git.oblomov.eu Git - linux-2.6/blob - 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/net_namespace.h>
  81 #include <net/netlink.h>
  82 #include <net/neighbour.h>
  83 #include <net/dst.h>
  84 #include <net/flow.h>
  85 #include <net/fib_rules.h>
  86 #include <net/dn.h>
  87 #include <net/dn_dev.h>
  88 #include <net/dn_nsp.h>
  89 #include <net/dn_route.h>
  90 #include <net/dn_neigh.h>
  91 #include <net/dn_fib.h>
  92
  93 struct dn_rt_hash_bucket
  94 {
  95         struct dn_route *chain;
  96         spinlock_t lock;
  97 };
  98
  99 extern struct neigh_table dn_neigh_table;
 100
 101
 102 static unsigned char dn_hiord_addr[6] = {0xAA,0x00,0x04,0x00,0x00,0x00};
 103
 104 static const int dn_rt_min_delay = 2 * HZ;
 105 static const int dn_rt_max_delay = 10 * HZ;
 106 static const int dn_rt_mtu_expires = 10 * 60 * HZ;
 107
 108 static unsigned long dn_rt_deadline;
 109
 110 static int dn_dst_gc(struct dst_ops *ops);
 111 static struct dst_entry *dn_dst_check(struct dst_entry *, __u32);
 112 static struct dst_entry *dn_dst_negative_advice(struct dst_entry *);
 113 static void dn_dst_link_failure(struct sk_buff *);
 114 static void dn_dst_update_pmtu(struct dst_entry *dst, u32 mtu);
 115 static int dn_route_input(struct sk_buff *);
 116 static void dn_run_flush(unsigned long dummy);
 117
 118 static struct dn_rt_hash_bucket *dn_rt_hash_table;
 119 static unsigned dn_rt_hash_mask;
 120
 121 static struct timer_list dn_route_timer;
 122 static DEFINE_TIMER(dn_rt_flush_timer, dn_run_flush, 0, 0);
 123 int decnet_dst_gc_interval = 2;
 124
 125 static struct dst_ops dn_dst_ops = {
 126         .family =               PF_DECnet,
 127         .protocol =             __constant_htons(ETH_P_DNA_RT),
 128         .gc_thresh =            128,
 129         .gc =                   dn_dst_gc,
 130         .check =                dn_dst_check,
 131         .negative_advice =      dn_dst_negative_advice,
 132         .link_failure =         dn_dst_link_failure,
 133         .update_pmtu =          dn_dst_update_pmtu,
 134         .entries =              ATOMIC_INIT(0),
 135 };
 136
 137 static __inline__ unsigned dn_hash(__le16 src, __le16 dst)
 138 {
 139         __u16 tmp = (__u16 __force)(src ^ dst);
 140         tmp ^= (tmp >> 3);
 141         tmp ^= (tmp >> 5);
 142         tmp ^= (tmp >> 10);
 143         return dn_rt_hash_mask & (unsigned)tmp;
 144 }
 145
 146 static inline void dnrt_free(struct dn_route *rt)
 147 {
 148         call_rcu_bh(&rt->u.dst.rcu_head, dst_rcu_free);
 149 }
 150
 151 static inline void dnrt_drop(struct dn_route *rt)
 152 {
 153         dst_release(&rt->u.dst);
 154         call_rcu_bh(&rt->u.dst.rcu_head, dst_rcu_free);
 155 }
 156
 157 static void dn_dst_check_expire(unsigned long dummy)
 158 {
 159         int i;
 160         struct dn_route *rt, **rtp;
 161         unsigned long now = jiffies;
 162         unsigned long expire = 120 * HZ;
 163
 164         for(i = 0; i <= dn_rt_hash_mask; i++) {
 165                 rtp = &dn_rt_hash_table[i].chain;
 166
 167                 spin_lock(&dn_rt_hash_table[i].lock);
 168                 while((rt=*rtp) != NULL) {
 169                         if (atomic_read(&rt->u.dst.__refcnt) ||
 170                                         (now - rt->u.dst.lastuse) < expire) {
 171                                 rtp = &rt->u.dst.dn_next;
 172                                 continue;
 173                         }
 174                         *rtp = rt->u.dst.dn_next;
 175                         rt->u.dst.dn_next = NULL;
 176                         dnrt_free(rt);
 177                 }
 178                 spin_unlock(&dn_rt_hash_table[i].lock);
 179
 180                 if ((jiffies - now) > 0)
 181                         break;
 182         }
 183
 184         mod_timer(&dn_route_timer, now + decnet_dst_gc_interval * HZ);
 185 }
 186
 187 static int dn_dst_gc(struct dst_ops *ops)
 188 {
 189         struct dn_route *rt, **rtp;
 190         int i;
 191         unsigned long now = jiffies;
 192         unsigned long expire = 10 * HZ;
 193
 194         for(i = 0; i <= dn_rt_hash_mask; i++) {
 195
 196                 spin_lock_bh(&dn_rt_hash_table[i].lock);
 197                 rtp = &dn_rt_hash_table[i].chain;
 198
 199                 while((rt=*rtp) != NULL) {
 200                         if (atomic_read(&rt->u.dst.__refcnt) ||
 201                                         (now - rt->u.dst.lastuse) < expire) {
 202                                 rtp = &rt->u.dst.dn_next;
 203                                 continue;
 204                         }
 205                         *rtp = rt->u.dst.dn_next;
 206                         rt->u.dst.dn_next = NULL;
 207                         dnrt_drop(rt);
 208                         break;
 209                 }
 210                 spin_unlock_bh(&dn_rt_hash_table[i].lock);
 211         }
 212
 213         return 0;
 214 }
 215
 216 /*
 217  * The decnet standards don't impose a particular minimum mtu, what they
 218  * do insist on is that the routing layer accepts a datagram of at least
 219  * 230 bytes long. Here we have to subtract the routing header length from
 220  * 230 to get the minimum acceptable mtu. If there is no neighbour, then we
 221  * assume the worst and use a long header size.
 222  *
 223  * We update both the mtu and the advertised mss (i.e. the segment size we
 224  * advertise to the other end).
 225  */
 226 static void dn_dst_update_pmtu(struct dst_entry *dst, u32 mtu)
 227 {
 228         u32 min_mtu = 230;
 229         struct dn_dev *dn = dst->neighbour ?
 230                             (struct dn_dev *)dst->neighbour->dev->dn_ptr : NULL;
 231
 232         if (dn && dn->use_long == 0)
 233                 min_mtu -= 6;
 234         else
 235                 min_mtu -= 21;
 236
 237         if (dst_metric(dst, RTAX_MTU) > mtu && mtu >= min_mtu) {
 238                 if (!(dst_metric_locked(dst, RTAX_MTU))) {
 239                         dst->metrics[RTAX_MTU-1] = mtu;
 240                         dst_set_expires(dst, dn_rt_mtu_expires);
 241                 }
 242                 if (!(dst_metric_locked(dst, RTAX_ADVMSS))) {
 243                         u32 mss = mtu - DN_MAX_NSP_DATA_HEADER;
 244                         if (dst_metric(dst, RTAX_ADVMSS) > mss)
 245                                 dst->metrics[RTAX_ADVMSS-1] = mss;
 246                 }
 247         }
 248 }
 249
 250 /*
 251  * When a route has been marked obsolete. (e.g. routing cache flush)
 252  */
 253 static struct dst_entry *dn_dst_check(struct dst_entry *dst, __u32 cookie)
 254 {
 255         return NULL;
 256 }
 257
 258 static struct dst_entry *dn_dst_negative_advice(struct dst_entry *dst)
 259 {
 260         dst_release(dst);
 261         return NULL;
 262 }
 263
 264 static void dn_dst_link_failure(struct sk_buff *skb)
 265 {
 266         return;
 267 }
 268
 269 static inline int compare_keys(struct flowi *fl1, struct flowi *fl2)
 270 {
 271         return ((fl1->nl_u.dn_u.daddr ^ fl2->nl_u.dn_u.daddr) |
 272                 (fl1->nl_u.dn_u.saddr ^ fl2->nl_u.dn_u.saddr) |
 273                 (fl1->mark ^ fl2->mark) |
 274                 (fl1->nl_u.dn_u.scope ^ fl2->nl_u.dn_u.scope) |
 275                 (fl1->oif ^ fl2->oif) |
 276                 (fl1->iif ^ fl2->iif)) == 0;
 277 }
 278
 279 static int dn_insert_route(struct dn_route *rt, unsigned hash, struct dn_route **rp)
 280 {
 281         struct dn_route *rth, **rthp;
 282         unsigned long now = jiffies;
 283
 284         rthp = &dn_rt_hash_table[hash].chain;
 285
 286         spin_lock_bh(&dn_rt_hash_table[hash].lock);
 287         while((rth = *rthp) != NULL) {
 288                 if (compare_keys(&rth->fl, &rt->fl)) {
 289                         /* Put it first */
 290                         *rthp = rth->u.dst.dn_next;
 291                         rcu_assign_pointer(rth->u.dst.dn_next,
 292                                            dn_rt_hash_table[hash].chain);
 293                         rcu_assign_pointer(dn_rt_hash_table[hash].chain, rth);
 294
 295                         dst_use(&rth->u.dst, now);
 296                         spin_unlock_bh(&dn_rt_hash_table[hash].lock);
 297
 298                         dnrt_drop(rt);
 299                         *rp = rth;
 300                         return 0;
 301                 }
 302                 rthp = &rth->u.dst.dn_next;
 303         }
 304
 305         rcu_assign_pointer(rt->u.dst.dn_next, dn_rt_hash_table[hash].chain);
 306         rcu_assign_pointer(dn_rt_hash_table[hash].chain, rt);
 307
 308         dst_use(&rt->u.dst, now);
 309         spin_unlock_bh(&dn_rt_hash_table[hash].lock);
 310         *rp = rt;
 311         return 0;
 312 }
 313
 314 void dn_run_flush(unsigned long dummy)
 315 {
 316         int i;
 317         struct dn_route *rt, *next;
 318
 319         for(i = 0; i < dn_rt_hash_mask; i++) {
 320                 spin_lock_bh(&dn_rt_hash_table[i].lock);
 321
 322                 if ((rt = xchg(&dn_rt_hash_table[i].chain, NULL)) == NULL)
 323                         goto nothing_to_declare;
 324
 325                 for(; rt; rt=next) {
 326                         next = rt->u.dst.dn_next;
 327                         rt->u.dst.dn_next = NULL;
 328                         dst_free((struct dst_entry *)rt);
 329                 }
 330
 331 nothing_to_declare:
 332                 spin_unlock_bh(&dn_rt_hash_table[i].lock);
 333         }
 334 }
 335
 336 static DEFINE_SPINLOCK(dn_rt_flush_lock);
 337
 338 void dn_rt_cache_flush(int delay)
 339 {
 340         unsigned long now = jiffies;
 341         int user_mode = !in_interrupt();
 342
 343         if (delay < 0)
 344                 delay = dn_rt_min_delay;
 345
 346         spin_lock_bh(&dn_rt_flush_lock);
 347
 348         if (del_timer(&dn_rt_flush_timer) && delay > 0 && dn_rt_deadline) {
 349                 long tmo = (long)(dn_rt_deadline - now);
 350
 351                 if (user_mode && tmo < dn_rt_max_delay - dn_rt_min_delay)
 352                         tmo = 0;
 353
 354                 if (delay > tmo)
 355                         delay = tmo;
 356         }
 357
 358         if (delay <= 0) {
 359                 spin_unlock_bh(&dn_rt_flush_lock);
 360                 dn_run_flush(0);
 361                 return;
 362         }
 363
 364         if (dn_rt_deadline == 0)
 365                 dn_rt_deadline = now + dn_rt_max_delay;
 366
 367         dn_rt_flush_timer.expires = now + delay;
 368         add_timer(&dn_rt_flush_timer);
 369         spin_unlock_bh(&dn_rt_flush_lock);
 370 }
 371
 372 /**
 373  * dn_return_short - Return a short packet to its sender
 374  * @skb: The packet to return
 375  *
 376  */
 377 static int dn_return_short(struct sk_buff *skb)
 378 {
 379         struct dn_skb_cb *cb;
 380         unsigned char *ptr;
 381         __le16 *src;
 382         __le16 *dst;
 383         __le16 tmp;
 384
 385         /* Add back headers */
 386         skb_push(skb, skb->data - skb_network_header(skb));
 387
 388         if ((skb = skb_unshare(skb, GFP_ATOMIC)) == NULL)
 389                 return NET_RX_DROP;
 390
 391         cb = DN_SKB_CB(skb);
 392         /* Skip packet length and point to flags */
 393         ptr = skb->data + 2;
 394         *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
 395
 396         dst = (__le16 *)ptr;
 397         ptr += 2;
 398         src = (__le16 *)ptr;
 399         ptr += 2;
 400         *ptr = 0; /* Zero hop count */
 401
 402         /* Swap source and destination */
 403         tmp  = *src;
 404         *src = *dst;
 405         *dst = tmp;
 406
 407         skb->pkt_type = PACKET_OUTGOING;
 408         dn_rt_finish_output(skb, NULL, NULL);
 409         return NET_RX_SUCCESS;
 410 }
 411
 412 /**
 413  * dn_return_long - Return a long packet to its sender
 414  * @skb: The long format packet to return
 415  *
 416  */
 417 static int dn_return_long(struct sk_buff *skb)
 418 {
 419         struct dn_skb_cb *cb;
 420         unsigned char *ptr;
 421         unsigned char *src_addr, *dst_addr;
 422         unsigned char tmp[ETH_ALEN];
 423
 424         /* Add back all headers */
 425         skb_push(skb, skb->data - skb_network_header(skb));
 426
 427         if ((skb = skb_unshare(skb, GFP_ATOMIC)) == NULL)
 428                 return NET_RX_DROP;
 429
 430         cb = DN_SKB_CB(skb);
 431         /* Ignore packet length and point to flags */
 432         ptr = skb->data + 2;
 433
 434         /* Skip padding */
 435         if (*ptr & DN_RT_F_PF) {
 436                 char padlen = (*ptr & ~DN_RT_F_PF);
 437                 ptr += padlen;
 438         }
 439
 440         *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
 441         ptr += 2;
 442         dst_addr = ptr;
 443         ptr += 8;
 444         src_addr = ptr;
 445         ptr += 6;
 446         *ptr = 0; /* Zero hop count */
 447
 448         /* Swap source and destination */
 449         memcpy(tmp, src_addr, ETH_ALEN);
 450         memcpy(src_addr, dst_addr, ETH_ALEN);
 451         memcpy(dst_addr, tmp, ETH_ALEN);
 452
 453         skb->pkt_type = PACKET_OUTGOING;
 454         dn_rt_finish_output(skb, dst_addr, src_addr);
 455         return NET_RX_SUCCESS;
 456 }
 457
 458 /**
 459  * dn_route_rx_packet - Try and find a route for an incoming packet
 460  * @skb: The packet to find a route for
 461  *
 462  * Returns: result of input function if route is found, error code otherwise
 463  */
 464 static int dn_route_rx_packet(struct sk_buff *skb)
 465 {
 466         struct dn_skb_cb *cb = DN_SKB_CB(skb);
 467         int err;
 468
 469         if ((err = dn_route_input(skb)) == 0)
 470                 return dst_input(skb);
 471
 472         if (decnet_debug_level & 4) {
 473                 char *devname = skb->dev ? skb->dev->name : "???";
 474                 struct dn_skb_cb *cb = DN_SKB_CB(skb);
 475                 printk(KERN_DEBUG
 476                         "DECnet: dn_route_rx_packet: rt_flags=0x%02x dev=%s len=%d src=0x%04hx dst=0x%04hx err=%d type=%d\n",
 477                         (int)cb->rt_flags, devname, skb->len,
 478                         le16_to_cpu(cb->src), le16_to_cpu(cb->dst),
 479                         err, skb->pkt_type);
 480         }
 481
 482         if ((skb->pkt_type == PACKET_HOST) && (cb->rt_flags & DN_RT_F_RQR)) {
 483                 switch(cb->rt_flags & DN_RT_PKT_MSK) {
 484                         case DN_RT_PKT_SHORT:
 485                                 return dn_return_short(skb);
 486                         case DN_RT_PKT_LONG:
 487                                 return dn_return_long(skb);
 488                 }
 489         }
 490
 491         kfree_skb(skb);
 492         return NET_RX_DROP;
 493 }
 494
 495 static int dn_route_rx_long(struct sk_buff *skb)
 496 {
 497         struct dn_skb_cb *cb = DN_SKB_CB(skb);
 498         unsigned char *ptr = skb->data;
 499
 500         if (!pskb_may_pull(skb, 21)) /* 20 for long header, 1 for shortest nsp */
 501                 goto drop_it;
 502
 503         skb_pull(skb, 20);
 504         skb_reset_transport_header(skb);
 505
 506         /* Destination info */
 507         ptr += 2;
 508         cb->dst = dn_eth2dn(ptr);
 509         if (memcmp(ptr, dn_hiord_addr, 4) != 0)
 510                 goto drop_it;
 511         ptr += 6;
 512
 513
 514         /* Source info */
 515         ptr += 2;
 516         cb->src = dn_eth2dn(ptr);
 517         if (memcmp(ptr, dn_hiord_addr, 4) != 0)
 518                 goto drop_it;
 519         ptr += 6;
 520         /* Other junk */
 521         ptr++;
 522         cb->hops = *ptr++; /* Visit Count */
 523
 524         return NF_HOOK(PF_DECnet, NF_DN_PRE_ROUTING, skb, skb->dev, NULL, dn_route_rx_packet);
 525
 526 drop_it:
 527         kfree_skb(skb);
 528         return NET_RX_DROP;
 529 }
 530
 531
 532
 533 static int dn_route_rx_short(struct sk_buff *skb)
 534 {
 535         struct dn_skb_cb *cb = DN_SKB_CB(skb);
 536         unsigned char *ptr = skb->data;
 537
 538         if (!pskb_may_pull(skb, 6)) /* 5 for short header + 1 for shortest nsp */
 539                 goto drop_it;
 540
 541         skb_pull(skb, 5);
 542         skb_reset_transport_header(skb);
 543
 544         cb->dst = *(__le16 *)ptr;
 545         ptr += 2;
 546         cb->src = *(__le16 *)ptr;
 547         ptr += 2;
 548         cb->hops = *ptr & 0x3f;
 549
 550         return NF_HOOK(PF_DECnet, NF_DN_PRE_ROUTING, skb, skb->dev, NULL, dn_route_rx_packet);
 551
 552 drop_it:
 553         kfree_skb(skb);
 554         return NET_RX_DROP;
 555 }
 556
 557 static int dn_route_discard(struct sk_buff *skb)
 558 {
 559         /*
 560          * I know we drop the packet here, but thats considered success in
 561          * this case
 562          */
 563         kfree_skb(skb);
 564         return NET_RX_SUCCESS;
 565 }
 566
 567 static int dn_route_ptp_hello(struct sk_buff *skb)
 568 {
 569         dn_dev_hello(skb);
 570         dn_neigh_pointopoint_hello(skb);
 571         return NET_RX_SUCCESS;
 572 }
 573
 574 int dn_route_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
 575 {
 576         struct dn_skb_cb *cb;
 577         unsigned char flags = 0;
 578         __u16 len = le16_to_cpu(*(__le16 *)skb->data);
 579         struct dn_dev *dn = (struct dn_dev *)dev->dn_ptr;
 580         unsigned char padlen = 0;
 581
 582         if (!net_eq(dev_net(dev), &init_net))
 583                 goto dump_it;
 584
 585         if (dn == NULL)
 586                 goto dump_it;
 587
 588         if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
 589                 goto out;
 590
 591         if (!pskb_may_pull(skb, 3))
 592                 goto dump_it;
 593
 594         skb_pull(skb, 2);
 595
 596         if (len > skb->len)
 597                 goto dump_it;
 598
 599         skb_trim(skb, len);
 600
 601         flags = *skb->data;
 602
 603         cb = DN_SKB_CB(skb);
 604         cb->stamp = jiffies;
 605         cb->iif = dev->ifindex;
 606
 607         /*
 608          * If we have padding, remove it.
 609          */
 610         if (flags & DN_RT_F_PF) {
 611                 padlen = flags & ~DN_RT_F_PF;
 612                 if (!pskb_may_pull(skb, padlen + 1))
 613                         goto dump_it;
 614                 skb_pull(skb, padlen);
 615                 flags = *skb->data;
 616         }
 617
 618         skb_reset_network_header(skb);
 619
 620         /*
 621          * Weed out future version DECnet
 622          */
 623         if (flags & DN_RT_F_VER)
 624                 goto dump_it;
 625
 626         cb->rt_flags = flags;
 627
 628         if (decnet_debug_level & 1)
 629                 printk(KERN_DEBUG
 630                         "dn_route_rcv: got 0x%02x from %s [%d %d %d]\n",
 631                         (int)flags, (dev) ? dev->name : "???", len, skb->len,
 632                         padlen);
 633
 634         if (flags & DN_RT_PKT_CNTL) {
 635                 if (unlikely(skb_linearize(skb)))
 636                         goto dump_it;
 637
 638                 switch(flags & DN_RT_CNTL_MSK) {
 639                         case DN_RT_PKT_INIT:
 640                                 dn_dev_init_pkt(skb);
 641                                 break;
 642                         case DN_RT_PKT_VERI:
 643                                 dn_dev_veri_pkt(skb);
 644                                 break;
 645                 }
 646
 647                 if (dn->parms.state != DN_DEV_S_RU)
 648                         goto dump_it;
 649
 650                 switch(flags & DN_RT_CNTL_MSK) {
 651                         case DN_RT_PKT_HELO:
 652                                 return NF_HOOK(PF_DECnet, NF_DN_HELLO, skb, skb->dev, NULL, dn_route_ptp_hello);
 653
 654                         case DN_RT_PKT_L1RT:
 655                         case DN_RT_PKT_L2RT:
 656                                 return NF_HOOK(PF_DECnet, NF_DN_ROUTE, skb, skb->dev, NULL, dn_route_discard);
 657                         case DN_RT_PKT_ERTH:
 658                                 return NF_HOOK(PF_DECnet, NF_DN_HELLO, skb, skb->dev, NULL, dn_neigh_router_hello);
 659
 660                         case DN_RT_PKT_EEDH:
 661                                 return NF_HOOK(PF_DECnet, NF_DN_HELLO, skb, skb->dev, NULL, dn_neigh_endnode_hello);
 662                 }
 663         } else {
 664                 if (dn->parms.state != DN_DEV_S_RU)
 665                         goto dump_it;
 666
 667                 skb_pull(skb, 1); /* Pull flags */
 668
 669                 switch(flags & DN_RT_PKT_MSK) {
 670                         case DN_RT_PKT_LONG:
 671                                 return dn_route_rx_long(skb);
 672                         case DN_RT_PKT_SHORT:
 673                                 return dn_route_rx_short(skb);
 674                 }
 675         }
 676
 677 dump_it:
 678         kfree_skb(skb);
 679 out:
 680         return NET_RX_DROP;
 681 }
 682
 683 static int dn_output(struct sk_buff *skb)
 684 {
 685         struct dst_entry *dst = skb->dst;
 686         struct dn_route *rt = (struct dn_route *)dst;
 687         struct net_device *dev = dst->dev;
 688         struct dn_skb_cb *cb = DN_SKB_CB(skb);
 689         struct neighbour *neigh;
 690
 691         int err = -EINVAL;
 692
 693         if ((neigh = dst->neighbour) == NULL)
 694                 goto error;
 695
 696         skb->dev = dev;
 697
 698         cb->src = rt->rt_saddr;
 699         cb->dst = rt->rt_daddr;
 700
 701         /*
 702          * Always set the Intra-Ethernet bit on all outgoing packets
 703          * originated on this node. Only valid flag from upper layers
 704          * is return-to-sender-requested. Set hop count to 0 too.
 705          */
 706         cb->rt_flags &= ~DN_RT_F_RQR;
 707         cb->rt_flags |= DN_RT_F_IE;
 708         cb->hops = 0;
 709
 710         return NF_HOOK(PF_DECnet, NF_DN_LOCAL_OUT, skb, NULL, dev, neigh->output);
 711
 712 error:
 713         if (net_ratelimit())
 714                 printk(KERN_DEBUG "dn_output: This should not happen\n");
 715
 716         kfree_skb(skb);
 717
 718         return err;
 719 }
 720
 721 static int dn_forward(struct sk_buff *skb)
 722 {
 723         struct dn_skb_cb *cb = DN_SKB_CB(skb);
 724         struct dst_entry *dst = skb->dst;
 725         struct dn_dev *dn_db = dst->dev->dn_ptr;
 726         struct dn_route *rt;
 727         struct neighbour *neigh = dst->neighbour;
 728         int header_len;
 729 #ifdef CONFIG_NETFILTER
 730         struct net_device *dev = skb->dev;
 731 #endif
 732
 733         if (skb->pkt_type != PACKET_HOST)
 734                 goto drop;
 735
 736         /* Ensure that we have enough space for headers */
 737         rt = (struct dn_route *)skb->dst;
 738         header_len = dn_db->use_long ? 21 : 6;
 739         if (skb_cow(skb, LL_RESERVED_SPACE(rt->u.dst.dev)+header_len))
 740                 goto drop;
 741
 742         /*
 743          * Hop count exceeded.
 744          */
 745         if (++cb->hops > 30)
 746                 goto drop;
 747
 748         skb->dev = rt->u.dst.dev;
 749
 750         /*
 751          * If packet goes out same interface it came in on, then set
 752          * the Intra-Ethernet bit. This has no effect for short
 753          * packets, so we don't need to test for them here.
 754          */
 755         cb->rt_flags &= ~DN_RT_F_IE;
 756         if (rt->rt_flags & RTCF_DOREDIRECT)
 757                 cb->rt_flags |= DN_RT_F_IE;
 758
 759         return NF_HOOK(PF_DECnet, NF_DN_FORWARD, skb, dev, skb->dev, neigh->output);
 760
 761 drop:
 762         kfree_skb(skb);
 763         return NET_RX_DROP;
 764 }
 765
 766 /*
 767  * Used to catch bugs. This should never normally get
 768  * called.
 769  */
 770 static int dn_rt_bug(struct sk_buff *skb)
 771 {
 772         if (net_ratelimit()) {
 773                 struct dn_skb_cb *cb = DN_SKB_CB(skb);
 774
 775                 printk(KERN_DEBUG "dn_rt_bug: skb from:%04x to:%04x\n",
 776                                 le16_to_cpu(cb->src), le16_to_cpu(cb->dst));
 777         }
 778
 779         kfree_skb(skb);
 780
 781         return NET_RX_BAD;
 782 }
 783
 784 static int dn_rt_set_next_hop(struct dn_route *rt, struct dn_fib_res *res)
 785 {
 786         struct dn_fib_info *fi = res->fi;
 787         struct net_device *dev = rt->u.dst.dev;
 788         struct neighbour *n;
 789         unsigned mss;
 790
 791         if (fi) {
 792                 if (DN_FIB_RES_GW(*res) &&
 793                     DN_FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
 794                         rt->rt_gateway = DN_FIB_RES_GW(*res);
 795                 memcpy(rt->u.dst.metrics, fi->fib_metrics,
 796                        sizeof(rt->u.dst.metrics));
 797         }
 798         rt->rt_type = res->type;
 799
 800         if (dev != NULL && rt->u.dst.neighbour == NULL) {
 801                 n = __neigh_lookup_errno(&dn_neigh_table, &rt->rt_gateway, dev);
 802                 if (IS_ERR(n))
 803                         return PTR_ERR(n);
 804                 rt->u.dst.neighbour = n;
 805         }
 806
 807         if (dst_metric(&rt->u.dst, RTAX_MTU) == 0 ||
 808             dst_metric(&rt->u.dst, RTAX_MTU) > rt->u.dst.dev->mtu)
 809                 rt->u.dst.metrics[RTAX_MTU-1] = rt->u.dst.dev->mtu;
 810         mss = dn_mss_from_pmtu(dev, dst_mtu(&rt->u.dst));
 811         if (dst_metric(&rt->u.dst, RTAX_ADVMSS) == 0 ||
 812             dst_metric(&rt->u.dst, RTAX_ADVMSS) > mss)
 813                 rt->u.dst.metrics[RTAX_ADVMSS-1] = mss;
 814         return 0;
 815 }
 816
 817 static inline int dn_match_addr(__le16 addr1, __le16 addr2)
 818 {
 819         __u16 tmp = le16_to_cpu(addr1) ^ le16_to_cpu(addr2);
 820         int match = 16;
 821         while(tmp) {
 822                 tmp >>= 1;
 823                 match--;
 824         }
 825         return match;
 826 }
 827
 828 static __le16 dnet_select_source(const struct net_device *dev, __le16 daddr, int scope)
 829 {
 830         __le16 saddr = 0;
 831         struct dn_dev *dn_db = dev->dn_ptr;
 832         struct dn_ifaddr *ifa;
 833         int best_match = 0;
 834         int ret;
 835
 836         read_lock(&dev_base_lock);
 837         for(ifa = dn_db->ifa_list; ifa; ifa = ifa->ifa_next) {
 838                 if (ifa->ifa_scope > scope)
 839                         continue;
 840                 if (!daddr) {
 841                         saddr = ifa->ifa_local;
 842                         break;
 843                 }
 844                 ret = dn_match_addr(daddr, ifa->ifa_local);
 845                 if (ret > best_match)
 846                         saddr = ifa->ifa_local;
 847                 if (best_match == 0)
 848                         saddr = ifa->ifa_local;
 849         }
 850         read_unlock(&dev_base_lock);
 851
 852         return saddr;
 853 }
 854
 855 static inline __le16 __dn_fib_res_prefsrc(struct dn_fib_res *res)
 856 {
 857         return dnet_select_source(DN_FIB_RES_DEV(*res), DN_FIB_RES_GW(*res), res->scope);
 858 }
 859
 860 static inline __le16 dn_fib_rules_map_destination(__le16 daddr, struct dn_fib_res *res)
 861 {
 862         __le16 mask = dnet_make_mask(res->prefixlen);
 863         return (daddr&~mask)|res->fi->fib_nh->nh_gw;
 864 }
 865
 866 static int dn_route_output_slow(struct dst_entry **pprt, const struct flowi *oldflp, int try_hard)
 867 {
 868         struct flowi fl = { .nl_u = { .dn_u =
 869                                       { .daddr = oldflp->fld_dst,
 870                                         .saddr = oldflp->fld_src,
 871                                         .scope = RT_SCOPE_UNIVERSE,
 872                                      } },
 873                             .mark = oldflp->mark,
 874                             .iif = init_net.loopback_dev->ifindex,
 875                             .oif = oldflp->oif };
 876         struct dn_route *rt = NULL;
 877         struct net_device *dev_out = NULL, *dev;
 878         struct neighbour *neigh = NULL;
 879         unsigned hash;
 880         unsigned flags = 0;
 881         struct dn_fib_res res = { .fi = NULL, .type = RTN_UNICAST };
 882         int err;
 883         int free_res = 0;
 884         __le16 gateway = 0;
 885
 886         if (decnet_debug_level & 16)
 887                 printk(KERN_DEBUG
 888                        "dn_route_output_slow: dst=%04x src=%04x mark=%d"
 889                        " iif=%d oif=%d\n", le16_to_cpu(oldflp->fld_dst),
 890                        le16_to_cpu(oldflp->fld_src),
 891                        oldflp->mark, init_net.loopback_dev->ifindex, oldflp->oif);
 892
 893         /* If we have an output interface, verify its a DECnet device */
 894         if (oldflp->oif) {
 895                 dev_out = dev_get_by_index(&init_net, oldflp->oif);
 896                 err = -ENODEV;
 897                 if (dev_out && dev_out->dn_ptr == NULL) {
 898                         dev_put(dev_out);
 899                         dev_out = NULL;
 900                 }
 901                 if (dev_out == NULL)
 902                         goto out;
 903         }
 904
 905         /* If we have a source address, verify that its a local address */
 906         if (oldflp->fld_src) {
 907                 err = -EADDRNOTAVAIL;
 908
 909                 if (dev_out) {
 910                         if (dn_dev_islocal(dev_out, oldflp->fld_src))
 911                                 goto source_ok;
 912                         dev_put(dev_out);
 913                         goto out;
 914                 }
 915                 read_lock(&dev_base_lock);
 916                 for_each_netdev(&init_net, dev) {
 917                         if (!dev->dn_ptr)
 918                                 continue;
 919                         if (!dn_dev_islocal(dev, oldflp->fld_src))
 920                                 continue;
 921                         if ((dev->flags & IFF_LOOPBACK) &&
 922                             oldflp->fld_dst &&
 923                             !dn_dev_islocal(dev, oldflp->fld_dst))
 924                                 continue;
 925
 926                         dev_out = dev;
 927                         break;
 928                 }
 929                 read_unlock(&dev_base_lock);
 930                 if (dev_out == NULL)
 931                         goto out;
 932                 dev_hold(dev_out);
 933 source_ok:
 934                 ;
 935         }
 936
 937         /* No destination? Assume its local */
 938         if (!fl.fld_dst) {
 939                 fl.fld_dst = fl.fld_src;
 940
 941                 err = -EADDRNOTAVAIL;
 942                 if (dev_out)
 943                         dev_put(dev_out);
 944                 dev_out = init_net.loopback_dev;
 945                 dev_hold(dev_out);
 946                 if (!fl.fld_dst) {
 947                         fl.fld_dst =
 948                         fl.fld_src = dnet_select_source(dev_out, 0,
 949                                                        RT_SCOPE_HOST);
 950                         if (!fl.fld_dst)
 951                                 goto out;
 952                 }
 953                 fl.oif = init_net.loopback_dev->ifindex;
 954                 res.type = RTN_LOCAL;
 955                 goto make_route;
 956         }
 957
 958         if (decnet_debug_level & 16)
 959                 printk(KERN_DEBUG
 960                        "dn_route_output_slow: initial checks complete."
 961                        " dst=%o4x src=%04x oif=%d try_hard=%d\n",
 962                        le16_to_cpu(fl.fld_dst), le16_to_cpu(fl.fld_src),
 963                        fl.oif, try_hard);
 964
 965         /*
 966          * N.B. If the kernel is compiled without router support then
 967          * dn_fib_lookup() will evaluate to non-zero so this if () block
 968          * will always be executed.
 969          */
 970         err = -ESRCH;
 971         if (try_hard || (err = dn_fib_lookup(&fl, &res)) != 0) {
 972                 struct dn_dev *dn_db;
 973                 if (err != -ESRCH)
 974                         goto out;
 975                 /*
 976                  * Here the fallback is basically the standard algorithm for
 977                  * routing in endnodes which is described in the DECnet routing
 978                  * docs
 979                  *
 980                  * If we are not trying hard, look in neighbour cache.
 981                  * The result is tested to ensure that if a specific output
 982                  * device/source address was requested, then we honour that
 983                  * here
 984                  */
 985                 if (!try_hard) {
 986                         neigh = neigh_lookup_nodev(&dn_neigh_table, &init_net, &fl.fld_dst);
 987                         if (neigh) {
 988                                 if ((oldflp->oif &&
 989                                     (neigh->dev->ifindex != oldflp->oif)) ||
 990                                     (oldflp->fld_src &&
 991                                     (!dn_dev_islocal(neigh->dev,
 992                                                       oldflp->fld_src)))) {
 993                                         neigh_release(neigh);
 994                                         neigh = NULL;
 995                                 } else {
 996                                         if (dev_out)
 997                                                 dev_put(dev_out);
 998                                         if (dn_dev_islocal(neigh->dev, fl.fld_dst)) {
 999                                                 dev_out = init_net.loopback_dev;
1000                                                 res.type = RTN_LOCAL;
1001                                         } else {
1002                                                 dev_out = neigh->dev;
1003                                         }
1004                                         dev_hold(dev_out);
1005                                         goto select_source;
1006                                 }
1007                         }
1008                 }
1009
1010                 /* Not there? Perhaps its a local address */
1011                 if (dev_out == NULL)
1012                         dev_out = dn_dev_get_default();
1013                 err = -ENODEV;
1014                 if (dev_out == NULL)
1015                         goto out;
1016                 dn_db = dev_out->dn_ptr;
1017                 /* Possible improvement - check all devices for local addr */
1018                 if (dn_dev_islocal(dev_out, fl.fld_dst)) {
1019                         dev_put(dev_out);
1020                         dev_out = init_net.loopback_dev;
1021                         dev_hold(dev_out);
1022                         res.type = RTN_LOCAL;
1023                         goto select_source;
1024                 }
1025                 /* Not local either.... try sending it to the default router */
1026                 neigh = neigh_clone(dn_db->router);
1027                 BUG_ON(neigh && neigh->dev != dev_out);
1028
1029                 /* Ok then, we assume its directly connected and move on */
1030 select_source:
1031                 if (neigh)
1032                         gateway = ((struct dn_neigh *)neigh)->addr;
1033                 if (gateway == 0)
1034                         gateway = fl.fld_dst;
1035                 if (fl.fld_src == 0) {
1036                         fl.fld_src = dnet_select_source(dev_out, gateway,
1037                                                          res.type == RTN_LOCAL ?
1038                                                          RT_SCOPE_HOST :
1039                                                          RT_SCOPE_LINK);
1040                         if (fl.fld_src == 0 && res.type != RTN_LOCAL)
1041                                 goto e_addr;
1042                 }
1043                 fl.oif = dev_out->ifindex;
1044                 goto make_route;
1045         }
1046         free_res = 1;
1047
1048         if (res.type == RTN_NAT)
1049                 goto e_inval;
1050
1051         if (res.type == RTN_LOCAL) {
1052                 if (!fl.fld_src)
1053                         fl.fld_src = fl.fld_dst;
1054                 if (dev_out)
1055                         dev_put(dev_out);
1056                 dev_out = init_net.loopback_dev;
1057                 dev_hold(dev_out);
1058                 fl.oif = dev_out->ifindex;
1059                 if (res.fi)
1060                         dn_fib_info_put(res.fi);
1061                 res.fi = NULL;
1062                 goto make_route;
1063         }
1064
1065         if (res.fi->fib_nhs > 1 && fl.oif == 0)
1066                 dn_fib_select_multipath(&fl, &res);
1067
1068         /*
1069          * We could add some logic to deal with default routes here and
1070          * get rid of some of the special casing above.
1071          */
1072
1073         if (!fl.fld_src)
1074                 fl.fld_src = DN_FIB_RES_PREFSRC(res);
1075
1076         if (dev_out)
1077                 dev_put(dev_out);
1078         dev_out = DN_FIB_RES_DEV(res);
1079         dev_hold(dev_out);
1080         fl.oif = dev_out->ifindex;
1081         gateway = DN_FIB_RES_GW(res);
1082
1083 make_route:
1084         if (dev_out->flags & IFF_LOOPBACK)
1085                 flags |= RTCF_LOCAL;
1086
1087         rt = dst_alloc(&dn_dst_ops);
1088         if (rt == NULL)
1089                 goto e_nobufs;
1090
1091         atomic_set(&rt->u.dst.__refcnt, 1);
1092         rt->u.dst.flags   = DST_HOST;
1093
1094         rt->fl.fld_src    = oldflp->fld_src;
1095         rt->fl.fld_dst    = oldflp->fld_dst;
1096         rt->fl.oif        = oldflp->oif;
1097         rt->fl.iif        = 0;
1098         rt->fl.mark       = oldflp->mark;
1099
1100         rt->rt_saddr      = fl.fld_src;
1101         rt->rt_daddr      = fl.fld_dst;
1102         rt->rt_gateway    = gateway ? gateway : fl.fld_dst;
1103         rt->rt_local_src  = fl.fld_src;
1104
1105         rt->rt_dst_map    = fl.fld_dst;
1106         rt->rt_src_map    = fl.fld_src;
1107
1108         rt->u.dst.dev = dev_out;
1109         dev_hold(dev_out);
1110         rt->u.dst.neighbour = neigh;
1111         neigh = NULL;
1112
1113         rt->u.dst.lastuse = jiffies;
1114         rt->u.dst.output  = dn_output;
1115         rt->u.dst.input   = dn_rt_bug;
1116         rt->rt_flags      = flags;
1117         if (flags & RTCF_LOCAL)
1118                 rt->u.dst.input = dn_nsp_rx;
1119
1120         err = dn_rt_set_next_hop(rt, &res);
1121         if (err)
1122                 goto e_neighbour;
1123
1124         hash = dn_hash(rt->fl.fld_src, rt->fl.fld_dst);
1125         dn_insert_route(rt, hash, (struct dn_route **)pprt);
1126
1127 done:
1128         if (neigh)
1129                 neigh_release(neigh);
1130         if (free_res)
1131                 dn_fib_res_put(&res);
1132         if (dev_out)
1133                 dev_put(dev_out);
1134 out:
1135         return err;
1136
1137 e_addr:
1138         err = -EADDRNOTAVAIL;
1139         goto done;
1140 e_inval:
1141         err = -EINVAL;
1142         goto done;
1143 e_nobufs:
1144         err = -ENOBUFS;
1145         goto done;
1146 e_neighbour:
1147         dst_free(&rt->u.dst);
1148         goto e_nobufs;
1149 }
1150
1151
1152 /*
1153  * N.B. The flags may be moved into the flowi at some future stage.
1154  */
1155 static int __dn_route_output_key(struct dst_entry **pprt, const struct flowi *flp, int flags)
1156 {
1157         unsigned hash = dn_hash(flp->fld_src, flp->fld_dst);
1158         struct dn_route *rt = NULL;
1159
1160         if (!(flags & MSG_TRYHARD)) {
1161                 rcu_read_lock_bh();
1162                 for(rt = rcu_dereference(dn_rt_hash_table[hash].chain); rt;
1163                         rt = rcu_dereference(rt->u.dst.dn_next)) {
1164                         if ((flp->fld_dst == rt->fl.fld_dst) &&
1165                             (flp->fld_src == rt->fl.fld_src) &&
1166                             (flp->mark == rt->fl.mark) &&
1167                             (rt->fl.iif == 0) &&
1168                             (rt->fl.oif == flp->oif)) {
1169                                 dst_use(&rt->u.dst, jiffies);
1170                                 rcu_read_unlock_bh();
1171                                 *pprt = &rt->u.dst;
1172                                 return 0;
1173                         }
1174                 }
1175                 rcu_read_unlock_bh();
1176         }
1177
1178         return dn_route_output_slow(pprt, flp, flags);
1179 }
1180
1181 static int dn_route_output_key(struct dst_entry **pprt, struct flowi *flp, int flags)
1182 {
1183         int err;
1184
1185         err = __dn_route_output_key(pprt, flp, flags);
1186         if (err == 0 && flp->proto) {
1187                 err = xfrm_lookup(&init_net, pprt, flp, NULL, 0);
1188         }
1189         return err;
1190 }
1191
1192 int dn_route_output_sock(struct dst_entry **pprt, struct flowi *fl, struct sock *sk, int flags)
1193 {
1194         int err;
1195
1196         err = __dn_route_output_key(pprt, fl, flags & MSG_TRYHARD);
1197         if (err == 0 && fl->proto) {
1198                 err = xfrm_lookup(&init_net, pprt, fl, sk,
1199                                  (flags & MSG_DONTWAIT) ? 0 : XFRM_LOOKUP_WAIT);
1200         }
1201         return err;
1202 }
1203
1204 static int dn_route_input_slow(struct sk_buff *skb)
1205 {
1206         struct dn_route *rt = NULL;
1207         struct dn_skb_cb *cb = DN_SKB_CB(skb);
1208         struct net_device *in_dev = skb->dev;
1209         struct net_device *out_dev = NULL;
1210         struct dn_dev *dn_db;
1211         struct neighbour *neigh = NULL;
1212         unsigned hash;
1213         int flags = 0;
1214         __le16 gateway = 0;
1215         __le16 local_src = 0;
1216         struct flowi fl = { .nl_u = { .dn_u =
1217                                      { .daddr = cb->dst,
1218                                        .saddr = cb->src,
1219                                        .scope = RT_SCOPE_UNIVERSE,
1220                                     } },
1221                             .mark = skb->mark,
1222                             .iif = skb->dev->ifindex };
1223         struct dn_fib_res res = { .fi = NULL, .type = RTN_UNREACHABLE };
1224         int err = -EINVAL;
1225         int free_res = 0;
1226
1227         dev_hold(in_dev);
1228
1229         if ((dn_db = in_dev->dn_ptr) == NULL)
1230                 goto out;
1231
1232         /* Zero source addresses are not allowed */
1233         if (fl.fld_src == 0)
1234                 goto out;
1235
1236         /*
1237          * In this case we've just received a packet from a source
1238          * outside ourselves pretending to come from us. We don't
1239          * allow it any further to prevent routing loops, spoofing and
1240          * other nasties. Loopback packets already have the dst attached
1241          * so this only affects packets which have originated elsewhere.
1242          */
1243         err  = -ENOTUNIQ;
1244         if (dn_dev_islocal(in_dev, cb->src))
1245                 goto out;
1246
1247         err = dn_fib_lookup(&fl, &res);
1248         if (err) {
1249                 if (err != -ESRCH)
1250                         goto out;
1251                 /*
1252                  * Is the destination us ?
1253                  */
1254                 if (!dn_dev_islocal(in_dev, cb->dst))
1255                         goto e_inval;
1256
1257                 res.type = RTN_LOCAL;
1258         } else {
1259                 __le16 src_map = fl.fld_src;
1260                 free_res = 1;
1261
1262                 out_dev = DN_FIB_RES_DEV(res);
1263                 if (out_dev == NULL) {
1264                         if (net_ratelimit())
1265                                 printk(KERN_CRIT "Bug in dn_route_input_slow() "
1266                                                  "No output device\n");
1267                         goto e_inval;
1268                 }
1269                 dev_hold(out_dev);
1270
1271                 if (res.r)
1272                         src_map = fl.fld_src; /* no NAT support for now */
1273
1274                 gateway = DN_FIB_RES_GW(res);
1275                 if (res.type == RTN_NAT) {
1276                         fl.fld_dst = dn_fib_rules_map_destination(fl.fld_dst, &res);
1277                         dn_fib_res_put(&res);
1278                         free_res = 0;
1279                         if (dn_fib_lookup(&fl, &res))
1280                                 goto e_inval;
1281                         free_res = 1;
1282                         if (res.type != RTN_UNICAST)
1283                                 goto e_inval;
1284                         flags |= RTCF_DNAT;
1285                         gateway = fl.fld_dst;
1286                 }
1287                 fl.fld_src = src_map;
1288         }
1289
1290         switch(res.type) {
1291         case RTN_UNICAST:
1292                 /*
1293                  * Forwarding check here, we only check for forwarding
1294                  * being turned off, if you want to only forward intra
1295                  * area, its up to you to set the routing tables up
1296                  * correctly.
1297                  */
1298                 if (dn_db->parms.forwarding == 0)
1299                         goto e_inval;
1300
1301                 if (res.fi->fib_nhs > 1 && fl.oif == 0)
1302                         dn_fib_select_multipath(&fl, &res);
1303
1304                 /*
1305                  * Check for out_dev == in_dev. We use the RTCF_DOREDIRECT
1306                  * flag as a hint to set the intra-ethernet bit when
1307                  * forwarding. If we've got NAT in operation, we don't do
1308                  * this optimisation.
1309                  */
1310                 if (out_dev == in_dev && !(flags & RTCF_NAT))
1311                         flags |= RTCF_DOREDIRECT;
1312
1313                 local_src = DN_FIB_RES_PREFSRC(res);
1314
1315         case RTN_BLACKHOLE:
1316         case RTN_UNREACHABLE:
1317                 break;
1318         case RTN_LOCAL:
1319                 flags |= RTCF_LOCAL;
1320                 fl.fld_src = cb->dst;
1321                 fl.fld_dst = cb->src;
1322
1323                 /* Routing tables gave us a gateway */
1324                 if (gateway)
1325                         goto make_route;
1326
1327                 /* Packet was intra-ethernet, so we know its on-link */
1328                 if (cb->rt_flags & DN_RT_F_IE) {
1329                         gateway = cb->src;
1330                         flags |= RTCF_DIRECTSRC;
1331                         goto make_route;
1332                 }
1333
1334                 /* Use the default router if there is one */
1335                 neigh = neigh_clone(dn_db->router);
1336                 if (neigh) {
1337                         gateway = ((struct dn_neigh *)neigh)->addr;
1338                         goto make_route;
1339                 }
1340
1341                 /* Close eyes and pray */
1342                 gateway = cb->src;
1343                 flags |= RTCF_DIRECTSRC;
1344                 goto make_route;
1345         default:
1346                 goto e_inval;
1347         }
1348
1349 make_route:
1350         rt = dst_alloc(&dn_dst_ops);
1351         if (rt == NULL)
1352                 goto e_nobufs;
1353
1354         rt->rt_saddr      = fl.fld_src;
1355         rt->rt_daddr      = fl.fld_dst;
1356         rt->rt_gateway    = fl.fld_dst;
1357         if (gateway)
1358                 rt->rt_gateway = gateway;
1359         rt->rt_local_src  = local_src ? local_src : rt->rt_saddr;
1360
1361         rt->rt_dst_map    = fl.fld_dst;
1362         rt->rt_src_map    = fl.fld_src;
1363
1364         rt->fl.fld_src    = cb->src;
1365         rt->fl.fld_dst    = cb->dst;
1366         rt->fl.oif        = 0;
1367         rt->fl.iif        = in_dev->ifindex;
1368         rt->fl.mark       = fl.mark;
1369
1370         rt->u.dst.flags = DST_HOST;
1371         rt->u.dst.neighbour = neigh;
1372         rt->u.dst.dev = out_dev;
1373         rt->u.dst.lastuse = jiffies;
1374         rt->u.dst.output = dn_rt_bug;
1375         switch(res.type) {
1376                 case RTN_UNICAST:
1377                         rt->u.dst.input = dn_forward;
1378                         break;
1379                 case RTN_LOCAL:
1380                         rt->u.dst.output = dn_output;
1381                         rt->u.dst.input = dn_nsp_rx;
1382                         rt->u.dst.dev = in_dev;
1383                         flags |= RTCF_LOCAL;
1384                         break;
1385                 default:
1386                 case RTN_UNREACHABLE:
1387                 case RTN_BLACKHOLE:
1388                         rt->u.dst.input = dst_discard;
1389         }
1390         rt->rt_flags = flags;
1391         if (rt->u.dst.dev)
1392                 dev_hold(rt->u.dst.dev);
1393
1394         err = dn_rt_set_next_hop(rt, &res);
1395         if (err)
1396                 goto e_neighbour;
1397
1398         hash = dn_hash(rt->fl.fld_src, rt->fl.fld_dst);
1399         dn_insert_route(rt, hash, (struct dn_route **)&skb->dst);
1400
1401 done:
1402         if (neigh)
1403                 neigh_release(neigh);
1404         if (free_res)
1405                 dn_fib_res_put(&res);
1406         dev_put(in_dev);
1407         if (out_dev)
1408                 dev_put(out_dev);
1409 out:
1410         return err;
1411
1412 e_inval:
1413         err = -EINVAL;
1414         goto done;
1415
1416 e_nobufs:
1417         err = -ENOBUFS;
1418         goto done;
1419
1420 e_neighbour:
1421         dst_free(&rt->u.dst);
1422         goto done;
1423 }
1424
1425 int dn_route_input(struct sk_buff *skb)
1426 {
1427         struct dn_route *rt;
1428         struct dn_skb_cb *cb = DN_SKB_CB(skb);
1429         unsigned hash = dn_hash(cb->src, cb->dst);
1430
1431         if (skb->dst)
1432                 return 0;
1433
1434         rcu_read_lock();
1435         for(rt = rcu_dereference(dn_rt_hash_table[hash].chain); rt != NULL;
1436             rt = rcu_dereference(rt->u.dst.dn_next)) {
1437                 if ((rt->fl.fld_src == cb->src) &&
1438                     (rt->fl.fld_dst == cb->dst) &&
1439                     (rt->fl.oif == 0) &&
1440                     (rt->fl.mark == skb->mark) &&
1441                     (rt->fl.iif == cb->iif)) {
1442                         dst_use(&rt->u.dst, jiffies);
1443                         rcu_read_unlock();
1444                         skb->dst = (struct dst_entry *)rt;
1445                         return 0;
1446                 }
1447         }
1448         rcu_read_unlock();
1449
1450         return dn_route_input_slow(skb);
1451 }
1452
1453 static int dn_rt_fill_info(struct sk_buff *skb, u32 pid, u32 seq,
1454                            int event, int nowait, unsigned int flags)
1455 {
1456         struct dn_route *rt = (struct dn_route *)skb->dst;
1457         struct rtmsg *r;
1458         struct nlmsghdr *nlh;
1459         unsigned char *b = skb_tail_pointer(skb);
1460         long expires;
1461
1462         nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*r), flags);
1463         r = NLMSG_DATA(nlh);
1464         r->rtm_family = AF_DECnet;
1465         r->rtm_dst_len = 16;
1466         r->rtm_src_len = 0;
1467         r->rtm_tos = 0;
1468         r->rtm_table = RT_TABLE_MAIN;
1469         RTA_PUT_U32(skb, RTA_TABLE, RT_TABLE_MAIN);
1470         r->rtm_type = rt->rt_type;
1471         r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
1472         r->rtm_scope = RT_SCOPE_UNIVERSE;
1473         r->rtm_protocol = RTPROT_UNSPEC;
1474         if (rt->rt_flags & RTCF_NOTIFY)
1475                 r->rtm_flags |= RTM_F_NOTIFY;
1476         RTA_PUT(skb, RTA_DST, 2, &rt->rt_daddr);
1477         if (rt->fl.fld_src) {
1478                 r->rtm_src_len = 16;
1479                 RTA_PUT(skb, RTA_SRC, 2, &rt->fl.fld_src);
1480         }
1481         if (rt->u.dst.dev)
1482                 RTA_PUT(skb, RTA_OIF, sizeof(int), &rt->u.dst.dev->ifindex);
1483         /*
1484          * Note to self - change this if input routes reverse direction when
1485          * they deal only with inputs and not with replies like they do
1486          * currently.
1487          */
1488         RTA_PUT(skb, RTA_PREFSRC, 2, &rt->rt_local_src);
1489         if (rt->rt_daddr != rt->rt_gateway)
1490                 RTA_PUT(skb, RTA_GATEWAY, 2, &rt->rt_gateway);
1491         if (rtnetlink_put_metrics(skb, rt->u.dst.metrics) < 0)
1492                 goto rtattr_failure;
1493         expires = rt->u.dst.expires ? rt->u.dst.expires - jiffies : 0;
1494         if (rtnl_put_cacheinfo(skb, &rt->u.dst, 0, 0, 0, expires,
1495                                rt->u.dst.error) < 0)
1496                 goto rtattr_failure;
1497         if (rt->fl.iif)
1498                 RTA_PUT(skb, RTA_IIF, sizeof(int), &rt->fl.iif);
1499
1500         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1501         return skb->len;
1502
1503 nlmsg_failure:
1504 rtattr_failure:
1505         nlmsg_trim(skb, b);
1506         return -1;
1507 }
1508
1509 /*
1510  * This is called by both endnodes and routers now.
1511  */
1512 static int dn_cache_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh, void *arg)
1513 {
1514         struct net *net = sock_net(in_skb->sk);
1515         struct rtattr **rta = arg;
1516         struct rtmsg *rtm = NLMSG_DATA(nlh);
1517         struct dn_route *rt = NULL;
1518         struct dn_skb_cb *cb;
1519         int err;
1520         struct sk_buff *skb;
1521         struct flowi fl;
1522
1523         if (net != &init_net)
1524                 return -EINVAL;
1525
1526         memset(&fl, 0, sizeof(fl));
1527         fl.proto = DNPROTO_NSP;
1528
1529         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1530         if (skb == NULL)
1531                 return -ENOBUFS;
1532         skb_reset_mac_header(skb);
1533         cb = DN_SKB_CB(skb);
1534
1535         if (rta[RTA_SRC-1])
1536                 memcpy(&fl.fld_src, RTA_DATA(rta[RTA_SRC-1]), 2);
1537         if (rta[RTA_DST-1])
1538                 memcpy(&fl.fld_dst, RTA_DATA(rta[RTA_DST-1]), 2);
1539         if (rta[RTA_IIF-1])
1540                 memcpy(&fl.iif, RTA_DATA(rta[RTA_IIF-1]), sizeof(int));
1541
1542         if (fl.iif) {
1543                 struct net_device *dev;
1544                 if ((dev = dev_get_by_index(&init_net, fl.iif)) == NULL) {
1545                         kfree_skb(skb);
1546                         return -ENODEV;
1547                 }
1548                 if (!dev->dn_ptr) {
1549                         dev_put(dev);
1550                         kfree_skb(skb);
1551                         return -ENODEV;
1552                 }
1553                 skb->protocol = htons(ETH_P_DNA_RT);
1554                 skb->dev = dev;
1555                 cb->src = fl.fld_src;
1556                 cb->dst = fl.fld_dst;
1557                 local_bh_disable();
1558                 err = dn_route_input(skb);
1559                 local_bh_enable();
1560                 memset(cb, 0, sizeof(struct dn_skb_cb));
1561                 rt = (struct dn_route *)skb->dst;
1562                 if (!err && -rt->u.dst.error)
1563                         err = rt->u.dst.error;
1564         } else {
1565                 int oif = 0;
1566                 if (rta[RTA_OIF - 1])
1567                         memcpy(&oif, RTA_DATA(rta[RTA_OIF - 1]), sizeof(int));
1568                 fl.oif = oif;
1569                 err = dn_route_output_key((struct dst_entry **)&rt, &fl, 0);
1570         }
1571
1572         if (skb->dev)
1573                 dev_put(skb->dev);
1574         skb->dev = NULL;
1575         if (err)
1576                 goto out_free;
1577         skb->dst = &rt->u.dst;
1578         if (rtm->rtm_flags & RTM_F_NOTIFY)
1579                 rt->rt_flags |= RTCF_NOTIFY;
1580
1581         err = dn_rt_fill_info(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq, RTM_NEWROUTE, 0, 0);
1582
1583         if (err == 0)
1584                 goto out_free;
1585         if (err < 0) {
1586                 err = -EMSGSIZE;
1587                 goto out_free;
1588         }
1589
1590         return rtnl_unicast(skb, &init_net, NETLINK_CB(in_skb).pid);
1591
1592 out_free:
1593         kfree_skb(skb);
1594         return err;
1595 }
1596
1597 /*
1598  * For routers, this is called from dn_fib_dump, but for endnodes its
1599  * called directly from the rtnetlink dispatch table.
1600  */
1601 int dn_cache_dump(struct sk_buff *skb, struct netlink_callback *cb)
1602 {
1603         struct net *net = sock_net(skb->sk);
1604         struct dn_route *rt;
1605         int h, s_h;
1606         int idx, s_idx;
1607
1608         if (net != &init_net)
1609                 return 0;
1610
1611         if (NLMSG_PAYLOAD(cb->nlh, 0) < sizeof(struct rtmsg))
1612                 return -EINVAL;
1613         if (!(((struct rtmsg *)NLMSG_DATA(cb->nlh))->rtm_flags&RTM_F_CLONED))
1614                 return 0;
1615
1616         s_h = cb->args[0];
1617         s_idx = idx = cb->args[1];
1618         for(h = 0; h <= dn_rt_hash_mask; h++) {
1619                 if (h < s_h)
1620                         continue;
1621                 if (h > s_h)
1622                         s_idx = 0;
1623                 rcu_read_lock_bh();
1624                 for(rt = rcu_dereference(dn_rt_hash_table[h].chain), idx = 0;
1625                         rt;
1626                         rt = rcu_dereference(rt->u.dst.dn_next), idx++) {
1627                         if (idx < s_idx)
1628                                 continue;
1629                         skb->dst = dst_clone(&rt->u.dst);
1630                         if (dn_rt_fill_info(skb, NETLINK_CB(cb->skb).pid,
1631                                         cb->nlh->nlmsg_seq, RTM_NEWROUTE,
1632                                         1, NLM_F_MULTI) <= 0) {
1633                                 dst_release(xchg(&skb->dst, NULL));
1634                                 rcu_read_unlock_bh();
1635                                 goto done;
1636                         }
1637                         dst_release(xchg(&skb->dst, NULL));
1638                 }
1639                 rcu_read_unlock_bh();
1640         }
1641
1642 done:
1643         cb->args[0] = h;
1644         cb->args[1] = idx;
1645         return skb->len;
1646 }
1647
1648 #ifdef CONFIG_PROC_FS
1649 struct dn_rt_cache_iter_state {
1650         int bucket;
1651 };
1652
1653 static struct dn_route *dn_rt_cache_get_first(struct seq_file *seq)
1654 {
1655         struct dn_route *rt = NULL;
1656         struct dn_rt_cache_iter_state *s = seq->private;
1657
1658         for(s->bucket = dn_rt_hash_mask; s->bucket >= 0; --s->bucket) {
1659                 rcu_read_lock_bh();
1660                 rt = dn_rt_hash_table[s->bucket].chain;
1661                 if (rt)
1662                         break;
1663                 rcu_read_unlock_bh();
1664         }
1665         return rcu_dereference(rt);
1666 }
1667
1668 static struct dn_route *dn_rt_cache_get_next(struct seq_file *seq, struct dn_route *rt)
1669 {
1670         struct dn_rt_cache_iter_state *s = seq->private;
1671
1672         rt = rt->u.dst.dn_next;
1673         while(!rt) {
1674                 rcu_read_unlock_bh();
1675                 if (--s->bucket < 0)
1676                         break;
1677                 rcu_read_lock_bh();
1678                 rt = dn_rt_hash_table[s->bucket].chain;
1679         }
1680         return rcu_dereference(rt);
1681 }
1682
1683 static void *dn_rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
1684 {
1685         struct dn_route *rt = dn_rt_cache_get_first(seq);
1686
1687         if (rt) {
1688                 while(*pos && (rt = dn_rt_cache_get_next(seq, rt)))
1689                         --*pos;
1690         }
1691         return *pos ? NULL : rt;
1692 }
1693
1694 static void *dn_rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1695 {
1696         struct dn_route *rt = dn_rt_cache_get_next(seq, v);
1697         ++*pos;
1698         return rt;
1699 }
1700
1701 static void dn_rt_cache_seq_stop(struct seq_file *seq, void *v)
1702 {
1703         if (v)
1704                 rcu_read_unlock_bh();
1705 }
1706
1707 static int dn_rt_cache_seq_show(struct seq_file *seq, void *v)
1708 {
1709         struct dn_route *rt = v;
1710         char buf1[DN_ASCBUF_LEN], buf2[DN_ASCBUF_LEN];
1711
1712         seq_printf(seq, "%-8s %-7s %-7s %04d %04d %04d\n",
1713                         rt->u.dst.dev ? rt->u.dst.dev->name : "*",
1714                         dn_addr2asc(le16_to_cpu(rt->rt_daddr), buf1),
1715                         dn_addr2asc(le16_to_cpu(rt->rt_saddr), buf2),
1716                         atomic_read(&rt->u.dst.__refcnt),
1717                         rt->u.dst.__use,
1718                         (int) dst_metric(&rt->u.dst, RTAX_RTT));
1719         return 0;
1720 }
1721
1722 static const struct seq_operations dn_rt_cache_seq_ops = {
1723         .start  = dn_rt_cache_seq_start,
1724         .next   = dn_rt_cache_seq_next,
1725         .stop   = dn_rt_cache_seq_stop,
1726         .show   = dn_rt_cache_seq_show,
1727 };
1728
1729 static int dn_rt_cache_seq_open(struct inode *inode, struct file *file)
1730 {
1731         return seq_open_private(file, &dn_rt_cache_seq_ops,
1732                         sizeof(struct dn_rt_cache_iter_state));
1733 }
1734
1735 static const struct file_operations dn_rt_cache_seq_fops = {
1736         .owner   = THIS_MODULE,
1737         .open    = dn_rt_cache_seq_open,
1738         .read    = seq_read,
1739         .llseek  = seq_lseek,
1740         .release = seq_release_private,
1741 };
1742
1743 #endif /* CONFIG_PROC_FS */
1744
1745 void __init dn_route_init(void)
1746 {
1747         int i, goal, order;
1748
1749         dn_dst_ops.kmem_cachep =
1750                 kmem_cache_create("dn_dst_cache", sizeof(struct dn_route), 0,
1751                                   SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1752         setup_timer(&dn_route_timer, dn_dst_check_expire, 0);
1753         dn_route_timer.expires = jiffies + decnet_dst_gc_interval * HZ;
1754         add_timer(&dn_route_timer);
1755
1756         goal = num_physpages >> (26 - PAGE_SHIFT);
1757
1758         for(order = 0; (1UL << order) < goal; order++)
1759                 /* NOTHING */;
1760
1761         /*
1762          * Only want 1024 entries max, since the table is very, very unlikely
1763          * to be larger than that.
1764          */
1765         while(order && ((((1UL << order) * PAGE_SIZE) /
1766                                 sizeof(struct dn_rt_hash_bucket)) >= 2048))
1767                 order--;
1768
1769         do {
1770                 dn_rt_hash_mask = (1UL << order) * PAGE_SIZE /
1771                         sizeof(struct dn_rt_hash_bucket);
1772                 while(dn_rt_hash_mask & (dn_rt_hash_mask - 1))
1773                         dn_rt_hash_mask--;
1774                 dn_rt_hash_table = (struct dn_rt_hash_bucket *)
1775                         __get_free_pages(GFP_ATOMIC, order);
1776         } while (dn_rt_hash_table == NULL && --order > 0);
1777
1778         if (!dn_rt_hash_table)
1779                 panic("Failed to allocate DECnet route cache hash table\n");
1780
1781         printk(KERN_INFO
1782                 "DECnet: Routing cache hash table of %u buckets, %ldKbytes\n",
1783                 dn_rt_hash_mask,
1784                 (long)(dn_rt_hash_mask*sizeof(struct dn_rt_hash_bucket))/1024);
1785
1786         dn_rt_hash_mask--;
1787         for(i = 0; i <= dn_rt_hash_mask; i++) {
1788                 spin_lock_init(&dn_rt_hash_table[i].lock);
1789                 dn_rt_hash_table[i].chain = NULL;
1790         }
1791
1792         dn_dst_ops.gc_thresh = (dn_rt_hash_mask + 1);
1793
1794         proc_net_fops_create(&init_net, "decnet_cache", S_IRUGO, &dn_rt_cache_seq_fops);
1795
1796 #ifdef CONFIG_DECNET_ROUTER
1797         rtnl_register(PF_DECnet, RTM_GETROUTE, dn_cache_getroute, dn_fib_dump);
1798 #else
1799         rtnl_register(PF_DECnet, RTM_GETROUTE, dn_cache_getroute,
1800                       dn_cache_dump);
1801 #endif
1802 }
1803
1804 void __exit dn_route_cleanup(void)
1805 {
1806         del_timer(&dn_route_timer);
1807         dn_run_flush(0);
1808
1809         proc_net_remove(&init_net, "decnet_cache");
1810 }
1811