make oldconfig will rebuild these...
[linux-2.4.21-pre4.git] / net / decnet / dn_route.c
1
2 /*
3  * DECnet       An implementation of the DECnet protocol suite for the LINUX
4  *              operating system.  DECnet is implemented using the  BSD Socket
5  *              interface as the means of communication with the user level.
6  *
7  *              DECnet Routing Functions (Endnode and Router)
8  *
9  * Authors:     Steve Whitehouse <SteveW@ACM.org>
10  *              Eduardo Marcelo Serrat <emserrat@geocities.com>
11  *
12  * Changes:
13  *              Steve Whitehouse : Fixes to allow "intra-ethernet" and
14  *                                 "return-to-sender" bits on outgoing
15  *                                 packets.
16  *              Steve Whitehouse : Timeouts for cached routes.
17  *              Steve Whitehouse : Use dst cache for input routes too.
18  *              Steve Whitehouse : Fixed error values in dn_send_skb.
19  *              Steve Whitehouse : Rework routing functions to better fit
20  *                                 DECnet routing design
21  *              Alexey Kuznetsov : New SMP locking
22  *              Steve Whitehouse : More SMP locking changes & dn_cache_dump()
23  *              Steve Whitehouse : Prerouting NF hook, now really is prerouting.
24  *                                 Fixed possible skb leak in rtnetlink funcs.
25  *              Steve Whitehouse : Dave Miller's dynamic hash table sizing and
26  *                                 Alexey Kuznetsov's finer grained locking
27  *                                 from ipv4/route.c.
28  *              Steve Whitehouse : Routing is now starting to look like a
29  *                                 sensible set of code now, mainly due to
30  *                                 my copying the IPv4 routing code. The
31  *                                 hooks here are modified and will continue
32  *                                 to evolve for a while.
33  *              Steve Whitehouse : Real SMP at last :-) Also new netfilter
34  *                                 stuff. Look out raw sockets your days
35  *                                 are numbered!
36  *              Steve Whitehouse : Added return-to-sender functions. Added
37  *                                 backlog congestion level return codes.
38  *              Steve Whitehouse : Fixed bug where routes were set up with
39  *                                 no ref count on net devices.
40  *                                 
41  */
42
43 /******************************************************************************
44     (c) 1995-1998 E.M. Serrat           emserrat@geocities.com
45     
46     This program is free software; you can redistribute it and/or modify
47     it under the terms of the GNU General Public License as published by
48     the Free Software Foundation; either version 2 of the License, or
49     any later version.
50
51     This program is distributed in the hope that it will be useful,
52     but WITHOUT ANY WARRANTY; without even the implied warranty of
53     MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
54     GNU General Public License for more details.
55 *******************************************************************************/
56
57 #include <linux/config.h>
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 <net/sock.h>
69 #include <linux/fcntl.h>
70 #include <linux/mm.h>
71 #include <linux/proc_fs.h>
72 #include <linux/init.h>
73 #include <linux/rtnetlink.h>
74 #include <linux/string.h>
75 #include <linux/netfilter_decnet.h>
76 #include <asm/errno.h>
77 #include <net/neighbour.h>
78 #include <net/dst.h>
79 #include <net/dn.h>
80 #include <net/dn_dev.h>
81 #include <net/dn_nsp.h>
82 #include <net/dn_route.h>
83 #include <net/dn_neigh.h>
84 #include <net/dn_fib.h>
85
86 struct dn_rt_hash_bucket
87 {
88         struct dn_route *chain;
89         rwlock_t lock;
90 } __attribute__((__aligned__(8)));
91
92 extern struct neigh_table dn_neigh_table;
93
94
95 static unsigned char dn_hiord_addr[6] = {0xAA,0x00,0x04,0x00,0x00,0x00};
96
97 int dn_rt_min_delay = 2*HZ;
98 int dn_rt_max_delay = 10*HZ;
99 static unsigned long dn_rt_deadline = 0;
100
101 static int dn_dst_gc(void);
102 static struct dst_entry *dn_dst_check(struct dst_entry *, __u32);
103 static struct dst_entry *dn_dst_reroute(struct dst_entry *, struct sk_buff *skb);
104 static struct dst_entry *dn_dst_negative_advice(struct dst_entry *);
105 static void dn_dst_link_failure(struct sk_buff *);
106 static int dn_route_input(struct sk_buff *);
107 static void dn_run_flush(unsigned long dummy);
108
109 static struct dn_rt_hash_bucket *dn_rt_hash_table;
110 static unsigned dn_rt_hash_mask;
111
112 static struct timer_list dn_route_timer;
113 static struct timer_list dn_rt_flush_timer = { function: dn_run_flush };
114 int decnet_dst_gc_interval = 2;
115
116 static struct dst_ops dn_dst_ops = {
117         family:                 PF_DECnet,
118         protocol:               __constant_htons(ETH_P_DNA_RT),
119         gc_thresh:              128,
120         gc:                     dn_dst_gc,
121         check:                  dn_dst_check,
122         reroute:                dn_dst_reroute,
123         negative_advice:        dn_dst_negative_advice,
124         link_failure:           dn_dst_link_failure,
125         entry_size:             sizeof(struct dn_route),
126         entries:                ATOMIC_INIT(0),
127 };
128
129 static __inline__ unsigned dn_hash(unsigned short src, unsigned short dst)
130 {
131         unsigned short tmp = src ^ dst;
132         tmp ^= (tmp >> 3);
133         tmp ^= (tmp >> 5);
134         tmp ^= (tmp >> 10);
135         return dn_rt_hash_mask & (unsigned)tmp;
136 }
137
138 static void SMP_TIMER_NAME(dn_dst_check_expire)(unsigned long dummy)
139 {
140         int i;
141         struct dn_route *rt, **rtp;
142         unsigned long now = jiffies;
143         unsigned long expire = 120 * HZ;
144
145         for(i = 0; i <= dn_rt_hash_mask; i++) {
146                 rtp = &dn_rt_hash_table[i].chain;
147
148                 write_lock(&dn_rt_hash_table[i].lock);
149                 while((rt=*rtp) != NULL) {
150                         if (atomic_read(&rt->u.dst.__refcnt) ||
151                                         (now - rt->u.dst.lastuse) < expire) {
152                                 rtp = &rt->u.rt_next;
153                                 continue;
154                         }
155                         *rtp = rt->u.rt_next;
156                         rt->u.rt_next = NULL;
157                         dst_free(&rt->u.dst);
158                 }
159                 write_unlock(&dn_rt_hash_table[i].lock);
160
161                 if ((jiffies - now) > 0)
162                         break;
163         }
164
165         mod_timer(&dn_route_timer, now + decnet_dst_gc_interval * HZ);
166 }
167
168 SMP_TIMER_DEFINE(dn_dst_check_expire, dn_dst_task);
169
170 static int dn_dst_gc(void)
171 {
172         struct dn_route *rt, **rtp;
173         int i;
174         unsigned long now = jiffies;
175         unsigned long expire = 10 * HZ;
176
177         for(i = 0; i <= dn_rt_hash_mask; i++) {
178
179                 write_lock_bh(&dn_rt_hash_table[i].lock);
180                 rtp = &dn_rt_hash_table[i].chain;
181
182                 while((rt=*rtp) != NULL) {
183                         if (atomic_read(&rt->u.dst.__refcnt) ||
184                                         (now - rt->u.dst.lastuse) < expire) {
185                                 rtp = &rt->u.rt_next;
186                                 continue;
187                         }
188                         *rtp = rt->u.rt_next;
189                         rt->u.rt_next = NULL;
190                         dst_free(&rt->u.dst);
191                         break;
192                 }
193                 write_unlock_bh(&dn_rt_hash_table[i].lock);
194         }
195
196         return 0;
197 }
198
199 static struct dst_entry *dn_dst_check(struct dst_entry *dst, __u32 cookie)
200 {
201         dst_release(dst);
202         return NULL;
203 }
204
205 static struct dst_entry *dn_dst_reroute(struct dst_entry *dst,
206                                         struct sk_buff *skb)
207 {
208         return NULL;
209 }
210
211 /*
212  * This is called through sendmsg() when you specify MSG_TRYHARD
213  * and there is already a route in cache.
214  */
215 static struct dst_entry *dn_dst_negative_advice(struct dst_entry *dst)
216 {
217         dst_release(dst);
218         return NULL;
219 }
220
221 static void dn_dst_link_failure(struct sk_buff *skb)
222 {
223         return;
224 }
225
226 static void dn_insert_route(struct dn_route *rt, unsigned hash)
227 {
228         unsigned long now = jiffies;
229
230         write_lock_bh(&dn_rt_hash_table[hash].lock);
231         rt->u.rt_next = dn_rt_hash_table[hash].chain;
232         dn_rt_hash_table[hash].chain = rt;
233         
234         dst_hold(&rt->u.dst);
235         rt->u.dst.__use++;
236         rt->u.dst.lastuse = now;
237
238         write_unlock_bh(&dn_rt_hash_table[hash].lock);
239 }
240
241 void SMP_TIMER_NAME(dn_run_flush)(unsigned long dummy)
242 {
243         int i;
244         struct dn_route *rt, *next;
245
246         for(i = 0; i < dn_rt_hash_mask; i++) {
247                 write_lock_bh(&dn_rt_hash_table[i].lock);
248
249                 if ((rt = xchg(&dn_rt_hash_table[i].chain, NULL)) == NULL)
250                         goto nothing_to_declare;
251
252                 for(; rt; rt=next) {
253                         next = rt->u.rt_next;
254                         rt->u.rt_next = NULL;
255                         dst_free((struct dst_entry *)rt);
256                 }
257
258 nothing_to_declare:
259                 write_unlock_bh(&dn_rt_hash_table[i].lock);
260         }
261 }
262
263 SMP_TIMER_DEFINE(dn_run_flush, dn_flush_task);
264
265 static spinlock_t dn_rt_flush_lock = SPIN_LOCK_UNLOCKED;
266
267 void dn_rt_cache_flush(int delay)
268 {
269         unsigned long now = jiffies;
270         int user_mode = !in_interrupt();
271
272         if (delay < 0)
273                 delay = dn_rt_min_delay;
274
275         spin_lock_bh(&dn_rt_flush_lock);
276
277         if (del_timer(&dn_rt_flush_timer) && delay > 0 && dn_rt_deadline) {
278                 long tmo = (long)(dn_rt_deadline - now);
279
280                 if (user_mode && tmo < dn_rt_max_delay - dn_rt_min_delay)
281                         tmo = 0;
282
283                 if (delay > tmo)
284                         delay = tmo;
285         }
286
287         if (delay <= 0) {
288                 spin_unlock_bh(&dn_rt_flush_lock);
289                 dn_run_flush(0);
290                 return;
291         }
292
293         if (dn_rt_deadline == 0)
294                 dn_rt_deadline = now + dn_rt_max_delay;
295
296         dn_rt_flush_timer.expires = now + delay;
297         add_timer(&dn_rt_flush_timer);
298         spin_unlock_bh(&dn_rt_flush_lock);
299 }
300
301 /**
302  * dn_return_short - Return a short packet to its sender
303  * @skb: The packet to return
304  *
305  */
306 static int dn_return_short(struct sk_buff *skb)
307 {
308         struct dn_skb_cb *cb;
309         unsigned char *ptr;
310         dn_address *src;
311         dn_address *dst;
312         dn_address tmp;
313
314         /* Add back headers */
315         skb_push(skb, skb->data - skb->nh.raw);
316
317         if ((skb = skb_unshare(skb, GFP_ATOMIC)) == NULL)
318                 return NET_RX_DROP;
319
320         cb = DN_SKB_CB(skb);
321         /* Skip packet length and point to flags */
322         ptr = skb->data + 2;
323         *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
324
325         dst = (dn_address *)ptr;
326         ptr += 2;
327         src = (dn_address *)ptr;
328         ptr += 2;
329         *ptr = 0; /* Zero hop count */
330
331         /* Swap source and destination */
332         tmp  = *src;
333         *src = *dst;
334         *dst = tmp;
335
336         skb->pkt_type = PACKET_OUTGOING;
337         dn_rt_finish_output(skb, NULL);
338         return NET_RX_SUCCESS;
339 }
340
341 /**
342  * dn_return_long - Return a long packet to its sender
343  * @skb: The long format packet to return
344  *
345  */
346 static int dn_return_long(struct sk_buff *skb)
347 {
348         struct dn_skb_cb *cb;
349         unsigned char *ptr;
350         unsigned char *src_addr, *dst_addr;
351         unsigned char tmp[ETH_ALEN];
352
353         /* Add back all headers */
354         skb_push(skb, skb->data - skb->nh.raw);
355
356         if ((skb = skb_unshare(skb, GFP_ATOMIC)) == NULL)
357                 return NET_RX_DROP;
358
359         cb = DN_SKB_CB(skb);
360         /* Ignore packet length and point to flags */
361         ptr = skb->data + 2;
362
363         /* Skip padding */
364         if (*ptr & DN_RT_F_PF) {
365                 char padlen = (*ptr & ~DN_RT_F_PF);
366                 ptr += padlen;
367         }
368
369         *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
370         ptr += 2;
371         dst_addr = ptr;
372         ptr += 8;
373         src_addr = ptr;
374         ptr += 6;
375         *ptr = 0; /* Zero hop count */
376
377         /* Swap source and destination */
378         memcpy(tmp, src_addr, ETH_ALEN);
379         memcpy(src_addr, dst_addr, ETH_ALEN);
380         memcpy(dst_addr, tmp, ETH_ALEN);
381
382         skb->pkt_type = PACKET_OUTGOING;
383         dn_rt_finish_output(skb, tmp);
384         return NET_RX_SUCCESS;
385 }
386
387 /**
388  * dn_route_rx_packet - Try and find a route for an incoming packet
389  * @skb: The packet to find a route for
390  *
391  * Returns: result of input function if route is found, error code otherwise
392  */
393 static int dn_route_rx_packet(struct sk_buff *skb)
394 {
395         struct dn_skb_cb *cb = DN_SKB_CB(skb);
396         int err;
397
398         if ((err = dn_route_input(skb)) == 0)
399                 return skb->dst->input(skb);
400
401         if (decnet_debug_level & 4) {
402                 char *devname = skb->dev ? skb->dev->name : "???";
403                 struct dn_skb_cb *cb = DN_SKB_CB(skb);
404                 printk(KERN_DEBUG
405                         "DECnet: dn_route_rx_packet: rt_flags=0x%02x dev=%s len=%d src=0x%04hx dst=0x%04hx err=%d type=%d\n",
406                         (int)cb->rt_flags, devname, skb->len, cb->src, cb->dst, 
407                         err, skb->pkt_type);
408         }
409
410         if ((skb->pkt_type == PACKET_HOST) && (cb->rt_flags & DN_RT_F_RQR)) {
411                 switch(cb->rt_flags & DN_RT_PKT_MSK) {
412                         case DN_RT_PKT_SHORT:
413                                 return dn_return_short(skb);
414                         case DN_RT_PKT_LONG:
415                                 return dn_return_long(skb);
416                 }
417         }
418
419         kfree_skb(skb);
420         return NET_RX_DROP;
421 }
422
423 static int dn_route_rx_long(struct sk_buff *skb)
424 {
425         struct dn_skb_cb *cb = DN_SKB_CB(skb);
426         unsigned char *ptr = skb->data;
427
428         if (skb->len < 21) /* 20 for long header, 1 for shortest nsp */
429                 goto drop_it;
430
431         skb_pull(skb, 20);
432         skb->h.raw = skb->data;
433
434         /* Destination info */
435         ptr += 2;
436         cb->dst = dn_htons(dn_eth2dn(ptr));
437         if (memcmp(ptr, dn_hiord_addr, 4) != 0)
438                 goto drop_it;
439         ptr += 6;
440
441
442         /* Source info */
443         ptr += 2;
444         cb->src = dn_htons(dn_eth2dn(ptr));
445         if (memcmp(ptr, dn_hiord_addr, 4) != 0)
446                 goto drop_it;
447         ptr += 6;
448         /* Other junk */
449         ptr++;
450         cb->hops = *ptr++; /* Visit Count */
451
452         return NF_HOOK(PF_DECnet, NF_DN_PRE_ROUTING, skb, skb->dev, NULL, dn_route_rx_packet);
453
454 drop_it:
455         kfree_skb(skb);
456         return NET_RX_DROP;
457 }
458
459
460
461 static int dn_route_rx_short(struct sk_buff *skb)
462 {
463         struct dn_skb_cb *cb = DN_SKB_CB(skb);
464         unsigned char *ptr = skb->data;
465
466         if (skb->len < 6) /* 5 for short header + 1 for shortest nsp */
467                 goto drop_it;
468
469         skb_pull(skb, 5);
470         skb->h.raw = skb->data;
471
472         cb->dst = *(dn_address *)ptr;
473         ptr += 2;
474         cb->src = *(dn_address *)ptr;
475         ptr += 2;
476         cb->hops = *ptr & 0x3f;
477
478         return NF_HOOK(PF_DECnet, NF_DN_PRE_ROUTING, skb, skb->dev, NULL, dn_route_rx_packet);
479
480 drop_it:
481         kfree_skb(skb);
482         return NET_RX_DROP;
483 }
484
485 static int dn_route_discard(struct sk_buff *skb)
486 {
487         /*
488          * I know we drop the packet here, but thats considered success in
489          * this case
490          */
491         kfree_skb(skb);
492         return NET_RX_SUCCESS;
493 }
494
495 static int dn_route_ptp_hello(struct sk_buff *skb)
496 {
497         dn_dev_hello(skb);
498         dn_neigh_pointopoint_hello(skb);
499         return NET_RX_SUCCESS;
500 }
501
502 int dn_route_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt)
503 {
504         struct dn_skb_cb *cb;
505         unsigned char flags = 0;
506         __u16 len = dn_ntohs(*(__u16 *)skb->data);
507         struct dn_dev *dn = (struct dn_dev *)dev->dn_ptr;
508         unsigned char padlen = 0;
509
510         if (dn == NULL)
511                 goto dump_it;
512
513         if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
514                 goto out;
515
516         skb_pull(skb, 2);
517
518         if (len > skb->len)
519                 goto dump_it;
520
521         skb_trim(skb, len);
522
523         flags = *skb->data;
524
525         cb = DN_SKB_CB(skb);
526         cb->stamp = jiffies;
527         cb->iif = dev->ifindex;
528
529         /*
530          * If we have padding, remove it.
531          */
532         if (flags & DN_RT_F_PF) {
533                 padlen = flags & ~DN_RT_F_PF;
534                 skb_pull(skb, padlen);
535                 flags = *skb->data;
536         }
537
538         skb->nh.raw = skb->data;
539
540         /*
541          * Weed out future version DECnet
542          */
543         if (flags & DN_RT_F_VER)
544                 goto dump_it;
545
546         cb->rt_flags = flags;
547
548         if (decnet_debug_level & 1)
549                 printk(KERN_DEBUG 
550                         "dn_route_rcv: got 0x%02x from %s [%d %d %d]\n",
551                         (int)flags, (dev) ? dev->name : "???", len, skb->len, 
552                         padlen);
553
554         if (flags & DN_RT_PKT_CNTL) {
555                 switch(flags & DN_RT_CNTL_MSK) {
556                         case DN_RT_PKT_INIT:
557                                 dn_dev_init_pkt(skb);
558                                 break;
559                         case DN_RT_PKT_VERI:
560                                 dn_dev_veri_pkt(skb);
561                                 break;
562                 }
563
564                 if (dn->parms.state != DN_DEV_S_RU)
565                         goto dump_it;
566
567                 switch(flags & DN_RT_CNTL_MSK) {
568                         case DN_RT_PKT_HELO:
569                                 return NF_HOOK(PF_DECnet, NF_DN_HELLO, skb, skb->dev, NULL, dn_route_ptp_hello);
570
571                         case DN_RT_PKT_L1RT:
572                         case DN_RT_PKT_L2RT:
573                                 return NF_HOOK(PF_DECnet, NF_DN_ROUTE, skb, skb->dev, NULL, dn_route_discard);
574                         case DN_RT_PKT_ERTH:
575                                 return NF_HOOK(PF_DECnet, NF_DN_HELLO, skb, skb->dev, NULL, dn_neigh_router_hello);
576
577                         case DN_RT_PKT_EEDH:
578                                 return NF_HOOK(PF_DECnet, NF_DN_HELLO, skb, skb->dev, NULL, dn_neigh_endnode_hello);
579                 }
580         } else {
581                 if (dn->parms.state != DN_DEV_S_RU)
582                         goto dump_it;
583
584                 skb_pull(skb, 1); /* Pull flags */
585
586                 switch(flags & DN_RT_PKT_MSK) {
587                         case DN_RT_PKT_LONG:
588                                 return dn_route_rx_long(skb);
589                         case DN_RT_PKT_SHORT:
590                                 return dn_route_rx_short(skb);
591                 }
592         }
593
594 dump_it:
595         kfree_skb(skb);
596 out:
597         return NET_RX_DROP;
598 }
599
600 static int dn_output(struct sk_buff *skb)
601 {
602         struct dst_entry *dst = skb->dst;
603         struct dn_route *rt = (struct dn_route *)dst;
604         struct net_device *dev = dst->dev;
605         struct dn_skb_cb *cb = DN_SKB_CB(skb);
606         struct neighbour *neigh;
607
608         int err = -EINVAL;
609
610         if ((neigh = dst->neighbour) == NULL)
611                 goto error;
612
613         skb->dev = dev;
614
615         cb->src = rt->rt_saddr;
616         cb->dst = rt->rt_daddr;
617
618         /*
619          * Always set the Intra-Ethernet bit on all outgoing packets
620          * originated on this node. Only valid flag from upper layers
621          * is return-to-sender-requested. Set hop count to 0 too.
622          */
623         cb->rt_flags &= ~DN_RT_F_RQR;
624         cb->rt_flags |= DN_RT_F_IE;
625         cb->hops = 0;
626
627         return NF_HOOK(PF_DECnet, NF_DN_LOCAL_OUT, skb, NULL, dev, neigh->output);
628
629 error:
630         if (net_ratelimit())
631                 printk(KERN_DEBUG "dn_output: This should not happen\n");
632
633         kfree_skb(skb);
634
635         return err;
636 }
637
638 #ifdef CONFIG_DECNET_ROUTER
639 static int dn_forward(struct sk_buff *skb)
640 {
641         struct dn_skb_cb *cb = DN_SKB_CB(skb);
642         struct dst_entry *dst = skb->dst;
643         struct neighbour *neigh;
644         struct net_device *dev = skb->dev;
645         int err = -EINVAL;
646
647         if ((neigh = dst->neighbour) == NULL)
648                 goto error;
649
650         /*
651          * Hop count exceeded.
652          */
653         err = NET_RX_DROP;
654         if (++cb->hops > 30)
655                 goto drop;
656
657         skb->dev = dst->dev;
658
659         /*
660          * If packet goes out same interface it came in on, then set
661          * the Intra-Ethernet bit. This has no effect for short
662          * packets, so we don't need to test for them here.
663          */
664         if (cb->iif == dst->dev->ifindex)
665                 cb->rt_flags |= DN_RT_F_IE;
666         else
667                 cb->rt_flags &= ~DN_RT_F_IE;
668
669         return NF_HOOK(PF_DECnet, NF_DN_FORWARD, skb, dev, skb->dev, neigh->output);
670
671
672 error:
673         if (net_ratelimit())
674                 printk(KERN_DEBUG "dn_forward: This should not happen\n");
675 drop:
676         kfree_skb(skb);
677
678         return err;
679 }
680 #endif
681
682 /*
683  * Drop packet. This is used for endnodes and for
684  * when we should not be forwarding packets from
685  * this dest.
686  */
687 static int dn_blackhole(struct sk_buff *skb)
688 {
689         kfree_skb(skb);
690         return NET_RX_DROP;
691 }
692
693 /*
694  * Used to catch bugs. This should never normally get
695  * called.
696  */
697 static int dn_rt_bug(struct sk_buff *skb)
698 {
699         if (net_ratelimit()) {
700                 struct dn_skb_cb *cb = DN_SKB_CB(skb);
701
702                 printk(KERN_DEBUG "dn_rt_bug: skb from:%04x to:%04x\n",
703                                 cb->src, cb->dst);
704         }
705
706         kfree_skb(skb);
707
708         return NET_RX_BAD;
709 }
710
711 static int dn_route_output_slow(struct dst_entry **pprt, dn_address dst, dn_address src, int flags)
712 {
713         struct dn_route *rt = NULL;
714         struct net_device *dev = decnet_default_device;
715         struct neighbour *neigh = NULL;
716         struct dn_dev *dn_db;
717         unsigned hash;
718 #ifdef CONFIG_DECNET_ROUTER
719         struct dn_fib_key key;
720         struct dn_fib_res res;
721         int err;
722
723         key.src = src;
724         key.dst = dst;
725         key.iif = 0;
726         key.oif = 0;
727         key.fwmark = 0;
728         key.scope = RT_SCOPE_UNIVERSE;
729
730         if ((err = dn_fib_lookup(&key, &res)) == 0) {
731                 switch(res.type) {
732                         case RTN_UNICAST:
733                                 /*
734                                  * This method of handling multipath
735                                  * routes is a hack and will change.
736                                  * It works for now though.
737                                  */
738                                 if (res.fi->fib_nhs)
739                                         dn_fib_select_multipath(&key, &res);
740                                 neigh = __neigh_lookup(&dn_neigh_table, &DN_FIB_RES_GW(res), DN_FIB_RES_DEV(res), 1);
741                                 err = -ENOBUFS;
742                                 if (!neigh)
743                                         break;
744                                 err = 0;
745                                 break;
746                         case RTN_UNREACHABLE:
747                                 err = -EHOSTUNREACH;
748                                 break;
749                         default:
750                                 err = -EINVAL;
751                 }
752                 dn_fib_res_put(&res);
753                 if (err < 0)
754                         return err;
755                 goto got_route;
756         }
757
758         if (err != -ESRCH)
759                 return err;
760 #endif 
761
762         /* Look in On-Ethernet cache first */
763         if (!(flags & MSG_TRYHARD)) {
764                 if ((neigh = dn_neigh_lookup(&dn_neigh_table, &dst)) != NULL)
765                         goto got_route;
766         }
767
768         if (dev == NULL)
769                 return -EINVAL;
770
771         dn_db = dev->dn_ptr;
772
773         if (dn_db == NULL)
774                 return -EINVAL;
775
776         /* Try default router */
777         if ((neigh = neigh_clone(dn_db->router)) != NULL)
778                 goto got_route;
779
780         /* Send to default device (and hope for the best) if above fail */
781         if ((neigh = __neigh_lookup(&dn_neigh_table, &dst, dev, 1)) != NULL)
782                 goto got_route;
783
784
785         return -EINVAL;
786
787 got_route:
788
789         if ((rt = dst_alloc(&dn_dst_ops)) == NULL) {
790                 neigh_release(neigh);
791                 return -EINVAL;
792         }
793
794         dn_db = (struct dn_dev *)neigh->dev->dn_ptr;
795         
796         rt->key.saddr  = src;
797         rt->rt_saddr   = src;
798         rt->key.daddr  = dst;
799         rt->rt_daddr   = dst;
800         rt->key.oif    = neigh ? neigh->dev->ifindex : -1;
801         rt->key.iif    = 0;
802         rt->key.fwmark = 0;
803
804         rt->u.dst.neighbour = neigh;
805         rt->u.dst.dev = neigh ? neigh->dev : NULL;
806         if (rt->u.dst.dev)
807                 dev_hold(rt->u.dst.dev);
808         rt->u.dst.lastuse = jiffies;
809         rt->u.dst.output = dn_output;
810         rt->u.dst.input  = dn_rt_bug;
811
812         if (dn_dev_islocal(neigh->dev, rt->rt_daddr))
813                 rt->u.dst.input = dn_nsp_rx;
814
815         hash = dn_hash(rt->key.saddr, rt->key.daddr);
816         dn_insert_route(rt, hash);
817         *pprt = &rt->u.dst;
818
819         return 0;
820 }
821
822 int dn_route_output(struct dst_entry **pprt, dn_address dst, dn_address src, int flags)
823 {
824         unsigned hash = dn_hash(src, dst);
825         struct dn_route *rt = NULL;
826
827         if (!(flags & MSG_TRYHARD)) {
828                 read_lock_bh(&dn_rt_hash_table[hash].lock);
829                 for(rt = dn_rt_hash_table[hash].chain; rt; rt = rt->u.rt_next) {
830                         if ((dst == rt->key.daddr) &&
831                                         (src == rt->key.saddr) &&
832                                         (rt->key.iif == 0) &&
833                                         (rt->key.oif != 0)) {
834                                 rt->u.dst.lastuse = jiffies;
835                                 dst_hold(&rt->u.dst);
836                                 rt->u.dst.__use++;
837                                 read_unlock_bh(&dn_rt_hash_table[hash].lock);
838                                 *pprt = &rt->u.dst;
839                                 return 0;
840                         }
841                 }
842                 read_unlock_bh(&dn_rt_hash_table[hash].lock);
843         }
844
845         return dn_route_output_slow(pprt, dst, src, flags);
846 }
847
848 static int dn_route_input_slow(struct sk_buff *skb)
849 {
850         struct dn_route *rt = NULL;
851         struct dn_skb_cb *cb = DN_SKB_CB(skb);
852         struct net_device *dev = skb->dev;
853         struct dn_dev *dn_db;
854         struct neighbour *neigh = NULL;
855         int (*dnrt_input)(struct sk_buff *skb);
856         int (*dnrt_output)(struct sk_buff *skb);
857         u32 fwmark = 0;
858         unsigned hash;
859         dn_address saddr = cb->src;
860         dn_address daddr = cb->dst;
861 #ifdef CONFIG_DECNET_ROUTER
862         struct dn_fib_key key;
863         struct dn_fib_res res;
864         int err;
865 #endif
866
867         if (dev == NULL)
868                 return -EINVAL;
869
870         if ((dn_db = dev->dn_ptr) == NULL)
871                 return -EINVAL;
872
873         /*
874          * In this case we've just received a packet from a source
875          * outside ourselves pretending to come from us. We don't
876          * allow it any further to prevent routing loops, spoofing and
877          * other nasties. Loopback packets already have the dst attached
878          * so this only affects packets which have originated elsewhere.
879          */
880         if (dn_dev_islocal(dev, cb->src))
881                 return -ENOTUNIQ;
882
883         /*
884          * Default is to create a drop everything entry
885          */
886         dnrt_input  = dn_blackhole;
887         dnrt_output = dn_rt_bug;
888
889         /*
890          * Is the destination us ?
891          */
892         if (!dn_dev_islocal(dev, cb->dst))
893                 goto non_local_input;
894
895         /*
896          * Local input... find source of skb
897          */
898         dnrt_input  = dn_nsp_rx;
899         dnrt_output = dn_output;
900         saddr = cb->dst;
901         daddr = cb->src;
902
903         if ((neigh = neigh_lookup(&dn_neigh_table, &cb->src, dev)) != NULL)
904                 goto add_entry;
905
906         if (dn_db->router && ((neigh = neigh_clone(dn_db->router)) != NULL))
907                 goto add_entry;
908
909         neigh = neigh_create(&dn_neigh_table, &cb->src, dev);
910         if (!IS_ERR(neigh)) {
911                 if (dev->type == ARPHRD_ETHER)
912                         memcpy(neigh->ha, skb->mac.ethernet->h_source, ETH_ALEN);
913                 goto add_entry;
914         }
915
916         return PTR_ERR(neigh);
917
918 non_local_input:
919
920 #ifdef CONFIG_DECNET_ROUTER
921         /*
922          * Destination is another node... find next hop in
923          * routing table here.
924          */
925
926         key.src = cb->src;
927         key.dst = cb->dst;
928         key.iif = dev->ifindex;
929         key.oif = 0;
930         key.scope = RT_SCOPE_UNIVERSE;
931
932 #ifdef CONFIG_DECNET_ROUTE_FWMARK
933         key.fwmark = skb->nfmark;
934 #else
935         key.fwmark = 0;
936 #endif
937
938         if ((err = dn_fib_lookup(&key, &res)) == 0) {
939                 switch(res.type) {
940                         case RTN_UNICAST:
941                                 if (res.fi->fib_nhs)
942                                         dn_fib_select_multipath(&key, &res);
943                                 neigh = __neigh_lookup(&dn_neigh_table, &DN_FIB_RES_GW(res), DN_FIB_RES_DEV(res), 1);
944                                 err = -ENOBUFS;
945                                 if (!neigh)
946                                         break;
947                                 err = 0;
948                                 dnrt_input = dn_forward;
949                                 fwmark = key.fwmark;
950                                 break;
951                         case RTN_UNREACHABLE:
952                                 dnrt_input = dn_blackhole;
953                                 fwmark = key.fwmark;
954                                 break;
955                         default:
956                                 err = -EINVAL;
957                 }
958                 dn_fib_res_put(&res);
959                 if (err < 0)
960                         return err;
961                 goto add_entry;
962         }
963
964         return err;
965
966 #endif /* CONFIG_DECNET_ROUTER */
967
968 add_entry:
969
970         if ((rt = dst_alloc(&dn_dst_ops)) == NULL) {
971                 neigh_release(neigh);
972                 return -EINVAL;
973         }
974
975         rt->key.saddr  = cb->src;
976         rt->rt_saddr   = saddr;
977         rt->key.daddr  = cb->dst;
978         rt->rt_daddr   = daddr;
979         rt->key.oif    = 0;
980         rt->key.iif    = dev->ifindex;
981         rt->key.fwmark = fwmark;
982
983         rt->u.dst.neighbour = neigh;
984         rt->u.dst.dev = neigh ? neigh->dev : NULL;
985         if (rt->u.dst.dev)
986                 dev_hold(rt->u.dst.dev);
987         rt->u.dst.lastuse = jiffies;
988         rt->u.dst.output = dnrt_output;
989         rt->u.dst.input = dnrt_input;
990
991         hash = dn_hash(rt->key.saddr, rt->key.daddr);
992         dn_insert_route(rt, hash);
993         skb->dst = (struct dst_entry *)rt;
994
995         return 0;
996 }
997
998 int dn_route_input(struct sk_buff *skb)
999 {
1000         struct dn_route *rt;
1001         struct dn_skb_cb *cb = DN_SKB_CB(skb);
1002         unsigned hash = dn_hash(cb->src, cb->dst);
1003
1004         if (skb->dst)
1005                 return 0;
1006
1007         read_lock(&dn_rt_hash_table[hash].lock);
1008         for(rt = dn_rt_hash_table[hash].chain; rt != NULL; rt = rt->u.rt_next) {
1009                 if ((rt->key.saddr == cb->src) &&
1010                                 (rt->key.daddr == cb->dst) &&
1011                                 (rt->key.oif == 0) &&
1012 #ifdef CONFIG_DECNET_ROUTE_FWMARK
1013                                 (rt->key.fwmark == skb->nfmark) &&
1014 #endif
1015                                 (rt->key.iif == cb->iif)) {
1016                         rt->u.dst.lastuse = jiffies;
1017                         dst_hold(&rt->u.dst);
1018                         rt->u.dst.__use++;
1019                         read_unlock(&dn_rt_hash_table[hash].lock);
1020                         skb->dst = (struct dst_entry *)rt;
1021                         return 0;
1022                 }
1023         }
1024         read_unlock(&dn_rt_hash_table[hash].lock);
1025
1026         return dn_route_input_slow(skb);
1027 }
1028
1029 static int dn_rt_fill_info(struct sk_buff *skb, u32 pid, u32 seq, int event, int nowait)
1030 {
1031         struct dn_route *rt = (struct dn_route *)skb->dst;
1032         struct rtmsg *r;
1033         struct nlmsghdr *nlh;
1034         unsigned char *b = skb->tail;
1035
1036         nlh = NLMSG_PUT(skb, pid, seq, event, sizeof(*r));
1037         r = NLMSG_DATA(nlh);
1038         nlh->nlmsg_flags = nowait ? NLM_F_MULTI : 0;
1039         r->rtm_family = AF_DECnet;
1040         r->rtm_dst_len = 16;
1041         r->rtm_src_len = 16;
1042         r->rtm_tos = 0;
1043         r->rtm_table = 0;
1044         r->rtm_type = 0;
1045         r->rtm_flags = 0;
1046         r->rtm_scope = RT_SCOPE_UNIVERSE;
1047         r->rtm_protocol = RTPROT_UNSPEC;
1048         RTA_PUT(skb, RTA_DST, 2, &rt->rt_daddr);
1049         RTA_PUT(skb, RTA_SRC, 2, &rt->rt_saddr);
1050         if (rt->u.dst.dev)
1051                 RTA_PUT(skb, RTA_OIF, sizeof(int), &rt->u.dst.dev->ifindex);
1052         if (rt->u.dst.window)
1053                 RTA_PUT(skb, RTAX_WINDOW, sizeof(unsigned), &rt->u.dst.window);
1054         if (rt->u.dst.rtt)
1055                 RTA_PUT(skb, RTAX_RTT, sizeof(unsigned), &rt->u.dst.rtt);
1056
1057         nlh->nlmsg_len = skb->tail - b;
1058         return skb->len;
1059
1060 nlmsg_failure:
1061 rtattr_failure:
1062         skb_trim(skb, b - skb->data);
1063         return -1;
1064 }
1065
1066 /*
1067  * This is called by both endnodes and routers now.
1068  */
1069 int dn_cache_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh, void *arg)
1070 {
1071         struct rtattr **rta = arg;
1072         struct dn_route *rt = NULL;
1073         struct dn_skb_cb *cb;
1074         dn_address dst = 0;
1075         dn_address src = 0;
1076         int iif = 0;
1077         int err;
1078         struct sk_buff *skb;
1079
1080         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1081         if (skb == NULL)
1082                 return -ENOBUFS;
1083         skb->mac.raw = skb->data;
1084         cb = DN_SKB_CB(skb);
1085
1086         if (rta[RTA_SRC-1])
1087                 memcpy(&src, RTA_DATA(rta[RTA_SRC-1]), 2);
1088         if (rta[RTA_DST-1])
1089                 memcpy(&dst, RTA_DATA(rta[RTA_DST-1]), 2);
1090         if (rta[RTA_IIF-1])
1091                 memcpy(&iif, RTA_DATA(rta[RTA_IIF-1]), sizeof(int));
1092
1093         if (iif) {
1094                 struct net_device *dev;
1095                 if ((dev = dev_get_by_index(iif)) == NULL) {
1096                         kfree_skb(skb);
1097                         return -ENODEV;
1098                 }
1099                 if (!dev->dn_ptr) {
1100                         dev_put(dev);
1101                         kfree_skb(skb);
1102                         return -ENODEV;
1103                 }
1104                 skb->protocol = __constant_htons(ETH_P_DNA_RT);
1105                 skb->dev = dev;
1106                 cb->src = src;
1107                 cb->dst = dst;
1108                 local_bh_disable();
1109                 err = dn_route_input(skb);
1110                 local_bh_enable();
1111                 memset(cb, 0, sizeof(struct dn_skb_cb));
1112                 rt = (struct dn_route *)skb->dst;
1113         } else {
1114                 err = dn_route_output((struct dst_entry **)&rt, dst, src, 0);
1115         }
1116
1117         if (!err && rt->u.dst.error)
1118                 err = rt->u.dst.error;
1119         if (skb->dev)
1120                 dev_put(skb->dev);
1121         skb->dev = NULL;
1122         if (err)
1123                 goto out_free;
1124         skb->dst = &rt->u.dst;
1125
1126         NETLINK_CB(skb).dst_pid = NETLINK_CB(in_skb).pid;
1127
1128         err = dn_rt_fill_info(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq, RTM_NEWROUTE, 0);
1129
1130         if (err == 0)
1131                 goto out_free;
1132         if (err < 0) {
1133                 err = -EMSGSIZE;
1134                 goto out_free;
1135         }
1136
1137         err = netlink_unicast(rtnl, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1138
1139         return err;
1140
1141 out_free:
1142         kfree_skb(skb);
1143         return err;
1144 }
1145
1146 /*
1147  * For routers, this is called from dn_fib_dump, but for endnodes its
1148  * called directly from the rtnetlink dispatch table.
1149  */
1150 int dn_cache_dump(struct sk_buff *skb, struct netlink_callback *cb)
1151 {
1152         struct dn_route *rt;
1153         int h, s_h;
1154         int idx, s_idx;
1155
1156         if (NLMSG_PAYLOAD(cb->nlh, 0) < sizeof(struct rtmsg))
1157                 return -EINVAL;
1158         if (!(((struct rtmsg *)NLMSG_DATA(cb->nlh))->rtm_flags&RTM_F_CLONED))
1159                 return 0;
1160
1161         s_h = cb->args[0];
1162         s_idx = idx = cb->args[1];
1163         for(h = 0; h <= dn_rt_hash_mask; h++) {
1164                 if (h < s_h)
1165                         continue;
1166                 if (h > s_h)
1167                         s_idx = 0;
1168                 read_lock_bh(&dn_rt_hash_table[h].lock);
1169                 for(rt = dn_rt_hash_table[h].chain, idx = 0; rt; rt = rt->u.rt_next, idx++) {
1170                         if (idx < s_idx)
1171                                 continue;
1172                         skb->dst = dst_clone(&rt->u.dst);
1173                         if (dn_rt_fill_info(skb, NETLINK_CB(cb->skb).pid,
1174                                         cb->nlh->nlmsg_seq, RTM_NEWROUTE, 1) <= 0) {
1175                                 dst_release(xchg(&skb->dst, NULL));
1176                                 read_unlock_bh(&dn_rt_hash_table[h].lock);
1177                                 goto done;
1178                         }
1179                         dst_release(xchg(&skb->dst, NULL));
1180                 }
1181                 read_unlock_bh(&dn_rt_hash_table[h].lock);
1182         }
1183
1184 done:
1185         cb->args[0] = h;
1186         cb->args[1] = idx;
1187         return skb->len;
1188 }
1189
1190 #ifdef CONFIG_PROC_FS
1191
1192 static int decnet_cache_get_info(char *buffer, char **start, off_t offset, int length)
1193 {
1194         int len = 0;
1195         off_t pos = 0;
1196         off_t begin = 0;
1197         struct dn_route *rt;
1198         int i;
1199         char buf1[DN_ASCBUF_LEN], buf2[DN_ASCBUF_LEN];
1200
1201         for(i = 0; i <= dn_rt_hash_mask; i++) {
1202                 read_lock_bh(&dn_rt_hash_table[i].lock);
1203                 rt = dn_rt_hash_table[i].chain;
1204                 for(; rt != NULL; rt = rt->u.rt_next) {
1205                         len += sprintf(buffer + len, "%-8s %-7s %-7s %04d %04d %04d\n",
1206                                         rt->u.dst.dev ? rt->u.dst.dev->name : "*",
1207                                         dn_addr2asc(dn_ntohs(rt->rt_daddr), buf1),
1208                                         dn_addr2asc(dn_ntohs(rt->rt_saddr), buf2),
1209                                         atomic_read(&rt->u.dst.__refcnt),
1210                                         rt->u.dst.__use,
1211                                         (int)rt->u.dst.rtt
1212                                         );
1213
1214
1215
1216                         pos = begin + len;
1217         
1218                         if (pos < offset) {
1219                                 len   = 0;
1220                                 begin = pos;
1221                         }
1222                         if (pos > offset + length)
1223                                 break;
1224                 }
1225                 read_unlock_bh(&dn_rt_hash_table[i].lock);
1226                 if (pos > offset + length)
1227                         break;
1228         }
1229
1230         *start = buffer + (offset - begin);
1231         len   -= (offset - begin);
1232
1233         if (len > length) len = length;
1234
1235         return(len);
1236
1237
1238 #endif /* CONFIG_PROC_FS */
1239
1240 void __init dn_route_init(void)
1241 {
1242         int i, goal, order;
1243
1244         dn_dst_ops.kmem_cachep = kmem_cache_create("dn_dst_cache",
1245                                                    sizeof(struct dn_route),
1246                                                    0, SLAB_HWCACHE_ALIGN,
1247                                                    NULL, NULL);
1248
1249         if (!dn_dst_ops.kmem_cachep)
1250                 panic("DECnet: Failed to allocate dn_dst_cache\n");
1251
1252         dn_route_timer.function = dn_dst_check_expire;
1253         dn_route_timer.expires = jiffies + decnet_dst_gc_interval * HZ;
1254         add_timer(&dn_route_timer);
1255
1256         goal = num_physpages >> (26 - PAGE_SHIFT);
1257
1258         for(order = 0; (1UL << order) < goal; order++)
1259                 /* NOTHING */;
1260
1261         /*
1262          * Only want 1024 entries max, since the table is very, very unlikely
1263          * to be larger than that.
1264          */
1265         while(order && ((((1UL << order) * PAGE_SIZE) / 
1266                                 sizeof(struct dn_rt_hash_bucket)) >= 2048))
1267                 order--;
1268
1269         do {
1270                 dn_rt_hash_mask = (1UL << order) * PAGE_SIZE /
1271                         sizeof(struct dn_rt_hash_bucket);
1272                 while(dn_rt_hash_mask & (dn_rt_hash_mask - 1))
1273                         dn_rt_hash_mask--;
1274                 dn_rt_hash_table = (struct dn_rt_hash_bucket *)
1275                         __get_free_pages(GFP_ATOMIC, order);
1276         } while (dn_rt_hash_table == NULL && --order > 0);
1277
1278         if (!dn_rt_hash_table)
1279                 panic("Failed to allocate DECnet route cache hash table\n");
1280
1281         printk(KERN_INFO 
1282                 "DECnet: Routing cache hash table of %u buckets, %ldKbytes\n", 
1283                 dn_rt_hash_mask, 
1284                 (long)(dn_rt_hash_mask*sizeof(struct dn_rt_hash_bucket))/1024);
1285
1286         dn_rt_hash_mask--;
1287         for(i = 0; i <= dn_rt_hash_mask; i++) {
1288                 dn_rt_hash_table[i].lock = RW_LOCK_UNLOCKED;
1289                 dn_rt_hash_table[i].chain = NULL;
1290         }
1291
1292         dn_dst_ops.gc_thresh = (dn_rt_hash_mask + 1);
1293
1294 #ifdef CONFIG_PROC_FS
1295         proc_net_create("decnet_cache",0,decnet_cache_get_info);
1296 #endif /* CONFIG_PROC_FS */
1297 }
1298
1299 void __exit dn_route_cleanup(void)
1300 {
1301         del_timer(&dn_route_timer);
1302         dn_run_flush(0);
1303
1304         proc_net_remove("decnet_cache");
1305 }
1306