5d7337bcf0fecde048d6916ac33f9919a72f5a41
[linux-2.6.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 <net/sock.h>
70 #include <linux/mm.h>
71 #include <linux/proc_fs.h>
72 #include <linux/seq_file.h>
73 #include <linux/init.h>
74 #include <linux/rtnetlink.h>
75 #include <linux/string.h>
76 #include <linux/netfilter_decnet.h>
77 #include <linux/rcupdate.h>
78 #include <linux/times.h>
79 #include <asm/errno.h>
80 #include <net/netlink.h>
81 #include <net/neighbour.h>
82 #include <net/dst.h>
83 #include <net/flow.h>
84 #include <net/fib_rules.h>
85 #include <net/dn.h>
86 #include <net/dn_dev.h>
87 #include <net/dn_nsp.h>
88 #include <net/dn_route.h>
89 #include <net/dn_neigh.h>
90 #include <net/dn_fib.h>
91
92 struct dn_rt_hash_bucket
93 {
94         struct dn_route *chain;
95         spinlock_t lock;
96 } __attribute__((__aligned__(8)));
97
98 extern struct neigh_table dn_neigh_table;
99
100
101 static unsigned char dn_hiord_addr[6] = {0xAA,0x00,0x04,0x00,0x00,0x00};
102
103 static const int dn_rt_min_delay = 2 * HZ;
104 static const int dn_rt_max_delay = 10 * HZ;
105 static const int dn_rt_mtu_expires = 10 * 60 * HZ;
106
107 static unsigned long dn_rt_deadline;
108
109 static int dn_dst_gc(void);
110 static struct dst_entry *dn_dst_check(struct dst_entry *, __u32);
111 static struct dst_entry *dn_dst_negative_advice(struct dst_entry *);
112 static void dn_dst_link_failure(struct sk_buff *);
113 static void dn_dst_update_pmtu(struct dst_entry *dst, u32 mtu);
114 static int dn_route_input(struct sk_buff *);
115 static void dn_run_flush(unsigned long dummy);
116
117 static struct dn_rt_hash_bucket *dn_rt_hash_table;
118 static unsigned dn_rt_hash_mask;
119
120 static struct timer_list dn_route_timer;
121 static DEFINE_TIMER(dn_rt_flush_timer, dn_run_flush, 0, 0);
122 int decnet_dst_gc_interval = 2;
123
124 static struct dst_ops dn_dst_ops = {
125         .family =               PF_DECnet,
126         .protocol =             __constant_htons(ETH_P_DNA_RT),
127         .gc_thresh =            128,
128         .gc =                   dn_dst_gc,
129         .check =                dn_dst_check,
130         .negative_advice =      dn_dst_negative_advice,
131         .link_failure =         dn_dst_link_failure,
132         .update_pmtu =          dn_dst_update_pmtu,
133         .entry_size =           sizeof(struct dn_route),
134         .entries =              ATOMIC_INIT(0),
135 };
136
137 static __inline__ unsigned dn_hash(__le16 src, __le16 dst)
138 {
139         __u16 tmp = (__u16 __force)(src ^ dst);
140         tmp ^= (tmp >> 3);
141         tmp ^= (tmp >> 5);
142         tmp ^= (tmp >> 10);
143         return dn_rt_hash_mask & (unsigned)tmp;
144 }
145
146 static inline void dnrt_free(struct dn_route *rt)
147 {
148         call_rcu_bh(&rt->u.dst.rcu_head, dst_rcu_free);
149 }
150
151 static inline void dnrt_drop(struct dn_route *rt)
152 {
153         dst_release(&rt->u.dst);
154         call_rcu_bh(&rt->u.dst.rcu_head, dst_rcu_free);
155 }
156
157 static void dn_dst_check_expire(unsigned long dummy)
158 {
159         int i;
160         struct dn_route *rt, **rtp;
161         unsigned long now = jiffies;
162         unsigned long expire = 120 * HZ;
163
164         for(i = 0; i <= dn_rt_hash_mask; i++) {
165                 rtp = &dn_rt_hash_table[i].chain;
166
167                 spin_lock(&dn_rt_hash_table[i].lock);
168                 while((rt=*rtp) != NULL) {
169                         if (atomic_read(&rt->u.dst.__refcnt) ||
170                                         (now - rt->u.dst.lastuse) < expire) {
171                                 rtp = &rt->u.dst.dn_next;
172                                 continue;
173                         }
174                         *rtp = rt->u.dst.dn_next;
175                         rt->u.dst.dn_next = NULL;
176                         dnrt_free(rt);
177                 }
178                 spin_unlock(&dn_rt_hash_table[i].lock);
179
180                 if ((jiffies - now) > 0)
181                         break;
182         }
183
184         mod_timer(&dn_route_timer, now + decnet_dst_gc_interval * HZ);
185 }
186
187 static int dn_dst_gc(void)
188 {
189         struct dn_route *rt, **rtp;
190         int i;
191         unsigned long now = jiffies;
192         unsigned long expire = 10 * HZ;
193
194         for(i = 0; i <= dn_rt_hash_mask; i++) {
195
196                 spin_lock_bh(&dn_rt_hash_table[i].lock);
197                 rtp = &dn_rt_hash_table[i].chain;
198
199                 while((rt=*rtp) != NULL) {
200                         if (atomic_read(&rt->u.dst.__refcnt) ||
201                                         (now - rt->u.dst.lastuse) < expire) {
202                                 rtp = &rt->u.dst.dn_next;
203                                 continue;
204                         }
205                         *rtp = rt->u.dst.dn_next;
206                         rt->u.dst.dn_next = NULL;
207                         dnrt_drop(rt);
208                         break;
209                 }
210                 spin_unlock_bh(&dn_rt_hash_table[i].lock);
211         }
212
213         return 0;
214 }
215
216 /*
217  * The decnet standards don't impose a particular minimum mtu, what they
218  * do insist on is that the routing layer accepts a datagram of at least
219  * 230 bytes long. Here we have to subtract the routing header length from
220  * 230 to get the minimum acceptable mtu. If there is no neighbour, then we
221  * assume the worst and use a long header size.
222  *
223  * We update both the mtu and the advertised mss (i.e. the segment size we
224  * advertise to the other end).
225  */
226 static void dn_dst_update_pmtu(struct dst_entry *dst, u32 mtu)
227 {
228         u32 min_mtu = 230;
229         struct dn_dev *dn = dst->neighbour ?
230                             (struct dn_dev *)dst->neighbour->dev->dn_ptr : NULL;
231
232         if (dn && dn->use_long == 0)
233                 min_mtu -= 6;
234         else
235                 min_mtu -= 21;
236
237         if (dst->metrics[RTAX_MTU-1] > mtu && mtu >= min_mtu) {
238                 if (!(dst_metric_locked(dst, RTAX_MTU))) {
239                         dst->metrics[RTAX_MTU-1] = mtu;
240                         dst_set_expires(dst, dn_rt_mtu_expires);
241                 }
242                 if (!(dst_metric_locked(dst, RTAX_ADVMSS))) {
243                         u32 mss = mtu - DN_MAX_NSP_DATA_HEADER;
244                         if (dst->metrics[RTAX_ADVMSS-1] > mss)
245                                 dst->metrics[RTAX_ADVMSS-1] = mss;
246                 }
247         }
248 }
249
250 /*
251  * When a route has been marked obsolete. (e.g. routing cache flush)
252  */
253 static struct dst_entry *dn_dst_check(struct dst_entry *dst, __u32 cookie)
254 {
255         return NULL;
256 }
257
258 static struct dst_entry *dn_dst_negative_advice(struct dst_entry *dst)
259 {
260         dst_release(dst);
261         return NULL;
262 }
263
264 static void dn_dst_link_failure(struct sk_buff *skb)
265 {
266         return;
267 }
268
269 static inline int compare_keys(struct flowi *fl1, struct flowi *fl2)
270 {
271         return ((fl1->nl_u.dn_u.daddr ^ fl2->nl_u.dn_u.daddr) |
272                 (fl1->nl_u.dn_u.saddr ^ fl2->nl_u.dn_u.saddr) |
273                 (fl1->mark ^ fl2->mark) |
274                 (fl1->nl_u.dn_u.scope ^ fl2->nl_u.dn_u.scope) |
275                 (fl1->oif ^ fl2->oif) |
276                 (fl1->iif ^ fl2->iif)) == 0;
277 }
278
279 static int dn_insert_route(struct dn_route *rt, unsigned hash, struct dn_route **rp)
280 {
281         struct dn_route *rth, **rthp;
282         unsigned long now = jiffies;
283
284         rthp = &dn_rt_hash_table[hash].chain;
285
286         spin_lock_bh(&dn_rt_hash_table[hash].lock);
287         while((rth = *rthp) != NULL) {
288                 if (compare_keys(&rth->fl, &rt->fl)) {
289                         /* Put it first */
290                         *rthp = rth->u.dst.dn_next;
291                         rcu_assign_pointer(rth->u.dst.dn_next,
292                                            dn_rt_hash_table[hash].chain);
293                         rcu_assign_pointer(dn_rt_hash_table[hash].chain, rth);
294
295                         rth->u.dst.__use++;
296                         dst_hold(&rth->u.dst);
297                         rth->u.dst.lastuse = now;
298                         spin_unlock_bh(&dn_rt_hash_table[hash].lock);
299
300                         dnrt_drop(rt);
301                         *rp = rth;
302                         return 0;
303                 }
304                 rthp = &rth->u.dst.dn_next;
305         }
306
307         rcu_assign_pointer(rt->u.dst.dn_next, dn_rt_hash_table[hash].chain);
308         rcu_assign_pointer(dn_rt_hash_table[hash].chain, rt);
309
310         dst_hold(&rt->u.dst);
311         rt->u.dst.__use++;
312         rt->u.dst.lastuse = now;
313         spin_unlock_bh(&dn_rt_hash_table[hash].lock);
314         *rp = rt;
315         return 0;
316 }
317
318 void dn_run_flush(unsigned long dummy)
319 {
320         int i;
321         struct dn_route *rt, *next;
322
323         for(i = 0; i < dn_rt_hash_mask; i++) {
324                 spin_lock_bh(&dn_rt_hash_table[i].lock);
325
326                 if ((rt = xchg(&dn_rt_hash_table[i].chain, NULL)) == NULL)
327                         goto nothing_to_declare;
328
329                 for(; rt; rt=next) {
330                         next = rt->u.dst.dn_next;
331                         rt->u.dst.dn_next = NULL;
332                         dst_free((struct dst_entry *)rt);
333                 }
334
335 nothing_to_declare:
336                 spin_unlock_bh(&dn_rt_hash_table[i].lock);
337         }
338 }
339
340 static DEFINE_SPINLOCK(dn_rt_flush_lock);
341
342 void dn_rt_cache_flush(int delay)
343 {
344         unsigned long now = jiffies;
345         int user_mode = !in_interrupt();
346
347         if (delay < 0)
348                 delay = dn_rt_min_delay;
349
350         spin_lock_bh(&dn_rt_flush_lock);
351
352         if (del_timer(&dn_rt_flush_timer) && delay > 0 && dn_rt_deadline) {
353                 long tmo = (long)(dn_rt_deadline - now);
354
355                 if (user_mode && tmo < dn_rt_max_delay - dn_rt_min_delay)
356                         tmo = 0;
357
358                 if (delay > tmo)
359                         delay = tmo;
360         }
361
362         if (delay <= 0) {
363                 spin_unlock_bh(&dn_rt_flush_lock);
364                 dn_run_flush(0);
365                 return;
366         }
367
368         if (dn_rt_deadline == 0)
369                 dn_rt_deadline = now + dn_rt_max_delay;
370
371         dn_rt_flush_timer.expires = now + delay;
372         add_timer(&dn_rt_flush_timer);
373         spin_unlock_bh(&dn_rt_flush_lock);
374 }
375
376 /**
377  * dn_return_short - Return a short packet to its sender
378  * @skb: The packet to return
379  *
380  */
381 static int dn_return_short(struct sk_buff *skb)
382 {
383         struct dn_skb_cb *cb;
384         unsigned char *ptr;
385         __le16 *src;
386         __le16 *dst;
387         __le16 tmp;
388
389         /* Add back headers */
390         skb_push(skb, skb->data - skb_network_header(skb));
391
392         if ((skb = skb_unshare(skb, GFP_ATOMIC)) == NULL)
393                 return NET_RX_DROP;
394
395         cb = DN_SKB_CB(skb);
396         /* Skip packet length and point to flags */
397         ptr = skb->data + 2;
398         *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
399
400         dst = (__le16 *)ptr;
401         ptr += 2;
402         src = (__le16 *)ptr;
403         ptr += 2;
404         *ptr = 0; /* Zero hop count */
405
406         /* Swap source and destination */
407         tmp  = *src;
408         *src = *dst;
409         *dst = tmp;
410
411         skb->pkt_type = PACKET_OUTGOING;
412         dn_rt_finish_output(skb, NULL, NULL);
413         return NET_RX_SUCCESS;
414 }
415
416 /**
417  * dn_return_long - Return a long packet to its sender
418  * @skb: The long format packet to return
419  *
420  */
421 static int dn_return_long(struct sk_buff *skb)
422 {
423         struct dn_skb_cb *cb;
424         unsigned char *ptr;
425         unsigned char *src_addr, *dst_addr;
426         unsigned char tmp[ETH_ALEN];
427
428         /* Add back all headers */
429         skb_push(skb, skb->data - skb_network_header(skb));
430
431         if ((skb = skb_unshare(skb, GFP_ATOMIC)) == NULL)
432                 return NET_RX_DROP;
433
434         cb = DN_SKB_CB(skb);
435         /* Ignore packet length and point to flags */
436         ptr = skb->data + 2;
437
438         /* Skip padding */
439         if (*ptr & DN_RT_F_PF) {
440                 char padlen = (*ptr & ~DN_RT_F_PF);
441                 ptr += padlen;
442         }
443
444         *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
445         ptr += 2;
446         dst_addr = ptr;
447         ptr += 8;
448         src_addr = ptr;
449         ptr += 6;
450         *ptr = 0; /* Zero hop count */
451
452         /* Swap source and destination */
453         memcpy(tmp, src_addr, ETH_ALEN);
454         memcpy(src_addr, dst_addr, ETH_ALEN);
455         memcpy(dst_addr, tmp, ETH_ALEN);
456
457         skb->pkt_type = PACKET_OUTGOING;
458         dn_rt_finish_output(skb, dst_addr, src_addr);
459         return NET_RX_SUCCESS;
460 }
461
462 /**
463  * dn_route_rx_packet - Try and find a route for an incoming packet
464  * @skb: The packet to find a route for
465  *
466  * Returns: result of input function if route is found, error code otherwise
467  */
468 static int dn_route_rx_packet(struct sk_buff *skb)
469 {
470         struct dn_skb_cb *cb = DN_SKB_CB(skb);
471         int err;
472
473         if ((err = dn_route_input(skb)) == 0)
474                 return dst_input(skb);
475
476         if (decnet_debug_level & 4) {
477                 char *devname = skb->dev ? skb->dev->name : "???";
478                 struct dn_skb_cb *cb = DN_SKB_CB(skb);
479                 printk(KERN_DEBUG
480                         "DECnet: dn_route_rx_packet: rt_flags=0x%02x dev=%s len=%d src=0x%04hx dst=0x%04hx err=%d type=%d\n",
481                         (int)cb->rt_flags, devname, skb->len,
482                         dn_ntohs(cb->src), dn_ntohs(cb->dst),
483                         err, skb->pkt_type);
484         }
485
486         if ((skb->pkt_type == PACKET_HOST) && (cb->rt_flags & DN_RT_F_RQR)) {
487                 switch(cb->rt_flags & DN_RT_PKT_MSK) {
488                         case DN_RT_PKT_SHORT:
489                                 return dn_return_short(skb);
490                         case DN_RT_PKT_LONG:
491                                 return dn_return_long(skb);
492                 }
493         }
494
495         kfree_skb(skb);
496         return NET_RX_DROP;
497 }
498
499 static int dn_route_rx_long(struct sk_buff *skb)
500 {
501         struct dn_skb_cb *cb = DN_SKB_CB(skb);
502         unsigned char *ptr = skb->data;
503
504         if (!pskb_may_pull(skb, 21)) /* 20 for long header, 1 for shortest nsp */
505                 goto drop_it;
506
507         skb_pull(skb, 20);
508         skb_reset_transport_header(skb);
509
510         /* Destination info */
511         ptr += 2;
512         cb->dst = dn_eth2dn(ptr);
513         if (memcmp(ptr, dn_hiord_addr, 4) != 0)
514                 goto drop_it;
515         ptr += 6;
516
517
518         /* Source info */
519         ptr += 2;
520         cb->src = dn_eth2dn(ptr);
521         if (memcmp(ptr, dn_hiord_addr, 4) != 0)
522                 goto drop_it;
523         ptr += 6;
524         /* Other junk */
525         ptr++;
526         cb->hops = *ptr++; /* Visit Count */
527
528         return NF_HOOK(PF_DECnet, NF_DN_PRE_ROUTING, skb, skb->dev, NULL, dn_route_rx_packet);
529
530 drop_it:
531         kfree_skb(skb);
532         return NET_RX_DROP;
533 }
534
535
536
537 static int dn_route_rx_short(struct sk_buff *skb)
538 {
539         struct dn_skb_cb *cb = DN_SKB_CB(skb);
540         unsigned char *ptr = skb->data;
541
542         if (!pskb_may_pull(skb, 6)) /* 5 for short header + 1 for shortest nsp */
543                 goto drop_it;
544
545         skb_pull(skb, 5);
546         skb_reset_transport_header(skb);
547
548         cb->dst = *(__le16 *)ptr;
549         ptr += 2;
550         cb->src = *(__le16 *)ptr;
551         ptr += 2;
552         cb->hops = *ptr & 0x3f;
553
554         return NF_HOOK(PF_DECnet, NF_DN_PRE_ROUTING, skb, skb->dev, NULL, dn_route_rx_packet);
555
556 drop_it:
557         kfree_skb(skb);
558         return NET_RX_DROP;
559 }
560
561 static int dn_route_discard(struct sk_buff *skb)
562 {
563         /*
564          * I know we drop the packet here, but thats considered success in
565          * this case
566          */
567         kfree_skb(skb);
568         return NET_RX_SUCCESS;
569 }
570
571 static int dn_route_ptp_hello(struct sk_buff *skb)
572 {
573         dn_dev_hello(skb);
574         dn_neigh_pointopoint_hello(skb);
575         return NET_RX_SUCCESS;
576 }
577
578 int dn_route_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
579 {
580         struct dn_skb_cb *cb;
581         unsigned char flags = 0;
582         __u16 len = dn_ntohs(*(__le16 *)skb->data);
583         struct dn_dev *dn = (struct dn_dev *)dev->dn_ptr;
584         unsigned char padlen = 0;
585
586         if (dn == NULL)
587                 goto dump_it;
588
589         if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
590                 goto out;
591
592         if (!pskb_may_pull(skb, 3))
593                 goto dump_it;
594
595         skb_pull(skb, 2);
596
597         if (len > skb->len)
598                 goto dump_it;
599
600         skb_trim(skb, len);
601
602         flags = *skb->data;
603
604         cb = DN_SKB_CB(skb);
605         cb->stamp = jiffies;
606         cb->iif = dev->ifindex;
607
608         /*
609          * If we have padding, remove it.
610          */
611         if (flags & DN_RT_F_PF) {
612                 padlen = flags & ~DN_RT_F_PF;
613                 if (!pskb_may_pull(skb, padlen + 1))
614                         goto dump_it;
615                 skb_pull(skb, padlen);
616                 flags = *skb->data;
617         }
618
619         skb_reset_network_header(skb);
620
621         /*
622          * Weed out future version DECnet
623          */
624         if (flags & DN_RT_F_VER)
625                 goto dump_it;
626
627         cb->rt_flags = flags;
628
629         if (decnet_debug_level & 1)
630                 printk(KERN_DEBUG
631                         "dn_route_rcv: got 0x%02x from %s [%d %d %d]\n",
632                         (int)flags, (dev) ? dev->name : "???", len, skb->len,
633                         padlen);
634
635         if (flags & DN_RT_PKT_CNTL) {
636                 if (unlikely(skb_linearize(skb)))
637                         goto dump_it;
638
639                 switch(flags & DN_RT_CNTL_MSK) {
640                         case DN_RT_PKT_INIT:
641                                 dn_dev_init_pkt(skb);
642                                 break;
643                         case DN_RT_PKT_VERI:
644                                 dn_dev_veri_pkt(skb);
645                                 break;
646                 }
647
648                 if (dn->parms.state != DN_DEV_S_RU)
649                         goto dump_it;
650
651                 switch(flags & DN_RT_CNTL_MSK) {
652                         case DN_RT_PKT_HELO:
653                                 return NF_HOOK(PF_DECnet, NF_DN_HELLO, skb, skb->dev, NULL, dn_route_ptp_hello);
654
655                         case DN_RT_PKT_L1RT:
656                         case DN_RT_PKT_L2RT:
657                                 return NF_HOOK(PF_DECnet, NF_DN_ROUTE, skb, skb->dev, NULL, dn_route_discard);
658                         case DN_RT_PKT_ERTH:
659                                 return NF_HOOK(PF_DECnet, NF_DN_HELLO, skb, skb->dev, NULL, dn_neigh_router_hello);
660
661                         case DN_RT_PKT_EEDH:
662                                 return NF_HOOK(PF_DECnet, NF_DN_HELLO, skb, skb->dev, NULL, dn_neigh_endnode_hello);
663                 }
664         } else {
665                 if (dn->parms.state != DN_DEV_S_RU)
666                         goto dump_it;
667
668                 skb_pull(skb, 1); /* Pull flags */
669
670                 switch(flags & DN_RT_PKT_MSK) {
671                         case DN_RT_PKT_LONG:
672                                 return dn_route_rx_long(skb);
673                         case DN_RT_PKT_SHORT:
674                                 return dn_route_rx_short(skb);
675                 }
676         }
677
678 dump_it:
679         kfree_skb(skb);
680 out:
681         return NET_RX_DROP;
682 }
683
684 static int dn_output(struct sk_buff *skb)
685 {
686         struct dst_entry *dst = skb->dst;
687         struct dn_route *rt = (struct dn_route *)dst;
688         struct net_device *dev = dst->dev;
689         struct dn_skb_cb *cb = DN_SKB_CB(skb);
690         struct neighbour *neigh;
691
692         int err = -EINVAL;
693
694         if ((neigh = dst->neighbour) == NULL)
695                 goto error;
696
697         skb->dev = dev;
698
699         cb->src = rt->rt_saddr;
700         cb->dst = rt->rt_daddr;
701
702         /*
703          * Always set the Intra-Ethernet bit on all outgoing packets
704          * originated on this node. Only valid flag from upper layers
705          * is return-to-sender-requested. Set hop count to 0 too.
706          */
707         cb->rt_flags &= ~DN_RT_F_RQR;
708         cb->rt_flags |= DN_RT_F_IE;
709         cb->hops = 0;
710
711         return NF_HOOK(PF_DECnet, NF_DN_LOCAL_OUT, skb, NULL, dev, neigh->output);
712
713 error:
714         if (net_ratelimit())
715                 printk(KERN_DEBUG "dn_output: This should not happen\n");
716
717         kfree_skb(skb);
718
719         return err;
720 }
721
722 static int dn_forward(struct sk_buff *skb)
723 {
724         struct dn_skb_cb *cb = DN_SKB_CB(skb);
725         struct dst_entry *dst = skb->dst;
726         struct dn_dev *dn_db = dst->dev->dn_ptr;
727         struct dn_route *rt;
728         struct neighbour *neigh = dst->neighbour;
729         int header_len;
730 #ifdef CONFIG_NETFILTER
731         struct net_device *dev = skb->dev;
732 #endif
733
734         if (skb->pkt_type != PACKET_HOST)
735                 goto drop;
736
737         /* Ensure that we have enough space for headers */
738         rt = (struct dn_route *)skb->dst;
739         header_len = dn_db->use_long ? 21 : 6;
740         if (skb_cow(skb, LL_RESERVED_SPACE(rt->u.dst.dev)+header_len))
741                 goto drop;
742
743         /*
744          * Hop count exceeded.
745          */
746         if (++cb->hops > 30)
747                 goto drop;
748
749         skb->dev = rt->u.dst.dev;
750
751         /*
752          * If packet goes out same interface it came in on, then set
753          * the Intra-Ethernet bit. This has no effect for short
754          * packets, so we don't need to test for them here.
755          */
756         cb->rt_flags &= ~DN_RT_F_IE;
757         if (rt->rt_flags & RTCF_DOREDIRECT)
758                 cb->rt_flags |= DN_RT_F_IE;
759
760         return NF_HOOK(PF_DECnet, NF_DN_FORWARD, skb, dev, skb->dev, neigh->output);
761
762 drop:
763         kfree_skb(skb);
764         return NET_RX_DROP;
765 }
766
767 /*
768  * Drop packet. This is used for endnodes and for
769  * when we should not be forwarding packets from
770  * this dest.
771  */
772 static int dn_blackhole(struct sk_buff *skb)
773 {
774         kfree_skb(skb);
775         return NET_RX_DROP;
776 }
777
778 /*
779  * Used to catch bugs. This should never normally get
780  * called.
781  */
782 static int dn_rt_bug(struct sk_buff *skb)
783 {
784         if (net_ratelimit()) {
785                 struct dn_skb_cb *cb = DN_SKB_CB(skb);
786
787                 printk(KERN_DEBUG "dn_rt_bug: skb from:%04x to:%04x\n",
788                                 dn_ntohs(cb->src), dn_ntohs(cb->dst));
789         }
790
791         kfree_skb(skb);
792
793         return NET_RX_BAD;
794 }
795
796 static int dn_rt_set_next_hop(struct dn_route *rt, struct dn_fib_res *res)
797 {
798         struct dn_fib_info *fi = res->fi;
799         struct net_device *dev = rt->u.dst.dev;
800         struct neighbour *n;
801         unsigned mss;
802
803         if (fi) {
804                 if (DN_FIB_RES_GW(*res) &&
805                     DN_FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
806                         rt->rt_gateway = DN_FIB_RES_GW(*res);
807                 memcpy(rt->u.dst.metrics, fi->fib_metrics,
808                        sizeof(rt->u.dst.metrics));
809         }
810         rt->rt_type = res->type;
811
812         if (dev != NULL && rt->u.dst.neighbour == NULL) {
813                 n = __neigh_lookup_errno(&dn_neigh_table, &rt->rt_gateway, dev);
814                 if (IS_ERR(n))
815                         return PTR_ERR(n);
816                 rt->u.dst.neighbour = n;
817         }
818
819         if (rt->u.dst.metrics[RTAX_MTU-1] == 0 ||
820             rt->u.dst.metrics[RTAX_MTU-1] > rt->u.dst.dev->mtu)
821                 rt->u.dst.metrics[RTAX_MTU-1] = rt->u.dst.dev->mtu;
822         mss = dn_mss_from_pmtu(dev, dst_mtu(&rt->u.dst));
823         if (rt->u.dst.metrics[RTAX_ADVMSS-1] == 0 ||
824             rt->u.dst.metrics[RTAX_ADVMSS-1] > mss)
825                 rt->u.dst.metrics[RTAX_ADVMSS-1] = mss;
826         return 0;
827 }
828
829 static inline int dn_match_addr(__le16 addr1, __le16 addr2)
830 {
831         __u16 tmp = dn_ntohs(addr1) ^ dn_ntohs(addr2);
832         int match = 16;
833         while(tmp) {
834                 tmp >>= 1;
835                 match--;
836         }
837         return match;
838 }
839
840 static __le16 dnet_select_source(const struct net_device *dev, __le16 daddr, int scope)
841 {
842         __le16 saddr = 0;
843         struct dn_dev *dn_db = dev->dn_ptr;
844         struct dn_ifaddr *ifa;
845         int best_match = 0;
846         int ret;
847
848         read_lock(&dev_base_lock);
849         for(ifa = dn_db->ifa_list; ifa; ifa = ifa->ifa_next) {
850                 if (ifa->ifa_scope > scope)
851                         continue;
852                 if (!daddr) {
853                         saddr = ifa->ifa_local;
854                         break;
855                 }
856                 ret = dn_match_addr(daddr, ifa->ifa_local);
857                 if (ret > best_match)
858                         saddr = ifa->ifa_local;
859                 if (best_match == 0)
860                         saddr = ifa->ifa_local;
861         }
862         read_unlock(&dev_base_lock);
863
864         return saddr;
865 }
866
867 static inline __le16 __dn_fib_res_prefsrc(struct dn_fib_res *res)
868 {
869         return dnet_select_source(DN_FIB_RES_DEV(*res), DN_FIB_RES_GW(*res), res->scope);
870 }
871
872 static inline __le16 dn_fib_rules_map_destination(__le16 daddr, struct dn_fib_res *res)
873 {
874         __le16 mask = dnet_make_mask(res->prefixlen);
875         return (daddr&~mask)|res->fi->fib_nh->nh_gw;
876 }
877
878 static int dn_route_output_slow(struct dst_entry **pprt, const struct flowi *oldflp, int try_hard)
879 {
880         struct flowi fl = { .nl_u = { .dn_u =
881                                       { .daddr = oldflp->fld_dst,
882                                         .saddr = oldflp->fld_src,
883                                         .scope = RT_SCOPE_UNIVERSE,
884                                      } },
885                             .mark = oldflp->mark,
886                             .iif = loopback_dev.ifindex,
887                             .oif = oldflp->oif };
888         struct dn_route *rt = NULL;
889         struct net_device *dev_out = NULL;
890         struct neighbour *neigh = NULL;
891         unsigned hash;
892         unsigned flags = 0;
893         struct dn_fib_res res = { .fi = NULL, .type = RTN_UNICAST };
894         int err;
895         int free_res = 0;
896         __le16 gateway = 0;
897
898         if (decnet_debug_level & 16)
899                 printk(KERN_DEBUG
900                        "dn_route_output_slow: dst=%04x src=%04x mark=%d"
901                        " iif=%d oif=%d\n", dn_ntohs(oldflp->fld_dst),
902                        dn_ntohs(oldflp->fld_src),
903                        oldflp->mark, loopback_dev.ifindex, oldflp->oif);
904
905         /* If we have an output interface, verify its a DECnet device */
906         if (oldflp->oif) {
907                 dev_out = dev_get_by_index(oldflp->oif);
908                 err = -ENODEV;
909                 if (dev_out && dev_out->dn_ptr == NULL) {
910                         dev_put(dev_out);
911                         dev_out = NULL;
912                 }
913                 if (dev_out == NULL)
914                         goto out;
915         }
916
917         /* If we have a source address, verify that its a local address */
918         if (oldflp->fld_src) {
919                 err = -EADDRNOTAVAIL;
920
921                 if (dev_out) {
922                         if (dn_dev_islocal(dev_out, oldflp->fld_src))
923                                 goto source_ok;
924                         dev_put(dev_out);
925                         goto out;
926                 }
927                 read_lock(&dev_base_lock);
928                 for(dev_out = dev_base; dev_out; dev_out = dev_out->next) {
929                         if (!dev_out->dn_ptr)
930                                 continue;
931                         if (!dn_dev_islocal(dev_out, oldflp->fld_src))
932                                 continue;
933                         if ((dev_out->flags & IFF_LOOPBACK) &&
934                             oldflp->fld_dst &&
935                             !dn_dev_islocal(dev_out, oldflp->fld_dst))
936                                 continue;
937                         break;
938                 }
939                 read_unlock(&dev_base_lock);
940                 if (dev_out == NULL)
941                         goto out;
942                 dev_hold(dev_out);
943 source_ok:
944                 ;
945         }
946
947         /* No destination? Assume its local */
948         if (!fl.fld_dst) {
949                 fl.fld_dst = fl.fld_src;
950
951                 err = -EADDRNOTAVAIL;
952                 if (dev_out)
953                         dev_put(dev_out);
954                 dev_out = &loopback_dev;
955                 dev_hold(dev_out);
956                 if (!fl.fld_dst) {
957                         fl.fld_dst =
958                         fl.fld_src = dnet_select_source(dev_out, 0,
959                                                        RT_SCOPE_HOST);
960                         if (!fl.fld_dst)
961                                 goto out;
962                 }
963                 fl.oif = loopback_dev.ifindex;
964                 res.type = RTN_LOCAL;
965                 goto make_route;
966         }
967
968         if (decnet_debug_level & 16)
969                 printk(KERN_DEBUG
970                        "dn_route_output_slow: initial checks complete."
971                        " dst=%o4x src=%04x oif=%d try_hard=%d\n",
972                        dn_ntohs(fl.fld_dst), dn_ntohs(fl.fld_src),
973                        fl.oif, try_hard);
974
975         /*
976          * N.B. If the kernel is compiled without router support then
977          * dn_fib_lookup() will evaluate to non-zero so this if () block
978          * will always be executed.
979          */
980         err = -ESRCH;
981         if (try_hard || (err = dn_fib_lookup(&fl, &res)) != 0) {
982                 struct dn_dev *dn_db;
983                 if (err != -ESRCH)
984                         goto out;
985                 /*
986                  * Here the fallback is basically the standard algorithm for
987                  * routing in endnodes which is described in the DECnet routing
988                  * docs
989                  *
990                  * If we are not trying hard, look in neighbour cache.
991                  * The result is tested to ensure that if a specific output
992                  * device/source address was requested, then we honour that
993                  * here
994                  */
995                 if (!try_hard) {
996                         neigh = neigh_lookup_nodev(&dn_neigh_table, &fl.fld_dst);
997                         if (neigh) {
998                                 if ((oldflp->oif &&
999                                     (neigh->dev->ifindex != oldflp->oif)) ||
1000                                     (oldflp->fld_src &&
1001                                     (!dn_dev_islocal(neigh->dev,
1002                                                       oldflp->fld_src)))) {
1003                                         neigh_release(neigh);
1004                                         neigh = NULL;
1005                                 } else {
1006                                         if (dev_out)
1007                                                 dev_put(dev_out);
1008                                         if (dn_dev_islocal(neigh->dev, fl.fld_dst)) {
1009                                                 dev_out = &loopback_dev;
1010                                                 res.type = RTN_LOCAL;
1011                                         } else {
1012                                                 dev_out = neigh->dev;
1013                                         }
1014                                         dev_hold(dev_out);
1015                                         goto select_source;
1016                                 }
1017                         }
1018                 }
1019
1020                 /* Not there? Perhaps its a local address */
1021                 if (dev_out == NULL)
1022                         dev_out = dn_dev_get_default();
1023                 err = -ENODEV;
1024                 if (dev_out == NULL)
1025                         goto out;
1026                 dn_db = dev_out->dn_ptr;
1027                 /* Possible improvement - check all devices for local addr */
1028                 if (dn_dev_islocal(dev_out, fl.fld_dst)) {
1029                         dev_put(dev_out);
1030                         dev_out = &loopback_dev;
1031                         dev_hold(dev_out);
1032                         res.type = RTN_LOCAL;
1033                         goto select_source;
1034                 }
1035                 /* Not local either.... try sending it to the default router */
1036                 neigh = neigh_clone(dn_db->router);
1037                 BUG_ON(neigh && neigh->dev != dev_out);
1038
1039                 /* Ok then, we assume its directly connected and move on */
1040 select_source:
1041                 if (neigh)
1042                         gateway = ((struct dn_neigh *)neigh)->addr;
1043                 if (gateway == 0)
1044                         gateway = fl.fld_dst;
1045                 if (fl.fld_src == 0) {
1046                         fl.fld_src = dnet_select_source(dev_out, gateway,
1047                                                          res.type == RTN_LOCAL ?
1048                                                          RT_SCOPE_HOST :
1049                                                          RT_SCOPE_LINK);
1050                         if (fl.fld_src == 0 && res.type != RTN_LOCAL)
1051                                 goto e_addr;
1052                 }
1053                 fl.oif = dev_out->ifindex;
1054                 goto make_route;
1055         }
1056         free_res = 1;
1057
1058         if (res.type == RTN_NAT)
1059                 goto e_inval;
1060
1061         if (res.type == RTN_LOCAL) {
1062                 if (!fl.fld_src)
1063                         fl.fld_src = fl.fld_dst;
1064                 if (dev_out)
1065                         dev_put(dev_out);
1066                 dev_out = &loopback_dev;
1067                 dev_hold(dev_out);
1068                 fl.oif = dev_out->ifindex;
1069                 if (res.fi)
1070                         dn_fib_info_put(res.fi);
1071                 res.fi = NULL;
1072                 goto make_route;
1073         }
1074
1075         if (res.fi->fib_nhs > 1 && fl.oif == 0)
1076                 dn_fib_select_multipath(&fl, &res);
1077
1078         /*
1079          * We could add some logic to deal with default routes here and
1080          * get rid of some of the special casing above.
1081          */
1082
1083         if (!fl.fld_src)
1084                 fl.fld_src = DN_FIB_RES_PREFSRC(res);
1085
1086         if (dev_out)
1087                 dev_put(dev_out);
1088         dev_out = DN_FIB_RES_DEV(res);
1089         dev_hold(dev_out);
1090         fl.oif = dev_out->ifindex;
1091         gateway = DN_FIB_RES_GW(res);
1092
1093 make_route:
1094         if (dev_out->flags & IFF_LOOPBACK)
1095                 flags |= RTCF_LOCAL;
1096
1097         rt = dst_alloc(&dn_dst_ops);
1098         if (rt == NULL)
1099                 goto e_nobufs;
1100
1101         atomic_set(&rt->u.dst.__refcnt, 1);
1102         rt->u.dst.flags   = DST_HOST;
1103
1104         rt->fl.fld_src    = oldflp->fld_src;
1105         rt->fl.fld_dst    = oldflp->fld_dst;
1106         rt->fl.oif        = oldflp->oif;
1107         rt->fl.iif        = 0;
1108         rt->fl.mark       = oldflp->mark;
1109
1110         rt->rt_saddr      = fl.fld_src;
1111         rt->rt_daddr      = fl.fld_dst;
1112         rt->rt_gateway    = gateway ? gateway : fl.fld_dst;
1113         rt->rt_local_src  = fl.fld_src;
1114
1115         rt->rt_dst_map    = fl.fld_dst;
1116         rt->rt_src_map    = fl.fld_src;
1117
1118         rt->u.dst.dev = dev_out;
1119         dev_hold(dev_out);
1120         rt->u.dst.neighbour = neigh;
1121         neigh = NULL;
1122
1123         rt->u.dst.lastuse = jiffies;
1124         rt->u.dst.output  = dn_output;
1125         rt->u.dst.input   = dn_rt_bug;
1126         rt->rt_flags      = flags;
1127         if (flags & RTCF_LOCAL)
1128                 rt->u.dst.input = dn_nsp_rx;
1129
1130         err = dn_rt_set_next_hop(rt, &res);
1131         if (err)
1132                 goto e_neighbour;
1133
1134         hash = dn_hash(rt->fl.fld_src, rt->fl.fld_dst);
1135         dn_insert_route(rt, hash, (struct dn_route **)pprt);
1136
1137 done:
1138         if (neigh)
1139                 neigh_release(neigh);
1140         if (free_res)
1141                 dn_fib_res_put(&res);
1142         if (dev_out)
1143                 dev_put(dev_out);
1144 out:
1145         return err;
1146
1147 e_addr:
1148         err = -EADDRNOTAVAIL;
1149         goto done;
1150 e_inval:
1151         err = -EINVAL;
1152         goto done;
1153 e_nobufs:
1154         err = -ENOBUFS;
1155         goto done;
1156 e_neighbour:
1157         dst_free(&rt->u.dst);
1158         goto e_nobufs;
1159 }
1160
1161
1162 /*
1163  * N.B. The flags may be moved into the flowi at some future stage.
1164  */
1165 static int __dn_route_output_key(struct dst_entry **pprt, const struct flowi *flp, int flags)
1166 {
1167         unsigned hash = dn_hash(flp->fld_src, flp->fld_dst);
1168         struct dn_route *rt = NULL;
1169
1170         if (!(flags & MSG_TRYHARD)) {
1171                 rcu_read_lock_bh();
1172                 for(rt = rcu_dereference(dn_rt_hash_table[hash].chain); rt;
1173                         rt = rcu_dereference(rt->u.dst.dn_next)) {
1174                         if ((flp->fld_dst == rt->fl.fld_dst) &&
1175                             (flp->fld_src == rt->fl.fld_src) &&
1176                             (flp->mark == rt->fl.mark) &&
1177                             (rt->fl.iif == 0) &&
1178                             (rt->fl.oif == flp->oif)) {
1179                                 rt->u.dst.lastuse = jiffies;
1180                                 dst_hold(&rt->u.dst);
1181                                 rt->u.dst.__use++;
1182                                 rcu_read_unlock_bh();
1183                                 *pprt = &rt->u.dst;
1184                                 return 0;
1185                         }
1186                 }
1187                 rcu_read_unlock_bh();
1188         }
1189
1190         return dn_route_output_slow(pprt, flp, flags);
1191 }
1192
1193 static int dn_route_output_key(struct dst_entry **pprt, struct flowi *flp, int flags)
1194 {
1195         int err;
1196
1197         err = __dn_route_output_key(pprt, flp, flags);
1198         if (err == 0 && flp->proto) {
1199                 err = xfrm_lookup(pprt, flp, NULL, 0);
1200         }
1201         return err;
1202 }
1203
1204 int dn_route_output_sock(struct dst_entry **pprt, struct flowi *fl, struct sock *sk, int flags)
1205 {
1206         int err;
1207
1208         err = __dn_route_output_key(pprt, fl, flags & MSG_TRYHARD);
1209         if (err == 0 && fl->proto) {
1210                 err = xfrm_lookup(pprt, fl, sk, !(flags & MSG_DONTWAIT));
1211         }
1212         return err;
1213 }
1214
1215 static int dn_route_input_slow(struct sk_buff *skb)
1216 {
1217         struct dn_route *rt = NULL;
1218         struct dn_skb_cb *cb = DN_SKB_CB(skb);
1219         struct net_device *in_dev = skb->dev;
1220         struct net_device *out_dev = NULL;
1221         struct dn_dev *dn_db;
1222         struct neighbour *neigh = NULL;
1223         unsigned hash;
1224         int flags = 0;
1225         __le16 gateway = 0;
1226         __le16 local_src = 0;
1227         struct flowi fl = { .nl_u = { .dn_u =
1228                                      { .daddr = cb->dst,
1229                                        .saddr = cb->src,
1230                                        .scope = RT_SCOPE_UNIVERSE,
1231                                     } },
1232                             .mark = skb->mark,
1233                             .iif = skb->dev->ifindex };
1234         struct dn_fib_res res = { .fi = NULL, .type = RTN_UNREACHABLE };
1235         int err = -EINVAL;
1236         int free_res = 0;
1237
1238         dev_hold(in_dev);
1239
1240         if ((dn_db = in_dev->dn_ptr) == NULL)
1241                 goto out;
1242
1243         /* Zero source addresses are not allowed */
1244         if (fl.fld_src == 0)
1245                 goto out;
1246
1247         /*
1248          * In this case we've just received a packet from a source
1249          * outside ourselves pretending to come from us. We don't
1250          * allow it any further to prevent routing loops, spoofing and
1251          * other nasties. Loopback packets already have the dst attached
1252          * so this only affects packets which have originated elsewhere.
1253          */
1254         err  = -ENOTUNIQ;
1255         if (dn_dev_islocal(in_dev, cb->src))
1256                 goto out;
1257
1258         err = dn_fib_lookup(&fl, &res);
1259         if (err) {
1260                 if (err != -ESRCH)
1261                         goto out;
1262                 /*
1263                  * Is the destination us ?
1264                  */
1265                 if (!dn_dev_islocal(in_dev, cb->dst))
1266                         goto e_inval;
1267
1268                 res.type = RTN_LOCAL;
1269         } else {
1270                 __le16 src_map = fl.fld_src;
1271                 free_res = 1;
1272
1273                 out_dev = DN_FIB_RES_DEV(res);
1274                 if (out_dev == NULL) {
1275                         if (net_ratelimit())
1276                                 printk(KERN_CRIT "Bug in dn_route_input_slow() "
1277                                                  "No output device\n");
1278                         goto e_inval;
1279                 }
1280                 dev_hold(out_dev);
1281
1282                 if (res.r)
1283                         src_map = fl.fld_src; /* no NAT support for now */
1284
1285                 gateway = DN_FIB_RES_GW(res);
1286                 if (res.type == RTN_NAT) {
1287                         fl.fld_dst = dn_fib_rules_map_destination(fl.fld_dst, &res);
1288                         dn_fib_res_put(&res);
1289                         free_res = 0;
1290                         if (dn_fib_lookup(&fl, &res))
1291                                 goto e_inval;
1292                         free_res = 1;
1293                         if (res.type != RTN_UNICAST)
1294                                 goto e_inval;
1295                         flags |= RTCF_DNAT;
1296                         gateway = fl.fld_dst;
1297                 }
1298                 fl.fld_src = src_map;
1299         }
1300
1301         switch(res.type) {
1302         case RTN_UNICAST:
1303                 /*
1304                  * Forwarding check here, we only check for forwarding
1305                  * being turned off, if you want to only forward intra
1306                  * area, its up to you to set the routing tables up
1307                  * correctly.
1308                  */
1309                 if (dn_db->parms.forwarding == 0)
1310                         goto e_inval;
1311
1312                 if (res.fi->fib_nhs > 1 && fl.oif == 0)
1313                         dn_fib_select_multipath(&fl, &res);
1314
1315                 /*
1316                  * Check for out_dev == in_dev. We use the RTCF_DOREDIRECT
1317                  * flag as a hint to set the intra-ethernet bit when
1318                  * forwarding. If we've got NAT in operation, we don't do
1319                  * this optimisation.
1320                  */
1321                 if (out_dev == in_dev && !(flags & RTCF_NAT))
1322                         flags |= RTCF_DOREDIRECT;
1323
1324                 local_src = DN_FIB_RES_PREFSRC(res);
1325
1326         case RTN_BLACKHOLE:
1327         case RTN_UNREACHABLE:
1328                 break;
1329         case RTN_LOCAL:
1330                 flags |= RTCF_LOCAL;
1331                 fl.fld_src = cb->dst;
1332                 fl.fld_dst = cb->src;
1333
1334                 /* Routing tables gave us a gateway */
1335                 if (gateway)
1336                         goto make_route;
1337
1338                 /* Packet was intra-ethernet, so we know its on-link */
1339                 if (cb->rt_flags & DN_RT_F_IE) {
1340                         gateway = cb->src;
1341                         flags |= RTCF_DIRECTSRC;
1342                         goto make_route;
1343                 }
1344
1345                 /* Use the default router if there is one */
1346                 neigh = neigh_clone(dn_db->router);
1347                 if (neigh) {
1348                         gateway = ((struct dn_neigh *)neigh)->addr;
1349                         goto make_route;
1350                 }
1351
1352                 /* Close eyes and pray */
1353                 gateway = cb->src;
1354                 flags |= RTCF_DIRECTSRC;
1355                 goto make_route;
1356         default:
1357                 goto e_inval;
1358         }
1359
1360 make_route:
1361         rt = dst_alloc(&dn_dst_ops);
1362         if (rt == NULL)
1363                 goto e_nobufs;
1364
1365         rt->rt_saddr      = fl.fld_src;
1366         rt->rt_daddr      = fl.fld_dst;
1367         rt->rt_gateway    = fl.fld_dst;
1368         if (gateway)
1369                 rt->rt_gateway = gateway;
1370         rt->rt_local_src  = local_src ? local_src : rt->rt_saddr;
1371
1372         rt->rt_dst_map    = fl.fld_dst;
1373         rt->rt_src_map    = fl.fld_src;
1374
1375         rt->fl.fld_src    = cb->src;
1376         rt->fl.fld_dst    = cb->dst;
1377         rt->fl.oif        = 0;
1378         rt->fl.iif        = in_dev->ifindex;
1379         rt->fl.mark       = fl.mark;
1380
1381         rt->u.dst.flags = DST_HOST;
1382         rt->u.dst.neighbour = neigh;
1383         rt->u.dst.dev = out_dev;
1384         rt->u.dst.lastuse = jiffies;
1385         rt->u.dst.output = dn_rt_bug;
1386         switch(res.type) {
1387                 case RTN_UNICAST:
1388                         rt->u.dst.input = dn_forward;
1389                         break;
1390                 case RTN_LOCAL:
1391                         rt->u.dst.output = dn_output;
1392                         rt->u.dst.input = dn_nsp_rx;
1393                         rt->u.dst.dev = in_dev;
1394                         flags |= RTCF_LOCAL;
1395                         break;
1396                 default:
1397                 case RTN_UNREACHABLE:
1398                 case RTN_BLACKHOLE:
1399                         rt->u.dst.input = dn_blackhole;
1400         }
1401         rt->rt_flags = flags;
1402         if (rt->u.dst.dev)
1403                 dev_hold(rt->u.dst.dev);
1404
1405         err = dn_rt_set_next_hop(rt, &res);
1406         if (err)
1407                 goto e_neighbour;
1408
1409         hash = dn_hash(rt->fl.fld_src, rt->fl.fld_dst);
1410         dn_insert_route(rt, hash, (struct dn_route **)&skb->dst);
1411
1412 done:
1413         if (neigh)
1414                 neigh_release(neigh);
1415         if (free_res)
1416                 dn_fib_res_put(&res);
1417         dev_put(in_dev);
1418         if (out_dev)
1419                 dev_put(out_dev);
1420 out:
1421         return err;
1422
1423 e_inval:
1424         err = -EINVAL;
1425         goto done;
1426
1427 e_nobufs:
1428         err = -ENOBUFS;
1429         goto done;
1430
1431 e_neighbour:
1432         dst_free(&rt->u.dst);
1433         goto done;
1434 }
1435
1436 int dn_route_input(struct sk_buff *skb)
1437 {
1438         struct dn_route *rt;
1439         struct dn_skb_cb *cb = DN_SKB_CB(skb);
1440         unsigned hash = dn_hash(cb->src, cb->dst);
1441
1442         if (skb->dst)
1443                 return 0;
1444
1445         rcu_read_lock();
1446         for(rt = rcu_dereference(dn_rt_hash_table[hash].chain); rt != NULL;
1447             rt = rcu_dereference(rt->u.dst.dn_next)) {
1448                 if ((rt->fl.fld_src == cb->src) &&
1449                     (rt->fl.fld_dst == cb->dst) &&
1450                     (rt->fl.oif == 0) &&
1451                     (rt->fl.mark == skb->mark) &&
1452                     (rt->fl.iif == cb->iif)) {
1453                         rt->u.dst.lastuse = jiffies;
1454                         dst_hold(&rt->u.dst);
1455                         rt->u.dst.__use++;
1456                         rcu_read_unlock();
1457                         skb->dst = (struct dst_entry *)rt;
1458                         return 0;
1459                 }
1460         }
1461         rcu_read_unlock();
1462
1463         return dn_route_input_slow(skb);
1464 }
1465
1466 static int dn_rt_fill_info(struct sk_buff *skb, u32 pid, u32 seq,
1467                            int event, int nowait, unsigned int flags)
1468 {
1469         struct dn_route *rt = (struct dn_route *)skb->dst;
1470         struct rtmsg *r;
1471         struct nlmsghdr *nlh;
1472         unsigned char *b = skb_tail_pointer(skb);
1473         long expires;
1474
1475         nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*r), flags);
1476         r = NLMSG_DATA(nlh);
1477         r->rtm_family = AF_DECnet;
1478         r->rtm_dst_len = 16;
1479         r->rtm_src_len = 0;
1480         r->rtm_tos = 0;
1481         r->rtm_table = RT_TABLE_MAIN;
1482         RTA_PUT_U32(skb, RTA_TABLE, RT_TABLE_MAIN);
1483         r->rtm_type = rt->rt_type;
1484         r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
1485         r->rtm_scope = RT_SCOPE_UNIVERSE;
1486         r->rtm_protocol = RTPROT_UNSPEC;
1487         if (rt->rt_flags & RTCF_NOTIFY)
1488                 r->rtm_flags |= RTM_F_NOTIFY;
1489         RTA_PUT(skb, RTA_DST, 2, &rt->rt_daddr);
1490         if (rt->fl.fld_src) {
1491                 r->rtm_src_len = 16;
1492                 RTA_PUT(skb, RTA_SRC, 2, &rt->fl.fld_src);
1493         }
1494         if (rt->u.dst.dev)
1495                 RTA_PUT(skb, RTA_OIF, sizeof(int), &rt->u.dst.dev->ifindex);
1496         /*
1497          * Note to self - change this if input routes reverse direction when
1498          * they deal only with inputs and not with replies like they do
1499          * currently.
1500          */
1501         RTA_PUT(skb, RTA_PREFSRC, 2, &rt->rt_local_src);
1502         if (rt->rt_daddr != rt->rt_gateway)
1503                 RTA_PUT(skb, RTA_GATEWAY, 2, &rt->rt_gateway);
1504         if (rtnetlink_put_metrics(skb, rt->u.dst.metrics) < 0)
1505                 goto rtattr_failure;
1506         expires = rt->u.dst.expires ? rt->u.dst.expires - jiffies : 0;
1507         if (rtnl_put_cacheinfo(skb, &rt->u.dst, 0, 0, 0, expires,
1508                                rt->u.dst.error) < 0)
1509                 goto rtattr_failure;
1510         if (rt->fl.iif)
1511                 RTA_PUT(skb, RTA_IIF, sizeof(int), &rt->fl.iif);
1512
1513         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1514         return skb->len;
1515
1516 nlmsg_failure:
1517 rtattr_failure:
1518         nlmsg_trim(skb, b);
1519         return -1;
1520 }
1521
1522 /*
1523  * This is called by both endnodes and routers now.
1524  */
1525 static int dn_cache_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh, void *arg)
1526 {
1527         struct rtattr **rta = arg;
1528         struct rtmsg *rtm = NLMSG_DATA(nlh);
1529         struct dn_route *rt = NULL;
1530         struct dn_skb_cb *cb;
1531         int err;
1532         struct sk_buff *skb;
1533         struct flowi fl;
1534
1535         memset(&fl, 0, sizeof(fl));
1536         fl.proto = DNPROTO_NSP;
1537
1538         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1539         if (skb == NULL)
1540                 return -ENOBUFS;
1541         skb_reset_mac_header(skb);
1542         cb = DN_SKB_CB(skb);
1543
1544         if (rta[RTA_SRC-1])
1545                 memcpy(&fl.fld_src, RTA_DATA(rta[RTA_SRC-1]), 2);
1546         if (rta[RTA_DST-1])
1547                 memcpy(&fl.fld_dst, RTA_DATA(rta[RTA_DST-1]), 2);
1548         if (rta[RTA_IIF-1])
1549                 memcpy(&fl.iif, RTA_DATA(rta[RTA_IIF-1]), sizeof(int));
1550
1551         if (fl.iif) {
1552                 struct net_device *dev;
1553                 if ((dev = dev_get_by_index(fl.iif)) == NULL) {
1554                         kfree_skb(skb);
1555                         return -ENODEV;
1556                 }
1557                 if (!dev->dn_ptr) {
1558                         dev_put(dev);
1559                         kfree_skb(skb);
1560                         return -ENODEV;
1561                 }
1562                 skb->protocol = __constant_htons(ETH_P_DNA_RT);
1563                 skb->dev = dev;
1564                 cb->src = fl.fld_src;
1565                 cb->dst = fl.fld_dst;
1566                 local_bh_disable();
1567                 err = dn_route_input(skb);
1568                 local_bh_enable();
1569                 memset(cb, 0, sizeof(struct dn_skb_cb));
1570                 rt = (struct dn_route *)skb->dst;
1571                 if (!err && -rt->u.dst.error)
1572                         err = rt->u.dst.error;
1573         } else {
1574                 int oif = 0;
1575                 if (rta[RTA_OIF - 1])
1576                         memcpy(&oif, RTA_DATA(rta[RTA_OIF - 1]), sizeof(int));
1577                 fl.oif = oif;
1578                 err = dn_route_output_key((struct dst_entry **)&rt, &fl, 0);
1579         }
1580
1581         if (skb->dev)
1582                 dev_put(skb->dev);
1583         skb->dev = NULL;
1584         if (err)
1585                 goto out_free;
1586         skb->dst = &rt->u.dst;
1587         if (rtm->rtm_flags & RTM_F_NOTIFY)
1588                 rt->rt_flags |= RTCF_NOTIFY;
1589
1590         err = dn_rt_fill_info(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq, RTM_NEWROUTE, 0, 0);
1591
1592         if (err == 0)
1593                 goto out_free;
1594         if (err < 0) {
1595                 err = -EMSGSIZE;
1596                 goto out_free;
1597         }
1598
1599         return rtnl_unicast(skb, NETLINK_CB(in_skb).pid);
1600
1601 out_free:
1602         kfree_skb(skb);
1603         return err;
1604 }
1605
1606 /*
1607  * For routers, this is called from dn_fib_dump, but for endnodes its
1608  * called directly from the rtnetlink dispatch table.
1609  */
1610 int dn_cache_dump(struct sk_buff *skb, struct netlink_callback *cb)
1611 {
1612         struct dn_route *rt;
1613         int h, s_h;
1614         int idx, s_idx;
1615
1616         if (NLMSG_PAYLOAD(cb->nlh, 0) < sizeof(struct rtmsg))
1617                 return -EINVAL;
1618         if (!(((struct rtmsg *)NLMSG_DATA(cb->nlh))->rtm_flags&RTM_F_CLONED))
1619                 return 0;
1620
1621         s_h = cb->args[0];
1622         s_idx = idx = cb->args[1];
1623         for(h = 0; h <= dn_rt_hash_mask; h++) {
1624                 if (h < s_h)
1625                         continue;
1626                 if (h > s_h)
1627                         s_idx = 0;
1628                 rcu_read_lock_bh();
1629                 for(rt = rcu_dereference(dn_rt_hash_table[h].chain), idx = 0;
1630                         rt;
1631                         rt = rcu_dereference(rt->u.dst.dn_next), idx++) {
1632                         if (idx < s_idx)
1633                                 continue;
1634                         skb->dst = dst_clone(&rt->u.dst);
1635                         if (dn_rt_fill_info(skb, NETLINK_CB(cb->skb).pid,
1636                                         cb->nlh->nlmsg_seq, RTM_NEWROUTE,
1637                                         1, NLM_F_MULTI) <= 0) {
1638                                 dst_release(xchg(&skb->dst, NULL));
1639                                 rcu_read_unlock_bh();
1640                                 goto done;
1641                         }
1642                         dst_release(xchg(&skb->dst, NULL));
1643                 }
1644                 rcu_read_unlock_bh();
1645         }
1646
1647 done:
1648         cb->args[0] = h;
1649         cb->args[1] = idx;
1650         return skb->len;
1651 }
1652
1653 #ifdef CONFIG_PROC_FS
1654 struct dn_rt_cache_iter_state {
1655         int bucket;
1656 };
1657
1658 static struct dn_route *dn_rt_cache_get_first(struct seq_file *seq)
1659 {
1660         struct dn_route *rt = NULL;
1661         struct dn_rt_cache_iter_state *s = seq->private;
1662
1663         for(s->bucket = dn_rt_hash_mask; s->bucket >= 0; --s->bucket) {
1664                 rcu_read_lock_bh();
1665                 rt = dn_rt_hash_table[s->bucket].chain;
1666                 if (rt)
1667                         break;
1668                 rcu_read_unlock_bh();
1669         }
1670         return rt;
1671 }
1672
1673 static struct dn_route *dn_rt_cache_get_next(struct seq_file *seq, struct dn_route *rt)
1674 {
1675         struct dn_rt_cache_iter_state *s = rcu_dereference(seq->private);
1676
1677         rt = rt->u.dst.dn_next;
1678         while(!rt) {
1679                 rcu_read_unlock_bh();
1680                 if (--s->bucket < 0)
1681                         break;
1682                 rcu_read_lock_bh();
1683                 rt = dn_rt_hash_table[s->bucket].chain;
1684         }
1685         return rt;
1686 }
1687
1688 static void *dn_rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
1689 {
1690         struct dn_route *rt = dn_rt_cache_get_first(seq);
1691
1692         if (rt) {
1693                 while(*pos && (rt = dn_rt_cache_get_next(seq, rt)))
1694                         --*pos;
1695         }
1696         return *pos ? NULL : rt;
1697 }
1698
1699 static void *dn_rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1700 {
1701         struct dn_route *rt = dn_rt_cache_get_next(seq, v);
1702         ++*pos;
1703         return rt;
1704 }
1705
1706 static void dn_rt_cache_seq_stop(struct seq_file *seq, void *v)
1707 {
1708         if (v)
1709                 rcu_read_unlock_bh();
1710 }
1711
1712 static int dn_rt_cache_seq_show(struct seq_file *seq, void *v)
1713 {
1714         struct dn_route *rt = v;
1715         char buf1[DN_ASCBUF_LEN], buf2[DN_ASCBUF_LEN];
1716
1717         seq_printf(seq, "%-8s %-7s %-7s %04d %04d %04d\n",
1718                         rt->u.dst.dev ? rt->u.dst.dev->name : "*",
1719                         dn_addr2asc(dn_ntohs(rt->rt_daddr), buf1),
1720                         dn_addr2asc(dn_ntohs(rt->rt_saddr), buf2),
1721                         atomic_read(&rt->u.dst.__refcnt),
1722                         rt->u.dst.__use,
1723                         (int) dst_metric(&rt->u.dst, RTAX_RTT));
1724         return 0;
1725 }
1726
1727 static struct seq_operations dn_rt_cache_seq_ops = {
1728         .start  = dn_rt_cache_seq_start,
1729         .next   = dn_rt_cache_seq_next,
1730         .stop   = dn_rt_cache_seq_stop,
1731         .show   = dn_rt_cache_seq_show,
1732 };
1733
1734 static int dn_rt_cache_seq_open(struct inode *inode, struct file *file)
1735 {
1736         struct seq_file *seq;
1737         int rc = -ENOMEM;
1738         struct dn_rt_cache_iter_state *s = kmalloc(sizeof(*s), GFP_KERNEL);
1739
1740         if (!s)
1741                 goto out;
1742         rc = seq_open(file, &dn_rt_cache_seq_ops);
1743         if (rc)
1744                 goto out_kfree;
1745         seq             = file->private_data;
1746         seq->private    = s;
1747         memset(s, 0, sizeof(*s));
1748 out:
1749         return rc;
1750 out_kfree:
1751         kfree(s);
1752         goto out;
1753 }
1754
1755 static const struct file_operations dn_rt_cache_seq_fops = {
1756         .owner   = THIS_MODULE,
1757         .open    = dn_rt_cache_seq_open,
1758         .read    = seq_read,
1759         .llseek  = seq_lseek,
1760         .release = seq_release_private,
1761 };
1762
1763 #endif /* CONFIG_PROC_FS */
1764
1765 void __init dn_route_init(void)
1766 {
1767         int i, goal, order;
1768
1769         dn_dst_ops.kmem_cachep =
1770                 kmem_cache_create("dn_dst_cache", sizeof(struct dn_route), 0,
1771                                   SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1772         init_timer(&dn_route_timer);
1773         dn_route_timer.function = dn_dst_check_expire;
1774         dn_route_timer.expires = jiffies + decnet_dst_gc_interval * HZ;
1775         add_timer(&dn_route_timer);
1776
1777         goal = num_physpages >> (26 - PAGE_SHIFT);
1778
1779         for(order = 0; (1UL << order) < goal; order++)
1780                 /* NOTHING */;
1781
1782         /*
1783          * Only want 1024 entries max, since the table is very, very unlikely
1784          * to be larger than that.
1785          */
1786         while(order && ((((1UL << order) * PAGE_SIZE) /
1787                                 sizeof(struct dn_rt_hash_bucket)) >= 2048))
1788                 order--;
1789
1790         do {
1791                 dn_rt_hash_mask = (1UL << order) * PAGE_SIZE /
1792                         sizeof(struct dn_rt_hash_bucket);
1793                 while(dn_rt_hash_mask & (dn_rt_hash_mask - 1))
1794                         dn_rt_hash_mask--;
1795                 dn_rt_hash_table = (struct dn_rt_hash_bucket *)
1796                         __get_free_pages(GFP_ATOMIC, order);
1797         } while (dn_rt_hash_table == NULL && --order > 0);
1798
1799         if (!dn_rt_hash_table)
1800                 panic("Failed to allocate DECnet route cache hash table\n");
1801
1802         printk(KERN_INFO
1803                 "DECnet: Routing cache hash table of %u buckets, %ldKbytes\n",
1804                 dn_rt_hash_mask,
1805                 (long)(dn_rt_hash_mask*sizeof(struct dn_rt_hash_bucket))/1024);
1806
1807         dn_rt_hash_mask--;
1808         for(i = 0; i <= dn_rt_hash_mask; i++) {
1809                 spin_lock_init(&dn_rt_hash_table[i].lock);
1810                 dn_rt_hash_table[i].chain = NULL;
1811         }
1812
1813         dn_dst_ops.gc_thresh = (dn_rt_hash_mask + 1);
1814
1815         proc_net_fops_create("decnet_cache", S_IRUGO, &dn_rt_cache_seq_fops);
1816
1817 #ifdef CONFIG_DECNET_ROUTER
1818         rtnl_register(PF_DECnet, RTM_GETROUTE, dn_cache_getroute, dn_fib_dump);
1819 #else
1820         rtnl_register(PF_DECnet, RTM_GETROUTE, dn_cache_getroute,
1821                       dn_cache_dump);
1822 #endif
1823 }
1824
1825 void __exit dn_route_cleanup(void)
1826 {
1827         del_timer(&dn_route_timer);
1828         dn_run_flush(0);
1829
1830         proc_net_remove("decnet_cache");
1831 }
1832