SAFE public projects git trees. - safe/jmp/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/config.h>
  59 #include <linux/errno.h>
  60 #include <linux/types.h>
  61 #include <linux/socket.h>
  62 #include <linux/in.h>
  63 #include <linux/kernel.h>
  64 #include <linux/sockios.h>
  65 #include <linux/net.h>
  66 #include <linux/netdevice.h>
  67 #include <linux/inet.h>
  68 #include <linux/route.h>
  69 #include <linux/in_route.h>
  70 #include <net/sock.h>
  71 #include <linux/mm.h>
  72 #include <linux/proc_fs.h>
  73 #include <linux/seq_file.h>
  74 #include <linux/init.h>
  75 #include <linux/rtnetlink.h>
  76 #include <linux/string.h>
  77 #include <linux/netfilter_decnet.h>
  78 #include <linux/rcupdate.h>
  79 #include <linux/times.h>
  80 #include <asm/errno.h>
  81 #include <net/neighbour.h>
  82 #include <net/dst.h>
  83 #include <net/flow.h>
  84 #include <net/dn.h>
  85 #include <net/dn_dev.h>
  86 #include <net/dn_nsp.h>
  87 #include <net/dn_route.h>
  88 #include <net/dn_neigh.h>
  89 #include <net/dn_fib.h>
  90
  91 struct dn_rt_hash_bucket
  92 {
  93         struct dn_route *chain;
  94         spinlock_t lock;
  95 } __attribute__((__aligned__(8)));
  96
  97 extern struct neigh_table dn_neigh_table;
  98
  99
 100 static unsigned char dn_hiord_addr[6] = {0xAA,0x00,0x04,0x00,0x00,0x00};
 101
 102 static const int dn_rt_min_delay = 2 * HZ;
 103 static const int dn_rt_max_delay = 10 * HZ;
 104 static const int dn_rt_mtu_expires = 10 * 60 * HZ;
 105
 106 static unsigned long dn_rt_deadline;
 107
 108 static int dn_dst_gc(void);
 109 static struct dst_entry *dn_dst_check(struct dst_entry *, __u32);
 110 static struct dst_entry *dn_dst_negative_advice(struct dst_entry *);
 111 static void dn_dst_link_failure(struct sk_buff *);
 112 static void dn_dst_update_pmtu(struct dst_entry *dst, u32 mtu);
 113 static int dn_route_input(struct sk_buff *);
 114 static void dn_run_flush(unsigned long dummy);
 115
 116 static struct dn_rt_hash_bucket *dn_rt_hash_table;
 117 static unsigned dn_rt_hash_mask;
 118
 119 static struct timer_list dn_route_timer;
 120 static DEFINE_TIMER(dn_rt_flush_timer, dn_run_flush, 0, 0);
 121 int decnet_dst_gc_interval = 2;
 122
 123 static struct dst_ops dn_dst_ops = {
 124         .family =               PF_DECnet,
 125         .protocol =             __constant_htons(ETH_P_DNA_RT),
 126         .gc_thresh =            128,
 127         .gc =                   dn_dst_gc,
 128         .check =                dn_dst_check,
 129         .negative_advice =      dn_dst_negative_advice,
 130         .link_failure =         dn_dst_link_failure,
 131         .update_pmtu =          dn_dst_update_pmtu,
 132         .entry_size =           sizeof(struct dn_route),
 133         .entries =              ATOMIC_INIT(0),
 134 };
 135
 136 static __inline__ unsigned dn_hash(__le16 src, __le16 dst)
 137 {
 138         __u16 tmp = (__u16 __force)(src ^ dst);
 139         tmp ^= (tmp >> 3);
 140         tmp ^= (tmp >> 5);
 141         tmp ^= (tmp >> 10);
 142         return dn_rt_hash_mask & (unsigned)tmp;
 143 }
 144
 145 static inline void dnrt_free(struct dn_route *rt)
 146 {
 147         call_rcu_bh(&rt->u.dst.rcu_head, dst_rcu_free);
 148 }
 149
 150 static inline void dnrt_drop(struct dn_route *rt)
 151 {
 152         if (rt)
 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.rt_next;
 172                                 continue;
 173                         }
 174                         *rtp = rt->u.rt_next;
 175                         rt->u.rt_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(void)
 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.rt_next;
 203                                 continue;
 204                         }
 205                         *rtp = rt->u.rt_next;
 206                         rt->u.rt_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->metrics[RTAX_MTU-1] > 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->metrics[RTAX_ADVMSS-1] > 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 memcmp(&fl1->nl_u.dn_u, &fl2->nl_u.dn_u, sizeof(fl1->nl_u.dn_u)) == 0 &&
 272                 fl1->oif == fl2->oif &&
 273                 fl1->iif == fl2->iif;
 274 }
 275
 276 static int dn_insert_route(struct dn_route *rt, unsigned hash, struct dn_route **rp)
 277 {
 278         struct dn_route *rth, **rthp;
 279         unsigned long now = jiffies;
 280
 281         rthp = &dn_rt_hash_table[hash].chain;
 282
 283         spin_lock_bh(&dn_rt_hash_table[hash].lock);
 284         while((rth = *rthp) != NULL) {
 285                 if (compare_keys(&rth->fl, &rt->fl)) {
 286                         /* Put it first */
 287                         *rthp = rth->u.rt_next;
 288                         rcu_assign_pointer(rth->u.rt_next,
 289                                            dn_rt_hash_table[hash].chain);
 290                         rcu_assign_pointer(dn_rt_hash_table[hash].chain, rth);
 291
 292                         rth->u.dst.__use++;
 293                         dst_hold(&rth->u.dst);
 294                         rth->u.dst.lastuse = now;
 295                         spin_unlock_bh(&dn_rt_hash_table[hash].lock);
 296
 297                         dnrt_drop(rt);
 298                         *rp = rth;
 299                         return 0;
 300                 }
 301                 rthp = &rth->u.rt_next;
 302         }
 303
 304         rcu_assign_pointer(rt->u.rt_next, dn_rt_hash_table[hash].chain);
 305         rcu_assign_pointer(dn_rt_hash_table[hash].chain, rt);
 306
 307         dst_hold(&rt->u.dst);
 308         rt->u.dst.__use++;
 309         rt->u.dst.lastuse = now;
 310         spin_unlock_bh(&dn_rt_hash_table[hash].lock);
 311         *rp = rt;
 312         return 0;
 313 }
 314
 315 void dn_run_flush(unsigned long dummy)
 316 {
 317         int i;
 318         struct dn_route *rt, *next;
 319
 320         for(i = 0; i < dn_rt_hash_mask; i++) {
 321                 spin_lock_bh(&dn_rt_hash_table[i].lock);
 322
 323                 if ((rt = xchg(&dn_rt_hash_table[i].chain, NULL)) == NULL)
 324                         goto nothing_to_declare;
 325
 326                 for(; rt; rt=next) {
 327                         next = rt->u.rt_next;
 328                         rt->u.rt_next = NULL;
 329                         dst_free((struct dst_entry *)rt);
 330                 }
 331
 332 nothing_to_declare:
 333                 spin_unlock_bh(&dn_rt_hash_table[i].lock);
 334         }
 335 }
 336
 337 static DEFINE_SPINLOCK(dn_rt_flush_lock);
 338
 339 void dn_rt_cache_flush(int delay)
 340 {
 341         unsigned long now = jiffies;
 342         int user_mode = !in_interrupt();
 343
 344         if (delay < 0)
 345                 delay = dn_rt_min_delay;
 346
 347         spin_lock_bh(&dn_rt_flush_lock);
 348
 349         if (del_timer(&dn_rt_flush_timer) && delay > 0 && dn_rt_deadline) {
 350                 long tmo = (long)(dn_rt_deadline - now);
 351
 352                 if (user_mode && tmo < dn_rt_max_delay - dn_rt_min_delay)
 353                         tmo = 0;
 354
 355                 if (delay > tmo)
 356                         delay = tmo;
 357         }
 358
 359         if (delay <= 0) {
 360                 spin_unlock_bh(&dn_rt_flush_lock);
 361                 dn_run_flush(0);
 362                 return;
 363         }
 364
 365         if (dn_rt_deadline == 0)
 366                 dn_rt_deadline = now + dn_rt_max_delay;
 367
 368         dn_rt_flush_timer.expires = now + delay;
 369         add_timer(&dn_rt_flush_timer);
 370         spin_unlock_bh(&dn_rt_flush_lock);
 371 }
 372
 373 /**
 374  * dn_return_short - Return a short packet to its sender
 375  * @skb: The packet to return
 376  *
 377  */
 378 static int dn_return_short(struct sk_buff *skb)
 379 {
 380         struct dn_skb_cb *cb;
 381         unsigned char *ptr;
 382         __le16 *src;
 383         __le16 *dst;
 384         __le16 tmp;
 385
 386         /* Add back headers */
 387         skb_push(skb, skb->data - skb->nh.raw);
 388
 389         if ((skb = skb_unshare(skb, GFP_ATOMIC)) == NULL)
 390                 return NET_RX_DROP;
 391
 392         cb = DN_SKB_CB(skb);
 393         /* Skip packet length and point to flags */
 394         ptr = skb->data + 2;
 395         *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
 396
 397         dst = (__le16 *)ptr;
 398         ptr += 2;
 399         src = (__le16 *)ptr;
 400         ptr += 2;
 401         *ptr = 0; /* Zero hop count */
 402
 403         /* Swap source and destination */
 404         tmp  = *src;
 405         *src = *dst;
 406         *dst = tmp;
 407
 408         skb->pkt_type = PACKET_OUTGOING;
 409         dn_rt_finish_output(skb, NULL, NULL);
 410         return NET_RX_SUCCESS;
 411 }
 412
 413 /**
 414  * dn_return_long - Return a long packet to its sender
 415  * @skb: The long format packet to return
 416  *
 417  */
 418 static int dn_return_long(struct sk_buff *skb)
 419 {
 420         struct dn_skb_cb *cb;
 421         unsigned char *ptr;
 422         unsigned char *src_addr, *dst_addr;
 423         unsigned char tmp[ETH_ALEN];
 424
 425         /* Add back all headers */
 426         skb_push(skb, skb->data - skb->nh.raw);
 427
 428         if ((skb = skb_unshare(skb, GFP_ATOMIC)) == NULL)
 429                 return NET_RX_DROP;
 430
 431         cb = DN_SKB_CB(skb);
 432         /* Ignore packet length and point to flags */
 433         ptr = skb->data + 2;
 434
 435         /* Skip padding */
 436         if (*ptr & DN_RT_F_PF) {
 437                 char padlen = (*ptr & ~DN_RT_F_PF);
 438                 ptr += padlen;
 439         }
 440
 441         *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
 442         ptr += 2;
 443         dst_addr = ptr;
 444         ptr += 8;
 445         src_addr = ptr;
 446         ptr += 6;
 447         *ptr = 0; /* Zero hop count */
 448
 449         /* Swap source and destination */
 450         memcpy(tmp, src_addr, ETH_ALEN);
 451         memcpy(src_addr, dst_addr, ETH_ALEN);
 452         memcpy(dst_addr, tmp, ETH_ALEN);
 453
 454         skb->pkt_type = PACKET_OUTGOING;
 455         dn_rt_finish_output(skb, dst_addr, src_addr);
 456         return NET_RX_SUCCESS;
 457 }
 458
 459 /**
 460  * dn_route_rx_packet - Try and find a route for an incoming packet
 461  * @skb: The packet to find a route for
 462  *
 463  * Returns: result of input function if route is found, error code otherwise
 464  */
 465 static int dn_route_rx_packet(struct sk_buff *skb)
 466 {
 467         struct dn_skb_cb *cb = DN_SKB_CB(skb);
 468         int err;
 469
 470         if ((err = dn_route_input(skb)) == 0)
 471                 return dst_input(skb);
 472
 473         if (decnet_debug_level & 4) {
 474                 char *devname = skb->dev ? skb->dev->name : "???";
 475                 struct dn_skb_cb *cb = DN_SKB_CB(skb);
 476                 printk(KERN_DEBUG
 477                         "DECnet: dn_route_rx_packet: rt_flags=0x%02x dev=%s len=%d src=0x%04hx dst=0x%04hx err=%d type=%d\n",
 478                         (int)cb->rt_flags, devname, skb->len,
 479                         dn_ntohs(cb->src), dn_ntohs(cb->dst),
 480                         err, skb->pkt_type);
 481         }
 482
 483         if ((skb->pkt_type == PACKET_HOST) && (cb->rt_flags & DN_RT_F_RQR)) {
 484                 switch(cb->rt_flags & DN_RT_PKT_MSK) {
 485                         case DN_RT_PKT_SHORT:
 486                                 return dn_return_short(skb);
 487                         case DN_RT_PKT_LONG:
 488                                 return dn_return_long(skb);
 489                 }
 490         }
 491
 492         kfree_skb(skb);
 493         return NET_RX_DROP;
 494 }
 495
 496 static int dn_route_rx_long(struct sk_buff *skb)
 497 {
 498         struct dn_skb_cb *cb = DN_SKB_CB(skb);
 499         unsigned char *ptr = skb->data;
 500
 501         if (!pskb_may_pull(skb, 21)) /* 20 for long header, 1 for shortest nsp */
 502                 goto drop_it;
 503
 504         skb_pull(skb, 20);
 505         skb->h.raw = skb->data;
 506
 507         /* Destination info */
 508         ptr += 2;
 509         cb->dst = dn_eth2dn(ptr);
 510         if (memcmp(ptr, dn_hiord_addr, 4) != 0)
 511                 goto drop_it;
 512         ptr += 6;
 513
 514
 515         /* Source info */
 516         ptr += 2;
 517         cb->src = dn_eth2dn(ptr);
 518         if (memcmp(ptr, dn_hiord_addr, 4) != 0)
 519                 goto drop_it;
 520         ptr += 6;
 521         /* Other junk */
 522         ptr++;
 523         cb->hops = *ptr++; /* Visit Count */
 524
 525         return NF_HOOK(PF_DECnet, NF_DN_PRE_ROUTING, skb, skb->dev, NULL, dn_route_rx_packet);
 526
 527 drop_it:
 528         kfree_skb(skb);
 529         return NET_RX_DROP;
 530 }
 531
 532
 533
 534 static int dn_route_rx_short(struct sk_buff *skb)
 535 {
 536         struct dn_skb_cb *cb = DN_SKB_CB(skb);
 537         unsigned char *ptr = skb->data;
 538
 539         if (!pskb_may_pull(skb, 6)) /* 5 for short header + 1 for shortest nsp */
 540                 goto drop_it;
 541
 542         skb_pull(skb, 5);
 543         skb->h.raw = skb->data;
 544
 545         cb->dst = *(__le16 *)ptr;
 546         ptr += 2;
 547         cb->src = *(__le16 *)ptr;
 548         ptr += 2;
 549         cb->hops = *ptr & 0x3f;
 550
 551         return NF_HOOK(PF_DECnet, NF_DN_PRE_ROUTING, skb, skb->dev, NULL, dn_route_rx_packet);
 552
 553 drop_it:
 554         kfree_skb(skb);
 555         return NET_RX_DROP;
 556 }
 557
 558 static int dn_route_discard(struct sk_buff *skb)
 559 {
 560         /*
 561          * I know we drop the packet here, but thats considered success in
 562          * this case
 563          */
 564         kfree_skb(skb);
 565         return NET_RX_SUCCESS;
 566 }
 567
 568 static int dn_route_ptp_hello(struct sk_buff *skb)
 569 {
 570         dn_dev_hello(skb);
 571         dn_neigh_pointopoint_hello(skb);
 572         return NET_RX_SUCCESS;
 573 }
 574
 575 int dn_route_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
 576 {
 577         struct dn_skb_cb *cb;
 578         unsigned char flags = 0;
 579         __u16 len = dn_ntohs(*(__le16 *)skb->data);
 580         struct dn_dev *dn = (struct dn_dev *)dev->dn_ptr;
 581         unsigned char padlen = 0;
 582
 583         if (dn == NULL)
 584                 goto dump_it;
 585
 586         if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
 587                 goto out;
 588
 589         if (!pskb_may_pull(skb, 3))
 590                 goto dump_it;
 591
 592         skb_pull(skb, 2);
 593
 594         if (len > skb->len)
 595                 goto dump_it;
 596
 597         skb_trim(skb, len);
 598
 599         flags = *skb->data;
 600
 601         cb = DN_SKB_CB(skb);
 602         cb->stamp = jiffies;
 603         cb->iif = dev->ifindex;
 604
 605         /*
 606          * If we have padding, remove it.
 607          */
 608         if (flags & DN_RT_F_PF) {
 609                 padlen = flags & ~DN_RT_F_PF;
 610                 if (!pskb_may_pull(skb, padlen + 1))
 611                         goto dump_it;
 612                 skb_pull(skb, padlen);
 613                 flags = *skb->data;
 614         }
 615
 616         skb->nh.raw = skb->data;
 617
 618         /*
 619          * Weed out future version DECnet
 620          */
 621         if (flags & DN_RT_F_VER)
 622                 goto dump_it;
 623
 624         cb->rt_flags = flags;
 625
 626         if (decnet_debug_level & 1)
 627                 printk(KERN_DEBUG
 628                         "dn_route_rcv: got 0x%02x from %s [%d %d %d]\n",
 629                         (int)flags, (dev) ? dev->name : "???", len, skb->len,
 630                         padlen);
 631
 632         if (flags & DN_RT_PKT_CNTL) {
 633                 if (unlikely(skb_is_nonlinear(skb)) &&
 634                     skb_linearize(skb, GFP_ATOMIC) != 0)
 635                         goto dump_it;
 636
 637                 switch(flags & DN_RT_CNTL_MSK) {
 638                         case DN_RT_PKT_INIT:
 639                                 dn_dev_init_pkt(skb);
 640                                 break;
 641                         case DN_RT_PKT_VERI:
 642                                 dn_dev_veri_pkt(skb);
 643                                 break;
 644                 }
 645
 646                 if (dn->parms.state != DN_DEV_S_RU)
 647                         goto dump_it;
 648
 649                 switch(flags & DN_RT_CNTL_MSK) {
 650                         case DN_RT_PKT_HELO:
 651                                 return NF_HOOK(PF_DECnet, NF_DN_HELLO, skb, skb->dev, NULL, dn_route_ptp_hello);
 652
 653                         case DN_RT_PKT_L1RT:
 654                         case DN_RT_PKT_L2RT:
 655                                 return NF_HOOK(PF_DECnet, NF_DN_ROUTE, skb, skb->dev, NULL, dn_route_discard);
 656                         case DN_RT_PKT_ERTH:
 657                                 return NF_HOOK(PF_DECnet, NF_DN_HELLO, skb, skb->dev, NULL, dn_neigh_router_hello);
 658
 659                         case DN_RT_PKT_EEDH:
 660                                 return NF_HOOK(PF_DECnet, NF_DN_HELLO, skb, skb->dev, NULL, dn_neigh_endnode_hello);
 661                 }
 662         } else {
 663                 if (dn->parms.state != DN_DEV_S_RU)
 664                         goto dump_it;
 665
 666                 skb_pull(skb, 1); /* Pull flags */
 667
 668                 switch(flags & DN_RT_PKT_MSK) {
 669                         case DN_RT_PKT_LONG:
 670                                 return dn_route_rx_long(skb);
 671                         case DN_RT_PKT_SHORT:
 672                                 return dn_route_rx_short(skb);
 673                 }
 674         }
 675
 676 dump_it:
 677         kfree_skb(skb);
 678 out:
 679         return NET_RX_DROP;
 680 }
 681
 682 static int dn_output(struct sk_buff *skb)
 683 {
 684         struct dst_entry *dst = skb->dst;
 685         struct dn_route *rt = (struct dn_route *)dst;
 686         struct net_device *dev = dst->dev;
 687         struct dn_skb_cb *cb = DN_SKB_CB(skb);
 688         struct neighbour *neigh;
 689
 690         int err = -EINVAL;
 691
 692         if ((neigh = dst->neighbour) == NULL)
 693                 goto error;
 694
 695         skb->dev = dev;
 696
 697         cb->src = rt->rt_saddr;
 698         cb->dst = rt->rt_daddr;
 699
 700         /*
 701          * Always set the Intra-Ethernet bit on all outgoing packets
 702          * originated on this node. Only valid flag from upper layers
 703          * is return-to-sender-requested. Set hop count to 0 too.
 704          */
 705         cb->rt_flags &= ~DN_RT_F_RQR;
 706         cb->rt_flags |= DN_RT_F_IE;
 707         cb->hops = 0;
 708
 709         return NF_HOOK(PF_DECnet, NF_DN_LOCAL_OUT, skb, NULL, dev, neigh->output);
 710
 711 error:
 712         if (net_ratelimit())
 713                 printk(KERN_DEBUG "dn_output: This should not happen\n");
 714
 715         kfree_skb(skb);
 716
 717         return err;
 718 }
 719
 720 static int dn_forward(struct sk_buff *skb)
 721 {
 722         struct dn_skb_cb *cb = DN_SKB_CB(skb);
 723         struct dst_entry *dst = skb->dst;
 724         struct dn_dev *dn_db = dst->dev->dn_ptr;
 725         struct dn_route *rt;
 726         struct neighbour *neigh = dst->neighbour;
 727         int header_len;
 728 #ifdef CONFIG_NETFILTER
 729         struct net_device *dev = skb->dev;
 730 #endif
 731
 732         if (skb->pkt_type != PACKET_HOST)
 733                 goto drop;
 734
 735         /* Ensure that we have enough space for headers */
 736         rt = (struct dn_route *)skb->dst;
 737         header_len = dn_db->use_long ? 21 : 6;
 738         if (skb_cow(skb, LL_RESERVED_SPACE(rt->u.dst.dev)+header_len))
 739                 goto drop;
 740
 741         /*
 742          * Hop count exceeded.
 743          */
 744         if (++cb->hops > 30)
 745                 goto drop;
 746
 747         skb->dev = rt->u.dst.dev;
 748
 749         /*
 750          * If packet goes out same interface it came in on, then set
 751          * the Intra-Ethernet bit. This has no effect for short
 752          * packets, so we don't need to test for them here.
 753          */
 754         cb->rt_flags &= ~DN_RT_F_IE;
 755         if (rt->rt_flags & RTCF_DOREDIRECT)
 756                 cb->rt_flags |= DN_RT_F_IE;
 757
 758         return NF_HOOK(PF_DECnet, NF_DN_FORWARD, skb, dev, skb->dev, neigh->output);
 759
 760 drop:
 761         kfree_skb(skb);
 762         return NET_RX_DROP;
 763 }
 764
 765 /*
 766  * Drop packet. This is used for endnodes and for
 767  * when we should not be forwarding packets from
 768  * this dest.
 769  */
 770 static int dn_blackhole(struct sk_buff *skb)
 771 {
 772         kfree_skb(skb);
 773         return NET_RX_DROP;
 774 }
 775
 776 /*
 777  * Used to catch bugs. This should never normally get
 778  * called.
 779  */
 780 static int dn_rt_bug(struct sk_buff *skb)
 781 {
 782         if (net_ratelimit()) {
 783                 struct dn_skb_cb *cb = DN_SKB_CB(skb);
 784
 785                 printk(KERN_DEBUG "dn_rt_bug: skb from:%04x to:%04x\n",
 786                                 dn_ntohs(cb->src), dn_ntohs(cb->dst));
 787         }
 788
 789         kfree_skb(skb);
 790
 791         return NET_RX_BAD;
 792 }
 793
 794 static int dn_rt_set_next_hop(struct dn_route *rt, struct dn_fib_res *res)
 795 {
 796         struct dn_fib_info *fi = res->fi;
 797         struct net_device *dev = rt->u.dst.dev;
 798         struct neighbour *n;
 799         unsigned mss;
 800
 801         if (fi) {
 802                 if (DN_FIB_RES_GW(*res) &&
 803                     DN_FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
 804                         rt->rt_gateway = DN_FIB_RES_GW(*res);
 805                 memcpy(rt->u.dst.metrics, fi->fib_metrics,
 806                        sizeof(rt->u.dst.metrics));
 807         }
 808         rt->rt_type = res->type;
 809
 810         if (dev != NULL && rt->u.dst.neighbour == NULL) {
 811                 n = __neigh_lookup_errno(&dn_neigh_table, &rt->rt_gateway, dev);
 812                 if (IS_ERR(n))
 813                         return PTR_ERR(n);
 814                 rt->u.dst.neighbour = n;
 815         }
 816
 817         if (rt->u.dst.metrics[RTAX_MTU-1] == 0 ||
 818             rt->u.dst.metrics[RTAX_MTU-1] > rt->u.dst.dev->mtu)
 819                 rt->u.dst.metrics[RTAX_MTU-1] = rt->u.dst.dev->mtu;
 820         mss = dn_mss_from_pmtu(dev, dst_mtu(&rt->u.dst));
 821         if (rt->u.dst.metrics[RTAX_ADVMSS-1] == 0 ||
 822             rt->u.dst.metrics[RTAX_ADVMSS-1] > mss)
 823                 rt->u.dst.metrics[RTAX_ADVMSS-1] = mss;
 824         return 0;
 825 }
 826
 827 static inline int dn_match_addr(__le16 addr1, __le16 addr2)
 828 {
 829         __u16 tmp = dn_ntohs(addr1) ^ dn_ntohs(addr2);
 830         int match = 16;
 831         while(tmp) {
 832                 tmp >>= 1;
 833                 match--;
 834         }
 835         return match;
 836 }
 837
 838 static __le16 dnet_select_source(const struct net_device *dev, __le16 daddr, int scope)
 839 {
 840         __le16 saddr = 0;
 841         struct dn_dev *dn_db = dev->dn_ptr;
 842         struct dn_ifaddr *ifa;
 843         int best_match = 0;
 844         int ret;
 845
 846         read_lock(&dev_base_lock);
 847         for(ifa = dn_db->ifa_list; ifa; ifa = ifa->ifa_next) {
 848                 if (ifa->ifa_scope > scope)
 849                         continue;
 850                 if (!daddr) {
 851                         saddr = ifa->ifa_local;
 852                         break;
 853                 }
 854                 ret = dn_match_addr(daddr, ifa->ifa_local);
 855                 if (ret > best_match)
 856                         saddr = ifa->ifa_local;
 857                 if (best_match == 0)
 858                         saddr = ifa->ifa_local;
 859         }
 860         read_unlock(&dev_base_lock);
 861
 862         return saddr;
 863 }
 864
 865 static inline __le16 __dn_fib_res_prefsrc(struct dn_fib_res *res)
 866 {
 867         return dnet_select_source(DN_FIB_RES_DEV(*res), DN_FIB_RES_GW(*res), res->scope);
 868 }
 869
 870 static inline __le16 dn_fib_rules_map_destination(__le16 daddr, struct dn_fib_res *res)
 871 {
 872         __le16 mask = dnet_make_mask(res->prefixlen);
 873         return (daddr&~mask)|res->fi->fib_nh->nh_gw;
 874 }
 875
 876 static int dn_route_output_slow(struct dst_entry **pprt, const struct flowi *oldflp, int try_hard)
 877 {
 878         struct flowi fl = { .nl_u = { .dn_u =
 879                                       { .daddr = oldflp->fld_dst,
 880                                         .saddr = oldflp->fld_src,
 881                                         .scope = RT_SCOPE_UNIVERSE,
 882 #ifdef CONFIG_DECNET_ROUTE_FWMARK
 883                                         .fwmark = oldflp->fld_fwmark
 884 #endif
 885                                      } },
 886                             .iif = loopback_dev.ifindex,
 887                             .oif = oldflp->oif };
 888         struct dn_route *rt = NULL;
 889         struct net_device *dev_out = NULL;
 890         struct neighbour *neigh = NULL;
 891         unsigned hash;
 892         unsigned flags = 0;
 893         struct dn_fib_res res = { .fi = NULL, .type = RTN_UNICAST };
 894         int err;
 895         int free_res = 0;
 896         __le16 gateway = 0;
 897
 898         if (decnet_debug_level & 16)
 899                 printk(KERN_DEBUG
 900                        "dn_route_output_slow: dst=%04x src=%04x mark=%d"
 901                        " iif=%d oif=%d\n", dn_ntohs(oldflp->fld_dst),
 902                        dn_ntohs(oldflp->fld_src),
 903                        oldflp->fld_fwmark, loopback_dev.ifindex, oldflp->oif);
 904
 905         /* If we have an output interface, verify its a DECnet device */
 906         if (oldflp->oif) {
 907                 dev_out = dev_get_by_index(oldflp->oif);
 908                 err = -ENODEV;
 909                 if (dev_out && dev_out->dn_ptr == NULL) {
 910                         dev_put(dev_out);
 911                         dev_out = NULL;
 912                 }
 913                 if (dev_out == NULL)
 914                         goto out;
 915         }
 916
 917         /* If we have a source address, verify that its a local address */
 918         if (oldflp->fld_src) {
 919                 err = -EADDRNOTAVAIL;
 920
 921                 if (dev_out) {
 922                         if (dn_dev_islocal(dev_out, oldflp->fld_src))
 923                                 goto source_ok;
 924                         dev_put(dev_out);
 925                         goto out;
 926                 }
 927                 read_lock(&dev_base_lock);
 928                 for(dev_out = dev_base; dev_out; dev_out = dev_out->next) {
 929                         if (!dev_out->dn_ptr)
 930                                 continue;
 931                         if (dn_dev_islocal(dev_out, oldflp->fld_src))
 932                                 break;
 933                 }
 934                 read_unlock(&dev_base_lock);
 935                 if (dev_out == NULL)
 936                         goto out;
 937                 dev_hold(dev_out);
 938 source_ok:
 939                 ;
 940         }
 941
 942         /* No destination? Assume its local */
 943         if (!fl.fld_dst) {
 944                 fl.fld_dst = fl.fld_src;
 945
 946                 err = -EADDRNOTAVAIL;
 947                 if (dev_out)
 948                         dev_put(dev_out);
 949                 dev_out = &loopback_dev;
 950                 dev_hold(dev_out);
 951                 if (!fl.fld_dst) {
 952                         fl.fld_dst =
 953                         fl.fld_src = dnet_select_source(dev_out, 0,
 954                                                        RT_SCOPE_HOST);
 955                         if (!fl.fld_dst)
 956                                 goto out;
 957                 }
 958                 fl.oif = loopback_dev.ifindex;
 959                 res.type = RTN_LOCAL;
 960                 goto make_route;
 961         }
 962
 963         if (decnet_debug_level & 16)
 964                 printk(KERN_DEBUG
 965                        "dn_route_output_slow: initial checks complete."
 966                        " dst=%o4x src=%04x oif=%d try_hard=%d\n",
 967                        dn_ntohs(fl.fld_dst), dn_ntohs(fl.fld_src),
 968                        fl.oif, try_hard);
 969
 970         /*
 971          * N.B. If the kernel is compiled without router support then
 972          * dn_fib_lookup() will evaluate to non-zero so this if () block
 973          * will always be executed.
 974          */
 975         err = -ESRCH;
 976         if (try_hard || (err = dn_fib_lookup(&fl, &res)) != 0) {
 977                 struct dn_dev *dn_db;
 978                 if (err != -ESRCH)
 979                         goto out;
 980                 /*
 981                  * Here the fallback is basically the standard algorithm for
 982                  * routing in endnodes which is described in the DECnet routing
 983                  * docs
 984                  *
 985                  * If we are not trying hard, look in neighbour cache.
 986                  * The result is tested to ensure that if a specific output
 987                  * device/source address was requested, then we honour that
 988                  * here
 989                  */
 990                 if (!try_hard) {
 991                         neigh = neigh_lookup_nodev(&dn_neigh_table, &fl.fld_dst);
 992                         if (neigh) {
 993                                 if ((oldflp->oif &&
 994                                     (neigh->dev->ifindex != oldflp->oif)) ||
 995                                     (oldflp->fld_src &&
 996                                     (!dn_dev_islocal(neigh->dev,
 997                                                       oldflp->fld_src)))) {
 998                                         neigh_release(neigh);
 999                                         neigh = NULL;
1000                                 } else {
1001                                         if (dev_out)
1002                                                 dev_put(dev_out);
1003                                         if (dn_dev_islocal(neigh->dev, fl.fld_dst)) {
1004                                                 dev_out = &loopback_dev;
1005                                                 res.type = RTN_LOCAL;
1006                                         } else {
1007                                                 dev_out = neigh->dev;
1008                                         }
1009                                         dev_hold(dev_out);
1010                                         goto select_source;
1011                                 }
1012                         }
1013                 }
1014
1015                 /* Not there? Perhaps its a local address */
1016                 if (dev_out == NULL)
1017                         dev_out = dn_dev_get_default();
1018                 err = -ENODEV;
1019                 if (dev_out == NULL)
1020                         goto out;
1021                 dn_db = dev_out->dn_ptr;
1022                 /* Possible improvement - check all devices for local addr */
1023                 if (dn_dev_islocal(dev_out, fl.fld_dst)) {
1024                         dev_put(dev_out);
1025                         dev_out = &loopback_dev;
1026                         dev_hold(dev_out);
1027                         res.type = RTN_LOCAL;
1028                         goto select_source;
1029                 }
1030                 /* Not local either.... try sending it to the default router */
1031                 neigh = neigh_clone(dn_db->router);
1032                 BUG_ON(neigh && neigh->dev != dev_out);
1033
1034                 /* Ok then, we assume its directly connected and move on */
1035 select_source:
1036                 if (neigh)
1037                         gateway = ((struct dn_neigh *)neigh)->addr;
1038                 if (gateway == 0)
1039                         gateway = fl.fld_dst;
1040                 if (fl.fld_src == 0) {
1041                         fl.fld_src = dnet_select_source(dev_out, gateway,
1042                                                          res.type == RTN_LOCAL ?
1043                                                          RT_SCOPE_HOST :
1044                                                          RT_SCOPE_LINK);
1045                         if (fl.fld_src == 0 && res.type != RTN_LOCAL)
1046                                 goto e_addr;
1047                 }
1048                 fl.oif = dev_out->ifindex;
1049                 goto make_route;
1050         }
1051         free_res = 1;
1052
1053         if (res.type == RTN_NAT)
1054                 goto e_inval;
1055
1056         if (res.type == RTN_LOCAL) {
1057                 if (!fl.fld_src)
1058                         fl.fld_src = fl.fld_dst;
1059                 if (dev_out)
1060                         dev_put(dev_out);
1061                 dev_out = &loopback_dev;
1062                 dev_hold(dev_out);
1063                 fl.oif = dev_out->ifindex;
1064                 if (res.fi)
1065                         dn_fib_info_put(res.fi);
1066                 res.fi = NULL;
1067                 goto make_route;
1068         }
1069
1070         if (res.fi->fib_nhs > 1 && fl.oif == 0)
1071                 dn_fib_select_multipath(&fl, &res);
1072
1073         /*
1074          * We could add some logic to deal with default routes here and
1075          * get rid of some of the special casing above.
1076          */
1077
1078         if (!fl.fld_src)
1079                 fl.fld_src = DN_FIB_RES_PREFSRC(res);
1080
1081         if (dev_out)
1082                 dev_put(dev_out);
1083         dev_out = DN_FIB_RES_DEV(res);
1084         dev_hold(dev_out);
1085         fl.oif = dev_out->ifindex;
1086         gateway = DN_FIB_RES_GW(res);
1087
1088 make_route:
1089         if (dev_out->flags & IFF_LOOPBACK)
1090                 flags |= RTCF_LOCAL;
1091
1092         rt = dst_alloc(&dn_dst_ops);
1093         if (rt == NULL)
1094                 goto e_nobufs;
1095
1096         atomic_set(&rt->u.dst.__refcnt, 1);
1097         rt->u.dst.flags   = DST_HOST;
1098
1099         rt->fl.fld_src    = oldflp->fld_src;
1100         rt->fl.fld_dst    = oldflp->fld_dst;
1101         rt->fl.oif        = oldflp->oif;
1102         rt->fl.iif        = 0;
1103 #ifdef CONFIG_DECNET_ROUTE_FWMARK
1104         rt->fl.fld_fwmark = oldflp->fld_fwmark;
1105 #endif
1106
1107         rt->rt_saddr      = fl.fld_src;
1108         rt->rt_daddr      = fl.fld_dst;
1109         rt->rt_gateway    = gateway ? gateway : fl.fld_dst;
1110         rt->rt_local_src  = fl.fld_src;
1111
1112         rt->rt_dst_map    = fl.fld_dst;
1113         rt->rt_src_map    = fl.fld_src;
1114
1115         rt->u.dst.dev = dev_out;
1116         dev_hold(dev_out);
1117         rt->u.dst.neighbour = neigh;
1118         neigh = NULL;
1119
1120         rt->u.dst.lastuse = jiffies;
1121         rt->u.dst.output  = dn_output;
1122         rt->u.dst.input   = dn_rt_bug;
1123         rt->rt_flags      = flags;
1124         if (flags & RTCF_LOCAL)
1125                 rt->u.dst.input = dn_nsp_rx;
1126
1127         err = dn_rt_set_next_hop(rt, &res);
1128         if (err)
1129                 goto e_neighbour;
1130
1131         hash = dn_hash(rt->fl.fld_src, rt->fl.fld_dst);
1132         dn_insert_route(rt, hash, (struct dn_route **)pprt);
1133
1134 done:
1135         if (neigh)
1136                 neigh_release(neigh);
1137         if (free_res)
1138                 dn_fib_res_put(&res);
1139         if (dev_out)
1140                 dev_put(dev_out);
1141 out:
1142         return err;
1143
1144 e_addr:
1145         err = -EADDRNOTAVAIL;
1146         goto done;
1147 e_inval:
1148         err = -EINVAL;
1149         goto done;
1150 e_nobufs:
1151         err = -ENOBUFS;
1152         goto done;
1153 e_neighbour:
1154         dst_free(&rt->u.dst);
1155         goto e_nobufs;
1156 }
1157
1158
1159 /*
1160  * N.B. The flags may be moved into the flowi at some future stage.
1161  */
1162 static int __dn_route_output_key(struct dst_entry **pprt, const struct flowi *flp, int flags)
1163 {
1164         unsigned hash = dn_hash(flp->fld_src, flp->fld_dst);
1165         struct dn_route *rt = NULL;
1166
1167         if (!(flags & MSG_TRYHARD)) {
1168                 rcu_read_lock_bh();
1169                 for(rt = rcu_dereference(dn_rt_hash_table[hash].chain); rt;
1170                         rt = rcu_dereference(rt->u.rt_next)) {
1171                         if ((flp->fld_dst == rt->fl.fld_dst) &&
1172                             (flp->fld_src == rt->fl.fld_src) &&
1173 #ifdef CONFIG_DECNET_ROUTE_FWMARK
1174                             (flp->fld_fwmark == rt->fl.fld_fwmark) &&
1175 #endif
1176                             (rt->fl.iif == 0) &&
1177                             (rt->fl.oif == flp->oif)) {
1178                                 rt->u.dst.lastuse = jiffies;
1179                                 dst_hold(&rt->u.dst);
1180                                 rt->u.dst.__use++;
1181                                 rcu_read_unlock_bh();
1182                                 *pprt = &rt->u.dst;
1183                                 return 0;
1184                         }
1185                 }
1186                 rcu_read_unlock_bh();
1187         }
1188
1189         return dn_route_output_slow(pprt, flp, flags);
1190 }
1191
1192 static int dn_route_output_key(struct dst_entry **pprt, struct flowi *flp, int flags)
1193 {
1194         int err;
1195
1196         err = __dn_route_output_key(pprt, flp, flags);
1197         if (err == 0 && flp->proto) {
1198                 err = xfrm_lookup(pprt, flp, NULL, 0);
1199         }
1200         return err;
1201 }
1202
1203 int dn_route_output_sock(struct dst_entry **pprt, struct flowi *fl, struct sock *sk, int flags)
1204 {
1205         int err;
1206
1207         err = __dn_route_output_key(pprt, fl, flags & MSG_TRYHARD);
1208         if (err == 0 && fl->proto) {
1209                 err = xfrm_lookup(pprt, fl, sk, !(flags & MSG_DONTWAIT));
1210         }
1211         return err;
1212 }
1213
1214 static int dn_route_input_slow(struct sk_buff *skb)
1215 {
1216         struct dn_route *rt = NULL;
1217         struct dn_skb_cb *cb = DN_SKB_CB(skb);
1218         struct net_device *in_dev = skb->dev;
1219         struct net_device *out_dev = NULL;
1220         struct dn_dev *dn_db;
1221         struct neighbour *neigh = NULL;
1222         unsigned hash;
1223         int flags = 0;
1224         __le16 gateway = 0;
1225         __le16 local_src = 0;
1226         struct flowi fl = { .nl_u = { .dn_u =
1227                                      { .daddr = cb->dst,
1228                                        .saddr = cb->src,
1229                                        .scope = RT_SCOPE_UNIVERSE,
1230 #ifdef CONFIG_DECNET_ROUTE_FWMARK
1231                                        .fwmark = skb->nfmark
1232 #endif
1233                                     } },
1234                             .iif = skb->dev->ifindex };
1235         struct dn_fib_res res = { .fi = NULL, .type = RTN_UNREACHABLE };
1236         int err = -EINVAL;
1237         int free_res = 0;
1238
1239         dev_hold(in_dev);
1240
1241         if ((dn_db = in_dev->dn_ptr) == NULL)
1242                 goto out;
1243
1244         /* Zero source addresses are not allowed */
1245         if (fl.fld_src == 0)
1246                 goto out;
1247
1248         /*
1249          * In this case we've just received a packet from a source
1250          * outside ourselves pretending to come from us. We don't
1251          * allow it any further to prevent routing loops, spoofing and
1252          * other nasties. Loopback packets already have the dst attached
1253          * so this only affects packets which have originated elsewhere.
1254          */
1255         err  = -ENOTUNIQ;
1256         if (dn_dev_islocal(in_dev, cb->src))
1257                 goto out;
1258
1259         err = dn_fib_lookup(&fl, &res);
1260         if (err) {
1261                 if (err != -ESRCH)
1262                         goto out;
1263                 /*
1264                  * Is the destination us ?
1265                  */
1266                 if (!dn_dev_islocal(in_dev, cb->dst))
1267                         goto e_inval;
1268
1269                 res.type = RTN_LOCAL;
1270                 flags |= RTCF_DIRECTSRC;
1271         } else {
1272                 __le16 src_map = fl.fld_src;
1273                 free_res = 1;
1274
1275                 out_dev = DN_FIB_RES_DEV(res);
1276                 if (out_dev == NULL) {
1277                         if (net_ratelimit())
1278                                 printk(KERN_CRIT "Bug in dn_route_input_slow() "
1279                                                  "No output device\n");
1280                         goto e_inval;
1281                 }
1282                 dev_hold(out_dev);
1283
1284                 if (res.r)
1285                         src_map = dn_fib_rules_policy(fl.fld_src, &res, &flags);
1286
1287                 gateway = DN_FIB_RES_GW(res);
1288                 if (res.type == RTN_NAT) {
1289                         fl.fld_dst = dn_fib_rules_map_destination(fl.fld_dst, &res);
1290                         dn_fib_res_put(&res);
1291                         free_res = 0;
1292                         if (dn_fib_lookup(&fl, &res))
1293                                 goto e_inval;
1294                         free_res = 1;
1295                         if (res.type != RTN_UNICAST)
1296                                 goto e_inval;
1297                         flags |= RTCF_DNAT;
1298                         gateway = fl.fld_dst;
1299                 }
1300                 fl.fld_src = src_map;
1301         }
1302
1303         switch(res.type) {
1304         case RTN_UNICAST:
1305                 /*
1306                  * Forwarding check here, we only check for forwarding
1307                  * being turned off, if you want to only forward intra
1308                  * area, its up to you to set the routing tables up
1309                  * correctly.
1310                  */
1311                 if (dn_db->parms.forwarding == 0)
1312                         goto e_inval;
1313
1314                 if (res.fi->fib_nhs > 1 && fl.oif == 0)
1315                         dn_fib_select_multipath(&fl, &res);
1316
1317                 /*
1318                  * Check for out_dev == in_dev. We use the RTCF_DOREDIRECT
1319                  * flag as a hint to set the intra-ethernet bit when
1320                  * forwarding. If we've got NAT in operation, we don't do
1321                  * this optimisation.
1322                  */
1323                 if (out_dev == in_dev && !(flags & RTCF_NAT))
1324                         flags |= RTCF_DOREDIRECT;
1325
1326                 local_src = DN_FIB_RES_PREFSRC(res);
1327
1328         case RTN_BLACKHOLE:
1329         case RTN_UNREACHABLE:
1330                 break;
1331         case RTN_LOCAL:
1332                 flags |= RTCF_LOCAL;
1333                 fl.fld_src = cb->dst;
1334                 fl.fld_dst = cb->src;
1335
1336                 /* Routing tables gave us a gateway */
1337                 if (gateway)
1338                         goto make_route;
1339
1340                 /* Packet was intra-ethernet, so we know its on-link */
1341                 if (cb->rt_flags | DN_RT_F_IE) {
1342                         gateway = cb->src;
1343                         flags |= RTCF_DIRECTSRC;
1344                         goto make_route;
1345                 }
1346
1347                 /* Use the default router if there is one */
1348                 neigh = neigh_clone(dn_db->router);
1349                 if (neigh) {
1350                         gateway = ((struct dn_neigh *)neigh)->addr;
1351                         goto make_route;
1352                 }
1353
1354                 /* Close eyes and pray */
1355                 gateway = cb->src;
1356                 flags |= RTCF_DIRECTSRC;
1357                 goto make_route;
1358         default:
1359                 goto e_inval;
1360         }
1361
1362 make_route:
1363         rt = dst_alloc(&dn_dst_ops);
1364         if (rt == NULL)
1365                 goto e_nobufs;
1366
1367         rt->rt_saddr      = fl.fld_src;
1368         rt->rt_daddr      = fl.fld_dst;
1369         rt->rt_gateway    = fl.fld_dst;
1370         if (gateway)
1371                 rt->rt_gateway = gateway;
1372         rt->rt_local_src  = local_src ? local_src : rt->rt_saddr;
1373
1374         rt->rt_dst_map    = fl.fld_dst;
1375         rt->rt_src_map    = fl.fld_src;
1376
1377         rt->fl.fld_src    = cb->src;
1378         rt->fl.fld_dst    = cb->dst;
1379         rt->fl.oif        = 0;
1380         rt->fl.iif        = in_dev->ifindex;
1381         rt->fl.fld_fwmark = fl.fld_fwmark;
1382
1383         rt->u.dst.flags = DST_HOST;
1384         rt->u.dst.neighbour = neigh;
1385         rt->u.dst.dev = out_dev;
1386         rt->u.dst.lastuse = jiffies;
1387         rt->u.dst.output = dn_rt_bug;
1388         switch(res.type) {
1389                 case RTN_UNICAST:
1390                         rt->u.dst.input = dn_forward;
1391                         break;
1392                 case RTN_LOCAL:
1393                         rt->u.dst.output = dn_output;
1394                         rt->u.dst.input = dn_nsp_rx;
1395                         rt->u.dst.dev = in_dev;
1396                         flags |= RTCF_LOCAL;
1397                         break;
1398                 default:
1399                 case RTN_UNREACHABLE:
1400                 case RTN_BLACKHOLE:
1401                         rt->u.dst.input = dn_blackhole;
1402         }
1403         rt->rt_flags = flags;
1404         if (rt->u.dst.dev)
1405                 dev_hold(rt->u.dst.dev);
1406
1407         err = dn_rt_set_next_hop(rt, &res);
1408         if (err)
1409                 goto e_neighbour;
1410
1411         hash = dn_hash(rt->fl.fld_src, rt->fl.fld_dst);
1412         dn_insert_route(rt, hash, (struct dn_route **)&skb->dst);
1413
1414 done:
1415         if (neigh)
1416                 neigh_release(neigh);
1417         if (free_res)
1418                 dn_fib_res_put(&res);
1419         dev_put(in_dev);
1420         if (out_dev)
1421                 dev_put(out_dev);
1422 out:
1423         return err;
1424
1425 e_inval:
1426         err = -EINVAL;
1427         goto done;
1428
1429 e_nobufs:
1430         err = -ENOBUFS;
1431         goto done;
1432
1433 e_neighbour:
1434         dst_free(&rt->u.dst);
1435         goto done;
1436 }
1437
1438 int dn_route_input(struct sk_buff *skb)
1439 {
1440         struct dn_route *rt;
1441         struct dn_skb_cb *cb = DN_SKB_CB(skb);
1442         unsigned hash = dn_hash(cb->src, cb->dst);
1443
1444         if (skb->dst)
1445                 return 0;
1446
1447         rcu_read_lock();
1448         for(rt = rcu_dereference(dn_rt_hash_table[hash].chain); rt != NULL;
1449             rt = rcu_dereference(rt->u.rt_next)) {
1450                 if ((rt->fl.fld_src == cb->src) &&
1451                     (rt->fl.fld_dst == cb->dst) &&
1452                     (rt->fl.oif == 0) &&
1453 #ifdef CONFIG_DECNET_ROUTE_FWMARK
1454                     (rt->fl.fld_fwmark == skb->nfmark) &&
1455 #endif
1456                     (rt->fl.iif == cb->iif)) {
1457                         rt->u.dst.lastuse = jiffies;
1458                         dst_hold(&rt->u.dst);
1459                         rt->u.dst.__use++;
1460                         rcu_read_unlock();
1461                         skb->dst = (struct dst_entry *)rt;
1462                         return 0;
1463                 }
1464         }
1465         rcu_read_unlock();
1466
1467         return dn_route_input_slow(skb);
1468 }
1469
1470 static int dn_rt_fill_info(struct sk_buff *skb, u32 pid, u32 seq,
1471                            int event, int nowait, unsigned int flags)
1472 {
1473         struct dn_route *rt = (struct dn_route *)skb->dst;
1474         struct rtmsg *r;
1475         struct nlmsghdr *nlh;
1476         unsigned char *b = skb->tail;
1477         struct rta_cacheinfo ci;
1478
1479         nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*r), flags);
1480         r = NLMSG_DATA(nlh);
1481         r->rtm_family = AF_DECnet;
1482         r->rtm_dst_len = 16;
1483         r->rtm_src_len = 0;
1484         r->rtm_tos = 0;
1485         r->rtm_table = RT_TABLE_MAIN;
1486         r->rtm_type = rt->rt_type;
1487         r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
1488         r->rtm_scope = RT_SCOPE_UNIVERSE;
1489         r->rtm_protocol = RTPROT_UNSPEC;
1490         if (rt->rt_flags & RTCF_NOTIFY)
1491                 r->rtm_flags |= RTM_F_NOTIFY;
1492         RTA_PUT(skb, RTA_DST, 2, &rt->rt_daddr);
1493         if (rt->fl.fld_src) {
1494                 r->rtm_src_len = 16;
1495                 RTA_PUT(skb, RTA_SRC, 2, &rt->fl.fld_src);
1496         }
1497         if (rt->u.dst.dev)
1498                 RTA_PUT(skb, RTA_OIF, sizeof(int), &rt->u.dst.dev->ifindex);
1499         /*
1500          * Note to self - change this if input routes reverse direction when
1501          * they deal only with inputs and not with replies like they do
1502          * currently.
1503          */
1504         RTA_PUT(skb, RTA_PREFSRC, 2, &rt->rt_local_src);
1505         if (rt->rt_daddr != rt->rt_gateway)
1506                 RTA_PUT(skb, RTA_GATEWAY, 2, &rt->rt_gateway);
1507         if (rtnetlink_put_metrics(skb, rt->u.dst.metrics) < 0)
1508                 goto rtattr_failure;
1509         ci.rta_lastuse = jiffies_to_clock_t(jiffies - rt->u.dst.lastuse);
1510         ci.rta_used     = rt->u.dst.__use;
1511         ci.rta_clntref  = atomic_read(&rt->u.dst.__refcnt);
1512         if (rt->u.dst.expires)
1513                 ci.rta_expires = jiffies_to_clock_t(rt->u.dst.expires - jiffies);
1514         else
1515                 ci.rta_expires = 0;
1516         ci.rta_error    = rt->u.dst.error;
1517         ci.rta_id       = ci.rta_ts = ci.rta_tsage = 0;
1518         RTA_PUT(skb, RTA_CACHEINFO, sizeof(ci), &ci);
1519         if (rt->fl.iif)
1520                 RTA_PUT(skb, RTA_IIF, sizeof(int), &rt->fl.iif);
1521
1522         nlh->nlmsg_len = skb->tail - b;
1523         return skb->len;
1524
1525 nlmsg_failure:
1526 rtattr_failure:
1527         skb_trim(skb, b - skb->data);
1528         return -1;
1529 }
1530
1531 /*
1532  * This is called by both endnodes and routers now.
1533  */
1534 int dn_cache_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh, void *arg)
1535 {
1536         struct rtattr **rta = arg;
1537         struct rtmsg *rtm = NLMSG_DATA(nlh);
1538         struct dn_route *rt = NULL;
1539         struct dn_skb_cb *cb;
1540         int err;
1541         struct sk_buff *skb;
1542         struct flowi fl;
1543
1544         memset(&fl, 0, sizeof(fl));
1545         fl.proto = DNPROTO_NSP;
1546
1547         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1548         if (skb == NULL)
1549                 return -ENOBUFS;
1550         skb->mac.raw = skb->data;
1551         cb = DN_SKB_CB(skb);
1552
1553         if (rta[RTA_SRC-1])
1554                 memcpy(&fl.fld_src, RTA_DATA(rta[RTA_SRC-1]), 2);
1555         if (rta[RTA_DST-1])
1556                 memcpy(&fl.fld_dst, RTA_DATA(rta[RTA_DST-1]), 2);
1557         if (rta[RTA_IIF-1])
1558                 memcpy(&fl.iif, RTA_DATA(rta[RTA_IIF-1]), sizeof(int));
1559
1560         if (fl.iif) {
1561                 struct net_device *dev;
1562                 if ((dev = dev_get_by_index(fl.iif)) == NULL) {
1563                         kfree_skb(skb);
1564                         return -ENODEV;
1565                 }
1566                 if (!dev->dn_ptr) {
1567                         dev_put(dev);
1568                         kfree_skb(skb);
1569                         return -ENODEV;
1570                 }
1571                 skb->protocol = __constant_htons(ETH_P_DNA_RT);
1572                 skb->dev = dev;
1573                 cb->src = fl.fld_src;
1574                 cb->dst = fl.fld_dst;
1575                 local_bh_disable();
1576                 err = dn_route_input(skb);
1577                 local_bh_enable();
1578                 memset(cb, 0, sizeof(struct dn_skb_cb));
1579                 rt = (struct dn_route *)skb->dst;
1580                 if (!err && -rt->u.dst.error)
1581                         err = rt->u.dst.error;
1582         } else {
1583                 int oif = 0;
1584                 if (rta[RTA_OIF - 1])
1585                         memcpy(&oif, RTA_DATA(rta[RTA_OIF - 1]), sizeof(int));
1586                 fl.oif = oif;
1587                 err = dn_route_output_key((struct dst_entry **)&rt, &fl, 0);
1588         }
1589
1590         if (skb->dev)
1591                 dev_put(skb->dev);
1592         skb->dev = NULL;
1593         if (err)
1594                 goto out_free;
1595         skb->dst = &rt->u.dst;
1596         if (rtm->rtm_flags & RTM_F_NOTIFY)
1597                 rt->rt_flags |= RTCF_NOTIFY;
1598
1599         NETLINK_CB(skb).dst_pid = NETLINK_CB(in_skb).pid;
1600
1601         err = dn_rt_fill_info(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq, RTM_NEWROUTE, 0, 0);
1602
1603         if (err == 0)
1604                 goto out_free;
1605         if (err < 0) {
1606                 err = -EMSGSIZE;
1607                 goto out_free;
1608         }
1609
1610         err = netlink_unicast(rtnl, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1611
1612         return err;
1613
1614 out_free:
1615         kfree_skb(skb);
1616         return err;
1617 }
1618
1619 /*
1620  * For routers, this is called from dn_fib_dump, but for endnodes its
1621  * called directly from the rtnetlink dispatch table.
1622  */
1623 int dn_cache_dump(struct sk_buff *skb, struct netlink_callback *cb)
1624 {
1625         struct dn_route *rt;
1626         int h, s_h;
1627         int idx, s_idx;
1628
1629         if (NLMSG_PAYLOAD(cb->nlh, 0) < sizeof(struct rtmsg))
1630                 return -EINVAL;
1631         if (!(((struct rtmsg *)NLMSG_DATA(cb->nlh))->rtm_flags&RTM_F_CLONED))
1632                 return 0;
1633
1634         s_h = cb->args[0];
1635         s_idx = idx = cb->args[1];
1636         for(h = 0; h <= dn_rt_hash_mask; h++) {
1637                 if (h < s_h)
1638                         continue;
1639                 if (h > s_h)
1640                         s_idx = 0;
1641                 rcu_read_lock_bh();
1642                 for(rt = rcu_dereference(dn_rt_hash_table[h].chain), idx = 0;
1643                         rt;
1644                         rt = rcu_dereference(rt->u.rt_next), idx++) {
1645                         if (idx < s_idx)
1646                                 continue;
1647                         skb->dst = dst_clone(&rt->u.dst);
1648                         if (dn_rt_fill_info(skb, NETLINK_CB(cb->skb).pid,
1649                                         cb->nlh->nlmsg_seq, RTM_NEWROUTE,
1650                                         1, NLM_F_MULTI) <= 0) {
1651                                 dst_release(xchg(&skb->dst, NULL));
1652                                 rcu_read_unlock_bh();
1653                                 goto done;
1654                         }
1655                         dst_release(xchg(&skb->dst, NULL));
1656                 }
1657                 rcu_read_unlock_bh();
1658         }
1659
1660 done:
1661         cb->args[0] = h;
1662         cb->args[1] = idx;
1663         return skb->len;
1664 }
1665
1666 #ifdef CONFIG_PROC_FS
1667 struct dn_rt_cache_iter_state {
1668         int bucket;
1669 };
1670
1671 static struct dn_route *dn_rt_cache_get_first(struct seq_file *seq)
1672 {
1673         struct dn_route *rt = NULL;
1674         struct dn_rt_cache_iter_state *s = seq->private;
1675
1676         for(s->bucket = dn_rt_hash_mask; s->bucket >= 0; --s->bucket) {
1677                 rcu_read_lock_bh();
1678                 rt = dn_rt_hash_table[s->bucket].chain;
1679                 if (rt)
1680                         break;
1681                 rcu_read_unlock_bh();
1682         }
1683         return rt;
1684 }
1685
1686 static struct dn_route *dn_rt_cache_get_next(struct seq_file *seq, struct dn_route *rt)
1687 {
1688         struct dn_rt_cache_iter_state *s = rcu_dereference(seq->private);
1689
1690         rt = rt->u.rt_next;
1691         while(!rt) {
1692                 rcu_read_unlock_bh();
1693                 if (--s->bucket < 0)
1694                         break;
1695                 rcu_read_lock_bh();
1696                 rt = dn_rt_hash_table[s->bucket].chain;
1697         }
1698         return rt;
1699 }
1700
1701 static void *dn_rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
1702 {
1703         struct dn_route *rt = dn_rt_cache_get_first(seq);
1704
1705         if (rt) {
1706                 while(*pos && (rt = dn_rt_cache_get_next(seq, rt)))
1707                         --*pos;
1708         }
1709         return *pos ? NULL : rt;
1710 }
1711
1712 static void *dn_rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1713 {
1714         struct dn_route *rt = dn_rt_cache_get_next(seq, v);
1715         ++*pos;
1716         return rt;
1717 }
1718
1719 static void dn_rt_cache_seq_stop(struct seq_file *seq, void *v)
1720 {
1721         if (v)
1722                 rcu_read_unlock_bh();
1723 }
1724
1725 static int dn_rt_cache_seq_show(struct seq_file *seq, void *v)
1726 {
1727         struct dn_route *rt = v;
1728         char buf1[DN_ASCBUF_LEN], buf2[DN_ASCBUF_LEN];
1729
1730         seq_printf(seq, "%-8s %-7s %-7s %04d %04d %04d\n",
1731                         rt->u.dst.dev ? rt->u.dst.dev->name : "*",
1732                         dn_addr2asc(dn_ntohs(rt->rt_daddr), buf1),
1733                         dn_addr2asc(dn_ntohs(rt->rt_saddr), buf2),
1734                         atomic_read(&rt->u.dst.__refcnt),
1735                         rt->u.dst.__use,
1736                         (int) dst_metric(&rt->u.dst, RTAX_RTT));
1737         return 0;
1738 }
1739
1740 static struct seq_operations dn_rt_cache_seq_ops = {
1741         .start  = dn_rt_cache_seq_start,
1742         .next   = dn_rt_cache_seq_next,
1743         .stop   = dn_rt_cache_seq_stop,
1744         .show   = dn_rt_cache_seq_show,
1745 };
1746
1747 static int dn_rt_cache_seq_open(struct inode *inode, struct file *file)
1748 {
1749         struct seq_file *seq;
1750         int rc = -ENOMEM;
1751         struct dn_rt_cache_iter_state *s = kmalloc(sizeof(*s), GFP_KERNEL);
1752
1753         if (!s)
1754                 goto out;
1755         rc = seq_open(file, &dn_rt_cache_seq_ops);
1756         if (rc)
1757                 goto out_kfree;
1758         seq             = file->private_data;
1759         seq->private    = s;
1760         memset(s, 0, sizeof(*s));
1761 out:
1762         return rc;
1763 out_kfree:
1764         kfree(s);
1765         goto out;
1766 }
1767
1768 static struct file_operations dn_rt_cache_seq_fops = {
1769         .owner   = THIS_MODULE,
1770         .open    = dn_rt_cache_seq_open,
1771         .read    = seq_read,
1772         .llseek  = seq_lseek,
1773         .release = seq_release_private,
1774 };
1775
1776 #endif /* CONFIG_PROC_FS */
1777
1778 void __init dn_route_init(void)
1779 {
1780         int i, goal, order;
1781
1782         dn_dst_ops.kmem_cachep = kmem_cache_create("dn_dst_cache",
1783                                                    sizeof(struct dn_route),
1784                                                    0, SLAB_HWCACHE_ALIGN,
1785                                                    NULL, NULL);
1786
1787         if (!dn_dst_ops.kmem_cachep)
1788                 panic("DECnet: Failed to allocate dn_dst_cache\n");
1789
1790         init_timer(&dn_route_timer);
1791         dn_route_timer.function = dn_dst_check_expire;
1792         dn_route_timer.expires = jiffies + decnet_dst_gc_interval * HZ;
1793         add_timer(&dn_route_timer);
1794
1795         goal = num_physpages >> (26 - PAGE_SHIFT);
1796
1797         for(order = 0; (1UL << order) < goal; order++)
1798                 /* NOTHING */;
1799
1800         /*
1801          * Only want 1024 entries max, since the table is very, very unlikely
1802          * to be larger than that.
1803          */
1804         while(order && ((((1UL << order) * PAGE_SIZE) /
1805                                 sizeof(struct dn_rt_hash_bucket)) >= 2048))
1806                 order--;
1807
1808         do {
1809                 dn_rt_hash_mask = (1UL << order) * PAGE_SIZE /
1810                         sizeof(struct dn_rt_hash_bucket);
1811                 while(dn_rt_hash_mask & (dn_rt_hash_mask - 1))
1812                         dn_rt_hash_mask--;
1813                 dn_rt_hash_table = (struct dn_rt_hash_bucket *)
1814                         __get_free_pages(GFP_ATOMIC, order);
1815         } while (dn_rt_hash_table == NULL && --order > 0);
1816
1817         if (!dn_rt_hash_table)
1818                 panic("Failed to allocate DECnet route cache hash table\n");
1819
1820         printk(KERN_INFO
1821                 "DECnet: Routing cache hash table of %u buckets, %ldKbytes\n",
1822                 dn_rt_hash_mask,
1823                 (long)(dn_rt_hash_mask*sizeof(struct dn_rt_hash_bucket))/1024);
1824
1825         dn_rt_hash_mask--;
1826         for(i = 0; i <= dn_rt_hash_mask; i++) {
1827                 spin_lock_init(&dn_rt_hash_table[i].lock);
1828                 dn_rt_hash_table[i].chain = NULL;
1829         }
1830
1831         dn_dst_ops.gc_thresh = (dn_rt_hash_mask + 1);
1832
1833         proc_net_fops_create("decnet_cache", S_IRUGO, &dn_rt_cache_seq_fops);
1834 }
1835
1836 void __exit dn_route_cleanup(void)
1837 {
1838         del_timer(&dn_route_timer);
1839         dn_run_flush(0);
1840
1841         proc_net_remove("decnet_cache");
1842 }
1843