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 <net/sock.h>
  70 #include <linux/mm.h>
  71 #include <linux/proc_fs.h>
  72 #include <linux/seq_file.h>
  73 #include <linux/init.h>
  74 #include <linux/rtnetlink.h>
  75 #include <linux/string.h>
  76 #include <linux/netfilter_decnet.h>
  77 #include <linux/rcupdate.h>
  78 #include <linux/times.h>
  79 #include <asm/errno.h>
  80 #include <net/neighbour.h>
  81 #include <net/dst.h>
  82 #include <net/flow.h>
  83 #include <net/dn.h>
  84 #include <net/dn_dev.h>
  85 #include <net/dn_nsp.h>
  86 #include <net/dn_route.h>
  87 #include <net/dn_neigh.h>
  88 #include <net/dn_fib.h>
  89
  90 struct dn_rt_hash_bucket
  91 {
  92         struct dn_route *chain;
  93         spinlock_t lock;
  94 } __attribute__((__aligned__(8)));
  95
  96 extern struct neigh_table dn_neigh_table;
  97
  98
  99 static unsigned char dn_hiord_addr[6] = {0xAA,0x00,0x04,0x00,0x00,0x00};
 100
 101 static const int dn_rt_min_delay = 2 * HZ;
 102 static const int dn_rt_max_delay = 10 * HZ;
 103 static const int dn_rt_mtu_expires = 10 * 60 * HZ;
 104
 105 static unsigned long dn_rt_deadline;
 106
 107 static int dn_dst_gc(void);
 108 static struct dst_entry *dn_dst_check(struct dst_entry *, __u32);
 109 static struct dst_entry *dn_dst_negative_advice(struct dst_entry *);
 110 static void dn_dst_link_failure(struct sk_buff *);
 111 static void dn_dst_update_pmtu(struct dst_entry *dst, u32 mtu);
 112 static int dn_route_input(struct sk_buff *);
 113 static void dn_run_flush(unsigned long dummy);
 114
 115 static struct dn_rt_hash_bucket *dn_rt_hash_table;
 116 static unsigned dn_rt_hash_mask;
 117
 118 static struct timer_list dn_route_timer;
 119 static DEFINE_TIMER(dn_rt_flush_timer, dn_run_flush, 0, 0);
 120 int decnet_dst_gc_interval = 2;
 121
 122 static struct dst_ops dn_dst_ops = {
 123         .family =               PF_DECnet,
 124         .protocol =             __constant_htons(ETH_P_DNA_RT),
 125         .gc_thresh =            128,
 126         .gc =                   dn_dst_gc,
 127         .check =                dn_dst_check,
 128         .negative_advice =      dn_dst_negative_advice,
 129         .link_failure =         dn_dst_link_failure,
 130         .update_pmtu =          dn_dst_update_pmtu,
 131         .entry_size =           sizeof(struct dn_route),
 132         .entries =              ATOMIC_INIT(0),
 133 };
 134
 135 static __inline__ unsigned dn_hash(__le16 src, __le16 dst)
 136 {
 137         __u16 tmp = (__u16 __force)(src ^ dst);
 138         tmp ^= (tmp >> 3);
 139         tmp ^= (tmp >> 5);
 140         tmp ^= (tmp >> 10);
 141         return dn_rt_hash_mask & (unsigned)tmp;
 142 }
 143
 144 static inline void dnrt_free(struct dn_route *rt)
 145 {
 146         call_rcu_bh(&rt->u.dst.rcu_head, dst_rcu_free);
 147 }
 148
 149 static inline void dnrt_drop(struct dn_route *rt)
 150 {
 151         dst_release(&rt->u.dst);
 152         call_rcu_bh(&rt->u.dst.rcu_head, dst_rcu_free);
 153 }
 154
 155 static void dn_dst_check_expire(unsigned long dummy)
 156 {
 157         int i;
 158         struct dn_route *rt, **rtp;
 159         unsigned long now = jiffies;
 160         unsigned long expire = 120 * HZ;
 161
 162         for(i = 0; i <= dn_rt_hash_mask; i++) {
 163                 rtp = &dn_rt_hash_table[i].chain;
 164
 165                 spin_lock(&dn_rt_hash_table[i].lock);
 166                 while((rt=*rtp) != NULL) {
 167                         if (atomic_read(&rt->u.dst.__refcnt) ||
 168                                         (now - rt->u.dst.lastuse) < expire) {
 169                                 rtp = &rt->u.rt_next;
 170                                 continue;
 171                         }
 172                         *rtp = rt->u.rt_next;
 173                         rt->u.rt_next = NULL;
 174                         dnrt_free(rt);
 175                 }
 176                 spin_unlock(&dn_rt_hash_table[i].lock);
 177
 178                 if ((jiffies - now) > 0)
 179                         break;
 180         }
 181
 182         mod_timer(&dn_route_timer, now + decnet_dst_gc_interval * HZ);
 183 }
 184
 185 static int dn_dst_gc(void)
 186 {
 187         struct dn_route *rt, **rtp;
 188         int i;
 189         unsigned long now = jiffies;
 190         unsigned long expire = 10 * HZ;
 191
 192         for(i = 0; i <= dn_rt_hash_mask; i++) {
 193
 194                 spin_lock_bh(&dn_rt_hash_table[i].lock);
 195                 rtp = &dn_rt_hash_table[i].chain;
 196
 197                 while((rt=*rtp) != NULL) {
 198                         if (atomic_read(&rt->u.dst.__refcnt) ||
 199                                         (now - rt->u.dst.lastuse) < expire) {
 200                                 rtp = &rt->u.rt_next;
 201                                 continue;
 202                         }
 203                         *rtp = rt->u.rt_next;
 204                         rt->u.rt_next = NULL;
 205                         dnrt_drop(rt);
 206                         break;
 207                 }
 208                 spin_unlock_bh(&dn_rt_hash_table[i].lock);
 209         }
 210
 211         return 0;
 212 }
 213
 214 /*
 215  * The decnet standards don't impose a particular minimum mtu, what they
 216  * do insist on is that the routing layer accepts a datagram of at least
 217  * 230 bytes long. Here we have to subtract the routing header length from
 218  * 230 to get the minimum acceptable mtu. If there is no neighbour, then we
 219  * assume the worst and use a long header size.
 220  *
 221  * We update both the mtu and the advertised mss (i.e. the segment size we
 222  * advertise to the other end).
 223  */
 224 static void dn_dst_update_pmtu(struct dst_entry *dst, u32 mtu)
 225 {
 226         u32 min_mtu = 230;
 227         struct dn_dev *dn = dst->neighbour ?
 228                             (struct dn_dev *)dst->neighbour->dev->dn_ptr : NULL;
 229
 230         if (dn && dn->use_long == 0)
 231                 min_mtu -= 6;
 232         else
 233                 min_mtu -= 21;
 234
 235         if (dst->metrics[RTAX_MTU-1] > mtu && mtu >= min_mtu) {
 236                 if (!(dst_metric_locked(dst, RTAX_MTU))) {
 237                         dst->metrics[RTAX_MTU-1] = mtu;
 238                         dst_set_expires(dst, dn_rt_mtu_expires);
 239                 }
 240                 if (!(dst_metric_locked(dst, RTAX_ADVMSS))) {
 241                         u32 mss = mtu - DN_MAX_NSP_DATA_HEADER;
 242                         if (dst->metrics[RTAX_ADVMSS-1] > mss)
 243                                 dst->metrics[RTAX_ADVMSS-1] = mss;
 244                 }
 245         }
 246 }
 247
 248 /*
 249  * When a route has been marked obsolete. (e.g. routing cache flush)
 250  */
 251 static struct dst_entry *dn_dst_check(struct dst_entry *dst, __u32 cookie)
 252 {
 253         return NULL;
 254 }
 255
 256 static struct dst_entry *dn_dst_negative_advice(struct dst_entry *dst)
 257 {
 258         dst_release(dst);
 259         return NULL;
 260 }
 261
 262 static void dn_dst_link_failure(struct sk_buff *skb)
 263 {
 264         return;
 265 }
 266
 267 static inline int compare_keys(struct flowi *fl1, struct flowi *fl2)
 268 {
 269         return memcmp(&fl1->nl_u.dn_u, &fl2->nl_u.dn_u, sizeof(fl1->nl_u.dn_u)) == 0 &&
 270                 fl1->oif == fl2->oif &&
 271                 fl1->iif == fl2->iif;
 272 }
 273
 274 static int dn_insert_route(struct dn_route *rt, unsigned hash, struct dn_route **rp)
 275 {
 276         struct dn_route *rth, **rthp;
 277         unsigned long now = jiffies;
 278
 279         rthp = &dn_rt_hash_table[hash].chain;
 280
 281         spin_lock_bh(&dn_rt_hash_table[hash].lock);
 282         while((rth = *rthp) != NULL) {
 283                 if (compare_keys(&rth->fl, &rt->fl)) {
 284                         /* Put it first */
 285                         *rthp = rth->u.rt_next;
 286                         rcu_assign_pointer(rth->u.rt_next,
 287                                            dn_rt_hash_table[hash].chain);
 288                         rcu_assign_pointer(dn_rt_hash_table[hash].chain, rth);
 289
 290                         rth->u.dst.__use++;
 291                         dst_hold(&rth->u.dst);
 292                         rth->u.dst.lastuse = now;
 293                         spin_unlock_bh(&dn_rt_hash_table[hash].lock);
 294
 295                         dnrt_drop(rt);
 296                         *rp = rth;
 297                         return 0;
 298                 }
 299                 rthp = &rth->u.rt_next;
 300         }
 301
 302         rcu_assign_pointer(rt->u.rt_next, dn_rt_hash_table[hash].chain);
 303         rcu_assign_pointer(dn_rt_hash_table[hash].chain, rt);
 304
 305         dst_hold(&rt->u.dst);
 306         rt->u.dst.__use++;
 307         rt->u.dst.lastuse = now;
 308         spin_unlock_bh(&dn_rt_hash_table[hash].lock);
 309         *rp = rt;
 310         return 0;
 311 }
 312
 313 void dn_run_flush(unsigned long dummy)
 314 {
 315         int i;
 316         struct dn_route *rt, *next;
 317
 318         for(i = 0; i < dn_rt_hash_mask; i++) {
 319                 spin_lock_bh(&dn_rt_hash_table[i].lock);
 320
 321                 if ((rt = xchg(&dn_rt_hash_table[i].chain, NULL)) == NULL)
 322                         goto nothing_to_declare;
 323
 324                 for(; rt; rt=next) {
 325                         next = rt->u.rt_next;
 326                         rt->u.rt_next = NULL;
 327                         dst_free((struct dst_entry *)rt);
 328                 }
 329
 330 nothing_to_declare:
 331                 spin_unlock_bh(&dn_rt_hash_table[i].lock);
 332         }
 333 }
 334
 335 static DEFINE_SPINLOCK(dn_rt_flush_lock);
 336
 337 void dn_rt_cache_flush(int delay)
 338 {
 339         unsigned long now = jiffies;
 340         int user_mode = !in_interrupt();
 341
 342         if (delay < 0)
 343                 delay = dn_rt_min_delay;
 344
 345         spin_lock_bh(&dn_rt_flush_lock);
 346
 347         if (del_timer(&dn_rt_flush_timer) && delay > 0 && dn_rt_deadline) {
 348                 long tmo = (long)(dn_rt_deadline - now);
 349
 350                 if (user_mode && tmo < dn_rt_max_delay - dn_rt_min_delay)
 351                         tmo = 0;
 352
 353                 if (delay > tmo)
 354                         delay = tmo;
 355         }
 356
 357         if (delay <= 0) {
 358                 spin_unlock_bh(&dn_rt_flush_lock);
 359                 dn_run_flush(0);
 360                 return;
 361         }
 362
 363         if (dn_rt_deadline == 0)
 364                 dn_rt_deadline = now + dn_rt_max_delay;
 365
 366         dn_rt_flush_timer.expires = now + delay;
 367         add_timer(&dn_rt_flush_timer);
 368         spin_unlock_bh(&dn_rt_flush_lock);
 369 }
 370
 371 /**
 372  * dn_return_short - Return a short packet to its sender
 373  * @skb: The packet to return
 374  *
 375  */
 376 static int dn_return_short(struct sk_buff *skb)
 377 {
 378         struct dn_skb_cb *cb;
 379         unsigned char *ptr;
 380         __le16 *src;
 381         __le16 *dst;
 382         __le16 tmp;
 383
 384         /* Add back headers */
 385         skb_push(skb, skb->data - skb->nh.raw);
 386
 387         if ((skb = skb_unshare(skb, GFP_ATOMIC)) == NULL)
 388                 return NET_RX_DROP;
 389
 390         cb = DN_SKB_CB(skb);
 391         /* Skip packet length and point to flags */
 392         ptr = skb->data + 2;
 393         *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
 394
 395         dst = (__le16 *)ptr;
 396         ptr += 2;
 397         src = (__le16 *)ptr;
 398         ptr += 2;
 399         *ptr = 0; /* Zero hop count */
 400
 401         /* Swap source and destination */
 402         tmp  = *src;
 403         *src = *dst;
 404         *dst = tmp;
 405
 406         skb->pkt_type = PACKET_OUTGOING;
 407         dn_rt_finish_output(skb, NULL, NULL);
 408         return NET_RX_SUCCESS;
 409 }
 410
 411 /**
 412  * dn_return_long - Return a long packet to its sender
 413  * @skb: The long format packet to return
 414  *
 415  */
 416 static int dn_return_long(struct sk_buff *skb)
 417 {
 418         struct dn_skb_cb *cb;
 419         unsigned char *ptr;
 420         unsigned char *src_addr, *dst_addr;
 421         unsigned char tmp[ETH_ALEN];
 422
 423         /* Add back all headers */
 424         skb_push(skb, skb->data - skb->nh.raw);
 425
 426         if ((skb = skb_unshare(skb, GFP_ATOMIC)) == NULL)
 427                 return NET_RX_DROP;
 428
 429         cb = DN_SKB_CB(skb);
 430         /* Ignore packet length and point to flags */
 431         ptr = skb->data + 2;
 432
 433         /* Skip padding */
 434         if (*ptr & DN_RT_F_PF) {
 435                 char padlen = (*ptr & ~DN_RT_F_PF);
 436                 ptr += padlen;
 437         }
 438
 439         *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
 440         ptr += 2;
 441         dst_addr = ptr;
 442         ptr += 8;
 443         src_addr = ptr;
 444         ptr += 6;
 445         *ptr = 0; /* Zero hop count */
 446
 447         /* Swap source and destination */
 448         memcpy(tmp, src_addr, ETH_ALEN);
 449         memcpy(src_addr, dst_addr, ETH_ALEN);
 450         memcpy(dst_addr, tmp, ETH_ALEN);
 451
 452         skb->pkt_type = PACKET_OUTGOING;
 453         dn_rt_finish_output(skb, dst_addr, src_addr);
 454         return NET_RX_SUCCESS;
 455 }
 456
 457 /**
 458  * dn_route_rx_packet - Try and find a route for an incoming packet
 459  * @skb: The packet to find a route for
 460  *
 461  * Returns: result of input function if route is found, error code otherwise
 462  */
 463 static int dn_route_rx_packet(struct sk_buff *skb)
 464 {
 465         struct dn_skb_cb *cb = DN_SKB_CB(skb);
 466         int err;
 467
 468         if ((err = dn_route_input(skb)) == 0)
 469                 return dst_input(skb);
 470
 471         if (decnet_debug_level & 4) {
 472                 char *devname = skb->dev ? skb->dev->name : "???";
 473                 struct dn_skb_cb *cb = DN_SKB_CB(skb);
 474                 printk(KERN_DEBUG
 475                         "DECnet: dn_route_rx_packet: rt_flags=0x%02x dev=%s len=%d src=0x%04hx dst=0x%04hx err=%d type=%d\n",
 476                         (int)cb->rt_flags, devname, skb->len,
 477                         dn_ntohs(cb->src), dn_ntohs(cb->dst),
 478                         err, skb->pkt_type);
 479         }
 480
 481         if ((skb->pkt_type == PACKET_HOST) && (cb->rt_flags & DN_RT_F_RQR)) {
 482                 switch(cb->rt_flags & DN_RT_PKT_MSK) {
 483                         case DN_RT_PKT_SHORT:
 484                                 return dn_return_short(skb);
 485                         case DN_RT_PKT_LONG:
 486                                 return dn_return_long(skb);
 487                 }
 488         }
 489
 490         kfree_skb(skb);
 491         return NET_RX_DROP;
 492 }
 493
 494 static int dn_route_rx_long(struct sk_buff *skb)
 495 {
 496         struct dn_skb_cb *cb = DN_SKB_CB(skb);
 497         unsigned char *ptr = skb->data;
 498
 499         if (!pskb_may_pull(skb, 21)) /* 20 for long header, 1 for shortest nsp */
 500                 goto drop_it;
 501
 502         skb_pull(skb, 20);
 503         skb->h.raw = skb->data;
 504
 505         /* Destination info */
 506         ptr += 2;
 507         cb->dst = dn_eth2dn(ptr);
 508         if (memcmp(ptr, dn_hiord_addr, 4) != 0)
 509                 goto drop_it;
 510         ptr += 6;
 511
 512
 513         /* Source info */
 514         ptr += 2;
 515         cb->src = dn_eth2dn(ptr);
 516         if (memcmp(ptr, dn_hiord_addr, 4) != 0)
 517                 goto drop_it;
 518         ptr += 6;
 519         /* Other junk */
 520         ptr++;
 521         cb->hops = *ptr++; /* Visit Count */
 522
 523         return NF_HOOK(PF_DECnet, NF_DN_PRE_ROUTING, skb, skb->dev, NULL, dn_route_rx_packet);
 524
 525 drop_it:
 526         kfree_skb(skb);
 527         return NET_RX_DROP;
 528 }
 529
 530
 531
 532 static int dn_route_rx_short(struct sk_buff *skb)
 533 {
 534         struct dn_skb_cb *cb = DN_SKB_CB(skb);
 535         unsigned char *ptr = skb->data;
 536
 537         if (!pskb_may_pull(skb, 6)) /* 5 for short header + 1 for shortest nsp */
 538                 goto drop_it;
 539
 540         skb_pull(skb, 5);
 541         skb->h.raw = skb->data;
 542
 543         cb->dst = *(__le16 *)ptr;
 544         ptr += 2;
 545         cb->src = *(__le16 *)ptr;
 546         ptr += 2;
 547         cb->hops = *ptr & 0x3f;
 548
 549         return NF_HOOK(PF_DECnet, NF_DN_PRE_ROUTING, skb, skb->dev, NULL, 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 = dn_ntohs(*(__le16 *)skb->data);
 578         struct dn_dev *dn = (struct dn_dev *)dev->dn_ptr;
 579         unsigned char padlen = 0;
 580
 581         if (dn == NULL)
 582                 goto dump_it;
 583
 584         if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
 585                 goto out;
 586
 587         if (!pskb_may_pull(skb, 3))
 588                 goto dump_it;
 589
 590         skb_pull(skb, 2);
 591
 592         if (len > skb->len)
 593                 goto dump_it;
 594
 595         skb_trim(skb, len);
 596
 597         flags = *skb->data;
 598
 599         cb = DN_SKB_CB(skb);
 600         cb->stamp = jiffies;
 601         cb->iif = dev->ifindex;
 602
 603         /*
 604          * If we have padding, remove it.
 605          */
 606         if (flags & DN_RT_F_PF) {
 607                 padlen = flags & ~DN_RT_F_PF;
 608                 if (!pskb_may_pull(skb, padlen + 1))
 609                         goto dump_it;
 610                 skb_pull(skb, padlen);
 611                 flags = *skb->data;
 612         }
 613
 614         skb->nh.raw = skb->data;
 615
 616         /*
 617          * Weed out future version DECnet
 618          */
 619         if (flags & DN_RT_F_VER)
 620                 goto dump_it;
 621
 622         cb->rt_flags = flags;
 623
 624         if (decnet_debug_level & 1)
 625                 printk(KERN_DEBUG
 626                         "dn_route_rcv: got 0x%02x from %s [%d %d %d]\n",
 627                         (int)flags, (dev) ? dev->name : "???", len, skb->len,
 628                         padlen);
 629
 630         if (flags & DN_RT_PKT_CNTL) {
 631                 if (unlikely(skb_linearize(skb)))
 632                         goto dump_it;
 633
 634                 switch(flags & DN_RT_CNTL_MSK) {
 635                         case DN_RT_PKT_INIT:
 636                                 dn_dev_init_pkt(skb);
 637                                 break;
 638                         case DN_RT_PKT_VERI:
 639                                 dn_dev_veri_pkt(skb);
 640                                 break;
 641                 }
 642
 643                 if (dn->parms.state != DN_DEV_S_RU)
 644                         goto dump_it;
 645
 646                 switch(flags & DN_RT_CNTL_MSK) {
 647                         case DN_RT_PKT_HELO:
 648                                 return NF_HOOK(PF_DECnet, NF_DN_HELLO, skb, skb->dev, NULL, dn_route_ptp_hello);
 649
 650                         case DN_RT_PKT_L1RT:
 651                         case DN_RT_PKT_L2RT:
 652                                 return NF_HOOK(PF_DECnet, NF_DN_ROUTE, skb, skb->dev, NULL, dn_route_discard);
 653                         case DN_RT_PKT_ERTH:
 654                                 return NF_HOOK(PF_DECnet, NF_DN_HELLO, skb, skb->dev, NULL, dn_neigh_router_hello);
 655
 656                         case DN_RT_PKT_EEDH:
 657                                 return NF_HOOK(PF_DECnet, NF_DN_HELLO, skb, skb->dev, NULL, dn_neigh_endnode_hello);
 658                 }
 659         } else {
 660                 if (dn->parms.state != DN_DEV_S_RU)
 661                         goto dump_it;
 662
 663                 skb_pull(skb, 1); /* Pull flags */
 664
 665                 switch(flags & DN_RT_PKT_MSK) {
 666                         case DN_RT_PKT_LONG:
 667                                 return dn_route_rx_long(skb);
 668                         case DN_RT_PKT_SHORT:
 669                                 return dn_route_rx_short(skb);
 670                 }
 671         }
 672
 673 dump_it:
 674         kfree_skb(skb);
 675 out:
 676         return NET_RX_DROP;
 677 }
 678
 679 static int dn_output(struct sk_buff *skb)
 680 {
 681         struct dst_entry *dst = skb->dst;
 682         struct dn_route *rt = (struct dn_route *)dst;
 683         struct net_device *dev = dst->dev;
 684         struct dn_skb_cb *cb = DN_SKB_CB(skb);
 685         struct neighbour *neigh;
 686
 687         int err = -EINVAL;
 688
 689         if ((neigh = dst->neighbour) == NULL)
 690                 goto error;
 691
 692         skb->dev = dev;
 693
 694         cb->src = rt->rt_saddr;
 695         cb->dst = rt->rt_daddr;
 696
 697         /*
 698          * Always set the Intra-Ethernet bit on all outgoing packets
 699          * originated on this node. Only valid flag from upper layers
 700          * is return-to-sender-requested. Set hop count to 0 too.
 701          */
 702         cb->rt_flags &= ~DN_RT_F_RQR;
 703         cb->rt_flags |= DN_RT_F_IE;
 704         cb->hops = 0;
 705
 706         return NF_HOOK(PF_DECnet, NF_DN_LOCAL_OUT, skb, NULL, dev, neigh->output);
 707
 708 error:
 709         if (net_ratelimit())
 710                 printk(KERN_DEBUG "dn_output: This should not happen\n");
 711
 712         kfree_skb(skb);
 713
 714         return err;
 715 }
 716
 717 static int dn_forward(struct sk_buff *skb)
 718 {
 719         struct dn_skb_cb *cb = DN_SKB_CB(skb);
 720         struct dst_entry *dst = skb->dst;
 721         struct dn_dev *dn_db = dst->dev->dn_ptr;
 722         struct dn_route *rt;
 723         struct neighbour *neigh = dst->neighbour;
 724         int header_len;
 725 #ifdef CONFIG_NETFILTER
 726         struct net_device *dev = skb->dev;
 727 #endif
 728
 729         if (skb->pkt_type != PACKET_HOST)
 730                 goto drop;
 731
 732         /* Ensure that we have enough space for headers */
 733         rt = (struct dn_route *)skb->dst;
 734         header_len = dn_db->use_long ? 21 : 6;
 735         if (skb_cow(skb, LL_RESERVED_SPACE(rt->u.dst.dev)+header_len))
 736                 goto drop;
 737
 738         /*
 739          * Hop count exceeded.
 740          */
 741         if (++cb->hops > 30)
 742                 goto drop;
 743
 744         skb->dev = rt->u.dst.dev;
 745
 746         /*
 747          * If packet goes out same interface it came in on, then set
 748          * the Intra-Ethernet bit. This has no effect for short
 749          * packets, so we don't need to test for them here.
 750          */
 751         cb->rt_flags &= ~DN_RT_F_IE;
 752         if (rt->rt_flags & RTCF_DOREDIRECT)
 753                 cb->rt_flags |= DN_RT_F_IE;
 754
 755         return NF_HOOK(PF_DECnet, NF_DN_FORWARD, skb, dev, skb->dev, neigh->output);
 756
 757 drop:
 758         kfree_skb(skb);
 759         return NET_RX_DROP;
 760 }
 761
 762 /*
 763  * Drop packet. This is used for endnodes and for
 764  * when we should not be forwarding packets from
 765  * this dest.
 766  */
 767 static int dn_blackhole(struct sk_buff *skb)
 768 {
 769         kfree_skb(skb);
 770         return NET_RX_DROP;
 771 }
 772
 773 /*
 774  * Used to catch bugs. This should never normally get
 775  * called.
 776  */
 777 static int dn_rt_bug(struct sk_buff *skb)
 778 {
 779         if (net_ratelimit()) {
 780                 struct dn_skb_cb *cb = DN_SKB_CB(skb);
 781
 782                 printk(KERN_DEBUG "dn_rt_bug: skb from:%04x to:%04x\n",
 783                                 dn_ntohs(cb->src), dn_ntohs(cb->dst));
 784         }
 785
 786         kfree_skb(skb);
 787
 788         return NET_RX_BAD;
 789 }
 790
 791 static int dn_rt_set_next_hop(struct dn_route *rt, struct dn_fib_res *res)
 792 {
 793         struct dn_fib_info *fi = res->fi;
 794         struct net_device *dev = rt->u.dst.dev;
 795         struct neighbour *n;
 796         unsigned mss;
 797
 798         if (fi) {
 799                 if (DN_FIB_RES_GW(*res) &&
 800                     DN_FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
 801                         rt->rt_gateway = DN_FIB_RES_GW(*res);
 802                 memcpy(rt->u.dst.metrics, fi->fib_metrics,
 803                        sizeof(rt->u.dst.metrics));
 804         }
 805         rt->rt_type = res->type;
 806
 807         if (dev != NULL && rt->u.dst.neighbour == NULL) {
 808                 n = __neigh_lookup_errno(&dn_neigh_table, &rt->rt_gateway, dev);
 809                 if (IS_ERR(n))
 810                         return PTR_ERR(n);
 811                 rt->u.dst.neighbour = n;
 812         }
 813
 814         if (rt->u.dst.metrics[RTAX_MTU-1] == 0 ||
 815             rt->u.dst.metrics[RTAX_MTU-1] > rt->u.dst.dev->mtu)
 816                 rt->u.dst.metrics[RTAX_MTU-1] = rt->u.dst.dev->mtu;
 817         mss = dn_mss_from_pmtu(dev, dst_mtu(&rt->u.dst));
 818         if (rt->u.dst.metrics[RTAX_ADVMSS-1] == 0 ||
 819             rt->u.dst.metrics[RTAX_ADVMSS-1] > mss)
 820                 rt->u.dst.metrics[RTAX_ADVMSS-1] = mss;
 821         return 0;
 822 }
 823
 824 static inline int dn_match_addr(__le16 addr1, __le16 addr2)
 825 {
 826         __u16 tmp = dn_ntohs(addr1) ^ dn_ntohs(addr2);
 827         int match = 16;
 828         while(tmp) {
 829                 tmp >>= 1;
 830                 match--;
 831         }
 832         return match;
 833 }
 834
 835 static __le16 dnet_select_source(const struct net_device *dev, __le16 daddr, int scope)
 836 {
 837         __le16 saddr = 0;
 838         struct dn_dev *dn_db = dev->dn_ptr;
 839         struct dn_ifaddr *ifa;
 840         int best_match = 0;
 841         int ret;
 842
 843         read_lock(&dev_base_lock);
 844         for(ifa = dn_db->ifa_list; ifa; ifa = ifa->ifa_next) {
 845                 if (ifa->ifa_scope > scope)
 846                         continue;
 847                 if (!daddr) {
 848                         saddr = ifa->ifa_local;
 849                         break;
 850                 }
 851                 ret = dn_match_addr(daddr, ifa->ifa_local);
 852                 if (ret > best_match)
 853                         saddr = ifa->ifa_local;
 854                 if (best_match == 0)
 855                         saddr = ifa->ifa_local;
 856         }
 857         read_unlock(&dev_base_lock);
 858
 859         return saddr;
 860 }
 861
 862 static inline __le16 __dn_fib_res_prefsrc(struct dn_fib_res *res)
 863 {
 864         return dnet_select_source(DN_FIB_RES_DEV(*res), DN_FIB_RES_GW(*res), res->scope);
 865 }
 866
 867 static inline __le16 dn_fib_rules_map_destination(__le16 daddr, struct dn_fib_res *res)
 868 {
 869         __le16 mask = dnet_make_mask(res->prefixlen);
 870         return (daddr&~mask)|res->fi->fib_nh->nh_gw;
 871 }
 872
 873 static int dn_route_output_slow(struct dst_entry **pprt, const struct flowi *oldflp, int try_hard)
 874 {
 875         struct flowi fl = { .nl_u = { .dn_u =
 876                                       { .daddr = oldflp->fld_dst,
 877                                         .saddr = oldflp->fld_src,
 878                                         .scope = RT_SCOPE_UNIVERSE,
 879 #ifdef CONFIG_DECNET_ROUTE_FWMARK
 880                                         .fwmark = oldflp->fld_fwmark
 881 #endif
 882                                      } },
 883                             .iif = loopback_dev.ifindex,
 884                             .oif = oldflp->oif };
 885         struct dn_route *rt = NULL;
 886         struct net_device *dev_out = NULL;
 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", dn_ntohs(oldflp->fld_dst),
 899                        dn_ntohs(oldflp->fld_src),
 900                        oldflp->fld_fwmark, 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(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                 read_lock(&dev_base_lock);
 925                 for(dev_out = dev_base; dev_out; dev_out = dev_out->next) {
 926                         if (!dev_out->dn_ptr)
 927                                 continue;
 928                         if (dn_dev_islocal(dev_out, oldflp->fld_src))
 929                                 break;
 930                 }
 931                 read_unlock(&dev_base_lock);
 932                 if (dev_out == NULL)
 933                         goto out;
 934                 dev_hold(dev_out);
 935 source_ok:
 936                 ;
 937         }
 938
 939         /* No destination? Assume its local */
 940         if (!fl.fld_dst) {
 941                 fl.fld_dst = fl.fld_src;
 942
 943                 err = -EADDRNOTAVAIL;
 944                 if (dev_out)
 945                         dev_put(dev_out);
 946                 dev_out = &loopback_dev;
 947                 dev_hold(dev_out);
 948                 if (!fl.fld_dst) {
 949                         fl.fld_dst =
 950                         fl.fld_src = dnet_select_source(dev_out, 0,
 951                                                        RT_SCOPE_HOST);
 952                         if (!fl.fld_dst)
 953                                 goto out;
 954                 }
 955                 fl.oif = loopback_dev.ifindex;
 956                 res.type = RTN_LOCAL;
 957                 goto make_route;
 958         }
 959
 960         if (decnet_debug_level & 16)
 961                 printk(KERN_DEBUG
 962                        "dn_route_output_slow: initial checks complete."
 963                        " dst=%o4x src=%04x oif=%d try_hard=%d\n",
 964                        dn_ntohs(fl.fld_dst), dn_ntohs(fl.fld_src),
 965                        fl.oif, try_hard);
 966
 967         /*
 968          * N.B. If the kernel is compiled without router support then
 969          * dn_fib_lookup() will evaluate to non-zero so this if () block
 970          * will always be executed.
 971          */
 972         err = -ESRCH;
 973         if (try_hard || (err = dn_fib_lookup(&fl, &res)) != 0) {
 974                 struct dn_dev *dn_db;
 975                 if (err != -ESRCH)
 976                         goto out;
 977                 /*
 978                  * Here the fallback is basically the standard algorithm for
 979                  * routing in endnodes which is described in the DECnet routing
 980                  * docs
 981                  *
 982                  * If we are not trying hard, look in neighbour cache.
 983                  * The result is tested to ensure that if a specific output
 984                  * device/source address was requested, then we honour that
 985                  * here
 986                  */
 987                 if (!try_hard) {
 988                         neigh = neigh_lookup_nodev(&dn_neigh_table, &fl.fld_dst);
 989                         if (neigh) {
 990                                 if ((oldflp->oif &&
 991                                     (neigh->dev->ifindex != oldflp->oif)) ||
 992                                     (oldflp->fld_src &&
 993                                     (!dn_dev_islocal(neigh->dev,
 994                                                       oldflp->fld_src)))) {
 995                                         neigh_release(neigh);
 996                                         neigh = NULL;
 997                                 } else {
 998                                         if (dev_out)
 999                                                 dev_put(dev_out);
1000                                         if (dn_dev_islocal(neigh->dev, fl.fld_dst)) {
1001                                                 dev_out = &loopback_dev;
1002                                                 res.type = RTN_LOCAL;
1003                                         } else {
1004                                                 dev_out = neigh->dev;
1005                                         }
1006                                         dev_hold(dev_out);
1007                                         goto select_source;
1008                                 }
1009                         }
1010                 }
1011
1012                 /* Not there? Perhaps its a local address */
1013                 if (dev_out == NULL)
1014                         dev_out = dn_dev_get_default();
1015                 err = -ENODEV;
1016                 if (dev_out == NULL)
1017                         goto out;
1018                 dn_db = dev_out->dn_ptr;
1019                 /* Possible improvement - check all devices for local addr */
1020                 if (dn_dev_islocal(dev_out, fl.fld_dst)) {
1021                         dev_put(dev_out);
1022                         dev_out = &loopback_dev;
1023                         dev_hold(dev_out);
1024                         res.type = RTN_LOCAL;
1025                         goto select_source;
1026                 }
1027                 /* Not local either.... try sending it to the default router */
1028                 neigh = neigh_clone(dn_db->router);
1029                 BUG_ON(neigh && neigh->dev != dev_out);
1030
1031                 /* Ok then, we assume its directly connected and move on */
1032 select_source:
1033                 if (neigh)
1034                         gateway = ((struct dn_neigh *)neigh)->addr;
1035                 if (gateway == 0)
1036                         gateway = fl.fld_dst;
1037                 if (fl.fld_src == 0) {
1038                         fl.fld_src = dnet_select_source(dev_out, gateway,
1039                                                          res.type == RTN_LOCAL ?
1040                                                          RT_SCOPE_HOST :
1041                                                          RT_SCOPE_LINK);
1042                         if (fl.fld_src == 0 && res.type != RTN_LOCAL)
1043                                 goto e_addr;
1044                 }
1045                 fl.oif = dev_out->ifindex;
1046                 goto make_route;
1047         }
1048         free_res = 1;
1049
1050         if (res.type == RTN_NAT)
1051                 goto e_inval;
1052
1053         if (res.type == RTN_LOCAL) {
1054                 if (!fl.fld_src)
1055                         fl.fld_src = fl.fld_dst;
1056                 if (dev_out)
1057                         dev_put(dev_out);
1058                 dev_out = &loopback_dev;
1059                 dev_hold(dev_out);
1060                 fl.oif = dev_out->ifindex;
1061                 if (res.fi)
1062                         dn_fib_info_put(res.fi);
1063                 res.fi = NULL;
1064                 goto make_route;
1065         }
1066
1067         if (res.fi->fib_nhs > 1 && fl.oif == 0)
1068                 dn_fib_select_multipath(&fl, &res);
1069
1070         /*
1071          * We could add some logic to deal with default routes here and
1072          * get rid of some of the special casing above.
1073          */
1074
1075         if (!fl.fld_src)
1076                 fl.fld_src = DN_FIB_RES_PREFSRC(res);
1077
1078         if (dev_out)
1079                 dev_put(dev_out);
1080         dev_out = DN_FIB_RES_DEV(res);
1081         dev_hold(dev_out);
1082         fl.oif = dev_out->ifindex;
1083         gateway = DN_FIB_RES_GW(res);
1084
1085 make_route:
1086         if (dev_out->flags & IFF_LOOPBACK)
1087                 flags |= RTCF_LOCAL;
1088
1089         rt = dst_alloc(&dn_dst_ops);
1090         if (rt == NULL)
1091                 goto e_nobufs;
1092
1093         atomic_set(&rt->u.dst.__refcnt, 1);
1094         rt->u.dst.flags   = DST_HOST;
1095
1096         rt->fl.fld_src    = oldflp->fld_src;
1097         rt->fl.fld_dst    = oldflp->fld_dst;
1098         rt->fl.oif        = oldflp->oif;
1099         rt->fl.iif        = 0;
1100 #ifdef CONFIG_DECNET_ROUTE_FWMARK
1101         rt->fl.fld_fwmark = oldflp->fld_fwmark;
1102 #endif
1103
1104         rt->rt_saddr      = fl.fld_src;
1105         rt->rt_daddr      = fl.fld_dst;
1106         rt->rt_gateway    = gateway ? gateway : fl.fld_dst;
1107         rt->rt_local_src  = fl.fld_src;
1108
1109         rt->rt_dst_map    = fl.fld_dst;
1110         rt->rt_src_map    = fl.fld_src;
1111
1112         rt->u.dst.dev = dev_out;
1113         dev_hold(dev_out);
1114         rt->u.dst.neighbour = neigh;
1115         neigh = NULL;
1116
1117         rt->u.dst.lastuse = jiffies;
1118         rt->u.dst.output  = dn_output;
1119         rt->u.dst.input   = dn_rt_bug;
1120         rt->rt_flags      = flags;
1121         if (flags & RTCF_LOCAL)
1122                 rt->u.dst.input = dn_nsp_rx;
1123
1124         err = dn_rt_set_next_hop(rt, &res);
1125         if (err)
1126                 goto e_neighbour;
1127
1128         hash = dn_hash(rt->fl.fld_src, rt->fl.fld_dst);
1129         dn_insert_route(rt, hash, (struct dn_route **)pprt);
1130
1131 done:
1132         if (neigh)
1133                 neigh_release(neigh);
1134         if (free_res)
1135                 dn_fib_res_put(&res);
1136         if (dev_out)
1137                 dev_put(dev_out);
1138 out:
1139         return err;
1140
1141 e_addr:
1142         err = -EADDRNOTAVAIL;
1143         goto done;
1144 e_inval:
1145         err = -EINVAL;
1146         goto done;
1147 e_nobufs:
1148         err = -ENOBUFS;
1149         goto done;
1150 e_neighbour:
1151         dst_free(&rt->u.dst);
1152         goto e_nobufs;
1153 }
1154
1155
1156 /*
1157  * N.B. The flags may be moved into the flowi at some future stage.
1158  */
1159 static int __dn_route_output_key(struct dst_entry **pprt, const struct flowi *flp, int flags)
1160 {
1161         unsigned hash = dn_hash(flp->fld_src, flp->fld_dst);
1162         struct dn_route *rt = NULL;
1163
1164         if (!(flags & MSG_TRYHARD)) {
1165                 rcu_read_lock_bh();
1166                 for(rt = rcu_dereference(dn_rt_hash_table[hash].chain); rt;
1167                         rt = rcu_dereference(rt->u.rt_next)) {
1168                         if ((flp->fld_dst == rt->fl.fld_dst) &&
1169                             (flp->fld_src == rt->fl.fld_src) &&
1170 #ifdef CONFIG_DECNET_ROUTE_FWMARK
1171                             (flp->fld_fwmark == rt->fl.fld_fwmark) &&
1172 #endif
1173                             (rt->fl.iif == 0) &&
1174                             (rt->fl.oif == flp->oif)) {
1175                                 rt->u.dst.lastuse = jiffies;
1176                                 dst_hold(&rt->u.dst);
1177                                 rt->u.dst.__use++;
1178                                 rcu_read_unlock_bh();
1179                                 *pprt = &rt->u.dst;
1180                                 return 0;
1181                         }
1182                 }
1183                 rcu_read_unlock_bh();
1184         }
1185
1186         return dn_route_output_slow(pprt, flp, flags);
1187 }
1188
1189 static int dn_route_output_key(struct dst_entry **pprt, struct flowi *flp, int flags)
1190 {
1191         int err;
1192
1193         err = __dn_route_output_key(pprt, flp, flags);
1194         if (err == 0 && flp->proto) {
1195                 err = xfrm_lookup(pprt, flp, NULL, 0);
1196         }
1197         return err;
1198 }
1199
1200 int dn_route_output_sock(struct dst_entry **pprt, struct flowi *fl, struct sock *sk, int flags)
1201 {
1202         int err;
1203
1204         err = __dn_route_output_key(pprt, fl, flags & MSG_TRYHARD);
1205         if (err == 0 && fl->proto) {
1206                 err = xfrm_lookup(pprt, fl, sk, !(flags & MSG_DONTWAIT));
1207         }
1208         return err;
1209 }
1210
1211 static int dn_route_input_slow(struct sk_buff *skb)
1212 {
1213         struct dn_route *rt = NULL;
1214         struct dn_skb_cb *cb = DN_SKB_CB(skb);
1215         struct net_device *in_dev = skb->dev;
1216         struct net_device *out_dev = NULL;
1217         struct dn_dev *dn_db;
1218         struct neighbour *neigh = NULL;
1219         unsigned hash;
1220         int flags = 0;
1221         __le16 gateway = 0;
1222         __le16 local_src = 0;
1223         struct flowi fl = { .nl_u = { .dn_u =
1224                                      { .daddr = cb->dst,
1225                                        .saddr = cb->src,
1226                                        .scope = RT_SCOPE_UNIVERSE,
1227 #ifdef CONFIG_DECNET_ROUTE_FWMARK
1228                                        .fwmark = skb->nfmark
1229 #endif
1230                                     } },
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                 flags |= RTCF_DIRECTSRC;
1268         } else {
1269                 __le16 src_map = fl.fld_src;
1270                 free_res = 1;
1271
1272                 out_dev = DN_FIB_RES_DEV(res);
1273                 if (out_dev == NULL) {
1274                         if (net_ratelimit())
1275                                 printk(KERN_CRIT "Bug in dn_route_input_slow() "
1276                                                  "No output device\n");
1277                         goto e_inval;
1278                 }
1279                 dev_hold(out_dev);
1280
1281                 if (res.r)
1282                         src_map = dn_fib_rules_policy(fl.fld_src, &res, &flags);
1283
1284                 gateway = DN_FIB_RES_GW(res);
1285                 if (res.type == RTN_NAT) {
1286                         fl.fld_dst = dn_fib_rules_map_destination(fl.fld_dst, &res);
1287                         dn_fib_res_put(&res);
1288                         free_res = 0;
1289                         if (dn_fib_lookup(&fl, &res))
1290                                 goto e_inval;
1291                         free_res = 1;
1292                         if (res.type != RTN_UNICAST)
1293                                 goto e_inval;
1294                         flags |= RTCF_DNAT;
1295                         gateway = fl.fld_dst;
1296                 }
1297                 fl.fld_src = src_map;
1298         }
1299
1300         switch(res.type) {
1301         case RTN_UNICAST:
1302                 /*
1303                  * Forwarding check here, we only check for forwarding
1304                  * being turned off, if you want to only forward intra
1305                  * area, its up to you to set the routing tables up
1306                  * correctly.
1307                  */
1308                 if (dn_db->parms.forwarding == 0)
1309                         goto e_inval;
1310
1311                 if (res.fi->fib_nhs > 1 && fl.oif == 0)
1312                         dn_fib_select_multipath(&fl, &res);
1313
1314                 /*
1315                  * Check for out_dev == in_dev. We use the RTCF_DOREDIRECT
1316                  * flag as a hint to set the intra-ethernet bit when
1317                  * forwarding. If we've got NAT in operation, we don't do
1318                  * this optimisation.
1319                  */
1320                 if (out_dev == in_dev && !(flags & RTCF_NAT))
1321                         flags |= RTCF_DOREDIRECT;
1322
1323                 local_src = DN_FIB_RES_PREFSRC(res);
1324
1325         case RTN_BLACKHOLE:
1326         case RTN_UNREACHABLE:
1327                 break;
1328         case RTN_LOCAL:
1329                 flags |= RTCF_LOCAL;
1330                 fl.fld_src = cb->dst;
1331                 fl.fld_dst = cb->src;
1332
1333                 /* Routing tables gave us a gateway */
1334                 if (gateway)
1335                         goto make_route;
1336
1337                 /* Packet was intra-ethernet, so we know its on-link */
1338                 if (cb->rt_flags | DN_RT_F_IE) {
1339                         gateway = cb->src;
1340                         flags |= RTCF_DIRECTSRC;
1341                         goto make_route;
1342                 }
1343
1344                 /* Use the default router if there is one */
1345                 neigh = neigh_clone(dn_db->router);
1346                 if (neigh) {
1347                         gateway = ((struct dn_neigh *)neigh)->addr;
1348                         goto make_route;
1349                 }
1350
1351                 /* Close eyes and pray */
1352                 gateway = cb->src;
1353                 flags |= RTCF_DIRECTSRC;
1354                 goto make_route;
1355         default:
1356                 goto e_inval;
1357         }
1358
1359 make_route:
1360         rt = dst_alloc(&dn_dst_ops);
1361         if (rt == NULL)
1362                 goto e_nobufs;
1363
1364         rt->rt_saddr      = fl.fld_src;
1365         rt->rt_daddr      = fl.fld_dst;
1366         rt->rt_gateway    = fl.fld_dst;
1367         if (gateway)
1368                 rt->rt_gateway = gateway;
1369         rt->rt_local_src  = local_src ? local_src : rt->rt_saddr;
1370
1371         rt->rt_dst_map    = fl.fld_dst;
1372         rt->rt_src_map    = fl.fld_src;
1373
1374         rt->fl.fld_src    = cb->src;
1375         rt->fl.fld_dst    = cb->dst;
1376         rt->fl.oif        = 0;
1377         rt->fl.iif        = in_dev->ifindex;
1378         rt->fl.fld_fwmark = fl.fld_fwmark;
1379
1380         rt->u.dst.flags = DST_HOST;
1381         rt->u.dst.neighbour = neigh;
1382         rt->u.dst.dev = out_dev;
1383         rt->u.dst.lastuse = jiffies;
1384         rt->u.dst.output = dn_rt_bug;
1385         switch(res.type) {
1386                 case RTN_UNICAST:
1387                         rt->u.dst.input = dn_forward;
1388                         break;
1389                 case RTN_LOCAL:
1390                         rt->u.dst.output = dn_output;
1391                         rt->u.dst.input = dn_nsp_rx;
1392                         rt->u.dst.dev = in_dev;
1393                         flags |= RTCF_LOCAL;
1394                         break;
1395                 default:
1396                 case RTN_UNREACHABLE:
1397                 case RTN_BLACKHOLE:
1398                         rt->u.dst.input = dn_blackhole;
1399         }
1400         rt->rt_flags = flags;
1401         if (rt->u.dst.dev)
1402                 dev_hold(rt->u.dst.dev);
1403
1404         err = dn_rt_set_next_hop(rt, &res);
1405         if (err)
1406                 goto e_neighbour;
1407
1408         hash = dn_hash(rt->fl.fld_src, rt->fl.fld_dst);
1409         dn_insert_route(rt, hash, (struct dn_route **)&skb->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 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)
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.rt_next)) {
1447                 if ((rt->fl.fld_src == cb->src) &&
1448                     (rt->fl.fld_dst == cb->dst) &&
1449                     (rt->fl.oif == 0) &&
1450 #ifdef CONFIG_DECNET_ROUTE_FWMARK
1451                     (rt->fl.fld_fwmark == skb->nfmark) &&
1452 #endif
1453                     (rt->fl.iif == cb->iif)) {
1454                         rt->u.dst.lastuse = jiffies;
1455                         dst_hold(&rt->u.dst);
1456                         rt->u.dst.__use++;
1457                         rcu_read_unlock();
1458                         skb->dst = (struct dst_entry *)rt;
1459                         return 0;
1460                 }
1461         }
1462         rcu_read_unlock();
1463
1464         return dn_route_input_slow(skb);
1465 }
1466
1467 static int dn_rt_fill_info(struct sk_buff *skb, u32 pid, u32 seq,
1468                            int event, int nowait, unsigned int flags)
1469 {
1470         struct dn_route *rt = (struct dn_route *)skb->dst;
1471         struct rtmsg *r;
1472         struct nlmsghdr *nlh;
1473         unsigned char *b = skb->tail;
1474         struct rta_cacheinfo ci;
1475
1476         nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*r), flags);
1477         r = NLMSG_DATA(nlh);
1478         r->rtm_family = AF_DECnet;
1479         r->rtm_dst_len = 16;
1480         r->rtm_src_len = 0;
1481         r->rtm_tos = 0;
1482         r->rtm_table = RT_TABLE_MAIN;
1483         r->rtm_type = rt->rt_type;
1484         r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
1485         r->rtm_scope = RT_SCOPE_UNIVERSE;
1486         r->rtm_protocol = RTPROT_UNSPEC;
1487         if (rt->rt_flags & RTCF_NOTIFY)
1488                 r->rtm_flags |= RTM_F_NOTIFY;
1489         RTA_PUT(skb, RTA_DST, 2, &rt->rt_daddr);
1490         if (rt->fl.fld_src) {
1491                 r->rtm_src_len = 16;
1492                 RTA_PUT(skb, RTA_SRC, 2, &rt->fl.fld_src);
1493         }
1494         if (rt->u.dst.dev)
1495                 RTA_PUT(skb, RTA_OIF, sizeof(int), &rt->u.dst.dev->ifindex);
1496         /*
1497          * Note to self - change this if input routes reverse direction when
1498          * they deal only with inputs and not with replies like they do
1499          * currently.
1500          */
1501         RTA_PUT(skb, RTA_PREFSRC, 2, &rt->rt_local_src);
1502         if (rt->rt_daddr != rt->rt_gateway)
1503                 RTA_PUT(skb, RTA_GATEWAY, 2, &rt->rt_gateway);
1504         if (rtnetlink_put_metrics(skb, rt->u.dst.metrics) < 0)
1505                 goto rtattr_failure;
1506         ci.rta_lastuse = jiffies_to_clock_t(jiffies - rt->u.dst.lastuse);
1507         ci.rta_used     = rt->u.dst.__use;
1508         ci.rta_clntref  = atomic_read(&rt->u.dst.__refcnt);
1509         if (rt->u.dst.expires)
1510                 ci.rta_expires = jiffies_to_clock_t(rt->u.dst.expires - jiffies);
1511         else
1512                 ci.rta_expires = 0;
1513         ci.rta_error    = rt->u.dst.error;
1514         ci.rta_id       = ci.rta_ts = ci.rta_tsage = 0;
1515         RTA_PUT(skb, RTA_CACHEINFO, sizeof(ci), &ci);
1516         if (rt->fl.iif)
1517                 RTA_PUT(skb, RTA_IIF, sizeof(int), &rt->fl.iif);
1518
1519         nlh->nlmsg_len = skb->tail - b;
1520         return skb->len;
1521
1522 nlmsg_failure:
1523 rtattr_failure:
1524         skb_trim(skb, b - skb->data);
1525         return -1;
1526 }
1527
1528 /*
1529  * This is called by both endnodes and routers now.
1530  */
1531 int dn_cache_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh, void *arg)
1532 {
1533         struct rtattr **rta = arg;
1534         struct rtmsg *rtm = NLMSG_DATA(nlh);
1535         struct dn_route *rt = NULL;
1536         struct dn_skb_cb *cb;
1537         int err;
1538         struct sk_buff *skb;
1539         struct flowi fl;
1540
1541         memset(&fl, 0, sizeof(fl));
1542         fl.proto = DNPROTO_NSP;
1543
1544         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1545         if (skb == NULL)
1546                 return -ENOBUFS;
1547         skb->mac.raw = skb->data;
1548         cb = DN_SKB_CB(skb);
1549
1550         if (rta[RTA_SRC-1])
1551                 memcpy(&fl.fld_src, RTA_DATA(rta[RTA_SRC-1]), 2);
1552         if (rta[RTA_DST-1])
1553                 memcpy(&fl.fld_dst, RTA_DATA(rta[RTA_DST-1]), 2);
1554         if (rta[RTA_IIF-1])
1555                 memcpy(&fl.iif, RTA_DATA(rta[RTA_IIF-1]), sizeof(int));
1556
1557         if (fl.iif) {
1558                 struct net_device *dev;
1559                 if ((dev = dev_get_by_index(fl.iif)) == NULL) {
1560                         kfree_skb(skb);
1561                         return -ENODEV;
1562                 }
1563                 if (!dev->dn_ptr) {
1564                         dev_put(dev);
1565                         kfree_skb(skb);
1566                         return -ENODEV;
1567                 }
1568                 skb->protocol = __constant_htons(ETH_P_DNA_RT);
1569                 skb->dev = dev;
1570                 cb->src = fl.fld_src;
1571                 cb->dst = fl.fld_dst;
1572                 local_bh_disable();
1573                 err = dn_route_input(skb);
1574                 local_bh_enable();
1575                 memset(cb, 0, sizeof(struct dn_skb_cb));
1576                 rt = (struct dn_route *)skb->dst;
1577                 if (!err && -rt->u.dst.error)
1578                         err = rt->u.dst.error;
1579         } else {
1580                 int oif = 0;
1581                 if (rta[RTA_OIF - 1])
1582                         memcpy(&oif, RTA_DATA(rta[RTA_OIF - 1]), sizeof(int));
1583                 fl.oif = oif;
1584                 err = dn_route_output_key((struct dst_entry **)&rt, &fl, 0);
1585         }
1586
1587         if (skb->dev)
1588                 dev_put(skb->dev);
1589         skb->dev = NULL;
1590         if (err)
1591                 goto out_free;
1592         skb->dst = &rt->u.dst;
1593         if (rtm->rtm_flags & RTM_F_NOTIFY)
1594                 rt->rt_flags |= RTCF_NOTIFY;
1595
1596         NETLINK_CB(skb).dst_pid = NETLINK_CB(in_skb).pid;
1597
1598         err = dn_rt_fill_info(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq, RTM_NEWROUTE, 0, 0);
1599
1600         if (err == 0)
1601                 goto out_free;
1602         if (err < 0) {
1603                 err = -EMSGSIZE;
1604                 goto out_free;
1605         }
1606
1607         err = netlink_unicast(rtnl, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1608
1609         return err;
1610
1611 out_free:
1612         kfree_skb(skb);
1613         return err;
1614 }
1615
1616 /*
1617  * For routers, this is called from dn_fib_dump, but for endnodes its
1618  * called directly from the rtnetlink dispatch table.
1619  */
1620 int dn_cache_dump(struct sk_buff *skb, struct netlink_callback *cb)
1621 {
1622         struct dn_route *rt;
1623         int h, s_h;
1624         int idx, s_idx;
1625
1626         if (NLMSG_PAYLOAD(cb->nlh, 0) < sizeof(struct rtmsg))
1627                 return -EINVAL;
1628         if (!(((struct rtmsg *)NLMSG_DATA(cb->nlh))->rtm_flags&RTM_F_CLONED))
1629                 return 0;
1630
1631         s_h = cb->args[0];
1632         s_idx = idx = cb->args[1];
1633         for(h = 0; h <= dn_rt_hash_mask; h++) {
1634                 if (h < s_h)
1635                         continue;
1636                 if (h > s_h)
1637                         s_idx = 0;
1638                 rcu_read_lock_bh();
1639                 for(rt = rcu_dereference(dn_rt_hash_table[h].chain), idx = 0;
1640                         rt;
1641                         rt = rcu_dereference(rt->u.rt_next), idx++) {
1642                         if (idx < s_idx)
1643                                 continue;
1644                         skb->dst = dst_clone(&rt->u.dst);
1645                         if (dn_rt_fill_info(skb, NETLINK_CB(cb->skb).pid,
1646                                         cb->nlh->nlmsg_seq, RTM_NEWROUTE,
1647                                         1, NLM_F_MULTI) <= 0) {
1648                                 dst_release(xchg(&skb->dst, NULL));
1649                                 rcu_read_unlock_bh();
1650                                 goto done;
1651                         }
1652                         dst_release(xchg(&skb->dst, NULL));
1653                 }
1654                 rcu_read_unlock_bh();
1655         }
1656
1657 done:
1658         cb->args[0] = h;
1659         cb->args[1] = idx;
1660         return skb->len;
1661 }
1662
1663 #ifdef CONFIG_PROC_FS
1664 struct dn_rt_cache_iter_state {
1665         int bucket;
1666 };
1667
1668 static struct dn_route *dn_rt_cache_get_first(struct seq_file *seq)
1669 {
1670         struct dn_route *rt = NULL;
1671         struct dn_rt_cache_iter_state *s = seq->private;
1672
1673         for(s->bucket = dn_rt_hash_mask; s->bucket >= 0; --s->bucket) {
1674                 rcu_read_lock_bh();
1675                 rt = dn_rt_hash_table[s->bucket].chain;
1676                 if (rt)
1677                         break;
1678                 rcu_read_unlock_bh();
1679         }
1680         return rt;
1681 }
1682
1683 static struct dn_route *dn_rt_cache_get_next(struct seq_file *seq, struct dn_route *rt)
1684 {
1685         struct dn_rt_cache_iter_state *s = rcu_dereference(seq->private);
1686
1687         rt = rt->u.rt_next;
1688         while(!rt) {
1689                 rcu_read_unlock_bh();
1690                 if (--s->bucket < 0)
1691                         break;
1692                 rcu_read_lock_bh();
1693                 rt = dn_rt_hash_table[s->bucket].chain;
1694         }
1695         return rt;
1696 }
1697
1698 static void *dn_rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
1699 {
1700         struct dn_route *rt = dn_rt_cache_get_first(seq);
1701
1702         if (rt) {
1703                 while(*pos && (rt = dn_rt_cache_get_next(seq, rt)))
1704                         --*pos;
1705         }
1706         return *pos ? NULL : rt;
1707 }
1708
1709 static void *dn_rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1710 {
1711         struct dn_route *rt = dn_rt_cache_get_next(seq, v);
1712         ++*pos;
1713         return rt;
1714 }
1715
1716 static void dn_rt_cache_seq_stop(struct seq_file *seq, void *v)
1717 {
1718         if (v)
1719                 rcu_read_unlock_bh();
1720 }
1721
1722 static int dn_rt_cache_seq_show(struct seq_file *seq, void *v)
1723 {
1724         struct dn_route *rt = v;
1725         char buf1[DN_ASCBUF_LEN], buf2[DN_ASCBUF_LEN];
1726
1727         seq_printf(seq, "%-8s %-7s %-7s %04d %04d %04d\n",
1728                         rt->u.dst.dev ? rt->u.dst.dev->name : "*",
1729                         dn_addr2asc(dn_ntohs(rt->rt_daddr), buf1),
1730                         dn_addr2asc(dn_ntohs(rt->rt_saddr), buf2),
1731                         atomic_read(&rt->u.dst.__refcnt),
1732                         rt->u.dst.__use,
1733                         (int) dst_metric(&rt->u.dst, RTAX_RTT));
1734         return 0;
1735 }
1736
1737 static struct seq_operations dn_rt_cache_seq_ops = {
1738         .start  = dn_rt_cache_seq_start,
1739         .next   = dn_rt_cache_seq_next,
1740         .stop   = dn_rt_cache_seq_stop,
1741         .show   = dn_rt_cache_seq_show,
1742 };
1743
1744 static int dn_rt_cache_seq_open(struct inode *inode, struct file *file)
1745 {
1746         struct seq_file *seq;
1747         int rc = -ENOMEM;
1748         struct dn_rt_cache_iter_state *s = kmalloc(sizeof(*s), GFP_KERNEL);
1749
1750         if (!s)
1751                 goto out;
1752         rc = seq_open(file, &dn_rt_cache_seq_ops);
1753         if (rc)
1754                 goto out_kfree;
1755         seq             = file->private_data;
1756         seq->private    = s;
1757         memset(s, 0, sizeof(*s));
1758 out:
1759         return rc;
1760 out_kfree:
1761         kfree(s);
1762         goto out;
1763 }
1764
1765 static struct file_operations dn_rt_cache_seq_fops = {
1766         .owner   = THIS_MODULE,
1767         .open    = dn_rt_cache_seq_open,
1768         .read    = seq_read,
1769         .llseek  = seq_lseek,
1770         .release = seq_release_private,
1771 };
1772
1773 #endif /* CONFIG_PROC_FS */
1774
1775 void __init dn_route_init(void)
1776 {
1777         int i, goal, order;
1778
1779         dn_dst_ops.kmem_cachep = kmem_cache_create("dn_dst_cache",
1780                                                    sizeof(struct dn_route),
1781                                                    0, SLAB_HWCACHE_ALIGN,
1782                                                    NULL, NULL);
1783
1784         if (!dn_dst_ops.kmem_cachep)
1785                 panic("DECnet: Failed to allocate dn_dst_cache\n");
1786
1787         init_timer(&dn_route_timer);
1788         dn_route_timer.function = dn_dst_check_expire;
1789         dn_route_timer.expires = jiffies + decnet_dst_gc_interval * HZ;
1790         add_timer(&dn_route_timer);
1791
1792         goal = num_physpages >> (26 - PAGE_SHIFT);
1793
1794         for(order = 0; (1UL << order) < goal; order++)
1795                 /* NOTHING */;
1796
1797         /*
1798          * Only want 1024 entries max, since the table is very, very unlikely
1799          * to be larger than that.
1800          */
1801         while(order && ((((1UL << order) * PAGE_SIZE) /
1802                                 sizeof(struct dn_rt_hash_bucket)) >= 2048))
1803                 order--;
1804
1805         do {
1806                 dn_rt_hash_mask = (1UL << order) * PAGE_SIZE /
1807                         sizeof(struct dn_rt_hash_bucket);
1808                 while(dn_rt_hash_mask & (dn_rt_hash_mask - 1))
1809                         dn_rt_hash_mask--;
1810                 dn_rt_hash_table = (struct dn_rt_hash_bucket *)
1811                         __get_free_pages(GFP_ATOMIC, order);
1812         } while (dn_rt_hash_table == NULL && --order > 0);
1813
1814         if (!dn_rt_hash_table)
1815                 panic("Failed to allocate DECnet route cache hash table\n");
1816
1817         printk(KERN_INFO
1818                 "DECnet: Routing cache hash table of %u buckets, %ldKbytes\n",
1819                 dn_rt_hash_mask,
1820                 (long)(dn_rt_hash_mask*sizeof(struct dn_rt_hash_bucket))/1024);
1821
1822         dn_rt_hash_mask--;
1823         for(i = 0; i <= dn_rt_hash_mask; i++) {
1824                 spin_lock_init(&dn_rt_hash_table[i].lock);
1825                 dn_rt_hash_table[i].chain = NULL;
1826         }
1827
1828         dn_dst_ops.gc_thresh = (dn_rt_hash_mask + 1);
1829
1830         proc_net_fops_create("decnet_cache", S_IRUGO, &dn_rt_cache_seq_fops);
1831 }
1832
1833 void __exit dn_route_cleanup(void)
1834 {
1835         del_timer(&dn_route_timer);
1836         dn_run_flush(0);
1837
1838         proc_net_remove("decnet_cache");
1839 }
1840