1 diff -urp v2.6.33/linux/include/linux/rtnetlink.h linux/include/linux/rtnetlink.h
2 --- v2.6.33/linux/include/linux/rtnetlink.h 2010-02-25 09:01:36.000000000 +0200
3 +++ linux/include/linux/rtnetlink.h 2010-02-25 11:11:52.000000000 +0200
4 @@ -304,6 +304,8 @@ struct rtnexthop {
5 #define RTNH_F_DEAD 1 /* Nexthop is dead (used by multipath) */
6 #define RTNH_F_PERVASIVE 2 /* Do recursive gateway lookup */
7 #define RTNH_F_ONLINK 4 /* Gateway is forced on link */
8 +#define RTNH_F_SUSPECT 8 /* We don't know the real state */
9 +#define RTNH_F_BADSTATE (RTNH_F_DEAD | RTNH_F_SUSPECT)
11 /* Macros to handle hexthops */
13 diff -urp v2.6.33/linux/include/net/flow.h linux/include/net/flow.h
14 --- v2.6.33/linux/include/net/flow.h 2009-03-25 09:48:32.000000000 +0200
15 +++ linux/include/net/flow.h 2010-02-25 11:11:52.000000000 +0200
16 @@ -19,6 +19,8 @@ struct flowi {
25 @@ -43,6 +45,8 @@ struct flowi {
26 #define fl6_flowlabel nl_u.ip6_u.flowlabel
27 #define fl4_dst nl_u.ip4_u.daddr
28 #define fl4_src nl_u.ip4_u.saddr
29 +#define fl4_lsrc nl_u.ip4_u.lsrc
30 +#define fl4_gw nl_u.ip4_u.gw
31 #define fl4_tos nl_u.ip4_u.tos
32 #define fl4_scope nl_u.ip4_u.scope
34 diff -urp v2.6.33/linux/include/net/ip_fib.h linux/include/net/ip_fib.h
35 --- v2.6.33/linux/include/net/ip_fib.h 2010-02-25 09:01:36.000000000 +0200
36 +++ linux/include/net/ip_fib.h 2010-02-25 11:11:52.000000000 +0200
37 @@ -207,6 +207,8 @@ extern int fib_lookup(struct net *n, str
38 extern struct fib_table *fib_new_table(struct net *net, u32 id);
39 extern struct fib_table *fib_get_table(struct net *net, u32 id);
41 +extern int fib_result_table(struct fib_result *res);
43 #endif /* CONFIG_IP_MULTIPLE_TABLES */
45 /* Exported by fib_frontend.c */
46 @@ -277,4 +279,6 @@ static inline void fib_proc_exit(struct
50 +extern rwlock_t fib_nhflags_lock;
52 #endif /* _NET_FIB_H */
53 diff -urp v2.6.33/linux/include/net/netfilter/nf_nat.h linux/include/net/netfilter/nf_nat.h
54 --- v2.6.33/linux/include/net/netfilter/nf_nat.h 2010-02-25 09:01:36.000000000 +0200
55 +++ linux/include/net/netfilter/nf_nat.h 2010-02-25 11:11:52.000000000 +0200
56 @@ -73,6 +73,13 @@ struct nf_conn_nat {
60 +/* Call input routing for SNAT-ed traffic */
61 +extern unsigned int ip_nat_route_input(unsigned int hooknum,
62 + struct sk_buff *skb,
63 + const struct net_device *in,
64 + const struct net_device *out,
65 + int (*okfn)(struct sk_buff *));
67 /* Set up the info structure to map into this range. */
68 extern unsigned int nf_nat_setup_info(struct nf_conn *ct,
69 const struct nf_nat_range *range,
70 diff -urp v2.6.33/linux/include/net/route.h linux/include/net/route.h
71 --- v2.6.33/linux/include/net/route.h 2010-02-25 09:01:36.000000000 +0200
72 +++ linux/include/net/route.h 2010-02-25 11:11:52.000000000 +0200
73 @@ -113,6 +113,7 @@ extern int __ip_route_output_key(struct
74 extern int ip_route_output_key(struct net *, struct rtable **, struct flowi *flp);
75 extern int ip_route_output_flow(struct net *, struct rtable **rp, struct flowi *flp, struct sock *sk, int flags);
76 extern int ip_route_input(struct sk_buff*, __be32 dst, __be32 src, u8 tos, struct net_device *devin);
77 +extern int ip_route_input_lookup(struct sk_buff*, __be32 dst, __be32 src, u8 tos, struct net_device *devin, __be32 lsrc);
78 extern unsigned short ip_rt_frag_needed(struct net *net, struct iphdr *iph, unsigned short new_mtu, struct net_device *dev);
79 extern void ip_rt_send_redirect(struct sk_buff *skb);
81 diff -urp v2.6.33/linux/net/bridge/br_netfilter.c linux/net/bridge/br_netfilter.c
82 --- v2.6.33/linux/net/bridge/br_netfilter.c 2010-02-25 09:01:36.000000000 +0200
83 +++ linux/net/bridge/br_netfilter.c 2010-02-25 11:11:52.000000000 +0200
84 @@ -343,6 +343,9 @@ static int br_nf_pre_routing_finish(stru
88 + /* Old skb->dst is not expected, it is lost in all cases */
91 if (nf_bridge->mask & BRNF_PKT_TYPE) {
92 skb->pkt_type = PACKET_OTHERHOST;
93 nf_bridge->mask ^= BRNF_PKT_TYPE;
94 diff -urp v2.6.33/linux/net/ipv4/fib_frontend.c linux/net/ipv4/fib_frontend.c
95 --- v2.6.33/linux/net/ipv4/fib_frontend.c 2010-02-25 09:01:36.000000000 +0200
96 +++ linux/net/ipv4/fib_frontend.c 2010-02-25 11:13:43.000000000 +0200
99 #ifndef CONFIG_IP_MULTIPLE_TABLES
101 +#define FIB_RES_TABLE(r) (RT_TABLE_MAIN)
103 static int __net_init fib4_rules_init(struct net *net)
105 struct fib_table *local_table, *main_table;
106 @@ -70,6 +72,8 @@ fail:
110 +#define FIB_RES_TABLE(r) (fib_result_table(r))
112 struct fib_table *fib_new_table(struct net *net, u32 id)
114 struct fib_table *tb;
115 @@ -124,7 +128,8 @@ void fib_select_default(struct net *net,
116 table = res->r->table;
118 tb = fib_get_table(net, table);
119 - if (FIB_RES_GW(*res) && FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK)
120 + if ((FIB_RES_GW(*res) && FIB_RES_NH(*res).nh_scope == RT_SCOPE_LINK) ||
121 + FIB_RES_NH(*res).nh_scope == RT_SCOPE_HOST)
122 fib_table_select_default(tb, flp, res);
125 @@ -241,6 +246,9 @@ int fib_validate_source(__be32 src, __be
128 struct fib_result res;
130 + unsigned char prefixlen;
131 + unsigned char scope;
132 int no_addr, rpf, accept_local;
135 @@ -269,31 +277,35 @@ int fib_validate_source(__be32 src, __be
137 *spec_dst = FIB_RES_PREFSRC(res);
138 fib_combine_itag(itag, &res);
139 -#ifdef CONFIG_IP_ROUTE_MULTIPATH
140 - if (FIB_RES_DEV(res) == dev || res.fi->fib_nhs > 1)
142 if (FIB_RES_DEV(res) == dev)
145 ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST;
149 + table = FIB_RES_TABLE(&res);
150 + prefixlen = res.prefixlen;
157 fl.oif = dev->ifindex;
160 if (fib_lookup(net, &fl, &res) == 0) {
161 - if (res.type == RTN_UNICAST) {
162 + if (res.type == RTN_UNICAST &&
163 + ((table == FIB_RES_TABLE(&res) &&
164 + res.prefixlen >= prefixlen && res.scope >= scope) ||
166 *spec_dst = FIB_RES_PREFSRC(res);
167 ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST;
178 @@ -916,9 +928,7 @@ static int fib_inetaddr_event(struct not
182 -#ifdef CONFIG_IP_ROUTE_MULTIPATH
185 rt_cache_flush(dev_net(dev), -1);
188 @@ -954,9 +964,7 @@ static int fib_netdev_event(struct notif
191 } endfor_ifa(in_dev);
192 -#ifdef CONFIG_IP_ROUTE_MULTIPATH
195 rt_cache_flush(dev_net(dev), -1);
198 diff -urp v2.6.33/linux/net/ipv4/fib_hash.c linux/net/ipv4/fib_hash.c
199 --- v2.6.33/linux/net/ipv4/fib_hash.c 2010-02-25 09:01:36.000000000 +0200
200 +++ linux/net/ipv4/fib_hash.c 2010-02-25 11:11:52.000000000 +0200
201 @@ -277,25 +277,35 @@ out:
202 void fib_table_select_default(struct fib_table *tb,
203 const struct flowi *flp, struct fib_result *res)
205 - int order, last_idx;
206 + int order, last_idx, last_dflt, last_nhsel;
207 + struct fib_alias *first_fa = NULL;
208 + struct hlist_head *head;
209 struct hlist_node *node;
211 struct fib_info *fi = NULL;
212 struct fib_info *last_resort;
213 struct fn_hash *t = (struct fn_hash *)tb->tb_data;
214 - struct fn_zone *fz = t->fn_zones[0];
215 + struct fn_zone *fz = t->fn_zones[res->prefixlen];
221 + k = fz_key(flp->fl4_dst, fz);
228 read_lock(&fib_hash_lock);
229 - hlist_for_each_entry(f, node, &fz->fz_hash[0], fn_hash) {
230 + head = &fz->fz_hash[fn_hash(k, fz)];
231 + hlist_for_each_entry(f, node, head, fn_hash) {
232 struct fib_alias *fa;
234 + if (f->fn_key != k)
237 list_for_each_entry(fa, &f->fn_alias, fa_list) {
238 struct fib_info *next_fi = fa->fa_info;
240 @@ -303,42 +313,56 @@ void fib_table_select_default(struct fib
241 fa->fa_type != RTN_UNICAST)
245 + fa->fa_tos != flp->fl4_tos)
247 if (next_fi->fib_priority > res->fi->fib_priority)
249 - if (!next_fi->fib_nh[0].nh_gw ||
250 - next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
252 fa->fa_state |= FA_S_ACCESSED;
255 - if (next_fi != res->fi)
257 - } else if (!fib_detect_death(fi, order, &last_resort,
258 - &last_idx, tb->tb_default)) {
260 + last_dflt = fa->fa_last_dflt;
263 + if (fi && !fib_detect_death(fi, order, &last_resort,
264 + &last_idx, &last_dflt, &last_nhsel, flp)) {
265 fib_result_assign(res, fi);
266 - tb->tb_default = order;
267 + first_fa->fa_last_dflt = order;
276 if (order <= 0 || fi == NULL) {
277 - tb->tb_default = -1;
278 + if (fi && fi->fib_nhs > 1 &&
279 + fib_detect_death(fi, order, &last_resort, &last_idx,
280 + &last_dflt, &last_nhsel, flp) &&
281 + last_resort == fi) {
282 + read_lock_bh(&fib_nhflags_lock);
283 + fi->fib_nh[last_nhsel].nh_flags &= ~RTNH_F_SUSPECT;
284 + read_unlock_bh(&fib_nhflags_lock);
286 + if (first_fa) first_fa->fa_last_dflt = -1;
290 if (!fib_detect_death(fi, order, &last_resort, &last_idx,
292 + &last_dflt, &last_nhsel, flp)) {
293 fib_result_assign(res, fi);
294 - tb->tb_default = order;
295 + first_fa->fa_last_dflt = order;
300 + if (last_idx >= 0) {
301 fib_result_assign(res, last_resort);
302 - tb->tb_default = last_idx;
303 + read_lock_bh(&fib_nhflags_lock);
304 + last_resort->fib_nh[last_nhsel].nh_flags &= ~RTNH_F_SUSPECT;
305 + read_unlock_bh(&fib_nhflags_lock);
306 + first_fa->fa_last_dflt = last_idx;
309 read_unlock(&fib_hash_lock);
311 @@ -462,6 +486,7 @@ int fib_table_insert(struct fib_table *t
312 write_lock_bh(&fib_hash_lock);
313 fi_drop = fa->fa_info;
315 + fa->fa_last_dflt = -1;
316 fa->fa_type = cfg->fc_type;
317 fa->fa_scope = cfg->fc_scope;
318 state = fa->fa_state;
319 @@ -516,6 +541,7 @@ int fib_table_insert(struct fib_table *t
320 new_fa->fa_type = cfg->fc_type;
321 new_fa->fa_scope = cfg->fc_scope;
322 new_fa->fa_state = 0;
323 + new_fa->fa_last_dflt = -1;
326 * Insert new entry to the list.
327 diff -urp v2.6.33/linux/net/ipv4/fib_lookup.h linux/net/ipv4/fib_lookup.h
328 --- v2.6.33/linux/net/ipv4/fib_lookup.h 2009-09-11 10:27:17.000000000 +0300
329 +++ linux/net/ipv4/fib_lookup.h 2010-02-25 11:11:52.000000000 +0200
332 struct list_head fa_list;
333 struct fib_info *fa_info;
338 @@ -37,7 +38,8 @@ extern struct fib_alias *fib_find_alias(
340 extern int fib_detect_death(struct fib_info *fi, int order,
341 struct fib_info **last_resort,
342 - int *last_idx, int dflt);
343 + int *last_idx, int *dflt, int *last_nhsel,
344 + const struct flowi *flp);
346 static inline void fib_result_assign(struct fib_result *res,
348 diff -urp v2.6.33/linux/net/ipv4/fib_rules.c linux/net/ipv4/fib_rules.c
349 --- v2.6.33/linux/net/ipv4/fib_rules.c 2010-02-25 09:01:36.000000000 +0200
350 +++ linux/net/ipv4/fib_rules.c 2010-02-25 11:11:52.000000000 +0200
351 @@ -54,6 +54,11 @@ u32 fib_rules_tclass(struct fib_result *
355 +int fib_result_table(struct fib_result *res)
357 + return res->r->table;
360 int fib_lookup(struct net *net, struct flowi *flp, struct fib_result *res)
362 struct fib_lookup_arg arg = {
363 diff -urp v2.6.33/linux/net/ipv4/fib_semantics.c linux/net/ipv4/fib_semantics.c
364 --- v2.6.33/linux/net/ipv4/fib_semantics.c 2010-02-25 09:01:36.000000000 +0200
365 +++ linux/net/ipv4/fib_semantics.c 2010-02-25 11:11:52.000000000 +0200
366 @@ -50,6 +50,7 @@ static struct hlist_head *fib_info_hash;
367 static struct hlist_head *fib_info_laddrhash;
368 static unsigned int fib_hash_size;
369 static unsigned int fib_info_cnt;
370 +rwlock_t fib_nhflags_lock = RW_LOCK_UNLOCKED;
372 #define DEVINDEX_HASHBITS 8
373 #define DEVINDEX_HASHSIZE (1U << DEVINDEX_HASHBITS)
374 @@ -186,7 +187,7 @@ static __inline__ int nh_comp(const stru
375 #ifdef CONFIG_NET_CLS_ROUTE
376 nh->nh_tclassid != onh->nh_tclassid ||
378 - ((nh->nh_flags^onh->nh_flags)&~RTNH_F_DEAD))
379 + ((nh->nh_flags^onh->nh_flags)&~RTNH_F_BADSTATE))
382 } endfor_nexthops(fi);
383 @@ -237,7 +238,7 @@ static struct fib_info *fib_find_info(co
384 nfi->fib_priority == fi->fib_priority &&
385 memcmp(nfi->fib_metrics, fi->fib_metrics,
386 sizeof(fi->fib_metrics)) == 0 &&
387 - ((nfi->fib_flags^fi->fib_flags)&~RTNH_F_DEAD) == 0 &&
388 + ((nfi->fib_flags^fi->fib_flags)&~RTNH_F_BADSTATE) == 0 &&
389 (nfi->fib_nhs == 0 || nh_comp(fi, nfi) == 0))
392 @@ -349,26 +350,70 @@ struct fib_alias *fib_find_alias(struct
395 int fib_detect_death(struct fib_info *fi, int order,
396 - struct fib_info **last_resort, int *last_idx, int dflt)
397 + struct fib_info **last_resort, int *last_idx, int *dflt,
398 + int *last_nhsel, const struct flowi *flp)
401 - int state = NUD_NONE;
404 + struct fib_nh * nh;
406 + int flag, dead = 1;
408 + /* change_nexthops(fi) { */
409 + for (nhsel = 0, nh = fi->fib_nh; nhsel < fi->fib_nhs; nh++, nhsel++) {
410 + if (flp->oif && flp->oif != nh->nh_oif)
412 + if (flp->fl4_gw && flp->fl4_gw != nh->nh_gw && nh->nh_gw &&
413 + nh->nh_scope == RT_SCOPE_LINK)
415 + if (nh->nh_flags & RTNH_F_DEAD)
418 - n = neigh_lookup(&arp_tbl, &fi->fib_nh[0].nh_gw, fi->fib_dev);
420 - state = n->nud_state;
423 - if (state == NUD_REACHABLE)
425 - if ((state&NUD_VALID) && order != dflt)
427 - if ((state&NUD_VALID) ||
428 - (*last_idx<0 && order > dflt)) {
432 + if (nh->nh_dev->flags & IFF_NOARP) {
438 + if (!nh->nh_gw || nh->nh_scope != RT_SCOPE_LINK)
439 + dst = flp->fl4_dst;
442 + n = neigh_lookup(&arp_tbl, &dst, nh->nh_dev);
444 + state = n->nud_state;
447 + if (state==NUD_REACHABLE ||
448 + ((state&NUD_VALID) && order != *dflt)) {
452 + if (!(state&NUD_VALID))
456 + if ((state&NUD_VALID) ||
457 + (*last_idx<0 && order >= *dflt)) {
460 + *last_nhsel = nhsel;
465 + read_lock_bh(&fib_nhflags_lock);
467 + nh->nh_flags |= RTNH_F_SUSPECT;
469 + nh->nh_flags &= ~RTNH_F_SUSPECT;
470 + read_unlock_bh(&fib_nhflags_lock);
473 + /* } endfor_nexthops(fi) */
478 #ifdef CONFIG_IP_ROUTE_MULTIPATH
479 @@ -540,8 +585,11 @@ static int fib_check_nh(struct fib_confi
481 if ((dev = __dev_get_by_index(net, nh->nh_oif)) == NULL)
483 - if (!(dev->flags&IFF_UP))
485 + if (!(dev->flags&IFF_UP)) {
486 + if (fi->fib_protocol != RTPROT_STATIC)
488 + nh->nh_flags |= RTNH_F_DEAD;
492 nh->nh_scope = RT_SCOPE_LINK;
493 @@ -561,24 +609,48 @@ static int fib_check_nh(struct fib_confi
494 /* It is not necessary, but requires a bit of thinking */
495 if (fl.fl4_scope < RT_SCOPE_LINK)
496 fl.fl4_scope = RT_SCOPE_LINK;
497 - if ((err = fib_lookup(net, &fl, &res)) != 0)
499 + err = fib_lookup(net, &fl, &res);
502 - if (res.type != RTN_UNICAST && res.type != RTN_LOCAL)
504 - nh->nh_scope = res.scope;
505 - nh->nh_oif = FIB_RES_OIF(res);
506 - if ((nh->nh_dev = FIB_RES_DEV(res)) == NULL)
508 - dev_hold(nh->nh_dev);
510 - if (!(nh->nh_dev->flags & IFF_UP))
514 + struct in_device *in_dev;
516 + if (err != -ENETUNREACH ||
517 + fi->fib_protocol != RTPROT_STATIC)
520 + in_dev = inetdev_by_index(net, nh->nh_oif);
521 + if (in_dev == NULL ||
522 + in_dev->dev->flags & IFF_UP) {
524 + in_dev_put(in_dev);
527 + nh->nh_flags |= RTNH_F_DEAD;
528 + nh->nh_scope = RT_SCOPE_LINK;
529 + nh->nh_dev = in_dev->dev;
530 + dev_hold(nh->nh_dev);
531 + in_dev_put(in_dev);
534 + if (res.type != RTN_UNICAST && res.type != RTN_LOCAL)
536 + nh->nh_scope = res.scope;
537 + nh->nh_oif = FIB_RES_OIF(res);
538 + if ((nh->nh_dev = FIB_RES_DEV(res)) == NULL)
540 + dev_hold(nh->nh_dev);
541 + if (!(nh->nh_dev->flags & IFF_UP)) {
542 + if (fi->fib_protocol != RTPROT_STATIC) {
546 + nh->nh_flags |= RTNH_F_DEAD;
556 struct in_device *in_dev;
558 @@ -589,8 +661,11 @@ out:
561 if (!(in_dev->dev->flags&IFF_UP)) {
562 - in_dev_put(in_dev);
564 + if (fi->fib_protocol != RTPROT_STATIC) {
565 + in_dev_put(in_dev);
568 + nh->nh_flags |= RTNH_F_DEAD;
570 nh->nh_dev = in_dev->dev;
571 dev_hold(nh->nh_dev);
572 @@ -899,8 +974,12 @@ int fib_semantic_match(struct list_head
574 if (nh->nh_flags&RTNH_F_DEAD)
576 - if (!flp->oif || flp->oif == nh->nh_oif)
578 + if (flp->oif && flp->oif != nh->nh_oif)
580 + if (flp->fl4_gw && flp->fl4_gw != nh->nh_gw &&
581 + nh->nh_gw && nh->nh_scope == RT_SCOPE_LINK)
585 #ifdef CONFIG_IP_ROUTE_MULTIPATH
586 if (nhsel < fi->fib_nhs) {
587 @@ -1080,18 +1159,29 @@ int fib_sync_down_dev(struct net_device
590 change_nexthops(fi) {
591 - if (nh->nh_flags&RTNH_F_DEAD)
593 - else if (nh->nh_dev == dev &&
594 - nh->nh_scope != scope) {
595 - nh->nh_flags |= RTNH_F_DEAD;
596 + if (nh->nh_flags&RTNH_F_DEAD) {
597 + if (fi->fib_protocol!=RTPROT_STATIC ||
598 + nh->nh_dev == NULL ||
599 + __in_dev_get_rtnl(nh->nh_dev) == NULL ||
600 + nh->nh_dev->flags&IFF_UP)
602 + } else if (nh->nh_dev == dev &&
603 + nh->nh_scope != scope) {
604 + write_lock_bh(&fib_nhflags_lock);
605 #ifdef CONFIG_IP_ROUTE_MULTIPATH
606 - spin_lock_bh(&fib_multipath_lock);
607 + spin_lock(&fib_multipath_lock);
608 + nh->nh_flags |= RTNH_F_DEAD;
609 fi->fib_power -= nh->nh_power;
611 - spin_unlock_bh(&fib_multipath_lock);
612 + spin_unlock(&fib_multipath_lock);
614 + nh->nh_flags |= RTNH_F_DEAD;
617 + write_unlock_bh(&fib_nhflags_lock);
618 + if (fi->fib_protocol!=RTPROT_STATIC ||
620 + __in_dev_get_rtnl(dev) == NULL)
623 #ifdef CONFIG_IP_ROUTE_MULTIPATH
624 if (force > 1 && nh->nh_dev == dev) {
625 @@ -1109,11 +1199,8 @@ int fib_sync_down_dev(struct net_device
629 -#ifdef CONFIG_IP_ROUTE_MULTIPATH
632 - Dead device goes up. We wake up dead nexthops.
633 - It takes sense only on multipath routes.
634 + Dead device goes up or new address is added. We wake up dead nexthops.
637 int fib_sync_up(struct net_device *dev)
638 @@ -1123,8 +1210,10 @@ int fib_sync_up(struct net_device *dev)
639 struct hlist_head *head;
640 struct hlist_node *node;
643 + struct fib_result res;
647 if (!(dev->flags&IFF_UP))
650 @@ -1132,6 +1221,7 @@ int fib_sync_up(struct net_device *dev)
651 hash = fib_devindex_hashfn(dev->ifindex);
652 head = &fib_info_devhash[hash];
656 hlist_for_each_entry(nh, node, head, nh_hash) {
657 struct fib_info *fi = nh->nh_parent;
658 @@ -1144,19 +1234,39 @@ int fib_sync_up(struct net_device *dev)
661 change_nexthops(fi) {
662 - if (!(nh->nh_flags&RTNH_F_DEAD)) {
664 + if (!(nh->nh_flags&RTNH_F_DEAD))
667 if (nh->nh_dev == NULL || !(nh->nh_dev->flags&IFF_UP))
669 if (nh->nh_dev != dev || !__in_dev_get_rtnl(dev))
671 + if (nh->nh_gw && fi->fib_protocol == RTPROT_STATIC) {
672 + struct flowi fl = {
674 + { .daddr = nh->nh_gw,
675 + .scope = nh->nh_scope } },
678 + if (fib_lookup(dev_net(dev), &fl, &res) != 0)
680 + if (res.type != RTN_UNICAST &&
681 + res.type != RTN_LOCAL) {
685 + nh->nh_scope = res.scope;
690 +#ifdef CONFIG_IP_ROUTE_MULTIPATH
691 spin_lock_bh(&fib_multipath_lock);
694 nh->nh_flags &= ~RTNH_F_DEAD;
695 +#ifdef CONFIG_IP_ROUTE_MULTIPATH
696 spin_unlock_bh(&fib_multipath_lock);
698 } endfor_nexthops(fi)
701 @@ -1164,10 +1274,14 @@ int fib_sync_up(struct net_device *dev)
711 +#ifdef CONFIG_IP_ROUTE_MULTIPATH
714 The algorithm is suboptimal, but it provides really
715 fair weighted route distribution.
716 @@ -1176,24 +1290,45 @@ int fib_sync_up(struct net_device *dev)
717 void fib_select_multipath(const struct flowi *flp, struct fib_result *res)
719 struct fib_info *fi = res->fi;
723 spin_lock_bh(&fib_multipath_lock);
727 + change_nexthops(fi) {
728 + if (flp->oif != nh->nh_oif)
730 + if (flp->fl4_gw && flp->fl4_gw != nh->nh_gw &&
731 + nh->nh_gw && nh->nh_scope == RT_SCOPE_LINK)
733 + if (!(nh->nh_flags&RTNH_F_BADSTATE)) {
734 + if (nh->nh_power > w) {
739 + } endfor_nexthops(fi);
741 + spin_unlock_bh(&fib_multipath_lock);
749 if (fi->fib_power <= 0) {
751 change_nexthops(fi) {
752 - if (!(nh->nh_flags&RTNH_F_DEAD)) {
753 + if (!(nh->nh_flags&RTNH_F_BADSTATE)) {
754 power += nh->nh_weight;
755 nh->nh_power = nh->nh_weight;
757 } endfor_nexthops(fi);
758 fi->fib_power = power;
760 - spin_unlock_bh(&fib_multipath_lock);
761 - /* Race condition: route has just become dead. */
770 @@ -1203,20 +1338,40 @@ void fib_select_multipath(const struct f
772 w = jiffies % fi->fib_power;
775 change_nexthops(fi) {
776 - if (!(nh->nh_flags&RTNH_F_DEAD) && nh->nh_power) {
777 + if (!(nh->nh_flags&RTNH_F_BADSTATE) && nh->nh_power) {
778 if ((w -= nh->nh_power) <= 0) {
781 - res->nh_sel = nhsel;
782 spin_unlock_bh(&fib_multipath_lock);
783 + res->nh_sel = nhsel;
788 + } endfor_nexthops(fi);
797 + if (!(nh->nh_flags&RTNH_F_DEAD)) {
798 + if (flp->oif && flp->oif != nh->nh_oif)
800 + if (flp->fl4_gw && flp->fl4_gw != nh->nh_gw &&
801 + nh->nh_gw && nh->nh_scope == RT_SCOPE_LINK)
803 + spin_unlock_bh(&fib_multipath_lock);
804 + res->nh_sel = nhsel;
807 } endfor_nexthops(fi);
809 /* Race condition: route has just become dead. */
811 spin_unlock_bh(&fib_multipath_lock);
814 diff -urp v2.6.33/linux/net/ipv4/fib_trie.c linux/net/ipv4/fib_trie.c
815 --- v2.6.33/linux/net/ipv4/fib_trie.c 2010-02-25 09:01:36.000000000 +0200
816 +++ linux/net/ipv4/fib_trie.c 2010-02-25 11:11:52.000000000 +0200
817 @@ -1272,6 +1272,7 @@ int fib_table_insert(struct fib_table *t
818 fi_drop = fa->fa_info;
819 new_fa->fa_tos = fa->fa_tos;
820 new_fa->fa_info = fi;
821 + new_fa->fa_last_dflt = -1;
822 new_fa->fa_type = cfg->fc_type;
823 new_fa->fa_scope = cfg->fc_scope;
824 state = fa->fa_state;
825 @@ -1312,6 +1313,7 @@ int fib_table_insert(struct fib_table *t
826 new_fa->fa_type = cfg->fc_type;
827 new_fa->fa_scope = cfg->fc_scope;
828 new_fa->fa_state = 0;
829 + new_fa->fa_last_dflt = -1;
831 * Insert new entry to the list.
833 @@ -1812,24 +1814,31 @@ void fib_table_select_default(struct fib
834 struct fib_result *res)
836 struct trie *t = (struct trie *) tb->tb_data;
837 - int order, last_idx;
838 + int order, last_idx, last_dflt, last_nhsel;
839 + struct fib_alias *first_fa = NULL;
840 struct fib_info *fi = NULL;
841 struct fib_info *last_resort;
842 struct fib_alias *fa = NULL;
843 struct list_head *fa_head;
853 + mask = inet_make_mask(res->prefixlen);
854 + key = ntohl(flp->fl4_dst & mask);
858 - l = fib_find_node(t, 0);
859 + l = fib_find_node(t, key);
863 - fa_head = get_fa_head(l, 0);
864 + fa_head = get_fa_head(l, res->prefixlen);
868 @@ -1843,39 +1852,52 @@ void fib_table_select_default(struct fib
869 fa->fa_type != RTN_UNICAST)
873 + fa->fa_tos != flp->fl4_tos)
875 if (next_fi->fib_priority > res->fi->fib_priority)
877 - if (!next_fi->fib_nh[0].nh_gw ||
878 - next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
880 fa->fa_state |= FA_S_ACCESSED;
883 - if (next_fi != res->fi)
885 - } else if (!fib_detect_death(fi, order, &last_resort,
886 - &last_idx, tb->tb_default)) {
888 + last_dflt = fa->fa_last_dflt;
891 + if (fi && !fib_detect_death(fi, order, &last_resort,
892 + &last_idx, &last_dflt, &last_nhsel, flp)) {
893 fib_result_assign(res, fi);
894 - tb->tb_default = order;
895 + first_fa->fa_last_dflt = order;
901 if (order <= 0 || fi == NULL) {
902 - tb->tb_default = -1;
903 + if (fi && fi->fib_nhs > 1 &&
904 + fib_detect_death(fi, order, &last_resort, &last_idx,
905 + &last_dflt, &last_nhsel, flp) &&
906 + last_resort == fi) {
907 + read_lock_bh(&fib_nhflags_lock);
908 + fi->fib_nh[last_nhsel].nh_flags &= ~RTNH_F_SUSPECT;
909 + read_unlock_bh(&fib_nhflags_lock);
911 + if (first_fa) first_fa->fa_last_dflt = -1;
915 if (!fib_detect_death(fi, order, &last_resort, &last_idx,
917 + &last_dflt, &last_nhsel, flp)) {
918 fib_result_assign(res, fi);
919 - tb->tb_default = order;
920 + first_fa->fa_last_dflt = order;
924 + if (last_idx >= 0) {
925 fib_result_assign(res, last_resort);
926 - tb->tb_default = last_idx;
927 + read_lock_bh(&fib_nhflags_lock);
928 + last_resort->fib_nh[last_nhsel].nh_flags &= ~RTNH_F_SUSPECT;
929 + read_unlock_bh(&fib_nhflags_lock);
930 + first_fa->fa_last_dflt = last_idx;
935 diff -urp v2.6.33/linux/net/ipv4/netfilter/ipt_MASQUERADE.c linux/net/ipv4/netfilter/ipt_MASQUERADE.c
936 --- v2.6.33/linux/net/ipv4/netfilter/ipt_MASQUERADE.c 2010-02-25 09:01:36.000000000 +0200
937 +++ linux/net/ipv4/netfilter/ipt_MASQUERADE.c 2010-02-25 11:11:52.000000000 +0200
938 @@ -51,7 +51,7 @@ masquerade_tg(struct sk_buff *skb, const
939 enum ip_conntrack_info ctinfo;
940 struct nf_nat_range newrange;
941 const struct nf_nat_multi_range_compat *mr;
942 - const struct rtable *rt;
946 NF_CT_ASSERT(par->hooknum == NF_INET_POST_ROUTING);
947 @@ -69,13 +69,28 @@ masquerade_tg(struct sk_buff *skb, const
951 - rt = skb_rtable(skb);
952 - newsrc = inet_select_addr(par->out, rt->rt_gateway, RT_SCOPE_UNIVERSE);
954 - printk("MASQUERADE: %s ate my IP address\n", par->out->name);
958 + struct flowi fl = { .nl_u = { .ip4_u =
959 + { .daddr = ip_hdr(skb)->daddr,
960 + .tos = (RT_TOS(ip_hdr(skb)->tos) |
962 + .gw = skb_rtable(skb)->rt_gateway,
965 + .oif = par->out->ifindex };
966 + if (ip_route_output_key(dev_net(par->out), &rt, &fl) != 0) {
967 + /* Funky routing can do this. */
968 + if (net_ratelimit())
969 + printk("MASQUERADE:"
970 + " No route: Rusty's brain broke!\n");
975 + newsrc = rt->rt_src;
978 nat->masq_index = par->out->ifindex;
980 /* Transfer from original range. */
981 diff -urp v2.6.33/linux/net/ipv4/netfilter/nf_nat_core.c linux/net/ipv4/netfilter/nf_nat_core.c
982 --- v2.6.33/linux/net/ipv4/netfilter/nf_nat_core.c 2010-02-25 09:01:36.000000000 +0200
983 +++ linux/net/ipv4/netfilter/nf_nat_core.c 2010-02-25 11:11:52.000000000 +0200
984 @@ -710,6 +710,52 @@ static struct pernet_operations nf_nat_n
985 .exit = nf_nat_net_exit,
989 +ip_nat_route_input(unsigned int hooknum,
990 + struct sk_buff *skb,
991 + const struct net_device *in,
992 + const struct net_device *out,
993 + int (*okfn)(struct sk_buff *))
996 + struct nf_conn *conn;
997 + enum ip_conntrack_info ctinfo;
998 + enum ip_conntrack_dir dir;
999 + unsigned long statusbit;
1002 + if (!(conn = nf_ct_get(skb, &ctinfo)))
1005 + if (!(conn->status & IPS_NAT_DONE_MASK))
1007 + dir = CTINFO2DIR(ctinfo);
1008 + statusbit = IPS_SRC_NAT;
1009 + if (dir == IP_CT_DIR_REPLY)
1010 + statusbit ^= IPS_NAT_MASK;
1011 + if (!(conn->status & statusbit))
1017 + if (skb->len < sizeof(struct iphdr))
1020 + /* use daddr in other direction as masquerade address (lsrc) */
1021 + iph = ip_hdr(skb);
1022 + saddr = conn->tuplehash[!dir].tuple.dst.u3.ip;
1023 + if (saddr == iph->saddr)
1026 + if (ip_route_input_lookup(skb, iph->daddr, iph->saddr, iph->tos,
1032 +EXPORT_SYMBOL_GPL(ip_nat_route_input);
1034 static int __init nf_nat_init(void)
1037 diff -urp v2.6.33/linux/net/ipv4/netfilter/nf_nat_standalone.c linux/net/ipv4/netfilter/nf_nat_standalone.c
1038 --- v2.6.33/linux/net/ipv4/netfilter/nf_nat_standalone.c 2010-02-25 09:01:36.000000000 +0200
1039 +++ linux/net/ipv4/netfilter/nf_nat_standalone.c 2010-02-25 11:11:52.000000000 +0200
1040 @@ -255,6 +255,14 @@ static struct nf_hook_ops nf_nat_ops[] _
1041 .hooknum = NF_INET_PRE_ROUTING,
1042 .priority = NF_IP_PRI_NAT_DST,
1044 + /* Before routing, route before mangling */
1046 + .hook = ip_nat_route_input,
1047 + .owner = THIS_MODULE,
1048 + .pf = NFPROTO_IPV4,
1049 + .hooknum = NF_INET_PRE_ROUTING,
1050 + .priority = NF_IP_PRI_LAST-1,
1052 /* After packet filtering, change source */
1055 diff -urp v2.6.33/linux/net/ipv4/route.c linux/net/ipv4/route.c
1056 --- v2.6.33/linux/net/ipv4/route.c 2010-02-25 09:01:36.000000000 +0200
1057 +++ linux/net/ipv4/route.c 2010-02-25 11:11:52.000000000 +0200
1058 @@ -695,6 +695,8 @@ static inline int compare_keys(struct fl
1059 return ((__force u32)((fl1->nl_u.ip4_u.daddr ^ fl2->nl_u.ip4_u.daddr) |
1060 (fl1->nl_u.ip4_u.saddr ^ fl2->nl_u.ip4_u.saddr)) |
1061 (fl1->mark ^ fl2->mark) |
1062 + ((__force u32)(fl1->nl_u.ip4_u.lsrc ^ fl2->nl_u.ip4_u.lsrc)) |
1063 + ((__force u32)(fl1->nl_u.ip4_u.gw ^ fl2->nl_u.ip4_u.gw)) |
1064 (*(u16 *)&fl1->nl_u.ip4_u.tos ^
1065 *(u16 *)&fl2->nl_u.ip4_u.tos) |
1066 (fl1->oif ^ fl2->oif) |
1067 @@ -1430,6 +1432,7 @@ void ip_rt_redirect(__be32 old_gw, __be3
1069 /* Gateway is different ... */
1070 rt->rt_gateway = new_gw;
1071 + if (rt->fl.fl4_gw) rt->fl.fl4_gw = new_gw;
1073 /* Redirect received -> path was valid */
1074 dst_confirm(&rth->u.dst);
1075 @@ -1875,6 +1878,7 @@ static int ip_route_input_mc(struct sk_b
1076 rth->fl.fl4_tos = tos;
1077 rth->fl.mark = skb->mark;
1078 rth->fl.fl4_src = saddr;
1079 + rth->fl.fl4_lsrc = 0;
1080 rth->rt_src = saddr;
1081 #ifdef CONFIG_NET_CLS_ROUTE
1082 rth->u.dst.tclassid = itag;
1083 @@ -1885,6 +1889,7 @@ static int ip_route_input_mc(struct sk_b
1084 dev_hold(rth->u.dst.dev);
1085 rth->idev = in_dev_get(rth->u.dst.dev);
1087 + rth->fl.fl4_gw = 0;
1088 rth->rt_gateway = daddr;
1089 rth->rt_spec_dst= spec_dst;
1090 rth->rt_genid = rt_genid(dev_net(dev));
1091 @@ -1949,7 +1954,7 @@ static int __mkroute_input(struct sk_buf
1092 struct fib_result *res,
1093 struct in_device *in_dev,
1094 __be32 daddr, __be32 saddr, u32 tos,
1095 - struct rtable **result)
1096 + __be32 lsrc, struct rtable **result)
1100 @@ -1983,6 +1988,7 @@ static int __mkroute_input(struct sk_buf
1101 flags |= RTCF_DIRECTSRC;
1103 if (out_dev == in_dev && err &&
1105 (IN_DEV_SHARED_MEDIA(out_dev) ||
1106 inet_addr_onlink(out_dev, saddr, FIB_RES_GW(*res))))
1107 flags |= RTCF_DOREDIRECT;
1108 @@ -2016,6 +2022,7 @@ static int __mkroute_input(struct sk_buf
1109 rth->fl.mark = skb->mark;
1110 rth->fl.fl4_src = saddr;
1111 rth->rt_src = saddr;
1112 + rth->fl.fl4_lsrc = lsrc;
1113 rth->rt_gateway = daddr;
1115 rth->fl.iif = in_dev->dev->ifindex;
1116 @@ -2023,6 +2030,7 @@ static int __mkroute_input(struct sk_buf
1117 dev_hold(rth->u.dst.dev);
1118 rth->idev = in_dev_get(rth->u.dst.dev);
1120 + rth->fl.fl4_gw = 0;
1121 rth->rt_spec_dst= spec_dst;
1123 rth->u.dst.input = ip_forward;
1124 @@ -2043,21 +2051,23 @@ static int __mkroute_input(struct sk_buf
1126 static int ip_mkroute_input(struct sk_buff *skb,
1127 struct fib_result *res,
1129 const struct flowi *fl,
1130 struct in_device *in_dev,
1131 - __be32 daddr, __be32 saddr, u32 tos)
1132 + __be32 daddr, __be32 saddr, u32 tos, __be32 lsrc)
1134 struct rtable* rth = NULL;
1138 + fib_select_default(net, fl, res);
1139 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1140 - if (res->fi && res->fi->fib_nhs > 1 && fl->oif == 0)
1141 + if (res->fi && res->fi->fib_nhs > 1)
1142 fib_select_multipath(fl, res);
1145 /* create a routing cache entry */
1146 - err = __mkroute_input(skb, res, in_dev, daddr, saddr, tos, &rth);
1147 + err = __mkroute_input(skb, res, in_dev, daddr, saddr, tos, lsrc, &rth);
1151 @@ -2078,18 +2088,19 @@ static int ip_mkroute_input(struct sk_bu
1154 static int ip_route_input_slow(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1155 - u8 tos, struct net_device *dev)
1156 + u8 tos, struct net_device *dev, __be32 lsrc)
1158 struct fib_result res;
1159 struct in_device *in_dev = in_dev_get(dev);
1160 struct flowi fl = { .nl_u = { .ip4_u =
1163 + .saddr = lsrc? : saddr,
1165 .scope = RT_SCOPE_UNIVERSE,
1168 - .iif = dev->ifindex };
1170 + dev_net(dev)->loopback_dev->ifindex : dev->ifindex };
1173 struct rtable * rth;
1174 @@ -2125,6 +2136,12 @@ static int ip_route_input_slow(struct sk
1175 ipv4_is_loopback(daddr))
1176 goto martian_destination;
1179 + if (ipv4_is_multicast(lsrc) || ipv4_is_lbcast(lsrc) ||
1180 + ipv4_is_zeronet(lsrc) || ipv4_is_loopback(lsrc))
1185 * Now we are ready to route packet.
1187 @@ -2134,6 +2151,8 @@ static int ip_route_input_slow(struct sk
1191 + fl.iif = dev->ifindex;
1192 + fl.fl4_src = saddr;
1194 RT_CACHE_STAT_INC(in_slow_tot);
1196 @@ -2158,7 +2177,7 @@ static int ip_route_input_slow(struct sk
1197 if (res.type != RTN_UNICAST)
1198 goto martian_destination;
1200 - err = ip_mkroute_input(skb, &res, &fl, in_dev, daddr, saddr, tos);
1201 + err = ip_mkroute_input(skb, &res, net, &fl, in_dev, daddr, saddr, tos, lsrc);
1205 @@ -2168,6 +2187,8 @@ out: return err;
1207 if (skb->protocol != htons(ETH_P_IP))
1212 if (ipv4_is_zeronet(saddr))
1213 spec_dst = inet_select_addr(dev, 0, RT_SCOPE_LINK);
1214 @@ -2209,6 +2230,7 @@ local_input:
1215 rth->u.dst.dev = net->loopback_dev;
1216 dev_hold(rth->u.dst.dev);
1217 rth->idev = in_dev_get(rth->u.dst.dev);
1218 + rth->fl.fl4_gw = 0;
1219 rth->rt_gateway = daddr;
1220 rth->rt_spec_dst= spec_dst;
1221 rth->u.dst.input= ip_local_deliver;
1222 @@ -2259,8 +2281,9 @@ martian_source:
1226 -int ip_route_input(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1227 - u8 tos, struct net_device *dev)
1229 +ip_route_input_cached(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1230 + u8 tos, struct net_device *dev, __be32 lsrc)
1232 struct rtable * rth;
1234 @@ -2281,6 +2304,7 @@ int ip_route_input(struct sk_buff *skb,
1235 if (((rth->fl.fl4_dst ^ daddr) |
1236 (rth->fl.fl4_src ^ saddr) |
1237 (rth->fl.iif ^ iif) |
1238 + (rth->fl.fl4_lsrc ^ lsrc) |
1240 (rth->fl.fl4_tos ^ tos)) == 0 &&
1241 rth->fl.mark == skb->mark &&
1242 @@ -2330,7 +2354,19 @@ skip_cache:
1246 - return ip_route_input_slow(skb, daddr, saddr, tos, dev);
1247 + return ip_route_input_slow(skb, daddr, saddr, tos, dev, lsrc);
1250 +int ip_route_input(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1251 + u8 tos, struct net_device *dev)
1253 + return ip_route_input_cached(skb, daddr, saddr, tos, dev, 0);
1256 +int ip_route_input_lookup(struct sk_buff *skb, __be32 daddr, __be32 saddr,
1257 + u8 tos, struct net_device *dev, __be32 lsrc)
1259 + return ip_route_input_cached(skb, daddr, saddr, tos, dev, lsrc);
1262 static int __mkroute_output(struct rtable **result,
1263 @@ -2402,6 +2438,7 @@ static int __mkroute_output(struct rtabl
1264 rth->fl.fl4_tos = tos;
1265 rth->fl.fl4_src = oldflp->fl4_src;
1266 rth->fl.oif = oldflp->oif;
1267 + rth->fl.fl4_gw = oldflp->fl4_gw;
1268 rth->fl.mark = oldflp->mark;
1269 rth->rt_dst = fl->fl4_dst;
1270 rth->rt_src = fl->fl4_src;
1271 @@ -2483,6 +2520,7 @@ static int ip_route_output_slow(struct n
1272 struct flowi fl = { .nl_u = { .ip4_u =
1273 { .daddr = oldflp->fl4_dst,
1274 .saddr = oldflp->fl4_src,
1275 + .gw = oldflp->fl4_gw,
1276 .tos = tos & IPTOS_RT_MASK,
1277 .scope = ((tos & RTO_ONLINK) ?
1279 @@ -2594,6 +2632,7 @@ static int ip_route_output_slow(struct n
1280 dev_out = net->loopback_dev;
1282 fl.oif = net->loopback_dev->ifindex;
1284 res.type = RTN_LOCAL;
1285 flags |= RTCF_LOCAL;
1287 @@ -2601,7 +2640,7 @@ static int ip_route_output_slow(struct n
1289 if (fib_lookup(net, &fl, &res)) {
1291 - if (oldflp->oif) {
1292 + if (oldflp->oif && dev_out->flags & IFF_UP) {
1293 /* Apparently, routing tables are wrong. Assume,
1294 that the destination is on link.
1296 @@ -2641,6 +2680,7 @@ static int ip_route_output_slow(struct n
1297 dev_out = net->loopback_dev;
1299 fl.oif = dev_out->ifindex;
1302 fib_info_put(res.fi);
1304 @@ -2648,13 +2688,12 @@ static int ip_route_output_slow(struct n
1308 + if (res.type == RTN_UNICAST)
1309 + fib_select_default(net, &fl, &res);
1310 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1311 - if (res.fi->fib_nhs > 1 && fl.oif == 0)
1312 + if (res.fi->fib_nhs > 1)
1313 fib_select_multipath(&fl, &res);
1316 - if (!res.prefixlen && res.type == RTN_UNICAST && !fl.oif)
1317 - fib_select_default(net, &fl, &res);
1320 fl.fl4_src = FIB_RES_PREFSRC(res);
1321 @@ -2695,6 +2734,7 @@ int __ip_route_output_key(struct net *ne
1322 rth->fl.fl4_src == flp->fl4_src &&
1324 rth->fl.oif == flp->oif &&
1325 + rth->fl.fl4_gw == flp->fl4_gw &&
1326 rth->fl.mark == flp->mark &&
1327 !((rth->fl.fl4_tos ^ flp->fl4_tos) &
1328 (IPTOS_RT_MASK | RTO_ONLINK)) &&
1329 @@ -3416,3 +3456,4 @@ void __init ip_static_sysctl_init(void)
1330 EXPORT_SYMBOL(__ip_select_ident);
1331 EXPORT_SYMBOL(ip_route_input);
1332 EXPORT_SYMBOL(ip_route_output_key);
1333 +EXPORT_SYMBOL(ip_route_input_lookup);