9383d3e5a1abb2d8916704c6f7e6a0a4fed0a32f
[linux-2.6.git] / net / decnet / dn_route.c
1 /*
2  * DECnet       An implementation of the DECnet protocol suite for the LINUX
3  *              operating system.  DECnet is implemented using the  BSD Socket
4  *              interface as the means of communication with the user level.
5  *
6  *              DECnet Routing Functions (Endnode and Router)
7  *
8  * Authors:     Steve Whitehouse <SteveW@ACM.org>
9  *              Eduardo Marcelo Serrat <emserrat@geocities.com>
10  *
11  * Changes:
12  *              Steve Whitehouse : Fixes to allow "intra-ethernet" and
13  *                                 "return-to-sender" bits on outgoing
14  *                                 packets.
15  *              Steve Whitehouse : Timeouts for cached routes.
16  *              Steve Whitehouse : Use dst cache for input routes too.
17  *              Steve Whitehouse : Fixed error values in dn_send_skb.
18  *              Steve Whitehouse : Rework routing functions to better fit
19  *                                 DECnet routing design
20  *              Alexey Kuznetsov : New SMP locking
21  *              Steve Whitehouse : More SMP locking changes & dn_cache_dump()
22  *              Steve Whitehouse : Prerouting NF hook, now really is prerouting.
23  *                                 Fixed possible skb leak in rtnetlink funcs.
24  *              Steve Whitehouse : Dave Miller's dynamic hash table sizing and
25  *                                 Alexey Kuznetsov's finer grained locking
26  *                                 from ipv4/route.c.
27  *              Steve Whitehouse : Routing is now starting to look like a
28  *                                 sensible set of code now, mainly due to
29  *                                 my copying the IPv4 routing code. The
30  *                                 hooks here are modified and will continue
31  *                                 to evolve for a while.
32  *              Steve Whitehouse : Real SMP at last :-) Also new netfilter
33  *                                 stuff. Look out raw sockets your days
34  *                                 are numbered!
35  *              Steve Whitehouse : Added return-to-sender functions. Added
36  *                                 backlog congestion level return codes.
37  *              Steve Whitehouse : Fixed bug where routes were set up with
38  *                                 no ref count on net devices.
39  *              Steve Whitehouse : RCU for the route cache
40  *              Steve Whitehouse : Preparations for the flow cache
41  *              Steve Whitehouse : Prepare for nonlinear skbs
42  */
43
44 /******************************************************************************
45     (c) 1995-1998 E.M. Serrat           emserrat@geocities.com
46
47     This program is free software; you can redistribute it and/or modify
48     it under the terms of the GNU General Public License as published by
49     the Free Software Foundation; either version 2 of the License, or
50     any later version.
51
52     This program is distributed in the hope that it will be useful,
53     but WITHOUT ANY WARRANTY; without even the implied warranty of
54     MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
55     GNU General Public License for more details.
56 *******************************************************************************/
57
58 #include <linux/errno.h>
59 #include <linux/types.h>
60 #include <linux/socket.h>
61 #include <linux/in.h>
62 #include <linux/kernel.h>
63 #include <linux/sockios.h>
64 #include <linux/net.h>
65 #include <linux/netdevice.h>
66 #include <linux/inet.h>
67 #include <linux/route.h>
68 #include <linux/in_route.h>
69 #include <net/sock.h>
70 #include <linux/mm.h>
71 #include <linux/proc_fs.h>
72 #include <linux/seq_file.h>
73 #include <linux/init.h>
74 #include <linux/rtnetlink.h>
75 #include <linux/string.h>
76 #include <linux/netfilter_decnet.h>
77 #include <linux/rcupdate.h>
78 #include <linux/times.h>
79 #include <asm/errno.h>
80 #include <net/net_namespace.h>
81 #include <net/netlink.h>
82 #include <net/neighbour.h>
83 #include <net/dst.h>
84 #include <net/flow.h>
85 #include <net/fib_rules.h>
86 #include <net/dn.h>
87 #include <net/dn_dev.h>
88 #include <net/dn_nsp.h>
89 #include <net/dn_route.h>
90 #include <net/dn_neigh.h>
91 #include <net/dn_fib.h>
92
93 struct dn_rt_hash_bucket
94 {
95         struct dn_route *chain;
96         spinlock_t lock;
97 };
98
99 extern struct neigh_table dn_neigh_table;
100
101
102 static unsigned char dn_hiord_addr[6] = {0xAA,0x00,0x04,0x00,0x00,0x00};
103
104 static const int dn_rt_min_delay = 2 * HZ;
105 static const int dn_rt_max_delay = 10 * HZ;
106 static const int dn_rt_mtu_expires = 10 * 60 * HZ;
107
108 static unsigned long dn_rt_deadline;
109
110 static int dn_dst_gc(struct dst_ops *ops);
111 static struct dst_entry *dn_dst_check(struct dst_entry *, __u32);
112 static struct dst_entry *dn_dst_negative_advice(struct dst_entry *);
113 static void dn_dst_link_failure(struct sk_buff *);
114 static void dn_dst_update_pmtu(struct dst_entry *dst, u32 mtu);
115 static int dn_route_input(struct sk_buff *);
116 static void dn_run_flush(unsigned long dummy);
117
118 static struct dn_rt_hash_bucket *dn_rt_hash_table;
119 static unsigned dn_rt_hash_mask;
120
121 static struct timer_list dn_route_timer;
122 static DEFINE_TIMER(dn_rt_flush_timer, dn_run_flush, 0, 0);
123 int decnet_dst_gc_interval = 2;
124
125 static struct dst_ops dn_dst_ops = {
126         .family =               PF_DECnet,
127         .protocol =             cpu_to_be16(ETH_P_DNA_RT),
128         .gc_thresh =            128,
129         .gc =                   dn_dst_gc,
130         .check =                dn_dst_check,
131         .negative_advice =      dn_dst_negative_advice,
132         .link_failure =         dn_dst_link_failure,
133         .update_pmtu =          dn_dst_update_pmtu,
134         .entries =              ATOMIC_INIT(0),
135 };
136
137 static __inline__ unsigned dn_hash(__le16 src, __le16 dst)
138 {
139         __u16 tmp = (__u16 __force)(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         dst_release(&rt->u.dst);
154         call_rcu_bh(&rt->u.dst.rcu_head, dst_rcu_free);
155 }
156
157 static void dn_dst_check_expire(unsigned long dummy)
158 {
159         int i;
160         struct dn_route *rt, **rtp;
161         unsigned long now = jiffies;
162         unsigned long expire = 120 * HZ;
163
164         for(i = 0; i <= dn_rt_hash_mask; i++) {
165                 rtp = &dn_rt_hash_table[i].chain;
166
167                 spin_lock(&dn_rt_hash_table[i].lock);
168                 while((rt=*rtp) != NULL) {
169                         if (atomic_read(&rt->u.dst.__refcnt) ||
170                                         (now - rt->u.dst.lastuse) < expire) {
171                                 rtp = &rt->u.dst.dn_next;
172                                 continue;
173                         }
174                         *rtp = rt->u.dst.dn_next;
175                         rt->u.dst.dn_next = NULL;
176                         dnrt_free(rt);
177                 }
178                 spin_unlock(&dn_rt_hash_table[i].lock);
179
180                 if ((jiffies - now) > 0)
181                         break;
182         }
183
184         mod_timer(&dn_route_timer, now + decnet_dst_gc_interval * HZ);
185 }
186
187 static int dn_dst_gc(struct dst_ops *ops)
188 {
189         struct dn_route *rt, **rtp;
190         int i;
191         unsigned long now = jiffies;
192         unsigned long expire = 10 * HZ;
193
194         for(i = 0; i <= dn_rt_hash_mask; i++) {
195
196                 spin_lock_bh(&dn_rt_hash_table[i].lock);
197                 rtp = &dn_rt_hash_table[i].chain;
198
199                 while((rt=*rtp) != NULL) {
200                         if (atomic_read(&rt->u.dst.__refcnt) ||
201                                         (now - rt->u.dst.lastuse) < expire) {
202                                 rtp = &rt->u.dst.dn_next;
203                                 continue;
204                         }
205                         *rtp = rt->u.dst.dn_next;
206                         rt->u.dst.dn_next = NULL;
207                         dnrt_drop(rt);
208                         break;
209                 }
210                 spin_unlock_bh(&dn_rt_hash_table[i].lock);
211         }
212
213         return 0;
214 }
215
216 /*
217  * The decnet standards don't impose a particular minimum mtu, what they
218  * do insist on is that the routing layer accepts a datagram of at least
219  * 230 bytes long. Here we have to subtract the routing header length from
220  * 230 to get the minimum acceptable mtu. If there is no neighbour, then we
221  * assume the worst and use a long header size.
222  *
223  * We update both the mtu and the advertised mss (i.e. the segment size we
224  * advertise to the other end).
225  */
226 static void dn_dst_update_pmtu(struct dst_entry *dst, u32 mtu)
227 {
228         u32 min_mtu = 230;
229         struct dn_dev *dn = dst->neighbour ?
230                             (struct dn_dev *)dst->neighbour->dev->dn_ptr : NULL;
231
232         if (dn && dn->use_long == 0)
233                 min_mtu -= 6;
234         else
235                 min_mtu -= 21;
236
237         if (dst_metric(dst, RTAX_MTU) > mtu && mtu >= min_mtu) {
238                 if (!(dst_metric_locked(dst, RTAX_MTU))) {
239                         dst->metrics[RTAX_MTU-1] = mtu;
240                         dst_set_expires(dst, dn_rt_mtu_expires);
241                 }
242                 if (!(dst_metric_locked(dst, RTAX_ADVMSS))) {
243                         u32 mss = mtu - DN_MAX_NSP_DATA_HEADER;
244                         if (dst_metric(dst, RTAX_ADVMSS) > mss)
245                                 dst->metrics[RTAX_ADVMSS-1] = mss;
246                 }
247         }
248 }
249
250 /*
251  * When a route has been marked obsolete. (e.g. routing cache flush)
252  */
253 static struct dst_entry *dn_dst_check(struct dst_entry *dst, __u32 cookie)
254 {
255         return NULL;
256 }
257
258 static struct dst_entry *dn_dst_negative_advice(struct dst_entry *dst)
259 {
260         dst_release(dst);
261         return NULL;
262 }
263
264 static void dn_dst_link_failure(struct sk_buff *skb)
265 {
266         return;
267 }
268
269 static inline int compare_keys(struct flowi *fl1, struct flowi *fl2)
270 {
271         return ((fl1->nl_u.dn_u.daddr ^ fl2->nl_u.dn_u.daddr) |
272                 (fl1->nl_u.dn_u.saddr ^ fl2->nl_u.dn_u.saddr) |
273                 (fl1->mark ^ fl2->mark) |
274                 (fl1->nl_u.dn_u.scope ^ fl2->nl_u.dn_u.scope) |
275                 (fl1->oif ^ fl2->oif) |
276                 (fl1->iif ^ fl2->iif)) == 0;
277 }
278
279 static int dn_insert_route(struct dn_route *rt, unsigned hash, struct dn_route **rp)
280 {
281         struct dn_route *rth, **rthp;
282         unsigned long now = jiffies;
283
284         rthp = &dn_rt_hash_table[hash].chain;
285
286         spin_lock_bh(&dn_rt_hash_table[hash].lock);
287         while((rth = *rthp) != NULL) {
288                 if (compare_keys(&rth->fl, &rt->fl)) {
289                         /* Put it first */
290                         *rthp = rth->u.dst.dn_next;
291                         rcu_assign_pointer(rth->u.dst.dn_next,
292                                            dn_rt_hash_table[hash].chain);
293                         rcu_assign_pointer(dn_rt_hash_table[hash].chain, rth);
294
295                         dst_use(&rth->u.dst, 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.dst.dn_next;
303         }
304
305         rcu_assign_pointer(rt->u.dst.dn_next, dn_rt_hash_table[hash].chain);
306         rcu_assign_pointer(dn_rt_hash_table[hash].chain, rt);
307
308         dst_use(&rt->u.dst, now);
309         spin_unlock_bh(&dn_rt_hash_table[hash].lock);
310         *rp = rt;
311         return 0;
312 }
313
314 static void dn_run_flush(unsigned long dummy)
315 {
316         int i;
317         struct dn_route *rt, *next;
318
319         for(i = 0; i < dn_rt_hash_mask; i++) {
320                 spin_lock_bh(&dn_rt_hash_table[i].lock);
321
322                 if ((rt = xchg(&dn_rt_hash_table[i].chain, NULL)) == NULL)
323                         goto nothing_to_declare;
324
325                 for(; rt; rt=next) {
326                         next = rt->u.dst.dn_next;
327                         rt->u.dst.dn_next = NULL;
328                         dst_free((struct dst_entry *)rt);
329                 }
330
331 nothing_to_declare:
332                 spin_unlock_bh(&dn_rt_hash_table[i].lock);
333         }
334 }
335
336 static DEFINE_SPINLOCK(dn_rt_flush_lock);
337
338 void dn_rt_cache_flush(int delay)
339 {
340         unsigned long now = jiffies;
341         int user_mode = !in_interrupt();
342
343         if (delay < 0)
344                 delay = dn_rt_min_delay;
345
346         spin_lock_bh(&dn_rt_flush_lock);
347
348         if (del_timer(&dn_rt_flush_timer) && delay > 0 && dn_rt_deadline) {
349                 long tmo = (long)(dn_rt_deadline - now);
350
351                 if (user_mode && tmo < dn_rt_max_delay - dn_rt_min_delay)
352                         tmo = 0;
353
354                 if (delay > tmo)
355                         delay = tmo;
356         }
357
358         if (delay <= 0) {
359                 spin_unlock_bh(&dn_rt_flush_lock);
360                 dn_run_flush(0);
361                 return;
362         }
363
364         if (dn_rt_deadline == 0)
365                 dn_rt_deadline = now + dn_rt_max_delay;
366
367         dn_rt_flush_timer.expires = now + delay;
368         add_timer(&dn_rt_flush_timer);
369         spin_unlock_bh(&dn_rt_flush_lock);
370 }
371
372 /**
373  * dn_return_short - Return a short packet to its sender
374  * @skb: The packet to return
375  *
376  */
377 static int dn_return_short(struct sk_buff *skb)
378 {
379         struct dn_skb_cb *cb;
380         unsigned char *ptr;
381         __le16 *src;
382         __le16 *dst;
383
384         /* Add back headers */
385         skb_push(skb, skb->data - skb_network_header(skb));
386
387         if ((skb = skb_unshare(skb, GFP_ATOMIC)) == NULL)
388                 return NET_RX_DROP;
389
390         cb = DN_SKB_CB(skb);
391         /* Skip packet length and point to flags */
392         ptr = skb->data + 2;
393         *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
394
395         dst = (__le16 *)ptr;
396         ptr += 2;
397         src = (__le16 *)ptr;
398         ptr += 2;
399         *ptr = 0; /* Zero hop count */
400
401         swap(*src, *dst);
402
403         skb->pkt_type = PACKET_OUTGOING;
404         dn_rt_finish_output(skb, NULL, NULL);
405         return NET_RX_SUCCESS;
406 }
407
408 /**
409  * dn_return_long - Return a long packet to its sender
410  * @skb: The long format packet to return
411  *
412  */
413 static int dn_return_long(struct sk_buff *skb)
414 {
415         struct dn_skb_cb *cb;
416         unsigned char *ptr;
417         unsigned char *src_addr, *dst_addr;
418         unsigned char tmp[ETH_ALEN];
419
420         /* Add back all headers */
421         skb_push(skb, skb->data - skb_network_header(skb));
422
423         if ((skb = skb_unshare(skb, GFP_ATOMIC)) == NULL)
424                 return NET_RX_DROP;
425
426         cb = DN_SKB_CB(skb);
427         /* Ignore packet length and point to flags */
428         ptr = skb->data + 2;
429
430         /* Skip padding */
431         if (*ptr & DN_RT_F_PF) {
432                 char padlen = (*ptr & ~DN_RT_F_PF);
433                 ptr += padlen;
434         }
435
436         *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
437         ptr += 2;
438         dst_addr = ptr;
439         ptr += 8;
440         src_addr = ptr;
441         ptr += 6;
442         *ptr = 0; /* Zero hop count */
443
444         /* Swap source and destination */
445         memcpy(tmp, src_addr, ETH_ALEN);
446         memcpy(src_addr, dst_addr, ETH_ALEN);
447         memcpy(dst_addr, tmp, ETH_ALEN);
448
449         skb->pkt_type = PACKET_OUTGOING;
450         dn_rt_finish_output(skb, dst_addr, src_addr);
451         return NET_RX_SUCCESS;
452 }
453
454 /**
455  * dn_route_rx_packet - Try and find a route for an incoming packet
456  * @skb: The packet to find a route for
457  *
458  * Returns: result of input function if route is found, error code otherwise
459  */
460 static int dn_route_rx_packet(struct sk_buff *skb)
461 {
462         struct dn_skb_cb *cb = DN_SKB_CB(skb);
463         int err;
464
465         if ((err = dn_route_input(skb)) == 0)
466                 return dst_input(skb);
467
468         if (decnet_debug_level & 4) {
469                 char *devname = skb->dev ? skb->dev->name : "???";
470                 struct dn_skb_cb *cb = DN_SKB_CB(skb);
471                 printk(KERN_DEBUG
472                         "DECnet: dn_route_rx_packet: rt_flags=0x%02x dev=%s len=%d src=0x%04hx dst=0x%04hx err=%d type=%d\n",
473                         (int)cb->rt_flags, devname, skb->len,
474                         le16_to_cpu(cb->src), le16_to_cpu(cb->dst),
475                         err, skb->pkt_type);
476         }
477
478         if ((skb->pkt_type == PACKET_HOST) && (cb->rt_flags & DN_RT_F_RQR)) {
479                 switch(cb->rt_flags & DN_RT_PKT_MSK) {
480                         case DN_RT_PKT_SHORT:
481                                 return dn_return_short(skb);
482                         case DN_RT_PKT_LONG:
483                                 return dn_return_long(skb);
484                 }
485         }
486
487         kfree_skb(skb);
488         return NET_RX_DROP;
489 }
490
491 static int dn_route_rx_long(struct sk_buff *skb)
492 {
493         struct dn_skb_cb *cb = DN_SKB_CB(skb);
494         unsigned char *ptr = skb->data;
495
496         if (!pskb_may_pull(skb, 21)) /* 20 for long header, 1 for shortest nsp */
497                 goto drop_it;
498
499         skb_pull(skb, 20);
500         skb_reset_transport_header(skb);
501
502         /* Destination info */
503         ptr += 2;
504         cb->dst = dn_eth2dn(ptr);
505         if (memcmp(ptr, dn_hiord_addr, 4) != 0)
506                 goto drop_it;
507         ptr += 6;
508
509
510         /* Source info */
511         ptr += 2;
512         cb->src = dn_eth2dn(ptr);
513         if (memcmp(ptr, dn_hiord_addr, 4) != 0)
514                 goto drop_it;
515         ptr += 6;
516         /* Other junk */
517         ptr++;
518         cb->hops = *ptr++; /* Visit Count */
519
520         return NF_HOOK(PF_DECnet, NF_DN_PRE_ROUTING, skb, skb->dev, NULL, dn_route_rx_packet);
521
522 drop_it:
523         kfree_skb(skb);
524         return NET_RX_DROP;
525 }
526
527
528
529 static int dn_route_rx_short(struct sk_buff *skb)
530 {
531         struct dn_skb_cb *cb = DN_SKB_CB(skb);
532         unsigned char *ptr = skb->data;
533
534         if (!pskb_may_pull(skb, 6)) /* 5 for short header + 1 for shortest nsp */
535                 goto drop_it;
536
537         skb_pull(skb, 5);
538         skb_reset_transport_header(skb);
539
540         cb->dst = *(__le16 *)ptr;
541         ptr += 2;
542         cb->src = *(__le16 *)ptr;
543         ptr += 2;
544         cb->hops = *ptr & 0x3f;
545
546         return NF_HOOK(PF_DECnet, NF_DN_PRE_ROUTING, skb, skb->dev, NULL, dn_route_rx_packet);
547
548 drop_it:
549         kfree_skb(skb);
550         return NET_RX_DROP;
551 }
552
553 static int dn_route_discard(struct sk_buff *skb)
554 {
555         /*
556          * I know we drop the packet here, but thats considered success in
557          * this case
558          */
559         kfree_skb(skb);
560         return NET_RX_SUCCESS;
561 }
562
563 static int dn_route_ptp_hello(struct sk_buff *skb)
564 {
565         dn_dev_hello(skb);
566         dn_neigh_pointopoint_hello(skb);
567         return NET_RX_SUCCESS;
568 }
569
570 int dn_route_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
571 {
572         struct dn_skb_cb *cb;
573         unsigned char flags = 0;
574         __u16 len = le16_to_cpu(*(__le16 *)skb->data);
575         struct dn_dev *dn = (struct dn_dev *)dev->dn_ptr;
576         unsigned char padlen = 0;
577
578         if (!net_eq(dev_net(dev), &init_net))
579                 goto dump_it;
580
581         if (dn == NULL)
582                 goto dump_it;
583
584         if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
585                 goto out;
586
587         if (!pskb_may_pull(skb, 3))
588                 goto dump_it;
589
590         skb_pull(skb, 2);
591
592         if (len > skb->len)
593                 goto dump_it;
594
595         skb_trim(skb, len);
596
597         flags = *skb->data;
598
599         cb = DN_SKB_CB(skb);
600         cb->stamp = jiffies;
601         cb->iif = dev->ifindex;
602
603         /*
604          * If we have padding, remove it.
605          */
606         if (flags & DN_RT_F_PF) {
607                 padlen = flags & ~DN_RT_F_PF;
608                 if (!pskb_may_pull(skb, padlen + 1))
609                         goto dump_it;
610                 skb_pull(skb, padlen);
611                 flags = *skb->data;
612         }
613
614         skb_reset_network_header(skb);
615
616         /*
617          * Weed out future version DECnet
618          */
619         if (flags & DN_RT_F_VER)
620                 goto dump_it;
621
622         cb->rt_flags = flags;
623
624         if (decnet_debug_level & 1)
625                 printk(KERN_DEBUG
626                         "dn_route_rcv: got 0x%02x from %s [%d %d %d]\n",
627                         (int)flags, (dev) ? dev->name : "???", len, skb->len,
628                         padlen);
629
630         if (flags & DN_RT_PKT_CNTL) {
631                 if (unlikely(skb_linearize(skb)))
632                         goto dump_it;
633
634                 switch(flags & DN_RT_CNTL_MSK) {
635                         case DN_RT_PKT_INIT:
636                                 dn_dev_init_pkt(skb);
637                                 break;
638                         case DN_RT_PKT_VERI:
639                                 dn_dev_veri_pkt(skb);
640                                 break;
641                 }
642
643                 if (dn->parms.state != DN_DEV_S_RU)
644                         goto dump_it;
645
646                 switch(flags & DN_RT_CNTL_MSK) {
647                         case DN_RT_PKT_HELO:
648                                 return NF_HOOK(PF_DECnet, NF_DN_HELLO, skb, skb->dev, NULL, dn_route_ptp_hello);
649
650                         case DN_RT_PKT_L1RT:
651                         case DN_RT_PKT_L2RT:
652                                 return NF_HOOK(PF_DECnet, NF_DN_ROUTE, skb, skb->dev, NULL, dn_route_discard);
653                         case DN_RT_PKT_ERTH:
654                                 return NF_HOOK(PF_DECnet, NF_DN_HELLO, skb, skb->dev, NULL, dn_neigh_router_hello);
655
656                         case DN_RT_PKT_EEDH:
657                                 return NF_HOOK(PF_DECnet, NF_DN_HELLO, skb, skb->dev, NULL, dn_neigh_endnode_hello);
658                 }
659         } else {
660                 if (dn->parms.state != DN_DEV_S_RU)
661                         goto dump_it;
662
663                 skb_pull(skb, 1); /* Pull flags */
664
665                 switch(flags & DN_RT_PKT_MSK) {
666                         case DN_RT_PKT_LONG:
667                                 return dn_route_rx_long(skb);
668                         case DN_RT_PKT_SHORT:
669                                 return dn_route_rx_short(skb);
670                 }
671         }
672
673 dump_it:
674         kfree_skb(skb);
675 out:
676         return NET_RX_DROP;
677 }
678
679 static int dn_output(struct sk_buff *skb)
680 {
681         struct dst_entry *dst = skb_dst(skb);
682         struct dn_route *rt = (struct dn_route *)dst;
683         struct net_device *dev = dst->dev;
684         struct dn_skb_cb *cb = DN_SKB_CB(skb);
685         struct neighbour *neigh;
686
687         int err = -EINVAL;
688
689         if ((neigh = dst->neighbour) == NULL)
690                 goto error;
691
692         skb->dev = dev;
693
694         cb->src = rt->rt_saddr;
695         cb->dst = rt->rt_daddr;
696
697         /*
698          * Always set the Intra-Ethernet bit on all outgoing packets
699          * originated on this node. Only valid flag from upper layers
700          * is return-to-sender-requested. Set hop count to 0 too.
701          */
702         cb->rt_flags &= ~DN_RT_F_RQR;
703         cb->rt_flags |= DN_RT_F_IE;
704         cb->hops = 0;
705
706         return NF_HOOK(PF_DECnet, NF_DN_LOCAL_OUT, skb, NULL, dev, neigh->output);
707
708 error:
709         if (net_ratelimit())
710                 printk(KERN_DEBUG "dn_output: This should not happen\n");
711
712         kfree_skb(skb);
713
714         return err;
715 }
716
717 static int dn_forward(struct sk_buff *skb)
718 {
719         struct dn_skb_cb *cb = DN_SKB_CB(skb);
720         struct dst_entry *dst = skb_dst(skb);
721         struct dn_dev *dn_db = dst->dev->dn_ptr;
722         struct dn_route *rt;
723         struct neighbour *neigh = dst->neighbour;
724         int header_len;
725 #ifdef CONFIG_NETFILTER
726         struct net_device *dev = skb->dev;
727 #endif
728
729         if (skb->pkt_type != PACKET_HOST)
730                 goto drop;
731
732         /* Ensure that we have enough space for headers */
733         rt = (struct dn_route *)skb_dst(skb);
734         header_len = dn_db->use_long ? 21 : 6;
735         if (skb_cow(skb, LL_RESERVED_SPACE(rt->u.dst.dev)+header_len))
736                 goto drop;
737
738         /*
739          * Hop count exceeded.
740          */
741         if (++cb->hops > 30)
742                 goto drop;
743
744         skb->dev = rt->u.dst.dev;
745
746         /*
747          * If packet goes out same interface it came in on, then set
748          * the Intra-Ethernet bit. This has no effect for short
749          * packets, so we don't need to test for them here.
750          */
751         cb->rt_flags &= ~DN_RT_F_IE;
752         if (rt->rt_flags & RTCF_DOREDIRECT)
753                 cb->rt_flags |= DN_RT_F_IE;
754
755         return NF_HOOK(PF_DECnet, NF_DN_FORWARD, skb, dev, skb->dev, neigh->output);
756
757 drop:
758         kfree_skb(skb);
759         return NET_RX_DROP;
760 }
761
762 /*
763  * Used to catch bugs. This should never normally get
764  * called.
765  */
766 static int dn_rt_bug(struct sk_buff *skb)
767 {
768         if (net_ratelimit()) {
769                 struct dn_skb_cb *cb = DN_SKB_CB(skb);
770
771                 printk(KERN_DEBUG "dn_rt_bug: skb from:%04x to:%04x\n",
772                                 le16_to_cpu(cb->src), le16_to_cpu(cb->dst));
773         }
774
775         kfree_skb(skb);
776
777         return NET_RX_DROP;
778 }
779
780 static int dn_rt_set_next_hop(struct dn_route *rt, struct dn_fib_res *res)
781 {
782         struct dn_fib_info *fi = res->fi;
783         struct net_device *dev = rt->u.dst.dev;
784         struct neighbour *n;
785         unsigned mss;
786
787         if (fi) {
788                 if (DN_FIB_RES_GW(*res) &&
789                     DN_FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
790                         rt->rt_gateway = DN_FIB_RES_GW(*res);
791                 memcpy(rt->u.dst.metrics, fi->fib_metrics,
792                        sizeof(rt->u.dst.metrics));
793         }
794         rt->rt_type = res->type;
795
796         if (dev != NULL && rt->u.dst.neighbour == NULL) {
797                 n = __neigh_lookup_errno(&dn_neigh_table, &rt->rt_gateway, dev);
798                 if (IS_ERR(n))
799                         return PTR_ERR(n);
800                 rt->u.dst.neighbour = n;
801         }
802
803         if (dst_metric(&rt->u.dst, RTAX_MTU) == 0 ||
804             dst_metric(&rt->u.dst, RTAX_MTU) > rt->u.dst.dev->mtu)
805                 rt->u.dst.metrics[RTAX_MTU-1] = rt->u.dst.dev->mtu;
806         mss = dn_mss_from_pmtu(dev, dst_mtu(&rt->u.dst));
807         if (dst_metric(&rt->u.dst, RTAX_ADVMSS) == 0 ||
808             dst_metric(&rt->u.dst, RTAX_ADVMSS) > mss)
809                 rt->u.dst.metrics[RTAX_ADVMSS-1] = mss;
810         return 0;
811 }
812
813 static inline int dn_match_addr(__le16 addr1, __le16 addr2)
814 {
815         __u16 tmp = le16_to_cpu(addr1) ^ le16_to_cpu(addr2);
816         int match = 16;
817         while(tmp) {
818                 tmp >>= 1;
819                 match--;
820         }
821         return match;
822 }
823
824 static __le16 dnet_select_source(const struct net_device *dev, __le16 daddr, int scope)
825 {
826         __le16 saddr = 0;
827         struct dn_dev *dn_db = dev->dn_ptr;
828         struct dn_ifaddr *ifa;
829         int best_match = 0;
830         int ret;
831
832         read_lock(&dev_base_lock);
833         for(ifa = dn_db->ifa_list; ifa; ifa = ifa->ifa_next) {
834                 if (ifa->ifa_scope > scope)
835                         continue;
836                 if (!daddr) {
837                         saddr = ifa->ifa_local;
838                         break;
839                 }
840                 ret = dn_match_addr(daddr, ifa->ifa_local);
841                 if (ret > best_match)
842                         saddr = ifa->ifa_local;
843                 if (best_match == 0)
844                         saddr = ifa->ifa_local;
845         }
846         read_unlock(&dev_base_lock);
847
848         return saddr;
849 }
850
851 static inline __le16 __dn_fib_res_prefsrc(struct dn_fib_res *res)
852 {
853         return dnet_select_source(DN_FIB_RES_DEV(*res), DN_FIB_RES_GW(*res), res->scope);
854 }
855
856 static inline __le16 dn_fib_rules_map_destination(__le16 daddr, struct dn_fib_res *res)
857 {
858         __le16 mask = dnet_make_mask(res->prefixlen);
859         return (daddr&~mask)|res->fi->fib_nh->nh_gw;
860 }
861
862 static int dn_route_output_slow(struct dst_entry **pprt, const struct flowi *oldflp, int try_hard)
863 {
864         struct flowi fl = { .nl_u = { .dn_u =
865                                       { .daddr = oldflp->fld_dst,
866                                         .saddr = oldflp->fld_src,
867                                         .scope = RT_SCOPE_UNIVERSE,
868                                      } },
869                             .mark = oldflp->mark,
870                             .iif = init_net.loopback_dev->ifindex,
871                             .oif = oldflp->oif };
872         struct dn_route *rt = NULL;
873         struct net_device *dev_out = NULL, *dev;
874         struct neighbour *neigh = NULL;
875         unsigned hash;
876         unsigned flags = 0;
877         struct dn_fib_res res = { .fi = NULL, .type = RTN_UNICAST };
878         int err;
879         int free_res = 0;
880         __le16 gateway = 0;
881
882         if (decnet_debug_level & 16)
883                 printk(KERN_DEBUG
884                        "dn_route_output_slow: dst=%04x src=%04x mark=%d"
885                        " iif=%d oif=%d\n", le16_to_cpu(oldflp->fld_dst),
886                        le16_to_cpu(oldflp->fld_src),
887                        oldflp->mark, init_net.loopback_dev->ifindex, oldflp->oif);
888
889         /* If we have an output interface, verify its a DECnet device */
890         if (oldflp->oif) {
891                 dev_out = dev_get_by_index(&init_net, oldflp->oif);
892                 err = -ENODEV;
893                 if (dev_out && dev_out->dn_ptr == NULL) {
894                         dev_put(dev_out);
895                         dev_out = NULL;
896                 }
897                 if (dev_out == NULL)
898                         goto out;
899         }
900
901         /* If we have a source address, verify that its a local address */
902         if (oldflp->fld_src) {
903                 err = -EADDRNOTAVAIL;
904
905                 if (dev_out) {
906                         if (dn_dev_islocal(dev_out, oldflp->fld_src))
907                                 goto source_ok;
908                         dev_put(dev_out);
909                         goto out;
910                 }
911                 read_lock(&dev_base_lock);
912                 for_each_netdev(&init_net, dev) {
913                         if (!dev->dn_ptr)
914                                 continue;
915                         if (!dn_dev_islocal(dev, oldflp->fld_src))
916                                 continue;
917                         if ((dev->flags & IFF_LOOPBACK) &&
918                             oldflp->fld_dst &&
919                             !dn_dev_islocal(dev, oldflp->fld_dst))
920                                 continue;
921
922                         dev_out = dev;
923                         break;
924                 }
925                 read_unlock(&dev_base_lock);
926                 if (dev_out == NULL)
927                         goto out;
928                 dev_hold(dev_out);
929 source_ok:
930                 ;
931         }
932
933         /* No destination? Assume its local */
934         if (!fl.fld_dst) {
935                 fl.fld_dst = fl.fld_src;
936
937                 err = -EADDRNOTAVAIL;
938                 if (dev_out)
939                         dev_put(dev_out);
940                 dev_out = init_net.loopback_dev;
941                 dev_hold(dev_out);
942                 if (!fl.fld_dst) {
943                         fl.fld_dst =
944                         fl.fld_src = dnet_select_source(dev_out, 0,
945                                                        RT_SCOPE_HOST);
946                         if (!fl.fld_dst)
947                                 goto out;
948                 }
949                 fl.oif = init_net.loopback_dev->ifindex;
950                 res.type = RTN_LOCAL;
951                 goto make_route;
952         }
953
954         if (decnet_debug_level & 16)
955                 printk(KERN_DEBUG
956                        "dn_route_output_slow: initial checks complete."
957                        " dst=%o4x src=%04x oif=%d try_hard=%d\n",
958                        le16_to_cpu(fl.fld_dst), le16_to_cpu(fl.fld_src),
959                        fl.oif, try_hard);
960
961         /*
962          * N.B. If the kernel is compiled without router support then
963          * dn_fib_lookup() will evaluate to non-zero so this if () block
964          * will always be executed.
965          */
966         err = -ESRCH;
967         if (try_hard || (err = dn_fib_lookup(&fl, &res)) != 0) {
968                 struct dn_dev *dn_db;
969                 if (err != -ESRCH)
970                         goto out;
971                 /*
972                  * Here the fallback is basically the standard algorithm for
973                  * routing in endnodes which is described in the DECnet routing
974                  * docs
975                  *
976                  * If we are not trying hard, look in neighbour cache.
977                  * The result is tested to ensure that if a specific output
978                  * device/source address was requested, then we honour that
979                  * here
980                  */
981                 if (!try_hard) {
982                         neigh = neigh_lookup_nodev(&dn_neigh_table, &init_net, &fl.fld_dst);
983                         if (neigh) {
984                                 if ((oldflp->oif &&
985                                     (neigh->dev->ifindex != oldflp->oif)) ||
986                                     (oldflp->fld_src &&
987                                     (!dn_dev_islocal(neigh->dev,
988                                                       oldflp->fld_src)))) {
989                                         neigh_release(neigh);
990                                         neigh = NULL;
991                                 } else {
992                                         if (dev_out)
993                                                 dev_put(dev_out);
994                                         if (dn_dev_islocal(neigh->dev, fl.fld_dst)) {
995                                                 dev_out = init_net.loopback_dev;
996                                                 res.type = RTN_LOCAL;
997                                         } else {
998                                                 dev_out = neigh->dev;
999                                         }
1000                                         dev_hold(dev_out);
1001                                         goto select_source;
1002                                 }
1003                         }
1004                 }
1005
1006                 /* Not there? Perhaps its a local address */
1007                 if (dev_out == NULL)
1008                         dev_out = dn_dev_get_default();
1009                 err = -ENODEV;
1010                 if (dev_out == NULL)
1011                         goto out;
1012                 dn_db = dev_out->dn_ptr;
1013                 /* Possible improvement - check all devices for local addr */
1014                 if (dn_dev_islocal(dev_out, fl.fld_dst)) {
1015                         dev_put(dev_out);
1016                         dev_out = init_net.loopback_dev;
1017                         dev_hold(dev_out);
1018                         res.type = RTN_LOCAL;
1019                         goto select_source;
1020                 }
1021                 /* Not local either.... try sending it to the default router */
1022                 neigh = neigh_clone(dn_db->router);
1023                 BUG_ON(neigh && neigh->dev != dev_out);
1024
1025                 /* Ok then, we assume its directly connected and move on */
1026 select_source:
1027                 if (neigh)
1028                         gateway = ((struct dn_neigh *)neigh)->addr;
1029                 if (gateway == 0)
1030                         gateway = fl.fld_dst;
1031                 if (fl.fld_src == 0) {
1032                         fl.fld_src = dnet_select_source(dev_out, gateway,
1033                                                          res.type == RTN_LOCAL ?
1034                                                          RT_SCOPE_HOST :
1035                                                          RT_SCOPE_LINK);
1036                         if (fl.fld_src == 0 && res.type != RTN_LOCAL)
1037                                 goto e_addr;
1038                 }
1039                 fl.oif = dev_out->ifindex;
1040                 goto make_route;
1041         }
1042         free_res = 1;
1043
1044         if (res.type == RTN_NAT)
1045                 goto e_inval;
1046
1047         if (res.type == RTN_LOCAL) {
1048                 if (!fl.fld_src)
1049                         fl.fld_src = fl.fld_dst;
1050                 if (dev_out)
1051                         dev_put(dev_out);
1052                 dev_out = init_net.loopback_dev;
1053                 dev_hold(dev_out);
1054                 fl.oif = dev_out->ifindex;
1055                 if (res.fi)
1056                         dn_fib_info_put(res.fi);
1057                 res.fi = NULL;
1058                 goto make_route;
1059         }
1060
1061         if (res.fi->fib_nhs > 1 && fl.oif == 0)
1062                 dn_fib_select_multipath(&fl, &res);
1063
1064         /*
1065          * We could add some logic to deal with default routes here and
1066          * get rid of some of the special casing above.
1067          */
1068
1069         if (!fl.fld_src)
1070                 fl.fld_src = DN_FIB_RES_PREFSRC(res);
1071
1072         if (dev_out)
1073                 dev_put(dev_out);
1074         dev_out = DN_FIB_RES_DEV(res);
1075         dev_hold(dev_out);
1076         fl.oif = dev_out->ifindex;
1077         gateway = DN_FIB_RES_GW(res);
1078
1079 make_route:
1080         if (dev_out->flags & IFF_LOOPBACK)
1081                 flags |= RTCF_LOCAL;
1082
1083         rt = dst_alloc(&dn_dst_ops);
1084         if (rt == NULL)
1085                 goto e_nobufs;
1086
1087         atomic_set(&rt->u.dst.__refcnt, 1);
1088         rt->u.dst.flags   = DST_HOST;
1089
1090         rt->fl.fld_src    = oldflp->fld_src;
1091         rt->fl.fld_dst    = oldflp->fld_dst;
1092         rt->fl.oif        = oldflp->oif;
1093         rt->fl.iif        = 0;
1094         rt->fl.mark       = oldflp->mark;
1095
1096         rt->rt_saddr      = fl.fld_src;
1097         rt->rt_daddr      = fl.fld_dst;
1098         rt->rt_gateway    = gateway ? gateway : fl.fld_dst;
1099         rt->rt_local_src  = fl.fld_src;
1100
1101         rt->rt_dst_map    = fl.fld_dst;
1102         rt->rt_src_map    = fl.fld_src;
1103
1104         rt->u.dst.dev = dev_out;
1105         dev_hold(dev_out);
1106         rt->u.dst.neighbour = neigh;
1107         neigh = NULL;
1108
1109         rt->u.dst.lastuse = jiffies;
1110         rt->u.dst.output  = dn_output;
1111         rt->u.dst.input   = dn_rt_bug;
1112         rt->rt_flags      = flags;
1113         if (flags & RTCF_LOCAL)
1114                 rt->u.dst.input = dn_nsp_rx;
1115
1116         err = dn_rt_set_next_hop(rt, &res);
1117         if (err)
1118                 goto e_neighbour;
1119
1120         hash = dn_hash(rt->fl.fld_src, rt->fl.fld_dst);
1121         dn_insert_route(rt, hash, (struct dn_route **)pprt);
1122
1123 done:
1124         if (neigh)
1125                 neigh_release(neigh);
1126         if (free_res)
1127                 dn_fib_res_put(&res);
1128         if (dev_out)
1129                 dev_put(dev_out);
1130 out:
1131         return err;
1132
1133 e_addr:
1134         err = -EADDRNOTAVAIL;
1135         goto done;
1136 e_inval:
1137         err = -EINVAL;
1138         goto done;
1139 e_nobufs:
1140         err = -ENOBUFS;
1141         goto done;
1142 e_neighbour:
1143         dst_free(&rt->u.dst);
1144         goto e_nobufs;
1145 }
1146
1147
1148 /*
1149  * N.B. The flags may be moved into the flowi at some future stage.
1150  */
1151 static int __dn_route_output_key(struct dst_entry **pprt, const struct flowi *flp, int flags)
1152 {
1153         unsigned hash = dn_hash(flp->fld_src, flp->fld_dst);
1154         struct dn_route *rt = NULL;
1155
1156         if (!(flags & MSG_TRYHARD)) {
1157                 rcu_read_lock_bh();
1158                 for(rt = rcu_dereference(dn_rt_hash_table[hash].chain); rt;
1159                         rt = rcu_dereference(rt->u.dst.dn_next)) {
1160                         if ((flp->fld_dst == rt->fl.fld_dst) &&
1161                             (flp->fld_src == rt->fl.fld_src) &&
1162                             (flp->mark == rt->fl.mark) &&
1163                             (rt->fl.iif == 0) &&
1164                             (rt->fl.oif == flp->oif)) {
1165                                 dst_use(&rt->u.dst, jiffies);
1166                                 rcu_read_unlock_bh();
1167                                 *pprt = &rt->u.dst;
1168                                 return 0;
1169                         }
1170                 }
1171                 rcu_read_unlock_bh();
1172         }
1173
1174         return dn_route_output_slow(pprt, flp, flags);
1175 }
1176
1177 static int dn_route_output_key(struct dst_entry **pprt, struct flowi *flp, int flags)
1178 {
1179         int err;
1180
1181         err = __dn_route_output_key(pprt, flp, flags);
1182         if (err == 0 && flp->proto) {
1183                 err = xfrm_lookup(&init_net, pprt, flp, NULL, 0);
1184         }
1185         return err;
1186 }
1187
1188 int dn_route_output_sock(struct dst_entry **pprt, struct flowi *fl, struct sock *sk, int flags)
1189 {
1190         int err;
1191
1192         err = __dn_route_output_key(pprt, fl, flags & MSG_TRYHARD);
1193         if (err == 0 && fl->proto) {
1194                 err = xfrm_lookup(&init_net, pprt, fl, sk,
1195                                  (flags & MSG_DONTWAIT) ? 0 : XFRM_LOOKUP_WAIT);
1196         }
1197         return err;
1198 }
1199
1200 static int dn_route_input_slow(struct sk_buff *skb)
1201 {
1202         struct dn_route *rt = NULL;
1203         struct dn_skb_cb *cb = DN_SKB_CB(skb);
1204         struct net_device *in_dev = skb->dev;
1205         struct net_device *out_dev = NULL;
1206         struct dn_dev *dn_db;
1207         struct neighbour *neigh = NULL;
1208         unsigned hash;
1209         int flags = 0;
1210         __le16 gateway = 0;
1211         __le16 local_src = 0;
1212         struct flowi fl = { .nl_u = { .dn_u =
1213                                      { .daddr = cb->dst,
1214                                        .saddr = cb->src,
1215                                        .scope = RT_SCOPE_UNIVERSE,
1216                                     } },
1217                             .mark = skb->mark,
1218                             .iif = skb->dev->ifindex };
1219         struct dn_fib_res res = { .fi = NULL, .type = RTN_UNREACHABLE };
1220         int err = -EINVAL;
1221         int free_res = 0;
1222
1223         dev_hold(in_dev);
1224
1225         if ((dn_db = in_dev->dn_ptr) == NULL)
1226                 goto out;
1227
1228         /* Zero source addresses are not allowed */
1229         if (fl.fld_src == 0)
1230                 goto out;
1231
1232         /*
1233          * In this case we've just received a packet from a source
1234          * outside ourselves pretending to come from us. We don't
1235          * allow it any further to prevent routing loops, spoofing and
1236          * other nasties. Loopback packets already have the dst attached
1237          * so this only affects packets which have originated elsewhere.
1238          */
1239         err  = -ENOTUNIQ;
1240         if (dn_dev_islocal(in_dev, cb->src))
1241                 goto out;
1242
1243         err = dn_fib_lookup(&fl, &res);
1244         if (err) {
1245                 if (err != -ESRCH)
1246                         goto out;
1247                 /*
1248                  * Is the destination us ?
1249                  */
1250                 if (!dn_dev_islocal(in_dev, cb->dst))
1251                         goto e_inval;
1252
1253                 res.type = RTN_LOCAL;
1254         } else {
1255                 __le16 src_map = fl.fld_src;
1256                 free_res = 1;
1257
1258                 out_dev = DN_FIB_RES_DEV(res);
1259                 if (out_dev == NULL) {
1260                         if (net_ratelimit())
1261                                 printk(KERN_CRIT "Bug in dn_route_input_slow() "
1262                                                  "No output device\n");
1263                         goto e_inval;
1264                 }
1265                 dev_hold(out_dev);
1266
1267                 if (res.r)
1268                         src_map = fl.fld_src; /* no NAT support for now */
1269
1270                 gateway = DN_FIB_RES_GW(res);
1271                 if (res.type == RTN_NAT) {
1272                         fl.fld_dst = dn_fib_rules_map_destination(fl.fld_dst, &res);
1273                         dn_fib_res_put(&res);
1274                         free_res = 0;
1275                         if (dn_fib_lookup(&fl, &res))
1276                                 goto e_inval;
1277                         free_res = 1;
1278                         if (res.type != RTN_UNICAST)
1279                                 goto e_inval;
1280                         flags |= RTCF_DNAT;
1281                         gateway = fl.fld_dst;
1282                 }
1283                 fl.fld_src = src_map;
1284         }
1285
1286         switch(res.type) {
1287         case RTN_UNICAST:
1288                 /*
1289                  * Forwarding check here, we only check for forwarding
1290                  * being turned off, if you want to only forward intra
1291                  * area, its up to you to set the routing tables up
1292                  * correctly.
1293                  */
1294                 if (dn_db->parms.forwarding == 0)
1295                         goto e_inval;
1296
1297                 if (res.fi->fib_nhs > 1 && fl.oif == 0)
1298                         dn_fib_select_multipath(&fl, &res);
1299
1300                 /*
1301                  * Check for out_dev == in_dev. We use the RTCF_DOREDIRECT
1302                  * flag as a hint to set the intra-ethernet bit when
1303                  * forwarding. If we've got NAT in operation, we don't do
1304                  * this optimisation.
1305                  */
1306                 if (out_dev == in_dev && !(flags & RTCF_NAT))
1307                         flags |= RTCF_DOREDIRECT;
1308
1309                 local_src = DN_FIB_RES_PREFSRC(res);
1310
1311         case RTN_BLACKHOLE:
1312         case RTN_UNREACHABLE:
1313                 break;
1314         case RTN_LOCAL:
1315                 flags |= RTCF_LOCAL;
1316                 fl.fld_src = cb->dst;
1317                 fl.fld_dst = cb->src;
1318
1319                 /* Routing tables gave us a gateway */
1320                 if (gateway)
1321                         goto make_route;
1322
1323                 /* Packet was intra-ethernet, so we know its on-link */
1324                 if (cb->rt_flags & DN_RT_F_IE) {
1325                         gateway = cb->src;
1326                         flags |= RTCF_DIRECTSRC;
1327                         goto make_route;
1328                 }
1329
1330                 /* Use the default router if there is one */
1331                 neigh = neigh_clone(dn_db->router);
1332                 if (neigh) {
1333                         gateway = ((struct dn_neigh *)neigh)->addr;
1334                         goto make_route;
1335                 }
1336
1337                 /* Close eyes and pray */
1338                 gateway = cb->src;
1339                 flags |= RTCF_DIRECTSRC;
1340                 goto make_route;
1341         default:
1342                 goto e_inval;
1343         }
1344
1345 make_route:
1346         rt = dst_alloc(&dn_dst_ops);
1347         if (rt == NULL)
1348                 goto e_nobufs;
1349
1350         rt->rt_saddr      = fl.fld_src;
1351         rt->rt_daddr      = fl.fld_dst;
1352         rt->rt_gateway    = fl.fld_dst;
1353         if (gateway)
1354                 rt->rt_gateway = gateway;
1355         rt->rt_local_src  = local_src ? local_src : rt->rt_saddr;
1356
1357         rt->rt_dst_map    = fl.fld_dst;
1358         rt->rt_src_map    = fl.fld_src;
1359
1360         rt->fl.fld_src    = cb->src;
1361         rt->fl.fld_dst    = cb->dst;
1362         rt->fl.oif        = 0;
1363         rt->fl.iif        = in_dev->ifindex;
1364         rt->fl.mark       = fl.mark;
1365
1366         rt->u.dst.flags = DST_HOST;
1367         rt->u.dst.neighbour = neigh;
1368         rt->u.dst.dev = out_dev;
1369         rt->u.dst.lastuse = jiffies;
1370         rt->u.dst.output = dn_rt_bug;
1371         switch(res.type) {
1372                 case RTN_UNICAST:
1373                         rt->u.dst.input = dn_forward;
1374                         break;
1375                 case RTN_LOCAL:
1376                         rt->u.dst.output = dn_output;
1377                         rt->u.dst.input = dn_nsp_rx;
1378                         rt->u.dst.dev = in_dev;
1379                         flags |= RTCF_LOCAL;
1380                         break;
1381                 default:
1382                 case RTN_UNREACHABLE:
1383                 case RTN_BLACKHOLE:
1384                         rt->u.dst.input = dst_discard;
1385         }
1386         rt->rt_flags = flags;
1387         if (rt->u.dst.dev)
1388                 dev_hold(rt->u.dst.dev);
1389
1390         err = dn_rt_set_next_hop(rt, &res);
1391         if (err)
1392                 goto e_neighbour;
1393
1394         hash = dn_hash(rt->fl.fld_src, rt->fl.fld_dst);
1395         dn_insert_route(rt, hash, &rt);
1396         skb_dst_set(skb, &rt->u.dst);
1397
1398 done:
1399         if (neigh)
1400                 neigh_release(neigh);
1401         if (free_res)
1402                 dn_fib_res_put(&res);
1403         dev_put(in_dev);
1404         if (out_dev)
1405                 dev_put(out_dev);
1406 out:
1407         return err;
1408
1409 e_inval:
1410         err = -EINVAL;
1411         goto done;
1412
1413 e_nobufs:
1414         err = -ENOBUFS;
1415         goto done;
1416
1417 e_neighbour:
1418         dst_free(&rt->u.dst);
1419         goto done;
1420 }
1421
1422 static int dn_route_input(struct sk_buff *skb)
1423 {
1424         struct dn_route *rt;
1425         struct dn_skb_cb *cb = DN_SKB_CB(skb);
1426         unsigned hash = dn_hash(cb->src, cb->dst);
1427
1428         if (skb_dst(skb))
1429                 return 0;
1430
1431         rcu_read_lock();
1432         for(rt = rcu_dereference(dn_rt_hash_table[hash].chain); rt != NULL;
1433             rt = rcu_dereference(rt->u.dst.dn_next)) {
1434                 if ((rt->fl.fld_src == cb->src) &&
1435                     (rt->fl.fld_dst == cb->dst) &&
1436                     (rt->fl.oif == 0) &&
1437                     (rt->fl.mark == skb->mark) &&
1438                     (rt->fl.iif == cb->iif)) {
1439                         dst_use(&rt->u.dst, jiffies);
1440                         rcu_read_unlock();
1441                         skb_dst_set(skb, (struct dst_entry *)rt);
1442                         return 0;
1443                 }
1444         }
1445         rcu_read_unlock();
1446
1447         return dn_route_input_slow(skb);
1448 }
1449
1450 static int dn_rt_fill_info(struct sk_buff *skb, u32 pid, u32 seq,
1451                            int event, int nowait, unsigned int flags)
1452 {
1453         struct dn_route *rt = (struct dn_route *)skb_dst(skb);
1454         struct rtmsg *r;
1455         struct nlmsghdr *nlh;
1456         unsigned char *b = skb_tail_pointer(skb);
1457         long expires;
1458
1459         nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*r), flags);
1460         r = NLMSG_DATA(nlh);
1461         r->rtm_family = AF_DECnet;
1462         r->rtm_dst_len = 16;
1463         r->rtm_src_len = 0;
1464         r->rtm_tos = 0;
1465         r->rtm_table = RT_TABLE_MAIN;
1466         RTA_PUT_U32(skb, RTA_TABLE, RT_TABLE_MAIN);
1467         r->rtm_type = rt->rt_type;
1468         r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
1469         r->rtm_scope = RT_SCOPE_UNIVERSE;
1470         r->rtm_protocol = RTPROT_UNSPEC;
1471         if (rt->rt_flags & RTCF_NOTIFY)
1472                 r->rtm_flags |= RTM_F_NOTIFY;
1473         RTA_PUT(skb, RTA_DST, 2, &rt->rt_daddr);
1474         if (rt->fl.fld_src) {
1475                 r->rtm_src_len = 16;
1476                 RTA_PUT(skb, RTA_SRC, 2, &rt->fl.fld_src);
1477         }
1478         if (rt->u.dst.dev)
1479                 RTA_PUT(skb, RTA_OIF, sizeof(int), &rt->u.dst.dev->ifindex);
1480         /*
1481          * Note to self - change this if input routes reverse direction when
1482          * they deal only with inputs and not with replies like they do
1483          * currently.
1484          */
1485         RTA_PUT(skb, RTA_PREFSRC, 2, &rt->rt_local_src);
1486         if (rt->rt_daddr != rt->rt_gateway)
1487                 RTA_PUT(skb, RTA_GATEWAY, 2, &rt->rt_gateway);
1488         if (rtnetlink_put_metrics(skb, rt->u.dst.metrics) < 0)
1489                 goto rtattr_failure;
1490         expires = rt->u.dst.expires ? rt->u.dst.expires - jiffies : 0;
1491         if (rtnl_put_cacheinfo(skb, &rt->u.dst, 0, 0, 0, expires,
1492                                rt->u.dst.error) < 0)
1493                 goto rtattr_failure;
1494         if (rt->fl.iif)
1495                 RTA_PUT(skb, RTA_IIF, sizeof(int), &rt->fl.iif);
1496
1497         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1498         return skb->len;
1499
1500 nlmsg_failure:
1501 rtattr_failure:
1502         nlmsg_trim(skb, b);
1503         return -1;
1504 }
1505
1506 /*
1507  * This is called by both endnodes and routers now.
1508  */
1509 static int dn_cache_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh, void *arg)
1510 {
1511         struct net *net = sock_net(in_skb->sk);
1512         struct rtattr **rta = arg;
1513         struct rtmsg *rtm = NLMSG_DATA(nlh);
1514         struct dn_route *rt = NULL;
1515         struct dn_skb_cb *cb;
1516         int err;
1517         struct sk_buff *skb;
1518         struct flowi fl;
1519
1520         if (net != &init_net)
1521                 return -EINVAL;
1522
1523         memset(&fl, 0, sizeof(fl));
1524         fl.proto = DNPROTO_NSP;
1525
1526         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1527         if (skb == NULL)
1528                 return -ENOBUFS;
1529         skb_reset_mac_header(skb);
1530         cb = DN_SKB_CB(skb);
1531
1532         if (rta[RTA_SRC-1])
1533                 memcpy(&fl.fld_src, RTA_DATA(rta[RTA_SRC-1]), 2);
1534         if (rta[RTA_DST-1])
1535                 memcpy(&fl.fld_dst, RTA_DATA(rta[RTA_DST-1]), 2);
1536         if (rta[RTA_IIF-1])
1537                 memcpy(&fl.iif, RTA_DATA(rta[RTA_IIF-1]), sizeof(int));
1538
1539         if (fl.iif) {
1540                 struct net_device *dev;
1541                 if ((dev = dev_get_by_index(&init_net, fl.iif)) == NULL) {
1542                         kfree_skb(skb);
1543                         return -ENODEV;
1544                 }
1545                 if (!dev->dn_ptr) {
1546                         dev_put(dev);
1547                         kfree_skb(skb);
1548                         return -ENODEV;
1549                 }
1550                 skb->protocol = htons(ETH_P_DNA_RT);
1551                 skb->dev = dev;
1552                 cb->src = fl.fld_src;
1553                 cb->dst = fl.fld_dst;
1554                 local_bh_disable();
1555                 err = dn_route_input(skb);
1556                 local_bh_enable();
1557                 memset(cb, 0, sizeof(struct dn_skb_cb));
1558                 rt = (struct dn_route *)skb_dst(skb);
1559                 if (!err && -rt->u.dst.error)
1560                         err = rt->u.dst.error;
1561         } else {
1562                 int oif = 0;
1563                 if (rta[RTA_OIF - 1])
1564                         memcpy(&oif, RTA_DATA(rta[RTA_OIF - 1]), sizeof(int));
1565                 fl.oif = oif;
1566                 err = dn_route_output_key((struct dst_entry **)&rt, &fl, 0);
1567         }
1568
1569         if (skb->dev)
1570                 dev_put(skb->dev);
1571         skb->dev = NULL;
1572         if (err)
1573                 goto out_free;
1574         skb_dst_set(skb, &rt->u.dst);
1575         if (rtm->rtm_flags & RTM_F_NOTIFY)
1576                 rt->rt_flags |= RTCF_NOTIFY;
1577
1578         err = dn_rt_fill_info(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq, RTM_NEWROUTE, 0, 0);
1579
1580         if (err == 0)
1581                 goto out_free;
1582         if (err < 0) {
1583                 err = -EMSGSIZE;
1584                 goto out_free;
1585         }
1586
1587         return rtnl_unicast(skb, &init_net, NETLINK_CB(in_skb).pid);
1588
1589 out_free:
1590         kfree_skb(skb);
1591         return err;
1592 }
1593
1594 /*
1595  * For routers, this is called from dn_fib_dump, but for endnodes its
1596  * called directly from the rtnetlink dispatch table.
1597  */
1598 int dn_cache_dump(struct sk_buff *skb, struct netlink_callback *cb)
1599 {
1600         struct net *net = sock_net(skb->sk);
1601         struct dn_route *rt;
1602         int h, s_h;
1603         int idx, s_idx;
1604
1605         if (net != &init_net)
1606                 return 0;
1607
1608         if (NLMSG_PAYLOAD(cb->nlh, 0) < sizeof(struct rtmsg))
1609                 return -EINVAL;
1610         if (!(((struct rtmsg *)NLMSG_DATA(cb->nlh))->rtm_flags&RTM_F_CLONED))
1611                 return 0;
1612
1613         s_h = cb->args[0];
1614         s_idx = idx = cb->args[1];
1615         for(h = 0; h <= dn_rt_hash_mask; h++) {
1616                 if (h < s_h)
1617                         continue;
1618                 if (h > s_h)
1619                         s_idx = 0;
1620                 rcu_read_lock_bh();
1621                 for(rt = rcu_dereference(dn_rt_hash_table[h].chain), idx = 0;
1622                         rt;
1623                         rt = rcu_dereference(rt->u.dst.dn_next), idx++) {
1624                         if (idx < s_idx)
1625                                 continue;
1626                         skb_dst_set(skb, dst_clone(&rt->u.dst));
1627                         if (dn_rt_fill_info(skb, NETLINK_CB(cb->skb).pid,
1628                                         cb->nlh->nlmsg_seq, RTM_NEWROUTE,
1629                                         1, NLM_F_MULTI) <= 0) {
1630                                 skb_dst_drop(skb);
1631                                 rcu_read_unlock_bh();
1632                                 goto done;
1633                         }
1634                         skb_dst_drop(skb);
1635                 }
1636                 rcu_read_unlock_bh();
1637         }
1638
1639 done:
1640         cb->args[0] = h;
1641         cb->args[1] = idx;
1642         return skb->len;
1643 }
1644
1645 #ifdef CONFIG_PROC_FS
1646 struct dn_rt_cache_iter_state {
1647         int bucket;
1648 };
1649
1650 static struct dn_route *dn_rt_cache_get_first(struct seq_file *seq)
1651 {
1652         struct dn_route *rt = NULL;
1653         struct dn_rt_cache_iter_state *s = seq->private;
1654
1655         for(s->bucket = dn_rt_hash_mask; s->bucket >= 0; --s->bucket) {
1656                 rcu_read_lock_bh();
1657                 rt = dn_rt_hash_table[s->bucket].chain;
1658                 if (rt)
1659                         break;
1660                 rcu_read_unlock_bh();
1661         }
1662         return rcu_dereference(rt);
1663 }
1664
1665 static struct dn_route *dn_rt_cache_get_next(struct seq_file *seq, struct dn_route *rt)
1666 {
1667         struct dn_rt_cache_iter_state *s = seq->private;
1668
1669         rt = rt->u.dst.dn_next;
1670         while(!rt) {
1671                 rcu_read_unlock_bh();
1672                 if (--s->bucket < 0)
1673                         break;
1674                 rcu_read_lock_bh();
1675                 rt = dn_rt_hash_table[s->bucket].chain;
1676         }
1677         return rcu_dereference(rt);
1678 }
1679
1680 static void *dn_rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
1681 {
1682         struct dn_route *rt = dn_rt_cache_get_first(seq);
1683
1684         if (rt) {
1685                 while(*pos && (rt = dn_rt_cache_get_next(seq, rt)))
1686                         --*pos;
1687         }
1688         return *pos ? NULL : rt;
1689 }
1690
1691 static void *dn_rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1692 {
1693         struct dn_route *rt = dn_rt_cache_get_next(seq, v);
1694         ++*pos;
1695         return rt;
1696 }
1697
1698 static void dn_rt_cache_seq_stop(struct seq_file *seq, void *v)
1699 {
1700         if (v)
1701                 rcu_read_unlock_bh();
1702 }
1703
1704 static int dn_rt_cache_seq_show(struct seq_file *seq, void *v)
1705 {
1706         struct dn_route *rt = v;
1707         char buf1[DN_ASCBUF_LEN], buf2[DN_ASCBUF_LEN];
1708
1709         seq_printf(seq, "%-8s %-7s %-7s %04d %04d %04d\n",
1710                         rt->u.dst.dev ? rt->u.dst.dev->name : "*",
1711                         dn_addr2asc(le16_to_cpu(rt->rt_daddr), buf1),
1712                         dn_addr2asc(le16_to_cpu(rt->rt_saddr), buf2),
1713                         atomic_read(&rt->u.dst.__refcnt),
1714                         rt->u.dst.__use,
1715                         (int) dst_metric(&rt->u.dst, RTAX_RTT));
1716         return 0;
1717 }
1718
1719 static const struct seq_operations dn_rt_cache_seq_ops = {
1720         .start  = dn_rt_cache_seq_start,
1721         .next   = dn_rt_cache_seq_next,
1722         .stop   = dn_rt_cache_seq_stop,
1723         .show   = dn_rt_cache_seq_show,
1724 };
1725
1726 static int dn_rt_cache_seq_open(struct inode *inode, struct file *file)
1727 {
1728         return seq_open_private(file, &dn_rt_cache_seq_ops,
1729                         sizeof(struct dn_rt_cache_iter_state));
1730 }
1731
1732 static const struct file_operations dn_rt_cache_seq_fops = {
1733         .owner   = THIS_MODULE,
1734         .open    = dn_rt_cache_seq_open,
1735         .read    = seq_read,
1736         .llseek  = seq_lseek,
1737         .release = seq_release_private,
1738 };
1739
1740 #endif /* CONFIG_PROC_FS */
1741
1742 void __init dn_route_init(void)
1743 {
1744         int i, goal, order;
1745
1746         dn_dst_ops.kmem_cachep =
1747                 kmem_cache_create("dn_dst_cache", sizeof(struct dn_route), 0,
1748                                   SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1749         setup_timer(&dn_route_timer, dn_dst_check_expire, 0);
1750         dn_route_timer.expires = jiffies + decnet_dst_gc_interval * HZ;
1751         add_timer(&dn_route_timer);
1752
1753         goal = num_physpages >> (26 - PAGE_SHIFT);
1754
1755         for(order = 0; (1UL << order) < goal; order++)
1756                 /* NOTHING */;
1757
1758         /*
1759          * Only want 1024 entries max, since the table is very, very unlikely
1760          * to be larger than that.
1761          */
1762         while(order && ((((1UL << order) * PAGE_SIZE) /
1763                                 sizeof(struct dn_rt_hash_bucket)) >= 2048))
1764                 order--;
1765
1766         do {
1767                 dn_rt_hash_mask = (1UL << order) * PAGE_SIZE /
1768                         sizeof(struct dn_rt_hash_bucket);
1769                 while(dn_rt_hash_mask & (dn_rt_hash_mask - 1))
1770                         dn_rt_hash_mask--;
1771                 dn_rt_hash_table = (struct dn_rt_hash_bucket *)
1772                         __get_free_pages(GFP_ATOMIC, order);
1773         } while (dn_rt_hash_table == NULL && --order > 0);
1774
1775         if (!dn_rt_hash_table)
1776                 panic("Failed to allocate DECnet route cache hash table\n");
1777
1778         printk(KERN_INFO
1779                 "DECnet: Routing cache hash table of %u buckets, %ldKbytes\n",
1780                 dn_rt_hash_mask,
1781                 (long)(dn_rt_hash_mask*sizeof(struct dn_rt_hash_bucket))/1024);
1782
1783         dn_rt_hash_mask--;
1784         for(i = 0; i <= dn_rt_hash_mask; i++) {
1785                 spin_lock_init(&dn_rt_hash_table[i].lock);
1786                 dn_rt_hash_table[i].chain = NULL;
1787         }
1788
1789         dn_dst_ops.gc_thresh = (dn_rt_hash_mask + 1);
1790
1791         proc_net_fops_create(&init_net, "decnet_cache", S_IRUGO, &dn_rt_cache_seq_fops);
1792
1793 #ifdef CONFIG_DECNET_ROUTER
1794         rtnl_register(PF_DECnet, RTM_GETROUTE, dn_cache_getroute, dn_fib_dump);
1795 #else
1796         rtnl_register(PF_DECnet, RTM_GETROUTE, dn_cache_getroute,
1797                       dn_cache_dump);
1798 #endif
1799 }
1800
1801 void __exit dn_route_cleanup(void)
1802 {
1803         del_timer(&dn_route_timer);
1804         dn_run_flush(0);
1805
1806         proc_net_remove(&init_net, "decnet_cache");
1807 }
1808