Linux-2.6.12-rc2
[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/config.h>
59 #include <linux/errno.h>
60 #include <linux/types.h>
61 #include <linux/socket.h>
62 #include <linux/in.h>
63 #include <linux/kernel.h>
64 #include <linux/sockios.h>
65 #include <linux/net.h>
66 #include <linux/netdevice.h>
67 #include <linux/inet.h>
68 #include <linux/route.h>
69 #include <linux/in_route.h>
70 #include <net/sock.h>
71 #include <linux/mm.h>
72 #include <linux/proc_fs.h>
73 #include <linux/seq_file.h>
74 #include <linux/init.h>
75 #include <linux/rtnetlink.h>
76 #include <linux/string.h>
77 #include <linux/netfilter_decnet.h>
78 #include <linux/rcupdate.h>
79 #include <linux/times.h>
80 #include <asm/errno.h>
81 #include <net/neighbour.h>
82 #include <net/dst.h>
83 #include <net/flow.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 struct timer_list dn_rt_flush_timer =
121                 TIMER_INITIALIZER(dn_run_flush, 0, 0);
122 int decnet_dst_gc_interval = 2;
123
124 static struct dst_ops dn_dst_ops = {
125         .family =               PF_DECnet,
126         .protocol =             __constant_htons(ETH_P_DNA_RT),
127         .gc_thresh =            128,
128         .gc =                   dn_dst_gc,
129         .check =                dn_dst_check,
130         .negative_advice =      dn_dst_negative_advice,
131         .link_failure =         dn_dst_link_failure,
132         .update_pmtu =          dn_dst_update_pmtu,
133         .entry_size =           sizeof(struct dn_route),
134         .entries =              ATOMIC_INIT(0),
135 };
136
137 static __inline__ unsigned dn_hash(unsigned short src, unsigned short dst)
138 {
139         unsigned short tmp = src ^ dst;
140         tmp ^= (tmp >> 3);
141         tmp ^= (tmp >> 5);
142         tmp ^= (tmp >> 10);
143         return dn_rt_hash_mask & (unsigned)tmp;
144 }
145
146 static inline void dnrt_free(struct dn_route *rt)
147 {
148         call_rcu_bh(&rt->u.dst.rcu_head, dst_rcu_free);
149 }
150
151 static inline void dnrt_drop(struct dn_route *rt)
152 {
153         if (rt)
154                 dst_release(&rt->u.dst);
155         call_rcu_bh(&rt->u.dst.rcu_head, dst_rcu_free);
156 }
157
158 static void dn_dst_check_expire(unsigned long dummy)
159 {
160         int i;
161         struct dn_route *rt, **rtp;
162         unsigned long now = jiffies;
163         unsigned long expire = 120 * HZ;
164
165         for(i = 0; i <= dn_rt_hash_mask; i++) {
166                 rtp = &dn_rt_hash_table[i].chain;
167
168                 spin_lock(&dn_rt_hash_table[i].lock);
169                 while((rt=*rtp) != NULL) {
170                         if (atomic_read(&rt->u.dst.__refcnt) ||
171                                         (now - rt->u.dst.lastuse) < expire) {
172                                 rtp = &rt->u.rt_next;
173                                 continue;
174                         }
175                         *rtp = rt->u.rt_next;
176                         rt->u.rt_next = NULL;
177                         dnrt_free(rt);
178                 }
179                 spin_unlock(&dn_rt_hash_table[i].lock);
180
181                 if ((jiffies - now) > 0)
182                         break;
183         }
184
185         mod_timer(&dn_route_timer, now + decnet_dst_gc_interval * HZ);
186 }
187
188 static int dn_dst_gc(void)
189 {
190         struct dn_route *rt, **rtp;
191         int i;
192         unsigned long now = jiffies;
193         unsigned long expire = 10 * HZ;
194
195         for(i = 0; i <= dn_rt_hash_mask; i++) {
196
197                 spin_lock_bh(&dn_rt_hash_table[i].lock);
198                 rtp = &dn_rt_hash_table[i].chain;
199
200                 while((rt=*rtp) != NULL) {
201                         if (atomic_read(&rt->u.dst.__refcnt) ||
202                                         (now - rt->u.dst.lastuse) < expire) {
203                                 rtp = &rt->u.rt_next;
204                                 continue;
205                         }
206                         *rtp = rt->u.rt_next;
207                         rt->u.rt_next = NULL;
208                         dnrt_drop(rt);
209                         break;
210                 }
211                 spin_unlock_bh(&dn_rt_hash_table[i].lock);
212         }
213
214         return 0;
215 }
216
217 /*
218  * The decnet standards don't impose a particular minimum mtu, what they
219  * do insist on is that the routing layer accepts a datagram of at least
220  * 230 bytes long. Here we have to subtract the routing header length from
221  * 230 to get the minimum acceptable mtu. If there is no neighbour, then we
222  * assume the worst and use a long header size.
223  *
224  * We update both the mtu and the advertised mss (i.e. the segment size we
225  * advertise to the other end).
226  */
227 static void dn_dst_update_pmtu(struct dst_entry *dst, u32 mtu)
228 {
229         u32 min_mtu = 230;
230         struct dn_dev *dn = dst->neighbour ?
231                             (struct dn_dev *)dst->neighbour->dev->dn_ptr : NULL;
232
233         if (dn && dn->use_long == 0)
234                 min_mtu -= 6;
235         else
236                 min_mtu -= 21;
237
238         if (dst->metrics[RTAX_MTU-1] > mtu && mtu >= min_mtu) {
239                 if (!(dst_metric_locked(dst, RTAX_MTU))) {
240                         dst->metrics[RTAX_MTU-1] = mtu;
241                         dst_set_expires(dst, dn_rt_mtu_expires);
242                 }
243                 if (!(dst_metric_locked(dst, RTAX_ADVMSS))) {
244                         u32 mss = mtu - DN_MAX_NSP_DATA_HEADER;
245                         if (dst->metrics[RTAX_ADVMSS-1] > mss)
246                                 dst->metrics[RTAX_ADVMSS-1] = mss;
247                 }
248         }
249 }
250
251 /* 
252  * When a route has been marked obsolete. (e.g. routing cache flush)
253  */
254 static struct dst_entry *dn_dst_check(struct dst_entry *dst, __u32 cookie)
255 {
256         return NULL;
257 }
258
259 static struct dst_entry *dn_dst_negative_advice(struct dst_entry *dst)
260 {
261         dst_release(dst);
262         return NULL;
263 }
264
265 static void dn_dst_link_failure(struct sk_buff *skb)
266 {
267         return;
268 }
269
270 static inline int compare_keys(struct flowi *fl1, struct flowi *fl2)
271 {
272         return memcmp(&fl1->nl_u.dn_u, &fl2->nl_u.dn_u, sizeof(fl1->nl_u.dn_u)) == 0 &&
273                 fl1->oif == fl2->oif &&
274                 fl1->iif == fl2->iif;
275 }
276
277 static int dn_insert_route(struct dn_route *rt, unsigned hash, struct dn_route **rp)
278 {
279         struct dn_route *rth, **rthp;
280         unsigned long now = jiffies;
281
282         rthp = &dn_rt_hash_table[hash].chain;
283
284         spin_lock_bh(&dn_rt_hash_table[hash].lock);
285         while((rth = *rthp) != NULL) {
286                 if (compare_keys(&rth->fl, &rt->fl)) {
287                         /* Put it first */
288                         *rthp = rth->u.rt_next;
289                         rcu_assign_pointer(rth->u.rt_next,
290                                            dn_rt_hash_table[hash].chain);
291                         rcu_assign_pointer(dn_rt_hash_table[hash].chain, rth);
292
293                         rth->u.dst.__use++;
294                         dst_hold(&rth->u.dst);
295                         rth->u.dst.lastuse = now;
296                         spin_unlock_bh(&dn_rt_hash_table[hash].lock);
297
298                         dnrt_drop(rt);
299                         *rp = rth;
300                         return 0;
301                 }
302                 rthp = &rth->u.rt_next;
303         }
304
305         rcu_assign_pointer(rt->u.rt_next, dn_rt_hash_table[hash].chain);
306         rcu_assign_pointer(dn_rt_hash_table[hash].chain, rt);
307         
308         dst_hold(&rt->u.dst);
309         rt->u.dst.__use++;
310         rt->u.dst.lastuse = now;
311         spin_unlock_bh(&dn_rt_hash_table[hash].lock);
312         *rp = rt;
313         return 0;
314 }
315
316 void dn_run_flush(unsigned long dummy)
317 {
318         int i;
319         struct dn_route *rt, *next;
320
321         for(i = 0; i < dn_rt_hash_mask; i++) {
322                 spin_lock_bh(&dn_rt_hash_table[i].lock);
323
324                 if ((rt = xchg(&dn_rt_hash_table[i].chain, NULL)) == NULL)
325                         goto nothing_to_declare;
326
327                 for(; rt; rt=next) {
328                         next = rt->u.rt_next;
329                         rt->u.rt_next = NULL;
330                         dst_free((struct dst_entry *)rt);
331                 }
332
333 nothing_to_declare:
334                 spin_unlock_bh(&dn_rt_hash_table[i].lock);
335         }
336 }
337
338 static DEFINE_SPINLOCK(dn_rt_flush_lock);
339
340 void dn_rt_cache_flush(int delay)
341 {
342         unsigned long now = jiffies;
343         int user_mode = !in_interrupt();
344
345         if (delay < 0)
346                 delay = dn_rt_min_delay;
347
348         spin_lock_bh(&dn_rt_flush_lock);
349
350         if (del_timer(&dn_rt_flush_timer) && delay > 0 && dn_rt_deadline) {
351                 long tmo = (long)(dn_rt_deadline - now);
352
353                 if (user_mode && tmo < dn_rt_max_delay - dn_rt_min_delay)
354                         tmo = 0;
355
356                 if (delay > tmo)
357                         delay = tmo;
358         }
359
360         if (delay <= 0) {
361                 spin_unlock_bh(&dn_rt_flush_lock);
362                 dn_run_flush(0);
363                 return;
364         }
365
366         if (dn_rt_deadline == 0)
367                 dn_rt_deadline = now + dn_rt_max_delay;
368
369         dn_rt_flush_timer.expires = now + delay;
370         add_timer(&dn_rt_flush_timer);
371         spin_unlock_bh(&dn_rt_flush_lock);
372 }
373
374 /**
375  * dn_return_short - Return a short packet to its sender
376  * @skb: The packet to return
377  *
378  */
379 static int dn_return_short(struct sk_buff *skb)
380 {
381         struct dn_skb_cb *cb;
382         unsigned char *ptr;
383         dn_address *src;
384         dn_address *dst;
385         dn_address tmp;
386
387         /* Add back headers */
388         skb_push(skb, skb->data - skb->nh.raw);
389
390         if ((skb = skb_unshare(skb, GFP_ATOMIC)) == NULL)
391                 return NET_RX_DROP;
392
393         cb = DN_SKB_CB(skb);
394         /* Skip packet length and point to flags */
395         ptr = skb->data + 2;
396         *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
397
398         dst = (dn_address *)ptr;
399         ptr += 2;
400         src = (dn_address *)ptr;
401         ptr += 2;
402         *ptr = 0; /* Zero hop count */
403
404         /* Swap source and destination */
405         tmp  = *src;
406         *src = *dst;
407         *dst = tmp;
408
409         skb->pkt_type = PACKET_OUTGOING;
410         dn_rt_finish_output(skb, NULL, NULL);
411         return NET_RX_SUCCESS;
412 }
413
414 /**
415  * dn_return_long - Return a long packet to its sender
416  * @skb: The long format packet to return
417  *
418  */
419 static int dn_return_long(struct sk_buff *skb)
420 {
421         struct dn_skb_cb *cb;
422         unsigned char *ptr;
423         unsigned char *src_addr, *dst_addr;
424         unsigned char tmp[ETH_ALEN];
425
426         /* Add back all headers */
427         skb_push(skb, skb->data - skb->nh.raw);
428
429         if ((skb = skb_unshare(skb, GFP_ATOMIC)) == NULL)
430                 return NET_RX_DROP;
431
432         cb = DN_SKB_CB(skb);
433         /* Ignore packet length and point to flags */
434         ptr = skb->data + 2;
435
436         /* Skip padding */
437         if (*ptr & DN_RT_F_PF) {
438                 char padlen = (*ptr & ~DN_RT_F_PF);
439                 ptr += padlen;
440         }
441
442         *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
443         ptr += 2;
444         dst_addr = ptr;
445         ptr += 8;
446         src_addr = ptr;
447         ptr += 6;
448         *ptr = 0; /* Zero hop count */
449
450         /* Swap source and destination */
451         memcpy(tmp, src_addr, ETH_ALEN);
452         memcpy(src_addr, dst_addr, ETH_ALEN);
453         memcpy(dst_addr, tmp, ETH_ALEN);
454
455         skb->pkt_type = PACKET_OUTGOING;
456         dn_rt_finish_output(skb, dst_addr, src_addr);
457         return NET_RX_SUCCESS;
458 }
459
460 /**
461  * dn_route_rx_packet - Try and find a route for an incoming packet
462  * @skb: The packet to find a route for
463  *
464  * Returns: result of input function if route is found, error code otherwise
465  */
466 static int dn_route_rx_packet(struct sk_buff *skb)
467 {
468         struct dn_skb_cb *cb = DN_SKB_CB(skb);
469         int err;
470
471         if ((err = dn_route_input(skb)) == 0)
472                 return dst_input(skb);
473
474         if (decnet_debug_level & 4) {
475                 char *devname = skb->dev ? skb->dev->name : "???";
476                 struct dn_skb_cb *cb = DN_SKB_CB(skb);
477                 printk(KERN_DEBUG
478                         "DECnet: dn_route_rx_packet: rt_flags=0x%02x dev=%s len=%d src=0x%04hx dst=0x%04hx err=%d type=%d\n",
479                         (int)cb->rt_flags, devname, skb->len, cb->src, cb->dst, 
480                         err, skb->pkt_type);
481         }
482
483         if ((skb->pkt_type == PACKET_HOST) && (cb->rt_flags & DN_RT_F_RQR)) {
484                 switch(cb->rt_flags & DN_RT_PKT_MSK) {
485                         case DN_RT_PKT_SHORT:
486                                 return dn_return_short(skb);
487                         case DN_RT_PKT_LONG:
488                                 return dn_return_long(skb);
489                 }
490         }
491
492         kfree_skb(skb);
493         return NET_RX_DROP;
494 }
495
496 static int dn_route_rx_long(struct sk_buff *skb)
497 {
498         struct dn_skb_cb *cb = DN_SKB_CB(skb);
499         unsigned char *ptr = skb->data;
500
501         if (!pskb_may_pull(skb, 21)) /* 20 for long header, 1 for shortest nsp */
502                 goto drop_it;
503
504         skb_pull(skb, 20);
505         skb->h.raw = skb->data;
506
507         /* Destination info */
508         ptr += 2;
509         cb->dst = dn_htons(dn_eth2dn(ptr));
510         if (memcmp(ptr, dn_hiord_addr, 4) != 0)
511                 goto drop_it;
512         ptr += 6;
513
514
515         /* Source info */
516         ptr += 2;
517         cb->src = dn_htons(dn_eth2dn(ptr));
518         if (memcmp(ptr, dn_hiord_addr, 4) != 0)
519                 goto drop_it;
520         ptr += 6;
521         /* Other junk */
522         ptr++;
523         cb->hops = *ptr++; /* Visit Count */
524
525         return NF_HOOK(PF_DECnet, NF_DN_PRE_ROUTING, skb, skb->dev, NULL, dn_route_rx_packet);
526
527 drop_it:
528         kfree_skb(skb);
529         return NET_RX_DROP;
530 }
531
532
533
534 static int dn_route_rx_short(struct sk_buff *skb)
535 {
536         struct dn_skb_cb *cb = DN_SKB_CB(skb);
537         unsigned char *ptr = skb->data;
538
539         if (!pskb_may_pull(skb, 6)) /* 5 for short header + 1 for shortest nsp */
540                 goto drop_it;
541
542         skb_pull(skb, 5);
543         skb->h.raw = skb->data;
544
545         cb->dst = *(dn_address *)ptr;
546         ptr += 2;
547         cb->src = *(dn_address *)ptr;
548         ptr += 2;
549         cb->hops = *ptr & 0x3f;
550
551         return NF_HOOK(PF_DECnet, NF_DN_PRE_ROUTING, skb, skb->dev, NULL, dn_route_rx_packet);
552
553 drop_it:
554         kfree_skb(skb);
555         return NET_RX_DROP;
556 }
557
558 static int dn_route_discard(struct sk_buff *skb)
559 {
560         /*
561          * I know we drop the packet here, but thats considered success in
562          * this case
563          */
564         kfree_skb(skb);
565         return NET_RX_SUCCESS;
566 }
567
568 static int dn_route_ptp_hello(struct sk_buff *skb)
569 {
570         dn_dev_hello(skb);
571         dn_neigh_pointopoint_hello(skb);
572         return NET_RX_SUCCESS;
573 }
574
575 int dn_route_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt)
576 {
577         struct dn_skb_cb *cb;
578         unsigned char flags = 0;
579         __u16 len = dn_ntohs(*(__u16 *)skb->data);
580         struct dn_dev *dn = (struct dn_dev *)dev->dn_ptr;
581         unsigned char padlen = 0;
582
583         if (dn == NULL)
584                 goto dump_it;
585
586         if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
587                 goto out;
588
589         if (!pskb_may_pull(skb, 3))
590                 goto dump_it;
591
592         skb_pull(skb, 2);
593
594         if (len > skb->len)
595                 goto dump_it;
596
597         skb_trim(skb, len);
598
599         flags = *skb->data;
600
601         cb = DN_SKB_CB(skb);
602         cb->stamp = jiffies;
603         cb->iif = dev->ifindex;
604
605         /*
606          * If we have padding, remove it.
607          */
608         if (flags & DN_RT_F_PF) {
609                 padlen = flags & ~DN_RT_F_PF;
610                 if (!pskb_may_pull(skb, padlen + 1))
611                         goto dump_it;
612                 skb_pull(skb, padlen);
613                 flags = *skb->data;
614         }
615
616         skb->nh.raw = skb->data;
617
618         /*
619          * Weed out future version DECnet
620          */
621         if (flags & DN_RT_F_VER)
622                 goto dump_it;
623
624         cb->rt_flags = flags;
625
626         if (decnet_debug_level & 1)
627                 printk(KERN_DEBUG 
628                         "dn_route_rcv: got 0x%02x from %s [%d %d %d]\n",
629                         (int)flags, (dev) ? dev->name : "???", len, skb->len, 
630                         padlen);
631
632         if (flags & DN_RT_PKT_CNTL) {
633                 if (unlikely(skb_is_nonlinear(skb)) &&
634                     skb_linearize(skb, GFP_ATOMIC) != 0)
635                         goto dump_it;
636
637                 switch(flags & DN_RT_CNTL_MSK) {
638                         case DN_RT_PKT_INIT:
639                                 dn_dev_init_pkt(skb);
640                                 break;
641                         case DN_RT_PKT_VERI:
642                                 dn_dev_veri_pkt(skb);
643                                 break;
644                 }
645
646                 if (dn->parms.state != DN_DEV_S_RU)
647                         goto dump_it;
648
649                 switch(flags & DN_RT_CNTL_MSK) {
650                         case DN_RT_PKT_HELO:
651                                 return NF_HOOK(PF_DECnet, NF_DN_HELLO, skb, skb->dev, NULL, dn_route_ptp_hello);
652
653                         case DN_RT_PKT_L1RT:
654                         case DN_RT_PKT_L2RT:
655                                 return NF_HOOK(PF_DECnet, NF_DN_ROUTE, skb, skb->dev, NULL, dn_route_discard);
656                         case DN_RT_PKT_ERTH:
657                                 return NF_HOOK(PF_DECnet, NF_DN_HELLO, skb, skb->dev, NULL, dn_neigh_router_hello);
658
659                         case DN_RT_PKT_EEDH:
660                                 return NF_HOOK(PF_DECnet, NF_DN_HELLO, skb, skb->dev, NULL, dn_neigh_endnode_hello);
661                 }
662         } else {
663                 if (dn->parms.state != DN_DEV_S_RU)
664                         goto dump_it;
665
666                 skb_pull(skb, 1); /* Pull flags */
667
668                 switch(flags & DN_RT_PKT_MSK) {
669                         case DN_RT_PKT_LONG:
670                                 return dn_route_rx_long(skb);
671                         case DN_RT_PKT_SHORT:
672                                 return dn_route_rx_short(skb);
673                 }
674         }
675
676 dump_it:
677         kfree_skb(skb);
678 out:
679         return NET_RX_DROP;
680 }
681
682 static int dn_output(struct sk_buff *skb)
683 {
684         struct dst_entry *dst = skb->dst;
685         struct dn_route *rt = (struct dn_route *)dst;
686         struct net_device *dev = dst->dev;
687         struct dn_skb_cb *cb = DN_SKB_CB(skb);
688         struct neighbour *neigh;
689
690         int err = -EINVAL;
691
692         if ((neigh = dst->neighbour) == NULL)
693                 goto error;
694
695         skb->dev = dev;
696
697         cb->src = rt->rt_saddr;
698         cb->dst = rt->rt_daddr;
699
700         /*
701          * Always set the Intra-Ethernet bit on all outgoing packets
702          * originated on this node. Only valid flag from upper layers
703          * is return-to-sender-requested. Set hop count to 0 too.
704          */
705         cb->rt_flags &= ~DN_RT_F_RQR;
706         cb->rt_flags |= DN_RT_F_IE;
707         cb->hops = 0;
708
709         return NF_HOOK(PF_DECnet, NF_DN_LOCAL_OUT, skb, NULL, dev, neigh->output);
710
711 error:
712         if (net_ratelimit())
713                 printk(KERN_DEBUG "dn_output: This should not happen\n");
714
715         kfree_skb(skb);
716
717         return err;
718 }
719
720 static int dn_forward(struct sk_buff *skb)
721 {
722         struct dn_skb_cb *cb = DN_SKB_CB(skb);
723         struct dst_entry *dst = skb->dst;
724         struct dn_dev *dn_db = dst->dev->dn_ptr;
725         struct dn_route *rt;
726         struct neighbour *neigh = dst->neighbour;
727         int header_len;
728 #ifdef CONFIG_NETFILTER
729         struct net_device *dev = skb->dev;
730 #endif
731
732         if (skb->pkt_type != PACKET_HOST)
733                 goto drop;
734
735         /* Ensure that we have enough space for headers */
736         rt = (struct dn_route *)skb->dst;
737         header_len = dn_db->use_long ? 21 : 6;
738         if (skb_cow(skb, LL_RESERVED_SPACE(rt->u.dst.dev)+header_len))
739                 goto drop;
740
741         /*
742          * Hop count exceeded.
743          */
744         if (++cb->hops > 30)
745                 goto drop;
746
747         skb->dev = rt->u.dst.dev;
748
749         /*
750          * If packet goes out same interface it came in on, then set
751          * the Intra-Ethernet bit. This has no effect for short
752          * packets, so we don't need to test for them here.
753          */
754         cb->rt_flags &= ~DN_RT_F_IE;
755         if (rt->rt_flags & RTCF_DOREDIRECT)
756                 cb->rt_flags |= DN_RT_F_IE;
757
758         return NF_HOOK(PF_DECnet, NF_DN_FORWARD, skb, dev, skb->dev, neigh->output);
759
760 drop:
761         kfree_skb(skb);
762         return NET_RX_DROP;
763 }
764
765 /*
766  * Drop packet. This is used for endnodes and for
767  * when we should not be forwarding packets from
768  * this dest.
769  */
770 static int dn_blackhole(struct sk_buff *skb)
771 {
772         kfree_skb(skb);
773         return NET_RX_DROP;
774 }
775
776 /*
777  * Used to catch bugs. This should never normally get
778  * called.
779  */
780 static int dn_rt_bug(struct sk_buff *skb)
781 {
782         if (net_ratelimit()) {
783                 struct dn_skb_cb *cb = DN_SKB_CB(skb);
784
785                 printk(KERN_DEBUG "dn_rt_bug: skb from:%04x to:%04x\n",
786                                 cb->src, cb->dst);
787         }
788
789         kfree_skb(skb);
790
791         return NET_RX_BAD;
792 }
793
794 static int dn_rt_set_next_hop(struct dn_route *rt, struct dn_fib_res *res)
795 {
796         struct dn_fib_info *fi = res->fi;
797         struct net_device *dev = rt->u.dst.dev;
798         struct neighbour *n;
799         unsigned mss;
800
801         if (fi) {
802                 if (DN_FIB_RES_GW(*res) &&
803                     DN_FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
804                         rt->rt_gateway = DN_FIB_RES_GW(*res);
805                 memcpy(rt->u.dst.metrics, fi->fib_metrics,
806                        sizeof(rt->u.dst.metrics));
807         }
808         rt->rt_type = res->type;
809
810         if (dev != NULL && rt->u.dst.neighbour == NULL) {
811                 n = __neigh_lookup_errno(&dn_neigh_table, &rt->rt_gateway, dev);
812                 if (IS_ERR(n))
813                         return PTR_ERR(n);
814                 rt->u.dst.neighbour = n;
815         }
816
817         if (rt->u.dst.metrics[RTAX_MTU-1] == 0 || 
818             rt->u.dst.metrics[RTAX_MTU-1] > rt->u.dst.dev->mtu)
819                 rt->u.dst.metrics[RTAX_MTU-1] = rt->u.dst.dev->mtu;
820         mss = dn_mss_from_pmtu(dev, dst_mtu(&rt->u.dst));
821         if (rt->u.dst.metrics[RTAX_ADVMSS-1] == 0 ||
822             rt->u.dst.metrics[RTAX_ADVMSS-1] > mss)
823                 rt->u.dst.metrics[RTAX_ADVMSS-1] = mss;
824         return 0;
825 }
826
827 static inline int dn_match_addr(__u16 addr1, __u16 addr2)
828 {
829         __u16 tmp = dn_ntohs(addr1) ^ dn_ntohs(addr2);
830         int match = 16;
831         while(tmp) {
832                 tmp >>= 1;
833                 match--;
834         }
835         return match;
836 }
837
838 static __u16 dnet_select_source(const struct net_device *dev, __u16 daddr, int scope)
839 {
840         __u16 saddr = 0;
841         struct dn_dev *dn_db = dev->dn_ptr;
842         struct dn_ifaddr *ifa;
843         int best_match = 0;
844         int ret;
845
846         read_lock(&dev_base_lock);
847         for(ifa = dn_db->ifa_list; ifa; ifa = ifa->ifa_next) {
848                 if (ifa->ifa_scope > scope)
849                         continue;
850                 if (!daddr) {
851                         saddr = ifa->ifa_local;
852                         break;
853                 }
854                 ret = dn_match_addr(daddr, ifa->ifa_local);
855                 if (ret > best_match)
856                         saddr = ifa->ifa_local;
857                 if (best_match == 0)
858                         saddr = ifa->ifa_local;
859         }
860         read_unlock(&dev_base_lock);
861
862         return saddr;
863 }
864
865 static inline __u16 __dn_fib_res_prefsrc(struct dn_fib_res *res)
866 {
867         return dnet_select_source(DN_FIB_RES_DEV(*res), DN_FIB_RES_GW(*res), res->scope);
868 }
869
870 static inline __u16 dn_fib_rules_map_destination(__u16 daddr, struct dn_fib_res *res)
871 {
872         __u16 mask = dnet_make_mask(res->prefixlen);
873         return (daddr&~mask)|res->fi->fib_nh->nh_gw;
874 }
875
876 static int dn_route_output_slow(struct dst_entry **pprt, const struct flowi *oldflp, int try_hard)
877 {
878         struct flowi fl = { .nl_u = { .dn_u = 
879                                       { .daddr = oldflp->fld_dst,
880                                         .saddr = oldflp->fld_src,
881                                         .scope = RT_SCOPE_UNIVERSE,
882 #ifdef CONFIG_DECNET_ROUTE_FWMARK
883                                         .fwmark = oldflp->fld_fwmark
884 #endif
885                                      } },
886                             .iif = loopback_dev.ifindex,
887                             .oif = oldflp->oif };
888         struct dn_route *rt = NULL;
889         struct net_device *dev_out = NULL;
890         struct neighbour *neigh = NULL;
891         unsigned hash;
892         unsigned flags = 0;
893         struct dn_fib_res res = { .fi = NULL, .type = RTN_UNICAST };
894         int err;
895         int free_res = 0;
896         __u16 gateway = 0;
897
898         if (decnet_debug_level & 16)
899                 printk(KERN_DEBUG
900                        "dn_route_output_slow: dst=%04x src=%04x mark=%d"
901                        " iif=%d oif=%d\n", oldflp->fld_dst, oldflp->fld_src,
902                        oldflp->fld_fwmark, 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                                 break;
932                 }
933                 read_unlock(&dev_base_lock);
934                 if (dev_out == NULL)
935                         goto out;
936                 dev_hold(dev_out);
937 source_ok:
938                 ;
939         }
940
941         /* No destination? Assume its local */
942         if (!fl.fld_dst) {
943                 fl.fld_dst = fl.fld_src;
944
945                 err = -EADDRNOTAVAIL;
946                 if (dev_out)
947                         dev_put(dev_out);
948                 dev_out = &loopback_dev;
949                 dev_hold(dev_out);
950                 if (!fl.fld_dst) {
951                         fl.fld_dst =
952                         fl.fld_src = dnet_select_source(dev_out, 0,
953                                                        RT_SCOPE_HOST);
954                         if (!fl.fld_dst)
955                                 goto out;
956                 }
957                 fl.oif = loopback_dev.ifindex;
958                 res.type = RTN_LOCAL;
959                 goto make_route;
960         }
961
962         if (decnet_debug_level & 16)
963                 printk(KERN_DEBUG
964                        "dn_route_output_slow: initial checks complete."
965                        " dst=%o4x src=%04x oif=%d try_hard=%d\n", fl.fld_dst,
966                        fl.fld_src, fl.oif, try_hard);
967
968         /*
969          * N.B. If the kernel is compiled without router support then
970          * dn_fib_lookup() will evaluate to non-zero so this if () block
971          * will always be executed.
972          */
973         err = -ESRCH;
974         if (try_hard || (err = dn_fib_lookup(&fl, &res)) != 0) {
975                 struct dn_dev *dn_db;
976                 if (err != -ESRCH)
977                         goto out;
978                 /*
979                  * Here the fallback is basically the standard algorithm for 
980                  * routing in endnodes which is described in the DECnet routing
981                  * docs
982                  *
983                  * If we are not trying hard, look in neighbour cache.
984                  * The result is tested to ensure that if a specific output
985                  * device/source address was requested, then we honour that 
986                  * here
987                  */
988                 if (!try_hard) {
989                         neigh = neigh_lookup_nodev(&dn_neigh_table, &fl.fld_dst);
990                         if (neigh) {
991                                 if ((oldflp->oif && 
992                                     (neigh->dev->ifindex != oldflp->oif)) ||
993                                     (oldflp->fld_src &&
994                                     (!dn_dev_islocal(neigh->dev,
995                                                       oldflp->fld_src)))) {
996                                         neigh_release(neigh);
997                                         neigh = NULL;
998                                 } else {
999                                         if (dev_out)
1000                                                 dev_put(dev_out);
1001                                         if (dn_dev_islocal(neigh->dev, fl.fld_dst)) {
1002                                                 dev_out = &loopback_dev;
1003                                                 res.type = RTN_LOCAL;
1004                                         } else {
1005                                                 dev_out = neigh->dev;
1006                                         }
1007                                         dev_hold(dev_out);
1008                                         goto select_source;
1009                                 }
1010                         }
1011                 }
1012
1013                 /* Not there? Perhaps its a local address */
1014                 if (dev_out == NULL)
1015                         dev_out = dn_dev_get_default();
1016                 err = -ENODEV;
1017                 if (dev_out == NULL)
1018                         goto out;
1019                 dn_db = dev_out->dn_ptr;
1020                 /* Possible improvement - check all devices for local addr */
1021                 if (dn_dev_islocal(dev_out, fl.fld_dst)) {
1022                         dev_put(dev_out);
1023                         dev_out = &loopback_dev;
1024                         dev_hold(dev_out);
1025                         res.type = RTN_LOCAL;
1026                         goto select_source;
1027                 }
1028                 /* Not local either.... try sending it to the default router */
1029                 neigh = neigh_clone(dn_db->router);
1030                 BUG_ON(neigh && neigh->dev != dev_out);
1031
1032                 /* Ok then, we assume its directly connected and move on */
1033 select_source:
1034                 if (neigh)
1035                         gateway = ((struct dn_neigh *)neigh)->addr;
1036                 if (gateway == 0)
1037                         gateway = fl.fld_dst;
1038                 if (fl.fld_src == 0) {
1039                         fl.fld_src = dnet_select_source(dev_out, gateway,
1040                                                          res.type == RTN_LOCAL ?
1041                                                          RT_SCOPE_HOST : 
1042                                                          RT_SCOPE_LINK);
1043                         if (fl.fld_src == 0 && res.type != RTN_LOCAL)
1044                                 goto e_addr;
1045                 }
1046                 fl.oif = dev_out->ifindex;
1047                 goto make_route;
1048         }
1049         free_res = 1;
1050
1051         if (res.type == RTN_NAT)
1052                 goto e_inval;
1053
1054         if (res.type == RTN_LOCAL) {
1055                 if (!fl.fld_src)
1056                         fl.fld_src = fl.fld_dst;
1057                 if (dev_out)
1058                         dev_put(dev_out);
1059                 dev_out = &loopback_dev;
1060                 dev_hold(dev_out);
1061                 fl.oif = dev_out->ifindex;
1062                 if (res.fi)
1063                         dn_fib_info_put(res.fi);
1064                 res.fi = NULL;
1065                 goto make_route;
1066         }
1067
1068         if (res.fi->fib_nhs > 1 && fl.oif == 0)
1069                 dn_fib_select_multipath(&fl, &res);
1070
1071         /* 
1072          * We could add some logic to deal with default routes here and
1073          * get rid of some of the special casing above.
1074          */
1075
1076         if (!fl.fld_src)
1077                 fl.fld_src = DN_FIB_RES_PREFSRC(res);
1078         
1079         if (dev_out)
1080                 dev_put(dev_out);
1081         dev_out = DN_FIB_RES_DEV(res);
1082         dev_hold(dev_out);
1083         fl.oif = dev_out->ifindex;
1084         gateway = DN_FIB_RES_GW(res);
1085
1086 make_route:
1087         if (dev_out->flags & IFF_LOOPBACK)
1088                 flags |= RTCF_LOCAL;
1089
1090         rt = dst_alloc(&dn_dst_ops);
1091         if (rt == NULL)
1092                 goto e_nobufs;
1093
1094         atomic_set(&rt->u.dst.__refcnt, 1);
1095         rt->u.dst.flags   = DST_HOST;
1096
1097         rt->fl.fld_src    = oldflp->fld_src;
1098         rt->fl.fld_dst    = oldflp->fld_dst;
1099         rt->fl.oif        = oldflp->oif;
1100         rt->fl.iif        = 0;
1101 #ifdef CONFIG_DECNET_ROUTE_FWMARK
1102         rt->fl.fld_fwmark = oldflp->fld_fwmark;
1103 #endif
1104
1105         rt->rt_saddr      = fl.fld_src;
1106         rt->rt_daddr      = fl.fld_dst;
1107         rt->rt_gateway    = gateway ? gateway : fl.fld_dst;
1108         rt->rt_local_src  = fl.fld_src;
1109
1110         rt->rt_dst_map    = fl.fld_dst;
1111         rt->rt_src_map    = fl.fld_src;
1112
1113         rt->u.dst.dev = dev_out;
1114         dev_hold(dev_out);
1115         rt->u.dst.neighbour = neigh;
1116         neigh = NULL;
1117
1118         rt->u.dst.lastuse = jiffies;
1119         rt->u.dst.output  = dn_output;
1120         rt->u.dst.input   = dn_rt_bug;
1121         rt->rt_flags      = flags;
1122         if (flags & RTCF_LOCAL)
1123                 rt->u.dst.input = dn_nsp_rx;
1124
1125         err = dn_rt_set_next_hop(rt, &res);
1126         if (err)
1127                 goto e_neighbour;
1128
1129         hash = dn_hash(rt->fl.fld_src, rt->fl.fld_dst);
1130         dn_insert_route(rt, hash, (struct dn_route **)pprt);
1131
1132 done:
1133         if (neigh)
1134                 neigh_release(neigh);
1135         if (free_res)
1136                 dn_fib_res_put(&res);
1137         if (dev_out)
1138                 dev_put(dev_out);
1139 out:
1140         return err;
1141
1142 e_addr:
1143         err = -EADDRNOTAVAIL;
1144         goto done;
1145 e_inval:
1146         err = -EINVAL;
1147         goto done;
1148 e_nobufs:
1149         err = -ENOBUFS;
1150         goto done;
1151 e_neighbour:
1152         dst_free(&rt->u.dst);
1153         goto e_nobufs;
1154 }
1155
1156
1157 /*
1158  * N.B. The flags may be moved into the flowi at some future stage.
1159  */
1160 static int __dn_route_output_key(struct dst_entry **pprt, const struct flowi *flp, int flags)
1161 {
1162         unsigned hash = dn_hash(flp->fld_src, flp->fld_dst);
1163         struct dn_route *rt = NULL;
1164
1165         if (!(flags & MSG_TRYHARD)) {
1166                 rcu_read_lock_bh();
1167                 for(rt = rcu_dereference(dn_rt_hash_table[hash].chain); rt;
1168                         rt = rcu_dereference(rt->u.rt_next)) {
1169                         if ((flp->fld_dst == rt->fl.fld_dst) &&
1170                             (flp->fld_src == rt->fl.fld_src) &&
1171 #ifdef CONFIG_DECNET_ROUTE_FWMARK
1172                             (flp->fld_fwmark == rt->fl.fld_fwmark) &&
1173 #endif
1174                             (rt->fl.iif == 0) &&
1175                             (rt->fl.oif == flp->oif)) {
1176                                 rt->u.dst.lastuse = jiffies;
1177                                 dst_hold(&rt->u.dst);
1178                                 rt->u.dst.__use++;
1179                                 rcu_read_unlock_bh();
1180                                 *pprt = &rt->u.dst;
1181                                 return 0;
1182                         }
1183                 }
1184                 rcu_read_unlock_bh();
1185         }
1186
1187         return dn_route_output_slow(pprt, flp, flags);
1188 }
1189
1190 static int dn_route_output_key(struct dst_entry **pprt, struct flowi *flp, int flags)
1191 {
1192         int err;
1193
1194         err = __dn_route_output_key(pprt, flp, flags);
1195         if (err == 0 && flp->proto) {
1196                 err = xfrm_lookup(pprt, flp, NULL, 0);
1197         }
1198         return err;
1199 }
1200
1201 int dn_route_output_sock(struct dst_entry **pprt, struct flowi *fl, struct sock *sk, int flags)
1202 {
1203         int err;
1204
1205         err = __dn_route_output_key(pprt, fl, flags & MSG_TRYHARD);
1206         if (err == 0 && fl->proto) {
1207                 err = xfrm_lookup(pprt, fl, sk, !(flags & MSG_DONTWAIT));
1208         }
1209         return err;
1210 }
1211
1212 static int dn_route_input_slow(struct sk_buff *skb)
1213 {
1214         struct dn_route *rt = NULL;
1215         struct dn_skb_cb *cb = DN_SKB_CB(skb);
1216         struct net_device *in_dev = skb->dev;
1217         struct net_device *out_dev = NULL;
1218         struct dn_dev *dn_db;
1219         struct neighbour *neigh = NULL;
1220         unsigned hash;
1221         int flags = 0;
1222         __u16 gateway = 0;
1223         __u16 local_src = 0;
1224         struct flowi fl = { .nl_u = { .dn_u = 
1225                                      { .daddr = cb->dst,
1226                                        .saddr = cb->src,
1227                                        .scope = RT_SCOPE_UNIVERSE,
1228 #ifdef CONFIG_DECNET_ROUTE_FWMARK
1229                                        .fwmark = skb->nfmark
1230 #endif
1231                                     } },
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                 flags |= RTCF_DIRECTSRC;
1269         } else {
1270                 __u16 src_map = fl.fld_src;
1271                 free_res = 1;
1272
1273                 out_dev = DN_FIB_RES_DEV(res);
1274                 if (out_dev == NULL) {
1275                         if (net_ratelimit())
1276                                 printk(KERN_CRIT "Bug in dn_route_input_slow() "
1277                                                  "No output device\n");
1278                         goto e_inval;
1279                 }
1280                 dev_hold(out_dev);
1281
1282                 if (res.r)
1283                         src_map = dn_fib_rules_policy(fl.fld_src, &res, &flags);
1284
1285                 gateway = DN_FIB_RES_GW(res);
1286                 if (res.type == RTN_NAT) {
1287                         fl.fld_dst = dn_fib_rules_map_destination(fl.fld_dst, &res);
1288                         dn_fib_res_put(&res);
1289                         free_res = 0;
1290                         if (dn_fib_lookup(&fl, &res))
1291                                 goto e_inval;
1292                         free_res = 1;
1293                         if (res.type != RTN_UNICAST)
1294                                 goto e_inval;
1295                         flags |= RTCF_DNAT;
1296                         gateway = fl.fld_dst;
1297                 }
1298                 fl.fld_src = src_map;
1299         }
1300
1301         switch(res.type) {
1302         case RTN_UNICAST:
1303                 /*
1304                  * Forwarding check here, we only check for forwarding
1305                  * being turned off, if you want to only forward intra
1306                  * area, its up to you to set the routing tables up
1307                  * correctly.
1308                  */
1309                 if (dn_db->parms.forwarding == 0)
1310                         goto e_inval;
1311
1312                 if (res.fi->fib_nhs > 1 && fl.oif == 0)
1313                         dn_fib_select_multipath(&fl, &res);
1314
1315                 /* 
1316                  * Check for out_dev == in_dev. We use the RTCF_DOREDIRECT
1317                  * flag as a hint to set the intra-ethernet bit when
1318                  * forwarding. If we've got NAT in operation, we don't do
1319                  * this optimisation.
1320                  */
1321                 if (out_dev == in_dev && !(flags & RTCF_NAT))
1322                         flags |= RTCF_DOREDIRECT;
1323
1324                 local_src = DN_FIB_RES_PREFSRC(res);
1325
1326         case RTN_BLACKHOLE:
1327         case RTN_UNREACHABLE:
1328                 break;
1329         case RTN_LOCAL:
1330                 flags |= RTCF_LOCAL;
1331                 fl.fld_src = cb->dst;
1332                 fl.fld_dst = cb->src;
1333
1334                 /* Routing tables gave us a gateway */
1335                 if (gateway)
1336                         goto make_route;
1337
1338                 /* Packet was intra-ethernet, so we know its on-link */
1339                 if (cb->rt_flags | DN_RT_F_IE) {
1340                         gateway = cb->src;
1341                         flags |= RTCF_DIRECTSRC;
1342                         goto make_route;
1343                 }
1344
1345                 /* Use the default router if there is one */
1346                 neigh = neigh_clone(dn_db->router);
1347                 if (neigh) {
1348                         gateway = ((struct dn_neigh *)neigh)->addr;
1349                         goto make_route;
1350                 }
1351
1352                 /* Close eyes and pray */
1353                 gateway = cb->src;
1354                 flags |= RTCF_DIRECTSRC;
1355                 goto make_route;
1356         default:
1357                 goto e_inval;
1358         }
1359
1360 make_route:
1361         rt = dst_alloc(&dn_dst_ops);
1362         if (rt == NULL)
1363                 goto e_nobufs;
1364
1365         rt->rt_saddr      = fl.fld_src;
1366         rt->rt_daddr      = fl.fld_dst;
1367         rt->rt_gateway    = fl.fld_dst;
1368         if (gateway)
1369                 rt->rt_gateway = gateway;
1370         rt->rt_local_src  = local_src ? local_src : rt->rt_saddr;
1371
1372         rt->rt_dst_map    = fl.fld_dst;
1373         rt->rt_src_map    = fl.fld_src;
1374
1375         rt->fl.fld_src    = cb->src;
1376         rt->fl.fld_dst    = cb->dst;
1377         rt->fl.oif        = 0;
1378         rt->fl.iif        = in_dev->ifindex;
1379         rt->fl.fld_fwmark = fl.fld_fwmark;
1380
1381         rt->u.dst.flags = DST_HOST;
1382         rt->u.dst.neighbour = neigh;
1383         rt->u.dst.dev = out_dev;
1384         rt->u.dst.lastuse = jiffies;
1385         rt->u.dst.output = dn_rt_bug;
1386         switch(res.type) {
1387                 case RTN_UNICAST:
1388                         rt->u.dst.input = dn_forward;
1389                         break;
1390                 case RTN_LOCAL:
1391                         rt->u.dst.output = dn_output;
1392                         rt->u.dst.input = dn_nsp_rx;
1393                         rt->u.dst.dev = in_dev;
1394                         flags |= RTCF_LOCAL;
1395                         break;
1396                 default:
1397                 case RTN_UNREACHABLE:
1398                 case RTN_BLACKHOLE:
1399                         rt->u.dst.input = dn_blackhole;
1400         }
1401         rt->rt_flags = flags;
1402         if (rt->u.dst.dev)
1403                 dev_hold(rt->u.dst.dev);
1404
1405         err = dn_rt_set_next_hop(rt, &res);
1406         if (err)
1407                 goto e_neighbour;
1408
1409         hash = dn_hash(rt->fl.fld_src, rt->fl.fld_dst);
1410         dn_insert_route(rt, hash, (struct dn_route **)&skb->dst);
1411
1412 done:
1413         if (neigh)
1414                 neigh_release(neigh);
1415         if (free_res)
1416                 dn_fib_res_put(&res);
1417         dev_put(in_dev);
1418         if (out_dev)
1419                 dev_put(out_dev);
1420 out:
1421         return err;
1422
1423 e_inval:
1424         err = -EINVAL;
1425         goto done;
1426
1427 e_nobufs:
1428         err = -ENOBUFS;
1429         goto done;
1430
1431 e_neighbour:
1432         dst_free(&rt->u.dst);
1433         goto done;
1434 }
1435
1436 int dn_route_input(struct sk_buff *skb)
1437 {
1438         struct dn_route *rt;
1439         struct dn_skb_cb *cb = DN_SKB_CB(skb);
1440         unsigned hash = dn_hash(cb->src, cb->dst);
1441
1442         if (skb->dst)
1443                 return 0;
1444
1445         rcu_read_lock();
1446         for(rt = rcu_dereference(dn_rt_hash_table[hash].chain); rt != NULL;
1447             rt = rcu_dereference(rt->u.rt_next)) {
1448                 if ((rt->fl.fld_src == cb->src) &&
1449                     (rt->fl.fld_dst == cb->dst) &&
1450                     (rt->fl.oif == 0) &&
1451 #ifdef CONFIG_DECNET_ROUTE_FWMARK
1452                     (rt->fl.fld_fwmark == skb->nfmark) &&
1453 #endif
1454                     (rt->fl.iif == cb->iif)) {
1455                         rt->u.dst.lastuse = jiffies;
1456                         dst_hold(&rt->u.dst);
1457                         rt->u.dst.__use++;
1458                         rcu_read_unlock();
1459                         skb->dst = (struct dst_entry *)rt;
1460                         return 0;
1461                 }
1462         }
1463         rcu_read_unlock();
1464
1465         return dn_route_input_slow(skb);
1466 }
1467
1468 static int dn_rt_fill_info(struct sk_buff *skb, u32 pid, u32 seq, int event, int nowait)
1469 {
1470         struct dn_route *rt = (struct dn_route *)skb->dst;
1471         struct rtmsg *r;
1472         struct nlmsghdr *nlh;
1473         unsigned char *b = skb->tail;
1474         struct rta_cacheinfo ci;
1475
1476         nlh = NLMSG_PUT(skb, pid, seq, event, sizeof(*r));
1477         r = NLMSG_DATA(nlh);
1478         nlh->nlmsg_flags = (nowait && pid) ? NLM_F_MULTI : 0;
1479         r->rtm_family = AF_DECnet;
1480         r->rtm_dst_len = 16;
1481         r->rtm_src_len = 0;
1482         r->rtm_tos = 0;
1483         r->rtm_table = RT_TABLE_MAIN;
1484         r->rtm_type = rt->rt_type;
1485         r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
1486         r->rtm_scope = RT_SCOPE_UNIVERSE;
1487         r->rtm_protocol = RTPROT_UNSPEC;
1488         if (rt->rt_flags & RTCF_NOTIFY)
1489                 r->rtm_flags |= RTM_F_NOTIFY;
1490         RTA_PUT(skb, RTA_DST, 2, &rt->rt_daddr);
1491         if (rt->fl.fld_src) {
1492                 r->rtm_src_len = 16;
1493                 RTA_PUT(skb, RTA_SRC, 2, &rt->fl.fld_src);
1494         }
1495         if (rt->u.dst.dev)
1496                 RTA_PUT(skb, RTA_OIF, sizeof(int), &rt->u.dst.dev->ifindex);
1497         /*
1498          * Note to self - change this if input routes reverse direction when
1499          * they deal only with inputs and not with replies like they do
1500          * currently.
1501          */
1502         RTA_PUT(skb, RTA_PREFSRC, 2, &rt->rt_local_src);
1503         if (rt->rt_daddr != rt->rt_gateway)
1504                 RTA_PUT(skb, RTA_GATEWAY, 2, &rt->rt_gateway);
1505         if (rtnetlink_put_metrics(skb, rt->u.dst.metrics) < 0)
1506                 goto rtattr_failure;
1507         ci.rta_lastuse = jiffies_to_clock_t(jiffies - rt->u.dst.lastuse);
1508         ci.rta_used     = rt->u.dst.__use;
1509         ci.rta_clntref  = atomic_read(&rt->u.dst.__refcnt);
1510         if (rt->u.dst.expires)
1511                 ci.rta_expires = jiffies_to_clock_t(rt->u.dst.expires - jiffies);
1512         else
1513                 ci.rta_expires = 0;
1514         ci.rta_error    = rt->u.dst.error;
1515         ci.rta_id       = ci.rta_ts = ci.rta_tsage = 0;
1516         RTA_PUT(skb, RTA_CACHEINFO, sizeof(ci), &ci);
1517         if (rt->fl.iif)
1518                 RTA_PUT(skb, RTA_IIF, sizeof(int), &rt->fl.iif);
1519
1520         nlh->nlmsg_len = skb->tail - b;
1521         return skb->len;
1522
1523 nlmsg_failure:
1524 rtattr_failure:
1525         skb_trim(skb, b - skb->data);
1526         return -1;
1527 }
1528
1529 /*
1530  * This is called by both endnodes and routers now.
1531  */
1532 int dn_cache_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh, void *arg)
1533 {
1534         struct rtattr **rta = arg;
1535         struct rtmsg *rtm = NLMSG_DATA(nlh);
1536         struct dn_route *rt = NULL;
1537         struct dn_skb_cb *cb;
1538         int err;
1539         struct sk_buff *skb;
1540         struct flowi fl;
1541
1542         memset(&fl, 0, sizeof(fl));
1543         fl.proto = DNPROTO_NSP;
1544
1545         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1546         if (skb == NULL)
1547                 return -ENOBUFS;
1548         skb->mac.raw = skb->data;
1549         cb = DN_SKB_CB(skb);
1550
1551         if (rta[RTA_SRC-1])
1552                 memcpy(&fl.fld_src, RTA_DATA(rta[RTA_SRC-1]), 2);
1553         if (rta[RTA_DST-1])
1554                 memcpy(&fl.fld_dst, RTA_DATA(rta[RTA_DST-1]), 2);
1555         if (rta[RTA_IIF-1])
1556                 memcpy(&fl.iif, RTA_DATA(rta[RTA_IIF-1]), sizeof(int));
1557
1558         if (fl.iif) {
1559                 struct net_device *dev;
1560                 if ((dev = dev_get_by_index(fl.iif)) == NULL) {
1561                         kfree_skb(skb);
1562                         return -ENODEV;
1563                 }
1564                 if (!dev->dn_ptr) {
1565                         dev_put(dev);
1566                         kfree_skb(skb);
1567                         return -ENODEV;
1568                 }
1569                 skb->protocol = __constant_htons(ETH_P_DNA_RT);
1570                 skb->dev = dev;
1571                 cb->src = fl.fld_src;
1572                 cb->dst = fl.fld_dst;
1573                 local_bh_disable();
1574                 err = dn_route_input(skb);
1575                 local_bh_enable();
1576                 memset(cb, 0, sizeof(struct dn_skb_cb));
1577                 rt = (struct dn_route *)skb->dst;
1578                 if (!err && -rt->u.dst.error)
1579                         err = rt->u.dst.error;
1580         } else {
1581                 int oif = 0;
1582                 if (rta[RTA_OIF - 1])
1583                         memcpy(&oif, RTA_DATA(rta[RTA_OIF - 1]), sizeof(int));
1584                 fl.oif = oif;
1585                 err = dn_route_output_key((struct dst_entry **)&rt, &fl, 0);
1586         }
1587
1588         if (skb->dev)
1589                 dev_put(skb->dev);
1590         skb->dev = NULL;
1591         if (err)
1592                 goto out_free;
1593         skb->dst = &rt->u.dst;
1594         if (rtm->rtm_flags & RTM_F_NOTIFY)
1595                 rt->rt_flags |= RTCF_NOTIFY;
1596
1597         NETLINK_CB(skb).dst_pid = NETLINK_CB(in_skb).pid;
1598
1599         err = dn_rt_fill_info(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq, RTM_NEWROUTE, 0);
1600
1601         if (err == 0)
1602                 goto out_free;
1603         if (err < 0) {
1604                 err = -EMSGSIZE;
1605                 goto out_free;
1606         }
1607
1608         err = netlink_unicast(rtnl, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1609
1610         return err;
1611
1612 out_free:
1613         kfree_skb(skb);
1614         return err;
1615 }
1616
1617 /*
1618  * For routers, this is called from dn_fib_dump, but for endnodes its
1619  * called directly from the rtnetlink dispatch table.
1620  */
1621 int dn_cache_dump(struct sk_buff *skb, struct netlink_callback *cb)
1622 {
1623         struct dn_route *rt;
1624         int h, s_h;
1625         int idx, s_idx;
1626
1627         if (NLMSG_PAYLOAD(cb->nlh, 0) < sizeof(struct rtmsg))
1628                 return -EINVAL;
1629         if (!(((struct rtmsg *)NLMSG_DATA(cb->nlh))->rtm_flags&RTM_F_CLONED))
1630                 return 0;
1631
1632         s_h = cb->args[0];
1633         s_idx = idx = cb->args[1];
1634         for(h = 0; h <= dn_rt_hash_mask; h++) {
1635                 if (h < s_h)
1636                         continue;
1637                 if (h > s_h)
1638                         s_idx = 0;
1639                 rcu_read_lock_bh();
1640                 for(rt = rcu_dereference(dn_rt_hash_table[h].chain), idx = 0;
1641                         rt;
1642                         rt = rcu_dereference(rt->u.rt_next), idx++) {
1643                         if (idx < s_idx)
1644                                 continue;
1645                         skb->dst = dst_clone(&rt->u.dst);
1646                         if (dn_rt_fill_info(skb, NETLINK_CB(cb->skb).pid,
1647                                         cb->nlh->nlmsg_seq, RTM_NEWROUTE, 1) <= 0) {
1648                                 dst_release(xchg(&skb->dst, NULL));
1649                                 rcu_read_unlock_bh();
1650                                 goto done;
1651                         }
1652                         dst_release(xchg(&skb->dst, NULL));
1653                 }
1654                 rcu_read_unlock_bh();
1655         }
1656
1657 done:
1658         cb->args[0] = h;
1659         cb->args[1] = idx;
1660         return skb->len;
1661 }
1662
1663 #ifdef CONFIG_PROC_FS
1664 struct dn_rt_cache_iter_state {
1665         int bucket;
1666 };
1667
1668 static struct dn_route *dn_rt_cache_get_first(struct seq_file *seq)
1669 {
1670         struct dn_route *rt = NULL;
1671         struct dn_rt_cache_iter_state *s = seq->private;
1672
1673         for(s->bucket = dn_rt_hash_mask; s->bucket >= 0; --s->bucket) {
1674                 rcu_read_lock_bh();
1675                 rt = dn_rt_hash_table[s->bucket].chain;
1676                 if (rt)
1677                         break;
1678                 rcu_read_unlock_bh();
1679         }
1680         return rt;
1681 }
1682
1683 static struct dn_route *dn_rt_cache_get_next(struct seq_file *seq, struct dn_route *rt)
1684 {
1685         struct dn_rt_cache_iter_state *s = rcu_dereference(seq->private);
1686
1687         rt = rt->u.rt_next;
1688         while(!rt) {
1689                 rcu_read_unlock_bh();
1690                 if (--s->bucket < 0)
1691                         break;
1692                 rcu_read_lock_bh();
1693                 rt = dn_rt_hash_table[s->bucket].chain;
1694         }
1695         return rt;
1696 }
1697
1698 static void *dn_rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
1699 {
1700         struct dn_route *rt = dn_rt_cache_get_first(seq);
1701
1702         if (rt) {
1703                 while(*pos && (rt = dn_rt_cache_get_next(seq, rt)))
1704                         --*pos;
1705         }
1706         return *pos ? NULL : rt;
1707 }
1708
1709 static void *dn_rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1710 {
1711         struct dn_route *rt = dn_rt_cache_get_next(seq, v);
1712         ++*pos;
1713         return rt;
1714 }
1715
1716 static void dn_rt_cache_seq_stop(struct seq_file *seq, void *v)
1717 {
1718         if (v)
1719                 rcu_read_unlock_bh();
1720 }
1721
1722 static int dn_rt_cache_seq_show(struct seq_file *seq, void *v)
1723 {
1724         struct dn_route *rt = v;
1725         char buf1[DN_ASCBUF_LEN], buf2[DN_ASCBUF_LEN];
1726
1727         seq_printf(seq, "%-8s %-7s %-7s %04d %04d %04d\n",
1728                         rt->u.dst.dev ? rt->u.dst.dev->name : "*",
1729                         dn_addr2asc(dn_ntohs(rt->rt_daddr), buf1),
1730                         dn_addr2asc(dn_ntohs(rt->rt_saddr), buf2),
1731                         atomic_read(&rt->u.dst.__refcnt),
1732                         rt->u.dst.__use,
1733                         (int) dst_metric(&rt->u.dst, RTAX_RTT));
1734         return 0;
1735
1736
1737 static struct seq_operations dn_rt_cache_seq_ops = {
1738         .start  = dn_rt_cache_seq_start,
1739         .next   = dn_rt_cache_seq_next,
1740         .stop   = dn_rt_cache_seq_stop,
1741         .show   = dn_rt_cache_seq_show,
1742 };
1743
1744 static int dn_rt_cache_seq_open(struct inode *inode, struct file *file)
1745 {
1746         struct seq_file *seq;
1747         int rc = -ENOMEM;
1748         struct dn_rt_cache_iter_state *s = kmalloc(sizeof(*s), GFP_KERNEL);
1749
1750         if (!s)
1751                 goto out;
1752         rc = seq_open(file, &dn_rt_cache_seq_ops);
1753         if (rc)
1754                 goto out_kfree;
1755         seq             = file->private_data;
1756         seq->private    = s;
1757         memset(s, 0, sizeof(*s));
1758 out:
1759         return rc;
1760 out_kfree:
1761         kfree(s);
1762         goto out;
1763 }
1764
1765 static struct file_operations dn_rt_cache_seq_fops = {
1766         .owner   = THIS_MODULE,
1767         .open    = dn_rt_cache_seq_open,
1768         .read    = seq_read,
1769         .llseek  = seq_lseek,
1770         .release = seq_release_private,
1771 };
1772
1773 #endif /* CONFIG_PROC_FS */
1774
1775 void __init dn_route_init(void)
1776 {
1777         int i, goal, order;
1778
1779         dn_dst_ops.kmem_cachep = kmem_cache_create("dn_dst_cache",
1780                                                    sizeof(struct dn_route),
1781                                                    0, SLAB_HWCACHE_ALIGN,
1782                                                    NULL, NULL);
1783
1784         if (!dn_dst_ops.kmem_cachep)
1785                 panic("DECnet: Failed to allocate dn_dst_cache\n");
1786
1787         init_timer(&dn_route_timer);
1788         dn_route_timer.function = dn_dst_check_expire;
1789         dn_route_timer.expires = jiffies + decnet_dst_gc_interval * HZ;
1790         add_timer(&dn_route_timer);
1791
1792         goal = num_physpages >> (26 - PAGE_SHIFT);
1793
1794         for(order = 0; (1UL << order) < goal; order++)
1795                 /* NOTHING */;
1796
1797         /*
1798          * Only want 1024 entries max, since the table is very, very unlikely
1799          * to be larger than that.
1800          */
1801         while(order && ((((1UL << order) * PAGE_SIZE) / 
1802                                 sizeof(struct dn_rt_hash_bucket)) >= 2048))
1803                 order--;
1804
1805         do {
1806                 dn_rt_hash_mask = (1UL << order) * PAGE_SIZE /
1807                         sizeof(struct dn_rt_hash_bucket);
1808                 while(dn_rt_hash_mask & (dn_rt_hash_mask - 1))
1809                         dn_rt_hash_mask--;
1810                 dn_rt_hash_table = (struct dn_rt_hash_bucket *)
1811                         __get_free_pages(GFP_ATOMIC, order);
1812         } while (dn_rt_hash_table == NULL && --order > 0);
1813
1814         if (!dn_rt_hash_table)
1815                 panic("Failed to allocate DECnet route cache hash table\n");
1816
1817         printk(KERN_INFO 
1818                 "DECnet: Routing cache hash table of %u buckets, %ldKbytes\n", 
1819                 dn_rt_hash_mask, 
1820                 (long)(dn_rt_hash_mask*sizeof(struct dn_rt_hash_bucket))/1024);
1821
1822         dn_rt_hash_mask--;
1823         for(i = 0; i <= dn_rt_hash_mask; i++) {
1824                 spin_lock_init(&dn_rt_hash_table[i].lock);
1825                 dn_rt_hash_table[i].chain = NULL;
1826         }
1827
1828         dn_dst_ops.gc_thresh = (dn_rt_hash_mask + 1);
1829
1830         proc_net_fops_create("decnet_cache", S_IRUGO, &dn_rt_cache_seq_fops);
1831 }
1832
1833 void __exit dn_route_cleanup(void)
1834 {
1835         del_timer(&dn_route_timer);
1836         dn_run_flush(0);
1837
1838         proc_net_remove("decnet_cache");
1839 }
1840