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