[SK_BUFF]: Introduce skb_network_header()
[linux-2.6.git] / net / decnet / dn_route.c
1 /*
2  * DECnet       An implementation of the DECnet protocol suite for the LINUX
3  *              operating system.  DECnet is implemented using the  BSD Socket
4  *              interface as the means of communication with the user level.
5  *
6  *              DECnet Routing Functions (Endnode and Router)
7  *
8  * Authors:     Steve Whitehouse <SteveW@ACM.org>
9  *              Eduardo Marcelo Serrat <emserrat@geocities.com>
10  *
11  * Changes:
12  *              Steve Whitehouse : Fixes to allow "intra-ethernet" and
13  *                                 "return-to-sender" bits on outgoing
14  *                                 packets.
15  *              Steve Whitehouse : Timeouts for cached routes.
16  *              Steve Whitehouse : Use dst cache for input routes too.
17  *              Steve Whitehouse : Fixed error values in dn_send_skb.
18  *              Steve Whitehouse : Rework routing functions to better fit
19  *                                 DECnet routing design
20  *              Alexey Kuznetsov : New SMP locking
21  *              Steve Whitehouse : More SMP locking changes & dn_cache_dump()
22  *              Steve Whitehouse : Prerouting NF hook, now really is prerouting.
23  *                                 Fixed possible skb leak in rtnetlink funcs.
24  *              Steve Whitehouse : Dave Miller's dynamic hash table sizing and
25  *                                 Alexey Kuznetsov's finer grained locking
26  *                                 from ipv4/route.c.
27  *              Steve Whitehouse : Routing is now starting to look like a
28  *                                 sensible set of code now, mainly due to
29  *                                 my copying the IPv4 routing code. The
30  *                                 hooks here are modified and will continue
31  *                                 to evolve for a while.
32  *              Steve Whitehouse : Real SMP at last :-) Also new netfilter
33  *                                 stuff. Look out raw sockets your days
34  *                                 are numbered!
35  *              Steve Whitehouse : Added return-to-sender functions. Added
36  *                                 backlog congestion level return codes.
37  *              Steve Whitehouse : Fixed bug where routes were set up with
38  *                                 no ref count on net devices.
39  *              Steve Whitehouse : RCU for the route cache
40  *              Steve Whitehouse : Preparations for the flow cache
41  *              Steve Whitehouse : Prepare for nonlinear skbs
42  */
43
44 /******************************************************************************
45     (c) 1995-1998 E.M. Serrat           emserrat@geocities.com
46
47     This program is free software; you can redistribute it and/or modify
48     it under the terms of the GNU General Public License as published by
49     the Free Software Foundation; either version 2 of the License, or
50     any later version.
51
52     This program is distributed in the hope that it will be useful,
53     but WITHOUT ANY WARRANTY; without even the implied warranty of
54     MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
55     GNU General Public License for more details.
56 *******************************************************************************/
57
58 #include <linux/errno.h>
59 #include <linux/types.h>
60 #include <linux/socket.h>
61 #include <linux/in.h>
62 #include <linux/kernel.h>
63 #include <linux/sockios.h>
64 #include <linux/net.h>
65 #include <linux/netdevice.h>
66 #include <linux/inet.h>
67 #include <linux/route.h>
68 #include <linux/in_route.h>
69 #include <net/sock.h>
70 #include <linux/mm.h>
71 #include <linux/proc_fs.h>
72 #include <linux/seq_file.h>
73 #include <linux/init.h>
74 #include <linux/rtnetlink.h>
75 #include <linux/string.h>
76 #include <linux/netfilter_decnet.h>
77 #include <linux/rcupdate.h>
78 #include <linux/times.h>
79 #include <asm/errno.h>
80 #include <net/neighbour.h>
81 #include <net/dst.h>
82 #include <net/flow.h>
83 #include <net/fib_rules.h>
84 #include <net/dn.h>
85 #include <net/dn_dev.h>
86 #include <net/dn_nsp.h>
87 #include <net/dn_route.h>
88 #include <net/dn_neigh.h>
89 #include <net/dn_fib.h>
90
91 struct dn_rt_hash_bucket
92 {
93         struct dn_route *chain;
94         spinlock_t lock;
95 } __attribute__((__aligned__(8)));
96
97 extern struct neigh_table dn_neigh_table;
98
99
100 static unsigned char dn_hiord_addr[6] = {0xAA,0x00,0x04,0x00,0x00,0x00};
101
102 static const int dn_rt_min_delay = 2 * HZ;
103 static const int dn_rt_max_delay = 10 * HZ;
104 static const int dn_rt_mtu_expires = 10 * 60 * HZ;
105
106 static unsigned long dn_rt_deadline;
107
108 static int dn_dst_gc(void);
109 static struct dst_entry *dn_dst_check(struct dst_entry *, __u32);
110 static struct dst_entry *dn_dst_negative_advice(struct dst_entry *);
111 static void dn_dst_link_failure(struct sk_buff *);
112 static void dn_dst_update_pmtu(struct dst_entry *dst, u32 mtu);
113 static int dn_route_input(struct sk_buff *);
114 static void dn_run_flush(unsigned long dummy);
115
116 static struct dn_rt_hash_bucket *dn_rt_hash_table;
117 static unsigned dn_rt_hash_mask;
118
119 static struct timer_list dn_route_timer;
120 static DEFINE_TIMER(dn_rt_flush_timer, dn_run_flush, 0, 0);
121 int decnet_dst_gc_interval = 2;
122
123 static struct dst_ops dn_dst_ops = {
124         .family =               PF_DECnet,
125         .protocol =             __constant_htons(ETH_P_DNA_RT),
126         .gc_thresh =            128,
127         .gc =                   dn_dst_gc,
128         .check =                dn_dst_check,
129         .negative_advice =      dn_dst_negative_advice,
130         .link_failure =         dn_dst_link_failure,
131         .update_pmtu =          dn_dst_update_pmtu,
132         .entry_size =           sizeof(struct dn_route),
133         .entries =              ATOMIC_INIT(0),
134 };
135
136 static __inline__ unsigned dn_hash(__le16 src, __le16 dst)
137 {
138         __u16 tmp = (__u16 __force)(src ^ dst);
139         tmp ^= (tmp >> 3);
140         tmp ^= (tmp >> 5);
141         tmp ^= (tmp >> 10);
142         return dn_rt_hash_mask & (unsigned)tmp;
143 }
144
145 static inline void dnrt_free(struct dn_route *rt)
146 {
147         call_rcu_bh(&rt->u.dst.rcu_head, dst_rcu_free);
148 }
149
150 static inline void dnrt_drop(struct dn_route *rt)
151 {
152         dst_release(&rt->u.dst);
153         call_rcu_bh(&rt->u.dst.rcu_head, dst_rcu_free);
154 }
155
156 static void dn_dst_check_expire(unsigned long dummy)
157 {
158         int i;
159         struct dn_route *rt, **rtp;
160         unsigned long now = jiffies;
161         unsigned long expire = 120 * HZ;
162
163         for(i = 0; i <= dn_rt_hash_mask; i++) {
164                 rtp = &dn_rt_hash_table[i].chain;
165
166                 spin_lock(&dn_rt_hash_table[i].lock);
167                 while((rt=*rtp) != NULL) {
168                         if (atomic_read(&rt->u.dst.__refcnt) ||
169                                         (now - rt->u.dst.lastuse) < expire) {
170                                 rtp = &rt->u.dst.dn_next;
171                                 continue;
172                         }
173                         *rtp = rt->u.dst.dn_next;
174                         rt->u.dst.dn_next = NULL;
175                         dnrt_free(rt);
176                 }
177                 spin_unlock(&dn_rt_hash_table[i].lock);
178
179                 if ((jiffies - now) > 0)
180                         break;
181         }
182
183         mod_timer(&dn_route_timer, now + decnet_dst_gc_interval * HZ);
184 }
185
186 static int dn_dst_gc(void)
187 {
188         struct dn_route *rt, **rtp;
189         int i;
190         unsigned long now = jiffies;
191         unsigned long expire = 10 * HZ;
192
193         for(i = 0; i <= dn_rt_hash_mask; i++) {
194
195                 spin_lock_bh(&dn_rt_hash_table[i].lock);
196                 rtp = &dn_rt_hash_table[i].chain;
197
198                 while((rt=*rtp) != NULL) {
199                         if (atomic_read(&rt->u.dst.__refcnt) ||
200                                         (now - rt->u.dst.lastuse) < expire) {
201                                 rtp = &rt->u.dst.dn_next;
202                                 continue;
203                         }
204                         *rtp = rt->u.dst.dn_next;
205                         rt->u.dst.dn_next = NULL;
206                         dnrt_drop(rt);
207                         break;
208                 }
209                 spin_unlock_bh(&dn_rt_hash_table[i].lock);
210         }
211
212         return 0;
213 }
214
215 /*
216  * The decnet standards don't impose a particular minimum mtu, what they
217  * do insist on is that the routing layer accepts a datagram of at least
218  * 230 bytes long. Here we have to subtract the routing header length from
219  * 230 to get the minimum acceptable mtu. If there is no neighbour, then we
220  * assume the worst and use a long header size.
221  *
222  * We update both the mtu and the advertised mss (i.e. the segment size we
223  * advertise to the other end).
224  */
225 static void dn_dst_update_pmtu(struct dst_entry *dst, u32 mtu)
226 {
227         u32 min_mtu = 230;
228         struct dn_dev *dn = dst->neighbour ?
229                             (struct dn_dev *)dst->neighbour->dev->dn_ptr : NULL;
230
231         if (dn && dn->use_long == 0)
232                 min_mtu -= 6;
233         else
234                 min_mtu -= 21;
235
236         if (dst->metrics[RTAX_MTU-1] > mtu && mtu >= min_mtu) {
237                 if (!(dst_metric_locked(dst, RTAX_MTU))) {
238                         dst->metrics[RTAX_MTU-1] = mtu;
239                         dst_set_expires(dst, dn_rt_mtu_expires);
240                 }
241                 if (!(dst_metric_locked(dst, RTAX_ADVMSS))) {
242                         u32 mss = mtu - DN_MAX_NSP_DATA_HEADER;
243                         if (dst->metrics[RTAX_ADVMSS-1] > mss)
244                                 dst->metrics[RTAX_ADVMSS-1] = mss;
245                 }
246         }
247 }
248
249 /*
250  * When a route has been marked obsolete. (e.g. routing cache flush)
251  */
252 static struct dst_entry *dn_dst_check(struct dst_entry *dst, __u32 cookie)
253 {
254         return NULL;
255 }
256
257 static struct dst_entry *dn_dst_negative_advice(struct dst_entry *dst)
258 {
259         dst_release(dst);
260         return NULL;
261 }
262
263 static void dn_dst_link_failure(struct sk_buff *skb)
264 {
265         return;
266 }
267
268 static inline int compare_keys(struct flowi *fl1, struct flowi *fl2)
269 {
270         return ((fl1->nl_u.dn_u.daddr ^ fl2->nl_u.dn_u.daddr) |
271                 (fl1->nl_u.dn_u.saddr ^ fl2->nl_u.dn_u.saddr) |
272                 (fl1->mark ^ fl2->mark) |
273                 (fl1->nl_u.dn_u.scope ^ fl2->nl_u.dn_u.scope) |
274                 (fl1->oif ^ fl2->oif) |
275                 (fl1->iif ^ fl2->iif)) == 0;
276 }
277
278 static int dn_insert_route(struct dn_route *rt, unsigned hash, struct dn_route **rp)
279 {
280         struct dn_route *rth, **rthp;
281         unsigned long now = jiffies;
282
283         rthp = &dn_rt_hash_table[hash].chain;
284
285         spin_lock_bh(&dn_rt_hash_table[hash].lock);
286         while((rth = *rthp) != NULL) {
287                 if (compare_keys(&rth->fl, &rt->fl)) {
288                         /* Put it first */
289                         *rthp = rth->u.dst.dn_next;
290                         rcu_assign_pointer(rth->u.dst.dn_next,
291                                            dn_rt_hash_table[hash].chain);
292                         rcu_assign_pointer(dn_rt_hash_table[hash].chain, rth);
293
294                         rth->u.dst.__use++;
295                         dst_hold(&rth->u.dst);
296                         rth->u.dst.lastuse = now;
297                         spin_unlock_bh(&dn_rt_hash_table[hash].lock);
298
299                         dnrt_drop(rt);
300                         *rp = rth;
301                         return 0;
302                 }
303                 rthp = &rth->u.dst.dn_next;
304         }
305
306         rcu_assign_pointer(rt->u.dst.dn_next, dn_rt_hash_table[hash].chain);
307         rcu_assign_pointer(dn_rt_hash_table[hash].chain, rt);
308
309         dst_hold(&rt->u.dst);
310         rt->u.dst.__use++;
311         rt->u.dst.lastuse = now;
312         spin_unlock_bh(&dn_rt_hash_table[hash].lock);
313         *rp = rt;
314         return 0;
315 }
316
317 void dn_run_flush(unsigned long dummy)
318 {
319         int i;
320         struct dn_route *rt, *next;
321
322         for(i = 0; i < dn_rt_hash_mask; i++) {
323                 spin_lock_bh(&dn_rt_hash_table[i].lock);
324
325                 if ((rt = xchg(&dn_rt_hash_table[i].chain, NULL)) == NULL)
326                         goto nothing_to_declare;
327
328                 for(; rt; rt=next) {
329                         next = rt->u.dst.dn_next;
330                         rt->u.dst.dn_next = NULL;
331                         dst_free((struct dst_entry *)rt);
332                 }
333
334 nothing_to_declare:
335                 spin_unlock_bh(&dn_rt_hash_table[i].lock);
336         }
337 }
338
339 static DEFINE_SPINLOCK(dn_rt_flush_lock);
340
341 void dn_rt_cache_flush(int delay)
342 {
343         unsigned long now = jiffies;
344         int user_mode = !in_interrupt();
345
346         if (delay < 0)
347                 delay = dn_rt_min_delay;
348
349         spin_lock_bh(&dn_rt_flush_lock);
350
351         if (del_timer(&dn_rt_flush_timer) && delay > 0 && dn_rt_deadline) {
352                 long tmo = (long)(dn_rt_deadline - now);
353
354                 if (user_mode && tmo < dn_rt_max_delay - dn_rt_min_delay)
355                         tmo = 0;
356
357                 if (delay > tmo)
358                         delay = tmo;
359         }
360
361         if (delay <= 0) {
362                 spin_unlock_bh(&dn_rt_flush_lock);
363                 dn_run_flush(0);
364                 return;
365         }
366
367         if (dn_rt_deadline == 0)
368                 dn_rt_deadline = now + dn_rt_max_delay;
369
370         dn_rt_flush_timer.expires = now + delay;
371         add_timer(&dn_rt_flush_timer);
372         spin_unlock_bh(&dn_rt_flush_lock);
373 }
374
375 /**
376  * dn_return_short - Return a short packet to its sender
377  * @skb: The packet to return
378  *
379  */
380 static int dn_return_short(struct sk_buff *skb)
381 {
382         struct dn_skb_cb *cb;
383         unsigned char *ptr;
384         __le16 *src;
385         __le16 *dst;
386         __le16 tmp;
387
388         /* Add back headers */
389         skb_push(skb, skb->data - skb_network_header(skb));
390
391         if ((skb = skb_unshare(skb, GFP_ATOMIC)) == NULL)
392                 return NET_RX_DROP;
393
394         cb = DN_SKB_CB(skb);
395         /* Skip packet length and point to flags */
396         ptr = skb->data + 2;
397         *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
398
399         dst = (__le16 *)ptr;
400         ptr += 2;
401         src = (__le16 *)ptr;
402         ptr += 2;
403         *ptr = 0; /* Zero hop count */
404
405         /* Swap source and destination */
406         tmp  = *src;
407         *src = *dst;
408         *dst = tmp;
409
410         skb->pkt_type = PACKET_OUTGOING;
411         dn_rt_finish_output(skb, NULL, NULL);
412         return NET_RX_SUCCESS;
413 }
414
415 /**
416  * dn_return_long - Return a long packet to its sender
417  * @skb: The long format packet to return
418  *
419  */
420 static int dn_return_long(struct sk_buff *skb)
421 {
422         struct dn_skb_cb *cb;
423         unsigned char *ptr;
424         unsigned char *src_addr, *dst_addr;
425         unsigned char tmp[ETH_ALEN];
426
427         /* Add back all headers */
428         skb_push(skb, skb->data - skb_network_header(skb));
429
430         if ((skb = skb_unshare(skb, GFP_ATOMIC)) == NULL)
431                 return NET_RX_DROP;
432
433         cb = DN_SKB_CB(skb);
434         /* Ignore packet length and point to flags */
435         ptr = skb->data + 2;
436
437         /* Skip padding */
438         if (*ptr & DN_RT_F_PF) {
439                 char padlen = (*ptr & ~DN_RT_F_PF);
440                 ptr += padlen;
441         }
442
443         *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
444         ptr += 2;
445         dst_addr = ptr;
446         ptr += 8;
447         src_addr = ptr;
448         ptr += 6;
449         *ptr = 0; /* Zero hop count */
450
451         /* Swap source and destination */
452         memcpy(tmp, src_addr, ETH_ALEN);
453         memcpy(src_addr, dst_addr, ETH_ALEN);
454         memcpy(dst_addr, tmp, ETH_ALEN);
455
456         skb->pkt_type = PACKET_OUTGOING;
457         dn_rt_finish_output(skb, dst_addr, src_addr);
458         return NET_RX_SUCCESS;
459 }
460
461 /**
462  * dn_route_rx_packet - Try and find a route for an incoming packet
463  * @skb: The packet to find a route for
464  *
465  * Returns: result of input function if route is found, error code otherwise
466  */
467 static int dn_route_rx_packet(struct sk_buff *skb)
468 {
469         struct dn_skb_cb *cb = DN_SKB_CB(skb);
470         int err;
471
472         if ((err = dn_route_input(skb)) == 0)
473                 return dst_input(skb);
474
475         if (decnet_debug_level & 4) {
476                 char *devname = skb->dev ? skb->dev->name : "???";
477                 struct dn_skb_cb *cb = DN_SKB_CB(skb);
478                 printk(KERN_DEBUG
479                         "DECnet: dn_route_rx_packet: rt_flags=0x%02x dev=%s len=%d src=0x%04hx dst=0x%04hx err=%d type=%d\n",
480                         (int)cb->rt_flags, devname, skb->len,
481                         dn_ntohs(cb->src), dn_ntohs(cb->dst),
482                         err, skb->pkt_type);
483         }
484
485         if ((skb->pkt_type == PACKET_HOST) && (cb->rt_flags & DN_RT_F_RQR)) {
486                 switch(cb->rt_flags & DN_RT_PKT_MSK) {
487                         case DN_RT_PKT_SHORT:
488                                 return dn_return_short(skb);
489                         case DN_RT_PKT_LONG:
490                                 return dn_return_long(skb);
491                 }
492         }
493
494         kfree_skb(skb);
495         return NET_RX_DROP;
496 }
497
498 static int dn_route_rx_long(struct sk_buff *skb)
499 {
500         struct dn_skb_cb *cb = DN_SKB_CB(skb);
501         unsigned char *ptr = skb->data;
502
503         if (!pskb_may_pull(skb, 21)) /* 20 for long header, 1 for shortest nsp */
504                 goto drop_it;
505
506         skb_pull(skb, 20);
507         skb->h.raw = skb->data;
508
509         /* Destination info */
510         ptr += 2;
511         cb->dst = dn_eth2dn(ptr);
512         if (memcmp(ptr, dn_hiord_addr, 4) != 0)
513                 goto drop_it;
514         ptr += 6;
515
516
517         /* Source info */
518         ptr += 2;
519         cb->src = dn_eth2dn(ptr);
520         if (memcmp(ptr, dn_hiord_addr, 4) != 0)
521                 goto drop_it;
522         ptr += 6;
523         /* Other junk */
524         ptr++;
525         cb->hops = *ptr++; /* Visit Count */
526
527         return NF_HOOK(PF_DECnet, NF_DN_PRE_ROUTING, skb, skb->dev, NULL, dn_route_rx_packet);
528
529 drop_it:
530         kfree_skb(skb);
531         return NET_RX_DROP;
532 }
533
534
535
536 static int dn_route_rx_short(struct sk_buff *skb)
537 {
538         struct dn_skb_cb *cb = DN_SKB_CB(skb);
539         unsigned char *ptr = skb->data;
540
541         if (!pskb_may_pull(skb, 6)) /* 5 for short header + 1 for shortest nsp */
542                 goto drop_it;
543
544         skb_pull(skb, 5);
545         skb->h.raw = skb->data;
546
547         cb->dst = *(__le16 *)ptr;
548         ptr += 2;
549         cb->src = *(__le16 *)ptr;
550         ptr += 2;
551         cb->hops = *ptr & 0x3f;
552
553         return NF_HOOK(PF_DECnet, NF_DN_PRE_ROUTING, skb, skb->dev, NULL, dn_route_rx_packet);
554
555 drop_it:
556         kfree_skb(skb);
557         return NET_RX_DROP;
558 }
559
560 static int dn_route_discard(struct sk_buff *skb)
561 {
562         /*
563          * I know we drop the packet here, but thats considered success in
564          * this case
565          */
566         kfree_skb(skb);
567         return NET_RX_SUCCESS;
568 }
569
570 static int dn_route_ptp_hello(struct sk_buff *skb)
571 {
572         dn_dev_hello(skb);
573         dn_neigh_pointopoint_hello(skb);
574         return NET_RX_SUCCESS;
575 }
576
577 int dn_route_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
578 {
579         struct dn_skb_cb *cb;
580         unsigned char flags = 0;
581         __u16 len = dn_ntohs(*(__le16 *)skb->data);
582         struct dn_dev *dn = (struct dn_dev *)dev->dn_ptr;
583         unsigned char padlen = 0;
584
585         if (dn == NULL)
586                 goto dump_it;
587
588         if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
589                 goto out;
590
591         if (!pskb_may_pull(skb, 3))
592                 goto dump_it;
593
594         skb_pull(skb, 2);
595
596         if (len > skb->len)
597                 goto dump_it;
598
599         skb_trim(skb, len);
600
601         flags = *skb->data;
602
603         cb = DN_SKB_CB(skb);
604         cb->stamp = jiffies;
605         cb->iif = dev->ifindex;
606
607         /*
608          * If we have padding, remove it.
609          */
610         if (flags & DN_RT_F_PF) {
611                 padlen = flags & ~DN_RT_F_PF;
612                 if (!pskb_may_pull(skb, padlen + 1))
613                         goto dump_it;
614                 skb_pull(skb, padlen);
615                 flags = *skb->data;
616         }
617
618         skb_reset_network_header(skb);
619
620         /*
621          * Weed out future version DECnet
622          */
623         if (flags & DN_RT_F_VER)
624                 goto dump_it;
625
626         cb->rt_flags = flags;
627
628         if (decnet_debug_level & 1)
629                 printk(KERN_DEBUG
630                         "dn_route_rcv: got 0x%02x from %s [%d %d %d]\n",
631                         (int)flags, (dev) ? dev->name : "???", len, skb->len,
632                         padlen);
633
634         if (flags & DN_RT_PKT_CNTL) {
635                 if (unlikely(skb_linearize(skb)))
636                         goto dump_it;
637
638                 switch(flags & DN_RT_CNTL_MSK) {
639                         case DN_RT_PKT_INIT:
640                                 dn_dev_init_pkt(skb);
641                                 break;
642                         case DN_RT_PKT_VERI:
643                                 dn_dev_veri_pkt(skb);
644                                 break;
645                 }
646
647                 if (dn->parms.state != DN_DEV_S_RU)
648                         goto dump_it;
649
650                 switch(flags & DN_RT_CNTL_MSK) {
651                         case DN_RT_PKT_HELO:
652                                 return NF_HOOK(PF_DECnet, NF_DN_HELLO, skb, skb->dev, NULL, dn_route_ptp_hello);
653
654                         case DN_RT_PKT_L1RT:
655                         case DN_RT_PKT_L2RT:
656                                 return NF_HOOK(PF_DECnet, NF_DN_ROUTE, skb, skb->dev, NULL, dn_route_discard);
657                         case DN_RT_PKT_ERTH:
658                                 return NF_HOOK(PF_DECnet, NF_DN_HELLO, skb, skb->dev, NULL, dn_neigh_router_hello);
659
660                         case DN_RT_PKT_EEDH:
661                                 return NF_HOOK(PF_DECnet, NF_DN_HELLO, skb, skb->dev, NULL, dn_neigh_endnode_hello);
662                 }
663         } else {
664                 if (dn->parms.state != DN_DEV_S_RU)
665                         goto dump_it;
666
667                 skb_pull(skb, 1); /* Pull flags */
668
669                 switch(flags & DN_RT_PKT_MSK) {
670                         case DN_RT_PKT_LONG:
671                                 return dn_route_rx_long(skb);
672                         case DN_RT_PKT_SHORT:
673                                 return dn_route_rx_short(skb);
674                 }
675         }
676
677 dump_it:
678         kfree_skb(skb);
679 out:
680         return NET_RX_DROP;
681 }
682
683 static int dn_output(struct sk_buff *skb)
684 {
685         struct dst_entry *dst = skb->dst;
686         struct dn_route *rt = (struct dn_route *)dst;
687         struct net_device *dev = dst->dev;
688         struct dn_skb_cb *cb = DN_SKB_CB(skb);
689         struct neighbour *neigh;
690
691         int err = -EINVAL;
692
693         if ((neigh = dst->neighbour) == NULL)
694                 goto error;
695
696         skb->dev = dev;
697
698         cb->src = rt->rt_saddr;
699         cb->dst = rt->rt_daddr;
700
701         /*
702          * Always set the Intra-Ethernet bit on all outgoing packets
703          * originated on this node. Only valid flag from upper layers
704          * is return-to-sender-requested. Set hop count to 0 too.
705          */
706         cb->rt_flags &= ~DN_RT_F_RQR;
707         cb->rt_flags |= DN_RT_F_IE;
708         cb->hops = 0;
709
710         return NF_HOOK(PF_DECnet, NF_DN_LOCAL_OUT, skb, NULL, dev, neigh->output);
711
712 error:
713         if (net_ratelimit())
714                 printk(KERN_DEBUG "dn_output: This should not happen\n");
715
716         kfree_skb(skb);
717
718         return err;
719 }
720
721 static int dn_forward(struct sk_buff *skb)
722 {
723         struct dn_skb_cb *cb = DN_SKB_CB(skb);
724         struct dst_entry *dst = skb->dst;
725         struct dn_dev *dn_db = dst->dev->dn_ptr;
726         struct dn_route *rt;
727         struct neighbour *neigh = dst->neighbour;
728         int header_len;
729 #ifdef CONFIG_NETFILTER
730         struct net_device *dev = skb->dev;
731 #endif
732
733         if (skb->pkt_type != PACKET_HOST)
734                 goto drop;
735
736         /* Ensure that we have enough space for headers */
737         rt = (struct dn_route *)skb->dst;
738         header_len = dn_db->use_long ? 21 : 6;
739         if (skb_cow(skb, LL_RESERVED_SPACE(rt->u.dst.dev)+header_len))
740                 goto drop;
741
742         /*
743          * Hop count exceeded.
744          */
745         if (++cb->hops > 30)
746                 goto drop;
747
748         skb->dev = rt->u.dst.dev;
749
750         /*
751          * If packet goes out same interface it came in on, then set
752          * the Intra-Ethernet bit. This has no effect for short
753          * packets, so we don't need to test for them here.
754          */
755         cb->rt_flags &= ~DN_RT_F_IE;
756         if (rt->rt_flags & RTCF_DOREDIRECT)
757                 cb->rt_flags |= DN_RT_F_IE;
758
759         return NF_HOOK(PF_DECnet, NF_DN_FORWARD, skb, dev, skb->dev, neigh->output);
760
761 drop:
762         kfree_skb(skb);
763         return NET_RX_DROP;
764 }
765
766 /*
767  * Drop packet. This is used for endnodes and for
768  * when we should not be forwarding packets from
769  * this dest.
770  */
771 static int dn_blackhole(struct sk_buff *skb)
772 {
773         kfree_skb(skb);
774         return NET_RX_DROP;
775 }
776
777 /*
778  * Used to catch bugs. This should never normally get
779  * called.
780  */
781 static int dn_rt_bug(struct sk_buff *skb)
782 {
783         if (net_ratelimit()) {
784                 struct dn_skb_cb *cb = DN_SKB_CB(skb);
785
786                 printk(KERN_DEBUG "dn_rt_bug: skb from:%04x to:%04x\n",
787                                 dn_ntohs(cb->src), dn_ntohs(cb->dst));
788         }
789
790         kfree_skb(skb);
791
792         return NET_RX_BAD;
793 }
794
795 static int dn_rt_set_next_hop(struct dn_route *rt, struct dn_fib_res *res)
796 {
797         struct dn_fib_info *fi = res->fi;
798         struct net_device *dev = rt->u.dst.dev;
799         struct neighbour *n;
800         unsigned mss;
801
802         if (fi) {
803                 if (DN_FIB_RES_GW(*res) &&
804                     DN_FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
805                         rt->rt_gateway = DN_FIB_RES_GW(*res);
806                 memcpy(rt->u.dst.metrics, fi->fib_metrics,
807                        sizeof(rt->u.dst.metrics));
808         }
809         rt->rt_type = res->type;
810
811         if (dev != NULL && rt->u.dst.neighbour == NULL) {
812                 n = __neigh_lookup_errno(&dn_neigh_table, &rt->rt_gateway, dev);
813                 if (IS_ERR(n))
814                         return PTR_ERR(n);
815                 rt->u.dst.neighbour = n;
816         }
817
818         if (rt->u.dst.metrics[RTAX_MTU-1] == 0 ||
819             rt->u.dst.metrics[RTAX_MTU-1] > rt->u.dst.dev->mtu)
820                 rt->u.dst.metrics[RTAX_MTU-1] = rt->u.dst.dev->mtu;
821         mss = dn_mss_from_pmtu(dev, dst_mtu(&rt->u.dst));
822         if (rt->u.dst.metrics[RTAX_ADVMSS-1] == 0 ||
823             rt->u.dst.metrics[RTAX_ADVMSS-1] > mss)
824                 rt->u.dst.metrics[RTAX_ADVMSS-1] = mss;
825         return 0;
826 }
827
828 static inline int dn_match_addr(__le16 addr1, __le16 addr2)
829 {
830         __u16 tmp = dn_ntohs(addr1) ^ dn_ntohs(addr2);
831         int match = 16;
832         while(tmp) {
833                 tmp >>= 1;
834                 match--;
835         }
836         return match;
837 }
838
839 static __le16 dnet_select_source(const struct net_device *dev, __le16 daddr, int scope)
840 {
841         __le16 saddr = 0;
842         struct dn_dev *dn_db = dev->dn_ptr;
843         struct dn_ifaddr *ifa;
844         int best_match = 0;
845         int ret;
846
847         read_lock(&dev_base_lock);
848         for(ifa = dn_db->ifa_list; ifa; ifa = ifa->ifa_next) {
849                 if (ifa->ifa_scope > scope)
850                         continue;
851                 if (!daddr) {
852                         saddr = ifa->ifa_local;
853                         break;
854                 }
855                 ret = dn_match_addr(daddr, ifa->ifa_local);
856                 if (ret > best_match)
857                         saddr = ifa->ifa_local;
858                 if (best_match == 0)
859                         saddr = ifa->ifa_local;
860         }
861         read_unlock(&dev_base_lock);
862
863         return saddr;
864 }
865
866 static inline __le16 __dn_fib_res_prefsrc(struct dn_fib_res *res)
867 {
868         return dnet_select_source(DN_FIB_RES_DEV(*res), DN_FIB_RES_GW(*res), res->scope);
869 }
870
871 static inline __le16 dn_fib_rules_map_destination(__le16 daddr, struct dn_fib_res *res)
872 {
873         __le16 mask = dnet_make_mask(res->prefixlen);
874         return (daddr&~mask)|res->fi->fib_nh->nh_gw;
875 }
876
877 static int dn_route_output_slow(struct dst_entry **pprt, const struct flowi *oldflp, int try_hard)
878 {
879         struct flowi fl = { .nl_u = { .dn_u =
880                                       { .daddr = oldflp->fld_dst,
881                                         .saddr = oldflp->fld_src,
882                                         .scope = RT_SCOPE_UNIVERSE,
883                                      } },
884                             .mark = oldflp->mark,
885                             .iif = loopback_dev.ifindex,
886                             .oif = oldflp->oif };
887         struct dn_route *rt = NULL;
888         struct net_device *dev_out = NULL;
889         struct neighbour *neigh = NULL;
890         unsigned hash;
891         unsigned flags = 0;
892         struct dn_fib_res res = { .fi = NULL, .type = RTN_UNICAST };
893         int err;
894         int free_res = 0;
895         __le16 gateway = 0;
896
897         if (decnet_debug_level & 16)
898                 printk(KERN_DEBUG
899                        "dn_route_output_slow: dst=%04x src=%04x mark=%d"
900                        " iif=%d oif=%d\n", dn_ntohs(oldflp->fld_dst),
901                        dn_ntohs(oldflp->fld_src),
902                        oldflp->mark, loopback_dev.ifindex, oldflp->oif);
903
904         /* If we have an output interface, verify its a DECnet device */
905         if (oldflp->oif) {
906                 dev_out = dev_get_by_index(oldflp->oif);
907                 err = -ENODEV;
908                 if (dev_out && dev_out->dn_ptr == NULL) {
909                         dev_put(dev_out);
910                         dev_out = NULL;
911                 }
912                 if (dev_out == NULL)
913                         goto out;
914         }
915
916         /* If we have a source address, verify that its a local address */
917         if (oldflp->fld_src) {
918                 err = -EADDRNOTAVAIL;
919
920                 if (dev_out) {
921                         if (dn_dev_islocal(dev_out, oldflp->fld_src))
922                                 goto source_ok;
923                         dev_put(dev_out);
924                         goto out;
925                 }
926                 read_lock(&dev_base_lock);
927                 for(dev_out = dev_base; dev_out; dev_out = dev_out->next) {
928                         if (!dev_out->dn_ptr)
929                                 continue;
930                         if (!dn_dev_islocal(dev_out, oldflp->fld_src))
931                                 continue;
932                         if ((dev_out->flags & IFF_LOOPBACK) &&
933                             oldflp->fld_dst &&
934                             !dn_dev_islocal(dev_out, oldflp->fld_dst))
935                                 continue;
936                         break;
937                 }
938                 read_unlock(&dev_base_lock);
939                 if (dev_out == NULL)
940                         goto out;
941                 dev_hold(dev_out);
942 source_ok:
943                 ;
944         }
945
946         /* No destination? Assume its local */
947         if (!fl.fld_dst) {
948                 fl.fld_dst = fl.fld_src;
949
950                 err = -EADDRNOTAVAIL;
951                 if (dev_out)
952                         dev_put(dev_out);
953                 dev_out = &loopback_dev;
954                 dev_hold(dev_out);
955                 if (!fl.fld_dst) {
956                         fl.fld_dst =
957                         fl.fld_src = dnet_select_source(dev_out, 0,
958                                                        RT_SCOPE_HOST);
959                         if (!fl.fld_dst)
960                                 goto out;
961                 }
962                 fl.oif = loopback_dev.ifindex;
963                 res.type = RTN_LOCAL;
964                 goto make_route;
965         }
966
967         if (decnet_debug_level & 16)
968                 printk(KERN_DEBUG
969                        "dn_route_output_slow: initial checks complete."
970                        " dst=%o4x src=%04x oif=%d try_hard=%d\n",
971                        dn_ntohs(fl.fld_dst), dn_ntohs(fl.fld_src),
972                        fl.oif, try_hard);
973
974         /*
975          * N.B. If the kernel is compiled without router support then
976          * dn_fib_lookup() will evaluate to non-zero so this if () block
977          * will always be executed.
978          */
979         err = -ESRCH;
980         if (try_hard || (err = dn_fib_lookup(&fl, &res)) != 0) {
981                 struct dn_dev *dn_db;
982                 if (err != -ESRCH)
983                         goto out;
984                 /*
985                  * Here the fallback is basically the standard algorithm for
986                  * routing in endnodes which is described in the DECnet routing
987                  * docs
988                  *
989                  * If we are not trying hard, look in neighbour cache.
990                  * The result is tested to ensure that if a specific output
991                  * device/source address was requested, then we honour that
992                  * here
993                  */
994                 if (!try_hard) {
995                         neigh = neigh_lookup_nodev(&dn_neigh_table, &fl.fld_dst);
996                         if (neigh) {
997                                 if ((oldflp->oif &&
998                                     (neigh->dev->ifindex != oldflp->oif)) ||
999                                     (oldflp->fld_src &&
1000                                     (!dn_dev_islocal(neigh->dev,
1001                                                       oldflp->fld_src)))) {
1002                                         neigh_release(neigh);
1003                                         neigh = NULL;
1004                                 } else {
1005                                         if (dev_out)
1006                                                 dev_put(dev_out);
1007                                         if (dn_dev_islocal(neigh->dev, fl.fld_dst)) {
1008                                                 dev_out = &loopback_dev;
1009                                                 res.type = RTN_LOCAL;
1010                                         } else {
1011                                                 dev_out = neigh->dev;
1012                                         }
1013                                         dev_hold(dev_out);
1014                                         goto select_source;
1015                                 }
1016                         }
1017                 }
1018
1019                 /* Not there? Perhaps its a local address */
1020                 if (dev_out == NULL)
1021                         dev_out = dn_dev_get_default();
1022                 err = -ENODEV;
1023                 if (dev_out == NULL)
1024                         goto out;
1025                 dn_db = dev_out->dn_ptr;
1026                 /* Possible improvement - check all devices for local addr */
1027                 if (dn_dev_islocal(dev_out, fl.fld_dst)) {
1028                         dev_put(dev_out);
1029                         dev_out = &loopback_dev;
1030                         dev_hold(dev_out);
1031                         res.type = RTN_LOCAL;
1032                         goto select_source;
1033                 }
1034                 /* Not local either.... try sending it to the default router */
1035                 neigh = neigh_clone(dn_db->router);
1036                 BUG_ON(neigh && neigh->dev != dev_out);
1037
1038                 /* Ok then, we assume its directly connected and move on */
1039 select_source:
1040                 if (neigh)
1041                         gateway = ((struct dn_neigh *)neigh)->addr;
1042                 if (gateway == 0)
1043                         gateway = fl.fld_dst;
1044                 if (fl.fld_src == 0) {
1045                         fl.fld_src = dnet_select_source(dev_out, gateway,
1046                                                          res.type == RTN_LOCAL ?
1047                                                          RT_SCOPE_HOST :
1048                                                          RT_SCOPE_LINK);
1049                         if (fl.fld_src == 0 && res.type != RTN_LOCAL)
1050                                 goto e_addr;
1051                 }
1052                 fl.oif = dev_out->ifindex;
1053                 goto make_route;
1054         }
1055         free_res = 1;
1056
1057         if (res.type == RTN_NAT)
1058                 goto e_inval;
1059
1060         if (res.type == RTN_LOCAL) {
1061                 if (!fl.fld_src)
1062                         fl.fld_src = fl.fld_dst;
1063                 if (dev_out)
1064                         dev_put(dev_out);
1065                 dev_out = &loopback_dev;
1066                 dev_hold(dev_out);
1067                 fl.oif = dev_out->ifindex;
1068                 if (res.fi)
1069                         dn_fib_info_put(res.fi);
1070                 res.fi = NULL;
1071                 goto make_route;
1072         }
1073
1074         if (res.fi->fib_nhs > 1 && fl.oif == 0)
1075                 dn_fib_select_multipath(&fl, &res);
1076
1077         /*
1078          * We could add some logic to deal with default routes here and
1079          * get rid of some of the special casing above.
1080          */
1081
1082         if (!fl.fld_src)
1083                 fl.fld_src = DN_FIB_RES_PREFSRC(res);
1084
1085         if (dev_out)
1086                 dev_put(dev_out);
1087         dev_out = DN_FIB_RES_DEV(res);
1088         dev_hold(dev_out);
1089         fl.oif = dev_out->ifindex;
1090         gateway = DN_FIB_RES_GW(res);
1091
1092 make_route:
1093         if (dev_out->flags & IFF_LOOPBACK)
1094                 flags |= RTCF_LOCAL;
1095
1096         rt = dst_alloc(&dn_dst_ops);
1097         if (rt == NULL)
1098                 goto e_nobufs;
1099
1100         atomic_set(&rt->u.dst.__refcnt, 1);
1101         rt->u.dst.flags   = DST_HOST;
1102
1103         rt->fl.fld_src    = oldflp->fld_src;
1104         rt->fl.fld_dst    = oldflp->fld_dst;
1105         rt->fl.oif        = oldflp->oif;
1106         rt->fl.iif        = 0;
1107         rt->fl.mark       = oldflp->mark;
1108
1109         rt->rt_saddr      = fl.fld_src;
1110         rt->rt_daddr      = fl.fld_dst;
1111         rt->rt_gateway    = gateway ? gateway : fl.fld_dst;
1112         rt->rt_local_src  = fl.fld_src;
1113
1114         rt->rt_dst_map    = fl.fld_dst;
1115         rt->rt_src_map    = fl.fld_src;
1116
1117         rt->u.dst.dev = dev_out;
1118         dev_hold(dev_out);
1119         rt->u.dst.neighbour = neigh;
1120         neigh = NULL;
1121
1122         rt->u.dst.lastuse = jiffies;
1123         rt->u.dst.output  = dn_output;
1124         rt->u.dst.input   = dn_rt_bug;
1125         rt->rt_flags      = flags;
1126         if (flags & RTCF_LOCAL)
1127                 rt->u.dst.input = dn_nsp_rx;
1128
1129         err = dn_rt_set_next_hop(rt, &res);
1130         if (err)
1131                 goto e_neighbour;
1132
1133         hash = dn_hash(rt->fl.fld_src, rt->fl.fld_dst);
1134         dn_insert_route(rt, hash, (struct dn_route **)pprt);
1135
1136 done:
1137         if (neigh)
1138                 neigh_release(neigh);
1139         if (free_res)
1140                 dn_fib_res_put(&res);
1141         if (dev_out)
1142                 dev_put(dev_out);
1143 out:
1144         return err;
1145
1146 e_addr:
1147         err = -EADDRNOTAVAIL;
1148         goto done;
1149 e_inval:
1150         err = -EINVAL;
1151         goto done;
1152 e_nobufs:
1153         err = -ENOBUFS;
1154         goto done;
1155 e_neighbour:
1156         dst_free(&rt->u.dst);
1157         goto e_nobufs;
1158 }
1159
1160
1161 /*
1162  * N.B. The flags may be moved into the flowi at some future stage.
1163  */
1164 static int __dn_route_output_key(struct dst_entry **pprt, const struct flowi *flp, int flags)
1165 {
1166         unsigned hash = dn_hash(flp->fld_src, flp->fld_dst);
1167         struct dn_route *rt = NULL;
1168
1169         if (!(flags & MSG_TRYHARD)) {
1170                 rcu_read_lock_bh();
1171                 for(rt = rcu_dereference(dn_rt_hash_table[hash].chain); rt;
1172                         rt = rcu_dereference(rt->u.dst.dn_next)) {
1173                         if ((flp->fld_dst == rt->fl.fld_dst) &&
1174                             (flp->fld_src == rt->fl.fld_src) &&
1175                             (flp->mark == rt->fl.mark) &&
1176                             (rt->fl.iif == 0) &&
1177                             (rt->fl.oif == flp->oif)) {
1178                                 rt->u.dst.lastuse = jiffies;
1179                                 dst_hold(&rt->u.dst);
1180                                 rt->u.dst.__use++;
1181                                 rcu_read_unlock_bh();
1182                                 *pprt = &rt->u.dst;
1183                                 return 0;
1184                         }
1185                 }
1186                 rcu_read_unlock_bh();
1187         }
1188
1189         return dn_route_output_slow(pprt, flp, flags);
1190 }
1191
1192 static int dn_route_output_key(struct dst_entry **pprt, struct flowi *flp, int flags)
1193 {
1194         int err;
1195
1196         err = __dn_route_output_key(pprt, flp, flags);
1197         if (err == 0 && flp->proto) {
1198                 err = xfrm_lookup(pprt, flp, NULL, 0);
1199         }
1200         return err;
1201 }
1202
1203 int dn_route_output_sock(struct dst_entry **pprt, struct flowi *fl, struct sock *sk, int flags)
1204 {
1205         int err;
1206
1207         err = __dn_route_output_key(pprt, fl, flags & MSG_TRYHARD);
1208         if (err == 0 && fl->proto) {
1209                 err = xfrm_lookup(pprt, fl, sk, !(flags & MSG_DONTWAIT));
1210         }
1211         return err;
1212 }
1213
1214 static int dn_route_input_slow(struct sk_buff *skb)
1215 {
1216         struct dn_route *rt = NULL;
1217         struct dn_skb_cb *cb = DN_SKB_CB(skb);
1218         struct net_device *in_dev = skb->dev;
1219         struct net_device *out_dev = NULL;
1220         struct dn_dev *dn_db;
1221         struct neighbour *neigh = NULL;
1222         unsigned hash;
1223         int flags = 0;
1224         __le16 gateway = 0;
1225         __le16 local_src = 0;
1226         struct flowi fl = { .nl_u = { .dn_u =
1227                                      { .daddr = cb->dst,
1228                                        .saddr = cb->src,
1229                                        .scope = RT_SCOPE_UNIVERSE,
1230                                     } },
1231                             .mark = skb->mark,
1232                             .iif = skb->dev->ifindex };
1233         struct dn_fib_res res = { .fi = NULL, .type = RTN_UNREACHABLE };
1234         int err = -EINVAL;
1235         int free_res = 0;
1236
1237         dev_hold(in_dev);
1238
1239         if ((dn_db = in_dev->dn_ptr) == NULL)
1240                 goto out;
1241
1242         /* Zero source addresses are not allowed */
1243         if (fl.fld_src == 0)
1244                 goto out;
1245
1246         /*
1247          * In this case we've just received a packet from a source
1248          * outside ourselves pretending to come from us. We don't
1249          * allow it any further to prevent routing loops, spoofing and
1250          * other nasties. Loopback packets already have the dst attached
1251          * so this only affects packets which have originated elsewhere.
1252          */
1253         err  = -ENOTUNIQ;
1254         if (dn_dev_islocal(in_dev, cb->src))
1255                 goto out;
1256
1257         err = dn_fib_lookup(&fl, &res);
1258         if (err) {
1259                 if (err != -ESRCH)
1260                         goto out;
1261                 /*
1262                  * Is the destination us ?
1263                  */
1264                 if (!dn_dev_islocal(in_dev, cb->dst))
1265                         goto e_inval;
1266
1267                 res.type = RTN_LOCAL;
1268         } else {
1269                 __le16 src_map = fl.fld_src;
1270                 free_res = 1;
1271
1272                 out_dev = DN_FIB_RES_DEV(res);
1273                 if (out_dev == NULL) {
1274                         if (net_ratelimit())
1275                                 printk(KERN_CRIT "Bug in dn_route_input_slow() "
1276                                                  "No output device\n");
1277                         goto e_inval;
1278                 }
1279                 dev_hold(out_dev);
1280
1281                 if (res.r)
1282                         src_map = fl.fld_src; /* no NAT support for now */
1283
1284                 gateway = DN_FIB_RES_GW(res);
1285                 if (res.type == RTN_NAT) {
1286                         fl.fld_dst = dn_fib_rules_map_destination(fl.fld_dst, &res);
1287                         dn_fib_res_put(&res);
1288                         free_res = 0;
1289                         if (dn_fib_lookup(&fl, &res))
1290                                 goto e_inval;
1291                         free_res = 1;
1292                         if (res.type != RTN_UNICAST)
1293                                 goto e_inval;
1294                         flags |= RTCF_DNAT;
1295                         gateway = fl.fld_dst;
1296                 }
1297                 fl.fld_src = src_map;
1298         }
1299
1300         switch(res.type) {
1301         case RTN_UNICAST:
1302                 /*
1303                  * Forwarding check here, we only check for forwarding
1304                  * being turned off, if you want to only forward intra
1305                  * area, its up to you to set the routing tables up
1306                  * correctly.
1307                  */
1308                 if (dn_db->parms.forwarding == 0)
1309                         goto e_inval;
1310
1311                 if (res.fi->fib_nhs > 1 && fl.oif == 0)
1312                         dn_fib_select_multipath(&fl, &res);
1313
1314                 /*
1315                  * Check for out_dev == in_dev. We use the RTCF_DOREDIRECT
1316                  * flag as a hint to set the intra-ethernet bit when
1317                  * forwarding. If we've got NAT in operation, we don't do
1318                  * this optimisation.
1319                  */
1320                 if (out_dev == in_dev && !(flags & RTCF_NAT))
1321                         flags |= RTCF_DOREDIRECT;
1322
1323                 local_src = DN_FIB_RES_PREFSRC(res);
1324
1325         case RTN_BLACKHOLE:
1326         case RTN_UNREACHABLE:
1327                 break;
1328         case RTN_LOCAL:
1329                 flags |= RTCF_LOCAL;
1330                 fl.fld_src = cb->dst;
1331                 fl.fld_dst = cb->src;
1332
1333                 /* Routing tables gave us a gateway */
1334                 if (gateway)
1335                         goto make_route;
1336
1337                 /* Packet was intra-ethernet, so we know its on-link */
1338                 if (cb->rt_flags & DN_RT_F_IE) {
1339                         gateway = cb->src;
1340                         flags |= RTCF_DIRECTSRC;
1341                         goto make_route;
1342                 }
1343
1344                 /* Use the default router if there is one */
1345                 neigh = neigh_clone(dn_db->router);
1346                 if (neigh) {
1347                         gateway = ((struct dn_neigh *)neigh)->addr;
1348                         goto make_route;
1349                 }
1350
1351                 /* Close eyes and pray */
1352                 gateway = cb->src;
1353                 flags |= RTCF_DIRECTSRC;
1354                 goto make_route;
1355         default:
1356                 goto e_inval;
1357         }
1358
1359 make_route:
1360         rt = dst_alloc(&dn_dst_ops);
1361         if (rt == NULL)
1362                 goto e_nobufs;
1363
1364         rt->rt_saddr      = fl.fld_src;
1365         rt->rt_daddr      = fl.fld_dst;
1366         rt->rt_gateway    = fl.fld_dst;
1367         if (gateway)
1368                 rt->rt_gateway = gateway;
1369         rt->rt_local_src  = local_src ? local_src : rt->rt_saddr;
1370
1371         rt->rt_dst_map    = fl.fld_dst;
1372         rt->rt_src_map    = fl.fld_src;
1373
1374         rt->fl.fld_src    = cb->src;
1375         rt->fl.fld_dst    = cb->dst;
1376         rt->fl.oif        = 0;
1377         rt->fl.iif        = in_dev->ifindex;
1378         rt->fl.mark       = fl.mark;
1379
1380         rt->u.dst.flags = DST_HOST;
1381         rt->u.dst.neighbour = neigh;
1382         rt->u.dst.dev = out_dev;
1383         rt->u.dst.lastuse = jiffies;
1384         rt->u.dst.output = dn_rt_bug;
1385         switch(res.type) {
1386                 case RTN_UNICAST:
1387                         rt->u.dst.input = dn_forward;
1388                         break;
1389                 case RTN_LOCAL:
1390                         rt->u.dst.output = dn_output;
1391                         rt->u.dst.input = dn_nsp_rx;
1392                         rt->u.dst.dev = in_dev;
1393                         flags |= RTCF_LOCAL;
1394                         break;
1395                 default:
1396                 case RTN_UNREACHABLE:
1397                 case RTN_BLACKHOLE:
1398                         rt->u.dst.input = dn_blackhole;
1399         }
1400         rt->rt_flags = flags;
1401         if (rt->u.dst.dev)
1402                 dev_hold(rt->u.dst.dev);
1403
1404         err = dn_rt_set_next_hop(rt, &res);
1405         if (err)
1406                 goto e_neighbour;
1407
1408         hash = dn_hash(rt->fl.fld_src, rt->fl.fld_dst);
1409         dn_insert_route(rt, hash, (struct dn_route **)&skb->dst);
1410
1411 done:
1412         if (neigh)
1413                 neigh_release(neigh);
1414         if (free_res)
1415                 dn_fib_res_put(&res);
1416         dev_put(in_dev);
1417         if (out_dev)
1418                 dev_put(out_dev);
1419 out:
1420         return err;
1421
1422 e_inval:
1423         err = -EINVAL;
1424         goto done;
1425
1426 e_nobufs:
1427         err = -ENOBUFS;
1428         goto done;
1429
1430 e_neighbour:
1431         dst_free(&rt->u.dst);
1432         goto done;
1433 }
1434
1435 int dn_route_input(struct sk_buff *skb)
1436 {
1437         struct dn_route *rt;
1438         struct dn_skb_cb *cb = DN_SKB_CB(skb);
1439         unsigned hash = dn_hash(cb->src, cb->dst);
1440
1441         if (skb->dst)
1442                 return 0;
1443
1444         rcu_read_lock();
1445         for(rt = rcu_dereference(dn_rt_hash_table[hash].chain); rt != NULL;
1446             rt = rcu_dereference(rt->u.dst.dn_next)) {
1447                 if ((rt->fl.fld_src == cb->src) &&
1448                     (rt->fl.fld_dst == cb->dst) &&
1449                     (rt->fl.oif == 0) &&
1450                     (rt->fl.mark == skb->mark) &&
1451                     (rt->fl.iif == cb->iif)) {
1452                         rt->u.dst.lastuse = jiffies;
1453                         dst_hold(&rt->u.dst);
1454                         rt->u.dst.__use++;
1455                         rcu_read_unlock();
1456                         skb->dst = (struct dst_entry *)rt;
1457                         return 0;
1458                 }
1459         }
1460         rcu_read_unlock();
1461
1462         return dn_route_input_slow(skb);
1463 }
1464
1465 static int dn_rt_fill_info(struct sk_buff *skb, u32 pid, u32 seq,
1466                            int event, int nowait, unsigned int flags)
1467 {
1468         struct dn_route *rt = (struct dn_route *)skb->dst;
1469         struct rtmsg *r;
1470         struct nlmsghdr *nlh;
1471         unsigned char *b = skb->tail;
1472         long expires;
1473
1474         nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*r), flags);
1475         r = NLMSG_DATA(nlh);
1476         r->rtm_family = AF_DECnet;
1477         r->rtm_dst_len = 16;
1478         r->rtm_src_len = 0;
1479         r->rtm_tos = 0;
1480         r->rtm_table = RT_TABLE_MAIN;
1481         RTA_PUT_U32(skb, RTA_TABLE, RT_TABLE_MAIN);
1482         r->rtm_type = rt->rt_type;
1483         r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
1484         r->rtm_scope = RT_SCOPE_UNIVERSE;
1485         r->rtm_protocol = RTPROT_UNSPEC;
1486         if (rt->rt_flags & RTCF_NOTIFY)
1487                 r->rtm_flags |= RTM_F_NOTIFY;
1488         RTA_PUT(skb, RTA_DST, 2, &rt->rt_daddr);
1489         if (rt->fl.fld_src) {
1490                 r->rtm_src_len = 16;
1491                 RTA_PUT(skb, RTA_SRC, 2, &rt->fl.fld_src);
1492         }
1493         if (rt->u.dst.dev)
1494                 RTA_PUT(skb, RTA_OIF, sizeof(int), &rt->u.dst.dev->ifindex);
1495         /*
1496          * Note to self - change this if input routes reverse direction when
1497          * they deal only with inputs and not with replies like they do
1498          * currently.
1499          */
1500         RTA_PUT(skb, RTA_PREFSRC, 2, &rt->rt_local_src);
1501         if (rt->rt_daddr != rt->rt_gateway)
1502                 RTA_PUT(skb, RTA_GATEWAY, 2, &rt->rt_gateway);
1503         if (rtnetlink_put_metrics(skb, rt->u.dst.metrics) < 0)
1504                 goto rtattr_failure;
1505         expires = rt->u.dst.expires ? rt->u.dst.expires - jiffies : 0;
1506         if (rtnl_put_cacheinfo(skb, &rt->u.dst, 0, 0, 0, expires,
1507                                rt->u.dst.error) < 0)
1508                 goto rtattr_failure;
1509         if (rt->fl.iif)
1510                 RTA_PUT(skb, RTA_IIF, sizeof(int), &rt->fl.iif);
1511
1512         nlh->nlmsg_len = skb->tail - b;
1513         return skb->len;
1514
1515 nlmsg_failure:
1516 rtattr_failure:
1517         skb_trim(skb, b - skb->data);
1518         return -1;
1519 }
1520
1521 /*
1522  * This is called by both endnodes and routers now.
1523  */
1524 int dn_cache_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh, void *arg)
1525 {
1526         struct rtattr **rta = arg;
1527         struct rtmsg *rtm = NLMSG_DATA(nlh);
1528         struct dn_route *rt = NULL;
1529         struct dn_skb_cb *cb;
1530         int err;
1531         struct sk_buff *skb;
1532         struct flowi fl;
1533
1534         memset(&fl, 0, sizeof(fl));
1535         fl.proto = DNPROTO_NSP;
1536
1537         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1538         if (skb == NULL)
1539                 return -ENOBUFS;
1540         skb_reset_mac_header(skb);
1541         cb = DN_SKB_CB(skb);
1542
1543         if (rta[RTA_SRC-1])
1544                 memcpy(&fl.fld_src, RTA_DATA(rta[RTA_SRC-1]), 2);
1545         if (rta[RTA_DST-1])
1546                 memcpy(&fl.fld_dst, RTA_DATA(rta[RTA_DST-1]), 2);
1547         if (rta[RTA_IIF-1])
1548                 memcpy(&fl.iif, RTA_DATA(rta[RTA_IIF-1]), sizeof(int));
1549
1550         if (fl.iif) {
1551                 struct net_device *dev;
1552                 if ((dev = dev_get_by_index(fl.iif)) == NULL) {
1553                         kfree_skb(skb);
1554                         return -ENODEV;
1555                 }
1556                 if (!dev->dn_ptr) {
1557                         dev_put(dev);
1558                         kfree_skb(skb);
1559                         return -ENODEV;
1560                 }
1561                 skb->protocol = __constant_htons(ETH_P_DNA_RT);
1562                 skb->dev = dev;
1563                 cb->src = fl.fld_src;
1564                 cb->dst = fl.fld_dst;
1565                 local_bh_disable();
1566                 err = dn_route_input(skb);
1567                 local_bh_enable();
1568                 memset(cb, 0, sizeof(struct dn_skb_cb));
1569                 rt = (struct dn_route *)skb->dst;
1570                 if (!err && -rt->u.dst.error)
1571                         err = rt->u.dst.error;
1572         } else {
1573                 int oif = 0;
1574                 if (rta[RTA_OIF - 1])
1575                         memcpy(&oif, RTA_DATA(rta[RTA_OIF - 1]), sizeof(int));
1576                 fl.oif = oif;
1577                 err = dn_route_output_key((struct dst_entry **)&rt, &fl, 0);
1578         }
1579
1580         if (skb->dev)
1581                 dev_put(skb->dev);
1582         skb->dev = NULL;
1583         if (err)
1584                 goto out_free;
1585         skb->dst = &rt->u.dst;
1586         if (rtm->rtm_flags & RTM_F_NOTIFY)
1587                 rt->rt_flags |= RTCF_NOTIFY;
1588
1589         err = dn_rt_fill_info(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq, RTM_NEWROUTE, 0, 0);
1590
1591         if (err == 0)
1592                 goto out_free;
1593         if (err < 0) {
1594                 err = -EMSGSIZE;
1595                 goto out_free;
1596         }
1597
1598         return rtnl_unicast(skb, NETLINK_CB(in_skb).pid);
1599
1600 out_free:
1601         kfree_skb(skb);
1602         return err;
1603 }
1604
1605 /*
1606  * For routers, this is called from dn_fib_dump, but for endnodes its
1607  * called directly from the rtnetlink dispatch table.
1608  */
1609 int dn_cache_dump(struct sk_buff *skb, struct netlink_callback *cb)
1610 {
1611         struct dn_route *rt;
1612         int h, s_h;
1613         int idx, s_idx;
1614
1615         if (NLMSG_PAYLOAD(cb->nlh, 0) < sizeof(struct rtmsg))
1616                 return -EINVAL;
1617         if (!(((struct rtmsg *)NLMSG_DATA(cb->nlh))->rtm_flags&RTM_F_CLONED))
1618                 return 0;
1619
1620         s_h = cb->args[0];
1621         s_idx = idx = cb->args[1];
1622         for(h = 0; h <= dn_rt_hash_mask; h++) {
1623                 if (h < s_h)
1624                         continue;
1625                 if (h > s_h)
1626                         s_idx = 0;
1627                 rcu_read_lock_bh();
1628                 for(rt = rcu_dereference(dn_rt_hash_table[h].chain), idx = 0;
1629                         rt;
1630                         rt = rcu_dereference(rt->u.dst.dn_next), idx++) {
1631                         if (idx < s_idx)
1632                                 continue;
1633                         skb->dst = dst_clone(&rt->u.dst);
1634                         if (dn_rt_fill_info(skb, NETLINK_CB(cb->skb).pid,
1635                                         cb->nlh->nlmsg_seq, RTM_NEWROUTE,
1636                                         1, NLM_F_MULTI) <= 0) {
1637                                 dst_release(xchg(&skb->dst, NULL));
1638                                 rcu_read_unlock_bh();
1639                                 goto done;
1640                         }
1641                         dst_release(xchg(&skb->dst, NULL));
1642                 }
1643                 rcu_read_unlock_bh();
1644         }
1645
1646 done:
1647         cb->args[0] = h;
1648         cb->args[1] = idx;
1649         return skb->len;
1650 }
1651
1652 #ifdef CONFIG_PROC_FS
1653 struct dn_rt_cache_iter_state {
1654         int bucket;
1655 };
1656
1657 static struct dn_route *dn_rt_cache_get_first(struct seq_file *seq)
1658 {
1659         struct dn_route *rt = NULL;
1660         struct dn_rt_cache_iter_state *s = seq->private;
1661
1662         for(s->bucket = dn_rt_hash_mask; s->bucket >= 0; --s->bucket) {
1663                 rcu_read_lock_bh();
1664                 rt = dn_rt_hash_table[s->bucket].chain;
1665                 if (rt)
1666                         break;
1667                 rcu_read_unlock_bh();
1668         }
1669         return rt;
1670 }
1671
1672 static struct dn_route *dn_rt_cache_get_next(struct seq_file *seq, struct dn_route *rt)
1673 {
1674         struct dn_rt_cache_iter_state *s = rcu_dereference(seq->private);
1675
1676         rt = rt->u.dst.dn_next;
1677         while(!rt) {
1678                 rcu_read_unlock_bh();
1679                 if (--s->bucket < 0)
1680                         break;
1681                 rcu_read_lock_bh();
1682                 rt = dn_rt_hash_table[s->bucket].chain;
1683         }
1684         return rt;
1685 }
1686
1687 static void *dn_rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
1688 {
1689         struct dn_route *rt = dn_rt_cache_get_first(seq);
1690
1691         if (rt) {
1692                 while(*pos && (rt = dn_rt_cache_get_next(seq, rt)))
1693                         --*pos;
1694         }
1695         return *pos ? NULL : rt;
1696 }
1697
1698 static void *dn_rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1699 {
1700         struct dn_route *rt = dn_rt_cache_get_next(seq, v);
1701         ++*pos;
1702         return rt;
1703 }
1704
1705 static void dn_rt_cache_seq_stop(struct seq_file *seq, void *v)
1706 {
1707         if (v)
1708                 rcu_read_unlock_bh();
1709 }
1710
1711 static int dn_rt_cache_seq_show(struct seq_file *seq, void *v)
1712 {
1713         struct dn_route *rt = v;
1714         char buf1[DN_ASCBUF_LEN], buf2[DN_ASCBUF_LEN];
1715
1716         seq_printf(seq, "%-8s %-7s %-7s %04d %04d %04d\n",
1717                         rt->u.dst.dev ? rt->u.dst.dev->name : "*",
1718                         dn_addr2asc(dn_ntohs(rt->rt_daddr), buf1),
1719                         dn_addr2asc(dn_ntohs(rt->rt_saddr), buf2),
1720                         atomic_read(&rt->u.dst.__refcnt),
1721                         rt->u.dst.__use,
1722                         (int) dst_metric(&rt->u.dst, RTAX_RTT));
1723         return 0;
1724 }
1725
1726 static struct seq_operations dn_rt_cache_seq_ops = {
1727         .start  = dn_rt_cache_seq_start,
1728         .next   = dn_rt_cache_seq_next,
1729         .stop   = dn_rt_cache_seq_stop,
1730         .show   = dn_rt_cache_seq_show,
1731 };
1732
1733 static int dn_rt_cache_seq_open(struct inode *inode, struct file *file)
1734 {
1735         struct seq_file *seq;
1736         int rc = -ENOMEM;
1737         struct dn_rt_cache_iter_state *s = kmalloc(sizeof(*s), GFP_KERNEL);
1738
1739         if (!s)
1740                 goto out;
1741         rc = seq_open(file, &dn_rt_cache_seq_ops);
1742         if (rc)
1743                 goto out_kfree;
1744         seq             = file->private_data;
1745         seq->private    = s;
1746         memset(s, 0, sizeof(*s));
1747 out:
1748         return rc;
1749 out_kfree:
1750         kfree(s);
1751         goto out;
1752 }
1753
1754 static const struct file_operations dn_rt_cache_seq_fops = {
1755         .owner   = THIS_MODULE,
1756         .open    = dn_rt_cache_seq_open,
1757         .read    = seq_read,
1758         .llseek  = seq_lseek,
1759         .release = seq_release_private,
1760 };
1761
1762 #endif /* CONFIG_PROC_FS */
1763
1764 void __init dn_route_init(void)
1765 {
1766         int i, goal, order;
1767
1768         dn_dst_ops.kmem_cachep =
1769                 kmem_cache_create("dn_dst_cache", sizeof(struct dn_route), 0,
1770                                   SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1771         init_timer(&dn_route_timer);
1772         dn_route_timer.function = dn_dst_check_expire;
1773         dn_route_timer.expires = jiffies + decnet_dst_gc_interval * HZ;
1774         add_timer(&dn_route_timer);
1775
1776         goal = num_physpages >> (26 - PAGE_SHIFT);
1777
1778         for(order = 0; (1UL << order) < goal; order++)
1779                 /* NOTHING */;
1780
1781         /*
1782          * Only want 1024 entries max, since the table is very, very unlikely
1783          * to be larger than that.
1784          */
1785         while(order && ((((1UL << order) * PAGE_SIZE) /
1786                                 sizeof(struct dn_rt_hash_bucket)) >= 2048))
1787                 order--;
1788
1789         do {
1790                 dn_rt_hash_mask = (1UL << order) * PAGE_SIZE /
1791                         sizeof(struct dn_rt_hash_bucket);
1792                 while(dn_rt_hash_mask & (dn_rt_hash_mask - 1))
1793                         dn_rt_hash_mask--;
1794                 dn_rt_hash_table = (struct dn_rt_hash_bucket *)
1795                         __get_free_pages(GFP_ATOMIC, order);
1796         } while (dn_rt_hash_table == NULL && --order > 0);
1797
1798         if (!dn_rt_hash_table)
1799                 panic("Failed to allocate DECnet route cache hash table\n");
1800
1801         printk(KERN_INFO
1802                 "DECnet: Routing cache hash table of %u buckets, %ldKbytes\n",
1803                 dn_rt_hash_mask,
1804                 (long)(dn_rt_hash_mask*sizeof(struct dn_rt_hash_bucket))/1024);
1805
1806         dn_rt_hash_mask--;
1807         for(i = 0; i <= dn_rt_hash_mask; i++) {
1808                 spin_lock_init(&dn_rt_hash_table[i].lock);
1809                 dn_rt_hash_table[i].chain = NULL;
1810         }
1811
1812         dn_dst_ops.gc_thresh = (dn_rt_hash_mask + 1);
1813
1814         proc_net_fops_create("decnet_cache", S_IRUGO, &dn_rt_cache_seq_fops);
1815 }
1816
1817 void __exit dn_route_cleanup(void)
1818 {
1819         del_timer(&dn_route_timer);
1820         dn_run_flush(0);
1821
1822         proc_net_remove("decnet_cache");
1823 }
1824