Merge git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi-misc-2.6
[platform/adaptation/renesas_rcar/renesas_kernel.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 <linux/slab.h>
70 #include <net/sock.h>
71 #include <linux/mm.h>
72 #include <linux/proc_fs.h>
73 #include <linux/seq_file.h>
74 #include <linux/init.h>
75 #include <linux/rtnetlink.h>
76 #include <linux/string.h>
77 #include <linux/netfilter_decnet.h>
78 #include <linux/rcupdate.h>
79 #include <linux/times.h>
80 #include <asm/errno.h>
81 #include <net/net_namespace.h>
82 #include <net/netlink.h>
83 #include <net/neighbour.h>
84 #include <net/dst.h>
85 #include <net/flow.h>
86 #include <net/fib_rules.h>
87 #include <net/dn.h>
88 #include <net/dn_dev.h>
89 #include <net/dn_nsp.h>
90 #include <net/dn_route.h>
91 #include <net/dn_neigh.h>
92 #include <net/dn_fib.h>
93
94 struct dn_rt_hash_bucket
95 {
96         struct dn_route __rcu *chain;
97         spinlock_t lock;
98 };
99
100 extern struct neigh_table dn_neigh_table;
101
102
103 static unsigned char dn_hiord_addr[6] = {0xAA,0x00,0x04,0x00,0x00,0x00};
104
105 static const int dn_rt_min_delay = 2 * HZ;
106 static const int dn_rt_max_delay = 10 * HZ;
107 static const int dn_rt_mtu_expires = 10 * 60 * HZ;
108
109 static unsigned long dn_rt_deadline;
110
111 static int dn_dst_gc(struct dst_ops *ops);
112 static struct dst_entry *dn_dst_check(struct dst_entry *, __u32);
113 static unsigned int dn_dst_default_advmss(const struct dst_entry *dst);
114 static unsigned int dn_dst_default_mtu(const struct dst_entry *dst);
115 static void dn_dst_destroy(struct dst_entry *);
116 static struct dst_entry *dn_dst_negative_advice(struct dst_entry *);
117 static void dn_dst_link_failure(struct sk_buff *);
118 static void dn_dst_update_pmtu(struct dst_entry *dst, u32 mtu);
119 static struct neighbour *dn_dst_neigh_lookup(const struct dst_entry *dst, const void *daddr);
120 static int dn_route_input(struct sk_buff *);
121 static void dn_run_flush(unsigned long dummy);
122
123 static struct dn_rt_hash_bucket *dn_rt_hash_table;
124 static unsigned dn_rt_hash_mask;
125
126 static struct timer_list dn_route_timer;
127 static DEFINE_TIMER(dn_rt_flush_timer, dn_run_flush, 0, 0);
128 int decnet_dst_gc_interval = 2;
129
130 static struct dst_ops dn_dst_ops = {
131         .family =               PF_DECnet,
132         .protocol =             cpu_to_be16(ETH_P_DNA_RT),
133         .gc_thresh =            128,
134         .gc =                   dn_dst_gc,
135         .check =                dn_dst_check,
136         .default_advmss =       dn_dst_default_advmss,
137         .default_mtu =          dn_dst_default_mtu,
138         .cow_metrics =          dst_cow_metrics_generic,
139         .destroy =              dn_dst_destroy,
140         .negative_advice =      dn_dst_negative_advice,
141         .link_failure =         dn_dst_link_failure,
142         .update_pmtu =          dn_dst_update_pmtu,
143         .neigh_lookup =         dn_dst_neigh_lookup,
144 };
145
146 static void dn_dst_destroy(struct dst_entry *dst)
147 {
148         dst_destroy_metrics_generic(dst);
149 }
150
151 static __inline__ unsigned dn_hash(__le16 src, __le16 dst)
152 {
153         __u16 tmp = (__u16 __force)(src ^ dst);
154         tmp ^= (tmp >> 3);
155         tmp ^= (tmp >> 5);
156         tmp ^= (tmp >> 10);
157         return dn_rt_hash_mask & (unsigned)tmp;
158 }
159
160 static inline void dnrt_free(struct dn_route *rt)
161 {
162         call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
163 }
164
165 static inline void dnrt_drop(struct dn_route *rt)
166 {
167         dst_release(&rt->dst);
168         call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
169 }
170
171 static void dn_dst_check_expire(unsigned long dummy)
172 {
173         int i;
174         struct dn_route *rt;
175         struct dn_route __rcu **rtp;
176         unsigned long now = jiffies;
177         unsigned long expire = 120 * HZ;
178
179         for (i = 0; i <= dn_rt_hash_mask; i++) {
180                 rtp = &dn_rt_hash_table[i].chain;
181
182                 spin_lock(&dn_rt_hash_table[i].lock);
183                 while ((rt = rcu_dereference_protected(*rtp,
184                                                 lockdep_is_held(&dn_rt_hash_table[i].lock))) != NULL) {
185                         if (atomic_read(&rt->dst.__refcnt) ||
186                                         (now - rt->dst.lastuse) < expire) {
187                                 rtp = &rt->dst.dn_next;
188                                 continue;
189                         }
190                         *rtp = rt->dst.dn_next;
191                         rt->dst.dn_next = NULL;
192                         dnrt_free(rt);
193                 }
194                 spin_unlock(&dn_rt_hash_table[i].lock);
195
196                 if ((jiffies - now) > 0)
197                         break;
198         }
199
200         mod_timer(&dn_route_timer, now + decnet_dst_gc_interval * HZ);
201 }
202
203 static int dn_dst_gc(struct dst_ops *ops)
204 {
205         struct dn_route *rt;
206         struct dn_route __rcu **rtp;
207         int i;
208         unsigned long now = jiffies;
209         unsigned long expire = 10 * HZ;
210
211         for (i = 0; i <= dn_rt_hash_mask; i++) {
212
213                 spin_lock_bh(&dn_rt_hash_table[i].lock);
214                 rtp = &dn_rt_hash_table[i].chain;
215
216                 while ((rt = rcu_dereference_protected(*rtp,
217                                                 lockdep_is_held(&dn_rt_hash_table[i].lock))) != NULL) {
218                         if (atomic_read(&rt->dst.__refcnt) ||
219                                         (now - rt->dst.lastuse) < expire) {
220                                 rtp = &rt->dst.dn_next;
221                                 continue;
222                         }
223                         *rtp = rt->dst.dn_next;
224                         rt->dst.dn_next = NULL;
225                         dnrt_drop(rt);
226                         break;
227                 }
228                 spin_unlock_bh(&dn_rt_hash_table[i].lock);
229         }
230
231         return 0;
232 }
233
234 /*
235  * The decnet standards don't impose a particular minimum mtu, what they
236  * do insist on is that the routing layer accepts a datagram of at least
237  * 230 bytes long. Here we have to subtract the routing header length from
238  * 230 to get the minimum acceptable mtu. If there is no neighbour, then we
239  * assume the worst and use a long header size.
240  *
241  * We update both the mtu and the advertised mss (i.e. the segment size we
242  * advertise to the other end).
243  */
244 static void dn_dst_update_pmtu(struct dst_entry *dst, u32 mtu)
245 {
246         struct neighbour *n = dst_get_neighbour(dst);
247         u32 min_mtu = 230;
248         struct dn_dev *dn;
249
250         dn = n ? rcu_dereference_raw(n->dev->dn_ptr) : NULL;
251
252         if (dn && dn->use_long == 0)
253                 min_mtu -= 6;
254         else
255                 min_mtu -= 21;
256
257         if (dst_metric(dst, RTAX_MTU) > mtu && mtu >= min_mtu) {
258                 if (!(dst_metric_locked(dst, RTAX_MTU))) {
259                         dst_metric_set(dst, RTAX_MTU, mtu);
260                         dst_set_expires(dst, dn_rt_mtu_expires);
261                 }
262                 if (!(dst_metric_locked(dst, RTAX_ADVMSS))) {
263                         u32 mss = mtu - DN_MAX_NSP_DATA_HEADER;
264                         u32 existing_mss = dst_metric_raw(dst, RTAX_ADVMSS);
265                         if (!existing_mss || existing_mss > mss)
266                                 dst_metric_set(dst, RTAX_ADVMSS, mss);
267                 }
268         }
269 }
270
271 /*
272  * When a route has been marked obsolete. (e.g. routing cache flush)
273  */
274 static struct dst_entry *dn_dst_check(struct dst_entry *dst, __u32 cookie)
275 {
276         return NULL;
277 }
278
279 static struct dst_entry *dn_dst_negative_advice(struct dst_entry *dst)
280 {
281         dst_release(dst);
282         return NULL;
283 }
284
285 static void dn_dst_link_failure(struct sk_buff *skb)
286 {
287 }
288
289 static inline int compare_keys(struct flowidn *fl1, struct flowidn *fl2)
290 {
291         return ((fl1->daddr ^ fl2->daddr) |
292                 (fl1->saddr ^ fl2->saddr) |
293                 (fl1->flowidn_mark ^ fl2->flowidn_mark) |
294                 (fl1->flowidn_scope ^ fl2->flowidn_scope) |
295                 (fl1->flowidn_oif ^ fl2->flowidn_oif) |
296                 (fl1->flowidn_iif ^ fl2->flowidn_iif)) == 0;
297 }
298
299 static int dn_insert_route(struct dn_route *rt, unsigned hash, struct dn_route **rp)
300 {
301         struct dn_route *rth;
302         struct dn_route __rcu **rthp;
303         unsigned long now = jiffies;
304
305         rthp = &dn_rt_hash_table[hash].chain;
306
307         spin_lock_bh(&dn_rt_hash_table[hash].lock);
308         while ((rth = rcu_dereference_protected(*rthp,
309                                                 lockdep_is_held(&dn_rt_hash_table[hash].lock))) != NULL) {
310                 if (compare_keys(&rth->fld, &rt->fld)) {
311                         /* Put it first */
312                         *rthp = rth->dst.dn_next;
313                         rcu_assign_pointer(rth->dst.dn_next,
314                                            dn_rt_hash_table[hash].chain);
315                         rcu_assign_pointer(dn_rt_hash_table[hash].chain, rth);
316
317                         dst_use(&rth->dst, now);
318                         spin_unlock_bh(&dn_rt_hash_table[hash].lock);
319
320                         dnrt_drop(rt);
321                         *rp = rth;
322                         return 0;
323                 }
324                 rthp = &rth->dst.dn_next;
325         }
326
327         rcu_assign_pointer(rt->dst.dn_next, dn_rt_hash_table[hash].chain);
328         rcu_assign_pointer(dn_rt_hash_table[hash].chain, rt);
329
330         dst_use(&rt->dst, now);
331         spin_unlock_bh(&dn_rt_hash_table[hash].lock);
332         *rp = rt;
333         return 0;
334 }
335
336 static void dn_run_flush(unsigned long dummy)
337 {
338         int i;
339         struct dn_route *rt, *next;
340
341         for (i = 0; i < dn_rt_hash_mask; i++) {
342                 spin_lock_bh(&dn_rt_hash_table[i].lock);
343
344                 if ((rt = xchg((struct dn_route **)&dn_rt_hash_table[i].chain, NULL)) == NULL)
345                         goto nothing_to_declare;
346
347                 for(; rt; rt = next) {
348                         next = rcu_dereference_raw(rt->dst.dn_next);
349                         RCU_INIT_POINTER(rt->dst.dn_next, NULL);
350                         dst_free((struct dst_entry *)rt);
351                 }
352
353 nothing_to_declare:
354                 spin_unlock_bh(&dn_rt_hash_table[i].lock);
355         }
356 }
357
358 static DEFINE_SPINLOCK(dn_rt_flush_lock);
359
360 void dn_rt_cache_flush(int delay)
361 {
362         unsigned long now = jiffies;
363         int user_mode = !in_interrupt();
364
365         if (delay < 0)
366                 delay = dn_rt_min_delay;
367
368         spin_lock_bh(&dn_rt_flush_lock);
369
370         if (del_timer(&dn_rt_flush_timer) && delay > 0 && dn_rt_deadline) {
371                 long tmo = (long)(dn_rt_deadline - now);
372
373                 if (user_mode && tmo < dn_rt_max_delay - dn_rt_min_delay)
374                         tmo = 0;
375
376                 if (delay > tmo)
377                         delay = tmo;
378         }
379
380         if (delay <= 0) {
381                 spin_unlock_bh(&dn_rt_flush_lock);
382                 dn_run_flush(0);
383                 return;
384         }
385
386         if (dn_rt_deadline == 0)
387                 dn_rt_deadline = now + dn_rt_max_delay;
388
389         dn_rt_flush_timer.expires = now + delay;
390         add_timer(&dn_rt_flush_timer);
391         spin_unlock_bh(&dn_rt_flush_lock);
392 }
393
394 /**
395  * dn_return_short - Return a short packet to its sender
396  * @skb: The packet to return
397  *
398  */
399 static int dn_return_short(struct sk_buff *skb)
400 {
401         struct dn_skb_cb *cb;
402         unsigned char *ptr;
403         __le16 *src;
404         __le16 *dst;
405
406         /* Add back headers */
407         skb_push(skb, skb->data - skb_network_header(skb));
408
409         if ((skb = skb_unshare(skb, GFP_ATOMIC)) == NULL)
410                 return NET_RX_DROP;
411
412         cb = DN_SKB_CB(skb);
413         /* Skip packet length and point to flags */
414         ptr = skb->data + 2;
415         *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
416
417         dst = (__le16 *)ptr;
418         ptr += 2;
419         src = (__le16 *)ptr;
420         ptr += 2;
421         *ptr = 0; /* Zero hop count */
422
423         swap(*src, *dst);
424
425         skb->pkt_type = PACKET_OUTGOING;
426         dn_rt_finish_output(skb, NULL, NULL);
427         return NET_RX_SUCCESS;
428 }
429
430 /**
431  * dn_return_long - Return a long packet to its sender
432  * @skb: The long format packet to return
433  *
434  */
435 static int dn_return_long(struct sk_buff *skb)
436 {
437         struct dn_skb_cb *cb;
438         unsigned char *ptr;
439         unsigned char *src_addr, *dst_addr;
440         unsigned char tmp[ETH_ALEN];
441
442         /* Add back all headers */
443         skb_push(skb, skb->data - skb_network_header(skb));
444
445         if ((skb = skb_unshare(skb, GFP_ATOMIC)) == NULL)
446                 return NET_RX_DROP;
447
448         cb = DN_SKB_CB(skb);
449         /* Ignore packet length and point to flags */
450         ptr = skb->data + 2;
451
452         /* Skip padding */
453         if (*ptr & DN_RT_F_PF) {
454                 char padlen = (*ptr & ~DN_RT_F_PF);
455                 ptr += padlen;
456         }
457
458         *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
459         ptr += 2;
460         dst_addr = ptr;
461         ptr += 8;
462         src_addr = ptr;
463         ptr += 6;
464         *ptr = 0; /* Zero hop count */
465
466         /* Swap source and destination */
467         memcpy(tmp, src_addr, ETH_ALEN);
468         memcpy(src_addr, dst_addr, ETH_ALEN);
469         memcpy(dst_addr, tmp, ETH_ALEN);
470
471         skb->pkt_type = PACKET_OUTGOING;
472         dn_rt_finish_output(skb, dst_addr, src_addr);
473         return NET_RX_SUCCESS;
474 }
475
476 /**
477  * dn_route_rx_packet - Try and find a route for an incoming packet
478  * @skb: The packet to find a route for
479  *
480  * Returns: result of input function if route is found, error code otherwise
481  */
482 static int dn_route_rx_packet(struct sk_buff *skb)
483 {
484         struct dn_skb_cb *cb;
485         int err;
486
487         if ((err = dn_route_input(skb)) == 0)
488                 return dst_input(skb);
489
490         cb = DN_SKB_CB(skb);
491         if (decnet_debug_level & 4) {
492                 char *devname = skb->dev ? skb->dev->name : "???";
493
494                 printk(KERN_DEBUG
495                         "DECnet: dn_route_rx_packet: rt_flags=0x%02x dev=%s len=%d src=0x%04hx dst=0x%04hx err=%d type=%d\n",
496                         (int)cb->rt_flags, devname, skb->len,
497                         le16_to_cpu(cb->src), le16_to_cpu(cb->dst),
498                         err, skb->pkt_type);
499         }
500
501         if ((skb->pkt_type == PACKET_HOST) && (cb->rt_flags & DN_RT_F_RQR)) {
502                 switch (cb->rt_flags & DN_RT_PKT_MSK) {
503                 case DN_RT_PKT_SHORT:
504                         return dn_return_short(skb);
505                 case DN_RT_PKT_LONG:
506                         return dn_return_long(skb);
507                 }
508         }
509
510         kfree_skb(skb);
511         return NET_RX_DROP;
512 }
513
514 static int dn_route_rx_long(struct sk_buff *skb)
515 {
516         struct dn_skb_cb *cb = DN_SKB_CB(skb);
517         unsigned char *ptr = skb->data;
518
519         if (!pskb_may_pull(skb, 21)) /* 20 for long header, 1 for shortest nsp */
520                 goto drop_it;
521
522         skb_pull(skb, 20);
523         skb_reset_transport_header(skb);
524
525         /* Destination info */
526         ptr += 2;
527         cb->dst = dn_eth2dn(ptr);
528         if (memcmp(ptr, dn_hiord_addr, 4) != 0)
529                 goto drop_it;
530         ptr += 6;
531
532
533         /* Source info */
534         ptr += 2;
535         cb->src = dn_eth2dn(ptr);
536         if (memcmp(ptr, dn_hiord_addr, 4) != 0)
537                 goto drop_it;
538         ptr += 6;
539         /* Other junk */
540         ptr++;
541         cb->hops = *ptr++; /* Visit Count */
542
543         return NF_HOOK(NFPROTO_DECNET, NF_DN_PRE_ROUTING, skb, skb->dev, NULL,
544                        dn_route_rx_packet);
545
546 drop_it:
547         kfree_skb(skb);
548         return NET_RX_DROP;
549 }
550
551
552
553 static int dn_route_rx_short(struct sk_buff *skb)
554 {
555         struct dn_skb_cb *cb = DN_SKB_CB(skb);
556         unsigned char *ptr = skb->data;
557
558         if (!pskb_may_pull(skb, 6)) /* 5 for short header + 1 for shortest nsp */
559                 goto drop_it;
560
561         skb_pull(skb, 5);
562         skb_reset_transport_header(skb);
563
564         cb->dst = *(__le16 *)ptr;
565         ptr += 2;
566         cb->src = *(__le16 *)ptr;
567         ptr += 2;
568         cb->hops = *ptr & 0x3f;
569
570         return NF_HOOK(NFPROTO_DECNET, NF_DN_PRE_ROUTING, skb, skb->dev, NULL,
571                        dn_route_rx_packet);
572
573 drop_it:
574         kfree_skb(skb);
575         return NET_RX_DROP;
576 }
577
578 static int dn_route_discard(struct sk_buff *skb)
579 {
580         /*
581          * I know we drop the packet here, but thats considered success in
582          * this case
583          */
584         kfree_skb(skb);
585         return NET_RX_SUCCESS;
586 }
587
588 static int dn_route_ptp_hello(struct sk_buff *skb)
589 {
590         dn_dev_hello(skb);
591         dn_neigh_pointopoint_hello(skb);
592         return NET_RX_SUCCESS;
593 }
594
595 int dn_route_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
596 {
597         struct dn_skb_cb *cb;
598         unsigned char flags = 0;
599         __u16 len = le16_to_cpu(*(__le16 *)skb->data);
600         struct dn_dev *dn = rcu_dereference(dev->dn_ptr);
601         unsigned char padlen = 0;
602
603         if (!net_eq(dev_net(dev), &init_net))
604                 goto dump_it;
605
606         if (dn == NULL)
607                 goto dump_it;
608
609         if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
610                 goto out;
611
612         if (!pskb_may_pull(skb, 3))
613                 goto dump_it;
614
615         skb_pull(skb, 2);
616
617         if (len > skb->len)
618                 goto dump_it;
619
620         skb_trim(skb, len);
621
622         flags = *skb->data;
623
624         cb = DN_SKB_CB(skb);
625         cb->stamp = jiffies;
626         cb->iif = dev->ifindex;
627
628         /*
629          * If we have padding, remove it.
630          */
631         if (flags & DN_RT_F_PF) {
632                 padlen = flags & ~DN_RT_F_PF;
633                 if (!pskb_may_pull(skb, padlen + 1))
634                         goto dump_it;
635                 skb_pull(skb, padlen);
636                 flags = *skb->data;
637         }
638
639         skb_reset_network_header(skb);
640
641         /*
642          * Weed out future version DECnet
643          */
644         if (flags & DN_RT_F_VER)
645                 goto dump_it;
646
647         cb->rt_flags = flags;
648
649         if (decnet_debug_level & 1)
650                 printk(KERN_DEBUG
651                         "dn_route_rcv: got 0x%02x from %s [%d %d %d]\n",
652                         (int)flags, (dev) ? dev->name : "???", len, skb->len,
653                         padlen);
654
655         if (flags & DN_RT_PKT_CNTL) {
656                 if (unlikely(skb_linearize(skb)))
657                         goto dump_it;
658
659                 switch (flags & DN_RT_CNTL_MSK) {
660                 case DN_RT_PKT_INIT:
661                         dn_dev_init_pkt(skb);
662                         break;
663                 case DN_RT_PKT_VERI:
664                         dn_dev_veri_pkt(skb);
665                         break;
666                 }
667
668                 if (dn->parms.state != DN_DEV_S_RU)
669                         goto dump_it;
670
671                 switch (flags & DN_RT_CNTL_MSK) {
672                 case DN_RT_PKT_HELO:
673                         return NF_HOOK(NFPROTO_DECNET, NF_DN_HELLO,
674                                        skb, skb->dev, NULL,
675                                        dn_route_ptp_hello);
676
677                 case DN_RT_PKT_L1RT:
678                 case DN_RT_PKT_L2RT:
679                         return NF_HOOK(NFPROTO_DECNET, NF_DN_ROUTE,
680                                        skb, skb->dev, NULL,
681                                        dn_route_discard);
682                 case DN_RT_PKT_ERTH:
683                         return NF_HOOK(NFPROTO_DECNET, NF_DN_HELLO,
684                                        skb, skb->dev, NULL,
685                                        dn_neigh_router_hello);
686
687                 case DN_RT_PKT_EEDH:
688                         return NF_HOOK(NFPROTO_DECNET, NF_DN_HELLO,
689                                        skb, skb->dev, NULL,
690                                        dn_neigh_endnode_hello);
691                 }
692         } else {
693                 if (dn->parms.state != DN_DEV_S_RU)
694                         goto dump_it;
695
696                 skb_pull(skb, 1); /* Pull flags */
697
698                 switch (flags & DN_RT_PKT_MSK) {
699                 case DN_RT_PKT_LONG:
700                         return dn_route_rx_long(skb);
701                 case DN_RT_PKT_SHORT:
702                         return dn_route_rx_short(skb);
703                 }
704         }
705
706 dump_it:
707         kfree_skb(skb);
708 out:
709         return NET_RX_DROP;
710 }
711
712 static int dn_to_neigh_output(struct sk_buff *skb)
713 {
714         struct dst_entry *dst = skb_dst(skb);
715         struct neighbour *n = dst_get_neighbour(dst);
716
717         return n->output(n, skb);
718 }
719
720 static int dn_output(struct sk_buff *skb)
721 {
722         struct dst_entry *dst = skb_dst(skb);
723         struct dn_route *rt = (struct dn_route *)dst;
724         struct net_device *dev = dst->dev;
725         struct dn_skb_cb *cb = DN_SKB_CB(skb);
726         struct neighbour *neigh;
727
728         int err = -EINVAL;
729
730         if ((neigh = dst_get_neighbour(dst)) == NULL)
731                 goto error;
732
733         skb->dev = dev;
734
735         cb->src = rt->rt_saddr;
736         cb->dst = rt->rt_daddr;
737
738         /*
739          * Always set the Intra-Ethernet bit on all outgoing packets
740          * originated on this node. Only valid flag from upper layers
741          * is return-to-sender-requested. Set hop count to 0 too.
742          */
743         cb->rt_flags &= ~DN_RT_F_RQR;
744         cb->rt_flags |= DN_RT_F_IE;
745         cb->hops = 0;
746
747         return NF_HOOK(NFPROTO_DECNET, NF_DN_LOCAL_OUT, skb, NULL, dev,
748                        dn_to_neigh_output);
749
750 error:
751         if (net_ratelimit())
752                 printk(KERN_DEBUG "dn_output: This should not happen\n");
753
754         kfree_skb(skb);
755
756         return err;
757 }
758
759 static int dn_forward(struct sk_buff *skb)
760 {
761         struct dn_skb_cb *cb = DN_SKB_CB(skb);
762         struct dst_entry *dst = skb_dst(skb);
763         struct dn_dev *dn_db = rcu_dereference(dst->dev->dn_ptr);
764         struct dn_route *rt;
765         int header_len;
766 #ifdef CONFIG_NETFILTER
767         struct net_device *dev = skb->dev;
768 #endif
769
770         if (skb->pkt_type != PACKET_HOST)
771                 goto drop;
772
773         /* Ensure that we have enough space for headers */
774         rt = (struct dn_route *)skb_dst(skb);
775         header_len = dn_db->use_long ? 21 : 6;
776         if (skb_cow(skb, LL_RESERVED_SPACE(rt->dst.dev)+header_len))
777                 goto drop;
778
779         /*
780          * Hop count exceeded.
781          */
782         if (++cb->hops > 30)
783                 goto drop;
784
785         skb->dev = rt->dst.dev;
786
787         /*
788          * If packet goes out same interface it came in on, then set
789          * the Intra-Ethernet bit. This has no effect for short
790          * packets, so we don't need to test for them here.
791          */
792         cb->rt_flags &= ~DN_RT_F_IE;
793         if (rt->rt_flags & RTCF_DOREDIRECT)
794                 cb->rt_flags |= DN_RT_F_IE;
795
796         return NF_HOOK(NFPROTO_DECNET, NF_DN_FORWARD, skb, dev, skb->dev,
797                        dn_to_neigh_output);
798
799 drop:
800         kfree_skb(skb);
801         return NET_RX_DROP;
802 }
803
804 /*
805  * Used to catch bugs. This should never normally get
806  * called.
807  */
808 static int dn_rt_bug(struct sk_buff *skb)
809 {
810         if (net_ratelimit()) {
811                 struct dn_skb_cb *cb = DN_SKB_CB(skb);
812
813                 printk(KERN_DEBUG "dn_rt_bug: skb from:%04x to:%04x\n",
814                                 le16_to_cpu(cb->src), le16_to_cpu(cb->dst));
815         }
816
817         kfree_skb(skb);
818
819         return NET_RX_DROP;
820 }
821
822 static unsigned int dn_dst_default_advmss(const struct dst_entry *dst)
823 {
824         return dn_mss_from_pmtu(dst->dev, dst_mtu(dst));
825 }
826
827 static unsigned int dn_dst_default_mtu(const struct dst_entry *dst)
828 {
829         return dst->dev->mtu;
830 }
831
832 static struct neighbour *dn_dst_neigh_lookup(const struct dst_entry *dst, const void *daddr)
833 {
834         return __neigh_lookup_errno(&dn_neigh_table, daddr, dst->dev);
835 }
836
837 static int dn_rt_set_next_hop(struct dn_route *rt, struct dn_fib_res *res)
838 {
839         struct dn_fib_info *fi = res->fi;
840         struct net_device *dev = rt->dst.dev;
841         unsigned int mss_metric;
842         struct neighbour *n;
843
844         if (fi) {
845                 if (DN_FIB_RES_GW(*res) &&
846                     DN_FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
847                         rt->rt_gateway = DN_FIB_RES_GW(*res);
848                 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
849         }
850         rt->rt_type = res->type;
851
852         if (dev != NULL && dst_get_neighbour(&rt->dst) == NULL) {
853                 n = __neigh_lookup_errno(&dn_neigh_table, &rt->rt_gateway, dev);
854                 if (IS_ERR(n))
855                         return PTR_ERR(n);
856                 dst_set_neighbour(&rt->dst, n);
857         }
858
859         if (dst_metric(&rt->dst, RTAX_MTU) > rt->dst.dev->mtu)
860                 dst_metric_set(&rt->dst, RTAX_MTU, rt->dst.dev->mtu);
861         mss_metric = dst_metric_raw(&rt->dst, RTAX_ADVMSS);
862         if (mss_metric) {
863                 unsigned int mss = dn_mss_from_pmtu(dev, dst_mtu(&rt->dst));
864                 if (mss_metric > mss)
865                         dst_metric_set(&rt->dst, RTAX_ADVMSS, mss);
866         }
867         return 0;
868 }
869
870 static inline int dn_match_addr(__le16 addr1, __le16 addr2)
871 {
872         __u16 tmp = le16_to_cpu(addr1) ^ le16_to_cpu(addr2);
873         int match = 16;
874         while(tmp) {
875                 tmp >>= 1;
876                 match--;
877         }
878         return match;
879 }
880
881 static __le16 dnet_select_source(const struct net_device *dev, __le16 daddr, int scope)
882 {
883         __le16 saddr = 0;
884         struct dn_dev *dn_db;
885         struct dn_ifaddr *ifa;
886         int best_match = 0;
887         int ret;
888
889         rcu_read_lock();
890         dn_db = rcu_dereference(dev->dn_ptr);
891         for (ifa = rcu_dereference(dn_db->ifa_list);
892              ifa != NULL;
893              ifa = rcu_dereference(ifa->ifa_next)) {
894                 if (ifa->ifa_scope > scope)
895                         continue;
896                 if (!daddr) {
897                         saddr = ifa->ifa_local;
898                         break;
899                 }
900                 ret = dn_match_addr(daddr, ifa->ifa_local);
901                 if (ret > best_match)
902                         saddr = ifa->ifa_local;
903                 if (best_match == 0)
904                         saddr = ifa->ifa_local;
905         }
906         rcu_read_unlock();
907
908         return saddr;
909 }
910
911 static inline __le16 __dn_fib_res_prefsrc(struct dn_fib_res *res)
912 {
913         return dnet_select_source(DN_FIB_RES_DEV(*res), DN_FIB_RES_GW(*res), res->scope);
914 }
915
916 static inline __le16 dn_fib_rules_map_destination(__le16 daddr, struct dn_fib_res *res)
917 {
918         __le16 mask = dnet_make_mask(res->prefixlen);
919         return (daddr&~mask)|res->fi->fib_nh->nh_gw;
920 }
921
922 static int dn_route_output_slow(struct dst_entry **pprt, const struct flowidn *oldflp, int try_hard)
923 {
924         struct flowidn fld = {
925                 .daddr = oldflp->daddr,
926                 .saddr = oldflp->saddr,
927                 .flowidn_scope = RT_SCOPE_UNIVERSE,
928                 .flowidn_mark = oldflp->flowidn_mark,
929                 .flowidn_iif = init_net.loopback_dev->ifindex,
930                 .flowidn_oif = oldflp->flowidn_oif,
931         };
932         struct dn_route *rt = NULL;
933         struct net_device *dev_out = NULL, *dev;
934         struct neighbour *neigh = NULL;
935         unsigned hash;
936         unsigned flags = 0;
937         struct dn_fib_res res = { .fi = NULL, .type = RTN_UNICAST };
938         int err;
939         int free_res = 0;
940         __le16 gateway = 0;
941
942         if (decnet_debug_level & 16)
943                 printk(KERN_DEBUG
944                        "dn_route_output_slow: dst=%04x src=%04x mark=%d"
945                        " iif=%d oif=%d\n", le16_to_cpu(oldflp->daddr),
946                        le16_to_cpu(oldflp->saddr),
947                        oldflp->flowidn_mark, init_net.loopback_dev->ifindex,
948                        oldflp->flowidn_oif);
949
950         /* If we have an output interface, verify its a DECnet device */
951         if (oldflp->flowidn_oif) {
952                 dev_out = dev_get_by_index(&init_net, oldflp->flowidn_oif);
953                 err = -ENODEV;
954                 if (dev_out && dev_out->dn_ptr == NULL) {
955                         dev_put(dev_out);
956                         dev_out = NULL;
957                 }
958                 if (dev_out == NULL)
959                         goto out;
960         }
961
962         /* If we have a source address, verify that its a local address */
963         if (oldflp->saddr) {
964                 err = -EADDRNOTAVAIL;
965
966                 if (dev_out) {
967                         if (dn_dev_islocal(dev_out, oldflp->saddr))
968                                 goto source_ok;
969                         dev_put(dev_out);
970                         goto out;
971                 }
972                 rcu_read_lock();
973                 for_each_netdev_rcu(&init_net, dev) {
974                         if (!dev->dn_ptr)
975                                 continue;
976                         if (!dn_dev_islocal(dev, oldflp->saddr))
977                                 continue;
978                         if ((dev->flags & IFF_LOOPBACK) &&
979                             oldflp->daddr &&
980                             !dn_dev_islocal(dev, oldflp->daddr))
981                                 continue;
982
983                         dev_out = dev;
984                         break;
985                 }
986                 rcu_read_unlock();
987                 if (dev_out == NULL)
988                         goto out;
989                 dev_hold(dev_out);
990 source_ok:
991                 ;
992         }
993
994         /* No destination? Assume its local */
995         if (!fld.daddr) {
996                 fld.daddr = fld.saddr;
997
998                 err = -EADDRNOTAVAIL;
999                 if (dev_out)
1000                         dev_put(dev_out);
1001                 dev_out = init_net.loopback_dev;
1002                 dev_hold(dev_out);
1003                 if (!fld.daddr) {
1004                         fld.daddr =
1005                         fld.saddr = dnet_select_source(dev_out, 0,
1006                                                        RT_SCOPE_HOST);
1007                         if (!fld.daddr)
1008                                 goto out;
1009                 }
1010                 fld.flowidn_oif = init_net.loopback_dev->ifindex;
1011                 res.type = RTN_LOCAL;
1012                 goto make_route;
1013         }
1014
1015         if (decnet_debug_level & 16)
1016                 printk(KERN_DEBUG
1017                        "dn_route_output_slow: initial checks complete."
1018                        " dst=%o4x src=%04x oif=%d try_hard=%d\n",
1019                        le16_to_cpu(fld.daddr), le16_to_cpu(fld.saddr),
1020                        fld.flowidn_oif, try_hard);
1021
1022         /*
1023          * N.B. If the kernel is compiled without router support then
1024          * dn_fib_lookup() will evaluate to non-zero so this if () block
1025          * will always be executed.
1026          */
1027         err = -ESRCH;
1028         if (try_hard || (err = dn_fib_lookup(&fld, &res)) != 0) {
1029                 struct dn_dev *dn_db;
1030                 if (err != -ESRCH)
1031                         goto out;
1032                 /*
1033                  * Here the fallback is basically the standard algorithm for
1034                  * routing in endnodes which is described in the DECnet routing
1035                  * docs
1036                  *
1037                  * If we are not trying hard, look in neighbour cache.
1038                  * The result is tested to ensure that if a specific output
1039                  * device/source address was requested, then we honour that
1040                  * here
1041                  */
1042                 if (!try_hard) {
1043                         neigh = neigh_lookup_nodev(&dn_neigh_table, &init_net, &fld.daddr);
1044                         if (neigh) {
1045                                 if ((oldflp->flowidn_oif &&
1046                                     (neigh->dev->ifindex != oldflp->flowidn_oif)) ||
1047                                     (oldflp->saddr &&
1048                                     (!dn_dev_islocal(neigh->dev,
1049                                                      oldflp->saddr)))) {
1050                                         neigh_release(neigh);
1051                                         neigh = NULL;
1052                                 } else {
1053                                         if (dev_out)
1054                                                 dev_put(dev_out);
1055                                         if (dn_dev_islocal(neigh->dev, fld.daddr)) {
1056                                                 dev_out = init_net.loopback_dev;
1057                                                 res.type = RTN_LOCAL;
1058                                         } else {
1059                                                 dev_out = neigh->dev;
1060                                         }
1061                                         dev_hold(dev_out);
1062                                         goto select_source;
1063                                 }
1064                         }
1065                 }
1066
1067                 /* Not there? Perhaps its a local address */
1068                 if (dev_out == NULL)
1069                         dev_out = dn_dev_get_default();
1070                 err = -ENODEV;
1071                 if (dev_out == NULL)
1072                         goto out;
1073                 dn_db = rcu_dereference_raw(dev_out->dn_ptr);
1074                 /* Possible improvement - check all devices for local addr */
1075                 if (dn_dev_islocal(dev_out, fld.daddr)) {
1076                         dev_put(dev_out);
1077                         dev_out = init_net.loopback_dev;
1078                         dev_hold(dev_out);
1079                         res.type = RTN_LOCAL;
1080                         goto select_source;
1081                 }
1082                 /* Not local either.... try sending it to the default router */
1083                 neigh = neigh_clone(dn_db->router);
1084                 BUG_ON(neigh && neigh->dev != dev_out);
1085
1086                 /* Ok then, we assume its directly connected and move on */
1087 select_source:
1088                 if (neigh)
1089                         gateway = ((struct dn_neigh *)neigh)->addr;
1090                 if (gateway == 0)
1091                         gateway = fld.daddr;
1092                 if (fld.saddr == 0) {
1093                         fld.saddr = dnet_select_source(dev_out, gateway,
1094                                                        res.type == RTN_LOCAL ?
1095                                                        RT_SCOPE_HOST :
1096                                                        RT_SCOPE_LINK);
1097                         if (fld.saddr == 0 && res.type != RTN_LOCAL)
1098                                 goto e_addr;
1099                 }
1100                 fld.flowidn_oif = dev_out->ifindex;
1101                 goto make_route;
1102         }
1103         free_res = 1;
1104
1105         if (res.type == RTN_NAT)
1106                 goto e_inval;
1107
1108         if (res.type == RTN_LOCAL) {
1109                 if (!fld.saddr)
1110                         fld.saddr = fld.daddr;
1111                 if (dev_out)
1112                         dev_put(dev_out);
1113                 dev_out = init_net.loopback_dev;
1114                 dev_hold(dev_out);
1115                 fld.flowidn_oif = dev_out->ifindex;
1116                 if (res.fi)
1117                         dn_fib_info_put(res.fi);
1118                 res.fi = NULL;
1119                 goto make_route;
1120         }
1121
1122         if (res.fi->fib_nhs > 1 && fld.flowidn_oif == 0)
1123                 dn_fib_select_multipath(&fld, &res);
1124
1125         /*
1126          * We could add some logic to deal with default routes here and
1127          * get rid of some of the special casing above.
1128          */
1129
1130         if (!fld.saddr)
1131                 fld.saddr = DN_FIB_RES_PREFSRC(res);
1132
1133         if (dev_out)
1134                 dev_put(dev_out);
1135         dev_out = DN_FIB_RES_DEV(res);
1136         dev_hold(dev_out);
1137         fld.flowidn_oif = dev_out->ifindex;
1138         gateway = DN_FIB_RES_GW(res);
1139
1140 make_route:
1141         if (dev_out->flags & IFF_LOOPBACK)
1142                 flags |= RTCF_LOCAL;
1143
1144         rt = dst_alloc(&dn_dst_ops, dev_out, 1, 0, DST_HOST);
1145         if (rt == NULL)
1146                 goto e_nobufs;
1147
1148         memset(&rt->fld, 0, sizeof(rt->fld));
1149         rt->fld.saddr        = oldflp->saddr;
1150         rt->fld.daddr        = oldflp->daddr;
1151         rt->fld.flowidn_oif  = oldflp->flowidn_oif;
1152         rt->fld.flowidn_iif  = 0;
1153         rt->fld.flowidn_mark = oldflp->flowidn_mark;
1154
1155         rt->rt_saddr      = fld.saddr;
1156         rt->rt_daddr      = fld.daddr;
1157         rt->rt_gateway    = gateway ? gateway : fld.daddr;
1158         rt->rt_local_src  = fld.saddr;
1159
1160         rt->rt_dst_map    = fld.daddr;
1161         rt->rt_src_map    = fld.saddr;
1162
1163         dst_set_neighbour(&rt->dst, neigh);
1164         neigh = NULL;
1165
1166         rt->dst.lastuse = jiffies;
1167         rt->dst.output  = dn_output;
1168         rt->dst.input   = dn_rt_bug;
1169         rt->rt_flags      = flags;
1170         if (flags & RTCF_LOCAL)
1171                 rt->dst.input = dn_nsp_rx;
1172
1173         err = dn_rt_set_next_hop(rt, &res);
1174         if (err)
1175                 goto e_neighbour;
1176
1177         hash = dn_hash(rt->fld.saddr, rt->fld.daddr);
1178         dn_insert_route(rt, hash, (struct dn_route **)pprt);
1179
1180 done:
1181         if (neigh)
1182                 neigh_release(neigh);
1183         if (free_res)
1184                 dn_fib_res_put(&res);
1185         if (dev_out)
1186                 dev_put(dev_out);
1187 out:
1188         return err;
1189
1190 e_addr:
1191         err = -EADDRNOTAVAIL;
1192         goto done;
1193 e_inval:
1194         err = -EINVAL;
1195         goto done;
1196 e_nobufs:
1197         err = -ENOBUFS;
1198         goto done;
1199 e_neighbour:
1200         dst_free(&rt->dst);
1201         goto e_nobufs;
1202 }
1203
1204
1205 /*
1206  * N.B. The flags may be moved into the flowi at some future stage.
1207  */
1208 static int __dn_route_output_key(struct dst_entry **pprt, const struct flowidn *flp, int flags)
1209 {
1210         unsigned hash = dn_hash(flp->saddr, flp->daddr);
1211         struct dn_route *rt = NULL;
1212
1213         if (!(flags & MSG_TRYHARD)) {
1214                 rcu_read_lock_bh();
1215                 for (rt = rcu_dereference_bh(dn_rt_hash_table[hash].chain); rt;
1216                         rt = rcu_dereference_bh(rt->dst.dn_next)) {
1217                         if ((flp->daddr == rt->fld.daddr) &&
1218                             (flp->saddr == rt->fld.saddr) &&
1219                             (flp->flowidn_mark == rt->fld.flowidn_mark) &&
1220                             dn_is_output_route(rt) &&
1221                             (rt->fld.flowidn_oif == flp->flowidn_oif)) {
1222                                 dst_use(&rt->dst, jiffies);
1223                                 rcu_read_unlock_bh();
1224                                 *pprt = &rt->dst;
1225                                 return 0;
1226                         }
1227                 }
1228                 rcu_read_unlock_bh();
1229         }
1230
1231         return dn_route_output_slow(pprt, flp, flags);
1232 }
1233
1234 static int dn_route_output_key(struct dst_entry **pprt, struct flowidn *flp, int flags)
1235 {
1236         int err;
1237
1238         err = __dn_route_output_key(pprt, flp, flags);
1239         if (err == 0 && flp->flowidn_proto) {
1240                 *pprt = xfrm_lookup(&init_net, *pprt,
1241                                     flowidn_to_flowi(flp), NULL, 0);
1242                 if (IS_ERR(*pprt)) {
1243                         err = PTR_ERR(*pprt);
1244                         *pprt = NULL;
1245                 }
1246         }
1247         return err;
1248 }
1249
1250 int dn_route_output_sock(struct dst_entry **pprt, struct flowidn *fl, struct sock *sk, int flags)
1251 {
1252         int err;
1253
1254         err = __dn_route_output_key(pprt, fl, flags & MSG_TRYHARD);
1255         if (err == 0 && fl->flowidn_proto) {
1256                 if (!(flags & MSG_DONTWAIT))
1257                         fl->flowidn_flags |= FLOWI_FLAG_CAN_SLEEP;
1258                 *pprt = xfrm_lookup(&init_net, *pprt,
1259                                     flowidn_to_flowi(fl), sk, 0);
1260                 if (IS_ERR(*pprt)) {
1261                         err = PTR_ERR(*pprt);
1262                         *pprt = NULL;
1263                 }
1264         }
1265         return err;
1266 }
1267
1268 static int dn_route_input_slow(struct sk_buff *skb)
1269 {
1270         struct dn_route *rt = NULL;
1271         struct dn_skb_cb *cb = DN_SKB_CB(skb);
1272         struct net_device *in_dev = skb->dev;
1273         struct net_device *out_dev = NULL;
1274         struct dn_dev *dn_db;
1275         struct neighbour *neigh = NULL;
1276         unsigned hash;
1277         int flags = 0;
1278         __le16 gateway = 0;
1279         __le16 local_src = 0;
1280         struct flowidn fld = {
1281                 .daddr = cb->dst,
1282                 .saddr = cb->src,
1283                 .flowidn_scope = RT_SCOPE_UNIVERSE,
1284                 .flowidn_mark = skb->mark,
1285                 .flowidn_iif = skb->dev->ifindex,
1286         };
1287         struct dn_fib_res res = { .fi = NULL, .type = RTN_UNREACHABLE };
1288         int err = -EINVAL;
1289         int free_res = 0;
1290
1291         dev_hold(in_dev);
1292
1293         if ((dn_db = rcu_dereference(in_dev->dn_ptr)) == NULL)
1294                 goto out;
1295
1296         /* Zero source addresses are not allowed */
1297         if (fld.saddr == 0)
1298                 goto out;
1299
1300         /*
1301          * In this case we've just received a packet from a source
1302          * outside ourselves pretending to come from us. We don't
1303          * allow it any further to prevent routing loops, spoofing and
1304          * other nasties. Loopback packets already have the dst attached
1305          * so this only affects packets which have originated elsewhere.
1306          */
1307         err  = -ENOTUNIQ;
1308         if (dn_dev_islocal(in_dev, cb->src))
1309                 goto out;
1310
1311         err = dn_fib_lookup(&fld, &res);
1312         if (err) {
1313                 if (err != -ESRCH)
1314                         goto out;
1315                 /*
1316                  * Is the destination us ?
1317                  */
1318                 if (!dn_dev_islocal(in_dev, cb->dst))
1319                         goto e_inval;
1320
1321                 res.type = RTN_LOCAL;
1322         } else {
1323                 __le16 src_map = fld.saddr;
1324                 free_res = 1;
1325
1326                 out_dev = DN_FIB_RES_DEV(res);
1327                 if (out_dev == NULL) {
1328                         if (net_ratelimit())
1329                                 printk(KERN_CRIT "Bug in dn_route_input_slow() "
1330                                                  "No output device\n");
1331                         goto e_inval;
1332                 }
1333                 dev_hold(out_dev);
1334
1335                 if (res.r)
1336                         src_map = fld.saddr; /* no NAT support for now */
1337
1338                 gateway = DN_FIB_RES_GW(res);
1339                 if (res.type == RTN_NAT) {
1340                         fld.daddr = dn_fib_rules_map_destination(fld.daddr, &res);
1341                         dn_fib_res_put(&res);
1342                         free_res = 0;
1343                         if (dn_fib_lookup(&fld, &res))
1344                                 goto e_inval;
1345                         free_res = 1;
1346                         if (res.type != RTN_UNICAST)
1347                                 goto e_inval;
1348                         flags |= RTCF_DNAT;
1349                         gateway = fld.daddr;
1350                 }
1351                 fld.saddr = src_map;
1352         }
1353
1354         switch(res.type) {
1355         case RTN_UNICAST:
1356                 /*
1357                  * Forwarding check here, we only check for forwarding
1358                  * being turned off, if you want to only forward intra
1359                  * area, its up to you to set the routing tables up
1360                  * correctly.
1361                  */
1362                 if (dn_db->parms.forwarding == 0)
1363                         goto e_inval;
1364
1365                 if (res.fi->fib_nhs > 1 && fld.flowidn_oif == 0)
1366                         dn_fib_select_multipath(&fld, &res);
1367
1368                 /*
1369                  * Check for out_dev == in_dev. We use the RTCF_DOREDIRECT
1370                  * flag as a hint to set the intra-ethernet bit when
1371                  * forwarding. If we've got NAT in operation, we don't do
1372                  * this optimisation.
1373                  */
1374                 if (out_dev == in_dev && !(flags & RTCF_NAT))
1375                         flags |= RTCF_DOREDIRECT;
1376
1377                 local_src = DN_FIB_RES_PREFSRC(res);
1378
1379         case RTN_BLACKHOLE:
1380         case RTN_UNREACHABLE:
1381                 break;
1382         case RTN_LOCAL:
1383                 flags |= RTCF_LOCAL;
1384                 fld.saddr = cb->dst;
1385                 fld.daddr = cb->src;
1386
1387                 /* Routing tables gave us a gateway */
1388                 if (gateway)
1389                         goto make_route;
1390
1391                 /* Packet was intra-ethernet, so we know its on-link */
1392                 if (cb->rt_flags & DN_RT_F_IE) {
1393                         gateway = cb->src;
1394                         flags |= RTCF_DIRECTSRC;
1395                         goto make_route;
1396                 }
1397
1398                 /* Use the default router if there is one */
1399                 neigh = neigh_clone(dn_db->router);
1400                 if (neigh) {
1401                         gateway = ((struct dn_neigh *)neigh)->addr;
1402                         goto make_route;
1403                 }
1404
1405                 /* Close eyes and pray */
1406                 gateway = cb->src;
1407                 flags |= RTCF_DIRECTSRC;
1408                 goto make_route;
1409         default:
1410                 goto e_inval;
1411         }
1412
1413 make_route:
1414         rt = dst_alloc(&dn_dst_ops, out_dev, 0, 0, DST_HOST);
1415         if (rt == NULL)
1416                 goto e_nobufs;
1417
1418         memset(&rt->fld, 0, sizeof(rt->fld));
1419         rt->rt_saddr      = fld.saddr;
1420         rt->rt_daddr      = fld.daddr;
1421         rt->rt_gateway    = fld.daddr;
1422         if (gateway)
1423                 rt->rt_gateway = gateway;
1424         rt->rt_local_src  = local_src ? local_src : rt->rt_saddr;
1425
1426         rt->rt_dst_map    = fld.daddr;
1427         rt->rt_src_map    = fld.saddr;
1428
1429         rt->fld.saddr        = cb->src;
1430         rt->fld.daddr        = cb->dst;
1431         rt->fld.flowidn_oif  = 0;
1432         rt->fld.flowidn_iif  = in_dev->ifindex;
1433         rt->fld.flowidn_mark = fld.flowidn_mark;
1434
1435         dst_set_neighbour(&rt->dst, neigh);
1436         rt->dst.lastuse = jiffies;
1437         rt->dst.output = dn_rt_bug;
1438         switch (res.type) {
1439         case RTN_UNICAST:
1440                 rt->dst.input = dn_forward;
1441                 break;
1442         case RTN_LOCAL:
1443                 rt->dst.output = dn_output;
1444                 rt->dst.input = dn_nsp_rx;
1445                 rt->dst.dev = in_dev;
1446                 flags |= RTCF_LOCAL;
1447                 break;
1448         default:
1449         case RTN_UNREACHABLE:
1450         case RTN_BLACKHOLE:
1451                 rt->dst.input = dst_discard;
1452         }
1453         rt->rt_flags = flags;
1454
1455         err = dn_rt_set_next_hop(rt, &res);
1456         if (err)
1457                 goto e_neighbour;
1458
1459         hash = dn_hash(rt->fld.saddr, rt->fld.daddr);
1460         dn_insert_route(rt, hash, &rt);
1461         skb_dst_set(skb, &rt->dst);
1462
1463 done:
1464         if (neigh)
1465                 neigh_release(neigh);
1466         if (free_res)
1467                 dn_fib_res_put(&res);
1468         dev_put(in_dev);
1469         if (out_dev)
1470                 dev_put(out_dev);
1471 out:
1472         return err;
1473
1474 e_inval:
1475         err = -EINVAL;
1476         goto done;
1477
1478 e_nobufs:
1479         err = -ENOBUFS;
1480         goto done;
1481
1482 e_neighbour:
1483         dst_free(&rt->dst);
1484         goto done;
1485 }
1486
1487 static int dn_route_input(struct sk_buff *skb)
1488 {
1489         struct dn_route *rt;
1490         struct dn_skb_cb *cb = DN_SKB_CB(skb);
1491         unsigned hash = dn_hash(cb->src, cb->dst);
1492
1493         if (skb_dst(skb))
1494                 return 0;
1495
1496         rcu_read_lock();
1497         for(rt = rcu_dereference(dn_rt_hash_table[hash].chain); rt != NULL;
1498             rt = rcu_dereference(rt->dst.dn_next)) {
1499                 if ((rt->fld.saddr == cb->src) &&
1500                     (rt->fld.daddr == cb->dst) &&
1501                     (rt->fld.flowidn_oif == 0) &&
1502                     (rt->fld.flowidn_mark == skb->mark) &&
1503                     (rt->fld.flowidn_iif == cb->iif)) {
1504                         dst_use(&rt->dst, jiffies);
1505                         rcu_read_unlock();
1506                         skb_dst_set(skb, (struct dst_entry *)rt);
1507                         return 0;
1508                 }
1509         }
1510         rcu_read_unlock();
1511
1512         return dn_route_input_slow(skb);
1513 }
1514
1515 static int dn_rt_fill_info(struct sk_buff *skb, u32 pid, u32 seq,
1516                            int event, int nowait, unsigned int flags)
1517 {
1518         struct dn_route *rt = (struct dn_route *)skb_dst(skb);
1519         struct rtmsg *r;
1520         struct nlmsghdr *nlh;
1521         unsigned char *b = skb_tail_pointer(skb);
1522         long expires;
1523
1524         nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*r), flags);
1525         r = NLMSG_DATA(nlh);
1526         r->rtm_family = AF_DECnet;
1527         r->rtm_dst_len = 16;
1528         r->rtm_src_len = 0;
1529         r->rtm_tos = 0;
1530         r->rtm_table = RT_TABLE_MAIN;
1531         RTA_PUT_U32(skb, RTA_TABLE, RT_TABLE_MAIN);
1532         r->rtm_type = rt->rt_type;
1533         r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
1534         r->rtm_scope = RT_SCOPE_UNIVERSE;
1535         r->rtm_protocol = RTPROT_UNSPEC;
1536         if (rt->rt_flags & RTCF_NOTIFY)
1537                 r->rtm_flags |= RTM_F_NOTIFY;
1538         RTA_PUT(skb, RTA_DST, 2, &rt->rt_daddr);
1539         if (rt->fld.saddr) {
1540                 r->rtm_src_len = 16;
1541                 RTA_PUT(skb, RTA_SRC, 2, &rt->fld.saddr);
1542         }
1543         if (rt->dst.dev)
1544                 RTA_PUT(skb, RTA_OIF, sizeof(int), &rt->dst.dev->ifindex);
1545         /*
1546          * Note to self - change this if input routes reverse direction when
1547          * they deal only with inputs and not with replies like they do
1548          * currently.
1549          */
1550         RTA_PUT(skb, RTA_PREFSRC, 2, &rt->rt_local_src);
1551         if (rt->rt_daddr != rt->rt_gateway)
1552                 RTA_PUT(skb, RTA_GATEWAY, 2, &rt->rt_gateway);
1553         if (rtnetlink_put_metrics(skb, dst_metrics_ptr(&rt->dst)) < 0)
1554                 goto rtattr_failure;
1555         expires = rt->dst.expires ? rt->dst.expires - jiffies : 0;
1556         if (rtnl_put_cacheinfo(skb, &rt->dst, 0, 0, 0, expires,
1557                                rt->dst.error) < 0)
1558                 goto rtattr_failure;
1559         if (dn_is_input_route(rt))
1560                 RTA_PUT(skb, RTA_IIF, sizeof(int), &rt->fld.flowidn_iif);
1561
1562         nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1563         return skb->len;
1564
1565 nlmsg_failure:
1566 rtattr_failure:
1567         nlmsg_trim(skb, b);
1568         return -1;
1569 }
1570
1571 /*
1572  * This is called by both endnodes and routers now.
1573  */
1574 static int dn_cache_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh, void *arg)
1575 {
1576         struct net *net = sock_net(in_skb->sk);
1577         struct rtattr **rta = arg;
1578         struct rtmsg *rtm = NLMSG_DATA(nlh);
1579         struct dn_route *rt = NULL;
1580         struct dn_skb_cb *cb;
1581         int err;
1582         struct sk_buff *skb;
1583         struct flowidn fld;
1584
1585         if (!net_eq(net, &init_net))
1586                 return -EINVAL;
1587
1588         memset(&fld, 0, sizeof(fld));
1589         fld.flowidn_proto = DNPROTO_NSP;
1590
1591         skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1592         if (skb == NULL)
1593                 return -ENOBUFS;
1594         skb_reset_mac_header(skb);
1595         cb = DN_SKB_CB(skb);
1596
1597         if (rta[RTA_SRC-1])
1598                 memcpy(&fld.saddr, RTA_DATA(rta[RTA_SRC-1]), 2);
1599         if (rta[RTA_DST-1])
1600                 memcpy(&fld.daddr, RTA_DATA(rta[RTA_DST-1]), 2);
1601         if (rta[RTA_IIF-1])
1602                 memcpy(&fld.flowidn_iif, RTA_DATA(rta[RTA_IIF-1]), sizeof(int));
1603
1604         if (fld.flowidn_iif) {
1605                 struct net_device *dev;
1606                 if ((dev = dev_get_by_index(&init_net, fld.flowidn_iif)) == NULL) {
1607                         kfree_skb(skb);
1608                         return -ENODEV;
1609                 }
1610                 if (!dev->dn_ptr) {
1611                         dev_put(dev);
1612                         kfree_skb(skb);
1613                         return -ENODEV;
1614                 }
1615                 skb->protocol = htons(ETH_P_DNA_RT);
1616                 skb->dev = dev;
1617                 cb->src = fld.saddr;
1618                 cb->dst = fld.daddr;
1619                 local_bh_disable();
1620                 err = dn_route_input(skb);
1621                 local_bh_enable();
1622                 memset(cb, 0, sizeof(struct dn_skb_cb));
1623                 rt = (struct dn_route *)skb_dst(skb);
1624                 if (!err && -rt->dst.error)
1625                         err = rt->dst.error;
1626         } else {
1627                 int oif = 0;
1628                 if (rta[RTA_OIF - 1])
1629                         memcpy(&oif, RTA_DATA(rta[RTA_OIF - 1]), sizeof(int));
1630                 fld.flowidn_oif = oif;
1631                 err = dn_route_output_key((struct dst_entry **)&rt, &fld, 0);
1632         }
1633
1634         if (skb->dev)
1635                 dev_put(skb->dev);
1636         skb->dev = NULL;
1637         if (err)
1638                 goto out_free;
1639         skb_dst_set(skb, &rt->dst);
1640         if (rtm->rtm_flags & RTM_F_NOTIFY)
1641                 rt->rt_flags |= RTCF_NOTIFY;
1642
1643         err = dn_rt_fill_info(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq, RTM_NEWROUTE, 0, 0);
1644
1645         if (err == 0)
1646                 goto out_free;
1647         if (err < 0) {
1648                 err = -EMSGSIZE;
1649                 goto out_free;
1650         }
1651
1652         return rtnl_unicast(skb, &init_net, NETLINK_CB(in_skb).pid);
1653
1654 out_free:
1655         kfree_skb(skb);
1656         return err;
1657 }
1658
1659 /*
1660  * For routers, this is called from dn_fib_dump, but for endnodes its
1661  * called directly from the rtnetlink dispatch table.
1662  */
1663 int dn_cache_dump(struct sk_buff *skb, struct netlink_callback *cb)
1664 {
1665         struct net *net = sock_net(skb->sk);
1666         struct dn_route *rt;
1667         int h, s_h;
1668         int idx, s_idx;
1669
1670         if (!net_eq(net, &init_net))
1671                 return 0;
1672
1673         if (NLMSG_PAYLOAD(cb->nlh, 0) < sizeof(struct rtmsg))
1674                 return -EINVAL;
1675         if (!(((struct rtmsg *)NLMSG_DATA(cb->nlh))->rtm_flags&RTM_F_CLONED))
1676                 return 0;
1677
1678         s_h = cb->args[0];
1679         s_idx = idx = cb->args[1];
1680         for(h = 0; h <= dn_rt_hash_mask; h++) {
1681                 if (h < s_h)
1682                         continue;
1683                 if (h > s_h)
1684                         s_idx = 0;
1685                 rcu_read_lock_bh();
1686                 for(rt = rcu_dereference_bh(dn_rt_hash_table[h].chain), idx = 0;
1687                         rt;
1688                         rt = rcu_dereference_bh(rt->dst.dn_next), idx++) {
1689                         if (idx < s_idx)
1690                                 continue;
1691                         skb_dst_set(skb, dst_clone(&rt->dst));
1692                         if (dn_rt_fill_info(skb, NETLINK_CB(cb->skb).pid,
1693                                         cb->nlh->nlmsg_seq, RTM_NEWROUTE,
1694                                         1, NLM_F_MULTI) <= 0) {
1695                                 skb_dst_drop(skb);
1696                                 rcu_read_unlock_bh();
1697                                 goto done;
1698                         }
1699                         skb_dst_drop(skb);
1700                 }
1701                 rcu_read_unlock_bh();
1702         }
1703
1704 done:
1705         cb->args[0] = h;
1706         cb->args[1] = idx;
1707         return skb->len;
1708 }
1709
1710 #ifdef CONFIG_PROC_FS
1711 struct dn_rt_cache_iter_state {
1712         int bucket;
1713 };
1714
1715 static struct dn_route *dn_rt_cache_get_first(struct seq_file *seq)
1716 {
1717         struct dn_route *rt = NULL;
1718         struct dn_rt_cache_iter_state *s = seq->private;
1719
1720         for(s->bucket = dn_rt_hash_mask; s->bucket >= 0; --s->bucket) {
1721                 rcu_read_lock_bh();
1722                 rt = rcu_dereference_bh(dn_rt_hash_table[s->bucket].chain);
1723                 if (rt)
1724                         break;
1725                 rcu_read_unlock_bh();
1726         }
1727         return rt;
1728 }
1729
1730 static struct dn_route *dn_rt_cache_get_next(struct seq_file *seq, struct dn_route *rt)
1731 {
1732         struct dn_rt_cache_iter_state *s = seq->private;
1733
1734         rt = rcu_dereference_bh(rt->dst.dn_next);
1735         while (!rt) {
1736                 rcu_read_unlock_bh();
1737                 if (--s->bucket < 0)
1738                         break;
1739                 rcu_read_lock_bh();
1740                 rt = rcu_dereference_bh(dn_rt_hash_table[s->bucket].chain);
1741         }
1742         return rt;
1743 }
1744
1745 static void *dn_rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
1746 {
1747         struct dn_route *rt = dn_rt_cache_get_first(seq);
1748
1749         if (rt) {
1750                 while(*pos && (rt = dn_rt_cache_get_next(seq, rt)))
1751                         --*pos;
1752         }
1753         return *pos ? NULL : rt;
1754 }
1755
1756 static void *dn_rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1757 {
1758         struct dn_route *rt = dn_rt_cache_get_next(seq, v);
1759         ++*pos;
1760         return rt;
1761 }
1762
1763 static void dn_rt_cache_seq_stop(struct seq_file *seq, void *v)
1764 {
1765         if (v)
1766                 rcu_read_unlock_bh();
1767 }
1768
1769 static int dn_rt_cache_seq_show(struct seq_file *seq, void *v)
1770 {
1771         struct dn_route *rt = v;
1772         char buf1[DN_ASCBUF_LEN], buf2[DN_ASCBUF_LEN];
1773
1774         seq_printf(seq, "%-8s %-7s %-7s %04d %04d %04d\n",
1775                         rt->dst.dev ? rt->dst.dev->name : "*",
1776                         dn_addr2asc(le16_to_cpu(rt->rt_daddr), buf1),
1777                         dn_addr2asc(le16_to_cpu(rt->rt_saddr), buf2),
1778                         atomic_read(&rt->dst.__refcnt),
1779                         rt->dst.__use,
1780                         (int) dst_metric(&rt->dst, RTAX_RTT));
1781         return 0;
1782 }
1783
1784 static const struct seq_operations dn_rt_cache_seq_ops = {
1785         .start  = dn_rt_cache_seq_start,
1786         .next   = dn_rt_cache_seq_next,
1787         .stop   = dn_rt_cache_seq_stop,
1788         .show   = dn_rt_cache_seq_show,
1789 };
1790
1791 static int dn_rt_cache_seq_open(struct inode *inode, struct file *file)
1792 {
1793         return seq_open_private(file, &dn_rt_cache_seq_ops,
1794                         sizeof(struct dn_rt_cache_iter_state));
1795 }
1796
1797 static const struct file_operations dn_rt_cache_seq_fops = {
1798         .owner   = THIS_MODULE,
1799         .open    = dn_rt_cache_seq_open,
1800         .read    = seq_read,
1801         .llseek  = seq_lseek,
1802         .release = seq_release_private,
1803 };
1804
1805 #endif /* CONFIG_PROC_FS */
1806
1807 void __init dn_route_init(void)
1808 {
1809         int i, goal, order;
1810
1811         dn_dst_ops.kmem_cachep =
1812                 kmem_cache_create("dn_dst_cache", sizeof(struct dn_route), 0,
1813                                   SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1814         dst_entries_init(&dn_dst_ops);
1815         setup_timer(&dn_route_timer, dn_dst_check_expire, 0);
1816         dn_route_timer.expires = jiffies + decnet_dst_gc_interval * HZ;
1817         add_timer(&dn_route_timer);
1818
1819         goal = totalram_pages >> (26 - PAGE_SHIFT);
1820
1821         for(order = 0; (1UL << order) < goal; order++)
1822                 /* NOTHING */;
1823
1824         /*
1825          * Only want 1024 entries max, since the table is very, very unlikely
1826          * to be larger than that.
1827          */
1828         while(order && ((((1UL << order) * PAGE_SIZE) /
1829                                 sizeof(struct dn_rt_hash_bucket)) >= 2048))
1830                 order--;
1831
1832         do {
1833                 dn_rt_hash_mask = (1UL << order) * PAGE_SIZE /
1834                         sizeof(struct dn_rt_hash_bucket);
1835                 while(dn_rt_hash_mask & (dn_rt_hash_mask - 1))
1836                         dn_rt_hash_mask--;
1837                 dn_rt_hash_table = (struct dn_rt_hash_bucket *)
1838                         __get_free_pages(GFP_ATOMIC, order);
1839         } while (dn_rt_hash_table == NULL && --order > 0);
1840
1841         if (!dn_rt_hash_table)
1842                 panic("Failed to allocate DECnet route cache hash table\n");
1843
1844         printk(KERN_INFO
1845                 "DECnet: Routing cache hash table of %u buckets, %ldKbytes\n",
1846                 dn_rt_hash_mask,
1847                 (long)(dn_rt_hash_mask*sizeof(struct dn_rt_hash_bucket))/1024);
1848
1849         dn_rt_hash_mask--;
1850         for(i = 0; i <= dn_rt_hash_mask; i++) {
1851                 spin_lock_init(&dn_rt_hash_table[i].lock);
1852                 dn_rt_hash_table[i].chain = NULL;
1853         }
1854
1855         dn_dst_ops.gc_thresh = (dn_rt_hash_mask + 1);
1856
1857         proc_net_fops_create(&init_net, "decnet_cache", S_IRUGO, &dn_rt_cache_seq_fops);
1858
1859 #ifdef CONFIG_DECNET_ROUTER
1860         rtnl_register(PF_DECnet, RTM_GETROUTE, dn_cache_getroute,
1861                       dn_fib_dump, NULL);
1862 #else
1863         rtnl_register(PF_DECnet, RTM_GETROUTE, dn_cache_getroute,
1864                       dn_cache_dump, NULL);
1865 #endif
1866 }
1867
1868 void __exit dn_route_cleanup(void)
1869 {
1870         del_timer(&dn_route_timer);
1871         dn_run_flush(0);
1872
1873         proc_net_remove(&init_net, "decnet_cache");
1874         dst_entries_destroy(&dn_dst_ops);
1875 }
1876