?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
dm thin: wake worker when discard is prepared
[linux-2.6.git] / net / decnet / dn_route.c
blobb57419cc41a486b3ab2ddc82643169c97b19d132
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 <linux/slab.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 <linux/export.h>
81 #include <asm/errno.h>
82 #include <net/net_namespace.h>
83 #include <net/netlink.h>
84 #include <net/neighbour.h>
85 #include <net/dst.h>
86 #include <net/flow.h>
87 #include <net/fib_rules.h>
88 #include <net/dn.h>
89 #include <net/dn_dev.h>
90 #include <net/dn_nsp.h>
91 #include <net/dn_route.h>
92 #include <net/dn_neigh.h>
93 #include <net/dn_fib.h>
94
95 struct dn_rt_hash_bucket
96 {
97 struct dn_route __rcu *chain;
98 spinlock_t lock;
99 };
100
101 extern struct neigh_table dn_neigh_table;
102
103
104 static unsigned char dn_hiord_addr[6] = {0xAA,0x00,0x04,0x00,0x00,0x00};
105
106 static const int dn_rt_min_delay = 2 * HZ;
107 static const int dn_rt_max_delay = 10 * HZ;
108 static const int dn_rt_mtu_expires = 10 * 60 * HZ;
109
110 static unsigned long dn_rt_deadline;
111
112 static int dn_dst_gc(struct dst_ops *ops);
113 static struct dst_entry *dn_dst_check(struct dst_entry *, __u32);
114 static unsigned int dn_dst_default_advmss(const struct dst_entry *dst);
115 static unsigned int dn_dst_mtu(const struct dst_entry *dst);
116 static void dn_dst_destroy(struct dst_entry *);
117 static void dn_dst_ifdown(struct dst_entry *, struct net_device *dev, int how);
118 static struct dst_entry *dn_dst_negative_advice(struct dst_entry *);
119 static void dn_dst_link_failure(struct sk_buff *);
120 static void dn_dst_update_pmtu(struct dst_entry *dst, struct sock *sk,
121 struct sk_buff *skb , u32 mtu);
122 static void dn_dst_redirect(struct dst_entry *dst, struct sock *sk,
123 struct sk_buff *skb);
124 static struct neighbour *dn_dst_neigh_lookup(const struct dst_entry *dst,
125 struct sk_buff *skb,
126 const void *daddr);
127 static int dn_route_input(struct sk_buff *);
128 static void dn_run_flush(unsigned long dummy);
129
130 static struct dn_rt_hash_bucket *dn_rt_hash_table;
131 static unsigned int dn_rt_hash_mask;
132
133 static struct timer_list dn_route_timer;
134 static DEFINE_TIMER(dn_rt_flush_timer, dn_run_flush, 0, 0);
135 int decnet_dst_gc_interval = 2;
136
137 static struct dst_ops dn_dst_ops = {
138 .family = PF_DECnet,
139 .protocol = cpu_to_be16(ETH_P_DNA_RT),
140 .gc_thresh = 128,
141 .gc = dn_dst_gc,
142 .check = dn_dst_check,
143 .default_advmss = dn_dst_default_advmss,
144 .mtu = dn_dst_mtu,
145 .cow_metrics = dst_cow_metrics_generic,
146 .destroy = dn_dst_destroy,
147 .ifdown = dn_dst_ifdown,
148 .negative_advice = dn_dst_negative_advice,
149 .link_failure = dn_dst_link_failure,
150 .update_pmtu = dn_dst_update_pmtu,
151 .redirect = dn_dst_redirect,
152 .neigh_lookup = dn_dst_neigh_lookup,
153 };
154
155 static void dn_dst_destroy(struct dst_entry *dst)
156 {
157 struct dn_route *rt = (struct dn_route *) dst;
158
159 if (rt->n)
160 neigh_release(rt->n);
161 dst_destroy_metrics_generic(dst);
162 }
163
164 static void dn_dst_ifdown(struct dst_entry *dst, struct net_device *dev, int how)
165 {
166 if (how) {
167 struct dn_route *rt = (struct dn_route *) dst;
168 struct neighbour *n = rt->n;
169
170 if (n && n->dev == dev) {
171 n->dev = dev_net(dev)->loopback_dev;
172 dev_hold(n->dev);
173 dev_put(dev);
174 }
175 }
176 }
177
178 static __inline__ unsigned int dn_hash(__le16 src, __le16 dst)
179 {
180 __u16 tmp = (__u16 __force)(src ^ dst);
181 tmp ^= (tmp >> 3);
182 tmp ^= (tmp >> 5);
183 tmp ^= (tmp >> 10);
184 return dn_rt_hash_mask & (unsigned int)tmp;
185 }
186
187 static inline void dnrt_free(struct dn_route *rt)
188 {
189 call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
190 }
191
192 static inline void dnrt_drop(struct dn_route *rt)
193 {
194 dst_release(&rt->dst);
195 call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
196 }
197
198 static void dn_dst_check_expire(unsigned long dummy)
199 {
200 int i;
201 struct dn_route *rt;
202 struct dn_route __rcu **rtp;
203 unsigned long now = jiffies;
204 unsigned long expire = 120 * HZ;
205
206 for (i = 0; i <= dn_rt_hash_mask; i++) {
207 rtp = &dn_rt_hash_table[i].chain;
208
209 spin_lock(&dn_rt_hash_table[i].lock);
210 while ((rt = rcu_dereference_protected(*rtp,
211 lockdep_is_held(&dn_rt_hash_table[i].lock))) != NULL) {
212 if (atomic_read(&rt->dst.__refcnt) ||
213 (now - rt->dst.lastuse) < expire) {
214 rtp = &rt->dst.dn_next;
215 continue;
216 }
217 *rtp = rt->dst.dn_next;
218 rt->dst.dn_next = NULL;
219 dnrt_free(rt);
220 }
221 spin_unlock(&dn_rt_hash_table[i].lock);
222
223 if ((jiffies - now) > 0)
224 break;
225 }
226
227 mod_timer(&dn_route_timer, now + decnet_dst_gc_interval * HZ);
228 }
229
230 static int dn_dst_gc(struct dst_ops *ops)
231 {
232 struct dn_route *rt;
233 struct dn_route __rcu **rtp;
234 int i;
235 unsigned long now = jiffies;
236 unsigned long expire = 10 * HZ;
237
238 for (i = 0; i <= dn_rt_hash_mask; i++) {
239
240 spin_lock_bh(&dn_rt_hash_table[i].lock);
241 rtp = &dn_rt_hash_table[i].chain;
242
243 while ((rt = rcu_dereference_protected(*rtp,
244 lockdep_is_held(&dn_rt_hash_table[i].lock))) != NULL) {
245 if (atomic_read(&rt->dst.__refcnt) ||
246 (now - rt->dst.lastuse) < expire) {
247 rtp = &rt->dst.dn_next;
248 continue;
249 }
250 *rtp = rt->dst.dn_next;
251 rt->dst.dn_next = NULL;
252 dnrt_drop(rt);
253 break;
254 }
255 spin_unlock_bh(&dn_rt_hash_table[i].lock);
256 }
257
258 return 0;
259 }
260
261 /*
262 * The decnet standards don't impose a particular minimum mtu, what they
263 * do insist on is that the routing layer accepts a datagram of at least
264 * 230 bytes long. Here we have to subtract the routing header length from
265 * 230 to get the minimum acceptable mtu. If there is no neighbour, then we
266 * assume the worst and use a long header size.
267 *
268 * We update both the mtu and the advertised mss (i.e. the segment size we
269 * advertise to the other end).
270 */
271 static void dn_dst_update_pmtu(struct dst_entry *dst, struct sock *sk,
272 struct sk_buff *skb, u32 mtu)
273 {
274 struct dn_route *rt = (struct dn_route *) dst;
275 struct neighbour *n = rt->n;
276 u32 min_mtu = 230;
277 struct dn_dev *dn;
278
279 dn = n ? rcu_dereference_raw(n->dev->dn_ptr) : NULL;
280
281 if (dn && dn->use_long == 0)
282 min_mtu -= 6;
283 else
284 min_mtu -= 21;
285
286 if (dst_metric(dst, RTAX_MTU) > mtu && mtu >= min_mtu) {
287 if (!(dst_metric_locked(dst, RTAX_MTU))) {
288 dst_metric_set(dst, RTAX_MTU, mtu);
289 dst_set_expires(dst, dn_rt_mtu_expires);
290 }
291 if (!(dst_metric_locked(dst, RTAX_ADVMSS))) {
292 u32 mss = mtu - DN_MAX_NSP_DATA_HEADER;
293 u32 existing_mss = dst_metric_raw(dst, RTAX_ADVMSS);
294 if (!existing_mss || existing_mss > mss)
295 dst_metric_set(dst, RTAX_ADVMSS, mss);
296 }
297 }
298 }
299
300 static void dn_dst_redirect(struct dst_entry *dst, struct sock *sk,
301 struct sk_buff *skb)
302 {
303 }
304
305 /*
306 * When a route has been marked obsolete. (e.g. routing cache flush)
307 */
308 static struct dst_entry *dn_dst_check(struct dst_entry *dst, __u32 cookie)
309 {
310 return NULL;
311 }
312
313 static struct dst_entry *dn_dst_negative_advice(struct dst_entry *dst)
314 {
315 dst_release(dst);
316 return NULL;
317 }
318
319 static void dn_dst_link_failure(struct sk_buff *skb)
320 {
321 }
322
323 static inline int compare_keys(struct flowidn *fl1, struct flowidn *fl2)
324 {
325 return ((fl1->daddr ^ fl2->daddr) |
326 (fl1->saddr ^ fl2->saddr) |
327 (fl1->flowidn_mark ^ fl2->flowidn_mark) |
328 (fl1->flowidn_scope ^ fl2->flowidn_scope) |
329 (fl1->flowidn_oif ^ fl2->flowidn_oif) |
330 (fl1->flowidn_iif ^ fl2->flowidn_iif)) == 0;
331 }
332
333 static int dn_insert_route(struct dn_route *rt, unsigned int hash, struct dn_route **rp)
334 {
335 struct dn_route *rth;
336 struct dn_route __rcu **rthp;
337 unsigned long now = jiffies;
338
339 rthp = &dn_rt_hash_table[hash].chain;
340
341 spin_lock_bh(&dn_rt_hash_table[hash].lock);
342 while ((rth = rcu_dereference_protected(*rthp,
343 lockdep_is_held(&dn_rt_hash_table[hash].lock))) != NULL) {
344 if (compare_keys(&rth->fld, &rt->fld)) {
345 /* Put it first */
346 *rthp = rth->dst.dn_next;
347 rcu_assign_pointer(rth->dst.dn_next,
348 dn_rt_hash_table[hash].chain);
349 rcu_assign_pointer(dn_rt_hash_table[hash].chain, rth);
350
351 dst_use(&rth->dst, now);
352 spin_unlock_bh(&dn_rt_hash_table[hash].lock);
353
354 dnrt_drop(rt);
355 *rp = rth;
356 return 0;
357 }
358 rthp = &rth->dst.dn_next;
359 }
360
361 rcu_assign_pointer(rt->dst.dn_next, dn_rt_hash_table[hash].chain);
362 rcu_assign_pointer(dn_rt_hash_table[hash].chain, rt);
363
364 dst_use(&rt->dst, now);
365 spin_unlock_bh(&dn_rt_hash_table[hash].lock);
366 *rp = rt;
367 return 0;
368 }
369
370 static void dn_run_flush(unsigned long dummy)
371 {
372 int i;
373 struct dn_route *rt, *next;
374
375 for (i = 0; i < dn_rt_hash_mask; i++) {
376 spin_lock_bh(&dn_rt_hash_table[i].lock);
377
378 if ((rt = xchg((struct dn_route **)&dn_rt_hash_table[i].chain, NULL)) == NULL)
379 goto nothing_to_declare;
380
381 for(; rt; rt = next) {
382 next = rcu_dereference_raw(rt->dst.dn_next);
383 RCU_INIT_POINTER(rt->dst.dn_next, NULL);
384 dst_free((struct dst_entry *)rt);
385 }
386
387 nothing_to_declare:
388 spin_unlock_bh(&dn_rt_hash_table[i].lock);
389 }
390 }
391
392 static DEFINE_SPINLOCK(dn_rt_flush_lock);
393
394 void dn_rt_cache_flush(int delay)
395 {
396 unsigned long now = jiffies;
397 int user_mode = !in_interrupt();
398
399 if (delay < 0)
400 delay = dn_rt_min_delay;
401
402 spin_lock_bh(&dn_rt_flush_lock);
403
404 if (del_timer(&dn_rt_flush_timer) && delay > 0 && dn_rt_deadline) {
405 long tmo = (long)(dn_rt_deadline - now);
406
407 if (user_mode && tmo < dn_rt_max_delay - dn_rt_min_delay)
408 tmo = 0;
409
410 if (delay > tmo)
411 delay = tmo;
412 }
413
414 if (delay <= 0) {
415 spin_unlock_bh(&dn_rt_flush_lock);
416 dn_run_flush(0);
417 return;
418 }
419
420 if (dn_rt_deadline == 0)
421 dn_rt_deadline = now + dn_rt_max_delay;
422
423 dn_rt_flush_timer.expires = now + delay;
424 add_timer(&dn_rt_flush_timer);
425 spin_unlock_bh(&dn_rt_flush_lock);
426 }
427
428 /**
429 * dn_return_short - Return a short packet to its sender
430 * @skb: The packet to return
431 *
432 */
433 static int dn_return_short(struct sk_buff *skb)
434 {
435 struct dn_skb_cb *cb;
436 unsigned char *ptr;
437 __le16 *src;
438 __le16 *dst;
439
440 /* Add back headers */
441 skb_push(skb, skb->data - skb_network_header(skb));
442
443 if ((skb = skb_unshare(skb, GFP_ATOMIC)) == NULL)
444 return NET_RX_DROP;
445
446 cb = DN_SKB_CB(skb);
447 /* Skip packet length and point to flags */
448 ptr = skb->data + 2;
449 *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
450
451 dst = (__le16 *)ptr;
452 ptr += 2;
453 src = (__le16 *)ptr;
454 ptr += 2;
455 *ptr = 0; /* Zero hop count */
456
457 swap(*src, *dst);
458
459 skb->pkt_type = PACKET_OUTGOING;
460 dn_rt_finish_output(skb, NULL, NULL);
461 return NET_RX_SUCCESS;
462 }
463
464 /**
465 * dn_return_long - Return a long packet to its sender
466 * @skb: The long format packet to return
467 *
468 */
469 static int dn_return_long(struct sk_buff *skb)
470 {
471 struct dn_skb_cb *cb;
472 unsigned char *ptr;
473 unsigned char *src_addr, *dst_addr;
474 unsigned char tmp[ETH_ALEN];
475
476 /* Add back all headers */
477 skb_push(skb, skb->data - skb_network_header(skb));
478
479 if ((skb = skb_unshare(skb, GFP_ATOMIC)) == NULL)
480 return NET_RX_DROP;
481
482 cb = DN_SKB_CB(skb);
483 /* Ignore packet length and point to flags */
484 ptr = skb->data + 2;
485
486 /* Skip padding */
487 if (*ptr & DN_RT_F_PF) {
488 char padlen = (*ptr & ~DN_RT_F_PF);
489 ptr += padlen;
490 }
491
492 *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
493 ptr += 2;
494 dst_addr = ptr;
495 ptr += 8;
496 src_addr = ptr;
497 ptr += 6;
498 *ptr = 0; /* Zero hop count */
499
500 /* Swap source and destination */
501 memcpy(tmp, src_addr, ETH_ALEN);
502 memcpy(src_addr, dst_addr, ETH_ALEN);
503 memcpy(dst_addr, tmp, ETH_ALEN);
504
505 skb->pkt_type = PACKET_OUTGOING;
506 dn_rt_finish_output(skb, dst_addr, src_addr);
507 return NET_RX_SUCCESS;
508 }
509
510 /**
511 * dn_route_rx_packet - Try and find a route for an incoming packet
512 * @skb: The packet to find a route for
513 *
514 * Returns: result of input function if route is found, error code otherwise
515 */
516 static int dn_route_rx_packet(struct sk_buff *skb)
517 {
518 struct dn_skb_cb *cb;
519 int err;
520
521 if ((err = dn_route_input(skb)) == 0)
522 return dst_input(skb);
523
524 cb = DN_SKB_CB(skb);
525 if (decnet_debug_level & 4) {
526 char *devname = skb->dev ? skb->dev->name : "???";
527
528 printk(KERN_DEBUG
529 "DECnet: dn_route_rx_packet: rt_flags=0x%02x dev=%s len=%d src=0x%04hx dst=0x%04hx err=%d type=%d\n",
530 (int)cb->rt_flags, devname, skb->len,
531 le16_to_cpu(cb->src), le16_to_cpu(cb->dst),
532 err, skb->pkt_type);
533 }
534
535 if ((skb->pkt_type == PACKET_HOST) && (cb->rt_flags & DN_RT_F_RQR)) {
536 switch (cb->rt_flags & DN_RT_PKT_MSK) {
537 case DN_RT_PKT_SHORT:
538 return dn_return_short(skb);
539 case DN_RT_PKT_LONG:
540 return dn_return_long(skb);
541 }
542 }
543
544 kfree_skb(skb);
545 return NET_RX_DROP;
546 }
547
548 static int dn_route_rx_long(struct sk_buff *skb)
549 {
550 struct dn_skb_cb *cb = DN_SKB_CB(skb);
551 unsigned char *ptr = skb->data;
552
553 if (!pskb_may_pull(skb, 21)) /* 20 for long header, 1 for shortest nsp */
554 goto drop_it;
555
556 skb_pull(skb, 20);
557 skb_reset_transport_header(skb);
558
559 /* Destination info */
560 ptr += 2;
561 cb->dst = dn_eth2dn(ptr);
562 if (memcmp(ptr, dn_hiord_addr, 4) != 0)
563 goto drop_it;
564 ptr += 6;
565
566
567 /* Source info */
568 ptr += 2;
569 cb->src = dn_eth2dn(ptr);
570 if (memcmp(ptr, dn_hiord_addr, 4) != 0)
571 goto drop_it;
572 ptr += 6;
573 /* Other junk */
574 ptr++;
575 cb->hops = *ptr++; /* Visit Count */
576
577 return NF_HOOK(NFPROTO_DECNET, NF_DN_PRE_ROUTING, skb, skb->dev, NULL,
578 dn_route_rx_packet);
579
580 drop_it:
581 kfree_skb(skb);
582 return NET_RX_DROP;
583 }
584
585
586
587 static int dn_route_rx_short(struct sk_buff *skb)
588 {
589 struct dn_skb_cb *cb = DN_SKB_CB(skb);
590 unsigned char *ptr = skb->data;
591
592 if (!pskb_may_pull(skb, 6)) /* 5 for short header + 1 for shortest nsp */
593 goto drop_it;
594
595 skb_pull(skb, 5);
596 skb_reset_transport_header(skb);
597
598 cb->dst = *(__le16 *)ptr;
599 ptr += 2;
600 cb->src = *(__le16 *)ptr;
601 ptr += 2;
602 cb->hops = *ptr & 0x3f;
603
604 return NF_HOOK(NFPROTO_DECNET, NF_DN_PRE_ROUTING, skb, skb->dev, NULL,
605 dn_route_rx_packet);
606
607 drop_it:
608 kfree_skb(skb);
609 return NET_RX_DROP;
610 }
611
612 static int dn_route_discard(struct sk_buff *skb)
613 {
614 /*
615 * I know we drop the packet here, but thats considered success in
616 * this case
617 */
618 kfree_skb(skb);
619 return NET_RX_SUCCESS;
620 }
621
622 static int dn_route_ptp_hello(struct sk_buff *skb)
623 {
624 dn_dev_hello(skb);
625 dn_neigh_pointopoint_hello(skb);
626 return NET_RX_SUCCESS;
627 }
628
629 int dn_route_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
630 {
631 struct dn_skb_cb *cb;
632 unsigned char flags = 0;
633 __u16 len = le16_to_cpu(*(__le16 *)skb->data);
634 struct dn_dev *dn = rcu_dereference(dev->dn_ptr);
635 unsigned char padlen = 0;
636
637 if (!net_eq(dev_net(dev), &init_net))
638 goto dump_it;
639
640 if (dn == NULL)
641 goto dump_it;
642
643 if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
644 goto out;
645
646 if (!pskb_may_pull(skb, 3))
647 goto dump_it;
648
649 skb_pull(skb, 2);
650
651 if (len > skb->len)
652 goto dump_it;
653
654 skb_trim(skb, len);
655
656 flags = *skb->data;
657
658 cb = DN_SKB_CB(skb);
659 cb->stamp = jiffies;
660 cb->iif = dev->ifindex;
661
662 /*
663 * If we have padding, remove it.
664 */
665 if (flags & DN_RT_F_PF) {
666 padlen = flags & ~DN_RT_F_PF;
667 if (!pskb_may_pull(skb, padlen + 1))
668 goto dump_it;
669 skb_pull(skb, padlen);
670 flags = *skb->data;
671 }
672
673 skb_reset_network_header(skb);
674
675 /*
676 * Weed out future version DECnet
677 */
678 if (flags & DN_RT_F_VER)
679 goto dump_it;
680
681 cb->rt_flags = flags;
682
683 if (decnet_debug_level & 1)
684 printk(KERN_DEBUG
685 "dn_route_rcv: got 0x%02x from %s [%d %d %d]\n",
686 (int)flags, (dev) ? dev->name : "???", len, skb->len,
687 padlen);
688
689 if (flags & DN_RT_PKT_CNTL) {
690 if (unlikely(skb_linearize(skb)))
691 goto dump_it;
692
693 switch (flags & DN_RT_CNTL_MSK) {
694 case DN_RT_PKT_INIT:
695 dn_dev_init_pkt(skb);
696 break;
697 case DN_RT_PKT_VERI:
698 dn_dev_veri_pkt(skb);
699 break;
700 }
701
702 if (dn->parms.state != DN_DEV_S_RU)
703 goto dump_it;
704
705 switch (flags & DN_RT_CNTL_MSK) {
706 case DN_RT_PKT_HELO:
707 return NF_HOOK(NFPROTO_DECNET, NF_DN_HELLO,
708 skb, skb->dev, NULL,
709 dn_route_ptp_hello);
710
711 case DN_RT_PKT_L1RT:
712 case DN_RT_PKT_L2RT:
713 return NF_HOOK(NFPROTO_DECNET, NF_DN_ROUTE,
714 skb, skb->dev, NULL,
715 dn_route_discard);
716 case DN_RT_PKT_ERTH:
717 return NF_HOOK(NFPROTO_DECNET, NF_DN_HELLO,
718 skb, skb->dev, NULL,
719 dn_neigh_router_hello);
720
721 case DN_RT_PKT_EEDH:
722 return NF_HOOK(NFPROTO_DECNET, NF_DN_HELLO,
723 skb, skb->dev, NULL,
724 dn_neigh_endnode_hello);
725 }
726 } else {
727 if (dn->parms.state != DN_DEV_S_RU)
728 goto dump_it;
729
730 skb_pull(skb, 1); /* Pull flags */
731
732 switch (flags & DN_RT_PKT_MSK) {
733 case DN_RT_PKT_LONG:
734 return dn_route_rx_long(skb);
735 case DN_RT_PKT_SHORT:
736 return dn_route_rx_short(skb);
737 }
738 }
739
740 dump_it:
741 kfree_skb(skb);
742 out:
743 return NET_RX_DROP;
744 }
745
746 static int dn_to_neigh_output(struct sk_buff *skb)
747 {
748 struct dst_entry *dst = skb_dst(skb);
749 struct dn_route *rt = (struct dn_route *) dst;
750 struct neighbour *n = rt->n;
751
752 return n->output(n, skb);
753 }
754
755 static int dn_output(struct sk_buff *skb)
756 {
757 struct dst_entry *dst = skb_dst(skb);
758 struct dn_route *rt = (struct dn_route *)dst;
759 struct net_device *dev = dst->dev;
760 struct dn_skb_cb *cb = DN_SKB_CB(skb);
761
762 int err = -EINVAL;
763
764 if (rt->n == NULL)
765 goto error;
766
767 skb->dev = dev;
768
769 cb->src = rt->rt_saddr;
770 cb->dst = rt->rt_daddr;
771
772 /*
773 * Always set the Intra-Ethernet bit on all outgoing packets
774 * originated on this node. Only valid flag from upper layers
775 * is return-to-sender-requested. Set hop count to 0 too.
776 */
777 cb->rt_flags &= ~DN_RT_F_RQR;
778 cb->rt_flags |= DN_RT_F_IE;
779 cb->hops = 0;
780
781 return NF_HOOK(NFPROTO_DECNET, NF_DN_LOCAL_OUT, skb, NULL, dev,
782 dn_to_neigh_output);
783
784 error:
785 net_dbg_ratelimited("dn_output: This should not happen\n");
786
787 kfree_skb(skb);
788
789 return err;
790 }
791
792 static int dn_forward(struct sk_buff *skb)
793 {
794 struct dn_skb_cb *cb = DN_SKB_CB(skb);
795 struct dst_entry *dst = skb_dst(skb);
796 struct dn_dev *dn_db = rcu_dereference(dst->dev->dn_ptr);
797 struct dn_route *rt;
798 int header_len;
799 #ifdef CONFIG_NETFILTER
800 struct net_device *dev = skb->dev;
801 #endif
802
803 if (skb->pkt_type != PACKET_HOST)
804 goto drop;
805
806 /* Ensure that we have enough space for headers */
807 rt = (struct dn_route *)skb_dst(skb);
808 header_len = dn_db->use_long ? 21 : 6;
809 if (skb_cow(skb, LL_RESERVED_SPACE(rt->dst.dev)+header_len))
810 goto drop;
811
812 /*
813 * Hop count exceeded.
814 */
815 if (++cb->hops > 30)
816 goto drop;
817
818 skb->dev = rt->dst.dev;
819
820 /*
821 * If packet goes out same interface it came in on, then set
822 * the Intra-Ethernet bit. This has no effect for short
823 * packets, so we don't need to test for them here.
824 */
825 cb->rt_flags &= ~DN_RT_F_IE;
826 if (rt->rt_flags & RTCF_DOREDIRECT)
827 cb->rt_flags |= DN_RT_F_IE;
828
829 return NF_HOOK(NFPROTO_DECNET, NF_DN_FORWARD, skb, dev, skb->dev,
830 dn_to_neigh_output);
831
832 drop:
833 kfree_skb(skb);
834 return NET_RX_DROP;
835 }
836
837 /*
838 * Used to catch bugs. This should never normally get
839 * called.
840 */
841 static int dn_rt_bug(struct sk_buff *skb)
842 {
843 struct dn_skb_cb *cb = DN_SKB_CB(skb);
844
845 net_dbg_ratelimited("dn_rt_bug: skb from:%04x to:%04x\n",
846 le16_to_cpu(cb->src), le16_to_cpu(cb->dst));
847
848 kfree_skb(skb);
849
850 return NET_RX_DROP;
851 }
852
853 static unsigned int dn_dst_default_advmss(const struct dst_entry *dst)
854 {
855 return dn_mss_from_pmtu(dst->dev, dst_mtu(dst));
856 }
857
858 static unsigned int dn_dst_mtu(const struct dst_entry *dst)
859 {
860 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
861
862 return mtu ? : dst->dev->mtu;
863 }
864
865 static struct neighbour *dn_dst_neigh_lookup(const struct dst_entry *dst,
866 struct sk_buff *skb,
867 const void *daddr)
868 {
869 return __neigh_lookup_errno(&dn_neigh_table, daddr, dst->dev);
870 }
871
872 static int dn_rt_set_next_hop(struct dn_route *rt, struct dn_fib_res *res)
873 {
874 struct dn_fib_info *fi = res->fi;
875 struct net_device *dev = rt->dst.dev;
876 unsigned int mss_metric;
877 struct neighbour *n;
878
879 if (fi) {
880 if (DN_FIB_RES_GW(*res) &&
881 DN_FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
882 rt->rt_gateway = DN_FIB_RES_GW(*res);
883 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
884 }
885 rt->rt_type = res->type;
886
887 if (dev != NULL && rt->n == NULL) {
888 n = __neigh_lookup_errno(&dn_neigh_table, &rt->rt_gateway, dev);
889 if (IS_ERR(n))
890 return PTR_ERR(n);
891 rt->n = n;
892 }
893
894 if (dst_metric(&rt->dst, RTAX_MTU) > rt->dst.dev->mtu)
895 dst_metric_set(&rt->dst, RTAX_MTU, rt->dst.dev->mtu);
896 mss_metric = dst_metric_raw(&rt->dst, RTAX_ADVMSS);
897 if (mss_metric) {
898 unsigned int mss = dn_mss_from_pmtu(dev, dst_mtu(&rt->dst));
899 if (mss_metric > mss)
900 dst_metric_set(&rt->dst, RTAX_ADVMSS, mss);
901 }
902 return 0;
903 }
904
905 static inline int dn_match_addr(__le16 addr1, __le16 addr2)
906 {
907 __u16 tmp = le16_to_cpu(addr1) ^ le16_to_cpu(addr2);
908 int match = 16;
909 while(tmp) {
910 tmp >>= 1;
911 match--;
912 }
913 return match;
914 }
915
916 static __le16 dnet_select_source(const struct net_device *dev, __le16 daddr, int scope)
917 {
918 __le16 saddr = 0;
919 struct dn_dev *dn_db;
920 struct dn_ifaddr *ifa;
921 int best_match = 0;
922 int ret;
923
924 rcu_read_lock();
925 dn_db = rcu_dereference(dev->dn_ptr);
926 for (ifa = rcu_dereference(dn_db->ifa_list);
927 ifa != NULL;
928 ifa = rcu_dereference(ifa->ifa_next)) {
929 if (ifa->ifa_scope > scope)
930 continue;
931 if (!daddr) {
932 saddr = ifa->ifa_local;
933 break;
934 }
935 ret = dn_match_addr(daddr, ifa->ifa_local);
936 if (ret > best_match)
937 saddr = ifa->ifa_local;
938 if (best_match == 0)
939 saddr = ifa->ifa_local;
940 }
941 rcu_read_unlock();
942
943 return saddr;
944 }
945
946 static inline __le16 __dn_fib_res_prefsrc(struct dn_fib_res *res)
947 {
948 return dnet_select_source(DN_FIB_RES_DEV(*res), DN_FIB_RES_GW(*res), res->scope);
949 }
950
951 static inline __le16 dn_fib_rules_map_destination(__le16 daddr, struct dn_fib_res *res)
952 {
953 __le16 mask = dnet_make_mask(res->prefixlen);
954 return (daddr&~mask)|res->fi->fib_nh->nh_gw;
955 }
956
957 static int dn_route_output_slow(struct dst_entry **pprt, const struct flowidn *oldflp, int try_hard)
958 {
959 struct flowidn fld = {
960 .daddr = oldflp->daddr,
961 .saddr = oldflp->saddr,
962 .flowidn_scope = RT_SCOPE_UNIVERSE,
963 .flowidn_mark = oldflp->flowidn_mark,
964 .flowidn_iif = LOOPBACK_IFINDEX,
965 .flowidn_oif = oldflp->flowidn_oif,
966 };
967 struct dn_route *rt = NULL;
968 struct net_device *dev_out = NULL, *dev;
969 struct neighbour *neigh = NULL;
970 unsigned int hash;
971 unsigned int flags = 0;
972 struct dn_fib_res res = { .fi = NULL, .type = RTN_UNICAST };
973 int err;
974 int free_res = 0;
975 __le16 gateway = 0;
976
977 if (decnet_debug_level & 16)
978 printk(KERN_DEBUG
979 "dn_route_output_slow: dst=%04x src=%04x mark=%d"
980 " iif=%d oif=%d\n", le16_to_cpu(oldflp->daddr),
981 le16_to_cpu(oldflp->saddr),
982 oldflp->flowidn_mark, LOOPBACK_IFINDEX,
983 oldflp->flowidn_oif);
984
985 /* If we have an output interface, verify its a DECnet device */
986 if (oldflp->flowidn_oif) {
987 dev_out = dev_get_by_index(&init_net, oldflp->flowidn_oif);
988 err = -ENODEV;
989 if (dev_out && dev_out->dn_ptr == NULL) {
990 dev_put(dev_out);
991 dev_out = NULL;
992 }
993 if (dev_out == NULL)
994 goto out;
995 }
996
997 /* If we have a source address, verify that its a local address */
998 if (oldflp->saddr) {
999 err = -EADDRNOTAVAIL;
1000
1001 if (dev_out) {
1002 if (dn_dev_islocal(dev_out, oldflp->saddr))
1003 goto source_ok;
1004 dev_put(dev_out);
1005 goto out;
1006 }
1007 rcu_read_lock();
1008 for_each_netdev_rcu(&init_net, dev) {
1009 if (!dev->dn_ptr)
1010 continue;
1011 if (!dn_dev_islocal(dev, oldflp->saddr))
1012 continue;
1013 if ((dev->flags & IFF_LOOPBACK) &&
1014 oldflp->daddr &&
1015 !dn_dev_islocal(dev, oldflp->daddr))
1016 continue;
1017
1018 dev_out = dev;
1019 break;
1020 }
1021 rcu_read_unlock();
1022 if (dev_out == NULL)
1023 goto out;
1024 dev_hold(dev_out);
1025 source_ok:
1026 ;
1027 }
1028
1029 /* No destination? Assume its local */
1030 if (!fld.daddr) {
1031 fld.daddr = fld.saddr;
1032
1033 err = -EADDRNOTAVAIL;
1034 if (dev_out)
1035 dev_put(dev_out);
1036 dev_out = init_net.loopback_dev;
1037 dev_hold(dev_out);
1038 if (!fld.daddr) {
1039 fld.daddr =
1040 fld.saddr = dnet_select_source(dev_out, 0,
1041 RT_SCOPE_HOST);
1042 if (!fld.daddr)
1043 goto out;
1044 }
1045 fld.flowidn_oif = LOOPBACK_IFINDEX;
1046 res.type = RTN_LOCAL;
1047 goto make_route;
1048 }
1049
1050 if (decnet_debug_level & 16)
1051 printk(KERN_DEBUG
1052 "dn_route_output_slow: initial checks complete."
1053 " dst=%o4x src=%04x oif=%d try_hard=%d\n",
1054 le16_to_cpu(fld.daddr), le16_to_cpu(fld.saddr),
1055 fld.flowidn_oif, try_hard);
1056
1057 /*
1058 * N.B. If the kernel is compiled without router support then
1059 * dn_fib_lookup() will evaluate to non-zero so this if () block
1060 * will always be executed.
1061 */
1062 err = -ESRCH;
1063 if (try_hard || (err = dn_fib_lookup(&fld, &res)) != 0) {
1064 struct dn_dev *dn_db;
1065 if (err != -ESRCH)
1066 goto out;
1067 /*
1068 * Here the fallback is basically the standard algorithm for
1069 * routing in endnodes which is described in the DECnet routing
1070 * docs
1071 *
1072 * If we are not trying hard, look in neighbour cache.
1073 * The result is tested to ensure that if a specific output
1074 * device/source address was requested, then we honour that
1075 * here
1076 */
1077 if (!try_hard) {
1078 neigh = neigh_lookup_nodev(&dn_neigh_table, &init_net, &fld.daddr);
1079 if (neigh) {
1080 if ((oldflp->flowidn_oif &&
1081 (neigh->dev->ifindex != oldflp->flowidn_oif)) ||
1082 (oldflp->saddr &&
1083 (!dn_dev_islocal(neigh->dev,
1084 oldflp->saddr)))) {
1085 neigh_release(neigh);
1086 neigh = NULL;
1087 } else {
1088 if (dev_out)
1089 dev_put(dev_out);
1090 if (dn_dev_islocal(neigh->dev, fld.daddr)) {
1091 dev_out = init_net.loopback_dev;
1092 res.type = RTN_LOCAL;
1093 } else {
1094 dev_out = neigh->dev;
1095 }
1096 dev_hold(dev_out);
1097 goto select_source;
1098 }
1099 }
1100 }
1101
1102 /* Not there? Perhaps its a local address */
1103 if (dev_out == NULL)
1104 dev_out = dn_dev_get_default();
1105 err = -ENODEV;
1106 if (dev_out == NULL)
1107 goto out;
1108 dn_db = rcu_dereference_raw(dev_out->dn_ptr);
1109 /* Possible improvement - check all devices for local addr */
1110 if (dn_dev_islocal(dev_out, fld.daddr)) {
1111 dev_put(dev_out);
1112 dev_out = init_net.loopback_dev;
1113 dev_hold(dev_out);
1114 res.type = RTN_LOCAL;
1115 goto select_source;
1116 }
1117 /* Not local either.... try sending it to the default router */
1118 neigh = neigh_clone(dn_db->router);
1119 BUG_ON(neigh && neigh->dev != dev_out);
1120
1121 /* Ok then, we assume its directly connected and move on */
1122 select_source:
1123 if (neigh)
1124 gateway = ((struct dn_neigh *)neigh)->addr;
1125 if (gateway == 0)
1126 gateway = fld.daddr;
1127 if (fld.saddr == 0) {
1128 fld.saddr = dnet_select_source(dev_out, gateway,
1129 res.type == RTN_LOCAL ?
1130 RT_SCOPE_HOST :
1131 RT_SCOPE_LINK);
1132 if (fld.saddr == 0 && res.type != RTN_LOCAL)
1133 goto e_addr;
1134 }
1135 fld.flowidn_oif = dev_out->ifindex;
1136 goto make_route;
1137 }
1138 free_res = 1;
1139
1140 if (res.type == RTN_NAT)
1141 goto e_inval;
1142
1143 if (res.type == RTN_LOCAL) {
1144 if (!fld.saddr)
1145 fld.saddr = fld.daddr;
1146 if (dev_out)
1147 dev_put(dev_out);
1148 dev_out = init_net.loopback_dev;
1149 dev_hold(dev_out);
1150 fld.flowidn_oif = dev_out->ifindex;
1151 if (res.fi)
1152 dn_fib_info_put(res.fi);
1153 res.fi = NULL;
1154 goto make_route;
1155 }
1156
1157 if (res.fi->fib_nhs > 1 && fld.flowidn_oif == 0)
1158 dn_fib_select_multipath(&fld, &res);
1159
1160 /*
1161 * We could add some logic to deal with default routes here and
1162 * get rid of some of the special casing above.
1163 */
1164
1165 if (!fld.saddr)
1166 fld.saddr = DN_FIB_RES_PREFSRC(res);
1167
1168 if (dev_out)
1169 dev_put(dev_out);
1170 dev_out = DN_FIB_RES_DEV(res);
1171 dev_hold(dev_out);
1172 fld.flowidn_oif = dev_out->ifindex;
1173 gateway = DN_FIB_RES_GW(res);
1174
1175 make_route:
1176 if (dev_out->flags & IFF_LOOPBACK)
1177 flags |= RTCF_LOCAL;
1178
1179 rt = dst_alloc(&dn_dst_ops, dev_out, 1, DST_OBSOLETE_NONE, DST_HOST);
1180 if (rt == NULL)
1181 goto e_nobufs;
1182
1183 memset(&rt->fld, 0, sizeof(rt->fld));
1184 rt->fld.saddr = oldflp->saddr;
1185 rt->fld.daddr = oldflp->daddr;
1186 rt->fld.flowidn_oif = oldflp->flowidn_oif;
1187 rt->fld.flowidn_iif = 0;
1188 rt->fld.flowidn_mark = oldflp->flowidn_mark;
1189
1190 rt->rt_saddr = fld.saddr;
1191 rt->rt_daddr = fld.daddr;
1192 rt->rt_gateway = gateway ? gateway : fld.daddr;
1193 rt->rt_local_src = fld.saddr;
1194
1195 rt->rt_dst_map = fld.daddr;
1196 rt->rt_src_map = fld.saddr;
1197
1198 rt->n = neigh;
1199 neigh = NULL;
1200
1201 rt->dst.lastuse = jiffies;
1202 rt->dst.output = dn_output;
1203 rt->dst.input = dn_rt_bug;
1204 rt->rt_flags = flags;
1205 if (flags & RTCF_LOCAL)
1206 rt->dst.input = dn_nsp_rx;
1207
1208 err = dn_rt_set_next_hop(rt, &res);
1209 if (err)
1210 goto e_neighbour;
1211
1212 hash = dn_hash(rt->fld.saddr, rt->fld.daddr);
1213 dn_insert_route(rt, hash, (struct dn_route **)pprt);
1214
1215 done:
1216 if (neigh)
1217 neigh_release(neigh);
1218 if (free_res)
1219 dn_fib_res_put(&res);
1220 if (dev_out)
1221 dev_put(dev_out);
1222 out:
1223 return err;
1224
1225 e_addr:
1226 err = -EADDRNOTAVAIL;
1227 goto done;
1228 e_inval:
1229 err = -EINVAL;
1230 goto done;
1231 e_nobufs:
1232 err = -ENOBUFS;
1233 goto done;
1234 e_neighbour:
1235 dst_free(&rt->dst);
1236 goto e_nobufs;
1237 }
1238
1239
1240 /*
1241 * N.B. The flags may be moved into the flowi at some future stage.
1242 */
1243 static int __dn_route_output_key(struct dst_entry **pprt, const struct flowidn *flp, int flags)
1244 {
1245 unsigned int hash = dn_hash(flp->saddr, flp->daddr);
1246 struct dn_route *rt = NULL;
1247
1248 if (!(flags & MSG_TRYHARD)) {
1249 rcu_read_lock_bh();
1250 for (rt = rcu_dereference_bh(dn_rt_hash_table[hash].chain); rt;
1251 rt = rcu_dereference_bh(rt->dst.dn_next)) {
1252 if ((flp->daddr == rt->fld.daddr) &&
1253 (flp->saddr == rt->fld.saddr) &&
1254 (flp->flowidn_mark == rt->fld.flowidn_mark) &&
1255 dn_is_output_route(rt) &&
1256 (rt->fld.flowidn_oif == flp->flowidn_oif)) {
1257 dst_use(&rt->dst, jiffies);
1258 rcu_read_unlock_bh();
1259 *pprt = &rt->dst;
1260 return 0;
1261 }
1262 }
1263 rcu_read_unlock_bh();
1264 }
1265
1266 return dn_route_output_slow(pprt, flp, flags);
1267 }
1268
1269 static int dn_route_output_key(struct dst_entry **pprt, struct flowidn *flp, int flags)
1270 {
1271 int err;
1272
1273 err = __dn_route_output_key(pprt, flp, flags);
1274 if (err == 0 && flp->flowidn_proto) {
1275 *pprt = xfrm_lookup(&init_net, *pprt,
1276 flowidn_to_flowi(flp), NULL, 0);
1277 if (IS_ERR(*pprt)) {
1278 err = PTR_ERR(*pprt);
1279 *pprt = NULL;
1280 }
1281 }
1282 return err;
1283 }
1284
1285 int dn_route_output_sock(struct dst_entry **pprt, struct flowidn *fl, struct sock *sk, int flags)
1286 {
1287 int err;
1288
1289 err = __dn_route_output_key(pprt, fl, flags & MSG_TRYHARD);
1290 if (err == 0 && fl->flowidn_proto) {
1291 if (!(flags & MSG_DONTWAIT))
1292 fl->flowidn_flags |= FLOWI_FLAG_CAN_SLEEP;
1293 *pprt = xfrm_lookup(&init_net, *pprt,
1294 flowidn_to_flowi(fl), sk, 0);
1295 if (IS_ERR(*pprt)) {
1296 err = PTR_ERR(*pprt);
1297 *pprt = NULL;
1298 }
1299 }
1300 return err;
1301 }
1302
1303 static int dn_route_input_slow(struct sk_buff *skb)
1304 {
1305 struct dn_route *rt = NULL;
1306 struct dn_skb_cb *cb = DN_SKB_CB(skb);
1307 struct net_device *in_dev = skb->dev;
1308 struct net_device *out_dev = NULL;
1309 struct dn_dev *dn_db;
1310 struct neighbour *neigh = NULL;
1311 unsigned int hash;
1312 int flags = 0;
1313 __le16 gateway = 0;
1314 __le16 local_src = 0;
1315 struct flowidn fld = {
1316 .daddr = cb->dst,
1317 .saddr = cb->src,
1318 .flowidn_scope = RT_SCOPE_UNIVERSE,
1319 .flowidn_mark = skb->mark,
1320 .flowidn_iif = skb->dev->ifindex,
1321 };
1322 struct dn_fib_res res = { .fi = NULL, .type = RTN_UNREACHABLE };
1323 int err = -EINVAL;
1324 int free_res = 0;
1325
1326 dev_hold(in_dev);
1327
1328 if ((dn_db = rcu_dereference(in_dev->dn_ptr)) == NULL)
1329 goto out;
1330
1331 /* Zero source addresses are not allowed */
1332 if (fld.saddr == 0)
1333 goto out;
1334
1335 /*
1336 * In this case we've just received a packet from a source
1337 * outside ourselves pretending to come from us. We don't
1338 * allow it any further to prevent routing loops, spoofing and
1339 * other nasties. Loopback packets already have the dst attached
1340 * so this only affects packets which have originated elsewhere.
1341 */
1342 err = -ENOTUNIQ;
1343 if (dn_dev_islocal(in_dev, cb->src))
1344 goto out;
1345
1346 err = dn_fib_lookup(&fld, &res);
1347 if (err) {
1348 if (err != -ESRCH)
1349 goto out;
1350 /*
1351 * Is the destination us ?
1352 */
1353 if (!dn_dev_islocal(in_dev, cb->dst))
1354 goto e_inval;
1355
1356 res.type = RTN_LOCAL;
1357 } else {
1358 __le16 src_map = fld.saddr;
1359 free_res = 1;
1360
1361 out_dev = DN_FIB_RES_DEV(res);
1362 if (out_dev == NULL) {
1363 net_crit_ratelimited("Bug in dn_route_input_slow() No output device\n");
1364 goto e_inval;
1365 }
1366 dev_hold(out_dev);
1367
1368 if (res.r)
1369 src_map = fld.saddr; /* no NAT support for now */
1370
1371 gateway = DN_FIB_RES_GW(res);
1372 if (res.type == RTN_NAT) {
1373 fld.daddr = dn_fib_rules_map_destination(fld.daddr, &res);
1374 dn_fib_res_put(&res);
1375 free_res = 0;
1376 if (dn_fib_lookup(&fld, &res))
1377 goto e_inval;
1378 free_res = 1;
1379 if (res.type != RTN_UNICAST)
1380 goto e_inval;
1381 flags |= RTCF_DNAT;
1382 gateway = fld.daddr;
1383 }
1384 fld.saddr = src_map;
1385 }
1386
1387 switch(res.type) {
1388 case RTN_UNICAST:
1389 /*
1390 * Forwarding check here, we only check for forwarding
1391 * being turned off, if you want to only forward intra
1392 * area, its up to you to set the routing tables up
1393 * correctly.
1394 */
1395 if (dn_db->parms.forwarding == 0)
1396 goto e_inval;
1397
1398 if (res.fi->fib_nhs > 1 && fld.flowidn_oif == 0)
1399 dn_fib_select_multipath(&fld, &res);
1400
1401 /*
1402 * Check for out_dev == in_dev. We use the RTCF_DOREDIRECT
1403 * flag as a hint to set the intra-ethernet bit when
1404 * forwarding. If we've got NAT in operation, we don't do
1405 * this optimisation.
1406 */
1407 if (out_dev == in_dev && !(flags & RTCF_NAT))
1408 flags |= RTCF_DOREDIRECT;
1409
1410 local_src = DN_FIB_RES_PREFSRC(res);
1411
1412 case RTN_BLACKHOLE:
1413 case RTN_UNREACHABLE:
1414 break;
1415 case RTN_LOCAL:
1416 flags |= RTCF_LOCAL;
1417 fld.saddr = cb->dst;
1418 fld.daddr = cb->src;
1419
1420 /* Routing tables gave us a gateway */
1421 if (gateway)
1422 goto make_route;
1423
1424 /* Packet was intra-ethernet, so we know its on-link */
1425 if (cb->rt_flags & DN_RT_F_IE) {
1426 gateway = cb->src;
1427 goto make_route;
1428 }
1429
1430 /* Use the default router if there is one */
1431 neigh = neigh_clone(dn_db->router);
1432 if (neigh) {
1433 gateway = ((struct dn_neigh *)neigh)->addr;
1434 goto make_route;
1435 }
1436
1437 /* Close eyes and pray */
1438 gateway = cb->src;
1439 goto make_route;
1440 default:
1441 goto e_inval;
1442 }
1443
1444 make_route:
1445 rt = dst_alloc(&dn_dst_ops, out_dev, 0, DST_OBSOLETE_NONE, DST_HOST);
1446 if (rt == NULL)
1447 goto e_nobufs;
1448
1449 memset(&rt->fld, 0, sizeof(rt->fld));
1450 rt->rt_saddr = fld.saddr;
1451 rt->rt_daddr = fld.daddr;
1452 rt->rt_gateway = fld.daddr;
1453 if (gateway)
1454 rt->rt_gateway = gateway;
1455 rt->rt_local_src = local_src ? local_src : rt->rt_saddr;
1456
1457 rt->rt_dst_map = fld.daddr;
1458 rt->rt_src_map = fld.saddr;
1459
1460 rt->fld.saddr = cb->src;
1461 rt->fld.daddr = cb->dst;
1462 rt->fld.flowidn_oif = 0;
1463 rt->fld.flowidn_iif = in_dev->ifindex;
1464 rt->fld.flowidn_mark = fld.flowidn_mark;
1465
1466 rt->n = neigh;
1467 rt->dst.lastuse = jiffies;
1468 rt->dst.output = dn_rt_bug;
1469 switch (res.type) {
1470 case RTN_UNICAST:
1471 rt->dst.input = dn_forward;
1472 break;
1473 case RTN_LOCAL:
1474 rt->dst.output = dn_output;
1475 rt->dst.input = dn_nsp_rx;
1476 rt->dst.dev = in_dev;
1477 flags |= RTCF_LOCAL;
1478 break;
1479 default:
1480 case RTN_UNREACHABLE:
1481 case RTN_BLACKHOLE:
1482 rt->dst.input = dst_discard;
1483 }
1484 rt->rt_flags = flags;
1485
1486 err = dn_rt_set_next_hop(rt, &res);
1487 if (err)
1488 goto e_neighbour;
1489
1490 hash = dn_hash(rt->fld.saddr, rt->fld.daddr);
1491 dn_insert_route(rt, hash, &rt);
1492 skb_dst_set(skb, &rt->dst);
1493
1494 done:
1495 if (neigh)
1496 neigh_release(neigh);
1497 if (free_res)
1498 dn_fib_res_put(&res);
1499 dev_put(in_dev);
1500 if (out_dev)
1501 dev_put(out_dev);
1502 out:
1503 return err;
1504
1505 e_inval:
1506 err = -EINVAL;
1507 goto done;
1508
1509 e_nobufs:
1510 err = -ENOBUFS;
1511 goto done;
1512
1513 e_neighbour:
1514 dst_free(&rt->dst);
1515 goto done;
1516 }
1517
1518 static int dn_route_input(struct sk_buff *skb)
1519 {
1520 struct dn_route *rt;
1521 struct dn_skb_cb *cb = DN_SKB_CB(skb);
1522 unsigned int hash = dn_hash(cb->src, cb->dst);
1523
1524 if (skb_dst(skb))
1525 return 0;
1526
1527 rcu_read_lock();
1528 for(rt = rcu_dereference(dn_rt_hash_table[hash].chain); rt != NULL;
1529 rt = rcu_dereference(rt->dst.dn_next)) {
1530 if ((rt->fld.saddr == cb->src) &&
1531 (rt->fld.daddr == cb->dst) &&
1532 (rt->fld.flowidn_oif == 0) &&
1533 (rt->fld.flowidn_mark == skb->mark) &&
1534 (rt->fld.flowidn_iif == cb->iif)) {
1535 dst_use(&rt->dst, jiffies);
1536 rcu_read_unlock();
1537 skb_dst_set(skb, (struct dst_entry *)rt);
1538 return 0;
1539 }
1540 }
1541 rcu_read_unlock();
1542
1543 return dn_route_input_slow(skb);
1544 }
1545
1546 static int dn_rt_fill_info(struct sk_buff *skb, u32 portid, u32 seq,
1547 int event, int nowait, unsigned int flags)
1548 {
1549 struct dn_route *rt = (struct dn_route *)skb_dst(skb);
1550 struct rtmsg *r;
1551 struct nlmsghdr *nlh;
1552 long expires;
1553
1554 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*r), flags);
1555 if (!nlh)
1556 return -EMSGSIZE;
1557
1558 r = nlmsg_data(nlh);
1559 r->rtm_family = AF_DECnet;
1560 r->rtm_dst_len = 16;
1561 r->rtm_src_len = 0;
1562 r->rtm_tos = 0;
1563 r->rtm_table = RT_TABLE_MAIN;
1564 r->rtm_type = rt->rt_type;
1565 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
1566 r->rtm_scope = RT_SCOPE_UNIVERSE;
1567 r->rtm_protocol = RTPROT_UNSPEC;
1568
1569 if (rt->rt_flags & RTCF_NOTIFY)
1570 r->rtm_flags |= RTM_F_NOTIFY;
1571
1572 if (nla_put_u32(skb, RTA_TABLE, RT_TABLE_MAIN) < 0 ||
1573 nla_put_le16(skb, RTA_DST, rt->rt_daddr) < 0)
1574 goto errout;
1575
1576 if (rt->fld.saddr) {
1577 r->rtm_src_len = 16;
1578 if (nla_put_le16(skb, RTA_SRC, rt->fld.saddr) < 0)
1579 goto errout;
1580 }
1581 if (rt->dst.dev &&
1582 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex) < 0)
1583 goto errout;
1584
1585 /*
1586 * Note to self - change this if input routes reverse direction when
1587 * they deal only with inputs and not with replies like they do
1588 * currently.
1589 */
1590 if (nla_put_le16(skb, RTA_PREFSRC, rt->rt_local_src) < 0)
1591 goto errout;
1592
1593 if (rt->rt_daddr != rt->rt_gateway &&
1594 nla_put_le16(skb, RTA_GATEWAY, rt->rt_gateway) < 0)
1595 goto errout;
1596
1597 if (rtnetlink_put_metrics(skb, dst_metrics_ptr(&rt->dst)) < 0)
1598 goto errout;
1599
1600 expires = rt->dst.expires ? rt->dst.expires - jiffies : 0;
1601 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires,
1602 rt->dst.error) < 0)
1603 goto errout;
1604
1605 if (dn_is_input_route(rt) &&
1606 nla_put_u32(skb, RTA_IIF, rt->fld.flowidn_iif) < 0)
1607 goto errout;
1608
1609 return nlmsg_end(skb, nlh);
1610
1611 errout:
1612 nlmsg_cancel(skb, nlh);
1613 return -EMSGSIZE;
1614 }
1615
1616 /*
1617 * This is called by both endnodes and routers now.
1618 */
1619 static int dn_cache_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh, void *arg)
1620 {
1621 struct net *net = sock_net(in_skb->sk);
1622 struct rtattr **rta = arg;
1623 struct rtmsg *rtm = nlmsg_data(nlh);
1624 struct dn_route *rt = NULL;
1625 struct dn_skb_cb *cb;
1626 int err;
1627 struct sk_buff *skb;
1628 struct flowidn fld;
1629
1630 if (!net_eq(net, &init_net))
1631 return -EINVAL;
1632
1633 memset(&fld, 0, sizeof(fld));
1634 fld.flowidn_proto = DNPROTO_NSP;
1635
1636 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1637 if (skb == NULL)
1638 return -ENOBUFS;
1639 skb_reset_mac_header(skb);
1640 cb = DN_SKB_CB(skb);
1641
1642 if (rta[RTA_SRC-1])
1643 memcpy(&fld.saddr, RTA_DATA(rta[RTA_SRC-1]), 2);
1644 if (rta[RTA_DST-1])
1645 memcpy(&fld.daddr, RTA_DATA(rta[RTA_DST-1]), 2);
1646 if (rta[RTA_IIF-1])
1647 memcpy(&fld.flowidn_iif, RTA_DATA(rta[RTA_IIF-1]), sizeof(int));
1648
1649 if (fld.flowidn_iif) {
1650 struct net_device *dev;
1651 if ((dev = dev_get_by_index(&init_net, fld.flowidn_iif)) == NULL) {
1652 kfree_skb(skb);
1653 return -ENODEV;
1654 }
1655 if (!dev->dn_ptr) {
1656 dev_put(dev);
1657 kfree_skb(skb);
1658 return -ENODEV;
1659 }
1660 skb->protocol = htons(ETH_P_DNA_RT);
1661 skb->dev = dev;
1662 cb->src = fld.saddr;
1663 cb->dst = fld.daddr;
1664 local_bh_disable();
1665 err = dn_route_input(skb);
1666 local_bh_enable();
1667 memset(cb, 0, sizeof(struct dn_skb_cb));
1668 rt = (struct dn_route *)skb_dst(skb);
1669 if (!err && -rt->dst.error)
1670 err = rt->dst.error;
1671 } else {
1672 int oif = 0;
1673 if (rta[RTA_OIF - 1])
1674 memcpy(&oif, RTA_DATA(rta[RTA_OIF - 1]), sizeof(int));
1675 fld.flowidn_oif = oif;
1676 err = dn_route_output_key((struct dst_entry **)&rt, &fld, 0);
1677 }
1678
1679 if (skb->dev)
1680 dev_put(skb->dev);
1681 skb->dev = NULL;
1682 if (err)
1683 goto out_free;
1684 skb_dst_set(skb, &rt->dst);
1685 if (rtm->rtm_flags & RTM_F_NOTIFY)
1686 rt->rt_flags |= RTCF_NOTIFY;
1687
1688 err = dn_rt_fill_info(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq, RTM_NEWROUTE, 0, 0);
1689
1690 if (err == 0)
1691 goto out_free;
1692 if (err < 0) {
1693 err = -EMSGSIZE;
1694 goto out_free;
1695 }
1696
1697 return rtnl_unicast(skb, &init_net, NETLINK_CB(in_skb).portid);
1698
1699 out_free:
1700 kfree_skb(skb);
1701 return err;
1702 }
1703
1704 /*
1705 * For routers, this is called from dn_fib_dump, but for endnodes its
1706 * called directly from the rtnetlink dispatch table.
1707 */
1708 int dn_cache_dump(struct sk_buff *skb, struct netlink_callback *cb)
1709 {
1710 struct net *net = sock_net(skb->sk);
1711 struct dn_route *rt;
1712 int h, s_h;
1713 int idx, s_idx;
1714 struct rtmsg *rtm;
1715
1716 if (!net_eq(net, &init_net))
1717 return 0;
1718
1719 if (nlmsg_len(cb->nlh) < sizeof(struct rtmsg))
1720 return -EINVAL;
1721
1722 rtm = nlmsg_data(cb->nlh);
1723 if (!(rtm->rtm_flags & RTM_F_CLONED))
1724 return 0;
1725
1726 s_h = cb->args[0];
1727 s_idx = idx = cb->args[1];
1728 for(h = 0; h <= dn_rt_hash_mask; h++) {
1729 if (h < s_h)
1730 continue;
1731 if (h > s_h)
1732 s_idx = 0;
1733 rcu_read_lock_bh();
1734 for(rt = rcu_dereference_bh(dn_rt_hash_table[h].chain), idx = 0;
1735 rt;
1736 rt = rcu_dereference_bh(rt->dst.dn_next), idx++) {
1737 if (idx < s_idx)
1738 continue;
1739 skb_dst_set(skb, dst_clone(&rt->dst));
1740 if (dn_rt_fill_info(skb, NETLINK_CB(cb->skb).portid,
1741 cb->nlh->nlmsg_seq, RTM_NEWROUTE,
1742 1, NLM_F_MULTI) <= 0) {
1743 skb_dst_drop(skb);
1744 rcu_read_unlock_bh();
1745 goto done;
1746 }
1747 skb_dst_drop(skb);
1748 }
1749 rcu_read_unlock_bh();
1750 }
1751
1752 done:
1753 cb->args[0] = h;
1754 cb->args[1] = idx;
1755 return skb->len;
1756 }
1757
1758 #ifdef CONFIG_PROC_FS
1759 struct dn_rt_cache_iter_state {
1760 int bucket;
1761 };
1762
1763 static struct dn_route *dn_rt_cache_get_first(struct seq_file *seq)
1764 {
1765 struct dn_route *rt = NULL;
1766 struct dn_rt_cache_iter_state *s = seq->private;
1767
1768 for(s->bucket = dn_rt_hash_mask; s->bucket >= 0; --s->bucket) {
1769 rcu_read_lock_bh();
1770 rt = rcu_dereference_bh(dn_rt_hash_table[s->bucket].chain);
1771 if (rt)
1772 break;
1773 rcu_read_unlock_bh();
1774 }
1775 return rt;
1776 }
1777
1778 static struct dn_route *dn_rt_cache_get_next(struct seq_file *seq, struct dn_route *rt)
1779 {
1780 struct dn_rt_cache_iter_state *s = seq->private;
1781
1782 rt = rcu_dereference_bh(rt->dst.dn_next);
1783 while (!rt) {
1784 rcu_read_unlock_bh();
1785 if (--s->bucket < 0)
1786 break;
1787 rcu_read_lock_bh();
1788 rt = rcu_dereference_bh(dn_rt_hash_table[s->bucket].chain);
1789 }
1790 return rt;
1791 }
1792
1793 static void *dn_rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
1794 {
1795 struct dn_route *rt = dn_rt_cache_get_first(seq);
1796
1797 if (rt) {
1798 while(*pos && (rt = dn_rt_cache_get_next(seq, rt)))
1799 --*pos;
1800 }
1801 return *pos ? NULL : rt;
1802 }
1803
1804 static void *dn_rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1805 {
1806 struct dn_route *rt = dn_rt_cache_get_next(seq, v);
1807 ++*pos;
1808 return rt;
1809 }
1810
1811 static void dn_rt_cache_seq_stop(struct seq_file *seq, void *v)
1812 {
1813 if (v)
1814 rcu_read_unlock_bh();
1815 }
1816
1817 static int dn_rt_cache_seq_show(struct seq_file *seq, void *v)
1818 {
1819 struct dn_route *rt = v;
1820 char buf1[DN_ASCBUF_LEN], buf2[DN_ASCBUF_LEN];
1821
1822 seq_printf(seq, "%-8s %-7s %-7s %04d %04d %04d\n",
1823 rt->dst.dev ? rt->dst.dev->name : "*",
1824 dn_addr2asc(le16_to_cpu(rt->rt_daddr), buf1),
1825 dn_addr2asc(le16_to_cpu(rt->rt_saddr), buf2),
1826 atomic_read(&rt->dst.__refcnt),
1827 rt->dst.__use, 0);
1828 return 0;
1829 }
1830
1831 static const struct seq_operations dn_rt_cache_seq_ops = {
1832 .start = dn_rt_cache_seq_start,
1833 .next = dn_rt_cache_seq_next,
1834 .stop = dn_rt_cache_seq_stop,
1835 .show = dn_rt_cache_seq_show,
1836 };
1837
1838 static int dn_rt_cache_seq_open(struct inode *inode, struct file *file)
1839 {
1840 return seq_open_private(file, &dn_rt_cache_seq_ops,
1841 sizeof(struct dn_rt_cache_iter_state));
1842 }
1843
1844 static const struct file_operations dn_rt_cache_seq_fops = {
1845 .owner = THIS_MODULE,
1846 .open = dn_rt_cache_seq_open,
1847 .read = seq_read,
1848 .llseek = seq_lseek,
1849 .release = seq_release_private,
1850 };
1851
1852 #endif /* CONFIG_PROC_FS */
1853
1854 void __init dn_route_init(void)
1855 {
1856 int i, goal, order;
1857
1858 dn_dst_ops.kmem_cachep =
1859 kmem_cache_create("dn_dst_cache", sizeof(struct dn_route), 0,
1860 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1861 dst_entries_init(&dn_dst_ops);
1862 setup_timer(&dn_route_timer, dn_dst_check_expire, 0);
1863 dn_route_timer.expires = jiffies + decnet_dst_gc_interval * HZ;
1864 add_timer(&dn_route_timer);
1865
1866 goal = totalram_pages >> (26 - PAGE_SHIFT);
1867
1868 for(order = 0; (1UL << order) < goal; order++)
1869 /* NOTHING */;
1870
1871 /*
1872 * Only want 1024 entries max, since the table is very, very unlikely
1873 * to be larger than that.
1874 */
1875 while(order && ((((1UL << order) * PAGE_SIZE) /
1876 sizeof(struct dn_rt_hash_bucket)) >= 2048))
1877 order--;
1878
1879 do {
1880 dn_rt_hash_mask = (1UL << order) * PAGE_SIZE /
1881 sizeof(struct dn_rt_hash_bucket);
1882 while(dn_rt_hash_mask & (dn_rt_hash_mask - 1))
1883 dn_rt_hash_mask--;
1884 dn_rt_hash_table = (struct dn_rt_hash_bucket *)
1885 __get_free_pages(GFP_ATOMIC, order);
1886 } while (dn_rt_hash_table == NULL && --order > 0);
1887
1888 if (!dn_rt_hash_table)
1889 panic("Failed to allocate DECnet route cache hash table\n");
1890
1891 printk(KERN_INFO
1892 "DECnet: Routing cache hash table of %u buckets, %ldKbytes\n",
1893 dn_rt_hash_mask,
1894 (long)(dn_rt_hash_mask*sizeof(struct dn_rt_hash_bucket))/1024);
1895
1896 dn_rt_hash_mask--;
1897 for(i = 0; i <= dn_rt_hash_mask; i++) {
1898 spin_lock_init(&dn_rt_hash_table[i].lock);
1899 dn_rt_hash_table[i].chain = NULL;
1900 }
1901
1902 dn_dst_ops.gc_thresh = (dn_rt_hash_mask + 1);
1903
1904 proc_net_fops_create(&init_net, "decnet_cache", S_IRUGO, &dn_rt_cache_seq_fops);
1905
1906 #ifdef CONFIG_DECNET_ROUTER
1907 rtnl_register(PF_DECnet, RTM_GETROUTE, dn_cache_getroute,
1908 dn_fib_dump, NULL);
1909 #else
1910 rtnl_register(PF_DECnet, RTM_GETROUTE, dn_cache_getroute,
1911 dn_cache_dump, NULL);
1912 #endif
1913 }
1914
1915 void __exit dn_route_cleanup(void)
1916 {
1917 del_timer(&dn_route_timer);
1918 dn_run_flush(0);
1919
1920 proc_net_remove(&init_net, "decnet_cache");
1921 dst_entries_destroy(&dn_dst_ops);
1922 }
1923
Linus' kernel tree