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