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 <linux/export.h>
81#include <asm/errno.h>
82#include <net/net_namespace.h>
83#include <net/netlink.h>
84#include <net/neighbour.h>
85#include <net/dst.h>
86#include <net/flow.h>
87#include <net/fib_rules.h>
88#include <net/dn.h>
89#include <net/dn_dev.h>
90#include <net/dn_nsp.h>
91#include <net/dn_route.h>
92#include <net/dn_neigh.h>
93#include <net/dn_fib.h>
94
95struct dn_rt_hash_bucket
96{
97	struct dn_route __rcu *chain;
98	spinlock_t lock;
99};
100
101extern struct neigh_table dn_neigh_table;
102
103
104static unsigned char dn_hiord_addr[6] = {0xAA,0x00,0x04,0x00,0x00,0x00};
105
106static const int dn_rt_min_delay = 2 * HZ;
107static const int dn_rt_max_delay = 10 * HZ;
108static const int dn_rt_mtu_expires = 10 * 60 * HZ;
109
110static unsigned long dn_rt_deadline;
111
112static int dn_dst_gc(struct dst_ops *ops);
113static struct dst_entry *dn_dst_check(struct dst_entry *, __u32);
114static unsigned int dn_dst_default_advmss(const struct dst_entry *dst);
115static unsigned int dn_dst_mtu(const struct dst_entry *dst);
116static void dn_dst_destroy(struct dst_entry *);
117static void dn_dst_ifdown(struct dst_entry *, struct net_device *dev, int how);
118static struct dst_entry *dn_dst_negative_advice(struct dst_entry *);
119static void dn_dst_link_failure(struct sk_buff *);
120static void dn_dst_update_pmtu(struct dst_entry *dst, struct sock *sk,
121			       struct sk_buff *skb , u32 mtu);
122static void dn_dst_redirect(struct dst_entry *dst, struct sock *sk,
123			    struct sk_buff *skb);
124static struct neighbour *dn_dst_neigh_lookup(const struct dst_entry *dst,
125					     struct sk_buff *skb,
126					     const void *daddr);
127static int dn_route_input(struct sk_buff *);
128static void dn_run_flush(unsigned long dummy);
129
130static struct dn_rt_hash_bucket *dn_rt_hash_table;
131static unsigned int dn_rt_hash_mask;
132
133static struct timer_list dn_route_timer;
134static DEFINE_TIMER(dn_rt_flush_timer, dn_run_flush, 0, 0);
135int decnet_dst_gc_interval = 2;
136
137static struct dst_ops dn_dst_ops = {
138	.family =		PF_DECnet,
139	.gc_thresh =		128,
140	.gc =			dn_dst_gc,
141	.check =		dn_dst_check,
142	.default_advmss =	dn_dst_default_advmss,
143	.mtu =			dn_dst_mtu,
144	.cow_metrics =		dst_cow_metrics_generic,
145	.destroy =		dn_dst_destroy,
146	.ifdown =		dn_dst_ifdown,
147	.negative_advice =	dn_dst_negative_advice,
148	.link_failure =		dn_dst_link_failure,
149	.update_pmtu =		dn_dst_update_pmtu,
150	.redirect =		dn_dst_redirect,
151	.neigh_lookup =		dn_dst_neigh_lookup,
152};
153
154static void dn_dst_destroy(struct dst_entry *dst)
155{
156	struct dn_route *rt = (struct dn_route *) dst;
157
158	if (rt->n)
159		neigh_release(rt->n);
160	dst_destroy_metrics_generic(dst);
161}
162
163static void dn_dst_ifdown(struct dst_entry *dst, struct net_device *dev, int how)
164{
165	if (how) {
166		struct dn_route *rt = (struct dn_route *) dst;
167		struct neighbour *n = rt->n;
168
169		if (n && n->dev == dev) {
170			n->dev = dev_net(dev)->loopback_dev;
171			dev_hold(n->dev);
172			dev_put(dev);
173		}
174	}
175}
176
177static __inline__ unsigned int dn_hash(__le16 src, __le16 dst)
178{
179	__u16 tmp = (__u16 __force)(src ^ dst);
180	tmp ^= (tmp >> 3);
181	tmp ^= (tmp >> 5);
182	tmp ^= (tmp >> 10);
183	return dn_rt_hash_mask & (unsigned int)tmp;
184}
185
186static inline void dnrt_free(struct dn_route *rt)
187{
188	call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
189}
190
191static inline void dnrt_drop(struct dn_route *rt)
192{
193	dst_release(&rt->dst);
194	call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
195}
196
197static void dn_dst_check_expire(unsigned long dummy)
198{
199	int i;
200	struct dn_route *rt;
201	struct dn_route __rcu **rtp;
202	unsigned long now = jiffies;
203	unsigned long expire = 120 * HZ;
204
205	for (i = 0; i <= dn_rt_hash_mask; i++) {
206		rtp = &dn_rt_hash_table[i].chain;
207
208		spin_lock(&dn_rt_hash_table[i].lock);
209		while ((rt = rcu_dereference_protected(*rtp,
210						lockdep_is_held(&dn_rt_hash_table[i].lock))) != NULL) {
211			if (atomic_read(&rt->dst.__refcnt) ||
212					(now - rt->dst.lastuse) < expire) {
213				rtp = &rt->dst.dn_next;
214				continue;
215			}
216			*rtp = rt->dst.dn_next;
217			rt->dst.dn_next = NULL;
218			dnrt_free(rt);
219		}
220		spin_unlock(&dn_rt_hash_table[i].lock);
221
222		if ((jiffies - now) > 0)
223			break;
224	}
225
226	mod_timer(&dn_route_timer, now + decnet_dst_gc_interval * HZ);
227}
228
229static int dn_dst_gc(struct dst_ops *ops)
230{
231	struct dn_route *rt;
232	struct dn_route __rcu **rtp;
233	int i;
234	unsigned long now = jiffies;
235	unsigned long expire = 10 * HZ;
236
237	for (i = 0; i <= dn_rt_hash_mask; i++) {
238
239		spin_lock_bh(&dn_rt_hash_table[i].lock);
240		rtp = &dn_rt_hash_table[i].chain;
241
242		while ((rt = rcu_dereference_protected(*rtp,
243						lockdep_is_held(&dn_rt_hash_table[i].lock))) != NULL) {
244			if (atomic_read(&rt->dst.__refcnt) ||
245					(now - rt->dst.lastuse) < expire) {
246				rtp = &rt->dst.dn_next;
247				continue;
248			}
249			*rtp = rt->dst.dn_next;
250			rt->dst.dn_next = NULL;
251			dnrt_drop(rt);
252			break;
253		}
254		spin_unlock_bh(&dn_rt_hash_table[i].lock);
255	}
256
257	return 0;
258}
259
260/*
261 * The decnet standards don't impose a particular minimum mtu, what they
262 * do insist on is that the routing layer accepts a datagram of at least
263 * 230 bytes long. Here we have to subtract the routing header length from
264 * 230 to get the minimum acceptable mtu. If there is no neighbour, then we
265 * assume the worst and use a long header size.
266 *
267 * We update both the mtu and the advertised mss (i.e. the segment size we
268 * advertise to the other end).
269 */
270static void dn_dst_update_pmtu(struct dst_entry *dst, struct sock *sk,
271			       struct sk_buff *skb, u32 mtu)
272{
273	struct dn_route *rt = (struct dn_route *) dst;
274	struct neighbour *n = rt->n;
275	u32 min_mtu = 230;
276	struct dn_dev *dn;
277
278	dn = n ? rcu_dereference_raw(n->dev->dn_ptr) : NULL;
279
280	if (dn && dn->use_long == 0)
281		min_mtu -= 6;
282	else
283		min_mtu -= 21;
284
285	if (dst_metric(dst, RTAX_MTU) > mtu && mtu >= min_mtu) {
286		if (!(dst_metric_locked(dst, RTAX_MTU))) {
287			dst_metric_set(dst, RTAX_MTU, mtu);
288			dst_set_expires(dst, dn_rt_mtu_expires);
289		}
290		if (!(dst_metric_locked(dst, RTAX_ADVMSS))) {
291			u32 mss = mtu - DN_MAX_NSP_DATA_HEADER;
292			u32 existing_mss = dst_metric_raw(dst, RTAX_ADVMSS);
293			if (!existing_mss || existing_mss > mss)
294				dst_metric_set(dst, RTAX_ADVMSS, mss);
295		}
296	}
297}
298
299static void dn_dst_redirect(struct dst_entry *dst, struct sock *sk,
300			    struct sk_buff *skb)
301{
302}
303
304/*
305 * When a route has been marked obsolete. (e.g. routing cache flush)
306 */
307static struct dst_entry *dn_dst_check(struct dst_entry *dst, __u32 cookie)
308{
309	return NULL;
310}
311
312static struct dst_entry *dn_dst_negative_advice(struct dst_entry *dst)
313{
314	dst_release(dst);
315	return NULL;
316}
317
318static void dn_dst_link_failure(struct sk_buff *skb)
319{
320}
321
322static inline int compare_keys(struct flowidn *fl1, struct flowidn *fl2)
323{
324	return ((fl1->daddr ^ fl2->daddr) |
325		(fl1->saddr ^ fl2->saddr) |
326		(fl1->flowidn_mark ^ fl2->flowidn_mark) |
327		(fl1->flowidn_scope ^ fl2->flowidn_scope) |
328		(fl1->flowidn_oif ^ fl2->flowidn_oif) |
329		(fl1->flowidn_iif ^ fl2->flowidn_iif)) == 0;
330}
331
332static int dn_insert_route(struct dn_route *rt, unsigned int hash, struct dn_route **rp)
333{
334	struct dn_route *rth;
335	struct dn_route __rcu **rthp;
336	unsigned long now = jiffies;
337
338	rthp = &dn_rt_hash_table[hash].chain;
339
340	spin_lock_bh(&dn_rt_hash_table[hash].lock);
341	while ((rth = rcu_dereference_protected(*rthp,
342						lockdep_is_held(&dn_rt_hash_table[hash].lock))) != NULL) {
343		if (compare_keys(&rth->fld, &rt->fld)) {
344			/* Put it first */
345			*rthp = rth->dst.dn_next;
346			rcu_assign_pointer(rth->dst.dn_next,
347					   dn_rt_hash_table[hash].chain);
348			rcu_assign_pointer(dn_rt_hash_table[hash].chain, rth);
349
350			dst_use(&rth->dst, now);
351			spin_unlock_bh(&dn_rt_hash_table[hash].lock);
352
353			dnrt_drop(rt);
354			*rp = rth;
355			return 0;
356		}
357		rthp = &rth->dst.dn_next;
358	}
359
360	rcu_assign_pointer(rt->dst.dn_next, dn_rt_hash_table[hash].chain);
361	rcu_assign_pointer(dn_rt_hash_table[hash].chain, rt);
362
363	dst_use(&rt->dst, now);
364	spin_unlock_bh(&dn_rt_hash_table[hash].lock);
365	*rp = rt;
366	return 0;
367}
368
369static void dn_run_flush(unsigned long dummy)
370{
371	int i;
372	struct dn_route *rt, *next;
373
374	for (i = 0; i < dn_rt_hash_mask; i++) {
375		spin_lock_bh(&dn_rt_hash_table[i].lock);
376
377		if ((rt = xchg((struct dn_route **)&dn_rt_hash_table[i].chain, NULL)) == NULL)
378			goto nothing_to_declare;
379
380		for(; rt; rt = next) {
381			next = rcu_dereference_raw(rt->dst.dn_next);
382			RCU_INIT_POINTER(rt->dst.dn_next, NULL);
383			dst_free((struct dst_entry *)rt);
384		}
385
386nothing_to_declare:
387		spin_unlock_bh(&dn_rt_hash_table[i].lock);
388	}
389}
390
391static DEFINE_SPINLOCK(dn_rt_flush_lock);
392
393void dn_rt_cache_flush(int delay)
394{
395	unsigned long now = jiffies;
396	int user_mode = !in_interrupt();
397
398	if (delay < 0)
399		delay = dn_rt_min_delay;
400
401	spin_lock_bh(&dn_rt_flush_lock);
402
403	if (del_timer(&dn_rt_flush_timer) && delay > 0 && dn_rt_deadline) {
404		long tmo = (long)(dn_rt_deadline - now);
405
406		if (user_mode && tmo < dn_rt_max_delay - dn_rt_min_delay)
407			tmo = 0;
408
409		if (delay > tmo)
410			delay = tmo;
411	}
412
413	if (delay <= 0) {
414		spin_unlock_bh(&dn_rt_flush_lock);
415		dn_run_flush(0);
416		return;
417	}
418
419	if (dn_rt_deadline == 0)
420		dn_rt_deadline = now + dn_rt_max_delay;
421
422	dn_rt_flush_timer.expires = now + delay;
423	add_timer(&dn_rt_flush_timer);
424	spin_unlock_bh(&dn_rt_flush_lock);
425}
426
427/**
428 * dn_return_short - Return a short packet to its sender
429 * @skb: The packet to return
430 *
431 */
432static int dn_return_short(struct sk_buff *skb)
433{
434	struct dn_skb_cb *cb;
435	unsigned char *ptr;
436	__le16 *src;
437	__le16 *dst;
438
439	/* Add back headers */
440	skb_push(skb, skb->data - skb_network_header(skb));
441
442	if ((skb = skb_unshare(skb, GFP_ATOMIC)) == NULL)
443		return NET_RX_DROP;
444
445	cb = DN_SKB_CB(skb);
446	/* Skip packet length and point to flags */
447	ptr = skb->data + 2;
448	*ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
449
450	dst = (__le16 *)ptr;
451	ptr += 2;
452	src = (__le16 *)ptr;
453	ptr += 2;
454	*ptr = 0; /* Zero hop count */
455
456	swap(*src, *dst);
457
458	skb->pkt_type = PACKET_OUTGOING;
459	dn_rt_finish_output(skb, NULL, NULL);
460	return NET_RX_SUCCESS;
461}
462
463/**
464 * dn_return_long - Return a long packet to its sender
465 * @skb: The long format packet to return
466 *
467 */
468static int dn_return_long(struct sk_buff *skb)
469{
470	struct dn_skb_cb *cb;
471	unsigned char *ptr;
472	unsigned char *src_addr, *dst_addr;
473	unsigned char tmp[ETH_ALEN];
474
475	/* Add back all headers */
476	skb_push(skb, skb->data - skb_network_header(skb));
477
478	if ((skb = skb_unshare(skb, GFP_ATOMIC)) == NULL)
479		return NET_RX_DROP;
480
481	cb = DN_SKB_CB(skb);
482	/* Ignore packet length and point to flags */
483	ptr = skb->data + 2;
484
485	/* Skip padding */
486	if (*ptr & DN_RT_F_PF) {
487		char padlen = (*ptr & ~DN_RT_F_PF);
488		ptr += padlen;
489	}
490
491	*ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
492	ptr += 2;
493	dst_addr = ptr;
494	ptr += 8;
495	src_addr = ptr;
496	ptr += 6;
497	*ptr = 0; /* Zero hop count */
498
499	/* Swap source and destination */
500	memcpy(tmp, src_addr, ETH_ALEN);
501	memcpy(src_addr, dst_addr, ETH_ALEN);
502	memcpy(dst_addr, tmp, ETH_ALEN);
503
504	skb->pkt_type = PACKET_OUTGOING;
505	dn_rt_finish_output(skb, dst_addr, src_addr);
506	return NET_RX_SUCCESS;
507}
508
509/**
510 * dn_route_rx_packet - Try and find a route for an incoming packet
511 * @skb: The packet to find a route for
512 *
513 * Returns: result of input function if route is found, error code otherwise
514 */
515static int dn_route_rx_packet(struct sock *sk, struct sk_buff *skb)
516{
517	struct dn_skb_cb *cb;
518	int err;
519
520	if ((err = dn_route_input(skb)) == 0)
521		return dst_input(skb);
522
523	cb = DN_SKB_CB(skb);
524	if (decnet_debug_level & 4) {
525		char *devname = skb->dev ? skb->dev->name : "???";
526
527		printk(KERN_DEBUG
528			"DECnet: dn_route_rx_packet: rt_flags=0x%02x dev=%s len=%d src=0x%04hx dst=0x%04hx err=%d type=%d\n",
529			(int)cb->rt_flags, devname, skb->len,
530			le16_to_cpu(cb->src), le16_to_cpu(cb->dst),
531			err, skb->pkt_type);
532	}
533
534	if ((skb->pkt_type == PACKET_HOST) && (cb->rt_flags & DN_RT_F_RQR)) {
535		switch (cb->rt_flags & DN_RT_PKT_MSK) {
536		case DN_RT_PKT_SHORT:
537			return dn_return_short(skb);
538		case DN_RT_PKT_LONG:
539			return dn_return_long(skb);
540		}
541	}
542
543	kfree_skb(skb);
544	return NET_RX_DROP;
545}
546
547static int dn_route_rx_long(struct sk_buff *skb)
548{
549	struct dn_skb_cb *cb = DN_SKB_CB(skb);
550	unsigned char *ptr = skb->data;
551
552	if (!pskb_may_pull(skb, 21)) /* 20 for long header, 1 for shortest nsp */
553		goto drop_it;
554
555	skb_pull(skb, 20);
556	skb_reset_transport_header(skb);
557
558	/* Destination info */
559	ptr += 2;
560	cb->dst = dn_eth2dn(ptr);
561	if (memcmp(ptr, dn_hiord_addr, 4) != 0)
562		goto drop_it;
563	ptr += 6;
564
565
566	/* Source info */
567	ptr += 2;
568	cb->src = dn_eth2dn(ptr);
569	if (memcmp(ptr, dn_hiord_addr, 4) != 0)
570		goto drop_it;
571	ptr += 6;
572	/* Other junk */
573	ptr++;
574	cb->hops = *ptr++; /* Visit Count */
575
576	return NF_HOOK(NFPROTO_DECNET, NF_DN_PRE_ROUTING, NULL, skb,
577		       skb->dev, NULL,
578		       dn_route_rx_packet);
579
580drop_it:
581	kfree_skb(skb);
582	return NET_RX_DROP;
583}
584
585
586
587static int dn_route_rx_short(struct sk_buff *skb)
588{
589	struct dn_skb_cb *cb = DN_SKB_CB(skb);
590	unsigned char *ptr = skb->data;
591
592	if (!pskb_may_pull(skb, 6)) /* 5 for short header + 1 for shortest nsp */
593		goto drop_it;
594
595	skb_pull(skb, 5);
596	skb_reset_transport_header(skb);
597
598	cb->dst = *(__le16 *)ptr;
599	ptr += 2;
600	cb->src = *(__le16 *)ptr;
601	ptr += 2;
602	cb->hops = *ptr & 0x3f;
603
604	return NF_HOOK(NFPROTO_DECNET, NF_DN_PRE_ROUTING, NULL, skb,
605		       skb->dev, NULL,
606		       dn_route_rx_packet);
607
608drop_it:
609	kfree_skb(skb);
610	return NET_RX_DROP;
611}
612
613static int dn_route_discard(struct sock *sk, struct sk_buff *skb)
614{
615	/*
616	 * I know we drop the packet here, but thats considered success in
617	 * this case
618	 */
619	kfree_skb(skb);
620	return NET_RX_SUCCESS;
621}
622
623static int dn_route_ptp_hello(struct sock *sk, struct sk_buff *skb)
624{
625	dn_dev_hello(skb);
626	dn_neigh_pointopoint_hello(skb);
627	return NET_RX_SUCCESS;
628}
629
630int dn_route_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
631{
632	struct dn_skb_cb *cb;
633	unsigned char flags = 0;
634	__u16 len = le16_to_cpu(*(__le16 *)skb->data);
635	struct dn_dev *dn = rcu_dereference(dev->dn_ptr);
636	unsigned char padlen = 0;
637
638	if (!net_eq(dev_net(dev), &init_net))
639		goto dump_it;
640
641	if (dn == NULL)
642		goto dump_it;
643
644	if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
645		goto out;
646
647	if (!pskb_may_pull(skb, 3))
648		goto dump_it;
649
650	skb_pull(skb, 2);
651
652	if (len > skb->len)
653		goto dump_it;
654
655	skb_trim(skb, len);
656
657	flags = *skb->data;
658
659	cb = DN_SKB_CB(skb);
660	cb->stamp = jiffies;
661	cb->iif = dev->ifindex;
662
663	/*
664	 * If we have padding, remove it.
665	 */
666	if (flags & DN_RT_F_PF) {
667		padlen = flags & ~DN_RT_F_PF;
668		if (!pskb_may_pull(skb, padlen + 1))
669			goto dump_it;
670		skb_pull(skb, padlen);
671		flags = *skb->data;
672	}
673
674	skb_reset_network_header(skb);
675
676	/*
677	 * Weed out future version DECnet
678	 */
679	if (flags & DN_RT_F_VER)
680		goto dump_it;
681
682	cb->rt_flags = flags;
683
684	if (decnet_debug_level & 1)
685		printk(KERN_DEBUG
686			"dn_route_rcv: got 0x%02x from %s [%d %d %d]\n",
687			(int)flags, (dev) ? dev->name : "???", len, skb->len,
688			padlen);
689
690	if (flags & DN_RT_PKT_CNTL) {
691		if (unlikely(skb_linearize(skb)))
692			goto dump_it;
693
694		switch (flags & DN_RT_CNTL_MSK) {
695		case DN_RT_PKT_INIT:
696			dn_dev_init_pkt(skb);
697			break;
698		case DN_RT_PKT_VERI:
699			dn_dev_veri_pkt(skb);
700			break;
701		}
702
703		if (dn->parms.state != DN_DEV_S_RU)
704			goto dump_it;
705
706		switch (flags & DN_RT_CNTL_MSK) {
707		case DN_RT_PKT_HELO:
708			return NF_HOOK(NFPROTO_DECNET, NF_DN_HELLO,
709				       NULL, skb, skb->dev, NULL,
710				       dn_route_ptp_hello);
711
712		case DN_RT_PKT_L1RT:
713		case DN_RT_PKT_L2RT:
714			return NF_HOOK(NFPROTO_DECNET, NF_DN_ROUTE,
715				       NULL, skb, skb->dev, NULL,
716				       dn_route_discard);
717		case DN_RT_PKT_ERTH:
718			return NF_HOOK(NFPROTO_DECNET, NF_DN_HELLO,
719				       NULL, skb, skb->dev, NULL,
720				       dn_neigh_router_hello);
721
722		case DN_RT_PKT_EEDH:
723			return NF_HOOK(NFPROTO_DECNET, NF_DN_HELLO,
724				       NULL, skb, skb->dev, NULL,
725				       dn_neigh_endnode_hello);
726		}
727	} else {
728		if (dn->parms.state != DN_DEV_S_RU)
729			goto dump_it;
730
731		skb_pull(skb, 1); /* Pull flags */
732
733		switch (flags & DN_RT_PKT_MSK) {
734		case DN_RT_PKT_LONG:
735			return dn_route_rx_long(skb);
736		case DN_RT_PKT_SHORT:
737			return dn_route_rx_short(skb);
738		}
739	}
740
741dump_it:
742	kfree_skb(skb);
743out:
744	return NET_RX_DROP;
745}
746
747static int dn_output(struct sock *sk, struct sk_buff *skb)
748{
749	struct dst_entry *dst = skb_dst(skb);
750	struct dn_route *rt = (struct dn_route *)dst;
751	struct net_device *dev = dst->dev;
752	struct dn_skb_cb *cb = DN_SKB_CB(skb);
753
754	int err = -EINVAL;
755
756	if (rt->n == NULL)
757		goto error;
758
759	skb->dev = dev;
760
761	cb->src = rt->rt_saddr;
762	cb->dst = rt->rt_daddr;
763
764	/*
765	 * Always set the Intra-Ethernet bit on all outgoing packets
766	 * originated on this node. Only valid flag from upper layers
767	 * is return-to-sender-requested. Set hop count to 0 too.
768	 */
769	cb->rt_flags &= ~DN_RT_F_RQR;
770	cb->rt_flags |= DN_RT_F_IE;
771	cb->hops = 0;
772
773	return NF_HOOK(NFPROTO_DECNET, NF_DN_LOCAL_OUT, sk, skb,
774		       NULL, dev,
775		       dn_to_neigh_output);
776
777error:
778	net_dbg_ratelimited("dn_output: This should not happen\n");
779
780	kfree_skb(skb);
781
782	return err;
783}
784
785static int dn_forward(struct sk_buff *skb)
786{
787	struct dn_skb_cb *cb = DN_SKB_CB(skb);
788	struct dst_entry *dst = skb_dst(skb);
789	struct dn_dev *dn_db = rcu_dereference(dst->dev->dn_ptr);
790	struct dn_route *rt;
791	int header_len;
792#ifdef CONFIG_NETFILTER
793	struct net_device *dev = skb->dev;
794#endif
795
796	if (skb->pkt_type != PACKET_HOST)
797		goto drop;
798
799	/* Ensure that we have enough space for headers */
800	rt = (struct dn_route *)skb_dst(skb);
801	header_len = dn_db->use_long ? 21 : 6;
802	if (skb_cow(skb, LL_RESERVED_SPACE(rt->dst.dev)+header_len))
803		goto drop;
804
805	/*
806	 * Hop count exceeded.
807	 */
808	if (++cb->hops > 30)
809		goto drop;
810
811	skb->dev = rt->dst.dev;
812
813	/*
814	 * If packet goes out same interface it came in on, then set
815	 * the Intra-Ethernet bit. This has no effect for short
816	 * packets, so we don't need to test for them here.
817	 */
818	cb->rt_flags &= ~DN_RT_F_IE;
819	if (rt->rt_flags & RTCF_DOREDIRECT)
820		cb->rt_flags |= DN_RT_F_IE;
821
822	return NF_HOOK(NFPROTO_DECNET, NF_DN_FORWARD, NULL, skb,
823		       dev, skb->dev,
824		       dn_to_neigh_output);
825
826drop:
827	kfree_skb(skb);
828	return NET_RX_DROP;
829}
830
831/*
832 * Used to catch bugs. This should never normally get
833 * called.
834 */
835static int dn_rt_bug_sk(struct sock *sk, struct sk_buff *skb)
836{
837	struct dn_skb_cb *cb = DN_SKB_CB(skb);
838
839	net_dbg_ratelimited("dn_rt_bug: skb from:%04x to:%04x\n",
840			    le16_to_cpu(cb->src), le16_to_cpu(cb->dst));
841
842	kfree_skb(skb);
843
844	return NET_RX_DROP;
845}
846
847static int dn_rt_bug(struct sk_buff *skb)
848{
849	struct dn_skb_cb *cb = DN_SKB_CB(skb);
850
851	net_dbg_ratelimited("dn_rt_bug: skb from:%04x to:%04x\n",
852			    le16_to_cpu(cb->src), le16_to_cpu(cb->dst));
853
854	kfree_skb(skb);
855
856	return NET_RX_DROP;
857}
858
859static unsigned int dn_dst_default_advmss(const struct dst_entry *dst)
860{
861	return dn_mss_from_pmtu(dst->dev, dst_mtu(dst));
862}
863
864static unsigned int dn_dst_mtu(const struct dst_entry *dst)
865{
866	unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
867
868	return mtu ? : dst->dev->mtu;
869}
870
871static struct neighbour *dn_dst_neigh_lookup(const struct dst_entry *dst,
872					     struct sk_buff *skb,
873					     const void *daddr)
874{
875	return __neigh_lookup_errno(&dn_neigh_table, daddr, dst->dev);
876}
877
878static int dn_rt_set_next_hop(struct dn_route *rt, struct dn_fib_res *res)
879{
880	struct dn_fib_info *fi = res->fi;
881	struct net_device *dev = rt->dst.dev;
882	unsigned int mss_metric;
883	struct neighbour *n;
884
885	if (fi) {
886		if (DN_FIB_RES_GW(*res) &&
887		    DN_FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
888			rt->rt_gateway = DN_FIB_RES_GW(*res);
889		dst_init_metrics(&rt->dst, fi->fib_metrics, true);
890	}
891	rt->rt_type = res->type;
892
893	if (dev != NULL && rt->n == NULL) {
894		n = __neigh_lookup_errno(&dn_neigh_table, &rt->rt_gateway, dev);
895		if (IS_ERR(n))
896			return PTR_ERR(n);
897		rt->n = n;
898	}
899
900	if (dst_metric(&rt->dst, RTAX_MTU) > rt->dst.dev->mtu)
901		dst_metric_set(&rt->dst, RTAX_MTU, rt->dst.dev->mtu);
902	mss_metric = dst_metric_raw(&rt->dst, RTAX_ADVMSS);
903	if (mss_metric) {
904		unsigned int mss = dn_mss_from_pmtu(dev, dst_mtu(&rt->dst));
905		if (mss_metric > mss)
906			dst_metric_set(&rt->dst, RTAX_ADVMSS, mss);
907	}
908	return 0;
909}
910
911static inline int dn_match_addr(__le16 addr1, __le16 addr2)
912{
913	__u16 tmp = le16_to_cpu(addr1) ^ le16_to_cpu(addr2);
914	int match = 16;
915	while(tmp) {
916		tmp >>= 1;
917		match--;
918	}
919	return match;
920}
921
922static __le16 dnet_select_source(const struct net_device *dev, __le16 daddr, int scope)
923{
924	__le16 saddr = 0;
925	struct dn_dev *dn_db;
926	struct dn_ifaddr *ifa;
927	int best_match = 0;
928	int ret;
929
930	rcu_read_lock();
931	dn_db = rcu_dereference(dev->dn_ptr);
932	for (ifa = rcu_dereference(dn_db->ifa_list);
933	     ifa != NULL;
934	     ifa = rcu_dereference(ifa->ifa_next)) {
935		if (ifa->ifa_scope > scope)
936			continue;
937		if (!daddr) {
938			saddr = ifa->ifa_local;
939			break;
940		}
941		ret = dn_match_addr(daddr, ifa->ifa_local);
942		if (ret > best_match)
943			saddr = ifa->ifa_local;
944		if (best_match == 0)
945			saddr = ifa->ifa_local;
946	}
947	rcu_read_unlock();
948
949	return saddr;
950}
951
952static inline __le16 __dn_fib_res_prefsrc(struct dn_fib_res *res)
953{
954	return dnet_select_source(DN_FIB_RES_DEV(*res), DN_FIB_RES_GW(*res), res->scope);
955}
956
957static inline __le16 dn_fib_rules_map_destination(__le16 daddr, struct dn_fib_res *res)
958{
959	__le16 mask = dnet_make_mask(res->prefixlen);
960	return (daddr&~mask)|res->fi->fib_nh->nh_gw;
961}
962
963static int dn_route_output_slow(struct dst_entry **pprt, const struct flowidn *oldflp, int try_hard)
964{
965	struct flowidn fld = {
966		.daddr = oldflp->daddr,
967		.saddr = oldflp->saddr,
968		.flowidn_scope = RT_SCOPE_UNIVERSE,
969		.flowidn_mark = oldflp->flowidn_mark,
970		.flowidn_iif = LOOPBACK_IFINDEX,
971		.flowidn_oif = oldflp->flowidn_oif,
972	};
973	struct dn_route *rt = NULL;
974	struct net_device *dev_out = NULL, *dev;
975	struct neighbour *neigh = NULL;
976	unsigned int hash;
977	unsigned int flags = 0;
978	struct dn_fib_res res = { .fi = NULL, .type = RTN_UNICAST };
979	int err;
980	int free_res = 0;
981	__le16 gateway = 0;
982
983	if (decnet_debug_level & 16)
984		printk(KERN_DEBUG
985		       "dn_route_output_slow: dst=%04x src=%04x mark=%d"
986		       " iif=%d oif=%d\n", le16_to_cpu(oldflp->daddr),
987		       le16_to_cpu(oldflp->saddr),
988		       oldflp->flowidn_mark, LOOPBACK_IFINDEX,
989		       oldflp->flowidn_oif);
990
991	/* If we have an output interface, verify its a DECnet device */
992	if (oldflp->flowidn_oif) {
993		dev_out = dev_get_by_index(&init_net, oldflp->flowidn_oif);
994		err = -ENODEV;
995		if (dev_out && dev_out->dn_ptr == NULL) {
996			dev_put(dev_out);
997			dev_out = NULL;
998		}
999		if (dev_out == NULL)
1000			goto out;
1001	}
1002
1003	/* If we have a source address, verify that its a local address */
1004	if (oldflp->saddr) {
1005		err = -EADDRNOTAVAIL;
1006
1007		if (dev_out) {
1008			if (dn_dev_islocal(dev_out, oldflp->saddr))
1009				goto source_ok;
1010			dev_put(dev_out);
1011			goto out;
1012		}
1013		rcu_read_lock();
1014		for_each_netdev_rcu(&init_net, dev) {
1015			if (!dev->dn_ptr)
1016				continue;
1017			if (!dn_dev_islocal(dev, oldflp->saddr))
1018				continue;
1019			if ((dev->flags & IFF_LOOPBACK) &&
1020			    oldflp->daddr &&
1021			    !dn_dev_islocal(dev, oldflp->daddr))
1022				continue;
1023
1024			dev_out = dev;
1025			break;
1026		}
1027		rcu_read_unlock();
1028		if (dev_out == NULL)
1029			goto out;
1030		dev_hold(dev_out);
1031source_ok:
1032		;
1033	}
1034
1035	/* No destination? Assume its local */
1036	if (!fld.daddr) {
1037		fld.daddr = fld.saddr;
1038
1039		err = -EADDRNOTAVAIL;
1040		if (dev_out)
1041			dev_put(dev_out);
1042		dev_out = init_net.loopback_dev;
1043		dev_hold(dev_out);
1044		if (!fld.daddr) {
1045			fld.daddr =
1046			fld.saddr = dnet_select_source(dev_out, 0,
1047						       RT_SCOPE_HOST);
1048			if (!fld.daddr)
1049				goto out;
1050		}
1051		fld.flowidn_oif = LOOPBACK_IFINDEX;
1052		res.type = RTN_LOCAL;
1053		goto make_route;
1054	}
1055
1056	if (decnet_debug_level & 16)
1057		printk(KERN_DEBUG
1058		       "dn_route_output_slow: initial checks complete."
1059		       " dst=%04x src=%04x oif=%d try_hard=%d\n",
1060		       le16_to_cpu(fld.daddr), le16_to_cpu(fld.saddr),
1061		       fld.flowidn_oif, try_hard);
1062
1063	/*
1064	 * N.B. If the kernel is compiled without router support then
1065	 * dn_fib_lookup() will evaluate to non-zero so this if () block
1066	 * will always be executed.
1067	 */
1068	err = -ESRCH;
1069	if (try_hard || (err = dn_fib_lookup(&fld, &res)) != 0) {
1070		struct dn_dev *dn_db;
1071		if (err != -ESRCH)
1072			goto out;
1073		/*
1074		 * Here the fallback is basically the standard algorithm for
1075		 * routing in endnodes which is described in the DECnet routing
1076		 * docs
1077		 *
1078		 * If we are not trying hard, look in neighbour cache.
1079		 * The result is tested to ensure that if a specific output
1080		 * device/source address was requested, then we honour that
1081		 * here
1082		 */
1083		if (!try_hard) {
1084			neigh = neigh_lookup_nodev(&dn_neigh_table, &init_net, &fld.daddr);
1085			if (neigh) {
1086				if ((oldflp->flowidn_oif &&
1087				    (neigh->dev->ifindex != oldflp->flowidn_oif)) ||
1088				    (oldflp->saddr &&
1089				    (!dn_dev_islocal(neigh->dev,
1090						     oldflp->saddr)))) {
1091					neigh_release(neigh);
1092					neigh = NULL;
1093				} else {
1094					if (dev_out)
1095						dev_put(dev_out);
1096					if (dn_dev_islocal(neigh->dev, fld.daddr)) {
1097						dev_out = init_net.loopback_dev;
1098						res.type = RTN_LOCAL;
1099					} else {
1100						dev_out = neigh->dev;
1101					}
1102					dev_hold(dev_out);
1103					goto select_source;
1104				}
1105			}
1106		}
1107
1108		/* Not there? Perhaps its a local address */
1109		if (dev_out == NULL)
1110			dev_out = dn_dev_get_default();
1111		err = -ENODEV;
1112		if (dev_out == NULL)
1113			goto out;
1114		dn_db = rcu_dereference_raw(dev_out->dn_ptr);
1115		/* Possible improvement - check all devices for local addr */
1116		if (dn_dev_islocal(dev_out, fld.daddr)) {
1117			dev_put(dev_out);
1118			dev_out = init_net.loopback_dev;
1119			dev_hold(dev_out);
1120			res.type = RTN_LOCAL;
1121			goto select_source;
1122		}
1123		/* Not local either.... try sending it to the default router */
1124		neigh = neigh_clone(dn_db->router);
1125		BUG_ON(neigh && neigh->dev != dev_out);
1126
1127		/* Ok then, we assume its directly connected and move on */
1128select_source:
1129		if (neigh)
1130			gateway = ((struct dn_neigh *)neigh)->addr;
1131		if (gateway == 0)
1132			gateway = fld.daddr;
1133		if (fld.saddr == 0) {
1134			fld.saddr = dnet_select_source(dev_out, gateway,
1135						       res.type == RTN_LOCAL ?
1136						       RT_SCOPE_HOST :
1137						       RT_SCOPE_LINK);
1138			if (fld.saddr == 0 && res.type != RTN_LOCAL)
1139				goto e_addr;
1140		}
1141		fld.flowidn_oif = dev_out->ifindex;
1142		goto make_route;
1143	}
1144	free_res = 1;
1145
1146	if (res.type == RTN_NAT)
1147		goto e_inval;
1148
1149	if (res.type == RTN_LOCAL) {
1150		if (!fld.saddr)
1151			fld.saddr = fld.daddr;
1152		if (dev_out)
1153			dev_put(dev_out);
1154		dev_out = init_net.loopback_dev;
1155		dev_hold(dev_out);
1156		fld.flowidn_oif = dev_out->ifindex;
1157		if (res.fi)
1158			dn_fib_info_put(res.fi);
1159		res.fi = NULL;
1160		goto make_route;
1161	}
1162
1163	if (res.fi->fib_nhs > 1 && fld.flowidn_oif == 0)
1164		dn_fib_select_multipath(&fld, &res);
1165
1166	/*
1167	 * We could add some logic to deal with default routes here and
1168	 * get rid of some of the special casing above.
1169	 */
1170
1171	if (!fld.saddr)
1172		fld.saddr = DN_FIB_RES_PREFSRC(res);
1173
1174	if (dev_out)
1175		dev_put(dev_out);
1176	dev_out = DN_FIB_RES_DEV(res);
1177	dev_hold(dev_out);
1178	fld.flowidn_oif = dev_out->ifindex;
1179	gateway = DN_FIB_RES_GW(res);
1180
1181make_route:
1182	if (dev_out->flags & IFF_LOOPBACK)
1183		flags |= RTCF_LOCAL;
1184
1185	rt = dst_alloc(&dn_dst_ops, dev_out, 1, DST_OBSOLETE_NONE, DST_HOST);
1186	if (rt == NULL)
1187		goto e_nobufs;
1188
1189	memset(&rt->fld, 0, sizeof(rt->fld));
1190	rt->fld.saddr        = oldflp->saddr;
1191	rt->fld.daddr        = oldflp->daddr;
1192	rt->fld.flowidn_oif  = oldflp->flowidn_oif;
1193	rt->fld.flowidn_iif  = 0;
1194	rt->fld.flowidn_mark = oldflp->flowidn_mark;
1195
1196	rt->rt_saddr      = fld.saddr;
1197	rt->rt_daddr      = fld.daddr;
1198	rt->rt_gateway    = gateway ? gateway : fld.daddr;
1199	rt->rt_local_src  = fld.saddr;
1200
1201	rt->rt_dst_map    = fld.daddr;
1202	rt->rt_src_map    = fld.saddr;
1203
1204	rt->n = neigh;
1205	neigh = NULL;
1206
1207	rt->dst.lastuse = jiffies;
1208	rt->dst.output  = dn_output;
1209	rt->dst.input   = dn_rt_bug;
1210	rt->rt_flags      = flags;
1211	if (flags & RTCF_LOCAL)
1212		rt->dst.input = dn_nsp_rx;
1213
1214	err = dn_rt_set_next_hop(rt, &res);
1215	if (err)
1216		goto e_neighbour;
1217
1218	hash = dn_hash(rt->fld.saddr, rt->fld.daddr);
1219	dn_insert_route(rt, hash, (struct dn_route **)pprt);
1220
1221done:
1222	if (neigh)
1223		neigh_release(neigh);
1224	if (free_res)
1225		dn_fib_res_put(&res);
1226	if (dev_out)
1227		dev_put(dev_out);
1228out:
1229	return err;
1230
1231e_addr:
1232	err = -EADDRNOTAVAIL;
1233	goto done;
1234e_inval:
1235	err = -EINVAL;
1236	goto done;
1237e_nobufs:
1238	err = -ENOBUFS;
1239	goto done;
1240e_neighbour:
1241	dst_free(&rt->dst);
1242	goto e_nobufs;
1243}
1244
1245
1246/*
1247 * N.B. The flags may be moved into the flowi at some future stage.
1248 */
1249static int __dn_route_output_key(struct dst_entry **pprt, const struct flowidn *flp, int flags)
1250{
1251	unsigned int hash = dn_hash(flp->saddr, flp->daddr);
1252	struct dn_route *rt = NULL;
1253
1254	if (!(flags & MSG_TRYHARD)) {
1255		rcu_read_lock_bh();
1256		for (rt = rcu_dereference_bh(dn_rt_hash_table[hash].chain); rt;
1257			rt = rcu_dereference_bh(rt->dst.dn_next)) {
1258			if ((flp->daddr == rt->fld.daddr) &&
1259			    (flp->saddr == rt->fld.saddr) &&
1260			    (flp->flowidn_mark == rt->fld.flowidn_mark) &&
1261			    dn_is_output_route(rt) &&
1262			    (rt->fld.flowidn_oif == flp->flowidn_oif)) {
1263				dst_use(&rt->dst, jiffies);
1264				rcu_read_unlock_bh();
1265				*pprt = &rt->dst;
1266				return 0;
1267			}
1268		}
1269		rcu_read_unlock_bh();
1270	}
1271
1272	return dn_route_output_slow(pprt, flp, flags);
1273}
1274
1275static int dn_route_output_key(struct dst_entry **pprt, struct flowidn *flp, int flags)
1276{
1277	int err;
1278
1279	err = __dn_route_output_key(pprt, flp, flags);
1280	if (err == 0 && flp->flowidn_proto) {
1281		*pprt = xfrm_lookup(&init_net, *pprt,
1282				    flowidn_to_flowi(flp), NULL, 0);
1283		if (IS_ERR(*pprt)) {
1284			err = PTR_ERR(*pprt);
1285			*pprt = NULL;
1286		}
1287	}
1288	return err;
1289}
1290
1291int dn_route_output_sock(struct dst_entry __rcu **pprt, struct flowidn *fl, struct sock *sk, int flags)
1292{
1293	int err;
1294
1295	err = __dn_route_output_key(pprt, fl, flags & MSG_TRYHARD);
1296	if (err == 0 && fl->flowidn_proto) {
1297		*pprt = xfrm_lookup(&init_net, *pprt,
1298				    flowidn_to_flowi(fl), sk, 0);
1299		if (IS_ERR(*pprt)) {
1300			err = PTR_ERR(*pprt);
1301			*pprt = NULL;
1302		}
1303	}
1304	return err;
1305}
1306
1307static int dn_route_input_slow(struct sk_buff *skb)
1308{
1309	struct dn_route *rt = NULL;
1310	struct dn_skb_cb *cb = DN_SKB_CB(skb);
1311	struct net_device *in_dev = skb->dev;
1312	struct net_device *out_dev = NULL;
1313	struct dn_dev *dn_db;
1314	struct neighbour *neigh = NULL;
1315	unsigned int hash;
1316	int flags = 0;
1317	__le16 gateway = 0;
1318	__le16 local_src = 0;
1319	struct flowidn fld = {
1320		.daddr = cb->dst,
1321		.saddr = cb->src,
1322		.flowidn_scope = RT_SCOPE_UNIVERSE,
1323		.flowidn_mark = skb->mark,
1324		.flowidn_iif = skb->dev->ifindex,
1325	};
1326	struct dn_fib_res res = { .fi = NULL, .type = RTN_UNREACHABLE };
1327	int err = -EINVAL;
1328	int free_res = 0;
1329
1330	dev_hold(in_dev);
1331
1332	if ((dn_db = rcu_dereference(in_dev->dn_ptr)) == NULL)
1333		goto out;
1334
1335	/* Zero source addresses are not allowed */
1336	if (fld.saddr == 0)
1337		goto out;
1338
1339	/*
1340	 * In this case we've just received a packet from a source
1341	 * outside ourselves pretending to come from us. We don't
1342	 * allow it any further to prevent routing loops, spoofing and
1343	 * other nasties. Loopback packets already have the dst attached
1344	 * so this only affects packets which have originated elsewhere.
1345	 */
1346	err  = -ENOTUNIQ;
1347	if (dn_dev_islocal(in_dev, cb->src))
1348		goto out;
1349
1350	err = dn_fib_lookup(&fld, &res);
1351	if (err) {
1352		if (err != -ESRCH)
1353			goto out;
1354		/*
1355		 * Is the destination us ?
1356		 */
1357		if (!dn_dev_islocal(in_dev, cb->dst))
1358			goto e_inval;
1359
1360		res.type = RTN_LOCAL;
1361	} else {
1362		__le16 src_map = fld.saddr;
1363		free_res = 1;
1364
1365		out_dev = DN_FIB_RES_DEV(res);
1366		if (out_dev == NULL) {
1367			net_crit_ratelimited("Bug in dn_route_input_slow() No output device\n");
1368			goto e_inval;
1369		}
1370		dev_hold(out_dev);
1371
1372		if (res.r)
1373			src_map = fld.saddr; /* no NAT support for now */
1374
1375		gateway = DN_FIB_RES_GW(res);
1376		if (res.type == RTN_NAT) {
1377			fld.daddr = dn_fib_rules_map_destination(fld.daddr, &res);
1378			dn_fib_res_put(&res);
1379			free_res = 0;
1380			if (dn_fib_lookup(&fld, &res))
1381				goto e_inval;
1382			free_res = 1;
1383			if (res.type != RTN_UNICAST)
1384				goto e_inval;
1385			flags |= RTCF_DNAT;
1386			gateway = fld.daddr;
1387		}
1388		fld.saddr = src_map;
1389	}
1390
1391	switch(res.type) {
1392	case RTN_UNICAST:
1393		/*
1394		 * Forwarding check here, we only check for forwarding
1395		 * being turned off, if you want to only forward intra
1396		 * area, its up to you to set the routing tables up
1397		 * correctly.
1398		 */
1399		if (dn_db->parms.forwarding == 0)
1400			goto e_inval;
1401
1402		if (res.fi->fib_nhs > 1 && fld.flowidn_oif == 0)
1403			dn_fib_select_multipath(&fld, &res);
1404
1405		/*
1406		 * Check for out_dev == in_dev. We use the RTCF_DOREDIRECT
1407		 * flag as a hint to set the intra-ethernet bit when
1408		 * forwarding. If we've got NAT in operation, we don't do
1409		 * this optimisation.
1410		 */
1411		if (out_dev == in_dev && !(flags & RTCF_NAT))
1412			flags |= RTCF_DOREDIRECT;
1413
1414		local_src = DN_FIB_RES_PREFSRC(res);
1415
1416	case RTN_BLACKHOLE:
1417	case RTN_UNREACHABLE:
1418		break;
1419	case RTN_LOCAL:
1420		flags |= RTCF_LOCAL;
1421		fld.saddr = cb->dst;
1422		fld.daddr = cb->src;
1423
1424		/* Routing tables gave us a gateway */
1425		if (gateway)
1426			goto make_route;
1427
1428		/* Packet was intra-ethernet, so we know its on-link */
1429		if (cb->rt_flags & DN_RT_F_IE) {
1430			gateway = cb->src;
1431			goto make_route;
1432		}
1433
1434		/* Use the default router if there is one */
1435		neigh = neigh_clone(dn_db->router);
1436		if (neigh) {
1437			gateway = ((struct dn_neigh *)neigh)->addr;
1438			goto make_route;
1439		}
1440
1441		/* Close eyes and pray */
1442		gateway = cb->src;
1443		goto make_route;
1444	default:
1445		goto e_inval;
1446	}
1447
1448make_route:
1449	rt = dst_alloc(&dn_dst_ops, out_dev, 0, DST_OBSOLETE_NONE, DST_HOST);
1450	if (rt == NULL)
1451		goto e_nobufs;
1452
1453	memset(&rt->fld, 0, sizeof(rt->fld));
1454	rt->rt_saddr      = fld.saddr;
1455	rt->rt_daddr      = fld.daddr;
1456	rt->rt_gateway    = fld.daddr;
1457	if (gateway)
1458		rt->rt_gateway = gateway;
1459	rt->rt_local_src  = local_src ? local_src : rt->rt_saddr;
1460
1461	rt->rt_dst_map    = fld.daddr;
1462	rt->rt_src_map    = fld.saddr;
1463
1464	rt->fld.saddr        = cb->src;
1465	rt->fld.daddr        = cb->dst;
1466	rt->fld.flowidn_oif  = 0;
1467	rt->fld.flowidn_iif  = in_dev->ifindex;
1468	rt->fld.flowidn_mark = fld.flowidn_mark;
1469
1470	rt->n = neigh;
1471	rt->dst.lastuse = jiffies;
1472	rt->dst.output = dn_rt_bug_sk;
1473	switch (res.type) {
1474	case RTN_UNICAST:
1475		rt->dst.input = dn_forward;
1476		break;
1477	case RTN_LOCAL:
1478		rt->dst.output = dn_output;
1479		rt->dst.input = dn_nsp_rx;
1480		rt->dst.dev = in_dev;
1481		flags |= RTCF_LOCAL;
1482		break;
1483	default:
1484	case RTN_UNREACHABLE:
1485	case RTN_BLACKHOLE:
1486		rt->dst.input = dst_discard;
1487	}
1488	rt->rt_flags = flags;
1489
1490	err = dn_rt_set_next_hop(rt, &res);
1491	if (err)
1492		goto e_neighbour;
1493
1494	hash = dn_hash(rt->fld.saddr, rt->fld.daddr);
1495	dn_insert_route(rt, hash, &rt);
1496	skb_dst_set(skb, &rt->dst);
1497
1498done:
1499	if (neigh)
1500		neigh_release(neigh);
1501	if (free_res)
1502		dn_fib_res_put(&res);
1503	dev_put(in_dev);
1504	if (out_dev)
1505		dev_put(out_dev);
1506out:
1507	return err;
1508
1509e_inval:
1510	err = -EINVAL;
1511	goto done;
1512
1513e_nobufs:
1514	err = -ENOBUFS;
1515	goto done;
1516
1517e_neighbour:
1518	dst_free(&rt->dst);
1519	goto done;
1520}
1521
1522static int dn_route_input(struct sk_buff *skb)
1523{
1524	struct dn_route *rt;
1525	struct dn_skb_cb *cb = DN_SKB_CB(skb);
1526	unsigned int hash = dn_hash(cb->src, cb->dst);
1527
1528	if (skb_dst(skb))
1529		return 0;
1530
1531	rcu_read_lock();
1532	for(rt = rcu_dereference(dn_rt_hash_table[hash].chain); rt != NULL;
1533	    rt = rcu_dereference(rt->dst.dn_next)) {
1534		if ((rt->fld.saddr == cb->src) &&
1535		    (rt->fld.daddr == cb->dst) &&
1536		    (rt->fld.flowidn_oif == 0) &&
1537		    (rt->fld.flowidn_mark == skb->mark) &&
1538		    (rt->fld.flowidn_iif == cb->iif)) {
1539			dst_use(&rt->dst, jiffies);
1540			rcu_read_unlock();
1541			skb_dst_set(skb, (struct dst_entry *)rt);
1542			return 0;
1543		}
1544	}
1545	rcu_read_unlock();
1546
1547	return dn_route_input_slow(skb);
1548}
1549
1550static int dn_rt_fill_info(struct sk_buff *skb, u32 portid, u32 seq,
1551			   int event, int nowait, unsigned int flags)
1552{
1553	struct dn_route *rt = (struct dn_route *)skb_dst(skb);
1554	struct rtmsg *r;
1555	struct nlmsghdr *nlh;
1556	long expires;
1557
1558	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*r), flags);
1559	if (!nlh)
1560		return -EMSGSIZE;
1561
1562	r = nlmsg_data(nlh);
1563	r->rtm_family = AF_DECnet;
1564	r->rtm_dst_len = 16;
1565	r->rtm_src_len = 0;
1566	r->rtm_tos = 0;
1567	r->rtm_table = RT_TABLE_MAIN;
1568	r->rtm_type = rt->rt_type;
1569	r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
1570	r->rtm_scope = RT_SCOPE_UNIVERSE;
1571	r->rtm_protocol = RTPROT_UNSPEC;
1572
1573	if (rt->rt_flags & RTCF_NOTIFY)
1574		r->rtm_flags |= RTM_F_NOTIFY;
1575
1576	if (nla_put_u32(skb, RTA_TABLE, RT_TABLE_MAIN) < 0 ||
1577	    nla_put_le16(skb, RTA_DST, rt->rt_daddr) < 0)
1578		goto errout;
1579
1580	if (rt->fld.saddr) {
1581		r->rtm_src_len = 16;
1582		if (nla_put_le16(skb, RTA_SRC, rt->fld.saddr) < 0)
1583			goto errout;
1584	}
1585	if (rt->dst.dev &&
1586	    nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex) < 0)
1587		goto errout;
1588
1589	/*
1590	 * Note to self - change this if input routes reverse direction when
1591	 * they deal only with inputs and not with replies like they do
1592	 * currently.
1593	 */
1594	if (nla_put_le16(skb, RTA_PREFSRC, rt->rt_local_src) < 0)
1595		goto errout;
1596
1597	if (rt->rt_daddr != rt->rt_gateway &&
1598	    nla_put_le16(skb, RTA_GATEWAY, rt->rt_gateway) < 0)
1599		goto errout;
1600
1601	if (rtnetlink_put_metrics(skb, dst_metrics_ptr(&rt->dst)) < 0)
1602		goto errout;
1603
1604	expires = rt->dst.expires ? rt->dst.expires - jiffies : 0;
1605	if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires,
1606			       rt->dst.error) < 0)
1607		goto errout;
1608
1609	if (dn_is_input_route(rt) &&
1610	    nla_put_u32(skb, RTA_IIF, rt->fld.flowidn_iif) < 0)
1611		goto errout;
1612
1613	nlmsg_end(skb, nlh);
1614	return 0;
1615
1616errout:
1617	nlmsg_cancel(skb, nlh);
1618	return -EMSGSIZE;
1619}
1620
1621const struct nla_policy rtm_dn_policy[RTA_MAX + 1] = {
1622	[RTA_DST]		= { .type = NLA_U16 },
1623	[RTA_SRC]		= { .type = NLA_U16 },
1624	[RTA_IIF]		= { .type = NLA_U32 },
1625	[RTA_OIF]		= { .type = NLA_U32 },
1626	[RTA_GATEWAY]		= { .type = NLA_U16 },
1627	[RTA_PRIORITY]		= { .type = NLA_U32 },
1628	[RTA_PREFSRC]		= { .type = NLA_U16 },
1629	[RTA_METRICS]		= { .type = NLA_NESTED },
1630	[RTA_MULTIPATH]		= { .type = NLA_NESTED },
1631	[RTA_TABLE]		= { .type = NLA_U32 },
1632	[RTA_MARK]		= { .type = NLA_U32 },
1633};
1634
1635/*
1636 * This is called by both endnodes and routers now.
1637 */
1638static int dn_cache_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh)
1639{
1640	struct net *net = sock_net(in_skb->sk);
1641	struct rtmsg *rtm = nlmsg_data(nlh);
1642	struct dn_route *rt = NULL;
1643	struct dn_skb_cb *cb;
1644	int err;
1645	struct sk_buff *skb;
1646	struct flowidn fld;
1647	struct nlattr *tb[RTA_MAX+1];
1648
1649	if (!net_eq(net, &init_net))
1650		return -EINVAL;
1651
1652	err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_dn_policy);
1653	if (err < 0)
1654		return err;
1655
1656	memset(&fld, 0, sizeof(fld));
1657	fld.flowidn_proto = DNPROTO_NSP;
1658
1659	skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1660	if (skb == NULL)
1661		return -ENOBUFS;
1662	skb_reset_mac_header(skb);
1663	cb = DN_SKB_CB(skb);
1664
1665	if (tb[RTA_SRC])
1666		fld.saddr = nla_get_le16(tb[RTA_SRC]);
1667
1668	if (tb[RTA_DST])
1669		fld.daddr = nla_get_le16(tb[RTA_DST]);
1670
1671	if (tb[RTA_IIF])
1672		fld.flowidn_iif = nla_get_u32(tb[RTA_IIF]);
1673
1674	if (fld.flowidn_iif) {
1675		struct net_device *dev;
1676		dev = __dev_get_by_index(&init_net, fld.flowidn_iif);
1677		if (!dev || !dev->dn_ptr) {
1678			kfree_skb(skb);
1679			return -ENODEV;
1680		}
1681		skb->protocol = htons(ETH_P_DNA_RT);
1682		skb->dev = dev;
1683		cb->src = fld.saddr;
1684		cb->dst = fld.daddr;
1685		local_bh_disable();
1686		err = dn_route_input(skb);
1687		local_bh_enable();
1688		memset(cb, 0, sizeof(struct dn_skb_cb));
1689		rt = (struct dn_route *)skb_dst(skb);
1690		if (!err && -rt->dst.error)
1691			err = rt->dst.error;
1692	} else {
1693		if (tb[RTA_OIF])
1694			fld.flowidn_oif = nla_get_u32(tb[RTA_OIF]);
1695
1696		err = dn_route_output_key((struct dst_entry **)&rt, &fld, 0);
1697	}
1698
1699	skb->dev = NULL;
1700	if (err)
1701		goto out_free;
1702	skb_dst_set(skb, &rt->dst);
1703	if (rtm->rtm_flags & RTM_F_NOTIFY)
1704		rt->rt_flags |= RTCF_NOTIFY;
1705
1706	err = dn_rt_fill_info(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq, RTM_NEWROUTE, 0, 0);
1707	if (err < 0) {
1708		err = -EMSGSIZE;
1709		goto out_free;
1710	}
1711
1712	return rtnl_unicast(skb, &init_net, NETLINK_CB(in_skb).portid);
1713
1714out_free:
1715	kfree_skb(skb);
1716	return err;
1717}
1718
1719/*
1720 * For routers, this is called from dn_fib_dump, but for endnodes its
1721 * called directly from the rtnetlink dispatch table.
1722 */
1723int dn_cache_dump(struct sk_buff *skb, struct netlink_callback *cb)
1724{
1725	struct net *net = sock_net(skb->sk);
1726	struct dn_route *rt;
1727	int h, s_h;
1728	int idx, s_idx;
1729	struct rtmsg *rtm;
1730
1731	if (!net_eq(net, &init_net))
1732		return 0;
1733
1734	if (nlmsg_len(cb->nlh) < sizeof(struct rtmsg))
1735		return -EINVAL;
1736
1737	rtm = nlmsg_data(cb->nlh);
1738	if (!(rtm->rtm_flags & RTM_F_CLONED))
1739		return 0;
1740
1741	s_h = cb->args[0];
1742	s_idx = idx = cb->args[1];
1743	for(h = 0; h <= dn_rt_hash_mask; h++) {
1744		if (h < s_h)
1745			continue;
1746		if (h > s_h)
1747			s_idx = 0;
1748		rcu_read_lock_bh();
1749		for(rt = rcu_dereference_bh(dn_rt_hash_table[h].chain), idx = 0;
1750			rt;
1751			rt = rcu_dereference_bh(rt->dst.dn_next), idx++) {
1752			if (idx < s_idx)
1753				continue;
1754			skb_dst_set(skb, dst_clone(&rt->dst));
1755			if (dn_rt_fill_info(skb, NETLINK_CB(cb->skb).portid,
1756					cb->nlh->nlmsg_seq, RTM_NEWROUTE,
1757					1, NLM_F_MULTI) < 0) {
1758				skb_dst_drop(skb);
1759				rcu_read_unlock_bh();
1760				goto done;
1761			}
1762			skb_dst_drop(skb);
1763		}
1764		rcu_read_unlock_bh();
1765	}
1766
1767done:
1768	cb->args[0] = h;
1769	cb->args[1] = idx;
1770	return skb->len;
1771}
1772
1773#ifdef CONFIG_PROC_FS
1774struct dn_rt_cache_iter_state {
1775	int bucket;
1776};
1777
1778static struct dn_route *dn_rt_cache_get_first(struct seq_file *seq)
1779{
1780	struct dn_route *rt = NULL;
1781	struct dn_rt_cache_iter_state *s = seq->private;
1782
1783	for(s->bucket = dn_rt_hash_mask; s->bucket >= 0; --s->bucket) {
1784		rcu_read_lock_bh();
1785		rt = rcu_dereference_bh(dn_rt_hash_table[s->bucket].chain);
1786		if (rt)
1787			break;
1788		rcu_read_unlock_bh();
1789	}
1790	return rt;
1791}
1792
1793static struct dn_route *dn_rt_cache_get_next(struct seq_file *seq, struct dn_route *rt)
1794{
1795	struct dn_rt_cache_iter_state *s = seq->private;
1796
1797	rt = rcu_dereference_bh(rt->dst.dn_next);
1798	while (!rt) {
1799		rcu_read_unlock_bh();
1800		if (--s->bucket < 0)
1801			break;
1802		rcu_read_lock_bh();
1803		rt = rcu_dereference_bh(dn_rt_hash_table[s->bucket].chain);
1804	}
1805	return rt;
1806}
1807
1808static void *dn_rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
1809{
1810	struct dn_route *rt = dn_rt_cache_get_first(seq);
1811
1812	if (rt) {
1813		while(*pos && (rt = dn_rt_cache_get_next(seq, rt)))
1814			--*pos;
1815	}
1816	return *pos ? NULL : rt;
1817}
1818
1819static void *dn_rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1820{
1821	struct dn_route *rt = dn_rt_cache_get_next(seq, v);
1822	++*pos;
1823	return rt;
1824}
1825
1826static void dn_rt_cache_seq_stop(struct seq_file *seq, void *v)
1827{
1828	if (v)
1829		rcu_read_unlock_bh();
1830}
1831
1832static int dn_rt_cache_seq_show(struct seq_file *seq, void *v)
1833{
1834	struct dn_route *rt = v;
1835	char buf1[DN_ASCBUF_LEN], buf2[DN_ASCBUF_LEN];
1836
1837	seq_printf(seq, "%-8s %-7s %-7s %04d %04d %04d\n",
1838		   rt->dst.dev ? rt->dst.dev->name : "*",
1839		   dn_addr2asc(le16_to_cpu(rt->rt_daddr), buf1),
1840		   dn_addr2asc(le16_to_cpu(rt->rt_saddr), buf2),
1841		   atomic_read(&rt->dst.__refcnt),
1842		   rt->dst.__use, 0);
1843	return 0;
1844}
1845
1846static const struct seq_operations dn_rt_cache_seq_ops = {
1847	.start	= dn_rt_cache_seq_start,
1848	.next	= dn_rt_cache_seq_next,
1849	.stop	= dn_rt_cache_seq_stop,
1850	.show	= dn_rt_cache_seq_show,
1851};
1852
1853static int dn_rt_cache_seq_open(struct inode *inode, struct file *file)
1854{
1855	return seq_open_private(file, &dn_rt_cache_seq_ops,
1856			sizeof(struct dn_rt_cache_iter_state));
1857}
1858
1859static const struct file_operations dn_rt_cache_seq_fops = {
1860	.owner	 = THIS_MODULE,
1861	.open	 = dn_rt_cache_seq_open,
1862	.read	 = seq_read,
1863	.llseek	 = seq_lseek,
1864	.release = seq_release_private,
1865};
1866
1867#endif /* CONFIG_PROC_FS */
1868
1869void __init dn_route_init(void)
1870{
1871	int i, goal, order;
1872
1873	dn_dst_ops.kmem_cachep =
1874		kmem_cache_create("dn_dst_cache", sizeof(struct dn_route), 0,
1875				  SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1876	dst_entries_init(&dn_dst_ops);
1877	setup_timer(&dn_route_timer, dn_dst_check_expire, 0);
1878	dn_route_timer.expires = jiffies + decnet_dst_gc_interval * HZ;
1879	add_timer(&dn_route_timer);
1880
1881	goal = totalram_pages >> (26 - PAGE_SHIFT);
1882
1883	for(order = 0; (1UL << order) < goal; order++)
1884		/* NOTHING */;
1885
1886	/*
1887	 * Only want 1024 entries max, since the table is very, very unlikely
1888	 * to be larger than that.
1889	 */
1890	while(order && ((((1UL << order) * PAGE_SIZE) /
1891				sizeof(struct dn_rt_hash_bucket)) >= 2048))
1892		order--;
1893
1894	do {
1895		dn_rt_hash_mask = (1UL << order) * PAGE_SIZE /
1896			sizeof(struct dn_rt_hash_bucket);
1897		while(dn_rt_hash_mask & (dn_rt_hash_mask - 1))
1898			dn_rt_hash_mask--;
1899		dn_rt_hash_table = (struct dn_rt_hash_bucket *)
1900			__get_free_pages(GFP_ATOMIC, order);
1901	} while (dn_rt_hash_table == NULL && --order > 0);
1902
1903	if (!dn_rt_hash_table)
1904		panic("Failed to allocate DECnet route cache hash table\n");
1905
1906	printk(KERN_INFO
1907		"DECnet: Routing cache hash table of %u buckets, %ldKbytes\n",
1908		dn_rt_hash_mask,
1909		(long)(dn_rt_hash_mask*sizeof(struct dn_rt_hash_bucket))/1024);
1910
1911	dn_rt_hash_mask--;
1912	for(i = 0; i <= dn_rt_hash_mask; i++) {
1913		spin_lock_init(&dn_rt_hash_table[i].lock);
1914		dn_rt_hash_table[i].chain = NULL;
1915	}
1916
1917	dn_dst_ops.gc_thresh = (dn_rt_hash_mask + 1);
1918
1919	proc_create("decnet_cache", S_IRUGO, init_net.proc_net,
1920		    &dn_rt_cache_seq_fops);
1921
1922#ifdef CONFIG_DECNET_ROUTER
1923	rtnl_register(PF_DECnet, RTM_GETROUTE, dn_cache_getroute,
1924		      dn_fib_dump, NULL);
1925#else
1926	rtnl_register(PF_DECnet, RTM_GETROUTE, dn_cache_getroute,
1927		      dn_cache_dump, NULL);
1928#endif
1929}
1930
1931void __exit dn_route_cleanup(void)
1932{
1933	del_timer(&dn_route_timer);
1934	dn_run_flush(0);
1935
1936	remove_proc_entry("decnet_cache", init_net.proc_net);
1937	dst_entries_destroy(&dn_dst_ops);
1938}
1939
1940