?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
[NETFILTER]: remove unused include file
[hh.org.git] / net / decnet / dn_route.c
blobdd0761e3d280cc311cbdf88b278b5e91cd4951ab
1 /*
2 * DECnet An implementation of the DECnet protocol suite for the LINUX
3 * operating system. DECnet is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * DECnet Routing Functions (Endnode and Router)
7 *
8 * Authors: Steve Whitehouse <SteveW@ACM.org>
9 * Eduardo Marcelo Serrat <emserrat@geocities.com>
10 *
11 * Changes:
12 * Steve Whitehouse : Fixes to allow "intra-ethernet" and
13 * "return-to-sender" bits on outgoing
14 * packets.
15 * Steve Whitehouse : Timeouts for cached routes.
16 * Steve Whitehouse : Use dst cache for input routes too.
17 * Steve Whitehouse : Fixed error values in dn_send_skb.
18 * Steve Whitehouse : Rework routing functions to better fit
19 * DECnet routing design
20 * Alexey Kuznetsov : New SMP locking
21 * Steve Whitehouse : More SMP locking changes & dn_cache_dump()
22 * Steve Whitehouse : Prerouting NF hook, now really is prerouting.
23 * Fixed possible skb leak in rtnetlink funcs.
24 * Steve Whitehouse : Dave Miller's dynamic hash table sizing and
25 * Alexey Kuznetsov's finer grained locking
26 * from ipv4/route.c.
27 * Steve Whitehouse : Routing is now starting to look like a
28 * sensible set of code now, mainly due to
29 * my copying the IPv4 routing code. The
30 * hooks here are modified and will continue
31 * to evolve for a while.
32 * Steve Whitehouse : Real SMP at last :-) Also new netfilter
33 * stuff. Look out raw sockets your days
34 * are numbered!
35 * Steve Whitehouse : Added return-to-sender functions. Added
36 * backlog congestion level return codes.
37 * Steve Whitehouse : Fixed bug where routes were set up with
38 * no ref count on net devices.
39 * Steve Whitehouse : RCU for the route cache
40 * Steve Whitehouse : Preparations for the flow cache
41 * Steve Whitehouse : Prepare for nonlinear skbs
42 */
43
44 /******************************************************************************
45 (c) 1995-1998 E.M. Serrat emserrat@geocities.com
46
47 This program is free software; you can redistribute it and/or modify
48 it under the terms of the GNU General Public License as published by
49 the Free Software Foundation; either version 2 of the License, or
50 any later version.
51
52 This program is distributed in the hope that it will be useful,
53 but WITHOUT ANY WARRANTY; without even the implied warranty of
54 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
55 GNU General Public License for more details.
56 *******************************************************************************/
57
58 #include <linux/errno.h>
59 #include <linux/types.h>
60 #include <linux/socket.h>
61 #include <linux/in.h>
62 #include <linux/kernel.h>
63 #include <linux/sockios.h>
64 #include <linux/net.h>
65 #include <linux/netdevice.h>
66 #include <linux/inet.h>
67 #include <linux/route.h>
68 #include <linux/in_route.h>
69 #include <net/sock.h>
70 #include <linux/mm.h>
71 #include <linux/proc_fs.h>
72 #include <linux/seq_file.h>
73 #include <linux/init.h>
74 #include <linux/rtnetlink.h>
75 #include <linux/string.h>
76 #include <linux/netfilter_decnet.h>
77 #include <linux/rcupdate.h>
78 #include <linux/times.h>
79 #include <asm/errno.h>
80 #include <net/neighbour.h>
81 #include <net/dst.h>
82 #include <net/flow.h>
83 #include <net/fib_rules.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
91 struct dn_rt_hash_bucket
92 {
93 struct dn_route *chain;
94 spinlock_t lock;
95 } __attribute__((__aligned__(8)));
96
97 extern struct neigh_table dn_neigh_table;
98
99
100 static unsigned char dn_hiord_addr[6] = {0xAA,0x00,0x04,0x00,0x00,0x00};
101
102 static const int dn_rt_min_delay = 2 * HZ;
103 static const int dn_rt_max_delay = 10 * HZ;
104 static const int dn_rt_mtu_expires = 10 * 60 * HZ;
105
106 static unsigned long dn_rt_deadline;
107
108 static int dn_dst_gc(void);
109 static struct dst_entry *dn_dst_check(struct dst_entry *, __u32);
110 static struct dst_entry *dn_dst_negative_advice(struct dst_entry *);
111 static void dn_dst_link_failure(struct sk_buff *);
112 static void dn_dst_update_pmtu(struct dst_entry *dst, u32 mtu);
113 static int dn_route_input(struct sk_buff *);
114 static void dn_run_flush(unsigned long dummy);
115
116 static struct dn_rt_hash_bucket *dn_rt_hash_table;
117 static unsigned dn_rt_hash_mask;
118
119 static struct timer_list dn_route_timer;
120 static DEFINE_TIMER(dn_rt_flush_timer, dn_run_flush, 0, 0);
121 int decnet_dst_gc_interval = 2;
122
123 static 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
136 static __inline__ unsigned dn_hash(__le16 src, __le16 dst)
137 {
138 __u16 tmp = (__u16 __force)(src ^ dst);
139 tmp ^= (tmp >> 3);
140 tmp ^= (tmp >> 5);
141 tmp ^= (tmp >> 10);
142 return dn_rt_hash_mask & (unsigned)tmp;
143 }
144
145 static inline void dnrt_free(struct dn_route *rt)
146 {
147 call_rcu_bh(&rt->u.dst.rcu_head, dst_rcu_free);
148 }
149
150 static inline void dnrt_drop(struct dn_route *rt)
151 {
152 dst_release(&rt->u.dst);
153 call_rcu_bh(&rt->u.dst.rcu_head, dst_rcu_free);
154 }
155
156 static 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
186 static 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 */
225 static 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 */
252 static struct dst_entry *dn_dst_check(struct dst_entry *dst, __u32 cookie)
253 {
254 return NULL;
255 }
256
257 static struct dst_entry *dn_dst_negative_advice(struct dst_entry *dst)
258 {
259 dst_release(dst);
260 return NULL;
261 }
262
263 static void dn_dst_link_failure(struct sk_buff *skb)
264 {
265 return;
266 }
267
268 static 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
275 static 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
314 void 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
331 nothing_to_declare:
332 spin_unlock_bh(&dn_rt_hash_table[i].lock);
333 }
334 }
335
336 static DEFINE_SPINLOCK(dn_rt_flush_lock);
337
338 void 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 */
377 static int dn_return_short(struct sk_buff *skb)
378 {
379 struct dn_skb_cb *cb;
380 unsigned char *ptr;
381 __le16 *src;
382 __le16 *dst;
383 __le16 tmp;
384
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
396 dst = (__le16 *)ptr;
397 ptr += 2;
398 src = (__le16 *)ptr;
399 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 */
417 static 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 */
464 static 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",
477 (int)cb->rt_flags, devname, skb->len,
478 dn_ntohs(cb->src), dn_ntohs(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
495 static 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_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_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
526 drop_it:
527 kfree_skb(skb);
528 return NET_RX_DROP;
529 }
530
531
532
533 static 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 = *(__le16 *)ptr;
545 ptr += 2;
546 cb->src = *(__le16 *)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
552 drop_it:
553 kfree_skb(skb);
554 return NET_RX_DROP;
555 }
556
557 static 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
567 static 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
574 int dn_route_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
575 {
576 struct dn_skb_cb *cb;
577 unsigned char flags = 0;
578 __u16 len = dn_ntohs(*(__le16 *)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_linearize(skb)))
633 goto dump_it;
634
635 switch(flags & DN_RT_CNTL_MSK) {
636 case DN_RT_PKT_INIT:
637 dn_dev_init_pkt(skb);
638 break;
639 case DN_RT_PKT_VERI:
640 dn_dev_veri_pkt(skb);
641 break;
642 }
643
644 if (dn->parms.state != DN_DEV_S_RU)
645 goto dump_it;
646
647 switch(flags & DN_RT_CNTL_MSK) {
648 case DN_RT_PKT_HELO:
649 return NF_HOOK(PF_DECnet, NF_DN_HELLO, skb, skb->dev, NULL, dn_route_ptp_hello);
650
651 case DN_RT_PKT_L1RT:
652 case DN_RT_PKT_L2RT:
653 return NF_HOOK(PF_DECnet, NF_DN_ROUTE, skb, skb->dev, NULL, dn_route_discard);
654 case DN_RT_PKT_ERTH:
655 return NF_HOOK(PF_DECnet, NF_DN_HELLO, skb, skb->dev, NULL, dn_neigh_router_hello);
656
657 case DN_RT_PKT_EEDH:
658 return NF_HOOK(PF_DECnet, NF_DN_HELLO, skb, skb->dev, NULL, dn_neigh_endnode_hello);
659 }
660 } else {
661 if (dn->parms.state != DN_DEV_S_RU)
662 goto dump_it;
663
664 skb_pull(skb, 1); /* Pull flags */
665
666 switch(flags & DN_RT_PKT_MSK) {
667 case DN_RT_PKT_LONG:
668 return dn_route_rx_long(skb);
669 case DN_RT_PKT_SHORT:
670 return dn_route_rx_short(skb);
671 }
672 }
673
674 dump_it:
675 kfree_skb(skb);
676 out:
677 return NET_RX_DROP;
678 }
679
680 static int dn_output(struct sk_buff *skb)
681 {
682 struct dst_entry *dst = skb->dst;
683 struct dn_route *rt = (struct dn_route *)dst;
684 struct net_device *dev = dst->dev;
685 struct dn_skb_cb *cb = DN_SKB_CB(skb);
686 struct neighbour *neigh;
687
688 int err = -EINVAL;
689
690 if ((neigh = dst->neighbour) == NULL)
691 goto error;
692
693 skb->dev = dev;
694
695 cb->src = rt->rt_saddr;
696 cb->dst = rt->rt_daddr;
697
698 /*
699 * Always set the Intra-Ethernet bit on all outgoing packets
700 * originated on this node. Only valid flag from upper layers
701 * is return-to-sender-requested. Set hop count to 0 too.
702 */
703 cb->rt_flags &= ~DN_RT_F_RQR;
704 cb->rt_flags |= DN_RT_F_IE;
705 cb->hops = 0;
706
707 return NF_HOOK(PF_DECnet, NF_DN_LOCAL_OUT, skb, NULL, dev, neigh->output);
708
709 error:
710 if (net_ratelimit())
711 printk(KERN_DEBUG "dn_output: This should not happen\n");
712
713 kfree_skb(skb);
714
715 return err;
716 }
717
718 static int dn_forward(struct sk_buff *skb)
719 {
720 struct dn_skb_cb *cb = DN_SKB_CB(skb);
721 struct dst_entry *dst = skb->dst;
722 struct dn_dev *dn_db = dst->dev->dn_ptr;
723 struct dn_route *rt;
724 struct neighbour *neigh = dst->neighbour;
725 int header_len;
726 #ifdef CONFIG_NETFILTER
727 struct net_device *dev = skb->dev;
728 #endif
729
730 if (skb->pkt_type != PACKET_HOST)
731 goto drop;
732
733 /* Ensure that we have enough space for headers */
734 rt = (struct dn_route *)skb->dst;
735 header_len = dn_db->use_long ? 21 : 6;
736 if (skb_cow(skb, LL_RESERVED_SPACE(rt->u.dst.dev)+header_len))
737 goto drop;
738
739 /*
740 * Hop count exceeded.
741 */
742 if (++cb->hops > 30)
743 goto drop;
744
745 skb->dev = rt->u.dst.dev;
746
747 /*
748 * If packet goes out same interface it came in on, then set
749 * the Intra-Ethernet bit. This has no effect for short
750 * packets, so we don't need to test for them here.
751 */
752 cb->rt_flags &= ~DN_RT_F_IE;
753 if (rt->rt_flags & RTCF_DOREDIRECT)
754 cb->rt_flags |= DN_RT_F_IE;
755
756 return NF_HOOK(PF_DECnet, NF_DN_FORWARD, skb, dev, skb->dev, neigh->output);
757
758 drop:
759 kfree_skb(skb);
760 return NET_RX_DROP;
761 }
762
763 /*
764 * Drop packet. This is used for endnodes and for
765 * when we should not be forwarding packets from
766 * this dest.
767 */
768 static int dn_blackhole(struct sk_buff *skb)
769 {
770 kfree_skb(skb);
771 return NET_RX_DROP;
772 }
773
774 /*
775 * Used to catch bugs. This should never normally get
776 * called.
777 */
778 static int dn_rt_bug(struct sk_buff *skb)
779 {
780 if (net_ratelimit()) {
781 struct dn_skb_cb *cb = DN_SKB_CB(skb);
782
783 printk(KERN_DEBUG "dn_rt_bug: skb from:%04x to:%04x\n",
784 dn_ntohs(cb->src), dn_ntohs(cb->dst));
785 }
786
787 kfree_skb(skb);
788
789 return NET_RX_BAD;
790 }
791
792 static int dn_rt_set_next_hop(struct dn_route *rt, struct dn_fib_res *res)
793 {
794 struct dn_fib_info *fi = res->fi;
795 struct net_device *dev = rt->u.dst.dev;
796 struct neighbour *n;
797 unsigned mss;
798
799 if (fi) {
800 if (DN_FIB_RES_GW(*res) &&
801 DN_FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
802 rt->rt_gateway = DN_FIB_RES_GW(*res);
803 memcpy(rt->u.dst.metrics, fi->fib_metrics,
804 sizeof(rt->u.dst.metrics));
805 }
806 rt->rt_type = res->type;
807
808 if (dev != NULL && rt->u.dst.neighbour == NULL) {
809 n = __neigh_lookup_errno(&dn_neigh_table, &rt->rt_gateway, dev);
810 if (IS_ERR(n))
811 return PTR_ERR(n);
812 rt->u.dst.neighbour = n;
813 }
814
815 if (rt->u.dst.metrics[RTAX_MTU-1] == 0 ||
816 rt->u.dst.metrics[RTAX_MTU-1] > rt->u.dst.dev->mtu)
817 rt->u.dst.metrics[RTAX_MTU-1] = rt->u.dst.dev->mtu;
818 mss = dn_mss_from_pmtu(dev, dst_mtu(&rt->u.dst));
819 if (rt->u.dst.metrics[RTAX_ADVMSS-1] == 0 ||
820 rt->u.dst.metrics[RTAX_ADVMSS-1] > mss)
821 rt->u.dst.metrics[RTAX_ADVMSS-1] = mss;
822 return 0;
823 }
824
825 static inline int dn_match_addr(__le16 addr1, __le16 addr2)
826 {
827 __u16 tmp = dn_ntohs(addr1) ^ dn_ntohs(addr2);
828 int match = 16;
829 while(tmp) {
830 tmp >>= 1;
831 match--;
832 }
833 return match;
834 }
835
836 static __le16 dnet_select_source(const struct net_device *dev, __le16 daddr, int scope)
837 {
838 __le16 saddr = 0;
839 struct dn_dev *dn_db = dev->dn_ptr;
840 struct dn_ifaddr *ifa;
841 int best_match = 0;
842 int ret;
843
844 read_lock(&dev_base_lock);
845 for(ifa = dn_db->ifa_list; ifa; ifa = ifa->ifa_next) {
846 if (ifa->ifa_scope > scope)
847 continue;
848 if (!daddr) {
849 saddr = ifa->ifa_local;
850 break;
851 }
852 ret = dn_match_addr(daddr, ifa->ifa_local);
853 if (ret > best_match)
854 saddr = ifa->ifa_local;
855 if (best_match == 0)
856 saddr = ifa->ifa_local;
857 }
858 read_unlock(&dev_base_lock);
859
860 return saddr;
861 }
862
863 static inline __le16 __dn_fib_res_prefsrc(struct dn_fib_res *res)
864 {
865 return dnet_select_source(DN_FIB_RES_DEV(*res), DN_FIB_RES_GW(*res), res->scope);
866 }
867
868 static inline __le16 dn_fib_rules_map_destination(__le16 daddr, struct dn_fib_res *res)
869 {
870 __le16 mask = dnet_make_mask(res->prefixlen);
871 return (daddr&~mask)|res->fi->fib_nh->nh_gw;
872 }
873
874 static int dn_route_output_slow(struct dst_entry **pprt, const struct flowi *oldflp, int try_hard)
875 {
876 struct flowi fl = { .nl_u = { .dn_u =
877 { .daddr = oldflp->fld_dst,
878 .saddr = oldflp->fld_src,
879 .scope = RT_SCOPE_UNIVERSE,
880 #ifdef CONFIG_DECNET_ROUTE_FWMARK
881 .fwmark = oldflp->fld_fwmark
882 #endif
883 } },
884 .iif = loopback_dev.ifindex,
885 .oif = oldflp->oif };
886 struct dn_route *rt = NULL;
887 struct net_device *dev_out = NULL;
888 struct neighbour *neigh = NULL;
889 unsigned hash;
890 unsigned flags = 0;
891 struct dn_fib_res res = { .fi = NULL, .type = RTN_UNICAST };
892 int err;
893 int free_res = 0;
894 __le16 gateway = 0;
895
896 if (decnet_debug_level & 16)
897 printk(KERN_DEBUG
898 "dn_route_output_slow: dst=%04x src=%04x mark=%d"
899 " iif=%d oif=%d\n", dn_ntohs(oldflp->fld_dst),
900 dn_ntohs(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 continue;
931 if ((dev_out->flags & IFF_LOOPBACK) &&
932 oldflp->fld_dst &&
933 !dn_dev_islocal(dev_out, oldflp->fld_dst))
934 continue;
935 break;
936 }
937 read_unlock(&dev_base_lock);
938 if (dev_out == NULL)
939 goto out;
940 dev_hold(dev_out);
941 source_ok:
942 ;
943 }
944
945 /* No destination? Assume its local */
946 if (!fl.fld_dst) {
947 fl.fld_dst = fl.fld_src;
948
949 err = -EADDRNOTAVAIL;
950 if (dev_out)
951 dev_put(dev_out);
952 dev_out = &loopback_dev;
953 dev_hold(dev_out);
954 if (!fl.fld_dst) {
955 fl.fld_dst =
956 fl.fld_src = dnet_select_source(dev_out, 0,
957 RT_SCOPE_HOST);
958 if (!fl.fld_dst)
959 goto out;
960 }
961 fl.oif = loopback_dev.ifindex;
962 res.type = RTN_LOCAL;
963 goto make_route;
964 }
965
966 if (decnet_debug_level & 16)
967 printk(KERN_DEBUG
968 "dn_route_output_slow: initial checks complete."
969 " dst=%o4x src=%04x oif=%d try_hard=%d\n",
970 dn_ntohs(fl.fld_dst), dn_ntohs(fl.fld_src),
971 fl.oif, try_hard);
972
973 /*
974 * N.B. If the kernel is compiled without router support then
975 * dn_fib_lookup() will evaluate to non-zero so this if () block
976 * will always be executed.
977 */
978 err = -ESRCH;
979 if (try_hard || (err = dn_fib_lookup(&fl, &res)) != 0) {
980 struct dn_dev *dn_db;
981 if (err != -ESRCH)
982 goto out;
983 /*
984 * Here the fallback is basically the standard algorithm for
985 * routing in endnodes which is described in the DECnet routing
986 * docs
987 *
988 * If we are not trying hard, look in neighbour cache.
989 * The result is tested to ensure that if a specific output
990 * device/source address was requested, then we honour that
991 * here
992 */
993 if (!try_hard) {
994 neigh = neigh_lookup_nodev(&dn_neigh_table, &fl.fld_dst);
995 if (neigh) {
996 if ((oldflp->oif &&
997 (neigh->dev->ifindex != oldflp->oif)) ||
998 (oldflp->fld_src &&
999 (!dn_dev_islocal(neigh->dev,
1000 oldflp->fld_src)))) {
1001 neigh_release(neigh);
1002 neigh = NULL;
1003 } else {
1004 if (dev_out)
1005 dev_put(dev_out);
1006 if (dn_dev_islocal(neigh->dev, fl.fld_dst)) {
1007 dev_out = &loopback_dev;
1008 res.type = RTN_LOCAL;
1009 } else {
1010 dev_out = neigh->dev;
1011 }
1012 dev_hold(dev_out);
1013 goto select_source;
1014 }
1015 }
1016 }
1017
1018 /* Not there? Perhaps its a local address */
1019 if (dev_out == NULL)
1020 dev_out = dn_dev_get_default();
1021 err = -ENODEV;
1022 if (dev_out == NULL)
1023 goto out;
1024 dn_db = dev_out->dn_ptr;
1025 /* Possible improvement - check all devices for local addr */
1026 if (dn_dev_islocal(dev_out, fl.fld_dst)) {
1027 dev_put(dev_out);
1028 dev_out = &loopback_dev;
1029 dev_hold(dev_out);
1030 res.type = RTN_LOCAL;
1031 goto select_source;
1032 }
1033 /* Not local either.... try sending it to the default router */
1034 neigh = neigh_clone(dn_db->router);
1035 BUG_ON(neigh && neigh->dev != dev_out);
1036
1037 /* Ok then, we assume its directly connected and move on */
1038 select_source:
1039 if (neigh)
1040 gateway = ((struct dn_neigh *)neigh)->addr;
1041 if (gateway == 0)
1042 gateway = fl.fld_dst;
1043 if (fl.fld_src == 0) {
1044 fl.fld_src = dnet_select_source(dev_out, gateway,
1045 res.type == RTN_LOCAL ?
1046 RT_SCOPE_HOST :
1047 RT_SCOPE_LINK);
1048 if (fl.fld_src == 0 && res.type != RTN_LOCAL)
1049 goto e_addr;
1050 }
1051 fl.oif = dev_out->ifindex;
1052 goto make_route;
1053 }
1054 free_res = 1;
1055
1056 if (res.type == RTN_NAT)
1057 goto e_inval;
1058
1059 if (res.type == RTN_LOCAL) {
1060 if (!fl.fld_src)
1061 fl.fld_src = fl.fld_dst;
1062 if (dev_out)
1063 dev_put(dev_out);
1064 dev_out = &loopback_dev;
1065 dev_hold(dev_out);
1066 fl.oif = dev_out->ifindex;
1067 if (res.fi)
1068 dn_fib_info_put(res.fi);
1069 res.fi = NULL;
1070 goto make_route;
1071 }
1072
1073 if (res.fi->fib_nhs > 1 && fl.oif == 0)
1074 dn_fib_select_multipath(&fl, &res);
1075
1076 /*
1077 * We could add some logic to deal with default routes here and
1078 * get rid of some of the special casing above.
1079 */
1080
1081 if (!fl.fld_src)
1082 fl.fld_src = DN_FIB_RES_PREFSRC(res);
1083
1084 if (dev_out)
1085 dev_put(dev_out);
1086 dev_out = DN_FIB_RES_DEV(res);
1087 dev_hold(dev_out);
1088 fl.oif = dev_out->ifindex;
1089 gateway = DN_FIB_RES_GW(res);
1090
1091 make_route:
1092 if (dev_out->flags & IFF_LOOPBACK)
1093 flags |= RTCF_LOCAL;
1094
1095 rt = dst_alloc(&dn_dst_ops);
1096 if (rt == NULL)
1097 goto e_nobufs;
1098
1099 atomic_set(&rt->u.dst.__refcnt, 1);
1100 rt->u.dst.flags = DST_HOST;
1101
1102 rt->fl.fld_src = oldflp->fld_src;
1103 rt->fl.fld_dst = oldflp->fld_dst;
1104 rt->fl.oif = oldflp->oif;
1105 rt->fl.iif = 0;
1106 #ifdef CONFIG_DECNET_ROUTE_FWMARK
1107 rt->fl.fld_fwmark = oldflp->fld_fwmark;
1108 #endif
1109
1110 rt->rt_saddr = fl.fld_src;
1111 rt->rt_daddr = fl.fld_dst;
1112 rt->rt_gateway = gateway ? gateway : fl.fld_dst;
1113 rt->rt_local_src = fl.fld_src;
1114
1115 rt->rt_dst_map = fl.fld_dst;
1116 rt->rt_src_map = fl.fld_src;
1117
1118 rt->u.dst.dev = dev_out;
1119 dev_hold(dev_out);
1120 rt->u.dst.neighbour = neigh;
1121 neigh = NULL;
1122
1123 rt->u.dst.lastuse = jiffies;
1124 rt->u.dst.output = dn_output;
1125 rt->u.dst.input = dn_rt_bug;
1126 rt->rt_flags = flags;
1127 if (flags & RTCF_LOCAL)
1128 rt->u.dst.input = dn_nsp_rx;
1129
1130 err = dn_rt_set_next_hop(rt, &res);
1131 if (err)
1132 goto e_neighbour;
1133
1134 hash = dn_hash(rt->fl.fld_src, rt->fl.fld_dst);
1135 dn_insert_route(rt, hash, (struct dn_route **)pprt);
1136
1137 done:
1138 if (neigh)
1139 neigh_release(neigh);
1140 if (free_res)
1141 dn_fib_res_put(&res);
1142 if (dev_out)
1143 dev_put(dev_out);
1144 out:
1145 return err;
1146
1147 e_addr:
1148 err = -EADDRNOTAVAIL;
1149 goto done;
1150 e_inval:
1151 err = -EINVAL;
1152 goto done;
1153 e_nobufs:
1154 err = -ENOBUFS;
1155 goto done;
1156 e_neighbour:
1157 dst_free(&rt->u.dst);
1158 goto e_nobufs;
1159 }
1160
1161
1162 /*
1163 * N.B. The flags may be moved into the flowi at some future stage.
1164 */
1165 static int __dn_route_output_key(struct dst_entry **pprt, const struct flowi *flp, int flags)
1166 {
1167 unsigned hash = dn_hash(flp->fld_src, flp->fld_dst);
1168 struct dn_route *rt = NULL;
1169
1170 if (!(flags & MSG_TRYHARD)) {
1171 rcu_read_lock_bh();
1172 for(rt = rcu_dereference(dn_rt_hash_table[hash].chain); rt;
1173 rt = rcu_dereference(rt->u.rt_next)) {
1174 if ((flp->fld_dst == rt->fl.fld_dst) &&
1175 (flp->fld_src == rt->fl.fld_src) &&
1176 #ifdef CONFIG_DECNET_ROUTE_FWMARK
1177 (flp->fld_fwmark == rt->fl.fld_fwmark) &&
1178 #endif
1179 (rt->fl.iif == 0) &&
1180 (rt->fl.oif == flp->oif)) {
1181 rt->u.dst.lastuse = jiffies;
1182 dst_hold(&rt->u.dst);
1183 rt->u.dst.__use++;
1184 rcu_read_unlock_bh();
1185 *pprt = &rt->u.dst;
1186 return 0;
1187 }
1188 }
1189 rcu_read_unlock_bh();
1190 }
1191
1192 return dn_route_output_slow(pprt, flp, flags);
1193 }
1194
1195 static int dn_route_output_key(struct dst_entry **pprt, struct flowi *flp, int flags)
1196 {
1197 int err;
1198
1199 err = __dn_route_output_key(pprt, flp, flags);
1200 if (err == 0 && flp->proto) {
1201 err = xfrm_lookup(pprt, flp, NULL, 0);
1202 }
1203 return err;
1204 }
1205
1206 int dn_route_output_sock(struct dst_entry **pprt, struct flowi *fl, struct sock *sk, int flags)
1207 {
1208 int err;
1209
1210 err = __dn_route_output_key(pprt, fl, flags & MSG_TRYHARD);
1211 if (err == 0 && fl->proto) {
1212 err = xfrm_lookup(pprt, fl, sk, !(flags & MSG_DONTWAIT));
1213 }
1214 return err;
1215 }
1216
1217 static int dn_route_input_slow(struct sk_buff *skb)
1218 {
1219 struct dn_route *rt = NULL;
1220 struct dn_skb_cb *cb = DN_SKB_CB(skb);
1221 struct net_device *in_dev = skb->dev;
1222 struct net_device *out_dev = NULL;
1223 struct dn_dev *dn_db;
1224 struct neighbour *neigh = NULL;
1225 unsigned hash;
1226 int flags = 0;
1227 __le16 gateway = 0;
1228 __le16 local_src = 0;
1229 struct flowi fl = { .nl_u = { .dn_u =
1230 { .daddr = cb->dst,
1231 .saddr = cb->src,
1232 .scope = RT_SCOPE_UNIVERSE,
1233 #ifdef CONFIG_DECNET_ROUTE_FWMARK
1234 .fwmark = skb->nfmark
1235 #endif
1236 } },
1237 .iif = skb->dev->ifindex };
1238 struct dn_fib_res res = { .fi = NULL, .type = RTN_UNREACHABLE };
1239 int err = -EINVAL;
1240 int free_res = 0;
1241
1242 dev_hold(in_dev);
1243
1244 if ((dn_db = in_dev->dn_ptr) == NULL)
1245 goto out;
1246
1247 /* Zero source addresses are not allowed */
1248 if (fl.fld_src == 0)
1249 goto out;
1250
1251 /*
1252 * In this case we've just received a packet from a source
1253 * outside ourselves pretending to come from us. We don't
1254 * allow it any further to prevent routing loops, spoofing and
1255 * other nasties. Loopback packets already have the dst attached
1256 * so this only affects packets which have originated elsewhere.
1257 */
1258 err = -ENOTUNIQ;
1259 if (dn_dev_islocal(in_dev, cb->src))
1260 goto out;
1261
1262 err = dn_fib_lookup(&fl, &res);
1263 if (err) {
1264 if (err != -ESRCH)
1265 goto out;
1266 /*
1267 * Is the destination us ?
1268 */
1269 if (!dn_dev_islocal(in_dev, cb->dst))
1270 goto e_inval;
1271
1272 res.type = RTN_LOCAL;
1273 flags |= RTCF_DIRECTSRC;
1274 } else {
1275 __le16 src_map = fl.fld_src;
1276 free_res = 1;
1277
1278 out_dev = DN_FIB_RES_DEV(res);
1279 if (out_dev == NULL) {
1280 if (net_ratelimit())
1281 printk(KERN_CRIT "Bug in dn_route_input_slow() "
1282 "No output device\n");
1283 goto e_inval;
1284 }
1285 dev_hold(out_dev);
1286
1287 if (res.r)
1288 src_map = fl.fld_src; /* no NAT support for now */
1289
1290 gateway = DN_FIB_RES_GW(res);
1291 if (res.type == RTN_NAT) {
1292 fl.fld_dst = dn_fib_rules_map_destination(fl.fld_dst, &res);
1293 dn_fib_res_put(&res);
1294 free_res = 0;
1295 if (dn_fib_lookup(&fl, &res))
1296 goto e_inval;
1297 free_res = 1;
1298 if (res.type != RTN_UNICAST)
1299 goto e_inval;
1300 flags |= RTCF_DNAT;
1301 gateway = fl.fld_dst;
1302 }
1303 fl.fld_src = src_map;
1304 }
1305
1306 switch(res.type) {
1307 case RTN_UNICAST:
1308 /*
1309 * Forwarding check here, we only check for forwarding
1310 * being turned off, if you want to only forward intra
1311 * area, its up to you to set the routing tables up
1312 * correctly.
1313 */
1314 if (dn_db->parms.forwarding == 0)
1315 goto e_inval;
1316
1317 if (res.fi->fib_nhs > 1 && fl.oif == 0)
1318 dn_fib_select_multipath(&fl, &res);
1319
1320 /*
1321 * Check for out_dev == in_dev. We use the RTCF_DOREDIRECT
1322 * flag as a hint to set the intra-ethernet bit when
1323 * forwarding. If we've got NAT in operation, we don't do
1324 * this optimisation.
1325 */
1326 if (out_dev == in_dev && !(flags & RTCF_NAT))
1327 flags |= RTCF_DOREDIRECT;
1328
1329 local_src = DN_FIB_RES_PREFSRC(res);
1330
1331 case RTN_BLACKHOLE:
1332 case RTN_UNREACHABLE:
1333 break;
1334 case RTN_LOCAL:
1335 flags |= RTCF_LOCAL;
1336 fl.fld_src = cb->dst;
1337 fl.fld_dst = cb->src;
1338
1339 /* Routing tables gave us a gateway */
1340 if (gateway)
1341 goto make_route;
1342
1343 /* Packet was intra-ethernet, so we know its on-link */
1344 if (cb->rt_flags | DN_RT_F_IE) {
1345 gateway = cb->src;
1346 flags |= RTCF_DIRECTSRC;
1347 goto make_route;
1348 }
1349
1350 /* Use the default router if there is one */
1351 neigh = neigh_clone(dn_db->router);
1352 if (neigh) {
1353 gateway = ((struct dn_neigh *)neigh)->addr;
1354 goto make_route;
1355 }
1356
1357 /* Close eyes and pray */
1358 gateway = cb->src;
1359 flags |= RTCF_DIRECTSRC;
1360 goto make_route;
1361 default:
1362 goto e_inval;
1363 }
1364
1365 make_route:
1366 rt = dst_alloc(&dn_dst_ops);
1367 if (rt == NULL)
1368 goto e_nobufs;
1369
1370 rt->rt_saddr = fl.fld_src;
1371 rt->rt_daddr = fl.fld_dst;
1372 rt->rt_gateway = fl.fld_dst;
1373 if (gateway)
1374 rt->rt_gateway = gateway;
1375 rt->rt_local_src = local_src ? local_src : rt->rt_saddr;
1376
1377 rt->rt_dst_map = fl.fld_dst;
1378 rt->rt_src_map = fl.fld_src;
1379
1380 rt->fl.fld_src = cb->src;
1381 rt->fl.fld_dst = cb->dst;
1382 rt->fl.oif = 0;
1383 rt->fl.iif = in_dev->ifindex;
1384 rt->fl.fld_fwmark = fl.fld_fwmark;
1385
1386 rt->u.dst.flags = DST_HOST;
1387 rt->u.dst.neighbour = neigh;
1388 rt->u.dst.dev = out_dev;
1389 rt->u.dst.lastuse = jiffies;
1390 rt->u.dst.output = dn_rt_bug;
1391 switch(res.type) {
1392 case RTN_UNICAST:
1393 rt->u.dst.input = dn_forward;
1394 break;
1395 case RTN_LOCAL:
1396 rt->u.dst.output = dn_output;
1397 rt->u.dst.input = dn_nsp_rx;
1398 rt->u.dst.dev = in_dev;
1399 flags |= RTCF_LOCAL;
1400 break;
1401 default:
1402 case RTN_UNREACHABLE:
1403 case RTN_BLACKHOLE:
1404 rt->u.dst.input = dn_blackhole;
1405 }
1406 rt->rt_flags = flags;
1407 if (rt->u.dst.dev)
1408 dev_hold(rt->u.dst.dev);
1409
1410 err = dn_rt_set_next_hop(rt, &res);
1411 if (err)
1412 goto e_neighbour;
1413
1414 hash = dn_hash(rt->fl.fld_src, rt->fl.fld_dst);
1415 dn_insert_route(rt, hash, (struct dn_route **)&skb->dst);
1416
1417 done:
1418 if (neigh)
1419 neigh_release(neigh);
1420 if (free_res)
1421 dn_fib_res_put(&res);
1422 dev_put(in_dev);
1423 if (out_dev)
1424 dev_put(out_dev);
1425 out:
1426 return err;
1427
1428 e_inval:
1429 err = -EINVAL;
1430 goto done;
1431
1432 e_nobufs:
1433 err = -ENOBUFS;
1434 goto done;
1435
1436 e_neighbour:
1437 dst_free(&rt->u.dst);
1438 goto done;
1439 }
1440
1441 int dn_route_input(struct sk_buff *skb)
1442 {
1443 struct dn_route *rt;
1444 struct dn_skb_cb *cb = DN_SKB_CB(skb);
1445 unsigned hash = dn_hash(cb->src, cb->dst);
1446
1447 if (skb->dst)
1448 return 0;
1449
1450 rcu_read_lock();
1451 for(rt = rcu_dereference(dn_rt_hash_table[hash].chain); rt != NULL;
1452 rt = rcu_dereference(rt->u.rt_next)) {
1453 if ((rt->fl.fld_src == cb->src) &&
1454 (rt->fl.fld_dst == cb->dst) &&
1455 (rt->fl.oif == 0) &&
1456 #ifdef CONFIG_DECNET_ROUTE_FWMARK
1457 (rt->fl.fld_fwmark == skb->nfmark) &&
1458 #endif
1459 (rt->fl.iif == cb->iif)) {
1460 rt->u.dst.lastuse = jiffies;
1461 dst_hold(&rt->u.dst);
1462 rt->u.dst.__use++;
1463 rcu_read_unlock();
1464 skb->dst = (struct dst_entry *)rt;
1465 return 0;
1466 }
1467 }
1468 rcu_read_unlock();
1469
1470 return dn_route_input_slow(skb);
1471 }
1472
1473 static int dn_rt_fill_info(struct sk_buff *skb, u32 pid, u32 seq,
1474 int event, int nowait, unsigned int flags)
1475 {
1476 struct dn_route *rt = (struct dn_route *)skb->dst;
1477 struct rtmsg *r;
1478 struct nlmsghdr *nlh;
1479 unsigned char *b = skb->tail;
1480 struct rta_cacheinfo ci;
1481
1482 nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*r), flags);
1483 r = NLMSG_DATA(nlh);
1484 r->rtm_family = AF_DECnet;
1485 r->rtm_dst_len = 16;
1486 r->rtm_src_len = 0;
1487 r->rtm_tos = 0;
1488 r->rtm_table = RT_TABLE_MAIN;
1489 RTA_PUT_U32(skb, RTA_TABLE, RT_TABLE_MAIN);
1490 r->rtm_type = rt->rt_type;
1491 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
1492 r->rtm_scope = RT_SCOPE_UNIVERSE;
1493 r->rtm_protocol = RTPROT_UNSPEC;
1494 if (rt->rt_flags & RTCF_NOTIFY)
1495 r->rtm_flags |= RTM_F_NOTIFY;
1496 RTA_PUT(skb, RTA_DST, 2, &rt->rt_daddr);
1497 if (rt->fl.fld_src) {
1498 r->rtm_src_len = 16;
1499 RTA_PUT(skb, RTA_SRC, 2, &rt->fl.fld_src);
1500 }
1501 if (rt->u.dst.dev)
1502 RTA_PUT(skb, RTA_OIF, sizeof(int), &rt->u.dst.dev->ifindex);
1503 /*
1504 * Note to self - change this if input routes reverse direction when
1505 * they deal only with inputs and not with replies like they do
1506 * currently.
1507 */
1508 RTA_PUT(skb, RTA_PREFSRC, 2, &rt->rt_local_src);
1509 if (rt->rt_daddr != rt->rt_gateway)
1510 RTA_PUT(skb, RTA_GATEWAY, 2, &rt->rt_gateway);
1511 if (rtnetlink_put_metrics(skb, rt->u.dst.metrics) < 0)
1512 goto rtattr_failure;
1513 ci.rta_lastuse = jiffies_to_clock_t(jiffies - rt->u.dst.lastuse);
1514 ci.rta_used = rt->u.dst.__use;
1515 ci.rta_clntref = atomic_read(&rt->u.dst.__refcnt);
1516 if (rt->u.dst.expires)
1517 ci.rta_expires = jiffies_to_clock_t(rt->u.dst.expires - jiffies);
1518 else
1519 ci.rta_expires = 0;
1520 ci.rta_error = rt->u.dst.error;
1521 ci.rta_id = ci.rta_ts = ci.rta_tsage = 0;
1522 RTA_PUT(skb, RTA_CACHEINFO, sizeof(ci), &ci);
1523 if (rt->fl.iif)
1524 RTA_PUT(skb, RTA_IIF, sizeof(int), &rt->fl.iif);
1525
1526 nlh->nlmsg_len = skb->tail - b;
1527 return skb->len;
1528
1529 nlmsg_failure:
1530 rtattr_failure:
1531 skb_trim(skb, b - skb->data);
1532 return -1;
1533 }
1534
1535 /*
1536 * This is called by both endnodes and routers now.
1537 */
1538 int dn_cache_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh, void *arg)
1539 {
1540 struct rtattr **rta = arg;
1541 struct rtmsg *rtm = NLMSG_DATA(nlh);
1542 struct dn_route *rt = NULL;
1543 struct dn_skb_cb *cb;
1544 int err;
1545 struct sk_buff *skb;
1546 struct flowi fl;
1547
1548 memset(&fl, 0, sizeof(fl));
1549 fl.proto = DNPROTO_NSP;
1550
1551 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1552 if (skb == NULL)
1553 return -ENOBUFS;
1554 skb->mac.raw = skb->data;
1555 cb = DN_SKB_CB(skb);
1556
1557 if (rta[RTA_SRC-1])
1558 memcpy(&fl.fld_src, RTA_DATA(rta[RTA_SRC-1]), 2);
1559 if (rta[RTA_DST-1])
1560 memcpy(&fl.fld_dst, RTA_DATA(rta[RTA_DST-1]), 2);
1561 if (rta[RTA_IIF-1])
1562 memcpy(&fl.iif, RTA_DATA(rta[RTA_IIF-1]), sizeof(int));
1563
1564 if (fl.iif) {
1565 struct net_device *dev;
1566 if ((dev = dev_get_by_index(fl.iif)) == NULL) {
1567 kfree_skb(skb);
1568 return -ENODEV;
1569 }
1570 if (!dev->dn_ptr) {
1571 dev_put(dev);
1572 kfree_skb(skb);
1573 return -ENODEV;
1574 }
1575 skb->protocol = __constant_htons(ETH_P_DNA_RT);
1576 skb->dev = dev;
1577 cb->src = fl.fld_src;
1578 cb->dst = fl.fld_dst;
1579 local_bh_disable();
1580 err = dn_route_input(skb);
1581 local_bh_enable();
1582 memset(cb, 0, sizeof(struct dn_skb_cb));
1583 rt = (struct dn_route *)skb->dst;
1584 if (!err && -rt->u.dst.error)
1585 err = rt->u.dst.error;
1586 } else {
1587 int oif = 0;
1588 if (rta[RTA_OIF - 1])
1589 memcpy(&oif, RTA_DATA(rta[RTA_OIF - 1]), sizeof(int));
1590 fl.oif = oif;
1591 err = dn_route_output_key((struct dst_entry **)&rt, &fl, 0);
1592 }
1593
1594 if (skb->dev)
1595 dev_put(skb->dev);
1596 skb->dev = NULL;
1597 if (err)
1598 goto out_free;
1599 skb->dst = &rt->u.dst;
1600 if (rtm->rtm_flags & RTM_F_NOTIFY)
1601 rt->rt_flags |= RTCF_NOTIFY;
1602
1603 NETLINK_CB(skb).dst_pid = NETLINK_CB(in_skb).pid;
1604
1605 err = dn_rt_fill_info(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq, RTM_NEWROUTE, 0, 0);
1606
1607 if (err == 0)
1608 goto out_free;
1609 if (err < 0) {
1610 err = -EMSGSIZE;
1611 goto out_free;
1612 }
1613
1614 return rtnl_unicast(skb, NETLINK_CB(in_skb).pid);
1615
1616 out_free:
1617 kfree_skb(skb);
1618 return err;
1619 }
1620
1621 /*
1622 * For routers, this is called from dn_fib_dump, but for endnodes its
1623 * called directly from the rtnetlink dispatch table.
1624 */
1625 int dn_cache_dump(struct sk_buff *skb, struct netlink_callback *cb)
1626 {
1627 struct dn_route *rt;
1628 int h, s_h;
1629 int idx, s_idx;
1630
1631 if (NLMSG_PAYLOAD(cb->nlh, 0) < sizeof(struct rtmsg))
1632 return -EINVAL;
1633 if (!(((struct rtmsg *)NLMSG_DATA(cb->nlh))->rtm_flags&RTM_F_CLONED))
1634 return 0;
1635
1636 s_h = cb->args[0];
1637 s_idx = idx = cb->args[1];
1638 for(h = 0; h <= dn_rt_hash_mask; h++) {
1639 if (h < s_h)
1640 continue;
1641 if (h > s_h)
1642 s_idx = 0;
1643 rcu_read_lock_bh();
1644 for(rt = rcu_dereference(dn_rt_hash_table[h].chain), idx = 0;
1645 rt;
1646 rt = rcu_dereference(rt->u.rt_next), idx++) {
1647 if (idx < s_idx)
1648 continue;
1649 skb->dst = dst_clone(&rt->u.dst);
1650 if (dn_rt_fill_info(skb, NETLINK_CB(cb->skb).pid,
1651 cb->nlh->nlmsg_seq, RTM_NEWROUTE,
1652 1, NLM_F_MULTI) <= 0) {
1653 dst_release(xchg(&skb->dst, NULL));
1654 rcu_read_unlock_bh();
1655 goto done;
1656 }
1657 dst_release(xchg(&skb->dst, NULL));
1658 }
1659 rcu_read_unlock_bh();
1660 }
1661
1662 done:
1663 cb->args[0] = h;
1664 cb->args[1] = idx;
1665 return skb->len;
1666 }
1667
1668 #ifdef CONFIG_PROC_FS
1669 struct dn_rt_cache_iter_state {
1670 int bucket;
1671 };
1672
1673 static struct dn_route *dn_rt_cache_get_first(struct seq_file *seq)
1674 {
1675 struct dn_route *rt = NULL;
1676 struct dn_rt_cache_iter_state *s = seq->private;
1677
1678 for(s->bucket = dn_rt_hash_mask; s->bucket >= 0; --s->bucket) {
1679 rcu_read_lock_bh();
1680 rt = dn_rt_hash_table[s->bucket].chain;
1681 if (rt)
1682 break;
1683 rcu_read_unlock_bh();
1684 }
1685 return rt;
1686 }
1687
1688 static struct dn_route *dn_rt_cache_get_next(struct seq_file *seq, struct dn_route *rt)
1689 {
1690 struct dn_rt_cache_iter_state *s = rcu_dereference(seq->private);
1691
1692 rt = rt->u.rt_next;
1693 while(!rt) {
1694 rcu_read_unlock_bh();
1695 if (--s->bucket < 0)
1696 break;
1697 rcu_read_lock_bh();
1698 rt = dn_rt_hash_table[s->bucket].chain;
1699 }
1700 return rt;
1701 }
1702
1703 static void *dn_rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
1704 {
1705 struct dn_route *rt = dn_rt_cache_get_first(seq);
1706
1707 if (rt) {
1708 while(*pos && (rt = dn_rt_cache_get_next(seq, rt)))
1709 --*pos;
1710 }
1711 return *pos ? NULL : rt;
1712 }
1713
1714 static void *dn_rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1715 {
1716 struct dn_route *rt = dn_rt_cache_get_next(seq, v);
1717 ++*pos;
1718 return rt;
1719 }
1720
1721 static void dn_rt_cache_seq_stop(struct seq_file *seq, void *v)
1722 {
1723 if (v)
1724 rcu_read_unlock_bh();
1725 }
1726
1727 static int dn_rt_cache_seq_show(struct seq_file *seq, void *v)
1728 {
1729 struct dn_route *rt = v;
1730 char buf1[DN_ASCBUF_LEN], buf2[DN_ASCBUF_LEN];
1731
1732 seq_printf(seq, "%-8s %-7s %-7s %04d %04d %04d\n",
1733 rt->u.dst.dev ? rt->u.dst.dev->name : "*",
1734 dn_addr2asc(dn_ntohs(rt->rt_daddr), buf1),
1735 dn_addr2asc(dn_ntohs(rt->rt_saddr), buf2),
1736 atomic_read(&rt->u.dst.__refcnt),
1737 rt->u.dst.__use,
1738 (int) dst_metric(&rt->u.dst, RTAX_RTT));
1739 return 0;
1740 }
1741
1742 static struct seq_operations dn_rt_cache_seq_ops = {
1743 .start = dn_rt_cache_seq_start,
1744 .next = dn_rt_cache_seq_next,
1745 .stop = dn_rt_cache_seq_stop,
1746 .show = dn_rt_cache_seq_show,
1747 };
1748
1749 static int dn_rt_cache_seq_open(struct inode *inode, struct file *file)
1750 {
1751 struct seq_file *seq;
1752 int rc = -ENOMEM;
1753 struct dn_rt_cache_iter_state *s = kmalloc(sizeof(*s), GFP_KERNEL);
1754
1755 if (!s)
1756 goto out;
1757 rc = seq_open(file, &dn_rt_cache_seq_ops);
1758 if (rc)
1759 goto out_kfree;
1760 seq = file->private_data;
1761 seq->private = s;
1762 memset(s, 0, sizeof(*s));
1763 out:
1764 return rc;
1765 out_kfree:
1766 kfree(s);
1767 goto out;
1768 }
1769
1770 static struct file_operations dn_rt_cache_seq_fops = {
1771 .owner = THIS_MODULE,
1772 .open = dn_rt_cache_seq_open,
1773 .read = seq_read,
1774 .llseek = seq_lseek,
1775 .release = seq_release_private,
1776 };
1777
1778 #endif /* CONFIG_PROC_FS */
1779
1780 void __init dn_route_init(void)
1781 {
1782 int i, goal, order;
1783
1784 dn_dst_ops.kmem_cachep =
1785 kmem_cache_create("dn_dst_cache", sizeof(struct dn_route), 0,
1786 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
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
1833 void __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
hh.org GIT migration