net/decnet/dn_route.c

   1
   2 /*
   3  * DECnet       An implementation of the DECnet protocol suite for the LINUX
   4  *              operating system.  DECnet is implemented using the  BSD Socket
   5  *              interface as the means of communication with the user level.
   6  *
   7  *              DECnet Routing Functions (Endnode and Router)
   8  *
   9  * Authors:     Steve Whitehouse <SteveW@ACM.org>
  10  *              Eduardo Marcelo Serrat <emserrat@geocities.com>
  11  *
  12  * Changes:
  13  *              Steve Whitehouse : Fixes to allow "intra-ethernet" and
  14  *                                 "return-to-sender" bits on outgoing
  15  *                                 packets.
  16  *              Steve Whitehouse : Timeouts for cached routes.
  17  *              Steve Whitehouse : Use dst cache for input routes too.
  18  *              Steve Whitehouse : Fixed error values in dn_send_skb.
  19  *              Steve Whitehouse : Rework routing functions to better fit
  20  *                                 DECnet routing design
  21  *              Alexey Kuznetsov : New SMP locking
  22  *              Steve Whitehouse : More SMP locking changes & dn_cache_dump()
  23  *              Steve Whitehouse : Prerouting NF hook, now really is prerouting.
  24  *                                 Fixed possible skb leak in rtnetlink funcs.
  25  *              Steve Whitehouse : Dave Miller's dynamic hash table sizing and
  26  *                                 Alexey Kuznetsov's finer grained locking
  27  *                                 from ipv4/route.c.
  28  *              Steve Whitehouse : Routing is now starting to look like a
  29  *                                 sensible set of code now, mainly due to
  30  *                                 my copying the IPv4 routing code. The
  31  *                                 hooks here are modified and will continue
  32  *                                 to evolve for a while.
  33  *              Steve Whitehouse : Real SMP at last :-) Also new netfilter
  34  *                                 stuff. Look out raw sockets your days
  35  *                                 are numbered!
  36  *              Steve Whitehouse : Added return-to-sender functions. Added
  37  *                                 backlog congestion level return codes.
  38  *              Steve Whitehouse : Fixed bug where routes were set up with
  39  *                                 no ref count on net devices.
  40  *              Steve Whitehouse : RCU for the route cache
  41  *              Steve Whitehouse : Preparations for the flow cache
  42  *              Steve Whitehouse : Prepare for nonlinear skbs
  43  */
  44
  45 /******************************************************************************
  46     (c) 1995-1998 E.M. Serrat           emserrat@geocities.com
  47
  48     This program is free software; you can redistribute it and/or modify
  49     it under the terms of the GNU General Public License as published by
  50     the Free Software Foundation; either version 2 of the License, or
  51     any later version.
  52
  53     This program is distributed in the hope that it will be useful,
  54     but WITHOUT ANY WARRANTY; without even the implied warranty of
  55     MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  56     GNU General Public License for more details.
  57 *******************************************************************************/
  58
  59 #include <linux/config.h>
  60 #include <linux/errno.h>
  61 #include <linux/types.h>
  62 #include <linux/socket.h>
  63 #include <linux/in.h>
  64 #include <linux/kernel.h>
  65 #include <linux/sockios.h>
  66 #include <linux/net.h>
  67 #include <linux/netdevice.h>
  68 #include <linux/inet.h>
  69 #include <linux/route.h>
  70 #include <linux/in_route.h>
  71 #include <net/sock.h>
  72 #include <linux/mm.h>
  73 #include <linux/proc_fs.h>
  74 #include <linux/seq_file.h>
  75 #include <linux/init.h>
  76 #include <linux/rtnetlink.h>
  77 #include <linux/string.h>
  78 #include <linux/netfilter_decnet.h>
  79 #include <linux/rcupdate.h>
  80 #include <linux/times.h>
  81 #include <asm/errno.h>
  82 #include <net/neighbour.h>
  83 #include <net/dst.h>
  84 #include <net/flow.h>
  85 #include <net/dn.h>
  86 #include <net/dn_dev.h>
  87 #include <net/dn_nsp.h>
  88 #include <net/dn_route.h>
  89 #include <net/dn_neigh.h>
  90 #include <net/dn_fib.h>
  91
  92 struct dn_rt_hash_bucket
  93 {
  94         struct dn_route *chain;
  95         spinlock_t lock;
  96 } __attribute__((__aligned__(8)));
  97
  98 extern struct neigh_table dn_neigh_table;
  99
 100
 101 static unsigned char dn_hiord_addr[6] = {0xAA,0x00,0x04,0x00,0x00,0x00};
 102
 103 int dn_rt_min_delay = 2 * HZ;
 104 int dn_rt_max_delay = 10 * HZ;
 105 int dn_rt_mtu_expires = 10 * 60 * HZ;
 106
 107 static unsigned long dn_rt_deadline;
 108
 109 static int dn_dst_gc(void);
 110 static struct dst_entry *dn_dst_check(struct dst_entry *, __u32);
 111 static struct dst_entry *dn_dst_negative_advice(struct dst_entry *);
 112 static void dn_dst_link_failure(struct sk_buff *);
 113 static void dn_dst_update_pmtu(struct dst_entry *dst, u32 mtu);
 114 static int dn_route_input(struct sk_buff *);
 115 static void dn_run_flush(unsigned long dummy);
 116
 117 static struct dn_rt_hash_bucket *dn_rt_hash_table;
 118 static unsigned dn_rt_hash_mask;
 119
 120 static struct timer_list dn_route_timer;
 121 static struct timer_list dn_rt_flush_timer =
 122                 TIMER_INITIALIZER(dn_run_flush, 0, 0);
 123 int decnet_dst_gc_interval = 2;
 124
 125 static struct dst_ops dn_dst_ops = {
 126         .family =               PF_DECnet,
 127         .protocol =             __constant_htons(ETH_P_DNA_RT),
 128         .gc_thresh =            128,
 129         .gc =                   dn_dst_gc,
 130         .check =                dn_dst_check,
 131         .negative_advice =      dn_dst_negative_advice,
 132         .link_failure =         dn_dst_link_failure,
 133         .update_pmtu =          dn_dst_update_pmtu,
 134         .entry_size =           sizeof(struct dn_route),
 135         .entries =              ATOMIC_INIT(0),
 136 };
 137
 138 static __inline__ unsigned dn_hash(unsigned short src, unsigned short dst)
 139 {
 140         unsigned short tmp = src ^ dst;
 141         tmp ^= (tmp >> 3);
 142         tmp ^= (tmp >> 5);
 143         tmp ^= (tmp >> 10);
 144         return dn_rt_hash_mask & (unsigned)tmp;
 145 }
 146
 147 static inline void dnrt_free(struct dn_route *rt)
 148 {
 149         call_rcu(&rt->u.dst.rcu_head, (void (*)(void *))dst_free, &rt->u.dst);
 150 }
 151
 152 static inline void dnrt_drop(struct dn_route *rt)
 153 {
 154         if (rt)
 155                 dst_release(&rt->u.dst);
 156         call_rcu(&rt->u.dst.rcu_head, (void (*)(void *))dst_free, &rt->u.dst);
 157 }
 158
 159 static void dn_dst_check_expire(unsigned long dummy)
 160 {
 161         int i;
 162         struct dn_route *rt, **rtp;
 163         unsigned long now = jiffies;
 164         unsigned long expire = 120 * HZ;
 165
 166         for(i = 0; i <= dn_rt_hash_mask; i++) {
 167                 rtp = &dn_rt_hash_table[i].chain;
 168
 169                 spin_lock(&dn_rt_hash_table[i].lock);
 170                 while((rt=*rtp) != NULL) {
 171                         if (atomic_read(&rt->u.dst.__refcnt) ||
 172                                         (now - rt->u.dst.lastuse) < expire) {
 173                                 rtp = &rt->u.rt_next;
 174                                 continue;
 175                         }
 176                         *rtp = rt->u.rt_next;
 177                         rt->u.rt_next = NULL;
 178                         dnrt_free(rt);
 179                 }
 180                 spin_unlock(&dn_rt_hash_table[i].lock);
 181
 182                 if ((jiffies - now) > 0)
 183                         break;
 184         }
 185
 186         mod_timer(&dn_route_timer, now + decnet_dst_gc_interval * HZ);
 187 }
 188
 189 static int dn_dst_gc(void)
 190 {
 191         struct dn_route *rt, **rtp;
 192         int i;
 193         unsigned long now = jiffies;
 194         unsigned long expire = 10 * HZ;
 195
 196         for(i = 0; i <= dn_rt_hash_mask; i++) {
 197
 198                 spin_lock_bh(&dn_rt_hash_table[i].lock);
 199                 rtp = &dn_rt_hash_table[i].chain;
 200
 201                 while((rt=*rtp) != NULL) {
 202                         if (atomic_read(&rt->u.dst.__refcnt) ||
 203                                         (now - rt->u.dst.lastuse) < expire) {
 204                                 rtp = &rt->u.rt_next;
 205                                 continue;
 206                         }
 207                         *rtp = rt->u.rt_next;
 208                         rt->u.rt_next = NULL;
 209                         dnrt_drop(rt);
 210                         break;
 211                 }
 212                 spin_unlock_bh(&dn_rt_hash_table[i].lock);
 213         }
 214
 215         return 0;
 216 }
 217
 218 /*
 219  * The decnet standards don't impose a particular minimum mtu, what they
 220  * do insist on is that the routing layer accepts a datagram of at least
 221  * 230 bytes long. Here we have to subtract the routing header length from
 222  * 230 to get the minimum acceptable mtu. If there is no neighbour, then we
 223  * assume the worst and use a long header size.
 224  *
 225  * We update both the mtu and the advertised mss (i.e. the segment size we
 226  * advertise to the other end).
 227  */
 228 static void dn_dst_update_pmtu(struct dst_entry *dst, u32 mtu)
 229 {
 230         u32 min_mtu = 230;
 231         struct dn_dev *dn = dst->neighbour ?
 232                             (struct dn_dev *)dst->neighbour->dev->dn_ptr : NULL;
 233
 234         if (dn && dn->use_long == 0)
 235                 min_mtu -= 6;
 236         else
 237                 min_mtu -= 21;
 238
 239         if (dst->metrics[RTAX_MTU-1] > mtu && mtu >= min_mtu) {
 240                 if (!(dst_metric_locked(dst, RTAX_MTU))) {
 241                         dst->metrics[RTAX_MTU-1] = mtu;
 242                         dst_set_expires(dst, dn_rt_mtu_expires);
 243                 }
 244                 if (!(dst_metric_locked(dst, RTAX_ADVMSS))) {
 245                         u32 mss = mtu - DN_MAX_NSP_DATA_HEADER;
 246                         if (dst->metrics[RTAX_ADVMSS-1] > mss)
 247                                 dst->metrics[RTAX_ADVMSS-1] = mss;
 248                 }
 249         }
 250 }
 251
 252 /*
 253  * When a route has been marked obsolete. (e.g. routing cache flush)
 254  */
 255 static struct dst_entry *dn_dst_check(struct dst_entry *dst, __u32 cookie)
 256 {
 257         dst_release(dst);
 258         return NULL;
 259 }
 260
 261 static struct dst_entry *dn_dst_negative_advice(struct dst_entry *dst)
 262 {
 263         dst_release(dst);
 264         return NULL;
 265 }
 266
 267 static void dn_dst_link_failure(struct sk_buff *skb)
 268 {
 269         return;
 270 }
 271
 272 static inline int compare_keys(struct flowi *fl1, struct flowi *fl2)
 273 {
 274         return memcmp(&fl1->nl_u.dn_u, &fl2->nl_u.dn_u, sizeof(fl1->nl_u.dn_u)) == 0 &&
 275                 fl1->oif == fl2->oif &&
 276                 fl1->iif == fl2->iif;
 277 }
 278
 279 static int dn_insert_route(struct dn_route *rt, unsigned hash, struct dn_route **rp)
 280 {
 281         struct dn_route *rth, **rthp;
 282         unsigned long now = jiffies;
 283
 284         rthp = &dn_rt_hash_table[hash].chain;
 285
 286         spin_lock_bh(&dn_rt_hash_table[hash].lock);
 287         while((rth = *rthp) != NULL) {
 288                 if (compare_keys(&rth->fl, &rt->fl)) {
 289                         /* Put it first */
 290                         *rthp = rth->u.rt_next;
 291                         smp_wmb();
 292                         rth->u.rt_next = dn_rt_hash_table[hash].chain;
 293                         smp_wmb();
 294                         dn_rt_hash_table[hash].chain = rth;
 295
 296                         rth->u.dst.__use++;
 297                         dst_hold(&rth->u.dst);
 298                         rth->u.dst.lastuse = now;
 299                         spin_unlock_bh(&dn_rt_hash_table[hash].lock);
 300
 301                         dnrt_drop(rt);
 302                         *rp = rth;
 303                         return 0;
 304                 }
 305                 rthp = &rth->u.rt_next;
 306         }
 307
 308         smp_wmb();
 309         rt->u.rt_next = dn_rt_hash_table[hash].chain;
 310         smp_wmb();
 311         dn_rt_hash_table[hash].chain = rt;
 312
 313         dst_hold(&rt->u.dst);
 314         rt->u.dst.__use++;
 315         rt->u.dst.lastuse = now;
 316         spin_unlock_bh(&dn_rt_hash_table[hash].lock);
 317         *rp = rt;
 318         return 0;
 319 }
 320
 321 void dn_run_flush(unsigned long dummy)
 322 {
 323         int i;
 324         struct dn_route *rt, *next;
 325
 326         for(i = 0; i < dn_rt_hash_mask; i++) {
 327                 spin_lock_bh(&dn_rt_hash_table[i].lock);
 328
 329                 if ((rt = xchg(&dn_rt_hash_table[i].chain, NULL)) == NULL)
 330                         goto nothing_to_declare;
 331
 332                 for(; rt; rt=next) {
 333                         next = rt->u.rt_next;
 334                         rt->u.rt_next = NULL;
 335                         dst_free((struct dst_entry *)rt);
 336                 }
 337
 338 nothing_to_declare:
 339                 spin_unlock_bh(&dn_rt_hash_table[i].lock);
 340         }
 341 }
 342
 343 static spinlock_t dn_rt_flush_lock = SPIN_LOCK_UNLOCKED;
 344
 345 void dn_rt_cache_flush(int delay)
 346 {
 347         unsigned long now = jiffies;
 348         int user_mode = !in_interrupt();
 349
 350         if (delay < 0)
 351                 delay = dn_rt_min_delay;
 352
 353         spin_lock_bh(&dn_rt_flush_lock);
 354
 355         if (del_timer(&dn_rt_flush_timer) && delay > 0 && dn_rt_deadline) {
 356                 long tmo = (long)(dn_rt_deadline - now);
 357
 358                 if (user_mode && tmo < dn_rt_max_delay - dn_rt_min_delay)
 359                         tmo = 0;
 360
 361                 if (delay > tmo)
 362                         delay = tmo;
 363         }
 364
 365         if (delay <= 0) {
 366                 spin_unlock_bh(&dn_rt_flush_lock);
 367                 dn_run_flush(0);
 368                 return;
 369         }
 370
 371         if (dn_rt_deadline == 0)
 372                 dn_rt_deadline = now + dn_rt_max_delay;
 373
 374         dn_rt_flush_timer.expires = now + delay;
 375         add_timer(&dn_rt_flush_timer);
 376         spin_unlock_bh(&dn_rt_flush_lock);
 377 }
 378
 379 /**
 380  * dn_return_short - Return a short packet to its sender
 381  * @skb: The packet to return
 382  *
 383  */
 384 static int dn_return_short(struct sk_buff *skb)
 385 {
 386         struct dn_skb_cb *cb;
 387         unsigned char *ptr;
 388         dn_address *src;
 389         dn_address *dst;
 390         dn_address tmp;
 391
 392         /* Add back headers */
 393         skb_push(skb, skb->data - skb->nh.raw);
 394
 395         if ((skb = skb_unshare(skb, GFP_ATOMIC)) == NULL)
 396                 return NET_RX_DROP;
 397
 398         cb = DN_SKB_CB(skb);
 399         /* Skip packet length and point to flags */
 400         ptr = skb->data + 2;
 401         *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
 402
 403         dst = (dn_address *)ptr;
 404         ptr += 2;
 405         src = (dn_address *)ptr;
 406         ptr += 2;
 407         *ptr = 0; /* Zero hop count */
 408
 409         /* Swap source and destination */
 410         tmp  = *src;
 411         *src = *dst;
 412         *dst = tmp;
 413
 414         skb->pkt_type = PACKET_OUTGOING;
 415         dn_rt_finish_output(skb, NULL, NULL);
 416         return NET_RX_SUCCESS;
 417 }
 418
 419 /**
 420  * dn_return_long - Return a long packet to its sender
 421  * @skb: The long format packet to return
 422  *
 423  */
 424 static int dn_return_long(struct sk_buff *skb)
 425 {
 426         struct dn_skb_cb *cb;
 427         unsigned char *ptr;
 428         unsigned char *src_addr, *dst_addr;
 429         unsigned char tmp[ETH_ALEN];
 430
 431         /* Add back all headers */
 432         skb_push(skb, skb->data - skb->nh.raw);
 433
 434         if ((skb = skb_unshare(skb, GFP_ATOMIC)) == NULL)
 435                 return NET_RX_DROP;
 436
 437         cb = DN_SKB_CB(skb);
 438         /* Ignore packet length and point to flags */
 439         ptr = skb->data + 2;
 440
 441         /* Skip padding */
 442         if (*ptr & DN_RT_F_PF) {
 443                 char padlen = (*ptr & ~DN_RT_F_PF);
 444                 ptr += padlen;
 445         }
 446
 447         *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
 448         ptr += 2;
 449         dst_addr = ptr;
 450         ptr += 8;
 451         src_addr = ptr;
 452         ptr += 6;
 453         *ptr = 0; /* Zero hop count */
 454
 455         /* Swap source and destination */
 456         memcpy(tmp, src_addr, ETH_ALEN);
 457         memcpy(src_addr, dst_addr, ETH_ALEN);
 458         memcpy(dst_addr, tmp, ETH_ALEN);
 459
 460         skb->pkt_type = PACKET_OUTGOING;
 461         dn_rt_finish_output(skb, dst_addr, src_addr);
 462         return NET_RX_SUCCESS;
 463 }
 464
 465 /**
 466  * dn_route_rx_packet - Try and find a route for an incoming packet
 467  * @skb: The packet to find a route for
 468  *
 469  * Returns: result of input function if route is found, error code otherwise
 470  */
 471 static int dn_route_rx_packet(struct sk_buff *skb)
 472 {
 473         struct dn_skb_cb *cb = DN_SKB_CB(skb);
 474         int err;
 475
 476         if ((err = dn_route_input(skb)) == 0)
 477                 return dst_input(skb);
 478
 479         if (decnet_debug_level & 4) {
 480                 char *devname = skb->dev ? skb->dev->name : "???";
 481                 struct dn_skb_cb *cb = DN_SKB_CB(skb);
 482                 printk(KERN_DEBUG
 483                         "DECnet: dn_route_rx_packet: rt_flags=0x%02x dev=%s len=%d src=0x%04hx dst=0x%04hx err=%d type=%d\n",
 484                         (int)cb->rt_flags, devname, skb->len, cb->src, cb->dst,
 485                         err, skb->pkt_type);
 486         }
 487
 488         if ((skb->pkt_type == PACKET_HOST) && (cb->rt_flags & DN_RT_F_RQR)) {
 489                 switch(cb->rt_flags & DN_RT_PKT_MSK) {
 490                         case DN_RT_PKT_SHORT:
 491                                 return dn_return_short(skb);
 492                         case DN_RT_PKT_LONG:
 493                                 return dn_return_long(skb);
 494                 }
 495         }
 496
 497         kfree_skb(skb);
 498         return NET_RX_DROP;
 499 }
 500
 501 static int dn_route_rx_long(struct sk_buff *skb)
 502 {
 503         struct dn_skb_cb *cb = DN_SKB_CB(skb);
 504         unsigned char *ptr = skb->data;
 505
 506         if (!pskb_may_pull(skb, 21)) /* 20 for long header, 1 for shortest nsp */
 507                 goto drop_it;
 508
 509         skb_pull(skb, 20);
 510         skb->h.raw = skb->data;
 511
 512         /* Destination info */
 513         ptr += 2;
 514         cb->dst = dn_htons(dn_eth2dn(ptr));
 515         if (memcmp(ptr, dn_hiord_addr, 4) != 0)
 516                 goto drop_it;
 517         ptr += 6;
 518
 519
 520         /* Source info */
 521         ptr += 2;
 522         cb->src = dn_htons(dn_eth2dn(ptr));
 523         if (memcmp(ptr, dn_hiord_addr, 4) != 0)
 524                 goto drop_it;
 525         ptr += 6;
 526         /* Other junk */
 527         ptr++;
 528         cb->hops = *ptr++; /* Visit Count */
 529
 530         return NF_HOOK(PF_DECnet, NF_DN_PRE_ROUTING, skb, skb->dev, NULL, dn_route_rx_packet);
 531
 532 drop_it:
 533         kfree_skb(skb);
 534         return NET_RX_DROP;
 535 }
 536
 537
 538
 539 static int dn_route_rx_short(struct sk_buff *skb)
 540 {
 541         struct dn_skb_cb *cb = DN_SKB_CB(skb);
 542         unsigned char *ptr = skb->data;
 543
 544         if (!pskb_may_pull(skb, 6)) /* 5 for short header + 1 for shortest nsp */
 545                 goto drop_it;
 546
 547         skb_pull(skb, 5);
 548         skb->h.raw = skb->data;
 549
 550         cb->dst = *(dn_address *)ptr;
 551         ptr += 2;
 552         cb->src = *(dn_address *)ptr;
 553         ptr += 2;
 554         cb->hops = *ptr & 0x3f;
 555
 556         return NF_HOOK(PF_DECnet, NF_DN_PRE_ROUTING, skb, skb->dev, NULL, dn_route_rx_packet);
 557
 558 drop_it:
 559         kfree_skb(skb);
 560         return NET_RX_DROP;
 561 }
 562
 563 static int dn_route_discard(struct sk_buff *skb)
 564 {
 565         /*
 566          * I know we drop the packet here, but thats considered success in
 567          * this case
 568          */
 569         kfree_skb(skb);
 570         return NET_RX_SUCCESS;
 571 }
 572
 573 static int dn_route_ptp_hello(struct sk_buff *skb)
 574 {
 575         dn_dev_hello(skb);
 576         dn_neigh_pointopoint_hello(skb);
 577         return NET_RX_SUCCESS;
 578 }
 579
 580 int dn_route_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt)
 581 {
 582         struct dn_skb_cb *cb;
 583         unsigned char flags = 0;
 584         __u16 len = dn_ntohs(*(__u16 *)skb->data);
 585         struct dn_dev *dn = (struct dn_dev *)dev->dn_ptr;
 586         unsigned char padlen = 0;
 587
 588         if (dn == NULL)
 589                 goto dump_it;
 590
 591         if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
 592                 goto out;
 593
 594         if (!pskb_may_pull(skb, 3))
 595                 goto dump_it;
 596
 597         skb_pull(skb, 2);
 598
 599         if (len > skb->len)
 600                 goto dump_it;
 601
 602         skb_trim(skb, len);
 603
 604         flags = *skb->data;
 605
 606         cb = DN_SKB_CB(skb);
 607         cb->stamp = jiffies;
 608         cb->iif = dev->ifindex;
 609
 610         /*
 611          * If we have padding, remove it.
 612          */
 613         if (flags & DN_RT_F_PF) {
 614                 padlen = flags & ~DN_RT_F_PF;
 615                 if (!pskb_may_pull(skb, padlen + 1))
 616                         goto dump_it;
 617                 skb_pull(skb, padlen);
 618                 flags = *skb->data;
 619         }
 620
 621         skb->nh.raw = skb->data;
 622
 623         /*
 624          * Weed out future version DECnet
 625          */
 626         if (flags & DN_RT_F_VER)
 627                 goto dump_it;
 628
 629         cb->rt_flags = flags;
 630
 631         if (decnet_debug_level & 1)
 632                 printk(KERN_DEBUG
 633                         "dn_route_rcv: got 0x%02x from %s [%d %d %d]\n",
 634                         (int)flags, (dev) ? dev->name : "???", len, skb->len,
 635                         padlen);
 636
 637         if (flags & DN_RT_PKT_CNTL) {
 638                 if (unlikely(skb_is_nonlinear(skb)) &&
 639                     skb_linearize(skb, GFP_ATOMIC) != 0)
 640                         goto dump_it;
 641
 642                 switch(flags & DN_RT_CNTL_MSK) {
 643                         case DN_RT_PKT_INIT:
 644                                 dn_dev_init_pkt(skb);
 645                                 break;
 646                         case DN_RT_PKT_VERI:
 647                                 dn_dev_veri_pkt(skb);
 648                                 break;
 649                 }
 650
 651                 if (dn->parms.state != DN_DEV_S_RU)
 652                         goto dump_it;
 653
 654                 switch(flags & DN_RT_CNTL_MSK) {
 655                         case DN_RT_PKT_HELO:
 656                                 return NF_HOOK(PF_DECnet, NF_DN_HELLO, skb, skb->dev, NULL, dn_route_ptp_hello);
 657
 658                         case DN_RT_PKT_L1RT:
 659                         case DN_RT_PKT_L2RT:
 660                                 return NF_HOOK(PF_DECnet, NF_DN_ROUTE, skb, skb->dev, NULL, dn_route_discard);
 661                         case DN_RT_PKT_ERTH:
 662                                 return NF_HOOK(PF_DECnet, NF_DN_HELLO, skb, skb->dev, NULL, dn_neigh_router_hello);
 663
 664                         case DN_RT_PKT_EEDH:
 665                                 return NF_HOOK(PF_DECnet, NF_DN_HELLO, skb, skb->dev, NULL, dn_neigh_endnode_hello);
 666                 }
 667         } else {
 668                 if (dn->parms.state != DN_DEV_S_RU)
 669                         goto dump_it;
 670
 671                 skb_pull(skb, 1); /* Pull flags */
 672
 673                 switch(flags & DN_RT_PKT_MSK) {
 674                         case DN_RT_PKT_LONG:
 675                                 return dn_route_rx_long(skb);
 676                         case DN_RT_PKT_SHORT:
 677                                 return dn_route_rx_short(skb);
 678                 }
 679         }
 680
 681 dump_it:
 682         kfree_skb(skb);
 683 out:
 684         return NET_RX_DROP;
 685 }
 686
 687 static int dn_output(struct sk_buff **pskb)
 688 {
 689         struct sk_buff *skb = *pskb;
 690         struct dst_entry *dst = skb->dst;
 691         struct dn_route *rt = (struct dn_route *)dst;
 692         struct net_device *dev = dst->dev;
 693         struct dn_skb_cb *cb = DN_SKB_CB(skb);
 694         struct neighbour *neigh;
 695
 696         int err = -EINVAL;
 697
 698         if ((neigh = dst->neighbour) == NULL)
 699                 goto error;
 700
 701         skb->dev = dev;
 702
 703         cb->src = rt->rt_saddr;
 704         cb->dst = rt->rt_daddr;
 705
 706         /*
 707          * Always set the Intra-Ethernet bit on all outgoing packets
 708          * originated on this node. Only valid flag from upper layers
 709          * is return-to-sender-requested. Set hop count to 0 too.
 710          */
 711         cb->rt_flags &= ~DN_RT_F_RQR;
 712         cb->rt_flags |= DN_RT_F_IE;
 713         cb->hops = 0;
 714
 715         return NF_HOOK(PF_DECnet, NF_DN_LOCAL_OUT, skb, NULL, dev, neigh->output);
 716
 717 error:
 718         if (net_ratelimit())
 719                 printk(KERN_DEBUG "dn_output: This should not happen\n");
 720
 721         kfree_skb(skb);
 722
 723         return err;
 724 }
 725
 726 static int dn_forward(struct sk_buff *skb)
 727 {
 728         struct dn_skb_cb *cb = DN_SKB_CB(skb);
 729         struct dst_entry *dst = skb->dst;
 730         struct dn_dev *dn_db = dst->dev->dn_ptr;
 731         struct dn_route *rt;
 732         struct neighbour *neigh = dst->neighbour;
 733         int header_len;
 734 #ifdef CONFIG_NETFILTER
 735         struct net_device *dev = skb->dev;
 736 #endif
 737
 738         if (skb->pkt_type != PACKET_HOST)
 739                 goto drop;
 740
 741         /* Ensure that we have enough space for headers */
 742         rt = (struct dn_route *)skb->dst;
 743         header_len = dn_db->use_long ? 21 : 6;
 744         if (skb_cow(skb, LL_RESERVED_SPACE(rt->u.dst.dev)+header_len))
 745                 goto drop;
 746
 747         /*
 748          * Hop count exceeded.
 749          */
 750         if (++cb->hops > 30)
 751                 goto drop;
 752
 753         skb->dev = rt->u.dst.dev;
 754
 755         /*
 756          * If packet goes out same interface it came in on, then set
 757          * the Intra-Ethernet bit. This has no effect for short
 758          * packets, so we don't need to test for them here.
 759          */
 760         cb->rt_flags &= ~DN_RT_F_IE;
 761         if (rt->rt_flags & RTCF_DOREDIRECT)
 762                 cb->rt_flags |= DN_RT_F_IE;
 763
 764         return NF_HOOK(PF_DECnet, NF_DN_FORWARD, skb, dev, skb->dev, neigh->output);
 765
 766 drop:
 767         kfree_skb(skb);
 768         return NET_RX_DROP;
 769 }
 770
 771 /*
 772  * Drop packet. This is used for endnodes and for
 773  * when we should not be forwarding packets from
 774  * this dest.
 775  */
 776 static int dn_blackhole(struct sk_buff *skb)
 777 {
 778         kfree_skb(skb);
 779         return NET_RX_DROP;
 780 }
 781
 782 /*
 783  * Used to catch bugs. This should never normally get
 784  * called.
 785  */
 786 static int dn_rt_bug(struct sk_buff *skb)
 787 {
 788         if (net_ratelimit()) {
 789                 struct dn_skb_cb *cb = DN_SKB_CB(skb);
 790
 791                 printk(KERN_DEBUG "dn_rt_bug: skb from:%04x to:%04x\n",
 792                                 cb->src, cb->dst);
 793         }
 794
 795         kfree_skb(skb);
 796
 797         return NET_RX_BAD;
 798 }
 799
 800 static int dn_rt_bug_out(struct sk_buff **pskb)
 801 {
 802         return dn_rt_bug(*pskb);
 803 }
 804
 805 static int dn_rt_set_next_hop(struct dn_route *rt, struct dn_fib_res *res)
 806 {
 807         struct dn_fib_info *fi = res->fi;
 808         struct net_device *dev = rt->u.dst.dev;
 809         struct neighbour *n;
 810         unsigned mss;
 811
 812         if (fi) {
 813                 if (DN_FIB_RES_GW(*res) &&
 814                     DN_FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
 815                         rt->rt_gateway = DN_FIB_RES_GW(*res);
 816                 memcpy(rt->u.dst.metrics, fi->fib_metrics,
 817                        sizeof(rt->u.dst.metrics));
 818         }
 819         rt->rt_type = res->type;
 820
 821         if (dev != NULL && rt->u.dst.neighbour == NULL) {
 822                 n = __neigh_lookup_errno(&dn_neigh_table, &rt->rt_gateway, dev);
 823                 if (IS_ERR(n))
 824                         return PTR_ERR(n);
 825                 rt->u.dst.neighbour = n;
 826         }
 827
 828         if (rt->u.dst.metrics[RTAX_MTU-1] == 0 ||
 829             rt->u.dst.metrics[RTAX_MTU-1] > rt->u.dst.dev->mtu)
 830                 rt->u.dst.metrics[RTAX_MTU-1] = rt->u.dst.dev->mtu;
 831         mss = dn_mss_from_pmtu(dev, dst_pmtu(&rt->u.dst));
 832         if (rt->u.dst.metrics[RTAX_ADVMSS-1] == 0 ||
 833             rt->u.dst.metrics[RTAX_ADVMSS-1] > mss)
 834                 rt->u.dst.metrics[RTAX_ADVMSS-1] = mss;
 835         return 0;
 836 }
 837
 838 static inline int dn_match_addr(__u16 addr1, __u16 addr2)
 839 {
 840         __u16 tmp = dn_ntohs(addr1) ^ dn_ntohs(addr2);
 841         int match = 16;
 842         while(tmp) {
 843                 tmp >>= 1;
 844                 match--;
 845         }
 846         return match;
 847 }
 848
 849 static __u16 dnet_select_source(const struct net_device *dev, __u16 daddr, int scope)
 850 {
 851         __u16 saddr = 0;
 852         struct dn_dev *dn_db = dev->dn_ptr;
 853         struct dn_ifaddr *ifa;
 854         int best_match = 0;
 855         int ret;
 856
 857         read_lock(&dev_base_lock);
 858         for(ifa = dn_db->ifa_list; ifa; ifa = ifa->ifa_next) {
 859                 if (ifa->ifa_scope > scope)
 860                         continue;
 861                 if (!daddr) {
 862                         saddr = ifa->ifa_local;
 863                         break;
 864                 }
 865                 ret = dn_match_addr(daddr, ifa->ifa_local);
 866                 if (ret > best_match)
 867                         saddr = ifa->ifa_local;
 868                 if (best_match == 0)
 869                         saddr = ifa->ifa_local;
 870         }
 871         read_unlock(&dev_base_lock);
 872
 873         return saddr;
 874 }
 875
 876 static inline __u16 __dn_fib_res_prefsrc(struct dn_fib_res *res)
 877 {
 878         return dnet_select_source(DN_FIB_RES_DEV(*res), DN_FIB_RES_GW(*res), res->scope);
 879 }
 880
 881 static inline __u16 dn_fib_rules_map_destination(__u16 daddr, struct dn_fib_res *res)
 882 {
 883         __u16 mask = dnet_make_mask(res->prefixlen);
 884         return (daddr&~mask)|res->fi->fib_nh->nh_gw;
 885 }
 886
 887 static int dn_route_output_slow(struct dst_entry **pprt, const struct flowi *oldflp, int try_hard)
 888 {
 889         struct flowi fl = { .nl_u = { .dn_u =
 890                                       { .daddr = oldflp->fld_dst,
 891                                         .saddr = oldflp->fld_src,
 892                                         .scope = RT_SCOPE_UNIVERSE,
 893 #ifdef CONFIG_DECNET_ROUTE_FWMARK
 894                                         .fwmark = oldflp->fld_fwmark
 895 #endif
 896                                      } },
 897                             .iif = loopback_dev.ifindex,
 898                             .oif = oldflp->oif };
 899         struct dn_route *rt = NULL;
 900         struct net_device *dev_out = NULL;
 901         struct neighbour *neigh = NULL;
 902         unsigned hash;
 903         unsigned flags = 0;
 904         struct dn_fib_res res = { .fi = NULL, .type = RTN_UNICAST };
 905         int err;
 906         int free_res = 0;
 907         __u16 gateway = 0;
 908
 909         if (decnet_debug_level & 16)
 910                 printk(KERN_DEBUG
 911                        "dn_route_output_slow: dst=%04x src=%04x mark=%d"
 912                        " iif=%d oif=%d\n", oldflp->fld_dst, oldflp->fld_src,
 913                        oldflp->fld_fwmark, loopback_dev.ifindex, oldflp->oif);
 914
 915         /* If we have an output interface, verify its a DECnet device */
 916         if (oldflp->oif) {
 917                 dev_out = dev_get_by_index(oldflp->oif);
 918                 err = -ENODEV;
 919                 if (dev_out && dev_out->dn_ptr == NULL) {
 920                         dev_put(dev_out);
 921                         dev_out = NULL;
 922                 }
 923                 if (dev_out == NULL)
 924                         goto out;
 925         }
 926
 927         /* If we have a source address, verify that its a local address */
 928         if (oldflp->fld_src) {
 929                 err = -EADDRNOTAVAIL;
 930
 931                 if (dev_out) {
 932                         if (dn_dev_islocal(dev_out, oldflp->fld_src))
 933                                 goto source_ok;
 934                         dev_put(dev_out);
 935                         goto out;
 936                 }
 937                 read_lock(&dev_base_lock);
 938                 for(dev_out = dev_base; dev_out; dev_out = dev_out->next) {
 939                         if (!dev_out->dn_ptr)
 940                                 continue;
 941                         if (dn_dev_islocal(dev_out, oldflp->fld_src))
 942                                 break;
 943                 }
 944                 read_unlock(&dev_base_lock);
 945                 if (dev_out == NULL)
 946                         goto out;
 947                 dev_hold(dev_out);
 948 source_ok:
 949                 ;
 950         }
 951
 952         /* No destination? Assume its local */
 953         if (!fl.fld_dst) {
 954                 fl.fld_dst = fl.fld_src;
 955
 956                 err = -EADDRNOTAVAIL;
 957                 if (dev_out)
 958                         dev_put(dev_out);
 959                 dev_out = &loopback_dev;
 960                 dev_hold(dev_out);
 961                 if (!fl.fld_dst) {
 962                         fl.fld_dst =
 963                         fl.fld_src = dnet_select_source(dev_out, 0,
 964                                                        RT_SCOPE_HOST);
 965                         if (!fl.fld_dst)
 966                                 goto out;
 967                 }
 968                 fl.oif = loopback_dev.ifindex;
 969                 res.type = RTN_LOCAL;
 970                 goto make_route;
 971         }
 972
 973         if (decnet_debug_level & 16)
 974                 printk(KERN_DEBUG
 975                        "dn_route_output_slow: initial checks complete."
 976                        " dst=%o4x src=%04x oif=%d try_hard=%d\n", fl.fld_dst,
 977                        fl.fld_src, fl.oif, try_hard);
 978
 979         /*
 980          * N.B. If the kernel is compiled without router support then
 981          * dn_fib_lookup() will evaluate to non-zero so this if () block
 982          * will always be executed.
 983          */
 984         err = -ESRCH;
 985         if (try_hard || (err = dn_fib_lookup(&fl, &res)) != 0) {
 986                 struct dn_dev *dn_db;
 987                 if (err != -ESRCH)
 988                         goto out;
 989                 /*
 990                  * Here the fallback is basically the standard algorithm for
 991                  * routing in endnodes which is described in the DECnet routing
 992                  * docs
 993                  *
 994                  * If we are not trying hard, look in neighbour cache.
 995                  * The result is tested to ensure that if a specific output
 996                  * device/source address was requested, then we honour that
 997                  * here
 998                  */
 999                 if (!try_hard) {
1000                         neigh = dn_neigh_lookup(&dn_neigh_table, &fl.fld_dst);
1001                         if (neigh) {
1002                                 if ((oldflp->oif &&
1003                                     (neigh->dev->ifindex != oldflp->oif)) ||
1004                                     (oldflp->fld_src &&
1005                                     (!dn_dev_islocal(neigh->dev,
1006                                                       oldflp->fld_src)))) {
1007                                         neigh_release(neigh);
1008                                         neigh = NULL;
1009                                 } else {
1010                                         if (dev_out)
1011                                                 dev_put(dev_out);
1012                                         if (dn_dev_islocal(neigh->dev, fl.fld_dst)) {
1013                                                 dev_out = &loopback_dev;
1014                                                 res.type = RTN_LOCAL;
1015                                         } else {
1016                                                 dev_out = neigh->dev;
1017                                         }
1018                                         dev_hold(dev_out);
1019                                         goto select_source;
1020                                 }
1021                         }
1022                 }
1023
1024                 /* Not there? Perhaps its a local address */
1025                 if (dev_out == NULL)
1026                         dev_out = dn_dev_get_default();
1027                 err = -ENODEV;
1028                 if (dev_out == NULL)
1029                         goto out;
1030                 dn_db = dev_out->dn_ptr;
1031                 /* Possible improvement - check all devices for local addr */
1032                 if (dn_dev_islocal(dev_out, fl.fld_dst)) {
1033                         dev_put(dev_out);
1034                         dev_out = &loopback_dev;
1035                         dev_hold(dev_out);
1036                         res.type = RTN_LOCAL;
1037                         goto select_source;
1038                 }
1039                 /* Not local either.... try sending it to the default router */
1040                 neigh = neigh_clone(dn_db->router);
1041                 BUG_ON(neigh && neigh->dev != dev_out);
1042
1043                 /* Ok then, we assume its directly connected and move on */
1044 select_source:
1045                 if (neigh)
1046                         gateway = ((struct dn_neigh *)neigh)->addr;
1047                 if (gateway == 0)
1048                         gateway = fl.fld_dst;
1049                 if (fl.fld_src == 0) {
1050                         fl.fld_src = dnet_select_source(dev_out, gateway,
1051                                                          res.type == RTN_LOCAL ?
1052                                                          RT_SCOPE_HOST :
1053                                                          RT_SCOPE_LINK);
1054                         if (fl.fld_src == 0 && res.type != RTN_LOCAL)
1055                                 goto e_addr;
1056                 }
1057                 fl.oif = dev_out->ifindex;
1058                 goto make_route;
1059         }
1060         free_res = 1;
1061
1062         if (res.type == RTN_NAT)
1063                 goto e_inval;
1064
1065         if (res.type == RTN_LOCAL) {
1066                 if (!fl.fld_src)
1067                         fl.fld_src = fl.fld_dst;
1068                 if (dev_out)
1069                         dev_put(dev_out);
1070                 dev_out = &loopback_dev;
1071                 dev_hold(dev_out);
1072                 fl.oif = dev_out->ifindex;
1073                 if (res.fi)
1074                         dn_fib_info_put(res.fi);
1075                 res.fi = NULL;
1076                 goto make_route;
1077         }
1078
1079         if (res.fi->fib_nhs > 1 && fl.oif == 0)
1080                 dn_fib_select_multipath(&fl, &res);
1081
1082         /*
1083          * We could add some logic to deal with default routes here and
1084          * get rid of some of the special casing above.
1085          */
1086
1087         if (!fl.fld_src)
1088                 fl.fld_src = DN_FIB_RES_PREFSRC(res);
1089
1090         if (dev_out)
1091                 dev_put(dev_out);
1092         dev_out = DN_FIB_RES_DEV(res);
1093         dev_hold(dev_out);
1094         fl.oif = dev_out->ifindex;
1095         gateway = DN_FIB_RES_GW(res);
1096
1097 make_route:
1098         if (dev_out->flags & IFF_LOOPBACK)
1099                 flags |= RTCF_LOCAL;
1100
1101         rt = dst_alloc(&dn_dst_ops);
1102         if (rt == NULL)
1103                 goto e_nobufs;
1104
1105         atomic_set(&rt->u.dst.__refcnt, 1);
1106         rt->u.dst.flags   = DST_HOST;
1107
1108         rt->fl.fld_src    = oldflp->fld_src;
1109         rt->fl.fld_dst    = oldflp->fld_dst;
1110         rt->fl.oif        = oldflp->oif;
1111         rt->fl.iif        = 0;
1112 #ifdef CONFIG_DECNET_ROUTE_FWMARK
1113         rt->fl.fld_fwmark = oldflp->fld_fwmark;
1114 #endif
1115
1116         rt->rt_saddr      = fl.fld_src;
1117         rt->rt_daddr      = fl.fld_dst;
1118         rt->rt_gateway    = gateway ? gateway : fl.fld_dst;
1119         rt->rt_local_src  = fl.fld_src;
1120
1121         rt->rt_dst_map    = fl.fld_dst;
1122         rt->rt_src_map    = fl.fld_src;
1123
1124         rt->u.dst.dev = dev_out;
1125         dev_hold(dev_out);
1126         rt->u.dst.neighbour = neigh;
1127         neigh = NULL;
1128
1129         rt->u.dst.lastuse = jiffies;
1130         rt->u.dst.output  = dn_output;
1131         rt->u.dst.input   = dn_rt_bug;
1132         rt->rt_flags      = flags;
1133         if (flags & RTCF_LOCAL)
1134                 rt->u.dst.input = dn_nsp_rx;
1135
1136         err = dn_rt_set_next_hop(rt, &res);
1137         if (err)
1138                 goto e_neighbour;
1139
1140         hash = dn_hash(rt->fl.fld_src, rt->fl.fld_dst);
1141         dn_insert_route(rt, hash, (struct dn_route **)pprt);
1142
1143 done:
1144         if (neigh)
1145                 neigh_release(neigh);
1146         if (free_res)
1147                 dn_fib_res_put(&res);
1148         if (dev_out)
1149                 dev_put(dev_out);
1150 out:
1151         return err;
1152
1153 e_addr:
1154         err = -EADDRNOTAVAIL;
1155         goto done;
1156 e_inval:
1157         err = -EINVAL;
1158         goto done;
1159 e_nobufs:
1160         err = -ENOBUFS;
1161         goto done;
1162 e_neighbour:
1163         dst_free(&rt->u.dst);
1164         goto e_nobufs;
1165 }
1166
1167
1168 /*
1169  * N.B. The flags may be moved into the flowi at some future stage.
1170  */
1171 static int __dn_route_output_key(struct dst_entry **pprt, const struct flowi *flp, int flags)
1172 {
1173         unsigned hash = dn_hash(flp->fld_src, flp->fld_dst);
1174         struct dn_route *rt = NULL;
1175
1176         if (!(flags & MSG_TRYHARD)) {
1177                 rcu_read_lock();
1178                 for(rt = dn_rt_hash_table[hash].chain; rt; rt = rt->u.rt_next) {
1179                         read_barrier_depends();
1180                         if ((flp->fld_dst == rt->fl.fld_dst) &&
1181                             (flp->fld_src == rt->fl.fld_src) &&
1182 #ifdef CONFIG_DECNET_ROUTE_FWMARK
1183                             (flp->fld_fwmark == rt->fl.fld_fwmark) &&
1184 #endif
1185                             (rt->fl.iif == 0) &&
1186                             (rt->fl.oif == flp->oif)) {
1187                                 rt->u.dst.lastuse = jiffies;
1188                                 dst_hold(&rt->u.dst);
1189                                 rt->u.dst.__use++;
1190                                 rcu_read_unlock();
1191                                 *pprt = &rt->u.dst;
1192                                 return 0;
1193                         }
1194                 }
1195                 rcu_read_unlock();
1196         }
1197
1198         return dn_route_output_slow(pprt, flp, flags);
1199 }
1200
1201 static int dn_route_output_key(struct dst_entry **pprt, struct flowi *flp, int flags)
1202 {
1203         int err;
1204
1205         err = __dn_route_output_key(pprt, flp, flags);
1206         if (err == 0 && flp->proto) {
1207                 err = xfrm_lookup(pprt, flp, NULL, 0);
1208         }
1209         return err;
1210 }
1211
1212 int dn_route_output_sock(struct dst_entry **pprt, struct flowi *fl, struct sock *sk, int flags)
1213 {
1214         int err;
1215
1216         err = __dn_route_output_key(pprt, fl, flags & MSG_TRYHARD);
1217         if (err == 0 && fl->proto) {
1218                 err = xfrm_lookup(pprt, fl, sk, !(flags & MSG_DONTWAIT));
1219         }
1220         return err;
1221 }
1222
1223 static int dn_route_input_slow(struct sk_buff *skb)
1224 {
1225         struct dn_route *rt = NULL;
1226         struct dn_skb_cb *cb = DN_SKB_CB(skb);
1227         struct net_device *in_dev = skb->dev;
1228         struct net_device *out_dev = NULL;
1229         struct dn_dev *dn_db;
1230         struct neighbour *neigh = NULL;
1231         unsigned hash;
1232         int flags = 0;
1233         __u16 gateway = 0;
1234         __u16 local_src = 0;
1235         struct flowi fl = { .nl_u = { .dn_u =
1236                                      { .daddr = cb->dst,
1237                                        .saddr = cb->src,
1238                                        .scope = RT_SCOPE_UNIVERSE,
1239 #ifdef CONFIG_DECNET_ROUTE_FWMARK
1240                                        .fwmark = skb->nfmark
1241 #endif
1242                                     } },
1243                             .iif = skb->dev->ifindex };
1244         struct dn_fib_res res = { .fi = NULL, .type = RTN_UNREACHABLE };
1245         int err = -EINVAL;
1246         int free_res = 0;
1247
1248         dev_hold(in_dev);
1249
1250         if ((dn_db = in_dev->dn_ptr) == NULL)
1251                 goto out;
1252
1253         /* Zero source addresses are not allowed */
1254         if (fl.fld_src == 0)
1255                 goto out;
1256
1257         /*
1258          * In this case we've just received a packet from a source
1259          * outside ourselves pretending to come from us. We don't
1260          * allow it any further to prevent routing loops, spoofing and
1261          * other nasties. Loopback packets already have the dst attached
1262          * so this only affects packets which have originated elsewhere.
1263          */
1264         err  = -ENOTUNIQ;
1265         if (dn_dev_islocal(in_dev, cb->src))
1266                 goto out;
1267
1268         err = dn_fib_lookup(&fl, &res);
1269         if (err) {
1270                 if (err != -ESRCH)
1271                         goto out;
1272                 /*
1273                  * Is the destination us ?
1274                  */
1275                 if (!dn_dev_islocal(in_dev, cb->dst))
1276                         goto e_inval;
1277
1278                 res.type = RTN_LOCAL;
1279                 flags |= RTCF_DIRECTSRC;
1280         } else {
1281                 __u16 src_map = fl.fld_src;
1282                 free_res = 1;
1283
1284                 out_dev = DN_FIB_RES_DEV(res);
1285                 if (out_dev == NULL) {
1286                         if (net_ratelimit())
1287                                 printk(KERN_CRIT "Bug in dn_route_input_slow() "
1288                                                  "No output device\n");
1289                         goto e_inval;
1290                 }
1291                 dev_hold(out_dev);
1292
1293                 if (res.r)
1294                         src_map = dn_fib_rules_policy(fl.fld_src, &res, &flags);
1295
1296                 gateway = DN_FIB_RES_GW(res);
1297                 if (res.type == RTN_NAT) {
1298                         fl.fld_dst = dn_fib_rules_map_destination(fl.fld_dst, &res);
1299                         dn_fib_res_put(&res);
1300                         free_res = 0;
1301                         if (dn_fib_lookup(&fl, &res))
1302                                 goto e_inval;
1303                         free_res = 1;
1304                         if (res.type != RTN_UNICAST)
1305                                 goto e_inval;
1306                         flags |= RTCF_DNAT;
1307                         gateway = fl.fld_dst;
1308                 }
1309                 fl.fld_src = src_map;
1310         }
1311
1312         switch(res.type) {
1313         case RTN_UNICAST:
1314                 /*
1315                  * Forwarding check here, we only check for forwarding
1316                  * being turned off, if you want to only forward intra
1317                  * area, its up to you to set the routing tables up
1318                  * correctly.
1319                  */
1320                 if (dn_db->parms.forwarding == 0)
1321                         goto e_inval;
1322
1323                 if (res.fi->fib_nhs > 1 && fl.oif == 0)
1324                         dn_fib_select_multipath(&fl, &res);
1325
1326                 /*
1327                  * Check for out_dev == in_dev. We use the RTCF_DOREDIRECT
1328                  * flag as a hint to set the intra-ethernet bit when
1329                  * forwarding. If we've got NAT in operation, we don't do
1330                  * this optimisation.
1331                  */
1332                 if (out_dev == in_dev && !(flags & RTCF_NAT))
1333                         flags |= RTCF_DOREDIRECT;
1334
1335                 local_src = DN_FIB_RES_PREFSRC(res);
1336
1337         case RTN_BLACKHOLE:
1338         case RTN_UNREACHABLE:
1339                 break;
1340         case RTN_LOCAL:
1341                 flags |= RTCF_LOCAL;
1342                 fl.fld_src = cb->dst;
1343                 fl.fld_dst = cb->src;
1344
1345                 /* Routing tables gave us a gateway */
1346                 if (gateway)
1347                         goto make_route;
1348
1349                 /* Packet was intra-ethernet, so we know its on-link */
1350                 if (cb->rt_flags | DN_RT_F_IE) {
1351                         gateway = cb->src;
1352                         flags |= RTCF_DIRECTSRC;
1353                         goto make_route;
1354                 }
1355
1356                 /* Use the default router if there is one */
1357                 neigh = neigh_clone(dn_db->router);
1358                 if (neigh) {
1359                         gateway = ((struct dn_neigh *)neigh)->addr;
1360                         goto make_route;
1361                 }
1362
1363                 /* Close eyes and pray */
1364                 gateway = cb->src;
1365                 flags |= RTCF_DIRECTSRC;
1366                 goto make_route;
1367         default:
1368                 goto e_inval;
1369         }
1370
1371 make_route:
1372         rt = dst_alloc(&dn_dst_ops);
1373         if (rt == NULL)
1374                 goto e_nobufs;
1375
1376         rt->rt_saddr      = fl.fld_src;
1377         rt->rt_daddr      = fl.fld_dst;
1378         rt->rt_gateway    = fl.fld_dst;
1379         if (gateway)
1380                 rt->rt_gateway = gateway;
1381         rt->rt_local_src  = local_src ? local_src : rt->rt_saddr;
1382
1383         rt->rt_dst_map    = fl.fld_dst;
1384         rt->rt_src_map    = fl.fld_src;
1385
1386         rt->fl.fld_src    = cb->src;
1387         rt->fl.fld_dst    = cb->dst;
1388         rt->fl.oif        = 0;
1389         rt->fl.iif        = in_dev->ifindex;
1390         rt->fl.fld_fwmark = fl.fld_fwmark;
1391
1392         rt->u.dst.flags = DST_HOST;
1393         rt->u.dst.neighbour = neigh;
1394         rt->u.dst.dev = out_dev;
1395         rt->u.dst.lastuse = jiffies;
1396         rt->u.dst.output = dn_rt_bug_out;
1397         switch(res.type) {
1398                 case RTN_UNICAST:
1399                         rt->u.dst.input = dn_forward;
1400                         break;
1401                 case RTN_LOCAL:
1402                         rt->u.dst.output = dn_output;
1403                         rt->u.dst.input = dn_nsp_rx;
1404                         rt->u.dst.dev = in_dev;
1405                         flags |= RTCF_LOCAL;
1406                         break;
1407                 default:
1408                 case RTN_UNREACHABLE:
1409                 case RTN_BLACKHOLE:
1410                         rt->u.dst.input = dn_blackhole;
1411         }
1412         rt->rt_flags = flags;
1413         if (rt->u.dst.dev)
1414                 dev_hold(rt->u.dst.dev);
1415
1416         err = dn_rt_set_next_hop(rt, &res);
1417         if (err)
1418                 goto e_neighbour;
1419
1420         hash = dn_hash(rt->fl.fld_src, rt->fl.fld_dst);
1421         dn_insert_route(rt, hash, (struct dn_route **)&skb->dst);
1422
1423 done:
1424         if (neigh)
1425                 neigh_release(neigh);
1426         if (free_res)
1427                 dn_fib_res_put(&res);
1428         dev_put(in_dev);
1429         if (out_dev)
1430                 dev_put(out_dev);
1431 out:
1432         return err;
1433
1434 e_inval:
1435         err = -EINVAL;
1436         goto done;
1437
1438 e_nobufs:
1439         err = -ENOBUFS;
1440         goto done;
1441
1442 e_neighbour:
1443         dst_free(&rt->u.dst);
1444         goto done;
1445 }
1446
1447 int dn_route_input(struct sk_buff *skb)
1448 {
1449         struct dn_route *rt;
1450         struct dn_skb_cb *cb = DN_SKB_CB(skb);
1451         unsigned hash = dn_hash(cb->src, cb->dst);
1452
1453         if (skb->dst)
1454                 return 0;
1455
1456         rcu_read_lock();
1457         for(rt = dn_rt_hash_table[hash].chain; rt != NULL; rt = rt->u.rt_next) {
1458                 read_barrier_depends();
1459                 if ((rt->fl.fld_src == cb->src) &&
1460                     (rt->fl.fld_dst == cb->dst) &&
1461                     (rt->fl.oif == 0) &&
1462 #ifdef CONFIG_DECNET_ROUTE_FWMARK
1463                     (rt->fl.fld_fwmark == skb->nfmark) &&
1464 #endif
1465                     (rt->fl.iif == cb->iif)) {
1466                         rt->u.dst.lastuse = jiffies;
1467                         dst_hold(&rt->u.dst);
1468                         rt->u.dst.__use++;
1469                         rcu_read_unlock();
1470                         skb->dst = (struct dst_entry *)rt;
1471                         return 0;
1472                 }
1473         }
1474         rcu_read_unlock();
1475
1476         return dn_route_input_slow(skb);
1477 }
1478
1479 static int dn_rt_fill_info(struct sk_buff *skb, u32 pid, u32 seq, int event, int nowait)
1480 {
1481         struct dn_route *rt = (struct dn_route *)skb->dst;
1482         struct rtmsg *r;
1483         struct nlmsghdr *nlh;
1484         unsigned char *b = skb->tail;
1485         struct rta_cacheinfo ci;
1486
1487         nlh = NLMSG_PUT(skb, pid, seq, event, sizeof(*r));
1488         r = NLMSG_DATA(nlh);
1489         nlh->nlmsg_flags = (nowait && pid) ? NLM_F_MULTI : 0;
1490         r->rtm_family = AF_DECnet;
1491         r->rtm_dst_len = 16;
1492         r->rtm_src_len = 0;
1493         r->rtm_tos = 0;
1494         r->rtm_table = RT_TABLE_MAIN;
1495         r->rtm_type = rt->rt_type;
1496         r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
1497         r->rtm_scope = RT_SCOPE_UNIVERSE;
1498         r->rtm_protocol = RTPROT_UNSPEC;
1499         if (rt->rt_flags & RTCF_NOTIFY)
1500                 r->rtm_flags |= RTM_F_NOTIFY;
1501         RTA_PUT(skb, RTA_DST, 2, &rt->rt_daddr);
1502         if (rt->fl.fld_src) {
1503                 r->rtm_src_len = 16;
1504                 RTA_PUT(skb, RTA_SRC, 2, &rt->fl.fld_src);
1505         }
1506         if (rt->u.dst.dev)
1507                 RTA_PUT(skb, RTA_OIF, sizeof(int), &rt->u.dst.dev->ifindex);
1508         /*
1509          * Note to self - change this if input routes reverse direction when
1510          * they deal only with inputs and not with replies like they do
1511          * currently.
1512          */
1513         RTA_PUT(skb, RTA_PREFSRC, 2, &rt->rt_local_src);
1514         if (rt->rt_daddr != rt->rt_gateway)
1515                 RTA_PUT(skb, RTA_GATEWAY, 2, &rt->rt_gateway);
1516         if (rtnetlink_put_metrics(skb, rt->u.dst.metrics) < 0)
1517                 goto rtattr_failure;
1518         ci.rta_lastuse = jiffies_to_clock_t(jiffies - rt->u.dst.lastuse);
1519         ci.rta_used     = rt->u.dst.__use;
1520         ci.rta_clntref  = atomic_read(&rt->u.dst.__refcnt);
1521         if (rt->u.dst.expires)
1522                 ci.rta_expires = jiffies_to_clock_t(rt->u.dst.expires - jiffies);
1523         else
1524                 ci.rta_expires = 0;
1525         ci.rta_error    = rt->u.dst.error;
1526         ci.rta_id       = ci.rta_ts = ci.rta_tsage = 0;
1527         RTA_PUT(skb, RTA_CACHEINFO, sizeof(ci), &ci);
1528         if (rt->fl.iif)
1529                 RTA_PUT(skb, RTA_IIF, sizeof(int), &rt->fl.iif);
1530
1531         nlh->nlmsg_len = skb->tail - b;
1532         return skb->len;
1533
1534 nlmsg_failure:
1535 rtattr_failure:
1536         skb_trim(skb, b - skb->data);
1537         return -1;
1538 }
1539
1540 /*
1541  * This is called by both endnodes and routers now.
1542  */
1543 int dn_cache_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh, void *arg)
1544 {
1545         struct rtattr **rta = arg;
1546         struct rtmsg *rtm = NLMSG_DATA(nlh);
1547         struct dn_route *rt = NULL;
1548         struct dn_skb_cb *cb;
1549         int err;
1550         struct sk_buff *skb;
1551         struct flowi fl;
1552
1553         memset(&fl, 0, sizeof(fl));
1554         fl.proto = DNPROTO_NSP;
1555
1556         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1557         if (skb == NULL)
1558                 return -ENOBUFS;
1559         skb->mac.raw = skb->data;
1560         cb = DN_SKB_CB(skb);
1561
1562         if (rta[RTA_SRC-1])
1563                 memcpy(&fl.fld_src, RTA_DATA(rta[RTA_SRC-1]), 2);
1564         if (rta[RTA_DST-1])
1565                 memcpy(&fl.fld_dst, RTA_DATA(rta[RTA_DST-1]), 2);
1566         if (rta[RTA_IIF-1])
1567                 memcpy(&fl.iif, RTA_DATA(rta[RTA_IIF-1]), sizeof(int));
1568
1569         if (fl.iif) {
1570                 struct net_device *dev;
1571                 if ((dev = dev_get_by_index(fl.iif)) == NULL) {
1572                         kfree_skb(skb);
1573                         return -ENODEV;
1574                 }
1575                 if (!dev->dn_ptr) {
1576                         dev_put(dev);
1577                         kfree_skb(skb);
1578                         return -ENODEV;
1579                 }
1580                 skb->protocol = __constant_htons(ETH_P_DNA_RT);
1581                 skb->dev = dev;
1582                 cb->src = fl.fld_src;
1583                 cb->dst = fl.fld_dst;
1584                 local_bh_disable();
1585                 err = dn_route_input(skb);
1586                 local_bh_enable();
1587                 memset(cb, 0, sizeof(struct dn_skb_cb));
1588                 rt = (struct dn_route *)skb->dst;
1589                 if (!err && -rt->u.dst.error)
1590                         err = rt->u.dst.error;
1591         } else {
1592                 int oif = 0;
1593                 if (rta[RTA_OIF - 1])
1594                         memcpy(&oif, RTA_DATA(rta[RTA_OIF - 1]), sizeof(int));
1595                 fl.oif = oif;
1596                 err = dn_route_output_key((struct dst_entry **)&rt, &fl, 0);
1597         }
1598
1599         if (skb->dev)
1600                 dev_put(skb->dev);
1601         skb->dev = NULL;
1602         if (err)
1603                 goto out_free;
1604         skb->dst = &rt->u.dst;
1605         if (rtm->rtm_flags & RTM_F_NOTIFY)
1606                 rt->rt_flags |= RTCF_NOTIFY;
1607
1608         NETLINK_CB(skb).dst_pid = NETLINK_CB(in_skb).pid;
1609
1610         err = dn_rt_fill_info(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq, RTM_NEWROUTE, 0);
1611
1612         if (err == 0)
1613                 goto out_free;
1614         if (err < 0) {
1615                 err = -EMSGSIZE;
1616                 goto out_free;
1617         }
1618
1619         err = netlink_unicast(rtnl, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1620
1621         return err;
1622
1623 out_free:
1624         kfree_skb(skb);
1625         return err;
1626 }
1627
1628 /*
1629  * For routers, this is called from dn_fib_dump, but for endnodes its
1630  * called directly from the rtnetlink dispatch table.
1631  */
1632 int dn_cache_dump(struct sk_buff *skb, struct netlink_callback *cb)
1633 {
1634         struct dn_route *rt;
1635         int h, s_h;
1636         int idx, s_idx;
1637
1638         if (NLMSG_PAYLOAD(cb->nlh, 0) < sizeof(struct rtmsg))
1639                 return -EINVAL;
1640         if (!(((struct rtmsg *)NLMSG_DATA(cb->nlh))->rtm_flags&RTM_F_CLONED))
1641                 return 0;
1642
1643         s_h = cb->args[0];
1644         s_idx = idx = cb->args[1];
1645         for(h = 0; h <= dn_rt_hash_mask; h++) {
1646                 if (h < s_h)
1647                         continue;
1648                 if (h > s_h)
1649                         s_idx = 0;
1650                 rcu_read_lock();
1651                 for(rt = dn_rt_hash_table[h].chain, idx = 0; rt; rt = rt->u.rt_next, idx++) {
1652                         read_barrier_depends();
1653                         if (idx < s_idx)
1654                                 continue;
1655                         skb->dst = dst_clone(&rt->u.dst);
1656                         if (dn_rt_fill_info(skb, NETLINK_CB(cb->skb).pid,
1657                                         cb->nlh->nlmsg_seq, RTM_NEWROUTE, 1) <= 0) {
1658                                 dst_release(xchg(&skb->dst, NULL));
1659                                 rcu_read_unlock();
1660                                 goto done;
1661                         }
1662                         dst_release(xchg(&skb->dst, NULL));
1663                 }
1664                 rcu_read_unlock();
1665         }
1666
1667 done:
1668         cb->args[0] = h;
1669         cb->args[1] = idx;
1670         return skb->len;
1671 }
1672
1673 #ifdef CONFIG_PROC_FS
1674 struct dn_rt_cache_iter_state {
1675         int bucket;
1676 };
1677
1678 static struct dn_route *dn_rt_cache_get_first(struct seq_file *seq)
1679 {
1680         struct dn_route *rt = NULL;
1681         struct dn_rt_cache_iter_state *s = seq->private;
1682
1683         for(s->bucket = dn_rt_hash_mask; s->bucket >= 0; --s->bucket) {
1684                 rcu_read_lock();
1685                 rt = dn_rt_hash_table[s->bucket].chain;
1686                 if (rt)
1687                         break;
1688                 rcu_read_unlock();
1689         }
1690         return rt;
1691 }
1692
1693 static struct dn_route *dn_rt_cache_get_next(struct seq_file *seq, struct dn_route *rt)
1694 {
1695         struct dn_rt_cache_iter_state *s = seq->private;
1696
1697         smp_read_barrier_depends();
1698         rt = rt->u.rt_next;
1699         while(!rt) {
1700                 rcu_read_unlock();
1701                 if (--s->bucket < 0)
1702                         break;
1703                 rcu_read_lock();
1704                 rt = dn_rt_hash_table[s->bucket].chain;
1705         }
1706         return rt;
1707 }
1708
1709 static void *dn_rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
1710 {
1711         struct dn_route *rt = dn_rt_cache_get_first(seq);
1712
1713         if (rt) {
1714                 while(*pos && (rt = dn_rt_cache_get_next(seq, rt)))
1715                         --*pos;
1716         }
1717         return *pos ? NULL : rt;
1718 }
1719
1720 static void *dn_rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1721 {
1722         struct dn_route *rt = dn_rt_cache_get_next(seq, v);
1723         ++*pos;
1724         return rt;
1725 }
1726
1727 static void dn_rt_cache_seq_stop(struct seq_file *seq, void *v)
1728 {
1729         if (v)
1730                 rcu_read_unlock();
1731 }
1732
1733 static int dn_rt_cache_seq_show(struct seq_file *seq, void *v)
1734 {
1735         struct dn_route *rt = v;
1736         char buf1[DN_ASCBUF_LEN], buf2[DN_ASCBUF_LEN];
1737
1738         seq_printf(seq, "%-8s %-7s %-7s %04d %04d %04d\n",
1739                         rt->u.dst.dev ? rt->u.dst.dev->name : "*",
1740                         dn_addr2asc(dn_ntohs(rt->rt_daddr), buf1),
1741                         dn_addr2asc(dn_ntohs(rt->rt_saddr), buf2),
1742                         atomic_read(&rt->u.dst.__refcnt),
1743                         rt->u.dst.__use,
1744                         (int) dst_metric(&rt->u.dst, RTAX_RTT));
1745         return 0;
1746 }
1747
1748 static struct seq_operations dn_rt_cache_seq_ops = {
1749         .start  = dn_rt_cache_seq_start,
1750         .next   = dn_rt_cache_seq_next,
1751         .stop   = dn_rt_cache_seq_stop,
1752         .show   = dn_rt_cache_seq_show,
1753 };
1754
1755 static int dn_rt_cache_seq_open(struct inode *inode, struct file *file)
1756 {
1757         struct seq_file *seq;
1758         int rc = -ENOMEM;
1759         struct dn_rt_cache_iter_state *s = kmalloc(sizeof(*s), GFP_KERNEL);
1760
1761         if (!s)
1762                 goto out;
1763         rc = seq_open(file, &dn_rt_cache_seq_ops);
1764         if (rc)
1765                 goto out_kfree;
1766         seq             = file->private_data;
1767         seq->private    = s;
1768         memset(s, 0, sizeof(*s));
1769 out:
1770         return rc;
1771 out_kfree:
1772         kfree(s);
1773         goto out;
1774 }
1775
1776 static struct file_operations dn_rt_cache_seq_fops = {
1777         .owner   = THIS_MODULE,
1778         .open    = dn_rt_cache_seq_open,
1779         .read    = seq_read,
1780         .llseek  = seq_lseek,
1781         .release = seq_release_private,
1782 };
1783
1784 #endif /* CONFIG_PROC_FS */
1785
1786 void __init dn_route_init(void)
1787 {
1788         int i, goal, order;
1789
1790         dn_dst_ops.kmem_cachep = kmem_cache_create("dn_dst_cache",
1791                                                    sizeof(struct dn_route),
1792                                                    0, SLAB_HWCACHE_ALIGN,
1793                                                    NULL, NULL);
1794
1795         if (!dn_dst_ops.kmem_cachep)
1796                 panic("DECnet: Failed to allocate dn_dst_cache\n");
1797
1798         init_timer(&dn_route_timer);
1799         dn_route_timer.function = dn_dst_check_expire;
1800         dn_route_timer.expires = jiffies + decnet_dst_gc_interval * HZ;
1801         add_timer(&dn_route_timer);
1802
1803         goal = num_physpages >> (26 - PAGE_SHIFT);
1804
1805         for(order = 0; (1UL << order) < goal; order++)
1806                 /* NOTHING */;
1807
1808         /*
1809          * Only want 1024 entries max, since the table is very, very unlikely
1810          * to be larger than that.
1811          */
1812         while(order && ((((1UL << order) * PAGE_SIZE) /
1813                                 sizeof(struct dn_rt_hash_bucket)) >= 2048))
1814                 order--;
1815
1816         do {
1817                 dn_rt_hash_mask = (1UL << order) * PAGE_SIZE /
1818                         sizeof(struct dn_rt_hash_bucket);
1819                 while(dn_rt_hash_mask & (dn_rt_hash_mask - 1))
1820                         dn_rt_hash_mask--;
1821                 dn_rt_hash_table = (struct dn_rt_hash_bucket *)
1822                         __get_free_pages(GFP_ATOMIC, order);
1823         } while (dn_rt_hash_table == NULL && --order > 0);
1824
1825         if (!dn_rt_hash_table)
1826                 panic("Failed to allocate DECnet route cache hash table\n");
1827
1828         printk(KERN_INFO
1829                 "DECnet: Routing cache hash table of %u buckets, %ldKbytes\n",
1830                 dn_rt_hash_mask,
1831                 (long)(dn_rt_hash_mask*sizeof(struct dn_rt_hash_bucket))/1024);
1832
1833         dn_rt_hash_mask--;
1834         for(i = 0; i <= dn_rt_hash_mask; i++) {
1835                 dn_rt_hash_table[i].lock = SPIN_LOCK_UNLOCKED;
1836                 dn_rt_hash_table[i].chain = NULL;
1837         }
1838
1839         dn_dst_ops.gc_thresh = (dn_rt_hash_mask + 1);
1840
1841         proc_net_fops_create("decnet_cache", S_IRUGO, &dn_rt_cache_seq_fops);
1842 }
1843
1844 void __exit dn_route_cleanup(void)
1845 {
1846         del_timer(&dn_route_timer);
1847         dn_run_flush(0);
1848
1849         proc_net_remove("decnet_cache");
1850 }
1851