dev *dn_db;
1014                 if (err != -ESRCH)
1015                         goto out;
1016                 /*
1017                  * Here the fallback is basically the standard algorithm for
1018                  * routing in endnodes which is described in the DECnet routing
1019                  * docs
1020                  *
1021                  * If we are not trying hard, look in neighbour cache.
1022                  * The result is tested to ensure that if a specific output
1023                  * device/source address was requested, then we honour that
1024                  * here
1025                  */
1026                 if (!try_hard) {
1027                         neigh = neigh_lookup_nodev(&dn_neigh_table, &init_net, &fld.daddr);
1028                         if (neigh) {
1029                                 if ((oldflp->flowidn_oif &&
1030                                     (neigh->dev->ifindex != oldflp->flowidn_oif)) ||
1031                                     (oldflp->saddr &&
1032                                     (!dn_dev_islocal(neigh->dev,
1033                                                      oldflp->saddr)))) {
1034                                         neigh_release(neigh);
1035                                         neigh = NULL;
1036                                 } else {
1037                                         if (dev_out)
1038                                                 dev_put(dev_out);
1039                                         if (dn_dev_islocal(neigh->dev, fld.daddr)) {
1040                                                 dev_out = init_net.loopback_dev;
1041                                                 res.type = RTN_LOCAL;
1042                                         } else {
1043                                                 dev_out = neigh->dev;
1044                                         }
1045                                         dev_hold(dev_out);
1046                                         goto select_source;
1047                                 }
1048                         }
1049                 }
1050
1051                 /* Not there? Perhaps its a local address */
1052                 if (dev_out == NULL)
1053                         dev_out = dn_dev_get_default();
1054                 err = -ENODEV;
1055                 if (dev_out == NULL)
1056                         goto out;
1057                 dn_db = rcu_dereference_raw(dev_out->dn_ptr);
1058                 /* Possible improvement - check all devices for local addr */
1059                 if (dn_dev_islocal(dev_out, fld.daddr)) {
1060                         dev_put(dev_out);
1061                         dev_out = init_net.loopback_dev;
1062                         dev_hold(dev_out);
1063                         res.type = RTN_LOCAL;
1064                         goto select_source;
1065                 }
1066                 /* Not local either.... try sending it to the default router */
1067                 neigh = neigh_clone(dn_db->router);
1068                 BUG_ON(neigh && neigh->dev != dev_out);
1069
1070                 /* Ok then, we assume its directly connected and move on */
1071 select_source:
1072                 if (neigh)
1073                         gateway = ((struct dn_neigh *)neigh)->addr;
1074                 if (gateway == 0)
1075                         gateway = fld.daddr;
1076                 if (fld.saddr == 0) {
1077                         fld.saddr = dnet_select_source(dev_out, gateway,
1078                                                        res.type == RTN_LOCAL ?
1079                                                        RT_SCOPE_HOST :
1080                                                        RT_SCOPE_LINK);
1081                         if (fld.saddr == 0 && res.type != RTN_LOCAL)
1082                                 goto e_addr;
1083                 }
1084                 fld.flowidn_oif = dev_out->ifindex;
1085                 goto make_route;
1086         }
1087         free_res = 1;
1088
1089         if (res.type == RTN_NAT)
1090                 goto e_inval;
1091
1092         if (res.type == RTN_LOCAL) {
1093                 if (!fld.saddr)
1094                         fld.saddr = fld.daddr;
1095                 if (dev_out)
1096                         dev_put(dev_out);
1097                 dev_out = init_net.loopback_dev;
1098                 dev_hold(dev_out);
1099                 fld.flowidn_oif = dev_out->ifindex;
1100                 if (res.fi)
1101                         dn_fib_info_put(res.fi);
1102                 res.fi = NULL;
1103                 goto make_route;
1104         }
1105
1106         if (res.fi->fib_nhs > 1 && fld.flowidn_oif == 0)
1107                 dn_fib_select_multipath(&fld, &res);
1108
1109         /*
1110          * We could add some logic to deal with default routes here and
1111          * get rid of some of the special casing above.
1112          */
1113
1114         if (!fld.saddr)
1115                 fld.saddr = DN_FIB_RES_PREFSRC(res);
1116
1117         if (dev_out)
1118                 dev_put(dev_out);
1119         dev_out = DN_FIB_RES_DEV(res);
1120         dev_hold(dev_out);
1121         fld.flowidn_oif = dev_out->ifindex;
1122         gateway = DN_FIB_RES_GW(res);
1123
1124 make_route:
1125         if (dev_out->flags & IFF_LOOPBACK)
1126                 flags |= RTCF_LOCAL;
1127
1128         rt = dst_alloc(&dn_dst_ops, dev_out, 1, 0, DST_HOST);
1129         if (rt == NULL)
1130                 goto e_nobufs;
1131
1132         memset(&rt->fld, 0, sizeof(rt->fld));
1133         rt->fld.saddr        = oldflp->saddr;
1134         rt->fld.daddr        = oldflp->daddr;
1135         rt->fld.flowidn_oif  = oldflp->flowidn_oif;
1136         rt->fld.flowidn_iif  = 0;
1137         rt->fld.flowidn_mark = oldflp->flowidn_mark;
1138
1139         rt->rt_saddr      = fld.saddr;
1140         rt->rt_daddr      = fld.daddr;
1141         rt->rt_gateway    = gateway ? gateway : fld.daddr;
1142         rt->rt_local_src  = fld.saddr;
1143
1144         rt->rt_dst_map    = fld.daddr;
1145         rt->rt_src_map    = fld.saddr;
1146
1147         rt->dst.neighbour = neigh;
1148         neigh = NULL;
1149
1150         rt->dst.lastuse = jiffies;
1151         rt->dst.output  = dn_output;
1152         rt->dst.input   = dn_rt_bug;
1153         rt->rt_flags      = flags;
1154         if (flags & RTCF_LOCAL)
1155                 rt->dst.input = dn_nsp_rx;
1156
1157         err = dn_rt_set_next_hop(rt, &res);
1158         if (err)
1159                 goto e_neighbour;
1160
1161         hash = dn_hash(rt->fld.saddr, rt->fld.daddr);
1162         dn_insert_route(rt, hash, (struct dn_route **)pprt);
1163
1164 done:
1165         if (neigh)
1166                 neigh_release(neigh);
1167         if (free_res)
1168                 dn_fib_res_put(&res);
1169         if (dev_out)
1170                 dev_put(dev_out);
1171 out:
1172         return err;
1173
1174 e_addr:
1175         err = -EADDRNOTAVAIL;
1176         goto done;
1177 e_inval:
1178         err = -EINVAL;
1179         goto done;
1180 e_nobufs:
1181         err = -ENOBUFS;
1182         goto done;
1183 e_neighbour:
1184         dst_free(&rt->dst);
1185         goto e_nobufs;
1186 }
1187
1188
1189 /*
1190  * N.B. The flags may be moved into the flowi at some future stage.
1191  */
1192 static int __dn_route_output_key(struct dst_entry **pprt, const struct flowidn *flp, int flags)
1193 {
1194         unsigned hash = dn_hash(flp->saddr, flp->daddr);
1195         struct dn_route *rt = NULL;
1196
1197         if (!(flags & MSG_TRYHARD)) {
1198                 rcu_read_lock_bh();
1199                 for (rt = rcu_dereference_bh(dn_rt_hash_table[hash].chain); rt;
1200                         rt = rcu_dereference_bh(rt->dst.dn_next)) {
1201                         if ((flp->daddr == rt->fld.daddr) &&
1202                             (flp->saddr == rt->fld.saddr) &&
1203                             (flp->flowidn_mark == rt->fld.flowidn_mark) &&
1204                             dn_is_output_route(rt) &&
1205                             (rt->fld.flowidn_oif == flp->flowidn_oif)) {
1206                                 dst_use(&rt->dst, jiffies);
1207                                 rcu_read_unlock_bh();
1208                                 *pprt = &rt->dst;
1209                                 return 0;
1210                         }
1211                 }
1212                 rcu_read_unlock_bh();
1213         }
1214
1215         return dn_route_output_slow(pprt, flp, flags);
1216 }
1217
1218 static int dn_route_output_key(struct dst_entry **pprt, struct flowidn *flp, int flags)
1219 {
1220         int err;
1221
1222         err = __dn_route_output_key(pprt, flp, flags);
1223         if (err == 0 && flp->flowidn_proto) {
1224                 *pprt = xfrm_lookup(&init_net, *pprt,
1225                                     flowidn_to_flowi(flp), NULL, 0);
1226                 if (IS_ERR(*pprt)) {
1227                         err = PTR_ERR(*pprt);
1228                         *pprt = NULL;
1229                 }
1230         }
1231         return err;
1232 }
1233
1234 int dn_route_output_sock(struct dst_entry **pprt, struct flowidn *fl, struct sock *sk, int flags)
1235 {
1236         int err;
1237
1238         err = __dn_route_output_key(pprt, fl, flags & MSG_TRYHARD);
1239         if (err == 0 && fl->flowidn_proto) {
1240                 if (!(flags & MSG_DONTWAIT))
1241                         fl->flowidn_flags |= FLOWI_FLAG_CAN_SLEEP;
1242                 *pprt = xfrm_lookup(&init_net, *pprt,
1243                                     flowidn_to_flowi(fl), sk, 0);
1244                 if (IS_ERR(*pprt)) {
1245                         err = PTR_ERR(*pprt);
1246                         *pprt = NULL;
1247                 }
1248         }
1249         return err;
1250 }
1251
1252 static int dn_route_input_slow(struct sk_buff *skb)
1253 {
1254         struct dn_route *rt = NULL;
1255         struct dn_skb_cb *cb = DN_SKB_CB(skb);
1256         struct net_device *in_dev = skb->dev;
1257         struct net_device *out_dev = NULL;
1258         struct dn_dev *dn_db;
1259         struct neighbour *neigh = NULL;
1260         unsigned hash;
1261         int flags = 0;
1262         __le16 gateway = 0;
1263         __le16 local_src = 0;
1264         struct flowidn fld = {
1265                 .daddr = cb->dst,
1266                 .saddr = cb->src,
1267                 .flowidn_scope = RT_SCOPE_UNIVERSE,
1268                 .flowidn_mark = skb->mark,
1269                 .flowidn_iif = skb->dev->ifindex,
1270         };
1271         struct dn_fib_res res = { .fi = NULL, .type = RTN_UNREACHABLE };
1272         int err = -EINVAL;
1273         int free_res = 0;
1274
1275         dev_hold(in_dev);
1276
1277         if ((dn_db = rcu_dereference(in_dev->dn_ptr)) == NULL)
1278                 goto out;
1279
1280         /* Zero source addresses are not allowed */
1281         if (fld.saddr == 0)
1282                 goto out;
1283
1284         /*
1285          * In this case we've just received a packet from a source
1286          * outside ourselves pretending to come from us. We don't
1287          * allow it any further to prevent routing loops, spoofing and
1288          * other nasties. Loopback packets already have the dst attached
1289          * so this only affects packets which have originated elsewhere.
1290          */
1291         err  = -ENOTUNIQ;
1292         if (dn_dev_islocal(in_dev, cb->src))
1293                 goto out;
1294
1295         err = dn_fib_lookup(&fld, &res);
1296         if (err) {
1297                 if (err != -ESRCH)
1298                         goto out;
1299                 /*
1300                  * Is the destination us ?
1301                  */
1302                 if (!dn_dev_islocal(in_dev, cb->dst))
1303                         goto e_inval;
1304
1305                 res.type = RTN_LOCAL;
1306         } else {
1307                 __le16 src_map = fld.saddr;
1308                 free_res = 1;
1309
1310                 out_dev = DN_FIB_RES_DEV(res);
1311                 if (out_dev == NULL) {
1312                         if (net_ratelimit())
1313                                 printk(KERN_CRIT "Bug in dn_route_input_slow() "
1314                                                  "No output device\n");
1315                         goto e_inval;
1316                 }
1317                 dev_hold(out_dev);
1318
1319                 if (res.r)
1320                         src_map = fld.saddr; /* no NAT support for now */
1321
1322                 gateway = DN_FIB_RES_GW(res);
1323                 if (res.type == RTN_NAT) {
1324                         fld.daddr = dn_fib_rules_map_destination(fld.daddr, &res);
1325                         dn_fib_res_put(&res);
1326                         free_res = 0;
1327                         if (dn_fib_lookup(&fld, &res))
1328                                 goto e_inval;
1329                         free_res = 1;
1330                         if (res.type != RTN_UNICAST)
1331                                 goto e_inval;
1332                         flags |= RTCF_DNAT;
1333                         gateway = fld.daddr;
1334                 }
1335                 fld.saddr = src_map;
1336         }
1337
1338         switch(res.type) {
1339         case RTN_UNICAST:
1340                 /*
1341                  * Forwarding check here, we only check for forwarding
1342                  * being turned off, if you want to only forward intra
1343                  * area, its up to you to set the routing tables up
1344                  * correctly.
1345                  */
1346                 if (dn_db->parms.forwarding == 0)
1347                         goto e_inval;
1348
1349                 if (res.fi->fib_nhs > 1 && fld.flowidn_oif == 0)
1350                         dn_fib_select_multipath(&fld, &res);
1351
1352                 /*
1353                  * Check for out_dev == in_dev. We use the RTCF_DOREDIRECT
1354                  * flag as a hint to set the intra-ethernet bit when
1355                  * forwarding. If we've got NAT in operation, we don't do
1356                  * this optimisation.
1357                  */
1358                 if (out_dev == in_dev && !(flags & RTCF_NAT))
1359                         flags |= RTCF_DOREDIRECT;
1360
1361                 local_src = DN_FIB_RES_PREFSRC(res);
1362
1363         case RTN_BLACKHOLE:
1364         case RTN_UNREACHABLE:
1365                 break;
1366         case RTN_LOCAL:
1367                 flags |= RTCF_LOCAL;
1368                 fld.saddr = cb->dst;
1369                 fld.daddr = cb->src;
1370
1371                 /* Routing tables gave us a gateway */
1372                 if (gateway)
1373                         goto make_route;
1374
1375                 /* Packet was intra-ethernet, so we know its on-link */
1376                 if (cb->rt_flags & DN_RT_F_IE) {
1377                         gateway = cb->src;
1378                         flags |= RTCF_DIRECTSRC;
1379                         goto make_route;
1380                 }
1381
1382                 /* Use the default router if there is one */
1383                 neigh = neigh_clone(dn_db->router);
1384                 if (neigh) {
1385                         gateway = ((struct dn_neigh *)neigh)->addr;
1386                         goto make_route;
1387                 }
1388
1389                 /* Close eyes and pray */
1390                 gateway = cb->src;
1391                 flags |= RTCF_DIRECTSRC;
1392                 goto make_route;
1393         default:
1394                 goto e_inval;
1395         }
1396
1397 make_route:
1398         rt = dst_alloc(&dn_dst_ops, out_dev, 0, 0, DST_HOST);
1399         if (rt == NULL)
1400                 goto e_nobufs;
1401
1402         memset(&rt->fld, 0, sizeof(rt->fld));
1403         rt->rt_saddr      = fld.saddr;
1404         rt->rt_daddr      = fld.daddr;
1405         rt->rt_gateway    = fld.daddr;
1406         if (gateway)
1407                 rt->rt_gateway = gateway;
1408         rt->rt_local_src  = local_src ? local_src : rt->rt_saddr;
1409
1410         rt->rt_dst_map    = fld.daddr;
1411         rt->rt_src_map    = fld.saddr;
1412
1413         rt->fld.saddr        = cb->src;
1414         rt->fld.daddr        = cb->dst;
1415         rt->fld.flowidn_oif  = 0;
1416         rt->fld.flowidn_iif  = in_dev->ifindex;
1417         rt->fld.flowidn_mark = fld.flowidn_mark;
1418
1419         rt->dst.neighbour = neigh;
1420         rt->dst.lastuse = jiffies;
1421         rt->dst.output = dn_rt_bug;
1422         switch(res.type) {
1423                 case RTN_UNICAST:
1424                         rt->dst.input = dn_forward;
1425                         break;
1426                 case RTN_LOCAL:
1427                         rt->dst.output = dn_output;
1428                         rt->dst.input = dn_nsp_rx;
1429                         rt->dst.dev = in_dev;
1430                         flags |= RTCF_LOCAL;
1431                         break;
1432                 default:
1433                 case RTN_UNREACHABLE:
1434                 case RTN_BLACKHOLE:
1435                         rt->dst.input = dst_discard;
1436         }
1437         rt->rt_flags = flags;
1438
1439         err = dn_rt_set_next_hop(rt, &res);
1440         if (err)
1441                 goto e_neighbour;
1442
1443         hash = dn_hash(rt->fld.saddr, rt->fld.daddr);
1444         dn_insert_route(rt, hash, &rt);
1445         skb_dst_set(skb, &rt->dst);
1446
1447 done:
1448         if (neigh)
1449                 neigh_release(neigh);
1450         if (free_res)
1451                 dn_fib_res_put(&res);
1452         dev_put(in_dev);
1453         if (out_dev)
1454                 dev_put(out_dev);
1455 out:
1456         return err;
1457
1458 e_inval:
1459         err = -EINVAL;
1460         goto done;
1461
1462 e_nobufs:
1463         err = -ENOBUFS;
1464         goto done;
1465
1466 e_neighbour:
1467         dst_free(&rt->dst);
1468         goto done;
1469 }
1470
1471 static int dn_route_input(struct sk_buff *skb)
1472 {
1473         struct dn_route *rt;
1474         struct dn_skb_cb *cb = DN_SKB_CB(skb);
1475         unsigned hash = dn_hash(cb->src, cb->dst);
1476
1477         if (skb_dst(skb))
1478                 return 0;
1479
1480         rcu_read_lock();
1481         for(rt = rcu_dereference(dn_rt_hash_table[hash].chain); rt != NULL;
1482             rt = rcu_dereference(rt->dst.dn_next)) {
1483                 if ((rt->fld.saddr == cb->src) &&
1484                     (rt->fld.daddr == cb->dst) &&
1485                     (rt->fld.flowidn_oif == 0) &&
1486                     (rt->fld.flowidn_mark == skb->mark) &&
1487                     (rt->fld.flowidn_iif == cb->iif)) {
1488                         dst_use(&rt->dst, jiffies);
1489                         rcu_read_unlock();
1490                         skb_dst_set(skb, (struct dst_entry *)rt);
1491                         return 0;
1492                 }
1493         }
1494         rcu_read_unlock();
1495
1496         return dn_route_input_slow(skb);
1497 }
1498
1499 static int dn_rt_fill_info(struct sk_buff *skb, u32 pid, u32 seq,
1500                            int event, int nowait, unsigned int flags)
1501 {
1502         struct dn_route *rt = (struct dn_route *)skb_dst(skb);
1503         struct rtmsg *r;
1504         struct nlmsghdr *nlh;
1505         unsigned char *b = skb_tail_pointer(skb);
1506         long expires;
1507
1508         nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*r), flags);
1509         r = NLMSG_DATA(nlh);
1510         r->rtm_family = AF_DECnet;
1511         r->rtm_dst_len = 16;
1512         r->rtm_src_len = 0;
1513         r->rtm_tos = 0;
1514         r->rtm_table = RT_TABLE_MAIN;
1515         RTA_PUT_U32(skb, RTA_TABLE, RT_TABLE_MAIN);
1516         r->rtm_type = rt->rt_type;
1517         r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
1518         r->rtm_scope = RT_SCOPE_UNIVERSE;
1519         r->rtm_protocol = RTPROT_UNSPEC;
1520         if (rt->rt_flags & RTCF_NOTIFY)
1521                 r->rtm_flags |= RTM_F_NOTIFY;
1522         RTA_PUT(skb, RTA_DST, 2, &rt->rt_daddr);
1523         if (rt->fld.saddr) {
1524                 r->rtm_src_len = 16;
1525                 RTA_PUT(skb, RTA_SRC, 2, &rt->fld.saddr);
1526         }
1527         if (rt->dst.dev)
1528                 RTA_PUT(skb, RTA_OIF, sizeof(int), &rt->dst.dev->ifindex);
1529         /*
1530          * Note to self - change this if input routes reverse direction when
1531          * they deal only with inputs and not with replies like they do
1532          * currently.
1533          */
1534         RTA_PUT(skb, RTA_PREFSRC, 2, &rt->rt_local_src);
1535         if (rt->rt_daddr != rt->rt_gateway)
1536                 RTA_PUT(skb, RTA_GATEWAY, 2, &rt->rt_gateway);
1537         if (rtnetlink_put_metrics(skb, dst_metrics_ptr(&rt->dst)) < 0)
1538                 goto rtattr_failure;
1539         expires = rt->dst.expires ? rt->dst.expires - jiffies : 0;
1540         if (rtnl_put_cacheinfo(skb, &rt->dst, 0, 0, 0, expires,
1541                                rt->dst.error) < 0)
1542                 goto rtattr_failure;
1543         if (dn_is_input_route(rt))
1544                 RTA_PUT(skb, RTA_IIF, sizeof(int), &rt->fld.flowidn_iif);
1545
1546         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1547         return skb->len;
1548
1549 nlmsg_failure:
1550 rtattr_failure:
1551         nlmsg_trim(skb, b);
1552         return -1;
1553 }
1554
1555 /*
1556  * This is called by both endnodes and routers now.
1557  */
1558 static int dn_cache_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh, void *arg)
1559 {
1560         struct net *net = sock_net(in_skb->sk);
1561         struct rtattr **rta = arg;
1562         struct rtmsg *rtm = NLMSG_DATA(nlh);
1563         struct dn_route *rt = NULL;
1564         struct dn_skb_cb *cb;
1565         int err;
1566         struct sk_buff *skb;
1567         struct flowidn fld;
1568
1569         if (!net_eq(net, &init_net))
1570                 return -EINVAL;
1571
1572         memset(&fld, 0, sizeof(fld));
1573         fld.flowidn_proto = DNPROTO_NSP;
1574
1575         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1576         if (skb == NULL)
1577                 return -ENOBUFS;
1578         skb_reset_mac_header(skb);
1579         cb = DN_SKB_CB(skb);
1580
1581         if (rta[RTA_SRC-1])
1582                 memcpy(&fld.saddr, RTA_DATA(rta[RTA_SRC-1]), 2);
1583         if (rta[RTA_DST-1])
1584                 memcpy(&fld.daddr, RTA_DATA(rta[RTA_DST-1]), 2);
1585         if (rta[RTA_IIF-1])
1586                 memcpy(&fld.flowidn_iif, RTA_DATA(rta[RTA_IIF-1]), sizeof(int));
1587
1588         if (fld.flowidn_iif) {
1589                 struct net_device *dev;
1590                 if ((dev = dev_get_by_index(&init_net, fld.flowidn_iif)) == NULL) {
1591                         kfree_skb(skb);
1592                         return -ENODEV;
1593                 }
1594                 if (!dev->dn_ptr) {
1595                         dev_put(dev);
1596                         kfree_skb(skb);
1597                         return -ENODEV;
1598                 }
1599                 skb->protocol = htons(ETH_P_DNA_RT);
1600                 skb->dev = dev;
1601                 cb->src = fld.saddr;
1602                 cb->dst = fld.daddr;
1603                 local_bh_disable();
1604                 err = dn_route_input(skb);
1605                 local_bh_enable();
1606                 memset(cb, 0, sizeof(struct dn_skb_cb));
1607                 rt = (struct dn_route *)skb_dst(skb);
1608                 if (!err && -rt->dst.error)
1609                         err = rt->dst.error;
1610         } else {
1611                 int oif = 0;
1612                 if (rta[RTA_OIF - 1])
1613                         memcpy(&oif, RTA_DATA(rta[RTA_OIF - 1]), sizeof(int));
1614                 fld.flowidn_oif = oif;
1615                 err = dn_route_output_key((struct dst_entry **)&rt, &fld, 0);
1616         }
1617
1618         if (skb->dev)
1619                 dev_put(skb->dev);
1620         skb->dev = NULL;
1621         if (err)
1622                 goto out_free;
1623         skb_dst_set(skb, &rt->dst);
1624         if (rtm->rtm_flags & RTM_F_NOTIFY)
1625                 rt->rt_flags |= RTCF_NOTIFY;
1626
1627         err = dn_rt_fill_info(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq, RTM_NEWROUTE, 0, 0);
1628
1629         if (err == 0)
1630                 goto out_free;
1631         if (err < 0) {
1632                 err = -EMSGSIZE;
1633                 goto out_free;
1634         }
1635
1636         return rtnl_unicast(skb, &init_net, NETLINK_CB(in_skb).pid);
1637
1638 out_free:
1639         kfree_skb(skb);
1640         return err;
1641 }
1642
1643 /*
1644  * For routers, this is called from dn_fib_dump, but for endnodes its
1645  * called directly from the rtnetlink dispatch table.
1646  */
1647 int dn_cache_dump(struct sk_buff *skb, struct netlink_callback *cb)
1648 {
1649         struct net *net = sock_net(skb->sk);
1650         struct dn_route *rt;
1651         int h, s_h;
1652         int idx, s_idx;
1653
1654         if (!net_eq(net, &init_net))
1655                 return 0;
1656
1657         if (NLMSG_PAYLOAD(cb->nlh, 0) < sizeof(struct rtmsg))
1658                 return -EINVAL;
1659         if (!(((struct rtmsg *)NLMSG_DATA(cb->nlh))->rtm_flags&RTM_F_CLONED))
1660                 return 0;
1661
1662         s_h = cb->args[0];
1663         s_idx = idx = cb->args[1];
1664         for(h = 0; h <= dn_rt_hash_mask; h++) {
1665                 if (h < s_h)
1666                         continue;
1667                 if (h > s_h)
1668                         s_idx = 0;
1669                 rcu_read_lock_bh();
1670                 for(rt = rcu_dereference_bh(dn_rt_hash_table[h].chain), idx = 0;
1671                         rt;
1672                         rt = rcu_dereference_bh(rt->dst.dn_next), idx++) {
1673                         if (idx < s_idx)
1674                                 continue;
1675                         skb_dst_set(skb, dst_clone(&rt->dst));
1676                         if (dn_rt_fill_info(skb, NETLINK_CB(cb->skb).pid,
1677                                         cb->nlh->nlmsg_seq, RTM_NEWROUTE,
1678                                         1, NLM_F_MULTI) <= 0) {
1679                                 skb_dst_drop(skb);
1680                                 rcu_read_unlock_bh();
1681                                 goto done;
1682                         }
1683                         skb_dst_drop(skb);
1684                 }
1685                 rcu_read_unlock_bh();
1686         }
1687
1688 done:
1689         cb->args[0] = h;
1690         cb->args[1] = idx;
1691         return skb->len;
1692 }
1693
1694 #ifdef CONFIG_PROC_FS
1695 struct dn_rt_cache_iter_state {
1696         int bucket;
1697 };
1698
1699 static struct dn_route *dn_rt_cache_get_first(struct seq_file *seq)
1700 {
1701         struct dn_route *rt = NULL;
1702         struct dn_rt_cache_iter_state *s = seq->private;
1703
1704         for(s->bucket = dn_rt_hash_mask; s->bucket >= 0; --s->bucket) {
1705                 rcu_read_lock_bh();
1706                 rt = rcu_dereference_bh(dn_rt_hash_table[s->bucket].chain);
1707                 if (rt)
1708                         break;
1709                 rcu_read_unlock_bh();
1710         }
1711         return rt;
1712 }
1713
1714 static struct dn_route *dn_rt_cache_get_next(struct seq_file *seq, struct dn_route *rt)
1715 {
1716         struct dn_rt_cache_iter_state *s = seq->private;
1717
1718         rt = rcu_dereference_bh(rt->dst.dn_next);
1719         while (!rt) {
1720                 rcu_read_unlock_bh();
1721                 if (--s->bucket < 0)
1722                         break;
1723                 rcu_read_lock_bh();
1724                 rt = rcu_dereference_bh(dn_rt_hash_table[s->bucket].chain);
1725         }
1726         return rt;
1727 }
1728
1729 static void *dn_rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
1730 {
1731         struct dn_route *rt = dn_rt_cache_get_first(seq);
1732
1733         if (rt) {
1734                 while(*pos && (rt = dn_rt_cache_get_next(seq, rt)))
1735                         --*pos;
1736         }
1737         return *pos ? NULL : rt;
1738 }
1739
1740 static void *dn_rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1741 {
1742         struct dn_route *rt = dn_rt_cache_get_next(seq, v);
1743         ++*pos;
1744         return rt;
1745 }
1746
1747 static void dn_rt_cache_seq_stop(struct seq_file *seq, void *v)
1748 {
1749         if (v)
1750                 rcu_read_unlock_bh();
1751 }
1752
1753 static int dn_rt_cache_seq_show(struct seq_file *seq, void *v)
1754 {
1755         struct dn_route *rt = v;
1756         char buf1[DN_ASCBUF_LEN], buf2[DN_ASCBUF_LEN];
1757
1758         seq_printf(seq, "%-8s %-7s %-7s %04d %04d %04d\n",
1759                         rt->dst.dev ? rt->dst.dev->name : "*",
1760                         dn_addr2asc(le16_to_cpu(rt->rt_daddr), buf1),
1761                         dn_addr2asc(le16_to_cpu(rt->rt_saddr), buf2),
1762                         atomic_read(&rt->dst.__refcnt),
1763                         rt->dst.__use,
1764                         (int) dst_metric(&rt->dst, RTAX_RTT));
1765         return 0;
1766 }
1767
1768 static const struct seq_operations dn_rt_cache_seq_ops = {
1769         .start  = dn_rt_cache_seq_start,
1770         .next   = dn_rt_cache_seq_next,
1771         .stop   = dn_rt_cache_seq_stop,
1772         .show   = dn_rt_cache_seq_show,
1773 };
1774
1775 static int dn_rt_cache_seq_open(struct inode *inode, struct file *file)
1776 {
1777         return seq_open_private(file, &dn_rt_cache_seq_ops,
1778                         sizeof(struct dn_rt_cache_iter_state));
1779 }
1780
1781 static const struct file_operations dn_rt_cache_seq_fops = {
1782         .owner   = THIS_MODULE,
1783         .open    = dn_rt_cache_seq_open,
1784         .read    = seq_read,
1785         .llseek  = seq_lseek,
1786         .release = seq_release_private,
1787 };
1788
1789 #endif /* CONFIG_PROC_FS */
1790
1791 void __init dn_route_init(void)
1792 {
1793         int i, goal, order;
1794
1795         dn_dst_ops.kmem_cachep =
1796                 kmem_cache_create("dn_dst_cache", sizeof(struct dn_route), 0,
1797                                   SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1798         dst_entries_init(&dn_dst_ops);
1799         setup_timer(&dn_route_timer, dn_dst_check_expire, 0);
1800         dn_route_timer.expires = jiffies + decnet_dst_gc_interval * HZ;
1801         add_timer(&dn_route_timer);
1802
1803         goal = totalram_pages >> (26 - PAGE_SHIFT);
1804
1805         for(order = 0; (1UL << order) < goal; order++)
1806                 /* NOTHING */;
1807
1808         /*
1809          * Only want 1024 entries max, since the table is very, very unlikely
1810          * to be larger than that.
1811          */
1812         while(order && ((((1UL << order) * PAGE_SIZE) /
1813                                 sizeof(struct dn_rt_hash_bucket)) >= 2048))
1814                 order--;
1815
1816         do {
1817                 dn_rt_hash_mask = (1UL << order) * PAGE_SIZE /
1818                         sizeof(struct dn_rt_hash_bucket);
1819                 while(dn_rt_hash_mask & (dn_rt_hash_mask - 1))
1820                         dn_rt_hash_mask--;
1821                 dn_rt_hash_table = (struct dn_rt_hash_bucket *)
1822                         __get_free_pages(GFP_ATOMIC, order);
1823         } while (dn_rt_hash_table == NULL && --order > 0);
1824
1825         if (!dn_rt_hash_table)
1826                 panic("Failed to allocate DECnet route cache hash table\n");
1827
1828         printk(KERN_INFO
1829                 "DECnet: Routing cache hash table of %u buckets, %ldKbytes\n",
1830                 dn_rt_hash_mask,
1831                 (long)(dn_rt_hash_mask*sizeof(struct dn_rt_hash_bucket))/1024);
1832
1833         dn_rt_hash_mask--;
1834         for(i = 0; i <= dn_rt_hash_mask; i++) {
1835                 spin_lock_init(&dn_rt_hash_table[i].lock);
1836                 dn_rt_hash_table[i].chain = NULL;
1837         }
1838
1839         dn_dst_ops.gc_thresh = (dn_rt_hash_mask + 1);
1840
1841         proc_net_fops_create(&init_net, "decnet_cache", S_IRUGO, &dn_rt_cache_seq_fops);
1842
1843 #ifdef CONFIG_DECNET_ROUTER
1844         rtnl_register(PF_DECnet, RTM_GETROUTE, dn_cache_getroute, dn_fib_dump);
1845 #else
1846         rtnl_register(PF_DECnet, RTM_GETROUTE, dn_cache_getroute,
1847                       dn_cache_dump);
1848 #endif
1849 }
1850
1851 void __exit dn_route_cleanup(void)
1852 {
1853         del_timer(&dn_route_timer);
1854         dn_run_flush(0);
1855
1856         proc_net_remove(&init_net, "decnet_cache");
1857         dst_entries_destroy(&dn_dst_ops);
1858 }
1859