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