Loading...
Note: File does not exist in v6.13.7.
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * DECnet An implementation of the DECnet protocol suite for the LINUX
4 * operating system. DECnet is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
6 *
7 * DECnet Routing Functions (Endnode and Router)
8 *
9 * Authors: Steve Whitehouse <SteveW@ACM.org>
10 * Eduardo Marcelo Serrat <emserrat@geocities.com>
11 *
12 * Changes:
13 * Steve Whitehouse : Fixes to allow "intra-ethernet" and
14 * "return-to-sender" bits on outgoing
15 * packets.
16 * Steve Whitehouse : Timeouts for cached routes.
17 * Steve Whitehouse : Use dst cache for input routes too.
18 * Steve Whitehouse : Fixed error values in dn_send_skb.
19 * Steve Whitehouse : Rework routing functions to better fit
20 * DECnet routing design
21 * Alexey Kuznetsov : New SMP locking
22 * Steve Whitehouse : More SMP locking changes & dn_cache_dump()
23 * Steve Whitehouse : Prerouting NF hook, now really is prerouting.
24 * Fixed possible skb leak in rtnetlink funcs.
25 * Steve Whitehouse : Dave Miller's dynamic hash table sizing and
26 * Alexey Kuznetsov's finer grained locking
27 * from ipv4/route.c.
28 * Steve Whitehouse : Routing is now starting to look like a
29 * sensible set of code now, mainly due to
30 * my copying the IPv4 routing code. The
31 * hooks here are modified and will continue
32 * to evolve for a while.
33 * Steve Whitehouse : Real SMP at last :-) Also new netfilter
34 * stuff. Look out raw sockets your days
35 * are numbered!
36 * Steve Whitehouse : Added return-to-sender functions. Added
37 * backlog congestion level return codes.
38 * Steve Whitehouse : Fixed bug where routes were set up with
39 * no ref count on net devices.
40 * Steve Whitehouse : RCU for the route cache
41 * Steve Whitehouse : Preparations for the flow cache
42 * Steve Whitehouse : Prepare for nonlinear skbs
43 */
44
45/******************************************************************************
46 (c) 1995-1998 E.M. Serrat emserrat@geocities.com
47
48*******************************************************************************/
49
50#include <linux/errno.h>
51#include <linux/types.h>
52#include <linux/socket.h>
53#include <linux/in.h>
54#include <linux/kernel.h>
55#include <linux/sockios.h>
56#include <linux/net.h>
57#include <linux/netdevice.h>
58#include <linux/inet.h>
59#include <linux/route.h>
60#include <linux/in_route.h>
61#include <linux/slab.h>
62#include <net/sock.h>
63#include <linux/mm.h>
64#include <linux/proc_fs.h>
65#include <linux/seq_file.h>
66#include <linux/init.h>
67#include <linux/rtnetlink.h>
68#include <linux/string.h>
69#include <linux/netfilter_decnet.h>
70#include <linux/rcupdate.h>
71#include <linux/times.h>
72#include <linux/export.h>
73#include <asm/errno.h>
74#include <net/net_namespace.h>
75#include <net/netlink.h>
76#include <net/neighbour.h>
77#include <net/dst.h>
78#include <net/flow.h>
79#include <net/fib_rules.h>
80#include <net/dn.h>
81#include <net/dn_dev.h>
82#include <net/dn_nsp.h>
83#include <net/dn_route.h>
84#include <net/dn_neigh.h>
85#include <net/dn_fib.h>
86
87struct dn_rt_hash_bucket
88{
89 struct dn_route __rcu *chain;
90 spinlock_t lock;
91};
92
93extern struct neigh_table dn_neigh_table;
94
95
96static unsigned char dn_hiord_addr[6] = {0xAA,0x00,0x04,0x00,0x00,0x00};
97
98static const int dn_rt_min_delay = 2 * HZ;
99static const int dn_rt_max_delay = 10 * HZ;
100static const int dn_rt_mtu_expires = 10 * 60 * HZ;
101
102static unsigned long dn_rt_deadline;
103
104static int dn_dst_gc(struct dst_ops *ops);
105static struct dst_entry *dn_dst_check(struct dst_entry *, __u32);
106static unsigned int dn_dst_default_advmss(const struct dst_entry *dst);
107static unsigned int dn_dst_mtu(const struct dst_entry *dst);
108static void dn_dst_destroy(struct dst_entry *);
109static void dn_dst_ifdown(struct dst_entry *, struct net_device *dev, int how);
110static struct dst_entry *dn_dst_negative_advice(struct dst_entry *);
111static void dn_dst_link_failure(struct sk_buff *);
112static void dn_dst_update_pmtu(struct dst_entry *dst, struct sock *sk,
113 struct sk_buff *skb , u32 mtu);
114static void dn_dst_redirect(struct dst_entry *dst, struct sock *sk,
115 struct sk_buff *skb);
116static struct neighbour *dn_dst_neigh_lookup(const struct dst_entry *dst,
117 struct sk_buff *skb,
118 const void *daddr);
119static int dn_route_input(struct sk_buff *);
120static void dn_run_flush(struct timer_list *unused);
121
122static struct dn_rt_hash_bucket *dn_rt_hash_table;
123static unsigned int dn_rt_hash_mask;
124
125static struct timer_list dn_route_timer;
126static DEFINE_TIMER(dn_rt_flush_timer, dn_run_flush);
127int decnet_dst_gc_interval = 2;
128
129static struct dst_ops dn_dst_ops = {
130 .family = PF_DECnet,
131 .gc_thresh = 128,
132 .gc = dn_dst_gc,
133 .check = dn_dst_check,
134 .default_advmss = dn_dst_default_advmss,
135 .mtu = dn_dst_mtu,
136 .cow_metrics = dst_cow_metrics_generic,
137 .destroy = dn_dst_destroy,
138 .ifdown = dn_dst_ifdown,
139 .negative_advice = dn_dst_negative_advice,
140 .link_failure = dn_dst_link_failure,
141 .update_pmtu = dn_dst_update_pmtu,
142 .redirect = dn_dst_redirect,
143 .neigh_lookup = dn_dst_neigh_lookup,
144};
145
146static void dn_dst_destroy(struct dst_entry *dst)
147{
148 struct dn_route *rt = (struct dn_route *) dst;
149
150 if (rt->n)
151 neigh_release(rt->n);
152 dst_destroy_metrics_generic(dst);
153}
154
155static void dn_dst_ifdown(struct dst_entry *dst, struct net_device *dev, int how)
156{
157 if (how) {
158 struct dn_route *rt = (struct dn_route *) dst;
159 struct neighbour *n = rt->n;
160
161 if (n && n->dev == dev) {
162 n->dev = dev_net(dev)->loopback_dev;
163 dev_hold(n->dev);
164 dev_put(dev);
165 }
166 }
167}
168
169static __inline__ unsigned int dn_hash(__le16 src, __le16 dst)
170{
171 __u16 tmp = (__u16 __force)(src ^ dst);
172 tmp ^= (tmp >> 3);
173 tmp ^= (tmp >> 5);
174 tmp ^= (tmp >> 10);
175 return dn_rt_hash_mask & (unsigned int)tmp;
176}
177
178static void dn_dst_check_expire(struct timer_list *unused)
179{
180 int i;
181 struct dn_route *rt;
182 struct dn_route __rcu **rtp;
183 unsigned long now = jiffies;
184 unsigned long expire = 120 * HZ;
185
186 for (i = 0; i <= dn_rt_hash_mask; i++) {
187 rtp = &dn_rt_hash_table[i].chain;
188
189 spin_lock(&dn_rt_hash_table[i].lock);
190 while ((rt = rcu_dereference_protected(*rtp,
191 lockdep_is_held(&dn_rt_hash_table[i].lock))) != NULL) {
192 if (atomic_read(&rt->dst.__refcnt) > 1 ||
193 (now - rt->dst.lastuse) < expire) {
194 rtp = &rt->dn_next;
195 continue;
196 }
197 *rtp = rt->dn_next;
198 rt->dn_next = NULL;
199 dst_dev_put(&rt->dst);
200 dst_release(&rt->dst);
201 }
202 spin_unlock(&dn_rt_hash_table[i].lock);
203
204 if ((jiffies - now) > 0)
205 break;
206 }
207
208 mod_timer(&dn_route_timer, now + decnet_dst_gc_interval * HZ);
209}
210
211static int dn_dst_gc(struct dst_ops *ops)
212{
213 struct dn_route *rt;
214 struct dn_route __rcu **rtp;
215 int i;
216 unsigned long now = jiffies;
217 unsigned long expire = 10 * HZ;
218
219 for (i = 0; i <= dn_rt_hash_mask; i++) {
220
221 spin_lock_bh(&dn_rt_hash_table[i].lock);
222 rtp = &dn_rt_hash_table[i].chain;
223
224 while ((rt = rcu_dereference_protected(*rtp,
225 lockdep_is_held(&dn_rt_hash_table[i].lock))) != NULL) {
226 if (atomic_read(&rt->dst.__refcnt) > 1 ||
227 (now - rt->dst.lastuse) < expire) {
228 rtp = &rt->dn_next;
229 continue;
230 }
231 *rtp = rt->dn_next;
232 rt->dn_next = NULL;
233 dst_dev_put(&rt->dst);
234 dst_release(&rt->dst);
235 break;
236 }
237 spin_unlock_bh(&dn_rt_hash_table[i].lock);
238 }
239
240 return 0;
241}
242
243/*
244 * The decnet standards don't impose a particular minimum mtu, what they
245 * do insist on is that the routing layer accepts a datagram of at least
246 * 230 bytes long. Here we have to subtract the routing header length from
247 * 230 to get the minimum acceptable mtu. If there is no neighbour, then we
248 * assume the worst and use a long header size.
249 *
250 * We update both the mtu and the advertised mss (i.e. the segment size we
251 * advertise to the other end).
252 */
253static void dn_dst_update_pmtu(struct dst_entry *dst, struct sock *sk,
254 struct sk_buff *skb, u32 mtu)
255{
256 struct dn_route *rt = (struct dn_route *) dst;
257 struct neighbour *n = rt->n;
258 u32 min_mtu = 230;
259 struct dn_dev *dn;
260
261 dn = n ? rcu_dereference_raw(n->dev->dn_ptr) : NULL;
262
263 if (dn && dn->use_long == 0)
264 min_mtu -= 6;
265 else
266 min_mtu -= 21;
267
268 if (dst_metric(dst, RTAX_MTU) > mtu && mtu >= min_mtu) {
269 if (!(dst_metric_locked(dst, RTAX_MTU))) {
270 dst_metric_set(dst, RTAX_MTU, mtu);
271 dst_set_expires(dst, dn_rt_mtu_expires);
272 }
273 if (!(dst_metric_locked(dst, RTAX_ADVMSS))) {
274 u32 mss = mtu - DN_MAX_NSP_DATA_HEADER;
275 u32 existing_mss = dst_metric_raw(dst, RTAX_ADVMSS);
276 if (!existing_mss || existing_mss > mss)
277 dst_metric_set(dst, RTAX_ADVMSS, mss);
278 }
279 }
280}
281
282static void dn_dst_redirect(struct dst_entry *dst, struct sock *sk,
283 struct sk_buff *skb)
284{
285}
286
287/*
288 * When a route has been marked obsolete. (e.g. routing cache flush)
289 */
290static struct dst_entry *dn_dst_check(struct dst_entry *dst, __u32 cookie)
291{
292 return NULL;
293}
294
295static struct dst_entry *dn_dst_negative_advice(struct dst_entry *dst)
296{
297 dst_release(dst);
298 return NULL;
299}
300
301static void dn_dst_link_failure(struct sk_buff *skb)
302{
303}
304
305static inline int compare_keys(struct flowidn *fl1, struct flowidn *fl2)
306{
307 return ((fl1->daddr ^ fl2->daddr) |
308 (fl1->saddr ^ fl2->saddr) |
309 (fl1->flowidn_mark ^ fl2->flowidn_mark) |
310 (fl1->flowidn_scope ^ fl2->flowidn_scope) |
311 (fl1->flowidn_oif ^ fl2->flowidn_oif) |
312 (fl1->flowidn_iif ^ fl2->flowidn_iif)) == 0;
313}
314
315static int dn_insert_route(struct dn_route *rt, unsigned int hash, struct dn_route **rp)
316{
317 struct dn_route *rth;
318 struct dn_route __rcu **rthp;
319 unsigned long now = jiffies;
320
321 rthp = &dn_rt_hash_table[hash].chain;
322
323 spin_lock_bh(&dn_rt_hash_table[hash].lock);
324 while ((rth = rcu_dereference_protected(*rthp,
325 lockdep_is_held(&dn_rt_hash_table[hash].lock))) != NULL) {
326 if (compare_keys(&rth->fld, &rt->fld)) {
327 /* Put it first */
328 *rthp = rth->dn_next;
329 rcu_assign_pointer(rth->dn_next,
330 dn_rt_hash_table[hash].chain);
331 rcu_assign_pointer(dn_rt_hash_table[hash].chain, rth);
332
333 dst_hold_and_use(&rth->dst, now);
334 spin_unlock_bh(&dn_rt_hash_table[hash].lock);
335
336 dst_release_immediate(&rt->dst);
337 *rp = rth;
338 return 0;
339 }
340 rthp = &rth->dn_next;
341 }
342
343 rcu_assign_pointer(rt->dn_next, dn_rt_hash_table[hash].chain);
344 rcu_assign_pointer(dn_rt_hash_table[hash].chain, rt);
345
346 dst_hold_and_use(&rt->dst, now);
347 spin_unlock_bh(&dn_rt_hash_table[hash].lock);
348 *rp = rt;
349 return 0;
350}
351
352static void dn_run_flush(struct timer_list *unused)
353{
354 int i;
355 struct dn_route *rt, *next;
356
357 for (i = 0; i < dn_rt_hash_mask; i++) {
358 spin_lock_bh(&dn_rt_hash_table[i].lock);
359
360 if ((rt = xchg((struct dn_route **)&dn_rt_hash_table[i].chain, NULL)) == NULL)
361 goto nothing_to_declare;
362
363 for(; rt; rt = next) {
364 next = rcu_dereference_raw(rt->dn_next);
365 RCU_INIT_POINTER(rt->dn_next, NULL);
366 dst_dev_put(&rt->dst);
367 dst_release(&rt->dst);
368 }
369
370nothing_to_declare:
371 spin_unlock_bh(&dn_rt_hash_table[i].lock);
372 }
373}
374
375static DEFINE_SPINLOCK(dn_rt_flush_lock);
376
377void dn_rt_cache_flush(int delay)
378{
379 unsigned long now = jiffies;
380 int user_mode = !in_interrupt();
381
382 if (delay < 0)
383 delay = dn_rt_min_delay;
384
385 spin_lock_bh(&dn_rt_flush_lock);
386
387 if (del_timer(&dn_rt_flush_timer) && delay > 0 && dn_rt_deadline) {
388 long tmo = (long)(dn_rt_deadline - now);
389
390 if (user_mode && tmo < dn_rt_max_delay - dn_rt_min_delay)
391 tmo = 0;
392
393 if (delay > tmo)
394 delay = tmo;
395 }
396
397 if (delay <= 0) {
398 spin_unlock_bh(&dn_rt_flush_lock);
399 dn_run_flush(NULL);
400 return;
401 }
402
403 if (dn_rt_deadline == 0)
404 dn_rt_deadline = now + dn_rt_max_delay;
405
406 dn_rt_flush_timer.expires = now + delay;
407 add_timer(&dn_rt_flush_timer);
408 spin_unlock_bh(&dn_rt_flush_lock);
409}
410
411/**
412 * dn_return_short - Return a short packet to its sender
413 * @skb: The packet to return
414 *
415 */
416static int dn_return_short(struct sk_buff *skb)
417{
418 struct dn_skb_cb *cb;
419 unsigned char *ptr;
420 __le16 *src;
421 __le16 *dst;
422
423 /* Add back headers */
424 skb_push(skb, skb->data - skb_network_header(skb));
425
426 if ((skb = skb_unshare(skb, GFP_ATOMIC)) == NULL)
427 return NET_RX_DROP;
428
429 cb = DN_SKB_CB(skb);
430 /* Skip packet length and point to flags */
431 ptr = skb->data + 2;
432 *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
433
434 dst = (__le16 *)ptr;
435 ptr += 2;
436 src = (__le16 *)ptr;
437 ptr += 2;
438 *ptr = 0; /* Zero hop count */
439
440 swap(*src, *dst);
441
442 skb->pkt_type = PACKET_OUTGOING;
443 dn_rt_finish_output(skb, NULL, NULL);
444 return NET_RX_SUCCESS;
445}
446
447/**
448 * dn_return_long - Return a long packet to its sender
449 * @skb: The long format packet to return
450 *
451 */
452static int dn_return_long(struct sk_buff *skb)
453{
454 struct dn_skb_cb *cb;
455 unsigned char *ptr;
456 unsigned char *src_addr, *dst_addr;
457 unsigned char tmp[ETH_ALEN];
458
459 /* Add back all headers */
460 skb_push(skb, skb->data - skb_network_header(skb));
461
462 if ((skb = skb_unshare(skb, GFP_ATOMIC)) == NULL)
463 return NET_RX_DROP;
464
465 cb = DN_SKB_CB(skb);
466 /* Ignore packet length and point to flags */
467 ptr = skb->data + 2;
468
469 /* Skip padding */
470 if (*ptr & DN_RT_F_PF) {
471 char padlen = (*ptr & ~DN_RT_F_PF);
472 ptr += padlen;
473 }
474
475 *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
476 ptr += 2;
477 dst_addr = ptr;
478 ptr += 8;
479 src_addr = ptr;
480 ptr += 6;
481 *ptr = 0; /* Zero hop count */
482
483 /* Swap source and destination */
484 memcpy(tmp, src_addr, ETH_ALEN);
485 memcpy(src_addr, dst_addr, ETH_ALEN);
486 memcpy(dst_addr, tmp, ETH_ALEN);
487
488 skb->pkt_type = PACKET_OUTGOING;
489 dn_rt_finish_output(skb, dst_addr, src_addr);
490 return NET_RX_SUCCESS;
491}
492
493/**
494 * dn_route_rx_packet - Try and find a route for an incoming packet
495 * @skb: The packet to find a route for
496 *
497 * Returns: result of input function if route is found, error code otherwise
498 */
499static int dn_route_rx_packet(struct net *net, struct sock *sk, struct sk_buff *skb)
500{
501 struct dn_skb_cb *cb;
502 int err;
503
504 if ((err = dn_route_input(skb)) == 0)
505 return dst_input(skb);
506
507 cb = DN_SKB_CB(skb);
508 if (decnet_debug_level & 4) {
509 char *devname = skb->dev ? skb->dev->name : "???";
510
511 printk(KERN_DEBUG
512 "DECnet: dn_route_rx_packet: rt_flags=0x%02x dev=%s len=%d src=0x%04hx dst=0x%04hx err=%d type=%d\n",
513 (int)cb->rt_flags, devname, skb->len,
514 le16_to_cpu(cb->src), le16_to_cpu(cb->dst),
515 err, skb->pkt_type);
516 }
517
518 if ((skb->pkt_type == PACKET_HOST) && (cb->rt_flags & DN_RT_F_RQR)) {
519 switch (cb->rt_flags & DN_RT_PKT_MSK) {
520 case DN_RT_PKT_SHORT:
521 return dn_return_short(skb);
522 case DN_RT_PKT_LONG:
523 return dn_return_long(skb);
524 }
525 }
526
527 kfree_skb(skb);
528 return NET_RX_DROP;
529}
530
531static int dn_route_rx_long(struct sk_buff *skb)
532{
533 struct dn_skb_cb *cb = DN_SKB_CB(skb);
534 unsigned char *ptr = skb->data;
535
536 if (!pskb_may_pull(skb, 21)) /* 20 for long header, 1 for shortest nsp */
537 goto drop_it;
538
539 skb_pull(skb, 20);
540 skb_reset_transport_header(skb);
541
542 /* Destination info */
543 ptr += 2;
544 cb->dst = dn_eth2dn(ptr);
545 if (memcmp(ptr, dn_hiord_addr, 4) != 0)
546 goto drop_it;
547 ptr += 6;
548
549
550 /* Source info */
551 ptr += 2;
552 cb->src = dn_eth2dn(ptr);
553 if (memcmp(ptr, dn_hiord_addr, 4) != 0)
554 goto drop_it;
555 ptr += 6;
556 /* Other junk */
557 ptr++;
558 cb->hops = *ptr++; /* Visit Count */
559
560 return NF_HOOK(NFPROTO_DECNET, NF_DN_PRE_ROUTING,
561 &init_net, NULL, skb, skb->dev, NULL,
562 dn_route_rx_packet);
563
564drop_it:
565 kfree_skb(skb);
566 return NET_RX_DROP;
567}
568
569
570
571static int dn_route_rx_short(struct sk_buff *skb)
572{
573 struct dn_skb_cb *cb = DN_SKB_CB(skb);
574 unsigned char *ptr = skb->data;
575
576 if (!pskb_may_pull(skb, 6)) /* 5 for short header + 1 for shortest nsp */
577 goto drop_it;
578
579 skb_pull(skb, 5);
580 skb_reset_transport_header(skb);
581
582 cb->dst = *(__le16 *)ptr;
583 ptr += 2;
584 cb->src = *(__le16 *)ptr;
585 ptr += 2;
586 cb->hops = *ptr & 0x3f;
587
588 return NF_HOOK(NFPROTO_DECNET, NF_DN_PRE_ROUTING,
589 &init_net, NULL, skb, skb->dev, NULL,
590 dn_route_rx_packet);
591
592drop_it:
593 kfree_skb(skb);
594 return NET_RX_DROP;
595}
596
597static int dn_route_discard(struct net *net, struct sock *sk, struct sk_buff *skb)
598{
599 /*
600 * I know we drop the packet here, but thats considered success in
601 * this case
602 */
603 kfree_skb(skb);
604 return NET_RX_SUCCESS;
605}
606
607static int dn_route_ptp_hello(struct net *net, struct sock *sk, struct sk_buff *skb)
608{
609 dn_dev_hello(skb);
610 dn_neigh_pointopoint_hello(skb);
611 return NET_RX_SUCCESS;
612}
613
614int dn_route_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
615{
616 struct dn_skb_cb *cb;
617 unsigned char flags = 0;
618 __u16 len = le16_to_cpu(*(__le16 *)skb->data);
619 struct dn_dev *dn = rcu_dereference(dev->dn_ptr);
620 unsigned char padlen = 0;
621
622 if (!net_eq(dev_net(dev), &init_net))
623 goto dump_it;
624
625 if (dn == NULL)
626 goto dump_it;
627
628 if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
629 goto out;
630
631 if (!pskb_may_pull(skb, 3))
632 goto dump_it;
633
634 skb_pull(skb, 2);
635
636 if (len > skb->len)
637 goto dump_it;
638
639 skb_trim(skb, len);
640
641 flags = *skb->data;
642
643 cb = DN_SKB_CB(skb);
644 cb->stamp = jiffies;
645 cb->iif = dev->ifindex;
646
647 /*
648 * If we have padding, remove it.
649 */
650 if (flags & DN_RT_F_PF) {
651 padlen = flags & ~DN_RT_F_PF;
652 if (!pskb_may_pull(skb, padlen + 1))
653 goto dump_it;
654 skb_pull(skb, padlen);
655 flags = *skb->data;
656 }
657
658 skb_reset_network_header(skb);
659
660 /*
661 * Weed out future version DECnet
662 */
663 if (flags & DN_RT_F_VER)
664 goto dump_it;
665
666 cb->rt_flags = flags;
667
668 if (decnet_debug_level & 1)
669 printk(KERN_DEBUG
670 "dn_route_rcv: got 0x%02x from %s [%d %d %d]\n",
671 (int)flags, (dev) ? dev->name : "???", len, skb->len,
672 padlen);
673
674 if (flags & DN_RT_PKT_CNTL) {
675 if (unlikely(skb_linearize(skb)))
676 goto dump_it;
677
678 switch (flags & DN_RT_CNTL_MSK) {
679 case DN_RT_PKT_INIT:
680 dn_dev_init_pkt(skb);
681 break;
682 case DN_RT_PKT_VERI:
683 dn_dev_veri_pkt(skb);
684 break;
685 }
686
687 if (dn->parms.state != DN_DEV_S_RU)
688 goto dump_it;
689
690 switch (flags & DN_RT_CNTL_MSK) {
691 case DN_RT_PKT_HELO:
692 return NF_HOOK(NFPROTO_DECNET, NF_DN_HELLO,
693 &init_net, NULL, skb, skb->dev, NULL,
694 dn_route_ptp_hello);
695
696 case DN_RT_PKT_L1RT:
697 case DN_RT_PKT_L2RT:
698 return NF_HOOK(NFPROTO_DECNET, NF_DN_ROUTE,
699 &init_net, NULL, skb, skb->dev, NULL,
700 dn_route_discard);
701 case DN_RT_PKT_ERTH:
702 return NF_HOOK(NFPROTO_DECNET, NF_DN_HELLO,
703 &init_net, NULL, skb, skb->dev, NULL,
704 dn_neigh_router_hello);
705
706 case DN_RT_PKT_EEDH:
707 return NF_HOOK(NFPROTO_DECNET, NF_DN_HELLO,
708 &init_net, NULL, skb, skb->dev, NULL,
709 dn_neigh_endnode_hello);
710 }
711 } else {
712 if (dn->parms.state != DN_DEV_S_RU)
713 goto dump_it;
714
715 skb_pull(skb, 1); /* Pull flags */
716
717 switch (flags & DN_RT_PKT_MSK) {
718 case DN_RT_PKT_LONG:
719 return dn_route_rx_long(skb);
720 case DN_RT_PKT_SHORT:
721 return dn_route_rx_short(skb);
722 }
723 }
724
725dump_it:
726 kfree_skb(skb);
727out:
728 return NET_RX_DROP;
729}
730
731static int dn_output(struct net *net, struct sock *sk, struct sk_buff *skb)
732{
733 struct dst_entry *dst = skb_dst(skb);
734 struct dn_route *rt = (struct dn_route *)dst;
735 struct net_device *dev = dst->dev;
736 struct dn_skb_cb *cb = DN_SKB_CB(skb);
737
738 int err = -EINVAL;
739
740 if (rt->n == NULL)
741 goto error;
742
743 skb->dev = dev;
744
745 cb->src = rt->rt_saddr;
746 cb->dst = rt->rt_daddr;
747
748 /*
749 * Always set the Intra-Ethernet bit on all outgoing packets
750 * originated on this node. Only valid flag from upper layers
751 * is return-to-sender-requested. Set hop count to 0 too.
752 */
753 cb->rt_flags &= ~DN_RT_F_RQR;
754 cb->rt_flags |= DN_RT_F_IE;
755 cb->hops = 0;
756
757 return NF_HOOK(NFPROTO_DECNET, NF_DN_LOCAL_OUT,
758 &init_net, sk, skb, NULL, dev,
759 dn_to_neigh_output);
760
761error:
762 net_dbg_ratelimited("dn_output: This should not happen\n");
763
764 kfree_skb(skb);
765
766 return err;
767}
768
769static int dn_forward(struct sk_buff *skb)
770{
771 struct dn_skb_cb *cb = DN_SKB_CB(skb);
772 struct dst_entry *dst = skb_dst(skb);
773 struct dn_dev *dn_db = rcu_dereference(dst->dev->dn_ptr);
774 struct dn_route *rt;
775 int header_len;
776 struct net_device *dev = skb->dev;
777
778 if (skb->pkt_type != PACKET_HOST)
779 goto drop;
780
781 /* Ensure that we have enough space for headers */
782 rt = (struct dn_route *)skb_dst(skb);
783 header_len = dn_db->use_long ? 21 : 6;
784 if (skb_cow(skb, LL_RESERVED_SPACE(rt->dst.dev)+header_len))
785 goto drop;
786
787 /*
788 * Hop count exceeded.
789 */
790 if (++cb->hops > 30)
791 goto drop;
792
793 skb->dev = rt->dst.dev;
794
795 /*
796 * If packet goes out same interface it came in on, then set
797 * the Intra-Ethernet bit. This has no effect for short
798 * packets, so we don't need to test for them here.
799 */
800 cb->rt_flags &= ~DN_RT_F_IE;
801 if (rt->rt_flags & RTCF_DOREDIRECT)
802 cb->rt_flags |= DN_RT_F_IE;
803
804 return NF_HOOK(NFPROTO_DECNET, NF_DN_FORWARD,
805 &init_net, NULL, skb, dev, skb->dev,
806 dn_to_neigh_output);
807
808drop:
809 kfree_skb(skb);
810 return NET_RX_DROP;
811}
812
813/*
814 * Used to catch bugs. This should never normally get
815 * called.
816 */
817static int dn_rt_bug_out(struct net *net, struct sock *sk, struct sk_buff *skb)
818{
819 struct dn_skb_cb *cb = DN_SKB_CB(skb);
820
821 net_dbg_ratelimited("dn_rt_bug: skb from:%04x to:%04x\n",
822 le16_to_cpu(cb->src), le16_to_cpu(cb->dst));
823
824 kfree_skb(skb);
825
826 return NET_RX_DROP;
827}
828
829static int dn_rt_bug(struct sk_buff *skb)
830{
831 struct dn_skb_cb *cb = DN_SKB_CB(skb);
832
833 net_dbg_ratelimited("dn_rt_bug: skb from:%04x to:%04x\n",
834 le16_to_cpu(cb->src), le16_to_cpu(cb->dst));
835
836 kfree_skb(skb);
837
838 return NET_RX_DROP;
839}
840
841static unsigned int dn_dst_default_advmss(const struct dst_entry *dst)
842{
843 return dn_mss_from_pmtu(dst->dev, dst_mtu(dst));
844}
845
846static unsigned int dn_dst_mtu(const struct dst_entry *dst)
847{
848 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
849
850 return mtu ? : dst->dev->mtu;
851}
852
853static struct neighbour *dn_dst_neigh_lookup(const struct dst_entry *dst,
854 struct sk_buff *skb,
855 const void *daddr)
856{
857 return __neigh_lookup_errno(&dn_neigh_table, daddr, dst->dev);
858}
859
860static int dn_rt_set_next_hop(struct dn_route *rt, struct dn_fib_res *res)
861{
862 struct dn_fib_info *fi = res->fi;
863 struct net_device *dev = rt->dst.dev;
864 unsigned int mss_metric;
865 struct neighbour *n;
866
867 if (fi) {
868 if (DN_FIB_RES_GW(*res) &&
869 DN_FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
870 rt->rt_gateway = DN_FIB_RES_GW(*res);
871 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
872 }
873 rt->rt_type = res->type;
874
875 if (dev != NULL && rt->n == NULL) {
876 n = __neigh_lookup_errno(&dn_neigh_table, &rt->rt_gateway, dev);
877 if (IS_ERR(n))
878 return PTR_ERR(n);
879 rt->n = n;
880 }
881
882 if (dst_metric(&rt->dst, RTAX_MTU) > rt->dst.dev->mtu)
883 dst_metric_set(&rt->dst, RTAX_MTU, rt->dst.dev->mtu);
884 mss_metric = dst_metric_raw(&rt->dst, RTAX_ADVMSS);
885 if (mss_metric) {
886 unsigned int mss = dn_mss_from_pmtu(dev, dst_mtu(&rt->dst));
887 if (mss_metric > mss)
888 dst_metric_set(&rt->dst, RTAX_ADVMSS, mss);
889 }
890 return 0;
891}
892
893static inline int dn_match_addr(__le16 addr1, __le16 addr2)
894{
895 __u16 tmp = le16_to_cpu(addr1) ^ le16_to_cpu(addr2);
896 int match = 16;
897 while(tmp) {
898 tmp >>= 1;
899 match--;
900 }
901 return match;
902}
903
904static __le16 dnet_select_source(const struct net_device *dev, __le16 daddr, int scope)
905{
906 __le16 saddr = 0;
907 struct dn_dev *dn_db;
908 struct dn_ifaddr *ifa;
909 int best_match = 0;
910 int ret;
911
912 rcu_read_lock();
913 dn_db = rcu_dereference(dev->dn_ptr);
914 for (ifa = rcu_dereference(dn_db->ifa_list);
915 ifa != NULL;
916 ifa = rcu_dereference(ifa->ifa_next)) {
917 if (ifa->ifa_scope > scope)
918 continue;
919 if (!daddr) {
920 saddr = ifa->ifa_local;
921 break;
922 }
923 ret = dn_match_addr(daddr, ifa->ifa_local);
924 if (ret > best_match)
925 saddr = ifa->ifa_local;
926 if (best_match == 0)
927 saddr = ifa->ifa_local;
928 }
929 rcu_read_unlock();
930
931 return saddr;
932}
933
934static inline __le16 __dn_fib_res_prefsrc(struct dn_fib_res *res)
935{
936 return dnet_select_source(DN_FIB_RES_DEV(*res), DN_FIB_RES_GW(*res), res->scope);
937}
938
939static inline __le16 dn_fib_rules_map_destination(__le16 daddr, struct dn_fib_res *res)
940{
941 __le16 mask = dnet_make_mask(res->prefixlen);
942 return (daddr&~mask)|res->fi->fib_nh->nh_gw;
943}
944
945static int dn_route_output_slow(struct dst_entry **pprt, const struct flowidn *oldflp, int try_hard)
946{
947 struct flowidn fld = {
948 .daddr = oldflp->daddr,
949 .saddr = oldflp->saddr,
950 .flowidn_scope = RT_SCOPE_UNIVERSE,
951 .flowidn_mark = oldflp->flowidn_mark,
952 .flowidn_iif = LOOPBACK_IFINDEX,
953 .flowidn_oif = oldflp->flowidn_oif,
954 };
955 struct dn_route *rt = NULL;
956 struct net_device *dev_out = NULL, *dev;
957 struct neighbour *neigh = NULL;
958 unsigned int hash;
959 unsigned int flags = 0;
960 struct dn_fib_res res = { .fi = NULL, .type = RTN_UNICAST };
961 int err;
962 int free_res = 0;
963 __le16 gateway = 0;
964
965 if (decnet_debug_level & 16)
966 printk(KERN_DEBUG
967 "dn_route_output_slow: dst=%04x src=%04x mark=%d"
968 " iif=%d oif=%d\n", le16_to_cpu(oldflp->daddr),
969 le16_to_cpu(oldflp->saddr),
970 oldflp->flowidn_mark, LOOPBACK_IFINDEX,
971 oldflp->flowidn_oif);
972
973 /* If we have an output interface, verify its a DECnet device */
974 if (oldflp->flowidn_oif) {
975 dev_out = dev_get_by_index(&init_net, oldflp->flowidn_oif);
976 err = -ENODEV;
977 if (dev_out && dev_out->dn_ptr == NULL) {
978 dev_put(dev_out);
979 dev_out = NULL;
980 }
981 if (dev_out == NULL)
982 goto out;
983 }
984
985 /* If we have a source address, verify that its a local address */
986 if (oldflp->saddr) {
987 err = -EADDRNOTAVAIL;
988
989 if (dev_out) {
990 if (dn_dev_islocal(dev_out, oldflp->saddr))
991 goto source_ok;
992 dev_put(dev_out);
993 goto out;
994 }
995 rcu_read_lock();
996 for_each_netdev_rcu(&init_net, dev) {
997 if (!dev->dn_ptr)
998 continue;
999 if (!dn_dev_islocal(dev, oldflp->saddr))
1000 continue;
1001 if ((dev->flags & IFF_LOOPBACK) &&
1002 oldflp->daddr &&
1003 !dn_dev_islocal(dev, oldflp->daddr))
1004 continue;
1005
1006 dev_out = dev;
1007 break;
1008 }
1009 rcu_read_unlock();
1010 if (dev_out == NULL)
1011 goto out;
1012 dev_hold(dev_out);
1013source_ok:
1014 ;
1015 }
1016
1017 /* No destination? Assume its local */
1018 if (!fld.daddr) {
1019 fld.daddr = fld.saddr;
1020
1021 if (dev_out)
1022 dev_put(dev_out);
1023 err = -EINVAL;
1024 dev_out = init_net.loopback_dev;
1025 if (!dev_out->dn_ptr)
1026 goto out;
1027 err = -EADDRNOTAVAIL;
1028 dev_hold(dev_out);
1029 if (!fld.daddr) {
1030 fld.daddr =
1031 fld.saddr = dnet_select_source(dev_out, 0,
1032 RT_SCOPE_HOST);
1033 if (!fld.daddr)
1034 goto out;
1035 }
1036 fld.flowidn_oif = LOOPBACK_IFINDEX;
1037 res.type = RTN_LOCAL;
1038 goto make_route;
1039 }
1040
1041 if (decnet_debug_level & 16)
1042 printk(KERN_DEBUG
1043 "dn_route_output_slow: initial checks complete."
1044 " dst=%04x src=%04x oif=%d try_hard=%d\n",
1045 le16_to_cpu(fld.daddr), le16_to_cpu(fld.saddr),
1046 fld.flowidn_oif, try_hard);
1047
1048 /*
1049 * N.B. If the kernel is compiled without router support then
1050 * dn_fib_lookup() will evaluate to non-zero so this if () block
1051 * will always be executed.
1052 */
1053 err = -ESRCH;
1054 if (try_hard || (err = dn_fib_lookup(&fld, &res)) != 0) {
1055 struct dn_dev *dn_db;
1056 if (err != -ESRCH)
1057 goto out;
1058 /*
1059 * Here the fallback is basically the standard algorithm for
1060 * routing in endnodes which is described in the DECnet routing
1061 * docs
1062 *
1063 * If we are not trying hard, look in neighbour cache.
1064 * The result is tested to ensure that if a specific output
1065 * device/source address was requested, then we honour that
1066 * here
1067 */
1068 if (!try_hard) {
1069 neigh = neigh_lookup_nodev(&dn_neigh_table, &init_net, &fld.daddr);
1070 if (neigh) {
1071 if ((oldflp->flowidn_oif &&
1072 (neigh->dev->ifindex != oldflp->flowidn_oif)) ||
1073 (oldflp->saddr &&
1074 (!dn_dev_islocal(neigh->dev,
1075 oldflp->saddr)))) {
1076 neigh_release(neigh);
1077 neigh = NULL;
1078 } else {
1079 if (dev_out)
1080 dev_put(dev_out);
1081 if (dn_dev_islocal(neigh->dev, fld.daddr)) {
1082 dev_out = init_net.loopback_dev;
1083 res.type = RTN_LOCAL;
1084 } else {
1085 dev_out = neigh->dev;
1086 }
1087 dev_hold(dev_out);
1088 goto select_source;
1089 }
1090 }
1091 }
1092
1093 /* Not there? Perhaps its a local address */
1094 if (dev_out == NULL)
1095 dev_out = dn_dev_get_default();
1096 err = -ENODEV;
1097 if (dev_out == NULL)
1098 goto out;
1099 dn_db = rcu_dereference_raw(dev_out->dn_ptr);
1100 if (!dn_db)
1101 goto e_inval;
1102 /* Possible improvement - check all devices for local addr */
1103 if (dn_dev_islocal(dev_out, fld.daddr)) {
1104 dev_put(dev_out);
1105 dev_out = init_net.loopback_dev;
1106 dev_hold(dev_out);
1107 res.type = RTN_LOCAL;
1108 goto select_source;
1109 }
1110 /* Not local either.... try sending it to the default router */
1111 neigh = neigh_clone(dn_db->router);
1112 BUG_ON(neigh && neigh->dev != dev_out);
1113
1114 /* Ok then, we assume its directly connected and move on */
1115select_source:
1116 if (neigh)
1117 gateway = ((struct dn_neigh *)neigh)->addr;
1118 if (gateway == 0)
1119 gateway = fld.daddr;
1120 if (fld.saddr == 0) {
1121 fld.saddr = dnet_select_source(dev_out, gateway,
1122 res.type == RTN_LOCAL ?
1123 RT_SCOPE_HOST :
1124 RT_SCOPE_LINK);
1125 if (fld.saddr == 0 && res.type != RTN_LOCAL)
1126 goto e_addr;
1127 }
1128 fld.flowidn_oif = dev_out->ifindex;
1129 goto make_route;
1130 }
1131 free_res = 1;
1132
1133 if (res.type == RTN_NAT)
1134 goto e_inval;
1135
1136 if (res.type == RTN_LOCAL) {
1137 if (!fld.saddr)
1138 fld.saddr = fld.daddr;
1139 if (dev_out)
1140 dev_put(dev_out);
1141 dev_out = init_net.loopback_dev;
1142 dev_hold(dev_out);
1143 if (!dev_out->dn_ptr)
1144 goto e_inval;
1145 fld.flowidn_oif = dev_out->ifindex;
1146 if (res.fi)
1147 dn_fib_info_put(res.fi);
1148 res.fi = NULL;
1149 goto make_route;
1150 }
1151
1152 if (res.fi->fib_nhs > 1 && fld.flowidn_oif == 0)
1153 dn_fib_select_multipath(&fld, &res);
1154
1155 /*
1156 * We could add some logic to deal with default routes here and
1157 * get rid of some of the special casing above.
1158 */
1159
1160 if (!fld.saddr)
1161 fld.saddr = DN_FIB_RES_PREFSRC(res);
1162
1163 if (dev_out)
1164 dev_put(dev_out);
1165 dev_out = DN_FIB_RES_DEV(res);
1166 dev_hold(dev_out);
1167 fld.flowidn_oif = dev_out->ifindex;
1168 gateway = DN_FIB_RES_GW(res);
1169
1170make_route:
1171 if (dev_out->flags & IFF_LOOPBACK)
1172 flags |= RTCF_LOCAL;
1173
1174 rt = dst_alloc(&dn_dst_ops, dev_out, 0, DST_OBSOLETE_NONE, DST_HOST);
1175 if (rt == NULL)
1176 goto e_nobufs;
1177
1178 rt->dn_next = NULL;
1179 memset(&rt->fld, 0, sizeof(rt->fld));
1180 rt->fld.saddr = oldflp->saddr;
1181 rt->fld.daddr = oldflp->daddr;
1182 rt->fld.flowidn_oif = oldflp->flowidn_oif;
1183 rt->fld.flowidn_iif = 0;
1184 rt->fld.flowidn_mark = oldflp->flowidn_mark;
1185
1186 rt->rt_saddr = fld.saddr;
1187 rt->rt_daddr = fld.daddr;
1188 rt->rt_gateway = gateway ? gateway : fld.daddr;
1189 rt->rt_local_src = fld.saddr;
1190
1191 rt->rt_dst_map = fld.daddr;
1192 rt->rt_src_map = fld.saddr;
1193
1194 rt->n = neigh;
1195 neigh = NULL;
1196
1197 rt->dst.lastuse = jiffies;
1198 rt->dst.output = dn_output;
1199 rt->dst.input = dn_rt_bug;
1200 rt->rt_flags = flags;
1201 if (flags & RTCF_LOCAL)
1202 rt->dst.input = dn_nsp_rx;
1203
1204 err = dn_rt_set_next_hop(rt, &res);
1205 if (err)
1206 goto e_neighbour;
1207
1208 hash = dn_hash(rt->fld.saddr, rt->fld.daddr);
1209 /* dn_insert_route() increments dst->__refcnt */
1210 dn_insert_route(rt, hash, (struct dn_route **)pprt);
1211
1212done:
1213 if (neigh)
1214 neigh_release(neigh);
1215 if (free_res)
1216 dn_fib_res_put(&res);
1217 if (dev_out)
1218 dev_put(dev_out);
1219out:
1220 return err;
1221
1222e_addr:
1223 err = -EADDRNOTAVAIL;
1224 goto done;
1225e_inval:
1226 err = -EINVAL;
1227 goto done;
1228e_nobufs:
1229 err = -ENOBUFS;
1230 goto done;
1231e_neighbour:
1232 dst_release_immediate(&rt->dst);
1233 goto e_nobufs;
1234}
1235
1236
1237/*
1238 * N.B. The flags may be moved into the flowi at some future stage.
1239 */
1240static int __dn_route_output_key(struct dst_entry **pprt, const struct flowidn *flp, int flags)
1241{
1242 unsigned int hash = dn_hash(flp->saddr, flp->daddr);
1243 struct dn_route *rt = NULL;
1244
1245 if (!(flags & MSG_TRYHARD)) {
1246 rcu_read_lock_bh();
1247 for (rt = rcu_dereference_bh(dn_rt_hash_table[hash].chain); rt;
1248 rt = rcu_dereference_bh(rt->dn_next)) {
1249 if ((flp->daddr == rt->fld.daddr) &&
1250 (flp->saddr == rt->fld.saddr) &&
1251 (flp->flowidn_mark == rt->fld.flowidn_mark) &&
1252 dn_is_output_route(rt) &&
1253 (rt->fld.flowidn_oif == flp->flowidn_oif)) {
1254 dst_hold_and_use(&rt->dst, jiffies);
1255 rcu_read_unlock_bh();
1256 *pprt = &rt->dst;
1257 return 0;
1258 }
1259 }
1260 rcu_read_unlock_bh();
1261 }
1262
1263 return dn_route_output_slow(pprt, flp, flags);
1264}
1265
1266static int dn_route_output_key(struct dst_entry **pprt, struct flowidn *flp, int flags)
1267{
1268 int err;
1269
1270 err = __dn_route_output_key(pprt, flp, flags);
1271 if (err == 0 && flp->flowidn_proto) {
1272 *pprt = xfrm_lookup(&init_net, *pprt,
1273 flowidn_to_flowi(flp), NULL, 0);
1274 if (IS_ERR(*pprt)) {
1275 err = PTR_ERR(*pprt);
1276 *pprt = NULL;
1277 }
1278 }
1279 return err;
1280}
1281
1282int dn_route_output_sock(struct dst_entry __rcu **pprt, struct flowidn *fl, struct sock *sk, int flags)
1283{
1284 int err;
1285
1286 err = __dn_route_output_key(pprt, fl, flags & MSG_TRYHARD);
1287 if (err == 0 && fl->flowidn_proto) {
1288 *pprt = xfrm_lookup(&init_net, *pprt,
1289 flowidn_to_flowi(fl), sk, 0);
1290 if (IS_ERR(*pprt)) {
1291 err = PTR_ERR(*pprt);
1292 *pprt = NULL;
1293 }
1294 }
1295 return err;
1296}
1297
1298static int dn_route_input_slow(struct sk_buff *skb)
1299{
1300 struct dn_route *rt = NULL;
1301 struct dn_skb_cb *cb = DN_SKB_CB(skb);
1302 struct net_device *in_dev = skb->dev;
1303 struct net_device *out_dev = NULL;
1304 struct dn_dev *dn_db;
1305 struct neighbour *neigh = NULL;
1306 unsigned int hash;
1307 int flags = 0;
1308 __le16 gateway = 0;
1309 __le16 local_src = 0;
1310 struct flowidn fld = {
1311 .daddr = cb->dst,
1312 .saddr = cb->src,
1313 .flowidn_scope = RT_SCOPE_UNIVERSE,
1314 .flowidn_mark = skb->mark,
1315 .flowidn_iif = skb->dev->ifindex,
1316 };
1317 struct dn_fib_res res = { .fi = NULL, .type = RTN_UNREACHABLE };
1318 int err = -EINVAL;
1319 int free_res = 0;
1320
1321 dev_hold(in_dev);
1322
1323 if ((dn_db = rcu_dereference(in_dev->dn_ptr)) == NULL)
1324 goto out;
1325
1326 /* Zero source addresses are not allowed */
1327 if (fld.saddr == 0)
1328 goto out;
1329
1330 /*
1331 * In this case we've just received a packet from a source
1332 * outside ourselves pretending to come from us. We don't
1333 * allow it any further to prevent routing loops, spoofing and
1334 * other nasties. Loopback packets already have the dst attached
1335 * so this only affects packets which have originated elsewhere.
1336 */
1337 err = -ENOTUNIQ;
1338 if (dn_dev_islocal(in_dev, cb->src))
1339 goto out;
1340
1341 err = dn_fib_lookup(&fld, &res);
1342 if (err) {
1343 if (err != -ESRCH)
1344 goto out;
1345 /*
1346 * Is the destination us ?
1347 */
1348 if (!dn_dev_islocal(in_dev, cb->dst))
1349 goto e_inval;
1350
1351 res.type = RTN_LOCAL;
1352 } else {
1353 __le16 src_map = fld.saddr;
1354 free_res = 1;
1355
1356 out_dev = DN_FIB_RES_DEV(res);
1357 if (out_dev == NULL) {
1358 net_crit_ratelimited("Bug in dn_route_input_slow() No output device\n");
1359 goto e_inval;
1360 }
1361 dev_hold(out_dev);
1362
1363 if (res.r)
1364 src_map = fld.saddr; /* no NAT support for now */
1365
1366 gateway = DN_FIB_RES_GW(res);
1367 if (res.type == RTN_NAT) {
1368 fld.daddr = dn_fib_rules_map_destination(fld.daddr, &res);
1369 dn_fib_res_put(&res);
1370 free_res = 0;
1371 if (dn_fib_lookup(&fld, &res))
1372 goto e_inval;
1373 free_res = 1;
1374 if (res.type != RTN_UNICAST)
1375 goto e_inval;
1376 flags |= RTCF_DNAT;
1377 gateway = fld.daddr;
1378 }
1379 fld.saddr = src_map;
1380 }
1381
1382 switch(res.type) {
1383 case RTN_UNICAST:
1384 /*
1385 * Forwarding check here, we only check for forwarding
1386 * being turned off, if you want to only forward intra
1387 * area, its up to you to set the routing tables up
1388 * correctly.
1389 */
1390 if (dn_db->parms.forwarding == 0)
1391 goto e_inval;
1392
1393 if (res.fi->fib_nhs > 1 && fld.flowidn_oif == 0)
1394 dn_fib_select_multipath(&fld, &res);
1395
1396 /*
1397 * Check for out_dev == in_dev. We use the RTCF_DOREDIRECT
1398 * flag as a hint to set the intra-ethernet bit when
1399 * forwarding. If we've got NAT in operation, we don't do
1400 * this optimisation.
1401 */
1402 if (out_dev == in_dev && !(flags & RTCF_NAT))
1403 flags |= RTCF_DOREDIRECT;
1404
1405 local_src = DN_FIB_RES_PREFSRC(res);
1406
1407 case RTN_BLACKHOLE:
1408 case RTN_UNREACHABLE:
1409 break;
1410 case RTN_LOCAL:
1411 flags |= RTCF_LOCAL;
1412 fld.saddr = cb->dst;
1413 fld.daddr = cb->src;
1414
1415 /* Routing tables gave us a gateway */
1416 if (gateway)
1417 goto make_route;
1418
1419 /* Packet was intra-ethernet, so we know its on-link */
1420 if (cb->rt_flags & DN_RT_F_IE) {
1421 gateway = cb->src;
1422 goto make_route;
1423 }
1424
1425 /* Use the default router if there is one */
1426 neigh = neigh_clone(dn_db->router);
1427 if (neigh) {
1428 gateway = ((struct dn_neigh *)neigh)->addr;
1429 goto make_route;
1430 }
1431
1432 /* Close eyes and pray */
1433 gateway = cb->src;
1434 goto make_route;
1435 default:
1436 goto e_inval;
1437 }
1438
1439make_route:
1440 rt = dst_alloc(&dn_dst_ops, out_dev, 1, DST_OBSOLETE_NONE, DST_HOST);
1441 if (rt == NULL)
1442 goto e_nobufs;
1443
1444 rt->dn_next = NULL;
1445 memset(&rt->fld, 0, sizeof(rt->fld));
1446 rt->rt_saddr = fld.saddr;
1447 rt->rt_daddr = fld.daddr;
1448 rt->rt_gateway = fld.daddr;
1449 if (gateway)
1450 rt->rt_gateway = gateway;
1451 rt->rt_local_src = local_src ? local_src : rt->rt_saddr;
1452
1453 rt->rt_dst_map = fld.daddr;
1454 rt->rt_src_map = fld.saddr;
1455
1456 rt->fld.saddr = cb->src;
1457 rt->fld.daddr = cb->dst;
1458 rt->fld.flowidn_oif = 0;
1459 rt->fld.flowidn_iif = in_dev->ifindex;
1460 rt->fld.flowidn_mark = fld.flowidn_mark;
1461
1462 rt->n = neigh;
1463 rt->dst.lastuse = jiffies;
1464 rt->dst.output = dn_rt_bug_out;
1465 switch (res.type) {
1466 case RTN_UNICAST:
1467 rt->dst.input = dn_forward;
1468 break;
1469 case RTN_LOCAL:
1470 rt->dst.output = dn_output;
1471 rt->dst.input = dn_nsp_rx;
1472 rt->dst.dev = in_dev;
1473 flags |= RTCF_LOCAL;
1474 break;
1475 default:
1476 case RTN_UNREACHABLE:
1477 case RTN_BLACKHOLE:
1478 rt->dst.input = dst_discard;
1479 }
1480 rt->rt_flags = flags;
1481
1482 err = dn_rt_set_next_hop(rt, &res);
1483 if (err)
1484 goto e_neighbour;
1485
1486 hash = dn_hash(rt->fld.saddr, rt->fld.daddr);
1487 /* dn_insert_route() increments dst->__refcnt */
1488 dn_insert_route(rt, hash, &rt);
1489 skb_dst_set(skb, &rt->dst);
1490
1491done:
1492 if (neigh)
1493 neigh_release(neigh);
1494 if (free_res)
1495 dn_fib_res_put(&res);
1496 dev_put(in_dev);
1497 if (out_dev)
1498 dev_put(out_dev);
1499out:
1500 return err;
1501
1502e_inval:
1503 err = -EINVAL;
1504 goto done;
1505
1506e_nobufs:
1507 err = -ENOBUFS;
1508 goto done;
1509
1510e_neighbour:
1511 dst_release_immediate(&rt->dst);
1512 goto done;
1513}
1514
1515static int dn_route_input(struct sk_buff *skb)
1516{
1517 struct dn_route *rt;
1518 struct dn_skb_cb *cb = DN_SKB_CB(skb);
1519 unsigned int hash = dn_hash(cb->src, cb->dst);
1520
1521 if (skb_dst(skb))
1522 return 0;
1523
1524 rcu_read_lock();
1525 for(rt = rcu_dereference(dn_rt_hash_table[hash].chain); rt != NULL;
1526 rt = rcu_dereference(rt->dn_next)) {
1527 if ((rt->fld.saddr == cb->src) &&
1528 (rt->fld.daddr == cb->dst) &&
1529 (rt->fld.flowidn_oif == 0) &&
1530 (rt->fld.flowidn_mark == skb->mark) &&
1531 (rt->fld.flowidn_iif == cb->iif)) {
1532 dst_hold_and_use(&rt->dst, jiffies);
1533 rcu_read_unlock();
1534 skb_dst_set(skb, (struct dst_entry *)rt);
1535 return 0;
1536 }
1537 }
1538 rcu_read_unlock();
1539
1540 return dn_route_input_slow(skb);
1541}
1542
1543static int dn_rt_fill_info(struct sk_buff *skb, u32 portid, u32 seq,
1544 int event, int nowait, unsigned int flags)
1545{
1546 struct dn_route *rt = (struct dn_route *)skb_dst(skb);
1547 struct rtmsg *r;
1548 struct nlmsghdr *nlh;
1549 long expires;
1550
1551 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*r), flags);
1552 if (!nlh)
1553 return -EMSGSIZE;
1554
1555 r = nlmsg_data(nlh);
1556 r->rtm_family = AF_DECnet;
1557 r->rtm_dst_len = 16;
1558 r->rtm_src_len = 0;
1559 r->rtm_tos = 0;
1560 r->rtm_table = RT_TABLE_MAIN;
1561 r->rtm_type = rt->rt_type;
1562 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
1563 r->rtm_scope = RT_SCOPE_UNIVERSE;
1564 r->rtm_protocol = RTPROT_UNSPEC;
1565
1566 if (rt->rt_flags & RTCF_NOTIFY)
1567 r->rtm_flags |= RTM_F_NOTIFY;
1568
1569 if (nla_put_u32(skb, RTA_TABLE, RT_TABLE_MAIN) < 0 ||
1570 nla_put_le16(skb, RTA_DST, rt->rt_daddr) < 0)
1571 goto errout;
1572
1573 if (rt->fld.saddr) {
1574 r->rtm_src_len = 16;
1575 if (nla_put_le16(skb, RTA_SRC, rt->fld.saddr) < 0)
1576 goto errout;
1577 }
1578 if (rt->dst.dev &&
1579 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex) < 0)
1580 goto errout;
1581
1582 /*
1583 * Note to self - change this if input routes reverse direction when
1584 * they deal only with inputs and not with replies like they do
1585 * currently.
1586 */
1587 if (nla_put_le16(skb, RTA_PREFSRC, rt->rt_local_src) < 0)
1588 goto errout;
1589
1590 if (rt->rt_daddr != rt->rt_gateway &&
1591 nla_put_le16(skb, RTA_GATEWAY, rt->rt_gateway) < 0)
1592 goto errout;
1593
1594 if (rtnetlink_put_metrics(skb, dst_metrics_ptr(&rt->dst)) < 0)
1595 goto errout;
1596
1597 expires = rt->dst.expires ? rt->dst.expires - jiffies : 0;
1598 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires,
1599 rt->dst.error) < 0)
1600 goto errout;
1601
1602 if (dn_is_input_route(rt) &&
1603 nla_put_u32(skb, RTA_IIF, rt->fld.flowidn_iif) < 0)
1604 goto errout;
1605
1606 nlmsg_end(skb, nlh);
1607 return 0;
1608
1609errout:
1610 nlmsg_cancel(skb, nlh);
1611 return -EMSGSIZE;
1612}
1613
1614const struct nla_policy rtm_dn_policy[RTA_MAX + 1] = {
1615 [RTA_DST] = { .type = NLA_U16 },
1616 [RTA_SRC] = { .type = NLA_U16 },
1617 [RTA_IIF] = { .type = NLA_U32 },
1618 [RTA_OIF] = { .type = NLA_U32 },
1619 [RTA_GATEWAY] = { .type = NLA_U16 },
1620 [RTA_PRIORITY] = { .type = NLA_U32 },
1621 [RTA_PREFSRC] = { .type = NLA_U16 },
1622 [RTA_METRICS] = { .type = NLA_NESTED },
1623 [RTA_MULTIPATH] = { .type = NLA_NESTED },
1624 [RTA_TABLE] = { .type = NLA_U32 },
1625 [RTA_MARK] = { .type = NLA_U32 },
1626};
1627
1628/*
1629 * This is called by both endnodes and routers now.
1630 */
1631static int dn_cache_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
1632 struct netlink_ext_ack *extack)
1633{
1634 struct net *net = sock_net(in_skb->sk);
1635 struct rtmsg *rtm = nlmsg_data(nlh);
1636 struct dn_route *rt = NULL;
1637 struct dn_skb_cb *cb;
1638 int err;
1639 struct sk_buff *skb;
1640 struct flowidn fld;
1641 struct nlattr *tb[RTA_MAX+1];
1642
1643 if (!net_eq(net, &init_net))
1644 return -EINVAL;
1645
1646 err = nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
1647 rtm_dn_policy, extack);
1648 if (err < 0)
1649 return err;
1650
1651 memset(&fld, 0, sizeof(fld));
1652 fld.flowidn_proto = DNPROTO_NSP;
1653
1654 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1655 if (skb == NULL)
1656 return -ENOBUFS;
1657 skb_reset_mac_header(skb);
1658 cb = DN_SKB_CB(skb);
1659
1660 if (tb[RTA_SRC])
1661 fld.saddr = nla_get_le16(tb[RTA_SRC]);
1662
1663 if (tb[RTA_DST])
1664 fld.daddr = nla_get_le16(tb[RTA_DST]);
1665
1666 if (tb[RTA_IIF])
1667 fld.flowidn_iif = nla_get_u32(tb[RTA_IIF]);
1668
1669 if (fld.flowidn_iif) {
1670 struct net_device *dev;
1671 dev = __dev_get_by_index(&init_net, fld.flowidn_iif);
1672 if (!dev || !dev->dn_ptr) {
1673 kfree_skb(skb);
1674 return -ENODEV;
1675 }
1676 skb->protocol = htons(ETH_P_DNA_RT);
1677 skb->dev = dev;
1678 cb->src = fld.saddr;
1679 cb->dst = fld.daddr;
1680 local_bh_disable();
1681 err = dn_route_input(skb);
1682 local_bh_enable();
1683 memset(cb, 0, sizeof(struct dn_skb_cb));
1684 rt = (struct dn_route *)skb_dst(skb);
1685 if (!err && -rt->dst.error)
1686 err = rt->dst.error;
1687 } else {
1688 if (tb[RTA_OIF])
1689 fld.flowidn_oif = nla_get_u32(tb[RTA_OIF]);
1690
1691 err = dn_route_output_key((struct dst_entry **)&rt, &fld, 0);
1692 }
1693
1694 skb->dev = NULL;
1695 if (err)
1696 goto out_free;
1697 skb_dst_set(skb, &rt->dst);
1698 if (rtm->rtm_flags & RTM_F_NOTIFY)
1699 rt->rt_flags |= RTCF_NOTIFY;
1700
1701 err = dn_rt_fill_info(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq, RTM_NEWROUTE, 0, 0);
1702 if (err < 0) {
1703 err = -EMSGSIZE;
1704 goto out_free;
1705 }
1706
1707 return rtnl_unicast(skb, &init_net, NETLINK_CB(in_skb).portid);
1708
1709out_free:
1710 kfree_skb(skb);
1711 return err;
1712}
1713
1714/*
1715 * For routers, this is called from dn_fib_dump, but for endnodes its
1716 * called directly from the rtnetlink dispatch table.
1717 */
1718int dn_cache_dump(struct sk_buff *skb, struct netlink_callback *cb)
1719{
1720 struct net *net = sock_net(skb->sk);
1721 struct dn_route *rt;
1722 int h, s_h;
1723 int idx, s_idx;
1724 struct rtmsg *rtm;
1725
1726 if (!net_eq(net, &init_net))
1727 return 0;
1728
1729 if (nlmsg_len(cb->nlh) < sizeof(struct rtmsg))
1730 return -EINVAL;
1731
1732 rtm = nlmsg_data(cb->nlh);
1733 if (!(rtm->rtm_flags & RTM_F_CLONED))
1734 return 0;
1735
1736 s_h = cb->args[0];
1737 s_idx = idx = cb->args[1];
1738 for(h = 0; h <= dn_rt_hash_mask; h++) {
1739 if (h < s_h)
1740 continue;
1741 if (h > s_h)
1742 s_idx = 0;
1743 rcu_read_lock_bh();
1744 for(rt = rcu_dereference_bh(dn_rt_hash_table[h].chain), idx = 0;
1745 rt;
1746 rt = rcu_dereference_bh(rt->dn_next), idx++) {
1747 if (idx < s_idx)
1748 continue;
1749 skb_dst_set(skb, dst_clone(&rt->dst));
1750 if (dn_rt_fill_info(skb, NETLINK_CB(cb->skb).portid,
1751 cb->nlh->nlmsg_seq, RTM_NEWROUTE,
1752 1, NLM_F_MULTI) < 0) {
1753 skb_dst_drop(skb);
1754 rcu_read_unlock_bh();
1755 goto done;
1756 }
1757 skb_dst_drop(skb);
1758 }
1759 rcu_read_unlock_bh();
1760 }
1761
1762done:
1763 cb->args[0] = h;
1764 cb->args[1] = idx;
1765 return skb->len;
1766}
1767
1768#ifdef CONFIG_PROC_FS
1769struct dn_rt_cache_iter_state {
1770 int bucket;
1771};
1772
1773static struct dn_route *dn_rt_cache_get_first(struct seq_file *seq)
1774{
1775 struct dn_route *rt = NULL;
1776 struct dn_rt_cache_iter_state *s = seq->private;
1777
1778 for(s->bucket = dn_rt_hash_mask; s->bucket >= 0; --s->bucket) {
1779 rcu_read_lock_bh();
1780 rt = rcu_dereference_bh(dn_rt_hash_table[s->bucket].chain);
1781 if (rt)
1782 break;
1783 rcu_read_unlock_bh();
1784 }
1785 return rt;
1786}
1787
1788static struct dn_route *dn_rt_cache_get_next(struct seq_file *seq, struct dn_route *rt)
1789{
1790 struct dn_rt_cache_iter_state *s = seq->private;
1791
1792 rt = rcu_dereference_bh(rt->dn_next);
1793 while (!rt) {
1794 rcu_read_unlock_bh();
1795 if (--s->bucket < 0)
1796 break;
1797 rcu_read_lock_bh();
1798 rt = rcu_dereference_bh(dn_rt_hash_table[s->bucket].chain);
1799 }
1800 return rt;
1801}
1802
1803static void *dn_rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
1804{
1805 struct dn_route *rt = dn_rt_cache_get_first(seq);
1806
1807 if (rt) {
1808 while(*pos && (rt = dn_rt_cache_get_next(seq, rt)))
1809 --*pos;
1810 }
1811 return *pos ? NULL : rt;
1812}
1813
1814static void *dn_rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1815{
1816 struct dn_route *rt = dn_rt_cache_get_next(seq, v);
1817 ++*pos;
1818 return rt;
1819}
1820
1821static void dn_rt_cache_seq_stop(struct seq_file *seq, void *v)
1822{
1823 if (v)
1824 rcu_read_unlock_bh();
1825}
1826
1827static int dn_rt_cache_seq_show(struct seq_file *seq, void *v)
1828{
1829 struct dn_route *rt = v;
1830 char buf1[DN_ASCBUF_LEN], buf2[DN_ASCBUF_LEN];
1831
1832 seq_printf(seq, "%-8s %-7s %-7s %04d %04d %04d\n",
1833 rt->dst.dev ? rt->dst.dev->name : "*",
1834 dn_addr2asc(le16_to_cpu(rt->rt_daddr), buf1),
1835 dn_addr2asc(le16_to_cpu(rt->rt_saddr), buf2),
1836 atomic_read(&rt->dst.__refcnt),
1837 rt->dst.__use, 0);
1838 return 0;
1839}
1840
1841static const struct seq_operations dn_rt_cache_seq_ops = {
1842 .start = dn_rt_cache_seq_start,
1843 .next = dn_rt_cache_seq_next,
1844 .stop = dn_rt_cache_seq_stop,
1845 .show = dn_rt_cache_seq_show,
1846};
1847#endif /* CONFIG_PROC_FS */
1848
1849void __init dn_route_init(void)
1850{
1851 int i, goal, order;
1852
1853 dn_dst_ops.kmem_cachep =
1854 kmem_cache_create("dn_dst_cache", sizeof(struct dn_route), 0,
1855 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1856 dst_entries_init(&dn_dst_ops);
1857 timer_setup(&dn_route_timer, dn_dst_check_expire, 0);
1858 dn_route_timer.expires = jiffies + decnet_dst_gc_interval * HZ;
1859 add_timer(&dn_route_timer);
1860
1861 goal = totalram_pages() >> (26 - PAGE_SHIFT);
1862
1863 for(order = 0; (1UL << order) < goal; order++)
1864 /* NOTHING */;
1865
1866 /*
1867 * Only want 1024 entries max, since the table is very, very unlikely
1868 * to be larger than that.
1869 */
1870 while(order && ((((1UL << order) * PAGE_SIZE) /
1871 sizeof(struct dn_rt_hash_bucket)) >= 2048))
1872 order--;
1873
1874 do {
1875 dn_rt_hash_mask = (1UL << order) * PAGE_SIZE /
1876 sizeof(struct dn_rt_hash_bucket);
1877 while(dn_rt_hash_mask & (dn_rt_hash_mask - 1))
1878 dn_rt_hash_mask--;
1879 dn_rt_hash_table = (struct dn_rt_hash_bucket *)
1880 __get_free_pages(GFP_ATOMIC, order);
1881 } while (dn_rt_hash_table == NULL && --order > 0);
1882
1883 if (!dn_rt_hash_table)
1884 panic("Failed to allocate DECnet route cache hash table\n");
1885
1886 printk(KERN_INFO
1887 "DECnet: Routing cache hash table of %u buckets, %ldKbytes\n",
1888 dn_rt_hash_mask,
1889 (long)(dn_rt_hash_mask*sizeof(struct dn_rt_hash_bucket))/1024);
1890
1891 dn_rt_hash_mask--;
1892 for(i = 0; i <= dn_rt_hash_mask; i++) {
1893 spin_lock_init(&dn_rt_hash_table[i].lock);
1894 dn_rt_hash_table[i].chain = NULL;
1895 }
1896
1897 dn_dst_ops.gc_thresh = (dn_rt_hash_mask + 1);
1898
1899 proc_create_seq_private("decnet_cache", 0444, init_net.proc_net,
1900 &dn_rt_cache_seq_ops,
1901 sizeof(struct dn_rt_cache_iter_state), NULL);
1902
1903#ifdef CONFIG_DECNET_ROUTER
1904 rtnl_register_module(THIS_MODULE, PF_DECnet, RTM_GETROUTE,
1905 dn_cache_getroute, dn_fib_dump, 0);
1906#else
1907 rtnl_register_module(THIS_MODULE, PF_DECnet, RTM_GETROUTE,
1908 dn_cache_getroute, dn_cache_dump, 0);
1909#endif
1910}
1911
1912void __exit dn_route_cleanup(void)
1913{
1914 del_timer(&dn_route_timer);
1915 dn_run_flush(NULL);
1916
1917 remove_proc_entry("decnet_cache", init_net.proc_net);
1918 dst_entries_destroy(&dn_dst_ops);
1919}