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 struct dst_entry *dn_dst_negative_advice(struct dst_entry *);
 115 static void dn_dst_link_failure(struct sk_buff *);
 116 static void dn_dst_update_pmtu(struct dst_entry *dst, u32 mtu);
 117 static int dn_route_input(struct sk_buff *);
 118 static void dn_run_flush(unsigned long dummy);
 119
 120 static struct dn_rt_hash_bucket *dn_rt_hash_table;
 121 static unsigned dn_rt_hash_mask;
 122
 123 static struct timer_list dn_route_timer;
 124 static DEFINE_TIMER(dn_rt_flush_timer, dn_run_flush, 0, 0);
 125 int decnet_dst_gc_interval = 2;
 126
 127 static struct dst_ops dn_dst_ops = {
 128         .family =               PF_DECnet,
 129         .protocol =             cpu_to_be16(ETH_P_DNA_RT),
 130         .gc_thresh =            128,
 131         .gc =                   dn_dst_gc,
 132         .check =                dn_dst_check,
 133         .default_advmss =       dn_dst_default_advmss,
 134         .negative_advice =      dn_dst_negative_advice,
 135         .link_failure =         dn_dst_link_failure,
 136         .update_pmtu =          dn_dst_update_pmtu,
 137 };
 138
 139 static __inline__ unsigned dn_hash(__le16 src, __le16 dst)
 140 {
 141         __u16 tmp = (__u16 __force)(src ^ dst);
 142         tmp ^= (tmp >> 3);
 143         tmp ^= (tmp >> 5);
 144         tmp ^= (tmp >> 10);
 145         return dn_rt_hash_mask & (unsigned)tmp;
 146 }
 147
 148 static inline void dnrt_free(struct dn_route *rt)
 149 {
 150         call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
 151 }
 152
 153 static inline void dnrt_drop(struct dn_route *rt)
 154 {
 155         dst_release(&rt->dst);
 156         call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
 157 }
 158
 159 static void dn_dst_check_expire(unsigned long dummy)
 160 {
 161         int i;
 162         struct dn_route *rt;
 163         struct dn_route __rcu **rtp;
 164         unsigned long now = jiffies;
 165         unsigned long expire = 120 * HZ;
 166
 167         for (i = 0; i <= dn_rt_hash_mask; i++) {
 168                 rtp = &dn_rt_hash_table[i].chain;
 169
 170                 spin_lock(&dn_rt_hash_table[i].lock);
 171                 while ((rt = rcu_dereference_protected(*rtp,
 172                                                 lockdep_is_held(&dn_rt_hash_table[i].lock))) != NULL) {
 173                         if (atomic_read(&rt->dst.__refcnt) ||
 174                                         (now - rt->dst.lastuse) < expire) {
 175                                 rtp = &rt->dst.dn_next;
 176                                 continue;
 177                         }
 178                         *rtp = rt->dst.dn_next;
 179                         rt->dst.dn_next = NULL;
 180                         dnrt_free(rt);
 181                 }
 182                 spin_unlock(&dn_rt_hash_table[i].lock);
 183
 184                 if ((jiffies - now) > 0)
 185                         break;
 186         }
 187
 188         mod_timer(&dn_route_timer, now + decnet_dst_gc_interval * HZ);
 189 }
 190
 191 static int dn_dst_gc(struct dst_ops *ops)
 192 {
 193         struct dn_route *rt;
 194         struct dn_route __rcu **rtp;
 195         int i;
 196         unsigned long now = jiffies;
 197         unsigned long expire = 10 * HZ;
 198
 199         for (i = 0; i <= dn_rt_hash_mask; i++) {
 200
 201                 spin_lock_bh(&dn_rt_hash_table[i].lock);
 202                 rtp = &dn_rt_hash_table[i].chain;
 203
 204                 while ((rt = rcu_dereference_protected(*rtp,
 205                                                 lockdep_is_held(&dn_rt_hash_table[i].lock))) != NULL) {
 206                         if (atomic_read(&rt->dst.__refcnt) ||
 207                                         (now - rt->dst.lastuse) < expire) {
 208                                 rtp = &rt->dst.dn_next;
 209                                 continue;
 210                         }
 211                         *rtp = rt->dst.dn_next;
 212                         rt->dst.dn_next = NULL;
 213                         dnrt_drop(rt);
 214                         break;
 215                 }
 216                 spin_unlock_bh(&dn_rt_hash_table[i].lock);
 217         }
 218
 219         return 0;
 220 }
 221
 222 /*
 223  * The decnet standards don't impose a particular minimum mtu, what they
 224  * do insist on is that the routing layer accepts a datagram of at least
 225  * 230 bytes long. Here we have to subtract the routing header length from
 226  * 230 to get the minimum acceptable mtu. If there is no neighbour, then we
 227  * assume the worst and use a long header size.
 228  *
 229  * We update both the mtu and the advertised mss (i.e. the segment size we
 230  * advertise to the other end).
 231  */
 232 static void dn_dst_update_pmtu(struct dst_entry *dst, u32 mtu)
 233 {
 234         u32 min_mtu = 230;
 235         struct dn_dev *dn = dst->neighbour ?
 236                             rcu_dereference_raw(dst->neighbour->dev->dn_ptr) : NULL;
 237
 238         if (dn && dn->use_long == 0)
 239                 min_mtu -= 6;
 240         else
 241                 min_mtu -= 21;
 242
 243         if (dst_metric(dst, RTAX_MTU) > mtu && mtu >= min_mtu) {
 244                 if (!(dst_metric_locked(dst, RTAX_MTU))) {
 245                         dst_metric_set(dst, RTAX_MTU, mtu);
 246                         dst_set_expires(dst, dn_rt_mtu_expires);
 247                 }
 248                 if (!(dst_metric_locked(dst, RTAX_ADVMSS))) {
 249                         u32 mss = mtu - DN_MAX_NSP_DATA_HEADER;
 250                         u32 existing_mss = dst_metric_raw(dst, RTAX_ADVMSS);
 251                         if (!existing_mss || existing_mss > mss)
 252                                 dst_metric_set(dst, RTAX_ADVMSS, mss);
 253                 }
 254         }
 255 }
 256
 257 /*
 258  * When a route has been marked obsolete. (e.g. routing cache flush)
 259  */
 260 static struct dst_entry *dn_dst_check(struct dst_entry *dst, __u32 cookie)
 261 {
 262         return NULL;
 263 }
 264
 265 static struct dst_entry *dn_dst_negative_advice(struct dst_entry *dst)
 266 {
 267         dst_release(dst);
 268         return NULL;
 269 }
 270
 271 static void dn_dst_link_failure(struct sk_buff *skb)
 272 {
 273 }
 274
 275 static inline int compare_keys(struct flowi *fl1, struct flowi *fl2)
 276 {
 277         return ((fl1->fld_dst ^ fl2->fld_dst) |
 278                 (fl1->fld_src ^ fl2->fld_src) |
 279                 (fl1->mark ^ fl2->mark) |
 280                 (fl1->fld_scope ^ fl2->fld_scope) |
 281                 (fl1->oif ^ fl2->oif) |
 282                 (fl1->iif ^ fl2->iif)) == 0;
 283 }
 284
 285 static int dn_insert_route(struct dn_route *rt, unsigned hash, struct dn_route **rp)
 286 {
 287         struct dn_route *rth;
 288         struct dn_route __rcu **rthp;
 289         unsigned long now = jiffies;
 290
 291         rthp = &dn_rt_hash_table[hash].chain;
 292
 293         spin_lock_bh(&dn_rt_hash_table[hash].lock);
 294         while ((rth = rcu_dereference_protected(*rthp,
 295                                                 lockdep_is_held(&dn_rt_hash_table[hash].lock))) != NULL) {
 296                 if (compare_keys(&rth->fl, &rt->fl)) {
 297                         /* Put it first */
 298                         *rthp = rth->dst.dn_next;
 299                         rcu_assign_pointer(rth->dst.dn_next,
 300                                            dn_rt_hash_table[hash].chain);
 301                         rcu_assign_pointer(dn_rt_hash_table[hash].chain, rth);
 302
 303                         dst_use(&rth->dst, now);
 304                         spin_unlock_bh(&dn_rt_hash_table[hash].lock);
 305
 306                         dnrt_drop(rt);
 307                         *rp = rth;
 308                         return 0;
 309                 }
 310                 rthp = &rth->dst.dn_next;
 311         }
 312
 313         rcu_assign_pointer(rt->dst.dn_next, dn_rt_hash_table[hash].chain);
 314         rcu_assign_pointer(dn_rt_hash_table[hash].chain, rt);
 315
 316         dst_use(&rt->dst, now);
 317         spin_unlock_bh(&dn_rt_hash_table[hash].lock);
 318         *rp = rt;
 319         return 0;
 320 }
 321
 322 static void dn_run_flush(unsigned long dummy)
 323 {
 324         int i;
 325         struct dn_route *rt, *next;
 326
 327         for (i = 0; i < dn_rt_hash_mask; i++) {
 328                 spin_lock_bh(&dn_rt_hash_table[i].lock);
 329
 330                 if ((rt = xchg((struct dn_route **)&dn_rt_hash_table[i].chain, NULL)) == NULL)
 331                         goto nothing_to_declare;
 332
 333                 for(; rt; rt = next) {
 334                         next = rcu_dereference_raw(rt->dst.dn_next);
 335                         RCU_INIT_POINTER(rt->dst.dn_next, NULL);
 336                         dst_free((struct dst_entry *)rt);
 337                 }
 338
 339 nothing_to_declare:
 340                 spin_unlock_bh(&dn_rt_hash_table[i].lock);
 341         }
 342 }
 343
 344 static DEFINE_SPINLOCK(dn_rt_flush_lock);
 345
 346 void dn_rt_cache_flush(int delay)
 347 {
 348         unsigned long now = jiffies;
 349         int user_mode = !in_interrupt();
 350
 351         if (delay < 0)
 352                 delay = dn_rt_min_delay;
 353
 354         spin_lock_bh(&dn_rt_flush_lock);
 355
 356         if (del_timer(&dn_rt_flush_timer) && delay > 0 && dn_rt_deadline) {
 357                 long tmo = (long)(dn_rt_deadline - now);
 358
 359                 if (user_mode && tmo < dn_rt_max_delay - dn_rt_min_delay)
 360                         tmo = 0;
 361
 362                 if (delay > tmo)
 363                         delay = tmo;
 364         }
 365
 366         if (delay <= 0) {
 367                 spin_unlock_bh(&dn_rt_flush_lock);
 368                 dn_run_flush(0);
 369                 return;
 370         }
 371
 372         if (dn_rt_deadline == 0)
 373                 dn_rt_deadline = now + dn_rt_max_delay;
 374
 375         dn_rt_flush_timer.expires = now + delay;
 376         add_timer(&dn_rt_flush_timer);
 377         spin_unlock_bh(&dn_rt_flush_lock);
 378 }
 379
 380 /**
 381  * dn_return_short - Return a short packet to its sender
 382  * @skb: The packet to return
 383  *
 384  */
 385 static int dn_return_short(struct sk_buff *skb)
 386 {
 387         struct dn_skb_cb *cb;
 388         unsigned char *ptr;
 389         __le16 *src;
 390         __le16 *dst;
 391
 392         /* Add back headers */
 393         skb_push(skb, skb->data - skb_network_header(skb));
 394
 395         if ((skb = skb_unshare(skb, GFP_ATOMIC)) == NULL)
 396                 return NET_RX_DROP;
 397
 398         cb = DN_SKB_CB(skb);
 399         /* Skip packet length and point to flags */
 400         ptr = skb->data + 2;
 401         *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
 402
 403         dst = (__le16 *)ptr;
 404         ptr += 2;
 405         src = (__le16 *)ptr;
 406         ptr += 2;
 407         *ptr = 0; /* Zero hop count */
 408
 409         swap(*src, *dst);
 410
 411         skb->pkt_type = PACKET_OUTGOING;
 412         dn_rt_finish_output(skb, NULL, NULL);
 413         return NET_RX_SUCCESS;
 414 }
 415
 416 /**
 417  * dn_return_long - Return a long packet to its sender
 418  * @skb: The long format packet to return
 419  *
 420  */
 421 static int dn_return_long(struct sk_buff *skb)
 422 {
 423         struct dn_skb_cb *cb;
 424         unsigned char *ptr;
 425         unsigned char *src_addr, *dst_addr;
 426         unsigned char tmp[ETH_ALEN];
 427
 428         /* Add back all headers */
 429         skb_push(skb, skb->data - skb_network_header(skb));
 430
 431         if ((skb = skb_unshare(skb, GFP_ATOMIC)) == NULL)
 432                 return NET_RX_DROP;
 433
 434         cb = DN_SKB_CB(skb);
 435         /* Ignore packet length and point to flags */
 436         ptr = skb->data + 2;
 437
 438         /* Skip padding */
 439         if (*ptr & DN_RT_F_PF) {
 440                 char padlen = (*ptr & ~DN_RT_F_PF);
 441                 ptr += padlen;
 442         }
 443
 444         *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
 445         ptr += 2;
 446         dst_addr = ptr;
 447         ptr += 8;
 448         src_addr = ptr;
 449         ptr += 6;
 450         *ptr = 0; /* Zero hop count */
 451
 452         /* Swap source and destination */
 453         memcpy(tmp, src_addr, ETH_ALEN);
 454         memcpy(src_addr, dst_addr, ETH_ALEN);
 455         memcpy(dst_addr, tmp, ETH_ALEN);
 456
 457         skb->pkt_type = PACKET_OUTGOING;
 458         dn_rt_finish_output(skb, dst_addr, src_addr);
 459         return NET_RX_SUCCESS;
 460 }
 461
 462 /**
 463  * dn_route_rx_packet - Try and find a route for an incoming packet
 464  * @skb: The packet to find a route for
 465  *
 466  * Returns: result of input function if route is found, error code otherwise
 467  */
 468 static int dn_route_rx_packet(struct sk_buff *skb)
 469 {
 470         struct dn_skb_cb *cb;
 471         int err;
 472
 473         if ((err = dn_route_input(skb)) == 0)
 474                 return dst_input(skb);
 475
 476         cb = DN_SKB_CB(skb);
 477         if (decnet_debug_level & 4) {
 478                 char *devname = skb->dev ? skb->dev->name : "???";
 479
 480                 printk(KERN_DEBUG
 481                         "DECnet: dn_route_rx_packet: rt_flags=0x%02x dev=%s len=%d src=0x%04hx dst=0x%04hx err=%d type=%d\n",
 482                         (int)cb->rt_flags, devname, skb->len,
 483                         le16_to_cpu(cb->src), le16_to_cpu(cb->dst),
 484                         err, skb->pkt_type);
 485         }
 486
 487         if ((skb->pkt_type == PACKET_HOST) && (cb->rt_flags & DN_RT_F_RQR)) {
 488                 switch(cb->rt_flags & DN_RT_PKT_MSK) {
 489                         case DN_RT_PKT_SHORT:
 490                                 return dn_return_short(skb);
 491                         case DN_RT_PKT_LONG:
 492                                 return dn_return_long(skb);
 493                 }
 494         }
 495
 496         kfree_skb(skb);
 497         return NET_RX_DROP;
 498 }
 499
 500 static int dn_route_rx_long(struct sk_buff *skb)
 501 {
 502         struct dn_skb_cb *cb = DN_SKB_CB(skb);
 503         unsigned char *ptr = skb->data;
 504
 505         if (!pskb_may_pull(skb, 21)) /* 20 for long header, 1 for shortest nsp */
 506                 goto drop_it;
 507
 508         skb_pull(skb, 20);
 509         skb_reset_transport_header(skb);
 510
 511         /* Destination info */
 512         ptr += 2;
 513         cb->dst = dn_eth2dn(ptr);
 514         if (memcmp(ptr, dn_hiord_addr, 4) != 0)
 515                 goto drop_it;
 516         ptr += 6;
 517
 518
 519         /* Source info */
 520         ptr += 2;
 521         cb->src = dn_eth2dn(ptr);
 522         if (memcmp(ptr, dn_hiord_addr, 4) != 0)
 523                 goto drop_it;
 524         ptr += 6;
 525         /* Other junk */
 526         ptr++;
 527         cb->hops = *ptr++; /* Visit Count */
 528
 529         return NF_HOOK(NFPROTO_DECNET, NF_DN_PRE_ROUTING, skb, skb->dev, NULL,
 530                        dn_route_rx_packet);
 531
 532 drop_it:
 533         kfree_skb(skb);
 534         return NET_RX_DROP;
 535 }
 536
 537
 538
 539 static int dn_route_rx_short(struct sk_buff *skb)
 540 {
 541         struct dn_skb_cb *cb = DN_SKB_CB(skb);
 542         unsigned char *ptr = skb->data;
 543
 544         if (!pskb_may_pull(skb, 6)) /* 5 for short header + 1 for shortest nsp */
 545                 goto drop_it;
 546
 547         skb_pull(skb, 5);
 548         skb_reset_transport_header(skb);
 549
 550         cb->dst = *(__le16 *)ptr;
 551         ptr += 2;
 552         cb->src = *(__le16 *)ptr;
 553         ptr += 2;
 554         cb->hops = *ptr & 0x3f;
 555
 556         return NF_HOOK(NFPROTO_DECNET, NF_DN_PRE_ROUTING, skb, skb->dev, NULL,
 557                        dn_route_rx_packet);
 558
 559 drop_it:
 560         kfree_skb(skb);
 561         return NET_RX_DROP;
 562 }
 563
 564 static int dn_route_discard(struct sk_buff *skb)
 565 {
 566         /*
 567          * I know we drop the packet here, but thats considered success in
 568          * this case
 569          */
 570         kfree_skb(skb);
 571         return NET_RX_SUCCESS;
 572 }
 573
 574 static int dn_route_ptp_hello(struct sk_buff *skb)
 575 {
 576         dn_dev_hello(skb);
 577         dn_neigh_pointopoint_hello(skb);
 578         return NET_RX_SUCCESS;
 579 }
 580
 581 int dn_route_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
 582 {
 583         struct dn_skb_cb *cb;
 584         unsigned char flags = 0;
 585         __u16 len = le16_to_cpu(*(__le16 *)skb->data);
 586         struct dn_dev *dn = rcu_dereference(dev->dn_ptr);
 587         unsigned char padlen = 0;
 588
 589         if (!net_eq(dev_net(dev), &init_net))
 590                 goto dump_it;
 591
 592         if (dn == NULL)
 593                 goto dump_it;
 594
 595         if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
 596                 goto out;
 597
 598         if (!pskb_may_pull(skb, 3))
 599                 goto dump_it;
 600
 601         skb_pull(skb, 2);
 602
 603         if (len > skb->len)
 604                 goto dump_it;
 605
 606         skb_trim(skb, len);
 607
 608         flags = *skb->data;
 609
 610         cb = DN_SKB_CB(skb);
 611         cb->stamp = jiffies;
 612         cb->iif = dev->ifindex;
 613
 614         /*
 615          * If we have padding, remove it.
 616          */
 617         if (flags & DN_RT_F_PF) {
 618                 padlen = flags & ~DN_RT_F_PF;
 619                 if (!pskb_may_pull(skb, padlen + 1))
 620                         goto dump_it;
 621                 skb_pull(skb, padlen);
 622                 flags = *skb->data;
 623         }
 624
 625         skb_reset_network_header(skb);
 626
 627         /*
 628          * Weed out future version DECnet
 629          */
 630         if (flags & DN_RT_F_VER)
 631                 goto dump_it;
 632
 633         cb->rt_flags = flags;
 634
 635         if (decnet_debug_level & 1)
 636                 printk(KERN_DEBUG
 637                         "dn_route_rcv: got 0x%02x from %s [%d %d %d]\n",
 638                         (int)flags, (dev) ? dev->name : "???", len, skb->len,
 639                         padlen);
 640
 641         if (flags & DN_RT_PKT_CNTL) {
 642                 if (unlikely(skb_linearize(skb)))
 643                         goto dump_it;
 644
 645                 switch(flags & DN_RT_CNTL_MSK) {
 646                         case DN_RT_PKT_INIT:
 647                                 dn_dev_init_pkt(skb);
 648                                 break;
 649                         case DN_RT_PKT_VERI:
 650                                 dn_dev_veri_pkt(skb);
 651                                 break;
 652                 }
 653
 654                 if (dn->parms.state != DN_DEV_S_RU)
 655                         goto dump_it;
 656
 657                 switch(flags & DN_RT_CNTL_MSK) {
 658                         case DN_RT_PKT_HELO:
 659                                 return NF_HOOK(NFPROTO_DECNET, NF_DN_HELLO,
 660                                                skb, skb->dev, NULL,
 661                                                dn_route_ptp_hello);
 662
 663                         case DN_RT_PKT_L1RT:
 664                         case DN_RT_PKT_L2RT:
 665                                 return NF_HOOK(NFPROTO_DECNET, NF_DN_ROUTE,
 666                                                skb, skb->dev, NULL,
 667                                                dn_route_discard);
 668                         case DN_RT_PKT_ERTH:
 669                                 return NF_HOOK(NFPROTO_DECNET, NF_DN_HELLO,
 670                                                skb, skb->dev, NULL,
 671                                                dn_neigh_router_hello);
 672
 673                         case DN_RT_PKT_EEDH:
 674                                 return NF_HOOK(NFPROTO_DECNET, NF_DN_HELLO,
 675                                                skb, skb->dev, NULL,
 676                                                dn_neigh_endnode_hello);
 677                 }
 678         } else {
 679                 if (dn->parms.state != DN_DEV_S_RU)
 680                         goto dump_it;
 681
 682                 skb_pull(skb, 1); /* Pull flags */
 683
 684                 switch(flags & DN_RT_PKT_MSK) {
 685                         case DN_RT_PKT_LONG:
 686                                 return dn_route_rx_long(skb);
 687                         case DN_RT_PKT_SHORT:
 688                                 return dn_route_rx_short(skb);
 689                 }
 690         }
 691
 692 dump_it:
 693         kfree_skb(skb);
 694 out:
 695         return NET_RX_DROP;
 696 }
 697
 698 static int dn_output(struct sk_buff *skb)
 699 {
 700         struct dst_entry *dst = skb_dst(skb);
 701         struct dn_route *rt = (struct dn_route *)dst;
 702         struct net_device *dev = dst->dev;
 703         struct dn_skb_cb *cb = DN_SKB_CB(skb);
 704         struct neighbour *neigh;
 705
 706         int err = -EINVAL;
 707
 708         if ((neigh = dst->neighbour) == NULL)
 709                 goto error;
 710
 711         skb->dev = dev;
 712
 713         cb->src = rt->rt_saddr;
 714         cb->dst = rt->rt_daddr;
 715
 716         /*
 717          * Always set the Intra-Ethernet bit on all outgoing packets
 718          * originated on this node. Only valid flag from upper layers
 719          * is return-to-sender-requested. Set hop count to 0 too.
 720          */
 721         cb->rt_flags &= ~DN_RT_F_RQR;
 722         cb->rt_flags |= DN_RT_F_IE;
 723         cb->hops = 0;
 724
 725         return NF_HOOK(NFPROTO_DECNET, NF_DN_LOCAL_OUT, skb, NULL, dev,
 726                        neigh->output);
 727
 728 error:
 729         if (net_ratelimit())
 730                 printk(KERN_DEBUG "dn_output: This should not happen\n");
 731
 732         kfree_skb(skb);
 733
 734         return err;
 735 }
 736
 737 static int dn_forward(struct sk_buff *skb)
 738 {
 739         struct dn_skb_cb *cb = DN_SKB_CB(skb);
 740         struct dst_entry *dst = skb_dst(skb);
 741         struct dn_dev *dn_db = rcu_dereference(dst->dev->dn_ptr);
 742         struct dn_route *rt;
 743         struct neighbour *neigh = dst->neighbour;
 744         int header_len;
 745 #ifdef CONFIG_NETFILTER
 746         struct net_device *dev = skb->dev;
 747 #endif
 748
 749         if (skb->pkt_type != PACKET_HOST)
 750                 goto drop;
 751
 752         /* Ensure that we have enough space for headers */
 753         rt = (struct dn_route *)skb_dst(skb);
 754         header_len = dn_db->use_long ? 21 : 6;
 755         if (skb_cow(skb, LL_RESERVED_SPACE(rt->dst.dev)+header_len))
 756                 goto drop;
 757
 758         /*
 759          * Hop count exceeded.
 760          */
 761         if (++cb->hops > 30)
 762                 goto drop;
 763
 764         skb->dev = rt->dst.dev;
 765
 766         /*
 767          * If packet goes out same interface it came in on, then set
 768          * the Intra-Ethernet bit. This has no effect for short
 769          * packets, so we don't need to test for them here.
 770          */
 771         cb->rt_flags &= ~DN_RT_F_IE;
 772         if (rt->rt_flags & RTCF_DOREDIRECT)
 773                 cb->rt_flags |= DN_RT_F_IE;
 774
 775         return NF_HOOK(NFPROTO_DECNET, NF_DN_FORWARD, skb, dev, skb->dev,
 776                        neigh->output);
 777
 778 drop:
 779         kfree_skb(skb);
 780         return NET_RX_DROP;
 781 }
 782
 783 /*
 784  * Used to catch bugs. This should never normally get
 785  * called.
 786  */
 787 static int dn_rt_bug(struct sk_buff *skb)
 788 {
 789         if (net_ratelimit()) {
 790                 struct dn_skb_cb *cb = DN_SKB_CB(skb);
 791
 792                 printk(KERN_DEBUG "dn_rt_bug: skb from:%04x to:%04x\n",
 793                                 le16_to_cpu(cb->src), le16_to_cpu(cb->dst));
 794         }
 795
 796         kfree_skb(skb);
 797
 798         return NET_RX_DROP;
 799 }
 800
 801 static unsigned int dn_dst_default_advmss(const struct dst_entry *dst)
 802 {
 803         return dn_mss_from_pmtu(dst->dev, dst_mtu(dst));
 804 }
 805
 806 static int dn_rt_set_next_hop(struct dn_route *rt, struct dn_fib_res *res)
 807 {
 808         struct dn_fib_info *fi = res->fi;
 809         struct net_device *dev = rt->dst.dev;
 810         struct neighbour *n;
 811         unsigned int metric;
 812
 813         if (fi) {
 814                 if (DN_FIB_RES_GW(*res) &&
 815                     DN_FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
 816                         rt->rt_gateway = DN_FIB_RES_GW(*res);
 817                 dst_import_metrics(&rt->dst, fi->fib_metrics);
 818         }
 819         rt->rt_type = res->type;
 820
 821         if (dev != NULL && rt->dst.neighbour == NULL) {
 822                 n = __neigh_lookup_errno(&dn_neigh_table, &rt->rt_gateway, dev);
 823                 if (IS_ERR(n))
 824                         return PTR_ERR(n);
 825                 rt->dst.neighbour = n;
 826         }
 827
 828         if (dst_metric(&rt->dst, RTAX_MTU) == 0 ||
 829             dst_metric(&rt->dst, RTAX_MTU) > rt->dst.dev->mtu)
 830                 dst_metric_set(&rt->dst, RTAX_MTU, rt->dst.dev->mtu);
 831         metric = dst_metric_raw(&rt->dst, RTAX_ADVMSS);
 832         if (metric) {
 833                 unsigned int mss = dn_mss_from_pmtu(dev, dst_mtu(&rt->dst));
 834                 if (metric > mss)
 835                         dst_metric_set(&rt->dst, RTAX_ADVMSS, mss);
 836         }
 837         return 0;
 838 }
 839
 840 static inline int dn_match_addr(__le16 addr1, __le16 addr2)
 841 {
 842         __u16 tmp = le16_to_cpu(addr1) ^ le16_to_cpu(addr2);
 843         int match = 16;
 844         while(tmp) {
 845                 tmp >>= 1;
 846                 match--;
 847         }
 848         return match;
 849 }
 850
 851 static __le16 dnet_select_source(const struct net_device *dev, __le16 daddr, int scope)
 852 {
 853         __le16 saddr = 0;
 854         struct dn_dev *dn_db;
 855         struct dn_ifaddr *ifa;
 856         int best_match = 0;
 857         int ret;
 858
 859         rcu_read_lock();
 860         dn_db = rcu_dereference(dev->dn_ptr);
 861         for (ifa = rcu_dereference(dn_db->ifa_list);
 862              ifa != NULL;
 863              ifa = rcu_dereference(ifa->ifa_next)) {
 864                 if (ifa->ifa_scope > scope)
 865                         continue;
 866                 if (!daddr) {
 867                         saddr = ifa->ifa_local;
 868                         break;
 869                 }
 870                 ret = dn_match_addr(daddr, ifa->ifa_local);
 871                 if (ret > best_match)
 872                         saddr = ifa->ifa_local;
 873                 if (best_match == 0)
 874                         saddr = ifa->ifa_local;
 875         }
 876         rcu_read_unlock();
 877
 878         return saddr;
 879 }
 880
 881 static inline __le16 __dn_fib_res_prefsrc(struct dn_fib_res *res)
 882 {
 883         return dnet_select_source(DN_FIB_RES_DEV(*res), DN_FIB_RES_GW(*res), res->scope);
 884 }
 885
 886 static inline __le16 dn_fib_rules_map_destination(__le16 daddr, struct dn_fib_res *res)
 887 {
 888         __le16 mask = dnet_make_mask(res->prefixlen);
 889         return (daddr&~mask)|res->fi->fib_nh->nh_gw;
 890 }
 891
 892 static int dn_route_output_slow(struct dst_entry **pprt, const struct flowi *oldflp, int try_hard)
 893 {
 894         struct flowi fl = { .fld_dst = oldflp->fld_dst,
 895                             .fld_src = oldflp->fld_src,
 896                             .fld_scope = RT_SCOPE_UNIVERSE,
 897                             .mark = oldflp->mark,
 898                             .iif = init_net.loopback_dev->ifindex,
 899                             .oif = oldflp->oif };
 900         struct dn_route *rt = NULL;
 901         struct net_device *dev_out = NULL, *dev;
 902         struct neighbour *neigh = NULL;
 903         unsigned hash;
 904         unsigned flags = 0;
 905         struct dn_fib_res res = { .fi = NULL, .type = RTN_UNICAST };
 906         int err;
 907         int free_res = 0;
 908         __le16 gateway = 0;
 909
 910         if (decnet_debug_level & 16)
 911                 printk(KERN_DEBUG
 912                        "dn_route_output_slow: dst=%04x src=%04x mark=%d"
 913                        " iif=%d oif=%d\n", le16_to_cpu(oldflp->fld_dst),
 914                        le16_to_cpu(oldflp->fld_src),
 915                        oldflp->mark, init_net.loopback_dev->ifindex, oldflp->oif);
 916
 917         /* If we have an output interface, verify its a DECnet device */
 918         if (oldflp->oif) {
 919                 dev_out = dev_get_by_index(&init_net, oldflp->oif);
 920                 err = -ENODEV;
 921                 if (dev_out && dev_out->dn_ptr == NULL) {
 922                         dev_put(dev_out);
 923                         dev_out = NULL;
 924                 }
 925                 if (dev_out == NULL)
 926                         goto out;
 927         }
 928
 929         /* If we have a source address, verify that its a local address */
 930         if (oldflp->fld_src) {
 931                 err = -EADDRNOTAVAIL;
 932
 933                 if (dev_out) {
 934                         if (dn_dev_islocal(dev_out, oldflp->fld_src))
 935                                 goto source_ok;
 936                         dev_put(dev_out);
 937                         goto out;
 938                 }
 939                 rcu_read_lock();
 940                 for_each_netdev_rcu(&init_net, dev) {
 941                         if (!dev->dn_ptr)
 942                                 continue;
 943                         if (!dn_dev_islocal(dev, oldflp->fld_src))
 944                                 continue;
 945                         if ((dev->flags & IFF_LOOPBACK) &&
 946                             oldflp->fld_dst &&
 947                             !dn_dev_islocal(dev, oldflp->fld_dst))
 948                                 continue;
 949
 950                         dev_out = dev;
 951                         break;
 952                 }
 953                 rcu_read_unlock();
 954                 if (dev_out == NULL)
 955                         goto out;
 956                 dev_hold(dev_out);
 957 source_ok:
 958                 ;
 959         }
 960
 961         /* No destination? Assume its local */
 962         if (!fl.fld_dst) {
 963                 fl.fld_dst = fl.fld_src;
 964
 965                 err = -EADDRNOTAVAIL;
 966                 if (dev_out)
 967                         dev_put(dev_out);
 968                 dev_out = init_net.loopback_dev;
 969                 dev_hold(dev_out);
 970                 if (!fl.fld_dst) {
 971                         fl.fld_dst =
 972                         fl.fld_src = dnet_select_source(dev_out, 0,
 973                                                        RT_SCOPE_HOST);
 974                         if (!fl.fld_dst)
 975                                 goto out;
 976                 }
 977                 fl.oif = init_net.loopback_dev->ifindex;
 978                 res.type = RTN_LOCAL;
 979                 goto make_route;
 980         }
 981
 982         if (decnet_debug_level & 16)
 983                 printk(KERN_DEBUG
 984                        "dn_route_output_slow: initial checks complete."
 985                        " dst=%o4x src=%04x oif=%d try_hard=%d\n",
 986                        le16_to_cpu(fl.fld_dst), le16_to_cpu(fl.fld_src),
 987                        fl.oif, try_hard);
 988
 989         /*
 990          * N.B. If the kernel is compiled without router support then
 991          * dn_fib_lookup() will evaluate to non-zero so this if () block
 992          * will always be executed.
 993          */
 994         err = -ESRCH;
 995         if (try_hard || (err = dn_fib_lookup(&fl, &res)) != 0) {
 996                 struct dn_dev *dn_db;
 997                 if (err != -ESRCH)
 998                         goto out;
 999                 /*
1000                  * Here the fallback is basically the standard algorithm for
1001                  * routing in endnodes which is described in the DECnet routing
1002                  * docs
1003                  *
1004                  * If we are not trying hard, look in neighbour cache.
1005                  * The result is tested to ensure that if a specific output
1006                  * device/source address was requested, then we honour that
1007                  * here
1008                  */
1009                 if (!try_hard) {
1010                         neigh = neigh_lookup_nodev(&dn_neigh_table, &init_net, &fl.fld_dst);
1011                         if (neigh) {
1012                                 if ((oldflp->oif &&
1013                                     (neigh->dev->ifindex != oldflp->oif)) ||
1014                                     (oldflp->fld_src &&
1015                                     (!dn_dev_islocal(neigh->dev,
1016                                                       oldflp->fld_src)))) {
1017                                         neigh_release(neigh);
1018                                         neigh = NULL;
1019                                 } else {
1020                                         if (dev_out)
1021                                                 dev_put(dev_out);
1022                                         if (dn_dev_islocal(neigh->dev, fl.fld_dst)) {
1023                                                 dev_out = init_net.loopback_dev;
1024                                                 res.type = RTN_LOCAL;
1025                                         } else {
1026                                                 dev_out = neigh->dev;
1027                                         }
1028                                         dev_hold(dev_out);
1029                                         goto select_source;
1030                                 }
1031                         }
1032                 }
1033
1034                 /* Not there? Perhaps its a local address */
1035                 if (dev_out == NULL)
1036                         dev_out = dn_dev_get_default();
1037                 err = -ENODEV;
1038                 if (dev_out == NULL)
1039                         goto out;
1040                 dn_db = rcu_dereference_raw(dev_out->dn_ptr);
1041                 /* Possible improvement - check all devices for local addr */
1042                 if (dn_dev_islocal(dev_out, fl.fld_dst)) {
1043                         dev_put(dev_out);
1044                         dev_out = init_net.loopback_dev;
1045                         dev_hold(dev_out);
1046                         res.type = RTN_LOCAL;
1047                         goto select_source;
1048                 }
1049                 /* Not local either.... try sending it to the default router */
1050                 neigh = neigh_clone(dn_db->router);
1051                 BUG_ON(neigh && neigh->dev != dev_out);
1052
1053                 /* Ok then, we assume its directly connected and move on */
1054 select_source:
1055                 if (neigh)
1056                         gateway = ((struct dn_neigh *)neigh)->addr;
1057                 if (gateway == 0)
1058                         gateway = fl.fld_dst;
1059                 if (fl.fld_src == 0) {
1060                         fl.fld_src = dnet_select_source(dev_out, gateway,
1061                                                          res.type == RTN_LOCAL ?
1062                                                          RT_SCOPE_HOST :
1063                                                          RT_SCOPE_LINK);
1064                         if (fl.fld_src == 0 && res.type != RTN_LOCAL)
1065                                 goto e_addr;
1066                 }
1067                 fl.oif = dev_out->ifindex;
1068                 goto make_route;
1069         }
1070         free_res = 1;
1071
1072         if (res.type == RTN_NAT)
1073                 goto e_inval;
1074
1075         if (res.type == RTN_LOCAL) {
1076                 if (!fl.fld_src)
1077                         fl.fld_src = fl.fld_dst;
1078                 if (dev_out)
1079                         dev_put(dev_out);
1080                 dev_out = init_net.loopback_dev;
1081                 dev_hold(dev_out);
1082                 fl.oif = dev_out->ifindex;
1083                 if (res.fi)
1084                         dn_fib_info_put(res.fi);
1085                 res.fi = NULL;
1086                 goto make_route;
1087         }
1088
1089         if (res.fi->fib_nhs > 1 && fl.oif == 0)
1090                 dn_fib_select_multipath(&fl, &res);
1091
1092         /*
1093          * We could add some logic to deal with default routes here and
1094          * get rid of some of the special casing above.
1095          */
1096
1097         if (!fl.fld_src)
1098                 fl.fld_src = DN_FIB_RES_PREFSRC(res);
1099
1100         if (dev_out)
1101                 dev_put(dev_out);
1102         dev_out = DN_FIB_RES_DEV(res);
1103         dev_hold(dev_out);
1104         fl.oif = dev_out->ifindex;
1105         gateway = DN_FIB_RES_GW(res);
1106
1107 make_route:
1108         if (dev_out->flags & IFF_LOOPBACK)
1109                 flags |= RTCF_LOCAL;
1110
1111         rt = dst_alloc(&dn_dst_ops);
1112         if (rt == NULL)
1113                 goto e_nobufs;
1114
1115         atomic_set(&rt->dst.__refcnt, 1);
1116         rt->dst.flags   = DST_HOST;
1117
1118         rt->fl.fld_src    = oldflp->fld_src;
1119         rt->fl.fld_dst    = oldflp->fld_dst;
1120         rt->fl.oif        = oldflp->oif;
1121         rt->fl.iif        = 0;
1122         rt->fl.mark       = oldflp->mark;
1123
1124         rt->rt_saddr      = fl.fld_src;
1125         rt->rt_daddr      = fl.fld_dst;
1126         rt->rt_gateway    = gateway ? gateway : fl.fld_dst;
1127         rt->rt_local_src  = fl.fld_src;
1128
1129         rt->rt_dst_map    = fl.fld_dst;
1130         rt->rt_src_map    = fl.fld_src;
1131
1132         rt->dst.dev = dev_out;
1133         dev_hold(dev_out);
1134         rt->dst.neighbour = neigh;
1135         neigh = NULL;
1136
1137         rt->dst.lastuse = jiffies;
1138         rt->dst.output  = dn_output;
1139         rt->dst.input   = dn_rt_bug;
1140         rt->rt_flags      = flags;
1141         if (flags & RTCF_LOCAL)
1142                 rt->dst.input = dn_nsp_rx;
1143
1144         err = dn_rt_set_next_hop(rt, &res);
1145         if (err)
1146                 goto e_neighbour;
1147
1148         hash = dn_hash(rt->fl.fld_src, rt->fl.fld_dst);
1149         dn_insert_route(rt, hash, (struct dn_route **)pprt);
1150
1151 done:
1152         if (neigh)
1153                 neigh_release(neigh);
1154         if (free_res)
1155                 dn_fib_res_put(&res);
1156         if (dev_out)
1157                 dev_put(dev_out);
1158 out:
1159         return err;
1160
1161 e_addr:
1162         err = -EADDRNOTAVAIL;
1163         goto done;
1164 e_inval:
1165         err = -EINVAL;
1166         goto done;
1167 e_nobufs:
1168         err = -ENOBUFS;
1169         goto done;
1170 e_neighbour:
1171         dst_free(&rt->dst);
1172         goto e_nobufs;
1173 }
1174
1175
1176 /*
1177  * N.B. The flags may be moved into the flowi at some future stage.
1178  */
1179 static int __dn_route_output_key(struct dst_entry **pprt, const struct flowi *flp, int flags)
1180 {
1181         unsigned hash = dn_hash(flp->fld_src, flp->fld_dst);
1182         struct dn_route *rt = NULL;
1183
1184         if (!(flags & MSG_TRYHARD)) {
1185                 rcu_read_lock_bh();
1186                 for (rt = rcu_dereference_bh(dn_rt_hash_table[hash].chain); rt;
1187                         rt = rcu_dereference_bh(rt->dst.dn_next)) {
1188                         if ((flp->fld_dst == rt->fl.fld_dst) &&
1189                             (flp->fld_src == rt->fl.fld_src) &&
1190                             (flp->mark == rt->fl.mark) &&
1191                             dn_is_output_route(rt) &&
1192                             (rt->fl.oif == flp->oif)) {
1193                                 dst_use(&rt->dst, jiffies);
1194                                 rcu_read_unlock_bh();
1195                                 *pprt = &rt->dst;
1196                                 return 0;
1197                         }
1198                 }
1199                 rcu_read_unlock_bh();
1200         }
1201
1202         return dn_route_output_slow(pprt, flp, flags);
1203 }
1204
1205 static int dn_route_output_key(struct dst_entry **pprt, struct flowi *flp, int flags)
1206 {
1207         int err;
1208
1209         err = __dn_route_output_key(pprt, flp, flags);
1210         if (err == 0 && flp->proto) {
1211                 err = xfrm_lookup(&init_net, pprt, flp, NULL, 0);
1212         }
1213         return err;
1214 }
1215
1216 int dn_route_output_sock(struct dst_entry **pprt, struct flowi *fl, struct sock *sk, int flags)
1217 {
1218         int err;
1219
1220         err = __dn_route_output_key(pprt, fl, flags & MSG_TRYHARD);
1221         if (err == 0 && fl->proto) {
1222                 err = xfrm_lookup(&init_net, pprt, fl, sk,
1223                                  (flags & MSG_DONTWAIT) ? 0 : XFRM_LOOKUP_WAIT);
1224         }
1225         return err;
1226 }
1227
1228 static int dn_route_input_slow(struct sk_buff *skb)
1229 {
1230         struct dn_route *rt = NULL;
1231         struct dn_skb_cb *cb = DN_SKB_CB(skb);
1232         struct net_device *in_dev = skb->dev;
1233         struct net_device *out_dev = NULL;
1234         struct dn_dev *dn_db;
1235         struct neighbour *neigh = NULL;
1236         unsigned hash;
1237         int flags = 0;
1238         __le16 gateway = 0;
1239         __le16 local_src = 0;
1240         struct flowi fl = { .fld_dst = cb->dst,
1241                             .fld_src = cb->src,
1242                             .fld_scope = RT_SCOPE_UNIVERSE,
1243                             .mark = skb->mark,
1244                             .iif = skb->dev->ifindex };
1245         struct dn_fib_res res = { .fi = NULL, .type = RTN_UNREACHABLE };
1246         int err = -EINVAL;
1247         int free_res = 0;
1248
1249         dev_hold(in_dev);
1250
1251         if ((dn_db = rcu_dereference(in_dev->dn_ptr)) == NULL)
1252                 goto out;
1253
1254         /* Zero source addresses are not allowed */
1255         if (fl.fld_src == 0)
1256                 goto out;
1257
1258         /*
1259          * In this case we've just received a packet from a source
1260          * outside ourselves pretending to come from us. We don't
1261          * allow it any further to prevent routing loops, spoofing and
1262          * other nasties. Loopback packets already have the dst attached
1263          * so this only affects packets which have originated elsewhere.
1264          */
1265         err  = -ENOTUNIQ;
1266         if (dn_dev_islocal(in_dev, cb->src))
1267                 goto out;
1268
1269         err = dn_fib_lookup(&fl, &res);
1270         if (err) {
1271                 if (err != -ESRCH)
1272                         goto out;
1273                 /*
1274                  * Is the destination us ?
1275                  */
1276                 if (!dn_dev_islocal(in_dev, cb->dst))
1277                         goto e_inval;
1278
1279                 res.type = RTN_LOCAL;
1280         } else {
1281                 __le16 src_map = fl.fld_src;
1282                 free_res = 1;
1283
1284                 out_dev = DN_FIB_RES_DEV(res);
1285                 if (out_dev == NULL) {
1286                         if (net_ratelimit())
1287                                 printk(KERN_CRIT "Bug in dn_route_input_slow() "
1288                                                  "No output device\n");
1289                         goto e_inval;
1290                 }
1291                 dev_hold(out_dev);
1292
1293                 if (res.r)
1294                         src_map = fl.fld_src; /* no NAT support for now */
1295
1296                 gateway = DN_FIB_RES_GW(res);
1297                 if (res.type == RTN_NAT) {
1298                         fl.fld_dst = dn_fib_rules_map_destination(fl.fld_dst, &res);
1299                         dn_fib_res_put(&res);
1300                         free_res = 0;
1301                         if (dn_fib_lookup(&fl, &res))
1302                                 goto e_inval;
1303                         free_res = 1;
1304                         if (res.type != RTN_UNICAST)
1305                                 goto e_inval;
1306                         flags |= RTCF_DNAT;
1307                         gateway = fl.fld_dst;
1308                 }
1309                 fl.fld_src = src_map;
1310         }
1311
1312         switch(res.type) {
1313         case RTN_UNICAST:
1314                 /*
1315                  * Forwarding check here, we only check for forwarding
1316                  * being turned off, if you want to only forward intra
1317                  * area, its up to you to set the routing tables up
1318                  * correctly.
1319                  */
1320                 if (dn_db->parms.forwarding == 0)
1321                         goto e_inval;
1322
1323                 if (res.fi->fib_nhs > 1 && fl.oif == 0)
1324                         dn_fib_select_multipath(&fl, &res);
1325
1326                 /*
1327                  * Check for out_dev == in_dev. We use the RTCF_DOREDIRECT
1328                  * flag as a hint to set the intra-ethernet bit when
1329                  * forwarding. If we've got NAT in operation, we don't do
1330                  * this optimisation.
1331                  */
1332                 if (out_dev == in_dev && !(flags & RTCF_NAT))
1333                         flags |= RTCF_DOREDIRECT;
1334
1335                 local_src = DN_FIB_RES_PREFSRC(res);
1336
1337         case RTN_BLACKHOLE:
1338         case RTN_UNREACHABLE:
1339                 break;
1340         case RTN_LOCAL:
1341                 flags |= RTCF_LOCAL;
1342                 fl.fld_src = cb->dst;
1343                 fl.fld_dst = cb->src;
1344
1345                 /* Routing tables gave us a gateway */
1346                 if (gateway)
1347                         goto make_route;
1348
1349                 /* Packet was intra-ethernet, so we know its on-link */
1350                 if (cb->rt_flags & DN_RT_F_IE) {
1351                         gateway = cb->src;
1352                         flags |= RTCF_DIRECTSRC;
1353                         goto make_route;
1354                 }
1355
1356                 /* Use the default router if there is one */
1357                 neigh = neigh_clone(dn_db->router);
1358                 if (neigh) {
1359                         gateway = ((struct dn_neigh *)neigh)->addr;
1360                         goto make_route;
1361                 }
1362
1363                 /* Close eyes and pray */
1364                 gateway = cb->src;
1365                 flags |= RTCF_DIRECTSRC;
1366                 goto make_route;
1367         default:
1368                 goto e_inval;
1369         }
1370
1371 make_route:
1372         rt = dst_alloc(&dn_dst_ops);
1373         if (rt == NULL)
1374                 goto e_nobufs;
1375
1376         rt->rt_saddr      = fl.fld_src;
1377         rt->rt_daddr      = fl.fld_dst;
1378         rt->rt_gateway    = fl.fld_dst;
1379         if (gateway)
1380                 rt->rt_gateway = gateway;
1381         rt->rt_local_src  = local_src ? local_src : rt->rt_saddr;
1382
1383         rt->rt_dst_map    = fl.fld_dst;
1384         rt->rt_src_map    = fl.fld_src;
1385
1386         rt->fl.fld_src    = cb->src;
1387         rt->fl.fld_dst    = cb->dst;
1388         rt->fl.oif        = 0;
1389         rt->fl.iif        = in_dev->ifindex;
1390         rt->fl.mark       = fl.mark;
1391
1392         rt->dst.flags = DST_HOST;
1393         rt->dst.neighbour = neigh;
1394         rt->dst.dev = out_dev;
1395         rt->dst.lastuse = jiffies;
1396         rt->dst.output = dn_rt_bug;
1397         switch(res.type) {
1398                 case RTN_UNICAST:
1399                         rt->dst.input = dn_forward;
1400                         break;
1401                 case RTN_LOCAL:
1402                         rt->dst.output = dn_output;
1403                         rt->dst.input = dn_nsp_rx;
1404                         rt->dst.dev = in_dev;
1405                         flags |= RTCF_LOCAL;
1406                         break;
1407                 default:
1408                 case RTN_UNREACHABLE:
1409                 case RTN_BLACKHOLE:
1410                         rt->dst.input = dst_discard;
1411         }
1412         rt->rt_flags = flags;
1413         if (rt->dst.dev)
1414                 dev_hold(rt->dst.dev);
1415
1416         err = dn_rt_set_next_hop(rt, &res);
1417         if (err)
1418                 goto e_neighbour;
1419
1420         hash = dn_hash(rt->fl.fld_src, rt->fl.fld_dst);
1421         dn_insert_route(rt, hash, &rt);
1422         skb_dst_set(skb, &rt->dst);
1423
1424 done:
1425         if (neigh)
1426                 neigh_release(neigh);
1427         if (free_res)
1428                 dn_fib_res_put(&res);
1429         dev_put(in_dev);
1430         if (out_dev)
1431                 dev_put(out_dev);
1432 out:
1433         return err;
1434
1435 e_inval:
1436         err = -EINVAL;
1437         goto done;
1438
1439 e_nobufs:
1440         err = -ENOBUFS;
1441         goto done;
1442
1443 e_neighbour:
1444         dst_free(&rt->dst);
1445         goto done;
1446 }
1447
1448 static int dn_route_input(struct sk_buff *skb)
1449 {
1450         struct dn_route *rt;
1451         struct dn_skb_cb *cb = DN_SKB_CB(skb);
1452         unsigned hash = dn_hash(cb->src, cb->dst);
1453
1454         if (skb_dst(skb))
1455                 return 0;
1456
1457         rcu_read_lock();
1458         for(rt = rcu_dereference(dn_rt_hash_table[hash].chain); rt != NULL;
1459             rt = rcu_dereference(rt->dst.dn_next)) {
1460                 if ((rt->fl.fld_src == cb->src) &&
1461                     (rt->fl.fld_dst == cb->dst) &&
1462                     (rt->fl.oif == 0) &&
1463                     (rt->fl.mark == skb->mark) &&
1464                     (rt->fl.iif == cb->iif)) {
1465                         dst_use(&rt->dst, jiffies);
1466                         rcu_read_unlock();
1467                         skb_dst_set(skb, (struct dst_entry *)rt);
1468                         return 0;
1469                 }
1470         }
1471         rcu_read_unlock();
1472
1473         return dn_route_input_slow(skb);
1474 }
1475
1476 static int dn_rt_fill_info(struct sk_buff *skb, u32 pid, u32 seq,
1477                            int event, int nowait, unsigned int flags)
1478 {
1479         struct dn_route *rt = (struct dn_route *)skb_dst(skb);
1480         struct rtmsg *r;
1481         struct nlmsghdr *nlh;
1482         unsigned char *b = skb_tail_pointer(skb);
1483         long expires;
1484
1485         nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*r), flags);
1486         r = NLMSG_DATA(nlh);
1487         r->rtm_family = AF_DECnet;
1488         r->rtm_dst_len = 16;
1489         r->rtm_src_len = 0;
1490         r->rtm_tos = 0;
1491         r->rtm_table = RT_TABLE_MAIN;
1492         RTA_PUT_U32(skb, RTA_TABLE, RT_TABLE_MAIN);
1493         r->rtm_type = rt->rt_type;
1494         r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
1495         r->rtm_scope = RT_SCOPE_UNIVERSE;
1496         r->rtm_protocol = RTPROT_UNSPEC;
1497         if (rt->rt_flags & RTCF_NOTIFY)
1498                 r->rtm_flags |= RTM_F_NOTIFY;
1499         RTA_PUT(skb, RTA_DST, 2, &rt->rt_daddr);
1500         if (rt->fl.fld_src) {
1501                 r->rtm_src_len = 16;
1502                 RTA_PUT(skb, RTA_SRC, 2, &rt->fl.fld_src);
1503         }
1504         if (rt->dst.dev)
1505                 RTA_PUT(skb, RTA_OIF, sizeof(int), &rt->dst.dev->ifindex);
1506         /*
1507          * Note to self - change this if input routes reverse direction when
1508          * they deal only with inputs and not with replies like they do
1509          * currently.
1510          */
1511         RTA_PUT(skb, RTA_PREFSRC, 2, &rt->rt_local_src);
1512         if (rt->rt_daddr != rt->rt_gateway)
1513                 RTA_PUT(skb, RTA_GATEWAY, 2, &rt->rt_gateway);
1514         if (rtnetlink_put_metrics(skb, dst_metrics_ptr(&rt->dst)) < 0)
1515                 goto rtattr_failure;
1516         expires = rt->dst.expires ? rt->dst.expires - jiffies : 0;
1517         if (rtnl_put_cacheinfo(skb, &rt->dst, 0, 0, 0, expires,
1518                                rt->dst.error) < 0)
1519                 goto rtattr_failure;
1520         if (dn_is_input_route(rt))
1521                 RTA_PUT(skb, RTA_IIF, sizeof(int), &rt->fl.iif);
1522
1523         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1524         return skb->len;
1525
1526 nlmsg_failure:
1527 rtattr_failure:
1528         nlmsg_trim(skb, b);
1529         return -1;
1530 }
1531
1532 /*
1533  * This is called by both endnodes and routers now.
1534  */
1535 static int dn_cache_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh, void *arg)
1536 {
1537         struct net *net = sock_net(in_skb->sk);
1538         struct rtattr **rta = arg;
1539         struct rtmsg *rtm = NLMSG_DATA(nlh);
1540         struct dn_route *rt = NULL;
1541         struct dn_skb_cb *cb;
1542         int err;
1543         struct sk_buff *skb;
1544         struct flowi fl;
1545
1546         if (!net_eq(net, &init_net))
1547                 return -EINVAL;
1548
1549         memset(&fl, 0, sizeof(fl));
1550         fl.proto = DNPROTO_NSP;
1551
1552         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1553         if (skb == NULL)
1554                 return -ENOBUFS;
1555         skb_reset_mac_header(skb);
1556         cb = DN_SKB_CB(skb);
1557
1558         if (rta[RTA_SRC-1])
1559                 memcpy(&fl.fld_src, RTA_DATA(rta[RTA_SRC-1]), 2);
1560         if (rta[RTA_DST-1])
1561                 memcpy(&fl.fld_dst, RTA_DATA(rta[RTA_DST-1]), 2);
1562         if (rta[RTA_IIF-1])
1563                 memcpy(&fl.iif, RTA_DATA(rta[RTA_IIF-1]), sizeof(int));
1564
1565         if (fl.iif) {
1566                 struct net_device *dev;
1567                 if ((dev = dev_get_by_index(&init_net, fl.iif)) == NULL) {
1568                         kfree_skb(skb);
1569                         return -ENODEV;
1570                 }
1571                 if (!dev->dn_ptr) {
1572                         dev_put(dev);
1573                         kfree_skb(skb);
1574                         return -ENODEV;
1575                 }
1576                 skb->protocol = htons(ETH_P_DNA_RT);
1577                 skb->dev = dev;
1578                 cb->src = fl.fld_src;
1579                 cb->dst = fl.fld_dst;
1580                 local_bh_disable();
1581                 err = dn_route_input(skb);
1582                 local_bh_enable();
1583                 memset(cb, 0, sizeof(struct dn_skb_cb));
1584                 rt = (struct dn_route *)skb_dst(skb);
1585                 if (!err && -rt->dst.error)
1586                         err = rt->dst.error;
1587         } else {
1588                 int oif = 0;
1589                 if (rta[RTA_OIF - 1])
1590                         memcpy(&oif, RTA_DATA(rta[RTA_OIF - 1]), sizeof(int));
1591                 fl.oif = oif;
1592                 err = dn_route_output_key((struct dst_entry **)&rt, &fl, 0);
1593         }
1594
1595         if (skb->dev)
1596                 dev_put(skb->dev);
1597         skb->dev = NULL;
1598         if (err)
1599                 goto out_free;
1600         skb_dst_set(skb, &rt->dst);
1601         if (rtm->rtm_flags & RTM_F_NOTIFY)
1602                 rt->rt_flags |= RTCF_NOTIFY;
1603
1604         err = dn_rt_fill_info(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq, RTM_NEWROUTE, 0, 0);
1605
1606         if (err == 0)
1607                 goto out_free;
1608         if (err < 0) {
1609                 err = -EMSGSIZE;
1610                 goto out_free;
1611         }
1612
1613         return rtnl_unicast(skb, &init_net, NETLINK_CB(in_skb).pid);
1614
1615 out_free:
1616         kfree_skb(skb);
1617         return err;
1618 }
1619
1620 /*
1621  * For routers, this is called from dn_fib_dump, but for endnodes its
1622  * called directly from the rtnetlink dispatch table.
1623  */
1624 int dn_cache_dump(struct sk_buff *skb, struct netlink_callback *cb)
1625 {
1626         struct net *net = sock_net(skb->sk);
1627         struct dn_route *rt;
1628         int h, s_h;
1629         int idx, s_idx;
1630
1631         if (!net_eq(net, &init_net))
1632                 return 0;
1633
1634         if (NLMSG_PAYLOAD(cb->nlh, 0) < sizeof(struct rtmsg))
1635                 return -EINVAL;
1636         if (!(((struct rtmsg *)NLMSG_DATA(cb->nlh))->rtm_flags&RTM_F_CLONED))
1637                 return 0;
1638
1639         s_h = cb->args[0];
1640         s_idx = idx = cb->args[1];
1641         for(h = 0; h <= dn_rt_hash_mask; h++) {
1642                 if (h < s_h)
1643                         continue;
1644                 if (h > s_h)
1645                         s_idx = 0;
1646                 rcu_read_lock_bh();
1647                 for(rt = rcu_dereference_bh(dn_rt_hash_table[h].chain), idx = 0;
1648                         rt;
1649                         rt = rcu_dereference_bh(rt->dst.dn_next), idx++) {
1650                         if (idx < s_idx)
1651                                 continue;
1652                         skb_dst_set(skb, dst_clone(&rt->dst));
1653                         if (dn_rt_fill_info(skb, NETLINK_CB(cb->skb).pid,
1654                                         cb->nlh->nlmsg_seq, RTM_NEWROUTE,
1655                                         1, NLM_F_MULTI) <= 0) {
1656                                 skb_dst_drop(skb);
1657                                 rcu_read_unlock_bh();
1658                                 goto done;
1659                         }
1660                         skb_dst_drop(skb);
1661                 }
1662                 rcu_read_unlock_bh();
1663         }
1664
1665 done:
1666         cb->args[0] = h;
1667         cb->args[1] = idx;
1668         return skb->len;
1669 }
1670
1671 #ifdef CONFIG_PROC_FS
1672 struct dn_rt_cache_iter_state {
1673         int bucket;
1674 };
1675
1676 static struct dn_route *dn_rt_cache_get_first(struct seq_file *seq)
1677 {
1678         struct dn_route *rt = NULL;
1679         struct dn_rt_cache_iter_state *s = seq->private;
1680
1681         for(s->bucket = dn_rt_hash_mask; s->bucket >= 0; --s->bucket) {
1682                 rcu_read_lock_bh();
1683                 rt = rcu_dereference_bh(dn_rt_hash_table[s->bucket].chain);
1684                 if (rt)
1685                         break;
1686                 rcu_read_unlock_bh();
1687         }
1688         return rt;
1689 }
1690
1691 static struct dn_route *dn_rt_cache_get_next(struct seq_file *seq, struct dn_route *rt)
1692 {
1693         struct dn_rt_cache_iter_state *s = seq->private;
1694
1695         rt = rcu_dereference_bh(rt->dst.dn_next);
1696         while (!rt) {
1697                 rcu_read_unlock_bh();
1698                 if (--s->bucket < 0)
1699                         break;
1700                 rcu_read_lock_bh();
1701                 rt = rcu_dereference_bh(dn_rt_hash_table[s->bucket].chain);
1702         }
1703         return rt;
1704 }
1705
1706 static void *dn_rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
1707 {
1708         struct dn_route *rt = dn_rt_cache_get_first(seq);
1709
1710         if (rt) {
1711                 while(*pos && (rt = dn_rt_cache_get_next(seq, rt)))
1712                         --*pos;
1713         }
1714         return *pos ? NULL : rt;
1715 }
1716
1717 static void *dn_rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1718 {
1719         struct dn_route *rt = dn_rt_cache_get_next(seq, v);
1720         ++*pos;
1721         return rt;
1722 }
1723
1724 static void dn_rt_cache_seq_stop(struct seq_file *seq, void *v)
1725 {
1726         if (v)
1727                 rcu_read_unlock_bh();
1728 }
1729
1730 static int dn_rt_cache_seq_show(struct seq_file *seq, void *v)
1731 {
1732         struct dn_route *rt = v;
1733         char buf1[DN_ASCBUF_LEN], buf2[DN_ASCBUF_LEN];
1734
1735         seq_printf(seq, "%-8s %-7s %-7s %04d %04d %04d\n",
1736                         rt->dst.dev ? rt->dst.dev->name : "*",
1737                         dn_addr2asc(le16_to_cpu(rt->rt_daddr), buf1),
1738                         dn_addr2asc(le16_to_cpu(rt->rt_saddr), buf2),
1739                         atomic_read(&rt->dst.__refcnt),
1740                         rt->dst.__use,
1741                         (int) dst_metric(&rt->dst, RTAX_RTT));
1742         return 0;
1743 }
1744
1745 static const struct seq_operations dn_rt_cache_seq_ops = {
1746         .start  = dn_rt_cache_seq_start,
1747         .next   = dn_rt_cache_seq_next,
1748         .stop   = dn_rt_cache_seq_stop,
1749         .show   = dn_rt_cache_seq_show,
1750 };
1751
1752 static int dn_rt_cache_seq_open(struct inode *inode, struct file *file)
1753 {
1754         return seq_open_private(file, &dn_rt_cache_seq_ops,
1755                         sizeof(struct dn_rt_cache_iter_state));
1756 }
1757
1758 static const struct file_operations dn_rt_cache_seq_fops = {
1759         .owner   = THIS_MODULE,
1760         .open    = dn_rt_cache_seq_open,
1761         .read    = seq_read,
1762         .llseek  = seq_lseek,
1763         .release = seq_release_private,
1764 };
1765
1766 #endif /* CONFIG_PROC_FS */
1767
1768 void __init dn_route_init(void)
1769 {
1770         int i, goal, order;
1771
1772         dn_dst_ops.kmem_cachep =
1773                 kmem_cache_create("dn_dst_cache", sizeof(struct dn_route), 0,
1774                                   SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1775         dst_entries_init(&dn_dst_ops);
1776         setup_timer(&dn_route_timer, dn_dst_check_expire, 0);
1777         dn_route_timer.expires = jiffies + decnet_dst_gc_interval * HZ;
1778         add_timer(&dn_route_timer);
1779
1780         goal = totalram_pages >> (26 - PAGE_SHIFT);
1781
1782         for(order = 0; (1UL << order) < goal; order++)
1783                 /* NOTHING */;
1784
1785         /*
1786          * Only want 1024 entries max, since the table is very, very unlikely
1787          * to be larger than that.
1788          */
1789         while(order && ((((1UL << order) * PAGE_SIZE) /
1790                                 sizeof(struct dn_rt_hash_bucket)) >= 2048))
1791                 order--;
1792
1793         do {
1794                 dn_rt_hash_mask = (1UL << order) * PAGE_SIZE /
1795                         sizeof(struct dn_rt_hash_bucket);
1796                 while(dn_rt_hash_mask & (dn_rt_hash_mask - 1))
1797                         dn_rt_hash_mask--;
1798                 dn_rt_hash_table = (struct dn_rt_hash_bucket *)
1799                         __get_free_pages(GFP_ATOMIC, order);
1800         } while (dn_rt_hash_table == NULL && --order > 0);
1801
1802         if (!dn_rt_hash_table)
1803                 panic("Failed to allocate DECnet route cache hash table\n");
1804
1805         printk(KERN_INFO
1806                 "DECnet: Routing cache hash table of %u buckets, %ldKbytes\n",
1807                 dn_rt_hash_mask,
1808                 (long)(dn_rt_hash_mask*sizeof(struct dn_rt_hash_bucket))/1024);
1809
1810         dn_rt_hash_mask--;
1811         for(i = 0; i <= dn_rt_hash_mask; i++) {
1812                 spin_lock_init(&dn_rt_hash_table[i].lock);
1813                 dn_rt_hash_table[i].chain = NULL;
1814         }
1815
1816         dn_dst_ops.gc_thresh = (dn_rt_hash_mask + 1);
1817
1818         proc_net_fops_create(&init_net, "decnet_cache", S_IRUGO, &dn_rt_cache_seq_fops);
1819
1820 #ifdef CONFIG_DECNET_ROUTER
1821         rtnl_register(PF_DECnet, RTM_GETROUTE, dn_cache_getroute, dn_fib_dump);
1822 #else
1823         rtnl_register(PF_DECnet, RTM_GETROUTE, dn_cache_getroute,
1824                       dn_cache_dump);
1825 #endif
1826 }
1827
1828 void __exit dn_route_cleanup(void)
1829 {
1830         del_timer(&dn_route_timer);
1831         dn_run_flush(0);
1832
1833         proc_net_remove(&init_net, "decnet_cache");
1834         dst_entries_destroy(&dn_dst_ops);
1835 }
1836