Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jbarnes/pci
[pandora-kernel.git] / net / decnet / dn_route.c
1 /*
2  * DECnet       An implementation of the DECnet protocol suite for the LINUX
3  *              operating system.  DECnet is implemented using the  BSD Socket
4  *              interface as the means of communication with the user level.
5  *
6  *              DECnet Routing Functions (Endnode and Router)
7  *
8  * Authors:     Steve Whitehouse <SteveW@ACM.org>
9  *              Eduardo Marcelo Serrat <emserrat@geocities.com>
10  *
11  * Changes:
12  *              Steve Whitehouse : Fixes to allow "intra-ethernet" and
13  *                                 "return-to-sender" bits on outgoing
14  *                                 packets.
15  *              Steve Whitehouse : Timeouts for cached routes.
16  *              Steve Whitehouse : Use dst cache for input routes too.
17  *              Steve Whitehouse : Fixed error values in dn_send_skb.
18  *              Steve Whitehouse : Rework routing functions to better fit
19  *                                 DECnet routing design
20  *              Alexey Kuznetsov : New SMP locking
21  *              Steve Whitehouse : More SMP locking changes & dn_cache_dump()
22  *              Steve Whitehouse : Prerouting NF hook, now really is prerouting.
23  *                                 Fixed possible skb leak in rtnetlink funcs.
24  *              Steve Whitehouse : Dave Miller's dynamic hash table sizing and
25  *                                 Alexey Kuznetsov's finer grained locking
26  *                                 from ipv4/route.c.
27  *              Steve Whitehouse : Routing is now starting to look like a
28  *                                 sensible set of code now, mainly due to
29  *                                 my copying the IPv4 routing code. The
30  *                                 hooks here are modified and will continue
31  *                                 to evolve for a while.
32  *              Steve Whitehouse : Real SMP at last :-) Also new netfilter
33  *                                 stuff. Look out raw sockets your days
34  *                                 are numbered!
35  *              Steve Whitehouse : Added return-to-sender functions. Added
36  *                                 backlog congestion level return codes.
37  *              Steve Whitehouse : Fixed bug where routes were set up with
38  *                                 no ref count on net devices.
39  *              Steve Whitehouse : RCU for the route cache
40  *              Steve Whitehouse : Preparations for the flow cache
41  *              Steve Whitehouse : Prepare for nonlinear skbs
42  */
43
44 /******************************************************************************
45     (c) 1995-1998 E.M. Serrat           emserrat@geocities.com
46
47     This program is free software; you can redistribute it and/or modify
48     it under the terms of the GNU General Public License as published by
49     the Free Software Foundation; either version 2 of the License, or
50     any later version.
51
52     This program is distributed in the hope that it will be useful,
53     but WITHOUT ANY WARRANTY; without even the implied warranty of
54     MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
55     GNU General Public License for more details.
56 *******************************************************************************/
57
58 #include <linux/errno.h>
59 #include <linux/types.h>
60 #include <linux/socket.h>
61 #include <linux/in.h>
62 #include <linux/kernel.h>
63 #include <linux/sockios.h>
64 #include <linux/net.h>
65 #include <linux/netdevice.h>
66 #include <linux/inet.h>
67 #include <linux/route.h>
68 #include <linux/in_route.h>
69 #include <linux/slab.h>
70 #include <net/sock.h>
71 #include <linux/mm.h>
72 #include <linux/proc_fs.h>
73 #include <linux/seq_file.h>
74 #include <linux/init.h>
75 #include <linux/rtnetlink.h>
76 #include <linux/string.h>
77 #include <linux/netfilter_decnet.h>
78 #include <linux/rcupdate.h>
79 #include <linux/times.h>
80 #include <linux/export.h>
81 #include <asm/errno.h>
82 #include <net/net_namespace.h>
83 #include <net/netlink.h>
84 #include <net/neighbour.h>
85 #include <net/dst.h>
86 #include <net/flow.h>
87 #include <net/fib_rules.h>
88 #include <net/dn.h>
89 #include <net/dn_dev.h>
90 #include <net/dn_nsp.h>
91 #include <net/dn_route.h>
92 #include <net/dn_neigh.h>
93 #include <net/dn_fib.h>
94
95 struct dn_rt_hash_bucket
96 {
97         struct dn_route __rcu *chain;
98         spinlock_t lock;
99 };
100
101 extern struct neigh_table dn_neigh_table;
102
103
104 static unsigned char dn_hiord_addr[6] = {0xAA,0x00,0x04,0x00,0x00,0x00};
105
106 static const int dn_rt_min_delay = 2 * HZ;
107 static const int dn_rt_max_delay = 10 * HZ;
108 static const int dn_rt_mtu_expires = 10 * 60 * HZ;
109
110 static unsigned long dn_rt_deadline;
111
112 static int dn_dst_gc(struct dst_ops *ops);
113 static struct dst_entry *dn_dst_check(struct dst_entry *, __u32);
114 static unsigned int dn_dst_default_advmss(const struct dst_entry *dst);
115 static unsigned int dn_dst_default_mtu(const struct dst_entry *dst);
116 static void dn_dst_destroy(struct dst_entry *);
117 static struct dst_entry *dn_dst_negative_advice(struct dst_entry *);
118 static void dn_dst_link_failure(struct sk_buff *);
119 static void dn_dst_update_pmtu(struct dst_entry *dst, u32 mtu);
120 static struct neighbour *dn_dst_neigh_lookup(const struct dst_entry *dst, const void *daddr);
121 static int dn_route_input(struct sk_buff *);
122 static void dn_run_flush(unsigned long dummy);
123
124 static struct dn_rt_hash_bucket *dn_rt_hash_table;
125 static unsigned dn_rt_hash_mask;
126
127 static struct timer_list dn_route_timer;
128 static DEFINE_TIMER(dn_rt_flush_timer, dn_run_flush, 0, 0);
129 int decnet_dst_gc_interval = 2;
130
131 static struct dst_ops dn_dst_ops = {
132         .family =               PF_DECnet,
133         .protocol =             cpu_to_be16(ETH_P_DNA_RT),
134         .gc_thresh =            128,
135         .gc =                   dn_dst_gc,
136         .check =                dn_dst_check,
137         .default_advmss =       dn_dst_default_advmss,
138         .default_mtu =          dn_dst_default_mtu,
139         .cow_metrics =          dst_cow_metrics_generic,
140         .destroy =              dn_dst_destroy,
141         .negative_advice =      dn_dst_negative_advice,
142         .link_failure =         dn_dst_link_failure,
143         .update_pmtu =          dn_dst_update_pmtu,
144         .neigh_lookup =         dn_dst_neigh_lookup,
145 };
146
147 static void dn_dst_destroy(struct dst_entry *dst)
148 {
149         dst_destroy_metrics_generic(dst);
150 }
151
152 static __inline__ unsigned dn_hash(__le16 src, __le16 dst)
153 {
154         __u16 tmp = (__u16 __force)(src ^ dst);
155         tmp ^= (tmp >> 3);
156         tmp ^= (tmp >> 5);
157         tmp ^= (tmp >> 10);
158         return dn_rt_hash_mask & (unsigned)tmp;
159 }
160
161 static inline void dnrt_free(struct dn_route *rt)
162 {
163         call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
164 }
165
166 static inline void dnrt_drop(struct dn_route *rt)
167 {
168         dst_release(&rt->dst);
169         call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
170 }
171
172 static void dn_dst_check_expire(unsigned long dummy)
173 {
174         int i;
175         struct dn_route *rt;
176         struct dn_route __rcu **rtp;
177         unsigned long now = jiffies;
178         unsigned long expire = 120 * HZ;
179
180         for (i = 0; i <= dn_rt_hash_mask; i++) {
181                 rtp = &dn_rt_hash_table[i].chain;
182
183                 spin_lock(&dn_rt_hash_table[i].lock);
184                 while ((rt = rcu_dereference_protected(*rtp,
185                                                 lockdep_is_held(&dn_rt_hash_table[i].lock))) != NULL) {
186                         if (atomic_read(&rt->dst.__refcnt) ||
187                                         (now - rt->dst.lastuse) < expire) {
188                                 rtp = &rt->dst.dn_next;
189                                 continue;
190                         }
191                         *rtp = rt->dst.dn_next;
192                         rt->dst.dn_next = NULL;
193                         dnrt_free(rt);
194                 }
195                 spin_unlock(&dn_rt_hash_table[i].lock);
196
197                 if ((jiffies - now) > 0)
198                         break;
199         }
200
201         mod_timer(&dn_route_timer, now + decnet_dst_gc_interval * HZ);
202 }
203
204 static int dn_dst_gc(struct dst_ops *ops)
205 {
206         struct dn_route *rt;
207         struct dn_route __rcu **rtp;
208         int i;
209         unsigned long now = jiffies;
210         unsigned long expire = 10 * HZ;
211
212         for (i = 0; i <= dn_rt_hash_mask; i++) {
213
214                 spin_lock_bh(&dn_rt_hash_table[i].lock);
215                 rtp = &dn_rt_hash_table[i].chain;
216
217                 while ((rt = rcu_dereference_protected(*rtp,
218                                                 lockdep_is_held(&dn_rt_hash_table[i].lock))) != NULL) {
219                         if (atomic_read(&rt->dst.__refcnt) ||
220                                         (now - rt->dst.lastuse) < expire) {
221                                 rtp = &rt->dst.dn_next;
222                                 continue;
223                         }
224                         *rtp = rt->dst.dn_next;
225                         rt->dst.dn_next = NULL;
226                         dnrt_drop(rt);
227                         break;
228                 }
229                 spin_unlock_bh(&dn_rt_hash_table[i].lock);
230         }
231
232         return 0;
233 }
234
235 /*
236  * The decnet standards don't impose a particular minimum mtu, what they
237  * do insist on is that the routing layer accepts a datagram of at least
238  * 230 bytes long. Here we have to subtract the routing header length from
239  * 230 to get the minimum acceptable mtu. If there is no neighbour, then we
240  * assume the worst and use a long header size.
241  *
242  * We update both the mtu and the advertised mss (i.e. the segment size we
243  * advertise to the other end).
244  */
245 static void dn_dst_update_pmtu(struct dst_entry *dst, u32 mtu)
246 {
247         struct neighbour *n = dst_get_neighbour(dst);
248         u32 min_mtu = 230;
249         struct dn_dev *dn;
250
251         dn = n ? rcu_dereference_raw(n->dev->dn_ptr) : NULL;
252
253         if (dn && dn->use_long == 0)
254                 min_mtu -= 6;
255         else
256                 min_mtu -= 21;
257
258         if (dst_metric(dst, RTAX_MTU) > mtu && mtu >= min_mtu) {
259                 if (!(dst_metric_locked(dst, RTAX_MTU))) {
260                         dst_metric_set(dst, RTAX_MTU, mtu);
261                         dst_set_expires(dst, dn_rt_mtu_expires);
262                 }
263                 if (!(dst_metric_locked(dst, RTAX_ADVMSS))) {
264                         u32 mss = mtu - DN_MAX_NSP_DATA_HEADER;
265                         u32 existing_mss = dst_metric_raw(dst, RTAX_ADVMSS);
266                         if (!existing_mss || existing_mss > mss)
267                                 dst_metric_set(dst, RTAX_ADVMSS, mss);
268                 }
269         }
270 }
271
272 /*
273  * When a route has been marked obsolete. (e.g. routing cache flush)
274  */
275 static struct dst_entry *dn_dst_check(struct dst_entry *dst, __u32 cookie)
276 {
277         return NULL;
278 }
279
280 static struct dst_entry *dn_dst_negative_advice(struct dst_entry *dst)
281 {
282         dst_release(dst);
283         return NULL;
284 }
285
286 static void dn_dst_link_failure(struct sk_buff *skb)
287 {
288 }
289
290 static inline int compare_keys(struct flowidn *fl1, struct flowidn *fl2)
291 {
292         return ((fl1->daddr ^ fl2->daddr) |
293                 (fl1->saddr ^ fl2->saddr) |
294                 (fl1->flowidn_mark ^ fl2->flowidn_mark) |
295                 (fl1->flowidn_scope ^ fl2->flowidn_scope) |
296                 (fl1->flowidn_oif ^ fl2->flowidn_oif) |
297                 (fl1->flowidn_iif ^ fl2->flowidn_iif)) == 0;
298 }
299
300 static int dn_insert_route(struct dn_route *rt, unsigned hash, struct dn_route **rp)
301 {
302         struct dn_route *rth;
303         struct dn_route __rcu **rthp;
304         unsigned long now = jiffies;
305
306         rthp = &dn_rt_hash_table[hash].chain;
307
308         spin_lock_bh(&dn_rt_hash_table[hash].lock);
309         while ((rth = rcu_dereference_protected(*rthp,
310                                                 lockdep_is_held(&dn_rt_hash_table[hash].lock))) != NULL) {
311                 if (compare_keys(&rth->fld, &rt->fld)) {
312                         /* Put it first */
313                         *rthp = rth->dst.dn_next;
314                         rcu_assign_pointer(rth->dst.dn_next,
315                                            dn_rt_hash_table[hash].chain);
316                         rcu_assign_pointer(dn_rt_hash_table[hash].chain, rth);
317
318                         dst_use(&rth->dst, now);
319                         spin_unlock_bh(&dn_rt_hash_table[hash].lock);
320
321                         dnrt_drop(rt);
322                         *rp = rth;
323                         return 0;
324                 }
325                 rthp = &rth->dst.dn_next;
326         }
327
328         rcu_assign_pointer(rt->dst.dn_next, dn_rt_hash_table[hash].chain);
329         rcu_assign_pointer(dn_rt_hash_table[hash].chain, rt);
330
331         dst_use(&rt->dst, now);
332         spin_unlock_bh(&dn_rt_hash_table[hash].lock);
333         *rp = rt;
334         return 0;
335 }
336
337 static void dn_run_flush(unsigned long dummy)
338 {
339         int i;
340         struct dn_route *rt, *next;
341
342         for (i = 0; i < dn_rt_hash_mask; i++) {
343                 spin_lock_bh(&dn_rt_hash_table[i].lock);
344
345                 if ((rt = xchg((struct dn_route **)&dn_rt_hash_table[i].chain, NULL)) == NULL)
346                         goto nothing_to_declare;
347
348                 for(; rt; rt = next) {
349                         next = rcu_dereference_raw(rt->dst.dn_next);
350                         RCU_INIT_POINTER(rt->dst.dn_next, NULL);
351                         dst_free((struct dst_entry *)rt);
352                 }
353
354 nothing_to_declare:
355                 spin_unlock_bh(&dn_rt_hash_table[i].lock);
356         }
357 }
358
359 static DEFINE_SPINLOCK(dn_rt_flush_lock);
360
361 void dn_rt_cache_flush(int delay)
362 {
363         unsigned long now = jiffies;
364         int user_mode = !in_interrupt();
365
366         if (delay < 0)
367                 delay = dn_rt_min_delay;
368
369         spin_lock_bh(&dn_rt_flush_lock);
370
371         if (del_timer(&dn_rt_flush_timer) && delay > 0 && dn_rt_deadline) {
372                 long tmo = (long)(dn_rt_deadline - now);
373
374                 if (user_mode && tmo < dn_rt_max_delay - dn_rt_min_delay)
375                         tmo = 0;
376
377                 if (delay > tmo)
378                         delay = tmo;
379         }
380
381         if (delay <= 0) {
382                 spin_unlock_bh(&dn_rt_flush_lock);
383                 dn_run_flush(0);
384                 return;
385         }
386
387         if (dn_rt_deadline == 0)
388                 dn_rt_deadline = now + dn_rt_max_delay;
389
390         dn_rt_flush_timer.expires = now + delay;
391         add_timer(&dn_rt_flush_timer);
392         spin_unlock_bh(&dn_rt_flush_lock);
393 }
394
395 /**
396  * dn_return_short - Return a short packet to its sender
397  * @skb: The packet to return
398  *
399  */
400 static int dn_return_short(struct sk_buff *skb)
401 {
402         struct dn_skb_cb *cb;
403         unsigned char *ptr;
404         __le16 *src;
405         __le16 *dst;
406
407         /* Add back headers */
408         skb_push(skb, skb->data - skb_network_header(skb));
409
410         if ((skb = skb_unshare(skb, GFP_ATOMIC)) == NULL)
411                 return NET_RX_DROP;
412
413         cb = DN_SKB_CB(skb);
414         /* Skip packet length and point to flags */
415         ptr = skb->data + 2;
416         *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
417
418         dst = (__le16 *)ptr;
419         ptr += 2;
420         src = (__le16 *)ptr;
421         ptr += 2;
422         *ptr = 0; /* Zero hop count */
423
424         swap(*src, *dst);
425
426         skb->pkt_type = PACKET_OUTGOING;
427         dn_rt_finish_output(skb, NULL, NULL);
428         return NET_RX_SUCCESS;
429 }
430
431 /**
432  * dn_return_long - Return a long packet to its sender
433  * @skb: The long format packet to return
434  *
435  */
436 static int dn_return_long(struct sk_buff *skb)
437 {
438         struct dn_skb_cb *cb;
439         unsigned char *ptr;
440         unsigned char *src_addr, *dst_addr;
441         unsigned char tmp[ETH_ALEN];
442
443         /* Add back all headers */
444         skb_push(skb, skb->data - skb_network_header(skb));
445
446         if ((skb = skb_unshare(skb, GFP_ATOMIC)) == NULL)
447                 return NET_RX_DROP;
448
449         cb = DN_SKB_CB(skb);
450         /* Ignore packet length and point to flags */
451         ptr = skb->data + 2;
452
453         /* Skip padding */
454         if (*ptr & DN_RT_F_PF) {
455                 char padlen = (*ptr & ~DN_RT_F_PF);
456                 ptr += padlen;
457         }
458
459         *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
460         ptr += 2;
461         dst_addr = ptr;
462         ptr += 8;
463         src_addr = ptr;
464         ptr += 6;
465         *ptr = 0; /* Zero hop count */
466
467         /* Swap source and destination */
468         memcpy(tmp, src_addr, ETH_ALEN);
469         memcpy(src_addr, dst_addr, ETH_ALEN);
470         memcpy(dst_addr, tmp, ETH_ALEN);
471
472         skb->pkt_type = PACKET_OUTGOING;
473         dn_rt_finish_output(skb, dst_addr, src_addr);
474         return NET_RX_SUCCESS;
475 }
476
477 /**
478  * dn_route_rx_packet - Try and find a route for an incoming packet
479  * @skb: The packet to find a route for
480  *
481  * Returns: result of input function if route is found, error code otherwise
482  */
483 static int dn_route_rx_packet(struct sk_buff *skb)
484 {
485         struct dn_skb_cb *cb;
486         int err;
487
488         if ((err = dn_route_input(skb)) == 0)
489                 return dst_input(skb);
490
491         cb = DN_SKB_CB(skb);
492         if (decnet_debug_level & 4) {
493                 char *devname = skb->dev ? skb->dev->name : "???";
494
495                 printk(KERN_DEBUG
496                         "DECnet: dn_route_rx_packet: rt_flags=0x%02x dev=%s len=%d src=0x%04hx dst=0x%04hx err=%d type=%d\n",
497                         (int)cb->rt_flags, devname, skb->len,
498                         le16_to_cpu(cb->src), le16_to_cpu(cb->dst),
499                         err, skb->pkt_type);
500         }
501
502         if ((skb->pkt_type == PACKET_HOST) && (cb->rt_flags & DN_RT_F_RQR)) {
503                 switch (cb->rt_flags & DN_RT_PKT_MSK) {
504                 case DN_RT_PKT_SHORT:
505                         return dn_return_short(skb);
506                 case DN_RT_PKT_LONG:
507                         return dn_return_long(skb);
508                 }
509         }
510
511         kfree_skb(skb);
512         return NET_RX_DROP;
513 }
514
515 static int dn_route_rx_long(struct sk_buff *skb)
516 {
517         struct dn_skb_cb *cb = DN_SKB_CB(skb);
518         unsigned char *ptr = skb->data;
519
520         if (!pskb_may_pull(skb, 21)) /* 20 for long header, 1 for shortest nsp */
521                 goto drop_it;
522
523         skb_pull(skb, 20);
524         skb_reset_transport_header(skb);
525
526         /* Destination info */
527         ptr += 2;
528         cb->dst = dn_eth2dn(ptr);
529         if (memcmp(ptr, dn_hiord_addr, 4) != 0)
530                 goto drop_it;
531         ptr += 6;
532
533
534         /* Source info */
535         ptr += 2;
536         cb->src = dn_eth2dn(ptr);
537         if (memcmp(ptr, dn_hiord_addr, 4) != 0)
538                 goto drop_it;
539         ptr += 6;
540         /* Other junk */
541         ptr++;
542         cb->hops = *ptr++; /* Visit Count */
543
544         return NF_HOOK(NFPROTO_DECNET, NF_DN_PRE_ROUTING, skb, skb->dev, NULL,
545                        dn_route_rx_packet);
546
547 drop_it:
548         kfree_skb(skb);
549         return NET_RX_DROP;
550 }
551
552
553
554 static int dn_route_rx_short(struct sk_buff *skb)
555 {
556         struct dn_skb_cb *cb = DN_SKB_CB(skb);
557         unsigned char *ptr = skb->data;
558
559         if (!pskb_may_pull(skb, 6)) /* 5 for short header + 1 for shortest nsp */
560                 goto drop_it;
561
562         skb_pull(skb, 5);
563         skb_reset_transport_header(skb);
564
565         cb->dst = *(__le16 *)ptr;
566         ptr += 2;
567         cb->src = *(__le16 *)ptr;
568         ptr += 2;
569         cb->hops = *ptr & 0x3f;
570
571         return NF_HOOK(NFPROTO_DECNET, NF_DN_PRE_ROUTING, skb, skb->dev, NULL,
572                        dn_route_rx_packet);
573
574 drop_it:
575         kfree_skb(skb);
576         return NET_RX_DROP;
577 }
578
579 static int dn_route_discard(struct sk_buff *skb)
580 {
581         /*
582          * I know we drop the packet here, but thats considered success in
583          * this case
584          */
585         kfree_skb(skb);
586         return NET_RX_SUCCESS;
587 }
588
589 static int dn_route_ptp_hello(struct sk_buff *skb)
590 {
591         dn_dev_hello(skb);
592         dn_neigh_pointopoint_hello(skb);
593         return NET_RX_SUCCESS;
594 }
595
596 int dn_route_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
597 {
598         struct dn_skb_cb *cb;
599         unsigned char flags = 0;
600         __u16 len = le16_to_cpu(*(__le16 *)skb->data);
601         struct dn_dev *dn = rcu_dereference(dev->dn_ptr);
602         unsigned char padlen = 0;
603
604         if (!net_eq(dev_net(dev), &init_net))
605                 goto dump_it;
606
607         if (dn == NULL)
608                 goto dump_it;
609
610         if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
611                 goto out;
612
613         if (!pskb_may_pull(skb, 3))
614                 goto dump_it;
615
616         skb_pull(skb, 2);
617
618         if (len > skb->len)
619                 goto dump_it;
620
621         skb_trim(skb, len);
622
623         flags = *skb->data;
624
625         cb = DN_SKB_CB(skb);
626         cb->stamp = jiffies;
627         cb->iif = dev->ifindex;
628
629         /*
630          * If we have padding, remove it.
631          */
632         if (flags & DN_RT_F_PF) {
633                 padlen = flags & ~DN_RT_F_PF;
634                 if (!pskb_may_pull(skb, padlen + 1))
635                         goto dump_it;
636                 skb_pull(skb, padlen);
637                 flags = *skb->data;
638         }
639
640         skb_reset_network_header(skb);
641
642         /*
643          * Weed out future version DECnet
644          */
645         if (flags & DN_RT_F_VER)
646                 goto dump_it;
647
648         cb->rt_flags = flags;
649
650         if (decnet_debug_level & 1)
651                 printk(KERN_DEBUG
652                         "dn_route_rcv: got 0x%02x from %s [%d %d %d]\n",
653                         (int)flags, (dev) ? dev->name : "???", len, skb->len,
654                         padlen);
655
656         if (flags & DN_RT_PKT_CNTL) {
657                 if (unlikely(skb_linearize(skb)))
658                         goto dump_it;
659
660                 switch (flags & DN_RT_CNTL_MSK) {
661                 case DN_RT_PKT_INIT:
662                         dn_dev_init_pkt(skb);
663                         break;
664                 case DN_RT_PKT_VERI:
665                         dn_dev_veri_pkt(skb);
666                         break;
667                 }
668
669                 if (dn->parms.state != DN_DEV_S_RU)
670                         goto dump_it;
671
672                 switch (flags & DN_RT_CNTL_MSK) {
673                 case DN_RT_PKT_HELO:
674                         return NF_HOOK(NFPROTO_DECNET, NF_DN_HELLO,
675                                        skb, skb->dev, NULL,
676                                        dn_route_ptp_hello);
677
678                 case DN_RT_PKT_L1RT:
679                 case DN_RT_PKT_L2RT:
680                         return NF_HOOK(NFPROTO_DECNET, NF_DN_ROUTE,
681                                        skb, skb->dev, NULL,
682                                        dn_route_discard);
683                 case DN_RT_PKT_ERTH:
684                         return NF_HOOK(NFPROTO_DECNET, NF_DN_HELLO,
685                                        skb, skb->dev, NULL,
686                                        dn_neigh_router_hello);
687
688                 case DN_RT_PKT_EEDH:
689                         return NF_HOOK(NFPROTO_DECNET, NF_DN_HELLO,
690                                        skb, skb->dev, NULL,
691                                        dn_neigh_endnode_hello);
692                 }
693         } else {
694                 if (dn->parms.state != DN_DEV_S_RU)
695                         goto dump_it;
696
697                 skb_pull(skb, 1); /* Pull flags */
698
699                 switch (flags & DN_RT_PKT_MSK) {
700                 case DN_RT_PKT_LONG:
701                         return dn_route_rx_long(skb);
702                 case DN_RT_PKT_SHORT:
703                         return dn_route_rx_short(skb);
704                 }
705         }
706
707 dump_it:
708         kfree_skb(skb);
709 out:
710         return NET_RX_DROP;
711 }
712
713 static int dn_to_neigh_output(struct sk_buff *skb)
714 {
715         struct dst_entry *dst = skb_dst(skb);
716         struct neighbour *n = dst_get_neighbour(dst);
717
718         return n->output(n, skb);
719 }
720
721 static int dn_output(struct sk_buff *skb)
722 {
723         struct dst_entry *dst = skb_dst(skb);
724         struct dn_route *rt = (struct dn_route *)dst;
725         struct net_device *dev = dst->dev;
726         struct dn_skb_cb *cb = DN_SKB_CB(skb);
727         struct neighbour *neigh;
728
729         int err = -EINVAL;
730
731         if ((neigh = dst_get_neighbour(dst)) == NULL)
732                 goto error;
733
734         skb->dev = dev;
735
736         cb->src = rt->rt_saddr;
737         cb->dst = rt->rt_daddr;
738
739         /*
740          * Always set the Intra-Ethernet bit on all outgoing packets
741          * originated on this node. Only valid flag from upper layers
742          * is return-to-sender-requested. Set hop count to 0 too.
743          */
744         cb->rt_flags &= ~DN_RT_F_RQR;
745         cb->rt_flags |= DN_RT_F_IE;
746         cb->hops = 0;
747
748         return NF_HOOK(NFPROTO_DECNET, NF_DN_LOCAL_OUT, skb, NULL, dev,
749                        dn_to_neigh_output);
750
751 error:
752         if (net_ratelimit())
753                 printk(KERN_DEBUG "dn_output: This should not happen\n");
754
755         kfree_skb(skb);
756
757         return err;
758 }
759
760 static int dn_forward(struct sk_buff *skb)
761 {
762         struct dn_skb_cb *cb = DN_SKB_CB(skb);
763         struct dst_entry *dst = skb_dst(skb);
764         struct dn_dev *dn_db = rcu_dereference(dst->dev->dn_ptr);
765         struct dn_route *rt;
766         int header_len;
767 #ifdef CONFIG_NETFILTER
768         struct net_device *dev = skb->dev;
769 #endif
770
771         if (skb->pkt_type != PACKET_HOST)
772                 goto drop;
773
774         /* Ensure that we have enough space for headers */
775         rt = (struct dn_route *)skb_dst(skb);
776         header_len = dn_db->use_long ? 21 : 6;
777         if (skb_cow(skb, LL_RESERVED_SPACE(rt->dst.dev)+header_len))
778                 goto drop;
779
780         /*
781          * Hop count exceeded.
782          */
783         if (++cb->hops > 30)
784                 goto drop;
785
786         skb->dev = rt->dst.dev;
787
788         /*
789          * If packet goes out same interface it came in on, then set
790          * the Intra-Ethernet bit. This has no effect for short
791          * packets, so we don't need to test for them here.
792          */
793         cb->rt_flags &= ~DN_RT_F_IE;
794         if (rt->rt_flags & RTCF_DOREDIRECT)
795                 cb->rt_flags |= DN_RT_F_IE;
796
797         return NF_HOOK(NFPROTO_DECNET, NF_DN_FORWARD, skb, dev, skb->dev,
798                        dn_to_neigh_output);
799
800 drop:
801         kfree_skb(skb);
802         return NET_RX_DROP;
803 }
804
805 /*
806  * Used to catch bugs. This should never normally get
807  * called.
808  */
809 static int dn_rt_bug(struct sk_buff *skb)
810 {
811         if (net_ratelimit()) {
812                 struct dn_skb_cb *cb = DN_SKB_CB(skb);
813
814                 printk(KERN_DEBUG "dn_rt_bug: skb from:%04x to:%04x\n",
815                                 le16_to_cpu(cb->src), le16_to_cpu(cb->dst));
816         }
817
818         kfree_skb(skb);
819
820         return NET_RX_DROP;
821 }
822
823 static unsigned int dn_dst_default_advmss(const struct dst_entry *dst)
824 {
825         return dn_mss_from_pmtu(dst->dev, dst_mtu(dst));
826 }
827
828 static unsigned int dn_dst_default_mtu(const struct dst_entry *dst)
829 {
830         return dst->dev->mtu;
831 }
832
833 static struct neighbour *dn_dst_neigh_lookup(const struct dst_entry *dst, const void *daddr)
834 {
835         return __neigh_lookup_errno(&dn_neigh_table, daddr, dst->dev);
836 }
837
838 static int dn_rt_set_next_hop(struct dn_route *rt, struct dn_fib_res *res)
839 {
840         struct dn_fib_info *fi = res->fi;
841         struct net_device *dev = rt->dst.dev;
842         unsigned int mss_metric;
843         struct neighbour *n;
844
845         if (fi) {
846                 if (DN_FIB_RES_GW(*res) &&
847                     DN_FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
848                         rt->rt_gateway = DN_FIB_RES_GW(*res);
849                 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
850         }
851         rt->rt_type = res->type;
852
853         if (dev != NULL && dst_get_neighbour(&rt->dst) == NULL) {
854                 n = __neigh_lookup_errno(&dn_neigh_table, &rt->rt_gateway, dev);
855                 if (IS_ERR(n))
856                         return PTR_ERR(n);
857                 dst_set_neighbour(&rt->dst, n);
858         }
859
860         if (dst_metric(&rt->dst, RTAX_MTU) > rt->dst.dev->mtu)
861                 dst_metric_set(&rt->dst, RTAX_MTU, rt->dst.dev->mtu);
862         mss_metric = dst_metric_raw(&rt->dst, RTAX_ADVMSS);
863         if (mss_metric) {
864                 unsigned int mss = dn_mss_from_pmtu(dev, dst_mtu(&rt->dst));
865                 if (mss_metric > mss)
866                         dst_metric_set(&rt->dst, RTAX_ADVMSS, mss);
867         }
868         return 0;
869 }
870
871 static inline int dn_match_addr(__le16 addr1, __le16 addr2)
872 {
873         __u16 tmp = le16_to_cpu(addr1) ^ le16_to_cpu(addr2);
874         int match = 16;
875         while(tmp) {
876                 tmp >>= 1;
877                 match--;
878         }
879         return match;
880 }
881
882 static __le16 dnet_select_source(const struct net_device *dev, __le16 daddr, int scope)
883 {
884         __le16 saddr = 0;
885         struct dn_dev *dn_db;
886         struct dn_ifaddr *ifa;
887         int best_match = 0;
888         int ret;
889
890         rcu_read_lock();
891         dn_db = rcu_dereference(dev->dn_ptr);
892         for (ifa = rcu_dereference(dn_db->ifa_list);
893              ifa != NULL;
894              ifa = rcu_dereference(ifa->ifa_next)) {
895                 if (ifa->ifa_scope > scope)
896                         continue;
897                 if (!daddr) {
898                         saddr = ifa->ifa_local;
899                         break;
900                 }
901                 ret = dn_match_addr(daddr, ifa->ifa_local);
902                 if (ret > best_match)
903                         saddr = ifa->ifa_local;
904                 if (best_match == 0)
905                         saddr = ifa->ifa_local;
906         }
907         rcu_read_unlock();
908
909         return saddr;
910 }
911
912 static inline __le16 __dn_fib_res_prefsrc(struct dn_fib_res *res)
913 {
914         return dnet_select_source(DN_FIB_RES_DEV(*res), DN_FIB_RES_GW(*res), res->scope);
915 }
916
917 static inline __le16 dn_fib_rules_map_destination(__le16 daddr, struct dn_fib_res *res)
918 {
919         __le16 mask = dnet_make_mask(res->prefixlen);
920         return (daddr&~mask)|res->fi->fib_nh->nh_gw;
921 }
922
923 static int dn_route_output_slow(struct dst_entry **pprt, const struct flowidn *oldflp, int try_hard)
924 {
925         struct flowidn fld = {
926                 .daddr = oldflp->daddr,
927                 .saddr = oldflp->saddr,
928                 .flowidn_scope = RT_SCOPE_UNIVERSE,
929                 .flowidn_mark = oldflp->flowidn_mark,
930                 .flowidn_iif = init_net.loopback_dev->ifindex,
931                 .flowidn_oif = oldflp->flowidn_oif,
932         };
933         struct dn_route *rt = NULL;
934         struct net_device *dev_out = NULL, *dev;
935         struct neighbour *neigh = NULL;
936         unsigned hash;
937         unsigned flags = 0;
938         struct dn_fib_res res = { .fi = NULL, .type = RTN_UNICAST };
939         int err;
940         int free_res = 0;
941         __le16 gateway = 0;
942
943         if (decnet_debug_level & 16)
944                 printk(KERN_DEBUG
945                        "dn_route_output_slow: dst=%04x src=%04x mark=%d"
946                        " iif=%d oif=%d\n", le16_to_cpu(oldflp->daddr),
947                        le16_to_cpu(oldflp->saddr),
948                        oldflp->flowidn_mark, init_net.loopback_dev->ifindex,
949                        oldflp->flowidn_oif);
950
951         /* If we have an output interface, verify its a DECnet device */
952         if (oldflp->flowidn_oif) {
953                 dev_out = dev_get_by_index(&init_net, oldflp->flowidn_oif);
954                 err = -ENODEV;
955                 if (dev_out && dev_out->dn_ptr == NULL) {
956                         dev_put(dev_out);
957                         dev_out = NULL;
958                 }
959                 if (dev_out == NULL)
960                         goto out;
961         }
962
963         /* If we have a source address, verify that its a local address */
964         if (oldflp->saddr) {
965                 err = -EADDRNOTAVAIL;
966
967                 if (dev_out) {
968                         if (dn_dev_islocal(dev_out, oldflp->saddr))
969                                 goto source_ok;
970                         dev_put(dev_out);
971                         goto out;
972                 }
973                 rcu_read_lock();
974                 for_each_netdev_rcu(&init_net, dev) {
975                         if (!dev->dn_ptr)
976                                 continue;
977                         if (!dn_dev_islocal(dev, oldflp->saddr))
978                                 continue;
979                         if ((dev->flags & IFF_LOOPBACK) &&
980                             oldflp->daddr &&
981                             !dn_dev_islocal(dev, oldflp->daddr))
982                                 continue;
983
984                         dev_out = dev;
985                         break;
986                 }
987                 rcu_read_unlock();
988                 if (dev_out == NULL)
989                         goto out;
990                 dev_hold(dev_out);
991 source_ok:
992                 ;
993         }
994
995         /* No destination? Assume its local */
996         if (!fld.daddr) {
997                 fld.daddr = fld.saddr;
998
999                 err = -EADDRNOTAVAIL;
1000                 if (dev_out)
1001                         dev_put(dev_out);
1002                 dev_out = init_net.loopback_dev;
1003                 dev_hold(dev_out);
1004                 if (!fld.daddr) {
1005                         fld.daddr =
1006                         fld.saddr = dnet_select_source(dev_out, 0,
1007                                                        RT_SCOPE_HOST);
1008                         if (!fld.daddr)
1009                                 goto out;
1010                 }
1011                 fld.flowidn_oif = init_net.loopback_dev->ifindex;
1012                 res.type = RTN_LOCAL;
1013                 goto make_route;
1014         }
1015
1016         if (decnet_debug_level & 16)
1017                 printk(KERN_DEBUG
1018                        "dn_route_output_slow: initial checks complete."
1019                        " dst=%o4x src=%04x oif=%d try_hard=%d\n",
1020                        le16_to_cpu(fld.daddr), le16_to_cpu(fld.saddr),
1021                        fld.flowidn_oif, try_hard);
1022
1023         /*
1024          * N.B. If the kernel is compiled without router support then
1025          * dn_fib_lookup() will evaluate to non-zero so this if () block
1026          * will always be executed.
1027          */
1028         err = -ESRCH;
1029         if (try_hard || (err = dn_fib_lookup(&fld, &res)) != 0) {
1030                 struct dn_dev *dn_db;
1031                 if (err != -ESRCH)
1032                         goto out;
1033                 /*
1034                  * Here the fallback is basically the standard algorithm for
1035                  * routing in endnodes which is described in the DECnet routing
1036                  * docs
1037                  *
1038                  * If we are not trying hard, look in neighbour cache.
1039                  * The result is tested to ensure that if a specific output
1040                  * device/source address was requested, then we honour that
1041                  * here
1042                  */
1043                 if (!try_hard) {
1044                         neigh = neigh_lookup_nodev(&dn_neigh_table, &init_net, &fld.daddr);
1045                         if (neigh) {
1046                                 if ((oldflp->flowidn_oif &&
1047                                     (neigh->dev->ifindex != oldflp->flowidn_oif)) ||
1048                                     (oldflp->saddr &&
1049                                     (!dn_dev_islocal(neigh->dev,
1050                                                      oldflp->saddr)))) {
1051                                         neigh_release(neigh);
1052                                         neigh = NULL;
1053                                 } else {
1054                                         if (dev_out)
1055                                                 dev_put(dev_out);
1056                                         if (dn_dev_islocal(neigh->dev, fld.daddr)) {
1057                                                 dev_out = init_net.loopback_dev;
1058                                                 res.type = RTN_LOCAL;
1059                                         } else {
1060                                                 dev_out = neigh->dev;
1061                                         }
1062                                         dev_hold(dev_out);
1063                                         goto select_source;
1064                                 }
1065                         }
1066                 }
1067
1068                 /* Not there? Perhaps its a local address */
1069                 if (dev_out == NULL)
1070                         dev_out = dn_dev_get_default();
1071                 err = -ENODEV;
1072                 if (dev_out == NULL)
1073                         goto out;
1074                 dn_db = rcu_dereference_raw(dev_out->dn_ptr);
1075                 /* Possible improvement - check all devices for local addr */
1076                 if (dn_dev_islocal(dev_out, fld.daddr)) {
1077                         dev_put(dev_out);
1078                         dev_out = init_net.loopback_dev;
1079                         dev_hold(dev_out);
1080                         res.type = RTN_LOCAL;
1081                         goto select_source;
1082                 }
1083                 /* Not local either.... try sending it to the default router */
1084                 neigh = neigh_clone(dn_db->router);
1085                 BUG_ON(neigh && neigh->dev != dev_out);
1086
1087                 /* Ok then, we assume its directly connected and move on */
1088 select_source:
1089                 if (neigh)
1090                         gateway = ((struct dn_neigh *)neigh)->addr;
1091                 if (gateway == 0)
1092                         gateway = fld.daddr;
1093                 if (fld.saddr == 0) {
1094                         fld.saddr = dnet_select_source(dev_out, gateway,
1095                                                        res.type == RTN_LOCAL ?
1096                                                        RT_SCOPE_HOST :
1097                                                        RT_SCOPE_LINK);
1098                         if (fld.saddr == 0 && res.type != RTN_LOCAL)
1099                                 goto e_addr;
1100                 }
1101                 fld.flowidn_oif = dev_out->ifindex;
1102                 goto make_route;
1103         }
1104         free_res = 1;
1105
1106         if (res.type == RTN_NAT)
1107                 goto e_inval;
1108
1109         if (res.type == RTN_LOCAL) {
1110                 if (!fld.saddr)
1111                         fld.saddr = fld.daddr;
1112                 if (dev_out)
1113                         dev_put(dev_out);
1114                 dev_out = init_net.loopback_dev;
1115                 dev_hold(dev_out);
1116                 fld.flowidn_oif = dev_out->ifindex;
1117                 if (res.fi)
1118                         dn_fib_info_put(res.fi);
1119                 res.fi = NULL;
1120                 goto make_route;
1121         }
1122
1123         if (res.fi->fib_nhs > 1 && fld.flowidn_oif == 0)
1124                 dn_fib_select_multipath(&fld, &res);
1125
1126         /*
1127          * We could add some logic to deal with default routes here and
1128          * get rid of some of the special casing above.
1129          */
1130
1131         if (!fld.saddr)
1132                 fld.saddr = DN_FIB_RES_PREFSRC(res);
1133
1134         if (dev_out)
1135                 dev_put(dev_out);
1136         dev_out = DN_FIB_RES_DEV(res);
1137         dev_hold(dev_out);
1138         fld.flowidn_oif = dev_out->ifindex;
1139         gateway = DN_FIB_RES_GW(res);
1140
1141 make_route:
1142         if (dev_out->flags & IFF_LOOPBACK)
1143                 flags |= RTCF_LOCAL;
1144
1145         rt = dst_alloc(&dn_dst_ops, dev_out, 1, 0, DST_HOST);
1146         if (rt == NULL)
1147                 goto e_nobufs;
1148
1149         memset(&rt->fld, 0, sizeof(rt->fld));
1150         rt->fld.saddr        = oldflp->saddr;
1151         rt->fld.daddr        = oldflp->daddr;
1152         rt->fld.flowidn_oif  = oldflp->flowidn_oif;
1153         rt->fld.flowidn_iif  = 0;
1154         rt->fld.flowidn_mark = oldflp->flowidn_mark;
1155
1156         rt->rt_saddr      = fld.saddr;
1157         rt->rt_daddr      = fld.daddr;
1158         rt->rt_gateway    = gateway ? gateway : fld.daddr;
1159         rt->rt_local_src  = fld.saddr;
1160
1161         rt->rt_dst_map    = fld.daddr;
1162         rt->rt_src_map    = fld.saddr;
1163
1164         dst_set_neighbour(&rt->dst, neigh);
1165         neigh = NULL;
1166
1167         rt->dst.lastuse = jiffies;
1168         rt->dst.output  = dn_output;
1169         rt->dst.input   = dn_rt_bug;
1170         rt->rt_flags      = flags;
1171         if (flags & RTCF_LOCAL)
1172                 rt->dst.input = dn_nsp_rx;
1173
1174         err = dn_rt_set_next_hop(rt, &res);
1175         if (err)
1176                 goto e_neighbour;
1177
1178         hash = dn_hash(rt->fld.saddr, rt->fld.daddr);
1179         dn_insert_route(rt, hash, (struct dn_route **)pprt);
1180
1181 done:
1182         if (neigh)
1183                 neigh_release(neigh);
1184         if (free_res)
1185                 dn_fib_res_put(&res);
1186         if (dev_out)
1187                 dev_put(dev_out);
1188 out:
1189         return err;
1190
1191 e_addr:
1192         err = -EADDRNOTAVAIL;
1193         goto done;
1194 e_inval:
1195         err = -EINVAL;
1196         goto done;
1197 e_nobufs:
1198         err = -ENOBUFS;
1199         goto done;
1200 e_neighbour:
1201         dst_free(&rt->dst);
1202         goto e_nobufs;
1203 }
1204
1205
1206 /*
1207  * N.B. The flags may be moved into the flowi at some future stage.
1208  */
1209 static int __dn_route_output_key(struct dst_entry **pprt, const struct flowidn *flp, int flags)
1210 {
1211         unsigned hash = dn_hash(flp->saddr, flp->daddr);
1212         struct dn_route *rt = NULL;
1213
1214         if (!(flags & MSG_TRYHARD)) {
1215                 rcu_read_lock_bh();
1216                 for (rt = rcu_dereference_bh(dn_rt_hash_table[hash].chain); rt;
1217                         rt = rcu_dereference_bh(rt->dst.dn_next)) {
1218                         if ((flp->daddr == rt->fld.daddr) &&
1219                             (flp->saddr == rt->fld.saddr) &&
1220                             (flp->flowidn_mark == rt->fld.flowidn_mark) &&
1221                             dn_is_output_route(rt) &&
1222                             (rt->fld.flowidn_oif == flp->flowidn_oif)) {
1223                                 dst_use(&rt->dst, jiffies);
1224                                 rcu_read_unlock_bh();
1225                                 *pprt = &rt->dst;
1226                                 return 0;
1227                         }
1228                 }
1229                 rcu_read_unlock_bh();
1230         }
1231
1232         return dn_route_output_slow(pprt, flp, flags);
1233 }
1234
1235 static int dn_route_output_key(struct dst_entry **pprt, struct flowidn *flp, int flags)
1236 {
1237         int err;
1238
1239         err = __dn_route_output_key(pprt, flp, flags);
1240         if (err == 0 && flp->flowidn_proto) {
1241                 *pprt = xfrm_lookup(&init_net, *pprt,
1242                                     flowidn_to_flowi(flp), NULL, 0);
1243                 if (IS_ERR(*pprt)) {
1244                         err = PTR_ERR(*pprt);
1245                         *pprt = NULL;
1246                 }
1247         }
1248         return err;
1249 }
1250
1251 int dn_route_output_sock(struct dst_entry **pprt, struct flowidn *fl, struct sock *sk, int flags)
1252 {
1253         int err;
1254
1255         err = __dn_route_output_key(pprt, fl, flags & MSG_TRYHARD);
1256         if (err == 0 && fl->flowidn_proto) {
1257                 if (!(flags & MSG_DONTWAIT))
1258                         fl->flowidn_flags |= FLOWI_FLAG_CAN_SLEEP;
1259                 *pprt = xfrm_lookup(&init_net, *pprt,
1260                                     flowidn_to_flowi(fl), sk, 0);
1261                 if (IS_ERR(*pprt)) {
1262                         err = PTR_ERR(*pprt);
1263                         *pprt = NULL;
1264                 }
1265         }
1266         return err;
1267 }
1268
1269 static int dn_route_input_slow(struct sk_buff *skb)
1270 {
1271         struct dn_route *rt = NULL;
1272         struct dn_skb_cb *cb = DN_SKB_CB(skb);
1273         struct net_device *in_dev = skb->dev;
1274         struct net_device *out_dev = NULL;
1275         struct dn_dev *dn_db;
1276         struct neighbour *neigh = NULL;
1277         unsigned hash;
1278         int flags = 0;
1279         __le16 gateway = 0;
1280         __le16 local_src = 0;
1281         struct flowidn fld = {
1282                 .daddr = cb->dst,
1283                 .saddr = cb->src,
1284                 .flowidn_scope = RT_SCOPE_UNIVERSE,
1285                 .flowidn_mark = skb->mark,
1286                 .flowidn_iif = skb->dev->ifindex,
1287         };
1288         struct dn_fib_res res = { .fi = NULL, .type = RTN_UNREACHABLE };
1289         int err = -EINVAL;
1290         int free_res = 0;
1291
1292         dev_hold(in_dev);
1293
1294         if ((dn_db = rcu_dereference(in_dev->dn_ptr)) == NULL)
1295                 goto out;
1296
1297         /* Zero source addresses are not allowed */
1298         if (fld.saddr == 0)
1299                 goto out;
1300
1301         /*
1302          * In this case we've just received a packet from a source
1303          * outside ourselves pretending to come from us. We don't
1304          * allow it any further to prevent routing loops, spoofing and
1305          * other nasties. Loopback packets already have the dst attached
1306          * so this only affects packets which have originated elsewhere.
1307          */
1308         err  = -ENOTUNIQ;
1309         if (dn_dev_islocal(in_dev, cb->src))
1310                 goto out;
1311
1312         err = dn_fib_lookup(&fld, &res);
1313         if (err) {
1314                 if (err != -ESRCH)
1315                         goto out;
1316                 /*
1317                  * Is the destination us ?
1318                  */
1319                 if (!dn_dev_islocal(in_dev, cb->dst))
1320                         goto e_inval;
1321
1322                 res.type = RTN_LOCAL;
1323         } else {
1324                 __le16 src_map = fld.saddr;
1325                 free_res = 1;
1326
1327                 out_dev = DN_FIB_RES_DEV(res);
1328                 if (out_dev == NULL) {
1329                         if (net_ratelimit())
1330                                 printk(KERN_CRIT "Bug in dn_route_input_slow() "
1331                                                  "No output device\n");
1332                         goto e_inval;
1333                 }
1334                 dev_hold(out_dev);
1335
1336                 if (res.r)
1337                         src_map = fld.saddr; /* no NAT support for now */
1338
1339                 gateway = DN_FIB_RES_GW(res);
1340                 if (res.type == RTN_NAT) {
1341                         fld.daddr = dn_fib_rules_map_destination(fld.daddr, &res);
1342                         dn_fib_res_put(&res);
1343                         free_res = 0;
1344                         if (dn_fib_lookup(&fld, &res))
1345                                 goto e_inval;
1346                         free_res = 1;
1347                         if (res.type != RTN_UNICAST)
1348                                 goto e_inval;
1349                         flags |= RTCF_DNAT;
1350                         gateway = fld.daddr;
1351                 }
1352                 fld.saddr = src_map;
1353         }
1354
1355         switch(res.type) {
1356         case RTN_UNICAST:
1357                 /*
1358                  * Forwarding check here, we only check for forwarding
1359                  * being turned off, if you want to only forward intra
1360                  * area, its up to you to set the routing tables up
1361                  * correctly.
1362                  */
1363                 if (dn_db->parms.forwarding == 0)
1364                         goto e_inval;
1365
1366                 if (res.fi->fib_nhs > 1 && fld.flowidn_oif == 0)
1367                         dn_fib_select_multipath(&fld, &res);
1368
1369                 /*
1370                  * Check for out_dev == in_dev. We use the RTCF_DOREDIRECT
1371                  * flag as a hint to set the intra-ethernet bit when
1372                  * forwarding. If we've got NAT in operation, we don't do
1373                  * this optimisation.
1374                  */
1375                 if (out_dev == in_dev && !(flags & RTCF_NAT))
1376                         flags |= RTCF_DOREDIRECT;
1377
1378                 local_src = DN_FIB_RES_PREFSRC(res);
1379
1380         case RTN_BLACKHOLE:
1381         case RTN_UNREACHABLE:
1382                 break;
1383         case RTN_LOCAL:
1384                 flags |= RTCF_LOCAL;
1385                 fld.saddr = cb->dst;
1386                 fld.daddr = cb->src;
1387
1388                 /* Routing tables gave us a gateway */
1389                 if (gateway)
1390                         goto make_route;
1391
1392                 /* Packet was intra-ethernet, so we know its on-link */
1393                 if (cb->rt_flags & DN_RT_F_IE) {
1394                         gateway = cb->src;
1395                         flags |= RTCF_DIRECTSRC;
1396                         goto make_route;
1397                 }
1398
1399                 /* Use the default router if there is one */
1400                 neigh = neigh_clone(dn_db->router);
1401                 if (neigh) {
1402                         gateway = ((struct dn_neigh *)neigh)->addr;
1403                         goto make_route;
1404                 }
1405
1406                 /* Close eyes and pray */
1407                 gateway = cb->src;
1408                 flags |= RTCF_DIRECTSRC;
1409                 goto make_route;
1410         default:
1411                 goto e_inval;
1412         }
1413
1414 make_route:
1415         rt = dst_alloc(&dn_dst_ops, out_dev, 0, 0, DST_HOST);
1416         if (rt == NULL)
1417                 goto e_nobufs;
1418
1419         memset(&rt->fld, 0, sizeof(rt->fld));
1420         rt->rt_saddr      = fld.saddr;
1421         rt->rt_daddr      = fld.daddr;
1422         rt->rt_gateway    = fld.daddr;
1423         if (gateway)
1424                 rt->rt_gateway = gateway;
1425         rt->rt_local_src  = local_src ? local_src : rt->rt_saddr;
1426
1427         rt->rt_dst_map    = fld.daddr;
1428         rt->rt_src_map    = fld.saddr;
1429
1430         rt->fld.saddr        = cb->src;
1431         rt->fld.daddr        = cb->dst;
1432         rt->fld.flowidn_oif  = 0;
1433         rt->fld.flowidn_iif  = in_dev->ifindex;
1434         rt->fld.flowidn_mark = fld.flowidn_mark;
1435
1436         dst_set_neighbour(&rt->dst, neigh);
1437         rt->dst.lastuse = jiffies;
1438         rt->dst.output = dn_rt_bug;
1439         switch (res.type) {
1440         case RTN_UNICAST:
1441                 rt->dst.input = dn_forward;
1442                 break;
1443         case RTN_LOCAL:
1444                 rt->dst.output = dn_output;
1445                 rt->dst.input = dn_nsp_rx;
1446                 rt->dst.dev = in_dev;
1447                 flags |= RTCF_LOCAL;
1448                 break;
1449         default:
1450         case RTN_UNREACHABLE:
1451         case RTN_BLACKHOLE:
1452                 rt->dst.input = dst_discard;
1453         }
1454         rt->rt_flags = flags;
1455
1456         err = dn_rt_set_next_hop(rt, &res);
1457         if (err)
1458                 goto e_neighbour;
1459
1460         hash = dn_hash(rt->fld.saddr, rt->fld.daddr);
1461         dn_insert_route(rt, hash, &rt);
1462         skb_dst_set(skb, &rt->dst);
1463
1464 done:
1465         if (neigh)
1466                 neigh_release(neigh);
1467         if (free_res)
1468                 dn_fib_res_put(&res);
1469         dev_put(in_dev);
1470         if (out_dev)
1471                 dev_put(out_dev);
1472 out:
1473         return err;
1474
1475 e_inval:
1476         err = -EINVAL;
1477         goto done;
1478
1479 e_nobufs:
1480         err = -ENOBUFS;
1481         goto done;
1482
1483 e_neighbour:
1484         dst_free(&rt->dst);
1485         goto done;
1486 }
1487
1488 static int dn_route_input(struct sk_buff *skb)
1489 {
1490         struct dn_route *rt;
1491         struct dn_skb_cb *cb = DN_SKB_CB(skb);
1492         unsigned hash = dn_hash(cb->src, cb->dst);
1493
1494         if (skb_dst(skb))
1495                 return 0;
1496
1497         rcu_read_lock();
1498         for(rt = rcu_dereference(dn_rt_hash_table[hash].chain); rt != NULL;
1499             rt = rcu_dereference(rt->dst.dn_next)) {
1500                 if ((rt->fld.saddr == cb->src) &&
1501                     (rt->fld.daddr == cb->dst) &&
1502                     (rt->fld.flowidn_oif == 0) &&
1503                     (rt->fld.flowidn_mark == skb->mark) &&
1504                     (rt->fld.flowidn_iif == cb->iif)) {
1505                         dst_use(&rt->dst, jiffies);
1506                         rcu_read_unlock();
1507                         skb_dst_set(skb, (struct dst_entry *)rt);
1508                         return 0;
1509                 }
1510         }
1511         rcu_read_unlock();
1512
1513         return dn_route_input_slow(skb);
1514 }
1515
1516 static int dn_rt_fill_info(struct sk_buff *skb, u32 pid, u32 seq,
1517                            int event, int nowait, unsigned int flags)
1518 {
1519         struct dn_route *rt = (struct dn_route *)skb_dst(skb);
1520         struct rtmsg *r;
1521         struct nlmsghdr *nlh;
1522         unsigned char *b = skb_tail_pointer(skb);
1523         long expires;
1524
1525         nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*r), flags);
1526         r = NLMSG_DATA(nlh);
1527         r->rtm_family = AF_DECnet;
1528         r->rtm_dst_len = 16;
1529         r->rtm_src_len = 0;
1530         r->rtm_tos = 0;
1531         r->rtm_table = RT_TABLE_MAIN;
1532         RTA_PUT_U32(skb, RTA_TABLE, RT_TABLE_MAIN);
1533         r->rtm_type = rt->rt_type;
1534         r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
1535         r->rtm_scope = RT_SCOPE_UNIVERSE;
1536         r->rtm_protocol = RTPROT_UNSPEC;
1537         if (rt->rt_flags & RTCF_NOTIFY)
1538                 r->rtm_flags |= RTM_F_NOTIFY;
1539         RTA_PUT(skb, RTA_DST, 2, &rt->rt_daddr);
1540         if (rt->fld.saddr) {
1541                 r->rtm_src_len = 16;
1542                 RTA_PUT(skb, RTA_SRC, 2, &rt->fld.saddr);
1543         }
1544         if (rt->dst.dev)
1545                 RTA_PUT(skb, RTA_OIF, sizeof(int), &rt->dst.dev->ifindex);
1546         /*
1547          * Note to self - change this if input routes reverse direction when
1548          * they deal only with inputs and not with replies like they do
1549          * currently.
1550          */
1551         RTA_PUT(skb, RTA_PREFSRC, 2, &rt->rt_local_src);
1552         if (rt->rt_daddr != rt->rt_gateway)
1553                 RTA_PUT(skb, RTA_GATEWAY, 2, &rt->rt_gateway);
1554         if (rtnetlink_put_metrics(skb, dst_metrics_ptr(&rt->dst)) < 0)
1555                 goto rtattr_failure;
1556         expires = rt->dst.expires ? rt->dst.expires - jiffies : 0;
1557         if (rtnl_put_cacheinfo(skb, &rt->dst, 0, 0, 0, expires,
1558                                rt->dst.error) < 0)
1559                 goto rtattr_failure;
1560         if (dn_is_input_route(rt))
1561                 RTA_PUT(skb, RTA_IIF, sizeof(int), &rt->fld.flowidn_iif);
1562
1563         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1564         return skb->len;
1565
1566 nlmsg_failure:
1567 rtattr_failure:
1568         nlmsg_trim(skb, b);
1569         return -1;
1570 }
1571
1572 /*
1573  * This is called by both endnodes and routers now.
1574  */
1575 static int dn_cache_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh, void *arg)
1576 {
1577         struct net *net = sock_net(in_skb->sk);
1578         struct rtattr **rta = arg;
1579         struct rtmsg *rtm = NLMSG_DATA(nlh);
1580         struct dn_route *rt = NULL;
1581         struct dn_skb_cb *cb;
1582         int err;
1583         struct sk_buff *skb;
1584         struct flowidn fld;
1585
1586         if (!net_eq(net, &init_net))
1587                 return -EINVAL;
1588
1589         memset(&fld, 0, sizeof(fld));
1590         fld.flowidn_proto = DNPROTO_NSP;
1591
1592         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1593         if (skb == NULL)
1594                 return -ENOBUFS;
1595         skb_reset_mac_header(skb);
1596         cb = DN_SKB_CB(skb);
1597
1598         if (rta[RTA_SRC-1])
1599                 memcpy(&fld.saddr, RTA_DATA(rta[RTA_SRC-1]), 2);
1600         if (rta[RTA_DST-1])
1601                 memcpy(&fld.daddr, RTA_DATA(rta[RTA_DST-1]), 2);
1602         if (rta[RTA_IIF-1])
1603                 memcpy(&fld.flowidn_iif, RTA_DATA(rta[RTA_IIF-1]), sizeof(int));
1604
1605         if (fld.flowidn_iif) {
1606                 struct net_device *dev;
1607                 if ((dev = dev_get_by_index(&init_net, fld.flowidn_iif)) == NULL) {
1608                         kfree_skb(skb);
1609                         return -ENODEV;
1610                 }
1611                 if (!dev->dn_ptr) {
1612                         dev_put(dev);
1613                         kfree_skb(skb);
1614                         return -ENODEV;
1615                 }
1616                 skb->protocol = htons(ETH_P_DNA_RT);
1617                 skb->dev = dev;
1618                 cb->src = fld.saddr;
1619                 cb->dst = fld.daddr;
1620                 local_bh_disable();
1621                 err = dn_route_input(skb);
1622                 local_bh_enable();
1623                 memset(cb, 0, sizeof(struct dn_skb_cb));
1624                 rt = (struct dn_route *)skb_dst(skb);
1625                 if (!err && -rt->dst.error)
1626                         err = rt->dst.error;
1627         } else {
1628                 int oif = 0;
1629                 if (rta[RTA_OIF - 1])
1630                         memcpy(&oif, RTA_DATA(rta[RTA_OIF - 1]), sizeof(int));
1631                 fld.flowidn_oif = oif;
1632                 err = dn_route_output_key((struct dst_entry **)&rt, &fld, 0);
1633         }
1634
1635         if (skb->dev)
1636                 dev_put(skb->dev);
1637         skb->dev = NULL;
1638         if (err)
1639                 goto out_free;
1640         skb_dst_set(skb, &rt->dst);
1641         if (rtm->rtm_flags & RTM_F_NOTIFY)
1642                 rt->rt_flags |= RTCF_NOTIFY;
1643
1644         err = dn_rt_fill_info(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq, RTM_NEWROUTE, 0, 0);
1645
1646         if (err == 0)
1647                 goto out_free;
1648         if (err < 0) {
1649                 err = -EMSGSIZE;
1650                 goto out_free;
1651         }
1652
1653         return rtnl_unicast(skb, &init_net, NETLINK_CB(in_skb).pid);
1654
1655 out_free:
1656         kfree_skb(skb);
1657         return err;
1658 }
1659
1660 /*
1661  * For routers, this is called from dn_fib_dump, but for endnodes its
1662  * called directly from the rtnetlink dispatch table.
1663  */
1664 int dn_cache_dump(struct sk_buff *skb, struct netlink_callback *cb)
1665 {
1666         struct net *net = sock_net(skb->sk);
1667         struct dn_route *rt;
1668         int h, s_h;
1669         int idx, s_idx;
1670
1671         if (!net_eq(net, &init_net))
1672                 return 0;
1673
1674         if (NLMSG_PAYLOAD(cb->nlh, 0) < sizeof(struct rtmsg))
1675                 return -EINVAL;
1676         if (!(((struct rtmsg *)NLMSG_DATA(cb->nlh))->rtm_flags&RTM_F_CLONED))
1677                 return 0;
1678
1679         s_h = cb->args[0];
1680         s_idx = idx = cb->args[1];
1681         for(h = 0; h <= dn_rt_hash_mask; h++) {
1682                 if (h < s_h)
1683                         continue;
1684                 if (h > s_h)
1685                         s_idx = 0;
1686                 rcu_read_lock_bh();
1687                 for(rt = rcu_dereference_bh(dn_rt_hash_table[h].chain), idx = 0;
1688                         rt;
1689                         rt = rcu_dereference_bh(rt->dst.dn_next), idx++) {
1690                         if (idx < s_idx)
1691                                 continue;
1692                         skb_dst_set(skb, dst_clone(&rt->dst));
1693                         if (dn_rt_fill_info(skb, NETLINK_CB(cb->skb).pid,
1694                                         cb->nlh->nlmsg_seq, RTM_NEWROUTE,
1695                                         1, NLM_F_MULTI) <= 0) {
1696                                 skb_dst_drop(skb);
1697                                 rcu_read_unlock_bh();
1698                                 goto done;
1699                         }
1700                         skb_dst_drop(skb);
1701                 }
1702                 rcu_read_unlock_bh();
1703         }
1704
1705 done:
1706         cb->args[0] = h;
1707         cb->args[1] = idx;
1708         return skb->len;
1709 }
1710
1711 #ifdef CONFIG_PROC_FS
1712 struct dn_rt_cache_iter_state {
1713         int bucket;
1714 };
1715
1716 static struct dn_route *dn_rt_cache_get_first(struct seq_file *seq)
1717 {
1718         struct dn_route *rt = NULL;
1719         struct dn_rt_cache_iter_state *s = seq->private;
1720
1721         for(s->bucket = dn_rt_hash_mask; s->bucket >= 0; --s->bucket) {
1722                 rcu_read_lock_bh();
1723                 rt = rcu_dereference_bh(dn_rt_hash_table[s->bucket].chain);
1724                 if (rt)
1725                         break;
1726                 rcu_read_unlock_bh();
1727         }
1728         return rt;
1729 }
1730
1731 static struct dn_route *dn_rt_cache_get_next(struct seq_file *seq, struct dn_route *rt)
1732 {
1733         struct dn_rt_cache_iter_state *s = seq->private;
1734
1735         rt = rcu_dereference_bh(rt->dst.dn_next);
1736         while (!rt) {
1737                 rcu_read_unlock_bh();
1738                 if (--s->bucket < 0)
1739                         break;
1740                 rcu_read_lock_bh();
1741                 rt = rcu_dereference_bh(dn_rt_hash_table[s->bucket].chain);
1742         }
1743         return rt;
1744 }
1745
1746 static void *dn_rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
1747 {
1748         struct dn_route *rt = dn_rt_cache_get_first(seq);
1749
1750         if (rt) {
1751                 while(*pos && (rt = dn_rt_cache_get_next(seq, rt)))
1752                         --*pos;
1753         }
1754         return *pos ? NULL : rt;
1755 }
1756
1757 static void *dn_rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1758 {
1759         struct dn_route *rt = dn_rt_cache_get_next(seq, v);
1760         ++*pos;
1761         return rt;
1762 }
1763
1764 static void dn_rt_cache_seq_stop(struct seq_file *seq, void *v)
1765 {
1766         if (v)
1767                 rcu_read_unlock_bh();
1768 }
1769
1770 static int dn_rt_cache_seq_show(struct seq_file *seq, void *v)
1771 {
1772         struct dn_route *rt = v;
1773         char buf1[DN_ASCBUF_LEN], buf2[DN_ASCBUF_LEN];
1774
1775         seq_printf(seq, "%-8s %-7s %-7s %04d %04d %04d\n",
1776                         rt->dst.dev ? rt->dst.dev->name : "*",
1777                         dn_addr2asc(le16_to_cpu(rt->rt_daddr), buf1),
1778                         dn_addr2asc(le16_to_cpu(rt->rt_saddr), buf2),
1779                         atomic_read(&rt->dst.__refcnt),
1780                         rt->dst.__use,
1781                         (int) dst_metric(&rt->dst, RTAX_RTT));
1782         return 0;
1783 }
1784
1785 static const struct seq_operations dn_rt_cache_seq_ops = {
1786         .start  = dn_rt_cache_seq_start,
1787         .next   = dn_rt_cache_seq_next,
1788         .stop   = dn_rt_cache_seq_stop,
1789         .show   = dn_rt_cache_seq_show,
1790 };
1791
1792 static int dn_rt_cache_seq_open(struct inode *inode, struct file *file)
1793 {
1794         return seq_open_private(file, &dn_rt_cache_seq_ops,
1795                         sizeof(struct dn_rt_cache_iter_state));
1796 }
1797
1798 static const struct file_operations dn_rt_cache_seq_fops = {
1799         .owner   = THIS_MODULE,
1800         .open    = dn_rt_cache_seq_open,
1801         .read    = seq_read,
1802         .llseek  = seq_lseek,
1803         .release = seq_release_private,
1804 };
1805
1806 #endif /* CONFIG_PROC_FS */
1807
1808 void __init dn_route_init(void)
1809 {
1810         int i, goal, order;
1811
1812         dn_dst_ops.kmem_cachep =
1813                 kmem_cache_create("dn_dst_cache", sizeof(struct dn_route), 0,
1814                                   SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1815         dst_entries_init(&dn_dst_ops);
1816         setup_timer(&dn_route_timer, dn_dst_check_expire, 0);
1817         dn_route_timer.expires = jiffies + decnet_dst_gc_interval * HZ;
1818         add_timer(&dn_route_timer);
1819
1820         goal = totalram_pages >> (26 - PAGE_SHIFT);
1821
1822         for(order = 0; (1UL << order) < goal; order++)
1823                 /* NOTHING */;
1824
1825         /*
1826          * Only want 1024 entries max, since the table is very, very unlikely
1827          * to be larger than that.
1828          */
1829         while(order && ((((1UL << order) * PAGE_SIZE) /
1830                                 sizeof(struct dn_rt_hash_bucket)) >= 2048))
1831                 order--;
1832
1833         do {
1834                 dn_rt_hash_mask = (1UL << order) * PAGE_SIZE /
1835                         sizeof(struct dn_rt_hash_bucket);
1836                 while(dn_rt_hash_mask & (dn_rt_hash_mask - 1))
1837                         dn_rt_hash_mask--;
1838                 dn_rt_hash_table = (struct dn_rt_hash_bucket *)
1839                         __get_free_pages(GFP_ATOMIC, order);
1840         } while (dn_rt_hash_table == NULL && --order > 0);
1841
1842         if (!dn_rt_hash_table)
1843                 panic("Failed to allocate DECnet route cache hash table\n");
1844
1845         printk(KERN_INFO
1846                 "DECnet: Routing cache hash table of %u buckets, %ldKbytes\n",
1847                 dn_rt_hash_mask,
1848                 (long)(dn_rt_hash_mask*sizeof(struct dn_rt_hash_bucket))/1024);
1849
1850         dn_rt_hash_mask--;
1851         for(i = 0; i <= dn_rt_hash_mask; i++) {
1852                 spin_lock_init(&dn_rt_hash_table[i].lock);
1853                 dn_rt_hash_table[i].chain = NULL;
1854         }
1855
1856         dn_dst_ops.gc_thresh = (dn_rt_hash_mask + 1);
1857
1858         proc_net_fops_create(&init_net, "decnet_cache", S_IRUGO, &dn_rt_cache_seq_fops);
1859
1860 #ifdef CONFIG_DECNET_ROUTER
1861         rtnl_register(PF_DECnet, RTM_GETROUTE, dn_cache_getroute,
1862                       dn_fib_dump, NULL);
1863 #else
1864         rtnl_register(PF_DECnet, RTM_GETROUTE, dn_cache_getroute,
1865                       dn_cache_dump, NULL);
1866 #endif
1867 }
1868
1869 void __exit dn_route_cleanup(void)
1870 {
1871         del_timer(&dn_route_timer);
1872         dn_run_flush(0);
1873
1874         proc_net_remove(&init_net, "decnet_cache");
1875         dst_entries_destroy(&dn_dst_ops);
1876 }
1877