net/decnet/dn_route.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * DECnet       An implementation of the DECnet protocol suite for the LINUX
   4  *              operating system.  DECnet is implemented using the  BSD Socket
   5  *              interface as the means of communication with the user level.
   6  *
   7  *              DECnet Routing Functions (Endnode and Router)
   8  *
   9  * Authors:     Steve Whitehouse <SteveW@ACM.org>
  10  *              Eduardo Marcelo Serrat <emserrat@geocities.com>
  11  *
  12  * Changes:
  13  *              Steve Whitehouse : Fixes to allow "intra-ethernet" and
  14  *                                 "return-to-sender" bits on outgoing
  15  *                                 packets.
  16  *              Steve Whitehouse : Timeouts for cached routes.
  17  *              Steve Whitehouse : Use dst cache for input routes too.
  18  *              Steve Whitehouse : Fixed error values in dn_send_skb.
  19  *              Steve Whitehouse : Rework routing functions to better fit
  20  *                                 DECnet routing design
  21  *              Alexey Kuznetsov : New SMP locking
  22  *              Steve Whitehouse : More SMP locking changes & dn_cache_dump()
  23  *              Steve Whitehouse : Prerouting NF hook, now really is prerouting.
  24  *                                 Fixed possible skb leak in rtnetlink funcs.
  25  *              Steve Whitehouse : Dave Miller's dynamic hash table sizing and
  26  *                                 Alexey Kuznetsov's finer grained locking
  27  *                                 from ipv4/route.c.
  28  *              Steve Whitehouse : Routing is now starting to look like a
  29  *                                 sensible set of code now, mainly due to
  30  *                                 my copying the IPv4 routing code. The
  31  *                                 hooks here are modified and will continue
  32  *                                 to evolve for a while.
  33  *              Steve Whitehouse : Real SMP at last :-) Also new netfilter
  34  *                                 stuff. Look out raw sockets your days
  35  *                                 are numbered!
  36  *              Steve Whitehouse : Added return-to-sender functions. Added
  37  *                                 backlog congestion level return codes.
  38  *              Steve Whitehouse : Fixed bug where routes were set up with
  39  *                                 no ref count on net devices.
  40  *              Steve Whitehouse : RCU for the route cache
  41  *              Steve Whitehouse : Preparations for the flow cache
  42  *              Steve Whitehouse : Prepare for nonlinear skbs
  43  */
  44
  45 /******************************************************************************
  46     (c) 1995-1998 E.M. Serrat           emserrat@geocities.com
  47
  48 *******************************************************************************/
  49
  50 #include <linux/errno.h>
  51 #include <linux/types.h>
  52 #include <linux/socket.h>
  53 #include <linux/in.h>
  54 #include <linux/kernel.h>
  55 #include <linux/sockios.h>
  56 #include <linux/net.h>
  57 #include <linux/netdevice.h>
  58 #include <linux/inet.h>
  59 #include <linux/route.h>
  60 #include <linux/in_route.h>
  61 #include <linux/slab.h>
  62 #include <net/sock.h>
  63 #include <linux/mm.h>
  64 #include <linux/proc_fs.h>
  65 #include <linux/seq_file.h>
  66 #include <linux/init.h>
  67 #include <linux/rtnetlink.h>
  68 #include <linux/string.h>
  69 #include <linux/netfilter_decnet.h>
  70 #include <linux/rcupdate.h>
  71 #include <linux/times.h>
  72 #include <linux/export.h>
  73 #include <asm/errno.h>
  74 #include <net/net_namespace.h>
  75 #include <net/netlink.h>
  76 #include <net/neighbour.h>
  77 #include <net/dst.h>
  78 #include <net/flow.h>
  79 #include <net/fib_rules.h>
  80 #include <net/dn.h>
  81 #include <net/dn_dev.h>
  82 #include <net/dn_nsp.h>
  83 #include <net/dn_route.h>
  84 #include <net/dn_neigh.h>
  85 #include <net/dn_fib.h>
  86
  87 struct dn_rt_hash_bucket {
  88         struct dn_route __rcu *chain;
  89         spinlock_t lock;
  90 };
  91
  92 extern struct neigh_table dn_neigh_table;
  93
  94
  95 static unsigned char dn_hiord_addr[6] = {0xAA, 0x00, 0x04, 0x00, 0x00, 0x00};
  96
  97 static const int dn_rt_min_delay = 2 * HZ;
  98 static const int dn_rt_max_delay = 10 * HZ;
  99 static const int dn_rt_mtu_expires = 10 * 60 * HZ;
 100
 101 static unsigned long dn_rt_deadline;
 102
 103 static int dn_dst_gc(struct dst_ops *ops);
 104 static struct dst_entry *dn_dst_check(struct dst_entry *, __u32);
 105 static unsigned int dn_dst_default_advmss(const struct dst_entry *dst);
 106 static unsigned int dn_dst_mtu(const struct dst_entry *dst);
 107 static void dn_dst_destroy(struct dst_entry *);
 108 static void dn_dst_ifdown(struct dst_entry *, struct net_device *dev, int how);
 109 static struct dst_entry *dn_dst_negative_advice(struct dst_entry *);
 110 static void dn_dst_link_failure(struct sk_buff *);
 111 static void dn_dst_update_pmtu(struct dst_entry *dst, struct sock *sk,
 112                                struct sk_buff *skb , u32 mtu,
 113                                bool confirm_neigh);
 114 static void dn_dst_redirect(struct dst_entry *dst, struct sock *sk,
 115                             struct sk_buff *skb);
 116 static struct neighbour *dn_dst_neigh_lookup(const struct dst_entry *dst,
 117                                              struct sk_buff *skb,
 118                                              const void *daddr);
 119 static int dn_route_input(struct sk_buff *);
 120 static void dn_run_flush(struct timer_list *unused);
 121
 122 static struct dn_rt_hash_bucket *dn_rt_hash_table;
 123 static unsigned int dn_rt_hash_mask;
 124
 125 static struct timer_list dn_route_timer;
 126 static DEFINE_TIMER(dn_rt_flush_timer, dn_run_flush);
 127 int decnet_dst_gc_interval = 2;
 128
 129 static struct dst_ops dn_dst_ops = {
 130         .family =               PF_DECnet,
 131         .gc_thresh =            128,
 132         .gc =                   dn_dst_gc,
 133         .check =                dn_dst_check,
 134         .default_advmss =       dn_dst_default_advmss,
 135         .mtu =                  dn_dst_mtu,
 136         .cow_metrics =          dst_cow_metrics_generic,
 137         .destroy =              dn_dst_destroy,
 138         .ifdown =               dn_dst_ifdown,
 139         .negative_advice =      dn_dst_negative_advice,
 140         .link_failure =         dn_dst_link_failure,
 141         .update_pmtu =          dn_dst_update_pmtu,
 142         .redirect =             dn_dst_redirect,
 143         .neigh_lookup =         dn_dst_neigh_lookup,
 144 };
 145
 146 static void dn_dst_destroy(struct dst_entry *dst)
 147 {
 148         struct dn_route *rt = (struct dn_route *) dst;
 149
 150         if (rt->n)
 151                 neigh_release(rt->n);
 152         dst_destroy_metrics_generic(dst);
 153 }
 154
 155 static void dn_dst_ifdown(struct dst_entry *dst, struct net_device *dev, int how)
 156 {
 157         if (how) {
 158                 struct dn_route *rt = (struct dn_route *) dst;
 159                 struct neighbour *n = rt->n;
 160
 161                 if (n && n->dev == dev) {
 162                         n->dev = dev_net(dev)->loopback_dev;
 163                         dev_hold(n->dev);
 164                         dev_put(dev);
 165                 }
 166         }
 167 }
 168
 169 static __inline__ unsigned int dn_hash(__le16 src, __le16 dst)
 170 {
 171         __u16 tmp = (__u16 __force)(src ^ dst);
 172         tmp ^= (tmp >> 3);
 173         tmp ^= (tmp >> 5);
 174         tmp ^= (tmp >> 10);
 175         return dn_rt_hash_mask & (unsigned int)tmp;
 176 }
 177
 178 static void dn_dst_check_expire(struct timer_list *unused)
 179 {
 180         int i;
 181         struct dn_route *rt;
 182         struct dn_route __rcu **rtp;
 183         unsigned long now = jiffies;
 184         unsigned long expire = 120 * HZ;
 185
 186         for (i = 0; i <= dn_rt_hash_mask; i++) {
 187                 rtp = &dn_rt_hash_table[i].chain;
 188
 189                 spin_lock(&dn_rt_hash_table[i].lock);
 190                 while ((rt = rcu_dereference_protected(*rtp,
 191                                                 lockdep_is_held(&dn_rt_hash_table[i].lock))) != NULL) {
 192                         if (atomic_read(&rt->dst.__refcnt) > 1 ||
 193                             (now - rt->dst.lastuse) < expire) {
 194                                 rtp = &rt->dn_next;
 195                                 continue;
 196                         }
 197                         *rtp = rt->dn_next;
 198                         rt->dn_next = NULL;
 199                         dst_dev_put(&rt->dst);
 200                         dst_release(&rt->dst);
 201                 }
 202                 spin_unlock(&dn_rt_hash_table[i].lock);
 203
 204                 if ((jiffies - now) > 0)
 205                         break;
 206         }
 207
 208         mod_timer(&dn_route_timer, now + decnet_dst_gc_interval * HZ);
 209 }
 210
 211 static int dn_dst_gc(struct dst_ops *ops)
 212 {
 213         struct dn_route *rt;
 214         struct dn_route __rcu **rtp;
 215         int i;
 216         unsigned long now = jiffies;
 217         unsigned long expire = 10 * HZ;
 218
 219         for (i = 0; i <= dn_rt_hash_mask; i++) {
 220
 221                 spin_lock_bh(&dn_rt_hash_table[i].lock);
 222                 rtp = &dn_rt_hash_table[i].chain;
 223
 224                 while ((rt = rcu_dereference_protected(*rtp,
 225                                                 lockdep_is_held(&dn_rt_hash_table[i].lock))) != NULL) {
 226                         if (atomic_read(&rt->dst.__refcnt) > 1 ||
 227                             (now - rt->dst.lastuse) < expire) {
 228                                 rtp = &rt->dn_next;
 229                                 continue;
 230                         }
 231                         *rtp = rt->dn_next;
 232                         rt->dn_next = NULL;
 233                         dst_dev_put(&rt->dst);
 234                         dst_release(&rt->dst);
 235                         break;
 236                 }
 237                 spin_unlock_bh(&dn_rt_hash_table[i].lock);
 238         }
 239
 240         return 0;
 241 }
 242
 243 /*
 244  * The decnet standards don't impose a particular minimum mtu, what they
 245  * do insist on is that the routing layer accepts a datagram of at least
 246  * 230 bytes long. Here we have to subtract the routing header length from
 247  * 230 to get the minimum acceptable mtu. If there is no neighbour, then we
 248  * assume the worst and use a long header size.
 249  *
 250  * We update both the mtu and the advertised mss (i.e. the segment size we
 251  * advertise to the other end).
 252  */
 253 static void dn_dst_update_pmtu(struct dst_entry *dst, struct sock *sk,
 254                                struct sk_buff *skb, u32 mtu,
 255                                bool confirm_neigh)
 256 {
 257         struct dn_route *rt = (struct dn_route *) dst;
 258         struct neighbour *n = rt->n;
 259         u32 min_mtu = 230;
 260         struct dn_dev *dn;
 261
 262         dn = n ? rcu_dereference_raw(n->dev->dn_ptr) : NULL;
 263
 264         if (dn && dn->use_long == 0)
 265                 min_mtu -= 6;
 266         else
 267                 min_mtu -= 21;
 268
 269         if (dst_metric(dst, RTAX_MTU) > mtu && mtu >= min_mtu) {
 270                 if (!(dst_metric_locked(dst, RTAX_MTU))) {
 271                         dst_metric_set(dst, RTAX_MTU, mtu);
 272                         dst_set_expires(dst, dn_rt_mtu_expires);
 273                 }
 274                 if (!(dst_metric_locked(dst, RTAX_ADVMSS))) {
 275                         u32 mss = mtu - DN_MAX_NSP_DATA_HEADER;
 276                         u32 existing_mss = dst_metric_raw(dst, RTAX_ADVMSS);
 277                         if (!existing_mss || existing_mss > mss)
 278                                 dst_metric_set(dst, RTAX_ADVMSS, mss);
 279                 }
 280         }
 281 }
 282
 283 static void dn_dst_redirect(struct dst_entry *dst, struct sock *sk,
 284                             struct sk_buff *skb)
 285 {
 286 }
 287
 288 /*
 289  * When a route has been marked obsolete. (e.g. routing cache flush)
 290  */
 291 static struct dst_entry *dn_dst_check(struct dst_entry *dst, __u32 cookie)
 292 {
 293         return NULL;
 294 }
 295
 296 static struct dst_entry *dn_dst_negative_advice(struct dst_entry *dst)
 297 {
 298         dst_release(dst);
 299         return NULL;
 300 }
 301
 302 static void dn_dst_link_failure(struct sk_buff *skb)
 303 {
 304 }
 305
 306 static inline int compare_keys(struct flowidn *fl1, struct flowidn *fl2)
 307 {
 308         return ((fl1->daddr ^ fl2->daddr) |
 309                 (fl1->saddr ^ fl2->saddr) |
 310                 (fl1->flowidn_mark ^ fl2->flowidn_mark) |
 311                 (fl1->flowidn_scope ^ fl2->flowidn_scope) |
 312                 (fl1->flowidn_oif ^ fl2->flowidn_oif) |
 313                 (fl1->flowidn_iif ^ fl2->flowidn_iif)) == 0;
 314 }
 315
 316 static int dn_insert_route(struct dn_route *rt, unsigned int hash, struct dn_route **rp)
 317 {
 318         struct dn_route *rth;
 319         struct dn_route __rcu **rthp;
 320         unsigned long now = jiffies;
 321
 322         rthp = &dn_rt_hash_table[hash].chain;
 323
 324         spin_lock_bh(&dn_rt_hash_table[hash].lock);
 325         while ((rth = rcu_dereference_protected(*rthp,
 326                                                 lockdep_is_held(&dn_rt_hash_table[hash].lock))) != NULL) {
 327                 if (compare_keys(&rth->fld, &rt->fld)) {
 328                         /* Put it first */
 329                         *rthp = rth->dn_next;
 330                         rcu_assign_pointer(rth->dn_next,
 331                                            dn_rt_hash_table[hash].chain);
 332                         rcu_assign_pointer(dn_rt_hash_table[hash].chain, rth);
 333
 334                         dst_hold_and_use(&rth->dst, now);
 335                         spin_unlock_bh(&dn_rt_hash_table[hash].lock);
 336
 337                         dst_release_immediate(&rt->dst);
 338                         *rp = rth;
 339                         return 0;
 340                 }
 341                 rthp = &rth->dn_next;
 342         }
 343
 344         rcu_assign_pointer(rt->dn_next, dn_rt_hash_table[hash].chain);
 345         rcu_assign_pointer(dn_rt_hash_table[hash].chain, rt);
 346
 347         dst_hold_and_use(&rt->dst, now);
 348         spin_unlock_bh(&dn_rt_hash_table[hash].lock);
 349         *rp = rt;
 350         return 0;
 351 }
 352
 353 static void dn_run_flush(struct timer_list *unused)
 354 {
 355         int i;
 356         struct dn_route *rt, *next;
 357
 358         for (i = 0; i < dn_rt_hash_mask; i++) {
 359                 spin_lock_bh(&dn_rt_hash_table[i].lock);
 360
 361                 rt = xchg((struct dn_route **)&dn_rt_hash_table[i].chain, NULL);
 362                 if (!rt)
 363                         goto nothing_to_declare;
 364
 365                 for (; rt; rt = next) {
 366                         next = rcu_dereference_raw(rt->dn_next);
 367                         RCU_INIT_POINTER(rt->dn_next, NULL);
 368                         dst_dev_put(&rt->dst);
 369                         dst_release(&rt->dst);
 370                 }
 371
 372 nothing_to_declare:
 373                 spin_unlock_bh(&dn_rt_hash_table[i].lock);
 374         }
 375 }
 376
 377 static DEFINE_SPINLOCK(dn_rt_flush_lock);
 378
 379 void dn_rt_cache_flush(int delay)
 380 {
 381         unsigned long now = jiffies;
 382         int user_mode = !in_interrupt();
 383
 384         if (delay < 0)
 385                 delay = dn_rt_min_delay;
 386
 387         spin_lock_bh(&dn_rt_flush_lock);
 388
 389         if (del_timer(&dn_rt_flush_timer) && delay > 0 && dn_rt_deadline) {
 390                 long tmo = (long)(dn_rt_deadline - now);
 391
 392                 if (user_mode && tmo < dn_rt_max_delay - dn_rt_min_delay)
 393                         tmo = 0;
 394
 395                 if (delay > tmo)
 396                         delay = tmo;
 397         }
 398
 399         if (delay <= 0) {
 400                 spin_unlock_bh(&dn_rt_flush_lock);
 401                 dn_run_flush(NULL);
 402                 return;
 403         }
 404
 405         if (dn_rt_deadline == 0)
 406                 dn_rt_deadline = now + dn_rt_max_delay;
 407
 408         dn_rt_flush_timer.expires = now + delay;
 409         add_timer(&dn_rt_flush_timer);
 410         spin_unlock_bh(&dn_rt_flush_lock);
 411 }
 412
 413 /**
 414  * dn_return_short - Return a short packet to its sender
 415  * @skb: The packet to return
 416  *
 417  */
 418 static int dn_return_short(struct sk_buff *skb)
 419 {
 420         struct dn_skb_cb *cb;
 421         unsigned char *ptr;
 422         __le16 *src;
 423         __le16 *dst;
 424
 425         /* Add back headers */
 426         skb_push(skb, skb->data - skb_network_header(skb));
 427
 428         skb = skb_unshare(skb, GFP_ATOMIC);
 429         if (!skb)
 430                 return NET_RX_DROP;
 431
 432         cb = DN_SKB_CB(skb);
 433         /* Skip packet length and point to flags */
 434         ptr = skb->data + 2;
 435         *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
 436
 437         dst = (__le16 *)ptr;
 438         ptr += 2;
 439         src = (__le16 *)ptr;
 440         ptr += 2;
 441         *ptr = 0; /* Zero hop count */
 442
 443         swap(*src, *dst);
 444
 445         skb->pkt_type = PACKET_OUTGOING;
 446         dn_rt_finish_output(skb, NULL, NULL);
 447         return NET_RX_SUCCESS;
 448 }
 449
 450 /**
 451  * dn_return_long - Return a long packet to its sender
 452  * @skb: The long format packet to return
 453  *
 454  */
 455 static int dn_return_long(struct sk_buff *skb)
 456 {
 457         struct dn_skb_cb *cb;
 458         unsigned char *ptr;
 459         unsigned char *src_addr, *dst_addr;
 460         unsigned char tmp[ETH_ALEN];
 461
 462         /* Add back all headers */
 463         skb_push(skb, skb->data - skb_network_header(skb));
 464
 465         skb = skb_unshare(skb, GFP_ATOMIC);
 466         if (!skb)
 467                 return NET_RX_DROP;
 468
 469         cb = DN_SKB_CB(skb);
 470         /* Ignore packet length and point to flags */
 471         ptr = skb->data + 2;
 472
 473         /* Skip padding */
 474         if (*ptr & DN_RT_F_PF) {
 475                 char padlen = (*ptr & ~DN_RT_F_PF);
 476                 ptr += padlen;
 477         }
 478
 479         *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
 480         ptr += 2;
 481         dst_addr = ptr;
 482         ptr += 8;
 483         src_addr = ptr;
 484         ptr += 6;
 485         *ptr = 0; /* Zero hop count */
 486
 487         /* Swap source and destination */
 488         memcpy(tmp, src_addr, ETH_ALEN);
 489         memcpy(src_addr, dst_addr, ETH_ALEN);
 490         memcpy(dst_addr, tmp, ETH_ALEN);
 491
 492         skb->pkt_type = PACKET_OUTGOING;
 493         dn_rt_finish_output(skb, dst_addr, src_addr);
 494         return NET_RX_SUCCESS;
 495 }
 496
 497 /**
 498  * dn_route_rx_packet - Try and find a route for an incoming packet
 499  * @net: The applicable net namespace
 500  * @sk: Socket packet transmitted on
 501  * @skb: The packet to find a route for
 502  *
 503  * Returns: result of input function if route is found, error code otherwise
 504  */
 505 static int dn_route_rx_packet(struct net *net, struct sock *sk, struct sk_buff *skb)
 506 {
 507         struct dn_skb_cb *cb;
 508         int err;
 509
 510         err = dn_route_input(skb);
 511         if (err == 0)
 512                 return dst_input(skb);
 513
 514         cb = DN_SKB_CB(skb);
 515         if (decnet_debug_level & 4) {
 516                 char *devname = skb->dev ? skb->dev->name : "???";
 517
 518                 printk(KERN_DEBUG
 519                         "DECnet: dn_route_rx_packet: rt_flags=0x%02x dev=%s len=%d src=0x%04hx dst=0x%04hx err=%d type=%d\n",
 520                         (int)cb->rt_flags, devname, skb->len,
 521                         le16_to_cpu(cb->src), le16_to_cpu(cb->dst),
 522                         err, skb->pkt_type);
 523         }
 524
 525         if ((skb->pkt_type == PACKET_HOST) && (cb->rt_flags & DN_RT_F_RQR)) {
 526                 switch (cb->rt_flags & DN_RT_PKT_MSK) {
 527                 case DN_RT_PKT_SHORT:
 528                         return dn_return_short(skb);
 529                 case DN_RT_PKT_LONG:
 530                         return dn_return_long(skb);
 531                 }
 532         }
 533
 534         kfree_skb(skb);
 535         return NET_RX_DROP;
 536 }
 537
 538 static int dn_route_rx_long(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, 21)) /* 20 for long header, 1 for shortest nsp */
 544                 goto drop_it;
 545
 546         skb_pull(skb, 20);
 547         skb_reset_transport_header(skb);
 548
 549         /* Destination info */
 550         ptr += 2;
 551         cb->dst = dn_eth2dn(ptr);
 552         if (memcmp(ptr, dn_hiord_addr, 4) != 0)
 553                 goto drop_it;
 554         ptr += 6;
 555
 556
 557         /* Source info */
 558         ptr += 2;
 559         cb->src = dn_eth2dn(ptr);
 560         if (memcmp(ptr, dn_hiord_addr, 4) != 0)
 561                 goto drop_it;
 562         ptr += 6;
 563         /* Other junk */
 564         ptr++;
 565         cb->hops = *ptr++; /* Visit Count */
 566
 567         return NF_HOOK(NFPROTO_DECNET, NF_DN_PRE_ROUTING,
 568                        &init_net, NULL, skb, skb->dev, NULL,
 569                        dn_route_rx_packet);
 570
 571 drop_it:
 572         kfree_skb(skb);
 573         return NET_RX_DROP;
 574 }
 575
 576
 577
 578 static int dn_route_rx_short(struct sk_buff *skb)
 579 {
 580         struct dn_skb_cb *cb = DN_SKB_CB(skb);
 581         unsigned char *ptr = skb->data;
 582
 583         if (!pskb_may_pull(skb, 6)) /* 5 for short header + 1 for shortest nsp */
 584                 goto drop_it;
 585
 586         skb_pull(skb, 5);
 587         skb_reset_transport_header(skb);
 588
 589         cb->dst = *(__le16 *)ptr;
 590         ptr += 2;
 591         cb->src = *(__le16 *)ptr;
 592         ptr += 2;
 593         cb->hops = *ptr & 0x3f;
 594
 595         return NF_HOOK(NFPROTO_DECNET, NF_DN_PRE_ROUTING,
 596                        &init_net, NULL, skb, skb->dev, NULL,
 597                        dn_route_rx_packet);
 598
 599 drop_it:
 600         kfree_skb(skb);
 601         return NET_RX_DROP;
 602 }
 603
 604 static int dn_route_discard(struct net *net, struct sock *sk, struct sk_buff *skb)
 605 {
 606         /*
 607          * I know we drop the packet here, but that's considered success in
 608          * this case
 609          */
 610         kfree_skb(skb);
 611         return NET_RX_SUCCESS;
 612 }
 613
 614 static int dn_route_ptp_hello(struct net *net, struct sock *sk, struct sk_buff *skb)
 615 {
 616         dn_dev_hello(skb);
 617         dn_neigh_pointopoint_hello(skb);
 618         return NET_RX_SUCCESS;
 619 }
 620
 621 int dn_route_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
 622 {
 623         struct dn_skb_cb *cb;
 624         unsigned char flags = 0;
 625         __u16 len = le16_to_cpu(*(__le16 *)skb->data);
 626         struct dn_dev *dn = rcu_dereference(dev->dn_ptr);
 627         unsigned char padlen = 0;
 628
 629         if (!net_eq(dev_net(dev), &init_net))
 630                 goto dump_it;
 631
 632         if (dn == NULL)
 633                 goto dump_it;
 634
 635         skb = skb_share_check(skb, GFP_ATOMIC);
 636         if (!skb)
 637                 goto out;
 638
 639         if (!pskb_may_pull(skb, 3))
 640                 goto dump_it;
 641
 642         skb_pull(skb, 2);
 643
 644         if (len > skb->len)
 645                 goto dump_it;
 646
 647         skb_trim(skb, len);
 648
 649         flags = *skb->data;
 650
 651         cb = DN_SKB_CB(skb);
 652         cb->stamp = jiffies;
 653         cb->iif = dev->ifindex;
 654
 655         /*
 656          * If we have padding, remove it.
 657          */
 658         if (flags & DN_RT_F_PF) {
 659                 padlen = flags & ~DN_RT_F_PF;
 660                 if (!pskb_may_pull(skb, padlen + 1))
 661                         goto dump_it;
 662                 skb_pull(skb, padlen);
 663                 flags = *skb->data;
 664         }
 665
 666         skb_reset_network_header(skb);
 667
 668         /*
 669          * Weed out future version DECnet
 670          */
 671         if (flags & DN_RT_F_VER)
 672                 goto dump_it;
 673
 674         cb->rt_flags = flags;
 675
 676         if (decnet_debug_level & 1)
 677                 printk(KERN_DEBUG
 678                         "dn_route_rcv: got 0x%02x from %s [%d %d %d]\n",
 679                         (int)flags, dev->name, len, skb->len,
 680                         padlen);
 681
 682         if (flags & DN_RT_PKT_CNTL) {
 683                 if (unlikely(skb_linearize(skb)))
 684                         goto dump_it;
 685
 686                 switch (flags & DN_RT_CNTL_MSK) {
 687                 case DN_RT_PKT_INIT:
 688                         dn_dev_init_pkt(skb);
 689                         break;
 690                 case DN_RT_PKT_VERI:
 691                         dn_dev_veri_pkt(skb);
 692                         break;
 693                 }
 694
 695                 if (dn->parms.state != DN_DEV_S_RU)
 696                         goto dump_it;
 697
 698                 switch (flags & DN_RT_CNTL_MSK) {
 699                 case DN_RT_PKT_HELO:
 700                         return NF_HOOK(NFPROTO_DECNET, NF_DN_HELLO,
 701                                        &init_net, NULL, skb, skb->dev, NULL,
 702                                        dn_route_ptp_hello);
 703
 704                 case DN_RT_PKT_L1RT:
 705                 case DN_RT_PKT_L2RT:
 706                         return NF_HOOK(NFPROTO_DECNET, NF_DN_ROUTE,
 707                                        &init_net, NULL, skb, skb->dev, NULL,
 708                                        dn_route_discard);
 709                 case DN_RT_PKT_ERTH:
 710                         return NF_HOOK(NFPROTO_DECNET, NF_DN_HELLO,
 711                                        &init_net, NULL, skb, skb->dev, NULL,
 712                                        dn_neigh_router_hello);
 713
 714                 case DN_RT_PKT_EEDH:
 715                         return NF_HOOK(NFPROTO_DECNET, NF_DN_HELLO,
 716                                        &init_net, NULL, skb, skb->dev, NULL,
 717                                        dn_neigh_endnode_hello);
 718                 }
 719         } else {
 720                 if (dn->parms.state != DN_DEV_S_RU)
 721                         goto dump_it;
 722
 723                 skb_pull(skb, 1); /* Pull flags */
 724
 725                 switch (flags & DN_RT_PKT_MSK) {
 726                 case DN_RT_PKT_LONG:
 727                         return dn_route_rx_long(skb);
 728                 case DN_RT_PKT_SHORT:
 729                         return dn_route_rx_short(skb);
 730                 }
 731         }
 732
 733 dump_it:
 734         kfree_skb(skb);
 735 out:
 736         return NET_RX_DROP;
 737 }
 738
 739 static int dn_output(struct net *net, struct sock *sk, struct sk_buff *skb)
 740 {
 741         struct dst_entry *dst = skb_dst(skb);
 742         struct dn_route *rt = (struct dn_route *)dst;
 743         struct net_device *dev = dst->dev;
 744         struct dn_skb_cb *cb = DN_SKB_CB(skb);
 745
 746         int err = -EINVAL;
 747
 748         if (rt->n == NULL)
 749                 goto error;
 750
 751         skb->dev = dev;
 752
 753         cb->src = rt->rt_saddr;
 754         cb->dst = rt->rt_daddr;
 755
 756         /*
 757          * Always set the Intra-Ethernet bit on all outgoing packets
 758          * originated on this node. Only valid flag from upper layers
 759          * is return-to-sender-requested. Set hop count to 0 too.
 760          */
 761         cb->rt_flags &= ~DN_RT_F_RQR;
 762         cb->rt_flags |= DN_RT_F_IE;
 763         cb->hops = 0;
 764
 765         return NF_HOOK(NFPROTO_DECNET, NF_DN_LOCAL_OUT,
 766                        &init_net, sk, skb, NULL, dev,
 767                        dn_to_neigh_output);
 768
 769 error:
 770         net_dbg_ratelimited("dn_output: This should not happen\n");
 771
 772         kfree_skb(skb);
 773
 774         return err;
 775 }
 776
 777 static int dn_forward(struct sk_buff *skb)
 778 {
 779         struct dn_skb_cb *cb = DN_SKB_CB(skb);
 780         struct dst_entry *dst = skb_dst(skb);
 781         struct dn_dev *dn_db = rcu_dereference(dst->dev->dn_ptr);
 782         struct dn_route *rt;
 783         int header_len;
 784         struct net_device *dev = skb->dev;
 785
 786         if (skb->pkt_type != PACKET_HOST)
 787                 goto drop;
 788
 789         /* Ensure that we have enough space for headers */
 790         rt = (struct dn_route *)skb_dst(skb);
 791         header_len = dn_db->use_long ? 21 : 6;
 792         if (skb_cow(skb, LL_RESERVED_SPACE(rt->dst.dev)+header_len))
 793                 goto drop;
 794
 795         /*
 796          * Hop count exceeded.
 797          */
 798         if (++cb->hops > 30)
 799                 goto drop;
 800
 801         skb->dev = rt->dst.dev;
 802
 803         /*
 804          * If packet goes out same interface it came in on, then set
 805          * the Intra-Ethernet bit. This has no effect for short
 806          * packets, so we don't need to test for them here.
 807          */
 808         cb->rt_flags &= ~DN_RT_F_IE;
 809         if (rt->rt_flags & RTCF_DOREDIRECT)
 810                 cb->rt_flags |= DN_RT_F_IE;
 811
 812         return NF_HOOK(NFPROTO_DECNET, NF_DN_FORWARD,
 813                        &init_net, NULL, skb, dev, skb->dev,
 814                        dn_to_neigh_output);
 815
 816 drop:
 817         kfree_skb(skb);
 818         return NET_RX_DROP;
 819 }
 820
 821 /*
 822  * Used to catch bugs. This should never normally get
 823  * called.
 824  */
 825 static int dn_rt_bug_out(struct net *net, struct sock *sk, struct sk_buff *skb)
 826 {
 827         struct dn_skb_cb *cb = DN_SKB_CB(skb);
 828
 829         net_dbg_ratelimited("dn_rt_bug: skb from:%04x to:%04x\n",
 830                             le16_to_cpu(cb->src), le16_to_cpu(cb->dst));
 831
 832         kfree_skb(skb);
 833
 834         return NET_RX_DROP;
 835 }
 836
 837 static int dn_rt_bug(struct sk_buff *skb)
 838 {
 839         struct dn_skb_cb *cb = DN_SKB_CB(skb);
 840
 841         net_dbg_ratelimited("dn_rt_bug: skb from:%04x to:%04x\n",
 842                             le16_to_cpu(cb->src), le16_to_cpu(cb->dst));
 843
 844         kfree_skb(skb);
 845
 846         return NET_RX_DROP;
 847 }
 848
 849 static unsigned int dn_dst_default_advmss(const struct dst_entry *dst)
 850 {
 851         return dn_mss_from_pmtu(dst->dev, dst_mtu(dst));
 852 }
 853
 854 static unsigned int dn_dst_mtu(const struct dst_entry *dst)
 855 {
 856         unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
 857
 858         return mtu ? : dst->dev->mtu;
 859 }
 860
 861 static struct neighbour *dn_dst_neigh_lookup(const struct dst_entry *dst,
 862                                              struct sk_buff *skb,
 863                                              const void *daddr)
 864 {
 865         return __neigh_lookup_errno(&dn_neigh_table, daddr, dst->dev);
 866 }
 867
 868 static int dn_rt_set_next_hop(struct dn_route *rt, struct dn_fib_res *res)
 869 {
 870         struct dn_fib_info *fi = res->fi;
 871         struct net_device *dev = rt->dst.dev;
 872         unsigned int mss_metric;
 873         struct neighbour *n;
 874
 875         if (fi) {
 876                 if (DN_FIB_RES_GW(*res) &&
 877                     DN_FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
 878                         rt->rt_gateway = DN_FIB_RES_GW(*res);
 879                 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
 880         }
 881         rt->rt_type = res->type;
 882
 883         if (dev != NULL && rt->n == NULL) {
 884                 n = __neigh_lookup_errno(&dn_neigh_table, &rt->rt_gateway, dev);
 885                 if (IS_ERR(n))
 886                         return PTR_ERR(n);
 887                 rt->n = n;
 888         }
 889
 890         if (dst_metric(&rt->dst, RTAX_MTU) > rt->dst.dev->mtu)
 891                 dst_metric_set(&rt->dst, RTAX_MTU, rt->dst.dev->mtu);
 892         mss_metric = dst_metric_raw(&rt->dst, RTAX_ADVMSS);
 893         if (mss_metric) {
 894                 unsigned int mss = dn_mss_from_pmtu(dev, dst_mtu(&rt->dst));
 895                 if (mss_metric > mss)
 896                         dst_metric_set(&rt->dst, RTAX_ADVMSS, mss);
 897         }
 898         return 0;
 899 }
 900
 901 static inline int dn_match_addr(__le16 addr1, __le16 addr2)
 902 {
 903         __u16 tmp = le16_to_cpu(addr1) ^ le16_to_cpu(addr2);
 904         int match = 16;
 905         while (tmp) {
 906                 tmp >>= 1;
 907                 match--;
 908         }
 909         return match;
 910 }
 911
 912 static __le16 dnet_select_source(const struct net_device *dev, __le16 daddr, int scope)
 913 {
 914         __le16 saddr = 0;
 915         struct dn_dev *dn_db;
 916         struct dn_ifaddr *ifa;
 917         int best_match = 0;
 918         int ret;
 919
 920         rcu_read_lock();
 921         dn_db = rcu_dereference(dev->dn_ptr);
 922         for (ifa = rcu_dereference(dn_db->ifa_list);
 923              ifa != NULL;
 924              ifa = rcu_dereference(ifa->ifa_next)) {
 925                 if (ifa->ifa_scope > scope)
 926                         continue;
 927                 if (!daddr) {
 928                         saddr = ifa->ifa_local;
 929                         break;
 930                 }
 931                 ret = dn_match_addr(daddr, ifa->ifa_local);
 932                 if (ret > best_match)
 933                         saddr = ifa->ifa_local;
 934                 if (best_match == 0)
 935                         saddr = ifa->ifa_local;
 936         }
 937         rcu_read_unlock();
 938
 939         return saddr;
 940 }
 941
 942 static inline __le16 __dn_fib_res_prefsrc(struct dn_fib_res *res)
 943 {
 944         return dnet_select_source(DN_FIB_RES_DEV(*res), DN_FIB_RES_GW(*res), res->scope);
 945 }
 946
 947 static inline __le16 dn_fib_rules_map_destination(__le16 daddr, struct dn_fib_res *res)
 948 {
 949         __le16 mask = dnet_make_mask(res->prefixlen);
 950         return (daddr&~mask)|res->fi->fib_nh->nh_gw;
 951 }
 952
 953 static int dn_route_output_slow(struct dst_entry **pprt, const struct flowidn *oldflp, int try_hard)
 954 {
 955         struct flowidn fld = {
 956                 .daddr = oldflp->daddr,
 957                 .saddr = oldflp->saddr,
 958                 .flowidn_scope = RT_SCOPE_UNIVERSE,
 959                 .flowidn_mark = oldflp->flowidn_mark,
 960                 .flowidn_iif = LOOPBACK_IFINDEX,
 961                 .flowidn_oif = oldflp->flowidn_oif,
 962         };
 963         struct dn_route *rt = NULL;
 964         struct net_device *dev_out = NULL, *dev;
 965         struct neighbour *neigh = NULL;
 966         unsigned int hash;
 967         unsigned int flags = 0;
 968         struct dn_fib_res res = { .fi = NULL, .type = RTN_UNICAST };
 969         int err;
 970         int free_res = 0;
 971         __le16 gateway = 0;
 972
 973         if (decnet_debug_level & 16)
 974                 printk(KERN_DEBUG
 975                        "dn_route_output_slow: dst=%04x src=%04x mark=%d"
 976                        " iif=%d oif=%d\n", le16_to_cpu(oldflp->daddr),
 977                        le16_to_cpu(oldflp->saddr),
 978                        oldflp->flowidn_mark, LOOPBACK_IFINDEX,
 979                        oldflp->flowidn_oif);
 980
 981         /* If we have an output interface, verify its a DECnet device */
 982         if (oldflp->flowidn_oif) {
 983                 dev_out = dev_get_by_index(&init_net, oldflp->flowidn_oif);
 984                 err = -ENODEV;
 985                 if (dev_out && dev_out->dn_ptr == NULL) {
 986                         dev_put(dev_out);
 987                         dev_out = NULL;
 988                 }
 989                 if (dev_out == NULL)
 990                         goto out;
 991         }
 992
 993         /* If we have a source address, verify that its a local address */
 994         if (oldflp->saddr) {
 995                 err = -EADDRNOTAVAIL;
 996
 997                 if (dev_out) {
 998                         if (dn_dev_islocal(dev_out, oldflp->saddr))
 999                                 goto source_ok;
1000                         dev_put(dev_out);
1001                         goto out;
1002                 }
1003                 rcu_read_lock();
1004                 for_each_netdev_rcu(&init_net, dev) {
1005                         if (!dev->dn_ptr)
1006                                 continue;
1007                         if (!dn_dev_islocal(dev, oldflp->saddr))
1008                                 continue;
1009                         if ((dev->flags & IFF_LOOPBACK) &&
1010                             oldflp->daddr &&
1011                             !dn_dev_islocal(dev, oldflp->daddr))
1012                                 continue;
1013
1014                         dev_out = dev;
1015                         break;
1016                 }
1017                 rcu_read_unlock();
1018                 if (dev_out == NULL)
1019                         goto out;
1020                 dev_hold(dev_out);
1021 source_ok:
1022                 ;
1023         }
1024
1025         /* No destination? Assume its local */
1026         if (!fld.daddr) {
1027                 fld.daddr = fld.saddr;
1028
1029                 if (dev_out)
1030                         dev_put(dev_out);
1031                 err = -EINVAL;
1032                 dev_out = init_net.loopback_dev;
1033                 if (!dev_out->dn_ptr)
1034                         goto out;
1035                 err = -EADDRNOTAVAIL;
1036                 dev_hold(dev_out);
1037                 if (!fld.daddr) {
1038                         fld.daddr =
1039                         fld.saddr = dnet_select_source(dev_out, 0,
1040                                                        RT_SCOPE_HOST);
1041                         if (!fld.daddr)
1042                                 goto done;
1043                 }
1044                 fld.flowidn_oif = LOOPBACK_IFINDEX;
1045                 res.type = RTN_LOCAL;
1046                 goto make_route;
1047         }
1048
1049         if (decnet_debug_level & 16)
1050                 printk(KERN_DEBUG
1051                        "dn_route_output_slow: initial checks complete."
1052                        " dst=%04x src=%04x oif=%d try_hard=%d\n",
1053                        le16_to_cpu(fld.daddr), le16_to_cpu(fld.saddr),
1054                        fld.flowidn_oif, try_hard);
1055
1056         /*
1057          * N.B. If the kernel is compiled without router support then
1058          * dn_fib_lookup() will evaluate to non-zero so this if () block
1059          * will always be executed.
1060          */
1061         err = -ESRCH;
1062         if (try_hard || (err = dn_fib_lookup(&fld, &res)) != 0) {
1063                 struct dn_dev *dn_db;
1064                 if (err != -ESRCH)
1065                         goto out;
1066                 /*
1067                  * Here the fallback is basically the standard algorithm for
1068                  * routing in endnodes which is described in the DECnet routing
1069                  * docs
1070                  *
1071                  * If we are not trying hard, look in neighbour cache.
1072                  * The result is tested to ensure that if a specific output
1073                  * device/source address was requested, then we honour that
1074                  * here
1075                  */
1076                 if (!try_hard) {
1077                         neigh = neigh_lookup_nodev(&dn_neigh_table, &init_net, &fld.daddr);
1078                         if (neigh) {
1079                                 if ((oldflp->flowidn_oif &&
1080                                     (neigh->dev->ifindex != oldflp->flowidn_oif)) ||
1081                                     (oldflp->saddr &&
1082                                     (!dn_dev_islocal(neigh->dev,
1083                                                      oldflp->saddr)))) {
1084                                         neigh_release(neigh);
1085                                         neigh = NULL;
1086                                 } else {
1087                                         if (dev_out)
1088                                                 dev_put(dev_out);
1089                                         if (dn_dev_islocal(neigh->dev, fld.daddr)) {
1090                                                 dev_out = init_net.loopback_dev;
1091                                                 res.type = RTN_LOCAL;
1092                                         } else {
1093                                                 dev_out = neigh->dev;
1094                                         }
1095                                         dev_hold(dev_out);
1096                                         goto select_source;
1097                                 }
1098                         }
1099                 }
1100
1101                 /* Not there? Perhaps its a local address */
1102                 if (dev_out == NULL)
1103                         dev_out = dn_dev_get_default();
1104                 err = -ENODEV;
1105                 if (dev_out == NULL)
1106                         goto out;
1107                 dn_db = rcu_dereference_raw(dev_out->dn_ptr);
1108                 if (!dn_db)
1109                         goto e_inval;
1110                 /* Possible improvement - check all devices for local addr */
1111                 if (dn_dev_islocal(dev_out, fld.daddr)) {
1112                         dev_put(dev_out);
1113                         dev_out = init_net.loopback_dev;
1114                         dev_hold(dev_out);
1115                         res.type = RTN_LOCAL;
1116                         goto select_source;
1117                 }
1118                 /* Not local either.... try sending it to the default router */
1119                 neigh = neigh_clone(dn_db->router);
1120                 BUG_ON(neigh && neigh->dev != dev_out);
1121
1122                 /* Ok then, we assume its directly connected and move on */
1123 select_source:
1124                 if (neigh)
1125                         gateway = ((struct dn_neigh *)neigh)->addr;
1126                 if (gateway == 0)
1127                         gateway = fld.daddr;
1128                 if (fld.saddr == 0) {
1129                         fld.saddr = dnet_select_source(dev_out, gateway,
1130                                                        res.type == RTN_LOCAL ?
1131                                                        RT_SCOPE_HOST :
1132                                                        RT_SCOPE_LINK);
1133                         if (fld.saddr == 0 && res.type != RTN_LOCAL)
1134                                 goto e_addr;
1135                 }
1136                 fld.flowidn_oif = dev_out->ifindex;
1137                 goto make_route;
1138         }
1139         free_res = 1;
1140
1141         if (res.type == RTN_NAT)
1142                 goto e_inval;
1143
1144         if (res.type == RTN_LOCAL) {
1145                 if (!fld.saddr)
1146                         fld.saddr = fld.daddr;
1147                 if (dev_out)
1148                         dev_put(dev_out);
1149                 dev_out = init_net.loopback_dev;
1150                 dev_hold(dev_out);
1151                 if (!dev_out->dn_ptr)
1152                         goto e_inval;
1153                 fld.flowidn_oif = dev_out->ifindex;
1154                 if (res.fi)
1155                         dn_fib_info_put(res.fi);
1156                 res.fi = NULL;
1157                 goto make_route;
1158         }
1159
1160         if (res.fi->fib_nhs > 1 && fld.flowidn_oif == 0)
1161                 dn_fib_select_multipath(&fld, &res);
1162
1163         /*
1164          * We could add some logic to deal with default routes here and
1165          * get rid of some of the special casing above.
1166          */
1167
1168         if (!fld.saddr)
1169                 fld.saddr = DN_FIB_RES_PREFSRC(res);
1170
1171         if (dev_out)
1172                 dev_put(dev_out);
1173         dev_out = DN_FIB_RES_DEV(res);
1174         dev_hold(dev_out);
1175         fld.flowidn_oif = dev_out->ifindex;
1176         gateway = DN_FIB_RES_GW(res);
1177
1178 make_route:
1179         if (dev_out->flags & IFF_LOOPBACK)
1180                 flags |= RTCF_LOCAL;
1181
1182         rt = dst_alloc(&dn_dst_ops, dev_out, 0, DST_OBSOLETE_NONE, 0);
1183         if (rt == NULL)
1184                 goto e_nobufs;
1185
1186         rt->dn_next = NULL;
1187         memset(&rt->fld, 0, sizeof(rt->fld));
1188         rt->fld.saddr        = oldflp->saddr;
1189         rt->fld.daddr        = oldflp->daddr;
1190         rt->fld.flowidn_oif  = oldflp->flowidn_oif;
1191         rt->fld.flowidn_iif  = 0;
1192         rt->fld.flowidn_mark = oldflp->flowidn_mark;
1193
1194         rt->rt_saddr      = fld.saddr;
1195         rt->rt_daddr      = fld.daddr;
1196         rt->rt_gateway    = gateway ? gateway : fld.daddr;
1197         rt->rt_local_src  = fld.saddr;
1198
1199         rt->rt_dst_map    = fld.daddr;
1200         rt->rt_src_map    = fld.saddr;
1201
1202         rt->n = neigh;
1203         neigh = NULL;
1204
1205         rt->dst.lastuse = jiffies;
1206         rt->dst.output  = dn_output;
1207         rt->dst.input   = dn_rt_bug;
1208         rt->rt_flags      = flags;
1209         if (flags & RTCF_LOCAL)
1210                 rt->dst.input = dn_nsp_rx;
1211
1212         err = dn_rt_set_next_hop(rt, &res);
1213         if (err)
1214                 goto e_neighbour;
1215
1216         hash = dn_hash(rt->fld.saddr, rt->fld.daddr);
1217         /* dn_insert_route() increments dst->__refcnt */
1218         dn_insert_route(rt, hash, (struct dn_route **)pprt);
1219
1220 done:
1221         if (neigh)
1222                 neigh_release(neigh);
1223         if (free_res)
1224                 dn_fib_res_put(&res);
1225         if (dev_out)
1226                 dev_put(dev_out);
1227 out:
1228         return err;
1229
1230 e_addr:
1231         err = -EADDRNOTAVAIL;
1232         goto done;
1233 e_inval:
1234         err = -EINVAL;
1235         goto done;
1236 e_nobufs:
1237         err = -ENOBUFS;
1238         goto done;
1239 e_neighbour:
1240         dst_release_immediate(&rt->dst);
1241         goto e_nobufs;
1242 }
1243
1244
1245 /*
1246  * N.B. The flags may be moved into the flowi at some future stage.
1247  */
1248 static int __dn_route_output_key(struct dst_entry **pprt, const struct flowidn *flp, int flags)
1249 {
1250         unsigned int hash = dn_hash(flp->saddr, flp->daddr);
1251         struct dn_route *rt = NULL;
1252
1253         if (!(flags & MSG_TRYHARD)) {
1254                 rcu_read_lock_bh();
1255                 for (rt = rcu_dereference_bh(dn_rt_hash_table[hash].chain); rt;
1256                         rt = rcu_dereference_bh(rt->dn_next)) {
1257                         if ((flp->daddr == rt->fld.daddr) &&
1258                             (flp->saddr == rt->fld.saddr) &&
1259                             (flp->flowidn_mark == rt->fld.flowidn_mark) &&
1260                             dn_is_output_route(rt) &&
1261                             (rt->fld.flowidn_oif == flp->flowidn_oif)) {
1262                                 dst_hold_and_use(&rt->dst, jiffies);
1263                                 rcu_read_unlock_bh();
1264                                 *pprt = &rt->dst;
1265                                 return 0;
1266                         }
1267                 }
1268                 rcu_read_unlock_bh();
1269         }
1270
1271         return dn_route_output_slow(pprt, flp, flags);
1272 }
1273
1274 static int dn_route_output_key(struct dst_entry **pprt, struct flowidn *flp, int flags)
1275 {
1276         int err;
1277
1278         err = __dn_route_output_key(pprt, flp, flags);
1279         if (err == 0 && flp->flowidn_proto) {
1280                 *pprt = xfrm_lookup(&init_net, *pprt,
1281                                     flowidn_to_flowi(flp), NULL, 0);
1282                 if (IS_ERR(*pprt)) {
1283                         err = PTR_ERR(*pprt);
1284                         *pprt = NULL;
1285                 }
1286         }
1287         return err;
1288 }
1289
1290 int dn_route_output_sock(struct dst_entry __rcu **pprt, struct flowidn *fl, struct sock *sk, int flags)
1291 {
1292         int err;
1293
1294         err = __dn_route_output_key(pprt, fl, flags & MSG_TRYHARD);
1295         if (err == 0 && fl->flowidn_proto) {
1296                 *pprt = xfrm_lookup(&init_net, *pprt,
1297                                     flowidn_to_flowi(fl), sk, 0);
1298                 if (IS_ERR(*pprt)) {
1299                         err = PTR_ERR(*pprt);
1300                         *pprt = NULL;
1301                 }
1302         }
1303         return err;
1304 }
1305
1306 static int dn_route_input_slow(struct sk_buff *skb)
1307 {
1308         struct dn_route *rt = NULL;
1309         struct dn_skb_cb *cb = DN_SKB_CB(skb);
1310         struct net_device *in_dev = skb->dev;
1311         struct net_device *out_dev = NULL;
1312         struct dn_dev *dn_db;
1313         struct neighbour *neigh = NULL;
1314         unsigned int hash;
1315         int flags = 0;
1316         __le16 gateway = 0;
1317         __le16 local_src = 0;
1318         struct flowidn fld = {
1319                 .daddr = cb->dst,
1320                 .saddr = cb->src,
1321                 .flowidn_scope = RT_SCOPE_UNIVERSE,
1322                 .flowidn_mark = skb->mark,
1323                 .flowidn_iif = skb->dev->ifindex,
1324         };
1325         struct dn_fib_res res = { .fi = NULL, .type = RTN_UNREACHABLE };
1326         int err = -EINVAL;
1327         int free_res = 0;
1328
1329         dev_hold(in_dev);
1330
1331         dn_db = rcu_dereference(in_dev->dn_ptr);
1332         if (!dn_db)
1333                 goto out;
1334
1335         /* Zero source addresses are not allowed */
1336         if (fld.saddr == 0)
1337                 goto out;
1338
1339         /*
1340          * In this case we've just received a packet from a source
1341          * outside ourselves pretending to come from us. We don't
1342          * allow it any further to prevent routing loops, spoofing and
1343          * other nasties. Loopback packets already have the dst attached
1344          * so this only affects packets which have originated elsewhere.
1345          */
1346         err  = -ENOTUNIQ;
1347         if (dn_dev_islocal(in_dev, cb->src))
1348                 goto out;
1349
1350         err = dn_fib_lookup(&fld, &res);
1351         if (err) {
1352                 if (err != -ESRCH)
1353                         goto out;
1354                 /*
1355                  * Is the destination us ?
1356                  */
1357                 if (!dn_dev_islocal(in_dev, cb->dst))
1358                         goto e_inval;
1359
1360                 res.type = RTN_LOCAL;
1361         } else {
1362                 __le16 src_map = fld.saddr;
1363                 free_res = 1;
1364
1365                 out_dev = DN_FIB_RES_DEV(res);
1366                 if (out_dev == NULL) {
1367                         net_crit_ratelimited("Bug in dn_route_input_slow() No output device\n");
1368                         goto e_inval;
1369                 }
1370                 dev_hold(out_dev);
1371
1372                 if (res.r)
1373                         src_map = fld.saddr; /* no NAT support for now */
1374
1375                 gateway = DN_FIB_RES_GW(res);
1376                 if (res.type == RTN_NAT) {
1377                         fld.daddr = dn_fib_rules_map_destination(fld.daddr, &res);
1378                         dn_fib_res_put(&res);
1379                         free_res = 0;
1380                         if (dn_fib_lookup(&fld, &res))
1381                                 goto e_inval;
1382                         free_res = 1;
1383                         if (res.type != RTN_UNICAST)
1384                                 goto e_inval;
1385                         flags |= RTCF_DNAT;
1386                         gateway = fld.daddr;
1387                 }
1388                 fld.saddr = src_map;
1389         }
1390
1391         switch (res.type) {
1392         case RTN_UNICAST:
1393                 /*
1394                  * Forwarding check here, we only check for forwarding
1395                  * being turned off, if you want to only forward intra
1396                  * area, its up to you to set the routing tables up
1397                  * correctly.
1398                  */
1399                 if (dn_db->parms.forwarding == 0)
1400                         goto e_inval;
1401
1402                 if (res.fi->fib_nhs > 1 && fld.flowidn_oif == 0)
1403                         dn_fib_select_multipath(&fld, &res);
1404
1405                 /*
1406                  * Check for out_dev == in_dev. We use the RTCF_DOREDIRECT
1407                  * flag as a hint to set the intra-ethernet bit when
1408                  * forwarding. If we've got NAT in operation, we don't do
1409                  * this optimisation.
1410                  */
1411                 if (out_dev == in_dev && !(flags & RTCF_NAT))
1412                         flags |= RTCF_DOREDIRECT;
1413
1414                 local_src = DN_FIB_RES_PREFSRC(res);
1415                 break;
1416         case RTN_BLACKHOLE:
1417         case RTN_UNREACHABLE:
1418                 break;
1419         case RTN_LOCAL:
1420                 flags |= RTCF_LOCAL;
1421                 fld.saddr = cb->dst;
1422                 fld.daddr = cb->src;
1423
1424                 /* Routing tables gave us a gateway */
1425                 if (gateway)
1426                         goto make_route;
1427
1428                 /* Packet was intra-ethernet, so we know its on-link */
1429                 if (cb->rt_flags & DN_RT_F_IE) {
1430                         gateway = cb->src;
1431                         goto make_route;
1432                 }
1433
1434                 /* Use the default router if there is one */
1435                 neigh = neigh_clone(dn_db->router);
1436                 if (neigh) {
1437                         gateway = ((struct dn_neigh *)neigh)->addr;
1438                         goto make_route;
1439                 }
1440
1441                 /* Close eyes and pray */
1442                 gateway = cb->src;
1443                 goto make_route;
1444         default:
1445                 goto e_inval;
1446         }
1447
1448 make_route:
1449         rt = dst_alloc(&dn_dst_ops, out_dev, 1, DST_OBSOLETE_NONE, 0);
1450         if (rt == NULL)
1451                 goto e_nobufs;
1452
1453         rt->dn_next = NULL;
1454         memset(&rt->fld, 0, sizeof(rt->fld));
1455         rt->rt_saddr      = fld.saddr;
1456         rt->rt_daddr      = fld.daddr;
1457         rt->rt_gateway    = fld.daddr;
1458         if (gateway)
1459                 rt->rt_gateway = gateway;
1460         rt->rt_local_src  = local_src ? local_src : rt->rt_saddr;
1461
1462         rt->rt_dst_map    = fld.daddr;
1463         rt->rt_src_map    = fld.saddr;
1464
1465         rt->fld.saddr        = cb->src;
1466         rt->fld.daddr        = cb->dst;
1467         rt->fld.flowidn_oif  = 0;
1468         rt->fld.flowidn_iif  = in_dev->ifindex;
1469         rt->fld.flowidn_mark = fld.flowidn_mark;
1470
1471         rt->n = neigh;
1472         rt->dst.lastuse = jiffies;
1473         rt->dst.output = dn_rt_bug_out;
1474         switch (res.type) {
1475         case RTN_UNICAST:
1476                 rt->dst.input = dn_forward;
1477                 break;
1478         case RTN_LOCAL:
1479                 rt->dst.output = dn_output;
1480                 rt->dst.input = dn_nsp_rx;
1481                 rt->dst.dev = in_dev;
1482                 flags |= RTCF_LOCAL;
1483                 break;
1484         default:
1485         case RTN_UNREACHABLE:
1486         case RTN_BLACKHOLE:
1487                 rt->dst.input = dst_discard;
1488         }
1489         rt->rt_flags = flags;
1490
1491         err = dn_rt_set_next_hop(rt, &res);
1492         if (err)
1493                 goto e_neighbour;
1494
1495         hash = dn_hash(rt->fld.saddr, rt->fld.daddr);
1496         /* dn_insert_route() increments dst->__refcnt */
1497         dn_insert_route(rt, hash, &rt);
1498         skb_dst_set(skb, &rt->dst);
1499
1500 done:
1501         if (neigh)
1502                 neigh_release(neigh);
1503         if (free_res)
1504                 dn_fib_res_put(&res);
1505         dev_put(in_dev);
1506         if (out_dev)
1507                 dev_put(out_dev);
1508 out:
1509         return err;
1510
1511 e_inval:
1512         err = -EINVAL;
1513         goto done;
1514
1515 e_nobufs:
1516         err = -ENOBUFS;
1517         goto done;
1518
1519 e_neighbour:
1520         dst_release_immediate(&rt->dst);
1521         goto done;
1522 }
1523
1524 static int dn_route_input(struct sk_buff *skb)
1525 {
1526         struct dn_route *rt;
1527         struct dn_skb_cb *cb = DN_SKB_CB(skb);
1528         unsigned int hash = dn_hash(cb->src, cb->dst);
1529
1530         if (skb_dst(skb))
1531                 return 0;
1532
1533         rcu_read_lock();
1534         for (rt = rcu_dereference(dn_rt_hash_table[hash].chain); rt != NULL;
1535             rt = rcu_dereference(rt->dn_next)) {
1536                 if ((rt->fld.saddr == cb->src) &&
1537                     (rt->fld.daddr == cb->dst) &&
1538                     (rt->fld.flowidn_oif == 0) &&
1539                     (rt->fld.flowidn_mark == skb->mark) &&
1540                     (rt->fld.flowidn_iif == cb->iif)) {
1541                         dst_hold_and_use(&rt->dst, jiffies);
1542                         rcu_read_unlock();
1543                         skb_dst_set(skb, (struct dst_entry *)rt);
1544                         return 0;
1545                 }
1546         }
1547         rcu_read_unlock();
1548
1549         return dn_route_input_slow(skb);
1550 }
1551
1552 static int dn_rt_fill_info(struct sk_buff *skb, u32 portid, u32 seq,
1553                            int event, int nowait, unsigned int flags)
1554 {
1555         struct dn_route *rt = (struct dn_route *)skb_dst(skb);
1556         struct rtmsg *r;
1557         struct nlmsghdr *nlh;
1558         long expires;
1559
1560         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*r), flags);
1561         if (!nlh)
1562                 return -EMSGSIZE;
1563
1564         r = nlmsg_data(nlh);
1565         r->rtm_family = AF_DECnet;
1566         r->rtm_dst_len = 16;
1567         r->rtm_src_len = 0;
1568         r->rtm_tos = 0;
1569         r->rtm_table = RT_TABLE_MAIN;
1570         r->rtm_type = rt->rt_type;
1571         r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
1572         r->rtm_scope = RT_SCOPE_UNIVERSE;
1573         r->rtm_protocol = RTPROT_UNSPEC;
1574
1575         if (rt->rt_flags & RTCF_NOTIFY)
1576                 r->rtm_flags |= RTM_F_NOTIFY;
1577
1578         if (nla_put_u32(skb, RTA_TABLE, RT_TABLE_MAIN) < 0 ||
1579             nla_put_le16(skb, RTA_DST, rt->rt_daddr) < 0)
1580                 goto errout;
1581
1582         if (rt->fld.saddr) {
1583                 r->rtm_src_len = 16;
1584                 if (nla_put_le16(skb, RTA_SRC, rt->fld.saddr) < 0)
1585                         goto errout;
1586         }
1587         if (rt->dst.dev &&
1588             nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex) < 0)
1589                 goto errout;
1590
1591         /*
1592          * Note to self - change this if input routes reverse direction when
1593          * they deal only with inputs and not with replies like they do
1594          * currently.
1595          */
1596         if (nla_put_le16(skb, RTA_PREFSRC, rt->rt_local_src) < 0)
1597                 goto errout;
1598
1599         if (rt->rt_daddr != rt->rt_gateway &&
1600             nla_put_le16(skb, RTA_GATEWAY, rt->rt_gateway) < 0)
1601                 goto errout;
1602
1603         if (rtnetlink_put_metrics(skb, dst_metrics_ptr(&rt->dst)) < 0)
1604                 goto errout;
1605
1606         expires = rt->dst.expires ? rt->dst.expires - jiffies : 0;
1607         if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires,
1608                                rt->dst.error) < 0)
1609                 goto errout;
1610
1611         if (dn_is_input_route(rt) &&
1612             nla_put_u32(skb, RTA_IIF, rt->fld.flowidn_iif) < 0)
1613                 goto errout;
1614
1615         nlmsg_end(skb, nlh);
1616         return 0;
1617
1618 errout:
1619         nlmsg_cancel(skb, nlh);
1620         return -EMSGSIZE;
1621 }
1622
1623 const struct nla_policy rtm_dn_policy[RTA_MAX + 1] = {
1624         [RTA_DST]               = { .type = NLA_U16 },
1625         [RTA_SRC]               = { .type = NLA_U16 },
1626         [RTA_IIF]               = { .type = NLA_U32 },
1627         [RTA_OIF]               = { .type = NLA_U32 },
1628         [RTA_GATEWAY]           = { .type = NLA_U16 },
1629         [RTA_PRIORITY]          = { .type = NLA_U32 },
1630         [RTA_PREFSRC]           = { .type = NLA_U16 },
1631         [RTA_METRICS]           = { .type = NLA_NESTED },
1632         [RTA_MULTIPATH]         = { .type = NLA_NESTED },
1633         [RTA_TABLE]             = { .type = NLA_U32 },
1634         [RTA_MARK]              = { .type = NLA_U32 },
1635 };
1636
1637 /*
1638  * This is called by both endnodes and routers now.
1639  */
1640 static int dn_cache_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
1641                              struct netlink_ext_ack *extack)
1642 {
1643         struct net *net = sock_net(in_skb->sk);
1644         struct rtmsg *rtm = nlmsg_data(nlh);
1645         struct dn_route *rt = NULL;
1646         struct dn_skb_cb *cb;
1647         int err;
1648         struct sk_buff *skb;
1649         struct flowidn fld;
1650         struct nlattr *tb[RTA_MAX+1];
1651
1652         if (!net_eq(net, &init_net))
1653                 return -EINVAL;
1654
1655         err = nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
1656                                      rtm_dn_policy, extack);
1657         if (err < 0)
1658                 return err;
1659
1660         memset(&fld, 0, sizeof(fld));
1661         fld.flowidn_proto = DNPROTO_NSP;
1662
1663         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1664         if (skb == NULL)
1665                 return -ENOBUFS;
1666         skb_reset_mac_header(skb);
1667         cb = DN_SKB_CB(skb);
1668
1669         if (tb[RTA_SRC])
1670                 fld.saddr = nla_get_le16(tb[RTA_SRC]);
1671
1672         if (tb[RTA_DST])
1673                 fld.daddr = nla_get_le16(tb[RTA_DST]);
1674
1675         if (tb[RTA_IIF])
1676                 fld.flowidn_iif = nla_get_u32(tb[RTA_IIF]);
1677
1678         if (fld.flowidn_iif) {
1679                 struct net_device *dev;
1680                 dev = __dev_get_by_index(&init_net, fld.flowidn_iif);
1681                 if (!dev || !dev->dn_ptr) {
1682                         kfree_skb(skb);
1683                         return -ENODEV;
1684                 }
1685                 skb->protocol = htons(ETH_P_DNA_RT);
1686                 skb->dev = dev;
1687                 cb->src = fld.saddr;
1688                 cb->dst = fld.daddr;
1689                 local_bh_disable();
1690                 err = dn_route_input(skb);
1691                 local_bh_enable();
1692                 memset(cb, 0, sizeof(struct dn_skb_cb));
1693                 rt = (struct dn_route *)skb_dst(skb);
1694                 if (!err && -rt->dst.error)
1695                         err = rt->dst.error;
1696         } else {
1697                 if (tb[RTA_OIF])
1698                         fld.flowidn_oif = nla_get_u32(tb[RTA_OIF]);
1699
1700                 err = dn_route_output_key((struct dst_entry **)&rt, &fld, 0);
1701         }
1702
1703         skb->dev = NULL;
1704         if (err)
1705                 goto out_free;
1706         skb_dst_set(skb, &rt->dst);
1707         if (rtm->rtm_flags & RTM_F_NOTIFY)
1708                 rt->rt_flags |= RTCF_NOTIFY;
1709
1710         err = dn_rt_fill_info(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq, RTM_NEWROUTE, 0, 0);
1711         if (err < 0) {
1712                 err = -EMSGSIZE;
1713                 goto out_free;
1714         }
1715
1716         return rtnl_unicast(skb, &init_net, NETLINK_CB(in_skb).portid);
1717
1718 out_free:
1719         kfree_skb(skb);
1720         return err;
1721 }
1722
1723 /*
1724  * For routers, this is called from dn_fib_dump, but for endnodes its
1725  * called directly from the rtnetlink dispatch table.
1726  */
1727 int dn_cache_dump(struct sk_buff *skb, struct netlink_callback *cb)
1728 {
1729         struct net *net = sock_net(skb->sk);
1730         struct dn_route *rt;
1731         int h, s_h;
1732         int idx, s_idx;
1733         struct rtmsg *rtm;
1734
1735         if (!net_eq(net, &init_net))
1736                 return 0;
1737
1738         if (nlmsg_len(cb->nlh) < sizeof(struct rtmsg))
1739                 return -EINVAL;
1740
1741         rtm = nlmsg_data(cb->nlh);
1742         if (!(rtm->rtm_flags & RTM_F_CLONED))
1743                 return 0;
1744
1745         s_h = cb->args[0];
1746         s_idx = idx = cb->args[1];
1747         for (h = 0; h <= dn_rt_hash_mask; h++) {
1748                 if (h < s_h)
1749                         continue;
1750                 if (h > s_h)
1751                         s_idx = 0;
1752                 rcu_read_lock_bh();
1753                 for (rt = rcu_dereference_bh(dn_rt_hash_table[h].chain), idx = 0;
1754                         rt;
1755                         rt = rcu_dereference_bh(rt->dn_next), idx++) {
1756                         if (idx < s_idx)
1757                                 continue;
1758                         skb_dst_set(skb, dst_clone(&rt->dst));
1759                         if (dn_rt_fill_info(skb, NETLINK_CB(cb->skb).portid,
1760                                         cb->nlh->nlmsg_seq, RTM_NEWROUTE,
1761                                         1, NLM_F_MULTI) < 0) {
1762                                 skb_dst_drop(skb);
1763                                 rcu_read_unlock_bh();
1764                                 goto done;
1765                         }
1766                         skb_dst_drop(skb);
1767                 }
1768                 rcu_read_unlock_bh();
1769         }
1770
1771 done:
1772         cb->args[0] = h;
1773         cb->args[1] = idx;
1774         return skb->len;
1775 }
1776
1777 #ifdef CONFIG_PROC_FS
1778 struct dn_rt_cache_iter_state {
1779         int bucket;
1780 };
1781
1782 static struct dn_route *dn_rt_cache_get_first(struct seq_file *seq)
1783 {
1784         struct dn_route *rt = NULL;
1785         struct dn_rt_cache_iter_state *s = seq->private;
1786
1787         for (s->bucket = dn_rt_hash_mask; s->bucket >= 0; --s->bucket) {
1788                 rcu_read_lock_bh();
1789                 rt = rcu_dereference_bh(dn_rt_hash_table[s->bucket].chain);
1790                 if (rt)
1791                         break;
1792                 rcu_read_unlock_bh();
1793         }
1794         return rt;
1795 }
1796
1797 static struct dn_route *dn_rt_cache_get_next(struct seq_file *seq, struct dn_route *rt)
1798 {
1799         struct dn_rt_cache_iter_state *s = seq->private;
1800
1801         rt = rcu_dereference_bh(rt->dn_next);
1802         while (!rt) {
1803                 rcu_read_unlock_bh();
1804                 if (--s->bucket < 0)
1805                         break;
1806                 rcu_read_lock_bh();
1807                 rt = rcu_dereference_bh(dn_rt_hash_table[s->bucket].chain);
1808         }
1809         return rt;
1810 }
1811
1812 static void *dn_rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
1813 {
1814         struct dn_route *rt = dn_rt_cache_get_first(seq);
1815
1816         if (rt) {
1817                 while (*pos && (rt = dn_rt_cache_get_next(seq, rt)))
1818                         --*pos;
1819         }
1820         return *pos ? NULL : rt;
1821 }
1822
1823 static void *dn_rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1824 {
1825         struct dn_route *rt = dn_rt_cache_get_next(seq, v);
1826         ++*pos;
1827         return rt;
1828 }
1829
1830 static void dn_rt_cache_seq_stop(struct seq_file *seq, void *v)
1831 {
1832         if (v)
1833                 rcu_read_unlock_bh();
1834 }
1835
1836 static int dn_rt_cache_seq_show(struct seq_file *seq, void *v)
1837 {
1838         struct dn_route *rt = v;
1839         char buf1[DN_ASCBUF_LEN], buf2[DN_ASCBUF_LEN];
1840
1841         seq_printf(seq, "%-8s %-7s %-7s %04d %04d %04d\n",
1842                    rt->dst.dev ? rt->dst.dev->name : "*",
1843                    dn_addr2asc(le16_to_cpu(rt->rt_daddr), buf1),
1844                    dn_addr2asc(le16_to_cpu(rt->rt_saddr), buf2),
1845                    atomic_read(&rt->dst.__refcnt),
1846                    rt->dst.__use, 0);
1847         return 0;
1848 }
1849
1850 static const struct seq_operations dn_rt_cache_seq_ops = {
1851         .start  = dn_rt_cache_seq_start,
1852         .next   = dn_rt_cache_seq_next,
1853         .stop   = dn_rt_cache_seq_stop,
1854         .show   = dn_rt_cache_seq_show,
1855 };
1856 #endif /* CONFIG_PROC_FS */
1857
1858 void __init dn_route_init(void)
1859 {
1860         int i, goal, order;
1861
1862         dn_dst_ops.kmem_cachep =
1863                 kmem_cache_create("dn_dst_cache", sizeof(struct dn_route), 0,
1864                                   SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1865         dst_entries_init(&dn_dst_ops);
1866         timer_setup(&dn_route_timer, dn_dst_check_expire, 0);
1867         dn_route_timer.expires = jiffies + decnet_dst_gc_interval * HZ;
1868         add_timer(&dn_route_timer);
1869
1870         goal = totalram_pages() >> (26 - PAGE_SHIFT);
1871
1872         for (order = 0; (1UL << order) < goal; order++)
1873                 /* NOTHING */;
1874
1875         /*
1876          * Only want 1024 entries max, since the table is very, very unlikely
1877          * to be larger than that.
1878          */
1879         while (order && ((((1UL << order) * PAGE_SIZE) /
1880                                 sizeof(struct dn_rt_hash_bucket)) >= 2048))
1881                 order--;
1882
1883         do {
1884                 dn_rt_hash_mask = (1UL << order) * PAGE_SIZE /
1885                         sizeof(struct dn_rt_hash_bucket);
1886                 while (dn_rt_hash_mask & (dn_rt_hash_mask - 1))
1887                         dn_rt_hash_mask--;
1888                 dn_rt_hash_table = (struct dn_rt_hash_bucket *)
1889                         __get_free_pages(GFP_ATOMIC, order);
1890         } while (dn_rt_hash_table == NULL && --order > 0);
1891
1892         if (!dn_rt_hash_table)
1893                 panic("Failed to allocate DECnet route cache hash table\n");
1894
1895         printk(KERN_INFO
1896                 "DECnet: Routing cache hash table of %u buckets, %ldKbytes\n",
1897                 dn_rt_hash_mask,
1898                 (long)(dn_rt_hash_mask*sizeof(struct dn_rt_hash_bucket))/1024);
1899
1900         dn_rt_hash_mask--;
1901         for (i = 0; i <= dn_rt_hash_mask; i++) {
1902                 spin_lock_init(&dn_rt_hash_table[i].lock);
1903                 dn_rt_hash_table[i].chain = NULL;
1904         }
1905
1906         dn_dst_ops.gc_thresh = (dn_rt_hash_mask + 1);
1907
1908         proc_create_seq_private("decnet_cache", 0444, init_net.proc_net,
1909                         &dn_rt_cache_seq_ops,
1910                         sizeof(struct dn_rt_cache_iter_state), NULL);
1911
1912 #ifdef CONFIG_DECNET_ROUTER
1913         rtnl_register_module(THIS_MODULE, PF_DECnet, RTM_GETROUTE,
1914                              dn_cache_getroute, dn_fib_dump, 0);
1915 #else
1916         rtnl_register_module(THIS_MODULE, PF_DECnet, RTM_GETROUTE,
1917                              dn_cache_getroute, dn_cache_dump, 0);
1918 #endif
1919 }
1920
1921 void __exit dn_route_cleanup(void)
1922 {
1923         del_timer(&dn_route_timer);
1924         dn_run_flush(NULL);
1925
1926         remove_proc_entry("decnet_cache", init_net.proc_net);
1927         dst_entries_destroy(&dn_dst_ops);
1928 }