Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * DECnet An implementation of the DECnet protocol suite for the LINUX
4 * operating system. DECnet is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
6 *
7 * DECnet Routing Functions (Endnode and Router)
8 *
9 * Authors: Steve Whitehouse <SteveW@ACM.org>
10 * Eduardo Marcelo Serrat <emserrat@geocities.com>
11 *
12 * Changes:
13 * Steve Whitehouse : Fixes to allow "intra-ethernet" and
14 * "return-to-sender" bits on outgoing
15 * packets.
16 * Steve Whitehouse : Timeouts for cached routes.
17 * Steve Whitehouse : Use dst cache for input routes too.
18 * Steve Whitehouse : Fixed error values in dn_send_skb.
19 * Steve Whitehouse : Rework routing functions to better fit
20 * DECnet routing design
21 * Alexey Kuznetsov : New SMP locking
22 * Steve Whitehouse : More SMP locking changes & dn_cache_dump()
23 * Steve Whitehouse : Prerouting NF hook, now really is prerouting.
24 * Fixed possible skb leak in rtnetlink funcs.
25 * Steve Whitehouse : Dave Miller's dynamic hash table sizing and
26 * Alexey Kuznetsov's finer grained locking
27 * from ipv4/route.c.
28 * Steve Whitehouse : Routing is now starting to look like a
29 * sensible set of code now, mainly due to
30 * my copying the IPv4 routing code. The
31 * hooks here are modified and will continue
32 * to evolve for a while.
33 * Steve Whitehouse : Real SMP at last :-) Also new netfilter
34 * stuff. Look out raw sockets your days
35 * are numbered!
36 * Steve Whitehouse : Added return-to-sender functions. Added
37 * backlog congestion level return codes.
38 * Steve Whitehouse : Fixed bug where routes were set up with
39 * no ref count on net devices.
40 * Steve Whitehouse : RCU for the route cache
41 * Steve Whitehouse : Preparations for the flow cache
42 * Steve Whitehouse : Prepare for nonlinear skbs
43 */
44
45/******************************************************************************
46 (c) 1995-1998 E.M. Serrat emserrat@geocities.com
47
48*******************************************************************************/
49
50#include <linux/errno.h>
51#include <linux/types.h>
52#include <linux/socket.h>
53#include <linux/in.h>
54#include <linux/kernel.h>
55#include <linux/sockios.h>
56#include <linux/net.h>
57#include <linux/netdevice.h>
58#include <linux/inet.h>
59#include <linux/route.h>
60#include <linux/in_route.h>
61#include <linux/slab.h>
62#include <net/sock.h>
63#include <linux/mm.h>
64#include <linux/proc_fs.h>
65#include <linux/seq_file.h>
66#include <linux/init.h>
67#include <linux/rtnetlink.h>
68#include <linux/string.h>
69#include <linux/netfilter_decnet.h>
70#include <linux/rcupdate.h>
71#include <linux/times.h>
72#include <linux/export.h>
73#include <asm/errno.h>
74#include <net/net_namespace.h>
75#include <net/netlink.h>
76#include <net/neighbour.h>
77#include <net/dst.h>
78#include <net/flow.h>
79#include <net/fib_rules.h>
80#include <net/dn.h>
81#include <net/dn_dev.h>
82#include <net/dn_nsp.h>
83#include <net/dn_route.h>
84#include <net/dn_neigh.h>
85#include <net/dn_fib.h>
86
87struct dn_rt_hash_bucket {
88 struct dn_route __rcu *chain;
89 spinlock_t lock;
90};
91
92extern struct neigh_table dn_neigh_table;
93
94
95static unsigned char dn_hiord_addr[6] = {0xAA, 0x00, 0x04, 0x00, 0x00, 0x00};
96
97static const int dn_rt_min_delay = 2 * HZ;
98static const int dn_rt_max_delay = 10 * HZ;
99static const int dn_rt_mtu_expires = 10 * 60 * HZ;
100
101static unsigned long dn_rt_deadline;
102
103static int dn_dst_gc(struct dst_ops *ops);
104static struct dst_entry *dn_dst_check(struct dst_entry *, __u32);
105static unsigned int dn_dst_default_advmss(const struct dst_entry *dst);
106static unsigned int dn_dst_mtu(const struct dst_entry *dst);
107static void dn_dst_destroy(struct dst_entry *);
108static void dn_dst_ifdown(struct dst_entry *, struct net_device *dev, int how);
109static struct dst_entry *dn_dst_negative_advice(struct dst_entry *);
110static void dn_dst_link_failure(struct sk_buff *);
111static void dn_dst_update_pmtu(struct dst_entry *dst, struct sock *sk,
112 struct sk_buff *skb , u32 mtu,
113 bool confirm_neigh);
114static void dn_dst_redirect(struct dst_entry *dst, struct sock *sk,
115 struct sk_buff *skb);
116static struct neighbour *dn_dst_neigh_lookup(const struct dst_entry *dst,
117 struct sk_buff *skb,
118 const void *daddr);
119static int dn_route_input(struct sk_buff *);
120static void dn_run_flush(struct timer_list *unused);
121
122static struct dn_rt_hash_bucket *dn_rt_hash_table;
123static unsigned int dn_rt_hash_mask;
124
125static struct timer_list dn_route_timer;
126static DEFINE_TIMER(dn_rt_flush_timer, dn_run_flush);
127int decnet_dst_gc_interval = 2;
128
129static struct dst_ops dn_dst_ops = {
130 .family = PF_DECnet,
131 .gc_thresh = 128,
132 .gc = dn_dst_gc,
133 .check = dn_dst_check,
134 .default_advmss = dn_dst_default_advmss,
135 .mtu = dn_dst_mtu,
136 .cow_metrics = dst_cow_metrics_generic,
137 .destroy = dn_dst_destroy,
138 .ifdown = dn_dst_ifdown,
139 .negative_advice = dn_dst_negative_advice,
140 .link_failure = dn_dst_link_failure,
141 .update_pmtu = dn_dst_update_pmtu,
142 .redirect = dn_dst_redirect,
143 .neigh_lookup = dn_dst_neigh_lookup,
144};
145
146static void dn_dst_destroy(struct dst_entry *dst)
147{
148 struct dn_route *rt = (struct dn_route *) dst;
149
150 if (rt->n)
151 neigh_release(rt->n);
152 dst_destroy_metrics_generic(dst);
153}
154
155static void dn_dst_ifdown(struct dst_entry *dst, struct net_device *dev, int how)
156{
157 if (how) {
158 struct dn_route *rt = (struct dn_route *) dst;
159 struct neighbour *n = rt->n;
160
161 if (n && n->dev == dev) {
162 n->dev = dev_net(dev)->loopback_dev;
163 dev_hold(n->dev);
164 dev_put(dev);
165 }
166 }
167}
168
169static __inline__ unsigned int dn_hash(__le16 src, __le16 dst)
170{
171 __u16 tmp = (__u16 __force)(src ^ dst);
172 tmp ^= (tmp >> 3);
173 tmp ^= (tmp >> 5);
174 tmp ^= (tmp >> 10);
175 return dn_rt_hash_mask & (unsigned int)tmp;
176}
177
178static void dn_dst_check_expire(struct timer_list *unused)
179{
180 int i;
181 struct dn_route *rt;
182 struct dn_route __rcu **rtp;
183 unsigned long now = jiffies;
184 unsigned long expire = 120 * HZ;
185
186 for (i = 0; i <= dn_rt_hash_mask; i++) {
187 rtp = &dn_rt_hash_table[i].chain;
188
189 spin_lock(&dn_rt_hash_table[i].lock);
190 while ((rt = rcu_dereference_protected(*rtp,
191 lockdep_is_held(&dn_rt_hash_table[i].lock))) != NULL) {
192 if (atomic_read(&rt->dst.__refcnt) > 1 ||
193 (now - rt->dst.lastuse) < expire) {
194 rtp = &rt->dn_next;
195 continue;
196 }
197 *rtp = rt->dn_next;
198 rt->dn_next = NULL;
199 dst_dev_put(&rt->dst);
200 dst_release(&rt->dst);
201 }
202 spin_unlock(&dn_rt_hash_table[i].lock);
203
204 if ((jiffies - now) > 0)
205 break;
206 }
207
208 mod_timer(&dn_route_timer, now + decnet_dst_gc_interval * HZ);
209}
210
211static int dn_dst_gc(struct dst_ops *ops)
212{
213 struct dn_route *rt;
214 struct dn_route __rcu **rtp;
215 int i;
216 unsigned long now = jiffies;
217 unsigned long expire = 10 * HZ;
218
219 for (i = 0; i <= dn_rt_hash_mask; i++) {
220
221 spin_lock_bh(&dn_rt_hash_table[i].lock);
222 rtp = &dn_rt_hash_table[i].chain;
223
224 while ((rt = rcu_dereference_protected(*rtp,
225 lockdep_is_held(&dn_rt_hash_table[i].lock))) != NULL) {
226 if (atomic_read(&rt->dst.__refcnt) > 1 ||
227 (now - rt->dst.lastuse) < expire) {
228 rtp = &rt->dn_next;
229 continue;
230 }
231 *rtp = rt->dn_next;
232 rt->dn_next = NULL;
233 dst_dev_put(&rt->dst);
234 dst_release(&rt->dst);
235 break;
236 }
237 spin_unlock_bh(&dn_rt_hash_table[i].lock);
238 }
239
240 return 0;
241}
242
243/*
244 * The decnet standards don't impose a particular minimum mtu, what they
245 * do insist on is that the routing layer accepts a datagram of at least
246 * 230 bytes long. Here we have to subtract the routing header length from
247 * 230 to get the minimum acceptable mtu. If there is no neighbour, then we
248 * assume the worst and use a long header size.
249 *
250 * We update both the mtu and the advertised mss (i.e. the segment size we
251 * advertise to the other end).
252 */
253static void dn_dst_update_pmtu(struct dst_entry *dst, struct sock *sk,
254 struct sk_buff *skb, u32 mtu,
255 bool confirm_neigh)
256{
257 struct dn_route *rt = (struct dn_route *) dst;
258 struct neighbour *n = rt->n;
259 u32 min_mtu = 230;
260 struct dn_dev *dn;
261
262 dn = n ? rcu_dereference_raw(n->dev->dn_ptr) : NULL;
263
264 if (dn && dn->use_long == 0)
265 min_mtu -= 6;
266 else
267 min_mtu -= 21;
268
269 if (dst_metric(dst, RTAX_MTU) > mtu && mtu >= min_mtu) {
270 if (!(dst_metric_locked(dst, RTAX_MTU))) {
271 dst_metric_set(dst, RTAX_MTU, mtu);
272 dst_set_expires(dst, dn_rt_mtu_expires);
273 }
274 if (!(dst_metric_locked(dst, RTAX_ADVMSS))) {
275 u32 mss = mtu - DN_MAX_NSP_DATA_HEADER;
276 u32 existing_mss = dst_metric_raw(dst, RTAX_ADVMSS);
277 if (!existing_mss || existing_mss > mss)
278 dst_metric_set(dst, RTAX_ADVMSS, mss);
279 }
280 }
281}
282
283static void dn_dst_redirect(struct dst_entry *dst, struct sock *sk,
284 struct sk_buff *skb)
285{
286}
287
288/*
289 * When a route has been marked obsolete. (e.g. routing cache flush)
290 */
291static struct dst_entry *dn_dst_check(struct dst_entry *dst, __u32 cookie)
292{
293 return NULL;
294}
295
296static struct dst_entry *dn_dst_negative_advice(struct dst_entry *dst)
297{
298 dst_release(dst);
299 return NULL;
300}
301
302static void dn_dst_link_failure(struct sk_buff *skb)
303{
304}
305
306static inline int compare_keys(struct flowidn *fl1, struct flowidn *fl2)
307{
308 return ((fl1->daddr ^ fl2->daddr) |
309 (fl1->saddr ^ fl2->saddr) |
310 (fl1->flowidn_mark ^ fl2->flowidn_mark) |
311 (fl1->flowidn_scope ^ fl2->flowidn_scope) |
312 (fl1->flowidn_oif ^ fl2->flowidn_oif) |
313 (fl1->flowidn_iif ^ fl2->flowidn_iif)) == 0;
314}
315
316static int dn_insert_route(struct dn_route *rt, unsigned int hash, struct dn_route **rp)
317{
318 struct dn_route *rth;
319 struct dn_route __rcu **rthp;
320 unsigned long now = jiffies;
321
322 rthp = &dn_rt_hash_table[hash].chain;
323
324 spin_lock_bh(&dn_rt_hash_table[hash].lock);
325 while ((rth = rcu_dereference_protected(*rthp,
326 lockdep_is_held(&dn_rt_hash_table[hash].lock))) != NULL) {
327 if (compare_keys(&rth->fld, &rt->fld)) {
328 /* Put it first */
329 *rthp = rth->dn_next;
330 rcu_assign_pointer(rth->dn_next,
331 dn_rt_hash_table[hash].chain);
332 rcu_assign_pointer(dn_rt_hash_table[hash].chain, rth);
333
334 dst_hold_and_use(&rth->dst, now);
335 spin_unlock_bh(&dn_rt_hash_table[hash].lock);
336
337 dst_release_immediate(&rt->dst);
338 *rp = rth;
339 return 0;
340 }
341 rthp = &rth->dn_next;
342 }
343
344 rcu_assign_pointer(rt->dn_next, dn_rt_hash_table[hash].chain);
345 rcu_assign_pointer(dn_rt_hash_table[hash].chain, rt);
346
347 dst_hold_and_use(&rt->dst, now);
348 spin_unlock_bh(&dn_rt_hash_table[hash].lock);
349 *rp = rt;
350 return 0;
351}
352
353static void dn_run_flush(struct timer_list *unused)
354{
355 int i;
356 struct dn_route *rt, *next;
357
358 for (i = 0; i < dn_rt_hash_mask; i++) {
359 spin_lock_bh(&dn_rt_hash_table[i].lock);
360
361 rt = xchg((struct dn_route **)&dn_rt_hash_table[i].chain, NULL);
362 if (!rt)
363 goto nothing_to_declare;
364
365 for (; rt; rt = next) {
366 next = rcu_dereference_raw(rt->dn_next);
367 RCU_INIT_POINTER(rt->dn_next, NULL);
368 dst_dev_put(&rt->dst);
369 dst_release(&rt->dst);
370 }
371
372nothing_to_declare:
373 spin_unlock_bh(&dn_rt_hash_table[i].lock);
374 }
375}
376
377static DEFINE_SPINLOCK(dn_rt_flush_lock);
378
379void dn_rt_cache_flush(int delay)
380{
381 unsigned long now = jiffies;
382 int user_mode = !in_interrupt();
383
384 if (delay < 0)
385 delay = dn_rt_min_delay;
386
387 spin_lock_bh(&dn_rt_flush_lock);
388
389 if (del_timer(&dn_rt_flush_timer) && delay > 0 && dn_rt_deadline) {
390 long tmo = (long)(dn_rt_deadline - now);
391
392 if (user_mode && tmo < dn_rt_max_delay - dn_rt_min_delay)
393 tmo = 0;
394
395 if (delay > tmo)
396 delay = tmo;
397 }
398
399 if (delay <= 0) {
400 spin_unlock_bh(&dn_rt_flush_lock);
401 dn_run_flush(NULL);
402 return;
403 }
404
405 if (dn_rt_deadline == 0)
406 dn_rt_deadline = now + dn_rt_max_delay;
407
408 dn_rt_flush_timer.expires = now + delay;
409 add_timer(&dn_rt_flush_timer);
410 spin_unlock_bh(&dn_rt_flush_lock);
411}
412
413/**
414 * dn_return_short - Return a short packet to its sender
415 * @skb: The packet to return
416 *
417 */
418static int dn_return_short(struct sk_buff *skb)
419{
420 struct dn_skb_cb *cb;
421 unsigned char *ptr;
422 __le16 *src;
423 __le16 *dst;
424
425 /* Add back headers */
426 skb_push(skb, skb->data - skb_network_header(skb));
427
428 skb = skb_unshare(skb, GFP_ATOMIC);
429 if (!skb)
430 return NET_RX_DROP;
431
432 cb = DN_SKB_CB(skb);
433 /* Skip packet length and point to flags */
434 ptr = skb->data + 2;
435 *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
436
437 dst = (__le16 *)ptr;
438 ptr += 2;
439 src = (__le16 *)ptr;
440 ptr += 2;
441 *ptr = 0; /* Zero hop count */
442
443 swap(*src, *dst);
444
445 skb->pkt_type = PACKET_OUTGOING;
446 dn_rt_finish_output(skb, NULL, NULL);
447 return NET_RX_SUCCESS;
448}
449
450/**
451 * dn_return_long - Return a long packet to its sender
452 * @skb: The long format packet to return
453 *
454 */
455static int dn_return_long(struct sk_buff *skb)
456{
457 struct dn_skb_cb *cb;
458 unsigned char *ptr;
459 unsigned char *src_addr, *dst_addr;
460 unsigned char tmp[ETH_ALEN];
461
462 /* Add back all headers */
463 skb_push(skb, skb->data - skb_network_header(skb));
464
465 skb = skb_unshare(skb, GFP_ATOMIC);
466 if (!skb)
467 return NET_RX_DROP;
468
469 cb = DN_SKB_CB(skb);
470 /* Ignore packet length and point to flags */
471 ptr = skb->data + 2;
472
473 /* Skip padding */
474 if (*ptr & DN_RT_F_PF) {
475 char padlen = (*ptr & ~DN_RT_F_PF);
476 ptr += padlen;
477 }
478
479 *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
480 ptr += 2;
481 dst_addr = ptr;
482 ptr += 8;
483 src_addr = ptr;
484 ptr += 6;
485 *ptr = 0; /* Zero hop count */
486
487 /* Swap source and destination */
488 memcpy(tmp, src_addr, ETH_ALEN);
489 memcpy(src_addr, dst_addr, ETH_ALEN);
490 memcpy(dst_addr, tmp, ETH_ALEN);
491
492 skb->pkt_type = PACKET_OUTGOING;
493 dn_rt_finish_output(skb, dst_addr, src_addr);
494 return NET_RX_SUCCESS;
495}
496
497/**
498 * dn_route_rx_packet - Try and find a route for an incoming packet
499 * @net: The applicable net namespace
500 * @sk: Socket packet transmitted on
501 * @skb: The packet to find a route for
502 *
503 * Returns: result of input function if route is found, error code otherwise
504 */
505static int dn_route_rx_packet(struct net *net, struct sock *sk, struct sk_buff *skb)
506{
507 struct dn_skb_cb *cb;
508 int err;
509
510 err = dn_route_input(skb);
511 if (err == 0)
512 return dst_input(skb);
513
514 cb = DN_SKB_CB(skb);
515 if (decnet_debug_level & 4) {
516 char *devname = skb->dev ? skb->dev->name : "???";
517
518 printk(KERN_DEBUG
519 "DECnet: dn_route_rx_packet: rt_flags=0x%02x dev=%s len=%d src=0x%04hx dst=0x%04hx err=%d type=%d\n",
520 (int)cb->rt_flags, devname, skb->len,
521 le16_to_cpu(cb->src), le16_to_cpu(cb->dst),
522 err, skb->pkt_type);
523 }
524
525 if ((skb->pkt_type == PACKET_HOST) && (cb->rt_flags & DN_RT_F_RQR)) {
526 switch (cb->rt_flags & DN_RT_PKT_MSK) {
527 case DN_RT_PKT_SHORT:
528 return dn_return_short(skb);
529 case DN_RT_PKT_LONG:
530 return dn_return_long(skb);
531 }
532 }
533
534 kfree_skb(skb);
535 return NET_RX_DROP;
536}
537
538static int dn_route_rx_long(struct sk_buff *skb)
539{
540 struct dn_skb_cb *cb = DN_SKB_CB(skb);
541 unsigned char *ptr = skb->data;
542
543 if (!pskb_may_pull(skb, 21)) /* 20 for long header, 1 for shortest nsp */
544 goto drop_it;
545
546 skb_pull(skb, 20);
547 skb_reset_transport_header(skb);
548
549 /* Destination info */
550 ptr += 2;
551 cb->dst = dn_eth2dn(ptr);
552 if (memcmp(ptr, dn_hiord_addr, 4) != 0)
553 goto drop_it;
554 ptr += 6;
555
556
557 /* Source info */
558 ptr += 2;
559 cb->src = dn_eth2dn(ptr);
560 if (memcmp(ptr, dn_hiord_addr, 4) != 0)
561 goto drop_it;
562 ptr += 6;
563 /* Other junk */
564 ptr++;
565 cb->hops = *ptr++; /* Visit Count */
566
567 return NF_HOOK(NFPROTO_DECNET, NF_DN_PRE_ROUTING,
568 &init_net, NULL, skb, skb->dev, NULL,
569 dn_route_rx_packet);
570
571drop_it:
572 kfree_skb(skb);
573 return NET_RX_DROP;
574}
575
576
577
578static int dn_route_rx_short(struct sk_buff *skb)
579{
580 struct dn_skb_cb *cb = DN_SKB_CB(skb);
581 unsigned char *ptr = skb->data;
582
583 if (!pskb_may_pull(skb, 6)) /* 5 for short header + 1 for shortest nsp */
584 goto drop_it;
585
586 skb_pull(skb, 5);
587 skb_reset_transport_header(skb);
588
589 cb->dst = *(__le16 *)ptr;
590 ptr += 2;
591 cb->src = *(__le16 *)ptr;
592 ptr += 2;
593 cb->hops = *ptr & 0x3f;
594
595 return NF_HOOK(NFPROTO_DECNET, NF_DN_PRE_ROUTING,
596 &init_net, NULL, skb, skb->dev, NULL,
597 dn_route_rx_packet);
598
599drop_it:
600 kfree_skb(skb);
601 return NET_RX_DROP;
602}
603
604static int dn_route_discard(struct net *net, struct sock *sk, struct sk_buff *skb)
605{
606 /*
607 * I know we drop the packet here, but that's considered success in
608 * this case
609 */
610 kfree_skb(skb);
611 return NET_RX_SUCCESS;
612}
613
614static int dn_route_ptp_hello(struct net *net, struct sock *sk, struct sk_buff *skb)
615{
616 dn_dev_hello(skb);
617 dn_neigh_pointopoint_hello(skb);
618 return NET_RX_SUCCESS;
619}
620
621int dn_route_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
622{
623 struct dn_skb_cb *cb;
624 unsigned char flags = 0;
625 __u16 len = le16_to_cpu(*(__le16 *)skb->data);
626 struct dn_dev *dn = rcu_dereference(dev->dn_ptr);
627 unsigned char padlen = 0;
628
629 if (!net_eq(dev_net(dev), &init_net))
630 goto dump_it;
631
632 if (dn == NULL)
633 goto dump_it;
634
635 skb = skb_share_check(skb, GFP_ATOMIC);
636 if (!skb)
637 goto out;
638
639 if (!pskb_may_pull(skb, 3))
640 goto dump_it;
641
642 skb_pull(skb, 2);
643
644 if (len > skb->len)
645 goto dump_it;
646
647 skb_trim(skb, len);
648
649 flags = *skb->data;
650
651 cb = DN_SKB_CB(skb);
652 cb->stamp = jiffies;
653 cb->iif = dev->ifindex;
654
655 /*
656 * If we have padding, remove it.
657 */
658 if (flags & DN_RT_F_PF) {
659 padlen = flags & ~DN_RT_F_PF;
660 if (!pskb_may_pull(skb, padlen + 1))
661 goto dump_it;
662 skb_pull(skb, padlen);
663 flags = *skb->data;
664 }
665
666 skb_reset_network_header(skb);
667
668 /*
669 * Weed out future version DECnet
670 */
671 if (flags & DN_RT_F_VER)
672 goto dump_it;
673
674 cb->rt_flags = flags;
675
676 if (decnet_debug_level & 1)
677 printk(KERN_DEBUG
678 "dn_route_rcv: got 0x%02x from %s [%d %d %d]\n",
679 (int)flags, dev->name, len, skb->len,
680 padlen);
681
682 if (flags & DN_RT_PKT_CNTL) {
683 if (unlikely(skb_linearize(skb)))
684 goto dump_it;
685
686 switch (flags & DN_RT_CNTL_MSK) {
687 case DN_RT_PKT_INIT:
688 dn_dev_init_pkt(skb);
689 break;
690 case DN_RT_PKT_VERI:
691 dn_dev_veri_pkt(skb);
692 break;
693 }
694
695 if (dn->parms.state != DN_DEV_S_RU)
696 goto dump_it;
697
698 switch (flags & DN_RT_CNTL_MSK) {
699 case DN_RT_PKT_HELO:
700 return NF_HOOK(NFPROTO_DECNET, NF_DN_HELLO,
701 &init_net, NULL, skb, skb->dev, NULL,
702 dn_route_ptp_hello);
703
704 case DN_RT_PKT_L1RT:
705 case DN_RT_PKT_L2RT:
706 return NF_HOOK(NFPROTO_DECNET, NF_DN_ROUTE,
707 &init_net, NULL, skb, skb->dev, NULL,
708 dn_route_discard);
709 case DN_RT_PKT_ERTH:
710 return NF_HOOK(NFPROTO_DECNET, NF_DN_HELLO,
711 &init_net, NULL, skb, skb->dev, NULL,
712 dn_neigh_router_hello);
713
714 case DN_RT_PKT_EEDH:
715 return NF_HOOK(NFPROTO_DECNET, NF_DN_HELLO,
716 &init_net, NULL, skb, skb->dev, NULL,
717 dn_neigh_endnode_hello);
718 }
719 } else {
720 if (dn->parms.state != DN_DEV_S_RU)
721 goto dump_it;
722
723 skb_pull(skb, 1); /* Pull flags */
724
725 switch (flags & DN_RT_PKT_MSK) {
726 case DN_RT_PKT_LONG:
727 return dn_route_rx_long(skb);
728 case DN_RT_PKT_SHORT:
729 return dn_route_rx_short(skb);
730 }
731 }
732
733dump_it:
734 kfree_skb(skb);
735out:
736 return NET_RX_DROP;
737}
738
739static int dn_output(struct net *net, struct sock *sk, struct sk_buff *skb)
740{
741 struct dst_entry *dst = skb_dst(skb);
742 struct dn_route *rt = (struct dn_route *)dst;
743 struct net_device *dev = dst->dev;
744 struct dn_skb_cb *cb = DN_SKB_CB(skb);
745
746 int err = -EINVAL;
747
748 if (rt->n == NULL)
749 goto error;
750
751 skb->dev = dev;
752
753 cb->src = rt->rt_saddr;
754 cb->dst = rt->rt_daddr;
755
756 /*
757 * Always set the Intra-Ethernet bit on all outgoing packets
758 * originated on this node. Only valid flag from upper layers
759 * is return-to-sender-requested. Set hop count to 0 too.
760 */
761 cb->rt_flags &= ~DN_RT_F_RQR;
762 cb->rt_flags |= DN_RT_F_IE;
763 cb->hops = 0;
764
765 return NF_HOOK(NFPROTO_DECNET, NF_DN_LOCAL_OUT,
766 &init_net, sk, skb, NULL, dev,
767 dn_to_neigh_output);
768
769error:
770 net_dbg_ratelimited("dn_output: This should not happen\n");
771
772 kfree_skb(skb);
773
774 return err;
775}
776
777static int dn_forward(struct sk_buff *skb)
778{
779 struct dn_skb_cb *cb = DN_SKB_CB(skb);
780 struct dst_entry *dst = skb_dst(skb);
781 struct dn_dev *dn_db = rcu_dereference(dst->dev->dn_ptr);
782 struct dn_route *rt;
783 int header_len;
784 struct net_device *dev = skb->dev;
785
786 if (skb->pkt_type != PACKET_HOST)
787 goto drop;
788
789 /* Ensure that we have enough space for headers */
790 rt = (struct dn_route *)skb_dst(skb);
791 header_len = dn_db->use_long ? 21 : 6;
792 if (skb_cow(skb, LL_RESERVED_SPACE(rt->dst.dev)+header_len))
793 goto drop;
794
795 /*
796 * Hop count exceeded.
797 */
798 if (++cb->hops > 30)
799 goto drop;
800
801 skb->dev = rt->dst.dev;
802
803 /*
804 * If packet goes out same interface it came in on, then set
805 * the Intra-Ethernet bit. This has no effect for short
806 * packets, so we don't need to test for them here.
807 */
808 cb->rt_flags &= ~DN_RT_F_IE;
809 if (rt->rt_flags & RTCF_DOREDIRECT)
810 cb->rt_flags |= DN_RT_F_IE;
811
812 return NF_HOOK(NFPROTO_DECNET, NF_DN_FORWARD,
813 &init_net, NULL, skb, dev, skb->dev,
814 dn_to_neigh_output);
815
816drop:
817 kfree_skb(skb);
818 return NET_RX_DROP;
819}
820
821/*
822 * Used to catch bugs. This should never normally get
823 * called.
824 */
825static int dn_rt_bug_out(struct net *net, struct sock *sk, struct sk_buff *skb)
826{
827 struct dn_skb_cb *cb = DN_SKB_CB(skb);
828
829 net_dbg_ratelimited("dn_rt_bug: skb from:%04x to:%04x\n",
830 le16_to_cpu(cb->src), le16_to_cpu(cb->dst));
831
832 kfree_skb(skb);
833
834 return NET_RX_DROP;
835}
836
837static int dn_rt_bug(struct sk_buff *skb)
838{
839 struct dn_skb_cb *cb = DN_SKB_CB(skb);
840
841 net_dbg_ratelimited("dn_rt_bug: skb from:%04x to:%04x\n",
842 le16_to_cpu(cb->src), le16_to_cpu(cb->dst));
843
844 kfree_skb(skb);
845
846 return NET_RX_DROP;
847}
848
849static unsigned int dn_dst_default_advmss(const struct dst_entry *dst)
850{
851 return dn_mss_from_pmtu(dst->dev, dst_mtu(dst));
852}
853
854static unsigned int dn_dst_mtu(const struct dst_entry *dst)
855{
856 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
857
858 return mtu ? : dst->dev->mtu;
859}
860
861static struct neighbour *dn_dst_neigh_lookup(const struct dst_entry *dst,
862 struct sk_buff *skb,
863 const void *daddr)
864{
865 return __neigh_lookup_errno(&dn_neigh_table, daddr, dst->dev);
866}
867
868static int dn_rt_set_next_hop(struct dn_route *rt, struct dn_fib_res *res)
869{
870 struct dn_fib_info *fi = res->fi;
871 struct net_device *dev = rt->dst.dev;
872 unsigned int mss_metric;
873 struct neighbour *n;
874
875 if (fi) {
876 if (DN_FIB_RES_GW(*res) &&
877 DN_FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
878 rt->rt_gateway = DN_FIB_RES_GW(*res);
879 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
880 }
881 rt->rt_type = res->type;
882
883 if (dev != NULL && rt->n == NULL) {
884 n = __neigh_lookup_errno(&dn_neigh_table, &rt->rt_gateway, dev);
885 if (IS_ERR(n))
886 return PTR_ERR(n);
887 rt->n = n;
888 }
889
890 if (dst_metric(&rt->dst, RTAX_MTU) > rt->dst.dev->mtu)
891 dst_metric_set(&rt->dst, RTAX_MTU, rt->dst.dev->mtu);
892 mss_metric = dst_metric_raw(&rt->dst, RTAX_ADVMSS);
893 if (mss_metric) {
894 unsigned int mss = dn_mss_from_pmtu(dev, dst_mtu(&rt->dst));
895 if (mss_metric > mss)
896 dst_metric_set(&rt->dst, RTAX_ADVMSS, mss);
897 }
898 return 0;
899}
900
901static inline int dn_match_addr(__le16 addr1, __le16 addr2)
902{
903 __u16 tmp = le16_to_cpu(addr1) ^ le16_to_cpu(addr2);
904 int match = 16;
905 while (tmp) {
906 tmp >>= 1;
907 match--;
908 }
909 return match;
910}
911
912static __le16 dnet_select_source(const struct net_device *dev, __le16 daddr, int scope)
913{
914 __le16 saddr = 0;
915 struct dn_dev *dn_db;
916 struct dn_ifaddr *ifa;
917 int best_match = 0;
918 int ret;
919
920 rcu_read_lock();
921 dn_db = rcu_dereference(dev->dn_ptr);
922 for (ifa = rcu_dereference(dn_db->ifa_list);
923 ifa != NULL;
924 ifa = rcu_dereference(ifa->ifa_next)) {
925 if (ifa->ifa_scope > scope)
926 continue;
927 if (!daddr) {
928 saddr = ifa->ifa_local;
929 break;
930 }
931 ret = dn_match_addr(daddr, ifa->ifa_local);
932 if (ret > best_match)
933 saddr = ifa->ifa_local;
934 if (best_match == 0)
935 saddr = ifa->ifa_local;
936 }
937 rcu_read_unlock();
938
939 return saddr;
940}
941
942static inline __le16 __dn_fib_res_prefsrc(struct dn_fib_res *res)
943{
944 return dnet_select_source(DN_FIB_RES_DEV(*res), DN_FIB_RES_GW(*res), res->scope);
945}
946
947static inline __le16 dn_fib_rules_map_destination(__le16 daddr, struct dn_fib_res *res)
948{
949 __le16 mask = dnet_make_mask(res->prefixlen);
950 return (daddr&~mask)|res->fi->fib_nh->nh_gw;
951}
952
953static int dn_route_output_slow(struct dst_entry **pprt, const struct flowidn *oldflp, int try_hard)
954{
955 struct flowidn fld = {
956 .daddr = oldflp->daddr,
957 .saddr = oldflp->saddr,
958 .flowidn_scope = RT_SCOPE_UNIVERSE,
959 .flowidn_mark = oldflp->flowidn_mark,
960 .flowidn_iif = LOOPBACK_IFINDEX,
961 .flowidn_oif = oldflp->flowidn_oif,
962 };
963 struct dn_route *rt = NULL;
964 struct net_device *dev_out = NULL, *dev;
965 struct neighbour *neigh = NULL;
966 unsigned int hash;
967 unsigned int flags = 0;
968 struct dn_fib_res res = { .fi = NULL, .type = RTN_UNICAST };
969 int err;
970 int free_res = 0;
971 __le16 gateway = 0;
972
973 if (decnet_debug_level & 16)
974 printk(KERN_DEBUG
975 "dn_route_output_slow: dst=%04x src=%04x mark=%d"
976 " iif=%d oif=%d\n", le16_to_cpu(oldflp->daddr),
977 le16_to_cpu(oldflp->saddr),
978 oldflp->flowidn_mark, LOOPBACK_IFINDEX,
979 oldflp->flowidn_oif);
980
981 /* If we have an output interface, verify its a DECnet device */
982 if (oldflp->flowidn_oif) {
983 dev_out = dev_get_by_index(&init_net, oldflp->flowidn_oif);
984 err = -ENODEV;
985 if (dev_out && dev_out->dn_ptr == NULL) {
986 dev_put(dev_out);
987 dev_out = NULL;
988 }
989 if (dev_out == NULL)
990 goto out;
991 }
992
993 /* If we have a source address, verify that its a local address */
994 if (oldflp->saddr) {
995 err = -EADDRNOTAVAIL;
996
997 if (dev_out) {
998 if (dn_dev_islocal(dev_out, oldflp->saddr))
999 goto source_ok;
1000 dev_put(dev_out);
1001 goto out;
1002 }
1003 rcu_read_lock();
1004 for_each_netdev_rcu(&init_net, dev) {
1005 if (!dev->dn_ptr)
1006 continue;
1007 if (!dn_dev_islocal(dev, oldflp->saddr))
1008 continue;
1009 if ((dev->flags & IFF_LOOPBACK) &&
1010 oldflp->daddr &&
1011 !dn_dev_islocal(dev, oldflp->daddr))
1012 continue;
1013
1014 dev_out = dev;
1015 break;
1016 }
1017 rcu_read_unlock();
1018 if (dev_out == NULL)
1019 goto out;
1020 dev_hold(dev_out);
1021source_ok:
1022 ;
1023 }
1024
1025 /* No destination? Assume its local */
1026 if (!fld.daddr) {
1027 fld.daddr = fld.saddr;
1028
1029 if (dev_out)
1030 dev_put(dev_out);
1031 err = -EINVAL;
1032 dev_out = init_net.loopback_dev;
1033 if (!dev_out->dn_ptr)
1034 goto out;
1035 err = -EADDRNOTAVAIL;
1036 dev_hold(dev_out);
1037 if (!fld.daddr) {
1038 fld.daddr =
1039 fld.saddr = dnet_select_source(dev_out, 0,
1040 RT_SCOPE_HOST);
1041 if (!fld.daddr)
1042 goto done;
1043 }
1044 fld.flowidn_oif = LOOPBACK_IFINDEX;
1045 res.type = RTN_LOCAL;
1046 goto make_route;
1047 }
1048
1049 if (decnet_debug_level & 16)
1050 printk(KERN_DEBUG
1051 "dn_route_output_slow: initial checks complete."
1052 " dst=%04x src=%04x oif=%d try_hard=%d\n",
1053 le16_to_cpu(fld.daddr), le16_to_cpu(fld.saddr),
1054 fld.flowidn_oif, try_hard);
1055
1056 /*
1057 * N.B. If the kernel is compiled without router support then
1058 * dn_fib_lookup() will evaluate to non-zero so this if () block
1059 * will always be executed.
1060 */
1061 err = -ESRCH;
1062 if (try_hard || (err = dn_fib_lookup(&fld, &res)) != 0) {
1063 struct dn_dev *dn_db;
1064 if (err != -ESRCH)
1065 goto out;
1066 /*
1067 * Here the fallback is basically the standard algorithm for
1068 * routing in endnodes which is described in the DECnet routing
1069 * docs
1070 *
1071 * If we are not trying hard, look in neighbour cache.
1072 * The result is tested to ensure that if a specific output
1073 * device/source address was requested, then we honour that
1074 * here
1075 */
1076 if (!try_hard) {
1077 neigh = neigh_lookup_nodev(&dn_neigh_table, &init_net, &fld.daddr);
1078 if (neigh) {
1079 if ((oldflp->flowidn_oif &&
1080 (neigh->dev->ifindex != oldflp->flowidn_oif)) ||
1081 (oldflp->saddr &&
1082 (!dn_dev_islocal(neigh->dev,
1083 oldflp->saddr)))) {
1084 neigh_release(neigh);
1085 neigh = NULL;
1086 } else {
1087 if (dev_out)
1088 dev_put(dev_out);
1089 if (dn_dev_islocal(neigh->dev, fld.daddr)) {
1090 dev_out = init_net.loopback_dev;
1091 res.type = RTN_LOCAL;
1092 } else {
1093 dev_out = neigh->dev;
1094 }
1095 dev_hold(dev_out);
1096 goto select_source;
1097 }
1098 }
1099 }
1100
1101 /* Not there? Perhaps its a local address */
1102 if (dev_out == NULL)
1103 dev_out = dn_dev_get_default();
1104 err = -ENODEV;
1105 if (dev_out == NULL)
1106 goto out;
1107 dn_db = rcu_dereference_raw(dev_out->dn_ptr);
1108 if (!dn_db)
1109 goto e_inval;
1110 /* Possible improvement - check all devices for local addr */
1111 if (dn_dev_islocal(dev_out, fld.daddr)) {
1112 dev_put(dev_out);
1113 dev_out = init_net.loopback_dev;
1114 dev_hold(dev_out);
1115 res.type = RTN_LOCAL;
1116 goto select_source;
1117 }
1118 /* Not local either.... try sending it to the default router */
1119 neigh = neigh_clone(dn_db->router);
1120 BUG_ON(neigh && neigh->dev != dev_out);
1121
1122 /* Ok then, we assume its directly connected and move on */
1123select_source:
1124 if (neigh)
1125 gateway = ((struct dn_neigh *)neigh)->addr;
1126 if (gateway == 0)
1127 gateway = fld.daddr;
1128 if (fld.saddr == 0) {
1129 fld.saddr = dnet_select_source(dev_out, gateway,
1130 res.type == RTN_LOCAL ?
1131 RT_SCOPE_HOST :
1132 RT_SCOPE_LINK);
1133 if (fld.saddr == 0 && res.type != RTN_LOCAL)
1134 goto e_addr;
1135 }
1136 fld.flowidn_oif = dev_out->ifindex;
1137 goto make_route;
1138 }
1139 free_res = 1;
1140
1141 if (res.type == RTN_NAT)
1142 goto e_inval;
1143
1144 if (res.type == RTN_LOCAL) {
1145 if (!fld.saddr)
1146 fld.saddr = fld.daddr;
1147 if (dev_out)
1148 dev_put(dev_out);
1149 dev_out = init_net.loopback_dev;
1150 dev_hold(dev_out);
1151 if (!dev_out->dn_ptr)
1152 goto e_inval;
1153 fld.flowidn_oif = dev_out->ifindex;
1154 if (res.fi)
1155 dn_fib_info_put(res.fi);
1156 res.fi = NULL;
1157 goto make_route;
1158 }
1159
1160 if (res.fi->fib_nhs > 1 && fld.flowidn_oif == 0)
1161 dn_fib_select_multipath(&fld, &res);
1162
1163 /*
1164 * We could add some logic to deal with default routes here and
1165 * get rid of some of the special casing above.
1166 */
1167
1168 if (!fld.saddr)
1169 fld.saddr = DN_FIB_RES_PREFSRC(res);
1170
1171 if (dev_out)
1172 dev_put(dev_out);
1173 dev_out = DN_FIB_RES_DEV(res);
1174 dev_hold(dev_out);
1175 fld.flowidn_oif = dev_out->ifindex;
1176 gateway = DN_FIB_RES_GW(res);
1177
1178make_route:
1179 if (dev_out->flags & IFF_LOOPBACK)
1180 flags |= RTCF_LOCAL;
1181
1182 rt = dst_alloc(&dn_dst_ops, dev_out, 0, DST_OBSOLETE_NONE, 0);
1183 if (rt == NULL)
1184 goto e_nobufs;
1185
1186 rt->dn_next = NULL;
1187 memset(&rt->fld, 0, sizeof(rt->fld));
1188 rt->fld.saddr = oldflp->saddr;
1189 rt->fld.daddr = oldflp->daddr;
1190 rt->fld.flowidn_oif = oldflp->flowidn_oif;
1191 rt->fld.flowidn_iif = 0;
1192 rt->fld.flowidn_mark = oldflp->flowidn_mark;
1193
1194 rt->rt_saddr = fld.saddr;
1195 rt->rt_daddr = fld.daddr;
1196 rt->rt_gateway = gateway ? gateway : fld.daddr;
1197 rt->rt_local_src = fld.saddr;
1198
1199 rt->rt_dst_map = fld.daddr;
1200 rt->rt_src_map = fld.saddr;
1201
1202 rt->n = neigh;
1203 neigh = NULL;
1204
1205 rt->dst.lastuse = jiffies;
1206 rt->dst.output = dn_output;
1207 rt->dst.input = dn_rt_bug;
1208 rt->rt_flags = flags;
1209 if (flags & RTCF_LOCAL)
1210 rt->dst.input = dn_nsp_rx;
1211
1212 err = dn_rt_set_next_hop(rt, &res);
1213 if (err)
1214 goto e_neighbour;
1215
1216 hash = dn_hash(rt->fld.saddr, rt->fld.daddr);
1217 /* dn_insert_route() increments dst->__refcnt */
1218 dn_insert_route(rt, hash, (struct dn_route **)pprt);
1219
1220done:
1221 if (neigh)
1222 neigh_release(neigh);
1223 if (free_res)
1224 dn_fib_res_put(&res);
1225 if (dev_out)
1226 dev_put(dev_out);
1227out:
1228 return err;
1229
1230e_addr:
1231 err = -EADDRNOTAVAIL;
1232 goto done;
1233e_inval:
1234 err = -EINVAL;
1235 goto done;
1236e_nobufs:
1237 err = -ENOBUFS;
1238 goto done;
1239e_neighbour:
1240 dst_release_immediate(&rt->dst);
1241 goto e_nobufs;
1242}
1243
1244
1245/*
1246 * N.B. The flags may be moved into the flowi at some future stage.
1247 */
1248static int __dn_route_output_key(struct dst_entry **pprt, const struct flowidn *flp, int flags)
1249{
1250 unsigned int hash = dn_hash(flp->saddr, flp->daddr);
1251 struct dn_route *rt = NULL;
1252
1253 if (!(flags & MSG_TRYHARD)) {
1254 rcu_read_lock_bh();
1255 for (rt = rcu_dereference_bh(dn_rt_hash_table[hash].chain); rt;
1256 rt = rcu_dereference_bh(rt->dn_next)) {
1257 if ((flp->daddr == rt->fld.daddr) &&
1258 (flp->saddr == rt->fld.saddr) &&
1259 (flp->flowidn_mark == rt->fld.flowidn_mark) &&
1260 dn_is_output_route(rt) &&
1261 (rt->fld.flowidn_oif == flp->flowidn_oif)) {
1262 dst_hold_and_use(&rt->dst, jiffies);
1263 rcu_read_unlock_bh();
1264 *pprt = &rt->dst;
1265 return 0;
1266 }
1267 }
1268 rcu_read_unlock_bh();
1269 }
1270
1271 return dn_route_output_slow(pprt, flp, flags);
1272}
1273
1274static int dn_route_output_key(struct dst_entry **pprt, struct flowidn *flp, int flags)
1275{
1276 int err;
1277
1278 err = __dn_route_output_key(pprt, flp, flags);
1279 if (err == 0 && flp->flowidn_proto) {
1280 *pprt = xfrm_lookup(&init_net, *pprt,
1281 flowidn_to_flowi(flp), NULL, 0);
1282 if (IS_ERR(*pprt)) {
1283 err = PTR_ERR(*pprt);
1284 *pprt = NULL;
1285 }
1286 }
1287 return err;
1288}
1289
1290int dn_route_output_sock(struct dst_entry __rcu **pprt, struct flowidn *fl, struct sock *sk, int flags)
1291{
1292 int err;
1293
1294 err = __dn_route_output_key(pprt, fl, flags & MSG_TRYHARD);
1295 if (err == 0 && fl->flowidn_proto) {
1296 *pprt = xfrm_lookup(&init_net, *pprt,
1297 flowidn_to_flowi(fl), sk, 0);
1298 if (IS_ERR(*pprt)) {
1299 err = PTR_ERR(*pprt);
1300 *pprt = NULL;
1301 }
1302 }
1303 return err;
1304}
1305
1306static int dn_route_input_slow(struct sk_buff *skb)
1307{
1308 struct dn_route *rt = NULL;
1309 struct dn_skb_cb *cb = DN_SKB_CB(skb);
1310 struct net_device *in_dev = skb->dev;
1311 struct net_device *out_dev = NULL;
1312 struct dn_dev *dn_db;
1313 struct neighbour *neigh = NULL;
1314 unsigned int hash;
1315 int flags = 0;
1316 __le16 gateway = 0;
1317 __le16 local_src = 0;
1318 struct flowidn fld = {
1319 .daddr = cb->dst,
1320 .saddr = cb->src,
1321 .flowidn_scope = RT_SCOPE_UNIVERSE,
1322 .flowidn_mark = skb->mark,
1323 .flowidn_iif = skb->dev->ifindex,
1324 };
1325 struct dn_fib_res res = { .fi = NULL, .type = RTN_UNREACHABLE };
1326 int err = -EINVAL;
1327 int free_res = 0;
1328
1329 dev_hold(in_dev);
1330
1331 dn_db = rcu_dereference(in_dev->dn_ptr);
1332 if (!dn_db)
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 break;
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, 1, DST_OBSOLETE_NONE, 0);
1450 if (rt == NULL)
1451 goto e_nobufs;
1452
1453 rt->dn_next = NULL;
1454 memset(&rt->fld, 0, sizeof(rt->fld));
1455 rt->rt_saddr = fld.saddr;
1456 rt->rt_daddr = fld.daddr;
1457 rt->rt_gateway = fld.daddr;
1458 if (gateway)
1459 rt->rt_gateway = gateway;
1460 rt->rt_local_src = local_src ? local_src : rt->rt_saddr;
1461
1462 rt->rt_dst_map = fld.daddr;
1463 rt->rt_src_map = fld.saddr;
1464
1465 rt->fld.saddr = cb->src;
1466 rt->fld.daddr = cb->dst;
1467 rt->fld.flowidn_oif = 0;
1468 rt->fld.flowidn_iif = in_dev->ifindex;
1469 rt->fld.flowidn_mark = fld.flowidn_mark;
1470
1471 rt->n = neigh;
1472 rt->dst.lastuse = jiffies;
1473 rt->dst.output = dn_rt_bug_out;
1474 switch (res.type) {
1475 case RTN_UNICAST:
1476 rt->dst.input = dn_forward;
1477 break;
1478 case RTN_LOCAL:
1479 rt->dst.output = dn_output;
1480 rt->dst.input = dn_nsp_rx;
1481 rt->dst.dev = in_dev;
1482 flags |= RTCF_LOCAL;
1483 break;
1484 default:
1485 case RTN_UNREACHABLE:
1486 case RTN_BLACKHOLE:
1487 rt->dst.input = dst_discard;
1488 }
1489 rt->rt_flags = flags;
1490
1491 err = dn_rt_set_next_hop(rt, &res);
1492 if (err)
1493 goto e_neighbour;
1494
1495 hash = dn_hash(rt->fld.saddr, rt->fld.daddr);
1496 /* dn_insert_route() increments dst->__refcnt */
1497 dn_insert_route(rt, hash, &rt);
1498 skb_dst_set(skb, &rt->dst);
1499
1500done:
1501 if (neigh)
1502 neigh_release(neigh);
1503 if (free_res)
1504 dn_fib_res_put(&res);
1505 dev_put(in_dev);
1506 if (out_dev)
1507 dev_put(out_dev);
1508out:
1509 return err;
1510
1511e_inval:
1512 err = -EINVAL;
1513 goto done;
1514
1515e_nobufs:
1516 err = -ENOBUFS;
1517 goto done;
1518
1519e_neighbour:
1520 dst_release_immediate(&rt->dst);
1521 goto done;
1522}
1523
1524static int dn_route_input(struct sk_buff *skb)
1525{
1526 struct dn_route *rt;
1527 struct dn_skb_cb *cb = DN_SKB_CB(skb);
1528 unsigned int hash = dn_hash(cb->src, cb->dst);
1529
1530 if (skb_dst(skb))
1531 return 0;
1532
1533 rcu_read_lock();
1534 for (rt = rcu_dereference(dn_rt_hash_table[hash].chain); rt != NULL;
1535 rt = rcu_dereference(rt->dn_next)) {
1536 if ((rt->fld.saddr == cb->src) &&
1537 (rt->fld.daddr == cb->dst) &&
1538 (rt->fld.flowidn_oif == 0) &&
1539 (rt->fld.flowidn_mark == skb->mark) &&
1540 (rt->fld.flowidn_iif == cb->iif)) {
1541 dst_hold_and_use(&rt->dst, jiffies);
1542 rcu_read_unlock();
1543 skb_dst_set(skb, (struct dst_entry *)rt);
1544 return 0;
1545 }
1546 }
1547 rcu_read_unlock();
1548
1549 return dn_route_input_slow(skb);
1550}
1551
1552static int dn_rt_fill_info(struct sk_buff *skb, u32 portid, u32 seq,
1553 int event, int nowait, unsigned int flags)
1554{
1555 struct dn_route *rt = (struct dn_route *)skb_dst(skb);
1556 struct rtmsg *r;
1557 struct nlmsghdr *nlh;
1558 long expires;
1559
1560 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*r), flags);
1561 if (!nlh)
1562 return -EMSGSIZE;
1563
1564 r = nlmsg_data(nlh);
1565 r->rtm_family = AF_DECnet;
1566 r->rtm_dst_len = 16;
1567 r->rtm_src_len = 0;
1568 r->rtm_tos = 0;
1569 r->rtm_table = RT_TABLE_MAIN;
1570 r->rtm_type = rt->rt_type;
1571 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
1572 r->rtm_scope = RT_SCOPE_UNIVERSE;
1573 r->rtm_protocol = RTPROT_UNSPEC;
1574
1575 if (rt->rt_flags & RTCF_NOTIFY)
1576 r->rtm_flags |= RTM_F_NOTIFY;
1577
1578 if (nla_put_u32(skb, RTA_TABLE, RT_TABLE_MAIN) < 0 ||
1579 nla_put_le16(skb, RTA_DST, rt->rt_daddr) < 0)
1580 goto errout;
1581
1582 if (rt->fld.saddr) {
1583 r->rtm_src_len = 16;
1584 if (nla_put_le16(skb, RTA_SRC, rt->fld.saddr) < 0)
1585 goto errout;
1586 }
1587 if (rt->dst.dev &&
1588 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex) < 0)
1589 goto errout;
1590
1591 /*
1592 * Note to self - change this if input routes reverse direction when
1593 * they deal only with inputs and not with replies like they do
1594 * currently.
1595 */
1596 if (nla_put_le16(skb, RTA_PREFSRC, rt->rt_local_src) < 0)
1597 goto errout;
1598
1599 if (rt->rt_daddr != rt->rt_gateway &&
1600 nla_put_le16(skb, RTA_GATEWAY, rt->rt_gateway) < 0)
1601 goto errout;
1602
1603 if (rtnetlink_put_metrics(skb, dst_metrics_ptr(&rt->dst)) < 0)
1604 goto errout;
1605
1606 expires = rt->dst.expires ? rt->dst.expires - jiffies : 0;
1607 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires,
1608 rt->dst.error) < 0)
1609 goto errout;
1610
1611 if (dn_is_input_route(rt) &&
1612 nla_put_u32(skb, RTA_IIF, rt->fld.flowidn_iif) < 0)
1613 goto errout;
1614
1615 nlmsg_end(skb, nlh);
1616 return 0;
1617
1618errout:
1619 nlmsg_cancel(skb, nlh);
1620 return -EMSGSIZE;
1621}
1622
1623const struct nla_policy rtm_dn_policy[RTA_MAX + 1] = {
1624 [RTA_DST] = { .type = NLA_U16 },
1625 [RTA_SRC] = { .type = NLA_U16 },
1626 [RTA_IIF] = { .type = NLA_U32 },
1627 [RTA_OIF] = { .type = NLA_U32 },
1628 [RTA_GATEWAY] = { .type = NLA_U16 },
1629 [RTA_PRIORITY] = { .type = NLA_U32 },
1630 [RTA_PREFSRC] = { .type = NLA_U16 },
1631 [RTA_METRICS] = { .type = NLA_NESTED },
1632 [RTA_MULTIPATH] = { .type = NLA_NESTED },
1633 [RTA_TABLE] = { .type = NLA_U32 },
1634 [RTA_MARK] = { .type = NLA_U32 },
1635};
1636
1637/*
1638 * This is called by both endnodes and routers now.
1639 */
1640static int dn_cache_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
1641 struct netlink_ext_ack *extack)
1642{
1643 struct net *net = sock_net(in_skb->sk);
1644 struct rtmsg *rtm = nlmsg_data(nlh);
1645 struct dn_route *rt = NULL;
1646 struct dn_skb_cb *cb;
1647 int err;
1648 struct sk_buff *skb;
1649 struct flowidn fld;
1650 struct nlattr *tb[RTA_MAX+1];
1651
1652 if (!net_eq(net, &init_net))
1653 return -EINVAL;
1654
1655 err = nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
1656 rtm_dn_policy, extack);
1657 if (err < 0)
1658 return err;
1659
1660 memset(&fld, 0, sizeof(fld));
1661 fld.flowidn_proto = DNPROTO_NSP;
1662
1663 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1664 if (skb == NULL)
1665 return -ENOBUFS;
1666 skb_reset_mac_header(skb);
1667 cb = DN_SKB_CB(skb);
1668
1669 if (tb[RTA_SRC])
1670 fld.saddr = nla_get_le16(tb[RTA_SRC]);
1671
1672 if (tb[RTA_DST])
1673 fld.daddr = nla_get_le16(tb[RTA_DST]);
1674
1675 if (tb[RTA_IIF])
1676 fld.flowidn_iif = nla_get_u32(tb[RTA_IIF]);
1677
1678 if (fld.flowidn_iif) {
1679 struct net_device *dev;
1680 dev = __dev_get_by_index(&init_net, fld.flowidn_iif);
1681 if (!dev || !dev->dn_ptr) {
1682 kfree_skb(skb);
1683 return -ENODEV;
1684 }
1685 skb->protocol = htons(ETH_P_DNA_RT);
1686 skb->dev = dev;
1687 cb->src = fld.saddr;
1688 cb->dst = fld.daddr;
1689 local_bh_disable();
1690 err = dn_route_input(skb);
1691 local_bh_enable();
1692 memset(cb, 0, sizeof(struct dn_skb_cb));
1693 rt = (struct dn_route *)skb_dst(skb);
1694 if (!err && -rt->dst.error)
1695 err = rt->dst.error;
1696 } else {
1697 if (tb[RTA_OIF])
1698 fld.flowidn_oif = nla_get_u32(tb[RTA_OIF]);
1699
1700 err = dn_route_output_key((struct dst_entry **)&rt, &fld, 0);
1701 }
1702
1703 skb->dev = NULL;
1704 if (err)
1705 goto out_free;
1706 skb_dst_set(skb, &rt->dst);
1707 if (rtm->rtm_flags & RTM_F_NOTIFY)
1708 rt->rt_flags |= RTCF_NOTIFY;
1709
1710 err = dn_rt_fill_info(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq, RTM_NEWROUTE, 0, 0);
1711 if (err < 0) {
1712 err = -EMSGSIZE;
1713 goto out_free;
1714 }
1715
1716 return rtnl_unicast(skb, &init_net, NETLINK_CB(in_skb).portid);
1717
1718out_free:
1719 kfree_skb(skb);
1720 return err;
1721}
1722
1723/*
1724 * For routers, this is called from dn_fib_dump, but for endnodes its
1725 * called directly from the rtnetlink dispatch table.
1726 */
1727int dn_cache_dump(struct sk_buff *skb, struct netlink_callback *cb)
1728{
1729 struct net *net = sock_net(skb->sk);
1730 struct dn_route *rt;
1731 int h, s_h;
1732 int idx, s_idx;
1733 struct rtmsg *rtm;
1734
1735 if (!net_eq(net, &init_net))
1736 return 0;
1737
1738 if (nlmsg_len(cb->nlh) < sizeof(struct rtmsg))
1739 return -EINVAL;
1740
1741 rtm = nlmsg_data(cb->nlh);
1742 if (!(rtm->rtm_flags & RTM_F_CLONED))
1743 return 0;
1744
1745 s_h = cb->args[0];
1746 s_idx = idx = cb->args[1];
1747 for (h = 0; h <= dn_rt_hash_mask; h++) {
1748 if (h < s_h)
1749 continue;
1750 if (h > s_h)
1751 s_idx = 0;
1752 rcu_read_lock_bh();
1753 for (rt = rcu_dereference_bh(dn_rt_hash_table[h].chain), idx = 0;
1754 rt;
1755 rt = rcu_dereference_bh(rt->dn_next), idx++) {
1756 if (idx < s_idx)
1757 continue;
1758 skb_dst_set(skb, dst_clone(&rt->dst));
1759 if (dn_rt_fill_info(skb, NETLINK_CB(cb->skb).portid,
1760 cb->nlh->nlmsg_seq, RTM_NEWROUTE,
1761 1, NLM_F_MULTI) < 0) {
1762 skb_dst_drop(skb);
1763 rcu_read_unlock_bh();
1764 goto done;
1765 }
1766 skb_dst_drop(skb);
1767 }
1768 rcu_read_unlock_bh();
1769 }
1770
1771done:
1772 cb->args[0] = h;
1773 cb->args[1] = idx;
1774 return skb->len;
1775}
1776
1777#ifdef CONFIG_PROC_FS
1778struct dn_rt_cache_iter_state {
1779 int bucket;
1780};
1781
1782static struct dn_route *dn_rt_cache_get_first(struct seq_file *seq)
1783{
1784 struct dn_route *rt = NULL;
1785 struct dn_rt_cache_iter_state *s = seq->private;
1786
1787 for (s->bucket = dn_rt_hash_mask; s->bucket >= 0; --s->bucket) {
1788 rcu_read_lock_bh();
1789 rt = rcu_dereference_bh(dn_rt_hash_table[s->bucket].chain);
1790 if (rt)
1791 break;
1792 rcu_read_unlock_bh();
1793 }
1794 return rt;
1795}
1796
1797static struct dn_route *dn_rt_cache_get_next(struct seq_file *seq, struct dn_route *rt)
1798{
1799 struct dn_rt_cache_iter_state *s = seq->private;
1800
1801 rt = rcu_dereference_bh(rt->dn_next);
1802 while (!rt) {
1803 rcu_read_unlock_bh();
1804 if (--s->bucket < 0)
1805 break;
1806 rcu_read_lock_bh();
1807 rt = rcu_dereference_bh(dn_rt_hash_table[s->bucket].chain);
1808 }
1809 return rt;
1810}
1811
1812static void *dn_rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
1813{
1814 struct dn_route *rt = dn_rt_cache_get_first(seq);
1815
1816 if (rt) {
1817 while (*pos && (rt = dn_rt_cache_get_next(seq, rt)))
1818 --*pos;
1819 }
1820 return *pos ? NULL : rt;
1821}
1822
1823static void *dn_rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1824{
1825 struct dn_route *rt = dn_rt_cache_get_next(seq, v);
1826 ++*pos;
1827 return rt;
1828}
1829
1830static void dn_rt_cache_seq_stop(struct seq_file *seq, void *v)
1831{
1832 if (v)
1833 rcu_read_unlock_bh();
1834}
1835
1836static int dn_rt_cache_seq_show(struct seq_file *seq, void *v)
1837{
1838 struct dn_route *rt = v;
1839 char buf1[DN_ASCBUF_LEN], buf2[DN_ASCBUF_LEN];
1840
1841 seq_printf(seq, "%-8s %-7s %-7s %04d %04d %04d\n",
1842 rt->dst.dev ? rt->dst.dev->name : "*",
1843 dn_addr2asc(le16_to_cpu(rt->rt_daddr), buf1),
1844 dn_addr2asc(le16_to_cpu(rt->rt_saddr), buf2),
1845 atomic_read(&rt->dst.__refcnt),
1846 rt->dst.__use, 0);
1847 return 0;
1848}
1849
1850static const struct seq_operations dn_rt_cache_seq_ops = {
1851 .start = dn_rt_cache_seq_start,
1852 .next = dn_rt_cache_seq_next,
1853 .stop = dn_rt_cache_seq_stop,
1854 .show = dn_rt_cache_seq_show,
1855};
1856#endif /* CONFIG_PROC_FS */
1857
1858void __init dn_route_init(void)
1859{
1860 int i, goal, order;
1861
1862 dn_dst_ops.kmem_cachep =
1863 kmem_cache_create("dn_dst_cache", sizeof(struct dn_route), 0,
1864 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1865 dst_entries_init(&dn_dst_ops);
1866 timer_setup(&dn_route_timer, dn_dst_check_expire, 0);
1867 dn_route_timer.expires = jiffies + decnet_dst_gc_interval * HZ;
1868 add_timer(&dn_route_timer);
1869
1870 goal = totalram_pages() >> (26 - PAGE_SHIFT);
1871
1872 for (order = 0; (1UL << order) < goal; order++)
1873 /* NOTHING */;
1874
1875 /*
1876 * Only want 1024 entries max, since the table is very, very unlikely
1877 * to be larger than that.
1878 */
1879 while (order && ((((1UL << order) * PAGE_SIZE) /
1880 sizeof(struct dn_rt_hash_bucket)) >= 2048))
1881 order--;
1882
1883 do {
1884 dn_rt_hash_mask = (1UL << order) * PAGE_SIZE /
1885 sizeof(struct dn_rt_hash_bucket);
1886 while (dn_rt_hash_mask & (dn_rt_hash_mask - 1))
1887 dn_rt_hash_mask--;
1888 dn_rt_hash_table = (struct dn_rt_hash_bucket *)
1889 __get_free_pages(GFP_ATOMIC, order);
1890 } while (dn_rt_hash_table == NULL && --order > 0);
1891
1892 if (!dn_rt_hash_table)
1893 panic("Failed to allocate DECnet route cache hash table\n");
1894
1895 printk(KERN_INFO
1896 "DECnet: Routing cache hash table of %u buckets, %ldKbytes\n",
1897 dn_rt_hash_mask,
1898 (long)(dn_rt_hash_mask*sizeof(struct dn_rt_hash_bucket))/1024);
1899
1900 dn_rt_hash_mask--;
1901 for (i = 0; i <= dn_rt_hash_mask; i++) {
1902 spin_lock_init(&dn_rt_hash_table[i].lock);
1903 dn_rt_hash_table[i].chain = NULL;
1904 }
1905
1906 dn_dst_ops.gc_thresh = (dn_rt_hash_mask + 1);
1907
1908 proc_create_seq_private("decnet_cache", 0444, init_net.proc_net,
1909 &dn_rt_cache_seq_ops,
1910 sizeof(struct dn_rt_cache_iter_state), NULL);
1911
1912#ifdef CONFIG_DECNET_ROUTER
1913 rtnl_register_module(THIS_MODULE, PF_DECnet, RTM_GETROUTE,
1914 dn_cache_getroute, dn_fib_dump, 0);
1915#else
1916 rtnl_register_module(THIS_MODULE, PF_DECnet, RTM_GETROUTE,
1917 dn_cache_getroute, dn_cache_dump, 0);
1918#endif
1919}
1920
1921void __exit dn_route_cleanup(void)
1922{
1923 del_timer(&dn_route_timer);
1924 dn_run_flush(NULL);
1925
1926 remove_proc_entry("decnet_cache", init_net.proc_net);
1927 dst_entries_destroy(&dn_dst_ops);
1928}
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 .protocol = cpu_to_be16(ETH_P_DNA_RT),
140 .gc_thresh = 128,
141 .gc = dn_dst_gc,
142 .check = dn_dst_check,
143 .default_advmss = dn_dst_default_advmss,
144 .mtu = dn_dst_mtu,
145 .cow_metrics = dst_cow_metrics_generic,
146 .destroy = dn_dst_destroy,
147 .ifdown = dn_dst_ifdown,
148 .negative_advice = dn_dst_negative_advice,
149 .link_failure = dn_dst_link_failure,
150 .update_pmtu = dn_dst_update_pmtu,
151 .redirect = dn_dst_redirect,
152 .neigh_lookup = dn_dst_neigh_lookup,
153};
154
155static void dn_dst_destroy(struct dst_entry *dst)
156{
157 struct dn_route *rt = (struct dn_route *) dst;
158
159 if (rt->n)
160 neigh_release(rt->n);
161 dst_destroy_metrics_generic(dst);
162}
163
164static void dn_dst_ifdown(struct dst_entry *dst, struct net_device *dev, int how)
165{
166 if (how) {
167 struct dn_route *rt = (struct dn_route *) dst;
168 struct neighbour *n = rt->n;
169
170 if (n && n->dev == dev) {
171 n->dev = dev_net(dev)->loopback_dev;
172 dev_hold(n->dev);
173 dev_put(dev);
174 }
175 }
176}
177
178static __inline__ unsigned int dn_hash(__le16 src, __le16 dst)
179{
180 __u16 tmp = (__u16 __force)(src ^ dst);
181 tmp ^= (tmp >> 3);
182 tmp ^= (tmp >> 5);
183 tmp ^= (tmp >> 10);
184 return dn_rt_hash_mask & (unsigned int)tmp;
185}
186
187static inline void dnrt_free(struct dn_route *rt)
188{
189 call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
190}
191
192static inline void dnrt_drop(struct dn_route *rt)
193{
194 dst_release(&rt->dst);
195 call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
196}
197
198static void dn_dst_check_expire(unsigned long dummy)
199{
200 int i;
201 struct dn_route *rt;
202 struct dn_route __rcu **rtp;
203 unsigned long now = jiffies;
204 unsigned long expire = 120 * HZ;
205
206 for (i = 0; i <= dn_rt_hash_mask; i++) {
207 rtp = &dn_rt_hash_table[i].chain;
208
209 spin_lock(&dn_rt_hash_table[i].lock);
210 while ((rt = rcu_dereference_protected(*rtp,
211 lockdep_is_held(&dn_rt_hash_table[i].lock))) != NULL) {
212 if (atomic_read(&rt->dst.__refcnt) ||
213 (now - rt->dst.lastuse) < expire) {
214 rtp = &rt->dst.dn_next;
215 continue;
216 }
217 *rtp = rt->dst.dn_next;
218 rt->dst.dn_next = NULL;
219 dnrt_free(rt);
220 }
221 spin_unlock(&dn_rt_hash_table[i].lock);
222
223 if ((jiffies - now) > 0)
224 break;
225 }
226
227 mod_timer(&dn_route_timer, now + decnet_dst_gc_interval * HZ);
228}
229
230static int dn_dst_gc(struct dst_ops *ops)
231{
232 struct dn_route *rt;
233 struct dn_route __rcu **rtp;
234 int i;
235 unsigned long now = jiffies;
236 unsigned long expire = 10 * HZ;
237
238 for (i = 0; i <= dn_rt_hash_mask; i++) {
239
240 spin_lock_bh(&dn_rt_hash_table[i].lock);
241 rtp = &dn_rt_hash_table[i].chain;
242
243 while ((rt = rcu_dereference_protected(*rtp,
244 lockdep_is_held(&dn_rt_hash_table[i].lock))) != NULL) {
245 if (atomic_read(&rt->dst.__refcnt) ||
246 (now - rt->dst.lastuse) < expire) {
247 rtp = &rt->dst.dn_next;
248 continue;
249 }
250 *rtp = rt->dst.dn_next;
251 rt->dst.dn_next = NULL;
252 dnrt_drop(rt);
253 break;
254 }
255 spin_unlock_bh(&dn_rt_hash_table[i].lock);
256 }
257
258 return 0;
259}
260
261/*
262 * The decnet standards don't impose a particular minimum mtu, what they
263 * do insist on is that the routing layer accepts a datagram of at least
264 * 230 bytes long. Here we have to subtract the routing header length from
265 * 230 to get the minimum acceptable mtu. If there is no neighbour, then we
266 * assume the worst and use a long header size.
267 *
268 * We update both the mtu and the advertised mss (i.e. the segment size we
269 * advertise to the other end).
270 */
271static void dn_dst_update_pmtu(struct dst_entry *dst, struct sock *sk,
272 struct sk_buff *skb, u32 mtu)
273{
274 struct dn_route *rt = (struct dn_route *) dst;
275 struct neighbour *n = rt->n;
276 u32 min_mtu = 230;
277 struct dn_dev *dn;
278
279 dn = n ? rcu_dereference_raw(n->dev->dn_ptr) : NULL;
280
281 if (dn && dn->use_long == 0)
282 min_mtu -= 6;
283 else
284 min_mtu -= 21;
285
286 if (dst_metric(dst, RTAX_MTU) > mtu && mtu >= min_mtu) {
287 if (!(dst_metric_locked(dst, RTAX_MTU))) {
288 dst_metric_set(dst, RTAX_MTU, mtu);
289 dst_set_expires(dst, dn_rt_mtu_expires);
290 }
291 if (!(dst_metric_locked(dst, RTAX_ADVMSS))) {
292 u32 mss = mtu - DN_MAX_NSP_DATA_HEADER;
293 u32 existing_mss = dst_metric_raw(dst, RTAX_ADVMSS);
294 if (!existing_mss || existing_mss > mss)
295 dst_metric_set(dst, RTAX_ADVMSS, mss);
296 }
297 }
298}
299
300static void dn_dst_redirect(struct dst_entry *dst, struct sock *sk,
301 struct sk_buff *skb)
302{
303}
304
305/*
306 * When a route has been marked obsolete. (e.g. routing cache flush)
307 */
308static struct dst_entry *dn_dst_check(struct dst_entry *dst, __u32 cookie)
309{
310 return NULL;
311}
312
313static struct dst_entry *dn_dst_negative_advice(struct dst_entry *dst)
314{
315 dst_release(dst);
316 return NULL;
317}
318
319static void dn_dst_link_failure(struct sk_buff *skb)
320{
321}
322
323static inline int compare_keys(struct flowidn *fl1, struct flowidn *fl2)
324{
325 return ((fl1->daddr ^ fl2->daddr) |
326 (fl1->saddr ^ fl2->saddr) |
327 (fl1->flowidn_mark ^ fl2->flowidn_mark) |
328 (fl1->flowidn_scope ^ fl2->flowidn_scope) |
329 (fl1->flowidn_oif ^ fl2->flowidn_oif) |
330 (fl1->flowidn_iif ^ fl2->flowidn_iif)) == 0;
331}
332
333static int dn_insert_route(struct dn_route *rt, unsigned int hash, struct dn_route **rp)
334{
335 struct dn_route *rth;
336 struct dn_route __rcu **rthp;
337 unsigned long now = jiffies;
338
339 rthp = &dn_rt_hash_table[hash].chain;
340
341 spin_lock_bh(&dn_rt_hash_table[hash].lock);
342 while ((rth = rcu_dereference_protected(*rthp,
343 lockdep_is_held(&dn_rt_hash_table[hash].lock))) != NULL) {
344 if (compare_keys(&rth->fld, &rt->fld)) {
345 /* Put it first */
346 *rthp = rth->dst.dn_next;
347 rcu_assign_pointer(rth->dst.dn_next,
348 dn_rt_hash_table[hash].chain);
349 rcu_assign_pointer(dn_rt_hash_table[hash].chain, rth);
350
351 dst_use(&rth->dst, now);
352 spin_unlock_bh(&dn_rt_hash_table[hash].lock);
353
354 dnrt_drop(rt);
355 *rp = rth;
356 return 0;
357 }
358 rthp = &rth->dst.dn_next;
359 }
360
361 rcu_assign_pointer(rt->dst.dn_next, dn_rt_hash_table[hash].chain);
362 rcu_assign_pointer(dn_rt_hash_table[hash].chain, rt);
363
364 dst_use(&rt->dst, now);
365 spin_unlock_bh(&dn_rt_hash_table[hash].lock);
366 *rp = rt;
367 return 0;
368}
369
370static void dn_run_flush(unsigned long dummy)
371{
372 int i;
373 struct dn_route *rt, *next;
374
375 for (i = 0; i < dn_rt_hash_mask; i++) {
376 spin_lock_bh(&dn_rt_hash_table[i].lock);
377
378 if ((rt = xchg((struct dn_route **)&dn_rt_hash_table[i].chain, NULL)) == NULL)
379 goto nothing_to_declare;
380
381 for(; rt; rt = next) {
382 next = rcu_dereference_raw(rt->dst.dn_next);
383 RCU_INIT_POINTER(rt->dst.dn_next, NULL);
384 dst_free((struct dst_entry *)rt);
385 }
386
387nothing_to_declare:
388 spin_unlock_bh(&dn_rt_hash_table[i].lock);
389 }
390}
391
392static DEFINE_SPINLOCK(dn_rt_flush_lock);
393
394void dn_rt_cache_flush(int delay)
395{
396 unsigned long now = jiffies;
397 int user_mode = !in_interrupt();
398
399 if (delay < 0)
400 delay = dn_rt_min_delay;
401
402 spin_lock_bh(&dn_rt_flush_lock);
403
404 if (del_timer(&dn_rt_flush_timer) && delay > 0 && dn_rt_deadline) {
405 long tmo = (long)(dn_rt_deadline - now);
406
407 if (user_mode && tmo < dn_rt_max_delay - dn_rt_min_delay)
408 tmo = 0;
409
410 if (delay > tmo)
411 delay = tmo;
412 }
413
414 if (delay <= 0) {
415 spin_unlock_bh(&dn_rt_flush_lock);
416 dn_run_flush(0);
417 return;
418 }
419
420 if (dn_rt_deadline == 0)
421 dn_rt_deadline = now + dn_rt_max_delay;
422
423 dn_rt_flush_timer.expires = now + delay;
424 add_timer(&dn_rt_flush_timer);
425 spin_unlock_bh(&dn_rt_flush_lock);
426}
427
428/**
429 * dn_return_short - Return a short packet to its sender
430 * @skb: The packet to return
431 *
432 */
433static int dn_return_short(struct sk_buff *skb)
434{
435 struct dn_skb_cb *cb;
436 unsigned char *ptr;
437 __le16 *src;
438 __le16 *dst;
439
440 /* Add back headers */
441 skb_push(skb, skb->data - skb_network_header(skb));
442
443 if ((skb = skb_unshare(skb, GFP_ATOMIC)) == NULL)
444 return NET_RX_DROP;
445
446 cb = DN_SKB_CB(skb);
447 /* Skip packet length and point to flags */
448 ptr = skb->data + 2;
449 *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
450
451 dst = (__le16 *)ptr;
452 ptr += 2;
453 src = (__le16 *)ptr;
454 ptr += 2;
455 *ptr = 0; /* Zero hop count */
456
457 swap(*src, *dst);
458
459 skb->pkt_type = PACKET_OUTGOING;
460 dn_rt_finish_output(skb, NULL, NULL);
461 return NET_RX_SUCCESS;
462}
463
464/**
465 * dn_return_long - Return a long packet to its sender
466 * @skb: The long format packet to return
467 *
468 */
469static int dn_return_long(struct sk_buff *skb)
470{
471 struct dn_skb_cb *cb;
472 unsigned char *ptr;
473 unsigned char *src_addr, *dst_addr;
474 unsigned char tmp[ETH_ALEN];
475
476 /* Add back all headers */
477 skb_push(skb, skb->data - skb_network_header(skb));
478
479 if ((skb = skb_unshare(skb, GFP_ATOMIC)) == NULL)
480 return NET_RX_DROP;
481
482 cb = DN_SKB_CB(skb);
483 /* Ignore packet length and point to flags */
484 ptr = skb->data + 2;
485
486 /* Skip padding */
487 if (*ptr & DN_RT_F_PF) {
488 char padlen = (*ptr & ~DN_RT_F_PF);
489 ptr += padlen;
490 }
491
492 *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
493 ptr += 2;
494 dst_addr = ptr;
495 ptr += 8;
496 src_addr = ptr;
497 ptr += 6;
498 *ptr = 0; /* Zero hop count */
499
500 /* Swap source and destination */
501 memcpy(tmp, src_addr, ETH_ALEN);
502 memcpy(src_addr, dst_addr, ETH_ALEN);
503 memcpy(dst_addr, tmp, ETH_ALEN);
504
505 skb->pkt_type = PACKET_OUTGOING;
506 dn_rt_finish_output(skb, dst_addr, src_addr);
507 return NET_RX_SUCCESS;
508}
509
510/**
511 * dn_route_rx_packet - Try and find a route for an incoming packet
512 * @skb: The packet to find a route for
513 *
514 * Returns: result of input function if route is found, error code otherwise
515 */
516static int dn_route_rx_packet(struct sk_buff *skb)
517{
518 struct dn_skb_cb *cb;
519 int err;
520
521 if ((err = dn_route_input(skb)) == 0)
522 return dst_input(skb);
523
524 cb = DN_SKB_CB(skb);
525 if (decnet_debug_level & 4) {
526 char *devname = skb->dev ? skb->dev->name : "???";
527
528 printk(KERN_DEBUG
529 "DECnet: dn_route_rx_packet: rt_flags=0x%02x dev=%s len=%d src=0x%04hx dst=0x%04hx err=%d type=%d\n",
530 (int)cb->rt_flags, devname, skb->len,
531 le16_to_cpu(cb->src), le16_to_cpu(cb->dst),
532 err, skb->pkt_type);
533 }
534
535 if ((skb->pkt_type == PACKET_HOST) && (cb->rt_flags & DN_RT_F_RQR)) {
536 switch (cb->rt_flags & DN_RT_PKT_MSK) {
537 case DN_RT_PKT_SHORT:
538 return dn_return_short(skb);
539 case DN_RT_PKT_LONG:
540 return dn_return_long(skb);
541 }
542 }
543
544 kfree_skb(skb);
545 return NET_RX_DROP;
546}
547
548static int dn_route_rx_long(struct sk_buff *skb)
549{
550 struct dn_skb_cb *cb = DN_SKB_CB(skb);
551 unsigned char *ptr = skb->data;
552
553 if (!pskb_may_pull(skb, 21)) /* 20 for long header, 1 for shortest nsp */
554 goto drop_it;
555
556 skb_pull(skb, 20);
557 skb_reset_transport_header(skb);
558
559 /* Destination info */
560 ptr += 2;
561 cb->dst = dn_eth2dn(ptr);
562 if (memcmp(ptr, dn_hiord_addr, 4) != 0)
563 goto drop_it;
564 ptr += 6;
565
566
567 /* Source info */
568 ptr += 2;
569 cb->src = dn_eth2dn(ptr);
570 if (memcmp(ptr, dn_hiord_addr, 4) != 0)
571 goto drop_it;
572 ptr += 6;
573 /* Other junk */
574 ptr++;
575 cb->hops = *ptr++; /* Visit Count */
576
577 return NF_HOOK(NFPROTO_DECNET, NF_DN_PRE_ROUTING, skb, 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, skb, skb->dev, NULL,
605 dn_route_rx_packet);
606
607drop_it:
608 kfree_skb(skb);
609 return NET_RX_DROP;
610}
611
612static int dn_route_discard(struct sk_buff *skb)
613{
614 /*
615 * I know we drop the packet here, but thats considered success in
616 * this case
617 */
618 kfree_skb(skb);
619 return NET_RX_SUCCESS;
620}
621
622static int dn_route_ptp_hello(struct sk_buff *skb)
623{
624 dn_dev_hello(skb);
625 dn_neigh_pointopoint_hello(skb);
626 return NET_RX_SUCCESS;
627}
628
629int dn_route_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
630{
631 struct dn_skb_cb *cb;
632 unsigned char flags = 0;
633 __u16 len = le16_to_cpu(*(__le16 *)skb->data);
634 struct dn_dev *dn = rcu_dereference(dev->dn_ptr);
635 unsigned char padlen = 0;
636
637 if (!net_eq(dev_net(dev), &init_net))
638 goto dump_it;
639
640 if (dn == NULL)
641 goto dump_it;
642
643 if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
644 goto out;
645
646 if (!pskb_may_pull(skb, 3))
647 goto dump_it;
648
649 skb_pull(skb, 2);
650
651 if (len > skb->len)
652 goto dump_it;
653
654 skb_trim(skb, len);
655
656 flags = *skb->data;
657
658 cb = DN_SKB_CB(skb);
659 cb->stamp = jiffies;
660 cb->iif = dev->ifindex;
661
662 /*
663 * If we have padding, remove it.
664 */
665 if (flags & DN_RT_F_PF) {
666 padlen = flags & ~DN_RT_F_PF;
667 if (!pskb_may_pull(skb, padlen + 1))
668 goto dump_it;
669 skb_pull(skb, padlen);
670 flags = *skb->data;
671 }
672
673 skb_reset_network_header(skb);
674
675 /*
676 * Weed out future version DECnet
677 */
678 if (flags & DN_RT_F_VER)
679 goto dump_it;
680
681 cb->rt_flags = flags;
682
683 if (decnet_debug_level & 1)
684 printk(KERN_DEBUG
685 "dn_route_rcv: got 0x%02x from %s [%d %d %d]\n",
686 (int)flags, (dev) ? dev->name : "???", len, skb->len,
687 padlen);
688
689 if (flags & DN_RT_PKT_CNTL) {
690 if (unlikely(skb_linearize(skb)))
691 goto dump_it;
692
693 switch (flags & DN_RT_CNTL_MSK) {
694 case DN_RT_PKT_INIT:
695 dn_dev_init_pkt(skb);
696 break;
697 case DN_RT_PKT_VERI:
698 dn_dev_veri_pkt(skb);
699 break;
700 }
701
702 if (dn->parms.state != DN_DEV_S_RU)
703 goto dump_it;
704
705 switch (flags & DN_RT_CNTL_MSK) {
706 case DN_RT_PKT_HELO:
707 return NF_HOOK(NFPROTO_DECNET, NF_DN_HELLO,
708 skb, skb->dev, NULL,
709 dn_route_ptp_hello);
710
711 case DN_RT_PKT_L1RT:
712 case DN_RT_PKT_L2RT:
713 return NF_HOOK(NFPROTO_DECNET, NF_DN_ROUTE,
714 skb, skb->dev, NULL,
715 dn_route_discard);
716 case DN_RT_PKT_ERTH:
717 return NF_HOOK(NFPROTO_DECNET, NF_DN_HELLO,
718 skb, skb->dev, NULL,
719 dn_neigh_router_hello);
720
721 case DN_RT_PKT_EEDH:
722 return NF_HOOK(NFPROTO_DECNET, NF_DN_HELLO,
723 skb, skb->dev, NULL,
724 dn_neigh_endnode_hello);
725 }
726 } else {
727 if (dn->parms.state != DN_DEV_S_RU)
728 goto dump_it;
729
730 skb_pull(skb, 1); /* Pull flags */
731
732 switch (flags & DN_RT_PKT_MSK) {
733 case DN_RT_PKT_LONG:
734 return dn_route_rx_long(skb);
735 case DN_RT_PKT_SHORT:
736 return dn_route_rx_short(skb);
737 }
738 }
739
740dump_it:
741 kfree_skb(skb);
742out:
743 return NET_RX_DROP;
744}
745
746static int dn_to_neigh_output(struct sk_buff *skb)
747{
748 struct dst_entry *dst = skb_dst(skb);
749 struct dn_route *rt = (struct dn_route *) dst;
750 struct neighbour *n = rt->n;
751
752 return n->output(n, skb);
753}
754
755static int dn_output(struct sock *sk, struct sk_buff *skb)
756{
757 struct dst_entry *dst = skb_dst(skb);
758 struct dn_route *rt = (struct dn_route *)dst;
759 struct net_device *dev = dst->dev;
760 struct dn_skb_cb *cb = DN_SKB_CB(skb);
761
762 int err = -EINVAL;
763
764 if (rt->n == NULL)
765 goto error;
766
767 skb->dev = dev;
768
769 cb->src = rt->rt_saddr;
770 cb->dst = rt->rt_daddr;
771
772 /*
773 * Always set the Intra-Ethernet bit on all outgoing packets
774 * originated on this node. Only valid flag from upper layers
775 * is return-to-sender-requested. Set hop count to 0 too.
776 */
777 cb->rt_flags &= ~DN_RT_F_RQR;
778 cb->rt_flags |= DN_RT_F_IE;
779 cb->hops = 0;
780
781 return NF_HOOK(NFPROTO_DECNET, NF_DN_LOCAL_OUT, skb, NULL, dev,
782 dn_to_neigh_output);
783
784error:
785 net_dbg_ratelimited("dn_output: This should not happen\n");
786
787 kfree_skb(skb);
788
789 return err;
790}
791
792static int dn_forward(struct sk_buff *skb)
793{
794 struct dn_skb_cb *cb = DN_SKB_CB(skb);
795 struct dst_entry *dst = skb_dst(skb);
796 struct dn_dev *dn_db = rcu_dereference(dst->dev->dn_ptr);
797 struct dn_route *rt;
798 int header_len;
799#ifdef CONFIG_NETFILTER
800 struct net_device *dev = skb->dev;
801#endif
802
803 if (skb->pkt_type != PACKET_HOST)
804 goto drop;
805
806 /* Ensure that we have enough space for headers */
807 rt = (struct dn_route *)skb_dst(skb);
808 header_len = dn_db->use_long ? 21 : 6;
809 if (skb_cow(skb, LL_RESERVED_SPACE(rt->dst.dev)+header_len))
810 goto drop;
811
812 /*
813 * Hop count exceeded.
814 */
815 if (++cb->hops > 30)
816 goto drop;
817
818 skb->dev = rt->dst.dev;
819
820 /*
821 * If packet goes out same interface it came in on, then set
822 * the Intra-Ethernet bit. This has no effect for short
823 * packets, so we don't need to test for them here.
824 */
825 cb->rt_flags &= ~DN_RT_F_IE;
826 if (rt->rt_flags & RTCF_DOREDIRECT)
827 cb->rt_flags |= DN_RT_F_IE;
828
829 return NF_HOOK(NFPROTO_DECNET, NF_DN_FORWARD, skb, dev, skb->dev,
830 dn_to_neigh_output);
831
832drop:
833 kfree_skb(skb);
834 return NET_RX_DROP;
835}
836
837/*
838 * Used to catch bugs. This should never normally get
839 * called.
840 */
841static int dn_rt_bug_sk(struct sock *sk, struct sk_buff *skb)
842{
843 struct dn_skb_cb *cb = DN_SKB_CB(skb);
844
845 net_dbg_ratelimited("dn_rt_bug: skb from:%04x to:%04x\n",
846 le16_to_cpu(cb->src), le16_to_cpu(cb->dst));
847
848 kfree_skb(skb);
849
850 return NET_RX_DROP;
851}
852
853static int dn_rt_bug(struct sk_buff *skb)
854{
855 struct dn_skb_cb *cb = DN_SKB_CB(skb);
856
857 net_dbg_ratelimited("dn_rt_bug: skb from:%04x to:%04x\n",
858 le16_to_cpu(cb->src), le16_to_cpu(cb->dst));
859
860 kfree_skb(skb);
861
862 return NET_RX_DROP;
863}
864
865static unsigned int dn_dst_default_advmss(const struct dst_entry *dst)
866{
867 return dn_mss_from_pmtu(dst->dev, dst_mtu(dst));
868}
869
870static unsigned int dn_dst_mtu(const struct dst_entry *dst)
871{
872 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
873
874 return mtu ? : dst->dev->mtu;
875}
876
877static struct neighbour *dn_dst_neigh_lookup(const struct dst_entry *dst,
878 struct sk_buff *skb,
879 const void *daddr)
880{
881 return __neigh_lookup_errno(&dn_neigh_table, daddr, dst->dev);
882}
883
884static int dn_rt_set_next_hop(struct dn_route *rt, struct dn_fib_res *res)
885{
886 struct dn_fib_info *fi = res->fi;
887 struct net_device *dev = rt->dst.dev;
888 unsigned int mss_metric;
889 struct neighbour *n;
890
891 if (fi) {
892 if (DN_FIB_RES_GW(*res) &&
893 DN_FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
894 rt->rt_gateway = DN_FIB_RES_GW(*res);
895 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
896 }
897 rt->rt_type = res->type;
898
899 if (dev != NULL && rt->n == NULL) {
900 n = __neigh_lookup_errno(&dn_neigh_table, &rt->rt_gateway, dev);
901 if (IS_ERR(n))
902 return PTR_ERR(n);
903 rt->n = n;
904 }
905
906 if (dst_metric(&rt->dst, RTAX_MTU) > rt->dst.dev->mtu)
907 dst_metric_set(&rt->dst, RTAX_MTU, rt->dst.dev->mtu);
908 mss_metric = dst_metric_raw(&rt->dst, RTAX_ADVMSS);
909 if (mss_metric) {
910 unsigned int mss = dn_mss_from_pmtu(dev, dst_mtu(&rt->dst));
911 if (mss_metric > mss)
912 dst_metric_set(&rt->dst, RTAX_ADVMSS, mss);
913 }
914 return 0;
915}
916
917static inline int dn_match_addr(__le16 addr1, __le16 addr2)
918{
919 __u16 tmp = le16_to_cpu(addr1) ^ le16_to_cpu(addr2);
920 int match = 16;
921 while(tmp) {
922 tmp >>= 1;
923 match--;
924 }
925 return match;
926}
927
928static __le16 dnet_select_source(const struct net_device *dev, __le16 daddr, int scope)
929{
930 __le16 saddr = 0;
931 struct dn_dev *dn_db;
932 struct dn_ifaddr *ifa;
933 int best_match = 0;
934 int ret;
935
936 rcu_read_lock();
937 dn_db = rcu_dereference(dev->dn_ptr);
938 for (ifa = rcu_dereference(dn_db->ifa_list);
939 ifa != NULL;
940 ifa = rcu_dereference(ifa->ifa_next)) {
941 if (ifa->ifa_scope > scope)
942 continue;
943 if (!daddr) {
944 saddr = ifa->ifa_local;
945 break;
946 }
947 ret = dn_match_addr(daddr, ifa->ifa_local);
948 if (ret > best_match)
949 saddr = ifa->ifa_local;
950 if (best_match == 0)
951 saddr = ifa->ifa_local;
952 }
953 rcu_read_unlock();
954
955 return saddr;
956}
957
958static inline __le16 __dn_fib_res_prefsrc(struct dn_fib_res *res)
959{
960 return dnet_select_source(DN_FIB_RES_DEV(*res), DN_FIB_RES_GW(*res), res->scope);
961}
962
963static inline __le16 dn_fib_rules_map_destination(__le16 daddr, struct dn_fib_res *res)
964{
965 __le16 mask = dnet_make_mask(res->prefixlen);
966 return (daddr&~mask)|res->fi->fib_nh->nh_gw;
967}
968
969static int dn_route_output_slow(struct dst_entry **pprt, const struct flowidn *oldflp, int try_hard)
970{
971 struct flowidn fld = {
972 .daddr = oldflp->daddr,
973 .saddr = oldflp->saddr,
974 .flowidn_scope = RT_SCOPE_UNIVERSE,
975 .flowidn_mark = oldflp->flowidn_mark,
976 .flowidn_iif = LOOPBACK_IFINDEX,
977 .flowidn_oif = oldflp->flowidn_oif,
978 };
979 struct dn_route *rt = NULL;
980 struct net_device *dev_out = NULL, *dev;
981 struct neighbour *neigh = NULL;
982 unsigned int hash;
983 unsigned int flags = 0;
984 struct dn_fib_res res = { .fi = NULL, .type = RTN_UNICAST };
985 int err;
986 int free_res = 0;
987 __le16 gateway = 0;
988
989 if (decnet_debug_level & 16)
990 printk(KERN_DEBUG
991 "dn_route_output_slow: dst=%04x src=%04x mark=%d"
992 " iif=%d oif=%d\n", le16_to_cpu(oldflp->daddr),
993 le16_to_cpu(oldflp->saddr),
994 oldflp->flowidn_mark, LOOPBACK_IFINDEX,
995 oldflp->flowidn_oif);
996
997 /* If we have an output interface, verify its a DECnet device */
998 if (oldflp->flowidn_oif) {
999 dev_out = dev_get_by_index(&init_net, oldflp->flowidn_oif);
1000 err = -ENODEV;
1001 if (dev_out && dev_out->dn_ptr == NULL) {
1002 dev_put(dev_out);
1003 dev_out = NULL;
1004 }
1005 if (dev_out == NULL)
1006 goto out;
1007 }
1008
1009 /* If we have a source address, verify that its a local address */
1010 if (oldflp->saddr) {
1011 err = -EADDRNOTAVAIL;
1012
1013 if (dev_out) {
1014 if (dn_dev_islocal(dev_out, oldflp->saddr))
1015 goto source_ok;
1016 dev_put(dev_out);
1017 goto out;
1018 }
1019 rcu_read_lock();
1020 for_each_netdev_rcu(&init_net, dev) {
1021 if (!dev->dn_ptr)
1022 continue;
1023 if (!dn_dev_islocal(dev, oldflp->saddr))
1024 continue;
1025 if ((dev->flags & IFF_LOOPBACK) &&
1026 oldflp->daddr &&
1027 !dn_dev_islocal(dev, oldflp->daddr))
1028 continue;
1029
1030 dev_out = dev;
1031 break;
1032 }
1033 rcu_read_unlock();
1034 if (dev_out == NULL)
1035 goto out;
1036 dev_hold(dev_out);
1037source_ok:
1038 ;
1039 }
1040
1041 /* No destination? Assume its local */
1042 if (!fld.daddr) {
1043 fld.daddr = fld.saddr;
1044
1045 err = -EADDRNOTAVAIL;
1046 if (dev_out)
1047 dev_put(dev_out);
1048 dev_out = init_net.loopback_dev;
1049 dev_hold(dev_out);
1050 if (!fld.daddr) {
1051 fld.daddr =
1052 fld.saddr = dnet_select_source(dev_out, 0,
1053 RT_SCOPE_HOST);
1054 if (!fld.daddr)
1055 goto out;
1056 }
1057 fld.flowidn_oif = LOOPBACK_IFINDEX;
1058 res.type = RTN_LOCAL;
1059 goto make_route;
1060 }
1061
1062 if (decnet_debug_level & 16)
1063 printk(KERN_DEBUG
1064 "dn_route_output_slow: initial checks complete."
1065 " dst=%o4x src=%04x oif=%d try_hard=%d\n",
1066 le16_to_cpu(fld.daddr), le16_to_cpu(fld.saddr),
1067 fld.flowidn_oif, try_hard);
1068
1069 /*
1070 * N.B. If the kernel is compiled without router support then
1071 * dn_fib_lookup() will evaluate to non-zero so this if () block
1072 * will always be executed.
1073 */
1074 err = -ESRCH;
1075 if (try_hard || (err = dn_fib_lookup(&fld, &res)) != 0) {
1076 struct dn_dev *dn_db;
1077 if (err != -ESRCH)
1078 goto out;
1079 /*
1080 * Here the fallback is basically the standard algorithm for
1081 * routing in endnodes which is described in the DECnet routing
1082 * docs
1083 *
1084 * If we are not trying hard, look in neighbour cache.
1085 * The result is tested to ensure that if a specific output
1086 * device/source address was requested, then we honour that
1087 * here
1088 */
1089 if (!try_hard) {
1090 neigh = neigh_lookup_nodev(&dn_neigh_table, &init_net, &fld.daddr);
1091 if (neigh) {
1092 if ((oldflp->flowidn_oif &&
1093 (neigh->dev->ifindex != oldflp->flowidn_oif)) ||
1094 (oldflp->saddr &&
1095 (!dn_dev_islocal(neigh->dev,
1096 oldflp->saddr)))) {
1097 neigh_release(neigh);
1098 neigh = NULL;
1099 } else {
1100 if (dev_out)
1101 dev_put(dev_out);
1102 if (dn_dev_islocal(neigh->dev, fld.daddr)) {
1103 dev_out = init_net.loopback_dev;
1104 res.type = RTN_LOCAL;
1105 } else {
1106 dev_out = neigh->dev;
1107 }
1108 dev_hold(dev_out);
1109 goto select_source;
1110 }
1111 }
1112 }
1113
1114 /* Not there? Perhaps its a local address */
1115 if (dev_out == NULL)
1116 dev_out = dn_dev_get_default();
1117 err = -ENODEV;
1118 if (dev_out == NULL)
1119 goto out;
1120 dn_db = rcu_dereference_raw(dev_out->dn_ptr);
1121 /* Possible improvement - check all devices for local addr */
1122 if (dn_dev_islocal(dev_out, fld.daddr)) {
1123 dev_put(dev_out);
1124 dev_out = init_net.loopback_dev;
1125 dev_hold(dev_out);
1126 res.type = RTN_LOCAL;
1127 goto select_source;
1128 }
1129 /* Not local either.... try sending it to the default router */
1130 neigh = neigh_clone(dn_db->router);
1131 BUG_ON(neigh && neigh->dev != dev_out);
1132
1133 /* Ok then, we assume its directly connected and move on */
1134select_source:
1135 if (neigh)
1136 gateway = ((struct dn_neigh *)neigh)->addr;
1137 if (gateway == 0)
1138 gateway = fld.daddr;
1139 if (fld.saddr == 0) {
1140 fld.saddr = dnet_select_source(dev_out, gateway,
1141 res.type == RTN_LOCAL ?
1142 RT_SCOPE_HOST :
1143 RT_SCOPE_LINK);
1144 if (fld.saddr == 0 && res.type != RTN_LOCAL)
1145 goto e_addr;
1146 }
1147 fld.flowidn_oif = dev_out->ifindex;
1148 goto make_route;
1149 }
1150 free_res = 1;
1151
1152 if (res.type == RTN_NAT)
1153 goto e_inval;
1154
1155 if (res.type == RTN_LOCAL) {
1156 if (!fld.saddr)
1157 fld.saddr = fld.daddr;
1158 if (dev_out)
1159 dev_put(dev_out);
1160 dev_out = init_net.loopback_dev;
1161 dev_hold(dev_out);
1162 fld.flowidn_oif = dev_out->ifindex;
1163 if (res.fi)
1164 dn_fib_info_put(res.fi);
1165 res.fi = NULL;
1166 goto make_route;
1167 }
1168
1169 if (res.fi->fib_nhs > 1 && fld.flowidn_oif == 0)
1170 dn_fib_select_multipath(&fld, &res);
1171
1172 /*
1173 * We could add some logic to deal with default routes here and
1174 * get rid of some of the special casing above.
1175 */
1176
1177 if (!fld.saddr)
1178 fld.saddr = DN_FIB_RES_PREFSRC(res);
1179
1180 if (dev_out)
1181 dev_put(dev_out);
1182 dev_out = DN_FIB_RES_DEV(res);
1183 dev_hold(dev_out);
1184 fld.flowidn_oif = dev_out->ifindex;
1185 gateway = DN_FIB_RES_GW(res);
1186
1187make_route:
1188 if (dev_out->flags & IFF_LOOPBACK)
1189 flags |= RTCF_LOCAL;
1190
1191 rt = dst_alloc(&dn_dst_ops, dev_out, 1, DST_OBSOLETE_NONE, DST_HOST);
1192 if (rt == NULL)
1193 goto e_nobufs;
1194
1195 memset(&rt->fld, 0, sizeof(rt->fld));
1196 rt->fld.saddr = oldflp->saddr;
1197 rt->fld.daddr = oldflp->daddr;
1198 rt->fld.flowidn_oif = oldflp->flowidn_oif;
1199 rt->fld.flowidn_iif = 0;
1200 rt->fld.flowidn_mark = oldflp->flowidn_mark;
1201
1202 rt->rt_saddr = fld.saddr;
1203 rt->rt_daddr = fld.daddr;
1204 rt->rt_gateway = gateway ? gateway : fld.daddr;
1205 rt->rt_local_src = fld.saddr;
1206
1207 rt->rt_dst_map = fld.daddr;
1208 rt->rt_src_map = fld.saddr;
1209
1210 rt->n = neigh;
1211 neigh = NULL;
1212
1213 rt->dst.lastuse = jiffies;
1214 rt->dst.output = dn_output;
1215 rt->dst.input = dn_rt_bug;
1216 rt->rt_flags = flags;
1217 if (flags & RTCF_LOCAL)
1218 rt->dst.input = dn_nsp_rx;
1219
1220 err = dn_rt_set_next_hop(rt, &res);
1221 if (err)
1222 goto e_neighbour;
1223
1224 hash = dn_hash(rt->fld.saddr, rt->fld.daddr);
1225 dn_insert_route(rt, hash, (struct dn_route **)pprt);
1226
1227done:
1228 if (neigh)
1229 neigh_release(neigh);
1230 if (free_res)
1231 dn_fib_res_put(&res);
1232 if (dev_out)
1233 dev_put(dev_out);
1234out:
1235 return err;
1236
1237e_addr:
1238 err = -EADDRNOTAVAIL;
1239 goto done;
1240e_inval:
1241 err = -EINVAL;
1242 goto done;
1243e_nobufs:
1244 err = -ENOBUFS;
1245 goto done;
1246e_neighbour:
1247 dst_free(&rt->dst);
1248 goto e_nobufs;
1249}
1250
1251
1252/*
1253 * N.B. The flags may be moved into the flowi at some future stage.
1254 */
1255static int __dn_route_output_key(struct dst_entry **pprt, const struct flowidn *flp, int flags)
1256{
1257 unsigned int hash = dn_hash(flp->saddr, flp->daddr);
1258 struct dn_route *rt = NULL;
1259
1260 if (!(flags & MSG_TRYHARD)) {
1261 rcu_read_lock_bh();
1262 for (rt = rcu_dereference_bh(dn_rt_hash_table[hash].chain); rt;
1263 rt = rcu_dereference_bh(rt->dst.dn_next)) {
1264 if ((flp->daddr == rt->fld.daddr) &&
1265 (flp->saddr == rt->fld.saddr) &&
1266 (flp->flowidn_mark == rt->fld.flowidn_mark) &&
1267 dn_is_output_route(rt) &&
1268 (rt->fld.flowidn_oif == flp->flowidn_oif)) {
1269 dst_use(&rt->dst, jiffies);
1270 rcu_read_unlock_bh();
1271 *pprt = &rt->dst;
1272 return 0;
1273 }
1274 }
1275 rcu_read_unlock_bh();
1276 }
1277
1278 return dn_route_output_slow(pprt, flp, flags);
1279}
1280
1281static int dn_route_output_key(struct dst_entry **pprt, struct flowidn *flp, int flags)
1282{
1283 int err;
1284
1285 err = __dn_route_output_key(pprt, flp, flags);
1286 if (err == 0 && flp->flowidn_proto) {
1287 *pprt = xfrm_lookup(&init_net, *pprt,
1288 flowidn_to_flowi(flp), NULL, 0);
1289 if (IS_ERR(*pprt)) {
1290 err = PTR_ERR(*pprt);
1291 *pprt = NULL;
1292 }
1293 }
1294 return err;
1295}
1296
1297int dn_route_output_sock(struct dst_entry __rcu **pprt, struct flowidn *fl, struct sock *sk, int flags)
1298{
1299 int err;
1300
1301 err = __dn_route_output_key(pprt, fl, flags & MSG_TRYHARD);
1302 if (err == 0 && fl->flowidn_proto) {
1303 *pprt = xfrm_lookup(&init_net, *pprt,
1304 flowidn_to_flowi(fl), sk, 0);
1305 if (IS_ERR(*pprt)) {
1306 err = PTR_ERR(*pprt);
1307 *pprt = NULL;
1308 }
1309 }
1310 return err;
1311}
1312
1313static int dn_route_input_slow(struct sk_buff *skb)
1314{
1315 struct dn_route *rt = NULL;
1316 struct dn_skb_cb *cb = DN_SKB_CB(skb);
1317 struct net_device *in_dev = skb->dev;
1318 struct net_device *out_dev = NULL;
1319 struct dn_dev *dn_db;
1320 struct neighbour *neigh = NULL;
1321 unsigned int hash;
1322 int flags = 0;
1323 __le16 gateway = 0;
1324 __le16 local_src = 0;
1325 struct flowidn fld = {
1326 .daddr = cb->dst,
1327 .saddr = cb->src,
1328 .flowidn_scope = RT_SCOPE_UNIVERSE,
1329 .flowidn_mark = skb->mark,
1330 .flowidn_iif = skb->dev->ifindex,
1331 };
1332 struct dn_fib_res res = { .fi = NULL, .type = RTN_UNREACHABLE };
1333 int err = -EINVAL;
1334 int free_res = 0;
1335
1336 dev_hold(in_dev);
1337
1338 if ((dn_db = rcu_dereference(in_dev->dn_ptr)) == NULL)
1339 goto out;
1340
1341 /* Zero source addresses are not allowed */
1342 if (fld.saddr == 0)
1343 goto out;
1344
1345 /*
1346 * In this case we've just received a packet from a source
1347 * outside ourselves pretending to come from us. We don't
1348 * allow it any further to prevent routing loops, spoofing and
1349 * other nasties. Loopback packets already have the dst attached
1350 * so this only affects packets which have originated elsewhere.
1351 */
1352 err = -ENOTUNIQ;
1353 if (dn_dev_islocal(in_dev, cb->src))
1354 goto out;
1355
1356 err = dn_fib_lookup(&fld, &res);
1357 if (err) {
1358 if (err != -ESRCH)
1359 goto out;
1360 /*
1361 * Is the destination us ?
1362 */
1363 if (!dn_dev_islocal(in_dev, cb->dst))
1364 goto e_inval;
1365
1366 res.type = RTN_LOCAL;
1367 } else {
1368 __le16 src_map = fld.saddr;
1369 free_res = 1;
1370
1371 out_dev = DN_FIB_RES_DEV(res);
1372 if (out_dev == NULL) {
1373 net_crit_ratelimited("Bug in dn_route_input_slow() No output device\n");
1374 goto e_inval;
1375 }
1376 dev_hold(out_dev);
1377
1378 if (res.r)
1379 src_map = fld.saddr; /* no NAT support for now */
1380
1381 gateway = DN_FIB_RES_GW(res);
1382 if (res.type == RTN_NAT) {
1383 fld.daddr = dn_fib_rules_map_destination(fld.daddr, &res);
1384 dn_fib_res_put(&res);
1385 free_res = 0;
1386 if (dn_fib_lookup(&fld, &res))
1387 goto e_inval;
1388 free_res = 1;
1389 if (res.type != RTN_UNICAST)
1390 goto e_inval;
1391 flags |= RTCF_DNAT;
1392 gateway = fld.daddr;
1393 }
1394 fld.saddr = src_map;
1395 }
1396
1397 switch(res.type) {
1398 case RTN_UNICAST:
1399 /*
1400 * Forwarding check here, we only check for forwarding
1401 * being turned off, if you want to only forward intra
1402 * area, its up to you to set the routing tables up
1403 * correctly.
1404 */
1405 if (dn_db->parms.forwarding == 0)
1406 goto e_inval;
1407
1408 if (res.fi->fib_nhs > 1 && fld.flowidn_oif == 0)
1409 dn_fib_select_multipath(&fld, &res);
1410
1411 /*
1412 * Check for out_dev == in_dev. We use the RTCF_DOREDIRECT
1413 * flag as a hint to set the intra-ethernet bit when
1414 * forwarding. If we've got NAT in operation, we don't do
1415 * this optimisation.
1416 */
1417 if (out_dev == in_dev && !(flags & RTCF_NAT))
1418 flags |= RTCF_DOREDIRECT;
1419
1420 local_src = DN_FIB_RES_PREFSRC(res);
1421
1422 case RTN_BLACKHOLE:
1423 case RTN_UNREACHABLE:
1424 break;
1425 case RTN_LOCAL:
1426 flags |= RTCF_LOCAL;
1427 fld.saddr = cb->dst;
1428 fld.daddr = cb->src;
1429
1430 /* Routing tables gave us a gateway */
1431 if (gateway)
1432 goto make_route;
1433
1434 /* Packet was intra-ethernet, so we know its on-link */
1435 if (cb->rt_flags & DN_RT_F_IE) {
1436 gateway = cb->src;
1437 goto make_route;
1438 }
1439
1440 /* Use the default router if there is one */
1441 neigh = neigh_clone(dn_db->router);
1442 if (neigh) {
1443 gateway = ((struct dn_neigh *)neigh)->addr;
1444 goto make_route;
1445 }
1446
1447 /* Close eyes and pray */
1448 gateway = cb->src;
1449 goto make_route;
1450 default:
1451 goto e_inval;
1452 }
1453
1454make_route:
1455 rt = dst_alloc(&dn_dst_ops, out_dev, 0, DST_OBSOLETE_NONE, DST_HOST);
1456 if (rt == NULL)
1457 goto e_nobufs;
1458
1459 memset(&rt->fld, 0, sizeof(rt->fld));
1460 rt->rt_saddr = fld.saddr;
1461 rt->rt_daddr = fld.daddr;
1462 rt->rt_gateway = fld.daddr;
1463 if (gateway)
1464 rt->rt_gateway = gateway;
1465 rt->rt_local_src = local_src ? local_src : rt->rt_saddr;
1466
1467 rt->rt_dst_map = fld.daddr;
1468 rt->rt_src_map = fld.saddr;
1469
1470 rt->fld.saddr = cb->src;
1471 rt->fld.daddr = cb->dst;
1472 rt->fld.flowidn_oif = 0;
1473 rt->fld.flowidn_iif = in_dev->ifindex;
1474 rt->fld.flowidn_mark = fld.flowidn_mark;
1475
1476 rt->n = neigh;
1477 rt->dst.lastuse = jiffies;
1478 rt->dst.output = dn_rt_bug_sk;
1479 switch (res.type) {
1480 case RTN_UNICAST:
1481 rt->dst.input = dn_forward;
1482 break;
1483 case RTN_LOCAL:
1484 rt->dst.output = dn_output;
1485 rt->dst.input = dn_nsp_rx;
1486 rt->dst.dev = in_dev;
1487 flags |= RTCF_LOCAL;
1488 break;
1489 default:
1490 case RTN_UNREACHABLE:
1491 case RTN_BLACKHOLE:
1492 rt->dst.input = dst_discard;
1493 }
1494 rt->rt_flags = flags;
1495
1496 err = dn_rt_set_next_hop(rt, &res);
1497 if (err)
1498 goto e_neighbour;
1499
1500 hash = dn_hash(rt->fld.saddr, rt->fld.daddr);
1501 dn_insert_route(rt, hash, &rt);
1502 skb_dst_set(skb, &rt->dst);
1503
1504done:
1505 if (neigh)
1506 neigh_release(neigh);
1507 if (free_res)
1508 dn_fib_res_put(&res);
1509 dev_put(in_dev);
1510 if (out_dev)
1511 dev_put(out_dev);
1512out:
1513 return err;
1514
1515e_inval:
1516 err = -EINVAL;
1517 goto done;
1518
1519e_nobufs:
1520 err = -ENOBUFS;
1521 goto done;
1522
1523e_neighbour:
1524 dst_free(&rt->dst);
1525 goto done;
1526}
1527
1528static int dn_route_input(struct sk_buff *skb)
1529{
1530 struct dn_route *rt;
1531 struct dn_skb_cb *cb = DN_SKB_CB(skb);
1532 unsigned int hash = dn_hash(cb->src, cb->dst);
1533
1534 if (skb_dst(skb))
1535 return 0;
1536
1537 rcu_read_lock();
1538 for(rt = rcu_dereference(dn_rt_hash_table[hash].chain); rt != NULL;
1539 rt = rcu_dereference(rt->dst.dn_next)) {
1540 if ((rt->fld.saddr == cb->src) &&
1541 (rt->fld.daddr == cb->dst) &&
1542 (rt->fld.flowidn_oif == 0) &&
1543 (rt->fld.flowidn_mark == skb->mark) &&
1544 (rt->fld.flowidn_iif == cb->iif)) {
1545 dst_use(&rt->dst, jiffies);
1546 rcu_read_unlock();
1547 skb_dst_set(skb, (struct dst_entry *)rt);
1548 return 0;
1549 }
1550 }
1551 rcu_read_unlock();
1552
1553 return dn_route_input_slow(skb);
1554}
1555
1556static int dn_rt_fill_info(struct sk_buff *skb, u32 portid, u32 seq,
1557 int event, int nowait, unsigned int flags)
1558{
1559 struct dn_route *rt = (struct dn_route *)skb_dst(skb);
1560 struct rtmsg *r;
1561 struct nlmsghdr *nlh;
1562 long expires;
1563
1564 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*r), flags);
1565 if (!nlh)
1566 return -EMSGSIZE;
1567
1568 r = nlmsg_data(nlh);
1569 r->rtm_family = AF_DECnet;
1570 r->rtm_dst_len = 16;
1571 r->rtm_src_len = 0;
1572 r->rtm_tos = 0;
1573 r->rtm_table = RT_TABLE_MAIN;
1574 r->rtm_type = rt->rt_type;
1575 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
1576 r->rtm_scope = RT_SCOPE_UNIVERSE;
1577 r->rtm_protocol = RTPROT_UNSPEC;
1578
1579 if (rt->rt_flags & RTCF_NOTIFY)
1580 r->rtm_flags |= RTM_F_NOTIFY;
1581
1582 if (nla_put_u32(skb, RTA_TABLE, RT_TABLE_MAIN) < 0 ||
1583 nla_put_le16(skb, RTA_DST, rt->rt_daddr) < 0)
1584 goto errout;
1585
1586 if (rt->fld.saddr) {
1587 r->rtm_src_len = 16;
1588 if (nla_put_le16(skb, RTA_SRC, rt->fld.saddr) < 0)
1589 goto errout;
1590 }
1591 if (rt->dst.dev &&
1592 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex) < 0)
1593 goto errout;
1594
1595 /*
1596 * Note to self - change this if input routes reverse direction when
1597 * they deal only with inputs and not with replies like they do
1598 * currently.
1599 */
1600 if (nla_put_le16(skb, RTA_PREFSRC, rt->rt_local_src) < 0)
1601 goto errout;
1602
1603 if (rt->rt_daddr != rt->rt_gateway &&
1604 nla_put_le16(skb, RTA_GATEWAY, rt->rt_gateway) < 0)
1605 goto errout;
1606
1607 if (rtnetlink_put_metrics(skb, dst_metrics_ptr(&rt->dst)) < 0)
1608 goto errout;
1609
1610 expires = rt->dst.expires ? rt->dst.expires - jiffies : 0;
1611 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires,
1612 rt->dst.error) < 0)
1613 goto errout;
1614
1615 if (dn_is_input_route(rt) &&
1616 nla_put_u32(skb, RTA_IIF, rt->fld.flowidn_iif) < 0)
1617 goto errout;
1618
1619 return nlmsg_end(skb, nlh);
1620
1621errout:
1622 nlmsg_cancel(skb, nlh);
1623 return -EMSGSIZE;
1624}
1625
1626const struct nla_policy rtm_dn_policy[RTA_MAX + 1] = {
1627 [RTA_DST] = { .type = NLA_U16 },
1628 [RTA_SRC] = { .type = NLA_U16 },
1629 [RTA_IIF] = { .type = NLA_U32 },
1630 [RTA_OIF] = { .type = NLA_U32 },
1631 [RTA_GATEWAY] = { .type = NLA_U16 },
1632 [RTA_PRIORITY] = { .type = NLA_U32 },
1633 [RTA_PREFSRC] = { .type = NLA_U16 },
1634 [RTA_METRICS] = { .type = NLA_NESTED },
1635 [RTA_MULTIPATH] = { .type = NLA_NESTED },
1636 [RTA_TABLE] = { .type = NLA_U32 },
1637 [RTA_MARK] = { .type = NLA_U32 },
1638};
1639
1640/*
1641 * This is called by both endnodes and routers now.
1642 */
1643static int dn_cache_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh)
1644{
1645 struct net *net = sock_net(in_skb->sk);
1646 struct rtmsg *rtm = nlmsg_data(nlh);
1647 struct dn_route *rt = NULL;
1648 struct dn_skb_cb *cb;
1649 int err;
1650 struct sk_buff *skb;
1651 struct flowidn fld;
1652 struct nlattr *tb[RTA_MAX+1];
1653
1654 if (!net_eq(net, &init_net))
1655 return -EINVAL;
1656
1657 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_dn_policy);
1658 if (err < 0)
1659 return err;
1660
1661 memset(&fld, 0, sizeof(fld));
1662 fld.flowidn_proto = DNPROTO_NSP;
1663
1664 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1665 if (skb == NULL)
1666 return -ENOBUFS;
1667 skb_reset_mac_header(skb);
1668 cb = DN_SKB_CB(skb);
1669
1670 if (tb[RTA_SRC])
1671 fld.saddr = nla_get_le16(tb[RTA_SRC]);
1672
1673 if (tb[RTA_DST])
1674 fld.daddr = nla_get_le16(tb[RTA_DST]);
1675
1676 if (tb[RTA_IIF])
1677 fld.flowidn_iif = nla_get_u32(tb[RTA_IIF]);
1678
1679 if (fld.flowidn_iif) {
1680 struct net_device *dev;
1681 dev = __dev_get_by_index(&init_net, fld.flowidn_iif);
1682 if (!dev || !dev->dn_ptr) {
1683 kfree_skb(skb);
1684 return -ENODEV;
1685 }
1686 skb->protocol = htons(ETH_P_DNA_RT);
1687 skb->dev = dev;
1688 cb->src = fld.saddr;
1689 cb->dst = fld.daddr;
1690 local_bh_disable();
1691 err = dn_route_input(skb);
1692 local_bh_enable();
1693 memset(cb, 0, sizeof(struct dn_skb_cb));
1694 rt = (struct dn_route *)skb_dst(skb);
1695 if (!err && -rt->dst.error)
1696 err = rt->dst.error;
1697 } else {
1698 if (tb[RTA_OIF])
1699 fld.flowidn_oif = nla_get_u32(tb[RTA_OIF]);
1700
1701 err = dn_route_output_key((struct dst_entry **)&rt, &fld, 0);
1702 }
1703
1704 skb->dev = NULL;
1705 if (err)
1706 goto out_free;
1707 skb_dst_set(skb, &rt->dst);
1708 if (rtm->rtm_flags & RTM_F_NOTIFY)
1709 rt->rt_flags |= RTCF_NOTIFY;
1710
1711 err = dn_rt_fill_info(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq, RTM_NEWROUTE, 0, 0);
1712
1713 if (err == 0)
1714 goto out_free;
1715 if (err < 0) {
1716 err = -EMSGSIZE;
1717 goto out_free;
1718 }
1719
1720 return rtnl_unicast(skb, &init_net, NETLINK_CB(in_skb).portid);
1721
1722out_free:
1723 kfree_skb(skb);
1724 return err;
1725}
1726
1727/*
1728 * For routers, this is called from dn_fib_dump, but for endnodes its
1729 * called directly from the rtnetlink dispatch table.
1730 */
1731int dn_cache_dump(struct sk_buff *skb, struct netlink_callback *cb)
1732{
1733 struct net *net = sock_net(skb->sk);
1734 struct dn_route *rt;
1735 int h, s_h;
1736 int idx, s_idx;
1737 struct rtmsg *rtm;
1738
1739 if (!net_eq(net, &init_net))
1740 return 0;
1741
1742 if (nlmsg_len(cb->nlh) < sizeof(struct rtmsg))
1743 return -EINVAL;
1744
1745 rtm = nlmsg_data(cb->nlh);
1746 if (!(rtm->rtm_flags & RTM_F_CLONED))
1747 return 0;
1748
1749 s_h = cb->args[0];
1750 s_idx = idx = cb->args[1];
1751 for(h = 0; h <= dn_rt_hash_mask; h++) {
1752 if (h < s_h)
1753 continue;
1754 if (h > s_h)
1755 s_idx = 0;
1756 rcu_read_lock_bh();
1757 for(rt = rcu_dereference_bh(dn_rt_hash_table[h].chain), idx = 0;
1758 rt;
1759 rt = rcu_dereference_bh(rt->dst.dn_next), idx++) {
1760 if (idx < s_idx)
1761 continue;
1762 skb_dst_set(skb, dst_clone(&rt->dst));
1763 if (dn_rt_fill_info(skb, NETLINK_CB(cb->skb).portid,
1764 cb->nlh->nlmsg_seq, RTM_NEWROUTE,
1765 1, NLM_F_MULTI) <= 0) {
1766 skb_dst_drop(skb);
1767 rcu_read_unlock_bh();
1768 goto done;
1769 }
1770 skb_dst_drop(skb);
1771 }
1772 rcu_read_unlock_bh();
1773 }
1774
1775done:
1776 cb->args[0] = h;
1777 cb->args[1] = idx;
1778 return skb->len;
1779}
1780
1781#ifdef CONFIG_PROC_FS
1782struct dn_rt_cache_iter_state {
1783 int bucket;
1784};
1785
1786static struct dn_route *dn_rt_cache_get_first(struct seq_file *seq)
1787{
1788 struct dn_route *rt = NULL;
1789 struct dn_rt_cache_iter_state *s = seq->private;
1790
1791 for(s->bucket = dn_rt_hash_mask; s->bucket >= 0; --s->bucket) {
1792 rcu_read_lock_bh();
1793 rt = rcu_dereference_bh(dn_rt_hash_table[s->bucket].chain);
1794 if (rt)
1795 break;
1796 rcu_read_unlock_bh();
1797 }
1798 return rt;
1799}
1800
1801static struct dn_route *dn_rt_cache_get_next(struct seq_file *seq, struct dn_route *rt)
1802{
1803 struct dn_rt_cache_iter_state *s = seq->private;
1804
1805 rt = rcu_dereference_bh(rt->dst.dn_next);
1806 while (!rt) {
1807 rcu_read_unlock_bh();
1808 if (--s->bucket < 0)
1809 break;
1810 rcu_read_lock_bh();
1811 rt = rcu_dereference_bh(dn_rt_hash_table[s->bucket].chain);
1812 }
1813 return rt;
1814}
1815
1816static void *dn_rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
1817{
1818 struct dn_route *rt = dn_rt_cache_get_first(seq);
1819
1820 if (rt) {
1821 while(*pos && (rt = dn_rt_cache_get_next(seq, rt)))
1822 --*pos;
1823 }
1824 return *pos ? NULL : rt;
1825}
1826
1827static void *dn_rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1828{
1829 struct dn_route *rt = dn_rt_cache_get_next(seq, v);
1830 ++*pos;
1831 return rt;
1832}
1833
1834static void dn_rt_cache_seq_stop(struct seq_file *seq, void *v)
1835{
1836 if (v)
1837 rcu_read_unlock_bh();
1838}
1839
1840static int dn_rt_cache_seq_show(struct seq_file *seq, void *v)
1841{
1842 struct dn_route *rt = v;
1843 char buf1[DN_ASCBUF_LEN], buf2[DN_ASCBUF_LEN];
1844
1845 seq_printf(seq, "%-8s %-7s %-7s %04d %04d %04d\n",
1846 rt->dst.dev ? rt->dst.dev->name : "*",
1847 dn_addr2asc(le16_to_cpu(rt->rt_daddr), buf1),
1848 dn_addr2asc(le16_to_cpu(rt->rt_saddr), buf2),
1849 atomic_read(&rt->dst.__refcnt),
1850 rt->dst.__use, 0);
1851 return 0;
1852}
1853
1854static const struct seq_operations dn_rt_cache_seq_ops = {
1855 .start = dn_rt_cache_seq_start,
1856 .next = dn_rt_cache_seq_next,
1857 .stop = dn_rt_cache_seq_stop,
1858 .show = dn_rt_cache_seq_show,
1859};
1860
1861static int dn_rt_cache_seq_open(struct inode *inode, struct file *file)
1862{
1863 return seq_open_private(file, &dn_rt_cache_seq_ops,
1864 sizeof(struct dn_rt_cache_iter_state));
1865}
1866
1867static const struct file_operations dn_rt_cache_seq_fops = {
1868 .owner = THIS_MODULE,
1869 .open = dn_rt_cache_seq_open,
1870 .read = seq_read,
1871 .llseek = seq_lseek,
1872 .release = seq_release_private,
1873};
1874
1875#endif /* CONFIG_PROC_FS */
1876
1877void __init dn_route_init(void)
1878{
1879 int i, goal, order;
1880
1881 dn_dst_ops.kmem_cachep =
1882 kmem_cache_create("dn_dst_cache", sizeof(struct dn_route), 0,
1883 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1884 dst_entries_init(&dn_dst_ops);
1885 setup_timer(&dn_route_timer, dn_dst_check_expire, 0);
1886 dn_route_timer.expires = jiffies + decnet_dst_gc_interval * HZ;
1887 add_timer(&dn_route_timer);
1888
1889 goal = totalram_pages >> (26 - PAGE_SHIFT);
1890
1891 for(order = 0; (1UL << order) < goal; order++)
1892 /* NOTHING */;
1893
1894 /*
1895 * Only want 1024 entries max, since the table is very, very unlikely
1896 * to be larger than that.
1897 */
1898 while(order && ((((1UL << order) * PAGE_SIZE) /
1899 sizeof(struct dn_rt_hash_bucket)) >= 2048))
1900 order--;
1901
1902 do {
1903 dn_rt_hash_mask = (1UL << order) * PAGE_SIZE /
1904 sizeof(struct dn_rt_hash_bucket);
1905 while(dn_rt_hash_mask & (dn_rt_hash_mask - 1))
1906 dn_rt_hash_mask--;
1907 dn_rt_hash_table = (struct dn_rt_hash_bucket *)
1908 __get_free_pages(GFP_ATOMIC, order);
1909 } while (dn_rt_hash_table == NULL && --order > 0);
1910
1911 if (!dn_rt_hash_table)
1912 panic("Failed to allocate DECnet route cache hash table\n");
1913
1914 printk(KERN_INFO
1915 "DECnet: Routing cache hash table of %u buckets, %ldKbytes\n",
1916 dn_rt_hash_mask,
1917 (long)(dn_rt_hash_mask*sizeof(struct dn_rt_hash_bucket))/1024);
1918
1919 dn_rt_hash_mask--;
1920 for(i = 0; i <= dn_rt_hash_mask; i++) {
1921 spin_lock_init(&dn_rt_hash_table[i].lock);
1922 dn_rt_hash_table[i].chain = NULL;
1923 }
1924
1925 dn_dst_ops.gc_thresh = (dn_rt_hash_mask + 1);
1926
1927 proc_create("decnet_cache", S_IRUGO, init_net.proc_net,
1928 &dn_rt_cache_seq_fops);
1929
1930#ifdef CONFIG_DECNET_ROUTER
1931 rtnl_register(PF_DECnet, RTM_GETROUTE, dn_cache_getroute,
1932 dn_fib_dump, NULL);
1933#else
1934 rtnl_register(PF_DECnet, RTM_GETROUTE, dn_cache_getroute,
1935 dn_cache_dump, NULL);
1936#endif
1937}
1938
1939void __exit dn_route_cleanup(void)
1940{
1941 del_timer(&dn_route_timer);
1942 dn_run_flush(0);
1943
1944 remove_proc_entry("decnet_cache", init_net.proc_net);
1945 dst_entries_destroy(&dn_dst_ops);
1946}
1947