]> bbs.cooldavid.org Git - net-next-2.6.git/blame - net/decnet/dn_route.c
git://bbs.cooldavid.org / net-next-2.6.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
[NET]: Turn nfmark into generic mark
[net-next-2.6.git] / net / decnet / dn_route.c
CommitLineData
1da177e4
LT		 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
1da177e4
LT		 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 <net/sock.h>
70#include <linux/mm.h>
71#include <linux/proc_fs.h>
72#include <linux/seq_file.h>
73#include <linux/init.h>
74#include <linux/rtnetlink.h>
75#include <linux/string.h>
76#include <linux/netfilter_decnet.h>
77#include <linux/rcupdate.h>
78#include <linux/times.h>
79#include <asm/errno.h>
80#include <net/neighbour.h>
81#include <net/dst.h>
82#include <net/flow.h>
a8731cbf 83#include <net/fib_rules.h>
1da177e4
LT		 84#include <net/dn.h>
85#include <net/dn_dev.h>
86#include <net/dn_nsp.h>
87#include <net/dn_route.h>
88#include <net/dn_neigh.h>
89#include <net/dn_fib.h>
90
91struct dn_rt_hash_bucket
92{
93 struct dn_route *chain;
94 spinlock_t lock;
95} __attribute__((__aligned__(8)));
96
97extern struct neigh_table dn_neigh_table;
98
99
100static unsigned char dn_hiord_addr[6] = {0xAA,0x00,0x04,0x00,0x00,0x00};
101
102static const int dn_rt_min_delay = 2 * HZ;
103static const int dn_rt_max_delay = 10 * HZ;
104static const int dn_rt_mtu_expires = 10 * 60 * HZ;
105
106static unsigned long dn_rt_deadline;
107
108static int dn_dst_gc(void);
109static struct dst_entry *dn_dst_check(struct dst_entry *, __u32);
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, u32 mtu);
113static int dn_route_input(struct sk_buff *);
114static void dn_run_flush(unsigned long dummy);
115
116static struct dn_rt_hash_bucket *dn_rt_hash_table;
117static unsigned dn_rt_hash_mask;
118
119static struct timer_list dn_route_timer;
8d06afab 120static DEFINE_TIMER(dn_rt_flush_timer, dn_run_flush, 0, 0);
1da177e4
LT		 121int decnet_dst_gc_interval = 2;
122
123static struct dst_ops dn_dst_ops = {
124 .family = PF_DECnet,
125 .protocol = __constant_htons(ETH_P_DNA_RT),
126 .gc_thresh = 128,
127 .gc = dn_dst_gc,
128 .check = dn_dst_check,
129 .negative_advice = dn_dst_negative_advice,
130 .link_failure = dn_dst_link_failure,
131 .update_pmtu = dn_dst_update_pmtu,
132 .entry_size = sizeof(struct dn_route),
133 .entries = ATOMIC_INIT(0),
134};
135
c4ea94ab 136static __inline__ unsigned dn_hash(__le16 src, __le16 dst)
1da177e4 137{
c4ea94ab 138 __u16 tmp = (__u16 __force)(src ^ dst);
1da177e4
LT		 139 tmp ^= (tmp >> 3);
140 tmp ^= (tmp >> 5);
141 tmp ^= (tmp >> 10);
142 return dn_rt_hash_mask & (unsigned)tmp;
143}
144
145static inline void dnrt_free(struct dn_route *rt)
146{
147 call_rcu_bh(&rt->u.dst.rcu_head, dst_rcu_free);
148}
149
150static inline void dnrt_drop(struct dn_route *rt)
151{
34001127 152 dst_release(&rt->u.dst);
1da177e4
LT		 153 call_rcu_bh(&rt->u.dst.rcu_head, dst_rcu_free);
154}
155
156static void dn_dst_check_expire(unsigned long dummy)
157{
158 int i;
159 struct dn_route *rt, **rtp;
160 unsigned long now = jiffies;
161 unsigned long expire = 120 * HZ;
162
163 for(i = 0; i <= dn_rt_hash_mask; i++) {
164 rtp = &dn_rt_hash_table[i].chain;
165
166 spin_lock(&dn_rt_hash_table[i].lock);
167 while((rt=*rtp) != NULL) {
168 if (atomic_read(&rt->u.dst.__refcnt) ||
169 (now - rt->u.dst.lastuse) < expire) {
170 rtp = &rt->u.rt_next;
171 continue;
172 }
173 *rtp = rt->u.rt_next;
174 rt->u.rt_next = NULL;
175 dnrt_free(rt);
176 }
177 spin_unlock(&dn_rt_hash_table[i].lock);
178
179 if ((jiffies - now) > 0)
180 break;
181 }
182
183 mod_timer(&dn_route_timer, now + decnet_dst_gc_interval * HZ);
184}
185
186static int dn_dst_gc(void)
187{
188 struct dn_route *rt, **rtp;
189 int i;
190 unsigned long now = jiffies;
191 unsigned long expire = 10 * HZ;
192
193 for(i = 0; i <= dn_rt_hash_mask; i++) {
194
195 spin_lock_bh(&dn_rt_hash_table[i].lock);
196 rtp = &dn_rt_hash_table[i].chain;
197
198 while((rt=*rtp) != NULL) {
199 if (atomic_read(&rt->u.dst.__refcnt) ||
200 (now - rt->u.dst.lastuse) < expire) {
201 rtp = &rt->u.rt_next;
202 continue;
203 }
204 *rtp = rt->u.rt_next;
205 rt->u.rt_next = NULL;
206 dnrt_drop(rt);
207 break;
208 }
209 spin_unlock_bh(&dn_rt_hash_table[i].lock);
210 }
211
212 return 0;
213}
214
215/*
216 * The decnet standards don't impose a particular minimum mtu, what they
217 * do insist on is that the routing layer accepts a datagram of at least
218 * 230 bytes long. Here we have to subtract the routing header length from
219 * 230 to get the minimum acceptable mtu. If there is no neighbour, then we
220 * assume the worst and use a long header size.
221 *
222 * We update both the mtu and the advertised mss (i.e. the segment size we
223 * advertise to the other end).
224 */
225static void dn_dst_update_pmtu(struct dst_entry *dst, u32 mtu)
226{
227 u32 min_mtu = 230;
228 struct dn_dev *dn = dst->neighbour ?
229 (struct dn_dev *)dst->neighbour->dev->dn_ptr : NULL;
230
231 if (dn && dn->use_long == 0)
232 min_mtu -= 6;
233 else
234 min_mtu -= 21;
235
236 if (dst->metrics[RTAX_MTU-1] > mtu && mtu >= min_mtu) {
237 if (!(dst_metric_locked(dst, RTAX_MTU))) {
238 dst->metrics[RTAX_MTU-1] = mtu;
239 dst_set_expires(dst, dn_rt_mtu_expires);
240 }
241 if (!(dst_metric_locked(dst, RTAX_ADVMSS))) {
242 u32 mss = mtu - DN_MAX_NSP_DATA_HEADER;
243 if (dst->metrics[RTAX_ADVMSS-1] > mss)
244 dst->metrics[RTAX_ADVMSS-1] = mss;
245 }
246 }
247}
248
249/*
250 * When a route has been marked obsolete. (e.g. routing cache flush)
251 */
252static struct dst_entry *dn_dst_check(struct dst_entry *dst, __u32 cookie)
253{
254 return NULL;
255}
256
257static struct dst_entry *dn_dst_negative_advice(struct dst_entry *dst)
258{
259 dst_release(dst);
260 return NULL;
261}
262
263static void dn_dst_link_failure(struct sk_buff *skb)
264{
265 return;
266}
267
268static inline int compare_keys(struct flowi *fl1, struct flowi *fl2)
269{
8238b218
DM		 270 return ((fl1->nl_u.dn_u.daddr ^ fl2->nl_u.dn_u.daddr) |
271 (fl1->nl_u.dn_u.saddr ^ fl2->nl_u.dn_u.saddr) |
be60358e 272#ifdef CONFIG_DECNET_ROUTE_FWMARK
8238b218
DM		 273 (fl1->nl_u.dn_u.fwmark ^ fl2->nl_u.dn_u.fwmark) |
274#endif
275 (fl1->nl_u.dn_u.scope ^ fl2->nl_u.dn_u.scope) |
276 (fl1->oif ^ fl2->oif) |
277 (fl1->iif ^ fl2->iif)) == 0;
1da177e4
LT		 278}
279
280static int dn_insert_route(struct dn_route *rt, unsigned hash, struct dn_route **rp)
281{
282 struct dn_route *rth, **rthp;
283 unsigned long now = jiffies;
284
285 rthp = &dn_rt_hash_table[hash].chain;
286
287 spin_lock_bh(&dn_rt_hash_table[hash].lock);
288 while((rth = *rthp) != NULL) {
289 if (compare_keys(&rth->fl, &rt->fl)) {
290 /* Put it first */
291 *rthp = rth->u.rt_next;
292 rcu_assign_pointer(rth->u.rt_next,
293 dn_rt_hash_table[hash].chain);
294 rcu_assign_pointer(dn_rt_hash_table[hash].chain, rth);
295
296 rth->u.dst.__use++;
297 dst_hold(&rth->u.dst);
298 rth->u.dst.lastuse = now;
299 spin_unlock_bh(&dn_rt_hash_table[hash].lock);
300
301 dnrt_drop(rt);
302 *rp = rth;
303 return 0;
304 }
305 rthp = &rth->u.rt_next;
306 }
307
308 rcu_assign_pointer(rt->u.rt_next, dn_rt_hash_table[hash].chain);
309 rcu_assign_pointer(dn_rt_hash_table[hash].chain, rt);
310
311 dst_hold(&rt->u.dst);
312 rt->u.dst.__use++;
313 rt->u.dst.lastuse = now;
314 spin_unlock_bh(&dn_rt_hash_table[hash].lock);
315 *rp = rt;
316 return 0;
317}
318
319void dn_run_flush(unsigned long dummy)
320{
321 int i;
322 struct dn_route *rt, *next;
323
324 for(i = 0; i < dn_rt_hash_mask; i++) {
325 spin_lock_bh(&dn_rt_hash_table[i].lock);
326
327 if ((rt = xchg(&dn_rt_hash_table[i].chain, NULL)) == NULL)
328 goto nothing_to_declare;
329
330 for(; rt; rt=next) {
331 next = rt->u.rt_next;
332 rt->u.rt_next = NULL;
333 dst_free((struct dst_entry *)rt);
334 }
335
336nothing_to_declare:
337 spin_unlock_bh(&dn_rt_hash_table[i].lock);
338 }
339}
340
341static DEFINE_SPINLOCK(dn_rt_flush_lock);
342
343void dn_rt_cache_flush(int delay)
344{
345 unsigned long now = jiffies;
346 int user_mode = !in_interrupt();
347
348 if (delay < 0)
349 delay = dn_rt_min_delay;
350
351 spin_lock_bh(&dn_rt_flush_lock);
352
353 if (del_timer(&dn_rt_flush_timer) && delay > 0 && dn_rt_deadline) {
354 long tmo = (long)(dn_rt_deadline - now);
355
356 if (user_mode && tmo < dn_rt_max_delay - dn_rt_min_delay)
357 tmo = 0;
358
359 if (delay > tmo)
360 delay = tmo;
361 }
362
363 if (delay <= 0) {
364 spin_unlock_bh(&dn_rt_flush_lock);
365 dn_run_flush(0);
366 return;
367 }
368
369 if (dn_rt_deadline == 0)
370 dn_rt_deadline = now + dn_rt_max_delay;
371
372 dn_rt_flush_timer.expires = now + delay;
373 add_timer(&dn_rt_flush_timer);
374 spin_unlock_bh(&dn_rt_flush_lock);
375}
376
377/**
378 * dn_return_short - Return a short packet to its sender
379 * @skb: The packet to return
380 *
381 */
382static int dn_return_short(struct sk_buff *skb)
383{
384 struct dn_skb_cb *cb;
385 unsigned char *ptr;
c4ea94ab
SW		 386 __le16 *src;
387 __le16 *dst;
388 __le16 tmp;
1da177e4
LT		 389
390 /* Add back headers */
391 skb_push(skb, skb->data - skb->nh.raw);
392
393 if ((skb = skb_unshare(skb, GFP_ATOMIC)) == NULL)
394 return NET_RX_DROP;
395
396 cb = DN_SKB_CB(skb);
397 /* Skip packet length and point to flags */
398 ptr = skb->data + 2;
399 *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
400
c4ea94ab 401 dst = (__le16 *)ptr;
1da177e4 402 ptr += 2;
c4ea94ab 403 src = (__le16 *)ptr;
1da177e4
LT		 404 ptr += 2;
405 *ptr = 0; /* Zero hop count */
406
407 /* Swap source and destination */
408 tmp = *src;
409 *src = *dst;
410 *dst = tmp;
411
412 skb->pkt_type = PACKET_OUTGOING;
413 dn_rt_finish_output(skb, NULL, NULL);
414 return NET_RX_SUCCESS;
415}
416
417/**
418 * dn_return_long - Return a long packet to its sender
419 * @skb: The long format packet to return
420 *
421 */
422static int dn_return_long(struct sk_buff *skb)
423{
424 struct dn_skb_cb *cb;
425 unsigned char *ptr;
426 unsigned char *src_addr, *dst_addr;
427 unsigned char tmp[ETH_ALEN];
428
429 /* Add back all headers */
430 skb_push(skb, skb->data - skb->nh.raw);
431
432 if ((skb = skb_unshare(skb, GFP_ATOMIC)) == NULL)
433 return NET_RX_DROP;
434
435 cb = DN_SKB_CB(skb);
436 /* Ignore packet length and point to flags */
437 ptr = skb->data + 2;
438
439 /* Skip padding */
440 if (*ptr & DN_RT_F_PF) {
441 char padlen = (*ptr & ~DN_RT_F_PF);
442 ptr += padlen;
443 }
444
445 *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
446 ptr += 2;
447 dst_addr = ptr;
448 ptr += 8;
449 src_addr = ptr;
450 ptr += 6;
451 *ptr = 0; /* Zero hop count */
452
453 /* Swap source and destination */
454 memcpy(tmp, src_addr, ETH_ALEN);
455 memcpy(src_addr, dst_addr, ETH_ALEN);
456 memcpy(dst_addr, tmp, ETH_ALEN);
457
458 skb->pkt_type = PACKET_OUTGOING;
459 dn_rt_finish_output(skb, dst_addr, src_addr);
460 return NET_RX_SUCCESS;
461}
462
463/**
464 * dn_route_rx_packet - Try and find a route for an incoming packet
465 * @skb: The packet to find a route for
466 *
467 * Returns: result of input function if route is found, error code otherwise
468 */
469static int dn_route_rx_packet(struct sk_buff *skb)
470{
471 struct dn_skb_cb *cb = DN_SKB_CB(skb);
472 int err;
473
474 if ((err = dn_route_input(skb)) == 0)
475 return dst_input(skb);
476
477 if (decnet_debug_level & 4) {
478 char *devname = skb->dev ? skb->dev->name : "???";
479 struct dn_skb_cb *cb = DN_SKB_CB(skb);
480 printk(KERN_DEBUG
481 "DECnet: dn_route_rx_packet: rt_flags=0x%02x dev=%s len=%d src=0x%04hx dst=0x%04hx err=%d type=%d\n",
c4ea94ab
SW		 482 (int)cb->rt_flags, devname, skb->len,
483 dn_ntohs(cb->src), dn_ntohs(cb->dst),
1da177e4
LT		 484 err, skb->pkt_type);
485 }
486
487 if ((skb->pkt_type == PACKET_HOST) && (cb->rt_flags & DN_RT_F_RQR)) {
488 switch(cb->rt_flags & DN_RT_PKT_MSK) {
489 case DN_RT_PKT_SHORT:
490 return dn_return_short(skb);
491 case DN_RT_PKT_LONG:
492 return dn_return_long(skb);
493 }
494 }
495
496 kfree_skb(skb);
497 return NET_RX_DROP;
498}
499
500static int dn_route_rx_long(struct sk_buff *skb)
501{
502 struct dn_skb_cb *cb = DN_SKB_CB(skb);
503 unsigned char *ptr = skb->data;
504
505 if (!pskb_may_pull(skb, 21)) /* 20 for long header, 1 for shortest nsp */
506 goto drop_it;
507
508 skb_pull(skb, 20);
509 skb->h.raw = skb->data;
510
511 /* Destination info */
512 ptr += 2;
c4ea94ab 513 cb->dst = dn_eth2dn(ptr);
1da177e4
LT		 514 if (memcmp(ptr, dn_hiord_addr, 4) != 0)
515 goto drop_it;
516 ptr += 6;
517
518
519 /* Source info */
520 ptr += 2;
c4ea94ab 521 cb->src = dn_eth2dn(ptr);
1da177e4
LT		 522 if (memcmp(ptr, dn_hiord_addr, 4) != 0)
523 goto drop_it;
524 ptr += 6;
525 /* Other junk */
526 ptr++;
527 cb->hops = *ptr++; /* Visit Count */
528
529 return NF_HOOK(PF_DECnet, NF_DN_PRE_ROUTING, skb, skb->dev, NULL, dn_route_rx_packet);
530
531drop_it:
532 kfree_skb(skb);
533 return NET_RX_DROP;
534}
535
536
537
538static int dn_route_rx_short(struct sk_buff *skb)
539{
540 struct dn_skb_cb *cb = DN_SKB_CB(skb);
541 unsigned char *ptr = skb->data;
542
543 if (!pskb_may_pull(skb, 6)) /* 5 for short header + 1 for shortest nsp */
544 goto drop_it;
545
546 skb_pull(skb, 5);
547 skb->h.raw = skb->data;
548
c4ea94ab 549 cb->dst = *(__le16 *)ptr;
1da177e4 550 ptr += 2;
c4ea94ab 551 cb->src = *(__le16 *)ptr;
1da177e4
LT		 552 ptr += 2;
553 cb->hops = *ptr & 0x3f;
554
555 return NF_HOOK(PF_DECnet, NF_DN_PRE_ROUTING, skb, skb->dev, NULL, dn_route_rx_packet);
556
557drop_it:
558 kfree_skb(skb);
559 return NET_RX_DROP;
560}
561
562static int dn_route_discard(struct sk_buff *skb)
563{
564 /*
565 * I know we drop the packet here, but thats considered success in
566 * this case
567 */
568 kfree_skb(skb);
569 return NET_RX_SUCCESS;
570}
571
572static int dn_route_ptp_hello(struct sk_buff *skb)
573{
574 dn_dev_hello(skb);
575 dn_neigh_pointopoint_hello(skb);
576 return NET_RX_SUCCESS;
577}
578
f2ccd8fa 579int dn_route_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
1da177e4
LT		 580{
581 struct dn_skb_cb *cb;
582 unsigned char flags = 0;
c4ea94ab 583 __u16 len = dn_ntohs(*(__le16 *)skb->data);
1da177e4
LT		 584 struct dn_dev *dn = (struct dn_dev *)dev->dn_ptr;
585 unsigned char padlen = 0;
586
587 if (dn == NULL)
588 goto dump_it;
589
590 if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
591 goto out;
592
593 if (!pskb_may_pull(skb, 3))
594 goto dump_it;
595
596 skb_pull(skb, 2);
597
598 if (len > skb->len)
599 goto dump_it;
600
601 skb_trim(skb, len);
602
603 flags = *skb->data;
604
605 cb = DN_SKB_CB(skb);
606 cb->stamp = jiffies;
607 cb->iif = dev->ifindex;
608
609 /*
610 * If we have padding, remove it.
611 */
612 if (flags & DN_RT_F_PF) {
613 padlen = flags & ~DN_RT_F_PF;
614 if (!pskb_may_pull(skb, padlen + 1))
615 goto dump_it;
616 skb_pull(skb, padlen);
617 flags = *skb->data;
618 }
619
620 skb->nh.raw = skb->data;
621
622 /*
623 * Weed out future version DECnet
624 */
625 if (flags & DN_RT_F_VER)
626 goto dump_it;
627
628 cb->rt_flags = flags;
629
630 if (decnet_debug_level & 1)
631 printk(KERN_DEBUG
632 "dn_route_rcv: got 0x%02x from %s [%d %d %d]\n",
633 (int)flags, (dev) ? dev->name : "???", len, skb->len,
634 padlen);
635
636 if (flags & DN_RT_PKT_CNTL) {
364c6bad 637 if (unlikely(skb_linearize(skb)))
1da177e4
LT		 638 goto dump_it;
639
640 switch(flags & DN_RT_CNTL_MSK) {
641 case DN_RT_PKT_INIT:
642 dn_dev_init_pkt(skb);
643 break;
644 case DN_RT_PKT_VERI:
645 dn_dev_veri_pkt(skb);
646 break;
647 }
648
649 if (dn->parms.state != DN_DEV_S_RU)
650 goto dump_it;
651
652 switch(flags & DN_RT_CNTL_MSK) {
653 case DN_RT_PKT_HELO:
654 return NF_HOOK(PF_DECnet, NF_DN_HELLO, skb, skb->dev, NULL, dn_route_ptp_hello);
655
656 case DN_RT_PKT_L1RT:
657 case DN_RT_PKT_L2RT:
658 return NF_HOOK(PF_DECnet, NF_DN_ROUTE, skb, skb->dev, NULL, dn_route_discard);
659 case DN_RT_PKT_ERTH:
660 return NF_HOOK(PF_DECnet, NF_DN_HELLO, skb, skb->dev, NULL, dn_neigh_router_hello);
661
662 case DN_RT_PKT_EEDH:
663 return NF_HOOK(PF_DECnet, NF_DN_HELLO, skb, skb->dev, NULL, dn_neigh_endnode_hello);
664 }
665 } else {
666 if (dn->parms.state != DN_DEV_S_RU)
667 goto dump_it;
668
669 skb_pull(skb, 1); /* Pull flags */
670
671 switch(flags & DN_RT_PKT_MSK) {
672 case DN_RT_PKT_LONG:
673 return dn_route_rx_long(skb);
674 case DN_RT_PKT_SHORT:
675 return dn_route_rx_short(skb);
676 }
677 }
678
679dump_it:
680 kfree_skb(skb);
681out:
682 return NET_RX_DROP;
683}
684
685static int dn_output(struct sk_buff *skb)
686{
687 struct dst_entry *dst = skb->dst;
688 struct dn_route *rt = (struct dn_route *)dst;
689 struct net_device *dev = dst->dev;
690 struct dn_skb_cb *cb = DN_SKB_CB(skb);
691 struct neighbour *neigh;
692
693 int err = -EINVAL;
694
695 if ((neigh = dst->neighbour) == NULL)
696 goto error;
697
698 skb->dev = dev;
699
700 cb->src = rt->rt_saddr;
701 cb->dst = rt->rt_daddr;
702
703 /*
704 * Always set the Intra-Ethernet bit on all outgoing packets
705 * originated on this node. Only valid flag from upper layers
706 * is return-to-sender-requested. Set hop count to 0 too.
707 */
708 cb->rt_flags &= ~DN_RT_F_RQR;
709 cb->rt_flags |= DN_RT_F_IE;
710 cb->hops = 0;
711
712 return NF_HOOK(PF_DECnet, NF_DN_LOCAL_OUT, skb, NULL, dev, neigh->output);
713
714error:
715 if (net_ratelimit())
716 printk(KERN_DEBUG "dn_output: This should not happen\n");
717
718 kfree_skb(skb);
719
720 return err;
721}
722
723static int dn_forward(struct sk_buff *skb)
724{
725 struct dn_skb_cb *cb = DN_SKB_CB(skb);
726 struct dst_entry *dst = skb->dst;
727 struct dn_dev *dn_db = dst->dev->dn_ptr;
728 struct dn_route *rt;
729 struct neighbour *neigh = dst->neighbour;
730 int header_len;
731#ifdef CONFIG_NETFILTER
732 struct net_device *dev = skb->dev;
733#endif
734
735 if (skb->pkt_type != PACKET_HOST)
736 goto drop;
737
738 /* Ensure that we have enough space for headers */
739 rt = (struct dn_route *)skb->dst;
740 header_len = dn_db->use_long ? 21 : 6;
741 if (skb_cow(skb, LL_RESERVED_SPACE(rt->u.dst.dev)+header_len))
742 goto drop;
743
744 /*
745 * Hop count exceeded.
746 */
747 if (++cb->hops > 30)
748 goto drop;
749
750 skb->dev = rt->u.dst.dev;
751
752 /*
753 * If packet goes out same interface it came in on, then set
754 * the Intra-Ethernet bit. This has no effect for short
755 * packets, so we don't need to test for them here.
756 */
757 cb->rt_flags &= ~DN_RT_F_IE;
758 if (rt->rt_flags & RTCF_DOREDIRECT)
759 cb->rt_flags |= DN_RT_F_IE;
760
761 return NF_HOOK(PF_DECnet, NF_DN_FORWARD, skb, dev, skb->dev, neigh->output);
762
763drop:
764 kfree_skb(skb);
765 return NET_RX_DROP;
766}
767
768/*
769 * Drop packet. This is used for endnodes and for
770 * when we should not be forwarding packets from
771 * this dest.
772 */
773static int dn_blackhole(struct sk_buff *skb)
774{
775 kfree_skb(skb);
776 return NET_RX_DROP;
777}
778
779/*
780 * Used to catch bugs. This should never normally get
781 * called.
782 */
783static int dn_rt_bug(struct sk_buff *skb)
784{
785 if (net_ratelimit()) {
786 struct dn_skb_cb *cb = DN_SKB_CB(skb);
787
788 printk(KERN_DEBUG "dn_rt_bug: skb from:%04x to:%04x\n",
c4ea94ab 789 dn_ntohs(cb->src), dn_ntohs(cb->dst));
1da177e4
LT		 790 }
791
792 kfree_skb(skb);
793
794 return NET_RX_BAD;
795}
796
797static int dn_rt_set_next_hop(struct dn_route *rt, struct dn_fib_res *res)
798{
799 struct dn_fib_info *fi = res->fi;
800 struct net_device *dev = rt->u.dst.dev;
801 struct neighbour *n;
802 unsigned mss;
803
804 if (fi) {
805 if (DN_FIB_RES_GW(*res) &&
806 DN_FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
807 rt->rt_gateway = DN_FIB_RES_GW(*res);
808 memcpy(rt->u.dst.metrics, fi->fib_metrics,
809 sizeof(rt->u.dst.metrics));
810 }
811 rt->rt_type = res->type;
812
813 if (dev != NULL && rt->u.dst.neighbour == NULL) {
814 n = __neigh_lookup_errno(&dn_neigh_table, &rt->rt_gateway, dev);
815 if (IS_ERR(n))
816 return PTR_ERR(n);
817 rt->u.dst.neighbour = n;
818 }
819
820 if (rt->u.dst.metrics[RTAX_MTU-1] == 0 ||
821 rt->u.dst.metrics[RTAX_MTU-1] > rt->u.dst.dev->mtu)
822 rt->u.dst.metrics[RTAX_MTU-1] = rt->u.dst.dev->mtu;
823 mss = dn_mss_from_pmtu(dev, dst_mtu(&rt->u.dst));
824 if (rt->u.dst.metrics[RTAX_ADVMSS-1] == 0 ||
825 rt->u.dst.metrics[RTAX_ADVMSS-1] > mss)
826 rt->u.dst.metrics[RTAX_ADVMSS-1] = mss;
827 return 0;
828}
829
c4ea94ab 830static inline int dn_match_addr(__le16 addr1, __le16 addr2)
1da177e4
LT		 831{
832 __u16 tmp = dn_ntohs(addr1) ^ dn_ntohs(addr2);
833 int match = 16;
834 while(tmp) {
835 tmp >>= 1;
836 match--;
837 }
838 return match;
839}
840
c4ea94ab 841static __le16 dnet_select_source(const struct net_device *dev, __le16 daddr, int scope)
1da177e4 842{
c4ea94ab 843 __le16 saddr = 0;
1da177e4
LT		 844 struct dn_dev *dn_db = dev->dn_ptr;
845 struct dn_ifaddr *ifa;
846 int best_match = 0;
847 int ret;
848
849 read_lock(&dev_base_lock);
850 for(ifa = dn_db->ifa_list; ifa; ifa = ifa->ifa_next) {
851 if (ifa->ifa_scope > scope)
852 continue;
853 if (!daddr) {
854 saddr = ifa->ifa_local;
855 break;
856 }
857 ret = dn_match_addr(daddr, ifa->ifa_local);
858 if (ret > best_match)
859 saddr = ifa->ifa_local;
860 if (best_match == 0)
861 saddr = ifa->ifa_local;
862 }
863 read_unlock(&dev_base_lock);
864
865 return saddr;
866}
867
c4ea94ab 868static inline __le16 __dn_fib_res_prefsrc(struct dn_fib_res *res)
1da177e4
LT		 869{
870 return dnet_select_source(DN_FIB_RES_DEV(*res), DN_FIB_RES_GW(*res), res->scope);
871}
872
c4ea94ab 873static inline __le16 dn_fib_rules_map_destination(__le16 daddr, struct dn_fib_res *res)
1da177e4 874{
c4ea94ab 875 __le16 mask = dnet_make_mask(res->prefixlen);
1da177e4
LT		 876 return (daddr&~mask)|res->fi->fib_nh->nh_gw;
877}
878
879static int dn_route_output_slow(struct dst_entry **pprt, const struct flowi *oldflp, int try_hard)
880{
881 struct flowi fl = { .nl_u = { .dn_u =
882 { .daddr = oldflp->fld_dst,
883 .saddr = oldflp->fld_src,
884 .scope = RT_SCOPE_UNIVERSE,
885#ifdef CONFIG_DECNET_ROUTE_FWMARK
886 .fwmark = oldflp->fld_fwmark
887#endif
888 } },
889 .iif = loopback_dev.ifindex,
890 .oif = oldflp->oif };
891 struct dn_route *rt = NULL;
892 struct net_device *dev_out = NULL;
893 struct neighbour *neigh = NULL;
894 unsigned hash;
895 unsigned flags = 0;
896 struct dn_fib_res res = { .fi = NULL, .type = RTN_UNICAST };
897 int err;
898 int free_res = 0;
c4ea94ab 899 __le16 gateway = 0;
1da177e4
LT		 900
901 if (decnet_debug_level & 16)
902 printk(KERN_DEBUG
903 "dn_route_output_slow: dst=%04x src=%04x mark=%d"
c4ea94ab
SW		 904 " iif=%d oif=%d\n", dn_ntohs(oldflp->fld_dst),
905 dn_ntohs(oldflp->fld_src),
1da177e4
LT		 906 oldflp->fld_fwmark, loopback_dev.ifindex, oldflp->oif);
907
908 /* If we have an output interface, verify its a DECnet device */
909 if (oldflp->oif) {
910 dev_out = dev_get_by_index(oldflp->oif);
911 err = -ENODEV;
912 if (dev_out && dev_out->dn_ptr == NULL) {
913 dev_put(dev_out);
914 dev_out = NULL;
915 }
916 if (dev_out == NULL)
917 goto out;
918 }
919
920 /* If we have a source address, verify that its a local address */
921 if (oldflp->fld_src) {
922 err = -EADDRNOTAVAIL;
923
924 if (dev_out) {
925 if (dn_dev_islocal(dev_out, oldflp->fld_src))
926 goto source_ok;
927 dev_put(dev_out);
928 goto out;
929 }
930 read_lock(&dev_base_lock);
931 for(dev_out = dev_base; dev_out; dev_out = dev_out->next) {
932 if (!dev_out->dn_ptr)
933 continue;
9bbf28a1
PC		 934 if (!dn_dev_islocal(dev_out, oldflp->fld_src))
935 continue;
936 if ((dev_out->flags & IFF_LOOPBACK) &&
937 oldflp->fld_dst &&
938 !dn_dev_islocal(dev_out, oldflp->fld_dst))
939 continue;
940 break;
1da177e4
LT		 941 }
942 read_unlock(&dev_base_lock);
943 if (dev_out == NULL)
944 goto out;
945 dev_hold(dev_out);
946source_ok:
947 ;
948 }
949
950 /* No destination? Assume its local */
951 if (!fl.fld_dst) {
952 fl.fld_dst = fl.fld_src;
953
954 err = -EADDRNOTAVAIL;
955 if (dev_out)
956 dev_put(dev_out);
957 dev_out = &loopback_dev;
958 dev_hold(dev_out);
959 if (!fl.fld_dst) {
960 fl.fld_dst =
961 fl.fld_src = dnet_select_source(dev_out, 0,
962 RT_SCOPE_HOST);
963 if (!fl.fld_dst)
964 goto out;
965 }
966 fl.oif = loopback_dev.ifindex;
967 res.type = RTN_LOCAL;
968 goto make_route;
969 }
970
971 if (decnet_debug_level & 16)
972 printk(KERN_DEBUG
973 "dn_route_output_slow: initial checks complete."
c4ea94ab
SW		 974 " dst=%o4x src=%04x oif=%d try_hard=%d\n",
975 dn_ntohs(fl.fld_dst), dn_ntohs(fl.fld_src),
976 fl.oif, try_hard);
1da177e4
LT		 977
978 /*
979 * N.B. If the kernel is compiled without router support then
980 * dn_fib_lookup() will evaluate to non-zero so this if () block
981 * will always be executed.
982 */
983 err = -ESRCH;
984 if (try_hard || (err = dn_fib_lookup(&fl, &res)) != 0) {
985 struct dn_dev *dn_db;
986 if (err != -ESRCH)
987 goto out;
988 /*
989 * Here the fallback is basically the standard algorithm for
990 * routing in endnodes which is described in the DECnet routing
991 * docs
992 *
993 * If we are not trying hard, look in neighbour cache.
994 * The result is tested to ensure that if a specific output
995 * device/source address was requested, then we honour that
996 * here
997 */
998 if (!try_hard) {
999 neigh = neigh_lookup_nodev(&dn_neigh_table, &fl.fld_dst);
1000 if (neigh) {
1001 if ((oldflp->oif &&
1002 (neigh->dev->ifindex != oldflp->oif)) ||
1003 (oldflp->fld_src &&
1004 (!dn_dev_islocal(neigh->dev,
1005 oldflp->fld_src)))) {
1006 neigh_release(neigh);
1007 neigh = NULL;
1008 } else {
1009 if (dev_out)
1010 dev_put(dev_out);
1011 if (dn_dev_islocal(neigh->dev, fl.fld_dst)) {
1012 dev_out = &loopback_dev;
1013 res.type = RTN_LOCAL;
1014 } else {
1015 dev_out = neigh->dev;
1016 }
1017 dev_hold(dev_out);
1018 goto select_source;
1019 }
1020 }
1021 }
1022
1023 /* Not there? Perhaps its a local address */
1024 if (dev_out == NULL)
1025 dev_out = dn_dev_get_default();
1026 err = -ENODEV;
1027 if (dev_out == NULL)
1028 goto out;
1029 dn_db = dev_out->dn_ptr;
1030 /* Possible improvement - check all devices for local addr */
1031 if (dn_dev_islocal(dev_out, fl.fld_dst)) {
1032 dev_put(dev_out);
1033 dev_out = &loopback_dev;
1034 dev_hold(dev_out);
1035 res.type = RTN_LOCAL;
1036 goto select_source;
1037 }
1038 /* Not local either.... try sending it to the default router */
1039 neigh = neigh_clone(dn_db->router);
1040 BUG_ON(neigh && neigh->dev != dev_out);
1041
1042 /* Ok then, we assume its directly connected and move on */
1043select_source:
1044 if (neigh)
1045 gateway = ((struct dn_neigh *)neigh)->addr;
1046 if (gateway == 0)
1047 gateway = fl.fld_dst;
1048 if (fl.fld_src == 0) {
1049 fl.fld_src = dnet_select_source(dev_out, gateway,
1050 res.type == RTN_LOCAL ?
1051 RT_SCOPE_HOST :
1052 RT_SCOPE_LINK);
1053 if (fl.fld_src == 0 && res.type != RTN_LOCAL)
1054 goto e_addr;
1055 }
1056 fl.oif = dev_out->ifindex;
1057 goto make_route;
1058 }
1059 free_res = 1;
1060
1061 if (res.type == RTN_NAT)
1062 goto e_inval;
1063
1064 if (res.type == RTN_LOCAL) {
1065 if (!fl.fld_src)
1066 fl.fld_src = fl.fld_dst;
1067 if (dev_out)
1068 dev_put(dev_out);
1069 dev_out = &loopback_dev;
1070 dev_hold(dev_out);
1071 fl.oif = dev_out->ifindex;
1072 if (res.fi)
1073 dn_fib_info_put(res.fi);
1074 res.fi = NULL;
1075 goto make_route;
1076 }
1077
1078 if (res.fi->fib_nhs > 1 && fl.oif == 0)
1079 dn_fib_select_multipath(&fl, &res);
1080
1081 /*
1082 * We could add some logic to deal with default routes here and
1083 * get rid of some of the special casing above.
1084 */
1085
1086 if (!fl.fld_src)
1087 fl.fld_src = DN_FIB_RES_PREFSRC(res);
1088
1089 if (dev_out)
1090 dev_put(dev_out);
1091 dev_out = DN_FIB_RES_DEV(res);
1092 dev_hold(dev_out);
1093 fl.oif = dev_out->ifindex;
1094 gateway = DN_FIB_RES_GW(res);
1095
1096make_route:
1097 if (dev_out->flags & IFF_LOOPBACK)
1098 flags |= RTCF_LOCAL;
1099
1100 rt = dst_alloc(&dn_dst_ops);
1101 if (rt == NULL)
1102 goto e_nobufs;
1103
1104 atomic_set(&rt->u.dst.__refcnt, 1);
1105 rt->u.dst.flags = DST_HOST;
1106
1107 rt->fl.fld_src = oldflp->fld_src;
1108 rt->fl.fld_dst = oldflp->fld_dst;
1109 rt->fl.oif = oldflp->oif;
1110 rt->fl.iif = 0;
1111#ifdef CONFIG_DECNET_ROUTE_FWMARK
1112 rt->fl.fld_fwmark = oldflp->fld_fwmark;
1113#endif
1114
1115 rt->rt_saddr = fl.fld_src;
1116 rt->rt_daddr = fl.fld_dst;
1117 rt->rt_gateway = gateway ? gateway : fl.fld_dst;
1118 rt->rt_local_src = fl.fld_src;
1119
1120 rt->rt_dst_map = fl.fld_dst;
1121 rt->rt_src_map = fl.fld_src;
1122
1123 rt->u.dst.dev = dev_out;
1124 dev_hold(dev_out);
1125 rt->u.dst.neighbour = neigh;
1126 neigh = NULL;
1127
1128 rt->u.dst.lastuse = jiffies;
1129 rt->u.dst.output = dn_output;
1130 rt->u.dst.input = dn_rt_bug;
1131 rt->rt_flags = flags;
1132 if (flags & RTCF_LOCAL)
1133 rt->u.dst.input = dn_nsp_rx;
1134
1135 err = dn_rt_set_next_hop(rt, &res);
1136 if (err)
1137 goto e_neighbour;
1138
1139 hash = dn_hash(rt->fl.fld_src, rt->fl.fld_dst);
1140 dn_insert_route(rt, hash, (struct dn_route **)pprt);
1141
1142done:
1143 if (neigh)
1144 neigh_release(neigh);
1145 if (free_res)
1146 dn_fib_res_put(&res);
1147 if (dev_out)
1148 dev_put(dev_out);
1149out:
1150 return err;
1151
1152e_addr:
1153 err = -EADDRNOTAVAIL;
1154 goto done;
1155e_inval:
1156 err = -EINVAL;
1157 goto done;
1158e_nobufs:
1159 err = -ENOBUFS;
1160 goto done;
1161e_neighbour:
1162 dst_free(&rt->u.dst);
1163 goto e_nobufs;
1164}
1165
1166
1167/*
1168 * N.B. The flags may be moved into the flowi at some future stage.
1169 */
1170static int __dn_route_output_key(struct dst_entry **pprt, const struct flowi *flp, int flags)
1171{
1172 unsigned hash = dn_hash(flp->fld_src, flp->fld_dst);
1173 struct dn_route *rt = NULL;
1174
1175 if (!(flags & MSG_TRYHARD)) {
1176 rcu_read_lock_bh();
1177 for(rt = rcu_dereference(dn_rt_hash_table[hash].chain); rt;
1178 rt = rcu_dereference(rt->u.rt_next)) {
1179 if ((flp->fld_dst == rt->fl.fld_dst) &&
1180 (flp->fld_src == rt->fl.fld_src) &&
1181#ifdef CONFIG_DECNET_ROUTE_FWMARK
1182 (flp->fld_fwmark == rt->fl.fld_fwmark) &&
1183#endif
1184 (rt->fl.iif == 0) &&
1185 (rt->fl.oif == flp->oif)) {
1186 rt->u.dst.lastuse = jiffies;
1187 dst_hold(&rt->u.dst);
1188 rt->u.dst.__use++;
1189 rcu_read_unlock_bh();
1190 *pprt = &rt->u.dst;
1191 return 0;
1192 }
1193 }
1194 rcu_read_unlock_bh();
1195 }
1196
1197 return dn_route_output_slow(pprt, flp, flags);
1198}
1199
1200static int dn_route_output_key(struct dst_entry **pprt, struct flowi *flp, int flags)
1201{
1202 int err;
1203
1204 err = __dn_route_output_key(pprt, flp, flags);
1205 if (err == 0 && flp->proto) {
1206 err = xfrm_lookup(pprt, flp, NULL, 0);
1207 }
1208 return err;
1209}
1210
1211int dn_route_output_sock(struct dst_entry **pprt, struct flowi *fl, struct sock *sk, int flags)
1212{
1213 int err;
1214
1215 err = __dn_route_output_key(pprt, fl, flags & MSG_TRYHARD);
1216 if (err == 0 && fl->proto) {
1217 err = xfrm_lookup(pprt, fl, sk, !(flags & MSG_DONTWAIT));
1218 }
1219 return err;
1220}
1221
1222static int dn_route_input_slow(struct sk_buff *skb)
1223{
1224 struct dn_route *rt = NULL;
1225 struct dn_skb_cb *cb = DN_SKB_CB(skb);
1226 struct net_device *in_dev = skb->dev;
1227 struct net_device *out_dev = NULL;
1228 struct dn_dev *dn_db;
1229 struct neighbour *neigh = NULL;
1230 unsigned hash;
1231 int flags = 0;
c4ea94ab
SW		 1232 __le16 gateway = 0;
1233 __le16 local_src = 0;
1da177e4
LT		 1234 struct flowi fl = { .nl_u = { .dn_u =
1235 { .daddr = cb->dst,
1236 .saddr = cb->src,
1237 .scope = RT_SCOPE_UNIVERSE,
1238#ifdef CONFIG_DECNET_ROUTE_FWMARK
82e91ffe 1239 .fwmark = skb->mark
1da177e4
LT		 1240#endif
1241 } },
1242 .iif = skb->dev->ifindex };
1243 struct dn_fib_res res = { .fi = NULL, .type = RTN_UNREACHABLE };
1244 int err = -EINVAL;
1245 int free_res = 0;
1246
1247 dev_hold(in_dev);
1248
1249 if ((dn_db = in_dev->dn_ptr) == NULL)
1250 goto out;
1251
1252 /* Zero source addresses are not allowed */
1253 if (fl.fld_src == 0)
1254 goto out;
1255
1256 /*
1257 * In this case we've just received a packet from a source
1258 * outside ourselves pretending to come from us. We don't
1259 * allow it any further to prevent routing loops, spoofing and
1260 * other nasties. Loopback packets already have the dst attached
1261 * so this only affects packets which have originated elsewhere.
1262 */
1263 err = -ENOTUNIQ;
1264 if (dn_dev_islocal(in_dev, cb->src))
1265 goto out;
1266
1267 err = dn_fib_lookup(&fl, &res);
1268 if (err) {
1269 if (err != -ESRCH)
1270 goto out;
1271 /*
1272 * Is the destination us ?
1273 */
1274 if (!dn_dev_islocal(in_dev, cb->dst))
1275 goto e_inval;
1276
1277 res.type = RTN_LOCAL;
1da177e4 1278 } else {
c4ea94ab 1279 __le16 src_map = fl.fld_src;
1da177e4
LT		 1280 free_res = 1;
1281
1282 out_dev = DN_FIB_RES_DEV(res);
1283 if (out_dev == NULL) {
1284 if (net_ratelimit())
1285 printk(KERN_CRIT "Bug in dn_route_input_slow() "
1286 "No output device\n");
1287 goto e_inval;
1288 }
1289 dev_hold(out_dev);
1290
1291 if (res.r)
a8731cbf 1292 src_map = fl.fld_src; /* no NAT support for now */
1da177e4
LT		 1293
1294 gateway = DN_FIB_RES_GW(res);
1295 if (res.type == RTN_NAT) {
1296 fl.fld_dst = dn_fib_rules_map_destination(fl.fld_dst, &res);
1297 dn_fib_res_put(&res);
1298 free_res = 0;
1299 if (dn_fib_lookup(&fl, &res))
1300 goto e_inval;
1301 free_res = 1;
1302 if (res.type != RTN_UNICAST)
1303 goto e_inval;
1304 flags |= RTCF_DNAT;
1305 gateway = fl.fld_dst;
1306 }
1307 fl.fld_src = src_map;
1308 }
1309
1310 switch(res.type) {
1311 case RTN_UNICAST:
1312 /*
1313 * Forwarding check here, we only check for forwarding
1314 * being turned off, if you want to only forward intra
1315 * area, its up to you to set the routing tables up
1316 * correctly.
1317 */
1318 if (dn_db->parms.forwarding == 0)
1319 goto e_inval;
1320
1321 if (res.fi->fib_nhs > 1 && fl.oif == 0)
1322 dn_fib_select_multipath(&fl, &res);
1323
1324 /*
1325 * Check for out_dev == in_dev. We use the RTCF_DOREDIRECT
1326 * flag as a hint to set the intra-ethernet bit when
1327 * forwarding. If we've got NAT in operation, we don't do
1328 * this optimisation.
1329 */
1330 if (out_dev == in_dev && !(flags & RTCF_NAT))
1331 flags |= RTCF_DOREDIRECT;
1332
1333 local_src = DN_FIB_RES_PREFSRC(res);
1334
1335 case RTN_BLACKHOLE:
1336 case RTN_UNREACHABLE:
1337 break;
1338 case RTN_LOCAL:
1339 flags |= RTCF_LOCAL;
1340 fl.fld_src = cb->dst;
1341 fl.fld_dst = cb->src;
1342
1343 /* Routing tables gave us a gateway */
1344 if (gateway)
1345 goto make_route;
1346
1347 /* Packet was intra-ethernet, so we know its on-link */
3a31b9d2 1348 if (cb->rt_flags & DN_RT_F_IE) {
1da177e4
LT		 1349 gateway = cb->src;
1350 flags |= RTCF_DIRECTSRC;
1351 goto make_route;
1352 }
1353
1354 /* Use the default router if there is one */
1355 neigh = neigh_clone(dn_db->router);
1356 if (neigh) {
1357 gateway = ((struct dn_neigh *)neigh)->addr;
1358 goto make_route;
1359 }
1360
1361 /* Close eyes and pray */
1362 gateway = cb->src;
1363 flags |= RTCF_DIRECTSRC;
1364 goto make_route;
1365 default:
1366 goto e_inval;
1367 }
1368
1369make_route:
1370 rt = dst_alloc(&dn_dst_ops);
1371 if (rt == NULL)
1372 goto e_nobufs;
1373
1374 rt->rt_saddr = fl.fld_src;
1375 rt->rt_daddr = fl.fld_dst;
1376 rt->rt_gateway = fl.fld_dst;
1377 if (gateway)
1378 rt->rt_gateway = gateway;
1379 rt->rt_local_src = local_src ? local_src : rt->rt_saddr;
1380
1381 rt->rt_dst_map = fl.fld_dst;
1382 rt->rt_src_map = fl.fld_src;
1383
1384 rt->fl.fld_src = cb->src;
1385 rt->fl.fld_dst = cb->dst;
1386 rt->fl.oif = 0;
1387 rt->fl.iif = in_dev->ifindex;
1388 rt->fl.fld_fwmark = fl.fld_fwmark;
1389
1390 rt->u.dst.flags = DST_HOST;
1391 rt->u.dst.neighbour = neigh;
1392 rt->u.dst.dev = out_dev;
1393 rt->u.dst.lastuse = jiffies;
1394 rt->u.dst.output = dn_rt_bug;
1395 switch(res.type) {
1396 case RTN_UNICAST:
1397 rt->u.dst.input = dn_forward;
1398 break;
1399 case RTN_LOCAL:
1400 rt->u.dst.output = dn_output;
1401 rt->u.dst.input = dn_nsp_rx;
1402 rt->u.dst.dev = in_dev;
1403 flags |= RTCF_LOCAL;
1404 break;
1405 default:
1406 case RTN_UNREACHABLE:
1407 case RTN_BLACKHOLE:
1408 rt->u.dst.input = dn_blackhole;
1409 }
1410 rt->rt_flags = flags;
1411 if (rt->u.dst.dev)
1412 dev_hold(rt->u.dst.dev);
1413
1414 err = dn_rt_set_next_hop(rt, &res);
1415 if (err)
1416 goto e_neighbour;
1417
1418 hash = dn_hash(rt->fl.fld_src, rt->fl.fld_dst);
1419 dn_insert_route(rt, hash, (struct dn_route **)&skb->dst);
1420
1421done:
1422 if (neigh)
1423 neigh_release(neigh);
1424 if (free_res)
1425 dn_fib_res_put(&res);
1426 dev_put(in_dev);
1427 if (out_dev)
1428 dev_put(out_dev);
1429out:
1430 return err;
1431
1432e_inval:
1433 err = -EINVAL;
1434 goto done;
1435
1436e_nobufs:
1437 err = -ENOBUFS;
1438 goto done;
1439
1440e_neighbour:
1441 dst_free(&rt->u.dst);
1442 goto done;
1443}
1444
1445int dn_route_input(struct sk_buff *skb)
1446{
1447 struct dn_route *rt;
1448 struct dn_skb_cb *cb = DN_SKB_CB(skb);
1449 unsigned hash = dn_hash(cb->src, cb->dst);
1450
1451 if (skb->dst)
1452 return 0;
1453
1454 rcu_read_lock();
1455 for(rt = rcu_dereference(dn_rt_hash_table[hash].chain); rt != NULL;
1456 rt = rcu_dereference(rt->u.rt_next)) {
1457 if ((rt->fl.fld_src == cb->src) &&
1458 (rt->fl.fld_dst == cb->dst) &&
1459 (rt->fl.oif == 0) &&
1460#ifdef CONFIG_DECNET_ROUTE_FWMARK
82e91ffe 1461 (rt->fl.fld_fwmark == skb->mark) &&
1da177e4
LT		 1462#endif
1463 (rt->fl.iif == cb->iif)) {
1464 rt->u.dst.lastuse = jiffies;
1465 dst_hold(&rt->u.dst);
1466 rt->u.dst.__use++;
1467 rcu_read_unlock();
1468 skb->dst = (struct dst_entry *)rt;
1469 return 0;
1470 }
1471 }
1472 rcu_read_unlock();
1473
1474 return dn_route_input_slow(skb);
1475}
1476
b6544c0b
JHS		 1477static int dn_rt_fill_info(struct sk_buff *skb, u32 pid, u32 seq,
1478 int event, int nowait, unsigned int flags)
1da177e4
LT		 1479{
1480 struct dn_route *rt = (struct dn_route *)skb->dst;
1481 struct rtmsg *r;
1482 struct nlmsghdr *nlh;
1483 unsigned char *b = skb->tail;
1484 struct rta_cacheinfo ci;
1485
b6544c0b 1486 nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*r), flags);
1da177e4 1487 r = NLMSG_DATA(nlh);
1da177e4
LT		 1488 r->rtm_family = AF_DECnet;
1489 r->rtm_dst_len = 16;
1490 r->rtm_src_len = 0;
1491 r->rtm_tos = 0;
1492 r->rtm_table = RT_TABLE_MAIN;
9e762a4a 1493 RTA_PUT_U32(skb, RTA_TABLE, RT_TABLE_MAIN);
1da177e4
LT		 1494 r->rtm_type = rt->rt_type;
1495 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
1496 r->rtm_scope = RT_SCOPE_UNIVERSE;
1497 r->rtm_protocol = RTPROT_UNSPEC;
1498 if (rt->rt_flags & RTCF_NOTIFY)
1499 r->rtm_flags |= RTM_F_NOTIFY;
1500 RTA_PUT(skb, RTA_DST, 2, &rt->rt_daddr);
1501 if (rt->fl.fld_src) {
1502 r->rtm_src_len = 16;
1503 RTA_PUT(skb, RTA_SRC, 2, &rt->fl.fld_src);
1504 }
1505 if (rt->u.dst.dev)
1506 RTA_PUT(skb, RTA_OIF, sizeof(int), &rt->u.dst.dev->ifindex);
1507 /*
1508 * Note to self - change this if input routes reverse direction when
1509 * they deal only with inputs and not with replies like they do
1510 * currently.
1511 */
1512 RTA_PUT(skb, RTA_PREFSRC, 2, &rt->rt_local_src);
1513 if (rt->rt_daddr != rt->rt_gateway)
1514 RTA_PUT(skb, RTA_GATEWAY, 2, &rt->rt_gateway);
1515 if (rtnetlink_put_metrics(skb, rt->u.dst.metrics) < 0)
1516 goto rtattr_failure;
1517 ci.rta_lastuse = jiffies_to_clock_t(jiffies - rt->u.dst.lastuse);
1518 ci.rta_used = rt->u.dst.__use;
1519 ci.rta_clntref = atomic_read(&rt->u.dst.__refcnt);
1520 if (rt->u.dst.expires)
1521 ci.rta_expires = jiffies_to_clock_t(rt->u.dst.expires - jiffies);
1522 else
1523 ci.rta_expires = 0;
1524 ci.rta_error = rt->u.dst.error;
1525 ci.rta_id = ci.rta_ts = ci.rta_tsage = 0;
1526 RTA_PUT(skb, RTA_CACHEINFO, sizeof(ci), &ci);
1527 if (rt->fl.iif)
1528 RTA_PUT(skb, RTA_IIF, sizeof(int), &rt->fl.iif);
1529
1530 nlh->nlmsg_len = skb->tail - b;
1531 return skb->len;
1532
1533nlmsg_failure:
1534rtattr_failure:
1535 skb_trim(skb, b - skb->data);
1536 return -1;
1537}
1538
1539/*
1540 * This is called by both endnodes and routers now.
1541 */
1542int dn_cache_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh, void *arg)
1543{
1544 struct rtattr **rta = arg;
1545 struct rtmsg *rtm = NLMSG_DATA(nlh);
1546 struct dn_route *rt = NULL;
1547 struct dn_skb_cb *cb;
1548 int err;
1549 struct sk_buff *skb;
1550 struct flowi fl;
1551
1552 memset(&fl, 0, sizeof(fl));
1553 fl.proto = DNPROTO_NSP;
1554
1555 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1556 if (skb == NULL)
1557 return -ENOBUFS;
1558 skb->mac.raw = skb->data;
1559 cb = DN_SKB_CB(skb);
1560
1561 if (rta[RTA_SRC-1])
1562 memcpy(&fl.fld_src, RTA_DATA(rta[RTA_SRC-1]), 2);
1563 if (rta[RTA_DST-1])
1564 memcpy(&fl.fld_dst, RTA_DATA(rta[RTA_DST-1]), 2);
1565 if (rta[RTA_IIF-1])
1566 memcpy(&fl.iif, RTA_DATA(rta[RTA_IIF-1]), sizeof(int));
1567
1568 if (fl.iif) {
1569 struct net_device *dev;
1570 if ((dev = dev_get_by_index(fl.iif)) == NULL) {
1571 kfree_skb(skb);
1572 return -ENODEV;
1573 }
1574 if (!dev->dn_ptr) {
1575 dev_put(dev);
1576 kfree_skb(skb);
1577 return -ENODEV;
1578 }
1579 skb->protocol = __constant_htons(ETH_P_DNA_RT);
1580 skb->dev = dev;
1581 cb->src = fl.fld_src;
1582 cb->dst = fl.fld_dst;
1583 local_bh_disable();
1584 err = dn_route_input(skb);
1585 local_bh_enable();
1586 memset(cb, 0, sizeof(struct dn_skb_cb));
1587 rt = (struct dn_route *)skb->dst;
1588 if (!err && -rt->u.dst.error)
1589 err = rt->u.dst.error;
1590 } else {
1591 int oif = 0;
1592 if (rta[RTA_OIF - 1])
1593 memcpy(&oif, RTA_DATA(rta[RTA_OIF - 1]), sizeof(int));
1594 fl.oif = oif;
1595 err = dn_route_output_key((struct dst_entry **)&rt, &fl, 0);
1596 }
1597
1598 if (skb->dev)
1599 dev_put(skb->dev);
1600 skb->dev = NULL;
1601 if (err)
1602 goto out_free;
1603 skb->dst = &rt->u.dst;
1604 if (rtm->rtm_flags & RTM_F_NOTIFY)
1605 rt->rt_flags |= RTCF_NOTIFY;
1606
1607 NETLINK_CB(skb).dst_pid = NETLINK_CB(in_skb).pid;
1608
b6544c0b 1609 err = dn_rt_fill_info(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq, RTM_NEWROUTE, 0, 0);
1da177e4
LT		 1610
1611 if (err == 0)
1612 goto out_free;
1613 if (err < 0) {
1614 err = -EMSGSIZE;
1615 goto out_free;
1616 }
1617
2942e900 1618 return rtnl_unicast(skb, NETLINK_CB(in_skb).pid);
1da177e4
LT		 1619
1620out_free:
1621 kfree_skb(skb);
1622 return err;
1623}
1624
1625/*
1626 * For routers, this is called from dn_fib_dump, but for endnodes its
1627 * called directly from the rtnetlink dispatch table.
1628 */
1629int dn_cache_dump(struct sk_buff *skb, struct netlink_callback *cb)
1630{
1631 struct dn_route *rt;
1632 int h, s_h;
1633 int idx, s_idx;
1634
1635 if (NLMSG_PAYLOAD(cb->nlh, 0) < sizeof(struct rtmsg))
1636 return -EINVAL;
1637 if (!(((struct rtmsg *)NLMSG_DATA(cb->nlh))->rtm_flags&RTM_F_CLONED))
1638 return 0;
1639
1640 s_h = cb->args[0];
1641 s_idx = idx = cb->args[1];
1642 for(h = 0; h <= dn_rt_hash_mask; h++) {
1643 if (h < s_h)
1644 continue;
1645 if (h > s_h)
1646 s_idx = 0;
1647 rcu_read_lock_bh();
1648 for(rt = rcu_dereference(dn_rt_hash_table[h].chain), idx = 0;
1649 rt;
1650 rt = rcu_dereference(rt->u.rt_next), idx++) {
1651 if (idx < s_idx)
1652 continue;
1653 skb->dst = dst_clone(&rt->u.dst);
1654 if (dn_rt_fill_info(skb, NETLINK_CB(cb->skb).pid,
b6544c0b
JHS		 1655 cb->nlh->nlmsg_seq, RTM_NEWROUTE,
1656 1, NLM_F_MULTI) <= 0) {
1da177e4
LT		 1657 dst_release(xchg(&skb->dst, NULL));
1658 rcu_read_unlock_bh();
1659 goto done;
1660 }
1661 dst_release(xchg(&skb->dst, NULL));
1662 }
1663 rcu_read_unlock_bh();
1664 }
1665
1666done:
1667 cb->args[0] = h;
1668 cb->args[1] = idx;
1669 return skb->len;
1670}
1671
1672#ifdef CONFIG_PROC_FS
1673struct dn_rt_cache_iter_state {
1674 int bucket;
1675};
1676
1677static struct dn_route *dn_rt_cache_get_first(struct seq_file *seq)
1678{
1679 struct dn_route *rt = NULL;
1680 struct dn_rt_cache_iter_state *s = seq->private;
1681
1682 for(s->bucket = dn_rt_hash_mask; s->bucket >= 0; --s->bucket) {
1683 rcu_read_lock_bh();
1684 rt = dn_rt_hash_table[s->bucket].chain;
1685 if (rt)
1686 break;
1687 rcu_read_unlock_bh();
1688 }
1689 return rt;
1690}
1691
1692static struct dn_route *dn_rt_cache_get_next(struct seq_file *seq, struct dn_route *rt)
1693{
1694 struct dn_rt_cache_iter_state *s = rcu_dereference(seq->private);
1695
1696 rt = rt->u.rt_next;
1697 while(!rt) {
1698 rcu_read_unlock_bh();
1699 if (--s->bucket < 0)
1700 break;
1701 rcu_read_lock_bh();
1702 rt = dn_rt_hash_table[s->bucket].chain;
1703 }
1704 return rt;
1705}
1706
1707static void *dn_rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
1708{
1709 struct dn_route *rt = dn_rt_cache_get_first(seq);
1710
1711 if (rt) {
1712 while(*pos && (rt = dn_rt_cache_get_next(seq, rt)))
1713 --*pos;
1714 }
1715 return *pos ? NULL : rt;
1716}
1717
1718static void *dn_rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1719{
1720 struct dn_route *rt = dn_rt_cache_get_next(seq, v);
1721 ++*pos;
1722 return rt;
1723}
1724
1725static void dn_rt_cache_seq_stop(struct seq_file *seq, void *v)
1726{
1727 if (v)
1728 rcu_read_unlock_bh();
1729}
1730
1731static int dn_rt_cache_seq_show(struct seq_file *seq, void *v)
1732{
1733 struct dn_route *rt = v;
1734 char buf1[DN_ASCBUF_LEN], buf2[DN_ASCBUF_LEN];
1735
1736 seq_printf(seq, "%-8s %-7s %-7s %04d %04d %04d\n",
1737 rt->u.dst.dev ? rt->u.dst.dev->name : "*",
1738 dn_addr2asc(dn_ntohs(rt->rt_daddr), buf1),
1739 dn_addr2asc(dn_ntohs(rt->rt_saddr), buf2),
1740 atomic_read(&rt->u.dst.__refcnt),
1741 rt->u.dst.__use,
1742 (int) dst_metric(&rt->u.dst, RTAX_RTT));
1743 return 0;
1744}
1745
1746static struct seq_operations dn_rt_cache_seq_ops = {
1747 .start = dn_rt_cache_seq_start,
1748 .next = dn_rt_cache_seq_next,
1749 .stop = dn_rt_cache_seq_stop,
1750 .show = dn_rt_cache_seq_show,
1751};
1752
1753static int dn_rt_cache_seq_open(struct inode *inode, struct file *file)
1754{
1755 struct seq_file *seq;
1756 int rc = -ENOMEM;
1757 struct dn_rt_cache_iter_state *s = kmalloc(sizeof(*s), GFP_KERNEL);
1758
1759 if (!s)
1760 goto out;
1761 rc = seq_open(file, &dn_rt_cache_seq_ops);
1762 if (rc)
1763 goto out_kfree;
1764 seq = file->private_data;
1765 seq->private = s;
1766 memset(s, 0, sizeof(*s));
1767out:
1768 return rc;
1769out_kfree:
1770 kfree(s);
1771 goto out;
1772}
1773
1774static struct file_operations dn_rt_cache_seq_fops = {
1775 .owner = THIS_MODULE,
1776 .open = dn_rt_cache_seq_open,
1777 .read = seq_read,
1778 .llseek = seq_lseek,
1779 .release = seq_release_private,
1780};
1781
1782#endif /* CONFIG_PROC_FS */
1783
1784void __init dn_route_init(void)
1785{
1786 int i, goal, order;
1787
e5d679f3
AD		 1788 dn_dst_ops.kmem_cachep =
1789 kmem_cache_create("dn_dst_cache", sizeof(struct dn_route), 0,
1790 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1da177e4
LT		 1791 init_timer(&dn_route_timer);
1792 dn_route_timer.function = dn_dst_check_expire;
1793 dn_route_timer.expires = jiffies + decnet_dst_gc_interval * HZ;
1794 add_timer(&dn_route_timer);
1795
1796 goal = num_physpages >> (26 - PAGE_SHIFT);
1797
1798 for(order = 0; (1UL << order) < goal; order++)
1799 /* NOTHING */;
1800
1801 /*
1802 * Only want 1024 entries max, since the table is very, very unlikely
1803 * to be larger than that.
1804 */
1805 while(order && ((((1UL << order) * PAGE_SIZE) /
1806 sizeof(struct dn_rt_hash_bucket)) >= 2048))
1807 order--;
1808
1809 do {
1810 dn_rt_hash_mask = (1UL << order) * PAGE_SIZE /
1811 sizeof(struct dn_rt_hash_bucket);
1812 while(dn_rt_hash_mask & (dn_rt_hash_mask - 1))
1813 dn_rt_hash_mask--;
1814 dn_rt_hash_table = (struct dn_rt_hash_bucket *)
1815 __get_free_pages(GFP_ATOMIC, order);
1816 } while (dn_rt_hash_table == NULL && --order > 0);
1817
1818 if (!dn_rt_hash_table)
1819 panic("Failed to allocate DECnet route cache hash table\n");
1820
1821 printk(KERN_INFO
1822 "DECnet: Routing cache hash table of %u buckets, %ldKbytes\n",
1823 dn_rt_hash_mask,
1824 (long)(dn_rt_hash_mask*sizeof(struct dn_rt_hash_bucket))/1024);
1825
1826 dn_rt_hash_mask--;
1827 for(i = 0; i <= dn_rt_hash_mask; i++) {
1828 spin_lock_init(&dn_rt_hash_table[i].lock);
1829 dn_rt_hash_table[i].chain = NULL;
1830 }
1831
1832 dn_dst_ops.gc_thresh = (dn_rt_hash_mask + 1);
1833
1834 proc_net_fops_create("decnet_cache", S_IRUGO, &dn_rt_cache_seq_fops);
1835}
1836
1837void __exit dn_route_cleanup(void)
1838{
1839 del_timer(&dn_route_timer);
1840 dn_run_flush(0);
1841
1842 proc_net_remove("decnet_cache");
1843}
1844
Clone of davem's net-next-2.6 tree