[NET]: Make packet reception network namespace safe
[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 } __attribute__((__aligned__(8)));
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(void);
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(void)
189 {
190         struct dn_route *rt, **rtp;
191         int i;
192         unsigned long now = jiffies;
193         unsigned long expire = 10 * HZ;
194
195         for(i = 0; i <= dn_rt_hash_mask; i++) {
196
197                 spin_lock_bh(&dn_rt_hash_table[i].lock);
198                 rtp = &dn_rt_hash_table[i].chain;
199
200                 while((rt=*rtp) != NULL) {
201                         if (atomic_read(&rt->u.dst.__refcnt) ||
202                                         (now - rt->u.dst.lastuse) < expire) {
203                                 rtp = &rt->u.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                         rth->u.dst.__use++;
297                         dst_hold(&rth->u.dst);
298                         rth->u.dst.lastuse = now;
299                         spin_unlock_bh(&dn_rt_hash_table[hash].lock);
300
301                         dnrt_drop(rt);
302                         *rp = rth;
303                         return 0;
304                 }
305                 rthp = &rth->u.dst.dn_next;
306         }
307
308         rcu_assign_pointer(rt->u.dst.dn_next, dn_rt_hash_table[hash].chain);
309         rcu_assign_pointer(dn_rt_hash_table[hash].chain, rt);
310
311         dst_hold(&rt->u.dst);
312         rt->u.dst.__use++;
313         rt->u.dst.lastuse = now;
314         spin_unlock_bh(&dn_rt_hash_table[hash].lock);
315         *rp = rt;
316         return 0;
317 }
318
319 void dn_run_flush(unsigned long dummy)
320 {
321         int i;
322         struct dn_route *rt, *next;
323
324         for(i = 0; i < dn_rt_hash_mask; i++) {
325                 spin_lock_bh(&dn_rt_hash_table[i].lock);
326
327                 if ((rt = xchg(&dn_rt_hash_table[i].chain, NULL)) == NULL)
328                         goto nothing_to_declare;
329
330                 for(; rt; rt=next) {
331                         next = rt->u.dst.dn_next;
332                         rt->u.dst.dn_next = NULL;
333                         dst_free((struct dst_entry *)rt);
334                 }
335
336 nothing_to_declare:
337                 spin_unlock_bh(&dn_rt_hash_table[i].lock);
338         }
339 }
340
341 static DEFINE_SPINLOCK(dn_rt_flush_lock);
342
343 void dn_rt_cache_flush(int delay)
344 {
345         unsigned long now = jiffies;
346         int user_mode = !in_interrupt();
347
348         if (delay < 0)
349                 delay = dn_rt_min_delay;
350
351         spin_lock_bh(&dn_rt_flush_lock);
352
353         if (del_timer(&dn_rt_flush_timer) && delay > 0 && dn_rt_deadline) {
354                 long tmo = (long)(dn_rt_deadline - now);
355
356                 if (user_mode && tmo < dn_rt_max_delay - dn_rt_min_delay)
357                         tmo = 0;
358
359                 if (delay > tmo)
360                         delay = tmo;
361         }
362
363         if (delay <= 0) {
364                 spin_unlock_bh(&dn_rt_flush_lock);
365                 dn_run_flush(0);
366                 return;
367         }
368
369         if (dn_rt_deadline == 0)
370                 dn_rt_deadline = now + dn_rt_max_delay;
371
372         dn_rt_flush_timer.expires = now + delay;
373         add_timer(&dn_rt_flush_timer);
374         spin_unlock_bh(&dn_rt_flush_lock);
375 }
376
377 /**
378  * dn_return_short - Return a short packet to its sender
379  * @skb: The packet to return
380  *
381  */
382 static int dn_return_short(struct sk_buff *skb)
383 {
384         struct dn_skb_cb *cb;
385         unsigned char *ptr;
386         __le16 *src;
387         __le16 *dst;
388         __le16 tmp;
389
390         /* Add back headers */
391         skb_push(skb, skb->data - skb_network_header(skb));
392
393         if ((skb = skb_unshare(skb, GFP_ATOMIC)) == NULL)
394                 return NET_RX_DROP;
395
396         cb = DN_SKB_CB(skb);
397         /* Skip packet length and point to flags */
398         ptr = skb->data + 2;
399         *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
400
401         dst = (__le16 *)ptr;
402         ptr += 2;
403         src = (__le16 *)ptr;
404         ptr += 2;
405         *ptr = 0; /* Zero hop count */
406
407         /* Swap source and destination */
408         tmp  = *src;
409         *src = *dst;
410         *dst = tmp;
411
412         skb->pkt_type = PACKET_OUTGOING;
413         dn_rt_finish_output(skb, NULL, NULL);
414         return NET_RX_SUCCESS;
415 }
416
417 /**
418  * dn_return_long - Return a long packet to its sender
419  * @skb: The long format packet to return
420  *
421  */
422 static int dn_return_long(struct sk_buff *skb)
423 {
424         struct dn_skb_cb *cb;
425         unsigned char *ptr;
426         unsigned char *src_addr, *dst_addr;
427         unsigned char tmp[ETH_ALEN];
428
429         /* Add back all headers */
430         skb_push(skb, skb->data - skb_network_header(skb));
431
432         if ((skb = skb_unshare(skb, GFP_ATOMIC)) == NULL)
433                 return NET_RX_DROP;
434
435         cb = DN_SKB_CB(skb);
436         /* Ignore packet length and point to flags */
437         ptr = skb->data + 2;
438
439         /* Skip padding */
440         if (*ptr & DN_RT_F_PF) {
441                 char padlen = (*ptr & ~DN_RT_F_PF);
442                 ptr += padlen;
443         }
444
445         *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
446         ptr += 2;
447         dst_addr = ptr;
448         ptr += 8;
449         src_addr = ptr;
450         ptr += 6;
451         *ptr = 0; /* Zero hop count */
452
453         /* Swap source and destination */
454         memcpy(tmp, src_addr, ETH_ALEN);
455         memcpy(src_addr, dst_addr, ETH_ALEN);
456         memcpy(dst_addr, tmp, ETH_ALEN);
457
458         skb->pkt_type = PACKET_OUTGOING;
459         dn_rt_finish_output(skb, dst_addr, src_addr);
460         return NET_RX_SUCCESS;
461 }
462
463 /**
464  * dn_route_rx_packet - Try and find a route for an incoming packet
465  * @skb: The packet to find a route for
466  *
467  * Returns: result of input function if route is found, error code otherwise
468  */
469 static int dn_route_rx_packet(struct sk_buff *skb)
470 {
471         struct dn_skb_cb *cb = DN_SKB_CB(skb);
472         int err;
473
474         if ((err = dn_route_input(skb)) == 0)
475                 return dst_input(skb);
476
477         if (decnet_debug_level & 4) {
478                 char *devname = skb->dev ? skb->dev->name : "???";
479                 struct dn_skb_cb *cb = DN_SKB_CB(skb);
480                 printk(KERN_DEBUG
481                         "DECnet: dn_route_rx_packet: rt_flags=0x%02x dev=%s len=%d src=0x%04hx dst=0x%04hx err=%d type=%d\n",
482                         (int)cb->rt_flags, devname, skb->len,
483                         dn_ntohs(cb->src), dn_ntohs(cb->dst),
484                         err, skb->pkt_type);
485         }
486
487         if ((skb->pkt_type == PACKET_HOST) && (cb->rt_flags & DN_RT_F_RQR)) {
488                 switch(cb->rt_flags & DN_RT_PKT_MSK) {
489                         case DN_RT_PKT_SHORT:
490                                 return dn_return_short(skb);
491                         case DN_RT_PKT_LONG:
492                                 return dn_return_long(skb);
493                 }
494         }
495
496         kfree_skb(skb);
497         return NET_RX_DROP;
498 }
499
500 static int dn_route_rx_long(struct sk_buff *skb)
501 {
502         struct dn_skb_cb *cb = DN_SKB_CB(skb);
503         unsigned char *ptr = skb->data;
504
505         if (!pskb_may_pull(skb, 21)) /* 20 for long header, 1 for shortest nsp */
506                 goto drop_it;
507
508         skb_pull(skb, 20);
509         skb_reset_transport_header(skb);
510
511         /* Destination info */
512         ptr += 2;
513         cb->dst = dn_eth2dn(ptr);
514         if (memcmp(ptr, dn_hiord_addr, 4) != 0)
515                 goto drop_it;
516         ptr += 6;
517
518
519         /* Source info */
520         ptr += 2;
521         cb->src = dn_eth2dn(ptr);
522         if (memcmp(ptr, dn_hiord_addr, 4) != 0)
523                 goto drop_it;
524         ptr += 6;
525         /* Other junk */
526         ptr++;
527         cb->hops = *ptr++; /* Visit Count */
528
529         return NF_HOOK(PF_DECnet, NF_DN_PRE_ROUTING, skb, skb->dev, NULL, dn_route_rx_packet);
530
531 drop_it:
532         kfree_skb(skb);
533         return NET_RX_DROP;
534 }
535
536
537
538 static int dn_route_rx_short(struct sk_buff *skb)
539 {
540         struct dn_skb_cb *cb = DN_SKB_CB(skb);
541         unsigned char *ptr = skb->data;
542
543         if (!pskb_may_pull(skb, 6)) /* 5 for short header + 1 for shortest nsp */
544                 goto drop_it;
545
546         skb_pull(skb, 5);
547         skb_reset_transport_header(skb);
548
549         cb->dst = *(__le16 *)ptr;
550         ptr += 2;
551         cb->src = *(__le16 *)ptr;
552         ptr += 2;
553         cb->hops = *ptr & 0x3f;
554
555         return NF_HOOK(PF_DECnet, NF_DN_PRE_ROUTING, skb, skb->dev, NULL, dn_route_rx_packet);
556
557 drop_it:
558         kfree_skb(skb);
559         return NET_RX_DROP;
560 }
561
562 static int dn_route_discard(struct sk_buff *skb)
563 {
564         /*
565          * I know we drop the packet here, but thats considered success in
566          * this case
567          */
568         kfree_skb(skb);
569         return NET_RX_SUCCESS;
570 }
571
572 static int dn_route_ptp_hello(struct sk_buff *skb)
573 {
574         dn_dev_hello(skb);
575         dn_neigh_pointopoint_hello(skb);
576         return NET_RX_SUCCESS;
577 }
578
579 int dn_route_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
580 {
581         struct dn_skb_cb *cb;
582         unsigned char flags = 0;
583         __u16 len = dn_ntohs(*(__le16 *)skb->data);
584         struct dn_dev *dn = (struct dn_dev *)dev->dn_ptr;
585         unsigned char padlen = 0;
586
587         if (dev->nd_net != &init_net)
588                 goto dump_it;
589
590         if (dn == NULL)
591                 goto dump_it;
592
593         if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
594                 goto out;
595
596         if (!pskb_may_pull(skb, 3))
597                 goto dump_it;
598
599         skb_pull(skb, 2);
600
601         if (len > skb->len)
602                 goto dump_it;
603
604         skb_trim(skb, len);
605
606         flags = *skb->data;
607
608         cb = DN_SKB_CB(skb);
609         cb->stamp = jiffies;
610         cb->iif = dev->ifindex;
611
612         /*
613          * If we have padding, remove it.
614          */
615         if (flags & DN_RT_F_PF) {
616                 padlen = flags & ~DN_RT_F_PF;
617                 if (!pskb_may_pull(skb, padlen + 1))
618                         goto dump_it;
619                 skb_pull(skb, padlen);
620                 flags = *skb->data;
621         }
622
623         skb_reset_network_header(skb);
624
625         /*
626          * Weed out future version DECnet
627          */
628         if (flags & DN_RT_F_VER)
629                 goto dump_it;
630
631         cb->rt_flags = flags;
632
633         if (decnet_debug_level & 1)
634                 printk(KERN_DEBUG
635                         "dn_route_rcv: got 0x%02x from %s [%d %d %d]\n",
636                         (int)flags, (dev) ? dev->name : "???", len, skb->len,
637                         padlen);
638
639         if (flags & DN_RT_PKT_CNTL) {
640                 if (unlikely(skb_linearize(skb)))
641                         goto dump_it;
642
643                 switch(flags & DN_RT_CNTL_MSK) {
644                         case DN_RT_PKT_INIT:
645                                 dn_dev_init_pkt(skb);
646                                 break;
647                         case DN_RT_PKT_VERI:
648                                 dn_dev_veri_pkt(skb);
649                                 break;
650                 }
651
652                 if (dn->parms.state != DN_DEV_S_RU)
653                         goto dump_it;
654
655                 switch(flags & DN_RT_CNTL_MSK) {
656                         case DN_RT_PKT_HELO:
657                                 return NF_HOOK(PF_DECnet, NF_DN_HELLO, skb, skb->dev, NULL, dn_route_ptp_hello);
658
659                         case DN_RT_PKT_L1RT:
660                         case DN_RT_PKT_L2RT:
661                                 return NF_HOOK(PF_DECnet, NF_DN_ROUTE, skb, skb->dev, NULL, dn_route_discard);
662                         case DN_RT_PKT_ERTH:
663                                 return NF_HOOK(PF_DECnet, NF_DN_HELLO, skb, skb->dev, NULL, dn_neigh_router_hello);
664
665                         case DN_RT_PKT_EEDH:
666                                 return NF_HOOK(PF_DECnet, NF_DN_HELLO, skb, skb->dev, NULL, dn_neigh_endnode_hello);
667                 }
668         } else {
669                 if (dn->parms.state != DN_DEV_S_RU)
670                         goto dump_it;
671
672                 skb_pull(skb, 1); /* Pull flags */
673
674                 switch(flags & DN_RT_PKT_MSK) {
675                         case DN_RT_PKT_LONG:
676                                 return dn_route_rx_long(skb);
677                         case DN_RT_PKT_SHORT:
678                                 return dn_route_rx_short(skb);
679                 }
680         }
681
682 dump_it:
683         kfree_skb(skb);
684 out:
685         return NET_RX_DROP;
686 }
687
688 static int dn_output(struct sk_buff *skb)
689 {
690         struct dst_entry *dst = skb->dst;
691         struct dn_route *rt = (struct dn_route *)dst;
692         struct net_device *dev = dst->dev;
693         struct dn_skb_cb *cb = DN_SKB_CB(skb);
694         struct neighbour *neigh;
695
696         int err = -EINVAL;
697
698         if ((neigh = dst->neighbour) == NULL)
699                 goto error;
700
701         skb->dev = dev;
702
703         cb->src = rt->rt_saddr;
704         cb->dst = rt->rt_daddr;
705
706         /*
707          * Always set the Intra-Ethernet bit on all outgoing packets
708          * originated on this node. Only valid flag from upper layers
709          * is return-to-sender-requested. Set hop count to 0 too.
710          */
711         cb->rt_flags &= ~DN_RT_F_RQR;
712         cb->rt_flags |= DN_RT_F_IE;
713         cb->hops = 0;
714
715         return NF_HOOK(PF_DECnet, NF_DN_LOCAL_OUT, skb, NULL, dev, neigh->output);
716
717 error:
718         if (net_ratelimit())
719                 printk(KERN_DEBUG "dn_output: This should not happen\n");
720
721         kfree_skb(skb);
722
723         return err;
724 }
725
726 static int dn_forward(struct sk_buff *skb)
727 {
728         struct dn_skb_cb *cb = DN_SKB_CB(skb);
729         struct dst_entry *dst = skb->dst;
730         struct dn_dev *dn_db = dst->dev->dn_ptr;
731         struct dn_route *rt;
732         struct neighbour *neigh = dst->neighbour;
733         int header_len;
734 #ifdef CONFIG_NETFILTER
735         struct net_device *dev = skb->dev;
736 #endif
737
738         if (skb->pkt_type != PACKET_HOST)
739                 goto drop;
740
741         /* Ensure that we have enough space for headers */
742         rt = (struct dn_route *)skb->dst;
743         header_len = dn_db->use_long ? 21 : 6;
744         if (skb_cow(skb, LL_RESERVED_SPACE(rt->u.dst.dev)+header_len))
745                 goto drop;
746
747         /*
748          * Hop count exceeded.
749          */
750         if (++cb->hops > 30)
751                 goto drop;
752
753         skb->dev = rt->u.dst.dev;
754
755         /*
756          * If packet goes out same interface it came in on, then set
757          * the Intra-Ethernet bit. This has no effect for short
758          * packets, so we don't need to test for them here.
759          */
760         cb->rt_flags &= ~DN_RT_F_IE;
761         if (rt->rt_flags & RTCF_DOREDIRECT)
762                 cb->rt_flags |= DN_RT_F_IE;
763
764         return NF_HOOK(PF_DECnet, NF_DN_FORWARD, skb, dev, skb->dev, neigh->output);
765
766 drop:
767         kfree_skb(skb);
768         return NET_RX_DROP;
769 }
770
771 /*
772  * Drop packet. This is used for endnodes and for
773  * when we should not be forwarding packets from
774  * this dest.
775  */
776 static int dn_blackhole(struct sk_buff *skb)
777 {
778         kfree_skb(skb);
779         return NET_RX_DROP;
780 }
781
782 /*
783  * Used to catch bugs. This should never normally get
784  * called.
785  */
786 static int dn_rt_bug(struct sk_buff *skb)
787 {
788         if (net_ratelimit()) {
789                 struct dn_skb_cb *cb = DN_SKB_CB(skb);
790
791                 printk(KERN_DEBUG "dn_rt_bug: skb from:%04x to:%04x\n",
792                                 dn_ntohs(cb->src), dn_ntohs(cb->dst));
793         }
794
795         kfree_skb(skb);
796
797         return NET_RX_BAD;
798 }
799
800 static int dn_rt_set_next_hop(struct dn_route *rt, struct dn_fib_res *res)
801 {
802         struct dn_fib_info *fi = res->fi;
803         struct net_device *dev = rt->u.dst.dev;
804         struct neighbour *n;
805         unsigned mss;
806
807         if (fi) {
808                 if (DN_FIB_RES_GW(*res) &&
809                     DN_FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
810                         rt->rt_gateway = DN_FIB_RES_GW(*res);
811                 memcpy(rt->u.dst.metrics, fi->fib_metrics,
812                        sizeof(rt->u.dst.metrics));
813         }
814         rt->rt_type = res->type;
815
816         if (dev != NULL && rt->u.dst.neighbour == NULL) {
817                 n = __neigh_lookup_errno(&dn_neigh_table, &rt->rt_gateway, dev);
818                 if (IS_ERR(n))
819                         return PTR_ERR(n);
820                 rt->u.dst.neighbour = n;
821         }
822
823         if (rt->u.dst.metrics[RTAX_MTU-1] == 0 ||
824             rt->u.dst.metrics[RTAX_MTU-1] > rt->u.dst.dev->mtu)
825                 rt->u.dst.metrics[RTAX_MTU-1] = rt->u.dst.dev->mtu;
826         mss = dn_mss_from_pmtu(dev, dst_mtu(&rt->u.dst));
827         if (rt->u.dst.metrics[RTAX_ADVMSS-1] == 0 ||
828             rt->u.dst.metrics[RTAX_ADVMSS-1] > mss)
829                 rt->u.dst.metrics[RTAX_ADVMSS-1] = mss;
830         return 0;
831 }
832
833 static inline int dn_match_addr(__le16 addr1, __le16 addr2)
834 {
835         __u16 tmp = dn_ntohs(addr1) ^ dn_ntohs(addr2);
836         int match = 16;
837         while(tmp) {
838                 tmp >>= 1;
839                 match--;
840         }
841         return match;
842 }
843
844 static __le16 dnet_select_source(const struct net_device *dev, __le16 daddr, int scope)
845 {
846         __le16 saddr = 0;
847         struct dn_dev *dn_db = dev->dn_ptr;
848         struct dn_ifaddr *ifa;
849         int best_match = 0;
850         int ret;
851
852         read_lock(&dev_base_lock);
853         for(ifa = dn_db->ifa_list; ifa; ifa = ifa->ifa_next) {
854                 if (ifa->ifa_scope > scope)
855                         continue;
856                 if (!daddr) {
857                         saddr = ifa->ifa_local;
858                         break;
859                 }
860                 ret = dn_match_addr(daddr, ifa->ifa_local);
861                 if (ret > best_match)
862                         saddr = ifa->ifa_local;
863                 if (best_match == 0)
864                         saddr = ifa->ifa_local;
865         }
866         read_unlock(&dev_base_lock);
867
868         return saddr;
869 }
870
871 static inline __le16 __dn_fib_res_prefsrc(struct dn_fib_res *res)
872 {
873         return dnet_select_source(DN_FIB_RES_DEV(*res), DN_FIB_RES_GW(*res), res->scope);
874 }
875
876 static inline __le16 dn_fib_rules_map_destination(__le16 daddr, struct dn_fib_res *res)
877 {
878         __le16 mask = dnet_make_mask(res->prefixlen);
879         return (daddr&~mask)|res->fi->fib_nh->nh_gw;
880 }
881
882 static int dn_route_output_slow(struct dst_entry **pprt, const struct flowi *oldflp, int try_hard)
883 {
884         struct flowi fl = { .nl_u = { .dn_u =
885                                       { .daddr = oldflp->fld_dst,
886                                         .saddr = oldflp->fld_src,
887                                         .scope = RT_SCOPE_UNIVERSE,
888                                      } },
889                             .mark = oldflp->mark,
890                             .iif = loopback_dev.ifindex,
891                             .oif = oldflp->oif };
892         struct dn_route *rt = NULL;
893         struct net_device *dev_out = NULL, *dev;
894         struct neighbour *neigh = NULL;
895         unsigned hash;
896         unsigned flags = 0;
897         struct dn_fib_res res = { .fi = NULL, .type = RTN_UNICAST };
898         int err;
899         int free_res = 0;
900         __le16 gateway = 0;
901
902         if (decnet_debug_level & 16)
903                 printk(KERN_DEBUG
904                        "dn_route_output_slow: dst=%04x src=%04x mark=%d"
905                        " iif=%d oif=%d\n", dn_ntohs(oldflp->fld_dst),
906                        dn_ntohs(oldflp->fld_src),
907                        oldflp->mark, loopback_dev.ifindex, oldflp->oif);
908
909         /* If we have an output interface, verify its a DECnet device */
910         if (oldflp->oif) {
911                 dev_out = dev_get_by_index(oldflp->oif);
912                 err = -ENODEV;
913                 if (dev_out && dev_out->dn_ptr == NULL) {
914                         dev_put(dev_out);
915                         dev_out = NULL;
916                 }
917                 if (dev_out == NULL)
918                         goto out;
919         }
920
921         /* If we have a source address, verify that its a local address */
922         if (oldflp->fld_src) {
923                 err = -EADDRNOTAVAIL;
924
925                 if (dev_out) {
926                         if (dn_dev_islocal(dev_out, oldflp->fld_src))
927                                 goto source_ok;
928                         dev_put(dev_out);
929                         goto out;
930                 }
931                 read_lock(&dev_base_lock);
932                 for_each_netdev(dev) {
933                         if (!dev->dn_ptr)
934                                 continue;
935                         if (!dn_dev_islocal(dev, oldflp->fld_src))
936                                 continue;
937                         if ((dev->flags & IFF_LOOPBACK) &&
938                             oldflp->fld_dst &&
939                             !dn_dev_islocal(dev, oldflp->fld_dst))
940                                 continue;
941
942                         dev_out = dev;
943                         break;
944                 }
945                 read_unlock(&dev_base_lock);
946                 if (dev_out == NULL)
947                         goto out;
948                 dev_hold(dev_out);
949 source_ok:
950                 ;
951         }
952
953         /* No destination? Assume its local */
954         if (!fl.fld_dst) {
955                 fl.fld_dst = fl.fld_src;
956
957                 err = -EADDRNOTAVAIL;
958                 if (dev_out)
959                         dev_put(dev_out);
960                 dev_out = &loopback_dev;
961                 dev_hold(dev_out);
962                 if (!fl.fld_dst) {
963                         fl.fld_dst =
964                         fl.fld_src = dnet_select_source(dev_out, 0,
965                                                        RT_SCOPE_HOST);
966                         if (!fl.fld_dst)
967                                 goto out;
968                 }
969                 fl.oif = loopback_dev.ifindex;
970                 res.type = RTN_LOCAL;
971                 goto make_route;
972         }
973
974         if (decnet_debug_level & 16)
975                 printk(KERN_DEBUG
976                        "dn_route_output_slow: initial checks complete."
977                        " dst=%o4x src=%04x oif=%d try_hard=%d\n",
978                        dn_ntohs(fl.fld_dst), dn_ntohs(fl.fld_src),
979                        fl.oif, try_hard);
980
981         /*
982          * N.B. If the kernel is compiled without router support then
983          * dn_fib_lookup() will evaluate to non-zero so this if () block
984          * will always be executed.
985          */
986         err = -ESRCH;
987         if (try_hard || (err = dn_fib_lookup(&fl, &res)) != 0) {
988                 struct dn_dev *dn_db;
989                 if (err != -ESRCH)
990                         goto out;
991                 /*
992                  * Here the fallback is basically the standard algorithm for
993                  * routing in endnodes which is described in the DECnet routing
994                  * docs
995                  *
996                  * If we are not trying hard, look in neighbour cache.
997                  * The result is tested to ensure that if a specific output
998                  * device/source address was requested, then we honour that
999                  * here
1000                  */
1001                 if (!try_hard) {
1002                         neigh = neigh_lookup_nodev(&dn_neigh_table, &fl.fld_dst);
1003                         if (neigh) {
1004                                 if ((oldflp->oif &&
1005                                     (neigh->dev->ifindex != oldflp->oif)) ||
1006                                     (oldflp->fld_src &&
1007                                     (!dn_dev_islocal(neigh->dev,
1008                                                       oldflp->fld_src)))) {
1009                                         neigh_release(neigh);
1010                                         neigh = NULL;
1011                                 } else {
1012                                         if (dev_out)
1013                                                 dev_put(dev_out);
1014                                         if (dn_dev_islocal(neigh->dev, fl.fld_dst)) {
1015                                                 dev_out = &loopback_dev;
1016                                                 res.type = RTN_LOCAL;
1017                                         } else {
1018                                                 dev_out = neigh->dev;
1019                                         }
1020                                         dev_hold(dev_out);
1021                                         goto select_source;
1022                                 }
1023                         }
1024                 }
1025
1026                 /* Not there? Perhaps its a local address */
1027                 if (dev_out == NULL)
1028                         dev_out = dn_dev_get_default();
1029                 err = -ENODEV;
1030                 if (dev_out == NULL)
1031                         goto out;
1032                 dn_db = dev_out->dn_ptr;
1033                 /* Possible improvement - check all devices for local addr */
1034                 if (dn_dev_islocal(dev_out, fl.fld_dst)) {
1035                         dev_put(dev_out);
1036                         dev_out = &loopback_dev;
1037                         dev_hold(dev_out);
1038                         res.type = RTN_LOCAL;
1039                         goto select_source;
1040                 }
1041                 /* Not local either.... try sending it to the default router */
1042                 neigh = neigh_clone(dn_db->router);
1043                 BUG_ON(neigh && neigh->dev != dev_out);
1044
1045                 /* Ok then, we assume its directly connected and move on */
1046 select_source:
1047                 if (neigh)
1048                         gateway = ((struct dn_neigh *)neigh)->addr;
1049                 if (gateway == 0)
1050                         gateway = fl.fld_dst;
1051                 if (fl.fld_src == 0) {
1052                         fl.fld_src = dnet_select_source(dev_out, gateway,
1053                                                          res.type == RTN_LOCAL ?
1054                                                          RT_SCOPE_HOST :
1055                                                          RT_SCOPE_LINK);
1056                         if (fl.fld_src == 0 && res.type != RTN_LOCAL)
1057                                 goto e_addr;
1058                 }
1059                 fl.oif = dev_out->ifindex;
1060                 goto make_route;
1061         }
1062         free_res = 1;
1063
1064         if (res.type == RTN_NAT)
1065                 goto e_inval;
1066
1067         if (res.type == RTN_LOCAL) {
1068                 if (!fl.fld_src)
1069                         fl.fld_src = fl.fld_dst;
1070                 if (dev_out)
1071                         dev_put(dev_out);
1072                 dev_out = &loopback_dev;
1073                 dev_hold(dev_out);
1074                 fl.oif = dev_out->ifindex;
1075                 if (res.fi)
1076                         dn_fib_info_put(res.fi);
1077                 res.fi = NULL;
1078                 goto make_route;
1079         }
1080
1081         if (res.fi->fib_nhs > 1 && fl.oif == 0)
1082                 dn_fib_select_multipath(&fl, &res);
1083
1084         /*
1085          * We could add some logic to deal with default routes here and
1086          * get rid of some of the special casing above.
1087          */
1088
1089         if (!fl.fld_src)
1090                 fl.fld_src = DN_FIB_RES_PREFSRC(res);
1091
1092         if (dev_out)
1093                 dev_put(dev_out);
1094         dev_out = DN_FIB_RES_DEV(res);
1095         dev_hold(dev_out);
1096         fl.oif = dev_out->ifindex;
1097         gateway = DN_FIB_RES_GW(res);
1098
1099 make_route:
1100         if (dev_out->flags & IFF_LOOPBACK)
1101                 flags |= RTCF_LOCAL;
1102
1103         rt = dst_alloc(&dn_dst_ops);
1104         if (rt == NULL)
1105                 goto e_nobufs;
1106
1107         atomic_set(&rt->u.dst.__refcnt, 1);
1108         rt->u.dst.flags   = DST_HOST;
1109
1110         rt->fl.fld_src    = oldflp->fld_src;
1111         rt->fl.fld_dst    = oldflp->fld_dst;
1112         rt->fl.oif        = oldflp->oif;
1113         rt->fl.iif        = 0;
1114         rt->fl.mark       = oldflp->mark;
1115
1116         rt->rt_saddr      = fl.fld_src;
1117         rt->rt_daddr      = fl.fld_dst;
1118         rt->rt_gateway    = gateway ? gateway : fl.fld_dst;
1119         rt->rt_local_src  = fl.fld_src;
1120
1121         rt->rt_dst_map    = fl.fld_dst;
1122         rt->rt_src_map    = fl.fld_src;
1123
1124         rt->u.dst.dev = dev_out;
1125         dev_hold(dev_out);
1126         rt->u.dst.neighbour = neigh;
1127         neigh = NULL;
1128
1129         rt->u.dst.lastuse = jiffies;
1130         rt->u.dst.output  = dn_output;
1131         rt->u.dst.input   = dn_rt_bug;
1132         rt->rt_flags      = flags;
1133         if (flags & RTCF_LOCAL)
1134                 rt->u.dst.input = dn_nsp_rx;
1135
1136         err = dn_rt_set_next_hop(rt, &res);
1137         if (err)
1138                 goto e_neighbour;
1139
1140         hash = dn_hash(rt->fl.fld_src, rt->fl.fld_dst);
1141         dn_insert_route(rt, hash, (struct dn_route **)pprt);
1142
1143 done:
1144         if (neigh)
1145                 neigh_release(neigh);
1146         if (free_res)
1147                 dn_fib_res_put(&res);
1148         if (dev_out)
1149                 dev_put(dev_out);
1150 out:
1151         return err;
1152
1153 e_addr:
1154         err = -EADDRNOTAVAIL;
1155         goto done;
1156 e_inval:
1157         err = -EINVAL;
1158         goto done;
1159 e_nobufs:
1160         err = -ENOBUFS;
1161         goto done;
1162 e_neighbour:
1163         dst_free(&rt->u.dst);
1164         goto e_nobufs;
1165 }
1166
1167
1168 /*
1169  * N.B. The flags may be moved into the flowi at some future stage.
1170  */
1171 static int __dn_route_output_key(struct dst_entry **pprt, const struct flowi *flp, int flags)
1172 {
1173         unsigned hash = dn_hash(flp->fld_src, flp->fld_dst);
1174         struct dn_route *rt = NULL;
1175
1176         if (!(flags & MSG_TRYHARD)) {
1177                 rcu_read_lock_bh();
1178                 for(rt = rcu_dereference(dn_rt_hash_table[hash].chain); rt;
1179                         rt = rcu_dereference(rt->u.dst.dn_next)) {
1180                         if ((flp->fld_dst == rt->fl.fld_dst) &&
1181                             (flp->fld_src == rt->fl.fld_src) &&
1182                             (flp->mark == rt->fl.mark) &&
1183                             (rt->fl.iif == 0) &&
1184                             (rt->fl.oif == flp->oif)) {
1185                                 rt->u.dst.lastuse = jiffies;
1186                                 dst_hold(&rt->u.dst);
1187                                 rt->u.dst.__use++;
1188                                 rcu_read_unlock_bh();
1189                                 *pprt = &rt->u.dst;
1190                                 return 0;
1191                         }
1192                 }
1193                 rcu_read_unlock_bh();
1194         }
1195
1196         return dn_route_output_slow(pprt, flp, flags);
1197 }
1198
1199 static int dn_route_output_key(struct dst_entry **pprt, struct flowi *flp, int flags)
1200 {
1201         int err;
1202
1203         err = __dn_route_output_key(pprt, flp, flags);
1204         if (err == 0 && flp->proto) {
1205                 err = xfrm_lookup(pprt, flp, NULL, 0);
1206         }
1207         return err;
1208 }
1209
1210 int dn_route_output_sock(struct dst_entry **pprt, struct flowi *fl, struct sock *sk, int flags)
1211 {
1212         int err;
1213
1214         err = __dn_route_output_key(pprt, fl, flags & MSG_TRYHARD);
1215         if (err == 0 && fl->proto) {
1216                 err = xfrm_lookup(pprt, fl, sk, !(flags & MSG_DONTWAIT));
1217         }
1218         return err;
1219 }
1220
1221 static int dn_route_input_slow(struct sk_buff *skb)
1222 {
1223         struct dn_route *rt = NULL;
1224         struct dn_skb_cb *cb = DN_SKB_CB(skb);
1225         struct net_device *in_dev = skb->dev;
1226         struct net_device *out_dev = NULL;
1227         struct dn_dev *dn_db;
1228         struct neighbour *neigh = NULL;
1229         unsigned hash;
1230         int flags = 0;
1231         __le16 gateway = 0;
1232         __le16 local_src = 0;
1233         struct flowi fl = { .nl_u = { .dn_u =
1234                                      { .daddr = cb->dst,
1235                                        .saddr = cb->src,
1236                                        .scope = RT_SCOPE_UNIVERSE,
1237                                     } },
1238                             .mark = skb->mark,
1239                             .iif = skb->dev->ifindex };
1240         struct dn_fib_res res = { .fi = NULL, .type = RTN_UNREACHABLE };
1241         int err = -EINVAL;
1242         int free_res = 0;
1243
1244         dev_hold(in_dev);
1245
1246         if ((dn_db = in_dev->dn_ptr) == NULL)
1247                 goto out;
1248
1249         /* Zero source addresses are not allowed */
1250         if (fl.fld_src == 0)
1251                 goto out;
1252
1253         /*
1254          * In this case we've just received a packet from a source
1255          * outside ourselves pretending to come from us. We don't
1256          * allow it any further to prevent routing loops, spoofing and
1257          * other nasties. Loopback packets already have the dst attached
1258          * so this only affects packets which have originated elsewhere.
1259          */
1260         err  = -ENOTUNIQ;
1261         if (dn_dev_islocal(in_dev, cb->src))
1262                 goto out;
1263
1264         err = dn_fib_lookup(&fl, &res);
1265         if (err) {
1266                 if (err != -ESRCH)
1267                         goto out;
1268                 /*
1269                  * Is the destination us ?
1270                  */
1271                 if (!dn_dev_islocal(in_dev, cb->dst))
1272                         goto e_inval;
1273
1274                 res.type = RTN_LOCAL;
1275         } else {
1276                 __le16 src_map = fl.fld_src;
1277                 free_res = 1;
1278
1279                 out_dev = DN_FIB_RES_DEV(res);
1280                 if (out_dev == NULL) {
1281                         if (net_ratelimit())
1282                                 printk(KERN_CRIT "Bug in dn_route_input_slow() "
1283                                                  "No output device\n");
1284                         goto e_inval;
1285                 }
1286                 dev_hold(out_dev);
1287
1288                 if (res.r)
1289                         src_map = fl.fld_src; /* no NAT support for now */
1290
1291                 gateway = DN_FIB_RES_GW(res);
1292                 if (res.type == RTN_NAT) {
1293                         fl.fld_dst = dn_fib_rules_map_destination(fl.fld_dst, &res);
1294                         dn_fib_res_put(&res);
1295                         free_res = 0;
1296                         if (dn_fib_lookup(&fl, &res))
1297                                 goto e_inval;
1298                         free_res = 1;
1299                         if (res.type != RTN_UNICAST)
1300                                 goto e_inval;
1301                         flags |= RTCF_DNAT;
1302                         gateway = fl.fld_dst;
1303                 }
1304                 fl.fld_src = src_map;
1305         }
1306
1307         switch(res.type) {
1308         case RTN_UNICAST:
1309                 /*
1310                  * Forwarding check here, we only check for forwarding
1311                  * being turned off, if you want to only forward intra
1312                  * area, its up to you to set the routing tables up
1313                  * correctly.
1314                  */
1315                 if (dn_db->parms.forwarding == 0)
1316                         goto e_inval;
1317
1318                 if (res.fi->fib_nhs > 1 && fl.oif == 0)
1319                         dn_fib_select_multipath(&fl, &res);
1320
1321                 /*
1322                  * Check for out_dev == in_dev. We use the RTCF_DOREDIRECT
1323                  * flag as a hint to set the intra-ethernet bit when
1324                  * forwarding. If we've got NAT in operation, we don't do
1325                  * this optimisation.
1326                  */
1327                 if (out_dev == in_dev && !(flags & RTCF_NAT))
1328                         flags |= RTCF_DOREDIRECT;
1329
1330                 local_src = DN_FIB_RES_PREFSRC(res);
1331
1332         case RTN_BLACKHOLE:
1333         case RTN_UNREACHABLE:
1334                 break;
1335         case RTN_LOCAL:
1336                 flags |= RTCF_LOCAL;
1337                 fl.fld_src = cb->dst;
1338                 fl.fld_dst = cb->src;
1339
1340                 /* Routing tables gave us a gateway */
1341                 if (gateway)
1342                         goto make_route;
1343
1344                 /* Packet was intra-ethernet, so we know its on-link */
1345                 if (cb->rt_flags & DN_RT_F_IE) {
1346                         gateway = cb->src;
1347                         flags |= RTCF_DIRECTSRC;
1348                         goto make_route;
1349                 }
1350
1351                 /* Use the default router if there is one */
1352                 neigh = neigh_clone(dn_db->router);
1353                 if (neigh) {
1354                         gateway = ((struct dn_neigh *)neigh)->addr;
1355                         goto make_route;
1356                 }
1357
1358                 /* Close eyes and pray */
1359                 gateway = cb->src;
1360                 flags |= RTCF_DIRECTSRC;
1361                 goto make_route;
1362         default:
1363                 goto e_inval;
1364         }
1365
1366 make_route:
1367         rt = dst_alloc(&dn_dst_ops);
1368         if (rt == NULL)
1369                 goto e_nobufs;
1370
1371         rt->rt_saddr      = fl.fld_src;
1372         rt->rt_daddr      = fl.fld_dst;
1373         rt->rt_gateway    = fl.fld_dst;
1374         if (gateway)
1375                 rt->rt_gateway = gateway;
1376         rt->rt_local_src  = local_src ? local_src : rt->rt_saddr;
1377
1378         rt->rt_dst_map    = fl.fld_dst;
1379         rt->rt_src_map    = fl.fld_src;
1380
1381         rt->fl.fld_src    = cb->src;
1382         rt->fl.fld_dst    = cb->dst;
1383         rt->fl.oif        = 0;
1384         rt->fl.iif        = in_dev->ifindex;
1385         rt->fl.mark       = fl.mark;
1386
1387         rt->u.dst.flags = DST_HOST;
1388         rt->u.dst.neighbour = neigh;
1389         rt->u.dst.dev = out_dev;
1390         rt->u.dst.lastuse = jiffies;
1391         rt->u.dst.output = dn_rt_bug;
1392         switch(res.type) {
1393                 case RTN_UNICAST:
1394                         rt->u.dst.input = dn_forward;
1395                         break;
1396                 case RTN_LOCAL:
1397                         rt->u.dst.output = dn_output;
1398                         rt->u.dst.input = dn_nsp_rx;
1399                         rt->u.dst.dev = in_dev;
1400                         flags |= RTCF_LOCAL;
1401                         break;
1402                 default:
1403                 case RTN_UNREACHABLE:
1404                 case RTN_BLACKHOLE:
1405                         rt->u.dst.input = dn_blackhole;
1406         }
1407         rt->rt_flags = flags;
1408         if (rt->u.dst.dev)
1409                 dev_hold(rt->u.dst.dev);
1410
1411         err = dn_rt_set_next_hop(rt, &res);
1412         if (err)
1413                 goto e_neighbour;
1414
1415         hash = dn_hash(rt->fl.fld_src, rt->fl.fld_dst);
1416         dn_insert_route(rt, hash, (struct dn_route **)&skb->dst);
1417
1418 done:
1419         if (neigh)
1420                 neigh_release(neigh);
1421         if (free_res)
1422                 dn_fib_res_put(&res);
1423         dev_put(in_dev);
1424         if (out_dev)
1425                 dev_put(out_dev);
1426 out:
1427         return err;
1428
1429 e_inval:
1430         err = -EINVAL;
1431         goto done;
1432
1433 e_nobufs:
1434         err = -ENOBUFS;
1435         goto done;
1436
1437 e_neighbour:
1438         dst_free(&rt->u.dst);
1439         goto done;
1440 }
1441
1442 int dn_route_input(struct sk_buff *skb)
1443 {
1444         struct dn_route *rt;
1445         struct dn_skb_cb *cb = DN_SKB_CB(skb);
1446         unsigned hash = dn_hash(cb->src, cb->dst);
1447
1448         if (skb->dst)
1449                 return 0;
1450
1451         rcu_read_lock();
1452         for(rt = rcu_dereference(dn_rt_hash_table[hash].chain); rt != NULL;
1453             rt = rcu_dereference(rt->u.dst.dn_next)) {
1454                 if ((rt->fl.fld_src == cb->src) &&
1455                     (rt->fl.fld_dst == cb->dst) &&
1456                     (rt->fl.oif == 0) &&
1457                     (rt->fl.mark == skb->mark) &&
1458                     (rt->fl.iif == cb->iif)) {
1459                         rt->u.dst.lastuse = jiffies;
1460                         dst_hold(&rt->u.dst);
1461                         rt->u.dst.__use++;
1462                         rcu_read_unlock();
1463                         skb->dst = (struct dst_entry *)rt;
1464                         return 0;
1465                 }
1466         }
1467         rcu_read_unlock();
1468
1469         return dn_route_input_slow(skb);
1470 }
1471
1472 static int dn_rt_fill_info(struct sk_buff *skb, u32 pid, u32 seq,
1473                            int event, int nowait, unsigned int flags)
1474 {
1475         struct dn_route *rt = (struct dn_route *)skb->dst;
1476         struct rtmsg *r;
1477         struct nlmsghdr *nlh;
1478         unsigned char *b = skb_tail_pointer(skb);
1479         long expires;
1480
1481         nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*r), flags);
1482         r = NLMSG_DATA(nlh);
1483         r->rtm_family = AF_DECnet;
1484         r->rtm_dst_len = 16;
1485         r->rtm_src_len = 0;
1486         r->rtm_tos = 0;
1487         r->rtm_table = RT_TABLE_MAIN;
1488         RTA_PUT_U32(skb, RTA_TABLE, RT_TABLE_MAIN);
1489         r->rtm_type = rt->rt_type;
1490         r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
1491         r->rtm_scope = RT_SCOPE_UNIVERSE;
1492         r->rtm_protocol = RTPROT_UNSPEC;
1493         if (rt->rt_flags & RTCF_NOTIFY)
1494                 r->rtm_flags |= RTM_F_NOTIFY;
1495         RTA_PUT(skb, RTA_DST, 2, &rt->rt_daddr);
1496         if (rt->fl.fld_src) {
1497                 r->rtm_src_len = 16;
1498                 RTA_PUT(skb, RTA_SRC, 2, &rt->fl.fld_src);
1499         }
1500         if (rt->u.dst.dev)
1501                 RTA_PUT(skb, RTA_OIF, sizeof(int), &rt->u.dst.dev->ifindex);
1502         /*
1503          * Note to self - change this if input routes reverse direction when
1504          * they deal only with inputs and not with replies like they do
1505          * currently.
1506          */
1507         RTA_PUT(skb, RTA_PREFSRC, 2, &rt->rt_local_src);
1508         if (rt->rt_daddr != rt->rt_gateway)
1509                 RTA_PUT(skb, RTA_GATEWAY, 2, &rt->rt_gateway);
1510         if (rtnetlink_put_metrics(skb, rt->u.dst.metrics) < 0)
1511                 goto rtattr_failure;
1512         expires = rt->u.dst.expires ? rt->u.dst.expires - jiffies : 0;
1513         if (rtnl_put_cacheinfo(skb, &rt->u.dst, 0, 0, 0, expires,
1514                                rt->u.dst.error) < 0)
1515                 goto rtattr_failure;
1516         if (rt->fl.iif)
1517                 RTA_PUT(skb, RTA_IIF, sizeof(int), &rt->fl.iif);
1518
1519         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1520         return skb->len;
1521
1522 nlmsg_failure:
1523 rtattr_failure:
1524         nlmsg_trim(skb, b);
1525         return -1;
1526 }
1527
1528 /*
1529  * This is called by both endnodes and routers now.
1530  */
1531 static int dn_cache_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh, void *arg)
1532 {
1533         struct rtattr **rta = arg;
1534         struct rtmsg *rtm = NLMSG_DATA(nlh);
1535         struct dn_route *rt = NULL;
1536         struct dn_skb_cb *cb;
1537         int err;
1538         struct sk_buff *skb;
1539         struct flowi fl;
1540
1541         memset(&fl, 0, sizeof(fl));
1542         fl.proto = DNPROTO_NSP;
1543
1544         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1545         if (skb == NULL)
1546                 return -ENOBUFS;
1547         skb_reset_mac_header(skb);
1548         cb = DN_SKB_CB(skb);
1549
1550         if (rta[RTA_SRC-1])
1551                 memcpy(&fl.fld_src, RTA_DATA(rta[RTA_SRC-1]), 2);
1552         if (rta[RTA_DST-1])
1553                 memcpy(&fl.fld_dst, RTA_DATA(rta[RTA_DST-1]), 2);
1554         if (rta[RTA_IIF-1])
1555                 memcpy(&fl.iif, RTA_DATA(rta[RTA_IIF-1]), sizeof(int));
1556
1557         if (fl.iif) {
1558                 struct net_device *dev;
1559                 if ((dev = dev_get_by_index(fl.iif)) == NULL) {
1560                         kfree_skb(skb);
1561                         return -ENODEV;
1562                 }
1563                 if (!dev->dn_ptr) {
1564                         dev_put(dev);
1565                         kfree_skb(skb);
1566                         return -ENODEV;
1567                 }
1568                 skb->protocol = __constant_htons(ETH_P_DNA_RT);
1569                 skb->dev = dev;
1570                 cb->src = fl.fld_src;
1571                 cb->dst = fl.fld_dst;
1572                 local_bh_disable();
1573                 err = dn_route_input(skb);
1574                 local_bh_enable();
1575                 memset(cb, 0, sizeof(struct dn_skb_cb));
1576                 rt = (struct dn_route *)skb->dst;
1577                 if (!err && -rt->u.dst.error)
1578                         err = rt->u.dst.error;
1579         } else {
1580                 int oif = 0;
1581                 if (rta[RTA_OIF - 1])
1582                         memcpy(&oif, RTA_DATA(rta[RTA_OIF - 1]), sizeof(int));
1583                 fl.oif = oif;
1584                 err = dn_route_output_key((struct dst_entry **)&rt, &fl, 0);
1585         }
1586
1587         if (skb->dev)
1588                 dev_put(skb->dev);
1589         skb->dev = NULL;
1590         if (err)
1591                 goto out_free;
1592         skb->dst = &rt->u.dst;
1593         if (rtm->rtm_flags & RTM_F_NOTIFY)
1594                 rt->rt_flags |= RTCF_NOTIFY;
1595
1596         err = dn_rt_fill_info(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq, RTM_NEWROUTE, 0, 0);
1597
1598         if (err == 0)
1599                 goto out_free;
1600         if (err < 0) {
1601                 err = -EMSGSIZE;
1602                 goto out_free;
1603         }
1604
1605         return rtnl_unicast(skb, NETLINK_CB(in_skb).pid);
1606
1607 out_free:
1608         kfree_skb(skb);
1609         return err;
1610 }
1611
1612 /*
1613  * For routers, this is called from dn_fib_dump, but for endnodes its
1614  * called directly from the rtnetlink dispatch table.
1615  */
1616 int dn_cache_dump(struct sk_buff *skb, struct netlink_callback *cb)
1617 {
1618         struct dn_route *rt;
1619         int h, s_h;
1620         int idx, s_idx;
1621
1622         if (NLMSG_PAYLOAD(cb->nlh, 0) < sizeof(struct rtmsg))
1623                 return -EINVAL;
1624         if (!(((struct rtmsg *)NLMSG_DATA(cb->nlh))->rtm_flags&RTM_F_CLONED))
1625                 return 0;
1626
1627         s_h = cb->args[0];
1628         s_idx = idx = cb->args[1];
1629         for(h = 0; h <= dn_rt_hash_mask; h++) {
1630                 if (h < s_h)
1631                         continue;
1632                 if (h > s_h)
1633                         s_idx = 0;
1634                 rcu_read_lock_bh();
1635                 for(rt = rcu_dereference(dn_rt_hash_table[h].chain), idx = 0;
1636                         rt;
1637                         rt = rcu_dereference(rt->u.dst.dn_next), idx++) {
1638                         if (idx < s_idx)
1639                                 continue;
1640                         skb->dst = dst_clone(&rt->u.dst);
1641                         if (dn_rt_fill_info(skb, NETLINK_CB(cb->skb).pid,
1642                                         cb->nlh->nlmsg_seq, RTM_NEWROUTE,
1643                                         1, NLM_F_MULTI) <= 0) {
1644                                 dst_release(xchg(&skb->dst, NULL));
1645                                 rcu_read_unlock_bh();
1646                                 goto done;
1647                         }
1648                         dst_release(xchg(&skb->dst, NULL));
1649                 }
1650                 rcu_read_unlock_bh();
1651         }
1652
1653 done:
1654         cb->args[0] = h;
1655         cb->args[1] = idx;
1656         return skb->len;
1657 }
1658
1659 #ifdef CONFIG_PROC_FS
1660 struct dn_rt_cache_iter_state {
1661         int bucket;
1662 };
1663
1664 static struct dn_route *dn_rt_cache_get_first(struct seq_file *seq)
1665 {
1666         struct dn_route *rt = NULL;
1667         struct dn_rt_cache_iter_state *s = seq->private;
1668
1669         for(s->bucket = dn_rt_hash_mask; s->bucket >= 0; --s->bucket) {
1670                 rcu_read_lock_bh();
1671                 rt = dn_rt_hash_table[s->bucket].chain;
1672                 if (rt)
1673                         break;
1674                 rcu_read_unlock_bh();
1675         }
1676         return rt;
1677 }
1678
1679 static struct dn_route *dn_rt_cache_get_next(struct seq_file *seq, struct dn_route *rt)
1680 {
1681         struct dn_rt_cache_iter_state *s = rcu_dereference(seq->private);
1682
1683         rt = rt->u.dst.dn_next;
1684         while(!rt) {
1685                 rcu_read_unlock_bh();
1686                 if (--s->bucket < 0)
1687                         break;
1688                 rcu_read_lock_bh();
1689                 rt = dn_rt_hash_table[s->bucket].chain;
1690         }
1691         return rt;
1692 }
1693
1694 static void *dn_rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
1695 {
1696         struct dn_route *rt = dn_rt_cache_get_first(seq);
1697
1698         if (rt) {
1699                 while(*pos && (rt = dn_rt_cache_get_next(seq, rt)))
1700                         --*pos;
1701         }
1702         return *pos ? NULL : rt;
1703 }
1704
1705 static void *dn_rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1706 {
1707         struct dn_route *rt = dn_rt_cache_get_next(seq, v);
1708         ++*pos;
1709         return rt;
1710 }
1711
1712 static void dn_rt_cache_seq_stop(struct seq_file *seq, void *v)
1713 {
1714         if (v)
1715                 rcu_read_unlock_bh();
1716 }
1717
1718 static int dn_rt_cache_seq_show(struct seq_file *seq, void *v)
1719 {
1720         struct dn_route *rt = v;
1721         char buf1[DN_ASCBUF_LEN], buf2[DN_ASCBUF_LEN];
1722
1723         seq_printf(seq, "%-8s %-7s %-7s %04d %04d %04d\n",
1724                         rt->u.dst.dev ? rt->u.dst.dev->name : "*",
1725                         dn_addr2asc(dn_ntohs(rt->rt_daddr), buf1),
1726                         dn_addr2asc(dn_ntohs(rt->rt_saddr), buf2),
1727                         atomic_read(&rt->u.dst.__refcnt),
1728                         rt->u.dst.__use,
1729                         (int) dst_metric(&rt->u.dst, RTAX_RTT));
1730         return 0;
1731 }
1732
1733 static const struct seq_operations dn_rt_cache_seq_ops = {
1734         .start  = dn_rt_cache_seq_start,
1735         .next   = dn_rt_cache_seq_next,
1736         .stop   = dn_rt_cache_seq_stop,
1737         .show   = dn_rt_cache_seq_show,
1738 };
1739
1740 static int dn_rt_cache_seq_open(struct inode *inode, struct file *file)
1741 {
1742         struct seq_file *seq;
1743         int rc = -ENOMEM;
1744         struct dn_rt_cache_iter_state *s;
1745
1746         s = kzalloc(sizeof(*s), GFP_KERNEL);
1747         if (!s)
1748                 goto out;
1749         rc = seq_open(file, &dn_rt_cache_seq_ops);
1750         if (rc)
1751                 goto out_kfree;
1752         seq             = file->private_data;
1753         seq->private    = s;
1754 out:
1755         return rc;
1756 out_kfree:
1757         kfree(s);
1758         goto out;
1759 }
1760
1761 static const struct file_operations dn_rt_cache_seq_fops = {
1762         .owner   = THIS_MODULE,
1763         .open    = dn_rt_cache_seq_open,
1764         .read    = seq_read,
1765         .llseek  = seq_lseek,
1766         .release = seq_release_private,
1767 };
1768
1769 #endif /* CONFIG_PROC_FS */
1770
1771 void __init dn_route_init(void)
1772 {
1773         int i, goal, order;
1774
1775         dn_dst_ops.kmem_cachep =
1776                 kmem_cache_create("dn_dst_cache", sizeof(struct dn_route), 0,
1777                                   SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1778         init_timer(&dn_route_timer);
1779         dn_route_timer.function = dn_dst_check_expire;
1780         dn_route_timer.expires = jiffies + decnet_dst_gc_interval * HZ;
1781         add_timer(&dn_route_timer);
1782
1783         goal = num_physpages >> (26 - PAGE_SHIFT);
1784
1785         for(order = 0; (1UL << order) < goal; order++)
1786                 /* NOTHING */;
1787
1788         /*
1789          * Only want 1024 entries max, since the table is very, very unlikely
1790          * to be larger than that.
1791          */
1792         while(order && ((((1UL << order) * PAGE_SIZE) /
1793                                 sizeof(struct dn_rt_hash_bucket)) >= 2048))
1794                 order--;
1795
1796         do {
1797                 dn_rt_hash_mask = (1UL << order) * PAGE_SIZE /
1798                         sizeof(struct dn_rt_hash_bucket);
1799                 while(dn_rt_hash_mask & (dn_rt_hash_mask - 1))
1800                         dn_rt_hash_mask--;
1801                 dn_rt_hash_table = (struct dn_rt_hash_bucket *)
1802                         __get_free_pages(GFP_ATOMIC, order);
1803         } while (dn_rt_hash_table == NULL && --order > 0);
1804
1805         if (!dn_rt_hash_table)
1806                 panic("Failed to allocate DECnet route cache hash table\n");
1807
1808         printk(KERN_INFO
1809                 "DECnet: Routing cache hash table of %u buckets, %ldKbytes\n",
1810                 dn_rt_hash_mask,
1811                 (long)(dn_rt_hash_mask*sizeof(struct dn_rt_hash_bucket))/1024);
1812
1813         dn_rt_hash_mask--;
1814         for(i = 0; i <= dn_rt_hash_mask; i++) {
1815                 spin_lock_init(&dn_rt_hash_table[i].lock);
1816                 dn_rt_hash_table[i].chain = NULL;
1817         }
1818
1819         dn_dst_ops.gc_thresh = (dn_rt_hash_mask + 1);
1820
1821         proc_net_fops_create(&init_net, "decnet_cache", S_IRUGO, &dn_rt_cache_seq_fops);
1822
1823 #ifdef CONFIG_DECNET_ROUTER
1824         rtnl_register(PF_DECnet, RTM_GETROUTE, dn_cache_getroute, dn_fib_dump);
1825 #else
1826         rtnl_register(PF_DECnet, RTM_GETROUTE, dn_cache_getroute,
1827                       dn_cache_dump);
1828 #endif
1829 }
1830
1831 void __exit dn_route_cleanup(void)
1832 {
1833         del_timer(&dn_route_timer);
1834         dn_run_flush(0);
1835
1836         proc_net_remove(&init_net, "decnet_cache");
1837 }
1838