94f4ec036669cc152af1a9ff9639889016235228
[~shefty/rdma-dev.git] / 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 <linux/export.h>
81 #include <asm/errno.h>
82 #include <net/net_namespace.h>
83 #include <net/netlink.h>
84 #include <net/neighbour.h>
85 #include <net/dst.h>
86 #include <net/flow.h>
87 #include <net/fib_rules.h>
88 #include <net/dn.h>
89 #include <net/dn_dev.h>
90 #include <net/dn_nsp.h>
91 #include <net/dn_route.h>
92 #include <net/dn_neigh.h>
93 #include <net/dn_fib.h>
94
95 struct dn_rt_hash_bucket
96 {
97         struct dn_route __rcu *chain;
98         spinlock_t lock;
99 };
100
101 extern struct neigh_table dn_neigh_table;
102
103
104 static unsigned char dn_hiord_addr[6] = {0xAA,0x00,0x04,0x00,0x00,0x00};
105
106 static const int dn_rt_min_delay = 2 * HZ;
107 static const int dn_rt_max_delay = 10 * HZ;
108 static const int dn_rt_mtu_expires = 10 * 60 * HZ;
109
110 static unsigned long dn_rt_deadline;
111
112 static int dn_dst_gc(struct dst_ops *ops);
113 static struct dst_entry *dn_dst_check(struct dst_entry *, __u32);
114 static unsigned int dn_dst_default_advmss(const struct dst_entry *dst);
115 static unsigned int dn_dst_mtu(const struct dst_entry *dst);
116 static void dn_dst_destroy(struct dst_entry *);
117 static struct dst_entry *dn_dst_negative_advice(struct dst_entry *);
118 static void dn_dst_link_failure(struct sk_buff *);
119 static void dn_dst_update_pmtu(struct dst_entry *dst, u32 mtu);
120 static struct neighbour *dn_dst_neigh_lookup(const struct dst_entry *dst, const void *daddr);
121 static int dn_route_input(struct sk_buff *);
122 static void dn_run_flush(unsigned long dummy);
123
124 static struct dn_rt_hash_bucket *dn_rt_hash_table;
125 static unsigned dn_rt_hash_mask;
126
127 static struct timer_list dn_route_timer;
128 static DEFINE_TIMER(dn_rt_flush_timer, dn_run_flush, 0, 0);
129 int decnet_dst_gc_interval = 2;
130
131 static struct dst_ops dn_dst_ops = {
132         .family =               PF_DECnet,
133         .protocol =             cpu_to_be16(ETH_P_DNA_RT),
134         .gc_thresh =            128,
135         .gc =                   dn_dst_gc,
136         .check =                dn_dst_check,
137         .default_advmss =       dn_dst_default_advmss,
138         .mtu =                  dn_dst_mtu,
139         .cow_metrics =          dst_cow_metrics_generic,
140         .destroy =              dn_dst_destroy,
141         .negative_advice =      dn_dst_negative_advice,
142         .link_failure =         dn_dst_link_failure,
143         .update_pmtu =          dn_dst_update_pmtu,
144         .neigh_lookup =         dn_dst_neigh_lookup,
145 };
146
147 static void dn_dst_destroy(struct dst_entry *dst)
148 {
149         dst_destroy_metrics_generic(dst);
150 }
151
152 static __inline__ unsigned dn_hash(__le16 src, __le16 dst)
153 {
154         __u16 tmp = (__u16 __force)(src ^ dst);
155         tmp ^= (tmp >> 3);
156         tmp ^= (tmp >> 5);
157         tmp ^= (tmp >> 10);
158         return dn_rt_hash_mask & (unsigned)tmp;
159 }
160
161 static inline void dnrt_free(struct dn_route *rt)
162 {
163         call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
164 }
165
166 static inline void dnrt_drop(struct dn_route *rt)
167 {
168         dst_release(&rt->dst);
169         call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
170 }
171
172 static void dn_dst_check_expire(unsigned long dummy)
173 {
174         int i;
175         struct dn_route *rt;
176         struct dn_route __rcu **rtp;
177         unsigned long now = jiffies;
178         unsigned long expire = 120 * HZ;
179
180         for (i = 0; i <= dn_rt_hash_mask; i++) {
181                 rtp = &dn_rt_hash_table[i].chain;
182
183                 spin_lock(&dn_rt_hash_table[i].lock);
184                 while ((rt = rcu_dereference_protected(*rtp,
185                                                 lockdep_is_held(&dn_rt_hash_table[i].lock))) != NULL) {
186                         if (atomic_read(&rt->dst.__refcnt) ||
187                                         (now - rt->dst.lastuse) < expire) {
188                                 rtp = &rt->dst.dn_next;
189                                 continue;
190                         }
191                         *rtp = rt->dst.dn_next;
192                         rt->dst.dn_next = NULL;
193                         dnrt_free(rt);
194                 }
195                 spin_unlock(&dn_rt_hash_table[i].lock);
196
197                 if ((jiffies - now) > 0)
198                         break;
199         }
200
201         mod_timer(&dn_route_timer, now + decnet_dst_gc_interval * HZ);
202 }
203
204 static int dn_dst_gc(struct dst_ops *ops)
205 {
206         struct dn_route *rt;
207         struct dn_route __rcu **rtp;
208         int i;
209         unsigned long now = jiffies;
210         unsigned long expire = 10 * HZ;
211
212         for (i = 0; i <= dn_rt_hash_mask; i++) {
213
214                 spin_lock_bh(&dn_rt_hash_table[i].lock);
215                 rtp = &dn_rt_hash_table[i].chain;
216
217                 while ((rt = rcu_dereference_protected(*rtp,
218                                                 lockdep_is_held(&dn_rt_hash_table[i].lock))) != NULL) {
219                         if (atomic_read(&rt->dst.__refcnt) ||
220                                         (now - rt->dst.lastuse) < expire) {
221                                 rtp = &rt->dst.dn_next;
222                                 continue;
223                         }
224                         *rtp = rt->dst.dn_next;
225                         rt->dst.dn_next = NULL;
226                         dnrt_drop(rt);
227                         break;
228                 }
229                 spin_unlock_bh(&dn_rt_hash_table[i].lock);
230         }
231
232         return 0;
233 }
234
235 /*
236  * The decnet standards don't impose a particular minimum mtu, what they
237  * do insist on is that the routing layer accepts a datagram of at least
238  * 230 bytes long. Here we have to subtract the routing header length from
239  * 230 to get the minimum acceptable mtu. If there is no neighbour, then we
240  * assume the worst and use a long header size.
241  *
242  * We update both the mtu and the advertised mss (i.e. the segment size we
243  * advertise to the other end).
244  */
245 static void dn_dst_update_pmtu(struct dst_entry *dst, u32 mtu)
246 {
247         struct neighbour *n = dst_get_neighbour(dst);
248         u32 min_mtu = 230;
249         struct dn_dev *dn;
250
251         dn = n ? rcu_dereference_raw(n->dev->dn_ptr) : NULL;
252
253         if (dn && dn->use_long == 0)
254                 min_mtu -= 6;
255         else
256                 min_mtu -= 21;
257
258         if (dst_metric(dst, RTAX_MTU) > mtu && mtu >= min_mtu) {
259                 if (!(dst_metric_locked(dst, RTAX_MTU))) {
260                         dst_metric_set(dst, RTAX_MTU, mtu);
261                         dst_set_expires(dst, dn_rt_mtu_expires);
262                 }
263                 if (!(dst_metric_locked(dst, RTAX_ADVMSS))) {
264                         u32 mss = mtu - DN_MAX_NSP_DATA_HEADER;
265                         u32 existing_mss = dst_metric_raw(dst, RTAX_ADVMSS);
266                         if (!existing_mss || existing_mss > mss)
267                                 dst_metric_set(dst, RTAX_ADVMSS, mss);
268                 }
269         }
270 }
271
272 /*
273  * When a route has been marked obsolete. (e.g. routing cache flush)
274  */
275 static struct dst_entry *dn_dst_check(struct dst_entry *dst, __u32 cookie)
276 {
277         return NULL;
278 }
279
280 static struct dst_entry *dn_dst_negative_advice(struct dst_entry *dst)
281 {
282         dst_release(dst);
283         return NULL;
284 }
285
286 static void dn_dst_link_failure(struct sk_buff *skb)
287 {
288 }
289
290 static inline int compare_keys(struct flowidn *fl1, struct flowidn *fl2)
291 {
292         return ((fl1->daddr ^ fl2->daddr) |
293                 (fl1->saddr ^ fl2->saddr) |
294                 (fl1->flowidn_mark ^ fl2->flowidn_mark) |
295                 (fl1->flowidn_scope ^ fl2->flowidn_scope) |
296                 (fl1->flowidn_oif ^ fl2->flowidn_oif) |
297                 (fl1->flowidn_iif ^ fl2->flowidn_iif)) == 0;
298 }
299
300 static int dn_insert_route(struct dn_route *rt, unsigned hash, struct dn_route **rp)
301 {
302         struct dn_route *rth;
303         struct dn_route __rcu **rthp;
304         unsigned long now = jiffies;
305
306         rthp = &dn_rt_hash_table[hash].chain;
307
308         spin_lock_bh(&dn_rt_hash_table[hash].lock);
309         while ((rth = rcu_dereference_protected(*rthp,
310                                                 lockdep_is_held(&dn_rt_hash_table[hash].lock))) != NULL) {
311                 if (compare_keys(&rth->fld, &rt->fld)) {
312                         /* Put it first */
313                         *rthp = rth->dst.dn_next;
314                         rcu_assign_pointer(rth->dst.dn_next,
315                                            dn_rt_hash_table[hash].chain);
316                         rcu_assign_pointer(dn_rt_hash_table[hash].chain, rth);
317
318                         dst_use(&rth->dst, now);
319                         spin_unlock_bh(&dn_rt_hash_table[hash].lock);
320
321                         dnrt_drop(rt);
322                         *rp = rth;
323                         return 0;
324                 }
325                 rthp = &rth->dst.dn_next;
326         }
327
328         rcu_assign_pointer(rt->dst.dn_next, dn_rt_hash_table[hash].chain);
329         rcu_assign_pointer(dn_rt_hash_table[hash].chain, rt);
330
331         dst_use(&rt->dst, now);
332         spin_unlock_bh(&dn_rt_hash_table[hash].lock);
333         *rp = rt;
334         return 0;
335 }
336
337 static void dn_run_flush(unsigned long dummy)
338 {
339         int i;
340         struct dn_route *rt, *next;
341
342         for (i = 0; i < dn_rt_hash_mask; i++) {
343                 spin_lock_bh(&dn_rt_hash_table[i].lock);
344
345                 if ((rt = xchg((struct dn_route **)&dn_rt_hash_table[i].chain, NULL)) == NULL)
346                         goto nothing_to_declare;
347
348                 for(; rt; rt = next) {
349                         next = rcu_dereference_raw(rt->dst.dn_next);
350                         RCU_INIT_POINTER(rt->dst.dn_next, NULL);
351                         dst_free((struct dst_entry *)rt);
352                 }
353
354 nothing_to_declare:
355                 spin_unlock_bh(&dn_rt_hash_table[i].lock);
356         }
357 }
358
359 static DEFINE_SPINLOCK(dn_rt_flush_lock);
360
361 void dn_rt_cache_flush(int delay)
362 {
363         unsigned long now = jiffies;
364         int user_mode = !in_interrupt();
365
366         if (delay < 0)
367                 delay = dn_rt_min_delay;
368
369         spin_lock_bh(&dn_rt_flush_lock);
370
371         if (del_timer(&dn_rt_flush_timer) && delay > 0 && dn_rt_deadline) {
372                 long tmo = (long)(dn_rt_deadline - now);
373
374                 if (user_mode && tmo < dn_rt_max_delay - dn_rt_min_delay)
375                         tmo = 0;
376
377                 if (delay > tmo)
378                         delay = tmo;
379         }
380
381         if (delay <= 0) {
382                 spin_unlock_bh(&dn_rt_flush_lock);
383                 dn_run_flush(0);
384                 return;
385         }
386
387         if (dn_rt_deadline == 0)
388                 dn_rt_deadline = now + dn_rt_max_delay;
389
390         dn_rt_flush_timer.expires = now + delay;
391         add_timer(&dn_rt_flush_timer);
392         spin_unlock_bh(&dn_rt_flush_lock);
393 }
394
395 /**
396  * dn_return_short - Return a short packet to its sender
397  * @skb: The packet to return
398  *
399  */
400 static int dn_return_short(struct sk_buff *skb)
401 {
402         struct dn_skb_cb *cb;
403         unsigned char *ptr;
404         __le16 *src;
405         __le16 *dst;
406
407         /* Add back headers */
408         skb_push(skb, skb->data - skb_network_header(skb));
409
410         if ((skb = skb_unshare(skb, GFP_ATOMIC)) == NULL)
411                 return NET_RX_DROP;
412
413         cb = DN_SKB_CB(skb);
414         /* Skip packet length and point to flags */
415         ptr = skb->data + 2;
416         *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
417
418         dst = (__le16 *)ptr;
419         ptr += 2;
420         src = (__le16 *)ptr;
421         ptr += 2;
422         *ptr = 0; /* Zero hop count */
423
424         swap(*src, *dst);
425
426         skb->pkt_type = PACKET_OUTGOING;
427         dn_rt_finish_output(skb, NULL, NULL);
428         return NET_RX_SUCCESS;
429 }
430
431 /**
432  * dn_return_long - Return a long packet to its sender
433  * @skb: The long format packet to return
434  *
435  */
436 static int dn_return_long(struct sk_buff *skb)
437 {
438         struct dn_skb_cb *cb;
439         unsigned char *ptr;
440         unsigned char *src_addr, *dst_addr;
441         unsigned char tmp[ETH_ALEN];
442
443         /* Add back all headers */
444         skb_push(skb, skb->data - skb_network_header(skb));
445
446         if ((skb = skb_unshare(skb, GFP_ATOMIC)) == NULL)
447                 return NET_RX_DROP;
448
449         cb = DN_SKB_CB(skb);
450         /* Ignore packet length and point to flags */
451         ptr = skb->data + 2;
452
453         /* Skip padding */
454         if (*ptr & DN_RT_F_PF) {
455                 char padlen = (*ptr & ~DN_RT_F_PF);
456                 ptr += padlen;
457         }
458
459         *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
460         ptr += 2;
461         dst_addr = ptr;
462         ptr += 8;
463         src_addr = ptr;
464         ptr += 6;
465         *ptr = 0; /* Zero hop count */
466
467         /* Swap source and destination */
468         memcpy(tmp, src_addr, ETH_ALEN);
469         memcpy(src_addr, dst_addr, ETH_ALEN);
470         memcpy(dst_addr, tmp, ETH_ALEN);
471
472         skb->pkt_type = PACKET_OUTGOING;
473         dn_rt_finish_output(skb, dst_addr, src_addr);
474         return NET_RX_SUCCESS;
475 }
476
477 /**
478  * dn_route_rx_packet - Try and find a route for an incoming packet
479  * @skb: The packet to find a route for
480  *
481  * Returns: result of input function if route is found, error code otherwise
482  */
483 static int dn_route_rx_packet(struct sk_buff *skb)
484 {
485         struct dn_skb_cb *cb;
486         int err;
487
488         if ((err = dn_route_input(skb)) == 0)
489                 return dst_input(skb);
490
491         cb = DN_SKB_CB(skb);
492         if (decnet_debug_level & 4) {
493                 char *devname = skb->dev ? skb->dev->name : "???";
494
495                 printk(KERN_DEBUG
496                         "DECnet: dn_route_rx_packet: rt_flags=0x%02x dev=%s len=%d src=0x%04hx dst=0x%04hx err=%d type=%d\n",
497                         (int)cb->rt_flags, devname, skb->len,
498                         le16_to_cpu(cb->src), le16_to_cpu(cb->dst),
499                         err, skb->pkt_type);
500         }
501
502         if ((skb->pkt_type == PACKET_HOST) && (cb->rt_flags & DN_RT_F_RQR)) {
503                 switch (cb->rt_flags & DN_RT_PKT_MSK) {
504                 case DN_RT_PKT_SHORT:
505                         return dn_return_short(skb);
506                 case DN_RT_PKT_LONG:
507                         return dn_return_long(skb);
508                 }
509         }
510
511         kfree_skb(skb);
512         return NET_RX_DROP;
513 }
514
515 static int dn_route_rx_long(struct sk_buff *skb)
516 {
517         struct dn_skb_cb *cb = DN_SKB_CB(skb);
518         unsigned char *ptr = skb->data;
519
520         if (!pskb_may_pull(skb, 21)) /* 20 for long header, 1 for shortest nsp */
521                 goto drop_it;
522
523         skb_pull(skb, 20);
524         skb_reset_transport_header(skb);
525
526         /* Destination info */
527         ptr += 2;
528         cb->dst = dn_eth2dn(ptr);
529         if (memcmp(ptr, dn_hiord_addr, 4) != 0)
530                 goto drop_it;
531         ptr += 6;
532
533
534         /* Source info */
535         ptr += 2;
536         cb->src = dn_eth2dn(ptr);
537         if (memcmp(ptr, dn_hiord_addr, 4) != 0)
538                 goto drop_it;
539         ptr += 6;
540         /* Other junk */
541         ptr++;
542         cb->hops = *ptr++; /* Visit Count */
543
544         return NF_HOOK(NFPROTO_DECNET, NF_DN_PRE_ROUTING, skb, skb->dev, NULL,
545                        dn_route_rx_packet);
546
547 drop_it:
548         kfree_skb(skb);
549         return NET_RX_DROP;
550 }
551
552
553
554 static int dn_route_rx_short(struct sk_buff *skb)
555 {
556         struct dn_skb_cb *cb = DN_SKB_CB(skb);
557         unsigned char *ptr = skb->data;
558
559         if (!pskb_may_pull(skb, 6)) /* 5 for short header + 1 for shortest nsp */
560                 goto drop_it;
561
562         skb_pull(skb, 5);
563         skb_reset_transport_header(skb);
564
565         cb->dst = *(__le16 *)ptr;
566         ptr += 2;
567         cb->src = *(__le16 *)ptr;
568         ptr += 2;
569         cb->hops = *ptr & 0x3f;
570
571         return NF_HOOK(NFPROTO_DECNET, NF_DN_PRE_ROUTING, skb, skb->dev, NULL,
572                        dn_route_rx_packet);
573
574 drop_it:
575         kfree_skb(skb);
576         return NET_RX_DROP;
577 }
578
579 static int dn_route_discard(struct sk_buff *skb)
580 {
581         /*
582          * I know we drop the packet here, but thats considered success in
583          * this case
584          */
585         kfree_skb(skb);
586         return NET_RX_SUCCESS;
587 }
588
589 static int dn_route_ptp_hello(struct sk_buff *skb)
590 {
591         dn_dev_hello(skb);
592         dn_neigh_pointopoint_hello(skb);
593         return NET_RX_SUCCESS;
594 }
595
596 int dn_route_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
597 {
598         struct dn_skb_cb *cb;
599         unsigned char flags = 0;
600         __u16 len = le16_to_cpu(*(__le16 *)skb->data);
601         struct dn_dev *dn = rcu_dereference(dev->dn_ptr);
602         unsigned char padlen = 0;
603
604         if (!net_eq(dev_net(dev), &init_net))
605                 goto dump_it;
606
607         if (dn == NULL)
608                 goto dump_it;
609
610         if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
611                 goto out;
612
613         if (!pskb_may_pull(skb, 3))
614                 goto dump_it;
615
616         skb_pull(skb, 2);
617
618         if (len > skb->len)
619                 goto dump_it;
620
621         skb_trim(skb, len);
622
623         flags = *skb->data;
624
625         cb = DN_SKB_CB(skb);
626         cb->stamp = jiffies;
627         cb->iif = dev->ifindex;
628
629         /*
630          * If we have padding, remove it.
631          */
632         if (flags & DN_RT_F_PF) {
633                 padlen = flags & ~DN_RT_F_PF;
634                 if (!pskb_may_pull(skb, padlen + 1))
635                         goto dump_it;
636                 skb_pull(skb, padlen);
637                 flags = *skb->data;
638         }
639
640         skb_reset_network_header(skb);
641
642         /*
643          * Weed out future version DECnet
644          */
645         if (flags & DN_RT_F_VER)
646                 goto dump_it;
647
648         cb->rt_flags = flags;
649
650         if (decnet_debug_level & 1)
651                 printk(KERN_DEBUG
652                         "dn_route_rcv: got 0x%02x from %s [%d %d %d]\n",
653                         (int)flags, (dev) ? dev->name : "???", len, skb->len,
654                         padlen);
655
656         if (flags & DN_RT_PKT_CNTL) {
657                 if (unlikely(skb_linearize(skb)))
658                         goto dump_it;
659
660                 switch (flags & DN_RT_CNTL_MSK) {
661                 case DN_RT_PKT_INIT:
662                         dn_dev_init_pkt(skb);
663                         break;
664                 case DN_RT_PKT_VERI:
665                         dn_dev_veri_pkt(skb);
666                         break;
667                 }
668
669                 if (dn->parms.state != DN_DEV_S_RU)
670                         goto dump_it;
671
672                 switch (flags & DN_RT_CNTL_MSK) {
673                 case DN_RT_PKT_HELO:
674                         return NF_HOOK(NFPROTO_DECNET, NF_DN_HELLO,
675                                        skb, skb->dev, NULL,
676                                        dn_route_ptp_hello);
677
678                 case DN_RT_PKT_L1RT:
679                 case DN_RT_PKT_L2RT:
680                         return NF_HOOK(NFPROTO_DECNET, NF_DN_ROUTE,
681                                        skb, skb->dev, NULL,
682                                        dn_route_discard);
683                 case DN_RT_PKT_ERTH:
684                         return NF_HOOK(NFPROTO_DECNET, NF_DN_HELLO,
685                                        skb, skb->dev, NULL,
686                                        dn_neigh_router_hello);
687
688                 case DN_RT_PKT_EEDH:
689                         return NF_HOOK(NFPROTO_DECNET, NF_DN_HELLO,
690                                        skb, skb->dev, NULL,
691                                        dn_neigh_endnode_hello);
692                 }
693         } else {
694                 if (dn->parms.state != DN_DEV_S_RU)
695                         goto dump_it;
696
697                 skb_pull(skb, 1); /* Pull flags */
698
699                 switch (flags & DN_RT_PKT_MSK) {
700                 case DN_RT_PKT_LONG:
701                         return dn_route_rx_long(skb);
702                 case DN_RT_PKT_SHORT:
703                         return dn_route_rx_short(skb);
704                 }
705         }
706
707 dump_it:
708         kfree_skb(skb);
709 out:
710         return NET_RX_DROP;
711 }
712
713 static int dn_to_neigh_output(struct sk_buff *skb)
714 {
715         struct dst_entry *dst = skb_dst(skb);
716         struct neighbour *n = dst_get_neighbour(dst);
717
718         return n->output(n, skb);
719 }
720
721 static int dn_output(struct sk_buff *skb)
722 {
723         struct dst_entry *dst = skb_dst(skb);
724         struct dn_route *rt = (struct dn_route *)dst;
725         struct net_device *dev = dst->dev;
726         struct dn_skb_cb *cb = DN_SKB_CB(skb);
727         struct neighbour *neigh;
728
729         int err = -EINVAL;
730
731         if ((neigh = dst_get_neighbour(dst)) == NULL)
732                 goto error;
733
734         skb->dev = dev;
735
736         cb->src = rt->rt_saddr;
737         cb->dst = rt->rt_daddr;
738
739         /*
740          * Always set the Intra-Ethernet bit on all outgoing packets
741          * originated on this node. Only valid flag from upper layers
742          * is return-to-sender-requested. Set hop count to 0 too.
743          */
744         cb->rt_flags &= ~DN_RT_F_RQR;
745         cb->rt_flags |= DN_RT_F_IE;
746         cb->hops = 0;
747
748         return NF_HOOK(NFPROTO_DECNET, NF_DN_LOCAL_OUT, skb, NULL, dev,
749                        dn_to_neigh_output);
750
751 error:
752         if (net_ratelimit())
753                 printk(KERN_DEBUG "dn_output: This should not happen\n");
754
755         kfree_skb(skb);
756
757         return err;
758 }
759
760 static int dn_forward(struct sk_buff *skb)
761 {
762         struct dn_skb_cb *cb = DN_SKB_CB(skb);
763         struct dst_entry *dst = skb_dst(skb);
764         struct dn_dev *dn_db = rcu_dereference(dst->dev->dn_ptr);
765         struct dn_route *rt;
766         int header_len;
767 #ifdef CONFIG_NETFILTER
768         struct net_device *dev = skb->dev;
769 #endif
770
771         if (skb->pkt_type != PACKET_HOST)
772                 goto drop;
773
774         /* Ensure that we have enough space for headers */
775         rt = (struct dn_route *)skb_dst(skb);
776         header_len = dn_db->use_long ? 21 : 6;
777         if (skb_cow(skb, LL_RESERVED_SPACE(rt->dst.dev)+header_len))
778                 goto drop;
779
780         /*
781          * Hop count exceeded.
782          */
783         if (++cb->hops > 30)
784                 goto drop;
785
786         skb->dev = rt->dst.dev;
787
788         /*
789          * If packet goes out same interface it came in on, then set
790          * the Intra-Ethernet bit. This has no effect for short
791          * packets, so we don't need to test for them here.
792          */
793         cb->rt_flags &= ~DN_RT_F_IE;
794         if (rt->rt_flags & RTCF_DOREDIRECT)
795                 cb->rt_flags |= DN_RT_F_IE;
796
797         return NF_HOOK(NFPROTO_DECNET, NF_DN_FORWARD, skb, dev, skb->dev,
798                        dn_to_neigh_output);
799
800 drop:
801         kfree_skb(skb);
802         return NET_RX_DROP;
803 }
804
805 /*
806  * Used to catch bugs. This should never normally get
807  * called.
808  */
809 static int dn_rt_bug(struct sk_buff *skb)
810 {
811         if (net_ratelimit()) {
812                 struct dn_skb_cb *cb = DN_SKB_CB(skb);
813
814                 printk(KERN_DEBUG "dn_rt_bug: skb from:%04x to:%04x\n",
815                                 le16_to_cpu(cb->src), le16_to_cpu(cb->dst));
816         }
817
818         kfree_skb(skb);
819
820         return NET_RX_DROP;
821 }
822
823 static unsigned int dn_dst_default_advmss(const struct dst_entry *dst)
824 {
825         return dn_mss_from_pmtu(dst->dev, dst_mtu(dst));
826 }
827
828 static unsigned int dn_dst_mtu(const struct dst_entry *dst)
829 {
830         unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
831
832         return mtu ? : dst->dev->mtu;
833 }
834
835 static struct neighbour *dn_dst_neigh_lookup(const struct dst_entry *dst, const void *daddr)
836 {
837         return __neigh_lookup_errno(&dn_neigh_table, daddr, dst->dev);
838 }
839
840 static int dn_rt_set_next_hop(struct dn_route *rt, struct dn_fib_res *res)
841 {
842         struct dn_fib_info *fi = res->fi;
843         struct net_device *dev = rt->dst.dev;
844         unsigned int mss_metric;
845         struct neighbour *n;
846
847         if (fi) {
848                 if (DN_FIB_RES_GW(*res) &&
849                     DN_FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
850                         rt->rt_gateway = DN_FIB_RES_GW(*res);
851                 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
852         }
853         rt->rt_type = res->type;
854
855         if (dev != NULL && dst_get_neighbour(&rt->dst) == NULL) {
856                 n = __neigh_lookup_errno(&dn_neigh_table, &rt->rt_gateway, dev);
857                 if (IS_ERR(n))
858                         return PTR_ERR(n);
859                 dst_set_neighbour(&rt->dst, n);
860         }
861
862         if (dst_metric(&rt->dst, RTAX_MTU) > rt->dst.dev->mtu)
863                 dst_metric_set(&rt->dst, RTAX_MTU, rt->dst.dev->mtu);
864         mss_metric = dst_metric_raw(&rt->dst, RTAX_ADVMSS);
865         if (mss_metric) {
866                 unsigned int mss = dn_mss_from_pmtu(dev, dst_mtu(&rt->dst));
867                 if (mss_metric > mss)
868                         dst_metric_set(&rt->dst, RTAX_ADVMSS, mss);
869         }
870         return 0;
871 }
872
873 static inline int dn_match_addr(__le16 addr1, __le16 addr2)
874 {
875         __u16 tmp = le16_to_cpu(addr1) ^ le16_to_cpu(addr2);
876         int match = 16;
877         while(tmp) {
878                 tmp >>= 1;
879                 match--;
880         }
881         return match;
882 }
883
884 static __le16 dnet_select_source(const struct net_device *dev, __le16 daddr, int scope)
885 {
886         __le16 saddr = 0;
887         struct dn_dev *dn_db;
888         struct dn_ifaddr *ifa;
889         int best_match = 0;
890         int ret;
891
892         rcu_read_lock();
893         dn_db = rcu_dereference(dev->dn_ptr);
894         for (ifa = rcu_dereference(dn_db->ifa_list);
895              ifa != NULL;
896              ifa = rcu_dereference(ifa->ifa_next)) {
897                 if (ifa->ifa_scope > scope)
898                         continue;
899                 if (!daddr) {
900                         saddr = ifa->ifa_local;
901                         break;
902                 }
903                 ret = dn_match_addr(daddr, ifa->ifa_local);
904                 if (ret > best_match)
905                         saddr = ifa->ifa_local;
906                 if (best_match == 0)
907                         saddr = ifa->ifa_local;
908         }
909         rcu_read_unlock();
910
911         return saddr;
912 }
913
914 static inline __le16 __dn_fib_res_prefsrc(struct dn_fib_res *res)
915 {
916         return dnet_select_source(DN_FIB_RES_DEV(*res), DN_FIB_RES_GW(*res), res->scope);
917 }
918
919 static inline __le16 dn_fib_rules_map_destination(__le16 daddr, struct dn_fib_res *res)
920 {
921         __le16 mask = dnet_make_mask(res->prefixlen);
922         return (daddr&~mask)|res->fi->fib_nh->nh_gw;
923 }
924
925 static int dn_route_output_slow(struct dst_entry **pprt, const struct flowidn *oldflp, int try_hard)
926 {
927         struct flowidn fld = {
928                 .daddr = oldflp->daddr,
929                 .saddr = oldflp->saddr,
930                 .flowidn_scope = RT_SCOPE_UNIVERSE,
931                 .flowidn_mark = oldflp->flowidn_mark,
932                 .flowidn_iif = init_net.loopback_dev->ifindex,
933                 .flowidn_oif = oldflp->flowidn_oif,
934         };
935         struct dn_route *rt = NULL;
936         struct net_device *dev_out = NULL, *dev;
937         struct neighbour *neigh = NULL;
938         unsigned hash;
939         unsigned flags = 0;
940         struct dn_fib_res res = { .fi = NULL, .type = RTN_UNICAST };
941         int err;
942         int free_res = 0;
943         __le16 gateway = 0;
944
945         if (decnet_debug_level & 16)
946                 printk(KERN_DEBUG
947                        "dn_route_output_slow: dst=%04x src=%04x mark=%d"
948                        " iif=%d oif=%d\n", le16_to_cpu(oldflp->daddr),
949                        le16_to_cpu(oldflp->saddr),
950                        oldflp->flowidn_mark, init_net.loopback_dev->ifindex,
951                        oldflp->flowidn_oif);
952
953         /* If we have an output interface, verify its a DECnet device */
954         if (oldflp->flowidn_oif) {
955                 dev_out = dev_get_by_index(&init_net, oldflp->flowidn_oif);
956                 err = -ENODEV;
957                 if (dev_out && dev_out->dn_ptr == NULL) {
958                         dev_put(dev_out);
959                         dev_out = NULL;
960                 }
961                 if (dev_out == NULL)
962                         goto out;
963         }
964
965         /* If we have a source address, verify that its a local address */
966         if (oldflp->saddr) {
967                 err = -EADDRNOTAVAIL;
968
969                 if (dev_out) {
970                         if (dn_dev_islocal(dev_out, oldflp->saddr))
971                                 goto source_ok;
972                         dev_put(dev_out);
973                         goto out;
974                 }
975                 rcu_read_lock();
976                 for_each_netdev_rcu(&init_net, dev) {
977                         if (!dev->dn_ptr)
978                                 continue;
979                         if (!dn_dev_islocal(dev, oldflp->saddr))
980                                 continue;
981                         if ((dev->flags & IFF_LOOPBACK) &&
982                             oldflp->daddr &&
983                             !dn_dev_islocal(dev, oldflp->daddr))
984                                 continue;
985
986                         dev_out = dev;
987                         break;
988                 }
989                 rcu_read_unlock();
990                 if (dev_out == NULL)
991                         goto out;
992                 dev_hold(dev_out);
993 source_ok:
994                 ;
995         }
996
997         /* No destination? Assume its local */
998         if (!fld.daddr) {
999                 fld.daddr = fld.saddr;
1000
1001                 err = -EADDRNOTAVAIL;
1002                 if (dev_out)
1003                         dev_put(dev_out);
1004                 dev_out = init_net.loopback_dev;
1005                 dev_hold(dev_out);
1006                 if (!fld.daddr) {
1007                         fld.daddr =
1008                         fld.saddr = dnet_select_source(dev_out, 0,
1009                                                        RT_SCOPE_HOST);
1010                         if (!fld.daddr)
1011                                 goto out;
1012                 }
1013                 fld.flowidn_oif = init_net.loopback_dev->ifindex;
1014                 res.type = RTN_LOCAL;
1015                 goto make_route;
1016         }
1017
1018         if (decnet_debug_level & 16)
1019                 printk(KERN_DEBUG
1020                        "dn_route_output_slow: initial checks complete."
1021                        " dst=%o4x src=%04x oif=%d try_hard=%d\n",
1022                        le16_to_cpu(fld.daddr), le16_to_cpu(fld.saddr),
1023                        fld.flowidn_oif, try_hard);
1024
1025         /*
1026          * N.B. If the kernel is compiled without router support then
1027          * dn_fib_lookup() will evaluate to non-zero so this if () block
1028          * will always be executed.
1029          */
1030         err = -ESRCH;
1031         if (try_hard || (err = dn_fib_lookup(&fld, &res)) != 0) {
1032                 struct dn_dev *dn_db;
1033                 if (err != -ESRCH)
1034                         goto out;
1035                 /*
1036                  * Here the fallback is basically the standard algorithm for
1037                  * routing in endnodes which is described in the DECnet routing
1038                  * docs
1039                  *
1040                  * If we are not trying hard, look in neighbour cache.
1041                  * The result is tested to ensure that if a specific output
1042                  * device/source address was requested, then we honour that
1043                  * here
1044                  */
1045                 if (!try_hard) {
1046                         neigh = neigh_lookup_nodev(&dn_neigh_table, &init_net, &fld.daddr);
1047                         if (neigh) {
1048                                 if ((oldflp->flowidn_oif &&
1049                                     (neigh->dev->ifindex != oldflp->flowidn_oif)) ||
1050                                     (oldflp->saddr &&
1051                                     (!dn_dev_islocal(neigh->dev,
1052                                                      oldflp->saddr)))) {
1053                                         neigh_release(neigh);
1054                                         neigh = NULL;
1055                                 } else {
1056                                         if (dev_out)
1057                                                 dev_put(dev_out);
1058                                         if (dn_dev_islocal(neigh->dev, fld.daddr)) {
1059                                                 dev_out = init_net.loopback_dev;
1060                                                 res.type = RTN_LOCAL;
1061                                         } else {
1062                                                 dev_out = neigh->dev;
1063                                         }
1064                                         dev_hold(dev_out);
1065                                         goto select_source;
1066                                 }
1067                         }
1068                 }
1069
1070                 /* Not there? Perhaps its a local address */
1071                 if (dev_out == NULL)
1072                         dev_out = dn_dev_get_default();
1073                 err = -ENODEV;
1074                 if (dev_out == NULL)
1075                         goto out;
1076                 dn_db = rcu_dereference_raw(dev_out->dn_ptr);
1077                 /* Possible improvement - check all devices for local addr */
1078                 if (dn_dev_islocal(dev_out, fld.daddr)) {
1079                         dev_put(dev_out);
1080                         dev_out = init_net.loopback_dev;
1081                         dev_hold(dev_out);
1082                         res.type = RTN_LOCAL;
1083                         goto select_source;
1084                 }
1085                 /* Not local either.... try sending it to the default router */
1086                 neigh = neigh_clone(dn_db->router);
1087                 BUG_ON(neigh && neigh->dev != dev_out);
1088
1089                 /* Ok then, we assume its directly connected and move on */
1090 select_source:
1091                 if (neigh)
1092                         gateway = ((struct dn_neigh *)neigh)->addr;
1093                 if (gateway == 0)
1094                         gateway = fld.daddr;
1095                 if (fld.saddr == 0) {
1096                         fld.saddr = dnet_select_source(dev_out, gateway,
1097                                                        res.type == RTN_LOCAL ?
1098                                                        RT_SCOPE_HOST :
1099                                                        RT_SCOPE_LINK);
1100                         if (fld.saddr == 0 && res.type != RTN_LOCAL)
1101                                 goto e_addr;
1102                 }
1103                 fld.flowidn_oif = dev_out->ifindex;
1104                 goto make_route;
1105         }
1106         free_res = 1;
1107
1108         if (res.type == RTN_NAT)
1109                 goto e_inval;
1110
1111         if (res.type == RTN_LOCAL) {
1112                 if (!fld.saddr)
1113                         fld.saddr = fld.daddr;
1114                 if (dev_out)
1115                         dev_put(dev_out);
1116                 dev_out = init_net.loopback_dev;
1117                 dev_hold(dev_out);
1118                 fld.flowidn_oif = dev_out->ifindex;
1119                 if (res.fi)
1120                         dn_fib_info_put(res.fi);
1121                 res.fi = NULL;
1122                 goto make_route;
1123         }
1124
1125         if (res.fi->fib_nhs > 1 && fld.flowidn_oif == 0)
1126                 dn_fib_select_multipath(&fld, &res);
1127
1128         /*
1129          * We could add some logic to deal with default routes here and
1130          * get rid of some of the special casing above.
1131          */
1132
1133         if (!fld.saddr)
1134                 fld.saddr = DN_FIB_RES_PREFSRC(res);
1135
1136         if (dev_out)
1137                 dev_put(dev_out);
1138         dev_out = DN_FIB_RES_DEV(res);
1139         dev_hold(dev_out);
1140         fld.flowidn_oif = dev_out->ifindex;
1141         gateway = DN_FIB_RES_GW(res);
1142
1143 make_route:
1144         if (dev_out->flags & IFF_LOOPBACK)
1145                 flags |= RTCF_LOCAL;
1146
1147         rt = dst_alloc(&dn_dst_ops, dev_out, 1, 0, DST_HOST);
1148         if (rt == NULL)
1149                 goto e_nobufs;
1150
1151         memset(&rt->fld, 0, sizeof(rt->fld));
1152         rt->fld.saddr        = oldflp->saddr;
1153         rt->fld.daddr        = oldflp->daddr;
1154         rt->fld.flowidn_oif  = oldflp->flowidn_oif;
1155         rt->fld.flowidn_iif  = 0;
1156         rt->fld.flowidn_mark = oldflp->flowidn_mark;
1157
1158         rt->rt_saddr      = fld.saddr;
1159         rt->rt_daddr      = fld.daddr;
1160         rt->rt_gateway    = gateway ? gateway : fld.daddr;
1161         rt->rt_local_src  = fld.saddr;
1162
1163         rt->rt_dst_map    = fld.daddr;
1164         rt->rt_src_map    = fld.saddr;
1165
1166         dst_set_neighbour(&rt->dst, neigh);
1167         neigh = NULL;
1168
1169         rt->dst.lastuse = jiffies;
1170         rt->dst.output  = dn_output;
1171         rt->dst.input   = dn_rt_bug;
1172         rt->rt_flags      = flags;
1173         if (flags & RTCF_LOCAL)
1174                 rt->dst.input = dn_nsp_rx;
1175
1176         err = dn_rt_set_next_hop(rt, &res);
1177         if (err)
1178                 goto e_neighbour;
1179
1180         hash = dn_hash(rt->fld.saddr, rt->fld.daddr);
1181         dn_insert_route(rt, hash, (struct dn_route **)pprt);
1182
1183 done:
1184         if (neigh)
1185                 neigh_release(neigh);
1186         if (free_res)
1187                 dn_fib_res_put(&res);
1188         if (dev_out)
1189                 dev_put(dev_out);
1190 out:
1191         return err;
1192
1193 e_addr:
1194         err = -EADDRNOTAVAIL;
1195         goto done;
1196 e_inval:
1197         err = -EINVAL;
1198         goto done;
1199 e_nobufs:
1200         err = -ENOBUFS;
1201         goto done;
1202 e_neighbour:
1203         dst_free(&rt->dst);
1204         goto e_nobufs;
1205 }
1206
1207
1208 /*
1209  * N.B. The flags may be moved into the flowi at some future stage.
1210  */
1211 static int __dn_route_output_key(struct dst_entry **pprt, const struct flowidn *flp, int flags)
1212 {
1213         unsigned hash = dn_hash(flp->saddr, flp->daddr);
1214         struct dn_route *rt = NULL;
1215
1216         if (!(flags & MSG_TRYHARD)) {
1217                 rcu_read_lock_bh();
1218                 for (rt = rcu_dereference_bh(dn_rt_hash_table[hash].chain); rt;
1219                         rt = rcu_dereference_bh(rt->dst.dn_next)) {
1220                         if ((flp->daddr == rt->fld.daddr) &&
1221                             (flp->saddr == rt->fld.saddr) &&
1222                             (flp->flowidn_mark == rt->fld.flowidn_mark) &&
1223                             dn_is_output_route(rt) &&
1224                             (rt->fld.flowidn_oif == flp->flowidn_oif)) {
1225                                 dst_use(&rt->dst, jiffies);
1226                                 rcu_read_unlock_bh();
1227                                 *pprt = &rt->dst;
1228                                 return 0;
1229                         }
1230                 }
1231                 rcu_read_unlock_bh();
1232         }
1233
1234         return dn_route_output_slow(pprt, flp, flags);
1235 }
1236
1237 static int dn_route_output_key(struct dst_entry **pprt, struct flowidn *flp, int flags)
1238 {
1239         int err;
1240
1241         err = __dn_route_output_key(pprt, flp, flags);
1242         if (err == 0 && flp->flowidn_proto) {
1243                 *pprt = xfrm_lookup(&init_net, *pprt,
1244                                     flowidn_to_flowi(flp), NULL, 0);
1245                 if (IS_ERR(*pprt)) {
1246                         err = PTR_ERR(*pprt);
1247                         *pprt = NULL;
1248                 }
1249         }
1250         return err;
1251 }
1252
1253 int dn_route_output_sock(struct dst_entry **pprt, struct flowidn *fl, struct sock *sk, int flags)
1254 {
1255         int err;
1256
1257         err = __dn_route_output_key(pprt, fl, flags & MSG_TRYHARD);
1258         if (err == 0 && fl->flowidn_proto) {
1259                 if (!(flags & MSG_DONTWAIT))
1260                         fl->flowidn_flags |= FLOWI_FLAG_CAN_SLEEP;
1261                 *pprt = xfrm_lookup(&init_net, *pprt,
1262                                     flowidn_to_flowi(fl), sk, 0);
1263                 if (IS_ERR(*pprt)) {
1264                         err = PTR_ERR(*pprt);
1265                         *pprt = NULL;
1266                 }
1267         }
1268         return err;
1269 }
1270
1271 static int dn_route_input_slow(struct sk_buff *skb)
1272 {
1273         struct dn_route *rt = NULL;
1274         struct dn_skb_cb *cb = DN_SKB_CB(skb);
1275         struct net_device *in_dev = skb->dev;
1276         struct net_device *out_dev = NULL;
1277         struct dn_dev *dn_db;
1278         struct neighbour *neigh = NULL;
1279         unsigned hash;
1280         int flags = 0;
1281         __le16 gateway = 0;
1282         __le16 local_src = 0;
1283         struct flowidn fld = {
1284                 .daddr = cb->dst,
1285                 .saddr = cb->src,
1286                 .flowidn_scope = RT_SCOPE_UNIVERSE,
1287                 .flowidn_mark = skb->mark,
1288                 .flowidn_iif = skb->dev->ifindex,
1289         };
1290         struct dn_fib_res res = { .fi = NULL, .type = RTN_UNREACHABLE };
1291         int err = -EINVAL;
1292         int free_res = 0;
1293
1294         dev_hold(in_dev);
1295
1296         if ((dn_db = rcu_dereference(in_dev->dn_ptr)) == NULL)
1297                 goto out;
1298
1299         /* Zero source addresses are not allowed */
1300         if (fld.saddr == 0)
1301                 goto out;
1302
1303         /*
1304          * In this case we've just received a packet from a source
1305          * outside ourselves pretending to come from us. We don't
1306          * allow it any further to prevent routing loops, spoofing and
1307          * other nasties. Loopback packets already have the dst attached
1308          * so this only affects packets which have originated elsewhere.
1309          */
1310         err  = -ENOTUNIQ;
1311         if (dn_dev_islocal(in_dev, cb->src))
1312                 goto out;
1313
1314         err = dn_fib_lookup(&fld, &res);
1315         if (err) {
1316                 if (err != -ESRCH)
1317                         goto out;
1318                 /*
1319                  * Is the destination us ?
1320                  */
1321                 if (!dn_dev_islocal(in_dev, cb->dst))
1322                         goto e_inval;
1323
1324                 res.type = RTN_LOCAL;
1325         } else {
1326                 __le16 src_map = fld.saddr;
1327                 free_res = 1;
1328
1329                 out_dev = DN_FIB_RES_DEV(res);
1330                 if (out_dev == NULL) {
1331                         if (net_ratelimit())
1332                                 printk(KERN_CRIT "Bug in dn_route_input_slow() "
1333                                                  "No output device\n");
1334                         goto e_inval;
1335                 }
1336                 dev_hold(out_dev);
1337
1338                 if (res.r)
1339                         src_map = fld.saddr; /* no NAT support for now */
1340
1341                 gateway = DN_FIB_RES_GW(res);
1342                 if (res.type == RTN_NAT) {
1343                         fld.daddr = dn_fib_rules_map_destination(fld.daddr, &res);
1344                         dn_fib_res_put(&res);
1345                         free_res = 0;
1346                         if (dn_fib_lookup(&fld, &res))
1347                                 goto e_inval;
1348                         free_res = 1;
1349                         if (res.type != RTN_UNICAST)
1350                                 goto e_inval;
1351                         flags |= RTCF_DNAT;
1352                         gateway = fld.daddr;
1353                 }
1354                 fld.saddr = src_map;
1355         }
1356
1357         switch(res.type) {
1358         case RTN_UNICAST:
1359                 /*
1360                  * Forwarding check here, we only check for forwarding
1361                  * being turned off, if you want to only forward intra
1362                  * area, its up to you to set the routing tables up
1363                  * correctly.
1364                  */
1365                 if (dn_db->parms.forwarding == 0)
1366                         goto e_inval;
1367
1368                 if (res.fi->fib_nhs > 1 && fld.flowidn_oif == 0)
1369                         dn_fib_select_multipath(&fld, &res);
1370
1371                 /*
1372                  * Check for out_dev == in_dev. We use the RTCF_DOREDIRECT
1373                  * flag as a hint to set the intra-ethernet bit when
1374                  * forwarding. If we've got NAT in operation, we don't do
1375                  * this optimisation.
1376                  */
1377                 if (out_dev == in_dev && !(flags & RTCF_NAT))
1378                         flags |= RTCF_DOREDIRECT;
1379
1380                 local_src = DN_FIB_RES_PREFSRC(res);
1381
1382         case RTN_BLACKHOLE:
1383         case RTN_UNREACHABLE:
1384                 break;
1385         case RTN_LOCAL:
1386                 flags |= RTCF_LOCAL;
1387                 fld.saddr = cb->dst;
1388                 fld.daddr = cb->src;
1389
1390                 /* Routing tables gave us a gateway */
1391                 if (gateway)
1392                         goto make_route;
1393
1394                 /* Packet was intra-ethernet, so we know its on-link */
1395                 if (cb->rt_flags & DN_RT_F_IE) {
1396                         gateway = cb->src;
1397                         flags |= RTCF_DIRECTSRC;
1398                         goto make_route;
1399                 }
1400
1401                 /* Use the default router if there is one */
1402                 neigh = neigh_clone(dn_db->router);
1403                 if (neigh) {
1404                         gateway = ((struct dn_neigh *)neigh)->addr;
1405                         goto make_route;
1406                 }
1407
1408                 /* Close eyes and pray */
1409                 gateway = cb->src;
1410                 flags |= RTCF_DIRECTSRC;
1411                 goto make_route;
1412         default:
1413                 goto e_inval;
1414         }
1415
1416 make_route:
1417         rt = dst_alloc(&dn_dst_ops, out_dev, 0, 0, DST_HOST);
1418         if (rt == NULL)
1419                 goto e_nobufs;
1420
1421         memset(&rt->fld, 0, sizeof(rt->fld));
1422         rt->rt_saddr      = fld.saddr;
1423         rt->rt_daddr      = fld.daddr;
1424         rt->rt_gateway    = fld.daddr;
1425         if (gateway)
1426                 rt->rt_gateway = gateway;
1427         rt->rt_local_src  = local_src ? local_src : rt->rt_saddr;
1428
1429         rt->rt_dst_map    = fld.daddr;
1430         rt->rt_src_map    = fld.saddr;
1431
1432         rt->fld.saddr        = cb->src;
1433         rt->fld.daddr        = cb->dst;
1434         rt->fld.flowidn_oif  = 0;
1435         rt->fld.flowidn_iif  = in_dev->ifindex;
1436         rt->fld.flowidn_mark = fld.flowidn_mark;
1437
1438         dst_set_neighbour(&rt->dst, neigh);
1439         rt->dst.lastuse = jiffies;
1440         rt->dst.output = dn_rt_bug;
1441         switch (res.type) {
1442         case RTN_UNICAST:
1443                 rt->dst.input = dn_forward;
1444                 break;
1445         case RTN_LOCAL:
1446                 rt->dst.output = dn_output;
1447                 rt->dst.input = dn_nsp_rx;
1448                 rt->dst.dev = in_dev;
1449                 flags |= RTCF_LOCAL;
1450                 break;
1451         default:
1452         case RTN_UNREACHABLE:
1453         case RTN_BLACKHOLE:
1454                 rt->dst.input = dst_discard;
1455         }
1456         rt->rt_flags = flags;
1457
1458         err = dn_rt_set_next_hop(rt, &res);
1459         if (err)
1460                 goto e_neighbour;
1461
1462         hash = dn_hash(rt->fld.saddr, rt->fld.daddr);
1463         dn_insert_route(rt, hash, &rt);
1464         skb_dst_set(skb, &rt->dst);
1465
1466 done:
1467         if (neigh)
1468                 neigh_release(neigh);
1469         if (free_res)
1470                 dn_fib_res_put(&res);
1471         dev_put(in_dev);
1472         if (out_dev)
1473                 dev_put(out_dev);
1474 out:
1475         return err;
1476
1477 e_inval:
1478         err = -EINVAL;
1479         goto done;
1480
1481 e_nobufs:
1482         err = -ENOBUFS;
1483         goto done;
1484
1485 e_neighbour:
1486         dst_free(&rt->dst);
1487         goto done;
1488 }
1489
1490 static int dn_route_input(struct sk_buff *skb)
1491 {
1492         struct dn_route *rt;
1493         struct dn_skb_cb *cb = DN_SKB_CB(skb);
1494         unsigned hash = dn_hash(cb->src, cb->dst);
1495
1496         if (skb_dst(skb))
1497                 return 0;
1498
1499         rcu_read_lock();
1500         for(rt = rcu_dereference(dn_rt_hash_table[hash].chain); rt != NULL;
1501             rt = rcu_dereference(rt->dst.dn_next)) {
1502                 if ((rt->fld.saddr == cb->src) &&
1503                     (rt->fld.daddr == cb->dst) &&
1504                     (rt->fld.flowidn_oif == 0) &&
1505                     (rt->fld.flowidn_mark == skb->mark) &&
1506                     (rt->fld.flowidn_iif == cb->iif)) {
1507                         dst_use(&rt->dst, jiffies);
1508                         rcu_read_unlock();
1509                         skb_dst_set(skb, (struct dst_entry *)rt);
1510                         return 0;
1511                 }
1512         }
1513         rcu_read_unlock();
1514
1515         return dn_route_input_slow(skb);
1516 }
1517
1518 static int dn_rt_fill_info(struct sk_buff *skb, u32 pid, u32 seq,
1519                            int event, int nowait, unsigned int flags)
1520 {
1521         struct dn_route *rt = (struct dn_route *)skb_dst(skb);
1522         struct rtmsg *r;
1523         struct nlmsghdr *nlh;
1524         unsigned char *b = skb_tail_pointer(skb);
1525         long expires;
1526
1527         nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*r), flags);
1528         r = NLMSG_DATA(nlh);
1529         r->rtm_family = AF_DECnet;
1530         r->rtm_dst_len = 16;
1531         r->rtm_src_len = 0;
1532         r->rtm_tos = 0;
1533         r->rtm_table = RT_TABLE_MAIN;
1534         RTA_PUT_U32(skb, RTA_TABLE, RT_TABLE_MAIN);
1535         r->rtm_type = rt->rt_type;
1536         r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
1537         r->rtm_scope = RT_SCOPE_UNIVERSE;
1538         r->rtm_protocol = RTPROT_UNSPEC;
1539         if (rt->rt_flags & RTCF_NOTIFY)
1540                 r->rtm_flags |= RTM_F_NOTIFY;
1541         RTA_PUT(skb, RTA_DST, 2, &rt->rt_daddr);
1542         if (rt->fld.saddr) {
1543                 r->rtm_src_len = 16;
1544                 RTA_PUT(skb, RTA_SRC, 2, &rt->fld.saddr);
1545         }
1546         if (rt->dst.dev)
1547                 RTA_PUT(skb, RTA_OIF, sizeof(int), &rt->dst.dev->ifindex);
1548         /*
1549          * Note to self - change this if input routes reverse direction when
1550          * they deal only with inputs and not with replies like they do
1551          * currently.
1552          */
1553         RTA_PUT(skb, RTA_PREFSRC, 2, &rt->rt_local_src);
1554         if (rt->rt_daddr != rt->rt_gateway)
1555                 RTA_PUT(skb, RTA_GATEWAY, 2, &rt->rt_gateway);
1556         if (rtnetlink_put_metrics(skb, dst_metrics_ptr(&rt->dst)) < 0)
1557                 goto rtattr_failure;
1558         expires = rt->dst.expires ? rt->dst.expires - jiffies : 0;
1559         if (rtnl_put_cacheinfo(skb, &rt->dst, 0, 0, 0, expires,
1560                                rt->dst.error) < 0)
1561                 goto rtattr_failure;
1562         if (dn_is_input_route(rt))
1563                 RTA_PUT(skb, RTA_IIF, sizeof(int), &rt->fld.flowidn_iif);
1564
1565         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1566         return skb->len;
1567
1568 nlmsg_failure:
1569 rtattr_failure:
1570         nlmsg_trim(skb, b);
1571         return -1;
1572 }
1573
1574 /*
1575  * This is called by both endnodes and routers now.
1576  */
1577 static int dn_cache_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh, void *arg)
1578 {
1579         struct net *net = sock_net(in_skb->sk);
1580         struct rtattr **rta = arg;
1581         struct rtmsg *rtm = NLMSG_DATA(nlh);
1582         struct dn_route *rt = NULL;
1583         struct dn_skb_cb *cb;
1584         int err;
1585         struct sk_buff *skb;
1586         struct flowidn fld;
1587
1588         if (!net_eq(net, &init_net))
1589                 return -EINVAL;
1590
1591         memset(&fld, 0, sizeof(fld));
1592         fld.flowidn_proto = DNPROTO_NSP;
1593
1594         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1595         if (skb == NULL)
1596                 return -ENOBUFS;
1597         skb_reset_mac_header(skb);
1598         cb = DN_SKB_CB(skb);
1599
1600         if (rta[RTA_SRC-1])
1601                 memcpy(&fld.saddr, RTA_DATA(rta[RTA_SRC-1]), 2);
1602         if (rta[RTA_DST-1])
1603                 memcpy(&fld.daddr, RTA_DATA(rta[RTA_DST-1]), 2);
1604         if (rta[RTA_IIF-1])
1605                 memcpy(&fld.flowidn_iif, RTA_DATA(rta[RTA_IIF-1]), sizeof(int));
1606
1607         if (fld.flowidn_iif) {
1608                 struct net_device *dev;
1609                 if ((dev = dev_get_by_index(&init_net, fld.flowidn_iif)) == NULL) {
1610                         kfree_skb(skb);
1611                         return -ENODEV;
1612                 }
1613                 if (!dev->dn_ptr) {
1614                         dev_put(dev);
1615                         kfree_skb(skb);
1616                         return -ENODEV;
1617                 }
1618                 skb->protocol = htons(ETH_P_DNA_RT);
1619                 skb->dev = dev;
1620                 cb->src = fld.saddr;
1621                 cb->dst = fld.daddr;
1622                 local_bh_disable();
1623                 err = dn_route_input(skb);
1624                 local_bh_enable();
1625                 memset(cb, 0, sizeof(struct dn_skb_cb));
1626                 rt = (struct dn_route *)skb_dst(skb);
1627                 if (!err && -rt->dst.error)
1628                         err = rt->dst.error;
1629         } else {
1630                 int oif = 0;
1631                 if (rta[RTA_OIF - 1])
1632                         memcpy(&oif, RTA_DATA(rta[RTA_OIF - 1]), sizeof(int));
1633                 fld.flowidn_oif = oif;
1634                 err = dn_route_output_key((struct dst_entry **)&rt, &fld, 0);
1635         }
1636
1637         if (skb->dev)
1638                 dev_put(skb->dev);
1639         skb->dev = NULL;
1640         if (err)
1641                 goto out_free;
1642         skb_dst_set(skb, &rt->dst);
1643         if (rtm->rtm_flags & RTM_F_NOTIFY)
1644                 rt->rt_flags |= RTCF_NOTIFY;
1645
1646         err = dn_rt_fill_info(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq, RTM_NEWROUTE, 0, 0);
1647
1648         if (err == 0)
1649                 goto out_free;
1650         if (err < 0) {
1651                 err = -EMSGSIZE;
1652                 goto out_free;
1653         }
1654
1655         return rtnl_unicast(skb, &init_net, NETLINK_CB(in_skb).pid);
1656
1657 out_free:
1658         kfree_skb(skb);
1659         return err;
1660 }
1661
1662 /*
1663  * For routers, this is called from dn_fib_dump, but for endnodes its
1664  * called directly from the rtnetlink dispatch table.
1665  */
1666 int dn_cache_dump(struct sk_buff *skb, struct netlink_callback *cb)
1667 {
1668         struct net *net = sock_net(skb->sk);
1669         struct dn_route *rt;
1670         int h, s_h;
1671         int idx, s_idx;
1672
1673         if (!net_eq(net, &init_net))
1674                 return 0;
1675
1676         if (NLMSG_PAYLOAD(cb->nlh, 0) < sizeof(struct rtmsg))
1677                 return -EINVAL;
1678         if (!(((struct rtmsg *)NLMSG_DATA(cb->nlh))->rtm_flags&RTM_F_CLONED))
1679                 return 0;
1680
1681         s_h = cb->args[0];
1682         s_idx = idx = cb->args[1];
1683         for(h = 0; h <= dn_rt_hash_mask; h++) {
1684                 if (h < s_h)
1685                         continue;
1686                 if (h > s_h)
1687                         s_idx = 0;
1688                 rcu_read_lock_bh();
1689                 for(rt = rcu_dereference_bh(dn_rt_hash_table[h].chain), idx = 0;
1690                         rt;
1691                         rt = rcu_dereference_bh(rt->dst.dn_next), idx++) {
1692                         if (idx < s_idx)
1693                                 continue;
1694                         skb_dst_set(skb, dst_clone(&rt->dst));
1695                         if (dn_rt_fill_info(skb, NETLINK_CB(cb->skb).pid,
1696                                         cb->nlh->nlmsg_seq, RTM_NEWROUTE,
1697                                         1, NLM_F_MULTI) <= 0) {
1698                                 skb_dst_drop(skb);
1699                                 rcu_read_unlock_bh();
1700                                 goto done;
1701                         }
1702                         skb_dst_drop(skb);
1703                 }
1704                 rcu_read_unlock_bh();
1705         }
1706
1707 done:
1708         cb->args[0] = h;
1709         cb->args[1] = idx;
1710         return skb->len;
1711 }
1712
1713 #ifdef CONFIG_PROC_FS
1714 struct dn_rt_cache_iter_state {
1715         int bucket;
1716 };
1717
1718 static struct dn_route *dn_rt_cache_get_first(struct seq_file *seq)
1719 {
1720         struct dn_route *rt = NULL;
1721         struct dn_rt_cache_iter_state *s = seq->private;
1722
1723         for(s->bucket = dn_rt_hash_mask; s->bucket >= 0; --s->bucket) {
1724                 rcu_read_lock_bh();
1725                 rt = rcu_dereference_bh(dn_rt_hash_table[s->bucket].chain);
1726                 if (rt)
1727                         break;
1728                 rcu_read_unlock_bh();
1729         }
1730         return rt;
1731 }
1732
1733 static struct dn_route *dn_rt_cache_get_next(struct seq_file *seq, struct dn_route *rt)
1734 {
1735         struct dn_rt_cache_iter_state *s = seq->private;
1736
1737         rt = rcu_dereference_bh(rt->dst.dn_next);
1738         while (!rt) {
1739                 rcu_read_unlock_bh();
1740                 if (--s->bucket < 0)
1741                         break;
1742                 rcu_read_lock_bh();
1743                 rt = rcu_dereference_bh(dn_rt_hash_table[s->bucket].chain);
1744         }
1745         return rt;
1746 }
1747
1748 static void *dn_rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
1749 {
1750         struct dn_route *rt = dn_rt_cache_get_first(seq);
1751
1752         if (rt) {
1753                 while(*pos && (rt = dn_rt_cache_get_next(seq, rt)))
1754                         --*pos;
1755         }
1756         return *pos ? NULL : rt;
1757 }
1758
1759 static void *dn_rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1760 {
1761         struct dn_route *rt = dn_rt_cache_get_next(seq, v);
1762         ++*pos;
1763         return rt;
1764 }
1765
1766 static void dn_rt_cache_seq_stop(struct seq_file *seq, void *v)
1767 {
1768         if (v)
1769                 rcu_read_unlock_bh();
1770 }
1771
1772 static int dn_rt_cache_seq_show(struct seq_file *seq, void *v)
1773 {
1774         struct dn_route *rt = v;
1775         char buf1[DN_ASCBUF_LEN], buf2[DN_ASCBUF_LEN];
1776
1777         seq_printf(seq, "%-8s %-7s %-7s %04d %04d %04d\n",
1778                         rt->dst.dev ? rt->dst.dev->name : "*",
1779                         dn_addr2asc(le16_to_cpu(rt->rt_daddr), buf1),
1780                         dn_addr2asc(le16_to_cpu(rt->rt_saddr), buf2),
1781                         atomic_read(&rt->dst.__refcnt),
1782                         rt->dst.__use,
1783                         (int) dst_metric(&rt->dst, RTAX_RTT));
1784         return 0;
1785 }
1786
1787 static const struct seq_operations dn_rt_cache_seq_ops = {
1788         .start  = dn_rt_cache_seq_start,
1789         .next   = dn_rt_cache_seq_next,
1790         .stop   = dn_rt_cache_seq_stop,
1791         .show   = dn_rt_cache_seq_show,
1792 };
1793
1794 static int dn_rt_cache_seq_open(struct inode *inode, struct file *file)
1795 {
1796         return seq_open_private(file, &dn_rt_cache_seq_ops,
1797                         sizeof(struct dn_rt_cache_iter_state));
1798 }
1799
1800 static const struct file_operations dn_rt_cache_seq_fops = {
1801         .owner   = THIS_MODULE,
1802         .open    = dn_rt_cache_seq_open,
1803         .read    = seq_read,
1804         .llseek  = seq_lseek,
1805         .release = seq_release_private,
1806 };
1807
1808 #endif /* CONFIG_PROC_FS */
1809
1810 void __init dn_route_init(void)
1811 {
1812         int i, goal, order;
1813
1814         dn_dst_ops.kmem_cachep =
1815                 kmem_cache_create("dn_dst_cache", sizeof(struct dn_route), 0,
1816                                   SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1817         dst_entries_init(&dn_dst_ops);
1818         setup_timer(&dn_route_timer, dn_dst_check_expire, 0);
1819         dn_route_timer.expires = jiffies + decnet_dst_gc_interval * HZ;
1820         add_timer(&dn_route_timer);
1821
1822         goal = totalram_pages >> (26 - PAGE_SHIFT);
1823
1824         for(order = 0; (1UL << order) < goal; order++)
1825                 /* NOTHING */;
1826
1827         /*
1828          * Only want 1024 entries max, since the table is very, very unlikely
1829          * to be larger than that.
1830          */
1831         while(order && ((((1UL << order) * PAGE_SIZE) /
1832                                 sizeof(struct dn_rt_hash_bucket)) >= 2048))
1833                 order--;
1834
1835         do {
1836                 dn_rt_hash_mask = (1UL << order) * PAGE_SIZE /
1837                         sizeof(struct dn_rt_hash_bucket);
1838                 while(dn_rt_hash_mask & (dn_rt_hash_mask - 1))
1839                         dn_rt_hash_mask--;
1840                 dn_rt_hash_table = (struct dn_rt_hash_bucket *)
1841                         __get_free_pages(GFP_ATOMIC, order);
1842         } while (dn_rt_hash_table == NULL && --order > 0);
1843
1844         if (!dn_rt_hash_table)
1845                 panic("Failed to allocate DECnet route cache hash table\n");
1846
1847         printk(KERN_INFO
1848                 "DECnet: Routing cache hash table of %u buckets, %ldKbytes\n",
1849                 dn_rt_hash_mask,
1850                 (long)(dn_rt_hash_mask*sizeof(struct dn_rt_hash_bucket))/1024);
1851
1852         dn_rt_hash_mask--;
1853         for(i = 0; i <= dn_rt_hash_mask; i++) {
1854                 spin_lock_init(&dn_rt_hash_table[i].lock);
1855                 dn_rt_hash_table[i].chain = NULL;
1856         }
1857
1858         dn_dst_ops.gc_thresh = (dn_rt_hash_mask + 1);
1859
1860         proc_net_fops_create(&init_net, "decnet_cache", S_IRUGO, &dn_rt_cache_seq_fops);
1861
1862 #ifdef CONFIG_DECNET_ROUTER
1863         rtnl_register(PF_DECnet, RTM_GETROUTE, dn_cache_getroute,
1864                       dn_fib_dump, NULL);
1865 #else
1866         rtnl_register(PF_DECnet, RTM_GETROUTE, dn_cache_getroute,
1867                       dn_cache_dump, NULL);
1868 #endif
1869 }
1870
1871 void __exit dn_route_cleanup(void)
1872 {
1873         del_timer(&dn_route_timer);
1874         dn_run_flush(0);
1875
1876         proc_net_remove(&init_net, "decnet_cache");
1877         dst_entries_destroy(&dn_dst_ops);
1878 }
1879