blob: 3407f190afe8892699025bf504c73111a082b8b5 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * DECnet An implementation of the DECnet protocol suite for the LINUX
3 * operating system. DECnet is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * DECnet Routing Functions (Endnode and Router)
7 *
8 * Authors: Steve Whitehouse <SteveW@ACM.org>
9 * Eduardo Marcelo Serrat <emserrat@geocities.com>
10 *
11 * Changes:
12 * Steve Whitehouse : Fixes to allow "intra-ethernet" and
13 * "return-to-sender" bits on outgoing
14 * packets.
15 * Steve Whitehouse : Timeouts for cached routes.
16 * Steve Whitehouse : Use dst cache for input routes too.
17 * Steve Whitehouse : Fixed error values in dn_send_skb.
18 * Steve Whitehouse : Rework routing functions to better fit
19 * DECnet routing design
20 * Alexey Kuznetsov : New SMP locking
21 * Steve Whitehouse : More SMP locking changes & dn_cache_dump()
22 * Steve Whitehouse : Prerouting NF hook, now really is prerouting.
23 * Fixed possible skb leak in rtnetlink funcs.
24 * Steve Whitehouse : Dave Miller's dynamic hash table sizing and
25 * Alexey Kuznetsov's finer grained locking
26 * from ipv4/route.c.
27 * Steve Whitehouse : Routing is now starting to look like a
28 * sensible set of code now, mainly due to
29 * my copying the IPv4 routing code. The
30 * hooks here are modified and will continue
31 * to evolve for a while.
32 * Steve Whitehouse : Real SMP at last :-) Also new netfilter
33 * stuff. Look out raw sockets your days
34 * are numbered!
35 * Steve Whitehouse : Added return-to-sender functions. Added
36 * backlog congestion level return codes.
37 * Steve Whitehouse : Fixed bug where routes were set up with
38 * no ref count on net devices.
39 * Steve Whitehouse : RCU for the route cache
40 * Steve Whitehouse : Preparations for the flow cache
41 * Steve Whitehouse : Prepare for nonlinear skbs
42 */
43
44/******************************************************************************
45 (c) 1995-1998 E.M. Serrat emserrat@geocities.com
46
47 This program is free software; you can redistribute it and/or modify
48 it under the terms of the GNU General Public License as published by
49 the Free Software Foundation; either version 2 of the License, or
50 any later version.
51
52 This program is distributed in the hope that it will be useful,
53 but WITHOUT ANY WARRANTY; without even the implied warranty of
54 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
55 GNU General Public License for more details.
56*******************************************************************************/
57
58#include <linux/config.h>
59#include <linux/errno.h>
60#include <linux/types.h>
61#include <linux/socket.h>
62#include <linux/in.h>
63#include <linux/kernel.h>
64#include <linux/sockios.h>
65#include <linux/net.h>
66#include <linux/netdevice.h>
67#include <linux/inet.h>
68#include <linux/route.h>
69#include <linux/in_route.h>
70#include <net/sock.h>
71#include <linux/mm.h>
72#include <linux/proc_fs.h>
73#include <linux/seq_file.h>
74#include <linux/init.h>
75#include <linux/rtnetlink.h>
76#include <linux/string.h>
77#include <linux/netfilter_decnet.h>
78#include <linux/rcupdate.h>
79#include <linux/times.h>
80#include <asm/errno.h>
81#include <net/neighbour.h>
82#include <net/dst.h>
83#include <net/flow.h>
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;
Ingo Molnar8d06afa2005-09-09 13:10:40 -0700120static DEFINE_TIMER(dn_rt_flush_timer, dn_run_flush, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121int 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
136static __inline__ unsigned dn_hash(unsigned short src, unsigned short dst)
137{
138 unsigned short tmp = src ^ dst;
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{
152 if (rt)
153 dst_release(&rt->u.dst);
154 call_rcu_bh(&rt->u.dst.rcu_head, dst_rcu_free);
155}
156
157static void dn_dst_check_expire(unsigned long dummy)
158{
159 int i;
160 struct dn_route *rt, **rtp;
161 unsigned long now = jiffies;
162 unsigned long expire = 120 * HZ;
163
164 for(i = 0; i <= dn_rt_hash_mask; i++) {
165 rtp = &dn_rt_hash_table[i].chain;
166
167 spin_lock(&dn_rt_hash_table[i].lock);
168 while((rt=*rtp) != NULL) {
169 if (atomic_read(&rt->u.dst.__refcnt) ||
170 (now - rt->u.dst.lastuse) < expire) {
171 rtp = &rt->u.rt_next;
172 continue;
173 }
174 *rtp = rt->u.rt_next;
175 rt->u.rt_next = NULL;
176 dnrt_free(rt);
177 }
178 spin_unlock(&dn_rt_hash_table[i].lock);
179
180 if ((jiffies - now) > 0)
181 break;
182 }
183
184 mod_timer(&dn_route_timer, now + decnet_dst_gc_interval * HZ);
185}
186
187static int dn_dst_gc(void)
188{
189 struct dn_route *rt, **rtp;
190 int i;
191 unsigned long now = jiffies;
192 unsigned long expire = 10 * HZ;
193
194 for(i = 0; i <= dn_rt_hash_mask; i++) {
195
196 spin_lock_bh(&dn_rt_hash_table[i].lock);
197 rtp = &dn_rt_hash_table[i].chain;
198
199 while((rt=*rtp) != NULL) {
200 if (atomic_read(&rt->u.dst.__refcnt) ||
201 (now - rt->u.dst.lastuse) < expire) {
202 rtp = &rt->u.rt_next;
203 continue;
204 }
205 *rtp = rt->u.rt_next;
206 rt->u.rt_next = NULL;
207 dnrt_drop(rt);
208 break;
209 }
210 spin_unlock_bh(&dn_rt_hash_table[i].lock);
211 }
212
213 return 0;
214}
215
216/*
217 * The decnet standards don't impose a particular minimum mtu, what they
218 * do insist on is that the routing layer accepts a datagram of at least
219 * 230 bytes long. Here we have to subtract the routing header length from
220 * 230 to get the minimum acceptable mtu. If there is no neighbour, then we
221 * assume the worst and use a long header size.
222 *
223 * We update both the mtu and the advertised mss (i.e. the segment size we
224 * advertise to the other end).
225 */
226static void dn_dst_update_pmtu(struct dst_entry *dst, u32 mtu)
227{
228 u32 min_mtu = 230;
229 struct dn_dev *dn = dst->neighbour ?
230 (struct dn_dev *)dst->neighbour->dev->dn_ptr : NULL;
231
232 if (dn && dn->use_long == 0)
233 min_mtu -= 6;
234 else
235 min_mtu -= 21;
236
237 if (dst->metrics[RTAX_MTU-1] > mtu && mtu >= min_mtu) {
238 if (!(dst_metric_locked(dst, RTAX_MTU))) {
239 dst->metrics[RTAX_MTU-1] = mtu;
240 dst_set_expires(dst, dn_rt_mtu_expires);
241 }
242 if (!(dst_metric_locked(dst, RTAX_ADVMSS))) {
243 u32 mss = mtu - DN_MAX_NSP_DATA_HEADER;
244 if (dst->metrics[RTAX_ADVMSS-1] > mss)
245 dst->metrics[RTAX_ADVMSS-1] = mss;
246 }
247 }
248}
249
250/*
251 * When a route has been marked obsolete. (e.g. routing cache flush)
252 */
253static struct dst_entry *dn_dst_check(struct dst_entry *dst, __u32 cookie)
254{
255 return NULL;
256}
257
258static struct dst_entry *dn_dst_negative_advice(struct dst_entry *dst)
259{
260 dst_release(dst);
261 return NULL;
262}
263
264static void dn_dst_link_failure(struct sk_buff *skb)
265{
266 return;
267}
268
269static inline int compare_keys(struct flowi *fl1, struct flowi *fl2)
270{
271 return memcmp(&fl1->nl_u.dn_u, &fl2->nl_u.dn_u, sizeof(fl1->nl_u.dn_u)) == 0 &&
272 fl1->oif == fl2->oif &&
273 fl1->iif == fl2->iif;
274}
275
276static int dn_insert_route(struct dn_route *rt, unsigned hash, struct dn_route **rp)
277{
278 struct dn_route *rth, **rthp;
279 unsigned long now = jiffies;
280
281 rthp = &dn_rt_hash_table[hash].chain;
282
283 spin_lock_bh(&dn_rt_hash_table[hash].lock);
284 while((rth = *rthp) != NULL) {
285 if (compare_keys(&rth->fl, &rt->fl)) {
286 /* Put it first */
287 *rthp = rth->u.rt_next;
288 rcu_assign_pointer(rth->u.rt_next,
289 dn_rt_hash_table[hash].chain);
290 rcu_assign_pointer(dn_rt_hash_table[hash].chain, rth);
291
292 rth->u.dst.__use++;
293 dst_hold(&rth->u.dst);
294 rth->u.dst.lastuse = now;
295 spin_unlock_bh(&dn_rt_hash_table[hash].lock);
296
297 dnrt_drop(rt);
298 *rp = rth;
299 return 0;
300 }
301 rthp = &rth->u.rt_next;
302 }
303
304 rcu_assign_pointer(rt->u.rt_next, dn_rt_hash_table[hash].chain);
305 rcu_assign_pointer(dn_rt_hash_table[hash].chain, rt);
306
307 dst_hold(&rt->u.dst);
308 rt->u.dst.__use++;
309 rt->u.dst.lastuse = now;
310 spin_unlock_bh(&dn_rt_hash_table[hash].lock);
311 *rp = rt;
312 return 0;
313}
314
315void dn_run_flush(unsigned long dummy)
316{
317 int i;
318 struct dn_route *rt, *next;
319
320 for(i = 0; i < dn_rt_hash_mask; i++) {
321 spin_lock_bh(&dn_rt_hash_table[i].lock);
322
323 if ((rt = xchg(&dn_rt_hash_table[i].chain, NULL)) == NULL)
324 goto nothing_to_declare;
325
326 for(; rt; rt=next) {
327 next = rt->u.rt_next;
328 rt->u.rt_next = NULL;
329 dst_free((struct dst_entry *)rt);
330 }
331
332nothing_to_declare:
333 spin_unlock_bh(&dn_rt_hash_table[i].lock);
334 }
335}
336
337static DEFINE_SPINLOCK(dn_rt_flush_lock);
338
339void dn_rt_cache_flush(int delay)
340{
341 unsigned long now = jiffies;
342 int user_mode = !in_interrupt();
343
344 if (delay < 0)
345 delay = dn_rt_min_delay;
346
347 spin_lock_bh(&dn_rt_flush_lock);
348
349 if (del_timer(&dn_rt_flush_timer) && delay > 0 && dn_rt_deadline) {
350 long tmo = (long)(dn_rt_deadline - now);
351
352 if (user_mode && tmo < dn_rt_max_delay - dn_rt_min_delay)
353 tmo = 0;
354
355 if (delay > tmo)
356 delay = tmo;
357 }
358
359 if (delay <= 0) {
360 spin_unlock_bh(&dn_rt_flush_lock);
361 dn_run_flush(0);
362 return;
363 }
364
365 if (dn_rt_deadline == 0)
366 dn_rt_deadline = now + dn_rt_max_delay;
367
368 dn_rt_flush_timer.expires = now + delay;
369 add_timer(&dn_rt_flush_timer);
370 spin_unlock_bh(&dn_rt_flush_lock);
371}
372
373/**
374 * dn_return_short - Return a short packet to its sender
375 * @skb: The packet to return
376 *
377 */
378static int dn_return_short(struct sk_buff *skb)
379{
380 struct dn_skb_cb *cb;
381 unsigned char *ptr;
382 dn_address *src;
383 dn_address *dst;
384 dn_address tmp;
385
386 /* Add back headers */
387 skb_push(skb, skb->data - skb->nh.raw);
388
389 if ((skb = skb_unshare(skb, GFP_ATOMIC)) == NULL)
390 return NET_RX_DROP;
391
392 cb = DN_SKB_CB(skb);
393 /* Skip packet length and point to flags */
394 ptr = skb->data + 2;
395 *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
396
397 dst = (dn_address *)ptr;
398 ptr += 2;
399 src = (dn_address *)ptr;
400 ptr += 2;
401 *ptr = 0; /* Zero hop count */
402
403 /* Swap source and destination */
404 tmp = *src;
405 *src = *dst;
406 *dst = tmp;
407
408 skb->pkt_type = PACKET_OUTGOING;
409 dn_rt_finish_output(skb, NULL, NULL);
410 return NET_RX_SUCCESS;
411}
412
413/**
414 * dn_return_long - Return a long packet to its sender
415 * @skb: The long format packet to return
416 *
417 */
418static int dn_return_long(struct sk_buff *skb)
419{
420 struct dn_skb_cb *cb;
421 unsigned char *ptr;
422 unsigned char *src_addr, *dst_addr;
423 unsigned char tmp[ETH_ALEN];
424
425 /* Add back all headers */
426 skb_push(skb, skb->data - skb->nh.raw);
427
428 if ((skb = skb_unshare(skb, GFP_ATOMIC)) == NULL)
429 return NET_RX_DROP;
430
431 cb = DN_SKB_CB(skb);
432 /* Ignore packet length and point to flags */
433 ptr = skb->data + 2;
434
435 /* Skip padding */
436 if (*ptr & DN_RT_F_PF) {
437 char padlen = (*ptr & ~DN_RT_F_PF);
438 ptr += padlen;
439 }
440
441 *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
442 ptr += 2;
443 dst_addr = ptr;
444 ptr += 8;
445 src_addr = ptr;
446 ptr += 6;
447 *ptr = 0; /* Zero hop count */
448
449 /* Swap source and destination */
450 memcpy(tmp, src_addr, ETH_ALEN);
451 memcpy(src_addr, dst_addr, ETH_ALEN);
452 memcpy(dst_addr, tmp, ETH_ALEN);
453
454 skb->pkt_type = PACKET_OUTGOING;
455 dn_rt_finish_output(skb, dst_addr, src_addr);
456 return NET_RX_SUCCESS;
457}
458
459/**
460 * dn_route_rx_packet - Try and find a route for an incoming packet
461 * @skb: The packet to find a route for
462 *
463 * Returns: result of input function if route is found, error code otherwise
464 */
465static int dn_route_rx_packet(struct sk_buff *skb)
466{
467 struct dn_skb_cb *cb = DN_SKB_CB(skb);
468 int err;
469
470 if ((err = dn_route_input(skb)) == 0)
471 return dst_input(skb);
472
473 if (decnet_debug_level & 4) {
474 char *devname = skb->dev ? skb->dev->name : "???";
475 struct dn_skb_cb *cb = DN_SKB_CB(skb);
476 printk(KERN_DEBUG
477 "DECnet: dn_route_rx_packet: rt_flags=0x%02x dev=%s len=%d src=0x%04hx dst=0x%04hx err=%d type=%d\n",
478 (int)cb->rt_flags, devname, skb->len, cb->src, cb->dst,
479 err, skb->pkt_type);
480 }
481
482 if ((skb->pkt_type == PACKET_HOST) && (cb->rt_flags & DN_RT_F_RQR)) {
483 switch(cb->rt_flags & DN_RT_PKT_MSK) {
484 case DN_RT_PKT_SHORT:
485 return dn_return_short(skb);
486 case DN_RT_PKT_LONG:
487 return dn_return_long(skb);
488 }
489 }
490
491 kfree_skb(skb);
492 return NET_RX_DROP;
493}
494
495static int dn_route_rx_long(struct sk_buff *skb)
496{
497 struct dn_skb_cb *cb = DN_SKB_CB(skb);
498 unsigned char *ptr = skb->data;
499
500 if (!pskb_may_pull(skb, 21)) /* 20 for long header, 1 for shortest nsp */
501 goto drop_it;
502
503 skb_pull(skb, 20);
504 skb->h.raw = skb->data;
505
506 /* Destination info */
507 ptr += 2;
508 cb->dst = dn_htons(dn_eth2dn(ptr));
509 if (memcmp(ptr, dn_hiord_addr, 4) != 0)
510 goto drop_it;
511 ptr += 6;
512
513
514 /* Source info */
515 ptr += 2;
516 cb->src = dn_htons(dn_eth2dn(ptr));
517 if (memcmp(ptr, dn_hiord_addr, 4) != 0)
518 goto drop_it;
519 ptr += 6;
520 /* Other junk */
521 ptr++;
522 cb->hops = *ptr++; /* Visit Count */
523
524 return NF_HOOK(PF_DECnet, NF_DN_PRE_ROUTING, skb, skb->dev, NULL, dn_route_rx_packet);
525
526drop_it:
527 kfree_skb(skb);
528 return NET_RX_DROP;
529}
530
531
532
533static int dn_route_rx_short(struct sk_buff *skb)
534{
535 struct dn_skb_cb *cb = DN_SKB_CB(skb);
536 unsigned char *ptr = skb->data;
537
538 if (!pskb_may_pull(skb, 6)) /* 5 for short header + 1 for shortest nsp */
539 goto drop_it;
540
541 skb_pull(skb, 5);
542 skb->h.raw = skb->data;
543
544 cb->dst = *(dn_address *)ptr;
545 ptr += 2;
546 cb->src = *(dn_address *)ptr;
547 ptr += 2;
548 cb->hops = *ptr & 0x3f;
549
550 return NF_HOOK(PF_DECnet, NF_DN_PRE_ROUTING, skb, skb->dev, NULL, dn_route_rx_packet);
551
552drop_it:
553 kfree_skb(skb);
554 return NET_RX_DROP;
555}
556
557static int dn_route_discard(struct sk_buff *skb)
558{
559 /*
560 * I know we drop the packet here, but thats considered success in
561 * this case
562 */
563 kfree_skb(skb);
564 return NET_RX_SUCCESS;
565}
566
567static int dn_route_ptp_hello(struct sk_buff *skb)
568{
569 dn_dev_hello(skb);
570 dn_neigh_pointopoint_hello(skb);
571 return NET_RX_SUCCESS;
572}
573
David S. Millerf2ccd8f2005-08-09 19:34:12 -0700574int dn_route_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575{
576 struct dn_skb_cb *cb;
577 unsigned char flags = 0;
578 __u16 len = dn_ntohs(*(__u16 *)skb->data);
579 struct dn_dev *dn = (struct dn_dev *)dev->dn_ptr;
580 unsigned char padlen = 0;
581
582 if (dn == NULL)
583 goto dump_it;
584
585 if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
586 goto out;
587
588 if (!pskb_may_pull(skb, 3))
589 goto dump_it;
590
591 skb_pull(skb, 2);
592
593 if (len > skb->len)
594 goto dump_it;
595
596 skb_trim(skb, len);
597
598 flags = *skb->data;
599
600 cb = DN_SKB_CB(skb);
601 cb->stamp = jiffies;
602 cb->iif = dev->ifindex;
603
604 /*
605 * If we have padding, remove it.
606 */
607 if (flags & DN_RT_F_PF) {
608 padlen = flags & ~DN_RT_F_PF;
609 if (!pskb_may_pull(skb, padlen + 1))
610 goto dump_it;
611 skb_pull(skb, padlen);
612 flags = *skb->data;
613 }
614
615 skb->nh.raw = skb->data;
616
617 /*
618 * Weed out future version DECnet
619 */
620 if (flags & DN_RT_F_VER)
621 goto dump_it;
622
623 cb->rt_flags = flags;
624
625 if (decnet_debug_level & 1)
626 printk(KERN_DEBUG
627 "dn_route_rcv: got 0x%02x from %s [%d %d %d]\n",
628 (int)flags, (dev) ? dev->name : "???", len, skb->len,
629 padlen);
630
631 if (flags & DN_RT_PKT_CNTL) {
632 if (unlikely(skb_is_nonlinear(skb)) &&
633 skb_linearize(skb, GFP_ATOMIC) != 0)
634 goto dump_it;
635
636 switch(flags & DN_RT_CNTL_MSK) {
637 case DN_RT_PKT_INIT:
638 dn_dev_init_pkt(skb);
639 break;
640 case DN_RT_PKT_VERI:
641 dn_dev_veri_pkt(skb);
642 break;
643 }
644
645 if (dn->parms.state != DN_DEV_S_RU)
646 goto dump_it;
647
648 switch(flags & DN_RT_CNTL_MSK) {
649 case DN_RT_PKT_HELO:
650 return NF_HOOK(PF_DECnet, NF_DN_HELLO, skb, skb->dev, NULL, dn_route_ptp_hello);
651
652 case DN_RT_PKT_L1RT:
653 case DN_RT_PKT_L2RT:
654 return NF_HOOK(PF_DECnet, NF_DN_ROUTE, skb, skb->dev, NULL, dn_route_discard);
655 case DN_RT_PKT_ERTH:
656 return NF_HOOK(PF_DECnet, NF_DN_HELLO, skb, skb->dev, NULL, dn_neigh_router_hello);
657
658 case DN_RT_PKT_EEDH:
659 return NF_HOOK(PF_DECnet, NF_DN_HELLO, skb, skb->dev, NULL, dn_neigh_endnode_hello);
660 }
661 } else {
662 if (dn->parms.state != DN_DEV_S_RU)
663 goto dump_it;
664
665 skb_pull(skb, 1); /* Pull flags */
666
667 switch(flags & DN_RT_PKT_MSK) {
668 case DN_RT_PKT_LONG:
669 return dn_route_rx_long(skb);
670 case DN_RT_PKT_SHORT:
671 return dn_route_rx_short(skb);
672 }
673 }
674
675dump_it:
676 kfree_skb(skb);
677out:
678 return NET_RX_DROP;
679}
680
681static int dn_output(struct sk_buff *skb)
682{
683 struct dst_entry *dst = skb->dst;
684 struct dn_route *rt = (struct dn_route *)dst;
685 struct net_device *dev = dst->dev;
686 struct dn_skb_cb *cb = DN_SKB_CB(skb);
687 struct neighbour *neigh;
688
689 int err = -EINVAL;
690
691 if ((neigh = dst->neighbour) == NULL)
692 goto error;
693
694 skb->dev = dev;
695
696 cb->src = rt->rt_saddr;
697 cb->dst = rt->rt_daddr;
698
699 /*
700 * Always set the Intra-Ethernet bit on all outgoing packets
701 * originated on this node. Only valid flag from upper layers
702 * is return-to-sender-requested. Set hop count to 0 too.
703 */
704 cb->rt_flags &= ~DN_RT_F_RQR;
705 cb->rt_flags |= DN_RT_F_IE;
706 cb->hops = 0;
707
708 return NF_HOOK(PF_DECnet, NF_DN_LOCAL_OUT, skb, NULL, dev, neigh->output);
709
710error:
711 if (net_ratelimit())
712 printk(KERN_DEBUG "dn_output: This should not happen\n");
713
714 kfree_skb(skb);
715
716 return err;
717}
718
719static int dn_forward(struct sk_buff *skb)
720{
721 struct dn_skb_cb *cb = DN_SKB_CB(skb);
722 struct dst_entry *dst = skb->dst;
723 struct dn_dev *dn_db = dst->dev->dn_ptr;
724 struct dn_route *rt;
725 struct neighbour *neigh = dst->neighbour;
726 int header_len;
727#ifdef CONFIG_NETFILTER
728 struct net_device *dev = skb->dev;
729#endif
730
731 if (skb->pkt_type != PACKET_HOST)
732 goto drop;
733
734 /* Ensure that we have enough space for headers */
735 rt = (struct dn_route *)skb->dst;
736 header_len = dn_db->use_long ? 21 : 6;
737 if (skb_cow(skb, LL_RESERVED_SPACE(rt->u.dst.dev)+header_len))
738 goto drop;
739
740 /*
741 * Hop count exceeded.
742 */
743 if (++cb->hops > 30)
744 goto drop;
745
746 skb->dev = rt->u.dst.dev;
747
748 /*
749 * If packet goes out same interface it came in on, then set
750 * the Intra-Ethernet bit. This has no effect for short
751 * packets, so we don't need to test for them here.
752 */
753 cb->rt_flags &= ~DN_RT_F_IE;
754 if (rt->rt_flags & RTCF_DOREDIRECT)
755 cb->rt_flags |= DN_RT_F_IE;
756
757 return NF_HOOK(PF_DECnet, NF_DN_FORWARD, skb, dev, skb->dev, neigh->output);
758
759drop:
760 kfree_skb(skb);
761 return NET_RX_DROP;
762}
763
764/*
765 * Drop packet. This is used for endnodes and for
766 * when we should not be forwarding packets from
767 * this dest.
768 */
769static int dn_blackhole(struct sk_buff *skb)
770{
771 kfree_skb(skb);
772 return NET_RX_DROP;
773}
774
775/*
776 * Used to catch bugs. This should never normally get
777 * called.
778 */
779static int dn_rt_bug(struct sk_buff *skb)
780{
781 if (net_ratelimit()) {
782 struct dn_skb_cb *cb = DN_SKB_CB(skb);
783
784 printk(KERN_DEBUG "dn_rt_bug: skb from:%04x to:%04x\n",
785 cb->src, cb->dst);
786 }
787
788 kfree_skb(skb);
789
790 return NET_RX_BAD;
791}
792
793static int dn_rt_set_next_hop(struct dn_route *rt, struct dn_fib_res *res)
794{
795 struct dn_fib_info *fi = res->fi;
796 struct net_device *dev = rt->u.dst.dev;
797 struct neighbour *n;
798 unsigned mss;
799
800 if (fi) {
801 if (DN_FIB_RES_GW(*res) &&
802 DN_FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
803 rt->rt_gateway = DN_FIB_RES_GW(*res);
804 memcpy(rt->u.dst.metrics, fi->fib_metrics,
805 sizeof(rt->u.dst.metrics));
806 }
807 rt->rt_type = res->type;
808
809 if (dev != NULL && rt->u.dst.neighbour == NULL) {
810 n = __neigh_lookup_errno(&dn_neigh_table, &rt->rt_gateway, dev);
811 if (IS_ERR(n))
812 return PTR_ERR(n);
813 rt->u.dst.neighbour = n;
814 }
815
816 if (rt->u.dst.metrics[RTAX_MTU-1] == 0 ||
817 rt->u.dst.metrics[RTAX_MTU-1] > rt->u.dst.dev->mtu)
818 rt->u.dst.metrics[RTAX_MTU-1] = rt->u.dst.dev->mtu;
819 mss = dn_mss_from_pmtu(dev, dst_mtu(&rt->u.dst));
820 if (rt->u.dst.metrics[RTAX_ADVMSS-1] == 0 ||
821 rt->u.dst.metrics[RTAX_ADVMSS-1] > mss)
822 rt->u.dst.metrics[RTAX_ADVMSS-1] = mss;
823 return 0;
824}
825
826static inline int dn_match_addr(__u16 addr1, __u16 addr2)
827{
828 __u16 tmp = dn_ntohs(addr1) ^ dn_ntohs(addr2);
829 int match = 16;
830 while(tmp) {
831 tmp >>= 1;
832 match--;
833 }
834 return match;
835}
836
837static __u16 dnet_select_source(const struct net_device *dev, __u16 daddr, int scope)
838{
839 __u16 saddr = 0;
840 struct dn_dev *dn_db = dev->dn_ptr;
841 struct dn_ifaddr *ifa;
842 int best_match = 0;
843 int ret;
844
845 read_lock(&dev_base_lock);
846 for(ifa = dn_db->ifa_list; ifa; ifa = ifa->ifa_next) {
847 if (ifa->ifa_scope > scope)
848 continue;
849 if (!daddr) {
850 saddr = ifa->ifa_local;
851 break;
852 }
853 ret = dn_match_addr(daddr, ifa->ifa_local);
854 if (ret > best_match)
855 saddr = ifa->ifa_local;
856 if (best_match == 0)
857 saddr = ifa->ifa_local;
858 }
859 read_unlock(&dev_base_lock);
860
861 return saddr;
862}
863
864static inline __u16 __dn_fib_res_prefsrc(struct dn_fib_res *res)
865{
866 return dnet_select_source(DN_FIB_RES_DEV(*res), DN_FIB_RES_GW(*res), res->scope);
867}
868
869static inline __u16 dn_fib_rules_map_destination(__u16 daddr, struct dn_fib_res *res)
870{
871 __u16 mask = dnet_make_mask(res->prefixlen);
872 return (daddr&~mask)|res->fi->fib_nh->nh_gw;
873}
874
875static int dn_route_output_slow(struct dst_entry **pprt, const struct flowi *oldflp, int try_hard)
876{
877 struct flowi fl = { .nl_u = { .dn_u =
878 { .daddr = oldflp->fld_dst,
879 .saddr = oldflp->fld_src,
880 .scope = RT_SCOPE_UNIVERSE,
881#ifdef CONFIG_DECNET_ROUTE_FWMARK
882 .fwmark = oldflp->fld_fwmark
883#endif
884 } },
885 .iif = loopback_dev.ifindex,
886 .oif = oldflp->oif };
887 struct dn_route *rt = NULL;
888 struct net_device *dev_out = NULL;
889 struct neighbour *neigh = NULL;
890 unsigned hash;
891 unsigned flags = 0;
892 struct dn_fib_res res = { .fi = NULL, .type = RTN_UNICAST };
893 int err;
894 int free_res = 0;
895 __u16 gateway = 0;
896
897 if (decnet_debug_level & 16)
898 printk(KERN_DEBUG
899 "dn_route_output_slow: dst=%04x src=%04x mark=%d"
900 " iif=%d oif=%d\n", oldflp->fld_dst, oldflp->fld_src,
901 oldflp->fld_fwmark, loopback_dev.ifindex, oldflp->oif);
902
903 /* If we have an output interface, verify its a DECnet device */
904 if (oldflp->oif) {
905 dev_out = dev_get_by_index(oldflp->oif);
906 err = -ENODEV;
907 if (dev_out && dev_out->dn_ptr == NULL) {
908 dev_put(dev_out);
909 dev_out = NULL;
910 }
911 if (dev_out == NULL)
912 goto out;
913 }
914
915 /* If we have a source address, verify that its a local address */
916 if (oldflp->fld_src) {
917 err = -EADDRNOTAVAIL;
918
919 if (dev_out) {
920 if (dn_dev_islocal(dev_out, oldflp->fld_src))
921 goto source_ok;
922 dev_put(dev_out);
923 goto out;
924 }
925 read_lock(&dev_base_lock);
926 for(dev_out = dev_base; dev_out; dev_out = dev_out->next) {
927 if (!dev_out->dn_ptr)
928 continue;
929 if (dn_dev_islocal(dev_out, oldflp->fld_src))
930 break;
931 }
932 read_unlock(&dev_base_lock);
933 if (dev_out == NULL)
934 goto out;
935 dev_hold(dev_out);
936source_ok:
937 ;
938 }
939
940 /* No destination? Assume its local */
941 if (!fl.fld_dst) {
942 fl.fld_dst = fl.fld_src;
943
944 err = -EADDRNOTAVAIL;
945 if (dev_out)
946 dev_put(dev_out);
947 dev_out = &loopback_dev;
948 dev_hold(dev_out);
949 if (!fl.fld_dst) {
950 fl.fld_dst =
951 fl.fld_src = dnet_select_source(dev_out, 0,
952 RT_SCOPE_HOST);
953 if (!fl.fld_dst)
954 goto out;
955 }
956 fl.oif = loopback_dev.ifindex;
957 res.type = RTN_LOCAL;
958 goto make_route;
959 }
960
961 if (decnet_debug_level & 16)
962 printk(KERN_DEBUG
963 "dn_route_output_slow: initial checks complete."
964 " dst=%o4x src=%04x oif=%d try_hard=%d\n", fl.fld_dst,
965 fl.fld_src, fl.oif, try_hard);
966
967 /*
968 * N.B. If the kernel is compiled without router support then
969 * dn_fib_lookup() will evaluate to non-zero so this if () block
970 * will always be executed.
971 */
972 err = -ESRCH;
973 if (try_hard || (err = dn_fib_lookup(&fl, &res)) != 0) {
974 struct dn_dev *dn_db;
975 if (err != -ESRCH)
976 goto out;
977 /*
978 * Here the fallback is basically the standard algorithm for
979 * routing in endnodes which is described in the DECnet routing
980 * docs
981 *
982 * If we are not trying hard, look in neighbour cache.
983 * The result is tested to ensure that if a specific output
984 * device/source address was requested, then we honour that
985 * here
986 */
987 if (!try_hard) {
988 neigh = neigh_lookup_nodev(&dn_neigh_table, &fl.fld_dst);
989 if (neigh) {
990 if ((oldflp->oif &&
991 (neigh->dev->ifindex != oldflp->oif)) ||
992 (oldflp->fld_src &&
993 (!dn_dev_islocal(neigh->dev,
994 oldflp->fld_src)))) {
995 neigh_release(neigh);
996 neigh = NULL;
997 } else {
998 if (dev_out)
999 dev_put(dev_out);
1000 if (dn_dev_islocal(neigh->dev, fl.fld_dst)) {
1001 dev_out = &loopback_dev;
1002 res.type = RTN_LOCAL;
1003 } else {
1004 dev_out = neigh->dev;
1005 }
1006 dev_hold(dev_out);
1007 goto select_source;
1008 }
1009 }
1010 }
1011
1012 /* Not there? Perhaps its a local address */
1013 if (dev_out == NULL)
1014 dev_out = dn_dev_get_default();
1015 err = -ENODEV;
1016 if (dev_out == NULL)
1017 goto out;
1018 dn_db = dev_out->dn_ptr;
1019 /* Possible improvement - check all devices for local addr */
1020 if (dn_dev_islocal(dev_out, fl.fld_dst)) {
1021 dev_put(dev_out);
1022 dev_out = &loopback_dev;
1023 dev_hold(dev_out);
1024 res.type = RTN_LOCAL;
1025 goto select_source;
1026 }
1027 /* Not local either.... try sending it to the default router */
1028 neigh = neigh_clone(dn_db->router);
1029 BUG_ON(neigh && neigh->dev != dev_out);
1030
1031 /* Ok then, we assume its directly connected and move on */
1032select_source:
1033 if (neigh)
1034 gateway = ((struct dn_neigh *)neigh)->addr;
1035 if (gateway == 0)
1036 gateway = fl.fld_dst;
1037 if (fl.fld_src == 0) {
1038 fl.fld_src = dnet_select_source(dev_out, gateway,
1039 res.type == RTN_LOCAL ?
1040 RT_SCOPE_HOST :
1041 RT_SCOPE_LINK);
1042 if (fl.fld_src == 0 && res.type != RTN_LOCAL)
1043 goto e_addr;
1044 }
1045 fl.oif = dev_out->ifindex;
1046 goto make_route;
1047 }
1048 free_res = 1;
1049
1050 if (res.type == RTN_NAT)
1051 goto e_inval;
1052
1053 if (res.type == RTN_LOCAL) {
1054 if (!fl.fld_src)
1055 fl.fld_src = fl.fld_dst;
1056 if (dev_out)
1057 dev_put(dev_out);
1058 dev_out = &loopback_dev;
1059 dev_hold(dev_out);
1060 fl.oif = dev_out->ifindex;
1061 if (res.fi)
1062 dn_fib_info_put(res.fi);
1063 res.fi = NULL;
1064 goto make_route;
1065 }
1066
1067 if (res.fi->fib_nhs > 1 && fl.oif == 0)
1068 dn_fib_select_multipath(&fl, &res);
1069
1070 /*
1071 * We could add some logic to deal with default routes here and
1072 * get rid of some of the special casing above.
1073 */
1074
1075 if (!fl.fld_src)
1076 fl.fld_src = DN_FIB_RES_PREFSRC(res);
1077
1078 if (dev_out)
1079 dev_put(dev_out);
1080 dev_out = DN_FIB_RES_DEV(res);
1081 dev_hold(dev_out);
1082 fl.oif = dev_out->ifindex;
1083 gateway = DN_FIB_RES_GW(res);
1084
1085make_route:
1086 if (dev_out->flags & IFF_LOOPBACK)
1087 flags |= RTCF_LOCAL;
1088
1089 rt = dst_alloc(&dn_dst_ops);
1090 if (rt == NULL)
1091 goto e_nobufs;
1092
1093 atomic_set(&rt->u.dst.__refcnt, 1);
1094 rt->u.dst.flags = DST_HOST;
1095
1096 rt->fl.fld_src = oldflp->fld_src;
1097 rt->fl.fld_dst = oldflp->fld_dst;
1098 rt->fl.oif = oldflp->oif;
1099 rt->fl.iif = 0;
1100#ifdef CONFIG_DECNET_ROUTE_FWMARK
1101 rt->fl.fld_fwmark = oldflp->fld_fwmark;
1102#endif
1103
1104 rt->rt_saddr = fl.fld_src;
1105 rt->rt_daddr = fl.fld_dst;
1106 rt->rt_gateway = gateway ? gateway : fl.fld_dst;
1107 rt->rt_local_src = fl.fld_src;
1108
1109 rt->rt_dst_map = fl.fld_dst;
1110 rt->rt_src_map = fl.fld_src;
1111
1112 rt->u.dst.dev = dev_out;
1113 dev_hold(dev_out);
1114 rt->u.dst.neighbour = neigh;
1115 neigh = NULL;
1116
1117 rt->u.dst.lastuse = jiffies;
1118 rt->u.dst.output = dn_output;
1119 rt->u.dst.input = dn_rt_bug;
1120 rt->rt_flags = flags;
1121 if (flags & RTCF_LOCAL)
1122 rt->u.dst.input = dn_nsp_rx;
1123
1124 err = dn_rt_set_next_hop(rt, &res);
1125 if (err)
1126 goto e_neighbour;
1127
1128 hash = dn_hash(rt->fl.fld_src, rt->fl.fld_dst);
1129 dn_insert_route(rt, hash, (struct dn_route **)pprt);
1130
1131done:
1132 if (neigh)
1133 neigh_release(neigh);
1134 if (free_res)
1135 dn_fib_res_put(&res);
1136 if (dev_out)
1137 dev_put(dev_out);
1138out:
1139 return err;
1140
1141e_addr:
1142 err = -EADDRNOTAVAIL;
1143 goto done;
1144e_inval:
1145 err = -EINVAL;
1146 goto done;
1147e_nobufs:
1148 err = -ENOBUFS;
1149 goto done;
1150e_neighbour:
1151 dst_free(&rt->u.dst);
1152 goto e_nobufs;
1153}
1154
1155
1156/*
1157 * N.B. The flags may be moved into the flowi at some future stage.
1158 */
1159static int __dn_route_output_key(struct dst_entry **pprt, const struct flowi *flp, int flags)
1160{
1161 unsigned hash = dn_hash(flp->fld_src, flp->fld_dst);
1162 struct dn_route *rt = NULL;
1163
1164 if (!(flags & MSG_TRYHARD)) {
1165 rcu_read_lock_bh();
1166 for(rt = rcu_dereference(dn_rt_hash_table[hash].chain); rt;
1167 rt = rcu_dereference(rt->u.rt_next)) {
1168 if ((flp->fld_dst == rt->fl.fld_dst) &&
1169 (flp->fld_src == rt->fl.fld_src) &&
1170#ifdef CONFIG_DECNET_ROUTE_FWMARK
1171 (flp->fld_fwmark == rt->fl.fld_fwmark) &&
1172#endif
1173 (rt->fl.iif == 0) &&
1174 (rt->fl.oif == flp->oif)) {
1175 rt->u.dst.lastuse = jiffies;
1176 dst_hold(&rt->u.dst);
1177 rt->u.dst.__use++;
1178 rcu_read_unlock_bh();
1179 *pprt = &rt->u.dst;
1180 return 0;
1181 }
1182 }
1183 rcu_read_unlock_bh();
1184 }
1185
1186 return dn_route_output_slow(pprt, flp, flags);
1187}
1188
1189static int dn_route_output_key(struct dst_entry **pprt, struct flowi *flp, int flags)
1190{
1191 int err;
1192
1193 err = __dn_route_output_key(pprt, flp, flags);
1194 if (err == 0 && flp->proto) {
1195 err = xfrm_lookup(pprt, flp, NULL, 0);
1196 }
1197 return err;
1198}
1199
1200int dn_route_output_sock(struct dst_entry **pprt, struct flowi *fl, struct sock *sk, int flags)
1201{
1202 int err;
1203
1204 err = __dn_route_output_key(pprt, fl, flags & MSG_TRYHARD);
1205 if (err == 0 && fl->proto) {
1206 err = xfrm_lookup(pprt, fl, sk, !(flags & MSG_DONTWAIT));
1207 }
1208 return err;
1209}
1210
1211static int dn_route_input_slow(struct sk_buff *skb)
1212{
1213 struct dn_route *rt = NULL;
1214 struct dn_skb_cb *cb = DN_SKB_CB(skb);
1215 struct net_device *in_dev = skb->dev;
1216 struct net_device *out_dev = NULL;
1217 struct dn_dev *dn_db;
1218 struct neighbour *neigh = NULL;
1219 unsigned hash;
1220 int flags = 0;
1221 __u16 gateway = 0;
1222 __u16 local_src = 0;
1223 struct flowi fl = { .nl_u = { .dn_u =
1224 { .daddr = cb->dst,
1225 .saddr = cb->src,
1226 .scope = RT_SCOPE_UNIVERSE,
1227#ifdef CONFIG_DECNET_ROUTE_FWMARK
1228 .fwmark = skb->nfmark
1229#endif
1230 } },
1231 .iif = skb->dev->ifindex };
1232 struct dn_fib_res res = { .fi = NULL, .type = RTN_UNREACHABLE };
1233 int err = -EINVAL;
1234 int free_res = 0;
1235
1236 dev_hold(in_dev);
1237
1238 if ((dn_db = in_dev->dn_ptr) == NULL)
1239 goto out;
1240
1241 /* Zero source addresses are not allowed */
1242 if (fl.fld_src == 0)
1243 goto out;
1244
1245 /*
1246 * In this case we've just received a packet from a source
1247 * outside ourselves pretending to come from us. We don't
1248 * allow it any further to prevent routing loops, spoofing and
1249 * other nasties. Loopback packets already have the dst attached
1250 * so this only affects packets which have originated elsewhere.
1251 */
1252 err = -ENOTUNIQ;
1253 if (dn_dev_islocal(in_dev, cb->src))
1254 goto out;
1255
1256 err = dn_fib_lookup(&fl, &res);
1257 if (err) {
1258 if (err != -ESRCH)
1259 goto out;
1260 /*
1261 * Is the destination us ?
1262 */
1263 if (!dn_dev_islocal(in_dev, cb->dst))
1264 goto e_inval;
1265
1266 res.type = RTN_LOCAL;
1267 flags |= RTCF_DIRECTSRC;
1268 } else {
1269 __u16 src_map = fl.fld_src;
1270 free_res = 1;
1271
1272 out_dev = DN_FIB_RES_DEV(res);
1273 if (out_dev == NULL) {
1274 if (net_ratelimit())
1275 printk(KERN_CRIT "Bug in dn_route_input_slow() "
1276 "No output device\n");
1277 goto e_inval;
1278 }
1279 dev_hold(out_dev);
1280
1281 if (res.r)
1282 src_map = dn_fib_rules_policy(fl.fld_src, &res, &flags);
1283
1284 gateway = DN_FIB_RES_GW(res);
1285 if (res.type == RTN_NAT) {
1286 fl.fld_dst = dn_fib_rules_map_destination(fl.fld_dst, &res);
1287 dn_fib_res_put(&res);
1288 free_res = 0;
1289 if (dn_fib_lookup(&fl, &res))
1290 goto e_inval;
1291 free_res = 1;
1292 if (res.type != RTN_UNICAST)
1293 goto e_inval;
1294 flags |= RTCF_DNAT;
1295 gateway = fl.fld_dst;
1296 }
1297 fl.fld_src = src_map;
1298 }
1299
1300 switch(res.type) {
1301 case RTN_UNICAST:
1302 /*
1303 * Forwarding check here, we only check for forwarding
1304 * being turned off, if you want to only forward intra
1305 * area, its up to you to set the routing tables up
1306 * correctly.
1307 */
1308 if (dn_db->parms.forwarding == 0)
1309 goto e_inval;
1310
1311 if (res.fi->fib_nhs > 1 && fl.oif == 0)
1312 dn_fib_select_multipath(&fl, &res);
1313
1314 /*
1315 * Check for out_dev == in_dev. We use the RTCF_DOREDIRECT
1316 * flag as a hint to set the intra-ethernet bit when
1317 * forwarding. If we've got NAT in operation, we don't do
1318 * this optimisation.
1319 */
1320 if (out_dev == in_dev && !(flags & RTCF_NAT))
1321 flags |= RTCF_DOREDIRECT;
1322
1323 local_src = DN_FIB_RES_PREFSRC(res);
1324
1325 case RTN_BLACKHOLE:
1326 case RTN_UNREACHABLE:
1327 break;
1328 case RTN_LOCAL:
1329 flags |= RTCF_LOCAL;
1330 fl.fld_src = cb->dst;
1331 fl.fld_dst = cb->src;
1332
1333 /* Routing tables gave us a gateway */
1334 if (gateway)
1335 goto make_route;
1336
1337 /* Packet was intra-ethernet, so we know its on-link */
1338 if (cb->rt_flags | DN_RT_F_IE) {
1339 gateway = cb->src;
1340 flags |= RTCF_DIRECTSRC;
1341 goto make_route;
1342 }
1343
1344 /* Use the default router if there is one */
1345 neigh = neigh_clone(dn_db->router);
1346 if (neigh) {
1347 gateway = ((struct dn_neigh *)neigh)->addr;
1348 goto make_route;
1349 }
1350
1351 /* Close eyes and pray */
1352 gateway = cb->src;
1353 flags |= RTCF_DIRECTSRC;
1354 goto make_route;
1355 default:
1356 goto e_inval;
1357 }
1358
1359make_route:
1360 rt = dst_alloc(&dn_dst_ops);
1361 if (rt == NULL)
1362 goto e_nobufs;
1363
1364 rt->rt_saddr = fl.fld_src;
1365 rt->rt_daddr = fl.fld_dst;
1366 rt->rt_gateway = fl.fld_dst;
1367 if (gateway)
1368 rt->rt_gateway = gateway;
1369 rt->rt_local_src = local_src ? local_src : rt->rt_saddr;
1370
1371 rt->rt_dst_map = fl.fld_dst;
1372 rt->rt_src_map = fl.fld_src;
1373
1374 rt->fl.fld_src = cb->src;
1375 rt->fl.fld_dst = cb->dst;
1376 rt->fl.oif = 0;
1377 rt->fl.iif = in_dev->ifindex;
1378 rt->fl.fld_fwmark = fl.fld_fwmark;
1379
1380 rt->u.dst.flags = DST_HOST;
1381 rt->u.dst.neighbour = neigh;
1382 rt->u.dst.dev = out_dev;
1383 rt->u.dst.lastuse = jiffies;
1384 rt->u.dst.output = dn_rt_bug;
1385 switch(res.type) {
1386 case RTN_UNICAST:
1387 rt->u.dst.input = dn_forward;
1388 break;
1389 case RTN_LOCAL:
1390 rt->u.dst.output = dn_output;
1391 rt->u.dst.input = dn_nsp_rx;
1392 rt->u.dst.dev = in_dev;
1393 flags |= RTCF_LOCAL;
1394 break;
1395 default:
1396 case RTN_UNREACHABLE:
1397 case RTN_BLACKHOLE:
1398 rt->u.dst.input = dn_blackhole;
1399 }
1400 rt->rt_flags = flags;
1401 if (rt->u.dst.dev)
1402 dev_hold(rt->u.dst.dev);
1403
1404 err = dn_rt_set_next_hop(rt, &res);
1405 if (err)
1406 goto e_neighbour;
1407
1408 hash = dn_hash(rt->fl.fld_src, rt->fl.fld_dst);
1409 dn_insert_route(rt, hash, (struct dn_route **)&skb->dst);
1410
1411done:
1412 if (neigh)
1413 neigh_release(neigh);
1414 if (free_res)
1415 dn_fib_res_put(&res);
1416 dev_put(in_dev);
1417 if (out_dev)
1418 dev_put(out_dev);
1419out:
1420 return err;
1421
1422e_inval:
1423 err = -EINVAL;
1424 goto done;
1425
1426e_nobufs:
1427 err = -ENOBUFS;
1428 goto done;
1429
1430e_neighbour:
1431 dst_free(&rt->u.dst);
1432 goto done;
1433}
1434
1435int dn_route_input(struct sk_buff *skb)
1436{
1437 struct dn_route *rt;
1438 struct dn_skb_cb *cb = DN_SKB_CB(skb);
1439 unsigned hash = dn_hash(cb->src, cb->dst);
1440
1441 if (skb->dst)
1442 return 0;
1443
1444 rcu_read_lock();
1445 for(rt = rcu_dereference(dn_rt_hash_table[hash].chain); rt != NULL;
1446 rt = rcu_dereference(rt->u.rt_next)) {
1447 if ((rt->fl.fld_src == cb->src) &&
1448 (rt->fl.fld_dst == cb->dst) &&
1449 (rt->fl.oif == 0) &&
1450#ifdef CONFIG_DECNET_ROUTE_FWMARK
1451 (rt->fl.fld_fwmark == skb->nfmark) &&
1452#endif
1453 (rt->fl.iif == cb->iif)) {
1454 rt->u.dst.lastuse = jiffies;
1455 dst_hold(&rt->u.dst);
1456 rt->u.dst.__use++;
1457 rcu_read_unlock();
1458 skb->dst = (struct dst_entry *)rt;
1459 return 0;
1460 }
1461 }
1462 rcu_read_unlock();
1463
1464 return dn_route_input_slow(skb);
1465}
1466
Jamal Hadi Salimb6544c02005-06-18 22:54:12 -07001467static int dn_rt_fill_info(struct sk_buff *skb, u32 pid, u32 seq,
1468 int event, int nowait, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469{
1470 struct dn_route *rt = (struct dn_route *)skb->dst;
1471 struct rtmsg *r;
1472 struct nlmsghdr *nlh;
1473 unsigned char *b = skb->tail;
1474 struct rta_cacheinfo ci;
1475
Jamal Hadi Salimb6544c02005-06-18 22:54:12 -07001476 nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*r), flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477 r = NLMSG_DATA(nlh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478 r->rtm_family = AF_DECnet;
1479 r->rtm_dst_len = 16;
1480 r->rtm_src_len = 0;
1481 r->rtm_tos = 0;
1482 r->rtm_table = RT_TABLE_MAIN;
1483 r->rtm_type = rt->rt_type;
1484 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
1485 r->rtm_scope = RT_SCOPE_UNIVERSE;
1486 r->rtm_protocol = RTPROT_UNSPEC;
1487 if (rt->rt_flags & RTCF_NOTIFY)
1488 r->rtm_flags |= RTM_F_NOTIFY;
1489 RTA_PUT(skb, RTA_DST, 2, &rt->rt_daddr);
1490 if (rt->fl.fld_src) {
1491 r->rtm_src_len = 16;
1492 RTA_PUT(skb, RTA_SRC, 2, &rt->fl.fld_src);
1493 }
1494 if (rt->u.dst.dev)
1495 RTA_PUT(skb, RTA_OIF, sizeof(int), &rt->u.dst.dev->ifindex);
1496 /*
1497 * Note to self - change this if input routes reverse direction when
1498 * they deal only with inputs and not with replies like they do
1499 * currently.
1500 */
1501 RTA_PUT(skb, RTA_PREFSRC, 2, &rt->rt_local_src);
1502 if (rt->rt_daddr != rt->rt_gateway)
1503 RTA_PUT(skb, RTA_GATEWAY, 2, &rt->rt_gateway);
1504 if (rtnetlink_put_metrics(skb, rt->u.dst.metrics) < 0)
1505 goto rtattr_failure;
1506 ci.rta_lastuse = jiffies_to_clock_t(jiffies - rt->u.dst.lastuse);
1507 ci.rta_used = rt->u.dst.__use;
1508 ci.rta_clntref = atomic_read(&rt->u.dst.__refcnt);
1509 if (rt->u.dst.expires)
1510 ci.rta_expires = jiffies_to_clock_t(rt->u.dst.expires - jiffies);
1511 else
1512 ci.rta_expires = 0;
1513 ci.rta_error = rt->u.dst.error;
1514 ci.rta_id = ci.rta_ts = ci.rta_tsage = 0;
1515 RTA_PUT(skb, RTA_CACHEINFO, sizeof(ci), &ci);
1516 if (rt->fl.iif)
1517 RTA_PUT(skb, RTA_IIF, sizeof(int), &rt->fl.iif);
1518
1519 nlh->nlmsg_len = skb->tail - b;
1520 return skb->len;
1521
1522nlmsg_failure:
1523rtattr_failure:
1524 skb_trim(skb, b - skb->data);
1525 return -1;
1526}
1527
1528/*
1529 * This is called by both endnodes and routers now.
1530 */
1531int dn_cache_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh, void *arg)
1532{
1533 struct rtattr **rta = arg;
1534 struct rtmsg *rtm = NLMSG_DATA(nlh);
1535 struct dn_route *rt = NULL;
1536 struct dn_skb_cb *cb;
1537 int err;
1538 struct sk_buff *skb;
1539 struct flowi fl;
1540
1541 memset(&fl, 0, sizeof(fl));
1542 fl.proto = DNPROTO_NSP;
1543
1544 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1545 if (skb == NULL)
1546 return -ENOBUFS;
1547 skb->mac.raw = skb->data;
1548 cb = DN_SKB_CB(skb);
1549
1550 if (rta[RTA_SRC-1])
1551 memcpy(&fl.fld_src, RTA_DATA(rta[RTA_SRC-1]), 2);
1552 if (rta[RTA_DST-1])
1553 memcpy(&fl.fld_dst, RTA_DATA(rta[RTA_DST-1]), 2);
1554 if (rta[RTA_IIF-1])
1555 memcpy(&fl.iif, RTA_DATA(rta[RTA_IIF-1]), sizeof(int));
1556
1557 if (fl.iif) {
1558 struct net_device *dev;
1559 if ((dev = dev_get_by_index(fl.iif)) == NULL) {
1560 kfree_skb(skb);
1561 return -ENODEV;
1562 }
1563 if (!dev->dn_ptr) {
1564 dev_put(dev);
1565 kfree_skb(skb);
1566 return -ENODEV;
1567 }
1568 skb->protocol = __constant_htons(ETH_P_DNA_RT);
1569 skb->dev = dev;
1570 cb->src = fl.fld_src;
1571 cb->dst = fl.fld_dst;
1572 local_bh_disable();
1573 err = dn_route_input(skb);
1574 local_bh_enable();
1575 memset(cb, 0, sizeof(struct dn_skb_cb));
1576 rt = (struct dn_route *)skb->dst;
1577 if (!err && -rt->u.dst.error)
1578 err = rt->u.dst.error;
1579 } else {
1580 int oif = 0;
1581 if (rta[RTA_OIF - 1])
1582 memcpy(&oif, RTA_DATA(rta[RTA_OIF - 1]), sizeof(int));
1583 fl.oif = oif;
1584 err = dn_route_output_key((struct dst_entry **)&rt, &fl, 0);
1585 }
1586
1587 if (skb->dev)
1588 dev_put(skb->dev);
1589 skb->dev = NULL;
1590 if (err)
1591 goto out_free;
1592 skb->dst = &rt->u.dst;
1593 if (rtm->rtm_flags & RTM_F_NOTIFY)
1594 rt->rt_flags |= RTCF_NOTIFY;
1595
1596 NETLINK_CB(skb).dst_pid = NETLINK_CB(in_skb).pid;
1597
Jamal Hadi Salimb6544c02005-06-18 22:54:12 -07001598 err = dn_rt_fill_info(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq, RTM_NEWROUTE, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599
1600 if (err == 0)
1601 goto out_free;
1602 if (err < 0) {
1603 err = -EMSGSIZE;
1604 goto out_free;
1605 }
1606
1607 err = netlink_unicast(rtnl, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1608
1609 return err;
1610
1611out_free:
1612 kfree_skb(skb);
1613 return err;
1614}
1615
1616/*
1617 * For routers, this is called from dn_fib_dump, but for endnodes its
1618 * called directly from the rtnetlink dispatch table.
1619 */
1620int dn_cache_dump(struct sk_buff *skb, struct netlink_callback *cb)
1621{
1622 struct dn_route *rt;
1623 int h, s_h;
1624 int idx, s_idx;
1625
1626 if (NLMSG_PAYLOAD(cb->nlh, 0) < sizeof(struct rtmsg))
1627 return -EINVAL;
1628 if (!(((struct rtmsg *)NLMSG_DATA(cb->nlh))->rtm_flags&RTM_F_CLONED))
1629 return 0;
1630
1631 s_h = cb->args[0];
1632 s_idx = idx = cb->args[1];
1633 for(h = 0; h <= dn_rt_hash_mask; h++) {
1634 if (h < s_h)
1635 continue;
1636 if (h > s_h)
1637 s_idx = 0;
1638 rcu_read_lock_bh();
1639 for(rt = rcu_dereference(dn_rt_hash_table[h].chain), idx = 0;
1640 rt;
1641 rt = rcu_dereference(rt->u.rt_next), idx++) {
1642 if (idx < s_idx)
1643 continue;
1644 skb->dst = dst_clone(&rt->u.dst);
1645 if (dn_rt_fill_info(skb, NETLINK_CB(cb->skb).pid,
Jamal Hadi Salimb6544c02005-06-18 22:54:12 -07001646 cb->nlh->nlmsg_seq, RTM_NEWROUTE,
1647 1, NLM_F_MULTI) <= 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648 dst_release(xchg(&skb->dst, NULL));
1649 rcu_read_unlock_bh();
1650 goto done;
1651 }
1652 dst_release(xchg(&skb->dst, NULL));
1653 }
1654 rcu_read_unlock_bh();
1655 }
1656
1657done:
1658 cb->args[0] = h;
1659 cb->args[1] = idx;
1660 return skb->len;
1661}
1662
1663#ifdef CONFIG_PROC_FS
1664struct dn_rt_cache_iter_state {
1665 int bucket;
1666};
1667
1668static struct dn_route *dn_rt_cache_get_first(struct seq_file *seq)
1669{
1670 struct dn_route *rt = NULL;
1671 struct dn_rt_cache_iter_state *s = seq->private;
1672
1673 for(s->bucket = dn_rt_hash_mask; s->bucket >= 0; --s->bucket) {
1674 rcu_read_lock_bh();
1675 rt = dn_rt_hash_table[s->bucket].chain;
1676 if (rt)
1677 break;
1678 rcu_read_unlock_bh();
1679 }
1680 return rt;
1681}
1682
1683static struct dn_route *dn_rt_cache_get_next(struct seq_file *seq, struct dn_route *rt)
1684{
1685 struct dn_rt_cache_iter_state *s = rcu_dereference(seq->private);
1686
1687 rt = rt->u.rt_next;
1688 while(!rt) {
1689 rcu_read_unlock_bh();
1690 if (--s->bucket < 0)
1691 break;
1692 rcu_read_lock_bh();
1693 rt = dn_rt_hash_table[s->bucket].chain;
1694 }
1695 return rt;
1696}
1697
1698static void *dn_rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
1699{
1700 struct dn_route *rt = dn_rt_cache_get_first(seq);
1701
1702 if (rt) {
1703 while(*pos && (rt = dn_rt_cache_get_next(seq, rt)))
1704 --*pos;
1705 }
1706 return *pos ? NULL : rt;
1707}
1708
1709static void *dn_rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1710{
1711 struct dn_route *rt = dn_rt_cache_get_next(seq, v);
1712 ++*pos;
1713 return rt;
1714}
1715
1716static void dn_rt_cache_seq_stop(struct seq_file *seq, void *v)
1717{
1718 if (v)
1719 rcu_read_unlock_bh();
1720}
1721
1722static int dn_rt_cache_seq_show(struct seq_file *seq, void *v)
1723{
1724 struct dn_route *rt = v;
1725 char buf1[DN_ASCBUF_LEN], buf2[DN_ASCBUF_LEN];
1726
1727 seq_printf(seq, "%-8s %-7s %-7s %04d %04d %04d\n",
1728 rt->u.dst.dev ? rt->u.dst.dev->name : "*",
1729 dn_addr2asc(dn_ntohs(rt->rt_daddr), buf1),
1730 dn_addr2asc(dn_ntohs(rt->rt_saddr), buf2),
1731 atomic_read(&rt->u.dst.__refcnt),
1732 rt->u.dst.__use,
1733 (int) dst_metric(&rt->u.dst, RTAX_RTT));
1734 return 0;
1735}
1736
1737static struct seq_operations dn_rt_cache_seq_ops = {
1738 .start = dn_rt_cache_seq_start,
1739 .next = dn_rt_cache_seq_next,
1740 .stop = dn_rt_cache_seq_stop,
1741 .show = dn_rt_cache_seq_show,
1742};
1743
1744static int dn_rt_cache_seq_open(struct inode *inode, struct file *file)
1745{
1746 struct seq_file *seq;
1747 int rc = -ENOMEM;
1748 struct dn_rt_cache_iter_state *s = kmalloc(sizeof(*s), GFP_KERNEL);
1749
1750 if (!s)
1751 goto out;
1752 rc = seq_open(file, &dn_rt_cache_seq_ops);
1753 if (rc)
1754 goto out_kfree;
1755 seq = file->private_data;
1756 seq->private = s;
1757 memset(s, 0, sizeof(*s));
1758out:
1759 return rc;
1760out_kfree:
1761 kfree(s);
1762 goto out;
1763}
1764
1765static struct file_operations dn_rt_cache_seq_fops = {
1766 .owner = THIS_MODULE,
1767 .open = dn_rt_cache_seq_open,
1768 .read = seq_read,
1769 .llseek = seq_lseek,
1770 .release = seq_release_private,
1771};
1772
1773#endif /* CONFIG_PROC_FS */
1774
1775void __init dn_route_init(void)
1776{
1777 int i, goal, order;
1778
1779 dn_dst_ops.kmem_cachep = kmem_cache_create("dn_dst_cache",
1780 sizeof(struct dn_route),
1781 0, SLAB_HWCACHE_ALIGN,
1782 NULL, NULL);
1783
1784 if (!dn_dst_ops.kmem_cachep)
1785 panic("DECnet: Failed to allocate dn_dst_cache\n");
1786
1787 init_timer(&dn_route_timer);
1788 dn_route_timer.function = dn_dst_check_expire;
1789 dn_route_timer.expires = jiffies + decnet_dst_gc_interval * HZ;
1790 add_timer(&dn_route_timer);
1791
1792 goal = num_physpages >> (26 - PAGE_SHIFT);
1793
1794 for(order = 0; (1UL << order) < goal; order++)
1795 /* NOTHING */;
1796
1797 /*
1798 * Only want 1024 entries max, since the table is very, very unlikely
1799 * to be larger than that.
1800 */
1801 while(order && ((((1UL << order) * PAGE_SIZE) /
1802 sizeof(struct dn_rt_hash_bucket)) >= 2048))
1803 order--;
1804
1805 do {
1806 dn_rt_hash_mask = (1UL << order) * PAGE_SIZE /
1807 sizeof(struct dn_rt_hash_bucket);
1808 while(dn_rt_hash_mask & (dn_rt_hash_mask - 1))
1809 dn_rt_hash_mask--;
1810 dn_rt_hash_table = (struct dn_rt_hash_bucket *)
1811 __get_free_pages(GFP_ATOMIC, order);
1812 } while (dn_rt_hash_table == NULL && --order > 0);
1813
1814 if (!dn_rt_hash_table)
1815 panic("Failed to allocate DECnet route cache hash table\n");
1816
1817 printk(KERN_INFO
1818 "DECnet: Routing cache hash table of %u buckets, %ldKbytes\n",
1819 dn_rt_hash_mask,
1820 (long)(dn_rt_hash_mask*sizeof(struct dn_rt_hash_bucket))/1024);
1821
1822 dn_rt_hash_mask--;
1823 for(i = 0; i <= dn_rt_hash_mask; i++) {
1824 spin_lock_init(&dn_rt_hash_table[i].lock);
1825 dn_rt_hash_table[i].chain = NULL;
1826 }
1827
1828 dn_dst_ops.gc_thresh = (dn_rt_hash_mask + 1);
1829
1830 proc_net_fops_create("decnet_cache", S_IRUGO, &dn_rt_cache_seq_fops);
1831}
1832
1833void __exit dn_route_cleanup(void)
1834{
1835 del_timer(&dn_route_timer);
1836 dn_run_flush(0);
1837
1838 proc_net_remove("decnet_cache");
1839}
1840