net/decnet/dn_route.c

   1 /*
   2  * DECnet       An implementation of the DECnet protocol suite for the LINUX
   3  *              operating system.  DECnet is implemented using the  BSD Socket
   4  *              interface as the means of communication with the user level.
   5  *
   6  *              DECnet Routing Functions (Endnode and Router)
   7  *
   8  * Authors:     Steve Whitehouse <SteveW@ACM.org>
   9  *              Eduardo Marcelo Serrat <emserrat@geocities.com>
  10  *
  11  * Changes:
  12  *              Steve Whitehouse : Fixes to allow "intra-ethernet" and
  13  *                                 "return-to-sender" bits on outgoing
  14  *                                 packets.
  15  *              Steve Whitehouse : Timeouts for cached routes.
  16  *              Steve Whitehouse : Use dst cache for input routes too.
  17  *              Steve Whitehouse : Fixed error values in dn_send_skb.
  18  *              Steve Whitehouse : Rework routing functions to better fit
  19  *                                 DECnet routing design
  20  *              Alexey Kuznetsov : New SMP locking
  21  *              Steve Whitehouse : More SMP locking changes & dn_cache_dump()
  22  *              Steve Whitehouse : Prerouting NF hook, now really is prerouting.
  23  *                                 Fixed possible skb leak in rtnetlink funcs.
  24  *              Steve Whitehouse : Dave Miller's dynamic hash table sizing and
  25  *                                 Alexey Kuznetsov's finer grained locking
  26  *                                 from ipv4/route.c.
  27  *              Steve Whitehouse : Routing is now starting to look like a
  28  *                                 sensible set of code now, mainly due to
  29  *                                 my copying the IPv4 routing code. The
  30  *                                 hooks here are modified and will continue
  31  *                                 to evolve for a while.
  32  *              Steve Whitehouse : Real SMP at last :-) Also new netfilter
  33  *                                 stuff. Look out raw sockets your days
  34  *                                 are numbered!
  35  *              Steve Whitehouse : Added return-to-sender functions. Added
  36  *                                 backlog congestion level return codes.
  37  *              Steve Whitehouse : Fixed bug where routes were set up with
  38  *                                 no ref count on net devices.
  39  *              Steve Whitehouse : RCU for the route cache
  40  *              Steve Whitehouse : Preparations for the flow cache
  41  *              Steve Whitehouse : Prepare for nonlinear skbs
  42  */
  43
  44 /******************************************************************************
  45     (c) 1995-1998 E.M. Serrat           emserrat@geocities.com
  46
  47     This program is free software; you can redistribute it and/or modify
  48     it under the terms of the GNU General Public License as published by
  49     the Free Software Foundation; either version 2 of the License, or
  50     any later version.
  51
  52     This program is distributed in the hope that it will be useful,
  53     but WITHOUT ANY WARRANTY; without even the implied warranty of
  54     MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  55     GNU General Public License for more details.
  56 *******************************************************************************/
  57
  58 #include <linux/errno.h>
  59 #include <linux/types.h>
  60 #include <linux/socket.h>
  61 #include <linux/in.h>
  62 #include <linux/kernel.h>
  63 #include <linux/sockios.h>
  64 #include <linux/net.h>
  65 #include <linux/netdevice.h>
  66 #include <linux/inet.h>
  67 #include <linux/route.h>
  68 #include <linux/in_route.h>
  69 #include <linux/slab.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/net_namespace.h>
  82 #include <net/netlink.h>
  83 #include <net/neighbour.h>
  84 #include <net/dst.h>
  85 #include <net/flow.h>
  86 #include <net/fib_rules.h>
  87 #include <net/dn.h>
  88 #include <net/dn_dev.h>
  89 #include <net/dn_nsp.h>
  90 #include <net/dn_route.h>
  91 #include <net/dn_neigh.h>
  92 #include <net/dn_fib.h>
  93
  94 struct dn_rt_hash_bucket
  95 {
  96         struct dn_route __rcu *chain;
  97         spinlock_t lock;
  98 };
  99
 100 extern struct neigh_table dn_neigh_table;
 101
 102
 103 static unsigned char dn_hiord_addr[6] = {0xAA,0x00,0x04,0x00,0x00,0x00};
 104
 105 static const int dn_rt_min_delay = 2 * HZ;
 106 static const int dn_rt_max_delay = 10 * HZ;
 107 static const int dn_rt_mtu_expires = 10 * 60 * HZ;
 108
 109 static unsigned long dn_rt_deadline;
 110
 111 static int dn_dst_gc(struct dst_ops *ops);
 112 static struct dst_entry *dn_dst_check(struct dst_entry *, __u32);
 113 static unsigned int dn_dst_default_advmss(const struct dst_entry *dst);
 114 static unsigned int dn_dst_default_mtu(const struct dst_entry *dst);
 115 static struct dst_entry *dn_dst_negative_advice(struct dst_entry *);
 116 static void dn_dst_link_failure(struct sk_buff *);
 117 static void dn_dst_update_pmtu(struct dst_entry *dst, u32 mtu);
 118 static int dn_route_input(struct sk_buff *);
 119 static void dn_run_flush(unsigned long dummy);
 120
 121 static struct dn_rt_hash_bucket *dn_rt_hash_table;
 122 static unsigned dn_rt_hash_mask;
 123
 124 static struct timer_list dn_route_timer;
 125 static DEFINE_TIMER(dn_rt_flush_timer, dn_run_flush, 0, 0);
 126 int decnet_dst_gc_interval = 2;
 127
 128 static struct dst_ops dn_dst_ops = {
 129         .family =               PF_DECnet,
 130         .protocol =             cpu_to_be16(ETH_P_DNA_RT),
 131         .gc_thresh =            128,
 132         .gc =                   dn_dst_gc,
 133         .check =                dn_dst_check,
 134         .default_advmss =       dn_dst_default_advmss,
 135         .default_mtu =          dn_dst_default_mtu,
 136         .negative_advice =      dn_dst_negative_advice,
 137         .link_failure =         dn_dst_link_failure,
 138         .update_pmtu =          dn_dst_update_pmtu,
 139 };
 140
 141 static __inline__ unsigned dn_hash(__le16 src, __le16 dst)
 142 {
 143         __u16 tmp = (__u16 __force)(src ^ dst);
 144         tmp ^= (tmp >> 3);
 145         tmp ^= (tmp >> 5);
 146         tmp ^= (tmp >> 10);
 147         return dn_rt_hash_mask & (unsigned)tmp;
 148 }
 149
 150 static inline void dnrt_free(struct dn_route *rt)
 151 {
 152         call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
 153 }
 154
 155 static inline void dnrt_drop(struct dn_route *rt)
 156 {
 157         dst_release(&rt->dst);
 158         call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
 159 }
 160
 161 static void dn_dst_check_expire(unsigned long dummy)
 162 {
 163         int i;
 164         struct dn_route *rt;
 165         struct dn_route __rcu **rtp;
 166         unsigned long now = jiffies;
 167         unsigned long expire = 120 * HZ;
 168
 169         for (i = 0; i <= dn_rt_hash_mask; i++) {
 170                 rtp = &dn_rt_hash_table[i].chain;
 171
 172                 spin_lock(&dn_rt_hash_table[i].lock);
 173                 while ((rt = rcu_dereference_protected(*rtp,
 174                                                 lockdep_is_held(&dn_rt_hash_table[i].lock))) != NULL) {
 175                         if (atomic_read(&rt->dst.__refcnt) ||
 176                                         (now - rt->dst.lastuse) < expire) {
 177                                 rtp = &rt->dst.dn_next;
 178                                 continue;
 179                         }
 180                         *rtp = rt->dst.dn_next;
 181                         rt->dst.dn_next = NULL;
 182                         dnrt_free(rt);
 183                 }
 184                 spin_unlock(&dn_rt_hash_table[i].lock);
 185
 186                 if ((jiffies - now) > 0)
 187                         break;
 188         }
 189
 190         mod_timer(&dn_route_timer, now + decnet_dst_gc_interval * HZ);
 191 }
 192
 193 static int dn_dst_gc(struct dst_ops *ops)
 194 {
 195         struct dn_route *rt;
 196         struct dn_route __rcu **rtp;
 197         int i;
 198         unsigned long now = jiffies;
 199         unsigned long expire = 10 * HZ;
 200
 201         for (i = 0; i <= dn_rt_hash_mask; i++) {
 202
 203                 spin_lock_bh(&dn_rt_hash_table[i].lock);
 204                 rtp = &dn_rt_hash_table[i].chain;
 205
 206                 while ((rt = rcu_dereference_protected(*rtp,
 207                                                 lockdep_is_held(&dn_rt_hash_table[i].lock))) != NULL) {
 208                         if (atomic_read(&rt->dst.__refcnt) ||
 209                                         (now - rt->dst.lastuse) < expire) {
 210                                 rtp = &rt->dst.dn_next;
 211                                 continue;
 212                         }
 213                         *rtp = rt->dst.dn_next;
 214                         rt->dst.dn_next = NULL;
 215                         dnrt_drop(rt);
 216                         break;
 217                 }
 218                 spin_unlock_bh(&dn_rt_hash_table[i].lock);
 219         }
 220
 221         return 0;
 222 }
 223
 224 /*
 225  * The decnet standards don't impose a particular minimum mtu, what they
 226  * do insist on is that the routing layer accepts a datagram of at least
 227  * 230 bytes long. Here we have to subtract the routing header length from
 228  * 230 to get the minimum acceptable mtu. If there is no neighbour, then we
 229  * assume the worst and use a long header size.
 230  *
 231  * We update both the mtu and the advertised mss (i.e. the segment size we
 232  * advertise to the other end).
 233  */
 234 static void dn_dst_update_pmtu(struct dst_entry *dst, u32 mtu)
 235 {
 236         u32 min_mtu = 230;
 237         struct dn_dev *dn = dst->neighbour ?
 238                             rcu_dereference_raw(dst->neighbour->dev->dn_ptr) : NULL;
 239
 240         if (dn && dn->use_long == 0)
 241                 min_mtu -= 6;
 242         else
 243                 min_mtu -= 21;
 244
 245         if (dst_metric(dst, RTAX_MTU) > mtu && mtu >= min_mtu) {
 246                 if (!(dst_metric_locked(dst, RTAX_MTU))) {
 247                         dst_metric_set(dst, RTAX_MTU, mtu);
 248                         dst_set_expires(dst, dn_rt_mtu_expires);
 249                 }
 250                 if (!(dst_metric_locked(dst, RTAX_ADVMSS))) {
 251                         u32 mss = mtu - DN_MAX_NSP_DATA_HEADER;
 252                         u32 existing_mss = dst_metric_raw(dst, RTAX_ADVMSS);
 253                         if (!existing_mss || existing_mss > mss)
 254                                 dst_metric_set(dst, RTAX_ADVMSS, mss);
 255                 }
 256         }
 257 }
 258
 259 /*
 260  * When a route has been marked obsolete. (e.g. routing cache flush)
 261  */
 262 static struct dst_entry *dn_dst_check(struct dst_entry *dst, __u32 cookie)
 263 {
 264         return NULL;
 265 }
 266
 267 static struct dst_entry *dn_dst_negative_advice(struct dst_entry *dst)
 268 {
 269         dst_release(dst);
 270         return NULL;
 271 }
 272
 273 static void dn_dst_link_failure(struct sk_buff *skb)
 274 {
 275 }
 276
 277 static inline int compare_keys(struct flowi *fl1, struct flowi *fl2)
 278 {
 279         return ((fl1->fld_dst ^ fl2->fld_dst) |
 280                 (fl1->fld_src ^ fl2->fld_src) |
 281                 (fl1->mark ^ fl2->mark) |
 282                 (fl1->fld_scope ^ fl2->fld_scope) |
 283                 (fl1->oif ^ fl2->oif) |
 284                 (fl1->iif ^ fl2->iif)) == 0;
 285 }
 286
 287 static int dn_insert_route(struct dn_route *rt, unsigned hash, struct dn_route **rp)
 288 {
 289         struct dn_route *rth;
 290         struct dn_route __rcu **rthp;
 291         unsigned long now = jiffies;
 292
 293         rthp = &dn_rt_hash_table[hash].chain;
 294
 295         spin_lock_bh(&dn_rt_hash_table[hash].lock);
 296         while ((rth = rcu_dereference_protected(*rthp,
 297                                                 lockdep_is_held(&dn_rt_hash_table[hash].lock))) != NULL) {
 298                 if (compare_keys(&rth->fl, &rt->fl)) {
 299                         /* Put it first */
 300                         *rthp = rth->dst.dn_next;
 301                         rcu_assign_pointer(rth->dst.dn_next,
 302                                            dn_rt_hash_table[hash].chain);
 303                         rcu_assign_pointer(dn_rt_hash_table[hash].chain, rth);
 304
 305                         dst_use(&rth->dst, now);
 306                         spin_unlock_bh(&dn_rt_hash_table[hash].lock);
 307
 308                         dnrt_drop(rt);
 309                         *rp = rth;
 310                         return 0;
 311                 }
 312                 rthp = &rth->dst.dn_next;
 313         }
 314
 315         rcu_assign_pointer(rt->dst.dn_next, dn_rt_hash_table[hash].chain);
 316         rcu_assign_pointer(dn_rt_hash_table[hash].chain, rt);
 317
 318         dst_use(&rt->dst, now);
 319         spin_unlock_bh(&dn_rt_hash_table[hash].lock);
 320         *rp = rt;
 321         return 0;
 322 }
 323
 324 static void dn_run_flush(unsigned long dummy)
 325 {
 326         int i;
 327         struct dn_route *rt, *next;
 328
 329         for (i = 0; i < dn_rt_hash_mask; i++) {
 330                 spin_lock_bh(&dn_rt_hash_table[i].lock);
 331
 332                 if ((rt = xchg((struct dn_route **)&dn_rt_hash_table[i].chain, NULL)) == NULL)
 333                         goto nothing_to_declare;
 334
 335                 for(; rt; rt = next) {
 336                         next = rcu_dereference_raw(rt->dst.dn_next);
 337                         RCU_INIT_POINTER(rt->dst.dn_next, NULL);
 338                         dst_free((struct dst_entry *)rt);
 339                 }
 340
 341 nothing_to_declare:
 342                 spin_unlock_bh(&dn_rt_hash_table[i].lock);
 343         }
 344 }
 345
 346 static DEFINE_SPINLOCK(dn_rt_flush_lock);
 347
 348 void dn_rt_cache_flush(int delay)
 349 {
 350         unsigned long now = jiffies;
 351         int user_mode = !in_interrupt();
 352
 353         if (delay < 0)
 354                 delay = dn_rt_min_delay;
 355
 356         spin_lock_bh(&dn_rt_flush_lock);
 357
 358         if (del_timer(&dn_rt_flush_timer) && delay > 0 && dn_rt_deadline) {
 359                 long tmo = (long)(dn_rt_deadline - now);
 360
 361                 if (user_mode && tmo < dn_rt_max_delay - dn_rt_min_delay)
 362                         tmo = 0;
 363
 364                 if (delay > tmo)
 365                         delay = tmo;
 366         }
 367
 368         if (delay <= 0) {
 369                 spin_unlock_bh(&dn_rt_flush_lock);
 370                 dn_run_flush(0);
 371                 return;
 372         }
 373
 374         if (dn_rt_deadline == 0)
 375                 dn_rt_deadline = now + dn_rt_max_delay;
 376
 377         dn_rt_flush_timer.expires = now + delay;
 378         add_timer(&dn_rt_flush_timer);
 379         spin_unlock_bh(&dn_rt_flush_lock);
 380 }
 381
 382 /**
 383  * dn_return_short - Return a short packet to its sender
 384  * @skb: The packet to return
 385  *
 386  */
 387 static int dn_return_short(struct sk_buff *skb)
 388 {
 389         struct dn_skb_cb *cb;
 390         unsigned char *ptr;
 391         __le16 *src;
 392         __le16 *dst;
 393
 394         /* Add back headers */
 395         skb_push(skb, skb->data - skb_network_header(skb));
 396
 397         if ((skb = skb_unshare(skb, GFP_ATOMIC)) == NULL)
 398                 return NET_RX_DROP;
 399
 400         cb = DN_SKB_CB(skb);
 401         /* Skip packet length and point to flags */
 402         ptr = skb->data + 2;
 403         *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
 404
 405         dst = (__le16 *)ptr;
 406         ptr += 2;
 407         src = (__le16 *)ptr;
 408         ptr += 2;
 409         *ptr = 0; /* Zero hop count */
 410
 411         swap(*src, *dst);
 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_network_header(skb));
 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;
 473         int err;
 474
 475         if ((err = dn_route_input(skb)) == 0)
 476                 return dst_input(skb);
 477
 478         cb = DN_SKB_CB(skb);
 479         if (decnet_debug_level & 4) {
 480                 char *devname = skb->dev ? skb->dev->name : "???";
 481
 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,
 485                         le16_to_cpu(cb->src), le16_to_cpu(cb->dst),
 486                         err, skb->pkt_type);
 487         }
 488
 489         if ((skb->pkt_type == PACKET_HOST) && (cb->rt_flags & DN_RT_F_RQR)) {
 490                 switch(cb->rt_flags & DN_RT_PKT_MSK) {
 491                         case DN_RT_PKT_SHORT:
 492                                 return dn_return_short(skb);
 493                         case DN_RT_PKT_LONG:
 494                                 return dn_return_long(skb);
 495                 }
 496         }
 497
 498         kfree_skb(skb);
 499         return NET_RX_DROP;
 500 }
 501
 502 static int dn_route_rx_long(struct sk_buff *skb)
 503 {
 504         struct dn_skb_cb *cb = DN_SKB_CB(skb);
 505         unsigned char *ptr = skb->data;
 506
 507         if (!pskb_may_pull(skb, 21)) /* 20 for long header, 1 for shortest nsp */
 508                 goto drop_it;
 509
 510         skb_pull(skb, 20);
 511         skb_reset_transport_header(skb);
 512
 513         /* Destination info */
 514         ptr += 2;
 515         cb->dst = dn_eth2dn(ptr);
 516         if (memcmp(ptr, dn_hiord_addr, 4) != 0)
 517                 goto drop_it;
 518         ptr += 6;
 519
 520
 521         /* Source info */
 522         ptr += 2;
 523         cb->src = dn_eth2dn(ptr);
 524         if (memcmp(ptr, dn_hiord_addr, 4) != 0)
 525                 goto drop_it;
 526         ptr += 6;
 527         /* Other junk */
 528         ptr++;
 529         cb->hops = *ptr++; /* Visit Count */
 530
 531         return NF_HOOK(NFPROTO_DECNET, NF_DN_PRE_ROUTING, skb, skb->dev, NULL,
 532                        dn_route_rx_packet);
 533
 534 drop_it:
 535         kfree_skb(skb);
 536         return NET_RX_DROP;
 537 }
 538
 539
 540
 541 static int dn_route_rx_short(struct sk_buff *skb)
 542 {
 543         struct dn_skb_cb *cb = DN_SKB_CB(skb);
 544         unsigned char *ptr = skb->data;
 545
 546         if (!pskb_may_pull(skb, 6)) /* 5 for short header + 1 for shortest nsp */
 547                 goto drop_it;
 548
 549         skb_pull(skb, 5);
 550         skb_reset_transport_header(skb);
 551
 552         cb->dst = *(__le16 *)ptr;
 553         ptr += 2;
 554         cb->src = *(__le16 *)ptr;
 555         ptr += 2;
 556         cb->hops = *ptr & 0x3f;
 557
 558         return NF_HOOK(NFPROTO_DECNET, NF_DN_PRE_ROUTING, skb, skb->dev, NULL,
 559                        dn_route_rx_packet);
 560
 561 drop_it:
 562         kfree_skb(skb);
 563         return NET_RX_DROP;
 564 }
 565
 566 static int dn_route_discard(struct sk_buff *skb)
 567 {
 568         /*
 569          * I know we drop the packet here, but thats considered success in
 570          * this case
 571          */
 572         kfree_skb(skb);
 573         return NET_RX_SUCCESS;
 574 }
 575
 576 static int dn_route_ptp_hello(struct sk_buff *skb)
 577 {
 578         dn_dev_hello(skb);
 579         dn_neigh_pointopoint_hello(skb);
 580         return NET_RX_SUCCESS;
 581 }
 582
 583 int dn_route_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
 584 {
 585         struct dn_skb_cb *cb;
 586         unsigned char flags = 0;
 587         __u16 len = le16_to_cpu(*(__le16 *)skb->data);
 588         struct dn_dev *dn = rcu_dereference(dev->dn_ptr);
 589         unsigned char padlen = 0;
 590
 591         if (!net_eq(dev_net(dev), &init_net))
 592                 goto dump_it;
 593
 594         if (dn == NULL)
 595                 goto dump_it;
 596
 597         if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
 598                 goto out;
 599
 600         if (!pskb_may_pull(skb, 3))
 601                 goto dump_it;
 602
 603         skb_pull(skb, 2);
 604
 605         if (len > skb->len)
 606                 goto dump_it;
 607
 608         skb_trim(skb, len);
 609
 610         flags = *skb->data;
 611
 612         cb = DN_SKB_CB(skb);
 613         cb->stamp = jiffies;
 614         cb->iif = dev->ifindex;
 615
 616         /*
 617          * If we have padding, remove it.
 618          */
 619         if (flags & DN_RT_F_PF) {
 620                 padlen = flags & ~DN_RT_F_PF;
 621                 if (!pskb_may_pull(skb, padlen + 1))
 622                         goto dump_it;
 623                 skb_pull(skb, padlen);
 624                 flags = *skb->data;
 625         }
 626
 627         skb_reset_network_header(skb);
 628
 629         /*
 630          * Weed out future version DECnet
 631          */
 632         if (flags & DN_RT_F_VER)
 633                 goto dump_it;
 634
 635         cb->rt_flags = flags;
 636
 637         if (decnet_debug_level & 1)
 638                 printk(KERN_DEBUG
 639                         "dn_route_rcv: got 0x%02x from %s [%d %d %d]\n",
 640                         (int)flags, (dev) ? dev->name : "???", len, skb->len,
 641                         padlen);
 642
 643         if (flags & DN_RT_PKT_CNTL) {
 644                 if (unlikely(skb_linearize(skb)))
 645                         goto dump_it;
 646
 647                 switch(flags & DN_RT_CNTL_MSK) {
 648                         case DN_RT_PKT_INIT:
 649                                 dn_dev_init_pkt(skb);
 650                                 break;
 651                         case DN_RT_PKT_VERI:
 652                                 dn_dev_veri_pkt(skb);
 653                                 break;
 654                 }
 655
 656                 if (dn->parms.state != DN_DEV_S_RU)
 657                         goto dump_it;
 658
 659                 switch(flags & DN_RT_CNTL_MSK) {
 660                         case DN_RT_PKT_HELO:
 661                                 return NF_HOOK(NFPROTO_DECNET, NF_DN_HELLO,
 662                                                skb, skb->dev, NULL,
 663                                                dn_route_ptp_hello);
 664
 665                         case DN_RT_PKT_L1RT:
 666                         case DN_RT_PKT_L2RT:
 667                                 return NF_HOOK(NFPROTO_DECNET, NF_DN_ROUTE,
 668                                                skb, skb->dev, NULL,
 669                                                dn_route_discard);
 670                         case DN_RT_PKT_ERTH:
 671                                 return NF_HOOK(NFPROTO_DECNET, NF_DN_HELLO,
 672                                                skb, skb->dev, NULL,
 673                                                dn_neigh_router_hello);
 674
 675                         case DN_RT_PKT_EEDH:
 676                                 return NF_HOOK(NFPROTO_DECNET, NF_DN_HELLO,
 677                                                skb, skb->dev, NULL,
 678                                                dn_neigh_endnode_hello);
 679                 }
 680         } else {
 681                 if (dn->parms.state != DN_DEV_S_RU)
 682                         goto dump_it;
 683
 684                 skb_pull(skb, 1); /* Pull flags */
 685
 686                 switch(flags & DN_RT_PKT_MSK) {
 687                         case DN_RT_PKT_LONG:
 688                                 return dn_route_rx_long(skb);
 689                         case DN_RT_PKT_SHORT:
 690                                 return dn_route_rx_short(skb);
 691                 }
 692         }
 693
 694 dump_it:
 695         kfree_skb(skb);
 696 out:
 697         return NET_RX_DROP;
 698 }
 699
 700 static int dn_output(struct sk_buff *skb)
 701 {
 702         struct dst_entry *dst = skb_dst(skb);
 703         struct dn_route *rt = (struct dn_route *)dst;
 704         struct net_device *dev = dst->dev;
 705         struct dn_skb_cb *cb = DN_SKB_CB(skb);
 706         struct neighbour *neigh;
 707
 708         int err = -EINVAL;
 709
 710         if ((neigh = dst->neighbour) == NULL)
 711                 goto error;
 712
 713         skb->dev = dev;
 714
 715         cb->src = rt->rt_saddr;
 716         cb->dst = rt->rt_daddr;
 717
 718         /*
 719          * Always set the Intra-Ethernet bit on all outgoing packets
 720          * originated on this node. Only valid flag from upper layers
 721          * is return-to-sender-requested. Set hop count to 0 too.
 722          */
 723         cb->rt_flags &= ~DN_RT_F_RQR;
 724         cb->rt_flags |= DN_RT_F_IE;
 725         cb->hops = 0;
 726
 727         return NF_HOOK(NFPROTO_DECNET, NF_DN_LOCAL_OUT, skb, NULL, dev,
 728                        neigh->output);
 729
 730 error:
 731         if (net_ratelimit())
 732                 printk(KERN_DEBUG "dn_output: This should not happen\n");
 733
 734         kfree_skb(skb);
 735
 736         return err;
 737 }
 738
 739 static int dn_forward(struct sk_buff *skb)
 740 {
 741         struct dn_skb_cb *cb = DN_SKB_CB(skb);
 742         struct dst_entry *dst = skb_dst(skb);
 743         struct dn_dev *dn_db = rcu_dereference(dst->dev->dn_ptr);
 744         struct dn_route *rt;
 745         struct neighbour *neigh = dst->neighbour;
 746         int header_len;
 747 #ifdef CONFIG_NETFILTER
 748         struct net_device *dev = skb->dev;
 749 #endif
 750
 751         if (skb->pkt_type != PACKET_HOST)
 752                 goto drop;
 753
 754         /* Ensure that we have enough space for headers */
 755         rt = (struct dn_route *)skb_dst(skb);
 756         header_len = dn_db->use_long ? 21 : 6;
 757         if (skb_cow(skb, LL_RESERVED_SPACE(rt->dst.dev)+header_len))
 758                 goto drop;
 759
 760         /*
 761          * Hop count exceeded.
 762          */
 763         if (++cb->hops > 30)
 764                 goto drop;
 765
 766         skb->dev = rt->dst.dev;
 767
 768         /*
 769          * If packet goes out same interface it came in on, then set
 770          * the Intra-Ethernet bit. This has no effect for short
 771          * packets, so we don't need to test for them here.
 772          */
 773         cb->rt_flags &= ~DN_RT_F_IE;
 774         if (rt->rt_flags & RTCF_DOREDIRECT)
 775                 cb->rt_flags |= DN_RT_F_IE;
 776
 777         return NF_HOOK(NFPROTO_DECNET, NF_DN_FORWARD, skb, dev, skb->dev,
 778                        neigh->output);
 779
 780 drop:
 781         kfree_skb(skb);
 782         return NET_RX_DROP;
 783 }
 784
 785 /*
 786  * Used to catch bugs. This should never normally get
 787  * called.
 788  */
 789 static int dn_rt_bug(struct sk_buff *skb)
 790 {
 791         if (net_ratelimit()) {
 792                 struct dn_skb_cb *cb = DN_SKB_CB(skb);
 793
 794                 printk(KERN_DEBUG "dn_rt_bug: skb from:%04x to:%04x\n",
 795                                 le16_to_cpu(cb->src), le16_to_cpu(cb->dst));
 796         }
 797
 798         kfree_skb(skb);
 799
 800         return NET_RX_DROP;
 801 }
 802
 803 static unsigned int dn_dst_default_advmss(const struct dst_entry *dst)
 804 {
 805         return dn_mss_from_pmtu(dst->dev, dst_mtu(dst));
 806 }
 807
 808 static unsigned int dn_dst_default_mtu(const struct dst_entry *dst)
 809 {
 810         return dst->dev->mtu;
 811 }
 812
 813 static int dn_rt_set_next_hop(struct dn_route *rt, struct dn_fib_res *res)
 814 {
 815         struct dn_fib_info *fi = res->fi;
 816         struct net_device *dev = rt->dst.dev;
 817         struct neighbour *n;
 818         unsigned int metric;
 819
 820         if (fi) {
 821                 if (DN_FIB_RES_GW(*res) &&
 822                     DN_FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
 823                         rt->rt_gateway = DN_FIB_RES_GW(*res);
 824                 dst_import_metrics(&rt->dst, fi->fib_metrics);
 825         }
 826         rt->rt_type = res->type;
 827
 828         if (dev != NULL && rt->dst.neighbour == NULL) {
 829                 n = __neigh_lookup_errno(&dn_neigh_table, &rt->rt_gateway, dev);
 830                 if (IS_ERR(n))
 831                         return PTR_ERR(n);
 832                 rt->dst.neighbour = n;
 833         }
 834
 835         if (dst_metric(&rt->dst, RTAX_MTU) > rt->dst.dev->mtu)
 836                 dst_metric_set(&rt->dst, RTAX_MTU, rt->dst.dev->mtu);
 837         metric = dst_metric_raw(&rt->dst, RTAX_ADVMSS);
 838         if (metric) {
 839                 unsigned int mss = dn_mss_from_pmtu(dev, dst_mtu(&rt->dst));
 840                 if (metric > mss)
 841                         dst_metric_set(&rt->dst, RTAX_ADVMSS, mss);
 842         }
 843         return 0;
 844 }
 845
 846 static inline int dn_match_addr(__le16 addr1, __le16 addr2)
 847 {
 848         __u16 tmp = le16_to_cpu(addr1) ^ le16_to_cpu(addr2);
 849         int match = 16;
 850         while(tmp) {
 851                 tmp >>= 1;
 852                 match--;
 853         }
 854         return match;
 855 }
 856
 857 static __le16 dnet_select_source(const struct net_device *dev, __le16 daddr, int scope)
 858 {
 859         __le16 saddr = 0;
 860         struct dn_dev *dn_db;
 861         struct dn_ifaddr *ifa;
 862         int best_match = 0;
 863         int ret;
 864
 865         rcu_read_lock();
 866         dn_db = rcu_dereference(dev->dn_ptr);
 867         for (ifa = rcu_dereference(dn_db->ifa_list);
 868              ifa != NULL;
 869              ifa = rcu_dereference(ifa->ifa_next)) {
 870                 if (ifa->ifa_scope > scope)
 871                         continue;
 872                 if (!daddr) {
 873                         saddr = ifa->ifa_local;
 874                         break;
 875                 }
 876                 ret = dn_match_addr(daddr, ifa->ifa_local);
 877                 if (ret > best_match)
 878                         saddr = ifa->ifa_local;
 879                 if (best_match == 0)
 880                         saddr = ifa->ifa_local;
 881         }
 882         rcu_read_unlock();
 883
 884         return saddr;
 885 }
 886
 887 static inline __le16 __dn_fib_res_prefsrc(struct dn_fib_res *res)
 888 {
 889         return dnet_select_source(DN_FIB_RES_DEV(*res), DN_FIB_RES_GW(*res), res->scope);
 890 }
 891
 892 static inline __le16 dn_fib_rules_map_destination(__le16 daddr, struct dn_fib_res *res)
 893 {
 894         __le16 mask = dnet_make_mask(res->prefixlen);
 895         return (daddr&~mask)|res->fi->fib_nh->nh_gw;
 896 }
 897
 898 static int dn_route_output_slow(struct dst_entry **pprt, const struct flowi *oldflp, int try_hard)
 899 {
 900         struct flowi fl = { .fld_dst = oldflp->fld_dst,
 901                             .fld_src = oldflp->fld_src,
 902                             .fld_scope = RT_SCOPE_UNIVERSE,
 903                             .mark = oldflp->mark,
 904                             .iif = init_net.loopback_dev->ifindex,
 905                             .oif = oldflp->oif };
 906         struct dn_route *rt = NULL;
 907         struct net_device *dev_out = NULL, *dev;
 908         struct neighbour *neigh = NULL;
 909         unsigned hash;
 910         unsigned flags = 0;
 911         struct dn_fib_res res = { .fi = NULL, .type = RTN_UNICAST };
 912         int err;
 913         int free_res = 0;
 914         __le16 gateway = 0;
 915
 916         if (decnet_debug_level & 16)
 917                 printk(KERN_DEBUG
 918                        "dn_route_output_slow: dst=%04x src=%04x mark=%d"
 919                        " iif=%d oif=%d\n", le16_to_cpu(oldflp->fld_dst),
 920                        le16_to_cpu(oldflp->fld_src),
 921                        oldflp->mark, init_net.loopback_dev->ifindex, oldflp->oif);
 922
 923         /* If we have an output interface, verify its a DECnet device */
 924         if (oldflp->oif) {
 925                 dev_out = dev_get_by_index(&init_net, oldflp->oif);
 926                 err = -ENODEV;
 927                 if (dev_out && dev_out->dn_ptr == NULL) {
 928                         dev_put(dev_out);
 929                         dev_out = NULL;
 930                 }
 931                 if (dev_out == NULL)
 932                         goto out;
 933         }
 934
 935         /* If we have a source address, verify that its a local address */
 936         if (oldflp->fld_src) {
 937                 err = -EADDRNOTAVAIL;
 938
 939                 if (dev_out) {
 940                         if (dn_dev_islocal(dev_out, oldflp->fld_src))
 941                                 goto source_ok;
 942                         dev_put(dev_out);
 943                         goto out;
 944                 }
 945                 rcu_read_lock();
 946                 for_each_netdev_rcu(&init_net, dev) {
 947                         if (!dev->dn_ptr)
 948                                 continue;
 949                         if (!dn_dev_islocal(dev, oldflp->fld_src))
 950                                 continue;
 951                         if ((dev->flags & IFF_LOOPBACK) &&
 952                             oldflp->fld_dst &&
 953                             !dn_dev_islocal(dev, oldflp->fld_dst))
 954                                 continue;
 955
 956                         dev_out = dev;
 957                         break;
 958                 }
 959                 rcu_read_unlock();
 960                 if (dev_out == NULL)
 961                         goto out;
 962                 dev_hold(dev_out);
 963 source_ok:
 964                 ;
 965         }
 966
 967         /* No destination? Assume its local */
 968         if (!fl.fld_dst) {
 969                 fl.fld_dst = fl.fld_src;
 970
 971                 err = -EADDRNOTAVAIL;
 972                 if (dev_out)
 973                         dev_put(dev_out);
 974                 dev_out = init_net.loopback_dev;
 975                 dev_hold(dev_out);
 976                 if (!fl.fld_dst) {
 977                         fl.fld_dst =
 978                         fl.fld_src = dnet_select_source(dev_out, 0,
 979                                                        RT_SCOPE_HOST);
 980                         if (!fl.fld_dst)
 981                                 goto out;
 982                 }
 983                 fl.oif = init_net.loopback_dev->ifindex;
 984                 res.type = RTN_LOCAL;
 985                 goto make_route;
 986         }
 987
 988         if (decnet_debug_level & 16)
 989                 printk(KERN_DEBUG
 990                        "dn_route_output_slow: initial checks complete."
 991                        " dst=%o4x src=%04x oif=%d try_hard=%d\n",
 992                        le16_to_cpu(fl.fld_dst), le16_to_cpu(fl.fld_src),
 993                        fl.oif, try_hard);
 994
 995         /*
 996          * N.B. If the kernel is compiled without router support then
 997          * dn_fib_lookup() will evaluate to non-zero so this if () block
 998          * will always be executed.
 999          */
1000         err = -ESRCH;
1001         if (try_hard || (err = dn_fib_lookup(&fl, &res)) != 0) {
1002                 struct dn_dev *dn_db;
1003                 if (err != -ESRCH)
1004                         goto out;
1005                 /*
1006                  * Here the fallback is basically the standard algorithm for
1007                  * routing in endnodes which is described in the DECnet routing
1008                  * docs
1009                  *
1010                  * If we are not trying hard, look in neighbour cache.
1011                  * The result is tested to ensure that if a specific output
1012                  * device/source address was requested, then we honour that
1013                  * here
1014                  */
1015                 if (!try_hard) {
1016                         neigh = neigh_lookup_nodev(&dn_neigh_table, &init_net, &fl.fld_dst);
1017                         if (neigh) {
1018                                 if ((oldflp->oif &&
1019                                     (neigh->dev->ifindex != oldflp->oif)) ||
1020                                     (oldflp->fld_src &&
1021                                     (!dn_dev_islocal(neigh->dev,
1022                                                       oldflp->fld_src)))) {
1023                                         neigh_release(neigh);
1024                                         neigh = NULL;
1025                                 } else {
1026                                         if (dev_out)
1027                                                 dev_put(dev_out);
1028                                         if (dn_dev_islocal(neigh->dev, fl.fld_dst)) {
1029                                                 dev_out = init_net.loopback_dev;
1030                                                 res.type = RTN_LOCAL;
1031                                         } else {
1032                                                 dev_out = neigh->dev;
1033                                         }
1034                                         dev_hold(dev_out);
1035                                         goto select_source;
1036                                 }
1037                         }
1038                 }
1039
1040                 /* Not there? Perhaps its a local address */
1041                 if (dev_out == NULL)
1042                         dev_out = dn_dev_get_default();
1043                 err = -ENODEV;
1044                 if (dev_out == NULL)
1045                         goto out;
1046                 dn_db = rcu_dereference_raw(dev_out->dn_ptr);
1047                 /* Possible improvement - check all devices for local addr */
1048                 if (dn_dev_islocal(dev_out, fl.fld_dst)) {
1049                         dev_put(dev_out);
1050                         dev_out = init_net.loopback_dev;
1051                         dev_hold(dev_out);
1052                         res.type = RTN_LOCAL;
1053                         goto select_source;
1054                 }
1055                 /* Not local either.... try sending it to the default router */
1056                 neigh = neigh_clone(dn_db->router);
1057                 BUG_ON(neigh && neigh->dev != dev_out);
1058
1059                 /* Ok then, we assume its directly connected and move on */
1060 select_source:
1061                 if (neigh)
1062                         gateway = ((struct dn_neigh *)neigh)->addr;
1063                 if (gateway == 0)
1064                         gateway = fl.fld_dst;
1065                 if (fl.fld_src == 0) {
1066                         fl.fld_src = dnet_select_source(dev_out, gateway,
1067                                                          res.type == RTN_LOCAL ?
1068                                                          RT_SCOPE_HOST :
1069                                                          RT_SCOPE_LINK);
1070                         if (fl.fld_src == 0 && res.type != RTN_LOCAL)
1071                                 goto e_addr;
1072                 }
1073                 fl.oif = dev_out->ifindex;
1074                 goto make_route;
1075         }
1076         free_res = 1;
1077
1078         if (res.type == RTN_NAT)
1079                 goto e_inval;
1080
1081         if (res.type == RTN_LOCAL) {
1082                 if (!fl.fld_src)
1083                         fl.fld_src = fl.fld_dst;
1084                 if (dev_out)
1085                         dev_put(dev_out);
1086                 dev_out = init_net.loopback_dev;
1087                 dev_hold(dev_out);
1088                 fl.oif = dev_out->ifindex;
1089                 if (res.fi)
1090                         dn_fib_info_put(res.fi);
1091                 res.fi = NULL;
1092                 goto make_route;
1093         }
1094
1095         if (res.fi->fib_nhs > 1 && fl.oif == 0)
1096                 dn_fib_select_multipath(&fl, &res);
1097
1098         /*
1099          * We could add some logic to deal with default routes here and
1100          * get rid of some of the special casing above.
1101          */
1102
1103         if (!fl.fld_src)
1104                 fl.fld_src = DN_FIB_RES_PREFSRC(res);
1105
1106         if (dev_out)
1107                 dev_put(dev_out);
1108         dev_out = DN_FIB_RES_DEV(res);
1109         dev_hold(dev_out);
1110         fl.oif = dev_out->ifindex;
1111         gateway = DN_FIB_RES_GW(res);
1112
1113 make_route:
1114         if (dev_out->flags & IFF_LOOPBACK)
1115                 flags |= RTCF_LOCAL;
1116
1117         rt = dst_alloc(&dn_dst_ops);
1118         if (rt == NULL)
1119                 goto e_nobufs;
1120
1121         atomic_set(&rt->dst.__refcnt, 1);
1122         rt->dst.flags   = DST_HOST;
1123
1124         rt->fl.fld_src    = oldflp->fld_src;
1125         rt->fl.fld_dst    = oldflp->fld_dst;
1126         rt->fl.oif        = oldflp->oif;
1127         rt->fl.iif        = 0;
1128         rt->fl.mark       = oldflp->mark;
1129
1130         rt->rt_saddr      = fl.fld_src;
1131         rt->rt_daddr      = fl.fld_dst;
1132         rt->rt_gateway    = gateway ? gateway : fl.fld_dst;
1133         rt->rt_local_src  = fl.fld_src;
1134
1135         rt->rt_dst_map    = fl.fld_dst;
1136         rt->rt_src_map    = fl.fld_src;
1137
1138         rt->dst.dev = dev_out;
1139         dev_hold(dev_out);
1140         rt->dst.neighbour = neigh;
1141         neigh = NULL;
1142
1143         rt->dst.lastuse = jiffies;
1144         rt->dst.output  = dn_output;
1145         rt->dst.input   = dn_rt_bug;
1146         rt->rt_flags      = flags;
1147         if (flags & RTCF_LOCAL)
1148                 rt->dst.input = dn_nsp_rx;
1149
1150         err = dn_rt_set_next_hop(rt, &res);
1151         if (err)
1152                 goto e_neighbour;
1153
1154         hash = dn_hash(rt->fl.fld_src, rt->fl.fld_dst);
1155         dn_insert_route(rt, hash, (struct dn_route **)pprt);
1156
1157 done:
1158         if (neigh)
1159                 neigh_release(neigh);
1160         if (free_res)
1161                 dn_fib_res_put(&res);
1162         if (dev_out)
1163                 dev_put(dev_out);
1164 out:
1165         return err;
1166
1167 e_addr:
1168         err = -EADDRNOTAVAIL;
1169         goto done;
1170 e_inval:
1171         err = -EINVAL;
1172         goto done;
1173 e_nobufs:
1174         err = -ENOBUFS;
1175         goto done;
1176 e_neighbour:
1177         dst_free(&rt->dst);
1178         goto e_nobufs;
1179 }
1180
1181
1182 /*
1183  * N.B. The flags may be moved into the flowi at some future stage.
1184  */
1185 static int __dn_route_output_key(struct dst_entry **pprt, const struct flowi *flp, int flags)
1186 {
1187         unsigned hash = dn_hash(flp->fld_src, flp->fld_dst);
1188         struct dn_route *rt = NULL;
1189
1190         if (!(flags & MSG_TRYHARD)) {
1191                 rcu_read_lock_bh();
1192                 for (rt = rcu_dereference_bh(dn_rt_hash_table[hash].chain); rt;
1193                         rt = rcu_dereference_bh(rt->dst.dn_next)) {
1194                         if ((flp->fld_dst == rt->fl.fld_dst) &&
1195                             (flp->fld_src == rt->fl.fld_src) &&
1196                             (flp->mark == rt->fl.mark) &&
1197                             dn_is_output_route(rt) &&
1198                             (rt->fl.oif == flp->oif)) {
1199                                 dst_use(&rt->dst, jiffies);
1200                                 rcu_read_unlock_bh();
1201                                 *pprt = &rt->dst;
1202                                 return 0;
1203                         }
1204                 }
1205                 rcu_read_unlock_bh();
1206         }
1207
1208         return dn_route_output_slow(pprt, flp, flags);
1209 }
1210
1211 static int dn_route_output_key(struct dst_entry **pprt, struct flowi *flp, int flags)
1212 {
1213         int err;
1214
1215         err = __dn_route_output_key(pprt, flp, flags);
1216         if (err == 0 && flp->proto) {
1217                 err = xfrm_lookup(&init_net, pprt, flp, NULL, 0);
1218         }
1219         return err;
1220 }
1221
1222 int dn_route_output_sock(struct dst_entry **pprt, struct flowi *fl, struct sock *sk, int flags)
1223 {
1224         int err;
1225
1226         err = __dn_route_output_key(pprt, fl, flags & MSG_TRYHARD);
1227         if (err == 0 && fl->proto) {
1228                 err = xfrm_lookup(&init_net, pprt, fl, sk,
1229                                  (flags & MSG_DONTWAIT) ? 0 : XFRM_LOOKUP_WAIT);
1230         }
1231         return err;
1232 }
1233
1234 static int dn_route_input_slow(struct sk_buff *skb)
1235 {
1236         struct dn_route *rt = NULL;
1237         struct dn_skb_cb *cb = DN_SKB_CB(skb);
1238         struct net_device *in_dev = skb->dev;
1239         struct net_device *out_dev = NULL;
1240         struct dn_dev *dn_db;
1241         struct neighbour *neigh = NULL;
1242         unsigned hash;
1243         int flags = 0;
1244         __le16 gateway = 0;
1245         __le16 local_src = 0;
1246         struct flowi fl = { .fld_dst = cb->dst,
1247                             .fld_src = cb->src,
1248                             .fld_scope = RT_SCOPE_UNIVERSE,
1249                             .mark = skb->mark,
1250                             .iif = skb->dev->ifindex };
1251         struct dn_fib_res res = { .fi = NULL, .type = RTN_UNREACHABLE };
1252         int err = -EINVAL;
1253         int free_res = 0;
1254
1255         dev_hold(in_dev);
1256
1257         if ((dn_db = rcu_dereference(in_dev->dn_ptr)) == NULL)
1258                 goto out;
1259
1260         /* Zero source addresses are not allowed */
1261         if (fl.fld_src == 0)
1262                 goto out;
1263
1264         /*
1265          * In this case we've just received a packet from a source
1266          * outside ourselves pretending to come from us. We don't
1267          * allow it any further to prevent routing loops, spoofing and
1268          * other nasties. Loopback packets already have the dst attached
1269          * so this only affects packets which have originated elsewhere.
1270          */
1271         err  = -ENOTUNIQ;
1272         if (dn_dev_islocal(in_dev, cb->src))
1273                 goto out;
1274
1275         err = dn_fib_lookup(&fl, &res);
1276         if (err) {
1277                 if (err != -ESRCH)
1278                         goto out;
1279                 /*
1280                  * Is the destination us ?
1281                  */
1282                 if (!dn_dev_islocal(in_dev, cb->dst))
1283                         goto e_inval;
1284
1285                 res.type = RTN_LOCAL;
1286         } else {
1287                 __le16 src_map = fl.fld_src;
1288                 free_res = 1;
1289
1290                 out_dev = DN_FIB_RES_DEV(res);
1291                 if (out_dev == NULL) {
1292                         if (net_ratelimit())
1293                                 printk(KERN_CRIT "Bug in dn_route_input_slow() "
1294                                                  "No output device\n");
1295                         goto e_inval;
1296                 }
1297                 dev_hold(out_dev);
1298
1299                 if (res.r)
1300                         src_map = fl.fld_src; /* no NAT support for now */
1301
1302                 gateway = DN_FIB_RES_GW(res);
1303                 if (res.type == RTN_NAT) {
1304                         fl.fld_dst = dn_fib_rules_map_destination(fl.fld_dst, &res);
1305                         dn_fib_res_put(&res);
1306                         free_res = 0;
1307                         if (dn_fib_lookup(&fl, &res))
1308                                 goto e_inval;
1309                         free_res = 1;
1310                         if (res.type != RTN_UNICAST)
1311                                 goto e_inval;
1312                         flags |= RTCF_DNAT;
1313                         gateway = fl.fld_dst;
1314                 }
1315                 fl.fld_src = src_map;
1316         }
1317
1318         switch(res.type) {
1319         case RTN_UNICAST:
1320                 /*
1321                  * Forwarding check here, we only check for forwarding
1322                  * being turned off, if you want to only forward intra
1323                  * area, its up to you to set the routing tables up
1324                  * correctly.
1325                  */
1326                 if (dn_db->parms.forwarding == 0)
1327                         goto e_inval;
1328
1329                 if (res.fi->fib_nhs > 1 && fl.oif == 0)
1330                         dn_fib_select_multipath(&fl, &res);
1331
1332                 /*
1333                  * Check for out_dev == in_dev. We use the RTCF_DOREDIRECT
1334                  * flag as a hint to set the intra-ethernet bit when
1335                  * forwarding. If we've got NAT in operation, we don't do
1336                  * this optimisation.
1337                  */
1338                 if (out_dev == in_dev && !(flags & RTCF_NAT))
1339                         flags |= RTCF_DOREDIRECT;
1340
1341                 local_src = DN_FIB_RES_PREFSRC(res);
1342
1343         case RTN_BLACKHOLE:
1344         case RTN_UNREACHABLE:
1345                 break;
1346         case RTN_LOCAL:
1347                 flags |= RTCF_LOCAL;
1348                 fl.fld_src = cb->dst;
1349                 fl.fld_dst = cb->src;
1350
1351                 /* Routing tables gave us a gateway */
1352                 if (gateway)
1353                         goto make_route;
1354
1355                 /* Packet was intra-ethernet, so we know its on-link */
1356                 if (cb->rt_flags & DN_RT_F_IE) {
1357                         gateway = cb->src;
1358                         flags |= RTCF_DIRECTSRC;
1359                         goto make_route;
1360                 }
1361
1362                 /* Use the default router if there is one */
1363                 neigh = neigh_clone(dn_db->router);
1364                 if (neigh) {
1365                         gateway = ((struct dn_neigh *)neigh)->addr;
1366                         goto make_route;
1367                 }
1368
1369                 /* Close eyes and pray */
1370                 gateway = cb->src;
1371                 flags |= RTCF_DIRECTSRC;
1372                 goto make_route;
1373         default:
1374                 goto e_inval;
1375         }
1376
1377 make_route:
1378         rt = dst_alloc(&dn_dst_ops);
1379         if (rt == NULL)
1380                 goto e_nobufs;
1381
1382         rt->rt_saddr      = fl.fld_src;
1383         rt->rt_daddr      = fl.fld_dst;
1384         rt->rt_gateway    = fl.fld_dst;
1385         if (gateway)
1386                 rt->rt_gateway = gateway;
1387         rt->rt_local_src  = local_src ? local_src : rt->rt_saddr;
1388
1389         rt->rt_dst_map    = fl.fld_dst;
1390         rt->rt_src_map    = fl.fld_src;
1391
1392         rt->fl.fld_src    = cb->src;
1393         rt->fl.fld_dst    = cb->dst;
1394         rt->fl.oif        = 0;
1395         rt->fl.iif        = in_dev->ifindex;
1396         rt->fl.mark       = fl.mark;
1397
1398         rt->dst.flags = DST_HOST;
1399         rt->dst.neighbour = neigh;
1400         rt->dst.dev = out_dev;
1401         rt->dst.lastuse = jiffies;
1402         rt->dst.output = dn_rt_bug;
1403         switch(res.type) {
1404                 case RTN_UNICAST:
1405                         rt->dst.input = dn_forward;
1406                         break;
1407                 case RTN_LOCAL:
1408                         rt->dst.output = dn_output;
1409                         rt->dst.input = dn_nsp_rx;
1410                         rt->dst.dev = in_dev;
1411                         flags |= RTCF_LOCAL;
1412                         break;
1413                 default:
1414                 case RTN_UNREACHABLE:
1415                 case RTN_BLACKHOLE:
1416                         rt->dst.input = dst_discard;
1417         }
1418         rt->rt_flags = flags;
1419         if (rt->dst.dev)
1420                 dev_hold(rt->dst.dev);
1421
1422         err = dn_rt_set_next_hop(rt, &res);
1423         if (err)
1424                 goto e_neighbour;
1425
1426         hash = dn_hash(rt->fl.fld_src, rt->fl.fld_dst);
1427         dn_insert_route(rt, hash, &rt);
1428         skb_dst_set(skb, &rt->dst);
1429
1430 done:
1431         if (neigh)
1432                 neigh_release(neigh);
1433         if (free_res)
1434                 dn_fib_res_put(&res);
1435         dev_put(in_dev);
1436         if (out_dev)
1437                 dev_put(out_dev);
1438 out:
1439         return err;
1440
1441 e_inval:
1442         err = -EINVAL;
1443         goto done;
1444
1445 e_nobufs:
1446         err = -ENOBUFS;
1447         goto done;
1448
1449 e_neighbour:
1450         dst_free(&rt->dst);
1451         goto done;
1452 }
1453
1454 static int dn_route_input(struct sk_buff *skb)
1455 {
1456         struct dn_route *rt;
1457         struct dn_skb_cb *cb = DN_SKB_CB(skb);
1458         unsigned hash = dn_hash(cb->src, cb->dst);
1459
1460         if (skb_dst(skb))
1461                 return 0;
1462
1463         rcu_read_lock();
1464         for(rt = rcu_dereference(dn_rt_hash_table[hash].chain); rt != NULL;
1465             rt = rcu_dereference(rt->dst.dn_next)) {
1466                 if ((rt->fl.fld_src == cb->src) &&
1467                     (rt->fl.fld_dst == cb->dst) &&
1468                     (rt->fl.oif == 0) &&
1469                     (rt->fl.mark == skb->mark) &&
1470                     (rt->fl.iif == cb->iif)) {
1471                         dst_use(&rt->dst, jiffies);
1472                         rcu_read_unlock();
1473                         skb_dst_set(skb, (struct dst_entry *)rt);
1474                         return 0;
1475                 }
1476         }
1477         rcu_read_unlock();
1478
1479         return dn_route_input_slow(skb);
1480 }
1481
1482 static int dn_rt_fill_info(struct sk_buff *skb, u32 pid, u32 seq,
1483                            int event, int nowait, unsigned int flags)
1484 {
1485         struct dn_route *rt = (struct dn_route *)skb_dst(skb);
1486         struct rtmsg *r;
1487         struct nlmsghdr *nlh;
1488         unsigned char *b = skb_tail_pointer(skb);
1489         long expires;
1490
1491         nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*r), flags);
1492         r = NLMSG_DATA(nlh);
1493         r->rtm_family = AF_DECnet;
1494         r->rtm_dst_len = 16;
1495         r->rtm_src_len = 0;
1496         r->rtm_tos = 0;
1497         r->rtm_table = RT_TABLE_MAIN;
1498         RTA_PUT_U32(skb, RTA_TABLE, RT_TABLE_MAIN);
1499         r->rtm_type = rt->rt_type;
1500         r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
1501         r->rtm_scope = RT_SCOPE_UNIVERSE;
1502         r->rtm_protocol = RTPROT_UNSPEC;
1503         if (rt->rt_flags & RTCF_NOTIFY)
1504                 r->rtm_flags |= RTM_F_NOTIFY;
1505         RTA_PUT(skb, RTA_DST, 2, &rt->rt_daddr);
1506         if (rt->fl.fld_src) {
1507                 r->rtm_src_len = 16;
1508                 RTA_PUT(skb, RTA_SRC, 2, &rt->fl.fld_src);
1509         }
1510         if (rt->dst.dev)
1511                 RTA_PUT(skb, RTA_OIF, sizeof(int), &rt->dst.dev->ifindex);
1512         /*
1513          * Note to self - change this if input routes reverse direction when
1514          * they deal only with inputs and not with replies like they do
1515          * currently.
1516          */
1517         RTA_PUT(skb, RTA_PREFSRC, 2, &rt->rt_local_src);
1518         if (rt->rt_daddr != rt->rt_gateway)
1519                 RTA_PUT(skb, RTA_GATEWAY, 2, &rt->rt_gateway);
1520         if (rtnetlink_put_metrics(skb, dst_metrics_ptr(&rt->dst)) < 0)
1521                 goto rtattr_failure;
1522         expires = rt->dst.expires ? rt->dst.expires - jiffies : 0;
1523         if (rtnl_put_cacheinfo(skb, &rt->dst, 0, 0, 0, expires,
1524                                rt->dst.error) < 0)
1525                 goto rtattr_failure;
1526         if (dn_is_input_route(rt))
1527                 RTA_PUT(skb, RTA_IIF, sizeof(int), &rt->fl.iif);
1528
1529         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1530         return skb->len;
1531
1532 nlmsg_failure:
1533 rtattr_failure:
1534         nlmsg_trim(skb, b);
1535         return -1;
1536 }
1537
1538 /*
1539  * This is called by both endnodes and routers now.
1540  */
1541 static int dn_cache_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh, void *arg)
1542 {
1543         struct net *net = sock_net(in_skb->sk);
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         if (!net_eq(net, &init_net))
1553                 return -EINVAL;
1554
1555         memset(&fl, 0, sizeof(fl));
1556         fl.proto = DNPROTO_NSP;
1557
1558         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1559         if (skb == NULL)
1560                 return -ENOBUFS;
1561         skb_reset_mac_header(skb);
1562         cb = DN_SKB_CB(skb);
1563
1564         if (rta[RTA_SRC-1])
1565                 memcpy(&fl.fld_src, RTA_DATA(rta[RTA_SRC-1]), 2);
1566         if (rta[RTA_DST-1])
1567                 memcpy(&fl.fld_dst, RTA_DATA(rta[RTA_DST-1]), 2);
1568         if (rta[RTA_IIF-1])
1569                 memcpy(&fl.iif, RTA_DATA(rta[RTA_IIF-1]), sizeof(int));
1570
1571         if (fl.iif) {
1572                 struct net_device *dev;
1573                 if ((dev = dev_get_by_index(&init_net, fl.iif)) == NULL) {
1574                         kfree_skb(skb);
1575                         return -ENODEV;
1576                 }
1577                 if (!dev->dn_ptr) {
1578                         dev_put(dev);
1579                         kfree_skb(skb);
1580                         return -ENODEV;
1581                 }
1582                 skb->protocol = htons(ETH_P_DNA_RT);
1583                 skb->dev = dev;
1584                 cb->src = fl.fld_src;
1585                 cb->dst = fl.fld_dst;
1586                 local_bh_disable();
1587                 err = dn_route_input(skb);
1588                 local_bh_enable();
1589                 memset(cb, 0, sizeof(struct dn_skb_cb));
1590                 rt = (struct dn_route *)skb_dst(skb);
1591                 if (!err && -rt->dst.error)
1592                         err = rt->dst.error;
1593         } else {
1594                 int oif = 0;
1595                 if (rta[RTA_OIF - 1])
1596                         memcpy(&oif, RTA_DATA(rta[RTA_OIF - 1]), sizeof(int));
1597                 fl.oif = oif;
1598                 err = dn_route_output_key((struct dst_entry **)&rt, &fl, 0);
1599         }
1600
1601         if (skb->dev)
1602                 dev_put(skb->dev);
1603         skb->dev = NULL;
1604         if (err)
1605                 goto out_free;
1606         skb_dst_set(skb, &rt->dst);
1607         if (rtm->rtm_flags & RTM_F_NOTIFY)
1608                 rt->rt_flags |= RTCF_NOTIFY;
1609
1610         err = dn_rt_fill_info(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq, RTM_NEWROUTE, 0, 0);
1611
1612         if (err == 0)
1613                 goto out_free;
1614         if (err < 0) {
1615                 err = -EMSGSIZE;
1616                 goto out_free;
1617         }
1618
1619         return rtnl_unicast(skb, &init_net, NETLINK_CB(in_skb).pid);
1620
1621 out_free:
1622         kfree_skb(skb);
1623         return err;
1624 }
1625
1626 /*
1627  * For routers, this is called from dn_fib_dump, but for endnodes its
1628  * called directly from the rtnetlink dispatch table.
1629  */
1630 int dn_cache_dump(struct sk_buff *skb, struct netlink_callback *cb)
1631 {
1632         struct net *net = sock_net(skb->sk);
1633         struct dn_route *rt;
1634         int h, s_h;
1635         int idx, s_idx;
1636
1637         if (!net_eq(net, &init_net))
1638                 return 0;
1639
1640         if (NLMSG_PAYLOAD(cb->nlh, 0) < sizeof(struct rtmsg))
1641                 return -EINVAL;
1642         if (!(((struct rtmsg *)NLMSG_DATA(cb->nlh))->rtm_flags&RTM_F_CLONED))
1643                 return 0;
1644
1645         s_h = cb->args[0];
1646         s_idx = idx = cb->args[1];
1647         for(h = 0; h <= dn_rt_hash_mask; h++) {
1648                 if (h < s_h)
1649                         continue;
1650                 if (h > s_h)
1651                         s_idx = 0;
1652                 rcu_read_lock_bh();
1653                 for(rt = rcu_dereference_bh(dn_rt_hash_table[h].chain), idx = 0;
1654                         rt;
1655                         rt = rcu_dereference_bh(rt->dst.dn_next), idx++) {
1656                         if (idx < s_idx)
1657                                 continue;
1658                         skb_dst_set(skb, dst_clone(&rt->dst));
1659                         if (dn_rt_fill_info(skb, NETLINK_CB(cb->skb).pid,
1660                                         cb->nlh->nlmsg_seq, RTM_NEWROUTE,
1661                                         1, NLM_F_MULTI) <= 0) {
1662                                 skb_dst_drop(skb);
1663                                 rcu_read_unlock_bh();
1664                                 goto done;
1665                         }
1666                         skb_dst_drop(skb);
1667                 }
1668                 rcu_read_unlock_bh();
1669         }
1670
1671 done:
1672         cb->args[0] = h;
1673         cb->args[1] = idx;
1674         return skb->len;
1675 }
1676
1677 #ifdef CONFIG_PROC_FS
1678 struct dn_rt_cache_iter_state {
1679         int bucket;
1680 };
1681
1682 static struct dn_route *dn_rt_cache_get_first(struct seq_file *seq)
1683 {
1684         struct dn_route *rt = NULL;
1685         struct dn_rt_cache_iter_state *s = seq->private;
1686
1687         for(s->bucket = dn_rt_hash_mask; s->bucket >= 0; --s->bucket) {
1688                 rcu_read_lock_bh();
1689                 rt = rcu_dereference_bh(dn_rt_hash_table[s->bucket].chain);
1690                 if (rt)
1691                         break;
1692                 rcu_read_unlock_bh();
1693         }
1694         return rt;
1695 }
1696
1697 static struct dn_route *dn_rt_cache_get_next(struct seq_file *seq, struct dn_route *rt)
1698 {
1699         struct dn_rt_cache_iter_state *s = seq->private;
1700
1701         rt = rcu_dereference_bh(rt->dst.dn_next);
1702         while (!rt) {
1703                 rcu_read_unlock_bh();
1704                 if (--s->bucket < 0)
1705                         break;
1706                 rcu_read_lock_bh();
1707                 rt = rcu_dereference_bh(dn_rt_hash_table[s->bucket].chain);
1708         }
1709         return rt;
1710 }
1711
1712 static void *dn_rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
1713 {
1714         struct dn_route *rt = dn_rt_cache_get_first(seq);
1715
1716         if (rt) {
1717                 while(*pos && (rt = dn_rt_cache_get_next(seq, rt)))
1718                         --*pos;
1719         }
1720         return *pos ? NULL : rt;
1721 }
1722
1723 static void *dn_rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1724 {
1725         struct dn_route *rt = dn_rt_cache_get_next(seq, v);
1726         ++*pos;
1727         return rt;
1728 }
1729
1730 static void dn_rt_cache_seq_stop(struct seq_file *seq, void *v)
1731 {
1732         if (v)
1733                 rcu_read_unlock_bh();
1734 }
1735
1736 static int dn_rt_cache_seq_show(struct seq_file *seq, void *v)
1737 {
1738         struct dn_route *rt = v;
1739         char buf1[DN_ASCBUF_LEN], buf2[DN_ASCBUF_LEN];
1740
1741         seq_printf(seq, "%-8s %-7s %-7s %04d %04d %04d\n",
1742                         rt->dst.dev ? rt->dst.dev->name : "*",
1743                         dn_addr2asc(le16_to_cpu(rt->rt_daddr), buf1),
1744                         dn_addr2asc(le16_to_cpu(rt->rt_saddr), buf2),
1745                         atomic_read(&rt->dst.__refcnt),
1746                         rt->dst.__use,
1747                         (int) dst_metric(&rt->dst, RTAX_RTT));
1748         return 0;
1749 }
1750
1751 static const struct seq_operations dn_rt_cache_seq_ops = {
1752         .start  = dn_rt_cache_seq_start,
1753         .next   = dn_rt_cache_seq_next,
1754         .stop   = dn_rt_cache_seq_stop,
1755         .show   = dn_rt_cache_seq_show,
1756 };
1757
1758 static int dn_rt_cache_seq_open(struct inode *inode, struct file *file)
1759 {
1760         return seq_open_private(file, &dn_rt_cache_seq_ops,
1761                         sizeof(struct dn_rt_cache_iter_state));
1762 }
1763
1764 static const struct file_operations dn_rt_cache_seq_fops = {
1765         .owner   = THIS_MODULE,
1766         .open    = dn_rt_cache_seq_open,
1767         .read    = seq_read,
1768         .llseek  = seq_lseek,
1769         .release = seq_release_private,
1770 };
1771
1772 #endif /* CONFIG_PROC_FS */
1773
1774 void __init dn_route_init(void)
1775 {
1776         int i, goal, order;
1777
1778         dn_dst_ops.kmem_cachep =
1779                 kmem_cache_create("dn_dst_cache", sizeof(struct dn_route), 0,
1780                                   SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1781         dst_entries_init(&dn_dst_ops);
1782         setup_timer(&dn_route_timer, dn_dst_check_expire, 0);
1783         dn_route_timer.expires = jiffies + decnet_dst_gc_interval * HZ;
1784         add_timer(&dn_route_timer);
1785
1786         goal = totalram_pages >> (26 - PAGE_SHIFT);
1787
1788         for(order = 0; (1UL << order) < goal; order++)
1789                 /* NOTHING */;
1790
1791         /*
1792          * Only want 1024 entries max, since the table is very, very unlikely
1793          * to be larger than that.
1794          */
1795         while(order && ((((1UL << order) * PAGE_SIZE) /
1796                                 sizeof(struct dn_rt_hash_bucket)) >= 2048))
1797                 order--;
1798
1799         do {
1800                 dn_rt_hash_mask = (1UL << order) * PAGE_SIZE /
1801                         sizeof(struct dn_rt_hash_bucket);
1802                 while(dn_rt_hash_mask & (dn_rt_hash_mask - 1))
1803                         dn_rt_hash_mask--;
1804                 dn_rt_hash_table = (struct dn_rt_hash_bucket *)
1805                         __get_free_pages(GFP_ATOMIC, order);
1806         } while (dn_rt_hash_table == NULL && --order > 0);
1807
1808         if (!dn_rt_hash_table)
1809                 panic("Failed to allocate DECnet route cache hash table\n");
1810
1811         printk(KERN_INFO
1812                 "DECnet: Routing cache hash table of %u buckets, %ldKbytes\n",
1813                 dn_rt_hash_mask,
1814                 (long)(dn_rt_hash_mask*sizeof(struct dn_rt_hash_bucket))/1024);
1815
1816         dn_rt_hash_mask--;
1817         for(i = 0; i <= dn_rt_hash_mask; i++) {
1818                 spin_lock_init(&dn_rt_hash_table[i].lock);
1819                 dn_rt_hash_table[i].chain = NULL;
1820         }
1821
1822         dn_dst_ops.gc_thresh = (dn_rt_hash_mask + 1);
1823
1824         proc_net_fops_create(&init_net, "decnet_cache", S_IRUGO, &dn_rt_cache_seq_fops);
1825
1826 #ifdef CONFIG_DECNET_ROUTER
1827         rtnl_register(PF_DECnet, RTM_GETROUTE, dn_cache_getroute, dn_fib_dump);
1828 #else
1829         rtnl_register(PF_DECnet, RTM_GETROUTE, dn_cache_getroute,
1830                       dn_cache_dump);
1831 #endif
1832 }
1833
1834 void __exit dn_route_cleanup(void)
1835 {
1836         del_timer(&dn_route_timer);
1837         dn_run_flush(0);
1838
1839         proc_net_remove(&init_net, "decnet_cache");
1840         dst_entries_destroy(&dn_dst_ops);
1841 }
1842