neigh: Pass neighbour entry to output ops.
[linux-3.10.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 <asm/errno.h>
81 #include <net/net_namespace.h>
82 #include <net/netlink.h>
83 #include <net/neighbour.h>
84 #include <net/dst.h>
85 #include <net/flow.h>
86 #include <net/fib_rules.h>
87 #include <net/dn.h>
88 #include <net/dn_dev.h>
89 #include <net/dn_nsp.h>
90 #include <net/dn_route.h>
91 #include <net/dn_neigh.h>
92 #include <net/dn_fib.h>
93
94 struct dn_rt_hash_bucket
95 {
96         struct dn_route __rcu *chain;
97         spinlock_t lock;
98 };
99
100 extern struct neigh_table dn_neigh_table;
101
102
103 static unsigned char dn_hiord_addr[6] = {0xAA,0x00,0x04,0x00,0x00,0x00};
104
105 static const int dn_rt_min_delay = 2 * HZ;
106 static const int dn_rt_max_delay = 10 * HZ;
107 static const int dn_rt_mtu_expires = 10 * 60 * HZ;
108
109 static unsigned long dn_rt_deadline;
110
111 static int dn_dst_gc(struct dst_ops *ops);
112 static struct dst_entry *dn_dst_check(struct dst_entry *, __u32);
113 static unsigned int dn_dst_default_advmss(const struct dst_entry *dst);
114 static unsigned int dn_dst_default_mtu(const struct dst_entry *dst);
115 static void dn_dst_destroy(struct dst_entry *);
116 static struct dst_entry *dn_dst_negative_advice(struct dst_entry *);
117 static void dn_dst_link_failure(struct sk_buff *);
118 static void dn_dst_update_pmtu(struct dst_entry *dst, u32 mtu);
119 static int dn_route_input(struct sk_buff *);
120 static void dn_run_flush(unsigned long dummy);
121
122 static struct dn_rt_hash_bucket *dn_rt_hash_table;
123 static unsigned dn_rt_hash_mask;
124
125 static struct timer_list dn_route_timer;
126 static DEFINE_TIMER(dn_rt_flush_timer, dn_run_flush, 0, 0);
127 int decnet_dst_gc_interval = 2;
128
129 static struct dst_ops dn_dst_ops = {
130         .family =               PF_DECnet,
131         .protocol =             cpu_to_be16(ETH_P_DNA_RT),
132         .gc_thresh =            128,
133         .gc =                   dn_dst_gc,
134         .check =                dn_dst_check,
135         .default_advmss =       dn_dst_default_advmss,
136         .default_mtu =          dn_dst_default_mtu,
137         .cow_metrics =          dst_cow_metrics_generic,
138         .destroy =              dn_dst_destroy,
139         .negative_advice =      dn_dst_negative_advice,
140         .link_failure =         dn_dst_link_failure,
141         .update_pmtu =          dn_dst_update_pmtu,
142 };
143
144 static void dn_dst_destroy(struct dst_entry *dst)
145 {
146         dst_destroy_metrics_generic(dst);
147 }
148
149 static __inline__ unsigned dn_hash(__le16 src, __le16 dst)
150 {
151         __u16 tmp = (__u16 __force)(src ^ dst);
152         tmp ^= (tmp >> 3);
153         tmp ^= (tmp >> 5);
154         tmp ^= (tmp >> 10);
155         return dn_rt_hash_mask & (unsigned)tmp;
156 }
157
158 static inline void dnrt_free(struct dn_route *rt)
159 {
160         call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
161 }
162
163 static inline void dnrt_drop(struct dn_route *rt)
164 {
165         dst_release(&rt->dst);
166         call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
167 }
168
169 static void dn_dst_check_expire(unsigned long dummy)
170 {
171         int i;
172         struct dn_route *rt;
173         struct dn_route __rcu **rtp;
174         unsigned long now = jiffies;
175         unsigned long expire = 120 * HZ;
176
177         for (i = 0; i <= dn_rt_hash_mask; i++) {
178                 rtp = &dn_rt_hash_table[i].chain;
179
180                 spin_lock(&dn_rt_hash_table[i].lock);
181                 while ((rt = rcu_dereference_protected(*rtp,
182                                                 lockdep_is_held(&dn_rt_hash_table[i].lock))) != NULL) {
183                         if (atomic_read(&rt->dst.__refcnt) ||
184                                         (now - rt->dst.lastuse) < expire) {
185                                 rtp = &rt->dst.dn_next;
186                                 continue;
187                         }
188                         *rtp = rt->dst.dn_next;
189                         rt->dst.dn_next = NULL;
190                         dnrt_free(rt);
191                 }
192                 spin_unlock(&dn_rt_hash_table[i].lock);
193
194                 if ((jiffies - now) > 0)
195                         break;
196         }
197
198         mod_timer(&dn_route_timer, now + decnet_dst_gc_interval * HZ);
199 }
200
201 static int dn_dst_gc(struct dst_ops *ops)
202 {
203         struct dn_route *rt;
204         struct dn_route __rcu **rtp;
205         int i;
206         unsigned long now = jiffies;
207         unsigned long expire = 10 * HZ;
208
209         for (i = 0; i <= dn_rt_hash_mask; i++) {
210
211                 spin_lock_bh(&dn_rt_hash_table[i].lock);
212                 rtp = &dn_rt_hash_table[i].chain;
213
214                 while ((rt = rcu_dereference_protected(*rtp,
215                                                 lockdep_is_held(&dn_rt_hash_table[i].lock))) != NULL) {
216                         if (atomic_read(&rt->dst.__refcnt) ||
217                                         (now - rt->dst.lastuse) < expire) {
218                                 rtp = &rt->dst.dn_next;
219                                 continue;
220                         }
221                         *rtp = rt->dst.dn_next;
222                         rt->dst.dn_next = NULL;
223                         dnrt_drop(rt);
224                         break;
225                 }
226                 spin_unlock_bh(&dn_rt_hash_table[i].lock);
227         }
228
229         return 0;
230 }
231
232 /*
233  * The decnet standards don't impose a particular minimum mtu, what they
234  * do insist on is that the routing layer accepts a datagram of at least
235  * 230 bytes long. Here we have to subtract the routing header length from
236  * 230 to get the minimum acceptable mtu. If there is no neighbour, then we
237  * assume the worst and use a long header size.
238  *
239  * We update both the mtu and the advertised mss (i.e. the segment size we
240  * advertise to the other end).
241  */
242 static void dn_dst_update_pmtu(struct dst_entry *dst, u32 mtu)
243 {
244         u32 min_mtu = 230;
245         struct dn_dev *dn = dst->neighbour ?
246                             rcu_dereference_raw(dst->neighbour->dev->dn_ptr) : NULL;
247
248         if (dn && dn->use_long == 0)
249                 min_mtu -= 6;
250         else
251                 min_mtu -= 21;
252
253         if (dst_metric(dst, RTAX_MTU) > mtu && mtu >= min_mtu) {
254                 if (!(dst_metric_locked(dst, RTAX_MTU))) {
255                         dst_metric_set(dst, RTAX_MTU, mtu);
256                         dst_set_expires(dst, dn_rt_mtu_expires);
257                 }
258                 if (!(dst_metric_locked(dst, RTAX_ADVMSS))) {
259                         u32 mss = mtu - DN_MAX_NSP_DATA_HEADER;
260                         u32 existing_mss = dst_metric_raw(dst, RTAX_ADVMSS);
261                         if (!existing_mss || existing_mss > mss)
262                                 dst_metric_set(dst, RTAX_ADVMSS, mss);
263                 }
264         }
265 }
266
267 /*
268  * When a route has been marked obsolete. (e.g. routing cache flush)
269  */
270 static struct dst_entry *dn_dst_check(struct dst_entry *dst, __u32 cookie)
271 {
272         return NULL;
273 }
274
275 static struct dst_entry *dn_dst_negative_advice(struct dst_entry *dst)
276 {
277         dst_release(dst);
278         return NULL;
279 }
280
281 static void dn_dst_link_failure(struct sk_buff *skb)
282 {
283 }
284
285 static inline int compare_keys(struct flowidn *fl1, struct flowidn *fl2)
286 {
287         return ((fl1->daddr ^ fl2->daddr) |
288                 (fl1->saddr ^ fl2->saddr) |
289                 (fl1->flowidn_mark ^ fl2->flowidn_mark) |
290                 (fl1->flowidn_scope ^ fl2->flowidn_scope) |
291                 (fl1->flowidn_oif ^ fl2->flowidn_oif) |
292                 (fl1->flowidn_iif ^ fl2->flowidn_iif)) == 0;
293 }
294
295 static int dn_insert_route(struct dn_route *rt, unsigned hash, struct dn_route **rp)
296 {
297         struct dn_route *rth;
298         struct dn_route __rcu **rthp;
299         unsigned long now = jiffies;
300
301         rthp = &dn_rt_hash_table[hash].chain;
302
303         spin_lock_bh(&dn_rt_hash_table[hash].lock);
304         while ((rth = rcu_dereference_protected(*rthp,
305                                                 lockdep_is_held(&dn_rt_hash_table[hash].lock))) != NULL) {
306                 if (compare_keys(&rth->fld, &rt->fld)) {
307                         /* Put it first */
308                         *rthp = rth->dst.dn_next;
309                         rcu_assign_pointer(rth->dst.dn_next,
310                                            dn_rt_hash_table[hash].chain);
311                         rcu_assign_pointer(dn_rt_hash_table[hash].chain, rth);
312
313                         dst_use(&rth->dst, now);
314                         spin_unlock_bh(&dn_rt_hash_table[hash].lock);
315
316                         dnrt_drop(rt);
317                         *rp = rth;
318                         return 0;
319                 }
320                 rthp = &rth->dst.dn_next;
321         }
322
323         rcu_assign_pointer(rt->dst.dn_next, dn_rt_hash_table[hash].chain);
324         rcu_assign_pointer(dn_rt_hash_table[hash].chain, rt);
325
326         dst_use(&rt->dst, now);
327         spin_unlock_bh(&dn_rt_hash_table[hash].lock);
328         *rp = rt;
329         return 0;
330 }
331
332 static void dn_run_flush(unsigned long dummy)
333 {
334         int i;
335         struct dn_route *rt, *next;
336
337         for (i = 0; i < dn_rt_hash_mask; i++) {
338                 spin_lock_bh(&dn_rt_hash_table[i].lock);
339
340                 if ((rt = xchg((struct dn_route **)&dn_rt_hash_table[i].chain, NULL)) == NULL)
341                         goto nothing_to_declare;
342
343                 for(; rt; rt = next) {
344                         next = rcu_dereference_raw(rt->dst.dn_next);
345                         RCU_INIT_POINTER(rt->dst.dn_next, NULL);
346                         dst_free((struct dst_entry *)rt);
347                 }
348
349 nothing_to_declare:
350                 spin_unlock_bh(&dn_rt_hash_table[i].lock);
351         }
352 }
353
354 static DEFINE_SPINLOCK(dn_rt_flush_lock);
355
356 void dn_rt_cache_flush(int delay)
357 {
358         unsigned long now = jiffies;
359         int user_mode = !in_interrupt();
360
361         if (delay < 0)
362                 delay = dn_rt_min_delay;
363
364         spin_lock_bh(&dn_rt_flush_lock);
365
366         if (del_timer(&dn_rt_flush_timer) && delay > 0 && dn_rt_deadline) {
367                 long tmo = (long)(dn_rt_deadline - now);
368
369                 if (user_mode && tmo < dn_rt_max_delay - dn_rt_min_delay)
370                         tmo = 0;
371
372                 if (delay > tmo)
373                         delay = tmo;
374         }
375
376         if (delay <= 0) {
377                 spin_unlock_bh(&dn_rt_flush_lock);
378                 dn_run_flush(0);
379                 return;
380         }
381
382         if (dn_rt_deadline == 0)
383                 dn_rt_deadline = now + dn_rt_max_delay;
384
385         dn_rt_flush_timer.expires = now + delay;
386         add_timer(&dn_rt_flush_timer);
387         spin_unlock_bh(&dn_rt_flush_lock);
388 }
389
390 /**
391  * dn_return_short - Return a short packet to its sender
392  * @skb: The packet to return
393  *
394  */
395 static int dn_return_short(struct sk_buff *skb)
396 {
397         struct dn_skb_cb *cb;
398         unsigned char *ptr;
399         __le16 *src;
400         __le16 *dst;
401
402         /* Add back headers */
403         skb_push(skb, skb->data - skb_network_header(skb));
404
405         if ((skb = skb_unshare(skb, GFP_ATOMIC)) == NULL)
406                 return NET_RX_DROP;
407
408         cb = DN_SKB_CB(skb);
409         /* Skip packet length and point to flags */
410         ptr = skb->data + 2;
411         *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
412
413         dst = (__le16 *)ptr;
414         ptr += 2;
415         src = (__le16 *)ptr;
416         ptr += 2;
417         *ptr = 0; /* Zero hop count */
418
419         swap(*src, *dst);
420
421         skb->pkt_type = PACKET_OUTGOING;
422         dn_rt_finish_output(skb, NULL, NULL);
423         return NET_RX_SUCCESS;
424 }
425
426 /**
427  * dn_return_long - Return a long packet to its sender
428  * @skb: The long format packet to return
429  *
430  */
431 static int dn_return_long(struct sk_buff *skb)
432 {
433         struct dn_skb_cb *cb;
434         unsigned char *ptr;
435         unsigned char *src_addr, *dst_addr;
436         unsigned char tmp[ETH_ALEN];
437
438         /* Add back all headers */
439         skb_push(skb, skb->data - skb_network_header(skb));
440
441         if ((skb = skb_unshare(skb, GFP_ATOMIC)) == NULL)
442                 return NET_RX_DROP;
443
444         cb = DN_SKB_CB(skb);
445         /* Ignore packet length and point to flags */
446         ptr = skb->data + 2;
447
448         /* Skip padding */
449         if (*ptr & DN_RT_F_PF) {
450                 char padlen = (*ptr & ~DN_RT_F_PF);
451                 ptr += padlen;
452         }
453
454         *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
455         ptr += 2;
456         dst_addr = ptr;
457         ptr += 8;
458         src_addr = ptr;
459         ptr += 6;
460         *ptr = 0; /* Zero hop count */
461
462         /* Swap source and destination */
463         memcpy(tmp, src_addr, ETH_ALEN);
464         memcpy(src_addr, dst_addr, ETH_ALEN);
465         memcpy(dst_addr, tmp, ETH_ALEN);
466
467         skb->pkt_type = PACKET_OUTGOING;
468         dn_rt_finish_output(skb, dst_addr, src_addr);
469         return NET_RX_SUCCESS;
470 }
471
472 /**
473  * dn_route_rx_packet - Try and find a route for an incoming packet
474  * @skb: The packet to find a route for
475  *
476  * Returns: result of input function if route is found, error code otherwise
477  */
478 static int dn_route_rx_packet(struct sk_buff *skb)
479 {
480         struct dn_skb_cb *cb;
481         int err;
482
483         if ((err = dn_route_input(skb)) == 0)
484                 return dst_input(skb);
485
486         cb = DN_SKB_CB(skb);
487         if (decnet_debug_level & 4) {
488                 char *devname = skb->dev ? skb->dev->name : "???";
489
490                 printk(KERN_DEBUG
491                         "DECnet: dn_route_rx_packet: rt_flags=0x%02x dev=%s len=%d src=0x%04hx dst=0x%04hx err=%d type=%d\n",
492                         (int)cb->rt_flags, devname, skb->len,
493                         le16_to_cpu(cb->src), le16_to_cpu(cb->dst),
494                         err, skb->pkt_type);
495         }
496
497         if ((skb->pkt_type == PACKET_HOST) && (cb->rt_flags & DN_RT_F_RQR)) {
498                 switch (cb->rt_flags & DN_RT_PKT_MSK) {
499                 case DN_RT_PKT_SHORT:
500                         return dn_return_short(skb);
501                 case DN_RT_PKT_LONG:
502                         return dn_return_long(skb);
503                 }
504         }
505
506         kfree_skb(skb);
507         return NET_RX_DROP;
508 }
509
510 static int dn_route_rx_long(struct sk_buff *skb)
511 {
512         struct dn_skb_cb *cb = DN_SKB_CB(skb);
513         unsigned char *ptr = skb->data;
514
515         if (!pskb_may_pull(skb, 21)) /* 20 for long header, 1 for shortest nsp */
516                 goto drop_it;
517
518         skb_pull(skb, 20);
519         skb_reset_transport_header(skb);
520
521         /* Destination info */
522         ptr += 2;
523         cb->dst = dn_eth2dn(ptr);
524         if (memcmp(ptr, dn_hiord_addr, 4) != 0)
525                 goto drop_it;
526         ptr += 6;
527
528
529         /* Source info */
530         ptr += 2;
531         cb->src = dn_eth2dn(ptr);
532         if (memcmp(ptr, dn_hiord_addr, 4) != 0)
533                 goto drop_it;
534         ptr += 6;
535         /* Other junk */
536         ptr++;
537         cb->hops = *ptr++; /* Visit Count */
538
539         return NF_HOOK(NFPROTO_DECNET, NF_DN_PRE_ROUTING, skb, skb->dev, NULL,
540                        dn_route_rx_packet);
541
542 drop_it:
543         kfree_skb(skb);
544         return NET_RX_DROP;
545 }
546
547
548
549 static int dn_route_rx_short(struct sk_buff *skb)
550 {
551         struct dn_skb_cb *cb = DN_SKB_CB(skb);
552         unsigned char *ptr = skb->data;
553
554         if (!pskb_may_pull(skb, 6)) /* 5 for short header + 1 for shortest nsp */
555                 goto drop_it;
556
557         skb_pull(skb, 5);
558         skb_reset_transport_header(skb);
559
560         cb->dst = *(__le16 *)ptr;
561         ptr += 2;
562         cb->src = *(__le16 *)ptr;
563         ptr += 2;
564         cb->hops = *ptr & 0x3f;
565
566         return NF_HOOK(NFPROTO_DECNET, NF_DN_PRE_ROUTING, skb, skb->dev, NULL,
567                        dn_route_rx_packet);
568
569 drop_it:
570         kfree_skb(skb);
571         return NET_RX_DROP;
572 }
573
574 static int dn_route_discard(struct sk_buff *skb)
575 {
576         /*
577          * I know we drop the packet here, but thats considered success in
578          * this case
579          */
580         kfree_skb(skb);
581         return NET_RX_SUCCESS;
582 }
583
584 static int dn_route_ptp_hello(struct sk_buff *skb)
585 {
586         dn_dev_hello(skb);
587         dn_neigh_pointopoint_hello(skb);
588         return NET_RX_SUCCESS;
589 }
590
591 int dn_route_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
592 {
593         struct dn_skb_cb *cb;
594         unsigned char flags = 0;
595         __u16 len = le16_to_cpu(*(__le16 *)skb->data);
596         struct dn_dev *dn = rcu_dereference(dev->dn_ptr);
597         unsigned char padlen = 0;
598
599         if (!net_eq(dev_net(dev), &init_net))
600                 goto dump_it;
601
602         if (dn == NULL)
603                 goto dump_it;
604
605         if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
606                 goto out;
607
608         if (!pskb_may_pull(skb, 3))
609                 goto dump_it;
610
611         skb_pull(skb, 2);
612
613         if (len > skb->len)
614                 goto dump_it;
615
616         skb_trim(skb, len);
617
618         flags = *skb->data;
619
620         cb = DN_SKB_CB(skb);
621         cb->stamp = jiffies;
622         cb->iif = dev->ifindex;
623
624         /*
625          * If we have padding, remove it.
626          */
627         if (flags & DN_RT_F_PF) {
628                 padlen = flags & ~DN_RT_F_PF;
629                 if (!pskb_may_pull(skb, padlen + 1))
630                         goto dump_it;
631                 skb_pull(skb, padlen);
632                 flags = *skb->data;
633         }
634
635         skb_reset_network_header(skb);
636
637         /*
638          * Weed out future version DECnet
639          */
640         if (flags & DN_RT_F_VER)
641                 goto dump_it;
642
643         cb->rt_flags = flags;
644
645         if (decnet_debug_level & 1)
646                 printk(KERN_DEBUG
647                         "dn_route_rcv: got 0x%02x from %s [%d %d %d]\n",
648                         (int)flags, (dev) ? dev->name : "???", len, skb->len,
649                         padlen);
650
651         if (flags & DN_RT_PKT_CNTL) {
652                 if (unlikely(skb_linearize(skb)))
653                         goto dump_it;
654
655                 switch (flags & DN_RT_CNTL_MSK) {
656                 case DN_RT_PKT_INIT:
657                         dn_dev_init_pkt(skb);
658                         break;
659                 case DN_RT_PKT_VERI:
660                         dn_dev_veri_pkt(skb);
661                         break;
662                 }
663
664                 if (dn->parms.state != DN_DEV_S_RU)
665                         goto dump_it;
666
667                 switch (flags & DN_RT_CNTL_MSK) {
668                 case DN_RT_PKT_HELO:
669                         return NF_HOOK(NFPROTO_DECNET, NF_DN_HELLO,
670                                        skb, skb->dev, NULL,
671                                        dn_route_ptp_hello);
672
673                 case DN_RT_PKT_L1RT:
674                 case DN_RT_PKT_L2RT:
675                         return NF_HOOK(NFPROTO_DECNET, NF_DN_ROUTE,
676                                        skb, skb->dev, NULL,
677                                        dn_route_discard);
678                 case DN_RT_PKT_ERTH:
679                         return NF_HOOK(NFPROTO_DECNET, NF_DN_HELLO,
680                                        skb, skb->dev, NULL,
681                                        dn_neigh_router_hello);
682
683                 case DN_RT_PKT_EEDH:
684                         return NF_HOOK(NFPROTO_DECNET, NF_DN_HELLO,
685                                        skb, skb->dev, NULL,
686                                        dn_neigh_endnode_hello);
687                 }
688         } else {
689                 if (dn->parms.state != DN_DEV_S_RU)
690                         goto dump_it;
691
692                 skb_pull(skb, 1); /* Pull flags */
693
694                 switch (flags & DN_RT_PKT_MSK) {
695                 case DN_RT_PKT_LONG:
696                         return dn_route_rx_long(skb);
697                 case DN_RT_PKT_SHORT:
698                         return dn_route_rx_short(skb);
699                 }
700         }
701
702 dump_it:
703         kfree_skb(skb);
704 out:
705         return NET_RX_DROP;
706 }
707
708 static int dn_to_neigh_output(struct sk_buff *skb)
709 {
710         struct dst_entry *dst = skb_dst(skb);
711         struct neighbour *n = dst->neighbour;
712
713         return n->output(n, skb);
714 }
715
716 static int dn_output(struct sk_buff *skb)
717 {
718         struct dst_entry *dst = skb_dst(skb);
719         struct dn_route *rt = (struct dn_route *)dst;
720         struct net_device *dev = dst->dev;
721         struct dn_skb_cb *cb = DN_SKB_CB(skb);
722         struct neighbour *neigh;
723
724         int err = -EINVAL;
725
726         if ((neigh = dst->neighbour) == NULL)
727                 goto error;
728
729         skb->dev = dev;
730
731         cb->src = rt->rt_saddr;
732         cb->dst = rt->rt_daddr;
733
734         /*
735          * Always set the Intra-Ethernet bit on all outgoing packets
736          * originated on this node. Only valid flag from upper layers
737          * is return-to-sender-requested. Set hop count to 0 too.
738          */
739         cb->rt_flags &= ~DN_RT_F_RQR;
740         cb->rt_flags |= DN_RT_F_IE;
741         cb->hops = 0;
742
743         return NF_HOOK(NFPROTO_DECNET, NF_DN_LOCAL_OUT, skb, NULL, dev,
744                        dn_to_neigh_output);
745
746 error:
747         if (net_ratelimit())
748                 printk(KERN_DEBUG "dn_output: This should not happen\n");
749
750         kfree_skb(skb);
751
752         return err;
753 }
754
755 static int dn_forward(struct sk_buff *skb)
756 {
757         struct dn_skb_cb *cb = DN_SKB_CB(skb);
758         struct dst_entry *dst = skb_dst(skb);
759         struct dn_dev *dn_db = rcu_dereference(dst->dev->dn_ptr);
760         struct dn_route *rt;
761         int header_len;
762 #ifdef CONFIG_NETFILTER
763         struct net_device *dev = skb->dev;
764 #endif
765
766         if (skb->pkt_type != PACKET_HOST)
767                 goto drop;
768
769         /* Ensure that we have enough space for headers */
770         rt = (struct dn_route *)skb_dst(skb);
771         header_len = dn_db->use_long ? 21 : 6;
772         if (skb_cow(skb, LL_RESERVED_SPACE(rt->dst.dev)+header_len))
773                 goto drop;
774
775         /*
776          * Hop count exceeded.
777          */
778         if (++cb->hops > 30)
779                 goto drop;
780
781         skb->dev = rt->dst.dev;
782
783         /*
784          * If packet goes out same interface it came in on, then set
785          * the Intra-Ethernet bit. This has no effect for short
786          * packets, so we don't need to test for them here.
787          */
788         cb->rt_flags &= ~DN_RT_F_IE;
789         if (rt->rt_flags & RTCF_DOREDIRECT)
790                 cb->rt_flags |= DN_RT_F_IE;
791
792         return NF_HOOK(NFPROTO_DECNET, NF_DN_FORWARD, skb, dev, skb->dev,
793                        dn_to_neigh_output);
794
795 drop:
796         kfree_skb(skb);
797         return NET_RX_DROP;
798 }
799
800 /*
801  * Used to catch bugs. This should never normally get
802  * called.
803  */
804 static int dn_rt_bug(struct sk_buff *skb)
805 {
806         if (net_ratelimit()) {
807                 struct dn_skb_cb *cb = DN_SKB_CB(skb);
808
809                 printk(KERN_DEBUG "dn_rt_bug: skb from:%04x to:%04x\n",
810                                 le16_to_cpu(cb->src), le16_to_cpu(cb->dst));
811         }
812
813         kfree_skb(skb);
814
815         return NET_RX_DROP;
816 }
817
818 static unsigned int dn_dst_default_advmss(const struct dst_entry *dst)
819 {
820         return dn_mss_from_pmtu(dst->dev, dst_mtu(dst));
821 }
822
823 static unsigned int dn_dst_default_mtu(const struct dst_entry *dst)
824 {
825         return dst->dev->mtu;
826 }
827
828 static int dn_rt_set_next_hop(struct dn_route *rt, struct dn_fib_res *res)
829 {
830         struct dn_fib_info *fi = res->fi;
831         struct net_device *dev = rt->dst.dev;
832         unsigned int mss_metric;
833         struct neighbour *n;
834
835         if (fi) {
836                 if (DN_FIB_RES_GW(*res) &&
837                     DN_FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
838                         rt->rt_gateway = DN_FIB_RES_GW(*res);
839                 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
840         }
841         rt->rt_type = res->type;
842
843         if (dev != NULL && rt->dst.neighbour == NULL) {
844                 n = __neigh_lookup_errno(&dn_neigh_table, &rt->rt_gateway, dev);
845                 if (IS_ERR(n))
846                         return PTR_ERR(n);
847                 rt->dst.neighbour = n;
848         }
849
850         if (dst_metric(&rt->dst, RTAX_MTU) > rt->dst.dev->mtu)
851                 dst_metric_set(&rt->dst, RTAX_MTU, rt->dst.dev->mtu);
852         mss_metric = dst_metric_raw(&rt->dst, RTAX_ADVMSS);
853         if (mss_metric) {
854                 unsigned int mss = dn_mss_from_pmtu(dev, dst_mtu(&rt->dst));
855                 if (mss_metric > mss)
856                         dst_metric_set(&rt->dst, RTAX_ADVMSS, mss);
857         }
858         return 0;
859 }
860
861 static inline int dn_match_addr(__le16 addr1, __le16 addr2)
862 {
863         __u16 tmp = le16_to_cpu(addr1) ^ le16_to_cpu(addr2);
864         int match = 16;
865         while(tmp) {
866                 tmp >>= 1;
867                 match--;
868         }
869         return match;
870 }
871
872 static __le16 dnet_select_source(const struct net_device *dev, __le16 daddr, int scope)
873 {
874         __le16 saddr = 0;
875         struct dn_dev *dn_db;
876         struct dn_ifaddr *ifa;
877         int best_match = 0;
878         int ret;
879
880         rcu_read_lock();
881         dn_db = rcu_dereference(dev->dn_ptr);
882         for (ifa = rcu_dereference(dn_db->ifa_list);
883              ifa != NULL;
884              ifa = rcu_dereference(ifa->ifa_next)) {
885                 if (ifa->ifa_scope > scope)
886                         continue;
887                 if (!daddr) {
888                         saddr = ifa->ifa_local;
889                         break;
890                 }
891                 ret = dn_match_addr(daddr, ifa->ifa_local);
892                 if (ret > best_match)
893                         saddr = ifa->ifa_local;
894                 if (best_match == 0)
895                         saddr = ifa->ifa_local;
896         }
897         rcu_read_unlock();
898
899         return saddr;
900 }
901
902 static inline __le16 __dn_fib_res_prefsrc(struct dn_fib_res *res)
903 {
904         return dnet_select_source(DN_FIB_RES_DEV(*res), DN_FIB_RES_GW(*res), res->scope);
905 }
906
907 static inline __le16 dn_fib_rules_map_destination(__le16 daddr, struct dn_fib_res *res)
908 {
909         __le16 mask = dnet_make_mask(res->prefixlen);
910         return (daddr&~mask)|res->fi->fib_nh->nh_gw;
911 }
912
913 static int dn_route_output_slow(struct dst_entry **pprt, const struct flowidn *oldflp, int try_hard)
914 {
915         struct flowidn fld = {
916                 .daddr = oldflp->daddr,
917                 .saddr = oldflp->saddr,
918                 .flowidn_scope = RT_SCOPE_UNIVERSE,
919                 .flowidn_mark = oldflp->flowidn_mark,
920                 .flowidn_iif = init_net.loopback_dev->ifindex,
921                 .flowidn_oif = oldflp->flowidn_oif,
922         };
923         struct dn_route *rt = NULL;
924         struct net_device *dev_out = NULL, *dev;
925         struct neighbour *neigh = NULL;
926         unsigned hash;
927         unsigned flags = 0;
928         struct dn_fib_res res = { .fi = NULL, .type = RTN_UNICAST };
929         int err;
930         int free_res = 0;
931         __le16 gateway = 0;
932
933         if (decnet_debug_level & 16)
934                 printk(KERN_DEBUG
935                        "dn_route_output_slow: dst=%04x src=%04x mark=%d"
936                        " iif=%d oif=%d\n", le16_to_cpu(oldflp->daddr),
937                        le16_to_cpu(oldflp->saddr),
938                        oldflp->flowidn_mark, init_net.loopback_dev->ifindex,
939                        oldflp->flowidn_oif);
940
941         /* If we have an output interface, verify its a DECnet device */
942         if (oldflp->flowidn_oif) {
943                 dev_out = dev_get_by_index(&init_net, oldflp->flowidn_oif);
944                 err = -ENODEV;
945                 if (dev_out && dev_out->dn_ptr == NULL) {
946                         dev_put(dev_out);
947                         dev_out = NULL;
948                 }
949                 if (dev_out == NULL)
950                         goto out;
951         }
952
953         /* If we have a source address, verify that its a local address */
954         if (oldflp->saddr) {
955                 err = -EADDRNOTAVAIL;
956
957                 if (dev_out) {
958                         if (dn_dev_islocal(dev_out, oldflp->saddr))
959                                 goto source_ok;
960                         dev_put(dev_out);
961                         goto out;
962                 }
963                 rcu_read_lock();
964                 for_each_netdev_rcu(&init_net, dev) {
965                         if (!dev->dn_ptr)
966                                 continue;
967                         if (!dn_dev_islocal(dev, oldflp->saddr))
968                                 continue;
969                         if ((dev->flags & IFF_LOOPBACK) &&
970                             oldflp->daddr &&
971                             !dn_dev_islocal(dev, oldflp->daddr))
972                                 continue;
973
974                         dev_out = dev;
975                         break;
976                 }
977                 rcu_read_unlock();
978                 if (dev_out == NULL)
979                         goto out;
980                 dev_hold(dev_out);
981 source_ok:
982                 ;
983         }
984
985         /* No destination? Assume its local */
986         if (!fld.daddr) {
987                 fld.daddr = fld.saddr;
988
989                 err = -EADDRNOTAVAIL;
990                 if (dev_out)
991                         dev_put(dev_out);
992                 dev_out = init_net.loopback_dev;
993                 dev_hold(dev_out);
994                 if (!fld.daddr) {
995                         fld.daddr =
996                         fld.saddr = dnet_select_source(dev_out, 0,
997                                                        RT_SCOPE_HOST);
998                         if (!fld.daddr)
999                                 goto out;
1000                 }
1001                 fld.flowidn_oif = init_net.loopback_dev->ifindex;
1002                 res.type = RTN_LOCAL;
1003                 goto make_route;
1004         }
1005
1006         if (decnet_debug_level & 16)
1007                 printk(KERN_DEBUG
1008                        "dn_route_output_slow: initial checks complete."
1009                        " dst=%o4x src=%04x oif=%d try_hard=%d\n",
1010                        le16_to_cpu(fld.daddr), le16_to_cpu(fld.saddr),
1011                        fld.flowidn_oif, try_hard);
1012
1013         /*
1014          * N.B. If the kernel is compiled without router support then
1015          * dn_fib_lookup() will evaluate to non-zero so this if () block
1016          * will always be executed.
1017          */
1018         err = -ESRCH;
1019         if (try_hard || (err = dn_fib_lookup(&fld, &res)) != 0) {
1020                 struct dn_dev *dn_db;
1021                 if (err != -ESRCH)
1022                         goto out;
1023                 /*
1024                  * Here the fallback is basically the standard algorithm for
1025                  * routing in endnodes which is described in the DECnet routing
1026                  * docs
1027                  *
1028                  * If we are not trying hard, look in neighbour cache.
1029                  * The result is tested to ensure that if a specific output
1030                  * device/source address was requested, then we honour that
1031                  * here
1032                  */
1033                 if (!try_hard) {
1034                         neigh = neigh_lookup_nodev(&dn_neigh_table, &init_net, &fld.daddr);
1035                         if (neigh) {
1036                                 if ((oldflp->flowidn_oif &&
1037                                     (neigh->dev->ifindex != oldflp->flowidn_oif)) ||
1038                                     (oldflp->saddr &&
1039                                     (!dn_dev_islocal(neigh->dev,
1040                                                      oldflp->saddr)))) {
1041                                         neigh_release(neigh);
1042                                         neigh = NULL;
1043                                 } else {
1044                                         if (dev_out)
1045                                                 dev_put(dev_out);
1046                                         if (dn_dev_islocal(neigh->dev, fld.daddr)) {
1047                                                 dev_out = init_net.loopback_dev;
1048                                                 res.type = RTN_LOCAL;
1049                                         } else {
1050                                                 dev_out = neigh->dev;
1051                                         }
1052                                         dev_hold(dev_out);
1053                                         goto select_source;
1054                                 }
1055                         }
1056                 }
1057
1058                 /* Not there? Perhaps its a local address */
1059                 if (dev_out == NULL)
1060                         dev_out = dn_dev_get_default();
1061                 err = -ENODEV;
1062                 if (dev_out == NULL)
1063                         goto out;
1064                 dn_db = rcu_dereference_raw(dev_out->dn_ptr);
1065                 /* Possible improvement - check all devices for local addr */
1066                 if (dn_dev_islocal(dev_out, fld.daddr)) {
1067                         dev_put(dev_out);
1068                         dev_out = init_net.loopback_dev;
1069                         dev_hold(dev_out);
1070                         res.type = RTN_LOCAL;
1071                         goto select_source;
1072                 }
1073                 /* Not local either.... try sending it to the default router */
1074                 neigh = neigh_clone(dn_db->router);
1075                 BUG_ON(neigh && neigh->dev != dev_out);
1076
1077                 /* Ok then, we assume its directly connected and move on */
1078 select_source:
1079                 if (neigh)
1080                         gateway = ((struct dn_neigh *)neigh)->addr;
1081                 if (gateway == 0)
1082                         gateway = fld.daddr;
1083                 if (fld.saddr == 0) {
1084                         fld.saddr = dnet_select_source(dev_out, gateway,
1085                                                        res.type == RTN_LOCAL ?
1086                                                        RT_SCOPE_HOST :
1087                                                        RT_SCOPE_LINK);
1088                         if (fld.saddr == 0 && res.type != RTN_LOCAL)
1089                                 goto e_addr;
1090                 }
1091                 fld.flowidn_oif = dev_out->ifindex;
1092                 goto make_route;
1093         }
1094         free_res = 1;
1095
1096         if (res.type == RTN_NAT)
1097                 goto e_inval;
1098
1099         if (res.type == RTN_LOCAL) {
1100                 if (!fld.saddr)
1101                         fld.saddr = fld.daddr;
1102                 if (dev_out)
1103                         dev_put(dev_out);
1104                 dev_out = init_net.loopback_dev;
1105                 dev_hold(dev_out);
1106                 fld.flowidn_oif = dev_out->ifindex;
1107                 if (res.fi)
1108                         dn_fib_info_put(res.fi);
1109                 res.fi = NULL;
1110                 goto make_route;
1111         }
1112
1113         if (res.fi->fib_nhs > 1 && fld.flowidn_oif == 0)
1114                 dn_fib_select_multipath(&fld, &res);
1115
1116         /*
1117          * We could add some logic to deal with default routes here and
1118          * get rid of some of the special casing above.
1119          */
1120
1121         if (!fld.saddr)
1122                 fld.saddr = DN_FIB_RES_PREFSRC(res);
1123
1124         if (dev_out)
1125                 dev_put(dev_out);
1126         dev_out = DN_FIB_RES_DEV(res);
1127         dev_hold(dev_out);
1128         fld.flowidn_oif = dev_out->ifindex;
1129         gateway = DN_FIB_RES_GW(res);
1130
1131 make_route:
1132         if (dev_out->flags & IFF_LOOPBACK)
1133                 flags |= RTCF_LOCAL;
1134
1135         rt = dst_alloc(&dn_dst_ops, dev_out, 1, 0, DST_HOST);
1136         if (rt == NULL)
1137                 goto e_nobufs;
1138
1139         memset(&rt->fld, 0, sizeof(rt->fld));
1140         rt->fld.saddr        = oldflp->saddr;
1141         rt->fld.daddr        = oldflp->daddr;
1142         rt->fld.flowidn_oif  = oldflp->flowidn_oif;
1143         rt->fld.flowidn_iif  = 0;
1144         rt->fld.flowidn_mark = oldflp->flowidn_mark;
1145
1146         rt->rt_saddr      = fld.saddr;
1147         rt->rt_daddr      = fld.daddr;
1148         rt->rt_gateway    = gateway ? gateway : fld.daddr;
1149         rt->rt_local_src  = fld.saddr;
1150
1151         rt->rt_dst_map    = fld.daddr;
1152         rt->rt_src_map    = fld.saddr;
1153
1154         rt->dst.neighbour = neigh;
1155         neigh = NULL;
1156
1157         rt->dst.lastuse = jiffies;
1158         rt->dst.output  = dn_output;
1159         rt->dst.input   = dn_rt_bug;
1160         rt->rt_flags      = flags;
1161         if (flags & RTCF_LOCAL)
1162                 rt->dst.input = dn_nsp_rx;
1163
1164         err = dn_rt_set_next_hop(rt, &res);
1165         if (err)
1166                 goto e_neighbour;
1167
1168         hash = dn_hash(rt->fld.saddr, rt->fld.daddr);
1169         dn_insert_route(rt, hash, (struct dn_route **)pprt);
1170
1171 done:
1172         if (neigh)
1173                 neigh_release(neigh);
1174         if (free_res)
1175                 dn_fib_res_put(&res);
1176         if (dev_out)
1177                 dev_put(dev_out);
1178 out:
1179         return err;
1180
1181 e_addr:
1182         err = -EADDRNOTAVAIL;
1183         goto done;
1184 e_inval:
1185         err = -EINVAL;
1186         goto done;
1187 e_nobufs:
1188         err = -ENOBUFS;
1189         goto done;
1190 e_neighbour:
1191         dst_free(&rt->dst);
1192         goto e_nobufs;
1193 }
1194
1195
1196 /*
1197  * N.B. The flags may be moved into the flowi at some future stage.
1198  */
1199 static int __dn_route_output_key(struct dst_entry **pprt, const struct flowidn *flp, int flags)
1200 {
1201         unsigned hash = dn_hash(flp->saddr, flp->daddr);
1202         struct dn_route *rt = NULL;
1203
1204         if (!(flags & MSG_TRYHARD)) {
1205                 rcu_read_lock_bh();
1206                 for (rt = rcu_dereference_bh(dn_rt_hash_table[hash].chain); rt;
1207                         rt = rcu_dereference_bh(rt->dst.dn_next)) {
1208                         if ((flp->daddr == rt->fld.daddr) &&
1209                             (flp->saddr == rt->fld.saddr) &&
1210                             (flp->flowidn_mark == rt->fld.flowidn_mark) &&
1211                             dn_is_output_route(rt) &&
1212                             (rt->fld.flowidn_oif == flp->flowidn_oif)) {
1213                                 dst_use(&rt->dst, jiffies);
1214                                 rcu_read_unlock_bh();
1215                                 *pprt = &rt->dst;
1216                                 return 0;
1217                         }
1218                 }
1219                 rcu_read_unlock_bh();
1220         }
1221
1222         return dn_route_output_slow(pprt, flp, flags);
1223 }
1224
1225 static int dn_route_output_key(struct dst_entry **pprt, struct flowidn *flp, int flags)
1226 {
1227         int err;
1228
1229         err = __dn_route_output_key(pprt, flp, flags);
1230         if (err == 0 && flp->flowidn_proto) {
1231                 *pprt = xfrm_lookup(&init_net, *pprt,
1232                                     flowidn_to_flowi(flp), NULL, 0);
1233                 if (IS_ERR(*pprt)) {
1234                         err = PTR_ERR(*pprt);
1235                         *pprt = NULL;
1236                 }
1237         }
1238         return err;
1239 }
1240
1241 int dn_route_output_sock(struct dst_entry **pprt, struct flowidn *fl, struct sock *sk, int flags)
1242 {
1243         int err;
1244
1245         err = __dn_route_output_key(pprt, fl, flags & MSG_TRYHARD);
1246         if (err == 0 && fl->flowidn_proto) {
1247                 if (!(flags & MSG_DONTWAIT))
1248                         fl->flowidn_flags |= FLOWI_FLAG_CAN_SLEEP;
1249                 *pprt = xfrm_lookup(&init_net, *pprt,
1250                                     flowidn_to_flowi(fl), sk, 0);
1251                 if (IS_ERR(*pprt)) {
1252                         err = PTR_ERR(*pprt);
1253                         *pprt = NULL;
1254                 }
1255         }
1256         return err;
1257 }
1258
1259 static int dn_route_input_slow(struct sk_buff *skb)
1260 {
1261         struct dn_route *rt = NULL;
1262         struct dn_skb_cb *cb = DN_SKB_CB(skb);
1263         struct net_device *in_dev = skb->dev;
1264         struct net_device *out_dev = NULL;
1265         struct dn_dev *dn_db;
1266         struct neighbour *neigh = NULL;
1267         unsigned hash;
1268         int flags = 0;
1269         __le16 gateway = 0;
1270         __le16 local_src = 0;
1271         struct flowidn fld = {
1272                 .daddr = cb->dst,
1273                 .saddr = cb->src,
1274                 .flowidn_scope = RT_SCOPE_UNIVERSE,
1275                 .flowidn_mark = skb->mark,
1276                 .flowidn_iif = skb->dev->ifindex,
1277         };
1278         struct dn_fib_res res = { .fi = NULL, .type = RTN_UNREACHABLE };
1279         int err = -EINVAL;
1280         int free_res = 0;
1281
1282         dev_hold(in_dev);
1283
1284         if ((dn_db = rcu_dereference(in_dev->dn_ptr)) == NULL)
1285                 goto out;
1286
1287         /* Zero source addresses are not allowed */
1288         if (fld.saddr == 0)
1289                 goto out;
1290
1291         /*
1292          * In this case we've just received a packet from a source
1293          * outside ourselves pretending to come from us. We don't
1294          * allow it any further to prevent routing loops, spoofing and
1295          * other nasties. Loopback packets already have the dst attached
1296          * so this only affects packets which have originated elsewhere.
1297          */
1298         err  = -ENOTUNIQ;
1299         if (dn_dev_islocal(in_dev, cb->src))
1300                 goto out;
1301
1302         err = dn_fib_lookup(&fld, &res);
1303         if (err) {
1304                 if (err != -ESRCH)
1305                         goto out;
1306                 /*
1307                  * Is the destination us ?
1308                  */
1309                 if (!dn_dev_islocal(in_dev, cb->dst))
1310                         goto e_inval;
1311
1312                 res.type = RTN_LOCAL;
1313         } else {
1314                 __le16 src_map = fld.saddr;
1315                 free_res = 1;
1316
1317                 out_dev = DN_FIB_RES_DEV(res);
1318                 if (out_dev == NULL) {
1319                         if (net_ratelimit())
1320                                 printk(KERN_CRIT "Bug in dn_route_input_slow() "
1321                                                  "No output device\n");
1322                         goto e_inval;
1323                 }
1324                 dev_hold(out_dev);
1325
1326                 if (res.r)
1327                         src_map = fld.saddr; /* no NAT support for now */
1328
1329                 gateway = DN_FIB_RES_GW(res);
1330                 if (res.type == RTN_NAT) {
1331                         fld.daddr = dn_fib_rules_map_destination(fld.daddr, &res);
1332                         dn_fib_res_put(&res);
1333                         free_res = 0;
1334                         if (dn_fib_lookup(&fld, &res))
1335                                 goto e_inval;
1336                         free_res = 1;
1337                         if (res.type != RTN_UNICAST)
1338                                 goto e_inval;
1339                         flags |= RTCF_DNAT;
1340                         gateway = fld.daddr;
1341                 }
1342                 fld.saddr = src_map;
1343         }
1344
1345         switch(res.type) {
1346         case RTN_UNICAST:
1347                 /*
1348                  * Forwarding check here, we only check for forwarding
1349                  * being turned off, if you want to only forward intra
1350                  * area, its up to you to set the routing tables up
1351                  * correctly.
1352                  */
1353                 if (dn_db->parms.forwarding == 0)
1354                         goto e_inval;
1355
1356                 if (res.fi->fib_nhs > 1 && fld.flowidn_oif == 0)
1357                         dn_fib_select_multipath(&fld, &res);
1358
1359                 /*
1360                  * Check for out_dev == in_dev. We use the RTCF_DOREDIRECT
1361                  * flag as a hint to set the intra-ethernet bit when
1362                  * forwarding. If we've got NAT in operation, we don't do
1363                  * this optimisation.
1364                  */
1365                 if (out_dev == in_dev && !(flags & RTCF_NAT))
1366                         flags |= RTCF_DOREDIRECT;
1367
1368                 local_src = DN_FIB_RES_PREFSRC(res);
1369
1370         case RTN_BLACKHOLE:
1371         case RTN_UNREACHABLE:
1372                 break;
1373         case RTN_LOCAL:
1374                 flags |= RTCF_LOCAL;
1375                 fld.saddr = cb->dst;
1376                 fld.daddr = cb->src;
1377
1378                 /* Routing tables gave us a gateway */
1379                 if (gateway)
1380                         goto make_route;
1381
1382                 /* Packet was intra-ethernet, so we know its on-link */
1383                 if (cb->rt_flags & DN_RT_F_IE) {
1384                         gateway = cb->src;
1385                         flags |= RTCF_DIRECTSRC;
1386                         goto make_route;
1387                 }
1388
1389                 /* Use the default router if there is one */
1390                 neigh = neigh_clone(dn_db->router);
1391                 if (neigh) {
1392                         gateway = ((struct dn_neigh *)neigh)->addr;
1393                         goto make_route;
1394                 }
1395
1396                 /* Close eyes and pray */
1397                 gateway = cb->src;
1398                 flags |= RTCF_DIRECTSRC;
1399                 goto make_route;
1400         default:
1401                 goto e_inval;
1402         }
1403
1404 make_route:
1405         rt = dst_alloc(&dn_dst_ops, out_dev, 0, 0, DST_HOST);
1406         if (rt == NULL)
1407                 goto e_nobufs;
1408
1409         memset(&rt->fld, 0, sizeof(rt->fld));
1410         rt->rt_saddr      = fld.saddr;
1411         rt->rt_daddr      = fld.daddr;
1412         rt->rt_gateway    = fld.daddr;
1413         if (gateway)
1414                 rt->rt_gateway = gateway;
1415         rt->rt_local_src  = local_src ? local_src : rt->rt_saddr;
1416
1417         rt->rt_dst_map    = fld.daddr;
1418         rt->rt_src_map    = fld.saddr;
1419
1420         rt->fld.saddr        = cb->src;
1421         rt->fld.daddr        = cb->dst;
1422         rt->fld.flowidn_oif  = 0;
1423         rt->fld.flowidn_iif  = in_dev->ifindex;
1424         rt->fld.flowidn_mark = fld.flowidn_mark;
1425
1426         rt->dst.neighbour = neigh;
1427         rt->dst.lastuse = jiffies;
1428         rt->dst.output = dn_rt_bug;
1429         switch (res.type) {
1430         case RTN_UNICAST:
1431                 rt->dst.input = dn_forward;
1432                 break;
1433         case RTN_LOCAL:
1434                 rt->dst.output = dn_output;
1435                 rt->dst.input = dn_nsp_rx;
1436                 rt->dst.dev = in_dev;
1437                 flags |= RTCF_LOCAL;
1438                 break;
1439         default:
1440         case RTN_UNREACHABLE:
1441         case RTN_BLACKHOLE:
1442                 rt->dst.input = dst_discard;
1443         }
1444         rt->rt_flags = flags;
1445
1446         err = dn_rt_set_next_hop(rt, &res);
1447         if (err)
1448                 goto e_neighbour;
1449
1450         hash = dn_hash(rt->fld.saddr, rt->fld.daddr);
1451         dn_insert_route(rt, hash, &rt);
1452         skb_dst_set(skb, &rt->dst);
1453
1454 done:
1455         if (neigh)
1456                 neigh_release(neigh);
1457         if (free_res)
1458                 dn_fib_res_put(&res);
1459         dev_put(in_dev);
1460         if (out_dev)
1461                 dev_put(out_dev);
1462 out:
1463         return err;
1464
1465 e_inval:
1466         err = -EINVAL;
1467         goto done;
1468
1469 e_nobufs:
1470         err = -ENOBUFS;
1471         goto done;
1472
1473 e_neighbour:
1474         dst_free(&rt->dst);
1475         goto done;
1476 }
1477
1478 static int dn_route_input(struct sk_buff *skb)
1479 {
1480         struct dn_route *rt;
1481         struct dn_skb_cb *cb = DN_SKB_CB(skb);
1482         unsigned hash = dn_hash(cb->src, cb->dst);
1483
1484         if (skb_dst(skb))
1485                 return 0;
1486
1487         rcu_read_lock();
1488         for(rt = rcu_dereference(dn_rt_hash_table[hash].chain); rt != NULL;
1489             rt = rcu_dereference(rt->dst.dn_next)) {
1490                 if ((rt->fld.saddr == cb->src) &&
1491                     (rt->fld.daddr == cb->dst) &&
1492                     (rt->fld.flowidn_oif == 0) &&
1493                     (rt->fld.flowidn_mark == skb->mark) &&
1494                     (rt->fld.flowidn_iif == cb->iif)) {
1495                         dst_use(&rt->dst, jiffies);
1496                         rcu_read_unlock();
1497                         skb_dst_set(skb, (struct dst_entry *)rt);
1498                         return 0;
1499                 }
1500         }
1501         rcu_read_unlock();
1502
1503         return dn_route_input_slow(skb);
1504 }
1505
1506 static int dn_rt_fill_info(struct sk_buff *skb, u32 pid, u32 seq,
1507                            int event, int nowait, unsigned int flags)
1508 {
1509         struct dn_route *rt = (struct dn_route *)skb_dst(skb);
1510         struct rtmsg *r;
1511         struct nlmsghdr *nlh;
1512         unsigned char *b = skb_tail_pointer(skb);
1513         long expires;
1514
1515         nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*r), flags);
1516         r = NLMSG_DATA(nlh);
1517         r->rtm_family = AF_DECnet;
1518         r->rtm_dst_len = 16;
1519         r->rtm_src_len = 0;
1520         r->rtm_tos = 0;
1521         r->rtm_table = RT_TABLE_MAIN;
1522         RTA_PUT_U32(skb, RTA_TABLE, RT_TABLE_MAIN);
1523         r->rtm_type = rt->rt_type;
1524         r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
1525         r->rtm_scope = RT_SCOPE_UNIVERSE;
1526         r->rtm_protocol = RTPROT_UNSPEC;
1527         if (rt->rt_flags & RTCF_NOTIFY)
1528                 r->rtm_flags |= RTM_F_NOTIFY;
1529         RTA_PUT(skb, RTA_DST, 2, &rt->rt_daddr);
1530         if (rt->fld.saddr) {
1531                 r->rtm_src_len = 16;
1532                 RTA_PUT(skb, RTA_SRC, 2, &rt->fld.saddr);
1533         }
1534         if (rt->dst.dev)
1535                 RTA_PUT(skb, RTA_OIF, sizeof(int), &rt->dst.dev->ifindex);
1536         /*
1537          * Note to self - change this if input routes reverse direction when
1538          * they deal only with inputs and not with replies like they do
1539          * currently.
1540          */
1541         RTA_PUT(skb, RTA_PREFSRC, 2, &rt->rt_local_src);
1542         if (rt->rt_daddr != rt->rt_gateway)
1543                 RTA_PUT(skb, RTA_GATEWAY, 2, &rt->rt_gateway);
1544         if (rtnetlink_put_metrics(skb, dst_metrics_ptr(&rt->dst)) < 0)
1545                 goto rtattr_failure;
1546         expires = rt->dst.expires ? rt->dst.expires - jiffies : 0;
1547         if (rtnl_put_cacheinfo(skb, &rt->dst, 0, 0, 0, expires,
1548                                rt->dst.error) < 0)
1549                 goto rtattr_failure;
1550         if (dn_is_input_route(rt))
1551                 RTA_PUT(skb, RTA_IIF, sizeof(int), &rt->fld.flowidn_iif);
1552
1553         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1554         return skb->len;
1555
1556 nlmsg_failure:
1557 rtattr_failure:
1558         nlmsg_trim(skb, b);
1559         return -1;
1560 }
1561
1562 /*
1563  * This is called by both endnodes and routers now.
1564  */
1565 static int dn_cache_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh, void *arg)
1566 {
1567         struct net *net = sock_net(in_skb->sk);
1568         struct rtattr **rta = arg;
1569         struct rtmsg *rtm = NLMSG_DATA(nlh);
1570         struct dn_route *rt = NULL;
1571         struct dn_skb_cb *cb;
1572         int err;
1573         struct sk_buff *skb;
1574         struct flowidn fld;
1575
1576         if (!net_eq(net, &init_net))
1577                 return -EINVAL;
1578
1579         memset(&fld, 0, sizeof(fld));
1580         fld.flowidn_proto = DNPROTO_NSP;
1581
1582         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1583         if (skb == NULL)
1584                 return -ENOBUFS;
1585         skb_reset_mac_header(skb);
1586         cb = DN_SKB_CB(skb);
1587
1588         if (rta[RTA_SRC-1])
1589                 memcpy(&fld.saddr, RTA_DATA(rta[RTA_SRC-1]), 2);
1590         if (rta[RTA_DST-1])
1591                 memcpy(&fld.daddr, RTA_DATA(rta[RTA_DST-1]), 2);
1592         if (rta[RTA_IIF-1])
1593                 memcpy(&fld.flowidn_iif, RTA_DATA(rta[RTA_IIF-1]), sizeof(int));
1594
1595         if (fld.flowidn_iif) {
1596                 struct net_device *dev;
1597                 if ((dev = dev_get_by_index(&init_net, fld.flowidn_iif)) == NULL) {
1598                         kfree_skb(skb);
1599                         return -ENODEV;
1600                 }
1601                 if (!dev->dn_ptr) {
1602                         dev_put(dev);
1603                         kfree_skb(skb);
1604                         return -ENODEV;
1605                 }
1606                 skb->protocol = htons(ETH_P_DNA_RT);
1607                 skb->dev = dev;
1608                 cb->src = fld.saddr;
1609                 cb->dst = fld.daddr;
1610                 local_bh_disable();
1611                 err = dn_route_input(skb);
1612                 local_bh_enable();
1613                 memset(cb, 0, sizeof(struct dn_skb_cb));
1614                 rt = (struct dn_route *)skb_dst(skb);
1615                 if (!err && -rt->dst.error)
1616                         err = rt->dst.error;
1617         } else {
1618                 int oif = 0;
1619                 if (rta[RTA_OIF - 1])
1620                         memcpy(&oif, RTA_DATA(rta[RTA_OIF - 1]), sizeof(int));
1621                 fld.flowidn_oif = oif;
1622                 err = dn_route_output_key((struct dst_entry **)&rt, &fld, 0);
1623         }
1624
1625         if (skb->dev)
1626                 dev_put(skb->dev);
1627         skb->dev = NULL;
1628         if (err)
1629                 goto out_free;
1630         skb_dst_set(skb, &rt->dst);
1631         if (rtm->rtm_flags & RTM_F_NOTIFY)
1632                 rt->rt_flags |= RTCF_NOTIFY;
1633
1634         err = dn_rt_fill_info(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq, RTM_NEWROUTE, 0, 0);
1635
1636         if (err == 0)
1637                 goto out_free;
1638         if (err < 0) {
1639                 err = -EMSGSIZE;
1640                 goto out_free;
1641         }
1642
1643         return rtnl_unicast(skb, &init_net, NETLINK_CB(in_skb).pid);
1644
1645 out_free:
1646         kfree_skb(skb);
1647         return err;
1648 }
1649
1650 /*
1651  * For routers, this is called from dn_fib_dump, but for endnodes its
1652  * called directly from the rtnetlink dispatch table.
1653  */
1654 int dn_cache_dump(struct sk_buff *skb, struct netlink_callback *cb)
1655 {
1656         struct net *net = sock_net(skb->sk);
1657         struct dn_route *rt;
1658         int h, s_h;
1659         int idx, s_idx;
1660
1661         if (!net_eq(net, &init_net))
1662                 return 0;
1663
1664         if (NLMSG_PAYLOAD(cb->nlh, 0) < sizeof(struct rtmsg))
1665                 return -EINVAL;
1666         if (!(((struct rtmsg *)NLMSG_DATA(cb->nlh))->rtm_flags&RTM_F_CLONED))
1667                 return 0;
1668
1669         s_h = cb->args[0];
1670         s_idx = idx = cb->args[1];
1671         for(h = 0; h <= dn_rt_hash_mask; h++) {
1672                 if (h < s_h)
1673                         continue;
1674                 if (h > s_h)
1675                         s_idx = 0;
1676                 rcu_read_lock_bh();
1677                 for(rt = rcu_dereference_bh(dn_rt_hash_table[h].chain), idx = 0;
1678                         rt;
1679                         rt = rcu_dereference_bh(rt->dst.dn_next), idx++) {
1680                         if (idx < s_idx)
1681                                 continue;
1682                         skb_dst_set(skb, dst_clone(&rt->dst));
1683                         if (dn_rt_fill_info(skb, NETLINK_CB(cb->skb).pid,
1684                                         cb->nlh->nlmsg_seq, RTM_NEWROUTE,
1685                                         1, NLM_F_MULTI) <= 0) {
1686                                 skb_dst_drop(skb);
1687                                 rcu_read_unlock_bh();
1688                                 goto done;
1689                         }
1690                         skb_dst_drop(skb);
1691                 }
1692                 rcu_read_unlock_bh();
1693         }
1694
1695 done:
1696         cb->args[0] = h;
1697         cb->args[1] = idx;
1698         return skb->len;
1699 }
1700
1701 #ifdef CONFIG_PROC_FS
1702 struct dn_rt_cache_iter_state {
1703         int bucket;
1704 };
1705
1706 static struct dn_route *dn_rt_cache_get_first(struct seq_file *seq)
1707 {
1708         struct dn_route *rt = NULL;
1709         struct dn_rt_cache_iter_state *s = seq->private;
1710
1711         for(s->bucket = dn_rt_hash_mask; s->bucket >= 0; --s->bucket) {
1712                 rcu_read_lock_bh();
1713                 rt = rcu_dereference_bh(dn_rt_hash_table[s->bucket].chain);
1714                 if (rt)
1715                         break;
1716                 rcu_read_unlock_bh();
1717         }
1718         return rt;
1719 }
1720
1721 static struct dn_route *dn_rt_cache_get_next(struct seq_file *seq, struct dn_route *rt)
1722 {
1723         struct dn_rt_cache_iter_state *s = seq->private;
1724
1725         rt = rcu_dereference_bh(rt->dst.dn_next);
1726         while (!rt) {
1727                 rcu_read_unlock_bh();
1728                 if (--s->bucket < 0)
1729                         break;
1730                 rcu_read_lock_bh();
1731                 rt = rcu_dereference_bh(dn_rt_hash_table[s->bucket].chain);
1732         }
1733         return rt;
1734 }
1735
1736 static void *dn_rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
1737 {
1738         struct dn_route *rt = dn_rt_cache_get_first(seq);
1739
1740         if (rt) {
1741                 while(*pos && (rt = dn_rt_cache_get_next(seq, rt)))
1742                         --*pos;
1743         }
1744         return *pos ? NULL : rt;
1745 }
1746
1747 static void *dn_rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1748 {
1749         struct dn_route *rt = dn_rt_cache_get_next(seq, v);
1750         ++*pos;
1751         return rt;
1752 }
1753
1754 static void dn_rt_cache_seq_stop(struct seq_file *seq, void *v)
1755 {
1756         if (v)
1757                 rcu_read_unlock_bh();
1758 }
1759
1760 static int dn_rt_cache_seq_show(struct seq_file *seq, void *v)
1761 {
1762         struct dn_route *rt = v;
1763         char buf1[DN_ASCBUF_LEN], buf2[DN_ASCBUF_LEN];
1764
1765         seq_printf(seq, "%-8s %-7s %-7s %04d %04d %04d\n",
1766                         rt->dst.dev ? rt->dst.dev->name : "*",
1767                         dn_addr2asc(le16_to_cpu(rt->rt_daddr), buf1),
1768                         dn_addr2asc(le16_to_cpu(rt->rt_saddr), buf2),
1769                         atomic_read(&rt->dst.__refcnt),
1770                         rt->dst.__use,
1771                         (int) dst_metric(&rt->dst, RTAX_RTT));
1772         return 0;
1773 }
1774
1775 static const struct seq_operations dn_rt_cache_seq_ops = {
1776         .start  = dn_rt_cache_seq_start,
1777         .next   = dn_rt_cache_seq_next,
1778         .stop   = dn_rt_cache_seq_stop,
1779         .show   = dn_rt_cache_seq_show,
1780 };
1781
1782 static int dn_rt_cache_seq_open(struct inode *inode, struct file *file)
1783 {
1784         return seq_open_private(file, &dn_rt_cache_seq_ops,
1785                         sizeof(struct dn_rt_cache_iter_state));
1786 }
1787
1788 static const struct file_operations dn_rt_cache_seq_fops = {
1789         .owner   = THIS_MODULE,
1790         .open    = dn_rt_cache_seq_open,
1791         .read    = seq_read,
1792         .llseek  = seq_lseek,
1793         .release = seq_release_private,
1794 };
1795
1796 #endif /* CONFIG_PROC_FS */
1797
1798 void __init dn_route_init(void)
1799 {
1800         int i, goal, order;
1801
1802         dn_dst_ops.kmem_cachep =
1803                 kmem_cache_create("dn_dst_cache", sizeof(struct dn_route), 0,
1804                                   SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1805         dst_entries_init(&dn_dst_ops);
1806         setup_timer(&dn_route_timer, dn_dst_check_expire, 0);
1807         dn_route_timer.expires = jiffies + decnet_dst_gc_interval * HZ;
1808         add_timer(&dn_route_timer);
1809
1810         goal = totalram_pages >> (26 - PAGE_SHIFT);
1811
1812         for(order = 0; (1UL << order) < goal; order++)
1813                 /* NOTHING */;
1814
1815         /*
1816          * Only want 1024 entries max, since the table is very, very unlikely
1817          * to be larger than that.
1818          */
1819         while(order && ((((1UL << order) * PAGE_SIZE) /
1820                                 sizeof(struct dn_rt_hash_bucket)) >= 2048))
1821                 order--;
1822
1823         do {
1824                 dn_rt_hash_mask = (1UL << order) * PAGE_SIZE /
1825                         sizeof(struct dn_rt_hash_bucket);
1826                 while(dn_rt_hash_mask & (dn_rt_hash_mask - 1))
1827                         dn_rt_hash_mask--;
1828                 dn_rt_hash_table = (struct dn_rt_hash_bucket *)
1829                         __get_free_pages(GFP_ATOMIC, order);
1830         } while (dn_rt_hash_table == NULL && --order > 0);
1831
1832         if (!dn_rt_hash_table)
1833                 panic("Failed to allocate DECnet route cache hash table\n");
1834
1835         printk(KERN_INFO
1836                 "DECnet: Routing cache hash table of %u buckets, %ldKbytes\n",
1837                 dn_rt_hash_mask,
1838                 (long)(dn_rt_hash_mask*sizeof(struct dn_rt_hash_bucket))/1024);
1839
1840         dn_rt_hash_mask--;
1841         for(i = 0; i <= dn_rt_hash_mask; i++) {
1842                 spin_lock_init(&dn_rt_hash_table[i].lock);
1843                 dn_rt_hash_table[i].chain = NULL;
1844         }
1845
1846         dn_dst_ops.gc_thresh = (dn_rt_hash_mask + 1);
1847
1848         proc_net_fops_create(&init_net, "decnet_cache", S_IRUGO, &dn_rt_cache_seq_fops);
1849
1850 #ifdef CONFIG_DECNET_ROUTER
1851         rtnl_register(PF_DECnet, RTM_GETROUTE, dn_cache_getroute,
1852                       dn_fib_dump, NULL);
1853 #else
1854         rtnl_register(PF_DECnet, RTM_GETROUTE, dn_cache_getroute,
1855                       dn_cache_dump, NULL);
1856 #endif
1857 }
1858
1859 void __exit dn_route_cleanup(void)
1860 {
1861         del_timer(&dn_route_timer);
1862         dn_run_flush(0);
1863
1864         proc_net_remove(&init_net, "decnet_cache");
1865         dst_entries_destroy(&dn_dst_ops);
1866 }
1867