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