1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * DECnet An implementation of the DECnet protocol suite for the LINUX
4 * operating system. DECnet is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
7 * DECnet Routing Functions (Endnode and Router)
9 * Authors: Steve Whitehouse <SteveW@ACM.org>
10 * Eduardo Marcelo Serrat <emserrat@geocities.com>
13 * Steve Whitehouse : Fixes to allow "intra-ethernet" and
14 * "return-to-sender" bits on outgoing
16 * Steve Whitehouse : Timeouts for cached routes.
17 * Steve Whitehouse : Use dst cache for input routes too.
18 * Steve Whitehouse : Fixed error values in dn_send_skb.
19 * Steve Whitehouse : Rework routing functions to better fit
20 * DECnet routing design
21 * Alexey Kuznetsov : New SMP locking
22 * Steve Whitehouse : More SMP locking changes & dn_cache_dump()
23 * Steve Whitehouse : Prerouting NF hook, now really is prerouting.
24 * Fixed possible skb leak in rtnetlink funcs.
25 * Steve Whitehouse : Dave Miller's dynamic hash table sizing and
26 * Alexey Kuznetsov's finer grained locking
28 * Steve Whitehouse : Routing is now starting to look like a
29 * sensible set of code now, mainly due to
30 * my copying the IPv4 routing code. The
31 * hooks here are modified and will continue
32 * to evolve for a while.
33 * Steve Whitehouse : Real SMP at last :-) Also new netfilter
34 * stuff. Look out raw sockets your days
36 * Steve Whitehouse : Added return-to-sender functions. Added
37 * backlog congestion level return codes.
38 * Steve Whitehouse : Fixed bug where routes were set up with
39 * no ref count on net devices.
40 * Steve Whitehouse : RCU for the route cache
41 * Steve Whitehouse : Preparations for the flow cache
42 * Steve Whitehouse : Prepare for nonlinear skbs
45 /******************************************************************************
46 (c) 1995-1998 E.M. Serrat emserrat@geocities.com
48 *******************************************************************************/
50 #include <linux/errno.h>
51 #include <linux/types.h>
52 #include <linux/socket.h>
54 #include <linux/kernel.h>
55 #include <linux/sockios.h>
56 #include <linux/net.h>
57 #include <linux/netdevice.h>
58 #include <linux/inet.h>
59 #include <linux/route.h>
60 #include <linux/in_route.h>
61 #include <linux/slab.h>
64 #include <linux/proc_fs.h>
65 #include <linux/seq_file.h>
66 #include <linux/init.h>
67 #include <linux/rtnetlink.h>
68 #include <linux/string.h>
69 #include <linux/netfilter_decnet.h>
70 #include <linux/rcupdate.h>
71 #include <linux/times.h>
72 #include <linux/export.h>
73 #include <asm/errno.h>
74 #include <net/net_namespace.h>
75 #include <net/netlink.h>
76 #include <net/neighbour.h>
79 #include <net/fib_rules.h>
81 #include <net/dn_dev.h>
82 #include <net/dn_nsp.h>
83 #include <net/dn_route.h>
84 #include <net/dn_neigh.h>
85 #include <net/dn_fib.h>
87 struct dn_rt_hash_bucket
89 struct dn_route __rcu *chain;
93 extern struct neigh_table dn_neigh_table;
96 static unsigned char dn_hiord_addr[6] = {0xAA,0x00,0x04,0x00,0x00,0x00};
98 static const int dn_rt_min_delay = 2 * HZ;
99 static const int dn_rt_max_delay = 10 * HZ;
100 static const int dn_rt_mtu_expires = 10 * 60 * HZ;
102 static unsigned long dn_rt_deadline;
104 static int dn_dst_gc(struct dst_ops *ops);
105 static struct dst_entry *dn_dst_check(struct dst_entry *, __u32);
106 static unsigned int dn_dst_default_advmss(const struct dst_entry *dst);
107 static unsigned int dn_dst_mtu(const struct dst_entry *dst);
108 static void dn_dst_destroy(struct dst_entry *);
109 static void dn_dst_ifdown(struct dst_entry *, struct net_device *dev, int how);
110 static struct dst_entry *dn_dst_negative_advice(struct dst_entry *);
111 static void dn_dst_link_failure(struct sk_buff *);
112 static void dn_dst_update_pmtu(struct dst_entry *dst, struct sock *sk,
113 struct sk_buff *skb , u32 mtu);
114 static void dn_dst_redirect(struct dst_entry *dst, struct sock *sk,
115 struct sk_buff *skb);
116 static struct neighbour *dn_dst_neigh_lookup(const struct dst_entry *dst,
119 static int dn_route_input(struct sk_buff *);
120 static void dn_run_flush(struct timer_list *unused);
122 static struct dn_rt_hash_bucket *dn_rt_hash_table;
123 static unsigned int dn_rt_hash_mask;
125 static struct timer_list dn_route_timer;
126 static DEFINE_TIMER(dn_rt_flush_timer, dn_run_flush);
127 int decnet_dst_gc_interval = 2;
129 static struct dst_ops dn_dst_ops = {
133 .check = dn_dst_check,
134 .default_advmss = dn_dst_default_advmss,
136 .cow_metrics = dst_cow_metrics_generic,
137 .destroy = dn_dst_destroy,
138 .ifdown = dn_dst_ifdown,
139 .negative_advice = dn_dst_negative_advice,
140 .link_failure = dn_dst_link_failure,
141 .update_pmtu = dn_dst_update_pmtu,
142 .redirect = dn_dst_redirect,
143 .neigh_lookup = dn_dst_neigh_lookup,
146 static void dn_dst_destroy(struct dst_entry *dst)
148 struct dn_route *rt = (struct dn_route *) dst;
151 neigh_release(rt->n);
152 dst_destroy_metrics_generic(dst);
155 static void dn_dst_ifdown(struct dst_entry *dst, struct net_device *dev, int how)
158 struct dn_route *rt = (struct dn_route *) dst;
159 struct neighbour *n = rt->n;
161 if (n && n->dev == dev) {
162 n->dev = dev_net(dev)->loopback_dev;
169 static __inline__ unsigned int dn_hash(__le16 src, __le16 dst)
171 __u16 tmp = (__u16 __force)(src ^ dst);
175 return dn_rt_hash_mask & (unsigned int)tmp;
178 static void dn_dst_check_expire(struct timer_list *unused)
182 struct dn_route __rcu **rtp;
183 unsigned long now = jiffies;
184 unsigned long expire = 120 * HZ;
186 for (i = 0; i <= dn_rt_hash_mask; i++) {
187 rtp = &dn_rt_hash_table[i].chain;
189 spin_lock(&dn_rt_hash_table[i].lock);
190 while ((rt = rcu_dereference_protected(*rtp,
191 lockdep_is_held(&dn_rt_hash_table[i].lock))) != NULL) {
192 if (atomic_read(&rt->dst.__refcnt) > 1 ||
193 (now - rt->dst.lastuse) < expire) {
199 dst_dev_put(&rt->dst);
200 dst_release(&rt->dst);
202 spin_unlock(&dn_rt_hash_table[i].lock);
204 if ((jiffies - now) > 0)
208 mod_timer(&dn_route_timer, now + decnet_dst_gc_interval * HZ);
211 static int dn_dst_gc(struct dst_ops *ops)
214 struct dn_route __rcu **rtp;
216 unsigned long now = jiffies;
217 unsigned long expire = 10 * HZ;
219 for (i = 0; i <= dn_rt_hash_mask; i++) {
221 spin_lock_bh(&dn_rt_hash_table[i].lock);
222 rtp = &dn_rt_hash_table[i].chain;
224 while ((rt = rcu_dereference_protected(*rtp,
225 lockdep_is_held(&dn_rt_hash_table[i].lock))) != NULL) {
226 if (atomic_read(&rt->dst.__refcnt) > 1 ||
227 (now - rt->dst.lastuse) < expire) {
233 dst_dev_put(&rt->dst);
234 dst_release(&rt->dst);
237 spin_unlock_bh(&dn_rt_hash_table[i].lock);
244 * The decnet standards don't impose a particular minimum mtu, what they
245 * do insist on is that the routing layer accepts a datagram of at least
246 * 230 bytes long. Here we have to subtract the routing header length from
247 * 230 to get the minimum acceptable mtu. If there is no neighbour, then we
248 * assume the worst and use a long header size.
250 * We update both the mtu and the advertised mss (i.e. the segment size we
251 * advertise to the other end).
253 static void dn_dst_update_pmtu(struct dst_entry *dst, struct sock *sk,
254 struct sk_buff *skb, u32 mtu)
256 struct dn_route *rt = (struct dn_route *) dst;
257 struct neighbour *n = rt->n;
261 dn = n ? rcu_dereference_raw(n->dev->dn_ptr) : NULL;
263 if (dn && dn->use_long == 0)
268 if (dst_metric(dst, RTAX_MTU) > mtu && mtu >= min_mtu) {
269 if (!(dst_metric_locked(dst, RTAX_MTU))) {
270 dst_metric_set(dst, RTAX_MTU, mtu);
271 dst_set_expires(dst, dn_rt_mtu_expires);
273 if (!(dst_metric_locked(dst, RTAX_ADVMSS))) {
274 u32 mss = mtu - DN_MAX_NSP_DATA_HEADER;
275 u32 existing_mss = dst_metric_raw(dst, RTAX_ADVMSS);
276 if (!existing_mss || existing_mss > mss)
277 dst_metric_set(dst, RTAX_ADVMSS, mss);
282 static void dn_dst_redirect(struct dst_entry *dst, struct sock *sk,
288 * When a route has been marked obsolete. (e.g. routing cache flush)
290 static struct dst_entry *dn_dst_check(struct dst_entry *dst, __u32 cookie)
295 static struct dst_entry *dn_dst_negative_advice(struct dst_entry *dst)
301 static void dn_dst_link_failure(struct sk_buff *skb)
305 static inline int compare_keys(struct flowidn *fl1, struct flowidn *fl2)
307 return ((fl1->daddr ^ fl2->daddr) |
308 (fl1->saddr ^ fl2->saddr) |
309 (fl1->flowidn_mark ^ fl2->flowidn_mark) |
310 (fl1->flowidn_scope ^ fl2->flowidn_scope) |
311 (fl1->flowidn_oif ^ fl2->flowidn_oif) |
312 (fl1->flowidn_iif ^ fl2->flowidn_iif)) == 0;
315 static int dn_insert_route(struct dn_route *rt, unsigned int hash, struct dn_route **rp)
317 struct dn_route *rth;
318 struct dn_route __rcu **rthp;
319 unsigned long now = jiffies;
321 rthp = &dn_rt_hash_table[hash].chain;
323 spin_lock_bh(&dn_rt_hash_table[hash].lock);
324 while ((rth = rcu_dereference_protected(*rthp,
325 lockdep_is_held(&dn_rt_hash_table[hash].lock))) != NULL) {
326 if (compare_keys(&rth->fld, &rt->fld)) {
328 *rthp = rth->dn_next;
329 rcu_assign_pointer(rth->dn_next,
330 dn_rt_hash_table[hash].chain);
331 rcu_assign_pointer(dn_rt_hash_table[hash].chain, rth);
333 dst_hold_and_use(&rth->dst, now);
334 spin_unlock_bh(&dn_rt_hash_table[hash].lock);
336 dst_release_immediate(&rt->dst);
340 rthp = &rth->dn_next;
343 rcu_assign_pointer(rt->dn_next, dn_rt_hash_table[hash].chain);
344 rcu_assign_pointer(dn_rt_hash_table[hash].chain, rt);
346 dst_hold_and_use(&rt->dst, now);
347 spin_unlock_bh(&dn_rt_hash_table[hash].lock);
352 static void dn_run_flush(struct timer_list *unused)
355 struct dn_route *rt, *next;
357 for (i = 0; i < dn_rt_hash_mask; i++) {
358 spin_lock_bh(&dn_rt_hash_table[i].lock);
360 if ((rt = xchg((struct dn_route **)&dn_rt_hash_table[i].chain, NULL)) == NULL)
361 goto nothing_to_declare;
363 for(; rt; rt = next) {
364 next = rcu_dereference_raw(rt->dn_next);
365 RCU_INIT_POINTER(rt->dn_next, NULL);
366 dst_dev_put(&rt->dst);
367 dst_release(&rt->dst);
371 spin_unlock_bh(&dn_rt_hash_table[i].lock);
375 static DEFINE_SPINLOCK(dn_rt_flush_lock);
377 void dn_rt_cache_flush(int delay)
379 unsigned long now = jiffies;
380 int user_mode = !in_interrupt();
383 delay = dn_rt_min_delay;
385 spin_lock_bh(&dn_rt_flush_lock);
387 if (del_timer(&dn_rt_flush_timer) && delay > 0 && dn_rt_deadline) {
388 long tmo = (long)(dn_rt_deadline - now);
390 if (user_mode && tmo < dn_rt_max_delay - dn_rt_min_delay)
398 spin_unlock_bh(&dn_rt_flush_lock);
403 if (dn_rt_deadline == 0)
404 dn_rt_deadline = now + dn_rt_max_delay;
406 dn_rt_flush_timer.expires = now + delay;
407 add_timer(&dn_rt_flush_timer);
408 spin_unlock_bh(&dn_rt_flush_lock);
412 * dn_return_short - Return a short packet to its sender
413 * @skb: The packet to return
416 static int dn_return_short(struct sk_buff *skb)
418 struct dn_skb_cb *cb;
423 /* Add back headers */
424 skb_push(skb, skb->data - skb_network_header(skb));
426 if ((skb = skb_unshare(skb, GFP_ATOMIC)) == NULL)
430 /* Skip packet length and point to flags */
432 *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
438 *ptr = 0; /* Zero hop count */
442 skb->pkt_type = PACKET_OUTGOING;
443 dn_rt_finish_output(skb, NULL, NULL);
444 return NET_RX_SUCCESS;
448 * dn_return_long - Return a long packet to its sender
449 * @skb: The long format packet to return
452 static int dn_return_long(struct sk_buff *skb)
454 struct dn_skb_cb *cb;
456 unsigned char *src_addr, *dst_addr;
457 unsigned char tmp[ETH_ALEN];
459 /* Add back all headers */
460 skb_push(skb, skb->data - skb_network_header(skb));
462 if ((skb = skb_unshare(skb, GFP_ATOMIC)) == NULL)
466 /* Ignore packet length and point to flags */
470 if (*ptr & DN_RT_F_PF) {
471 char padlen = (*ptr & ~DN_RT_F_PF);
475 *ptr++ = (cb->rt_flags & ~DN_RT_F_RQR) | DN_RT_F_RTS;
481 *ptr = 0; /* Zero hop count */
483 /* Swap source and destination */
484 memcpy(tmp, src_addr, ETH_ALEN);
485 memcpy(src_addr, dst_addr, ETH_ALEN);
486 memcpy(dst_addr, tmp, ETH_ALEN);
488 skb->pkt_type = PACKET_OUTGOING;
489 dn_rt_finish_output(skb, dst_addr, src_addr);
490 return NET_RX_SUCCESS;
494 * dn_route_rx_packet - Try and find a route for an incoming packet
495 * @skb: The packet to find a route for
497 * Returns: result of input function if route is found, error code otherwise
499 static int dn_route_rx_packet(struct net *net, struct sock *sk, struct sk_buff *skb)
501 struct dn_skb_cb *cb;
504 if ((err = dn_route_input(skb)) == 0)
505 return dst_input(skb);
508 if (decnet_debug_level & 4) {
509 char *devname = skb->dev ? skb->dev->name : "???";
512 "DECnet: dn_route_rx_packet: rt_flags=0x%02x dev=%s len=%d src=0x%04hx dst=0x%04hx err=%d type=%d\n",
513 (int)cb->rt_flags, devname, skb->len,
514 le16_to_cpu(cb->src), le16_to_cpu(cb->dst),
518 if ((skb->pkt_type == PACKET_HOST) && (cb->rt_flags & DN_RT_F_RQR)) {
519 switch (cb->rt_flags & DN_RT_PKT_MSK) {
520 case DN_RT_PKT_SHORT:
521 return dn_return_short(skb);
523 return dn_return_long(skb);
531 static int dn_route_rx_long(struct sk_buff *skb)
533 struct dn_skb_cb *cb = DN_SKB_CB(skb);
534 unsigned char *ptr = skb->data;
536 if (!pskb_may_pull(skb, 21)) /* 20 for long header, 1 for shortest nsp */
540 skb_reset_transport_header(skb);
542 /* Destination info */
544 cb->dst = dn_eth2dn(ptr);
545 if (memcmp(ptr, dn_hiord_addr, 4) != 0)
552 cb->src = dn_eth2dn(ptr);
553 if (memcmp(ptr, dn_hiord_addr, 4) != 0)
558 cb->hops = *ptr++; /* Visit Count */
560 return NF_HOOK(NFPROTO_DECNET, NF_DN_PRE_ROUTING,
561 &init_net, NULL, skb, skb->dev, NULL,
571 static int dn_route_rx_short(struct sk_buff *skb)
573 struct dn_skb_cb *cb = DN_SKB_CB(skb);
574 unsigned char *ptr = skb->data;
576 if (!pskb_may_pull(skb, 6)) /* 5 for short header + 1 for shortest nsp */
580 skb_reset_transport_header(skb);
582 cb->dst = *(__le16 *)ptr;
584 cb->src = *(__le16 *)ptr;
586 cb->hops = *ptr & 0x3f;
588 return NF_HOOK(NFPROTO_DECNET, NF_DN_PRE_ROUTING,
589 &init_net, NULL, skb, skb->dev, NULL,
597 static int dn_route_discard(struct net *net, struct sock *sk, struct sk_buff *skb)
600 * I know we drop the packet here, but thats considered success in
604 return NET_RX_SUCCESS;
607 static int dn_route_ptp_hello(struct net *net, struct sock *sk, struct sk_buff *skb)
610 dn_neigh_pointopoint_hello(skb);
611 return NET_RX_SUCCESS;
614 int dn_route_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
616 struct dn_skb_cb *cb;
617 unsigned char flags = 0;
618 __u16 len = le16_to_cpu(*(__le16 *)skb->data);
619 struct dn_dev *dn = rcu_dereference(dev->dn_ptr);
620 unsigned char padlen = 0;
622 if (!net_eq(dev_net(dev), &init_net))
628 if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
631 if (!pskb_may_pull(skb, 3))
645 cb->iif = dev->ifindex;
648 * If we have padding, remove it.
650 if (flags & DN_RT_F_PF) {
651 padlen = flags & ~DN_RT_F_PF;
652 if (!pskb_may_pull(skb, padlen + 1))
654 skb_pull(skb, padlen);
658 skb_reset_network_header(skb);
661 * Weed out future version DECnet
663 if (flags & DN_RT_F_VER)
666 cb->rt_flags = flags;
668 if (decnet_debug_level & 1)
670 "dn_route_rcv: got 0x%02x from %s [%d %d %d]\n",
671 (int)flags, (dev) ? dev->name : "???", len, skb->len,
674 if (flags & DN_RT_PKT_CNTL) {
675 if (unlikely(skb_linearize(skb)))
678 switch (flags & DN_RT_CNTL_MSK) {
680 dn_dev_init_pkt(skb);
683 dn_dev_veri_pkt(skb);
687 if (dn->parms.state != DN_DEV_S_RU)
690 switch (flags & DN_RT_CNTL_MSK) {
692 return NF_HOOK(NFPROTO_DECNET, NF_DN_HELLO,
693 &init_net, NULL, skb, skb->dev, NULL,
698 return NF_HOOK(NFPROTO_DECNET, NF_DN_ROUTE,
699 &init_net, NULL, skb, skb->dev, NULL,
702 return NF_HOOK(NFPROTO_DECNET, NF_DN_HELLO,
703 &init_net, NULL, skb, skb->dev, NULL,
704 dn_neigh_router_hello);
707 return NF_HOOK(NFPROTO_DECNET, NF_DN_HELLO,
708 &init_net, NULL, skb, skb->dev, NULL,
709 dn_neigh_endnode_hello);
712 if (dn->parms.state != DN_DEV_S_RU)
715 skb_pull(skb, 1); /* Pull flags */
717 switch (flags & DN_RT_PKT_MSK) {
719 return dn_route_rx_long(skb);
720 case DN_RT_PKT_SHORT:
721 return dn_route_rx_short(skb);
731 static int dn_output(struct net *net, struct sock *sk, struct sk_buff *skb)
733 struct dst_entry *dst = skb_dst(skb);
734 struct dn_route *rt = (struct dn_route *)dst;
735 struct net_device *dev = dst->dev;
736 struct dn_skb_cb *cb = DN_SKB_CB(skb);
745 cb->src = rt->rt_saddr;
746 cb->dst = rt->rt_daddr;
749 * Always set the Intra-Ethernet bit on all outgoing packets
750 * originated on this node. Only valid flag from upper layers
751 * is return-to-sender-requested. Set hop count to 0 too.
753 cb->rt_flags &= ~DN_RT_F_RQR;
754 cb->rt_flags |= DN_RT_F_IE;
757 return NF_HOOK(NFPROTO_DECNET, NF_DN_LOCAL_OUT,
758 &init_net, sk, skb, NULL, dev,
762 net_dbg_ratelimited("dn_output: This should not happen\n");
769 static int dn_forward(struct sk_buff *skb)
771 struct dn_skb_cb *cb = DN_SKB_CB(skb);
772 struct dst_entry *dst = skb_dst(skb);
773 struct dn_dev *dn_db = rcu_dereference(dst->dev->dn_ptr);
776 struct net_device *dev = skb->dev;
778 if (skb->pkt_type != PACKET_HOST)
781 /* Ensure that we have enough space for headers */
782 rt = (struct dn_route *)skb_dst(skb);
783 header_len = dn_db->use_long ? 21 : 6;
784 if (skb_cow(skb, LL_RESERVED_SPACE(rt->dst.dev)+header_len))
788 * Hop count exceeded.
793 skb->dev = rt->dst.dev;
796 * If packet goes out same interface it came in on, then set
797 * the Intra-Ethernet bit. This has no effect for short
798 * packets, so we don't need to test for them here.
800 cb->rt_flags &= ~DN_RT_F_IE;
801 if (rt->rt_flags & RTCF_DOREDIRECT)
802 cb->rt_flags |= DN_RT_F_IE;
804 return NF_HOOK(NFPROTO_DECNET, NF_DN_FORWARD,
805 &init_net, NULL, skb, dev, skb->dev,
814 * Used to catch bugs. This should never normally get
817 static int dn_rt_bug_out(struct net *net, struct sock *sk, struct sk_buff *skb)
819 struct dn_skb_cb *cb = DN_SKB_CB(skb);
821 net_dbg_ratelimited("dn_rt_bug: skb from:%04x to:%04x\n",
822 le16_to_cpu(cb->src), le16_to_cpu(cb->dst));
829 static int dn_rt_bug(struct sk_buff *skb)
831 struct dn_skb_cb *cb = DN_SKB_CB(skb);
833 net_dbg_ratelimited("dn_rt_bug: skb from:%04x to:%04x\n",
834 le16_to_cpu(cb->src), le16_to_cpu(cb->dst));
841 static unsigned int dn_dst_default_advmss(const struct dst_entry *dst)
843 return dn_mss_from_pmtu(dst->dev, dst_mtu(dst));
846 static unsigned int dn_dst_mtu(const struct dst_entry *dst)
848 unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
850 return mtu ? : dst->dev->mtu;
853 static struct neighbour *dn_dst_neigh_lookup(const struct dst_entry *dst,
857 return __neigh_lookup_errno(&dn_neigh_table, daddr, dst->dev);
860 static int dn_rt_set_next_hop(struct dn_route *rt, struct dn_fib_res *res)
862 struct dn_fib_info *fi = res->fi;
863 struct net_device *dev = rt->dst.dev;
864 unsigned int mss_metric;
868 if (DN_FIB_RES_GW(*res) &&
869 DN_FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
870 rt->rt_gateway = DN_FIB_RES_GW(*res);
871 dst_init_metrics(&rt->dst, fi->fib_metrics, true);
873 rt->rt_type = res->type;
875 if (dev != NULL && rt->n == NULL) {
876 n = __neigh_lookup_errno(&dn_neigh_table, &rt->rt_gateway, dev);
882 if (dst_metric(&rt->dst, RTAX_MTU) > rt->dst.dev->mtu)
883 dst_metric_set(&rt->dst, RTAX_MTU, rt->dst.dev->mtu);
884 mss_metric = dst_metric_raw(&rt->dst, RTAX_ADVMSS);
886 unsigned int mss = dn_mss_from_pmtu(dev, dst_mtu(&rt->dst));
887 if (mss_metric > mss)
888 dst_metric_set(&rt->dst, RTAX_ADVMSS, mss);
893 static inline int dn_match_addr(__le16 addr1, __le16 addr2)
895 __u16 tmp = le16_to_cpu(addr1) ^ le16_to_cpu(addr2);
904 static __le16 dnet_select_source(const struct net_device *dev, __le16 daddr, int scope)
907 struct dn_dev *dn_db;
908 struct dn_ifaddr *ifa;
913 dn_db = rcu_dereference(dev->dn_ptr);
914 for (ifa = rcu_dereference(dn_db->ifa_list);
916 ifa = rcu_dereference(ifa->ifa_next)) {
917 if (ifa->ifa_scope > scope)
920 saddr = ifa->ifa_local;
923 ret = dn_match_addr(daddr, ifa->ifa_local);
924 if (ret > best_match)
925 saddr = ifa->ifa_local;
927 saddr = ifa->ifa_local;
934 static inline __le16 __dn_fib_res_prefsrc(struct dn_fib_res *res)
936 return dnet_select_source(DN_FIB_RES_DEV(*res), DN_FIB_RES_GW(*res), res->scope);
939 static inline __le16 dn_fib_rules_map_destination(__le16 daddr, struct dn_fib_res *res)
941 __le16 mask = dnet_make_mask(res->prefixlen);
942 return (daddr&~mask)|res->fi->fib_nh->nh_gw;
945 static int dn_route_output_slow(struct dst_entry **pprt, const struct flowidn *oldflp, int try_hard)
947 struct flowidn fld = {
948 .daddr = oldflp->daddr,
949 .saddr = oldflp->saddr,
950 .flowidn_scope = RT_SCOPE_UNIVERSE,
951 .flowidn_mark = oldflp->flowidn_mark,
952 .flowidn_iif = LOOPBACK_IFINDEX,
953 .flowidn_oif = oldflp->flowidn_oif,
955 struct dn_route *rt = NULL;
956 struct net_device *dev_out = NULL, *dev;
957 struct neighbour *neigh = NULL;
959 unsigned int flags = 0;
960 struct dn_fib_res res = { .fi = NULL, .type = RTN_UNICAST };
965 if (decnet_debug_level & 16)
967 "dn_route_output_slow: dst=%04x src=%04x mark=%d"
968 " iif=%d oif=%d\n", le16_to_cpu(oldflp->daddr),
969 le16_to_cpu(oldflp->saddr),
970 oldflp->flowidn_mark, LOOPBACK_IFINDEX,
971 oldflp->flowidn_oif);
973 /* If we have an output interface, verify its a DECnet device */
974 if (oldflp->flowidn_oif) {
975 dev_out = dev_get_by_index(&init_net, oldflp->flowidn_oif);
977 if (dev_out && dev_out->dn_ptr == NULL) {
985 /* If we have a source address, verify that its a local address */
987 err = -EADDRNOTAVAIL;
990 if (dn_dev_islocal(dev_out, oldflp->saddr))
996 for_each_netdev_rcu(&init_net, dev) {
999 if (!dn_dev_islocal(dev, oldflp->saddr))
1001 if ((dev->flags & IFF_LOOPBACK) &&
1003 !dn_dev_islocal(dev, oldflp->daddr))
1010 if (dev_out == NULL)
1017 /* No destination? Assume its local */
1019 fld.daddr = fld.saddr;
1024 dev_out = init_net.loopback_dev;
1025 if (!dev_out->dn_ptr)
1027 err = -EADDRNOTAVAIL;
1031 fld.saddr = dnet_select_source(dev_out, 0,
1036 fld.flowidn_oif = LOOPBACK_IFINDEX;
1037 res.type = RTN_LOCAL;
1041 if (decnet_debug_level & 16)
1043 "dn_route_output_slow: initial checks complete."
1044 " dst=%04x src=%04x oif=%d try_hard=%d\n",
1045 le16_to_cpu(fld.daddr), le16_to_cpu(fld.saddr),
1046 fld.flowidn_oif, try_hard);
1049 * N.B. If the kernel is compiled without router support then
1050 * dn_fib_lookup() will evaluate to non-zero so this if () block
1051 * will always be executed.
1054 if (try_hard || (err = dn_fib_lookup(&fld, &res)) != 0) {
1055 struct dn_dev *dn_db;
1059 * Here the fallback is basically the standard algorithm for
1060 * routing in endnodes which is described in the DECnet routing
1063 * If we are not trying hard, look in neighbour cache.
1064 * The result is tested to ensure that if a specific output
1065 * device/source address was requested, then we honour that
1069 neigh = neigh_lookup_nodev(&dn_neigh_table, &init_net, &fld.daddr);
1071 if ((oldflp->flowidn_oif &&
1072 (neigh->dev->ifindex != oldflp->flowidn_oif)) ||
1074 (!dn_dev_islocal(neigh->dev,
1076 neigh_release(neigh);
1081 if (dn_dev_islocal(neigh->dev, fld.daddr)) {
1082 dev_out = init_net.loopback_dev;
1083 res.type = RTN_LOCAL;
1085 dev_out = neigh->dev;
1093 /* Not there? Perhaps its a local address */
1094 if (dev_out == NULL)
1095 dev_out = dn_dev_get_default();
1097 if (dev_out == NULL)
1099 dn_db = rcu_dereference_raw(dev_out->dn_ptr);
1102 /* Possible improvement - check all devices for local addr */
1103 if (dn_dev_islocal(dev_out, fld.daddr)) {
1105 dev_out = init_net.loopback_dev;
1107 res.type = RTN_LOCAL;
1110 /* Not local either.... try sending it to the default router */
1111 neigh = neigh_clone(dn_db->router);
1112 BUG_ON(neigh && neigh->dev != dev_out);
1114 /* Ok then, we assume its directly connected and move on */
1117 gateway = ((struct dn_neigh *)neigh)->addr;
1119 gateway = fld.daddr;
1120 if (fld.saddr == 0) {
1121 fld.saddr = dnet_select_source(dev_out, gateway,
1122 res.type == RTN_LOCAL ?
1125 if (fld.saddr == 0 && res.type != RTN_LOCAL)
1128 fld.flowidn_oif = dev_out->ifindex;
1133 if (res.type == RTN_NAT)
1136 if (res.type == RTN_LOCAL) {
1138 fld.saddr = fld.daddr;
1141 dev_out = init_net.loopback_dev;
1143 if (!dev_out->dn_ptr)
1145 fld.flowidn_oif = dev_out->ifindex;
1147 dn_fib_info_put(res.fi);
1152 if (res.fi->fib_nhs > 1 && fld.flowidn_oif == 0)
1153 dn_fib_select_multipath(&fld, &res);
1156 * We could add some logic to deal with default routes here and
1157 * get rid of some of the special casing above.
1161 fld.saddr = DN_FIB_RES_PREFSRC(res);
1165 dev_out = DN_FIB_RES_DEV(res);
1167 fld.flowidn_oif = dev_out->ifindex;
1168 gateway = DN_FIB_RES_GW(res);
1171 if (dev_out->flags & IFF_LOOPBACK)
1172 flags |= RTCF_LOCAL;
1174 rt = dst_alloc(&dn_dst_ops, dev_out, 0, DST_OBSOLETE_NONE, DST_HOST);
1179 memset(&rt->fld, 0, sizeof(rt->fld));
1180 rt->fld.saddr = oldflp->saddr;
1181 rt->fld.daddr = oldflp->daddr;
1182 rt->fld.flowidn_oif = oldflp->flowidn_oif;
1183 rt->fld.flowidn_iif = 0;
1184 rt->fld.flowidn_mark = oldflp->flowidn_mark;
1186 rt->rt_saddr = fld.saddr;
1187 rt->rt_daddr = fld.daddr;
1188 rt->rt_gateway = gateway ? gateway : fld.daddr;
1189 rt->rt_local_src = fld.saddr;
1191 rt->rt_dst_map = fld.daddr;
1192 rt->rt_src_map = fld.saddr;
1197 rt->dst.lastuse = jiffies;
1198 rt->dst.output = dn_output;
1199 rt->dst.input = dn_rt_bug;
1200 rt->rt_flags = flags;
1201 if (flags & RTCF_LOCAL)
1202 rt->dst.input = dn_nsp_rx;
1204 err = dn_rt_set_next_hop(rt, &res);
1208 hash = dn_hash(rt->fld.saddr, rt->fld.daddr);
1209 /* dn_insert_route() increments dst->__refcnt */
1210 dn_insert_route(rt, hash, (struct dn_route **)pprt);
1214 neigh_release(neigh);
1216 dn_fib_res_put(&res);
1223 err = -EADDRNOTAVAIL;
1232 dst_release_immediate(&rt->dst);
1238 * N.B. The flags may be moved into the flowi at some future stage.
1240 static int __dn_route_output_key(struct dst_entry **pprt, const struct flowidn *flp, int flags)
1242 unsigned int hash = dn_hash(flp->saddr, flp->daddr);
1243 struct dn_route *rt = NULL;
1245 if (!(flags & MSG_TRYHARD)) {
1247 for (rt = rcu_dereference_bh(dn_rt_hash_table[hash].chain); rt;
1248 rt = rcu_dereference_bh(rt->dn_next)) {
1249 if ((flp->daddr == rt->fld.daddr) &&
1250 (flp->saddr == rt->fld.saddr) &&
1251 (flp->flowidn_mark == rt->fld.flowidn_mark) &&
1252 dn_is_output_route(rt) &&
1253 (rt->fld.flowidn_oif == flp->flowidn_oif)) {
1254 dst_hold_and_use(&rt->dst, jiffies);
1255 rcu_read_unlock_bh();
1260 rcu_read_unlock_bh();
1263 return dn_route_output_slow(pprt, flp, flags);
1266 static int dn_route_output_key(struct dst_entry **pprt, struct flowidn *flp, int flags)
1270 err = __dn_route_output_key(pprt, flp, flags);
1271 if (err == 0 && flp->flowidn_proto) {
1272 *pprt = xfrm_lookup(&init_net, *pprt,
1273 flowidn_to_flowi(flp), NULL, 0);
1274 if (IS_ERR(*pprt)) {
1275 err = PTR_ERR(*pprt);
1282 int dn_route_output_sock(struct dst_entry __rcu **pprt, struct flowidn *fl, struct sock *sk, int flags)
1286 err = __dn_route_output_key(pprt, fl, flags & MSG_TRYHARD);
1287 if (err == 0 && fl->flowidn_proto) {
1288 *pprt = xfrm_lookup(&init_net, *pprt,
1289 flowidn_to_flowi(fl), sk, 0);
1290 if (IS_ERR(*pprt)) {
1291 err = PTR_ERR(*pprt);
1298 static int dn_route_input_slow(struct sk_buff *skb)
1300 struct dn_route *rt = NULL;
1301 struct dn_skb_cb *cb = DN_SKB_CB(skb);
1302 struct net_device *in_dev = skb->dev;
1303 struct net_device *out_dev = NULL;
1304 struct dn_dev *dn_db;
1305 struct neighbour *neigh = NULL;
1309 __le16 local_src = 0;
1310 struct flowidn fld = {
1313 .flowidn_scope = RT_SCOPE_UNIVERSE,
1314 .flowidn_mark = skb->mark,
1315 .flowidn_iif = skb->dev->ifindex,
1317 struct dn_fib_res res = { .fi = NULL, .type = RTN_UNREACHABLE };
1323 if ((dn_db = rcu_dereference(in_dev->dn_ptr)) == NULL)
1326 /* Zero source addresses are not allowed */
1331 * In this case we've just received a packet from a source
1332 * outside ourselves pretending to come from us. We don't
1333 * allow it any further to prevent routing loops, spoofing and
1334 * other nasties. Loopback packets already have the dst attached
1335 * so this only affects packets which have originated elsewhere.
1338 if (dn_dev_islocal(in_dev, cb->src))
1341 err = dn_fib_lookup(&fld, &res);
1346 * Is the destination us ?
1348 if (!dn_dev_islocal(in_dev, cb->dst))
1351 res.type = RTN_LOCAL;
1353 __le16 src_map = fld.saddr;
1356 out_dev = DN_FIB_RES_DEV(res);
1357 if (out_dev == NULL) {
1358 net_crit_ratelimited("Bug in dn_route_input_slow() No output device\n");
1364 src_map = fld.saddr; /* no NAT support for now */
1366 gateway = DN_FIB_RES_GW(res);
1367 if (res.type == RTN_NAT) {
1368 fld.daddr = dn_fib_rules_map_destination(fld.daddr, &res);
1369 dn_fib_res_put(&res);
1371 if (dn_fib_lookup(&fld, &res))
1374 if (res.type != RTN_UNICAST)
1377 gateway = fld.daddr;
1379 fld.saddr = src_map;
1385 * Forwarding check here, we only check for forwarding
1386 * being turned off, if you want to only forward intra
1387 * area, its up to you to set the routing tables up
1390 if (dn_db->parms.forwarding == 0)
1393 if (res.fi->fib_nhs > 1 && fld.flowidn_oif == 0)
1394 dn_fib_select_multipath(&fld, &res);
1397 * Check for out_dev == in_dev. We use the RTCF_DOREDIRECT
1398 * flag as a hint to set the intra-ethernet bit when
1399 * forwarding. If we've got NAT in operation, we don't do
1400 * this optimisation.
1402 if (out_dev == in_dev && !(flags & RTCF_NAT))
1403 flags |= RTCF_DOREDIRECT;
1405 local_src = DN_FIB_RES_PREFSRC(res);
1408 case RTN_UNREACHABLE:
1411 flags |= RTCF_LOCAL;
1412 fld.saddr = cb->dst;
1413 fld.daddr = cb->src;
1415 /* Routing tables gave us a gateway */
1419 /* Packet was intra-ethernet, so we know its on-link */
1420 if (cb->rt_flags & DN_RT_F_IE) {
1425 /* Use the default router if there is one */
1426 neigh = neigh_clone(dn_db->router);
1428 gateway = ((struct dn_neigh *)neigh)->addr;
1432 /* Close eyes and pray */
1440 rt = dst_alloc(&dn_dst_ops, out_dev, 1, DST_OBSOLETE_NONE, DST_HOST);
1445 memset(&rt->fld, 0, sizeof(rt->fld));
1446 rt->rt_saddr = fld.saddr;
1447 rt->rt_daddr = fld.daddr;
1448 rt->rt_gateway = fld.daddr;
1450 rt->rt_gateway = gateway;
1451 rt->rt_local_src = local_src ? local_src : rt->rt_saddr;
1453 rt->rt_dst_map = fld.daddr;
1454 rt->rt_src_map = fld.saddr;
1456 rt->fld.saddr = cb->src;
1457 rt->fld.daddr = cb->dst;
1458 rt->fld.flowidn_oif = 0;
1459 rt->fld.flowidn_iif = in_dev->ifindex;
1460 rt->fld.flowidn_mark = fld.flowidn_mark;
1463 rt->dst.lastuse = jiffies;
1464 rt->dst.output = dn_rt_bug_out;
1467 rt->dst.input = dn_forward;
1470 rt->dst.output = dn_output;
1471 rt->dst.input = dn_nsp_rx;
1472 rt->dst.dev = in_dev;
1473 flags |= RTCF_LOCAL;
1476 case RTN_UNREACHABLE:
1478 rt->dst.input = dst_discard;
1480 rt->rt_flags = flags;
1482 err = dn_rt_set_next_hop(rt, &res);
1486 hash = dn_hash(rt->fld.saddr, rt->fld.daddr);
1487 /* dn_insert_route() increments dst->__refcnt */
1488 dn_insert_route(rt, hash, &rt);
1489 skb_dst_set(skb, &rt->dst);
1493 neigh_release(neigh);
1495 dn_fib_res_put(&res);
1511 dst_release_immediate(&rt->dst);
1515 static int dn_route_input(struct sk_buff *skb)
1517 struct dn_route *rt;
1518 struct dn_skb_cb *cb = DN_SKB_CB(skb);
1519 unsigned int hash = dn_hash(cb->src, cb->dst);
1525 for(rt = rcu_dereference(dn_rt_hash_table[hash].chain); rt != NULL;
1526 rt = rcu_dereference(rt->dn_next)) {
1527 if ((rt->fld.saddr == cb->src) &&
1528 (rt->fld.daddr == cb->dst) &&
1529 (rt->fld.flowidn_oif == 0) &&
1530 (rt->fld.flowidn_mark == skb->mark) &&
1531 (rt->fld.flowidn_iif == cb->iif)) {
1532 dst_hold_and_use(&rt->dst, jiffies);
1534 skb_dst_set(skb, (struct dst_entry *)rt);
1540 return dn_route_input_slow(skb);
1543 static int dn_rt_fill_info(struct sk_buff *skb, u32 portid, u32 seq,
1544 int event, int nowait, unsigned int flags)
1546 struct dn_route *rt = (struct dn_route *)skb_dst(skb);
1548 struct nlmsghdr *nlh;
1551 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*r), flags);
1555 r = nlmsg_data(nlh);
1556 r->rtm_family = AF_DECnet;
1557 r->rtm_dst_len = 16;
1560 r->rtm_table = RT_TABLE_MAIN;
1561 r->rtm_type = rt->rt_type;
1562 r->rtm_flags = (rt->rt_flags & ~0xFFFF) | RTM_F_CLONED;
1563 r->rtm_scope = RT_SCOPE_UNIVERSE;
1564 r->rtm_protocol = RTPROT_UNSPEC;
1566 if (rt->rt_flags & RTCF_NOTIFY)
1567 r->rtm_flags |= RTM_F_NOTIFY;
1569 if (nla_put_u32(skb, RTA_TABLE, RT_TABLE_MAIN) < 0 ||
1570 nla_put_le16(skb, RTA_DST, rt->rt_daddr) < 0)
1573 if (rt->fld.saddr) {
1574 r->rtm_src_len = 16;
1575 if (nla_put_le16(skb, RTA_SRC, rt->fld.saddr) < 0)
1579 nla_put_u32(skb, RTA_OIF, rt->dst.dev->ifindex) < 0)
1583 * Note to self - change this if input routes reverse direction when
1584 * they deal only with inputs and not with replies like they do
1587 if (nla_put_le16(skb, RTA_PREFSRC, rt->rt_local_src) < 0)
1590 if (rt->rt_daddr != rt->rt_gateway &&
1591 nla_put_le16(skb, RTA_GATEWAY, rt->rt_gateway) < 0)
1594 if (rtnetlink_put_metrics(skb, dst_metrics_ptr(&rt->dst)) < 0)
1597 expires = rt->dst.expires ? rt->dst.expires - jiffies : 0;
1598 if (rtnl_put_cacheinfo(skb, &rt->dst, 0, expires,
1602 if (dn_is_input_route(rt) &&
1603 nla_put_u32(skb, RTA_IIF, rt->fld.flowidn_iif) < 0)
1606 nlmsg_end(skb, nlh);
1610 nlmsg_cancel(skb, nlh);
1614 const struct nla_policy rtm_dn_policy[RTA_MAX + 1] = {
1615 [RTA_DST] = { .type = NLA_U16 },
1616 [RTA_SRC] = { .type = NLA_U16 },
1617 [RTA_IIF] = { .type = NLA_U32 },
1618 [RTA_OIF] = { .type = NLA_U32 },
1619 [RTA_GATEWAY] = { .type = NLA_U16 },
1620 [RTA_PRIORITY] = { .type = NLA_U32 },
1621 [RTA_PREFSRC] = { .type = NLA_U16 },
1622 [RTA_METRICS] = { .type = NLA_NESTED },
1623 [RTA_MULTIPATH] = { .type = NLA_NESTED },
1624 [RTA_TABLE] = { .type = NLA_U32 },
1625 [RTA_MARK] = { .type = NLA_U32 },
1629 * This is called by both endnodes and routers now.
1631 static int dn_cache_getroute(struct sk_buff *in_skb, struct nlmsghdr *nlh,
1632 struct netlink_ext_ack *extack)
1634 struct net *net = sock_net(in_skb->sk);
1635 struct rtmsg *rtm = nlmsg_data(nlh);
1636 struct dn_route *rt = NULL;
1637 struct dn_skb_cb *cb;
1639 struct sk_buff *skb;
1641 struct nlattr *tb[RTA_MAX+1];
1643 if (!net_eq(net, &init_net))
1646 err = nlmsg_parse_deprecated(nlh, sizeof(*rtm), tb, RTA_MAX,
1647 rtm_dn_policy, extack);
1651 memset(&fld, 0, sizeof(fld));
1652 fld.flowidn_proto = DNPROTO_NSP;
1654 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1657 skb_reset_mac_header(skb);
1658 cb = DN_SKB_CB(skb);
1661 fld.saddr = nla_get_le16(tb[RTA_SRC]);
1664 fld.daddr = nla_get_le16(tb[RTA_DST]);
1667 fld.flowidn_iif = nla_get_u32(tb[RTA_IIF]);
1669 if (fld.flowidn_iif) {
1670 struct net_device *dev;
1671 dev = __dev_get_by_index(&init_net, fld.flowidn_iif);
1672 if (!dev || !dev->dn_ptr) {
1676 skb->protocol = htons(ETH_P_DNA_RT);
1678 cb->src = fld.saddr;
1679 cb->dst = fld.daddr;
1681 err = dn_route_input(skb);
1683 memset(cb, 0, sizeof(struct dn_skb_cb));
1684 rt = (struct dn_route *)skb_dst(skb);
1685 if (!err && -rt->dst.error)
1686 err = rt->dst.error;
1689 fld.flowidn_oif = nla_get_u32(tb[RTA_OIF]);
1691 err = dn_route_output_key((struct dst_entry **)&rt, &fld, 0);
1697 skb_dst_set(skb, &rt->dst);
1698 if (rtm->rtm_flags & RTM_F_NOTIFY)
1699 rt->rt_flags |= RTCF_NOTIFY;
1701 err = dn_rt_fill_info(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq, RTM_NEWROUTE, 0, 0);
1707 return rtnl_unicast(skb, &init_net, NETLINK_CB(in_skb).portid);
1715 * For routers, this is called from dn_fib_dump, but for endnodes its
1716 * called directly from the rtnetlink dispatch table.
1718 int dn_cache_dump(struct sk_buff *skb, struct netlink_callback *cb)
1720 struct net *net = sock_net(skb->sk);
1721 struct dn_route *rt;
1726 if (!net_eq(net, &init_net))
1729 if (nlmsg_len(cb->nlh) < sizeof(struct rtmsg))
1732 rtm = nlmsg_data(cb->nlh);
1733 if (!(rtm->rtm_flags & RTM_F_CLONED))
1737 s_idx = idx = cb->args[1];
1738 for(h = 0; h <= dn_rt_hash_mask; h++) {
1744 for(rt = rcu_dereference_bh(dn_rt_hash_table[h].chain), idx = 0;
1746 rt = rcu_dereference_bh(rt->dn_next), idx++) {
1749 skb_dst_set(skb, dst_clone(&rt->dst));
1750 if (dn_rt_fill_info(skb, NETLINK_CB(cb->skb).portid,
1751 cb->nlh->nlmsg_seq, RTM_NEWROUTE,
1752 1, NLM_F_MULTI) < 0) {
1754 rcu_read_unlock_bh();
1759 rcu_read_unlock_bh();
1768 #ifdef CONFIG_PROC_FS
1769 struct dn_rt_cache_iter_state {
1773 static struct dn_route *dn_rt_cache_get_first(struct seq_file *seq)
1775 struct dn_route *rt = NULL;
1776 struct dn_rt_cache_iter_state *s = seq->private;
1778 for(s->bucket = dn_rt_hash_mask; s->bucket >= 0; --s->bucket) {
1780 rt = rcu_dereference_bh(dn_rt_hash_table[s->bucket].chain);
1783 rcu_read_unlock_bh();
1788 static struct dn_route *dn_rt_cache_get_next(struct seq_file *seq, struct dn_route *rt)
1790 struct dn_rt_cache_iter_state *s = seq->private;
1792 rt = rcu_dereference_bh(rt->dn_next);
1794 rcu_read_unlock_bh();
1795 if (--s->bucket < 0)
1798 rt = rcu_dereference_bh(dn_rt_hash_table[s->bucket].chain);
1803 static void *dn_rt_cache_seq_start(struct seq_file *seq, loff_t *pos)
1805 struct dn_route *rt = dn_rt_cache_get_first(seq);
1808 while(*pos && (rt = dn_rt_cache_get_next(seq, rt)))
1811 return *pos ? NULL : rt;
1814 static void *dn_rt_cache_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1816 struct dn_route *rt = dn_rt_cache_get_next(seq, v);
1821 static void dn_rt_cache_seq_stop(struct seq_file *seq, void *v)
1824 rcu_read_unlock_bh();
1827 static int dn_rt_cache_seq_show(struct seq_file *seq, void *v)
1829 struct dn_route *rt = v;
1830 char buf1[DN_ASCBUF_LEN], buf2[DN_ASCBUF_LEN];
1832 seq_printf(seq, "%-8s %-7s %-7s %04d %04d %04d\n",
1833 rt->dst.dev ? rt->dst.dev->name : "*",
1834 dn_addr2asc(le16_to_cpu(rt->rt_daddr), buf1),
1835 dn_addr2asc(le16_to_cpu(rt->rt_saddr), buf2),
1836 atomic_read(&rt->dst.__refcnt),
1841 static const struct seq_operations dn_rt_cache_seq_ops = {
1842 .start = dn_rt_cache_seq_start,
1843 .next = dn_rt_cache_seq_next,
1844 .stop = dn_rt_cache_seq_stop,
1845 .show = dn_rt_cache_seq_show,
1847 #endif /* CONFIG_PROC_FS */
1849 void __init dn_route_init(void)
1853 dn_dst_ops.kmem_cachep =
1854 kmem_cache_create("dn_dst_cache", sizeof(struct dn_route), 0,
1855 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1856 dst_entries_init(&dn_dst_ops);
1857 timer_setup(&dn_route_timer, dn_dst_check_expire, 0);
1858 dn_route_timer.expires = jiffies + decnet_dst_gc_interval * HZ;
1859 add_timer(&dn_route_timer);
1861 goal = totalram_pages() >> (26 - PAGE_SHIFT);
1863 for(order = 0; (1UL << order) < goal; order++)
1867 * Only want 1024 entries max, since the table is very, very unlikely
1868 * to be larger than that.
1870 while(order && ((((1UL << order) * PAGE_SIZE) /
1871 sizeof(struct dn_rt_hash_bucket)) >= 2048))
1875 dn_rt_hash_mask = (1UL << order) * PAGE_SIZE /
1876 sizeof(struct dn_rt_hash_bucket);
1877 while(dn_rt_hash_mask & (dn_rt_hash_mask - 1))
1879 dn_rt_hash_table = (struct dn_rt_hash_bucket *)
1880 __get_free_pages(GFP_ATOMIC, order);
1881 } while (dn_rt_hash_table == NULL && --order > 0);
1883 if (!dn_rt_hash_table)
1884 panic("Failed to allocate DECnet route cache hash table\n");
1887 "DECnet: Routing cache hash table of %u buckets, %ldKbytes\n",
1889 (long)(dn_rt_hash_mask*sizeof(struct dn_rt_hash_bucket))/1024);
1892 for(i = 0; i <= dn_rt_hash_mask; i++) {
1893 spin_lock_init(&dn_rt_hash_table[i].lock);
1894 dn_rt_hash_table[i].chain = NULL;
1897 dn_dst_ops.gc_thresh = (dn_rt_hash_mask + 1);
1899 proc_create_seq_private("decnet_cache", 0444, init_net.proc_net,
1900 &dn_rt_cache_seq_ops,
1901 sizeof(struct dn_rt_cache_iter_state), NULL);
1903 #ifdef CONFIG_DECNET_ROUTER
1904 rtnl_register_module(THIS_MODULE, PF_DECnet, RTM_GETROUTE,
1905 dn_cache_getroute, dn_fib_dump, 0);
1907 rtnl_register_module(THIS_MODULE, PF_DECnet, RTM_GETROUTE,
1908 dn_cache_getroute, dn_cache_dump, 0);
1912 void __exit dn_route_cleanup(void)
1914 del_timer(&dn_route_timer);
1917 remove_proc_entry("decnet_cache", init_net.proc_net);
1918 dst_entries_destroy(&dn_dst_ops);