Loading...
Note: File does not exist in v6.2.
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{
89 struct dn_route __rcu *chain;
90 spinlock_t lock;
91};
92
93extern struct neigh_table dn_neigh_table;
94
95
96static unsigned char dn_hiord_addr[6] = {0xAA,0x00,0x04,0x00,0x00,0x00};
97
98static const int dn_rt_min_delay = 2 * HZ;
99static const int dn_rt_max_delay = 10 * HZ;
100static const int dn_rt_mtu_expires = 10 * 60 * HZ;
101
102static unsigned long dn_rt_deadline;
103
104static int dn_dst_gc(struct dst_ops *ops);
105static struct dst_entry *dn_dst_check(struct dst_entry *, __u32);
106static unsigned int dn_dst_default_advmss(const struct dst_entry *dst);
107static unsigned int dn_dst_mtu(const struct dst_entry *dst);
108static void dn_dst_destroy(struct dst_entry *);
109static void dn_dst_ifdown(struct dst_entry *, struct net_device *dev, int how);
110static struct dst_entry *dn_dst_negative_advice(struct dst_entry *);
111static void dn_dst_link_failure(struct sk_buff *);
112static void dn_dst_update_pmtu(struct dst_entry *dst, struct sock *sk,
113 struct sk_buff *skb , u32 mtu,
114 bool confirm_neigh);
115static void dn_dst_redirect(struct dst_entry *dst, struct sock *sk,
116 struct sk_buff *skb);
117static struct neighbour *dn_dst_neigh_lookup(const struct dst_entry *dst,
118 struct sk_buff *skb,
119 const void *daddr);
120static int dn_route_input(struct sk_buff *);
121static void dn_run_flush(struct timer_list *unused);
122
123static struct dn_rt_hash_bucket *dn_rt_hash_table;
124static unsigned int dn_rt_hash_mask;
125
126static struct timer_list dn_route_timer;
127static DEFINE_TIMER(dn_rt_flush_timer, dn_run_flush);
128int decnet_dst_gc_interval = 2;
129
130static struct dst_ops dn_dst_ops = {
131 .family = PF_DECnet,
132 .gc_thresh = 128,
133 .gc = dn_dst_gc,
134 .check = dn_dst_check,
135 .default_advmss = dn_dst_default_advmss,
136 .mtu = dn_dst_mtu,
137 .cow_metrics = dst_cow_metrics_generic,
138 .destroy = dn_dst_destroy,
139 .ifdown = dn_dst_ifdown,
140 .negative_advice = dn_dst_negative_advice,
141 .link_failure = dn_dst_link_failure,
142 .update_pmtu = dn_dst_update_pmtu,
143 .redirect = dn_dst_redirect,
144 .neigh_lookup = dn_dst_neigh_lookup,
145};
146
147static void dn_dst_destroy(struct dst_entry *dst)
148{
149 struct dn_route *rt = (struct dn_route *) dst;
150
151 if (rt->n)
152 neigh_release(rt->n);
153 dst_destroy_metrics_generic(dst);
154}
155
156static void dn_dst_ifdown(struct dst_entry *dst, struct net_device *dev, int how)
157{
158 if (how) {
159 struct dn_route *rt = (struct dn_route *) dst;
160 struct neighbour *n = rt->n;
161
162 if (n && n->dev == dev) {
163 n->dev = dev_net(dev)->loopback_dev;
164 dev_hold(n->dev);
165 dev_put(dev);
166 }
167 }
168}
169
170static __inline__ unsigned int dn_hash(__le16 src, __le16 dst)
171{
172 __u16 tmp = (__u16 __force)(src ^ dst);
173 tmp ^= (tmp >> 3);
174 tmp ^= (tmp >> 5);
175 tmp ^= (tmp >> 10);
176 return dn_rt_hash_mask & (unsigned int)tmp;
177}
178
179static void dn_dst_check_expire(struct timer_list *unused)
180{
181 int i;
182 struct dn_route *rt;
183 struct dn_route __rcu **rtp;
184 unsigned long now = jiffies;
185 unsigned long expire = 120 * HZ;
186
187 for (i = 0; i <= dn_rt_hash_mask; i++) {
188 rtp = &dn_rt_hash_table[i].chain;
189
190 spin_lock(&dn_rt_hash_table[i].lock);
191 while ((rt = rcu_dereference_protected(*rtp,
192 lockdep_is_held(&dn_rt_hash_table[i].lock))) != NULL) {
193 if (atomic_read(&rt->dst.__refcnt) > 1 ||
194 (now - rt->dst.lastuse) < expire) {
195 rtp = &rt->dn_next;
196 continue;
197 }
198 *rtp = rt->dn_next;
199 rt->dn_next = NULL;
200 dst_dev_put(&rt->dst);
201 dst_release(&rt->dst);
202 }
203 spin_unlock(&dn_rt_hash_table[i].lock);
204
205 if ((jiffies - now) > 0)
206 break;
207 }
208
209 mod_timer(&dn_route_timer, now + decnet_dst_gc_interval * HZ);
210}
211
212static int dn_dst_gc(struct dst_ops *ops)
213{
214 struct dn_route *rt;
215 struct dn_route __rcu **rtp;
216 int i;
217 unsigned long now = jiffies;
218 unsigned long expire = 10 * HZ;
219
220 for (i = 0; i <= dn_rt_hash_mask; i++) {
221
222 spin_lock_bh(&dn_rt_hash_table[i].lock);
223 rtp = &dn_rt_hash_table[i].chain;
224
225 while ((rt = rcu_dereference_protected(*rtp,
226 lockdep_is_held(&dn_rt_hash_table[i].lock))) != NULL) {
227 if (atomic_read(&rt->dst.__refcnt) > 1 ||
228 (now - rt->dst.lastuse) < expire) {
229 rtp = &rt->dn_next;
230 continue;
231 }
232 *rtp = rt->dn_next;
233 rt->dn_next = NULL;
234 dst_dev_put(&rt->dst);
235 dst_release(&rt->dst);
236 break;
237 }
238 spin_unlock_bh(&dn_rt_hash_table[i].lock);
239 }
240
241 return 0;
242}
243
244/*
245 * The decnet standards don't impose a particular minimum mtu, what they
246 * do insist on is that the routing layer accepts a datagram of at least
247 * 230 bytes long. Here we have to subtract the routing header length from
248 * 230 to get the minimum acceptable mtu. If there is no neighbour, then we
249 * assume the worst and use a long header size.
250 *
251 * We update both the mtu and the advertised mss (i.e. the segment size we
252 * advertise to the other end).
253 */
254static void dn_dst_update_pmtu(struct dst_entry *dst, struct sock *sk,
255 struct sk_buff *skb, u32 mtu,
256 bool confirm_neigh)
257{
258 struct dn_route *rt = (struct dn_route *) dst;
259 struct neighbour *n = rt->n;
260 u32 min_mtu = 230;
261 struct dn_dev *dn;
262
263 dn = n ? rcu_dereference_raw(n->dev->dn_ptr) : NULL;
264
265 if (dn && dn->use_long == 0)
266 min_mtu -= 6;
267 else
268 min_mtu -= 21;
269
270 if (dst_metric(dst, RTAX_MTU) > mtu && mtu >= min_mtu) {
271 if (!(dst_metric_locked(dst, RTAX_MTU))) {
272 dst_metric_set(dst, RTAX_MTU, mtu);
273 dst_set_expires(dst, dn_rt_mtu_expires);
274 }
275 if (!(dst_metric_locked(dst, RTAX_ADVMSS))) {
276 u32 mss = mtu - DN_MAX_NSP_DATA_HEADER;
277 u32 existing_mss = dst_metric_raw(dst, RTAX_ADVMSS);
278 if (!existing_mss || existing_mss > mss)
279 dst_metric_set(dst, RTAX_ADVMSS, mss);
280 }
281 }
282}
283
284static void dn_dst_redirect(struct dst_entry *dst, struct sock *sk,
285 struct sk_buff *skb)
286{
287}
288
289/*
290 * When a route has been marked obsolete. (e.g. routing cache flush)
291 */
292static struct dst_entry *dn_dst_check(struct dst_entry *dst, __u32 cookie)
293{
294 return NULL;
295}
296
297static struct dst_entry *dn_dst_negative_advice(struct dst_entry *dst)
298{
299 dst_release(dst);
300 return NULL;
301}
302
303static void dn_dst_link_failure(struct sk_buff *skb)
304{
305}
306
307static inline int compare_keys(struct flowidn *fl1, struct flowidn *fl2)
308{
309 return ((fl1->daddr ^ fl2->daddr) |
310 (fl1->saddr ^ fl2->saddr) |
311 (fl1->flowidn_mark ^ fl2->flowidn_mark) |
312 (fl1->flowidn_scope ^ fl2->flowidn_scope) |
313 (fl1->flowidn_oif ^ fl2->flowidn_oif) |
314 (fl1->flowidn_iif ^ fl2->flowidn_iif)) == 0;
315}
316
317static int dn_insert_route(struct dn_route *rt, unsigned int hash, struct dn_route **rp)
318{
319 struct dn_route *rth;
320 struct dn_route __rcu **rthp;
321 unsigned long now = jiffies;
322
323 rthp = &dn_rt_hash_table[hash].chain;
324
325 spin_lock_bh(&dn_rt_hash_table[hash].lock);
326 while ((rth = rcu_dereference_protected(*rthp,
327 lockdep_is_held(&dn_rt_hash_table[hash].lock))) != NULL) {
328 if (compare_keys(&rth->fld, &rt->fld)) {
329 /* Put it first */
330 *rthp = rth->dn_next;
331 rcu_assign_pointer(rth->dn_next,
332 dn_rt_hash_table[hash].chain);
333 rcu_assign_pointer(dn_rt_hash_table[hash].chain, rth);
334
335 dst_hold_and_use(&rth->dst, now);
336 spin_unlock_bh(&dn_rt_hash_table[hash].lock);
337
338 dst_release_immediate(&rt->dst);
339 *rp = rth;
340 return 0;
341 }
342 rthp = &rth->dn_next;
343 }
344
345 rcu_assign_pointer(rt->dn_next, dn_rt_hash_table[hash].chain);
346 rcu_assign_pointer(dn_rt_hash_table[hash].chain, rt);
347
348 dst_hold_and_use(&rt->dst, now);
349 spin_unlock_bh(&dn_rt_hash_table[hash].lock);
350 *rp = rt;
351 return 0;
352}
353
354static void dn_run_flush(struct timer_list *unused)
355{
356 int i;
357 struct dn_route *rt, *next;
358
359 for (i = 0; i < dn_rt_hash_mask; i++) {
360 spin_lock_bh(&dn_rt_hash_table[i].lock);
361
362 if ((rt = xchg((struct dn_route **)&dn_rt_hash_table[i].chain, NULL)) == NULL)
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 if ((skb = skb_unshare(skb, GFP_ATOMIC)) == NULL)
429 return NET_RX_DROP;
430
431 cb = DN_SKB_CB(skb);
432 /* Skip packet length and point to flags */
433 ptr = skb->data + 2;
434 *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
435
436 dst = (__le16 *)ptr;
437 ptr += 2;
438 src = (__le16 *)ptr;
439 ptr += 2;
440 *ptr = 0; /* Zero hop count */
441
442 swap(*src, *dst);
443
444 skb->pkt_type = PACKET_OUTGOING;
445 dn_rt_finish_output(skb, NULL, NULL);
446 return NET_RX_SUCCESS;
447}
448
449/**
450 * dn_return_long - Return a long packet to its sender
451 * @skb: The long format packet to return
452 *
453 */
454static int dn_return_long(struct sk_buff *skb)
455{
456 struct dn_skb_cb *cb;
457 unsigned char *ptr;
458 unsigned char *src_addr, *dst_addr;
459 unsigned char tmp[ETH_ALEN];
460
461 /* Add back all headers */
462 skb_push(skb, skb->data - skb_network_header(skb));
463
464 if ((skb = skb_unshare(skb, GFP_ATOMIC)) == NULL)
465 return NET_RX_DROP;
466
467 cb = DN_SKB_CB(skb);
468 /* Ignore packet length and point to flags */
469 ptr = skb->data + 2;
470
471 /* Skip padding */
472 if (*ptr & DN_RT_F_PF) {
473 char padlen = (*ptr & ~DN_RT_F_PF);
474 ptr += padlen;
475 }
476
477 *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
478 ptr += 2;
479 dst_addr = ptr;
480 ptr += 8;
481 src_addr = ptr;
482 ptr += 6;
483 *ptr = 0; /* Zero hop count */
484
485 /* Swap source and destination */
486 memcpy(tmp, src_addr, ETH_ALEN);
487 memcpy(src_addr, dst_addr, ETH_ALEN);
488 memcpy(dst_addr, tmp, ETH_ALEN);
489
490 skb->pkt_type = PACKET_OUTGOING;
491 dn_rt_finish_output(skb, dst_addr, src_addr);
492 return NET_RX_SUCCESS;
493}
494
495/**
496 * dn_route_rx_packet - Try and find a route for an incoming packet
497 * @net: The applicable net namespace
498 * @sk: Socket packet transmitted on
499 * @skb: The packet to find a route for
500 *
501 * Returns: result of input function if route is found, error code otherwise
502 */
503static int dn_route_rx_packet(struct net *net, struct sock *sk, struct sk_buff *skb)
504{
505 struct dn_skb_cb *cb;
506 int err;
507
508 if ((err = dn_route_input(skb)) == 0)
509 return dst_input(skb);
510
511 cb = DN_SKB_CB(skb);
512 if (decnet_debug_level & 4) {
513 char *devname = skb->dev ? skb->dev->name : "???";
514
515 printk(KERN_DEBUG
516 "DECnet: dn_route_rx_packet: rt_flags=0x%02x dev=%s len=%d src=0x%04hx dst=0x%04hx err=%d type=%d\n",
517 (int)cb->rt_flags, devname, skb->len,
518 le16_to_cpu(cb->src), le16_to_cpu(cb->dst),
519 err, skb->pkt_type);
520 }
521
522 if ((skb->pkt_type == PACKET_HOST) && (cb->rt_flags & DN_RT_F_RQR)) {
523 switch (cb->rt_flags & DN_RT_PKT_MSK) {
524 case DN_RT_PKT_SHORT:
525 return dn_return_short(skb);
526 case DN_RT_PKT_LONG:
527 return dn_return_long(skb);
528 }
529 }
530
531 kfree_skb(skb);
532 return NET_RX_DROP;
533}
534
535static int dn_route_rx_long(struct sk_buff *skb)
536{
537 struct dn_skb_cb *cb = DN_SKB_CB(skb);
538 unsigned char *ptr = skb->data;
539
540 if (!pskb_may_pull(skb, 21)) /* 20 for long header, 1 for shortest nsp */
541 goto drop_it;
542
543 skb_pull(skb, 20);
544 skb_reset_transport_header(skb);
545
546 /* Destination info */
547 ptr += 2;
548 cb->dst = dn_eth2dn(ptr);
549 if (memcmp(ptr, dn_hiord_addr, 4) != 0)
550 goto drop_it;
551 ptr += 6;
552
553
554 /* Source info */
555 ptr += 2;
556 cb->src = dn_eth2dn(ptr);
557 if (memcmp(ptr, dn_hiord_addr, 4) != 0)
558 goto drop_it;
559 ptr += 6;
560 /* Other junk */
561 ptr++;
562 cb->hops = *ptr++; /* Visit Count */
563
564 return NF_HOOK(NFPROTO_DECNET, NF_DN_PRE_ROUTING,
565 &init_net, NULL, skb, skb->dev, NULL,
566 dn_route_rx_packet);
567
568drop_it:
569 kfree_skb(skb);
570 return NET_RX_DROP;
571}
572
573
574
575static int dn_route_rx_short(struct sk_buff *skb)
576{
577 struct dn_skb_cb *cb = DN_SKB_CB(skb);
578 unsigned char *ptr = skb->data;
579
580 if (!pskb_may_pull(skb, 6)) /* 5 for short header + 1 for shortest nsp */
581 goto drop_it;
582
583 skb_pull(skb, 5);
584 skb_reset_transport_header(skb);
585
586 cb->dst = *(__le16 *)ptr;
587 ptr += 2;
588 cb->src = *(__le16 *)ptr;
589 ptr += 2;
590 cb->hops = *ptr & 0x3f;
591
592 return NF_HOOK(NFPROTO_DECNET, NF_DN_PRE_ROUTING,
593 &init_net, NULL, skb, skb->dev, NULL,
594 dn_route_rx_packet);
595
596drop_it:
597 kfree_skb(skb);
598 return NET_RX_DROP;
599}
600
601static int dn_route_discard(struct net *net, struct sock *sk, struct sk_buff *skb)
602{
603 /*
604 * I know we drop the packet here, but thats considered success in
605 * this case
606 */
607 kfree_skb(skb);
608 return NET_RX_SUCCESS;
609}
610
611static int dn_route_ptp_hello(struct net *net, struct sock *sk, struct sk_buff *skb)
612{
613 dn_dev_hello(skb);
614 dn_neigh_pointopoint_hello(skb);
615 return NET_RX_SUCCESS;
616}
617
618int dn_route_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
619{
620 struct dn_skb_cb *cb;
621 unsigned char flags = 0;
622 __u16 len = le16_to_cpu(*(__le16 *)skb->data);
623 struct dn_dev *dn = rcu_dereference(dev->dn_ptr);
624 unsigned char padlen = 0;
625
626 if (!net_eq(dev_net(dev), &init_net))
627 goto dump_it;
628
629 if (dn == NULL)
630 goto dump_it;
631
632 if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
633 goto out;
634
635 if (!pskb_may_pull(skb, 3))
636 goto dump_it;
637
638 skb_pull(skb, 2);
639
640 if (len > skb->len)
641 goto dump_it;
642
643 skb_trim(skb, len);
644
645 flags = *skb->data;
646
647 cb = DN_SKB_CB(skb);
648 cb->stamp = jiffies;
649 cb->iif = dev->ifindex;
650
651 /*
652 * If we have padding, remove it.
653 */
654 if (flags & DN_RT_F_PF) {
655 padlen = flags & ~DN_RT_F_PF;
656 if (!pskb_may_pull(skb, padlen + 1))
657 goto dump_it;
658 skb_pull(skb, padlen);
659 flags = *skb->data;
660 }
661
662 skb_reset_network_header(skb);
663
664 /*
665 * Weed out future version DECnet
666 */
667 if (flags & DN_RT_F_VER)
668 goto dump_it;
669
670 cb->rt_flags = flags;
671
672 if (decnet_debug_level & 1)
673 printk(KERN_DEBUG
674 "dn_route_rcv: got 0x%02x from %s [%d %d %d]\n",
675 (int)flags, dev->name, len, skb->len,
676 padlen);
677
678 if (flags & DN_RT_PKT_CNTL) {
679 if (unlikely(skb_linearize(skb)))
680 goto dump_it;
681
682 switch (flags & DN_RT_CNTL_MSK) {
683 case DN_RT_PKT_INIT:
684 dn_dev_init_pkt(skb);
685 break;
686 case DN_RT_PKT_VERI:
687 dn_dev_veri_pkt(skb);
688 break;
689 }
690
691 if (dn->parms.state != DN_DEV_S_RU)
692 goto dump_it;
693
694 switch (flags & DN_RT_CNTL_MSK) {
695 case DN_RT_PKT_HELO:
696 return NF_HOOK(NFPROTO_DECNET, NF_DN_HELLO,
697 &init_net, NULL, skb, skb->dev, NULL,
698 dn_route_ptp_hello);
699
700 case DN_RT_PKT_L1RT:
701 case DN_RT_PKT_L2RT:
702 return NF_HOOK(NFPROTO_DECNET, NF_DN_ROUTE,
703 &init_net, NULL, skb, skb->dev, NULL,
704 dn_route_discard);
705 case DN_RT_PKT_ERTH:
706 return NF_HOOK(NFPROTO_DECNET, NF_DN_HELLO,
707 &init_net, NULL, skb, skb->dev, NULL,
708 dn_neigh_router_hello);
709
710 case DN_RT_PKT_EEDH:
711 return NF_HOOK(NFPROTO_DECNET, NF_DN_HELLO,
712 &init_net, NULL, skb, skb->dev, NULL,
713 dn_neigh_endnode_hello);
714 }
715 } else {
716 if (dn->parms.state != DN_DEV_S_RU)
717 goto dump_it;
718
719 skb_pull(skb, 1); /* Pull flags */
720
721 switch (flags & DN_RT_PKT_MSK) {
722 case DN_RT_PKT_LONG:
723 return dn_route_rx_long(skb);
724 case DN_RT_PKT_SHORT:
725 return dn_route_rx_short(skb);
726 }
727 }
728
729dump_it:
730 kfree_skb(skb);
731out:
732 return NET_RX_DROP;
733}
734
735static int dn_output(struct net *net, struct sock *sk, struct sk_buff *skb)
736{
737 struct dst_entry *dst = skb_dst(skb);
738 struct dn_route *rt = (struct dn_route *)dst;
739 struct net_device *dev = dst->dev;
740 struct dn_skb_cb *cb = DN_SKB_CB(skb);
741
742 int err = -EINVAL;
743
744 if (rt->n == NULL)
745 goto error;
746
747 skb->dev = dev;
748
749 cb->src = rt->rt_saddr;
750 cb->dst = rt->rt_daddr;
751
752 /*
753 * Always set the Intra-Ethernet bit on all outgoing packets
754 * originated on this node. Only valid flag from upper layers
755 * is return-to-sender-requested. Set hop count to 0 too.
756 */
757 cb->rt_flags &= ~DN_RT_F_RQR;
758 cb->rt_flags |= DN_RT_F_IE;
759 cb->hops = 0;
760
761 return NF_HOOK(NFPROTO_DECNET, NF_DN_LOCAL_OUT,
762 &init_net, sk, skb, NULL, dev,
763 dn_to_neigh_output);
764
765error:
766 net_dbg_ratelimited("dn_output: This should not happen\n");
767
768 kfree_skb(skb);
769
770 return err;
771}
772
773static int dn_forward(struct sk_buff *skb)
774{
775 struct dn_skb_cb *cb = DN_SKB_CB(skb);
776 struct dst_entry *dst = skb_dst(skb);
777 struct dn_dev *dn_db = rcu_dereference(dst->dev->dn_ptr);
778 struct dn_route *rt;
779 int header_len;
780 struct net_device *dev = skb->dev;
781
782 if (skb->pkt_type != PACKET_HOST)
783 goto drop;
784
785 /* Ensure that we have enough space for headers */
786 rt = (struct dn_route *)skb_dst(skb);
787 header_len = dn_db->use_long ? 21 : 6;
788 if (skb_cow(skb, LL_RESERVED_SPACE(rt->dst.dev)+header_len))
789 goto drop;
790
791 /*
792 * Hop count exceeded.
793 */
794 if (++cb->hops > 30)
795 goto drop;
796
797 skb->dev = rt->dst.dev;
798
799 /*
800 * If packet goes out same interface it came in on, then set
801 * the Intra-Ethernet bit. This has no effect for short
802 * packets, so we don't need to test for them here.
803 */
804 cb->rt_flags &= ~DN_RT_F_IE;
805 if (rt->rt_flags & RTCF_DOREDIRECT)
806 cb->rt_flags |= DN_RT_F_IE;
807
808 return NF_HOOK(NFPROTO_DECNET, NF_DN_FORWARD,
809 &init_net, NULL, skb, dev, skb->dev,
810 dn_to_neigh_output);
811
812drop:
813 kfree_skb(skb);
814 return NET_RX_DROP;
815}
816
817/*
818 * Used to catch bugs. This should never normally get
819 * called.
820 */
821static int dn_rt_bug_out(struct net *net, struct sock *sk, struct sk_buff *skb)
822{
823 struct dn_skb_cb *cb = DN_SKB_CB(skb);
824
825 net_dbg_ratelimited("dn_rt_bug: skb from:%04x to:%04x\n",
826 le16_to_cpu(cb->src), le16_to_cpu(cb->dst));
827
828 kfree_skb(skb);
829
830 return NET_RX_DROP;
831}
832
833static int dn_rt_bug(struct sk_buff *skb)
834{
835 struct dn_skb_cb *cb = DN_SKB_CB(skb);
836
837 net_dbg_ratelimited("dn_rt_bug: skb from:%04x to:%04x\n",
838 le16_to_cpu(cb->src), le16_to_cpu(cb->dst));
839
840 kfree_skb(skb);
841
842 return NET_RX_DROP;
843}
844
845static unsigned int dn_dst_default_advmss(const struct dst_entry *dst)
846{
847 return dn_mss_from_pmtu(dst->dev, dst_mtu(dst));
848}
849
850static unsigned int dn_dst_mtu(const struct dst_entry *dst)
851{
852 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
853
854 return mtu ? : dst->dev->mtu;
855}
856
857static struct neighbour *dn_dst_neigh_lookup(const struct dst_entry *dst,
858 struct sk_buff *skb,
859 const void *daddr)
860{
861 return __neigh_lookup_errno(&dn_neigh_table, daddr, dst->dev);
862}
863
864static int dn_rt_set_next_hop(struct dn_route *rt, struct dn_fib_res *res)
865{
866 struct dn_fib_info *fi = res->fi;
867 struct net_device *dev = rt->dst.dev;
868 unsigned int mss_metric;
869 struct neighbour *n;
870
871 if (fi) {
872 if (DN_FIB_RES_GW(*res) &&
873 DN_FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
874 rt->rt_gateway = DN_FIB_RES_GW(*res);
875 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
876 }
877 rt->rt_type = res->type;
878
879 if (dev != NULL && rt->n == NULL) {
880 n = __neigh_lookup_errno(&dn_neigh_table, &rt->rt_gateway, dev);
881 if (IS_ERR(n))
882 return PTR_ERR(n);
883 rt->n = n;
884 }
885
886 if (dst_metric(&rt->dst, RTAX_MTU) > rt->dst.dev->mtu)
887 dst_metric_set(&rt->dst, RTAX_MTU, rt->dst.dev->mtu);
888 mss_metric = dst_metric_raw(&rt->dst, RTAX_ADVMSS);
889 if (mss_metric) {
890 unsigned int mss = dn_mss_from_pmtu(dev, dst_mtu(&rt->dst));
891 if (mss_metric > mss)
892 dst_metric_set(&rt->dst, RTAX_ADVMSS, mss);
893 }
894 return 0;
895}
896
897static inline int dn_match_addr(__le16 addr1, __le16 addr2)
898{
899 __u16 tmp = le16_to_cpu(addr1) ^ le16_to_cpu(addr2);
900 int match = 16;
901 while(tmp) {
902 tmp >>= 1;
903 match--;
904 }
905 return match;
906}
907
908static __le16 dnet_select_source(const struct net_device *dev, __le16 daddr, int scope)
909{
910 __le16 saddr = 0;
911 struct dn_dev *dn_db;
912 struct dn_ifaddr *ifa;
913 int best_match = 0;
914 int ret;
915
916 rcu_read_lock();
917 dn_db = rcu_dereference(dev->dn_ptr);
918 for (ifa = rcu_dereference(dn_db->ifa_list);
919 ifa != NULL;
920 ifa = rcu_dereference(ifa->ifa_next)) {
921 if (ifa->ifa_scope > scope)
922 continue;
923 if (!daddr) {
924 saddr = ifa->ifa_local;
925 break;
926 }
927 ret = dn_match_addr(daddr, ifa->ifa_local);
928 if (ret > best_match)
929 saddr = ifa->ifa_local;
930 if (best_match == 0)
931 saddr = ifa->ifa_local;
932 }
933 rcu_read_unlock();
934
935 return saddr;
936}
937
938static inline __le16 __dn_fib_res_prefsrc(struct dn_fib_res *res)
939{
940 return dnet_select_source(DN_FIB_RES_DEV(*res), DN_FIB_RES_GW(*res), res->scope);
941}
942
943static inline __le16 dn_fib_rules_map_destination(__le16 daddr, struct dn_fib_res *res)
944{
945 __le16 mask = dnet_make_mask(res->prefixlen);
946 return (daddr&~mask)|res->fi->fib_nh->nh_gw;
947}
948
949static int dn_route_output_slow(struct dst_entry **pprt, const struct flowidn *oldflp, int try_hard)
950{
951 struct flowidn fld = {
952 .daddr = oldflp->daddr,
953 .saddr = oldflp->saddr,
954 .flowidn_scope = RT_SCOPE_UNIVERSE,
955 .flowidn_mark = oldflp->flowidn_mark,
956 .flowidn_iif = LOOPBACK_IFINDEX,
957 .flowidn_oif = oldflp->flowidn_oif,
958 };
959 struct dn_route *rt = NULL;
960 struct net_device *dev_out = NULL, *dev;
961 struct neighbour *neigh = NULL;
962 unsigned int hash;
963 unsigned int flags = 0;
964 struct dn_fib_res res = { .fi = NULL, .type = RTN_UNICAST };
965 int err;
966 int free_res = 0;
967 __le16 gateway = 0;
968
969 if (decnet_debug_level & 16)
970 printk(KERN_DEBUG
971 "dn_route_output_slow: dst=%04x src=%04x mark=%d"
972 " iif=%d oif=%d\n", le16_to_cpu(oldflp->daddr),
973 le16_to_cpu(oldflp->saddr),
974 oldflp->flowidn_mark, LOOPBACK_IFINDEX,
975 oldflp->flowidn_oif);
976
977 /* If we have an output interface, verify its a DECnet device */
978 if (oldflp->flowidn_oif) {
979 dev_out = dev_get_by_index(&init_net, oldflp->flowidn_oif);
980 err = -ENODEV;
981 if (dev_out && dev_out->dn_ptr == NULL) {
982 dev_put(dev_out);
983 dev_out = NULL;
984 }
985 if (dev_out == NULL)
986 goto out;
987 }
988
989 /* If we have a source address, verify that its a local address */
990 if (oldflp->saddr) {
991 err = -EADDRNOTAVAIL;
992
993 if (dev_out) {
994 if (dn_dev_islocal(dev_out, oldflp->saddr))
995 goto source_ok;
996 dev_put(dev_out);
997 goto out;
998 }
999 rcu_read_lock();
1000 for_each_netdev_rcu(&init_net, dev) {
1001 if (!dev->dn_ptr)
1002 continue;
1003 if (!dn_dev_islocal(dev, oldflp->saddr))
1004 continue;
1005 if ((dev->flags & IFF_LOOPBACK) &&
1006 oldflp->daddr &&
1007 !dn_dev_islocal(dev, oldflp->daddr))
1008 continue;
1009
1010 dev_out = dev;
1011 break;
1012 }
1013 rcu_read_unlock();
1014 if (dev_out == NULL)
1015 goto out;
1016 dev_hold(dev_out);
1017source_ok:
1018 ;
1019 }
1020
1021 /* No destination? Assume its local */
1022 if (!fld.daddr) {
1023 fld.daddr = fld.saddr;
1024
1025 if (dev_out)
1026 dev_put(dev_out);
1027 err = -EINVAL;
1028 dev_out = init_net.loopback_dev;
1029 if (!dev_out->dn_ptr)
1030 goto out;
1031 err = -EADDRNOTAVAIL;
1032 dev_hold(dev_out);
1033 if (!fld.daddr) {
1034 fld.daddr =
1035 fld.saddr = dnet_select_source(dev_out, 0,
1036 RT_SCOPE_HOST);
1037 if (!fld.daddr)
1038 goto out;
1039 }
1040 fld.flowidn_oif = LOOPBACK_IFINDEX;
1041 res.type = RTN_LOCAL;
1042 goto make_route;
1043 }
1044
1045 if (decnet_debug_level & 16)
1046 printk(KERN_DEBUG
1047 "dn_route_output_slow: initial checks complete."
1048 " dst=%04x src=%04x oif=%d try_hard=%d\n",
1049 le16_to_cpu(fld.daddr), le16_to_cpu(fld.saddr),
1050 fld.flowidn_oif, try_hard);
1051
1052 /*
1053 * N.B. If the kernel is compiled without router support then
1054 * dn_fib_lookup() will evaluate to non-zero so this if () block
1055 * will always be executed.
1056 */
1057 err = -ESRCH;
1058 if (try_hard || (err = dn_fib_lookup(&fld, &res)) != 0) {
1059 struct dn_dev *dn_db;
1060 if (err != -ESRCH)
1061 goto out;
1062 /*
1063 * Here the fallback is basically the standard algorithm for
1064 * routing in endnodes which is described in the DECnet routing
1065 * docs
1066 *
1067 * If we are not trying hard, look in neighbour cache.
1068 * The result is tested to ensure that if a specific output
1069 * device/source address was requested, then we honour that
1070 * here
1071 */
1072 if (!try_hard) {
1073 neigh = neigh_lookup_nodev(&dn_neigh_table, &init_net, &fld.daddr);
1074 if (neigh) {
1075 if ((oldflp->flowidn_oif &&
1076 (neigh->dev->ifindex != oldflp->flowidn_oif)) ||
1077 (oldflp->saddr &&
1078 (!dn_dev_islocal(neigh->dev,
1079 oldflp->saddr)))) {
1080 neigh_release(neigh);
1081 neigh = NULL;
1082 } else {
1083 if (dev_out)
1084 dev_put(dev_out);
1085 if (dn_dev_islocal(neigh->dev, fld.daddr)) {
1086 dev_out = init_net.loopback_dev;
1087 res.type = RTN_LOCAL;
1088 } else {
1089 dev_out = neigh->dev;
1090 }
1091 dev_hold(dev_out);
1092 goto select_source;
1093 }
1094 }
1095 }
1096
1097 /* Not there? Perhaps its a local address */
1098 if (dev_out == NULL)
1099 dev_out = dn_dev_get_default();
1100 err = -ENODEV;
1101 if (dev_out == NULL)
1102 goto out;
1103 dn_db = rcu_dereference_raw(dev_out->dn_ptr);
1104 if (!dn_db)
1105 goto e_inval;
1106 /* Possible improvement - check all devices for local addr */
1107 if (dn_dev_islocal(dev_out, fld.daddr)) {
1108 dev_put(dev_out);
1109 dev_out = init_net.loopback_dev;
1110 dev_hold(dev_out);
1111 res.type = RTN_LOCAL;
1112 goto select_source;
1113 }
1114 /* Not local either.... try sending it to the default router */
1115 neigh = neigh_clone(dn_db->router);
1116 BUG_ON(neigh && neigh->dev != dev_out);
1117
1118 /* Ok then, we assume its directly connected and move on */
1119select_source:
1120 if (neigh)
1121 gateway = ((struct dn_neigh *)neigh)->addr;
1122 if (gateway == 0)
1123 gateway = fld.daddr;
1124 if (fld.saddr == 0) {
1125 fld.saddr = dnet_select_source(dev_out, gateway,
1126 res.type == RTN_LOCAL ?
1127 RT_SCOPE_HOST :
1128 RT_SCOPE_LINK);
1129 if (fld.saddr == 0 && res.type != RTN_LOCAL)
1130 goto e_addr;
1131 }
1132 fld.flowidn_oif = dev_out->ifindex;
1133 goto make_route;
1134 }
1135 free_res = 1;
1136
1137 if (res.type == RTN_NAT)
1138 goto e_inval;
1139
1140 if (res.type == RTN_LOCAL) {
1141 if (!fld.saddr)
1142 fld.saddr = fld.daddr;
1143 if (dev_out)
1144 dev_put(dev_out);
1145 dev_out = init_net.loopback_dev;
1146 dev_hold(dev_out);
1147 if (!dev_out->dn_ptr)
1148 goto e_inval;
1149 fld.flowidn_oif = dev_out->ifindex;
1150 if (res.fi)
1151 dn_fib_info_put(res.fi);
1152 res.fi = NULL;
1153 goto make_route;
1154 }
1155
1156 if (res.fi->fib_nhs > 1 && fld.flowidn_oif == 0)
1157 dn_fib_select_multipath(&fld, &res);
1158
1159 /*
1160 * We could add some logic to deal with default routes here and
1161 * get rid of some of the special casing above.
1162 */
1163
1164 if (!fld.saddr)
1165 fld.saddr = DN_FIB_RES_PREFSRC(res);
1166
1167 if (dev_out)
1168 dev_put(dev_out);
1169 dev_out = DN_FIB_RES_DEV(res);
1170 dev_hold(dev_out);
1171 fld.flowidn_oif = dev_out->ifindex;
1172 gateway = DN_FIB_RES_GW(res);
1173
1174make_route:
1175 if (dev_out->flags & IFF_LOOPBACK)
1176 flags |= RTCF_LOCAL;
1177
1178 rt = dst_alloc(&dn_dst_ops, dev_out, 0, DST_OBSOLETE_NONE, 0);
1179 if (rt == NULL)
1180 goto e_nobufs;
1181
1182 rt->dn_next = NULL;
1183 memset(&rt->fld, 0, sizeof(rt->fld));
1184 rt->fld.saddr = oldflp->saddr;
1185 rt->fld.daddr = oldflp->daddr;
1186 rt->fld.flowidn_oif = oldflp->flowidn_oif;
1187 rt->fld.flowidn_iif = 0;
1188 rt->fld.flowidn_mark = oldflp->flowidn_mark;
1189
1190 rt->rt_saddr = fld.saddr;
1191 rt->rt_daddr = fld.daddr;
1192 rt->rt_gateway = gateway ? gateway : fld.daddr;
1193 rt->rt_local_src = fld.saddr;
1194
1195 rt->rt_dst_map = fld.daddr;
1196 rt->rt_src_map = fld.saddr;
1197
1198 rt->n = neigh;
1199 neigh = NULL;
1200
1201 rt->dst.lastuse = jiffies;
1202 rt->dst.output = dn_output;
1203 rt->dst.input = dn_rt_bug;
1204 rt->rt_flags = flags;
1205 if (flags & RTCF_LOCAL)
1206 rt->dst.input = dn_nsp_rx;
1207
1208 err = dn_rt_set_next_hop(rt, &res);
1209 if (err)
1210 goto e_neighbour;
1211
1212 hash = dn_hash(rt->fld.saddr, rt->fld.daddr);
1213 /* dn_insert_route() increments dst->__refcnt */
1214 dn_insert_route(rt, hash, (struct dn_route **)pprt);
1215
1216done:
1217 if (neigh)
1218 neigh_release(neigh);
1219 if (free_res)
1220 dn_fib_res_put(&res);
1221 if (dev_out)
1222 dev_put(dev_out);
1223out:
1224 return err;
1225
1226e_addr:
1227 err = -EADDRNOTAVAIL;
1228 goto done;
1229e_inval:
1230 err = -EINVAL;
1231 goto done;
1232e_nobufs:
1233 err = -ENOBUFS;
1234 goto done;
1235e_neighbour:
1236 dst_release_immediate(&rt->dst);
1237 goto e_nobufs;
1238}
1239
1240
1241/*
1242 * N.B. The flags may be moved into the flowi at some future stage.
1243 */
1244static int __dn_route_output_key(struct dst_entry **pprt, const struct flowidn *flp, int flags)
1245{
1246 unsigned int hash = dn_hash(flp->saddr, flp->daddr);
1247 struct dn_route *rt = NULL;
1248
1249 if (!(flags & MSG_TRYHARD)) {
1250 rcu_read_lock_bh();
1251 for (rt = rcu_dereference_bh(dn_rt_hash_table[hash].chain); rt;
1252 rt = rcu_dereference_bh(rt->dn_next)) {
1253 if ((flp->daddr == rt->fld.daddr) &&
1254 (flp->saddr == rt->fld.saddr) &&
1255 (flp->flowidn_mark == rt->fld.flowidn_mark) &&
1256 dn_is_output_route(rt) &&
1257 (rt->fld.flowidn_oif == flp->flowidn_oif)) {
1258 dst_hold_and_use(&rt->dst, jiffies);
1259 rcu_read_unlock_bh();
1260 *pprt = &rt->dst;
1261 return 0;
1262 }
1263 }
1264 rcu_read_unlock_bh();
1265 }
1266
1267 return dn_route_output_slow(pprt, flp, flags);
1268}
1269
1270static int dn_route_output_key(struct dst_entry **pprt, struct flowidn *flp, int flags)
1271{
1272 int err;
1273
1274 err = __dn_route_output_key(pprt, flp, flags);
1275 if (err == 0 && flp->flowidn_proto) {
1276 *pprt = xfrm_lookup(&init_net, *pprt,
1277 flowidn_to_flowi(flp), NULL, 0);
1278 if (IS_ERR(*pprt)) {
1279 err = PTR_ERR(*pprt);
1280 *pprt = NULL;
1281 }
1282 }
1283 return err;
1284}
1285
1286int dn_route_output_sock(struct dst_entry __rcu **pprt, struct flowidn *fl, struct sock *sk, int flags)
1287{
1288 int err;
1289
1290 err = __dn_route_output_key(pprt, fl, flags & MSG_TRYHARD);
1291 if (err == 0 && fl->flowidn_proto) {
1292 *pprt = xfrm_lookup(&init_net, *pprt,
1293 flowidn_to_flowi(fl), sk, 0);
1294 if (IS_ERR(*pprt)) {
1295 err = PTR_ERR(*pprt);
1296 *pprt = NULL;
1297 }
1298 }
1299 return err;
1300}
1301
1302static int dn_route_input_slow(struct sk_buff *skb)
1303{
1304 struct dn_route *rt = NULL;
1305 struct dn_skb_cb *cb = DN_SKB_CB(skb);
1306 struct net_device *in_dev = skb->dev;
1307 struct net_device *out_dev = NULL;
1308 struct dn_dev *dn_db;
1309 struct neighbour *neigh = NULL;
1310 unsigned int hash;
1311 int flags = 0;
1312 __le16 gateway = 0;
1313 __le16 local_src = 0;
1314 struct flowidn fld = {
1315 .daddr = cb->dst,
1316 .saddr = cb->src,
1317 .flowidn_scope = RT_SCOPE_UNIVERSE,
1318 .flowidn_mark = skb->mark,
1319 .flowidn_iif = skb->dev->ifindex,
1320 };
1321 struct dn_fib_res res = { .fi = NULL, .type = RTN_UNREACHABLE };
1322 int err = -EINVAL;
1323 int free_res = 0;
1324
1325 dev_hold(in_dev);
1326
1327 if ((dn_db = rcu_dereference(in_dev->dn_ptr)) == NULL)
1328 goto out;
1329
1330 /* Zero source addresses are not allowed */
1331 if (fld.saddr == 0)
1332 goto out;
1333
1334 /*
1335 * In this case we've just received a packet from a source
1336 * outside ourselves pretending to come from us. We don't
1337 * allow it any further to prevent routing loops, spoofing and
1338 * other nasties. Loopback packets already have the dst attached
1339 * so this only affects packets which have originated elsewhere.
1340 */
1341 err = -ENOTUNIQ;
1342 if (dn_dev_islocal(in_dev, cb->src))
1343 goto out;
1344
1345 err = dn_fib_lookup(&fld, &res);
1346 if (err) {
1347 if (err != -ESRCH)
1348 goto out;
1349 /*
1350 * Is the destination us ?
1351 */
1352 if (!dn_dev_islocal(in_dev, cb->dst))
1353 goto e_inval;
1354
1355 res.type = RTN_LOCAL;
1356 } else {
1357 __le16 src_map = fld.saddr;
1358 free_res = 1;
1359
1360 out_dev = DN_FIB_RES_DEV(res);
1361 if (out_dev == NULL) {
1362 net_crit_ratelimited("Bug in dn_route_input_slow() No output device\n");
1363 goto e_inval;
1364 }
1365 dev_hold(out_dev);
1366
1367 if (res.r)
1368 src_map = fld.saddr; /* no NAT support for now */
1369
1370 gateway = DN_FIB_RES_GW(res);
1371 if (res.type == RTN_NAT) {
1372 fld.daddr = dn_fib_rules_map_destination(fld.daddr, &res);
1373 dn_fib_res_put(&res);
1374 free_res = 0;
1375 if (dn_fib_lookup(&fld, &res))
1376 goto e_inval;
1377 free_res = 1;
1378 if (res.type != RTN_UNICAST)
1379 goto e_inval;
1380 flags |= RTCF_DNAT;
1381 gateway = fld.daddr;
1382 }
1383 fld.saddr = src_map;
1384 }
1385
1386 switch(res.type) {
1387 case RTN_UNICAST:
1388 /*
1389 * Forwarding check here, we only check for forwarding
1390 * being turned off, if you want to only forward intra
1391 * area, its up to you to set the routing tables up
1392 * correctly.
1393 */
1394 if (dn_db->parms.forwarding == 0)
1395 goto e_inval;
1396
1397 if (res.fi->fib_nhs > 1 && fld.flowidn_oif == 0)
1398 dn_fib_select_multipath(&fld, &res);
1399
1400 /*
1401 * Check for out_dev == in_dev. We use the RTCF_DOREDIRECT
1402 * flag as a hint to set the intra-ethernet bit when
1403 * forwarding. If we've got NAT in operation, we don't do
1404 * this optimisation.
1405 */
1406 if (out_dev == in_dev && !(flags & RTCF_NAT))
1407 flags |= RTCF_DOREDIRECT;
1408
1409 local_src = DN_FIB_RES_PREFSRC(res);
1410
1411 case RTN_BLACKHOLE:
1412 case RTN_UNREACHABLE:
1413 break;
1414 case RTN_LOCAL:
1415 flags |= RTCF_LOCAL;
1416 fld.saddr = cb->dst;
1417 fld.daddr = cb->src;
1418
1419 /* Routing tables gave us a gateway */
1420 if (gateway)
1421 goto make_route;
1422
1423 /* Packet was intra-ethernet, so we know its on-link */
1424 if (cb->rt_flags & DN_RT_F_IE) {
1425 gateway = cb->src;
1426 goto make_route;
1427 }
1428
1429 /* Use the default router if there is one */
1430 neigh = neigh_clone(dn_db->router);
1431 if (neigh) {
1432 gateway = ((struct dn_neigh *)neigh)->addr;
1433 goto make_route;
1434 }
1435
1436 /* Close eyes and pray */
1437 gateway = cb->src;
1438 goto make_route;
1439 default:
1440 goto e_inval;
1441 }
1442
1443make_route:
1444 rt = dst_alloc(&dn_dst_ops, out_dev, 1, DST_OBSOLETE_NONE, 0);
1445 if (rt == NULL)
1446 goto e_nobufs;
1447
1448 rt->dn_next = NULL;
1449 memset(&rt->fld, 0, sizeof(rt->fld));
1450 rt->rt_saddr = fld.saddr;
1451 rt->rt_daddr = fld.daddr;
1452 rt->rt_gateway = fld.daddr;
1453 if (gateway)
1454 rt->rt_gateway = gateway;
1455 rt->rt_local_src = local_src ? local_src : rt->rt_saddr;
1456
1457 rt->rt_dst_map = fld.daddr;
1458 rt->rt_src_map = fld.saddr;
1459
1460 rt->fld.saddr = cb->src;
1461 rt->fld.daddr = cb->dst;
1462 rt->fld.flowidn_oif = 0;
1463 rt->fld.flowidn_iif = in_dev->ifindex;
1464 rt->fld.flowidn_mark = fld.flowidn_mark;
1465
1466 rt->n = neigh;
1467 rt->dst.lastuse = jiffies;
1468 rt->dst.output = dn_rt_bug_out;
1469 switch (res.type) {
1470 case RTN_UNICAST:
1471 rt->dst.input = dn_forward;
1472 break;
1473 case RTN_LOCAL:
1474 rt->dst.output = dn_output;
1475 rt->dst.input = dn_nsp_rx;
1476 rt->dst.dev = in_dev;
1477 flags |= RTCF_LOCAL;
1478 break;
1479 default:
1480 case RTN_UNREACHABLE:
1481 case RTN_BLACKHOLE:
1482 rt->dst.input = dst_discard;
1483 }
1484 rt->rt_flags = flags;
1485
1486 err = dn_rt_set_next_hop(rt, &res);
1487 if (err)
1488 goto e_neighbour;
1489
1490 hash = dn_hash(rt->fld.saddr, rt->fld.daddr);
1491 /* dn_insert_route() increments dst->__refcnt */
1492 dn_insert_route(rt, hash, &rt);
1493 skb_dst_set(skb, &rt->dst);
1494
1495done:
1496 if (neigh)
1497 neigh_release(neigh);
1498 if (free_res)
1499 dn_fib_res_put(&res);
1500 dev_put(in_dev);
1501 if (out_dev)
1502 dev_put(out_dev);
1503out:
1504 return err;
1505
1506e_inval:
1507 err = -EINVAL;
1508 goto done;
1509
1510e_nobufs:
1511 err = -ENOBUFS;
1512 goto done;
1513
1514e_neighbour:
1515 dst_release_immediate(&rt->dst);
1516 goto done;
1517}
1518
1519static int dn_route_input(struct sk_buff *skb)
1520{
1521 struct dn_route *rt;
1522 struct dn_skb_cb *cb = DN_SKB_CB(skb);
1523 unsigned int hash = dn_hash(cb->src, cb->dst);
1524
1525 if (skb_dst(skb))
1526 return 0;
1527
1528 rcu_read_lock();
1529 for(rt = rcu_dereference(dn_rt_hash_table[hash].chain); rt != NULL;
1530 rt = rcu_dereference(rt->dn_next)) {
1531 if ((rt->fld.saddr == cb->src) &&
1532 (rt->fld.daddr == cb->dst) &&
1533 (rt->fld.flowidn_oif == 0) &&
1534 (rt->fld.flowidn_mark == skb->mark) &&
1535 (rt->fld.flowidn_iif == cb->iif)) {
1536 dst_hold_and_use(&rt->dst, jiffies);
1537 rcu_read_unlock();
1538 skb_dst_set(skb, (struct dst_entry *)rt);
1539 return 0;
1540 }
1541 }
1542 rcu_read_unlock();
1543
1544 return dn_route_input_slow(skb);
1545}
1546
1547static int dn_rt_fill_info(struct sk_buff *skb, u32 portid, u32 seq,
1548 int event, int nowait, unsigned int flags)
1549{
1550 struct dn_route *rt = (struct dn_route *)skb_dst(skb);
1551 struct rtmsg *r;
1552 struct nlmsghdr *nlh;
1553 long expires;
1554
1555 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*r), flags);
1556 if (!nlh)
1557 return -EMSGSIZE;
1558
1559 r = nlmsg_data(nlh);
1560 r->rtm_family = AF_DECnet;
1561 r->rtm_dst_len = 16;
1562 r->rtm_src_len = 0;
1563 r->rtm_tos = 0;
1564 r->rtm_table = RT_TABLE_MAIN;
1565 r->rtm_type = rt->rt_type;
1566 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
1567 r->rtm_scope = RT_SCOPE_UNIVERSE;
1568 r->rtm_protocol = RTPROT_UNSPEC;
1569
1570 if (rt->rt_flags & RTCF_NOTIFY)
1571 r->rtm_flags |= RTM_F_NOTIFY;
1572
1573 if (nla_put_u32(skb, RTA_TABLE, RT_TABLE_MAIN) < 0 ||
1574 nla_put_le16(skb, RTA_DST, rt->rt_daddr) < 0)
1575 goto errout;
1576
1577 if (rt->fld.saddr) {
1578 r->rtm_src_len = 16;
1579 if (nla_put_le16(skb, RTA_SRC, rt->fld.saddr) < 0)
1580 goto errout;
1581 }
1582 if (rt->dst.dev &&
1583 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex) < 0)
1584 goto errout;
1585
1586 /*
1587 * Note to self - change this if input routes reverse direction when
1588 * they deal only with inputs and not with replies like they do
1589 * currently.
1590 */
1591 if (nla_put_le16(skb, RTA_PREFSRC, rt->rt_local_src) < 0)
1592 goto errout;
1593
1594 if (rt->rt_daddr != rt->rt_gateway &&
1595 nla_put_le16(skb, RTA_GATEWAY, rt->rt_gateway) < 0)
1596 goto errout;
1597
1598 if (rtnetlink_put_metrics(skb, dst_metrics_ptr(&rt->dst)) < 0)
1599 goto errout;
1600
1601 expires = rt->dst.expires ? rt->dst.expires - jiffies : 0;
1602 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires,
1603 rt->dst.error) < 0)
1604 goto errout;
1605
1606 if (dn_is_input_route(rt) &&
1607 nla_put_u32(skb, RTA_IIF, rt->fld.flowidn_iif) < 0)
1608 goto errout;
1609
1610 nlmsg_end(skb, nlh);
1611 return 0;
1612
1613errout:
1614 nlmsg_cancel(skb, nlh);
1615 return -EMSGSIZE;
1616}
1617
1618const struct nla_policy rtm_dn_policy[RTA_MAX + 1] = {
1619 [RTA_DST] = { .type = NLA_U16 },
1620 [RTA_SRC] = { .type = NLA_U16 },
1621 [RTA_IIF] = { .type = NLA_U32 },
1622 [RTA_OIF] = { .type = NLA_U32 },
1623 [RTA_GATEWAY] = { .type = NLA_U16 },
1624 [RTA_PRIORITY] = { .type = NLA_U32 },
1625 [RTA_PREFSRC] = { .type = NLA_U16 },
1626 [RTA_METRICS] = { .type = NLA_NESTED },
1627 [RTA_MULTIPATH] = { .type = NLA_NESTED },
1628 [RTA_TABLE] = { .type = NLA_U32 },
1629 [RTA_MARK] = { .type = NLA_U32 },
1630};
1631
1632/*
1633 * This is called by both endnodes and routers now.
1634 */
1635static int dn_cache_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
1636 struct netlink_ext_ack *extack)
1637{
1638 struct net *net = sock_net(in_skb->sk);
1639 struct rtmsg *rtm = nlmsg_data(nlh);
1640 struct dn_route *rt = NULL;
1641 struct dn_skb_cb *cb;
1642 int err;
1643 struct sk_buff *skb;
1644 struct flowidn fld;
1645 struct nlattr *tb[RTA_MAX+1];
1646
1647 if (!net_eq(net, &init_net))
1648 return -EINVAL;
1649
1650 err = nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
1651 rtm_dn_policy, extack);
1652 if (err < 0)
1653 return err;
1654
1655 memset(&fld, 0, sizeof(fld));
1656 fld.flowidn_proto = DNPROTO_NSP;
1657
1658 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1659 if (skb == NULL)
1660 return -ENOBUFS;
1661 skb_reset_mac_header(skb);
1662 cb = DN_SKB_CB(skb);
1663
1664 if (tb[RTA_SRC])
1665 fld.saddr = nla_get_le16(tb[RTA_SRC]);
1666
1667 if (tb[RTA_DST])
1668 fld.daddr = nla_get_le16(tb[RTA_DST]);
1669
1670 if (tb[RTA_IIF])
1671 fld.flowidn_iif = nla_get_u32(tb[RTA_IIF]);
1672
1673 if (fld.flowidn_iif) {
1674 struct net_device *dev;
1675 dev = __dev_get_by_index(&init_net, fld.flowidn_iif);
1676 if (!dev || !dev->dn_ptr) {
1677 kfree_skb(skb);
1678 return -ENODEV;
1679 }
1680 skb->protocol = htons(ETH_P_DNA_RT);
1681 skb->dev = dev;
1682 cb->src = fld.saddr;
1683 cb->dst = fld.daddr;
1684 local_bh_disable();
1685 err = dn_route_input(skb);
1686 local_bh_enable();
1687 memset(cb, 0, sizeof(struct dn_skb_cb));
1688 rt = (struct dn_route *)skb_dst(skb);
1689 if (!err && -rt->dst.error)
1690 err = rt->dst.error;
1691 } else {
1692 if (tb[RTA_OIF])
1693 fld.flowidn_oif = nla_get_u32(tb[RTA_OIF]);
1694
1695 err = dn_route_output_key((struct dst_entry **)&rt, &fld, 0);
1696 }
1697
1698 skb->dev = NULL;
1699 if (err)
1700 goto out_free;
1701 skb_dst_set(skb, &rt->dst);
1702 if (rtm->rtm_flags & RTM_F_NOTIFY)
1703 rt->rt_flags |= RTCF_NOTIFY;
1704
1705 err = dn_rt_fill_info(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq, RTM_NEWROUTE, 0, 0);
1706 if (err < 0) {
1707 err = -EMSGSIZE;
1708 goto out_free;
1709 }
1710
1711 return rtnl_unicast(skb, &init_net, NETLINK_CB(in_skb).portid);
1712
1713out_free:
1714 kfree_skb(skb);
1715 return err;
1716}
1717
1718/*
1719 * For routers, this is called from dn_fib_dump, but for endnodes its
1720 * called directly from the rtnetlink dispatch table.
1721 */
1722int dn_cache_dump(struct sk_buff *skb, struct netlink_callback *cb)
1723{
1724 struct net *net = sock_net(skb->sk);
1725 struct dn_route *rt;
1726 int h, s_h;
1727 int idx, s_idx;
1728 struct rtmsg *rtm;
1729
1730 if (!net_eq(net, &init_net))
1731 return 0;
1732
1733 if (nlmsg_len(cb->nlh) < sizeof(struct rtmsg))
1734 return -EINVAL;
1735
1736 rtm = nlmsg_data(cb->nlh);
1737 if (!(rtm->rtm_flags & RTM_F_CLONED))
1738 return 0;
1739
1740 s_h = cb->args[0];
1741 s_idx = idx = cb->args[1];
1742 for(h = 0; h <= dn_rt_hash_mask; h++) {
1743 if (h < s_h)
1744 continue;
1745 if (h > s_h)
1746 s_idx = 0;
1747 rcu_read_lock_bh();
1748 for(rt = rcu_dereference_bh(dn_rt_hash_table[h].chain), idx = 0;
1749 rt;
1750 rt = rcu_dereference_bh(rt->dn_next), idx++) {
1751 if (idx < s_idx)
1752 continue;
1753 skb_dst_set(skb, dst_clone(&rt->dst));
1754 if (dn_rt_fill_info(skb, NETLINK_CB(cb->skb).portid,
1755 cb->nlh->nlmsg_seq, RTM_NEWROUTE,
1756 1, NLM_F_MULTI) < 0) {
1757 skb_dst_drop(skb);
1758 rcu_read_unlock_bh();
1759 goto done;
1760 }
1761 skb_dst_drop(skb);
1762 }
1763 rcu_read_unlock_bh();
1764 }
1765
1766done:
1767 cb->args[0] = h;
1768 cb->args[1] = idx;
1769 return skb->len;
1770}
1771
1772#ifdef CONFIG_PROC_FS
1773struct dn_rt_cache_iter_state {
1774 int bucket;
1775};
1776
1777static struct dn_route *dn_rt_cache_get_first(struct seq_file *seq)
1778{
1779 struct dn_route *rt = NULL;
1780 struct dn_rt_cache_iter_state *s = seq->private;
1781
1782 for(s->bucket = dn_rt_hash_mask; s->bucket >= 0; --s->bucket) {
1783 rcu_read_lock_bh();
1784 rt = rcu_dereference_bh(dn_rt_hash_table[s->bucket].chain);
1785 if (rt)
1786 break;
1787 rcu_read_unlock_bh();
1788 }
1789 return rt;
1790}
1791
1792static struct dn_route *dn_rt_cache_get_next(struct seq_file *seq, struct dn_route *rt)
1793{
1794 struct dn_rt_cache_iter_state *s = seq->private;
1795
1796 rt = rcu_dereference_bh(rt->dn_next);
1797 while (!rt) {
1798 rcu_read_unlock_bh();
1799 if (--s->bucket < 0)
1800 break;
1801 rcu_read_lock_bh();
1802 rt = rcu_dereference_bh(dn_rt_hash_table[s->bucket].chain);
1803 }
1804 return rt;
1805}
1806
1807static void *dn_rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
1808{
1809 struct dn_route *rt = dn_rt_cache_get_first(seq);
1810
1811 if (rt) {
1812 while(*pos && (rt = dn_rt_cache_get_next(seq, rt)))
1813 --*pos;
1814 }
1815 return *pos ? NULL : rt;
1816}
1817
1818static void *dn_rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1819{
1820 struct dn_route *rt = dn_rt_cache_get_next(seq, v);
1821 ++*pos;
1822 return rt;
1823}
1824
1825static void dn_rt_cache_seq_stop(struct seq_file *seq, void *v)
1826{
1827 if (v)
1828 rcu_read_unlock_bh();
1829}
1830
1831static int dn_rt_cache_seq_show(struct seq_file *seq, void *v)
1832{
1833 struct dn_route *rt = v;
1834 char buf1[DN_ASCBUF_LEN], buf2[DN_ASCBUF_LEN];
1835
1836 seq_printf(seq, "%-8s %-7s %-7s %04d %04d %04d\n",
1837 rt->dst.dev ? rt->dst.dev->name : "*",
1838 dn_addr2asc(le16_to_cpu(rt->rt_daddr), buf1),
1839 dn_addr2asc(le16_to_cpu(rt->rt_saddr), buf2),
1840 atomic_read(&rt->dst.__refcnt),
1841 rt->dst.__use, 0);
1842 return 0;
1843}
1844
1845static const struct seq_operations dn_rt_cache_seq_ops = {
1846 .start = dn_rt_cache_seq_start,
1847 .next = dn_rt_cache_seq_next,
1848 .stop = dn_rt_cache_seq_stop,
1849 .show = dn_rt_cache_seq_show,
1850};
1851#endif /* CONFIG_PROC_FS */
1852
1853void __init dn_route_init(void)
1854{
1855 int i, goal, order;
1856
1857 dn_dst_ops.kmem_cachep =
1858 kmem_cache_create("dn_dst_cache", sizeof(struct dn_route), 0,
1859 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1860 dst_entries_init(&dn_dst_ops);
1861 timer_setup(&dn_route_timer, dn_dst_check_expire, 0);
1862 dn_route_timer.expires = jiffies + decnet_dst_gc_interval * HZ;
1863 add_timer(&dn_route_timer);
1864
1865 goal = totalram_pages() >> (26 - PAGE_SHIFT);
1866
1867 for(order = 0; (1UL << order) < goal; order++)
1868 /* NOTHING */;
1869
1870 /*
1871 * Only want 1024 entries max, since the table is very, very unlikely
1872 * to be larger than that.
1873 */
1874 while(order && ((((1UL << order) * PAGE_SIZE) /
1875 sizeof(struct dn_rt_hash_bucket)) >= 2048))
1876 order--;
1877
1878 do {
1879 dn_rt_hash_mask = (1UL << order) * PAGE_SIZE /
1880 sizeof(struct dn_rt_hash_bucket);
1881 while(dn_rt_hash_mask & (dn_rt_hash_mask - 1))
1882 dn_rt_hash_mask--;
1883 dn_rt_hash_table = (struct dn_rt_hash_bucket *)
1884 __get_free_pages(GFP_ATOMIC, order);
1885 } while (dn_rt_hash_table == NULL && --order > 0);
1886
1887 if (!dn_rt_hash_table)
1888 panic("Failed to allocate DECnet route cache hash table\n");
1889
1890 printk(KERN_INFO
1891 "DECnet: Routing cache hash table of %u buckets, %ldKbytes\n",
1892 dn_rt_hash_mask,
1893 (long)(dn_rt_hash_mask*sizeof(struct dn_rt_hash_bucket))/1024);
1894
1895 dn_rt_hash_mask--;
1896 for(i = 0; i <= dn_rt_hash_mask; i++) {
1897 spin_lock_init(&dn_rt_hash_table[i].lock);
1898 dn_rt_hash_table[i].chain = NULL;
1899 }
1900
1901 dn_dst_ops.gc_thresh = (dn_rt_hash_mask + 1);
1902
1903 proc_create_seq_private("decnet_cache", 0444, init_net.proc_net,
1904 &dn_rt_cache_seq_ops,
1905 sizeof(struct dn_rt_cache_iter_state), NULL);
1906
1907#ifdef CONFIG_DECNET_ROUTER
1908 rtnl_register_module(THIS_MODULE, PF_DECnet, RTM_GETROUTE,
1909 dn_cache_getroute, dn_fib_dump, 0);
1910#else
1911 rtnl_register_module(THIS_MODULE, PF_DECnet, RTM_GETROUTE,
1912 dn_cache_getroute, dn_cache_dump, 0);
1913#endif
1914}
1915
1916void __exit dn_route_cleanup(void)
1917{
1918 del_timer(&dn_route_timer);
1919 dn_run_flush(NULL);
1920
1921 remove_proc_entry("decnet_cache", init_net.proc_net);
1922 dst_entries_destroy(&dn_dst_ops);
1923}