?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
x86/mm: Add TLB purge to free pmd/pte page interfaces
[linux/fpc-iii.git] / net / decnet / dn_route.c
blob403593bd2b83b0d717f8062e476f8fffb50f92dc
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 void dn_dst_check_expire(unsigned long dummy)
192 {
193 int i;
194 struct dn_route *rt;
195 struct dn_route __rcu **rtp;
196 unsigned long now = jiffies;
197 unsigned long expire = 120 * HZ;
198
199 for (i = 0; i <= dn_rt_hash_mask; i++) {
200 rtp = &dn_rt_hash_table[i].chain;
201
202 spin_lock(&dn_rt_hash_table[i].lock);
203 while ((rt = rcu_dereference_protected(*rtp,
204 lockdep_is_held(&dn_rt_hash_table[i].lock))) != NULL) {
205 if (atomic_read(&rt->dst.__refcnt) ||
206 (now - rt->dst.lastuse) < expire) {
207 rtp = &rt->dst.dn_next;
208 continue;
209 }
210 *rtp = rt->dst.dn_next;
211 rt->dst.dn_next = NULL;
212 dnrt_free(rt);
213 }
214 spin_unlock(&dn_rt_hash_table[i].lock);
215
216 if ((jiffies - now) > 0)
217 break;
218 }
219
220 mod_timer(&dn_route_timer, now + decnet_dst_gc_interval * HZ);
221 }
222
223 static int dn_dst_gc(struct dst_ops *ops)
224 {
225 struct dn_route *rt;
226 struct dn_route __rcu **rtp;
227 int i;
228 unsigned long now = jiffies;
229 unsigned long expire = 10 * HZ;
230
231 for (i = 0; i <= dn_rt_hash_mask; i++) {
232
233 spin_lock_bh(&dn_rt_hash_table[i].lock);
234 rtp = &dn_rt_hash_table[i].chain;
235
236 while ((rt = rcu_dereference_protected(*rtp,
237 lockdep_is_held(&dn_rt_hash_table[i].lock))) != NULL) {
238 if (atomic_read(&rt->dst.__refcnt) ||
239 (now - rt->dst.lastuse) < expire) {
240 rtp = &rt->dst.dn_next;
241 continue;
242 }
243 *rtp = rt->dst.dn_next;
244 rt->dst.dn_next = NULL;
245 dnrt_free(rt);
246 break;
247 }
248 spin_unlock_bh(&dn_rt_hash_table[i].lock);
249 }
250
251 return 0;
252 }
253
254 /*
255 * The decnet standards don't impose a particular minimum mtu, what they
256 * do insist on is that the routing layer accepts a datagram of at least
257 * 230 bytes long. Here we have to subtract the routing header length from
258 * 230 to get the minimum acceptable mtu. If there is no neighbour, then we
259 * assume the worst and use a long header size.
260 *
261 * We update both the mtu and the advertised mss (i.e. the segment size we
262 * advertise to the other end).
263 */
264 static void dn_dst_update_pmtu(struct dst_entry *dst, struct sock *sk,
265 struct sk_buff *skb, u32 mtu)
266 {
267 struct dn_route *rt = (struct dn_route *) dst;
268 struct neighbour *n = rt->n;
269 u32 min_mtu = 230;
270 struct dn_dev *dn;
271
272 dn = n ? rcu_dereference_raw(n->dev->dn_ptr) : NULL;
273
274 if (dn && dn->use_long == 0)
275 min_mtu -= 6;
276 else
277 min_mtu -= 21;
278
279 if (dst_metric(dst, RTAX_MTU) > mtu && mtu >= min_mtu) {
280 if (!(dst_metric_locked(dst, RTAX_MTU))) {
281 dst_metric_set(dst, RTAX_MTU, mtu);
282 dst_set_expires(dst, dn_rt_mtu_expires);
283 }
284 if (!(dst_metric_locked(dst, RTAX_ADVMSS))) {
285 u32 mss = mtu - DN_MAX_NSP_DATA_HEADER;
286 u32 existing_mss = dst_metric_raw(dst, RTAX_ADVMSS);
287 if (!existing_mss || existing_mss > mss)
288 dst_metric_set(dst, RTAX_ADVMSS, mss);
289 }
290 }
291 }
292
293 static void dn_dst_redirect(struct dst_entry *dst, struct sock *sk,
294 struct sk_buff *skb)
295 {
296 }
297
298 /*
299 * When a route has been marked obsolete. (e.g. routing cache flush)
300 */
301 static struct dst_entry *dn_dst_check(struct dst_entry *dst, __u32 cookie)
302 {
303 return NULL;
304 }
305
306 static struct dst_entry *dn_dst_negative_advice(struct dst_entry *dst)
307 {
308 dst_release(dst);
309 return NULL;
310 }
311
312 static void dn_dst_link_failure(struct sk_buff *skb)
313 {
314 }
315
316 static inline int compare_keys(struct flowidn *fl1, struct flowidn *fl2)
317 {
318 return ((fl1->daddr ^ fl2->daddr) |
319 (fl1->saddr ^ fl2->saddr) |
320 (fl1->flowidn_mark ^ fl2->flowidn_mark) |
321 (fl1->flowidn_scope ^ fl2->flowidn_scope) |
322 (fl1->flowidn_oif ^ fl2->flowidn_oif) |
323 (fl1->flowidn_iif ^ fl2->flowidn_iif)) == 0;
324 }
325
326 static int dn_insert_route(struct dn_route *rt, unsigned int hash, struct dn_route **rp)
327 {
328 struct dn_route *rth;
329 struct dn_route __rcu **rthp;
330 unsigned long now = jiffies;
331
332 rthp = &dn_rt_hash_table[hash].chain;
333
334 spin_lock_bh(&dn_rt_hash_table[hash].lock);
335 while ((rth = rcu_dereference_protected(*rthp,
336 lockdep_is_held(&dn_rt_hash_table[hash].lock))) != NULL) {
337 if (compare_keys(&rth->fld, &rt->fld)) {
338 /* Put it first */
339 *rthp = rth->dst.dn_next;
340 rcu_assign_pointer(rth->dst.dn_next,
341 dn_rt_hash_table[hash].chain);
342 rcu_assign_pointer(dn_rt_hash_table[hash].chain, rth);
343
344 dst_use(&rth->dst, now);
345 spin_unlock_bh(&dn_rt_hash_table[hash].lock);
346
347 dst_free(&rt->dst);
348 *rp = rth;
349 return 0;
350 }
351 rthp = &rth->dst.dn_next;
352 }
353
354 rcu_assign_pointer(rt->dst.dn_next, dn_rt_hash_table[hash].chain);
355 rcu_assign_pointer(dn_rt_hash_table[hash].chain, rt);
356
357 dst_use(&rt->dst, now);
358 spin_unlock_bh(&dn_rt_hash_table[hash].lock);
359 *rp = rt;
360 return 0;
361 }
362
363 static void dn_run_flush(unsigned long dummy)
364 {
365 int i;
366 struct dn_route *rt, *next;
367
368 for (i = 0; i < dn_rt_hash_mask; i++) {
369 spin_lock_bh(&dn_rt_hash_table[i].lock);
370
371 if ((rt = xchg((struct dn_route **)&dn_rt_hash_table[i].chain, NULL)) == NULL)
372 goto nothing_to_declare;
373
374 for(; rt; rt = next) {
375 next = rcu_dereference_raw(rt->dst.dn_next);
376 RCU_INIT_POINTER(rt->dst.dn_next, NULL);
377 dnrt_free(rt);
378 }
379
380 nothing_to_declare:
381 spin_unlock_bh(&dn_rt_hash_table[i].lock);
382 }
383 }
384
385 static DEFINE_SPINLOCK(dn_rt_flush_lock);
386
387 void dn_rt_cache_flush(int delay)
388 {
389 unsigned long now = jiffies;
390 int user_mode = !in_interrupt();
391
392 if (delay < 0)
393 delay = dn_rt_min_delay;
394
395 spin_lock_bh(&dn_rt_flush_lock);
396
397 if (del_timer(&dn_rt_flush_timer) && delay > 0 && dn_rt_deadline) {
398 long tmo = (long)(dn_rt_deadline - now);
399
400 if (user_mode && tmo < dn_rt_max_delay - dn_rt_min_delay)
401 tmo = 0;
402
403 if (delay > tmo)
404 delay = tmo;
405 }
406
407 if (delay <= 0) {
408 spin_unlock_bh(&dn_rt_flush_lock);
409 dn_run_flush(0);
410 return;
411 }
412
413 if (dn_rt_deadline == 0)
414 dn_rt_deadline = now + dn_rt_max_delay;
415
416 dn_rt_flush_timer.expires = now + delay;
417 add_timer(&dn_rt_flush_timer);
418 spin_unlock_bh(&dn_rt_flush_lock);
419 }
420
421 /**
422 * dn_return_short - Return a short packet to its sender
423 * @skb: The packet to return
424 *
425 */
426 static int dn_return_short(struct sk_buff *skb)
427 {
428 struct dn_skb_cb *cb;
429 unsigned char *ptr;
430 __le16 *src;
431 __le16 *dst;
432
433 /* Add back headers */
434 skb_push(skb, skb->data - skb_network_header(skb));
435
436 if ((skb = skb_unshare(skb, GFP_ATOMIC)) == NULL)
437 return NET_RX_DROP;
438
439 cb = DN_SKB_CB(skb);
440 /* Skip packet length and point to flags */
441 ptr = skb->data + 2;
442 *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
443
444 dst = (__le16 *)ptr;
445 ptr += 2;
446 src = (__le16 *)ptr;
447 ptr += 2;
448 *ptr = 0; /* Zero hop count */
449
450 swap(*src, *dst);
451
452 skb->pkt_type = PACKET_OUTGOING;
453 dn_rt_finish_output(skb, NULL, NULL);
454 return NET_RX_SUCCESS;
455 }
456
457 /**
458 * dn_return_long - Return a long packet to its sender
459 * @skb: The long format packet to return
460 *
461 */
462 static int dn_return_long(struct sk_buff *skb)
463 {
464 struct dn_skb_cb *cb;
465 unsigned char *ptr;
466 unsigned char *src_addr, *dst_addr;
467 unsigned char tmp[ETH_ALEN];
468
469 /* Add back all headers */
470 skb_push(skb, skb->data - skb_network_header(skb));
471
472 if ((skb = skb_unshare(skb, GFP_ATOMIC)) == NULL)
473 return NET_RX_DROP;
474
475 cb = DN_SKB_CB(skb);
476 /* Ignore packet length and point to flags */
477 ptr = skb->data + 2;
478
479 /* Skip padding */
480 if (*ptr & DN_RT_F_PF) {
481 char padlen = (*ptr & ~DN_RT_F_PF);
482 ptr += padlen;
483 }
484
485 *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
486 ptr += 2;
487 dst_addr = ptr;
488 ptr += 8;
489 src_addr = ptr;
490 ptr += 6;
491 *ptr = 0; /* Zero hop count */
492
493 /* Swap source and destination */
494 memcpy(tmp, src_addr, ETH_ALEN);
495 memcpy(src_addr, dst_addr, ETH_ALEN);
496 memcpy(dst_addr, tmp, ETH_ALEN);
497
498 skb->pkt_type = PACKET_OUTGOING;
499 dn_rt_finish_output(skb, dst_addr, src_addr);
500 return NET_RX_SUCCESS;
501 }
502
503 /**
504 * dn_route_rx_packet - Try and find a route for an incoming packet
505 * @skb: The packet to find a route for
506 *
507 * Returns: result of input function if route is found, error code otherwise
508 */
509 static int dn_route_rx_packet(struct net *net, struct sock *sk, struct sk_buff *skb)
510 {
511 struct dn_skb_cb *cb;
512 int err;
513
514 if ((err = dn_route_input(skb)) == 0)
515 return dst_input(skb);
516
517 cb = DN_SKB_CB(skb);
518 if (decnet_debug_level & 4) {
519 char *devname = skb->dev ? skb->dev->name : "???";
520
521 printk(KERN_DEBUG
522 "DECnet: dn_route_rx_packet: rt_flags=0x%02x dev=%s len=%d src=0x%04hx dst=0x%04hx err=%d type=%d\n",
523 (int)cb->rt_flags, devname, skb->len,
524 le16_to_cpu(cb->src), le16_to_cpu(cb->dst),
525 err, skb->pkt_type);
526 }
527
528 if ((skb->pkt_type == PACKET_HOST) && (cb->rt_flags & DN_RT_F_RQR)) {
529 switch (cb->rt_flags & DN_RT_PKT_MSK) {
530 case DN_RT_PKT_SHORT:
531 return dn_return_short(skb);
532 case DN_RT_PKT_LONG:
533 return dn_return_long(skb);
534 }
535 }
536
537 kfree_skb(skb);
538 return NET_RX_DROP;
539 }
540
541 static int dn_route_rx_long(struct sk_buff *skb)
542 {
543 struct dn_skb_cb *cb = DN_SKB_CB(skb);
544 unsigned char *ptr = skb->data;
545
546 if (!pskb_may_pull(skb, 21)) /* 20 for long header, 1 for shortest nsp */
547 goto drop_it;
548
549 skb_pull(skb, 20);
550 skb_reset_transport_header(skb);
551
552 /* Destination info */
553 ptr += 2;
554 cb->dst = dn_eth2dn(ptr);
555 if (memcmp(ptr, dn_hiord_addr, 4) != 0)
556 goto drop_it;
557 ptr += 6;
558
559
560 /* Source info */
561 ptr += 2;
562 cb->src = dn_eth2dn(ptr);
563 if (memcmp(ptr, dn_hiord_addr, 4) != 0)
564 goto drop_it;
565 ptr += 6;
566 /* Other junk */
567 ptr++;
568 cb->hops = *ptr++; /* Visit Count */
569
570 return NF_HOOK(NFPROTO_DECNET, NF_DN_PRE_ROUTING,
571 &init_net, NULL, skb, skb->dev, NULL,
572 dn_route_rx_packet);
573
574 drop_it:
575 kfree_skb(skb);
576 return NET_RX_DROP;
577 }
578
579
580
581 static int dn_route_rx_short(struct sk_buff *skb)
582 {
583 struct dn_skb_cb *cb = DN_SKB_CB(skb);
584 unsigned char *ptr = skb->data;
585
586 if (!pskb_may_pull(skb, 6)) /* 5 for short header + 1 for shortest nsp */
587 goto drop_it;
588
589 skb_pull(skb, 5);
590 skb_reset_transport_header(skb);
591
592 cb->dst = *(__le16 *)ptr;
593 ptr += 2;
594 cb->src = *(__le16 *)ptr;
595 ptr += 2;
596 cb->hops = *ptr & 0x3f;
597
598 return NF_HOOK(NFPROTO_DECNET, NF_DN_PRE_ROUTING,
599 &init_net, NULL, skb, skb->dev, NULL,
600 dn_route_rx_packet);
601
602 drop_it:
603 kfree_skb(skb);
604 return NET_RX_DROP;
605 }
606
607 static int dn_route_discard(struct net *net, struct sock *sk, struct sk_buff *skb)
608 {
609 /*
610 * I know we drop the packet here, but thats considered success in
611 * this case
612 */
613 kfree_skb(skb);
614 return NET_RX_SUCCESS;
615 }
616
617 static int dn_route_ptp_hello(struct net *net, struct sock *sk, struct sk_buff *skb)
618 {
619 dn_dev_hello(skb);
620 dn_neigh_pointopoint_hello(skb);
621 return NET_RX_SUCCESS;
622 }
623
624 int dn_route_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
625 {
626 struct dn_skb_cb *cb;
627 unsigned char flags = 0;
628 __u16 len = le16_to_cpu(*(__le16 *)skb->data);
629 struct dn_dev *dn = rcu_dereference(dev->dn_ptr);
630 unsigned char padlen = 0;
631
632 if (!net_eq(dev_net(dev), &init_net))
633 goto dump_it;
634
635 if (dn == NULL)
636 goto dump_it;
637
638 if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
639 goto out;
640
641 if (!pskb_may_pull(skb, 3))
642 goto dump_it;
643
644 skb_pull(skb, 2);
645
646 if (len > skb->len)
647 goto dump_it;
648
649 skb_trim(skb, len);
650
651 flags = *skb->data;
652
653 cb = DN_SKB_CB(skb);
654 cb->stamp = jiffies;
655 cb->iif = dev->ifindex;
656
657 /*
658 * If we have padding, remove it.
659 */
660 if (flags & DN_RT_F_PF) {
661 padlen = flags & ~DN_RT_F_PF;
662 if (!pskb_may_pull(skb, padlen + 1))
663 goto dump_it;
664 skb_pull(skb, padlen);
665 flags = *skb->data;
666 }
667
668 skb_reset_network_header(skb);
669
670 /*
671 * Weed out future version DECnet
672 */
673 if (flags & DN_RT_F_VER)
674 goto dump_it;
675
676 cb->rt_flags = flags;
677
678 if (decnet_debug_level & 1)
679 printk(KERN_DEBUG
680 "dn_route_rcv: got 0x%02x from %s [%d %d %d]\n",
681 (int)flags, (dev) ? dev->name : "???", len, skb->len,
682 padlen);
683
684 if (flags & DN_RT_PKT_CNTL) {
685 if (unlikely(skb_linearize(skb)))
686 goto dump_it;
687
688 switch (flags & DN_RT_CNTL_MSK) {
689 case DN_RT_PKT_INIT:
690 dn_dev_init_pkt(skb);
691 break;
692 case DN_RT_PKT_VERI:
693 dn_dev_veri_pkt(skb);
694 break;
695 }
696
697 if (dn->parms.state != DN_DEV_S_RU)
698 goto dump_it;
699
700 switch (flags & DN_RT_CNTL_MSK) {
701 case DN_RT_PKT_HELO:
702 return NF_HOOK(NFPROTO_DECNET, NF_DN_HELLO,
703 &init_net, NULL, skb, skb->dev, NULL,
704 dn_route_ptp_hello);
705
706 case DN_RT_PKT_L1RT:
707 case DN_RT_PKT_L2RT:
708 return NF_HOOK(NFPROTO_DECNET, NF_DN_ROUTE,
709 &init_net, NULL, skb, skb->dev, NULL,
710 dn_route_discard);
711 case DN_RT_PKT_ERTH:
712 return NF_HOOK(NFPROTO_DECNET, NF_DN_HELLO,
713 &init_net, NULL, skb, skb->dev, NULL,
714 dn_neigh_router_hello);
715
716 case DN_RT_PKT_EEDH:
717 return NF_HOOK(NFPROTO_DECNET, NF_DN_HELLO,
718 &init_net, NULL, skb, skb->dev, NULL,
719 dn_neigh_endnode_hello);
720 }
721 } else {
722 if (dn->parms.state != DN_DEV_S_RU)
723 goto dump_it;
724
725 skb_pull(skb, 1); /* Pull flags */
726
727 switch (flags & DN_RT_PKT_MSK) {
728 case DN_RT_PKT_LONG:
729 return dn_route_rx_long(skb);
730 case DN_RT_PKT_SHORT:
731 return dn_route_rx_short(skb);
732 }
733 }
734
735 dump_it:
736 kfree_skb(skb);
737 out:
738 return NET_RX_DROP;
739 }
740
741 static int dn_output(struct net *net, struct sock *sk, struct sk_buff *skb)
742 {
743 struct dst_entry *dst = skb_dst(skb);
744 struct dn_route *rt = (struct dn_route *)dst;
745 struct net_device *dev = dst->dev;
746 struct dn_skb_cb *cb = DN_SKB_CB(skb);
747
748 int err = -EINVAL;
749
750 if (rt->n == NULL)
751 goto error;
752
753 skb->dev = dev;
754
755 cb->src = rt->rt_saddr;
756 cb->dst = rt->rt_daddr;
757
758 /*
759 * Always set the Intra-Ethernet bit on all outgoing packets
760 * originated on this node. Only valid flag from upper layers
761 * is return-to-sender-requested. Set hop count to 0 too.
762 */
763 cb->rt_flags &= ~DN_RT_F_RQR;
764 cb->rt_flags |= DN_RT_F_IE;
765 cb->hops = 0;
766
767 return NF_HOOK(NFPROTO_DECNET, NF_DN_LOCAL_OUT,
768 &init_net, sk, skb, NULL, dev,
769 dn_to_neigh_output);
770
771 error:
772 net_dbg_ratelimited("dn_output: This should not happen\n");
773
774 kfree_skb(skb);
775
776 return err;
777 }
778
779 static int dn_forward(struct sk_buff *skb)
780 {
781 struct dn_skb_cb *cb = DN_SKB_CB(skb);
782 struct dst_entry *dst = skb_dst(skb);
783 struct dn_dev *dn_db = rcu_dereference(dst->dev->dn_ptr);
784 struct dn_route *rt;
785 int header_len;
786 struct net_device *dev = skb->dev;
787
788 if (skb->pkt_type != PACKET_HOST)
789 goto drop;
790
791 /* Ensure that we have enough space for headers */
792 rt = (struct dn_route *)skb_dst(skb);
793 header_len = dn_db->use_long ? 21 : 6;
794 if (skb_cow(skb, LL_RESERVED_SPACE(rt->dst.dev)+header_len))
795 goto drop;
796
797 /*
798 * Hop count exceeded.
799 */
800 if (++cb->hops > 30)
801 goto drop;
802
803 skb->dev = rt->dst.dev;
804
805 /*
806 * If packet goes out same interface it came in on, then set
807 * the Intra-Ethernet bit. This has no effect for short
808 * packets, so we don't need to test for them here.
809 */
810 cb->rt_flags &= ~DN_RT_F_IE;
811 if (rt->rt_flags & RTCF_DOREDIRECT)
812 cb->rt_flags |= DN_RT_F_IE;
813
814 return NF_HOOK(NFPROTO_DECNET, NF_DN_FORWARD,
815 &init_net, NULL, skb, dev, skb->dev,
816 dn_to_neigh_output);
817
818 drop:
819 kfree_skb(skb);
820 return NET_RX_DROP;
821 }
822
823 /*
824 * Used to catch bugs. This should never normally get
825 * called.
826 */
827 static int dn_rt_bug_out(struct net *net, struct sock *sk, struct sk_buff *skb)
828 {
829 struct dn_skb_cb *cb = DN_SKB_CB(skb);
830
831 net_dbg_ratelimited("dn_rt_bug: skb from:%04x to:%04x\n",
832 le16_to_cpu(cb->src), le16_to_cpu(cb->dst));
833
834 kfree_skb(skb);
835
836 return NET_RX_DROP;
837 }
838
839 static int dn_rt_bug(struct sk_buff *skb)
840 {
841 struct dn_skb_cb *cb = DN_SKB_CB(skb);
842
843 net_dbg_ratelimited("dn_rt_bug: skb from:%04x to:%04x\n",
844 le16_to_cpu(cb->src), le16_to_cpu(cb->dst));
845
846 kfree_skb(skb);
847
848 return NET_RX_DROP;
849 }
850
851 static unsigned int dn_dst_default_advmss(const struct dst_entry *dst)
852 {
853 return dn_mss_from_pmtu(dst->dev, dst_mtu(dst));
854 }
855
856 static unsigned int dn_dst_mtu(const struct dst_entry *dst)
857 {
858 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
859
860 return mtu ? : dst->dev->mtu;
861 }
862
863 static struct neighbour *dn_dst_neigh_lookup(const struct dst_entry *dst,
864 struct sk_buff *skb,
865 const void *daddr)
866 {
867 return __neigh_lookup_errno(&dn_neigh_table, daddr, dst->dev);
868 }
869
870 static int dn_rt_set_next_hop(struct dn_route *rt, struct dn_fib_res *res)
871 {
872 struct dn_fib_info *fi = res->fi;
873 struct net_device *dev = rt->dst.dev;
874 unsigned int mss_metric;
875 struct neighbour *n;
876
877 if (fi) {
878 if (DN_FIB_RES_GW(*res) &&
879 DN_FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
880 rt->rt_gateway = DN_FIB_RES_GW(*res);
881 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
882 }
883 rt->rt_type = res->type;
884
885 if (dev != NULL && rt->n == NULL) {
886 n = __neigh_lookup_errno(&dn_neigh_table, &rt->rt_gateway, dev);
887 if (IS_ERR(n))
888 return PTR_ERR(n);
889 rt->n = n;
890 }
891
892 if (dst_metric(&rt->dst, RTAX_MTU) > rt->dst.dev->mtu)
893 dst_metric_set(&rt->dst, RTAX_MTU, rt->dst.dev->mtu);
894 mss_metric = dst_metric_raw(&rt->dst, RTAX_ADVMSS);
895 if (mss_metric) {
896 unsigned int mss = dn_mss_from_pmtu(dev, dst_mtu(&rt->dst));
897 if (mss_metric > mss)
898 dst_metric_set(&rt->dst, RTAX_ADVMSS, mss);
899 }
900 return 0;
901 }
902
903 static inline int dn_match_addr(__le16 addr1, __le16 addr2)
904 {
905 __u16 tmp = le16_to_cpu(addr1) ^ le16_to_cpu(addr2);
906 int match = 16;
907 while(tmp) {
908 tmp >>= 1;
909 match--;
910 }
911 return match;
912 }
913
914 static __le16 dnet_select_source(const struct net_device *dev, __le16 daddr, int scope)
915 {
916 __le16 saddr = 0;
917 struct dn_dev *dn_db;
918 struct dn_ifaddr *ifa;
919 int best_match = 0;
920 int ret;
921
922 rcu_read_lock();
923 dn_db = rcu_dereference(dev->dn_ptr);
924 for (ifa = rcu_dereference(dn_db->ifa_list);
925 ifa != NULL;
926 ifa = rcu_dereference(ifa->ifa_next)) {
927 if (ifa->ifa_scope > scope)
928 continue;
929 if (!daddr) {
930 saddr = ifa->ifa_local;
931 break;
932 }
933 ret = dn_match_addr(daddr, ifa->ifa_local);
934 if (ret > best_match)
935 saddr = ifa->ifa_local;
936 if (best_match == 0)
937 saddr = ifa->ifa_local;
938 }
939 rcu_read_unlock();
940
941 return saddr;
942 }
943
944 static inline __le16 __dn_fib_res_prefsrc(struct dn_fib_res *res)
945 {
946 return dnet_select_source(DN_FIB_RES_DEV(*res), DN_FIB_RES_GW(*res), res->scope);
947 }
948
949 static inline __le16 dn_fib_rules_map_destination(__le16 daddr, struct dn_fib_res *res)
950 {
951 __le16 mask = dnet_make_mask(res->prefixlen);
952 return (daddr&~mask)|res->fi->fib_nh->nh_gw;
953 }
954
955 static int dn_route_output_slow(struct dst_entry **pprt, const struct flowidn *oldflp, int try_hard)
956 {
957 struct flowidn fld = {
958 .daddr = oldflp->daddr,
959 .saddr = oldflp->saddr,
960 .flowidn_scope = RT_SCOPE_UNIVERSE,
961 .flowidn_mark = oldflp->flowidn_mark,
962 .flowidn_iif = LOOPBACK_IFINDEX,
963 .flowidn_oif = oldflp->flowidn_oif,
964 };
965 struct dn_route *rt = NULL;
966 struct net_device *dev_out = NULL, *dev;
967 struct neighbour *neigh = NULL;
968 unsigned int hash;
969 unsigned int flags = 0;
970 struct dn_fib_res res = { .fi = NULL, .type = RTN_UNICAST };
971 int err;
972 int free_res = 0;
973 __le16 gateway = 0;
974
975 if (decnet_debug_level & 16)
976 printk(KERN_DEBUG
977 "dn_route_output_slow: dst=%04x src=%04x mark=%d"
978 " iif=%d oif=%d\n", le16_to_cpu(oldflp->daddr),
979 le16_to_cpu(oldflp->saddr),
980 oldflp->flowidn_mark, LOOPBACK_IFINDEX,
981 oldflp->flowidn_oif);
982
983 /* If we have an output interface, verify its a DECnet device */
984 if (oldflp->flowidn_oif) {
985 dev_out = dev_get_by_index(&init_net, oldflp->flowidn_oif);
986 err = -ENODEV;
987 if (dev_out && dev_out->dn_ptr == NULL) {
988 dev_put(dev_out);
989 dev_out = NULL;
990 }
991 if (dev_out == NULL)
992 goto out;
993 }
994
995 /* If we have a source address, verify that its a local address */
996 if (oldflp->saddr) {
997 err = -EADDRNOTAVAIL;
998
999 if (dev_out) {
1000 if (dn_dev_islocal(dev_out, oldflp->saddr))
1001 goto source_ok;
1002 dev_put(dev_out);
1003 goto out;
1004 }
1005 rcu_read_lock();
1006 for_each_netdev_rcu(&init_net, dev) {
1007 if (!dev->dn_ptr)
1008 continue;
1009 if (!dn_dev_islocal(dev, oldflp->saddr))
1010 continue;
1011 if ((dev->flags & IFF_LOOPBACK) &&
1012 oldflp->daddr &&
1013 !dn_dev_islocal(dev, oldflp->daddr))
1014 continue;
1015
1016 dev_out = dev;
1017 break;
1018 }
1019 rcu_read_unlock();
1020 if (dev_out == NULL)
1021 goto out;
1022 dev_hold(dev_out);
1023 source_ok:
1024 ;
1025 }
1026
1027 /* No destination? Assume its local */
1028 if (!fld.daddr) {
1029 fld.daddr = fld.saddr;
1030
1031 if (dev_out)
1032 dev_put(dev_out);
1033 err = -EINVAL;
1034 dev_out = init_net.loopback_dev;
1035 if (!dev_out->dn_ptr)
1036 goto out;
1037 err = -EADDRNOTAVAIL;
1038 dev_hold(dev_out);
1039 if (!fld.daddr) {
1040 fld.daddr =
1041 fld.saddr = dnet_select_source(dev_out, 0,
1042 RT_SCOPE_HOST);
1043 if (!fld.daddr)
1044 goto out;
1045 }
1046 fld.flowidn_oif = LOOPBACK_IFINDEX;
1047 res.type = RTN_LOCAL;
1048 goto make_route;
1049 }
1050
1051 if (decnet_debug_level & 16)
1052 printk(KERN_DEBUG
1053 "dn_route_output_slow: initial checks complete."
1054 " dst=%04x src=%04x oif=%d try_hard=%d\n",
1055 le16_to_cpu(fld.daddr), le16_to_cpu(fld.saddr),
1056 fld.flowidn_oif, try_hard);
1057
1058 /*
1059 * N.B. If the kernel is compiled without router support then
1060 * dn_fib_lookup() will evaluate to non-zero so this if () block
1061 * will always be executed.
1062 */
1063 err = -ESRCH;
1064 if (try_hard || (err = dn_fib_lookup(&fld, &res)) != 0) {
1065 struct dn_dev *dn_db;
1066 if (err != -ESRCH)
1067 goto out;
1068 /*
1069 * Here the fallback is basically the standard algorithm for
1070 * routing in endnodes which is described in the DECnet routing
1071 * docs
1072 *
1073 * If we are not trying hard, look in neighbour cache.
1074 * The result is tested to ensure that if a specific output
1075 * device/source address was requested, then we honour that
1076 * here
1077 */
1078 if (!try_hard) {
1079 neigh = neigh_lookup_nodev(&dn_neigh_table, &init_net, &fld.daddr);
1080 if (neigh) {
1081 if ((oldflp->flowidn_oif &&
1082 (neigh->dev->ifindex != oldflp->flowidn_oif)) ||
1083 (oldflp->saddr &&
1084 (!dn_dev_islocal(neigh->dev,
1085 oldflp->saddr)))) {
1086 neigh_release(neigh);
1087 neigh = NULL;
1088 } else {
1089 if (dev_out)
1090 dev_put(dev_out);
1091 if (dn_dev_islocal(neigh->dev, fld.daddr)) {
1092 dev_out = init_net.loopback_dev;
1093 res.type = RTN_LOCAL;
1094 } else {
1095 dev_out = neigh->dev;
1096 }
1097 dev_hold(dev_out);
1098 goto select_source;
1099 }
1100 }
1101 }
1102
1103 /* Not there? Perhaps its a local address */
1104 if (dev_out == NULL)
1105 dev_out = dn_dev_get_default();
1106 err = -ENODEV;
1107 if (dev_out == NULL)
1108 goto out;
1109 dn_db = rcu_dereference_raw(dev_out->dn_ptr);
1110 if (!dn_db)
1111 goto e_inval;
1112 /* Possible improvement - check all devices for local addr */
1113 if (dn_dev_islocal(dev_out, fld.daddr)) {
1114 dev_put(dev_out);
1115 dev_out = init_net.loopback_dev;
1116 dev_hold(dev_out);
1117 res.type = RTN_LOCAL;
1118 goto select_source;
1119 }
1120 /* Not local either.... try sending it to the default router */
1121 neigh = neigh_clone(dn_db->router);
1122 BUG_ON(neigh && neigh->dev != dev_out);
1123
1124 /* Ok then, we assume its directly connected and move on */
1125 select_source:
1126 if (neigh)
1127 gateway = ((struct dn_neigh *)neigh)->addr;
1128 if (gateway == 0)
1129 gateway = fld.daddr;
1130 if (fld.saddr == 0) {
1131 fld.saddr = dnet_select_source(dev_out, gateway,
1132 res.type == RTN_LOCAL ?
1133 RT_SCOPE_HOST :
1134 RT_SCOPE_LINK);
1135 if (fld.saddr == 0 && res.type != RTN_LOCAL)
1136 goto e_addr;
1137 }
1138 fld.flowidn_oif = dev_out->ifindex;
1139 goto make_route;
1140 }
1141 free_res = 1;
1142
1143 if (res.type == RTN_NAT)
1144 goto e_inval;
1145
1146 if (res.type == RTN_LOCAL) {
1147 if (!fld.saddr)
1148 fld.saddr = fld.daddr;
1149 if (dev_out)
1150 dev_put(dev_out);
1151 dev_out = init_net.loopback_dev;
1152 dev_hold(dev_out);
1153 if (!dev_out->dn_ptr)
1154 goto e_inval;
1155 fld.flowidn_oif = dev_out->ifindex;
1156 if (res.fi)
1157 dn_fib_info_put(res.fi);
1158 res.fi = NULL;
1159 goto make_route;
1160 }
1161
1162 if (res.fi->fib_nhs > 1 && fld.flowidn_oif == 0)
1163 dn_fib_select_multipath(&fld, &res);
1164
1165 /*
1166 * We could add some logic to deal with default routes here and
1167 * get rid of some of the special casing above.
1168 */
1169
1170 if (!fld.saddr)
1171 fld.saddr = DN_FIB_RES_PREFSRC(res);
1172
1173 if (dev_out)
1174 dev_put(dev_out);
1175 dev_out = DN_FIB_RES_DEV(res);
1176 dev_hold(dev_out);
1177 fld.flowidn_oif = dev_out->ifindex;
1178 gateway = DN_FIB_RES_GW(res);
1179
1180 make_route:
1181 if (dev_out->flags & IFF_LOOPBACK)
1182 flags |= RTCF_LOCAL;
1183
1184 rt = dst_alloc(&dn_dst_ops, dev_out, 0, DST_OBSOLETE_NONE, DST_HOST);
1185 if (rt == NULL)
1186 goto e_nobufs;
1187
1188 memset(&rt->fld, 0, sizeof(rt->fld));
1189 rt->fld.saddr = oldflp->saddr;
1190 rt->fld.daddr = oldflp->daddr;
1191 rt->fld.flowidn_oif = oldflp->flowidn_oif;
1192 rt->fld.flowidn_iif = 0;
1193 rt->fld.flowidn_mark = oldflp->flowidn_mark;
1194
1195 rt->rt_saddr = fld.saddr;
1196 rt->rt_daddr = fld.daddr;
1197 rt->rt_gateway = gateway ? gateway : fld.daddr;
1198 rt->rt_local_src = fld.saddr;
1199
1200 rt->rt_dst_map = fld.daddr;
1201 rt->rt_src_map = fld.saddr;
1202
1203 rt->n = neigh;
1204 neigh = NULL;
1205
1206 rt->dst.lastuse = jiffies;
1207 rt->dst.output = dn_output;
1208 rt->dst.input = dn_rt_bug;
1209 rt->rt_flags = flags;
1210 if (flags & RTCF_LOCAL)
1211 rt->dst.input = dn_nsp_rx;
1212
1213 err = dn_rt_set_next_hop(rt, &res);
1214 if (err)
1215 goto e_neighbour;
1216
1217 hash = dn_hash(rt->fld.saddr, rt->fld.daddr);
1218 dn_insert_route(rt, hash, (struct dn_route **)pprt);
1219
1220 done:
1221 if (neigh)
1222 neigh_release(neigh);
1223 if (free_res)
1224 dn_fib_res_put(&res);
1225 if (dev_out)
1226 dev_put(dev_out);
1227 out:
1228 return err;
1229
1230 e_addr:
1231 err = -EADDRNOTAVAIL;
1232 goto done;
1233 e_inval:
1234 err = -EINVAL;
1235 goto done;
1236 e_nobufs:
1237 err = -ENOBUFS;
1238 goto done;
1239 e_neighbour:
1240 dst_free(&rt->dst);
1241 goto e_nobufs;
1242 }
1243
1244
1245 /*
1246 * N.B. The flags may be moved into the flowi at some future stage.
1247 */
1248 static int __dn_route_output_key(struct dst_entry **pprt, const struct flowidn *flp, int flags)
1249 {
1250 unsigned int hash = dn_hash(flp->saddr, flp->daddr);
1251 struct dn_route *rt = NULL;
1252
1253 if (!(flags & MSG_TRYHARD)) {
1254 rcu_read_lock_bh();
1255 for (rt = rcu_dereference_bh(dn_rt_hash_table[hash].chain); rt;
1256 rt = rcu_dereference_bh(rt->dst.dn_next)) {
1257 if ((flp->daddr == rt->fld.daddr) &&
1258 (flp->saddr == rt->fld.saddr) &&
1259 (flp->flowidn_mark == rt->fld.flowidn_mark) &&
1260 dn_is_output_route(rt) &&
1261 (rt->fld.flowidn_oif == flp->flowidn_oif)) {
1262 dst_use(&rt->dst, jiffies);
1263 rcu_read_unlock_bh();
1264 *pprt = &rt->dst;
1265 return 0;
1266 }
1267 }
1268 rcu_read_unlock_bh();
1269 }
1270
1271 return dn_route_output_slow(pprt, flp, flags);
1272 }
1273
1274 static int dn_route_output_key(struct dst_entry **pprt, struct flowidn *flp, int flags)
1275 {
1276 int err;
1277
1278 err = __dn_route_output_key(pprt, flp, flags);
1279 if (err == 0 && flp->flowidn_proto) {
1280 *pprt = xfrm_lookup(&init_net, *pprt,
1281 flowidn_to_flowi(flp), NULL, 0);
1282 if (IS_ERR(*pprt)) {
1283 err = PTR_ERR(*pprt);
1284 *pprt = NULL;
1285 }
1286 }
1287 return err;
1288 }
1289
1290 int dn_route_output_sock(struct dst_entry __rcu **pprt, struct flowidn *fl, struct sock *sk, int flags)
1291 {
1292 int err;
1293
1294 err = __dn_route_output_key(pprt, fl, flags & MSG_TRYHARD);
1295 if (err == 0 && fl->flowidn_proto) {
1296 *pprt = xfrm_lookup(&init_net, *pprt,
1297 flowidn_to_flowi(fl), sk, 0);
1298 if (IS_ERR(*pprt)) {
1299 err = PTR_ERR(*pprt);
1300 *pprt = NULL;
1301 }
1302 }
1303 return err;
1304 }
1305
1306 static int dn_route_input_slow(struct sk_buff *skb)
1307 {
1308 struct dn_route *rt = NULL;
1309 struct dn_skb_cb *cb = DN_SKB_CB(skb);
1310 struct net_device *in_dev = skb->dev;
1311 struct net_device *out_dev = NULL;
1312 struct dn_dev *dn_db;
1313 struct neighbour *neigh = NULL;
1314 unsigned int hash;
1315 int flags = 0;
1316 __le16 gateway = 0;
1317 __le16 local_src = 0;
1318 struct flowidn fld = {
1319 .daddr = cb->dst,
1320 .saddr = cb->src,
1321 .flowidn_scope = RT_SCOPE_UNIVERSE,
1322 .flowidn_mark = skb->mark,
1323 .flowidn_iif = skb->dev->ifindex,
1324 };
1325 struct dn_fib_res res = { .fi = NULL, .type = RTN_UNREACHABLE };
1326 int err = -EINVAL;
1327 int free_res = 0;
1328
1329 dev_hold(in_dev);
1330
1331 if ((dn_db = rcu_dereference(in_dev->dn_ptr)) == NULL)
1332 goto out;
1333
1334 /* Zero source addresses are not allowed */
1335 if (fld.saddr == 0)
1336 goto out;
1337
1338 /*
1339 * In this case we've just received a packet from a source
1340 * outside ourselves pretending to come from us. We don't
1341 * allow it any further to prevent routing loops, spoofing and
1342 * other nasties. Loopback packets already have the dst attached
1343 * so this only affects packets which have originated elsewhere.
1344 */
1345 err = -ENOTUNIQ;
1346 if (dn_dev_islocal(in_dev, cb->src))
1347 goto out;
1348
1349 err = dn_fib_lookup(&fld, &res);
1350 if (err) {
1351 if (err != -ESRCH)
1352 goto out;
1353 /*
1354 * Is the destination us ?
1355 */
1356 if (!dn_dev_islocal(in_dev, cb->dst))
1357 goto e_inval;
1358
1359 res.type = RTN_LOCAL;
1360 } else {
1361 __le16 src_map = fld.saddr;
1362 free_res = 1;
1363
1364 out_dev = DN_FIB_RES_DEV(res);
1365 if (out_dev == NULL) {
1366 net_crit_ratelimited("Bug in dn_route_input_slow() No output device\n");
1367 goto e_inval;
1368 }
1369 dev_hold(out_dev);
1370
1371 if (res.r)
1372 src_map = fld.saddr; /* no NAT support for now */
1373
1374 gateway = DN_FIB_RES_GW(res);
1375 if (res.type == RTN_NAT) {
1376 fld.daddr = dn_fib_rules_map_destination(fld.daddr, &res);
1377 dn_fib_res_put(&res);
1378 free_res = 0;
1379 if (dn_fib_lookup(&fld, &res))
1380 goto e_inval;
1381 free_res = 1;
1382 if (res.type != RTN_UNICAST)
1383 goto e_inval;
1384 flags |= RTCF_DNAT;
1385 gateway = fld.daddr;
1386 }
1387 fld.saddr = src_map;
1388 }
1389
1390 switch(res.type) {
1391 case RTN_UNICAST:
1392 /*
1393 * Forwarding check here, we only check for forwarding
1394 * being turned off, if you want to only forward intra
1395 * area, its up to you to set the routing tables up
1396 * correctly.
1397 */
1398 if (dn_db->parms.forwarding == 0)
1399 goto e_inval;
1400
1401 if (res.fi->fib_nhs > 1 && fld.flowidn_oif == 0)
1402 dn_fib_select_multipath(&fld, &res);
1403
1404 /*
1405 * Check for out_dev == in_dev. We use the RTCF_DOREDIRECT
1406 * flag as a hint to set the intra-ethernet bit when
1407 * forwarding. If we've got NAT in operation, we don't do
1408 * this optimisation.
1409 */
1410 if (out_dev == in_dev && !(flags & RTCF_NAT))
1411 flags |= RTCF_DOREDIRECT;
1412
1413 local_src = DN_FIB_RES_PREFSRC(res);
1414
1415 case RTN_BLACKHOLE:
1416 case RTN_UNREACHABLE:
1417 break;
1418 case RTN_LOCAL:
1419 flags |= RTCF_LOCAL;
1420 fld.saddr = cb->dst;
1421 fld.daddr = cb->src;
1422
1423 /* Routing tables gave us a gateway */
1424 if (gateway)
1425 goto make_route;
1426
1427 /* Packet was intra-ethernet, so we know its on-link */
1428 if (cb->rt_flags & DN_RT_F_IE) {
1429 gateway = cb->src;
1430 goto make_route;
1431 }
1432
1433 /* Use the default router if there is one */
1434 neigh = neigh_clone(dn_db->router);
1435 if (neigh) {
1436 gateway = ((struct dn_neigh *)neigh)->addr;
1437 goto make_route;
1438 }
1439
1440 /* Close eyes and pray */
1441 gateway = cb->src;
1442 goto make_route;
1443 default:
1444 goto e_inval;
1445 }
1446
1447 make_route:
1448 rt = dst_alloc(&dn_dst_ops, out_dev, 0, DST_OBSOLETE_NONE, DST_HOST);
1449 if (rt == NULL)
1450 goto e_nobufs;
1451
1452 memset(&rt->fld, 0, sizeof(rt->fld));
1453 rt->rt_saddr = fld.saddr;
1454 rt->rt_daddr = fld.daddr;
1455 rt->rt_gateway = fld.daddr;
1456 if (gateway)
1457 rt->rt_gateway = gateway;
1458 rt->rt_local_src = local_src ? local_src : rt->rt_saddr;
1459
1460 rt->rt_dst_map = fld.daddr;
1461 rt->rt_src_map = fld.saddr;
1462
1463 rt->fld.saddr = cb->src;
1464 rt->fld.daddr = cb->dst;
1465 rt->fld.flowidn_oif = 0;
1466 rt->fld.flowidn_iif = in_dev->ifindex;
1467 rt->fld.flowidn_mark = fld.flowidn_mark;
1468
1469 rt->n = neigh;
1470 rt->dst.lastuse = jiffies;
1471 rt->dst.output = dn_rt_bug_out;
1472 switch (res.type) {
1473 case RTN_UNICAST:
1474 rt->dst.input = dn_forward;
1475 break;
1476 case RTN_LOCAL:
1477 rt->dst.output = dn_output;
1478 rt->dst.input = dn_nsp_rx;
1479 rt->dst.dev = in_dev;
1480 flags |= RTCF_LOCAL;
1481 break;
1482 default:
1483 case RTN_UNREACHABLE:
1484 case RTN_BLACKHOLE:
1485 rt->dst.input = dst_discard;
1486 }
1487 rt->rt_flags = flags;
1488
1489 err = dn_rt_set_next_hop(rt, &res);
1490 if (err)
1491 goto e_neighbour;
1492
1493 hash = dn_hash(rt->fld.saddr, rt->fld.daddr);
1494 dn_insert_route(rt, hash, &rt);
1495 skb_dst_set(skb, &rt->dst);
1496
1497 done:
1498 if (neigh)
1499 neigh_release(neigh);
1500 if (free_res)
1501 dn_fib_res_put(&res);
1502 dev_put(in_dev);
1503 if (out_dev)
1504 dev_put(out_dev);
1505 out:
1506 return err;
1507
1508 e_inval:
1509 err = -EINVAL;
1510 goto done;
1511
1512 e_nobufs:
1513 err = -ENOBUFS;
1514 goto done;
1515
1516 e_neighbour:
1517 dst_free(&rt->dst);
1518 goto done;
1519 }
1520
1521 static int dn_route_input(struct sk_buff *skb)
1522 {
1523 struct dn_route *rt;
1524 struct dn_skb_cb *cb = DN_SKB_CB(skb);
1525 unsigned int hash = dn_hash(cb->src, cb->dst);
1526
1527 if (skb_dst(skb))
1528 return 0;
1529
1530 rcu_read_lock();
1531 for(rt = rcu_dereference(dn_rt_hash_table[hash].chain); rt != NULL;
1532 rt = rcu_dereference(rt->dst.dn_next)) {
1533 if ((rt->fld.saddr == cb->src) &&
1534 (rt->fld.daddr == cb->dst) &&
1535 (rt->fld.flowidn_oif == 0) &&
1536 (rt->fld.flowidn_mark == skb->mark) &&
1537 (rt->fld.flowidn_iif == cb->iif)) {
1538 dst_use(&rt->dst, jiffies);
1539 rcu_read_unlock();
1540 skb_dst_set(skb, (struct dst_entry *)rt);
1541 return 0;
1542 }
1543 }
1544 rcu_read_unlock();
1545
1546 return dn_route_input_slow(skb);
1547 }
1548
1549 static int dn_rt_fill_info(struct sk_buff *skb, u32 portid, u32 seq,
1550 int event, int nowait, unsigned int flags)
1551 {
1552 struct dn_route *rt = (struct dn_route *)skb_dst(skb);
1553 struct rtmsg *r;
1554 struct nlmsghdr *nlh;
1555 long expires;
1556
1557 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*r), flags);
1558 if (!nlh)
1559 return -EMSGSIZE;
1560
1561 r = nlmsg_data(nlh);
1562 r->rtm_family = AF_DECnet;
1563 r->rtm_dst_len = 16;
1564 r->rtm_src_len = 0;
1565 r->rtm_tos = 0;
1566 r->rtm_table = RT_TABLE_MAIN;
1567 r->rtm_type = rt->rt_type;
1568 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
1569 r->rtm_scope = RT_SCOPE_UNIVERSE;
1570 r->rtm_protocol = RTPROT_UNSPEC;
1571
1572 if (rt->rt_flags & RTCF_NOTIFY)
1573 r->rtm_flags |= RTM_F_NOTIFY;
1574
1575 if (nla_put_u32(skb, RTA_TABLE, RT_TABLE_MAIN) < 0 ||
1576 nla_put_le16(skb, RTA_DST, rt->rt_daddr) < 0)
1577 goto errout;
1578
1579 if (rt->fld.saddr) {
1580 r->rtm_src_len = 16;
1581 if (nla_put_le16(skb, RTA_SRC, rt->fld.saddr) < 0)
1582 goto errout;
1583 }
1584 if (rt->dst.dev &&
1585 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex) < 0)
1586 goto errout;
1587
1588 /*
1589 * Note to self - change this if input routes reverse direction when
1590 * they deal only with inputs and not with replies like they do
1591 * currently.
1592 */
1593 if (nla_put_le16(skb, RTA_PREFSRC, rt->rt_local_src) < 0)
1594 goto errout;
1595
1596 if (rt->rt_daddr != rt->rt_gateway &&
1597 nla_put_le16(skb, RTA_GATEWAY, rt->rt_gateway) < 0)
1598 goto errout;
1599
1600 if (rtnetlink_put_metrics(skb, dst_metrics_ptr(&rt->dst)) < 0)
1601 goto errout;
1602
1603 expires = rt->dst.expires ? rt->dst.expires - jiffies : 0;
1604 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires,
1605 rt->dst.error) < 0)
1606 goto errout;
1607
1608 if (dn_is_input_route(rt) &&
1609 nla_put_u32(skb, RTA_IIF, rt->fld.flowidn_iif) < 0)
1610 goto errout;
1611
1612 nlmsg_end(skb, nlh);
1613 return 0;
1614
1615 errout:
1616 nlmsg_cancel(skb, nlh);
1617 return -EMSGSIZE;
1618 }
1619
1620 const struct nla_policy rtm_dn_policy[RTA_MAX + 1] = {
1621 [RTA_DST] = { .type = NLA_U16 },
1622 [RTA_SRC] = { .type = NLA_U16 },
1623 [RTA_IIF] = { .type = NLA_U32 },
1624 [RTA_OIF] = { .type = NLA_U32 },
1625 [RTA_GATEWAY] = { .type = NLA_U16 },
1626 [RTA_PRIORITY] = { .type = NLA_U32 },
1627 [RTA_PREFSRC] = { .type = NLA_U16 },
1628 [RTA_METRICS] = { .type = NLA_NESTED },
1629 [RTA_MULTIPATH] = { .type = NLA_NESTED },
1630 [RTA_TABLE] = { .type = NLA_U32 },
1631 [RTA_MARK] = { .type = NLA_U32 },
1632 };
1633
1634 /*
1635 * This is called by both endnodes and routers now.
1636 */
1637 static int dn_cache_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh)
1638 {
1639 struct net *net = sock_net(in_skb->sk);
1640 struct rtmsg *rtm = nlmsg_data(nlh);
1641 struct dn_route *rt = NULL;
1642 struct dn_skb_cb *cb;
1643 int err;
1644 struct sk_buff *skb;
1645 struct flowidn fld;
1646 struct nlattr *tb[RTA_MAX+1];
1647
1648 if (!net_eq(net, &init_net))
1649 return -EINVAL;
1650
1651 err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_dn_policy);
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
1713 out_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 */
1722 int 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->dst.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
1766 done:
1767 cb->args[0] = h;
1768 cb->args[1] = idx;
1769 return skb->len;
1770 }
1771
1772 #ifdef CONFIG_PROC_FS
1773 struct dn_rt_cache_iter_state {
1774 int bucket;
1775 };
1776
1777 static 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
1792 static 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->dst.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
1807 static 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
1818 static 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
1825 static void dn_rt_cache_seq_stop(struct seq_file *seq, void *v)
1826 {
1827 if (v)
1828 rcu_read_unlock_bh();
1829 }
1830
1831 static 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
1845 static 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
1852 static int dn_rt_cache_seq_open(struct inode *inode, struct file *file)
1853 {
1854 return seq_open_private(file, &dn_rt_cache_seq_ops,
1855 sizeof(struct dn_rt_cache_iter_state));
1856 }
1857
1858 static const struct file_operations dn_rt_cache_seq_fops = {
1859 .owner = THIS_MODULE,
1860 .open = dn_rt_cache_seq_open,
1861 .read = seq_read,
1862 .llseek = seq_lseek,
1863 .release = seq_release_private,
1864 };
1865
1866 #endif /* CONFIG_PROC_FS */
1867
1868 void __init dn_route_init(void)
1869 {
1870 int i, goal, order;
1871
1872 dn_dst_ops.kmem_cachep =
1873 kmem_cache_create("dn_dst_cache", sizeof(struct dn_route), 0,
1874 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1875 dst_entries_init(&dn_dst_ops);
1876 setup_timer(&dn_route_timer, dn_dst_check_expire, 0);
1877 dn_route_timer.expires = jiffies + decnet_dst_gc_interval * HZ;
1878 add_timer(&dn_route_timer);
1879
1880 goal = totalram_pages >> (26 - PAGE_SHIFT);
1881
1882 for(order = 0; (1UL << order) < goal; order++)
1883 /* NOTHING */;
1884
1885 /*
1886 * Only want 1024 entries max, since the table is very, very unlikely
1887 * to be larger than that.
1888 */
1889 while(order && ((((1UL << order) * PAGE_SIZE) /
1890 sizeof(struct dn_rt_hash_bucket)) >= 2048))
1891 order--;
1892
1893 do {
1894 dn_rt_hash_mask = (1UL << order) * PAGE_SIZE /
1895 sizeof(struct dn_rt_hash_bucket);
1896 while(dn_rt_hash_mask & (dn_rt_hash_mask - 1))
1897 dn_rt_hash_mask--;
1898 dn_rt_hash_table = (struct dn_rt_hash_bucket *)
1899 __get_free_pages(GFP_ATOMIC, order);
1900 } while (dn_rt_hash_table == NULL && --order > 0);
1901
1902 if (!dn_rt_hash_table)
1903 panic("Failed to allocate DECnet route cache hash table\n");
1904
1905 printk(KERN_INFO
1906 "DECnet: Routing cache hash table of %u buckets, %ldKbytes\n",
1907 dn_rt_hash_mask,
1908 (long)(dn_rt_hash_mask*sizeof(struct dn_rt_hash_bucket))/1024);
1909
1910 dn_rt_hash_mask--;
1911 for(i = 0; i <= dn_rt_hash_mask; i++) {
1912 spin_lock_init(&dn_rt_hash_table[i].lock);
1913 dn_rt_hash_table[i].chain = NULL;
1914 }
1915
1916 dn_dst_ops.gc_thresh = (dn_rt_hash_mask + 1);
1917
1918 proc_create("decnet_cache", S_IRUGO, init_net.proc_net,
1919 &dn_rt_cache_seq_fops);
1920
1921 #ifdef CONFIG_DECNET_ROUTER
1922 rtnl_register(PF_DECnet, RTM_GETROUTE, dn_cache_getroute,
1923 dn_fib_dump, NULL);
1924 #else
1925 rtnl_register(PF_DECnet, RTM_GETROUTE, dn_cache_getroute,
1926 dn_cache_dump, NULL);
1927 #endif
1928 }
1929
1930 void __exit dn_route_cleanup(void)
1931 {
1932 del_timer(&dn_route_timer);
1933 dn_run_flush(0);
1934
1935 remove_proc_entry("decnet_cache", init_net.proc_net);
1936 dst_entries_destroy(&dn_dst_ops);
1937 }
1938
Linux with added FPC-III support