blob: e172cf98d7fc3bdfda1831f06fed18fa20e097c6 [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
Steven Whitehousec4ea94a2006-03-20 22:42:39 -0800136static __inline__ unsigned dn_hash(__le16 src, __le16 dst)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137{
Steven Whitehousec4ea94a2006-03-20 22:42:39 -0800138 __u16 tmp = (__u16 __force)(src ^ dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139 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{
Adrian Bunk34001122006-03-20 23:00:29 -0800152 dst_release(&rt->u.dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153 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{
270 return memcmp(&fl1->nl_u.dn_u, &fl2->nl_u.dn_u, sizeof(fl1->nl_u.dn_u)) == 0 &&
271 fl1->oif == fl2->oif &&
272 fl1->iif == fl2->iif;
273}
274
275static int dn_insert_route(struct dn_route *rt, unsigned hash, struct dn_route **rp)
276{
277 struct dn_route *rth, **rthp;
278 unsigned long now = jiffies;
279
280 rthp = &dn_rt_hash_table[hash].chain;
281
282 spin_lock_bh(&dn_rt_hash_table[hash].lock);
283 while((rth = *rthp) != NULL) {
284 if (compare_keys(&rth->fl, &rt->fl)) {
285 /* Put it first */
286 *rthp = rth->u.rt_next;
287 rcu_assign_pointer(rth->u.rt_next,
288 dn_rt_hash_table[hash].chain);
289 rcu_assign_pointer(dn_rt_hash_table[hash].chain, rth);
290
291 rth->u.dst.__use++;
292 dst_hold(&rth->u.dst);
293 rth->u.dst.lastuse = now;
294 spin_unlock_bh(&dn_rt_hash_table[hash].lock);
295
296 dnrt_drop(rt);
297 *rp = rth;
298 return 0;
299 }
300 rthp = &rth->u.rt_next;
301 }
302
303 rcu_assign_pointer(rt->u.rt_next, dn_rt_hash_table[hash].chain);
304 rcu_assign_pointer(dn_rt_hash_table[hash].chain, rt);
305
306 dst_hold(&rt->u.dst);
307 rt->u.dst.__use++;
308 rt->u.dst.lastuse = now;
309 spin_unlock_bh(&dn_rt_hash_table[hash].lock);
310 *rp = rt;
311 return 0;
312}
313
314void dn_run_flush(unsigned long dummy)
315{
316 int i;
317 struct dn_route *rt, *next;
318
319 for(i = 0; i < dn_rt_hash_mask; i++) {
320 spin_lock_bh(&dn_rt_hash_table[i].lock);
321
322 if ((rt = xchg(&dn_rt_hash_table[i].chain, NULL)) == NULL)
323 goto nothing_to_declare;
324
325 for(; rt; rt=next) {
326 next = rt->u.rt_next;
327 rt->u.rt_next = NULL;
328 dst_free((struct dst_entry *)rt);
329 }
330
331nothing_to_declare:
332 spin_unlock_bh(&dn_rt_hash_table[i].lock);
333 }
334}
335
336static DEFINE_SPINLOCK(dn_rt_flush_lock);
337
338void dn_rt_cache_flush(int delay)
339{
340 unsigned long now = jiffies;
341 int user_mode = !in_interrupt();
342
343 if (delay < 0)
344 delay = dn_rt_min_delay;
345
346 spin_lock_bh(&dn_rt_flush_lock);
347
348 if (del_timer(&dn_rt_flush_timer) && delay > 0 && dn_rt_deadline) {
349 long tmo = (long)(dn_rt_deadline - now);
350
351 if (user_mode && tmo < dn_rt_max_delay - dn_rt_min_delay)
352 tmo = 0;
353
354 if (delay > tmo)
355 delay = tmo;
356 }
357
358 if (delay <= 0) {
359 spin_unlock_bh(&dn_rt_flush_lock);
360 dn_run_flush(0);
361 return;
362 }
363
364 if (dn_rt_deadline == 0)
365 dn_rt_deadline = now + dn_rt_max_delay;
366
367 dn_rt_flush_timer.expires = now + delay;
368 add_timer(&dn_rt_flush_timer);
369 spin_unlock_bh(&dn_rt_flush_lock);
370}
371
372/**
373 * dn_return_short - Return a short packet to its sender
374 * @skb: The packet to return
375 *
376 */
377static int dn_return_short(struct sk_buff *skb)
378{
379 struct dn_skb_cb *cb;
380 unsigned char *ptr;
Steven Whitehousec4ea94a2006-03-20 22:42:39 -0800381 __le16 *src;
382 __le16 *dst;
383 __le16 tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384
385 /* Add back headers */
386 skb_push(skb, skb->data - skb->nh.raw);
387
388 if ((skb = skb_unshare(skb, GFP_ATOMIC)) == NULL)
389 return NET_RX_DROP;
390
391 cb = DN_SKB_CB(skb);
392 /* Skip packet length and point to flags */
393 ptr = skb->data + 2;
394 *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
395
Steven Whitehousec4ea94a2006-03-20 22:42:39 -0800396 dst = (__le16 *)ptr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397 ptr += 2;
Steven Whitehousec4ea94a2006-03-20 22:42:39 -0800398 src = (__le16 *)ptr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 ptr += 2;
400 *ptr = 0; /* Zero hop count */
401
402 /* Swap source and destination */
403 tmp = *src;
404 *src = *dst;
405 *dst = tmp;
406
407 skb->pkt_type = PACKET_OUTGOING;
408 dn_rt_finish_output(skb, NULL, NULL);
409 return NET_RX_SUCCESS;
410}
411
412/**
413 * dn_return_long - Return a long packet to its sender
414 * @skb: The long format packet to return
415 *
416 */
417static int dn_return_long(struct sk_buff *skb)
418{
419 struct dn_skb_cb *cb;
420 unsigned char *ptr;
421 unsigned char *src_addr, *dst_addr;
422 unsigned char tmp[ETH_ALEN];
423
424 /* Add back all headers */
425 skb_push(skb, skb->data - skb->nh.raw);
426
427 if ((skb = skb_unshare(skb, GFP_ATOMIC)) == NULL)
428 return NET_RX_DROP;
429
430 cb = DN_SKB_CB(skb);
431 /* Ignore packet length and point to flags */
432 ptr = skb->data + 2;
433
434 /* Skip padding */
435 if (*ptr & DN_RT_F_PF) {
436 char padlen = (*ptr & ~DN_RT_F_PF);
437 ptr += padlen;
438 }
439
440 *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
441 ptr += 2;
442 dst_addr = ptr;
443 ptr += 8;
444 src_addr = ptr;
445 ptr += 6;
446 *ptr = 0; /* Zero hop count */
447
448 /* Swap source and destination */
449 memcpy(tmp, src_addr, ETH_ALEN);
450 memcpy(src_addr, dst_addr, ETH_ALEN);
451 memcpy(dst_addr, tmp, ETH_ALEN);
452
453 skb->pkt_type = PACKET_OUTGOING;
454 dn_rt_finish_output(skb, dst_addr, src_addr);
455 return NET_RX_SUCCESS;
456}
457
458/**
459 * dn_route_rx_packet - Try and find a route for an incoming packet
460 * @skb: The packet to find a route for
461 *
462 * Returns: result of input function if route is found, error code otherwise
463 */
464static int dn_route_rx_packet(struct sk_buff *skb)
465{
466 struct dn_skb_cb *cb = DN_SKB_CB(skb);
467 int err;
468
469 if ((err = dn_route_input(skb)) == 0)
470 return dst_input(skb);
471
472 if (decnet_debug_level & 4) {
473 char *devname = skb->dev ? skb->dev->name : "???";
474 struct dn_skb_cb *cb = DN_SKB_CB(skb);
475 printk(KERN_DEBUG
476 "DECnet: dn_route_rx_packet: rt_flags=0x%02x dev=%s len=%d src=0x%04hx dst=0x%04hx err=%d type=%d\n",
Steven Whitehousec4ea94a2006-03-20 22:42:39 -0800477 (int)cb->rt_flags, devname, skb->len,
478 dn_ntohs(cb->src), dn_ntohs(cb->dst),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479 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;
Steven Whitehousec4ea94a2006-03-20 22:42:39 -0800508 cb->dst = dn_eth2dn(ptr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509 if (memcmp(ptr, dn_hiord_addr, 4) != 0)
510 goto drop_it;
511 ptr += 6;
512
513
514 /* Source info */
515 ptr += 2;
Steven Whitehousec4ea94a2006-03-20 22:42:39 -0800516 cb->src = dn_eth2dn(ptr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517 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
Steven Whitehousec4ea94a2006-03-20 22:42:39 -0800544 cb->dst = *(__le16 *)ptr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545 ptr += 2;
Steven Whitehousec4ea94a2006-03-20 22:42:39 -0800546 cb->src = *(__le16 *)ptr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547 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;
Steven Whitehousec4ea94a2006-03-20 22:42:39 -0800578 __u16 len = dn_ntohs(*(__le16 *)skb->data);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 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",
Steven Whitehousec4ea94a2006-03-20 22:42:39 -0800785 dn_ntohs(cb->src), dn_ntohs(cb->dst));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 }
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
Steven Whitehousec4ea94a2006-03-20 22:42:39 -0800826static inline int dn_match_addr(__le16 addr1, __le16 addr2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827{
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
Steven Whitehousec4ea94a2006-03-20 22:42:39 -0800837static __le16 dnet_select_source(const struct net_device *dev, __le16 daddr, int scope)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838{
Steven Whitehousec4ea94a2006-03-20 22:42:39 -0800839 __le16 saddr = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840 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
Steven Whitehousec4ea94a2006-03-20 22:42:39 -0800864static inline __le16 __dn_fib_res_prefsrc(struct dn_fib_res *res)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865{
866 return dnet_select_source(DN_FIB_RES_DEV(*res), DN_FIB_RES_GW(*res), res->scope);
867}
868
Steven Whitehousec4ea94a2006-03-20 22:42:39 -0800869static inline __le16 dn_fib_rules_map_destination(__le16 daddr, struct dn_fib_res *res)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870{
Steven Whitehousec4ea94a2006-03-20 22:42:39 -0800871 __le16 mask = dnet_make_mask(res->prefixlen);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872 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;
Steven Whitehousec4ea94a2006-03-20 22:42:39 -0800895 __le16 gateway = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896
897 if (decnet_debug_level & 16)
898 printk(KERN_DEBUG
899 "dn_route_output_slow: dst=%04x src=%04x mark=%d"
Steven Whitehousec4ea94a2006-03-20 22:42:39 -0800900 " iif=%d oif=%d\n", dn_ntohs(oldflp->fld_dst),
901 dn_ntohs(oldflp->fld_src),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 oldflp->fld_fwmark, loopback_dev.ifindex, oldflp->oif);
903
904 /* If we have an output interface, verify its a DECnet device */
905 if (oldflp->oif) {
906 dev_out = dev_get_by_index(oldflp->oif);
907 err = -ENODEV;
908 if (dev_out && dev_out->dn_ptr == NULL) {
909 dev_put(dev_out);
910 dev_out = NULL;
911 }
912 if (dev_out == NULL)
913 goto out;
914 }
915
916 /* If we have a source address, verify that its a local address */
917 if (oldflp->fld_src) {
918 err = -EADDRNOTAVAIL;
919
920 if (dev_out) {
921 if (dn_dev_islocal(dev_out, oldflp->fld_src))
922 goto source_ok;
923 dev_put(dev_out);
924 goto out;
925 }
926 read_lock(&dev_base_lock);
927 for(dev_out = dev_base; dev_out; dev_out = dev_out->next) {
928 if (!dev_out->dn_ptr)
929 continue;
930 if (dn_dev_islocal(dev_out, oldflp->fld_src))
931 break;
932 }
933 read_unlock(&dev_base_lock);
934 if (dev_out == NULL)
935 goto out;
936 dev_hold(dev_out);
937source_ok:
938 ;
939 }
940
941 /* No destination? Assume its local */
942 if (!fl.fld_dst) {
943 fl.fld_dst = fl.fld_src;
944
945 err = -EADDRNOTAVAIL;
946 if (dev_out)
947 dev_put(dev_out);
948 dev_out = &loopback_dev;
949 dev_hold(dev_out);
950 if (!fl.fld_dst) {
951 fl.fld_dst =
952 fl.fld_src = dnet_select_source(dev_out, 0,
953 RT_SCOPE_HOST);
954 if (!fl.fld_dst)
955 goto out;
956 }
957 fl.oif = loopback_dev.ifindex;
958 res.type = RTN_LOCAL;
959 goto make_route;
960 }
961
962 if (decnet_debug_level & 16)
963 printk(KERN_DEBUG
964 "dn_route_output_slow: initial checks complete."
Steven Whitehousec4ea94a2006-03-20 22:42:39 -0800965 " dst=%o4x src=%04x oif=%d try_hard=%d\n",
966 dn_ntohs(fl.fld_dst), dn_ntohs(fl.fld_src),
967 fl.oif, try_hard);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968
969 /*
970 * N.B. If the kernel is compiled without router support then
971 * dn_fib_lookup() will evaluate to non-zero so this if () block
972 * will always be executed.
973 */
974 err = -ESRCH;
975 if (try_hard || (err = dn_fib_lookup(&fl, &res)) != 0) {
976 struct dn_dev *dn_db;
977 if (err != -ESRCH)
978 goto out;
979 /*
980 * Here the fallback is basically the standard algorithm for
981 * routing in endnodes which is described in the DECnet routing
982 * docs
983 *
984 * If we are not trying hard, look in neighbour cache.
985 * The result is tested to ensure that if a specific output
986 * device/source address was requested, then we honour that
987 * here
988 */
989 if (!try_hard) {
990 neigh = neigh_lookup_nodev(&dn_neigh_table, &fl.fld_dst);
991 if (neigh) {
992 if ((oldflp->oif &&
993 (neigh->dev->ifindex != oldflp->oif)) ||
994 (oldflp->fld_src &&
995 (!dn_dev_islocal(neigh->dev,
996 oldflp->fld_src)))) {
997 neigh_release(neigh);
998 neigh = NULL;
999 } else {
1000 if (dev_out)
1001 dev_put(dev_out);
1002 if (dn_dev_islocal(neigh->dev, fl.fld_dst)) {
1003 dev_out = &loopback_dev;
1004 res.type = RTN_LOCAL;
1005 } else {
1006 dev_out = neigh->dev;
1007 }
1008 dev_hold(dev_out);
1009 goto select_source;
1010 }
1011 }
1012 }
1013
1014 /* Not there? Perhaps its a local address */
1015 if (dev_out == NULL)
1016 dev_out = dn_dev_get_default();
1017 err = -ENODEV;
1018 if (dev_out == NULL)
1019 goto out;
1020 dn_db = dev_out->dn_ptr;
1021 /* Possible improvement - check all devices for local addr */
1022 if (dn_dev_islocal(dev_out, fl.fld_dst)) {
1023 dev_put(dev_out);
1024 dev_out = &loopback_dev;
1025 dev_hold(dev_out);
1026 res.type = RTN_LOCAL;
1027 goto select_source;
1028 }
1029 /* Not local either.... try sending it to the default router */
1030 neigh = neigh_clone(dn_db->router);
1031 BUG_ON(neigh && neigh->dev != dev_out);
1032
1033 /* Ok then, we assume its directly connected and move on */
1034select_source:
1035 if (neigh)
1036 gateway = ((struct dn_neigh *)neigh)->addr;
1037 if (gateway == 0)
1038 gateway = fl.fld_dst;
1039 if (fl.fld_src == 0) {
1040 fl.fld_src = dnet_select_source(dev_out, gateway,
1041 res.type == RTN_LOCAL ?
1042 RT_SCOPE_HOST :
1043 RT_SCOPE_LINK);
1044 if (fl.fld_src == 0 && res.type != RTN_LOCAL)
1045 goto e_addr;
1046 }
1047 fl.oif = dev_out->ifindex;
1048 goto make_route;
1049 }
1050 free_res = 1;
1051
1052 if (res.type == RTN_NAT)
1053 goto e_inval;
1054
1055 if (res.type == RTN_LOCAL) {
1056 if (!fl.fld_src)
1057 fl.fld_src = fl.fld_dst;
1058 if (dev_out)
1059 dev_put(dev_out);
1060 dev_out = &loopback_dev;
1061 dev_hold(dev_out);
1062 fl.oif = dev_out->ifindex;
1063 if (res.fi)
1064 dn_fib_info_put(res.fi);
1065 res.fi = NULL;
1066 goto make_route;
1067 }
1068
1069 if (res.fi->fib_nhs > 1 && fl.oif == 0)
1070 dn_fib_select_multipath(&fl, &res);
1071
1072 /*
1073 * We could add some logic to deal with default routes here and
1074 * get rid of some of the special casing above.
1075 */
1076
1077 if (!fl.fld_src)
1078 fl.fld_src = DN_FIB_RES_PREFSRC(res);
1079
1080 if (dev_out)
1081 dev_put(dev_out);
1082 dev_out = DN_FIB_RES_DEV(res);
1083 dev_hold(dev_out);
1084 fl.oif = dev_out->ifindex;
1085 gateway = DN_FIB_RES_GW(res);
1086
1087make_route:
1088 if (dev_out->flags & IFF_LOOPBACK)
1089 flags |= RTCF_LOCAL;
1090
1091 rt = dst_alloc(&dn_dst_ops);
1092 if (rt == NULL)
1093 goto e_nobufs;
1094
1095 atomic_set(&rt->u.dst.__refcnt, 1);
1096 rt->u.dst.flags = DST_HOST;
1097
1098 rt->fl.fld_src = oldflp->fld_src;
1099 rt->fl.fld_dst = oldflp->fld_dst;
1100 rt->fl.oif = oldflp->oif;
1101 rt->fl.iif = 0;
1102#ifdef CONFIG_DECNET_ROUTE_FWMARK
1103 rt->fl.fld_fwmark = oldflp->fld_fwmark;
1104#endif
1105
1106 rt->rt_saddr = fl.fld_src;
1107 rt->rt_daddr = fl.fld_dst;
1108 rt->rt_gateway = gateway ? gateway : fl.fld_dst;
1109 rt->rt_local_src = fl.fld_src;
1110
1111 rt->rt_dst_map = fl.fld_dst;
1112 rt->rt_src_map = fl.fld_src;
1113
1114 rt->u.dst.dev = dev_out;
1115 dev_hold(dev_out);
1116 rt->u.dst.neighbour = neigh;
1117 neigh = NULL;
1118
1119 rt->u.dst.lastuse = jiffies;
1120 rt->u.dst.output = dn_output;
1121 rt->u.dst.input = dn_rt_bug;
1122 rt->rt_flags = flags;
1123 if (flags & RTCF_LOCAL)
1124 rt->u.dst.input = dn_nsp_rx;
1125
1126 err = dn_rt_set_next_hop(rt, &res);
1127 if (err)
1128 goto e_neighbour;
1129
1130 hash = dn_hash(rt->fl.fld_src, rt->fl.fld_dst);
1131 dn_insert_route(rt, hash, (struct dn_route **)pprt);
1132
1133done:
1134 if (neigh)
1135 neigh_release(neigh);
1136 if (free_res)
1137 dn_fib_res_put(&res);
1138 if (dev_out)
1139 dev_put(dev_out);
1140out:
1141 return err;
1142
1143e_addr:
1144 err = -EADDRNOTAVAIL;
1145 goto done;
1146e_inval:
1147 err = -EINVAL;
1148 goto done;
1149e_nobufs:
1150 err = -ENOBUFS;
1151 goto done;
1152e_neighbour:
1153 dst_free(&rt->u.dst);
1154 goto e_nobufs;
1155}
1156
1157
1158/*
1159 * N.B. The flags may be moved into the flowi at some future stage.
1160 */
1161static int __dn_route_output_key(struct dst_entry **pprt, const struct flowi *flp, int flags)
1162{
1163 unsigned hash = dn_hash(flp->fld_src, flp->fld_dst);
1164 struct dn_route *rt = NULL;
1165
1166 if (!(flags & MSG_TRYHARD)) {
1167 rcu_read_lock_bh();
1168 for(rt = rcu_dereference(dn_rt_hash_table[hash].chain); rt;
1169 rt = rcu_dereference(rt->u.rt_next)) {
1170 if ((flp->fld_dst == rt->fl.fld_dst) &&
1171 (flp->fld_src == rt->fl.fld_src) &&
1172#ifdef CONFIG_DECNET_ROUTE_FWMARK
1173 (flp->fld_fwmark == rt->fl.fld_fwmark) &&
1174#endif
1175 (rt->fl.iif == 0) &&
1176 (rt->fl.oif == flp->oif)) {
1177 rt->u.dst.lastuse = jiffies;
1178 dst_hold(&rt->u.dst);
1179 rt->u.dst.__use++;
1180 rcu_read_unlock_bh();
1181 *pprt = &rt->u.dst;
1182 return 0;
1183 }
1184 }
1185 rcu_read_unlock_bh();
1186 }
1187
1188 return dn_route_output_slow(pprt, flp, flags);
1189}
1190
1191static int dn_route_output_key(struct dst_entry **pprt, struct flowi *flp, int flags)
1192{
1193 int err;
1194
1195 err = __dn_route_output_key(pprt, flp, flags);
1196 if (err == 0 && flp->proto) {
1197 err = xfrm_lookup(pprt, flp, NULL, 0);
1198 }
1199 return err;
1200}
1201
1202int dn_route_output_sock(struct dst_entry **pprt, struct flowi *fl, struct sock *sk, int flags)
1203{
1204 int err;
1205
1206 err = __dn_route_output_key(pprt, fl, flags & MSG_TRYHARD);
1207 if (err == 0 && fl->proto) {
1208 err = xfrm_lookup(pprt, fl, sk, !(flags & MSG_DONTWAIT));
1209 }
1210 return err;
1211}
1212
1213static int dn_route_input_slow(struct sk_buff *skb)
1214{
1215 struct dn_route *rt = NULL;
1216 struct dn_skb_cb *cb = DN_SKB_CB(skb);
1217 struct net_device *in_dev = skb->dev;
1218 struct net_device *out_dev = NULL;
1219 struct dn_dev *dn_db;
1220 struct neighbour *neigh = NULL;
1221 unsigned hash;
1222 int flags = 0;
Steven Whitehousec4ea94a2006-03-20 22:42:39 -08001223 __le16 gateway = 0;
1224 __le16 local_src = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225 struct flowi fl = { .nl_u = { .dn_u =
1226 { .daddr = cb->dst,
1227 .saddr = cb->src,
1228 .scope = RT_SCOPE_UNIVERSE,
1229#ifdef CONFIG_DECNET_ROUTE_FWMARK
1230 .fwmark = skb->nfmark
1231#endif
1232 } },
1233 .iif = skb->dev->ifindex };
1234 struct dn_fib_res res = { .fi = NULL, .type = RTN_UNREACHABLE };
1235 int err = -EINVAL;
1236 int free_res = 0;
1237
1238 dev_hold(in_dev);
1239
1240 if ((dn_db = in_dev->dn_ptr) == NULL)
1241 goto out;
1242
1243 /* Zero source addresses are not allowed */
1244 if (fl.fld_src == 0)
1245 goto out;
1246
1247 /*
1248 * In this case we've just received a packet from a source
1249 * outside ourselves pretending to come from us. We don't
1250 * allow it any further to prevent routing loops, spoofing and
1251 * other nasties. Loopback packets already have the dst attached
1252 * so this only affects packets which have originated elsewhere.
1253 */
1254 err = -ENOTUNIQ;
1255 if (dn_dev_islocal(in_dev, cb->src))
1256 goto out;
1257
1258 err = dn_fib_lookup(&fl, &res);
1259 if (err) {
1260 if (err != -ESRCH)
1261 goto out;
1262 /*
1263 * Is the destination us ?
1264 */
1265 if (!dn_dev_islocal(in_dev, cb->dst))
1266 goto e_inval;
1267
1268 res.type = RTN_LOCAL;
1269 flags |= RTCF_DIRECTSRC;
1270 } else {
Steven Whitehousec4ea94a2006-03-20 22:42:39 -08001271 __le16 src_map = fl.fld_src;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272 free_res = 1;
1273
1274 out_dev = DN_FIB_RES_DEV(res);
1275 if (out_dev == NULL) {
1276 if (net_ratelimit())
1277 printk(KERN_CRIT "Bug in dn_route_input_slow() "
1278 "No output device\n");
1279 goto e_inval;
1280 }
1281 dev_hold(out_dev);
1282
1283 if (res.r)
1284 src_map = dn_fib_rules_policy(fl.fld_src, &res, &flags);
1285
1286 gateway = DN_FIB_RES_GW(res);
1287 if (res.type == RTN_NAT) {
1288 fl.fld_dst = dn_fib_rules_map_destination(fl.fld_dst, &res);
1289 dn_fib_res_put(&res);
1290 free_res = 0;
1291 if (dn_fib_lookup(&fl, &res))
1292 goto e_inval;
1293 free_res = 1;
1294 if (res.type != RTN_UNICAST)
1295 goto e_inval;
1296 flags |= RTCF_DNAT;
1297 gateway = fl.fld_dst;
1298 }
1299 fl.fld_src = src_map;
1300 }
1301
1302 switch(res.type) {
1303 case RTN_UNICAST:
1304 /*
1305 * Forwarding check here, we only check for forwarding
1306 * being turned off, if you want to only forward intra
1307 * area, its up to you to set the routing tables up
1308 * correctly.
1309 */
1310 if (dn_db->parms.forwarding == 0)
1311 goto e_inval;
1312
1313 if (res.fi->fib_nhs > 1 && fl.oif == 0)
1314 dn_fib_select_multipath(&fl, &res);
1315
1316 /*
1317 * Check for out_dev == in_dev. We use the RTCF_DOREDIRECT
1318 * flag as a hint to set the intra-ethernet bit when
1319 * forwarding. If we've got NAT in operation, we don't do
1320 * this optimisation.
1321 */
1322 if (out_dev == in_dev && !(flags & RTCF_NAT))
1323 flags |= RTCF_DOREDIRECT;
1324
1325 local_src = DN_FIB_RES_PREFSRC(res);
1326
1327 case RTN_BLACKHOLE:
1328 case RTN_UNREACHABLE:
1329 break;
1330 case RTN_LOCAL:
1331 flags |= RTCF_LOCAL;
1332 fl.fld_src = cb->dst;
1333 fl.fld_dst = cb->src;
1334
1335 /* Routing tables gave us a gateway */
1336 if (gateway)
1337 goto make_route;
1338
1339 /* Packet was intra-ethernet, so we know its on-link */
1340 if (cb->rt_flags | DN_RT_F_IE) {
1341 gateway = cb->src;
1342 flags |= RTCF_DIRECTSRC;
1343 goto make_route;
1344 }
1345
1346 /* Use the default router if there is one */
1347 neigh = neigh_clone(dn_db->router);
1348 if (neigh) {
1349 gateway = ((struct dn_neigh *)neigh)->addr;
1350 goto make_route;
1351 }
1352
1353 /* Close eyes and pray */
1354 gateway = cb->src;
1355 flags |= RTCF_DIRECTSRC;
1356 goto make_route;
1357 default:
1358 goto e_inval;
1359 }
1360
1361make_route:
1362 rt = dst_alloc(&dn_dst_ops);
1363 if (rt == NULL)
1364 goto e_nobufs;
1365
1366 rt->rt_saddr = fl.fld_src;
1367 rt->rt_daddr = fl.fld_dst;
1368 rt->rt_gateway = fl.fld_dst;
1369 if (gateway)
1370 rt->rt_gateway = gateway;
1371 rt->rt_local_src = local_src ? local_src : rt->rt_saddr;
1372
1373 rt->rt_dst_map = fl.fld_dst;
1374 rt->rt_src_map = fl.fld_src;
1375
1376 rt->fl.fld_src = cb->src;
1377 rt->fl.fld_dst = cb->dst;
1378 rt->fl.oif = 0;
1379 rt->fl.iif = in_dev->ifindex;
1380 rt->fl.fld_fwmark = fl.fld_fwmark;
1381
1382 rt->u.dst.flags = DST_HOST;
1383 rt->u.dst.neighbour = neigh;
1384 rt->u.dst.dev = out_dev;
1385 rt->u.dst.lastuse = jiffies;
1386 rt->u.dst.output = dn_rt_bug;
1387 switch(res.type) {
1388 case RTN_UNICAST:
1389 rt->u.dst.input = dn_forward;
1390 break;
1391 case RTN_LOCAL:
1392 rt->u.dst.output = dn_output;
1393 rt->u.dst.input = dn_nsp_rx;
1394 rt->u.dst.dev = in_dev;
1395 flags |= RTCF_LOCAL;
1396 break;
1397 default:
1398 case RTN_UNREACHABLE:
1399 case RTN_BLACKHOLE:
1400 rt->u.dst.input = dn_blackhole;
1401 }
1402 rt->rt_flags = flags;
1403 if (rt->u.dst.dev)
1404 dev_hold(rt->u.dst.dev);
1405
1406 err = dn_rt_set_next_hop(rt, &res);
1407 if (err)
1408 goto e_neighbour;
1409
1410 hash = dn_hash(rt->fl.fld_src, rt->fl.fld_dst);
1411 dn_insert_route(rt, hash, (struct dn_route **)&skb->dst);
1412
1413done:
1414 if (neigh)
1415 neigh_release(neigh);
1416 if (free_res)
1417 dn_fib_res_put(&res);
1418 dev_put(in_dev);
1419 if (out_dev)
1420 dev_put(out_dev);
1421out:
1422 return err;
1423
1424e_inval:
1425 err = -EINVAL;
1426 goto done;
1427
1428e_nobufs:
1429 err = -ENOBUFS;
1430 goto done;
1431
1432e_neighbour:
1433 dst_free(&rt->u.dst);
1434 goto done;
1435}
1436
1437int dn_route_input(struct sk_buff *skb)
1438{
1439 struct dn_route *rt;
1440 struct dn_skb_cb *cb = DN_SKB_CB(skb);
1441 unsigned hash = dn_hash(cb->src, cb->dst);
1442
1443 if (skb->dst)
1444 return 0;
1445
1446 rcu_read_lock();
1447 for(rt = rcu_dereference(dn_rt_hash_table[hash].chain); rt != NULL;
1448 rt = rcu_dereference(rt->u.rt_next)) {
1449 if ((rt->fl.fld_src == cb->src) &&
1450 (rt->fl.fld_dst == cb->dst) &&
1451 (rt->fl.oif == 0) &&
1452#ifdef CONFIG_DECNET_ROUTE_FWMARK
1453 (rt->fl.fld_fwmark == skb->nfmark) &&
1454#endif
1455 (rt->fl.iif == cb->iif)) {
1456 rt->u.dst.lastuse = jiffies;
1457 dst_hold(&rt->u.dst);
1458 rt->u.dst.__use++;
1459 rcu_read_unlock();
1460 skb->dst = (struct dst_entry *)rt;
1461 return 0;
1462 }
1463 }
1464 rcu_read_unlock();
1465
1466 return dn_route_input_slow(skb);
1467}
1468
Jamal Hadi Salimb6544c02005-06-18 22:54:12 -07001469static int dn_rt_fill_info(struct sk_buff *skb, u32 pid, u32 seq,
1470 int event, int nowait, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471{
1472 struct dn_route *rt = (struct dn_route *)skb->dst;
1473 struct rtmsg *r;
1474 struct nlmsghdr *nlh;
1475 unsigned char *b = skb->tail;
1476 struct rta_cacheinfo ci;
1477
Jamal Hadi Salimb6544c02005-06-18 22:54:12 -07001478 nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*r), flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479 r = NLMSG_DATA(nlh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480 r->rtm_family = AF_DECnet;
1481 r->rtm_dst_len = 16;
1482 r->rtm_src_len = 0;
1483 r->rtm_tos = 0;
1484 r->rtm_table = RT_TABLE_MAIN;
1485 r->rtm_type = rt->rt_type;
1486 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
1487 r->rtm_scope = RT_SCOPE_UNIVERSE;
1488 r->rtm_protocol = RTPROT_UNSPEC;
1489 if (rt->rt_flags & RTCF_NOTIFY)
1490 r->rtm_flags |= RTM_F_NOTIFY;
1491 RTA_PUT(skb, RTA_DST, 2, &rt->rt_daddr);
1492 if (rt->fl.fld_src) {
1493 r->rtm_src_len = 16;
1494 RTA_PUT(skb, RTA_SRC, 2, &rt->fl.fld_src);
1495 }
1496 if (rt->u.dst.dev)
1497 RTA_PUT(skb, RTA_OIF, sizeof(int), &rt->u.dst.dev->ifindex);
1498 /*
1499 * Note to self - change this if input routes reverse direction when
1500 * they deal only with inputs and not with replies like they do
1501 * currently.
1502 */
1503 RTA_PUT(skb, RTA_PREFSRC, 2, &rt->rt_local_src);
1504 if (rt->rt_daddr != rt->rt_gateway)
1505 RTA_PUT(skb, RTA_GATEWAY, 2, &rt->rt_gateway);
1506 if (rtnetlink_put_metrics(skb, rt->u.dst.metrics) < 0)
1507 goto rtattr_failure;
1508 ci.rta_lastuse = jiffies_to_clock_t(jiffies - rt->u.dst.lastuse);
1509 ci.rta_used = rt->u.dst.__use;
1510 ci.rta_clntref = atomic_read(&rt->u.dst.__refcnt);
1511 if (rt->u.dst.expires)
1512 ci.rta_expires = jiffies_to_clock_t(rt->u.dst.expires - jiffies);
1513 else
1514 ci.rta_expires = 0;
1515 ci.rta_error = rt->u.dst.error;
1516 ci.rta_id = ci.rta_ts = ci.rta_tsage = 0;
1517 RTA_PUT(skb, RTA_CACHEINFO, sizeof(ci), &ci);
1518 if (rt->fl.iif)
1519 RTA_PUT(skb, RTA_IIF, sizeof(int), &rt->fl.iif);
1520
1521 nlh->nlmsg_len = skb->tail - b;
1522 return skb->len;
1523
1524nlmsg_failure:
1525rtattr_failure:
1526 skb_trim(skb, b - skb->data);
1527 return -1;
1528}
1529
1530/*
1531 * This is called by both endnodes and routers now.
1532 */
1533int dn_cache_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh, void *arg)
1534{
1535 struct rtattr **rta = arg;
1536 struct rtmsg *rtm = NLMSG_DATA(nlh);
1537 struct dn_route *rt = NULL;
1538 struct dn_skb_cb *cb;
1539 int err;
1540 struct sk_buff *skb;
1541 struct flowi fl;
1542
1543 memset(&fl, 0, sizeof(fl));
1544 fl.proto = DNPROTO_NSP;
1545
1546 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1547 if (skb == NULL)
1548 return -ENOBUFS;
1549 skb->mac.raw = skb->data;
1550 cb = DN_SKB_CB(skb);
1551
1552 if (rta[RTA_SRC-1])
1553 memcpy(&fl.fld_src, RTA_DATA(rta[RTA_SRC-1]), 2);
1554 if (rta[RTA_DST-1])
1555 memcpy(&fl.fld_dst, RTA_DATA(rta[RTA_DST-1]), 2);
1556 if (rta[RTA_IIF-1])
1557 memcpy(&fl.iif, RTA_DATA(rta[RTA_IIF-1]), sizeof(int));
1558
1559 if (fl.iif) {
1560 struct net_device *dev;
1561 if ((dev = dev_get_by_index(fl.iif)) == NULL) {
1562 kfree_skb(skb);
1563 return -ENODEV;
1564 }
1565 if (!dev->dn_ptr) {
1566 dev_put(dev);
1567 kfree_skb(skb);
1568 return -ENODEV;
1569 }
1570 skb->protocol = __constant_htons(ETH_P_DNA_RT);
1571 skb->dev = dev;
1572 cb->src = fl.fld_src;
1573 cb->dst = fl.fld_dst;
1574 local_bh_disable();
1575 err = dn_route_input(skb);
1576 local_bh_enable();
1577 memset(cb, 0, sizeof(struct dn_skb_cb));
1578 rt = (struct dn_route *)skb->dst;
1579 if (!err && -rt->u.dst.error)
1580 err = rt->u.dst.error;
1581 } else {
1582 int oif = 0;
1583 if (rta[RTA_OIF - 1])
1584 memcpy(&oif, RTA_DATA(rta[RTA_OIF - 1]), sizeof(int));
1585 fl.oif = oif;
1586 err = dn_route_output_key((struct dst_entry **)&rt, &fl, 0);
1587 }
1588
1589 if (skb->dev)
1590 dev_put(skb->dev);
1591 skb->dev = NULL;
1592 if (err)
1593 goto out_free;
1594 skb->dst = &rt->u.dst;
1595 if (rtm->rtm_flags & RTM_F_NOTIFY)
1596 rt->rt_flags |= RTCF_NOTIFY;
1597
1598 NETLINK_CB(skb).dst_pid = NETLINK_CB(in_skb).pid;
1599
Jamal Hadi Salimb6544c02005-06-18 22:54:12 -07001600 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 -07001601
1602 if (err == 0)
1603 goto out_free;
1604 if (err < 0) {
1605 err = -EMSGSIZE;
1606 goto out_free;
1607 }
1608
1609 err = netlink_unicast(rtnl, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1610
1611 return err;
1612
1613out_free:
1614 kfree_skb(skb);
1615 return err;
1616}
1617
1618/*
1619 * For routers, this is called from dn_fib_dump, but for endnodes its
1620 * called directly from the rtnetlink dispatch table.
1621 */
1622int dn_cache_dump(struct sk_buff *skb, struct netlink_callback *cb)
1623{
1624 struct dn_route *rt;
1625 int h, s_h;
1626 int idx, s_idx;
1627
1628 if (NLMSG_PAYLOAD(cb->nlh, 0) < sizeof(struct rtmsg))
1629 return -EINVAL;
1630 if (!(((struct rtmsg *)NLMSG_DATA(cb->nlh))->rtm_flags&RTM_F_CLONED))
1631 return 0;
1632
1633 s_h = cb->args[0];
1634 s_idx = idx = cb->args[1];
1635 for(h = 0; h <= dn_rt_hash_mask; h++) {
1636 if (h < s_h)
1637 continue;
1638 if (h > s_h)
1639 s_idx = 0;
1640 rcu_read_lock_bh();
1641 for(rt = rcu_dereference(dn_rt_hash_table[h].chain), idx = 0;
1642 rt;
1643 rt = rcu_dereference(rt->u.rt_next), idx++) {
1644 if (idx < s_idx)
1645 continue;
1646 skb->dst = dst_clone(&rt->u.dst);
1647 if (dn_rt_fill_info(skb, NETLINK_CB(cb->skb).pid,
Jamal Hadi Salimb6544c02005-06-18 22:54:12 -07001648 cb->nlh->nlmsg_seq, RTM_NEWROUTE,
1649 1, NLM_F_MULTI) <= 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650 dst_release(xchg(&skb->dst, NULL));
1651 rcu_read_unlock_bh();
1652 goto done;
1653 }
1654 dst_release(xchg(&skb->dst, NULL));
1655 }
1656 rcu_read_unlock_bh();
1657 }
1658
1659done:
1660 cb->args[0] = h;
1661 cb->args[1] = idx;
1662 return skb->len;
1663}
1664
1665#ifdef CONFIG_PROC_FS
1666struct dn_rt_cache_iter_state {
1667 int bucket;
1668};
1669
1670static struct dn_route *dn_rt_cache_get_first(struct seq_file *seq)
1671{
1672 struct dn_route *rt = NULL;
1673 struct dn_rt_cache_iter_state *s = seq->private;
1674
1675 for(s->bucket = dn_rt_hash_mask; s->bucket >= 0; --s->bucket) {
1676 rcu_read_lock_bh();
1677 rt = dn_rt_hash_table[s->bucket].chain;
1678 if (rt)
1679 break;
1680 rcu_read_unlock_bh();
1681 }
1682 return rt;
1683}
1684
1685static struct dn_route *dn_rt_cache_get_next(struct seq_file *seq, struct dn_route *rt)
1686{
1687 struct dn_rt_cache_iter_state *s = rcu_dereference(seq->private);
1688
1689 rt = rt->u.rt_next;
1690 while(!rt) {
1691 rcu_read_unlock_bh();
1692 if (--s->bucket < 0)
1693 break;
1694 rcu_read_lock_bh();
1695 rt = dn_rt_hash_table[s->bucket].chain;
1696 }
1697 return rt;
1698}
1699
1700static void *dn_rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
1701{
1702 struct dn_route *rt = dn_rt_cache_get_first(seq);
1703
1704 if (rt) {
1705 while(*pos && (rt = dn_rt_cache_get_next(seq, rt)))
1706 --*pos;
1707 }
1708 return *pos ? NULL : rt;
1709}
1710
1711static void *dn_rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1712{
1713 struct dn_route *rt = dn_rt_cache_get_next(seq, v);
1714 ++*pos;
1715 return rt;
1716}
1717
1718static void dn_rt_cache_seq_stop(struct seq_file *seq, void *v)
1719{
1720 if (v)
1721 rcu_read_unlock_bh();
1722}
1723
1724static int dn_rt_cache_seq_show(struct seq_file *seq, void *v)
1725{
1726 struct dn_route *rt = v;
1727 char buf1[DN_ASCBUF_LEN], buf2[DN_ASCBUF_LEN];
1728
1729 seq_printf(seq, "%-8s %-7s %-7s %04d %04d %04d\n",
1730 rt->u.dst.dev ? rt->u.dst.dev->name : "*",
1731 dn_addr2asc(dn_ntohs(rt->rt_daddr), buf1),
1732 dn_addr2asc(dn_ntohs(rt->rt_saddr), buf2),
1733 atomic_read(&rt->u.dst.__refcnt),
1734 rt->u.dst.__use,
1735 (int) dst_metric(&rt->u.dst, RTAX_RTT));
1736 return 0;
1737}
1738
1739static struct seq_operations dn_rt_cache_seq_ops = {
1740 .start = dn_rt_cache_seq_start,
1741 .next = dn_rt_cache_seq_next,
1742 .stop = dn_rt_cache_seq_stop,
1743 .show = dn_rt_cache_seq_show,
1744};
1745
1746static int dn_rt_cache_seq_open(struct inode *inode, struct file *file)
1747{
1748 struct seq_file *seq;
1749 int rc = -ENOMEM;
1750 struct dn_rt_cache_iter_state *s = kmalloc(sizeof(*s), GFP_KERNEL);
1751
1752 if (!s)
1753 goto out;
1754 rc = seq_open(file, &dn_rt_cache_seq_ops);
1755 if (rc)
1756 goto out_kfree;
1757 seq = file->private_data;
1758 seq->private = s;
1759 memset(s, 0, sizeof(*s));
1760out:
1761 return rc;
1762out_kfree:
1763 kfree(s);
1764 goto out;
1765}
1766
1767static struct file_operations dn_rt_cache_seq_fops = {
1768 .owner = THIS_MODULE,
1769 .open = dn_rt_cache_seq_open,
1770 .read = seq_read,
1771 .llseek = seq_lseek,
1772 .release = seq_release_private,
1773};
1774
1775#endif /* CONFIG_PROC_FS */
1776
1777void __init dn_route_init(void)
1778{
1779 int i, goal, order;
1780
1781 dn_dst_ops.kmem_cachep = kmem_cache_create("dn_dst_cache",
1782 sizeof(struct dn_route),
1783 0, SLAB_HWCACHE_ALIGN,
1784 NULL, NULL);
1785
1786 if (!dn_dst_ops.kmem_cachep)
1787 panic("DECnet: Failed to allocate dn_dst_cache\n");
1788
1789 init_timer(&dn_route_timer);
1790 dn_route_timer.function = dn_dst_check_expire;
1791 dn_route_timer.expires = jiffies + decnet_dst_gc_interval * HZ;
1792 add_timer(&dn_route_timer);
1793
1794 goal = num_physpages >> (26 - PAGE_SHIFT);
1795
1796 for(order = 0; (1UL << order) < goal; order++)
1797 /* NOTHING */;
1798
1799 /*
1800 * Only want 1024 entries max, since the table is very, very unlikely
1801 * to be larger than that.
1802 */
1803 while(order && ((((1UL << order) * PAGE_SIZE) /
1804 sizeof(struct dn_rt_hash_bucket)) >= 2048))
1805 order--;
1806
1807 do {
1808 dn_rt_hash_mask = (1UL << order) * PAGE_SIZE /
1809 sizeof(struct dn_rt_hash_bucket);
1810 while(dn_rt_hash_mask & (dn_rt_hash_mask - 1))
1811 dn_rt_hash_mask--;
1812 dn_rt_hash_table = (struct dn_rt_hash_bucket *)
1813 __get_free_pages(GFP_ATOMIC, order);
1814 } while (dn_rt_hash_table == NULL && --order > 0);
1815
1816 if (!dn_rt_hash_table)
1817 panic("Failed to allocate DECnet route cache hash table\n");
1818
1819 printk(KERN_INFO
1820 "DECnet: Routing cache hash table of %u buckets, %ldKbytes\n",
1821 dn_rt_hash_mask,
1822 (long)(dn_rt_hash_mask*sizeof(struct dn_rt_hash_bucket))/1024);
1823
1824 dn_rt_hash_mask--;
1825 for(i = 0; i <= dn_rt_hash_mask; i++) {
1826 spin_lock_init(&dn_rt_hash_table[i].lock);
1827 dn_rt_hash_table[i].chain = NULL;
1828 }
1829
1830 dn_dst_ops.gc_thresh = (dn_rt_hash_mask + 1);
1831
1832 proc_net_fops_create("decnet_cache", S_IRUGO, &dn_rt_cache_seq_fops);
1833}
1834
1835void __exit dn_route_cleanup(void)
1836{
1837 del_timer(&dn_route_timer);
1838 dn_run_flush(0);
1839
1840 proc_net_remove("decnet_cache");
1841}
1842