?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
cgroup: pass struct mem_cgroup instead of struct cgroup to socket memcg
[linux-2.6/libata-dev.git] / net / decnet / dn_route.c
blob80a3de4906d3d8bb4c598dde288e1decddf8e7c9
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 struct dst_entry *dn_dst_negative_advice(struct dst_entry *);
118 static void dn_dst_link_failure(struct sk_buff *);
119 static void dn_dst_update_pmtu(struct dst_entry *dst, u32 mtu);
120 static struct neighbour *dn_dst_neigh_lookup(const struct dst_entry *dst, const void *daddr);
121 static int dn_route_input(struct sk_buff *);
122 static void dn_run_flush(unsigned long dummy);
123
124 static struct dn_rt_hash_bucket *dn_rt_hash_table;
125 static unsigned dn_rt_hash_mask;
126
127 static struct timer_list dn_route_timer;
128 static DEFINE_TIMER(dn_rt_flush_timer, dn_run_flush, 0, 0);
129 int decnet_dst_gc_interval = 2;
130
131 static struct dst_ops dn_dst_ops = {
132 .family = PF_DECnet,
133 .protocol = cpu_to_be16(ETH_P_DNA_RT),
134 .gc_thresh = 128,
135 .gc = dn_dst_gc,
136 .check = dn_dst_check,
137 .default_advmss = dn_dst_default_advmss,
138 .mtu = dn_dst_mtu,
139 .cow_metrics = dst_cow_metrics_generic,
140 .destroy = dn_dst_destroy,
141 .negative_advice = dn_dst_negative_advice,
142 .link_failure = dn_dst_link_failure,
143 .update_pmtu = dn_dst_update_pmtu,
144 .neigh_lookup = dn_dst_neigh_lookup,
145 };
146
147 static void dn_dst_destroy(struct dst_entry *dst)
148 {
149 dst_destroy_metrics_generic(dst);
150 }
151
152 static __inline__ unsigned dn_hash(__le16 src, __le16 dst)
153 {
154 __u16 tmp = (__u16 __force)(src ^ dst);
155 tmp ^= (tmp >> 3);
156 tmp ^= (tmp >> 5);
157 tmp ^= (tmp >> 10);
158 return dn_rt_hash_mask & (unsigned)tmp;
159 }
160
161 static inline void dnrt_free(struct dn_route *rt)
162 {
163 call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
164 }
165
166 static inline void dnrt_drop(struct dn_route *rt)
167 {
168 dst_release(&rt->dst);
169 call_rcu_bh(&rt->dst.rcu_head, dst_rcu_free);
170 }
171
172 static void dn_dst_check_expire(unsigned long dummy)
173 {
174 int i;
175 struct dn_route *rt;
176 struct dn_route __rcu **rtp;
177 unsigned long now = jiffies;
178 unsigned long expire = 120 * HZ;
179
180 for (i = 0; i <= dn_rt_hash_mask; i++) {
181 rtp = &dn_rt_hash_table[i].chain;
182
183 spin_lock(&dn_rt_hash_table[i].lock);
184 while ((rt = rcu_dereference_protected(*rtp,
185 lockdep_is_held(&dn_rt_hash_table[i].lock))) != NULL) {
186 if (atomic_read(&rt->dst.__refcnt) ||
187 (now - rt->dst.lastuse) < expire) {
188 rtp = &rt->dst.dn_next;
189 continue;
190 }
191 *rtp = rt->dst.dn_next;
192 rt->dst.dn_next = NULL;
193 dnrt_free(rt);
194 }
195 spin_unlock(&dn_rt_hash_table[i].lock);
196
197 if ((jiffies - now) > 0)
198 break;
199 }
200
201 mod_timer(&dn_route_timer, now + decnet_dst_gc_interval * HZ);
202 }
203
204 static int dn_dst_gc(struct dst_ops *ops)
205 {
206 struct dn_route *rt;
207 struct dn_route __rcu **rtp;
208 int i;
209 unsigned long now = jiffies;
210 unsigned long expire = 10 * HZ;
211
212 for (i = 0; i <= dn_rt_hash_mask; i++) {
213
214 spin_lock_bh(&dn_rt_hash_table[i].lock);
215 rtp = &dn_rt_hash_table[i].chain;
216
217 while ((rt = rcu_dereference_protected(*rtp,
218 lockdep_is_held(&dn_rt_hash_table[i].lock))) != NULL) {
219 if (atomic_read(&rt->dst.__refcnt) ||
220 (now - rt->dst.lastuse) < expire) {
221 rtp = &rt->dst.dn_next;
222 continue;
223 }
224 *rtp = rt->dst.dn_next;
225 rt->dst.dn_next = NULL;
226 dnrt_drop(rt);
227 break;
228 }
229 spin_unlock_bh(&dn_rt_hash_table[i].lock);
230 }
231
232 return 0;
233 }
234
235 /*
236 * The decnet standards don't impose a particular minimum mtu, what they
237 * do insist on is that the routing layer accepts a datagram of at least
238 * 230 bytes long. Here we have to subtract the routing header length from
239 * 230 to get the minimum acceptable mtu. If there is no neighbour, then we
240 * assume the worst and use a long header size.
241 *
242 * We update both the mtu and the advertised mss (i.e. the segment size we
243 * advertise to the other end).
244 */
245 static void dn_dst_update_pmtu(struct dst_entry *dst, u32 mtu)
246 {
247 struct neighbour *n = dst_get_neighbour_noref(dst);
248 u32 min_mtu = 230;
249 struct dn_dev *dn;
250
251 dn = n ? rcu_dereference_raw(n->dev->dn_ptr) : NULL;
252
253 if (dn && dn->use_long == 0)
254 min_mtu -= 6;
255 else
256 min_mtu -= 21;
257
258 if (dst_metric(dst, RTAX_MTU) > mtu && mtu >= min_mtu) {
259 if (!(dst_metric_locked(dst, RTAX_MTU))) {
260 dst_metric_set(dst, RTAX_MTU, mtu);
261 dst_set_expires(dst, dn_rt_mtu_expires);
262 }
263 if (!(dst_metric_locked(dst, RTAX_ADVMSS))) {
264 u32 mss = mtu - DN_MAX_NSP_DATA_HEADER;
265 u32 existing_mss = dst_metric_raw(dst, RTAX_ADVMSS);
266 if (!existing_mss || existing_mss > mss)
267 dst_metric_set(dst, RTAX_ADVMSS, mss);
268 }
269 }
270 }
271
272 /*
273 * When a route has been marked obsolete. (e.g. routing cache flush)
274 */
275 static struct dst_entry *dn_dst_check(struct dst_entry *dst, __u32 cookie)
276 {
277 return NULL;
278 }
279
280 static struct dst_entry *dn_dst_negative_advice(struct dst_entry *dst)
281 {
282 dst_release(dst);
283 return NULL;
284 }
285
286 static void dn_dst_link_failure(struct sk_buff *skb)
287 {
288 }
289
290 static inline int compare_keys(struct flowidn *fl1, struct flowidn *fl2)
291 {
292 return ((fl1->daddr ^ fl2->daddr) |
293 (fl1->saddr ^ fl2->saddr) |
294 (fl1->flowidn_mark ^ fl2->flowidn_mark) |
295 (fl1->flowidn_scope ^ fl2->flowidn_scope) |
296 (fl1->flowidn_oif ^ fl2->flowidn_oif) |
297 (fl1->flowidn_iif ^ fl2->flowidn_iif)) == 0;
298 }
299
300 static int dn_insert_route(struct dn_route *rt, unsigned hash, struct dn_route **rp)
301 {
302 struct dn_route *rth;
303 struct dn_route __rcu **rthp;
304 unsigned long now = jiffies;
305
306 rthp = &dn_rt_hash_table[hash].chain;
307
308 spin_lock_bh(&dn_rt_hash_table[hash].lock);
309 while ((rth = rcu_dereference_protected(*rthp,
310 lockdep_is_held(&dn_rt_hash_table[hash].lock))) != NULL) {
311 if (compare_keys(&rth->fld, &rt->fld)) {
312 /* Put it first */
313 *rthp = rth->dst.dn_next;
314 rcu_assign_pointer(rth->dst.dn_next,
315 dn_rt_hash_table[hash].chain);
316 rcu_assign_pointer(dn_rt_hash_table[hash].chain, rth);
317
318 dst_use(&rth->dst, now);
319 spin_unlock_bh(&dn_rt_hash_table[hash].lock);
320
321 dnrt_drop(rt);
322 *rp = rth;
323 return 0;
324 }
325 rthp = &rth->dst.dn_next;
326 }
327
328 rcu_assign_pointer(rt->dst.dn_next, dn_rt_hash_table[hash].chain);
329 rcu_assign_pointer(dn_rt_hash_table[hash].chain, rt);
330
331 dst_use(&rt->dst, now);
332 spin_unlock_bh(&dn_rt_hash_table[hash].lock);
333 *rp = rt;
334 return 0;
335 }
336
337 static void dn_run_flush(unsigned long dummy)
338 {
339 int i;
340 struct dn_route *rt, *next;
341
342 for (i = 0; i < dn_rt_hash_mask; i++) {
343 spin_lock_bh(&dn_rt_hash_table[i].lock);
344
345 if ((rt = xchg((struct dn_route **)&dn_rt_hash_table[i].chain, NULL)) == NULL)
346 goto nothing_to_declare;
347
348 for(; rt; rt = next) {
349 next = rcu_dereference_raw(rt->dst.dn_next);
350 RCU_INIT_POINTER(rt->dst.dn_next, NULL);
351 dst_free((struct dst_entry *)rt);
352 }
353
354 nothing_to_declare:
355 spin_unlock_bh(&dn_rt_hash_table[i].lock);
356 }
357 }
358
359 static DEFINE_SPINLOCK(dn_rt_flush_lock);
360
361 void dn_rt_cache_flush(int delay)
362 {
363 unsigned long now = jiffies;
364 int user_mode = !in_interrupt();
365
366 if (delay < 0)
367 delay = dn_rt_min_delay;
368
369 spin_lock_bh(&dn_rt_flush_lock);
370
371 if (del_timer(&dn_rt_flush_timer) && delay > 0 && dn_rt_deadline) {
372 long tmo = (long)(dn_rt_deadline - now);
373
374 if (user_mode && tmo < dn_rt_max_delay - dn_rt_min_delay)
375 tmo = 0;
376
377 if (delay > tmo)
378 delay = tmo;
379 }
380
381 if (delay <= 0) {
382 spin_unlock_bh(&dn_rt_flush_lock);
383 dn_run_flush(0);
384 return;
385 }
386
387 if (dn_rt_deadline == 0)
388 dn_rt_deadline = now + dn_rt_max_delay;
389
390 dn_rt_flush_timer.expires = now + delay;
391 add_timer(&dn_rt_flush_timer);
392 spin_unlock_bh(&dn_rt_flush_lock);
393 }
394
395 /**
396 * dn_return_short - Return a short packet to its sender
397 * @skb: The packet to return
398 *
399 */
400 static int dn_return_short(struct sk_buff *skb)
401 {
402 struct dn_skb_cb *cb;
403 unsigned char *ptr;
404 __le16 *src;
405 __le16 *dst;
406
407 /* Add back headers */
408 skb_push(skb, skb->data - skb_network_header(skb));
409
410 if ((skb = skb_unshare(skb, GFP_ATOMIC)) == NULL)
411 return NET_RX_DROP;
412
413 cb = DN_SKB_CB(skb);
414 /* Skip packet length and point to flags */
415 ptr = skb->data + 2;
416 *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
417
418 dst = (__le16 *)ptr;
419 ptr += 2;
420 src = (__le16 *)ptr;
421 ptr += 2;
422 *ptr = 0; /* Zero hop count */
423
424 swap(*src, *dst);
425
426 skb->pkt_type = PACKET_OUTGOING;
427 dn_rt_finish_output(skb, NULL, NULL);
428 return NET_RX_SUCCESS;
429 }
430
431 /**
432 * dn_return_long - Return a long packet to its sender
433 * @skb: The long format packet to return
434 *
435 */
436 static int dn_return_long(struct sk_buff *skb)
437 {
438 struct dn_skb_cb *cb;
439 unsigned char *ptr;
440 unsigned char *src_addr, *dst_addr;
441 unsigned char tmp[ETH_ALEN];
442
443 /* Add back all headers */
444 skb_push(skb, skb->data - skb_network_header(skb));
445
446 if ((skb = skb_unshare(skb, GFP_ATOMIC)) == NULL)
447 return NET_RX_DROP;
448
449 cb = DN_SKB_CB(skb);
450 /* Ignore packet length and point to flags */
451 ptr = skb->data + 2;
452
453 /* Skip padding */
454 if (*ptr & DN_RT_F_PF) {
455 char padlen = (*ptr & ~DN_RT_F_PF);
456 ptr += padlen;
457 }
458
459 *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
460 ptr += 2;
461 dst_addr = ptr;
462 ptr += 8;
463 src_addr = ptr;
464 ptr += 6;
465 *ptr = 0; /* Zero hop count */
466
467 /* Swap source and destination */
468 memcpy(tmp, src_addr, ETH_ALEN);
469 memcpy(src_addr, dst_addr, ETH_ALEN);
470 memcpy(dst_addr, tmp, ETH_ALEN);
471
472 skb->pkt_type = PACKET_OUTGOING;
473 dn_rt_finish_output(skb, dst_addr, src_addr);
474 return NET_RX_SUCCESS;
475 }
476
477 /**
478 * dn_route_rx_packet - Try and find a route for an incoming packet
479 * @skb: The packet to find a route for
480 *
481 * Returns: result of input function if route is found, error code otherwise
482 */
483 static int dn_route_rx_packet(struct sk_buff *skb)
484 {
485 struct dn_skb_cb *cb;
486 int err;
487
488 if ((err = dn_route_input(skb)) == 0)
489 return dst_input(skb);
490
491 cb = DN_SKB_CB(skb);
492 if (decnet_debug_level & 4) {
493 char *devname = skb->dev ? skb->dev->name : "???";
494
495 printk(KERN_DEBUG
496 "DECnet: dn_route_rx_packet: rt_flags=0x%02x dev=%s len=%d src=0x%04hx dst=0x%04hx err=%d type=%d\n",
497 (int)cb->rt_flags, devname, skb->len,
498 le16_to_cpu(cb->src), le16_to_cpu(cb->dst),
499 err, skb->pkt_type);
500 }
501
502 if ((skb->pkt_type == PACKET_HOST) && (cb->rt_flags & DN_RT_F_RQR)) {
503 switch (cb->rt_flags & DN_RT_PKT_MSK) {
504 case DN_RT_PKT_SHORT:
505 return dn_return_short(skb);
506 case DN_RT_PKT_LONG:
507 return dn_return_long(skb);
508 }
509 }
510
511 kfree_skb(skb);
512 return NET_RX_DROP;
513 }
514
515 static int dn_route_rx_long(struct sk_buff *skb)
516 {
517 struct dn_skb_cb *cb = DN_SKB_CB(skb);
518 unsigned char *ptr = skb->data;
519
520 if (!pskb_may_pull(skb, 21)) /* 20 for long header, 1 for shortest nsp */
521 goto drop_it;
522
523 skb_pull(skb, 20);
524 skb_reset_transport_header(skb);
525
526 /* Destination info */
527 ptr += 2;
528 cb->dst = dn_eth2dn(ptr);
529 if (memcmp(ptr, dn_hiord_addr, 4) != 0)
530 goto drop_it;
531 ptr += 6;
532
533
534 /* Source info */
535 ptr += 2;
536 cb->src = dn_eth2dn(ptr);
537 if (memcmp(ptr, dn_hiord_addr, 4) != 0)
538 goto drop_it;
539 ptr += 6;
540 /* Other junk */
541 ptr++;
542 cb->hops = *ptr++; /* Visit Count */
543
544 return NF_HOOK(NFPROTO_DECNET, NF_DN_PRE_ROUTING, skb, skb->dev, NULL,
545 dn_route_rx_packet);
546
547 drop_it:
548 kfree_skb(skb);
549 return NET_RX_DROP;
550 }
551
552
553
554 static int dn_route_rx_short(struct sk_buff *skb)
555 {
556 struct dn_skb_cb *cb = DN_SKB_CB(skb);
557 unsigned char *ptr = skb->data;
558
559 if (!pskb_may_pull(skb, 6)) /* 5 for short header + 1 for shortest nsp */
560 goto drop_it;
561
562 skb_pull(skb, 5);
563 skb_reset_transport_header(skb);
564
565 cb->dst = *(__le16 *)ptr;
566 ptr += 2;
567 cb->src = *(__le16 *)ptr;
568 ptr += 2;
569 cb->hops = *ptr & 0x3f;
570
571 return NF_HOOK(NFPROTO_DECNET, NF_DN_PRE_ROUTING, skb, skb->dev, NULL,
572 dn_route_rx_packet);
573
574 drop_it:
575 kfree_skb(skb);
576 return NET_RX_DROP;
577 }
578
579 static int dn_route_discard(struct sk_buff *skb)
580 {
581 /*
582 * I know we drop the packet here, but thats considered success in
583 * this case
584 */
585 kfree_skb(skb);
586 return NET_RX_SUCCESS;
587 }
588
589 static int dn_route_ptp_hello(struct sk_buff *skb)
590 {
591 dn_dev_hello(skb);
592 dn_neigh_pointopoint_hello(skb);
593 return NET_RX_SUCCESS;
594 }
595
596 int dn_route_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
597 {
598 struct dn_skb_cb *cb;
599 unsigned char flags = 0;
600 __u16 len = le16_to_cpu(*(__le16 *)skb->data);
601 struct dn_dev *dn = rcu_dereference(dev->dn_ptr);
602 unsigned char padlen = 0;
603
604 if (!net_eq(dev_net(dev), &init_net))
605 goto dump_it;
606
607 if (dn == NULL)
608 goto dump_it;
609
610 if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
611 goto out;
612
613 if (!pskb_may_pull(skb, 3))
614 goto dump_it;
615
616 skb_pull(skb, 2);
617
618 if (len > skb->len)
619 goto dump_it;
620
621 skb_trim(skb, len);
622
623 flags = *skb->data;
624
625 cb = DN_SKB_CB(skb);
626 cb->stamp = jiffies;
627 cb->iif = dev->ifindex;
628
629 /*
630 * If we have padding, remove it.
631 */
632 if (flags & DN_RT_F_PF) {
633 padlen = flags & ~DN_RT_F_PF;
634 if (!pskb_may_pull(skb, padlen + 1))
635 goto dump_it;
636 skb_pull(skb, padlen);
637 flags = *skb->data;
638 }
639
640 skb_reset_network_header(skb);
641
642 /*
643 * Weed out future version DECnet
644 */
645 if (flags & DN_RT_F_VER)
646 goto dump_it;
647
648 cb->rt_flags = flags;
649
650 if (decnet_debug_level & 1)
651 printk(KERN_DEBUG
652 "dn_route_rcv: got 0x%02x from %s [%d %d %d]\n",
653 (int)flags, (dev) ? dev->name : "???", len, skb->len,
654 padlen);
655
656 if (flags & DN_RT_PKT_CNTL) {
657 if (unlikely(skb_linearize(skb)))
658 goto dump_it;
659
660 switch (flags & DN_RT_CNTL_MSK) {
661 case DN_RT_PKT_INIT:
662 dn_dev_init_pkt(skb);
663 break;
664 case DN_RT_PKT_VERI:
665 dn_dev_veri_pkt(skb);
666 break;
667 }
668
669 if (dn->parms.state != DN_DEV_S_RU)
670 goto dump_it;
671
672 switch (flags & DN_RT_CNTL_MSK) {
673 case DN_RT_PKT_HELO:
674 return NF_HOOK(NFPROTO_DECNET, NF_DN_HELLO,
675 skb, skb->dev, NULL,
676 dn_route_ptp_hello);
677
678 case DN_RT_PKT_L1RT:
679 case DN_RT_PKT_L2RT:
680 return NF_HOOK(NFPROTO_DECNET, NF_DN_ROUTE,
681 skb, skb->dev, NULL,
682 dn_route_discard);
683 case DN_RT_PKT_ERTH:
684 return NF_HOOK(NFPROTO_DECNET, NF_DN_HELLO,
685 skb, skb->dev, NULL,
686 dn_neigh_router_hello);
687
688 case DN_RT_PKT_EEDH:
689 return NF_HOOK(NFPROTO_DECNET, NF_DN_HELLO,
690 skb, skb->dev, NULL,
691 dn_neigh_endnode_hello);
692 }
693 } else {
694 if (dn->parms.state != DN_DEV_S_RU)
695 goto dump_it;
696
697 skb_pull(skb, 1); /* Pull flags */
698
699 switch (flags & DN_RT_PKT_MSK) {
700 case DN_RT_PKT_LONG:
701 return dn_route_rx_long(skb);
702 case DN_RT_PKT_SHORT:
703 return dn_route_rx_short(skb);
704 }
705 }
706
707 dump_it:
708 kfree_skb(skb);
709 out:
710 return NET_RX_DROP;
711 }
712
713 static int dn_to_neigh_output(struct sk_buff *skb)
714 {
715 struct dst_entry *dst = skb_dst(skb);
716 struct neighbour *n = dst_get_neighbour_noref(dst);
717
718 return n->output(n, skb);
719 }
720
721 static int dn_output(struct sk_buff *skb)
722 {
723 struct dst_entry *dst = skb_dst(skb);
724 struct dn_route *rt = (struct dn_route *)dst;
725 struct net_device *dev = dst->dev;
726 struct dn_skb_cb *cb = DN_SKB_CB(skb);
727
728 int err = -EINVAL;
729
730 if (dst_get_neighbour_noref(dst) == NULL)
731 goto error;
732
733 skb->dev = dev;
734
735 cb->src = rt->rt_saddr;
736 cb->dst = rt->rt_daddr;
737
738 /*
739 * Always set the Intra-Ethernet bit on all outgoing packets
740 * originated on this node. Only valid flag from upper layers
741 * is return-to-sender-requested. Set hop count to 0 too.
742 */
743 cb->rt_flags &= ~DN_RT_F_RQR;
744 cb->rt_flags |= DN_RT_F_IE;
745 cb->hops = 0;
746
747 return NF_HOOK(NFPROTO_DECNET, NF_DN_LOCAL_OUT, skb, NULL, dev,
748 dn_to_neigh_output);
749
750 error:
751 if (net_ratelimit())
752 printk(KERN_DEBUG "dn_output: This should not happen\n");
753
754 kfree_skb(skb);
755
756 return err;
757 }
758
759 static int dn_forward(struct sk_buff *skb)
760 {
761 struct dn_skb_cb *cb = DN_SKB_CB(skb);
762 struct dst_entry *dst = skb_dst(skb);
763 struct dn_dev *dn_db = rcu_dereference(dst->dev->dn_ptr);
764 struct dn_route *rt;
765 int header_len;
766 #ifdef CONFIG_NETFILTER
767 struct net_device *dev = skb->dev;
768 #endif
769
770 if (skb->pkt_type != PACKET_HOST)
771 goto drop;
772
773 /* Ensure that we have enough space for headers */
774 rt = (struct dn_route *)skb_dst(skb);
775 header_len = dn_db->use_long ? 21 : 6;
776 if (skb_cow(skb, LL_RESERVED_SPACE(rt->dst.dev)+header_len))
777 goto drop;
778
779 /*
780 * Hop count exceeded.
781 */
782 if (++cb->hops > 30)
783 goto drop;
784
785 skb->dev = rt->dst.dev;
786
787 /*
788 * If packet goes out same interface it came in on, then set
789 * the Intra-Ethernet bit. This has no effect for short
790 * packets, so we don't need to test for them here.
791 */
792 cb->rt_flags &= ~DN_RT_F_IE;
793 if (rt->rt_flags & RTCF_DOREDIRECT)
794 cb->rt_flags |= DN_RT_F_IE;
795
796 return NF_HOOK(NFPROTO_DECNET, NF_DN_FORWARD, skb, dev, skb->dev,
797 dn_to_neigh_output);
798
799 drop:
800 kfree_skb(skb);
801 return NET_RX_DROP;
802 }
803
804 /*
805 * Used to catch bugs. This should never normally get
806 * called.
807 */
808 static int dn_rt_bug(struct sk_buff *skb)
809 {
810 if (net_ratelimit()) {
811 struct dn_skb_cb *cb = DN_SKB_CB(skb);
812
813 printk(KERN_DEBUG "dn_rt_bug: skb from:%04x to:%04x\n",
814 le16_to_cpu(cb->src), le16_to_cpu(cb->dst));
815 }
816
817 kfree_skb(skb);
818
819 return NET_RX_DROP;
820 }
821
822 static unsigned int dn_dst_default_advmss(const struct dst_entry *dst)
823 {
824 return dn_mss_from_pmtu(dst->dev, dst_mtu(dst));
825 }
826
827 static unsigned int dn_dst_mtu(const struct dst_entry *dst)
828 {
829 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
830
831 return mtu ? : dst->dev->mtu;
832 }
833
834 static struct neighbour *dn_dst_neigh_lookup(const struct dst_entry *dst, const void *daddr)
835 {
836 return __neigh_lookup_errno(&dn_neigh_table, daddr, dst->dev);
837 }
838
839 static int dn_rt_set_next_hop(struct dn_route *rt, struct dn_fib_res *res)
840 {
841 struct dn_fib_info *fi = res->fi;
842 struct net_device *dev = rt->dst.dev;
843 unsigned int mss_metric;
844 struct neighbour *n;
845
846 if (fi) {
847 if (DN_FIB_RES_GW(*res) &&
848 DN_FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
849 rt->rt_gateway = DN_FIB_RES_GW(*res);
850 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
851 }
852 rt->rt_type = res->type;
853
854 if (dev != NULL && dst_get_neighbour_noref(&rt->dst) == NULL) {
855 n = __neigh_lookup_errno(&dn_neigh_table, &rt->rt_gateway, dev);
856 if (IS_ERR(n))
857 return PTR_ERR(n);
858 dst_set_neighbour(&rt->dst, n);
859 }
860
861 if (dst_metric(&rt->dst, RTAX_MTU) > rt->dst.dev->mtu)
862 dst_metric_set(&rt->dst, RTAX_MTU, rt->dst.dev->mtu);
863 mss_metric = dst_metric_raw(&rt->dst, RTAX_ADVMSS);
864 if (mss_metric) {
865 unsigned int mss = dn_mss_from_pmtu(dev, dst_mtu(&rt->dst));
866 if (mss_metric > mss)
867 dst_metric_set(&rt->dst, RTAX_ADVMSS, mss);
868 }
869 return 0;
870 }
871
872 static inline int dn_match_addr(__le16 addr1, __le16 addr2)
873 {
874 __u16 tmp = le16_to_cpu(addr1) ^ le16_to_cpu(addr2);
875 int match = 16;
876 while(tmp) {
877 tmp >>= 1;
878 match--;
879 }
880 return match;
881 }
882
883 static __le16 dnet_select_source(const struct net_device *dev, __le16 daddr, int scope)
884 {
885 __le16 saddr = 0;
886 struct dn_dev *dn_db;
887 struct dn_ifaddr *ifa;
888 int best_match = 0;
889 int ret;
890
891 rcu_read_lock();
892 dn_db = rcu_dereference(dev->dn_ptr);
893 for (ifa = rcu_dereference(dn_db->ifa_list);
894 ifa != NULL;
895 ifa = rcu_dereference(ifa->ifa_next)) {
896 if (ifa->ifa_scope > scope)
897 continue;
898 if (!daddr) {
899 saddr = ifa->ifa_local;
900 break;
901 }
902 ret = dn_match_addr(daddr, ifa->ifa_local);
903 if (ret > best_match)
904 saddr = ifa->ifa_local;
905 if (best_match == 0)
906 saddr = ifa->ifa_local;
907 }
908 rcu_read_unlock();
909
910 return saddr;
911 }
912
913 static inline __le16 __dn_fib_res_prefsrc(struct dn_fib_res *res)
914 {
915 return dnet_select_source(DN_FIB_RES_DEV(*res), DN_FIB_RES_GW(*res), res->scope);
916 }
917
918 static inline __le16 dn_fib_rules_map_destination(__le16 daddr, struct dn_fib_res *res)
919 {
920 __le16 mask = dnet_make_mask(res->prefixlen);
921 return (daddr&~mask)|res->fi->fib_nh->nh_gw;
922 }
923
924 static int dn_route_output_slow(struct dst_entry **pprt, const struct flowidn *oldflp, int try_hard)
925 {
926 struct flowidn fld = {
927 .daddr = oldflp->daddr,
928 .saddr = oldflp->saddr,
929 .flowidn_scope = RT_SCOPE_UNIVERSE,
930 .flowidn_mark = oldflp->flowidn_mark,
931 .flowidn_iif = init_net.loopback_dev->ifindex,
932 .flowidn_oif = oldflp->flowidn_oif,
933 };
934 struct dn_route *rt = NULL;
935 struct net_device *dev_out = NULL, *dev;
936 struct neighbour *neigh = NULL;
937 unsigned hash;
938 unsigned flags = 0;
939 struct dn_fib_res res = { .fi = NULL, .type = RTN_UNICAST };
940 int err;
941 int free_res = 0;
942 __le16 gateway = 0;
943
944 if (decnet_debug_level & 16)
945 printk(KERN_DEBUG
946 "dn_route_output_slow: dst=%04x src=%04x mark=%d"
947 " iif=%d oif=%d\n", le16_to_cpu(oldflp->daddr),
948 le16_to_cpu(oldflp->saddr),
949 oldflp->flowidn_mark, init_net.loopback_dev->ifindex,
950 oldflp->flowidn_oif);
951
952 /* If we have an output interface, verify its a DECnet device */
953 if (oldflp->flowidn_oif) {
954 dev_out = dev_get_by_index(&init_net, oldflp->flowidn_oif);
955 err = -ENODEV;
956 if (dev_out && dev_out->dn_ptr == NULL) {
957 dev_put(dev_out);
958 dev_out = NULL;
959 }
960 if (dev_out == NULL)
961 goto out;
962 }
963
964 /* If we have a source address, verify that its a local address */
965 if (oldflp->saddr) {
966 err = -EADDRNOTAVAIL;
967
968 if (dev_out) {
969 if (dn_dev_islocal(dev_out, oldflp->saddr))
970 goto source_ok;
971 dev_put(dev_out);
972 goto out;
973 }
974 rcu_read_lock();
975 for_each_netdev_rcu(&init_net, dev) {
976 if (!dev->dn_ptr)
977 continue;
978 if (!dn_dev_islocal(dev, oldflp->saddr))
979 continue;
980 if ((dev->flags & IFF_LOOPBACK) &&
981 oldflp->daddr &&
982 !dn_dev_islocal(dev, oldflp->daddr))
983 continue;
984
985 dev_out = dev;
986 break;
987 }
988 rcu_read_unlock();
989 if (dev_out == NULL)
990 goto out;
991 dev_hold(dev_out);
992 source_ok:
993 ;
994 }
995
996 /* No destination? Assume its local */
997 if (!fld.daddr) {
998 fld.daddr = fld.saddr;
999
1000 err = -EADDRNOTAVAIL;
1001 if (dev_out)
1002 dev_put(dev_out);
1003 dev_out = init_net.loopback_dev;
1004 dev_hold(dev_out);
1005 if (!fld.daddr) {
1006 fld.daddr =
1007 fld.saddr = dnet_select_source(dev_out, 0,
1008 RT_SCOPE_HOST);
1009 if (!fld.daddr)
1010 goto out;
1011 }
1012 fld.flowidn_oif = init_net.loopback_dev->ifindex;
1013 res.type = RTN_LOCAL;
1014 goto make_route;
1015 }
1016
1017 if (decnet_debug_level & 16)
1018 printk(KERN_DEBUG
1019 "dn_route_output_slow: initial checks complete."
1020 " dst=%o4x src=%04x oif=%d try_hard=%d\n",
1021 le16_to_cpu(fld.daddr), le16_to_cpu(fld.saddr),
1022 fld.flowidn_oif, try_hard);
1023
1024 /*
1025 * N.B. If the kernel is compiled without router support then
1026 * dn_fib_lookup() will evaluate to non-zero so this if () block
1027 * will always be executed.
1028 */
1029 err = -ESRCH;
1030 if (try_hard || (err = dn_fib_lookup(&fld, &res)) != 0) {
1031 struct dn_dev *dn_db;
1032 if (err != -ESRCH)
1033 goto out;
1034 /*
1035 * Here the fallback is basically the standard algorithm for
1036 * routing in endnodes which is described in the DECnet routing
1037 * docs
1038 *
1039 * If we are not trying hard, look in neighbour cache.
1040 * The result is tested to ensure that if a specific output
1041 * device/source address was requested, then we honour that
1042 * here
1043 */
1044 if (!try_hard) {
1045 neigh = neigh_lookup_nodev(&dn_neigh_table, &init_net, &fld.daddr);
1046 if (neigh) {
1047 if ((oldflp->flowidn_oif &&
1048 (neigh->dev->ifindex != oldflp->flowidn_oif)) ||
1049 (oldflp->saddr &&
1050 (!dn_dev_islocal(neigh->dev,
1051 oldflp->saddr)))) {
1052 neigh_release(neigh);
1053 neigh = NULL;
1054 } else {
1055 if (dev_out)
1056 dev_put(dev_out);
1057 if (dn_dev_islocal(neigh->dev, fld.daddr)) {
1058 dev_out = init_net.loopback_dev;
1059 res.type = RTN_LOCAL;
1060 } else {
1061 dev_out = neigh->dev;
1062 }
1063 dev_hold(dev_out);
1064 goto select_source;
1065 }
1066 }
1067 }
1068
1069 /* Not there? Perhaps its a local address */
1070 if (dev_out == NULL)
1071 dev_out = dn_dev_get_default();
1072 err = -ENODEV;
1073 if (dev_out == NULL)
1074 goto out;
1075 dn_db = rcu_dereference_raw(dev_out->dn_ptr);
1076 /* Possible improvement - check all devices for local addr */
1077 if (dn_dev_islocal(dev_out, fld.daddr)) {
1078 dev_put(dev_out);
1079 dev_out = init_net.loopback_dev;
1080 dev_hold(dev_out);
1081 res.type = RTN_LOCAL;
1082 goto select_source;
1083 }
1084 /* Not local either.... try sending it to the default router */
1085 neigh = neigh_clone(dn_db->router);
1086 BUG_ON(neigh && neigh->dev != dev_out);
1087
1088 /* Ok then, we assume its directly connected and move on */
1089 select_source:
1090 if (neigh)
1091 gateway = ((struct dn_neigh *)neigh)->addr;
1092 if (gateway == 0)
1093 gateway = fld.daddr;
1094 if (fld.saddr == 0) {
1095 fld.saddr = dnet_select_source(dev_out, gateway,
1096 res.type == RTN_LOCAL ?
1097 RT_SCOPE_HOST :
1098 RT_SCOPE_LINK);
1099 if (fld.saddr == 0 && res.type != RTN_LOCAL)
1100 goto e_addr;
1101 }
1102 fld.flowidn_oif = dev_out->ifindex;
1103 goto make_route;
1104 }
1105 free_res = 1;
1106
1107 if (res.type == RTN_NAT)
1108 goto e_inval;
1109
1110 if (res.type == RTN_LOCAL) {
1111 if (!fld.saddr)
1112 fld.saddr = fld.daddr;
1113 if (dev_out)
1114 dev_put(dev_out);
1115 dev_out = init_net.loopback_dev;
1116 dev_hold(dev_out);
1117 fld.flowidn_oif = dev_out->ifindex;
1118 if (res.fi)
1119 dn_fib_info_put(res.fi);
1120 res.fi = NULL;
1121 goto make_route;
1122 }
1123
1124 if (res.fi->fib_nhs > 1 && fld.flowidn_oif == 0)
1125 dn_fib_select_multipath(&fld, &res);
1126
1127 /*
1128 * We could add some logic to deal with default routes here and
1129 * get rid of some of the special casing above.
1130 */
1131
1132 if (!fld.saddr)
1133 fld.saddr = DN_FIB_RES_PREFSRC(res);
1134
1135 if (dev_out)
1136 dev_put(dev_out);
1137 dev_out = DN_FIB_RES_DEV(res);
1138 dev_hold(dev_out);
1139 fld.flowidn_oif = dev_out->ifindex;
1140 gateway = DN_FIB_RES_GW(res);
1141
1142 make_route:
1143 if (dev_out->flags & IFF_LOOPBACK)
1144 flags |= RTCF_LOCAL;
1145
1146 rt = dst_alloc(&dn_dst_ops, dev_out, 1, 0, DST_HOST);
1147 if (rt == NULL)
1148 goto e_nobufs;
1149
1150 memset(&rt->fld, 0, sizeof(rt->fld));
1151 rt->fld.saddr = oldflp->saddr;
1152 rt->fld.daddr = oldflp->daddr;
1153 rt->fld.flowidn_oif = oldflp->flowidn_oif;
1154 rt->fld.flowidn_iif = 0;
1155 rt->fld.flowidn_mark = oldflp->flowidn_mark;
1156
1157 rt->rt_saddr = fld.saddr;
1158 rt->rt_daddr = fld.daddr;
1159 rt->rt_gateway = gateway ? gateway : fld.daddr;
1160 rt->rt_local_src = fld.saddr;
1161
1162 rt->rt_dst_map = fld.daddr;
1163 rt->rt_src_map = fld.saddr;
1164
1165 dst_set_neighbour(&rt->dst, neigh);
1166 neigh = NULL;
1167
1168 rt->dst.lastuse = jiffies;
1169 rt->dst.output = dn_output;
1170 rt->dst.input = dn_rt_bug;
1171 rt->rt_flags = flags;
1172 if (flags & RTCF_LOCAL)
1173 rt->dst.input = dn_nsp_rx;
1174
1175 err = dn_rt_set_next_hop(rt, &res);
1176 if (err)
1177 goto e_neighbour;
1178
1179 hash = dn_hash(rt->fld.saddr, rt->fld.daddr);
1180 dn_insert_route(rt, hash, (struct dn_route **)pprt);
1181
1182 done:
1183 if (neigh)
1184 neigh_release(neigh);
1185 if (free_res)
1186 dn_fib_res_put(&res);
1187 if (dev_out)
1188 dev_put(dev_out);
1189 out:
1190 return err;
1191
1192 e_addr:
1193 err = -EADDRNOTAVAIL;
1194 goto done;
1195 e_inval:
1196 err = -EINVAL;
1197 goto done;
1198 e_nobufs:
1199 err = -ENOBUFS;
1200 goto done;
1201 e_neighbour:
1202 dst_free(&rt->dst);
1203 goto e_nobufs;
1204 }
1205
1206
1207 /*
1208 * N.B. The flags may be moved into the flowi at some future stage.
1209 */
1210 static int __dn_route_output_key(struct dst_entry **pprt, const struct flowidn *flp, int flags)
1211 {
1212 unsigned hash = dn_hash(flp->saddr, flp->daddr);
1213 struct dn_route *rt = NULL;
1214
1215 if (!(flags & MSG_TRYHARD)) {
1216 rcu_read_lock_bh();
1217 for (rt = rcu_dereference_bh(dn_rt_hash_table[hash].chain); rt;
1218 rt = rcu_dereference_bh(rt->dst.dn_next)) {
1219 if ((flp->daddr == rt->fld.daddr) &&
1220 (flp->saddr == rt->fld.saddr) &&
1221 (flp->flowidn_mark == rt->fld.flowidn_mark) &&
1222 dn_is_output_route(rt) &&
1223 (rt->fld.flowidn_oif == flp->flowidn_oif)) {
1224 dst_use(&rt->dst, jiffies);
1225 rcu_read_unlock_bh();
1226 *pprt = &rt->dst;
1227 return 0;
1228 }
1229 }
1230 rcu_read_unlock_bh();
1231 }
1232
1233 return dn_route_output_slow(pprt, flp, flags);
1234 }
1235
1236 static int dn_route_output_key(struct dst_entry **pprt, struct flowidn *flp, int flags)
1237 {
1238 int err;
1239
1240 err = __dn_route_output_key(pprt, flp, flags);
1241 if (err == 0 && flp->flowidn_proto) {
1242 *pprt = xfrm_lookup(&init_net, *pprt,
1243 flowidn_to_flowi(flp), NULL, 0);
1244 if (IS_ERR(*pprt)) {
1245 err = PTR_ERR(*pprt);
1246 *pprt = NULL;
1247 }
1248 }
1249 return err;
1250 }
1251
1252 int dn_route_output_sock(struct dst_entry **pprt, struct flowidn *fl, struct sock *sk, int flags)
1253 {
1254 int err;
1255
1256 err = __dn_route_output_key(pprt, fl, flags & MSG_TRYHARD);
1257 if (err == 0 && fl->flowidn_proto) {
1258 if (!(flags & MSG_DONTWAIT))
1259 fl->flowidn_flags |= FLOWI_FLAG_CAN_SLEEP;
1260 *pprt = xfrm_lookup(&init_net, *pprt,
1261 flowidn_to_flowi(fl), sk, 0);
1262 if (IS_ERR(*pprt)) {
1263 err = PTR_ERR(*pprt);
1264 *pprt = NULL;
1265 }
1266 }
1267 return err;
1268 }
1269
1270 static int dn_route_input_slow(struct sk_buff *skb)
1271 {
1272 struct dn_route *rt = NULL;
1273 struct dn_skb_cb *cb = DN_SKB_CB(skb);
1274 struct net_device *in_dev = skb->dev;
1275 struct net_device *out_dev = NULL;
1276 struct dn_dev *dn_db;
1277 struct neighbour *neigh = NULL;
1278 unsigned hash;
1279 int flags = 0;
1280 __le16 gateway = 0;
1281 __le16 local_src = 0;
1282 struct flowidn fld = {
1283 .daddr = cb->dst,
1284 .saddr = cb->src,
1285 .flowidn_scope = RT_SCOPE_UNIVERSE,
1286 .flowidn_mark = skb->mark,
1287 .flowidn_iif = skb->dev->ifindex,
1288 };
1289 struct dn_fib_res res = { .fi = NULL, .type = RTN_UNREACHABLE };
1290 int err = -EINVAL;
1291 int free_res = 0;
1292
1293 dev_hold(in_dev);
1294
1295 if ((dn_db = rcu_dereference(in_dev->dn_ptr)) == NULL)
1296 goto out;
1297
1298 /* Zero source addresses are not allowed */
1299 if (fld.saddr == 0)
1300 goto out;
1301
1302 /*
1303 * In this case we've just received a packet from a source
1304 * outside ourselves pretending to come from us. We don't
1305 * allow it any further to prevent routing loops, spoofing and
1306 * other nasties. Loopback packets already have the dst attached
1307 * so this only affects packets which have originated elsewhere.
1308 */
1309 err = -ENOTUNIQ;
1310 if (dn_dev_islocal(in_dev, cb->src))
1311 goto out;
1312
1313 err = dn_fib_lookup(&fld, &res);
1314 if (err) {
1315 if (err != -ESRCH)
1316 goto out;
1317 /*
1318 * Is the destination us ?
1319 */
1320 if (!dn_dev_islocal(in_dev, cb->dst))
1321 goto e_inval;
1322
1323 res.type = RTN_LOCAL;
1324 } else {
1325 __le16 src_map = fld.saddr;
1326 free_res = 1;
1327
1328 out_dev = DN_FIB_RES_DEV(res);
1329 if (out_dev == NULL) {
1330 if (net_ratelimit())
1331 printk(KERN_CRIT "Bug in dn_route_input_slow() "
1332 "No output device\n");
1333 goto e_inval;
1334 }
1335 dev_hold(out_dev);
1336
1337 if (res.r)
1338 src_map = fld.saddr; /* no NAT support for now */
1339
1340 gateway = DN_FIB_RES_GW(res);
1341 if (res.type == RTN_NAT) {
1342 fld.daddr = dn_fib_rules_map_destination(fld.daddr, &res);
1343 dn_fib_res_put(&res);
1344 free_res = 0;
1345 if (dn_fib_lookup(&fld, &res))
1346 goto e_inval;
1347 free_res = 1;
1348 if (res.type != RTN_UNICAST)
1349 goto e_inval;
1350 flags |= RTCF_DNAT;
1351 gateway = fld.daddr;
1352 }
1353 fld.saddr = src_map;
1354 }
1355
1356 switch(res.type) {
1357 case RTN_UNICAST:
1358 /*
1359 * Forwarding check here, we only check for forwarding
1360 * being turned off, if you want to only forward intra
1361 * area, its up to you to set the routing tables up
1362 * correctly.
1363 */
1364 if (dn_db->parms.forwarding == 0)
1365 goto e_inval;
1366
1367 if (res.fi->fib_nhs > 1 && fld.flowidn_oif == 0)
1368 dn_fib_select_multipath(&fld, &res);
1369
1370 /*
1371 * Check for out_dev == in_dev. We use the RTCF_DOREDIRECT
1372 * flag as a hint to set the intra-ethernet bit when
1373 * forwarding. If we've got NAT in operation, we don't do
1374 * this optimisation.
1375 */
1376 if (out_dev == in_dev && !(flags & RTCF_NAT))
1377 flags |= RTCF_DOREDIRECT;
1378
1379 local_src = DN_FIB_RES_PREFSRC(res);
1380
1381 case RTN_BLACKHOLE:
1382 case RTN_UNREACHABLE:
1383 break;
1384 case RTN_LOCAL:
1385 flags |= RTCF_LOCAL;
1386 fld.saddr = cb->dst;
1387 fld.daddr = cb->src;
1388
1389 /* Routing tables gave us a gateway */
1390 if (gateway)
1391 goto make_route;
1392
1393 /* Packet was intra-ethernet, so we know its on-link */
1394 if (cb->rt_flags & DN_RT_F_IE) {
1395 gateway = cb->src;
1396 flags |= RTCF_DIRECTSRC;
1397 goto make_route;
1398 }
1399
1400 /* Use the default router if there is one */
1401 neigh = neigh_clone(dn_db->router);
1402 if (neigh) {
1403 gateway = ((struct dn_neigh *)neigh)->addr;
1404 goto make_route;
1405 }
1406
1407 /* Close eyes and pray */
1408 gateway = cb->src;
1409 flags |= RTCF_DIRECTSRC;
1410 goto make_route;
1411 default:
1412 goto e_inval;
1413 }
1414
1415 make_route:
1416 rt = dst_alloc(&dn_dst_ops, out_dev, 0, 0, DST_HOST);
1417 if (rt == NULL)
1418 goto e_nobufs;
1419
1420 memset(&rt->fld, 0, sizeof(rt->fld));
1421 rt->rt_saddr = fld.saddr;
1422 rt->rt_daddr = fld.daddr;
1423 rt->rt_gateway = fld.daddr;
1424 if (gateway)
1425 rt->rt_gateway = gateway;
1426 rt->rt_local_src = local_src ? local_src : rt->rt_saddr;
1427
1428 rt->rt_dst_map = fld.daddr;
1429 rt->rt_src_map = fld.saddr;
1430
1431 rt->fld.saddr = cb->src;
1432 rt->fld.daddr = cb->dst;
1433 rt->fld.flowidn_oif = 0;
1434 rt->fld.flowidn_iif = in_dev->ifindex;
1435 rt->fld.flowidn_mark = fld.flowidn_mark;
1436
1437 dst_set_neighbour(&rt->dst, neigh);
1438 rt->dst.lastuse = jiffies;
1439 rt->dst.output = dn_rt_bug;
1440 switch (res.type) {
1441 case RTN_UNICAST:
1442 rt->dst.input = dn_forward;
1443 break;
1444 case RTN_LOCAL:
1445 rt->dst.output = dn_output;
1446 rt->dst.input = dn_nsp_rx;
1447 rt->dst.dev = in_dev;
1448 flags |= RTCF_LOCAL;
1449 break;
1450 default:
1451 case RTN_UNREACHABLE:
1452 case RTN_BLACKHOLE:
1453 rt->dst.input = dst_discard;
1454 }
1455 rt->rt_flags = flags;
1456
1457 err = dn_rt_set_next_hop(rt, &res);
1458 if (err)
1459 goto e_neighbour;
1460
1461 hash = dn_hash(rt->fld.saddr, rt->fld.daddr);
1462 dn_insert_route(rt, hash, &rt);
1463 skb_dst_set(skb, &rt->dst);
1464
1465 done:
1466 if (neigh)
1467 neigh_release(neigh);
1468 if (free_res)
1469 dn_fib_res_put(&res);
1470 dev_put(in_dev);
1471 if (out_dev)
1472 dev_put(out_dev);
1473 out:
1474 return err;
1475
1476 e_inval:
1477 err = -EINVAL;
1478 goto done;
1479
1480 e_nobufs:
1481 err = -ENOBUFS;
1482 goto done;
1483
1484 e_neighbour:
1485 dst_free(&rt->dst);
1486 goto done;
1487 }
1488
1489 static int dn_route_input(struct sk_buff *skb)
1490 {
1491 struct dn_route *rt;
1492 struct dn_skb_cb *cb = DN_SKB_CB(skb);
1493 unsigned hash = dn_hash(cb->src, cb->dst);
1494
1495 if (skb_dst(skb))
1496 return 0;
1497
1498 rcu_read_lock();
1499 for(rt = rcu_dereference(dn_rt_hash_table[hash].chain); rt != NULL;
1500 rt = rcu_dereference(rt->dst.dn_next)) {
1501 if ((rt->fld.saddr == cb->src) &&
1502 (rt->fld.daddr == cb->dst) &&
1503 (rt->fld.flowidn_oif == 0) &&
1504 (rt->fld.flowidn_mark == skb->mark) &&
1505 (rt->fld.flowidn_iif == cb->iif)) {
1506 dst_use(&rt->dst, jiffies);
1507 rcu_read_unlock();
1508 skb_dst_set(skb, (struct dst_entry *)rt);
1509 return 0;
1510 }
1511 }
1512 rcu_read_unlock();
1513
1514 return dn_route_input_slow(skb);
1515 }
1516
1517 static int dn_rt_fill_info(struct sk_buff *skb, u32 pid, u32 seq,
1518 int event, int nowait, unsigned int flags)
1519 {
1520 struct dn_route *rt = (struct dn_route *)skb_dst(skb);
1521 struct rtmsg *r;
1522 struct nlmsghdr *nlh;
1523 unsigned char *b = skb_tail_pointer(skb);
1524 long expires;
1525
1526 nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*r), flags);
1527 r = NLMSG_DATA(nlh);
1528 r->rtm_family = AF_DECnet;
1529 r->rtm_dst_len = 16;
1530 r->rtm_src_len = 0;
1531 r->rtm_tos = 0;
1532 r->rtm_table = RT_TABLE_MAIN;
1533 RTA_PUT_U32(skb, RTA_TABLE, RT_TABLE_MAIN);
1534 r->rtm_type = rt->rt_type;
1535 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
1536 r->rtm_scope = RT_SCOPE_UNIVERSE;
1537 r->rtm_protocol = RTPROT_UNSPEC;
1538 if (rt->rt_flags & RTCF_NOTIFY)
1539 r->rtm_flags |= RTM_F_NOTIFY;
1540 RTA_PUT(skb, RTA_DST, 2, &rt->rt_daddr);
1541 if (rt->fld.saddr) {
1542 r->rtm_src_len = 16;
1543 RTA_PUT(skb, RTA_SRC, 2, &rt->fld.saddr);
1544 }
1545 if (rt->dst.dev)
1546 RTA_PUT(skb, RTA_OIF, sizeof(int), &rt->dst.dev->ifindex);
1547 /*
1548 * Note to self - change this if input routes reverse direction when
1549 * they deal only with inputs and not with replies like they do
1550 * currently.
1551 */
1552 RTA_PUT(skb, RTA_PREFSRC, 2, &rt->rt_local_src);
1553 if (rt->rt_daddr != rt->rt_gateway)
1554 RTA_PUT(skb, RTA_GATEWAY, 2, &rt->rt_gateway);
1555 if (rtnetlink_put_metrics(skb, dst_metrics_ptr(&rt->dst)) < 0)
1556 goto rtattr_failure;
1557 expires = rt->dst.expires ? rt->dst.expires - jiffies : 0;
1558 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, 0, 0, expires,
1559 rt->dst.error) < 0)
1560 goto rtattr_failure;
1561 if (dn_is_input_route(rt))
1562 RTA_PUT(skb, RTA_IIF, sizeof(int), &rt->fld.flowidn_iif);
1563
1564 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1565 return skb->len;
1566
1567 nlmsg_failure:
1568 rtattr_failure:
1569 nlmsg_trim(skb, b);
1570 return -1;
1571 }
1572
1573 /*
1574 * This is called by both endnodes and routers now.
1575 */
1576 static int dn_cache_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh, void *arg)
1577 {
1578 struct net *net = sock_net(in_skb->sk);
1579 struct rtattr **rta = arg;
1580 struct rtmsg *rtm = NLMSG_DATA(nlh);
1581 struct dn_route *rt = NULL;
1582 struct dn_skb_cb *cb;
1583 int err;
1584 struct sk_buff *skb;
1585 struct flowidn fld;
1586
1587 if (!net_eq(net, &init_net))
1588 return -EINVAL;
1589
1590 memset(&fld, 0, sizeof(fld));
1591 fld.flowidn_proto = DNPROTO_NSP;
1592
1593 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1594 if (skb == NULL)
1595 return -ENOBUFS;
1596 skb_reset_mac_header(skb);
1597 cb = DN_SKB_CB(skb);
1598
1599 if (rta[RTA_SRC-1])
1600 memcpy(&fld.saddr, RTA_DATA(rta[RTA_SRC-1]), 2);
1601 if (rta[RTA_DST-1])
1602 memcpy(&fld.daddr, RTA_DATA(rta[RTA_DST-1]), 2);
1603 if (rta[RTA_IIF-1])
1604 memcpy(&fld.flowidn_iif, RTA_DATA(rta[RTA_IIF-1]), sizeof(int));
1605
1606 if (fld.flowidn_iif) {
1607 struct net_device *dev;
1608 if ((dev = dev_get_by_index(&init_net, fld.flowidn_iif)) == NULL) {
1609 kfree_skb(skb);
1610 return -ENODEV;
1611 }
1612 if (!dev->dn_ptr) {
1613 dev_put(dev);
1614 kfree_skb(skb);
1615 return -ENODEV;
1616 }
1617 skb->protocol = htons(ETH_P_DNA_RT);
1618 skb->dev = dev;
1619 cb->src = fld.saddr;
1620 cb->dst = fld.daddr;
1621 local_bh_disable();
1622 err = dn_route_input(skb);
1623 local_bh_enable();
1624 memset(cb, 0, sizeof(struct dn_skb_cb));
1625 rt = (struct dn_route *)skb_dst(skb);
1626 if (!err && -rt->dst.error)
1627 err = rt->dst.error;
1628 } else {
1629 int oif = 0;
1630 if (rta[RTA_OIF - 1])
1631 memcpy(&oif, RTA_DATA(rta[RTA_OIF - 1]), sizeof(int));
1632 fld.flowidn_oif = oif;
1633 err = dn_route_output_key((struct dst_entry **)&rt, &fld, 0);
1634 }
1635
1636 if (skb->dev)
1637 dev_put(skb->dev);
1638 skb->dev = NULL;
1639 if (err)
1640 goto out_free;
1641 skb_dst_set(skb, &rt->dst);
1642 if (rtm->rtm_flags & RTM_F_NOTIFY)
1643 rt->rt_flags |= RTCF_NOTIFY;
1644
1645 err = dn_rt_fill_info(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq, RTM_NEWROUTE, 0, 0);
1646
1647 if (err == 0)
1648 goto out_free;
1649 if (err < 0) {
1650 err = -EMSGSIZE;
1651 goto out_free;
1652 }
1653
1654 return rtnl_unicast(skb, &init_net, NETLINK_CB(in_skb).pid);
1655
1656 out_free:
1657 kfree_skb(skb);
1658 return err;
1659 }
1660
1661 /*
1662 * For routers, this is called from dn_fib_dump, but for endnodes its
1663 * called directly from the rtnetlink dispatch table.
1664 */
1665 int dn_cache_dump(struct sk_buff *skb, struct netlink_callback *cb)
1666 {
1667 struct net *net = sock_net(skb->sk);
1668 struct dn_route *rt;
1669 int h, s_h;
1670 int idx, s_idx;
1671
1672 if (!net_eq(net, &init_net))
1673 return 0;
1674
1675 if (NLMSG_PAYLOAD(cb->nlh, 0) < sizeof(struct rtmsg))
1676 return -EINVAL;
1677 if (!(((struct rtmsg *)NLMSG_DATA(cb->nlh))->rtm_flags&RTM_F_CLONED))
1678 return 0;
1679
1680 s_h = cb->args[0];
1681 s_idx = idx = cb->args[1];
1682 for(h = 0; h <= dn_rt_hash_mask; h++) {
1683 if (h < s_h)
1684 continue;
1685 if (h > s_h)
1686 s_idx = 0;
1687 rcu_read_lock_bh();
1688 for(rt = rcu_dereference_bh(dn_rt_hash_table[h].chain), idx = 0;
1689 rt;
1690 rt = rcu_dereference_bh(rt->dst.dn_next), idx++) {
1691 if (idx < s_idx)
1692 continue;
1693 skb_dst_set(skb, dst_clone(&rt->dst));
1694 if (dn_rt_fill_info(skb, NETLINK_CB(cb->skb).pid,
1695 cb->nlh->nlmsg_seq, RTM_NEWROUTE,
1696 1, NLM_F_MULTI) <= 0) {
1697 skb_dst_drop(skb);
1698 rcu_read_unlock_bh();
1699 goto done;
1700 }
1701 skb_dst_drop(skb);
1702 }
1703 rcu_read_unlock_bh();
1704 }
1705
1706 done:
1707 cb->args[0] = h;
1708 cb->args[1] = idx;
1709 return skb->len;
1710 }
1711
1712 #ifdef CONFIG_PROC_FS
1713 struct dn_rt_cache_iter_state {
1714 int bucket;
1715 };
1716
1717 static struct dn_route *dn_rt_cache_get_first(struct seq_file *seq)
1718 {
1719 struct dn_route *rt = NULL;
1720 struct dn_rt_cache_iter_state *s = seq->private;
1721
1722 for(s->bucket = dn_rt_hash_mask; s->bucket >= 0; --s->bucket) {
1723 rcu_read_lock_bh();
1724 rt = rcu_dereference_bh(dn_rt_hash_table[s->bucket].chain);
1725 if (rt)
1726 break;
1727 rcu_read_unlock_bh();
1728 }
1729 return rt;
1730 }
1731
1732 static struct dn_route *dn_rt_cache_get_next(struct seq_file *seq, struct dn_route *rt)
1733 {
1734 struct dn_rt_cache_iter_state *s = seq->private;
1735
1736 rt = rcu_dereference_bh(rt->dst.dn_next);
1737 while (!rt) {
1738 rcu_read_unlock_bh();
1739 if (--s->bucket < 0)
1740 break;
1741 rcu_read_lock_bh();
1742 rt = rcu_dereference_bh(dn_rt_hash_table[s->bucket].chain);
1743 }
1744 return rt;
1745 }
1746
1747 static void *dn_rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
1748 {
1749 struct dn_route *rt = dn_rt_cache_get_first(seq);
1750
1751 if (rt) {
1752 while(*pos && (rt = dn_rt_cache_get_next(seq, rt)))
1753 --*pos;
1754 }
1755 return *pos ? NULL : rt;
1756 }
1757
1758 static void *dn_rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1759 {
1760 struct dn_route *rt = dn_rt_cache_get_next(seq, v);
1761 ++*pos;
1762 return rt;
1763 }
1764
1765 static void dn_rt_cache_seq_stop(struct seq_file *seq, void *v)
1766 {
1767 if (v)
1768 rcu_read_unlock_bh();
1769 }
1770
1771 static int dn_rt_cache_seq_show(struct seq_file *seq, void *v)
1772 {
1773 struct dn_route *rt = v;
1774 char buf1[DN_ASCBUF_LEN], buf2[DN_ASCBUF_LEN];
1775
1776 seq_printf(seq, "%-8s %-7s %-7s %04d %04d %04d\n",
1777 rt->dst.dev ? rt->dst.dev->name : "*",
1778 dn_addr2asc(le16_to_cpu(rt->rt_daddr), buf1),
1779 dn_addr2asc(le16_to_cpu(rt->rt_saddr), buf2),
1780 atomic_read(&rt->dst.__refcnt),
1781 rt->dst.__use,
1782 (int) dst_metric(&rt->dst, RTAX_RTT));
1783 return 0;
1784 }
1785
1786 static const struct seq_operations dn_rt_cache_seq_ops = {
1787 .start = dn_rt_cache_seq_start,
1788 .next = dn_rt_cache_seq_next,
1789 .stop = dn_rt_cache_seq_stop,
1790 .show = dn_rt_cache_seq_show,
1791 };
1792
1793 static int dn_rt_cache_seq_open(struct inode *inode, struct file *file)
1794 {
1795 return seq_open_private(file, &dn_rt_cache_seq_ops,
1796 sizeof(struct dn_rt_cache_iter_state));
1797 }
1798
1799 static const struct file_operations dn_rt_cache_seq_fops = {
1800 .owner = THIS_MODULE,
1801 .open = dn_rt_cache_seq_open,
1802 .read = seq_read,
1803 .llseek = seq_lseek,
1804 .release = seq_release_private,
1805 };
1806
1807 #endif /* CONFIG_PROC_FS */
1808
1809 void __init dn_route_init(void)
1810 {
1811 int i, goal, order;
1812
1813 dn_dst_ops.kmem_cachep =
1814 kmem_cache_create("dn_dst_cache", sizeof(struct dn_route), 0,
1815 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1816 dst_entries_init(&dn_dst_ops);
1817 setup_timer(&dn_route_timer, dn_dst_check_expire, 0);
1818 dn_route_timer.expires = jiffies + decnet_dst_gc_interval * HZ;
1819 add_timer(&dn_route_timer);
1820
1821 goal = totalram_pages >> (26 - PAGE_SHIFT);
1822
1823 for(order = 0; (1UL << order) < goal; order++)
1824 /* NOTHING */;
1825
1826 /*
1827 * Only want 1024 entries max, since the table is very, very unlikely
1828 * to be larger than that.
1829 */
1830 while(order && ((((1UL << order) * PAGE_SIZE) /
1831 sizeof(struct dn_rt_hash_bucket)) >= 2048))
1832 order--;
1833
1834 do {
1835 dn_rt_hash_mask = (1UL << order) * PAGE_SIZE /
1836 sizeof(struct dn_rt_hash_bucket);
1837 while(dn_rt_hash_mask & (dn_rt_hash_mask - 1))
1838 dn_rt_hash_mask--;
1839 dn_rt_hash_table = (struct dn_rt_hash_bucket *)
1840 __get_free_pages(GFP_ATOMIC, order);
1841 } while (dn_rt_hash_table == NULL && --order > 0);
1842
1843 if (!dn_rt_hash_table)
1844 panic("Failed to allocate DECnet route cache hash table\n");
1845
1846 printk(KERN_INFO
1847 "DECnet: Routing cache hash table of %u buckets, %ldKbytes\n",
1848 dn_rt_hash_mask,
1849 (long)(dn_rt_hash_mask*sizeof(struct dn_rt_hash_bucket))/1024);
1850
1851 dn_rt_hash_mask--;
1852 for(i = 0; i <= dn_rt_hash_mask; i++) {
1853 spin_lock_init(&dn_rt_hash_table[i].lock);
1854 dn_rt_hash_table[i].chain = NULL;
1855 }
1856
1857 dn_dst_ops.gc_thresh = (dn_rt_hash_mask + 1);
1858
1859 proc_net_fops_create(&init_net, "decnet_cache", S_IRUGO, &dn_rt_cache_seq_fops);
1860
1861 #ifdef CONFIG_DECNET_ROUTER
1862 rtnl_register(PF_DECnet, RTM_GETROUTE, dn_cache_getroute,
1863 dn_fib_dump, NULL);
1864 #else
1865 rtnl_register(PF_DECnet, RTM_GETROUTE, dn_cache_getroute,
1866 dn_cache_dump, NULL);
1867 #endif
1868 }
1869
1870 void __exit dn_route_cleanup(void)
1871 {
1872 del_timer(&dn_route_timer);
1873 dn_run_flush(0);
1874
1875 proc_net_remove(&init_net, "decnet_cache");
1876 dst_entries_destroy(&dn_dst_ops);
1877 }
1878
Linux 2.6 libata-dev (mirror)