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